Multi-mode bridge vibration monitoring system
By synchronously collecting bridge acceleration and sound signals using multimodal sensors, and combining feature extraction and fusion evaluation, a vibration distribution cloud map is generated. This solves the problem of single-modal monitoring being susceptible to noise interference, and enables high-precision assessment and real-time early warning of bridge health status.
Patent Information
- Application Number
- CN202511483169.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-17
- Publication Date
- 2026-01-13
AI Technical Summary
Existing bridge vibration monitoring methods mostly rely on single-mode sensors, which are susceptible to noise interference and cannot fully reflect the true vibration state of the bridge. Furthermore, the health status assessment does not take into account external environmental factors, leading to false alarms or missed alarms, and lacks robustness.
Multimodal sensors are used to simultaneously collect acceleration and sound signals. By combining feature extraction, multimodal fusion and vibration state assessment, a vibration distribution cloud map is generated and an overall anomaly index is calculated, and the early warning threshold is dynamically adjusted.
It improves the accuracy and stability of signal feature extraction in bridge vibration monitoring, enhances the real-time nature and adaptability of health status assessment, and reduces the risk of false alarms and missed alarms.
Smart Images
Figure CN121323905A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of bridge engineering technology, and in particular to a multimodal bridge vibration monitoring system. Background Technology
[0002] As a crucial component of transportation infrastructure, the structural safety and long-term service performance of bridges directly impact the safety and efficiency of public transportation. During their service life, bridges are subject to various factors, including vehicle loads, changes in ambient temperature, wind forces, and material aging, which can easily lead to changes in their vibration characteristics. Real-time monitoring of bridge vibration response not only reflects the stress state of the bridge but also provides vital information for structural damage identification and health assessment.
[0003] Most existing bridge vibration monitoring methods rely on single-mode sensor data, such as using only accelerometers to collect bridge vibration responses or relying solely on acoustic sensors to capture structural anomalies. These single-mode monitoring methods are susceptible to noise interference in complex environments, leading to unstable signal feature extraction and an inability to comprehensively reflect the true vibration state of the bridge. Furthermore, existing methods often use fixed thresholds for health status assessment, failing to consider the dynamic influence of external environmental factors such as temperature and traffic flow, which can easily result in false alarms or missed alarms. In addition, some existing monitoring systems lack robustness in feature fusion and spatial distribution calculation, making it difficult to generate vibration distribution results covering the entire structure, thus limiting the accuracy and reliability of health status assessments. Summary of the Invention
[0004] This invention provides a multimodal bridge vibration monitoring system that simultaneously acquires acceleration and sound signals, and combines feature extraction, multimodal fusion, and vibration status assessment to achieve visualization of bridge vibration distribution and intelligent early warning of health status.
[0005] A multimodal bridge vibration monitoring system includes a signal synchronous acquisition module, a signal feature extraction module, a multimodal feature fusion module, and a vibration state assessment module, wherein;
[0006] The signal synchronization acquisition module is used to deploy acceleration sensors and sound sensors at multiple key parts of the bridge, and to synchronously acquire bridge acceleration signals and bridge sound signals within a predetermined time period.
[0007] The signal feature extraction module is used to perform time-frequency analysis on the collected bridge acceleration signal and bridge sound signal respectively, and obtain the acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix accordingly.
[0008] The multimodal feature fusion module is used to perform weighted fusion of the obtained acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix to generate a multimodal fusion feature matrix.
[0009] The vibration state assessment module is used to perform spatial interpolation calculation on the multimodal fusion feature matrix to generate a vibration distribution cloud map of the bridge, calculate the overall anomaly index of the bridge based on the vibration distribution cloud map, and assess the health status of the bridge based on the overall anomaly index.
[0010] Optionally, the execution process of the signal synchronization acquisition module includes the following steps:
[0011] Accelerometers and sound sensors are deployed at several key locations on the bridge’s main beams, piers, and hangers to form a sensor network.
