Bridge structure anomaly identification method based on data fusion

Abstract

This invention discloses a bridge structural anomaly identification method based on data fusion, belonging to the field of bridge health monitoring technology. The method includes constructing an initial finite element model containing geometric topology and mechanical parameters based on bridge design drawings and geological survey reports; deploying multi-source sensor arrays at the mid-span, supports, and pier roots of the bridge to simultaneously collect dynamic strain time series, vibration acceleration spectrum, and environmental temperature and humidity data. Wavelet packet decomposition is used to extract energy entropy features from the dynamic strain data, which are then fused with the modal frequency offset features of the vibration acceleration to generate a comprehensive health state vector. Spurious strain signals are eliminated through physical correction using environmental temperature and humidity data. The corrected feature set is input into the initial finite element model to invert and solve for the stiffness degradation matrix, which is then compared with the standard stiffness matrix to identify stiffness abrupt change units, generating an anomaly location coordinate set. This invention improves the accuracy of anomaly identification through multi-feature fusion and environmental physical correction, achieving precise location of bridge structural anomalies.

Description

Technical Field

[0001] This invention belongs to the field of bridge health monitoring technology, specifically a method for identifying bridge structural anomalies based on data fusion. Background Technology

[0002] Traditional bridge structural anomaly identification often relies on single-type sensor data analysis, typically selecting a single data source from dynamic strain time-series data or vibration acceleration spectrum data to construct a state characterization system. Simultaneously, it uses an initial finite element model without incorporating real-time operational data correction to complete structural mechanical performance analysis. The acquisition and application of multi-source sensor data remain at the level of independent analysis, failing to achieve collaborative processing of data from different dimensions. Temperature stress caused by changes in environmental temperature and humidity can generate spurious strain signals during structural response acquisition. Existing technologies do not implement targeted physical correction processing for such interference signals. Single sensor features cannot comprehensively map the health status of the bridge structure, and the stiffness inversion calculation of the finite element model does not eliminate the interference of environmental factors. Therefore, structural anomaly identification results are prone to misjudgment, and the accuracy of anomaly location coordinates cannot be guaranteed.

[0003] Existing technologies cannot achieve feature-level fusion of wavelet packet decomposition energy entropy features from dynamic strain time-series data and modal frequency offsets from vibration acceleration spectrum data, making it difficult to generate a comprehensive health state vector that fully reflects the structural state. Furthermore, they cannot utilize environmental temperature and humidity variation data to physically correct structural response characteristics, cannot eliminate spurious strain signals caused by temperature stress, and the corrected structural response feature set cannot be input into the initial finite element model to complete the inversion calculation of the stiffness degradation matrix. They also cannot accurately identify structural elements with abrupt stiffness changes through comparison with the standard stiffness matrix. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art;

[0005] Therefore, this invention proposes a bridge structural anomaly identification method based on data fusion, including:

[0006] Based on the existing design drawings and geological survey reports of the bridge structure, an initial finite element model of the bridge structure is constructed. The initial finite element model includes the geometric and topological relationships and material mechanical parameters of the beams, piers and supports.

[0007] Multi-source sensor arrays are deployed at the mid-span, supports, and pier roots of the bridge structure to simultaneously collect dynamic strain time-series data, vibration acceleration spectrum data, and environmental temperature and humidity change data of the bridge structure during operation.

[0008] Wavelet packet decomposition is performed on the dynamic strain time series data to extract energy entropy features in different frequency bands. The energy entropy features are then fused with the modal frequency offsets extracted from the vibration acceleration spectrum data to generate a comprehensive health state vector of the bridge structure.

[0009] The comprehensive health state vector is physically corrected using the environmental temperature and humidity change data to eliminate spurious strain signals caused by temperature stress and obtain the corrected structural response feature set.

[0010] The modified structural response feature set is input into the initial finite element model. The stiffness degradation matrix of the bridge structure in the current state is obtained by inversion calculation and compared with the standard stiffness matrix of the initial finite element model to identify structural elements with abrupt stiffness changes and generate a preliminary abnormal positioning coordinate set.

[0011] Furthermore, based on existing design drawings and geological survey reports of the bridge structure, an initial finite element model of the bridge structure is constructed, including:

[0012] The existing design drawings of the bridge structure are analyzed to extract the structural geometric dimensions, spatial topological connections and material designations of the beams, piers and supports. Based on the material designations, a material mechanics handbook is consulted to assign material mechanics parameters such as elastic modulus, Poisson's ratio and density to each material.

[0013] Import the geological survey report, establish a distribution model of different soil layers under the bridge pier foundation based on borehole sampling data, and calculate the equivalent spring stiffness coefficient of each soil layer in combination with soil mechanics parameters to simulate the vertical and horizontal constraints of the soil layers on the bridge pier.

[0014] The beam body and piers are discretized by using the beam grid method or spatial beam elements, the supports are modeled by connecting elements, and the soil constraints are modeled by a series of spring elements with corresponding equivalent spring stiffness coefficients, which are combined to form a complete structural finite element model.

[0015] Static analysis was performed on the established complete structural finite element model under its own weight, and the initial deformation results obtained from the solution were compared and verified with the precamber in the design drawings.

[0016] If the error between the calculated initial deformation result and the pre-camber is less than the preset tolerance threshold, the modeling is confirmed to be correct, and this complete structural finite element model and its material mechanical parameters and standard stiffness matrix are used as the initial finite element model; otherwise, the modeling parameters are iteratively adjusted until the error meets the standard.

[0017] Furthermore, wavelet packet decomposition is performed on the dynamic strain time series data to extract energy entropy features in different frequency bands. These energy entropy features are then fused with the modal frequency offsets extracted from the vibration acceleration spectrum data at the feature level to generate a comprehensive health state vector for the bridge structure, including:

[0018] A wavelet basis function with tight support characteristics is selected, and a four-level wavelet packet decomposition is performed on the dynamic strain time series data to decompose the signal into sixteen independent frequency band subspaces;

[0019] The probability distribution of signal energy in each frequency band subspace is calculated, and the energy entropy value of each frequency band is calculated based on the Shannon entropy definition to form a frequency band energy entropy sequence;

[0020] The peak picking algorithm is applied to the vibration acceleration spectrum data to extract the measured values ​​of the first six natural modal frequencies, and the relative percentage of offset between the measured values ​​and the theoretical values ​​obtained from the simulation of the initial finite element model is calculated to form a modal frequency offset sequence.

