Abstract

Dynamic multi-body simulation (MBS) models are important for the development and analysis of gearboxes. With the help of these models, the operational behavior of the gearbox can be examined in detail, without having to produce a prototype. This makes it possible to investigate the influence of various factors, such as misalignments and deformations of components, on the operational behavior of the gearbox. The modelling of the gear mesh is particularly relevant to be able to model the behavior of the gearbox realistically. If the gear mesh should be calculated with an enhanced level of detail, numerical FE-based tooth contact analyses (TCA) are used for each integration step of the dynamic model. These TCA models can be integrated in the Simpack MBS with the help of User Forces. However, the disadvantage of this highly precise method is the longer calculation time that results from solving the TCA for each calculation step. This prevents efficient use in the product development. To improve the calculation time of the MBS model, a neural network is developed in Tensorflow. In the trained range, the network predicts the behavior of the gear contact as accurate as the TCA can calculate it. A User Force is then used to load the Tensorflow network in the MBS and predict the resulting toques and forces of the gear mesh in each integration step. This speeds up the simulation enormously and enables the efficient use of highly accurate gear contact modelling also for complex models in the MBS.

Slide deck

Presenter Bio

Marius Willecke studied mechanical engineering from 2013 to 2019 at the RWTH Aachen University. During his studies he started working as a student assistant at the gear department of the Laboratory for Machine Tools and Production Engineering of the RWTH Aachen University in 2015. After graduating with honors, he continued his work in the gear departments acoustics group as a research assistant. His main fields of research are the development and programming of simulation models for quasi static and dynamic gear box models. Starting in 2025 he took the lead of the development of the Simulation Software “WZL Gear Toolbox”. Additionally, to his research work, he is also an active member of the VDI working group for the definition of the Gear Data Exchange Format (GDE).