Composite Materials Research Progress
Composite Materials Research Progress
Composite Materials Research Progress
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
COMPOSITE MATERIALS<br />
RESEARCH PROGRESS
COMPOSITE MATERIALS<br />
RESEARCH PROGRESS<br />
LUCAS P. DURAND<br />
EDITOR<br />
Nova Science Publishers, Inc.<br />
New York
Copyright © 2008 by Nova Science Publishers, Inc.<br />
All rights reserved. No part of this book may be reproduced, stored in a retrieval system or<br />
transmitted in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical<br />
photocopying, recording or otherwise without the written permission of the Publisher.<br />
For permission to use material from this book please contact us:<br />
Telephone 631-231-7269; Fax 631-231-8175<br />
Web Site: http://www.novapublishers.com<br />
NOTICE TO THE READER<br />
The Publisher has taken reasonable care in the preparation of this book, but makes no expressed or<br />
implied warranty of any kind and assumes no responsibility for any errors or omissions. No<br />
liability is assumed for incidental or consequential damages in connection with or arising out of<br />
information contained in this book. The Publisher shall not be liable for any special,<br />
consequential, or exemplary damages resulting, in whole or in part, from the readers’ use of, or<br />
reliance upon, this material. Any parts of this book based on government reports are so indicated<br />
and copyright is claimed for those parts to the extent applicable to compilations of such works.<br />
Independent verification should be sought for any data, advice or recommendations contained in<br />
this book. In addition, no responsibility is assumed by the publisher for any injury and/or damage<br />
to persons or property arising from any methods, products, instructions, ideas or otherwise<br />
contained in this publication.<br />
This publication is designed to provide accurate and authoritative information with regard to the<br />
subject matter covered herein. It is sold with the clear understanding that the Publisher is not<br />
engaged in rendering legal or any other professional services. If legal or any other expert<br />
assistance is required, the services of a competent person should be sought. FROM A<br />
DECLARATION OF PARTICIPANTS JOINTLY ADOPTED BY A COMMITTEE OF THE<br />
AMERICAN BAR ASSOCIATION AND A COMMITTEE OF PUBLISHERS.<br />
LIBRARY OF CONGRESS CATALOGING-IN-PUBLICATION DATA<br />
<strong>Composite</strong> materials research progress / Lucas P. Durand, Editor.<br />
p. cm.<br />
Includes index.<br />
ISBN-13: 978-1-60692-496-9<br />
1. <strong>Composite</strong> materials. I. Durand, Lucas P.<br />
TA418.9.C6C594 2008<br />
620.1'18--dc22 2007034054<br />
Published by Nova Science Publishers, Inc. � New York
CONTENTS<br />
Preface vii<br />
Chapter 1 Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts<br />
Submitted to Environmental and Mechanical Loads<br />
Jacquemin Frédéric and Fréour Sylvain<br />
Chapter 2 Optimization of Laminated <strong>Composite</strong> Structures: Problems,<br />
Solution Procedures and Applications<br />
Michaël Bruyneel<br />
Chapter 3 Major Trends in Polymeric <strong>Composite</strong>s Technology 109<br />
W.H. Zhong, R.G. Maguire, S.S. Sangari and P.H. Wu<br />
Chapter 4 An Experimental and Analytical Study of Unidirectional<br />
Carbon Fiber Reinforced Epoxy Modified<br />
by SiC Nanoparticle<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari<br />
and Shaik Jeelani<br />
Chapter 5 Damage Evaluation and Residual Strength Prediction of CFRP<br />
Laminates by Means of Acoustic Emission Techniques<br />
Giangiacomo Minak and Andrea Zucchelli<br />
Chapter 6 <strong>Research</strong> Directions in the Fatigue Testing of Polymer<br />
<strong>Composite</strong>s<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi,<br />
G. Luyckx and J. Degrieck<br />
Chapter 7 Damage Variables in Impact Testing<br />
of <strong>Composite</strong> Laminates<br />
Maria Pia Cavatorta and Davide Salvatore Paolino<br />
Chapter 8 Electromechanical Field Concentrations and Polarization<br />
Switching by Electrodes in Piezoelectric <strong>Composite</strong>s<br />
Yasuhide Shindo and Fumio Narita<br />
1<br />
51<br />
129<br />
165<br />
209<br />
237<br />
257
vi<br />
Contents<br />
Chapter 9 Recent Advances in Discontinuously Reinforced Aluminum<br />
Based Metal Matrix Nanocomposites<br />
S.C. Tjong<br />
Index 297<br />
275
PREFACE<br />
<strong>Composite</strong> materials are engineered materials made from two or more constituent<br />
materials with significantly different physical or chemical properties and which remain<br />
separate and distinct on a macroscopic level within the finished structure. Fiber Reinforced<br />
Polymers or FRPs include Wood comprising (cellulose fibers in a lignin and hemicellulose<br />
matrix), Carbon-fiber reinforced plastic or CFRP, Glass-fiber reinforced plastic or GFRP<br />
(also GRP). If classified by matrix then there are Thermoplastic <strong>Composite</strong>s, short fiber<br />
thermoplastics, long fiber thermoplastics or long fiber reinforced thermoplastics There are<br />
numerous thermoset composites, but advanced systems usually incorporate aramid fibre and<br />
carbon fibre in an epoxy resin matrix.<br />
<strong>Composite</strong>s can also utilise metal fibres reinforcing other metals, as in Metal matrix<br />
composites or MMC. Ceramic matrix composites include Bone (hydroxyapatite reinforced<br />
with collagen fibers), Cermet (ceramic and metal) and Concrete. Organic matrix/ceramic<br />
aggregate composites include Asphalt concrete, Mastic asphalt, Mastic roller hybrid, Dental<br />
composite, Syntactic foam and Mother of Pearl. Chobham armour is a special composite used<br />
in military applications. Engineered wood includes a wide variety of different products such<br />
as Plywood, Oriented strand board, Wood plastic composite (recycled wood fiber in<br />
polyethylene matrix), Pykrete (sawdust in ice matrix), Plastic-impregnated or laminated paper<br />
or textiles, Arborite, Formica (plastic) and Micarta.<br />
<strong>Composite</strong> materials have gained popularity (despite their generally high cost) in highperformance<br />
products such as aerospace components (tails, wings , fuselages, propellors),<br />
boat and scull hulls, and racing car bodies. More mundane uses include fishing rods and<br />
storage tanks.<br />
This new book presents the latest research from around the world.<br />
The purpose of Chapter 1 is to present various application of statistical scale transition<br />
models to the analysis of polymer-matrix composites submitted to thermo-hygro-mechanical<br />
loads. In order to achieve such a goal, two approaches, classically used in the field of<br />
modelling heterogeneous material are studied: Eshelby-Kröner self-consistent model on the<br />
one hand and Mori-Tanaka approximate, on the second hand. Both models manage to handle<br />
the question of the homogenization of the microscopic properties of the constituents (matrix<br />
and reinforcements) in order to express the effective macroscopic coefficients of moisture<br />
expansion, coefficients of thermal expansion and elastic stiffness of a uni-directionally<br />
reinforced single ply. Inversion scale transition relations are provided also, in order to identify<br />
the effective unknown behaviour of a constituent. The proposed method entails to inverse
viii<br />
Lucas P. Durand<br />
scale transition models usually employed in order to predict the homogenised macroscopic<br />
elastic/hygroscopic/thermal properties of the composite ply from those of the constituents.<br />
The identification procedure involves the coupling of the inverse scale transition models to<br />
macroscopic input data obtained through either experiments or in the already published<br />
literature. Applications of the proposed approach to practical cases are provided: in particular,<br />
a very satisfactory agreement between the fitted elastic constants and the corresponding<br />
properties expected in practice for the reinforcing fiber of typical composite plies is achieved.<br />
Another part of this work is devoted to the extensive analysis of macroscopic mechanical<br />
states concentration within the constituents of the plies of a composite structure submitted to<br />
thermo-hygro-elastic loads. Both numerical and a fully explicit version of Eshelby-Kröner<br />
model are detailed. The two approaches are applied in the viewpoint of predicting the<br />
mechanical states in both the fiber and the matrix of composites structures submitted to a<br />
transient hygro-elastic load. For this purpose, rigorous continuum mechanics formalisms are<br />
used for the determination of the required time and space dependent macroscopic stresses.<br />
The reliability of the new analytical approach is checked through a comparison between the<br />
local stress states calculated in both the resin and fiber according to the new closed form<br />
solutions and the equivalent numerical model: a very good agreement between the two<br />
models was obtained.<br />
The purpose of the final part of this work consists in the determination of microscopic<br />
(local) quadratic failure criterion (in stress space) in the matrix of a composite structure<br />
submitted to purely mechanical load. The local failure criterion of the pure matrix is deduced<br />
from the macroscopic strength of the composite ply (available from experiments), using an<br />
appropriate inverse model involving the explicit scale transition relations previously obtained<br />
for the macroscopic stress concentration at microscopic level. Convenient analytical forms are<br />
provided as often as possible, else procedures required to achieve numerical calculations are<br />
extensively explained. Applications of this model are achieved for two typical carbon-fiber<br />
reinforced epoxies: the previously unknown microscopic strength coefficients and ultimate<br />
strength of the considered epoxies are identified and compared to typical expected values<br />
published in the literature.<br />
In Chapter 2 the optimal design of laminated composite structures is considered. A<br />
review of the literature is proposed. It aims at giving a general overview of the problems that<br />
a designer must face when he works with laminated composite structures and the specific<br />
solutions that have been derived. Based on it and on the industrial needs an optimization<br />
method specially devoted to composite structures is developed and presented. The related<br />
solution procedure is general and reliable. It is based on fibers orientations and ply<br />
thicknesses as design variables. It is used daily in an (European) industrial context for the<br />
design of composite aircraft box structures located in the wings, the center wing box, and the<br />
vertical and horizontal tail plane. This approach is based on sequential convex programming<br />
and consists in replacing the original optimization problem by a sequence of approximated<br />
sub-problems. A very general and self adaptive approximation scheme is used. It can consider<br />
the particular structure of the mechanical responses of composites, which can be of a different<br />
nature when both fiber orientations and plies thickness are design variables. Several<br />
numerical applications illustrate the efficiency of the proposed approach.<br />
As explained in Chapter 3, composites have been growing exponentially in technology<br />
and applications for decades. The world of aerospace has been one of the earliest and<br />
strongest proponents of advanced composites and the culmination of the recent advances in
Preface ix<br />
composite technology are realized in the Boeing Model 787 with over 50% by weight of<br />
composites, bringing the application of composites in large structures into a new age. This<br />
mostly-composite Boeing 787 has been credited with putting an end to the era of the all-metal<br />
airplane on new designs, and it is perhaps the most visible manifestation of the fact that<br />
composites are having a profound and growing effect on all sectors of society.<br />
It is generally well-known that composite materials are made of reinforcement fibers and<br />
matrix materials, and light weight and high mechanical properties are the primary benefits of<br />
a composite structure. Accordingly, the development trends in composite technology lie in 1)<br />
new material technology specifically for developing novel fibers and matrices, enhancing<br />
interfacial adhesion between fiber and matrix, hybridization and multi-functionalization, and<br />
2) more reliable, high quality, rapid and low cost manufacturing technology.<br />
New reinforcement fiber technology including next generation carbon fibers and organic<br />
fibers with improved mechanical and physical properties, such as Spectra®, Dyneema®, and<br />
Zylon®, have been developing continuously. More significantly, various nanotechnology<br />
based novel fiber reinforcements have conspicuously and rapidly appeared in recent years.<br />
Matrix materials have become as complex as the fibers, satisfying increasing demands for<br />
impact resistant and damage tolerant structure. Various means of accomplishing this have<br />
ranged from elastomeric/thermoplastic minor phases to discrete layers of toughened<br />
materials. Nano-modified polymeric matrices are mostly involved in the development trends<br />
for matrix polymer materials. Technology for enhancing the interfacial adhesion properties<br />
between the reinforcement and matrix for a composite to provide high stress-transfer ability is<br />
more critically demanded and the science of the interface is expanding. Fiber/matrix<br />
interfacial adhesion is vital for the application of the newly developed advanced<br />
reinforcement materials. Effective approaches to improving new and non-traditional<br />
treatment methods for better adhesion have just started to receive sufficient attention. Multifunctionality<br />
is also an important trend for advanced composites, in particular, utilizing<br />
nanotechnology developments in recent years to provide greater opportunities for forcing<br />
materials to play a more comprehensive role in the designs of the future.<br />
More reliable and low cost manufacturing technology has been pursued by industry and<br />
academic researchers and the traditional material forms are being replaced by those which<br />
support the growing need for high quality, rapid production rates and lower recurring costs.<br />
Major trends include the recognition of the value of resin infusion methods, automated<br />
thermoplastic processing which takes advantage of the unique advantages of that material<br />
class, and the value of moving away from dependence on the large and expensive autoclaves.<br />
In Chapter 4, an innovative manufacturing process was developed to fabricate<br />
nanophased carbon prepregs used in the manufacturing of unidirectional composite laminates.<br />
In this technique, prepregs were manufactured using solution impregnation and filament<br />
winding methods and subsequently consolidated into laminates. Spherical silicon carbide<br />
nanoparticles (β-SiC) were first infused in a high temperature epoxy through an ultrasonic<br />
cavitation process. The loading of nanoparticles was 1.5% by weight of the resin. After<br />
infusion, the nano-phased resin was used to impregnate a continuous strand of dry carbon<br />
fiber tows in a filament winding set-up. In the next step, these nanophased prepregs were<br />
wrapped over a cylindrical foam mandrel especially built for this purpose using a filament<br />
winder. Once the desired thickness was achieved, the stacked prepregs were cut along the<br />
length of the cylindrical mandrel, removed from the mandrel, and laid out open to form a<br />
rectangular panel. The panel was then consolidated in a regular compression molding
x<br />
Lucas P. Durand<br />
machine. In parallel, control panels were also fabricated following similar routes without any<br />
nanoparticle infusion. Extensive thermal and mechanical characterizations were performed to<br />
evaluate the performances of the neat and nano-phased systems. Thermo Gravimetric<br />
Analysis (TGA) results indicate that there is an increase in the degradation temperature (about<br />
7 0 C) of the nano-phased composites. Similar results from Differential Scanning Calorimetry<br />
(DSC) and Dynamic Mechanical Analysis (DMA) tests were obtained. An improvement of<br />
about 5 0 C in glass transition temperature (Tg) of nano-phased systems were also seen.<br />
Mechanical tests on the laminates indicated improvement in flexural strength and stiffness by<br />
about 32% and 20% respectively whereas in tensile properties there was a nominal<br />
improvement between 7-10%. Finally, micro numerical constitutive model and damage<br />
constitutive equations were derived and an analytical approach combining the modified shearlag<br />
model and Monte Carlo simulation technique to simulate the tensile failure process of<br />
unidirectional layered composites were also established to describe stress-strain relationships.<br />
A new approach that integrates acoustic emission (AE) and the mechanical behaviour of<br />
composite materials is presented in Chapter 5. Usually AE information is used to evaluate<br />
qualitatively the damage progression in order to assess the structural integrity of a wide<br />
variety of mechanical elements such as pressure vessels. From the other side, the mechanical<br />
information, e.g. the stress-strain curve, is used to obtain a quantitative description of the<br />
material behaviour. In order to perform a deeper analysis, a function that combines AE and<br />
mechanical information is introduced. In particular, this function depends on the strain energy<br />
and on the AE events energy, and it was used to study the behaviour of CFRP composite<br />
laminates in different applications: (i) to describe the damage progression in tensile and<br />
transversal load testing; (ii) to predict residual tensile strength of transversally loaded<br />
laminates (condition that simulates a low velocity impact).<br />
For a long time, fatigue testing of composites was only focused on providing the S-N<br />
fatigue life data. No efforts were made to gather additional data from the same test by using<br />
more advanced instrumentation methods. The development of methods such as digital image<br />
correlation (strain mapping) and optical fibre sensing allows for much better instrumentation,<br />
combined with traditional equipment such as extensometers, thermocouples and resistance<br />
measurement. In addition, validation with finite element simulations of the realistic boundary<br />
conditions and loading conditions in the experimental set-up must maximize the generated<br />
data from one single fatigue test.<br />
This research paper presents a survey of the authors’ recent research activities on fatigue<br />
in polymer composites. For almost ten years now, combined fatigue testing and modelling has<br />
been done on glass and carbon polymer composites with different lay-ups and textile<br />
architectures. Chapter 6 wants to prove that a synergetic approach between instrumented<br />
testing, detailed damage inspection and advanced numerical modelling can provide an answer<br />
to the major challenges that are still present in the research on fatigue of composites.<br />
Chapter 7 presents an overview of the damage variables proposed in the literature over<br />
the years, including a new variable recently introduced by the Authors to specifically address<br />
the problem of thick laminates subject to repeated impacts. Numerous impact data are used as<br />
a basis to address and comment potentials and limitations of the different variables. Impact<br />
data refer to single impact events as well as repeated impact tests performed on laminates<br />
with different fiber and matrix combinations and various lay-ups. Laminates of different<br />
thickness are considered, ranging from tenths to tens of millimeters.
Preface xi<br />
The analysis shows that some of the variables can indeed be used for assessing the<br />
damage tolerance of the laminate. In single impact tests, results point out the existence of an<br />
energy threshold at about 40-50% of the penetration energy, below which the damage threat<br />
is quite negligible. Other variables are not directly related to the amount of damage induced in<br />
the laminate but rather give an indication of the laminate efficiency of energy absorption.<br />
The electromechanical field concentrations due to electrodes in piezoelectric composites<br />
are investigated through numerical and experimental characterization. Chapter 8 consists of<br />
two parts. In the first part, a nonlinear finite element analysis is carried out to discuss the<br />
electromechanical fields in rectangular piezoelectric composite actuators with partial<br />
electrodes, by introducing models for polarization switching in local areas of the field<br />
concentrations. Two criteria based on the work done by electromechanical loads and the<br />
internal energy density are used. Strain measurements are also presented for a four layered<br />
piezoelectric actuator, and a comparison of the predictions with experimental data is<br />
conducted. In the second part, the electromechanical fields in the neighborhood of circular<br />
electrodes in piezoelectric disk composites are reported. Nonlinear disk device behavior<br />
induced by localized polarization switching is discussed.<br />
Aluminum-based alloys reinforced with ceramic microparticles are attractive materials<br />
for many structural applications. However, large ceramic microparticles often act as stress<br />
concentrators in the composites during mechanical loading, giving rise to failure of materials<br />
via particle cracking. In recent years, increasing demand for high performance materials has<br />
led to the development of aluminum-based nanocomposites having functions and properties<br />
that are not achievable with monolithic materials and microcomposites. The incorporation of<br />
very low volume contents of ceramic reinforcements on a nanometer scale into aluminumbased<br />
alloys yields remarkable mechanical properties such as high tensile stiffness and<br />
strength as well as excellent creep resistance. However, agglomeration of nanoparticles<br />
occurs readily during the composite fabrication, leading to inferior mechanical performance<br />
of nanocomposites with higher filler content. Cryomilling and severe plastic deformation<br />
processes have emerged as the two important processes to form ultrafine grained composites<br />
with homogeneous dispersion of reinforcing particles. In Chapter 9, recent development in the<br />
processing, structure and mechanical properties of the aluminum-based nanocomposites are<br />
addressed and discussed.
In: <strong>Composite</strong> <strong>Materials</strong> <strong>Research</strong> <strong>Progress</strong> ISBN: 1-60021-994-2<br />
Editor: Lucas P. Durand, pp. 1-50 © 2008 Nova Science Publishers, Inc.<br />
Chapter 1<br />
MULTI-SCALE ANALYSIS OF FIBER-REINFORCED<br />
COMPOSITE PARTS SUBMITTED<br />
TO ENVIRONMENTAL AND MECHANICAL LOADS<br />
Jacquemin Frédéric and Fréour Sylvain *<br />
GeM -Institut de Recherche en Génie Civil et Mécanique, Université de Nantes-Ecole<br />
Centrale de Nantes-CNRS UMR 6183, 37 Boulevard de l’Université, BP 406,<br />
44 602 Saint-Nazaire, France<br />
Abstract<br />
The purpose of this work is to present various application of statistical scale transition<br />
models to the analysis of polymer-matrix composites submitted to thermo-hygro-mechanical<br />
loads. In order to achieve such a goal, two approaches, classically used in the field of<br />
modelling heterogeneous material are studied: Eshelby-Kröner self-consistent model on the<br />
one hand and Mori-Tanaka approximate, on the second hand. Both models manage to handle<br />
the question of the homogenization of the microscopic properties of the constituents (matrix<br />
and reinforcements) in order to express the effective macroscopic coefficients of moisture<br />
expansion, coefficients of thermal expansion and elastic stiffness of a uni-directionally<br />
reinforced single ply. Inversion scale transition relations are provided also, in order to identify<br />
the effective unknown behaviour of a constituent. The proposed method entails to inverse<br />
scale transition models usually employed in order to predict the homogenised macroscopic<br />
elastic/hygroscopic/thermal properties of the composite ply from those of the constituents.<br />
The identification procedure involves the coupling of the inverse scale transition models to<br />
macroscopic input data obtained through either experiments or in the already published<br />
literature. Applications of the proposed approach to practical cases are provided: in particular,<br />
a very satisfactory agreement between the fitted elastic constants and the corresponding<br />
properties expected in practice for the reinforcing fiber of typical composite plies is achieved.<br />
Another part of this work is devoted to the extensive analysis of macroscopic mechanical<br />
states concentration within the constituents of the plies of a composite structure submitted to<br />
thermo-hygro-elastic loads. Both numerical and a fully explicit version of Eshelby-Kröner<br />
model are detailed. The two approaches are applied in the viewpoint of predicting the<br />
mechanical states in both the fiber and the matrix of composites structures submitted to a<br />
* E-mail address: sylvain.freour@univ-nantes.fr. Fax number : +33240172618. (Corresponding author)
2<br />
Jacquemin Frédéric and Fréour Sylvain<br />
transient hygro-elastic load. For this purpose, rigorous continuum mechanics formalisms are<br />
used for the determination of the required time and space dependent macroscopic stresses. The<br />
reliability of the new analytical approach is checked through a comparison between the local<br />
stress states calculated in both the resin and fiber according to the new closed form solutions<br />
and the equivalent numerical model: a very good agreement between the two models was<br />
obtained.<br />
The purpose of the final part of this work consists in the determination of microscopic<br />
(local) quadratic failure criterion (in stress space) in the matrix of a composite structure<br />
submitted to purely mechanical load. The local failure criterion of the pure matrix is deduced<br />
from the macroscopic strength of the composite ply (available from experiments), using an<br />
appropriate inverse model involving the explicit scale transition relations previously obtained<br />
for the macroscopic stress concentration at microscopic level. Convenient analytical forms are<br />
provided as often as possible, else procedures required to achieve numerical calculations are<br />
extensively explained. Applications of this model are achieved for two typical carbon-fiber<br />
reinforced epoxies: the previously unknown microscopic strength coefficients and ultimate<br />
strength of the considered epoxies are identified and compared to typical expected values<br />
published in the literature.<br />
Keywords: scale transition modelling, homogenization, identification, polymer-matrix<br />
composites.<br />
1. Introduction<br />
Carbon-reinforced epoxy based composites offer design, processing, performance and cost<br />
advantages compared to metals for manufacturing structural parts. Among the advantages,<br />
provided by carbon-reinforced epoxies over metals and ceramics, that have been recognised<br />
for years, improved fracture toughness, impact resistance, strength to weight ratio as well as<br />
high resistance to corrosion and enhanced fatigue properties have often been put in good use<br />
for practical applications (Karakuzu et al., 2001).<br />
Now, the accurate design and sizing of any structure requires the knowledge of the<br />
mechanical states experienced by the material for the possibly various loads, expected to<br />
occur during service life. Since high performance composites are being increasingly used in<br />
aerospace and marine structural applications, where they are exposed to severe environmental<br />
conditions, these composites experience hygrothermal loads as well as more classical<br />
mechanical loads. Now, unlike metallic or ceramic materials, composites are susceptible to<br />
both temperature and moisture when exposed to such working environments. These<br />
environmental conditions are known to possibly induce sometimes critical stresses<br />
distributions within the plies of the composite structures or even within their very constituents<br />
(i.e. the reinforcements on the one hand and the matrix on the second hand). Actually,<br />
carbon/epoxy composites can absorb significant amount of water and exhibit heterogeneous<br />
Coefficients of Moisture Expansion (CME) and Coefficients of Thermal Expansion (CTE)<br />
(i.e. the CME/CTE of the epoxy matrix are strongly different from the CME/CTE of the<br />
carbon fibers, as shown in: Tsai, 1987; Agbossou and Pastor, 1997; Soden et al., 1998),<br />
moreover, the diffusion of moisture in such materials is a rather slow process, resulting in the<br />
occurrence of moisture concentration gradients within their depth, during at least the transient<br />
stage (Crank, 1975). As a consequence, local stresses take place from hygro-thermal loading<br />
of composite structures which closely depends on the experienced environmental conditions,<br />
on the local intrinsic properties of the constituents and on its microstructure (the morphology
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 3<br />
of the constituents, the lay-up configuration, ... fall in this last category of factors). Now, the<br />
knowledge of internal stresses is necessary to predict a possible damage occurrence in the<br />
material during its manufacturing process or service life. Thus, the study of the development<br />
of internal stresses due to thermo-hygro-elastic loads in composites is very important in<br />
regard to any engineering application. Numerous papers, available in the literature, deal with<br />
this question, using Finite Element Analysis or Continuum Mechanics-based formalisms.<br />
These methods allow the calculation of the macroscopic stresses in each ply constituting the<br />
composite (Jacquemin and Vautrin, 2002). But, they do not provide information on the local<br />
mechanical states, in the fibers and matrix of a given ply, and, consequently, do not allow to<br />
explain the phenomenon of matrix cracking and damage development in composite structures,<br />
which originate at the microscopic level. The present work is precisely focused on the study<br />
of the internal stresses in the constituents of the ply. In order to reach this goal, scale<br />
transition models are required.<br />
The present work underlines the potential of scale-transition models, as predictive tools,<br />
complementary to continuum mechanics in order to address: i) the estimation of the effective<br />
hygro-thermo-elastic properties of a composite ply from those of its constituents (section 2),<br />
ii) the identification of the hygro-thermo-elastic properties of one constituent of a composite<br />
ply (section 3), iii) the estimation of the local mechanical states experienced in each<br />
constituent of a composite structure (section 4), iv) the identification of the local strength of<br />
the constitutive matrix (section 5).<br />
Section 6 of this paper is mainly dedicated to conclusions about the above listed sections<br />
the whereas section 7 is devoted to the introducing some scientifically appealing perspectives<br />
of research in the field of composites materials which are highly considered for further<br />
investigation in the forthcoming years.<br />
2. Scale-Transition Model for Predicting the Macroscopic<br />
Thermo-Hygro-Elastic Properties of a <strong>Composite</strong> Ply<br />
2.1. Introduction<br />
Scale transition models are based on a multi-scale representation of materials. In the case of<br />
composite materials, for instance, a two-scale model is sufficient:<br />
- The properties and mechanical states of either the resin or its reinforcements are<br />
respectively indicated by the superscripts m and r . These constituents define the socalled<br />
“pseudo-macroscopic” scale of the material (Sprauel and Castex, 1991).<br />
- Homogenisation operations performed over its aforementioned constituents are<br />
assumed to provide the effective behaviour of the composite ply, which defines the<br />
macroscopic scale of the model. It is denoted by the superscript I . This definition<br />
also enables to consider an uni-directional reinforcement at macroscopic scale, which<br />
is a satisfactorily realistic statement, compared to the present design of composite<br />
structures (except for the particular case of woven-composites that will be<br />
specifically discussed in section 7.1).
4<br />
Jacquemin Frédéric and Fréour Sylvain<br />
As for the composite structure, it is actually constituted by an assembly of the above<br />
described composite plies, each of them possibly having the principal axis of their<br />
reinforcements differently oriented from one to another. This approach enables to treat the<br />
case of multi-directional laminates, as shown, for example, in (Fréour et al., 2005a).<br />
2.2. The Classical Practical Strategy for the Direct Application of<br />
Homogenisation Procedures<br />
Within scale transition modeling, the local properties of the i−superscripted constituents are<br />
usually considered to be known (i.e. the pseudo-macroscopic stiffnesses, L i , coefficients of<br />
thermal expansion M i and coefficients of moisture expansion β i ), whereas the corresponding<br />
effective macroscopic properties of the composite structure (respectively, L I , M I and β I ) are a<br />
priori unknown and results from (often numerical) computations.<br />
Among the numerous, available in the literature scale transition models, able to handle<br />
such a problem, most involve rough-and-ready theoretical frameworks: Voigt (Voigt, 1928),<br />
Reuss, (Reuss, 1929), Neerfeld-Hill (Neerfeld, 1942; Hill, 1952), Tsai-Hahn (Tsai and Hahn,<br />
1980) and Mori-Tanaka (Mori and Tanaka, 1973; Tanaka and Mori, 1970) approximates fall<br />
in this category. This is not satisfying, since such a model does not properly depict the real<br />
physical conditions experienced in practice by the material. In spite of this lack of physical<br />
realism, some of the aforementioned models do nevertheless provide a numerically satisfying<br />
estimation of the effective properties of a composite ply, by comparison with the<br />
experimental values or others, more rigorous models. Both Tsai-Hahn and Mori-Tanaka<br />
models fulfil this interesting condition (Jacquemin et al., 2005; Fréour et al., 2006a).<br />
Nevertheless, in the field of scale transition modelling, the best candidate remains Kröner-<br />
Eshelby self-consistent model, because only this model takes into account a rigorous<br />
treatment of the thermo-hygro-elastic interactions between the homogeneous macroscopic<br />
medium and its heterogeneous constituents, as well as this model enables handling the<br />
microstructure (i.e. the particular morphology of the constituents, especially that of the<br />
reinforcements).<br />
2.3. Estimating the Effective Properties of a <strong>Composite</strong> Ply through Eshelby-<br />
Kröner Self-consistent Model<br />
Self-consistent models based on the mathematical formalism proposed by Kröner (Kröner,<br />
1958) constitute a reliable method to predict the micromechanical behavior of heterogeneous<br />
materials. The method was initially introduced to treat the case of polycrystalline materials,<br />
i.e. duplex steels, aluminium alloys, etc., submitted to purely elastic loads.<br />
Estimations of homogenized elastic properties and related problems have been given in<br />
several works (François, 1991; Mabelly, 1996; Kocks et al., 1998). The model was thereafter<br />
extended to thermoelastic loads and gave satisfactory results on either single-phase (Turner<br />
and Tome, 1994; Gloaguen et al., 2002) or two-phases (Fréour et al., 2003a and 2003b)<br />
materials. More recently, this classical model was improved in order to take into account<br />
stresses and strains due to moisture in carbon fiber-reinforced polymer–matrix composites.
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 5<br />
Therefore, the formalism was extent so that homogenisation relations were established for<br />
estimating the macroscopic CME from those of the constituents (Jacquemin et al., 2005).<br />
Many previously published documents have been dedicated to the determination of (at<br />
least some of) the effective thermo-hygro-elastic properties of heterogeneous materials<br />
through Kröner-Eshelby self-consistent approach (Kocks et al., 1998; Gloaguen et al., 2002;<br />
Fréour et al., 2003a-b; Jacquemin et al., 2005). The main involved equations are:<br />
( I + E<br />
I<br />
: [ L<br />
i<br />
− L<br />
I ] )<br />
L<br />
I<br />
= L<br />
i<br />
:<br />
−1<br />
(1)<br />
i=<br />
r, m<br />
1<br />
−1<br />
i I I − I<br />
i I I<br />
( + L : R ) : L : ( L + L : R )<br />
I 1<br />
−1<br />
i i i<br />
β = L : L : β ΔC (2)<br />
I<br />
ΔC<br />
i=<br />
r, m<br />
i=<br />
r, m<br />
1<br />
−1<br />
i I I − I<br />
i I I<br />
[ L + L : R ] : L : [ L + L : R ]<br />
I<br />
M =<br />
−1<br />
i i<br />
: L : M<br />
(3)<br />
i=<br />
r, m<br />
i=<br />
r, m<br />
Where ΔC i is the moisture content of the studied i element of the composite structure.<br />
The superscripts r and m are considered as replacement rule for the general superscript i, in<br />
the cases that the properties of the reinforcements or those of the matrix have to be<br />
considered, respectively. Actually, the pseudo-macroscopic moisture contents ΔC r and ΔC m<br />
can be expressed as a function of the macroscopic hygroscopic load ΔC I (Loos and Springer,<br />
1981), so that the hygro-mechanical states cancels in relation (2) that can finally be rewritten<br />
as a function of the materials properties only, but at the exclusion of the ΔC i that are<br />
unexpected to appear in such an expression (Jacquemin et al., 2005). Relation (2), that is<br />
provided in the present work for predicting the macroscopic CME, is given for its enhanced<br />
readability, compared to the more rigorous state exclusive relation.<br />
In relations (1-3), the brackets < > stand for volume weighted averages (that in fact<br />
replace volume integrals that would require Finite Elements Methods instead). Empirically, as<br />
stated by Hill (Hill, 1952), arithmetic or geometric averages suggest themselves as good<br />
approximations. On the one hand, the geometric mean of a set of positive data is defined as<br />
the n th root of the product of all the members of the set, where n is the number of members.<br />
On the other hand, in mathematics and statistics, the arithmetic mean (or simply the mean) of<br />
a list of numbers is the sum of all the members of the list divided by the number of items in<br />
the list. For Young’s modulus, as an example, the Geometric Average Y GA of the moduli<br />
GA<br />
according to the Reuss (YR) and Voigt (YV) models is defined as Y = YR<br />
YV<br />
,<br />
whereas the corresponding Arithmetic Average Y AA AA Y Y<br />
is: Y R +<br />
= V<br />
.<br />
2
6<br />
Jacquemin Frédéric and Fréour Sylvain<br />
In statistics, given a set of data, X = { x1, x2, ..., xn} and corresponding weights, W = {<br />
w1, w2, ..., wn}, the weighted geometric (respectively, arithmetic) mean<br />
i<br />
X<br />
GA<br />
i=<br />
1,2,..., n<br />
(respectively,<br />
AA<br />
i<br />
X ) is calculated as:<br />
i=<br />
1,2,..., n<br />
⎛ n ⎞<br />
1/<br />
⎜ ∑ w ⎟<br />
GA ⎛ n ⎞ ⎜ i ⎟<br />
i ⎜ w<br />
X = ⎟ ⎝i=<br />
1 ⎠<br />
= ⎜∏<br />
x i<br />
i 1,2,..., n<br />
i ⎟<br />
⎝ i=<br />
1 ⎠<br />
∑<br />
∑ =<br />
i<br />
X<br />
AA 1 n<br />
= xi<br />
wi<br />
i=<br />
1,2,..., n n<br />
w i 1<br />
i<br />
i=<br />
1<br />
(5)<br />
Both averages have been extensively used in the field of materials science, in order to<br />
achieve various scale transition modelling over a wide range of materials. The interested<br />
reader can refer to: (Morawiec, 1989; Matthies and Humbert, 1993; Matthies et al., 1994) that<br />
can be considered as typical illustrations of works taking advantage of the geometric average<br />
for estimating the properties and mechanical states of polycrystals (nevertheless, Eshelby-<br />
Kröner self-consistent model was not involved in any of these articles), whereas the<br />
previously cited references (Kocks et al., 1998; Gloaguen et al., 2002; Fréour et al., 2003;<br />
Jacquemin et al., 2005) show applications of arithmetic averages for studying of polycrystals<br />
or composite structures.<br />
According to equations (4) and (5), the explicit writing of a volume weighted average<br />
directly depend on the averaging method chosen to perform this operation. Since the present<br />
work aims to express analytical forms involving such volume averages, it is necessary to<br />
select one average type in order to ensure a better understanding for the reader. Usually, in<br />
this field of research, the arithmetic and not the geometric volume weighted average is used.<br />
Moreover, in a recent work, the alternative geometric averages were also used for estimating<br />
the effective properties of carbon-epoxy composites (Fréour et al., to be published).<br />
Nevertheless, the obtained results were not found as satisfactory than in the previously<br />
studied cases of metallic polycrystals or metal ceramic assemblies. Actually, the very strongly<br />
heterogeneous properties presented by the constituents of carbon reinforced polymer matrix<br />
composites yields a strong underestimation of the effective properties of the composite ply<br />
predicted according to Eshelby-Kröner model involving the geometric average, by<br />
comparison to the expected (measured) properties. Thus, the geometric average should not be<br />
considered as a reliable alternate solution to the classical arithmetic average for achieving<br />
scale transition modelling of composite structures. Consequently, arithmetic averages<br />
satisfying to relation (4) only will be used in the following of this manuscript.<br />
(4)
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 7<br />
Now, in the present case, where the macroscopic behaviour is described by two separate<br />
heterogeneous inclusions only (i.e. one for the matrix and one for the reinforcements),<br />
convenient simplifications of equation (5) do occur.<br />
Actually, introducing v r and v m as the volume fractions of the ply constituents, and<br />
taking into account the classical relation on the summation over the volume fractions (i.e. v r +<br />
v m =1), equation (5) applied to the volume average of any tensor A writes:<br />
AA<br />
i<br />
A =<br />
i=<br />
r, m<br />
i<br />
r r m m<br />
A = v A + v A<br />
i=<br />
r, m<br />
In the following of the present work, the superscript AA denoting the selected volume<br />
average type will be omitted.<br />
According to equations (1-3), the effective properties expressed within Eshelby-Kröner<br />
self-consistent model involve a still undefined tensor, R I . This term is the so-called “reaction<br />
tensor” (Kocks et al., 1998). It satisfies:<br />
I<br />
I I −1<br />
−1<br />
−1<br />
( ) : ⎜<br />
⎛ I I<br />
−<br />
= − ⎟<br />
⎞ I<br />
I S S L E : E<br />
R = esh esh<br />
(7)<br />
⎝ ⎠<br />
In the very preceding equation, I stands for the fourth order identity tensor. Hill’s tensor<br />
E I , also known as Morris tensor (Morris, 1970), expresses the dependence of the reaction<br />
tensor on the morphology assumed for the matrix and its reinforcements (Hill, 1965). It can<br />
I<br />
I I I<br />
−1<br />
be expressed as a function of Eshelby’s tensor S esh , through E = Sesh<br />
: L . It has to be<br />
underlined that both Hill’s and Eshelby’s tensor components are functions of the macroscopic<br />
stiffness L I (some examples are given in Kocks et al., 1998; Mura, 1982).<br />
In the case, when ellipsoidal-shaped inclusions have to be taken into account, the<br />
following general form enables the calculation of the components of this tensor (see the<br />
works of Asaro and Barnett, 1975 or Kocks et al. 1998):<br />
⎧<br />
⎪E<br />
⎪<br />
⎨<br />
⎪<br />
⎪⎩<br />
γ<br />
I<br />
ijkl<br />
ikjl<br />
=<br />
=<br />
1<br />
4π<br />
π<br />
∫<br />
0<br />
I [ Kik<br />
() ξ ]<br />
sinθ dθ<br />
In the case of an orthotropic macroscopic symmetry, the components Kjp(ξ) were given in<br />
(Kröner, 1953):<br />
I<br />
K<br />
⎡ I<br />
L<br />
⎢<br />
11<br />
= ⎢<br />
⎢<br />
⎢⎣<br />
−1<br />
ξ<br />
( ) ( )<br />
( ) ( )<br />
( ) ( ) ⎥ ⎥⎥⎥<br />
2 I 2 I 2 I I I I<br />
ξ + +<br />
+<br />
+<br />
⎤<br />
1 L66ξ<br />
2 L55ξ3<br />
L12<br />
L66<br />
ξ1ξ<br />
2 L13<br />
L55<br />
ξ1ξ<br />
2<br />
I I I 2 I 2 I 2 I I<br />
L12<br />
+ L66<br />
ξ1ξ<br />
2 L66ξ1<br />
+ L22ξ<br />
2 + L44ξ3<br />
L23<br />
+ L44<br />
ξ 2ξ3<br />
I I I I I 2 I 2 I 2<br />
L13<br />
+ L55<br />
ξ1ξ<br />
2 L23<br />
+ L44<br />
ξ2ξ<br />
3 L55ξ1<br />
+ L44ξ<br />
2 + L33ξ3<br />
2π<br />
j<br />
∫<br />
0<br />
ξ<br />
γ<br />
l<br />
ikjl<br />
dφ<br />
⎦<br />
(6)<br />
(8)<br />
(9)
8<br />
with<br />
Jacquemin Frédéric and Fréour Sylvain<br />
sinθ cosφ sinθ sinφ cosθ<br />
ξ = ,ξ = ,ξ = (10)<br />
1 2 3<br />
a1 a2 a3<br />
where 2 a1, 2 a2, 2 a3 are the lengths of the principal axes of the ellipsoid (representing the<br />
considered inclusion) assumed to be respectively parallel to the longitudinal, transverse and<br />
normal directions of the sample reference frame.<br />
According to equations (2-3, 7), the determination of both the macroscopic coefficients of<br />
thermal and moisture expansion are somewhat straightforward, while the effective stiffness is<br />
known, because the involved expressions are explicit. On the contrary, the estimation of the<br />
macroscopic stiffness of the composite ply through (1) cannot be as easily handled.<br />
Expression (1) is implicit because it involves L I tensor in both its right and left members.<br />
Moreover, calculating the right member of equation (1) entails evaluating the reaction tensor<br />
(7) which also depends on the researched elastic stiffness, at least because of the occurrence<br />
of Hill’s tensor (or Eshelby’s tensor, if that notation is preferred) in relation (1). As a<br />
consequence, the effective elastic properties of a composite ply satisfying to Eshelby-Kröner<br />
self-consistent model constitutive relations are estimated at the end of an iterative numerical<br />
procedure. This is the main drawback of the self-consistent procedure preventing from<br />
achieving an analytical determination of the effective macroscopic thermo-hygro-elastic<br />
properties of a composite ply, in the case where this scale transition model is employed.<br />
Therefore, managing to express explicit solutions for estimating the macroscopic properties<br />
(or at least the macroscopic stiffness) requires focusing on a less intricate, less rigorous model<br />
but still providing realistic numerical values. Mori-Tanaka approach suggest itself as an<br />
appropriate candidate, for reasons that will be comprehensively explained in the next<br />
subsection.<br />
2.4. Introducing Mori-Tanaka Model as a Possible Alternate Solution to<br />
Eshelby-Kröner Model<br />
As Eshelby-Kröner self-consistent approach, Mori Tanaka estimate is a scale transition model<br />
derived from the pioneering mathematical work of Eshelby (Eshelby, 1957). Mori and Tanaka<br />
actually investigated the opportunity of extending Eshelby’s single-inclusion model (which is<br />
sometimes presented as an “infinitely dilute solution model”) to the case where the volume<br />
fraction of the ellipsoidal heterogeneous inclusion embedded in the matrix is not tending<br />
towards zero anymore, but admits a finite numerical value (Mori and Tanaka, 1973; Tanaka<br />
and Mori, 1970). Calculations show that, in many cases, the effective homogenised<br />
macroscopic properties deduced from Mori-Tanaka approximate are close to their<br />
counterparts, estimated from the previously described Eshelby-Kröner self-consistent<br />
procedure (Baptiste, 1996, Fréour et al., 2006a). Exceptions to this statement occur<br />
nevertheless in the cases where extreme heterogeneities in the constituents properties have to<br />
be accounted for. For example, handling a significant porosities volume fraction yields Mori-<br />
Tanaka estimations deviating considerably from the self-consistent corresponding
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 9<br />
calculations, according to (Benveniste, 1987). However, Mori-Tanaka approach is reported to<br />
remain reliable for treating cases similar to those aimed by the present work.<br />
It has previously been demonstrated that the effective macroscopic homogenised thermohygro-elastic<br />
properties exhibited by a composite ply, according to Mori and Tanaka<br />
approximation satisfy the following relations (Baptiste, 1996; Fréour et al., 2006a):<br />
−1<br />
−1<br />
−1<br />
I i i i<br />
i i<br />
i<br />
L = T : L : T<br />
= L : T : T<br />
(11)<br />
i=<br />
r, m<br />
i=<br />
r, m i=<br />
r, m<br />
i=<br />
r, m<br />
T<br />
I 1 ⎛<br />
−1<br />
⎞<br />
⎜ i i i i<br />
i<br />
: : : ⎟ i<br />
β = T L L T : β ΔC<br />
(12)<br />
I<br />
ΔC<br />
⎜<br />
⎟<br />
⎝<br />
i=<br />
r, m ⎠<br />
i=<br />
r, m<br />
T<br />
I ⎛<br />
−1<br />
⎞<br />
⎜ i i i i<br />
: : : ⎟ i<br />
M = T L L T : M<br />
(13)<br />
⎜<br />
⎟<br />
⎝<br />
i=<br />
r, m ⎠<br />
i=<br />
r, m<br />
The superscript T appearing in relations (12-13) denotes transposition operation.<br />
The same remarks as indicated in the preceding subsection holds for the determination of<br />
the effective macroscopic CME using relation (12). This equation can be rewritten as a<br />
function of the materials properties only, thus excluding the moisture contents.<br />
In equations (11-13), T i is the elastic strain localisation tensor, expressed for the isuperscripted<br />
phase that is considered to interact with the embedding phase (denoted by the<br />
superscript e). Actually, Mori-Tanaka model is based on a two-step scale-transition<br />
procedure. In this theory, contrary to the case of Eshelby-Kröner self-consistent model, the<br />
inclusions are not considered to be directly embedded in the effective material having the<br />
behaviour of the composite structure (and thus interacting with it). In Mori and Tanaka<br />
approximation, the n constituents of a n-phase composite ply are separated in two subclasses:<br />
one of them is designed as the embedding constituent, whereas the n-1 others are considered<br />
as inclusions of the first one. The inclusion particles are embedded in the matrix phase, itself<br />
being loaded at the infinite by the hygro-mechanical conditions applied on the composite<br />
structure. In consequence, the inclusion phase does not experience any interaction with the<br />
macroscopic scale, but with the matrix only. In consequence, Mori and Tanaka model<br />
corresponds to the direct extension of Eshelby’s single inclusion model (Eshelby, 1957) to the<br />
case that the volume fraction of inclusions does not remain infinitesimal anymore. Within<br />
Mori and Tanaka approach, this localisation tensor T i writes as follows:<br />
[ ( ) ] 1 −<br />
i i e<br />
I + E : L − L<br />
i<br />
T =<br />
(14)<br />
Contrary to the case of Eshelby-Kröner scale-transition model (refer to subsection 2.3.<br />
above), the localisation involved within Mori-Tanaka approximate does not explicitly involve
10<br />
Jacquemin Frédéric and Fréour Sylvain<br />
the macroscopic stiffness. Nevertheless, according to the already cited same subsection, the<br />
reaction tensor involved in Eshelby-Kröner model was also implicitely depending on the<br />
macroscopic stiffness through the calculation procedure entailed for estimating Hill’s tensor.<br />
Within Mori-Tanaka procedure (Benveniste, 1987; Baptiste 1996; Fréour et al., 2006a),<br />
Hill’s tensor E i expresses the dependence of the strain localization tensor on the morphology<br />
assumed for the embedding phase and the particulates it surrounds (Hill, 1965). It can be<br />
i<br />
expressed as a function of Eshelby’s tensor S esh , through:<br />
i i e<br />
−1<br />
E = Sesh<br />
: L<br />
(15)<br />
In practice, the calculation of Hill’s tensor for the embedded inclusions phase only would<br />
be necessary, since obvious simplifications of (14), leading to T I<br />
e = , occur in the case that<br />
the embedding constituent localisation tensor is considered. According to relations (14-15),<br />
the strain localization tensor T i does not involve the macroscopic stiffness tensor (or any other<br />
macroscopic property). As a consequence, contrary to Eshelby-Kröner self-consistent<br />
procedure, Mori-Tanaka approximation provides explicit relations (actually, the<br />
homogenization equations (11-13)) for estimating the researched macroscopic effective<br />
properties of a composite ply.<br />
2.5. Example of Homogenization: The Case of T300-N5208 <strong>Composite</strong>s<br />
The present subsection is focused on the application of the theoretical frameworks described<br />
in the above 2.3 and 2.4 sections to the numerical simulation of the effective properties of a<br />
typical, high-strength, fiber-reinforced composite made up of T300 carbon fibers and N5208<br />
epoxy resin. The choice of such a material is justified because of the strong heterogeneities of<br />
the hygro-thermo-elastic properties of its constituents (actually, the numerical deviation<br />
occurring among the macroscopic properties of composites determined through various scale<br />
transition relations rises with this factor, see Jacquemin et al, 2005; Herakovich, 1998). Table<br />
1 accounts for the pseudo-macroscopic properties reported in the literature for these<br />
constituents. The comparison between the results obtained through the two, considered in the<br />
present work alternate scale transition framework of Mori-Tanaka model are displayed on<br />
figure 1, for:<br />
- the longitudinal and transverse Young’s moduli<br />
- Coulomb’s moduli<br />
I<br />
G 12 ,<br />
I<br />
G 23 ,<br />
I I<br />
- the coefficients of thermal expansion M 11,<br />
M 22<br />
I I<br />
- the coefficients of moisture expansion β 11,<br />
β 22 .<br />
I<br />
Y 11 ,<br />
I<br />
Y 22 ,
T300 fibers<br />
(Soden et al.,<br />
1998;<br />
Agbossou and<br />
Pastor, 1997)<br />
N5208 epoxy<br />
matrix (Tsai,<br />
1987;<br />
Agbossou and<br />
Pastor, 1997)<br />
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 11<br />
Table 1. Hygro-thermo-mechanical properties of T300/5208 constituents.<br />
ρ<br />
[g/cm 3 ]<br />
Y 1<br />
[GPa]<br />
Y 2, Y3<br />
[GPa]<br />
ν12<br />
ν<br />
13<br />
G23<br />
[GPa]<br />
G 12<br />
[GPa]<br />
M11<br />
[10 -6 /K]<br />
M22, M 33<br />
[10 -6 /K]<br />
11 22 β , β<br />
1200 230 15 0.2 7 15 -1.5 27 0<br />
1867 4.5 4.5 0.4 6.4 6.4 60 60 0.6<br />
The calculations were achieved assuming that the reinforcements exhibit fiber-like<br />
morphology with an infinite length axis parallel to the longitudinal direction of the ply. For<br />
the determination of the CME, a perfect adhesion between the carbon fibers and the resin was<br />
assumed. Moreover, it also was assumed that the fibers do not absorb any moisture. Thus, the<br />
ratio between the pseudo-macroscopic and the macroscopic moisture contents is deduced<br />
from the expression given in (Loos and Springer, 1981):<br />
m<br />
I<br />
I<br />
ρ<br />
m m<br />
ΔC<br />
= (16)<br />
ΔC v ρ<br />
where ρ stands for the densities. The macroscopic density can be deduced form the classical<br />
rule of mixture:<br />
I m m r r<br />
= v ρ v ρ<br />
(17)<br />
ρ +<br />
The equations required for achieving Mori-Tanaka estimations involve relations (8-17).<br />
For the purpose of the strain localization, the embedding constituent was considered to be the<br />
epoxy matrix, whatever the considered volume fraction of reinforcements (thus, the<br />
e m<br />
transformation rule L = L was considered to be valid in any case). Figure 1 also reports<br />
the numerical results obtained through Kröner-Eshelby Self-Consistent model (1-3, 6-10, 16-<br />
17), in the same conditions (identical inclusion morphology and constituents properties as for<br />
Mori-Tanaka computations).<br />
β<br />
33
12<br />
Macroscopic stiffness<br />
component [GPa]<br />
CME component<br />
1,4<br />
1,2<br />
0,8<br />
0,6<br />
0,4<br />
0,2<br />
0, 25<br />
0,2<br />
0, 15<br />
0,1<br />
0, 05<br />
250<br />
200<br />
150<br />
100<br />
1<br />
50<br />
0<br />
Jacquemin Frédéric and Fréour Sylvain<br />
0 0,25 0,5 0,75 1<br />
matrix volume fraction<br />
0<br />
0 0,25 0,5 0,75 1<br />
0<br />
I<br />
Y22<br />
I<br />
Y11<br />
I<br />
β11<br />
I<br />
β22<br />
epoxy volume fraction<br />
Macroscopic stiffness component<br />
[GPa]<br />
CTE component [10 -6 K -1 ]<br />
80<br />
60<br />
40<br />
20<br />
-20<br />
16<br />
14<br />
12<br />
10<br />
8<br />
6<br />
4<br />
2<br />
0<br />
0 0,25 0,5 0,75 1<br />
epoxy volume fraction<br />
0<br />
0 0,25 0,5 0,75 1<br />
epoxy volume fraction<br />
Longitudinal (KESC) Transverse (KESC)<br />
Longitudinal (Mori-Tanaka) Transverse (Mori-Tanaka)<br />
0 0,25 0,5 0,75 1<br />
matrix volume fraction<br />
Figure 1. Macroscopic effective hygro-thermo-mechanical properties of T300/N5208 plies, estimated as<br />
a function of the epoxy volume fraction, through scale transition homogenisation procedures.<br />
Comparison between Mori-Tanaka approximate and Kröner-Eshelby self-consistent model.<br />
Figure 1 shows the following interesting results:<br />
1) In pure elasticity, both the investigated scale transition methods manage to reproduce<br />
the expected mechanical behaviour of the composite ply: the material is stiffer in the<br />
longitudinal direction than in the transverse direction. Moreover, the bounds are satisfying:<br />
the properties of the single constituents are correctly obtained for those of the composite ply<br />
in the cases where the epoxy volume fraction is either taken equal to v m =0 (transversely<br />
isotropic elastic properties of T300 fibers) or v m =1 (isotropic elastic properties of N5208<br />
resin).<br />
2) The curves drawn for each checked elastic constant are almost superposed, except for<br />
I<br />
Coulomb’s modulus G 12 . Thus Mori-Tanaka model constitutes a rather reliable alternate<br />
homogenization procedure to Eshelby-Kröner rigorous solution for estimating the<br />
macroscopic elastic properties of typical carbon-epoxies.<br />
I<br />
G 23<br />
I<br />
M11<br />
I<br />
G12<br />
I<br />
M22
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 13<br />
3) Kröner-Eshelby self-consistent model and Mori-Tanaka approach both also do manage<br />
to achieve a realistic prediction of the macroscopic coefficients of thermal expansion.<br />
Especially, the expected boundary values are attained when the conditions v m =1 (isotropic<br />
CTE of N5208 resin) or v m =0 (transversely isotropic thermal properties of T300 fibers) are<br />
taken into account.<br />
4) Mori-Tanaka approximate correctly reproduces the expected macroscopic coefficients<br />
of moisture expansion in the longitudinal direction. In the transverse direction, however,<br />
Mori-Tanaka model properly follows Eshelby Kröner model estimates while the epoxy<br />
m<br />
volume fraction is higher than 0.5. In the range 0 ≤ v ≤ 0.5,<br />
discrepancies occur between<br />
two considered scale transition models. In the case that the considered strain localization<br />
assumes the epoxy as the embedding constituent within Mori-Tanaka approximate, the<br />
relative error on I β 22 induced by this localization procedure, compared to Kröner-Eshelby<br />
reference values remains weaker than 9%, and falls below 6% in the range of epoxy volume<br />
fraction that is typical for designing composites structures for engineering applications<br />
m<br />
(0.3 ≤ v ≤ 0.7) .<br />
5) In the range of the epoxy volume fraction, that is typical for designing composites<br />
m<br />
structures for engineering applications ( i.e. 0.3 ≤ v ≤ 0.7)<br />
, according to the above<br />
discussed results 3) and 4), Mori-Tanaka model can be employed as an alternative to Eshelby-<br />
Kröner self-consistent model for estimating the effective macroscopic hygro-thermomechanical<br />
properties of composite plies.<br />
The above listed elements 1) to 5) finally indicate that the effective macroscopic thermohygro-elastic<br />
properties of composite plies can be estimated in a reliable fashion using Mori-<br />
Tanaka approximate, assuming the epoxy as the embedding constituent, instead of the more<br />
rigorous Kröner-Eshelby model. This statement is true while the epoxy volume fraction<br />
remains higher than 40%. Beyond this boundary value, some significant relative error (less<br />
than 10%) may be expected to occur in the estimated transverse CME.<br />
The results, obtained in the present section, will be used in the following as input<br />
parameters for estimating the mechanical states experienced at macroscopic but at<br />
microscopic scale also in composite structures submitted to various loads (the interested<br />
reader should refer to section 4 for details).<br />
3. Inverse Scale Transition Modelling for the Identification of the<br />
Hygro-Thermo-Elastic Properties of One Constituent of a<br />
<strong>Composite</strong> Ply<br />
3.1. Introduction<br />
The precise knowledge of the pseudo-macroscopic properties of each constituent of a<br />
composite structure is required in order to achieve the prediction of its behavior (and<br />
especially its mechanical states) through scale transition models. Nevertheless, the pseudomacroscopic<br />
stiffness, coefficients of thermal expansion and moisture expansion of the matrix
14<br />
Jacquemin Frédéric and Fréour Sylvain<br />
and its reinforcements are not always fully available in the already published literature. The<br />
practical determination of the hygro-thermo-mechanical properties of composite materials are<br />
most of the time achieved on unidirectionnaly reinforced composites and unreinforced<br />
matrices (Bowles et al., 1981; Dyer et al., 1992; Ferreira et al., 2006a; Ferreira et al., 2006b;<br />
Herakovich, 1998; Sims et al., 1977). In spite of the existence of several articles dedicated to<br />
the characterization of the properties of the isolated reinforcements (Tsai and Daniel, 1994;<br />
DiCarlo, 1986; Tsai and Chiang, 2000), the practical achieving of this task remains difficult<br />
to handle, and the available published data for typical reinforcing particulates employed in<br />
composite design are still very limited. As a consequence, the properties of the single<br />
reinforcements exhibiting extreme morphologies (such as fibers), are not often known from<br />
direct experiment, but more usually they are deduced from the knowledge of the properties of<br />
the pure matrices and those of the composite ply (which both are easier to determine), through<br />
appropriate calculation procedures. The question of determining the properties of some<br />
constituents of heterogeneous materials has been extensively addressed in the field of<br />
materials science, especially for studying complex polycrystalline metallic alloys (like<br />
titanium alloys, cf. Fréour et al., 2002 ; 2005b ; 2006b) or metal matrix composites (typically<br />
Aluminum-Silicon Carbide composites cf. Fréour et al., 2003a ; 2003b or iron oxides from<br />
the inner core of the Earth, cf. Matthies et al., 2001, for instance). The required calculation<br />
methods involved in order to achieve such a goal are either based on Finite Element Analysis<br />
(Han et al., 1995) or on the inversion of scale transition homogenization procedures similar to<br />
those already presented in section 2 of the present paper. It was shown in previous works that<br />
it was actually possible to identify the properties of one constituent of a heterogeneous<br />
material from available (measured) macroscopic quantities through inverse scale transition<br />
models. Such identification methods were successfully used in the field of metal-matrix<br />
composites for the determination of the average elastic (Freour et al., 2002) and thermal<br />
(Freour et al., 2006b) properties of the β-phase of (α+β) titanium alloys. The procedure was<br />
recently extended to the study of the anisotropic elastic properties of the single-crystal of the<br />
β-phase of (α+β) titanium alloys on the basis of the interpretation of X-Ray Diffraction strain<br />
measurements performed on heterogeneous polycrystalline samples in (Freour et al., 2005b).<br />
The question of determining the temperature dependent coefficients of thermal expansion of<br />
silicon carbide was handled using a similar approach from measurements performed on<br />
aluminum – silicon carbide metal matrix composites in (Freour et al., 2003a; Freour et al.,<br />
2003b).Numerical inversion of both Mori-Tanaka and Eshelby-Kröner self-consistent models<br />
will be developed and discussed here.<br />
3.2. Estimating Constituents Properties from Eshelby-Kröner Self-consistent<br />
or Mori-Tanaka Inverse Scale Transition Models<br />
3.2.1. Application of Eshelby-Kröner Self-consistent Framework to the<br />
Identification of the Pseudo-macroscopic Properties of one Constituent<br />
Embedded in a Two-Constituents <strong>Composite</strong> Material<br />
The pseudomacroscopic stiffness tensor of the reinforcements can be deduced from the<br />
inversion of the Eshelby-Kröner main homogenization form over the constituents elastic<br />
properties (1) as follows :
1 I<br />
L = L :<br />
r<br />
v<br />
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 15<br />
m<br />
I r I v m I m I −1<br />
I r I<br />
[ E : ( L − L ) + I]<br />
− L : [ E : ( L − L ) + I]<br />
: [ E : ( L − L ) + I]<br />
r<br />
v<br />
The application of this equation implies that both the macroscopic stiffness and the<br />
pseudomacroscopic mechanical behaviour of the matrix is perfectly determined. The elastic<br />
stiffness of the matrix constituting the composite ply will be assumed to be identical to the<br />
elastic stiffness of the pure single matrix, deduced in practice from measurements performed<br />
on bulk samples made up of pure matrix. It was demonstrated in (Fréour et al., 2002) that this<br />
assumption was not leading to significant errors in the case that polycrystalline multi-phase<br />
samples were considered. The similarities existing between multi-phase polycrystals and<br />
polymer based composites suggest that this assumption should be suitable in the present<br />
context, at least when scale factors do not occur. Nevertheless, in the case that significant<br />
edge effects, due for instance to a reduced thickness of the matrix layer constituting the<br />
composite ply, might be expected to occur, the identification of the ply embedded matrix<br />
elastic properties to those of the corresponding bulk material would not systematically be<br />
appropriate. Consequently, the application of inverse form (18) given above could lead to an<br />
erroneous estimation of the reinforcements elastic stiffness. Moreover, identification based on<br />
such inverse homogenization methods are sensitive to both the precise knowledge of the<br />
constituents volume fractions (i.e. v m and v r ) and to the presence of porosities (which lowers<br />
the effective stiffness L I of the composite ply).<br />
An expression, analogous to above-relation (18) can be found for the elastic stiffness of<br />
the matrix, through the following replacement rules over the superscripts/subscripts:<br />
m → r, r → m . Nevertheless, the situation, where the properties of the reinforcements are<br />
known, when those of the matrix are unknown is highly improbable.<br />
The pseudomacroscopic coefficients of moisture expansion of the matrix can be deduced<br />
from the inversion of the homogenization form (2) as follows :<br />
where G m writes :<br />
( ) m I I m<br />
L + L : R G<br />
(18)<br />
m<br />
m 1 −1<br />
β = L :<br />
:<br />
(19)<br />
m m<br />
v ΔC<br />
i I I −1<br />
I I r r I I −1<br />
r r r<br />
( L + L : R ) : L : β − v ( L + L : R ) : L : ΔC<br />
m I<br />
G = ΔC<br />
β (20)<br />
i=<br />
r, m<br />
An expression, analogous to above-relation (19) can also be found for the coefficients of<br />
moisture expansion of a permeable reinforcement type, through the following replacement<br />
rules over the superscripts/subscripts: m → r, r → m .<br />
In the particular case, where impermeable reinforcements are present in the composite<br />
structure, the coefficients of moisture expansion of the matrix simplifies as follows (an<br />
extensive study of this very question was achieved in Jacquemin et al., 2005):
16<br />
Jacquemin Frédéric and Fréour Sylvain<br />
m I I i I I −1<br />
I I<br />
( L + L : R ) : ( L + L : R ) : L<br />
I<br />
m ΔC m−1<br />
β = L :<br />
: β (21)<br />
m m<br />
v ΔC<br />
i=<br />
r, m<br />
The pseudomacroscopic coefficients of thermal expansion of the matrix can be deduced<br />
from the inversion of the homogenization form (3) as follows:<br />
( ) ( ) ( ) ⎥ ⎥<br />
⎡<br />
⎤<br />
m I I<br />
−<br />
−<br />
+ ⎢ i I I 1 I<br />
I r r I I 1 r<br />
L L : R : L + L : R : L : M − v L + L : R : L M<br />
m m−1<br />
r (22)<br />
M = L :<br />
:<br />
⎢<br />
⎣<br />
i=<br />
r, m<br />
⎦<br />
Form (22) can be easily rewritten for expressing the coefficients of thermal expansion of<br />
the reinforcements, using the same replacement rules over the superscripts/subscripts:<br />
m → r, r → m , than for the previous cases.<br />
3.2.2. Application of Mori-Tanaka Estimates to the Identification of the Pseudomacroscopic<br />
Properties of one Constituent Embedded in a Two-Constituents<br />
<strong>Composite</strong> Material<br />
3.2.2.1. Inverse Mori-Tanaka Elastic Model<br />
In the present work, it is be considered, that the reinforcements are surrounded by the matrix,<br />
thus, T m =I and (11) develops as follows:<br />
I<br />
( ) ( ) 1<br />
m m r r r m r r<br />
−<br />
v L + v L : T : v I + v : T<br />
L =<br />
(23)<br />
Thus, from (11) two alternate equations are obtained for identifying the pseudomacroscopic<br />
stiffness of the composite ply constituents:<br />
• On the first hand, the elastic properties of the matrix satisfies<br />
I r r<br />
( L - L ) : T<br />
m<br />
m I 1-<br />
v<br />
L = L +<br />
(25)<br />
m<br />
v<br />
Equation (25) is an implicit equation since both its left and right hand sides involve the<br />
researched stiffness tensor L m .<br />
• whereas, on the second hand, the elastic stiffness of the reinforcements respects<br />
I m r<br />
−1<br />
( L - L ) : T<br />
m<br />
r I v<br />
L = L +<br />
(26)<br />
m<br />
1-<br />
v
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 17<br />
For the same reasons as above (i.e. comments about equation (25)), expression (26) is an<br />
implicit relation. As a consequence, the need of an inverse modelling for achieving the<br />
identification of the elastic properties exhibited by any one constituent of a composite ply<br />
through Mori-Tanaka scale-transition approximate yields the loss of the main advantage of<br />
this very model over the more rigorous Eshelby-Kröner self-consistent approach: the<br />
opportunity to express analytical explicit relations instead of having to perform successive<br />
numerical calculations for solving implicit equations. Moreover, the general remarks about<br />
the sensitivity of identification methods to certain factors, expressed in subsection 3.2.1 are<br />
valid in the present context also.<br />
3.2.2.2. Inverse Mori-Tanaka Model for Identifying Coefficients of Moisture of Thermal<br />
Expansion<br />
Following the same line of reasoning as above, in the purely elastic case, one can inverse<br />
relation (12) in order to express the coefficients of moisture expansion of a constituent<br />
embedded in a composite ply according to Mori-Tanaka estimates, or its coefficients of<br />
thermal expansion, from the homogenization relation (13). In the case of the pure matrix, one<br />
gets:<br />
m 1 m<br />
−1<br />
⎡ i i<br />
I r r r<br />
M = L : ⎢ L : T : M − v L : T : M<br />
m<br />
v ⎢⎣<br />
i=<br />
r, m<br />
m 1 m<br />
−1<br />
⎡ i i<br />
I I r r r r r ⎤<br />
β = L : ⎢ L : T : β ΔC − v L : T : β ΔC ⎥ (28)<br />
m m<br />
v ΔC ⎢⎣<br />
i=<br />
r, m<br />
⎥⎦<br />
This last relation (valid for the general case of a possibly permeable reinforcement type)<br />
yields to the following simplified form if impermeable reinforcements are considered:<br />
r<br />
⎤<br />
⎥<br />
⎥⎦<br />
(27)<br />
m 1 m<br />
−1<br />
i i<br />
I I<br />
β = L : L : T : β ΔC<br />
(29)<br />
m m<br />
v ΔC<br />
i=<br />
r, m<br />
Due to the localization procedure which does not treat in an equivalent way the<br />
embedding matrix and the embedded inclusions (reinforcements) in the point of view of<br />
Mori-Tanaka scale-transition approach, the inverse forms satisfied by the coefficients of<br />
thermal expansion and coefficients of moisture expansion of the reinforcements are not<br />
anymore deduced from the above-relations established for the matrix through simple<br />
replacement rules. Actually, unlike the inverse forms obtained according to Eshelby-Kröner<br />
self-consistent model, Mori-Tanaka model yields non-equivalent inverse forms for the matrix<br />
one the one hand and for the reinforcements, on the second hand. The expressions, required<br />
for identifying the thermal or hygroscopic properties of reinforcements within Mori-Tanaka<br />
model are:
18<br />
Jacquemin Frédéric and Fréour Sylvain<br />
r 1 r 1 r<br />
−1<br />
⎡ i i<br />
I m m<br />
M = T : L : ⎢ L : T : M − v L : M<br />
r<br />
v<br />
⎢⎣<br />
i=<br />
r, m<br />
− m<br />
r 1 r −1 r<br />
−1<br />
⎡ i i<br />
I I m m m m ⎤<br />
β = T : L : ⎢ L : T : β ΔC − v L : β ΔC ⎥ (31)<br />
r r<br />
v ΔC<br />
⎢⎣<br />
i=<br />
r, m<br />
⎥⎦<br />
3.3. Examples of Properties Identification in <strong>Composite</strong> Structures Using<br />
Inverse Scale Transition Methods<br />
3.3.1. Determination of Reinforcing Fibers Elastic Properties<br />
The literature often provides elastic properties of carbon-fiber reinforced epoxies (see for<br />
instance Sai Ram and Sinha, 1991), that can be used in order to apply inverse scale transition<br />
model and thus identify the properties of the reinforcing fibers, as an example. Table 2 of the<br />
present work summarizes the previously published data for an unidirectional composite<br />
designed for aeronautic applications, containing a volume fraction v r =0.60 of reinforcing<br />
fibers. In order to achieve the calculations, according to relations (18) or (26) depending on<br />
whether Eshelby-Kröner model or Mori-Tanaka approximation, input values are required for<br />
the pseudo-macroscopic properties of the epoxy matrix constituting the composite ply. The<br />
elastic constants considered for this purpose are listed in Table 3 (from Herakovich, 1998).<br />
Both the above-cited inverse scale transition methods have been applied. The obtained results<br />
are provided in Table 4, where they are compared to typical values, reported in the literature,<br />
for high-strength reinforcing fibers (Herakovich, 1998). It is shown that a very good<br />
agreement between the two inverse models is obtained. Moreover, the calculated values are<br />
similar to those expected for typical reinforcements according to the literature. Nevertheless,<br />
some discrepancies between the identified moduli do exist, especially for<br />
r G 12 (that<br />
r<br />
corresponds to L 55 stiffness component). Actually, the value deduced for this component<br />
through Mori-Tanaka inverse model deviates from both the expected properties and the<br />
estimations of Eshelby-Kröner model. This deviation, occurring for this very component, is<br />
Table 2. Macroscopic elastic moduli (from the literature) and stiffness tensor<br />
components (calculated) considered for the composite ply at ΔC 0<br />
I = % and T I = 300<br />
K, according to (Sai Ram and Sinha, 1991).<br />
Elastic moduli<br />
Stiffness tensor<br />
components<br />
I<br />
1<br />
I<br />
Y 2 [GPa]<br />
I<br />
ν 12 [1]<br />
I<br />
G 12 [GPa]<br />
I<br />
23<br />
130 9.5 0.3 6.0 3.0<br />
Y [GPa]<br />
⎤<br />
⎥<br />
⎥⎦<br />
G [GPa]<br />
I<br />
11<br />
I<br />
L 22 [GPa]<br />
I<br />
L 12 [GPa]<br />
I<br />
L 44 [GPa]<br />
I<br />
L 55 [GPa]<br />
134.2 14.8 7.1 6.0 3.0<br />
L [GPa]<br />
(30)
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 19<br />
obviously directly related to the discrepancies previously underlined in subsection 2.5 where<br />
the question of comparing the homogenization relations of the two scale transition methods<br />
presented in this paper, was investigated.<br />
Table 3. Pseudomacroscopic elastic moduli and stiffness tensor components assumed for<br />
the epoxy matrix of the composite plies at ΔC 0<br />
I = % and T I = 300 K, according to<br />
(Herakovich, 1998).<br />
Elastic moduli<br />
Stiffness tensor<br />
components<br />
m<br />
1<br />
m<br />
Y 2 [GPa]<br />
m<br />
ν 12 [1]<br />
m<br />
G 12 [GPa]<br />
m<br />
23<br />
5.35 5.35 0.350 1.98 1.98<br />
Y [GPa]<br />
G [GPa]<br />
m<br />
11<br />
m<br />
L 22 [GPa]<br />
m<br />
L 12 [GPa]<br />
m<br />
L 44 [GPa]<br />
m<br />
L 55 [GPa]<br />
8.62 8.62 4.66 1.98 1.98<br />
L [GPa]<br />
Table 4. Pseudomacroscopic elastic moduli and stiffness tensor components identified<br />
for the carbon fiber reinforcing the composite plies at ΔC 0<br />
I = % and T I = 300 K,<br />
according to either Mori-Tanaka estimates, or Eshelby-Kröner self-consistent model.<br />
Comparison with the corresponding properties exhibited in practice by typical highstrength<br />
carbon fibers, according to (Herakovich, 1998).<br />
Elastic moduli<br />
r<br />
Y 1 [GPa]<br />
r<br />
Y 2 [GPa]<br />
r<br />
ν 12 [1]<br />
r<br />
G 23 [GPa]<br />
r<br />
G 12 [GPa]<br />
Mori-Tanaka estimate 213.1 13.7 0.27 4.1 22.7<br />
Eshelby-Kröner model 213.2 13.3 0.27 4.0 12.1<br />
Typical expected<br />
properties<br />
Stiffness tensor<br />
components<br />
232 15 0.279 5.0 15<br />
r<br />
L 11[GPa]<br />
r<br />
L 22 [GPa]<br />
r<br />
L 12 [GPa]<br />
r<br />
L 44 [GPa]<br />
r<br />
L 55 [GPa]<br />
Mori-Tanaka estimate 219.2 24.9 11.2 4.1 22.7<br />
Eshelby-Kröner model 219.2 23.9 10.8 4.0 12.1<br />
Typical expected<br />
properties<br />
236.7 20.1 8.4 5.02 15<br />
3.3.2. Determination of AS4/3501-6 Matrix Coefficients of Moisture Expansion<br />
Macroscopic values of the Coefficients of Moisture Expansion are sometimes available,<br />
contrary to the corresponding pure epoxy resin CME. Simulations were performed in the cas<br />
e of an AS4/3501-6 composite, with a reinforcing fiber volume fraction v r =0.60. The<br />
calculations were achieved using the elastic properties given in Table 5, and the macroscopic<br />
coefficients of moisture expansion listed in Table 6. The same table summarizes the results<br />
obtained with both inverse Eshelby-Kröner self-consistent model (21) and Mori-Tanaka
20<br />
Jacquemin Frédéric and Fréour Sylvain<br />
estimates (29) assuming a moisture content<br />
ΔC<br />
= 3.<br />
125 (the ratio between composite and<br />
I<br />
ΔC<br />
resin densities being 1.25 in this material, the moisture content ratio assumed in the present<br />
study corresponds to the maximum expected value), in the case that impermeable<br />
reinforcements are considered. According to Table 6, a very good agreement is obtained<br />
between the two inverse models. This result is compatible with the homogenisation<br />
calculation previously achieved in subsection 2.5: for such a volume fraction of<br />
reinforcements, Eshelby-Kröner and Mori-Tanaka models provide identical macroscopic<br />
coefficients of moisture expansion from the pseudomacroscopic data. As a consequence, the<br />
corresponding inverse forms (21) and (29) yields the same estimation for the<br />
pseudomacroscopic CME of the matrix constituting the composite ply.<br />
Table 5. Macroscopic and pseudo-macroscopic mechanical elastic properties of<br />
AS4/3501-6 constituents.<br />
m<br />
E 1 [GPa] E 2, E3<br />
[GPa] ν 12 , ν13<br />
ν23 G 12 [GPa]<br />
AS4 fibers<br />
(Soden et al., 1998) 225 15 0.2 0.40 15<br />
3501-6 epoxy matrix (Soden et<br />
al., 1998)<br />
AS4/3501-6<br />
(KESC homogenisation)<br />
4.2 4.2 0.34 0.34 1.567<br />
135.2 9.2 0.25 0.36 5.2<br />
Table 6. Macroscopic and pseudomacroscopic (3501-6 matrix only) coefficients of<br />
moisture expansion of AS4/3501-6 composite. The pseudomacroscopic values results<br />
from the two inverse scale transition models described in the present work.<br />
Moisture expansion coefficient β 11 β 22, β33<br />
AS4/3501-6 (Daniel and Ishai, 1994) 0.01 0.2<br />
3501-6 epoxy from Eshelby-Kröner self-consistent inverse model 0.148 0.148<br />
3501-6 epoxy from Mori-Tanaka inverse model 0.148 0.148<br />
4. From the Numerical Model to Analytical Solutions for<br />
Estimating the Pseudo-macroscopic Mechanical States<br />
4.1. Introduction<br />
It was extensively discussed in previously published works (the interested reader can, for<br />
instance refer to Benveniste, 1987 and Fréour et al., 2006a, where the question is addressed),<br />
that Mori and Tanaka constitutive assumptions were not suitable for a reliable estimation of<br />
the localization of the macroscopic mechanical states within the constituents of typical
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 21<br />
composites conceived for engineering applications, which often present a significant volume<br />
fraction of reinforcements. As a consequence, only Eshelby-Kröner approach will be<br />
considered in the present section.<br />
4.2. Numerical SC Model Extended to a Thermo-Hygro-Elastic Load<br />
Within Kröner and Eshelby self-consistent framework, the hygrothermal dilatation generated<br />
by a moisture content increment ΔC i is treated as a transformation strain exactly like the<br />
thermal dilatation occurring after a temperature increment ΔT i (that last case was extensively<br />
discussed in the literature, see for example Kocks et al., 1998). Thus, the pseudo-macroscopic<br />
stresses σ i in the considered constituent (i.e. i=r or i=m) are given by:<br />
i<br />
i<br />
i i i i i<br />
( ε − M ΔT ΔC )<br />
σ = L :<br />
− β<br />
(32)<br />
Where, ε stands for the strain tensor. In general case, the moisture content differs at<br />
macroscopic scale and pseudo-macroscopic scale, contrary to the temperature. Actually, the<br />
reinforcements generally do not absorb moisture. In consequence, the mass of water<br />
contained by the composite is: either found in the matrix, locally trapped in porosities<br />
(Mensitieri et al., 1995) or located where fiber debonding occurs.<br />
Replacing the superscripts i by I in (32) leads to the stress-strain relation that holds at<br />
macroscopic scale.<br />
I<br />
I<br />
I I I I I<br />
( ε − M ΔT ΔC )<br />
σ = L :<br />
− β<br />
(33)<br />
The so-called “scale-transition relation” enabling to determine the local stresses and<br />
strains from the macroscopic mechanical states was demonstrated in a fundamental work,<br />
starting from the assumption that the elementary inclusions (here the matrix and the fiber)<br />
have ellipsoidal shapes (Eshelby, 1957):<br />
i<br />
I<br />
i I ( ε )<br />
I I<br />
: R<br />
(34)<br />
σ − σ = −L<br />
: − ε<br />
Actually, (34) is not very useful, because both the unknown pseudo-macroscopic stresses<br />
and strains appear. Nevertheless, combining (32-34) enables to find the following expression<br />
for the pseudo-macroscopic strain (the demonstration is available in Jacquemin et al., 2005<br />
and Fréour et al., 2003b):<br />
i I I<br />
−1<br />
I I I I i i I I i i i I I I<br />
( L + L : R ) : [ ( L + L : R ) .. ε + ( L : M − L : M ) ΔT + L : β ΔC L : β ΔC ]<br />
i<br />
ε =<br />
−<br />
In relation (35), the classical replacement rule ΔT i =ΔT Ι =ΔT was introduced (i.e. the<br />
temperature field is considered to be uniform within the considered ply).<br />
(35)
22<br />
Jacquemin Frédéric and Fréour Sylvain<br />
Moreover, it was established in (Hill, 1967), that the self-consistent model was<br />
compatible with the following volume averages on both pseudo-macroscopic stresses and<br />
strains:<br />
σ<br />
ε<br />
i<br />
i<br />
i=<br />
r, m<br />
i=<br />
r, m<br />
= σ<br />
For a given applied macroscopic thermo-hygro-elastic load {σ I , ΔC I, ΔT} one can easily<br />
determine ε I through (33), provided that the effective elastic behaviour L I of the ply has been<br />
calculated using either the homogenization procedure corresponding to Eshelby-Kröner<br />
model or the corresponding Mori-Tanaka alternate solution (see previous developments<br />
provided in section 2 above). Then, the pseudo-macroscopic strains are determined through<br />
(35).<br />
4.3. Analytical Expression for Calculating the Mechanical States Experienced<br />
by the Constituents of Fiber-Reinforced <strong>Composite</strong>s According to<br />
Eshelby-Kröner Model<br />
The main impediment requiring to be overcome in order to achieve closed-forms from<br />
relation (35) is the determination of Morris’ tensor E I . Actually, according to the integrals<br />
appearing in relation (8), this tensor will admit only numerical solutions in most cases.<br />
However, some analytical forms for Morris’ tensor are actually available in the literature;<br />
the interested reader can for instance refer to (of Mura, 1982; Kocks et al., 1998; or Qiu and<br />
Weng 1991). Nevertheless, these forms were established considering either spherical, discshaped<br />
of fiber-shaped inclusions embedded in an ideally isotropic macroscopic medium, that<br />
is incompatible with the strong elastic anisotropy exhibited by fiber-reinforced composites at<br />
macroscopic scale (Tsai and Hahn, 1987).<br />
In the case of carbon-epoxy composites, a transversely isotropic macroscopic behaviour<br />
being coherent with fiber shape is actually expected (and predicted by the numerical<br />
computations). Assuming that the longitudinal (subscripted 1) axis is parallel to fiber axis, one<br />
obtains the following conditions for the semi-lengths of the microstructure representative<br />
ellipsoid: a1→∞, a2=a3. Moreover, the macroscopic elastic stiffness should satisfy :<br />
I I I I I I<br />
L11 ≠ L12<br />
≠ L22<br />
≠ L23<br />
≠ L44<br />
≠ L55<br />
. Now, it is obvious, that these additional<br />
hypotheses lead to drastic simplifications of Morris’ tensor (8), in the case that fiber<br />
morphology is considered for the reinforcements. The line of reasoning required to achieve<br />
the writing of analytical expressions for Morris’ tensor is extensively presented in (Welzel et<br />
= ε<br />
al., 2005; Fréour et al., 2005). Actually, one obtains (in contracted notation i.e,<br />
components are given here):<br />
I<br />
I<br />
(36)<br />
I<br />
E ij
I<br />
E<br />
⎡0<br />
⎢<br />
⎢0<br />
⎢<br />
⎢<br />
⎢<br />
⎢0<br />
⎢<br />
= ⎢<br />
⎢0<br />
⎢<br />
⎢<br />
⎢0<br />
⎢<br />
⎢<br />
⎢0<br />
⎢<br />
⎣<br />
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 23<br />
0<br />
3 1<br />
+<br />
I I I<br />
8L22<br />
4L22<br />
− 4L23<br />
I I<br />
L22<br />
+ L23<br />
2<br />
I I I<br />
8L22L<br />
23 − 8L22<br />
0<br />
0<br />
0<br />
0<br />
I I<br />
L22<br />
+ L23<br />
2<br />
I I I<br />
8L22L<br />
23 − 8L22<br />
3 1<br />
+<br />
I I I<br />
8L22<br />
4L22<br />
− 4L23<br />
0<br />
0<br />
0<br />
0<br />
0<br />
0<br />
1 1<br />
+<br />
I I I<br />
8L22<br />
4L22<br />
− 4L23<br />
0<br />
0<br />
0<br />
0<br />
0<br />
0<br />
1<br />
I<br />
8L55<br />
0<br />
0 ⎤<br />
⎥<br />
0 ⎥<br />
⎥<br />
⎥<br />
⎥<br />
0 ⎥<br />
⎥<br />
⎥<br />
0 ⎥<br />
⎥<br />
⎥<br />
0 ⎥<br />
⎥<br />
⎥<br />
1 ⎥<br />
I<br />
8L ⎥<br />
55 ⎦<br />
In fact, the epoxy matrix is usually isotropic, so that three components only have to be<br />
considered for its elastic constants: m m m<br />
L 11,<br />
L12<br />
and L44<br />
. One moisture expansion coefficient is<br />
sufficient to describe the hygroscopic behaviour of the matrix: m β 11.<br />
In the case of the carbon fibers, a transverse isotropy is generally observed. Thus, the<br />
corresponding elasticity constants depend on the following components:<br />
r r r r r r<br />
L 11,<br />
L12,<br />
L22,<br />
L23,<br />
L44,<br />
and L55<br />
. Moreover, since the carbon fiber does not absorb water,<br />
r r<br />
its CME β 11 and β22<br />
will not be involved in the mechanical states determination.<br />
Introducing these additional assumptions in (35), and taking into account the form (37)<br />
obtained for Morris’ tensor, one can deduce the following strain tensors for both the matrix<br />
and the fibers:<br />
⎡ i i i<br />
ε<br />
⎤<br />
⎢<br />
11 ε12<br />
ε13<br />
i i i i ⎥<br />
ε = ⎢ε12<br />
ε22<br />
ε23⎥<br />
(38)<br />
⎢ i i i ⎥<br />
⎢<br />
ε<br />
⎣ 13 ε23<br />
ε33<br />
⎥⎦<br />
where, in the case of the matrix,<br />
⎧ m I<br />
ε<br />
⎪<br />
11 = ε11<br />
⎪ I I<br />
m 2 L55<br />
ε12<br />
⎪ε12<br />
=<br />
I m<br />
⎪ L55<br />
+ L44<br />
⎪<br />
I I<br />
⎪ m 2 L55<br />
ε13<br />
⎪ε13<br />
=<br />
I m<br />
⎪ L55<br />
+ L44<br />
⎪ m m m m m<br />
⎨ m N1<br />
+ N2<br />
+ N3<br />
+ N4<br />
+ N<br />
ε<br />
5<br />
=<br />
⎪ 22<br />
m<br />
D<br />
⎪<br />
1<br />
⎪<br />
⎪ m<br />
ε23<br />
=<br />
⎪<br />
2<br />
I I<br />
⎪ 2 L22<br />
+ L23<br />
⎪<br />
⎪ m m I22<br />
ε = −<br />
⎪ 33 ε22<br />
4 L<br />
⎪⎩<br />
L<br />
I I I I<br />
2 L22<br />
( L22<br />
− L23<br />
) ε23<br />
I m I m I m ( L44<br />
− L44<br />
) + L22<br />
( 3 L44<br />
− 2 L23<br />
− 3 L44<br />
)<br />
I I I I<br />
( L22<br />
− L23<br />
)( ε22<br />
− ε33<br />
)<br />
2<br />
I I m m I m I m<br />
22 + 3 L22<br />
( L11<br />
− L12<br />
) - L23(<br />
L11<br />
+ L23<br />
- L12<br />
)<br />
(37)<br />
(39)
24<br />
Jacquemin Frédéric and Fréour Sylvain<br />
m m m m m<br />
( L11<br />
+ 2L12<br />
)( β11<br />
ΔC + M11<br />
ΔT)<br />
I I I I I I I I I<br />
−L12<br />
( β11ΔC<br />
+ M11ΔT)<br />
− ( L22<br />
+ L23<br />
)( β22ΔC<br />
+ M33ΔT)<br />
I m I<br />
( L12<br />
− L12<br />
) ε11<br />
I m m I I m m I<br />
I22 L22<br />
( 5 L11<br />
− L12<br />
+ 3 L22<br />
) − L23(<br />
3 L11<br />
+ L12<br />
+ 4 L22<br />
)<br />
L<br />
2 2<br />
I I m m I I ( 3 L22<br />
− L23<br />
)( L11<br />
− L12<br />
) + L22<br />
− L23<br />
I m m I I m m I<br />
I22 L22<br />
( L11<br />
− 5 L12<br />
− L22<br />
) + L23(<br />
L11<br />
+ 3 L12<br />
+ 4 L22<br />
) −<br />
L<br />
2 2<br />
I I m m I I ( 3 L22<br />
− L23<br />
)( L11<br />
− L12<br />
) + L22<br />
− L23<br />
⎧ m<br />
N<br />
⎪<br />
1 =<br />
⎪ m<br />
N2<br />
=<br />
⎪<br />
⎪ m<br />
N3<br />
=<br />
⎪<br />
⎪<br />
⎪ m<br />
N =<br />
⎨ 4<br />
⎪<br />
⎪<br />
⎪<br />
⎪ m<br />
N5<br />
=<br />
⎪<br />
⎪<br />
⎪ m m m I I<br />
⎩D1<br />
= L11<br />
+ L12<br />
+ L22<br />
− L23<br />
I<br />
2<br />
+ L23<br />
I<br />
ε22<br />
2<br />
I<br />
3 L23<br />
I<br />
ε33<br />
The pseudo-macroscopic stress tensors are deduced from the strains using (32). Thus, in<br />
the matrix, one will have:<br />
with<br />
⎧ m m<br />
σ<br />
⎪<br />
11 = L11<br />
m m<br />
⎨σ22<br />
= L11<br />
⎪ m m<br />
⎪σ<br />
=<br />
⎩ 33 L11<br />
(40)<br />
⎡ m m m m m<br />
σ<br />
⎤<br />
⎢<br />
11 2 L44ε12<br />
2 L44ε13<br />
m m m m m m ⎥<br />
σ = ⎢2<br />
L44ε12<br />
σ22<br />
2 L44ε<br />
23 ⎥<br />
(41)<br />
⎢ m m m m m ⎥<br />
⎢<br />
2 L<br />
⎣ 44ε13<br />
2 L44ε<br />
23 σ33<br />
⎥⎦<br />
m m m m m m m m m ( ε11<br />
− M11<br />
ΔT)<br />
+ L12<br />
( ε22<br />
+ ε33<br />
− 2 M11<br />
ΔT)<br />
− β11(<br />
L11<br />
+ 2 L12<br />
)<br />
m m m m m m m m m ( ε22<br />
− M11<br />
ΔT)<br />
+ L12<br />
( ε11<br />
+ ε33<br />
− 2 M11<br />
ΔT)<br />
− β11(<br />
L11<br />
+ 2 L12<br />
)<br />
m m m m m m m m m ( ε33<br />
− M11<br />
ΔT)<br />
+ L12<br />
( ε11<br />
+ ε22<br />
− 2 M11<br />
ΔT)<br />
− β11(<br />
L11<br />
+ 2 L12<br />
)<br />
m<br />
ΔC<br />
m<br />
ΔC<br />
m<br />
ΔC<br />
The local mechanical states in the fiber are provided by Hill’s strains and stresses average<br />
laws (36):<br />
ε<br />
σ<br />
(42)<br />
m<br />
r 1 I v m<br />
= ε − ε<br />
(43)<br />
r r<br />
v<br />
v<br />
m<br />
r 1 I v m<br />
= σ − σ<br />
(44)<br />
r r<br />
v<br />
v
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 25<br />
4.4. Examples of Multi-scale Stresses Estimations in <strong>Composite</strong> Structures:<br />
T300-N5208 <strong>Composite</strong> Pipe Submitted to Environmental Conditions<br />
4.4.1. Macroscopic Analysis<br />
4.4.1.1. Moisture Concentration<br />
Consider an initially dry, thin uni-directionally reinforced composite pipe, whose inner and<br />
outer radii are a and b respectively, and let the laminate be exposed to an ambient fluid with<br />
boundary concentration c0. The macroscopic moisture concentration, c I (r,t), is solution of the<br />
following system with Fick's equation (45), where D I is the transverse diffusion coefficient of<br />
the composite. Boundary and initial conditions are described in (46):<br />
I<br />
∂c<br />
∂t<br />
⎪⎧<br />
c<br />
⎨<br />
⎪⎩ c<br />
I<br />
I<br />
= D<br />
⎡ 2<br />
∂ c<br />
⎢ 2<br />
⎣ ∂r<br />
I<br />
( a,<br />
t)<br />
( r,<br />
0)<br />
= c<br />
= 0<br />
0<br />
I<br />
+<br />
I<br />
1 ∂c<br />
r ∂r<br />
and c<br />
I<br />
⎤<br />
⎥<br />
⎦<br />
( b,<br />
t)<br />
, a
26<br />
Jacquemin Frédéric and Fréour Sylvain<br />
plane tensors of hygroscopic expansion coefficients and moduli. Those tensors are assumed to<br />
be material constants.<br />
with,<br />
I<br />
⎧ I<br />
σ ⎫ ⎡ I<br />
L<br />
11<br />
⎪ ⎪ ⎢<br />
11<br />
I<br />
⎪σ<br />
⎪ ⎢ I<br />
22 L12<br />
⎨ ⎬ = ⎢<br />
I<br />
⎪σ33<br />
⎪ ⎢ I<br />
L12<br />
⎪ I ⎪ ⎢<br />
⎩τ12<br />
⎭ ⎢<br />
⎣0<br />
⎪⎧<br />
τ<br />
⎨<br />
⎪⎩ τ<br />
I<br />
32<br />
I<br />
13<br />
I<br />
L12<br />
I<br />
L22<br />
I<br />
L23<br />
0<br />
⎪⎫<br />
⎡L<br />
⎬ = ⎢<br />
⎪⎭ ⎢⎣<br />
0<br />
I<br />
L ⎤<br />
12 0<br />
⎥<br />
I<br />
L ⎥<br />
23 0<br />
⎥<br />
I<br />
L ⎥<br />
22 0<br />
⎥<br />
I<br />
0 L ⎥<br />
55 ⎦<br />
I<br />
44<br />
0<br />
L<br />
I<br />
55<br />
⎤<br />
⎥<br />
⎥⎦<br />
⎧ε<br />
⎪<br />
⎪ε<br />
⎨<br />
⎪ε<br />
⎪<br />
⎩γ<br />
⎪⎧<br />
γ<br />
⎨<br />
⎪⎩ γ<br />
I<br />
11<br />
I<br />
13<br />
I<br />
22<br />
I<br />
33<br />
I<br />
12<br />
I<br />
32<br />
⎪⎫<br />
⎬<br />
⎪⎭<br />
− β<br />
− β<br />
− β<br />
− β<br />
I<br />
11<br />
I<br />
22<br />
I<br />
12<br />
I<br />
ΔC<br />
⎫<br />
⎪<br />
I<br />
ΔC<br />
⎪<br />
⎬ I<br />
ΔC<br />
⎪<br />
I ⎪<br />
ΔC<br />
⎭<br />
I c<br />
Δ C = . c<br />
I<br />
ρ<br />
I and ρ I are respectively the macroscopic moisture concentration and the<br />
mass density of the dry material.<br />
To solve the hygromechanical problem, it is necessary to express the strains versus the<br />
displacements along with the compatibility and equilibrium equations.<br />
Introducing a characteristic modulus L 0 , we introduce the following dimensionless<br />
variables:<br />
σ<br />
Ι<br />
= σ<br />
Ι I I I I I I I I<br />
/ L0<br />
, L = L /L0<br />
, ( w , u , v ) = ( w , u , v ) / b.<br />
Displacements with respect to longitudinal and circumferencial directions, respectively<br />
u ( x,<br />
r)<br />
I<br />
and v ( x,<br />
r)<br />
I<br />
are then deduced:<br />
⎧ I<br />
u ( x,<br />
r)<br />
= R1x<br />
⎪<br />
I<br />
⎨v<br />
( x,<br />
r)<br />
= R 2xr<br />
⎪<br />
R1,<br />
R 2 are constants.<br />
⎩<br />
It is worth noticing that the displacements ( x,<br />
r)<br />
and ( x,<br />
r)<br />
do not depend on the<br />
moisture concentration field. Finally, to obtain the through-thickness or radial component of<br />
the displacement<br />
concentration (47).<br />
I<br />
w , we shall consider in the following the analytical transient<br />
I<br />
The radial component of the displacement field w satisfies the following equation:<br />
u I<br />
v I<br />
I<br />
22<br />
(48)<br />
(49)<br />
(50)
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 27<br />
r<br />
2<br />
∂<br />
2<br />
w<br />
∂r<br />
2<br />
I<br />
∂w<br />
+ r<br />
∂r<br />
I<br />
− w<br />
I<br />
2 ∂ΔC<br />
K1r<br />
=<br />
∂r<br />
L<br />
I I I I I I<br />
with, K1=<br />
L12<br />
β11<br />
+ L23<br />
β22<br />
+ L22<br />
β22<br />
It is shown that the general solution of equation (51) writes as the sum of a solution of the<br />
homogeneous equation and of a particular solution (Jacquemin et Vautrin, 2002).<br />
w<br />
I<br />
( r)<br />
R 4<br />
= R 3r<br />
+ −<br />
r<br />
∞<br />
∑<br />
m=<br />
1<br />
∞<br />
∑<br />
k=<br />
0<br />
2exp(- ωmτ)<br />
K<br />
ω Δ′ ( ω I ) L<br />
m<br />
u<br />
m<br />
22<br />
k 1 2k+<br />
1<br />
(-1) ( ) ( ωm<br />
)<br />
2<br />
k!<br />
( k + 1)!<br />
k 1 2k+<br />
1 2<br />
(-1) ( ) ( ω )<br />
∞<br />
m<br />
2<br />
2<br />
∑<br />
k=<br />
0 k!<br />
( k + 1)!<br />
2<br />
1<br />
k 1<br />
( −1)<br />
( )<br />
I<br />
22<br />
∞<br />
[ A<br />
2<br />
i<br />
∑<br />
+ {<br />
i<br />
k=<br />
0<br />
2k+<br />
2<br />
1<br />
[ 2ln(<br />
ω<br />
2<br />
k+<br />
2<br />
[<br />
ln( r)<br />
r<br />
(( 2k<br />
k!<br />
( k + 1)!<br />
m<br />
+ 3)<br />
2k+<br />
1<br />
( ω<br />
) − ψ(<br />
k + 1)<br />
− ψ(<br />
k + 2)]<br />
2k+<br />
3<br />
2<br />
−<br />
−1)<br />
2k+<br />
2<br />
m )<br />
2(<br />
2k<br />
(( 2k<br />
I<br />
(( 2k<br />
+ 3)<br />
+ 3)<br />
r<br />
+ 3)<br />
2<br />
r<br />
2k+<br />
3<br />
2k+<br />
3<br />
−1)<br />
2<br />
−1)<br />
(( 2k<br />
2<br />
] −<br />
r<br />
B<br />
π<br />
( 2k+<br />
3)<br />
+ 3)<br />
2<br />
r ln( r)<br />
+<br />
−1)<br />
Finally, the displacement field depends on four constants to be determined : Ri for i=1..4.<br />
These four constants result from the following conditions :<br />
• global force balance of the cylinder;<br />
• nullity of the normal stress on the two lateral surfaces.<br />
4.4.2. Numerical Simulations of Internal Stresses in T300/5208 <strong>Composite</strong><br />
Laminated Pipes<br />
4.4.2.1. Introduction<br />
Thin laminated composite pipes, with thickness 4 mm, initially dry then exposed to an<br />
ambient fluid, made up of T300/5208 carbon-epoxy plies, with a fiber volume fraction v r =0.6,<br />
were considered for the determination of both macroscopic stresses and moisture content as a<br />
function of time and space. The closed-form formalism used in order to determine the<br />
mechanical stresses and strains in each ply of the structure is described in subsection 4.4.1.<br />
This model ensures the calculation of the macroscopic moisture content, too.<br />
When the equilibrium state is reached, the maximum moisture content of the neat resin<br />
may be estimated from the maximum moisture content of the composite. By assuming that<br />
the fibers do not absorb any moisture, ΔC I and ΔC m are related by expression (16) given by<br />
(Loos and Springer, 1981). In the case of T300/5208, since the ratio between composite and<br />
resin densities is 1.33 (due to the constituents properties listed in table 1), the maximum<br />
moisture content ratio given by (16) is about 3.33.<br />
} ]<br />
(51)
28<br />
Jacquemin Frédéric and Fréour Sylvain<br />
Figure 2 shows the time-dependent concentration profiles, resulting from the application<br />
of a boundary concentration c0, as a function of the normalized radial distance from the inner<br />
radius rdim. At the beginning of the diffusion process important concentration gradients occur<br />
near the external surfaces. The permanent concentration (noticed perm in the caption) holds<br />
with a constant value because of the symmetrical hygroscopic loading. The macroscopic<br />
mechanical states were calculated for two types of composites structures: a) a unidirectionnaly<br />
reinforced cylinder, and b) a [55°/-55°]S laminated cylinder.<br />
c (%)<br />
1,5<br />
1,4<br />
1,3<br />
1,2<br />
1,1<br />
1<br />
0,9<br />
0,8<br />
0,7<br />
0,6<br />
0,5<br />
0,4<br />
0,3<br />
0,2<br />
0,1<br />
0<br />
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1<br />
r dim<br />
0.5 month<br />
1 month<br />
1.5 months<br />
2 months<br />
2.5 months<br />
3 months<br />
6 months<br />
Figure 2. Time dependent concentration profiles in T300/5208 as a function of the normalised radial<br />
distance from the inner radius r dim.<br />
Starting with the macroscopic stresses deduced from continuum mechanics, the local<br />
stresses in both the fiber and matrix were calculated either with the new analytical forms or<br />
the fully numerical model. The comparison between the two approaches is plotted on<br />
figures 3 and 4. These figures show the very good agreement between the numerical<br />
approach and the corresponding closed-forms solutions. The slight differences appearing<br />
are due to the small deviations on the components of Morris’ tensor calculated using the<br />
two approaches. Actually, it is not possible to assume the quasi-infinite length of the fiber<br />
along the longitudinal axis in the case of the numerical approach, because the numerical<br />
computation of Morris’ tensor is highly time-consuming. Thus, the numerical version of<br />
Eshelby-Kröner self-consistent model constitutes only an approximation of the real<br />
microstructure of the composite. In consequence, it seems that the new analytical forms,<br />
that are able to take into account the proper microstructure for the fibers, are not only more<br />
convenient, but also more reliable than the initially proposed numerical approach.<br />
4.4.2.2. Interpretation of the Simulations<br />
The highest level of macroscopic tensile stress is reached for the uni-directional composite, in<br />
the transverse direction and in the central ply of the structure (figure 3). The transverse<br />
stresses exceed probably the macroscopic tensile strength in this direction. The choice of a<br />
perm
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 29<br />
[+55°/-55°]S laminated allows to reduce the macroscopic stress in the transverse direction.<br />
Nevertheless, a high shear stress rises along the time in the fibers of the central ply of such a<br />
structure (figure 3).<br />
[MPa]<br />
[MPa]<br />
σ 11<br />
100<br />
50<br />
-50<br />
-100<br />
-150<br />
50<br />
-50<br />
-100<br />
-150<br />
-200<br />
200<br />
0<br />
0<br />
0<br />
0,5 1 1,5 2 2,5 3 6 perm<br />
0,5 1 1,5 2 2,5 3 6 perm<br />
b)<br />
a)<br />
month<br />
month<br />
[MPa]<br />
[MPa]<br />
100<br />
-100<br />
-150<br />
40<br />
35<br />
30<br />
25<br />
20<br />
15<br />
10<br />
-50<br />
5<br />
0<br />
50<br />
0<br />
AS4 50_1_1 ΔC m / ΔC I = 2<br />
1 2 3 4 5 6 7 8<br />
-200<br />
-400<br />
cas<br />
0,5 1 1,5 2 2,5 3 6 perm<br />
a)<br />
b)<br />
month<br />
0,5 1 1,5 2 2,5 3 6 perm<br />
month<br />
composite (CMF) matrix (numerical) fiber (numerical)<br />
matrix (analytical) fiber (analytical)<br />
Figure 3. Local stresses in T300/5208 composite for the central ply, in the case of a) the unidirectionaly<br />
reinforced composite and b) the [+55°/-55°] S symmetric laminate. CMF stands for<br />
Continuum Mechanics Formalisms.<br />
Moreover, the figure 4 shows that the micro-mechanical model always predict a very<br />
high compressive stress in the matrix of the inner ply whatever the laminate studied (the<br />
macroscopic stress is negligible in the radial direction because thin structures are considered).<br />
These local stresses could help to explain damage occurrence in the surface of composite<br />
structures in fatigue.
