Bridge health state monitoring and alarming method and system

Abstract

The application provides a bridge health state monitoring alarm method and system, relates to the technical field of data monitoring alarm, and the method comprises the following steps: collecting real-time strain data and historical strain data of each monitoring point, integrating the two into a real-time sequence, constructing a bridge full-life-cycle strain data historical sequence, obtaining peak point coordinates in the historical sequence, calculating the time interval of adjacent peak points and the mean value and standard deviation thereof, determining the weights of the two through Bayesian optimization and constructing a dynamic constraint window; based on the window, the similarity between the real-time sequence and the historical sequence is calculated by using a DTW algorithm, and an alarm is given when the similarity is less than a preset similarity threshold value; and the application can adapt to the complex changes of the bridge structure under different service stages and environmental conditions, and improve the accuracy of bridge health state monitoring.

Description

Technical Field

[0001] This application relates to the technical field of data monitoring and alarm, and in particular to a method and system for monitoring and alarming the health status of bridges. Background Technology

[0002] As a critical infrastructure in transportation networks, bridges' safety and reliability directly impact people's lives and property, as well as the stable operation of society. With the passage of time, bridge structures inevitably experience various forms of damage and performance degradation, such as cracks, deformation, and steel corrosion, due to factors including natural environmental changes (such as temperature variations, humidity, and earthquakes), traffic loads (long-term vehicle loads, overloading, etc.), and material aging. If these damages are not detected and addressed promptly, they may gradually develop, eventually leading to partial or even complete bridge collapse, causing severe and catastrophic consequences. Therefore, real-time and accurate monitoring of bridge health, timely detection of potential safety hazards, and implementation of corresponding measures are crucial for ensuring the safe operation of bridges.

[0003] Chinese invention patent application CN119984701A discloses a bridge monitoring method and system based on dynamic bridge patterns. This method acquires the dynamic strain response of the entire bridge under vehicle load excitation by constructing a grating array sensor network, and then constructs dynamic bridge patterns characterizing the bridge's structural properties. As strain change curves that vary along the bridge's mileage and characterize the inter-structural stress relationships, dynamic bridge patterns can reflect the overall structural characteristics of the bridge under specific loads. This method determines whether the bridge has problems such as abnormal traffic flow, heavy vehicles, sudden accidental impacts, or defects by comparing the characteristics of the dynamic bridge patterns (such as extreme values ​​and trends) with the baseline state.

[0004] The patented dynamic bridge pattern judgment criteria are relatively fixed (such as extreme value position difference and consistency of change trend), which makes it difficult to adapt to the complex changes in bridge structure under different service stages and environmental conditions, resulting in low flexibility. Summary of the Invention

[0005] In order to adapt to the complex changes in bridge structures under different service stages and environmental conditions, and to improve the accuracy of bridge health status monitoring, this application provides a bridge health status monitoring and alarm method and system.

[0006] Firstly, this application provides a bridge health status monitoring and alarm method, which adopts the following technical solution:

[0007] A bridge health status monitoring and alarm method includes the following steps:

[0008] Real-time and historical strain data from various monitoring points are collected, and the real-time and historical strain data belonging to the same monitoring point are integrated into a real-time sequence to construct a historical sequence containing strain data throughout the entire life cycle of the bridge.

[0009] Obtain the coordinates of peak points in the historical sequence, calculate the time interval between adjacent peak points based on the peak point coordinates, calculate the mean and standard deviation of the time interval, determine the weights of the mean and standard deviation through a Bayesian optimization algorithm, and calculate the dynamic constraint window based on the mean, standard deviation and their weights using a weighted summation algorithm.

[0010] Based on the dynamic constraint window, the DTW algorithm is used to calculate the similarity between the real-time sequence and the historical sequence. When the similarity is less than the preset similarity threshold, an alarm signal is issued.

[0011] This application first collects real-time and historical strain data from various monitoring points, integrating them into a real-time sequence, and constructs a historical sequence based on strain data throughout the bridge's entire lifecycle. This allows for more accurate capture of the changing trends and patterns of bridge strain through the accumulation of data over the entire lifecycle, revealing changes in strain characteristics caused by material aging, environmental erosion, and other factors during long-term use, thus helping to identify potential safety hazards in advance. Next, the coordinates of peak points in the historical sequence are identified, and the time intervals between adjacent peak points are calculated based on these coordinates. Furthermore, the mean and standard deviation of these time intervals are calculated. This application uses a Bayesian optimization algorithm to determine the weights of the mean and standard deviation, using the sum of the product of the mean, standard deviation, and their weights as a dynamic constraint window. This dynamic constraint window considers the time interval characteristics of peak points in the historical strain data, enabling it to adaptively adjust according to the actual strain patterns of the bridge, thus improving its applicability to different bridge structures.

[0012] This application uses a dynamic constraint window and the DTW (Dynamic Time Warping) algorithm to calculate the similarity between real-time and historical sequences. Since the DTW algorithm can handle the scaling and offset of time series on the time axis, this application can more accurately measure the similarity between real-time and historical sequences, which helps to detect abnormal bridge strain in a timely manner, adapt to the complex changes in bridge structures under different service stages and environmental conditions, and improve the accuracy of bridge health status monitoring.

[0013] Optionally, when the similarity is not less than a preset similarity threshold, the method further includes:

[0014] The strain deviation value is calculated based on real-time strain data and historical strain data, and a spatiotemporal correlation matrix is ​​constructed. The element in the i-th row and j-th column of the spatiotemporal correlation matrix represents the strain deviation value of the j-th monitoring point at time i.

[0015] Spatiotemporal coupling analysis is performed on the spatiotemporal correlation matrix to obtain analysis results. When the analysis results are abnormal, an early warning signal is issued.

[0016] This application calculates strain deviation values ​​based on real-time and historical strain data and constructs a spatiotemporal correlation matrix. This overcomes the limitations of analysis based solely on a single monitoring point or simple time series analysis, integrating the strain deviations of various monitoring points at different times into a single matrix. This allows for a direct visualization of the strain changes at different monitoring points at different times, providing a more comprehensive and detailed understanding of the overall strain state of the bridge. The element in the i-th row and j-th column of the spatiotemporal correlation matrix represents the strain deviation value of the j-th monitoring point at time i, achieving the fusion of temporal and spatial information. It reflects strain changes at different times in the temporal dimension and displays the locational differences of each monitoring point in the spatial dimension, helping to identify potential, localized strain anomalies.

