Główne pojęcia
This research leverages Bayesian machine learning algorithms to analyze and parameterize physics-based models of battery degradation, specifically focusing on solid-electrolyte interphase (SEI) growth mechanisms, demonstrating the effectiveness of this approach in identifying dominant degradation processes and quantifying uncertainties.
Streszczenie
Bibliographic Information:
Philipp, M. C. J., Kuhn, Y., Latz, A., & Horstmann, B. (2024). Physics-based inverse modeling of battery degradation with Bayesian methods. arXiv preprint arXiv:2410.19478.
Research Objective:
This study aims to determine the dominant transport mechanisms responsible for continuous SEI growth in lithium-ion batteries and accurately parameterize corresponding physics-based degradation models using Bayesian machine learning methods.
Methodology:
The researchers employed a physics-informed battery model (Single Particle Model with electrolyte effects - SPMe) coupled with theoretical SEI growth mechanisms (electron diffusion, electron conduction, electron migration, and solvent diffusion). They utilized Bayesian optimization (EP-BOLFI) to inversely model synthetic and real storage data, quantifying parameter uncertainties and correlations. Additionally, they applied Bayesian model selection (BASQ) to identify the most probable SEI growth mechanism based on model evidence.
Key Findings:
- EP-BOLFI successfully parameterized the SEI growth models, accurately capturing the SoC-dependent capacity loss trends in both synthetic and real storage data.
- Parameter correlations obtained from EP-BOLFI provided insights into the interplay between different SEI growth mechanisms.
- BASQ identified electron diffusion as the most likely mechanism for SoC-dependent SEI growth during storage, supporting previous experimental findings.
- The study demonstrated the effectiveness of feature selection in EP-BOLFI for improving parameter identification, particularly under noisy conditions.
Main Conclusions:
This research highlights the power of Bayesian machine learning for analyzing complex battery degradation processes. By combining physics-based models with advanced statistical methods, the study provides a robust framework for identifying dominant degradation mechanisms, quantifying uncertainties, and ultimately guiding the development of longer-lasting batteries.
Significance:
This work contributes significantly to the field of battery degradation modeling by demonstrating a practical and efficient approach for analyzing complex, coupled physico-chemical processes. The findings have implications for battery management strategies and the development of more accurate lifetime prediction models.
Limitations and Future Research:
The study primarily focused on SEI growth during storage. Future research should investigate the interplay of multiple degradation mechanisms during cycling and explore the applicability of these methods to other battery chemistries and aging processes.
Statystyki
The study analyzed storage data of lithium-ion batteries with a nickel-cobalt-aluminium oxide (NCA) cathode, stored for 9.5 months with check-ups every two months.
Synthetic cycling data was generated for 500 full cycles with 1C CC-CV charge and 1C constant discharge, with voltage cut-offs of 2.5V and 4.2V.
The study investigated different noise levels applied to the synthetic cycling data, with variances of 8 · 10−6Ah2 and 8 · 10−5Ah2.
Cytaty
"The increasing capabilities of machine learning (ML) algorithms point to a possible way to tackle this problem."
"The most natural incorporation of uncertainty is achieved by Bayesian algorithms, favoring them over other ML algorithms for these purposes."
"Despite decades-long research efforts, the scientific community cannot fully explain the SEI growth, suggesting that the SEI is complex and possibly based on multiple coupled mechanisms."
"BASQ identifies electron diffusion as the best transport mechanism to describe the SoC-dependent characteristics in this storage data."