Physics-Informed Condition Monitoring of SiC Power Modules

Abstract

Silicon carbide (SiC) power modules are increasingly deployed in automotive traction inverters, where reliable condition monitoring is essential to prevent in-service failures and reduce maintenance costs. Despite extensive qualification procedures standardized under AQG 324, no consolidated approach exists for in-field health state estimation of SiC devices. Existing methods range from physics-of-failure lifetime models, which lack real-time applicability, to purely data-driven architectures that require large labeled datasets and offer limited generalization, to physics-informed machine learning frameworks that, while promising, remain computationally demanding for embedded deployment.

This work addresses condition monitoring of SiC MOSFET power modules assembled with sintered packaging technology, which prevents solder degradation and thereby produces aging behavior qualitatively distinct from previously studied devices. In solder-based modules, the forward voltage drop $V_{DS}$ follows smooth quasi-exponential trajectories driven by progressive solder delamination. With sintered packaging this mechanism is suppressed, and $V_{DS}$ instead exhibits multi-regime degradation profiles; modules are additionally subject to wirebond liftoff events that introduce abrupt, non-monotonic perturbations directly onto $V_{DS}$, posing new challenges for condition monitoring approaches.

To address these challenges, we propose a condition monitoring framework integrating three complementary elements. First, physics-informed feature engineering replaces raw sensor signals with physically grounded inputs, including cumulative damage indicators derived from junction temperature swing and mean junction temperature, combined with a Miner rule accumulator, which encode degradation history in a form directly interpretable by the model. Second, a monotonicity constraint enforced via gradient penalty regularization embeds the expected degradation direction as a physics-guided prior, improving generalization and cross-validation stability. Third, the model output is parameterized as a heavy-tailed distribution rather than a point estimate, providing calibrated uncertainty quantification inherently robust to the out-of-distribution variance introduced by liftoff events.

The framework is evaluated on an industrial power cycling dataset provided by Infineon Technologies, acquired under multiple operating conditions. Multiple neural network architectures of varying complexity and sequential context are compared within a rigorous cross-validation protocol. The full physics-informed feature set combined with gradient penalty regularization consistently outperforms purely data-driven baselines, reducing mean absolute error by approximately 70\% and maintaining stable performance across all cross-validation folds, including conditions where baseline models degrade significantly. These results confirm that integrating physical prior knowledge at the feature, constraint, and output distribution levels is both necessary and sufficient to handle the complexity of SiC power module degradation, while remaining lightweight enough for practical embedded deployment.