Development of Deep Learning-based for Hot-rolled Seamless Steel Pipes' Outer Surface Defect Recognition System

Detecting surface defects is crucial during the production process of hot-rolled seamless steel pipes. In order to make up for the shortcomings of traditional inspection methods, this study developed a set of surface defect detection systems of hot-rolled seamless steel pipe by 3D point cloud and deep learning. A 3D point cloud-based defect detection equipment for seamless steel pipe outer surface was developed to achieve full coverage of seamless steel pipe outer surface detection. The defect detection model of seamless steel pipe surface was designed based on the YOLOF algorithm of correcting attention mechanism, and the corresponding software system was developed for online identification. To confirm the efficacy of the proposed approach, the defect pictures collected by a steel pipe factory were used as an implementation case. The experimental results confirm that all types of defects are effectively covered with a detection rate exceeding 99%, while maintaining a false detection rate of under 0.08%. The industrial application shows that compared with the current advanced detection method, the false positive rate of the outer surface is reduced by 9.5% on average. The proposed method has low installation and maintenance costs, which can lay a foundation for further application of seamless steel pipe quality determination, analysis, and control.