Stefan Holbach

• This thesis aims at investigating the capability and feasibility of Machine Learning algorithms for developing models simulating the behavior of E/E powertrain components. Machine learning based simulation models possess the advantage of being trained via real measurement data and no time-consuming manual set up of equation and parameter adaptions are needed to get a proper simulation model of the component. For this purpose, the thesis starts with the introduction of E/E powertrain components of interest. Moreover, Machine Learning algorithms are introduced that support model based and supervised training and are hence of interest for behavior simulation. The design, implementation, training and optimization of the different Machine Learning based simulation models according to the provided data is presented. These models are not only simulation models of the single introduced components but also models of the composition of these components. The resulting models are evaluated against test data which has not been used for training. This evaluation illustrates the ability and inability of the different Machine Learning algorithms to simulate and generalize specific powertrain components. It also illustrates the necessary scope of the models according the number of composite components and their accuracy.