[0012] Configure a unified high-precision synchronization clock signal for all deployed accelerometers and sound sensors, and initialize the timestamp of each sensor to align the timestamps of all sensors with the high-precision synchronization clock signal;
[0013] Based on the accelerometer and sound sensor that have completed timestamp alignment, the bridge acceleration signal and bridge sound signal are synchronously triggered and collected within the predetermined time period according to the preset sampling frequency and sampling duration.
[0014] Optionally, the specific process of initializing the timestamp of each sensor includes: broadcasting a synchronization pulse to all accelerometers and sound sensors, and resetting the timing start point of each sensor's internal clock to zero after receiving the synchronization pulse.
[0015] Optionally, the preset sampling frequency is not less than 512Hz for the accelerometer and not less than 16kHz for the sound sensor.
[0016] Optionally, the execution process of the signal feature extraction module includes the following steps:
[0017] Receive the bridge acceleration signal and the bridge sound signal from the signal synchronization acquisition module;
[0018] The received bridge acceleration signal and bridge sound signal are preprocessed to obtain preprocessed bridge acceleration signal and preprocessed bridge sound signal, respectively. The preprocessing includes detrending and filtering.
[0019] Time-frequency analysis was performed on the preprocessed bridge acceleration signal and the preprocessed bridge sound signal. The time-frequency analysis adopted short-time Fourier transform to convert the preprocessed bridge acceleration signal into an initial acceleration time-frequency domain matrix and the preprocessed bridge sound signal into an initial sound time-frequency domain matrix.
[0020] The initial acceleration time-frequency domain matrix and the initial sound time-frequency domain matrix are normalized to obtain the acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix, respectively, and the acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix are output to the multimodal feature fusion module.
[0021] Optionally, the filtering process employs low-pass filtering for the bridge acceleration signal with a cutoff frequency twice the highest frequency of interest for the bridge, and band-pass filtering for the bridge sound signal to preserve the main frequency bands of structural vibration.
[0022] Optionally, the execution process of the multimodal feature fusion module includes the following steps:
[0023] Receive the acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix from the signal feature extraction module, and calculate the global signal-to-noise ratio of the acceleration time-frequency domain feature matrix and the global signal-to-noise ratio of the sound time-frequency domain feature matrix;
[0024] Based on the calculated global signal-to-noise ratio of the acceleration time-frequency domain feature matrix and the global signal-to-noise ratio of the sound time-frequency domain feature matrix, an adaptive weight allocation strategy is adopted to assign a first weight to the acceleration time-frequency domain feature matrix and a second weight to the sound time-frequency domain feature matrix, respectively.
[0025] Each element in the received acceleration time-frequency domain feature matrix is multiplied by the assigned first weight, and each element in the sound time-frequency domain feature matrix is multiplied by the assigned second weight. The element values at corresponding positions in the two weighted matrices are then added together to generate an initial multimodal fusion feature matrix.
[0026] The generated initial multimodal fusion feature matrix is subjected to overall energy normalization to obtain the final multimodal fusion feature matrix, and the multimodal fusion feature matrix is output to the vibration state assessment module.
[0027] Optionally, the adaptive weight allocation strategy is as follows: the first weight is set to the ratio of the global signal-to-noise ratio of the acceleration time-frequency domain feature matrix to the sum of the two global signal-to-noise ratios, and the second weight is set to the ratio of the global signal-to-noise ratio of the sound time-frequency domain feature matrix to the sum of the two global signal-to-noise ratios.
[0028] Optionally, the execution process of the vibration state assessment module includes the following steps:
[0029] The system receives the multimodal fusion feature matrix from the multimodal feature fusion module and performs spatial interpolation calculation on the multimodal fusion feature matrix based on the coordinate information of the sensor deployment location to generate the vibration distribution cloud map covering the entire bridge structure.
[0030] The generated vibration distribution cloud map for the current monitoring period is compared pixel by pixel with the baseline vibration distribution cloud map of the baseline health state stored in the database, and the sum of squares of the differences of all pixels is calculated as the difference degree.
