The journey towards this paper started in 2017 and began after reading the book “The Art of Action: How Leaders Close the Gaps between Plans, Actions and Results” by Stephen Bungay. It wasn’t chance, it was the result of following a philosophy called directed opportunism. The next step was to have a strategy, and after a few false starts, our strategy in 2019 became to develop a universal model to be able to predict battery lifetime, for any chemistry and independent of how you use it. I estimated it would take us 10 years and require ~200 people.
We kept our strategy to ourselves for a few years whilst we built the foundations, with the first fruit being a landmark review paper in 2021 on “Lithium ion battery degradation: what you need to know”. This paper was about creating a map, of degradation mechanisms, so we knew where we were going. During this period the Faraday Institution in the UK was also started and I found myself running a >£10M multi-scale modelling project for batteries. Under this project we adopted PyBaMM which was born in Oxford, and started throwing resources at it to create a framework within which we could start coupling all the degradation equations. Also to create a community of collaborators to help us reach ~200 people working together. This resulted in another landmark paper on “Lithium ion battery degradation: how to model it” in 2022. This was the first paper to couple four degradation mechanisms simultaneously and could predict path dependent behaviour.
In 2020 we also kicked off a large scale (for an academic group) degradation study on commercial cells, which took 2 years to gather and another to analyse and use, which we later published in the paper on “Lithium ion battery degradation: Comprehensive cycle ageing data and analysis for commercial 21700 cells”. We did this knowing we would need data to parameterise and validate our models for a commercial cell if industry was going to believe these models could be useful and trustworthy. Our partners in the MSM project from Birmingham published the now famous ‘Chen 2020’ paper on “Development of Experimental Techniques for Parameterization of Multi-scale Lithium-ion Battery Models”. Crucially this paper had a full set of parameters for our physics-based models of the cell we were conducting the large-scale degradation study.
Another crucial conclusion that we made from the 2021 review paper, was that capacity fade and power fade were not enough. There was a missing layer of identifiable states called degradation modes, pioneered by the likes of Dubarry in "Synthesize battery degradation modes via a diagnostic and prognostic model". These needed to be measured during degradation experiments and included as states in the degradation models, leading to our paper on “Lithium-Ion Battery Degradation: Measuring Rapid Loss of Active Silicon in Silicon-Graphite Composite Electrodes”. The goal for the team was then to validate the degradation models against the experimental data for a commercial cell, where the model had to correctly predict the evolution of degradation modes over time, not just capacity and power fade, and for multiple temperatures and different ways of cycling the cell. A very difficult task.
To enable this we also had to work on additional upgrades, such as solvent consumption (coupled degradation mechanisms now reached 5) in the paper on Modelling solvent consumption from SEI layer growth in lithium-ion batteries. We also worked with the PyBaMM team to add composite electrodes, allowing us to answer the question “Is silicon worth it? Modelling degradation in composite silicon–graphite lithium-ion battery electrodes”, and sped up the models so we could simulate “A million cycles in a day: Enabling high-throughput computing of lithium-ion battery degradation with physics-based models”. This was necessary to enable us to find a parameter set for the degradation models, as we had to resort to brute force for the first attempt.
After many years we eventually succeeded, with the results shown in this paper. However, we are not done, there are still 4 years left from our initial estimate, and the more we learn the more we realise needs to be done, so we probably won’t be finished in 2029. There are still multiple degradation mechanisms and different physics that need adding to PyBaMM, and existing equations that need improving. We also need to continuously improve the parameterisation procedures and develop new accelerated degradation testing protocols that are needed to validate the models. So, although this paper is another important landmark on our journey, it is not the end.