Boosting battery life with hybrid machine learning of degradation mechanisms
Boosting battery life with hybrid machine learning of degradation mechanisms
Battery degradation poses a significant obstacle to efforts to decarbonise the economy. To meet electric vehicle targets for the next decade, design strategies are required to extend battery cycle lifetimes. The goal of the project is to quantitively describe the mechanism which traps Li-ions behind atomically thin surface layers formed by oxygen loss by applying machine learning methods to high quality X-ray data with the aim of identifying the surface physics models which agree with the measurements.Supervisors
Primary: Prof. Louis Piper, Warwick Manufacturing Group
Dr Florian Theil, Maths
A transcript of the video is available by clicking this link - transcript opens in another windowLink opens in a new window
Background
High-end electric vehicles use Ni-rich layered oxides in their Li-ion batteries, which offer high energy densities but suffer from accelerated degradation. Recent studies have shown that Li-ions can become trapped behind atomically thin surface layers formed by oxygen loss, but the reason is unclear. Modelling the transport properties across these boundaries is critical for identifying and evaluating engineering solutions. This PhD project will have access to unique battery studies at Warwick to test these models.
Approach
- Innovative operando X-ray data of WMG built cells under various cycling rate and aging conditions will be collected at experimental facilities at the Warwick Diffraction RTP e.g. Ref. [PRX Energy 2024, Joule 2024 in-press]. We also have ongoing long duration (6-month) operando studies at Diamond synchrotron facility.
- Use Physics-based surface diffusion models from the literature to develop continuum models that account for the growth of surface layers.
- Integrate the RS layer equations into standard electrochemical cell models, such as the Doyle-Fuller-Newman (DFN) model consisting of several coupled partial differential equations. Machine learning will be used to whittle down the complexity of the DFN model; otherwise, the necessary numerical simulations would take too long. This step relies on the HetSys module PX912 (Multiscale Modelling).