[0017] This application utilizes spatiotemporal coupling analysis of the spatiotemporal correlation matrix to uncover deeper patterns and characteristics in strain deviation values. Traditional analysis methods may only detect obvious strain anomalies, while spatiotemporal coupling analysis considers the interaction between temporal and spatial factors. For example, strain deviation values ​​at certain monitoring points may exhibit periodic changes over time and be spatially correlated with other monitoring points. Spatiotemporal coupling analysis can identify such complex patterns, thus more accurately determining the presence of anomalies. When the analysis results indicate an anomaly, an early warning signal is issued. Compared to relying solely on similarity judgment, this application further improves the accuracy of anomaly warnings. Similarity judgment mainly focuses on the overall similarity between real-time and historical sequences, while spatiotemporal coupling analysis examines strain data changes from a more microscopic perspective. Combining the two can reduce the probability of false alarms and missed alarms. The spatiotemporal correlation matrix constructed based on historical strain data can be used for long-term trend analysis. By observing the changes in the spatiotemporal correlation matrix over different time periods, the development trend of bridge structural strain can be understood, and potential structural problems can be predicted.

[0018] Optionally, the spatiotemporal coupling analysis of the spatiotemporal correlation matrix includes:

[0019] In the time dimension, the duration, fluctuation frequency and change rate of the abnormal state at the same monitoring point are tracked. If the duration of the abnormal state exceeds the preset duration threshold, or the fluctuation frequency is greater than the preset fluctuation threshold, or the change rate is greater than the preset change threshold, it is determined that there is a trend of deterioration.

[0020] In the spatial dimension, if multiple monitoring points show abnormalities one after another within a preset time period, and the direction of abnormality propagation is consistent with the force transmission path of the bridge structure, it is determined that there is a structurally related abnormality.

[0021] When there is a trend of deterioration or structural correlation anomalies, the analysis results are considered abnormal.

[0022] This application identifies progressive damage in localized areas of bridge structures by tracking the duration, frequency, and rate of change of abnormal states at the same monitoring point and setting corresponding thresholds to determine trends of deterioration. For critical load-bearing components of a bridge, such as a section of the main beam, if abnormal strain persists for a duration exceeding a preset threshold, it may indicate fatigue damage or material degradation in that area. This application can promptly detect such potential deterioration trends, providing a basis for subsequent maintenance and repair.

[0023] Considering the scenario where multiple monitoring points successively exhibit anomalies within a preset time period, and the direction of anomaly propagation aligns with the force transmission path of the bridge structure, structurally correlated anomalies can be effectively identified. In bridge structures, force transmission follows a pattern. If the direction of anomaly propagation matches the force transmission path, it indicates a possible change in the overall stress state of the structure. For example, foundation settlement could lead to a redistribution of forces in the superstructure, or damage to a critical component could trigger abnormal stress in adjacent components. This application can grasp the overall health status of the bridge structure from a macroscopic perspective. When structurally correlated anomalies are detected, combining the direction of anomaly propagation with the location of monitoring points can roughly pinpoint the area where damage may occur.

[0024] Optionally, the spatiotemporal coupling analysis of the spatiotemporal correlation matrix further includes:

[0025] Set the spatiotemporal weight value for each anomaly monitoring point, including: assigning a time weight adjustment factor to anomaly monitoring points whose duration exceeds a preset duration threshold, clustering each row of elements in the spatiotemporal correlation matrix using a clustering algorithm, dividing anomaly monitoring points into anomaly regions based on the clustering results, and assigning a spatial weight adjustment factor to anomaly monitoring points in anomaly regions.

[0026] The spatiotemporal weight values ​​are updated based on the time weight adjustment factor and the spatial weight adjustment factor. When the updated spatiotemporal weight values ​​exceed the linkage threshold, the location of the abnormal area is marked and the time trajectory of the abnormal development is displayed.

[0027] This application assigns a time-weighted adjustment factor to abnormal monitoring points whose duration exceeds a preset threshold, fully considering the impact of the temporal continuity of abnormal states on structural health. By assigning a time-weighted adjustment factor, this application can highlight the importance of such long-term anomalies, making the analysis results more accurately reflect the actual health status of the structure and minimizing misjudgments caused by short-term fluctuations.

[0028] This application utilizes a clustering algorithm to cluster each row of elements in the spatiotemporal correlation matrix, and divides the abnormal monitoring points into abnormal regions based on the clustering results. Then, it assigns spatial weight adjustment factors to the abnormal monitoring points in the abnormal regions, taking into account the correlation between monitoring points in different locations in the structure. In bridge structures, anomalies in monitoring points in adjacent or related parts may have similar causes. By dividing abnormal regions through clustering and assigning spatial weights, the comprehensive impact of spatially related anomalies can be assessed more reasonably, thereby improving the accuracy of identifying structural problems.

[0029] The spatiotemporal weight values ​​are updated based on time and spatial weight adjustment factors. The updated values ​​integrate information from both dimensions, making the analysis results more accurate and reliable. Accurate anomaly location and time trajectory display provide strong support for structural maintenance decisions. Based on the severity and development trend of anomaly areas, targeted maintenance plans can be formulated, improving the focus and effectiveness of maintenance and ensuring the safe operation of the bridge structure.

[0030] Optionally, the calculation of the strain deviation value based on real-time strain data and historical strain data further includes:

[0031] Set a time window and use the time window to segment the historical strain data. Integrate the historical strain data of each time window into a dataset. Calculate the time decay factor based on the time difference between the historical strain data and the real-time strain data in each dataset. The larger the time difference, the larger the time decay factor.

[0032] Based on the time decay factor, the processed historical strain data is calculated using a weighted moving average algorithm, and the strain deviation value is calculated based on the real-time strain data and the processed historical strain data.

[0033] This application segments historical strain data by setting time windows, integrates the historical strain data of each time window into a dataset, and calculates a time decay factor based on the time difference. The larger the time difference, the larger the time decay factor, which means that the historical strain data further away from the real-time strain data has a smaller weight in subsequent calculations. The above scheme fully considers the timeliness of the data, because the state of the bridge structure may change over time, and the representativeness of earlier historical data to the current structural state gradually decreases.

