Pith. sign in

REVIEW 3 major objections 4 minor 46 references

Embedding an Arrhenius capacity-fade law directly into the neural network forward pass gives the lowest battery health prediction error and avoids non-physical capacity regeneration in extrapolation.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 13:44 UTC pith:KRQLVYLA

load-bearing objection Solid engineering paper but the physics attribution is unproven — needs a control experiment before the 'physics-informed' label carries weight. the 3 major comments →

arxiv 2607.29095 v1 pith:KRQLVYLA submitted 2026-07-31 cs.LG

PiDDM: Physics-Informed Differentiable Degradation Modeling for Lithium-Ion Battery State-of-Health Prediction

Zeping Chen , Ruda Jian , Sachin Sigdel , Guoping Xiong , Jian-Xun Wang , Tengfei Luo This is my paper
classification cs.LG
keywords state-of-health predictionlithium-ion battery degradationphysics-informed neural networkdifferentiable degradation modelArrhenius kineticssolid electrolyte interphasecapacity extrapolationbattery lifetime forecasting
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tries to establish that battery state of health can be predicted more accurately and more physically when an empirical Arrhenius–power-law degradation model—capacity loss from solid-electrolyte-interphase growth and loss of lithium inventory—is hard-wired into the neural network as a differentiable rate module rather than applied as a soft penalty or left out. On a public set of 55 lithium-ion cells cycled under six protocols, this framework (PiDDM) averages the lowest error on all four metrics in ordinary prediction, and in a 90/10 extrapolation test it predicts the accelerating end-of-life knee while avoiding the upward capacity spikes that a plain multilayer perceptron and a published baseline physics-informed network exhibit. The authors' argument is that prescribing the degradation-rate structure acts as an inductive bias that improves cross-protocol generalization; they explicitly treat the learned coefficients as effective, data-calibrated parameters rather than a complete electrochemistry model. They also acknowledge a tradeoff: for the R2.5 protocol, where measured capacity contains upward fluctuations, PiDDM's absolute error is slightly higher because it does not chase those localized recoveries.

Core claim

The central claim is that Eq. (1), dQ/dt = -k_SEI exp(-E_SEI/RT) t^alpha - k_LLI exp(-E_LLI/RT) DoD^b1 Ah^b2, should be embedded as a differentiable degradation module in the forward pass. At each cycle, a fully connected network maps 16 routine charging features to five effective kinetic parameters, two global pre-exponential factors are learned, and forward-Euler integration of the rate yields the predicted capacity trajectory. On the 55-battery, six-protocol dataset, PiDDM achieves average MSE of 6.63e-5 versus 1.77e-4 for the baseline PINN and 2.42e-4 for the MLP in standard prediction; in the 90/10 temporal extrapolation setting the average MSEs are 1.11e-4, 2.44e-4, and 8.26e-4 respect

What carries the argument

The load-bearing mechanism is the degradation-rate module: the two-term Arrhenius–power-law capacity-fade relation (Eq. 1) composed with a first-order forward-Euler integrator (Eqs. 2-3) that converts per-cycle rates into a capacity trajectory. The network supplies the five effective kinetic parameters per cycle, while the pre-exponential factors are global trainable scalars. Because the forward pass is exactly this integrator, the SOH trajectory inherits the monotonic, thermally activated shape of the rate law, and because the integrator is differentiable the whole system is trained end-to-end with gradient descent. The contrast to the baseline PINN, which learns its degradation operator as

Load-bearing premise

The entire advantage rests on the premise that the specific Arrhenius–power-law form of Eq. (1), rather than merely the smooth monotone integrated-trajectory architecture, is what produces the accuracy and extrapolation gains.

