A deep learning-based approach for fault diagnosis of current-carrying ring in catenary system

In the Industrial Internet of Things, the deep learning-based methods are used to help solve various problems. The current-carrying ring as one of important components on the catenary system which is always small in the catenary image has the potential risk to be a defect to impact the train operation. To improve the detection performance for the faulted current-carrying ring, a fault diagnosis method for the current-carrying ring based on an improved CenterNet model is proposed. Through analyzing of the characteristics of the catenary images and the detection network, the catenary image is preprocessed firstly by a simple enhancement method, which is proposed based on the Retinex theory for improving the quality of the image and suppressing noise in some degree. The embedded attention modules denoted as spatial weight block and channel weight block are adopted to enhance the local and global features, respectively. The shallow characteristics are fused into the deep semantic features with adaptive learning weights to make the features abundant. The weighted loss is presented to improve the performance of the detection for the faulted current-carrying ring. The experimental results show that the proposed method has improved fault diagnosis accuracy for the current-carrying rings which presents higher precision and recall values compared with the other detection networks in the experiments. It could provide useful assistance for improving efficiency and stability of the railway transportation.

Automated summary

This study proposes a fault diagnosis method for the current-carrying ring based on an improved CenterNet model in the Industrial Internet of Things. By using attention modules and weighted loss, the accuracy of fault diagnosis for the current-carrying rings is improved.