[0034] The weighted moving average algorithm itself has the function of smoothing data. When preprocessing historical strain data, it takes into account multiple data points within a time window and calculates the weighted average by assigning different weights (i.e., time decay factors) to data at different time points. This reduces random fluctuations and noise interference in historical strain data, making the processed historical strain data more stable and reliable.

[0035] By using a weighted moving average algorithm to calculate processed historical strain data based on a time decay factor, and then combining this data with real-time strain data to calculate the strain deviation value, the accuracy of strain deviation calculation can be improved. Since the processed historical strain data better reflects the current structural state of the bridge and removes some noise and abnormal fluctuations, the strain deviation value calculated by comparing it with real-time strain data more accurately reflects the changes in structural strain. The time decay factor gives greater weight to recent historical data in the calculation, helping to highlight the recent trend of strain changes in the bridge structure. When calculating the strain deviation value, it can more sensitively capture subtle changes in the structural state and promptly detect potential anomalies.

[0036] Optionally, before constructing the spatiotemporal correlation matrix, the method further includes:

[0037] A three-dimensional model of the bridge was established using finite element software. The temperature field distribution and structural response were simulated using the three-dimensional model of the bridge. The temperature strain coefficient at each monitoring point was calculated. The strain deviation value was corrected using the temperature strain coefficient, and the corrected strain deviation value was recorded as the new strain deviation value.

[0038] This application establishes a three-dimensional model of the bridge using finite element method (FEM) software, which can accurately simulate the temperature field distribution of the bridge under different environmental conditions. Based on the simulated temperature field distribution and structural response, the temperature strain coefficient at each monitoring point is calculated. This coefficient clearly defines the contribution of temperature changes to the strain at each monitoring point. This application uses the temperature strain coefficient to correct the strain deviation value, and the corrected strain deviation value is recorded as the new strain deviation value. Through correction, this application can weaken the interference of temperature on the strain deviation value, making the new strain deviation value more accurately reflect the real strain changes of the bridge structure caused by loads, damage, and other factors, thereby improving the accuracy and reliability of the data.

[0039] In bridge structural health monitoring, strain changes caused by temperature variations are often superimposed on strain changes caused by structural damage or loads, making them difficult to distinguish. By correcting the strain deviation value, this application can make the data from different monitoring periods more consistent and comparable. When conducting long-term monitoring data analysis, it is possible to more clearly observe the changing trends of structural performance, providing a more reliable basis for bridge safety assessment and maintenance decisions.

[0040] Constructing a spatiotemporal correlation matrix requires accurate and reliable strain data as a foundation. The corrected strain deviation values ​​can more realistically reflect the spatiotemporal correlation between various monitoring points of the bridge structure, providing high-quality data input for the construction of the spatiotemporal correlation matrix and helping to discover potential abnormal areas and problem patterns in the structure.

[0041] Optionally, the method further includes:

[0042] The wavelet packet decomposition algorithm is used to perform multi-scale frequency domain analysis on the strain deviation value, decompose the strain deviation value into characteristic components of different frequency bands, and extract the deviation energy ratio of each frequency band.

[0043] Sensitivity coefficients are set for each frequency band based on the inherent frequency parameters of the bridge structure. The deviation energy ratio of each frequency band is weighted and calculated based on the sensitivity coefficients to obtain a comprehensive anomaly index. When the comprehensive anomaly index exceeds the preset index threshold, it is determined that the frequency characteristics of the strain deviation value are related to the structural resonance risk, and an early warning signal is issued.

[0044] This application employs a wavelet packet decomposition algorithm to perform multi-scale frequency domain analysis on strain deviation values. This decomposes the strain deviation values ​​into characteristic components of different frequency bands, thereby more comprehensively capturing the characteristics of strain deviation values ​​at different frequencies. Subsequently, this application quantifies the contribution of different frequency bands to the strain deviation values ​​by extracting the deviation energy proportion of each frequency band. The energy proportion reflects the signal strength of each frequency band. By analyzing the changes in the deviation energy proportion of each frequency band, anomalies in the frequency domain of the strain deviation values ​​can be detected in a timely manner.

[0045] This application sets sensitivity coefficients for each frequency band based on the natural frequency parameters of the bridge structure, taking into account the dynamic characteristics of the bridge structure itself. Different bridges have different natural frequencies due to differences in their structural form, materials, dimensions, and other factors. Correlating the sensitivity coefficients with the natural frequencies makes the weighted calculations more closely reflect the actual situation of the bridge and improves the ability to identify structural resonance risks. Subsequently, this application performs weighted calculations on the proportion of deviation energy in each frequency band based on the sensitivity coefficients to obtain a comprehensive anomaly index, which can comprehensively consider the contribution of different frequency bands to the structural resonance risk. The comprehensive anomaly index can comprehensively reflect the anomalies of strain deviation values ​​in multiple frequency bands, providing a comprehensive quantitative indicator for judging whether the structure is in a state of resonance risk. When the comprehensive anomaly index exceeds the preset index threshold, it is determined that the frequency characteristics of the strain deviation value are related to the structural resonance risk, and an early warning signal is issued, thereby timely detecting the potential resonance risk of the bridge structure and buying time for taking corresponding measures.

[0046] Optionally, the method further includes:

[0047] A trend identification window is set, and a sliding window linear regression algorithm is used to calculate the rate of change of the deviation energy ratio of each frequency band within the trend identification window. If the rate of change of the current frequency band in n consecutive trend identification windows is greater than the preset rate of change threshold, it is determined that the deviation energy ratio of the current frequency band is on a continuous upward trend, and the sensitivity coefficient of the current frequency band is increased.

[0048] This application, by setting a trend identification window and employing a sliding window linear regression algorithm, can dynamically monitor the changes in the proportion of deviation energy in each frequency band over a period of time. Compared with static analysis, this application can track the changing trends of frequency band characteristics in real time and promptly detect potential anomalies. If the rate of change of the current frequency band in n consecutive trend identification windows is greater than a preset rate of change threshold, it is determined that the proportion of deviation energy in the current frequency band is showing a continuous upward trend, thereby improving the accuracy and reliability of trend judgment and minimizing misjudgments caused by accidental fluctuations in a single window. When it is determined that the proportion of deviation energy in the current frequency band is showing a continuous upward trend, the sensitivity coefficient of the current frequency band is increased. Optimizing the sensitivity coefficient setting helps improve the calculation accuracy of the comprehensive anomaly index, thereby improving the accuracy of early warning.