What would settle it

A control experiment with the same forward-Euler cumulative-sum architecture and network capacity, but with the Arrhenius/power-law terms replaced by a flexible sign-constrained per-cycle rate function, could settle it: if the 90/10 extrapolation MSE stays near PiDDM's, the physics form is not the cause. Reporting the fitted activation energies would also test it—values far outside documented SEI/LLI ranges would undermine the mechanistic interpretation.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • In standard prediction, PiDDM's average MSE (6.63e-5) is about 3.7 times lower than the MLP's (2.42e-4) and about 2.7 times lower than the baseline PINN's (1.77e-4).
  • In the 90/10 extrapolation setting, PiDDM's average MSE (1.11e-4) is roughly 2.2 times lower than the baseline PINN's and 7.5 times lower than the MLP's, while producing smoother, nearly monotonic trajectories.
  • The gain is a genuine forecast property: integration starts once from the first measured capacity and never re-reads actual capacity in the held-out region, so the extrapolated values are generated recursively from predicted rates.
  • The benefit is not uniform: the MLP achieves lower pointwise error on R3 in extrapolation, and the baseline PINN has lower MAE/MAPE on R2.5 in standard prediction, because PiDDM trades noise-following for smoothness and physical consistency.
  • If the approach transfers, the same differentiable rate-module pattern could be applied to other degradation indicators, such as internal resistance growth, which the authors name as future work.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The paper never reports the learned values of k_SEI, k_LLI, E_SEI, E_LLI, alpha, b1, and b2, so the specific attribution to SEI and LLI remains unvalidated; the same accuracy gains might come from any smooth monotone rate law.
  • A structure-free control is absent: a version of PiDDM with the same Euler integration but a learned, sign-constrained per-cycle rate function would reveal whether the Arrhenius form or the integrative architecture drives the improvement.
  • Monotonicity is not guaranteed by construction, since no positivity constraint on the pre-exponential factors or network outputs is described; the observed smoothness may depend on training outcomes and would benefit from an explicit constraint.
  • The R2.5/R3 tradeoff points to a missing quantitative metric for smoothness and directional consistency; without one, practitioners cannot systematically choose between pointwise accuracy and physical plausibility.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes PiDDM, a physics-informed differentiable framework for lithium-ion battery state-of-health (SOH) prediction. The method embeds a semi-empirical Arrhenius–power-law degradation rate, dQ/dt = -k_SEI exp(-E_SEI/RT) t^α - k_LLI exp(-E_LLI/RT) DoD^b1 Ah^b2 (Eq. 1), into the forward pass. A neural network maps 16 charging features to five cycle-dependent kinetic parameters, which are combined with two global pre-exponential factors and known operating variables in a differentiable module; forward-Euler integration (Eqs. 2–3) produces the predicted capacity trajectory. The model is evaluated on the public Wang et al. dataset (55 NCM batteries, six protocols) and compared with an MLP and a baseline PINN. The authors report lower average MSE in both standard prediction (6.63e-5 vs 1.77e-4 and 2.42e-4) and 90/10 temporal extrapolation (1.11e-4 vs 2.44e-4 and 8.26e-4), while also claiming smoother, physically consistent monotone trajectories without the capacity regeneration observed in baselines. The paper discloses protocol-level exceptions (R2.5 MAPE/MAE, R3 extrapolation) and discusses the accuracy-versus-physical-consistency trade-off.

Significance. If the claimed improvements hold, PiDDM would offer a practical hybrid approach that combines a simple mechanistic prior with data-driven flexibility, with tangible benefits for cross-protocol generalization and end-of-life extrapolation. The paper has clear strengths: the forward-Euler integration is transparent and correct; the cumulative-sum formulation (Eq. 3) is simple and reproducible; the evaluation uses a public multi-protocol dataset; and the authors explicitly report the protocol-level exceptions rather than only favorable averages. However, the paper's central attribution — that the specific Arrhenius–power-law physics in Eq. (1) causes the gains — is not yet supported by a free-rate integrator control or any validation of the learned kinetic parameters. In addition, the claimed monotonicity is not guaranteed by the stated architecture because no positivity constraints are described for the trainable parameters. The absence of multiple-seed runs and confidence intervals further weakens the quantitative comparison. These issues are fixable but are load-bearing for the paper's main claims.

