Bayesian Analysis of Interpretable Aging across Thousands of Lithium-ion Battery Cycles
Marc D. Berliner, Minsu Kim, Xiao Cui, Vivek N. Lam, Patrick A. Asinger, Martin Z. Bazant, William C. Chueh, Richard D. Braatz · 2025
Read Bayesian Analysis of Interpretable Aging across Thousands of Lithium-ion Battery Cycles by Marc D. Berliner, Minsu Kim, Xiao Cui, Vivek N. Lam, Patrick A. Asinger, Martin Z. Bazant, William C. Chueh, Richard D. Braatz (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-10439.txt.
Introduction
Marc Berliner and seven co-authors track how key physics parameters age inside lithium-ion cells over thousands of cycles, in a paper from 2025. A standard battery model ties aging to reaction rates and diffusion speeds in each electrode, but whether those can be pinned down uniquely was open. Ninety-five Tesla cells supply the test bed.
Using Bayesian sampling across more than seven thousand diagnostic cycles, they found anode parameters identifiable throughout life. Cathode parameters became identifiable only from mid-life on, which signals growing cathode resistance. A simple power law then linked the key parameters to state of health with high consistency.
Catalog record
| CREATOR | Marc D. Berliner, Minsu Kim, Xiao Cui, Vivek N. Lam, Patrick A. Asinger, Martin Z. Bazant, William C. Chueh, Richard D. Braatz |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-10439
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Bayesian Analysis of Interpretable Aging across Thousands of Lithium-ion Battery Cycles Marc D. Berliner, Minsu Kim, Xiao Cui, Vivek N. Lam, Patrick A. Asinger, Martin Z. Bazant, William C. Chueh, Richard D. Braatz Abstract The Doyle-Fuller-Newman (DFN) model is a common mechanistic model for lithium-ion batteries. The reaction rate constant and diffusivity within the DFN model are key parameters that directly affect the movement of lithium ions, thereby offering explanations for cell aging. This work investigates the ability to uniquely estimate each electrode’s diffusion coefficients and reaction rate constants of 95 Tesla Model 3 cells with a nickel cobalt aluminum oxide (NCA) cathode and silicon oxide–graphite (LiC 6 {}_{\text{6}} –SiO x {}_{\text{x}} ) anode. The parameters are estimated at intermittent diagnostic cycles over the lifetime of each cell. The four parameters are estimated using Markov chain Monte Carlo (MCMC) for uncertainty quantification (UQ) for a total of 7776 cycles at discharge C-rates of C/5, 1C, and 2C. While one or more anode parameters are uniquely identifiable over every cell’s lifetime, cathode parameters become identifiable at mid- to end-of-life, indicating measurable resistive growth in the cathode. The contribution of key parameters to the state of health (SOH) is expressed as a power law. This model for SOH shows a high consistency with the MCMC results performed over the overall lifespan of each cell. Our approach suggests that effective diagnosis of aging can be achieved by predicting the trajectories of the parameters contributing to cell aging. As such, extending our analysis with more physically accurate models building on DFN may lead to more identifiable parameters and further improved aging predictions. 1 1 footnotetext: These authors contributed equally to this work. 1 Introduction Lithium-ion batteries have become ubiquitous in modern technology, including in personal consumer devices and automobiles. Batteries are commonly modeled using porous electrode theory (PET) [ 1 , 2 , 3 , 4 ] , which includes electrochemical kinetics at the solid-electrolyte interfaces in the porous electrodes, transport through the electrolyte and in the solid particles, and thermodynamics modeled by as function of concentration for the open-circuit voltage (OCV). The Doyle-Fuller-Newman (DFN) model (also called the PET model and the Pseudo-Two-Dimensional (P2D) model) has one dimension as the position between the two metal contact points on the opposite sides of the electrode-separator-electrode sandwich and the second dimension as the distance from the center of a solid particle. The common practice is to fit a half dozen effective transport and kinetic coefficients in the DFN model to battery cycling data [ 2 ] , and then to use the model to explore changes in the battery design or the operating conditions. When fitting model parameters to experimental data, an important consideration is whether the data contain sufficient information to uniquely specify the model parameters. Answering this question is referred to as an identifiability analysis . While past studies have employed structural