MA – Physics-Informed Neural Networks for Inverter Temperature Prediction

The scope of this master thesis is:

  • Develop a physics-informed neural network for junction and coolant outlet temperature prediction in an automotive inverter under varying electrical and cooling conditions
    • Based on the thermal governing equation introduced earlier.
    • If necessary, extend the model to include additional heat-transfer paths, such as ambient convection and thermal coupling effects.
    • Estimate power loss and thermal resistance from available reference data using AI modelling.
    • Ensure real-time model speed capability.
  • Define representative driving profiles and measurement conditions for testbench data acquisition for model training
    • Select operating points that cover relevant electrical and thermal conditions.
    • Ensure that the collected data are suitable for model training and validation.
  • Develop a scripted workflow for automated training and deployment
    • Automated model training.
    • Configurable network architecture, layer sizes, node numbers, and training strategy.
    • Automatic export of the corresponding Simulink model.

Bearbeiter: Krithi Nagarajan

Betreuer: Dragan Stojkovic, Dr. Stefan Weigl (Schaeffler)

Verantwortlicher: Prof. Dr.-Ing. Martin März