There is a great need for non-invasive methods to unravel the underlying causes of battery ageing. From the perspective of electrochemical models, battery ageing affects a number of parameters, but conventional parametrization of electrodes is no longer possible. This is because the re-parametrisation would require the fabrication of an aged half-cell. Although ageing models can explain and map some degradation mechanisms, their application in electric vehicles is an ongoing challenge.
This work presents a comparative study on a subset of electrochemical parameters of pristine, calendar aged and cycled aged Li-ion batteries. The goal is to identify the main degradation modes via differential voltage analysis and track changes throughout battery lifetime via electrochemical model parameter optimisation. For this purpose, an isothermal single particle model (SPM) is simulated in COMSOL Multiphysics® with MATLAB® via LiveLink(TM).
The estimated parameters affected by ageing include: diffusion time, stoichiometry, active material volume fraction and kinetic rate constant. It is observed that the diffusion time increases as the battery ages, resulting in a doubling of time needed to reach thermodynamic equilibrium following full charge. Both the model and experiments were conducted using current profiles derived from the Artemis motorway drive cycle. Parameters were adjusted to minimise the root mean squared error (RMSE) between the measurements and simulations. Although model accuracy decreases with battery life, it remains at an acceptable level (RMSE 14 – 32 mV). Some inaccuracy is due to cathode degradation, which is not accounted for in the model. To avoid identifiability problems only a specific subset of parameters can be optimised at a time.
As expected, in case of calendar aged cell which experienced loss of lithium inventory (LLI), the stoichiometric limit on the anode decreased. Also, the kinetic rate decreased, suggesting an increase in resistance, most likely caused by the SEI growth. The cycle aged cell experienced a decrease in the active material volume fraction corresponding to loss of active material in the anode and a stoichiometric shift due to LLI. Both diffusion and kinetics slowed down with cyclic age.
In conclusion, the SPM voltage prediction accuracy can be maintained throughout battery lifetime within good limits (14- 32mV) if parameters are updated accordingly. If the model is operated without parameter optimisation at the end of life, the RMSE would increase to 180 mV.