Unsupervised Vision‐Based Structural Anomaly Detection and Localization with Reverse Knowledge Distillation

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.

In principal, a proper analysis of the dynamic response of a structure can provide general indicators of its operational conditions. When the dynamic response changes due to a variation of the physical properties of a structure, then one may conclude that some kind of damages has occurred. This paper presents investigation of the robustness and comparison of four simple methodologies to both identify and quantify the damages in structures, based on the use of Frequency Response Functions (FRF) signals, Principal Component Analysis technique (PCA) and Transmissibility. A steel beam with constant rectangular cross-section is used to compare the proposed approaches. At first, nine damaged scenarios are created and for each of them numerical examples are discussed; a database of FRFs is measured using modal testing. Then, PCA theory is applied to the FRF matrix and global damage detection and quantification indices are defined by using the first 3 Principal Components; Hotelling's T-squared distribution is also applied and by using transmissibility two other indicators, Transmissibility Damage Indicator and Weighted Damage Indicator, are computed for the assessment of damage. The reported examples show that all proposed methods are able to detect and quantify damages at the initial stage.

Application of two-stage evaluation and optimization update methods for the structural damage detection of a portal beam structure

The [Formula: see text] regularization is usually used to deal with the problems of under-determinacy and measurement noise for the conventional sensitivity-based model updating damage detection methods. However, the [Formula: see text] regularization technique often provides overly smooth solutions and thus cannot exhibit the sparsity of the structural damage due to the promotion of the 2-norm term on smoothness. In the study, a structural damage detection method is proposed based on an improved modal flexibility sensitivity function and an iterative reweighted [Formula: see text] (IR[Formula: see text] regularization. Specifically, the sensitivity function is established by introducing changes in the mode shapes into the derivative of eigenvalue and can be applied to identify the localized damage more accurately. Additionally, IR[Formula: see text] regularization is proposed to deal with the ill-posed problem of damage detection in a noisy environment. The proposed IR[Formula: see text] regularization is compared with the [Formula: see text] and [Formula: see text] regularizations through a numerical and an experimental examples. The numerical and experimental results indicate that the IR[Formula: see text] regularization can more accurately locate and quantify the single and multiple damages under the noise situation. The maximum identification errors are only 5.16% and 5.67%, respectively. Moreover, compared to the basic modal flexibility sensitivity function, the improved function is more sensitive to the damage. The maximum identification error of the improved function is less than 6%, while the relative errors are significantly larger in the basic function.

The concept of using tunable piezoelectric transducer circuitry to enhance the performance of frequency-shift based damage detection has been recently explored. While previous studies have shown promising features, the advantage of such an approach has not been fully realized due to the limitations in the associated damage identification algorithms. This research aims at advancing the state-of-the-art of this concept, where an iterative eigenvector updating method is developed to improve the accuracy of the damage detection scheme. The basic principle of this method is to iteratively update the eigenvectors of the damaged structure such that the information of the damage-induced eigenvector change can be included in the damage detection formulation without actually measuring the vibration modes. The inductance tuning method is extended to structures with multiple transducers and circuits. Numerical analysis on a plate structure demonstrated the effectiveness and accuracy of the new method in detecting both the location and severity of the structural damage. Nomenclature K = generalized stiffness matrix of the electro-mechanical integrated system e j K = elemental stiffness matrix of the j th element c K = coupling matrix between the mechanical structure and piezoelectric circuit p K = inverse capacitance matrix of the piezoelectric circuit s K = stiffness matrix of the mechanical structure only M = generalized mass matrix of the electro-mechanical integrated system s M = mass matrix of the mechanical structure only q = displacement vector of the mechanical structure Q = electrical charge flow vector in the circuit i ω = i th resonant frequency of the healthy integrated electro-mechanical system () d i ω = i th resonant frequency of the damaged integrated electro-mechanical system i λ = i th eigenvalue of the healthy integrated electro-mechanical system

In this study, a two-stage damage detection procedure using a new approximation of modal strain energy (MSE) value is developed to boost the accuracy of structural damage localization in plate-type structures. The MSE is calculated using the center differential method for both intact and damaged states. The new formulas to approximate MSE are developed based on the nine-node isometric quadrilateral element that eliminates boundary conditions’ effect on damage detection results. The modal assurance criteria (MAC) is used as an indicator to choose the mode shapes with the best damage detection results. A global and a local MSE are proposed to enhance the accuracy of detecting the location of structural damage. An aluminum plate is used to verify the effectiveness of the proposed procedure with several damage scenarios. The results show that the new estimation of MSE value combined with MAC and the two-step MSE procedure can improve the accuracy of identifying the damage location compared to the original MSE procedure.KeywordsDamage detectionModal assurance criteriaModal strain energyPlate-type structureStructural health monitoring

