Machine learning-assisted modelling of stress concentration factor of unidirectional fiber composites for predicting their tensile strength

Abstract

Significant variations in the tensile strength of unidirectional (UD) fiber-reinforced composites are frequently observed due to randomness in the fiber arrays. Herein, we propose a novel method for predicting tensile strength capable of quantifying uncertainty based on a new recurrence relation for fiber fracture propagation and a determination algorithm for the fracture sequence for random fiber arrays (RFAs). We performed finite element simulations, calculating the stress concentration factor (SCF) for UD composites with various RFAs. Then, we trained an artificial neural network with the obtained SCF data and used it to predict the SCF for composites with an arbitrary RFA. The tensile strength of UD composites was predicted over a range of values, demonstrating that accuracy was superior to conventional prediction methods.