Mitigating Hotspots in Multi-disk Oil Cooled Brake Using Machine Learning

An accurate and rapid thermal model of an axle-brake system is crucial to the design process of reliable braking systems. Proper thermal management is necessary to avoid damaging effects, such as brake fade, thermal cracking, and lubricating oil degradation. In order to understand the thermal effects inside of a lubricated braking system, it is common to use Computational Fluid Dynamics (CFD) to calculate the heat generation and rejection. However, this is a difficult and time-consuming process, especially when trying to optimize a braking system. This article uses the results from several CFD runs to train a Stacked Ensemble Model (SEM), which allows the use of machine learning (ML) to predict the systems’ temperature based on several input design parameters. The robustness of the SEM was evaluated using uncertainty quantification. Next, a Particle Swarm Optimization method was applied to the parameterized gearbox, using the SEM to bypass the costly and slow CFD calculations. This delivered a gearbox optimized for the lowest disk contact temperature, using significantly less resources and time than a traditional CFD-based optimization process.