Bridge damage identification method and system based on random subspace and hypothesis testing

Abstract

The invention discloses a bridge damage identification method and system based on random subspace and hypothesis testing, and the method comprises the steps: firstly obtaining bridge structure response data, and then carrying out the modal identification of the data through a covariance-driven random subspace identification method, and obtaining a bridge modal parameter; then screening real modals based on the obtained modal parameters to generate a stability diagram, and obtaining final modal parameters through feature extraction; and finally, accurately identifying whether the bridge is damaged or not according to the final modal parameters by using a double-sample bilateral t test method. The method effectively overcomes the difficulty of the traditional monitoring technology in the aspects of estimation error and environmental factor interference, provides a more reliable and efficient solution for bridge damage identification, and has important significance for guaranteeing the safe operation of the bridge.

Description

Technical Field

[0001] This invention belongs to the field of bridge damage detection technology, and relates to a bridge damage identification method and system based on random subspace and hypothesis testing. Background Technology

[0002] In recent years, my country's urban construction and major infrastructure development have flourished, achieving remarkable results. Transportation networks have been continuously improved, and skyscrapers have sprung up, greatly enhancing the convenience and comfort of people's lives and powerfully promoting rapid economic and social development. However, behind this prosperous scene lie many hidden safety hazards that cannot be ignored. Over time, building structures are exposed to the natural environment, enduring the erosion of wind, sun, rain, and freezing temperatures, while building materials gradually age. During long-term operation, the performance of many building structures continuously deteriorates, internal damage accumulates, leading to a series of problems such as reduced load-bearing capacity and decreased structural durability. If these problems are not detected and addressed in a timely manner, they could potentially lead to catastrophic accidents, seriously threatening the lives and property of the people. Among various types of buildings, large-scale structures, especially bridges, play a vital role in people's lives and work due to their unique connectivity. As key nodes in transportation hubs, bridges bear a large volume of traffic, connecting cities and villages, and regions, and are irreplaceable in ensuring the smooth transport of goods and the movement of people. Research shows that, apart from human factors such as ship collisions, many bridge structures exhibit various signs of damage before accidents occur. Therefore, ensuring the safe operation of bridges and other large-scale structures, as well as monitoring their structural health, has become an important research area concerning the safety of people's lives and property. Currently, bridge health monitoring primarily relies on structural health monitoring technology. The core idea of ​​this technology is to extract parameters sensitive to structural damage from long-term dynamic signals collected by sensors, and to achieve health diagnosis through continuous observation of these damage-sensitive features. Theoretically, this method can effectively detect structural damage and achieve the goal of bridge health status monitoring. However, in practical applications, this method struggles to achieve the expected results. On the one hand, many structural features cannot be directly measured and must be estimated from measurement data using system identification techniques. For example, modal characteristics need to be estimated using vibration response data such as acceleration or strain, and inappropriate modal identification methods inevitably introduce estimation errors. On the other hand, almost all modal characteristics are not only sensitive to structural damage but also extremely sensitive to changes in environmental parameters such as atmospheric temperature, wind speed, and relative humidity. This necessitates ensuring the accuracy of the estimated features while fully considering the interference of complex environmental factors on structural damage assessment when applying this monitoring scheme, significantly increasing the difficulty and complexity of monitoring. Summary of the Invention

[0003] The purpose of this invention is to solve the technical problems in the prior art where many features of bridge structures cannot be measured, resulting in large errors in the estimation of damage-sensitive parameters, and damage identification results are easily affected by the environment. This invention provides a bridge damage identification method and system based on random subspace and hypothesis testing.

[0004] To achieve the above objectives, the present invention employs the following technical solution: The first aspect of this invention provides a bridge damage identification method based on random subspace and hypothesis testing, comprising the following steps: Obtain bridge structural response data; Modal identification was performed on the bridge structural response data using a covariance-driven random subspace identification method to obtain the bridge's modal parameters. Based on the modal parameters of the bridge, the true modes are selected and a stability graph is generated; the final modal parameters are obtained by feature extraction from the stability graph. Based on the final modal parameters, a two-sample bilateral method is used. Inspection methods are used to identify whether a bridge is damaged.

