Bridge real-time monitoring and evaluation method based on beam end displacement and rotation angle

Abstract

The invention relates to the technical field of bridge monitoring, and discloses a real-time bridge monitoring and evaluating method based on beam end displacement and rotation angle, which comprises the following steps: acquiring beam end displacement and rotation angle data acquired by a displacement sensor and a rotation angle sensor in a continuous monitoring period, and filtering and normalizing the acquired data to obtain processed data; constructing a discrete curve and performing polynomial curve fitting by using a settlement trajectory multi-cycle residual error clustering identification method, calculating a single-cycle residual error sequence, determining an edge fluctuation trend point by comparing adjacent cycle residual error values, and performing anomaly marking; inputting the marked abnormal data into an abnormal recognition model, predicting the risk level of the bridge structure through the abnormal recognition model, and outputting an abnormal recognition result; outputting a comprehensive structure response diagnosis report according to the abnormity identification result; the precision of monitoring and evaluation is improved, and safe operation of the bridge is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge monitoring, and in particular to a real-time monitoring and evaluation method for bridges based on beam end displacement and rotation angle. Background Art

[0002] As key nodes in transportation networks, the safety and reliability of bridges are crucial. As bridges age and traffic volumes continue to grow, they gradually experience damage and performance degradation. Traditional bridge monitoring methods, which often rely on manual inspections, suffer from low efficiency, strong subjectivity, and difficulty detecting early signs of subtle damage. These methods cannot meet the demands of modern bridge real-time monitoring and rapid assessment. While existing technologies for structural health monitoring widely utilize various sensors and data analysis methods, they are limited by single-parameter monitoring or linear models and struggle to fully capture the nonlinear behavior and subtle changes in structures under complex operating conditions. For example, traditional methods lack sufficient sensitivity and adaptability to handle data fluctuations caused by sudden loads or environmental changes, potentially leading to misjudgments or omissions of potential structural risks. Furthermore, reliance on a single data source or parameter limits the monitoring system's ability to identify structural behavior patterns under the influence of complex and variable factors. This limitation is particularly pronounced for large structures such as bridges, where structural responses are influenced by multiple factors, including load variations, material fatigue, and environmental conditions. Existing technologies fail to provide a systematic solution that fully utilizes the interactive responses of multiple parameters, which often leads to incomplete assessments of structural health status, affecting the timeliness and accuracy of maintenance decisions, and may increase maintenance costs or risks if problems are not identified in a timely manner. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and to design a real-time monitoring and evaluation method for bridges based on beam end displacement and rotation angle.

[0004] The present invention provides a real-time monitoring and evaluation method for a bridge based on beam end displacement and rotation angle, the method comprising the following steps: Obtaining the beam end displacement and rotation angle data collected by the displacement sensor and the rotation angle sensor during the continuous monitoring period, and filtering and normalizing the collected data to obtain processed data; The multi-period residual clustering identification method of sedimentation trajectory is used to construct a discrete curve and perform polynomial curve fitting, calculate the single-period residual sequence, and determine the edge fluctuation trend point by comparing the residual values ​​of adjacent periods and mark the anomaly; Input the marked abnormal data into the abnormality recognition model, predict the risk level of the bridge structure through the abnormality recognition model, and output the abnormality recognition results; Based on the anomaly identification results, a comprehensive structural response diagnosis report containing measurement point number, time index, risk type and level is output.

[0005] Optionally, in the first implementation of the present invention, the step of acquiring the beam end displacement and angle data collected by the displacement sensor and the angle sensor during the continuous monitoring period, and filtering and normalizing the collected data to obtain processed data includes: Obtaining the beam end displacement and rotation angle data collected by the displacement sensor and rotation angle sensor during the continuous monitoring period; The Kalman filter algorithm based on the maximum correlation entropy criterion is used to filter the collected data, and the data after Kalman filter processing is subjected to empirical mode decomposition to obtain the denoised beam end displacement and rotation angle data; Starting from the starting point of the time series, a sliding window is slid sequentially on the denoised beam end displacement and rotation angle data. In each window, the local minimum and maximum values ​​of the denoised beam end displacement and rotation angle data are calculated, and the data in the window are normalized to obtain the processed data.

[0006] Optionally, in a second implementation of the present invention, the filtering process of the collected data using a Kalman filter algorithm based on the maximum relevant entropy criterion includes: Initialize the state vector and covariance matrix, and at each time step, predict the current state and covariance based on the state estimate at the previous moment; Calculate the Kalman gain at the current moment, dynamically adjust the noise statistical characteristics using the maximum correlation entropy criterion, and correct the state estimate and covariance based on the difference between the measured and predicted values; In each iteration, the state estimation value and covariance matrix are continuously updated according to the new measurement value, and the iteration is continued until the filtering processing of all collected data is completed to obtain the filtered beam end displacement and rotation angle data.

