A data-assisted physics-informed neural network for predicting fatigue life of electronic components under complex shock loads

Abstract

Reusable spacecraft electronic components experience multiple, complex shock damage during their operational life, which is a primary contributor to mission failure. This study proposes a data-assisted physics-informed neural network (DA-PINN) model to assess fatigue damage in electronic components under complex shock loads. Unlike traditional PINN that solves partial differential equations, DA-PINN combines experimental with physics equations to enhance prediction accuracy. An autoregressive (AR) model was used to improve the shock fatigue life model, which was then integrated as a physical constraint into the loss function of DA-PINN. Subsequently, using ball grid array (BGA) solder joints as the research subject, complex shock fatigue experiments were conducted to train and validate the DA-PINN model. The results demonstrate the outstanding performance of the DA-PINN model, with all predicted shock fatigue life values falling within a scatter band of 1.5 times, surpassing the traditional shock fatigue life model and artificial neural networks. Notably, the physics-informed constraints embedded in DA-PINN enable it to maintain strong prediction accuracy and stability even when trained on small datasets. The proposed model can provide a reference for predicting the shock fatigue life of electronic components in reusable spacecraft.