Bridge expansion joint state monitoring method, device and system based on vibration voiceprint collaborative perception

Abstract

The invention provides a bridge expansion joint state monitoring method, device and system based on vibration voiceprint collaborative awareness, and the method comprises the steps: collecting a vibration signal sequence and a voiceprint signal sequence of a bridge expansion joint, generating an original monitoring data set, carrying out the time-space correlation modeling of the vibration signal sequence and the voiceprint signal sequence, and generating a time sequence collaborative evolution field. And performing causal sequential difference deconstruction on the time sequence co-evolution field, separating a causal lag mode of the vibration signal sequence relative to the voiceprint signal sequence in a target frequency band range, mapping the causal lag mode to a preset physical force transmission path network, generating a hidden damage topological graph for describing structural performance weakening characteristics, and determining the structural performance weakening characteristics. And based on connectivity characteristics of nodes and strength characteristics of edges in the hidden damage topological graph, quantitatively evaluating the integrity state level and the local damage degree of the bridge expansion joint, and generating a monitoring result containing a state level identifier. According to the invention, the monitoring efficiency can be improved, and the stability and applicability in a complex operation environment can be ensured.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, device and system for monitoring the condition of bridge expansion joints based on vibration acoustic signature collaborative sensing. Background Technology

[0002] Bridge expansion joints are critical components of bridge structures, used to absorb deformations caused by temperature changes, loads, and foundation settlement. Their performance directly affects the safety and service life of the bridge. Condition assessment of expansion joints is crucial for ensuring traffic safety and structural durability. Currently, two main methods are relied upon: manual inspection and continuous sensor monitoring. Manual inspection relies on on-site observation and tapping to determine the condition, which suffers from low efficiency, traffic interruption, insufficient ability to identify minor faults, and severe limitations due to adverse weather and lighting conditions. While continuous sensor monitoring achieves automated data acquisition, it generally uses single-type sensors such as displacement or vibration sensors for uninterrupted operation, resulting in large data redundancy, high system energy consumption, and difficulty in effectively distinguishing between actual structural damage and environmental interference signals due to the limited monitoring dimensions, leading to false alarms and missed alarms, and failing to accurately locate hidden damage and its evolution trend. Summary of the Invention

[0003] In view of this, the present invention provides a method, device, and system for monitoring the condition of bridge expansion joints based on vibration acoustic signature collaborative sensing. The technical solution of the present invention is implemented as follows: In a first aspect, embodiments of the present invention provide a method for monitoring the state of bridge expansion joints based on vibration-acoustic fingerprint collaborative sensing. The method includes: synchronously acquiring vibration signal sequences and acoustic fingerprint signal sequences of bridge expansion joints under operational loads using a distributed sensor array, generating a time-stamp-aligned original monitoring data set, wherein the original monitoring data set includes vibration time-history curves and acoustic fingerprint spectral characteristic curves at different acquisition locations; performing spatiotemporal correlation modeling on the vibration signal sequences and acoustic fingerprint signal sequences in the original monitoring data set to generate a temporal co-evolution field describing the dynamic coupling relationship between vibration and acoustic fingerprint signals, wherein the dimension of the temporal co-evolution field is correlated with the sampling frequency and spatial distribution density of the original monitoring data set; and further... The temporal co-evolution field is deconstructed causally and sequentially to separate the causal hysteresis modes of the vibration signal sequence relative to the acoustic signature signal sequence within the target frequency band. The causal hysteresis modes include the phase difference variation trend and amplitude coupling coefficient between the vibration signal and the acoustic signature signal. The causal hysteresis modes are mapped to a preset physical force transmission path network to generate a latent damage topology map describing the weakening characteristics of structural performance. The nodes of the latent damage topology map correspond to key force-bearing units in the force transmission path, and the edges correspond to the force flow transmission intensity between units. Based on the connectivity characteristics of the nodes and the strength characteristics of the edges in the latent damage topology map, the overall performance level and local damage degree of the bridge expansion joint are quantitatively evaluated, and monitoring results including state level identifiers are generated.

[0004] Secondly, embodiments of the present invention provide a bridge expansion joint status monitoring device, the device comprising: a data acquisition module, configured to synchronously acquire vibration signal sequences and acoustic signature signal sequences of bridge expansion joints under operational loads through a distributed sensor array, generating a raw monitoring data set with timestamp alignment, the raw monitoring data set including vibration time history curves and acoustic signature spectral characteristic curves at different acquisition locations; a data association module, configured to perform spatiotemporal association modeling on the vibration signal sequences and acoustic signature signal sequences in the raw monitoring data set, generating a temporal co-evolution field describing the dynamic coupling relationship between vibration and acoustic signature signals, the dimension of the temporal co-evolution field being correlated with the sampling frequency and spatial distribution density of the raw monitoring data set; and a data deconstruction module, configured to deconstruct the data... A temporal co-evolution field is used to perform causal sequential difference deconstruction, separating the causal hysteresis modes of the vibration signal sequence relative to the acoustic signature signal sequence within the target frequency band. The causal hysteresis modes include the phase difference change trend and amplitude coupling coefficient between the vibration signal and the acoustic signature signal. A damage characterization module is used to map the causal hysteresis modes to a preset physical force transmission path network, generating a latent damage topology map describing the weakening characteristics of structural performance. The nodes of the latent damage topology map correspond to key force-bearing units in the force transmission path, and the edges correspond to the force flow transmission intensity between units. A state recognition module is used to quantitatively evaluate the overall performance level and local damage degree of the bridge expansion joint based on the connectivity characteristics of the nodes and the strength characteristics of the edges in the latent damage topology map, generating monitoring results containing state level identifiers.

[0005] Thirdly, the present invention provides a bridge expansion joint condition monitoring system, including a memory and a processor. The memory stores a computer program that can run on the processor, and the processor executes the program to implement the steps in the above-described method.

[0006] This invention synchronously collects vibration and acoustic signature signals using a distributed sensor array and generates a set of original monitoring data with timestamp alignment. It then performs spatiotemporal correlation modeling to construct a temporal co-evolution field. By deconstructing causal sequential differences, it separates the causal hysteresis mode of the vibration signal relative to the acoustic signature signal and maps it to a physical force transmission path network to generate a latent damage topology map. Finally, based on the connectivity characteristics of nodes and the strength characteristics of edges in the topology map, it quantitatively evaluates the overall performance level and the degree of local damage and generates monitoring results with state level labels. This method effectively integrates the dynamic coupling relationships of multimodal signals, revealing the inherent hysteresis characteristics between vibration and acoustic signals through causal analysis, making damage identification more accurate and reliable. Simultaneously, by mapping signal features to a physical force transmission path network, it achieves intuitive visualization and quantitative analysis of damage location and extent, improving the accuracy and interpretability of condition assessment. Furthermore, unsupervised causal sequential difference deconstruction and topological evaluation avoid reliance on manual labeling or preset thresholds, reducing errors caused by subjective intervention and enhancing the adaptability and robustness of the monitoring system. Moreover, spatiotemporal correlation modeling and networked processing can efficiently handle large-scale distributed sensor data, improving monitoring efficiency and ensuring stability and applicability in complex operating environments. Attached Figure Description

[0007] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present invention and, together with the specification, serve to explain the technical solutions of the present invention.

[0008] Figure 1 This is a schematic diagram illustrating the implementation process of a bridge expansion joint status monitoring method based on vibration acoustic signature collaborative sensing, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structural composition of a bridge expansion joint condition monitoring device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the hardware entity of a computer system provided in an embodiment of the present invention. Detailed Implementation

[0009] This invention provides a method for monitoring the condition of bridge expansion joints based on vibration and acoustic signature collaborative sensing. This method can be executed by a processor of a computer system. The computer system can refer to a device with data processing capabilities, such as a server, laptop, tablet, or desktop computer. For example, the computer system can be a system communicatively connected to a sensing device on the bridge, capable of receiving signals collected by the sensing device, or enabling the sensing device to collect signals and transmit them back.

[0010] Figure 1 This is a schematic diagram illustrating the implementation process of a bridge expansion joint condition monitoring method based on vibration acoustic signature collaborative sensing, as provided in an embodiment of the present invention. Figure 1As shown, the method includes: Step S100: synchronously collecting the vibration signal sequence and acoustic signature signal sequence of the bridge expansion joint under operating load through a distributed sensor array, generating a raw monitoring data set with timestamp alignment, the raw monitoring data set containing vibration time history curves and acoustic signature spectrum characteristic curves at different collection locations.

[0011] In bridge expansion joint condition monitoring, vibration signal sequences reflect the changes in the vibration of the expansion joint under operational loads over time, including information such as the amplitude and frequency of the vibration. This information reflects the dynamic response characteristics of the expansion joint. Acoustic signal sequences show the changes in the sound signals generated by the expansion joint during operation over time. The acoustic spectrum characteristic curve can display the energy distribution of the sound signal at different frequencies, and different acoustic features may correspond to different operating states of the expansion joint. Timestamp alignment ensures the temporal consistency of vibration signal sequences and acoustic signal sequences collected by different sensing units, facilitating subsequent comprehensive data analysis. The original monitoring data set integrates vibration signal sequences and acoustic signal sequences collected by all sensing units, and adds a timestamp to each data point. The vibration time history curve, with time as the horizontal axis and vibration amplitude as the vertical axis, intuitively presents the dynamic process of expansion joint vibration; the acoustic spectrum characteristic curve, with frequency as the horizontal axis and the energy of the acoustic signal at that frequency as the vertical axis, reflects the frequency characteristics of the acoustic signal.

[0012] For example, the distributed sensor array consists of multiple vibration sensors and acoustic sensors. These sensors are distributed at different locations along the bridge expansion joint to comprehensively acquire vibration and acoustic information of the expansion joint. For instance, a vibration sensor and an acoustic sensor are installed at predetermined intervals along the bridge expansion joint, creating a sensor network covering the entire expansion joint. When operational loads such as vehicles pass over the bridge, these sensors synchronously start acquiring data, recording the vibration signal sequence and acoustic signal sequence of the expansion joint under the operational load. After timestamp alignment, the acquired data generates the original monitoring dataset.

[0013] Step S200: Perform spatiotemporal correlation modeling on the vibration signal sequence and acoustic signature signal sequence in the original monitoring data set to generate a temporal co-evolution field describing the dynamic coupling relationship between vibration and acoustic signature signals. The dimension of the temporal co-evolution field is related to the sampling frequency and spatial distribution density of the original monitoring data set.

[0014] Spatiotemporal correlation modeling considers the temporal and spatial relationships between vibration signal sequences and acoustic signature signal sequences. The dynamic coupling relationship between vibration and acoustic signature signals refers to the dynamic changes in the mutual influence and interaction between vibration and acoustic signature signals during the operation of bridge expansion joints. Spatiotemporal correlation modeling can reveal the inherent laws governing this coupling relationship. The temporal co-evolution field is a field used to describe the temporal and spatial changes in the dynamic coupling relationship between vibration and acoustic signature signals. Its dimension is related to the sampling frequency and spatial distribution density of the original monitoring data set. The higher the sampling frequency, the more refined the signal changes can be captured, and the higher the dimension of the temporal co-evolution field may be; the denser the spatial distribution of sensing units, the richer the spatial information obtained, and the higher the dimension of the temporal co-evolution field will be.

[0015] As one implementation method, step S200 may specifically include the following steps S210~S260: Step S210: Perform timestamp interpolation processing on the vibration time history curve and acoustic signature spectrum characteristic curve in the original monitoring data set, correct the sampling delay of different sensing units, and generate vibration resampling sequence and acoustic signature resampling sequence with equal time intervals. The time interval between the vibration resampling sequence and the acoustic signature resampling sequence is consistent with the highest sampling frequency of the original monitoring data set.

[0016] Due to the influence of factors such as hardware and transmission lines of different sensing units, sampling times may differ, resulting in sampling delay. Timestamp interpolation aims to correct this sampling delay, ensuring that data collected by different sensing units are evenly spaced in time. By performing timestamp interpolation on vibration time-history curves and acoustic signature spectral characteristic curves, vibration resampling sequences and acoustic signature resampling sequences with equal time intervals can be generated. The time intervals of these resampling sequences are consistent with the highest sampling frequency of the original monitoring data set, thus ensuring that the resampling sequences retain the highest temporal resolution information of the original data.

[0017] As one implementation method, step S210 may specifically include the following steps S211~S216: Step S211: Read the vibration time history curve, acoustic spectrum characteristic curve and corresponding timestamp sequence of each sensing unit in the original monitoring data set. The timestamp sequence contains the acquisition time information of each sampling point.

[0018] Step S212: Perform linear regression fitting on the timestamp sequence of each sensing unit to obtain the time base deviation model parameters. The time base deviation model parameters include time offset and sampling interval correction coefficient. Adjust the model parameters based on the system reference clock signal to synchronize the time base of each sensing unit with the system reference clock.

