Development of High-speed Surface Crack Detection Technique for Metal Panels via light control

Metal panels have been used extensively in automobile and home appliance industries owing to its excellent strength-to-weight ratio. Because the surface crack of the press panel affects the general strength, crack detection in the production line is vital to quality control. In an automated production line characterized by high speed within 3s, applying the conventional contact-type non-destructive inspection technologies is difficult; hence, diagnosis is performed via visual inspection by experts in most production lines after the completion of a process. The efficiency and accuracy of these diagnoses are affected by external factors such as human skills and concentration. To mitigate this issue, an automated and accurate vision-based surface defect detection technique is proposed herein. In this technique, products are passed in front of a blackout fabric to control the lighting for image acquisition. Subsequently, a clear edge line is extracted via weighted object variance (WOV) Otsu's method and percolation-based shape recognition. Next, an initial detection is performed to identify viable specific edges in low-resolution images via curvature evaluation and edge detection. Finally, for a more accurate detection, these edges are analyzed individually at high resolution, and the crack positions are identified. The efficacy of this detection system is verified in a laboratory-scale production line.