pith. sign in

arxiv: 2606.12226 · v1 · pith:TEUCWVLSnew · submitted 2026-06-10 · 💻 cs.CV · eess.IV

An Electric Potential-Augmented Benchmark Dataset for Physics-Guided Image Reconstruction of Electrical Capacitance Tomography

Xinqi Zhang , Qiming Ma , Lihui Peng This is my paper

Pith reviewed 2026-06-27 10:21 UTC · model grok-4.3

classification 💻 cs.CV eess.IV
keywords electrical capacitance tomographyelectric potential mapsphysics-guided machine learningimage reconstructionbenchmark datasetsoft-field effectdeep learning
0
0 comments X

The pith

Including electric potential maps in ECT training data improves deep learning reconstruction accuracy and robustness.

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

The paper creates a benchmark dataset for electrical capacitance tomography that supplies not only capacitance measurements and permittivity images but also the electric potential fields that physically link them. Standard deep learning methods map capacitances straight to images while ignoring this field, which governs the nonlinear soft-field behavior of the sensor. The new dataset supplies 20,000 simulated samples that each contain eight full-field potential maps generated for an eight-electrode sensor across four flow patterns. Tests on both in-distribution and out-of-distribution cases show that models given access to these maps achieve higher accuracy and better robustness than models that receive only capacitances. The explicit field data lowers the barrier to embedding physical laws directly into machine-learning pipelines for ECT image reconstruction.

Core claim

The paper presents an electric potential-augmented benchmark dataset that supplies eight excitation-wise full-field potential maps together with conventional capacitance vectors and permittivity images. Comprehensive evaluation on both in-distribution and out-of-distribution scenarios demonstrates that models using the added potential information produce more accurate and more robust reconstructions of permittivity distributions than models trained on capacitance data alone.

What carries the argument

The electric potential-augmented dataset, which preserves eight full-field potential maps per sample as the explicit physical link between capacitance measurements and permittivity distributions.

If this is right

  • Models that receive potential maps achieve higher reconstruction accuracy than capacitance-only baselines.
  • Robustness improves under out-of-distribution flow patterns when potential information is supplied.
  • The dataset supplies standardized evaluation protocols for both the forward and inverse problems of ECT.
  • Explicit field data reduces the effort required to embed physical constraints into machine-learning models for ECT.

Where Pith is reading between the lines

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

  • The same augmentation approach could be applied to other tomographic modalities that rely on soft-field measurements.
  • Hybrid architectures could use the supplied potential maps to enforce consistency with Maxwell's equations during training.
  • Extending the dataset with experimental measurements from real sensors would test whether simulation-derived gains transfer to hardware.

Load-bearing premise

Simulations of the eight-electrode sensor generate potential maps and capacitance values that match the latent physics of real ECT systems.

What would settle it

Train identical networks on the dataset both with and without the potential maps, then measure reconstruction error on measurements collected from physical ECT hardware.

Figures

Figures reproduced from arXiv: 2606.12226 by Lihui Peng, Qiming Ma, Xinqi Zhang.

