MagBridge-Battery: A Synthetic Bridge Dataset for Li-ion Magnetometry and State-of-Health Diagnostics

Prasanna Kumar Rangarajan; Sakthi Prabhu Gunasekar

arxiv: 2605.20240 · v1 · pith:MEBU6SCUnew · submitted 2026-05-17 · 💻 cs.LG

MagBridge-Battery: A Synthetic Bridge Dataset for Li-ion Magnetometry and State-of-Health Diagnostics

Sakthi Prabhu Gunasekar , Prasanna Kumar Rangarajan This is my paper

Pith reviewed 2026-05-21 07:33 UTC · model grok-4.3

classification 💻 cs.LG

keywords synthetic datasetbattery diagnosticsmagnetic sensingstate-of-healthLi-ion batteriesanomaly detectionbenchmarkmagnetometry

0 comments

The pith

A new synthetic dataset supplies 6760 magnetic signatures paired with battery state-of-health labels to support magnetic diagnostics.

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

The paper releases MagBridge-Battery v1.0 to address the shortage of public magnetic measurement data for lithium-ion batteries that is also labeled with degradation information. It constructs the dataset by conditioning real magnetic field morphologies with state-of-health values drawn from an existing electrochemical collection and adds controlled anomaly cases plus low-voltage extrapolations. Benchmark tasks are defined for regression of state of health, classification of second-life suitability, and anomaly detection, with a cell-disjoint split that avoids leakage. An ablation that shuffles the labels drives regression performance to zero, showing that the magnetic inputs carry non-trivial information about battery condition. This resource matters because magnetic sensing can reveal internal structural changes that terminal voltage measurements alone do not capture.

Core claim

We release MagBridge-Battery v1.0, a synthetic dataset of 6,760 magnetic-field signatures that bridges real magnetic morphology from the Mohammadi-Jerschow Open Science Framework archive with state-of-health labels from the PulseBat dataset. The release contains 5,600 PulseBat-conditioned grounded samples, 600 synthetic sensor-anomaly samples derived from clean parents, and 560 low-voltage Regime-B extrapolation samples. A cell-disjoint, parent-child-leakage-free primary benchmark split is verified to contain zero overlapping cells, zero cross-split parent-child pairs, and zero sample-ID overlap. We define three primary benchmark tasks: SOH regression, second-life classification, and anomaly

What carries the argument

The synthetic bridge that conditions real magnetic signatures with SOH labels from an electrochemical dataset while injecting controlled anomalies and enforcing cell-disjoint splits.

If this is right

SOH regression on the primary split reaches R squared of approximately 0.77.
The dataset directly supports second-life classification and anomaly detection benchmarks.
The verified cell-disjoint split eliminates train-test leakage from shared cells or parent samples.
A label-shuffle ablation confirms that performance relies on genuine input-label relationships rather than artifacts.

Where Pith is reading between the lines

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

The same bridging method could be applied to other scarce paired sensing modalities such as acoustic or optical battery monitoring.
Portable magnetic sensor prototypes could be trained on this data and then tested in field conditions on electric-vehicle fleets.
Future releases might add time-series evolution of the magnetic signatures across multiple charge-discharge cycles to enable predictive modeling.

Load-bearing premise

The synthetic conditioning and anomaly injection steps produce magnetic signatures whose statistical relationship to SOH matches what would be observed in actual paired magnetic-electrochemical experiments.

What would settle it

Simultaneous magnetic and electrochemical measurements on the same set of degraded cells, followed by direct comparison of the resulting statistical distributions against those in the synthetic dataset.

Figures

Figures reproduced from arXiv: 2605.20240 by Prasanna Kumar Rangarajan, Sakthi Prabhu Gunasekar.