[0021] The frequency band energy entropy sequence and the modal frequency offset sequence are aligned according to the time sequence and mapped to a unified feature space through normalization processing;

[0022] A weighting factor calculated from the environmental temperature and humidity change data is applied to each data point in the mapped feature space, and the weighted sum is used to obtain the comprehensive health status vector of the bridge structure.

[0023] Furthermore, the comprehensive health state vector is physically corrected using the environmental temperature and humidity change data to eliminate spurious strain signals caused by temperature stress, including:

[0024] The ambient temperature curve and relative humidity curve are separated from the ambient temperature and humidity change data, and the daily periodic variation component of the ambient temperature curve is identified.

[0025] Based on the concrete thermal expansion coefficient and elastic modulus of the bridge structure, a lookup table of the linear correspondence between ambient temperature and structural thermal strain is established.

[0026] Using the linear correspondence lookup table, the daily periodic variation components of the ambient temperature curve are converted into an equivalent thermal strain time series.

[0027] Subtracting the equivalent thermal strain time series from the dynamic strain time series data yields residual strain time series data free from temperature interference.

[0028] The residual strain time series data is resubmitted into the calculation process of the frequency band energy entropy, and the comprehensive health state vector is updated to obtain the corrected structural response feature set.

[0029] Further, the modified structural response feature set is input into the initial finite element model, and the stiffness degradation matrix of the bridge structure in the current state is obtained through inversion calculation, including:

[0030] A diagonal perturbation factor is introduced into the global stiffness matrix of the initial finite element model to construct the sensitivity analysis equation to be solved.

[0031] Using the measured modal frequencies in the modified structural response feature set as the target values ​​and the diagonal perturbation factor as the independent variable, an objective function is constructed.

[0032] The objective function is iteratively optimized using a sequential quadratic programming algorithm to find the optimal combination of diagonal perturbation factors that makes the simulated modal frequencies approximate the measured modal frequencies.

[0033] The obtained optimal diagonal perturbation factor combination is used to replace the elements in the global stiffness matrix of the initial finite element model according to the corresponding degree of freedom position, and the stiffness degradation matrix of the bridge structure in the current state is reconstructed.

[0034] Furthermore, by comparing the stiffness matrix with the standard stiffness matrix of the initial finite element model, structural elements with abrupt stiffness changes are identified, and a preliminary set of anomaly location coordinates is generated, including:

[0035] The stiffness variation matrix is ​​obtained by calculating the matrix difference between the stiffness degradation matrix and the standard stiffness matrix of the initial finite element model.

[0036] Iterate through each non-zero element in the stiffness variation matrix and calculate the absolute amount and relative percentage of its stiffness loss;

[0037] Set a threshold for the relative percentage of stiffness loss, and extract the degree of freedom numbers corresponding to stiffness loss values ​​that exceed the threshold;

[0038] Based on the degree of freedom number, the geometric topological relationship of the initial finite element model is traced back to the specific structural component number and its coordinate position in three-dimensional space.

[0039] All the parsed coordinate positions are clustered and deduplicated to form the preliminary abnormal location coordinate set.

[0040] Furthermore, it also includes a visually assisted verification step based on crack image data:

[0041] High-resolution crack images captured by an industrial camera array deployed at the bottom of the bridge are retrieved, and the high-resolution crack images are subjected to image enhancement and edge sharpening processing.

[0042] The Hough transform algorithm is used to detect the straight crack profile in the processed high-definition crack image, and the length and orientation angle of the straight crack profile are measured.

[0043] The length and orientation angle of the linear crack profile are spatially registered with the preliminary anomaly location coordinate set to determine whether there are overlapping or adjacent crack features.

[0044] If there are overlapping or adjacent crack features, the timestamp of the high-definition crack image is read and associated with the timestamp of the dynamic strain time series data.

[0045] Based on the correlation results, the dynamic strain time series data segments corresponding to the timestamp before and after the shooting are extracted, and it is analyzed whether the strain mutation rate in the data segments exceeds the preset safety limit, and an anomaly confirmation list with visual evidence markers is generated.

[0046] Further, spatial registration is performed between the length and orientation angle of the linear crack profile and the preliminary anomaly location coordinate set, including:

[0047] Metadata information containing camera intrinsic and extrinsic parameters is extracted from the high-resolution crack image, and a transformation matrix from the image pixel coordinate system to the bridge structure world coordinate system is constructed based on the metadata information;

[0048] Using the transformation matrix, the length and orientation angle of the straight crack profile are projected onto the world coordinate system of the bridge structure to obtain the spatial straight line equation of the crack in the world coordinate system.

[0049] Traverse each coordinate point in the preliminary anomaly location coordinate set and calculate the vertical distance from the coordinate point to the spatial straight line equation;

[0050] If the vertical distance is less than a preset spatial tolerance threshold, it is determined that there is a visually visible crack in the area where the coordinate point is located, and the confidence level of the coordinate point in the preliminary abnormal location coordinate set is increased by one level.

[0051] Furthermore, it also includes a time-series correlation analysis step based on traffic load data:

[0052] Access the real-time vehicle data stream from bridge toll stations or weigh stations, and extract the axle load, speed, and specific timestamp of each vehicle passing through the bridge.

[0053] Calculate the equivalent impact load spectrum of each vehicle on the bridge structure when it passes based on the axle load and vehicle speed;

[0054] The equivalent impact load spectrum is convolved with the dynamic strain time series data to simulate the theoretical strain response curve of the bridge under traffic load.

[0055] The cross-correlation coefficient between the theoretical strain response curve and the actually acquired dynamic strain time series data is calculated to obtain the waveform similarity index between the theoretical strain response curve and the dynamic strain time series data.

[0056] If the waveform similarity index is lower than the set lower threshold, it is determined that there is nonlinear response caused by damage inside the bridge structure, and the time period corresponding to the waveform similarity index being lower than the threshold is marked as a suspicious time period, which is fed back into the process of solving the stiffness degradation matrix as a constraint condition.

[0057] Furthermore, the equivalent impact load spectrum is convolved with the dynamic strain time series data to simulate the theoretical strain response curve of the bridge under traffic load, including:

[0058] The equivalent impact load spectrum of each vehicle is defined as a pulse excitation signal and applied to the corresponding lane position node of the initial finite element model.

[0059] The transient dynamic response of the initial finite element model under the pulse excitation signal is solved using the modal superposition method.

[0060] Extract the nodal displacement time history curves corresponding to the sensor positions at the mid-span of the bridge from the transient dynamic response;

[0061] The nodal displacement time history curves are converted into strain time history curves and spliced ​​together according to the chronological order of vehicle passage.