[0049] The comprehensive anomaly index is calculated by weighting the proportion of deviation energy in each frequency band with a sensitivity coefficient. Appropriate adjustment of the sensitivity coefficient can make the comprehensive anomaly index more accurately reflect the correlation between the frequency characteristics of strain deviation values ​​and structural resonance risk. When the proportion of deviation energy in a certain frequency band shows a continuous upward trend, increasing its sensitivity coefficient can make the comprehensive anomaly index reflect this change more promptly, thereby issuing a more accurate early warning signal.

[0050] Optionally, the method further includes:

[0051] The coordinates of the maximum value point are obtained from the coordinates of the peak point. The coordinates of the maximum value point are then input into the three-dimensional model of the bridge for simulation analysis to obtain the theoretical stress distribution cloud map of each monitoring point.

[0052] Real-time strain data is converted into actual stress distribution cloud maps. The overlap between theoretical stress distribution cloud maps and actual stress distribution cloud maps is calculated. When the overlap is lower than the overlap threshold, an alarm signal is issued.

[0053] This application uses the coordinates of the maximum value points in the historical sequence to simulate and analyze the bridge's three-dimensional model. Based on the actual structural parameters and stress conditions of the bridge, it can obtain more accurate theoretical stress distribution cloud maps of each monitoring point. By analyzing the theoretical stress distribution cloud maps corresponding to the maximum value points, it is possible to identify the stress concentration areas and potential weak points of the bridge structure under these extreme conditions in advance.

[0054] Converting real-time strain data into an actual stress distribution cloud map can reflect the current stress distribution of the bridge structure in real time and intuitively. Strain data is a direct reflection of the stress state of the bridge structure. By converting it into a stress distribution cloud map, the magnitude and distribution range of stress in various parts of the bridge can be displayed more clearly.

[0055] By calculating the overlap between the theoretical and actual stress distribution cloud maps and setting an overlap threshold, anomalies in the bridge structure can be accurately identified. When the overlap is below the threshold, it indicates a significant deviation between the actual and theoretical stress distributions, potentially suggesting damage, deformation, or a change in the stress state of the bridge structure. This overlap-based judgment method offers strong objectivity and accuracy, reducing misjudgments caused by anomalies in data from a single monitoring point.

[0056] Secondly, this application provides a bridge health status monitoring and alarm system, which adopts the following technical solution:

[0057] A bridge health status monitoring and alarm system includes: a memory and a processor.

[0058] The memory contains a computer-readable storage medium;

[0059] When the processor processes a computer program stored on the computer-readable storage medium, it implements the method as described in the first aspect.

[0060] In summary, this application includes at least one of the following beneficial technical effects:

[0061] 1. This application first collects real-time and historical strain data from various monitoring points, integrating them into a real-time sequence, and constructs a historical sequence based on strain data throughout the bridge's entire life cycle. This allows for more accurate capture of the changing trends and patterns of bridge strain through the accumulation of data over the entire life cycle, revealing changes in strain characteristics caused by material aging, environmental erosion, and other factors during long-term use, thus helping to identify potential safety hazards in advance. Next, the coordinates of peak points in the historical sequence are identified, and the time interval between adjacent peak points is calculated based on these coordinates. Furthermore, the mean and standard deviation of the time intervals are calculated. This application uses a Bayesian optimization algorithm to determine the weights of the mean and standard deviation, using the sum of the product of the mean, standard deviation, and their weights as a dynamic constraint window. This dynamic constraint window considers the time interval characteristics of peak points in the historical strain data, enabling the constraint window to adaptively adjust according to the actual strain patterns of the bridge, thus improving its applicability to different bridge structures.

[0062] 2. This application uses the Dynamic Time Warping (DTW) algorithm based on a dynamic constraint window to calculate the similarity between real-time and historical sequences. Since the DTW algorithm can handle the scaling and offset of time series on the time axis, this application can more accurately measure the similarity between real-time and historical sequences, which helps to detect abnormal bridge strain in a timely manner, adapt to the complex changes in bridge structures under different service stages and environmental conditions, and improve the accuracy of bridge health status monitoring. Attached Figure Description

[0063] Figure 1 This is a flowchart of Embodiment 1 of this application;

[0064] Figure 2 This is a flowchart of Embodiment 2 of this application;

[0065] Figure 3 This is a flowchart of Embodiment 3 of this application. Detailed Implementation

[0066] The following combination Figures 1 to 3 This application will be described in further detail.

[0067] Example 1: This example discloses a bridge health status monitoring and alarm method, referring to... Figure 1 The method includes: S11 data acquisition and processing, S12 setting a dynamic constraint window, and S13 alarm analysis. First, real-time strain data and historical strain data from each monitoring point are collected and integrated into a real-time sequence. A historical strain data sequence for the entire life cycle of the bridge is constructed. The coordinates of peak points in the historical sequence are obtained, and the time interval between adjacent peak points, their mean, and standard deviation are calculated. The weights of the two data points are determined through Bayesian optimization, and a dynamic constraint window is constructed. Based on this window, the DTW algorithm is used to calculate the similarity between the real-time sequence and the historical sequence, and an alarm is triggered when the similarity is less than a preset similarity threshold. The execution process of each step in this embodiment is as follows:

[0068] S11 Data Acquisition and Processing: Vibrating wire strain gauges are installed at various monitoring points on the bridge. These strain gauges can collect real-time strain data of the bridge structure under load.

[0069] The real-time strain data collected by the vibrating wire strain gauge is not only used for current analysis, but is also stored on local storage devices or cloud databases to form historical strain data. The data is organized in chronological order during storage, and an independent data file or database table is created for each monitoring point to record all strain data from the start of monitoring to the current moment.

[0070] Historical strain data for each monitoring point is acquired. Real-time strain data and historical strain data belonging to the same monitoring point are plotted on the same coordinate system with time as the horizontal axis and strain data as the vertical axis to form a real-time sequence. The real-time sequence can intuitively show the strain state of a certain monitoring point in the bridge structure at the current moment and its changing trend over time.

[0071] Strain data from various monitoring points on similar bridges are continuously collected from the construction phase to the demolition or reconstruction phase, covering the strain response of the bridge under different design loads and environmental conditions (such as temperature, humidity, wind speed, etc.). The collected life-cycle strain data from each monitoring point are organized and analyzed in chronological order to obtain a historical sequence of strain data for each monitoring point that includes the entire life cycle of the bridge. This historical sequence can reflect the strain change pattern of a certain monitoring point of the bridge structure at different stages.