[0007] Optionally, in a third implementation of the present invention, performing empirical mode decomposition on the data processed by the Kalman filter to obtain denoised beam end displacement and rotation angle data includes: The filtered beam end displacement and rotation data are used as the input of empirical mode decomposition. Each time point corresponds to a displacement or rotation measurement value to obtain the original signal. Decompose the original signal into several IMF components, perform wavelet threshold denoising on each IMF component, and remove the wavelet coefficients corresponding to high-frequency noise; The IMF components after wavelet threshold denoising are reconstructed to obtain the denoised beam end displacement and rotation angle data.

[0008] Optionally, in a fourth implementation of the present invention, the method of using the multi-cycle residual clustering identification method of the sedimentation trajectory to construct a discrete curve and perform polynomial curve fitting, calculate the single-cycle residual sequence, determine the edge fluctuation trend point by comparing the residual values ​​of adjacent cycles and perform abnormal marking, including: Collect the settlement data of the beam end in multiple consecutive monitoring cycles, arrange the settlement data in each monitoring cycle in chronological order to form a discrete data point sequence, and construct a discrete curve for each cycle; For the discrete curve of each cycle, a polynomial curve fitting is performed on the discrete curve of each cycle, and the actual measured settlement value is subtracted from the value on the corresponding polynomial fitting curve to obtain the residual value at each time point in the cycle; Arrange these residual values ​​in chronological order to form a single-period residual sequence of the period, and use DTW-BIRCH clustering to analyze the single-period residual sequence; The VB-CPD algorithm is applied to each single-cycle residual sequence to identify mutation points in the sequence. The residual sequences of multiple measurement points are modeled as graph signals. The graph signals are transformed from the spatial domain to the frequency domain through a graph Fourier transform. By analyzing the results of the graph Fourier transform, joint space-time anomalies are detected. Combining the DTW-BIRCH clustering results, the mutation points detected by VB-CPD, and the detected space-time joint anomalies, the marginal fluctuation trend points in the residual sequence are determined and marked as abnormal.

[0009] Optionally, in a fifth implementation of the present invention, the adopting DTW-BIRCH cluster analysis of a single-period residual sequence includes: Perform DTW operations on single-period residual sequences of different periods to find the optimal alignment path between different residual sequences; The DTW distance between residual sequences of different periods is calculated, and the residual sequence after DTW processing is input into the BIRCH hierarchical clustering algorithm. The BIRCH algorithm clusters the residual sequence, finds abnormal patterns in the data, and obtains clustering results.

[0010] Optionally, in a sixth implementation of the present invention, applying the VB-CPD algorithm to each single-period residual sequence to identify a mutation point in the sequence includes: Construct a state space model to describe the residual sequence, determine the observation equation and state transfer equation, and initialize the parameters of the variational distribution; Define the variational lower bound as the optimization objective and calculate the value of the variational lower bound based on the current variational parameters and state space model; Use the variational inference algorithm to update the variational parameters to maximize the variational lower bound. The variational parameters of the state variables are updated according to the current observation values ​​and the state transfer equation. The variational parameters of the mutation point are updated according to the characteristics of the residual sequence and prior information. By comparing the values ​​of the variational lower bound in two adjacent iterations, if the difference is less than the preset threshold, the variational lower bound is considered to have converged. According to the variational distribution after convergence, the position of the mutation point is determined, and the identified mutation point information is output.

[0011] Optionally, in a seventh implementation of the present invention, inputting the marked abnormal data into an abnormality recognition model, predicting the risk level of the bridge structure by the abnormality recognition model, and outputting the abnormality recognition result includes: The labeled anomaly data is normalized to obtain a displacement and rotation sequence, which is then fed into the Transformer encoder of the anomaly recognition model. Positional encoding is added to each element in the sequence. The anomaly recognition model is built based on the Transformer, GNN, and PINN networks. The Transformer encoder consists of multiple encoder layers, each of which contains a multi-head self-attention mechanism and a feedforward neural network to extract feature information; The feature information is input into the GNN as node features. Each node generates a message based on the features of its neighbor nodes and the information of the edges, and passes the generated message to the adjacent nodes. Each node updates its own feature representation based on the received message, learns the topological relationship by aggregating the information of neighboring nodes, deepens the modeling of the topological relationship through multiple GNN layers, and outputs features that reflect the topological relationship; The temporal features output by the Transformer encoder and the topological features output by the GNN are input into the cross-attention mechanism. The fused features are mapped to the risk level space through the fully connected layer to obtain a preliminary risk level prediction result. The PINN network is used to mechanically constrain the preliminary risk level prediction results and map them to the corresponding risk level. The risk level is output and the corresponding measurement point number and time index are recorded.

