Effective fault diagnosis in wind turbine operations is crucial for ensuring safety, reliability, and cost efficiency. This paper presents a machine learning-based approach for detecting multiple demagnetization faults including (unipolar, multi-magnets, adjacent, and uniform demagnetization) in Permanent Magnet Synchronous Generators (PMSG). The approach utilizes FEM 3D simulation models to analyze stator current and flux signals for fault detection. Discrete Wavelet Transform (DWT) is employed for feature extraction under both healthy and faulty conditions. We evaluate the performance of three primary machine learning classifiers—Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Ensemble methods—along with 17 sub-classifiers, for diagnosing faults using current and flux signals in PMSG. The results indicate that flux signals are more effective than stator current signals for detecting demagnetization faults. In the simulation, eight out of seventeen classifiers achieved 100% classification accuracy for all faults in PMSG using leakage flux signals, surpassing the accuracy obtained with current signals. However, Ensemble Bagged, Ensemble RUS, Coarse KNN, and Medium KNN classifiers demonstrated suboptimal performance.