Fault identification for permanent magnet synchronous generators of wind turbines by wavelet signal processing with machine learning classification algorithms

The increasing demand for wind energy has led to a rise in the application of Permanent Magnet Synchronous Generators (PMSG) which are used in wind turbines. Stator winding faults in this type of generator can deteriorate performance and lead to severe faults if not detected early. This study’s data set consists of the electrical signal having a percentage of faults generated for two types of coils turn to turn and turn to ground. Signal processing techniques namely MODWPT and MODWT are used to diagnose and denoise the signal consisting of the percentage of faults in PMSG. The base wavelet for both signal processing techniques has been identified using the criterion known as MESE. Statistical features were extracted from the coefficients to generate feature vectors, of which the feature vector generated using MODWPT was ranked using ReliefF. Finally, three machine learning models were trained, tested, and validated using these feature vectors to identify the percentage of faults in PMSG.