[0012] Optionally, in an eighth implementation of the present invention, extracting feature information through a multi-head self-attention mechanism and a feedforward neural network includes: Calculate the attention score of each element and other elements in the sequence, use the multi-head mechanism to focus on the dependencies in the sequence from different subspaces, perform weighted summation of the elements according to the attention score, and obtain the feature representation processed by the attention mechanism; The features output by the attention mechanism are input into the feedforward neural network, and nonlinear transformation is performed to extract features. The attention and feedforward operations are repeated, and the feature extraction is deepened through multiple encoder layers to output features that can reflect the temporal dependence of displacement and angle sequences.

[0013] Optionally, in a ninth implementation of the present invention, inputting the temporal features output by the Transformer encoder and the topological features output by the GNN into the cross attention mechanism includes: The attention score between temporal features and topological features is calculated, and the temporal features and topological features are weighted and summed according to the attention score. The spatiotemporal feature information is integrated to obtain a comprehensive feature representation.

[0014] The technical solution provided by the present invention obtains beam end displacement and rotation data collected by displacement sensors and rotation angle sensors within a continuous monitoring period, and filters and normalizes the collected data to obtain processed data. A multi-cycle residual clustering recognition method for settlement trajectories is used to construct a discrete curve and perform polynomial curve fitting, calculate a single-cycle residual sequence, and determine edge fluctuation trend points by comparing residual values ​​of adjacent cycles, and mark anomalies. The marked anomaly data is input into an anomaly recognition model, which predicts the risk level of the bridge structure and outputs an anomaly recognition result. Based on the anomaly recognition result, a comprehensive structural response diagnosis report containing a measurement point number, a time index, and a risk type and level is output. The present invention automatically learns the complex relationship between multiple parameters such as beam end displacement and rotation angle, and can more accurately capture the nonlinear behavior and subtle changes of bridge structures under complex working conditions. This improves the accuracy of monitoring and assessment, achieves comprehensive and accurate monitoring of the bridge structure status, reduces misjudgments and omissions of potential structural risks, and outputs a visual comprehensive structural response diagnosis report, providing intuitive and accurate data support for bridge operation and maintenance personnel, facilitating timely and reasonable maintenance decisions, reducing maintenance costs, and ensuring the safe operation of the bridge. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0016] Figure 1 A schematic diagram of a first embodiment of a bridge real-time monitoring and evaluation method based on beam end displacement and rotation provided by an embodiment of the present invention; Figure 2 A schematic diagram of a second embodiment of a bridge real-time monitoring and evaluation method based on beam end displacement and rotation provided by an embodiment of the present invention; Figure 3A schematic diagram of a third embodiment of a real-time bridge monitoring and evaluation method based on beam end displacement and rotation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The terms "first," "second," "third," "fourth," and so forth (if any) in the description and claims of the present invention and in the accompanying drawings are used to distinguish similar items and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that shown or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or device that includes a series of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, product, or device.

[0018] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 A schematic diagram of a first embodiment of a bridge real-time monitoring and evaluation method based on beam end displacement and rotation provided by an embodiment of the present invention, wherein the method specifically comprises the following steps: Step 101: Obtain beam end displacement and rotation angle data collected by the displacement sensor and the rotation angle sensor during a continuous monitoring period, and filter and normalize the collected data to obtain processed data; Step 102: Use the multi-cycle residual clustering identification method of the sedimentation trajectory to construct a discrete curve and perform polynomial curve fitting, calculate the single-cycle residual sequence, determine the edge fluctuation trend point by comparing the residual values ​​of adjacent cycles, and mark the abnormality; Step 103: input the marked abnormal data into the abnormality recognition model, predict the risk level of the bridge structure through the abnormality recognition model, and output the abnormality recognition result; In this embodiment, the marked abnormal data is normalized to obtain a displacement and angle sequence, which is input into the Transformer encoder of the abnormality recognition model, and a position code is added to each element in the sequence, wherein the abnormality recognition model is constructed based on the Transformer, GNN and PINN network; the Transformer encoder is composed of multiple encoder layers, each encoder layer contains a multi-head self-attention mechanism and a feedforward neural network, and feature information is extracted through the multi-head self-attention mechanism and the feedforward neural network; the feature information is input into the GNN as a node feature, and each node generates a message based on the features of its neighboring nodes and the information of the edges, and transmits the generated message to the adjacent nodes; each node updates its own feature representation based on the received message By aggregating the information of neighboring nodes, the topological relationship is learned, the modeling of the topological relationship is deepened through multiple GNN layers, and features reflecting the topological relationship are output; the temporal features output by the Transformer encoder and the topological features output by the GNN are input into the cross-attention mechanism, the attention score between the temporal features and the topological features is calculated, the temporal features and the topological features are weighted and summed according to the attention score, the spatiotemporal feature information is fused, and a comprehensive feature representation is obtained. The fused features are mapped to the risk level space through the fully connected layer to obtain a preliminary risk level prediction result; the PINN network is used to mechanically constrain the preliminary risk level prediction result and map it to the corresponding risk level, output the risk level, and record the corresponding measurement point number and time index.

