Computationally-efficient thermal simulations of large Li-ion battery packs using submodeling technique

Accurate and rapid prediction of temperature distribution in a large Li-ion battery pack comprising thousands of cells is critical for ensuring safety and performance of battery packs for electric vehicles. Due to the multiscale geometry and the large number of individual cells in an automotive battery pack, fullscale thermal simulations typically take a long time to complete. Approaches for rapidly computing the temperature field in a large battery pack without significant loss of accuracy is, therefore, a key technological need. This paper presents thermal simulations of a large, air-cooled Li-ion battery pack containing thousands of individual cells using the submodeling technique. A coarse model that neglects fine geometrical details is first solved, and the results are used to solve a more detailed sub-model. It is shown that this approach results in 7X reduction in computation time while preserving the accuracy of the predicted temperature field. Trade-offs between computation time and accuracy are examined. The submodeling technique is used to investigate the thermal design of a large Li-ion battery pack, including the effect of discharge rate and coolant flowrate on temperature field, and the thermal response to a pulsed spike in discharge rate. Limitations of the submodeling technique are discussed. The technique discussed here is a general one, and may help significantly reduce computation time for thermal design and optimization of large, realistic Li-ion battery packs. (C) 2020 Elsevier Ltd. All rights reserved.

Automated summary

This paper presents thermal simulations of a large, air-cooled Li-ion battery pack containing thousands of individual cells using the submodeling technique. A coarse model is first solved, and the results are used to solve a more detailed sub-model, resulting in a 7X reduction in computation time while preserving the accuracy of the predicted temperature field. Trade-offs between computation time and accuracy are examined, and limitations of the submodeling technique are discussed.