System Identification Technique and Neural Networks for Material Lifetime Assessment Application - Complex System Modelling and Control Through Intelligent Soft Computations - page 761

ratios were used in the training set, the NN model developed was able to generalize
well and gave reasonably accurate fatigue life prediction under a wide range of
stress ratio values. The reliability and accuracy of the NN prediction were quan-
ti

ed by a small MSE value.
Klemenc and Fajdiga ( 2012 ) have employed a class of evolutionary algorithms
to estimate S-N curves and their scatter using a differential ant-stigmergy algorithm
(DASA). In Klemenc and Fajdiga ( 2013 ), the authors have extended the use of
evolutionary algorithms of GA and DASA for estimating E-N curves and their
scatter.
From the recent investigations, soft computing techniques have been proven to
be a suf cient tool for modelling fatigue life of composite materials, ranging from

unidirectional to multidirectional laminate, with the potential for fatigue life
assessment under variable amplitude loadings. In addition, the use of soft com-
puting techniques has been also a new route in the task of fatigue life assessment
where the main aim is to develop soft computing models that produce reliable
prediction using a limited body of fatigue data, which in turn can support design
decisions very soon and reliably.
However, it is still necessary to optimize NN prediction of composite fatigue life
under variable amplitude loading by utilizing less fatigue data but at the same time
ensuring reasonably accurate prediction to take full advantage of the NN potential
for much more efficient fatigue life assessment. Therefore, further confirmation and
examination for various composite materials or different NN architectures and
frameworks are required. Moreover, it will be also valuable to

cient
training data set for various types of composites in relation to the full utilization of
fatigue data available. It is worth to point out here that fatigue behavior is still so
complicated that the problem requires more effort before NN can be used with more
con

find the suf

dence (Zhang and Friedrich 2003 ).

1.2 Motivation and Objectives

In the previous works, investigations were only focused on the application of
various soft computing models, in particular NN, in the task of material lifetime
assessment. As a matter of fact, however, no further attempt has been made to link
and ground the use of NN models to one important concept of fatigue life analysis
and design, namely constant life diagrams or CLD. It will be shown that the
utilization of CLD concept could lead to the use of non-linear auto-regressive
exogenous inputs (NARX) structure as a natural of choice of NN con guration for

the task of material lifetime assessment.
Moreover, in term of fatigue life assessment of composite materials under
multivariable amplitude loading, previous investigations were only focused on the
utilization of NN for multivariable amplitude loadings with respect to different
stress ratio values-R. No further attempt, however, so far has been devoted to the
utilization of NN for fatigue life assessment of composite materials under both