Fault Diagnosis Methods Based on Fuzzy Logic and Artificial Neural Networks Approaches for Wind Turbine PWM Inverter

Abstract

ABSTRACT In this work, we investigate fault diagnosis methods of the PWM inverter in a wind energy conversion system (WECS) employing a doubly fed induction generator (DFIG), using a training dataset of 21 scenarios (1 healthy, 6 single‐switch, 12 double‐switch and 2 triple‐switch faults). The proposed approach employs two intelligent diagnostic techniques, involving fuzzy logic and artificial neural networks, to detect open‐circuit and short‐circuit faults in the inverter switches. The first approach is based on implementing fuzzy logic to compute the Average Absolute Value of Current (AAVC) values using stator current signals across 20 healthy and faulty simulations, thereby monitoring (detecting and locating) inverter faults in real time, with phase currents of the generator being reported. The second approach employs a neural network with a 3D feature extraction to operational data from these 20 training scenarios generated under nominal online steady‐state conditions, effectively identifying system anomalies. These data train the neural model to identify anomalies and faults within a system effectively. Simulation results demonstrate that the proposed neural network–based diagnosis method outperforms the fuzzy logic–based approach in terms of fault coverage, defined as the proportion of 20 simulated fault patterns, including single‐switch, double‐switch and triple‐switch faults, correctly detected.