[0019] Linear regression fitting is used to find the linear relationship between the timestamp sequence of each sensing unit and the system reference clock. By performing linear regression analysis on the timestamp sequence of each sensing unit, a model describing the time reference deviation can be obtained. The parameters of this model include time offset and sampling interval correction coefficient. The time offset represents the time difference between the timestamp of the sensing unit and the system reference clock, and the sampling interval correction coefficient is used to correct the sampling interval of the sensing unit to be consistent with the sampling interval of the system reference clock. The system reference clock signal is an accurate time reference. By comparing and adjusting the model parameters with the system reference clock signal, the time reference of each sensing unit can be synchronized with the system reference clock. For example, the time offset and sampling interval correction coefficient can be calculated based on the correspondence between the timestamp sequence of each sensing unit and the system reference clock signal. Then, the timestamp sequence of each sensing unit is adjusted according to these parameters to synchronize it with the system reference clock. For example, for the timestamp sequence of a sensing unit, each timestamp is adjusted accordingly based on the calculated time offset and sampling interval correction coefficient to ensure that the time reference of that sensing unit is consistent with the system reference clock.

[0020] Step S213: Using the highest sampling frequency of the original monitoring data set as the reference sampling frequency, resample the vibration time history curve and the acoustic signature spectrum characteristic curve, calculate the interpolation results at equal time intervals, and generate the vibration resampling sequence and the acoustic signature resampling sequence.

[0021] Resampling is the process of resampling the vibration time history curve and the acoustic signature spectrum characteristic curve based on a reference sampling frequency. Using the highest sampling frequency of the original monitoring data set as the reference sampling frequency ensures that the resampled sequence retains the highest temporal resolution information of the original data. During resampling, interpolation results need to be calculated at equal time intervals. Interpolation can be implemented using interpolation algorithms such as linear interpolation and spline interpolation. Through interpolation calculation, the values ​​of the vibration signal and acoustic signature signal can be obtained at the new equal time intervals, thereby generating the vibration resampling sequence and the acoustic signature resampling sequence. For example, first, the highest sampling frequency of the original monitoring data set is determined, and this is used as a reference to determine the resampling time interval. Then, for the vibration time history curve and the acoustic signature spectrum characteristic curve, at each equal time interval, the interpolation result at that point is calculated using an interpolation algorithm based on the information of adjacent known data points. Arranging the interpolation results of all equal time interval points sequentially yields the vibration resampling sequence and the acoustic signature resampling sequence.

[0022] Step S214: Establish the spatial coordinate mapping relationship of the sensor array, convert the physical installation position of each sensor unit into a two-dimensional rectangular coordinate system with the center of the bridge expansion joint as the origin, and generate a spatial coordinate matrix. The elements of the spatial coordinate matrix are the two-dimensional coordinate values ​​of the sensor units.

[0023] The spatial coordinate mapping relationship of the sensor array is used to clarify the relative position of each sensor unit in space. Converting the physical installation position of each sensor unit into a two-dimensional Cartesian coordinate system with the center of the bridge expansion joint as the origin is to standardize the spatial position representation of each sensor unit, facilitating subsequent data analysis and processing. The spatial coordinate matrix is ​​a matrix containing the two-dimensional coordinate values ​​of each sensor unit. By establishing the spatial coordinate mapping relationship and coordinate transformation, the physical position information of each sensor unit can be converted into matrix elements. For example, first, the direction and unit length of the two-dimensional Cartesian coordinate system are determined based on the actual situation of the bridge expansion joint. Then, the physical installation position of each sensor unit is measured using measuring tools. Next, a coordinate transformation algorithm is used to convert the physical installation position of each sensor unit into coordinates in the two-dimensional Cartesian coordinate system. Finally, these coordinate values ​​are arranged in a certain order to generate the spatial coordinate matrix.

[0024] Step S215: Associate the spatial coordinate matrix with the vibration resampling sequence and the acoustic resampling sequence according to the sensor unit position to generate a synchronization signal matrix with spatial coordinate markers. Each element of the synchronization signal matrix contains the vibration signal amplitude, acoustic signal amplitude, spatial coordinates and a calibrated timestamp.

[0025] Associating the spatial coordinate matrix with the vibration resampling sequence and the acoustic resampling sequence is to combine the spatial location information of the sensing unit with the corresponding vibration signal and acoustic signal to form a more comprehensive data matrix. By matching according to the sensing unit's location, it can be ensured that the spatial coordinates, vibration signal amplitude, acoustic signal amplitude, and calibrated timestamp of each sensing unit can be accurately correlated. For example, by traversing the spatial coordinate matrix, vibration resampling sequence, and acoustic resampling sequence, and based on the sensing unit's location information, the corresponding spatial coordinates, vibration signal amplitude, acoustic signal amplitude, and calibrated timestamp are combined into an element of the synchronization signal matrix. Each sensing unit is processed sequentially, ultimately generating a synchronization signal matrix with spatial coordinate markers.

[0026] Step S216: Calculate the standard deviation of the signal amplitude difference between adjacent sensing units at the same timestamp. When the standard deviation is less than the preset threshold, output the vibration resampling sequence and the acoustic resampling sequence. Otherwise, repeat the timestamp interpolation and resampling steps until the standard deviation meets the threshold requirement.

[0027] Calculating the standard deviation of the signal amplitude difference between adjacent sensing units at the same timestamp is to verify the effectiveness of timestamp interpolation and resampling. If the standard deviation is too large, it indicates significant signal differences between different sensing units, potentially suggesting time asynchrony or inaccurate resampling. The preset threshold is a standard set based on actual conditions. When the standard deviation is less than this threshold, the timestamp interpolation and resampling results are considered satisfactory, and vibration resampling sequences and acoustic resampling sequences can be output. If the standard deviation does not meet the threshold requirement, the timestamp interpolation and resampling steps need to be repeated, continuously adjusted and optimized until the standard deviation meets the threshold requirement. For example, the synchronization signal matrix is ​​traversed, and for each timestamp, the amplitude difference of the vibration signal and acoustic signal between adjacent sensing units is calculated. Then, the standard deviation of these amplitude differences is calculated separately. The calculated standard deviation is compared with the preset threshold, and based on the comparison result, it is decided whether to output the vibration resampling sequence and acoustic resampling sequence, or to repeat the timestamp interpolation and resampling steps.

[0028] Step S220: Perform numerical multiplication on the vibration resampling sequence and the acoustic resampling sequence to generate coupled feature values ​​under the same spatiotemporal coordinates. Generate a continuously updated coupled feature matrix sequence along the time axis through a sliding window. The row dimension of the coupled feature matrix corresponds to the sensor array position, and the column dimension corresponds to the sampling point within the window.

[0029] Numerical multiplication of the vibration resampling sequence and the acoustic signature resampling sequence is performed to obtain coupling feature values ​​reflecting the interaction between the vibration signal and the acoustic signature signal. Multiplication at the same spatiotemporal coordinates reveals the degree of coupling between the two signals at that location. By moving a sliding window along the time axis, coupling feature values ​​can be calculated within different time windows, generating a continuously updated sequence of coupling feature matrices. The row dimension of the coupling feature matrix corresponds to different positions of the sensor array, and the column dimension corresponds to the sampling points within the window, thus displaying the coupling characteristics of the vibration signal and the acoustic signature signal from both spatial and temporal dimensions. For example, the elements at the same spatiotemporal coordinates in the vibration resampling sequence and the acoustic signature resampling sequence are numerically multiplied to obtain the coupling feature values. Then, a sliding window is set, the size of which is determined according to actual needs. The window is slid sequentially along the time axis, and within each window, the coupling feature values ​​are arranged in the order of sensor array position and sampling points, resulting in a coupling feature matrix. As the window moves, new coupling feature matrices are continuously generated, ultimately yielding a continuously updated sequence of coupling feature matrices.

[0030] Step S230: Calculate the difference between the matrix elements of the current window and the previous window in the coupling feature matrix sequence to obtain the difference matrix that reflects the rate of change of signal coupling strength. Use the spatial location of the sensor array as nodes and the elements of the difference matrix as edge weights to generate a dynamic correlation network data structure.

[0031] The difference calculation of matrix elements between the current window and the previous window in the coupling feature matrix sequence is used to obtain the change in signal coupling strength between adjacent time windows. The elements in the difference matrix reflect the rate of change of signal coupling strength. By analyzing these rates of change, the dynamic change of the coupling relationship between vibration and acoustic signals over time can be understood. A dynamic correlation network data structure is constructed using the spatial locations of the sensor array as nodes and the elements of the difference matrix as edge weights, which can intuitively display the correlation of signal coupling strength changes between different sensing units. For example, the coupling feature matrix sequence is traversed sequentially. For the coupling feature matrix of the current window and the coupling feature matrix of the previous window, the corresponding matrix elements are subtracted to obtain the difference matrix. Then, each spatial location of the sensor array is treated as a node, and the corresponding elements in the difference matrix are used as the weights of the edges between nodes to construct a dynamic correlation network. As time progresses, the coupling feature matrix sequence is continuously updated, and the difference matrix and the dynamic correlation network also change dynamically accordingly.

[0032] Step S240: Perform time-frequency domain tensor decomposition on the dynamic correlation network data structure, decompose the spatiotemporal variation characteristics of the network into frequency-dependent components, spatial distribution components and time evolution components, and combine them to generate a multi-dimensional time-frequency feature tensor. The dimensions of the multi-dimensional time-frequency feature tensor include the frequency axis, the sensor spatial axis and the time axis.

[0033] Time-frequency domain tensor decomposition (TF-TWD) is a method for decomposing the complex spatiotemporal variation characteristics of dynamically correlated network data structures. This decomposition separates the network's spatiotemporal variation characteristics into frequency-dependent components, spatial distribution components, and temporal evolution components. The frequency-dependent component reflects the signal's characteristics at different frequencies, the spatial distribution component reflects the spatial distribution of sensing units, and the temporal evolution component shows the signal's changes over time. Combining these three components generates a multi-dimensional time-frequency feature tensor with three dimensions: frequency axis, sensor spatial axis, and time axis, providing a more comprehensive description of the time-frequency characteristics of vibration and acoustic signature signals. For example, a specialized time-frequency domain tensor decomposition algorithm is used to process the dynamically correlated network data structure. This algorithm decomposes the dynamically correlated network into frequency-dependent components, spatial distribution components, and temporal evolution components based on its spatiotemporal variation characteristics. Then, these three components are combined in the order of frequency axis, sensor spatial axis, and time axis to generate a multi-dimensional time-frequency feature tensor.

[0034] Step S250: Standardize the multi-dimensional time-frequency feature tensor, map the feature values ​​of each dimension to the same numerical range, calculate the geodesic distance between any two points in the standardized tensor, determine the geodesic path through nearest neighbor graph search, and generate a set of coordinate points of the manifold surface representing the topological structure of the signal coupling relationship.

[0035] Standardizing multi-dimensional time-frequency feature tensors eliminates dimensional differences between feature values ​​of different dimensions, making them comparable. Mapping feature values ​​to the same numerical range facilitates subsequent calculations and comparisons. Calculating the geodesic distance between any two points in the standardized tensor measures the shortest path length between them in the manifold space. Determining the geodesic path through a nearest neighbor graph search finds the optimal path between the two points. Finally, a set of coordinate points on the manifold surface representing the topological structure of signal coupling relationships is generated, showcasing the topological structure and trends of these coupling relationships.

[0036] As one implementation method, step S250 may specifically include the following steps S251~S256: Step S251: Standardize the frequency axis, spatial axis and time axis feature values ​​of the multi-dimensional time-frequency feature tensor respectively, and map the feature values ​​of each dimension to the same numerical range through linear transformation to eliminate the dimensional differences of different features and generate a standardized time-frequency feature tensor.

[0037] Standardizing the eigenvalues ​​of the frequency, spatial, and time axes of a multi-dimensional time-frequency feature tensor separately aims to ensure numerical consistency across different dimensions. Linear transformation is a commonly used standardization method that maps eigenvalues ​​across dimensions to a predefined range of values, thus eliminating dimensional differences. For example, the frequency, spatial, and time axes are processed separately. For each axis's eigenvalue, a linear transformation algorithm is used to map it to a predefined range based on its value. For instance, for the frequency axis's eigenvalues, the maximum and minimum values ​​are found, and then a linear transformation is applied to convert each eigenvalue to the predefined range. This process is repeated for all three axes to generate the standardized time-frequency feature tensor.

[0038] Step S252: Perform eigenvalue decomposition on the standardized time-frequency feature tensor, determine the manifold embedding dimension based on the eigenvalue contribution rate, and ensure that the manifold embedding dimension retains the main information entropy of the original tensor to construct the embedding space.

[0039] Eigenvalue decomposition (EVD) is used to decompose a standardized time-frequency feature tensor into eigenvalues ​​and eigenvectors. By calculating the contribution rate of each eigenvalue, we can determine which eigenvalues ​​contribute significantly to the information of the original tensor. Determining the manifold embedding dimension based on the eigenvalue contribution rate aims to embed the data into a suitable low-dimensional space while preserving the main information entropy of the original tensor. Constructing the embedding space transforms high-dimensional tensor data into a low-dimensional representation, facilitating subsequent analysis and processing. For example, we use EVD to process the standardized time-frequency feature tensor to obtain eigenvalues ​​and eigenvectors. Then, we calculate the contribution rate of each eigenvalue, i.e., the proportion of that eigenvalue to the sum of all eigenvalues. The eigenvalues ​​are sorted from largest to smallest contribution rate, and the dimensions corresponding to the eigenvalues ​​with the largest contribution rates are selected as the manifold embedding dimensions. Finally, we construct the embedding space based on these selected dimensions and project the original tensor data into this embedding space.

