Pattern recognition and quality control: an intelligent approach to identify defects

This paper describes a knowledge-based system utilized to identify imperfections or defects in industrial products. The kind of defects we study appear on the piece external area (like spots, fractures, scratches, dark or white lines). The application of the system has been developed in wall or floor tile factories and has proved to be adequate. The system works with codified information from the wall or floor tile faces. This information is accessed by special devices which pick up the image and transform it in an array of numbers and codes. Therefore, the system behavior can be defined by this information. Initially the system detects the existence of imperfections using a first group of computational programs; after that, a second group of programs defines the gravity level of each detected defect (for instance, if it implies to reject the piece). Finally, a third group of programs (the identification system) informs its users of the type of imperfection detected, i.e. promotes its identification. We show the general idea of the identification system and some results. This can be seen as a useful and interesting application of knowledge-based systems in the quality control area.