30<br />
, , [MPa]<br />
, , [MPa]<br />
σ 11<br />
150<br />
100<br />
50<br />
-50<br />
-100<br />
-150<br />
-200<br />
-250<br />
50<br />
0<br />
-50<br />
-100<br />
-150<br />
-200<br />
-250<br />
200<br />
0<br />
0<br />
a)<br />
Jacquemin Frédéric and Fréour Sylvain<br />
0,5 1 1,5 2 2,5 3 6 perm<br />
0,5 1 1,5 2 2,5 3 6 perm<br />
b)<br />
month<br />
month<br />
, , [MPa]<br />
, , [MPa]<br />
150<br />
100<br />
50<br />
0<br />
-50<br />
-100<br />
-150<br />
-200<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
-30<br />
-35<br />
-40<br />
AS4 50_1_1 ΔC m / ΔC I = 2<br />
1 2 3 4 5 6 7 8<br />
-200<br />
-400<br />
cas<br />
0,5 1 1,5 2 2,5 3 6 perm<br />
a)<br />
month<br />
0,5 1 1,5 2 2,5 3 6 perm<br />
b)<br />
month<br />
composite (CMF) matrix (numerical) fiber (numerical)<br />
matrix (analytical) fiber (analytical)<br />
Figure 4. Local stresses in T300/5208 composite for the inner ply, in the case of a) the uni-directionaly<br />
reinforced composite and b) the [+55°/-55°] S symmetric laminate. CMF stands for Continuum<br />
Mechanics Formalisms.<br />
This work demonstrates the complementarities of continuum mechanics and micromechanical<br />
models for the prediction of a possible damage in composite structures submitted<br />
to hygro-elastic loads.<br />
In the following section, the analytical expressions presented here for the localization of<br />
the macroscopic mechanical states within the plies constituents, will be inversed in order to<br />
achieve the identification of the strength of the constitutive matrix of a composite ply.<br />
5. Identification of the Local Strength of the Constitutive Matrix<br />
of a <strong>Composite</strong> Ply<br />
5.1. Introduction<br />
Damage predictions are important for design and for guiding materials improvement for<br />
engineering applications. <strong>Composite</strong> structures encountered in engineering applications
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 31<br />
are designed to endure combined mechanical, thermal and hygroscopic loads during their<br />
service life. Besides, composite structures usually benefit from improved properties<br />
granted by a multidirectional arrangement of their plies. The multiplicity of both possible<br />
loads and ply arrangements is not compatible with an extensive experimental<br />
investigation of composite structures damage. As a result, only uniaxial and pure shear<br />
test data of unidirectional composites are usually available in the literature. By<br />
consequence, the estimation of damage occurrence in composite structures requires<br />
introducing adapted failure criteria extending the available data to the combined loads<br />
and composite laminates considered for one particular application. Many published<br />
papers have dealt with this problem: see for instance (Tsai, 1987; Cuntze, 2003).<br />
Nevertheless, it is established for a long time, that in composite structures the damage<br />
initiates at microscopic scale, either (and most of time) in the matrix or (sometimes) in<br />
the fibers. The failure of a ply is thus closely related and explained by the failure of its<br />
microscopic constituents (Tsai, 1987; Cuntze, 2003; Fleck and Jelf, 1995; Kaddour et al.,<br />
2003; Khashaba, 2004). As a consequence, the reliable prediction of a possible damage<br />
occurrence of multi-directionnal laminates submitted to complex loading requires the<br />
knowledge of the microscopic failure criteria of the epoxy matrix and carbon fibers<br />
constituting the plies. Nevertheless, previous published works have emphasised the<br />
following remarkable result: the strength of the pure constituents (i.e. pure epoxy resin)<br />
strongly depends on the size of the sample, and especially on its thickness (Fiedler et al.,<br />
2001). Besides, the thickness of a ply in thin laminates has the magnitude of 150 microns,<br />
that is generally strongly weaker than the thickness of the samples tested for the<br />
experimental determination of the strength of the pure constituents. As a consequence,<br />
the experimental strengths of pure carbon fibers and epoxy matrices, determined on bulk<br />
specimen can hardly be directly used to properly estimate microscopic failure criteria in<br />
real structures. In particular, as shown for instance in (Garett and Bailey, 1977;<br />
Christensen and Rinde, 1979), the effect of the matrix on transverse failure of composite<br />
structures is of interest. The strain to failure of the pure matrix in uniaxial tension varied<br />
from 1.5 to 70 % whereas transverse strains to failure of corresponding fiber reinforced<br />
composites were dramatically smaller and varied only in the range 0.2 to 0.9%.<br />
In the present study, an innovative method, dedicated to the determination of the<br />
microscopic stress/strain failure criteria of the epoxy matrix embedded in a composite<br />
structure is described. This method is based on the inversion of the analytical expressions<br />
presented in section 4.3. The present work describes developments relating the<br />
macroscopic failure envelopes to the microscopic ones. The conditions, indicated in<br />
already published literature, when the macroscopic failure can exclusively be attributed<br />
to matrix failure modes are taken into account as fundamental hypotheses of the present<br />
approach. The model enables the identification of both the strength coefficients and<br />
ultimate strength, so that the microscopic stress/strain failure envelopes can also be<br />
drawn. Applications to the case of two typical carbon/epoxy composites (T300/5208 and<br />
AS4/3501) are achieved: the failure conditions of the N5208 and 3501-6 epoxy resins<br />
will be determined and compared.
32<br />
Jacquemin Frédéric and Fréour Sylvain<br />
5.2. Determination of the Local Failure Criterion of the Matrix from the<br />
Macroscopic Strength Data of the <strong>Composite</strong> Ply<br />
5.2.1. Introduction – Choice of a Failure Criterion<br />
In this paper, failure is taken in the general sense previously defined in the literature,<br />
including fracture, but also yield, etc. Since this works aims applications to multidirectional<br />
structures submitted to triaxial stresses, general failure criteria are necessary to the description<br />
of the strength in both stress and strain spaces. Failure criteria serve important functions in the<br />
design and sizing of composite laminates. They should provide a convenient framework or<br />
model for mathematical operations. The framework should be the same for different<br />
definitions of failures, such as the ultimate strength, endurance limit, or a working stress<br />
based on design or reliability considerations. However, the criteria are not intended to explain<br />
the mechanisms of failure, that can occur concurrently or sequentially. The quadratic criterion<br />
will be used in the present study: it includes interactions among the stress or strain<br />
components analogous to the Von Mises criterion for isotropic materials, and is compatible<br />
with the existence of strength having the properties, often met in the case that composite<br />
structures are considered, to be anisotropic and also possibly different in tension or<br />
compression. The criterion, expressed in stress space writes as follows :<br />
F<br />
i<br />
mnop<br />
σ<br />
i mn<br />
σ<br />
i op<br />
i<br />
mn<br />
i mn<br />
+ F σ = 1<br />
(52)<br />
where F stands for the strength parameters respectively expressed in stress space. The<br />
superscript i represents the scale considered for failure prediction (macroscopic: i=Ι or<br />
pseudomacroscopic: i=m or i=r).<br />
In order to use the failure criteria (52) presented above, it is necessary to identify the<br />
i<br />
i<br />
quadratic ( F mnop ) and linear ( F mn ) strength parameters involved in the equation.<br />
In the present work, for helping fixing the ideas, the simplified case of three-dimensional<br />
stresses and strains (for both macroscopic and microscopic scales), with a single shear<br />
component, usually met in multi-directional composite laminates submitted to mechanical<br />
i i<br />
loads (see examples given in Tsai, 1987) will be assumed to hold (i.e. σ13<br />
= σ23<br />
= 0 MPa ,<br />
i i<br />
ε13<br />
= ε23<br />
= 0 , where the subscripts 1, 2 and 3 respectively denotes the directions parallel<br />
to the fiber axis, the transverse direction and the normal direction, in the orthogonal frame of<br />
reference of the considered ply). Besides, the strength should be unaffected by the direction or<br />
i<br />
sign of the shear stress component σ 12 : if shear stress is reversed, the strength should be kept<br />
i<br />
i<br />
constant. However, sign reversal for the longitudinal ( σ 11)<br />
and transverse ( σ 22 ) stresses<br />
components from tension to compression is expected to have a significant effect on both the<br />
macroscopic and microscopic strength of the composite. As a consequence, terms of equation<br />
(52) containing first-degree shear stress should be null. Finally, taking into account the<br />
definition chosen for the reference frame, and the properties of (at least) transverse isotropy
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 33<br />
exhibited at any (i.e. macroscopic or microscopic) scale in one ply, the strength parameters<br />
have to satisfy the following relations:<br />
⎧ i i<br />
F<br />
⎪<br />
2222 = F3333,<br />
⎪ i i<br />
⎨F1122<br />
= F1133,<br />
⎪ i i<br />
⎪F<br />
= F .<br />
⎩ 22 33<br />
Taking into account the above listed simplifications, equation (52) can be rewritten:<br />
i i ( σ22<br />
+ σ33<br />
)<br />
⎧ i i<br />
2<br />
i i<br />
2<br />
i ⎛ i<br />
2<br />
i<br />
2 ⎞ i i<br />
i i i<br />
⎪1<br />
= F1111σ11<br />
+ 2 F1212σ12<br />
+ F2222<br />
⎜<br />
⎜σ<br />
22 + σ33<br />
⎟ + 2 F1122σ11<br />
+ 2 F2233σ22σ33<br />
+ (54)<br />
⎨<br />
⎝ ⎠<br />
⎪ i i i i i<br />
⎪⎩<br />
F11σ11<br />
+ F22<br />
( σ22<br />
+ σ33<br />
)<br />
5.2.2. Direct Identification of the Macroscopic Strength Parameters<br />
Most of the unknown macroscopic strength parameters in stress space, appearing in equation<br />
(54) can be identified using information deduced from simple mechanical tests (uniaxial<br />
tension, compression or longitudinal shear tests Tsai, 1987):<br />
/<br />
I<br />
I 1 I 1 I 1 1 I 1 1 I 1<br />
F = , F = , F = - , F = - , F =<br />
1111<br />
I I / 2222<br />
I I<br />
/ 11 I<br />
I<br />
/ 22 I<br />
I<br />
/ 1212<br />
X X Y Y X X Y Y 2 S<br />
Where X I and Y I are respectively the longitudinal and transverse tensile stress strength,<br />
/<br />
I<br />
Y the longitudinal and transverse compressive stress strength, whereas S I is the<br />
X and<br />
longitudinal shear stress.<br />
The two unknown remaining terms, I<br />
F 1122 and I<br />
F 2233 are related to the interaction<br />
between two orthogonal stress components. The practical determination of these interaction<br />
terms requires performing biaxial tests, which are not as easy to achieve than uniaxial tests.<br />
As a consequence, the required data are often not available in the literature. There are,<br />
however, geometric and physical conditions fixing the mathematical form of the failure<br />
criterion (54): for instance, the failure envelope has to be closed so that the material cannot<br />
present infinite strength when submitted to any load. Let us introduce a dimensionless<br />
interaction term:<br />
i*<br />
mmnn<br />
i<br />
Fmmnn<br />
i i<br />
mmmmFnnnn<br />
I 2<br />
(53)<br />
(55)<br />
F = (56)<br />
F
34<br />
Jacquemin Frédéric and Fréour Sylvain<br />
i*<br />
For closed envelopes, the condition − 1 ≤ Fmmnn<br />
≤ 1 has to be satisfied. But a more<br />
detailed theoretical study (see Liu and Tsai, 1998) reduces the admissible range to the domain<br />
1<br />
[-1,0]. The same reference (Liu and Tsai, 1998) advises the choice of F -<br />
2<br />
* I<br />
mmnn = for the<br />
macroscopic interaction term (which corresponds to the generalised Von Mises model), since<br />
this value is reasonable for a wide range of laminates. Taking into account this additional<br />
I I<br />
F = F ensures the<br />
assumption in equation (56), the knowledge of I<br />
F 1111 and<br />
I<br />
determination of the last two missing interaction terms 1122<br />
F and<br />
2222 3333<br />
I<br />
F 2233 , in stress space.<br />
One similar method could be applied in order to determine the macroscopic strength<br />
parameters expressed in strain space from the ultimate strains. Nevertheless, this method is<br />
not useful in practice since uniaxial strains are difficult to apply to a sample. Thus, the<br />
ultimate strains are generally deduced from the ultimate stresses: to reach this goal, one has to<br />
introduce the macroscopic properties, i.e. the stiffness tensor L I , in order to relate both failure<br />
criteria through Hooke’s law (33) expressed at macroscopic scale assuming a purely elastic<br />
load.<br />
5.2.3. Identification of the Microscopic Strength Parameter (of the Matrix Only)<br />
Using an Inverse Method<br />
From the standpoint of the structural designer, it is desirable to have failure criteria which are<br />
applicable at the level of the lamina, the laminate, and the structural component.<br />
Nevertheless, failure at macroscopic scale is often the consequence of an accumulation of<br />
micro-level failure events (Tsai, 1987; Liu and Tsai, 1998). Laminated materials typically<br />
exhibit many local failures prior to rupture. Thus, it is important to build up tools enabling to<br />
enhance the understanding of micro-level failure mechanisms in order to develop higherstrength<br />
materials. The ultimate goal is to have a failure theory that the designer can use with<br />
confidence under the most general structural configuration and loading conditions and that the<br />
developer of materials can use to design and fabricate new products to meet specific needs. In<br />
order to reach this goal, the estimation of microscopic strength criteria would be of a valuable<br />
help.<br />
Since the epoxy resins involved in composite structures generally exhibit an isotropic<br />
hygro-mechanical behaviour, the microscopic strength criterion expressed in terms of stresses<br />
(54) simplifies as follows:<br />
⎧<br />
⎪1<br />
⎨<br />
⎪<br />
⎩<br />
= F<br />
m<br />
1111<br />
+ F<br />
m<br />
11<br />
⎛<br />
⎜σ<br />
⎝<br />
m<br />
2<br />
11<br />
+ σ<br />
+ σ<br />
m m m<br />
( σ + σ + σ )<br />
11<br />
22<br />
m<br />
2<br />
22<br />
33<br />
m<br />
2<br />
33<br />
⎞<br />
⎟ + 2F<br />
⎠<br />
m<br />
1212<br />
σ<br />
m<br />
2<br />
12<br />
+ 2 F<br />
m<br />
1122<br />
m m m m m<br />
[ σ ( σ + σ ) + σ σ ]<br />
Thus, only four strength parameters have to be determined in order to enable failure<br />
m m m m<br />
predictions at microscopic scale:<br />
F1111<br />
, F1212<br />
, F1122<br />
, F11<br />
.<br />
Hypotheses being compatible<br />
with the experimental observations are necessary to build an inverse model enabling the<br />
11<br />
22<br />
33<br />
22<br />
33<br />
(57)
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 35<br />
determination of these four parameters from the corresponding, available from practical<br />
mechanical tests, macroscopic strength stress failure criterion.<br />
The present work is focused on the development of modelling tools for the prediction of a<br />
possible damage occurrence in fiber-reinforced epoxy laminates submitted to mechanical<br />
loads. Actually, fibrous composite materials fail in a variety of mechanisms at the<br />
fiber/matrix microscopic scale. Besides, according to the literature, i) fiber-dominated failures<br />
usually occur when the plies are loaded in planes perpendicular to the fibers axis (longitudinal<br />
tension and compression), whereas ii) matrix-dominated failures often occur in the cases that<br />
the plies are loaded along the transverse and normal directions in tension and compression or<br />
when shear stresses are applied to the considered ply (Tsai, 1987; Liu and Tsai, 1998). Thus,<br />
matrix-dominated failure modes often occur in practice. As a consequence, the above listed i)<br />
and ii) statements will be used in order to identify microscopic strength parameters in stress<br />
and strain spaces for the matrix.<br />
According to the developments of section 4, it is possible to derive the pioneering<br />
numerical self-consistent model of Kröner and Eshelby in order to find the relation between<br />
the macroscopic mechanical states and the researched corresponding microscopic stresses and<br />
strains existing in the matrix of a composite material.<br />
In the present work, the strength parameters in either the matrix or the ply will be<br />
considered to remain independent from the magnitude of the applied mechanical load. Since<br />
the damage envelope has been defined as the strain or stress threshold beyond which nonlinearity<br />
occurs in the behaviour of the material at the scale concerned by damage, and in the<br />
case that a purely mechanical load is taken into account, the material is assumed to behave<br />
elastically until failure occurs. Now, in these conditions, both stress and strain ultimate<br />
strength are simultaneously reached, and satisfy either macroscopic elastic Hooke’s law (33)<br />
or the corresponding microscopic relations that are deduced from (38-42), assuming<br />
I m<br />
I m<br />
ΔC = ΔC = 0 and ΔT = ΔT = 0 K .<br />
It will be assumed that macroscopic failure occurring in the transverse and normal<br />
directions, for a longitudinal stress σ 0 MPa<br />
I<br />
11 = , is governed by local failure of the matrix.<br />
Various macroscopic stress states, compatible with that last hypothesis, are taken on the<br />
macroscopic strength envelope (54), expressed in stress space and, finally implemented in the<br />
scale transition relations (38-42). This leads to the determination of microscopic mechanical<br />
stresses and strains states in the matrix, that are, according to our hypotheses, responsible for<br />
macroscopic damage governed by matrix failure. As a consequence, these local mechanical<br />
states should be compatible with the microscopic failure envelopes of the matrix as written in<br />
equations (57).<br />
According to this relation, four, non equivalent, macroscopic stress states suffice to find<br />
m m m m<br />
the eight researched coefficients involved in (57): F 1111,<br />
F1212<br />
, F1122<br />
, F11<br />
. The whole method<br />
required to perform such estimation is described on table 7. Actually, four macroscopic<br />
loading states taken on the stress failure envelope (defined on table 7) σ , σ , σ and<br />
are required for the determination of the four coefficients of the failure envelopes since<br />
numerical tests shows that equation (56) rewritten at microscopic scale for the epoxy matrix<br />
does not provide an additional relation between m<br />
F 1111 and m<br />
F 1122 :<br />
I<br />
a<br />
I<br />
b<br />
I<br />
c<br />
I<br />
σ d
36<br />
Jacquemin Frédéric and Fréour Sylvain<br />
m*<br />
1122<br />
F<br />
m<br />
F1122<br />
1<br />
= ≠ −<br />
(58)<br />
m<br />
F 2<br />
1111<br />
Moreover, according to (38-42) an uniaxial macroscopic tension or compression along<br />
the transverse (or normal) direction induces local mechanical states in the matrix generally<br />
exhibiting no zero strain and stress on-diagonal components (see for instance the cases of the<br />
macroscopic loads<br />
I<br />
σ a and<br />
I<br />
σ b on Table 7). As a consequence, only the strength coefficient<br />
m<br />
F 1212 can be determined independently from the three others, from the single macroscopic<br />
I<br />
m m m<br />
load σ d . Concerning the calculation of F 1111,<br />
F1122<br />
, F11<br />
, one has to solve numerically the<br />
system (60) (cf. Table 7).<br />
Finally, the uniaxial microscopic ultimate stresses of the epoxy matrix embedded in the<br />
composite structure can be deduced from the set of equations (55) expressed at microscopic<br />
scale (i.e. replacing the subscripts I by the subscript m ), provided that the coefficients of the<br />
local failure envelope are already known:<br />
⎧<br />
⎪X<br />
⎪<br />
⎪<br />
⎪<br />
⎨X<br />
⎪<br />
⎪<br />
⎪S<br />
⎪<br />
⎩<br />
m<br />
/<br />
m<br />
m<br />
= Y<br />
=<br />
m<br />
= Y<br />
m<br />
/<br />
1<br />
2 F<br />
= Z<br />
= Z<br />
m<br />
1212<br />
m<br />
1<br />
=<br />
2 F<br />
m<br />
/<br />
m<br />
1111<br />
1<br />
=<br />
2 F<br />
⎛<br />
⎜<br />
⎝<br />
m<br />
1111<br />
⎛<br />
⎜<br />
⎝<br />
m<br />
2<br />
11<br />
+ 4 F<br />
m<br />
2<br />
11<br />
F<br />
m<br />
1111<br />
+ 4 F<br />
The method, developed in the present paragraph, enables the determination of a) the<br />
coefficients of the microscopic failure envelope of the epoxy matrix in stress and/or strain<br />
space from the macroscopic failure envelope of the ply and scale transition relations<br />
linking macroscopic loads to the corresponding local microscopic mechanical states<br />
experienced by the matrix, only thereafter, b) the local maximum strength of the matrix<br />
embedding the carbon fibers which can be evaluated from the classical formalism relating<br />
the strength to the coefficients of the failure envelope. This inverse method provides an<br />
alternative to the classical direct approach leading to the determination of the failure<br />
envelope from the maximum strength measured on pure epoxies, in the cases that the<br />
required data is not available or when the behaviour of the matrix embedded in the<br />
composite structure is expected to be significantly different from the behaviour of the<br />
pure matrix, as shown for example, in references (Garett and Bailey, 1977; Christensen<br />
and Rinde, 1979).<br />
F<br />
− F<br />
m<br />
1111<br />
m<br />
11<br />
⎞<br />
⎟<br />
⎠<br />
+ F<br />
m<br />
11<br />
⎞<br />
⎟<br />
⎠<br />
(62)
Applied<br />
macroscopic load<br />
Corresponding<br />
macroscopic strain<br />
Corresponding<br />
microscopic stress<br />
according to (15-17)<br />
Corresponding<br />
conditions for<br />
finding the<br />
microscopic strength<br />
coefficients in stress<br />
space from (10, 19)<br />
I<br />
σ<br />
a<br />
I<br />
ε<br />
i<br />
Table 7. One possible set of trials enabling the determination of the microscopic strength<br />
coefficients of the matrix expressed in stress space.<br />
⎡0<br />
⎢<br />
=<br />
⎢<br />
0<br />
⎢<br />
⎣0<br />
=<br />
⎡<br />
⎢<br />
ε<br />
⎢<br />
⎢<br />
⎢<br />
⎢<br />
⎣<br />
i = a, b<br />
0<br />
0<br />
0<br />
I<br />
Y<br />
I<br />
11i<br />
0<br />
0⎤<br />
⎥<br />
0 , I<br />
⎥<br />
σ<br />
b<br />
0⎥<br />
⎦<br />
0<br />
I<br />
ε<br />
22i<br />
0<br />
I<br />
ε<br />
33i<br />
⎡0<br />
⎢<br />
= ⎢0<br />
⎢<br />
⎣<br />
0<br />
0<br />
0<br />
⎤<br />
⎥<br />
⎥<br />
⎥<br />
⎥<br />
⎥<br />
⎦<br />
0<br />
I<br />
/<br />
Y<br />
0<br />
0⎤<br />
⎥<br />
0⎥<br />
0⎥<br />
⎦<br />
I I I<br />
ε11i<br />
= S12<br />
σ22i<br />
,<br />
I I I<br />
ε22i<br />
= S22<br />
σ22i<br />
,<br />
I I I<br />
ε33i<br />
= S23<br />
σ22i<br />
⎡0<br />
0 0 ⎤<br />
I ⎢ I ⎥<br />
σ c =<br />
⎢<br />
0 σ22c<br />
0<br />
⎥ ,<br />
⎢<br />
I<br />
⎣0<br />
0 σ ⎥<br />
33c⎦<br />
⎧ I ⎛ I2<br />
I2<br />
⎞<br />
⎪F2222<br />
⎜<br />
⎜σ<br />
22c + σ33c<br />
⎟ +<br />
⎪ ⎝ ⎠<br />
⎨<br />
⎪ I I I I ⎛ I2<br />
I2<br />
⎞<br />
⎪<br />
2 F2233<br />
σ22c<br />
σ33c<br />
+ F22<br />
⎜<br />
⎜σ22c<br />
+ σ33c<br />
⎟ −1<br />
= 0<br />
⎩<br />
⎝ ⎠<br />
⎡ I<br />
ε<br />
⎢ 11c<br />
I<br />
ε = ⎢<br />
c 0<br />
⎢<br />
⎢<br />
0<br />
⎢⎣<br />
0<br />
I<br />
ε 22c<br />
0<br />
0 ⎤<br />
⎥<br />
0 ⎥<br />
⎥<br />
I<br />
ε<br />
⎥<br />
33c ⎥⎦<br />
I I ( σ + σ )<br />
I I<br />
ε11c<br />
= S12<br />
22c 33c ,<br />
I I I I I<br />
ε22c<br />
= S22<br />
σ22c<br />
+ S23<br />
σ33c<br />
,<br />
I I I I I<br />
ε33c<br />
= S23<br />
σ22c<br />
+ S22<br />
σ33c<br />
⎡ m<br />
σ<br />
⎢ 11i<br />
m<br />
σ ⎢<br />
i = 0<br />
⎢<br />
⎢<br />
0<br />
⎢⎣<br />
0<br />
m<br />
σ 22i<br />
0<br />
0 ⎤<br />
⎥<br />
0 ⎥ , i = a, b, c<br />
⎥<br />
m<br />
σ<br />
⎥<br />
33i ⎥⎦<br />
(59)<br />
⎧<br />
F<br />
m<br />
+<br />
m<br />
+<br />
m<br />
⎪ 1111<br />
Ai<br />
F<br />
1122<br />
Bi<br />
F<br />
11<br />
Ci<br />
−1<br />
= 0<br />
⎪ 2 2 2<br />
⎪A<br />
=<br />
m<br />
+<br />
m<br />
+<br />
m<br />
⎪ i<br />
σ<br />
11i<br />
σ<br />
22i<br />
σ<br />
33i<br />
, i = a, b, c<br />
⎨ ⎡ ⎛<br />
⎞ ⎤<br />
⎪B<br />
= 2 ⎢σ<br />
m ⎜σ<br />
m<br />
+ σ<br />
m ⎟ + σ<br />
m<br />
σ<br />
m ⎥<br />
⎪ i<br />
⎢<br />
11i ⎜ 22i 33i ⎟ 22i 33i<br />
⎥<br />
⎪ ⎣ ⎝<br />
⎠ ⎦<br />
⎪<br />
⎩<br />
C = σ<br />
m<br />
+ σ<br />
m<br />
+ σ<br />
m<br />
i 11i 22i 33i<br />
(60)<br />
I<br />
σ d<br />
⎡ 0<br />
⎢ I<br />
= ⎢S<br />
⎢<br />
⎢<br />
0<br />
⎣<br />
S<br />
⎡ 0<br />
⎢<br />
I ⎢ I<br />
ε d = ε<br />
⎢ 12d<br />
⎢ 0<br />
⎣<br />
I<br />
12d<br />
ε = S<br />
m<br />
σ d<br />
I<br />
66<br />
⎡ 0<br />
⎢<br />
= ⎢S<br />
⎢<br />
⎢<br />
0<br />
⎣<br />
m<br />
S<br />
I<br />
0<br />
0<br />
I<br />
0⎤<br />
⎥ ⎥⎥⎥<br />
0<br />
0<br />
⎦<br />
I<br />
ε12d<br />
S<br />
m<br />
0<br />
0<br />
0<br />
0<br />
0⎤<br />
⎥<br />
0⎥<br />
0<br />
⎥<br />
⎥⎦<br />
0⎤<br />
⎥<br />
0⎥<br />
⎥<br />
0⎥<br />
⎦<br />
2<br />
2 F<br />
m<br />
S<br />
m<br />
1212<br />
-1<br />
= 0 (61)
38<br />
Jacquemin Frédéric and Fréour Sylvain<br />
5.3. Numerical Applications and Examples<br />
5.3.1. Identification of the Microscopic Failure Criteria of Two Typical Epoxies<br />
from the Knowledge of the Macroscopic Failure Envelope of AS4/3501-6 and<br />
T300/N5208 <strong>Composite</strong> Plies<br />
In the present paper, two types of high strength carbon fiber reinforced epoxies are<br />
considered: a) AS4/3501-6 and b) T300/N5208 composites having identical fiber volume<br />
fraction: v f =0.6. These two materials constitute good candidates for the present work, since<br />
the microscopic strength of their respective matrix is not yet available (at our knowledge) in<br />
the already published literature, in spite of they are quite often considered for illustrating<br />
scientific works in this field of research (Tsai, 1987).<br />
Strengths [MPa] X I<br />
Table 8. Macroscopic strength data.<br />
X I ´ Y I , Z I<br />
Y I ´, Z I ´ S I<br />
T300/5208 (Tsai, 1987) 1500 1500 40 246 68<br />
AS4/3501-6 (Liu and Tsai, 1998) 1950 1480 48 200 79<br />
Table 9. Quadratic macroscopic stress failure criteria deduced from the strength data.<br />
Quadratic ijkl subscripted coefficients [MPa -2 ] and linear ij subscripted coefficients<br />
[MPa -1 ].<br />
Strength<br />
parameters<br />
I<br />
F 1111<br />
T300/5208 4.44 10 -7<br />
I<br />
F , 2222<br />
I<br />
F 3333<br />
1.02 10 -4<br />
I<br />
F 1212<br />
I<br />
F , 1122<br />
I<br />
F 1133<br />
I<br />
F 2233<br />
I<br />
F 11<br />
I<br />
F ,<br />
22<br />
I<br />
F 33<br />
1.08 10 -4 -3.36 10 -6 -5.08 10 -5 0 0.0209<br />
AS4/3501-6 3.46 10 -7 1.04 10 -4 8.01 10 -5 -3.00 10 -6 -5.02 10 -5 -0.0002 0.0158<br />
The macroscopic strength of single plies are given in Table 8. The coefficients of the<br />
corresponding quadratic macroscopic stress failure criteria, deduced from the classical direct<br />
method, through equation (55) are listed in Table 9.<br />
Table 10. Macroscopic stiffness components [GPa] of 60% volume uni-directionally<br />
fiber reinforced plies. Fiber axis is parallel to longitudinal direction.<br />
Stiffness<br />
components<br />
I<br />
L 11<br />
T300/5208 142.72<br />
AS4/3501-6 137.27<br />
I<br />
L , 22<br />
13.92<br />
I<br />
L 33<br />
I<br />
L , 12<br />
I<br />
L 13<br />
I<br />
L 23<br />
I<br />
L 44<br />
I<br />
L ,<br />
55<br />
5.79 7.19 3.34 7.00<br />
11.60 4.20 5.22 3.68 6.45<br />
I<br />
L 66
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 39<br />
Table 11. Quadratic local stress failure criteria in N5208 and 3501-6 epoxy matrices<br />
respectively deduced from the macroscopic failure envelopes of T300/5208 and<br />
AS4/3501-6 plies, taking into account the microscopic elastic properties given on tables 1<br />
and 5. Quadratic ijkl subscripted coefficients [MPa -2 ] and linear ij subscripted<br />
coefficients [MPa -1 ].<br />
Strength<br />
parameters<br />
m<br />
F 1111 , m<br />
F 2222 , m<br />
F 3333<br />
N5208 2.18 10 -4<br />
3501-6 2.15 10 -4<br />
m<br />
F 1212<br />
m<br />
F 1122 , m<br />
F , 1133<br />
m<br />
F 2233<br />
m<br />
F 11 , m<br />
F ,<br />
22<br />
m<br />
F 33<br />
7.82 10 -4 -8.77 10 -5 0.0162<br />
5.04 10 -4 -8.07 10 -5 0.0143<br />
Table 12. Local (matrix embedded in a composite ply) strength data deduced from the<br />
local quadratic stress failure criteria of a N5208 and 3501-6 epoxy matrices respectively.<br />
Strengths [MPa] X m , Y m , Z m X m ´, Y m ´, Z m ´ S m<br />
N5208 40.1 114.7 25.3<br />
3501-6 42.6 108.9 30.9<br />
In order to achieve the identification of the coefficients of the quadratic microscopic<br />
failure criteria of the pure epoxies (3501-6 and N5208, respectively), the method<br />
previously explained in subsection 5.2.3 was applied. The macroscopic stiffnesses<br />
considered for the simulation are provided in Table 10, whereas the elastic constants of<br />
the elastically isotropic resins, required for localising the macroscopic stress/strain states<br />
at the microscopic scale in the matrices, according to equations (38-42), were previously<br />
given in tables 1 and 5. In order to find the microscopic strength coefficients, four<br />
independent macroscopic stress states<br />
I<br />
σ a ,<br />
I<br />
σ b ,<br />
I<br />
σ c ,<br />
I<br />
σ d located on the macroscopic<br />
failure envelope according to the conditions described on the first raw of Table 7. Table<br />
11 shows the strength coefficients found for the quadratic microscopic failure criterion in<br />
stress space of both epoxies by solving equations (60-61). Besides, the microscopic<br />
ultimate uniaxial stresses of the two studied epoxies have been determined by introducing<br />
in equation (62) the results of the previous identification of the strength coefficients of<br />
their respective quadratic failure criterion in stress space (still Table 11). The<br />
corresponding results have been listed in Table 12.<br />
Finally, instances of the microscopic failure envelopes have been drawn and<br />
superimposed to the corresponding macroscopic failure envelopes. Pictures of Figure 5<br />
compare the results obtained in stress space for each couple epoxy/composite.
40<br />
&22 [MPa]<br />
& & [MPa]<br />
& & [MPa]<br />
-4000 -3000 -2000 -1000-50 0 1000 2000<br />
T300/5208<br />
N5208<br />
Jacquemin Frédéric and Fréour Sylvain<br />
50<br />
-150<br />
-250<br />
-350<br />
& 11 [MPa]<br />
-500 -400 -300 -200 -100 0 100<br />
-100<br />
T300/5208<br />
N5208<br />
100<br />
50<br />
-250 -200 -150 -100 -50 0 50<br />
T300/5208<br />
N5208<br />
σ 22 [MPa]<br />
& 22 [MPa]<br />
200<br />
100<br />
0<br />
-200<br />
-300<br />
-400<br />
-500<br />
0<br />
-50<br />
-100<br />
& & &[MPa]<br />
σ 33 [MPa]<br />
& & [MPa]<br />
-4000 -3000 -2000 -1000-50 0 1000 2000<br />
AS4/3501<br />
3501-6<br />
50<br />
-150<br />
-250<br />
-350<br />
σ 11 [MPa]<br />
200<br />
-500 -400 -300 -200 -100 0 100<br />
AS4/3501<br />
3501-6<br />
σ 22 [MPa]<br />
0<br />
-200<br />
-400<br />
100<br />
50<br />
-250 -200 -150 -100 -50 0 50<br />
AS4/3501<br />
3501-6<br />
σ 22 [MPa]<br />
Figure 5. Examples of macroscopic and local (matrix only) stress failure envelopes of T300/5208 and<br />
AS4/3501-6 plies.<br />
5.3.2. Observations on Predicted Results and Discussion<br />
According to the identification procedure described in subsection 5.2.3, an infinite number of<br />
I I I I<br />
macroscopic stress states sets { σ a , σ b , σ c , σ d } can be considered for the determination<br />
I<br />
of the researched microscopic failure envelope strength coefficients. Actually, σ c only may<br />
I<br />
a<br />
I<br />
b<br />
I<br />
d<br />
vary whereas σ , σ and σ are fixed by the macroscopic ultimate stresses<br />
of the considered composite structure (see the first raw of Table 7). Several tests were<br />
0<br />
-50<br />
-100<br />
I<br />
Y ,<br />
/<br />
I<br />
Y ,<br />
I<br />
S
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 41<br />
performed, introducing various numerical stress states (compatible with the constitutive<br />
I<br />
hypotheses of the present work) for σ c . The tests showed that the microscopic strength<br />
coefficients are, as expected, independent from the choice of the initial macroscopic stress<br />
I<br />
state σ c : one set of coefficients only is found as the unique solution of system (60). This<br />
demonstrates that the inverse model presented here is reliable from a numerical point of view.<br />
The obtained results for the ultimate uniaxial stresses of 3501-6 and N5208 epoxies are<br />
close together (Table 12), whereas the macroscopic strength present significant discrepancies<br />
(Table 8). As an example, the relative deviation between the macroscopic longitudinal tensile<br />
ultimate stress of the two composites reaches around 25% when the relative deviation<br />
between the longitudinal tensile ultimate stress of the two epoxies is limited to 6%. Moreover,<br />
the representation of the microscopic failure envelopes are rather similar for the two<br />
considered resins, (Figure 5), whereas the macroscopic failure envelopes differ from one<br />
composite to the other (Figure 5, also). This could be interpreted as follows: for the<br />
considered composites, the observed deviation in the macroscopic failure envelopes comes<br />
from the choice of the reinforcing fibers and not from the choice of the resin. This is<br />
remarkable, since the considered epoxies exhibit a very different elastic mechanical behaviour<br />
(see Tables 1 and 5).<br />
Moreover, the predicted microscopic ultimate uniaxial stresses are coherent with<br />
experimental results measured on plain resins. For instance, reference (Fiedler et al., 2001)<br />
reports a strength value of 117 MPa in compression, and elastic limits reaching respectively<br />
29 MPa in tension and 31 MPa in torsion for small specimen of plain unreinforced Bisphenol-<br />
A type resin (i.e. “small” denotes a significantly reduced sized in normal and transverse<br />
directions compared to “bulk” specimen). These measured strength are of the same order of<br />
magnitude than the strength, calculated in the present work, for 3501-6 and N5208 epoxies.<br />
At the opposite, the strengths determined on bulk specimens of 5208 and 3501-6 plain<br />
epoxies are approximately two times higher than the values obtained in the present work, for<br />
the strength of the corresponding epoxies embedded in thin composite plies. This last result is<br />
also compatible with both the experimental comparison achieved in reference (Fiedler at al.,<br />
2001) on various sized pure epoxies and the practical comparisons of the failure mechanisms<br />
exhibited by composites structures and their constitutive epoxy resin (see Garett and Bailey,<br />
1977; Christensen and Rinde, 1979). The present work allows to represent the scale effects<br />
observed in practice on the composite constituents strengths, because the composite ply<br />
strengths involved in the calculations do actually depend on both the constituents properties<br />
and microstructure.<br />
6. Conclusions<br />
The present work dealt with the question of scale transition modelling of polymer matrix<br />
composites and its application to several fields of investigation. Therefore, Mori-Tanaka and<br />
Eshelby-Kröner self-consistent models, taking advantage of arithmetic averages, were both<br />
considered for achieving the determinaiton of the homogenized properties of composite ply as<br />
a function of the properties of its constituents (on the one hand, the matrix , and on the second<br />
hand, the reinforcements).
42<br />
Jacquemin Frédéric and Fréour Sylvain<br />
The theoretical models properly take into account the specific microstructure of such<br />
materials. Especially the extreme morphology of the reinforcements can be considered, while<br />
the morphology and orientation of the reinforcing inclusions are kept constant in a single ply.<br />
As a consequence, the models manage to reproduce realistically the strong macroscopic<br />
anisotropy observed in practice on uni-directionally fiber-reinforced epoxies. The obtained<br />
results have shown that the two approaches, presented here, yield close together estimations<br />
of the macroscopic coefficients of thermal expansion, coefficients of moisture expansion and<br />
elastic moduli, in the range of the epoxy volume fraction, that is typical for designing<br />
m<br />
composites structures for engineering applications ( i.e. 0.3 ≤ v ≤ 0.7)<br />
. Nevertheless, an<br />
I<br />
exception to this statement occurs for Coulomb modulus G 12 , that is strongly<br />
underestimated in the case that the calculations are performed according to Mori-Tanaka<br />
approximation, in the same range of epoxy volume fractions.<br />
Moreover, realistic inverse scale transition procedures based on Kröner-Eshelby selfconsistent<br />
model and Mori-Tanaka estimates were also provided for achieving the numerical<br />
determination of the mechanical, hygroscopic or thermal properties of one constituent of an<br />
uni-directionally reinforced composite ply. Both models were used in order to estimate the<br />
elastic stiffness of reinforcing fibers embedded in a composite ply, from the knowledge of the<br />
macroscopic properties and those of the matrix. The obtained numerical results were<br />
successfully compared with expected practical results. A similar study was achieved in the<br />
standpoint of estimating the coefficients of moisture expansion of the matrix constituting a<br />
composite ply. In both cases the proposed theoretical approaches led to similar results, which<br />
is satisfying. Thus, the two inverse models described in the present work can be equally used<br />
in order to achieve such an identification.<br />
Another section of this article was devoted to the analysis of the macroscopic mechanical<br />
states localization within the constituents of a composite ply. Since it was previously<br />
demonstrated in the literature, that Mori-Tanaka approximation was not reliable for handling<br />
such a task, only Eshelby-Kröner model was considered. A numerical model, valid for any<br />
morphology of the reinforcing inclusions, was provided. Moreover a rigorous fully analytical<br />
treatment of the classical Kröner and Eshelby Self-Consistent model including morphology<br />
effects was achieved also. Especially, the determination of Morris’ tensor was performed in a<br />
satisfactory agreement with the transverse macroscopic elastic anisotropy expected for the<br />
fiber shape that should be taken into account in order to satisfactory represent the specific<br />
microstructure of carbon-fiber reinforced composites. The new closed-form solutions<br />
obtained for the components of Morris’ tensor were introduced in the classical hygro-thermoelastic<br />
scale transition relation in order to express analytically the internal strains and stresses<br />
in both the fiber and the resin of a ply submitted to a hygro-thermo-elastic load. The closedform<br />
solution demonstrated in the present work was compared to the fully numerical selfconsistent<br />
model for various geometrical arrangements of the fibers: uni-directional or<br />
laminated composites. A very good agreement was obtained between the two models for any<br />
component of the local stress tensors. It was also demonstrated that continuum mechanics and<br />
micro-mechanical models give complementary information about the occurrence of a possible<br />
damage during the loading of the structure.<br />
In a last part, the present study explained a procedure enabling to achieve the<br />
identification of one single set of strength parameters defining completely the microscopic
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 43<br />
failure envelope of the matrix entering in the composition of a composite structure, in the<br />
cases that a pure mechanical load is applied. The identification method was built around an<br />
inverse scale transition method which requires the knowledge of the macroscopic strengths,<br />
and both the macroscopic and microscopic elastic stiffnesses. Besides, it was necessary to<br />
consider some hypotheses in order to proceed to the identification of the coefficients of the<br />
microscopic quadratic failure criteria. In the present work, it was assumed that the<br />
macroscopic failure of a uni-directionally reinforced ply is dominated by the local failure of<br />
the matrix when the external load is applied in planes perpendicular to the fiber axis.<br />
Numerical applications of the proposed inverse method were made considering the cases<br />
of two high-strength composites structures: AS4/3501-6 and T300/N5208. The determination<br />
of the microscopic quadratic failure criterion of the pure epoxies (3501-6 and N5208,<br />
respectively) was achieved. The obtained results are close together and present a good<br />
agreement with ultimate strengths measured on reduced sized plain resins (available from<br />
already published literature). This demonstrates the reliability of the present predictive<br />
method for estimating the local failure behaviour of epoxies whose experimental failure<br />
criterion has not yet been determined.<br />
In further works, the proposed approach will be extended to the more general case of<br />
hygro-thermo-mechanical loads. This will imply to take into account the stress free strains in<br />
order to keep consistency between the failure envelopes expressed in stress and strain spaces.<br />
Besides, the rigorous treatment of the hygro-thermo-mechanical load requires to consider the<br />
dependence on the temperature and moisture content of a) the elastic stiffness, coefficients of<br />
thermal expansion and coefficients of moisture expansion and b) the ultimate strength (and in<br />
general, the coefficients of the considered failure criterion), at both macroscopic and<br />
microscopic scales. Others perspectives of research are proposed in the following section<br />
below.<br />
7. Perspectives<br />
Scale transition modelling based theoretical analysis of composite structures constitutes an<br />
overexpanding field of research, due to multiple factors. Among them, the emergence of new<br />
materials exhibiting a specific, more advanced microstructure, the ambition to account for<br />
additional, sometimes only recently discovered, physical phenomena and the relentless<br />
research for building faster, more convenient but still reliable models stand for the three<br />
essential motivations for achieving further developments in the incoming years.<br />
7.1. Emergence of New <strong>Materials</strong><br />
The present development stage of Eshelby’s single inclusion theory involved in the<br />
mechanical modeling of composites is not intended for a rigorous treatment of the<br />
morphology presented by the reinforcements used for manufacturing woven-composites. As a<br />
consequence, answering to the question of a theoretical study, through scale transition<br />
models, of mechanical parts made of such composites will require a specific and still missing<br />
solution.
44<br />
Jacquemin Frédéric and Fréour Sylvain<br />
Since the recent discovery of carbon nanotubes in the 90’s, researchers worldwide have<br />
engaged in fundamental studies of this novel material (Treacy et al., 1996). The pioneering<br />
works have underlined the characteristics of carbon nanotubes such as an extraordinarily high<br />
stiffness (Salvetat et al., 1999) coupled to a high tensile strength (Demczyk et al., 2002)., high<br />
aspect ratio and an especially low density. Actually, for instance, the experimental direct<br />
mechanical measurement of the elastic properties of carbon nanotubes provided Young’s<br />
moduli in the range of 1 TPa, which considerably exceeds the corresponding modulus of any<br />
currently available fiber material (Salvetat et al., 1999; Demczyk et al., 2002).<br />
In consequence, the technological applications of carbon nanotubes as reinforcements for<br />
elastomers (Frogley et al., 2003) or polymer-based composites (Liu and Wagner, 2005;<br />
Breton et al., 2004; Xiao et al., 2006) was very recently investigated. Furthermore, multimaterials<br />
made up of polymer matrix, carbon fibers and carbon nanotubes are considered also<br />
for achieving a new generation of engineering composites.<br />
7.2. Accounting for Additional Physical Factors<br />
The present work is focused on the theoretical prediction of the mechanical behaviour of<br />
composite structures submitted also to environmental conditions. However, every aspect of<br />
the consequences of environmental loading on the constituents of composite materials have<br />
not always been considered in this paper, for the sake of simplicity. Nevertheless, accounting<br />
for some additional physical factors would improve the realism and the reliability of the<br />
predictions obtained through the scale-transition models.<br />
For instance, the moisture diffusion process was assumed, in the present work, to follow<br />
the linear, classical, established for a long time, Fickian model. Nevertheless, some valuable<br />
experimental results, already reported in (Gillat and Broutman, 1978), have shown that<br />
certain anomalies in the moisture sorption process, (i.e. discrepancies from the expected<br />
Fickian behaviour) could be explained from basic principles of irreversible thermodynamics,<br />
by a strong coupling between the moisture transport in polymers and the local stress state<br />
(Weitsman, 1990a, Weitsman, 1990b).<br />
The present work yields several perspectives of research concerning the application of<br />
scale transition model to the identification of composite materials properties. Moisture and<br />
temperature are not the only parameters leading to an evolution of the mechanical properties<br />
of epoxies. According to the literature, thermo-oxidation is reported to enhance the stiffness<br />
of the epoxies (Decelle et al., 2003 ; Ho and al., 2006). The inverse methods presented here<br />
could for instance be directly applied to the estimation of the epoxy stiffening from the<br />
knowledge of the macroscopic elastic properties evolution as a function of the mass loss<br />
during the thermo-oxidation process. Furthermore, extensions of the inverse models could be<br />
achieved in order to account for the variation of the coefficients of thermal and/or moisture<br />
expansion of the constituents of a composite ply, enabling to identify them and their<br />
evolutions as a function of the environmental conditions. Finally, a similar approach could be<br />
developed in order to identify the damage induced evolution of the mechanical behaviour of<br />
the constituents of composite plies from the inelastic part of macroscopic stress/strain curves.<br />
The experimental data required for achieving such analysis is already available in the<br />
literature (Soden et al., 1998). Nevertheless, local and macroscopic damage have still to be
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 45<br />
implemented in the theoretical laws. The above-listed perspectives of research will be<br />
successively considered in further works.<br />
7.3. Improving the Calculation Time While Ensuring the Most Reliable<br />
Predictions<br />
The present work underlines the sometime existing opportunity to replace purely numerical<br />
mathematical solutions by analytical forms enabling to significantly reduce both the time<br />
required for designing the software and the time necessary for achieving one simulation. It<br />
was demonstrated in this paper that Eshelby-Kröner could be, at least partially, presented as<br />
an analytical model, while it was used for predicting mechanical states. Nevertheless, the<br />
estimation of the macroscopic properties (elastic stiffness, coefficients of thermal expansion<br />
and coefficients of moisture expansion) through the homogenization relations deduced from<br />
this very model do still involve an implicit iterative procedure. It was already shown in the<br />
literature by Welzel and his co-authors, that under specific conditions, it was possible to build<br />
a model, numerically equivalent to Eshelby-Kröner model, from the combination of two<br />
(separately less successful) other models (Welzel, 2002 ; Welzel et al. 2003). The concept is<br />
similar to the idea based on empirical comparisons, historically proposed by Neerfeld<br />
(Neerfeld, 1942) and Hill (Hill, 1952) to average Reuss and Voigt rough hypotheses in order<br />
to get a numerically acceptable theoretical solution. In the field of micro-mechanical<br />
modelling of composite materials, a combination of the two possible localization procedures<br />
considered for Mori-Tanaka model in the present work would enable to numerically<br />
reproduce the homogenized properties obtained from Eshelby-Kröner model. Building an<br />
effective model from the two main ways of writing Mori-Tanaka model would mainly enable<br />
to obtain closed-form solutions for the elastic stiffness tensor, instead of having to<br />
numerically solve the iterative procedure involved in Eshelby-Kröner self-consistent model.<br />
Thus, a coupling of this numerically effective solution for predicting realistic hygrothermomechanical<br />
macroscopic properties to the already proposed in this very article analytical<br />
forms for the local mechanical states would yield to a faster but still extremely reliable<br />
innovative scale-transition approach for studying composite materials. The analytical forms<br />
required for achieving the effective Mori-Tanaka model should be derived and published in<br />
the near future.<br />
References<br />
Agbossou, A., Pastor J. (1997). Thermal stresses and thermal expansion coefficients of nlayered<br />
fiber-reinforced composites, <strong>Composite</strong>s Science and Technology, 57: 249-260.<br />
Asaro, R. J. and Barnett, D. M. (1975). The non-uniform transformation strain problem for an<br />
anisotropic ellipsoidal inclusion, Journal of the Mechanics and Physics of Solids, 23:<br />
77-83.<br />
Baptiste, D. (1996). Evolution des contraintes dans les matériaux composites – Modélisation<br />
micromécanique du comportement des composites à renforts discontinus, In: Analyse des<br />
contraintes résiduelles par diffraction des rayons X et des neutrons, Alain Lodini and<br />
Michel Perrin (Editors).
46<br />
Jacquemin Frédéric and Fréour Sylvain<br />
Benveniste, Y. (1987). A New Approach to the Application of Mori-Tanaka’s Theory in<br />
<strong>Composite</strong> <strong>Materials</strong>, Mechanics of <strong>Materials</strong>, 6: 147-157.<br />
Bowles, D.E., Post, D., Herakovich, C.T., and Tenney, D.R. (1981). Moiré Interferometry for<br />
Thermal Expansion of <strong>Composite</strong>s, Exp. Mech., 21: 441-447.<br />
Breton, Y., Désarmot, G., Salvetat, J.P., Delpeux, S., Sinturel, C., Béguin, F. et Bonnamy, S.<br />
(2004). Mechanical properties of multiwall carbon nanotubes/epoxy composites :<br />
influence of network morphology, Carbon, 42: 1027-1030.<br />
Christensen, R. M. and Rinde, J. A. (1979). Transverse tensile characteristics of fiber<br />
composites with flexible resins : theory and test results, Pol. Eng. Sci., 19: 506-511.<br />
Crank, J. (1975). The mathematics of diffusion, Clarendon Press, Oxford.<br />
Cuntze, R. G. (2003). The predictive capability of failure mode concept-based strength<br />
criteria for multi-directional laminates – part B, <strong>Composite</strong> Science and Technology, 64:<br />
487-516.<br />
Daniel, I. M., and Ishai, O. (1994). In: “Engineering Mechanics of <strong>Composite</strong> <strong>Materials</strong>”,<br />
Oxford University Press.<br />
Decelle, J., Huet, N. and Bellenger, V. (2003). Oxidation induced shrinkage for thermally<br />
aged epoxy networks, Polymer Degradation and Stability, 81: 239-248.<br />
Demczyk, B.G., Wang, Y.M., Cumings, J., Hetman, M., Han, W., Zettl, A., and Ritchie, R.<br />
O. (2002). « Direct mechanical measurement of the tensile strength and elastic modulus<br />
of multiwalled carbon nanotubes », Mat. Sci. Eng., A334: 173-178.<br />
DiCarlo, J.A. (1986). Creep of chemically vapor deposited SiC fiber, J. Mater. Sci., 21:<br />
217-224.<br />
Dyer, S.R.A., Lord, D., Hutchinson, I.J., Ward, I.M. and Duckett, R.A. (1992). Elastic<br />
anisotropy in unidirectional fibre reinforced composites, J. Phys. D: Appl. Phys., 25:<br />
66-73.<br />
Eshelby, J. D. (1957). The Determination of the Elastic Field of an Ellipsoidal Inclusion, and<br />
Related Problems, In: Proceedings of the Royal Society London, A241: 376–396.<br />
Ferreira, C., Casari, P., Bouzidi, R. and Jacquemin, F. (2006a). Identification of Young<br />
Modulus Profile in PVC Foam Core thickness using speckle interferometry and Inverse<br />
Method, Proceedings of SPIE - The International Society for Optical Engineering.<br />
Ferreira, C., Jacquemin, F. and Casari, P. (2006b). Measurement of the nonuniform thermal<br />
expansion coefficient of a PVC foam core by speckle interferometry - Influence on the<br />
mechanical behavior of sandwich structures, Journal of Cellular Plastics, 42 (5):<br />
393-404.<br />
Fiedler, B., Hojo, M., Ochiai, S., Schulte, K. and Ando, M. (2001), Failure behaviour of an<br />
epoxy matrix under different kinds of static loading, <strong>Composite</strong> Science and Technology,<br />
61: 1615-1624.<br />
Fleck, N. A. and Jelf, P. M. (1995). Deformation and failure of a carbon fibre composite<br />
under combined shear and transverse loading, Acta metall. mater., 43: 3001-3007.<br />
François, M. (1991). Détermination de contraintes résiduelles sur des fils d’acier eutectoïde<br />
de faible diamètre par diffraction des rayons X, Doctoral Thesis, ENSAM, Paris.<br />
Fréour, S., Gloaguen, D., François, M., Guillén, R., Girard, E. and Bouillo, J. (2002)<br />
Determination of the macroscopic elastic constants of a phase embedded in a multiphase<br />
polycrystal – application to the beta-phase of Ti17 titanium based alloy, <strong>Materials</strong><br />
Science Forum, 404-407: 723-728.
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 47<br />
Fréour, S., Gloaguen, D., François, M. and Guillén, R. (2003a). Study of the Coefficients of<br />
Thermal Expansion of Phases Embedded in Multiphase <strong>Materials</strong>, Material Science<br />
Forum, 426–432: 2083–2088.<br />
Fréour, S., Gloaguen, D., François, M. and Guillén, R. (2003b). Thermal properties of<br />
polycrystals - X-ray diffraction and scale transition modelling, Physica Status Solidi a,<br />
201: 59-71.<br />
Fréour, S., Jacquemin, F. and Guillén, R. (2005a). On an analytical Self-Consistent model for<br />
internal stress prediction in fiber-reinforced composites submitted to hygro-elastic load,<br />
Journal of Reinforced Plastics and <strong>Composite</strong>s, 24: 1365-1377.<br />
Fréour, S., Gloaguen, D., François, M., Perronnet, A. and Guillén, R. (2005b). Estimation of<br />
Ti-17 β−phase Single-Crystal Elasticity Constants using X-Ray Diffraction<br />
measurements and inverse scale transition modelling, Journal of Applied<br />
Crystallography, 38: 30-37.<br />
Fréour, S., Jacquemin, F. and Guillén, R. (2006a). Extension of Mori-Tanaka Approach to<br />
Hygroelastic Loading of Fiber-Reinforced <strong>Composite</strong>s – Comparison with Eshelby-<br />
Kröner Self-consistent Model, Journal of Reinforced Plastics and <strong>Composite</strong>s, 25: 1039-<br />
1052.<br />
Fréour, S., Gloaguen, D., François, M. and Guillén, R. (2006b). Application of inverse<br />
models and XRD analysis to the determination of Ti-17 β−phase Coefficients of Thermal<br />
Expansion, Scripta Materialia, 54: 1475-1478.<br />
Fréour, S., Jacquemin, F. and Guillén, R. (to be published). On the use of the geometric mean<br />
approximation in estimating the effective hygro-elastic behaviour of fiber-reinforced<br />
composites, Journal of <strong>Materials</strong> Science.<br />
Frogley, M.D., D. Ravich, D., and Wagner, H.D. (2003). “Mechanical properties of carbon<br />
nanoparticle-reinforced elastomers”, Comp. Sci. Tech., 63: 1647-1654.<br />
Garett, K. W. and Bailey, J. E. (1977). The effect of resin failure strain on the tensile<br />
properties of glass fiber-reinforced polyester cross-ply laminates, J. Mater. Sci., 12:<br />
2189-2194.<br />
Gillat, O. and Broutman, L.J. (1978). “Effect of External Stress on Moisture Diffusion and<br />
Degradation in a Graphite Reinforced Epoxy Laminate”, ASTM STP, 658: 61-83.<br />
Gloaguen, D., François, M., Guillén, R. and Royer, J. (2002). Evolution of Internal Stresses in<br />
Rolled Zr702, Acta Materialia, 50: 871–880.<br />
Han, J., Bertram, A., Olschewski, J., Hermann, W. and Sockel, H.G. (1995). Identification of<br />
elastic constants of alloys with sheet and fibre textures based on resonance measurements<br />
and finite element analysis. <strong>Materials</strong> Science and Engineering, A191: 105-111.<br />
Herakovitch, C. T. (1998). Mechanics of Fibrous <strong>Composite</strong>s, John Wiley and Sons Inc.,<br />
New York.<br />
Hill, R. (1952). The elastic behaviour of a crystalline aggregate. Proc. Phys. Soc., 65:<br />
349-354.<br />
Hill, R., (1965). Continuum micro-mechanics of elastoplastic polycrystals, J. Mech. Phys.<br />
Solids, 13: 89-101.<br />
Hill, R. (1967). The essential structure of constitutive laws for metals composites and<br />
polycrystals, Journal of the Mechanics and Physics of Solids, 15: 79-95.
48<br />
Jacquemin Frédéric and Fréour Sylvain<br />
Ho, N.Q., Pons, F. and Lafarie-Frenot, M.C. (2006). Characterization of an Oxidized Layer in<br />
Epoxy Resin and in Carbon Epoxy <strong>Composite</strong> for Aeronautic Applications, Proceedings<br />
of ECCM 12.<br />
Jacquemin, F. and Vautrin, A. (2002). A Closed-form Solution for the Internal Stresses in<br />
Thick <strong>Composite</strong> Cylinders Induced by Cyclical Environmental Conditions, <strong>Composite</strong><br />
Structures, 58: 1–9.<br />
Jacquemin, F., Fréour, S., and Guillén, R. (2005). A self-consistent approach for transient<br />
hygroscopic stresses and moisture expansion coefficients of fiber-reinforced composites,<br />
Journal of Reinforced Plastics and <strong>Composite</strong>s, 24: 485-502.<br />
Kaddour, A. S., Hinton, M. J. and Soden, P. D. (2003). Behaviour of ± 45° glass/epoxy<br />
filament wound composite tubes under quasi-static equal biaxial tension-compression<br />
loading: experimental results, <strong>Composite</strong>s : part B, 34: 689-704.<br />
Karakazu, R., Atas, C. and Akbulut, H. (2001). Elastic-plastic Behaviour of Woven-steelfiber-reinforced<br />
Thermoplastic Laminated Plates under In-plane Loading, <strong>Composite</strong>s<br />
Science and Technology, 61: 1475-1483.<br />
Khashaba, U. A., (2004). In-plane shear properties of cross-ply composite laminates with<br />
different off-axis angles, <strong>Composite</strong> Structures, 65: 167-177.<br />
Kocks, U. F., Tomé , C. N. and Wenk, H. R. (1998). Texture and Anisotropy, Cambridge<br />
University Press.<br />
Kröner E. (1953). Dissertation, Technischen Hochschule Stuttgart.<br />
Kröner, E. (1958). “Berechnung der elastischen Konstanten des Vielkristalls aus des<br />
Konstanten des Einkristalls”, Zeitschrift für Physik, 151: 504–518.<br />
Liu, K-S and Tsai, S. W. (1998). A progressive quadratic failure criterion for a laminate,<br />
<strong>Composite</strong> Science and Technology, 58: 1023-1032.<br />
Liu, L. et Wagner, H.D. (2005). Rubery and gassy epoxy resins reinforced with carbon<br />
nanotubes, Comp. Sci. Tech., 65: 1865-1868.<br />
Loos, A. C. and Springer, G. S. (1981). Environmental Effects on <strong>Composite</strong> <strong>Materials</strong>,<br />
Moisture Absorption of Graphite – Epoxy Composition Immersed in Liquids and in<br />
Humid Air, pp. 34–55, Technomic Publishing.<br />
Mabelly, P. (1996). Contribution à l’étude des pics de diffraction – Approche expérimentale<br />
et modélisation micromécanique, Doctoral Thesis, ENSAM, Aix en Provence.<br />
Matthies, S., and Humbert, M. (1993). The realization of the concept of a geometric mean for<br />
calculating physical constants of polycrystalline materials, Phys. Stat. Sol. b, 177: K47-<br />
K50.<br />
Matthies, S., Humbert, M., and Schuman, Ch. (1994). On the use of the geometric mean<br />
approximation in residual stress analysis”, Phys. Stat. Sol. b, 186 : K41-K44.<br />
Matthies, S. Merkel, S., Wenk, H.R., Hemley, R.J. and Mao, H. (2001). Effects of texture on<br />
the determination of elasticity of polycrystalline ε-iron from diffraction measurements,<br />
Earth and Planetary Science Letters, 194: 201-212.<br />
Mensitieri, G. M., Del Nobile, M. A., Apicella, A. and Nicolais, L. (1995). Moisture-matrix<br />
interactions in polymer based composite materials, Revue de l’Institut Français du<br />
Pétrole, 50: 551-571.<br />
Morawiec, A. (1989). Calculation of polycrystal elastic constants from single-crystal data,<br />
Phys. Stat. Sol. b, 154 : 535-541.
Multi-scale Analysis of Fiber-Reinforced <strong>Composite</strong> Parts… 49<br />
Mori, T. and Tanaka, K. (1973). Average Stress in Matrix and Average Elastic Energy of<br />
<strong>Materials</strong> with Misfitting Inclusions, Acta Metallurgica, 21: 571-574.<br />
Morris, R. (1970). Elastic constants of polycrystals, Int. J. Eng. Sci., 8: 49.<br />
Mura, T. (1982). Micromechanics of Defects in Solids, Martinus Nijhoff Publishers, The<br />
Hague, Netherlands.<br />
Neerfeld, H. (1942). Zur Spannungsberechnung aus röntgenographischen<br />
Dehnungsmessungen, Mitt. Kaiser-Wilhelm-Inst. Eisenforschung Düsseldorf 24: 61-70.<br />
Qiu, Y. P. and Weng, G. J. (1991). The influence of inclusion shape on the overall<br />
elastoplastic behavior of a two-phase isotropic composite, Int. J. Solids Structures, 27<br />
(12): 1537-1550.<br />
Reuss, A. (1929). Berechnung der Fliessgrenze von Mischkristallen auf Grund der<br />
Plastizitätsbedingung für Einkristalle, Zeitschrift für Angewandte und Mathematik und<br />
Mechanik, 9: 49–58.<br />
Sai Ram K.S., Sinha, P.K. (1991). Hygrothermal effects on the bending charactericstics of<br />
laminated composite plates, Computational Structure, 40 (4): 1009-1015.<br />
Salvetat, J.-P., Andrews, G., Briggs, D., Bonard, J.-M., Bacsa, R.R., Kulik, A.J., Stöckli, T.,<br />
Burnham, N.A., and Forro, L. (1999). “Elastic and Shear Moduli of Single-Walled<br />
Carbon Nanotube Ropes”, Phys. Rev. Lett., 82: 944-947.<br />
Sims, G.D., Dean, G.D., Read, B.E. and Western B.C. (1977). Assessment of Damage in<br />
GRP Laminates by Stress Wave Emission and Dynamic Mechanical Measurements,<br />
Journal of <strong>Materials</strong> Science, 12 (11): 2329-2342.<br />
Soden, P. D., Hinton M. J. and A. S. Kaddour, A. S. (1998). Lamina properties lay-up<br />
configurations and loading conditions for a range of fiber-reinforced composite<br />
laminates, <strong>Composite</strong>s Science and Technology, 58: 1011-1022.<br />
Sprauel, J. M., and Castex, L. (1991). First European Powder Diffraction International<br />
Conference on X-Ray stress analysis, Munich.<br />
Tanaka, K. and Mori, T. (1970). The Hardening of Crystals by Non-deforming Particules and<br />
Fibers, Acta Metallurgica, 18: 931-941.<br />
Treacy, M.M.J., Ebbesen, T.W. and Gibson, T.M. (1996). “Exceptionally high Young’s<br />
modulus observed for individual carbon nanotubes”, Nature, 381: 678-680.<br />
Tsai, C.L. and Daniel I.M. (1993). Measurement of longitudinal shear modulus of single<br />
fibers by means of a torsional pendulum. 38th International SAMPE Symposium<br />
1993:1861-1868.<br />
Tsai C.L. and Daniel I.M. (1994). Method for thermo-mechanical characterization of single<br />
fibers, <strong>Composite</strong>s Science and Technology, 50: 7-12.<br />
Tsai, C.L. and Chiang, C.H. (2000). Characterization of the hygric behavior of single fibers.<br />
<strong>Composite</strong>s Science and Technology, 60: 2725-2729<br />
Tsai, S. W. and Hahn, H. T. (1980). Introduction to composite materials, Technomic<br />
Publishing Co., Inc., Lancaster, Pennsylvania.<br />
Tsai, S. W. (1987). <strong>Composite</strong> Design, 3rd edn, Think <strong>Composite</strong>s.<br />
Turner, P. A. and Tome, C. N. (1994). “A Study of Residual Stresses in Zircaloy-2 with Rod<br />
Texture”, Acta Metallurgica and Materialia, 42: 4143–4153.<br />
Voigt, W. (1928). Lehrbuch der Kristallphysik, Teubner, Leipzig/Berlin.<br />
Weitsman, Y. (1990a). “A Continuum Diffusion Model for Viscoelastic <strong>Materials</strong>”, Journal<br />
of Physical Chemistry, 94: 961-968.
50<br />
Jacquemin Frédéric and Fréour Sylvain<br />
Weitsman, Y. (1990b). “Moisture in <strong>Composite</strong>s: Sorption and Damage”, in: Fatigue of<br />
<strong>Composite</strong> <strong>Materials</strong>. Elsevier Science Publisher, K.L. Reifsnider (editor), 385-429.<br />
Welzel, U. (2002). “Diffraction Analysis of Residual Stress; Modelling Elastic Grain<br />
Interaction.”, PhD thesis, University of Stuttgart, Germany.<br />
Welzel, U., Leoni, M. and Mittemeijer, E. J. (2003). « The determination of stresses in thin<br />
films; modelling elastic grain interaction », Philosophical Magazine, 83: 603-630.<br />
Welzel, U., Fréour, S. and Mittemeijer, E. J. (2005). « Direction-dependent elastic graininteraction<br />
models – a comparative study », Philosophical Magazine, 85: 2391-2414.<br />
Xiao, K.Q., Zhang, L.C. et Zarudi, I. (2006). « Mechanical and rheological properties of<br />
carbon nanotube-reinforced polyethylene composites », Comp. Sci. Tech., 67: 177-182.
In: <strong>Composite</strong> <strong>Materials</strong> <strong>Research</strong> <strong>Progress</strong> ISBN: 1-60021-994-2<br />
Editor: Lucas P. Durand, pp. 51-107 © 2008 Nova Science Publishers, Inc.<br />
Chapter 2<br />
OPTIMIZATION OF LAMINATED COMPOSITE<br />
STRUCTURES: PROBLEMS, SOLUTION PROCEDURES<br />
AND APPLICATIONS<br />
Michaël Bruyneel<br />
SAMTECH s.a., Liège Science Park<br />
Rue des Chasseurs-ardennais 8, 4031 Angleur, Belgium<br />
Abstract<br />
In this chapter the optimal design of laminated composite structures is considered. A<br />
review of the literature is proposed. It aims at giving a general overview of the problems that a<br />
designer must face when he works with laminated composite structures and the specific<br />
solutions that have been derived. Based on it and on the industrial needs an optimization<br />
method specially devoted to composite structures is developed and presented. The related<br />
solution procedure is general and reliable. It is based on fibers orientations and ply thicknesses<br />
as design variables. It is used daily in an (European) industrial context for the design of<br />
composite aircraft box structures located in the wings, the center wing box, and the vertical<br />
and horizontal tail plane. This approach is based on sequential convex programming and<br />
consists in replacing the original optimization problem by a sequence of approximated subproblems.<br />
A very general and self adaptive approximation scheme is used. It can consider the<br />
particular structure of the mechanical responses of composites, which can be of a different<br />
nature when both fiber orientations and plies thickness are design variables. Several numerical<br />
applications illustrate the efficiency of the proposed approach.<br />
1. Introduction<br />
According to their high stiffness and strength-to-weight ratios, composite materials are well<br />
suited for high-tech aeronautics applications. A large amount of parameters is needed to<br />
qualify a composite construction, e.g. the stacking sequence, the plies thickness and the fibers<br />
orientations. It results that the use of optimization techniques is necessary, especially to tailor<br />
the material to specific structural needs. The chapter will cover this subject and is divided in<br />
three main parts.
52<br />
Michaël Bruyneel<br />
After recalling the goal of optimization, the different laminates parameterizations will be<br />
presented with their limitations (the pros and the cons) in the frame of the optimal design of<br />
composite structures. The issues linked to the modeling of structures made of such materials<br />
and the problems solved in the literature will be reviewed. The key role of fibers orientations<br />
in the resulting laminate properties will be discussed. Finally the outlines of a pragmatic<br />
solution procedure for industrial applications will be drawn. Throughout this section, a<br />
profuse and state-of-the-art review of the literature will be provided.<br />
Secondly, a general solution procedure used daily in industrial problems including fibers<br />
reinforced composite materials will be described. The related optimization algorithm is based<br />
on sequential convex programming and has proven to be very reliable. This algorithm is<br />
presented in detail and validated by comparing its performances to other optimization<br />
methods of the literature.<br />
Finally, it will be shown how this optimization algorithm can efficiently solve several<br />
kinds of composite structure design problems: amongst others, solutions for topology<br />
optimization with orthotropic materials will be presented, important considerations about the<br />
optimal design of composites including buckling criteria will be discussed, optimization with<br />
respect to damage tolerance will be considered (crack delamination in a laminated structure).<br />
On top of that, some key points of the solution procedure based on this optimization<br />
algorithm applied to the pre-sizing of (European) industrial composite aircraft box structures<br />
will be presented.<br />
2. The Optimal Design Problem and Available Optimization<br />
Methods<br />
The goal of optimization is to reach the best solution of a problem under some restrictions. Its<br />
mathematical formulation is given in (2.1), where g0(x) is the objective function to be<br />
minimized, gj(x) are the constraints to be satisfied at the solution, and x={xi, i=1,…,n} is the<br />
set of design variables. The value of those design variables change during the optimization<br />
process but are limited by an upper and a lower bound when they are continuous, which will<br />
be the case in the sequel.<br />
min g ( x )<br />
0<br />
max<br />
g j ( x ) ≤ g j j = 1,...,<br />
m<br />
(2.1)<br />
xi ≤ xi<br />
≤ xi<br />
i<br />
= 1,...,<br />
The problem (2.1) is illustrated in Figure 2.1, where 2 design variables x1 and x2 are<br />
considered. The isovalues of the objective function are drawn, as well as the limiting values<br />
of the constraints. The solution is found via an iterative process. x k is the vector of design<br />
variables at the current iteration k, and x k+1 is the estimation of the solution at the iteration<br />
k+1. Typically a local solution xlocal will be reached when a gradient based optimization<br />
method is used. The best solution xglobal can only be found when all the design space is looked<br />
over: this last can be accessed with specific optimization methods that include a non<br />
deterministic procedure, as the genetic algorithms.<br />
n
Optimization of Laminated <strong>Composite</strong> Structures… 53<br />
x2<br />
* global<br />
X<br />
a<br />
(k)<br />
S<br />
no<br />
X<br />
( k+<br />
1)<br />
b<br />
(k)<br />
X<br />
α<br />
Initial design<br />
Structural analysis<br />
Optimization<br />
New design<br />
Optimal design ?<br />
yes<br />
End<br />
* local<br />
X<br />
g j (X)<br />
Figure 2.1. Illustration of an optimization problem and its solution.<br />
In structural optimization, the design functions can be global as the weight, the stiffness,<br />
the vibration frequencies, the buckling loads, or local as strength constraints, strains and<br />
failure criteria. When the design variables are linked to the transverse properties of the<br />
structural members (e.g. the cross-section area of a bar in a truss), the related optimization<br />
problem is called optimal sizing (Figure 2.2a). The value of some geometric items (e.g. a<br />
radius of an ellipse) can also be variable: in this case, we are talking about shape optimization<br />
(Figure 2.2b). Topology optimization aims at spreading a given amount of material in the<br />
structure for a maximum stiffness. Here, holes can be automatically created during the<br />
optimization process (Figure 2.2c). Finally, the optimization of the material can be addressed,<br />
e.g. the local design of laminated composite structure with respect to fibers orientations, ply<br />
thickness and stacking sequence (Figure 2.2d).<br />
x1
54<br />
Initial<br />
designs<br />
Final designs<br />
a) Optimal sizing b) Shape<br />
optimization<br />
Michaël Bruyneel<br />
c) Topology<br />
optimization<br />
Figure 2.2. The structural optimization problems.<br />
d) Material<br />
optimization<br />
The structural optimization problems are non linear and non convex, and several local<br />
minimum exist. It is usually accepted that a local solution xlocal gives satisfaction. The global<br />
solution xglobal can only be determined with very large computational resources. In some cases<br />
when the problem includes a very large amount of constraints, a feasible solution is<br />
acceptable.<br />
A lot of methods exist to solve the problem (2.1). Morris (1982), Vanderplaats (1984),<br />
and Haftka and Gurdal (1992) present techniques based on the mathematical programming<br />
approach used in structural optimization. Most of them are compared by Barthelemy and<br />
Haftka (1993), and Schittkowski et al. (1994). Non deterministic methods, such as the genetic<br />
algorithm (Goldberg, 1989), are studied by Potgieter and Stander (1998), and Arora et al.<br />
(1995). Those authors also present a review of the methods used in global optimization.<br />
Optimality criteria for the specific solution of fibers optimal orientations in membrane<br />
(Pedersen 1989) and in plates (Krog 1996) must be mentioned as well. Finally the response<br />
surfaces methods are also used for optimizing laminated structures (Harrison et al. 1995, Liu<br />
et al. 2000, Rikards et al. 2006, Lanzi and Giavotto 2006).<br />
The approximation concepts approach, also called Sequential Convex Programming,<br />
developed in the seventies by Fleury (1973), Schmit and Farschi (1974), and Schmit and<br />
Fleury (1980) has allowed to efficiently solve several structural optimization problems: the<br />
optimal sizing of trusses, shape optimization (Braibant and Fleury, 1985), topology<br />
optimization (Duysinx, 1996, 1997, and Duysinx and Bendsøe, 1998), composite structures<br />
optimization (Bruyneel and Fleury 2002, Bruyneel 2006), as well as multidisciplinary<br />
optimization problems (Zhang et al., 1995 and Sigmund, 2001). In sizing and shape<br />
optimization the solution is usually reached within 10 iterations. For topology optimization,<br />
since a very large number of design variables are included in the problem, a larger number of<br />
design cycles is needed for converging with respect to stabilized design variables values over<br />
2 iterations.<br />
Those approximation methods consist in replacing the solution of the initial optimization<br />
problem (2.1) by the solution of a sequence of approximated optimization problems, as<br />
illustrated in Figure 2.3.
