Pith. sign in

REVIEW 3 major objections 4 minor 39 references

Reinforcement Learning for Robust Ageing-Aware Control of Li-ion Battery Systems with Data-Driven Formal Verification

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A switched reinforcement-learning charging controller for Li-ion cells is certified—with probability at least 99.956% and confidence at least 99.9999%—to reach 90% SOC within 80 minutes while keeping voltage and temperature within safety li

desk verdict A genuine integration of RL and data-driven formal verification for battery control, but the headline probability guarantee depends on an under-specified SOC partition. read the letter →

arxiv 2509.04288 v2 pith:L4IVSU3L submitted 2025-09-04 eess.SY cs.AIcs.SY

Rudi Coppola , Hovsep Touloujian , Pierfrancesco Ombrini , Manuel Mazo Jr This is my paper · ORCID
classification eess.SYcs.AIcs.SY
keywords Li-ionbatteriesageing-awarechargingreinforcementlearningdata-drivenformalverificationreach-while-avoidspecificationswitchedcontrolsolid-electrolyteinterphasePACguarantees
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to show that reinforcement learning can produce a battery charging controller that is both fast and ageing-aware, and that the controller's safety and performance can be certified with formal probabilistic guarantees rather than relying on average test performance. It closes the loop with a data-driven finite abstraction of the closed-loop system: the abstraction over-approximates all output-label behaviors sampled from a high-fidelity electrochemical model, and a PAC-style bound turns that into a distribution-free guarantee. When the first RL controller fails verification, counterexamples drive a partition of the initial-condition space into regions, each with its own retrained RL controller; the switched controller then passes verification with a stated 99.956% probability. The practical payoff is a charging protocol that beats the standard CC-CV method on charging time and capacity loss, with formal backing for the improvement.

What carries the argument

The load-bearing object is the data-driven Strongest Asynchronous ℓ-complete Abstraction (SAℓCA): a finite automaton whose states are length-ℓ sequences of output labels observed from sampled trajectories, with transitions given by the domino rule—a successor state's first ℓ−1 labels must equal the predecessor's last ℓ−1 labels. It over-approximates the concrete battery's output behaviors, and a PAC-style bound (Proposition 2) says a new initial condition's label sequence lies inside the abstraction with probability at least 1−ε at confidence at least 1−β. This abstraction acts as the verifier inside a counterexample-guided inductive synthesis loop: it certifies the Reach-While-Avoid propert

What would settle it

Take the final switched controller and simulate a large batch of initial conditions drawn from the boundaries of the 8 partition cells, or with a deliberately biased SOC measurement of a few percent; if any trajectory violates voltage or temperature limits or fails to reach 90% SOC within 80 minutes, then the abstraction's label alignment or its coverage of the switched regions is wrong and the 99.956% claim does not transfer to those cases.

Watch

Extended reading notes

Core claim

The paper establishes that a switched output-feedback charging controller for a Li-ion cell, synthesized by iterating RL training against a data-driven formal verifier, can be certified to satisfy a time-bounded Reach-While-Avoid specification—reach 90% SOC within 80 minutes while keeping voltage at most 4.2 V and temperature at most 45 °C—with probability at least 99.956% and confidence at least 99.9999%, over random initial conditions, manufacturing parameter variations, and state-of-health values between 85% and 100%. The guarantee is distribution-free: it holds for any underlying distribution of initial conditions, not just the Gaussian used in sampling. The same finite abstraction is re

Load-bearing premise

The probabilistic certificate transfers to a real cell only if the output labels used by the verifier match the true physical sets—goal label t means SOC in [0.9,1] and safe label a means voltage at most 4.2 V and temperature at most 45 °C—so any SOC-estimation error or label discretization mismatch breaks the chain.

