Pith. sign in

REVIEW 3 major objections 7 minor 88 references

A physics-informed copula VAE recovers the true posterior over bridge stiffness damage from noisy modal data.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-11 09:53 UTC pith:ABGGYPSY

load-bearing objection Solid synthetic SciML method for bridge damage posteriors; the eigenvalue decoder and copula are real advances, but unimodal Gaussians miss the multimodality their own RQMC baselines show. the 3 major comments →

arxiv 2607.05025 v2 pith:ABGGYPSY submitted 2026-07-06 cs.LG

Uncertainty-aware damage identification in short-span bridges via physics-informed variational autoencoder

Ana Fernandez-Navamuel , A. Javier Omella , Diego Zamora-Sanchez , David Pardo This is my paper
classification cs.LG
keywords Scientific machine learningUncertainty quantificationVariational autoencoderStructural health monitoringGaussian copulaBridge damage identificationPhysics-informed neural networks
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Vibration-based damage identification for bridges is an ill-posed inverse problem: sparse sensors, measurement noise, and environmental variability mean many damage patterns can produce nearly the same frequencies and mode shapes. Deterministic neural networks return a single answer without uncertainty and can violate the underlying structural dynamics. This paper introduces a physics-informed Gaussian-copula variational autoencoder that maps noisy modal observations to a full posterior over element-wise stiffness reduction factors. The decoder is not a learned surrogate but a differentiable eigenvalue solver that enforces the free-vibration equations of the finite-element model; a Gaussian copula supplies the dependence structure among adjacent elements that ordinary diagonal-Gaussian VAEs ignore. On a synthetic simply-supported bridge with 2.5 % frequency noise and 5 % mode-shape noise the method recovers calibrated posteriors (95 % coverage near 0.79, low calibration error) while using far fewer trainable parameters than a purely data-driven counterpart. The result is an uncertainty-aware diagnosis that can flag early, localised damage and quantify how confident the assessment is.

Core claim

Embedding a differentiable numerical eigenvalue solver as the VAE decoder, together with a Gaussian-copula posterior over stiffness reduction factors, recovers the true Bayesian posterior for short-span bridge damage from noisy modal data, yielding well-calibrated credible intervals and higher average log-likelihood than a surrogate-decoder baseline while cutting the number of trainable parameters by roughly fourfold.

What carries the argument

Physics-informed Gaussian-copula variational autoencoder (PI-GCVAE): an encoder that outputs the parameters of truncated-Gaussian marginals and a Cholesky factor of a correlation matrix; a sampling layer that draws from the resulting Gaussian copula; and a decoder that assembles the updated stiffness matrix and solves the generalised eigenvalue problem exactly.

Load-bearing premise

The true damage posterior can be adequately captured by truncated univariate Gaussians linked only by a Gaussian copula; multimodal or strongly non-Gaussian marginals will be missed.

What would settle it

On the same synthetic bridge, replace the single-Gaussian marginals with a mixture that can represent the bimodal ground-truth densities shown for elements z2 and z4; if coverage, MACE and average log-likelihood do not improve, the expressiveness claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 7 minor

Summary. The paper proposes a physics-informed Gaussian copula variational autoencoder (PI-GCVAE) for vibration-based damage identification in short-span bridges. An encoder maps noisy modal features (frequencies and mode shapes) to parameters of a truncated-Gaussian-marginal / Gaussian-copula posterior over element stiffness reduction factors; samples are decoded by a differentiable generalized eigenvalue solver that enforces structural dynamics exactly. Training uses an ELBO-style loss with frequency, MAC, and posterior-density terms. Validation is entirely synthetic (simply supported beam, nel=5 and a 10-D check, 2.5% frequency / 5% mode-shape noise), with RQMC ground-truth posteriors, γ-sensitivity, and a surrogate-decoder baseline. At γ=0.30 the method reports MSE 0.034, 95% coverage 0.789, MACE 0.026, and higher ALL than the NN-decoder baseline, with fewer trainable parameters.