[0062] The complete time series obtained after splicing is the theoretical strain response curve of the bridge under traffic load.

[0063] Compared with the prior art, the beneficial effects of the present invention are:

[0064] Dynamic strain time-series data are decomposed into wavelet packet decomposition to extract energy entropy features in different frequency bands. These energy entropy features are then fused with modal frequency offsets extracted from vibration acceleration spectrum data at the feature level. The resulting comprehensive health state vector of the bridge structure can fully carry the core representation information of multi-source sensor data. The fusion processing of different types of sensor features can cover the local details of structural response changes and overall modal evolution information. The representation blind spots of single sensor data are effectively filled, and the integrity of structural state representation is enhanced. During feature fusion, the error components of different sensor data counterbalance each other, and the stability of structural state representation results is optimized. The collaborative representation of multi-dimensional features can more comprehensively map the mechanical response changes of each component of the bridge structure.

[0065] Environmental temperature and humidity change data are applied to the physical correction processing of the comprehensive health state vector. The pseudo-strain signal induced by temperature stress is separated and eliminated from the structural response features. The structural response feature set after the correction operation only corresponds to the real changes in the mechanical properties of the structure itself. The inversion calculation carried out after inputting this feature set into the initial finite element model can obtain the stiffness degradation matrix that fits the actual operating state of the structure. The comparison processing between the stiffness degradation matrix and the standard stiffness matrix of the initial finite element model can accurately identify structural units with abrupt changes in stiffness parameters. The directionality of structural anomaly location is enhanced, the identification deviation caused by environmental interference factors is effectively eliminated, the accuracy of the structural anomaly location coordinate set is optimized, the reliability of the structural anomaly identification results is improved, the identification misjudgment caused by pseudo-anomaly signals is continuously reduced, and the accuracy and effectiveness of structural anomaly identification are optimized simultaneously. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating the steps of the bridge structure anomaly identification method based on data fusion described in this invention.

[0067] Figure 2 A flowchart for generating a comprehensive health status vector;

[0068] Figure 3 The diagram shows the iterative optimization process of bridge modal frequencies.

[0069] Figure 4 The time-series analysis diagram of strain and strain rate of bridge structure is shown.

[0070] Figure 5 This is a time-series correlation diagram of traffic load and bridge strain response. Detailed Implementation

[0071] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] See Figure 1Based on existing design drawings and geological survey reports of the bridge structure, an initial finite element model of the bridge structure was constructed. This model accurately includes the geometric topological relationships and material mechanical parameters of the beams, piers, and supports. Multi-source sensor arrays were deployed at key locations such as the mid-span, supports, and pier roots of the bridge structure to simultaneously collect dynamic strain time-series data, vibration acceleration spectrum data, and environmental temperature and humidity variation data during operation. Wavelet packet decomposition was performed on the collected dynamic strain time-series data to extract energy entropy features at different frequency bands. Simultaneously, modal frequency offsets were extracted from the vibration acceleration spectrum data. The energy entropy features and modal frequency offsets were fused at the feature level to generate a comprehensive health state vector of the bridge structure. The comprehensive health state vector was physically corrected using environmental temperature and humidity variation data to eliminate spurious strain signals caused by temperature stress, resulting in a corrected structural response feature set. The modified structural response feature set is input into the initial finite element model. The stiffness degradation matrix of the bridge structure under the current state is obtained by inversion calculation. The matrix is ​​then compared with the standard stiffness matrix of the initial finite element model to identify structural elements with abrupt stiffness changes and generate a preliminary abnormal location coordinate set.

[0073] In one embodiment of the present invention, existing design drawings of bridge structures are typically provided in CAD format, containing geometric dimensional information such as beam cross-sectional dimensions, pier height and diameter, and bearing type and spacing. The parsing process extracts the span, beam height, flange width, pier center coordinates, abutment dimensions, and precise installation positions of the bearings on the cap beam by reading the layers and block attributes in the drawings, thereby constructing a complete spatial topological connection relationship. The bill of materials in the drawings indicates material specifications such as C50 concrete for the main beams, C40 concrete for the piers, and Q345 steel for the bearings.

[0074] In some embodiments, the geological survey report provides a columnar section of multiple boreholes at the bridge site. Borehole sampling data reveals that the underground soil layers, from top to bottom, consist of fill, silty clay, and strongly weathered rock. Based on the thickness of each soil layer, the standard penetration test blow count, and the compression modulus and cohesion measured by laboratory geotechnical tests, the equivalent spring stiffness coefficient of each soil layer can be calculated. For the silty clay layer, the vertical equivalent spring stiffness coefficient can be calculated using the layered summation method and the subgrade coefficient method, while the horizontal equivalent spring stiffness coefficient can be calculated using the m-method. The calculated equivalent spring stiffness coefficients of each soil layer will be used to simulate the actual constraint effect of the soil layers on the bridge pier foundation.

[0075] In practice, the main girder and piers of the bridge are discretized using spatial beam elements. The main girder is divided into multiple beam elements along the longitudinal direction, and the cross-sectional properties of each beam element are defined according to the cross-sectional dimensions in the design drawings. The piers are also divided into multiple beam elements along the height direction. The supports are modeled using connection elements, and the properties of the connection elements are set according to the vertical stiffness, horizontal stiffness, and rotational stiffness parameters in the support product manual. For soil constraints, spring elements with corresponding equivalent spring stiffness coefficients are set at the bottom and side nodes of the pier foundation, based on the soil layer distribution model. The vertical spring elements simulate the vertical support of the soil, and the horizontal spring elements simulate the lateral constraint of the soil. All the discretized beam elements, connection elements, and spring elements are combined to form a complete structural finite element model.

[0076] Optionally, a static analysis is performed by applying a self-weight load to the established complete structural finite element model to obtain the initial deformation results of the structure under self-weight. This result is usually represented by the vertical deflection curve of the main beam. The description section of the design drawings provides the pre-camber value of the main beam in the completed bridge state. The calculated mid-span deflection value of the main beam is compared with the design pre-camber value. The preset tolerance threshold can be set according to the design specifications, for example, 5% of the pre-camber value. If the error between the calculated deflection and the pre-camber is less than 5%, the finite element modeling is confirmed to be correct. This model, along with all its material mechanical parameters and the overall structural standard stiffness matrix calculated under a non-damaging state, is used as the initial finite element model for subsequent analyses. If the error exceeds 5%, it is necessary to check whether the input geometric dimensions are accurate, whether the material parameters are correct, and whether the spring stiffness coefficients of the boundary conditions are reasonable, and to perform iterative adjustments until the error meets the requirements.