Editorial extensions

If this is right

  • If correct, deploying the synthesized controller on cells within the specified parameter and SOH ranges carries a distribution-free guarantee that a fresh cell will follow a safe, goal-reaching charge with probability at least 99.956% at confidence at least 99.9999%.
  • The same data-driven abstraction can be reused to verify additional temporal properties without new sampling or training, e.g., time bounds to intermediate SOC levels such as 50% in at most 33.75 minutes and 70% in at most 49.25 minutes.
  • The CEGIS architecture turns verification failures into region-specific retraining, so the formal guarantee scales from a single controller to a switched family of controllers without abandoning the abstraction-based certificate.
  • The controller operates in output feedback using only measurable signals—SOC, voltage, temperature, past input current, and time—so the protocol is deployable in a battery management system without estimating internal electrochemical states.
  • Relative to the CC-CV benchmark at 3.5 A, the synthesized protocol saves up to 20 minutes of charging time in some initial-condition regions with comparable capacity loss, and improves the reward balance across all sampled initial conditions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because training and verification share the same electrochemical simulator, a model bias common to both—for example in SEI kinetics—would not be detected by the abstraction; independent hardware experiments would be needed before relying on the numeric probability in practice.
  • The switched controller selects a region from the initial measurements only, so initial measurement noise or sensor bias at the first step could place a cell in the wrong control region; the paper does not quantify sensitivity to that error.
  • The reusable-abstraction pattern suggests a broader recipe: once a certified finite model of a closed-loop system is built, many control objectives can be checked offline against the same PAC bound, at the cost of re-verifying when the plant distribution changes.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes a CEGIS-based framework for synthesizing an output-feedback charging controller for a DFN-model Li-ion battery. RL (soft actor-critic) trains candidate controllers, and a data-driven Strongest Asynchronous ℓ-complete Abstraction (SAℓCA) built from 10^5 sampled H-long behaviors is used to verify a time-bounded Reach-While-Avoid (RWA) specification. If the finite abstraction satisfies the specification, Proposition 2 gives a distribution-free PAC guarantee. The headline result, Eq. (25), is that with confidence at least 99.9999%, the probability that a new initial condition (voltage, temperature, manufacturing parameters, SOH) yields a behavior in the abstraction is at least 99.956%; since all abstract behaviors are verified to satisfy the RWA specification, the concrete behavior is concluded to satisfy RWA. One refinement step partitions the initial set by voltage/temperature and retrains eight agents, yielding the switched controller K1.

Significance. If the guarantee chain is valid, the paper makes a valuable contribution: it couples RL with data-driven formal verification for a realistic physics-based battery model, treats manufacturing and SOH variability explicitly, and demonstrates reuse of the abstraction for secondary properties (50%/70% SOC time bounds). The approach is distribution-free and avoids Gaussian assumptions on the surrogate. The main caveat is that the transfer of the abstract verification to the concrete battery depends on a precise alignment between abstract labels and the concrete goal/safe sets; the current manuscript does not establish this alignment reproducibly, and the claimed ageing guarantee is broader than what the verified RWA specification supports.