[0072] S12 sets a dynamic constraint window. First, the historical sequence is preprocessed, such as through smoothing filtering to remove noise interference. Then, the historical strain data of the preprocessed historical sequence is mapped to the grayscale values ​​of the image. An optimal threshold is calculated using the Otsu algorithm to divide the historical sequence into peak and non-peak regions. Points within the peak region are the peak points, and their coordinates are further obtained.

[0073] Based on the identified peak point coordinates, the time interval between adjacent peak points is calculated. The time interval reflects the periodic characteristics of the strain change of the bridge structure. The time interval between adjacent peak points may vary for bridges with different structures or under different working conditions.

[0074] Statistical analysis was performed on the calculated time intervals between adjacent peak points to calculate their mean and standard deviation. The mean represents the average level of the time intervals between adjacent peak points, reflecting the average period of strain change in the bridge structure; the standard deviation measures the dispersion of the time intervals. The larger the standard deviation, the greater the fluctuation of the time intervals, and the more unstable the strain change of the bridge structure.

[0075] Bayesian optimization is an efficient global optimization algorithm used to determine the weights of the mean and standard deviation. This algorithm constructs a probabilistic model to describe the relationship between the objective function and its parameters, continuously updating the probabilistic model based on existing observation data to gradually approach the optimal weight values. The process is as follows:

[0076] The evaluation metric for the dynamic constraint window is to maximize the F1 score of the similarity detection. The calculation model for the F1 score is as follows:

[0077] ;

[0078] in, False alarm rate; This represents the underreporting rate.

[0079] The optimization variables are the mean weight and the standard deviation weight, which satisfy the condition that their sum is 1.

[0080] Initialization: Randomly generate initial weight combinations.

[0081] Proxy model construction: Use Gaussian process (GP) to fit the relationship between the objective function and the weights to capture nonlinear features.

[0082] A Gaussian process (GP) is a set of random variables, where any finite number of these random variables follow a joint Gaussian distribution. In surrogate models, GP is used to model the objective function. With weight Relationship:

[0083] ;

[0084] in, It is a mean function, and its value is 0; Let covariance function be defined at two points in the parameter space. and The correlation is calculated using radial basis functions:

[0085] ;

[0086] in, is the sequence variance, which is the coefficient of the diagonal elements of the covariance matrix, representing the overall fluctuation intensity of the function values ​​in the parameter space; L is the sequence length; This represents the noise variance, used to characterize observation noise.

[0087] Acquisition function optimization: Select the next set of weights through Expected Improvement (EI), balance exploration (high uncertainty region) and utilization (high mean region), and find the weight combination that maximizes the expected improvement.

[0088] Iterative update: Validate the above weight combination in a real environment, update the surrogate model until convergence (such as reaching the maximum number of iterations or the change in the objective function is less than the threshold), and output the final weight combination.

[0089] The sum of the calculated mean, standard deviation, and the final weight combination product is used as the dynamic constraint window. The calculation model for the dynamic constraint window is as follows:

[0090] ;

[0091] in, and These are the weights of the mean and standard deviation in the final weighted combination; d is the dynamic constraint window. The mean of the time intervals; denoted as the standard deviation of the time interval.

[0092] The dynamic constraint window can be automatically adjusted according to the historical strain change characteristics of the bridge structure, which is highly flexible and adaptable.

[0093] The S13 alarm analysis uses the DTW algorithm to calculate the similarity between real-time and historical sequences. The DTW algorithm employs a dynamic constraint window to find the optimal alignment path between the two sequences, minimizing the sum of distances between corresponding points, thus calculating the similarity value. A similarity value closer to 1 indicates greater similarity between the two sequences; a similarity value closer to 0 indicates greater difference between the two sequences.

[0094] The DTW algorithm uses a dynamic constraint window to find the optimal alignment path between two sequences. This means that on the time axis, the a-th point of the real-time sequence can only be aligned with the max(1,ad) to min(b,a+d)-th historical strain data in the historical sequence, where b is the number of historical strain data in the historical sequence.

[0095] Based on the dynamic constraint window, a constrained distance matrix is ​​constructed for the a-th real-time strain data in the real-time sequence. and the c-th historical strain data in the historical sequence The calculation is performed only when c satisfies max(1,ad)≤c≤min(b,a+d). and The distance between them is calculated and stored in the distance matrix.

[0096] A path from the element in the first row and first column of the distance matrix to the element in the w-th row and b-th column is found using dynamic programming, minimizing the total distance along the path, where w is the number of real-time strain data points in the real-time sequence. The minimum total distance is then converted into a similarity score using the following formula:

[0097] ;

[0098] Where S represents the similarity; The minimum total distance; This is the sum of distances between all possible point pairs without considering the dynamic constraint window.

[0099] The calculated similarity between the real-time and historical sequences is compared with a preset similarity threshold. This threshold is determined based on the bridge's structural characteristics, historical data, and engineering experience, and ranges from 0.7 to 0.9. When the similarity is less than the preset threshold, it indicates a significant difference between the real-time and historical sequences, suggesting an abnormal strain state in the bridge structure. An alarm signal is then issued, which can be transmitted to relevant personnel through various means, such as SMS, email, and audible / visual alarms. The alarm information includes the monitoring point location, similarity value, and alarm time.

[0100] By adopting the above scheme, this embodiment can realize real-time monitoring, anomaly analysis and alarm of the strain state of bridge structure, providing strong protection for the safe operation of bridge.

[0101] Example 2: Refer to Figure 2 The difference between this embodiment and Embodiment 1 is that, when the similarity is not less than a preset similarity threshold, the method further includes:

[0102] S21 calculates the strain deviation. Through the path in the alarm analysis of S13, the correspondence between each real-time strain data in the real-time sequence and the historical strain data in the historical sequence is obtained. The process is as follows:

[0103] Starting from the bottom right corner C[w][b] of the cumulative distance matrix, backtracking is performed to determine the path direction according to the following rules:

[0104] If C[a][c] comes from C[a-1][c-1], then the a-th element of the real-time sequence is aligned with the c-th element of the historical sequence.

