Leveraging Historical Thermal Wind Tunnel Data for ML-Based Predictions of Component Temperatures for a New Vehicle Project

Abstract

<div class="section abstract"><div class="htmlview paragraph">The thermal operational safety (TOS) of a vehicle ensures that no component exceeds its critical temperature during vehicle operation. To enhance the current TOS validation process, a data-driven approach is proposed to predict maximum component temperatures of a new vehicle project by leveraging the historical thermal wind tunnel data from previous vehicle projects. The approach intends to support engineers with temperature predictions in the early phase and reduce the number of wind tunnel tests in the late phase of the TOS validation process. In the early phase, all measurements of the new vehicle project are predicted. In the late phase, a percentage of measurements with the test vehicle used for the model training and the remaining tests are predicted with the trained ML model. In a first step, data from all wind tunnel tests is extracted into a joint dataset together with metadata about the vehicle and the executed load case. With the extracted dataset, different Machine Learning (ML) models are trained and optimized to predict maximum component temperatures. The ML-based temperature prediction is evaluated by two vehicles with different vehicle platforms. Temperature predictions are made for engine and suspension bearings. The predictions in the early phase show very good results with a maximum mean absolute error (MAE) between 3 and 5 °C for the first vehicle. The evaluation of the second vehicle results in higher MAEs between 3 and 8 °C. In the late phase, the MAEs lie between 2 and 4 °C for the suspension bearings of both vehicles using 20 % of the wind tunnel tests. Despite the promising results, the ML models need to be further improved and tested to use them as an accurate and reliable predictor in the current TOS validation.</div></div>