[0019] In this embodiment, a physical information embedded neural network (PINN) is used: the bridge mechanics equations, such as the equilibrium equations and constitutive equations in structural mechanics, are determined, and the bridge mechanics equations are added as loss terms to the loss function of the model. During the model training process, the total loss function containing the physical information loss term is minimized, so that the model learns feature representations that conform to physical laws, thereby enhancing the generalization of the model.

[0020] In this embodiment, the attention score of each element and other elements in the sequence is calculated, and the dependency relationship in the sequence is paid attention to from different subspaces through a multi-head mechanism. The elements are weighted and summed according to the attention score to obtain the feature representation processed by the attention mechanism. The features output by the attention mechanism are input into a feedforward neural network, and nonlinear transformation is performed to extract features. The attention and feedforward operations are repeated, and the feature extraction is deepened through multiple encoder layers to output features that can reflect the temporal dependency of the displacement and angle sequences.

[0021] Step 104: Output a comprehensive structural response diagnosis report including the measurement point number, time index, risk type and level according to the abnormality identification result.

[0022] In this embodiment, the three typical links of identifying sudden change nodes in the settlement trajectory of the beam end, delineating the local deformation jump section, and locking the abnormal points of the interaction between strain and rotation are comprehensively optimized to construct a joint identification mechanism based on the multi-parameter cross-response characteristics of the time series; the target groups are the technical application end entities such as the bridge structure operation and maintenance units and the development team of the automatic identification system for the structural health status; multiple original physical parameters such as the vertical settlement value of the beam end during the continuous monitoring period, the displacement difference ratio in any time period, the strain differential characteristic sequence, the dynamic change slope of the rotation angle, etc.; key technical contents such as multi-period residual analysis of the settlement trajectory, identification of the critical area of ​​the displacement difference ratio of the beam end, setting of the mapping rule of the intersection of strain and rotation angle, and output of the classification level of abnormal nodes are included.

[0023] In this embodiment, multi-cycle residual clustering for settlement trajectory identification involves selecting representative values ​​from the beam end settlement displacement over five consecutive cycles, constructing a discrete curve with the periodic time series as the horizontal axis, performing a polynomial curve fitting operation, and calculating the point-by-point difference between the actual and fitted values ​​to obtain a single-cycle residual sequence. Based on this sequence, the relative positions of the residual values ​​at each node in adjacent cycles are compared. Residual values ​​at the same position in adjacent cycles are sorted by size, the absolute value of the difference is determined, and the residual continuity state is cross-checked. Points exhibiting marginal fluctuation trends over three or more cycles are identified and flagged as abnormal based on whether they cross the extreme values ​​of the historical residual interval.

[0024] In this embodiment, the displacement difference ratio trend mutation identification and segment expansion: the displacement values ​​of two adjacent sampling periods are obtained from the beam end monitoring time sequence, and after calculating the difference, the difference is deduced by ratio with the difference of the previous period to obtain the difference ratio sequence in the continuous time period. The monitoring rule is set as follows: if the difference ratio shows a numerical multiplication characteristic in three consecutive sampling cycles, and the direction of difference ratio change is consistent with the trend of the adjacent measuring points, then the measuring point is locked as the jump center point. With this point as the core, two points are extended to the left and right measuring points respectively to establish the initial local deformation segment. Determine whether the difference ratio in subsequent sampling of the segment tends to be monotonic. If the monotonic change condition is not met, the segment is frozen and no longer expanded to ensure that the area is included in the subsequent assessment as a key risk segment.

[0025] In this embodiment, the interactive mapping mechanism of the intersection of strain differential and angle slope is as follows: for the same beam end measuring point, its strain value in a continuous time period is extracted, the difference calculation between adjacent time points is performed to form a differential sequence, and the differential surge point is marked; in the corresponding period, the slope of the angle time series is calculated and the slope reversal node is extracted. The strain differential and the angle slope nodes are compared by time index, and the point pairs with overlapping indexes or intervals less than the set threshold are extracted to determine whether the point pairs meet the threshold conditions of consistent slope direction or strain surge amplitude exceeding twice the background mean. If so, it is used as an intersection anomaly point and input into the risk level division logic module. The response level of the anomaly point is set according to parameters such as strain peak value and node density.