[0005] Furthermore, a covariance-driven random subspace identification method is used to identify the modalities of the bridge structural response data, specifically as follows: The Hankel matrix is ​​constructed using bridge structural response data; the row space of the Hankel matrix is ​​divided into past and future parts. The covariance of the past and future parts of Hankel's row space is calculated separately to obtain the Toeplitz matrix. and ; Based on the Toeplitz matrix and Calculate the state transition matrix; The state transition matrix is ​​orthogonally similar to diagonalize, and the modal parameters of the bridge are calculated.

[0006] Furthermore, the Hankel matrix is ​​constructed using bridge structural response data; the row space of the Hankel matrix is ​​divided into past and future parts, specifically:

[0007] in, The Hankel matrix; the Hankel matrix includes Data collected from each channel Displacement data, with a time delay of [number] data points. , ; The past part is The future part is: or ;in, , including from arrive The line, but not including row; include to The line, excluding OK.

[0008] Furthermore, the aforementioned and The calculation methods are as follows:

[0009] in, This indicates transpose.

[0010] Furthermore, the state transition matrix is:

[0011] in, ; ; This is the state transition matrix; This is the output matrix; This is the input influence matrix.

[0012] Furthermore, the modal parameters include natural frequency, damping ratio, and mode shape.

[0013] Furthermore, the selection of true modes is based on modal confidence, damping ratio check, and complex conjugate pole check as discriminant indicators.

[0014] Furthermore, modal frequency features were extracted by using the DBSCAN clustering method to extract features from the stability graph.

[0015] Furthermore, the method of using a two-sample bilateral t-test to identify whether the bridge is damaged specifically involves: Collect reference samples of the bridge's health status to obtain health samples; Assuming the final modal parameters and the healthy samples come from the same population, the two-sample bilateral t-test method is used to calculate the t-value and the corresponding p-value. Assuming the final modal parameters are not from the same population as the healthy samples, a two-sample bilateral t-test is used to calculate the t-value and the corresponding p-value. Based on different assumptions and the calculated t-values ​​and corresponding p-values, it is determined whether the bridge is damaged.

[0016] A second aspect of the present invention provides a bridge damage identification system based on random subspace and hypothesis testing, comprising: The data acquisition module acquires bridge structural response data; The modal parameter identification module uses a covariance-driven random subspace identification method to identify the modal parameters of the bridge structure response data. The modal parameter filtering module filters out the true modes based on the modal parameters of the bridge and generates a stability map; the final modal parameters are obtained by extracting features from the stability map. The damage identification module, based on the final modal parameters, employs a two-sample, two-sided approach. Inspection methods are used to identify whether a bridge is damaged.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention discloses a bridge damage identification method based on random subspace and hypothesis testing. It employs a covariance-driven random subspace identification method, constructing a Hankel matrix and dividing its row space into past and future parts. The Toeplitz matrix and state transition matrix are then calculated, and the state transition matrix is ​​orthogonally similarly diagonalized to obtain modal parameters. This process fully utilizes the characteristics of bridge structural response data, resulting in more accurate modal parameters such as natural frequencies, damping ratios, and mode shapes. Simultaneously, modal confidence, damping ratio checks, and complex conjugate pole checks are used as discriminant indicators to screen for true modes. The DBSCAN clustering method is combined to extract features from the stability graph to obtain the final modal parameters, further eliminating spurious modal interference and ensuring the authenticity and stability of the modal parameters, laying a solid foundation for subsequent damage identification. In the damage identification stage, a two-sample bilateral t-test is used. By collecting reference samples from a healthy bridge state, a hypothesis test is performed to determine whether the final modal parameters and healthy samples originate from the same population. The t-value and corresponding p-value are calculated to determine whether the bridge is damaged. This statistical hypothesis testing approach effectively quantifies the differences in modal parameter variations, improving the scientific rigor and reliability of damage identification. Whether the bridge structure is in a healthy state or damaged, rigorous statistical analysis enables accurate judgments, reducing the probability of misjudgments and omissions. This method is adaptable to the damage identification needs of bridges under different operating conditions, exhibiting good versatility and adaptability. It provides strong technical support for bridge safety monitoring and maintenance, helping to promptly detect potential bridge damage, ensure bridge operational safety, and extend bridge service life. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a diagram showing the distribution of vibration sensors on the bridge deck and piers according to an embodiment of the present invention. Figure 2 This is a flowchart of the bridge damage identification method based on random subspace and hypothesis testing according to the present invention. Figure 3 As an embodiment of the present invention, a modal frequency stability diagram is shown. Figure 4 The DBSCAN clustering method in this embodiment of the invention identifies the first six stable modal frequencies. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and marked in the accompanying drawings can generally be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0022] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0023] The present invention will now be described in further detail with reference to the accompanying drawings: See Figure 1 This invention discloses a bridge damage identification method based on random subspace and hypothesis testing, comprising the following steps: The process consists of three stages: the first stage is the identification of modal parameters; the second stage is the screening and feature extraction of the modal identification results; and the third stage is the use of a two-sample bilateral t-test to eliminate the influence of the environment on damage assessment and accurately identify the damage status of the bridge structure.