Significance. If the claims hold under the stated assumptions, the work is a useful SciML contribution to SHM: embedding a classical eigenvalue solver removes surrogate decoder error and cuts parameter count, while the Gaussian copula gives a scalable way to encode spatial stiffness correlations without the cost of full Gaussian mixtures or flows. The synthetic protocol is stronger than typical SHM ML papers—RQMC baselines, coverage/MACE/ALL, and an explicit NN-decoder ablation—and code availability is stated. The multimodality limitation is already partially acknowledged; with tempered claims and clearer quantification of when the variational family fails, the method would be a credible early-stage diagnostic tool for linear, bending-dominated short-span bridges.

major comments (3)
  1. The abstract and §5 claim that PI-GCVAE “accurately recovers the true posterior,” but the variational family in §3.3 (Eqs. 11–14) forces unimodal truncated-Gaussian marginals. Section 5.4.2 and Figs. 5a/6a show that RQMC ground-truth marginals (e.g. z2, z4) are strongly bimodal; the encoder places mass on a single mode and assigns near-zero probability to secondary modes. Coverage (0.789) and ALL can remain acceptable when the true z lies near the dominant mode, yet the shape of qζ(z|m) is systematically wrong precisely where the inverse problem is non-unique. This is load-bearing for the central claim: either quantify mode-capture failure (fraction of multimodal cases, mass on secondary modes, Wasserstein/TV to RQMC), or soften “accurately recovers” throughout abstract, §5, and §6 and state that the method recovers a unimodal approximation concentrated on the dominant mode.
  2. Abstract reports “77.2% coverage” while Table 3 (γ=0.30) reports 0.789 (78.9%) and the reader’s strongest claim cites 0.789. Reconcile the number, state exactly which credible-interval definition and γ produce the headline figure, and keep abstract, results, and conclusions consistent.
  3. §3.5 / Eq. (27): the implemented objective is a weighted sum of reconstruction discrepancies plus γ² E[log q], justified as a weighted negative ELBO under a uniform prior. The duplicated “Posterior PDF term” paragraph and the prose that LPDF “forces the approximate posterior to match the prior” are confusing (under a uniform prior, E[log q] is an entropy/KL term, not a match to a structured prior). Clarify the exact objective being minimized, the role of γ as a noise/temperature parameter, and that the prior is flat on [z_lb,1]^nel so that the regularizer is entropy-like rather than a match to an informative damage prior.
minor comments (7)
  1. Notation: PI-GCV AE / PI-GCVAE / GCV AE-NN appear with inconsistent spacing and hyphenation; standardize (e.g. PI-GCVAE) in abstract, keywords, and body.
  2. Fig. 1 caption and §2.1: boundary conditions are stated as fixed vertical displacement (and in one place rotation); ensure consistency with the simply-supported conditions of Eq. (2) (zero displacement and moment).
  3. Table 2 lists “Delayed early stopping … Starting epoch 10,000” while Fig. 3 caption refers to ~11,000 epochs; state the actual stopping epoch and final Lval used for the reported metrics.
  4. §4.1: “restricting the vertical displacement and rotation at both ends” conflicts with simply-supported (moment-free) ends; correct the FE BC description.
  5. Several references appear twice or with near-duplicate entries (e.g. Simoen et al. uncertainty review); clean the bibliography.
  6. Typos and line breaks: “bri dges”, “V AE”, “GCV AE”, “stiffness”, and broken words from PDF extraction should be fixed in the camera-ready source.
  7. §5.3 10-D experiment: state whether γ, λ, noise levels, and encoder capacity were re-tuned or held fixed from the 5-D case; this affects interpretation of the ALL drop.

Circularity Check

0 steps flagged

No load-bearing circularity: architecture and ELBO are standard; quantitative claims rest on independent synthetic true-z and separate RQMC posteriors, with only a minor shared-γ modeling choice.