[0031] The calculated difference is input into a preset normalization function for processing, and the overall anomaly index, which is in the range of 0 to 1, is output.
[0032] The obtained overall anomaly index is compared with a preset warning threshold. When the overall anomaly index continuously exceeds the warning threshold, the bridge health status is determined to be abnormal and a structural warning signal is generated.
[0033] Optionally, the warning threshold is a dynamic threshold, which is dynamically adjusted based on historical monitoring data of the bridge under different temperature and traffic flow conditions.
[0034] The beneficial effects of this invention are:
[0035] This invention ensures strict synchronization of multi-source data by deploying accelerometers and acoustic sensors in key components such as the main beam, piers, and hangers, and utilizing a high-precision synchronous clock signal for unified time synchronization and timestamp alignment. Based on this, by combining preprocessing methods such as low-pass filtering, band-pass filtering, and detrending term processing, as well as time-frequency analysis and normalization processing based on short-time Fourier transform, the time-frequency domain feature matrices of both acceleration and acoustic signals can be obtained simultaneously. This scheme not only overcomes the problem of single-mode interference in complex environments but also improves the accuracy and stability of signal feature extraction, providing high-quality input for subsequent fusion analysis.
[0036] This invention calculates the global signal-to-noise ratio (SNR) of the acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix, and employs an adaptive weight allocation strategy for weighted fusion, ensuring that modes with higher SNR occupy a larger weight in the fusion result. Furthermore, by combining dimensionality verification and interpolation / truncation adjustment methods, it ensures that the feature matrices of different modes remain consistent before fusion, avoiding information loss caused by dimensionality mismatch. Finally, the multimodal fusion feature matrix generated through L2 norm normalization has a unified energy scale, thereby improving the robustness of the overall feature representation and the sensitivity to structural vibration details.
[0037] This invention utilizes an inverse distance weighted interpolation algorithm to generate a vibration distribution cloud map covering the entire bridge structure. The difference is calculated by comparing the pixel-by-pixel root mean square error with a benchmark vibration distribution cloud map, and then combined with a normalization function to obtain an overall anomaly index. This allows for a direct quantification of the degree of vibration anomaly in the bridge under different monitoring periods. Furthermore, a formulaic dynamic early warning threshold adjustment method is employed, incorporating temperature changes and traffic flow levels into the threshold calculation, ensuring that the early warning threshold dynamically changes with environmental conditions. When the overall anomaly index continuously exceeds the dynamic threshold, the system can promptly generate a structural early warning signal, effectively improving the real-time performance, accuracy, and adaptability of bridge health monitoring, and reducing the risk of false alarms and missed alarms. Attached Figure Description
[0038] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0039] Figure 1 This is a schematic diagram of the system flow according to an embodiment of the present invention;
[0040] Figure 2 This is a schematic diagram of the multimodal feature fusion module in an embodiment of the present invention. Detailed Implementation
[0041] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.
[0042] It should be noted that the use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.
[0043] Generally, terms can be understood at least partly from their use in context. For example, depending at least partly on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in a singular sense, or a combination of features, structures, or characteristics in a plural sense. Additionally, the term "based on" can be understood not necessarily to convey an exclusive set of factors, but rather, alternatively, depending at least partly on the context, to allow for the presence of other factors that are not necessarily explicitly described.
[0044] like Figures 1-2 As shown, a multimodal bridge vibration monitoring system includes a signal synchronous acquisition module, a signal feature extraction module, a multimodal feature fusion module, and a vibration state assessment module, wherein;
[0045] The signal synchronization acquisition module is used to deploy acceleration sensors and sound sensors at multiple key locations on the bridge, and to synchronously acquire the bridge's acceleration signals and sound signals within a predetermined time period. Specifically:
[0046] Sensor network deployment: Sensors are deployed at multiple key locations on the bridge, including the main girder, piers, and hangers. Specifically, accelerometers and acoustic sensors are installed at the mid-span and quarter-span locations of the main girder, the tops of the piers, and the lower anchor points of the hangers, forming a sensor network covering the bridge's critical load-bearing components. Both accelerometers and acoustic sensors are securely fixed to the structural surface to ensure the stability and accuracy of signal acquisition during long-term operation.