[0026] In this embodiment, a multi-source feature collaborative risk level determination and output mechanism is implemented: the set of abnormal points identified in the above three steps is unified on a timeline to determine whether there are areas where multiple types of abnormal features occur in a similar time period. If a concentrated area is found, the system uses these time points as the core window to extract the corresponding deformation parameters, including settlement amplitude, difference ratio series, strain jump rate, corner reversal frequency, etc., and scores them according to the set weights. Based on the level range to which the score belongs, three risk labels are assigned: high, secondary, and basic. A comprehensive structural response diagnostic report containing the measurement point number, time index, risk type, and level is output, providing direct data support for subsequent on-site management and structural maintenance decisions.

[0027] In this embodiment, high-precision displacement and rotation sensors are installed at the ends of the bridge beams. These sensors operate continuously over a continuous monitoring period, covering daily, monthly, quarterly, and annual periods. Displacement and rotation data from the beam ends are collected at pre-set intervals, such as every minute or every hour, to obtain raw data reflecting the real-time status of the bridge structure at different time scales. Data filtering: The collected data may contain various noises, such as sensor errors and external environmental interference. To remove this noise, a filtering algorithm is used to process the data. This filtering operation filters out the noise, making the data smoother and more accurate, and ensuring the reliability of subsequent analysis results. Data normalization: Data collected by different sensors may have different magnitudes and distribution ranges. To make subsequent analysis and processing more unified and effective, the filtered data needs to be normalized. Normalization involves mapping the data to a specific range, such as the [0, 1] interval. This eliminates magnitude differences between different data, improving data comparability and model stability. The preprocessed beam end displacement data is organized according to different time scales (daily, monthly, quarterly, and annual). The time point within each time scale is used as the horizontal axis, and the displacement value at the corresponding time point is used as the vertical axis. These data points form discrete curves at different time scales. Discrete curves can intuitively display the displacement changes of the beam end at different time scales. Polynomial curve fitting is performed on the discrete curves at different time scales. In this way, a suitable curve is found for each discrete data point at each time scale to approximate its variation pattern, resulting in a mathematical model that can describe the overall trend of the subsidence trajectory at different time scales. The single-cycle residual at each time scale is calculated, that is, the difference between the actual displacement data and the corresponding point on the polynomial fitting curve at the corresponding time scale. These residuals are arranged in chronological order to obtain a single-cycle residual sequence at different time scales. The residual sequence reflects the difference between the actual data and the fitting curve, indicating the portion of the data that deviates from the overall trend. Edge fluctuation trend points are identified and anomalies are marked. For single-cycle residual sequences at different time scales, the residual values ​​of adjacent cycles are compared. If the residual value of a cycle shows a significant change compared to adjacent cycles or deviates significantly from the initial value at that time scale, this point may be an edge fluctuation trend point. These edge fluctuation trend points are marked as anomalies for subsequent attention and analysis. At the same time, the residual sequence is statistically analyzed to observe the data distribution characteristics. If a sudden change in the data is found, it is also included in the anomaly mark range. The marked anomaly data at different time scales is input into the pre-trained anomaly recognition model.After being trained on a large amount of data, the model has the ability to identify different types of anomalies. The anomaly recognition model analyzes and judges the input anomaly data, and based on the data characteristics and the patterns learned by the model, predicts the current risk level of the bridge structure at different time scales. Risk levels can be divided into high, medium, and low levels, each representing a different degree of potential safety hazards that the bridge structure may face. After the model completes its prediction, it outputs anomaly recognition results at different time scales, including the corresponding risk level for each measuring point and related anomaly information. These results will provide an important basis for subsequent decision-making. Based on the anomaly recognition results at different time scales, information such as the number of each measuring point, the corresponding time index (accurate to the day, month, quarter, or year), the risk type, and the predicted risk level are collected and organized. At the same time, the results of statistical analysis and regression analysis of long-term monitoring data are integrated into it, and the changing trends of data, the relationship between abnormal situations and time, etc. are explained in detail. This information is the core content of the report and can comprehensively reflect the status of the bridge structure at different time scales. The sorted information is generated according to a certain format and specification to generate a comprehensive structural response diagnosis report containing measurement point number, time index, risk type and level. The report can be in various forms such as charts and text descriptions to intuitively display the risk situation of the bridge structure during the long-term monitoring process, and provide a scientific decision-making basis for the maintenance and management of the bridge. For situations judged to be failure or sub-health, they are highlighted in the report and corresponding recommended measures are given.

