Expression of gear-tooth-flank deviations for Hobbing-Machine-Diagnosis system (Learning-data collection through hobbing simulation and their compression with network representation)

Abstract

A hobbed gear has pitch, profile, and helix deviations caused by various factors. However it is hard for even experts to specify which factors cause individual deviations from the measured ones. Therefore, a hobbing-machine-diagnosis system could play a key role in identifying the factors from hobbed-gear deviations. Artificial intelligence for image recognition, which is comparatively easy to use, could be appropriate for the system. Generally, a larger number of training data for artificial intelligence could improve the system accuracy, however, it must take a longer time for learning. Besides, efficient learning requires small image data. Typical gear-deviation diagrams are large images that include unnecessary information for the diagnosis system. Therefore, the image data for training have only information on the correlation coefficients between the deviations of two teeth selected at a time. The present paper describes a method for providing image data for learning. The image is the arrangement of correlation coefficients in rows and columns as a grayscale. This paper also describes the relationship between the image and the hobbing problems; e.g., radial or face runouts. As a result, the image data showed clear characteristics for the difference in the periodicity of the hobbing problem, however, did not show clear characteristics for the difference in the directionality.

Keywords

• hobbing machine
• profile deviation
• helix deviations
• correlation coefficient
• artificial intelligence
• image recognition
• training data