[0040] Step S253: In the embedding space, the nearest neighbor number is determined by a certain appropriate ratio of the number of spatial locations of the sensing array. The nearest neighbor point of each tensor point is found by searching the nearest neighbor graph. The nearest neighbor points are connected to form a network structure. The shortest path between any two points in the network is calculated as the geodesic distance.

[0041] In the embedding space, it is necessary to determine the nearest neighbors of each tensor point. Using a suitable proportion of the number of spatial locations in the sensor array as the number of nearest neighbors is to reasonably determine the nearest neighbor range of each tensor point. The nearest neighbor graph search algorithm can be used to find the nearest neighbors of each tensor point. These nearest neighbors are connected to form a network structure. The shortest path between any two points in the network is calculated as the geodesic distance, which measures the shortest distance between the two points in the manifold space. For example, for each tensor point in the embedding space, the nearest neighbor graph search algorithm is used to find the nearest neighbors of that tensor point based on the predetermined number of nearest neighbors. Then, these nearest neighbors are connected to the tensor point to form edges, constructing a network. Next, the shortest path algorithm is used to calculate the shortest path between any two points in the network, which is then used as the geodesic distance between the two points.

[0042] Step S254: Construct a distance matrix based on geodesic distance, and convert the distance matrix into a set of coordinate points in Euclidean space through multidimensional scaling analysis to generate preliminary embedded coordinates. The dimension of the embedded coordinates is set to three dimensions.

[0043] Constructing a distance matrix based on geodesic distance involves storing the geodesic distance between any two points in the network within a matrix. Multidimensional scaling analysis (MSA) is a method that transforms the distance matrix into a set of coordinate points in Euclidean space. This analysis converts distance relationships in manifold space into Euclidean coordinate representations. Preliminary embedded coordinates are generated, and their dimensions are set to three dimensions for easy visualization and analysis. For example, based on the calculated geodesic distances, the geodesic distances between any two points are used as matrix elements to construct a distance matrix. Then, the MSA algorithm is used to process the distance matrix. This algorithm maps the data to Euclidean space based on the distance relationships in the distance matrix, generating preliminary embedded coordinates. Setting the dimensions of the embedded coordinates to three dimensions allows for visualization and further analysis of the data in three-dimensional space.

[0044] Step S255: Perform Riemannian metric correction on the initial embedded coordinates, calculate the Riemann curvature tensor of each coordinate point, adjust the coordinate position through gradient descent to minimize the error between the distance in Euclidean space and the geodesic distance on the Riemann manifold, and generate the corrected manifold coordinates.

[0045] The Riemannian metric correction of the initial embedded coordinates aims to make the distances in Euclidean space closer to the geodesic distances on the Riemannian manifold. Calculating the Riemannian curvature tensor at each coordinate point reveals the local geometric properties of the manifold. The coordinate positions are adjusted using a gradient descent algorithm, progressively optimizing the coordinates to minimize the error between the distances in Euclidean space and the geodesic distances on the Riemannian manifold. This ultimately generates the corrected manifold coordinates, improving the accuracy of the coordinate representation. For example, the Riemannian curvature tensor at each coordinate point in the initial embedded coordinates is first calculated using algorithms and formulas related to Riemannian geometry. Then, based on the Riemannian curvature tensor and the error between the distances in Euclidean space and the geodesic distances on the Riemannian manifold, the gradient descent algorithm is used to adjust the coordinate positions. The gradient descent algorithm iteratively updates the coordinate positions, gradually reducing the error until a preset error threshold is met, ultimately yielding the corrected manifold coordinates.

[0046] Step S256: Sort the corrected manifold coordinates according to the time series to generate a set of manifold trajectory coordinate points that evolve over time. Connect the discrete trajectory points through nonlinear interpolation to generate a set of manifold surface coordinate points that characterize the topological structure of signal coupling relationships.

[0047] Sort the corrected manifold coordinates by time series to show how the manifold coordinates change over time, generating a set of manifold trajectory coordinate points that evolve over time. Connecting the discrete trajectory points using nonlinear interpolation transforms the discrete data into a continuous manifold surface, generating a set of manifold surface coordinate points representing the topological structure of signal coupling relationships. This provides a more intuitive display of the topological structure and changing trends of signal coupling relationships. For example, the coordinate points are sorted according to the time information in the corrected manifold coordinates. The sorted coordinate points are then arranged sequentially to form a set of manifold trajectory coordinate points that evolve over time. Then, a nonlinear interpolation algorithm is used to estimate the coordinate values ​​of intermediate positions based on adjacent discrete trajectory points, connecting these points to form a continuous manifold surface. Finally, a set of manifold surface coordinate points representing the topological structure of signal coupling relationships is obtained.

[0048] Step S260: Extend the time axis of the manifold surface coordinate point set, fill the manifold state at adjacent sampling times with nonlinear interpolation, generate continuously evolving dynamic surface data, and obtain a time-series co-evolution field describing the dynamic coupling relationship between vibration and acoustic signals. The smoothness of the time-series co-evolution field is related to the sampling frequency of the original monitoring data set.

[0049] Extending the manifold surface coordinate point set along the time axis makes the manifold surface more continuous in time. Since sampling times are discrete, the manifold state between adjacent sampling times may be missing. Filling in the manifold state between adjacent sampling times using nonlinear interpolation can compensate for these missing values, generating continuously evolving dynamic surface data. This dynamic surface data constitutes a temporal co-evolution field describing the dynamic coupling relationship between vibration and acoustic signature signals. The smoothness of the temporal co-evolution field is related to the sampling frequency of the original monitoring data set; the higher the sampling frequency, the more refined the signal changes can be captured, and the smoother the temporal co-evolution field. For example, adjacent sampling times are determined based on the time information in the manifold surface coordinate point set. Then, using a nonlinear interpolation algorithm, the manifold state at intermediate times is estimated based on the manifold state at adjacent sampling times, filling in the manifold state data at these intermediate times. Combining all the filled data forms the continuously evolving dynamic surface data, thus obtaining the temporal co-evolution field describing the dynamic coupling relationship between vibration and acoustic signature signals.

[0050] Step S300: Perform causal sequential difference deconstruction on the temporal co-evolution field to separate the causal hysteresis mode of the vibration signal sequence relative to the acoustic pattern signal sequence in the target frequency band. The causal hysteresis mode includes the phase difference change trend and amplitude coupling coefficient between the vibration signal and the acoustic pattern signal.

[0051] Causal sequential difference deconstruction is used to determine the causal relationship and differences between vibration signal sequences and acoustic signature sequences from a temporal co-evolution field. Separating the causal hysteresis modes of the vibration signal sequence relative to the acoustic signature sequence within the target frequency band is to clarify the causal hysteresis characteristics between the vibration and acoustic signature signals within a specific frequency band. The phase difference variation trend contained in the causal hysteresis modes reflects the temporal sequence relationship between the vibration and acoustic signature signals, while the amplitude coupling coefficient reflects the degree of correlation between their amplitudes.

[0052] As one implementation method, step S300 may specifically include the following steps S310~S360: Step S310: Extract the signal coupling evolution trajectory at different spatial locations in the temporal co-evolution field. Each signal coupling evolution trajectory corresponds to the change curve of the vibration-acoustic coupling relationship of a sensing unit over time. The horizontal axis of the signal coupling evolution trajectory is the time sampling point, and the vertical axis is the geodesic coordinate value on the manifold surface.

[0053] Extracting signal coupling evolution trajectories at different spatial locations within the temporal co-evolution field is crucial for analyzing the temporal variation of the vibration-acoustic coupling relationship of each sensing unit. Each signal coupling evolution trajectory corresponds to a sensing unit, with its x-axis representing the time sampling point and its y-axis representing the geodesic coordinates on the manifold surface. This allows for the visualization of the temporal and spatial variations in the vibration-acoustic coupling relationship of the sensing unit within a single curve. For example, each spatial location within the temporal co-evolution field is traversed, and the signal coupling evolution trajectory at each location is extracted. Based on the temporal information within the temporal co-evolution field and the geodesic coordinates on the manifold surface, the corresponding data points are sequentially connected to form the signal coupling evolution trajectory.

[0054] Step S320: Filter each coupled evolution trajectory, denoise it by local weighted averaging within a sliding window, preserve the trend characteristics of the trajectory, recursively accumulate the distance between any two filtered trajectories to generate a distance value that reflects the similarity of the trajectories, and combine all the distance values ​​to form a trajectory distance matrix.

[0055] Filtering each coupled evolution trajectory removes noise interference while preserving its trend characteristics. Local weighted averaging within a sliding window is a common filtering method that smooths the trajectory curve and highlights its main trend. Recursively accumulating the distance between any two filtered trajectories measures their similarity. Combining the distance values ​​between all trajectories forms a trajectory distance matrix, facilitating subsequent clustering analysis.

[0056] As one implementation method, step S320 may specifically include the following steps S321 to S326: Step S321: Filter the sampling point data of each coupled evolution trajectory, set the window length to an appropriate proportion of the total number of trajectory time sampling points, use a polynomial to fit the data within the window, calculate the fitting result as the filtered trajectory data, and retain the trend characteristics of the trajectory.

[0057] Filtering the sampled data points for each coupled evolution trajectory removes noise while preserving trend characteristics. Setting the window length to a suitable proportion of the total number of trajectory time-series sampling points determines the filtering range. Polynomial fitting of the data within the window approximates the data within the window using a polynomial function, and the fitting result is used as the filtered trajectory data. This smooths the trajectory curve and highlights its trend characteristics. For example, for each coupled evolution trajectory, the window length is determined according to a preset proportion. Then, the window is slid sequentially across the trajectory, and within each window, a polynomial fitting algorithm is used to fit the sampled data points within that window. The fitting result is used as the filtered trajectory data for that window. Each window is processed sequentially to obtain the final filtered trajectory data.

[0058] Step S322: For the two filtered trajectories, construct a cumulative distance matrix with the trajectory sampling points as nodes. The matrix elements are the Euclidean distances between the corresponding nodes of the two trajectories. Calculate the minimum cumulative distance of each node using a recursive formula that considers the distances between the current node and the previous node, between the current node and the current node, and between the previous node and the current node.

[0059] For the two filtered trajectories, a cumulative distance matrix is ​​constructed using the sampled points as nodes to measure the distance relationship between the two trajectories. The matrix elements represent the Euclidean distances between corresponding nodes of the two trajectories, intuitively reflecting the distances between nodes. The minimum cumulative distance for each node is calculated using a recursive formula to find the optimal matching path between the two trajectories. The recursive formula considers the distances between the current node and the previous node, the current node and the current node, and the previous node and the current node to ensure that the calculated minimum cumulative distance is optimal. For example, first, the number of sampled points for the two filtered trajectories is determined. Then, a cumulative distance matrix is ​​constructed using the sampled points as nodes, with the number of rows and columns corresponding to the number of sampled points for the two trajectories. For each element in the matrix, the Euclidean distance between corresponding nodes of the two trajectories is calculated. Next, using the recursive formula, starting from the top left corner of the matrix, the minimum cumulative distance for each node is calculated sequentially. The recursive formula selects the minimum cumulative distance for the current node based on the distances between the current node and the previous node, the current node and the current node, and the previous node and the current node.

[0060] Step S323: Solve for the minimum path of the cumulative distance matrix using dynamic programming. The path extends from the upper left corner to the lower right corner of the matrix. The sum of the elements on the path is the distance between the two trajectories. The smaller the distance value, the higher the similarity between the trajectories.

[0061] Dynamic programming is a method for finding optimal paths. By finding the minimum path in the cumulative distance matrix using dynamic programming, the shortest matching path between two trajectories can be found. The path extends from the top left corner to the bottom right corner of the matrix, and the sum of the elements on the path is the distance between the two trajectories. The smaller the distance, the higher the similarity between the two trajectories. For example, dynamic programming is used to process the cumulative distance matrix. The algorithm starts from the top left corner of the matrix and calculates step by step to the bottom right corner to determine a path that minimizes the sum of the elements on the path. The elements on this path are recorded, and they are added together to obtain the distance between the two trajectories. This distance value can be used to determine the similarity between the two trajectories.

[0062] Step S324: Calculate the pairwise distance values ​​for the coupled evolution trajectories of all sensing units, arrange the distance values ​​according to the trajectory number, and generate a trajectory distance matrix. The rows and columns of the trajectory distance matrix correspond to the sensing unit numbers, the diagonal elements are zero, and the off-diagonal elements are the distance values ​​of the corresponding two trajectories.

[0063] Calculating pairwise distances for the coupled evolution trajectories of all sensing units is crucial for comprehensively assessing the similarity between trajectories. The distance values ​​are then arranged by trajectory index to generate a trajectory distance matrix. Rows and columns of the matrix correspond to sensing unit indices, diagonal elements are set to zero (because the distance between elements is zero), and off-diagonal elements represent the distances between the corresponding two trajectories. This allows for a single matrix to display the distance relationships between all trajectories, facilitating subsequent clustering analysis. For example, the coupled evolution trajectories of all sensing units are traversed, and distance values ​​are calculated for any two trajectories. The calculated distance values ​​are then arranged in order of trajectory index and stored in the trajectory distance matrix. The diagonal elements of the trajectory distance matrix are set to zero. This process ultimately generates the complete trajectory distance matrix.

