System Identification Technique and Neural Networks for Material Lifetime Assessment Application - Complex System Modelling and Control Through Intelligent Soft Computations - page 758

therefore has been always challenging task. Moreover, the ability to make accurate
predictions of fatigue durability is critical to the related design optimization process
(Post et al. 2008 ).
Recently, composite materials in particular

fiber reinforced polymer (FRP)
composites have become more popular materials instead of metals in many struc-
tural applications due to their excellent properties, such as high strength to weight
ratio, tailored properties along preferred direction and high corrosion resistance. For
instances, the use of composite materials has been common in many components of
automotive, aircraft, ship hull as well as wind turbine blade structures. Fatigue
characterization of the composite materials therefore is also important. In particular,
it is also desirable to understand and assess the fatigue behaviour of the materials
for an expected or anticipated spectrum or variable amplitude fatigue loading.
Nonetheless, modeling of composites fatigue life under complex and spectrum
loading conditions comes with a greater challenge to researchers in this eld.

Different with metals, more considerations must be taken into account in the
modeling of composites lifetime, such as wide variety of component materials or

fiber and matrix types, laminate design or lay-ups, anticipated failure modes, fatigue
states governed by stress ratios-R, on-axis/off-axis orientation as well as manu-
facturing methods. As a result, such a modeling task becomes complicated and
developing a universal understanding of the performance of composite materials
under spectrum fatigue loadings is also very difficult because many factors should
be included and anticipated in the model (Reifsnider 1991 ; Harris 2003 ; Passipo-
ularidis et al. 2011 ). On the other hand, in most cases, authors only had limited
experimental fatigue data in hands. It thus makes the model development is also
frequently impeded by a large amount of fatigue testing data needed, which is very
costly and time consuming to collect.
Numerous empirical and phenomenological models have been introduced over
the past 40 years of fatigue studies for composite materials (Post et al. 2008 ;
Philippidis and Passipoularidis 2007 ; Passipoularidis and Philippidis 2009 ). A well
known common approach is that the fatigue behaviour of composite materials is to
be modeled or predicted using readily collected constant amplitude fatigue data for
a material or system of materials of interest, as traditionally fatigue characterization
of a material is performed under constant amplitude sinusoidal loading. Nonethe-
less, as stated by Post et al. ( 2008 ), it is often that any empirically determined model
parameters are to be fitted to the variable amplitude fatigue data modeled. Thus the

relative accuracy remains uncertain between the empirical and phenomenological
approaches reported in the literature. As a result, many models are developed with a
speci

c material and loading con guration and their generalization to other cases

remains uncertain.
Driven by the requirement for speeding up time frame from research stage to
market place and also cutting down the associated cost, in recent years there has
been increasingly interest in pursuing and utilizing alternative approaches based
upon soft computing framework, in particular neural networks (NN), to develop
ef

cient and robust predictive model for fatigue life assessment of composite
materials. It is interesting to note that the characteristic and capability of soft