[0105] If C[a][c] comes from C[a-1][c], then the a-th element of the real-time sequence is aligned with an element before the c-th element in the historical sequence. The final result is determined by the path chosen during the backtracking process from the lower right corner of the cumulative distance matrix to the upper left corner.

[0106] If C[a][c] comes from C[a][c-1], then an element before the a-th element of the real-time sequence is aligned with the c-th element of the historical sequence.

[0107] By continuously backtracking until reaching the top left corner C[0][0] of the matrix, the complete alignment path is obtained.

[0108] Based on the correspondence in the path, calculate the strain deviation between the real-time strain data and the corresponding historical strain data.

[0109] The spatiotemporal correlation matrix is ​​a two-dimensional matrix. Given n monitoring points and m time intervals, the spatiotemporal correlation matrix is ​​an m×n matrix, where the element in the i-th row and j-th column represents the strain deviation value of the j-th monitoring point at time i. Following the chronological order and monitoring point number, the calculated strain deviation values ​​for each monitoring point at each time interval are sequentially filled into the spatiotemporal correlation matrix. The completed spatiotemporal correlation matrix simultaneously reflects the temporal and spatial distribution of the strain deviation.

[0110] S22 analysis involves spatiotemporal coupling analysis of the spatiotemporal correlation matrix, including:

[0111] S221 Time Dimension Analysis: In the time dimension, it tracks the duration, fluctuation frequency, and rate of change of abnormal states at the same monitoring point. The calculation process for the duration, fluctuation frequency, and rate of change is as follows:

[0112] For each monitoring point, the times when its strain deviation value exceeds the preset normal deviation threshold are found in the spatiotemporal correlation matrix, and the length of the continuous time period of these abnormal times is calculated. For example, if the strain deviation value of a certain monitoring point exceeds the threshold from time t1 to time t2, the duration is equal to the difference between t2 and t1.

[0113] The fluctuation frequency is calculated by counting the number of times the strain deviation value of a certain monitoring point exceeds the threshold within a certain time range and the total number of measurements.

[0114] Calculate the ratio of the change in strain deviation value between adjacent time points to the time interval, and use this ratio as the rate of change.

[0115] The calculated duration, fluctuation frequency, and rate of change are compared with preset duration thresholds, fluctuation thresholds, and change thresholds, respectively. If the duration exceeds the preset duration threshold, the fluctuation frequency is greater than the preset fluctuation threshold, or the rate of change is greater than the preset change threshold, then the monitoring point is determined to have a trend of deterioration.

[0116] S222 Spatial Dimension Analysis: In terms of spatial dimension, if multiple monitoring points successively exhibit anomalies within a preset time period, and the direction of anomaly propagation is consistent with the force transmission path of the bridge structure, then it is determined that there is a structurally correlated anomaly. The process is as follows:

[0117] Within a preset time period, observe the strain deviation of each monitoring point in the spatiotemporal correlation matrix. If more than a preset number of monitoring points are found to have strain deviations exceeding the threshold, record the location of these abnormal monitoring points and the order in which they appear.

[0118] By considering the structural force transmission path of the bridge, we analyze whether the direction of anomaly propagation at these monitoring points is consistent with this path. For example, for a simply supported beam bridge, under load, strain anomalies typically propagate from the load application point towards the supports. If the direction of anomaly propagation at multiple monitoring points matches this force transmission path, then a structurally correlated anomaly is identified.

[0119] The S223 warning is issued when a trend of deterioration is determined in the time dimension or a structural correlation anomaly is determined in the spatial dimension. In this case, the spatiotemporal coupling analysis result is considered abnormal, and a warning signal is sent to the operation and maintenance personnel.

[0120] Spatiotemporal coupling analysis of the spatiotemporal correlation matrix also includes:

[0121] S224 sets the spatiotemporal weights, specifying the spatiotemporal weight value for each anomaly monitoring point, including:

[0122] A preset duration threshold is set for each anomaly monitoring point. For each anomaly monitoring point, the duration of its abnormal state is continuously monitored. When the duration exceeds the preset duration threshold, a time weight adjustment factor is assigned to the monitoring point.

[0123] The time weight adjustment factor can be calculated based on the ratio of duration to a preset duration threshold. The calculation model for the time weight adjustment factor is as follows:

[0124] ;

[0125] Where A is the time weight adjustment factor; The duration of the abnormal state; The preset duration threshold is denoted by k; k is an adjustment coefficient used to control the rate at which the time weight adjustment factor increases with the duration, and its value ranges from 0.2 to 0.5. This represents the maximum duration of the abnormal state.

[0126] The elements in each row of the spatiotemporal correlation matrix are clustered using either hierarchical clustering or DBSCAN algorithms. Based on the results of the clustering algorithms, the abnormal monitoring points are divided into different abnormal regions.

[0127] For each anomalous region, a spatial weight adjustment factor is assigned based on factors such as the number and density of anomalous monitoring points within that region. In this embodiment, the more monitoring points there are within the anomalous region and the higher their density, the larger the spatial weight adjustment factor becomes.

[0128] S225 updates the weights, updating the spatiotemporal weight values ​​based on the time weight adjustment factor and the spatial weight adjustment factor. When the updated spatiotemporal weight values ​​exceed the linkage threshold, the location of the abnormal area is marked, and the time trajectory of the abnormal development is displayed.

[0129] A weighted summation algorithm is used to update the spatiotemporal weight value of each anomaly monitoring point based on time and spatial weight adjustment factors. The calculation formula is as follows:

[0130] ;

[0131] Where α and β are weighting coefficients used to balance the influence of time weight and spatial weight on the spatiotemporal weight value, and α+β=1; A is the time weight adjustment factor; B is the spatial weight adjustment factor.

[0132] When the spatiotemporal weight value of an anomaly monitoring point in an abnormal area exceeds the linkage threshold, the location of the abnormal area is marked on a map or other visualization interface, and the spatiotemporal weight value of each anomaly monitoring point at different times is recorded. These data are arranged in chronological order, and a curve of the anomaly weight value changing over time is plotted, which is the time trajectory of the anomaly development.

[0133] Example 3: Reference Figure 3 The difference between this embodiment and Embodiment 2 is that the calculation of the strain deviation value based on real-time strain data and historical strain data includes:

[0134] S31 data segmentation: Set a time window and use the time window to segment the historical strain data. For example, if the time window size is 1 hour and the sampling rate of the historical strain data is once per minute, then each time window will contain 60 historical strain data.

