Multiaxial fatigue life prediction using physics-informed neural networks with sensitive features

Abstract

Deep learning is a widely used tool for multiaxial fatigue life prediction. However, neural network still needs to solve various problems in the solution dominated by physical equations. This work proposes a physics-informed neural network framework incorporating sensitive features and life prediction models. The framework enhances the training process of the neural network under the constraint of life prediction model. Sensitive features are extracted to select input features with higher importance to the model output. The results show that the filtered sensitive features improve the predictive performance of the neural network. Introducing the Smith-Watson-Topper model as a physical loss degrades the predictive performance of the neural network. On the contrary, introducing Fatemi-Socie and Shang-Wang model as physical loss improves the predictive performance of neural network.