132 / 2023-09-20 20:47:09

Classification of Demagnetization Faults based on Performance Evaluation of Machine Learning Algorithms via Flux and Vibration Signals in Permanent Magnet Wind Turbine Generator

K-nearest neighbors, Machine learning, Matching pursuit, Mo-tor current signature analysis, and Permanent magnet generator

Special Sessions > SS18 - Advanced Sensing, Monitoring and Diagnosis in Smart Grid

Fault diagnosis and condition monitoring play a significant role in wind turbines as they guarantee safety and reliability and avoid perilous conditions; therefore, fault diagnosis prior to its existence saves both time and costs. This paper proposes a machine learning-based fault diagnosis technique using vibration and leakage flux for multiple demagnetization faults, including "healthy, 30% unipolar demagnetization, 50% multi-magnets demagnetization, 100% adjacent poles demagnetization, and 40% uniform demagnetization" in a 25kW Permanent Magnet Wind Turbine Generator. The feature extraction is achieved from the Matching Pursuit signal processing technique for healthy and faulty operations. Then, eight fundamental classifications comprised of 31 sub-classifiers are trained using MATLAB. Furthermore, the KNN technique is demonstrated in Python for classifying the abovementioned faults and comparing its accuracy per K values. Results illustrate that leakage flux performs better than vibration signal for demagnetization fault diagnosis in wind turbine generators. Of the 31 algorithms tested, it was discovered that 09 had complete accuracy (100%), and seventeen had an accuracy of at least 90%. Moreover, for classifying multiple demagnetization faults utilizing a leakage flux signal, the KNN method has a 100% accuracy rate when K=3 and almost 90% accuracy rate for K=1 to 16.