major comments (3)
  1. [Sec. 4.2, Sec. 5.1, Eq. (25)] The SOC partition is ambiguous and inconsistent with the stated RWA goal. The text says the [0,1] interval is 'partitioned in 19 identical sections plus 1 for the Goal set' while Sec. 5.1 fixes X_G as SOC∈[0.9,1]. If the 19 sections are width-0.05 intervals ending at 0.95, then the goal label t is [0.95,1], a strict subset of X_G; if instead the 19 sections cover [0,0.9), the later claim that labels g–t denote SOC≥50% is false under either arithmetic. The exact target of the 99.956% guarantee is therefore not reproducible. Please specify the interval for every label, reconcile the reuse claim, and provide the implementation or a precise partition table. This is load-bearing because Eq. (25) transfers to the concrete RWA only if the abstract goal/safe labels are aligned with X_G and X_S.
  2. [Sec. 1, Sec. 6] The introduction claims 'We obtain a charging protocol with formal guarantees on the maximum rate of ageing of the battery'. This is not supported by the verification machinery: the RWA specification (Sec. 3.2, Sec. 5.1) involves only SOC, V, and T, and the abstraction labels contain no capacity-loss or ageing variable. The capacity loss Q_l appears only in the RL reward (Eq. (17)–(20)) and in benchmarking plots. The formal guarantee (Eq. (25)) is about the RWA specification, not about ageing. Please revise the claims to state exactly what is formally guaranteed, or extend the specification to include a verified ageing metric.
  3. [Sec. 2.1 Eq. (2), Sec. 3.1.3, Sec. 4.2 Step 1] The closed-loop system is presented as autonomous in Eq. (2), but the controller observes the time step k (Sec. 3.1.3: z_k includes 'time-step, ... k'; Sec. 4.1: policy π(z_k) with k in the observation). Therefore the closed-loop dynamics depend explicitly on k and are not of the form x_{k+1}=f(x_k). The application of Proposition 2, which is stated for a time-invariant TS, is not formally justified as written. Either remove k from the controller observation, or explicitly augment the state with the time index and discuss how the abstraction handles the resulting non-autonomous system. This is needed for the rigor of the guarantee chain.
minor comments (4)
  1. [Sec. 4.2] The phrase 'the Goal set, which consists of the region [SOC,1]' uses 'SOC' both as a variable name and as the state-of-charge; this makes the intended lower bound unclear. Use a distinct symbol, e.g., SOC_goal.
  2. [Fig. 11 caption] The caption reads 'Effect of the CC-CV protocol on pristine and aged cell...' but the text refers to the K1 controller. This appears to be a copy-paste error from Fig. 6 and should be corrected.
  3. [Sec. 5.2, reuse claim] The statement 'labels from g to t and from o to t indicate SOCs greater or equal to 50% and 70%' is likely a typo (probably 'k to t' for 50% if the width is 0.05). Please correct and verify against the actual partition.
  4. [General] No code or data are provided. Given that the reported guarantee hinges on the exact implementation of the partition and the SAℓCA construction, the authors should make the implementation available or at least list the exact label intervals and verification scripts.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the PAC/SAℓCA guarantee is used as an external theorem and the battery-level claim is a new application.

full rationale

The derivation chain is: train an RL controller, close the loop with the DFN model, sample N=10^5 behaviors, construct a data-driven SAℓCA, verify the finite abstraction against RWA via safety/reachability games, and then invoke Proposition 2 (quoted from [14]) plus the scenario-optimization bound of [21] to state the 99.956%/99.9999% PAC guarantee. The load-bearing probabilistic step is a general theorem about deterministic systems with finite output labels; its stated assumptions (i.i.d. sampling, Assumption 2) do not include the battery RWA property, and the bound on ε is computed from N, s*_N, and β, not fitted to make the conclusion true. The finite-abstraction verification is a model-checking step on the abstraction, not a renaming of the sampled outcomes. The use of prior work by the same group ([13]–[15], [20]) is real evidence rather than circular: Proposition 1 comes from Schmuck & Raisch [20] and Proposition 2 is an established PAC result; no uniqueness claim or ansatz is smuggled in by citation. The only notable concern in the text is Assumption 1: the abstraction-to-concrete transfer requires the output labels to be exactly aligned with X_G and X_S, and Section 4.2's description of the SOC partition is ambiguous relative to Section 5.1's X_G=[0.9,1] and to the later claim that labels g–t denote SOC≥50%. That ambiguity affects the reproducibility and the precise target of the guarantee, but it is an assumption/consistency issue, not a circular reduction: the RWA conclusion is not equivalent to the partition definition by construction. No fitted parameter is disguised as a prediction, and the controller comparison against CC-CV is external benchmarking. Therefore the circularity score is 0.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The formal certificate rests on prior SAℓCA and PAC theorems, on the exact label alignment assumption, and on the simulation model. The free parameters are design choices that affect the achieved controller and the tightness of the abstraction, but they are not fitted to produce the RWA guarantee.

