Generation of Artificial Learning Data to Train Functional Meta-Models of Micro Gears

Abstract

The complexity of products increases considerably and key functions can often only be realized by using high-precision components. As a result, the requirements of high-precision components reach technological manufacturing limits. This is of particular importance for micro components, such as micro gears, where the manufacturing deviations are relatively large compared to the component size and therefore have a large influence on the functional characteristics of the assembled product. A potential solution is to develop a meta-model with machine learning (ML) methods, which enables the real-time prediction of functional characteristics of micro gears in terms of kinematic transmission error (KTE), based on optical in-line measurements of the entire topography of the gear. To train a suitable ML model large amounts of data are needed. This poses a challenge because real measurements cannot cover the entire spectrum of all deviations and also require a large amount of resources in terms of time, material and costs. In this paper, an approach for the generation of learning data on the basis of virtual and deviated gear models is developed with the help of Skin Model Shapes. The objective is to generate gear data that reflects the nominal shape as well as the process-related manufacturing deviations and the random scatter. For validation, the artificially generated gears are compared with a reference dataset both geometrically and functionally. Therefore, both the standard deviations and the KTE are calculated for both data sets.