CoChat

On the Consistent Classification and Treatment of Uncertainties in Structural Health Monitoring Applications

Open access
Kamariotis, Antonios, Vlachas, Konstantinos, Ntertimanis, Vasileios, Koune, Ioannis, Cicirello, Alice, Chatzi, Eleni · 2025 · ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part B: Mechanical Engineering

The DOI matches its Crossref record; 156 of the 156 references it cites were found in a registry or an index; 5 of the 44 sources its claims cite were checked for support.

Abstract

In this paper, we provide a comprehensive definition and classification of various sources of uncertainty within the fields of structural dynamics, system identification, and structural health monitoring (SHM), with a primary focus on the latter. Utilizing the classical input–output system representation as a main contextual framework, we present a taxonomy of uncertainties, intended for consistent classification of uncertainties in SHM applications: (i) input uncertainty; (ii) model form uncertainty; (iii) model parameter/variable uncertainty; (iv) measurement uncertainty; and (v) inherent variability. We then critically review methods and algorithms that address these uncertainties in the context of key SHM tasks: system identification and model inference, model updating, accounting for environmental and operational variability (EOV), virtual sensing, damage identification, and prognostic health management. A benchmark shear frame model with hysteretic links is employed as a running example to illustrate the application of selected methods and algorithmic tools. Finally, we discuss open challenges and future research directions in uncertainty quantification for SHM.

Overview

CoChat Summary

TL;DR

This review proposes a consistent taxonomy of uncertainty sources in structural dynamics, system identification, and structural health monitoring (SHM): input uncertainty, model form uncertainty, model parameter/variable uncertainty, measurement uncertainty, and inherent variability. It reviews methods for key SHM tasks and demonstrates selected approaches on a benchmark two-story shear frame with Bouc-Wen hysteretic links.

Key findings

  • The paper defines and classifies five uncertainty classes for SHM and maps them onto the classical input-output system representation, relating them to prior literature categories such as aleatory/epistemic, model inadequacy, parameter uncertainty, and measurement noise.
  • It reviews uncertainty treatment across SHM tasks including system identification/model inference, model updating, environmental and operational variability (EOV), state estimation/virtual sensing, damage identification, and prognostic health management.
  • In the benchmark demonstration, the linear time-invariant frame has 72 normal modes from 3.85 to 433.70 Hz; a Monte Carlo campaign of 100 experiments with 5% noise-to-signal ratio and 32,768 samples is used, with analysis limited to [0, 50 Hz] containing 32 translational and rotational normal modes.
  • For the time-varying/EOV case, 365 independent experiments are run with temperature span 30 °C, reference temperature 20 °C, and thermal coefficient \(K_T = 5 \times 10^{-3}\); ARX(42,42,0) identification gives percentage fitness concentrated around 98.95%.
  • In Bayesian model updating, the posterior for the Bouc-Wen parameter \(\alpha\) shifts toward the true value 0.2 and reduces uncertainty relative to the prior \(U(0.05,1.0)\), though posterior uncertainty remains nonnegligible.
  • In damage identification, with ground-truth storey stiffness ratios \(h_1 = h_2 = 0.8\), 10% RMS noise, and 250 measurements per sensor, the posterior reduces uncertainty but overestimates the first-storey damage factor.

Methods

  • Design: a review and taxonomy paper, not a primary experimental study; it synthesizes literature and uses a benchmark finite-element model of a three-dimensional two-story shear frame with nonlinear Bouc-Wen hysteretic joints as a running example.
  • Data/sample: synthetic simulation data from the benchmark frame, including Monte Carlo experiments for linear time-invariant identification, 365 temperature-varying experiments for EOV, synthetic acceleration data for Bayesian model updating, earthquake-excited response for virtual sensing, and synthetic damaged-state data for damage identification.
  • Main analysis: methods demonstrated include ARX/AR time-series identification, GPR-ARX modeling under EOV, Bayesian model updating with transitional MCMC, unscented Kalman filtering for joint state-parameter estimation, and variational Bayesian Monte Carlo for damage-parameter inference.

