A Data-Driven Remaining Useful Life Prediction Method for Power MOSFETs Considering Nonlinear Dynamical Behaviors

Abstract

Prognostic and health management (PHM) techniques for power MOSFETs are getting increasing attention recently. A variety of methods have been developed and implemented to conduct lifetime predictions for power MOSFETs. Nevertheless, most of the current studies seem to have limitations in a comprehensive understanding of the nonlinear dynamical degradation process. Single parameter-oriented prediction methods may ignore deeper dynamical behaviors during the degradation. Besides, the methods are incapable of tackling abnormal degradation paths such as a sudden rise. In view of the limitations, a data-driven prediction method taking into consideration the nonlinear dynamical behaviors is developed. To analyze nonlinear and chaotic properties, phase space reconstruction (PSR) is conducted on the time series degradation data. Then, the largest Lyapunov exponent and power spectrum are calculated against aging time. The evolution of nonlinear and chaotic behaviors during the degradation is investigated. Thereby, a novel health indicator (HI) taking into account nonlinear indices is constructed. Subsequently, a prediction method based on a long short-term memory (LSTM) network is proposed. The developed method is validated by an actual degradation dataset. The results show that the developed method is capable of addressing the limitations with desirable accuracies.