**Figure 1.** Figure 1: MagBridge-Battery v1.0 composition. The dataset contains 6,760 samples: 5,600 grounded samples, 600 synthetic anomaly samples, and 560 low [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Figure 2: Bridge architecture. Real magnetic morphology from the OSF archive and SOH/SOC/U-feature labels from PulseBat are combined through a morphology [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Benchmark and validation protocol for MagBridge-Battery v1.0. The release is evaluated through integrity checks, bridge sanity invariants, distributional [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

read the original abstract

Battery health diagnostics today rely overwhelmingly on electrochemical signals measured at the cell terminals. A parallel literature has shown that magnetic sensing can resolve information that terminal-only measurements miss, but method development is limited by the absence, to the best of our knowledge, of public battery magnetic-measurement datasets paired with degradation labels. We release MagBridge-Battery v1.0, a synthetic dataset of 6,760 magnetic-field signatures that bridges real magnetic morphology from the Mohammadi-Jerschow Open Science Framework (OSF) archive with state-of-health (SOH) labels from the PulseBat dataset. The release contains 5,600 PulseBat-conditioned grounded samples, 600 synthetic sensor-anomaly samples derived from clean parents, and 560 low-voltage Regime-B extrapolation samples. A cell-disjoint, parent-child-leakage-free primary benchmark split is verified to contain zero overlapping cells, zero cross-split parent-child pairs, and zero sample-ID overlap. We define three primary benchmark tasks: SOH regression, second-life classification, and anomaly detection, plus an auxiliary anomaly-subtype classification task. A controlled label-shuffle ablation collapses SOH regression from R^2 approximately 0.77 to approximately 0, confirming that the bridge encodes input SOH non-trivially rather than producing label-aligned artifacts. The dataset is released on Zenodo under CC-BY-4.0, and the bridge code and benchmark suite are released under Apache-2.0. This work provides a public benchmark for magnetic-sensing battery diagnostics while paired magnetic-electrochemical measurements remain scarce.

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 is a careful dataset release that gives the first public synthetic bridge between magnetic morphology and battery SOH labels, with solid splits and a shuffle ablation, but the physical realism of the pairing remains untested.

read the letter

The main thing here is that the authors have released MagBridge-Battery v1.0, a synthetic set of 6760 magnetic signatures that links real morphology data from the Mohammadi-Jerschow archive to SOH labels from PulseBat. They include cell-disjoint splits with no parent-child leakage, plus 600 anomaly samples and some low-voltage extrapolations. The label-shuffle test is straightforward and shows SOH regression dropping from R^2 around 0.77 to zero, which rules out trivial artifacts and confirms the bridge carries non-random structure. The code and data are out under open licenses, which makes it immediately usable. That combination of a documented gap, verifiable splits, and an informative ablation is what the paper actually delivers. The soft spot is the lack of any real paired magnetic and electrochemical measurements on the same degraded cells. The synthetic conditioning and anomaly injection steps are described, but without simultaneous lab data the statistical mapping between signatures and SOH is assumed rather than checked. The paper flags this and leaves validation to downstream users, which is honest but means the benchmark's reliability as a proxy still needs to be shown. This is for researchers working on magnetic sensing for battery health who need a public starting point while paired measurements stay scarce. A reader developing regression or anomaly methods for non-terminal diagnostics would get concrete splits and tasks to try right away. It deserves a serious referee because the release is technically careful and the gap it fills is real. I would send it to review so the community can use the resource and test the transfer to actual paired experiments as those data appear.

Referee Report

1 major / 2 minor

Summary. The paper claims to release MagBridge-Battery v1.0, a synthetic dataset of 6,760 magnetic-field signatures that bridges real magnetic morphology from the Mohammadi-Jerschow OSF archive with state-of-health (SOH) labels from the PulseBat dataset. It provides 5,600 PulseBat-conditioned grounded samples, 600 synthetic sensor-anomaly samples, and 560 low-voltage Regime-B extrapolation samples, with cell-disjoint, leakage-free benchmark splits. Three primary tasks are defined (SOH regression, second-life classification, anomaly detection) plus an auxiliary anomaly-subtype task. A label-shuffle ablation is included showing SOH regression R² collapsing from ~0.77 to ~0.

