The thermal operational safety (TOS) of a vehicle ensures that no component exceeds its critical temperature during operation. The current TOS validation process has two shortcomings. Firstly, temperature predictions in the early phase of the vehicle development process are solely based on expert knowledge. Secondly, the testing effort in the wind tunnel is time-consuming and expensive. This work proposes a data-driven approach to predict maximum component temperatures of a new vehicle project for different load cases by leveraging the historical thermal wind tunnel data from previous vehicle projects. In a first step, data from the single measurement files is extracted into a structured dataset including metadata about the vehicle and the executed load case. With the extracted dataset, different Machine Learning (ML) models are trained to predict maximum temperatures of engine and suspension bearings. The hyperparameters of each ML model are optimized and the trained ML models are compared regarding their cross-validation loss. Two new combustion engine vehicle projects are used as test data to evaluate the performance of the ML models. The predictions in the early phase result in MAEs between 3 and 7 °C. To investigate the potential of reducing the number of wind tunnel tests, a certain percentage of measurements with the test vehicle is used for the model training. The remaining measurements are predicted by the trained ML model. Already with a small percentage, the MAE for the predictions of the remaining measurements can be lowered for all components significantly. Thus, the MAE for all component lies below 4 °C. Despite the promising results for both applications, further investigations are required to reliably embed the data-driven temperature predictions into the current development process.
Session:
AERODYNAMICS
|
| 14:30 - 15:00