Information Technology Reference
In-Depth Information
Fig. 25 Bounds of
MLP-NARX prediction for
multivariable and multiaxial
fatigue life assessment of
E-glass/polyester
([0/(
±
45)
2
/0]
T
). See the
corresponding Fig.
18
and lower bounds of NN-NARX model prediction for the stress ratios of R = 0.5,
−
2, respectively. Note that the stress ratio values of the material were
chosen as representative examples as their corresponding fatigue lives scatters are
wide in comparison to those of other stress ratio values.
One can observe from Figs.
26
,
27
and
28
that the upper and lower bounds of
NN-NARX prediction can adequately describe the fatigue lives scatters produced
from fatigue testing, while at the same time the obtained mean lifetimes from the
NN-NARX model can give good prediction of the experimental data. Clearly, the
results show the applicability of the present NN modeling for multivariable and
multiaxial fatigue life assessment thus building spectrum fatigue analysis. More-
over, it is worth noting the clear concept of CLD which has been employed and
linked to the selection of NARX structure in the modeling approach.
In addition, considering that the present NARX modeling is made up of one-step
ahead prediction, it is also worth noting that the presented results give con
0.5 and
−
dence
level to the user in using NN with NARX structure for describing scatter of fatigue
lives of composite materials. Moreover, the simulation task can be also extended for
other probabilities of failures (10 and 90 % probabilities of failures). As has been
stated by Bucar et al. (
2007
), once an NN con
guration has been designed and
optimized, the network can be used suitably for describing the scatter of S-N curves
for arbitrary set of in
uential factors. Hence, it would be interesting to investigate
the fatigue lives scatter of composite materials associated to various probabilities of
failure in comparison to those of isotropic materials such as metals (Klemenc and
Fajdiga
2012
,
2013
) using the present NN-NARX modeling framework. Such a
modeling task will be the further subject of the NN-NARX modeling study.
fl
Search WWH ::
Custom Search