A Data-Driven Method for Current Sensor Fault Diagnosis in Single-Phase PWM Rectifier

In practical drive systems, the occurrence of sensor faults may lead to control signal deviation and system performance degradation. Thus, this paper designs a data-based approach for sensor fault diagnosis strategy. This data-driven method’s principle is derived from the sensor signal prediction in single-phase PWM rectifier. The faults are detected and diagnosed by the normalized residual between predicted and actual signals. As the most important part, the signal predictor is designed by combining nonlinear autoregressive exogenous (NARX) learning model and a randomized learning technique, Extreme Learning Machine (ELM). To validate the effectiveness and accuracy, several hardware-in-loop tests are carried out to demonstrate the feasibility and reliability of the proposed data-driven method.