Demagnetization Fault Diagnosis of Permanent Magnet Synchronous Generator for Wind Turbine Based on ResNet-50 Transfer Learning Model

This paper utilizes Convolutional Neural Networks (CNN) with Residual Networks (ResNet) to identify and classify demagnetization faults in images. Firstly, we collected current feature datasets corresponding to two types of permanent magnet wind turbines, namely 25 kW and 2 MW, representing the source and target domains. Using the source domain dataset, we trained a ResNet50 model specifically for demagnetization fault diagnosis of the 25 kW wind turbine. Subsequently, we applied a network-based transfer learning approach to adapt this model to the demagnetization fault diagnosis for the 2 MW wind turbine, completing the fine-tuning training required for transfer learning. Compared to models without transfer learning, this method significantly improves the accuracy and efficiency of demagnetization fault diagnosis.

I. INTRODUCTION

With the development of wind power generation technology, the proportion of wind power generation in the new energy field continues to increase. According to a research report released by the Global Wind Energy Council, by 2023, the global wind power installed capacity have reached 1000 GW and will grow well. At present, wind turbines are mainly divided into two types: doubly fed-and direct-driven. Since direct-driven wind turbines are based on permanent magnet synchronous motors, they are significantly better than doubly fed ones in terms of efficiency and reliability and, therefore, have attracted much attention from the wind power industry.