Physics-informed machine learning-based swelling models for future battery cells
Physics-informed machine learning-based swelling models for future battery cells
Efficient batteries for automotive industry are critical for achieving net zero goals and the future of our planet. During their lifetime, those energy storage systems can experience complex electrochemical-thermomechanical phenomena that can result in their volumetric changes (so called swelling).
Swollen batteries are at risk of rupturing which may significantly shorten their lifetime. Development of advanced computer models is critical for understanding and optimization of batteries against the swelling phenomenon.
This project aims at developing scientific machine learning approaches based on the Bayesian paradigm and electrochemical-thermomechanical models in close collaboration with our project partner Jaguar Land Rover.
Supervisors
Primary: Dr Lukasz Figiel, Warwick Manufacturing Group
Dr Ferran Brosa Planella, Maths
Project Partner: Jaguar Land Rover
A transcript of the video is available by clicking this link - transcript opens in another windowLink opens in a new window
Background
Clean future of our planet relies on the development of efficient propulsion systems for automotive industry. Energy storage devices such as lithium-ion batteries are an example of such a net-zero solution. Unfortunately, those systems can undergo significant volumetric changes (so called swelling) that originate at the material level as structural changes (frequently anisotropic) due to the absorption of lithium ions during electro-chemical cycling of batteries in operation. In turn, swollen battery cells are at risk of internal damage, thus potentially leading to significant reduction of their lifetime. Our ability to predict those volumetric changes right from the material level and to use that information to optimize battery cells is of practical importance to automotive industry. Classical electrochemical-thermomechanical (ECTM) models are typically deterministic and insufficient to identify most optimum battery materials within certain confidence level and required computational efficiency. Recent advances in Machine Learning (ML) offer promising means for developing hybrid ML-based ECTM models that can overcome computational deficiencies of classical models and account for uncertainty in their predictions.
Project goals
An ultimate aim of this project is the development of a physics-informed data-driven model for battery cells to capture their volumetric changes in a typical automotive battery in operation. The model will incorporate anisotropy effects at the active material level by extending particle-level models, and will combine it with the Bayesian framework to generate surrogate multiscale models for battery cells able to quantify uncertainties. The new modelling approach will be initially developed and experimentally validated for state-of-the-art cells based on graphite and NMC electrodes. It will be subsequently utilized to infer material properties that minimize cell swelling.
The main outcomes of the project will be as follows:
- New machine learning-based electrochemical-thermomechanical (ECTM) model capturing effects of material anisotropy on cell swelling across the scales with uncertainty quantification
- Experimentally-validated computational platform with numerical implementation of an ECTM model and Bayesian approach within a Python-based machine learning tool.