An Advanced Mission Profile-based Lifetime Assessment in Power Electronics Assembly Using Physics-informed Surrogate Modeling

Power modules under mission-profile operation develop thermo-mechanical fatigue in bond wires and solder joints due to temperature cycling. Empirical lifetime rules miss transient and sequential effects, while finite-element (FE) analysis, though physical, is computationally too expensive for application-close mission profile or system-level studies. We propose a physics-informed surrogate of an FE-modeled SiC power module that calculates in multi-physics domain to obtain damage indicators for bond-wire foot and solder interconnections, like plastic strain. Trained on FE simulations from ARTEMIS-derived profiles and evaluated on independent frames, Transformer and Temporal Convolution Network (TCN) were determined as the best deep models for bond wires and solder interconnection respectively. Moreover, random forest + AdaBoost (RF+ADA) was the most stable shallow alternative. All surrogate models can reach R2> 0.995 against FE results. Estimation of lifetime based on surrogate results stayed within a deviation of 10% in comparison to the strain calculated using FE model. It shows that the surrogate approaches may enable a mission profile-based reliability assessment for power electronics assembly.