18.06.2013 Views

LIBRARY ı6ıul 0) - Cranfield University

LIBRARY ı6ıul 0) - Cranfield University

LIBRARY ı6ıul 0) - Cranfield University

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

experimental conditions and are not valid outside the range of the experiment [refs.<br />

165,166]. They provide a fast approach to model building without the need for<br />

extensive and accurate knowledge of all process variables [ref. 165]. Theoretical<br />

models are developed from first principles, based on the process scientific and<br />

physical facts. They are usually more reliable and flexible in predicting the dynamic<br />

behaviour of a process over a wide operating range. However, depending on the<br />

complexity of the process phenomena, the resulting model equations may be difficult,<br />

if not impossible, to solve [ref. 164].<br />

Semi-empirical models are developed with regard to established fact or<br />

knowledge of the process. They combine the advantages of theoretical and empirical<br />

models to achieve modelling accuracy. They offer a standardised process modelling<br />

strategy, even when the basic principles of the process are not understood [ref. 51]<br />

2.6.4.1 Experimental design<br />

Experimental design is the process of planning experiments so that appropriate<br />

data will be collected for a representative statistical analysis to be performed, to reach<br />

valid and objective conclusions [ref. 167]. The experimental design objectives range<br />

from process analysis to developing models for process control and establishing<br />

correlation between process inputs and outputs.<br />

The most important part of planning an experiment, after defining the<br />

objectives of the investigation, is the identification of the inputs and outputs of the<br />

process, i. e. the factors that might affect the process behaviour, within a practical<br />

range relevant to the process, and the factors that can be used for assessing this<br />

change of behaviour [refs. 51,167]. Thought must be given to how the response will<br />

be measured and the probable accuracy of these measurements.<br />

Variables in an experimental design are usually categorized into two groups,<br />

namely independent variables and dependent variables. Generally, the parameters that<br />

are directly controllable, such as machine settings, are chosen as independent<br />

variables. The responses of the process to changes in the independent variables are<br />

considered as dependent variables. In gas metal arc welding, variables such as tip-to-<br />

workpiece distance, welding speed, welding voltage, wire feed rate and gap are often<br />

treated as independent variables, whereas welding current and bead geometry are<br />

normally considered as dependent variables [refs. 51,168].<br />

There are two approaches to experiment design. The planned and the intuitive<br />

sequential experiment. The planned approach is usually based on full or fractional<br />

factorial designs, while in a sequential experiment a trial and error approach is used,<br />

and various input variable combinations are selected using process knowledge [ref.<br />

51].<br />

Factorial design is an experimental design technique with which, for any<br />

complete trial or replication of the experiment, all possible combinations of factor<br />

levels are investigated [ref. 167]. It provides a systematic way of performing the least<br />

number of experiments to obtain a maximum amount of information in a multivariable<br />

environment. Factorial design is orthogonal by nature, or in other words no<br />

correlations exist between process independent variables. Consequently, experimental<br />

errors are normally distributed [refs. 51,169]. It is only effective, however, for a<br />

45

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!