System Identification Technique and Neural Networks for Material Lifetime Assessment Application - Complex System Modelling and Control Through Intelligent Soft Computations - page 777

the experimental data, the conservative prediction results were still preserved. For
Material II-45 1 were good

for all stress levels involved. As can be seen, the NN fatigue life prediction results
can closely follow the experimental data trend. The coef

° , the NN fatigue life prediction results of R = 0.5 and −

cient of determination (R 2 )
of the fatigue life prediction can be also observed. For Materials I and II, best values
of the coef cient of determination was 0.989 and 0.9653, respectively. For all the

materials examined, the values of R 2 was ranging from 0.7923 to 0.989.

5.2 Fatigue Life Assessment of Multivariable Amplitude
Loadings with RBFNN-NARX Model

In this section, fatigue life assessment of multivariable amplitude loading with the
RBFNN-NARX model is presented.
Figure 15 presents the NN fatigue life prediction of Material I at stress ratios R of
the testing set using the RBFNN-NARX model. It can be seen that the RBFNN-
NARX model prediction results were consistent with the experimental data
showing also the applicability of the RBFNN-NARX model for this problem. The
RBFNN-NARX model also showed its ability to dynamically predict the fatigue
lives sliding over each stress level in a fashion of spectrum loading made up by
several R values.
Figures 16 and 17 further depict respectively fatigue life prediction of Material
II-on-axis and Material II-45 . It can be also seen that in general the RBFNN-

NARX model prediction results can fairly follow the experimental data trend.

°

Fig. 15 Fatigue lives
predicted by the RBFNN-
NARX structure for the tested
sets: R = 0.9, 0.8, 0.7, 0.5,
− 2 of Material I

(from left to right)

0.5, − 1 and −