Machine Parts Visual Inspection System using Fuzzy-Neural Network

This paper treats the machine parts visual inspection system for surface defects, such as a crack, a pin-hole and other imperfections. The automated inspection system for defects on the three dimensional surface of machine parts has not been reported in the past, due to the complexity of analyses of light reflection data and the difficulty of measuring conditions. The features of defects are analyzed and classified. Also the features of defects can be indicated in terms of membership functions of the fuzzy technique. A new SPOT CHECK algorithm for defects on three dimensional surfaces of the machine parts has been developed. The moving average method, and the statistical treatment of light reflection data from the surfaces of the parts are applied to the SPOT CHECK method which has been developed. It is possible by using the SPOT CHECK method to measure the defects without influencing the measuring conditions, such as the illumination and the positioning of the parts. The neural network system has been developed, the system uses the membership functions of features of defects as the input data, while the kinds of defects as the output data. It is found that the neural network system has rapid convergence characteristics. The experimental verification results for the recognition of the kinds of defects are satisfactory.