Significance. If the synthetic conditioning faithfully reproduces the statistical relationships between magnetic signatures and SOH that would be observed in real paired experiments, the release would supply a valuable public benchmark for magnetic-sensing battery diagnostics where simultaneous measurements remain scarce. Credit is due for the open data release on Zenodo (CC-BY-4.0), the Apache-2.0 code and benchmark suite, the explicit verification of zero cell overlap and parent-child leakage across splits, and the controlled label-shuffle ablation that supplies direct evidence of non-trivial SOH encoding.

major comments (1)

[Dataset Generation] The dataset-generation description invokes the premise that conditioning and anomaly injection produce magnetic signatures whose statistical relationship to SOH matches what simultaneous real measurements on degraded cells would yield, yet no such paired real data is used for validation. This assumption is load-bearing for the claim that the released splits and benchmarks (e.g., SOH regression at R² ~0.77) can serve as reliable proxies for method development.

minor comments (2)

[Abstract] The abstract reports R² as 'approximately 0.77'; stating the precise value together with the number of runs or standard deviation would improve quantitative clarity.
[Benchmark Splits] A compact table summarizing split statistics (unique cells, samples per split, and leakage checks) would make the 'cell-disjoint, parent-child-leakage-free' claim easier to inspect at a glance.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive review, positive recognition of the open release, leakage-free splits, and label-shuffle ablation, and for identifying the key assumption in the dataset generation process. We respond to the major comment below.

read point-by-point responses

Referee: [Dataset Generation] The dataset-generation description invokes the premise that conditioning and anomaly injection produce magnetic signatures whose statistical relationship to SOH matches what simultaneous real measurements on degraded cells would yield, yet no such paired real data is used for validation. This assumption is load-bearing for the claim that the released splits and benchmarks (e.g., SOH regression at R² ~0.77) can serve as reliable proxies for method development.

Authors: We agree that direct validation against simultaneously acquired paired magnetic and electrochemical data on identical degraded cells would provide the strongest possible confirmation. Such paired public datasets do not currently exist, which is the central motivation for constructing the synthetic bridge. The generation procedure maps real magnetic morphologies drawn from the Mohammadi-Jerschow OSF archive onto SOH trajectories taken from PulseBat by applying established physical relationships between lithium distribution, current density, and the resulting external magnetic field. The controlled label-shuffle ablation already demonstrates that the generated signatures carry non-trivial SOH information rather than label-aligned artifacts. We will revise the manuscript to include an expanded limitations subsection that explicitly states the synthetic nature of the conditioning, the physical assumptions employed, and the absence of paired real-data validation, thereby clarifying the scope within which the benchmarks should be interpreted as proxies. revision: partial

Circularity Check

0 steps flagged

No circularity: dataset release from independent external sources

full rationale

The paper releases a synthetic dataset constructed by bridging two independent public external datasets (magnetic morphology from Mohammadi-Jerschow OSF archive and SOH labels from PulseBat). The label-shuffle ablation is a statistical control independent of any fitted parameters or author-defined quantities. No equations, predictions, or load-bearing steps reduce by construction to inputs defined within the paper or via self-citation chains. The work is self-contained as a data-release contribution with externally sourced inputs and falsifiable downstream benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper introduces no new free parameters or invented physical entities. It relies on two existing public datasets and standard synthetic generation steps for anomalies and regime extrapolation.

axioms (1)

domain assumption Magnetic morphology recorded on one set of cells can be realistically conditioned on SOH labels from a different set of cells to produce usable training examples.
This bridging premise is required for the entire synthetic construction and is stated in the dataset-generation section of the abstract.

pith-pipeline@v0.9.0 · 5824 in / 1527 out tokens · 80295 ms · 2026-05-21T07:33:34.709210+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

