Thermal Neural Networks for High-Resolution Temperature Modeling of Electric Traction Machines With Oil Spray Cooling
Niels Wiese, Konstantin Etzold, R. C. Reinhardt, Markus Henke
Abstract
The demand for competitive electric traction machines expands the boundaries of their overload capability and energy efficiency. This requires an increasingly precise thermal monitoring to ensure protection from thermal damage and temperature dependent control adjustments. Thermal neural networks represent a combination of solely data-driven neural networks and lumped-parameter thermal networks. They have been proven to manage the task of thermal modeling for real-time online applications and even outperform standalone neural networks or lumped-parameter thermal network approaches, while requiring less parametrization effort. However, there is a particular shortage of studies focusing on thermal neural networks for temperature modeling of electric drives with an oil spray cooling. This contribution shall provide a step towards closing this gap. Thereto, a thermal neural network is trained and validated on measurement data of an electric machine with a stator cooling jacket and an oil spray cooling. An analysis of the derived model points out measures to improve the prediction performance or interpretability. The complex thermal system with the oil cooling is adopted by the best network presented in this article with a mean squared error of 0.41 °C <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> on a validation dataset, consolidating the potential of thermal neural networks for online temperature monitoring of electric machines.