[0077] It is understandable that the accuracy of the initial finite element model is the foundation of the entire method, and its standard stiffness matrix is:

[0078]

[0079] in: Represents the standard stiffness matrix. This indicates that for all in the model Summing each unit Indicates the unit volume Accumulate points. This represents the strain-displacement matrix of the element. This represents the elasticity matrix composed of the material's elastic modulus and Poisson's ratio. This formula describes the process of integrating element stiffness into overall stiffness; the standard stiffness matrix... It serves as a benchmark for subsequent identification of stiffness degradation.

[0080] In one embodiment of the present invention, see [reference] Figure 2For a segment of dynamic strain time-series data acquired by sensors at mid-span of a prestressed concrete T-beam bridge, the db4 wavelet basis function with tight support characteristics was selected. Four-level wavelet packet decomposition was performed on the dynamic strain time-series data with a sampling frequency of 100Hz, decomposing the original signal into sixteen independent frequency band subspaces. These sixteen subspaces correspond to frequency ranges of 0-3.125Hz, 3.125-6.25Hz, 6.25-9.375Hz, ..., 46.875-50Hz, respectively. The energy of the reconstructed signal within each frequency band subspace was calculated, along with its probability distribution. Based on the Shannon entropy definition, the energy entropy value of each frequency band was calculated, forming a frequency band energy entropy sequence consisting of sixteen components. A peak-picking algorithm was applied to the vibration acceleration spectrum data collected within the same time period to identify the first six significant peaks in the spectrum. The corresponding frequency values ​​were extracted as the measured values ​​of the first six natural modal frequencies, which are 2.1Hz, 5.8Hz, 9.3Hz, 12.7Hz, 16.0Hz, and 19.2Hz, respectively. The relative percentage offset between these six measured frequency values ​​and the theoretical frequency values ​​of 3.0Hz, 6.5Hz, 10.1Hz, 13.5Hz, 16.8Hz, and 20.5Hz obtained from the modal analysis simulation of the initial finite element model was calculated to form a modal frequency offset sequence [-30.0%, -10.8%, -7.9%, -5.9%, -4.8%, -6.3%].

[0081] It is understandable that the frequency band energy entropy sequence and the modal frequency shift sequence are aligned with the same time series, both sequences corresponding to the monitoring data of the same hour, with a time resolution of one data point per minute, for a total of 60 time points. Using the max-min normalization method, the sixteen-dimensional frequency band energy entropy sequence and the six-dimensional modal frequency shift sequence are mapped to the [0,1] interval respectively, achieving mapping to a unified feature space. For each data point in the mapped feature space, a weighting factor derived from the environmental temperature and humidity change data is calculated. The weighting factor is calculated based on the temperature reduction effect on structural stiffness, and a weighted sum is obtained to obtain a comprehensive scalar value. This comprehensive scalar value sequence constitutes the comprehensive health state vector of the bridge structure.

[0082] In some embodiments, the ambient temperature curve and relative humidity curve are separated from the ambient temperature and humidity change data. The ambient temperature curve exhibits a clear diurnal periodic variation component, gradually rising from 15 degrees Celsius at night to 30 degrees Celsius during the day, and then falling back. The diurnal periodic variation component of the ambient temperature curve is identified by extracting the trend term through moving average filtering, and then subtracting the trend term from the original data to obtain the periodic term. Based on the thermal expansion coefficient and elastic modulus of the bridge structure's concrete, a lookup table of the linear correspondence between ambient temperature and structural thermal strain is established. The lookup table is established based on the formula:

[0083]

[0084] in: Indicates thermal strain, This represents the coefficient of thermal expansion of concrete. This represents the temperature change. A linear correspondence lookup table is used to identify the diurnal periodic components of the ambient temperature curve, i.e., the components corresponding to each temperature sampling point. This is converted into an equivalent thermal strain time series.

[0085] In practice, the equivalent thermal strain time series is subtracted from the original dynamic strain time series data point by point. Assuming the original dynamic strain value at a certain sampling point is 150 microstrain, the corresponding equivalent thermal strain value is -20 microstrain (cooling). The subtraction yields residual strain time series data of 170 microstrain. This residual strain time series data is then re-substituted into the calculation process of the frequency band energy entropy, performing the same four-layer wavelet packet decomposition, energy probability calculation, and Shannon entropy calculation to update the sixteen-dimensional frequency band energy entropy sequence. The updated frequency band energy entropy sequence replaces the original calculated sequence, and is then normalized and weighted again with the modal frequency shift sequence to obtain the corrected structural response feature set. This corrected structural response feature set is represented as a comprehensive health state vector sequence that eliminates the influence of daily temperature cycle fluctuations.

[0086] In one embodiment of the present invention, for a prestressed concrete T-beam bridge with suspected local damage, a diagonal perturbation factor is introduced into the global stiffness matrix of the initial finite element model. Specifically, a diagonal matrix with the same dimensions as the global stiffness matrix of the initial finite element model is constructed, and its diagonal elements are the perturbation factors to be solved. The constructed sensitivity analysis equation is as follows:

[0087]

[0088] in: It is the standard stiffness matrix of the initial finite element model. It is composed of diagonal perturbation factor The diagonal perturbation matrix formed It is the mode shape vector. It is an eigenvalue. It is the mass matrix. The target values ​​are the first four modal frequencies obtained through vibration testing from the corrected structural response characteristic set. These target values ​​are specific numerical values, including the first vertical frequency of 5.2Hz, the second vertical frequency of 12.8Hz, etc., and the diagonal perturbation factor. Construct the objective function with the independent variable as the objective variable. ,in The simulated value of the m-th modal frequency is obtained by calculating the stiffness matrix after perturbation. It is the measured m-th modal frequency.

[0089] In some embodiments, a sequential quadratic programming algorithm is used to construct the objective function. Iterative optimization is performed to find the optimal combination of diagonal perturbation factors that makes the simulated modal frequencies approximate the measured modal frequencies. The iterative process of the sequential quadratic programming algorithm includes constructing a quadratic approximation of the objective function and a linear approximation of the constraints in each iteration, solving a quadratic programming subproblem to obtain the search direction, determining the step size through line search, and updating the diagonal perturbation factors. The value of until the objective function The value is less than the preset tolerance or the maximum number of iterations is reached. The optimal combination of diagonal perturbation factors obtained after iterative optimization is then used to form a specific set of... The numerical values ​​replace the global stiffness matrix of the initial finite element model with the corresponding degrees of freedom. The corresponding elements on the diagonal are replaced using the following formula: ,in It is the element in the i-th row and i-th column of the reconstructed stiffness matrix. The element in the i-th row and i-th column of the initial stiffness matrix is ​​used to reconstruct the stiffness degradation matrix of the bridge structure in the current state. .