major comments (3)
  1. [§2.3, Eqs. (1)–(3)] The central qualitative claim of 'no capacity regeneration' is not guaranteed by the architecture as described. The negative sign in Eq. (1) only ensures dQ/dt ≤ 0 if k_SEI, k_LLI, both exponentials, t^α, DoD^b1, and Ah^b2 are all non-negative. The five kinetic parameters θ̂_n are outputs of an unconstrained neural network (no activation function is specified in §2.3), and κ are global trainable scalars with no described positivity constraint. Nothing prevents the optimizer from driving, e.g., k_SEI or α negative, which would produce an upward increment in Eq. (2). Please either impose explicit non-negativity constraints (e.g., softplus) on κ and the relevant outputs, or state clearly that monotonicity is a soft property that can be violated.
  2. [§3.1, Tables 3–4] The performance comparison rests entirely on point estimates from a single training run. No seed variation, confidence intervals, or statistical significance tests are reported. Given the disclosed exceptions (R2.5 MAPE/MAE and R3 extrapolation), the 'lowest average error' could plausibly be within run-to-run variability. Please report results over at least 3–5 random seeds (different network initializations and data shuffles) as mean ± std, and where possible a paired test across batteries. This is necessary to support the central accuracy claim.
  3. [§2.3, Eq. (3); §3.1] The attribution of the gains to the specific Arrhenius–power-law physics is untested. Because Eq. (3) is a cumulative sum of per-cycle rates, the architecture can represent any monotone trajectory by appropriately choosing the five per-cycle parameters; the functional form of Eq. (1) is not a binding constraint unless the parameter ranges are tied to known physics or a flexible-rate control is run. The paper does not report the learned k_SEI, k_LLI, E_SEI, E_LLI, α, b1, b2 for any protocol, and no comparison with literature SEI/LLI activation energies is given (the authors themselves state in §2.1 that coefficients should be interpreted as effective data-calibrated parameters). A minimal control — a model with the same Euler/cumulative-sum structure but a free per-cycle rate, e.g., a neural network directly outputting a signed or constrained rate — is needed to determine whether the phys
minor comments (4)
  1. [§2.4.5, Eq. (14)] The normalization expression writes a general interval [a,b] but the formula shown maps to [-1,1]. Clarify the notation or remove the unspecified interval.
  2. [§2.3] The input feature vector x_n is said to contain 16 statistical features, but the feature extraction from charging segments is only referenced to earlier papers. A brief description or table of the feature definitions would improve reproducibility.
  3. [Data and code availability] The statement 'Data and code are available upon request' is weaker than a permanent repository link. Given the emphasis on reproducibility, a DOI or public archive would be preferable.
  4. [§3.2] The text describing the R3 exception is clear, but Figure 5's trajectories are only qualitatively discussed. Adding a per-battery error table or a boxplot of per-battery errors for the extrapolation setting would help quantify the variability of the reported means.

Circularity Check

1 steps flagged

Quantitative benchmark gains are independent; only the no-regeneration/monotonicity claim is definitional.

specific steps
  1. self definitional [Abstract & §3.1, compared with construction in §2.1 Eq. (1) and §2.3 Eqs. (2)-(3)]
    "The negative sign in Eq. (1) enforces the physically expected direction of capacity loss... Because PiDDM is regularized by degradation physics that enforces non-increasing SOH trajectories, it does not track these local upward variations as closely as the baseline PINN."

    The paper's qualitative claim that PiDDM 'avoid[s] the non-physical capacity regeneration produced by the baseline models' is a direct consequence of the model construction: Eq. (3) is a cumulative sum of per-cycle rates, and Eq. (1) assigns those rates an explicit negative sign. With positive k_SEI/k_LLI and positive exponential/power-law terms, every increment is non-positive, so the SOH trajectory cannot increase. The paper itself describes the model as 'enforc[ing] non-increasing SOH trajectories.' Thus the 'no regeneration' result is not an independent empirical finding about the physics; it is restating the sign convention of the forward model. The quantitative accuracy gains, however, come from held-out-battery error comparisons and do not reduce to this construction.

full rationale

The paper's central quantitative claims are empirical benchmark results: PiDDM achieves lower average MSE than an MLP and a baseline PINN on held-out batteries, both in standard prediction and 90/10 temporal extrapolation. Those comparisons do not reduce to the model equations or to any fitted parameter renamed as a prediction. The Arrhenius–power-law form of Eq. (1) is explicitly presented as a semi-empirical prior rather than a unique or complete degradation model, so no uniqueness theorem is being imported. Citations to Wang et al. [27] are to an external baseline method, and the few self-citations ([16], [25], [26]) are background references, not load-bearing for the central derivation. The one definitional element is the monotonicity/no-regeneration claim: because the capacity trajectory is produced by the cumulative sum in Eq. (3) of a rate that Eq. (1) writes with a leading negative sign, the trajectory is non-increasing by construction (for positive coefficient values), as the paper itself acknowledges by saying the model 'enforces non-increasing SOH trajectories.' Reporting the absence of capacity regeneration as a physics-informed advantage is therefore partly restating the model definition rather than an independent validation of the physics. This does not undermine the quantitative comparisons, but it means the physics attribution for the monotonicity benefit is not independently tested by a sign-constrained integrator control. Overall, a minor self-definitional step, with the central predictive claims retaining independent empirical content.

