Predictive Temperature Distribution of Test Rig Electric Motors Using Machine Learning Algorithms

The consequences of climate change are increasingly noticeable in everyday life and require consistent and immediate action. A multitude of legislative restrictions regarding the permissible emission of pollutants in the transport sector requires a redesign of the powertrain architecture. The focus on hybrid and purely electric drive concepts boosts the demand for electric motor development and testing. The electric motor is the central component in the electric powertrain. Consequently, its efficiency is a significant lever for improving overall vehicle efficiency. A large part of the energy losses in the electric motor is due to the electrical resistance. The electrical conductivity of the windings is temperature-dependent, making thermal management crucial for the efficiency of the electric motor. Furthermore, thermal management has significant influence on the health condition and the expected lifetime of the electric motor. By analyzing and evaluating the temperature distribution, damage-relevant operating points can be identified and avoided in advance. The determination of a continuous temperature distribution opens up new possibilities for condition-based maintenance as well as the development of innovative operating strategies. A consistent temperature distribution within the motor map is the foundation for optimal electric motor efficiency. Depending on different load collectives and operating modes, initial experimental investigations on the test bench have shown that different temperature distributions occur within the motor map. The application of machine learning algorithms enables the prediction of the temperature distribution. This includes, on the one hand, the identification of critical temperature areas and, on the other hand, the identification of potential temperature peaks depending on the driving cycle. The overall goal is to determine a continuous temperature curve over the entire motor map. By using a few temperature measurement points in conjunction with suitable machine learning algorithms, results can be obtained regarding the temperature distribution at any point in the motor map. By applying and subsequently comparing standardized machine learning algorithms the most suitable algorithm is determined.