A Two-Stage Deep Learning Method with Novel Damage Sensitive Features for Structural Damage Identification

Fukunaga-Koontz feature transformation for statistical structural damage detection and hierarchical neuro-fuzzy damage localisation

Purpose Aiming at the problems of high cost, high risk and traffic impact of in situ load testing of Bridges after fire, this paper proposes a fire bridge structural damage identification method based on a fire dynamics simulator (FDS)-Kriging model modification and a unit modal strain energy damage index, so as to effectively assess the damage of bridge structures after fire. Determine whether it can continue to be used and whether it affects driving safety. Design/methodology/approach Propose a fire bridge structural damage identification method to establish the FDS-Kriging model correction. Findings The finite element model of the fire bridge modified using FDS-Kriging approach can reflect the actual situation of the bridge structure more accurately, and the damage location and degree identified by the modal strain energy method are consistent with the actual bridge. Therefore, the effectiveness of the proposed method has been verified, and a new strategy for fire bridge damage identification has been provided. Research demonstrates that the finite element model of the fire-damaged bridge, modified using the FDS-Kriging method, can more accurately reflect the actual structural state. Moreover, the damage location and severity identified via the modal strain energy method align well with the actual bridge conditions. Thus, the effectiveness and engineering applicability of the proposed method are validated, offering a comprehensive new strategy and technical support for the integrated performance evaluation, damage identification and reliability analysis of fire-damaged bridges. Originality/value By determining the finite element correction interval and combining it with the Kriging model, the method realizes the fine correction of the finite element model of the fire bridge. Structural damage identification is carried out by using the unit modal strain energy damage index. This verifies the effectiveness and engineering applicability of the proposed method, which offers a novel approach and technical support for the comprehensive performance evaluation, damage identification and reliability analysis of fire-damaged bridges.

Distributed long gauge strain sensing technology has solved the problem of difficult identification of local damage in traditional ‘point’ monitoring, and has received extensive attention in the field of structural damage identification. Owing to the inevitable presence of measurement noise and environmental factors in the macro strain response measurement, a single damage index has also underlined some drawbacks generally arising when multiple damages occur, or errors affect the identified dynamic properties of the systems. To address these challenges, this paper proposes a data fusion method based on the Dempster–Shafer evidence theory, relying on distributed strain sensing technology. The identification results of the modal macro strain-based index and quasi-static macro strain energy-based damage index are fused to make a comprehensive decision on structural damage location. Damage identification studies are conducted on different types of structures under impact loads and random wind loads to verify the effectiveness and accuracy of the proposed data fusion method in the case of single and multiple damage conditions. The results show that the proposed data fusion method can accurately identify the damage location and effectively reduce misjudgment on undamaged locations; it has potential application value in practical structural health monitoring.

Recent advances in deep learning have enhanced vibration-based damage detection in civil infrastructure. However, the limited availability of labeled damage data remains a major obstacle for real-world deployment. To address this, this study proposes a novel unsupervised anomaly detection framework that relies solely on vibration data from intact structures. The framework integrates a convolutional autoencoder (CAE) with deep support vector data description (DSVDD), combining reconstruction-based and one-class classification approaches to improve sensitivity to subtle damage-induced changes. Time–frequency features are first extracted using the Fourier Synchrosqueezed Transform, which is transformed into images that amplify damage-sensitive patterns. The CAE-DSVDD model, tailored for multisensor structural health monitoring data under real-world operational conditions, is trained exclusively on undamaged-state data. A damage index is then computed by combining reconstruction error and latent-space deviation. Model performance is optimized through hyperparameter tuning, including a balancing parameter between the dual objectives. Unsupervised thresholds are evaluated to enable label-free classification. The framework is validated using field monitoring data from a real-world damaged truss bridge, demonstrating its high detection accuracy to real-world datasets. Additional experiments assess the framework’s robustness under varying loading conditions, input representations, sensor configurations, and its potential for damage localization. The proposed method achieved an F1-score of 98.46% under forced vibration, with consistent performance across low-severity damage cases. The overall results indicate the proposed framework provides an effective solution for unsupervised structural damage detection, offering a scalable and data-efficient approach for real-world infrastructure monitoring.

