Arc Fault Diagnosis Method for Brush Slip Ring System of Doubly-Fed Induction Generator Based on Image Recognition

Abstract

This article proposes a method for diagnosing arc faults in the brush slip ring system of a Doubly-Fed Induction Generator (DFIG) using image recognition technology. The aim is to achieve automatic detection and diagnosis of arc faults. The experimental process of this diagnostic method includes collecting a large amount of arc faults image sample data through an industrial camera, preprocessing the collected data, completing data annotation, and creating a dataset. Build the YOLOv5 model using the PyTorch deep learning framework, train it using the arc faults image dataset, observe the convergence of the loss function during the model training process, and analyze the model testing results, taking into account evaluation indicators such as Precision, Recall, mean Average Precision (mAP), and Frames Per Second (FPS). This article proposes the K-YOLO++ model for diagnosing arc faults. The K-YOLO++ model is an improved model obtained by adding a small object detection layer on the basis of the YOLOv5 model, improving the multi-scale detection mechanism, and executing the K-means++ clustering algorithm to improve the anchors. The test results on the experimental dataset of arc faults in the brush slip ring system of Doubly-Fed Induction Generator indicate that the K-YOLO++ network model has improved Precision, Recall, and mAP. The mAP of the improved model is 88.36%, with a FPS of 50.62. The recognition effect of small targets is superior, effectively reducing missed and false detections, enabling efficient and accurate arc faults diagnosis.

Keywords

• Arc fault
• deep learning
• image recognition
• K-YOLO++
• small object detection