Machine Learning and EMI For MOSFET Aging Diagnosis

Abstract

MOSFETs play a key role in static power converters, where power MOSFETs operate at high switching frequencies and are susceptible to Electromagnetic Interference (EMI) due to parasitic inductive and capacitive elements inherent in the electric circuit. Despite being unwanted, EMI can offer valuable insight into the condition of the emitting agent, as their amplitude and frequency are heavily dependent on the agent’s intrinsic characteristics. As MOSFETs age, their intrinsic characteristics change, leading to corresponding changes in the emitted EMI. Previous studies have investigated the evolution of EMI in a DC-DC converter but did not quantitatively assess the extent of degradation. In this study, we propose a machine learning-based approach for predicting failures based on EMI. We demonstrate that various regression algorithms can accurately predict MOSFET failure based on EMI, including the degree of degradation, enabling the estimation of remaining useful life (RUL). We validate the effectiveness of our approach through various simulation results.