Thesis in Development of a Learning Based Compositional Electrical Drive Model

The identification of accurate simulation models of electric drive systems, comprising the inverter, an electric driver, and further components, is a crucial step for the design of high-performing controllers, fault diagnosis, and many other tasks. Goal of the thesis is to develop a compositional model for electric drives that allows for simulation, identification and control using automatic differentiation techniques. The main idea is to implement differentiable models for components of an electric drive that can be freely combined to an overall system model. You will familiarize yourself with physical models of electric drives (electric machines, inverters, ...). You will do literature research on existing (ML-based) approaches for the identification of electric drives. Furthermore, you will develop the dynamical physical electrical drive model combined with data-based models. Last but not least, you will implement the proof of concept to demonstrate the gradient-based optimization of the overall model for a given example system under using dynamical data with ODE solvers.