Axiom & Free-Parameter Ledger

2 free parameters · 6 axioms · 0 invented entities

PiDDM's physics content is Eq. (1): two global pre-exponential constants and five per-cycle exponents/activation energies, all fitted to the same capacity data being predicted. The monotone-decreasing direction relies on an unflagged positivity assumption on kappa and the NN outputs. The dataset split, features, and Euler scheme are standard choices. No new physical entities are introduced; the 'effective kinetic parameters' are fitted quantities, not independently evidenced objects.

free parameters (2)
  • k_SEI, k_LLI (global pre-exponential factors kappa) = not reported
    Global trainable scalars in Eq. (1), optimized jointly with the network (§2.3); fitted to capacity data, values never disclosed.
  • Per-cycle kinetic parameters theta_n = (E_SEI,n, alpha_n, E_LLI,n, b1,n, b2,n) = 5 values per cycle, not reported
    Neural-network output per cycle from the 16 charging features (§2.3). The 'physics' coefficients are learned quantities — effectively five fitted parameters per cycle — not independently measured kinetic constants.
axioms (6)
  • domain assumption Capacity-fade rate decomposes additively into SEI (Arrhenius * t^alpha) and LLI (Arrhenius * DoD^b1 * Ah^b2) terms (Eq. 1)
    The central inductive bias; motivated by cited semi-empirical aging models [37,41-45], but the exact additive functional form is a modeling choice for this paper (§2.1). The paper itself states it is 'not claimed to be a unique or complete description of battery aging'.
  • domain assumption Per-cycle degradation state is determined by the 16 charging features via a fixed NN mapping (x_n -> theta_n)
    Assumes the 16 CC/CV charging statistics fully determine the current degradation kinetics well enough to transfer across six protocols (§2.3).
  • domain assumption The 90/10 temporal split isolates extrapolation; the early-life fit governs late-life behavior
    The extrapolation result depends on the chosen split and on the assumption that degradation continues to follow the same functional family beyond the training horizon (§2.4.6, §3.2).
  • ad hoc to paper Sign structure dQ/dt <= 0 holds (positivity of kappa and the exponential/power-law terms)
    Monotonicity claims in the Abstract and §3.2 require kappa > 0; no positivity constraint on kappa or the NN outputs is described, and Adam is unconstrained, so the claim rests on an unflagged assumption (§2.3). Figure 5 shows PiDDM is only 'more nearly monotonic', not strictly.
  • standard math Forward-Euler with Delta n = 1 cycle is an adequate integrator
    Eq. (2) applies one Euler step per observed cycle; no convergence or accuracy analysis is given, but the choice is reasonable for cycle-indexed data.
  • domain assumption Per-cycle temperature T is available from the operating protocol
    T enters the Arrhenius factors; the paper says T and DoD are 'obtained from the operating protocol' (§2.3) but does not document how per-cycle temperature is extracted from the dataset.

pith-pipeline@v1.3.0-daily-deepseek · 14043 in / 25583 out tokens · 231528 ms · 2026-08-03T13:44:44.112679+00:00 · methodology

0 comments
read the original abstract

Accurate prediction of lithium-ion battery state of health (SOH) is essential for reliable energy storage operation. However, purely data-driven models may generalize poorly across cycling protocols and produce physically implausible behavior during long-term extrapolation. We developed a physics-informed differentiable degradation modeling framework (PiDDM) for battery SOH prediction. PiDDM incorporates empirical Arrhenius degradation kinetics associated with solid electrolyte interphase growth and loss of lithium inventory into the training objective, encouraging physically consistent capacity fade under diverse operating conditions. The framework was evaluated using a public dataset of 55 batteries cycled under six operating protocols. PiDDM achieved the lowest average prediction error among the evaluated models and substantially reduced mean squared error relative to a multilayer perceptron and a baseline physics-informed neural network. For extrapolation, the models were trained on the first 90% of each battery's cycle life and evaluated on the unseen final 10%. PiDDM captured accelerated end-of-life degradation while avoiding the nonphysical capacity regeneration produced by the baseline models. These results show that incorporating degradation physics into neural network training improves predictive accuracy and physical consistency, providing a promising approach for practical battery health monitoring.

Figures

Figures reproduced from arXiv: 2607.29095 by Guoping Xiong, Jian-Xun Wang, Ruda Jian, Sachin Sigdel, Tengfei Luo, Zeping Chen.