Limitations

  • The paper states open challenges including prior assumptions that affect uncertainty quantification, lack of metrics/benchmarks/validation strategies for comparing UQ algorithms, need for scalable UQ solutions, and strategies for reducing uncertainties.
  • The paper notes specific demonstration limitations: the BMU posterior uncertainty is nonnegligible; the hysteretic parameter \(\alpha\) remains unobservable until nonlinearity is excited; and damage-factor overestimation is attributed to limited data, measurement uncertainty, model form uncertainty, and variational approximation error.
  • The text also makes evident that the benchmark examples rely on synthetic data, simplified assumptions such as no spatial/temporal correlation in likelihoods, and in places a fictitious temperature-modulus relation and a toy problem with properties “not necessarily up to the real scale.”

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1115/1.4067140

Cites

156 of 156 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 High-level representation of a monitored dynamical system. S represents a physical engineering system subjected to an input u ( t ), while being exposed to disturbances d ( t ) and deterioration processes h ( t ). The system produces an output y ( t ), asubsetofwhichisusuallymonitoredviatheuseofsensors.The system representation is escorted by a prescribed or inferred system model M that aims to represent the actual physical system as closely as possible. This features an internal, usually latent, state vector x ( t ), and can be further described by a set of parameters N .
Fig. 1 High-level representation of a monitored dynamical system. S represents a physical engineering system subjected to an input u ( t ), while being exposed to disturbances d ( t ) and deterioration processes h ( t ). The system produces an output y ( t ), asubsetofwhichisusuallymonitoredviatheuseofsensors.The system representation is escorted by a prescribed or inferred system model M that aims to represent the actual physical system as closely as possible. This features an internal, usually latent, state vector x ( t ), and can be further described by a set of parameters N .CC-BY · From the PDF
Fig. 2 Illustration of the infiltration of the classes or our defined uncertainty taxonomy in the essential representation of a monitored dynamical system, as portrayed in Fig. 5. The defined classes include input uncertainty, model form uncertainty, model parameter/variable uncertainty, measurement uncertainty, and inherent variability.
Fig. 2 Illustration of the infiltration of the classes or our defined uncertainty taxonomy in the essential representation of a monitored dynamical system, as portrayed in Fig. 5. The defined classes include input uncertainty, model form uncertainty, model parameter/variable uncertainty, measurement uncertainty, and inherent variability.CC-BY · From the PDF
Fig. 3 Categorization of purely data-driven system methods for damage identification, organized according to the domain the method operates in, the method class, and the specific featurethat is producedtoservefordamageidentification.ThisfigureisadaptedfromRef.[64]. This table overviews methods rely on identification of the system shown in Fig. 1. The feature does not cover the deep learning based schemes for feature extraction.
Fig. 3 Categorization of purely data-driven system methods for damage identification, organized according to the domain the method operates in, the method class, and the specific featurethat is producedtoservefordamageidentification.ThisfigureisadaptedfromRef.[64]. This table overviews methods rely on identification of the system shown in Fig. 1. The feature does not cover the deep learning based schemes for feature extraction.CC-BY · From the PDF
Fig. 4 Graphical representation of the two-story frame with annotation of the hysteretic (nonlinear) link positions, as well as the positions of acceleration sensors that are assumed to be deployed for the state estimation/virtual sensing task of Sec. 4.4