[0047] High-precision synchronization clock signal configuration: A high-precision synchronization clock signal is uniformly configured for all deployed accelerometers and sound sensors. The high-precision synchronization clock signal is provided by a GPS timing module or a BeiDou timing module. The high-precision synchronization clock signal is distributed to each sensor via wired connection to ensure that all sensors obtain a unified time reference.
[0048] Timestamp Initialization: After the high-precision synchronization clock signal is distributed, the data acquisition center broadcasts a synchronization pulse to all accelerometers and sound sensors. Upon receiving the synchronization pulse, each accelerometer and sound sensor resets its internal clock's starting point to zero, ensuring that the timestamps of all sensors are strictly aligned with the high-precision synchronization clock signal.
[0049] Sampling parameter settings: Uniform sampling parameters are set based on the signal characteristics of different sensor types. The sampling frequency for the accelerometer is set to no less than 512Hz, and the sampling frequency for the sound sensor is set to no less than 16kHz. The sampling duration is set according to the monitoring task requirements to ensure continuous signal acquisition within a complete predetermined time period.
[0050] Synchronous Triggered Acquisition: Before sampling begins, the data acquisition center sends a unified acquisition start command to the sensor network. Upon receiving the acquisition start command, all acceleration and sound sensors simultaneously activate their acquisition functions and acquire bridge acceleration and sound signals at preset sampling frequencies and durations, achieving fully synchronous acquisition across regions and multiple sensors.
[0051] Signal preprocessing: The acquired raw bridge acceleration and sound signals are preprocessed. Preprocessing includes noise filtering and signal normalization.
[0052] Noise filtering removes electromagnetic interference noise and random environmental noise through a preset filtering algorithm, ensuring the authenticity of the signal waveform;
[0053] Signal normalization adjusts the output amplitudes of different sensors to a uniform range, which facilitates subsequent signal feature extraction.
[0054] The preprocessed bridge acceleration signal and bridge sound signal are transmitted to the signal feature extraction module for subsequent time-frequency analysis.
[0055] The signal feature extraction module is used to perform time-frequency analysis on the acquired bridge acceleration signal and bridge sound signal, respectively, to obtain the corresponding time-frequency domain feature matrix of acceleration and the time-frequency domain feature matrix of sound, specifically:
[0056] Signal Reception: The signal feature extraction module receives the bridge acceleration signal and bridge sound signal output from the signal synchronization acquisition module. Both the bridge acceleration signal and the bridge sound signal are aligned with a high-precision synchronous clock signal to ensure time consistency among multiple sensors and provide reliable input for subsequent signal analysis.
[0057] Preprocessing: The received bridge acceleration signal and bridge sound signal are preprocessed separately to obtain preprocessed bridge acceleration signal and preprocessed bridge sound signal. Preprocessing includes the following two steps:
[0058] Detrend term processing: The global trend term in the bridge acceleration signal and bridge sound signal is eliminated by linear fitting method to ensure the stability of the signal mean during the analysis process.
[0059] Filtering: Low-pass filtering is used on the bridge acceleration signal, with the cutoff frequency set to twice the highest frequency of interest of the bridge, in order to retain the main components of the bridge structural vibration and effectively suppress high-frequency noise; band-pass filtering is used on the bridge sound signal, with the passband of the filter covering the main frequency band of the bridge structural vibration, thereby filtering out irrelevant low-frequency and high-frequency noise in the environment.
[0060] Time-frequency analysis: The preprocessed bridge acceleration signal and bridge sound signal are input into the short-time Fourier transform processing module to perform time-frequency analysis. The short-time Fourier transform uses a Hanning window as the window function, with the window length set to include at least 5 bridge fundamental frequency cycles and the overlap rate set to 50%.