Optimization of Laminated <strong>Composite</strong> Structures… 55<br />
x2<br />
(k)<br />
*<br />
X<br />
* global<br />
X<br />
no<br />
(k)<br />
X<br />
Initial design<br />
End<br />
yes<br />
* local<br />
X<br />
g j (X)<br />
Approximated optimization problem<br />
Solution of the approximated problem<br />
Optimal design ?<br />
~ ( k)<br />
g j ( X)<br />
Figure 2.3. Definition of an approximated optimization problem based on the information at the current<br />
design point x (k) . The corresponding feasible domain is defined by the constraints of (2.2).<br />
Each function entering the problem (2.1) is replaced by a convex approximation<br />
~ ( k)<br />
g j ( X)<br />
based on a Taylor series expansion in terms of the direct design variables x i or<br />
intermediate ones as for example the inverse design variables 1 xi<br />
. For a current design x (k)<br />
at iteration k, the approximated optimization problem writes:<br />
~<br />
~ ( k)<br />
min g0<br />
( x)<br />
( k)<br />
max<br />
g j ( x ) ≤ g j<br />
j = 1,...,<br />
m<br />
(2.2)<br />
( k)<br />
( k)<br />
xi ≤ xi<br />
≤ xi<br />
i = 1,...,<br />
n<br />
where the symbol ~ is related to an approximated function. The explicit and convex<br />
optimization problem (2.2) is itself solved by dedicated methods of mathematical<br />
x1
56<br />
Michaël Bruyneel<br />
programming (see Section 7). Building an approximated problem requires to carry out a<br />
structural and a sensitivity analyses (via the finite elements method). Solving the related<br />
explicit problem does no longer necessitate a finite element analysis (expensive in CPU for<br />
large scale problems).<br />
The solution obtained with this approach doesn’t correspond to the global optimum, but<br />
to a local one, since gradients and deterministic information are used. Nevertheless this local<br />
solution is found very quickly and several initial designs could be used to try to find a better<br />
solution, as proposed by Cheng (1986). Finally it must be noted that when a very large<br />
number of constraints is considered in the optimal design problem (say more than 10 5 ) the<br />
user is often satisfied with a feasible solution.<br />
3. Parameterizations of Laminated <strong>Composite</strong> Structures<br />
Before presenting the several possible parameterizations of laminates, with their advantages<br />
and their disadvantages, the classical lamination theory is briefly recalled in order to<br />
introduce the notation that will be used throughout the chapter. See Tsai and Hahn (1980),<br />
Gay (1991) and Berthelot (1992) for details.<br />
3.1. The Classical Lamination Theory<br />
3.1.1. Constitutive Relations for a Ply<br />
Fibers reinforced composite materials are orthotropic along the fibers direction, that is in the<br />
local material axes (x,y,z) illustrated in Figure 3.1. Homogeneous macroscopic properties are<br />
assumed at the ply and at the laminate levels.<br />
y<br />
z,3<br />
2<br />
x<br />
θ<br />
1<br />
Material axes<br />
(orthotropy)<br />
Figure 3.1. The unidirectional ply with its material and structural axes.<br />
For a linear elastic behaviour, the stress-strain relations in the material axes are given by<br />
the Hook’s law Qε<br />
σ = where ε and σ are the strain and stress tensors, respectively, while<br />
Q is the matrix collecting the stiffness coefficients in the orthotropic axes. For a plane stress<br />
assumption, it comes that
⎧σ<br />
x ⎫ ⎡ mE<br />
⎪ ⎪ ⎢<br />
x<br />
⎨σ<br />
y ⎬ = ⎢mν<br />
xy E y<br />
⎪ ⎪ ⎢<br />
⎩<br />
σ xy ⎭ ⎣<br />
0<br />
Optimization of Laminated <strong>Composite</strong> Structures… 57<br />
mν<br />
yxE<br />
x<br />
mEy<br />
0<br />
0 ⎤⎧<br />
ε x ⎫ ⎡Qxx<br />
⎥⎪<br />
⎪<br />
0 =<br />
⎢<br />
⎥⎨<br />
ε y ⎬ ⎢<br />
Qyx<br />
G ⎥⎪<br />
⎪<br />
xy ⎢<br />
⎦⎩<br />
γ xy ⎭ ⎣ 0<br />
Qxy<br />
Qyy<br />
0<br />
0 ⎤⎧<br />
ε x ⎫<br />
⎪ ⎪<br />
0<br />
⎥<br />
⎥⎨<br />
ε ⎬ ,<br />
1<br />
y m =<br />
Q ⎥⎪<br />
⎪ 1 −ν<br />
xyν<br />
ss ⎦⎩<br />
γ xy ⎭<br />
yx<br />
(3.1)<br />
The stresses and strains can be written in the structural coordinates (1,2,3) as in (3.2) and<br />
(3.3) where θ is the angle between the local and structural axes, defined in Figure 3.1.<br />
⎡ 2<br />
⎧ε1<br />
⎫ cos θ<br />
⎪ ⎪ ⎢ 2<br />
⎨ε<br />
2 ⎬ = ⎢ sin θ<br />
⎪ ⎪ ⎢<br />
⎩ε<br />
6 ⎭ ⎢<br />
2 cosθ<br />
sinθ<br />
⎣<br />
⎡ 2<br />
⎧σ1<br />
⎫ cos θ<br />
⎪ ⎪ ⎢ 2<br />
⎨σ<br />
2 ⎬ = ⎢ sin θ<br />
⎪ ⎪ ⎢<br />
⎩σ<br />
6 ⎭ ⎢<br />
cosθ<br />
sinθ<br />
⎣<br />
sin<br />
cos<br />
2<br />
2<br />
θ<br />
θ<br />
− 2 cosθ<br />
sinθ<br />
sin<br />
cos<br />
2<br />
2<br />
θ<br />
θ<br />
− cosθ<br />
sinθ<br />
− cosθ<br />
sinθ<br />
⎤ ⎧ ε x ⎫<br />
⎥ ⎪ ⎪<br />
cosθ<br />
sinθ<br />
⎥ ⎨ ε y ⎬<br />
2 2<br />
cos θ − sin θ ⎥ ⎪ ⎪<br />
⎥⎦<br />
⎩<br />
γ xy ⎭<br />
− 2cosθ<br />
sinθ<br />
⎤ ⎧σ<br />
x ⎫<br />
⎥ ⎪ ⎪<br />
2 cosθ<br />
sinθ<br />
⎥ ⎨σ<br />
y ⎬<br />
2 2<br />
cos θ − sin θ ⎥ ⎪ ⎪<br />
⎥⎦<br />
⎩<br />
σ xy ⎭<br />
(3.2)<br />
(3.3)<br />
For a ply with an orientation θ with respect to the structural axes, the constitutive<br />
relations write:<br />
⎧σ1<br />
⎫ ⎡Q<br />
⎪ ⎪<br />
=<br />
⎢<br />
⎨σ<br />
2 ⎬ ⎢<br />
Q<br />
⎪ ⎪<br />
⎩σ<br />
6 ⎭ ⎢⎣<br />
Q<br />
11<br />
12<br />
16<br />
Q<br />
Q<br />
Q<br />
12<br />
22<br />
26<br />
Q<br />
Q<br />
Q<br />
16<br />
26<br />
66<br />
⎤⎧ε1<br />
⎫<br />
⎥⎪<br />
⎪<br />
⎥⎨ε<br />
2 ⎬<br />
⎥⎪<br />
⎪<br />
⎦⎩ε<br />
6 ⎭<br />
where the matrix of the stiffness coefficients in the structural axes takes the form:<br />
⎡<br />
(3.4)<br />
⎧Q<br />
4 4 2 2<br />
2 2<br />
11 ⎫ c s 2c<br />
s 4c<br />
s<br />
⎪<br />
Q<br />
⎪ ⎢ 4 4 2 2<br />
2 2 ⎥<br />
⎪ 22 ⎪ ⎢ s c 2c<br />
s 4c<br />
s ⎥ ⎧Qxx<br />
⎫<br />
⎪Q<br />
2 2 2 2 4 4<br />
2 2<br />
12 ⎪ ⎢<br />
⎪<br />
4 Q<br />
⎪<br />
c s c s c + s − c s ⎥ ⎪ yy ⎪<br />
Q ( 1,<br />
2,<br />
3)<br />
= ⎨ ⎬ = ⎢<br />
⎥<br />
2 2 2 2 2 2 2 2 2 ⎨ ⎬ (3.5)<br />
⎪Q66<br />
⎪ ⎢c<br />
s c s − 2c<br />
s ( c − s ) ⎥ ⎪Qxy<br />
⎪<br />
⎪Q<br />
⎪ ⎢ 3 3 3 3 3 3 ⎥<br />
16 c s cs cs c s 2(<br />
cs c s)<br />
⎪<br />
⎩Q<br />
⎪<br />
⎢<br />
− −<br />
−<br />
⎥ ss ⎭<br />
⎪ ⎪<br />
( x,<br />
y,<br />
z)<br />
⎪Q<br />
3 3 3 3 3 3<br />
⎩ 26 ⎪⎭<br />
⎢<br />
( 1,<br />
2,<br />
3)<br />
⎣ cs − c s ( c s − cs ) 2(<br />
c s − cs ) ⎥⎦<br />
with<br />
c = cos θ s = sinθ<br />
The variation of the Q’s with respect to the angle θ is plotted in Figure 3.2. It is observed<br />
that the stiffness coefficients are highly non linear in terms of the fibers orientation.<br />
⎤
58<br />
Michaël Bruyneel<br />
Figure 3.2. Stiffness coefficients in N/mm² in the structural axes for several values of the fibers<br />
orientation in a carbon/epoxy material T300/5208 (after Tsai and Hahn, 1980).<br />
Based on the fact that the trigonometric functions entering the matrix in (3.5) can be<br />
written in the following way:<br />
4 1<br />
cos θ = ( 3 + 4cos<br />
2θ<br />
+ cos 4θ<br />
)<br />
8<br />
3 1<br />
cos θ sinθ<br />
= ( 2sin<br />
2θ<br />
+ sin 4θ<br />
)<br />
8<br />
2 2 1<br />
cos θ sin θ = ( 1 − cos 4θ<br />
)<br />
8<br />
3 1<br />
cosθ<br />
sin θ = ( 2sin<br />
2θ<br />
− sin 4θ<br />
)<br />
8<br />
4 1<br />
sin θ = ( 3 − 4 cos 2θ<br />
+ cos 4θ<br />
)<br />
8<br />
(3.6)<br />
Tsai and Pagano (1968) derived an alternative expression for the Q’s coefficients in the<br />
structural axes given in (3.7):<br />
⎡Q11<br />
Q12<br />
Q16<br />
⎤<br />
Q ( 1,<br />
2,<br />
3)<br />
=<br />
⎢<br />
Q22<br />
Q<br />
⎥<br />
26 = γ0<br />
+ γ1<br />
cos2θ + γ2<br />
cos4θ<br />
+ γ3<br />
sin 2θ<br />
+ γ4<br />
sin 4θ<br />
⎢<br />
⎥<br />
⎢⎣<br />
sym Q66⎥⎦<br />
where the parameters γ are functions of the lamina invariants U1-U5:<br />
(3.7)
and<br />
γ<br />
2<br />
γ<br />
0<br />
Optimization of Laminated <strong>Composite</strong> Structures… 59<br />
⎡ U1<br />
=<br />
⎢<br />
⎢<br />
⎢⎣<br />
sym<br />
⎡ U3<br />
=<br />
⎢<br />
⎢<br />
⎢⎣<br />
sym<br />
U<br />
U<br />
−U<br />
U<br />
3<br />
3<br />
4<br />
1<br />
0 ⎤<br />
0<br />
⎥<br />
⎥<br />
U5⎥⎦<br />
0 ⎤<br />
0<br />
⎥<br />
⎥<br />
−U<br />
3⎥⎦<br />
1<br />
U1<br />
= ( 3Qxx<br />
+ 3Q<br />
yy + 2Qxy<br />
+ 4Qss<br />
)<br />
8<br />
1<br />
U2<br />
= ( Qxx<br />
− Qyy<br />
)<br />
2<br />
1<br />
U3<br />
= ( Qxx<br />
+ Qyy<br />
− 2Qxy<br />
− 4Qss<br />
)<br />
8<br />
⎡<br />
⎢<br />
0<br />
⎢<br />
γ 3 = ⎢<br />
⎢<br />
⎢sym<br />
⎢⎣<br />
3.1.2. Constitutive Relations for a Laminate<br />
⎡ U2<br />
0 0⎤<br />
γ<br />
⎢<br />
⎥<br />
1 = −U<br />
0<br />
⎢ 2<br />
(3.8)<br />
⎥<br />
⎢⎣<br />
sym 0⎥⎦<br />
0<br />
0<br />
U2<br />
⎤<br />
2 ⎥<br />
U ⎥<br />
2<br />
⎥<br />
2 ⎥<br />
0 ⎥<br />
⎥⎦<br />
⎡ 0<br />
γ<br />
⎢<br />
4 =<br />
⎢<br />
⎢⎣<br />
sym<br />
0<br />
0<br />
U3<br />
⎤<br />
−U<br />
⎥<br />
3⎥<br />
0 ⎥⎦<br />
1<br />
U4<br />
= ( Qxx<br />
+ Qyy<br />
+ 6Qxy<br />
− 4Qss<br />
)<br />
8<br />
1<br />
U5<br />
= ( Qxx<br />
+ Qyy<br />
− 2Qxy<br />
+ 4Qss<br />
)<br />
8<br />
<strong>Composite</strong> structures are thin membranes, plates or shells made of n unidirectional<br />
orthotropic plies stacked on the top of each other. Such structures can support in and out-of<br />
plane loadings. In the following the constitutive relations for a laminate made of several<br />
individual plies are derived. The notations are defined in Figure 3.3. In the case of plane<br />
stress, i.e. the effects of transverse shear is neglected, in-plane normal and shear loads N, as<br />
well as the flexural and torsional moments M are applied to the laminate. Those loadings are<br />
computed by considering the stress state in each ply with the relations (3.9):<br />
⎧N1<br />
⎫ ⎧σ<br />
⎫<br />
h / 2 1<br />
⎧M1<br />
⎫ ⎧σ<br />
⎫<br />
h / 2 1<br />
⎪ ⎪ ⎪ ⎪<br />
⎪ ⎪ ⎪ ⎪<br />
N = ⎨N<br />
2 ⎬ = ∫ ⎨σ<br />
2 ⎬dz<br />
M = ⎨M<br />
2 ⎬ = ∫ ⎨σ<br />
2 ⎬zdz<br />
(3.9)<br />
⎪N<br />
⎪ −h / 2 ⎪ ⎪<br />
⎩ 6 ⎭ ⎩σ<br />
⎪<br />
6 ⎭<br />
M ⎪ −h / 2 ⎪ ⎪<br />
⎩ 6 ⎭ ⎩σ<br />
6 ⎭<br />
For a first order cinematic theory, where the displacement through the laminate’s<br />
thickness is linear in the z coordinate measured with respect to the mid-plane of the plate/shell<br />
(Figure 3.3), the vector of laminate’s strains εl is linked to the in-plane strains and the<br />
curvatures via the relation εl = ε + zκ<br />
0<br />
. With this definition it turns that the constitutive<br />
relations for a laminate are given by (3.10) where A, B and D are the in-plane, coupling and<br />
bending stiffness matrices of the laminate.
60<br />
⎧ N ⎫ ⎡A<br />
⎨ ⎬ = ⎢<br />
⎩M⎭<br />
⎣B<br />
z k<br />
3<br />
n<br />
k<br />
2<br />
1<br />
⎧ N1<br />
⎫ ⎡ A11<br />
⎪ ⎪ ⎢<br />
⎪<br />
N2<br />
A<br />
⎪ ⎢ 12<br />
B⎤⎪⎧<br />
0<br />
ε ⎪⎫<br />
⎪ N6<br />
⎪ ⎢A16<br />
⎥⎨<br />
⎬ ⇔ ⎨ ⎬ = ⎢<br />
D⎦⎪⎩<br />
κ ⎪⎭ ⎪M1<br />
⎪ ⎢<br />
B11<br />
⎪M<br />
2 ⎪ ⎢B12<br />
⎪ ⎪ ⎢<br />
⎩M<br />
6 ⎭ ⎣B16<br />
Michaël Bruyneel<br />
A12<br />
A22<br />
A26<br />
B12<br />
B22<br />
B26<br />
A16<br />
A26<br />
A66<br />
B16<br />
B26<br />
B66<br />
B11<br />
B12<br />
B16<br />
D11<br />
D12<br />
D16<br />
B12<br />
B22<br />
B26<br />
D12<br />
D22<br />
D26<br />
(a) A laminate with its structural axes. h is the total thickness<br />
(b) Several unidirectional plies stacked on top of each other.<br />
Material axes related to the k th ply .<br />
t k<br />
B ⎤⎧<br />
0<br />
16 ε ⎫<br />
1<br />
⎪ ⎪<br />
B<br />
⎥ 0<br />
26 ⎥⎪ε<br />
2 ⎪<br />
B ⎥⎪<br />
0 ⎪<br />
66 ε6<br />
⎥⎨<br />
⎬<br />
D16<br />
⎥⎪κ1<br />
⎪<br />
D26⎥⎪κ<br />
⎪<br />
2<br />
⎥⎪<br />
⎪<br />
D66⎦⎪⎩<br />
κ6<br />
⎪⎭<br />
(c). Definition of the plies location through the laminate’s thickness.<br />
hk and hk-1 are used to locate the k th ply of the stacking sequence<br />
h k-1<br />
h 2<br />
h 1<br />
h k<br />
h 0<br />
(3.10)<br />
Figure 3.3. A laminate with n layers (a) Structural axes (b) Material axes of ply k (c) Position of each<br />
ply in the stacking sequence.<br />
1
Optimization of Laminated <strong>Composite</strong> Structures… 61<br />
3.2. The Possible Parameterizations of Laminates<br />
There exist several parameterizations for the laminates depending on the way the coefficients<br />
of the stiffness matrices in (3.10) are computed and depending on the definition of the design<br />
variables. The advantages and disadvantages of those different parameterizations are<br />
compared in the perspective of the optimal design of the laminated composite structures.<br />
3.2.1. Parameterization with Respect to Thickness and Orientation<br />
When the ply thickness and the related fibers orientation are chosen to describe the laminate,<br />
the coefficients of the stiffness matrices can be written as follows:<br />
= ∑<br />
=<br />
− −<br />
n<br />
Aij<br />
Qij<br />
k hk<br />
k 1<br />
hk<br />
1 )<br />
)]( ( [ θ ⇔ = ∑<br />
=<br />
n<br />
Aij<br />
[ Qij<br />
( θ k )] tk<br />
k 1<br />
1 n<br />
2 2<br />
Bij<br />
= ∑ [ Qij<br />
( θ k )]( hk<br />
− hk−1<br />
) ⇔<br />
2 k=<br />
1<br />
= ∑<br />
=<br />
n<br />
Bij<br />
[ Qij<br />
( θ k )] tk<br />
zk<br />
k 1<br />
(3.11)<br />
1 n<br />
3 3<br />
Dij<br />
= ∑ [ Qij<br />
( θk )]( hk<br />
− hk−1<br />
) ⇔<br />
3 k=<br />
1<br />
= ∑ +<br />
=<br />
n<br />
3<br />
2 tk<br />
Dij<br />
[ Qij<br />
( θ k )]( tk<br />
zk<br />
) , i , j = 1,<br />
2,<br />
6<br />
k 1<br />
12<br />
where zk and hk define the position of the k th ply in the stacking sequence. tk and k<br />
θ are the<br />
ply thickness and the fibers orientation, respectively (Figure 3.3).<br />
With such a parameterization the local values (e.g. the stresses in each ply of the<br />
laminate) are available via the relations (3.1) and (3.4). On top of that the design problem is<br />
written in terms of the physical parameters used for the manufacturing of the laminated<br />
structures. Finally several different materials can be considered in the laminate when the<br />
parameterization (3.11) is used.<br />
However when fibers orientations are allowed to change during the structural design<br />
process the resulting mechanical properties are generally strongly non linear (see Figure 3.2)<br />
and non convex, and local minima appear in the optimization problem. This is also illustrated<br />
in Figure 3.4 that draws the variation of the strain energy density in a laminate over 2 fibers<br />
orientations. In Figure 3.5 it is shown that the structural responses entirely differ when either<br />
ply thickness or ply orientation is considered in the design, resulting in mixed monotonousnon<br />
monotonous structural behaviors. It turns that the optimal design task is more<br />
complicated since the optimization method should be able to efficiently take into account<br />
simultaneously both different behaviors.<br />
Additionally using such a parameterization increases the number of design variables that<br />
may appear in the optimal design problem since the thickness and fibers orientation of each<br />
ply are possible variables. Finally optimizing with respect to the fibers orientations is known<br />
to be very difficult and few publications are available on the subject. For a sake of<br />
completion, the sensitivity analysis of the structural responses of composites with respect to<br />
those variables can be found in Mateus et al. (1991), Geier and Zimmerman (1994), and<br />
Dems (1996).
62<br />
Strain energy density<br />
(N/mm)<br />
Michaël Bruyneel<br />
θ2 θ1<br />
Figure 3.4. Variation of the strain energy density in a [θ 1/θ 2] S laminate with respect to the fibers<br />
orientations θ 1 and θ 2.<br />
Strain nenergy<br />
density (N/mm)<br />
θ<br />
1.2<br />
1.4<br />
1.6<br />
1.8<br />
Figure 3.5. Variation of the strain energy density in an unidirectional ply with respect to its thickness t<br />
and its fibers orientation θ.<br />
3.2.2. Parameterization with Sub-laminates<br />
The design parameters are no longer defined based on single unidirectional plies but instead<br />
on predefined sub-laminates. Each sub-laminate is itself made of several single unidirectional<br />
plies. The design parameters are assigned to the sub-laminates and no longer to each<br />
individual ply. Examples of sub-laminates may be [0/45/-45/90], [0/60/-60] or [0/90]. This<br />
parameterization allows to decrease the number of design variables. However the control at<br />
the ply level is lost. The previously presented parameterization in terms of ply thickness and<br />
orientation is a limiting case.<br />
2<br />
t
Optimization of Laminated <strong>Composite</strong> Structures… 63<br />
3<br />
Sub-laminate 2<br />
[0/45/-45/90]<br />
2<br />
Sub-laminate 1<br />
[30/-30]<br />
Figure 3.6. Parameterization with sub-laminates. Here the symmetric laminate is made of 2 sublaminates.<br />
3.2.3. The Lamination Parameters<br />
The stiffness matrix in (3.10) can be expressed with the lamina invariants defined in (3.8)<br />
together with the lamination parameters. For a given base material identical for each ply of<br />
the laminate the lamination parameters are given by (3.12) in the structural axes:<br />
ξ<br />
h / 2<br />
A,<br />
B,<br />
D 0,<br />
1,<br />
2<br />
[ ] = ∫ z [ cos 2θ<br />
( z),<br />
cos 4θ<br />
( z),<br />
sin 2θ<br />
( z),<br />
sin 4θ<br />
( z)<br />
]<br />
1,<br />
2,<br />
3,<br />
4<br />
−h<br />
/ 2<br />
1<br />
dz<br />
(3.12)<br />
The lamination parameters are the zero, first and second order moments relative to the<br />
plate mid-plane of the trigonometric functions (3.6) entering the rotation formulae for the ply<br />
stiffness coefficients (3.5). With this definition the stiffness matrices A, B and D in (3.10)<br />
write:<br />
A = hγ<br />
B = γ<br />
h<br />
D =<br />
12<br />
+ γ<br />
0<br />
B<br />
1ξ1<br />
3<br />
γ 0<br />
+<br />
A<br />
1ξ1<br />
B<br />
γ 2ξ2<br />
+ γ<br />
D<br />
1 1<br />
+<br />
A<br />
2ξ2<br />
B<br />
γ3ξ<br />
3<br />
+ γ ξ + γ ξ<br />
2<br />
+ γ<br />
D<br />
2<br />
+<br />
A<br />
3ξ3<br />
B<br />
γ 4ξ4<br />
+ γ<br />
+ γ ξ<br />
D<br />
3ξ3<br />
4<br />
A<br />
4<br />
+ γ ξ<br />
4<br />
D<br />
4<br />
(3.13)<br />
Twelve lamination parameters exist in total and characterize the global stiffness of the<br />
laminate. This number is independent of the number of plies that contains the laminate. In<br />
most applications the lamination parameters are normalized with respect to the total thickness<br />
of the laminate (Grenestedt, 1992, and Hammer, 1997). In the case of symmetric laminates<br />
the 4 lamination parameters B<br />
ξ defining the coupling stiffness B vanish. Moreover when the<br />
structure is either subjected to in-plane loads or to out-of-plane loads only the 4 lamination<br />
parameters related to the in-plane stiffness A<br />
ξ or the out-of-plane stiffness D<br />
ξ must be<br />
considered, respectively. In the case of composite membrane or plates presenting orthotropic<br />
material properties 2 lamination parameters are sufficient to characterize the problem.<br />
Lamination parameters are not independent variables. Feasible regions of the lamination<br />
parameters exist which provide realizable laminates. Grenestedt and Gudmundson (1993)
64<br />
Michaël Bruyneel<br />
demontrated that the set of the 12 lamination parameters is convex. It is also observed from<br />
(3.13) that the constitutive matrices A, B and D are linear with respect to the lamination<br />
parameters. This means that the optimization problem is convex if it includes functions<br />
related to the global stiffness of the laminate, as for example the structural stiffness, vibration<br />
frequencies and buckling loads (Foldager, 1999).<br />
Feasible regions were determined for specific laminate configurations (e.g. Miki, 1982<br />
and Grenestedt, 1992), but the region for the 12 lamination parameters has not yet been<br />
determined. Recently the relations between the lamination parameters were derived for ply<br />
angles restricted to 0, 90, 45 and -45 degrees by Liu et al. (2004) for membrane and bending<br />
effects, and by Diaconu and Sekine (2004) for membrane, coupling and bending effects.<br />
One of the feasible regions of lamination parameters is illustrated in Figure 3.7 in the<br />
case of a symmetric and orthotropic laminated plate subjected to bending. As the plate is<br />
assumed orthotropic in bending D ξ 1 and D ξ 2 are enough to identify the stiffness of such a<br />
problem. Those two lamination parameters take their values on the outline delimited by the<br />
points A, B, C, and in the dashed zone. Any combination of the lamination parameters that is<br />
outside of this region will produce a laminate which is not realizable. When this plate is<br />
simply supported and subjected to a uniform pressure, the vertical displacement is a function<br />
of D ξ 1 and D ξ 2 . The iso-values of this structural response are the parallel lines illustrated in<br />
Figure 3.7. According to Grenestedt (1990), the plate stiffness increases in the direction of the<br />
arrow. The stiffest plate is then characterized by the point D in Figure 3.7, which corresponds<br />
to a [(±θ)n]S laminate, defined by a single parameter θ.<br />
1<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0<br />
-0.2<br />
-0.4<br />
-0.6<br />
-0.8<br />
C<br />
D<br />
D<br />
ξ2<br />
E<br />
D<br />
ξ1<br />
-1<br />
-1 -0.5 0.5 1<br />
B<br />
Figure 3.7. Feasible domain (outline plus dashed zone) of the lamination parameters for a symmetric<br />
and orthotropic laminated plate subjected to a uniform pressure (after Grenestedt, 1990). The points A,<br />
B, C correspond to [0], [(±45) n] S and [90] laminates, respectively. The point D defines a [(±θ) n] S<br />
laminate. The point E is a combination of laminates defined on the outline. The laminate of maximum<br />
stiffness is located on the outline (point D)<br />
A
Optimization of Laminated <strong>Composite</strong> Structures… 65<br />
This kind of parameterization has allowed to show that optimal solutions – in terms of the<br />
stiffness – are often related to simple laminates with few different ply orientations. For<br />
example only one orientation is necessary for characterizing the optimal laminate in a flexural<br />
problem (Figure 3.7), and at most 3 different ply orientations are sufficient to define the<br />
optimal stacking sequence in the case of a membrane of maximum stiffness (Lipton, 1994).<br />
Table 3.1 summarizes some of those important results.<br />
When using such a parameterization the number of design variables is very small (12 in<br />
the most general case) irrespective to the number of plies that contains the laminate. As<br />
seen in Figure 3.7 the design space is convex, and only one set of lamination parameters<br />
characterizes the optimal solution. However acording to the relations (3.8) and (3.13) only<br />
one kind of material can be used in the laminate: defining a different material for the core<br />
of a sandwich panel is for example not allowed (Tsai and Hahn, 1980). Additionally the<br />
local structural responses (e.g. the stresses in each ply) can not be expressed in terms of the<br />
lamination parameters since those last are defined at the global (laminate) level and are<br />
linked to the structural stiffness. However the global strains of the laminate (but not in each<br />
ply) can be computed with relation (3.10) and used in the optimization, as is done by<br />
Herencia et al. (2006). The feasible regions of the 12 lamination parameters is not yet<br />
determined. As said before those regions are only known for specific laminate<br />
configurations. This strongly limit their use in the frame of the optimal design of composite<br />
structures. Finally when the optimal values of the lamination parameters are known,<br />
coming back to corresponding thicknesses and orientations is a difficult problem and the<br />
solution is non unique (Hammer, 1997). Foldager et al. (1998) proposed a technique based<br />
on a mathematical programming approach while Autio (2000) used a genetic algorithm to<br />
find this solution when the number of layers is limited or for prescribed standardized ply<br />
angles.<br />
Table 3.1. Summary of some important results obtained with the lamination parameters<br />
Kind of<br />
structure<br />
Laminate configuration Criteria Optimal sequence Reference<br />
Grenestedt (1990),<br />
Miki and Sugiyama<br />
(1993)<br />
Plate Symmetric/orthotropic<br />
Symmetric<br />
Stiffness<br />
Vibration<br />
Buckling<br />
Buckling<br />
[ ( ± θ ) n ] S<br />
[ ( ± θ ) n ] S<br />
[ ( ± θ ) n ] S<br />
[ θ ] S<br />
Grenestedt (1991)<br />
Membrane Symmetric<br />
General<br />
Stiffness<br />
Stiffness<br />
[ ( / 90 α)<br />
n ] S<br />
[ θ ] , [ α / 90 + α]<br />
(1993)<br />
Hammer (1997)<br />
Symmetric/orthotropic Buckling [ ( ± θ ) n ] S , [ 0 / 90]<br />
S ,<br />
Cylindrical<br />
shell<br />
3.2.4. Combined Parameterization<br />
α + Fukunaga and Sekine<br />
quasi-isotropic<br />
Fukunaga and<br />
Vanderplaats (1991b)<br />
As shown by Foldager et al. (1998) and Foldager (1999), composite structures can be<br />
designed by combining two parameterizations: the lamination parameters on one hand, and<br />
the plies thickness and fibers orientations on the other hand. The benefit of the approach
66<br />
Michaël Bruyneel<br />
relies on using a convex design space with respect to the lamination parameters, while<br />
keeping in the problem’s definition the physical variables in terms of thickness and<br />
orientation. This iterative procedure – between both design spaces – consists in determining<br />
a first (local) solution in terms of thicknesses and orientations. A new search direction<br />
towards the global optimum is then computed by evaluating the first order derivative of the<br />
objective function at the local solution with respect to the lamination parameters. The<br />
global optimum is reached when this sensitivity is close to zero. Otherwise a new design<br />
point is calculated in the space of the fibers orientations, and the process continues, usually<br />
by adding new plies in the laminate. As seen in Figure 3.8, the structural response is not<br />
convex with respect to θ while it is convex in terms of the lamination parameter ξ. With<br />
this technique the knowledge of the feasible regions of the lamination parameters is not<br />
mandatory.<br />
Although efficient, this solution procedure can only be used for global structural<br />
responses like the stiffness, the vibration frequencies and the buckling load.<br />
f<br />
1<br />
2<br />
3<br />
4<br />
Figure 3.8. Illustration of the optimization process after Foldager et al. (1998) in both spaces of the<br />
lamination parameters ξ and the fibers orientation θ.<br />
3.2.5. Alternative Parameterization<br />
In order to decrease the non linearities introduced by the fibers orientation variables,<br />
Fukunaga and Vanderplaats (1991a) proposed to parameterize the laminated composite<br />
membranes with the following intermediate variables:<br />
i = sin 2 i or xi = cos2θ<br />
i<br />
x θ<br />
based on the relation (3.12) and (3.13). This formulation was tested by Vermaut et al. (1998)<br />
for the optimal design of laminates with respect to strength and weight restrictions. As in the<br />
previous section, the main difficulty is to compute the orientations corresponding to the<br />
optimal intermediate variables values xi.<br />
f(θ)<br />
θ, ξ<br />
f(ξ)
Optimization of Laminated <strong>Composite</strong> Structures… 67<br />
4. Specific Problems in the Optimal Design of <strong>Composite</strong><br />
Structures<br />
For designing laminated composite structures a very large number of data must be considered<br />
(material properties, plies thickness and fibers orientation, stacking sequence) and complex<br />
geometries must be modelled (aircraft wings, car bodies). Therefore the finite element method<br />
is used for the computation of the structural mechanical responses. Usually mass, structural<br />
stiffness, ply strength and strain, as well as buckling loads are the functions used in the<br />
optimization problem. The design variables are classically the parameters defining the<br />
laminate: fibers orientations, plies thickness, and indirectly the number of plies and the<br />
stacking sequence. Some specific problems appear in the formulation of the optimization<br />
problem for laminated structures. They are reported hereafter.<br />
Large number of design variables. Even for a parameterization in terms of the lamination<br />
parameters, the number of design variables can easily reach a large value when the plies<br />
thickness and fibers orientations are allowed to change over the structure, leading to non<br />
homogenenous plies (Figure 4.1) and curvilinear fibers formats (Hyer and Charette 1991,<br />
Hyer and Lee 1991, Duvaut et al. 2000). In industrial applications (Krog et al. 2007),<br />
thicknesses related to specific orientations (0°, ±45°, 90°) are used and several independent<br />
regions are defined throughout the composite structure, what increases the number of design<br />
variables.<br />
Large number of design functions. Not only global structural responses related to the<br />
stiffness are relevant in a composite structure optimization, but also the local strength of each<br />
ply. Damage tolerance and local buckling restrictions are important as well. For an aircraft<br />
wing, it is usual to include about 300000 constraints in the optimization problem (Krog et al.<br />
2007).<br />
Non homogeneous ply<br />
Homogeneous ply<br />
Figure 4.1. Homogeneous and non homogeneous ply in a laminate.<br />
Problems related to the topology optimization of composite structures. In topology<br />
optimization one is looking for the optimal distribution of a given amount of material in a<br />
predefined design space that maximizes the structural stiffness (Figure 4.2).
68<br />
Domain where the<br />
material is<br />
distributed<br />
Solid<br />
Michaël Bruyneel<br />
Figure 4.2. Illustration of a topology optimization problem (after Bruyneel, 2002)<br />
For composite structures, and due to the stratification of the material, it results that 2<br />
topology optimization problems must be defined and solved simultaneously: the optimal<br />
distribution of plies at a given altitude in the laminate (Figure 4.3) and the transverse topology<br />
optimization where the optimal local stacking sequence is looked for (Figure 4.4). Continuity<br />
conditions between adjacent laminates should also be imposed.<br />
Figure 4.3. Topology optimization at a given altitude in the non homogeneous laminate.<br />
Figure 4.4. Transverse topology optimization in a composite structure.<br />
Void
Optimization of Laminated <strong>Composite</strong> Structures… 69<br />
Specific non linear behaviors of laminated structures. In order to improve the accuracy in<br />
the model, non linear effects, and especially the design with respect to the limit load, should<br />
be considered in the formulation of the optimization of composite structures. This<br />
dramatically increases the computational time of the finite element analysis, and can only be<br />
used for studying small structural parts such as super-stringers, i.e. some stiffeners and the<br />
panel (Colson et al., 2007). Although simple fracture mechanics criteria have been considered<br />
(Papila et al. 2001), damage tolerance and propagation of the cracks (delamination) should be<br />
taken into account in the same way.<br />
Uncertainties on the mechanical properties of composites. There is a larger dispersion in<br />
the mechanical properties of the fibers reinforced composite materials than for metals.<br />
Moreover, some uncertainties concerning the orientations and the plies thickness exist.<br />
Robust optimization should be used in these cases (Mahadevan and Liu, 1998, Chao et al.,<br />
1993, Chao, 1996, and Kristindottir et al., 1996).<br />
Strong link with the manufacturing process. Contrary to the design with metals, there is<br />
a strong link between the material design, the structural design and the manufacturing<br />
process when dealing with composite materials. The constraints linked to manufacturing<br />
can strongly influence the design and the structural performances (Henderson et al., 1999,<br />
Fine and Springer, 1997, Manne and Tsai, 1998) and should be taken into account to<br />
formulate in a rational way the design problem (Karandikar and Mistree, 1992).<br />
Singular optima in laminates design problems. When strength constraints are<br />
considered in the design problem, and if the lower bounds on the plies thickness is set close<br />
to 0 (i.e. some plies can disappear at the solution from the initial stacking sequence), it can<br />
be seen (Schmit and Farschi, 1973, Bruyneel and Fleury, 2001) that the design space can<br />
become degenerated. In this case the optimal design can not be reached with gradient based<br />
optimization methods. Such a degenerated design space is illustrated in Figure 4.5. It is<br />
divided into a feasible and an infeasible region according to the limiting value of the Tsai-<br />
Wu criteria. In this example a [0/90]S laminate’s weight is to be minimized under an inplane<br />
load N1. The optimal solution is a [0] laminate. Unfortunately this optimal laminate<br />
configuration can not be reached with a gradient based method since the 90 degree plies are<br />
still present in the problem even if their thickness is close to zero, and the related Tsai-Wu<br />
criterion penalizes the optimization process. A first solution consists in using the ε-relaxed<br />
approach (Cheng and Guo 1997), which slightly modifies the design space in the<br />
neighborhood of the solution and allows the optimization method to reach the true optimum<br />
[0] * . Alternatively (Bruyneel and Fleury, 2001, and Bruyneel and Duysinx, 2006) when<br />
fibers orientations are design variables the shape of the design space changes, the gap<br />
between the true optimal solution and the one constrained by plies with a vanishing<br />
thickness [0/x] * decreases and the real optimal solution becomes attainable (Figure 4.5).<br />
Optimizing over the fibers orientations allows to circumvent the singularity of the design<br />
space.
70<br />
* represents the obtained solutions, optimum or not<br />
Michaël Bruyneel<br />
Figure 4.5. Design space for [0/90] S and [0/10] S laminates.<br />
Importance of the fibers orientations in the laminate design. Besides their efficiency in<br />
avoiding the singularity in the optimization process as just explained before fibers<br />
orientations play a key role in the design of composite structures. Modifying their value<br />
allows for great weight savings, as illustrated in Figure 4.6. Let’s consider that the initial<br />
laminate design corresponds to fibers orientation and ply thickness at point A. A first way to<br />
obtain a feasible design with respect to strength restrictions is to increase the ply thickness<br />
and go to B, which penalizes the structural weight. Another solution consists in modifying the<br />
fibers orientation, here at constant thickness (point C). A better solution is to simultaneously<br />
optimize with respect to both kinds of design variables (point D). However taking into<br />
account such variables in the optimization problem is a real issue, and providing a reliable<br />
solution procedure is a challenge.<br />
Figure 4.6. Design space for an unidirectional laminate subjected to either N 1 or N 6. Iso-values of the<br />
Tsai-Wu criterion. The ply thickness and fibers orientations are the design variables.<br />
The optimal stacking sequence. A large part of the research effort on composites has been<br />
dedicated to the solution of the optimal stacking sequence problem. As it is a combinatorial
Optimization of Laminated <strong>Composite</strong> Structures… 71<br />
problem including integer variables, genetic algorithms have been used (Haftka and Gurdal,<br />
1992, Le Riche and Haftka, 1993). The topology optimization formulation of Figure 4.4 was<br />
used by Beckers (1999) and (Stegmann and Lund, 2005) to solve this problem with discrete<br />
and continuous design variables, respectively. Another approach, still based on the discrete<br />
character of the problem, is proposed by Carpentier et al. (2006). It consists in using a lay-up<br />
table defined based on buckling, geometric and industrial rules considerations. This table,<br />
which satisfies the ply drop-off continuity restrictions is determined numerically. Once it is<br />
obtained a given laminate total thickness corresponds to a stacking sequence (via a column of<br />
the table). The optimization process then consists in optimizing the local thickness of a set of<br />
contiguous laminates defining the structure. Each laminate has equivalent homogenized<br />
properties with 0, ±45 and 90° plies. Based on the lay-up table, the stacking sequence is<br />
therefore known everywhere in the structure for different local optimal thicknesses and the<br />
composite material can be drapped.<br />
Figure 4.7. Illustration of a lay-up table for 0, ±45 and 90° plies.<br />
5. Problems Solved in the Literature<br />
5.1. Structural Responses<br />
When designing laminated composite structures the functions entering the optimization<br />
problem (2.1) are classically the stiffness, the vibration frequencies, the structural stability<br />
and the plies’ strength. (see Abrate, 1994, for a detailed review of the literature). It is<br />
interesting to note that for orthotropic laminates maximizing the stiffness, the frequency or<br />
the first buckling load will provide the same solution (Pedersen, 1987 and Grenestedt, 1990).<br />
On top of that, it should be noted that optimizing a laminated structure against plies strength<br />
or stiffness will result in different designs. It results that the local (stress) effects are very<br />
important in the optimal design of composite structures (Tauchert and Adibhatla, 1985,<br />
Fukunaga and Sekine, 1993, and Hammer, 1997).
72<br />
Michaël Bruyneel<br />
5.2. Optimal Design with Respect to Fibers Orientations<br />
Determining the optimal fibers orientation is a very difficult problem since the structural<br />
responses in terms of such variables are highly non linear, non monotonous and non convex.<br />
However it has just been show in the previous section that the design of laminated composite<br />
structures is very sensitive with respect to those variables. As explained by the editors of<br />
commercial optimization software (Thomas et al., 2000) there is a need for an efficient<br />
treatment of such parameters.<br />
A small amount of work has been dedicated to the optimal design of laminated structures<br />
with respect to the fibers orientations. Several kinds of approaches have been investigated and<br />
are reported in the literature:<br />
• Approach by optimality criteria<br />
Optimal orientations of orthotropic materials that maximize the stiffness in membrane<br />
structures were obtained by Pedersen (1989, 1990 and 1991), and by Diaz and Bendsøe<br />
(1992) for multiple load cases. When the unidirectional ply is only subjected to in-plane<br />
loads, Pedersen (1989) proposed to place the fibers in the direction of the principal stresses.<br />
The resulting optimality criterion was used in topology optimization including rank-2<br />
materials (Bendsøe, 1995). This technique was used by Thomsen (1991) in the optimal design<br />
of non homogeneous composite disks. This criterion was extended by Krog (1996) to Mindlin<br />
plates and shells.<br />
• Approach based on the mathematical programming<br />
As soon as 1971, Kicher and Chao solved the problem with a gradients based method.<br />
Hirano (1979a and 1979b) used the zero order method of Powell (conjugate directions) for<br />
buckling optimization of laminated structures. Tauchert and Adibhatla (1984 and 1985) used<br />
a quasi-Newton technique (DFP) able to take into account linear constraints for minimizing<br />
the strain energy of a laminate for a given weight. Cheng (1986) minimized the compliance of<br />
plates in bending and determined the optimal orientations with an approach based on the<br />
steepest descent method.<br />
Martin (1987) found the minimum weight of a sandwich panel subjected to stiffness and<br />
strength restrictions with a method based on the Sequential Convex Programming<br />
(Vanderplaats, 1984). Watkins and Morris (1987) used a similar procedure with a robust<br />
move-limits strategy (see also Hammer 1997).<br />
In Foldager (1999), the method used for determining the optimal fibers orientations is not<br />
cited but belongs according to the author to the family of mathematical programming<br />
methods.<br />
SQP, the feasible directions method and the quasi-Newton BFGS were used by<br />
Mahadevan and Liu (1998), Fukunaga and Vanderplaats (1991a), and Mota Soares et al.<br />
(1993, 1995 and 1997), respectively. Those mathematical programming methods are reported<br />
and explained in Bonnans et al. (2003).<br />
• Approach with non deterministic methods
Optimization of Laminated <strong>Composite</strong> Structures… 73<br />
Genetic algorithms have been employed by several authors for determining the optimal<br />
stacking sequence of laminated structures (Le Riche and Haftka, 1993, Kogiso et al., 1994<br />
and Potgieter and Stander, 1998) or in the treatment of fibers orientations (Upadhyay and<br />
Kalyanarama, 2000).<br />
5.3. Formulations of the Optimization Problem<br />
Thickness and orientation variables were treated in several ways in the literature. They have<br />
been considered either simultaneously as in Pedersen (1991), and Fukunaga and Vanderplaats<br />
(1991a), or separately (Mota Soares et al. 1993, 1995 and 1997, and Franco Correia et al.<br />
1997).<br />
Weight, stiffness and strength criteria have been separately introduced in the design<br />
problem and taken into account in a bi-level approach by (Mota Soares et al., 1993, 1995,<br />
1997 and Franco Correia et al., 1997): at the first level the weight is kept constant and the<br />
stiffness is optimized over fibers orientations ; at the second level the ply thicknesses are the<br />
only variables in an optimization problem that aims at minimizing the weight with respect to<br />
strength and/or displacements restrictions. A similar approach can be found in Kam and Lai<br />
(1989), and Soeiro et al. (1994). Fukunaga and Sekine (1993) also used a bi-level approach<br />
for determining laminates with maximal stiffness and strength in non homogeneous<br />
composite structures (Figure 4.2) subjected to in-plane loads. In Hammer (1997), both<br />
problems are separately solved and the initial configuration for optimizing with respect to<br />
strength is the laminate previously obtained with a maximal stiffness consideration.<br />
6. Optimal Design of <strong>Composite</strong>s for Industrial Applications<br />
Based on the several possible laminate parameterizations and on the previous discussion it<br />
was concluded in Bruyneel (2002, 2006) that an industrial solution procedure for the design<br />
of laminated composite structures should preferably be based on fibers orientations and ply<br />
thicknesses, instead of intermediate non physical design variables such as the lamination<br />
parameters. Using those variables allows optimizing very general structures (membranes,<br />
shells, volumes, subjected to in- and out-of-plane loads, symmetric or not) and provides a<br />
solution that is directly interpretable by the user.<br />
On the other hand, an optimization procedure used for industrial applications should be<br />
able to consider a large number of design variables and constraints, and find the solution (or<br />
at least a feasible design) in a small number of design cycles. Additionally, the optimization<br />
formulation should be as much general as possible, and not only limited to specific cases (e.g.<br />
not only thicknesses, not only membrane structures, not only orthotropic configurations,…).<br />
For those reasons, a solution procedure based on the approximation concepts approach seems<br />
to be inevitable. Interesting local solutions can be found by resorting to other optimization<br />
methods (e.g. response surfaces coupled with a genetic algorithm) but on structures of limited<br />
size. For the pre-design of large composite structures like a full wing or a fuselage, or when<br />
non linear responses are defined in the analysis (post-buckling, non linear material behavior),<br />
the approximation concepts approach proved to be a fast method not expensive in CPU time<br />
for solving industrial problems (Krog and al, 2007, Colson et al., 2007).
74<br />
Michaël Bruyneel<br />
It results that robust approximation schemes must be available to efficiently optimize<br />
laminated structures. The characteristics of such a reliable approximation are explained in the<br />
following, and tests are carried out to show the efficiency and the applicability of the method.<br />
7. Optimization Algorithm for Industrial Applications<br />
7.1. The Approximation Concepts Approach<br />
In the approximation concepts approach, the solution of the primary optimization problem<br />
(2.1) is replaced with a sequence of explicit approximated problems generated through first<br />
order Taylor series expansion of the structural functions in terms of specific intermediate<br />
variables (e.g. direct xi or inverse 1/xi variables). The generated structural approximations<br />
built from the information known at least at the current design point (via a finite element<br />
analysis), are convex and separable. As will be explained latter a dual formulation can then be<br />
used in a very efficient way for solving each explicit approximated problem.<br />
According to section 2, it is apparent that the approximation concepts approach is well<br />
adapted to structural optimization including sizing, shape and topology optimization<br />
problems. However, the use of the existing schemes (section 7.2) can sometimes lead to bad<br />
approximations of the structural responses and slow convergence (or no convergence at all)<br />
can occur (Figure 7.1).<br />
x2<br />
* global<br />
X<br />
(k ) *<br />
X<br />
x2<br />
(k )<br />
X<br />
* global<br />
X<br />
* local<br />
X<br />
(k ) *<br />
X<br />
x1<br />
x2<br />
* global<br />
X<br />
a. A too conservative approximation b. A too few conservative approximation and unfeasible intermediate<br />
solutions c. An approximation not adapted to the problem, leading to zigzagging<br />
* local<br />
X<br />
Figure 7.1. Difficulties appearing in the approximation of highly non linear structural responses.<br />
x1<br />
(k ) *<br />
X<br />
(k )<br />
X<br />
* local<br />
X<br />
x1
Optimization of Laminated <strong>Composite</strong> Structures… 75<br />
Such difficulties are met for laminates optimization: their structural responses are mixed,<br />
i.e. monotonous with regard to plies thickness and non monotonous when fibers orientations<br />
are considered (Figure 3.5). Additionally, the non monotonous structural behaviors in terms<br />
of orientations are difficult to manage (Figure 3.4). It results that the selection of a right<br />
approximation scheme is a real challenge. In the next section a generalized approximation<br />
scheme is presented that is able to effectively treat those kinds of problems. This optimization<br />
algorithm will identify the structural behavior (monotonous or not) according to the involved<br />
design variable (orientation or thickness), and will automatically generate the most reliable<br />
approximation for each structural function included in the optimization problem. In section 8<br />
numerical tests will compare the efficiency of the proposed approximation scheme and the<br />
existing ones for laminates optimization including both thickness and orientation variables.<br />
7.2. Selection of an Accurate Approximation Scheme<br />
7.2.1. Monotonous Approximations<br />
Based on the first order derivatives of the structural responses included in the optimization<br />
problem, linear approximations can be built at the current design point x k . It is a first order<br />
Taylor series expansion in terms of the direct design variables xi (7.1).<br />
k<br />
k ∂g<br />
j<br />
k<br />
g ~ ( ) ( ) ( x ) ( )<br />
j ( x ) =g j ( x ) + ∑ ( xi<br />
− xi<br />
)<br />
(7.1)<br />
∂x<br />
i<br />
As it is very simple this approximation is most of the time not efficient for structural<br />
optimization but can anyway be used with some specific move-limits rules (Watkins and<br />
Morris, 1987) that prevent the intermediate design point to go too far from the current one<br />
and to generate large oscillations during the optimization process (Figures 7.1b and 7.1c).<br />
Since the stresses vary as 1/xi in isostatic trusses where xi is the cross section area of the<br />
bars, a linear approximation in terms of the inverse design variables is more reliable for the<br />
optimal sizing of thin structures. The resulting reciprocal approximation is given in (7.2).<br />
k<br />
g<br />
~ ( )<br />
j<br />
( k)<br />
i<br />
( k)<br />
⎛ ⎞<br />
( k)<br />
( k)<br />
2 ∂g<br />
j ( x )<br />
−<br />
⎜ 1 1<br />
( x ) =g<br />
⎟<br />
j ( x ) ∑ ( xi<br />
)<br />
−<br />
(7.2)<br />
∂ ⎜ k ⎟<br />
i<br />
xi<br />
x ( )<br />
⎝ i xi<br />
⎠<br />
The Conlin scheme developed by Fleury and Braibant (1986) is a convex approximation<br />
based on (7.1) and (7.2). It is reported in (7.3) and illustrated in Figure 7.2.<br />
g~<br />
( k)<br />
j<br />
( k)<br />
⎛ ⎞<br />
( k)<br />
∂g<br />
j ( x ) ( k)<br />
( k)<br />
2<br />
∂g<br />
j ( x )<br />
+<br />
⎜ 1 1<br />
( x ) =g<br />
⎟<br />
j ( x ) ∑ ( xi<br />
− xi<br />
) − ∑ ( xi<br />
)<br />
− (7.3)<br />
∂<br />
⎜ ⎟<br />
−<br />
∂<br />
( k)<br />
+ xi<br />
xi<br />
x<br />
⎝ i xi<br />
⎠<br />
( k)
76<br />
Michaël Bruyneel<br />
The symbols ∑ ( +) and ∑ ( −) in (7.3) denote the summations over terms having positive<br />
and negative first order derivatives. When the first order derivative of the considered<br />
structural response is positive a linear approximation in terms of the direct variables is built,<br />
while a reciprocal approximation is used on the contrary.<br />
145<br />
140<br />
135<br />
130<br />
125<br />
120<br />
115<br />
110<br />
105<br />
100<br />
g(x<br />
)<br />
~ ( )<br />
g k<br />
l<br />
( x)<br />
Strain energy density<br />
(N/mm)<br />
~ ( )<br />
g k<br />
r<br />
( x)<br />
45 90 (k )<br />
(k )<br />
180<br />
xr<br />
xl<br />
Figure 7.2. The Conlin approximation.<br />
Conlin can only work with positive design variables since an asymptote is imposed at<br />
xi=0. On top of that, the curvature of this approximation is imposed by the derivative at the<br />
current design point and can not be adapted to better fit the problem.<br />
The Method of Moving Asymptotes or MMA (Svanberg 1987) generalizes Conlin by<br />
introducing two sets of new parameters, the lower and upper asymptotes, Li and Ui, that can<br />
take positive or negative values, in order to adjust the convexity of the approximation in<br />
accordance with the problem under consideration. The asymptotes are updated following<br />
some rules provided by Svanberg (1987). The parameters pij and qij are built with the first<br />
order derivatives.<br />
Strain energy<br />
145 density (N/mm)<br />
140<br />
135<br />
130<br />
125<br />
120<br />
115<br />
110<br />
105<br />
g(x<br />
)<br />
~ ( )<br />
g ( x)<br />
k<br />
100<br />
(k )<br />
L<br />
100<br />
(k )<br />
U<br />
45 90 (k ) 135 (k )* 180 45 90 (k )*<br />
(k )<br />
180<br />
x<br />
x<br />
Strain energy<br />
145 density (N/mm)<br />
140<br />
135<br />
130<br />
125<br />
120<br />
115<br />
110<br />
105<br />
g(x<br />
)<br />
Figure 7.3. The MMA approximation.<br />
x<br />
~ ( )<br />
g ( x)<br />
k<br />
x
Optimization of Laminated <strong>Composite</strong> Structures… 77<br />
⎛<br />
⎞ ⎛<br />
⎞<br />
~ ( k)<br />
( k)<br />
( k)<br />
⎜ 1<br />
1 ⎟ ( k)<br />
⎜ 1 1<br />
g<br />
⎟<br />
j ( x ) =g j ( x ) + ∑ pij<br />
−<br />
+ ∑q−(7.4)<br />
⎜ ( k)<br />
( k)<br />
( k)<br />
⎟ ij ⎜ ( k)<br />
( k)<br />
( k)<br />
⎟<br />
+ ⎝U<br />
i − xi<br />
U i − xi<br />
⎠ − ⎝ xi<br />
− Li<br />
xi<br />
− Li<br />
⎠<br />
As it will be seen later those monotonous schemes are not efficient for optimizing<br />
structural functions presenting non monotonous behaviors, as in Figure 3.4.<br />
7.2.2. Non Monotonous Approximations<br />
Based on MMA, Svanberg (1995) developed the Globally Convergent MMA approximation<br />
(GCMMA). As illustrated in Figure 7.4 it is non monotonous and still only based on the<br />
information at the current design point (functions values, first order derivatives, asymptotes<br />
values). Here both Ui and Li are used simultaneously. It was not the case in (7.4).<br />
k<br />
g<br />
~ ( )<br />
j<br />
⎛<br />
⎞ ⎛<br />
⎞<br />
( k)<br />
( k)<br />
∑<br />
⎜ 1 1 ⎟ ( k)<br />
+ ∑<br />
⎜ 1 1<br />
( x ) =g +<br />
−<br />
− ⎟<br />
j ( x ) pij<br />
q<br />
(7.5)<br />
⎜ ( k)<br />
( k)<br />
( k)<br />
⎟ ij ⎜ ( k)<br />
( k)<br />
( k)<br />
⎟<br />
i ⎝U<br />
i − xi<br />
Ui<br />
− xi<br />
⎠ i ⎝ xi<br />
− Li<br />
xi<br />
− Li<br />
⎠<br />
Using this method can lead to slow convergence given that it can generated too<br />
conservative approximations of the design functions (Figure 7.1a).<br />
145<br />
140<br />
135<br />
130<br />
125<br />
120<br />
115<br />
110<br />
105<br />
100<br />
Strain energy<br />
density (N/mm)<br />
(k )<br />
L<br />
g(x<br />
)<br />
(k )<br />
U<br />
45 90 (k ) 135 (k )* 180<br />
x<br />
~ ( )<br />
g ( x)<br />
k<br />
Figure 7.4. The GCMMA approximation.<br />
In order to improve the quality of this approximation it was proposed in Bruyneel and<br />
Fleury (2002) and Bruyneel et al. (2002) to use the gradients at the previous iteration to<br />
improve the quality of the approximation, leading to the definition of the Gradient Based<br />
MMA approximations (GBMMA). In those methods the pij and qij parameters of (7.5) are<br />
computed based on the function value and gradient at the current design point and on the<br />
gradient at the previous iteration. The rules defined by Svanberg (1995) for updating the<br />
asymptotes are used.<br />
x
78<br />
Michaël Bruyneel<br />
7.2.3. Mixed Approximation of the MMA Family<br />
When dealing with structural optimization problems including design variables of two<br />
different natures, for example in problems mixing ply thickness and orientation variables, one<br />
is faced to a difficult task because of the simultaneous presence of monotonous and nonmonotonous<br />
behaviors with respect to the set of design variables. In these conditions, most of<br />
the usual approximation schemes presented before have poor convergence properties or even<br />
fail to solve these kinds of problems. Knowing that the MMA approximation is very reliable<br />
for approximating monotonous design functions and based on the GBMMA approximations,<br />
a mixed monotonous – non monotonous scheme is presented in Bruyneel and Fleury (2002)<br />
and Bruyneel et al. (2002), which will automatically adapt itself to the problem to be<br />
approximated (7.6).<br />
⎛<br />
k<br />
k<br />
k<br />
g<br />
~ ( )<br />
( ) ( )<br />
∑<br />
⎜ 1<br />
j ( x)<br />
= g j ( x ) + pij<br />
⎜ ( k)<br />
i∈A⎝Ui−xi 1<br />
⎞ ⎛<br />
⎟ ( k)<br />
+ ∑<br />
⎜ 1<br />
−<br />
q<br />
( k)<br />
( k)<br />
−<br />
⎟ ij ⎜ ( k)<br />
Ui<br />
xi<br />
⎠ i∈A<br />
⎝ xi<br />
− Li<br />
1<br />
⎞<br />
− ⎟<br />
( k)<br />
( k)<br />
x −<br />
⎟<br />
i Li<br />
⎠<br />
⎛<br />
( k)<br />
+ ∑<br />
⎜ 1<br />
pij<br />
⎜ ( k)<br />
+ , i∈B⎝Ui−xi 1 ⎞ ⎛<br />
−<br />
⎟ ( k)<br />
+ ∑ q ⎜ 1<br />
( k)<br />
( k)<br />
−<br />
⎟ ij ⎜ ( k)<br />
Ui<br />
xi<br />
⎠ −,<br />
i∈B<br />
⎝ xi<br />
− Li<br />
1 ⎞<br />
− ⎟<br />
( k)<br />
( k)<br />
x −<br />
⎟<br />
i Li<br />
⎠<br />
In (7.6) the symbols ∑ ( + , i) and ∑( − , i ) designate the summations over terms having<br />
positive and negative first order derivatives, respectively. A and B are the sets of design<br />
variables leading to a non monotonous and a monotonous behavior respectively, in the<br />
considered structural response. At a given stage k of the iterative optimization process, a<br />
monotonous, non monotonous or linear approximation is automatically selected, based on the<br />
tests (7.7), (7.8) and (7.9) computed for given structural response g (X)<br />
and design variable<br />
x i .<br />
j<br />
(7.6)<br />
k<br />
k −1<br />
∂g<br />
j ( x ) ∂g<br />
j ( x )<br />
× > 0 ⇒ MMA (monotonous)<br />
∂xi<br />
∂xi<br />
(7.7)<br />
k<br />
k −1<br />
∂g<br />
j ( x ) ∂g<br />
j ( x )<br />
× < 0 ⇒ GBMMA (non monotonous)<br />
∂xi<br />
∂xi<br />
(7.8)<br />
k<br />
k −1<br />
∂g<br />
j ( x ) ∂g<br />
j ( x )<br />
− = 0 ⇒ linear expansion<br />
∂x<br />
∂x<br />
(7.9)<br />
i<br />
i<br />
The selection of a right approximation is illustrated in Figure 7.5: when a monotonous<br />
approximation is used for approximating a non monotonous function, oscillations can appear,<br />
while a non monotonous approximation is too conservative when the function is monotnous.<br />
The best approximation is therefore selected based on tests (7.7) to (7.9). This strategy proved<br />
to be reliable for simple laminates design (Bruyneel and Fleury 2002) and for general<br />
laminated composite structures design problems (Bruyneel 2006, Bruyneel et al. 2007, Krog<br />
et al. 2007), for truss sizing and configuration (Bruyneel et al. 2002), for topology
Optimization of Laminated <strong>Composite</strong> Structures… 79<br />
optimization which includes a large amount of design variables (Bruyneel and Duysinx<br />
2005). It has been made available in the BOSS Quattro optimization toolbox (Radovcic and<br />
Remouchamps, 2002). In the following this solution procedure based on a mixed<br />
approximation scheme is called Self Adaptive Method (SAM). Based on this approximation<br />
scheme, it is possible to resort to the other ones (GBMMA, MMA, Conlin and the linear<br />
approximation) by setting specific values to the asymptotes and by limiting the<br />
approximations to the sets A or B in (7.6).<br />
145<br />
140<br />
135<br />
130<br />
125<br />
120<br />
115<br />
110<br />
105<br />
100<br />
Strain energy<br />
density<br />
(N/mm)<br />
(k )<br />
L<br />
g(θ<br />
)<br />
θ<br />
(k ) *<br />
MMA<br />
g GCMMA<br />
~<br />
g MMA<br />
~<br />
(k )<br />
U<br />
45 90 (k ) 135 (k ) * 180<br />
θ θGCMMA<br />
400<br />
350<br />
300<br />
250<br />
200<br />
150<br />
g MMA<br />
~<br />
g(t<br />
)<br />
Strain energy<br />
density<br />
(N/mm)<br />
1.2 1.3 1.4 1.5 (k ) 1.7<br />
Figure 7.5. The mixed SAM approximation.<br />
t<br />
g GCMMA<br />
~<br />
(k )*<br />
GCMMA<br />
A summary of the approximations that will be compared in the following is presented in<br />
Table 7.1.<br />
Table 7.1. Summary of the approximations that will be compared in the numerical tests<br />
Approximation Author Behavior<br />
MMA Svanberg (1987) Monotonous<br />
GCMMA Svanberg (1995) Non monotonous<br />
SAM Bruyneel (2006) Mixed monotonous/non monotonous<br />
7.3. Solution Procedure for Mono and Multi-objective Optimizations<br />
Since the approximations are convex and separable the solution of each optimization subproblem<br />
(Figure 2.3) is achieved by using a dual approach. Based on the theory of the duality,<br />
solving the problem (2.2) in the space of the primal variables xi is equivalent to maximize a<br />
function (7.10) that depends on the Lagrangian multipliers λ j , also called dual variables:<br />
max minL<br />
( x, λ)<br />
λ<br />
x<br />
t<br />
(k ) *<br />
MMA<br />
0 0,...,<br />
( 0 1)<br />
=<br />
= ≥ λ<br />
λ<br />
m j j (7.10)<br />
t
80<br />
Michaël Bruyneel<br />
Solving the primal problem (2.2) requires the manipulation of one design function, m<br />
structural restrictions and 2 × n side constraints (for mono-objective problems). When the<br />
dual formulation is used, the resulting quasi-unconstrained problem (7.10) includes one<br />
design function and m side constraints, if the side constraints in the primal problem are treated<br />
separately. In relation (7.10), L( x,<br />
λ)<br />
is the Lagrangian function of the optimization problem,<br />
which can be written<br />
pij<br />
qij<br />
L ( x,<br />
λ)<br />
= ∑λj( c j + ∑ + ∑ )<br />
(7.11)<br />
k<br />
U − x x − L<br />
j i i i i<br />
according to the general definition of the involved approximations g<br />
~<br />
j ( X ) of the functions.<br />
The parameter λj is the dual variable associated to each approximated function g<br />
~<br />
j ( X ) . Given<br />
that the approximations are separable, the Lagrangian function is separable too. It turns that:<br />
and the Lagrangian problem of (7.10)<br />
can be split in n one dimensional problems<br />
= ∑ x ) , ( ) ( λ<br />
λ x, L<br />
L<br />
i<br />
i<br />
k<br />
minL ( x, λ)<br />
x<br />
i<br />
i<br />
k<br />
minL ( x , λ)<br />
(7.12)<br />
x<br />
i<br />
The primal-dual relations are obtained by solving (7.12) for each primal variable xi:<br />
∂Li<br />
( xi<br />
, λ)<br />
= 0<br />
∂xi<br />
i<br />
i<br />
⇒<br />
xi<br />
= xi<br />
( λ)<br />
k i<br />
(7.13)<br />
Relation (7.13) asserts the stationnarity conditions of the Lagrangian function over the<br />
primal variables xi. Once the primal-dual relations (7.13) are known, (7.10) can be replaced<br />
by<br />
maxl ( λ)<br />
⇔ maxL(<br />
x(<br />
λ)<br />
, λ)<br />
(7.14)<br />
λ<br />
λ<br />
λ j ≥ 0 j = 1,...,<br />
m<br />
Solving problem (2.2) is then equivalent to maximize the dual function l (λ)<br />
with non<br />
negativity constraints on the dual variables (7.14). As it is explained by Fleury (1993), the
Optimization of Laminated <strong>Composite</strong> Structures… 81<br />
maximization (7.14) is replaced by a sequence of quadratic sub-problems. Each sub-problem<br />
is itself partially solved by a first order maximization algorithm in the dual space.<br />
In the case of a multi-objective formulation the optimization problem writes :<br />
min max gl0<br />
( x)<br />
X l = 1,...,<br />
nc<br />
g j (x ) ≤ g j j = 1,...,<br />
m<br />
(7.15)<br />
where nc is the number of load cases. Using the bound formulation (Olhoff, 1989) the<br />
problem (7.15) can be written as:<br />
g l0<br />
( x)<br />
≤ β<br />
1 2<br />
min β<br />
2<br />
l = 1,...,<br />
nc<br />
(7.16)<br />
g j (x ) ≤ g j<br />
j = 1,...,<br />
m<br />
where β is the multiobjective factor, that is an additional design variable in the optimization<br />
problem. Instead of solving (7.16) problem (7.17) is considered where a new variable δ is<br />
introduced for the possible relaxation of the set of constraints.<br />
( ) ( ) 2<br />
1 2 1 2 C ( k −1)<br />
min β + δ + p + ∑ xi<br />
− xi<br />
2 2 2 i<br />
g j0<br />
( x ) ≤ β j g j0<br />
j = 1,...,<br />
nobj<br />
(7.17)<br />
g j ( x ) ≤ g j ( 1+<br />
δ )<br />
j = 1,...,<br />
m<br />
g j0<br />
are target values on the objective functions. The dual approach described for monoobjective<br />
optimisation problems is then applied to (7.17).<br />
8. Applications of the Optimization Solution Procedure<br />
In the following examples (except the simple laminate designs and the topology optimization<br />
problem), the structural and semi-analytical sensitivity analyses are carried out with<br />
SAMCEF (http://www.samcef.com). The Boss Quattro optimisation tool box<br />
(http://www.samcef.com) is used for defining and solving the optimisation problem<br />
(Radovcic and Remouchamps 2002).<br />
8.1. Laminate Subjected to in- and out-of-plane Loadings<br />
A symmetric 4 plies laminate made of carbon/epoxy is considered. The load case and the<br />
initial configuration are provided in Table 8.1. The fibers orientations of each ply are the
82<br />
Michaël Bruyneel<br />
design variables, while plies thicknesses are kept constant. The optimization consists in<br />
minimizing the laminate’s strain energy density, i.e. maximizing its stiffness. The evolution<br />
of this objective function with respect to the 2 angles θ1 and θ2 is reported in Figure 8.1, with<br />
the initial and optimal design points. A restriction is imposed on the relative variation of the 2<br />
design variables. The optimization problem writes :<br />
1 T T 1 T<br />
min ε 0 Aε 0+<br />
κ Dκ<br />
θ 2 2<br />
θ 2 − θ1<br />
≤ 45<br />
(8.1)<br />
0. 001 θ ≤180<br />
i = 1,<br />
2<br />
≤ i<br />
where the stiffness matrices A, B and D, and the laminate’s strain and curvature were<br />
previously defined in Section 3.<br />
Strain energy<br />
density (N/mm)<br />
θ2<br />
θ1<br />
Initial design<br />
Optimal design<br />
Solution<br />
Figure 8.1. Variation of the strain energy density in the symmetric laminate subjected to the load case<br />
of Table 8.1.<br />
In-plane load case<br />
( N 1,<br />
N 2 , N6<br />
)<br />
in N/mm<br />
Table 8.1. Problem’s definition: load case and initial design<br />
Out-of-plane load case<br />
( M 1,<br />
M 2 , M 6 )<br />
in N<br />
Initial orientations<br />
θ = ( θ1,<br />
θ 2 )<br />
in degrees<br />
23.3°<br />
22.3°<br />
Initial thicknesses<br />
t = ( t1,<br />
t2<br />
)<br />
in mm<br />
(2000,0,1000) (0,500,0) (45,135) (1,2)<br />
In this application the laminate is subjected not only to in-plane but also to out-of-plane<br />
loadings. Since the plies thicknesses are not identical (Table 8.1) the objective function is not<br />
symmetric with regards to the axis θ 1 = θ 2 (Figure 8.2).
Optimization of Laminated <strong>Composite</strong> Structures… 83<br />
180<br />
160<br />
140<br />
120<br />
100<br />
θ2<br />
80<br />
60<br />
40<br />
20<br />
Strain energy density<br />
(N/mm)<br />
θ init<br />
θ opt<br />
θopt unconstrained<br />
0<br />
0 20 40 60 80 100 120 140 160 180<br />
θ1 Figure 8.2. Illustration of the design space. Staring point, unconstrained and constrained optimum.<br />
The iteration histories for the 3 approximation schemes are illustrated in the Figure 8.3.<br />
The convergence of the optimization process is controlled by the relative variation of the<br />
design variables at 2 successive iterations. The MMA approximation converges in 41<br />
iterations. 29 iterations are enough for GCMMA. When the SAM approximation is used the<br />
solution is reached in a very small number of iterations.<br />
25<br />
20<br />
15<br />
10<br />
5<br />
150<br />
100<br />
Objective function (N/mm)<br />
0<br />
0 20 40 60<br />
50<br />
Evolution of angles (deg.)<br />
0<br />
0 20 40 60<br />
20<br />
15<br />
10<br />
5<br />
150<br />
100<br />
Objective function (N/mm)<br />
0<br />
0 10 20 30<br />
50<br />
Evolution of angles (deg.)<br />
0<br />
0 10 20 30<br />
20<br />
15<br />
10<br />
5<br />
150<br />
100<br />
Objective function (N/mm)<br />
0<br />
0 5 10 15<br />
50<br />
Evolution of angles (deg.)<br />
0<br />
0 5 10 15<br />
MMA GCMMA SAM<br />
Figure 8.3. Iteration history for the 3 approximation methods.
84<br />
8.2. Non Homogeneous Laminate<br />
Michaël Bruyneel<br />
In this application a non homogeneous composite membrane divided in regions of constant<br />
thickness and fibre orientations is studied. Each region is defined with an unidirectional<br />
laminate made of a glass/epoxy material. The design over stiffness is only considered here.<br />
The solution with respect to strength and stiffness is provided in Bruyneel (2006).<br />
2<br />
2<br />
1<br />
1<br />
2<br />
P<br />
1<br />
P<br />
Figure 8.4. Initial configurations with 45 and -45 degrees plies orientations.<br />
The quasi-unconstrained optimization problem (8.2) consists in finding the optimal<br />
values of the plies thickness and fibers orientations in each region of the laminated composite<br />
structure that maximize the overall stiffness (i.e. that minimize the compliance – the potential<br />
energy of the applied loads). The vectors of the design variables are given by<br />
θ = { θi<br />
, i = 1,...,<br />
n}<br />
and t = { ti , i = 1,...,<br />
n}<br />
where n is the number of regions according to<br />
Figure 8.4. The initial thicknesses are of 1 mm.<br />
2<br />
2<br />
1<br />
P<br />
1<br />
min Compliance<br />
θ,t<br />
0° ≤θ<br />
i ≤180°<br />
i = 1,...,<br />
n<br />
(8.2)<br />
0. 01mm<br />
≤ ti<br />
≤ 5mm<br />
In this problem the optimal values of the thickness is 5 mm, that is their upper bound.<br />
Anyway this application illustrates the difficulties encountered when both kinds of design<br />
variables appear in the design problem. The optimal values of the compliances are reported in<br />
Figure 8.5 as a function of the number of regions. As already noticed by Foldager (1999) an<br />
increase of the number of regions of different orientations improves the overall optimal<br />
structural stiffness (i.e. it decreases the compliance).<br />
The optimal fibers orientations are illustrated in Figure 8.6, for the several membrane<br />
configurations of Figure 8.4. The iteration histories are reported in Figure 8.7. When the SAM<br />
method is used, about 10 iterations are enough for reaching a stationary solution with respect<br />
to a small relative variation of the objective at 2 successive iterations. The GCMMA<br />
approximation finds this solution in a larger number of design cycles. It is observed that when<br />
P<br />
P
Optimization of Laminated <strong>Composite</strong> Structures… 85<br />
the SAM method is used, the structural responses in terms of both the fibers orientations and<br />
the thicknesses are well approximated, while using GCMMA, the approximation in terms of<br />
the thicknesses is too conservative, what slows down the overall convergence speed of the<br />
optimization process.<br />
1<br />
0.95<br />
0.9<br />
0.85<br />
0.8<br />
0.75<br />
0.7<br />
0.65<br />
0.6<br />
0.55<br />
Relative compliances<br />
0.5<br />
1 4 8 12 20<br />
Number of regions : n<br />
Figure 8.5. Evolution of the compliances in the problem (8.2) for the structures illustrated in Figure 8.4.<br />
The compliance of the one region structure is the reference (n = 1)<br />
1 region<br />
4 regions<br />
Figure 8.6. Continued on next page.
86<br />
Michaël Bruyneel<br />
8 regions<br />
12 regions<br />
20 regions<br />
Figure 8.6. Illustration of the optimal fibers orientations for the different composite membranes<br />
illustrated in Figure 7.9.<br />
Pli 19<br />
Pli 5<br />
Figure 8.7. Continued on next page.
4<br />
x 10<br />
5<br />
4<br />
3<br />
2<br />
1<br />
Compliance (Nmm)<br />
0<br />
0 20 40 60 80<br />
4<br />
x 10<br />
5<br />
4<br />
3<br />
2<br />
1<br />
Compliance (Nmm)<br />
0<br />
0 5 10 15<br />
Optimization of Laminated <strong>Composite</strong> Structures… 87<br />
Mass (kg) and thickness of ply19 (mm)<br />
8<br />
7<br />
6<br />
5<br />
4<br />
3<br />
2<br />
1<br />
0 20 40 60 80<br />
7<br />
6<br />
5<br />
4<br />
3<br />
2<br />
GCMMA<br />
SAM<br />
Total mass<br />
Thickness of ply 19<br />
Mass (kg) and thickness of ply19 (mm)<br />
8<br />
Total mass<br />
Thickness of ply 19<br />
1<br />
0 5 10 15<br />
140<br />
120<br />
100<br />
80<br />
60<br />
Orientation of ply 5 (deg.)<br />
40<br />
0 20 40 60 80<br />
140<br />
120<br />
100<br />
80<br />
60<br />
Orientation of ply 5 (deg.)<br />
40<br />
0 5 10 15<br />
Figure 8.7. Convergence history for GCMMA and SAM for the membrane divided in 20 regions.<br />
Evolution of the thickness and the orientations of the plies number 5 and 19.<br />
2<br />
1.9<br />
1.8<br />
1.7<br />
1.6<br />
1.5<br />
1.4<br />
1.3<br />
1.2<br />
1.1<br />
Vertical displacement δ max under the load (mm)<br />
1<br />
0 20 40 60 80 100 120 140 160 180<br />
Fibers orientation (deg.)<br />
O 320 finite elements<br />
+ 80 finite elements<br />
* 20 finite elements<br />
Figure 8.8. Evolution of the vertical displacement under the applied load for several discretizations of<br />
the homogeneous composite membrane (Figure 8.4, n=1).
88<br />
Michaël Bruyneel<br />
In Figure 8.8 the evolution of the vertical displacement under the load is drawn with<br />
respect to the fibers orientation in the case of the homogeneous membrane (Figure 8.4, n=1).<br />
The global minimum displacement is obtained for a value of 170°. When the starting point of<br />
the optimization process of the problem (8.2) is close to 45°, 0° fibers orientation is found as<br />
a local optimum. As -45° is chosen here for the initial design (i.e. 135°), the global optimum<br />
can be reached. This illustrates the fact that a gradient based method is not able to reach the<br />
global optimum, unless the starting point is in its vicinity. In Figure 8.8, the influence of the<br />
mesh refinement on the solution is presented, as well.<br />
8.3. Multi-objective Optimization<br />
A symmetric laminate made of 4 plies and subjected to 2 in-plane load cases is considered.<br />
N 2<br />
3<br />
N 1<br />
1<br />
θ<br />
x<br />
N 6<br />
N 1<br />
Figure 8.9. Laminate subjected to in-plane loads.<br />
The applied loads and the initial configuration are reported in Table 8.2. The load case<br />
(2) is variable : the factor k takes the values 0,1,2,…,8. The extreme load cases are, on one<br />
hand (1000,0,0) and on the other hand the combination of (1000,0,0) and (0,2000,0) N/mm.<br />
Table 8.2. Definition of the problem: load case and starting point<br />
Load case (1) Load case (2) Initial orientations Initial thickness<br />
( N 1,<br />
N 2 , N6<br />
) ( N 1,<br />
N 2 , N6<br />
) θ = ( θ1,<br />
θ 2 )<br />
t = ( t1,<br />
t2<br />
)<br />
in N/mm<br />
in N/mm<br />
in degrés<br />
en mm<br />
(1000,0,0) (0, k × 250 ,0) (30,120) (1,2)<br />
The performance of three approximation schemes are compared : GCMMA, MMA and<br />
SAM. The optimization problem writes :<br />
1<br />
min max ε( ) ( j)<br />
2 j Aε<br />
j 1,2<br />
T<br />
θ, t =<br />
TW ( j)<br />
( θ i , ti<br />
) ≤1<br />
i<br />
, j = 1,<br />
2<br />
2<br />
N 2
Optimization of Laminated <strong>Composite</strong> Structures… 89<br />
4<br />
∑ t i ≤ 4<br />
i=<br />
1<br />
(8.3)<br />
0. 001 ≤ θ i ≤ 180 i = 1,<br />
2<br />
0. 001 t ≤ 10<br />
i = 1,<br />
2<br />
≤ i<br />
where j is the number of the load case. This problem is solved by resorting the its bound<br />
formulation (Olhoff, 1989) including here 5 design variables (2 orientations, 2 thicknesses<br />
and the multi-objective factor β) and 7 constraints:<br />
1<br />
min β<br />
2<br />
1 T<br />
ε(<br />
j)<br />
Aε(<br />
j)<br />
2<br />
TW ( j)<br />
( i , ti<br />
)<br />
4<br />
∑ t i<br />
i=<br />
1<br />
0. 001 ≤ i ≤<br />
0. 001 ≤ i ≤ 10<br />
≤ β<br />
2<br />
j = 1,<br />
2<br />
θ ≤1<br />
i , j = 1,<br />
2<br />
(8.4)<br />
≤ 4<br />
θ 180 i = 1,<br />
2<br />
t i = 1,<br />
2<br />
The results are reported in Figure 8.10 for the different values of k. The solution is<br />
obtained when the relative variation of the design variables at 2 successive iterations is lower<br />
than 0.01. It is seen that a large number of iterations is needed to reach the optimum when<br />
MMA is used. GCMMA converges in a lower number of iterations. As for mono-objective<br />
problems, SAM is the most effective optimization method.<br />
+ MMA<br />
o GCMMA<br />
Δ SAM<br />
Maximum strain energy density (N/mm)<br />
4<br />
3<br />
2<br />
1<br />
0<br />
0 2 4 6 8<br />
80<br />
60<br />
40<br />
20<br />
Number of iterations<br />
0<br />
0 2 4 6 8<br />
Load parameter k Load parameter k<br />
Figure 8.10. Variation of the strain energy density and number of iterations needed to reach the solution<br />
as a function of the parameter k.
90<br />
10 5<br />
10 0<br />
15<br />
10<br />
10 -5<br />
5<br />
10 -10<br />
Objective functions (N/mm)<br />
0<br />
0 10 20 30 40 50<br />
3<br />
2.5<br />
2<br />
1.5<br />
1<br />
Variations of the objective functions<br />
0 10 20 30 40 50<br />
Michaël Bruyneel<br />
2 Maximum constraints violations<br />
10<br />
10 0<br />
10 -2<br />
10 5<br />
10 0<br />
10 -5<br />
0 10 20 30 40 50<br />
Maximum variables variation<br />
0 10 20 30 40 50<br />
Figure 8.11. Convergence history for MMA. k is equal to 3.<br />
Objective functions (N/mm)<br />
0.5<br />
0 2 4 6 8 10 12<br />
10 0<br />
10 -2<br />
10 -4<br />
10 -6<br />
Variations of the objective functions<br />
2 4 6 8 10 12<br />
10 1<br />
10 0<br />
10 -1<br />
10 5<br />
10 0<br />
10 -5<br />
Maximum constraints violations<br />
0 2 4 6 8 10 12<br />
Maximum variables variation<br />
Figure 8.12. Convergence history for SAM. k is equal to 3.<br />
2 4 6 8 10 12<br />
Figure 8.13 illustrates the optimum stacking sequence for the different values of the load<br />
parameter k. The solution corresponds to a [0/90]S with a variable proportion of 90° plies<br />
(depending on k).<br />
Figure 8.14 describes the design space for k = 4. The iso-values of both objective<br />
functions are drawn. The arrow indicates the direction for an increase of the stiffness. The<br />
optimal solution is characterized here by identical values of both objective functions.
Optimization of Laminated <strong>Composite</strong> Structures… 91<br />
4<br />
3.5<br />
3<br />
2.5<br />
2<br />
1.5<br />
1<br />
Evolution of the strain<br />
energy density<br />
Laminate configuration<br />
for the several load<br />
cases<br />
0.5<br />
0 1 2 3 4 5 6 7 8<br />
Load parameter k<br />
Figure 8.13. Variation of the strain energy density and configuration of the corresponding optimal<br />
laminate.<br />
Figure 8.14. Evolution of the strain energy densities for the [0/90]S laminate Subjected to<br />
( N 1 , N 2 , N6<br />
) = ( 0,<br />
1000,<br />
0)<br />
N / mm and ( N 1 , N 2 , N6<br />
) = ( 1000,<br />
0,<br />
0)<br />
N / mm . t0° and t90° are the plies<br />
thickness.<br />
The variation of the strain energy density for each single load case is illustrated in<br />
Figures 8.15 and 8.16. In those particular cases, the optimal solutions are given by only 90° or<br />
0° orientations. This illustrates the need for a multi-objective formulation when several<br />
functions are considered as objective.
92<br />
Michaël Bruyneel<br />
Figure 8.15. Evolution of the strain energy density in the [0/90]S laminate subjected to<br />
N 2 = 1000N<br />
/ mm .<br />
Figure 8.16. Evolution of the strain energy density in the [0/90]S laminate subjected to<br />
N 1 = 1000N<br />
/ mm .<br />
8.4. Optimal Design with Respect to Stiffness and Strength Restrictions<br />
In this application a stiffened laminated composite panel subjected to a uniform pressure is<br />
considered. The geometry, the boundary conditions and the stacking sequence of the different<br />
parts of the panel are illustrated in Figure 8.17. The plies thickness is equal to 0.125 mm and<br />
the base material is carbon/epoxy.
3.5<br />
3<br />
2.5<br />
2<br />
1.5<br />
1<br />
Optimization of Laminated <strong>Composite</strong> Structures… 93<br />
laminate 1 :[(0/90/45/-45)2]S<br />
laminate 2 : [0/90/45/-45]4<br />
Figure 8.17. Geometry and initial stacking sequence of the stiffened panel.<br />
Relative compliance<br />
0.5<br />
0 10 20 30<br />
3<br />
2.5<br />
2<br />
1.5<br />
1<br />
0.5<br />
0 5 10 15<br />
2<br />
1.8<br />
1.6<br />
1.4<br />
1.2<br />
1<br />
GCMMA<br />
Relative mass<br />
0.8<br />
0 10 20 30<br />
1.8<br />
1.6<br />
1.4<br />
1.2<br />
1<br />
SAM<br />
0.8<br />
0 5 10 15<br />
60<br />
50<br />
40<br />
30<br />
20<br />
10<br />
Number of violations<br />
0<br />
0 10 20 30<br />
Relative compliance Relative mass Number of violations<br />
2<br />
60<br />
Figure 8.18. Convergence history for GCMMA and SAM.<br />
50<br />
40<br />
30<br />
20<br />
10<br />
0<br />
0 5 10 15
94<br />
Michaël Bruyneel<br />
The optimization problem consists in maximizing the structural stiffness for a given<br />
maximum weight, knowing that a safety margin of 0.15 on the Tsai-Hill criterion on the top<br />
and the bottom of each ply must be obtained at the solution. 64 strength restrictions are<br />
defined at the plies level. The design variables are the orientations of the plies initially<br />
oriented at 0, -45, +45 and 90 degrees and the related thicknesses. The problem includes 16<br />
design variables. The convergence histories of GCMMA and SAM are compared in Figure<br />
8.18.<br />
The SAM approximation succeeds in finding a solution in a very small number of<br />
iterations, with comparison to GCMMA. The optimal stacking sequence is illustrated in<br />
Figure 8.19. As already observed by Grenestedt (1990) and Foldager (1999), the optimal<br />
laminates include very few different orientations.<br />
laminate 1 :[90]<br />
laminate 2 : [0/93/96/93]4<br />
Figure. 8.19. Optimal design of the stiffened panel.<br />
8.5. Optimal Design under Buckling Considerations<br />
2.48 mm<br />
Anyone who has carried out optimal sizing with a buckling criterion has experienced an<br />
undesirable effect of very slow convergence speed and possibly large variations of the design<br />
functions during the iteration history. The reasons for the bad convergence of the buckling<br />
optimisation problem are multiple, and make it difficult to solve: discontinuous character of<br />
the problem due to the localized nature of local buckling, non differentiability of the eigenvalues<br />
and related problems in the sensitivity computation, modes crossing, selection of a<br />
right optimisation method, etc.<br />
A curved composite panel including 7 hat stiffeners is considered. The load case consists<br />
of a compression along the long curved sides, and in shear on the whole outline. The structure<br />
is simply supported on its edges. Bushing elements are used to fasten the stiffeners to the<br />
panel. In each super-stiffener (made of one stringer and the corresponding part of the whole<br />
panel), 3 design variables are used for defining the thickness of the 0°, 90° and ±45° plies in<br />
the panel and in the stiffener. 42 design variables are then defined. The goal is to find the<br />
structure of minimum weight with a minimum buckling load larger than 1.2. The results<br />
obtained in Bruyneel et al. (2007) are reported in Figures 8.20 for Conlin (Fleury and<br />
Braibant 1986) and SAM (Bruyneel 2006). The 12 first buckling loads are the design<br />
restrictions of the optimisation problem. In Figure 8.20, the evolutions of the weight and the
Optimization of Laminated <strong>Composite</strong> Structures… 95<br />
first buckling load λ1 over the iterations are plotted, as well as some characteristic buckling<br />
modes.<br />
Figure 8.20. Convergence history for the buckling optimisation with Conlin (left) and SAM (right)<br />
Bruyneel et al. (2007)<br />
It is seen that when Conlin is used (Figure 8.20, left) a solution can not be reached. With<br />
SAM (Figure 8.20, right), the solution is obtained after an erratic convergence history. Those<br />
oscillations come from the fact that local buckling modes appear during the optimisation<br />
process, and some parts of the structures are no longer sensitive to this criterion. A small<br />
thickness is therefore assigned to those parts to decrease the weight, what makes them very<br />
sensitive to buckling at the next iteration, leading to oscillations of the design variables and<br />
functions values. It was observed in Bruyneel et al. (2007) that when a large number of<br />
buckling loads are used in the optimization problem (say 100 for the problem of Figure 8.20),
96<br />
Michaël Bruyneel<br />
a solution with SAM is reached in 6 iterations, while Conlin is still no longer able to<br />
converge.<br />
8.6. Topology Optimization of Laminated <strong>Composite</strong> Structures<br />
The topology optimization problem of Figure 4.3 is here considered. In topology optimization<br />
of isotropic material (Bendsoe 1995), the design variable is a pseudo-density μi that varies<br />
between 0 and 1 in each finite element i (Figure 4.2). The so-called SIMP material law<br />
(Simply isotropic Material with Penalization) takes the following form:<br />
Ei<br />
= μ<br />
p 0<br />
i E<br />
0<br />
ρi = μi<br />
ρ<br />
(8.5)<br />
where E 0 and ρ 0 are the Young modulus and the density of the base material (e.g. steel), E<br />
and ρ are the effective material properties, and p is the exponent of the SIMP law, chosen by<br />
the user (1
Optimization of Laminated <strong>Composite</strong> Structures… 97<br />
The problem in Figure 8.21 is solved with this parameterization. It includes 3750 design<br />
variables. The optimal topology and orientations obtained for an half of the structure are<br />
given in Figure 8.22. A comparison of the convergence speed for several approximations is<br />
provided in Figure 8.23.<br />
?<br />
Figure 8.21. Definition of topology optimization problem. The initial structure is full of material.<br />
Figure 8.22. Optimal topology with orthotropic material. Only one half of the structure is drawn. The<br />
fibers orientation is plotted in the few elements that contain full material at the solution<br />
Figure 8.23. Convergence history for several approximation schemes for the topology optimization<br />
problem including orthotropic material.
98<br />
Michaël Bruyneel<br />
8.7. An Industrial Solution for the Pre-design of <strong>Composite</strong> Aircraft Boxes<br />
As reported in Krog et al. (2007), the pre-design of an aircraft wing is a large scale<br />
optimization problem including (up to now) about 1000 design variables and about 300000<br />
constraints. Those variables are linked to the total thickness of the laminate made of 0, ±45<br />
and 90° plies in the panel and to the dimensions of the cross section for the composite<br />
stiffener of each super-stringer defining the box structure (Figure 8.24). The constraints<br />
expressed as reserve factors (RF) are amongst others related to buckling and damage<br />
tolerance.<br />
Figure 8.24. The principle of a composite wing made of super-stringers (from Krog et al., 2007).<br />
Taking into account a so large number of design functions in the optimization problem<br />
will dramatically increase the CPU time spent in the optimizer. In order to decrease the size<br />
of the optimization problem, a technique for scanning the constraints (Figure 8.25) has been<br />
implemented in Boss Quattro (www.samcef.com). It consists in feeding the optimizer with<br />
the most critical constraints, based on their value at a given iteration. This leads to the<br />
definition of 2 sets of active and inactive constraints. The optimizer can only see the active<br />
restrictions. Those sets are not updated at each iteration but only when some inactive<br />
constraints tend to become violated after a given number of iterations (FREQ in Figure 8.25).<br />
When the SAM method (Bruyneel 2006) is used, the information at the previous design point<br />
is lost when the sets are updated, and the approximation is therefore only built based on the<br />
information at the current design point, that is with GCMMA (Svanberg 1995), for that<br />
specific iteration.<br />
The SAM approximation was found to be reliable in solving pre-design optimization<br />
problems of composite aircraft box structures in wings, center wing box, vertical and<br />
horizontal tail planes. Typically 30 iterations were enough to reach a stationary value of the<br />
weight and a nearly feasible design where very few constraints (less than 10) were still<br />
violated but of an amount of no more than 3 percents (RF larger than 0.97). Details of the<br />
results and of the implementation can be found in Krog et al. (2007).
Optimization of Laminated <strong>Composite</strong> Structures… 99<br />
Figure 8.25. Strategy for scanning the constraints in large scale optimization problems (Krog et al.<br />
2007).<br />
8.8. Optimal Design with Respect to Damage Tolerance<br />
A simple DCB beam is considered (Figure 8.26). The energy release rates of modes I, II and<br />
III are computed at the straight crack front with a specific virtual crack extension method<br />
described by Bruyneel et al. (2006). The stacking sequence composed of 32 plies is given by:<br />
[θ/−θ/0/−θ/0/θ/θ/04/θ/0/−θ/0/−θ/θ/d/−θ/θ/0/θ/0/−θ/04/−θ/0/θ/0/θ/-θ]<br />
where d is the location of the interface where delamination will take place and θ is a variable.<br />
The goal is to find the optimal value of the orientation that will decrease the maximum value<br />
of GI along the crack front.<br />
Figure 8.26. DCB beam and variation of GI along the crack front for the initial design. On the left the<br />
displacements of the lips are multiplied by 50.
