Star-vehicle collaborative physical constraint driving damage identification method for heavy load bridge in cold region

Abstract

A satellite-vehicle collaborative physical constraint-driven damage identification method for heavy-load bridges in cold regions is proposed. Currently, the damage locations of heavy-load bridges in cold regions are random, the damage types are diverse, and multiple parts of the damage are coupled and superimposed, making it difficult to extract damage features and accurately model the damage evolution law. Early damage identification and assessment are difficult. Satellite-vehicle collaboration is only a shallow superposition mode of InSAR initial screening and vehicle-mounted detailed investigation, without realizing a standardized process of deep data fusion and global physical constraint processing. This invention will sequentially perform data preprocessing and full-process physical constraint-driven damage inversion processing on the constructed satellite-vehicle dual-source data collaborative inversion to establish an L1-L2 hybrid regularized model for satellite-vehicle dual-source data collaborative inversion. Based on the L1-L2 hybrid regularized model for satellite-vehicle dual-source data collaborative inversion, the damage assessment and uncertainty quantification process will be completed.

Description

Technical Field

[0001] This invention specifically relates to a star-vehicle collaborative physical constraint-driven damage identification method for heavy-load bridges in cold regions, belonging to the field of road and bridge health monitoring technology. Background Technology

[0002] Heavy-load bridges are subjected to repeated vehicle traffic, overload impacts, and the combined effects of environmental corrosion and temperature changes over long periods. Damage to these bridges often exhibits characteristics of being insidious, cumulative, and progressive, with minimal changes in macroscopic appearance. Early damage such as minute cracks, internal steel reinforcement corrosion, and internal concrete voids is easily overlooked. Currently, industry practices primarily rely on manual visual inspection and periodic load tests for routine inspections. The adoption rate of structural health monitoring systems is low, the number of monitoring points is limited, and data processing capabilities are insufficient, making real-time tracking and precise location of damage difficult. Common identification methods include manual visual inspection, non-destructive testing such as ultrasonic and rebound tests, static and dynamic load tests, and sensor-based structural health monitoring and signal analysis. The core challenge lies in the complex and variable stress field of bridges caused by repeated heavy loads. Early structural response signals are weak and easily masked by external interference signals such as traffic loads, environmental noise, and temperature and humidity changes. Furthermore, the random location, diverse types of damage, and overlapping damage from multiple locations make damage feature extraction difficult and damage evolution patterns hard to model accurately, further increasing the difficulty of early damage identification and assessment.

[0003] Small and medium-span concrete beam bridges that have been in service for more than ten years in cold regions account for a high proportion. They are subjected to high-frequency heavy-load cycles with axle loads generally exceeding 30t, coupled with the deterioration effects of 100 to 200 freeze-thaw cycles per year, chloride corrosion from de-icing agents, and extreme environments with annual temperature differences exceeding 80℃. The incidence of defects is about 2 to 3 times that of warm regions, and the average service life is only 58% of the design reference period, resulting in extremely high maintenance and operation costs.

[0004] Existing bridge monitoring technologies suffer from three major, irreconcilable engineering challenges:

[0005] The main contradiction between cost and coverage is that fixed sensor health monitoring systems have high deployment costs per bridge and can only cover bridges with extra-large spans, leaving a large proportion of small and medium-sized bridges in a long-term monitoring blind spot; manual inspections are inefficient and highly subjective, and cannot achieve early damage warning.

[0006] The contradiction between accuracy and scenario: single vehicle-mounted mobile sensing technology is affected by road roughness, temperature distortion and strong noise interference in cold regions, resulting in insufficient accuracy of vehicle-bridge coupled response reconstruction and poor robustness of damage identification; single spaceborne InSAR technology can only capture macroscopic low-frequency deformation and cannot identify local high-frequency dynamic damage caused by heavy load.

[0007] The contradiction between physical consistency and generalization means that purely data-driven damage identification methods lack physical constraints, are prone to non-physical interpretations, and have poor generalization ability.

[0008] In summary, existing monitoring technologies for small-span bridges in cold regions suffer from drawbacks such as high cost, narrow coverage, insufficient depth of satellite-rover data fusion, poor physical consistency of inversion results, and weak adaptability to extreme environments. These limitations fail to meet the urgent needs of large-scale, refined operation and maintenance of numerous bridges in cold regions. Furthermore, satellite-rover collaboration remains at a shallow overlay mode of InSAR initial screening and vehicle-mounted detailed investigation, failing to achieve a standardized process of deep data fusion and global physical constraints. Summary of the Invention

[0009] To overcome the shortcomings of existing technologies, a star-vehicle cooperative physical constraint-driven damage identification method for heavy-load bridges in cold regions is provided to solve the above problems.