Structural damage identification using strain mode differences by the iFEM based on the convolutional neural network (CNN)

Damage identification in steel girder bridges using modal strain energy-based damage index method and artificial neural network

New Damage Identification Method for Operational Metro Tunnel Based on Perturbation Theory and Fuzzy Logic

In the past few decades, various techniques for structural damage identification based on dynamic properties have been widely studied, among which modal curvature is the research hotspot, due to its high sensitivity to local damage. To realize story-level structural damage identification of shear buildings, this paper first derives the patterns of variation in modal curvature of the damaged structure. Then, a new damage index sensitive to local damage called the Modal Curvature Change modified with Softmax (MCCSM for short), is proposed to improve the accuracy of structural damage identification. Afterward, the validity of the damage index is verified by a shaking table test of a scaled RC frame structure model and its corresponding numerical model. The research results show that the damage location indicated by MCCSM corresponds well with the damage phenomenon in the shaking table test and numerical simulation; compared with using the theoretical mode shape, the MCCSM index can give comparable damage identification results using the mode shape obtained through operational modal analysis from vibration time-history data, which proves the feasibility in engineering practices; through numerical simulation, it is proved that the MCCSM index is applicable in a wide range of damage cases with different story position, number of damaged story, and damage extent.

A damage detection method using BP neural network based on a novel damage index, the correlation characteristic of the acceleration response, is proposed, and is evaluated through the FEM simulation and experiment verification. On the basis of achievements in existence, the feasibility of using the correlation characteristic as damage index is validated theoretically. The damage detection for a simple-supported beam using the proposed method was FEM simulated. The results showed that the trained BP neural network can correctly detect the location and extent of damages in both single damage case and multi-damage case. A model test of a reinforced concrete simple-supported beam was performed to verify the validity and efficiency of the damage detection method. From the results of the model test, it is shown that the trained BP neural network can correctly locate the damage mostly detect the extent of damage. A conclusion is given that the novel damage detection method using the correlation characteristic of the acceleration response as damage index is feasible and efficient.

As one of the most common structural forms in port engineering, the operation environment of high-pile wharf is quite harsh and complex, and its pile foundation often produces structural damage of different degrees. Until now, there is a lack of efficient, safe and economic damage detection methods. A novel and precise real-time structural damage detection (SDD) method using both finite element modelling (FEM) and 1D convolutional neural networks (CNNs) is established in this study. The results indicate that the proposed method could accurately identify the presence and location of damage in real time. The results also demonstrated that the proposed 1D CNNs based model are more sensitive to the longitudinal and lateral displacement responses of the high-pile wharf structure.

The structural characteristic deflection shapes (CDS’s) such as mode shapes and operational deflection shapes are highly sensitive to structural damage in beam- or plate-type structures. Nevertheless, they are vulnerable to measurement noise and could result in unacceptable identification errors. In order to increase the accuracy and noise robustness of damage identification based on CDS’s using vibration responses of random excitation, joint approximate diagonalization (JAD) technique and gapped smoothing method (GSM) are combined to form a sensitive and robust damage index (DI), which can simultaneously detect the existence of damage and localize its position. In addition, it is possible to apply this approach to damage identification of structures under ambient excitation. First, JAD method which is an essential technique of blind source separation is investigated to simultaneously diagonalize a set of power spectral density matrices corresponding to frequencies near a certain natural frequency to estimate a joint unitary diagonalizer. The columns of this joint diagonalizer contain dominant CDS’s. With the identified dominant CDS’s around different natural frequencies, GSM is used to extract damage features and a robust damage identification index is then proposed. Numerical and experimental examples of beams with cracks are used to verify the validity and noise robustness of JAD based CDS estimation and the proposed DI. Furthermore, damage identification using dominant CDS’s estimated by JAD method is demonstrated to be more accurate and noise robust than by the commonly used singular value decomposition method.

