Flatness Defect Detection and Classification in Hot Rolled Steel Strips Using Convolutional Neural Networks

摘要

This paper addresses the improvement of flatness defect detection and classification in the steel industry. Localization and classification of the defects is respectively taken care of by a detector and a classifier. The pipeline can start with either CSV or image files coming straight from the plant sensors. To probe the performance of the system, it was used to detect and classify flatness defects in hot steel strips. A total of about 513 strips produced in a real steelworks were used for this purpose for a total of about 4806 defect images. A comparison between different traditional machine learning and deep learning models was carried out showing better performances with the latter approach.