Figure 1
Figure 1. Figure 1: Overview of the Physics-informed Differentiable degradation modeling (PiDDM) framework for battery state-of￾health (SOH) prediction. Charging and discharging protocol information is processed by the PiDDM, which combines data￾driven learning with empirical degradation kinetics to predict physically consistent SOH trajectories over the battery lifetime. 2.3. PiDDM Framework Purely data-driven neural network… view at source ↗
Figure 2
Figure 2. Figure 2: Physics-informed Differentiable degradation modeling (PiDDM). (a) Empirical capacity-fade relation containing SEI-associated and LLI-associated degradation contributions. (b) The neural network maps the 16-dimensional charging￾feature vector xn to five effective, cycle-dependent kinetic parameters θb n. These parameters are combined with the global trainable pre-exponential factors κ and the known operatin… view at source ↗
Figure 3
Figure 3. Figure 3: Predictive performance of PiDDM across diverse protocols. SOH predictions from PiDDM (green) are compared with those of MLP (blue) and PINN (red) across six cycling protocols: constant current (2C, 3C), randomized discharge (R2.5, R3), random walk (RW), and satellite missions. still shows visible deviations from the ground truth and does not always capture the onset of the end-of￾life degradation knee accu… view at source ↗
Figure 4
Figure 4. Figure 4: Extrapolation evaluation protocol for SOH prediction. Observed degradation data (gray points) are used for training, while the extrapolation region corresponds to the unseen portion of the SOH trajectory that must be predicted by the model. Under the extrapolation setup described above, the models are trained using only the initial 90% of each available training trajectory and are required to predict beyon… view at source ↗
Figure 5
Figure 5. Figure 5: Temporal extrapolation performance across diverse cycling protocols. SOH predictions from MLP, PINN, and PiDDM are compared across six protocols: varying C-rates (2C and 3C), resistance-based aging conditions (R2.5 and R3), and dynamic load profiles (RW and Satellite). Models are trained on the initial 90% of the cycle life and evaluated on the remaining unseen 10%. according to the four reported error met… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

46 extracted references · 19 canonical work pages

  1. [1]

    Tarascon, M

    J.-M. Tarascon, M. Armand, Issues and challenges facing rechargeable lithium batteries, Nature 414 (2001) 359–367.doi:10.1038/35104644. 23

    doi:10.1038/35104644 2001

  2. [2]

    Armand, J.-M

    M. Armand, J.-M. Tarascon, Building better batteries, Nature 451 (2008) 652–657.doi:10.1038/ 451652a

    2008

  3. [3]

    B. Dunn, H. Kamath, J.-M. Tarascon, Electrical energy storage for the grid: A battery of choices, Science 334 (6058) (2011) 928–935.doi:10.1126/science.1212741

    work page doi:10.1126/science.1212741 2011

  4. [4]

    J. B. Goodenough, K.-S. Park, The li-ion rechargeable battery: A perspective, Journal of the American Chemical Society 135 (4) (2013) 1167–1176.doi:10.1021/ja3091438

    doi:10.1021/ja3091438 2013

  5. [5]

    Nitta, F

    N. Nitta, F . Wu, J. T . Lee, G. Yushin, Li-ion battery materials: Present and future, Materials Today 18 (5) (2015) 252–264.doi:10.1016/j.mattod.2014.10.040

    doi:10.1016/j.mattod.2014.10.040 2015

  6. [6]

    URLhttps://www.iea.org/reports/global-ev-outlook-2024/trends-in-electric-cars

    International Energy Agency, Global ev outlook 2024: Trends in electric cars, accessed 2026-04-29 (2024). URLhttps://www.iea.org/reports/global-ev-outlook-2024/trends-in-electric-cars

    2024

  7. [7]

    URLhttps://www.iea.org/reports/global-ev-outlook-2024/trends-in-electric-vehicle-batteries

    International Energy Agency, Global ev outlook 2024: Trends in electric vehicle batteries, accessed 2026-04-29 (2024). URLhttps://www.iea.org/reports/global-ev-outlook-2024/trends-in-electric-vehicle-batteries

    2024

  8. [8]

    Vetter, P

    J. Vetter, P . Novák, M. R. Wagner, C. Veit, K.-C. Möller, J. O. Besenhard, M. Winter, M. Wohlfahrt- Mehrens, C. Vogler, A. Hammouche, Ageing mechanisms in lithium-ion batteries, Journal of Power Sources 147 (1–2) (2005) 269–281.doi:10.1016/j.jpowsour.2005.01.006

    doi:10.1016/j.jpowsour.2005.01.006 2005

  9. [9]

    Barré, B

    A. Barré, B. Deguilhem, S. Grolleau, M. Gérard, F . Suard, D. Riu, A review on lithium-ion battery age- ing mechanisms and estimations for automotive applications, Journal of Power Sources 241 (2013) 680–689.doi:10.1016/j.jpowsour.2013.05.040

    work page doi:10.1016/j.jpowsour.2013.05.040 2013

  10. [10]