Fig. 4 Graphical representation of the two-story frame with annotation of the hysteretic (nonlinear) link positions, as well as the positions of acceleration sensors that are assumed to be deployed for the state estimation/virtual sensing task of Sec. 4.4CC-BY · From the PDF
Fig. 5 Illustration of the nonlinear hysteretic mechanism that is applied for each link element, according to its DOF. The link elements are noticeable through the additional nodes (white circles) near the end of each beam/column element. The arrows indicate the local coordinate system assumed for each beam element.
Fig. 5 Illustration of the nonlinear hysteretic mechanism that is applied for each link element, according to its DOF. The link elements are noticeable through the additional nodes (white circles) near the end of each beam/column element. The arrows indicate the local coordinate system assumed for each beam element.CC-BY · From the PDF
Fig. 6 Welch-based coherence function estimates (Monte Carlo experiments; LTI frame, point O2): ( a ) along x -axis and ( b ) along y -axis
Fig. 6 Welch-based coherence function estimates (Monte Carlo experiments; LTI frame, point O2): ( a ) along x -axis and ( b ) along y -axisCC-BY · From the PDF
Fig. 7 Welch-based FRF estimates (Monte Carlo experiments; LTI frame, point O2): ( a ) magnitude along x -axis, ( b ) phase along x -axis, ( c ) magnitude along y -axis, and ( d ) phase along y -axis
Fig. 7 Welch-based FRF estimates (Monte Carlo experiments; LTI frame, point O2): ( a ) magnitude along x -axis, ( b ) phase along x -axis, ( c ) magnitude along y -axis, and ( d ) phase along y -axisCC-BY · From the PDF
Image 8
CC-BY · From the PDF
Image 9
CC-BY · From the PDF
Image 10
CC-BY · From the PDF
Fig. 8 ARX model order estimation (Monte Carlo experiments; LTI frame, point O2. The squares indicate mean value and the vertical bars standard deviation): ( a ) along x -axis and ( b ) along y -axis.
Fig. 8 ARX model order estimation (Monte Carlo experiments; LTI frame, point O2. The squares indicate mean value and the vertical bars standard deviation): ( a ) along x -axis and ( b ) along y -axis.CC-BY · From the PDF
Image 12
CC-BY · From the PDF
Fig. 9 ARXmodelparametervectors (Monte Carlo experiments; LTI frame, point O2. The squares indicate mean value and the vertical bars standard deviation): ( a ) ARX x ð 30, 30, 0 Þ models and ( b ) ARX y ð 42, 42, 0 Þ models.
Fig. 9 ARXmodelparametervectors (Monte Carlo experiments; LTI frame, point O2. The squares indicate mean value and the vertical bars standard deviation): ( a ) ARX x ð 30, 30, 0 Þ models and ( b ) ARX y ð 42, 42, 0 Þ models.CC-BY · From the PDF
Fig. 10 ARX-estimated vibration modes (Monte Carlo experiments; LTI frame, point O2): ( a ) natural frequency along x -axis, ( b ) modal damping ratio along x -axis, ( c ) natural frequency along y -axis, and ( d ) modal damping ratio along y -axis
Fig. 10 ARX-estimated vibration modes (Monte Carlo experiments; LTI frame, point O2): ( a ) natural frequency along x -axis, ( b ) modal damping ratio along x -axis, ( c ) natural frequency along y -axis, and ( d ) modal damping ratio along y -axisCC-BY · From the PDF
Fig. 11 AR-estimated vibration modes (Monte Carlo experiments; LTI frame, point O2): ( a ) natural frequency along x -axis and ( b ) modal damping ratio along x -axis
Fig. 11 AR-estimated vibration modes (Monte Carlo experiments; LTI frame, point O2): ( a ) natural frequency along x -axis and ( b ) modal damping ratio along x -axisCC-BY · From the PDF
Fig. 12 Temperatureandmodulusdataforthelineattimevariant (LTV) frame: ( a ) temperature time-series over the long temporal scale, ( b ) sample(bars)andassociatedkernel-based(continuous line) PDF estimates, and ( c ) elastic modulus versus temperature
Fig. 12 Temperatureandmodulusdataforthelineattimevariant (LTV) frame: ( a ) temperature time-series over the long temporal scale, ( b ) sample(bars)andassociatedkernel-based(continuous line) PDF estimates, and ( c ) elastic modulus versus temperatureCC-BY · From the PDF
Fig. 13 Behavior of the first three natural frequencies of the LTV frame: ( a ) versus temperature and ( b ) versus long time-scale