[0024] 1. First stage (modal recognition stage) Step 1-1: Construct the Hankel matrix using the raw data of the bridge structure response. Assume the acceleration sensor system has... Each acquisition channel (usually deployed along the bridge deck) acquires data over a period of time. Each displacement data point is used to arrange the data collected from each channel in rows to form a single displacement data point. A dimensional original data matrix, where each column represents the displacement data measured by each accelerometer at the same time. express, Take time delay At time n, the specific construction method of the Hankel matrix is ​​as follows: Let , will the size be The time window slides to the right along the original data matrix. The first to second parts of the Hankel matrix l OK, The first component of the Hankel matrixl +1~2 l Row by row, slide the time window sequentially until the window contains the first row. By listing the data, we can obtain a value of [size]. The Hankel matrix. To represent the structure of this matrix, this invention introduces the notation... It represents the vector arrive The rows that make up the matrix. Specifically, the complete Hankel matrix is ​​denoted as... This indicates that it contains from arrive All of the components OK:

[0025] This is the Hankel matrix. The row space of the Hankel matrix is ​​divided into two parts: "past" and "future." The "past" part is denoted as... From the front Line composition, from arrive The future part can be defined in two ways depending on the specific application: (1) , including from arrive The rows, but not including the final vector. (2) including to The rows, omitting intermediate vectors .

[0026] Steps 1-2: [Regarding...] and , and Covariance calculations were performed separately to obtain two Toeplitz matrices. and :

[0027] Steps 1-3: [Regarding...] Decompose it. First, decompose... Singular value decomposition is performed to extract the main features of the system and suppress the influence of measurement noise, thereby more accurately characterizing the dynamic behavior of the system. Let... The modal order of the bridge is indicated by superscript symbols. To emphasize the dependence of the matrix and parameters on the selected system order, superscript symbols are used throughout the identification process. . The singular value decomposition results are:

[0028] In the above formula, and The dimension of the matrix is , The dimension of the matrix is Perform the following identity transformations on the above equation:

[0029] in, and It can be represented as , .

[0030] The matrix in the above formula Since it is a non-singular matrix, it can function as a similarity transformation matrix. To simplify calculations, we take... ,have:

[0031]

[0032] and It can also be expressed as:

[0033]

[0034] in, The state transition matrix describes the dynamic relationship of the system state evolution over time. The output matrix maps the internal state of the system to the observable output. Let be the input influence matrix, representing the direct impact of external inputs on changes in the system state. Then we have:

[0035] The superscript "+" indicates that the Moore-Penrose pseudo-inverse operation is performed on the matrix.

