Design of Commutator Cylindrical Foreign Object Detection System Based on Improved YOLOv5

In the automatic production line of a motor, after the commutator has been turned and other processes, there may be copper chips and other foreign objects falling into the insulation slot to affect the performance of the motor. This paper adopts machine vision technology to automatically detect foreign objects in the commutator insulation slot and designs a commutator column surface foreign object detection system based on the improved YOLOv5. The composite illumination design with multiple light source combinations is used on the illumination system to collect pictures with obvious features in the insulation slot. The detection algorithm uses a PP-LCNet lightweight network to replace the backbone structure in the YOLOv5 network, introduces the SE module at the end of the network, and replaces the Relu activation function with the H-Swish activation function to speed up the detection speed. The experimental results show that the improved YOLOv5 algorithm reduces 54 misjudgments of qualified products per day compared with the traditional image recognition algorithm, and the correct recognition rate reaches 97.54%; the single inspection time is reduced by 66.29% compared with the YOLOv5 algorithm, which meets the accuracy and speed requirements of realtime inspection in automated production.