Digital twins are used to optimize digital assets in an industrial context using simulations and real-time data \cite{RaS20}. Digital twins of electrical machines require mathematical models that are sufficiently accurate for the design and sufficiently fast to be used for condition monitoring. This leads to a hierarchy of models that are needed by the digital twin \cite{ChSM23}. In the case of electrical machines, these models must in addition incorporate the physical coupling between electrical, mechanical, and thermal phenomena for more accurate computations. However, the computing overhead that these models typically bring makes real-time simulations unfeasible. In this presentation, we present how one can construct this model hierarchy by incorporating physics-based and data-driven modeling. This is enabled by the data that is collected by the digital twin from the electrical machine in real time. The software tools necessary to carry out simulations in the digital twin are also covered. The pipeline in which the simulation models are implemented enables the automatic configuration of new digital twins and the modeling of new machines. The digital twin's overall software architecture incorporates the implemented simulation models. The digital twin then selects the suitable simulation model from the model hierarchy based on the relevant task and function. In order to evaluate the created structure and the different services, we provide an experimental implementation of the concept for the digital twin and the data flows using a hardware demonstrator in conjunction with an implemented software demonstrator. The outcomes of the experimental results and open problems are finally discussed.