Data-Driven Approach for Fault Prognosis of SiC MOSFETs

Abstract

This article proposes an unsupervised learning approach for fault prognosis of silicon carbide (SiC) mosfets. The proposed approach utilizes the changing trend of a device's voltage, current, temperature, and other device characteristics with its degradation. The failure modes of semiconductors are reviewed along with existing methods for fault prognosis. The proposed approach is the first to address prognostics of SiC devices, and it can avoid the effects from system noise and data errors. It is not limited to offline analysis and is targeted at online implementation. It is easy to implement on standard digital platforms, and has fast computational speed. Offline data analysis is performed to verify the effectiveness of the proposed method, and a processor-in-the-loop system is used to verify its ability to perform online fault prognosis.