A machine learning-based prediction of biaxial failure envelope of a short fiber-reinforced polymer composite

Short-fiber reinforced polymer composites (SFRPs) consist of short and discontinuous fibers dispersed in polymer matrices. This study presents the determination of biaxial failure envelope of an SFRP specimen using an artificial neural network (ANN). The failure envelope defines the decision boundary under biaxial loading stress, distinguishing stress states inside as survival and outside as failure. The complex modeling and the high cost associated with the FE-analysis of SFRPs make the determination of the failure envelope computationally expensive. This study uses an ANN as a surrogate model to predict the biaxial failure envelopes of an SFRP specimen. The failure envelopes used for training and testing the ANN model are extracted for the SFRP specimen using a two-step homogenization, employing the first pseudo-grain failure model. The database is supplemented with experimental data from biaxial tests and FE analysis results available in the literature. An elastoplastic polymer matrix dispersed with short elastic fibers is taken for analysis. The strength parameters of the fiber and matrix and the geometrical parameters of the microstructure are varied over a range to develop a dataset for ANN training. The failure envelopes are predicted for two different unseen SFRPs using the ANN model. The ANN predictions are compared with the simulation and experimental results reported in the literature. Additionally, a parametric study is performed to investigate the effect of the key parameters of the SFRP, such as the volume fraction, aspect ratio, and orientation of the fiber.