
Abstract
The efficient design of battery cooling systems requires extensive testing and simulation. Current software possibilities allow for advanced CFD simulations or low fidelity, thermal mass-based models. In high fidelity simulations, the preparation of computational models requires a vast amount of time. Execution times are long. Obtained results often require time-consuming post-processing. Models created using this approach are not suited for the software-in-the-loop and system-level simulations. On the other end of the spectrum, low fidelity thermal lumped models are available. They offer fast execution but require the identification of parameters based on the experiment or CFD simulations. Low-fidelity models do not capture the spatial distribution of temperatures. They are significant for safety precautions, cell aging, and cooling systems design. In both cases, complete design iteration can span up to multiple days on the initial stage. To accelerate this design process, we suggest a reduced-order modeling approach. We present an efficient framework for the creation of reduced-order models for nonlinear thermal problems. Our key feature is the efficient management of ROMs of thermal components. We can assemble, run and post-process models with hundreds of components on a regular desktop computer. Equivalent full order meshes would reach a size of over 5 million degrees of freedom. We handle nonlinear material properties and fluid components. Our framework is capable of handling industrial-scale applications. We can generate digital twins with real-time execution. We present the methodology of assembling ROM of thermal systems out of ROMs generated for each unique component. We describe the assembly process of the ROM on component and system levels. Next, we present a methodology to handle nonlinear data, retaining low memory consumption. In the last part, we present Q-Bat- - the simulation environment based on our methodology. The framework is based on the MATLAB- scripting language. We show how to streamline and automate the ROM creation and analysis. The model can be compiled to the Simulink- environment, enabling system-level simulations. We present computational results and compare them, in terms of accuracy and performance, against high fidelity CFD analysis.