[0064] Step S325: Normalize the elements of the trajectory distance matrix and map the distance values ​​to the same numerical range through linear transformation to generate a normalized trajectory distance matrix, so that the distance values ​​of different sensing unit groups are comparable.

[0065] Normalizing the elements of the trajectory distance matrix eliminates dimensional differences in distance values ​​between different sensor unit groups. By mapping the distance values ​​to the same numerical range through a linear transformation, a normalized trajectory distance matrix is ​​generated, making the distance values ​​of different sensor unit groups comparable. For example, the maximum and minimum values ​​in the trajectory distance matrix are found. Then, a linear transformation formula is used to map each element in the matrix to a preset numerical range. After mapping all elements, the normalized trajectory distance matrix is ​​obtained.

[0066] Step S326: Calculate the row mean and column mean of the normalized trajectory distance matrix. If the difference between the row mean and the column mean is less than a preset threshold, the matrix symmetry is deemed to be qualified. Otherwise, the trajectory filtering and distance calculation steps are re-executed until the symmetry requirement is met.

[0067] Calculating the row and column means of the normalized trajectory distance matrix is ​​to check the matrix's symmetry. If the difference between the row and column means is less than a preset threshold, the matrix has good symmetry and is considered symmetric. If the symmetry requirement is not met, the trajectory filtering and distance calculation steps need to be repeated, continuously adjusted and optimized until the matrix meets the symmetry requirement. For example, the row and column means of the normalized trajectory distance matrix are calculated separately. Then, the difference between the row and column means is calculated. This difference is compared with a preset threshold, and the comparison result determines whether the matrix is ​​symmetric or whether the trajectory filtering and distance calculation steps need to be repeated.

[0068] Step S330: Perform clustering processing on the trajectory distance matrix, divide the trajectories with smaller distance values ​​into the same cluster, determine the number of clusters by calculating the contour coefficient, so that the similarity of trajectories within a cluster is higher than the similarity of trajectories between clusters, and each cluster corresponds to a group of sensing units with similar coupling characteristics.

[0069] Clustering the trajectory distance matrix aims to group trajectories with similar characteristics into the same cluster. Trajectories with smaller distance values ​​are grouped together because smaller distance values ​​indicate higher trajectory similarity. Determining the number of clusters using silhouette coefficient calculation aims to find the optimal clustering method, ensuring that intra-cluster trajectory similarity is higher than inter-cluster trajectory similarity. Each cluster corresponds to a group of sensor units with similar coupling characteristics, facilitating grouping of sensor units for subsequent analysis. For example, a clustering algorithm is used to process the trajectory distance matrix, experimenting with different numbers of clusters. For each number of clusters, a silhouette coefficient is calculated. The silhouette coefficient reflects the difference between intra-cluster and inter-cluster trajectory similarity; a larger silhouette coefficient indicates better clustering. The number of clusters corresponding to the largest silhouette coefficient is selected to divide the trajectories into different clusters, with each cluster containing sensor units with similar coupling characteristics.

[0070] Step S340: Perform time-frequency analysis on the coupled evolution trajectory of each cluster. Calculate the instantaneous frequency and energy value of the trajectory within the window by sliding a window function along the time axis, and generate a time-frequency energy spectrum. The horizontal axis of the time-frequency energy spectrum represents time, the vertical axis represents frequency, and the color intensity represents the magnitude of the energy value.

[0071] Time-frequency analysis of the coupled evolution trajectory of each cluster is performed to understand its characteristics in time and frequency. A window function is slid along the time axis to divide the trajectory into different time windows. Within each window, the instantaneous frequency and energy value of the trajectory are calculated. The time-frequency energy spectrum plot, with time on the horizontal axis and frequency on the vertical axis, uses color intensity to represent energy value magnitude, visually displaying the energy distribution of the trajectory in the time-frequency domain. For example, a suitable window function is selected for the coupled evolution trajectory of each cluster. The window function is slid sequentially along the time axis, and within each window, the instantaneous frequency and energy value of the trajectory are calculated using a time-frequency analysis algorithm. The calculated instantaneous frequencies and energy values ​​are arranged in order of time and frequency to generate a time-frequency energy spectrum plot. Different color intensities are used to represent the energy values ​​in the spectrum.

[0072] Step S350: Calculate the accumulated energy along the frequency axis in the time-frequency energy spectrum. When the accumulated energy reaches a preset proportion of the total energy, record the corresponding frequency range as the target frequency band. The preset proportion is preset according to the vibration characteristics of the bridge expansion joint.

[0073] Calculating the accumulated energy along the frequency axis in the time-frequency energy spectrum is to determine the energy accumulation at different frequencies. When the accumulated energy reaches a preset proportion of the total energy, the corresponding frequency range is recorded as the target frequency band. This target frequency band contains the main energy information of the vibration signal and acoustic signature signal. The preset proportion is determined based on the vibration characteristics of the bridge expansion joint. Different bridge expansion joints may have different vibration characteristics, so the preset proportion needs to be determined according to the actual situation. For example, starting from the low-frequency end of the time-frequency energy spectrum, the accumulated energy is calculated sequentially along the frequency axis. For each frequency point, the energy values ​​of that frequency point and the previous frequency points are added together to obtain the accumulated energy. When the accumulated energy reaches the preset proportion of the total energy, the frequency range at this point is recorded and used as the target frequency band.

[0074] Step S360: Perform causal relationship analysis on the coupling evolution trajectory within the target frequency band, calculate the probability value of the causal influence of the vibration signal on the acoustic pattern signal, and when the probability value is greater than the preset threshold, extract the phase difference sequence and amplitude ratio sequence of the vibration signal relative to the acoustic pattern signal within the frequency band, and combine them to generate a causal lag mode. The time length of the causal lag mode is consistent with the number of time sampling points of the coupling evolution trajectory.

[0075] Causal relationship analysis of the coupled evolution trajectory within the target frequency band is performed to determine the causal relationship between the vibration signal and the acoustic signature signal. Calculating the probability value of the causal influence of the vibration signal on the acoustic signature signal measures the degree of influence. When the probability value is greater than a preset threshold, it indicates a significant causal relationship. In this case, the phase difference sequence and amplitude ratio sequence of the vibration signal relative to the acoustic signature signal within that frequency band are extracted and combined to generate a causal hysteresis mode. The time length of the causal hysteresis mode is consistent with the number of time sampling points of the coupled evolution trajectory to ensure that the causal hysteresis mode accurately reflects the signal characteristics within the target frequency band.

[0076] As one implementation method, step S360 may specifically include the following steps S361 to S366: Step S361: Extract the vibration component sequence and acoustic component sequence within the target frequency band from the original monitoring data set, retain the signal component within the target frequency band through bandpass filtering, attenuate the signal component outside the frequency band, and reduce the amplitude of the signal outside the frequency band to below a preset ratio of the amplitude of the signal within the frequency band.

[0077] Extracting vibration and acoustic signature sequences within the target frequency band from the original monitoring data set aims to focus on the signal characteristics within that band. Bandpass filtering retains only the signal components within the target band while attenuating the signal components outside the band, reducing the amplitude of the out-of-band signal to below a preset proportion of the amplitude of the in-band signal, thus highlighting the signal within the target frequency band. For example, a bandpass filter is used to process the vibration and acoustic signature sequences from the original monitoring data set. The passband of the bandpass filter is set to the target frequency band, allowing the signal components within the target band to pass while attenuating the signal components outside the band. By adjusting the filter parameters, the amplitude of the out-of-band signal is reduced to below a preset proportion of the amplitude of the in-band signal. This ultimately yields the vibration and acoustic signature sequences within the target frequency band.

[0078] Step S362: Perform a stationarity test on the vibration component sequence and the acoustic component sequence, calculate the autocorrelation coefficient of the sequence, and determine that the sequence is stationary when the autocorrelation coefficient decays to zero with the lag order and meets the characteristics of white noise; otherwise, perform difference processing on the sequence until the sequence is stationary.

[0079] The stationarity test is performed on the vibration component sequence and the acoustic signature component sequence to ensure that the statistical characteristics of the sequence do not change over time. The autocorrelation coefficient of the sequence is calculated, and its change with lag order is observed to determine whether the sequence is stationary. The sequence is considered stationary when the autocorrelation coefficient decays to zero with lag order and meets the characteristics of white noise. If the sequence is not stationary, it is differencing the sequence, i.e., calculating the difference between adjacent data points, and performing multiple differencing operations until the sequence reaches a stationary state.

[0080] For example, the autocorrelation coefficient calculation method is used to calculate the vibration component sequence and the acoustic waveform component sequence. For each sequence, the autocorrelation coefficient at different lag orders is calculated. The change of the autocorrelation coefficient with the lag order is observed. If the autocorrelation coefficient gradually decays to zero and satisfies the characteristics of white noise, the sequence is determined to be stationary. If the sequence is not stationary, the sequence is differencing and stationarity is checked again until the sequence is stationary.

[0081] Step S363: Construct a regression model with the voiceprint component sequence as the dependent variable and the lag terms of the vibration component sequence and the voiceprint component sequence as independent variables. Calculate and determine the lag order of the regression model using the information criterion to minimize the information criterion value of the regression model, while ensuring that the residuals of the regression model satisfy the characteristics of white noise.

[0082] A regression model is constructed using the voiceprint component sequence as the dependent variable and the lagged terms of both the vibration and voiceprint component sequences as independent variables to analyze the causal influence of vibration signals on voiceprint signals. The lag order of the regression model is determined by calculating the information criterion to find the optimal model structure that minimizes the information criterion value while ensuring the residuals satisfy the white noise characteristic. This improves the accuracy and reliability of the regression model. For example, different lag orders are tried, and a regression model is constructed for each lag order. Information criterions, such as the Akaike information criterion or the Bayesian information criterion, are used to calculate the information criterion value for each regression model. The lag order with the smallest information criterion value is selected as the lag order of the regression model. After constructing the regression model, it is checked whether the model residuals satisfy the white noise characteristic. If not, the lag order may need to be adjusted or other methods may be used.

[0083] Step S364: Calculate the significance of the causal influence of the vibration component sequence on the acoustic component sequence by means of a test statistic. The calculation of the statistic takes into account the difference in the sum of squared residuals between the constrained and unconstrained models, as well as parameters such as lag order and sample size.

[0084] The purpose of calculating the causal significance of the vibration component sequence on the voiceprint component sequence using a test statistic is to determine whether the vibration signal has a significant causal effect on the voiceprint signal. The calculation of the statistic considers the difference in the sum of squared residuals between the constrained and unconstrained models, as well as parameters such as lag order and sample size. Integrating these factors allows for a more accurate assessment of the significance of the causal effect. For example, constrained and unconstrained models are constructed. The constrained model restricts the coefficients of certain independent variables to zero in the regression model, while the unconstrained model does not impose such restrictions. The sum of squared residuals of the two models is calculated, and the test statistic is calculated using the test statistic calculation method based on the difference in the sum of squared residuals, lag order, and sample size.

[0085] Step S365: Convert the statistic into a causal influence probability value, calculate the probability value through the distribution function. The probability value represents the significance of the causal influence of the vibration signal on the acoustic signal. When the probability value is greater than a preset threshold, it is determined that there is a significant causal relationship.

[0086] Converting a statistical statistic into a causal influence probability value is to transform the numerical value of the statistic into a probabilistic form, making it easier to determine the significance of the causal influence. The probability value is calculated using a distribution function, and its magnitude is used to determine whether the causal influence of the vibration signal on the voiceprint signal is significant. When the probability value is greater than a preset threshold, a significant causal relationship is determined to exist. For example, a statistical distribution function, such as the F-distribution function, is used to convert the test statistic into a causal influence probability value. The probability value is compared with a preset threshold; if the probability value is greater than the preset threshold, a significant causal relationship is determined to exist between the vibration signal and the voiceprint signal.

[0087] Step S366: Repeat the above process for all sampling points within the target frequency band to generate a causal influence probability sequence that changes over time. Remove isolated noise points in the probability sequence by median filtering. The filter window length is a preset proportion of the number of time sampling points to generate a smooth causal influence probability curve. Extract the phase difference sequence and amplitude ratio sequence corresponding to the peak position of the curve and combine them to generate a causal hysteresis mode.

[0088] The causal relationship analysis process described above is repeated for all sampling points within the target frequency band to generate a causal influence probability sequence that varies over time, demonstrating the change in the causal influence of the vibration signal on the acoustic signature signal. Median filtering removes isolated noise points from the probability sequence, with the filtering window length set to a preset proportion of the number of time sampling points, smoothing the probability sequence and generating a smooth causal influence probability curve. The phase difference sequence and amplitude ratio sequence corresponding to the peak positions of the curve are extracted and combined to generate a causal hysteresis mode, which contains the causal hysteresis characteristics of the vibration signal and acoustic signature signal within the target frequency band. For example, for each sampling point within the target frequency band, the causal relationship analysis process of steps S361-S365 is repeated to obtain the causal influence probability value for each sampling point. These probability values ​​are arranged in chronological order to generate a causal influence probability sequence. Then, a median filtering algorithm is used to filter the probability sequence, with the filtering window length set to a preset proportion of the number of time sampling points. Filtering removes isolated noise points from the probability sequence, generating a smooth causal influence probability curve. Find the peak positions in the curve, extract the phase difference sequence and amplitude ratio sequence corresponding to these peak positions, and combine them to generate causal hysteresis modes.