full rationale

The paper's central claims (physical consistency via embedded eigenvalue decoder; recovery of posterior over z via Gaussian-copula VAE; coverage/ALL/MACE metrics) do not reduce by construction to their inputs. The decoder (Eqs. 15–19) is the classical generalized eigenproblem, not a fitted surrogate that redefines the target. Training minimizes the ELBO (Eq. 27) on labeled synthetic pairs {noisy m, known z}; evaluation metrics (MSE/MAE/coverage/ALL in §5.1–5.3 and Tables 3–5) are computed directly against the known synthetic ztrue, which are independent of the encoder parameters. The RQMC ground-truth posteriors (§5.4.1, Eqs. 35–39) are generated by an independent numerical procedure (QMC sampling + likelihood) solely for visual corner-plot comparison; they are not used to fit θ or to compute the headline numbers. Self-citations ([25,26,40,42]) supply background on prior deterministic and copula VAEs but are not invoked as uniqueness theorems or load-bearing premises for the present recovery claims. The only minor modeling linkage is that the same scalar γ appears both as the ELBO regularizer weight and as the likelihood precision in the RQMC construction (Eq. 38); this is a consistent noise-model choice, not a fit-then-predict circularity, and does not force the reported coverage of true z. Multimodality limitations of the truncated-Gaussian marginals are a correctness/expressiveness issue, not circularity. The derivation chain is therefore self-contained against external synthetic benchmarks.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 1 invented entities

The central empirical claim rests on standard structural dynamics plus several modeling choices that are not independently validated on real bridges: linear undamped modal physics, invariant mass, truncated-Gaussian+Gaussian-copula variational family, synthetic noise levels, and loss weights. Free parameters control noise robustness and calibration; invented entity is the PI-GCVAE architecture itself.

free parameters (6)
  • γ (ELBO PDF regularizer weight) = 0.30
    Selected by grid search over {0.25,0.30,0.35,0.40}; γ=0.30 chosen for best accuracy/ALL trade-off (Table 3). Directly controls reported coverage and calibration.
  • λ (MAC loss weight) = 10
    Fixed by hand to balance mode-shape vs log-frequency terms (Table 2).
  • frequency noise σ_f = 0.025 (2.5%)
    Multiplicative Gaussian noise level on synthetic frequencies; sets aleatoric difficulty of the inverse problem.
  • mode-shape noise σ_φ = 0.05 (5%)
    Additive Gaussian noise level relative to peak mode amplitude.
  • z_lb (minimum stiffness reduction) = 0.5
    Lower bound of damage support and truncated-Gaussian domain; defines the prior and sampling box.
  • encoder width / depth and learning rate = 128-128-128, lr=1e-5
    Three×128 ReLU layers, Adam 1e-5, batch 1024, L2 1e-5; architectural free choices affecting capacity.
axioms (6)
  • domain assumption Euler-Bernoulli beam theory with simply-supported BCs governs the short-span bridge (Eqs. 1–3).
    Foundation of the FE model and eigenvalue decoder; vertical bending only, no torsion/transverse coupling.
  • domain assumption Damage is a linear-elastic element-wise stiffness reduction with mass matrix invariant (Eq. 8–9).
    Early-stage damage assumption; excludes nonlinear, mass-changing, or support-condition damage.
  • domain assumption OMA-extracted first nm modes equal the undamped generalized eigenproblem solution up to the stated noise model.
    Justifies training on eigenvalue solutions rather than time-history simulation (§2.3, §4.1).
  • ad hoc to paper Approximate posterior factorizes as Gaussian copula times truncated-Gaussian marginals (Eqs. 11–14).
    Variational family choice; paper later shows it cannot capture multimodal marginals.
  • ad hoc to paper Prior on z is uniform on [z_lb,1]^nel; measurement noise independent Gaussian on log-frequencies and (1-MAC).
    Used both in ELBO and in RQMC ground-truth likelihood (Eq. 38).
  • standard math Sklar’s theorem / Gaussian copula density formulas and standard VAE ELBO identity.
    Standard probability and variational inference machinery.
invented entities (1)
  • PI-GCVAE (physics-informed Gaussian copula variational autoencoder) no independent evidence
    purpose: Encoder maps modal data to copula parameters; sampling layer draws correlated stiffness vectors; differentiable eigenvalue decoder enforces dynamics.
    The named architecture is the paper’s proposed object; no external independent evidence beyond the synthetic experiments here.

pith-pipeline@v1.1.0-grok45 · 28907 in / 3584 out tokens · 30987 ms · 2026-07-11T09:53:37.145521+00:00 · methodology

