How battery thermal design can be accelerated

Nowadays, battery engineers who design cooling systems of the batteries of electric vehicles have a couple of ways how to accomplish their goals. We present a short review of each of the approaches in the first part of the article. Finally, we briefly describe the fast and accurate methodology that we use at QuickerSim and will elaborate on that in the next article soon.

When it comes to the thermal design of the batteries in electromobility, one can choose at least one of the following options to predict the thermal behavior of the complete battery pack:

• Iterative improvements of previous battery models,
• Prototyping with thermocouple or thermographic measurements,
• Simple 1-D computer calculations,
• Full, detailed 3-D CFD (Computational Fluid Dynamics) analyses,
• Fast reduced-order models of battery pack heat transfer.

We now briefly go through the pros and cons of each of the approaches.
 

Iterative improvements

This option probably requires little comment. Each company uses it. However, for battery-pack thermal design, it can be way too limiting. Electromobility develops rapidly, so do the electrochemical cells. It means that electrochemistries change (cathode, anode, electrolyte materials, etc.). Newer cells often have thermal properties (internal Ohmic resistance, heat capacities, thermal conductivities) different than those used a few years earlier. Iterative improvements are important but may lead to suboptimal designs due to outdating of some of the conclusions drawn from the previous projects and designs.
 

Prototyping and experimental testing

Often used by companies that are still premature in the design of the cooling systems or do not face large overheating problems in their batteries (e.g. the average charge and discharge currents are low and very small amounts of heat are dissipated within the battery pack). In such situations, experimental verification of the final design may be sufficient. It should also be done for at least one battery pack design when computational methodologies are introduced to the R&D workflow and the importance of experiments shall be reduced. As a sole R&D method, experimental testing is not recommended. It lasts a long to prepare the prototype, costs are high (materials, human work, time) and it is difficult to assess the influence of multiple parameters on the end result.
 

Simple 1-D calculations

Automotive or motive batteries never work in steady-state conditions. It means that proper thermal design and calculations must be done for transient cases. Once thermal properties of the cells are known or experimentally identified (thermal conductivity, heat capacity, internal resistance as a function of temperature, and the state of charge), one can quite easily build simple battery calculation models based on thermal energy balances. The advantage of the 1-D models is that they work very quickly and deliver transient results in seconds. There are two main drawbacks. First, they lose all the information of the temperature distribution in each cell (where temperature gradients in a cell e.g. during rapid charging can be as high as 10 degrees Celsius or higher between the top and bottom lid of the cell – as a result neglecting it may lead to wrong degradation predictions). The second drawback is that 1-D models are often either data-based or require properly tuned thermal resistance parameters. They are very difficult or impossible to estimate when we are still in the design phase. This causes the 1-D models to be good for quick predictions, but far too inaccurate for the serious design of the cooling system details and the R&D process.

3-D CFD modeling

CFD modeling allows getting very accurate temperature predictions for any CAD geometry. This is the largest advantage in the R&D process of the development of a cooling system of a battery pack. No preliminary operation data are required. Thermal properties of the components are sufficient to build an accurate simulation model and one can fully virtually investigate how design decisions will impact battery pack thermal behavior. However, preprocessing of the CFD model is very demanding. Preparation of the CAD geometry, simplification of the construction details, generation of the computational mesh, and setting up all the parameters in a complex CFD model often requires even a month of simulation engineer’s time. On top of that, transient runs take many hours (typically 2-5 days) on small supercomputers (around 100 CPUs) to get the results. That is a very accurate tool, but often too costly (even more in time-cost than financially) to be used effectively in the R&D workflow under stringent deadlines.

Reduced-order models

They can be built in a variety of ways. The main idea behind the approach is to get a simulation model that is significantly faster and much less computationally intensive than CFD. At the same time it is worth mentioning that there are at least two paths to follow when creating a reduced-order model:

• data-based model order reduction – it means that one needs to have a large database of results, experiments of real-life operation data to abstract a simulation model (often by means of regression, interpolation, or neural networks) that mimics the operation of the real component. This approach, however, is often insufficient for the R&D process – when we want to design something new – trusted data are usually not available yet. And if we have any data from the previous designs, it is rather not something that we want to use to mimic the new system and technology (especially in battery technologies that evolve quickly from year to year).
• numerical simulation models – this group of models is most important for battery designers. They are built based on detailed CFD discretizations (like finite element or finite volume discretization) and mathematically transformed into a properly designed vector or function space that is able to represent the very detailed 3-D temperature distribution in the battery pack using many times the smaller number of unknowns in the system. As a result, we accelerate all computations by a factor of around 100 and retain 95-98% accuracy of the temperature solution.

After drawing this landscape of methodologies, we dig in our next article into reduced-order models for thermal battery pack simulations that we use in QuikerSim and Q-Bat.