[0061] A short-time Fourier transform is performed on the preprocessed bridge acceleration signal to obtain the initial acceleration time-frequency domain matrix.
[0062] A short-time Fourier transform is performed on the preprocessed bridge sound signal to obtain the initial sound time-frequency domain matrix.
[0063] This time-frequency analysis can simultaneously characterize the energy distribution features of a signal in both the time and frequency domains, providing fundamental data for subsequent feature fusion.
[0064] Normalization: The initial acceleration time-frequency domain matrix and the initial sound time-frequency domain matrix are normalized separately to eliminate the influence of differences in the magnitude of different signal amplitudes on the analysis results. Normalization can be performed using one of the following two methods:
[0065] Maximum and minimum value normalization method: Scaling each element value in the matrix to the [0,1] interval to obtain the acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix.
[0066] Z-score normalization method: Normalize based on the mean and standard deviation of the initial acceleration time-frequency domain matrix and the initial sound time-frequency domain matrix respectively, to generate an acceleration time-frequency domain feature matrix and a sound time-frequency domain feature matrix with a mean of 0 and a standard deviation of 1.
[0067] Feature output: The final acceleration time-frequency domain feature matrix and sound time-frequency domain feature matrix are output to the multimodal feature fusion module to provide input data support for subsequent multimodal fusion and vibration state assessment.
[0068] The multimodal feature fusion module is used to perform weighted fusion of the obtained acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix to generate a multimodal fusion feature matrix, specifically:
[0069] Feature Matrix Reception: The multimodal feature fusion module receives the acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix output from the signal feature extraction module. To ensure the accuracy of subsequent fusion calculations, the system performs integrity verification on the data packets during reception to confirm that there is no data loss or transmission error in the matrix content.
[0070] Global signal-to-noise ratio (SNR) calculation: The global SNR is calculated for both the received acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix. The specific method is as follows:
[0071] The average energy of each time-frequency domain feature matrix within the main structural resonance frequency band in the frequency domain is extracted and defined as the signal energy.
[0072] The average energy of each time-frequency domain feature matrix outside the main structural resonance frequency band is extracted and defined as noise energy;
[0073] Using the ratio of signal energy to noise energy as the global signal-to-noise ratio (SNR), the global SNR of the acceleration time-frequency domain feature matrix and the global SNR of the sound time-frequency domain feature matrix are obtained respectively.
[0074] Adaptive weight allocation: Based on the calculated global signal-to-noise ratios (SNRs), an adaptive weight allocation strategy is used to determine the fusion weights. Specifically:
[0075] The first weight is set as the ratio of the global signal-to-noise ratio of the acceleration time-frequency domain feature matrix to the sum of the two global signal-to-noise ratios;
[0076] The second weight is set as the ratio of the global signal-to-noise ratio of the sound time-frequency domain feature matrix to the sum of the two global signal-to-noise ratios.
[0077] This weighting method can dynamically reflect the reliability of different modal signals under the current monitoring environment, thereby improving the overall effectiveness of the fusion features.
[0078] Matrix Dimension Verification and Adjustment: Before performing weighted fusion, the matrix dimensions of the acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix are first verified to be consistent. If the dimensions are inconsistent, an adjustment operation is performed:
[0079] When the matrix dimension is small, interpolation methods can be used to supplement data points on the time axis or frequency axis to expand the dimension;
[0080] When the matrix dimension is large, the matrix can be reduced by truncating it in non-critical frequency bands or redundant time segments.
[0081] Ultimately, this ensures that the two matrices have completely identical dimensions, providing the conditions for subsequent element-level operations.
[0082] A matrix is defined as "smaller" when the number of elements in the time or frequency dimension is less than 95% of the number of elements in the corresponding dimension of another matrix. In this case, interpolation is performed to expand the matrix to the same dimension as the other matrix.
[0083] A matrix is defined as "larger" when the number of elements in the time or frequency dimension exceeds 105% of the number of elements in the corresponding dimension of another matrix. In this case, a truncation operation is performed to trim the matrix to the same dimension as the other matrix.