[0010] A method for identifying damage driven by physical constraints in satellite-vehicle collaborative systems for heavy-load bridges in cold regions is proposed. The method involves performing data preprocessing and full-process physical constraint-driven damage inversion processing on a constructed satellite-vehicle dual-layer collaborative sensing system, and then establishing an L1-L2 hybrid regularized model for satellite-vehicle dual-source data collaborative inversion. The damage assessment and uncertainty quantification process are completed based on the L1-L2 hybrid regularized model for satellite-vehicle dual-source data collaborative inversion.

[0011] As a preferred option, the construction process of the spaceborne-vehicle dual-layer collaborative perception system is to establish a data system based on spaceborne data, vehicle data, and spaceborne deformation data and vehicle dynamic response data after spatiotemporal matching processing.

[0012] The satellite data acquisition process involves acquiring one year of satellite C-band SAR data, with a revisit period of 6 to 12 days and a spatial resolution of 5 to 20 meters. After inverting the long-term temporal deformation data of the bridge using PS-InSAR technology, multiple effective deformation observations are obtained.

[0013] The vehicle-mounted data collection process involves deploying vehicle-mounted terminals on multiple vehicles with a speed of 40-60 km / h. The vehicle-mounted terminals include a three-axis accelerometer, a Beidou positioning module, and an environmental sensor. They collect bridge crossing data in two periods, winter and summer, with 40-60 valid bridge crossing samples collected in each period.

[0014] The spatiotemporal matching processing is based on the spatiotemporal reference module, and uses timestamp alignment and spatial coordinate transformation algorithms to obtain spaceborne deformation data and vehicle-mounted dynamic response data. The time synchronization error between the spaceborne deformation data and the vehicle-mounted dynamic response data is less than or equal to 1 second, and the spatial registration error is less than or equal to 0.5 m.

[0015] As a preferred option, the data preprocessing process of the spaceborne-vehicle dual-layer collaborative perception system is to sequentially perform noise suppression processing, contact point response reconstruction processing, and environmental interference compensation processing on the spaceborne-vehicle dual-layer collaborative perception system.

[0016] The noise suppression process involves using variational mode decomposition (VMD) to decompose the vehicle acceleration signal into 6 to 12 intrinsic mode functions (IMFs). The IMF components containing the bridge response are selected by using the kurtosis criterion. Then, the residual noise is removed by wavelet thresholding. After the signal-to-noise ratio is improved by more than 12 dB, the first-level processed data is obtained.

[0017] The process of reconstructing the contact point response of the primary processing data is as follows: a physical constraint variational autoencoder PC-VAE is constructed through the physical constraint response reconstruction module. The vehicle-axle coupling dynamic equation is used as the network regularization term. The vehicle body acceleration signal is input and the wheel axle contact point response is output. After eliminating the interference of vehicle inherent vibration and road surface roughness, the secondary processing data is formed.

[0018] The process of environmental interference compensation for secondary processed data involves establishing a temperature-road roughness coupled correction model through transfer learning using the environmental interference compensation module, inputting environmental temperature and road surface condition data, standardizing damage features, and eliminating the impact of winter and summer environmental differences in cold regions on the identification results to form tertiary processed data.

[0019] As a preferred option: the process of establishing an L1-L2 hybrid regularized model for collaborative inversion of satellite-vehicle dual-source data after performing full-process physical constraint-driven damage inversion on the three-level processed data is based on the maximum deformation value at the mid-span of the bridge inverted by InSAR, establishing the feasible region of structural stiffness degradation parameters, limiting the stiffness reduction coefficient range to 0.8~1.2, using the dynamic differential equation of the vehicle-bridge coupled system as the core control equation for inversion, constraining the physical consistency of the inversion process, using the energy conservation integral method to discretize the nonlinear dynamic equation, ensuring energy conservation and numerical stability in the solution process, and then establishing an L1-L2 hybrid regularized model for collaborative inversion of satellite-vehicle dual-source data.

[0020] The process of establishing the L1-L2 hybrid regularization model for the collaborative inversion of satellite-vehicle dual-source data is as follows: The objective function corresponding to the L1-L2 hybrid regularization model for the collaborative inversion of satellite-vehicle dual-source data is established through an L1-L2 hybrid regularization unit:

[0021]

[0022] In the above formula, θ is the structural stiffness degradation parameter vector, α is the first regularization coefficient, β is the second regularization coefficient, and the first regularization coefficient α and the second regularization coefficient β are determined according to the local heavy-duty vehicle coefficient; A is the system matrix after linearization of the vehicle-bridge coupling system; y is the observation vector containing the spaceborne deformation and vehicle response; m represents the test value.

