28.02.2014 Views

The Development of Neural Network Based System Identification ...

The Development of Neural Network Based System Identification ...

The Development of Neural Network Based System Identification ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

2.3 NEURAL NETWORK BASED SYSTEM IDENTIFICATION 29<br />

complex problems in aeronautical and flight control system applications. For example,<br />

the NN approach has been used to identify aerodynamic coefficients <strong>of</strong> an aircraft and<br />

rotorcraft system in Suresh et al. [2003], Dennis and Stengel [1992]. NASA Dryden<br />

Flight Research Centre and Boeing company have developed an adaptive flight control<br />

system that utilises the NN model to predict the stability and control parameters <strong>of</strong> the<br />

F-15 tactical jet fighter. This prediction data was then continuously used to optimise<br />

the control system and to assist the pilot in damage or failure situations <strong>of</strong> the aircraft<br />

[Urnes et al., 2001]. Furthermore, NN is also used to detect faults in the helicopter<br />

transmission system and has been proven reliable to classify faults even with different<br />

vibration signatures data [Parker et al., 1993]. Review papers commenting on the usage<br />

<strong>of</strong> the NN approach in the aeronautics and flight control system design have been<br />

contributed by Calise and Rysdyk [1998] and Faller and Schreck [1996]. <strong>The</strong>re are two<br />

main advantages for transitioning neural network technology into UAS: (1) the potential<br />

to reduce the flight control system design and development costs, (2) possible reduction<br />

in cost associated with the development <strong>of</strong> large aerodynamic database [Calise and<br />

Rysdyk, 1998].<br />

2.3.1 NN Model Structure Used for Dynamics Modelling<br />

<strong>The</strong>re exist several ANN model structures that have been used to model the dynamics<br />

representation <strong>of</strong> a UAV helicopter. <strong>The</strong> NN model structures differ from each other<br />

depending on how their processing units or neurons are interconnected. This section<br />

covers the main types <strong>of</strong> network structures used for modelling the helicopter flight<br />

dynamics.<br />

Since the helicopter dynamics is non-linear, the NN system identification approach<br />

using the NNARX (<strong>Neural</strong> <strong>Network</strong>-Auto Regressive structure with eXtra inputs) model<br />

structure can be used to address such a problem. Here, the linear model structure<br />

such as the ARX model structure was introduced as the basic model structure while<br />

the NN network was used to introduce non-linearity to the model estimation. Similar<br />

to stability features <strong>of</strong> the ARX model structure, the prediction from the NNARX<br />

model was considered stable since there was only a pure algebraic relationship between

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

Saved successfully!

Ooh no, something went wrong!