[0090] Calculate the stiffness degradation matrix With the standard stiffness matrix of the initial finite element model The matrix difference is obtained by subtracting corresponding elements to obtain the stiffness variation matrix. ,Right now Traversing the stiffness variation matrix For each non-zero element in the equation, which mainly appears on the diagonal and in its vicinity, calculate the absolute amount of stiffness loss represented by each non-zero element, i.e. The value of is calculated, and its position relative to the corresponding element of the standard stiffness matrix is ​​determined. The relative percentage, the formula for calculating the relative percentage is: Set a threshold for the relative percentage of stiffness loss. The specific value of the threshold is set based on engineering experience and structural safety specifications, for example, 15%. Extract the degree of freedom numbers corresponding to stiffness loss values ​​exceeding the 15% threshold. The extracted degree of freedom numbers are a set of integer indices.

[0091] It is understandable that the extracted degrees of freedom (DOF) numbers are traced back to the geometric topology of the initial finite element model. The geometric topology of the initial finite element model records the structural element number and node number to which each DDF number belongs. The specific structural component number and its coordinate position in three-dimensional space are then analyzed. For example, the DDF with stiffness loss exceeding the threshold corresponds to the translational DDF in the Y direction of node 2 of beam element 5. Beam element 5 is the second T-beam of the third span, and the three-dimensional coordinates of its node 2 are (X=35.2m, Y=0m, Z=8.5m). All the coordinate positions obtained through analysis, such as multiple different DDFs potentially corresponding to different nodes of the same beam, are clustered for deduplication. The clustering deduplication uses a Euclidean distance-based clustering algorithm, grouping coordinate points with a spatial distance of less than 2 meters into one class and selecting the center point of each class as the representative coordinate, forming a preliminary anomaly location coordinate set. This coordinate set contains multiple three-dimensional coordinates such as (35.2, 0, 8.5).

[0092] Optionally, the threshold for the relative percentage of stiffness loss is not a single fixed value. Different thresholds can be used for different parts of the bridge structure. For example, a 10% threshold could be set for the mid-span region of the main girder, and a 20% threshold for the support region, while traversing the stiffness variation matrix. At that time, the structural part to which it belongs is determined according to the degree of freedom number, and the corresponding threshold is applied for filtering.

[0093] See Figure 3 This diagram illustrates the iterative optimization process of bridge modal frequencies, showcasing the optimization of bridge structural modal frequencies. It primarily reflects the convergence effect of the sequential quadratic programming algorithm on modal frequency optimization during stiffness matrix inversion in bridge structural anomaly identification. Simulated values ​​of each order gradually approach the corresponding measured values ​​with increasing iteration count, demonstrating the algorithm's convergence. The iteration count ranges from 0 to 20, and the error between simulated and measured values ​​continuously decreases with each iteration. This process is the core step in inverting the stiffness degradation matrix of the bridge structure. Iterative optimization brings the simulated modal frequencies closer to the measured values, providing a data foundation for subsequent calculations of the stiffness degradation matrix and locating structural anomalies. The fourth-order torsional frequency exhibits the largest iterative convergence amplitude, reflecting the higher sensitivity of this order of mode to changes in structural stiffness.

[0094] In one embodiment of the invention, high-resolution crack images captured by an industrial camera array deployed at the bottom of the bridge are retrieved. The industrial camera array includes multiple high-resolution cameras fixedly mounted on the bottom plate of the box girder and the surface of the piers. Image enhancement and edge sharpening are performed on the high-resolution crack images. Image enhancement uses histogram equalization, and edge sharpening uses Laplacian convolution. The Hough transform algorithm is used to detect straight crack contours in the processed high-resolution crack images. The Hough transform algorithm identifies multiple pixel segments from the processed image and measures the length and orientation angle of each straight crack contour. The orientation angle is defined as the angle between the pixel segment and the horizontal axis of the image. Metadata information containing camera intrinsic and extrinsic parameters is extracted from the high-resolution crack images. The metadata information includes the camera focal length, principal point coordinates, camera position in the world coordinate system, and attitude angle. Based on the metadata information of the camera intrinsic and extrinsic parameters, a transformation matrix is ​​constructed from the image pixel coordinate system to the bridge structure's world coordinate system. The transformation matrix is ​​calculated using camera calibration parameters and installation position parameters. Using the transformation matrix from the image pixel coordinate system to the bridge structure's world coordinate system, the pixel start and end coordinates of the linear crack profile, along with the calculated azimuth angle, are projected onto the bridge structure's world coordinate system. The projection calculation is completed by solving the collinearity equation, yielding the spatial linear equation of the crack in the world coordinate system. This spatial linear equation can be expressed in parametric form. ,in It is a vector of point coordinates on a straight line. It is the world coordinate of one end of the crack. It is a parameter. It is the direction vector of the crack direction.

[0095] In some embodiments, each coordinate point in the preliminary anomaly location coordinate set is traversed. The preliminary anomaly location coordinate set comes from the inversion result of the embodiment and may include coordinates such as (35.2,0,8.5), (35.5,0,8.3), (18.7,5.1,6.0), etc. The perpendicular distance from each coordinate point to the spatial straight line equation is calculated. The calculation formula is:

[0096]

[0097] in: These are the preliminary coordinates of the anomaly location. These are the world coordinates of the crack endpoints. It is the crack direction vector. Represents the cross product of vectors. This represents the magnitude of the vector. If the vertical distance is less than a preset spatial tolerance threshold, which is set based on camera measurement accuracy and structural dimensions (e.g., 0.5 meters), then the area where the coordinate point is located is determined to have a visually visible crack, and the confidence level of the coordinate point in the preliminary anomaly location coordinate set is increased by one level. The initial confidence level is 1, and it becomes 2 after being increased by one level. The length and orientation angle of the straight crack profile are spatially registered with the preliminary anomaly location coordinate set to determine whether there are overlapping or adjacent crack features. The results of calculating the vertical distance can be compiled into a comparison table to record the matching status. See Table 1.