0 comments
read the original abstract

Vibration-based damage identification in civil infrastructure is a challenging, ill-posed inverse problem due to measurement noise, sparse sensor arrays, and environmental variability. While deep learning is powerful for system identification, deterministic approaches lack reliable uncertainty quantification and can yield physically inconsistent results. This work proposes a robust probabilistic Scientific Machine Learning (SciML) framework: a physics-informed Gaussian copula variational autoencoder (PI-GCVAE) for structural health monitoring (SHM). First, we eliminate the need for data-driven surrogates by embedding a differentiable numerical eigenvalue solver directly into the VAE architecture. This ensures that latent space samples satisfy the governing equations of structural dynamics, reducing the trainable parameter space and improving generalization. Second, we replace the conventional independence assumption of latent variables with a Gaussian copula. This model captures complex, physics-dependent spatial cross-correlations between adjacent structural elements, defining feasible solutions while accounting for inherent system variability and measurement errors. Third, compared with alternatives such as Gaussian mixtures, our copula-based VAE provides an efficient distributional model for high-dimensional, strongly correlated latent spaces. We validate the approach using a synthetic dataset of a simply supported bridge subjected to various damage scenarios and corrupted with stochastic Gaussian noise. Synthetic data enables exhaustive validation against ground-truth stiffness values unavailable in practice. Results demonstrate that the PI-GCVAE accurately recovers the true posterior distribution, achieving 77.2% coverage. The proposed framework provides a reliable, scalable tool for early-stage damage diagnosis in operating bridges.

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

88 extracted references · 4 canonical work pages

  1. [1]

    Jayasundara, D

    N. Jayasundara, D. P. Thambiratnam, T. H. T. Chan, A. Nguyen , Damage detection and quantification in deck type arch bridges using vibration based methods and artificia l neural networks, Engineering Failure Analysis 109 (2020) 104265

    2020

  2. [2]

    O. Avci, O. Abdeljaber, S. Kiranyaz, M. Hussein, M. Gabbouj, D. J. Inman, A review of vibration- based damage detection in civil structures: From traditional meth ods to Machine Learning and Deep Learning applications, Mechanical Systems and Signal Processing 1 47 (2021) 107077

    2021

  3. [3]

    E. N. Chatzi, C. Papadimitriou, Identification methods for struc tural health monitoring, volume 567, Springer, 2016

    2016

  4. [4]

    D. Liu, Z. Tang, Y. Bao, H. Li, Machine-learning-based methods f or output-only structural modal identification, Structural Control and Health Monitoring 28 (12) ( 2021) e2843

    2021

  5. [5]

    J. R. Casas, J. J. Moughty, Bridge damage detection based on v ibration data: Past and new developments, Frontiers in Built Environment 3 (2017) 1–12

    2017

  6. [6]

    M. I. Friswell, Damage identification using inverse methods, Philoso phical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 365 (18 51) (2007) 393–410

    2007

  7. [7]

    Rainieri, G

    C. Rainieri, G. Fabbrocino, Operational Modal Analysis of Civil Eng ineering Structures, An introduc- tion and a guide for applications, Springer, 2014

    2014

  8. [8]

    M. He, P. Liang, J. Li, Y. Zhang, Y. Liu, Fully automated precise op erational modal identification, Engineering Structures 234 (2021) 111988

    2021

  9. [9]

    Reynders, M

    E. Reynders, M. Schevenels, G. De Roeck, Macec 3.3, Technical Report, KU Leuven, 2014

    2014

  10. [10]

    W. Fan, P. Qiao, Vibration-based damage identification methods : a review and comparative study, Structural Health Monitoring 10 (1) (2011) 83–111

    2011

  11. [11]

    Bayissa, N

    W. Bayissa, N. Haritos, Vibration-based structural damage id entification using wavelet transform, Mechanical Systems and Signal Processing 22 (5) (2008) 1194–12 15

    2008

  12. [12]

    Roveri, A

    N. Roveri, A. Carcaterra, Damage detection in structures un der traveling loads by Hilbert-Huang transform, Mechanical Systems and Signal Processing 28 (2012) 128–144

    2012

  13. [13]

    M. L. Fugate, H. Sohn, Vibration-based damage detection usin g statistical process control, Mechanical Systems and Signal Processing 15 (4) (2001) 707–721

    2001

  14. [14]

    C. M. Bishop, Pattern Recognition and Machine Learning, Spring er, New York, 2006

    2006

  15. [15]

