arxiv: 2604.14566 · v1 · submitted 2026-04-16 · 💻 cs.LG · cs.SY· eess.SY

Recognition: unknown

Physics-Informed Machine Learning for Pouch Cell Temperature Estimation

Zheng Liu

Authors on Pith no claims yet

Pith reviewed 2026-05-10 12:04 UTC · model grok-4.3

classification 💻 cs.LG cs.SYeess.SY

keywords physics-informed machine learningpouch cell temperature estimationbattery thermal managementneural networksheat transfer equationssurrogate modelingcooling channel geometries

0 comments

The pith

Embedding heat transfer equations in a neural network loss function enables faster and more accurate steady-state temperature estimation for pouch cells.

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

The paper introduces a physics-informed machine learning approach to estimate temperatures in pouch cells using indirect liquid cooling. It combines data from simulations with the physical heat transfer laws enforced in the model's training objective. This leads to quicker convergence during training and a 49.1 percent lower mean squared error than standard data-driven neural networks. The improvement is especially noticeable in areas distant from the cooling channels, where pure data models struggle more. Such a method could help in designing better thermal management for electric vehicle batteries by providing reliable predictions without full simulations each time.

Core claim

The PIML framework integrates the governing heat transfer equations directly into the neural network's loss function, resulting in high-fidelity predictions of steady-state temperature profiles that converge more rapidly and achieve a 49.1% reduction in mean squared error compared to purely data-driven models, with superior performance validated on independent test cases especially away from cooling channels.

What carries the argument

The physics-informed loss function, which adds terms enforcing the heat transfer governing equations to the neural network training objective.

If this is right

The PIML model can serve as an efficient surrogate for finite element simulations when optimizing battery cooling channel designs.
Higher accuracy away from cooling channels supports reliable predictions across varied thermal distributions in pouch cells.
Faster convergence during training lowers the computational resources needed to develop temperature estimation models.
Validation across multiple cooling geometries indicates the approach generalizes to different indirect liquid cooling configurations.

Where Pith is reading between the lines

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

The same loss-function approach could be adapted for time-dependent heat equations to handle transient battery operation.
Physics-informed training might lower the volume of simulation data required for other thermal surrogate models in engineering.
Efficient inference from the trained network opens paths to embedding temperature estimation in vehicle control systems.

Load-bearing premise

The governing heat transfer equations accurately capture the steady-state thermal behavior across all tested cooling channel geometries and the training data from finite element simulations is representative without significant modeling errors.

What would settle it

A new cooling channel geometry or boundary condition where experimental temperature measurements deviate substantially from PIML predictions while matching a data-driven model more closely would falsify the accuracy and convergence superiority.

Figures

Figures reproduced from arXiv: 2604.14566 by Zheng Liu.

**Figure 2.** Figure 2: Temperature estimation by data-driven and PIML methods. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

read the original abstract

Accurate temperature estimation of pouch cells with indirect liquid cooling is essential for optimizing battery thermal management systems for transportation electrification. However, it is challenging due to the computational expense of finite element simulations and the limitations of data-driven models. This paper presents a physics-informed machine learning (PIML) framework for the efficient and reliable estimation of steady-state temperature profiles. The PIML approach integrates the governing heat transfer equations directly into the neural network's loss function, enabling high-fidelity predictions with significantly faster convergence than purely data-driven methods. The framework is evaluated on a dataset of varying cooling channel geometries. Results demonstrate that the PIML model converges more rapidly and achieves markedly higher accuracy, with a 49.1% reduction in mean squared error over the data-driven model. Validation against independent test cases further confirms its superior performance, particularly in regions away from the cooling channels. These findings underscore the potential of PIML for surrogate modeling and design optimization in battery systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This applies PINNs to pouch cell temperature estimation with varying cooling geometries and reports a 49% MSE drop versus data-driven baselines, but the gains rest on thin methodological details.

