Evaluation System of Tooth Contact Patterns of Hypoid Gears Using Artificial Intelligence Available to Purchase

Tooth contact inspection is one of the most common methods for checking qualities of hypoid gear pairs. A change in machine setting parameters for cutting and lapping processes of a hypoid gear pair enables a tooth contact pattern of a hypoid gear pair to be varied. The deviation of the pattern from the target one is represented by a grade point. In the inspection, the qualities of hypoid gear pairs are usually classified into only two grades; OK or NG. However, in order to conduct a follow-up survey on problems of the products and to be useful to be trouble shooting tasks of the end products, finer classifications and more quantitative evaluations of tooth contact patterns could be effective. Such approaches have been tried, however, only experienced and well-trained technicians for the inspection of hypoid gear pairs can determine the point of each tooth contact pattern. And it is difficult to make this evaluation method automatic. To overcome this problem, an application of artificial intelligence system must be useful. The present paper describes a computer evaluation system using the neural network, which is a kind of the artificial intelligence systems, for tooth contact patterns of hypoid gear pairs which can evaluate the results of the inspections instead of experienced hypoid gear technicians. This system with the neural network has a capability to learn relationships between evaluation grade points of tooth contact patterns given by the hypoid gear technicians and graphics of tooth contact patterns of hypoid gear pairs. Moreover, it can return the evaluation grade points when a tooth contact pattern is input into the system. The evaluation performance of the developed system was discussed. And a quality of normative tooth contact patterns, which were used as the teacher signals for training the neural network system, greatly affected its performance. The comparison of evaluated grade points obtained from developed system with the technician’s ones showed that the correct answer ratio obtained from the developed system was about 90% in the best case.