System Identification Technique and Neural Networks for Material Lifetime Assessment Application - Complex System Modelling and Control Through Intelligent Soft Computations - page 770

Table 2 Materials examined with the orientations and the training and testing sets employed for
fatigue life assessment of multivariable and multiaxial loadings
Material Fatigue data as training

set: R and

Fatigue data as testing
set: R and ʸ values ʸ values

E-glass/polyester [0/( ± 45) 2 /0] T
(Vassilopoulos and Philippidis 2002 )

R = 0.1: ʸ =0 ° R = 10:
ʸ =0 °

R = 0.5: ʸ =0 °
R = − 1: ʸ =0 °

R = 0.1: °

R = − 1: ʸ =30 °
R = 10:

ʸ =15

ʸ =30 °

R = 0.1: °

R = 0.5: ʸ =45 °
R = − 1: ʸ =45 °
R = 10:

ʸ =45

ʸ =45 °

R = °

R = 10: ʸ =60 °
R = 0.1:

− 1: ʸ =60

ʸ =75 °

R = 0.1: °

R = − 1: ʸ =90 °
R = 10:

ʸ =90

ʸ =90 °

E-glass fabrics/epoxy [ 45/]

(Mandell and Samborsky 2010 )

± 45/0 4 / ± R = 0.1: ʸ =0 ° R = 10: R = 0.5: °

R = − 0.5: ʸ =0 °
R = − 1: ʸ =0 °
R =

ʸ =0

ʸ =0 °

− 2: ʸ =0 °

R = 0.1: °

R = 0.5: ʸ =90 °
R =

ʸ =90

− 0.5: ʸ =90 °

R = °

R = − 2: ʸ =90 °
R = 10:

− 1: ʸ =90

ʸ =90 °

been also shown in other field of application, for instance in Azar ( 2013 ) for

medical applications. Moreover, Bayesian regularization was incorporated (Foresee
and Hagan 1997 ) to accommodate the noise which may be present in the target data
as well as to deal with limited training data that may lead to ill-posed problem.

4.3.1 Adaptation of Bayesian Framework Within
the Levenberg-Marquardt Algorithm

Bayesian regularization was incorporated in the Levenberg-Marquardt algorithm
through the modi ed objective function of NN, E(w), as follows: