Ultra-fast Self parameterisation for lithium-ion battery models

Challenge Statement

Useful models of lithium-ion batteries are key to achieving a low-carbon economy. Models can both inform design improvements during manufacture and monitor working batteries in the field. Many models have been shown to reliably reproduce real battery behaviour [10.1149/1945-7111/ab9050], however there is a perpetual problem with finding time- and cost-efficient methods for accurately determining model parameters that maximise predictive power [10.1007/s10800-021-01579-5].

A model's efficacy depends upon having accurate parameters. Often these cannot be measured directly and must be inferred from potentially noisy or time varying voltage/current data. Inference requires either dedicated experiments and device teardown, or searching parameter space to find the best fitting model. When more than a handful of parameters are needed the speed of classical solvers make this search infeasible. This project will deliver robust and ultra-fast models to industry and tools which automate and radically accelerate the parameterisation process, making it deployable in the field.

Project Aims

Our aim is to combine ultra-fast surrogates (around one million times faster than classical solvers [arXiv:2312.17329]) with modern Bayesian inference techniques to automatically yield parameterised battery models from general and easy-to-measure data. The process will be sufficiently light-weight that it can either be used in active devices and in real-time to update model parameters as a device ages, or to entirely parameterise a relatively complicated physics-based model without the need for device teardown.

We will apply our methodology to the most industrially-relevant models (as guided by our partner Elysia). We will use their industrially-relevant field data to develop and test our methodology for both equivalent circuit models (ECMs) and physics-based models (PBMs). We will deliver the ultra-fast surrogate models and "selfparameterising" capability to Elysia in the form of user-friendly code that will readily integrate with their existing workflows and therefore be of immediate impact in their company.