    Teughels, J

    A. Teughels, J. Maeck, G. De Roeck, Damage assessment by FE model updating using damage functions, Computers and Structures 80 (25) (2002) 1869–1879

    2002

  16. [16]

    J. R. Wu, Q. S. Li, Finite element model updating for a high-rise st ructure based on ambient vibration measurements, Engineering Structures 26 (7) (2004) 979–990

    2004

  17. [17]

    J. Liu, X. Meng, D. Zhang, C. Jiang, X. Han, An efficient method t o reduce ill-posedness for structural dynamic load identification, Mechanical Systems and Signal Process ing 95 (2017) 273–285

    2017

  18. [19]

    Giagopoulos, A

    D. Giagopoulos, A. Arailopoulos, V. Dertimanis, C. Papadimitriou, E. Chatzi, K. Grompanopoulos, Structural health monitoring and fatigue damage estimation using v ibration measurements and finite element model updating, Structural Health Monitoring 18 (4) (201 9) 1189–1206. 21

  19. [20]

    Kamariotis, E

    A. Kamariotis, E. Chatzi, D. Straub, Value of information from v ibration-based structural health monitoring extracted via Bayesian model updating, Mechanical Sys tems and Signal Processing 166 (2022) 108465

    2022

  20. [21]

    Ching, Y.-C

    J. Ching, Y.-C. Chen, Transitional Markov chain Monte Carlo met hod for Bayesian model updating, Journal of Engineering Mechanics 133 (7) (2007) 816–832

    2007

  21. [22]

    P. Ma, Y. Zhang, A copula-based transitional markov chain mon te carlo method for bayesian model updating, Reliability Engineering & System Safety (2025) 111572

    2025

  22. [23]

    L. Sun, Z. Shang, Y. Xia, S. Bhowmick, S. Nagarajaiah, Review o f Bridge Structural Health Monitor- ing Aided by Big Data and Artificial Intelligence: From Condition Assess ment to Damage Detection, Journal of Structural Engineering 146 (5) (2020) 04020073

    2020

  23. [24]

    Zhang, K.-V

    Y. Zhang, K.-V. Yuen, Review of artificial intelligence-based brid ge damage detection, Advances in Mechanical Engineering 14 (9) (2022) 16878132221122770

    2022

  24. [25]

    Fernandez-Navamuel, D

    A. Fernandez-Navamuel, D. Zamora-S´ anchez, ´A. J. Omella, D. Pardo, D. Garcia-Sanchez, F. Mag- alh˜ aes, Supervised Deep Learning with Finite Element simulations for damage identification in bridges, Engineering Structures 257 (2022) 114016

    2022

  25. [26]

    Fernandez-Navamuel, D

    A. Fernandez-Navamuel, D. Pardo, F. Magalh˜ aes, D. Zamora-S´ anchez,´A. J. Omella, D. Garcia-Sanchez, Bridge damage identification under varying environmental and oper ational conditions combining Deep Learning and numerical simulations, Mechanical Systems and Signal Processing 200 (2023) 110471

    2023

  26. [27]

    Sanayei, B

    M. Sanayei, B. Arya, E. M. Santini, S. Wadia-Fascetti, Significan ce of Modeling Error in Structural Parameter Estimation, Computer-Aided Civil and Infrastructure Engineering 16 (1) (2004) 12–27

    2004

  27. [28]

    Simoen, G

    E. Simoen, G. De Roeck, G. Lombaert, Dealing with uncertainty in model updating for damage assessment: A review, Mechanical Systems and Signal Processing 56 (2015) 123–149

    2015

  28. [29]

    Bourlard, Y

    H. Bourlard, Y. Kamp, Auto-Association by Multilayer Perceptr ons and Singular Value Decomposition, Biological Cybernetics 59 (2) (1988) 291–294

    1988

  29. [30]

    P. Peng, S. Jalali, X. Yuan, Solving inverse problems via auto-enc oders, IEEE Journal on Selected Areas in Information Theory 1 (1) (2020) 312–323

    2020

  30. [31]

    Tarantola, Inverse Problem Theory and Methods for Model Parameter Estimation, SIAM, 2004