100<br />
Michaël Bruyneel<br />
The solution is provided in Figure 8.27. The optimal value for the angle θ is zero. The<br />
convergence is achieved in 5 iterations with the SAM approximation and in 15 for MMA<br />
(Figure 8.28). Although the solution of this problem is trivial, the procedure could be used for<br />
more realistic structures subjected to several complex load cases.<br />
Figure 8.27. DCB beam and variation of GI along the crack front for the optimal design. On the left the<br />
displacements of the lips are multiplied by 50<br />
Figure 8.28. Convergence history for the optimization with respect to damage tolerance. SAM<br />
converges in 5 iterations while MMA needs 15 iterations to reach the solution<br />
9. Conclusion<br />
In this chapter the optimal design of laminated composite structures was considered. After a<br />
review of the literature an optimization method specially devoted to composite structures was<br />
presented. This review helped us in selecting a formulation of the optimization problem that<br />
satisfies the industrial needs. In this context the fibers orientations and the ply thicknesses<br />
were selected as design variables. It was shown on the proposed applications that the<br />
developed solution procedure is general and reliable. It can be used for solving laminated<br />
composite problems including membrane, shells, solids, single and multiple load cases, in
Optimization of Laminated <strong>Composite</strong> Structures… 101<br />
stiffness, buckling and strength based designs. It is routinely used in an (European) industrial<br />
context for the design of composite aircraft box structures located in the wings, the center<br />
wing box, and the vertical and horizontal tail plane. This approach is based on sequential<br />
convex programming and consists in replacing the original optimization problem by a<br />
sequence of approximated sub-problems. A very general and self adaptive approximation<br />
scheme is used. It can consider the particular structure of the mechanical responses of<br />
composites, which can be of a different nature when both fiber orientations and plies<br />
thickness are design variables.<br />
References<br />
Abrate S. (1994). Optimal design of laminated plates and shells, <strong>Composite</strong> Structures, 29,<br />
269-286.<br />
Arora J.S., Elwakeil O.A., Chahande A.I. and Hsieh C.C. (1995). Global optimization<br />
methods for engineering applications: a review, Structural Optimization, 9, 137-159.<br />
Autio M. (2000). Determining the real lay-up of a laminate corresponding to optimal<br />
lamination parameters by genetic search, Structural and Multidisciplinary Optimization,<br />
20, 301-310.<br />
Barthelemy J.F.M. and Haftka R.T. (1993). Approximation concepts for optimum structural<br />
design – a review, Structural Optimization, 5, 129-144.<br />
Beckers M. (1999). A dual method for structural optimization involving discrete variables,<br />
Third World Congress of Structural and Multidisciplinary Optimization, Amherst, New<br />
York, May 17-21, 1999.<br />
Bendsøe M.P. (1995). Optimization of structural topology, shape, and material, Springer<br />
Verlag, Berlin.<br />
Berthelot J.M. (1992). Matériaux composites. Comportement mécanique et analyse des<br />
structures, Masson, Paris.<br />
Bonnans J.F., Gilbert J.C., Lemaréchal C. and Sagastizabal C.A. (2003). Numerical<br />
optimization: theoretical and practical aspects. Springer, Berlin, Heidelberg New York.<br />
BOSS Quattro. SAMTECH S.A., Liège, Belgium. www.samcef.com.<br />
Braibant V. and Fleury C. (1985). An approximate concepts approach to shape optimal design,<br />
Computer Methods in Applied Mechanics and Engineering, 53, 119-148.<br />
Bruyneel M. and Fleury C. (2001). The key role of fibers orientations in the optimization of<br />
composite structures, Internal Report OA-58, LTAS-Optimization Multidisciplinaire,<br />
Université de Liège, Belgium.<br />
Bruyneel M. (2002). Schémas d’approximation pour la conception optimale des structures en<br />
matériaux composites, Doctoral Thesis, Faculté des Sciences Appliquées, Univesrité de<br />
Liège, Belgium.<br />
Bruyneel M. and Fleury C. (2002). <strong>Composite</strong> structures optimization using sequential<br />
convex programming, Advances in Engineering Software, 33, 697-711.<br />
Bruyneel M., Duysinx P. and Fleury C. (2002). A family of MMA approximations for<br />
structural optimization, Structural & Multidisciplinary Optimization, 24, 263-276.<br />
Bruyneel M. and Duysinx P. (2005). Note on topology optimization of continuum structures<br />
including self-weight, Structural & Multidisciplinary Optimization, 29, 245-256.
102<br />
Michaël Bruyneel<br />
Bruyneel M. (2006). A general and effective approach for the optimal design of fiber<br />
reinforced composite structures, <strong>Composite</strong>s Science & Technology, 66, 1303-1314.<br />
Bruyneel M. and Duysinx P. (2006). Note on singular optima in laminates design problems,<br />
Structural & Multidisciplinary Optimization, 31, 156-159.<br />
Bruyneel M., Morelle P. and Delsemme J.P. (2006). Failure analysis of metallic and<br />
composite structures with SAMCEF, NAFEMS Seminar – <strong>Materials</strong> Modelling,<br />
Niedernhausen near Wiesbaden, Germany, December 5-6.<br />
Bruyneel M., Colson B. and Remouchamps A. (2007). Discussion on some convergence<br />
problems in buckling optimization, to appear in Structural & Multidisciplinary<br />
Optimization<br />
Carpentier A., Michel L., Grihon S. and Barreau J.J. (2006). Buckling optimization of<br />
composite panels via lay-up tables. III European Conference on Computational<br />
mechanics – Solids, Structures and Coupled Problems in engineering, C.A. Mota Soares<br />
et al. (editors), Lisbon, Portugal, 5-9 June 2006.<br />
Chao L.P., Gandhi M.V. and Thompson B.S. (1993). A design-for-manufacture methodology<br />
for incorporating manufacturing uncertainties in the robust design of fibrous laminated<br />
composite structures, Journal of <strong>Composite</strong> <strong>Materials</strong>, 27 (2), 175-194.<br />
Chao L.P. (1996). Multiobjective optimization design methodology for incorporating<br />
manufacturing uncertainties in advanced composite structures, Engineering Optimization,<br />
25, 309-323.<br />
Cheng K. (1986). Sensitivity analysis and a mixed approach to the optimization of symmetric<br />
layered composite plates, Engineering Optimization, 9, 233-247.<br />
Cheng G. and Guo X. (1997). ε-relaxed approach in structural topology optimization,<br />
Structural Optimization, 13, 258-266.<br />
Colson B., Remouchamps A. and Grihgon S. (2007). Advanced computational structures<br />
mechanics optimization, 10 th SAMTECH Users Conference, March 13-14, Liège,<br />
Belgium.<br />
Dems K. (1996). Sensitivity analysis and optimal design for fiber reinforced composite discs,<br />
Structural Optimization, 11, 176-186.<br />
Diaconu C.G. and Sekine H. (2004). Layup optimization for buckling of laminated composite<br />
shells with restricted layer angles, AIAA Journal, Vol. 42, No 10, 2153-2163.<br />
Diaz A. and Bendsøe M.P. (1992). Shape optimization of structures for multiple loading<br />
situations using a homogenization method, Structural Optimization, 4, 17-22.<br />
Duvaut G., Terrel G., Léné F. and Verijenko V.E. (2000). Optimization of fiber reinforced<br />
composites, <strong>Composite</strong> Structures, 48, 83-89.<br />
Duysinx P. (1996). Optimization topologique : du milieu continu à la structure élastique,<br />
Doctoral Thesis, Faculté des Sciences Appliquées, Université de Liège, Belgium.<br />
Duysinx P. (1997). Layout optimization: a mathematical programming approach, DCAMM<br />
Report 540, Technical University of Denmark.<br />
Duysinx P. and Bendsøe M.P. (1998). Topology optimization of continuum structures with local<br />
stress constraints, Int. J. Num. Meth. Engng., 43, 1453-1478.<br />
Fine A.S. and Springer G.S. (1997). Design of composite laminates for strength, weight, and<br />
manufacturability, Journal of <strong>Composite</strong> <strong>Materials</strong>, 31 (23), 2330-2390.<br />
Fleury C. (1973). Méthodes numériques d’optimisation des structures, Internal Report SF-19,<br />
Laboratoire d’Aéronautique, Université de Liège.
Optimization of Laminated <strong>Composite</strong> Structures… 103<br />
Fleury C. and Braibant V. (1986). Structural optimization: a new dual method using mixed<br />
variables, Int. J. Num. Meth. Engng., 23, 409-428.<br />
Fleury C. (1993). Dual methods for convex separable problems, Optimization of Large Structural<br />
Systems, Volume I (G.I.N. Rozvany, editor), 509-530, Kluwer Academic Publishers, The<br />
Netherlands.<br />
Foldager J., Hansen J.S. and Olhoff N. (1998). A general approach forcing convexity of ply angle<br />
optimization in composite laminates. Structural Optimization, 16, 201-211.<br />
Foldager J.P. (1999). Design of composite structures, Doctoral Thesis, Special Report N° 39,<br />
Institute of Mechanical Engineering, Aalborg University, Denmark.<br />
Franco Correia V.M., Mota Soares C.M. and Mota Soares C.A. (1997). Design sensitivity<br />
analysis and optimal design of composite structures using higher order discrete models,<br />
Engineering Optimization, 29, 85-11.<br />
Fukunaga H. and Vanderplaats G.N. (1991a). Strength optimization of laminated composites<br />
with respect to layer thickness and/or layer orientation angle, Computers & Structures, 40<br />
(6), 1429- 1439.<br />
Fukunaga H. and Vanderplaats G.N. (1991b). Stiffness optimization of orthotropic laminated<br />
composites using lamination parameters, AIAA Journal, 29 (4), 641- 646.<br />
Fukunaga H. and Sekine H. (1993). Optimum design of composite structures for shape, layer<br />
angle and layer thickness distributions, Journal of <strong>Composite</strong> <strong>Materials</strong>, 27 (15), 1479-<br />
1492.<br />
Gay D. (1991).Matériaux composites, Hermès, Paris.<br />
Geier B. and Zimmerman R. (1994). <strong>Composite</strong> laminate stiffnesses and their derivatives,<br />
Advances in Design and Automation, Volume II (Gilmore B.J., Hoeltzel D.A., Dutta D. and<br />
Eschenauer H.A., editors), ASME 1994, 237-246.<br />
Goldberg D. (1989). Genetic algorithms in search, optimization and machine learning, Addison-<br />
Wesley Publishing Company Inc., Reading, MA.<br />
Grenestedt J.L. (1990). <strong>Composite</strong> plate optimization only requires one parameter, Structural<br />
Optimization, 2, 29-37.<br />
Grenestedt J.L. (1991). Lay-up optimization against buckling of shear panels, Structural<br />
Optimization, 3, 115-120.<br />
Grenestedt J.L. (1992). Lay-up optimization of composite structures, Doctoral Thesis, Rapport<br />
92-94, Dept. of Lightweight Structures, Royal Institute of Technology of Stockholm,<br />
Sweden.<br />
Grenestdt J.L. and Gudmundson P. (1993). Lay-up optimization of composite material<br />
structures, In: Optimal design with Advanced <strong>Materials</strong> (P. Pedersen, editor), 311-336,<br />
Elsevier Science Publ. B.V.<br />
Haftka R.T. and Gürdal Z. (1992). Elements of structural optimization, Kluwer Academic<br />
Publishers.<br />
Hammer V.B. (1997). Design of composite laminates with optimized stiffness, strength, and<br />
damage properties, Doctoral Thesis, DCAMM report S72, Technical University of<br />
Denmark.<br />
Harrison P., Le Riche R. and Haftka R.T. (1995). Design of stiffened composite panels by<br />
genetic algorithm and response surface approximations, AIAA/ASME/ASCE/AHS/ASC<br />
Structures, Structural Dynamics and <strong>Materials</strong> Conference, 36th and AIAA/ASME<br />
Adaptive Structures Forum, New Orleans, LA; UNITED STATES; 10-13 Apr. 1995.<br />
58-68.
104<br />
Michaël Bruyneel<br />
Henderson J.L., Gürdal Z. and Loos A.C. (1999). Combined structural and manufacturing<br />
optimization of stiffened composite panels, Journal of Aircraft, 36 (1), 246-254.<br />
Herencia J.E., Weaver P.M. and Friswell M.I. (2006). Local optimization of long anisotropic<br />
laminated fibre composite panels with T shape stiffeners, 47 th<br />
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and <strong>Materials</strong><br />
Conference, 1 – 4 May 2006, Newport, Rhode Island.<br />
Hirano Y. (1979a). Optimum design of laminated plates under shear, Journal of <strong>Composite</strong><br />
<strong>Materials</strong>, 13, 329-335.<br />
Hirano Y. (1979b). Optimum design of laminated plates under axial compression, AIAA<br />
Journal, 17 (9), 1017-1019.<br />
Hyer M.W. and Charette R.F. (1991). Use of curvilinear fiber format in composite structure<br />
design, AIAA Journal, 29, 1011-1015.<br />
Hyer M.W. and Lee H.H. (1991). The use of curvilinear fiber format to improve buckling<br />
resistance of composite plates with central circular holes, <strong>Composite</strong> Structures, 18,<br />
239-261.<br />
Kam T.Y. and Lai M.D. (1989). Multilevel optimal design of laminated composite plate<br />
structures, Computers & Structures, 31 (2), 197-202.<br />
Karandikar H. and Mistree F. (1992). Designing a composite material pressure vessel for<br />
manufacture: a case study in concurrent engineering, Engineering Optimization, 18,<br />
235-262.<br />
Kicher T.P. and Chao T.L. (1971). Minimum weight design of stiffened fiber composite<br />
cylinders, Journal of Aircraft, 8, 562-569.<br />
Kogiso N., Watson L.T., Gürdal Z. and Haftka R.T. (1994). Genetic algorithms with local<br />
improvement for composite laminate design, Structural Optimization, 7, 207-218.<br />
Kristinsdottir B.P., Zabinsky Z.B., Tuttle M.E. and Csendes T. (1996). Incorporating<br />
manufacturing tolerances in near-optimal design of composite structures, Engineering<br />
Optimization, 23, 1-23.<br />
Krog L.A. (1996). Layout optimization of disk, plate, and shell structures, Doctoral Thesis,<br />
Special Report N° 27, Institute of Mechanical Engineering, Aalborg University, Denmark.<br />
Krog L., Bruyneel M., Remouchamps A. and Fleury C. (2007). COMBOX: a distributed<br />
computing process for optimum sizing of composite aircraft box structures. 10 th SAMTECH<br />
Users Conference, March 13-14, Liège, Belgium.<br />
Lanzi L. and Giavotta V. (2006). Post-buckling optimization of composite stiffened panels:<br />
computations and experiments, <strong>Composite</strong> Structures, 73, 206-220.<br />
Le Riche R. and Haftka R.T. (1993). Optimization of laminated stacking sequence for<br />
buckling load maximization by genetic algorithms, AIAA Journal, 31 (5), 951-956.<br />
Lipton R.P. (1994). On optimal reinforcement of plates and choice of design parameters, Control<br />
& Cybernetics, 23 (3), 481-493.<br />
Liu B., Haftka R.T. and Akgun M. (2000). Two-level composite wing structural optimization<br />
using response surfaces, Structural & Multidisciplinary Optimization, 20, 87-96.<br />
Liu B., Haftka R. and Trompette P. (2004). Single level composite wing optimization based<br />
on flexural lamination parameters, Structural & Multidisciplinary Optimization, 26,<br />
111-120.<br />
Mahadevan S. and Liu X. (1998). Probabilistic optimum design of composite laminates, Journal<br />
of <strong>Composite</strong> <strong>Materials</strong>, 32 (1), 68-82.
Optimization of Laminated <strong>Composite</strong> Structures… 105<br />
Manne P.M. and Tsai S.W. (1998). Design optimization of composite plates: Part I – Design<br />
criteria for strength, stiffness, and manufacturing complexity of composite laminates, Journal<br />
of <strong>Composite</strong> <strong>Materials</strong>, 32 (6), 544-571.<br />
Martin P.M. (1987). Optimum design of anisotropic sandwich panels with thin faces, Engineering<br />
Optimization, 11, 3-12.<br />
Mateus H.C., Mota Soares C.M. and Mota Soares C.A. (1991). Sensitivity analysis and optimal<br />
design of thin laminated composite structures, Computers & Structures, 41 (3), 501-508.<br />
Miki M. (1982). Material design of composite laminates with required in-plane elastic<br />
properties, <strong>Progress</strong> in Science and Engineering of <strong>Composite</strong>s, (Hayashi, T., Kawata, K.<br />
and Umekawa, M., editors), 1725-1731, ICCM-IV, Tokyo.<br />
Morris A.J. (1982). Foundations of structural optimization: a inified approach. John Willey and<br />
Sons.<br />
Mota Soares C.M., Mota Soares C.A. and Franco Correia V. (1993). Optimal design of thin<br />
laminated composite structures. Topology design of structures (M.P. Bendsøe and C.A.<br />
Mota Soares, editors), 313-327, Kluwer Academic Publishers, The Netherlands.<br />
Mota Soares C.M. and Mota Soares C.A. (1995). A model for the optimum design of thin<br />
laminated plate-shell structures for static, dynamic and buckling behavior, <strong>Composite</strong><br />
Structures, 32, 69-79.<br />
Mota Soares C.M., Mota Soares C.A. and Franco Correia V. (1997). Optimization of<br />
multilaminated structures using high-order deformation models, Computer Methods in<br />
Applied Mechanics and Engineering, 149, 133-152.<br />
Olhoff N. (1989). Multicriterion structural optimization via bound formulation and mathematical<br />
programming, Structural Optimization, 1, 11-17.<br />
Papila M., Vitali R., Haftka R.T., and Sankar V. Optimal weight of cracked composite panels by<br />
an equivalent strain constraint. 42 nd AIAA/ASME/ASCE/AHS/ASC Structures, Structural<br />
Dynamics, and <strong>Materials</strong> Conference and Exhibit, 16-19 April 2001, Seattle, Washington.<br />
Pedersen P. (1987). On sensitivity analysis and optimal design of specially orthotropic laminates,<br />
Engineering Optimization, 11, 305-316.<br />
Pedersen P. (1989). On optimal orientation of orthotropic materials, Structural Optimization, 1,<br />
101-106.<br />
Pedersen P. (1990). Bounds on elastic energy in solids of orthotropic materials. Structural<br />
Optimization, 2, 55-63.<br />
Pedersen P. (1991). On thickness and orientational design with orthotropic materials, Structural<br />
Optimization, 3, 69-78.<br />
Potgieter E. and Stander N. (1998). The genetic algorithm applied to stiffness maximization of<br />
laminated plates: review and comparison, Structural Optimization, 15, 221-229.<br />
Radovcic Y. and Remouchamps A. (2002). BOSS Quattro: an open system for parametric<br />
design, Structural & Multidisciplinary Optimization, 23, 140-152.<br />
Rikards R., Abramovich H., Kalnins K. and Auzins J. (2006). Surrogate modeling in design<br />
optimization of stiffened composite shells, <strong>Composite</strong> Structures, 73, 244-251.<br />
Rion V. and Bruyneel M. (2006). Topology optimization of membranes made of orthotropic<br />
material, in: Selected Papers of Prof. Nguyen Dang Hung’s Students and Collaborators,<br />
(DeSaxcé G., Moës N., eds), Vietnam National University-HCM Publishing House.<br />
SAMCEF. Système d’analyse des milieux continus par éléments finis, SAMTECH S.A., Liège,<br />
Belgium. www.samcef.com.
106<br />
Michaël Bruyneel<br />
Schittkowski K., Zillober C. and Zotemantel R. (1994). Numerical comparison of nonlinear<br />
programming algorithms for structural optimization, Structural Optimization, 7, 1-19.<br />
Schmit L.A. and Farshi B. (1973). Optimum laminate design for strength and stiffness, Int. J.<br />
Num. Meth. Engng., 7, 519-536.<br />
Schmit L.A. and Farshi B. (1974). Some approximation concepts for structural synthesis, AIAA<br />
Journal, 12 (5), 692-699.<br />
Schmit L.A. and Fleury C. (1980). Structural synthesis by combining approximation concepts<br />
and dual methods, AIAA Journal, 18, 1252-1260.<br />
Sigmund O. (2001). Design of multiphysics actuators using topology optimization – Part 1:<br />
one-material structures, Computer Methods in Applied Mechanics and Engineering, 199<br />
(49), 6577-6604.<br />
Soeiro A.V., Conceição Antonio C.A. and Torres Marques A. (1994). Multilevel optimization<br />
of laminated composite structures, Structural Optimization, 7, 55-60.<br />
Stegmann J. and Lund E. (2005). Discrete material optimization of general composite shell<br />
structures, Int. J. Num; Meth. Engng., 62 (14), 2009-2027.<br />
Svanberg K. (1987). The Method of Moving Asymptotes - A new method for structural<br />
optimization, Int. J. Num. Meth. Engng., 24, 359-373.<br />
Svanberg K. (1995). A globally convergent version of MMA without linesearch, First World<br />
Congress of Structural and Multidisciplinary Optimization (Olhoff N. and Rozvany G.I.N.,<br />
editors, Pergamon Press), Goslar, Germany, May 28 – June 2, 1995, 9-16.<br />
Tauchert T.R. and Adibhatla S. (1984). Design of laminated plates for maximum stiffness,<br />
Journal of <strong>Composite</strong> <strong>Materials</strong>, 18, 58-69.<br />
Tauchert T.R. and Adibhatla S. (1985). Design of laminated plates for maximum bending<br />
strength, Engineering Optimization, 8, 253-263.<br />
Thomas H.L., Zhou M., Shyy Y.K. and Pagaldipti N. (2000). Practical aspects of commercial<br />
composite topology optimization software development, Topology Optimization of<br />
Structures and <strong>Composite</strong> Continua (Rozvany G.I.N. and Olhoff N., editors), 269-278,<br />
Kluwer Academic Publishers, The Netherlands, 2000.<br />
Thomsen J. (1991). Optimization of composite discs, Structural Optimization, 3, 89-98.<br />
Tsai, S.W. and Pagano N.J. (1968). Invariant properties of composite materials, <strong>Composite</strong><br />
<strong>Materials</strong> Workshop, Technomic Publishing, Westport, CT, 233-253.<br />
Tsai .S.W. and Hahn H.T. (1980). Introduction to composite materials, Technomic Publication<br />
Co. Westport.<br />
Upadhyay A. and Kalyanarama V. (2000). Optimum design of fiber composite stiffened panels<br />
using genetic algorithms, Engineering Optimization, 33, 201-220.<br />
Vanderplaats G.N. (1984). Numerical optimization techniques for engineering design: with<br />
applications, McGraw-Hill.<br />
Vermaut O., Bruyneel M. and Fleury C. (1998). Strength optimization of laminated<br />
composites using the approximation concepts approach, International Conference on<br />
Advanced Computational Methods in Engineering ACOMEN98 (Van Keer R.,<br />
Verhegghe B., Hogge M. and Noldus E., editors, Shaker Publishing B.V.), Ghent,<br />
Belgium, September 2-4embre 1998, 243-250.<br />
Watkins R.I. and Morris A.J. (1987). A multicriteria objective function optimization scheme<br />
for laminated composites for use in multilevel structural optimization schemes, Computer<br />
Methods in Applied Mechanics and Engineering, 60, 233-251.
Optimization of Laminated <strong>Composite</strong> Structures… 107<br />
Zhang W.H., Fleury C. and Duysinx P. (1995). A generalized method of moving asymptotes<br />
(GMMA) including equality constraints, First World Congress of Structural and<br />
Multidisciplinary Optimization (Olhoff N. and Rozvany G.I.N., editors, Pergamon Press),<br />
Goslar, Germany, May 28 – June 2, 1995, 53-58.
In: <strong>Composite</strong> <strong>Materials</strong> <strong>Research</strong> <strong>Progress</strong> ISBN: 1-60021-994-2<br />
Editor: Lucas P. Durand, pp. 109-128 © 2008 Nova Science Publishers, Inc.<br />
Chapter 3<br />
MAJOR TRENDS IN POLYMERIC COMPOSITES<br />
TECHNOLOGY<br />
W.H. Zhong<br />
Department of Mechanical Engineering and Applied Mechanics<br />
North Dakota State University, Fargo, ND<br />
R.G. Maguire<br />
Boeing 787 Program/Phantom Works, The Boeing Co. Seattle WA<br />
S.S. Sangari<br />
Boeing <strong>Materials</strong> & Processes Technology, The Boeing Company, Seattle, WA<br />
P.H. Wu<br />
Spirit Co., Wichita KS<br />
Abstract<br />
<strong>Composite</strong>s have been growing exponentially in technology and applications for decades.<br />
The world of aerospace has been one of the earliest and strongest proponents of advanced<br />
composites and the culmination of the recent advances in composite technology are realized in<br />
the Boeing Model 787 with over 50% by weight of composites, bringing the application of<br />
composites in large structures into a new age. This mostly-composite Boeing 787 has been<br />
credited with putting an end to the era of the all-metal airplane on new designs, and it is<br />
perhaps the most visible manifestation of the fact that composites are having a profound and<br />
growing effect on all sectors of society.<br />
It is generally well-known that composite materials are made of reinforcement fibers and<br />
matrix materials, and light weight and high mechanical properties are the primary benefits of a<br />
composite structure. Accordingly, the development trends in composite technology lie in 1)<br />
new material technology specifically for developing novel fibers and matrices, enhancing<br />
interfacial adhesion between fiber and matrix, hybridization and multi-functionalization, and<br />
2) more reliable, high quality, rapid and low cost manufacturing technology.<br />
New reinforcement fiber technology including next generation carbon fibers and organic<br />
fibers with improved mechanical and physical properties, such as Spectra®, Dyneema®, and<br />
Zylon®, have been developing continuously. More significantly, various nanotechnology<br />
based novel fiber reinforcements have conspicuously and rapidly appeared in recent years.<br />
Matrix materials have become as complex as the fibers, satisfying increasing demands for
110<br />
W.H. Zhong, R.G. Maguire, S.S. Sangari et al.<br />
impact resistant and damage tolerant structure. Various means of accomplishing this have<br />
ranged from elastomeric/thermoplastic minor phases to discrete layers of toughened materials.<br />
Nano-modified polymeric matrices are mostly involved in the development trends for matrix<br />
polymer materials. Technology for enhancing the interfacial adhesion properties between the<br />
reinforcement and matrix for a composite to provide high stress-transfer ability is more<br />
critically demanded and the science of the interface is expanding. Fiber/matrix interfacial<br />
adhesion is vital for the application of the newly developed advanced reinforcement materials.<br />
Effective approaches to improving new and non-traditional treatment methods for better<br />
adhesion have just started to receive sufficient attention. Multi-functionality is also an<br />
important trend for advanced composites, in particular, utilizing nanotechnology<br />
developments in recent years to provide greater opportunities for forcing materials to play a<br />
more comprehensive role in the designs of the future.<br />
More reliable and low cost manufacturing technology has been pursued by industry and<br />
academic researchers and the traditional material forms are being replaced by those which<br />
support the growing need for high quality, rapid production rates and lower recurring costs.<br />
Major trends include the recognition of the value of resin infusion methods, automated<br />
thermoplastic processing which takes advantage of the unique advantages of that material<br />
class, and the value of moving away from dependence on the large and expensive autoclaves.<br />
Introduction<br />
In sectors such as aerospace, wind energy, power transmission, marine, automotive and<br />
trucking, composites have been moving into the primary structure of wings, fuselages,<br />
chassis, hulls, and towers. In products such as sports goods and equipment, medical<br />
equipment, civil infrastructure, and dentistry composites are contributing to a market growth<br />
that will soon relegate homogeneous and isotropic materials to a niche category. The word<br />
“composite” is becoming synonymous with greater design flexibility and optimized materials<br />
utilization, leading to more opportunities for monolithic structural designs, less fasteners and<br />
holes, optimization of overall structural element architecture, improved fatigue and corrosion<br />
behavior, and high efficiency and maintainability. <strong>Composite</strong>s are also particularly suitable<br />
for structural health monitoring systems with the associated advantages of reduced<br />
conservatism in designs.<br />
As they have evolved over the past several decades, composites now are spreading out<br />
and leaving their early material forms and traditional processes, and incorporating new<br />
constituents from nano particles to smart additives to hybridizing to capture the best of all<br />
technologies. This has led to lean and efficient automated processes that will enable these<br />
new developments to be cost-effective in production and performance-enhanced in products<br />
that serve us all.<br />
Over the past several decades polymeric composites have matured and evolved,<br />
sometimes fitfully, but much of the time in a steady development driven by the increasing<br />
awareness among industries of the values available from combinations of matrices and fibers.<br />
The world of aerospace has been one of the strongest proponents of advanced composites,<br />
eventually converging on a combination of carbon fibers and thermosetting materials as the<br />
preferred choice for the harsh environment and complex loading of aerostructures. Structural<br />
materials applied for airplane structures from metallic materials, to composites and then nanomodified<br />
composite materials are being developed, see Fig. 1.
• Mechanical performance<br />
enhancements through alloying<br />
& heat treatments<br />
Isotropic Metal<br />
<strong>Materials</strong><br />
Major Trends in Polymeric <strong>Composite</strong>s Technology 111<br />
Continuing Trend Toward Increasing<br />
Capability of Engineering <strong>Materials</strong><br />
Anisotropic <strong>Composite</strong><br />
<strong>Materials</strong><br />
• Fiber orientation optimizing effects on<br />
strength & stiffness<br />
• Laminate tailoring for coupled<br />
deformations<br />
• Independent toughness<br />
improvements through polymer alloying<br />
and controlling phase morphology<br />
Figure 1. <strong>Materials</strong> for airplane structures.<br />
Nano-Modified<br />
<strong>Composite</strong> <strong>Materials</strong><br />
• Broad mechanical property<br />
improvements<br />
• Flammability and solvent resistance<br />
enhancements<br />
• Conductivity and CTE tailoring<br />
• Inherent color, optical qualities, etc.<br />
• Multi-functionality<br />
<strong>Composite</strong>s have been applied in various areas including aerospace, automotives,<br />
renewable energy structures (e.g. windmill blades as shown in Fig. 2), marines, sports and<br />
construction. . The steps for pursuing composite materials with ultra-light weight, super<br />
mechanical properties and multi-functionalities have been developing fast, in particular, the<br />
nanotechnology speed up and provide revolutionary opportunities for the new trends of<br />
composite development, which can be summarized in the Fig. 3.<br />
Figure 2. 38-meter European fiber glass windmill blade.
112<br />
Reinforcement:<br />
Advanced materials<br />
technology<br />
a. New fibers: PBO,<br />
UHMWPE fibers, etc.<br />
b. Nano-tech based:<br />
• 1D reinforcement:<br />
(i) Spun nanofibers<br />
(nanocomposites)<br />
(ii) Nano-scaled fibers (e.g.<br />
nano-PAN based CF)<br />
(iii) Ropes, yarns, bundles<br />
(e.g. CNT ropes, yarns)<br />
• 2D reinforcement:<br />
Film, sheet, mat, etc.<br />
(e.g. Bucky paper,<br />
nano paper)<br />
c. Hybridization: e.g.ARALL<br />
W.H. Zhong, R.G. Maguire, S.S. Sangari et al.<br />
Improve properties for FRP composites<br />
Interface:<br />
a. Nano-coating fibers<br />
(e.g., CNTs coated GF)<br />
Improve interfacial adhesion<br />
b. Reactive nano-matrix<br />
(nanocomposites)<br />
(e.g. reactive nano-epoxy:<br />
improve wetting& adhesion)<br />
Reliable and low cost<br />
manufacturing technology<br />
Matrix:<br />
Nano-filled Resin:<br />
(nanocomposites)<br />
a. Simple physical mixture of<br />
polymer and nano-fillers<br />
b. Integration of nano<br />
polymer systems<br />
(e.g. CNTs-epoxy, GNFsepoxy)<br />
c. multi-functionalization<br />
Figure 3. Routes of property enhancement for macro-composites.<br />
With the success and proliferation of state-of-the-art composite materials in many areas<br />
including aerospace and renewable energy structures, there are new opportunities that are<br />
being considered now that the bar has been successfully raised with the 787. Some of these<br />
will be explored in this chapter.<br />
I. Nanocomposites and Multifunctional <strong>Materials</strong>: carbon nanotubes (CNTs),<br />
nanofibers, and nanoplatelets have a natural affinity for polymeric materials and their<br />
inclusion in composites offers the promise of multi-functionality: electrical/thermal<br />
conductivity, acoustic damping and optical functionalities may be combined with<br />
load bearing capabilities.<br />
II. Hybridization: we see a new generation of metal/composite combinations deriving<br />
from early attempts such as GLARE®, ARALL®, titanium/graphite (TiGr), but<br />
utilizing new processes for the applications of monodisperse metallic coatings of<br />
high mechanical properties and durability, as well as multi-functionality.<br />
III. Alternatives to Carbon Fibers and Next Generation Carbon Fibers: as the<br />
understanding of the shortcomings of organic fibers such as UHMWPE increases,<br />
new approaches to interface enhancements are being developed which offer the
Major Trends in Polymeric <strong>Composite</strong>s Technology 113<br />
benefit of a new supply chain for highly capable structural fibers. New generations<br />
of continuous carbon fibers are also being developed both on the micro and nano<br />
scales with improved performance and functionality.<br />
IV. Processing Technologies: the autoclave has been the mainstay for many years, but<br />
growing trends toward lean manufacturing make this a roadblock to improved<br />
efficiency. Continuous and inexpensive processing is making the curing step a part<br />
of the overall lean philosophy in composite manufacturing. A growing trend is the<br />
move away from prepregs to the family of materials and processing called resin<br />
infusion (RI). This includes Resin Transfer Molding (RTM), Vacuum-Assisted<br />
Resin Transfer Molding (VARTM), Resin Film Infusion (RFI), and the development<br />
of new continuous processes. All this is based on low-cost material forms of neat<br />
resin and fiber preforms that allow the manufacturer to put the two together in<br />
proprietary and efficient ways.<br />
V. Other trends include a new generation of thermoplastics being developed to serve the<br />
growing automation of TP parts, smart materials and structures which can de-couple<br />
requirements and reduce weight, low-cost carbon fibers from bio sources, and others.<br />
1. Nanocomposites and Multifunctional <strong>Materials</strong><br />
The definition of nanocomposites covers a variety of systems such as one-dimensional, twodimensional<br />
and, three-dimensional materials made of distinctly dissimilar components and<br />
mixed at the nanometer scale for achieving drastically enhanced properties. To obtain multifunctionality<br />
in nanocomposites, nanoparticles with high aspect ratio have been successfully<br />
employed. This denotes being functional in one property while either achieving new<br />
properties that are unknown in the individual components or improving and maintaining other<br />
intrinsic properties.<br />
Nanocomposites can be used as matrices for nano/macro-composites, or traditional<br />
composites when micro-scale fibers are included. Nano constituent composites have also been<br />
made into nanoscale fibers through spinning methods. To date, the creation of such new<br />
materials has resulted in:<br />
• Enhanced mechanical properties: strength, stiffness; toughness, impact resistance,<br />
structural durability, etc.<br />
• Improved electrical conductivity<br />
• Improved thermal conductivity and thermal management<br />
• Improved flame resistance, thermal stability and increased service temperatures<br />
• Enhanced acoustic damping<br />
• Improved dimensional stability (low or tailored coefficient of thermal expansion)<br />
• Enhanced tribological properties (wear, abrasion resistance, hardness)<br />
• Improved barrier properties and environmental controls<br />
• Decreased permeability<br />
• Reduced shrinkage
114<br />
W.H. Zhong, R.G. Maguire, S.S. Sangari et al.<br />
In many cases the product is a multifunctional material that has several of the above<br />
characteristics and properties in combination. Nanocomposites can improve many kinds of<br />
functionalities and typically can result in multi-functionalities. These functionalities include:<br />
This can be extremely attractive for engineered applications where costs of testing,<br />
evaluations, qualifications and certifications of each new material are high, and sometimes<br />
prohibitive. The prospect of a single material for multiple functions can therefore be very<br />
compelling, both for engineered performances and the non-recurring costs associated with<br />
bringing the material to readiness for the designer.<br />
Bulk form:<br />
Enhanced<br />
functionalities:<br />
- electrical<br />
- thermal<br />
- flame resistant<br />
- damping<br />
- mechanical …<br />
As loose fillers:<br />
Nano-fillers + polymers<br />
--> Nanocomposites<br />
…<br />
Nano-fillers: CNTs, CNFs, etc.<br />
Spinning fibers: 1D<br />
? 2D sheet/paper<br />
Enhanced<br />
functionalities:<br />
- electrical<br />
- thermal<br />
- flame resistant<br />
- damping<br />
- mechanical…<br />
+ Fiber<br />
Combined further:<br />
1D: ropes, yarns, bundles<br />
2D: sheet/paper forms<br />
+Matrix<br />
+Matrix<br />
FRP composites:<br />
(Hybrid composites)<br />
Enhanced<br />
functionalities:<br />
- electrical<br />
- thermal<br />
- flame resistant<br />
- damping<br />
- mechanical…<br />
…<br />
Figure 4. Nano-scale materials in different forms for applications in composites.<br />
Nano-sized materials as nano-fillers for nanocomposites can be classified into three<br />
categories according to the shape: particles (metals, metal compounds, organic and inorganic<br />
particles), fibrous materials (nanotubes, nanofibers, nanowires), and layered materials
Major Trends in Polymeric <strong>Composite</strong>s Technology 115<br />
(graphene platelets, clay). These nano-fillers have exceptionally high specific surface areas so<br />
the overall amount of interfacial area is enormous if the nano-fillers are adequately dispersed<br />
within the matrix. This can result in creating various functionalities including mechanical,<br />
physical and other classes of properties. The large specific surface areas are highly desirable<br />
for stress transfer between the nano-fillers and the matrix, as well as providing increased<br />
chemical reactivity and energy levels compared to conventional bulk materials. Almost all<br />
nano-scale materials can be used as loose fillers for making nanocomposites. Fibrous nanofillers<br />
can be made into yarns/bundles, mats, braids, sheets/papers, which could be used in<br />
fiber reinforced polymer (FRP) composites. Nanocomposites can be applied in various forms<br />
such as coatings, films/sheets, spinning fibers, bulk materials as well as matrices for FRP<br />
composites due to the nano-fillers’ dramatic capability in enhancing functionalities for the<br />
polymer materials. When these nano-fillers or nanocomposites are used in fiber reinforced<br />
polymer (FRP) composites they become hybrid composite materials, which may exhibit<br />
multifunctional properties as illustrated in Fig. 4.<br />
Nanocomposites and hybrid composites with various functionalities have vast<br />
applications in structural applications in aircraft, space vehicles and renewable energy<br />
assemblies; impact protection systems; thermal management components; fuel cells;<br />
electronic devices, sensors, actuators, various functional coatings, electrostatic dissipation<br />
(ESD) and electromagnetic interference (EMI) radiation protection, lightning strike<br />
protection, etc.<br />
It has been established that improvements in the properties of nanocomposites are<br />
strongly affected by many factors including nano-filler size distribution, shape, aspect ratio,<br />
concentration, degree of dispersion, characteristics of the matrix, interactions between the<br />
filler and the matrix, and interfaces between the nano-particles themselves.<br />
According to the potential functionalities, nano-scale fillers can also be divided into (1)<br />
carbon types, such as CNT, carbon nanofibers (CNF, VGCF, or GNF), and graphite<br />
nanoplatelets (GNP), and (2) non-carbon types, such as nano-clay, POSS, nano-silica, metal<br />
nanoparticles and nano metal oxide particles. Nanocomposites with carbon type nano-fillers<br />
are mainly utilized for improvements of damping, mechanical, electrical and thermal<br />
properties. Nanocomposites with non-carbon type nano-fillers are predominantly used for<br />
flame retardency, improved barrier property, creep resistant, tribological properties and to<br />
some extent in early works, mechanical property enhancements.<br />
Nanocomposites have been undergoing rapid developments and significant progress has<br />
been made in the fields of nanocomposites and nanocomposite multi-functionalities over the<br />
past few decades. However, there are an abundant amount of questions and challenges left to<br />
be solved before taking full advantage of nano-scale fillers for development of stable, highquality<br />
nanocomposites. These include types, purity levels and polymer types<br />
(thermoset/thermoplastic), structure characteristics, viscosity at room and/or elevated<br />
temperature, appropriate treatment methods to be applied to the nano-fillers which will affect<br />
the interaction between the nano-fillers and polymer matrix, etc. In order to create a blend<br />
with controlled ratios of components and a well-dispersed nano-filler into the polymer matrix<br />
effective mixing methods and processing parameters should be understood and applied. Only<br />
when a complete understanding of these issues is established will the performance of<br />
nanocomposites with desired properties/functionalities be fully realized.<br />
Although many researchers have conducted remarkably successful experiments for<br />
achieving high performance nanocomposites, and obtained many encouraging empirical
116<br />
W.H. Zhong, R.G. Maguire, S.S. Sangari et al.<br />
results, there is still a critical lack of comprehensive mathematical modeling that is needed to<br />
be used to make effective predictions for processing-structure-property relationship or when<br />
evaluating the multi-functionalities. An example is the wide array of electrical conductivity<br />
percolation threshold values for certain nano-fillers (e.g. CNT or CNF), when combined with<br />
the goal of improved strength and modulus. Modeling can increase the speed of selection and<br />
reduce the scale of actual testing, a huge advantage for bringing new products quickly to<br />
market. Mathematical modeling also provides practical benefits to industry in developing<br />
modeling capabilities for designing new materials. With the added dimensions of the<br />
nanocomposites options, conventional testing and down-selection for choices between the<br />
various and numerous materials can quickly become unfeasible.<br />
Nanostructures have unique physical and chemical properties different from bulk<br />
materials of the same chemical composition. The mechanical, electrical, thermal and<br />
magnetic properties of composites consisting of an insulating matrix and dispersed<br />
nanoparticles have been extensively studied over the past few decades. The significant<br />
progress in the understanding of nanocomposite systems within recent years has shown that<br />
multifunctional nanocomposites offer both great potential and great challenges, marking it as<br />
a highly active field of research. The research is continuing at an increasing pace, as the<br />
requirements for stronger and lighter materials are needed by a variety of industries.<br />
However, much research effort is continuing toward the development of new processing<br />
techniques that control the purity and dispersions of nanoparticles in the polymers.<br />
2. Hybridization<br />
As composites develop and improve, and their applications grow, there will be inevitably, the<br />
realization that there are limitations and compromises in changing from metals to “nonmetals”.<br />
In many cases, the logical thought process is to consider how the best of both<br />
materials can be in included in hybrids. One of the major subsets of this segment of materials<br />
is based on the concept of going with one of the composites most valuable characteristics, that<br />
is, lamination of individual plies. In some cases the concept that comes most readily to mind<br />
is the replacement of one or more composite plies with metal foil or sheet and these are<br />
generally referred to as Fiber Metal laminates or FMLs.<br />
FMLs were first developed at Delft University in the Netherlands in the early 1980s, and<br />
marketed by Aluminum Company of America (Alcoa), combining sheets of aluminum in an<br />
alternating pattern with plies of traditional composite. As a result of this hybridization FMLs<br />
can theoretically combine the best of both the metal and the composite materials. The first<br />
FML was ARALL® (ARamid-ALuminum Laminate), a combination of aluminum and<br />
aramid/epoxy. This fiber-aluminum adhesive-bonded laminate is a super-hybrid composite<br />
material, which has many attractive properties such as good damage tolerance property, very<br />
high fatigue crack growth resistance, and high static strength along the fiber direction. The<br />
characteristics of ARALL® also include low density and resistance to the effects of<br />
temperature, humidity and acidity/alkalinity, etc.<br />
But greater applications became apparent if the aramid fiber composites were replaced by<br />
the ubiquitous glass fibers composites. In the 1980s, Delft developed a glass/epoxy FML<br />
called GLARE (GLAss-REinforced) composed of thin layers of aluminum sheet or foil<br />
interspersed with layers of fiberglass composite prepreg. The pre-preg layers may be aligned
Major Trends in Polymeric <strong>Composite</strong>s Technology 117<br />
in various directions to accommodate the loading and because of this characteristic, it is truly<br />
a composite laminate with tailorable in-plane properties, and with the capability to add the<br />
aluminum plies in various locations and arrangements through the stack up, but with<br />
processing properties similar to bulk aluminum sheet metal. Its major advantages over<br />
conventional aluminum are lower density, better fatigue resistance (cracks are inhibited in<br />
growth due to the restraining effects of the adjacent composite plies) and better resistance to<br />
impact. An important consideration is the matching of the cure temperature of the composite<br />
with the thermal effects on the metal. For example, if the composite requires a 350°F degree<br />
cure, the aluminum alloy must be able to sustain its performace after exposure to that<br />
temperature because it is the total laminate that must go through the autoclave process. The<br />
original GLARE® used 250°F curing fiberglass composite, with an appropriate aluminum<br />
alloy, but in situations where a higher Tg is required for design thermal exposure, a 350F<br />
version was created and has been dubbed “New GLARE”.<br />
Another interesting version of FML is titanium-graphite (TiGr) laminates which consist<br />
of layers of titanium interleafed through the thickness of a Carbon Fiber Reinforced Plastic<br />
(CRFP) laminate. TiGr offers advantages over metallic structures in terms of weight, fatigue<br />
characteristics, damage tolerance, and design flexibility, and also advantages over traditional<br />
composite materials through higher bearing capabilities, greater toughness, and an expanded<br />
design space. There has been extensive testing in the industry to support development of<br />
mechanical properties for TiGr with various epoxy prepreg systems and success in<br />
optimization of surface preparation has resulted in extremely robust and environmentally<br />
durable TiGr. Production-related issues such as scale-up, compound contours, drilling and<br />
trimming, NDI, repair methods and even automation have been addressed for feasibility and<br />
optimization.<br />
These extreme hybrid composites have great potential for use in many applications<br />
including aircrafts. However, due to the large difference in thermal expansion coefficient of<br />
both the fiber and metal, and the anisotropy of the composites layers combined with isotropic<br />
metals, large residual stresses can be built up during the curing cycle which could cause an<br />
unsuitable residual stress system that may seriously hinder its outstanding performance. To<br />
regulate the residual stresses with controlled layups is necessary. In addition, there are many<br />
factors which influence the performance of these FMLs due to three kinds of material<br />
constituents involved, and the two interfaces: fiber/resin and metal/resin. To control the<br />
quality of the FML materials is critically important for the applications in practice.<br />
Other types of hybrids include co-mingling of different fibers for multifunctionality of<br />
the composite ply. Co-mingling of carbon fibers and glass fibers can add a softness to a<br />
laminate in places that need dimensional flexibility. Co-mingling of carbon and<br />
thermoplastic fibers can add toughness to strength and stiffness in a part. In some<br />
commercial products, such as the Cytec Priform technology, the thermoplastic actually melts<br />
during the cure and disperses in a controlled way throughout the matrix to form a minor phase<br />
of toughening material.<br />
A new trend is the coating of polymeric composites with metals for multiple purposes<br />
ranging from electrical conductivity to thermal management to surface hardness. MesoScribe<br />
Technologies has a Direct Write Thermal Spray in which fine powders are injected into a<br />
small thermal plasma and accelerated through an aperture to make patterns on composites<br />
without a cure cycle or masking, and is adaptable to large and highly contoured surfaces.
118<br />
W.H. Zhong, R.G. Maguire, S.S. Sangari et al.<br />
Integran has its Nanovar technology in which finely grained metal is applied to a composite<br />
surface to increase wear in leading edges and for Invar tool repair. The reduced grain size<br />
leads to increased hardness and strength by the inverse relationship to the square root of the<br />
grain size. Other methods are in development to apply monodisperse metallic coating to<br />
composites for weight reduction, improved surface finish, alternatives to environmentally<br />
unacceptable coatings, and greater design freedom.<br />
In the nano-scale composites, hybridization of nanoparticles offers the potential of a rich<br />
soup in which some additives e.g. CNTs, are added for mechanical and electrical properties,<br />
some, e.g. POSS (Polyhedral Oligomeric Silsesquioxane) or nanoclay can be added for flame<br />
resistance, and others, e.g. nano gold particles, can be added for color.<br />
As we become more fluent in the use of new materials and less prejudiced in the use of<br />
older materials hybridization will become a natural trend for those who want it all.<br />
3. Alternatives to Carbon Fibers and Next Generation Carbon<br />
Fibers<br />
Alternatives to Carbon Fibers With the growth of applications of composites using the high<br />
specific strength and stiffness of carbon fibers, comes also the growing demand for those<br />
fibers in many sectors of industry: aerospace, energy, transportation, infrastructure, medical,<br />
sports equipment, etc. There is also the issue of galvanic corrosion in some applications<br />
where carbon forms one of the three elements of a circuit with metals like aluminum, and<br />
water. As a result there is a growing interest in non-carbon fibers, many of which have been<br />
looked at in the past, but now being considered with a hungrier eye, if some of their<br />
shortcomings can be overcome without sacrificing their benefits, especially in tensile<br />
strength.<br />
One example is liquid crystal polymers (LCP). Celanese developed a thermotropic<br />
polyester-polyarylate LCP in the mid 1970’s and commercialized the Vectra family of resins<br />
in 1985. Vectran® is an example of an LCP fiber. The molecules of the liquid crystal<br />
polymer are rigid and position themselves into randomly oriented domains. The polymer<br />
exhibits anisotropic behavior in the melt state, leading to the term thus the term “liquid crystal<br />
polymer.” The molten polymer is extruded through spinneret holes and the molecules align<br />
parallel to each other along the fiber axis. The highly oriented fiber structure results in high<br />
tensile properties.<br />
Vectran differs from other high-performance non-carbon fibers, aramid and ultra-high<br />
molecular weight polyethylene (UHMWPE), in that is thermotropic, melt-spun, and melts at a<br />
high temperature. Aramid fiber is lyotropic, solvent-spun and does not melt at high<br />
temperature. UHMWPE fiber is gelspun and melts at a relatively low temperature. In all<br />
these fibers the high modulus/high tensile strength is achieved through the oriented linear<br />
molecules called microfibrils. And all these fibers have an order of magnitude lower<br />
compression strength than tensile strength.<br />
In general, organic fibers, such as aramid fiber (e.g. Kevlar®), ultra high molecular<br />
weight polyethylene (UHMWPE) fiber (e.g. Spectra®), and Poly(p-phenylene-2,6benzobisoxazole)<br />
(PBO) fiber (e.g. Zylon®), have excellent mechanical and physical<br />
properties. Kevlar® provides excellent impact resistance and is one of the lightest structural
Major Trends in Polymeric <strong>Composite</strong>s Technology 119<br />
fibers available on the market today, which has been widely used both in soft body armor<br />
applications, and as reinforcement for hard armor, helmets and electronic housing protection.<br />
UHWMPE fibers, such as Spectra® and Dyneema®, are a type of ultra lightweight, highstrength<br />
polyethylene fibers. High damage tolerance, non-conductivity and flexibility, a much<br />
higher specific strength and modulus and energy-to-break, low moisture sensitivity, and good<br />
UV resistance, all make this fiber a good aramid alternative. These fibers are typically used in<br />
ballistic and high impact composite applications. Zylon® consists of a rigid chain of<br />
molecules of poly(p-phenylene-2,6-benzobisoxazole, PBO. It has excellent tensile strength<br />
and modulus. Fabrics made from Zylon ® are found in both ballistic and composite<br />
applications.<br />
In addition to the attractive mechanical properties of organic fibers, albeit limited in their<br />
current form, it should be noted that they are highly valuable by the key industries of the U.S.<br />
and offer the significant advantage in the avoidance of galvanic corrosion with aluminum and<br />
certain other metals. With organic fiber composites, the corrosion threat is avoided and the<br />
cost and weight benefits would be enormous to the aerospace and other industries if other<br />
critical properties could be enhanced. In addition, there is the economic challenge now of the<br />
growing applications for carbon fibers and the resultant shortages of carbon composite<br />
materials, which may impact the economy of the US. Engineering conferences for the past<br />
two years have focused on that increasing shortage and what technical and scientific<br />
alternatives are available. How to make organic fibers a viable alternative to carbon fibers for<br />
structural applications is often discussed in these forums within the genre of nanocomposites.<br />
This category of research holds great promise to enable that technology sector and can<br />
accelerate the fulfillment of the promise of multi-functional materials.<br />
Typical characteristics of these organic fibers include non-polar chemical structures and<br />
crystalline chains. It is these structural characteristics that impart the advanced mechanical<br />
properties on to the fibers. For example, UHMWPE fiber obtains its high strength from the<br />
straightening of long polymer chains by taking advantage of the strong covalent bonds in the<br />
backbone of the monomer. The modulus of the fiber is proportional to the draw ratio which<br />
controls the degree of crystallinity. The main benefits of a UHMWPE continuous fiber<br />
include high specific strength and moduli, leading to a lower weight for a given design load.<br />
The chemical neutrality of the fiber surface leads to a high degree of corrosion resistance;<br />
there are no places to allow for a concentrated attack on the surface. In addition, the<br />
anisotropic nature of the fiber allows for low coefficients of thermal expansion, meaning<br />
dimensional stability of the finished composite product.<br />
However, the non-polar chemical structure and resulting lack of reactive groups on the<br />
organic fiber surface lead to low surface energy, and thus leads to difficulty in obtaining good<br />
wetting and adhesion at the fiber/matrix interface. This low surface energy requires that the<br />
matrix material be of an even lower energy to achieve sufficient wetting and adhesion,<br />
ultimately realizing strong bond at the fiber/matrix interface. This results in the limited<br />
applications of the organic fibers because many properties of the composites are determined<br />
by the transfer ability of the fiber/matrix interface. To tackle the problem, various surface<br />
treatments to improve the interfacial wetting and adhesion, are applied. There appears to be<br />
an absence of a good means to alter the fiber without sacrificing its desirable properties. It is<br />
concluded that novel cost-effective methods for improving the interfacial adhesion between<br />
the organic fibers and the polymer matrix are vital to the full realization of their potential as<br />
structural materials.
120<br />
W.H. Zhong, R.G. Maguire, S.S. Sangari et al.<br />
For aramid and UHMWPE fibers, silane coupling treatments (effective for glass fibers)<br />
and oxidation treatments (effective for carbon fibers) are not effective in improving the<br />
interfacial strength [1-6]. For UHMWPE fibers, many treatment methods including nitrogen<br />
ion implantation, nitrogen plasma, fast atom beams, laser ablation, chain disentanglement,<br />
high power ion beam treatments, and cold plasma [7-13] have been used. Such approaches<br />
show some improvement in interfacial properties, but also can degrade the mechanical<br />
properties of the fibers by damaging the chain structure of the UHMWPE fiber and result in<br />
formation of amorphous hydrogenated carbon. Recently, the effectiveness of an atmospheric<br />
plasma was demonstrated for dramatic improvement in the adhesion of polyetheretherketone<br />
(PEEK) composites to epoxy [14]. This atmospheric plasma was shown to be an important<br />
potential strategy for improving the interfacial adhesion between organic fibers and polymer<br />
resins.<br />
A nanotechnology approach was recently developed by Dr. Zhong’s group in which<br />
conventional epoxy resins are converted into reactive nano-epoxy resins. Unlike the<br />
conventional epoxy resins that require carbon fibers to be surface oxidized/treated before<br />
being impregnated, the nano-epoxy resins contain reactive graphite nanofibers which can<br />
improve the wettability and adhesion properties between UHMWPE fibers and the resin<br />
matrix [15-19].<br />
Next Generation Carbon Fibers – Continuous Nanoscale Carbon Fibers Traditional Carbon<br />
fibers have high strength, high modulus and attractively low density. The high strength-toweight<br />
ratio combined with superior stiffness has made carbon fibers the material of choice<br />
for high performance composite structures in the aerospace, defense and other industries.<br />
Polymer fibers, which leave a carbon residue and do not melt upon pyrolysis in an inert<br />
atmosphere, are generally considered candidates for carbon fiber production. It is known that<br />
the structural perfection of precursor fibers is the most crucial factors on the strength of<br />
carbon fibers. Imperfections (such as surface defects, bulk defects and others) in the precursor<br />
fibers are likely to be translated to the resulting carbon fibers, and the amounts and sizes of<br />
structural imperfections directly determine the final fiber strength. The fundamental<br />
approach/solution for improving the strength of carbon fibers is to reduce the amounts and<br />
sizes of numerous types of defects in the precursor and there has been a clear and continuing<br />
trend among commercial carbon fiber suppliers in achieving higher strengths through this<br />
approach.<br />
There is also a growing interest in having thinner plies of composite material without loss<br />
of strength or stiffness and thinner plies means thinner fibers. Historically the means of<br />
producing very small diameter (down to submicron range) has been electrospinning. Since the<br />
1930s electrospinning has been used on nylon and other polymers to achieve small fibers to<br />
provide filtering media and other applications. In the process a strong electric field acts on a<br />
polymer solution resulting in a polymer stream which solidifies through the evaporation of<br />
the solvent.<br />
This can also be applied to a polyacrylonitrile (PAN) copolymer (precursor) to produce<br />
nanofibers with diameters in the nanoscale range with the potential of ultimately producing<br />
continuous nano-scaled carbon fibers with strengths and stiffness much higher than<br />
conventional micro scale carbon fibers. Additionally, since diameters of the electrospun PAN<br />
nanofibers can be further reduced by stretching and carbonization processes, the resulting<br />
nano-scaled carbon fibers can have diameters of less than 100 nm. When incorporated into a
Major Trends in Polymeric <strong>Composite</strong>s Technology 121<br />
matrix composite this could yield a prepreg ply about two orders of magnitude less thickness<br />
with the same strength and stiffness as conventional prepreg offering benefits in weight<br />
critical structure.<br />
4. Processing Technologies<br />
Out-of-Autoclave Processing: Although the autoclave has served the composite industry well<br />
providing structural integrity as well as thermal curing, there is a growing demand for a leaner<br />
means of curing parts. Being able to cure parts in a continuous stream like a pizza oven is the<br />
Lean Manufacturing teams dream. But even a common oven can provide leaner flow.<br />
The economic advantage of an oven process for structural composites is large.<br />
Autoclaves are an order of magnitude more expensive than ovens with the same temperature<br />
uniformity. The process flow and batch constraints of autoclaves can potentially be<br />
eliminated with ovens. And liquid nitrogen systems used to prevent fires would not be<br />
needed. Also there is the possibility of part growth that may limit the autoclave usefulness.<br />
Such an example is seen in the windmill blades in which designs are growing faster than an<br />
autoclave can be depreciated. Blades of necessity have been hot bag cured for many years<br />
and as they surpassed the 38-meter length the designs have been using carbon fiber in place of<br />
glass fiber composites. There is also the issue of large-scale complex parts such as racing<br />
sailboat hulls that would require an autoclave of the scale that not many groups can afford.<br />
These also have been “cooked in a tent” for many years.<br />
In the past there have been attempts to utilize UV curing, e-beam curing, X-ray curing,<br />
oven curing, tent curing, hot press curing, continuous pultrusion curing, many forms of resin<br />
infusion, etc., all with various degrees of success or the lack of significant applications. The<br />
key to success is the material. If they can be developed to have no voids in the uncured<br />
laminate and no volatile components in the resin system then the pressure element becomes<br />
moot. The concerns include matching the fiber volume associated with autoclave cures and<br />
processing variable that can impact design allowables. One of the noted successes in this<br />
trend is oven cured epoxy carbon fiber systems approved by the FAA for structural<br />
applications on civil aircraft (Agate Program allowables database is FAA approved) [20].<br />
Some of these systems are improved with high (>30 in Hg). Generally if a void free<br />
uncured laminate can be achieved either through hot debulking, ultrasonic compaction or<br />
other means, vacuum bag pressure in an oven will be sufficient to achieve a structural part.<br />
There are many new methods that involve either single or double diaphragms to hold the<br />
part while being vacuum-cured. One of these is the double diaphragm resin infusion process<br />
RIDFT of Florida State University High Performance <strong>Materials</strong> Institute which should offer<br />
lower tooling costs and shorter cycle times. Another novel approach is that of Quickstep®<br />
developed jointly with the Australian research organization CSIRO, it is based on a liquid<br />
filled container in which a lightweight mold floats on one of the flexible faces of the pressure<br />
chambers [21]. The container is filled with a heat transfer fluid which is circulated through<br />
the chamber to rapidly heat and cool the mold. The process enables the cure cycle to be<br />
stopped and restarted at any time and parts of the laminate to be left uncured. Parts can be<br />
consolidated and formed and then final cured in-situ at a later time. It is being developed<br />
further with several universities and the National center for <strong>Composite</strong>s.
122<br />
W.H. Zhong, R.G. Maguire, S.S. Sangari et al.<br />
In-situ compaction and curing is another approach that will have great value for lean and<br />
rapid production of composites using unidirectional composite prepreg tape or tow forms.<br />
The uni-prepreg is wound onto a mandrel and heat applied from several IR sources to the<br />
material as it is placed. There is also work being done at NASA on E-Beam in-situ curing<br />
using low energy processing [22].<br />
Resin Infusion Processes: Resin infusion although it is a generally an out-of-autoclave set of<br />
processes, is itself a highly significant trend in composites processing. It covers many<br />
processes such as RTM, VARTM, RFI and each of these major subsections has many<br />
variations (including some which go back into the autoclave), but the common factor among<br />
all of these is that the resins and the fibers are marketed separately and the customer fulfills<br />
the impregnation process within their own manufacturing planning. This can offer<br />
opportunity for customization and optimization for specific user needs and as well a<br />
significant cost savings in materials since the costly impregnation process is now done inhouse.<br />
Several of the major composite material suppliers forecast growing markets for the<br />
infusion resins and dry fiber preforms as compared to the more historical prepreg material<br />
forms.<br />
Resin infusion is a closed low pressure process for the manufacture of complex-shape,<br />
high-strength and lightweight composite parts for a wide range of aerospace, automotive,<br />
marine and satellite applications. In this process, a resin system is drawn into a dry fiber<br />
laminate in a mold where it can cure to form the finished part. RTM is an infusion process<br />
that employs an injection system to transfer a mixture of liquid resin and catalyst into a closed<br />
mold containing a preform, which is preset fiber mats. The resin is injected under a controlled<br />
pressure by a carefully designed pattern of inlet ports and vent holes. This guarantees that the<br />
fibers become fully wet and produce a low void content and high fiber-volume composite<br />
part. Fiber volumes approaching the 65% values, which is typical for prepreg lay-up<br />
techniques, have been achieved with RTM. Subsequent curing of the resin forms a net-shape<br />
part with good dimensional tolerances. The size and complexity of the part significantly<br />
influences the cycle times.<br />
VARTM is an adaptation of the RTM process and is generally used to manufacture parts<br />
for marine, ground transportation and infrastructure applications. The process uses an open<br />
mold cavity which is laid up with a preform and covered with a vacuum bag made of air<br />
impervious films such as nylon or silicone film. The air is expelled from the preform<br />
assembly using a vacuum pump. A liquid resin is allowed to infuse into the mold from an<br />
external reservoir after all air leaks are eliminated and the system is equilibrated. A high<br />
permeability resin distribution medium is placed on top of the preform to facilitate the resin<br />
flow over the lateral extent of the part. The system is kept under vacuum until the resin is<br />
completely gelled. The part may then be cured at room temperature or in an oven. Bag leaks<br />
and bridging are common problems in VARTM. Bag leaks take place at the sealant-bag-film<br />
interface or as a result of film failure due to improper handling. Bridging is the failure of the<br />
sealing bag to conform to the shape of the mold. This leads to the part failing to receive<br />
uniform pressure during the cure cycle.<br />
Matrix viscosity and process time are the two main differences between RTM and<br />
VARTM. For RTM, the resin must travel through the "X" and "Y" directions while for<br />
VARTM it travels on top and only needs to impregnate the "T" or "Z" direction. This requires
Major Trends in Polymeric <strong>Composite</strong>s Technology 123<br />
shorter time and provides the advantage of the lower temperature and faster cure time<br />
combined with reduced thermal stress.<br />
Resin film infusion, RFI, is a composite manufacturing process which has advanced from<br />
earlier work on vacuum impregnation and RTM. In this process, a semi-cured resin film is<br />
liquefied and absorbed throughout the fiber. The mold filling is further assisted by vacuum to<br />
reduce the air voids remaining in the fabricated part. The resin and the fiber are generally<br />
placed together into the mold but are not initially combined. In some applications the fiber<br />
and the resin are placed in the mold in separate steps and are combined by applying pressure.<br />
Computer simulations are commonly used to determine processing details such as resin<br />
viscosity, preform permeability, resin/preform interactions, and the time to completely cure<br />
the composite part. The major difference between RFI and RTM is that former uses a hot<br />
melt resin film while the later utilizes a liquid resin. RFI does not require low minimum<br />
viscosity as in the RTM process.<br />
Orthophthalic, isophthalic polyesters and vinyl esters are primarily used in RTM<br />
processes. A variety of polymers are being developed specifically for RTM application, e.g.<br />
low-shrink and low-profile polyesters for improved surface appearance. New resins including<br />
epoxies, acrylic/polyester hybrids, urethanes, bismaleimides (BMI), and phenolic resins are<br />
also produced which require changes in the equipment and conditioning the resin prior to<br />
injection. These systems offer a whole new range of cost and performance options to the<br />
RTM process.<br />
Reinforcements used in RTM are normally glass fibers, continuous fiber mats and<br />
chopped strand preforms. Special mats that contain thermoplastic binders are heated and thermoformed<br />
into perfect preforms. Both woven and non-woven glass fibers and biaxial and<br />
triaxial mats have been produced for the RTM applications. Other high performance<br />
reinforcements such as carbon fiber and aramid can be incorporated in RTM laminates either<br />
alone or as part of a hybrid system.<br />
A typical RTM mold features oil connections, injection runner system, and self<br />
clamping/load devices. RTM surfaces offer high quality through a combination of<br />
appropriate resin reinforcements, molds and process conditions. A combination of mechanical<br />
clamping arrangements with special presses is necessary to secure the mold halves. This is<br />
required for applying pressure uniformly to the mold when pressurized resin is employed.<br />
RTM processing requires accurate and reliable injection of liquid resin. Resins must balance<br />
low viscosity at processing temperatures and long pot life without sacrificing the mechanical<br />
properties to alter the flow characteristics. The resin is injected until the mold is completely<br />
filled. Motionless or non-mechanical mixers are normally utilized to blend resin and catalyst.<br />
After curing for a required time the composite part is moved from the mold. A mixer flush<br />
system is also incorporated when required to purge non-disposable mixers.<br />
RTM is employed to reduce fill times and to fabricate large-scale composite structures<br />
with substantial laminate thicknesses. It fills the gap between hand lay-up and compression<br />
molding of sheet or bulk moldings in matched metal molds. In comparison with lay-up and<br />
spray-up processes RTM provides two finished surfaces on parts which can be similar or<br />
dissimilar, highly reproducible thickness, low monomer loss and higher output since it is less<br />
labor and material intensive. Compared with matched-metal-die compression molding, RTM<br />
enables the use of parts such as ribs and inserts, decreases lead times for molds, and has<br />
lower-cost molds and molding equipment.
124<br />
W.H. Zhong, R.G. Maguire, S.S. Sangari et al.<br />
RTM provides a lot of processing flexibility. Shrink control systems can be employed to<br />
produce improved surfaces. The fibers and resins are utilized at their lowest cost plus resin<br />
content can be controlled to a significant degree and reinforcements can be easily<br />
incorporated. Cores and other insets can also be positioned prior to resin injection to yield<br />
complex parts in one fabrication step. RTM can also be used to prototype parts for market<br />
evaluation since the initial investment costs such as tooling and operating expenses are low.<br />
In this case, RTM's short lead time and lower cost tooling is a real advantage. It also offers<br />
production of near-net-shape parts which in turn leads to low material wastage. Closed RTM<br />
molds release fewer volatile materials than open molds. Added benefits of this closed-mold<br />
process are greatly reduced volatile organic compound (VOC) emissions.<br />
VARTM can be used at room temperature with no heated mold and compared to RTM<br />
more consistent and uniform coverage can be obtained. VARTM provides significant savings<br />
in the tooling cost as it requires only a one-piece mold. The use of the vacuum bag eliminates<br />
the need for making a precise matched metal mold as in the conventional RTM process and<br />
thus reduces the cost and design difficulties associated with large metal tools. VARTM also<br />
beats the 5-10% accuracy of hand lay-up.<br />
For a given resin, mold filling using RFI is relatively faster than most other liquidcomposite-molding<br />
processes since the RFI products are usually thin and the infusion<br />
distance is short. In addition, RFI can provide parts with high mechanical properties due to<br />
solid state of initial polymer material and elevated cure temperature.<br />
Along with the advantages, RTM also has inevitable disadvantages including inability to<br />
manufacture very large parts and permeability issues which lead to increased processing time.<br />
It is an intermediate volume production process and its tooling and equipment costs are higher<br />
than the hand lay-up and spray-up. Tooling design and fabrication to handle injection<br />
pressures, clamping and sealing are more complex and manufacture of complex parts require<br />
trial and error experimentations combined with flow simulation modeling.<br />
VARTM also is a relatively complex process to perform well. The flexible nature of the<br />
vacuum bag brings about the difficulty of controlling the final thickness of the preform, and<br />
hence the fiber volume fraction of the composite. Due to the complex nature of VARTM, the<br />
trial and error experimentation is not only inefficient but also expensive for the process design<br />
and optimization.<br />
RTM is being used for a number of products including aircraft components, recreational<br />
vehicle, truck and sports car bodies, automobile panels, medical equipment, dish antenna,<br />
storage tanks, electrical covers, windmill blades, plumbing parts, transportation seating,<br />
chemical pumps, marine components such as hatches and small boats, bicycle frames and<br />
doors.<br />
Resin Infusion processes enable us to manufacture a wide range of complex as well as simple<br />
fiber reinforced composite products. Improvements are being made on resins and tooling<br />
developments which further expand market opportunities. The combination of pre-positioning<br />
a variety of reinforcements and incorporating secondary reinforcements and other design<br />
details with good surface control on both sides makes RTM a primary process candidate for<br />
structural applications in the aerospace industry as well as other markets. Due to its<br />
versatility, growing use in industry, and being environmentally friendly, experts believe that<br />
the future of this useful composite process should continue to grow. In many applications its<br />
replacement of hand layup/ prepreg material forms/autoclave processing production has
Major Trends in Polymeric <strong>Composite</strong>s Technology 125<br />
resulted in significant cost and rate benefits, further enhanced with the possibility of ganginfusion<br />
production scenarios and automated preform production and handling.<br />
Advanced Thermoplastic <strong>Composite</strong> Processing: Thermoplastics are having a resurgence of<br />
interest, largely due to a growing list of successful new processing methods. Reinforced<br />
Thermoplastic Laminates (RTL) is an economical means of producing a solid laminate if<br />
there is no constraint against constant thickness. The pre-consolidated laminate is heated<br />
above the melt temperature, usually by infrared lamps, and then automatically transferred into<br />
a pair of cool tools that rapidly close to form and cool the part. This achieved very rapid<br />
cycle time. Another process that has been growing in applications is the automated<br />
thermoplastic pultrusion process that can produce high volumes of long straight parts of<br />
various shapes [23]. These and other thermoforming processes are bringing new life to the<br />
applications of thermoplastics in significant structural applications that promise to realize the<br />
attractions of short cycle times and the possibility of recycling.<br />
5. Other Trends<br />
Smart <strong>Composite</strong>s: Smart materials have the ability to perform both sensing and actuation<br />
functions. The use of imbedded sensors such as piezoelectric, shape-memory alloys,<br />
magneto-strictive, or fiber optics with Bragg gratings (FOBG) to sense and mitigate the<br />
threats to the health of a structure, i.e. Structural Health Monitoring (SHM), holds great<br />
promise for the future of composite primary structure through the elimination of designed-in<br />
excess material for undetected damage events; being aware of damage when it happens and<br />
where it happens can eliminate much design conservatism. Other possibilities are the<br />
incorporation of self-healing or restorative abilities, active control of key functions such as<br />
vibration, etc. Smart composites face the challenges of effective dispersion and interfacial<br />
adhesion of the “smart” constituents. Smart composite materials can be obtained by mixing<br />
the polymer matrix with smart material used for health monitoring, active control and selfrestoration<br />
of structural and functional materials. Recent advances in optical glass fibers have<br />
produced a form which has the approximate same diameter as a carbon fiber so can be<br />
incorporated into a tape or fabric reinforcement without disruption of the load carrying<br />
capability.<br />
Bio-based composites: Increasing interest is developing in bio-based composite constituents.<br />
With shortages developing for the traditional petroleum based products, there are activities in<br />
US, China, Singapore and elsewhere to develop carbon fibers from renewable agricultural<br />
sources such as corn, soy, rice, wheat and other biomaterials that do not deplete the petroleum<br />
reserves. To date the efforts are still in their early stages of success, the quality is not that of<br />
the PAN or pitch based fibers but the costs are very attractive and the growing interest in<br />
greener processing will add impetus to these activities, particularly for those applications<br />
where lower performance is not critical and which are suffering from the current carbon fiber<br />
shortage. University of Delaware Affordable <strong>Composite</strong>s from Renewable Resources<br />
(ACRES) program is one source of development in this area, having been awarded a USDA<br />
National <strong>Research</strong> Initiative to investigate the possibility of making circuit boards from soy<br />
resins and chicken feather based carbon fibers rather than the conventional epoxy, PAN-based
126<br />
W.H. Zhong, R.G. Maguire, S.S. Sangari et al.<br />
composites. Michigan State University has a center for biocomposites with many projects<br />
underway on bio composites, green nanocomposites, biodegradeable thermoplastic polymers<br />
and soy based bioplastics among others [24].<br />
Short fiber composites: Originally short fibers in composites were very short, basically just<br />
additives with aspect ratios only slightly greater than one. Subsequently, the long<br />
discontinuous fiber (LDF) composites appeared with fibers either chopped, stretch broken, or<br />
otherwise made discontinuous. These LDFs could approach the performance of continuous<br />
fiber composites with fiber lengths of 2” to 4” and some degree of alignment. For applications<br />
with complex geometry, these materials can offer relief from the limitations of continuous<br />
fiber prepregs difficulties in conforming to bends and reentrant shapes. In many cases these<br />
materials and processes can replace traditional metals in parts with complex shapes. In<br />
aerospace applications the advantages offered in replacing small metal parts and in some<br />
cases conventional composites materials and processes are significant and growing [25].<br />
Chopped fibers can also be hybridized with continuous fibers to create an engineered form.<br />
Phoenix <strong>Composite</strong>s selectively uses continuous fiber uni-directional and woven<br />
reinforcement locally introduced into a parent structure consisting of chopped random fiber<br />
reinforcement for a form that is comparable to continuous fiber composites in strength and<br />
stiffness but still retaining the geometric flexibility of a chopped fiber process. Chopped fiber<br />
composites when combined with thermoplastics can be a very cost-effective process.<br />
University of Alabama-Birmingham has produced effective bus seats using Long Fiber<br />
Reinforced Thermoplastics (LFT: PP + glass fiber) in a compression molding process.<br />
Conclusions (Summary)<br />
Polymeric composites technology has been the vehicle of change in key industrial sectors for<br />
the past 30 years, growing from fiberglass reinforcement to more sophisticated polymeric<br />
fibers and the current champion, the carbon fiber/multi-phase matrix polymer composite<br />
materials. As applications have grown both in breadth and scale, new needs and visions have<br />
created strong and focused trends, in both materials and processing sciences and technologies,<br />
and emerging at an increasing rate. Market-driven pull and science-based push mechanisms<br />
have brought us to a richer landscape of increased dimensions and applications unimagined a<br />
few decades earlier. The composites community, unlike other industrial technolgies, is not<br />
complacent. Although autoclaves and prepreg have served us well, we want to get rid of<br />
them and process in a more optimized way with resin infusion and leaner manufacturing.<br />
While the current materials have enabled entire airplanes to be made of composite materials,<br />
we want those materials to now serve multiple functions, behave with intelligence, be<br />
greener, and are exploring the huge benefits of very small matters in the world of<br />
nanotechnology. Conventional materials such as thermoplastics take on a new life as vastly<br />
more efficient and focused processing methods are developed, and combining conventional<br />
materials in unconventional and novel ways is opening new possibilities. The compositeer<br />
has much to feel satisfied about but there is much to challenge them in the future.
Acknowledgements<br />
Major Trends in Polymeric <strong>Composite</strong>s Technology 127<br />
Dr. Zhong acknowledges the support from NASA through grant NNM04AA62G and from<br />
NSF through NIRT Grant 0506531.<br />
References<br />
[1] Delmonte, J.; Technology of Carbon and Graphitic <strong>Composite</strong>s; Van Nostrand<br />
Reinholdt Co., 1980.<br />
[2] Subramanian, R. V.; Jukubowski J. J. Polym. Eng. Sci. 1978, 18, 590-600.<br />
[3] Broutman, L. J.; Agarwal, B. D. Polym. Eng. Sci. 1974, 14, 581-588.<br />
[4] William Jr, J. H.; Kousiounelos, P. N. Fibre. Sci. Tech. 1978, 11, 83-88.<br />
[5] Peiffer, D. G. J. Appl. Polym. Sci. 1979, 24, 1451-1455.<br />
[6] Arridge, R. G. C. Polym. Eng. Sci, 1975, 15, 757-760.<br />
[7] Chen, J. S.; Lau, S. P.; Sun, Z.; Tay, B.K.; Yu, G. Q.; Zhu, F. Y.; Zhu, D. Z.; Xu, H. J.<br />
Surf. Coat. Tech. 2001, 138, 33-38.<br />
[8] Kostov, K. G.; Ueda, M.; Tan, I. H.; Leite, N. F.; Beleto, A. F.; Gomes, G. F. Surf. Coat.<br />
Tech. 2004, 186, 287-290.<br />
[9] Ujvari, T.; Toth, A.; Bertoti, I.; Nagy, P. M.; Juhasz, A. Solid State Ionics, 2001, 141-<br />
142, 225-229.<br />
[10] Torrisi, L.; Gammino, S; Mezzasalma, A. M.; Visco, A. M.; Badziak, J.; Parys, P.;<br />
Wolowski, J.; Woryna, E.; Krasa, J.; Laska, L.; Pfeifer, M.; Rohlena, K.; Boody; F. P.<br />
Appl. Surf. Sci. 2004, 227, 164-174.<br />
[11] Cohen, Y.; Rein, D. M.; Vaykhansky, L. E.; Porter, R. S. <strong>Composite</strong>s Part A.1999, 30,<br />
19-25.<br />
[12] Netravali, A. N. Fiber/resin interface modifiction techniques: A case study of ultra-high<br />
molecular weight polyethylene fibers, 50 th Intl. SAMPE, Long Beach, CA, 2005.<br />
[13] Nguyen, H. X.; Riahi, G.; Wood, G.; Poursartip, A. in 33 rd Intl. SAMPE Symp.,<br />
Anaheim, CA, 1988.<br />
[14] Hicks, R. F.; Babayan, S. E.; Penelon, J.; Truong, Q.; Cheng, S. F.; Le, V. V.;<br />
Ghilarducci, J.; Hsieh, A.; Deitzel, J. M.; Gillespie, J. W. Atmospheric Plasma<br />
Treatment of Polyetheretherketone <strong>Composite</strong>s for Improved Adhesion, SAMPE Fall<br />
Technical Conference Proceedings: Global Advances in <strong>Materials</strong> and Process<br />
Engineering, Dallas, TX, 2006; lCD-ROM, pp 9.<br />
[15] Neema, S.; Salehi-Khojin, A.; Zhamu, A.; Zhong, W. H.; Jana, S.; Gan,Y. X. J. Colloid<br />
Interf. Sci. 2006, 299, 332-341,<br />
[16] Jana, S.; Zhamu, A.; Zhong, W. H.; Gan,Y. X. J. Adhesion. 2006, 82, 1157-1175.<br />
[17] Salehi-Khojin, A.; Stone, J. J.; Zhong, W. H. J. Compos. Mater. 2007, 41, 1163-1176,.<br />
[18] Zhamu, A.; Wingert, M.; Jana, S.; Zhong, W. H.; Stone, J. J. <strong>Composite</strong>s Part A. 2007,<br />
38, 699-709.<br />
[19] Zhamu, A.; Zhong, W. H.; Stone, J. J. Compos. Sci. Tech. 2006, 66, 2736-2742.<br />
[20] Donnet, J. B.; Bansal, R. C. Carbon Fibers; 2 nd Edition; Marcel Dekker: New York,<br />
NY, 1990.<br />
[21] Figueiredo, J. L. et al. Carbon Fibers, Filaments and <strong>Composite</strong>s; Kluwer Academics<br />
Publishers: Netherlands, 1990.
128<br />
W.H. Zhong, R.G. Maguire, S.S. Sangari et al.<br />
[22] http://www.tc.faa.gov/its/cmd/visitors/data/AAR-430/advanced.pdf<br />
[23] http://www.quickstep.com.au<br />
[24] http://www.sti.nasa.gov/tto/spinoff2001/ip7.html<br />
[25] http://www.acm-fn.de/e_start.htm<br />
[26] http://www.egr.msu.edu/cmsc/biomaterials/star/star<br />
[27] http://www.hexcel.com/Products/Matrix+Products/Other+FRM/HexMC
In: <strong>Composite</strong> <strong>Materials</strong> <strong>Research</strong> <strong>Progress</strong> ISBN: 1-60021-994-2<br />
Editor: Lucas P. Durand, pp. 129-164 © 2008 Nova Science Publishers, Inc.<br />
Chapter 4<br />
AN EXPERIMENTAL AND ANALYTICAL STUDY<br />
OF UNIDIRECTIONAL CARBON FIBER REINFORCED<br />
EPOXY MODIFIED BY SIC NANOPARTICLE<br />
Yuanxin Zhou a , Hassan Mahfuz b , Vijaya Rangari a<br />
and Shaik Jeelani a<br />
a Center for Advanced <strong>Materials</strong> at Tuskegee University, Tuskegee, AL, 36088<br />
b Department of Ocean Engineering, Florida Atlantic University, Boca Raton, FL 33431<br />
Abstract<br />
In the present investigation, an innovative manufacturing process was developed to<br />
fabricate nanophased carbon prepregs used in the manufacturing of unidirectional composite<br />
laminates. In this technique, prepregs were manufactured using solution impregnation and<br />
filament winding methods and subsequently consolidated into laminates. Spherical silicon<br />
carbide nanoparticles (β-SiC) were first infused in a high temperature epoxy through an<br />
ultrasonic cavitation process. The loading of nanoparticles was 1.5% by weight of the resin.<br />
After infusion, the nano-phased resin was used to impregnate a continuous strand of dry<br />
carbon fiber tows in a filament winding set-up. In the next step, these nanophased prepregs<br />
were wrapped over a cylindrical foam mandrel especially built for this purpose using a<br />
filament winder. Once the desired thickness was achieved, the stacked prepregs were cut<br />
along the length of the cylindrical mandrel, removed from the mandrel, and laid out open to<br />
form a rectangular panel. The panel was then consolidated in a regular compression molding<br />
machine. In parallel, control panels were also fabricated following similar routes without any<br />
nanoparticle infusion. Extensive thermal and mechanical characterizations were performed to<br />
evaluate the performances of the neat and nano-phased systems. Thermo Gravimetric Analysis<br />
(TGA) results indicate that there is an increase in the degradation temperature (about 7 0 C) of<br />
the nano-phased composites. Similar results from Differential Scanning Calorimetry (DSC)<br />
and Dynamic Mechanical Analysis (DMA) tests were obtained. An improvement of about<br />
5 0 C in glass transition temperature (T g) of nano-phased systems were also seen. Mechanical<br />
tests on the laminates indicated improvement in flexural strength and stiffness by about 32%<br />
and 20% respectively whereas in tensile properties there was a nominal improvement between<br />
7-10%. Finally, micro numerical constitutive model and damage constitutive equations were<br />
derived and an analytical approach combining the modified shear-lag model and Monte Carlo
130<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
simulation technique to simulate the tensile failure process of unidirectional layered<br />
composites were also established to describe stress-strain relationships.<br />
Introduction<br />
Carbon fiber reinforced polymer matrix composites due to their high specific strength and<br />
specific stiffness have become attractive structural materials not only in the weight sensitive<br />
aerospace industry, but also in marine, armor, automobile, railways, civil engineering<br />
structures, sport goods, etc. Generally, the in-plane tensile properties of the fiber/polymer<br />
composite are defined by the fiber properties, while the compression properties and properties<br />
along the thickness dimension are defined by the characteristics of the matrix resin.<br />
Epoxy resin is the most commonly used polymer matrix for advanced composite<br />
materials. Over the years, many attempts have been made to modify the properties of epoxy<br />
by the addition of either rubber particles [1-2] or fillers [3-4] so that the matrix-dominated<br />
composite properties are improved. The addition of rubber particles improves the fracture<br />
toughness of epoxy, but decreases its modulus and strength. The addition of fillers, on the<br />
other hand, improves the modulus and strength of epoxy, but decreases its fracture toughness.<br />
Usually, the typical filler content needed for significant enhancement of these properties can<br />
be as high as 10-20% by volume. At such high particle volume fractions, the processing of<br />
the material often becomes difficult, and since the inorganic filler has a higher density than<br />
the resin, the density of the filled resin is also increased. Nanoparticle filled resins are<br />
attracting considerable attention since they can produce property enhancement that are<br />
sometimes even higher than the conventional filled polymers at volume fractions in the range<br />
of 1 to 5%. It has been established that adding small amounts of nano-particles (
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 131<br />
Monte Carlo simulation on the tensile failure process of unidirectional carbon prepreg<br />
laminates.<br />
<strong>Materials</strong><br />
T700SC 12000-50C carbon fiber used in this research is manufactured by Toray Carbon<br />
Fibers America, Inc., USA. It is the highest strength, standard modulus fiber in the form of<br />
continuous filament tows with outstanding processing characteristics for filament winding,<br />
weaving and prepregging produced using the PAN (polyacrylonitrile) process.<br />
The epoxy system used was CR46T. It’s a high temperature cure prepreg resin system.<br />
To avoid degradation of its properties, the resin is kept under sub-zero temperatures in a<br />
sealed atmosphere. This resin system is well suited for prepreg applications, and its properties<br />
are shown in Table 1.<br />
Table 1. Resin properties<br />
Density 0.0457 lb/in 3<br />
Gell (min) @ 350 0 F 6 – 10<br />
GIC<br />
0.733<br />
lb/in<br />
± 0.3<br />
2<br />
Tg DRY 2hr @ 375 0 F 437 0 F<br />
Tg WET 2hr @ 375 0 F 295 0 F<br />
Tensile strength @ RT 10 ksi<br />
Tensile modulus @ RT 643 ksi<br />
Poisson ratio 0.36<br />
Elongation 1.7 %<br />
The nanosized fillers for this present investigation were chosen as nano-sized silicon<br />
carbide particles. These are highly complex material existing primarily in amorphous or<br />
crystalline states. The amorphous SiC is mainly used in coating industries. In functional and<br />
structural applications, crystalline SiC are extensively used due to their excellent thermomechanical<br />
properties such as high hardness and stiffness, good corrosion and oxidation<br />
resistance, high thermal conductivity and high chemical and thermal stability. [10-12]. Such<br />
SiC is available in two different phases, namely alpha (α) and beta (β) phases. The formation<br />
of these two structures depends on the molecular organization of the basic structural unit, a 2layer<br />
planner unit of Si and C in tetrahedral coordination. β−SiC is formed when the planes of<br />
Si and C are rearranged in a cubic symmetry with a lattice constant a = 0.4358 nm. On the<br />
other hand, heating of β−SiC to high temperature causes the transformation of the cubic<br />
symmetry to a mixture of hexagonal (6H) and rhombohedral (15R) polytypes known as α-<br />
SiC. The corresponding lattice constant parameters of α-SiCs are: a = 0.3082 nm and c =
132<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
1.5117 nm [13-14]. The α-SiC is chemically unstable and as a result their application is very<br />
limited. A schematic showing the crystal structure of α and β−SiC is given in Figure 1.<br />
The nano sized β-SiC particles were obtained from MER Corporation, USA. These<br />
particles are spherical in shape with average diameter of about 30 nm as shown in Figure 2.<br />
The bulk material contains more than 95% of SiC with small traces of Oxygen and Carbon.<br />
(a)<br />
(b)<br />
Figure 1. Crystal structure of (a) α-SiC, (b) β-SiC Particles<br />
[www.a-e/englisch/lexikon/ siliciumcarbid-bild2.htm]
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 133<br />
Manufacturing<br />
Figure 2. TEM micrograph of nano SiC sized particles.<br />
Manufacturing of the Nano-phased Carbon Prepregs<br />
Online solution impregnation and filament winding were used as the method of<br />
manufacturing nano-phased unidirectional carbon prepregs. This method involves four<br />
principle steps: (1) uniform dispersion of nano particles in the resin system; (2) application of<br />
resin reaction mixture onto the reinforcing tows; (3) removal of solvent from the prepregs;<br />
and (4) filament winding.<br />
There are various techniques to disperse nanoparticles in the resin system. Acoustic<br />
cavitation is one of the efficient ways to disperse nano-particles into the virgin materials. In<br />
this case, the ultrasonic power supply (generator) converts 50/60 Hz voltage to a high<br />
frequency electrical energy. This voltage is applied to the piezoelectric crystals within the<br />
converter, where it is changed to small mechnical vibrations. The converter’s longitudinal<br />
vibrations are amplified by the probe (horn) and transmitted to the liquid as ultrasonic waves<br />
consisting of alternate compressions and rarefactions. These pressure fluctuations give rise to<br />
microscopic bubbles (cavities), which expand during the negative pressure excursions, and<br />
implode violently during the positive excursions. Some of these cavities oscillate at a<br />
frequency of the applied field (usually 20 kHz) while the gas content inside these cavities<br />
remains constant. As the bubble collapse, millions of shock waves, eddies and extremes in<br />
pressures and temperatures are generated at the implosion sites. Although this phenomenon<br />
known as cavitation, lasts but a few microseconds, and the amount of energy released by each<br />
individual bubble is minimal, the cumulative amount of energy generated is extremely high.<br />
During the operation, an active cavitation region is created close to the source of the<br />
ultrasound probe and that the ultrasonic processing produces high pitched noise in the form of<br />
harmonics which are above the human audible range, emanating from the container walls and<br />
the fluid surface. The development of cavitation processes in the ultrasonically processed<br />
melt creates favorable conditions for the intensification of various physio-chemical processes.
