Defect detection and classification using machine learning classifier

Abstract

In steel industry, hot rolling is the first and major step in flat strip production. In addition to standard measurement, precise information concerning the complete hot strip surface quality has become increasingly important for several reasons. Visual inspection of the hot strip by an inspector is, in most cases not possible because of the high speeds and high temperature involved. In recent times, only video monitors and video recorders have been used where inspectors check on-line or taped video sequences for defects. Developing automatic detection and classification of surface defects of hot strips has been really a challenging problem. Real time image processing typically involves the application of high-speed camera, which may give the defect images in real time. Human experts evaluate this information and give the level of defects, types of defects of the hot strips and will suggest remedial measure to over come those defects. This paper highlights the initial steps in developing a Decision Support System for hot strip evaluation in the manufacturing plant based on the process carried out in the practice. As a first step, a methodology is developed for defect image classifier using machine-learning model - Artificial Neural Networks (ANN). ANN has an excellent generalization capability to learn from the set of data obtained during real-time images. However, while developing ANN model for industrial application many unresolved issues come into pictures. In this study our objectives are issues related to data collection, data modeling, neural networks modeling, reliability of neural models, etc. for automatic detection and classification of defects of hot strips. The proposed study intends to develop general guidelines for developing ANN model for automatic surface inspection for hot strip mills.