Prediction of the fatigue life of unidirectional glass fiber/epoxy composite laminae using different neural network paradigms

Predicting the fatigue life of fiber-reinforced composite materials has been investigated from a number of viewpoints. Proposed methodologies have either been based on damage modeling or on some kind of mathematical relationship. Artificial neural networks (ANN) were used as an alternative non-linear modeling technique due to their ability to learn by example. Previous research has shown that, if trained adequately, the ANN can be used to predict composites fatigue life for a given set of conditions usually sought by designers. Similar to other reported attempts, the authorsʹ previous work used the classical feedforward ANN for the fatigue life prediction of composites. Although the use of this network model gave results comparable to current fatigue life prediction methods, other types of ANN such as modular, self-organizing, radial basis, and principal component analysis networks are considered for improving the prediction accuracy. A comparison of such ANN structures in predicting the fatigue behavior of unidirectional glass fiber/epoxy composite laminae for various fiber orientation angles and stress ratios is investigated in this work.