Predictive Control of Electric Machine Cooling System for Extended Lifetime and Performance

Sammanfattning

Thermal management is an important challenge in heavy-duty battery electric vehicles, since elevated electric machine temperatures can accelerate component aging and may limit available performance. Conventional cooling strategies are typically reactive and rely on present temperature measurements, which limits their ability to prepare the cooling system for upcoming thermal loads. This thesis investigates whether predictive cooling control can improve electric machine thermal management by using preview information about future road slope. A predictive cooling framework is developed by combining a preview-based operating-point predictor, a control-oriented grey-box thermal model, and a model predictive controller. The road-slope preview is used to estimate future vehicle speed, gear selection, electric machine torque and speed, which are then converted into a predicted heat-loss sequence. This predicted thermal load is supplied to the predictive controller, which coordinates the available actuators over a finite prediction horizon. The framework is evaluated in a simulation-based co-simulation environment using four driving scenarios. The predictive controller is compared against a reactive baseline controller. The results show that the predictive controller consistently reduces the maximum hotspot temperature, mean hotspot temperature, and equivalent thermal load temperature across the evaluated real-route scenarios. The equivalent thermal load temperature was reduced by approximately 6%, while the maximum hotspot temperature was reduced by up to 8.6%. The effect on thermal cycling was more scenario-dependent: the equivalent thermal cycling amplitude decreased in two scenarios but increased slightly in the most demanding route. Under severe thermal loading, the predictive controller delayed the onset of high hotspot-temperature regions and reduced the time spent above selected thresholds, indicating improved timing of cooling action in high-temperature operating regions. Overall, the thesis demonstrates that preview-based predictive cooling is technically feasible in a simulation-based framework and can reduce temperature-related thermal stress compared to a reactive benchmark. However, the results also show that the benefits depend on route severity, available cooling authority, prediction accuracy, and thermal model fidelity. Further work is required to validate the approach with broader datasets, assess sensitivity to preview uncertainty, and evaluate real-time implementation in an embedded vehicle environment.