A physics-guided modelling method of artificial neural network for multiaxial fatigue life prediction under irregular loading

Abstract

To improve the generalization of the artificial neural network (ANN) model on the prediction of multiaxial irregular cases, a physics-guided modelling method is proposed with inspiration from the Basquin–Coffin–Manson equation. The method suggested using two neurons in the last hidden layer of the ANN model and constraining the sign of weight and bias value. In this way, the prior physical knowledge of fatigue life distribution is introduced into the ANN model, which resulted in a satisfactory performance on the life prediction of multiaxial loading cases and better extrapolation ability. Furthermore, the physics-guided ANN model can also provide satisfactory prediction on irregular cases with the training of only regular cases. Compared with the conventional model, the average relative error and root mean squared error (RMSE) of prediction decreased by 33.29% and 44.29%, respectively. It greatly broadens the application scenarios of neural networks on multiaxial fatigue life prediction. This article is part of the theme issue 'Physics-informed machine learning and its structural integrity applications (Part 1)'.