Automatic Thermal Modeling

Dataset for reproducing the code of the work: Estimation of Semiconductor Power Losses Through Automatic Thermal Modeling.

Abstract:

Achieving the optimal design of power converters hinges on a deep understanding of the system's dissipation elements to meet desired performance and safety standards. Calorimetric techniques have outperformed classical electrical methods in estimating semiconductor power losses. However, they come with mechanical limitations and depend on analytical electrothermal equivalent circuits. These models are highly topology and technology-dependent, often resulting in either overly simplistic representations that underestimate thermal effects or complex sets of differential equations. To overcome these challenges, we present an innovative data-driven method for characterizing power converter thermal dynamics. This method empowers designers to calculate semiconductor power losses solely based on temperature measurements, which can eliminate the need or be combined with calorimeters. By analyzing sets of power vs. temperature profiles, our approach identifies the most appropriate linear model. This method is rooted in an optimization process that ensures not only precise identification but also the integration of desired modeling requirements, such as dynamics' invertibility for power loss estimation from temperature profiles. This versatile methodology is applicable to any power converter topology, and the derived linear model allows the use of standard control theory techniques for analyzing and controlling thermal dynamics. Real-world experiments validate the proposal's universality and accuracy.