[0098] Table 1: Preliminary Anomaly Point Matching Calculation Table with Visual Cracks

[0099] Preliminary anomaly coordinates Crack Space Linear Equation Calculate the vertical distance (m) Is it less than the 0.5m threshold? Post-processing confidence (35.2,0,8.5) L1:(35.0,0,8.4)+t(0.1,0,0.05) 0.22 yes 2 (35.5,0,8.3) L1:(35.0,0,8.4)+t(0.1,0,0.05) 0.35 yes 2 (18.7,5.1,6.0) L2:(20.1,5.0,6.0)+t(-0.05,0.02,0) 1.40 no 1

[0100] It is understandable that if overlapping or adjacent crack features exist, such as the two points (35.2,0,8.5) and (35.5,0,8.3) in Table 1, the timestamp of the high-resolution crack image containing the matching crack features is read. The timestamp records the precise time of image acquisition, for example, "2026-06-15 10:25:30.123". The timestamp is then correlated with the timestamp of the dynamic strain time-series data. The dynamic strain time-series data is continuously acquired, and each data point has a corresponding timestamp. Based on the correlation result, dynamic strain time-series data segments corresponding to a time window before and after the timestamp are extracted, for example, 5 minutes before and after. The strain mutation rate within the dynamic strain time-series data segment is analyzed to see if it exceeds a preset safety limit. The strain mutation rate is obtained by calculating the maximum absolute value of the strain change rate over time within the data segment. The preset safety limit is set based on material properties, for example, 5 microstrain / second. If the calculated strain mutation rate exceeds 5 microstrains / second, an anomaly confirmation list with visual evidence markers is generated. The list records the coordinates of the anomaly location, the matching crack information, the shooting time, the strain mutation rate, and the judgment result.

[0101] Optionally, for multiple adjacent cracks detected in the same area, the Hough transform algorithm may identify multiple spatial line equations. In this case, the vertical distance from the initial anomaly coordinates to each crack's spatial line equation is calculated. If any crack exists such that the vertical distance is less than the spatial tolerance threshold, it is determined that the area where the coordinates are located has visually visible crack features, and the confidence level is increased. The spatial registration process is executed automatically, with the calculation program traversing all anomaly coordinates and all identified crack features to complete distance calculations and threshold judgments in batches. When generating the anomaly confirmation list, the list summarizes all successfully matched records and integrates the initial anomaly location from data fusion inversion with crack evidence from visual detection to form a comprehensive report containing location, time, sensor data, and image evidence.

[0102] See Figure 4 This is a time-series analysis chart of strain and strain rate of a bridge structure, used to verify the changes in structural response before and after the discovery of cracks. It belongs to the visual-assisted verification link of bridge anomaly identification. The strain data from 0 to 5.5 minutes fluctuated in the range of 35 to 65 microstrains, showing an overall trend of slow decline followed by a rebound. The strain rate oscillated violently between -100 and 100, reflecting the normal structural response caused by traffic loads and environmental factors. There were no obvious abnormal abrupt changes, and the structure was in a normal working state. A significant peak appeared in the strain data from 5.5 to 10 minutes, reaching a maximum of nearly 70 microstrains, far exceeding the previous level, indicating that the local stiffness of the structure decreased and deformation intensified after the crack appeared. The strain rate simultaneously showed a violent positive peak, exceeding 100 microstrains / minute in some periods, indicating that crack propagation led to a sharp increase in the structural deformation rate. The strain and strain rate dropped rapidly from 8 to 10 minutes, possibly indicating that the structure entered a brief period of stability or that the detected local damage had not yet further expanded.

[0103] In one embodiment of the present invention, real-time vehicle data streams from bridge toll stations or weigh stations are accessed. These streams contain records of each vehicle passing through the bridge, with each record including license plate number, passage timestamp, axle load, total weight, speed, and lane information. The axle load, speed, and specific timestamp of each vehicle's passage through the bridge are extracted. For example, for a three-axle truck, the timestamps are: front axle load 7 tons, middle axle load 12 tons, rear axle load 12 tons, speed 60 km / h, and passage through the bridge entrance "2026-08-2014:25:10.500". Based on the axle load and speed, the equivalent impact load spectrum generated by each vehicle on the bridge structure is calculated. The calculation of the equivalent impact load spectrum considers the vehicle's axle load, wheelbase, speed, and bridge smoothness, transforming the moving concentrated axle load into a time-history force function acting on the bridge deck nodes. The equivalent impact load spectrum can be represented as a series of time-force sequences. Its peak value is proportional to the axle load and is related to vehicle speed and bridge surface smoothness spectrum.

[0104] The equivalent impact load spectrum of each vehicle is defined as a pulse excitation signal, and this pulse excitation signal is applied to the corresponding lane position nodes of the initial finite element model. The corresponding lane position nodes of the initial finite element model are predefined in the model based on lane markings. For example, if a vehicle is traveling in the second lane, its equivalent impact load spectrum... The force will act on all element nodes representing the second lane in the initial finite element model, and the location of the force will change dynamically with the movement of the vehicle. The transient dynamic response of the initial finite element model under the action of the pulse excitation signal is solved using the modal superposition method. The modal superposition method selects the first 20 modes of the initial finite element model for superposition calculation, and solves the equivalent impact load spectrum. The displacement, velocity, and acceleration response time histories of each node of the bridge under simulated vehicle loads were obtained. The node displacement time history curves corresponding to the sensor positions at the mid-span of the bridge were extracted from the transient dynamic response. These curves describe the vertical displacement of the measuring point at the mid-span over time under simulated vehicle loads. The node displacement time history curves were converted into strain time history curves. This conversion was based on material mechanics and geometric relationships. The bending strain time history at the node location was obtained by calculating the ratio of the displacement difference between the upper and lower edges of the beam element to the initial height of the element. The curves were then stitched together according to the chronological order of vehicle passage. The strain time history curves obtained from the individual simulations of each vehicle were arranged sequentially on the time axis based on the specific timestamp of its passage over the bridge. The intervals between the curves were determined according to the actual time intervals between vehicle passages.

[0105] In some embodiments, the complete time series obtained after splicing is the theoretical strain response curve of the bridge under traffic load. Theoretically, the theoretical strain response curve should be consistent with the actual measured dynamic strain time series data under the same traffic load in terms of waveform, amplitude, and phase. The cross-correlation coefficient is calculated between the theoretical strain response curve and the actually collected dynamic strain time series data. A time window, such as a 10-second window, is selected for the cross-correlation coefficient calculation. The calculation is performed synchronously across the theoretical strain response curve and the dynamic strain time series data, calculating the cross-correlation coefficient between the two curves within each window. Cross-correlation coefficient The calculation formula is:

[0106]

[0107] in: It is the value of the measured dynamic strain time series data at time t. The theoretical strain response curve is in The value at time, and These are the mean values ​​of their respective sequences within the calculation window. It is a time lag. This is the total number of data points within the calculation window. The maximum cross-correlation coefficient obtained is the waveform similarity index between the theoretical strain response curve and the dynamic strain time series data within that window. .

