GenVoid: Uncertainty-Aware Learning of Subsurface Material Defects with an Experimentally Validated Physics-Informed Generative Model
Trishit Mondal, Prajwal Bharadwaj, Nikhil Karanjgaokar, Ameya D. Jagtap
cs.LG
Quick verdict
GenVoid pairs a generative level-set prior with a real FEM solver for hidden-void identification; strong synthetic results, modest experimental IoU.
Summary
The paper introduces GenVoid, a physics-informed generative framework that infers the number, location, and shape of internal voids in 2D and 3D solids from boundary displacement measurements alone. Void geometry is represented by a parametric level set built from compactly supported RBFs, combining a variational latent decoder (trained as an auto-decoder on ~13,000 synthetic shapes) that supplies a shape prior with an MLP-head RBF correction refined during inference. Predictions are evaluated by a genuine FEniCS finite-element solver (linear elastic, Neo-Hookean hyperelastic, Ramberg–Osgood plasticity) and optimized with Levenberg–Marquardt using adjoint sensitivities; Monte Carlo sampling of the latent posterior yields void-probability maps. Across 16 synthetic 2D geometries IoU ranges 0.797–0.992 and across 9 3D cases 0.808–0.958; accuracy degrades with fewer sensors (3D two-void IoU 0.705 at 256 stations to 0.430 at 32) and with noise (2D IoUp 0.988 clean to 0.371 at 10% noise). An observability map based on adjoint topological sensitivity links defect location, sensor layout, and load ensembles, revealing stress shadows near existing voids. Validation on DIC measurements from PMMA and elastoplastic tensile specimens gives moderate IoUp of 0.596–0.789. Limitations: a planned baseline comparison (Section ??) is missing, no code release, no repeated trials or confidence intervals, known constitutive properties assumed, and only two experimental specimens.
Review complete
Assessment at a glance
4
risk flags identified
● high 1
● medium 2
● low 1
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- GenVoid identifies the number, location, and morphology of internal voids in 2D and 3D domains from surface displacement measurements alone, across linear-elastic, hyperelastic, and plastic material behaviors.
- The variational latent formulation propagates measurement noise into spatially resolved void-probability maps, quantifying uncertainty in the reconstructed geometry.
- Experimentally measured DIC displacement fields from physical tensile specimens validate void reconstruction with IoUp between 0.596 and 0.789.
- The proposed observability measure quantifies where defects are detectable and links sensor configuration and load ensembles to reconstruction fidelity, revealing stress shadows near existing defects.
Strengths
Genuine physics in the optimization loop
Predicted displacements are computed by a real FEniCS FEM solve at every Levenberg–Marquardt step with adjoint sensitivities (Section 3.3), not by a learned surrogate — the physics-informed label is substantive, and the same solver spans linear elastic, Neo-Hookean, and Ramberg–Osgood laws (Sections 2, 4.4).
Well-motivated two-stage latent design
The auto-decoder stage (Eq. 12) shapes the latent manifold on exact geometry targets before a frozen-decoder encoder stage (Eq. 15) maps measurements into it; Remark 1 gives a clear rationale, and permutation-invariant max+mean station pooling (Eq. 17) allows variable sensor counts without retraining.
Rare experimental validation with DIC
Practically useful observability measure
Candid degradation reporting
Weaknesses
Missing baseline comparison
No statistical rigor
Experimental scope mismatch
Steep noise sensitivity with unquantified real noise
Overbroad generality claims
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
GenVoid — Structured Preliminary Review
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 4 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is GenVoid?
GenVoid is a physics-informed generative framework that reconstructs hidden internal voids (defects) in 2D and 3D structures using only displacement measurements taken on the outer surface. It combines a learned shape prior (a variational latent decoder) with a real finite-element mechanics solver during optimization, and outputs a probability map of where the void is rather than a single guess.
How accurate is GenVoid?
On synthetic benchmarks it reaches IoU (overlap with the true void) of 0.797–0.992 across 16 two-dimensional cases and 0.808–0.958 across 9 three-dimensional cases. Accuracy drops with fewer sensors (e.g., IoU 0.705 to 0.430 in 3D when stations fall from 256 to 32) and with noise (IoUp 0.988 clean to 0.371 at 10% noise in 2D). On real experiments it achieves moderate IoUp of 0.596–0.789.
Was GenVoid validated on real physical experiments?
Yes. The authors manufactured tensile specimens (laser-cut acrylic and 3D-printed resin) with embedded T-shaped, L-shaped, and crescent-shaped voids, loaded them on an Instron machine, measured surface displacements with 2D digital image correlation (DIC), and fed those measurements to GenVoid. Two specimen sets were reconstructed under linear-elastic and elastoplastic material assumptions.
Does GenVoid need to know the material properties?
Yes. The current framework assumes the host material's constitutive law and properties (e.g., Young's modulus, hardening parameters) are known; only the void geometry is inferred. The authors identify joint geometry–material inference as future work.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures are available only as captions; the visual quality of reconstructions, probability maps, and observability fields could not be assessed.
- The baseline comparison referenced as 'Section ??' is absent from the provided text; it may be missing from arXiv v1 itself or lost in extraction, so the high-severity flag depends on this.
- Appendix B describes hyperelastic specimens and test protocols, but no hyperelastic experimental results appear in the provided text — this may be a manuscript gap rather than a claim of results.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.