free parameters (4)
  • Reward weights lambda1, lambda2, lambda3 = lambda1=10^2, lambda2=10^5, lambda3=2.5e-2 (Table 1)
    Hand-chosen weights balancing SOC progress, time penalty, and capacity loss in the RL reward; they shape the controller but are not fitted by the formal verification.
  • Abstraction memory length l = 6
    Chosen in Table 1 as a trade-off between abstraction tightness and state count; larger l gives a tighter over-approximation.
  • SOC partition granularity = 19 bins plus 1 goal bin
    Section 4.2: chosen finely to avoid self-loops in the abstraction, which is pivotal for reachability verification.
  • Refinement clustering grid = 4 initial-voltage segments x 2 initial-temperature segments (8 agents)
    Section 5.2.1: chosen after observing that counterexamples spread uniformly over the initial-condition set; a hyper-parameter of the CEGIS refinement.
assumptions (5)
  • standard math SAℓCA over-approximation: B^H(S) subset of B^H(S_l) for l >= 1 (Proposition 1, ref [20]).
    The finite abstraction conservatively covers all H-long behaviors of the concrete system; cited from Schmuck and Raisch.
  • standard math Data-driven PAC bound: P^N[ P[ B^H_x(S) in B^H(hat S_l) ] >= 1 - epsilon(s*, N, beta) ] >= 1 - beta (Proposition 2, refs [14,21]).
    Distribution-free guarantee that a new sample's behavior is covered by the data-driven abstraction; cited from the authors' own prior work and Garatti/Campi.
  • domain assumption Assumption 1: X0, XS, XG are aligned with the output partition map h, i.e., label sets exactly represent initial, goal, and safe sets.
    Section 2.1 after Definition 3; without this, label-level reachability and safety would not imply concrete-state satisfaction.
  • domain assumption Assumption 2: initial conditions x0 are drawn i.i.d. from a probability distribution P.
    Section 2.2; required for the PAC bound and for the reported confidence statement.
  • domain assumption The DFN/PyBaMM model with SEI-driven capacity loss is a faithful representation of the LG M50LT cell over the tested SOH range.
    Section 3.1 and Appendix A; all training, verification, and benchmarking are performed on this model, not on physical cells.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Reinforcement Learning for Robust Ageing-Aware Control of Li-ion Battery Systems with Data-Driven Formal Verification." pith.science (2026). https://pith.science/paper/L4IVSU3L

@misc{pith2026250904288,
  author       = {Pith},
  title        = {Pith review of: Reinforcement Learning for Robust Ageing-Aware Control of Li-ion Battery Systems with Data-Driven Formal Verification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L4IVSU3L}},
  note         = {Machine review of arXiv:2509.04288}
}
read the original abstract

Rechargeable lithium-ion (Li-ion) batteries are a ubiquitous element of modern technology. In the last decades, the production and design of such batteries and their adjacent embedded charging and safety protocols, denoted by Battery Management Systems (BMS), has taken central stage. A fundamental challenge to be addressed is the trade-off between the speed of charging and the ageing behavior, resulting in the loss of capacity in the battery cell. We rely on a high-fidelity physics-based battery model and propose an approach to data-driven charging and safety protocol design. Following a Counterexample-Guided Inductive Synthesis scheme, we combine Reinforcement Learning (RL) with recent developments in data-driven formal methods to obtain a hybrid control strategy: RL is used to synthesise the individual controllers, and a data-driven abstraction guides their partitioning into a switched structure, depending on the initial output measurements of the battery. The resulting discrete selection among RL-based controllers, coupled with the continuous battery dynamics, realises a hybrid system. When a design meets the desired criteria, the abstraction provides probabilistic guarantees on the closed-loop performance of the cell.

Figures

Figures reproduced from arXiv: 2509.04288 by the authors.

