基于机器学习辅助的IGBT模块键合界面疲劳损伤数值模拟方法研究

Bonded interface cracking is the primary failure mode of insulated gate bipolar transistor (IGBT) modules during prolonged operation. A numerical simulation method that effectively characterizes fatigue crack propagation at the bonded interface of IGBT modules under power cycling conditions was developed in this study. First, power cycling tests (PCTs) were performed for different numbers of cycles, followed by pull tests on the Al bond wires, obtaining load-displacement (F- ) curves for the bonded interface after different power cycles. Then, a finite element-fatigue cohesive zone model (FE-FCZM) was established, considering the cumulative effects of interface damage during the power cycling. To determine the fatigue cohesive zone model (FCZM) parameters, the influence of these parameters on the F- curves was investigated using sensitivity analysis, and the F- curves for different degrees of damage at the bonded interface were calculated after various numbers of power cycles. A machine learning (ML) model was subsequently established utilizing the simulated F- curve as input and the FCZM parameters as outputs, effectively identifying the FCZM parameters. Finally, the F- curves obtained from numerical simulations and experiments, as well as the degree of damage at the bonded interface, were compared to verify the accuracy of the ML method and the numerical simulation method for bonded interface fatigue damage proposed in this study. In addition, the proposed method was used to predict the lifetime of IGBT modules and compared with several commonly used lifetime models. The results show that the proposed method has the smallest error among all the models that were compared. This article provides a numerical method for describing the damage failure of the bonded interface in IGBT modules, which are significant for lifetime prediction and reliability assessment.