[0089] Step S400: Map the causal hysteresis mode to a preset physical force transmission path network to generate a latent damage topology map describing the weakening characteristics of structural performance. The nodes of the latent damage topology map correspond to the key force-bearing units in the force transmission path, and the edges correspond to the force flow transmission intensity between units.

[0090] Mapping causal hysteresis modes to a pre-defined physical force transmission path network is intended to link the causal hysteresis characteristics of vibration and acoustic signature signals with the physical force transmission structure of bridge expansion joints. The pre-defined physical force transmission path network describes the force transmission paths and relationships between load-bearing elements within the expansion joint. Through mapping, information from the causal hysteresis modes can be converted into features of nodes and edges in the physical force transmission path network, generating a latent damage topology map describing the weakening characteristics of structural performance. The nodes in the latent damage topology map correspond to key load-bearing elements in the force transmission path, and the edges correspond to the force flow transmission intensity between elements, visually demonstrating the latent damage status of the bridge expansion joint.

[0091] As one implementation method, step S400 may specifically include the following steps S410~S460: Step S410: Read the preset physical force transmission path network data. The nodes of the physical force transmission path network correspond to the force key points of the expansion joint key components, the edges correspond to the force transmission paths between components, and the edge attributes include the initial force flow transmission coefficient.

[0092] Reading the preset physical force transmission path network data is to obtain the physical force transmission structure information of bridge expansion joints. The nodes of the physical force transmission path network correspond to the key stress points of critical components of the expansion joint, clearly indicating the location of force application; edges correspond to the force transmission paths between components, showing the method of force transmission between components; edge attributes include the initial force flow transmission coefficient, reflecting the initial force flow transmission capacity between components. In practical applications, this data is stored in a designated data file or database. Specialized data reading tools or programs can be used to read the physical force transmission path network data into memory according to the data's storage format and structure for subsequent processing and analysis. For example, if the data is stored as a text file, the program can read the file content line by line, parsing out node information, edge information, and the initial force flow transmission coefficients of the edges, etc.

[0093] Step S420: Assign the phase difference sequence and amplitude ratio sequence in the causal hysteresis mode to the network nodes at the corresponding spatial locations. The mapping is based on the spatial distance between the sensor array installation location and the network node. The causal hysteresis mode of the nearest sensor unit is assigned to the corresponding node. When multiple sensor units correspond to the same node, a weighted average is used to fuse the mode parameters, with the weight being the reciprocal of the distance.

[0094] In this step, the phase difference sequence and amplitude ratio sequence in the causal hysteresis mode need to be accurately assigned to the corresponding nodes in the physical force transmission path network. First, the installation locations of the sensor array and the spatial locations of the network nodes must be clearly defined. By calculating the spatial distance between the sensor array installation location and the network nodes, the causal hysteresis modes of the sensor units are assigned to the corresponding network nodes based on the principle of proximity. When multiple sensor units correspond to the same node, a weighted average method is used to fuse the modal parameters for more reasonable integration of these modal information. The closer the sensor unit is, the greater the weight of its causal hysteresis mode in the fusion; the weight is set to the reciprocal of the distance. This allows the sensor unit closer to the node to have a greater impact on the node, more accurately reflecting the actual situation of the node. For example, each network node can be traversed, and for each node, its spatial distance to the installation locations of all sensor units can be calculated. The closest sensor unit is then identified, and its phase difference sequence and amplitude ratio sequence in the causal hysteresis mode are assigned to that node. If multiple sensing units are close to the node, their distances to the node are calculated separately, and their corresponding weights are obtained. Then, based on these weights, a weighted average is calculated on the phase difference sequence and amplitude ratio sequence of the multiple sensing units, and the fused result is assigned to the node.

[0095] Step S430: Calculate the damage characteristic value of each network node. The damage characteristic value is the Euclidean distance between the current causal hysteresis mode and the healthy baseline mode. The healthy baseline mode is obtained by collecting monitoring data when the bridge expansion joint is newly built and stored as a phase difference baseline sequence and an amplitude ratio baseline sequence. The larger the distance value, the more severe the node performance degradation.

[0096] Calculating the damage characteristic values ​​of network nodes is a crucial step in assessing the damage extent of key load-bearing units in bridge expansion joints. The healthy baseline mode, obtained through monitoring data collection during the construction of the bridge expansion joint, represents the characteristics of the nodes in a healthy state and is stored as a phase difference baseline sequence and an amplitude ratio baseline sequence. The current causal hysteresis mode is a characteristic sequence reflecting the current node state obtained in the previous steps. The damage characteristic values ​​are determined by calculating the Euclidean distance between the current causal hysteresis mode and the healthy baseline mode. The Euclidean distance measures the degree of difference between the two sequences; a larger distance indicates a greater difference between the current node's state and its healthy state, signifying more severe node performance degradation.

[0097] During the calculation, for each network node, the phase difference sequence and amplitude ratio sequence in its current causal hysteresis mode are compared with the reference phase difference sequence and amplitude ratio sequence in the healthy baseline mode, respectively. The distance between the two sequences is calculated using the Euclidean distance method, yielding the node's damage characteristic value. A specialized algorithm can be used to calculate the Euclidean distance, ensuring the accuracy of the results.

[0098] Step S440: Adjust the force flow transmission coefficient of the corresponding edge in the physical force transmission path network based on the node damage feature value. The adjustment method is to attenuate the initial coefficient through the damage feature value. The larger the damage feature value, the greater the attenuation of the weight of the corresponding edge, so as to ensure the overall force flow of the network is conserved after the coefficient adjustment.

[0099] Adjusting the force flow transmission coefficients of edges in a physical force transmission path network based on node damage characteristic values ​​is to reflect the impact of node damage on force flow transmission. Damage to a node leads to changes in the force flow transmission capacity of its surroundings; therefore, the initial force flow transmission coefficients of edges need to be adjusted according to the node's damage characteristic values. The adjustment method involves attenuating the initial coefficients using the damage characteristic values. A larger damage characteristic value indicates more severe node damage, resulting in a greater decrease in the force flow transmission capacity of the corresponding edge, and thus a greater degree of attenuation in the edge weight. Simultaneously, during the coefficient adjustment process, it is crucial to ensure the overall force flow conservation of the network; that is, the sum of the incoming edge force flow of all nodes in the network equals the sum of the outgoing edge force flow, to guarantee the rationality and physical reality of force flow transmission.

[0100] In one implementation, step S440 may specifically include the following steps S441 to S446: Step S441: Read the health baseline mode data from the hidden damage database. The health baseline mode includes the phase difference baseline sequence and amplitude ratio baseline sequence of each node of the bridge expansion joint in the newly built state. The sampling frequency of the sequence is consistent with the original monitoring data set, and the time length is multiple complete traffic load cycles.

[0101] The latent damage database is an important resource for storing health data related to bridge expansion joints. Health baseline modal data are read from this database, which includes the phase difference baseline sequence and amplitude ratio baseline sequence for each node of the bridge expansion joint in its newly constructed state. To ensure data consistency and comparability, the sampling frequency of the sequences is consistent with the original monitoring data set, and the time length is multiple complete traffic load cycles. This allows for a comprehensive and accurate reflection of the characteristics of bridge expansion joints in their healthy state.

[0102] Step S442: Calculate the difference between the current causal lag mode and the healthy baseline mode. The difference is achieved through dynamic time warping distance. Calculate the distance value for the phase difference sequence and amplitude ratio sequence of each node, and then obtain the comprehensive difference by weighted summation. The weights reflect the sensitivity of phase changes to damage.

[0103] Calculating the dissimilarity between the current causal lag mode and the healthy baseline mode is a crucial step in further assessing the degree of node damage. Dynamic time-warped distance (DTDM) is used to calculate the dissimilarity, a method that can handle differences in sequence length and time alignment, and more accurately measures the similarity between two sequences. For each node, DTDM is calculated separately for its phase difference sequence and amplitude ratio sequence, yielding two distance values. Then, considering the different sensitivities of phase changes to damage, different weights are assigned to these two distance values, and a weighted summation is used to obtain the comprehensive dissimilarity.

[0104] When calculating the dynamic time warping distance, the dynamic time warping algorithm can be used to find the optimal matching path between two sequences and calculate the distance based on the elements on the path. The weight settings need to be determined based on the actual situation and experience to accurately reflect the importance of phase changes and amplitude ratio changes to damage assessment.

[0105] Step S443: Determine the maximum damage threshold. Obtain the limit damage state parameters of the key components of the expansion joint through material fatigue test data. When the cumulative damage of the component reaches the material limit value, record the phase difference sequence and amplitude ratio sequence at this time, and calculate the difference between them and the healthy reference mode as the maximum damage threshold.

[0106] Determining the maximum damage threshold is to set an upper limit for damage, enabling a reasonable assessment and judgment of the damage level at the nodes. Ultimate damage state parameters of key components in the expansion joint are obtained using material fatigue test data. During the test, continuously varying loads are applied to the components to simulate actual usage conditions. When the cumulative damage of the component reaches the material's limit value, the phase difference sequence and amplitude ratio sequence are recorded. These sequences are compared with the healthy baseline mode, and the degree of difference between them is calculated. This degree of difference is used as the maximum damage threshold.

[0107] When conducting material fatigue tests, it is necessary to strictly control the test conditions and parameters to ensure the accuracy and reliability of the test results.

[0108] Step S444: For each node in the network, calculate the damage feature value based on the ratio of the current difference degree to the maximum damage threshold. The damage feature value is the ratio of the difference degree to the maximum damage threshold. When the difference degree exceeds the maximum damage threshold, the damage feature value is truncated to the ratio corresponding to the maximum threshold.

[0109] For each node in the network, a damage characteristic value is calculated based on the ratio of the currently calculated dissimilarity to the maximum damage threshold. This ratio calculation method quantifies the degree of damage to the node, facilitating comparison and analysis. When the dissimilarity of a node exceeds the maximum damage threshold, it indicates that the node's damage has reached or exceeded the limit state. To avoid the damage characteristic value being unreasonably too large, the damage characteristic value is truncated to the ratio corresponding to the maximum threshold.

[0110] During the calculation, each node in the network is traversed, and its current dissimilarity is divided by the maximum damage threshold to obtain the damage feature value. If the calculated damage feature value exceeds the ratio corresponding to the maximum threshold, it is adjusted to that ratio.

[0111] Step S445: Identify all edges directly connected to the node, including outgoing edges starting from the node and incoming edges ending at the node. Adjust the coefficients of both outgoing and incoming edges using the node's damage feature value. The adjusted coefficients are the product of the initial coefficients and (1 - damage feature value).

[0112] Identifying all edges directly connected to nodes is fundamental to adjusting the force flow transmission coefficients. For each node, it's necessary to determine the outgoing edges originating from it and the incoming edges ending at it. The force flow transmission coefficients of these edges are affected by the node's damage characteristic value. Both outgoing and incoming edges are adjusted using the node's damage characteristic value by multiplying the initial coefficient by (1 - damage characteristic value). The larger the damage characteristic value, the smaller the value of (1 - damage characteristic value), and the smaller the adjusted coefficient, reflecting the weakening effect of node damage on the edge's force flow transmission capacity. For example, each node in the network is traversed, and based on the node's connection relationships with edges, all edges directly connected to that node are identified. For each edge, its initial force flow transmission coefficient is obtained, and then adjusted according to the node's damage characteristic value to obtain the adjusted coefficient.

[0113] Step S446: Perform force flow conservation verification on the adjusted force flow transmission coefficients, calculate the difference between the sum of incoming force flow and the sum of outgoing force flow of each node in the network, and when the absolute value of the difference exceeds the preset ratio of the initial sum of force flow, perform a secondary adjustment on the coefficients of the surrounding edges of the node, and optimize iteratively to make the difference less than the preset ratio.

[0114] Verifying the conservation of force flow in the adjusted force flow transmission coefficients is a crucial step in ensuring the rationality of force flow transmission. The difference between the sum of incoming and outgoing force flows at each node in the network is calculated. If the absolute value of this difference exceeds a preset proportion of the initial total force flow, it indicates that force flow is not conserved, requiring a secondary adjustment of the coefficients of the edges surrounding the node. Through iterative optimization, the edge coefficients are continuously adjusted so that the difference between the sum of incoming and outgoing force flows gradually decreases until it falls below a preset proportion, thus ensuring the overall conservation of force flow in the network.

[0115] When performing force flow conservation verification, each node in the network is traversed, and the sum of its incoming and outgoing force flows is calculated to obtain the difference. The absolute value of the difference is compared with a preset ratio of the initial force flow sum. If it exceeds this ratio, an appropriate algorithm is used to adjust the coefficients of the edges surrounding the node. The edge coefficients can be continuously optimized through multiple iterations until the force flow conservation requirement is met.

[0116] Step S450: Reconstruct the topology of the adjusted physical force transmission path network, calculate the adjacency matrix and Laplace matrix of the network using graph theory algorithms, extract the eigenvalues ​​and eigenvectors of the matrix, and generate topology graph data that reflects the force flow distribution characteristics of the network.