134<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
Acoustic cavitation accelerates heat and mass transfer processes such as diffusion, wetting,<br />
dissolution, dispersion and emulsification [15-16].<br />
Optimization of the solution prepregging process begins with the appropriate choice of<br />
solvent. A high degree of wetting can only be expected from solvents that possess favorable<br />
thermodynamics regarding wetting of the particular solid material (carbon filaments, in this<br />
case) [17-18]. The process of wetting entails the contact and spreading of the solvent over the<br />
surface of the solid, i.e., liquids that possess a low contact angle for a particular solid show<br />
considerable wetting behavior (as opposed to liquids that display high contact angles). This<br />
solvent should be chosen from a list of candidate solvents capable of dissolving the matrix<br />
polymer. The differences in wetting action, coupled with other relevant parameters such as<br />
boiling point and general practicality of the particular solvent choice usage, will lead to an<br />
appropriate choice of solvent. In particular, solvent characteristics should include a much<br />
lower boiling point than melt flow point of the resin and a lower density then that of the resin<br />
for ease of residual solvent removal [19]. An example of the preceding contact angle analysis<br />
can be found in a study by Patel and Lee [17]. In their study, fiberglass tows were subjected<br />
to contact angle analysis using the Wilhelmy plate method. A series of liquids was used (not<br />
polymer solutions), each having differing values of viscosity and surface tension. The<br />
equilibrium contact angles for all of these liquids were not observed to be a function of<br />
solvent viscosity (viscosity range = 0.33 mPa – 1499.0). Furthermore, the liquid surface<br />
tension was found to be positively correlated with the contact angle, i.e., increases in surface<br />
tension generally yielded larger contact angle measurements. It should be stressed that these<br />
results only indicate trends in contact angles; they may not imply favorable conditions for<br />
capillary flow (in addition to wetting), which is another important consideration in the<br />
prepreg process [15]. Once the appropriate solvent is identified for solution prepregging,<br />
prepregged tapes can be manufactured. The objective in solution prepregging is to prepare a<br />
uniform tape in which every fiber surface is uniformly wetted with the polymeric matrix<br />
material. Another objective in solution prepregging is maximizing the amount of matrix<br />
material pick-up. This is easily quantifiable as the amount of matrix material adhering to the<br />
fiber surface after a single immersion into the resin bath. The nature of the relationship<br />
between fiber dispersion and matrix pick up is expected to be competitive. This can be<br />
inferred from the extremes of the process. In a polymer solution with a concentration<br />
approaching zero, every filament can be expected to be wetted (resulting in a good fiber<br />
dispersion), assuming that the thermodynamics are favorable. But the matrix pick up in this<br />
case is nearly zero since there is no polymer in solution. At the other extreme, the polymer<br />
weight fraction in solution approaches one. In this case, the fiber wetting upon dipping will be<br />
very poor given the extremely high viscosity of the resin (kinetic limitation). But upon<br />
wetting, a large amount of polymer will remain on the fiber surface (high matrix pick up).<br />
Therefore, intuition states that there will exist an intermediate polymer solution concentration<br />
in which a balance is obtained between the fiber dispersion and matrix pick up. The concepts<br />
in the preceding paragraph can be more easily visualized by using a model that approximates<br />
the wetting process of a fiber tow by a polymer solution. By combining the Kelvin equation,<br />
which describes wetting of a solution in micro-capillaries and Darcy’s Law, which describes<br />
flow in porous media, the following equation is obtained:<br />
f<br />
void<br />
{ 2S<br />
( 2 / R)<br />
γ θ}<br />
2<br />
t =<br />
l μV /<br />
cos<br />
b<br />
sizing
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 135<br />
where: tf = tow wetting time<br />
l = tow thickness<br />
µ�= solution viscosity<br />
Vvoid = tow void volume<br />
Sb = tow permeability (perpendicular to fiber direction)<br />
R = fiber-fiber separation<br />
γsizing= solution surface tension<br />
θ= contact angle<br />
A quick survey of the equation reveals the following three trends:<br />
• As the solution viscosity increases, the time of tow wetting increases.<br />
• As the surface tension of the solution increases, the time of tow wetting decreases.<br />
• As the contact angle increases from 0 0 �(complete wetting) to 90 0 �(mostly nonwetting),<br />
the cosine term decreases and thus increases the time of tow wetting.<br />
Prepreg residence time is also known to influence both the fiber dispersion and<br />
efficiency. In a study by Lacroix et al. [20], ultra-high modulus polyethylene fiber bundles<br />
were prepregged with a xylene/ low-density polyethylene solution. For a prepregging time<br />
range of 8 min. – 19.5 hours, it was noted that increasing prepreg time increased the layer<br />
thickness of deposited polymer around the fiber surfaces. Similar results were obtained in a<br />
study by Moon et al. [19] in which solvent prepregged fiber bundles were prepared from glass<br />
fibers and a high-density polyethylene/ toluene solution.<br />
After the fiber tapes are prepregged with the nano-phased resin, the solvent has to be<br />
driven off. In this case, since the tapes are not to be wound around a storage spool following<br />
prepregging, solvent elimination should be complete. This represents a crucial step in the<br />
overall composite manufacturing process, as residual solvent can result in voids during the<br />
melt consolidation process. How the solvent interaction with the fiber/matrix/nanoparticle<br />
interface is an important consideration, given the influence of the quality of the interface in<br />
determining the final mechanical properties of the composite. The presence of solvent is<br />
generally known to reduce the quality of the matrix/fiber interface. The reasons for this<br />
phenomenon are unclear, but can be explained by the following hypothesis [21]:<br />
• Solvent extraction can cause separation of the fiber/matrix interface<br />
• Solvent concentration at the interface will interfere with fiber/matrix contact; and<br />
• Phase separation of low molecular weight species at the interface may form a weak<br />
interface between the fiber and matrix.<br />
Solvent removal, in part, is regarded to proceed by solvent concentration at the interface,<br />
followed by solvent traversing the fiber surface and escaping from the ends of the composite.<br />
Obviously this will result in poor interfacial quality if this is to occur during melt<br />
consolidation or autoclave processing, as the case maybe. A study conducted by Wu et al.<br />
[22] illustrates how residual solvent negatively affects composite mechanical property<br />
quality. Solution prepregged carbon fiber reinforced polyethersulphone composites were<br />
prepared and compared with strictly hotmelt processed composites of the same nominal fiber
136<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
content. The transverse flexural strength of the solution prepregged material was only half<br />
that of the melt-processed material. Upon analysis of the solution prepregged material using<br />
differential scanning calorimetry (DSC), it was found that residual solvent remained in the<br />
sample, despite hotmelt consolidation of the prepreg. Residual solvent can most likely be<br />
attributable to difficulty in solvent diffusion during the consolidation process. The reasons for<br />
poor interfacial quality are thought to be attributable in the reasons outlined in the preceding<br />
paragraph.<br />
Commercially available high temperature prepreg resin CR46T was first dissolved in<br />
acetone (Dimethyl ketone, class 1B, Fisher Scientific Co. LLC, USA), at a ratio of 65:35 by<br />
mechanical stirring at 1500 RPM for about 4 hours as shown in Figure 3. Spherical shaped<br />
silicon carbide nanoparticles were carefully measured to have a 1.5% loading by weight of<br />
the resin and mechanically mixed with the liquid resin. The mixture was then irradiated with<br />
high intensity Sonic Vibra Cell ultrasonic liquid processor (Ti-horn, 20 kHz, 100W/cm 2 ) at<br />
50% amplitude for 30 minutes. This ensured uniform mixing of nanoparticles over the entire<br />
volume of the resin. To avoid temperature rise during sonication, cooling was employed by<br />
submerging the mixing beaker in a water bath maintained at 50 0 F as shown in Figure 4. The<br />
nano-phased resin reaction mixture was then transferred into a heating bath maintained at a<br />
constant temperature of 80 0 F throughout the fabrication as shown in Figure 5. A continuous<br />
strand of carbon fiber from a spool attached in the spindle bracket assembly was allowed to<br />
pass through the resin bath at a rate of about 1 meter per minute. In this case, the resin<br />
reaction mixture individually wet each filament within the fiber tow. Once the fiber was<br />
coated with nano-phased resin the excess solvent was removed from the prepreg by passing<br />
the wet strand through a high temperature heater maintained at 160 0 F. The nano-phased<br />
prepreg tow was then routed and fed through a fiber delivery system and was precisely hoopwound<br />
on a rotating foam mandrel on the filament winding machine. Figure 6 represents the<br />
schematic of solution impregnation and filament winding setup. During the fiber placement,<br />
the winding angle was kept at 89.875 0 to avoid excessive gaps or overlaps between adjacent<br />
courses.<br />
Figure 3. CR46T resin mixed with acetone.
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 137<br />
Figure 4. Ultrasonic cavitation.<br />
Figure 5. Nano-phased resin.<br />
Figure 6. Schematic of solution impregnation and filament winding.
138<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
Manufacturing of Nano-phased Carbon Prepreg Unidirectional Laminates<br />
The process of unidirectional laminate fabrication started with online prepregging. By this it<br />
means that when eight layers of the manufactured prepregs were hoop wound on the rotating<br />
foam mandrel during prepreg manufacturing, the remaining tow was cut and the mandrel was<br />
removed from the filament winding machine. During prepreg stacking, care was taken to<br />
place tows without allowing the previous layer to dry. The prepregs on the cylinder was then<br />
longitudinally cut open into a rectangular sheet as shown in Figure 7. These rectangular<br />
sheets were arranged in a compression molding setup by putting symmetric layers of plastic<br />
film, bleeder cloth and teflon on the top and bottom. The whole setup was then placed in<br />
Tetrahedron MTP press compression molder as shown in Figure 8. Mold temperature was<br />
ramped to get 350 0 F while the mold pressure was kept as 40 Psi and consolidated for about 4<br />
hours to obtain a 2mm thick SiC-carbon-epoxy nanophased unidirectional laminate (as shown<br />
in Figure 9). A typical consolidation cycle is shown in Figure 10.<br />
Figure 7. Schematic of unidirectional laminate preparation.<br />
Figure 8. Compression molder.
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 139<br />
Figure 9. Nano-phased unidirectional laminate.<br />
Figure 10. Consolidation cycle for laminates.<br />
Experimental Results and Discussion<br />
Differential Scanning Calorimetry<br />
The Differential Scanning calorimetric (DSC) studies have been carried out to understand the<br />
effect of nanoparticles on glass transition temperature of cured carbon-epoxy prepregs based<br />
on consolidated samples. Figure 11 represents the DSC curve of as-fabricated neat system.<br />
The curve exhibits an endothermic baseline shift at about 220 0 C and highly exothermic peak<br />
at about 390 0 C. The baseline shift at 220 0 C is assigned to the glass transition temperature<br />
and exothermic peak at 390 0 C is assigned to the decomposition temperature of the epoxy<br />
resin. These results match well with the supplier materials data sheet. Figure 12 represents the<br />
DSC curve of as-fabricated 1.5 wt.% SiC nano-phased system. The curve showed only one<br />
exothermic peak at about 397 0 C which is assigned to the decomposition temperature of the<br />
cured epoxy resin. The baseline shift corresponding to glass transition temperature was<br />
almost disappeared. This clearly indicated that the epoxy was highly cross-linked due to<br />
catalytic effect caused by SiC nanoparticles. The shift was not observed, we believe, because<br />
of the equipment sensitivity to the higher cross-linked polymers. To further validate the
140<br />
Heat Flow (W/g)<br />
Heat Flow (W/g)<br />
0.1000<br />
0.0812<br />
0.0625<br />
0.0437<br />
0.0250<br />
0.0062<br />
-0.0125<br />
-0.0313<br />
-0.0500<br />
-0.0688<br />
-0.0815<br />
-0.1063<br />
-0.1250<br />
-0.1438<br />
-0.1625<br />
-0.1812<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
-0.2000<br />
50 100<br />
150<br />
200<br />
250<br />
300<br />
350<br />
400<br />
0.0500<br />
0.0188<br />
-0.0816<br />
-0.1563<br />
-0.2250<br />
-0.2938<br />
-0.3625<br />
-0.4312<br />
Temperature (°C)<br />
Figure 11. DSC graph of neat prepreg system.<br />
-0.5000<br />
50 100<br />
150<br />
200<br />
250<br />
300<br />
350<br />
400<br />
Temperature (°C)<br />
Figure 12. DSC graph of 1.5 wt.% SiC nano-phased prepreg system.
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 141<br />
Figure 13. DMA graphs of neat and 1.5 wt.% SiC nano-phased prepreg systems.<br />
results, dynamic mechanical analysis (DMA) tests were also carried out under single<br />
cantilever test environment. The results indicated the existence of glass transition (Tg) for the<br />
nano-phased system which is about 5 0 C higher than the neat counterpart as shown in Figure<br />
13. The increase in Tg may be attributed to a loss in the mobility of chain segments of epoxy<br />
resin resulting from the high nanoparticle/matrix interaction. Impeded chain mobility is<br />
possible if the nanoparticles are well dispersed in the matrix. The particle surface-to-surface<br />
distances (‘matrix bridges’) should then be relatively small and chain segment movement may<br />
be restricted. Good adhesion of nanoparticles with the surrounding polymer matrix<br />
additionally may have benefited the dynamic modulus by hindering molecular motion to<br />
some extend. The hard particles incorporated into the polymer may also have acted as<br />
additional virtual ‘‘network nodes’’. In either situation it can be deduced that Tg increased as<br />
a result of more number of cross-linked polymer chains and restricted mobility of the chain<br />
segments in the presence of SiC nanoparticles.<br />
Thermo Gravimetric Analysis<br />
Thermo gravimetric analysis (TGA) has been carried out to find the degradation temperature<br />
or to estimate the thermal stability of neat and 1.5 wt.% nano-phased prepreg systems. Figure<br />
14 shows that in as-fabricated neat system, the resin decomposed at about 390 0 C which is<br />
represented by the peak of the derivative curve. The TGA curve shown in Figure 15 indicated<br />
that the decomposition temperature of the nano-phased system was about 397 0 C, which is<br />
almost 7 0 C higher than the neat counterpart. From the results it is clear that in nano-phased<br />
systems, epoxy was amply cross-linked and had minimum particle-to-particle interaction,
142<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
which resulted in increase in thermal stability of the system. These results are consistent with<br />
the DSC results as well.<br />
o<br />
Deriv. of Weight (%/ C)<br />
o<br />
Deriv. of Weight (%/ C)<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0.0<br />
390 C<br />
o<br />
200 300 400 500 600<br />
Temperature( oC)<br />
Figure 14. TGA graph of neat prepreg system.<br />
397 C<br />
o<br />
200 300 400 500 600<br />
Temperature( oC)<br />
Figure 15. TGA graph of 1.5 wt.% SiC nano-phased prepreg system.<br />
110<br />
100<br />
90<br />
80<br />
70<br />
110<br />
100<br />
90<br />
80<br />
70<br />
Weight (%)<br />
Weight (%)
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 143<br />
Flexure Response of Layered <strong>Composite</strong>s<br />
A typical flexure stress-strain plot of neat and nano-phased laminates is shown in Figure 16.<br />
The curves show considerable non-linear deformation and the irregularities in the curves were<br />
attributed to random fiber breakage with pinging noise during the test. The specimens failed<br />
rapidly after reaching the point of maximum stress. In general, the composites exhibited<br />
brittle-type failure. The curves also revealed that by infusing 1.5 wt.% SiC nanoparticles,<br />
strength and modulus significantly improved. Nano-phased system showed approximately<br />
32% increase in flexural strength and 20% in modulus when compared to the neat ones as<br />
shown in Table 2. This result was as expected because of the strong bonding between filler<br />
particles and matrix in nano-phased specimen which resulted in the static adhesion strength as<br />
well as the interfacial stiffness to transfer stresses and elastic deformation efficiently from the<br />
matrix to the fillers via the interface. In other words, large contact areas which translated into<br />
high interfacial stiffness and homogeneous dispersion of nanoparticles assisted in an efficient<br />
stress transfer between polymer and nanoparticles which lead the particles to carry a part of<br />
the external load and resulted in improved flexural strength and stiffness. In addition, the<br />
nanoparticles may have acted as stoppers to crack growth by pinning the cracks. It is also<br />
observed for the nanocomposites in the present study that the strain-to-break tends rather to<br />
slightly higher values in comparison with the neat systems. This increase suggests that the<br />
nanoparticles are able to introduce additional mechanisms of failure and energy consumption<br />
without blocking matrix deformation. Standard deviation for the set of neat and nano-phased<br />
system experimental data were shown to have lower values (Table 3). Lower standard<br />
deviation indicated the stability and consistency in the results as well.<br />
Figure 16. Engineering Stress-Strain curves of flexure test.
144<br />
Material<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
Table 2. Flexure test data for carbon prepreg laminates<br />
Flexural<br />
Strength<br />
(MPa)<br />
Average<br />
Strength<br />
(MPa)<br />
Gain/Loss<br />
Strength<br />
(%)<br />
Flexural<br />
Modulus<br />
(GPa)<br />
Neat Sample 1 765.32 82.14<br />
Neat Sample 2 806.46 80.43<br />
Neat Sample 3 789.38 787.45 -- 82.05<br />
Neat Sample 4 782.62 80.29<br />
Neat Sample 5 793.45<br />
76.91<br />
1.5wt% Sample1 1043.35 93.93<br />
1.5wt% Sample2 995.28 97.02<br />
1.5wt% Sample3 1012.59 1042.34 +32.37 98.25<br />
1.5wt% Sample4 1088.92 90.65<br />
1.5wt% Sample5 1071.55<br />
99.36<br />
Average<br />
Modulus<br />
(GPa)<br />
80.36 --<br />
Gain/Loss<br />
Modulus<br />
(%)<br />
95.84 +19.26<br />
Table 3. Standard deviation and Coefficient of variation of flexure test data<br />
Standard Deviation (±) Co-eff. of Variation (%)<br />
Neat system Strength 13.52 1.72<br />
Neat system Modulus 1.89 2.36<br />
+1.5wt% Nano-phased system Strength 34.99 3.36<br />
+1.5wt% Nano-phased system Modulus 3.17 3.31<br />
Tensile Response of Layered <strong>Composite</strong>s<br />
Typical curves for the tensile behavior of both neat and 1.5 wt.% nano-phased specimen are<br />
shown in Figure 17. The in-plane tensile behavior of both the composites shows linear<br />
behavior up to approximately 1.2% strain where initial fiber failure occurred. The behavior<br />
continued to be linear again till the final specimen failure. Both the elastic modulus and the<br />
strength of nano-phased composites were between 7-10% higher than their neat counterparts.<br />
The reason for such small improvement could be visualized in the sense that, in tension the<br />
fiber took maximum load and the nanoparticle infusion in the matrix did not contribute much<br />
in improving the tensile properties. The improvement of modulus in this study was mainly<br />
because of the improvement of the matrix modulus by filler dispersion. Therefore it can be<br />
deduced that higher tensile properties in the nanocomposite is due to higher nano-phased<br />
matrix properties. Average mechanical properties and their deviation are shown in Table 4<br />
and Table 5, respectively.
Material<br />
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 145<br />
Figure 17. Engineering Stress-Strain curves of tensile test.<br />
Table 4. Tensile test data for carbon prepreg laminates<br />
Tensile<br />
Strength<br />
(GPa)<br />
Average<br />
Strength<br />
(GPa)<br />
Gain/Loss<br />
Strength<br />
(%)<br />
Tensile<br />
Modulus<br />
(GPa)<br />
Neat Sample 1 1.32 86.44<br />
Neat Sample 2 1.41 84.35<br />
Neat Sample 3 1.39 1.38 -- 85.03<br />
Neat Sample 4 1.42 88.57<br />
Neat Sample 5 1.34<br />
86.61<br />
1.5wt% Sample1 1.48 96.39<br />
1.5wt% Sample2 1.39 90.68<br />
1.5wt% Sample3 1.51 1.48 +7.25 93.13<br />
1.5wt% Sample4 1.49 97.90<br />
1.5wt% Sample5 1.55<br />
94.37<br />
Average<br />
Modulus<br />
(GPa)<br />
86.20 --<br />
Gain/Loss<br />
Modulus<br />
(%)<br />
94.49 +9.62<br />
Table 5. Standard deviation and Coefficient of variation of tensile test data<br />
Standard Deviation<br />
(±)<br />
Neat system Strength 0.04 2.86<br />
Neat system Modulus 1.49 1.69<br />
+1.5wt% Nano-phased system Strength 0.05 3.56<br />
+1.5wt% Nano-phased system Modulus 2.51 2.66<br />
Co-eff. of Variation<br />
(%)
146<br />
SEM Analysis<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
SEM analysis was carried out on JSV 5800 JOEL Scanning Electron Microscope. Specimen<br />
from failed samples of flexure tests were selected, prepared and attached to the sample holder<br />
with a silver paint and coated with gold to avoid charge build-up by the electron. Figures 18<br />
and 19 show the SEM micrographs obtained. It is observed from Figures 18a-18b that most of<br />
the damage was located in the loading zone, including large intra and inter-layer delamination<br />
cracks as well as fiber/bundle failures. The damage and the fracture processes were mainly<br />
due to local shear components. They also show interfiber micro-cracks and delamination<br />
cracks. These micrographs also reveal that the carbon fibers were highly oriented with<br />
uniform resin distribution. Figure 19b shows the SEM micrograph in which SiC nanoparticles<br />
(white dots) are distributed uniformly without agglomeration. Also revealing the size of the<br />
filament to be 8-10 microns in diameter and SiC nanoparticle to be of about 30-40 nm range.<br />
(a) (b)<br />
Figure 18. SEM of failed flexure samples in thickness direction.<br />
(a) (b)<br />
Figure 19. SEM of failed flexure samples.
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 147<br />
Numerical Simulation of Tensile Failure Process of Layered<br />
<strong>Composite</strong>s<br />
The tensile failure of fiber reinforced composite material involves a complicated damage<br />
accumulation process resulting from random fiber breakage, stress transfer form broken to<br />
intact fiber, and interface debonding between the fiber and matrix. It is difficult to analyze<br />
such a complicated probabilistic failure phenomenon precisely by means of analytical<br />
methods. The Monte Carlo simulation technique coupled with a stress analysis method is one<br />
of the most effective tools for understanding the tensile failure process [23-27]. In past<br />
Monte Carlo simulations, a micro-composite unit with a coarse mesh and a few fibers of short<br />
length was always used as the numerical model. In practice, structural composites usually<br />
contain large quantities of fibers, when such micro-composite unit is applied to simulate the<br />
failure process of practical composites, it may result in a lack of statistical effects and<br />
magnification of boundary effects, causing errors in calculations of the stress concentration.<br />
Yuan et al. [28] presented a two-dimensional large-fine numerical micro-composite model<br />
with fine mesh, sufficient fibers and adequate length instead of the aforementioned model and<br />
developed a new Monte Carlo simulation method to study the tensile failure process of<br />
unidirectional composites. Based on the new model and method, the average statistical<br />
evolution of the composites deformation and failure, caused by the accumulation of the<br />
random breakages of large quantities of fibers, matrices and interfaces, is successfully<br />
simulated. By taking account of the inertial effect, strain-rate effect of components and the<br />
softening effect caused by the thermo-mechanical coupling in the simulation model, the<br />
tensile stress–strain curves of unidirectional fiber reinforced resin matrix composites CFRP<br />
and GFRP at different high strain-rates were successfully predicted, which agree well with the<br />
experimental results [29-31].<br />
All above Monte Carlo simulations were coupled with the classical shear-lag model. It is<br />
assumed that the fibers bear the whole axial load and the matrix only carries the shear stress.<br />
Ochiai et al. [32, 33] proposed a modified shear-lag model, which takes the axial load born by<br />
the matrix into account, to study the stress concentration in the elastic and elastic-plastic<br />
matrix caused by single fiber breakage. In the present study, Monte Carlo numerical<br />
constitutive model according to Ochiai's modified shear-lag model with fine mesh, sufficient<br />
fibers and adequate length was established to study the failure process of unidirectional<br />
layered composites, to predict the mechanical behavior of these composites with the prepreg<br />
epoxy matrix and to study the relationship between the interface and composite strengths.<br />
Model of <strong>Composite</strong>s<br />
Figure 20 shows the large-fine numerical model of unidirectional composite that consists of n<br />
fibers and n+1 matrices. Each fiber or matrix, at the length of L, is composed of m elements<br />
of length Δx=L/m in the longitudinal direction. The cross-section of the fiber and the matrix<br />
are simplified as rectangle, considering that the simplified fiber has the same sheared area as<br />
that of the actual one, we have
148<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
H = πD<br />
/ 2 df = D / 2<br />
dm =<br />
V<br />
1<br />
( −V<br />
f )<br />
df<br />
where H, df and dm are the thickness of composite, the width of fiber and the width of matrix,<br />
respectively (as shown in Figure 20). D and Vf are the diameter and the volume fraction of<br />
fiber. The displacement component at node (i,j) is expressed as u i,<br />
j .<br />
Figure 20. Model of Unidirectional carbon fiber reinforced matrix resin.<br />
The composite is pulled at one end and fixed at the other end. In figure 20, the left<br />
boundary is fixed, and the right moves with a constant speed V, namely<br />
The initial condition is<br />
⎪⎧<br />
u<br />
⎨<br />
⎪⎩ u<br />
k<br />
i,<br />
0<br />
k<br />
i,<br />
m<br />
= 0<br />
= VkΔt<br />
0<br />
u 0 ( 1 ≤ ≤ 2n<br />
+ 1)<br />
, = i j<br />
Constitutive Assumptions<br />
( 1 ≤ i ≤ 2n<br />
+ 1)<br />
( 1 ≤ i<br />
≤ 2n<br />
+<br />
i and ( ≤ j ≤ m)<br />
f<br />
(1)<br />
(2)<br />
0 (3)<br />
It is assumed that the fibers are homogeneous and linear elastic, the fiber strength is described<br />
statically by single Weibull distribution [34]:
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 149<br />
β<br />
⎡ L ⎛ σ ⎞ ⎤<br />
P ( σ ) = exp⎢−<br />
⎜<br />
⎟ ⎥<br />
(4)<br />
⎢ L0<br />
⎣ ⎝σ<br />
0 ⎠ ⎥⎦<br />
where, P is the survive probability of fiber at stress σ , and the σ 0 and β are Weibull scale<br />
L length and reference length of fiber.<br />
In addition, it is assumed that epoxy matrix is homogeneous and linear elastic.<br />
parameter and Weibull shape parameter, L and 0<br />
⎧σ<br />
m = Emε<br />
⎨<br />
⎩τ<br />
m = Gmγ<br />
where, E m and Gm are tensile modulus and shear modulus.<br />
Shear Stress on the Interface<br />
The shear stress at i−1/i,j interface (shown in Figure 21) τ i−1/i,j can be expressed as a function<br />
of the fiber displacement ui,j and the interface displacement ui−1/i,j:<br />
( u − u ) /( df / 2)<br />
τ =<br />
(6a)<br />
i−1 / i,<br />
j G f i,<br />
j i−1<br />
/ i,<br />
j<br />
where G f is the shear modulus of the fiber. τ i−1/i,j can be also expressed as :<br />
( u − u ) / ( dm / 2)<br />
τ i−1 / i,<br />
j = Gm i−1<br />
/ i,<br />
j i−1,<br />
j<br />
(6b)<br />
If the interface does not break, combine (6A) and (6B) to eliminate ui−1/i,j, then we get<br />
when the interface breaks, we have<br />
2GmG<br />
f<br />
τ i−1<br />
/ i,<br />
j =<br />
( ui,<br />
j − ui−1,<br />
j )<br />
(7)<br />
G df + G dm<br />
m<br />
τ = τ<br />
i−<br />
1 / i,<br />
j<br />
where, τ c is the friction between the fiber and the matrix when the matrix cracks or the<br />
interface debonds.<br />
f<br />
c<br />
(5)
150<br />
Governing Equation<br />
For the fiber element<br />
where,<br />
For the matrix<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
Figure 21. Interface shear stress between fiber element and matrix element.<br />
d u<br />
dfE τ τ<br />
(8a)<br />
f<br />
2<br />
i,<br />
j<br />
+ 2<br />
dx<br />
i / i+<br />
1,<br />
j i−1<br />
/ i,<br />
j<br />
( − ) = 0<br />
d u<br />
dmE τ τ<br />
(8b)<br />
m<br />
2<br />
i,<br />
j<br />
+ 2<br />
dx<br />
i / i+<br />
1,<br />
j i / i−1,<br />
j<br />
( − ) = 0<br />
d u<br />
dmE τ (8c)<br />
m<br />
2<br />
1,<br />
j<br />
+ 2<br />
dx<br />
1/<br />
2,<br />
j<br />
( − 0)<br />
= 0<br />
2<br />
d u2n<br />
+ 1,<br />
j<br />
dmEm + ( 0 −τ<br />
2 / 2 1,<br />
) = 0<br />
2<br />
n n+<br />
j<br />
(8d)<br />
dx<br />
Equation 8A-D can be expressed by the governing equation as follow<br />
A i<br />
A dfE<br />
i<br />
f<br />
d<br />
u<br />
2G<br />
G<br />
2<br />
i,<br />
j<br />
2 +<br />
dfGm<br />
m f<br />
+ dmG f<br />
i+<br />
1,<br />
j i,<br />
j i−1,<br />
j<br />
dx<br />
⎛1<br />
+ μ ⎞ ⎛1<br />
+ μ ⎞<br />
= ⎜ ⎟ + ⎜ ⎟( −1)<br />
⎝ 2 ⎠ ⎝ 2 ⎠<br />
i<br />
( u − 2u<br />
+ u ) = 0<br />
( 1<br />
− )<br />
E dm E V<br />
m<br />
μ = =<br />
E df E V<br />
f<br />
m f<br />
f f<br />
(9)
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 151<br />
Finite Difference Method<br />
We define the non-dimensional displacement coordinate as<br />
where<br />
U = u<br />
i,<br />
j i,<br />
j<br />
/ ξ<br />
, X x<br />
ξ =<br />
The Equation 9 can be rewriten as<br />
A<br />
i<br />
d<br />
2<br />
U<br />
dX<br />
= /ξ ,<br />
The second-order derivative at point (i,j) is:<br />
2 k<br />
dU i,j<br />
d X<br />
no element break<br />
element (i,j)~(i,j+1) break<br />
element (i,j-1)~(i,j) break<br />
2<br />
τ i / i+<br />
1,<br />
j = τ i / i+<br />
1,<br />
j<br />
( + )<br />
dfE f G fdm G mdf<br />
2G<br />
G<br />
f m<br />
( U − U + U ) = 0<br />
i,<br />
j<br />
+ 1,<br />
2 2 i+<br />
j i,<br />
j i−1,<br />
j<br />
Substituting Equation (13) into (12), one can obtain:<br />
UL =<br />
⎧<br />
⎨<br />
⎩<br />
⎪⎧<br />
UR<br />
= ⎨<br />
⎪⎩<br />
U<br />
0<br />
ξ /<br />
E f<br />
df<br />
(10)<br />
(11)<br />
(12)<br />
⎧ 1 k<br />
k k<br />
⎪ 2 ( U i,j − 1 − 2U i,j + U i,j + 1)<br />
( )<br />
⎪<br />
Δx<br />
⎪ 4 k k<br />
= ⎨ 2 ( U i,j − U i,j −1)<br />
⎪ 3(<br />
Δx)<br />
⎪ 4 k k<br />
2 ( U i,j + 1 − U i,j )<br />
⎩⎪<br />
3(<br />
Δx<br />
)<br />
(13)<br />
( UL + UR)<br />
+ ( ΔX<br />
)<br />
C<br />
C<br />
2<br />
k Ai<br />
(<br />
2<br />
4<br />
U = i,<br />
j<br />
C1Ai<br />
+ C2C<br />
3 Δ<br />
k<br />
i,<br />
j−1<br />
U , + 1<br />
k<br />
i j<br />
0<br />
element (i,j-1)~(i,j) unbroken<br />
element (i,j-1)~(i,j) broken<br />
( ) 2<br />
X<br />
UD +<br />
C<br />
5<br />
UU )<br />
(14)
152<br />
⎧<br />
UU = ⎨<br />
⎩<br />
⎧<br />
UD = ⎨<br />
⎩<br />
C1<br />
C 2<br />
C3<br />
⎧<br />
= ⎨<br />
⎩<br />
⎧<br />
= ⎨<br />
⎩<br />
*<br />
τ / + 1,<br />
k<br />
i i j<br />
0<br />
*<br />
τ −1/<br />
,<br />
k<br />
i i j<br />
1<br />
2<br />
1<br />
0<br />
0.<br />
75<br />
⎧0<br />
⎪<br />
= ⎨1<br />
⎪<br />
⎩2<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
element (i,j)~(i,j+1) unbroken<br />
element (i,j)~(i,j+1) broken<br />
interface (i,j)~(i+1,j) unbroken<br />
interface (i,j)~(i+1,j) broken<br />
interface (i-1,j)~(i,j) unbroken<br />
interface (i-1,j)~(i,j) broken<br />
element broken<br />
no element broken<br />
element broken<br />
no element broken<br />
both side matrix broken<br />
single side matrix brokes<br />
no matrix broken<br />
C 4<br />
C<br />
5<br />
1 i ≠ 1<br />
= ⎨⎧<br />
⎩0 i = 1<br />
1 i ≠ n<br />
= ⎨⎧<br />
⎩0 i = n<br />
Using the successive over-relaxation, we have<br />
[ ] [ ] ( )[ ] 1<br />
k q k q<br />
k q−<br />
U = U + 1−<br />
λ U<br />
i,<br />
j<br />
λ (15)<br />
i,<br />
j<br />
where, λ is the relaxation factor, which controls the convergence speed of solution, and q is<br />
the times of iteration.<br />
k<br />
k<br />
i,<br />
j<br />
U i,<br />
j is the right side of Equation (14). After U i,<br />
j is obtained, the<br />
stress of the segment of fiber and matrix can be calculated from following expression:
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 153<br />
For fiber element:<br />
For matrix element<br />
U i+<br />
1,<br />
j −U<br />
i,<br />
j<br />
σ i,<br />
j = E f<br />
(16a)<br />
X<br />
U i+<br />
1,<br />
j −U<br />
i,<br />
j<br />
σ i,<br />
j = Em<br />
(16b)<br />
X<br />
The strain and stress of composites were calculated from average stress of all elements.<br />
c<br />
n m n m<br />
1 ⎡<br />
⎤<br />
= ×<br />
( )<br />
( ) ⎢∑∑σ<br />
2i+<br />
1,<br />
j 1−<br />
V f + ∑∑σ<br />
2i,<br />
jV<br />
f<br />
2N<br />
+ 1 M<br />
⎥<br />
⎣ i−0<br />
1 1 1 ⎦<br />
σ (17)<br />
Strength of Fiber Element and the Failure Criterion<br />
VKΔt<br />
ε c =<br />
(18)<br />
L<br />
Strength assignment to the fiber elements<br />
In simulating, the strength of the fiber elements should be predetermined. According to<br />
the Weibull statistical constitutive model the strength of the fiber follow Equation (4). If we<br />
assume L in Equation (4) equal to mesh length Δx, here σΔx can be obtained from the scale<br />
parameter σ0 at experimental length Lo. n×m random array ηi,j, equally distributed in the<br />
range of (0,1), are produced by the computer, and we let<br />
η<br />
i,<br />
j<br />
β<br />
⎡ Δx<br />
⎛ S ⎤<br />
i,<br />
j ⎞<br />
= P ( Δx,<br />
S ) = ⎢−<br />
⎜<br />
⎟<br />
i,<br />
j exp<br />
⎥<br />
(19)<br />
⎢ L0<br />
⎣ ⎝ σ 0 ⎠ ⎥<br />
⎦<br />
From the Equation (19), we can get the strength of fiber element<br />
The failure criterion<br />
The failure criterion of fiber is<br />
i,<br />
j ≥ S i,<br />
j<br />
1<br />
i,<br />
j<br />
⎡ L0<br />
= − ln j σ<br />
⎤ β<br />
( ηi,<br />
) 0<br />
S ⎢ ⎥<br />
(20)<br />
⎣ Δx<br />
⎦<br />
σ fiber element broken (21a)
154<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
σ fiber element unbroken (21b)<br />
i,<br />
j ≤ S i,<br />
j<br />
The failure criterion of interface is<br />
τ ≥ τ<br />
i,<br />
j m<br />
interface broken (22a)<br />
τ ≤ τ<br />
i,<br />
j m<br />
interface unbroken (22b)<br />
where, τm is the ultimate shear stress of matrix.<br />
The failure criterion of matrix is<br />
ε ≥ ε<br />
i,<br />
j m<br />
matrix element broken (23a)<br />
ε ≤ ε<br />
i,<br />
j m<br />
matrix element unbroken (23b)<br />
where, εm is the failure strain of the matrix<br />
Simulation Procedure<br />
Based on the above numerical constitutive model, a computational program is compiled to<br />
simulate the microscopic dynamic failure process of unidirectional composites. The<br />
simulation procedure is illustrated as following:<br />
(A) Randomly assign a statistical strength Si,j (i=2,4,...2n, J=1,2,…m) to the fiber<br />
element, definitely assign the tensile strength to the matrix and the shear strength to<br />
the interface.<br />
(B) Solve Equation (14) by iteration formula (15) using Ui,j k−1 , Ui,j k−2 and the boundary<br />
conditon at time t=kΔt, obtain the displacement field Ui,j k (i=1,2,...2n+1,j=1,2,...m).<br />
(C) Determine whether the element or the interface breakage (or the matrix element<br />
unloading) has happened, if new breakage occurs, take the breakage into account and<br />
repeat steps (B) and (C) until no new breakage occurs, else calculate the apparent<br />
stress σc and strain εc using Equation (17) and (18).<br />
(D) Increase a time step and repeat step (B) and (C) till the composites failure happens.<br />
Here the "composites failure" is defined as the state when the stress σc drops from<br />
σmax to about 50% of σmax.<br />
Experimental Results of Fiber, Matrix and <strong>Composite</strong><br />
The statistical parameters of fiber were obtained from tension tests of T700 fiber bundles. The<br />
tensile stress-strain curve of fiber bundle in Figure 22 shows considerable amount of nonlinearity.<br />
The specimen failed gradually after reaching the maximum stress due to the tensile<br />
strength distribution of fibers. Three parameters were determined from each stress-strain
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 155<br />
curve: elastic modulus (E), tensile strength ( σ b ), and failure strain ( ε b ). Elastic modulus or<br />
Young's modulus is the initial slope of the stress-strain curve. Tensile strength is the<br />
maximum stress at the peak load and the strain corresponding to the tensile strength is the<br />
failure strain. The average values of these three properties are:<br />
E = 210GPa<br />
σ = 1.<br />
93GPa<br />
ε = 1.<br />
07%<br />
b<br />
Based on fiber bundles model and statistical theory of fiber strength [34], the Weibull<br />
parameters for tensile strength of carbon fibers also can be obatined:<br />
Stress (GPa)<br />
2.00<br />
1.60<br />
1.20<br />
0.80<br />
0.40<br />
0.00<br />
σ 2.<br />
70GPa<br />
β = 9.<br />
03 L 100mm<br />
0 =<br />
T700 Carbon Fiber<br />
Experimental Results<br />
Simulated Results<br />
0.000 0.005 0.010 0.015 0.020<br />
Strain<br />
Figure 22. Stress strain curve of carbon fiber.<br />
b<br />
0 =
156<br />
Stress (MPa)<br />
120<br />
80<br />
40<br />
0<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
Neat Epoxy<br />
+1.% SiC Nano Particle Filled Epoxy<br />
0.00 0.02 0.04 0.06<br />
Strain (mm/mm)<br />
Figure 23. Stress-strain curves of the epoxy.<br />
Figure 23 shows the stress-strain curves of epoxy and nanophased epoxy. It can be<br />
observed that both the modulus and strength have been improved by filling nano particle into<br />
matrix system. All parameters of fiber and matrix were listed in Table 6.<br />
Table 6. Parameters of T700 fiber and matrix<br />
Material T700 carbon fiber Neat epoxy Nano-phased epoxy<br />
E (GPa) 210 2.45 3.32<br />
G (GPa) 87.5 1.02 1.38<br />
D ( μ m ) 5 -------- --------<br />
Vf (or Vm) 49% 51% 51%<br />
β 9.03 -------- -------σ<br />
0 ( GPa)<br />
2.70 -------- -------σ<br />
b (MPa)<br />
-------- 89 110<br />
τ ( ) -------- 45 55<br />
max GPa
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 157<br />
Simulation Result and Discussion<br />
Figure 24 shows the simulation results of neat and 1.5wt.% SiC nano-phased layered<br />
composites along with the experimental tensile results. It can be observed that the numerical<br />
simulation results agree well with the experimental results. This indicates that the numerical<br />
model, the simulation method and the program are reliable. Simulated results and<br />
experimental results also show that the strength of nanocomposite was about 7-10% higher<br />
than that of the neat one. As already discussed in the previous chapter, it may have been<br />
contributed from the improvement of matrix and interface properties.<br />
Figure 24. Simulated Stress-strain curves of tensile test.<br />
The failure strain of matrix is higher than that of fiber in unidirectional composites, the<br />
fiber element with the lowest strength is first broken. Then, with increasing applied load, the<br />
breakage of fiber elements occurs randomly. On the other hand, high shear stress are<br />
generated in the matrix elements due to the fiber breakages, so that the matrix cracking and<br />
interfacial debonding occur, leading to the final failure of composites. Figure 25 shows the<br />
simulated failure process of composite from the above results. Figure 25a and 25b indicate<br />
the initial fracture occurring at fiber at low stress level. Stress concentration in the matrix
158<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
elements results in matrix cracking and interfacial debonding. One can observe the damage<br />
zone of nano-phased composite is smaller than that of neat system. Figure 25c and 25d show<br />
the failure appearances at about 60% Peak load of the composites. Figure 25e and 25f show<br />
the finial failure appearances of composites. In these figures, the fiber element breakage of<br />
two composite are almost same, but matrix element breakage of nano-phased composite is<br />
smaller than that of neat system.<br />
Figure 25. Continued on next page.<br />
(a) (b)<br />
(c) (d)
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 159<br />
(e) (f)<br />
Figure 25. Simulated tensile failure process of neat and nano-phased composites.<br />
Figure 26. Stress strain curves of CFRP with different interface strengths.
160<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
Figure 27. Relationship between composite’s tensile strength and interface strength.<br />
Interface strength is a very important parameter affecting the strength of composite.<br />
Based on the current numerical model, the effect of interface strength on the tensile failure<br />
process of unidirectional composites has been studied. Figure 26 shows the tensile stress–<br />
strain curves of composite with different interface strength (100, 50, 20 and 10MPa). The<br />
relationship between the strength of composites and interface strength are shown in Figure 27.<br />
From Figure 26 and 27, it can be observed that the tensile strength of the composites<br />
increases with interface strength, and when the value of interface strength is over 50 MPa, the<br />
tensile strength of composites tends to be a constant and not affected by interface strength.<br />
Figure 28 shows the simulated micro damage patterns of the composites with weak interface<br />
and strong interface. Comparing these patterns with the macro stress–strain curves, it is<br />
evident that: when the interface strength gets weaker, the adjacent interfaces get easier to<br />
break after the fibers break, the load transfer along the traverse direction gets more<br />
ineffective, the reinforced function of fiber gets more insufficient, the damage area gets<br />
larger, the damage pattern gets more disordered, the negative effect gets greater, thus the<br />
composite’s strength gets lower; when the interface strength gets stronger, the fiber breakage<br />
propagates throughout the composite along the straight line and the composites strength gets<br />
higher.
Conclusion<br />
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 161<br />
(a) (b)<br />
Figure 28. Simulated micro damage patterns. (a) Weak interface, (b) Strong interface.<br />
Using the solution impregnation and filament winding techniques, an innovative<br />
manufacturing method was introduced to manufacture nano-phased carbon/epoxy prepreg<br />
tapes/tows from a nanoparticle modified epoxy resin system.<br />
It was seen from TGA that the as-prepared nanocomposites were thermally more stable<br />
than their neat counterparts. An improvement of about 7 0 C was noticed. The improvement in<br />
thermal stability was believed to have been caused by increased cross-linking and better<br />
interactions between the epoxy and SiC nanoparticles.<br />
As seen with DSC and DMA, an improvement in glass transition temperature (Tg) of the<br />
nano-phased system was about 5 0 C. The improvement is due to the restricted mobility of the<br />
chain segments in the presence of SiC nanoparticles, as a result of a greater number of crosslinked<br />
polymer chains.<br />
Response of nano SiC infused composites under flexure loading showed significant<br />
improvements in strength as well as stiffness over the neat ones. On an average the increment<br />
in strength and stiffness was 32% and 20% respectively over the neat systems. It is believed<br />
that the higher strength of the SiC systems is attributed to better interfacial bonding and the<br />
resistance offered by the nanoparticles to crack propagation.<br />
In respect of tensile strength and stiffness, the nanocomposite systems offered<br />
improvements between 7 and 10%. It was seen that all the test coupons (neat and nanophased)<br />
failed in the gage length and failure modes were as described in the ASTM standard.<br />
This nominal improvement in the tensile properties was believed to have occurred because in<br />
tension, stresses in fibers were more dominant than in the nano-phased matrix and the<br />
property improvement is merely due to the improved properties of the resin.<br />
Monte Carlo simulation technique has been established to simulate unidirectional layered<br />
neat and nano-phased composites. The simulation results are in agreement with the<br />
experimental results. Tensile failure process was simulated and the damage zone and micro-
162<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
damage patterns were studied. The damage zone for the nano-phased system was found to be<br />
less than that of the neat system. It was also shown that the interface strength played an<br />
important role in the composite’s tensile failure.<br />
Acknowledgements<br />
The authors would like to gratefully acknowledge the support of USACERL through grant<br />
no.:W9132T-07-p-0011 and National Science Foundation.<br />
Reference<br />
[1] Imanaka M., Nakamura Y., Nishimura A. and Iida T., Fracture toughness of rubbermodified<br />
epoxy adhesives: effect of plastic deformability of the matrix phase.<br />
<strong>Composite</strong>s Science and Technology, Volume 63, Issue 1, 2003, Pages 41-51<br />
[2] Chikhi N., Fellahi S. and Bakar M. Modification of epoxy resin using reactive liquid<br />
(ATBN) rubber. European Polymer Journal, Volume 38, Issue 2, 2002, Pages 251-264<br />
[3] Xian G.J., Walter R. and Haupert F., Friction and wear of epoxy/TiO2 nanocomposites:<br />
Influence of additional short carbon fibers, Aramid and PTFE particles. <strong>Composite</strong>s<br />
Science and Technology, Volume 66, Issue 16, 2006, Pages 3199-3209<br />
[4] Vasconcelos P.V., Lino F.J., Magalhaes A. and Neto R.J.L., Impact fracture study of<br />
epoxy-based composites with aluminium particles and milled fibres, Journal of<br />
<strong>Materials</strong> Processing Technology, Volume 170, Issues 1-2, 2005, Pages 277-283<br />
[5] Reynaud E., Gauthier C. Perez J., Review Metallurgie 96(2), 1999, pp169-176.<br />
[6] Wang W. X., Takao Y., Matsubara T. and Kim H.S., Improvement of the interlaminar<br />
fracture toughness of composite laminates by whisker reinforced interlamination,<br />
<strong>Composite</strong>s Science and Technology 62 (2002), pp 767–774.<br />
[7] Sherman D., Lemaitre J. and Leckie F., The mechanical behavior of an alumina<br />
carbon/epoxy laminate, Acta Metall. Mater. Vol. 43 No. 12 (1995), pp 4483-4493.<br />
[8] Chisholm N., Mahfuz H,. Rangari V.K., Ashfaq A. and Jeelani S., Fabrication and<br />
mechanical characterization of carbon/SiC-epoxy Nanocomposites, <strong>Composite</strong><br />
Structures, Volume 67, Issue 1, 2005, pp115-124<br />
[9] Gojny F.H., Wichmann M.H.G, Fiedler B., Bauhofer W. and Schulte K., Influence of<br />
nano-modification on the mechanical and electrical properties of conventional fibrereinforced<br />
composites, <strong>Composite</strong>s Part A: Volume 36, Issue 11 , 2005, pp1525-1535.<br />
[10] Reznik B., Gerthsen D., Zhang W., Huttinger K. J. 2003. Microstructure of SiC<br />
deposited form methyletrichlorosilane. Journal of European Ceramic Society 23: 1499-<br />
1508.<br />
[11] Harris G. L., Yang C. Y. W. 1989. Amorphous and crystalline silicon carbide and<br />
related materials. Springer Proceedings in Physics 34, Berlin and New York.<br />
[12] Rahman M. M., Harris G. L., Yang C. Y. W. 1989. Amorphous and crystalline silicon<br />
carbide- II: Recent developments. Springer Proceedings in Physics 43, Berlin and New<br />
York.
An Experimental and Analytical Study of Unidirectional Carbon Fiber… 163<br />
[13] Heuner A. H., Fyburg G.A., Ogbuji L.U., Mitchel T.E. 1978. A transformation in<br />
polycrystalline SiC: I, microstructural aspects. Journal of American Ceramic Society 61:<br />
406-411.<br />
[14] Najajima Y. 1991. Silicon carbide ceramics-1, Fundamental and solid reaction. Editors<br />
Somiya S., Inometa Y. Elsivier Applied Science, London and New York.<br />
[15] Eskin G.I. 2001. Broad prospects for commercial application of the ultrasonoic<br />
(cavitation) melt treatment of light alloys. Ultrasononic Sonochemistry, Moscow:<br />
319-325.<br />
[16] Kumar R. V., Koltypin Y., Gadanken A. 2002. Preparation and characterization of<br />
nickel-polystyrene nanocomposite by ultrasound irradiation. Journal of Applied Polymer<br />
Science 86: 160.<br />
[17] Patel N., Lee L. J. 1996. Modeling of void formation and removal in liquid composite<br />
molding. Part I: Wettability Analysis. Polymer <strong>Composite</strong>s 17 (1): 96-103.<br />
[18] Good R. J. 1979. Contact Angles and the Surface Free Energy of Solids. From Surface<br />
and Colloid Science II – Experimental Methods. Editors: Good R. J., Stromberg R. R.<br />
Plenum Press, New York, USA.<br />
[19] Moon C. K., Um Y. S., Lee J. O., Cho H. H., Park C. H. 1993. Development of<br />
thermoplastic prepreg by the solution-bond method. Journal of Applied Polymer Science<br />
47: 195-188.<br />
[20] Lacroix F. V., Werwer M., Schulte K. 1998. Solution impregnation of polyethylene<br />
fiber/polyethylene matrix composites. <strong>Composite</strong>s Part A 29A: 371-376.<br />
[21] Smith F. C., Moloney L. D., Matthews F. L. 1996. Fabrication of woven carbon fibre/<br />
polycarbonate repair patches. <strong>Composite</strong>s Part A 27A: 1089-1095.<br />
[22] Wu G. M., Schultze J. M., Hodge D. J., Cogswell F. N. 1990. Solution impregnation of<br />
carbon fiber reinforced poly(ethersulphone) composites. From ANTEC 90 Conference<br />
Proceedings – Plastics in the Environment: Yesterday, Today and Tomorrow, Texas,<br />
U.S.A.<br />
[23] Fukuda H., Chou T. W. 1982. Monte Carlo simulation of the strength of hybrid<br />
composites. Journal of <strong>Composite</strong>s <strong>Materials</strong> 16: 371-385.<br />
[24] Oh K. P. 1979. A Monte Carlo study of the strength of unidirectional fiber reinforced<br />
composites. Journal of <strong>Composite</strong> <strong>Materials</strong> 13: 311–328.<br />
[25] Curtis P. T. 1986. A computer model of the tensile failure process in unidirectional fibre<br />
composites. <strong>Composite</strong>s Science and Technology 27: 63–86.<br />
[26] Lienbkamp M., Schwartz P. 1993. A Monte Carlo simulation of the failure of a seven<br />
fibre microcomosite. <strong>Composite</strong>s Science and Technology 46: 139–146.<br />
[27] Goda K., Phoenix S. L. 1994. Reliability approach to the tensile strength of<br />
unidirectional CFRP composites by Monte Carlo simulation in a shear-lag model.<br />
<strong>Composite</strong> Science and Technology 50: 457-468.<br />
[28] Yuan J., Xia Y., Yang B. 1994. A note on the Monte Carlo simulation of the tensile<br />
deformation and failure process of unidiretional composites. <strong>Composite</strong>s Science and<br />
Technology 52: 197–204.<br />
[29] Wang Z., Yuan J., Xia Y. 1998. A dynamic Monte Carlo simulation for unidirectional<br />
composites under tensile impact. <strong>Composite</strong>s Science and Technology 58: 487–495.<br />
[30] Xia Y., Wang Z. 1999. A dynamic Monte Carlo microscopic damage model<br />
incorporating thermo-mechanical coupling in a unidirectional composites. <strong>Composite</strong>s<br />
Science and Technology 59: 947–955.
164<br />
Yuanxin Zhou, Hassan Mahfuz, Vijaya Rangari et al.<br />
[31] Xia Y., Wang Y. 2001. An improved dynamic Monte Carlo microscopic numerical<br />
constitutive model incorporating thermal-mechanical coupling for unidirectional GRP<br />
under tensile impact. <strong>Composite</strong>s Science and Technology 61: 997–1003.<br />
[32] Ochiai S., Tokinori K., Osamura K., Nakatani E., Yamatsuta Y. 1991. Stress<br />
concentration at a notch-tip in unidirectional metal matrix composites. Metall Trans<br />
22A: 2085–2095.<br />
[33] Ochiai S., Osamura K. 1986. Stress distribution in a segmented coating film on metal<br />
fiber under tensile loading. Journal of <strong>Materials</strong> Science 21: 2744-2752.<br />
[34] Zhou Y., Jiang D., Xia Y. 2001. Tensile mechanical behavior of T300 and M40J fiber<br />
bundles at different strain rates. Journal of <strong>Materials</strong> Science 36: 919-922.<br />
[35] Mahesh S., Hanan J. C., Üstündag E., Beyerlein I. J. 2004. Shear-lag model for a single<br />
fiber metal matrix composite with an elasto-plastic matrix and a slipping interface.<br />
International Journal of Solids and Structures 41: 4197-4218.<br />
[36] Huang W., Nie X., Xia Y. 2003. An experimental study on the in situ strength of SiC<br />
fibre in unidirectional SiC/Al composites. <strong>Composite</strong>s Part A: Applied Science and<br />
Manufacturing 34: 1161-1166.<br />
[37] Zhou Y., Wang Y., Xia Y., Mallick P. K. 2003. An experimental study on the tensile<br />
behavior of a unidirectional carbon fiber reinforced aluminum composite at different<br />
strain rates. <strong>Materials</strong> Science and Engineering A, 362: 112-117.
In: <strong>Composite</strong> <strong>Materials</strong> <strong>Research</strong> <strong>Progress</strong> ISBN: 1-60021-994-2<br />
Editor: Lucas P. Durand, pp. 165-207 © 2008 Nova Science Publishers, Inc.<br />
Chapter 5<br />
DAMAGE EVALUATION AND RESIDUAL STRENGTH<br />
PREDICTION OF CFRP LAMINATES BY MEANS<br />
OF ACOUSTIC EMISSION TECHNIQUES<br />
Giangiacomo Minak 1 and Andrea Zucchelli 2<br />
Department of Mechanical Engineering, Alma Mater Studiorum - Università di Bologna,<br />
Viale Risorgimento 2, 40135 Bologna, Italia<br />
Abstract<br />
A new approach that integrates acoustic emission (AE) and the mechanical behaviour of<br />
composite materials is presented. Usually AE information is used to evaluate qualitatively the<br />
damage progression in order to assess the structural integrity of a wide variety of mechanical<br />
elements such as pressure vessels. From the other side, the mechanical information, e.g. the<br />
stress-strain curve, is used to obtain a quantitative description of the material behaviour. In<br />
order to perform a deeper analysis, a function that combines AE and mechanical information<br />
is introduced. In particular, this function depends on the strain energy and on the AE events<br />
energy, and it was used to study the behaviour of CFRP composite laminates in different<br />
applications: (i) to describe the damage progression in tensile and transversal load testing; (ii)<br />
to predict residual tensile strength of transversally loaded laminates (condition that simulates a<br />
low velocity impact).<br />
Introduction<br />
Long fibre reinforced composite laminates are a complex structure at the meso-scale. The<br />
fibres embedded in the matrix constitute the lamina and the overlapping of different laminas<br />
makes the composite laminate. A consequence of this architecture is the complex behaviour<br />
during loading and servicing of components realized by such material, and the multiplicity of<br />
different failure mechanisms that determine the damage progression.<br />
1 E-mail address: giangiacomo.minak@unibo.it<br />
2 E-mail address: a.zucchelli@unibo.it
166<br />
Giangiacomo Minak and Andrea Zucchelli<br />
In particular the prevision of the performances (e.g. stiffness, damping) and the strength<br />
limits (e.g. tensile, compressive and fatigue) of this kind of material is an important task for<br />
the material scientists and engineers.<br />
Numerical models for the composite laminate progressive failure are currently developed<br />
by the researchers for different applications [1-3] .<br />
These models require an experimental validation by means of tests in which damage<br />
progression is monitored in a suitable way.<br />
Nowadays, different techniques are proposed such as electrical resistance [4], fibre Bragg<br />
grating sensors [5], photo-elasticity [6] and acoustic emission [7-9].<br />
On the other hand, residual strength evaluation after fatigue or impact loading is<br />
important for the determination of composite components reliability. In fact, laminate<br />
composite materials have a wide application in light-weight structural members. In particular<br />
fibre-reinforced plastics are increasingly used in airborne structures and the long range<br />
passenger airplanes of the future may include many important parts of the fuselage and<br />
components made with Carbon Fibre Reinforced Plastic (CFRP). This class of materials is<br />
characterized by outstanding strength-to-weight and stiffness-to-weight ratios. Nevertheless<br />
their resistance to accidental damage is an important issue for the designer. In particular,<br />
CFRPs are very susceptible to internal damage caused by transverse loads such as indentation<br />
and impact, while the probability of such loadings occurring during the manufacture, service<br />
or maintenance of composite structures is very high [10]. This lack of resistance to low<br />
velocity and low energy impact damage [11-13] is one of the main obstacles to a more<br />
widespread application of these composite materials, especially in the case of a thermo-set<br />
matrix like epoxy.<br />
A threshold, conventionally located at 20 m/s, divides the impact problems into two<br />
fields, high and low velocity, due to the different types of induced damage [14].<br />
In the low-velocity impact field, a quasi-static loading can simulate the actual behaviour,<br />
since the vibrational effects are negligible [15, 16]. In fact, many researchers [17-23] use<br />
load-displacement histories to compare structural responses from impact and quasi-static tests<br />
and they find that both the dynamic and static responses have corresponding load drops due to<br />
failures in the laminates.<br />
Low velocity and low energy impact damage usually consists of matrix cracking [24, 25]<br />
and delamination [23, 26, 27], while debonding and fibre breaking occur for higher impact<br />
energy values [28, 29].<br />
As said, besides the behaviour of the material during an impact, an issue of great interest<br />
is the evaluation of the post-impact resistance characteristics of CFRP. In fact, damage due to<br />
impact often can be present in the component before it is put into service and loaded.<br />
To detect the damage level present in the laminate and the damaged zone area, several<br />
techniques are used, such as simple visual inspection , C-scan and X-ray. AE event counts are<br />
also utilized to predict the residual tensile strength (RTS) after impact [30].<br />
Due to the importance of delamination, which decreases locally the buckling load, much<br />
effort has been spent in researching the compression after impact (CAI) performances of<br />
composites [31, 32]. Nevertheless tensile [30, 33, 35] and fatigue properties [36, 37] are also<br />
important to predict the component failure.
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 167<br />
In this chapter after a brief introduction about AE, we define a function of the acoustic<br />
energy and of the strain energy that allowed us to characterize damage in two different load<br />
configurations:<br />
tensile testing, on different types of CFRP laminate specimens;<br />
transversal loading of quasi-isotropic CFRP laminate plates.<br />
Moreover in the second case, the residual tensile strength was related to the AE recorded<br />
during the transversal loading phase.<br />
Acoustic Emission Technique<br />
Nearly every scientist who makes mechanical experimental testing is familiar with the<br />
acoustic emission produced by the material during the loading phase, which sometimes can be<br />
heard simply by naked ears.<br />
In fact, during a material test or in general when a component is subject to external loads,<br />
a rapid stress redistribution can occur due to permanent and irreversible phenomena, caused<br />
by damage mechanisms, such as a matrix crack onset or growth, delamination and fibre<br />
fracture. During this redistribution, part of the strain energy stored in the material is released<br />
in the form of heat and of elastic waves that propagate in the material until they reach the free<br />
surface. These transient elastic waves are commonly detected as acoustic waves.<br />
Some acoustic emission can be also produced by mechanisms different from damage<br />
(such as sliding and friction of two surfaces in contact) and this must be taken into account.<br />
The elastic waves propagating at the component surfaces are detected by means of<br />
piezoelectric devices that convert the mechanical signal into an electrical one.<br />
Even if the AE physical principle is very simple and immediate, the use of this technique<br />
is not so straightforward because the acoustic wave propagation in solids, especially in the<br />
anisotropic ones as CFRP, is quite complicated. Multiple waves that propagate with different<br />
velocities, reflection, refraction, dispersion, and attenuation, may affect the measured signal.<br />
Nevertheless some advantages with respect to other non destructive testing techniques can be<br />
found in the possibility to monitor a large volume of material by means of few sensors able to<br />
locate the damage by triangulation and to make it continuous during real life service. In<br />
reality, the acoustic emission is produced within the material itself once loaded at a level that<br />
produces some form of damage. In this sense, it is not strictly a non destructive testing<br />
method since it is based on passive monitoring of acoustic energy released by the material or<br />
structure itself while under load.<br />
AE is also used to monitor damage onset and progression in laboratory tests [7-9, 38, 39].<br />
The most difficult AE analysis task is the identification of the damage mechanism,<br />
particularly when multiple damage mechanisms are present as in the case of CFRP.<br />
In fact, changes in AE due to propagation in the material and to the measurement system<br />
may mask characteristics that are related to the damage mechanism. If the AE source is<br />
known, as in the case of uniaxial specimens, the quality of AE data can be improved by noise<br />
discrimination and rejection.
168<br />
Giangiacomo Minak and Andrea Zucchelli<br />
Numerous methods have been attempted to identify damage mechanisms from AE data<br />
[40,41] and in general many carefully controlled laboratory experiments are necessary to<br />
develop relationships between measured AE signals and a specific damage mechanism.<br />
The results from AE monitoring have been used in attempts to estimate the residual<br />
strength or life of a structure [34]. Most strength assessments from AE are based on empirical<br />
correlations developed from failure tests on a large number of nominally identical structures<br />
[42].<br />
Even if recently the research on AE, especially as regards composite laminates focused<br />
on modal analysis [43-45] in this chapter we consider classical feature-based (also known as<br />
parametric) AE analysis [38], in which for each acoustic emission a set of meaningful<br />
parameters (shown in figure 1) are detected such as:<br />
- progressive event number<br />
- counts per event<br />
- maximum amplitude within the event<br />
- event duration<br />
- event energy<br />
Figure 1. Acoustic emission parameters.<br />
This approach has been used in composite laminates with different AE interpretations.<br />
Many authors (e.g. Siron and tsuda [40] ) report that fibre breakage produces large amplitude<br />
signals while matrix cracking results in much smaller amplitudes and delamination is thought<br />
to produce medium amplitude signals. Other studies conclude that matrix cracking causes<br />
large amplitude signals while fibre breakage produced low amplitudes [47].<br />
In reality the amplitude depends on a number of factors including the local stress<br />
conditions and the energy released. In fact, for example, in [43] a very small increment of<br />
matrix crack growth produces a much smaller amplitude signal than a large matrix crack.<br />
Moreover, long duration events are attributed to delamination [44].
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 169<br />
More details on AE methods can be found for example in [46].<br />
Our contribution in this field is the introduction of a novel function able to combine the<br />
acoustic energy released during an event and the strain energy stored in the material in that<br />
moment. This function will be called “sentry” function because it signals important material<br />
damage events during the tests, and its integral is related to the total damage and to the<br />
residual strength as it will be shown in the next paragraph and in the examples.<br />
The Sentry Function<br />
In order to perform a deeper analysis of the laminate behaviour, a function that combines both<br />
the mechanical and acoustic energy information [35, 47] is introduced. This function is<br />
expressed in terms of the logarithm of the ratio between the strain energy (Es) and the<br />
acoustic energy (Ea),<br />
( )<br />
( ) ⎟⎟<br />
x ⎞<br />
x<br />
⎛ Es<br />
f ( x)<br />
= Ln ⎜<br />
(1)<br />
⎝ Ea<br />
⎠<br />
where x is the test driving variable (usually displacement or strain).<br />
The function f(x) takes into account the continuous balancing between the stored strain<br />
energy and the released acoustic energy due to damage. The function f(x) is generally<br />
discontinuous and can be described by the combinations of four types of function, shown in<br />
figure 2: (I) an increasing function PI(x), (II) a sudden drop function PII(x), (III) a constant<br />
function PIII(x) and (IV) a decreasing function PIV(x).<br />
These functions are defined over an “acoustic emission domain” (ΩAE) that correspond to<br />
the displacement range over which the AE events were recorded. For all AE quantities ΩAE<br />
represents the definition domain and outside the function of AE cumulative events,<br />
cumulative counts, events energy and all other quantities related to the AE information are<br />
null.<br />
Dividing the AE domain ΩAE in sub-domain as reported in figure 2 it possible to write the<br />
following relation:<br />
Ω = Ω U Ω U Ω U Ω<br />
(2)<br />
AE AE,<br />
I AE,<br />
II AE,<br />
III AE,<br />
IV<br />
In that condition the function f can be written as follow:<br />
( x)<br />
( x)<br />
( x)<br />
( x)<br />
⎧ PI<br />
⇔ x ∈ ΩAE,<br />
I<br />
⎪<br />
⎪<br />
PII<br />
⇔ x ∈ ΩAE,<br />
II<br />
f = ⎨PIII<br />
⇔ x ∈ ΩAE,<br />
III<br />
(3)<br />
⎪PIV<br />
⇔ x ∈ ΩAE,<br />
IV<br />
⎪<br />
⎪⎩<br />
0 x ∉Ω<br />
AE
170<br />
⎟ ⎞<br />
⎠<br />
s<br />
⎜ ⎛<br />
⎝<br />
=<br />
a<br />
E<br />
E<br />
Ln<br />
f<br />
Giangiacomo Minak and Andrea Zucchelli<br />
PI<br />
Ω AE,I<br />
….<br />
PII<br />
….<br />
Ω AE,II<br />
Ω AE<br />
PIII<br />
Ω AE,III<br />
….<br />
PIV<br />
Ω AE,IV<br />
Displacement<br />
Figure 2. The basic functions P I , P II , P III and P IV, used to describe the function f.<br />
If there is more than one sub-domain ΩAE,k, k∈{I,II,III,IV}, it is possible to write the<br />
following relation:<br />
Ω<br />
n I ⎛<br />
= ⎜<br />
UΩ<br />
⎝ i=<br />
1<br />
n II<br />
n III<br />
n IV<br />
⎞ ⎛ i ⎞ ⎛ i ⎞ ⎛ i ⎞<br />
⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟<br />
, I ⎟<br />
U<br />
⎜U<br />
ΩAE,<br />
II ⎟<br />
U<br />
⎜U<br />
ΩAE,<br />
III ⎟<br />
U<br />
⎜U<br />
ΩAE<br />
VI ⎟<br />
(4)<br />
⎠ ⎝ i=<br />
1 ⎠ ⎝ i=<br />
1 ⎠ ⎝ i=<br />
1 ⎠<br />
AE<br />
i<br />
AE<br />
,<br />
where nI, nII, nIII, and nIV are the number sub-domain corresponding to the trend type I,II, III<br />
and IV respectively. Then functions PI,PII, PIII and PIV can be written as follow:<br />
P<br />
P<br />
I<br />
( x)<br />
III<br />
( x)<br />
⎧ PI<br />
⎪<br />
= ⎨<br />
⎪<br />
⎩PI<br />
,<br />
, 1<br />
n I<br />
⎧ P<br />
⎪<br />
= ⎨<br />
⎪<br />
⎩<br />
PIII<br />
( x)<br />
( x)<br />
III,<br />
1<br />
, n III<br />
/ x ∈ Ω<br />
...<br />
/ x ∈Ω<br />
( x)<br />
( x)<br />
1<br />
AE,<br />
I<br />
n I<br />
AE,<br />
I<br />
/ x ∈Ω<br />
...<br />
/ x ∈ Ω<br />
1<br />
AE,<br />
III<br />
n III<br />
AE,<br />
III<br />
P<br />
P<br />
II<br />
( x)<br />
IV<br />
( x)<br />
⎧ P<br />
⎪<br />
= ⎨<br />
⎪<br />
⎩PII<br />
II,<br />
1<br />
, n II<br />
⎧ P<br />
⎪<br />
= ⎨<br />
⎪<br />
⎩<br />
P<br />
( x)<br />
IV,<br />
1<br />
( x)<br />
IV,<br />
n IV<br />
/ x ∈ Ω<br />
( x)<br />
...<br />
/ x ∈ Ω<br />
( x)<br />
1<br />
AE,<br />
II<br />
/ x ∈ Ω<br />
...<br />
n II<br />
AE,<br />
II<br />
/ x ∈ Ω<br />
1<br />
AE,<br />
IV<br />
n IV<br />
AE,<br />
IV<br />
From the physical point of view the part of f(x) characterized by an increasing trend, type<br />
I, represents the strain energy storing phases. The slope of PI,i (x) functions decreases during<br />
(5)
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 171<br />
the test because the energy stored in the material tends to its limit and the AE cumulate<br />
energy (Ea) increases due to the progression of material damage.<br />
When a significant internal material failure happens there is an instantaneous release of<br />
the stored energy that produces an AE event with a high energy content. This fact is<br />
highlighted by the sudden drops of the function f(x) that can be described by functions of type<br />
II: PII(x).<br />
In figure 3 are reported some examples of combinations of parts of the function f(x), and<br />
in particular in figure 3a it possible to note three trends of type I (PI,1, PI,3 and PI,2)<br />
respectively connected by two functions of type II (PII,1 and PII,2).<br />
f<br />
f<br />
f<br />
PI,1<br />
Displacement<br />
PI,4<br />
Displacement<br />
Displacement<br />
S1+<br />
PI,5<br />
S1-<br />
PI,2<br />
PIII,<br />
PII,1<br />
S2+<br />
S2-<br />
1 PIV,1<br />
PIV,2<br />
PI,3<br />
PII,2<br />
PII,3 PII,4<br />
Figure 3. Examples of compositions of the basic function describing common parts of the function f.<br />
PII,5<br />
A<br />
B<br />
C
172<br />
Giangiacomo Minak and Andrea Zucchelli<br />
During the first phase, PI,1, the material is storing the strain energy, but when an internal<br />
limit is reached a failure happens and the function f presents a first sudden drop (PII,1). After<br />
this first drop the material starts again to store strain energy and it is interesting to notice that<br />
the slope S1- of PI,1, before the event, is equal to the initial lope of PI,2 (S1+). The failure that<br />
caused the first fall in this case does not affect significantly the material integrity (i.e. PII,1 is<br />
not related to an important material modification). So the value of f(x) is reduced due to the<br />
internal energy release, but the material strain energy capability is not compromised.<br />
Different is the case represented by the second fall PII,2 in which the slope before (S2-)<br />
and after (S2+) are different. This means that after PII,2 the material strain energy storing<br />
capability is changed and because of an important material modification. In particular it is<br />
interesting to notice that S2->S2+. After this event it is also reasonable to hypothesize that the<br />
material damage increases. It was also previously observed [47], that typically after one or<br />
two drops of f(x) characterized by the slope variation of PI(x) (S->S+) the next drop is<br />
followed by a function of type III or IV. In figure 3B and figure 3C are reported two possible<br />
situations that happen when the material damage has reached an important level. In the case<br />
reported in figure 3B at the end of the storing energy phase PI,4 there is an instantaneous strain<br />
energy release that causes the falling phase PII,3. The following constant behaviour of f(x),<br />
described by PIII,1, is due to a progressive strain energy storing phase that is superimposed to<br />
an equivalent material damage progression. The next fall PII,4 is followed by a decreasing<br />
function PIV,1. The decreasing function type PIV, is related to the fact that the AE activity is<br />
greater then the material strain energy storing capability: the damage has reached a maximum<br />
and the material has no resources to bear the load. Then the phase PIV,1 in figure 3B indicates<br />
that the material is totally damaged. In figure 3B between PIII,1 and PIV,1 there is a drop<br />
function PII,4 that is due to an important failure inside the material, but this situation is not the<br />
general one. In fact, sometimes it happens that after a constant trend of f(x) there are no more<br />
drops and f(x) gradually decreases. This behaviour is due to a gradual damage progression<br />
inside the material and no important failure events happen. At the opposite, the case<br />
represented in figure 3C is related to a critical event inside the material: the strain energy<br />
storing phase PI,5 is followed by a sudden drop PII,5 and then it follows a decreasing phase<br />
PIV,2. This situation generally happens at the end of a test or, if it happens at the beginning it<br />
reveals the presence of some defects inside the material.<br />
The described analysis shows that the function f(x) can be usefully implemented to<br />
describe the material damage progression because it takes into account both mechanical and<br />
acoustic information. Summarizing we have that the increasing part of f(x) reveals the<br />
material strain energy storing capability, the falls reveal the instantaneous release of the<br />
stored energy due to failures, the constant and decreasing trends of f(x) prelude and describe<br />
an important failure of the material structure. In the following sections two examples<br />
regarding the application of different strategies about the use of the function f(x) to describe<br />
the composite material damage are developed. In the first example the function f(x) is used to<br />
determine the damage progression of five different types of laminates under in plane tensile<br />
loading. In particular considering the function f(x) it was possible to identify the most<br />
important material failure highlighted by the sub function PII,i. Then considering the stressstrain<br />
information limited in the sub-domain ΩAE,I, ΩAE,III and ΩAE,IV, it was determined the<br />
progressive stiffness reduction of the material.<br />
In the first example the local structure of the function f(x) is used to interpret the stressstrain<br />
data in order to determine the material damage model. While in the second example
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 173<br />
the function f(x) is used to estimate the overall material damage when composite laminate are<br />
subjected to an out-of-plane load. Because of its synthesis capability the function f(x) can be<br />
used to summarize the whole material damage history and in the case of the transversal load<br />
tests the integral of the function f(x), Int(f), over the domain ΩAE was calculated:<br />
Int<br />
⎛ Es<br />
⎞<br />
f = ∫ Ln⎜ ⎟dΩ<br />
(6)<br />
⎝ Ea<br />
⎠<br />
( )<br />
Ω<br />
AE<br />
The values of Int(f) are influenced by the complete trend of f(x) and by the extension of<br />
the AE domain.<br />
Applications<br />
Case study 1: materials and method<br />
The composite laminates used for the tensile tests were different in terms of lay-up<br />
(unidirectional laminates (UD), angle-ply laminates (AP) and quasi-isotropic laminates (QI),<br />
fibre volume percentage and laminates thickness.<br />
The pre-pregs were made by T-300 graphite fibres and epoxy matrix. Specimens were<br />
cured in autoclave then cut by a diamond saw.<br />
The characteristics of these three types of laminates are reported in Table 1. For each type<br />
ten specimens were tested.<br />
Table 1. Types of laminates used for the tensile tests<br />
Laminate type ID Lay-up<br />
Fibre<br />
Volume<br />
(%)<br />
Thickness<br />
(mm)<br />
Unidirectional UD [0°]8 60 1.4<br />
Angle ply AP [±45°]4S 30 2.8<br />
Quasi isotropic<br />
QI1 [0°,90°, ±45°]4S 60 1.4<br />
QI2 [0°, ±45°, 90°]4S 60 1.4<br />
QI3 [0°,90°, ±45°]4S 30 2.8<br />
Specimen dimensions were 250 mm in length and 25 mm in width for all types of<br />
laminates as recommended by ASTM 3039M for AP and QI lay-ups and the gage length was<br />
140 mm, as shown in figure 4A.<br />
Uniaxial tensile tests were done under displacement control using an INSTRON 8032<br />
with a 100 kN load-cell, and speed of 0.05 mm/sec.<br />
In order to reduce the acquisition of spurious acoustic external signals [47] two noise<br />
gates were assembled in a series configuration with the specimens as shown in figure 4B.
174<br />
Giangiacomo Minak and Andrea Zucchelli<br />
During the test, the AE were monitored by a Physical Acoustic Corporation (PAC) PCI-<br />
DSP4 device with two transducers PAC R15 setting up the amplitude threshold at 40 dB.<br />
A<br />
B<br />
140 mm<br />
110 mm<br />
AE transducers<br />
Grips<br />
AE transducers<br />
Specimen<br />
Grips<br />
250 mm<br />
Figure 4. (A) specimen for tensile test and test set-up with AE transducers position, (B) the specimen<br />
fixing grips and external AE noise insulation devices.<br />
25 mm
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 175<br />
The AE transducers were placed in a linear configuration located at a distance of 110<br />
mm, as shown in figure 4A. The Elastic modulus was measured by means of HBM LY41-<br />
6/350 strain-gauges.<br />
Case study 1: results and discussion<br />
The experimental results analysis and discussion is here organized in two phases: the first<br />
one in which the only classical mechanical information (stress and strain) are considered, and<br />
the second one in which the AE information is analyzed and related to the material<br />
mechanical response.<br />
The stress-strain curves in Figure 5 show the effect of the fibre orientation and volume<br />
percentage and also the small, but not negligible, influence of the plies sequence in the<br />
laminates.<br />
In particular the AP diagram was qualitatively different from the other laminate ones<br />
because of the different failure mechanism that was fibre-dominated for UD and QI and<br />
matrix-dominated for AP.<br />
In fact, the mechanical response of AP laminates with respect to in plane tensile load was<br />
strongly nonlinear and the ultimate stress value was about 60 MPa. The two aspects that<br />
determined the behaviour of AP laminate were the fibre direction (±45°) and the low volume<br />
percentage of fibres. Visual inspection revealed also a marked necking due to the high<br />
percentage of matrix forming the composite and the sliding of fibres in it.<br />
Stress (MPa)<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
UD<br />
QI2<br />
QI1<br />
QI3<br />
AP<br />
0<br />
0.000 0.010 0.020 0.030 0.040 0.050 0.060 0.070 0.080 0.090<br />
Strain<br />
Figure 5. Examples of stress-strain diagram for the different types of laminates.<br />
From the stress-strain diagram it was possible to determine some of the most significant<br />
information regarding the mechanical response of laminates. In particular the elastic modulus<br />
(EX) and the ultimate strength were estimated considering all tested specimens. In table 2 are<br />
summarized the theoretical Elastic modulus, estimated by means of the lamination theory, and
176<br />
Giangiacomo Minak and Andrea Zucchelli<br />
the corresponding experimental values, while in table 3 the ultimate strain and stress values<br />
are reported.<br />
Table 2. Theoretical calculation and experimental measure of Elastic modulus of each<br />
laminate type<br />
Ex (MPa)<br />
Theoretical Experimental<br />
ID M.V. M.V. S.D.<br />
UD 197000 190000 16000<br />
AP 23000 22000 3000<br />
QI1 75000 74000 6000<br />
QI2 75000 76000 3000<br />
QI3 77000 78000 3000<br />
Table 3. Ultimate strain and stress values of tested laminates<br />
Ultimate Strain Ultimate Stress (MPa)<br />
ID M.V. S.D. M.V. S.D.<br />
UD 0.019 0.001 2157 104<br />
AP 0.084 0.008 62 7<br />
QI1 0.012 0.002 571 35<br />
QI2 0.010 0.001 600 44<br />
QI3 0.012 0.001 322 19<br />
The main acoustic parameters that have been considered are: counts per AE event and AE<br />
event energy (Ea). Both parameters have been related by means of double entry diagram with<br />
the stress-strain behaviour. In the following figures 6-10 examples of these diagrams, one for<br />
each type of laminate, are reported.
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 177<br />
Figure 6. UD laminate type, cumulative counts, AE energy, function f and stress versus displacement.
178<br />
Giangiacomo Minak and Andrea Zucchelli<br />
Figure 7. AP laminate type cumulative counts, AE energy, function f and stress versus displacement.
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 179<br />
Figure 8. QI1 laminate type, cumulative counts, AE energy, function f and stress versus displacement.
180<br />
Giangiacomo Minak and Andrea Zucchelli<br />
Figure 9. QI2 laminate type, cumulative counts, AE energy, function f and stress versus displacement.
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 181<br />
Figure 10. QI3 laminate type, cumulative counts, AE energy, function f and stress versus displacement.
182<br />
Giangiacomo Minak and Andrea Zucchelli<br />
From figures 6 to 10 it is possible to observe that the five types of laminates have<br />
different behaviours from the AE point of view. A preliminary observation can be done<br />
considering the AE domain that can be used to identify the Free Failure Domain (FFD), i.e.<br />
the strain domain over which no failures are detected. Considering the ratio between ΩAE and<br />
the strain at rupture it can be seen that the percentages of the FFD over the all strain domain<br />
are the following: 11% in the case of UD laminates, 0.5% in the case of AP laminates, 1% in<br />
the case of QI1 laminates, 30% in the case of QI2 laminates and 1.4% in the case of QI3<br />
laminates. The estimated percentage values of FFD indicate the different attitude to the<br />
damage onset of each type of laminate, and in particular it is interesting to note that the QI2<br />
laminate type is the one that has the greater capability to be strained without significant<br />
damage. On the contrary the AP laminate types are the most sensitive to the applied strain and<br />
reveal an early damage onset, probably due to the high matrix percentage content and to fibre<br />
orientation (±45°).<br />
Considering the diagram of AE event cumulative counts, figures 6A to 10A, it can be<br />
observed that only in the case of AP laminates the slope of the diagram is quite constant<br />
during the all test. For the other laminate types, UD and QI, the cumulative counts reveal an<br />
initial trend with low slope values that progressively increase during the test. Such behaviour<br />
can be interpreted considering the different damage attitude of the laminates and their<br />
structure. In the case of AP laminates the mechanical behaviour was dominated by the matrix<br />
deformation and cracking. The effect of fibres in AP laminates did not influence the material<br />
behaviour and, on the contrary, as observed during experiments at the early stage of tests,<br />
fibres promoted matrix breakage and spalling. Such interpretation of AP laminates behaviour<br />
is also supported by the AE energy diagram, figure 7B, where it is noticeable the presence of<br />
AE events with an energy content (the maximum AE event energy is about 4.0⋅10 -4 J) that is<br />
typical of composite laminate matrix failures [47]. Different behaviour was observed in the<br />
case of UD, QI1 and QI2 laminates. Considering, for example, diagrams of figures 6A, 8A<br />
and 9A, for the UD, QI1 and QI2 laminates respectively, it is possible to note the presence of<br />
a strain domain where the cumulative count rate is quite low. For these laminates, during the<br />
initial test stage no significant failures can be detected and considering also the energy<br />
diagrams, figures 6B, 8B and 9B it is possible to assume that the sources of AE event are<br />
mainly due to matrix cracks onset. Comparing in particular the cumulative counts and energy<br />
diagrams of QI1 and QI2 laminates it is interesting to note that in the case of QI2 the<br />
maximum number of cumulative counts (∼ 3⋅10 5 counts) is lower than the one of QI1<br />
laminate (∼ 2⋅10 6 counts), but, at the same time, the maximum AE event energy of QI2 (∼<br />
1.2⋅10 -3 J) is comparable to the one of QI1 laminate (∼ 2.2⋅10 -3 J). This behaviour can be<br />
understood considering the different delamination strength of the two laminates. In fact, as<br />
reported in [47, 48], delamination is a possible failure mechanism for laminates of type QI1,<br />
and, on the contrary, it is not a typical failure for laminates of type QI2. So in the case of QI1<br />
the maximum number of counts is greater than in the case of QI2 thanks to the contribution of<br />
events caused by inter-laminar fractures and delamination. Nevertheless the maximum AE<br />
energies for the QI1 and QI2 laminates are comparable because the final crisis of both<br />
materials is characterized by a fibre breaking process that determines the release of an AE<br />
event with an high energy content. The behaviour of QI3 laminate is quite different if<br />
compared to QI1 and QI2. In particular the cumulative counts trend, figure 10A, shows a<br />
consistent release of AE events at the early test stage, but the total number of cumulative
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 183<br />
counts over all test (∼ 1.4⋅10 4 counts) is lower than in the case of QI1 and QI2. Considering<br />
the AE energy diagram of QI3, figure 10B, it is possible to observe that events with a high<br />
energy content happens also in the early-middle stage of the test, differently than in the case<br />
of QI1 and QI2 where events with a high energy content appear only at the final test stage.<br />
All these facts are related to the QI3 ply composition: the matrix volumes of each ply<br />
influences the dominant material failure mode, the matrix cracking, as in the case of AP.<br />
All the previous considerations can be effectively summarized and completed considering<br />
the diagrams of the sentry functions f.<br />
In the case of UD laminate the construction of the function f reveals the initial material<br />
damage, figure 6C, that is highlighted by the sub-function PI,1 and PII,1. This initial damage<br />
can be related to some material internal adjustment and to the onset of some internal cracks<br />
but, as revealed by the following PI,2, such damages do not compromise the material<br />
capability of storing strain energy. The following PIV,1 sub-function reveals that the internal<br />
cracks propagate and that the material storing energy capability is progressively reduced until<br />
a sequence of drops due to splitting (PII,2, PII,3 and PII4) mixed with some strain energy storing<br />
phases (PI,3, PI,4, PI,5) that prelude to the first and most critical drop PII,5. This drop is mainly<br />
related to the free edge delamination and after this a new but small strain energy storing phase<br />
starts: PI,6. During this phase all the laminate plies work as springs in parallel and go on<br />
storing strain energy. After this phase next drops, PII,6 and PII,7, mixed with a slowly<br />
increasing part of f, PI,7, and two constant trends for f, PIII,1 and PIII,2, prelude the final crisis of<br />
the laminate.<br />
Considering the diagram in figure 7C of the AP laminate it can be observed that the<br />
structure of the sentry function is simpler if compared to the UD case. In fact only three subdomains<br />
of strain energy storing phases, PI,1, PI,2 and PII,6, and three drops, PI,1, PI,2 and PI,3,<br />
characterize the structure of the sentry function for the AP laminate reported in figure 10C.<br />
The smooth trend of the sub-function of type I reveal the modest capability of the material to<br />
store the strain energy and this fact is due to the high matrix volume percentage in each ply<br />
and to the fibre orientation.<br />
The initial trend of the sentry function of QI1 is characterized by a first material crisis,<br />
PIV,1, followed by a drop, PII,1. Such behaviour is due to an initial material adjustment and to<br />
some inter-laminar cracks onset that will contribute to delamination process. The next phase<br />
is characterized by two strain energy storing phases, PI,1 and PI,2, respectively followed by a<br />
sudden drop, PI,2, and a decreasing sub-function, PIV,2. In particular the sudden drop and the<br />
decreasing sub-function indicate the onset of the material crisis due to delamination and<br />
transversal cracks. After the sub-function PIV,2 the sentry function is characterized by a<br />
complex combination of sudden drops and constant sub-functions. This behaviour indicates<br />
that the material integrity is compromised and that each ply is progressively damaged until<br />
the final crisis. In this way it is interesting to notice that after the PIV,2 there are four sudden<br />
drops indicating the important crisis of each basic ply type (0°, +45°, -45°, 90°) that<br />
constitutes the original laminate QI1.<br />
In the case of QI2 the sentry function has a different trend with respect to the case of QI1.<br />
In fact at the test beginning damage is not appreciable and a sub-function of type I, PI,1,<br />
indicates a strain energy storing phase. After the first drop, PII,1, a consistent material damage<br />
indicates the reduced strain energy storing capability. Such material damage is mainly due to<br />
a global laminate loss of strength: the absence of delamination contributes to the cracks<br />
distribution and growth inside all the deformed material volume and this creates the condition
184<br />
Giangiacomo Minak and Andrea Zucchelli<br />
for a general crisis of the laminate. In fact the PII,1 is followed by a small series of strain<br />
energy storing phases, PI,2 and PI,3, and a combination of sub-function of type II, III and IV<br />
highlighting a great material damage.<br />
The QI3 is characterized by a sentry function that mixes the behaviour of the studied UD<br />
and AP laminates. At the test beginning is visible a combination of sub-functions of type I<br />
and of type II with a predominant strain energy storing phase. This phase is characterized by<br />
the sequence of the following sub-functions: PI,1, PII,1, PI,2, PII,2 and PI,3. A delamination<br />
failure is revealed by the slope change of f corresponding to PII,3. After this failure a reduced<br />
energy storing capability is revealed by the PI,4 and the following drop PII,4 is due to the<br />
failure of the weakest ply inside the laminate. This first ply crisis is followed by a subfunction<br />
of type I, PI,5, that has a smooth trend due to the previous material damage. The<br />
damage corresponding to the sub-function PII,6 is due to the crisis of one of the stronger plies.<br />
After this laminate crisis the energy storing phase represented by the PI,6 is due to the stresses<br />
redistribution between the undamaged plies that are now working as springs in parallel.<br />
The analysis here described has been used to perform a quantitative estimation of the<br />
laminate damage and in particular the sentry function of each laminate has been used to<br />
perform a discretization of the stress-strain curve. As an example of this analysis we report<br />
the case of UD laminate type.<br />
f<br />
40<br />
35<br />
30<br />
25<br />
20<br />
15<br />
10<br />
5<br />
f Stress-Strain<br />
Drops<br />
0<br />
0<br />
0.000 0.003 0.005 0.008 0.010<br />
Strain<br />
0.013 0.015 0.018 0.020<br />
Figure 11. Stress and f diagram versus strain, the most key drops of f are highlighted by means of gaps<br />
diagram.<br />
The analysis of the sentry function in figure 11 allows the identification of 10 drops on<br />
the basis of which the stress-strain curve was divided in order to calculate the Elastic<br />
modulus.<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
Stress (MPa)
Stress (MPa)<br />
Stress (MPa)<br />
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 185<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
2500<br />
2000<br />
1500<br />
1000<br />
Experimental<br />
Model<br />
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02<br />
Strain<br />
Figure 12. Stress-strain diagram and its discretization according to the key-drops.<br />
500<br />
Stress - Strain Model<br />
Young Modulus<br />
0<br />
0.0E+00<br />
0 0.0025 0.005 0.0075 0.01 0.0125 0.015 0.0175 0.02<br />
Strain<br />
2.0E+05<br />
1.8E+05<br />
1.6E+05<br />
1.4E+05<br />
1.2E+05<br />
1.0E+05<br />
8.0E+04<br />
6.0E+04<br />
4.0E+04<br />
2.0E+04<br />
Figure 13. Discretized stress-strain curve and Elastic modulus according to the key-drops analysis.<br />
By means of a linear regression the Elastic modulus of the stress-strain curve<br />
corresponding to each strain interval was estimated. In figure 12 the discretized stress-strain<br />
curve is reported and overlapped to the original diagram. In figure 13 the discretized stressstrain<br />
curve and Elastic modulus values are reported.<br />
Young modulus (MPa)
186<br />
Giangiacomo Minak and Andrea Zucchelli<br />
Table 4. Lower, upper and intermediate strain values used for the stress-strain curve<br />
discretization, Elastic modulus (Young) and Damage (D)<br />
UD<br />
AP<br />
QI1<br />
QI2<br />
QI3<br />
Low<br />
strain<br />
Up<br />
strain<br />
Mean Values<br />
Ref.<br />
Strain<br />
Young<br />
[Mpa]<br />
0.0000 0.0030 0.0015 191200 0.029<br />
0.0031 0.0097 0.0064 137500 0.302<br />
0.0104 0.0109 0.0106 105653 0.464<br />
0.0120 0.0154 0.0137 88522 0.551<br />
0.0154 0.0190 0.0172 31211 0.842<br />
0.0000 0.0006 0.0003 22865 0.006<br />
0.0012 0.0037 0.0025 6340 0.724<br />
0.0085 0.0137 0.0111 1666 0.928<br />
0.0252 0.0307 0.0280 329 0.986<br />
0.0308 0.0839 0.0573 182 0.992<br />
0.0000 0.0015 0.0008 74300 0.047<br />
0.0015 0.0031 0.0023 71600 0.082<br />
0.0031 0.0040 0.0035 62000 0.205<br />
0.0060 0.0072 0.0066 45175 0.421<br />
0.0081 0.0086 0.0083 32976 0.577<br />
0.0089 0.0115 0.0102 22151 0.716<br />
0.0000 0.0029 0.0015 72100 0.051<br />
0.0034 0.0041 0.0038 67050 0.118<br />
0.0044 0.0075 0.0060 59050 0.223<br />
0.0080 0.0087 0.0084 41022 0.460<br />
0.0087 0.0099 0.0093 28195 0.629<br />
0.0000 0.0003 0.0001 77608 0.005<br />
0.0003 0.0007 0.0005 53487 0.314<br />
0.0008 0.0016 0.0012 45048 0.422<br />
0.0016 0.0035 0.0025 27500 0.647<br />
0.0041 0.0048 0.0045 24973 0.680<br />
0.0050 0.0114 0.0082 20900 0.732<br />
0.0114 0.0120 0.0117 12997 0.833<br />
D
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 187<br />
It has to be noticed that building up the diagram in figure 13 the Elastic modulus values<br />
have been associated to the mean values of the strain range that have been used to discretize<br />
the stress-strain curve.<br />
Using the calculated values of the Elastic modulus it was also estimated the values of the<br />
damage by means of its conventional definition: D = 1- E(s)/E0, where E0 is the Young<br />
modulus of the undamaged material.<br />
In Table 4 are summarized the strain intervals, low-strain and up-strain, that have been<br />
used to discretize the stress-strain curves of all specimens for each type of laminates, the<br />
corresponding strain mean value to which the estimated Elastic modulus and Damage values<br />
are associated.<br />
Stress (MPa)<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
0<br />
0 0.0025 0.005 0.0075 0.01 0.0125 0.015 0.0175 0.02<br />
Strain<br />
Stress - Strain Model<br />
Damage<br />
Figure 14. Discretized stress-strain curve and Damage according to the key-drops analysis.<br />
The information summarized in Table 4 can be usefully implemented in FEA software to<br />
model the considered composite laminate progressive failure behaviour.<br />
In figure 15 are graphically represented the Elastic modulus and the Damage values<br />
plotted considering as strain values the mean values reported in Table 4<br />
1<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
Damage
188<br />
A<br />
B<br />
E (MPa)<br />
D<br />
250000<br />
200000<br />
150000<br />
100000<br />
50000<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
Giangiacomo Minak and Andrea Zucchelli<br />
0<br />
0.000 0.010 0.020 0.030 0.040 0.050 0.060 0.070 0.080 0.090<br />
Strain<br />
0.0<br />
0.000 0.010 0.020 0.030 0.040 0.050 0.060 0.070 0.080 0.090<br />
Strain<br />
E<br />
D() s = 1−<br />
E<br />
Figure 15. Trends of (A) Elastic modulus, E, and (B) Damage, D, versus strain for each type of<br />
laminate; for the damage calculation, of each laminate, E 0 is equal to the mean value of the Young<br />
modulus as reported in Table 2 that correspond to the Young modulus of the undamaged laminate.<br />
Details about the damage of each type of laminate are reported in figures 16 to 20.<br />
( s)<br />
0<br />
UD<br />
AP<br />
QI1<br />
QI2<br />
QI3<br />
UD<br />
AP<br />
QI1<br />
QI2<br />
QI3
D<br />
D<br />
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 189<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
0.000 0.002 0.004 0.006 0.008 0.010 0.012 0.014 0.016 0.018 0.020<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
Strain<br />
Figure 16. Damage for UD laminate types.<br />
0.0<br />
0.000 0.010 0.020 0.030 0.040 0.050 0.060 0.070<br />
Strain<br />
Figure 17. Damage for AP laminate types.<br />
UD<br />
AP
190<br />
D<br />
D<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
Giangiacomo Minak and Andrea Zucchelli<br />
0.0<br />
0.000 0.002 0.004 0.006 0.008 0.010 0.012<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
Strain<br />
Figure 18. Damage for QI1 laminate types.<br />
0.0<br />
0.000 0.002 0.004 0.006 0.008 0.010 0.012<br />
Strain<br />
Figure 19. Damage for QI2 laminate types.<br />
QI1<br />
QI2
D<br />
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 191<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
0.000 0.002 0.004 0.006 0.008 0.010 0.012 0.014<br />
Strain<br />
Figure 20. Damage for QI3 laminate types.<br />
Case study 2: materials and method<br />
Eighteen graphite/epoxy composite square laminate plates 250x250 mm 2 were studied.<br />
Their thickness was 1.6 mm. They have been made in autoclave from pre-pregs by stacking<br />
eight unidirectional plies with quasi-isotropic orientations [0,90,45,-45]s.<br />
The specimens were placed in a circular clamping fixture with an internal diameter of<br />
200 mm and they were loaded at the centre: an hemispherical hardened steel ball with a radius<br />
of 7 mm was indented in the top centre point of the laminate by means of a servo-hydraulic<br />
Instron 8033 testing machine controlled by an MTS Teststar II system and equipped by a<br />
25kN load cell.<br />
The specimens were loaded monotonically out-of-plane in control of displacement and<br />
the head speed was 0.05 mm/s.<br />
The tests were stopped at three different damage levels, one (Low) corresponding to the<br />
load value of 2 kN, the second (Medium) corresponding to the first load drop in the loaddisplacement<br />
curve and the third (High) to the complete perforation of the plate.<br />
During the test, the AE has been monitored by a Physical Acoustic Corporation (PAC)<br />
PCI-DSP4 device with four transducers PAC R15 setting up the amplitude threshold at 40 dB.<br />
In figure 21a it is possible to see the fixture system equipped with AE piezoelectric<br />
sensors and in figure 21b the complete experimental setup.<br />
After each quasi-static test, the damaged plate was sliced by a diamond saw to obtain<br />
tensile specimens with the same geometry suggested by ASTM D 5766 for open hole testing<br />
of CFRP, a width of 40 mm and a length of 250 mm. The indented zone was in the centre of<br />
these tensile specimens and the whole damaged zone was included in the specimen width.<br />
The damaged zones size was previously identified by means of the localization tool of the<br />
AE system as it is shown in figure 22 for the High damage level.<br />
QI3
192<br />
Giangiacomo Minak and Andrea Zucchelli<br />
Figure 21. (A) fixture system, (B)experimental device.<br />
AE sensors<br />
Cutting directions<br />
40 mm<br />
250 mm<br />
Detected AE<br />
sources<br />
External<br />
fibers direction<br />
Figure 22. Damaged zone area detected by AE emission and tensile specimens cutting directions for the<br />
High damage level.<br />
Some plates were also analyzed by C-Scan and MicroCT and in these cases the value of<br />
damaged area evaluated by AE was confirmed.<br />
Nine plates had the external ply fibres oriented in the direction of the specimen axis and<br />
other nine had the external ply fibres orthogonal to the specimen axis, so that two different
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 193<br />
stacking sequences were produced, respectively [0,90,45,-45]s and [90,0,45,-45]s , as shown<br />
in figure 22.<br />
Analogous tensile tests were run, for each stacking sequence, on nine undamaged<br />
specimens cut from undamaged zones of the same plates in order to get reference values.<br />
Tensile tests were performed by means of the same servo-hydraulic machine in<br />
displacement control with a head speed of 0.02 mm/s and a gauge length of 150 mm. Also in<br />
the case of tensile tests the AE have been monitored by a Physical Acoustic Corporation<br />
(PAC) PCI-DSP4 device with two transducers PAC R15, setting the amplitude threshold at<br />
40 dB.<br />
Case study 2: Results and Discussion<br />
Transversal load test<br />
In figure 23 are shown macro photos of the loaded side (a) and of the back side (b) of<br />
damaged plates for the three different damage levels.<br />
Figure 23. Damaged zones (inside the dotted circles) for the three damage levels (Low, Medium, High)<br />
on the loaded surface (a) and on the back surface (b).