[0036] Steps 1-4: Calculate the modal parameters of the bridge. For the state transition matrix... Perform orthogonal similarity diagonalization:

[0037] in, It is by Discrete-time complex eigenvalues The diagonal matrix formed Therefore The eigenvectors are orthogonal matrices composed of column vectors. Since bridge vibration can be considered a continuous system, based on the relationship between the eigenvalues ​​of the state matrices of discrete and continuous systems, the eigenvalues ​​of the bridge system can be... Represented as:

[0038] in, This refers to the sampling frequency of the bridge vibration monitoring sensor system.

[0039] The three modal parameters of a bridge system, namely natural frequency, damping ratio, and mode shape, are respectively represented by... , , This means that, based on the above, the modal order can be obtained. n Under the given conditions, the calculation expressions for the three candidate values ​​of modal parameters are as follows:

[0040] in, yes The first of the matrix k Column vectors.

[0041] 2. Second stage (pseudomodal removal and feature extraction stage) Step 2-1: Determine whether the modes obtained in the first stage can represent the characteristics of the real bridge modes, identify the real modes, and eliminate pseudo-modes. To distinguish between real and pseudo-modes, three modal discrimination indices are used to screen out pseudo-modes: Modal Assurance Criterion (MAC), damping ratio check, and complex conjugate pole check.

[0042] MAC is a widely used statistical metric for evaluating the consistency between two mode shapes. It is defined as follows:

[0043] Among them, and These are two mode shape vectors, with superscripts. This represents the conjugate transpose operation. The MAC value ranges from 0 to 1, with values ​​close to 1 indicating high correlation (i.e., the two modes are almost identical), and values ​​close to 0 indicating orthogonality or no correlation. In practical applications, a MAC threshold is typically set (usually between 0.9 and 0.99) to identify and eliminate redundant or numerically unstable modes. This criterion is particularly effective in clustering physically meaningful modes of different model orders and in discarding spurious modes introduced by numerical noise or overfitting.

[0044] The criteria for damping ratio testing and complex conjugate pole testing are as follows: Damping ratio check: Damping ratio in the actual bridge structure for each mode. It must be positive and below 0.2, otherwise the structure is unstable, therefore only 0 < 0 is retained. Modes <0.2.

[0045] Complex conjugate pole check: If the eigenvalue satisfy If the mode represents an unstable structure, it is identified as a pseudo-mode.

[0046] Based on the above method, the modal frequency data obtained in the first stage are filtered to remove pseudo-modes, retaining only the real modes with clear physical meaning, and the system order is then determined. n =2, 4, 6, ..., 100 (step size 2), repeat the above steps to finally form a stability plot of the bridge system order and natural frequencies. A stability plot is a two-dimensional distribution graph plotted with the system order as the horizontal axis and the corresponding modal frequencies as the vertical axis. It can intuitively reflect the variation patterns of identified modal frequencies at different orders. Typically, the physical modal frequencies exhibit a continuous and stable trajectory as the system order changes, while pseudo-modal frequencies show discrete, drifting, or discontinuous distribution characteristics. Therefore, by analyzing the stability plot, we can effectively distinguish between real and pseudo-modal frequencies, providing a solid data foundation for subsequent modal feature extraction and damage detection.

[0047] Step 2-2: Extract modal frequency features using the DBSCAN clustering method. After obtaining modal frequency samples across multiple model orders, the DBSCAN algorithm is used to automatically cluster the data. The DBSCAN algorithm can identify clusters of arbitrary shapes without pre-setting the number of clusters and has good noise robustness. In the modal identification process of bridge health monitoring, the DBSCAN algorithm can perform cluster analysis on feature points in the modal stability map, thereby effectively extracting physical modes with relatively stable frequencies and damping, and further eliminating pseudo-modes caused by numerical errors or non-structural features. The specific steps are as follows: First, the cluster radius parameter (eps) and minimum number of sample points parameter (minPts) of the DBSCAN algorithm are set, and Euclidean distance is used as the metric for the density relationship between data points. Based on the set parameters, the DBSCAN algorithm is used to cluster the modal frequency data, dividing the modal frequency samples into several dense clusters, while automatically removing isolated noise points. The mean of the data points within each cluster is calculated, and this mean is used as the center frequency of the corresponding cluster. The center frequency is used as the candidate physical modal frequency. All obtained cluster center frequencies are sorted according to frequency magnitude, and the top 6 stable modal frequencies are selected as the output modal parameters for the final identification of the bridge structure.

