Surrogate model based prediction of transmission error characteristics based on generalized topography deviations

Abstract

Simulation models of the tooth contact can provide valuable information of the gearbox operational behavior for deviated gears. It is therefore possible to gather geometry and deviation information of the two gears in contact and then estimate how they will perform in the assembled unit before even mounting them. This procedure could save time and costs used to disassemble gear boxes which fail the end of line tests because of acoustic reasons. The major challenge using sophisticated simulation models, which can describe the tooth contact is the tremendous amount of calculation time they need to produce suit-able results. In a productive gearbox manufacturing process this time is just not available. Therefore, a method to predict the operational behavior faster, than with the current used simulation models is needed. Firstly the size of the problem is downsized by introducing the sum deviation surface. It allows to reduce the number of necessary input parameters for a gear topography description to eighteen factors. With the help of that sum deviation surface, 3000 variants within a given variation space are calculated. The resulting training dataset is used to develop a deep neural network meta-model of the gear contact, which can predict the characteristics of the transmission error under load. With the help of that meta-model, the excitation of a gear pair can be predicted faster than real-time.