    A. Tarantola, Inverse Problem Theory and Methods for Model Parameter Estimation, SIAM, 2004

    2004

  31. [32]

    Shahriari, D

    M. Shahriari, D. Pardo, J. A. Rivera, C. Torres-Verd ´ ın, A. Picon, J. Del Ser, S. Ossand´ on, V. M. Calo, Error control and loss functions for the deep learning inversion of borehole resistivity measurements, International Journal for Numerical Methods in Engineering 122 ( 6) (2021) 1629–1657

    2021

  32. [33]

    H. Sohn, K. H. Law, A Bayesian probabilistic approach for struc ture damage detection, Earthquake Engineering & Structural Dynamics 26 (12) (1997) 1259–1281

    1997

  33. [34]

    Z. Chen, R. Zhang, J. Zheng, H. Sun, Sparse Bayesian learning for structural damage identification, Mechanical Systems and Signal Processing 140 (2020) 106689

    2020

  34. [35]

    Doersch, Tutorial on Variational Autoencoders, arXiv pre print arXiv:1606.05908 (2016)

    C. Doersch, Tutorial on Variational Autoencoders, arXiv pre print arXiv:1606.05908 (2016)

    Pith/arXiv arXiv 2016

  35. [36]

    Pinheiro Cinelli, M

    L. Pinheiro Cinelli, M. Ara´ ujo Marins, E. A. Barros da Silva, S. Lima Netto, Variational autoencoder, Variational methods for machine learning with applications to deep ne tworks, Springer, 2021, pp. 111–149

    2021

  36. [37]

    H. Goh, S. Sheriffdeen, Solving Bayesian Inverse Problems via Va riational Autoencoders, Proceedings of Machine Learning Research 145 (2021) 386–425

    2021

  37. [38]

    M. Liu, D. Grana, L. P. de Figueiredo, Uncertainty quantificatio n in stochastic inversion with dimensionality reduction using variational autoencoder, Geophysic s 87 (2) (2022) M43–M58. 22

    2022

  38. [39]

    Rodriguez, J

    O. Rodriguez, J. M. Taylor, D. Pardo, Multimodal variational au toencoder for inverse problems in geophysics: Application to a 1-D magnetotelluric problem, Geophysic al Journal International 235 (3) (2023) 2598–2613

    2023

  39. [40]

    Fernandez-Navamuel, N

    A. Fernandez-Navamuel, N. Gorostidi, D. Pardo, V. Nava, E. C hatzi, Gaussian mixture autoencoder for uncertainty-aware damage identification in a floating offshore w ind turbine, Wind Energy Science 10 (5) (2025) 857–885

    2025

  40. [41]

    Papamakarios, E

    G. Papamakarios, E. Nalisnick, Normalizing flows for probabilistic m odeling and inference, Journal of Machine Learning Research 22 (57) (2021) 1–64

    2021

  41. [42]

    Fernandez-Navamuel, M

    A. Fernandez-Navamuel, M. A. Diaz Viera, M. Croci, A Copula-ba sed variational autoencoder for uncertainty quantification in inverse problems: application to damag e identification in an offshore wind turbine, arXiv preprint arXiv:2510.02013 (2026)

    arXiv 2026

  42. [43]

    R. B. Nelsen, Properties and applications of copulas: A brief sur vey, Proceedings of the first brazilian conference on statistical modeling in insurance and finance, USP Sa o Paulo, 2003, pp. 10–28

    2003

  43. [44]

    How to start training: The effect of initialization and architecture

    Boris Hanin and David Rolnick. How to start training: The effect of initialization and architecture. Advances in Neural Information Processing Systems , 31, 2018

    2018

  44. [45]

    Joe, Dependence modeling with copulas, CRC press, 2014

    H. Joe, Dependence modeling with copulas, CRC press, 2014

    2014

  45. [46]

    Joe, Parametric families of multivariate distributions with given margins, Journal of Multivariate Analysis 46 (2) (1993) 262–282

    H. Joe, Parametric families of multivariate distributions with given margins, Journal of Multivariate Analysis 46 (2) (1993) 262–282

    1993

  46. [47]

    Okazaki, Scientific Machine Learning Seismology, arXiv preprin t arXiv:2409.18397 (2025)

