Datasets:

guanwencan
/

ASCE_JCCV

This repository contains sixteen high-resolution supplemental figures supporting the numerical validation, seismic damage-field analysis, sparse reconstruction, surrogate comparison, and spatial fidelity evaluation presented in the manuscript.

Dataset Description

A single elasto-plastic time-history analysis of a tunnel-soil system can require several hours of computation. The underlying study investigates whether surrogate models fitted to sparse samples of simulated damage fields can accelerate this computation while preserving the field peaks that govern structural assessment of the tunnel lining.

Three reconstruction mechanisms are evaluated using the same reference fields, observation locations, and performance metrics:

A mechanics-informed non-stationary Gaussian process with homoscedastic noise.
A mechanics-informed non-stationary Gaussian process with geometrically prescribed heteroscedastic noise.
An untrained deep image prior.

The methods are compared across three field quantities and four seismic instants using a common set of 3,200 observation points.

The reference fields originate from a plane-strain finite-element model of a shield tunnel embedded in soil. In the constitutive model, the Mohr-Coulomb cohesion, friction angle, and dilation angle degrade with accumulated equivalent plastic strain. Damage is therefore generated at the constitutive level rather than introduced through post-processing.

Figure Gallery

Figure S1: Dynamic-model validation

Verification of the dynamic model using an independent elastic case. Equivalent stress distributions in the soil and tunnel lining are shown at five time instants.

Figure S2: Evolution of the reference damage fields

Reference results for all six field measures at four recorded seismic instants.

Figure S3: Effect of the damage constitutive law

Difference in equivalent plastic strain between simulations performed with and without the damage constitutive law.

Figure S4: Faster-softening case

Damage-field evolution for the constitutive configuration with faster material softening.

Figure S5: Reconstruction of equivalent plastic strain

Sparse reconstruction of the equivalent plastic-strain field at all four seismic instants.

Figure S6: Reconstruction of the damage field

Sparse reconstruction of the damage field at all four seismic instants.

Figure S7: Reconstruction of plastic dissipation

Sparse reconstruction of the plastic-dissipation field at all four seismic instants.

Figure S8: Deep-image-prior early-stopping audit

Early-stopping analysis for the deep image prior across all twelve reconstructed frames.

Figure S9: Gradient sharpness ratio

Spatial gradient sharpness ratio for the three reconstructed field quantities, including the reference fields.

Figure S10: Signed relative error

Signed relative-error distributions for the three reconstructed field quantities, including the reference fields.

Figure S11: Damage-front orientation error

Damage-front orientation errors for the three reconstructed field quantities, including the reference fields.

Figure S12: Local peak fidelity

Local peak-fidelity evaluation for the three reconstructed field quantities, including the reference fields.

Figure S13: Neural-network surrogate on an independent case

Performance of the neural-network surrogate when evaluated on an independent case.

Figure S14: Stationary-kernel surrogates on an independent case

Performance of stationary-kernel surrogate models when evaluated on an independent case.

Figure S15: Predicted-versus-reference comparison

Predicted-versus-reference scatter plots for six surrogate configurations and two field quantities.

Figure S16: Best-performing surrogate by element

Element-wise assignment of the best-performing surrogate model.

File Index

Figure	File	Description
S1	`S01_validation_stress.png`	Independent elastic validation case
S2	`S02_damage_evolution_full.png`	Reference damage-field evolution
S3	`S03_depeq_difference.png`	Plastic-strain difference
S4	`S04_faster_softening.png`	Faster-softening case
S5	`S05_recon_kappa_allframes.png`	Equivalent plastic-strain reconstruction
S6	`S06_recon_D_allframes.png`	Damage-field reconstruction
S7	`S07_recon_Wp_allframes.png`	Plastic-dissipation reconstruction
S8	`S08_dip_earlystop.png`	Deep-image-prior early-stopping audit
S9	`S09_gsr_allfields.png`	Gradient sharpness ratio
S10	`S10_relerr_allfields.png`	Signed relative error
S11	`S11_orient_allfields.png`	Damage-front orientation error
S12	`S12_lpf_allfields.png`	Local peak fidelity
S13	`S13_nn_independent.png`	Neural-network surrogate
S14	`S14_kernels_independent.png`	Stationary-kernel surrogates
S15	`S15_scatter_both.png`	Predicted-versus-reference scatter
S16	`S16_dominance_map.png`	Best-performing surrogate by element

All figure files are located in the figs_supp directory.

Figure Production

The figures were rendered with Matplotlib at 300 dpi using cached finite-element and reconstruction results. No numerical model was rerun specifically to create this supplemental dataset.

Field plots were produced using tricontourf on the unstructured finite-element mesh. The tunnel opening was masked, and a focus window of approximately +/-6 m was used. A shared color scale was applied within each comparison row to ensure consistent visual interpretation.

Intended Use

This dataset is intended for:

Reproducibility and visual inspection of the manuscript results.
Comparison of sparse field-reconstruction methods.
Evaluation of spatial fidelity metrics.
Research and teaching in earthquake, tunnel, and computational engineering.
Development of surrogate models for finite-element field data.

Limitations

These files contain rendered supplemental figures rather than the complete finite-element model, raw simulation database, or executable reconstruction code. Numerical values should therefore not be extracted from the images when the original computational data are required.

The manuscript is currently under review. The bibliographic information in this dataset card will be updated after publication.

License

The supplemental figures are released under the Creative Commons Attribution 4.0 International License.

You may share and adapt the material provided that appropriate credit is given.

Citation

Until the final publication reference becomes available, please cite this dataset as:

@dataset{guan2026shield_tunnel_supplement,
  author    = {Guan, Wencan},
  title     = {Supplemental Figures for Surrogate-Accelerated Computation
               of Seismic Damage Fields in Shield Tunnels},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/guanwencan/ASCE_JCCV},
  license   = {CC BY 4.0}
}

Contact

Profile	Information
	Wencan Guan School of Civil Engineering, Tongji University Institute of Structural Mechanics, Bauhaus-Universität Weimar ORCID: 0009-0005-9398-1331 Hugging Face: guanwencan

Downloads last month: 124

Number of rows:

Total file size:

9.56 MB