[0117] Reconstructing the topology of the adjusted physical force transmission path network is a crucial step in deeply analyzing the force flow distribution characteristics of the network. Graph theory algorithms are used to calculate the adjacency matrix and Laplace matrix of the network. The adjacency matrix represents the connections between nodes in the network, while the Laplace matrix reflects the network's topology and force flow transmission characteristics. The eigenvalues ​​and eigenvectors of these two matrices are extracted; these eigenvalues ​​and eigenvectors contain important topological information of the network and can be used to generate topological graph data reflecting the force flow distribution characteristics of the network.

[0118] When calculating the adjacency matrix, the presence or absence of connections between nodes in the network is represented by matrix elements. Calculating the Laplacian matrix requires combining the adjacency matrix and the degree information of the nodes. Specialized graph theory algorithms are used to extract the eigenvalues ​​and eigenvectors of the matrix, and this information is then organized into a topological graph.

[0119] Step S460: Associate the node size in the topology graph data with the damage feature value, and associate the edge thickness with the adjusted force flow transmission coefficient to generate a latent damage topology graph that describes the weakening characteristics of structural performance. The node coordinates of the latent damage topology graph are consistent with the spatial location of the physical force transmission path network.

[0120] Associating node sizes with damage feature values ​​and edge thicknesses with adjusted force transmission coefficients in the topology graph data is to visually represent the network's topological and damage information. Node size reflects the degree of damage; a larger damage feature value corresponds to a larger node, indicating more severe performance degradation. Edge thickness reflects the magnitude of the force transmission coefficient; a larger coefficient corresponds to a thicker edge, indicating stronger force transmission capability. This association method generates a latent damage topology graph describing the weakening characteristics of structural performance. To accurately reflect the actual spatial layout of the network, the node coordinates of the latent damage topology graph are consistent with the spatial locations of the physical force transmission path network.

[0121] When generating a latent damage topology map, graphical drawing tools or visualization software can be used. Based on the node damage characteristic values ​​and edge force transmission coefficients in the topology map data, the size of the nodes and the thickness of the edges are set. Simultaneously, the nodes are placed on coordinates consistent with their spatial locations in the physical force transmission path network, ultimately generating an intuitive latent damage topology map.

[0122] Step S500: Based on the connectivity characteristics of nodes and the strength characteristics of edges in the latent damage topology graph, quantitatively evaluate the overall performance level and local damage degree of the bridge expansion joint, and generate monitoring results containing status level identifiers.

[0123] Quantitative assessment based on the connectivity characteristics of nodes and the strength characteristics of edges in a latent damage topology graph allows for a comprehensive and accurate understanding of the structural performance of bridge expansion joints. Node connectivity reflects the tightness of connections between nodes and the overall network connectivity, while edge strength reflects the capacity for force transmission and stability. Quantitative analysis of these characteristics determines the overall performance level of the bridge expansion joint, i.e., the health status and performance of the entire expansion joint structure. Simultaneously, it allows for the assessment of local damage levels, identifying areas with more severe damage. Finally, monitoring results containing status level indicators are generated, providing a clear basis for bridge maintenance and management.

[0124] As one implementation method, step S500 may specifically include the following steps S510~S560: Step S510: Extract node connectivity subgraphs from the latent damage topology graph, identify all mutually independent connected subgraphs in the graph, each connected subgraph contains a set of nodes directly or indirectly connected by edges, generate a connected subgraph set, and each element of the connected subgraph set contains the identifiers of all nodes in the subgraph and the connection relationships between the nodes.

[0125] Extracting node connectivity subgraphs from a latently damaged topology graph is fundamental to analyzing its connectivity structure, identifying all independent connected subgraphs. Nodes in each connected subgraph are directly or indirectly connected by edges, forming relatively independent connected regions. A set of connected subgraphs is generated, where each element contains identifiers of all nodes within the subgraph and the connections between them, clearly demonstrating the connectivity structure of the latently damaged topology graph.

[0126] When extracting connected subgraphs, graph traversal algorithms such as depth-first search or breadth-first search can be used. Starting from a node in the graph, traverse all nodes connected to it, forming a connected subgraph. Then, select a new starting node from unvisited nodes and repeat the process until all nodes in the graph have been visited. Organize all the obtained connected subgraphs into a connected subgraph set.

[0127] Step S520: Calculate the average degree and diameter of each node in each connected subgraph. The average degree is the average number of connected edges of all nodes in the subgraph, and the diameter is the maximum length of the shortest path between any two nodes in the subgraph. Generate the subgraph structure feature sequence.

[0128] Calculating the average degree and diameter of nodes in each connected subgraph is a crucial step in quantifying the structural characteristics of a connected subgraph. The average degree reflects the tightness of connections between nodes within the subgraph; a higher average degree indicates tighter connections. The diameter reflects the size and connectivity of the subgraph; a larger diameter indicates a wider subgraph. By calculating these two metrics, a sequence of subgraph structural characteristics is generated. This sequence contains information on the average degree and diameter of nodes in each connected subgraph, facilitating the comparison and analysis of the structural characteristics of different connected subgraphs.

[0129] When calculating the average degree of nodes, for each connected subgraph, the number of connecting edges between all nodes in the subgraph is counted, and then the average value is calculated. When calculating the diameter, the shortest path algorithm is used to calculate the shortest path length between any two nodes in the subgraph, and the longest length is determined as the diameter. The average degree of nodes and the diameter of each connected subgraph are recorded in the subgraph structure feature sequence.

[0130] Step S530: Extract the force flow transmission intensity values ​​of all edges in the hidden damage topology graph, calculate the distribution entropy and interquartile range of the intensity values. The distribution entropy reflects the uniformity of the intensity value distribution, and the interquartile range reflects the dispersion of the intensity values. Generate edge intensity distribution characteristics.

[0131] Extracting the force transmission intensity values ​​of all edges in the latent damage topology graph is a prerequisite for analyzing the edge strength distribution characteristics. By calculating the distribution entropy and interquartile range of these intensity values, the distribution of edge strength can be described from different perspectives. The distribution entropy reflects the uniformity of the intensity value distribution; the smaller the entropy value, the more uniform the intensity value distribution. The interquartile range reflects the dispersion of the intensity values; the smaller the interquartile range, the more concentrated the intensity values. Generating edge strength distribution features, which include distribution entropy and interquartile range information, can comprehensively describe the distribution characteristics of edge strength, providing a basis for evaluating the force transmission stability of bridge expansion joints.

[0132] When extracting the force flow transmission intensity value of the edges, all edges in the latent damage topology are traversed, the intensity value of each edge is recorded, and the distribution entropy and interquartile range of these intensity values ​​are calculated. The calculation results are then organized into edge intensity distribution characteristics.

[0133] Step S540: Identify abnormal edges in the edge strength distribution characteristics. Abnormal edges are edges whose strength values ​​are lower than a preset ratio of the corresponding edge strength benchmark value under healthy conditions. Generate a set of abnormal edges, which includes the identifier of the abnormal edges and their stress flow transmission strength values.

[0134] Identifying anomalous edges in the edge strength distribution characteristics is crucial for determining areas of abnormal force flow transmission. Anomalous edges are those whose strength values ​​are lower than a preset proportion of the corresponding edge strength benchmark value under healthy conditions. These edges may be damaged or have degraded performance. By identifying anomalous edges, regions with unstable force flow transmission in bridge expansion joints can be detected in a timely manner, generating a set of anomalous edges. This set includes the identifiers of the anomalous edges and their corresponding stress flow transmission strength values.

[0135] When identifying abnormal edges, the force transmission intensity value of each edge is compared with the corresponding edge intensity benchmark value under healthy conditions. If the intensity value is lower than a preset ratio, the edge is marked as an abnormal edge, its identifier and intensity value are recorded, and it is added to the abnormal edge set.

[0136] Step S550: Integrate the subgraph structural feature sequence, edge strength distribution features, and abnormal edge set to generate an overall performance evaluation index. The index comprehensively reflects the structural integrity, edge strength distribution uniformity, and abnormal edge ratio of the connected subgraph.

[0137] Integrating the subgraph structural feature sequence, edge strength distribution characteristics, and anomalous edge set is the core step in generating an overall performance evaluation index. By comprehensively considering factors such as the structural integrity of the connected subgraph, the uniformity of edge strength distribution, and the proportion of anomalous edges, the overall performance level of bridge expansion joints can be comprehensively and accurately assessed. Integrating this information generates a comprehensive evaluation index that can intuitively reflect the overall health and performance of bridge expansion joints.

[0138] As one implementation method, step S550 may specifically include the following steps S551~S556: Step S551: Standardize the average degree of nodes in the subgraph structure feature sequence, convert it to a numerical range, and generate a standardized average degree of nodes. The higher the standardized value, the closer the subgraph connection density is to a healthy state.

[0139] Standardizing the average node degree in the subgraph structural feature sequence aims to eliminate dimensional differences between the average node degrees of different subgraphs. This is achieved by converting the average node degree to a predetermined numerical range, generating a standardized average node degree. A higher standardized value indicates that the subgraph's connection density is closer to a healthy state, meaning the connections between nodes are tighter and the structure is more stable.

[0140] When performing standardization, a standardization algorithm, such as linear standardization, can be used. Based on the range of average node degrees in the subgraph's structural feature sequence, the average degree value of each node is mapped to a preset numerical range. The processed result is then used as the standardized node average degree.

[0141] Step S552: Standardize the distribution entropy in the edge strength distribution characteristics, convert it to the same numerical range, and generate standardized distribution entropy. The lower the standardized value, the more uniform the edge strength distribution.

[0142] Standardizing the distribution entropy in the edge strength distribution characteristics is to make the distribution entropy of different graphs comparable. The distribution entropy is transformed to a value range similar to the average degree of the nodes, generating a standardized distribution entropy. The lower the standardized value, the more uniform the edge strength distribution and the more stable the force flow transmission.

[0143] The distribution entropy is processed using a standardization algorithm, mapping it to a preset numerical range based on its value range. The processed result is then used as the standardized distribution entropy.

[0144] Step S553: ​​Calculate the ratio of the number of abnormal edges in the abnormal edge set to the total number of edges in the topology graph to obtain the abnormal edge ratio. Standardize the abnormal edge ratio to convert it to the same numerical range to generate a standardized abnormal edge ratio. The lower the standardized value, the smaller the proportion of abnormal edges.

[0145] The ratio of the number of abnormal edges in the abnormal edge set to the total number of edges in the topology graph is calculated to obtain the abnormal edge percentage, which reflects the proportion of abnormal edges in the entire topology graph. The abnormal edge percentage is then standardized to a value range similar to the average degree and distribution entropy of the nodes, generating a standardized abnormal edge percentage. The lower the standardized value, the smaller the proportion of abnormal edges, and the more normal the force flow transmission in the bridge expansion joints.

[0146] When calculating the percentage of outliers, the number of outliers in the outlier edge set is counted, and the total number of edges in the topology graph is compared. The ratio between the two is then calculated. A standardization algorithm is used to map the percentage of outliers to a preset numerical range, resulting in a standardized percentage of outliers.

[0147] Step S554: The standardized node average degree, standardized distribution entropy, and standardized outlier edge ratio are weighted and summed. The weights are determined based on the degree of influence of each index on the overall structural performance, generating intermediate evaluation indexes. The index values ​​comprehensively reflect the subgraph structure, edge strength distribution, and the impact of outlier edges.

[0148] The weighted summation of standardized node average degree, standardized distribution entropy, and standardized outlier edge proportion is used to comprehensively consider the impact of these three indicators on the overall structural performance of the bridge expansion joint. The weights are determined based on the degree of influence of each indicator on the overall structural performance, and the weight settings need to be determined in conjunction with actual conditions and experience. Through weighted summation, an intermediate evaluation index is generated, whose value can comprehensively reflect the impact of subgraph structure, edge strength distribution, and outlier edges on the overall performance of the bridge expansion joint. In the weighted summation, the standardized node average degree, standardized distribution entropy, and standardized outlier edge proportion are multiplied by their respective weights, and then the results are added together to obtain the intermediate evaluation index.

[0149] Step S555: Perform spatiotemporal stability verification on the intermediate evaluation indicators, compare the changes in the intermediate evaluation indicators in different time windows, and if the changes exceed the preset threshold, readjust the weights of each indicator and recalculate to generate stable intermediate evaluation indicators.

[0150] Verifying the spatiotemporal stability of intermediate evaluation indicators is a crucial step in ensuring the reliability of evaluation results. By comparing the changes in intermediate evaluation indicators across different time windows, if the changes exceed a preset threshold, it indicates instability in the evaluation results, potentially due to improper weighting of the indicators. In this case, it is necessary to readjust the weights of each indicator and recalculate the intermediate evaluation indicators until the changes in the intermediate evaluation indicators are less than the preset threshold, thus generating stable intermediate evaluation indicators.

[0151] During spatiotemporal stability verification, intermediate evaluation indicators across different time windows are compared, and their changes are calculated. These changes are then compared to preset thresholds. If the thresholds are exceeded, appropriate algorithms are used to readjust the weights of each indicator, and the intermediate evaluation indicators are recalculated until the stability requirements are met.

