Life Prediction and Modelling

Integrating AI Into Battery Performance Models For Accurate Extrapolation From Limited Data Identifying models that simultaneously predict battery performance (i.e., predicting battery response versus SOC, temperature, and C-rate) and degradation (change in performance over battery life) is hugely challenging. Empirical models of performance and lifetime, including equivalent circuit models, are easier to fit but extrapolate poorly to new operating conditions.

Zhou et al. 1] demonstrate augmenting of an extremely basic equivalent circuit model with Gaussian process learned states for performance and health parameters. The learned model is able to compensate for OCV changes over lifetime.

Kuzhiyila et al. 2] demonstrate using a Universal Differential Equation UDE) to augment a physics-based model of SEI growth to enable the simple model to predict SEI growth for a NMC811|Gr-Si cell.

Physics-based models extrapolate well to new operating conditions, but it is extremely challenging to define differential equations that accurately describe the changes in battery performance observed during testing. Integrating AI into battery performance models mitigates this downside by learning from available testing data to compensate for model errors, and shifts lifetime forecasts from deterministic to probabilistic. Source: 1 Zhou et al arXiv 2407.06639v2, 2 Ayyangatu Kuzhiyil et al SSRN 4948229

Extracted from the PDF; charts and images may read imperfectly.