Autoencoder Based Method for Detection of Steel Wire Defects

In the steel industry, quality control of the final product is usually done by manual visual inspection. Due to the disadvantages of manual visual inspection, research and development of new automatic inspection techniques for the detection of steel surface defects has gained more importance in recent years. The development of industrial processes and the need for new systems have revealed new defect detection methods based on computer vision. The importance of automatic inspection based on computer vision is based on its adoption as a quality control tool, which can inspect without damaging the material to be examined and without affecting the production speed. In this context, it is the main objective of the manufacturers that the inspection to be applied during steel production solves the tasks of the people currently carrying out the inspection work with an automatic inspection system. In this study, an autoencoder based defect detection method is proposed for the detection of steel wire defects. In the study, the defects on the steel were detected by using a deep automatic encoder. In autoencoders, abnormal data produces higher rendering error than normal data. This situation is used to detect abnormal situations in images. Using this method, an accuracy rate of 94.44% was obtained in classifying the anomalies on the steel surface.