[0084] If the difference is within ±5%, then fine-tuning can be done directly through linear interpolation or edge filling, without the need for large-scale interpolation or truncation.
[0085] Weighted fusion operation: Each element in the dimension-verified acceleration time-frequency domain feature matrix is multiplied by the first weight, and each element in the sound time-frequency domain feature matrix is multiplied by the second weight. Element-wise summation is then performed on corresponding elements in the two weighted matrices to generate the initial multimodal fusion feature matrix. This operation ensures that information from the two modal features at the same time-frequency position is effectively integrated.
[0086] Overall Energy Normalization: The generated initial multimodal fusion feature matrix undergoes overall energy normalization. The L2 norm normalization method is used, where the square root of the sum of the squares of all elements in the matrix is taken as the normalization factor. Each element in the matrix is then divided by this normalization factor to obtain the final multimodal fusion feature matrix. This method can suppress biases caused by differences in matrix size or energy, ensuring the fusion result has a uniform energy scale.
[0087] Fusion Feature Output: The final multimodal fusion feature matrix is output to the vibration state assessment module. This output process is transmitted via the data bus and undergoes secondary verification at the module interface to ensure the integrity and consistency of the fusion feature matrix during transmission.
[0088] The vibration status assessment module is used to perform spatial interpolation calculations on the multimodal fusion feature matrix, generate a vibration distribution cloud map of the bridge, calculate the overall anomaly index of the bridge based on the vibration distribution cloud map, and assess the health status of the bridge based on the overall anomaly index. Specifically:
[0089] Feature Matrix Reception and Coordinate Matching: The vibration state assessment module receives the multimodal fusion feature matrix from the multimodal feature fusion module. Combining this with the coordinate information of the sensor deployment locations, each set of feature values in the multimodal fusion feature matrix is matched with the corresponding sensor coordinates, establishing a one-to-one correspondence between time-frequency features and spatial location. This correspondence provides the basic data for subsequent spatial interpolation calculations.
[0090] Spatial interpolation calculation and vibration distribution cloud map generation: Based on the established relationship between coordinates and eigenvalues, the inverse distance weighted interpolation algorithm is used to perform spatial interpolation calculation on the multimodal fusion feature matrix.
[0091] During the interpolation process, the search radius is set to 1 / 10 of the main span length of the bridge to ensure that the coverage of the interpolation operation matches the bridge structural dimensions.
[0092] The minimum number of sampling points is set to 4 to ensure that each interpolation point is supported by at least 4 sensor data, thereby improving the stability and accuracy of the interpolation results.
[0093] After interpolation is completed, a vibration distribution cloud map covering the entire bridge structure is generated to characterize the vibration distribution status of key areas of the bridge during the current monitoring period.
[0094] Vibration distribution cloud map difference calculation: The obtained vibration distribution cloud map for the current monitoring period is compared pixel by pixel with the pre-stored benchmark vibration distribution cloud map in the database. The root mean square error (RMSE) calculation method is used to obtain the square root of the sum of the squares of the differences among all pixels as the difference degree. The specific formula is as follows:
[0095] ;
[0096] in, The current monitoring cycle vibration distribution cloud map is the [number]th ... pixel value, The first reference vibration distribution cloud map pixel value, This represents the total number of pixels. This difference is used to quantitatively characterize the degree of deviation between the current state of the bridge and its health baseline.
[0097] Overall anomaly index calculation: The calculated difference is input into a preset normalization function for processing to obtain an overall anomaly index ranging from 0 to 1.
[0098] When a linear normalization function is used, the degree of difference is directly mapped to the interval [0,1], where 0 represents no difference and 1 represents the maximum acceptable difference;
[0099] When using a normalization function trained based on historical monitoring data, the historical maximum difference is used as the normalization benchmark, and the ratio of the current difference to the historical maximum difference is used as the overall anomaly index.
[0100] This method ensures that the differences between different monitoring periods have a uniform measurement scale, which facilitates long-term tracking of health status.