    T. Okazaki, Scientific Machine Learning Seismology, arXiv preprin t arXiv:2409.18397 (2025)

    Pith/arXiv arXiv 2025

  47. [48]

    Raissi, P

    M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics-informed ne ural networks: A deep learning frame- work for solving forward and inverse problems involving nonlinear par tial differential equations, Journal of Computational Physics 378 (2019) 686–707

    2019

  48. [49]

    Mo, W.-J

    J. Mo, W.-J. Yan, Explainable Neural-Networked Variational Inf erence: A New and Fast Paradigm with Automatic Differentiation for High-Dimensional Bayesian Inverse Pr oblems, Reliability Engineering & System Safety (2025) 111337

    2025

  49. [50]

    Almanst¨ otter, R

    M. Almanst¨ otter, R. Vetter, D. Iber, PINNverse: Accurat e parameter estimation in differen- tial equations from noisy data with constrained physics-informed n eural networks, arXiv preprint arXiv:2504.05248 (2025)

    Pith/arXiv arXiv 2025

  50. [51]

    de O Teloli, R

    R. de O Teloli, R. Tittarelli, M. Bigot, L. Coelho, E. Ramasso, P. Le M oal, M. Ouisse, A physics- informed neural networks framework for model parameter ident ification of beam-like structures, Mechanical Systems and Signal Processing 224 (2025) 112189

    2025

  51. [52]

    Y. Shin, J. Darbon, G. E. Karniadakis, On the convergence of p hysics informed neural networks for linear second-order elliptic and parabolic type PDEs, arXiv preprint a rXiv:2004.01806 (2020)

    Pith/arXiv arXiv 2004

  52. [53]

    Haywood-Alexander, G

    M. Haywood-Alexander, G. Arcieri, A. Kamariotis, E. Chatzi, Re sponse estimation and system identifi- cation of dynamical systems via physics-informed neural network s, Advanced Modeling and Simulation in Engineering Sciences 12 (1) (2025) 8

    2025

  53. [54]

    S. S. Eshkevari, M. Tak´ aˇ c, S. N. Pakzad, M. Jahani, DynNet: Physics-based neural architecture design for nonlinear structural response modeling and prediction, Engine ering Structures 229 (2021) 111582

    2021

  54. [55]

    Bacsa, W

    K. Bacsa, W. Liu, I. Abdallah, E. Chatzi, Structural Dynamics F eature Learning Using a Supervised Variational Autoencoder, Journal of Engineering Mechanics 151 ( 2) (2025) 04024106

    2025

  55. [56]

    H.-y. Zhao, C. Wang, J.-s. Fan, Reverse Variational Autoenco der: A probabilistic inference framework for structural health monitoring and inverse analysis in engineering , Reliability Engineering & System Safety (2025) 111938. 23

    2025

  56. [57]

    J. S. Przemieniecki, Theory of Matrix Structural Analysis, McG raw-Hill, 1968

    1968

  57. [58]

    J. S. Przemieniecki, Theory of matrix structural analysis, Cou rier Corporation, 1985

    1985

  58. [59]

    D. W. Pepper, J. C. Heinrich, The finite element method: basic co ncepts and applications, Taylor & Francis, 2005

    2005

  59. [60]

    Strang, Linear algebra and its applications, Thomson, Brook s/Cole, 2006

    G. Strang, Linear algebra and its applications, Thomson, Brook s/Cole, 2006

    2006

  60. [61]

    Magalh˜ aes, ´A

    F. Magalh˜ aes, ´A. Cunha, Explaining operational modal analysis with data from an ar ch bridge, Mechanical Systems and Signal Processing 25 (5) (2011) 1431–14 50

    2011

  61. [62]

    Reynders, Rik Pintelon, G

    E. Reynders, Rik Pintelon, G. De Roeck, Uncertainty bounds on modal parameters obtained from stochastic subspace identification, Mechanical Systems and Signa l Processing 22 (4) (2008) 948–969

    2008

  62. [63]

    Abadi, P

    M. Abadi, P. Barham, J. Chen, Z. Chen, TensorFlow: A system f or large-scale machine learning, arXiv preprint arXiv:1605.08695 (2016)

    Pith/arXiv arXiv 2016

  63. [64]