Fig. 13 Behavior of the first three natural frequencies of the LTV frame: ( a ) versus temperature and ( b ) versus long time-scaleCC-BY · From the PDF
Fig. 14 Performance of the model order selection criteria (LTV frame, point O2, y direction): ( a ) the BIC and ( b ) the percentage fitness to data
Fig. 14 Performance of the model order selection criteria (LTV frame, point O2, y direction): ( a ) the BIC and ( b ) the percentage fitness to dataCC-BY · From the PDF
Fig. 15 Performance of the ARX(42, 42, 0) prediction errors (LTV frame, point O2, y direction): ( a ) sample autocorrelation function, ( b ) samplePDFofthenormalized(  e ½ k Š ‰ N ð 0, 1 Þ ) prediction errors, and ( c ) normal probability plot of  e ½ k Š
Fig. 15 Performance of the ARX(42, 42, 0) prediction errors (LTV frame, point O2, y direction): ( a ) sample autocorrelation function, ( b ) samplePDFofthenormalized(  e ½ k Š ‰ N ð 0, 1 Þ ) prediction errors, and ( c ) normal probability plot of  e ½ k ŠCC-BY · From the PDF
Fig. 16 Distribution of the percentage fitness to the short time estimation dataset during ARX modeling (LTV frame, point O2, y direction)
Fig. 16 Distribution of the percentage fitness to the short time estimation dataset during ARX modeling (LTV frame, point O2, y direction)CC-BY · From the PDF
Fig. 17 Indicative results of the AR/X parameters' GPR modeling (LTV frame, point O2, y direction): ( a ) AR coefficient and ( b ) exogenous coefficient
Fig. 17 Indicative results of the AR/X parameters' GPR modeling (LTV frame, point O2, y direction): ( a ) AR coefficient and ( b ) exogenous coefficientCC-BY · From the PDF
Fig. 18 BMU for inferring the parameter a of the Bouc-Wen model and quantifying the posterior uncertainty
Fig. 18 BMU for inferring the parameter a of the Bouc-Wen model and quantifying the posterior uncertaintyCC-BY · From the PDF
Fig. 19 Virtual sensing; estimation of unmeasured quantities of interest-displacement and velocities at the first (node O 1) and secondstorey(node O 2), along the x direction of Fig. 4. Quantities estimated via coupling of the M ROM with a UKF are denoted in dashed lines, reference (true) time histories, as simulated via the M ROM, in solid lines. It is reminded that only the second storey accelerations are assumed to be measured.
Fig. 19 Virtual sensing; estimation of unmeasured quantities of interest-displacement and velocities at the first (node O 1) and secondstorey(node O 2), along the x direction of Fig. 4. Quantities estimated via coupling of the M ROM with a UKF are denoted in dashed lines, reference (true) time histories, as simulated via the M ROM, in solid lines. It is reminded that only the second storey accelerations are assumed to be measured.CC-BY · From the PDF
Fig. 20 Hysteretic loop drawn for the link element experiencing the largest amount of relative deformation. In this case, this corresponding to a rotational spring of a second storey element. The plot displays the difference between the FOM and ROM simulation, reflecting the aspect of model form uncertainty.
Fig. 20 Hysteretic loop drawn for the link element experiencing the largest amount of relative deformation. In this case, this corresponding to a rotational spring of a second storey element. The plot displays the difference between the FOM and ROM simulation, reflecting the aspect of model form uncertainty.CC-BY · From the PDF
Fig. 21 Identification of the hysteretic (Bouc-Wen) model parameter a , in an effort to narrow model parameter uncertainty
Fig. 21 Identification of the hysteretic (Bouc-Wen) model parameter a , in an effort to narrow model parameter uncertaintyCC-BY · From the PDF
Fig. 22 Posterior distribution of the uncertain parameters h 1, h 2, and c . The vertical dashed lines denote the ground truth for parameters where it is known.
Fig. 22 Posterior distribution of the uncertain parameters h 1, h 2, and c . The vertical dashed lines denote the ground truth for parameters where it is known.CC-BY · From the PDF
Fig. 23 Probability of the damage factor at the i th storey exceeding a given threshold
Fig. 23 Probability of the damage factor at the i th storey exceeding a given thresholdCC-BY · From the PDF