This article investigates whether higher classification accuracy can always be achieved by utilizing a pre-trained deep learning model as the feature extractor in the Bag-of-Deep-Visual-Words (BoDVW) classification model, as opposed to directly using the new classification layer of the pre-trained model for classification. Considering the multiple factors related to the feature extractor -such as model architecture, fine-tuning strategy, number of training samples, feature extraction method, and feature encoding method—we investigate these factors through experiments and then provide detailed answers to the question. In our experiments, we use five feature encoding methods: hard-voting, soft-voting, locally constrained linear coding, super vector coding, and fisher vector (FV). We also employ two popular feature extraction methods: one (denoted as Ext-DFs(CP)) uses a convolutional or non-global pooling layer, and another (denoted as Ext-DFs(FC)) uses a fully-connected or global pooling layer. Three pre-trained models—VGGNet-16, ResNext-50(32×4d), and Swin-B—are utilized as feature extractors. Experimental results on six datasets (15-Scenes, TF-Flowers, MIT Indoor-67, COVID-19 CXR, NWPU-RESISC45, and Caltech-101) reveal that compared to using the pre-trained model with only the new classification layer re-trained for classification, employing it as the feature extractor in the BoDVW model improves the accuracy in 35 out of 36 experiments when using FV. With Ext-DFs(CP), the accuracy increases by 0.13% to 8.43% (averaged at 3.11%), and with Ext-DFs(FC), it increases by 1.06% to 14.63% (averaged at 5.66%). Furthermore, when all layers of the pre-trained model are fine-tuned and used as the feature extractor, the results vary depending on the methods used. If FV and Ext-DFs(FC) are used, the accuracy increases by 0.21% to 5.65% (averaged at 1.58%) in 14 out of 18 experiments. Our results suggest that while using a pre-trained deep learning model as the feature extractor does not always improve classification accuracy, it holds great potential as an accuracy improvement technique.

A structural damage detection method by integrating data fusion and random forests was proposed. The original acceleration signals were translated into energy features by wavelet packet decomposition. Then the processed energy features were fused into new energy features by data fusion. This can further enlarge the differences among all types of damages. Finally, random forests as an effective classifier was used to detect the multiclass damage. Numerical study on the benchmark model and an eight-storey steel shear frame structure model was carried out to validate the accuracy of the proposed damage detection method. The experiment results indicate that the damage detection method based on random forests and data fusion can improve damage detection accuracy in comparison with random forests alone, support vector machine alone, and support vector machine and data fusion techniques. Moreover, the proposed method has significantly better stability than several other methods.

In this paper, we propose a complementary self-supervised mask model based on teacher-student networks. This model contains a student network, a teacher network, and a mask prediction module. The student's network is an encoder structure, and the teacher's network consists of encoders and decoders. The teacher and student network encoders are used for learning image representations and have the same network structure and model parameters. The pre-training has two pre-tasks: First, the mask image block representation predicted by the decoder in the teacher network predicts the actual image pixels through the mask prediction module. Then, we introduce a comparative learning loss function to compare the outputs of the teacher and student modules in representation space. This paper proposes a complementary masking mechanism to reduce the gap between upstream and downstream mismatches in the pre-training model based on mask image modeling (MIM). For example, a complete picture is an input into the teacher and the student network. For the teacher network, the input picture is randomly masked off, for example, 75 %; the student network masks the remaining part of the input picture, 25 %. The student network masks the rest (25%) of the input image. The pre-trained model proposed in this paper has been pre-trained on COCO and other data sets, and downstream tasks are performed on four conventional data sets. By comparing with some of the latest self-supervised pre-trained models, it is proved that the pre-trained model proposed in this paper can learn better representational information.

A damage detection process can be significantly enhanced if the nonlinear effects can be used when extracting damage-sensitive features from measured signals. The coherence function is typically used for nonlinearity identification by determining the extent of the output power linearly correlated with the input. However, the excitations are usually difficult to measure in actual tests. To overcome this limit, this article presents a signal-segments cross-coherence method for nonlinearity identification. By defining a signal-segments cross-coherence matrix and signal-segments cross-coherence index, the method can visually and quantitatively indicate the presence of nonlinearity. The innovation of the new method is that the coherence analysis process only depends on a single output signal, where input and baseline signals are not required. Then, a novel structural damage localization index is constructed by multi-point comparing of the signal-segments cross-coherence indices, based on the assumption that all the measured signals from different points on the structure have the same frequency bandwidth and components. To meet this requirement, a newly proposed signal decomposition method called analytic mode decomposition method is adopted. Numerical studies on a duffing oscillator and a 10 degree-of-freedom spring-damping-mass system were performed to demonstrate the nonlinear identification process and investigate the effectiveness and robustness of the signal-segments cross-coherence-based damage detection method. The results show that the signal-segments cross-coherence method can effectively indicate the appearance of nonlinearity by the signal-segments cross-coherence matrix and signal-segments cross-coherence index with strong noise robustness. And the proposed damage localization index can accurately detect the weak nonlinear damage even with severely noise-polluted signals. To further investigate the applicability of the new method, an experimental study was conducted on a steel simplified scale model of a monopile offshore wind turbine support structure. The results demonstrate that the proposed signal-segments cross-coherence method and the new damage localization index can be used to detect the bolt-loosening damage of the steel structure only with output signals.