read the letter

The paper takes the standard physics-informed neural net setup and applies it to steady-state heat transfer in pouch cells with indirect liquid cooling and different channel layouts. It claims faster convergence and a 49.1% MSE reduction over a plain neural net on the same finite-element dataset, plus better accuracy away from the cooling channels on held-out cases. That is the core result worth noting. The physics residual in the loss appears to supply useful regularization for extrapolation, which is the practical payoff for surrogate modeling in battery design. The work is incremental rather than foundational, but the concrete engineering setting and the geometry variations make it a reasonable next step after the usual PINN demonstrations. What it does cleanly is tie the method to a real constraint (cooling channel placement) and show that the physics term helps where data is sparse. The soft spots are the missing specifics on network depth, loss weighting between data and physics terms, data splits, and whether the data-driven baseline received equivalent hyperparameter effort. Without those, the size of the reported gain is hard to interpret. The dataset comes from FEM, so any systematic bias in the simulations carries through; the paper would be stronger if it addressed that directly. This is aimed at engineers doing battery thermal management who need faster surrogates for design sweeps. A reader already working with scientific ML for heat transfer problems will see a usable example and can judge whether the approach transfers to their geometry. It is not a deep theoretical advance, but the comparison is straightforward enough that it merits referee time to check the implementation details and reproducibility.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a physics-informed machine learning (PIML) framework that embeds the steady-state heat transfer equation into the loss function of a neural network to predict temperature profiles in pouch cells with variable indirect liquid cooling channel geometries. Trained on finite-element simulation data, the PIML model is reported to converge faster and achieve a 49.1% reduction in mean squared error relative to a purely data-driven neural network baseline, with improved accuracy particularly in regions away from the cooling channels, as demonstrated on independent test cases.

Significance. If the quantitative claims are reproducible with full methodological transparency, the work would illustrate a practical benefit of physics-informed losses for surrogate modeling in battery thermal management, where FEM is computationally expensive and data-driven models can struggle with extrapolation. The emphasis on variable geometries and out-of-channel performance aligns with design-optimization needs, but the absence of architecture, weighting, and statistical details limits assessment of whether the reported gains are robust or generalizable beyond the specific FEM dataset.

major comments (2)

[Abstract / Results] Abstract and Results section: the central claim of a 49.1% MSE reduction is presented without any description of network architecture (depth, width, activation), loss weighting between data and physics residuals, training/validation/test split ratios, optimizer settings, or the precise configuration of the data-driven baseline. These omissions make it impossible to determine whether the improvement is attributable to the physics embedding or to differences in model capacity or hyper-parameters.
[Methods / Results] Methods and Results: no error bars, standard deviations across multiple random seeds, or statistical tests accompany the 49.1% figure or the convergence curves. With only a single scalar reported, it is unclear whether the performance difference is statistically significant or sensitive to the particular FEM mesh or cooling-channel parameter ranges used in the dataset.

minor comments (2)

[Abstract] The abstract states that the framework is 'evaluated on a dataset of varying cooling channel geometries' but provides no quantitative description of the parameter ranges or number of geometries; adding a table or figure summarizing the dataset statistics would improve clarity.
[Methods] Notation for the heat-equation residual and boundary conditions should be defined explicitly in the Methods section rather than assumed from standard heat-transfer literature, to allow readers to verify the exact form of the physics loss.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback, which has helped us improve the clarity and rigor of our work on the physics-informed machine learning framework for pouch cell temperature estimation. We have revised the manuscript to address the concerns about methodological transparency and statistical validation.

read point-by-point responses

Referee: [Abstract / Results] Abstract and Results section: the central claim of a 49.1% MSE reduction is presented without any description of network architecture (depth, width, activation), loss weighting between data and physics residuals, training/validation/test split ratios, optimizer settings, or the precise configuration of the data-driven baseline. These omissions make it impossible to determine whether the improvement is attributable to the physics embedding or to differences in model capacity or hyper-parameters.

Authors: We agree that the original manuscript omitted these essential implementation details, which hinders reproducibility and makes it difficult to isolate the contribution of the physics-informed loss. This was an oversight on our part. In the revised manuscript, we have added a dedicated subsection in the Methods section that fully specifies the network architecture (depth, width, and activation functions), the weighting coefficient between the data loss and physics residual terms, the training/validation/test split ratios, the optimizer and its hyperparameters, and explicit confirmation that the data-driven baseline employs an identical architecture and training configuration except for the removal of the physics term. These additions demonstrate that the reported 49.1% MSE reduction arises from the embedding of the steady-state heat transfer equation. revision: yes
Referee: [Methods / Results] Methods and Results: no error bars, standard deviations across multiple random seeds, or statistical tests accompany the 49.1% figure or the convergence curves. With only a single scalar reported, it is unclear whether the performance difference is statistically significant or sensitive to the particular FEM mesh or cooling-channel parameter ranges used in the dataset.

