Master Thesis Identification of Models for Electric Machines Using Differentiable ODE Solvers

The identification of accurate simulation models of electric machines is a crucial step for the design of high-performing controllers, fault diagnosis and many other tasks. The goal of your Master thesis is to investigate and develop effective optimization and training methods for the identification of electric machines. The focus lies on the design of physics-enhanced data-based models and the implementation of training algorithms. Of further interest is the evaluation of various optimization techniques and their impact on model accuracy and computational efficiency.

During your Master Thesis your task includes the familiarization with the physical models of electric machines.

• Familiarization with the physical models of electric machines.
• Conduct literature research on existing ML-based approaches for the identification of electric machines.
• Set up a differentiable parametric model of the electric machine.
• Implement and compare different training and optimization methods, such as quadrature-based approaches, multi-step ODE solvers utilizing various integration techniques and prediction horizons, neural ODEs and methods involving the differentiation of measured currents.
• Evaluate the training methods in terms of convergence speed, accuracy, robustness and computational cost.

Your profile

• Education: Master studies in the field of Electrical Engineering, Technical Cybernetics, Engineering Sciences, Computer Science or comparable
• Experience and Knowledge: proficient programming skills in Python or Julia, as well as experience with Machine Learning and frameworks for automatic differentiation (e.g., PyTorch, JAX or Flux.jl); basic knowledge of dynamical systems (differential equations); understanding of the physics of electric machines, system identification; experience with MATLAB/Simulink to interact with legacy simulation models
• Personality and Working Practice: you are a communicative and reliable individual with a critical mindset and a proactive approach
• Languages: good in English