and linearized identifiability analyses to show that multiple combinations of effective transport and kinetic coefficients can produce nearly identical voltage discharge curves, more recent articles take an alternative approach of carrying out a fully nonlinear quantitative identifiability analysis for the DFN model [ 5 , 6 ] . The recent approach provides precise information on the uncertainty of the combinations of the estimated model parameters and makes precise statements as to which of the original model parameters are unlikely to be practically identifiable for commercial lithium-ion batteries. Numerous publications have fit various lithium-ion battery parameters to cycling data using different models and methods. For example, a local linearized sensitivity analysis was applied to separate groups of thermodynamic and kinetic DFN model parameters regressed on low-current discharge curves and current pulses, respectively [ 7 ] . Jokar et al. [ 8 ] implemented a local linearized sensitivity analysis, which was coupled with a genetic algorithm applied to a simplified P2D model to establish the time periods where parameters greatly affect a discharge voltage curve under high and low C-rates. Another study [ 9 ] applied linearized local sensitivities and a Monte Carlo (MC)-based covariance analysis to study identifiability from discharge curves and the electrolyte concentration in the center of the separator. A fit of a hundred DFN model parameters to experimental battery cycling data using a genetic algorithm [ 10 ] found that only a small subset of the parameters was identifiable. The identifiability of parameters in the single-particle (SP) model with electrolyte dynamics under a constant state of charge (SOC), represented in terms of probability density functions (PDFs) of model parameters, has been quantified by Markov chain Monte Carlo (MCMC) [ 11 ] . Unidentifiability in the solid-state diffusion parameters was alleviated by applying a sinusoidal pulse to achieve a range of SOC. At low C-rates, the SP model has a comparable error to the P2D model [ 12 ] . The MCMC method has also been applied to quantify uncertainties in five effective transport and kinetic coefficients in the DFN model [ 6 , 13 ] . Of these five parameters, only the anode solid diffusion coefficient was found to be globally identifiable [ 6 ] . Reductions in the effective transport and kinetic coefficients over the cycle life of a Li-ion battery were plotted and shown to follow a power law [ 13 ] . The approach predicted voltage discharge profiles at future cycles from experimental data collected for the first 200 cycles. Our previous work [ 6 , 5 ] applied a nonlinear identifiability analysis to electrochemical battery models based on PET and multiphase PET (MPET) [ 14 ] , respectively. Ref. [ 6 ] estimated five diffusion and kinetic coefficients from synthetically generated data of a single discharge curve. Here we extend this work to a dataset with silicon oxide–graphite/nickel cobalt aluminium lithium-ion (NCA/LiC 6 –SiO {}_{\text{x}} ) cells from a disassembled Tesla Model 3. The a posteriori parameter distributions are estimated at every diagnostic cycle using MCMC. First, MCMC methods are presented and shown to provide global nonlinear identifiability trends for a small subject of parameters. Second, unknown physical properties of the NCA/LiC 6 –SiO {}_{\text{x}} cell, which are estimated experimentally by inverting the model, are discussed. Third, the model specifications are established for the standardized cycling conditions of the fleet of NCA cells. Fourth, results are presented for the identifiability analysis and Bayesian estimation of each cycle with simple parameter fittings as a function of the State of Health (SOH). This article is organized as follows: Section 2 provides theoretical background on the DFN model and Bayesian parameter identification. Section 3 describes the parameterization for explaining the cycling behavior of the NCA cell. Section 4 presents the overall scheme for parameter identification and provides details on the NCA cell data. Section 5 represents the main results, which consist of identifiability and degradation diagnosis, and Section 6 summarizes the article. 2 Background 2.1 DFN model The DFN model describes the microscopic physicochemical behavior of lithium-ion batteries and has been successfully applied to fast charging strategies [ 15 , 16 ] , aging analysis [ 17 , 18 ] , lifetime prediction [ 19 ] , and fault diagnosis [ 20 ] . Each porous electrode has an electrically conductive solid phase in close contact with a