[0101] Early warning threshold comparison and health status assessment: The obtained overall anomaly index is compared with the preset early warning threshold. The early warning threshold is a dynamic threshold, dynamically adjusted based on historical monitoring data of the bridge under different temperature and traffic flow conditions. The calculation formula is:
[0102] ;
[0103] Among them, the basic threshold is the warning threshold set under standard environmental conditions (reference temperature, normal traffic flow). This is the difference between the current temperature and the reference temperature, used to reflect the influence of temperature on the vibration characteristics of the bridge structure. The temperature influence coefficient, obtained through training with historical monitoring data, is used to characterize the sensitivity of temperature changes to the early warning threshold. The traffic flow impact coefficient, also learned from historical monitoring data, is used to characterize the sensitivity of traffic flow to the warning threshold. This represents the current traffic flow level, with values ranging from non-negative integers, and is used to reflect the impact of different traffic loads on the bridge's vibration response.
[0104] When the overall abnormality index is less than or equal to the warning threshold, the bridge is judged to be in normal health condition.
[0105] When the overall anomaly index continuously exceeds the warning threshold, the bridge's health status is determined to be abnormal, and a structural warning signal is generated.
[0106] The criteria for consecutive exceedances are: the overall abnormal index exceeds the warning threshold in the current monitoring period and the two previous monitoring periods.
[0107] Establishment and updating of benchmark seismic distribution cloud map: The benchmark seismic distribution cloud map is established by the following method: Under the healthy condition of the bridge, multimodal fusion feature matrices are continuously collected for 30 natural days, spatial interpolation is performed on each matrix, and the average value is calculated to generate the benchmark seismic distribution cloud map.
[0108] During long-term operation, the system can periodically update the baseline vibration distribution cloud map based on new health status monitoring data to ensure that it matches the actual condition of the bridge, thereby improving the reliability of health status assessment.
[0109] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.
[0110] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A multimodal bridge vibration monitoring system, characterized in that, It includes a signal synchronization acquisition module, a signal feature extraction module, a multimodal feature fusion module, and a vibration state assessment module, among which; The signal synchronization acquisition module is used to deploy acceleration sensors and sound sensors at multiple key parts of the bridge, and to synchronously acquire bridge acceleration signals and bridge sound signals within a predetermined time period. The signal feature extraction module is used to perform time-frequency analysis on the collected bridge acceleration signal and bridge sound signal respectively, and obtain the acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix accordingly; The multimodal feature fusion module is used to perform weighted fusion of the obtained acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix to generate a multimodal fusion feature matrix; The vibration state assessment module is used to perform spatial interpolation calculation on the multimodal fusion feature matrix to generate a vibration distribution cloud map of the bridge, calculate the overall anomaly index of the bridge based on the vibration distribution cloud map, and assess the health status of the bridge based on the overall anomaly index.
2. The multimodal bridge vibration monitoring system according to claim 1, characterized in that, The execution process of the signal synchronization acquisition module includes the following steps: Accelerometers and sound sensors are deployed at several key locations on the bridge’s main beams, piers, and hangers to form a sensor network. Configure a unified high-precision synchronization clock signal for all deployed accelerometers and sound sensors, and initialize the timestamp of each sensor to align the timestamps of all sensors with the high-precision synchronization clock signal; Based on the accelerometer and sound sensor that have completed timestamp alignment, the bridge acceleration signal and bridge sound signal are synchronously triggered and collected within the predetermined time period according to the preset sampling frequency and sampling duration.
3. The multimodal bridge vibration monitoring system according to claim 2, characterized in that, The specific process of initializing the timestamp of each sensor includes: broadcasting a synchronization pulse to all accelerometers and sound sensors, and resetting the timing start of each sensor's internal clock to zero after receiving the synchronization pulse.
4. The multimodal bridge vibration monitoring system according to claim 3, characterized in that, The preset sampling frequency is no less than 512Hz for the accelerometer and no less than 16kHz for the sound sensor.
