Neural Network-based Cold-rolled Strip Surface Defect Detection And Classification

Quality control has gained high significance in steel production. Defects on thesurface of cold steel strips are main factors to evaluate the quality of cold strips, sosurface defects inspection is of great importance to improve quality of cold steelstrips. Traditional surface defects inspection done by human inspectors is far fromsatisfactory. Developing automatic surface defects inspection systems is a commondemand of steel corporations.Surface defects inspection techniques based on neural network are investigated.The general design scheme of surface defects inspection system for cold rolled stripsis given, hardware and software of the system is described, and algorithms for detectdetection and defect classification are discussed. The details are as follows:1. Based on the technology requirement of surface defects inspection system forcold rolled strips, the article gives the general design scheme of the system.2. Algorithms for detect detection are developed. Experiments with samples fromproduction lines show that six types of surface defects, which are investigated, can bedetected, and the detection rate is higher than 90%.3. Features extracted by the system are discussed. Experiments show that thefeatures are effective for classification of detects.4. Algorithms for defect classification are developed. Classifiers are constructedbased on back-propagation network. Experiments with samples of surface detectsshow that classification rate is up to 90%.It's a new method of combine the digital image processing and network in thedefect inspection of the cold strips steel, it has the advantages that other methods havenot, and have high precision and high efficiency. And this method can be expandedaccording to requirement in the production practice.