System Identification Technique and Neural Networks for Material Lifetime Assessment Application - Complex System Modelling and Control Through Intelligent Soft Computations - page 759

computing techniques are lying on emulating relationships in sets of input data to
subsequently predict the outcome of another new set of input data, for examples,
another composite system or stress environment.
NN have been previously employed for elevated temperature creep-fatigue life
prediction (Venkatesh and Rack 1999 ), fracture toughness and tensile strength of
microalloy steel evaluation (Haque and Sudhakar 2002 ), prediction of fatigue crack
growth rate in welded tubular joints (Fathi and Aghakouchak 2007 ), while genetic
algorithm (GA) has been employed as parameterization tool for fatigue crack
growth of Al-5052 (Bukkapatnam and Sadananda 2005 ) as well as optimization
tool for fuzzy logic and NN models in life prediction of boiler tubes (Majidian and
Saidi 2007 ). Moreover, recently NN has been also employed to build a probability
distribution function for fatigue life prediction of steel under step-stress conditions
(Pujol and Pinto 2011 ).
In recent years, soft computing techniques have found their applications in the

field of fatigue life assessment of composite materials in particular under variable
amplitude loading conditions (Aymerich and Serra 1998 ; Lee and Almond 2003 ).
The use of soft computing techniques in fatigue life assessment of composite
materials has a wide range of applications from unidirectional (Al-Assaf and El-
Kadi 2001 ; El-Kadi and Al-Assaf 2002 ) to multidirectional laminate (Freire Junior
et al. 2005 ; Vassilopoulos et al. 2007 , 2008 ; Freire Junior et al. 2007 , 2009 ).
In the present chapter, a framework of system identification technique based
upon nonlinear autoregressive exogenous inputs (NARX) for material lifetime
assessment using neural networks (NN) will be presented. Using the proposed
framework, material lifetime assessment can be fashioned for a wide spectrum of
loading in an ef cient manner based upon limited material lifetime data as the basis

of the NARX regressor. The key aspect of the new approach is that sliding over
one-step to one-step of the stress level so that the task of prediction dynamically
covered all loading spectrum.
The remaining of this chapter is organized as follows. Comprehensive reviews in
the modeling of fatigue life of composite materials along with the motivation and
objective for the present study are presented in Sect. 1 . In Sect. 2 , concept of
constant life diagrams (CLD) as rational of the use of the NARX structure in the
present application is brie

y described. NN architectures developed and employed
in this study are presented in Sect. 3 . Section 4 describes composite materials
examined and numerical procedures employed for the developed NN structures.
Results and discussion are presented in Sect. 5 , followed by conclusions in Sect. 6 .

fl

1.1 Soft Computing Techniques for Fatigue Life Assessment
of Composite Materials

Al-Assaf and El-Kadi ( 2001 ) and El-Kadi and Al-Assaf ( 2002 ) assessed the fatigue
life of unidirectional glass

fiber/epoxy laminae using different neural network
paradigms, namely feed forward (FF), modular (MN), radial basis function (RBF)