Authors: We acknowledge that presenting the 49.1% MSE reduction as a single scalar without error bars, standard deviations, or statistical tests limits evaluation of significance and robustness. We have revised the Results section to report performance metrics with error bars and standard deviations computed over multiple independent training runs using different random seeds. We have also incorporated a statistical test comparing the two models and added a sensitivity analysis examining variations in FEM mesh resolution and cooling-channel parameter ranges within the dataset. These updates show that the performance advantage of the PIML model is consistent across these factors. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper's core method embeds standard steady-state heat transfer equations (from first principles) as a residual term in the neural network loss function, which is independent of the FEM-generated training data and target temperature outputs. The reported 49.1% MSE improvement is an empirical comparison to a purely data-driven baseline on the same dataset, with no evidence that any prediction reduces to a fitted parameter by construction, that governing equations are self-defined from outputs, or that load-bearing steps rely on self-citations or ansatzes smuggled from prior author work. The approach follows standard PINN methodology without internal reduction to inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard heat transfer physics and neural network optimization; no new entities are postulated and no free parameters are explicitly fitted beyond typical training choices.

axioms (1)

domain assumption The steady-state heat transfer equations govern the temperature distribution in the pouch cells with liquid cooling.
Invoked when integrating governing equations into the neural network loss function as described in the abstract.

pith-pipeline@v0.9.0 · 5457 in / 1390 out tokens · 49758 ms · 2026-05-10T12:04:19.229073+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

20 extracted references

[1]

Life cycle assessment of hydrometallurgical recycling for cathode active materials,

Z. Liu, J. G. Sederholm, K.-W. Lan, E. J. Cho, M. J. Dipto, Y . Guru- mukhi, K. F. Rabbi, M. C. Hatzell, N. H. Perry, N. Miljkovic,et al., “Life cycle assessment of hydrometallurgical recycling for cathode active materials,”Journal of Power Sources, vol. 580, p. 233345, 2023

2023
[2]

Machine learning enhanced control co-design optimization of an immersion cooled battery thermal management system,

Z. Liu, P. Kabirzadeh, H. Wu, W. Fu, H. Qiu, N. Miljkovic, Y . Li, and P. Wang, “Machine learning enhanced control co-design optimization of an immersion cooled battery thermal management system,”Journal of Applied Physics, vol. 136, no. 2, 2024

2024
[3]

Pouch-type lithium secondary battery,

J.-H. Kim, “Pouch-type lithium secondary battery,” Apr. 10 2007. US Patent 7,201,998

2007
[4]

Designing reliable cathode system for high-performance inorganic solid-state pouch cells,

S. Wang, S. Liu, W. Chen, Y . Hu, D. Chen, M. He, M. Zhou, T. Lei, Y . Zhang, and J. Xiong, “Designing reliable cathode system for high-performance inorganic solid-state pouch cells,”Advanced Science, vol. 11, no. 23, p. 2401889, 2024

2024
[5]

The effect of temperature on capacity and power in cycled lithium ion batteries,

J. R. Belt, C. D. Ho, T. J. Miller, M. A. Habib, and T. Q. Duong, “The effect of temperature on capacity and power in cycled lithium ion batteries,”Journal of power sources, vol. 142, no. 1-2, pp. 354–360, 2005

2005
[6]

Integrating heat transfer and control optimization: A comprehensive review of battery thermal management systems,

P. Kabirzadeh, Z. Liu, M. Olyaei, H. Qiu, Y . Gurumukhi, H. Tyagi, Y . Jiang, V . S. Garimella, B. P. Singh, Y . Li,et al., “Integrating heat transfer and control optimization: A comprehensive review of battery thermal management systems,”Journal of Energy Storage, vol. 131, p. 117289, 2025

2025
[7]

Comprehensive comparison study on battery thermal management modules with indirect and direct liquid cooling,

B. Tong, J. Shi, M. Cao, W. Xuan, J. Chen, K. Jin, J. Sun, and Q. Wang, “Comprehensive comparison study on battery thermal management modules with indirect and direct liquid cooling,”Applied Thermal Engineering, vol. 268, p. 125945, 2025

2025
[8]

Lithium- ion pouch cells: An overview,

S. V . Gopinadh, V . Anoopkumar, M. J. N. Ansari, D. Srivastava, A. Raj M, B. John, A. Samridh, P. Vijayakumar, and T. Mercy, “Lithium- ion pouch cells: An overview,”Energy Harvesting and Storage: Funda- mentals and Materials, pp. 209–224, 2022

