A Novel Dynamic Optical Imaging and Defect Detection Method for the Surfaces of Metal Rotary Parts

Abstract This paper presents a dynamic imaging and detection (DID) system to address the issue of surface defect detection in metal rotary parts (MRP) in industrial settings. First, a novel optical detection system is designed, which employs a multi-view imaging mechanism equipped with six cameras and incorporates a multi-lighting scheme (including semi-tunnel light and ring/coaxial light sources) to capture high-quality images of the MRP surfaces and end faces. To overcome the limitations of traditional static imaging, a high-dynamic multi-view computational imaging method is developed. This method utilizes adaptive video stream processing, dynamic texture modeling, and texture-driven enhancement techniques to generate panoramic fused images with improved defect visibility. After obtaining high-quality images, an improved YOLOv10 model was proposed, which integrates the designed C2f_MCP and Dynamic Upsampling (D_Upsample) modules to achieve high-performance defect detection. Experimental results demonstrate that this complete closed-loop framework of "Acquisition → Imaging → Detection" outperforms traditional imaging techniques in terms of imaging contrast and texture clarity. Additionally, the YOLOv10-based detection method exhibits robust performance in detecting multi-scale defects under complex conditions, providing reliable assurance for quality control in smart manufacturing.