Figure 1
Figure 1. Figure 1: Geometric configuration of the eight-electrode ECT sensor. 3 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Representative permittivity distributions of the proposed electric potential-augmented ECT dataset. The images depict two typical samples for each of the four flow patterns, namely single-bar, two-bar, annular, and stratified flows, shown on the full 78 × 78 computational grid [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Normalized capacitance vectors of the four representative samples shown in the first row of figure 2 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: presents the excitation-wise electric potential fields for two representative flow patterns, namely single-bar and stratified flow, corresponding to the first and fourth samples in the first row of figure 2. The left panel shows the permittivity distribution, and the right panels show the eight potential maps under sequential electrode excitations with equipotential contours overlaid. The local deflection … view at source ↗
Figure 5
Figure 5. Figure 5: Statistical and field-response diagnostics of the proposed dataset. (a) Average spatial occupancy of the high-permittivity phase for the four flow patterns, summarizing the statistical coverage of the stochastic geometry generators. (b) Potential distortion relative to the empty-pipe reference for representative samples from the four flow patterns. Bright white lines are overlaid to show the outlines of th… view at source ↗
Figure 6
Figure 6. Figure 6: Network architectures used for the potential-effectiveness experiments on forward and inverse ECT problems. The upper part shows the forward models, which predict the normalized 28-dimensional capacitance vector from the permittivity distribution. The lower part shows the inverse models, which reconstruct the permittivity distribution from the normalized capacitance vector. To isolate the role of electric … view at source ↗
Figure 7
Figure 7. Figure 7: Inverse reconstruction examples for representative stratified and two-bar samples. For each sample, the ground-truth permittivity distribution is compared with the reconstructed results of Models A–D [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Forward prediction results for representative stratified and OOD two-bar samples. For each sample, the permittivity distribution is shown on the left, while the normalized capacitance-vector comparison and the component-wise absolute errors are overlaid on the right. Line plots denote the ground-truth and predicted normalized capacitance vectors, and grouped bars denote the absolute errors over the 28 non-… view at source ↗
read the original abstract

While deep learning has significantly advanced image reconstruction of Electrical Capacitance Tomography (ECT), most data-driven methods map directly between capacitance and permittivity distribution, treating the sensor as a black box. This overlooks the electric potential field -- the fundamental physical link governing the nonlinear and ill-posed ``soft-field'' effect. To address this, we propose an electric potential-augmented ECT benchmark dataset designed to explicitly integrate latent physics behind ECT into the learning process. Generated via a COMSOL-MATLAB pipeline for an eight-electrode sensor as an example, the dataset comprises 20,000 randomized samples across four typical flow patterns. Crucially, alongside the conventional capacitance vectors and permittivity distributions depicted as images, each sample preserves eight excitation-wise full-field potential maps. Beyond data release, we provide illustrative evaluation protocols for both forward and inverse problems of ECT. Through comprehensive testing on both in-distribution (IID) and out-of-distribution (OOD) scenarios, we systematically demonstrate how the inclusion of electric potential maps enhances modeling accuracy and robustness. Fundamentally, the explicit inclusion of latent field information significantly lowers the barrier to integrating physical laws into ECT modeling, thereby establishing a standardized foundation for future physics-guided machine learning of ECT image reconstruction.

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.

Referee Report

1 major / 2 minor

Summary. The manuscript introduces an electric potential-augmented benchmark dataset for Electrical Capacitance Tomography (ECT) image reconstruction. Using a COMSOL-MATLAB pipeline for an eight-electrode sensor, it generates 20,000 randomized samples across four typical flow patterns, including capacitance vectors, permittivity distributions, and eight excitation-wise full-field potential maps. The paper provides illustrative evaluation protocols for forward and inverse problems and demonstrates through IID and OOD testing that including electric potential maps enhances modeling accuracy and robustness.

Significance. If the simulation pipeline faithfully represents real ECT physics and the reported improvements are reproducible, this dataset establishes a standardized foundation for physics-guided machine learning in ECT by explicitly incorporating the latent electric potential field. This could significantly advance the integration of physical laws into data-driven models for this ill-posed inverse problem.

major comments (1)
  1. [Abstract] Abstract: The central claim that 'the inclusion of electric potential maps enhances modeling accuracy and robustness' is asserted without quantitative results, model architectures, error bars, statistical tests, or exclusion criteria. This renders the strength of the IID/OOD demonstration unverifiable from the summary and places the load-bearing evidence entirely on the (unseen here) evaluation protocols section.
minor comments (2)
  1. The four typical flow patterns are referenced but not named or illustrated; adding a brief description or figure would improve clarity for readers unfamiliar with ECT.
  2. The COMSOL-MATLAB pipeline is described at a high level; a short paragraph on mesh convergence or boundary condition choices would help users assess simulation fidelity without requiring them to re-run the code.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive review and the recommendation of minor revision. We address the single major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that 'the inclusion of electric potential maps enhances modeling accuracy and robustness' is asserted without quantitative results, model architectures, error bars, statistical tests, or exclusion criteria. This renders the strength of the IID/OOD demonstration unverifiable from the summary and places the load-bearing evidence entirely on the (unseen here) evaluation protocols section.