[0048] Based on the above principles, accurate identification of natural frequencies can be achieved. Since the low-order modal frequencies (e.g., the 1st to 6th orders) of bridge structures exhibit similar regularities in their time-varying behavior, this invention uses the first modal frequency... As a damage-sensitive parameter, combined The numerical variation patterns under environmental influences are used to determine whether there is damage to the bridge structure.

[0049] 3. Third stage (structural damage assessment stage) In this stage, a two-sample bilateral t-test was used to determine bridge structural damage. The specific steps are as follows: Step 3-1: Collect reference samples under healthy conditions. First, collect a reference sample dataset when the bridge is in a healthy state, as a benchmark for subsequent damage assessment. The specific method is as follows: Select a specific time period during the bridge's service life when the environment is relatively stable (e.g., 2:00 AM to 3:00 AM daily in July) to collect vibration data; use accelerometers installed on the bridge structure to collect vibration signal data and extract the corresponding first-order modal frequencies; organize the first-order modal frequency samples obtained during the above time period to form a reference sample dataset under healthy conditions, denoted as... .

[0050] Step 3-2: Collect test samples during service. During the normal service period of the bridge, real-time vibration signals are continuously collected during the same time period as the reference sample dataset (e.g., 2:00 AM to 3:00 AM daily in July). Specifically, the first-order modal frequencies obtained from each monitoring during service are extracted; the extracted modal frequency samples are organized into a test sample dataset, denoted as... .

[0051] Step 3-3: Use a two-sample, two-sided t-test to determine bridge damage. The null and alternative hypotheses are set as follows: Null hypothesis The test sample and the healthy sample came from the same population; Opposing assumptions The test sample and the healthy sample did not come from the same population (i.e., the bridge was damaged).

[0052] Based on the above assumptions, a two-sample bilateral t-test was used to determine structural damage to the test samples, and the results for the reference samples were calculated separately. and test samples The mean, standard deviation, and sample size are determined, and the t-value is calculated using the following formula:

[0053] in, and These are the first-order frequency means of the test sample and the healthy sample, respectively. and These are the first-order frequency variances of the test sample and the healthy sample, respectively. and These represent the sample size for the test samples and the healthy samples, respectively.

[0054] Based on the calculated t-value and the corresponding p-value, determine whether to accept the null hypothesis: If we accept the original hypothesis (If the p-value is less than the preset value, such as 0.05), it indicates that the test sample comes from a healthy population and the bridge is in a healthy state; If we reject the null hypothesis If the p-value is greater than or equal to the preset value such as 0.05, it indicates that there is a significant difference between the test sample and the healthy sample, and the bridge may have structural damage.

[0055] The two-sample bilateral t-test method based on first-order frequency described above can sensitively and accurately identify the damage status of bridge structures, thereby issuing timely warnings before major damage occurs and providing effective guidance for bridge maintenance and repair work.

[0056] To verify the effectiveness of the method proposed in this invention, one embodiment of the invention uses the classic dataset of the Swiss Z24 bridge to establish a bridge structural damage identification model. Before the controlled and gradual introduction of actual structural damage, the bridge underwent nearly a year of continuous long-term monitoring. Even after the damage was introduced, the environmental monitoring system remained fully operational, thereby generating a dataset covering both undamaged (healthy) and damaged states of the bridge structure. Figure 1 The distribution of vibration sensors on the bridge deck and piers is shown, with numbers (e.g., 05, 06) indicating the sensor locations. The bridge vibration signals measured by these sensors can be used for structural modal identification and damage assessment to eliminate the influence of environmental factors on damage-sensitive parameters.

