Power electronic converters are essential in modern electrical and electronic applications, necessitating the study on power electronic reliability for dependable design and operation. In earlier research on power electronic reliability, the mission profiles have been mapped to thermal stresses and lifetimes, while its applications are generally focused on steady-state scenarios. It turns out to be time-consuming and inefficient to repeat the entire procedures for addressing the impact of stress variances or model uncertainties. In this context, this article defines the sensitivity for power electronic reliability and provides a sensitivity-based perspective on system optimization, where the concept is formulated through two steps, accounting for thermal stresses and lifetime, respectively. The applications of sensitivities for thermal stresses and lifetime are demonstrated, which serve as identifiers of critical factors and quantitative metrics of marginal performances. Upon this, possibilities are also emerging for enhancing the cost-effectiveness of reliability validations and encompassing model robustness into reliability analysis and prediction.
1 January 2026 in IEEE Transactions on Industrial Electronics
With the rapid shift toward the electrification of society, the reliability and lifespan prediction of power electronic components have become increasingly crucial. Thermal simulations serve as a fundamental tool to analyze the thermal behavior of these systems; however, real-time applications require a significant reduction in computational complexity. This study presents an improved Proper Orthogonal Decomposition method using Galerkin projection, offering enhanced numerical robustness and adaptability to arbitrary, time-dependent boundary conditions. The method is validated on an integrated metal substrate circuit board, demonstrating high accuracy with temperature prediction errors below 1 K during both steady-state and transient operations when compared to FE solutions and a maximum error of 2.1 K when compared to steady-state measurements. This work, which is publicly available on GitHub, contributes both academically and technically toward efficient, reliable thermal management in power electronics.