Figure 1
Figure 1. Illustration of the SAℓCA for Example 1 for ℓ = 2 (left) and ℓ = 3 (right). The upper portion of every node represents the state, the lower portion represents the state’s output according to Proposition 1. The edges represent the transition relation Eℓ, and every state of the SAℓCA is initial. RWA specifications demand that any trace starting from the set X0 reaches the target set XG in at most H steps, while remain… view at source ↗
Figure 2
Figure 2. Schematic of the CEGIS architecture. The learner (orange block) proposes a controller [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Schematic of the DFN Model. Upper: electrode-scale transport regulates the electrolyte and electric potential drops [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Example of the effects of various input currents on the voltage, current and temperature evolution during a CC-CV [PITH_FULL_IMAGE:figures/full_fig_p017_4.png]
Figure 5
Figure 5. Figure 5: Benchmarking of the CC-CV protocol. Relations between input current and charging times, maximum temperatures [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]
Figure 6
Figure 6. Figure 6: Effect of the CC-CV protocol on pristine and aged cell with average manufacturing parameters. [PITH_FULL_IMAGE:figures/full_fig_p019_6.png]
Figure 7
Figure 7. Figure 7: Trajectory distribution for the CC-CV Protocol [PITH_FULL_IMAGE:figures/full_fig_p020_7.png]
Figure 8
Figure 8. Figure 8: (Left) Average cumulative reward obtained by the battery controlled with [PITH_FULL_IMAGE:figures/full_fig_p023_8.png]
Figure 9
Figure 9. Figure 9: (Left) Average capacity loss obtained by the battery controlled with [PITH_FULL_IMAGE:figures/full_fig_p023_9.png]
Figure 10
Figure 10. Figure 10: (Left) Average charging time obtained by the battery controlled with [PITH_FULL_IMAGE:figures/full_fig_p024_10.png]
Figure 11
Figure 11. Figure 11: Effect of the CC-CV protocol on pristine and aged cell with average manufacturing parameters. [PITH_FULL_IMAGE:figures/full_fig_p025_11.png]
Figure 12
Figure 12. Figure 12: Trajectory distribution for the final synthesized controller [PITH_FULL_IMAGE:figures/full_fig_p026_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 37 canonical work pages

  1. [1]

    J. Weng, A. Jossen, A. Stefanopoulou, J. Li, X. Feng, G. Offer, Fast-charging lithium-ion batteries require a systems engineering approach, Nature Energy (2025) 1–2

    work page 2025

  2. [2]

    K. A. Smith, C. D. Rahn, C.-Y. Wang, Control oriented 1d electrochemical model of lithium ion battery, Energy Conversion and management 48 (9) (2007) 2565–2578

    work page 2007

  3. [3]

    Barré, B

    A. Barré, B. Deguilhem, S. Grolleau, M. Gérard, F. Suard, D. Riu, A review on lithium-ion battery ageing mechanisms and estimations for automotive applications, Journal of power sources 241 (2013) 680–689

    work page 2013

  4. [4]

    Darling, J

    R. Darling, J. Newman, Modeling side reactions in composite liymn2 o 4 electrodes, Journal of The Electrochemical Society 145 (3) (1998) 990

    work page 1998

  5. [5]

    Khalik, H

    Z. Khalik, H. J. Bergveld, M. Donkers, Ageing-aware charging of lithium-ion batteries using a surrogate model, in: 2021 American Control Conference (ACC), IEEE, 2021, pp. 4414–4420

    work page 2021

  6. [6]

    Khalik, H

    Z. Khalik, H. J. Bergveld, M. Donkers, Ageing-aware charging of lithium-ion batteries using an electrochemistry-based model with capacity-loss side reactions, in: 2020 American Control Conference (ACC), IEEE, 2020, pp. 2213–2218

    work page 2020

  7. [7]

    Khalik, Modeling and Optimal Control for Aging-Aware Charging of Batteries, PhD Thesis (research TU/e / graduation TU/e), Eindhoven University of Technology, Eindhoven (Nov

    Z. Khalik, Modeling and Optimal Control for Aging-Aware Charging of Batteries, PhD Thesis (research TU/e / graduation TU/e), Eindhoven University of Technology, Eindhoven (Nov. 2021)

    work page 2021

  8. [8]

    Kolluri, S

    S. Kolluri, S. V. Aduru, M. Pathak, R. D. Braatz, V. R. Subramanian, Real-time nonlinear model predictive control (nmpc) strategies using physics-based models for advanced lithium-ion battery man- agement system (bms), Journal of The Electrochemical Society 167 (6) (2020) 063505

    work page 2020

Show all 39 references
  1. [9]

    S. Park, A. Pozzi, M. Whitmeyer, H. Perez, A. Kandel, G. Kim, Y. Choi, W. T. Joe, D. M. Raimondo, S. Moura, A Deep Reinforcement Learning Framework for Fast Charging of Li-ion Batteries, IEEE Transactions on Transportation Electrification 8 (2) (2022) 2770–2784. 27