[0152] Step S556: Map stable intermediate evaluation indicators to preset state level classification standards to generate overall performance evaluation indicators, which include state level identifiers and corresponding quantitative scores.

[0153] Mapping stable intermediate evaluation indicators to preset state level classification standards is a key step in visually presenting the evaluation results. These preset state level classification standards are developed based on the actual conditions and experience of bridge expansion joints. By comparing stable intermediate evaluation indicators with these standards, their corresponding state levels are determined. Simultaneously, corresponding quantitative scores are generated based on the state levels, ultimately producing an overall performance evaluation indicator. This indicator includes a state level identifier and a quantitative score, clearly demonstrating the overall performance level of the bridge expansion joint.

[0154] During the mapping process, stable intermediate evaluation indicators are compared one by one with preset state level classification standards to determine their respective state levels. Based on the scoring rules corresponding to each state level, a quantitative score is generated. The state level identifier and the quantitative score are then combined to form an overall performance evaluation index.

[0155] Step S560: For each node, combine its performance weakening index, the average degree of nodes in the connected subgraph and the number of associated abnormal edges to determine the level of local damage, generate local damage distribution information containing node identifier, damage level and spatial coordinates, integrate the overall performance evaluation index and local damage distribution information to generate monitoring results containing state level identifiers.

[0156] For each node, determining the level of local damage by combining its performance weakening index, the average degree of nodes in its connected subgraph, and the number of associated anomalous edges is a crucial step in assessing the local damage status of bridge expansion joints. The performance weakening index reflects the degree of damage to the node itself, the average degree of nodes in its connected subgraph reflects the connectivity environment in which the node is located, and the number of associated anomalous edges indicates anomalies in force flow transmission around the node. By comprehensively considering these three factors, the level of local damage for each node can be accurately determined.

[0157] Local damage distribution information, including node identifiers, damage levels, and spatial coordinates, is generated, providing a clear visual representation of the damage status and spatial location of each node. Finally, the overall performance evaluation indicators and local damage distribution information are integrated to generate monitoring results containing status level identifiers. These results comprehensively reflect the overall and local health status of the bridge expansion joints.

[0158] As one implementation, step S560 may include the following steps S561~S566: Step S561: Extract the performance weakening index of each node from the node performance index sequence, extract the average degree of the nodes in the connected subgraph of the corresponding node from the subgraph structure feature sequence, count the number of abnormal edges connected to the node from the abnormal edge set, and generate a node local feature triplet, which includes the performance weakening index, the average degree of the node, and the number of associated abnormal edges.

[0159] The performance weakening index of each node is extracted from the node performance index sequence. This index reflects the degree of damage to the node itself. The average degree of the nodes in the connected subgraph to which the corresponding node belongs is extracted from the subgraph structural feature sequence. The average degree of the node reflects the connectivity environment in which the node is located; the higher the average degree, the better the connectivity of the node. The number of anomalous edges connected to the node is counted from the set of anomalous edges. The number of anomalous edges reflects the anomalous situation of force flow transmission around the node. These three pieces of information are combined into a node local feature triplet, which can comprehensively describe the local features of the node.

[0160] When extracting information, the corresponding information is found from the relevant sequences and sets based on the node's identifier. This information is then organized into triplet sets of local node features.

[0161] Step S562: Standardize the performance weakening index in the local feature triplet of the node, convert it to a numerical range, and generate a standardized weakening index.

[0162] Standardizing the performance degradation indices in the local feature triples of nodes aims to eliminate dimensional differences between these indices, making them comparable. Standardized indices are generated by transforming the performance degradation indices to a predetermined numerical range. A standardization algorithm is then used to process the indices, mapping each index value to a preset range based on its value. The processed result is then used as the standardized performance degradation index.

[0163] Step S563: Standardize the number of associated abnormal edges in the local feature triplet of the node, convert it to the same numerical range, and generate a standardized number of abnormal edges.

[0164] Standardizing the number of associated outlier edges in the local feature triples of nodes is to make the number of associated outlier edges comparable across different nodes. The number of associated outlier edges is converted to the same numerical range as the performance weakening metric to generate a standardized number of outlier edges.

[0165] The number of associated outlier edges is processed using a standardization algorithm. Based on the range of values, each value is mapped to a preset numerical range. The processed result is used as the standardized number of outlier edges.

[0166] Step S564: The standardized weakening index, the number of standardized abnormal edges, and the average degree of the nodes in the connected subgraph are weighted and fused. The weights are determined based on the degree of influence of each feature on local damage, and a node local damage score is generated. The higher the score, the more severe the local damage.

[0167] The weighted fusion of standardized weakening indicators, the number of standardized outlier edges, and the average degree of nodes in the connected subgraph is intended to comprehensively consider the impact of these three features on local node damage. The weights are determined based on the degree of influence of each feature on local damage; the weight settings need to be determined in conjunction with practical situations and experience. Through weighted fusion, a local node damage score is generated; the higher the score, the more severe the local damage to the node.

[0168] When performing weighted fusion, the standardized weakening index, the number of standardized abnormal edges, and the average degree of the nodes in the connected subgraph are multiplied by their respective weights, and then the results are added together to obtain the node local damage score.

[0169] Step S565: Compare the local damage score of the node with the preset damage level threshold to determine the local damage level of the node. The damage level is divided into multiple levels, and each level corresponds to a scoring range.

[0170] Comparing the local damage score of a node with preset damage level thresholds is a crucial step in determining the degree of local damage to the node. These preset damage level thresholds are established based on the actual conditions and experience of bridge expansion joints. The local damage score of a node is compared with these thresholds to determine its corresponding damage level. Damage levels are divided into multiple tiers, each corresponding to a scoring range, thus clearly defining the degree of damage to the node. During the comparison, the local damage score of the node is compared with the scoring range of each damage level to determine its corresponding level. This level is then used as the degree of local damage to the node.

[0171] Step S566: Summarize the local damage level and spatial coordinates of all nodes to generate local damage distribution information, which includes node identifier, damage level, spatial coordinates and corresponding score value.

[0172] Summarizing the local damage severity levels and spatial coordinates of all nodes is to comprehensively display the local damage distribution of bridge expansion joints. The identifier, damage level, spatial coordinates, and corresponding score of each node are compiled to generate local damage distribution information. This information can intuitively display the damage status and spatial location of each node, providing detailed reference for bridge maintenance and management.

[0173] When summarizing information, all nodes are traversed, and the relevant information of each node is recorded and organized into local damage distribution information.

[0174] Finally, the overall performance evaluation indicators and local damage distribution information are integrated to generate monitoring results that include status level identifiers. These monitoring results comprehensively reflect the overall and local health status of the bridge expansion joints, providing detailed and accurate data for bridge maintenance and management. Analysis of the monitoring results allows for the timely identification of problems with the bridge expansion joints, enabling appropriate maintenance measures to be taken and ensuring the safe operation of the bridge.

[0175] It is understood that the various algorithms involved in the above descriptions of the embodiments of the present invention can be obtained from relevant content in the prior art. To save space, they will not be elaborated on in the embodiments of the present invention. In addition, those skilled in the art can supplement the details based on common knowledge in the art when implementing the solutions of the present invention. For example, they can use normalization to eliminate dimensional conflicts before feature fusion, use interpolation to eliminate dimensional differences, reasonably set thresholds based on historical data, experience or business scenario requirements, train the model based on a general model training method, set the number of layers in the model structure based on actual needs, select activation functions, etc. The present invention will not provide redundant descriptions of overly detailed implementation processes here.

[0176] Figure 2 This is a schematic diagram of the structural composition of a bridge expansion joint condition monitoring device provided in an embodiment of the present invention, as shown below. Figure 2 As shown, the bridge expansion joint status monitoring device 200 includes: a data acquisition module 210, used to synchronously acquire vibration signal sequences and acoustic signature signal sequences of the bridge expansion joint under operational loads through a distributed sensor array, generating a raw monitoring data set with timestamp alignment, the raw monitoring data set containing vibration time history curves and acoustic signature spectral characteristic curves at different acquisition locations; a data association module 220, used to perform spatiotemporal association modeling on the vibration signal sequences and acoustic signature signal sequences in the raw monitoring data set, generating a time-series co-evolution field describing the dynamic coupling relationship between vibration and acoustic signature signals, the dimension of the time-series co-evolution field being correlated with the sampling frequency and spatial distribution density of the raw monitoring data set; and a data deconstruction module 230, used to analyze the time-series co-evolution... The field is subjected to causal sequential difference deconstruction to separate the causal hysteresis modes of the vibration signal sequence relative to the acoustic signal sequence in the target frequency band. The causal hysteresis modes include the phase difference change trend and amplitude coupling coefficient between the vibration signal and the acoustic signal. The damage characterization module 240 is used to map the causal hysteresis modes to a preset physical force transmission path network to generate a hidden damage topology map describing the weakening characteristics of structural performance. The nodes of the hidden damage topology map correspond to the key force-bearing units in the force transmission path, and the edges correspond to the force flow transmission intensity between units. The state recognition module 250 is used to quantitatively evaluate the overall performance level and local damage degree of the bridge expansion joint based on the connectivity characteristics of the nodes and the strength characteristics of the edges in the hidden damage topology map, and generate monitoring results containing state level identifiers.

[0177] The descriptions of the apparatus embodiments above are similar to those of the method embodiments above, and have similar beneficial effects. In some embodiments, the functions or modules included in the apparatus provided by the present invention can be used to perform the methods described in the method embodiments above. For technical details not disclosed in the apparatus embodiments of the present invention, please refer to the descriptions of the method embodiments of the present invention for understanding.

[0178] Figure 3 A hardware entity diagram of a computer system provided as an embodiment of the present invention, such as... Figure 3 As shown, the hardware entity of the computer system 1000 includes a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program that can run on the processor 1001, and the processor 1001 executes the program to implement the steps in the method of any of the above embodiments.

Claims

1. A bridge expansion joint state monitoring method based on vibration voiceprint collaborative perception, characterized in that, The method comprises: Synchronously collecting vibration signal sequences and acoustic fingerprint signal sequences of the bridge expansion joint under operating load through the distributed sensing array, generating a raw monitoring data set with timestamp alignment, and the raw monitoring data set containing vibration time history curves and acoustic fingerprint spectrum feature curves at different collection positions; Performing spatiotemporal correlation modeling on the vibration signal sequences and acoustic fingerprint signal sequences in the raw monitoring data set, generating a time sequence collaborative evolution field describing the dynamic coupling relationship between vibration and acoustic fingerprint signals, and the dimension of the time sequence collaborative evolution field being associated with the sampling frequency and spatial distribution density of the raw monitoring data set; Performing causal sequential difference decomposition on the time sequence collaborative evolution field, separating out a causal lag mode of the vibration signal sequences relative to the acoustic fingerprint signal sequences in a target frequency band range, and the causal lag mode containing a phase difference variation trend and an amplitude coupling coefficient between the vibration signal and the acoustic fingerprint signal; Mapping the causal lag mode to a preset physical force transmission path network, generating an implicit damage topology graph describing structural performance weakening characteristics, and the nodes of the implicit damage topology graph corresponding to key stress units in the force transmission path and the edges corresponding to the force flow transmission strength between the units; Based on the connectivity features of the nodes and the strength features of the edges in the implicit damage topology graph, quantitatively evaluating the overall state level and local damage degree of the bridge expansion joint, and generating a monitoring result containing a state level identifier.

2. The method of claim 1, wherein, The spatiotemporal correlation modeling on the vibration signal sequences and acoustic fingerprint signal sequences in the raw monitoring data set, generating a time sequence collaborative evolution field describing the dynamic coupling relationship between vibration and acoustic fingerprint signals, comprises: Performing timestamp interpolation processing on the vibration time history curves and acoustic fingerprint spectrum feature curves in the raw monitoring data set, correcting the sampling delay of different sensing units, generating vibration resampling sequences and acoustic fingerprint resampling sequences with equal time intervals, and the time intervals of the vibration resampling sequences and acoustic fingerprint resampling sequences being consistent with the highest sampling frequency of the raw monitoring data set; Performing numerical product operation on the vibration resampling sequences and acoustic fingerprint resampling sequences, generating coupling feature values under the same spatiotemporal coordinates, generating a continuously updated coupling feature matrix sequence along the time axis, and the row dimension of the coupling feature matrix corresponding to the sensing array position and the column dimension corresponding to the sampling points within the window; Performing difference value calculation on the matrix elements of the current window and the previous window in the coupling feature matrix sequence, obtaining a difference matrix reflecting the signal coupling strength variation rate, taking the sensing array spatial position as the node and the difference matrix element as the edge weight, and generating a dynamic correlation network data structure; Performing time-frequency domain tensor decomposition on the dynamic correlation network data structure, decomposing the spatiotemporal variation characteristics of the network into frequency-dependent components, spatial distribution components and time evolution components, and combining to generate a multi-dimensional time-frequency feature tensor, and the dimension of the multi-dimensional time-frequency feature tensor including the frequency axis, the sensor spatial axis and the time axis; The multi-dimensional time-frequency feature tensor is normalized, each dimension feature value is mapped to the same numerical interval, the geodesic distance between any two points in the normalized tensor is calculated, the geodesic path is determined through the near neighbor graph search, and a manifold surface coordinate point set representing the signal coupling relationship topological structure is generated. The manifold surface coordinate point set is extended along the time axis, the manifold state of adjacent sampling time is filled through nonlinear interpolation, continuous evolution dynamic surface data is generated, and a time sequence cooperative evolution field describing the dynamic coupling relationship of the vibration-acoustic print signal is obtained.

