A Paris-law-informed neural fatigue cohesive model and its application to open-hole composite laminates

Abstract

Recent developments in machine learning technologies offer new insights for solving traditional engineering problems as well as a new data-driven framework for novel in-situ material characterization scheme. However, the need for large training samples severely limits the applicability of models of this nature. In this work, a novel training strategy utilizing Paris law is proposed for a neural fatigue cohesive (NFCOH) model. By adding regulation terms deduced from Paris law to the loss function of the neural network, the number of training samples required for NFCOH is significantly reduced. In addition, compensation methods for interfacial strength parameters are also proposed and investigated to further increase the flexibility of NFCOH. Finally, the NFCOH model is utilized to predict the fatigue life of complex open-holed composite laminates to demonstrate the effectiveness of the model where good correlations between the numerical predictions and experimental measurements are achieved.