Discussion on classification methods for lifetime evaluation of a lab-scale SiC MOSFET power module
Résumé
This paper focuses on classification methods for evaluating the lifetimeconsumption (LC) of power electronics modules. The generalization of powerelectronics devices introduces new issues concerning the reliability of equipment,especially in the transportation field. To meet these expectations, this paper discussesan approach to evaluate the percentage of lifetime of a lab-scale SiC MOSFET powermodule, designed for an aircraft application. This module is based on a planartechnology, and presents typical failure modes concerning the SiC MOSFET chip itselfand its environment. The modules have been aged on a specific instrumented testbench to trigger the expected failure modes. Thanks to it, a large database ofparameters have been obtained in order to find a relevant failure signature. Once thesignature obtained, a comprehensive solution is required to classify the signaturesinto relevant classes related to the module LC. To meet the issue, three types ofclassification have been tested with learning data set: Support Vector Machine, kNearest Neighbors and neural network. The last contribution of this paper is adiscussion on the evaluation of the percentage of lifetime consumption of a new testmodule thanks to the most promising models obtained from the learning data set.