    D. P. Kingma, J. L. Ba, Adam: A method for stochastic optimizat ion, 3rd International Conference on Learning Representations, ICLR (2015) 1–15

    2015

  64. [65]

    C. P. Burgess, I. Higgins, A. Pal, L. Matthey, N. Watters, G. D esjardins, A. Lerchner, Understanding disentangling in Beta-V AE, arXiv preprint arXiv:1804.03599 (2018)

    Pith/arXiv arXiv 2018

  65. [66]

    J. E. Mottershead, M. I. Friswell, Model updating in structura l dynamics: a survey, Journal of Sound and Vibration 167 (2) (1993) 347–375

    1993

  66. [67]

    Y. Xia, H. Hao, J. M. W. Brownjohn, P.-Q. Xia, Damage identificat ion of structures with uncertain frequency and mode shape data, Earthquake Engineering & Struc tural Dynamics 31 (5) (2002) 1053– 1066

    2002

  67. [68]

    Bekta¸ s, Y

    S. Bekta¸ s, Y. S ¸i¸ sman, The comparison of L1 and L2-norm minimization methods, International Journal of the Physical Sciences 5 (11) (2010) 1721–1727

    2010

  68. [69]

    A. B. Owen, Monte Carlo, quasi-Monte Carlo, and randomized qu asi-Monte Carlo, Monte-Carlo and Quasi-Monte Carlo Methods 1998, Springer (2000) 86–97

    1998

  69. [70]

    L’Ecuyer, C

    P. L’Ecuyer, C. Lemieux, Recent advances in randomized quasi- Monte Carlo methods, Modeling uncertainty, Springer (2002) 419–474

    2002

  70. [71]

    Sklar, Random variables, joint distribution functions, and co pulas, Kybernetika 9 (6) (1973) 449–460

    A. Sklar, Random variables, joint distribution functions, and co pulas, Kybernetika 9 (6) (1973) 449–460

    1973

  71. [72]

    P. K. Trivedi, D. M. Zimmer, Copula modeling: an introduction for p ractitioners, Foundations and Trends in Econometrics 1 (1) (2007) 1–111

    2007

  72. [73]

    Wasserman, G

    M. Wasserman, G. Mateos, Stabilizing the kumaraswamy distribu tion, arXiv preprint arXiv:2410.00660 (2024)

    Pith/arXiv arXiv 2024

  73. [74]

    Kuleshov, N

    V. Kuleshov, N. Fenner, Accurate uncertainties for deep lear ning using calibrated regression, Interna- tional Conference on Machine Learning (2018) 2796–2804

    2018

  74. [75]

    Pastor, M

    M. Pastor, M. Binda, T. Harˇ car ´ ık, Modal assurance criterion, Procedia Engineering 48 (2012) 543–548

    2012

  75. [76]

    D. M. Blei, A. Kucukelbir, J. D. McAuliffe, Variational Inference: A Review for Statisticians, Journal of the American Statistical Association 112 (518) (2017) 859–877

    2017

  76. [77]

    D. P. Kingma, M. Welling, Auto-encoding variational bayes, arXiv preprint arXiv:1312.6114 (2013)

    Pith/arXiv arXiv 2013

  77. [78]

    Pandey, M

    A. Pandey, M. Biswas, Damage detection from changes in curva ture mode shapes, Journal of Sound and Vibration 145 (2) (1991) 321–332

    1991

  78. [79]

    Hammersley, Monte Carlo methods, Springer Science & Busine ss Media, 2013

    J. Hammersley, Monte Carlo methods, Springer Science & Busine ss Media, 2013. 24

    2013

  79. [80]

    Rahaman, A

    N. Rahaman, A. Baratin, On the spectral bias of neural netwo rks, International Conference on Machine Learning, PMLR (2019) 5301–5310

    2019

  80. [81]

    Explicit uncertainty propagation in Mahalanobis squared distance for reliable damage detection of brid ges using experimental data,

    G. A. del Pozo, B. T. Svendsen, Ø. W. Petersen, and O. Øiseth , “Explicit uncertainty propagation in Mahalanobis squared distance for reliable damage detection of brid ges using experimental data,” Journal of Civil Structural Health Monitoring , pp. 1–24, 2025

    2025

Showing first 80 references.