ABSTRACT

In the series production of gearboxes, most gearboxes are usually tested in an end-of-line test after assembly. Among other things, the noise of the gearbox is checked during such tests. A reason for an acoustical failure can be an unlucky combination of the manufacturing deviation of two meshing gears. If a gearbox appears faulty, it has to be disassembled again, which is labor and cost intensive. To reduce the number of gearboxes sorted out for acoustics reasons, the gearbox acoustic behavior can be simulated prior to its assembly. The simulation considers the specific deviated gears of the unit. This allows the detection of the combination of unfavorable manufacturing deviations. A simulation model domain, generally used to calculate the acoustic behavior of the whole gearbox, is an elastic multi body system model. This model type calculates the acoustic behavior in time domain to consider the nonlinear meshing behavior of gears and other components. The major challenge using this model type is the tremendous amount of calculation time needed to produce suitable results. That is why a method to predict the acoustic behavior of the gears faster than with the currently used simulation models is needed. Otherwise, the simulation cannot be performed in parallel to gearbox manufacturing. To achieve calculation speedup, the size of the mathematical problem is reduced by introducing the sum deviation surface as a generalization of gear topographies. It allows reducing the number of necessary input parameters for the new model. Using the sum deviation surface, variants within a given deviation space are calculated as training dataset. This dataset is utilized to develop a deep neural network model of the gearbox. The network then predicts the acoustic behavior of a whole gearbox based on the topography deviations faster than real-time. Prediction is verified by an independent test data set.

Author(s): Marius Willecke, Jens Brimmers, Christian Brecher

ISBN: 978-1-64353-161-8