Surface Defect Recognition and Invalidation Judgment of Remanufactured Gears Based on Machine Vision

To improve the disassembly automation of remanufactured products and solve the blindness problem in the manual disassembly process, this paper proposes a method about remanufactured product defect gear invalidation judgement based on the improved YOLOv5 target detection algorithm combined with image processing technology. Aiming at the background interference problem in the recognition process, Squeeze-and-Excitation Networks (SENet) is added to the detection model of You Only Look Once version 5 (YOLOv5) to improve the target detection capability of the model. Then exporting the target detection frame, the redundant background of the target image is eliminated by mask processing, which concentrates subsequent identifications’ attention on gear targets. By means of using convex hull detection to judge gear broken teeth defects, and calculating the actual length of cracks by object image coordinate transformation, the method that combines the improved YOLOv5 algorithm model and gear defect recognition realizes gear invalidation judgement. This paper focuses on two parts: identifying gears and gear defects using the YOLOv5 and image processing algorithms. The experiment shows that this method has a good recognition effect on gear broken teeth, and the crack length recognition error does not exceed 1.38%. The whole research improves the rationality and accuracy of gears in the process of remanufacturing and disassembly and provides an effective way for deep disassembly of remanufactured products.