194<br />
Giangiacomo Minak and Andrea Zucchelli<br />
In the picture of the low damage level lamina, both in the loaded side (L-a) and in the<br />
back side (L-b) the indentation is barely visible by the naked eye. The medium level damage<br />
laminas present a slightly larger mark on the loaded side (M-a) and some fibre breakage with<br />
matrix leakage on the back side (M-b). Finally the highly damaged lamina pictures (H-a) and<br />
(H-b) show fibre failure on both sides.<br />
Figure 24. Fibres failure on the loaded surface for the Low load level: fractures on fibres are evidenced<br />
by the arrows.<br />
Figure 25. Tensile fibre failure on the back surface for the Medium load level: fracture on one fibre is<br />
evidenced by the arrows.
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 195<br />
Investigating more deeply the loaded side of the low damage level laminate it was<br />
possible to find a number of broken fibres due to local shear [29], as it is shown in the SEM<br />
image of figure 24.<br />
A different failure mode for the fibre is shown in the SEM image of figure 25.<br />
3<br />
2.5<br />
2<br />
1.5<br />
1<br />
0.5<br />
Indentation: Maimum Load (kN)<br />
Load (kN)<br />
B<br />
3<br />
2.5<br />
2<br />
1.5<br />
1<br />
0.5<br />
0<br />
0 2 4 6 8 10 12 14<br />
0<br />
0 2 4 6 8 10 12 14<br />
M<br />
3<br />
2.5<br />
2<br />
1.5<br />
1<br />
0.5<br />
P<br />
0<br />
0 2 4 6 8 10 12 14<br />
Displacement (mm)<br />
Figure 26. Transversal load-displacement curves for the three damage levels.<br />
3.0<br />
2.5<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
Load<br />
Energy<br />
0.0<br />
0.0<br />
0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0<br />
Indentation: Maximum displacement (mm)<br />
30.0<br />
25.0<br />
20.0<br />
15.0<br />
10.0<br />
5.0<br />
Indentation: Total Strain Energy<br />
(J)<br />
Figure 27. Maximum displacement, maximum load and total strain energy of the transversal loading<br />
tests.
196<br />
Giangiacomo Minak and Andrea Zucchelli<br />
In this case, referred to the back side of a medium damage level lamina, matrix-fibre<br />
debonding is evident and the tensile fibre fracture surfaces are orthogonal to their axis and<br />
widely separated.<br />
Load-displacement curves for the three damage levels are reported in figure 26 while the<br />
maximum load, maximum displacements and total strain energy recorded in each test can be<br />
found in figure 27.<br />
Acoustic emission analysis of the transversal load test<br />
A parametric AE analysis was performed considering the acoustic energy, the cumulative<br />
events and the cumulative counts per event. As an example, in figure 28 these three AE<br />
parameters are plotted together with the respective load-displacement diagrams.<br />
For the AE diagrams reported in figure 28, the ΩAE is defined in the displacement range<br />
[0.7; 11.1] mm. It is interesting to notice the presence of the FFD that represents a phase of<br />
the test during which no damages are induced in the laminate (i.e. [0; 0.7]mm). The diagrams<br />
of cumulative events and cumulative counts versus the displacement are characterized by a<br />
general monotonic increasing trend, a quite similar shape and, for each diagram, at the same<br />
displacement value there are significant slope variations. In particular from figure 28A and<br />
figure 28B, inside the ΩAE, it is possible to notice a first part of the cumulative events and<br />
counts diagram characterized by low values and a smooth trend and dominated by AE events<br />
with a low number of counts and a low AE event rate. In figure 28 this first test phase has<br />
been identified in an AE sub-domain marked as Z1 limited, for the considered example, in the<br />
displacement range of [0.7; 6.3] mm. Considering also the energy diagram, figure 28C, it is<br />
possible to observe that only one event in Z1, at the displacement value of 3.6 mm, has an<br />
appreciable acoustic energy (over 3.0 10 -6 J) and from the statistical analysis it was observed<br />
that only 5 events have an energy over 1.0 10 -6 J, that can be considered typical for fibres<br />
breakage in bending. The sub-domain Z1 is physically dominated by material adjustment<br />
(especially fibre alignment) and by matrix deformation and matrix crack onset that are<br />
typically related to AE events with a small number of counts and low energy content [47].<br />
The presence of few AE events with a high value of energy (over 1.0 10 -6 J) can be physically<br />
related to the breakage of some fibres as was previously noticed by the SEM image (figure<br />
23) even if probably the energy content of these events in most cases should be low since<br />
these fibres are not loaded in tension. So during this first test phase (Z1) no significant<br />
damage is induced to the material, as it was noticed during the visual inspection of laminate<br />
surfaces, a small hemispherical mark in the matrix is appreciable (figure 23 L-a & L-b), and<br />
as pointed out by means of the SEM image, figure 24, a certain amount of fibres are broken.<br />
After this first phase there is a considerable increase of the AE activity represented by<br />
increased values of the slope in both cumulative event and counts diagrams. The test phase<br />
characterized by a high AE activity presents two other slope variations that can be used to<br />
define three other sub-domains: Z2 over the displacement range [6.3; 7.2] mm, Z3 over the<br />
displacement range [7.2; 8.1] mm and the Z4 over the displacement range [8.1; 11.1] mm. It is<br />
important to point out, as initially noticed, that all the slope variations in the cumulative event<br />
and counts diagram correspond to the same indenter displacement values. This is mainly<br />
related to the fact that, generally, the damage growth inside the material causes an increase of<br />
the total AE activity (events with an increased rating and an increased number of counts per
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 197<br />
Event CUM<br />
Count CUM<br />
Eac (J)<br />
1.0E+04<br />
9.0E+03<br />
8.0E+03<br />
7.0E+03<br />
6.0E+03<br />
5.0E+03<br />
4.0E+03<br />
3.0E+03<br />
2.0E+03<br />
1.0E+03<br />
0.0E+00<br />
3.0E+05<br />
2.5E+05<br />
2.0E+05<br />
1.5E+05<br />
1.0E+05<br />
5.0E+04<br />
Event CUM<br />
Load<br />
0 2 4 6 8 10 12 14<br />
4.5E+05<br />
Displacement (mm)<br />
3.0<br />
4.0E+05<br />
3.5E+05<br />
Count CUM<br />
Load<br />
B<br />
2.5<br />
6.0E-04<br />
5.0E-04<br />
4.0E-04<br />
3.0E-04<br />
2.0E-04<br />
1.0E-04<br />
0.0E+00<br />
Z 1 Z 2 Z 3 Z 4<br />
0.0E+00<br />
0.0<br />
9.0E-04<br />
8.0E-04<br />
7.0E-04<br />
0 2 4 6 8 10<br />
Eac Displacement (mm)<br />
Load<br />
12<br />
C<br />
14<br />
3.0<br />
2.5<br />
Ω AE<br />
0 2 4 6 8 10 12 14<br />
Displacement (mm)<br />
Figure 28. Main AE parameter and load diagram versus displacement; (A) cumulate of AE events, (B)<br />
cumulate of AE counts per event, (C) AE energy of each event.<br />
A<br />
3.0<br />
2.5<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
0.0<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
0.0<br />
Load (kN)<br />
Load (kN)<br />
Load (kN)
198<br />
Giangiacomo Minak and Andrea Zucchelli<br />
event [47]). The slope variations in all diagrams are associated to events with a high energy<br />
content. In fact, as shown in figure 28C, the transitions between Z1 and Z2 and then between<br />
Z2 and Z3 are determined by two events with an energy content higher then 1.0 10 -6 J<br />
probably caused by fibres breakage. It is also interesting to notice that in the sub-domain Z2<br />
there is a considerable number of events with a high energy content (22 events with an energy<br />
higher then 1.0 10 -6 J) probably related to the breakage of quite large number of fibres. This<br />
fact was also confirmed by the visual and SEM analysis (figure 23 M-a & M-b, figure 25).<br />
Similar considerations can be developed considering the sub-domains Z3 and Z4 with a<br />
number of 27 and 100 events with energy content higher then 1.0 10 -6 J respectively, that<br />
probably reveal the progressive fibres breaking process.<br />
Besides the analysis of the AE information it is interesting to relate the AE diagrams to<br />
the mechanical response of the laminate.<br />
In particular it is possible to notice that both the first zone Z1 and the second zone Z2 end<br />
at the two important load drops. In the sub-domain Z1 the load-displacement diagram has a<br />
monotonic increasing trend with an increasing slope and this is a direct consequence of the<br />
fact that the system stiffness is increased by the transition from the bending to the membrane<br />
behaviour [35]. So this sub-domain characterizes the test phase during which no important<br />
damage is induced and the main part of the mechanical energy is stored in the material as<br />
strain energy, in fact only a small part of the mechanical energy is dissipated by fibres<br />
adjustment or alignments and matrix crack onset.<br />
After the first load drop, the sub-domain Z2 begins, where the load displacement diagram<br />
is again increasing monotonically, but contrary to what is observed in Z1 the slope is<br />
decreasing. This is related to the material damage corresponding to the first load drop. In fact,<br />
as noticed by visual inspection and SEM analysis, after the first load drop the fibres breakage<br />
and the brittle matrix leakage reduce the local resistance of the laminate. So in the subdomain<br />
Z2 the strain energy storing capability of the laminate is reduced if compared with the<br />
laminate behaviour in Z1.<br />
In the sub-domain Z3 the load-displacement diagram is characterized by a monotonic<br />
increasing trend with a consistent decreasing of the slope. After the second load drop<br />
delamination and fibre breakages compromise the local out-of-plane resistance of the<br />
laminate and the energy storing capability is significantly reduced. The third zone ends when<br />
the load reaches a relative maximum value and then it decreases. The sub-domain Z4 is<br />
characterized by a slowly decreasing trend of the load with the total penetration of the<br />
indenter in the laminate and the AE event are mainly caused by the delamination, matrix<br />
cracking and leakage, fibre breaking and bending-pull-out. At the end of the loaddisplacement<br />
diagram there is a new increasing trend due to the contact of the support of the<br />
spherical indenter with the laminate surface.<br />
The physical evidence of the failure modes that happens during the loading history in Z4<br />
can be reconstructed by visual and SEM inspection as shown in figure 23 (H-a& H-b) and<br />
figure 24.<br />
An example of implementation of the function f(x) for this case study is shown in figure<br />
29 where the strain energy (Es), the cumulative AE event energy (Ea), the load and the f(x)<br />
diagrams relative to the same test reported in figure 28 are shown.
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 199<br />
Es (J)<br />
Ea Cumulate (J)<br />
f<br />
1.8E+01<br />
1.6E+01<br />
1.4E+01<br />
1.2E+01<br />
1.0E+01<br />
8.0E+00<br />
6.0E+00<br />
4.0E+00<br />
2.0E+00<br />
0.0E+00<br />
Es<br />
Load<br />
2.5E-03<br />
0 2 4 6 8 10 12 14<br />
3.0<br />
Displacement (mm)<br />
2.0E-03<br />
Eac CUM<br />
Load<br />
B<br />
2.5<br />
1.5E-03<br />
1.0E-03<br />
5.0E-04<br />
0.0E+00<br />
2.5E+01<br />
2.0E+01<br />
1.5E+01<br />
1.0E+01<br />
5.0E+00<br />
0.0E+00<br />
Z1<br />
0 2 4 6 8 10 12 14<br />
PII,1 PII,2<br />
PI,1 PI,2 PI,3<br />
ΩAE<br />
Displacement (mm)<br />
PIII,1<br />
0 2 4 6 8 10 12 14<br />
Displacement (mm)<br />
f<br />
Load<br />
Figure 29. (A) f general behaviour, (B) example of f diagram for an experimental indentation test.<br />
In figure 29A the sub-domain Z1,Z2, Z3 and Z4, as previously analysed and in figure 29B<br />
the AE domain, ΩAE, are reported. The strain energy diagram, figure 29A, is continuous,<br />
Z2 Z3<br />
PIV,1<br />
PII,3<br />
PII,4<br />
PIII,2<br />
Z4<br />
A<br />
C<br />
3.0<br />
2.5<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
0.0<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
0.0<br />
3.0<br />
2.5<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
0.0<br />
Load (kN)<br />
Load (kN)<br />
Load (kN)
200<br />
Giangiacomo Minak and Andrea Zucchelli<br />
monotonic with an increasing trend and it is characterized by four main parts: the first part<br />
has an exponential trend, the other parts have nearly linear trends with different slopes. The<br />
slope variations coincide to the sub-domain Z2, Z3 and Z4 limits. The cumulative AE event<br />
energy is a discrete function that monotonically increases. The sub-domain previously cited<br />
delimits some of the Ea diagram variations: the first part of Ea is characterized by low values<br />
and a smooth trend (inside Z1), the second domain, Z2, is characterized by an Ea increasing<br />
trend and slope and at the end of Z2 there is an evident gap in the Ea value. This gap, as it is<br />
clear in figure 28C, is due to an AE event with a high energy content and from the mechanical<br />
point of view is related to the second load drop: the material damage has reached a high level.<br />
In the sub-domain Z3 and Z4 the Ea diagram is characterized by an increasing trend but no<br />
particular behaviour can be noticed. On the contrary, it is interesting to notice that in the Z2<br />
domain the Ea trend is increasing with a general slope greater than the Ea slope in Z3 and Z4.<br />
In particular comparing the diagrams in figure 28 and the Ea diagram in figure 29B it is<br />
evident that the AE activity in Z2 is characterized by an increased AE event rate than in Z1<br />
and in Z2 the events have a mean number of counts and a mean energy content higher than the<br />
ones in Z1. This analysis of the AE information indicates that the test phase coinciding to the<br />
sub-domain Z2 is the prelude to the main material crisis.<br />
Considering the f(x) diagram it is possible to notice the presence of three increasing<br />
diagram parts (type I: PI,1; PI,2; PI,3), four falls (diagram of type II: PII,1; PII,2; PII,3; PII,4), two<br />
constant parts (PIII,1; PIII,2) and one decreasing part (PIV,1). The three increasing diagram parts<br />
of f(x) are limited in the sub-domain Z1 indicating that at this first test stage the material has a<br />
moderate attitude in storing the mechanical energy. Analysing the f(x) diagram in the subdomain<br />
Z1 it is possible to note that the first fall PII,1 that connect PI,1 and PI,2 is not related to<br />
an important material failure: there is no slope variation of f(x) before and after the fall PII,1<br />
(S1-=S1+). On the contrary, considering the second fall (PII,2) it is possible to note that the final<br />
slope of PI,2 and the starting slope of PI,3 are different (S2->s2+). This fact is due to the first<br />
important material damage and, as noticed during 28C analysis, it is related to the fibres<br />
breakage. It is worth noting that that the simple analysis of the load-displacement diagram<br />
does not single out this first damage even though it is important because it definitely<br />
influences the material strain energy storing capability and indicates the displacement value at<br />
which the damage significantly begins. The other three sub-domains are characterized by a<br />
general decreasing and constant trends of f(x). In particular in Z2, after the third fall (PII,3), the<br />
f(x) is characterized by an initial constant trend directly followed by a decreasing trend (no<br />
falls connect PIII,1 and PIV,1). This behaviour can be explained by the fact that the cumulated<br />
damage during the test phase in sub-domain Z1 and the first fall is great enough to<br />
compromise the material strain energy storing capability. The fourth fall, PII,4, that follows the<br />
decreasing trend of f(x) (PIV,1) is due to the final material local degradation. It is interesting to<br />
notice that in sub-domains Z3 and Z4 the f(x) is characterized by a constant trend while the<br />
load-displacement diagram reveals the presence of a local load maximum value. So even if<br />
the load diagram indicates a residual stiffness of the laminate the f(x) clearly indicates that the<br />
material damage is definitive. The constant behaviour of f(x) indicates that despite the local<br />
load maximum the material energy release is continuous and great enough to compensate the<br />
material strain energy storing attitude: the mechanical energy propagates the material damage.
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 201<br />
3.3. Residual Tensile Strength<br />
From the experimental results it was verified that the RTS of the laminates with [0,90,+45,-<br />
45]s lay-up was greater than the [90,0,+45,-45]s one for each damage level. In [30] this result<br />
is put into relation with the greater tensile fibre damage of the ply adjacent to the most<br />
external one on the back side.<br />
In figure 30 the tensile tests results are summarized in terms of RTS and the<br />
correspondent values of the displacement.<br />
RTS tests showed a reduction respect to undamaged specimens [35] even for the barely<br />
visible indentations corresponding with Low damage levels. This is explained by the shear<br />
fibre breakage shown in figure 24. This fibre failure mode, different from the tensile one, is<br />
characterized by low strain energy level and, consequently, low acoustic energy emission.<br />
Nevertheless the study of the function f can be useful to identify this kind of failure and an<br />
example of this analysis was previously cited and it is reported in figure 29 B. In particular<br />
from the diagram in figure 29B the important drop of f at a displacement of 3.5 takes into<br />
account a low value of strain energy stored in the laminate (Es = 1.3 J) and an AE event with<br />
low energy content (Ea = 3.810-6 J). In order to take into account the material damage it is<br />
necessary to evaluate all events that cause loss of structural integrity. Since the function f<br />
amplifies the most important material damage events and it is able consider at the same time<br />
the strain energy storing capability and the released internal energy, its integral was utilized<br />
as a damage indicator. In figure 31 the RTS data are plotted versus the respective values of<br />
Int(f) for each laminate type. In particular it is evident the negative relation between the RTS<br />
and the values of the f integrate, confirming, as presented by other authors using different<br />
damage indicators [49-52], that the variable Int(f) is a reliable instrument to evaluate the<br />
material damage during the indentation process.<br />
To represent mathematically the relations between RTS and the damage indicator many<br />
different approaches are utilized: discontinuous relations are composed by linear[49] or non<br />
linear [50] equations and they present a threshold at the damage indicator, so values of<br />
damage indicator lower to a specified threshold value do not change the RTS that is so equal<br />
to the virgin material tensile strength; on the contrary continuous relations[51, 52] have a<br />
plateau which value is equal to the virgin material tensile strength when the damage indicator<br />
in zero and they have a curvature inversion.<br />
In the present work in order to relate the Int(f) and the RTS a continuous relation was<br />
considered having the following form:<br />
C<br />
BInt ( ( f)<br />
)<br />
RTS Ae −<br />
= (2)<br />
Where the constant A is related to the ultimate load of the virgin material, and the<br />
constant B and C can be obtained by means of a linear regression based on the experimental<br />
data. Implementing the model in (2) to the experimental data it was estimated the following<br />
values for the coefficient of the continuous model:<br />
- A = 39 kN<br />
- laminate configuration [0,90,+45,-45]s: B = 8.8 10 -5 ; C = 2.0 (mm -1 );<br />
- laminate configuration [90,0,+45,-45]s: B = 8.3 10 -7 ; C = 2.8 (mm -1 );
202<br />
Giangiacomo Minak and Andrea Zucchelli<br />
In figure 32 the mathematical continuous models implementing the previous coefficient<br />
are represented by means of the continuous line showing a good fit.<br />
Maximum Displacement (mm)<br />
Maximum Event Cum<br />
14.0<br />
12.0<br />
10.0<br />
8.0<br />
6.0<br />
4.0<br />
2.0<br />
0.0<br />
1.8E+04<br />
1.6E+04<br />
1.4E+04<br />
1.2E+04<br />
1.0E+04<br />
8.0E+03<br />
6.0E+03<br />
4.0E+03<br />
2.0E+03<br />
A<br />
0.0 25.0 50.0 75.0 100.0 125.0 150.0<br />
Int(f) (mm)<br />
A<br />
MAX Event<br />
Max Count<br />
0.0E+00<br />
0.0E+00<br />
0.0 25.0 50.0 75.0 100.0 125.0 150.0<br />
Int(f) (mm)<br />
MAX Displacement<br />
Max Load<br />
3.0<br />
2.5<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
0.0<br />
1.2E+06<br />
1.0E+06<br />
8.0E+05<br />
6.0E+05<br />
4.0E+05<br />
2.0E+05<br />
Figure 30. Scatter diagrams of Int(f) and the main mechanical variables (A) and AE parameters (B).<br />
Maximum Load (kN)<br />
Maximum Count Cum
Tensile Test: Load at rupture (kN)<br />
Tensile Residual Strength (kN)<br />
40<br />
35<br />
30<br />
25<br />
20<br />
15<br />
10<br />
5<br />
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 203<br />
0<br />
0 1 2 3 4 5 6<br />
Tensile Test: Displacement at rupture (mm)<br />
[0,90,+-45]<br />
[90,0,+-45]<br />
Figure 31. Ultimate load and ultimate displacement from the residual strength tensile tests.<br />
45.0<br />
40.0<br />
35.0<br />
30.0<br />
25.0<br />
20.0<br />
15.0<br />
10.0<br />
5.0<br />
[0,90,+-45]<br />
[90,0,+-45]<br />
0.0<br />
0.0 20.0 40.0 60.0 80.0 100.0 120.0 140.0 160.0<br />
Int(f ) (mm)<br />
Figure 32. Tensile residual strength versus the integrate of the function f and the representation of the<br />
respective correlation model for the two damaged laminate configurations [0°,90°.+45°,-45°]s and<br />
[90°,0°.+45°,-45°] s
204<br />
Conclusion<br />
Giangiacomo Minak and Andrea Zucchelli<br />
In this chapter, a new approach to the evaluation of damage progression and of the residual<br />
strength of CFRP was presented.<br />
This approach is based on standard parametric AE and in particular on the acoustic<br />
energy.<br />
A function of the acoustic energy and of the strain energy, called Sentry function, was<br />
introduced and its application was illustrated in the case of:<br />
1) damage progression in tensile testing of different types of CFRP laminates;<br />
2) damage progression and residual strength evaluation in the case of CFRP plates loaded<br />
at the centre.<br />
In the first case, the Sentry function allowed us to single out important material failures<br />
and to calculate the corresponding damage values, while in the second case, after the damage<br />
identification phase, the residual tensile strength was related to the integral of the Sentry<br />
function over the acoustic domain defined in the transversal load test.<br />
References<br />
[1] Slight DW, <strong>Progress</strong>ive Failure Analysis Methodology for Laminated <strong>Composite</strong><br />
Structures, NASA/TP-1999-209107, 1999.<br />
[2] Basu S, Wass AM, Ambur AR, Prediction of progressive failure in multidirectional<br />
composite laminated panels, International Journal of Solids and Structures, 44 (2007)<br />
2648-2676<br />
[3] Lapczyk I, Hurtado JA, <strong>Progress</strong>ive damage modeling in fibre-reinforced materials,<br />
<strong>Composite</strong>s Part A, 38 (2007) 2333-2341<br />
[4] Abry JC, Bochard S, Chateauminois A, Salvia M, Giraud G, In situ detection of damage<br />
in CFRP laminates by electrical resistance measurements, <strong>Composite</strong>s Science and<br />
Technology, 59 (1999) 925-935<br />
[5] Tsuda H, Lee JR, Strain and damage monitoring of CFRP in impact loading using a<br />
fibre Bragg grating sensor system, <strong>Composite</strong>s Science and Technology, 67 (2007) 1353-<br />
1361<br />
[6] Deuschle HM, Wittel FK, Gerard H, Busse G, Kroplin BH, Investigation of progressive<br />
failure in composites by combined simulated and experimental photoelasticity,<br />
Computational Material Science, 38 (2006) 1-8<br />
[7] Benmedakhene S, Kenane M, Benzeggagh ML, Initiation and growth of delamination in<br />
glass/epoxy composites subjected to static and dynamic loading by acoustic emission<br />
monitoring, <strong>Composite</strong>s Science and technology, 59 (1999) 201-208<br />
[8] Bourchak M, Farrow IR, Bond IP, Rowland CW, Menan F, Acoustic Emission energy<br />
as a fatigue damage parameter for CFRP composites, International Journal of Fatigue,<br />
29 (2007) 458-470<br />
[9] Loutas TH, Kostopulos V, Ramirez-Jimenez C, Pharaoh M, Damage evolution in<br />
center-holed glass/polyester composites under quasi static loading using time-frequency
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 205<br />
analysis of acoustic emission monitored waveforms, <strong>Composite</strong>s Science and<br />
Technology, 66 (2006) 1366-1375<br />
[10] Hosur MV, Murty CRL, Ramamurthy TS, Shet A, Estimation of impact-induced<br />
damage of CFRP laminates through ultrasonic imaging, NDT&E International, 31(5),<br />
(1998) 359-374<br />
[11] Abrate S, Impact on laminated composite materials. Applied Mechanics Review, 44(4),<br />
(1991) 155–90.<br />
[12] Abrate S, Impact on laminated composites: recent advances. Applied Mechanics<br />
Reviews , 47(11), (1994) 517–44.<br />
[13] Abrate S, Impact on composite structures. Cambridge: Cambridge University Press,<br />
(1998).<br />
[14] Cantwell WJ, Morton J, The impact resistance of composite materials a review,<br />
<strong>Composite</strong>s, 22(5), (1991) 347-362<br />
[15] Caprino G, Langella A, Lopresto V, Prediction of first failure energy of circular carbon<br />
fibre reinforced plastic plates loaded at the centre, <strong>Composite</strong>s Part A, 34 (2003)<br />
349-357<br />
[16] Caprino G, Langella A, Lopresto V, Elastic behaviour of circular composite plates<br />
transversely loaded at the centre, <strong>Composite</strong>s Part A , 33 (2002) 1191–1197<br />
[17] Found MS, Holden GJ, Swamy RN. Static indentation and impact behaviour of GRP<br />
pultruded sections, <strong>Composite</strong> Structures, 39 (1997) 223-228<br />
[18] Lee SM, Zahuta P, Instrumented impact and static indentation of composites, Journal of<br />
<strong>Composite</strong> <strong>Materials</strong>, 25(2), (1991) 204–22<br />
[19] Kwon YS, Sankar BV, Indentation-flexure and low-velocity impact damage in graphite<br />
epoxy laminates, Journal of <strong>Composite</strong>s Technology and <strong>Research</strong>, 15(2), (1993)<br />
101–110<br />
[20] Wardle BL, Lagace PA, On the use of quasi-static testing to assess impact damage<br />
resistance of composite shell structures, Mechanics of <strong>Composite</strong> <strong>Materials</strong> and<br />
Structures, 5(1), (1998) 103–121<br />
[21] Sjöblom PO, Hartness TM, Cordell TM, On low-velocity impact testing of composite<br />
materials, Journal of <strong>Composite</strong> <strong>Materials</strong>, 22(1),(1988) 30–52.<br />
[22] Lagace PA, Williamson JE, Tsang PHW, Wolf E, Thomas S, A preliminary proposition<br />
for a test method to measure impact damage resistance, Journal of Reinforced Plastics<br />
and <strong>Composite</strong>s, 12(5), (1993) 584–601<br />
[23] Symons DD, Characterisation of indentation damage in 0/90 lay-up T300/914 CFRP,<br />
<strong>Composite</strong>s Science and Technology 60, (2000) 391-401<br />
[24] Alderson KL, Evans KE. Failure mechanisms during the transverse loading of filamentwound<br />
pipes under static and low velocity impact conditions, <strong>Composite</strong>s, 23(3), (1992)<br />
167–73<br />
[25] Hirai Y, Hamada H, Kim JK. Impact response of woven glass-fabric composites. I.<br />
Effect of fibre surface treatment, <strong>Composite</strong>s Science and Technology, 58(1), (1998) 91-<br />
105<br />
[26] Matemilola SA, Stronge WJ. Low speed impact damage in filament wound CFRP<br />
composite pressure vessels, Journal of Pressure Vessel Technology;119(4), (1997)<br />
435–43<br />
[27] Schoeppner GA, Abrate S, Delamination threshold loads for low velocity impact on<br />
composite laminates, <strong>Composite</strong>s Part A, 31(9), (2000) 903-915
206<br />
Giangiacomo Minak and Andrea Zucchelli<br />
[28] Found MS, Lamb JR, Damage assessment of impacted thin CFRP panels, <strong>Composite</strong>:<br />
Part A, 35(9), (2004) 1039–1047<br />
[29] Luo RK, Green ER, Morrison CJ, Impact damage analysis of composite plates,<br />
International Journal of Impact Engineering, 22 , (1999) 435-447<br />
[30] Caprino G, Teti R. Residual strength evaluation of impacted GRP laminates with<br />
acoustic emission monitoring, <strong>Composite</strong>s Science and Technology, 53, (1995) 13-19<br />
[31] Cartié DDR, Irving PE. Effect of Resin and Fibre Properties of Impact and Compression<br />
after Impact Performances of CFRP, <strong>Composite</strong>s Part A, 33 (4), (2002) 483-493<br />
[32] Zhang X, Davies GAO, Hitchings D, Impact damage with compressive preload and<br />
post-impact compression of carbon composite plates, International Journal of Impact<br />
Engineering, 22 , (1999) 485-509<br />
[33] Caprino G, Residual strength prediction of impacted CFRP laminates, Journal of<br />
<strong>Composite</strong> <strong>Materials</strong>,18, (1984) 508-518.<br />
[34] Caprino G, Teti R, de Iorio I, Predicting residual strength of pre-fatigued glass fibrereinforced<br />
plastic laminates through acoustic emission monitoring, <strong>Composite</strong>s Part B,<br />
36 (2005) 365-371<br />
[35] Cesari F, Dal Re V, Minak G, Zucchelli A, Damage and residual strength of laminated<br />
graphite-epoxy composite circular plates loaded at the centre, <strong>Composite</strong>s Part A, 38<br />
(2007), 1163-1173<br />
[36] Symons DD, Davis G, Fatigue testing of impact-damage T300/914 carbon-fibrereinforced<br />
plastic, <strong>Composite</strong> Science and Technology, 60(3), 2000, 379-389<br />
[37] Minak G, Morelli P, Zucchelli A, Fatigue residual strength of laminated graphite-epoxy<br />
circular plates damaged by transversal load, Proceedings of the 12 th European<br />
Conference on <strong>Composite</strong> <strong>Materials</strong>, Biarritz, (2006)<br />
[38] Dunegan HL, Harris DO, Tatro CA, Fracture analysis by use of acoustic emission<br />
Engineering Fracture Mechanics, 1 (1), (1968) 105-110<br />
[39] Shippen NC, Adams DF, Acoustic Emission monitoring of damage progression in<br />
graphite/epoxy laminates, Journal of Reinforced Plastics and <strong>Composite</strong>s, 4(1985)<br />
242-261<br />
[40] Siron O, Tsuda H, Acoustic Emission in Carbon Fibre-Reinforced Plastics <strong>Materials</strong>,<br />
Annales De Chimie et Science des Matériaux, 25, 7, (2000) 533-537,<br />
[41] Mizutani Y, Nagashima K, Takemoto M, Ono K, Fracture mechanism characterization<br />
of cross-ply carbon-fibre composites using acoustic emission analysis, NDT&E<br />
International, 33 (2000) 101-110<br />
[42] Walker JL, Hill E, Workman GL, Russell SS, A Neural Network-Acoustic Emission<br />
Analysis of Impact Damaged Graphite-Epoxy Pressure Vessels, Proceedings of the<br />
ASNT 1995 Spring Conference, Las Vegas (1995) 106–108.<br />
[43] Prosser WH, Waveform Analysis of AE from <strong>Composite</strong>s. Proceedings of the Sixth<br />
International Symposium on Acoustic Emission from <strong>Composite</strong> <strong>Materials</strong> (AECM-6),<br />
San Antonio (1998) 61–70.<br />
[44] Prosser WH, Jackson KE, Kellas S, Smith BT, McKeon J, Friedman A, Advanced,<br />
Waveform Based Acoustic Emission Detection of Matrix Cracking in <strong>Composite</strong>s.<br />
<strong>Materials</strong> Evaluation 53(9),(1995) 1052–1058<br />
[45] Woo SC, Choi NS, Analysis of fracture process in single-edge-notched laminated<br />
composites based on the high amplitude acoustic emission events, <strong>Composite</strong>s Science<br />
and Technology, 67 (2007) 1451-1458
Damage Evaluation and Residual Strength Prediction of CFRP Laminates … 207<br />
[46] Shull PJ, Non destructive Evaluation: Theory, Techniques, and Applications, Marcel<br />
Dekker Inc., 2002.<br />
[47] Zucchelli A, Dal Re V, Experimental analysis of composite laminate progressive failure<br />
by AE monitoring, Proceedings ICEM12 - 12th International Conference on<br />
Experimental Mechanics, Bari (2004)<br />
[48] MIL-HDBK-17-3E; Department of Defence Handbook, Polymer Matrix <strong>Composite</strong>s,<br />
volume 3. <strong>Materials</strong> Usage, Design, and Analysis.<br />
[49] Shim VPW, Yang LM, Characterization of the residual mechanical properties of woven<br />
fabric reinforced composites after low-velocity impact, International Journal of<br />
Mechanical Sciences, 47, 2005, 647-665<br />
[50] Caprino G, Lopresto V, The significance of indentation in the inspection of carbon<br />
fibre-reinforced plastic panels damaged by low velocity impact, <strong>Composite</strong>s Science and<br />
Technology, 60, (2000) 1003-1012<br />
[51] Sanchez-Saez S, Barbero E, Zaera R, Navarro E., Compression after impact of thin<br />
composite laminates, <strong>Composite</strong>s Science and Technology, 65, (2005) 1911-1919<br />
[52] Qi B, Herszberg I, An engineering approach for predicting residual strength of<br />
carbon/epoxy laminates after impact and hygro-thermal cycling, <strong>Composite</strong> Structures,<br />
47, (1999) 483-490
In: <strong>Composite</strong> <strong>Materials</strong> <strong>Research</strong> <strong>Progress</strong> ISBN: 1-60021-994-2<br />
Editor: Lucas P. Durand, pp. 209-236 © 2008 Nova Science Publishers, Inc.<br />
Chapter 6<br />
RESEARCH DIRECTIONS IN THE FATIGUE TESTING<br />
OF POLYMER COMPOSITES<br />
W. Van Paepegem * , I. De Baere, E. Lamkanfi,<br />
G. Luyckx and J. Degrieck<br />
Dept. of Mechanical Construction and Production, Sint-Pietersnieuwstraat 41,<br />
9000 Gent, Belgium<br />
Abstract<br />
For a long time, fatigue testing of composites was only focused on providing the S-N<br />
fatigue life data. No efforts were made to gather additional data from the same test by using<br />
more advanced instrumentation methods. The development of methods such as digital image<br />
correlation (strain mapping) and optical fibre sensing allows for much better instrumentation,<br />
combined with traditional equipment such as extensometers, thermocouples and resistance<br />
measurement. In addition, validation with finite element simulations of the realistic boundary<br />
conditions and loading conditions in the experimental set-up must maximize the generated<br />
data from one single fatigue test.<br />
This research paper presents a survey of the authors’ recent research activities on fatigue<br />
in polymer composites. For almost ten years now, combined fatigue testing and modelling has<br />
been done on glass and carbon polymer composites with different lay-ups and textile<br />
architectures. This paper wants to prove that a synergetic approach between instrumented<br />
testing, detailed damage inspection and advanced numerical modelling can provide an answer<br />
to the major challenges that are still present in the research on fatigue of composites.<br />
1. Introduction<br />
The research on fatigue in composites in general has been largely inspired by the research on<br />
fatigue in metals. Despite the advantages that this knowledge transfer has provided, it has also<br />
brought about that there is still a widespread belief that the fatigue behaviour of metals and<br />
composites is indeed very similar. As a consequence the aim of most fatigue tests on<br />
* E-mail address: Wim.VanPaepegem@UGent.be
210<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
composites is still to establish the S-N curve for that particular composite. The efforts to<br />
combine such fatigue tests with a variety of online and offline monitoring techniques and<br />
detailed numerical simulations of the experimental boundary conditions and observed<br />
material degradation, are much more limited.<br />
This paper wants to give a general overview of the different types of fatigue tests, the<br />
available online and offline monitoring techniques and the indispensable need of finite<br />
element calculations to understand the outcomes of these tests. As such, it should become<br />
clear that one single experimental fatigue test, if properly instrumented and simulated, can<br />
provide a lot more information about the fatigue behaviour of the tested composite material.<br />
The next paragraphs will discuss:<br />
• the different fatigue test set-ups and related online monitoring techniques,<br />
• the inspection of fatigue damage,<br />
• the finite element simulation of experimental boundary conditions.<br />
2. Fatigue Test Set-ups and Online Monitoring Techniques<br />
In this paragraph, a general overview of the most relevant fatigue test set-ups is given:<br />
(i) tension-tension fatigue, (ii) bending fatigue, and (iii) shear dominated fatigue. The related<br />
online monitoring techniques are discussed and some examples of measurements are briefly<br />
presented.<br />
An elaborate discussion of all types of fatigue testing, including tension-compression<br />
fatigue, biaxial fatigue and torsional fatigue, can be found elsewhere [1].<br />
2.1. Tension-Tension Fatigue<br />
The uni-axial tension-tension fatigue test is the most widely used fatigue test. The coupon<br />
geometry is a parallel-sided specimen, instrumented with tabs. The choice of the tabbing<br />
material differs among the testing laboratories. Some prefer steel or aluminium tabs, but most<br />
of them use glass/epoxy tabs, where the glass reinforcement has a [+45°/-45°]ns stacking<br />
sequence. In most cases, the tabs are straight-sided non-tapered tabs.<br />
A fatigue test is usually conducted with a servo-hydraulic testing machine, equipped with<br />
grips that clamp the specimen. The alignment of the specimen is very important. No bending<br />
loads must be induced in the specimen due to misalignment.<br />
In tension-tension fatigue tests, the stress ratio R (= σmin/σmax) is often chosen to be 0.1.<br />
The test frequency is always chosen as high as possible to limit the duration of the test and<br />
minimize the cost, but the fatigue response of some composites strongly depends on the<br />
frequency (especially in case of fibre-reinforced thermoplastics).<br />
In the international standards, the number of cycles to failure is considered as the main<br />
outcome of the tension-tension fatigue test. Yet it is worth the effort to use online<br />
instrumentation methods.<br />
The most simple and effective online measurement is the axial stiffness evolution. The<br />
axial stiffness can be directly calculated from the axial stress (loadcell) and the axial strain<br />
(extensometer). The axial strain must never be calculated from the axial displacement and the
<strong>Research</strong> Directions in the Fatigue Testing of Polymer <strong>Composite</strong>s 211<br />
gauge length, because the inevitable slip in the clamps can lead to serious errors in the strain<br />
calculation. Depending on the fibre and matrix type and the stacking sequence, the stiffness<br />
degradation can range from a few percent to several tens of percent [2-7].<br />
If the transverse strain is measured additionally, the Poisson’s ratio νxy can be followed as<br />
well. It has been recently showed by Van Paepegem et al. [8] that the evolution of the<br />
Poisson’s ratio is a very sensitive parameter for fatigue damage. Figure 1 shows the evolution<br />
of the Poisson’s ratio for a [0°/90°]2s unidirectional glass fabric/epoxy composite in tensiontension<br />
fatigue. The νxy – εxx curves in strain-controlled fatigue between 0.0006 and 0.006<br />
show a highly nonlinear behaviour and are upper-bounded by the static degradation of the<br />
Poisson’s ratio.<br />
ν xy [-]<br />
0.20<br />
0.15<br />
0.10<br />
0.05<br />
-0.00<br />
0.000 0.005 0.010 0.015 0.020<br />
-0.05<br />
-0.10<br />
-0.15<br />
-0.20<br />
ν xy versus ε xx for [0°/90°] 2s fatigue test W_090_8<br />
ε xx [-]<br />
static [0°/90°] 2s test IF4<br />
static [0°/90°] 2s test IF6<br />
[0°/90°] 2s fatigue test W_090_8: cycle 600 + 5<br />
[0°/90°] 2s fatigue test W_090_8: cycle 3600 + 5<br />
[0°/90°] 2s fatigue test W_090_8: cycle 37200 + 5<br />
Figure 1. Evolution of the Poisson’s ratio ν xy in function of the longitudinal strain ε xx for a glass/epoxy<br />
[0°/90°] 2s specimen at three chosen intervals in the fatigue test [8].<br />
Another online technique is the use of embedded optical fibre sensors with a Bragg<br />
grating. The Bragg grating is a periodical variation of the optical refractive index that is<br />
written in the core of the glass fibre and is typically a few millimetres in length (Figure 2).<br />
When broadband light is transmitted into the optical fibre, the Bragg grating acts as a<br />
wavelength selective mirror. For each grating only one wavelength, the Bragg wavelength, λB<br />
is reflected with a Full Width at Half Maximum of typically 100 pm, while all other<br />
wavelengths are transmitted. The Bragg wavelength is directly proportional with the period of<br />
the Bragg grating. If the optical fibre sensor is embedded in a composite laminate, the strain<br />
in the loaded laminate will cause the period of the Bragg grating to change, and hence the<br />
value of the reflected Bragg wavelength.
212<br />
The advantages are numerous:<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
Figure 2. Measurement principle of the optical fibre sensor.<br />
• the measurement is absolute and does not drift in time,<br />
• fibre optic sensors are rugged passive components resulting in a high lifetime<br />
(>20 years) and are insensitive to electromagnetic interference,<br />
• the fibre Bragg grating forms an intrinsic part of the optical fibre and has very<br />
small dimensions which makes it very suitable for embedding in composite plates,<br />
• many fibre Bragg gratings can be multiplexed employing only one optical line so<br />
more sensing points can be read out at the same time.<br />
Doyle et al. [9] experimented on the use of fibre optic sensors for tracking the cure<br />
reaction of a fibre reinforced epoxy, with success. They also successfully demonstrated the<br />
feasibility of these sensors for monitoring the stiffness reduction due to fatigue damage, for<br />
thermosetting matrix.<br />
De Baere et al. [10,11] have shown that the optical fibre sensors also survive the<br />
production process for carbon fabric thermoplastics (both autoclave and compression<br />
moulding) and that the correspondence between the axial strain measurements from the<br />
extensometer and the optical fibre sensor were identical in tension-tension fatigue tests (see<br />
Figure 3). That means that the adhesion of the embedded optical fibre sensor to the<br />
surrounding thermoplastic material is very good.<br />
The accumulation of permanent strain is another important phenomenon to monitor.<br />
Especially in composites with large residual stresses built up during manufacturing, the relief<br />
of thermal stresses due to fatigue cracking can result in accumulation of permanent strain.<br />
There again, optical fibre sensors are very sensitive sensors to measure these permanent<br />
strains. Figure 4 shows the stress-strain curves of intermediate static tensile tests during a<br />
tension-tension fatigue test on a carbon thermoplastic. The accumulation of permanent strain<br />
can be clearly seen.
<strong>Research</strong> Directions in the Fatigue Testing of Polymer <strong>Composite</strong>s 213<br />
Figure 3. Comparison of the longitudinal strain ε xx measurement from the optical fibre and the<br />
extensometer in a tension-tension fatigue test of a carbon fibre-reinforced thermoplastic [11].<br />
Figure 4. Intermediate static tests in a tension-tension fatigue experiment of a carbon fibre-reinforced<br />
thermoplastic [11].<br />
Resistance measurement is a well-established damage detection technique for<br />
unidirectional carbon composites [12]. For a long time, there has been disagreement between<br />
researchers whether the resistance should increase or decrease when local fibre fractures<br />
occur [13-15]. In a recent series of articles, it has been clearly demonstrated that the<br />
resistance must increase with increasing damage to the fibre yarns, but a lot of researchers<br />
observe a decrease of resistance, due to bad contact of the electrodes.
214<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
Recently, De Baere et al. [16,17] showed that resistance measurement also works very<br />
well for monitoring damage in carbon fabric reinforced thermoplastics under tension-tension<br />
fatigue loading. Current injection has been done with an innovative technique. Behind the<br />
tabs, in the strain-free area of the specimens, the current is injected by means of two rivets at<br />
both ends of the specimen, as shown in Figure 5.<br />
Figure 5. Use of rivets for electrical resistance measurement in carbon fibre composites [17].<br />
Figure 6 shows the evolution of relative resistance change ρ and axial fatigue stress σxx<br />
during fatigue cycles 4025 till 4030.<br />
Figure 6. Correspondence between applied sine wave of stress σ xx and measured resistance in a tensiontension<br />
fatigue test of a carbon fabric/PPS composite [17].<br />
2.2. Bending Fatigue<br />
Uni-axial fatigue experiments in tension/compression are most often used in fatigue research<br />
[18-20] and accepted as a standard fatigue test, while bending fatigue experiments are<br />
scarcely used to study the fatigue behaviour of composites [21-23]. Bending fatigue tests<br />
differ in several aspects:
<strong>Research</strong> Directions in the Fatigue Testing of Polymer <strong>Composite</strong>s 215<br />
• the bending moment is (piecewise) linear along the length of the specimen (3-point<br />
bending, 4-point bending, cantilever beam bending). Hence stresses, strains and<br />
damage distribution vary along the gauge length of the specimen. On the contrary,<br />
with tension/compression fatigue experiments, the stresses, strains and damage are<br />
assumed to be equal in each cross-section of the specimen,<br />
• due to the continuous stress redistribution, the neutral fibre (as defined in the classic<br />
beam theory) is moving in the cross-section because of changing damage<br />
distributions. Once a small area inside the composite material has moved for example<br />
from the compressive side to the tensile side, the damage behaviour of that area is<br />
altered considerably,<br />
• the finite element implementation of related damage models gives rise to several<br />
complications, because each material point is loaded with a different stress, strain<br />
and possibly stress ratio, so that damage growth can be different for each material<br />
point. In tension/compression fatigue tests, the stress- or strain-amplitude is constant<br />
during fatigue life and differential equations describing decrease of stiffness or<br />
strength, can often be simply integrated over the considered number of loading<br />
cycles,<br />
• smaller forces and larger displacements in bending allow a more slender design of<br />
the fatigue testing facility.<br />
Basically, three types of bending fatigue tests can be distinguished: (i) three-point<br />
bending [24,25], (ii) four-point bending [26], and (iii) cantilever bending [22,27-30]. The<br />
success of these tests for fatigue of polymer composites is quite limited, because the<br />
interpretation of the results is more difficult and in case of stiffness degradation, stress<br />
redistribution across the specimen height comes into play.<br />
Moreover, as long as the bending stiffness of the laminate is high enough (e.g. sandwich<br />
composites), the deflections are small and linear beam theory still applies, but once that the<br />
bending stiffness of the composite decreases (e.g. thin laminates), the deflections are large<br />
and geometric nonlinearities and friction at the roller supports affect the fatigue results.<br />
The authors designed a test set-up for cantilever bending fatigue tests as depicted in<br />
Figure 7.<br />
The power of the motor is transmitted by a V-belt to a second shaft. The second shaft<br />
bears a mechanism with crank and connecting rod, which imposes an alternating<br />
displacement on the hinge (point C in Figure 7) that connects the connecting rod with the<br />
lower clamp of the composite specimen. At the upper end the specimen is clamped (point A<br />
in Figure 7). Hence the sample is loaded as a composite cantilever beam.<br />
A full Wheatstone bridge on the connecting rod is used to measure the force acting on the<br />
composite specimen. Due to the (bending) stiffness degradation of the specimen during<br />
fatigue life, the measured force will gradually decrease as the amplitude umax of the prescribed<br />
displacement remains constant. In order to record the out-of-plane displacement profile, it<br />
was necessary to develop a mechanism to hold the specimen fixed in this state, because<br />
recording the profile while the test keeps running at a frequency of 2.2 Hz, gives rise to some<br />
practical problems. A rotary digital encoder was attached to the second shaft. Its angular<br />
position (relative to a certain reference angle) is directly related with the loading path of the
216<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
composite specimen. The frequency inverter reads the signal from the rotary encoder and can<br />
stop and hold the motor at a predetermined angular position of the encoder.<br />
Figure 7. Test set-up for cantilever bending fatigue [28].<br />
A digital photograph of the out-of-plane displacement profile is taken from the side view.<br />
To enhance the contrast, the edge of the composite specimen has been painted white. An<br />
example of such a digital photograph is given in Figure 8. When the number of pixels for a<br />
known distance is counted, the out-of-plane displacement profile can be calculated. Thereto<br />
an edge-detection algorithm is used which detects the edges of the composite specimen on the<br />
digital photograph. Figure 8(right) shows an example of the edge detection algorithm. Of<br />
course, the calculated out-of-plane displacement profile applies to the deformation of the<br />
specimen surface, not to the out-of-plane displacement of the midplane of the laminate.<br />
Figure 8. Use of image processing algorithms to track the out-of-plane displacement profile in<br />
cantilever bending fatigue.
<strong>Research</strong> Directions in the Fatigue Testing of Polymer <strong>Composite</strong>s 217<br />
2.3. Shear Dominated Fatigue<br />
Fatigue testing in pure shear is very difficult. Lessard et al. [31] modified the static three-rail<br />
shear test (ASTM D 4255/D 4255M – 01) to do fatigue testing on carbon/epoxy plates.<br />
A much more common method are the tension-tension fatigue tests on a [+45°/-45°]ns<br />
laminate, This test is based on the ASTM D3518/D3518M-94(2001) Standard Test Method<br />
for ‘In-Plane Shear Response of Polymer Matrix <strong>Composite</strong> <strong>Materials</strong> by Tensile Test of a<br />
±45° Laminate’. This standard explains how the shear stress-strain curve can be derived from<br />
a static tensile test on a ±45° laminate, by measuring the longitudinal and transverse strain.<br />
The test is also called a bias tension test, because the bias (or cross-grain) direction is the 45°<br />
direction between warp and weft direction in case of fabric reinforced composites.<br />
In both pure shear and shear dominated fatigue, the test frequency is a very important<br />
parameter. The shear stresses can lead to significant autogeneous heating and once the<br />
temperature exceeds the glass transition temperature, the deformations can be very large.<br />
Figure 9 shows the localized yielding of a [(+45°,-45°)]2s carbon fabric/PPS composite in<br />
tension-tension fatigue at 2 Hz. Temperature rises up to 90 °C were measured with a<br />
thermocouple at the top surface.<br />
Figure 9. Localized yielding of a [(+45°,-45°)] 4s carbon fabric/PPS composite in tension-tension fatigue<br />
at 2 Hz.<br />
Shear stress τ 12 [MPa]<br />
60<br />
50<br />
40<br />
30<br />
20<br />
10<br />
Shear stress-strain curve for cyclic [+45°/-45°] 2s test IH2<br />
IH2 cyclic test<br />
IH4 static test<br />
0<br />
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07<br />
Shear strain γ12 [-]<br />
Figure 10. Shear stress-strain curve for the cyclic tensile test on a [+45°/-45°] 2s glass/epoxy specimen<br />
and the envelope of the corresponding static test [32].
218<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
A hardly studied phenomenon is the accumulation of permanent strain during shear<br />
dominated fatigue loading. For composite materials with a thermoplastic matrix, creep effects<br />
seem to be dominant, while in case of thermosetting materials, permanent strain is simply<br />
neglected in most reported literature. Moreover, for both types of material, the phenomenon is<br />
not understood quite well.<br />
Van Paepegem et al. [32,33] studied the accumulation of permanent shear strain in<br />
[+45°/-45°]2s glass/epoxy laminates under cyclic loading. They showed that the shear<br />
modulus significantly degrades, but that the accumulation of permanent shear strain is even<br />
more important. Figure 10 shows the accumulation of permanent shear strain in cyclic loading<br />
of unidirectional glass fabric/epoxy composites.<br />
3. Visualization of Fatigue Damage<br />
3.1. Micrographs<br />
The most easy inspection technique is visual inspection. Depending on the difference in<br />
optical refractive index of the matrix and fibre materials, the transparency of the composite<br />
laminate can be very high. Gagel et al. [34] reported an extraordinary high transparency of Eglass<br />
multi-axial non-crimp fabric epoxy laminates. Matrix cracks, voids and inclusions could<br />
be detected easily by transmitted light.<br />
Optical or light microscopy provides a direct path from observations made with the naked<br />
eye, to what is visible at magnifications up to about 1000 × [35]. Fracture surfaces are<br />
embedded in resin and polished before observation. Figure 11 shows a microscopic image of<br />
the damage in a plain weave glass/epoxy composite loaded in bending fatigue [36].<br />
1 mm<br />
Figure 11. Micrograph of the fatigue damage at the clamped end of a composite specimen loaded in<br />
cantilever bending fatigue [36].<br />
3.2. Ultrasonic Inspection<br />
A very common inspection technique for fatigue damage in (textile) composites is<br />
ultrasonics. Ultrasonics can be performed in various modes of operation, but the most<br />
common for fatigue damage detection is the through-transmission (C-scan) technique.
<strong>Research</strong> Directions in the Fatigue Testing of Polymer <strong>Composite</strong>s 219<br />
Through-transmission ultrasonics basically consists of a transducer for emitting ultrasonic<br />
pulses that is placed at or near one surface and a receiver sensor that is located at the opposite<br />
surface. The technique applies to relatively low frequency sound beams, typically 0.5 MHz to<br />
15 MHz, having a small aperture. The transducer and receiver are coupled to the surfaces or<br />
they are immersed in water together with the composite. The ultrasound waves are attenuated<br />
by defects in the composite and the acoustic attenuation is monitored using the receiver [37].<br />
Figure 12 shows the C-scan of a thermoplastic composite specimen tested in three-point<br />
bending fatigue. The central area is clearly damaged.<br />
Figure 12. C-scan of the central damaged zone in a composite specimen loaded in three-point bending<br />
fatigue.<br />
With classical ultrasonic C-scans, the surface of the object under investigation is scanned<br />
point by point in order to detect and to localise possible defects or possible anomalies. In a Cscan<br />
the transducer is normally kept perpendicular and at a constant distance to the surface of<br />
the object.<br />
A less known but promising technique is the ultrasonic polar scan. With the use of polar<br />
scans we do not aim at the detection and localisation of defects or anomalies, but rather at the<br />
characterisation of the material. Therefore in a polar scan a single representative point of the<br />
object is scanned, under all possible angles θ and ϕ of incidence of the ultrasonic beam, as is<br />
shown in Figure 13. Due to the dimensions of a real ultrasonic beam, a small zone, rather than<br />
a single point of the object is scanned. The distance between transducer and scanned point is<br />
again kept constant, and an acoustic coupling medium, such as water, is used. As is also the<br />
case with classical C-scans, scanning is performed using pulsed signals. Obliquely incident<br />
ultrasonic waves have already been used more or less frequently for purposes of material<br />
characterisation. In each case wave velocities or arrival times of ultrasonic pulses were<br />
measured [38-40]. In a polar scan however, the amplitude of the transmitted beam is<br />
measured. Amplitude measurements are much easier to perform, and can be done with the<br />
most simple ultrasonic apparatus, an advantage for the possible application of the technique in<br />
industrial circumstances.<br />
In the early eighties Van Dreumel and Speijer [41] have shown that ultrasonic polar scans<br />
in principle can visualise in a non-destructive way fibre orientations of the layers in laminates<br />
stacked from unidirectional layers. Unfortunately, after these experiments, polar scans have<br />
been hardly studied or used any more, the reasons for this being mainly the complexity of the<br />
"formation" of a polar scan, and the lack of means at that time for the numerical simulation of<br />
a polar scan.
220<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
Figure 13. Schematic drawing of the polar scan set-up (left) and example of an experimentally<br />
measured polar scan of a unidirectional carbon/epoxy composite [42].<br />
Yet, Maes [43] showed that the recorded polar scans of a glass fabric/epoxy composite<br />
before and after fatigue damage clearly differ, as shown in Figure 14. Due to the degradation<br />
of the elastic properties, the propagation speed of ultrasound in the respective directions has<br />
changed.<br />
Figure 14. Polar scan of a glass fabric/epoxy composite before fatigue loading (left) and after fatigue<br />
loading (right) [43].<br />
3.3. X-ray Micro-tomography<br />
High-resolution 3D X-ray micro-tomography or micro-CT is a relatively new technique<br />
which allows scientists to investigate the internal structure of their samples without actually<br />
opening or cutting them [44]. Without any form of sample preparation, 3D computer models<br />
of the sample and its internal features can be produced with this technique. In order to<br />
perform tomography, digital radiographs of the sample are made from different orientations<br />
by rotating the sample along the scan axis from 0 to 360 degrees. After collecting all the
<strong>Research</strong> Directions in the Fatigue Testing of Polymer <strong>Composite</strong>s 221<br />
projection data, the reconstruction process is producing 2D horizontal cross-sections of the<br />
scanned sample. These 2D images can then be rendered into 3D models, which enable to<br />
virtually look into the object.<br />
Figure 15 shows the micro-tomography images of a fatigue damaged 5-harness satin<br />
weave carbon/PPS (left) and the embedded optical fibre sensors in a carbon thermoplastic<br />
composite (right).<br />
Figure 15. Micro-tomography images of a fatigue damaged carbon/PPS composite (left) and three<br />
composite samples with embedded optical fibre sensor (right) [11].<br />
4. Finite Element Simulation of Experimental Boundary<br />
Conditions<br />
In this paragraph it is shown that finite element simulations should support the experimental<br />
work in order to be able to discriminate between intrinsic material behaviour and induced<br />
effects by (insufficiently understood) experimental boundary conditions.<br />
Four examples are given where the strong correlation between experimental<br />
measurements and finite element simulations is proven.<br />
4.1. Clamping Conditions in Tension-Tension Fatigue<br />
As stated before, the composite coupons for tension-tension fatigue testing are parallel-sided<br />
specimens, instrumented with aluminium or composite tabs. One of the main concerns in<br />
tension-tension fatigue testing of composites is tab failure, i.e. the specimen fails just next to<br />
or inside the tabbing area. Such failures are due to the inevitable stress concentration near the<br />
clamped edges.<br />
In Figure 16, two types of standard tensile machine fixtures are shown, with the<br />
dimensions of the grips.<br />
In order to optimize the shape and length of the tabs, it is important to simulate the stress<br />
state near the clamped region. Therefore a simulation of part of the clamping mechanism has<br />
been done in ABAQUS/Standard v6.6-2.
222<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
Figure 16. Instron TM mechanical grips (left) and Instron TM servohydraulic grips (right).<br />
Figure 17 illustrates the simulated parts in the finite element model. Because of<br />
symmetry, only half of the clamps is modelled, which reduces calculation time. The<br />
corresponding symmetry boundary conditions have been imposed on the specimen. To further<br />
reduce computation time, a rigid body constraint is placed on part of the housing cylinder of<br />
the wedge grips, only the area where the cylinder makes contact with the grip is left<br />
deformable. Furthermore, a part that models the wedge is added, also with a rigid body<br />
constraint to reduce calculation time. The reference point of this part is given a certain<br />
downward displacement. This part represents the hydraulic plunger in the hydraulic clamps.<br />
Figure 17. The simulated clamps in the finite element model [45].
<strong>Research</strong> Directions in the Fatigue Testing of Polymer <strong>Composite</strong>s 223<br />
Two time steps were implemented: in the first, the wedge was given a downward motion<br />
of 0.75 mm, simulating the pre-stressing of the grips; in the second, the bottom of the<br />
specimen was pulled down over 1 mm, simulating a tensile test.<br />
Contact conditions were imposed between the surfaces of the specimen and the grip, the<br />
grip and the cylinder and the grip and the wedge. Since the grip first follows the movement of<br />
the wedge and then the movement of the specimen, the slave surfaces of all contact conditions<br />
mentioned, were placed on the grips. Between specimen and grip, the tangential behaviour<br />
‘rough’ was implemented, which means that no slip occurs once nodes make contact. For the<br />
other contact conditions, the ‘lagrange’ condition was used, which means that the tangential<br />
force is μ times the normal force, μ being the friction coefficient. The same friction<br />
coefficient was used for both conditions.<br />
The grip was meshed with a C3D8R element, a linear brick element with reduced<br />
integration, whereas all other parts were meshed with C3D20R, a quadratic brick element<br />
with reduced integration. The C3D8R of the grip is required instead of the C3D20R, since the<br />
slave surfaces require midface nodes and the C3D20R do not have one.<br />
For the grip, the wedge and the cylinder, steel was implemented with a Young’s modulus<br />
of 210000 MPa and a Poisson’s ratio of 0.3. The specimen was modelled in a composite<br />
material with the following elastic properties (Table 1).<br />
Table 1. The implemented engineering constants in the finite-element model for the<br />
specimen.<br />
E11 [MPa] 56000 ν12 [-] 0.033 G12 [MPa] 4175<br />
E22 [MPa] 57000 ν13 [-] 0.3 G13 [MPa] 4175<br />
E33 [MPa] 9000 ν23 [-] 0.3 G23 [MPa] 4175<br />
In [46], the authors have derived an analytical formula for the clamping setup illustrated<br />
in Figure 18 that describes the interaction between the load F on the specimen, the force RA of<br />
the plunger (see Figure 18) represented by part A and the contact force P on the specimen.<br />
Figure 18. Symbolic representation of the gripping principle of a clamp [45].
224<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
The following equation was derived, with μij the coefficient of friction between parts i<br />
and j (i,j=A,B,C):<br />
F cosα −μBCsin α (1 −μACμBC)cos α −( μBC −μAC<br />
)sinα<br />
P= + RA<br />
2 sinα + μ cosα sinα + μ cosα<br />
BC BC<br />
For both grips displayed in Figure 16, the angle α is set equal to 10 degrees.<br />
In [45] it was shown that the grips in Figure 17 can be replaced by their equivalent<br />
contact pressure, calculated from Equation (1), because the simulated contact pressure with<br />
the finite element model of Figure 17 perfectly corresponds with the analytically calculated<br />
contact pressure.<br />
Next a detailed finite element model of the end tab region has been developed. Figure 19<br />
shows the model of this setup, both mesh and boundary conditions are illustrated.<br />
The specimen is meshed using C3D20R elements using a global element size of 2 mm.<br />
Where stress concentrations were expected, the element size was reduced to 0.5 mm. The<br />
thickness of the specimen was 2.4 mm, which is also the thickness of the tabs, as has already<br />
been mentioned. The material properties for the composite specimen are given in Table 1.<br />
For the boundary conditions, the displacement along the 1 and 2 axis was inhibited for<br />
planes B1 (on top) and B2 (at the bottom), simulating the ‘rough’ boundary condition from the<br />
previous paragraph. Since contraction of the specimen is possible in the 3-direction due to the<br />
Poisson effect, the movement along the 3-axis was allowed for both planes. In order to<br />
prevent movement of the entire sample along the 3 axis, the central line of plane C (at the<br />
back) was fixed.<br />
Figure 19. Illustration of the model for the end tab region [45].<br />
Two time steps were modelled. In the first, the contact pressure p, calculated from<br />
Equation (1), was imposed. In the second, a tensile stress of 600 MPa was applied on surface<br />
A. The exact value of the stress does not matter, since the stress concentration factors are<br />
compared.<br />
(1)
<strong>Research</strong> Directions in the Fatigue Testing of Polymer <strong>Composite</strong>s 225<br />
In that way, a detailed analysis of the stress concentration factors in the end tab region<br />
was studied for several clamping conditions and end tab geometries. Detailed results can be<br />
found in [45].<br />
4.2. Friction in Single-Sided 3-Point Bending Fatigue<br />
In uni-axial fatigue tests on fibre-reinforced composites, the stress-strain hysteresis loop can<br />
be used as a measure for stiffness degradation and energy dissipation. In case of three-point<br />
bending fatigue tests, the hysteresis loop of the bending force versus midspan displacement<br />
can yield similar information. In this numerical study, it is shown that the shape of the<br />
hysteresis loop can be affected significantly by friction at the supports, especially for large<br />
deflections. As such, the area of the closed hysteresis loop is no longer a measure for energy<br />
dissipation and damage growth.<br />
Figure 20 shows typical hysteresis loops of the bending force versus midspan deflection<br />
at several times during fatigue life for the [90°/0°]2s carbon/PPS laminate. The amplitude of<br />
the midspan deflection was 14.5 mm and the testing frequency was 2.0 Hz. The hysteresis<br />
loops are gone through in clockwise direction (loading – unloading).<br />
The problem treated here, is the typical shape of the hysteresis curve. One would expect<br />
that the dissipated energy during such a hysteresis cycle is used for initiation and propagation<br />
of microscopic fatigue damage, but in the case of this material, ultrasonic C-scans could not<br />
detect any significant fatigue damage in the specimens (apart from the last stage in fatigue<br />
life). Therefore it was assumed that the effect could be induced by friction at the supports,<br />
given the very large midspan displacements for a short span length.<br />
Bending force [N]<br />
Typical hysteresis curves in bending for [90°/0°] 2s carbon/PPS laminate<br />
700<br />
600<br />
500<br />
400<br />
300<br />
200<br />
100<br />
Cycle 1<br />
Cycle 80 000<br />
Cycle 160 000<br />
Cycle 191 000<br />
0<br />
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15<br />
Deflection [mm]<br />
Figure 20. Typical hysteresis curves in bending for [90°/0°] 2s carbon/PPS laminate.
226<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
Finite element simulations have been done to prove the hypothesis of friction affecting<br />
the shape of the hysteresis loop.<br />
The simulations have been done with the commercial implicit finite element code<br />
SAMCEF TM . The finite element mesh is shown in Figure 21. Eight layers of composite have<br />
been modelled with isoparametric volumic elements, one element per layer through the<br />
thickness. The end supports and the load striking edge have been modelled as rigid body<br />
cilinders with radius 5 mm. The contact conditions between supports and composite elements<br />
can have a different friction coefficient.<br />
The material is assumed to behave in a linear elastic manner, but the geometric<br />
nonlinearity is taken into account.<br />
Figure 21. SAMCEF TM finite element model of the three-point bending test.<br />
Figure 22. Simulated displacement contours for a three-point bending test on a [90°/0°] 2s carbon/PPS<br />
laminate.<br />
Figure 22 shows the simulated deflection of the [90°/0°]2s specimen for a prescribed midspan<br />
displacement of 14.5 mm (in agreement with the imposed displacement in the three-point bending<br />
fatigue tests).
Bending force [N]<br />
<strong>Research</strong> Directions in the Fatigue Testing of Polymer <strong>Composite</strong>s 227<br />
700<br />
600<br />
500<br />
400<br />
300<br />
200<br />
100<br />
Simulated hysteresis curves for different friction conditions<br />
μ = 0.0<br />
μ = 0.1<br />
μ = 0.2<br />
μ = 0.3<br />
0<br />
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15<br />
Deflection [mm]<br />
Figure 23. Simulated hysteresis curves for a [90°/0°] 2s carbon/PPS laminate with different friction<br />
conditions at the supports: (i) μ = 0.0, (ii) μ = 0.1, (iii) μ = 0.2 and (iv) μ = 0.3.<br />
700<br />
600<br />
500<br />
400<br />
300<br />
200<br />
Bending force [N] Simulated force-displacement history for three-point bending test (μ = 0.3)<br />
100<br />
0<br />
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15<br />
Deflection [mm]<br />
Figure 24. Detailed simulation of the force-displacement curve of the [90°/0°] 2s carbon/PPS laminate<br />
for μ = 0.3.<br />
In Figure 23, the simulated hysteresis curves are plotted for different friction conditions.<br />
The complete loading-unloading path has been simulated, where the imposed midspan
228<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
displacement increases from 0.0 to 14.5 mm and decreases back to 0.0 mm. The curve of<br />
bending force versus midspan deflection is shown for four different friction conditions at the<br />
two end supports: (i) μ = 0.0, (ii) μ = 0.1, (iii) μ = 0.2 and (iv) μ = 0.3.<br />
It can be clearly seen that for μ = 0.0, there is no hysteresis. However, the curve is<br />
slightly nonlinear due to the geometric nonlinearity (large deflection). For μ = 0.3, the typical<br />
shape of the hysteresis curve is found back, although no material damage was taken into<br />
account. As a consequence, the shape variation is only due to the friction coefficient.<br />
The simulation for μ = 0.3 has been done again with a very small time step at the<br />
transition from loading to unloading. The effect is even more pronounced, as can be seen in<br />
Figure 24. It is worth to mention that the value of the maximum bending force is in very good<br />
agreement with the experimentally measured one during the three-point bending fatigue tests<br />
(see Figure 20).<br />
Finally, the simulated stress-strain history in Figure 25 for one of the integration points of<br />
the finite element at the tensile side in midspan proves that there is no material hysteresis.<br />
It has been shown that the friction between the composite specimen and the supports was<br />
the predominant cause of this phenomenon. Static bending tests with different support<br />
conditions were performed and three-dimensional finite element analyses were done with<br />
different friction coefficients. These tests confirmed the hypothesis.<br />
Stress σ 11 [MPa]<br />
Simulated stress-strain history for three-point bending test (μ = 0.3)<br />
700<br />
600<br />
500<br />
400<br />
300<br />
200<br />
100<br />
0<br />
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2<br />
Strain ε11 [%]<br />
Figure 25. Simulated stress-strain history of [90°/0°] 2s carbon/PPS laminate for μ = 0.3.<br />
Therefore, it can be concluded that the information from hysteresis loops in bending<br />
fatigue must be considered very carefully.
<strong>Research</strong> Directions in the Fatigue Testing of Polymer <strong>Composite</strong>s 229<br />
4.3. Fully-Reversed 3-Point Bending Fatigue<br />
This study investigates whether a three-point bending setup with fully reversed loading can be<br />
used for the validation of (a combination of) damage models for thin composite laminates in<br />
static or fatigue loading conditions.<br />
When fully reversed bending is used, each side of the specimen is successively loaded in<br />
tension as well as in compression. As a result, the material in the beam sees alternating<br />
tension and compression, which makes this setup ideal for the validation of tensioncompression<br />
fatigue models.<br />
If fully reversed bending is considered, some changes must be made to the original threepoint<br />
bending setup in Figure 26.<br />
Figure 26. Single-sided three-point bending test [47].<br />
Figure 27. The central roll and the rotating outer support (left) and the mounted fully reversed threepoint<br />
bending setup [47].<br />
(i) for the central indenter as well as the outer supports, two contact cylinders are<br />
required, one for the upward and one for the downward motion. Since the centre of the<br />
specimen remains horizontal (see Figure 26, right), no additional modifications are needed for<br />
the central indenter; ii) because the specimen rotates at its ends (see Figure 26, right), the<br />
outer supports need to allow for this rotation. Otherwise, this would induce unwanted reaction<br />
forces in the specimen, corrupting the fatigue data. The indenter and used supports for the
230<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
developed fully-reversed bending fatigue set-up are shown in Figure 27 and the total<br />
assembly in Figure 28.<br />
Figure 28. Total assembly of the fully reversed three-point bending setup [47].<br />
The correct modelling of the boundary conditions applied in this fully-reversed threepoint<br />
bending set-up is not straightforward. Below the detailed finite element model is<br />
discussed.<br />
Since the central indenting rolls do not require any rotation, they are modelled by two<br />
separate rolls (Figure 29, left). To reduce calculation time, rigid body conditions are applied<br />
on all areas that do not make contact with the specimen. The element type is the same linear<br />
brick element with reduced integration, C3D8R, as before, the element size is 0.5 mm.<br />
The easiest way to model the rotating support is by modelling it as a single part, which is<br />
depicted in Figure 29, right. The rolls are slightly longer than the width of the specimen, so<br />
that the specimen does not make contact with the connecting part between the two rolls. Extra<br />
partitions are created resulting in a better mesh. The distance between the two rolls is equal to<br />
the thickness of the specimen, the rolls have a diameter of 10, as was the case in the<br />
experimental setup.<br />
Figure 29. The model of the central indenter as two separate rolls (left) and the model of the rotating<br />
support as one part (right) [47].
<strong>Research</strong> Directions in the Fatigue Testing of Polymer <strong>Composite</strong>s 231<br />
Again there is a rigid-body constraint on all the partitions that do not make contact with<br />
the specimen, in order to save calculation time (Figure 29, left and right). The part is meshed<br />
with C3D8R elements; the element size is also 0.5 mm. The latter is done to assure that the<br />
calculation does not diverge as a result of contact problems.<br />
Figure 30 shows the final model of the fully-reversed three-point bending set-up.<br />
For the boundary conditions of the rotating support, only the rotation of the support<br />
around its ‘natural axis’ is allowed, all other movement is constrained.<br />
For the contact conditions, the slave surface is put on the support and the master surface<br />
is on the specimen. The latter helps the rotating of the support, since normally, the slave<br />
surface follows the movement of the master surface.<br />
The specimen is a beam with dimensions 2.4 mm x 15 mm x 80 mm and it is meshed<br />
with quadratic brick elements with reduced integration, C3D20R. The global size of the<br />
elements is 3 mm. However, in the zones of contact, the size is 1 mm to ensure that no<br />
convergence problems occur due to the contact conditions. The material model is the same<br />
linear elastic model as in paragraph 4.1 (see Table 1).<br />
Figure 30. Illustration of the mesh and the boundary conditions for the three-point bending setup with<br />
rotating supports [47].<br />
With this model, the bending stresses in the experimental fatigue tests could be simulated<br />
very well. More details can be found in [47].<br />
4.4. Cantilever Bending Fatigue<br />
This paragraph deals with the correct modelling of the set-up for cantilever bending fatigue<br />
that was already shown in Figure 7. It is tempting to model the clamped side of the specimen<br />
by a number of fully constraint nodes in the finite element mesh. However, this model<br />
appears to add unwanted stiffness to the specimen and the predicted force is considerably<br />
higher than the measured one.<br />
The calculations were done for a plain weave glass/epoxy specimen. The prescribed<br />
displacement umax (= 34.4 mm) was chosen rather large in order to assess the effect of<br />
geometrical non-linearities. The corresponding maximum bending force for the specimen,
232<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
measured by a strain gauge bridge, was 117.5 Newton at the first loading cycle. Table 2<br />
illustrates the influence of several modelling assumptions. 2-D and 3-D meshes have been<br />
used with “complete fixation” (fixing all nodes in the clamped cross-section) and “clamping<br />
surfaces” (modelling of the clamping plates with prestressing force). Due to symmetry<br />
conditions, the 3-D simulations were performed for the half width of the specimen and they<br />
were indicated in the table as “3-D symmetry models”. All simulations are quasi-static<br />
analyses, except the fourth simulation, which takes into account the inertia forces during<br />
fatigue loading.<br />
From the third and fourth simulation it is confirmed that a quasi-static analysis is<br />
sufficient. Indeed, since the fatigue experiments are performed at a frequency of 2.23 Hz and<br />
the mass of the reciprocating parts is very small due to the limited forces in bending, the<br />
inertia forces are negligible.<br />
Table 2. Comparison of the different finite element models for the bending fatigue setup.<br />
FE model type<br />
No. of<br />
elements<br />
Bending<br />
force [N]<br />
CPU time<br />
2D plane strain, complete fixation 445 155.1 0’17’’<br />
2D plane strain, clamping surfaces 517 141.2 0’46’’<br />
3D symmetry model, complete fixation 1461 139.2 32’45’’<br />
3D symmetry model, complete fixation, inertia forces 1461 138.9 4 h 35’57’’<br />
3D symmetry model, clamping surfaces, no 1765 146.7 21’53’’<br />
geometrical non-linearities<br />
3D symmetry model, clamping surfaces 1765 120.8 40’44’’<br />
Z<br />
Y<br />
X<br />
Figure 31. Finite element model of the bending fatigue setup [36].
<strong>Research</strong> Directions in the Fatigue Testing of Polymer <strong>Composite</strong>s 233<br />
Figure 31 shows the finite element mesh for a 3-D analysis with full modelling of the<br />
clamped surfaces and the prescribed displacement. The diagonal lines in the left part of the<br />
mesh are used by the SAMCEF preprocessor to indicate the presence of clamping conditions.<br />
Due to the symmetry conditions with respect to the (x,y)-plane, only one half of the specimen<br />
width has to be modelled. The lines in the bottom right part make up the rigid body part<br />
where the prescribed bending displacement is applied.<br />
5. Conclusion<br />
This paper has presented a collection of research efforts in the field of (i) fatigue test set-ups<br />
and related online monitoring techniques, (ii) inspection of fatigue damage and (iii) the finite<br />
element simulation of experimental boundary conditions.<br />
It has been shown that an integrated approach of these three research fields can benefit<br />
the knowledge and insight into the fatigue testing of fibre-reinforced composites.<br />
Acknowledgements<br />
The author W. Van Paepegem gratefully acknowledges his finance through a grant of the<br />
Fund for Scientific <strong>Research</strong> – Flanders (F.W.O.), and the advice and technical support of the<br />
Ten Cate company. The author I. De Baere is highly indebted to the university research fund<br />
BOF (Bijzonder Onderzoeksfonds UGent) for his research grant.<br />
References<br />
[1] Harris, B. (ed.) (2003). Fatigue in composites. Science and technology of the fatigue<br />
response of fibre-reinforced plastics. Cambridge, Woodhead Publishing Ltd., 742 pp.<br />
[2] Hashin, Z. (1985). Cumulative damage theory for composite materials: residual life and<br />
residual strength methods. <strong>Composite</strong>s Science and Technology, 23, 1-19.<br />
[3] Whitworth, H.A. (2000). Evaluation of the residual strength degradation in composite<br />
laminates under fatigue loading. <strong>Composite</strong> Structures, 48(4), 261-264.<br />
[4] Highsmith, A.L. and Reifsnider, K.L. (1982). Stiffness-reduction mechanisms in<br />
composite laminates. In : Reifsnider, K.L. (ed.). Damage in composite materials. ASTM<br />
STP 775. American Society for Testing and <strong>Materials</strong>, pp. 103-117.<br />
[5] Yang, J.N., Jones, D.L., Yang, S.H. and Meskini, A. (1990). A stiffness degradation<br />
model for graphite/epoxy laminates. Journal of <strong>Composite</strong> <strong>Materials</strong>, 24, 753-769.<br />
[6] Yang, J.N., Lee, L.J. and Sheu, D.Y. (1992). Modulus reduction and fatigue damage of<br />
matrix dominated composite laminates. <strong>Composite</strong> Structures, 21, 91-100.<br />
[7] Kedward, K.T. and Beaumont, P.W.R. (1992). The treatment of fatigue and damage<br />
accumulation in composite design. International Journal of Fatigue, 14(5), 283-294.<br />
[8] Van Paepegem, W., De Baere, I., Lamkanfi, E. and Degrieck, J. (2007). Poisson’s ratio<br />
as a sensitive indicator of (fatigue) damage in fibre-reinforced plastics. Fatigue and<br />
Fracture of Engineering <strong>Materials</strong> & Structures, 30, 269–276.
234<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
[9] Doyle, C., Martin, A., Liu, T., Wu, M., Hayes, S., Crosby, P.A., Powell, G.R., Brooks,<br />
D. and Fernando, G.F. (1998). In-situ process and condition monitoring of advanced<br />
fibre-reinforced composite materials using optical fibre sensors. Smart <strong>Materials</strong> &<br />
Structures, 7(2), 145-158 APR 1998.<br />
[10] De Baere, I., Voet, E., Van Paepegem, W., Vlekken, J., Cnudde, V., Masschaele, B. and<br />
Degrieck, J. (2007). Strain monitoring in thermoplastic composites with optical fibre<br />
sensors: embedding process, visualization with micro-tomography and fatigue results.<br />
Journal of Thermoplastic <strong>Composite</strong> <strong>Materials</strong>, 20 (September 2007), 453-472.<br />
[11] De Baere, I., Voet, E., Luyckx, G., Van Paepegem, W., Vlekken, J., Cnudde, V.,<br />
Masschaele, B. and Degrieck, J. (2007). On the feasibility of optical fibre sensors for<br />
strain monitoring in thermoplastic composites under fatigue loading conditions.<br />
Proceedings of the Third international workshop on Optical Measurement Techniques<br />
for Structures and Systems (OPTIMESS 2007), May 28-30, 2007, Leuven, Belgium.<br />
[12] Abry, J.C., Bochard, S., Chateauminois, A., Salvia, M. and Giraud, G. (1999). In situ<br />
detection of damage in CFRP laminates by electrical resistance measurements.<br />
<strong>Composite</strong>s Science and Technology, 59, 925-935.<br />
[13] Angelidis, N., Wei, C.Y. and Irving, P.E. (2004). The electrical resistance response of<br />
continuous carbon fibre composite laminates to mechanical strain. <strong>Composite</strong>s Part A,<br />
35, 1135–47.<br />
[14] Chung, D.D.L. and Wang, S. (2006). Discussion on paper ‘The electrical resistance<br />
response of continuous carbon fibre composite laminates to mechanical strain’ by<br />
Angelidis N, Wei CY, Irving PE, <strong>Composite</strong>s: Part A 2004;35:1135–47. <strong>Composite</strong>s<br />
Part A, 37, 1490–1494.<br />
[15] Angelidis, N., Wei, C.Y. and Irving, P.E. (2006). Short communication - Response to<br />
discussion of paper: The electrical resistance response of continuous carbon fibre<br />
composite laminates to mechanical strain. <strong>Composite</strong>s Part A, 37, 1495–1499.<br />
[16] De Baere, I., Van Paepegem, W. and Degrieck, J. (2007). The use of rivets for electrical<br />
resistance measurement on carbon fibre-reinforced thermoplastics. Smart <strong>Materials</strong> and<br />
Structures, 16(5), 1821-1828.<br />
[17] De Baere, I., Van Paepegem, W. and Degrieck, J. (2007). Electrical resistance<br />
measurement on carbon fibre-reinforced thermoplastics with rivets as electrodes.<br />
Proceedings of the Fourth International Conference Emerging Techonologies in Non-<br />
Destructive Testing (ETNDT 4), Stuttgart, Germany, 2 – 4 april 2007.<br />
[18] Fujii, T., Amijima, S. and Okubo, K. (1993). Microscopic fatigue processes in a plainweave<br />
glass-fibre composite. <strong>Composite</strong>s Science and Technology, 49, 327-333.<br />
[19] Schulte, K., Reese, E. and Chou, T.-W. (1987). Fatigue behaviour and damage<br />
development in woven fabric and hybrid fabric composites. In : Matthews, F.L., Buskell,<br />
N.C.R., Hodgkinson, J.M. and Morton, J. (eds.). Sixth International Conference on<br />
<strong>Composite</strong> <strong>Materials</strong> (ICCM-VI) & Second European Conference on <strong>Composite</strong><br />
<strong>Materials</strong> (ECCM-II) : Volume 4. Proceedings, 20-24 July 1987, London, UK, Elsevier,<br />
pp. 4.89-4.99.<br />
[20] Hansen, U. (1997). Damage development in woven fabric composites during tensiontension<br />
fatigue. In : Andersen, S.I., Brøndsted, P., Lilholt, H., Lystrup, Aa., Rheinländer,<br />
J.T., Sørensen, B.F. and Toftegaard, H. (eds.). Polymeric <strong>Composite</strong>s - Expanding the<br />
Limits. Proceedings of the 18th Risø International Symposium on <strong>Materials</strong> Science, 1-5<br />
September 1997, Roskilde, Denmark, Risø International Laboratory, pp. 345-351.
