关键词: YOLOv5s, 电梯曳引钢丝绳, 外表面缺陷, 缺陷检测, 特征提取, 特征融合

Abstract:

To improve the detection efficiency and automation level of hidden defects on the outer surface of elevator traction steel wire ropes, and reduce the incidence of elevator accidents, an online detection model for defects on the outer surface of elevator traction steel wire ropes based on improved YOLOv5s is constructed. Firstly, the GhostConv module is introduced into the feature extraction layer C3 module to reduce computational complexity, and the Convolutional Block Attention Module (CBAM) is integrated to enhance the feature extraction capability of small-scale defects. A feature extraction module GC-C3 (GhostConv and CBAM-C3) that integrates GhostConv and CBAM is constructed; Secondly, in the feature fusion layer, Path Aggregation Network (PANet) and Bidirectional Feature Pyramid Network (BiFPN) are used to construct a multi-scale feature fusion network PBNet (PANet BiFPN), which combines multi-scale weight allocation strategy to improve the fusion effect of multi-scale defect feature information; Then, dynamically adjusting the quality weights of prediction boxes using Weighted Intersection over Union(WIoU) loss function, reducing the interference of low-quality samples on training. Finally, the model will be deployed to the developed detection system to perform online testing on the surface defects of the traction steel wire rope on the elevator car roof, verify the improvement effect of the model, and provide grading reminders for the defects. The results showed that the average detection accuracy of the improved model was 96.2%, with a detection speed of 192 f/s, which was 4.1% and 12.3% higher than the original model, respectively. The model volume was reduced by 38.9%. According to the online visualization experiment, under the actual operating environment of the traction steel wire rope (light illumination of 200~400 lx, speed of 1.5~2 m/s), the average accuracy of the system for 8 typical external surface defects is still stable at 94.6% or above, which meets the application requirements of online detection of hidden dangers of external surface defects of the traction steel wire rope in service and reduces elevator accidents caused by external surface defects of the traction steel wire rope.

Key words: YOLOv5s, elevator traction steel wire rope, outer surface defects, defect detection, feature extraction, feature fusion