    2022

  2. [10]

    T. Gao, Y. Han, D. Fraggedakis, S. Das, T. Zhou, C.-N. Yeh, S. Xu, W. C. Chueh, J. Li, M. Z. Bazant, Interplay of lithium intercalation and plating on a single graphite particle, Joule 5 (2) (2021) 393–414

    2021

  3. [11]

    X. Lu, M. Lagnoni, A. Bertei, S. Das, R. E. Owen, Q. Li, K. O’Regan, A. Wade, D. P. Finegan, E.Kendrick, etal., Multiscaledynamicsofchargingandplatingingraphiteelectrodescouplingoperando microscopy and phase-field modelling, Nature Communications 14 (1) (2023) 5127

    2023

  4. [12]

    M. A. Chowdhury, S. S. Al-Wahaibi, Q. Lu, Adaptive safe reinforcement learning-enabled optimization of battery fast-charging protocols, AIChE Journal 71 (1) (2025) e18605

    2025

  5. [13]

    R.Coppola, A.Peruffo, M.MazoJr, Data-drivenAbstractionsforVerificationofDeterministicSystems, arXiv preprint arXiv:2211.01793 (2022)

    2022 arXiv

  6. [14]

    Coppola, A

    R. Coppola, A. Peruffo, M. Mazo, Data-driven abstractions for verification of linear systems, IEEE Control Systems Letters 7 (2023) 2737–2742

    2023

  7. [15]

    Coppola, A

    R. Coppola, A. Peruffo, M. Mazo Jr, Data-driven abstractions for control systems via random explo- ration, arXiv preprint arXiv:2402.10668 (2024)

    2024 arXiv

  8. [16]

    Abate, C

    A. Abate, C. David, P. Kesseli, D. Kroening, E. Polgreen, Counterexample guided inductive synthesis modulo theories, in: International Conference on Computer Aided Verification, Springer, 2018, pp. 270–288

    2018

  9. [17]

    Ravanbakhsh, S

    H. Ravanbakhsh, S. Sankaranarayanan, Learning control lyapunov functions from counterexamples and demonstrations, Autonomous Robots 43 (2) (2019) 275–307

    2019

  10. [18]

    Ahmed, A

    D. Ahmed, A. Peruffo, A. Abate, Automated and sound synthesis of lyapunov functions with smt solvers, in: Tools and Algorithms for the Construction and Analysis of Systems: 26th International Conference, TACAS 2020, Held as Part of the European Joint Conferences on Theory and ...

    2020

  11. [19]

    Tabuada, Verification and Control of Hybrid Systems: A Symbolic Approach, Springer Science & Business Media, 2009

    P. Tabuada, Verification and Control of Hybrid Systems: A Symbolic Approach, Springer Science & Business Media, 2009

    2009

  12. [20]

    Schmuck, J

    A.-K. Schmuck, J. Raisch, Asynchronous l-complete approximations, Systems & Control Letters 73 (2014) 67–75

    2014

  13. [21]

    Garatti, M

    S. Garatti, M. C. Campi, The risk of making decisions from data through the lens of the scenario approach, IFAC-PapersOnLine 54 (7) (2021) 607–612. 28

    2021

  14. [22]

    Doyle, T

    M. Doyle, T. F. Fuller, J. Newman, Modeling of galvanostatic charge and discharge of the lithium/polymer/insertion cell, Journal of the Electrochemical society 140 (6) (1993) 1526

    1993

  15. [23]

    Brosa Planella, W

    F. Brosa Planella, W. Ai, A. M. Boyce, A. Ghosh, I. Korotkin, S. Sahu, V. Sulzer, R. Timms, T. G. Tranter, M. Zyskin, et al., A continuum of physics-based lithium-ion battery models reviewed, Progress in Energy 4 (4) (2022) 042003

    2022

  16. [24]

    C.-H. Chen, F. B. Planella, K. O’regan, D. Gastol, W. D. Widanage, E. Kendrick, Development of experimental techniques for parameterization of multi-scale lithium-ion battery models, Journal of The Electrochemical Society 167 (8) (2020) 080534

