Physics-informed Data-driven Methods for Reliability Test and Analysis of Semiconductor Switches in Power Converters

Abstract:
Reliability is an important task in power electronic converters in particular of power semiconductor switches, which deeply rely on testing. However, most of the existing researchers in this area are independent of each other. For example, extensive testing efforts have been carried out for some specific devices from different sources but a comprehensive analysis among these testing results is missing today. Although extensive testing results have been accumulated in this domain, the knowledge extracted from the historical data is limited. As result, similar testings have been repeated for different devices for the long term. This project will apply physics-informed data-driven methods for this domain, which has three aspects, (1) Early degradation study: digging into the physical failure mechanisms of power devices to achieve early lifetime prediction, which serves to reduce the testing time of power switches; (2) Remaining useful lifetime (RUL) prediction: extend the research outcomes of aspect (1) to power converter applications, which serves to predict the RUL in condition monitoring; (3) Data-driven lifetime model generation: integrate both physical failure mechanisms of (1) and collection of historical testing data to establish a data-driven lifetime model generation tool, which serves to comprehensively analyze and utilize the extensive testing results that existed in this domain.

Funding:
CSC Scholarship