Development of a Virtual Sensor for Torque Prediction in Electric Machines by Machine Learning Methods and Physical Modeling

EasyChair Preprint 14559

Abstract

The automotive industry is in a rapidly transformation process with the influences of the electric mobility and autonomous driving, which has also an impact on the knowledge of developers about the behavior of electric motors. In order to provide an accurate prediction of efficiency of electric motors and condition monitoring, it is essential to precisely determine the torque of an electric motor depending on various influencing factors. In response to this circumstance, this paper presents a methodology for the development of a virtual sensor for torque prediction. According the state of the art, virtual sensors are a good method to handle the challenge for a torque prediction. In order to eliminate the influences of a test specimen, two asynchronous machines with the identical construction (440 kW nominal power) are braced against each other with a cardan shaft. The electrical machines are equipped with a speed sensor on one side and a torque sensor on the other side of the machine rotor shaft. In order to crystallize an optimal method for developing a virtual sensor, data-driven AI models were created and discussed in comparison to physical models. The results of the individual model approaches are compared with real data and discussed in this paper. The black box model has an optimized accuracy in the area of torque prediction. The significance of this endeavor lies in its potential to revolutionize the field of electric vehicle engineering. Traditional physical sensors have limitations in terms of cost, complexity, and scalability. Our proposed virtual sensor offers a promising alternative, circumventing these constraints while delivering a specified torque prediction.  Furthermore, the discussion and comparison between the physical and black box approach build up a good basement and shows the possibilities of the different methods.

Keyphrases: Test rig, Torque prediction, efficiency prediction, electric motors, machine learning, virtual sensor

@booklet{EasyChair:14559,
  author    = {Lennart Kopp and Lukas Steidle and Jan-Niklas Molan and Alexander Maier and Jan Kratschmann and Florian Schmid and Markus Kley},
  title     = {Development of a Virtual Sensor for Torque Prediction in Electric Machines by Machine Learning Methods and Physical Modeling},
  howpublished = {EasyChair Preprint 14559},
  year      = {EasyChair, 2024}}