[0057] To obtain damage-sensitive parameters and determine whether bridge structural damage has occurred based on their changes, this invention employs the Cov-SSI algorithm, which can extract parameters such as modal frequencies from acceleration data. After obtaining preliminary modal identification results, modal screening and feature extraction steps are used to eliminate spurious modes in the modal identification results, reducing the impact of noise on the identification results and obtaining accurate modal identification results. Subsequently, a two-sample bilateral t-test is used to determine whether the test samples and healthy samples come from the same population, ultimately achieving accurate identification of bridge structural damage. The specific implementation process of modal parameter identification and structural damage identification functions is as follows: Figure 2 As shown.

[0058] The specific method is described below: 1. Modality recognition and pseudomodality removal First, the bridge vibration acceleration data is downsampled and filtered to construct a data matrix suitable for SSI analysis. Using the Cov-SSI algorithm, modal parameters are extracted by constructing a Hankel matrix and applying singular value decomposition. Subsequently, the three modal discrimination indices introduced in the modal screening section are applied to filter out pseudo-modes in the modal identification results.

[0059] 2. Feature Extraction Scan the modal frequencies of different system orders. Figure 3 Modal frequency stability plots for systems with orders ranging from 2 to 100, obtained using the Cov-SSI algorithm, were plotted, providing an intuitive foundation for obtaining accurate modal identification results. To further extract stable and physically meaningful modal parameters, the DBSCAN algorithm from this method was applied to cluster the modal frequencies and effectively eliminate noise. Then, the cluster centers were sorted by frequency, and the top six stable modal frequency values ​​were selected as the final results of modal identification. The results of identifying the top six stable modal frequencies using the DBSCAN clustering method are plotted on [the graph / plot]. Figure 4 In the diagram, each color represents a different modality. As can be seen, the identified clusters exhibit vertical grouping of frequency points across multiple model orders, indicating the consistency and stability of these modal frequencies. The most prominent clusters are located around 3.9 Hz, 5.0 Hz, 9.9 Hz, 10.5 Hz, 12.5 Hz, and 13.3 Hz, representing the first six natural frequencies of the structure. The dense and consistent distribution of these clusters across model orders further confirms the robustness of the DBSCAN-based identification method.

[0060] 3. Structural damage assessment To assess the impact of damage on modal behavior, the modal frequencies in damaged and undamaged states were compared, and their trends with temperature were observed. Figure 4As shown, at similar temperatures, damage consistently leads to a decrease in modal frequencies. Therefore, a two-sample bilateral t-test can be used to determine whether the test dataset differs statistically from the healthy baseline, thus completing the damage detection.

[0061] Since structural damage was first introduced in early August, vibration data collected in July were used as the baseline (i.e., the healthy sample). Two-sample bilateral t-tests were performed on the August and July datasets, as well as the June and July datasets. The results are shown in Table 1. A significant statistical difference exists between the August and July datasets, indicating a shift in the first-order modal frequencies due to structural damage. In contrast, there is no significant difference between the June and July datasets, confirming the consistency of the undamaged state.

[0062] By systematically analyzing vibration data from different monitoring stages of the Swiss Z24 bridge, this method effectively distinguishes between dynamic changes caused by environmental factors and modal shifts caused by actual structural damage, thus enabling accurate assessment of the structure's health status. Furthermore, this method can effectively capture significant changes in first-order modal frequencies during damage detection, providing a scientifically reliable technical foundation for bridge structural safety management and possessing broad practical application potential.

[0063] Table 1. Results of the two-sample bilateral t-test between healthy and test samples.

