MFLP-PINN: A physics-informed neural network for multiaxial fatigue life prediction

Abstract

In this study, a physics-informed neural network (MFLP-PINN), combining multiaxial fatigue critical plane model and the neural network, is proposed for life prediction. First, a multiaxial fatigue life prediction model based on the critical plane approach is proposed, which takes the equivalent strain amplitude on the critical plane as the main damage parameter, and considers the normal strain energy on the critical plane. Then, a total of four prediction models including the new critical plane model are integrated into the loss function of a neural network to build the MFLP-PINN. The accuracy of the proposed critical plane criterion and the MFLP-PINN are respectively verified using multiaxial fatigue test data of three materials. Finally, the results show that the prediction model integrated into the loss function has a significant impact on the neural network prediction. For a specific material, integrating a life prediction model with good prediction ability to this material as the loss function into a neural network model is helpful to improve prediction accuracy. Conversely, integrating a life prediction model with poor prediction ability to this material as the loss function into a neural network model will reduce the prediction accuracy.