2022
[9]

Data-driven control co- design for indirect liquid cooling plate with microchannels for battery thermal management,

Z. Liu, Y . Xu, H. Wu, P. Wang, and Y . Li, “Data-driven control co- design for indirect liquid cooling plate with microchannels for battery thermal management,” inInternational Design Engineering Technical Conferences and Computers and Information in Engineering Confer- ence, vol. 87301, p. V03AT03A048, American Society of Mechanical Engineers, 2023

2023
[10]

Electrical and thermal active co-management for lithium-ion batteries,

Z. Zheng, Z. Liu, S. Kohtz, P. Wang, Y . Li, W. Fu, N. Miljkovic, and S. Smith, “Electrical and thermal active co-management for lithium-ion batteries,” in2022 IEEE Transportation Electrification Conference & Expo (ITEC), pp. 1159–1162, IEEE, 2022

2022
[11]

External liquid cooling method for lithium-ion battery modules under ultra-fast charging,

Y . Qin, Z. Xu, J. Du, H. Guo, L. Lu, and M. Ouyang, “External liquid cooling method for lithium-ion battery modules under ultra-fast charging,”IEEE Transactions on Industry Applications, vol. 58, no. 6, pp. 7658–7667, 2022

2022
[12]

Prognostic and health management for engineering systems: a review of the data-driven approach and algorithms,

T. Sutharssan, S. Stoyanov, C. Bailey, and C. Yin, “Prognostic and health management for engineering systems: a review of the data-driven approach and algorithms,”The Journal of engineering, vol. 2015, no. 7, pp. 215–222, 2015

2015
[13]

Physics-informed machine learning enhanced battery pack optimization,

Z. Liu, Y . Jiang, Y . Li, and P. Wang, “Physics-informed machine learning enhanced battery pack optimization,” in2025 IEEE/AIAA Transportation Electrification Conference and Electric Aircraft Technologies Sympo- sium (ITEC+ EATS), pp. 1–5, IEEE, 2025

2025
[14]

Physics-informed machine learning for battery pack thermal management,

Z. Liu, Y . Jiang, Y . Li, and P. Wang, “Physics-informed machine learning for battery pack thermal management,” in2025 Annual Reliability and Maintainability Symposium (RAMS), pp. 1–7, IEEE, 2025

2025
[15]

Multi-fidelity physics-informed convolutional neural network for heat map prediction of battery packs,

Y . Jiang, Z. Liu, P. Kabirzadeh, Y . Wu, Y . Li, N. Miljkovic, and P. Wang, “Multi-fidelity physics-informed convolutional neural network for heat map prediction of battery packs,”Reliability Engineering & System Safety, vol. 256, p. 110752, 2025

2025
[16]

Physics-informed machine learning,

G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, “Physics-informed machine learning,”Nature Reviews Physics, vol. 3, no. 6, pp. 422–440, 2021

2021
[17]

Measurement of two-dimensional heat generation rate of pouch type lithium-ion battery using a multifunc- tional calorimeter,

Y . Hu, S.-Y . Choe, and T. R. Garrick, “Measurement of two-dimensional heat generation rate of pouch type lithium-ion battery using a multifunc- tional calorimeter,”Journal of Power Sources, vol. 532, p. 231350, 2022

2022
[18]

Thermal performance of lithium-ion battery thermal management system by using mini-channel cooling,

Z. Qian, Y . Li, and Z. Rao, “Thermal performance of lithium-ion battery thermal management system by using mini-channel cooling,”Energy Conversion and Management, vol. 126, pp. 622–631, 2016

2016
[19]

A numerical analysis on multi-stage tesla valve based cold plate for cooling of pouch type li-ion batteries,

K. Monika, C. Chakraborty, S. Roy, R. Sujith, and S. P. Datta, “A numerical analysis on multi-stage tesla valve based cold plate for cooling of pouch type li-ion batteries,”International Journal of Heat and Mass Transfer, vol. 177, p. 121560, 2021

2021
[20]

Investigation of power battery thermal management by using mini-channel cold plate,

Y . Huo, Z. Rao, X. Liu, and J. Zhao, “Investigation of power battery thermal management by using mini-channel cold plate,”Energy Con- version and Management, vol. 89, pp. 387–395, 2015

2015