Machine vision-based segmentation and classification method for intelligent roller surface monitoring

Abstract

The surface quality of steel rollers is a key factor determining the quality of final products such as metal sheets and foils in the rolling industry. It is important to examine the surface quality of rollers since rollers with optical defects will always duplicate the defects onto the metal sheet or foil during rolling. The typical optical defects of rollers after finish grinding include speckles, chatter marks, swirl marks and combination of all of the above. They can hardly be modeled or shaped by the approach of micro topography or SEM (scanning electrical microscope). In this paper, an on-site machine vision system is firstly applied for stable inspection for the optical defects on roller surfaces. Then, an improved optical defect segmentation algorithm is developed based on the active contour model and the images including chatter marks and swirl marks. The normal surface state is classified by the combination of methods of Gabor filters, KPCA method and ELM neural networks. Finally, experiment are carried out to verify the efficiency of the improved segmentation method and the recognition rate of the combined classification algorithm.