Demagnetization Fault Diagnosis Based on Feature Extraction and Stacking Ensemble Learning for Permanent Magnet Synchronous Generator

During the operation of a permanent magnet wind turbine, magnet demagnetization failure may occur, which directly affects the regular operation of the wind turbine and adversely affects wind power generation. This paper proposes a demagnetization fault diagnosis method for permanent magnet generators based on feature extraction and stacking integrated learning. A permanent magnet generator with a power of 25 kW was used to conduct a demagnetization fault simulation experiment. Collect the current signal of the generator and extract features such as time domain, frequency domain, entropy, and singular value. The different extracted features are trained through the Stacking integrated learning framework to realize pattern recognition of demagnetization faults and determine the operating status of the generator, thereby realizing demagnetization fault diagnosis of permanent magnet generators.

I. Introduction

The permanent magnet wind turbine is located at a high altitude and has a harsh environment. It is susceptible to various faults due to long-term vibration, noise, and electromagnetic interference. Demagnetization fault is one of the most common types of faults in permanent magnet wind turbines. Demagnetization faults are mainly caused by electrical stress (armature reaction) [1]. Demagnetization can act on all magnetic poles or some magnetic poles. The former is called uniform demagnetization, while the latter is called non-uniform demagnetization or local demagnetization. When a permanent magnet demagnetization fault occurs in a permanent magnet wind turbine, it can decrease the performance features of the permanent magnet motor, increase temperature, increase vibration, and increase noise [2]. Suppose the fault cannot be detected and eliminated promptly. In that case, it can lead to damage and scrapping of the generator, seriously affecting the operation of the wind farm.