Physics-Based Inverse Modeling of Battery Degradation with Bayesian Methods

Area of Science:

• Materials Science
• Electrochemistry
• Computational Science

Background:

• Improving lithium-ion battery performance requires a deep understanding of complex internal processes.
• Physical models are essential for insights but face challenges in validation and parameterization.
• Degradation mechanisms, such as solid-electrolyte interphase (SEI) growth, limit battery lifetime.

Purpose of the Study:

• To apply advanced Bayesian machine learning methods for accurate parameterization of battery degradation models.
• To investigate the efficacy of Expectation Propagation + Bayesian Optimization for Likelihood-Free Inference (EP-BOLFI) for modeling SEI growth.
• To confirm the best theoretical model for SEI growth during battery storage using Bayesian Alternately Subsampled Quadrature (BASQ).

Main Methods:

• Utilized Expectation Propagation + Bayesian Optimization for Likelihood-Free Inference (EP-BOLFI) to parameterize SEI growth models.
• Incorporated human expertise through feature selection to enhance model parameterization.
• Employed Bayesian Alternately Subsampled Quadrature (BASQ) to calculate model probabilities and compare theoretical models.

Main Results:

• EP-BOLFI successfully parameterized SEI growth models using both synthetic and real battery degradation data.
• Achieved accurate parameterization with uncertainty quantification even under challenging conditions, outperforming standard Markov Chain Monte Carlo methods.
• Identified electron diffusion as the most suitable model for describing SEI growth during battery storage, validated by BASQ.

Conclusions:

• Bayesian inference methods provide a robust framework for validating and parameterizing complex battery models.
• EP-BOLFI offers an efficient and effective approach for understanding battery degradation and improving lifetime predictions.
• The study confirms electron diffusion as the leading model for SEI growth, advancing battery storage science.