The degradation modulator applies SOH-driven perturbation to the base morphology through a learned latent representation... LDA subspace... cone-restricted k-NN softmin retrieval... blend the decoded morphology with the base sample proportionally to (1−SOH).
IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean absolute_floor_iff_bare_distinguishability unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

A label-shuffle ablation collapses SOH regression from R² ≈0.77 to ≈0

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages

[1]

and Attia, Peter M

Severson, Kristen A. and Attia, Peter M. and Jin, Norman and Perkins, Nicholas and Jiang, Benben and Yang, Zi and Chen, Michael H. and Aykol, Muratahan and Herring, Patrick K. and Fraggedakis, Dimitrios and Bazant, Martin Z. and Harris, Stephen J. and Chueh, William C. and Braatz, Richard D. , title =. Nature Energy , volume =. 2019 , doi =

work page 2019
[2]

Nature Communications , volume =

Tao, Shengyu and Ma, Ruifei and Zhao, Zixi and Ma, Guangyuan and Su, Lin and Chang, Heng and Chen, Yuou and Liu, Haizhou and Liang, Zheng and Cao, Tingwei and Ji, Haocheng and Han, Zhiyuan and Lu, Minyan and Yang, Huixiong and Wen, Zongguo and Yao, Jianhua and Yu, Rong and Wei, Guodan and Li, Yang and Zhang, Xuan and Xu, Tingyang and Zhou, Guangmin , titl...

work page 2024
[3]

Lithium-ion battery data and where to find it , journal =

dos Reis, Gon. Lithium-ion battery data and where to find it , journal =. 2021 , doi =

work page 2021
[4]

and Tarascon, Jean-Marie , title =

Huang, Jiaqiang and Boles, Steven T. and Tarascon, Jean-Marie , title =. Nature Sustainability , volume =. 2022 , doi =

work page 2022
[5]

eTransportation , volume =

Lu, Xibin and Tarascon, Jean-Marie and Huang, Jiaqiang , title =. eTransportation , volume =. 2022 , doi =

work page 2022
[6]

and Mohammadi, Mohaddese and Schauerman, Christopher M

Ilott, Andrew J. and Mohammadi, Mohaddese and Schauerman, Christopher M. and Ganter, Matthew J. and Jerschow, Alexej , title =. Nature Communications , volume =. 2018 , doi =

work page 2018
[7]

Proceedings of the National Academy of Sciences , volume =

Romanenko, Konstantin and Jerschow, Alexej , title =. Proceedings of the National Academy of Sciences , volume =. 2019 , doi =

work page 2019
[8]

and Jerschow, Alexej , title =

Romanenko, Konstantin and Kuchel, Philip W. and Jerschow, Alexej , title =. Chemistry of Materials , volume =. 2020 , doi =

work page 2020
[9]

and Ilott, Andrew J

Mohammadi, Mohaddese and Silletta, Emilia V. and Ilott, Andrew J. and Jerschow, Alexej , title =. Journal of Magnetic Resonance , volume =. 2019 , doi =

work page 2019
[10]

and Mohammadi, Mohaddese and Silletta, Emilia V

Hu, Yinan and Iwata, Geoffrey Z. and Mohammadi, Mohaddese and Silletta, Emilia V. and Wickenbrock, Arne and Blanchard, John W. and Budker, Dmitry and Jerschow, Alexej , title =. Proceedings of the National Academy of Sciences , volume =. 2020 , doi =

work page 2020
[11]

and Jerschow, Alexej , title =

Ilott, Andrew J. and Jerschow, Alexej , title =. Scientific Reports , volume =. 2017 , doi =

work page 2017
[12]

Journal of Power Sources , volume =

Brauchle, Felix and Grimsmann, Florian and von Kessel, Otto and Birke, Kai Peter , title =. Journal of Power Sources , volume =. 2021 , doi =

work page 2021
[13]

and Rupp, Jennifer L

Pollok, Stefan and Khoshkalam, Mohamad and Ghaffari-Tabrizi, Fardin and Kurnia, Fran and Wang, Danni and Li, Siqi and Bucher, Dominik B. and Rupp, Jennifer L. M. and Christensen, Dennis Valbj. Magnetic microscopy for operando imaging of battery dynamics , journal =. 2025 , doi =

work page 2025
[14]