Effect of response type and excitation frequency range on the structural damage detection method using correlation functions of vibration responses

This article, written by Special Publications Editor Adam Wilson, contains highlights of paper SPE 190087, “Unsupervised Statistical Learning With Integrated Pattern-Based Geostatistical Simulation,” by Q. Li and R. Aguilera, SPE, University of Calgary, prepared for the 2018 SPE Western Regional Meeting, Garden Grove, California, USA, 22–27 April. The paper has not been peer reviewed. This paper presents a new geostatistics modeling methodology that connects geostatistics and machine-learning methodologies, uses nonlinear topological mapping to reduce the original high-dimensional data space, and uses unsupervised-learning algorithms to bypass problems with supervised-learning algorithms. The algorithm presented is a neural topology-preserving pattern-based geostatistical simulation algorithm that integrates the self-organizing map (SOM) concept and its updated version—growing self-organizing map (GSOM)—with an unsupervised competitive learning structure. Introduction In oil and gas reservoir modeling, any model construction faces challenges of limited data to some extent. The heuristic behind all geostatistical techniques is the implicit existence of statistical relationships among available data. “Data” here is a broad term; it could be discrete points, such as porosity or permeability at certain locations, but it also could be training images (TIs), which are used in this work. Using TIs as input data originated with multiple-point geostatistics. The aim was to overcome the limitations of using traditional two-point statistical variograms to describe geological continuity, especially in the case of curvilinear structures, which are quite common in nature, such as in fracture networks and geological fluvial structures. The authors write that geostatistics could benefit from the machine-learning or statistical-learning areas. Machine-learning tasks can be divided into two protocols, supervised learning and unsupervised learning. The difference depends on whether the input data have correct labels or not. For the investigation considered in this paper, after retrieving image patches from TIs, machine-learning algorithms were used to cluster those image patches into different classifications. If the correct clusters to which the image patches belong is known beforehand, the data can be said to have correct labels. Further, a large amount of these labeled data could be used for a model to learn. Supervised learning occurs when the model is finely tuned by guidance of these correct labels using an error-correction process. Unsupervised learning, on the other hand, occurs when neither how many clusters exist nor the correct clusters to which the image patches belong is known. Thus, without correct labels for guidance, this problem is identified as an unsupervised-learning problem. This is the first major drawback of applying machine-learning algorithms to geostatistical simulations. Two other important issues are encountered when performing pattern-based geostatistical simulations. The first is related to the large number of image patches. This large image-patch database contains high pattern redundancy, which could make pattern similarity comparison inefficient. The second issue is the high dimensionality of each image patch, which could typically be described by low-dimensional nonlinear-structure manifolds. To model and visualize these high-dimensional data, a nonlinearity dimensionality reduction should be sought.

Unity check method for structural damage detection

Non-linearities in the vibrations of elastic structures with a closing crack: A state of the art review

Structural health monitoring (SHM) based on long-term safety and efficacy has attracted much attention. The success of SHM methods depends on the information content of the measurements; in addition, there are challenges in processing SHM in terms of the optimal number and location of sensors. In this study, we provide a convenient method for structural damage detection using vibration data from plane steel frames. We apply 1-D convolutional neural networks (1D-CNNs) to present an influential sensor selection analysis to maintain performance and optimize the number and location of sensors. Finding the damage location based on the optimal number and location of sensors will significantly reduce the detection time and the number of installed sensors. The proposed structural damage detection method uses stress analysis to find positions with large displacements, then divides the detection zones, selects highly discriminate sensors, analyzes the influence of sensor combinations within each zone to achieve optimization of sensor number as well as position selection, and establishes CNN models of the overall structure for damage detection of the joints. The proposed method was practically validated on benchmark data from the Qatar University Grandstand Simulator (QUGS). The experimental results show that the sensor demand rate is only 16.67% compared to the previous optimal number and location of sensors with no restrictions on the mounting position. Moreover, only four CNN models are required to predict damage/undamaged for the entire structure (30 connections). The accuracy of damage detection is improved to 96.62% using sensor combinations with influence.