    Authors: The abstract serves as a high-level summary of the manuscript contributions and key findings, consistent with standard scientific writing conventions. The full quantitative results—including specific accuracy and robustness metrics with error bars, the model architectures employed, statistical tests, and detailed evaluation protocols for both IID and OOD scenarios—are presented in the evaluation protocols and results sections of the manuscript. We acknowledge that incorporating select quantitative highlights into the abstract could improve immediate verifiability of the central claim. We will therefore revise the abstract to include key numerical results demonstrating the reported improvements. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper is a data-release manuscript whose core contribution is a COMSOL-MATLAB-generated benchmark dataset of 20,000 samples that includes capacitance vectors, permittivity images, and eight excitation-wise potential maps. No derivation chain, fitted parameters, or uniqueness theorems appear in the provided text. The claimed benefit of including potential maps is shown via internal IID/OOD evaluation protocols on the released data itself; these protocols do not reduce to self-referential equations or self-citations that would force the result. The work is therefore self-contained against external benchmarks and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the assumption that the simulation pipeline accurately captures real ECT physics and that the potential maps constitute the key latent information missing from prior datasets.

axioms (1)
  • domain assumption The COMSOL-MATLAB pipeline for an eight-electrode sensor generates potential maps and capacitance values that are representative of physical ECT behavior.
    All 20,000 samples and the claimed benefits of the potential maps rest on the fidelity of this simulation.

pith-pipeline@v0.9.1-grok · 5755 in / 1197 out tokens · 42228 ms · 2026-06-27T10:21:43.178027+00:00 · methodology

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

31 extracted references · 31 canonical work pages

  1. [1]

    Huang S M, Plaskowski A B, Xie C G and Beck M S 1989 Tomographic imaging of two-component flow using capacitance sensorsJ. Phys. E: Sci. Instrum.22173–177 (doi: 10.1088/0022-3735/22/3/009)

    work page doi:10.1088/0022-3735/22/3/009 1989

  2. [2]

    G: Circuits Devices Syst

    Xie C G, Huang S M, Beck M S, Hoyle B S, Thorn R, Lenn C and Snowden D 1992 Electrical capacitance tomography for flow imaging: system model for development of image reconstruction algorithms and design of primary sensorsIEE Proc. G: Circuits Devices Syst. 13989–98 (doi: 10.1049/ip-g-2.1992.0015)

    work page doi:10.1049/ip-g-2.1992.0015 1992

  3. [3]

    Reinecke N and Mewes D 1996 Recent developments and industrial/research applications of capacitance tomographyMeas. Sci. Technol.7233–246 (doi: 10.1088/0957-0233/7/3/004)

    work page doi:10.1088/0957-0233/7/3/004 1996

  4. [4]

    Marashdeh Q, Fan L-S, Du B and Warsito W 2008 Electrical capacitance tomography – a perspectiveInd. Eng. Chem. Res.473708–3719 (doi: 10.1021/ie0713590)

    work page doi:10.1021/ie0713590 2008

  5. [5]

    Yang W 2010 Design of electrical capacitance tomography sensorsMeas. Sci. Technol.21 042001 (doi: 10.1088/0957-0233/21/4/042001)

    work page doi:10.1088/0957-0233/21/4/042001 2010

  6. [6]

    Isaksen Ø 1996 A review of reconstruction techniques for capacitance tomographyMeas. Sci. Technol.7325–337 (doi: 10.1088/0957-0233/7/3/013)

    work page doi:10.1088/0957-0233/7/3/013 1996

  7. [7]

    Yang W Q and Peng L 2003 Image reconstruction algorithms for electrical capacitance tomographyMeas. Sci. Technol.14R1–R13 (doi: 10.1088/0957-0233/14/1/201)

    work page doi:10.1088/0957-0233/14/1/201 2003

  8. [8]

