Health State Estimation of Power MOSFETs With Multistate Probability Prediction-Based Particle Filter

Power metal–oxide–semiconductor field-effect transistors (MOSFETs) are critical components in power electronics, drivers, and industrial automation fields. However, their health state can easily degrade due to the high voltage, current, frequency, and complex operating conditions, resulting in various faults and damages. Accurately estimating the health state of power MOSFETs is therefore crucial to improve the reliability and safety of circuits involving these components. Particle-filter-based methods can provide health state estimation distributions and reduce the influence of fluctuations. However, due to the complex degradation mechanism of power MOSFETs, determining the state transition equation for particle-filter-based methods is challenging. To address this issue, this article proposes a health state estimation method using multistate probability-prediction-based particle filter. The proposed method constructs multistate particles based on the statistical characteristics of the health state and maps these particles to the current health state using the probability prediction method. Finally, by combining the basic theory of particle filter with health state measurement, the proposed method achieves effective health state estimation. The effectiveness of the proposed method is validated using the public dataset of National Aeronautics and Space Administration (NASA), showing accurate health state estimation of power MOSFETs in complex environments. The experimental results show that for the health state estimation of device 36, the proposed method has the lowest root mean square error (RMSE)/mean compared with long short-term memory based particle filter (LSTM-PF), support vector machines (SVM), linear regression (LR), self-attention with LSTM (LSTM-SA), and Gaussian process regression (GPR), with a maximum accuracy improvement of 45.73%. For the health state estimation of device 38, the proposed method is still the most accurate method among the comparison methods, with a maximum accuracy improvement of 31.51%.