[0108] It is understandable that if the calculated waveform similarity index If the value falls below a set lower threshold, which can be determined based on historical health data statistical analysis (e.g., set to 0.75), it is determined that there is nonlinear response caused by damage within the bridge structure. The waveform similarity index will be used... Time periods with values ​​below a threshold of 0.75 are marked as suspicious periods, such as "14:25:10 to 14:25:20," and are fed back as constraints in the stiffness degradation matrix solution process. During the stiffness degradation matrix solution process, dynamic strain time-series data and vibration acceleration spectrum data collected within the suspicious periods are given higher weights, or the inversion solution is constrained to ensure that the theoretical model's response within the suspicious period has higher consistency with the measured data, thereby guiding the inversion algorithm to focus more on time periods exhibiting abnormal behavior under specific load conditions.

[0109] Optionally, the cross-correlation coefficient calculation is not limited to the overall waveform. Spectral analysis can be performed on the theoretical strain response curve and the measured dynamic strain time-series data separately to compare the similarity of their energy distribution within characteristic frequency bands, or to compare the differences in their impact response attenuation rates, serving as a supplement to the waveform similarity index. The extracted equivalent impact load spectrum includes not only vertical force but also, based on vehicle type and speed, the braking force or centrifugal force components generated. These components are also input into the model as part of the pulse excitation signal to more accurately simulate actual traffic loads. When multiple vehicles are present on the bridge simultaneously, their equivalent impact load spectra will superimpose in time. This multi-vehicle coupling effect needs to be considered when calculating the theoretical strain response curve. The vehicle load spectra of different lanes are superimposed in the time domain and then input into the model to calculate the transient dynamic response.

[0110] See Figure 5 This is a time-series correlation diagram of traffic load and bridge strain response. It's a core visualization result of the "Traffic Load Time-Series Correlation Analysis" step in bridge structural anomaly identification, intuitively demonstrating the mechanical response patterns of bridges under vehicle loads. Each time the traffic load peaks, the actual strain response also peaks simultaneously, verifying the fundamental mechanical law that "the greater the load, the more significant the structural deformation." The third load peak corresponds to the highest strain response, showing a positive correlation between load and strain, quantifying the mapping relationship between load and structural response. The strain response is not a single spike but exhibits a small-amplitude oscillation and decay, which is a typical dynamic response of a bridge under impact load within its elastic range. During non-load periods, the strain response approaches zero, eliminating interference from non-load factors such as ambient temperature and equipment noise, ensuring data validity.

[0111] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A bridge structural anomaly identification method based on data fusion, characterized in that, The method includes: Based on the existing design drawings and geological survey reports of the bridge structure, an initial finite element model of the bridge structure is constructed. The initial finite element model includes the geometric and topological relationships and material mechanical parameters of the beams, piers and supports. Multi-source sensor arrays are deployed at the mid-span, supports, and pier roots of the bridge structure to simultaneously collect dynamic strain time-series data, vibration acceleration spectrum data, and environmental temperature and humidity change data of the bridge structure during operation. Wavelet packet decomposition is performed on the dynamic strain time series data to extract energy entropy features in different frequency bands. The energy entropy features are then fused with the modal frequency offsets extracted from the vibration acceleration spectrum data to generate a comprehensive health state vector of the bridge structure. The comprehensive health state vector is physically corrected using the environmental temperature and humidity change data to eliminate spurious strain signals caused by temperature stress and obtain the corrected structural response feature set. The modified structural response feature set is input into the initial finite element model. The stiffness degradation matrix of the bridge structure in the current state is obtained by inversion calculation and compared with the standard stiffness matrix of the initial finite element model to identify structural elements with abrupt stiffness changes and generate a preliminary abnormal positioning coordinate set.

2. The bridge structural anomaly identification method based on data fusion according to claim 1, characterized in that, Based on the existing design drawings and geological survey report of the bridge structure, an initial finite element model of the bridge structure is constructed, including: The existing design drawings of the bridge structure are analyzed to extract the structural geometric dimensions, spatial topological connections and material designations of the beams, piers and supports. Based on the material designations, a material mechanics handbook is consulted to assign material mechanics parameters such as elastic modulus, Poisson's ratio and density to each material. Import the geological survey report, establish a distribution model of different soil layers under the bridge pier foundation based on borehole sampling data, and calculate the equivalent spring stiffness coefficient of each soil layer in combination with soil mechanics parameters to simulate the vertical and horizontal constraints of the soil layers on the bridge pier. The beam body and piers are discretized by using the beam grid method or spatial beam elements, the supports are modeled by connecting elements, and the soil constraints are modeled by a series of spring elements with corresponding equivalent spring stiffness coefficients, which are combined to form a complete structural finite element model. Static analysis was performed on the established complete structural finite element model under its own weight, and the initial deformation results obtained from the solution were compared and verified with the precamber in the design drawings. If the error between the calculated initial deformation result and the pre-camber is less than the preset tolerance threshold, the modeling is confirmed to be correct, and this complete structural finite element model and its material mechanical parameters and standard stiffness matrix are used as the initial finite element model; otherwise, the modeling parameters are iteratively adjusted until the error meets the standard.

3. The bridge structural anomaly identification method based on data fusion according to claim 2, characterized in that, The dynamic strain time-series data is subjected to wavelet packet decomposition to extract energy entropy features in different frequency bands. These energy entropy features are then fused with the modal frequency offsets extracted from the vibration acceleration spectrum data at the feature level to generate a comprehensive health state vector for the bridge structure, including: A wavelet basis function with tight support characteristics is selected, and a four-level wavelet packet decomposition is performed on the dynamic strain time series data to decompose the signal into sixteen independent frequency band subspaces; The probability distribution of signal energy in each frequency band subspace is calculated, and the energy entropy value of each frequency band is calculated based on the Shannon entropy definition to form a frequency band energy entropy sequence; The peak picking algorithm is applied to the vibration acceleration spectrum data to extract the measured values ​​of the first six natural modal frequencies, and the relative percentage of offset between the measured values ​​and the theoretical values ​​obtained from the simulation of the initial finite element model is calculated to form a modal frequency offset sequence. The frequency band energy entropy sequence and the modal frequency offset sequence are aligned according to the time sequence and mapped to a unified feature space through normalization processing; A weighting factor calculated from the environmental temperature and humidity change data is applied to each data point in the mapped feature space, and the weighted sum is used to obtain the comprehensive health status vector of the bridge structure.