    C. R. Birkl, M. R. Roberts, E. McTurk, P . G. Bruce, D. A. Howey, Degradation diagnostics for lithium ion cells, Journal of Power Sources 341 (2017) 373–386.doi:10.1016/j.jpowsour.2016.12.011

    doi:10.1016/j.jpowsour.2016.12.011 2017

  11. [11]

    J. S. Edge, S. O’Kane, R. Prosser, N. D. Kirkaldy, A. N. Patel, A. Hales, A. Ghosh, W . Ai, J. Chen, J. Yang, S. Li, M.-C. Pang, L. Bravo Diaz, A. Tomaszewska, M. W . Marzook, K. N. Radhakrishnan, H. Wang, Y. Patel, B. Wu, G. J. Offer, Lithium ion battery degradation: What you need to know, Physical Chem- istry Chemical Physics 23 (2021) 8200–8221.doi:10....

    doi:10.1039/d1cp00359c 2021

  12. [12]

    S. E. J. O’Kane, W . Ai, G. Madabattula, D. Alonso-Alvarez, R. Timms, V . Sulzer, J. S. Edge, G. J. Offer, Lithium-ion battery degradation: How to model it, Physical Chemistry Chemical Physics 24 (2022) 7909–7922.doi:10.1039/D2CP00417H

    doi:10.1039/d2cp00417h 2022

  13. [13]

    Newman, W

    J. Newman, W . Tiedemann, Porous-electrode theory with battery applications, AIChE Journal 21 (1) (1975) 25–41.doi:10.1002/aic.690210103

    doi:10.1002/aic.690210103 1975

  14. [14]

    Doyle, T

    M. Doyle, T . F . Fuller, J. Newman, Modeling of galvanostatic charge and discharge of the lithi- um/polymer/insertion cell, Journal of The Electrochemical Society 140 (6) (1993) 1526–1533.doi: 10.1149/1.2221597

    doi:10.1149/1.2221597 1993

  15. [15]

    T . F . Fuller, M. Doyle, J. Newman, Simulation and optimization of the dual lithium ion insertion cell, Journal of The Electrochemical Society 141 (1) (1994) 1–10.doi:10.1149/1.2054684

    doi:10.1149/1.2054684 1994

  16. [16]

    Z. Chen, D. L. Danilov, R.-A. Eichel, P . H. L. Notten, Porous electrode modeling and its applica- tions to li-ion batteries, Advanced Energy Materials 12 (32) (2022) 2201506.doi:10.1002/aenm. 202201506

    doi:10.1002/aenm 2022

  17. [17]

    H. Yu, H. Zhang, Z. Zhang, S. Yang, State estimation of lithium-ion batteries via physics-machine learning combined methods: A methodological review and future perspectives, eTransportation 24 (2025) 100420.doi:10.1016/j.etran.2025.100420

    arXiv 2025

  18. [18]

    Sulzer, S

    V . Sulzer, S. G. Marquis, R. Timms, M. Robinson, S. J. Chapman, Python Battery Mathematical Mod- elling (PyBaMM), Journal of Open Research Software 9 (1) (2021) 14.doi:10.5334/jors.309

    doi:10.5334/jors.309 2021

  19. [19]

    G. L. Plett, Extended kalman filtering for battery management systems of lipb-based hev battery packs: Part 3. state and parameter estimation, Journal of Power Sources 134 (2) (2004) 277–292. doi:10.1016/j.jpowsour.2004.02.033

    work page doi:10.1016/j.jpowsour.2004.02.033 2004

  20. [20]

    K. A. Severson, P . M. Attia, N. Jin, N. Perkins, B. Jiang, Z. Yang, M. H. Chen, M. Aykol, P . K. Her- ring, D. Fraggedakis, M. Z. Bazant, S. J. Harris, W . C. Chueh, R. D. Braatz, Data-driven predic- tion of battery cycle life before capacity degradation, Nature Energy 4 (5) (2019) 383–391.doi: 10.1038/s41560-019-0356-8

    doi:10.1038/s41560-019-0356-8 2019

  21. [21]

    P . M. Attia, A. Grover, N. Jin, K. A. Severson, T . M. Markov, Y.-H. Liao, M. H. Chen, B. Cheong, N. Perkins, Z. Yang, P . K. Herring, M. Aykol, S. J. Harris, R. D. Braatz, S. Ermon, W . C. Chueh, Closed- 25 loop optimization of fast-charging protocols for batteries with machine learning, Nature 578 (2020) 397–402.doi:10.1038/s41586-020-1994-5

    doi:10.1038/s41586-020-1994-5 2020

  22. [22]