3. The method of claim 2, wherein, The vibration time history curve and the acoustic print spectrum feature curve in the original monitoring data set are timestamped and interpolated, the sampling delay of different sensing units is corrected, and vibration resampling sequences and acoustic resampling sequences with equal time intervals are generated, including: The vibration time history curve, the acoustic print spectrum feature curve and the corresponding timestamp sequence of each sensing unit in the original monitoring data set are read, and the timestamp sequence includes the collection time information of each sampling point; The timestamp sequence of each sensing unit is linearly regressed and fitted to obtain time reference deviation model parameters, the time reference deviation model parameters include time offset and sampling interval correction coefficients, and the model parameters are adjusted based on the system reference clock signal to synchronize the time reference of each sensing unit with the system reference clock; The vibration time history curve and the acoustic print spectrum feature curve are resampled based on the highest sampling frequency of the original monitoring data set as the reference sampling frequency, the interpolation results are calculated at the equal time interval points, and vibration resampling sequences and acoustic resampling sequences are generated; The spatial coordinate mapping relationship of the sensing array is established, the physical installation position of each sensing unit is converted into the two-dimensional rectangular coordinate system coordinate with the bridge expansion joint center as the origin, and a spatial coordinate matrix is generated, the elements of the spatial coordinate matrix are two-dimensional coordinate values of the sensing units; The spatial coordinate matrix, the vibration resampling sequence and the acoustic resampling sequence are associated according to the sensing unit position to generate a synchronous signal matrix with spatial coordinate markers, each element of the synchronous signal matrix includes the vibration signal amplitude, the acoustic signal amplitude, the spatial coordinate and the calibrated timestamp; The standard deviation of the signal amplitude difference of adjacent sensing units at the same timestamp is calculated, when the standard deviation is less than a preset threshold, the vibration resampling sequence and the acoustic resampling sequence are output, otherwise the timestamp interpolation and resampling steps are re-executed until the standard deviation meets the threshold requirement.

4. The method of claim 2, wherein, The multi-dimensional time-frequency feature tensor is normalized, each dimension feature value is mapped to the same numerical interval, the geodesic distance between any two points in the normalized tensor is calculated, the geodesic path is determined through the near neighbor graph search, and a manifold surface coordinate point set representing the signal coupling relationship topological structure is generated, including: The frequency axis, spatial axis and time axis feature values of the multi-dimensional time-frequency feature tensor are normalized respectively, each dimension feature value is mapped to the same numerical interval through linear transformation, and a normalized time-frequency feature tensor is generated; Eigenvalue decomposition is performed on the standardized time-frequency feature tensor, a manifold embedding dimension is determined according to an eigenvalue contribution rate, the manifold embedding dimension retains main information entropy of the original tensor, and an embedding space is constructed; In the embedding space, a square root of a number of spatial positions of the sensing array is taken as a number of neighbors, neighbor points of each tensor point are found through neighbor graph search, the neighbor points are connected to form a network structure, and a shortest path between any two points in the network is calculated as a geodesic distance; A distance matrix is constructed based on the geodesic distance, the distance matrix is converted into a coordinate point set in an Euclidean space through multidimensional scaling analysis, and preliminary embedding coordinates are generated; Riemannian metric correction is performed on the preliminary embedding coordinates, a Riemannian curvature tensor of each coordinate point is calculated, and coordinate positions are adjusted through gradient descent to minimize an error between a distance in the Euclidean space and a geodesic distance on a Riemannian manifold, and corrected manifold coordinates are generated; The corrected manifold coordinates are sorted according to a time sequence to generate a manifold trajectory coordinate point set evolving over time, discrete trajectory points are connected through nonlinear interpolation to generate a manifold surface coordinate point set representing a coupling relationship topological structure of a signal.

5. The method of claim 1, wherein, The time sequence collaborative evolution field is sequentially and differentially decomposed according to causality to separate a causal lag mode of a vibration signal sequence relative to a voiceprint signal sequence in a target frequency band range, including: Signal coupling evolution trajectories at different spatial positions are extracted from the time sequence collaborative evolution field, each signal coupling evolution trajectory corresponds to a vibration-voiceprint coupling relationship curve of a sensing unit changing over time, a horizontal coordinate of the signal coupling evolution trajectory is a time sampling point, and a vertical coordinate is a geodesic coordinate value on a manifold surface; Each coupling evolution trajectory is filtered through local weighted average in a sliding window to remove noise and retain a trend feature of the trajectory, and a recursive cumulative distance of any two filtered trajectories is calculated to generate a distance value reflecting a similarity of the trajectories, and all distance values are combined to form a trajectory distance matrix; The trajectory distance matrix is clustered, trajectories with smaller distance values are divided into the same cluster, a number of clusters is determined through contour coefficient calculation, a similarity of trajectories in a cluster is higher than a similarity of trajectories between clusters, and each cluster corresponds to a sensing unit group with similar coupling characteristics; A time-frequency analysis is performed on the coupling evolution trajectories of each cluster, an instantaneous frequency and an energy value of a trajectory in a window are calculated through window function sliding along a time axis to generate a time-frequency energy spectrum; Cumulative energy is calculated along a frequency axis in the time-frequency energy spectrum, a frequency range corresponding to a preset proportion of total energy is recorded as a target frequency band when the cumulative energy reaches the preset proportion, and the preset proportion is preset according to vibration characteristics of a bridge expansion joint; A causal relationship analysis is performed on the coupling evolution trajectories in the target frequency band range, a causal influence probability value of a vibration signal on a voiceprint signal is calculated, a phase difference sequence and an amplitude ratio sequence of the vibration signal relative to the voiceprint signal in the frequency band are extracted when the probability value is greater than a preset threshold value, and a causal lag mode is generated by combination, and a time length of the causal lag mode is consistent with a time sampling point number of the coupling evolution trajectory.

6. The method of claim 5, wherein, The filtering processing is performed on each coupling evolution track, local weighted average in a sliding window is used for denoising, trend characteristics of the track are retained, recursive cumulative distance calculation is performed on any two filtered tracks, distance values reflecting track similarity are generated, all distance values are combined to form a track distance matrix, including: The sampling point data of each coupling evolution track is filtered, a window length is set as a preset proportion of the total number of track time sampling points, polynomial fitting window data is used, and the fitting result is calculated as the filtered track data, retaining the trend characteristics of the track; For the two filtered tracks, the track sampling points are taken as nodes to construct a cumulative distance matrix, and the matrix elements are the Euclidean distances of the corresponding nodes of the two tracks; The minimum path of the cumulative distance matrix is solved by dynamic programming, the path extends from the top left corner to the bottom right corner of the matrix, and the sum of the elements on the path is the distance value of the two tracks, and the smaller the distance value, the higher the track similarity; The distance values of all coupling evolution tracks of the sensing units are calculated, the distance values are arranged according to the track serial numbers, and a track distance matrix is generated, the row and column of the track distance matrix correspond to the serial numbers of the sensing units, the diagonal elements are zero, and the non-diagonal elements are the distance values of the corresponding two tracks; The elements of the track distance matrix are normalized, the distance values are mapped to the same numerical interval through linear transformation, and a normalized track distance matrix is generated; The row mean and column mean of the normalized track distance matrix are calculated, when the difference between the row mean and the column mean is less than a preset threshold, it is determined that the matrix symmetry is qualified, otherwise the track filtering and distance calculation steps are re-executed until the symmetry requirement is met.

7. The method of claim 5, wherein, The coupling evolution tracks in the target frequency band range are analyzed for causal relationship, the causal influence probability value of the vibration signal on the voiceprint signal is calculated, when the probability value is greater than a preset threshold, the phase difference sequence and the amplitude ratio sequence of the vibration signal relative to the voiceprint signal in the frequency band are extracted, and a causal lag mode is generated by combination, including: The vibration component sequence and the voiceprint component sequence in the target frequency band range are extracted from the original monitoring data set, the signal components in the target frequency band are retained through band-pass filtering, and the signal components outside the frequency band are attenuated, so that the amplitude of the signal outside the frequency band is reduced to below a preset proportion of the amplitude of the signal within the frequency band; The vibration component sequence and the voiceprint component sequence are subjected to stationarity test, and the autocorrelation coefficient of the sequence is calculated, when the autocorrelation coefficient decays to zero with the lag order and satisfies the white noise characteristic, it is determined that the sequence is stationary, otherwise the sequence is subjected to difference processing until the sequence is stationary; The vibration component sequence is taken as the dependent variable, the lag term of the vibration component sequence and the lag term of the voiceprint component sequence are taken as the independent variables, a regression model is constructed, the lag order of the regression model is determined through information criterion calculation to minimize the information criterion value of the regression model, and the regression model residual satisfies the white noise characteristic; The vibration component sequence is calculated for the causal influence significance of the vibration component sequence on the voiceprint component sequence through test statistic. The statistical quantity is converted into a causal influence probability value, the probability value is calculated through a distribution function, the probability value represents a significant causal influence of a vibration signal on a voiceprint signal, and when the probability value is greater than a preset threshold value, it is determined that there is a significant causal relationship; The above process is repeated for all sampling points in the target frequency band to generate a time-varying causal influence probability sequence, isolated noise points in the probability sequence are removed through median filtering, the filtering window length is a preset proportion of the number of time sampling points, a smooth causal influence probability curve is generated, a phase difference sequence and an amplitude ratio sequence corresponding to a peak value of the curve are extracted, and the causal lag mode is generated by combination.

8. The method of claim 1, wherein, The causal lag mode is mapped to a preset physical force transmission path network to generate an implicit damage topology graph describing structural performance weakening characteristics, including: reading preset physical force transmission path network data, the nodes of the physical force transmission path network correspond to the stress key points of the key components of the expansion joint, the edges correspond to the force transmission paths between the components, and the edge attributes include the initial force flow transmission coefficients; the phase difference sequence and the amplitude ratio sequence in the causal lag mode are assigned to the network nodes at the corresponding spatial positions, and the mapping basis is the spatial distance between the sensor array installation position and the network node, and the causal lag mode of the sensor unit closest in distance is assigned to the corresponding node; calculating the damage characteristic value of each network node, the damage characteristic value being the Euclidean distance between the current causal lag mode and the health reference mode, the health reference mode being obtained by monitoring data acquisition when the bridge expansion joint is newly built, and being stored as a phase difference reference sequence and an amplitude ratio reference sequence, and the greater the distance value, the more serious the performance degradation of the node; adjusting the force flow transmission coefficients of the corresponding edges in the physical force transmission path network based on the node damage characteristic value, wherein the adjustment mode is to attenuate and correct the initial coefficients by the damage characteristic value, and the greater the damage characteristic value, the greater the attenuation degree of the weight of the corresponding edge; topology structure reconstruction is performed on the adjusted physical force transmission path network, the adjacency matrix and the Laplacian matrix of the network are calculated, the eigenvalues and eigenvectors of the matrix are extracted, and topology graph data reflecting the force flow distribution characteristics of the network are generated; the node size in the topology graph data is associated with the damage characteristic value, and the thickness of the edge is associated with the adjusted force flow transmission coefficient, to generate an implicit damage topology graph describing the structural performance weakening characteristics, and the node coordinates of the implicit damage topology graph are consistent with the spatial positions of the physical force transmission path network.

9. A bridge joint condition monitoring device, characterized by The device comprises: a data acquisition module configured to synchronously acquire vibration signal sequences and voiceprint signal sequences of a bridge expansion joint under operating loads through a distributed sensor array, and generate an original monitoring data set with time stamp alignment, wherein the original monitoring data set comprises vibration time history curves and voiceprint spectral feature curves at different acquisition positions; a data correlation module configured to perform time-space correlation modeling on the vibration signal sequences and the voiceprint signal sequences in the original monitoring data set, and generate a time sequence collaborative evolution field describing the dynamic coupling relationship between the vibration and voiceprint signals, wherein the dimension of the time sequence collaborative evolution field is associated with the sampling frequency and the spatial distribution density of the original monitoring data set. a data deconstruction module configured to perform causal sequential difference deconstruction on the time-coordinated evolution field to separate a causal lag mode of the vibration signal sequence relative to the voiceprint signal sequence in a target frequency band range, the causal lag mode containing a phase difference variation trend and an amplitude coupling coefficient between the vibration signal and the voiceprint signal; a damage characterization module configured to map the causal lag mode to a preset physical force transmission path network to generate an implicit damage topology graph describing structural performance weakening features, nodes of the implicit damage topology graph corresponding to key stress units in the force transmission path and edges corresponding to force flow transmission strength between the units; a state recognition module configured to quantitatively evaluate an overall state level and a local damage degree of the bridge expansion joint based on connectivity features of the nodes and strength features of the edges in the implicit damage topology graph to generate a monitoring result containing a state grade identification.

10. A bridge joint condition monitoring system comprising a memory and a processor, the memory storing a computer program operable on the processor, characterised in that, The processor implements the steps in the method of any one of claims 1 to 9 when executing the program.

Images