[0135] S32 calculates the time decay factor by integrating historical strain data from each time window into a dataset. The time decay factor is calculated based on the time difference between the historical and real-time strain data in each dataset. The time decay factor reflects the temporal correlation between historical and real-time strain data; the larger the time difference, the smaller the impact of historical data on the current state, and thus the larger the time decay factor. The calculation process for the time decay factor is as follows:

[0136] For each time window, the time difference between the historical strain data and the real-time strain data is calculated. The time difference is equal to the timestamp of the real-time strain data minus the timestamp of the historical strain data. Based on the time difference, the time decay factor is calculated using the following formula:

[0137] ;

[0138] Where E is the time decay factor; The attenuation coefficient is... , The larger the value, the faster the decay. denoted as , where is the time difference; e is the base of the natural logarithm.

[0139] S33 updates historical data by using a weighted moving average algorithm based on a time decay factor to calculate the processed historical strain data. The calculation formula is as follows:

[0140] ;

[0141] in, This refers to the processed historical strain data within the u-th time window; For the f-th historical strain data in the u-th time window The time decay factor; g is the number of historical strain data in the u-th time window.

[0142] The strain deviation value is calculated based on real-time strain data and processed historical strain data.

[0143] S34 Correction: When bridge materials are affected by temperature changes, they undergo thermal expansion and contraction deformation. If the deformation is constrained (e.g., fixed at both ends of the bridge), internal thermal stress will be generated, leading to strain changes. This embodiment corrects the strain deviation value by calculating the temperature strain coefficient at each monitoring point. The temperature strain coefficient can be understood as the amount of strain change caused by a unit temperature change. This embodiment uses finite element software (such as ANSYS, ABAQUS) to establish a three-dimensional model of the bridge, simulate the temperature field distribution and structural response, and calculate the temperature strain coefficient. The specific process is as follows:

[0144] Based on the bridge design drawings, establish the bridge's geometric model (including main beams, piers, supports, etc.), define material properties (linear expansion coefficient, elastic modulus, Poisson's ratio, etc.), input measured or predicted temperature data (such as solar radiation temperature, seasonal temperature changes), for non-uniform temperature fields, define temperature gradients (such as temperature difference between the top and bottom of the beam), constrain support positions (such as fixed hinge supports, rolling supports), consider vehicle loads, wind loads, etc., run thermal-structural coupling analysis, obtain strain distribution, extract strain-temperature curves at monitoring points, and fit the temperature strain coefficient.

[0145] The strain deviation value is corrected using a temperature strain coefficient, and the corrected strain deviation value is recorded as the new strain deviation value. The calculation model for the new strain deviation value is as follows:

[0146] ;

[0147] in, This is the new strain deviation value; This represents the strain deviation value; This is the temperature strain coefficient.

[0148] S35 simulation analysis maps peak points (time-location-amplitude) in the historical sequence to equivalent nodal forces according to their location. Fixed / elastic constraints are set according to the bridge bearing type to simulate the actual boundary. The simulated actual boundary is input as the boundary condition into the finite element software. Based on static / dynamic analysis (such as ANSYS, MIDAS), the theoretical stress distribution of each monitoring point under a given load is calculated, and a color cloud map (RGB encoded stress value) is generated, which is the theoretical stress distribution cloud map.

[0149] Real-time strain data is converted into stress through the material's elastic modulus, and point cloud data is generated according to the spatial coordinates of the monitoring points. Kriging interpolation or inverse distance weighting (IDW) is used to expand the discrete point data into an actual stress distribution cloud map, maintaining the same resolution as the theoretical stress distribution cloud map.

[0150] The degree of similarity of stress values ​​between two contour maps in space is calculated using the Structural Similarity Index (SSIM) or Root Mean Square Error (RMSE). When the degree of similarity exceeds the threshold, an alarm signal is issued.

[0151] Example 4: This example differs from Example 2 in that the method further includes:

[0152] The strain deviation value reflects the difference between the structural strain and the reference value (such as the design value or historical average). The strain deviation values ​​are arranged in chronological order according to the timestamps of the historical strain data to form a time series. Using the wavelet packet decomposition algorithm, the time series is decomposed into sub-signals of different frequency bands. Each sub-signal corresponds to a specific frequency range, called a wavelet packet node. Each wavelet packet node corresponds to a set of wavelet packet coefficients. These wavelet packet coefficients reflect the characteristic components of the time series in that frequency band.

[0153] For each wavelet packet node, the energy of the wavelet packet coefficients is calculated. The formula for calculating the energy of the wavelet packet coefficients is:

[0154] ;

[0155] in, The energy of the coefficients of the p-th wavelet packet; It is the coefficient of the qth wavelet packet of the p-th wavelet packet node.

[0156] The calculation model for the percentage of deviation energy in each frequency band is as follows:

[0157] ;

[0158] in, The deviation energy percentage of the p-th wavelet packet; The energy sum of the time series is equal to the sum of the energies of all wavelet packet coefficients.

[0159] The modal parameters of the bridge structure are analyzed by finite element model to obtain the first few natural frequencies (such as the first vertical bending frequency and the first torsional frequency). Based on the natural frequency parameters of the bridge structure, sensitivity coefficients for each frequency band are set. High sensitivity coefficients are assigned to frequency bands close to the resonant frequency range of the bridge structure, and low sensitivity coefficients are assigned to frequency bands far from the resonant frequency range.

[0160] The comprehensive anomaly index is obtained by weighting the proportion of deviation energy in each frequency band based on the sensitivity coefficient. The comprehensive anomaly index reflects the energy weight of the frequency component related to resonance risk in the strain deviation value. The larger the value of the comprehensive anomaly index, the higher the resonance risk. The calculation model of the comprehensive anomaly index is as follows:

[0161] ;

[0162] Where I represents the comprehensive anomaly index; Let be the percentage of the bias energy of the r-th wavelet packet; Let be the sensitivity coefficient of the r-th wavelet packet.

[0163] When the comprehensive anomaly index exceeds the preset index threshold, it is determined that the frequency characteristics of the strain deviation value are related to the structural resonance risk, and an early warning signal is issued.

[0164] A trend identification window is set, which should cover the typical cycle of the dynamic response of the bridge structure, and its value is equal to the ratio of the preset number of sampling points to the sampling frequency.

