A Data-Driven Diagnosis Method of Open-Circuit Switch Faults for PMSG-Based Wind Generation System

Abstract

Wind energy conversion system technology has attracted worldwide attention, and the condition monitoring and fault diagnosis for the system become significant issues. A data-driven fault diagnosis method is presented to detect and locate open-circuit switch faults of the back-to-back converter in permanent magnet synchronous generator (PMSG)-based wind generation system. Convolutional neural network (CNN)-based neural network is applied as a fault diagnosis method, and the dropout process is employed to deal with the over-fitting problem. Twelve sensor signals of current and voltage in the back-to-back converter in various conditions are measured. A grid-connected PMSG-based wind generation model has been built in MATLAB/Simulink to estimate the proposed algorithm. Least squares support vector machine (LSSVM) and back-propagation artificial neural network (BPANN) are applied as comparison methods. Simulation results reveals that the proposed theory has a decent performance regarding the detection and location of different faulty switches in an assembly of various operating conditions.