Data-driven desiGn foR automotivE battEry optimizatioN

Electromobility plays a key role in the transition to a climate-neutral society, which is one of EU objectives for 2050. Recall that in Europe, passenger and freight transport produces a significant proportion of CO2 emissions. The need to minimize them is leading to the biggest change in modern individual transportation in the last century. Society will assimilate and incorporate electromobility if electric vehicles have a reasonable price and usability (i.e., vehicle range).

The high cost and the significant weight of batteries make them the key factor in determining price and range of the vehicle, but battery design needs to respond to the crucial requirement of battery thermal management because it is critical for both battery lifetime and range.

Optimal design (i.e., most favorable thermal-flow coupled arrangement) of the battery cooling system is critical to attain the ambitious emissions targets envisioned.

GREEN aims to develop an innovative computational framework to study and virtually prototype next generation of battery thermal management systems.

State-of-the-art cooling is based on a number of serpentine cooling channels in a baseplate, usually optimized manually by trial-and-error, a very time-consuming and inefficient process due to the complexity of the heat exchange problem.

The project aims to devise a surrogate model for the solution of a parametric optimization problem. This is a novel extension of the surrogate concept. Instead of constructing a surrogate to replace the full-order computation or to accelerate the optimization loop, the new reduced order model (ROM) aims directly at the optimal solution to meet the industry needs in the required timing constrains.

The coupled thermalfluid problem with optimized geometry implies another novelty: accounting for the coupled multi-physics character of the models to build ad-hoc surrogates using Scientific Machine Learning techniques (or Physics-Informed Machine Learning). That is, including domain awareness as a fundamental property of the surrogates. The resulting surrogates are expected to be robust and accurate.

Finally, this methodology should be applicable to design real battery cooling systems. That is, in real full-scale automotive problems (not simple academic examples) where the amount of data (spatial information for geometry and primal variables) is humongous. This will require novel contributions in application of Scientific Machine Learning techniques to reduce the dimensionality of the data flow within the multi-physics coupling. The iterative search of the multi-physics solution is expected to be simpler, more stable, and faster and the input/output complexity for the disciplinary surrogates will be therefore reduced.

These challenges will be addressed by the research team longstanding expertise and proved contributions in data-driven surrogate modeling (also for CFD). Moreover, the team includes a top researcher form the automotive industry.