    2020

  17. [25]

    Nyman, M

    A. Nyman, M. Behm, G. Lindbergh, Electrochemical characterisation and modelling of the mass trans- port phenomena in lipf6–ec–emc electrolyte, Electrochimica Acta 53 (22) (2008) 6356–6365

    2008

  18. [26]

    O’Regan, F

    K. O’Regan, F. B. Planella, W. D. Widanage, E. Kendrick, Thermal-electrochemical parameters of a high energy lithium-ion cylindrical battery, Electrochimica Acta 425 (2022) 140700

    2022

  19. [27]

    Ramadass, B

    P. Ramadass, B. Haran, P. M. Gomadam, R. White, B. N. Popov, Development of first principles capacity fade model for li-ion cells, Journal of the Electrochemical Society 151 (2) (2004) A196

    2004

  20. [28]

    Sulzer, S

    V. Sulzer, S. G. Marquis, R. Timms, M. Robinson, S. J. Chapman, Python battery mathematical modelling (pybamm), Journal of Open Research Software 9 (1) (2021)

    2021

  21. [29]

    Haarnoja, A

    T. Haarnoja, A. Zhou, P. Abbeel, S. Levine, Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor, in: International conference on machine learning, Pmlr, 2018, pp. 1861–1870

    2018

  22. [30]

    D. Xu, Y. Tian, A comprehensive survey of clustering algorithms, Annals of data science 2 (2015) 165–193

    2015

  23. [31]

    Delft High Performance Computing Centre (DHPC), DelftBlue Supercomputer (Phase 1),https:// www.tudelft.nl/dhpc/ark:/44463/DelftBluePhase1(2022)

    2022

  24. [32]

    Reijsbergen, P.-T

    D. Reijsbergen, P.-T. De Boer, W. Scheinhardt, B. Haverkort, On hypothesis testing for statistical model checking, International journal on software tools for technology transfer 17 (4) (2015) 377–395

    2015

  25. [33]

    Karanth, M

    P. Karanth, M. Weijers, P. Ombrini, D. Ripepi, F. Ooms, F. M. Mulder, A phase inversion strategy for low-tortuosity and ultrahigh-mass-loading nickel-rich layered oxide electrodes, Cell Reports Physical Science 5 (6) (2024)

    2024

  26. [34]

    M. Z. Bazant, Unified quantum theory of electrochemical kinetics by coupled ion–electron transfer, Faraday Discussions 246 (2023) 60–124. 29

    2023

  27. [35]

    P. Bai, D. A. Cogswell, M. Z. Bazant, Suppression of phase separation in lifepo4 nanoparticles during battery discharge, Nano letters 11 (11) (2011) 4890–4896

    2011

  28. [36]

    Ombrini, Q

    P. Ombrini, Q. Wang, A. Vasileiadis, F. Wu, Z. Gao, X. Hu, M. van Hulzen, B. Li, C. Zhao, M. Wage- maker, Kinetically induced memory effect in li-ion batteries, EES Batteries (2025)

    2025

  29. [37]

    Ombrini, M

    P. Ombrini, M. Z. Bazant, M. Wagemaker, A. Vasileiadis, Thermodynamics of multi-sublattice battery active materials: from an extended regular solution theory to a phase-field model of limnyfe1-ypo4, npj Computational Materials 9 (1) (2023) 148

    2023

  30. [38]

    Karger, S

    A. Karger, S. E. O’Kane, M. Rogge, C. Kirst, J. P. Singer, M. Marinescu, G. J. Offer, A. Jossen, Modeling particle versus sei cracking in lithium-ion battery degradation: Why calendar and cycle aging cannot simply be added, Journal of The Electrochemical Society 171 (9) (2024) 090512

    2024

  31. [39]

    Sulzer, S

    V. Sulzer, S. G. Marquis, R. Timms, M. Robinson, S. J. Chapman, Python Battery Mathematical Modelling (PyBaMM), Journal of Open Research Software 9 (1) (2021). 30

    2021

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.