    Yang W Q, Spink D M, York T A and McCann H 1999 An image-reconstruction algorithm based on Landweber’s iteration method for electrical-capacitance tomographyMeas. Sci. Technol.101065–1069 (doi: 10.1088/0957-0233/10/11/315)

    work page doi:10.1088/0957-0233/10/11/315 1999

  9. [9]

    Peng L, Merkus H and Scarlett B 2000 Using regularization methods for image reconstruction of electrical capacitance tomographyPart. Part. Syst. Charact.1796–104 (doi: 10.1002/1521-4117(200010)17:3<96::AID-PPSC96>3.0.CO;2-8)

    work page doi:10.1002/1521-4117(200010)17:3 2000

  10. [10]

    Fang W 2004 A nonlinear image reconstruction algorithm for electrical capacitance tomographyMeas. Sci. Technol.152124–2132 (doi: 10.1088/0957-0233/15/10/023)

    work page doi:10.1088/0957-0233/15/10/023 2004

  11. [11]

    Soleimani M and Lionheart W R B 2005 Nonlinear image reconstruction for electrical capacitance tomography using experimental dataMeas. Sci. Technol.161987–1996 (doi: 10.1088/0957-0233/16/10/014)

    work page doi:10.1088/0957-0233/16/10/014 2005

  12. [12]

    Li Y and Yang W Q 2008 Image reconstruction by nonlinear Landweber iteration for complicated distributionsMeas. Sci. Technol.19094014 (doi: 10.1088/0957-0233/19/9/094014)

    work page doi:10.1088/0957-0233/19/9/094014 2008

  13. [13]

    Watzenig D and Fox C 2009 A review of statistical modelling and inference for electrical capacitance tomographyMeas. Sci. Technol.20052002 (doi: 10.1088/0957-0233/20/5/052002)

    work page doi:10.1088/0957-0233/20/5/052002 2009

  14. [14]

    Zhao J, Xu Y, Tan C and Dong F 2014 A fast sparse reconstruction algorithm for electrical tomographyMeas. Sci. Technol.25085401 (doi: 10.1088/0957-0233/25/8/085401)

    work page doi:10.1088/0957-0233/25/8/085401 2014

  15. [15]

    Ye J, Wang H and Yang W 2015 Image reconstruction for electrical capacitance tomography based on sparse representationIEEE Trans. Instrum. Meas.6489–102 (doi: 10.1109/TIM.2014.2329738) 12 IOP PublishingJournalvv(yyyy) aaaaaa Authoret al

    work page doi:10.1109/tim.2014.2329738 2015

  16. [16]

    Nooralahiyan A Y and Hoyle B S 1997 Three-component tomographic flow imaging using artificial neural network reconstructionChem. Eng. Sci.522139–2148 (doi: 10.1016/S0009-2509(97)00040-7)

    work page doi:10.1016/s0009-2509(97)00040-7 1997

  17. [17]

    J.185464–5474 (doi: 10.1109/JSEN.2018.2836337)

    Zheng J and Peng L 2018 An autoencoder-based image reconstruction for electrical capacitance tomographyIEEE Sens. J.185464–5474 (doi: 10.1109/JSEN.2018.2836337)

    work page doi:10.1109/jsen.2018.2836337 2018

  18. [18]

    Zheng J, Ma H and Peng L 2019 A CNN-based image reconstruction for electrical capacitance tomography2019 IEEE International Conference on Imaging Systems and Techniques (IST)(doi: 10.1109/IST48021.2019.9010096)

    work page doi:10.1109/ist48021.2019.9010096 2019

  19. [19]

    J.204879–4890 (doi: 10.1109/JSEN.2020.2965731)

    Zheng J and Peng L 2020 A deep learning compensated back projection for image reconstruction of electrical capacitance tomographyIEEE Sens. J.204879–4890 (doi: 10.1109/JSEN.2020.2965731)

    work page doi:10.1109/jsen.2020.2965731 2020

  20. [20]