[0165] The sliding step size is set to half or one-third of the trend recognition window. The sliding window linear regression algorithm is used to calculate the rate of change of the deviation energy ratio within the trend recognition window for each frequency band. The process is as follows:

[0166] The deviation energy percentage of each frequency band within the trend identification window is extracted, and a linear regression model is constructed to fit the deviation energy percentage of the current trend identification window using the least squares method to obtain the rate of change of the deviation energy percentage of each frequency band within the trend identification window. In this embodiment, the linear regression model uses a linear function, with time as the independent variable and the deviation energy percentage as the dependent variable. The rate of change of the deviation energy percentage of each frequency band within the trend identification window is the coefficient of the first term of the linear function.

[0167] If the rate of change of the current frequency band is greater than the preset rate of change threshold for n consecutive trend recognition windows, it is determined that the deviation energy ratio of the current frequency band is on a continuous upward trend, and the sensitivity coefficient of the current frequency band is increased, for example, by 0.1.

[0168] Example 5: This example discloses a bridge health status monitoring and alarm system, the system including: a memory and a processor.

[0169] The memory contains a computer-readable storage medium;

[0170] When the processor processes the computer program stored on the computer-readable storage medium, it implements the bridge health status monitoring and alarm method.

[0171] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A bridge health state monitoring alarm method, characterized by, include: Real-time and historical strain data from various monitoring points are collected, and the real-time and historical strain data belonging to the same monitoring point are integrated into a real-time sequence to construct a historical sequence containing strain data throughout the entire life cycle of the bridge. Obtain the coordinates of peak points in the historical sequence, calculate the time interval between adjacent peak points based on the peak point coordinates, calculate the mean and standard deviation of the time interval, determine the weights of the mean and standard deviation through a Bayesian optimization algorithm, and calculate the dynamic constraint window based on the mean, standard deviation and their weights using a weighted summation algorithm. Based on the dynamic constraint window, the DTW algorithm is used to calculate the similarity between real-time sequences and historical sequences. When the similarity is less than the preset similarity threshold, an alarm signal is issued. When the similarity is not less than a preset similarity threshold, the strain deviation value is calculated based on real-time strain data and historical strain data, including: setting a time window, segmenting the historical strain data using the time window, integrating the historical strain data of each time window into a dataset, calculating the time decay factor based on the time difference between the historical strain data and the real-time strain data in each dataset, the larger the time difference, the larger the time decay factor; based on the time decay factor, calculating the processed historical strain data using a weighted moving average algorithm, and calculating the strain deviation value based on the real-time strain data and the processed historical strain data; Construct a spatiotemporal correlation matrix, wherein the element in the i-th row and j-th column of the spatiotemporal correlation matrix represents the strain deviation value of the j-th monitoring point at time i; Spatiotemporal coupling analysis of the spatiotemporal correlation matrix includes: In the time dimension, the duration, fluctuation frequency and change rate of the abnormal state at the same monitoring point are tracked. If the duration of the abnormal state exceeds the preset duration threshold, or the fluctuation frequency is greater than the preset fluctuation threshold, or the change rate is greater than the preset change threshold, it is determined that there is a trend of deterioration. In the spatial dimension, if multiple monitoring points show abnormalities one after another within a preset time period, and the direction of abnormality propagation is consistent with the force transmission path of the bridge structure, it is determined that there is a structurally related abnormality. Set the spatiotemporal weight value for each anomaly monitoring point, including: assigning a time weight adjustment factor to anomaly monitoring points whose duration exceeds a preset duration threshold, clustering each row of elements in the spatiotemporal correlation matrix using a clustering algorithm, dividing anomaly monitoring points into anomaly regions based on the clustering results, and assigning a spatial weight adjustment factor to anomaly monitoring points in anomaly regions. The spatiotemporal weight values ​​are updated based on the time weight adjustment factor and the spatial weight adjustment factor. When the updated spatiotemporal weight values ​​exceed the linkage threshold, the location of the abnormal area is marked and the time trajectory of the abnormal development is displayed. When there is a trend of deterioration or structural correlation anomalies, the analysis results are abnormal, and an early warning signal is issued.

2. The bridge health monitoring alarm method of claim 1, wherein, Before constructing the spatiotemporal correlation matrix, the method further includes: A three-dimensional model of the bridge was established using finite element software. The temperature field distribution and structural response were simulated using the three-dimensional model of the bridge. The temperature strain coefficient at each monitoring point was calculated. The strain deviation value was corrected using the temperature strain coefficient, and the corrected strain deviation value was recorded as the new strain deviation value.

3. The bridge health monitoring alarm method of claim 1, wherein, The method further includes: The wavelet packet decomposition algorithm is used for multi-scale frequency domain analysis of the strain deviation value, the strain deviation value is decomposed into characteristic components of different frequency bands, and the deviation energy proportion of each frequency band is extracted; According to the bridge structure inherent frequency parameter, the sensitive coefficient of each frequency band is set, the deviation energy proportion of each frequency band is weighted calculated based on the sensitive coefficient, and the comprehensive abnormal index is obtained; when the comprehensive abnormal index exceeds the preset index threshold, it is determined that the frequency characteristics of the strain deviation value are related to the structure resonance risk, and a warning signal is sent.

4. The bridge health monitoring alarm method of claim 3, wherein, The method further comprises: A trend identification window is set, a sliding window linear regression algorithm is used to calculate the change rate of the deviation energy proportion of each frequency band in the trend identification window, and if the change rate of the current frequency band in the continuous n trend identification windows is greater than the preset change rate threshold, it is determined that the deviation energy proportion of the current frequency band shows a continuous upward trend, and the sensitive coefficient of the current frequency band is increased.

5. The bridge health monitoring alarm method of claim 2, wherein, The method further comprises: According to the maximum value point coordinates, the maximum value point coordinates are input into the bridge three-dimensional model for simulation analysis, and the theoretical stress distribution cloud diagram of each monitoring point is obtained; The real-time strain data is converted into the actual stress distribution cloud diagram, the coincidence degree of the theoretical stress distribution cloud diagram and the actual stress distribution cloud diagram is calculated, and when the coincidence degree is lower than the coincidence degree threshold, an alarm signal is sent.

6. A bridge health monitoring and alert system, characterized by, It comprises: a memory and a processor, The memory has a computer readable storage medium stored therein; When the processor processes the computer program stored on the computer readable storage medium, the method of any one of claims 1-5 is realized.

Images