Current sensor fault diagnosis and fault‐tolerant control for single‐phase PWM rectifier based on a hybrid model‐based and data‐driven method

Abstract

In this study, a hybrid model-based and data-driven method is proposed for the current sensor fault diagnosis used in single-phase pulse width modulation (PWM) rectifier. According to the principle of model-based methods, the proposed diagnostic method is based on signal prediction and residual generation. Differently, instead of a mathematical model, the signal prediction model is developed based on a data-driven method. Non-linear autoregressive exogenous learning model, randomised learning technique, and extreme learning machine are utilised to generate the data-driven prediction model. Once the fault is detected, fault-tolerant control is activated by substituting the predicted signal for the information of faulty sensors. The offline test shows that the proposed method is able to predict the sensor signal accurately with the root mean square error of 4.276 × 10−5. In addition, hardware-in-the-loop tests are conducted to verify the feasibility and reliability of the proposed method in the real-time application.