5. The multimodal bridge vibration monitoring system according to claim 3, characterized in that, The execution process of the signal feature extraction module includes the following steps: Receive the bridge acceleration signal and the bridge sound signal from the signal synchronization acquisition module; The received bridge acceleration signal and bridge sound signal are preprocessed to obtain preprocessed bridge acceleration signal and preprocessed bridge sound signal, respectively. The preprocessing includes detrending and filtering. Time-frequency analysis was performed on the preprocessed bridge acceleration signal and the preprocessed bridge sound signal. The time-frequency analysis adopted short-time Fourier transform to convert the preprocessed bridge acceleration signal into an initial acceleration time-frequency domain matrix and the preprocessed bridge sound signal into an initial sound time-frequency domain matrix. The initial acceleration time-frequency domain matrix and the initial sound time-frequency domain matrix are normalized to obtain the acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix, respectively, and the acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix are output to the multimodal feature fusion module.
6. The multimodal bridge vibration monitoring system according to claim 5, characterized in that, The filtering process employs low-pass filtering for the bridge acceleration signal, with a cutoff frequency twice the highest frequency of interest for the bridge, and band-pass filtering for the bridge sound signal to preserve the main frequency bands of structural vibration.
7. The multimodal bridge vibration monitoring system according to claim 5, characterized in that, The execution process of the multimodal feature fusion module includes the following steps: Receive the acceleration time-frequency domain feature matrix and the sound time-frequency domain feature matrix from the signal feature extraction module, and calculate the global signal-to-noise ratio of the acceleration time-frequency domain feature matrix and the global signal-to-noise ratio of the sound time-frequency domain feature matrix; Based on the calculated global signal-to-noise ratio of the acceleration time-frequency domain feature matrix and the global signal-to-noise ratio of the sound time-frequency domain feature matrix, an adaptive weight allocation strategy is adopted to assign a first weight to the acceleration time-frequency domain feature matrix and a second weight to the sound time-frequency domain feature matrix, respectively. Each element in the received acceleration time-frequency domain feature matrix is multiplied by the assigned first weight, and each element in the sound time-frequency domain feature matrix is multiplied by the assigned second weight. The element values at corresponding positions in the two weighted matrices are then added together to generate an initial multimodal fusion feature matrix. The generated initial multimodal fusion feature matrix is subjected to overall energy normalization to obtain the final multimodal fusion feature matrix, and the multimodal fusion feature matrix is output to the vibration state assessment module.
8. The multimodal bridge vibration monitoring system according to claim 7, characterized in that, The adaptive weight allocation strategy is as follows: the first weight is set as the ratio of the global signal-to-noise ratio of the acceleration time-frequency domain feature matrix to the sum of the two global signal-to-noise ratios, and the second weight is set as the ratio of the global signal-to-noise ratio of the sound time-frequency domain feature matrix to the sum of the two global signal-to-noise ratios.
9. A multimodal bridge vibration monitoring system according to claim 7, characterized in that, The execution process of the vibration state assessment module includes the following steps: The system receives the multimodal fusion feature matrix from the multimodal feature fusion module and performs spatial interpolation calculation on the multimodal fusion feature matrix based on the coordinate information of the sensor deployment location to generate the vibration distribution cloud map covering the entire bridge structure. The generated vibration distribution cloud map for the current monitoring period is compared pixel by pixel with the baseline vibration distribution cloud map of the baseline health state stored in the database, and the sum of squares of the differences of all pixels is calculated as the difference degree. The calculated difference is input into a preset normalization function for processing, and the overall anomaly index, which is in the range of 0 to 1, is output. The obtained overall anomaly index is compared with a preset warning threshold. When the overall anomaly index continuously exceeds the warning threshold, the bridge health status is determined to be abnormal and a structural warning signal is generated.
10. A multimodal bridge vibration monitoring system according to claim 9, characterized in that, The warning threshold is a dynamic threshold, which is dynamically adjusted based on historical monitoring data of the bridge under different temperature and traffic flow conditions.