[0028] In this embodiment, by introducing a joint identification mechanism for multi-parameter cross-response characteristics in a time series, the accuracy and predictive capabilities of structural health monitoring are significantly enhanced. In particular, in the multi-cycle residual analysis of settlement trajectories, periodic data fitting and comparison of the clustering of residual values ​​within consecutive cycles enable the early identification of potential abnormal nodes, rather than waiting until structural damage manifests before taking action. Furthermore, critical region identification for beam end displacement difference ratios effectively identifies nonlinear jumps in displacement behavior by comparing displacement change ratios at consecutive time points, facilitating the timely detection and response to possible stress concentrations or localized damage. By setting rules for mapping intersections between strain and rotation slope, the system allows correlations between complex stresses and structural responses, increasing the dimensionality and complexity of monitoring data and providing an effective method for locating and preventing potential risks to critical structures. The combination of these technical aspects not only improves the accuracy of early warnings but also deepens problem analysis, thereby avoiding the limitations of over-reliance on a single parameter and achieving a more comprehensive structural health assessment.

[0029] See also Figure 2 , a schematic diagram of a second embodiment of a bridge real-time monitoring and evaluation method based on beam end displacement and rotation provided by an embodiment of the present invention, the method comprising: Step 201: Acquire beam end displacement and rotation angle data collected by the displacement sensor and rotation angle sensor during a continuous monitoring period; Step 202: Filter the collected data using a Kalman filter algorithm based on the maximum correlation entropy criterion, perform empirical mode decomposition on the data processed by the Kalman filter, and obtain denoised beam end displacement and rotation angle data; In this embodiment, the state vector and covariance matrix are initialized, and at each time step, the state and covariance at the current moment are predicted based on the state estimate at the previous moment; the Kalman gain at the current moment is calculated, and the noise statistical characteristics are dynamically adjusted using the maximum correlation entropy criterion. The state estimate and covariance are corrected based on the difference between the measured value and the predicted value; in each iteration process, the state estimate and covariance matrix are continuously updated based on the new measured value, and iteration is continued until the filtering processing of all collected data is completed to obtain the filtered beam end displacement and angle data.

[0030] In this embodiment, the filtered beam end displacement and rotation angle data are used as the input of empirical mode decomposition, and each time point corresponds to a displacement or rotation angle measurement value to obtain the original signal; the original signal is decomposed into several IMF components, and each IMF component is subjected to wavelet threshold denoising to remove the wavelet coefficients corresponding to the high-frequency noise; the IMF components after wavelet threshold denoising are reconstructed to obtain the denoised beam end displacement and rotation angle data.

[0031] Step 203: Starting from the starting point of the time series, a sliding window is sequentially slid over the denoised beam end displacement and rotation angle data. Within each window, the local minimum and maximum values ​​of the denoised beam end displacement and rotation angle data are calculated, and the data within the window are normalized to obtain the processed data.

[0032] See also Figure 3 , a schematic diagram of a third embodiment of a bridge real-time monitoring and evaluation method based on beam end displacement and rotation provided by an embodiment of the present invention, the method comprising: Step 301: Collect the settlement data of the beam end in a plurality of consecutive monitoring cycles, arrange the settlement data in each monitoring cycle in chronological order to form a discrete data point sequence, and construct a discrete curve for each cycle; Step 302: For each period's discrete curve, perform polynomial curve fitting on the discrete curve, subtract the actual measured settlement value from the value on the corresponding polynomial fitting curve, and obtain the residual value at each time point in the period; Step 303: Arrange the residual values ​​in chronological order to form a single-period residual sequence of the period, and use DTW-BIRCH clustering to analyze the single-period residual sequence; Step 304: Apply the VB-CPD algorithm to each single-cycle residual sequence to identify the mutation point in the sequence, model the residual sequences of multiple measurement points as a graph signal, perform a graph Fourier transform on the graph signal to convert the signal from the spatial domain to the frequency domain, and detect joint space-time anomalies by analyzing the results of the graph Fourier transform; Step 305: Combining the DTW-BIRCH clustering results, the mutation points detected by VB-CPD, and the detected space-time joint anomalies, determine the edge fluctuation trend points in the residual sequence and mark them as abnormal.

[0033] In this embodiment, DTW operations are performed on single-period residual sequences of different periods to find the optimal alignment path between different residual sequences; the DTW distances between residual sequences of different periods are calculated, and the residual sequences after DTW processing are input into the BIRCH hierarchical clustering algorithm. The BIRCH algorithm clusters the residual sequences, discovers abnormal patterns in the data, and obtains clustering results.