4. The bridge structural anomaly identification method based on data fusion according to claim 3, characterized in that, The comprehensive health state vector is physically corrected using the environmental temperature and humidity change data to eliminate spurious strain signals caused by temperature stress, including: The ambient temperature curve and relative humidity curve are separated from the ambient temperature and humidity change data, and the daily periodic variation component of the ambient temperature curve is identified. Based on the concrete thermal expansion coefficient and elastic modulus of the bridge structure, a lookup table of the linear correspondence between ambient temperature and structural thermal strain is established. Using the linear correspondence lookup table, the daily periodic variation components of the ambient temperature curve are converted into an equivalent thermal strain time series. Subtracting the equivalent thermal strain time series from the dynamic strain time series data yields residual strain time series data free from temperature interference. The residual strain time series data is resubmitted into the calculation process of the frequency band energy entropy, and the comprehensive health state vector is updated to obtain the corrected structural response feature set.

5. The bridge structural anomaly identification method based on data fusion according to claim 4, characterized in that, The modified structural response feature set is input into the initial finite element model, and the stiffness degradation matrix of the bridge structure in the current state is obtained by inversion calculation, including: A diagonal perturbation factor is introduced into the global stiffness matrix of the initial finite element model to construct the sensitivity analysis equation to be solved. Using the measured modal frequencies in the modified structural response feature set as the target values ​​and the diagonal perturbation factor as the independent variable, an objective function is constructed. The objective function is iteratively optimized using a sequential quadratic programming algorithm to find the optimal combination of diagonal perturbation factors that makes the simulated modal frequencies approximate the measured modal frequencies. The obtained optimal diagonal perturbation factor combination is used to replace the elements in the global stiffness matrix of the initial finite element model according to the corresponding degree of freedom position, and the stiffness degradation matrix of the bridge structure in the current state is reconstructed.

6. The bridge structural anomaly identification method based on data fusion according to claim 5, characterized in that, By comparing the standard stiffness matrix of the initial finite element model with the matrix, structural elements with abrupt stiffness changes are identified, and a preliminary set of anomaly location coordinates is generated, including: The stiffness variation matrix is ​​obtained by calculating the matrix difference between the stiffness degradation matrix and the standard stiffness matrix of the initial finite element model. Iterate through each non-zero element in the stiffness variation matrix and calculate the absolute amount and relative percentage of its stiffness loss; Set a threshold for the relative percentage of stiffness loss, and extract the degree of freedom numbers corresponding to stiffness loss values ​​that exceed the threshold; Based on the degree of freedom number, the geometric topological relationship of the initial finite element model is traced back to the specific structural component number and its coordinate position in three-dimensional space. All the parsed coordinate positions are clustered and deduplicated to form the preliminary abnormal location coordinate set.

7. The bridge structural anomaly identification method based on data fusion according to claim 6, characterized in that, It also includes a visually-assisted verification step based on crack image data: High-resolution crack images captured by an industrial camera array deployed at the bottom of the bridge are retrieved, and the high-resolution crack images are subjected to image enhancement and edge sharpening processing. The Hough transform algorithm is used to detect the straight crack profile in the processed high-definition crack image, and the length and orientation angle of the straight crack profile are measured. The length and orientation angle of the linear crack profile are spatially registered with the preliminary anomaly location coordinate set to determine whether there are overlapping or adjacent crack features. If there are overlapping or adjacent crack features, the timestamp of the high-definition crack image is read and associated with the timestamp of the dynamic strain time series data. Based on the correlation results, the dynamic strain time series data segments corresponding to the timestamp before and after the shooting are extracted, and it is analyzed whether the strain mutation rate in the data segments exceeds the preset safety limit, and an anomaly confirmation list with visual evidence markers is generated.

8. The bridge structural anomaly identification method based on data fusion according to claim 7, characterized in that, Spatial registration is performed between the length and orientation angle of the linear crack profile and the preliminary anomaly location coordinate set, including: Metadata information containing camera intrinsic and extrinsic parameters is extracted from the high-resolution crack image, and a transformation matrix from the image pixel coordinate system to the bridge structure world coordinate system is constructed based on the metadata information; Using the transformation matrix, the length and orientation angle of the straight crack profile are projected onto the world coordinate system of the bridge structure to obtain the spatial straight line equation of the crack in the world coordinate system. Traverse each coordinate point in the preliminary anomaly location coordinate set and calculate the vertical distance from the coordinate point to the spatial straight line equation; If the vertical distance is less than a preset spatial tolerance threshold, it is determined that there is a visually visible crack in the area where the coordinate point is located, and the confidence level of the coordinate point in the preliminary abnormal location coordinate set is increased by one level.

9. The bridge structural anomaly identification method based on data fusion according to claim 8, characterized in that, It also includes a time-series correlation analysis step based on traffic load data: Access the real-time vehicle data stream from bridge toll stations or weigh stations, and extract the axle load, speed, and specific timestamp of each vehicle passing through the bridge. Calculate the equivalent impact load spectrum of each vehicle on the bridge structure when it passes based on the axle load and vehicle speed; The equivalent impact load spectrum is convolved with the dynamic strain time series data to simulate the theoretical strain response curve of the bridge under traffic load. The cross-correlation coefficient between the theoretical strain response curve and the actually acquired dynamic strain time series data is calculated to obtain the waveform similarity index between the theoretical strain response curve and the dynamic strain time series data. If the waveform similarity index is lower than the set lower threshold, it is determined that there is nonlinear response caused by damage inside the bridge structure, and the time period corresponding to the waveform similarity index being lower than the threshold is marked as a suspicious time period, which is fed back into the process of solving the stiffness degradation matrix as a constraint condition.

10. The bridge structural anomaly identification method based on data fusion according to claim 9, characterized in that, The equivalent impact load spectrum is convolved with the dynamic strain time series data to simulate the theoretical strain response curve of the bridge under traffic load, including: The equivalent impact load spectrum of each vehicle is defined as a pulse excitation signal and applied to the corresponding lane position node of the initial finite element model. The transient dynamic response of the initial finite element model under the pulse excitation signal is solved using the modal superposition method. Extract the nodal displacement time history curves corresponding to the sensor positions at the mid-span of the bridge from the transient dynamic response; The nodal displacement time history curves are converted into strain time history curves and spliced ​​together according to the chronological order of vehicle passage. The complete time series obtained after splicing is the theoretical strain response curve of the bridge under traffic load.

Images