Hybrid system application to defect classification in cast aluminum

Resumen

This paper desecribes some A. I.-based techniques applied to the interpretation of images from aluminum surface. The whole process includes defect detection and defect classification into five to eight different types. The whole process is on-line performed. The developed Visual Inspection System includes: a defect detection module, a feature extraction module and a classification module. Images coming from the aluminum surface are preprocessed by means of local analysis to seek for defects. A local analysis is necessary due to the presence of texture and to the appearance of defects where clear points appear mixed up with darker ones. The image acquisition is performed by a set of CCD cameras and these are supported by a specially developed lighting system. Image processing for defect detection consists on getting simple statistical parameters, averages in the neighbourhood and local comparison with correct patterns. Also, we discuss about feature extraction and the use of direct and indirect methods for syntactic analysis and extraction of the feature vector. At this point, we put emphasis on transforming the set of defect primitives into a feature vector to reduce the spatial dimension of the input to the classifiers. Several classifiers are used together to improve the performance of the classification module. Online classification is achieved. A hybrid system has been developed for the structure recognition of defects in cast aluminum.