E-Bike Drive Units are typically viewed as standard components and therefore mounted in all bike categories as e.g., cargo, mountain, and trekking bikes. Previous investigations and measurements have already shown that this leads to a wide range of loads acting on the drive unit due to the different drivers, bicycles and driving situations. To make the engines available as a single standard component the whole powertrain must be installed in one housing that has to resist all impacts and loads out of any possible cycling situation. This paper is therefore intended to show the systematic development of a computationally efficient model and calculation process for the drive unit, which enables the determination of the operational stability and fatigue of its mechanical parts. The most important influencing parameters that must be modelled are the pedal forces, the asymmetrically acting chain force, gear and motor forces, thermal loads, and static loads from the assembly and pretensioning process. Another major influencing factor is the frame, which can either introduce external forces into the drive unit, or due to different stiffnesses, shifting boundary conditions are imposed at the interfaces connecting the drive and the frame, and thus strongly affects the flow of forces within the drive unit. Last, but not least, geometric parameters like the mounting angle, pedal length and the chain line should not be neglected. Because of this immense number of input factors a systematic approach had to be developed to get suitable combinations of these input datasets and to reduce the number of input parameters. On the one hand, the measurement data is clustered into individual driving situations with a known distribution and amplitude, on the other hand, systematic correlations are used to limit the possible combinations. Likewise, static and dynamic loads are calculated separately and then superimposed. In response to the high number of simulations, an automated workflow was developed for the generation of input data, their calculation, and a standardized evaluation. In addition, a surrogate model is formed in this program to estimate the results of additional load points so that the entire load spectrum can be represented with as few simulations as possible. In order to receive a runtime efficient FEM-model, several reductions and simplifications were performed and will be discussed. This includes the substructuring of the frame and a reduction to certain relevant stiffness parameters as well as an analytical model which replaces pedals, crankshaft, chainring and the gear system to represent them by a surface pressure in the outer shell of the bearings. To avoid time-consuming calculations for hyperplastic parts such as rubber sleeves and seals, their static behaviour was determined in a submodel and replaced by a nonlinear analytical connector. In the end this model allows the efficient simulation of all relevant load cases in the most common driving situations, taking into account the highly variable input variables. Finally, it will be shown how these modelling approaches can be used in different stages of product development, with particular emphasis on robust optimization, design validation and fatigue calculation.