When it comes to battery modeling, there’s no magic bullet. The right model for a specific case depends on what questions need to be answered, and which phenomena drive the results of interest. For example, perhaps we would like a model that can predict the full cycle life of a cell with data from only a small number of cycles. We could build a model that projects coulombic efficiency (CE) and cumulative lithium loss to predict capacity for the Nth cycle. But this model would be useless for a battery that has high CE until the SEI (solid electrolyte interphase) layer catastrophically fails after 200 cycles. In that case we would need a model that focuses on volumetric expansion, mechanical stress, and fatigue.
Is the coulombic efficiency model useless, then? If it can’t predict SEI failure, then what’s the point? If a cell starts out with poor CE, or if the CE trends in a way that we know there will be too much lithium loss, then our CE model can still free up test channels by ruling out some cells earlier. The CE model is very simple, but it can answer a very specific question quickly and easily. We can use a separate mechanical fatigue model to help us rule out certain cell designs before we build them, but that mechanical model might not tell us might about CE.
We could also build a model that very completely simultaneously models all the interesting phenomena for a specific cell. This is known as the “digital twin” approach, and it only works when applied to a narrow scope. We could take a single cell design that has been characterized experimentally and fit a comprehensive model. That might include time-series lithium transfer and diffusion, chemical potentials throughout the cell, and the effects of chemical potential on SEI growth to predict performance degradation. If this model is wrong, then we can try to figure out what we’ve missed and amend the model until it is correct. If this model predicts the terminal voltages, capacity, and degradation, then we probably have a good understanding of what’s going on in the cell.
The search for the “everything” model for all phenomena in all cells – the singing, dancing, all-knowing model – is usually a waste of resources. The most (cost-) effective models will answer specific questions in specific cases but not all questions in all cases.
Some examples of relevant battery modeling approaches are:
Data-based analytical models
This family of models can be used to extract knowledge from experiments and understand the performance and interactions of materials that the company has tested. These models can help scientists determine which cell designs to try next or predict the best cell designs for specific applications. Micantis has these battery modeling analytics features built-in to the WorkBook platform.
Finite Element Models and Finite Difference Methods in 1D, 2D, or 3D (or 0-D Lumped System)
These types of physics-based models can help scientists understand how cells function and are most useful for understanding single-cell-scale and pack-scale phenomena. Two specific models in this category are Lithium-diffusion/Nernst-based models (i.e. to understand which cell component is rate-limiting) and heat transfer (intra-cell, pack-level, or system-level).
Often these models are “built” with software tools like Comsol Multiphysics but require an expert to run and maintain them. A focused FEM-based tool that can be applied to narrow applications is available to be packaged into the WorkBook platform.
Density Functional Theory (DFT) Model
DFT can be useful to understand specific quantum chemical phenomena such as favored reaction kinetics, how that varies chemical potential, and the impact of introducing different molecules into the system. A DFT model can help scientists understand underlying irreversible chemical reactions and preselect candidate material design changes to improve performance. This information can also improve the scientist’s intuition and help them make better design choices.
There are a several DFT software libraries available. Micantis is currently using Quantum Espresso.
Multiscale and Composite Models
Multiscale models incorporate two or more models into a single system. This can model how phenomena at different physical or temporal scales interact to predict outcomes that can’t be resolved without considering all these phenomena simultaneously. The output of one model can feed input into another, or the models can operate concurrently. For example, a model of cell aging could use a DFT model to predict kinetics, and a lithium diffusion FEM to model the evolution of the chemical potential as the SEI grows.
The Micantis team has a wealth of battery modeling expertise in a range of applications including materials science and electrochemistry. The Micantis team can apply our repertoire of proven models and techniques to provide the tools you need to build better batteries faster.