Evans, William and Coussens, Thomas and Woodley, Mark T. M. and Fabricant, A. M. and Kendall, G. D. and Sonnet, M. and Wasylowski, D. and Sauer, Dirk Uwe and Oru. Quantum magnetic imaging of current density in lithium-ion batteries , journal =. 2025 , doi =. 2512.01125 , archivePrefix =

work page arXiv 2025
[15]

2020 , doi =

Mohammadi, Mohaddese and Jerschow, Alexej , title =. 2020 , doi =

work page 2020
[16]

Physical Review Applied , volume =

Fujii, Keisuke and Nakajima, Kohei , title =. Physical Review Applied , volume =. 2017 , doi =

work page 2017
[17]

, title =

Birkl, Christoph R. , title =. 2017 , doi =

work page 2017
[18]

Energy & Environmental Science , volume =

Ma, Guijun and Xu, Songpei and Jiang, Chen and Cheng, Yujie and Yang, Xiaosong and Shen, Yu and Chen, Zonghai , title =. Energy & Environmental Science , volume =. 2022 , doi =

work page 2022
[19]

Nature Communications , volume =

Lu, Jiahuan and Xiong, Rui and Tian, Jinpeng and Wang, Chenxu and Sun, Fengchun , title =. Nature Communications , volume =. 2023 , doi =

work page 2023
[20]

Nature Communications , volume =

Wang, Fujin and Zhai, Zhi and Zhao, Zhibin and Di, Yi and Chen, Xuefeng , title =. Nature Communications , volume =. 2024 , doi =

work page 2024
[21]

Journal of Energy Storage , volume =

Wang, Fujin and Zhai, Zhi and Liu, Bin and Chen, Xuefeng , title =. Journal of Energy Storage , volume =. 2024 , doi =

work page 2024

[1] [1]

and Attia, Peter M

Severson, Kristen A. and Attia, Peter M. and Jin, Norman and Perkins, Nicholas and Jiang, Benben and Yang, Zi and Chen, Michael H. and Aykol, Muratahan and Herring, Patrick K. and Fraggedakis, Dimitrios and Bazant, Martin Z. and Harris, Stephen J. and Chueh, William C. and Braatz, Richard D. , title =. Nature Energy , volume =. 2019 , doi =

work page 2019

[2] [2]

Nature Communications , volume =

Tao, Shengyu and Ma, Ruifei and Zhao, Zixi and Ma, Guangyuan and Su, Lin and Chang, Heng and Chen, Yuou and Liu, Haizhou and Liang, Zheng and Cao, Tingwei and Ji, Haocheng and Han, Zhiyuan and Lu, Minyan and Yang, Huixiong and Wen, Zongguo and Yao, Jianhua and Yu, Rong and Wei, Guodan and Li, Yang and Zhang, Xuan and Xu, Tingyang and Zhou, Guangmin , titl...

work page 2024

[3] [3]

Lithium-ion battery data and where to find it , journal =

dos Reis, Gon. Lithium-ion battery data and where to find it , journal =. 2021 , doi =

work page 2021

[4] [4]

and Tarascon, Jean-Marie , title =

Huang, Jiaqiang and Boles, Steven T. and Tarascon, Jean-Marie , title =. Nature Sustainability , volume =. 2022 , doi =

work page 2022

[5] [5]

eTransportation , volume =

Lu, Xibin and Tarascon, Jean-Marie and Huang, Jiaqiang , title =. eTransportation , volume =. 2022 , doi =

work page 2022

[6] [6]

and Mohammadi, Mohaddese and Schauerman, Christopher M

Ilott, Andrew J. and Mohammadi, Mohaddese and Schauerman, Christopher M. and Ganter, Matthew J. and Jerschow, Alexej , title =. Nature Communications , volume =. 2018 , doi =

work page 2018

[7] [7]

