System Identification Technique and Neural Networks for Material Lifetime Assessment Application - Complex System Modelling and Control Through Intelligent Soft Computations - page 784

Here, the informative bounds of NN prediction in the fatigue life assessment of
E-glass/polyester ([90/0/ 45) 2 /0] T ) as listed

respectively in Tables 1 and 2 are described as the fatigue life prediction tasks
represent multivariable and multiaxial fatigue life assessment.
Figures 24 and 25 depict the informative bounds of NN fatigue life prediction for
these materials. It can be seen from Figs. 24 and 25 that the NN-NARX model
employing the Bayesian regularization can conveniently provide the informative
upper and lower bounds of prediction as the error bars for the mean lifetimes. It is
interesting to note that the error bars for fatigue lives of the

± 45/0] S ) and E-glass/polyester ([0/( ±

first testing set (R = 0.9

in Fig. 24 and R0.5 in Fig. 25 ) are narrowest in comparison to the others.

The presented results clearly show the feasibility of the present spectrum fatigue
life analysis as the testing stress ratio sets are moving from the tensile-tensile to the
compressive-compressive sector in the CLD region, thus forming a spectrum of
loading conditions. In addition, as the lifetime prediction will slide over the suc-
cession of testing stress ratio sets, the error bars will also vary accordingly based
upon the one-step ahead prediction with respect to each stress level S corresponding
to stress ratio values R.
It should be pointed out here that the bounds of NN prediction are also useful in
describing the scatter of fatigue lives due to variability in aspects of material
(material from different batches or manufacturers), specimen (production and sur-
face quality), fatigue load (types of load, frequency or equipment), environment
(controlled temperature and humidity), personal or laboratory technicians skill
(Schijve 2009 ) as well as other arbitrary set of in

ʸ 0 °

uential factors (Bucar et al. 2007 ).
Figures 26 , 27 and 28 depict the fatigue lives scatter for E-glass/polyester ([90/0/

fl

± 45/0] S ) from experiments (Mandell and Samborsky 2010 ) along with the upper

Fig. 24 Bounds of MLP-
NARX prediction for
multivariable fatigue life
assessment of E-glass/
polyester ([90/0/ 45/0] S ) for

the testing sets: R = 0.9, 0.8,
0.7, 0.5,

±

− 0.5, − 1 and − 2 (see

Table 1 )