[0034] In this embodiment, a state-space model is constructed to describe the residual sequence, the observation equation and the state transition equation are determined, and the parameters of the variational distribution are initialized; a variational lower bound is defined as the optimization target, and the value of the variational lower bound is calculated based on the current variational parameters and the state-space model; a variational inference algorithm is used to update the variational parameters to maximize the variational lower bound, the variational parameters of the state variables are updated based on the current observation values ​​and the state transition equation, and the variational parameters of the mutation point position are updated based on the characteristics of the residual sequence and prior information; by comparing the values ​​of the variational lower bound in two adjacent iterations, if the difference is less than a preset threshold, it is considered that the variational lower bound has converged, and the position of the mutation point is determined based on the converged variational distribution, and the identified mutation point information is output.

[0035] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A real-time monitoring and evaluation method for bridges based on beam end displacement and rotation angle, characterized in that: The method comprises the following steps: Obtaining the beam end displacement and rotation angle data collected by the displacement sensor and the rotation angle sensor during the continuous monitoring period, and filtering and normalizing the collected data to obtain processed data; The multi-period residual clustering identification method of sedimentation trajectory is used to construct a discrete curve and perform polynomial curve fitting, calculate the single-period residual sequence, and determine the edge fluctuation trend point by comparing the residual values ​​of adjacent periods and mark the anomaly; Input the marked abnormal data into the abnormality recognition model, predict the risk level of the bridge structure through the abnormality recognition model, and output the abnormality recognition results; Based on the anomaly identification results, a comprehensive structural response diagnosis report containing measurement point number, time index, risk type and level is output.

2. A bridge real-time monitoring and evaluation method based on beam end displacement and rotation angle according to claim 1, characterized in that: The step of obtaining the beam end displacement and rotation angle data collected by the displacement sensor and the rotation angle sensor during the continuous monitoring period, and filtering and normalizing the collected data to obtain processed data includes: Obtaining the beam end displacement and rotation angle data collected by the displacement sensor and rotation angle sensor during the continuous monitoring period; The Kalman filter algorithm based on the maximum correlation entropy criterion is used to filter the collected data, and the data after Kalman filter processing is subjected to empirical mode decomposition to obtain the denoised beam end displacement and rotation angle data; Starting from the starting point of the time series, a sliding window is slid sequentially on the denoised beam end displacement and rotation angle data. In each window, the local minimum and maximum values ​​of the denoised beam end displacement and rotation angle data are calculated, and the data in the window are normalized to obtain the processed data.

3. A bridge real-time monitoring and evaluation method based on beam end displacement and rotation angle as claimed in claim 2, characterized in that: The method of filtering the collected data using a Kalman filter algorithm based on the maximum correlation entropy criterion includes: Initialize the state vector and covariance matrix, and at each time step, predict the current state and covariance based on the state estimate at the previous moment; Calculate the Kalman gain at the current moment, dynamically adjust the noise statistical characteristics using the maximum correlation entropy criterion, and correct the state estimate and covariance based on the difference between the measured and predicted values; In each iteration, the state estimation value and covariance matrix are continuously updated according to the new measurement value, and the iteration is continued until the filtering processing of all collected data is completed to obtain the filtered beam end displacement and rotation angle data.

4. The bridge real-time monitoring and evaluation method based on beam end displacement and rotation angle according to claim 2 is characterized in that: The data processed by Kalman filtering is subjected to empirical mode decomposition to obtain denoised beam end displacement and rotation angle data, including: The filtered beam end displacement and rotation data are used as the input of empirical mode decomposition. Each time point corresponds to a displacement or rotation measurement value to obtain the original signal. Decompose the original signal into several IMF components, perform wavelet threshold denoising on each IMF component, and remove the wavelet coefficients corresponding to high-frequency noise; The IMF components after wavelet threshold denoising are reconstructed to obtain the denoised beam end displacement and rotation angle data.

5. The bridge real-time monitoring and evaluation method based on beam end displacement and rotation angle according to claim 1 is characterized in that: The method of using the multi-cycle residual clustering identification method of the sedimentation trajectory to construct a discrete curve and perform polynomial curve fitting, calculate the single-cycle residual sequence, determine the edge fluctuation trend point by comparing the residual values ​​of adjacent cycles and perform abnormal marking, including: Collect the settlement data of the beam end in multiple consecutive monitoring cycles, arrange the settlement data in each monitoring cycle in chronological order to form a discrete data point sequence, and construct a discrete curve for each cycle; For the discrete curve of each cycle, a polynomial curve fitting is performed on the discrete curve of each cycle, and the actual measured settlement value is subtracted from the value on the corresponding polynomial fitting curve to obtain the residual value at each time point in the cycle; Arrange these residual values ​​in chronological order to form a single-period residual sequence of the period, and use DTW-BIRCH clustering to analyze the single-period residual sequence; The VB-CPD algorithm is applied to each single-cycle residual sequence to identify mutation points in the sequence. The residual sequences of multiple measurement points are modeled as graph signals. The graph signals are transformed from the spatial domain to the frequency domain through a graph Fourier transform. By analyzing the results of the graph Fourier transform, joint space-time anomalies are detected. Combining the DTW-BIRCH clustering results, the mutation points detected by VB-CPD, and the detected space-time joint anomalies, the marginal fluctuation trend points in the residual sequence are determined and marked as abnormal.

