Application of deep learning in workpiece defect detection

Abstract Due to problems in manufacturing process and forging technology, some of the workpieces production in the workshop have defects, which affects the ornamental value and utility of the workpieces. At present, the detection of workpiece defects is still mainly relying on manual detection, which consumes human resources and has a high detection error rate. To solve the above problems, deep learning is used to improve the detection of workpiece defects and reduce the cost of workpiece production. This article uses Faster R-CNN1 as the basic architecture and integrates Resnet502 as the backbone network. In addition, Deformable Convolutional Networks3 is introduced to extract defect features of different forms, and the multi-scale transformation problem is solved through Feature Pyramid Networks4. Comparing with YOLOv35, Faster R-CNN, Mask R-CNN6 and other models, the results show that the mAP of the object detection model based on Faster R-CNN is 89.21%, the detection effect is the best, and it can meet the actual production needs.