    R. R. Richardson, M. A. Osborne, D. A. Howey, Gaussian process regression for forecasting battery state of health, Journal of Power Sources 357 (2017) 209–219.doi:10.1016/j.jpowsour.2017. 05.004

    work page doi:10.1016/j.jpowsour.2017 2017

  23. [23]

    Roman, S

    D. Roman, S. Saxena, V . Robu, M. Pecht, D. Flynn, Machine learning pipeline for battery state-of-health estimation, Nature Machine Intelligence 3 (2021) 447–456.doi:10.1038/ s42256-021-00312-3

    2021

  24. [24]

    Sadler, R

    J. Sadler, R. Mohammed, M. Castle, K. Uddin, Diagnostic forecasting of battery degradation through contrastive learning, Scientific Reports 15 (1) (2025) 33360.doi:10.1038/s41598-025-17183-y

    work page doi:10.1038/s41598-025-17183-y 2025

  25. [25]

    D. P . Akhare, Z. Chen, R. Gulotty, T . Luo, J.-X. Wang, Probabilistic physics-integrated neural differ- entiable modeling for isothermal chemical vapor infiltration process, npj Computational Materials 10 (1) (2024) 265.doi:10.1038/s41524-024-01407-y

    work page doi:10.1038/s41524-024-01407-y 2024

  26. [26]

    Z. Chen, M. Yacouti, M. Shakiba, J.-X. Wang, T . Luo, V . Varshney, Predicting stress and damage in carbon fiber-reinforced composites deformation process using composite u-net surrogate model, Theoretical and Applied Mechanics Letters 16 (1) (2026) 100634.doi:https://doi.org/10.1016/ j.taml.2025.100634

    arXiv 2026

  27. [27]

    F . Wang, Z. Zhai, Z. Zhao, Y. Di, X. Chen, Physics-informed neural network for lithium-ion battery degradation stable modeling and prognosis, Nature Communications 15 (1) (2024) 4332.doi:10. 1038/s41467-024-48779-z

    2024

  28. [28]

    Ma, et al., A multi-stage dataset for lithium-ion battery aging study with various charging condi- tions, Scientific Data 11 (1) (2024) 651.doi:10.1038/s41597-024-03277-z

    L. Ma, et al., A multi-stage dataset for lithium-ion battery aging study with various charging condi- tions, Scientific Data 11 (1) (2024) 651.doi:10.1038/s41597-024-03277-z

    work page doi:10.1038/s41597-024-03277-z 2024

  29. [29]

    J. Zhu, Y. Wang, Y. Huang, B. R. Gopaluni, Y. Cao, M. Heere, M. J. Mühlbauer, L. Mereacre, H. Dai, X. Liu, A. Senyshyn, X. Wei, M. Knapp, H. Ehrenberg, Data-driven capacity estimation of commercial lithium-ion batteries from voltage relaxation, Nature Communications 13 (1) (2022) 2261.doi: 10.1038/s41467-022-29837-w. 26

    work page doi:10.1038/s41467-022-29837-w 2022

  30. [30]

    Raissi, P

    M. Raissi, P . Perdikaris, G. E. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equa- tions, Journal of Computational Physics 378 (2019) 686–707.doi:10.1016/j.jcp.2018.10.045

    doi:10.1016/j.jcp.2018.10.045 2019

  31. [31]

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P . Perdikaris, S. Wang, L. Yang, Physics-informed machine learning, Nature Reviews Physics 3 (2021) 422–440.doi:10.1038/s42254-021-00314-5

    doi:10.1038/s42254-021-00314-5 2021

  32. [32]

    Peled, S

    E. Peled, S. Menkin, Review–SEI: Past, present and future, Journal of The Electrochemical Society 164 (7) (2017) A1703–A1719.doi:10.1149/2.1441707jes

    work page doi:10.1149/2.1441707jes 2017

  33. [33]

    M. B. Pinson, M. Z. Bazant, Theory of SEI formation in rechargeable batteries: Capacity fade, accel- erated aging and lifetime prediction, Journal of The Electrochemical Society 160 (2) (2013) A243– A250.doi:10.1149/2.044302jes

    doi:10.1149/2.044302jes 2013

  34. [34]

