Title: Fatigue life prediction for carbon fibre/epoxy laminate composites under spectrum loading using two different neural network architectures
Authors: Wael A. Altabey; Mohammad Noori
Addresses: International Institute for Urban Systems Engineering, Southeast University, Nanjing, 210096, China; Department of Mechanical Engineering, Faculty of Engineering, Alexandria University, Alexandria, 21544, Egypt ' Department of Mechanical Engineering, California Polytechnic State University, San Luis Obispo, CA 93405, USA
Abstract: The objective of this study is to predict the fatigue life of carbon fibre/epoxy composite laminate sheets involving 12 balanced woven bidirectional layers with the same orientation angle [0/90°]. The composite sheets considered are subjected to variable amplitude block loadings with different negative and positive stress ratios. This objective is accomplished by designing an efficient artificial neural network (ANN) architecture, with taking into account effect of the residual strength from spectrum loading. The number of cycles to failure (N) is related to the residual strength of the structure for constant amplitude loading. A simple first order model is postulated that determines the residual strength at any point during the fatigue life as a function of the static strength and stress ratio by applying the two-parameter Weibull probability density distribution. Two neural network structures, a feed-forward neural network (FFNN) and a radial basis neural network (RBNN), are applied, trained and tested to predict the fatigue life based on four groups of data considered. These data include the maximum stress (σmax) and the stress ratio (R), the Smith-Watson-Topper (SWT) parameter, fatigue strength ratio (Ψ), or failure criterion with the fibre orientation. On the other hand, the validity of the SWT, Ψ and selected suitable failure criterion for present study including material, loading and orientation were checked and modified before they were used for training the designed ANNs. The results show improvement when using one input data (SWT, or Ψ, or failure criterion) instead of two input data (σmax and R) and for the case of one input data the best prediction is observed for failure criterion condition, followed by Ψ and SWT respectively. Moreover, the RBNN demonstrates better results as compared with those obtained by the FFNN.
Keywords: artificial neural networks; ANNs; fatigue life; residual strength; composite materials; spectrum loading.
International Journal of Sustainable Materials and Structural Systems, 2017 Vol.3 No.1, pp.53 - 78
Available online: 28 May 2018 *Full-text access for editors Access for subscribers Purchase this article Comment on this article