A Technique for Clustering Individual Defects from Images of Steel Strips with Periodical Defects

The steel strips produced in steel-making plants, are used as raw material in many other industries, so quality control is an essential aspect. One factor that indicates the quality of a steel strip is the number of defects, such as holes or scratches, on its surface. This paper describes a technique to detect an especially harmful type of surface defect called periodical defect. These defects are a periodic pattern on the surface of the strip. Using a backtracking algorithm all the individual defects contained in the strip are examined to determine which defects compose a set which constitutes a periodical defect. An implementation of this clustering technique was tested using a set of real steel strips, whose characteristics were previously stored in a database. A test environment to quantify the goodness of the results and to determine the best values to parameterize the clustering algorithm has also been developed.