    Yang X, Zhao C, Chen B, Zhang M and Li Y 2019 Big data driven U-Net based electrical capacitance image reconstruction algorithm2019 IEEE International Conference on Imaging Systems and Techniques (IST)(doi: 10.1109/IST48021.2019.9010423)

    work page doi:10.1109/ist48021.2019.9010423 2019

  21. [21]

    Zhu H, Sun J, Xu L, Tian W and Sun S 2020 Permittivity reconstruction in electrical capacitance tomography based on visual representation of deep neural networkIEEE Sens. J. 204803–4815 (doi: 10.1109/JSEN.2020.2964559)

    work page doi:10.1109/jsen.2020.2964559 2020

  22. [22]

    Deabes W, Abdel-Hakim A E, Bouazza K E and Althobaiti H 2022 Adversarial resolution enhancement for electrical capacitance tomography image reconstructionSensors223142 (doi: 10.3390/s22093142)

    work page doi:10.3390/s22093142 2022

  23. [23]

    Wanta D, Smolik A, Smolik W T, Midura M and Wr´ oblewski P 2025 Image reconstruction using machine-learned pseudoinverse in electrical capacitance tomographyEng. Appl. Artif. Intell.142109888 (doi: 10.1016/j.engappai.2024.109888)

    work page doi:10.1016/j.engappai.2024.109888 2025

  24. [24]

    Peng L, Yang Y, Li Y, Zhang M, Wang H and Yang W 2025 Deep learning-based image reconstruction for electrical capacitance tomographyMeas. Sci. Technol.36062003 (doi: 10.1088/1361-6501/add8ad)

    work page doi:10.1088/1361-6501/add8ad 2025

  25. [25]

    Zheng J, Li J, Li Y and Peng L 2018 A benchmark dataset and deep learning-based image reconstruction for electrical capacitance tomographySensors183701 (doi: 10.3390/s18113701)

    work page doi:10.3390/s18113701 2018

  26. [26]

    Data(doi: 10.1038/s41597-026-07433-7)

    Shi D, Wang Y, Li X, Sun T, Liu H and Guo D 2026 An electrical capacitance tomography dataset for image reconstruction benchmarkingSci. Data(doi: 10.1038/s41597-026-07433-7)

    work page doi:10.1038/s41597-026-07433-7 2026

  27. [27]

    Jin Y, Li Y, Zhang M and Peng L 2024 A physics-constrained deep learning-based image reconstruction for electrical capacitance tomographyIEEE Trans. Instrum. Meas.73 4500612 (doi: 10.1109/TIM.2023.3338673)

    work page doi:10.1109/tim.2023.3338673 2024

  28. [28]

    Soft Comput.155111436 (doi: 10.1016/j.asoc.2024.111436)

    Lei J and Liu Q 2024 Data and measurement mechanism integrated imaging method for electrical capacitance tomographyAppl. Soft Comput.155111436 (doi: 10.1016/j.asoc.2024.111436)

    work page doi:10.1016/j.asoc.2024.111436 2024

  29. [29]

    Loser T, Wajman R and Mewes D 2001 Electrical capacitance tomography: image reconstruction along electrical field linesMeas. Sci. Technol.121083–1091 (doi: 10.1088/0957-0233/12/8/314)

    work page doi:10.1088/0957-0233/12/8/314 2001

  30. [30]

    Wang S, Li Y, Chen Z, Kong M and Yang Y 2024 Digital twin-assisted 3-D electrical capacitance tomography for multiphase flow imagingIEEE Trans. Instrum. Meas.73 4507612 (doi: 10.1109/TIM.2024.3440409)

    work page doi:10.1109/tim.2024.3440409 2024

  31. [31]

    Banasiak R, Stawska Z and Fabija´ nska A 2025 Direct estimation of electric field distribution in circular ECT sensors using graph convolutional networksSensors256371 (doi: 10.3390/s25206371) 13

    work page doi:10.3390/s25206371 2025