Abstract
There are various causes for deviations in the tooth profile and flank line of hobbed gears, and it is difficult to estimate defects during hobbing from tooth profile deviation and helix deviation. Therefore, in this research, using artificial intelligence for image analysis, we aim to develop a diagnostic system that can identify motion errors contained in the hobbing process from the profile and helix deviations of hobbed teeth. The AI learns a classification of motion errors from a training dataset created through computer simulations of hobbing processes with various motion errors. In previous work, we developed a hobbing simulation and used it to create training data for artificial intelligence. This paper discusses the impact of training data on the classification accuracy of AI systems and determines the requirements for improving accuracy. As a result, it was found that the classification accuracy was improved by providing AI with training data containing noise. Also, in order to improve the classification accuracy for gears with multiple motion errors, training data created from the results of hobbing simulations with multiple motion errors is required.
