Automatic detection of surface defects on rolled steel using Computer Vision and Artificial Neural Networks

This work addresses the problem of automated visual inspection of surface defects on rolled steel, by using Computer Vision and Artificial Neural Networks. In recent years, the increasing throughput in the steel industry has become the visual inspection a critical production bottleneck. In this scenario, to assure a high rolled steel quality, novel sensor-based technologies have been developed. Unlike most common techniques, which are frequently based on manual estimations that lead to significant time and financial constraints, we present an automatic system based on (i) image analysis techniques, such as, Hough Transform to classify three defects with well-defined geometric shapes: welding, clamp and identification hole and (ii) two well-known feature extraction techniques: Principal Component Analysis and Self-Organizing Maps to classify three defects with complex shapes, specifically, oxidation, exfoliation and waveform defect. To demonstrate the effectiveness of our system, we tested it on challenging real-world datasets, acquired in a rolling mill of the steel company ArcelorMittal. The system was successfully validated, achieving overall accuracy of 87% and demonstrating its high potential to be applied in real scenarios.