<strong>Research</strong> Directions in the Fatigue Testing of Polymer <strong>Composite</strong>s 235<br />
[21] Ferry, L., Gabory, D., Sicot, N., Berard, J.Y., Perreux, D. and Varchon, D. (1997).<br />
Experimental study of glass-epoxy composite bars loaded in combined bending and<br />
torsion loads. Fatigue and characterisation of the damage growth. In : Degallaix, S.,<br />
Bathias, C. and Fougères, R. (eds.). International Conference on fatigue of composites.<br />
Proceedings, 3-5 June 1997, Paris, France, La Société Française de Métallurgie et de<br />
Matériaux, pp. 266-273.<br />
[22] Herrington, P.D. and Doucet, A.B. (1992). <strong>Progress</strong>ion of bending fatigue damage<br />
around a discontinuity in glass/epoxy composites. Journal of <strong>Composite</strong> <strong>Materials</strong>,<br />
26(14), 2045-2059.<br />
[23] Chen, A.S. and Matthews, F.L. (1993). Biaxial flexural fatigue of composite plates. In :<br />
Miravete, A. (ed.). ICCM/9 <strong>Composite</strong>s : properties and applications. Volume VI.<br />
Proceedings of the Ninth International Conference on <strong>Composite</strong> <strong>Materials</strong>, 12-16 July<br />
1993, Madrid, Spain, Woodhead Publishing Limited, pp. 899-906.<br />
[24] Sidoroff, F. and Subagio, B. (1987). Fatigue damage modelling of composite materials<br />
from bending tests. In : Matthews, F.L., Buskell, N.C.R., Hodgkinson, J.M. and Morton,<br />
J. (eds.). Sixth International Conference on <strong>Composite</strong> <strong>Materials</strong> (ICCM-VI) & Second<br />
European Conference on <strong>Composite</strong> <strong>Materials</strong> (ECCM-II) : Volume 4. Proceedings, 20-<br />
24 July 1987, London, UK, Elsevier, pp. 4.32-4.39.<br />
[25] El Mahi, A., Berthelot, J.-M. and Bezazi, A. (2002). The fatigue behaviour and damage<br />
development in cross-ply laminates in flexural tests. Proceedings of the Tenth European<br />
Conference on <strong>Composite</strong> <strong>Materials</strong> (ECCM-10), Brugge, Belgium, 3-7 June 2002.<br />
[26] Caprino, G. and D'Amore, A. (1998). Flexural fatigue behaviour of random continuousfibre-reinforced<br />
thermoplastic composites. <strong>Composite</strong>s Science and Technology, 58,<br />
957-965.<br />
[27] Van Paepegem, W. and Degrieck, J. (2002). A New Coupled Approach of Residual<br />
Stiffness and Strength for Fatigue of Fibre-reinforced <strong>Composite</strong>s. International Journal<br />
of Fatigue, 24(7), 747-762.<br />
[28] Van Paepegem, W. and Degrieck, J. (2001). Fatigue Degradation Modelling of Plain<br />
Woven Glass/epoxy <strong>Composite</strong>s. <strong>Composite</strong>s Part A, 32(10), 1433-1441.<br />
[29] Van Paepegem, W. and Degrieck, J. (2002). Modelling damage and permanent strain in<br />
fibre-reinforced composites under in-plane fatigue loading. <strong>Composite</strong>s Science and<br />
Technology, 63(5), 677-694.<br />
[30] Van Paepegem, W. and Degrieck, J. (2005). Simulating Damage and Permanent Strain<br />
in <strong>Composite</strong>s under In-plane Fatigue Loading. Computers & Structures, 83(23-24),<br />
1930-1942.<br />
[31] Lessard, L.B., Eilers, O.P. and Shokrieh, M.M. (1995). Testing of in-plane shear<br />
properties under fatigue loading. Journal of Reinforced Plastics and <strong>Composite</strong>s, 14,<br />
965-987.<br />
[32] Van Paepegem, W., De Baere, I. and Degrieck, J. (2006). Modelling the nonlinear shear<br />
stress-strain response of glass fibre-reinforced composites. Part I: Experimental results.<br />
<strong>Composite</strong>s Science and Technology, 66(10), 1455-1464.<br />
[33] Van Paepegem, W., De Baere, I. and Degrieck, J. (2006). Modelling the nonlinear shear<br />
stress-strain response of glass fibre-reinforced composites. Part II: Model development<br />
and finite element simulations. <strong>Composite</strong>s Science and Technology , 66(10), 1465-<br />
1478.
236<br />
W. Van Paepegem, I. De Baere, E. Lamkanfi et al.<br />
[34] Gagel, A., Fiedler, B. and Schulte, K. (2006). On modelling the mechanical degradation<br />
of fatigue loaded glass-fibre non-crimp fabric reinforced epoxy laminates. <strong>Composite</strong>s<br />
Science and Technology, 66(5), 657-664.<br />
[35] Hull, D. (1999). Fractography: observing, measuring and interpreting fracture surface<br />
topography. Cambridge, Cambridge University Press, 366 pp.<br />
[36] Van Paepegem, W. (2002). Development and finite element implementation of a damage<br />
model for fatigue of fibre-reinforced polymers. Ph.D. thesis. Gent, Belgium, Ghent<br />
University Architectural and Engineering Press (ISBN 90-76714-13-4), 403 p.<br />
[37] Mouritz, A.P. (2003). Non-destructive evaluation of damage accumulation. In: Harris, B.<br />
(ed.). Fatigue in <strong>Composite</strong>s. Cambridge, Woodhead Publishing and CRC Press, 2003,<br />
pp. 242-266.<br />
[38] Audoin, B. and Baste, S. (1994). Ultrasonic Evaluation of Stiffness Tensor Changes and<br />
Associated Anisotropic Damage in a Ceramic Matrix <strong>Composite</strong>. Journal of Applied<br />
Mechanics 61:309-316.<br />
[39] Kriz, R.D. & Stinchcomb, W.W. (1979). Elastic Moduli of Transversely Isotropic<br />
Graphite Fibers and Their <strong>Composite</strong>s. Experimental Mechanics 19:41-49.<br />
[40] Rokhlin, S.I. and Wang, W. (1992). Double throughtransmission bulk wave method for<br />
ultrasonic phase velocity measurements and determination of elastic constants of<br />
composite materials. J. Acoust. Soc. Am. 91:3303-3312.<br />
[41] van Dreumel, W.H. and Speijer, J.L. (1981). Nondestructive <strong>Composite</strong> Laminate<br />
Characterisation by Means of Ultrasonic Polar-Scan. <strong>Materials</strong> Evaluation 39:922-925.<br />
[42] Degrieck J. (1995). Some Possibilities for Non-Destructive Characterisation of<br />
<strong>Composite</strong> Plates by Means of Ultrasonic Polar Scans. Proceedings First Joint Belgian-<br />
Hellenic Conference on Non Destructive Testing, Patras, Greece, 22-23 May 1995.<br />
[43] Maes, K. (1998). Non-destructive evaluation of degradation in a fibre-reinforced plastic.<br />
Master thesis (in Dutch). Ghent University, Ghent, 105 pp.<br />
[44] Cnudde, V., Masschaele, B., Dierick, M., Vlassenbroeck, J., Van Hoorebeke, L. and<br />
Jacobs, P. (2006). Recent progress in X-ray CT as a geosciences tool. Applied<br />
Geochemistry, 21(5), 826-832.<br />
[45] De Baere, I., Van Paepegem, W. and Degrieck, J. (2007). On the design of end tabs for<br />
quasi-static and fatigue testing of fibre-reinforced composites. Accepted for Polymer<br />
<strong>Composite</strong>s.<br />
[46] De Baere, I., Van Paepegem, W. and Degrieck, J. (2007). Design of mechanical clamps<br />
with extra long wedge grips for static and fatigue testing of composite materials in<br />
tension and compression. Accepted for Experimental Techniques.<br />
[47] De Baere, I., Van Paepegem, W. and Degrieck, J. (2007). On the feasibility of a threepoint<br />
bending setup for the validation of (fatigue) damage models for thin composite<br />
laminates. Accepted for Polymer <strong>Composite</strong>s.
In: <strong>Composite</strong> <strong>Materials</strong> <strong>Research</strong> <strong>Progress</strong> ISBN: 1-60021-994-2<br />
Editor: Lucas P. Durand, pp. 237-256 © 2008 Nova Science Publishers, Inc.<br />
Chapter 7<br />
DAMAGE VARIABLES IN IMPACT TESTING<br />
OF COMPOSITE LAMINATES<br />
Maria Pia Cavatorta and Davide Salvatore Paolino<br />
Mechanical Engineering Department – Politecnico di Torino,<br />
Corso Duca degli Abruzzi, 24 – 10129 Torino (Italy)<br />
Abstract<br />
The Chapter presents an overview of the damage variables proposed in the literature over<br />
the years, including a new variable recently introduced by the Authors to specifically address<br />
the problem of thick laminates subject to repeated impacts. Numerous impact data are used as<br />
a basis to address and comment potentials and limitations of the different variables. Impact<br />
data refer to single impact events as well as repeated impact tests performed on laminates with<br />
different fiber and matrix combinations and various lay-ups. Laminates of different thickness<br />
are considered, ranging from tenths to tens of millimeters.<br />
The analysis shows that some of the variables can indeed be used for assessing the<br />
damage tolerance of the laminate. In single impact tests, results point out the existence of an<br />
energy threshold at about 40-50% of the penetration energy, below which the damage threat is<br />
quite negligible. Other variables are not directly related to the amount of damage induced in<br />
the laminate but rather give an indication of the laminate efficiency of energy absorption.<br />
Introduction<br />
<strong>Composite</strong> laminates are expected to absorb low velocity impacts either during assembling or<br />
use. Even when the impact damage is barely visible, the incurred micro-damage may have a<br />
significant effect on the laminate strength and durability. The impact energy can be absorbed<br />
at any point of the laminate, well away from the point of impact, and by means of various<br />
laminate level failure mechanisms including front face indentation (indicative of local matrix<br />
crushing and local fiber breakage), interlaminar delamination, back face splitting and fiber<br />
peeling. In the literature [1-11], it is acknowledged that matrix cracking is the first type of<br />
damage introduced during impact; however, the presence of matrix cracks per se does not<br />
significantly change the overall laminate stiffness. Rather, the matrix crack tips may act as
238<br />
Maria Pia Cavatorta and Davide Salvatore Paolino<br />
initiation point for delaminations and fiber breaks which may dramatically reduce the local<br />
and or global laminate stiffness thus affecting the load-time response. The literature also<br />
acknowledges that more damaging energy absorption mechanisms (such as delamination,<br />
fiber pull-out, fiber/matrix debonding and fiber fracture) follows matrix cracking and that<br />
they significantly reduce the strength and stiffness of the laminate. Considering the<br />
importance of damage assessment, there have been several attempts in the literature to look<br />
for measurable test quantities that could be correlated to the damage process [6, 12-19].<br />
Under low-velocity impact loading conditions, the time of contact between the impactor<br />
and the target is relative long. Even though vibratory load responses from the composite<br />
sample, the impactor and the specimen supporting fixture are common features of impact<br />
loading history, the load history can still yield important information concerning damage<br />
initiation and growth [20]. Several authors have used the force-time history to compare the<br />
structural response from impact tests: in particular, values of the First Damage Force (FDF)<br />
[14-21], the Delamination Threshold Load (DTL) [6] or Hertzian Failure Load (Ph) [15] as<br />
well as of the Peak Force (Fpeak) [2-3, 17, 22-24] have often been used to rank laminate<br />
performance. Identification of the FDF poses no troubles as it corresponds to the first load<br />
drop which can be detected in the load-time history. However, comparison of laminate<br />
performance on such basis can be risky since the level of laminate damage associated with the<br />
first load drop may be quite different for a given laminate tested under different impact<br />
energies, or for different laminates tested at a given impact energy. In this respect, definition<br />
of the DTL and of the Ph appear more suitable for ranking laminate performance as they are<br />
intended to identify a more specific damage condition, that is the initiation of significant<br />
damage. In the case of the DTL, significant damage is defined as predominately delamination,<br />
while for the Ph energy absorption mechanisms other than matrix cracking are considered.<br />
The DTL and Ph do not necessarily correspond to the first load drop; rather, they are<br />
associated to the load drop at which a significant change in the slope of the forcedisplacement<br />
curve may be detected and which signals a change in laminate stiffness.<br />
Experimental determination of these load thresholds, which are shown to vary with the<br />
laminate thickness to the 3/2 power, may prove helpful for damage tolerant design: no<br />
significant damage threat is associated to impact events for which Fpeak is below the laminate<br />
DTL or Ph. On the contrary, for impact events for which Fpeak is above the laminate DTL or<br />
Ph, a damage threat exists, even if no information can be obtained on the final amount of<br />
cumulative damage that will occur.<br />
Difficulties and possible ambiguities in determining the DTL or the Ph have often led<br />
researchers to use Fpeak instead, considering it as the turning point between rather limited and<br />
more significant forms of damage. In [17,25], Liu suggested that for any composite laminate<br />
there exists a maximum value of Fpeak. When the impact energy is such that Fpeak is below this<br />
maximum value, the laminate suffers indentation and local matrix cracking, whereas when<br />
loaded by the maximum Fpeak significant delamination starts to take place.<br />
The idea of Fpeak as the signaling point of significant damage initiation is the basis of the<br />
dimensionless parameter introduced in 1975 by Beaumont et al. [16] called the Ductility<br />
Index (DuI). The DuI, which is proposed as a useful tool for ranking the impact performance<br />
of different materials under similar testing conditions, is defined as the ratio between the<br />
propagation energy Epropagation and the initiation energy Einitiation and it is given by the<br />
expression:
Damage Variables in Impact Testing of <strong>Composite</strong> Laminates 239<br />
E<br />
propagation<br />
DuI = (1)<br />
E<br />
initiation<br />
where Einitiation and Epropagation correspond to the energies absorbed before and after Fpeak,<br />
respectively.<br />
The ductility index is small for brittle materials, where most of the energy is absorbed<br />
before Fpeak, and high for ductile materials, where most of the energy is absorbed after Fpeak is<br />
exceeded. The energy absorption mechanisms before Fpeak are crazing and microcracking of<br />
the matrix; whereas, after Fpeak, crack growth is observed via fiber pull-out, fiber/matrix<br />
debonding and fiber fracture [26-29].<br />
Other energy variables have been introduced since the DuI to rank impact performance.<br />
In [12-13] Belingardi and Vadori introduced the Damage Degree (DD) defined as the ratio<br />
between the absorbed energy (Ea) and the impact energy (Ei). Ei is the kinetic energy of the<br />
impactor right before contact takes place and it is indeed the energy introduced into the<br />
specimen. Ea can be calculated from the force-displacement curve as the area surrounded by<br />
the curve in case of closed force-displacement curves (impact event with rebound) or the area<br />
bounded by the force-displacement curve up to a constant level of force and the horizontal<br />
axis in case of open force-displacement curves (impact event with no rebound). Based on the<br />
energy viewpoint, penetration should take place the first time Ea reaches Ei. Therefore the DD<br />
is below one for impact events with rebound while it reaches the value of one in case the<br />
impactor is stopped with no rebound or specimen penetration is achieved. In [13-14], it was<br />
shown that the relationship between the DD and the impact energy increases monotonically<br />
until saturation and a fairly good data interpolation was achieved by a linear regression curve<br />
[14]. A saturation energy level (Esa) was defined as the impact energy at which the DD<br />
regression curve reaches the value of one. This energy threshold is of practical and theoretical<br />
interest since it defines the maximum energy level the laminate can dissipate with no<br />
penetration and by means of internal damage mechanisms only [12]. In synthesis, the DD is<br />
defined as:<br />
In [17,18], Liu proposed a second-order polynomial regression curve to describe the<br />
absorbed energy vs. impact energy curve up to penetration (named by Liu the energy profile):<br />
2<br />
Ea = aEi<br />
+ bEi<br />
+ c<br />
Depending on the laminate under study, the linear term and the constant c can be smaller<br />
than the quadratic term so that equation (2) can be simplified as:<br />
a<br />
2<br />
i<br />
(2)<br />
E ≅ aE<br />
(3)<br />
From the energy profile, Liu was able to define a Penetration Threshold (Pn) (in a series<br />
of “continuous” impacts at increasing impact energies, it represents the first condition of no<br />
impactor rebound and therefore of equality between impact and absorbed energy), and a<br />
Perforation Threshold (Pr) (first condition of laminate complete perforation). Between the<br />
penetration and perforation thresholds, there exists a range, named by Liu “the range of the<br />
penetration process”, in which the impact energy and the absorbed energy are equal to each
240<br />
Maria Pia Cavatorta and Davide Salvatore Paolino<br />
other but which represent different stages of the penetration process with the impactor<br />
moving deeper and deeper into the specimen as the impact energy increases.<br />
Penetration and perforation thresholds increase with thickness, so does the range of the<br />
penetration process. In other words, while for thin laminates the difference between the<br />
penetration and the perforation thresholds can be negligible, for thick laminates the same can<br />
become quite significant. For cross-ply glass-epoxy composite laminates, Liu [17] found:<br />
Pn<br />
= 0.<br />
8t<br />
P<br />
r<br />
0.<br />
0247<br />
where t is the laminate thickness. Equation (4) indicates that for the investigated glass/epoxy<br />
laminates, the penetration threshold is about 80% of the perforation threshold. In case of 3mm<br />
thin laminates, the range of penetration process (Pr–Pn) is less than 2 J. For 6-mm<br />
laminates, a difference of 15J is found, while for 12-mm thick laminates, (Pr –Pn) exceeds<br />
100J and by far can not be neglected.<br />
In addition to identification of the laminate Pn, Pr and range of penetration process, the<br />
energy profile was used by Liu to define a coefficient η, named the Efficiency of Energy<br />
Absorption. The coefficient is defined as the ratio between the area bounded by the<br />
polynomial regression line of equation (2) up to Pn and the horizontal axis and the area of the<br />
rectangular triangle having for hypotenuse the bisector from zero to Pn. The bisector of the<br />
energy profile represents the equal energy between impact and absorption; therefore, the<br />
triangular zone corresponds to the highest energy-area the material can possibly have. As all<br />
materials have an energy absorption capability less than 100%, the regression curve is always<br />
below the bisector. However, the closer the regression curve to the bisector, the higher the<br />
energy absorption capability of the laminate.<br />
An interesting analysis of the energy profile was provided in [19]. By normalizing the<br />
impact energy and the absorbed energy by the laminate Pn, Mian and Quaresimin were able to<br />
obtain a single master curve which proved to work very well when thin laminates were<br />
investigated. A direct consequence of the existence of a master curve is that, when normalized<br />
by the laminate Pn, the efficiency of energy absorption is basically constant for all laminates,<br />
i.e. η varies linearly with the laminate Pn.<br />
The range of penetration process yet remained to be investigated. To this aim, a new<br />
variable, named the Damage Index (DI), was recently introduced by the Authors [30-32]. The<br />
DI definition aroused considering that in the range of the penetration process, the impactor<br />
moves deeper and deeper into the specimen as the impact energy increases. On the contrary,<br />
pure energy variables as the DD by definition saturates to one over the entire penetration<br />
process.<br />
s<br />
≈<br />
0.<br />
8<br />
(4)<br />
MAX<br />
DI = DD<br />
(5)<br />
sQS<br />
The value sMAX in equation (5) refers to the displacement value recorded at the instant<br />
when the force approximately reaches a constant value, in case of impact tests that cause
Damage Variables in Impact Testing of <strong>Composite</strong> Laminates 241<br />
specimen perforation; while it corresponds to the maximum displacement recorded during the<br />
test below the perforation threshold Pr.<br />
In order to make the DI a non-dimensional quantity, the displacement sMAX was<br />
normalised by the corresponding displacement sQS measured in quasi-static perforation tests.<br />
The choice of normalising by sQS was taken so to define an absolute reference test and leave<br />
apart possible strain-rate effects on the sMAX values. For all the laminates investigated by the<br />
Authors, sMAX of perforation tests was constant regardless of the impact velocity and equal to<br />
the sQS value.<br />
Experimental<br />
Experimental impact tests were performed according to ASTM 3029 standard [33] using an<br />
instrumented free-fall drop dart testing machine. The impactor has a total mass of 20 kg; its<br />
head is hemispherical with a radius of 10 mm. Stainless steel was chosen for its high hardness<br />
and resistance to corrosion. The maximum falling height of the testing machine is 2 m, which<br />
corresponds to a maximum impact energy of 392 J. The drop-weight apparatus was equipped<br />
with a motorized lifting track. By means of a piezoelectric load cell, force-time curves were<br />
acquired. The acceleration history was calculated dividing the force term by the impactor<br />
mass. The displacement was obtained by double integration of the acceleration and thus<br />
force-displacement curves were plotted. By integration of the force-displacement curves,<br />
deformation energy-displacement curves were then obtained. Initial conditions were given<br />
with the time axis having its origin at the time of impact. At time t=0, the dart coordinate is<br />
zero and its initial velocity can be obtained by the well known relationship:<br />
v = 2gΔh<br />
0 (6)<br />
where Δh is defined as the height loss of the centre of mass of the dart with respect to the<br />
reference surface [12]. The impact velocity was also measured by an optoelectronic device.<br />
Agreement between measured and theoretical values was very good.<br />
The collected data were stored after each impact and the impactor was returned to its original<br />
starting height. Using this technique, the chosen impact velocity was consistently obtained in<br />
successive impacts. Because, the target holder was rigidly attached to the frame of the testing<br />
device, the tup struck the specimen each time at the same location.<br />
Square specimen panels, with 100 mm edge, were clamped through rigid plates having a<br />
central hole 76.2 mm in diameter, and fixed to a rigid base to prevent slippage of the<br />
specimen. The clamping system was designed to provide an uniform pressure all over the<br />
clamping area.<br />
Prior to impact tests, a series of quasi-static perforation tests were performed to get<br />
information on the laminate strength characteristics. Specimens were tested using a servohydraulic<br />
machine with maximum loading capacity of 100 kN. The hydraulic actuator was<br />
electronically controlled in order to perform constant velocity tests. Signals of the force<br />
applied by the actuator and of the actuator displacement were acquired in time with an<br />
appropriate sampling rate.<br />
Table 1 reports main characteristics of the laminates used in the study.
242<br />
Maria Pia Cavatorta and Davide Salvatore Paolino<br />
Table 1. Main characteristics of the laminates analyzed in the chapter<br />
Ref. Fiber /Matrix Lay-up<br />
Thickness<br />
[mm]<br />
Acronym<br />
used<br />
[30] Glass/Vinylester [random/02/90/random 12.31 GVP90_12.31<br />
+Polyester<br />
/90/02/random]<br />
[0/90] 15<br />
4.00 GE90s_4.00<br />
[32] Glass/Epoxy<br />
[0/90] 30<br />
8.00 GE90s_8.00<br />
[03 /903] 5<br />
4.00 GE90m_4.00<br />
[03 /903] 10<br />
8.00 GE90m_8.00<br />
[21] Glass/Epoxy [random/-45/+45/02] 2 4.50 GE45_4.50<br />
[0/90] 4<br />
0.35 CE90_0.35<br />
[0/90] 8<br />
0.75 CE90_0.75<br />
[14] Carbon/Epoxy<br />
[0/90] 16<br />
1.55 CE90_1.55<br />
[0/60/-60] 4<br />
0.40 CE60_0.40<br />
[0/60/-60] 8<br />
0.85 CE60_0.85<br />
[0/60/-60] 16<br />
1.75 CE60_1.75<br />
Results for Single Impact Tests<br />
Figures 1-3, 5-6 reports the results obtained for the analyzed laminates. For sake of<br />
comparison among the different laminates, in all graphs the impact energy Ei is divided by the<br />
penetration threshold Pn [34].<br />
F peak/F pen<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0<br />
E i/P n<br />
GVP90_12.31<br />
GE90s_8.00<br />
GE90m_8.00<br />
GE45_4.50<br />
GE90s_4.00<br />
GE90m_4.00<br />
CE60_1.75<br />
CE90_1.55<br />
CE60_0.85<br />
CE90_0.75<br />
CE60_0.40<br />
CE90_0.35<br />
1st linear trend<br />
2nd linear trend<br />
polynomial trend<br />
Figure 1. Normalized peak force plotted against non-dimensional impact energy E i/P n. Linear and<br />
polynomial trends.<br />
Figure 1 reports data for Fpeak vs. the non-dimensional impact energy Ei/Pn. Values of<br />
Fpeak are normalized by the peak force Fpen registered for an impact energy equal to Pn.
Damage Variables in Impact Testing of <strong>Composite</strong> Laminates 243<br />
Considering the two-order difference in laminate thickness, data scattering is fairly limited.<br />
Data for the thinner laminates are the most dispersed and show the lowest values. In [35], the<br />
impact force was shown to depend on the flexibility of the laminate: values of Fpeak decrease<br />
with increasing laminate flexibility.<br />
A general trend for Fpeak can be envisaged. In [36], Found et al. proposed a relationship<br />
between Fpeak and the square root of the impact velocity, i.e. between Fpeak and the impact<br />
energy to the ¼ power. The proposed relationship (dotted curve in Figure 1) well interpolates<br />
the experimental data.<br />
Interpolation by two straight lines also appears rather good, allowing to point out that, for<br />
impact energies above 40%-50% Pn, the rate of increase of Fpeak with increasing impact<br />
energies slows down, with the value of Fpeak approaching an asymptote. The asymptotic trend<br />
of Figure 1 well agrees with the idea of a maximum value for Fpeak [17,25]. In this respect,<br />
data of Figure 1 seems to suggest that no real damage threat is associated to impact events for<br />
which the impact energy is below 40%-50% of the laminate Pn. A concept of impact threshold<br />
energy has been put forward by many researchers [35, 37-39]. This threshold has been<br />
defined as a measure of the ability of a composite laminate to resist initial strength<br />
degradation [35].<br />
DD<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0<br />
E i/P n<br />
GVP90_12.31 GE90s_8.00<br />
GE90m_8.00 GE45_4.50<br />
GE90s_4.00 GE90m_4.00<br />
CE60_1.75 CE90_1.55<br />
CE60_0.85 CE90_0.75<br />
CE60_0.40 CE90_0.35<br />
Figure 2. DD values plotted against non-dimensional impact energy E i/P n.<br />
Figure 2 reports data for the DD, which appear fairly more dispersed, apart from the data<br />
points of the three thicker laminates (GVP90_12.31, GE90s_8.00; GE90m_8.00) that are<br />
basically overlapping. As a general rule, it can be said that the DD increases for increasing<br />
impact energies and shows notably higher values for thicker laminates. In this respect, it is<br />
important to note that high values of the DD do not imply severe damage within the<br />
laminates. Indeed, the DD is a measure of the percentage of impact energy absorbed by the<br />
laminate whereas no distinction is made on the absorption mechanisms as it is the case for the<br />
DuI. High values of the absorption energy Ea can indeed be desirable, for example in crash
244<br />
Maria Pia Cavatorta and Davide Salvatore Paolino<br />
events [40]. Strictly speaking, the DD is not a damage variable but rather a point by point<br />
measurement of the laminate efficiency of energy absorption, whereas the coefficient η<br />
proposed by Liu averages the efficiency of energy absorption over a wide range of impact<br />
energies (from very low energies to Pn).<br />
E a/P n<br />
1<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
GVP90_12.31<br />
GE90s_8.00<br />
GE90m_8.00<br />
GE45_4.50<br />
GE90s_4.00<br />
GE90m_4.00<br />
CE60_1.75<br />
CE90_1.55<br />
CE60_0.85<br />
CE90_0.75<br />
CE60_0.40<br />
CE90_0.35<br />
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0<br />
E i/P n<br />
Figure 3. Energy absorption Master Curve.<br />
Figure 3 plots the Master Curve proposed in [19]. Data for the three thicker laminates are<br />
again superimposed and very close to the bisector, meaning that the absorbed energy is about<br />
equal the impact energy (DD almost one). For thin laminates, the difference is more enhanced<br />
pointing out a lower efficiency of energy absorption [3].<br />
As said, no indications can be evinced from Figures 2 and 3 in terms of laminate damage<br />
tolerance. In this respect, definition of the DuI, which differentiates between the nature of<br />
energy absorption mechanisms, appears rather appealing. In its original definition, the DuI is<br />
meant to rank different laminates tested under similar impact conditions, on the basis of a<br />
more fragile or more ductile behavior under impact loading. It is worthwhile noticing that in<br />
the literature the DuI is basically used to rank different laminates at perforation or fracture<br />
(Charpy tests), for which determination of Einitiation and Epropagation poses no trouble. The forcedisplacement<br />
curves are open curves and Einitiation is calculated without ambiguity as the area<br />
bounded by the force-displacement curve up to Fpeak and the horizontal axis, while Epropagation<br />
is calculated as the area bounded by the force-displacement curve from Fpeak to a constant<br />
level of force and the horizontal axis (Figure 4a). In the attempt of proving the DuI against<br />
other damage variables, computation of the DuI was also applied to impact tests with<br />
rebound. When the dart rebounds, the force-displacement curves are closed curves and<br />
computation of Einitiation and Epropagation becomes troublesome as different areas could be<br />
considered. No references of DuI computation in impact tests with rebound were found in the<br />
literature. Considering that in impact events with rebound the energy absorbed by the<br />
laminate is equal to the area bounded by the force-displacement curve, in computing the DuI
Damage Variables in Impact Testing of <strong>Composite</strong> Laminates 245<br />
it was decided to calculate Einitiation as the area bounded by the force-displacement curve up to<br />
Fpeak and Epropagation as the area bounded by the force-displacement curve after Fpeak so that the<br />
sum of the two energies is equal to the overall energy absorbed by the laminate (Figure 4b).<br />
In this way, only the portion of impact energy in fact dissipated by the laminate was taken<br />
into account and differentiated by the nature of energy absorption mechanisms.<br />
Force [kN]<br />
12<br />
10<br />
8<br />
6<br />
4<br />
2<br />
0<br />
Initiation<br />
Energy<br />
F peak<br />
0 5 10 15 20 25<br />
Displacement [mm]<br />
Propagation<br />
Energy<br />
Figure 4a. An example of force-displacement curve with perforation: filled areas correspond to E initiation<br />
(Initiation Energy) and Epropagation (Propagation Energy).<br />
Force [kN]<br />
30<br />
25<br />
20<br />
15<br />
10<br />
5<br />
0<br />
Initiation<br />
Energy<br />
0 2 4 6 8 10<br />
Displacement [mm]<br />
F peak<br />
Propagation<br />
Energy<br />
Figure 4b. An example of force-displacement curve with rebound: filled areas correspond to E initiation<br />
(Initiation Energy) and Epropagation (Propagation Energy).
246<br />
DuI<br />
2.5<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
0.0<br />
Maria Pia Cavatorta and Davide Salvatore Paolino<br />
GVP90_12.31<br />
GE90s_8.00<br />
GE90m_8.00<br />
GE45_4.50<br />
GE90s_4.00<br />
GE90m_4.00<br />
CE60_1.75<br />
CE90_1.55<br />
CE60_0.85<br />
CE90_0.75<br />
CE60_0.40<br />
CE90_0.35<br />
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0<br />
E i/P n<br />
Figure 5. DuI values plotted against non-dimensional impact energy E i/P n.<br />
Figure 5 reports data for the DuI plotted against the non-dimensional impact energy Ei/Pn.<br />
By looking at the DuI values at penetration (Ei=Pn), thicker laminates appear to exhibit a<br />
more ductile behavior. However, considering the elevated heterogeneity of the laminates<br />
under study (in terms of type of fiber and matrix, orientation and percentage of fibers as well<br />
as laminate thickness), it is the Authors’ opinion that caution must be taken in ranking<br />
laminate performance. It should also be reminded that for the thicker laminates, penetration<br />
and perforation thresholds do not coincide but are quite distant from each other. Significance<br />
of Figure 5 is to show that, by extending computation of the DuI to impact energies below Pn,<br />
the DuI can be used as a damage variable. In particular, data on Figure 5 show that for impact<br />
energies up to 40% Pn, the amount of Epropagation is almost null meaning that the main energy<br />
absorbing mechanism is matrix cracking. Above 40% Pn, and especially in the case of thick<br />
laminates, the contribution of Epropagation becomes more and more important. This implies that<br />
Fpeak occurs at a value of displacement significantly lower than the maximum displacement<br />
reached by the laminate before dart rebound. As the impact energy increases, contribution of<br />
delamination and of fiber breakage to the energy absorption mechanisms becomes more and<br />
more important.<br />
Figure 6 reports data in terms of the DI. By taking into account the value of the maximum<br />
displacement, the DI is more of a damage variable than the DD was. Indeed, Figure 6 shows<br />
that at very low impact energies the DI is almost null to then increase monotonically for<br />
increasing impact energies. Up to impact energies of about 40-50% Pn, signaled by graphs of<br />
Fpeak and DuI as energy thresholds, the difference in the value of the DI for different<br />
laminates is quite limited. Above this threshold, the difference in DI data for different<br />
laminates increases significantly. Data on Figure 6 also show that DI values do not saturate to<br />
one at Pn, thus allowing to monitor the range of the penetration process. The effectiveness of<br />
the variable in distinguishing between the penetration and the perforation thresholds can be<br />
evinced from Figures 7 and 8 which report DD and DI data for two different laminates.
Damage Variables in Impact Testing of <strong>Composite</strong> Laminates 247<br />
Interestingly, up to Pn the DI increases linearly with the impact energy to then grow quite<br />
abruptly over the range of the penetration process. A linear relationship also exists between<br />
the DD data and the impact energy; however, the DD data can not be used beyond penetration<br />
as (by definition) the DD stays at the value of one over the entire range of the penetration<br />
process.<br />
DI<br />
DD, DI<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
GVP90_12.31 GE90s_8.00<br />
GE90m_8.00 GE45_4.50<br />
GE90s_4.00 GE90m_4.00<br />
CE60_1.75 CE90_1.55<br />
CE60_0.85 CE90_0.75<br />
CE60_0.40 CE90_0.35<br />
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0<br />
E i/P n<br />
Figure 6. DI values plotted against non-dimensional impact energy E i/P n<br />
y = 0.68x + 0.28<br />
R 2 = 0.96<br />
y = 0.52x - 0.14<br />
R 2 = 0.99<br />
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0<br />
E i/P n<br />
DD DI<br />
Figure 7. Comparison between DD and DI values for impact tests on glass/epoxy 6.25 mm thick<br />
laminates. Impact data taken from reference [17].
248<br />
DD, DI<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
Maria Pia Cavatorta and Davide Salvatore Paolino<br />
y = 0.80x + 0.21<br />
R 2 = 0.90<br />
y = 0.58x - 0.01<br />
R 2 = 0.98<br />
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6<br />
E i/P n<br />
DD DI<br />
Figure 8. Comparison between DD and DI values for impact tests on GE45_4.50 laminates.<br />
Results for Repeated Impact Tests<br />
Apart from monitoring the range of the penetration process, the DI has proven to provide<br />
important pieces of information in case of repeated impact tests, a loading conditions of<br />
particular relevance in marine applications [4,30-32,38-39,41-42]. Figures 9-14 report data<br />
obtained on two different laminates (GVP90_12_31, CE90m_4.00) tested under repeated<br />
impacts. The two depicted impact energies were selected to represent tests of no laminate<br />
perforation within 40 impacts and tests of laminate perforation. Figures 9-10 reports data for<br />
Fpeak. As it can be observed, for impact energies that cause no perforation within test duration,<br />
values of Fpeak slightly increase in the first few impacts to then reach an asymptote. On the<br />
contrary, for energies that cause perforation, values of Fpeak decrease impact after impact as a<br />
consequence of damage accumulation. For a given laminate, initial values of Fpeak reported in<br />
Figure 10 are obviously higher than those of Figure 9 due to the higher impact energy used in<br />
the test (Figure 1), while values just before perforation are lower than the asymptotic values<br />
of Figure 9 due to the significant damage induced in the laminate. With respect to the initial<br />
values of Fpeak, it is worthwhile noticing that for the 25 J tests and the 98 J test performed on<br />
the GVP90_12.31 laminate, the maximum in Fpeak is not reached at the first impact. This<br />
effect has already been observed in the literature. In a series of repeated impact tests run on<br />
carbon/epoxy composite laminate, Wyrick and Adams [22] commented the initial increase in<br />
Fpeak as the result of the compaction process of the thin layer of unreinforced resin at the<br />
impacted surface. At low impact energy levels, damage to the fibers near the surface is<br />
minimal and the compaction process provides a harder surface for the next impact. In this<br />
respect, it is worthwhile noticing that this initial increase in Fpeak was observed when the<br />
impact energy was below 40% Pn, once more confirming the existence of an energy threshold<br />
level.
F peak [kN]<br />
16<br />
14<br />
12<br />
10<br />
8<br />
6<br />
4<br />
2<br />
0<br />
Damage Variables in Impact Testing of <strong>Composite</strong> Laminates 249<br />
1 4 7 10 13 16 19 22 25 28 31 34 37 40<br />
Impact Number<br />
GVP90_12.31<br />
CE90m_4.00<br />
Figure 9. Values of peak force in repeated impact tests performed on GVP90_12.31 and CE90m_4.00<br />
laminates. Impact energy: 25 J.<br />
F peak [kN]<br />
30<br />
25<br />
20<br />
15<br />
10<br />
5<br />
0<br />
1 3 5 7 9 11 13 15 17 19<br />
Impact Number<br />
GVP90_12.31<br />
CE90m_4.00<br />
Figure 10. Values of peak force in repeated impact tests performed on GVP90_12.31 and CE90m_4.00<br />
laminates. Impact energy: 98 J.<br />
Figures 11-12 report data in terms of the DuI. As in Figure 5, for impact tests with<br />
rebound, contribution of Einitiation and Epropagation is calculated according to the definition<br />
illustrated in Figure 4b. Figure 11 shows that for impact energies that cause no perforation,<br />
the DuI is very low and constant throughout the test, signaling no significant damage
250<br />
Maria Pia Cavatorta and Davide Salvatore Paolino<br />
accumulation. For impact energies that cause perforation, the DuI maintains a very low value<br />
up to a few impacts before perforation when it rapidly increases (Figure 12).<br />
DuI<br />
0.10<br />
0.08<br />
0.06<br />
0.04<br />
0.02<br />
0.00<br />
1 4 7 10 13 16 19 22 25 28 31 34 37 40<br />
Impact Number<br />
GVP90_12.31<br />
CE90m_4.00<br />
Figure 11. DuI data for repeated impact tests performed on GVP90_12.31 and CE90m_4.00 laminates.<br />
Impact energy: 24.5 J.<br />
DuI<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
0.0<br />
1 3 5 7 9 11 13 15 17 19<br />
Impact Number<br />
GVP90_12.31<br />
CE90m_4.00<br />
Figure 12. DuI data for repeated impact tests performed on GVP90_12.31 and CE90m_4.00 laminates.<br />
Impact energy: 98 J.<br />
Data in terms of DD and DI are reported in Figures 13 and 14. To avoid confusion, data<br />
are organized for single impact energies and single laminates. DD and DI data are plotted<br />
together to favor a comparison between the two variables. For energies that cause no
Damage Variables in Impact Testing of <strong>Composite</strong> Laminates 251<br />
perforation (Figure 13), the DD data show an initial decrease thus suggesting a reduction in<br />
the percentage of impact energy that the laminate is able to absorb. For the thicker laminate<br />
(GVP90_12.31), this initial reduction is quite significant, going from a percentage of energy<br />
absorption of about 85% in the first impact to a quite stable value of about 70% in subsequent<br />
tests. Visual observation of the laminate after each impact pointed out that in the first few<br />
impacts the impactor indents the laminate and that the size of this indentation does not<br />
significantly change in subsequent tests. Existence of an initial localized damage that does not<br />
appear to significantly grow in subsequent tests is also what is conveyed by the DuI as well as<br />
the DI data, which keep to a constant low level throughout the test.<br />
DD, DI<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
DD DI<br />
1 4 7 10 13 16 19 22 25 28 31 34 37 40<br />
Impact Number<br />
Figure 13a. Comparison between DD and DI data in repeated impact tests performed on CE90m_4.00<br />
laminates. Impact energy: 24.5 J.<br />
DD, DI<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
DD DI<br />
1 4 7 10 13 16 19 22 25 28 31 34 37 40<br />
Impact Number<br />
Figure 13b. Comparison between DD and DI data in repeated impact tests performed on GVP90_12.31<br />
laminates. Impact energy: 24.5 J.
252<br />
DD, DI<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
Maria Pia Cavatorta and Davide Salvatore Paolino<br />
y = 1.1E-01x + 3.2E-01<br />
R 2 = 1.00<br />
1 2 3 4 5<br />
Impact Number<br />
DD DI<br />
Figure 14a. Comparison between DD and DI data in repeated impact tests performed on CE90m_4.00<br />
laminates. Impact energy: 98 J.<br />
DD, DI<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
y = 9.3E-03x + 1.9E-01<br />
R 2 = 0.96<br />
1 3 5 7 9 11 13 15 17 19<br />
Impact Number<br />
DD DI<br />
Figure 14b. Comparison between DD and DI data in repeated impact tests performed on GVP90_12.31<br />
laminates. Impact energy: 98 J.<br />
Figure 14a reports data for the CE90m_4.00 laminate tested at an impact energy of 98J.<br />
Perforation is achieved at the 5 th impact. As it can be observed, the DD constantly increases<br />
impact after impact to reach a value of about one at the 4 th impact, where laminate penetration<br />
is achieved. DI values increase at a constant rate up to the 3 rd impact to then grow very<br />
rapidly and reach the value of one at laminate perforation. This trend is more evident in<br />
Figure 14b, for which perforation is achieved at the 19 th impact. Up to the 17 th impact, the DI<br />
slowly increases at a constant rate impact after impact. In the last two impacts before
Damage Variables in Impact Testing of <strong>Composite</strong> Laminates 253<br />
perforation, the increase is on the contrary very rapid. Differently from the DuI which keeps<br />
to a constant low level up to a few impacts before perforation (Figure 12), the DI allows to<br />
monitor the initial phase of steady damage accumulation helping foreseen perforation. The<br />
initial slow decrease of DD values from the first to the second impact is followed by a<br />
constant phase up to the 17 th impact, after which the DD increases rapidly and reaches a value<br />
of one at perforation. Likewise the DuI, DD data are not very sensitive for predicting laminate<br />
perforation as, apart from the last 2-3 impacts, the constant phase of Figure 14b does not<br />
differ from the asymptotic trends of Figures 13a and 13b, where no perforation is achieved.<br />
Also, DD values at the first impact are about the same, regardless of the level of impact<br />
energy.<br />
Conclusion<br />
Impact test data obtained on different laminates are used to compare damage variables which<br />
have been proposed in the literature over the years. To this aim, definition of the two energy<br />
contributions used to compute the DuI has been extended to analyze impact tests with<br />
rebound.<br />
In single impact tests performed at different impact energies, data for Fpeak and DuI point<br />
out the existence of an impact energy threshold at about 40-50% Pn, below which the energy<br />
absorption mechanism is mainly matrix cracking. Graphs of Fpeak vs. impact energy show a<br />
bi-linear trend with a change in slope around the energy threshold; while values of the DuI are<br />
almost null below the energy threshold to then increase quite abruptly up to penetration.<br />
Interestingly, the energy threshold is about the same for all the laminates analyzed in the<br />
study, whose thickness varies from tenths to tens of millimeters. DI data increase<br />
monotonically for increasing impact energies and show very limited scattering up to the<br />
energy threshold. DD values and data in the Master Curve give no indications on the laminate<br />
damage tolerance; rather, they provide a measure of the absorption capability of the laminate.<br />
Results show that thicker laminates are characterized by a higher efficiency of energy<br />
absorption.<br />
Main advantage of the DI variable is the possibility to distinguish between the<br />
penetration and perforation energy thresholds. The distinction is essential when dealing with<br />
thick laminates, for which the impact energy that causes laminate perforation can by far<br />
exceed the penetration energy. In the range of the penetration process, the DI effectively<br />
monitors the impactor moving deeper and deeper into the laminate.<br />
Also in case of repeated impact tests, the DI provides important pieces of information.<br />
For impact energies that cause no laminate perforation within test duration, the DI stays at a<br />
constant low value throughout the test, owing to a negligible damage accumulation besides<br />
initial laminate indentation. For impact energies that cause perforation, the DI shows an initial<br />
phase of linear growth with the number of impacts, owing to a steady accumulation of<br />
damage. A few impacts before perforation, the DI starts raising quite abruptly, helping<br />
foreseeing laminate failure.<br />
Results for Fpeak show that it maintains a constant value when perforation is not achieved<br />
while it decreases rapidly otherwise. However, graphs of Fpeak versus impact number do not<br />
signal any change in the rate of damage accumulation. DuI values are almost null throughout<br />
the test when no perforation occurs. Low and constant values also characterize tests at higher
254<br />
Maria Pia Cavatorta and Davide Salvatore Paolino<br />
energies up to a few impacts before perforation, when data show a rapid increase. Therefore,<br />
by looking at the DuI values in the first impacts, no prediction can be made as to whether or<br />
not laminate perforation will occur within test duration. In case of repeated impacts, DD data<br />
can monitor the efficiency of energy absorption impact after impact. Results show that DD<br />
values at the first impact are about the same, regardless of the impact energy. Moreover,<br />
regardless of the final output of the test (no perforation/perforation), DD values show an<br />
initial slight decrease followed by a rather constant phase. When perforation is to be achieved,<br />
DD values start to increase a few impacts before final failure.<br />
Acknowledgments<br />
The Authors wish to acknowledge valuable discussions with professor Giovanni Belingardi.<br />
References<br />
[1] Roy, R; Sarkar BK; Bose NR. Impact fatigue of glass fibre-vinylester resin composites.<br />
<strong>Composite</strong>s Part A, 2001, 32, 871-876.<br />
[2] Zhou, G. Static behaviour and damage of thick composite laminates. <strong>Composite</strong><br />
Structures, 1996, 36, 13-22.<br />
[3] Zhou, G; Davies, GAO. Impact response of thick reinforced polyester laminates.<br />
International Journal of Impact Engineering, 1995, 3, 357-374.<br />
[4] Sutherland, LS; Guedes Soares, C. Effects of laminate thickness and reinforcement type<br />
on the impact behaviour of E-glass/polyester laminates. <strong>Composite</strong>s Science and<br />
Technology, 1999, 59, 2243-2260.<br />
[5] Azouaoui, K; Rechak, S; Azari, Z; Benmedakhene, S; Laksimi, A; Pluvinage, G.<br />
Modelling of damage and failure of glass/epoxy composite plates subject to impact<br />
fatigue. International Journal of Fatigue, 2001, 23, 877-885.<br />
[6] Schoeppner, GA; Abrate, S. Delamination threshold loads for low velocity impact on<br />
composite laminates. <strong>Composite</strong>s Part A, 2000, 31, 903-915.<br />
[8] Guden, M; Yildirim, U; Hall, IW. Effect of strain rate on the compression behaviour of a<br />
woven glass fiber/SC-15 composite. Polymer Testing, 2004, 23, 719-725.<br />
[9] Baucom, JN; Zikry, MA. Low-velocity impact damage progression in woven E-glass<br />
composite systems. <strong>Composite</strong>s Part A, 2005, 36, 658-664.<br />
[10] Ambu, R; Aymerich, F; Ginesu, F; Priolo, P. Assessment of NDT interferometric<br />
techniques for impact damage detection in composite laminates. <strong>Composite</strong>s Science and<br />
Technology, 2006, 66, 199-205.<br />
[11] Okoli, OI. The effect of strain rate and failure modes on the failure energy of fibre<br />
reinforced composites. <strong>Composite</strong> Structures, 2001, 54, 199-303.<br />
[12] Belingardi, G; Grasso, F; Vadori, R. Energy absorption and damage degree in impact<br />
testing of composite materials. Proceedings XI ICEM (Int. Conf. Experimental<br />
Mechanics), Oxford (UK), 1998, 279-285.<br />
[13] Belingardi, G; Vadori, R. Low velocity impact tests of laminate glass-fiber-epoxy matrix<br />
composite materials plates. International Journal of Impact Engineering, 2002, 27,<br />
213-229.
Damage Variables in Impact Testing of <strong>Composite</strong> Laminates 255<br />
[14] Belingardi, G; Vadori, R. Influence of the laminate thickness in low velocity impact<br />
behaviour of composite material plate. <strong>Composite</strong> Structures, 2003, 61, 27-38.<br />
[15] Shyr, TW; Pan YH. Impact resistance and damage characteristics of composite<br />
laminates. <strong>Composite</strong> Structures, 2003, 62, 193-203.<br />
[16] Beaumont, PWR; Reiwald, PG; Zweben, C. Methods for improving the impact<br />
resistance of composite materials. in Foreign object impact behaviour of composites.<br />
ASTM Spec Tech Publ, 1974, 568, 134–58.<br />
[17] Liu, D. Characterization of Impact Properties and Damage Process of Glass/Epoxy<br />
<strong>Composite</strong> Laminates. Journal of <strong>Composite</strong> <strong>Materials</strong>, 2004, 38, 1425-1442.<br />
[18] Liu, D; Raju, BB; Dang, X. Size effects on impact response of composite laminates.<br />
International Journal of Impact Engineering, 1988, 21, 837-854.<br />
[19] Mian, S; Quaresimin, M. A model for the energy absorption capability of composite<br />
laminates. Proceedings 8 th ASME Conference on ESDA (Engineering Systems Design<br />
and Analysis), Turin (Italy), July 4-7, 2006, Paper Number: 95788.<br />
[20] Lee, SM; Zahuta, P. Instrumented impact and static indentation of composites. Journal<br />
of <strong>Composite</strong> <strong>Materials</strong>, 1991, 25, 204-222.<br />
[21] Belingardi, G; Cavatorta, MP; Duella, R. Material characterization of a composite-foam<br />
sandwich for the front structure of a high speed train. <strong>Composite</strong> Structures, 2003, 61,<br />
13-25.<br />
[22] Wyrick, DA; Adams, DF. Residual Strength of a Carbon/Epoxy <strong>Composite</strong> Material<br />
Subjected to Repeated Impact. Journal of <strong>Composite</strong> <strong>Materials</strong>, 1988, 22, 749-765.<br />
[23] Caprino, G; Lopresto, V; Scarponi, C; Briotti, G. Influence of material thickness on the<br />
response of carbon-fabric/epoxy panel to low velocity impact. <strong>Composite</strong>s Science and<br />
Technology, 1999, 59, 2279-2286.<br />
[24] Zhou, G; Davies, GAO. Characterization of thick glass woven roving/polyester<br />
laminates: 1. Tension, compression and shear. <strong>Composite</strong>s, 1995, 26, 579-586.<br />
[25] Liu, D; Raju, BB; Dang, X. Impact perforation resistance of laminated and assembled<br />
composites plates. International Journal of Impact Engineering, 2000, 24, 733-746.<br />
[26] Harmia, T; Friedrich, K. Mechanical and thermomechanical properties of discontinuous<br />
long glass fiber reinforced PA66/PP blends. Plastics, Rubber and <strong>Composite</strong>s<br />
Processing and Applications, 1995, 23, 63-69.<br />
[27] Kishore; Kulkarni, SM; Sharathchandra, S. Sunil, D. On the use of an instrumented setup<br />
to characterize the impact behaviour of an epoxy system containing varying fly ash<br />
content. Polymer Testing, 2002, 21, 763-771.<br />
[28] Jang, J; Han, S. Mechanical properties of glass-fibre mat/PMMA functionally gradient<br />
composite. <strong>Composite</strong>s Part A, 1999, 30, 1045-1053.<br />
[29] Pegoretti, A; Zanolli, A; Migliaresi, C. Flexural and interlaminar mechanical properties<br />
of unidirectional liquid cristalline single-polymer composites. <strong>Composite</strong>s Science and<br />
Technology, 2006, 66, 1953-1962.<br />
[30] Belingardi, G; Cavatorta, MP; Paolino, DS. Comparative Response in Repeated Impact<br />
Tests of Hand Lay-up and Vacuum Infusion Glass Reinforced <strong>Composite</strong>s. International<br />
Journal of Impact Engineering, 2007. doi:10.1016/j.ijimpeng.2007.02.005<br />
[31] Belingardi, G, Cavatorta, MP; Paolino, DS. A new damage index to monitor the range of<br />
the penetration process in thick laminates. <strong>Composite</strong>s Science and Technology. In press.
256<br />
Maria Pia Cavatorta and Davide Salvatore Paolino<br />
[32] Belingardi, G, Cavatorta, MP; Paolino, DS. On the rate of growth and extent of the<br />
steady damage accumulation phase in repeated impact tests. <strong>Composite</strong>s Science and<br />
Technology. Submitted.<br />
[33] ASTM D3029. Standard Test Method for Impact Resistance of Rigid Plastic Sheeting or<br />
Parts by means of a Tup (Falling Weight). American Society for Testing <strong>Materials</strong>,<br />
1982.<br />
[34] Lopresto, V; Melito, V; Leone, C; Caprino, G. Effect of stitches on the impact behaviour<br />
of graphite/epoxy composites. <strong>Composite</strong>s Science and Technology, 2006, 66, 206-214.<br />
[35] Ying, Y. Analysis of the impact threshold energy for carbon fiber and fabric reinforced<br />
composites. Journal of Reinforced Plastics and <strong>Composite</strong>s, 1998, 17, 1056-1075.<br />
[36] Found, MS; Howard, IC. Single and multiple impact behaviour of a CFRP laminate.<br />
<strong>Composite</strong> Structures, 1995, 32, 159-163.<br />
[37] Luo, RK; Green, ER; Morrison, CJ. An approach to evaluate the impact damage<br />
initiation and propagation in composite plates. <strong>Composite</strong>s Part B, 2001, 32, 513-520.<br />
[38] Supratik Datta; Vamsee Krishna, A; Rao RMVGK. Low Velocity Impact Damage<br />
Tolerances Studies on Glass-Epoxy Laminates – Effects of Material, Process and Test<br />
Parameters. Journal of Reinforced Plastics and <strong>Composite</strong>s, 2004, 23, 327-345.<br />
[39] Jang, BP; Huang, CT; Hsieh, CY; Kowbel W; Jang, BZ. Repeated Impact Failure of<br />
Continuous Fiber Reinforced Thermoplastic and Thermoset <strong>Composite</strong>s. Journal of<br />
<strong>Composite</strong> <strong>Materials</strong>, 1991, 25, 1171-1203.<br />
[40] Sutherland, LS; Guedes Soares, C. Effect of laminate thickness and of matrix resin on<br />
the impact of low fibre-volume, woven roving E-glass composites. <strong>Composite</strong>s Science<br />
and Technology, 2004, 64, 1691-1700.<br />
[41] Kawaguchi, T; Nishimura, H; Ito, K; Sorimachi, H; Kuriyama, T; Narisawa, I. Impact<br />
fatigue properties of glass fiber- reinforced thermoplastics. <strong>Composite</strong> Science and<br />
Technology, 2004, 64, 1057-1067.<br />
[42] Bijoysri Kahn; Rao, RMVGK; Venkataraman, N. Low Velocity Fatigue Studies on<br />
Glass Epoxy <strong>Composite</strong>s Laminates with Varied Material and Test Parameters – Effect<br />
of Incident energy and Fibre Volume Fraction. Journal of Reinforced Plastics and<br />
<strong>Composite</strong>s, 1995, 14, 1150-1159.
In: <strong>Composite</strong> <strong>Materials</strong> <strong>Research</strong> <strong>Progress</strong><br />
Editor: Lucas P. Durand, pp. 257-273<br />
Chapter 8<br />
ISBN 1-60021-994-2<br />
c○ 2008 Nova Science Publishers, Inc.<br />
ELECTROMECHANICAL FIELD CONCENTRATIONS<br />
AND POLARIZATION SWITCHING BY ELECTRODES<br />
IN PIEZOELECTRIC COMPOSITES<br />
Yasuhide Shindo and Fumio Narita<br />
Department of <strong>Materials</strong> Processing, Graduate School of Engineering,<br />
Tohoku University<br />
Abstract<br />
The electromechanical field concentrations due to electrodes in piezoelectric composites<br />
are investigated through numerical and experimental characterization. This<br />
work consists of two parts. In the first part, a nonlinear finite element analysis is carried<br />
out to discuss the electromechanical fields in rectangular piezoelectric composite<br />
actuators with partial electrodes, by introducing models for polarization switching in<br />
local areas of the field concentrations. Two criteria based on the work done by electromechanical<br />
loads and the internal energy density are used. Strain measurements are<br />
also presented for a four layered piezoelectric actuator, and a comparison of the predictions<br />
with experimental data is conducted. In the second part, the electromechanical<br />
fields in the neighborhood of circular electrodes in piezoelectric disk composites are<br />
reported. Nonlinear disk device behavior induced by localized polarization switching<br />
is discussed.<br />
1. Introduction<br />
Sensor and actuator applications take advantage of the piezoelectric coupling converting<br />
electrical energy into mechanical energy and vice versa. Piezoelectric ceramics and composites<br />
play a significant role as active electronic components in many areas of science and<br />
technology, such as smart and MEMS devices. In some actuator applications, high values<br />
of stress and electric field arise in the neighborhood of an electrode tip in piezoelectric<br />
ceramics [1] and composites [2], and the field concentrations can result in electromechanical<br />
degradation [3, 4]. One of the limitations for practical use of piezoelectric ceramics<br />
and composites is also their nonlinear behavior, which occurs due to polarization switching<br />
and/or domain wall motion at high electromechanical field levels near the electrode<br />
tip. In order to optimize the performance of the piezoelectric devices, it is important to
258 Yasuhide Shindo and Fumio Narita<br />
understand the electromechanical field concentrations due to electrodes in piezoelectric ceramics<br />
and composites. Recently, Yoshida et al. [5] discussed the electromechanical field<br />
concentrations due to circular electrodes in piezoelectric ceramics through theoretical and<br />
experimental characterizations. Their model quantitatively predicted the nonlinear electromechanical<br />
fields induced by polarization switching near the circular electrode tip. Also,<br />
numerical predictions of strain concentration were in relatively good agreement with measured<br />
values.<br />
The main aim of this work is to evaluate the electromechanical fields in the neighborhood<br />
of surface and internal electrodes in piezoelectric composites. First, we study the<br />
effect of applied voltage on the electromechanical field concentrations near the electrodes<br />
in rectangular piezoelectric composite actuators. A nonlinear finite element analysis is performed<br />
to calculate the strain, stress, electric field and electric displacement by introducing<br />
models for polarization switching in local areas of the field concentrations. Two criteria<br />
based on the work done by electromechanical loads and the internal energy density are<br />
used and compared. Strain measurements are also presented to validate the predictions<br />
using a four layered piezoelectric actuator. A comparison of strain concentration is made<br />
between measurements and calculations, and a nonlinear behavior induced by localized polarization<br />
switching is discussed. The device performance and polarization switching zone<br />
near the electrodes are further predicted for some electrode configurations in the rectangular<br />
piezoelectric composites. Next, we discuss the electromechanical field concentrations due<br />
to circular electrodes in piezoelectric disk composites. The effects of applied voltage and<br />
localized polarization switching on the disk device performance are examined.<br />
2. Basic Equations<br />
Consider a piezoelectric material with no body force and free charge. The governing equations<br />
in the Cartesian coordinates xi(i =1, 2, 3) are given by<br />
σji,j =0 (1)<br />
Di,i =0 (2)<br />
where σij is the stress tensor, Di is the electric displacement vector, a comma denotes<br />
partial differentiation with respect to the coordinate xi, and the Einstein summation convention<br />
over repeated indices is used. The relation between the strain tensor εij and the<br />
displacement vector ui is given by<br />
εij = 1<br />
2 (uj,i<br />
and the electric field intensity vector is<br />
+ ui,j) (3)<br />
Ei = −φ,i (4)<br />
where φ is the electric potential. In a ferroelectric, polarization switching leads to a change<br />
in the remanent strain εr r<br />
ij and remanent polarization Pi . The total strain and electric displacement<br />
are<br />
εij = ε l ij + ε r ij (5)<br />
Di = D l i + P r<br />
i<br />
(6)
Eletromechanical Field Concentrations and Polarization Switching... 259<br />
where the superscript l denotes the linear contribution to the strain and electric displacement,<br />
and the linear piezoelectric relationships are given by<br />
ε l ij = sijklσkl + dkijEk (7)<br />
D l i = diklσkl + ɛikEk (8)<br />
In Eqs. (7) and (8), sijkl, dkij and ɛik are the elastic compliance tensor, direct piezoelectric<br />
tensor and dielectric permittivity tensor, which satisfy the following symmetry relations:<br />
sijkl = sjikl = sijlk = sklij, dkij = dkji, ɛik = ɛki<br />
σij and D l i are related to εl ij and Ei by<br />
σij = cijklε l kl − ekijEk (10)<br />
D l i = eiklε l kl + ɛikEk (11)<br />
where cijkl and eikl are the elastic and piezoelectric tensors, and<br />
cijkl = cjikl = cijlk = cklij, ekij = ekji<br />
For piezoceramics which exhibit symmetry of a hexagonal crystal of class 6 mm with respect<br />
to principal x1,x2, and x3 axes, the constitutive relations can be written in the following<br />
form:<br />
⎧<br />
⎪⎨<br />
⎪⎩<br />
⎧<br />
⎪⎨<br />
⎪⎩<br />
σ1<br />
σ2<br />
σ3<br />
σ4<br />
σ5<br />
σ6<br />
where<br />
D l 1<br />
D l 2<br />
D l 3<br />
⎫<br />
⎡<br />
⎢<br />
⎪⎬<br />
⎢<br />
= ⎢<br />
⎣<br />
⎪⎭<br />
⎫<br />
⎪⎬<br />
⎪⎭ =<br />
⎡<br />
⎢<br />
⎣<br />
c11 c12 c13 0 0 0<br />
c12 c11 c13 0 0 0<br />
c13 c13 c33 0 0 0<br />
0 0 0 c44 0 0<br />
0 0 0 0 c44 0<br />
0 0 0 0 0 c66<br />
0 0 0 0 e15 0<br />
0 0 0 e15 0 0<br />
e31 e31 e33 0 0 0<br />
⎤ ⎧<br />
⎥ ⎪⎨<br />
⎥<br />
⎦<br />
⎪⎩<br />
σ1 = σ11, σ2 = σ22, σ3 = σ33<br />
ε l 1<br />
ε l 2<br />
ε l 3<br />
ε l 4<br />
ε l 5<br />
ε l 6<br />
(9)<br />
(12)<br />
⎫ ⎡<br />
0<br />
⎢<br />
⎪⎬<br />
⎢ 0<br />
⎢ 0<br />
− ⎢ 0<br />
⎢<br />
⎣<br />
⎪⎭<br />
e15<br />
0<br />
0<br />
0<br />
0<br />
e15<br />
0<br />
0<br />
e31<br />
e31<br />
e33<br />
0<br />
0<br />
0<br />
⎤<br />
⎥ ⎧<br />
⎥ ⎪⎨<br />
⎥ ⎪⎩ ⎥<br />
⎦<br />
E1<br />
E2<br />
E3<br />
⎫<br />
⎪⎬<br />
⎪⎭<br />
(13)<br />
⎧<br />
ε<br />
⎤<br />
⎪⎨<br />
⎥<br />
⎦<br />
⎪⎩<br />
l 1<br />
εl 2<br />
εl 3<br />
εl 4<br />
εl 5<br />
εl ⎫<br />
⎡<br />
⎤ ⎧ ⎫<br />
⎪⎬ ɛ11 0 0 ⎪⎨ E1 ⎪⎬<br />
⎢<br />
⎥<br />
+ ⎣ 0 ɛ11 0 ⎦ E2<br />
⎪⎩ ⎪⎭<br />
0 0 ɛ33 E3<br />
⎪⎭<br />
6<br />
(14)<br />
σ4 = σ23 = σ32, σ5 = σ31 = σ13, σ6 = σ12 = σ21<br />
ε l 1 = ε l 11, ε l 2 = ε l 22, ε l 3 = ε l 33<br />
ε l 4 =2εl 23 =2εl 32 ,εl 5 =2εl 31 =2εl 13 ,εl 6 =2εl 12 =2εl 21<br />
�<br />
�<br />
(15)<br />
(16)
260 Yasuhide Shindo and Fumio Narita<br />
c11 = c1111 = c2222, c12 = c1122, c13 = c1133 = c2233, c33 = c3333<br />
c44 = c2323 = c3131, c66 = c1212 = 1<br />
2 (c11 − c12)<br />
e15 = e131 = e223, e31 = e311 = e322, e33 = e333<br />
The direction of the spontaneous polarization P s of each grain can change by 90 ◦ or<br />
180 ◦ for ferroelectric switching induced by a sufficiently large electric field. In order to<br />
develop a non-linear model incorporating the polarization switching mechanisms with the<br />
electromechanical fields calculations, two criteria are used. The first criterion for polarization<br />
switching is based on work down, and the second is internal energy density switching<br />
criterion.<br />
The first criterion [6] states that a polarization switches when the electrical and mechanical<br />
work exceeds a critical value<br />
σij∆εij + Ei∆Pi ≥ 2P s Ec<br />
where ∆εij and ∆Pi are the changes in the spontaneous strain and polarization during<br />
switching, respectively, and Ec is a coercive electric field. The changes in ∆εij = ε r ij and<br />
∆Pi = P r<br />
i for 180◦ switching can be expressed as<br />
⎫<br />
⎬<br />
⎭<br />
(17)<br />
(18)<br />
(19)<br />
∆ε11 =0, ∆ε22 =0, ∆ε33 =0, ∆ε12 =0, ∆ε23 =0, ∆ε31 =0 (20)<br />
∆P1 =0, ∆P2 =0, ∆P3 = −2P s (21)<br />
For 90 ◦ switching in the x3x1 plane, there results<br />
∆ε11 = γ s , ∆ε22 =0, ∆ε33 = −γ s , ∆ε12 =0, ∆ε23 =0, ∆ε31 =0 (22)<br />
∆P1 = ±P s , ∆P2 =0, ∆P3 = −P s<br />
For 90 ◦ switching in the x2x3 plane,<br />
(23)<br />
∆ε11 =0, ∆ε22 = γ s , ∆ε33 = −γ s , ∆ε12 =0, ∆ε23 =0, ∆ε31 =0 (24)<br />
∆P1 =0, ∆P2 = ±P s , ∆P3 = −P s<br />
The polarization switching criterion based on internal energy density (second criterion)<br />
[7] is defined as<br />
U = Uc<br />
where U is the internal energy density and Uc is a critical value of internal energy density<br />
corresponding to the switching mode. The internal energy density associated with 180 ◦<br />
switching can be written as<br />
U = 1<br />
2 D3E3<br />
In the case of 90 ◦ switching in the x3x1 plane, the internal energy density is<br />
(25)<br />
(26)<br />
(27)<br />
U = 1<br />
2 (σ11ε11 + σ33ε33 +2σ31ε31 + D1E1) (28)
Eletromechanical Field Concentrations and Polarization Switching... 261<br />
For 90 ◦ switching in the x2x3 plane,<br />
U = 1<br />
2 (σ22ε22 + σ33ε33 +2σ32ε32 + D2E2) (29)<br />
The critical value of internal energy density is assumed in the following form:<br />
where ɛ T 33<br />
Uc = 1<br />
2 ɛT 33(Ec) 2<br />
is the dielectric permittivity at constant stress.<br />
The constitutive equations (10) and (11) during polarization switching are<br />
The new piezoelectric constant e ′ ikl<br />
by<br />
(30)<br />
σij = cijklε l kl − e ′ kijEk (31)<br />
D l i = e ′ iklε l kl + ɛikEk (32)<br />
is related to the elastic and direct piezoelectric constants<br />
e ′ 111 = d′ 111 c11 + d ′ 122 c12 + d ′ 133 c13<br />
e ′ 122 = d ′ 111c12 + d ′ 122c11 + d ′ 133c13<br />
e ′ 133 = d′ 111 c13 + d ′ 122 c13 + d ′ 133 c33<br />
e ′ 123 =2d ′ 123c44<br />
e ′ 131 =2d′ 131 c44<br />
e ′ 112 =2d ′ 112c66<br />
e ′ 211 = d′ 211 c11 + d ′ 222 c12 + d ′ 233 c13<br />
e ′ 222 = d ′ 211c12 + d ′ 222c11 + d ′ 233c13<br />
e ′ 233 = d′ 211 c13 + d ′ 222 c13 + d ′ 233 c33<br />
e ′ 223 =2d ′ 223c44<br />
e ′ 231 =2d′ 231 c44<br />
e ′ 212 =2d′ 212 c66<br />
e ′ 311 = d ′ 311c11 + d ′ 322c12 + d ′ 333c13<br />
e ′ 322 = d′ 311 c12 + d ′ 322 c11 + d ′ 333 c13<br />
e ′ 333 = d ′ 311c13 + d ′ 322c13 + d ′ 333c33<br />
e ′ 323 =2d′ 323 c44<br />
e ′ 331 =2d ′ 331c44<br />
e ′ 312 =2d′ 312 c66<br />
The components of the piezoelectricity tensor d ′ ikl are<br />
d ′<br />
�<br />
ikl = d33ninknl + d31(niδil − ninknl)+ 1<br />
2 d15(δiknl<br />
�<br />
− 2ninknl + δilnk)<br />
where ni is the unit vector in the poling direction, δij is the Kroneker delta, and d33 = d333,<br />
d31 = d311, d15 =2d131 are the direct piezoelectric constants.<br />
(33)<br />
(34)
262 Yasuhide Shindo and Fumio Narita<br />
3. Rectangular Piezoelectric <strong>Composite</strong> Actuators<br />
3.1. Computational Model<br />
We performed 3D finite element calculations to present the electromechanical fields distributions<br />
around the electrode tip. The geometry used was a four layered piezoelectric<br />
composite actuator, as shown in Fig. 1. A rectangular Cartesian coordinate system ( x,y,z)<br />
is used with the z-axis coinciding with the poling direction. Electrodes with length a and<br />
width W are embedded in the piezoelectric actuator of length L and width W . An external<br />
electrode is attached on both sides of the actuator to address each electrode. The<br />
thickness of the layer h is chosen, and a L − a tab region exists on both sides of the layer.<br />
The total thickness is 4h. Because of the geometric and loading symmetry, only a half<br />
of the specimen needs to be analyzed. The electric potential on two electrode surfaces<br />
(−L/2 ≤ x ≤−L/2 +a, |y| ≤W/2, z =0, 2h) equals the applied voltage, φ = V0.<br />
The electrode surface (L/2 − a ≤ x ≤ L/2, |y| ≤W/2, z = h) is connected to the<br />
ground, so that φ =0. The normal displacement and shear stress on the surface (|x| ≤L/2,<br />
y = −W/2, |z| ≤2h) are zero. The surface (|x| ≤L/2, y = W/2, |z| ≤2h) is stress free.<br />
W<br />
h<br />
h<br />
Electrode<br />
Poling<br />
O<br />
y<br />
z<br />
O<br />
a<br />
a<br />
L<br />
a<br />
Figure 1. A rectangular piezoelectric composite actuator.<br />
Each element consists of many grains, and each grain is modeled as a uniformly polarized<br />
cell that contains a single domain. The model neglects the domain wall effects and<br />
interaction among different domains. In reality, this is not true, but the assumption does<br />
not affect the general conclusions drawn. The polarization of each grain initially aligns as<br />
closely as possible with the z- direction. The polarization switching is defined for each<br />
x<br />
x
Eletromechanical Field Concentrations and Polarization Switching... 263<br />
element in a material. The electric potential φ is applied, and the electromechanical fields<br />
of each element are computed from the finite element analysis (FEA). The switching criterion<br />
of Eq. (19) or (26) is checked for every element to see if switching will occur. After<br />
all possible polarization switches have occurred, the piezoelectric tensor of each element<br />
is rotated to the new polarization direction. The electroelastic fields are re-calculated, and<br />
the process is repeated until the solution converges. The macroscopic response of the material<br />
is determined by the finite element model, which is an aggregate of elements. The<br />
spontaneous polarization P s and strain γ s are assigned representative values of 0.3 C/m 2<br />
and 0.004, respectively. Our previous experiments [8] verified the accuracy of the above<br />
scheme, and showed that the results obtained are of general applicability.<br />
3.2. Experiments<br />
The actuator discussed in this section was fabricated using a soft lead zirconate titanate<br />
(PZT) C-91 [9]. The material properties are listed in Table 1, and the corcive electric field<br />
is approximately Ec = 0.35 MV/m. The dimensions of the specimen are L = 30 mm, W =<br />
10 mm, and 4h = 20 mm. The electrode length is a = 20 mm. The specimen was placed<br />
on the rigid floor.<br />
The high-voltage amplifier was limited to 1.25 kV so that a 0.25 MV/m field corresponded<br />
to a layer thickness of 5.0 mm. Strain gauges were placed around the electrode tip<br />
region. The sensors have an active length of 0.2 mm.<br />
Table 1. Material properties of C-91.<br />
Elastic stiffnesses Piezoelectric coefficients Dielectric permittivities<br />
(×1010N/m2 ) (C/m2 ) (×10−10C/Vm) c11 c12 c13 c33 c44 e31 e33 e15 ɛ11 ɛ33<br />
C-91 12.0 7.7 7.7 11.4 2.4 −17.3 21.2 20.2 226 235<br />
3.3. Results and Discussion<br />
We first present analytical and experimental results for L = 30 mm, W = 10 mm, and<br />
4h =20mm. The electrode length is a = 20 mm. Fig. 2 shows the finite element analysis<br />
results for the strain εzz versus electric field E0 = V0/h at the face of the actuator (at<br />
y =5mm plane) for x = 5 mm and z = 0.8 mm. For the polarization switching effect, the<br />
predictions based on work (Eq. (19)) and energy density (Eq. (26)) are shown. Also plotted<br />
are the experimental data in the range approximately ± 0.18 MV/m. Calculation results<br />
show that a monotonically increasing negative electric field causes polarization reversal.<br />
Polarization switching in a local region leads to a significant increase of compressive strain<br />
within the actuator when compared to the linear case. After the electric field reaches about<br />
−0.20 (0.24) MV/m, local polarization switching, based on work (energy density), can<br />
cause an unexpected decrease in compressive strain near the electrode tip during switching.
264 Yasuhide Shindo and Fumio Narita<br />
Strain,� zz (�10 -6 )<br />
200<br />
100<br />
0<br />
-100<br />
-200<br />
L = 30 m m<br />
W = 10 m m<br />
4h = 20 m m<br />
a = 20 m m<br />
Test<br />
FEA<br />
x = 5.0 m m<br />
y = 5.0 m m<br />
z = 0.8 m m<br />
W ork<br />
Energy density<br />
-0.3 -0.2 -0.1 0 0.1 0.2<br />
Electric field, E0 (M V/m )<br />
Figure 2. Strain versus electric field for laminated actuator.<br />
E = -0.34 M V/m<br />
0<br />
W ork<br />
Energy density<br />
Poling<br />
o<br />
90 switching<br />
o<br />
180 switching<br />
Figure 3. Polarization switching zone induced by electric field for laminated actuator.<br />
Fig. 3 shows the 180 ◦ and 90 ◦ switching zones near the electrode tip of the actuator under<br />
E0 = −0.34 MV/m. Predictions by different criteria are presented. 90 ◦ switching zone<br />
based on energy density are larger than that based on work. Fig. 4 displays the distribution<br />
of the normal component of stress σzz as a function of x at y =0mm and z =0, 1 and<br />
9 mm for laminated actuator with L = 30 mm, W = 10 mm, 4h =20mm and a =20<br />
mm under E0 =0.2 MV/m from the finite element analysis. The solid line represents the
Eletromechanical Field Concentrations and Polarization Switching... 265<br />
normal stress at the interface, the dashed line represents the normal stress near the internal<br />
electrode tip, and the alternate long and short dashed line denotes the value near the surface<br />
electrode tip. The normal stress at the interface is singular at the electrode tip. The stress<br />
ahead of the electrode tip is tensile, while the stress behind the electrode tip is compressive.<br />
The values of the normal stress near the internal electrode tip are higher than those near the<br />
Displacem ent,u z (�m )<br />
2<br />
1<br />
0<br />
-1<br />
-2<br />
FEA<br />
W ork<br />
x = 0 m m<br />
y = 0 m m<br />
z = 10 m m<br />
L = 30 m m<br />
W = 10 m m<br />
4h = 20 m m<br />
a = 20 m m<br />
24<br />
28<br />
-1000 0 1000<br />
Voltage,V 0 (V)<br />
Figure 4. Normal stress versus x for laminated actuator under E0 =0.2 MV/m.<br />
surface electrode tip. Fig. 5 shows the comparison of the distribution of the shear stress σzx<br />
against x near the internal electrode tip with that near the surface electrode tip for the same<br />
laminated actuator. The shear stress peaks at about x =5.5 mm, and the peak value near<br />
the internal electrode tip is higher than that near the surface electrode tip.<br />
Surface electrode<br />
h<br />
h<br />
x<br />
z<br />
b<br />
O<br />
b �<br />
b<br />
c<br />
r<br />
Internal electrode<br />
Figure 5. Shear stress versus x for laminated actuator under E0 =0.2 MV/m.<br />
y
266 Yasuhide Shindo and Fumio Narita<br />
!<br />
( )<br />
"# $<br />
%& ' &<br />
Figure 6. Displacement versus voltage for laminated actuator.<br />
Next, the effect of electrode length on the performance of the actuator with L = 30 mm,<br />
W = 10 mm and 4h = 20 mm is discussed. Fig. 6 shows the predictions of displacement<br />
uz at x =0mm, y =0mm and z =10mm as a function of applied voltage V0, based on<br />
work, for a =20, 24 and 28 mm. Non-linearity in the displacement versus voltage curves<br />
depends on the electrode length. The displacement increases with an increase of a from 20<br />
mm to 24 mm. Little difference is observed between the results for a =24mm and 28 mm.<br />
4. Piezoelectric Disk <strong>Composite</strong>s<br />
4.1. Problem Statement and Solution Procedure<br />
Consider a two-layered piezoelectric disk composite with radius c and thickness h as shown<br />
in Fig. 7. The origin of the coordinates (r,θ,z) is located at the center of the interface<br />
considered as z =0, 0 ≤ r ≤ c. The z axis is assumed to coincide with the six fold<br />
axis of symmetry in the class of a 6mm crystal class, or with the poling axis in the case<br />
of poled piezoelectric ceramics. Three parallel circular electrodes of radius b lie in the<br />
planes z =0, ±h. The bonding at z =0is assumed to be perfect so that the stresses and<br />
displacements are continuous along the interface of the composite. Let the voltage applied<br />
to the internal electrode surface be denoted by V0. The surface electrodes are grounded.<br />
The axisymmetric models were generated using the commercial FE method software<br />
package. The electrode layers were not incorporated into the model.
Eletromechanical Field Concentrations and Polarization Switching... 267<br />
" # $ %" #<br />
Figure 7. A piezoelectric disk composite actuator.<br />
4.2. Numerical Results and Discussion<br />
We consider C-91 + /C-91 − and C-91 + /C-91 + with c =10mm and 2h =2mm, corresponding<br />
to the tension and bending actuator models, respectively. The superscripts − and<br />
+ denote, respectively, the situations for negative and positive poling directions.<br />
Plotted in Fig. 8 are the numerical values of radial strain εrr near the circular internal<br />
electrode tip (r =9mm and z =0.2 mm) as a function of electric field E0 = V0/h for<br />
Strain,� rr (�10 -6 )<br />
60<br />
40<br />
20<br />
0<br />
C-91 + /C-91 +<br />
c= 10 m m<br />
2h = 2 m m<br />
b = 8 m m FEA<br />
W ork<br />
Energy density<br />
-0.2 0 0.2<br />
Electric field, E0 (M V/m )<br />
!<br />
r= 9 m m<br />
z= 0.2 m m<br />
Figure 8. Strain versus electric field for disk composite tension actuator.
268 Yasuhide Shindo and Fumio Narita<br />
E 0=<br />
-0.22 M V/m<br />
-0.30 M V/m<br />
Poling<br />
0.5 m m<br />
+0.22 M V/m<br />
+0.30 M V/m<br />
W ork<br />
o<br />
90 switching<br />
o<br />
180 switching<br />
Figure 9. Normal stress versus r for disk composite tension actuator under E0 =0.2<br />
MV/m.<br />
C-91 + /C-91 − disk tension actuator with b =8mm. The predictions based on work (Eq.<br />
(19)) and energy density (Eq. (26)) are shown. As the electric field is reduced from zero,<br />
the compressive strain increases. Local polarization switching can cause a decrease in<br />
compressive strain near the circular electrode tip. Little difference is observed between<br />
two criteria. As the positive electric field increases, polarization switching did not occur.<br />
Fig. 9 shows the distribution of the normal stress σzz as a function of r at z =0and 0.2<br />
mm for C-91 + /C-91 − disk tension actuator with b =8mm under E0 =0.2 MV/m. Near<br />
! " # $! " #<br />
Figure 10. Shear stress versus r for disk composite tension actuator under E0 = 0.2
MV/m.<br />
Eletromechanical Field Concentrations and Polarization Switching... 269<br />
" # $ %" #<br />
Figure 11. Displacement versus voltage for disk composite tension actuator.<br />
the circular electrode tip, the normal stress at the interface is singular, and the stress ahead of<br />
the circular electrode tip is tensile, while the stress behind the electrode tip is compressive.<br />
The normal stress, apart from the interface, near the electrode tip has smaller value than the<br />
interface stress. Fig. 10 gives the distribution of the shear stress σzr as a function of r at<br />
z =0.01 and 0.2 mm for the same disk tension actuator. The magnitudes of the shear stress<br />
increase toward the circular electrode tip as is expected. Fig. 11 shows the predictions of<br />
displacement uz at r =0mm and z =1mm as a function of applied voltage V0, based<br />
on work, for b =8and 10 mm. There is a small influence of the electrode radius on the<br />
displacement versus voltage curves.<br />
Fig. 12 shows the computed strain εrr of C-91 + /C-91 + disk bending actuator corresponding<br />
to Fig. 8. The negative electric field increases the compressive strain, similar to<br />
C-91 + /C-91 − disk tension actuator. After the electric field reaches about −0.25 MV/m,<br />
polarization switching leads to a decrease in the compressive strain. As the electric field<br />
E0 continues to be reduced, the strain becomes tensile. On the other hand, as the positive<br />
electric field is increased, the strain near the electrode tip increases gradually due to the<br />
piezoelectric effect and then sharply increases as switching occurs due to electromechanical<br />
field concentrations. Little difference is observed between two criteria. Fig. 13 shows<br />
the predicted switching zones, based on work, near the circular electrode tip. As the electric<br />
fields increase, the area of the switched region grows. Fig. 14 shows the normal stress<br />
distribution σzz as a function of r at z =0and 0.2 mm for C-91 + /C-91 + disk bending<br />
actuator with b =8mm. The interface normal stress of the disk bending actuator is singular<br />
at the circular electrode tip, similar to the disk tension actuator. The stress ahead of<br />
the circular electrode tip is tensile, while the stress behind the circular electrode tip changes<br />
from tensile to compressive in the neighborhood of the electrode tip.<br />
!
270 Yasuhide Shindo and Fumio Narita<br />
" # $ " # $<br />
%<br />
%<br />
% & ' (<br />
Figure 12. Strain versus electric field for disk composite bending actuator.<br />
Figure 13. Polarization switching zone induced by electric field for disk composite bending<br />
actuator.<br />
!<br />
% #<br />
%
Eletromechanical Field Concentrations and Polarization Switching... 271<br />
Figure 14. Normal stress versus r for disk composite bending actuator under E0 =0.2<br />
MV/m.<br />
Figure 15. Shear stress versus r for disk composite bending actuator under E0 =0.2<br />
MV/m.<br />
!
272 Yasuhide Shindo and Fumio Narita<br />
! " # $! " #<br />
Figure 16. Tip deflection versus voltage for disk composite bending actuator.<br />
Fig. 15 shows the similar results for the shear stress distribution σzr. A singularity in the<br />
interface shear stress also develops at the circular electrode tip. Note that for the C-91 + /C-<br />
91 + disk bending actuator, since the problem is unsymmetrical to the r-axis, the shear stress<br />
does not become zero along the whole r-axis. Fig.16 gives a plot of the tip deflection uz at<br />
r =10mm and z =0mm with applied voltage V0, based on work, for C-91 + /C-91 + disk<br />
bending actuator with b =8and 10 mm. The curve rises steeply at first when the voltage is<br />
increased from zero. The tip deflection then gradually levels off when the voltage reaches<br />
about 220 V, because of switching in the lower layer (see Fig. 13). A similar phenomenon<br />
can be observed for negative voltage. The bending actuator for b =10mm exhibits higher<br />
deflection.<br />
5. Conclusions<br />
The electromechanical field distributionsin the neighborhood of the electrodes in piezoelectric<br />
composites were investigated. Two criteria for polarization switching in piezoelectric<br />
materials were incorporated into a finite element procedure. The results indicated that high<br />
values of electromechanical fields cause the localized polarization switching near the electrode<br />
tip, and the strain vs electric field curves show the non-linear behavior. Also, the<br />
size of the switching zone in the piezoelectric composites increased with increasing electric<br />
fields. As a remark, we note that this study may be useful in designing advanced piezoelectric<br />
composite actuators.
Eletromechanical Field Concentrations and Polarization Switching... 273<br />
Acknowledgements<br />
This work was partially supported by the Grant-in-Aid for Scientific <strong>Research</strong> (B) and<br />
Young Scientists (B) from the Ministry of Education, Culture, Sports, Science and Technology,<br />
Japan.<br />
References<br />
[1] Shindo, Y., Narita, F. & Sosa, H. (1998). Electroelastic analysis of piezoelectric ceramics<br />
with surface electrodes. Int. J. Eng. Sci., 36, 1001-1009.<br />
[2] Narita, F., Yoshida, M. & Shindo, Y. (2004). Electroelastic effect induced by electrode<br />
embedded at the interface of two piezoelectric half-planes. Mech. Mater., 36, 999-<br />
1006.<br />
[3] Dos Santos e Lucato, S. L., Lupascu, D. C., Kamlah, M., Rödel, J. & Lynch, C. S.<br />
(2001). Constraint-induced crack initiation at electrode edges in piezoelectric ceramics.<br />
Acta Mater., 49, 2751-2759.<br />
[4] Qiu, W., Kang, Y.-L., Qin, Q.-H., Sun, Q.-C. & Xu, F.-Y. (2007). Study for multilayer<br />
piezoelectric composite structure as displacement actuator by Moiré interferometry<br />
and infrared thermography experiments. Mater. Sci. Eng. A, 15, 452-453.<br />
[5] Yoshida, M., Narita, F., Shindo, Y., Karaiwa, M. & Horiguchi, K. (2003). Electroelastic<br />
field concentration by circular electrodes in piezoelectric ceramics. Smart Mater.<br />
Struct., 12, 972-978.<br />
[6] Hwang, S. C., Lynch, C. S. & McMeeking, R. M. (1995). Ferroelectric/ferroelastic<br />
interactions and a polarization switching model. Acta Metall. Mater., 43, 2073-2084.<br />
[7] Kalyanam, S. & Sun, C. T. (2005). Modeling of electrical boundary condition and<br />
domain switching in piezoelectric materials. Mech. Mater., 37, 769-784.<br />
[8] Narita, F., Shindo, Y. & Hayashi, K. (2005). Bending and polarization switching of<br />
piezoelectric laminated actuators under electromechanical loading. Comput. Struct.,<br />
83, 1164-1170.<br />
[9] Shindo, Y., Yoshida, M., Narita, F. & Horiguchi, K. (2004). Electroelastic field concentrations<br />
ahead of electrodes in multilayer piezoelectric actuators: experiment and<br />
finite element simulation. J. Mech. Phys. Solids, 52, 1109-1124.
In: <strong>Composite</strong> <strong>Materials</strong> <strong>Research</strong> <strong>Progress</strong> ISBN: 1-60021-994-2<br />
Editor: Lucas P. Durand, pp. 275-296 © 2008 Nova Science Publishers, Inc.<br />
Chapter 9<br />
RECENT ADVANCES IN DISCONTINUOUSLY<br />
REINFORCED ALUMINUM BASED METAL MATRIX<br />
NANOCOMPOSITES<br />
S.C. Tjong *<br />
Department of Physics and <strong>Materials</strong> Science, City University of Hong Kong,<br />
Tat Chee Avenue, Kowloon, Hong Kong<br />
Abstract<br />
Aluminum-based alloys reinforced with ceramic microparticles are attractive materials<br />
for many structural applications. However, large ceramic microparticles often act as stress<br />
concentrators in the composites during mechanical loading, giving rise to failure of materials<br />
via particle cracking. In recent years, increasing demand for high performance materials has<br />
led to the development of aluminum-based nanocomposites having functions and properties<br />
that are not achievable with monolithic materials and microcomposites. The incorporation of<br />
very low volume contents of ceramic reinforcements on a nanometer scale into aluminumbased<br />
alloys yields remarkable mechanical properties such as high tensile stiffness and<br />
strength as well as excellent creep resistance. However, agglomeration of nanoparticles occurs<br />
readily during the composite fabrication, leading to inferior mechanical performance of<br />
nanocomposites with higher filler content. Cryomilling and severe plastic deformation<br />
processes have emerged as the two important processes to form ultrafine grained composites<br />
with homogeneous dispersion of reinforcing particles. In the present review article, recent<br />
development in the processing, structure and mechanical properties of the aluminum-based<br />
nanocomposites are addressed and discussed.<br />
Introduction<br />
Discontinuously reinforced aluminum (DRA) based metal matrix composites are of<br />
increasing interest because of their high specific stiffness and strength, high isotropic and<br />
excellent wear resistance as well as cost effective manufacturing. DRA composites have been<br />
* E-mail address: aptjong@cityu.edu.hk
276<br />
S.C. Tjong<br />
developed in the past two decades for various automobile, aerospace, electronic packaging<br />
and other structural applications. Many factors affect the mechanical properties of DRA<br />
composites including matrix alloy composition, reinforcement material, reinforcement size,<br />
shape, volume fraction and distribution, nature of the matrix-reinforcement interface, etc. The<br />
reinforcement materials generally should possess significantly higher specific and specific<br />
strength, as well as high melting temperature compared to the matrix alloy. Ceramic<br />
reinforcement has the advantage of a relatively low density and high elastic modulus. Typical<br />
ceramic particles commonly used to reinforce aluminum and its alloys including SiC, B4C,<br />
Si3N4, AlN, Al2O3, TiC, TiB2, etc Particle reinforced composites are conventionally prepared<br />
either via powder metallurgy (PM) or liquid metallurgy, in which the reinforcing particles<br />
with sizes of several microns are directly incorporated into solid or liquid aluminum,<br />
respectively. The composites thus prepared can be viewed as ex-situ MMCs. However,<br />
ceramic microparticles fracture readily during mechanical loading, leading to low toughness<br />
of the composites [1-3]. Figs. 1(a)-1(b) show typical fracture morphology of ceramic<br />
microparticles in Al-based composites during tensile loading. Furthermore, reinforcement<br />
material such as SiC is not thermodynamically stable and thus can react with aluminum<br />
matrix during the composite fabrication and service at elevated temperatures. Efforts have<br />
been made to overcome the occurrence of such difficulties by developing novel in-situ<br />
processing. In the process, the reinforcing particles are directly formed in a metallic matrix by<br />
chemical reactions between constituent elements during the composite fabrication [4, 5].<br />
Accordingly, very fine in-situ particles with diameters down to submicrometer scale ( > 100<br />
nm) can be synthesized and dispersed more uniformly within aluminum matrix [6]. The<br />
formation of clean, ultrafine and thermally stable ceramic reinforcements rendering the in-situ<br />
composites exhibit excellent mechanical properties.<br />
Figure 1. SEM fractographs showing fracture and decohesion of alumina particles of (a) 6061<br />
Al/20vol.%Al2O 3 and (b) 7005 Al /10vol.% Al 2O 3 composites tensile tested at room temperature [3].<br />
The successful synthesis of large-scale ceramic, metallic and intermetallic nanoparticles<br />
in recent years has motivated materials scientists to develop novel metal-matrix<br />
nanocomposites with excellent mechanical properties for advanced structural engineering
Recent Advances in Discontinuously Reinforced Aluminum… 277<br />
applications. Nanoparticles can be synthesized by several processes such as gas phase<br />
condensation, laser ablation, aerosol route, mechanochemical processing is well established<br />
[5, 7-9]. They reveal unique physical and mechanical properties that are different from those<br />
of bulk solids and microparticles. Due to their high specific surface area, nanoparticles exhibit<br />
a high reactivity and strong tendency towards agglomeration. It is necessary to disperse exsitu<br />
nanoparticles more uniformly in aluminum matrix in order to obtain desired mechanical<br />
properties. In the case of liquid metallurgy processing, high-intensity ultrasonic waves can be<br />
employed to disperse the SiC nanoparticles more uniformly in molten aluminum alloy [10,<br />
11]. In powder metallurgy route, mechanical alloying, particularly cryomilling has been used<br />
to refine and disperse the ceramic phase in the Al matrix [12 -16]. In most cases, ceramic<br />
particles with original sizes of several micrometers can be reduced to nanometer level after<br />
cryomilling [17].<br />
Recently, there has been a growing interest in the application of severe plastic<br />
deformation (SPD) such as high pressure torsion (HPT) and equal channel angular pressing<br />
(ECAP) for producing materials with ultrafine grain structure in submicrometer levels [18 -<br />
29]. ECAP is more attractive for industrial applications because it can be employed to<br />
produce large fully-dense samples or products. It consists of pressing the sample through a<br />
die into an L-shaped channel without changing its cross-section. The sample deforms by<br />
simple shear, thereby inducing a high density of dislocations that are subsequently arranged to<br />
the meta-stable sub-grains of high-angle boundaries. By repeating the pressing process, the<br />
strain is accumulated during each increment cycle. The ultra-fine grained composites<br />
processed by ECAP exhibit high yield strength and good ductility [27].<br />
Agglomeration of Particles<br />
Generally, ceramic particles of micrometer sizes are prone to cluster during the composite<br />
fabrication. Particle clustering is more prevalent in cast than in PM microcomposites [30, 31].<br />
This leads to the mechanical properties of microcomposites are far below the theoretical<br />
values. For the PM microcomposites, the particle size ratio of the matrix and reinforcement is<br />
the main factor controlling the degree of microstructural homogeneity [32-35]. Furthermore,<br />
secondary processing technique such as ECAP and HPT are reported to be very effective to<br />
improve the dispersion of reinforcing ceramic particles in the PM DRA composites [20, 24,<br />
27]. Figs. 2(a) -2(c) show the effect of ECAP extrusion cycles on the particle distribution in<br />
PM 6061 Al/20% Al2O3 composite. The composite in the as-fabricated condition shows<br />
extensive particle clustering as expected. The clusters are aligned along the extrusion<br />
direction (Fig. 2(a)). These clusters begin to dissolve and disperse into individual particles<br />
after four ECAP passes at 370 ºC. The particle distribution appears homogeneous after<br />
pressing for seven passes. In addition to declustering, ECAP treatment also yields grain<br />
refinement of the aluminum alloy matrix.<br />
It is well recognized that nanoparticles tend to agglomerate into large clusters during<br />
composite processing even under low loading levels of reinforcement. In this respect,<br />
appropriate processing procedures are needed to improve the dispersion of nanoparticles in<br />
aluminum matrix. Recently, Yang et al. used high-intensity ultrasonic waves to assist the<br />
dispersion of SiC nanoparticles (average size ≤ 30 nm) in molten aluminum alloy A356 [10,<br />
11]. Fig. 3 shows a typical experimental setup for the ultrasonic assisted melting. The
278<br />
S.C. Tjong<br />
ultrasonic waves generate nonlinear effects in molten metal such as transient cavitation and<br />
acoustic streaming. Acoustic cavitation involves the formation, growth, pulsating and<br />
collapsing of tiny bubbles, thereby yielding transient local hot spots and implosive impacts to<br />
break up the clustered particles. The strong impact and local high temperatures enhance the<br />
wettability between molten metal and nanoparticles. Consequently, cast Al-based<br />
nanocomposite with better dispersion of ceramic nanoparticles can be prepared (Fig. 4).<br />
(a)<br />
(b)<br />
Figure 2. Continued on next page.