    E. Peled, The electrochemical behavior of alkali and alkaline earth metals in nonaqueous battery systems–the solid electrolyte interphase model, Journal of The Electrochemical Society 126 (12) (1979) 2047–2051.doi:10.1149/1.2128859

    work page doi:10.1149/1.2128859 1979

  35. [35]

    Christensen, J

    J. Christensen, J. Newman, Cyclable lithium and capacity loss in li-ion cells, Journal of The Electro- chemical Society 152 (4) (2005) A818–A829.doi:10.1149/1.1870752

    work page doi:10.1149/1.1870752 2005

  36. [36]

    Ramadass, B

    P . Ramadass, B. Haran, R. White, B. N. Popov, Mathematical modeling of the capacity fade of Li-ion cells, Journal of Power Sources 123 (2) (2003) 230–240.doi:10.1016/S0378-7753(03)00531-7

    work page doi:10.1016/s0378-7753(03)00531-7 2003

  37. [37]

    P . Keil, A. Jossen, Calendar aging of lithium-ion batteries, Journal of The Electrochemical Society 163 (9) (2016) A1872–A1880.doi:10.1149/2.0411609jes

    work page doi:10.1149/2.0411609jes 2016

  38. [38]

    H. J. Ploehn, P . Ramadass, R. E. White, Solvent diffusion model for aging of lithium-ion battery cells, Journal of The Electrochemical Society 151 (3) (2004) A456–A462.doi:10.1149/1.1644601

    work page doi:10.1149/1.1644601 2004

  39. [39]

    Ramadass, B

    P . Ramadass, B. Haran, P . M. Gomadam, R. White, B. N. Popov, Development of first principles ca- pacity fade model for li-ion cells, Journal of The Electrochemical Society 151 (2) (2004) A196–A203. doi:10.1149/1.1634273

    work page doi:10.1149/1.1634273 2004

  40. [40]

    Single, A

    F . Single, A. Latz, B. Horstmann, Identifying the mechanism of continued growth of the solid- electrolyte interphase, ChemSusChem 11 (12) (2018) 1950–1955.doi:10.1002/cssc.201800077. 27

    work page doi:10.1002/cssc.201800077 2018

  41. [41]

    Bloom, B

    I. Bloom, B. W . Cole, J. J. Sohn, S. A. Jones, E. G. Polzin, V . S. Battaglia, G. L. Henriksen, C. G. Motloch, R. Richardson, T . Unkelhaeuser, D. Ingersoll, H. L. Case, An accelerated calendar and cycle life study of li-ion cells, Journal of Power Sources 101 (2) (2001) 238–247.doi:10.1016/S0378-7753(01) 00783-2

    doi:10.1016/s0378-7753(01 2001

  42. [42]

    J. Wang, P . Liu, J. Hicks-Garner, E. Sherman, S. Soukiazian, M. Verbrugge, H. Tataria, J. Musser, P . Fi- namore, Cycle-life model for graphite-LiFePO4 cells, Journal of Power Sources 196 (8) (2011) 3942– 3948.doi:10.1016/j.jpowsour.2010.11.134

    work page doi:10.1016/j.jpowsour.2010.11.134 2011

  43. [43]

    Schmalstieg, S

    J. Schmalstieg, S. Käbitz, M. Ecker, D. U. Sauer, From accelerated aging tests to a lifetime prediction model: Analyzing lithium-ion batteries, in: Proceedings of the 2013 World Electric Vehicle Sym- posium and Exhibition (EVS27), IEEE, Barcelona, Spain, 2013, pp. 1–12.doi:10.1109/EVS.2013. 6914753

    work page doi:10.1109/evs.2013 2013

  44. [44]

    Schmalstieg, S

    J. Schmalstieg, S. Käbitz, M. Ecker, D. U. Sauer, A holistic aging model for Li(NiMnCo)O2 based 18650 lithium-ion batteries, Journal of Power Sources 257 (2014) 325–334.doi:10.1016/j. jpowsour.2014.02.012

    doi:10.1016/j 2014

  45. [45]

    F . Leng, C. M. Tan, M. Pecht, Effect of temperature on the aging rate of li ion battery operating above room temperature, Scientific Reports 5 (2015) 12967.doi:10.1038/srep12967

    work page doi:10.1038/srep12967 2015

  46. [46]

    X. Lin, J. Park, L. Liu, Y. Lee, A. M. Sastry, W . Lu, A comprehensive capacity fade model and analysis for Li-ion batteries, Journal of The Electrochemical Society 160 (10) (2013) A1701–A1710.doi:10. 1149/2.040310jes. 28

    2013