[0064] One embodiment of the present invention provides a bridge damage identification system based on random subspace and hypothesis testing, comprising: The data acquisition module acquires bridge structural response data; The modal parameter identification module uses a covariance-driven random subspace identification method to identify the modal parameters of the bridge structure response data. The modal parameter filtering module filters out the true modes based on the modal parameters of the bridge and generates a stability map; the final modal parameters are obtained by extracting features from the stability map. The damage identification module, based on the final modal parameters, employs a two-sample, two-sided approach. Inspection methods are used to identify whether a bridge is damaged.

[0065] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A bridge damage identification method based on random subspace and hypothesis testing, characterized in that, Includes the following steps: Obtain bridge structural response data; Modal identification was performed on the bridge structural response data using a covariance-driven random subspace identification method to obtain the bridge's modal parameters. Based on the modal parameters of the bridge, the true modes are selected and a stability graph is generated; the final modal parameters are obtained by feature extraction from the stability graph. Based on the final modal parameters, a two-sample bilateral method is used. Inspection methods are used to identify whether a bridge is damaged.

2. The bridge damage identification method based on random subspace and hypothesis testing according to claim 1, characterized in that, Modal identification was performed on the bridge structural response data using a covariance-driven random subspace identification method, specifically as follows: The Hankel matrix is ​​constructed using bridge structural response data; the row space of the Hankel matrix is ​​divided into past and future parts. The covariance of the past and future parts of Hankel's row space is calculated separately to obtain the Toeplitz matrix. and ; Based on the Toeplitz matrix and Calculate the state transition matrix; The state transition matrix is ​​orthogonally similar to diagonalize, and the modal parameters of the bridge are calculated.

3. The bridge damage identification method based on random subspace and hypothesis testing according to claim 1, characterized in that, The Hankel matrix is ​​constructed using bridge structural response data; the row space of the Hankel matrix is ​​divided into past and future parts, specifically: in, The Hankel matrix; the Hankel matrix includes Data collected from each channel Displacement data, with a time delay of [number] data points. , ; The past part is The future part is: or ;in, , including from arrive The line, but not including row; include to The line, excluding OK.

4. The bridge damage identification method based on random subspace and hypothesis testing according to claim 3, characterized in that, The and The calculation methods are as follows: in, This indicates transpose.

5. The bridge damage identification method based on random subspace and hypothesis testing according to claim 4, characterized in that, The state transition matrix is: in, ; ; This is the state transition matrix; This is the output matrix; This is the input influence matrix.

6. The bridge damage identification method based on random subspace and hypothesis testing according to claim 1, characterized in that, The modal parameters include natural frequency, damping ratio, and mode shape.

7. The bridge damage identification method based on random subspace and hypothesis testing according to claim 1, characterized in that, The selection of true modes is based on modal confidence, damping ratio check, and complex conjugate pole check as criteria.

8. The bridge damage identification method based on random subspace and hypothesis testing according to claim 1, characterized in that, Modal frequency features were extracted by using the DBSCAN clustering method to extract features from the stability graph.

9. The bridge damage identification method based on random subspace and hypothesis testing according to claim 1, characterized in that, The method of using a two-sample bilateral t-test to identify whether the bridge is damaged is as follows: Collect reference samples of the bridge's health status to obtain health samples; Assuming the final modal parameters and the healthy samples come from the same population, the two-sample bilateral t-test method is used to calculate the t-value and the corresponding p-value. Assuming the final modal parameters are not from the same population as the healthy samples, a two-sample bilateral t-test is used to calculate the t-value and the corresponding p-value. Based on different assumptions and the calculated t-values ​​and corresponding p-values, it is determined whether the bridge is damaged.

10. A bridge damage identification system based on random subspace and hypothesis testing, characterized in that, include: The data acquisition module acquires bridge structural response data; The modal parameter identification module uses a covariance-driven random subspace identification method to identify the modal parameters of the bridge structure response data. The modal parameter filtering module filters out the true modes based on the modal parameters of the bridge and generates a stability map; the final modal parameters are obtained by extracting features from the stability map. The damage identification module, based on the final modal parameters, employs a two-sample, two-sided approach. Inspection methods are used to identify whether a bridge is damaged.

Images