A 2D+3D Vision-based Surface Defect Detection Method for Mandrel

Described here is a 2D+3D-based automatic detection system for mandrel surface defect. The system is functioning by means of using four sets of line structured light measurement devices to fulfill jointly calibration, establishing the relationship between the image pixel coordinate system and the world coordinate system, getting all the point clouds concatenated and fused to reconstruct the cross-sectional contour of the mandrel, and finally, having the mandrel real-time photographed and reconstructed to obtain the overall surface contour thereof. And the system is capable of detecting defects as needing depth information such as longitudinal scratch, blocky flaw, and surface peeling etc via setting such threshold values as defect depth and area. It is verified via relevant experiment that the system as developed features a detection speed of 2 m/s, a minimum detection accuracy of 0.3 mm, a defect missing rate ≤ 0.5%, and a false alarm rate ≤ 5%, having met the requirements for surface defect detection of the mandrel.