Verification

Claims the paper makes

4 Valid · 7 Not valid · 26 Not checked · 51 Own results and interpretations

29 more claims

Create a free account to see all 88 claims and to verify the 19 references still open.

See all claims — free account
  • Probably not a substantial claim. 32
  • It may state two separate claims as one. 26
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Show all 156 references

Research works printed but not linked: 14

  1. 1988, System Identification
  2. 1990, Time Series Analysis, Forecasting and Control
  3. 2019, Nonlinear System Identification: A User-Oriented Roadmap, 28
  4. 2021, Sensor Fault Label Identification for Robust Structural Health Monitoring, 159
  5. 1999, System Identification: Theory for the User
  6. 2002, New Introduction to Multiple Time Series Analysis
  7. 2007, Filtering and System Identification: A Least Squares Approach
  8. 1968, On-the-Line Analysis of Random Vibrations
  9. 2018, Variational Bayesian Monte Carlo, NIPS'18, 1
  10. 2013, UKF Estimation of SP-TARMA Models for the Identification of Time-Varying Structures, 1450
  11. 2023, On Off-Line and On-Line Bayesian Filtering for Uncertainty Quantification of Structural Deterioration, J. Struct. Dyn
  12. Rytter, A., 1993, “Vibrational Based Inspection of Civil Engineering Structures,” Ph.D. thesis, University of Aalborg, Denmark, p. 206.https://vbn.aau.dk/en/publications/vibrationalbasedinspectionofcivilengineeringstructures
  13. 2017, Prognostics and Health Management of Engineering Systems. An Introduction
  14. Harris, 2009, Vibration Data Analysis, 19.1

Cited by

Metadata

Authors
Kamariotis, A., Vlachas, K., Ntertimanis, V., Koune, I., Cicirello, A., & Chatzi, E.
Affiliations
  • ETH Zurich (CH)
  • Department of Civil, Environmental, and Geomatic Engineering, ETH Zurich , Stefano-Franscini-Platz 5, Zurich 8093, Switzerland
  • Delft University of Technology (NL)
  • University of Cambridge (GB)
Corresponding author
  • Kamariotis, Antonios
  • Vlachas, Konstantinos
  • Koune, Ioannis
  • Cicirello, Alice
  • Chatzi, Eleni
Venue
ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part B: Mechanical Engineering
Published
2024-12-09
Volume
11
Issue
1
ISSN
  • 2332-9017 (ISSN-L)
  • 2332-9025
OpenAlex
W4404301691
Cited by
15
References
145
Citations by year
2026: 6 · 2025: 9
FWCI
2.27
Type
journal article
Language
en
Open access
Locations
  • ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems Part B Mechanical Engineering (publishedVersion, cc-by)
  • Repository for Publications and Research Data (ETH Zurich) (publishedVersion, cc-by)
  • ETH Zürich Research Collection
Funders
  • UK Research and Innovation (100014013, 101086413)
  • Agence Nationale de la Recherche
  • Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (101086413, 200021)
  • Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWA.1431.20.002)
  • HORIZON EUROPE Framework Programme (101086413)
  • Staatssekretariat für Bildung, Forschung und Innovation
Topics
  • Physical Sciences › Engineering › Civil and Structural Engineering › Structural Health Monitoring Techniques
  • Physical Sciences › Engineering › Computational Mechanics › Probabilistic and Robust Engineering Design
  • Physical Sciences › Engineering › Control and Systems Engineering › Control Systems and Identification
Keywords
uncertainty quantification, structural health monitoring, uncertainty taxonomy, parameter uncertainty, model-form uncertainty, model parameter uncertainty, measurement uncertainty, inherent variability, system identification, model updating, soft sensing, damage detection, prognostics and health management

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-11-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.