Recent Advances in Discontinuously Reinforced Aluminum… 279<br />
(c)<br />
Figure 2. Microstructure of the PM 6061 Al /20% Al 2O 3 composite: (a) as-extruded condition. The<br />
reinforcing alumina size is ~ 1-5 μm, (b) after four ECAP passes and (c) after seven ECAP passes [24].<br />
Figure 3. Experimental setup of ultrasonic assisted melting [10].<br />
In PM nanocomposites, clustering of nanoparticles often occurs during processing and<br />
the degree of agglomeration increases with increasing filler content [36] (Fig. 5). Through a<br />
solid-state cryomilling route, better dispersion of nanoparticles in aluminum matrix can be<br />
achieved. In the process, collisions between the grinding media lead to repeated fracture and<br />
welding of the raw powders in a high-energy ball mill. Low temperature (liquid nitrogen)<br />
environment suppresses the recovery and recrystallization of matrix grains during milling,<br />
thereby yielding finer grain structures. The nature of the process allows the incorporation of<br />
large volume fractions of reinforcement into aluminum matrix with a homogeneous<br />
distribution [12,13, 15, 37, 38]. Consolidation of cryomilled powders via hot pressing, cold<br />
isostatic pressing (CIP), hot isostatic pressing (HIP), extrusion, spark plasma sintering, etc. is<br />
necessary to produce bulk composites with full density, useful shapes and sizes for practical
280<br />
S.C. Tjong<br />
applications. Goujon et al. prepared the Al 5000/AlN (4-30 vol.%) nanocomposites though<br />
cryomilling of 5000 Al powder (380 nm) and AlN particle (150 nm) followed by hot pressing<br />
[12, 13]. Cryomilling for 6 h is required to obtain a good homogeneity of powder mixtures.<br />
The crystallite sizes of Al and AlN in the powders are reduced to about 49 and 30 nm,<br />
respectively. Hot pressing leads to homogeneous dispersion of the AlN phase in the Al alloy<br />
matrix and to an increase of the crystallite size of Al to submicrometer regime but not of AlN<br />
(Fig. 6). The microstructure of this composite consists of UFG aluminum grains free of AlN<br />
particles and regions dispersed with AlN nanoparticles.<br />
Figure 4. SEM micrograph showing the microstructure of as-cast A356/2%SiC nanocomposite [10].<br />
Figure 5. TEM micrograph showing agglomeration of particulates at the grain boundary of Al/5 vol.%<br />
Al2O 3 nanocomposite. The mean size of alumina nanoparticles is ~ 50 nm [36].
Recent Advances in Discontinuously Reinforced Aluminum… 281<br />
Figure 6. Microstructure of Al 5000/20.6vol.% AlN nanocomposite prepared by cryomilling and hot<br />
pressing [13].<br />
Structure-Property Relationship<br />
Aluminum-based nanocomposites can be classified into two categories according to the size<br />
dimensions of reinforcing particle and aluminum matrix employed, i.e. micrograined matrix<br />
composites reinforced with nanoparticles and UFG matrix composites reinforced with<br />
submicron- or nanoparticles. In the former case, ceramic nanoparticles are introduced directly<br />
into aluminum matrix having grain sizes in micrometer level via PM or ingot casting. The<br />
latter relates the use of cryomilling to refine the reinforcing particles and aluminum matrix<br />
down to submicrometer of nanoscale regime. Alternatively, the matrix grains of the<br />
composites can also be refined to submicrometer level using the SPD process.<br />
It is well recognized that the deformation behavior of nanocrystalline metals is quite<br />
different from their micro-grained counterparts. According to the Hall-Petch relation, a<br />
substantial increase in yield strength can be achieved by reducing the grain size of metals<br />
to the submicrometer or nanometer regime. Nanocrystalline metals generally have very low<br />
tensile ductility, and exhibit creep and superplasticity at lower temperatures compared to their<br />
micro-grained counterparts [9]. This is attributed to large volume (more than 50%) of atoms<br />
are located at the grain boundaries or interfacial boundaries of nanometals. Consequently,<br />
grain boundary activity is a dominant factor for controlling the mechanical properties. It is of<br />
practical interest to understand the effect of particle additions on the mechanical properties of<br />
aluminum and its alloys having submicrometer or nanometer grain sizes.<br />
Micro-grained Matrices<br />
Tjong et al. investigated the microstructure and mechanical properties of pure aluminum<br />
reinforced with low loading levels of Si3N4 (15 nm) or Si-N-C (25 nm) nanoparticles. Such
282<br />
S.C. Tjong<br />
nanoparticles were prepared by means of the laser induced gas-phase reactions [39-41]. They<br />
reported that the mechanical strength of nanoparticle strengthened composites is far superior<br />
to that of microparticle reinforce composite with a similar volume content of particulate. In<br />
other words, the tensile strength of Al/1vol% Si3N4 (15 nm) and Al/1vol.% Si-N-C (25<br />
nm)nanocomposites is comparable to that of Al/15vol% SiC (3.5μm) composite, but the yield<br />
stress of such nanocomposites is significantly higher than that of the microcomposite. The<br />
tensile ductility of nanocomposites is also higher than that of microcomposite (Table 1).<br />
However, increasing the Si-N-C nanoparticle content to 5 vol.% leads to deterioration of<br />
mechanical properties as a result of the particle agglomeration. The strengthening mechanism<br />
of nanocomposites is derived from the Orowan stress. It is well known that the Orowan<br />
strengthening results from interaction between dislocation and the dispersed particles during<br />
mechanical loading. Recently, Kang and Chan [36] also reported that the tensile strength of<br />
the Al/1vol.% Al2O3 nanocomposite is similar to that of the Al/10 vol.%SiCp (13 μm)<br />
composite, and the yield strength of the former is higher than that of the latter (Fig. 7). This<br />
figure reveals that the yield and tensile strengths of Al reinforced with Al2O3 nanoparticles<br />
increase with increasing filler content up to 4 vol.% Al2O3 at the expense of tensile ductility.<br />
Above 4 vol.%, the strengthening effects level off owing to the agglomeration of alumina<br />
nanoparticles as shown in Fig. 5. The main strengthening effect in such nanocomposites also<br />
arises from the Orowan stress. It is worth-noting that both the tensile strength and tensile<br />
ductility of cast Al-based composites are improved considerably as a result of better<br />
dispersion of nanoparticles in the alloy matrix via laser assisted melting [10,11].<br />
Figure 7. Tensile properties of Al/Al 2O 3 nanocomposites prepared by conventional powder metallurgy<br />
method. The tensile properties of Al/10 vol.% SiC (10 μm) microcomposites are also show for the<br />
purposes of comparison [36].
Recent Advances in Discontinuously Reinforced Aluminum… 283<br />
Table 1. Tensile properties of Al-based micro- and nanocomposites [41].<br />
Specimen<br />
Tensile Strength,<br />
MPa<br />
Yield Strength,<br />
MPa<br />
Elongation at<br />
Break, %<br />
Pure Al 70 30 ---<br />
Al/15 vol.% SiC (3.5 μm) 176 94 14.5<br />
Al/1 vol.% Si3N4 (15 nm) 180 144 17.4<br />
Al/1 vol.% Si-N-C (25 nm) 178 134 19.7<br />
Al/5 vol.% Si-N-C (25 nm) 153 114 6.2<br />
It is widely known that DRA microcomposites exhibit higher creep resistance than their<br />
unreinforced matrix materials because the particulates acting as barriers to dislocation<br />
movement. There is no plastic flow occurs within ceramic reinforcing particles. Accordingly,<br />
plastic deformation of DRA composites is controlled exclusively by flow in the metallic<br />
matrices. The high temperature creep behavior of coarse-grained Al and its alloys reinforced<br />
with microparticles is characterized by high values of n and Q. The creep activation energy of<br />
microcomposites is often much larger than that for aluminum lattice self-diffusion (142<br />
kJ/mol) [42-45]. Such anomalous behavior can be rationalized by introducing a threshold<br />
stress (σo) opposing creep flow. In this respect, the observed creep deformation is not driven<br />
by the applied stress σ but rather by an effective stress σc (σc = σ -σo). The threshold stress<br />
may originate from several sources such as Orowan bowing between particles, attractive<br />
attraction between dislocations and particles as well as back-stress associated with local<br />
dislocation climb [5, 42]. The rate controlling equation can be written as follows:<br />
σ −σ<br />
o n Q<br />
ε& = A(<br />
) exp( − )<br />
[1]<br />
G RT<br />
where ε& is the creep rate, A is a constant, G is shear modulus, R is Universal gas constant<br />
and T is absolute temperature. The creep behavior of Al-based microcomposites is related to<br />
modified creep behavior of aluminum solid solution alloys, and the equations developed for<br />
solid solution alloys can be used to described the creep behavior of composites provided that<br />
the applied stress is replaced by an effective stress. Thus, the threshold stress for creep in Albased<br />
microcomposites is associated with interactions between dislocations and fine<br />
dispersion of particles. These particles may be fine oxides in PM MMCs or precipitates in the<br />
matrix alloy of cast composites [42-44]. Introducing a threshold stress and its temperature<br />
dependence into the creep rate analysis yields a true stress exponent, n, of 3, 5 or 8, and true<br />
creep activation energy. For composites with a true exponent close to 3, dislocation viscous<br />
glide is rate controlling with an activation energy for creep is associated with interdiffusion of<br />
the solute atoms. On the other hand, dislocation climb process predominates for n = 5 in<br />
which the activation energy for creep is associated with aluminum lattice self-diffusion. For<br />
the composites with a true exponent close to 8, creep deformation is controlled by the lattice<br />
diffusion and its rate is proportional to the third power of substructure grain size λ [45, 46].<br />
Mathematically, the phenomenological creep rate equation for n = 8 can be written as:<br />
ε& = S (DL/b 2 ) (λ/b) 3 [(σ- σo)/E] 8 [2]
284<br />
S.C. Tjong<br />
where DL is lattice self-diffusion coefficient, λ sub-grain size, E Young’s modulus and S a<br />
numerical constant. A substructure is formed due to an increase in the dislocation density as a<br />
consequence of the thermal mismatch between the matrix and the reinforcement. The size of<br />
substructure is controlled by the interparticle spacing. Generally, subgrains can be generated<br />
more easily in pure Al than in the Al solid solution alloys during creep deformation [45, 46].<br />
Figure 8. Creep rate vs applied stress for the Al/1vol.% Si-N-C nanocomposite at 573 -673 K [41].<br />
Figure 9. Arrhenius plot of steady creep rate against 10 3 /T for Al/1vol.%Si-N-C (25 nm)<br />
nanocomposite [41].
Recent Advances in Discontinuously Reinforced Aluminum… 285<br />
Figure 10. Comparison of creep behavior between Al/1vol.%Si-N-C (25 nm) nanocomposite (open<br />
symbol) and Al/15vol% SiC (3.5μm) composite (solid symbol) at 573 and 623 K [41].<br />
The creep behavior of the nanoparticle reinforced composites is mainly depended on the<br />
matrix materials selected, i.e pure aluminum and aluminum solid solution alloy. The high<br />
temperature creep strength of micrograined aluminum is also greatly improved by the<br />
addition of low volume content of ceramic nanoparticles. Tjong et al. demonstrated that the<br />
creep resistance of the Al/1vol.%Si3N4 (15nm) and Al/1vol.%Si-N-C (25 nm)<br />
nanocomposites is about two orders of magnitude higher than that of the [40, 41]. Fig. 8<br />
shows the variation of steady creep rate vs applied stress for the Al/1vol.%Si-N-C (25 nm)<br />
nanocomposite. The Arrhenius plot of creep rate against 10 3 /T for this nanocomposite is<br />
shown in Fig. 9. The nanocomposite exhibits an apparent stress exponent (n) varying from<br />
15.7 to 23.0 and an apparent creep activation energy (Q) of 248 kJ/mol. The apparent<br />
activation energy of the Al/1vol.%Si-N-C nanocomposite is much higher than that for lattice<br />
diffusion of aluminum (142 kJ/mol). Similar high apparent values of n and Q values are also<br />
observed for the Al/1vol.% Si3N4 nanocomposite. For the purposes of comparison, the creep<br />
rates of the Al/15vol.% SiCp (3.5 µm) microcomposite and the Al/1vol.%Si-N-C (25 nm)<br />
nanocomposite at 573 and 623 K are presented in Fig. 10. It is evident that the creep rate of<br />
the Al/1vol.%Si-N-C nanocomposite is about two orders of magnitude lower comparing to<br />
the Al/15vol.% SiCp (3.5 µm) microcomposite. To rationalize the high apparent values of n<br />
and Q of the Al/1vol.%Si-N-C nanocomposite, a slip creep mechanism of constant<br />
substructure as given in Eq (2) is applied to the Al/1vol.%Si-N-C nanocomposite (Fig. 11). It<br />
appears that the datum points at three temperatures can be fitted linearly. It is considered that<br />
the nanoparticles of very small volume content (1 vol.%) pin the subgrain boundaries<br />
effectively. Consequently, the microstructure of nanocomposite remains unchanged during<br />
creep deformation. The threshold stress can be determined from Fig. 12 by extrapolating the<br />
linear regression line to zero strain rates. The values of threshold stress are determined to be<br />
36.3, 25.7 and 17.3 MPa at 573, 623 and 673 K, respectively. It is obvious that the threshold
286<br />
S.C. Tjong<br />
stress is temperature dependent. By plotting the lattice diffusion compensated creep rate<br />
(ε& /DL) against the modulus-compensated effective stress (σ -σo/E), the datum points for all<br />
temperatures merge into a straight line with a slope of 8 (Fig. 12). This implies that the creep<br />
rate of the Al/1vol.%Si-N-C nanocomposite is subgrain forming dislocation creep controlled<br />
by lattice-diffusion.<br />
1 / 8<br />
Figure 11. Variation of ε& with applied stress on double linear scales for Al/1vol.%Si-N-C (25 nm)<br />
nanocomposite [41].<br />
Figure 12. Variation of diffusivity normalized steady creep rate, ε& /D L, with modulus-compensated<br />
effective stress, σ -σ o/E, on double logarithmic coordinates for Al/1vol.%Si-N-C (25 nm)<br />
nanocomposite [41].
Recent Advances in Discontinuously Reinforced Aluminum… 287<br />
Figure 13. TEM micrograph showing interaction between dislocations and alumina nanoparticles<br />
during creep of PM 2024 Al/Al2O 3 nanocomposite at 10 MPa, 678 K [47].<br />
Figure 14. Creep rate vs effective stress on a logarithmic scale for PM 2014Al/Al 2O 3 nanocomposites<br />
prepared at (a) 0.3 and (b) 1.0 % oxygen levels [47].<br />
Recently, Mohamed and coworkers studied the creep mechanism of the PM aluminum<br />
solid solution alloy (2014 Al) reinforced with alumina nanoparticles [47]. The alumina<br />
nanoparticles of 30 and 35 nm are intentionally formed in 2014 Al during sintering via the<br />
introduction of water moisture with oxygen levels controlled at 0.3 and 1.0 wt%, respectively.<br />
Analysis of the creep data of this alloy reveals the presence of temperature- dependent
288<br />
S.C. Tjong<br />
threshold stress resulting from the interaction between moving dislocations and alumina oxide<br />
nanoparticles (Fig. 13). Such dislocation-particle interaction would impede lattice dislocation<br />
movement, thereby reducing creep rate of the composite. By incorporating the threshold<br />
stress into analysis, plots of creep rate versus effective stress yield straight lines with different<br />
slopes, i.e. n = 3 for low stress region and 5 for high stress regime (Fig. 14). Hence, the creep<br />
behavior of nanocomposite is consistent with the behavior of Al-Cu solid solution alloy (2014<br />
Al) that exhibits a transition from viscous glide (n = 3) to the high-stress region (n = 5) where<br />
dislocations break away from the solute atom atmosphere. They indicated that the true creep<br />
characteristics of PM 2024 Al/Al2O3 nanocomposite are consistent with those reported for<br />
aluminum solid-solution alloys [43]. Therefore, the creep deformation of nanocomposite with<br />
a matrix containing solutes is controlled by a viscous glide slip mechanism.<br />
Ultrafine Grained Matrices<br />
As mentioned above, conventional PM blending method yields Al nanocomposites with<br />
inhomogeneous distribution of reinforcing particles within the metal matrix. Cryomilling can<br />
provide a homogeneous dispersion of reinforcing particles in submicrometer or<br />
nanocrystalline matrix. The subsequent hot consolidation of cryomilled nanopowders into<br />
final bulk products causes the composites to have an UFG structure as a result of grain growth<br />
of the matrix. Schoenung and coworkers investigated the microstructure and tensile behavior<br />
of bulk nanostructures 5083 Al/5 vol.% SiC (25 nm) composite prepared by cryomilling<br />
followed by hot isostatic pressing and hot rolling [48]. They reported that the hot rolled<br />
composite consists of regions dispersed with SiC nanoparticles (100 -200 nm) and regions<br />
free of SiC nanoparticles (~ 700 nm). Fig. 15 shows the TEM micrograph of the SiCdispersed<br />
region in which SiC nanoparticles are distributed homogeneously within the<br />
ultrafine grains of the 5053 al matrix. The tensile properties of such composite from room<br />
temperature to 573 K are shown in Fig. 16. The composite exhibits very high tensile strength<br />
at room temperature but extremely low ductility. The strength decreases but the ductility<br />
increases with increasing test temperatures. In a nanocomposite with an UFG matrix, the<br />
dislocation movement in the matrix is restricted by the high density of grain boundaries.<br />
Consequently, the composite exhibits high tensile strength but very low tensile ductility.<br />
It is well known that nanocrystalline (NC) materials exhibit very low tensile ductility and<br />
toughness due to the lack of strain hardening [9]. The presence of coarser grains within the<br />
nanocrystalline matrix can enhance the ductility of nanostructured materials at the expense of<br />
mechanical strength [49-51]. Different toughening approaches have been proposed to enhance<br />
the ductility of NC materials either via thermomechanical treatment or cryomilling. Recently,<br />
Lavernia and coworkers reported that the UFG Al-Mg alloys with a bimodal microstructure<br />
exhibit a combination of high strength and good ductility [52- 55]. Such alloys were<br />
synthesized by consolidation of a mixture of cryomilled Al-Mg and unmilled powders.<br />
Consequently, strain hardening is regained in CG regions while maintaining high strength in<br />
NC regions. The CG grains can provide more dislocation activity than the NC grains. Ductilephase<br />
toughening in bi-modal structured Al-Mg alloys is attributed to the occurrence of crack<br />
bridging as well as delamination between UFG and CG regions during plastic deformation<br />
[51].
Recent Advances in Discontinuously Reinforced Aluminum… 289<br />
Figure 15. Bright field TEM micrograph of the hot rolled 5053 Al/5vol.% SiC composite showing the<br />
dispersion of SiC nanoparticles in ultrafine grains [48].<br />
Figure 16. Tensile stress-strain curves for the hot rolled 5053 Al/5 vol.% SiC composite at various<br />
temperatures [48].<br />
Based on this approach, Schoenung and coworkers prepared bimodal 5083Al/10 wt%B4C<br />
composite by blending cryomilled composite powders with an equal amount of CG 5083 Al<br />
followed by CIP and extrusion [15]. Figs. 17(a) -17(b) show the microstructure of bimodal<br />
composite consisting of UFG (NC) Al and CG Al. The B4C are uniformly distributed in the
290<br />
S.C. Tjong<br />
NC Al, and the NC Al and the CG Al are alternately distributed. (Fig.17(a)). This bimodal<br />
composite exhibits very high compressive yield strength of 1065 MPa comparing to 504 MPa<br />
of bimodal 5083 alloy. However, the composite still exhibits low compressive ductility<br />
(0.8%). Annealing the composite at 723 K improves its ductility to 2.5%. Fig. 18 shows the<br />
temperature dependence of the yield strength for the tri-modal composite. The yield strength<br />
decreases rapidly with test temperatures up to 473 K, followed by a relatively slow decrease<br />
at higher temperatures. At 473 K, the compressive yield stress of the tri-modal composite is<br />
282 MPa, being higher than that of the heat-treated 5083 alloy at room temperature [15]. In<br />
another study, Scheonung and coworkers fabricated bimodal 5083Al/6.5vol.% SiC (25 nm)<br />
composite by blending cryomilled composite powders with an equal amount of CG 5083 Al<br />
followed by HIP and hot rolling [56]. Such nanocomposite exhibits improved tensile ductility<br />
of 2.6% when compared with the nanocomposite consolidated from 100% of the cryomilled<br />
composite powders having a tensile ductility of 0.5%. This is because the ductile coarsegrains<br />
can undergo larger extent of plastic deformation, while ultrafine grains exhibit limited<br />
deformation.<br />
Figure 17. Bright field TEM images for the bimodal 5083Al/10 wt% B 4C composite in the (a) extrusion<br />
direction and (b) transverse direction, with the inset being the selected area diffraction patterns taken at<br />
the interface between the NC Al and B 4C [15].<br />
The creep behavior of the UFG composites is now considered. Presently, little is known<br />
regarding the high temperature creep behavior and deformation mechanism of such<br />
composites. What is the effect of reinforcing particles on the UFG composites having large<br />
grain boundary areas? Would these particles act as effective obstacles to the dislocation<br />
movement and hinder grain boundary sliding and diffusional flow during high temperature<br />
creep? If they do, the creep rates of UFG composites would reduce dramatically. More<br />
research work is needed in this area in near future to elucidate these problems. Nevertheless,<br />
proper understanding of the creep behavior of near-nanostructured Al-based alloy shed light<br />
on the creep deformation of UFG composites. Very recently, Chauhan et al. investigated the<br />
creep behavior of an UFG Al 5083 alloy at 573 – 648 K [57]. The alloy was prepared by<br />
consolidating cryomilled powders via HIP and extrusion. Analysis of the creep date reveals<br />
the presence of a temperature dependence threshold stress. Incorporation of this threshold
Recent Advances in Discontinuously Reinforced Aluminum… 291<br />
stress into a modified creep equation yields a true stress of ~ 2 and a true activation energy<br />
close to that for boundary diffusion for Al, indicating that the rate controlling process is<br />
related to grain boundary sliding. In other words, the grain boundary activity becomes<br />
dominant in an UFG 5083 Al alloy during creep deformation at high temperatures.<br />
Processing of DRA composites through SPD is an effective route to refine the grain size<br />
of composites to submicrometer level and to disperse the reinforcing particles homogeneously<br />
within the UFG matrix. Langdon and coworkers studied microstructural development in an<br />
Al-6061 composite reinforced with 10 vol.% Al2O3 particulates by means of the HPT and<br />
ECAP techniques [23]. The average size of particulates is ~ 10 μm. For HPT, the samples<br />
were strained at room temperature to a total strain of ~ 7 under a pressure of 3.5 GPa. For<br />
ECAP, samples were pressed for eight passes at 673 K, and two additional passes at 473 K,<br />
giving a total strain of ~ 10. Substantial grain refinement of aluminum alloy matrix can be<br />
achieved using both techniques, i.e., a mean grain size of ~0.2 μm is attained after HPT and<br />
~0.6 μm after ECAP. The microstructures of strained composites consist of an array of very<br />
small grains with poorly defined boundaries. There is no refinement for the alumina<br />
microparticles after HPT or ECAP treatment. The strength of the ECAP 6061 Al/Al2O3<br />
composite is increased by almost two fold by ECAP and close to a factor of ~3 by torsion<br />
straining due to the grain refining of aluminum alloy matrix [23]. In general, ECAP treatment<br />
does not cause fracture of reinforcing particles during plastic straining, especially for finer<br />
particulates [28]. However, limited particle cracking is found for the 6061 Al composite<br />
reinforced with large alumina particulate (7.4 μm) subjected to ECAP at room temperature<br />
[58].<br />
Figure 18. Temperature dependence of the compressive yield strength for the tri-modal 5083Al/10 wt%<br />
B4C composite. The inset shows true stress-strain curves tested at elevated temperatures [15].
292<br />
S.C. Tjong<br />
(a)<br />
(b)<br />
Figure 19. TEM micrograhs of (a) unreinforced pure Al after eight passes and (b) Al/5 vol.% Gr<br />
composite after four passes at room temperature [29].<br />
ECAP treatment of Al-based composites at room temperature is particularly attractive<br />
from the economic viewpoint. Proper selection of reinforcing particulates that experience no<br />
cracking is of technological interest. Very recently, Saravanan et al. used ECAP to refine the<br />
matrix grains of the Al/5 vol.% Gr (65 μm) composite at room temperature [29]. The soft and<br />
self lubricating nature of graphite can prevent the fracture of particulates during ECAP<br />
treatment. Figs. 19(a) shows a typical TEM micrograph of pure aluminum after eight ECAP<br />
passes at room temperature. The microstructure is characterized by well defined subgrains<br />
with a size of ~ 620 nm. In contrast, a significant grain refinement, down to the<br />
submicrometer level of ~ 300 nm can be achieved by pressing the Al/5 vol.% Gr composite at
Recent Advances in Discontinuously Reinforced Aluminum… 293<br />
room temperature for only four passes (Fig. 19(b)). Moreover, the grain boundaries of<br />
submicron grains of the composite are diffused comparing to a well defined structure of Al<br />
grains. The selected area electron diffraction (SAD) patterns of pure Al and the composite<br />
reveal numerous spot features indicating the presence of an array of many ultarfine grains<br />
having random distribution of orientations (insets of Figs. 1(9a)-(b)). The tensile strength of<br />
the composite increases from 97 to 249 MPa after four ECAP passes.<br />
Figure 20. TEM micrograph of Al/5 vol.% Al 2O 3 nanocomposite fabricated by HPT consolidation of<br />
raw material powders under 1.5 GPa [21].<br />
Figure 21. Tensile stress-strain curves of Al samples (1,2) and Al/5 vol.% Al 2O 3 samples (3, 4, 5)<br />
fabricated by HPT consolidation under the pressure of 1.5 GPa and tested at 300 ºC at strain rates of 10 -<br />
4 s -1 (1,3) and 10 -3 s -1 (2, 4) and at 400 ºC at a strain rate of 10 -4 s -1 (5) [21].
294<br />
S.C. Tjong<br />
Apart from forming ultrafine grains in composites, ECAP treatment can also consolidate<br />
ultrafine raw powders to produce fully dense (> 98%) bulk composite materials. Alexandrov<br />
et al. used the HPT technique to consolidate the Al powder (50 μm) and Al2O3 nanoparticle<br />
(50 nm) to form the Al/5 vol.% Al2O3 nanocomposite under a pressure of 1.5 GPa at room<br />
temperature. The powder mixture of nanocomposite was ball-milled for 30 min to ensure a<br />
uniform distribution of ceramic particles [21]. Fig. 20 shows the TEM micrograph of the HPT<br />
consolidated Al/5 vol.% Al2O3 nanocomposite. The nanocomposite exhibits an UFG structure<br />
having an average gain size of 120 nm. Room temperature tensile tests showed that the Al/5<br />
vol.% Al2O3 nanocomposite have limited ductility of 1 to 2%. At 300 ºC, the nanocomposite<br />
tested at a strain rate of 10 -3 s -1 had a plastic flow stress of ~ 66 MPa and a tensile ductility of<br />
~ 20 % (Fig. 21). In contrast, pure Al had a flow stress of ~ 60 MPa and a tensile ductility of<br />
~ 40 % tested at the same strain rate. However, the Al/5 vol.% Al2O3 nanocomposite showed<br />
a high strain-rate sensitivity of flow stress at 400 K; the strain-rate sensitivity (m) was 0.35.<br />
Strain rate sensitivity defined as the slope of logarithmic plot of the flow stress vs. strain rate.<br />
It is an inverse of stress exponent (n) and an important parameter in superplasticity. The Al/5<br />
vol.% Al2O3 nanocomposite exhibited a low flow stress of 20 MPa but a high tensile ductility<br />
of ~ 200 %. The enhanced tensile ductility observed in the HPT consolidated nanocomposite<br />
with a total elongation of ~ 200 % indicating the occurrence of superplastic-like flow<br />
behavior. According to the literature, high strain rate super-plasticity can be achieved in<br />
ECAP processed aluminum alloys with UFG structures [59]. High strain rate superplasticity<br />
in the sub-micron metals is often characterized by very high flow stresses or pronounced<br />
strengthening. Grain boundary sliding is considered to be the dominant deformation mode for<br />
superplasticity in the sub-micron and nanocrystalline metals [60]. Future challenges for<br />
materials scientists are to elucidate the underlying creep and superplastic deformation<br />
mechanisms of aluminum based nanocomposites having UFG and nanocrystalline matrices.<br />
Conclusions<br />
The development of aluminum nanocomposites is still in embryonic stage and there are many<br />
challenges in this field in the years ahead. Considerable progress has been made in the<br />
fabrication, microstructural and mechanical characterization of novel aluminum-based metal<br />
matrix nanocomposites in recent years. The nanocomposites can be simply prepared by<br />
incorporating very low volume contents of ceramic nanoparticles into aluminum matrix via<br />
PM or ingot casting. The nanocomposites thus prepared exhibit excellent mechanical<br />
properties including high yield strength and superior creep resistance. However,<br />
agglomeration of nanoparticles occurs readily during the composite fabrication, leading to<br />
poorer mechanical performance of composites with higher filler content. This problem can be<br />
eliminated in cast nanocomposites by using high-intensity ultrasonic waves to disperse the<br />
nanoparticles in molten aluminum. In the case of PM nanocomposites, cryomilling and severe<br />
plastic deformation processes have emerged as the two major processes to produce aluminum<br />
based composites having ultrafine grained matrix structures and homogeneous dispersion of<br />
reinforcing particles within the matrices.
References<br />
Recent Advances in Discontinuously Reinforced Aluminum… 295<br />
[1] Dutta, I.; Quiles, F.N.; McNelley, T.R.; Nagarajan, R. Metall Mater Trans A 1998, vol<br />
29, 2433 – 2446.<br />
[2] Hong, S.J.; Kim, H.M.; Huh, D.; Suryanarayana, C.; Chun, B.S. Mater Sci Eng A 2003,<br />
vol 346, 198 -204.<br />
[3] Ceschini, L.; Minak, G.; Morri, A. Compos Sci Technol 2006, vol 66, 333 – 342.<br />
[4] Westwood , A. R. Metall Trans A 1988, vol 19, 749 -758.<br />
[5] Tjong, S. C.; Ma, Z. Y. Mater Sci Eng R 2000, vol 29, 49 -113.<br />
[6] Yu, P.; Mei, Z.; Tjong, S. C. Mater Chem Phys. 2005, vol 93, 109 - 116.<br />
[7] Choy, K. L. In Handbook of Nanostructured <strong>Materials</strong> and Nanotechnology; Nalwa,<br />
H.S.; Ed.; Academic Press, New York, 2000, Vol. 1, pp. 533- 577.<br />
[8] Dodd, A. C.; McCormick, P.G. Scripta Mater 2001, vol. 44, 1725 – 1729.<br />
[9] Tjong, S.C.; Chen, H. Mater Sci Eng R 2004, vol 45, 1- 88.<br />
[10] Li, X.; Yang, Y.; Cheng, X. J Mater Sci 2004, vol 39, 3211 – 3212.<br />
[11] Yang, Y., Lan, J., Li, X. Mater Sci Eng A 2004, vol 380, 378 -383.<br />
[12] Goujon, C.; Goeuriot, P.; Chedru, M.; Vicens, J.; Chermant, J.L.; Bernard, F.; Niepce,<br />
J.C.; Verdier, P.; Laurent, Y. Powder Technol 1999, vol 105, 328 -336.<br />
[13] Goujon, C.; Goeuriot, P. Mater Sci Eng A 2001, vol 315, 180 -188.<br />
[14] Goujon, C.; Goeuriot, P. Mater Sci Eng A 2003, vol 356, 399 - 404.<br />
[15] Ye, J.; Han, B.Q.; Lee, Z.; Ahn, B.; Nutt, S.R.; Schoenung, J.M. Scripta Mater 2005,<br />
vol 53, 481 -486.<br />
[16] Han, B.Q.; Ye, J.; Tang, F.; Schoenung, J.; Lavernia, E.J. J Mater Sci 2007, vol 42,<br />
1660 -1672.<br />
[17] Zhang, Z.; Han, B.Q.; Witkin, D.; Ajdelsztajn, L.; Lavernia, E.J. Scripta Mater 2006,<br />
vol 54, 869 – 874.<br />
[18] Wang, J.; Horita, Z.; Furukawa, M.; Nemoto, M.; Trenev, N.K.; Valiev, R.Z, Ma, Y.;<br />
Langdon, T.G. J Mater Res 1993, vol 8, 2810 -2814.<br />
[19] Chinh N.Q.; Szommer, P.; Horita, Z.; Langdon, T.G. Adv Mater 2006, vol 18, 34 – 39.<br />
[20] Pippan, R.; Wetscher, F.; Hofok, M.; Varhauer, A.; Sabirov, I. Adv Eng Mater 2006, vol<br />
8, 1046 – 1056.<br />
[21] Alexandrov, I.V.; Zhu, Y.T.; Lowe, T.C.; Islamgaliev, R.K.; Valiev, R.Z. Metall Mater<br />
Trans A 1998, 29, 2253 – 2260.<br />
[22] Leo, P., Cerri, E.; De Marco P.P.; Roven, H. J. J Mater Process Technol 2007, vol 182,<br />
207 – 214.<br />
[23] Valiev, R. Z.; Islamgaliev, R. K.; Kuzmina, N.F.; Li, Y.; Langdon, T.G. Scripta Mater.<br />
1998, vol 40, 117 - 122.<br />
[24] Sabirov, I.; Kolednik, O.; Valiev, R.Z.; Pippan, R. Acta Mater 2005, vol 53, 4919 –<br />
4930.<br />
[25] Chen, L.J.; Ma, C.Y.; Stoica, G.M.; Liaw, P.K.; Xu, C.; Langdon, T.G. Mater Sci Eng A<br />
2005, vol 410-411, 472 – 475.<br />
[26] Han, B. Q.; Langdon, T. G. Mater Sci Eng A 2005, vol 410-411, 430 – 434.<br />
[27] Lillo, T.M. Mater Sci Eng A 2005, vol 410-411, 443 – 446.<br />
[28] Kawasaki, M.; Huang, Y.; Xu, C.; Furukawa, M.; Horita, Z.; Langdon, T.G. Mater Sci<br />
Eng A 2005, vol 410- 411, 402 – 407.
296<br />
S.C. Tjong<br />
[29] Saravanan, M.; Pillai, R.M.; Ravi, K.R.; Pai, B.C.; Brahmakumar, M. Compos Sci<br />
Technol 2007, vol 67, 1275 – 1279.<br />
[30] Kennedy, A.R.; Wyatt, S.M. Compos Sci Technol 2000, vol 60, 307 – 314.<br />
[31] Youssef, Y.M.; Dashwood, R.J.; Lee, P.D. <strong>Composite</strong> A 2005, vol 36, 747 – 763.<br />
[32] Bhanu Prasad V.V.; Bhat, B. V.; Mahajan, Y.R.; Ramakrishnan, P. Mater Sci Eng A<br />
2002, vol 337, 179 -186.<br />
[33] Slipenyuk, A.; Kuprin, V.; Milman, Y.; Spowart, J. E.; Micracle, D. B. Mater Sci Eng A<br />
2004, vol 381, 165 – 170.<br />
[34] Tjong, S.C.; Tam, K.F. Mater Chem Phys 2006, vol 97, 91-97.<br />
[35] Lewandowski, J.J. In Powder Metallurgy <strong>Composite</strong>s, Kumar, P.; Vedula, K.; Eds;<br />
TMS, Warrendale, PA, 1988, pp. 117 -124.<br />
[36] Kam, Y.C.; Chan, S.L. Mater Chem Phys 2004, vol 85, 438 - 443.<br />
[37] Ye , J.; He, J.; Schoenung, J.M. Metall Mater Trans A 2006, vol 37, 3099- 3109.<br />
[38] Ye, J.; Lee, Z.; Ahn, B.; He, J.; Nutt, S.R.; Schoenung, J.M. Metall Mater Trans A 2006,<br />
vol 37, 3111- 3117.<br />
[39] Ma, Z.Y.; Li, .L.; Liang, Y., Zhen, F.; Bi, J.; Tjong, S.C. Mater Sci Eng A 1996, vol 219,<br />
229 - 231.<br />
[40] Ma, Z.Y.; Tjong, S.C.; Li, Y.L.; Liang, Y. Mater Sci Eng A 1997, vol 225, 125 -134.<br />
[41] Ma, Z.Y.; Tjong, S.C.; Li, Y. L. Compos Sci Technol 1999, vol 59, 263 – 270.<br />
[42] Park, K.T.; Lavernia, E. J.; Mohamed, F.A. Acta Metall Mater 1990, vol 38, 2149-2159.<br />
[43] Mohamed, F. A. Mater Sci Eng A, 1998, vol 245, 242 - 256.<br />
[44] Li, Y.; Langdon, T. G. Acta Mater 1998, vol 46, 1143-1155.<br />
[45] Sherby, O.D.; Klundt, R.H.; Miller, A.K. Metall Trans A 1977, vol 8, 843 -850.<br />
[46] Mishra, R.S.; Pandey, A.B.; Prasad, K.S.; Mukherjee, A.K. Scripta Metall Mater 1995,<br />
vol 33, 479 -484.<br />
[47] Lin, Z.; Chan, S. L.; Mohamed, F. A. Mater Sci Eng A 2005, vol 394, 103 -111.<br />
[48] Tang, F.; Han,B.Q.; Hagiwara, M.; Schoenung, J. M. Adv Eng Mater 2007, vol 9,<br />
286 -291.<br />
[49] Koch, C.C.; Morris, D.G.; Lu, K.; Inoue, A. MRS Bull 1999, vol 24, 54-48.<br />
[50] Hayes, R.W.; Rodriquez, R.; Lavernia, E.J. Acta Mater 2001, vol 49, 4055 - 4068.<br />
[51] Wang, Y.; Chen, M.; Zhou, F.; Ma, E. Nature 2002 vol 419, 912 -914.<br />
[52] Han, B.Q.; Lee, Z.; Witkin, D.; Nutt, S.R.; Lavernia, E.J. Metall Mater Trans A 2003,<br />
vol 34, 957 -965.<br />
[53] Lee, Z.; Witkinn, D.B.; Radmilovic, V.; Lavernia, E.J.; Nutt, S. R. Mater Sci Eng A<br />
2005, vol 410 -411, 462 – 467.<br />
[54] Fan, G.J.; Choo, H.; Liaw, P.K.; Lavernia, E.J. Acta Mater 2006, vol 54, 1759 -1766.<br />
[55] Han, B.Q.; Huang, J.Y.; Zhu, Y.T.; Lavernia, E.J. Acta Mater 2006, vol 54, 3015 - 3024.<br />
[56] Tang, F.; Hagiwara, M.; Schownung, J.M. Mater Sci Eng A 2005, vol 407, 306 -314.<br />
[57] Chuhan M.; Roy, I.; Mohamed, F.A. Mater Sci Eng A 2005, vol 410 -411, 24 -27.<br />
[58] Li, Y.; Langdon, T. G. J Mater Sci 2000, vol 35, 1201 -1204.<br />
[59] Valiev, R.Z.; Salimonenko, D.A.; Tsenev, N.K.; Berbon, P.B.; Langdon, T.G. Scripta<br />
Mater 1997, vol. 37, 1945 -1950.<br />
[60] Mohamed, F. A.; Li, Y. Mater Sci Eng A 2002, vol 298, 1-15.
A<br />
Aβ, 283<br />
accounting, 44<br />
accuracy, 69, 124, 263<br />
acetone, 136<br />
acidity, 116<br />
acoustic waves, 167<br />
activation energy, 283, 285, 291<br />
actuation, 125<br />
actuators, xi, 106, 115, 257, 258, 272, 273<br />
adaptation, 122<br />
additives, 110, 118, 126<br />
adhesion, ix, 11, 110, 112, 119, 120, 125, 141, 143,<br />
212<br />
adhesion properties, 120<br />
adhesion strength, 143<br />
adhesives, 162<br />
adjustment, 183, 196, 198<br />
aerospace, vii, viii, 2, 109, 110, 111, 112, 118, 119,<br />
120, 122, 124, 126, 130, 276<br />
age, ix, 109<br />
Alabama, 126<br />
algorithm, 52, 54, 65, 73, 75, 81, 103, 105, 216<br />
alkalinity, 116<br />
alloys, xi, 14, 47, 125, 163, 275, 276, 281, 283, 284,<br />
288, 294<br />
alternative(s), 6, 13, 36, 58, 118, 119<br />
aluminium, 4, 162, 210, 221<br />
aluminium alloys, 4<br />
aluminum, xi, 14, 116, 117, 118, 119, 164, 275, 276,<br />
277, 279, 280, 281, 283, 285, 287, 288, 289, 291,<br />
292, 293, 294, 295<br />
ambiguity, 244<br />
amplitude, 136, 168, 174, 191, 193, 206, 215, 219,<br />
225<br />
anisotropy, 22, 42, 46, 117<br />
annealing, 290<br />
INDEX<br />
AP, 173, 175, 176, 178, 182, 183, 184, 186, 188, 189<br />
Arborite, vii<br />
arithmetic, 5, 6, 41<br />
ash, 255<br />
aspect ratio, 44, 113, 115, 126, 130<br />
asphalt, vii<br />
assessment, 206, 238<br />
assignment, 154<br />
assumptions, 20, 23, 232<br />
asymptotic, 243, 248, 253<br />
atmosphere, 120, 131, 288<br />
atoms, 281, 283<br />
attention, ix, 110, 130<br />
automation, 113, 117<br />
averaging, 6<br />
avoidance, 119<br />
awareness, 110<br />
B<br />
barriers, 283<br />
beams, 120, 219<br />
behavior, xi, 4, 13, 49, 73, 75, 78, 105, 110, 118,<br />
134, 144, 164, 244, 246, 257, 258, 272, 281, 283,<br />
285, 288, 290, 294<br />
Belgium, 51, 101, 102, 104, 105, 106, 209, 234, 235,<br />
236<br />
bending, 49, 59, 64, 72, 106, 196, 198, 210, 214,<br />
215, 216, 218, 219, 225, 226, 227, 228, 229, 230,<br />
231, 232, 233, 235, 236, 267, 269, 270, 271, 272<br />
benefits, ix, 109, 116, 118, 119, 121, 124, 125, 126<br />
bias, 217<br />
biomaterials, 125, 128<br />
blends, 255<br />
BMI, 123<br />
bonding, 143, 266<br />
bounds, 12, 69<br />
Bragg grating, 125, 204, 211, 212
298<br />
braids, 115<br />
broadband, 211<br />
bubbles, 133, 278<br />
bulk materials, 115<br />
C<br />
candidates, 38, 120<br />
capillary, 134<br />
carbon, vii, viii, ix, x, 2, 4, 6, 10, 11, 12, 18, 19, 22,<br />
23, 27, 31, 36, 38, 42, 44, 46, 47, 48, 49, 50, 58,<br />
81, 92, 109, 110, 112, 113, 115, 117, 118, 119,<br />
120, 121, 123, 125, 126, 127, 129, 130, 131, 132,<br />
133, 134, 135, 136, 137, 138, 139, 141, 143, 144,<br />
145, 147, 149, 151, 153, 155, 156, 157, 159, 161,<br />
162, 163, 164, 166, 205, 206, 207, 209, 212, 213,<br />
214, 217, 220, 221, 225, 226, 227, 228, 234, 242,<br />
248, 255, 256<br />
carbon nanotubes, 44, 46, 49, 112<br />
carbonization, 120<br />
case study, 104, 127, 198<br />
cast(ing), 277, 278, 280, 281, 282, 283, 294<br />
catalyst, 122, 123<br />
catalytic effect, 139<br />
cell, 173, 191, 241, 262<br />
cellulose, vii<br />
ceramic(s), vii, xi, 2, 6, 163, 257, 258, 266, 273, 275,<br />
276, 277, 278, 281, 283, 285, 294<br />
chain mobility, 141<br />
chemical composition, 116<br />
chemical properties, vii, 116<br />
chemical reactions, 276<br />
chemical reactivity, 115<br />
chemical structures, 119<br />
chicken, 125<br />
China, 125<br />
civil engineering, 130<br />
classes, 115<br />
clustering, 277, 279<br />
clusters, 277<br />
coatings, 112, 115, 118<br />
collagen, vii<br />
collisions, 279<br />
commercial, 72, 106, 117, 120, 163, 226, 266<br />
communication, 234<br />
community, 126<br />
compatibility, 26<br />
complexity, 105, 122, 219<br />
compliance, 72, 84, 85, 93, 259<br />
complications, 215<br />
components, vii, 7, 18, 19, 22, 23, 28, 32, 33, 36, 38,<br />
42, 113, 115, 121, 124, 147, 148, 165, 166, 212,<br />
257, 261<br />
Index<br />
composites, vii, viii, ix, x, xi, 1, 2, 3, 4, 6, 10, 13, 14,<br />
15, 21, 22, 28, 31, 38, 41, 42, 43, 44, 45, 46, 47,<br />
48, 50, 51, 52, 61, 69, 70, 101, 102, 103, 106, 109,<br />
110, 112, 113, 114, 115, 116, 117, 118, 119, 120,<br />
121, 122, 125, 126, 129, 130, 135, 143, 144, 148,<br />
149, 154, 155, 156, 157, 158, 159, 160, 161, 162,<br />
163, 164, 166, 204, 205, 206, 207, 209, 210, 212,<br />
213, 214, 215, 217, 218, 221, 225, 233, 234, 235,<br />
236, 254, 255, 256, 257, 258, 272, 275, 276, 277,<br />
279, 281, 282, 283, 285, 288, 290, 291, 292, 294<br />
composition(s), 43, 171, 183, 276<br />
compounds, 114<br />
computation, 28, 67, 94, 222, 244, 246<br />
computer simulation, 123<br />
computing, 104, 244<br />
concentration, viii, 1, 2, 25, 26, 28, 115, 134, 135,<br />
148, 149, 158, 164, 221, 224, 225, 258, 273<br />
conception, 101<br />
concrete, vii<br />
concurrent engineering, 104<br />
condensation, 277<br />
conditioning, 123<br />
conductivity, 112, 113, 119, 132<br />
confidence, 34<br />
configuration, 3, 34, 65, 69, 73, 78, 81, 88, 91, 173,<br />
175, 201<br />
confusion, 250<br />
Congress, 101, 106, 107<br />
consolidation, 135, 136, 138, 288, 293<br />
constant rate, 252<br />
constraints, 52, 53, 54, 55, 56, 67, 69, 72, 73, 80, 81,<br />
89, 90, 98, 99, 102, 107, 121<br />
construction, 51, 111, 183<br />
continuity, 71<br />
control, x, 62, 116, 117, 124, 125, 129, 173, 191, 193<br />
conventional composite, 126<br />
convergence, 74, 77, 78, 83, 85, 94, 95, 97, 100, 102,<br />
154, 231<br />
convex, viii, 51, 52, 54, 55, 61, 64, 65, 66, 72, 74,<br />
75, 79, 101, 103<br />
cooling, 136<br />
corn, 125<br />
correlation(s), x, 168, 203, 209, 221<br />
corrosion, 2, 110, 118, 119, 132, 241<br />
cosine, 135<br />
cost saving, 122<br />
costs, ix, 110, 114, 121, 124, 125<br />
Coulomb, 42<br />
coupling, viii, 1, 44, 45, 59, 63, 64, 120, 148, 164,<br />
219, 257<br />
covalent bond, 119<br />
coverage, 124<br />
CPU, 56, 73, 98, 232
crack, 52, 99, 100, 116, 143, 162, 167, 168, 196,<br />
198, 237, 239, 273, 288<br />
creep, xi, 115, 218, 275, 281, 283, 284, 285, 286,<br />
287, 288, 290, 291, 294<br />
critical value, 260, 261<br />
cross-linked polymers, 139<br />
crystal polymers, 118<br />
crystal structure, 132<br />
crystalline, 47, 119, 132, 163<br />
crystals, 133<br />
curing, 113, 117, 121, 122, 123<br />
cybernetics, 104<br />
cycles, 54, 73, 84, 210, 214, 215, 277<br />
cycling, 207<br />
D<br />
damping, 112, 113, 114, 115, 166<br />
database, 121<br />
DD, 190, 205, 206, 239, 240, 243, 244, 246, 247,<br />
248, 250, 251, 252, 253, 254<br />
decomposition, 139, 141<br />
decomposition temperature, 139, 141<br />
defects, 120, 172, 219<br />
defense, 120<br />
definition, 3, 32, 59, 61, 63, 66, 77, 80, 82, 98, 113,<br />
169, 187, 238, 240, 244, 247, 249, 253<br />
deformability, 162<br />
deformation, 105, 143, 148, 164, 182, 196, 216, 241,<br />
277, 281, 283, 284, 285, 288, 290, 291, 294<br />
degradation, x, 129, 131, 141, 200, 211, 215, 220,<br />
225, 233, 236, 243, 257<br />
degree of crystallinity, 119<br />
Delaware, 125<br />
delivery, 136<br />
demand, xi, 118, 121, 275<br />
Denmark, 102, 103, 104, 234<br />
density, 11, 26, 44, 62, 76, 77, 79, 82, 96, 116, 117,<br />
120, 130, 134, 135, 260, 264, 267, 276, 277, 279,<br />
284, 288<br />
derivatives, 75, 76, 77, 78, 103<br />
detection, 204, 213, 216, 218, 219, 234, 254<br />
deterministic, 52, 54, 56, 72<br />
deviation, 10, 18, 41, 143, 144, 147<br />
diamond, 173, 191<br />
diaphragm, 121<br />
dielectric, 259, 261, 263<br />
dielectric permittivity, 259, 261<br />
differential equations, 215<br />
differential scanning calorimetry (DSC), x, 129, 136,<br />
139, 140, 142, 161<br />
differentiation, 258<br />
diffraction, 45, 46, 48, 290<br />
Index 299<br />
diffusion, 2, 25, 28, 44, 46, 134, 136, 283, 284, 285,<br />
286, 291<br />
diffusion process, 28, 44<br />
diffusivity, 286<br />
discontinuity, 235<br />
discrete variable, 101<br />
discretization, 184, 185, 186<br />
discrimination, 167<br />
discs, 102, 106<br />
dislocation, 282, 283, 284, 286, 288, 290<br />
dispersion, xi, 69, 115, 125, 133, 134, 135, 143, 144,<br />
167, 275, 277, 278, 279, 280, 282, 283, 288, 289,<br />
294<br />
displacement, 26, 27, 59, 64, 87, 88, 149, 150, 152,<br />
155, 166, 169, 173, 177, 178, 179, 180, 181, 191,<br />
193, 195, 196, 197, 198, 200, 201, 203, 210, 215,<br />
216, 222, 224, 225, 226, 227, 228, 231, 233, 238,<br />
239, 240, 241, 244, 245, 246, 258, 259, 262, 266,<br />
269, 273<br />
distribution, 67, 68, 115, 122, 147, 156, 164, 183,<br />
215, 264, 265, 268, 269, 272, 276, 277, 279, 288,<br />
293, 294<br />
doors, 124<br />
double logarithmic coordinates, 286<br />
dream, 121<br />
ductility, 239, 277, 281, 282, 288, 290, 294<br />
durability, 112, 113, 237<br />
duration, 168, 210, 248, 253, 254<br />
dynamic mechanical analysis, 141<br />
E<br />
ears, 167<br />
EI, 5, 7, 15, 22<br />
Einstein, Albert, 258<br />
elastic deformation, 143<br />
elasticity, 12, 23, 48, 96, 166<br />
elastomers, 44, 47<br />
electric field, 120, 257, 258, 260, 263, 264, 267, 268,<br />
269, 270, 272<br />
electric potential, 258, 262, 263<br />
electrical conductivity, 113, 116, 117<br />
electrical properties, 118, 163<br />
electrical resistance, 166, 204, 214, 234<br />
electrodes, xi, 213, 234, 257, 258, 266, 272, 273<br />
electromagnetic, 115, 212<br />
electron, 147<br />
electrospinning, 120<br />
elongation, 294<br />
embryonic, 294<br />
emission, x, 165, 166, 167, 168, 169, 192, 196, 204,<br />
205, 206<br />
emulsification, 134
300<br />
endothermic, 139<br />
endurance, 32<br />
energy, x, xi, 61, 62, 72, 76, 77, 79, 82, 83, 84, 89,<br />
91, 92, 99, 105, 110, 115, 118, 119, 122, 133, 143,<br />
165, 166, 167, 168, 169, 170, 171, 172, 176, 177,<br />
178, 179, 180, 181, 182, 183, 184, 195, 196, 197,<br />
198, 199, 200, 201, 204, 205, 225, 237, 238, 239,<br />
240, 241, 243, 244, 245, 246, 248, 249, 250, 251,<br />
252, 253, 254, 255, 256, 257, 258, 260, 261, 263,<br />
264, 268, 279<br />
energy consumption, 143<br />
energy density, xi, 61, 62, 76, 82, 83, 89, 91, 92, 257,<br />
258, 260, 261, 263, 264, 268<br />
energy emission, 201<br />
environment, 110, 141, 279<br />
environmental conditions, 2, 44<br />
environmental control, 113<br />
epoxy, vii, ix, 2, 6, 10, 11, 12, 13, 18, 19, 20, 22, 23,<br />
27, 31, 34, 35, 36, 39, 41, 42, 44, 46, 48, 58, 81,<br />
84, 92, 112, 116, 117, 120, 121, 125, 129, 130,<br />
131, 138, 139, 141, 149, 150, 157, 161, 162, 163,<br />
166, 173, 191, 204, 205, 206, 207, 210, 211, 212,<br />
217, 218, 220, 231, 233, 235, 236, 240, 247, 248,<br />
254, 255, 256<br />
epoxy resins, 31, 34, 48, 120<br />
equal channel angular pressing, 277<br />
equality, 107, 239<br />
equilibrium, 26, 27, 134<br />
equipment, x, 110, 118, 123, 124, 139, 209<br />
esters, 123<br />
estimating, 5, 6, 8, 10, 12, 13, 42, 43, 47<br />
European, viii, 49, 51, 52, 101, 102, 111, 162, 163,<br />
206, 234, 235<br />
evaporation, 120<br />
evidence, 198<br />
evolution, 44, 82, 88, 148, 204, 210, 211, 214<br />
exclusion, 5<br />
exothermic, 139<br />
exposure, 117<br />
extraction, 135<br />
extrusion, 277, 279, 289, 290<br />
F<br />
FAA, 121<br />
fabric, 125, 205, 207, 211, 212, 214, 217, 218, 220,<br />
234, 236, 255, 256<br />
fabrication, xi, 124, 136, 138, 275, 276, 277, 294<br />
failure, viii, x, xi, 2, 31, 32, 33, 34, 35, 36, 38, 39,<br />
40, 41, 43, 46, 47, 48, 53, 122, 130, 131, 143, 144,<br />
148, 149, 155, 156, 158, 159, 160, 162, 164, 165,<br />
166, 168, 171, 172, 175, 182, 183, 184, 187, 194,<br />
Index<br />
195, 198, 200, 201, 204, 205, 207, 210, 221, 237,<br />
253, 254, 275<br />
family, 72, 101, 113, 118<br />
fatigue, x, 2, 29, 110, 117, 166, 204, 209, 210, 211,<br />
212, 213, 214, 215, 216, 217, 218, 219, 220, 221,<br />
225, 226, 228, 229, 230, 231, 232, 233, 234, 235,<br />
236, 254, 256<br />
fiber bundles, 135, 156<br />
fiber optics, 125<br />
fibers, vii, viii, ix, 2, 3, 10, 11, 12, 13, 14, 18, 19, 20,<br />
23, 27, 28, 29, 31, 35, 36, 42, 44, 49, 51, 52, 53,<br />
54, 56, 57, 58, 61, 62, 65, 66, 67, 69, 70, 72, 73,<br />
75, 81, 84, 85, 86, 88, 96, 97, 100, 101, 109, 110,<br />
112, 113, 114, 115, 116, 117, 118, 119, 120, 122,<br />
123, 124, 125, 126, 127, 135, 147, 148, 149, 150,<br />
156, 161, 162, 192, 246, 248<br />
filament, ix, 48, 129, 131, 133, 134, 136, 137, 138,<br />
147, 161, 205<br />
filled polymers, 130<br />
fillers, 112, 114, 115, 116, 130, 132, 143<br />
film(s), 50, 115, 122, 123, 138, 164<br />
finance, 233<br />
finite element method, 67<br />
fires, 121<br />
First World, 106, 107<br />
fishing, vii<br />
fixation, 232<br />
flame, 113, 114, 115, 118<br />
flexibility, 110, 117, 119, 124, 126, 243<br />
flexural strength, x, 129, 130, 136, 143<br />
fluctuations, 133<br />
fluid, 25, 27, 121, 133<br />
focusing, 8<br />
Formica, vii<br />
fracture processes, 147<br />
fractures, 182, 194, 213<br />
France, 1, 235<br />
freedom, 118<br />
friction, 151, 167, 215, 223, 224, 225, 226, 227, 228<br />
fuel cell, 115<br />
fulfillment, 119<br />
functionalization, ix, 109, 112<br />
G<br />
GAO, 206, 254, 255<br />
gas phase, 277<br />
gauge, 193, 211, 215, 232<br />
generation, 44, 112, 113<br />
genetic algorithms, 52, 71, 104, 106<br />
genre, 119<br />
Germany, 50, 102, 106, 107, 234
glass, x, 47, 48, 84, 111, 116, 117, 120, 121, 123,<br />
125, 126, 129, 135, 139, 141, 161, 204, 205, 206,<br />
209, 210, 211, 217, 218, 220, 231, 234, 235, 236,<br />
240, 247, 254, 255, 256<br />
glass transition, x, 129, 139, 141, 161, 217<br />
glass transition temperature, x, 129, 139, 161, 217<br />
GNP, 115<br />
gold, 118, 147<br />
grain boundaries, 281, 288, 293<br />
grain refinement, 291, 292<br />
grains, 262, 277, 279, 280, 281, 288, 289, 290, 291,<br />
292, 293, 294<br />
graph, 140, 142<br />
graphite, 112, 115, 117, 120, 173, 191, 205, 206,<br />
233, 256, 292<br />
gravimetric analysis, 141<br />
Greece, 236<br />
groups, 121<br />
growth, 110, 116, 117, 118, 121, 143, 167, 168, 183,<br />
196, 204, 215, 225, 235, 238, 239, 253, 256, 278,<br />
288<br />
hardness, 113, 117, 118, 132, 241<br />
head, 191, 193, 241<br />
healing, 125<br />
health, 110, 125<br />
heat(ing), 111, 121, 122, 132, 134, 136, 167, 217,<br />
290<br />
heat transfer, 121<br />
height, 215, 241<br />
hemicellulose, vii<br />
heterogeneity, 246<br />
high fat, 116<br />
homogeneity, 277, 280<br />
Hong Kong, 275<br />
hot pressing, 279, 280<br />
hot spots, 278<br />
house(ing), 105, 119, 222<br />
humidity, 116<br />
hybrid, vii, 115, 116, 117, 123, 163, 234<br />
hybridization, ix, 109, 116, 118<br />
hydroxyapatite, vii<br />
hypothesis, 35, 135, 226, 228<br />
hysteresis, 225, 226, 227, 228<br />
hysteresis loop, 225, 226, 228<br />
H<br />
identification, viii, 1, 2, 3, 14, 15, 17, 30, 31, 39, 40,<br />
42, 43, 44, 167, 184, 204, 240<br />
I<br />
Index 301<br />
identity, 7<br />
images, 221, 290<br />
imaging, 205<br />
immersion, 134<br />
impact energy, 237, 238, 239, 240, 241, 242, 243,<br />
244, 245, 246, 247, 248, 251, 252, 253, 254<br />
implementation, 98, 198, 215, 236<br />
impregnation, ix, 122, 123, 129, 133, 136, 137, 161,<br />
163<br />
in situ, 117, 164<br />
incidence, 219<br />
inclusion, 8, 9, 11, 43, 45, 49, 112<br />
independent variable, 63<br />
indication, xi, 237<br />
indicators, 201<br />
indices, 258<br />
industrial application, 52, 67, 73, 277<br />
industrial sectors, 126<br />
industry, ix, 110, 116, 117, 118, 121, 124, 130<br />
inelastic, 44<br />
inertia, 232<br />
infinite, 9, 11, 28, 33, 40<br />
infrastructure, 110, 118, 122<br />
initiation, 225, 238, 256, 273<br />
insight, 233<br />
inspection, 218<br />
Instron, 191<br />
insulation, 174<br />
integration, 223, 228, 230, 231, 241<br />
integrity, x, 121, 165, 172, 183, 201<br />
intelligence, 126<br />
intensity, 136, 258, 277, 294<br />
interaction(s), 4, 9, 32, 33, 34, 48, 50, 115, 123, 135,<br />
141, 161, 223, 262, 273, 282, 283, 287, 288<br />
interface, ix, 99, 110, 112, 119, 122, 127, 130, 135,<br />
143, 148, 149, 150, 151, 153, 155, 156, 157, 160,<br />
161, 162, 164, 265, 266, 269, 272, 273, 276, 290<br />
interfacial adhesion, ix, 109, 110, 112, 119, 120<br />
interfacial bonding, 162<br />
interfacial properties, 120<br />
interference, 115, 212<br />
intermetallic nanoparticles, 276<br />
international standards, 210<br />
interpretation, 14, 182, 215<br />
interval, 185<br />
intuition, 134<br />
invariants, 58, 63<br />
inversion, 14, 15, 16, 31, 201<br />
investment, 124<br />
ion implantation, 120<br />
IR, 122, 204<br />
iron, 14, 48<br />
irradiation, 163
302<br />
isostatic pressing, 279, 288<br />
Italy, 237, 255<br />
iteration, 52, 55, 77, 83, 84, 94, 95, 98, 154, 155<br />
Japan, 273<br />
kinetic energy, 239<br />
knowledge transfer, 209<br />
J<br />
K<br />
L<br />
labor, 123<br />
laminar, 182, 183<br />
laminated composites, 42, 103, 106, 205<br />
lamination, 56, 63, 64, 65, 66, 67, 73, 101, 103, 104,<br />
116, 175<br />
laser, 120, 277, 282<br />
laser ablation, 120, 277<br />
laws, 24, 45, 47<br />
lead, 15, 22, 74, 77, 119, 123, 124, 134, 143, 211,<br />
217, 263, 279<br />
leakage, 194, 198<br />
leaks, 122<br />
lifetime, 212<br />
lignin, vii<br />
limitation, 134<br />
liquid nitrogen, 121, 279<br />
liquids, 134<br />
literature, viii, x, 1, 2, 3, 4, 10, 14, 18, 21, 22, 31, 32,<br />
33, 35, 38, 42, 43, 44, 45, 51, 52, 71, 72, 73, 100,<br />
218, 237, 238, 244, 248, 253, 294<br />
localization, 10, 11, 13, 17, 20, 30, 42, 45, 191<br />
location, 60, 99, 241<br />
London, 46, 163, 234, 235<br />
M<br />
machine learning, 103<br />
magnetic properties, 116<br />
management, 113, 115, 117<br />
manipulation, 80<br />
manufacturer, 113<br />
manufacturing, ix, 2, 3, 43, 61, 69, 102, 104, 105,<br />
109, 110, 112, 113, 122, 123, 126, 129, 130, 133,<br />
135, 138, 161, 212, 275<br />
mapping, x, 209<br />
market(s), 110, 116, 119, 122, 124<br />
Index<br />
masking, 117<br />
mass loss, 44<br />
mass transfer process, 134<br />
material degradation, 210<br />
materials science, 6, 14<br />
mathematical programming, 54, 65, 72, 102<br />
mathematics, 5, 46<br />
matrix, vii, viii, ix, x, 1, 2, 3, 4, 5, 7, 8, 9, 11, 12, 13,<br />
14, 15, 16, 17, 18, 19, 20, 21, 23, 24, 25, 28, 29,<br />
30, 31, 35, 36, 37, 38, 39, 40, 41, 42, 43, 46, 48,<br />
56, 57, 58, 63, 96, 109, 110, 112, 115, 116, 117,<br />
119, 120, 121, 126, 130, 134, 135, 141, 143, 144,<br />
148, 149, 150, 151, 153, 154, 155, 156, 157, 158,<br />
162, 163, 164, 165, 166, 167, 168, 173, 175, 182,<br />
183, 194, 196, 198, 211, 212, 218, 233, 237, 238,<br />
239, 246, 253, 254, 256, 275, 276, 277, 279, 280,<br />
281, 282, 283, 284, 285, 288, 291, 292, 294<br />
measurement, x, 44, 46, 167, 209, 210, 212, 213,<br />
214, 234, 244<br />
mechanical behavior, 46, 149, 162, 164<br />
mechanical degradation, 236<br />
mechanical energy, 198, 200, 257<br />
mechanical properties, ix, xi, 44, 61, 69, 109, 111,<br />
112, 113, 117, 119, 124, 130, 135, 144, 207, 255,<br />
275, 276, 277, 281, 282<br />
mechanical stress, 27<br />
media, 120, 279<br />
melt(s), 117, 118, 120, 123, 125, 133, 134, 135, 136,<br />
163<br />
melting, 276, 277, 279, 282<br />
melting temperature, 276<br />
membranes, 59, 66, 73, 86, 105<br />
MEMS, 257<br />
metal oxide, 115<br />
metallurgy, 276, 277, 282<br />
metals, vii, 2, 47, 69, 114, 116, 117, 118, 119, 126,<br />
209, 281, 294<br />
Micarta, vii<br />
micrometer, 277, 281<br />
microscopy, 218<br />
microstructure(s), 2, 4, 22, 28, 41, 42, 43, 280, 281,<br />
285, 288, 289, 291, 292<br />
military, vii<br />
Ministry of Education, 272<br />
mixing, 78, 115, 125, 136<br />
MMA, 76, 77, 78, 79, 83, 88, 89, 90, 100, 101, 106<br />
MMCs, 276, 283<br />
mobility, 141, 161<br />
modeling, 4, 43, 52, 105, 116, 124, 204<br />
models, vii, viii, xi, 1, 2, 3, 4, 5, 13, 14, 18, 20, 30,<br />
41, 42, 43, 44, 45, 47, 50, 103, 105, 166, 202, 215,<br />
220, 221, 222, 229, 232, 236, 257, 258, 266, 267
modulus, 5, 12, 26, 42, 44, 46, 49, 96, 116, 118, 119,<br />
120, 130, 131, 135, 141, 143, 144, 150, 156, 157,<br />
175, 176, 184, 185, 186, 187, 188, 218, 223, 276,<br />
283, 284, 286<br />
moisture, vii, 1, 2, 4, 5, 8, 9, 10, 11, 13, 15, 17, 19,<br />
20, 21, 23, 25, 26, 27, 42, 43, 44, 45, 48, 119, 287<br />
moisture content, 5, 9, 11, 20, 21, 27, 43<br />
moisture sorption, 44<br />
mold, 121, 122, 123, 124, 138<br />
moldings, 123<br />
molecular weight, 118, 127, 135<br />
molecules, 118, 119<br />
monomer, 119, 123<br />
Monte Carlo, x, 129, 131, 148, 149, 162, 163, 164<br />
Moon, 135, 163<br />
morphology, 2, 4, 7, 10, 11, 22, 42, 43, 46, 111, 276<br />
Moscow, 163<br />
motion, 141, 223, 229, 257<br />
moulding, 212<br />
movement, 141, 223, 224, 231, 283, 288, 290<br />
MTS, 191<br />
multiple factors, 43<br />
multiplicity, 31, 165<br />
multiwalled carbon nanotubes, 46<br />
N<br />
nanocomposites, xi, 112, 113, 114, 115, 116, 119,<br />
126, 130, 143, 161, 162, 163, 275, 276, 279, 280,<br />
281, 282, 283, 285, 287, 288, 294<br />
nanocrystalline metals, 281, 294<br />
nanofibers, 112, 114, 115, 120<br />
nanofillers, 115<br />
nanometer, xi, 113, 275, 277, 281<br />
nanometer scale, xi, 113, 275<br />
nanoparticles, ix, xi, 113, 115, 116, 118, 129, 130,<br />
133, 136, 139, 141, 143, 147, 161, 162, 275, 277,<br />
278, 279, 280, 281, 282, 285, 287, 288, 289, 294<br />
nanostructured materials, 288<br />
nanostructures, 116, 288<br />
nanotechnology, ix, 109, 110, 111, 120, 126<br />
nanotube(s), 48, 50, 114, 130<br />
nanowires, 114<br />
NASA, 122, 127, 204<br />
National Science Foundation, 162<br />
negative relation, 201<br />
negativity, 80<br />
Netherlands, 49, 103, 105, 106, 116, 127<br />
network, 46, 141<br />
neutrons, 45<br />
New Orleans, 103<br />
New York, 47, 101, 127, 163, 295<br />
Newton, 72, 232<br />
Index 303<br />
next generation, ix, 109<br />
nickel, 163<br />
nitrogen, 120<br />
nodes, 223, 231, 232<br />
noise, 133, 143, 167, 173, 174<br />
non-linear, 143, 231, 232, 260, 272<br />
O<br />
observations, 34, 218<br />
oil, 123<br />
one dimension, 80<br />
optimization, viii, 51, 52, 53, 54, 55, 61, 64, 65, 66,<br />
67, 68, 69, 70, 71, 72, 73, 74, 75, 78, 79, 80, 81,<br />
82, 83, 84, 85, 88, 89, 94, 95, 96, 97, 98, 99, 100,<br />
101, 102, 103, 104, 105, 106, 110, 117, 122, 124<br />
optimization method, 52, 61, 69, 89, 100<br />
optoelectronic, 241<br />
organic fibers, 112, 118, 119, 120<br />
organization, 121, 132<br />
orientation, 42, 57, 58, 61, 62, 65, 66, 67, 70, 72, 73,<br />
75, 78, 87, 88, 96, 97, 99, 103, 105, 111, 175, 182,<br />
183, 246<br />
oxidation, 44, 120, 132<br />
oxide(s), 14, 283, 288<br />
oxygen, 132, 287<br />
P<br />
packaging, 276<br />
PAN, 112, 120, 125, 131<br />
parameter, 64, 66, 80, 89, 90, 91, 103, 150, 154, 160,<br />
197, 204, 211, 217, 238, 294<br />
Paris, 46, 101, 103, 235<br />
particles, xi, 9, 110, 114, 115, 118, 130, 132, 133,<br />
141, 143, 162, 275, 276, 277, 278, 280, 281, 282,<br />
283, 288, 290, 291, 294<br />
passive, 167, 212<br />
PEEK, 120<br />
pendulum, 49<br />
percolation, 116<br />
perforation, 191, 239, 240, 241, 244, 245, 246, 248,<br />
249, 250, 251, 252, 253, 254, 255<br />
performance, vii, xi, 2, 88, 110, 111, 113, 115, 117,<br />
118, 120, 123, 125, 126, 130, 238, 239, 246, 257,<br />
258, 266, 275, 294<br />
permeability, 113, 122, 123, 124, 135<br />
phenolic resins, 123<br />
physical and mechanical properties, 277<br />
physical properties, ix, 109<br />
piezoelectric, xi, 125, 133, 167, 191, 241, 257, 258,<br />
259, 261, 262, 263, 266, 267, 269, 272, 273
304<br />
piezoelectricity, 261<br />
pitch, 125<br />
planning, 122<br />
plasma, 117, 120, 279<br />
plastic deformation, xi, 275, 283, 288, 290, 294<br />
plastic strain, 291<br />
plasticity, 294<br />
plastics, 166, 233<br />
platelets, 115, 130<br />
plywood, vii<br />
PM, 276, 277, 279, 281, 283, 287, 288, 294<br />
PMMA, 255<br />
Poisson ratio, 131<br />
polarization, xi, 257, 258, 260, 261, 262, 263, 268,<br />
269, 272, 273<br />
polyarylate, 118<br />
polycarbonate, 163<br />
polycrystalline, 4, 14, 15, 48, 163<br />
polyester(s), 47, 118, 123, 204, 254, 255<br />
polyetheretherketone, 120<br />
polyethylene, vii, 50, 118, 119, 127, 135, 163<br />
polymer(s), vii, ix, x, 1, 2, 4, 6, 15, 41, 44, 48, 110,<br />
111, 112, 114, 115, 116, 118, 119, 120, 123, 124,<br />
125, 126, 130, 134, 135, 141, 143, 161, 209, 215,<br />
236, 255<br />
polymer chains, 119, 141, 161<br />
polymer composites, x, 209, 215, 255<br />
polymer materials, ix, 110, 115<br />
polymer matrix, 6, 41, 44, 115, 119, 125, 130, 141<br />
polymer solutions, 134<br />
polymer systems, 112<br />
polymer-based composites, 44<br />
polymeric composites, 110, 117<br />
polymeric materials, 112<br />
polymeric matrices, ix, 110<br />
polystyrene, 163<br />
poor, 78, 134, 135, 136<br />
porous media, 134<br />
ports, 122<br />
Portugal, 102<br />
power, 110, 120, 133, 215, 238, 243, 283<br />
PPS, 214, 217, 221, 225, 226, 227, 228<br />
prediction, 13, 30, 31, 32, 35, 44, 47, 206, 254<br />
pressure, x, 64, 92, 104, 121, 122, 123, 133, 138,<br />
165, 205, 224, 241, 277, 291, 293, 294<br />
probability, 150, 166<br />
probe, 133<br />
production, ix, 110, 120, 122, 124, 125, 212<br />
program(ming), viii, 51, 52, 56, 101, 105, 106, 125,<br />
155, 157<br />
proliferation, 112<br />
propagation, 69, 162, 167, 220, 225, 238, 256<br />
proposition, 205<br />
Index<br />
prototype, 124<br />
PTFE, 162<br />
pulses, 219<br />
pultrusion, 121, 125<br />
pumps, 124<br />
PVC, 46<br />
pyrolysis, 120<br />
Q<br />
qualifications, 114<br />
quantitative estimation, 184<br />
R<br />
radial distance, 28<br />
radiation, 115<br />
radius, 28, 53, 191, 226, 241, 266, 269<br />
range, 6, 13, 31, 34, 42, 44, 49, 120, 122, 123, 124,<br />
130, 133, 134, 135, 147, 154, 166, 169, 187, 196,<br />
211, 239, 240, 244, 246, 247, 248, 253, 255, 263<br />
reactive groups, 119<br />
reactivity, 277<br />
realism, 4, 44<br />
reality, 167, 168, 262<br />
reasoning, 17, 22<br />
recalling, 52<br />
recognition, ix, 110<br />
reconstruction, 221<br />
recovery, 279<br />
recrystallization, 279<br />
recycling, 125<br />
redistribution, 167, 184, 215<br />
reduction, 172, 201, 212, 233, 251<br />
reference frame, 8, 32<br />
refining, 291<br />
reflection, 167<br />
refractive index, 211, 218<br />
regression, 185, 201, 239, 240, 285<br />
regression line, 240, 285<br />
reinforcement, ix, 3, 15, 17, 104, 109, 110, 112, 119,<br />
125, 126, 210, 254, 276, 277, 279, 284<br />
reinforcing fibers, 18, 41, 42<br />
rejection, 167<br />
relationship(s), x, 116, 118, 130, 134, 149, 160, 168,<br />
239, 241, 243, 247, 259<br />
relaxation, 81, 154<br />
relevance, 248<br />
reliability, viii, 2, 32, 43, 44, 166<br />
renewable energy, 111, 112, 115<br />
repair, 117, 118, 163<br />
reserves, 125
esin reaction, 133, 136<br />
resins, 39, 41, 43, 46, 118, 120, 122, 123, 124, 125,<br />
130<br />
resistance, x, xi, 2, 104, 111, 113, 116, 117, 118,<br />
119, 132, 162, 166, 198, 205, 209, 213, 214, 234,<br />
241, 255, 275, 283, 285, 294<br />
resolution, 220<br />
resources, 54, 172<br />
revolutionary, 111<br />
rheological properties, 50<br />
Rhode Island, 104<br />
rice, 125<br />
robust design, 102<br />
rods, vii<br />
rolling, 288, 290<br />
room temperature, 122, 124, 276, 288, 290, 291, 292,<br />
293<br />
Royal Society, 46<br />
RTS, 166, 201<br />
rubber, 130, 162<br />
S<br />
SA, 81, 102, 105, 163, 205, 233<br />
SAD, 293<br />
safety, 94<br />
sample(ing), 8, 31, 34, 136, 147, 215, 220, 221, 224,<br />
238, 241, 277<br />
satellite, 122<br />
satisfaction, 54<br />
saturation, 239<br />
savings, 70, 124<br />
sawdust, vii<br />
scattering, 243, 253<br />
science, ix, 110, 126, 257<br />
scull, vii<br />
search, 66, 101, 103<br />
selected area electron diffraction, 293<br />
selecting, 100<br />
SEM micrographs, 147<br />
sensing, x, 125, 209, 212<br />
sensitivity, 17, 56, 61, 66, 81, 94, 103, 105, 119, 139,<br />
294<br />
sensors, 115, 125, 166, 167, 191, 192, 211, 212, 221,<br />
234, 263<br />
separation, 135<br />
series, 134, 173, 184, 213, 239, 241, 248<br />
shape, 22, 42, 49, 53, 54, 69, 74, 101, 103, 104, 114,<br />
115, 122, 124, 125, 132, 150, 196, 221, 225, 226,<br />
228, 276<br />
shape-memory, 125<br />
shear, x, 29, 31, 32, 33, 35, 46, 48, 49, 59, 94, 103,<br />
104, 129, 147, 148, 149, 150, 151, 155, 158, 164,<br />
Index 305<br />
195, 201, 210, 217, 218, 235, 255, 262, 265, 269,<br />
272, 277, 283<br />
shear strength, 155<br />
shock waves, 133<br />
shortage, 119, 125<br />
Si3N4, 276, 281, 282, 283, 285<br />
SIC, 129<br />
sign, 32<br />
signaling, 238, 249<br />
signals, 168, 169, 173, 219, 238<br />
silane, 120<br />
silica, 115<br />
silicon, ix, 14, 129, 132, 136, 163<br />
Silicon carbide, 163<br />
silver, 147<br />
simulation, x, 10, 39, 45, 124, 130, 131, 148, 155,<br />
157, 162, 163, 164, 210, 219, 221, 227, 228, 232,<br />
233, 273<br />
sine wave, 214<br />
Singapore, 125<br />
sintering, 279, 287<br />
sites, 133<br />
smart materials, 113<br />
society, ix, 109<br />
software, 45, 72, 106, 187, 266<br />
solid state, 124<br />
solvent(s), 111, 118, 120, 133, 134, 135, 136<br />
Spain, 235<br />
species, 135<br />
specific surface, 115, 277<br />
speed, 85, 94, 97, 111, 116, 150, 154, 173, 191, 193,<br />
205, 220, 255<br />
spindle, 136<br />
sports, 110, 111, 118, 124<br />
stability, 71, 113, 119, 143<br />
stages, 125, 240<br />
statistical analysis, 196<br />
statistics, 5, 6<br />
steel, 96, 191, 210, 223, 241<br />
storage, vii, 124, 135<br />
strain, x, 9, 10, 11, 13, 14, 21, 23, 31, 32, 34, 35, 36,<br />
39, 43, 44, 45, 47, 56, 61, 62, 67, 72, 82, 89, 91,<br />
92, 105, 130, 143, 144, 148, 154, 155, 156, 157,<br />
158, 160, 161, 164, 165, 167, 169, 170, 172, 175,<br />
176, 182, 183, 184, 185, 186, 187, 188, 195, 196,<br />
198, 199, 200, 201, 204, 209, 210, 211, 212, 213,<br />
214, 215, 217, 218, 225, 228, 232, 234, 235, 241,<br />
254, 258, 259, 260, 263, 267, 268, 269, 272, 277,<br />
285, 288, 291, 293, 294<br />
strategies, 172<br />
stratification, 68<br />
strength, viii, xi, 2, 3, 10, 18, 19, 30, 31, 32, 33, 34,<br />
35, 36, 37, 38, 39, 40, 41, 42, 43, 46, 51, 53, 66,
306<br />
67, 69, 70, 71, 72, 73, 84, 94, 101, 102, 103, 105,<br />
106, 111, 113, 116, 117, 118, 119, 120, 121, 122,<br />
126, 130, 131, 143, 144, 150, 154, 155, 156, 157,<br />
158, 160, 161, 162, 163, 164, 166, 168, 169, 175,<br />
182, 183, 203, 204, 206, 207, 215, 233, 237, 238,<br />
241, 243, 275, 276, 277, 281, 282, 285, 288, 290,<br />
291, 294<br />
stress, viii, ix, x, xi, 2, 21, 24, 25, 27, 29, 31, 32, 33,<br />
34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 47, 48,<br />
49, 56, 59, 71, 96, 102, 110, 115, 117, 123, 130,<br />
143, 148, 149, 150, 151, 154, 155, 156, 157, 158,<br />
161, 165, 167, 168, 172, 175, 176, 177, 178, 179,<br />
180, 181, 184, 185, 186, 187, 210, 212, 214, 215,<br />
217, 221, 224, 225, 228, 235, 257, 258, 261, 262,<br />
264, 265, 268, 269, 271, 272, 275, 282, 283, 284,<br />
285, 286, 287, 288, 289, 290, 291, 293, 294<br />
stress-strain curves, 157, 175, 187, 212, 289, 291,<br />
293<br />
stretching, 120<br />
structural characteristics, 119<br />
suffering, 125<br />
Sun, 127, 273<br />
superplasticity, 281, 294<br />
suppliers, 120, 122<br />
supply, 113, 133<br />
supply chain, 113<br />
surface area, 130<br />
surface energy, 119<br />
surface tension, 134, 135<br />
surface treatment, 205<br />
Sweden, 103<br />
switching, xi, 257, 258, 260, 261, 262, 263, 264,<br />
268, 269, 270, 272, 273<br />
symbols, 76, 78<br />
symmetry, 7, 132, 222, 232, 233, 259, 262, 266<br />
synthesis, 106, 173, 239, 276<br />
systems, vii, x, 110, 113, 115, 116, 117, 121, 123,<br />
124, 129, 130, 141, 143, 162, 254<br />
T<br />
tanks, vii, 124<br />
Taylor series, 55, 74, 75<br />
technology, viii, ix, 109, 110, 112, 117, 118, 119,<br />
126, 204, 233, 257<br />
teflon, 138<br />
TEM, 133, 280, 287, 288, 289, 290, 292, 293, 294<br />
temperature, ix, x, 2, 14, 21, 43, 44, 115, 116, 117,<br />
118, 121, 123, 124, 125, 129, 131, 132, 136, 138,<br />
141, 217, 279, 283, 285, 286, 287, 288, 290, 294<br />
temperature dependence, 290<br />
Index<br />
tensile strength, x, 28, 44, 46, 118, 119, 155, 156,<br />
160, 161, 162, 164, 165, 166, 167, 201, 204, 282,<br />
288, 293<br />
tensile stress, 28, 33, 148, 156, 160, 224<br />
tension, 32, 33, 35, 36, 41, 48, 134, 144, 156, 162,<br />
196, 210, 212, 213, 214, 215, 217, 221, 229, 234,<br />
236, 267, 268, 269<br />
test data, 31, 144, 145, 147, 253<br />
Texas, 163<br />
textiles, vii<br />
TGA, x, 129, 141, 142, 161<br />
theory, 9, 25, 34, 43, 46, 56, 59, 79, 156, 175, 215,<br />
233<br />
thermal expansion, vii, 1, 4, 10, 13, 14, 16, 17, 42,<br />
43, 45, 113, 117, 119<br />
thermal properties, viii, 1, 13, 42<br />
thermal stability, 113, 132, 141, 142, 161<br />
thermodynamics, 44, 134<br />
thermo-mechanical, 11, 12, 14, 43, 49, 148, 164<br />
thermomechanical treatment, 288<br />
thermoplastic(s), vii, ix, 110, 113, 115, 117, 123,<br />
125, 126, 163, 210, 212, 213, 214, 218, 219, 221,<br />
234, 235, 256<br />
Third World, 101<br />
threat(s), xi, 119, 125, 237, 238, 243<br />
threshold(s), xi, 35, 116, 166, 174, 191, 193, 201,<br />
205, 237, 238, 239, 240, 241, 242, 243, 246, 248,<br />
253, 254, 256, 283, 285, 288, 290<br />
titanium, 14, 46, 112, 117<br />
Tokyo, 105<br />
toluene, 135<br />
topology, 52, 54, 67, 68, 71, 72, 74, 78, 81, 96, 97,<br />
101, 102, 106<br />
tracking, 212<br />
transducer, 219<br />
transformation, 11, 21, 45, 132, 163<br />
transition(s), vii, viii, 1, 2, 3, 4, 6, 8, 9, 10, 12, 13,<br />
14, 17, 18, 19, 20, 21, 35, 36, 41, 42, 43, 44, 45,<br />
47, 198, 228, 288<br />
transmission, 110, 218, 219<br />
transparency, 218<br />
transport, 44<br />
transportation, 118, 122, 124<br />
treatment methods, ix, 110, 115, 120<br />
trend, ix, 110, 113, 117, 118, 120, 121, 122, 170,<br />
172, 173, 182, 183, 184, 196, 198, 200, 242, 243,<br />
252, 253<br />
trial and error, 124<br />
triangulation, 167<br />
UK, 234, 235, 254<br />
U
ultrasonic waves, 133, 219, 277, 278, 294<br />
ultrasound, 133, 163, 219, 220<br />
uniaxial tension, 31<br />
uniform, 21, 45, 64, 92, 122, 124, 133, 134, 136,<br />
147, 241, 294<br />
universities, 121<br />
updating, 77<br />
USDA, 125<br />
UV, 119, 121<br />
V<br />
vacuum, 121, 122, 123, 124<br />
validation, x, 166, 209, 229, 236<br />
values, viii, 2, 4, 8, 13, 18, 19, 20, 41, 52, 54, 58, 61,<br />
64, 65, 66, 70, 76, 77, 79, 81, 84, 88, 89, 90, 95,<br />
110, 116, 122, 134, 143, 156, 166, 173, 176, 182,<br />
185, 186, 187, 193, 196, 200, 201, 204, 238, 241,<br />
243, 246, 247, 248, 252, 253, 254, 257, 258, 263,<br />
265, 267, 272, 277, 283, 285<br />
vapor, 46<br />
variable(s), viii, x, xi, 26, 51, 52, 53, 54, 55, 61, 62,<br />
65, 66, 67, 69, 70, 71, 72, 73, 74, 75, 76, 78, 79,<br />
80, 81, 82, 83, 84, 88, 89, 90, 94, 95, 96, 97, 98,<br />
99, 100, 101, 103, 121, 169, 201, 202, 237, 239,<br />
240, 244, 246, 250, 253<br />
variation, 44, 57, 61, 82, 83, 84, 89, 90, 91, 99, 100,<br />
144, 147, 172, 200, 211, 228, 285<br />
vector, 25, 52, 59, 258, 261<br />
vehicles, 115<br />
velocity, x, 165, 166, 205, 207, 236, 237, 238, 241,<br />
243, 254, 255<br />
versatility, 124<br />
vessels, x, 165, 205<br />
vibration, 53, 64, 66, 71, 125<br />
Index 307<br />
Vietnam, 105<br />
vinylester, 254<br />
Viscoelastic, 49<br />
viscosity, 115, 122, 123, 134, 135<br />
visualization, 234<br />
W<br />
Washington, 105<br />
wave propagation, 167<br />
wavelengths, 211<br />
wear, 113, 118, 162, 275<br />
Weibull distribution, 150<br />
weight ratio, 2, 51, 166<br />
weight reduction, 118<br />
welding, 279<br />
wettability, 120, 278<br />
wetting, 112, 119, 134, 135<br />
wheat, 125<br />
wind, 110<br />
wood, vii<br />
workers, 130<br />
writing, 6, 22, 45<br />
X<br />
X-ray, 47, 121, 166, 220, 236<br />
X-ray diffraction (XRD), 47<br />
xylene, 135<br />
Y<br />
yield, 32, 42, 45, 121, 124, 225, 238, 277, 281, 282,<br />
288, 290, 291, 294