Abstract
Gear noise emitted from transmission units for automobiles should be evaluated by an expert of gear noise. Although on some production lines, quietness performance estimates from measured noise levels of the units, the estimation must be severe. There is no definite relationship between the measured noise levels and the evaluations by the experts. Therefore, for standing on the safe side, the estimation should be severe. As a result, such an automatic gear noise diagnosis system must yield transmission units with over-quality. The present study deals with a new gear noise diagnosis system to which an artificial intelligence is applied; i.e., a neural net is applied. The former evaluations by the new gear noise diagnosis system were quite good when the statistical property of the teacher signals from which the system learned was similar to that of population. This fact means many teacher signals are necessary on the practical use. The present paper describes a new method for digitizing the measured noise levels. This method provided good evaluations of the system even when the statistical property of the teacher signals were not similar to that of the population. In addition, a new method, “Moment Method” for determining the construction of the neural net was introduced instead of “Back Propagation Method”. The Moment. Method contributed to the improvement of the system judgements.