6. A bridge real-time monitoring and evaluation method based on beam end displacement and rotation angle as claimed in claim 5, characterized in that: The DTW-BIRCH cluster analysis of a single-period residual sequence includes: Perform DTW operations on single-period residual sequences of different periods to find the optimal alignment path between different residual sequences; The DTW distance between residual sequences of different periods is calculated, and the residual sequence after DTW processing is input into the BIRCH hierarchical clustering algorithm. The BIRCH algorithm clusters the residual sequence, finds abnormal patterns in the data, and obtains clustering results.

7. The bridge real-time monitoring and evaluation method based on beam end displacement and rotation angle according to claim 5 is characterized in that: The step of applying the VB-CPD algorithm to each single-cycle residual sequence to identify a mutation point in the sequence includes: Construct a state space model to describe the residual sequence, determine the observation equation and state transfer equation, and initialize the parameters of the variational distribution; Define the variational lower bound as the optimization objective and calculate the value of the variational lower bound based on the current variational parameters and state space model; Use the variational inference algorithm to update the variational parameters to maximize the variational lower bound. The variational parameters of the state variables are updated according to the current observation values ​​and the state transfer equation. The variational parameters of the mutation point are updated according to the characteristics of the residual sequence and prior information. By comparing the values ​​of the variational lower bound in two adjacent iterations, if the difference is less than the preset threshold, the variational lower bound is considered to have converged. According to the variational distribution after convergence, the position of the mutation point is determined, and the identified mutation point information is output.

8. The bridge real-time monitoring and evaluation method based on beam end displacement and rotation angle according to claim 1 is characterized in that: The marked abnormal data is input into the abnormality recognition model, the risk level of the bridge structure is predicted by the abnormality recognition model, and the abnormality recognition result is output, including: The labeled anomaly data is normalized to obtain a displacement and rotation sequence, which is then fed into the Transformer encoder of the anomaly recognition model. Positional encoding is added to each element in the sequence. The anomaly recognition model is built based on the Transformer, GNN, and PINN networks. The Transformer encoder consists of multiple encoder layers, each of which contains a multi-head self-attention mechanism and a feedforward neural network to extract feature information; The feature information is input into the GNN as node features. Each node generates a message based on the features of its neighbor nodes and the information of the edges, and passes the generated message to the adjacent nodes. Each node updates its own feature representation based on the received message, learns the topological relationship by aggregating the information of neighboring nodes, deepens the modeling of the topological relationship through multiple GNN layers, and outputs features that reflect the topological relationship; The temporal features output by the Transformer encoder and the topological features output by the GNN are input into the cross-attention mechanism. The fused features are mapped to the risk level space through the fully connected layer to obtain a preliminary risk level prediction result. The PINN network is used to mechanically constrain the preliminary risk level prediction results and map them to the corresponding risk level. The risk level is output and the corresponding measurement point number and time index are recorded.

9. The bridge real-time monitoring and evaluation method based on beam end displacement and rotation angle according to claim 8, characterized in that: The feature information is extracted through the multi-head self-attention mechanism and feedforward neural network, including: Calculate the attention score of each element and other elements in the sequence, use the multi-head mechanism to focus on the dependencies in the sequence from different subspaces, perform weighted summation of the elements according to the attention score, and obtain the feature representation processed by the attention mechanism; The features output by the attention mechanism are input into the feedforward neural network, and nonlinear transformation is performed to extract features. The attention and feedforward operations are repeated, and the feature extraction is deepened through multiple encoder layers to output features that can reflect the temporal dependence of displacement and angle sequences.

10. The bridge real-time monitoring and evaluation method based on beam end displacement and rotation angle according to claim 8, characterized in that: The temporal features output by the Transformer encoder and the topological features output by the GNN are input into the cross attention mechanism, including: The attention score between temporal features and topological features is calculated, and the temporal features and topological features are weighted and summed according to the attention score. The spatiotemporal feature information is integrated to obtain a comprehensive feature representation.

Images