A Real-Time Bent Cable Detection Method for Fatigue Testing in Fast Drag Chain Machines

Abstract

Bending cables can cause irreversible damage to the tracks and rails of fast drag chain machines. To swiftly and precisely identify bent cables within these machines, an intelligent detection method based on improved YOLOv8n for bent cables is proposed. This method can simultaneously achieve clear detection and bend detection of cables. The YOLOv8n backbone network is augmented with a global attention mechanism (GAM) to adjust the importance weights of each channel, enabling more effective capture of key features and enhancing the feature maps’ expressive capacity. A P2 small-object detection layer is incorporated in the detection head to improve the model’s capability to detect minute curved areas. Moreover, the Wise_IoU (W_IoU) loss function is adopted in place of the traditional C_IoU loss function to minimize the impact of low-quality samples on model performance during training, thereby optimizing the training process and enhancing model accuracy. The refined YOLOv8n model demonstrated a mean average precision (mAP) of 92.1% in detecting bent cables, with a detection time of 2.1 ms, leading to a 0.8-ms reduction in detection time compared to the original YOLOv8n model. These improvements make the model particularly well-suited for rapid detection in fast drag chain machines. The detection method has already been applied in practice and helps avoid over 3 track damages within a quarter.