Uncertainty-Aware and Explainable Machine Learning for Early Prediction of Battery Cell Degradation

We want to develop batteries with long cycle life. Economics of energy storage improves with long lasting batteries whether it is for electric vehicles or grid scale storage. The process of designing high cycle life batteries or developing of battery management schemes to minimize degradation is very time consuming. One has to perform many cycling experiments for the full cell lifetime to study the effects of different design parameters or the effect of charging management. Battery degradation consists of many highly complex mechano-electro-chemical processes [1], for which we cannot just do physics based simulations for lifetime prediction but reply on actual experiments.<br/>We demonstrate that an uncertainty-aware deep autoregressive model (LSTM recurrent neural network) can be trained with very limited data to robustly model the capacity degradation over the entire lifetime, outperforming previous approaches [2] in accuracy for EOL (end of life) prediction. Our model is the first that is able to model the entire capacity fade trajectory from the early cycles without a fixed limit on the maximum lifetime while maintaining uncertainty awareness and explainability. With the model predicting uncertainty, we can also provide an uncertainty estimate over the trajectory as well as for the EOL, allowing appropriate and reliable deployment. An explainability analysis [3] of the proposed recurrent neural network model provides cognizance of the interplay between degradation mechanisms across multiple heterogeneous cells. The model aligns with existing chemical insights into the rationale for early EOL despite not being trained for this or having been provided prior chemical knowledge. We have intentionally developed the model based on information that is universally available from all types of battery cycling data and chemistry agnostic. Thus we foresee our model to generalize to all battery chemistries and formats.<br/>References<br/>1. S. Edge, J. et al. Lithium ion battery degradation: what you need to know. Physical Chemistry Chemical Physics 23, 8200–8221 (2021).<br/>2. Severson, K. A. et al. Data-driven prediction of battery cycle life before capacity degradation. Nature Energy 4, 383–391 (2019).<br/>3. Samek, W., Montavon, G., Lapuschkin, S., Anders, C. J. & Müller, K.-R. Explaining Deep Neural Networks and Beyond: A Review of Methods and Applications. Proceedings of the IEEE 109, 247–278 (2021).