Large scale tests with progressive damage on a prestressed concrete highway bridge have been performed to investigate the sensitivity of several damage detection, localization, and quantification methods based on modal parameters. To investigate the quality of modal parameters, the data set of one damage step was analyzed by several output-only identification techniques. Although the bridge was severely cracked, natural frequencies as well as mode shapes display only minor changes. However, the relative changes of mode shapes are larger than those observed for natural frequencies. A novel damage indicator, called mode shape area index, based on changes of mode shapes, has been developed and found as the most sensitive damage detection approach. Damage detection or localization via changes of the flexibility matrix performed better than natural frequencies or mode shapes alone. The application of the direct stiffness calculation and a sensitivity-based model update technique showed results having a high level of ambiguity about the location and quantification of damage also at the highest damage level. Evaluating the information collected in this study the test results indicate that an early stage damage identification in prestressed concrete bridges is hardly possible because of the nearly complete recovery of stiffness after closing of cracks in prestressed concrete and the effect of environmental parameters on modal data.

Considering the issue of misjudgment in railway bridge damage identification, a method combining the step-by-step damage detection method with the statistical pattern recognition is proposed to detect the structural damage of a railway continuous girder bridge. The whole process of damage identification is divided into three identification sub-steps, namely, damage early warning, damage location, and damage extent identification. The multi-class pattern classification algorithm of C-support vector machine and the regression algorithm of ε-support vector machine are engaged to identify the damage location and damage extent, respectively. For verifying the proposed method, both of the proposed method and the optimization method are used to deal with the measured data obtained from a specific railway continuous girder model bridge. The results show that the proposed method can not only identify the damage location correctly, but also obtain the damage extent which is consistent with the experimental results accurately. By uncoupling finite element analysis and damage identification, normalizing the index, and seeking the separation hyper plane with maximum margin, the proposed method has more favorable advantages in generalization and anti-noise. As a result, it has the ability to identify the damage location and extent, and can be applied to the damage identification in real bridge structures.

PurposeThe paper aims to identify both the location and severity of damage in complex framed buildings using limited noisy vibration measurements. The study aims to directly adopt incomplete measured mode shapes in structural damage identification and effectively reduce the influence of measurement errors on predictions of structural damage.Design/methodology/approachDamage indicators are properly chosen to reflect both the location and severity of damage in framed buildings at element level for braces and at critical point level for beams and columns. Basic equations for an iterative solution procedure are provided to be solved for the chosen damage indicators. The Tikhonov regularisation method incorporating the L‐curve criterion for determining the regularisation parameter is employed to produce stable and robust solutions for damage indicators.FindingsThe proposed method can correctly assess the quantification of structural damage at specific locations in complex framed buildings using only limited information on modal data measurements with errors, without requiring mode shape expansion techniques or model reduction processes.Research limitations/implicationsFurther work may be needed to improve the accuracy of inverse predictions for very small structural damage from noisy measurements.Practical implicationsThe paper includes implications for the development of reliable techniques for rapid and on‐line damage assessment and health monitoring of framed buildings.Originality/valueThe paper offers a practical approach and procedure for correctly detecting structural damage and assessing structural condition from limited noisy vibration measurements.

This paper presents an experimental study on the detection and localization of damage in a steel plate using vibration-based methods. Dynamic impulse tests were conducted on both undamaged and damaged plate to determine changes in modal parameters, such as eigenfrequencies and mode shapes. Several damage detection methods, including the Modal Assurance Criterion (MAC), Coordinate Modal Assurance Criterion (COMAC), Mode Shape Curvature (MSC), Mode Shape Curvature Square (MSCS), and Damage Index (DI), were applied. The results showed that while MAC and COMAC effectively detected damage, they did not provide information about the location. MSC and MSCS indices demonstrated improved damage localization, with the MSCS being particularly effective. The DI index provided the clearest and most accurate representation of the damage location. These findings highlight the potential of these methods for non-destructive testing and structural health monitoring.