Proceedings of the National Academy of Sciences , volume =

Romanenko, Konstantin and Jerschow, Alexej , title =. Proceedings of the National Academy of Sciences , volume =. 2019 , doi =

work page 2019

[8] [8]

and Jerschow, Alexej , title =

Romanenko, Konstantin and Kuchel, Philip W. and Jerschow, Alexej , title =. Chemistry of Materials , volume =. 2020 , doi =

work page 2020

[9] [9]

and Ilott, Andrew J

Mohammadi, Mohaddese and Silletta, Emilia V. and Ilott, Andrew J. and Jerschow, Alexej , title =. Journal of Magnetic Resonance , volume =. 2019 , doi =

work page 2019

[10] [10]

and Mohammadi, Mohaddese and Silletta, Emilia V

Hu, Yinan and Iwata, Geoffrey Z. and Mohammadi, Mohaddese and Silletta, Emilia V. and Wickenbrock, Arne and Blanchard, John W. and Budker, Dmitry and Jerschow, Alexej , title =. Proceedings of the National Academy of Sciences , volume =. 2020 , doi =

work page 2020

[11] [11]

and Jerschow, Alexej , title =

Ilott, Andrew J. and Jerschow, Alexej , title =. Scientific Reports , volume =. 2017 , doi =

work page 2017

[12] [12]

Journal of Power Sources , volume =

Brauchle, Felix and Grimsmann, Florian and von Kessel, Otto and Birke, Kai Peter , title =. Journal of Power Sources , volume =. 2021 , doi =

work page 2021

[13] [13]

and Rupp, Jennifer L

Pollok, Stefan and Khoshkalam, Mohamad and Ghaffari-Tabrizi, Fardin and Kurnia, Fran and Wang, Danni and Li, Siqi and Bucher, Dominik B. and Rupp, Jennifer L. M. and Christensen, Dennis Valbj. Magnetic microscopy for operando imaging of battery dynamics , journal =. 2025 , doi =

work page 2025

[14] [14]

Evans, William and Coussens, Thomas and Woodley, Mark T. M. and Fabricant, A. M. and Kendall, G. D. and Sonnet, M. and Wasylowski, D. and Sauer, Dirk Uwe and Oru. Quantum magnetic imaging of current density in lithium-ion batteries , journal =. 2025 , doi =. 2512.01125 , archivePrefix =

work page arXiv 2025

[15] [15]

2020 , doi =

Mohammadi, Mohaddese and Jerschow, Alexej , title =. 2020 , doi =

work page 2020

[16] [16]

Physical Review Applied , volume =

Fujii, Keisuke and Nakajima, Kohei , title =. Physical Review Applied , volume =. 2017 , doi =

work page 2017

[17] [17]

, title =

Birkl, Christoph R. , title =. 2017 , doi =

work page 2017

[18] [18]

Energy & Environmental Science , volume =

Ma, Guijun and Xu, Songpei and Jiang, Chen and Cheng, Yujie and Yang, Xiaosong and Shen, Yu and Chen, Zonghai , title =. Energy & Environmental Science , volume =. 2022 , doi =

work page 2022

[19] [19]

Nature Communications , volume =

Lu, Jiahuan and Xiong, Rui and Tian, Jinpeng and Wang, Chenxu and Sun, Fengchun , title =. Nature Communications , volume =. 2023 , doi =

work page 2023

[20] [20]

Nature Communications , volume =

Wang, Fujin and Zhai, Zhi and Zhao, Zhibin and Di, Yi and Chen, Xuefeng , title =. Nature Communications , volume =. 2024 , doi =

work page 2024

[21] [21]

Journal of Energy Storage , volume =

Wang, Fujin and Zhai, Zhi and Liu, Bin and Chen, Xuefeng , title =. Journal of Energy Storage , volume =. 2024 , doi =

work page 2024