[0023] The L1-L2 hybrid regularized model is solved by using the adaptive constrained unscented Kalman filter (UKF) solution unit. An adaptive forgetting factor is introduced to dynamically adjust the weights of historical data, thereby completing the process of obtaining the stiffness degradation parameters of each bridge element after solving the regularized model.

[0024] As a preferred option, the Alternating Directional Multiplier Method (ADMM) is used to solve the global optimal solution of the objective function during the process of solving the formula corresponding to the L1-L2 hybrid regularized model of the completed star-rocket dual-source data collaborative inversion.

[0025] As a preferred option: the damage assessment and uncertainty quantification process is completed by using the L1-L2 hybrid regularization model based on the collaborative inversion of star-vehicle dual-source data. The damage assessment and output module completes the damage identification and uncertainty quantification assessment process.

[0026] The damage identification process involves determining the degree of damage based on the stiffness degradation parameters of each bridge unit obtained from the data. At the same time, the calculated stiffness degradation parameters of each bridge unit are compared with the measured results of the static load test to determine the error rate, thereby verifying the accuracy of the damage assessment.

[0027] The uncertainty quantification process involves quantifying the uncertainty of the inversion results using the Monte Carlo-response surface methodology, classifying the damage level based on the degree of stiffness degradation, and outputting a bridge damage assessment report that includes damage location, damage degree, confidence interval, and maintenance recommendations.

[0028] As a preferred approach: the Monte Carlo random field generation process in the uncertainty quantification is a joint distribution of simulation process noise and observation noise. A surrogate model is established to quickly assess the sensitivity of parameter identification results to noise and output confidence intervals. Simulation process noise refers to model error, and observation noise refers to sensor error. Based on the sensitivity parameter threshold, the damage mutation rate index is set, and the weight of historical data is dynamically adjusted. The formula for calculating the forgetting factor, which enhances the tracking ability of time-varying parameters, is as follows:

[0029]

[0030] In the above formula, α∈(0,1) and tr represents the trace of the matrix.

[0031] As a preferred solution, the spaceborne-vehicle dual-layer collaborative sensing system uses multi-source data. During the preprocessing of the multi-source data, a variational mode decomposition and wavelet thresholding joint denoising method is used to separate environmental noise and the vehicle's inherent vibration components. A physically constrained variational autoencoder is constructed to reconstruct the wheel axle contact point response. After extracting a multi-dimensional damage-sensitive feature set, the adaptive compensation process for cold-region temperature-road roughness coupling interference is completed through transfer learning using the multi-dimensional damage-sensitive feature set. The multi-dimensional damage-sensitive feature set includes frequency domain modal features, time domain statistical features, and time-frequency domain energy features. The frequency domain modal features include frequency shift and mode shape curvature, the time domain statistical features include root mean square response and peak factor, and the time-frequency domain energy features include wavelet energy entropy and Hilbert-Huang transform marginal spectrum.

[0032] The beneficial effects of this invention are as follows:

[0033] This invention presents a quantitative identification method specifically adapted for the damage severity of heavily loaded bridges in cold regions. It utilizes a pre-constructed spaceborne-vehicle dual-layer collaborative sensing system as benchmark data, achieving high-precision cross-scale data matching based on the BeiDou spatiotemporal reference. Based on the benchmark data, the invention employs an adaptive constrained unscented Kalman filter algorithm to solve the L1-L2 hybrid regularized model generated through a full-process physical constraint-driven damage inversion process. This enables accurate damage identification and uncertainty quantification, providing precise assessment of heavily loaded bridges in cold regions and facilitating safety monitoring and effective early warning systems for these bridges.

[0034] The spaceborne-vehicle dual-layer collaborative perception system in this invention is a dual-layer collaborative perception system that can realize a spaceborne InSAR wide-area survey combined with heavy-duty vehicles and a vehicle-mounted sensing precise detailed survey of heavy-duty vehicles. It reconstructs the wheel bridge contact point response through a physically constrained variational autoencoder and completes adaptive compensation for interference in extreme cold environments by combining transfer learning. It innovatively constructs a full-process physical multi-constraint framework of boundary constraints, equation constraints, and energy constraints.

[0035] This invention eliminates the need for fixed sensors on the bridge structure, reducing the cost of monitoring a single bridge by more than 90%, and achieving a damage location accuracy of ≥95% at a noise level of 10%. It is particularly suitable for large-scale intelligent operation and maintenance of a large number of small and medium-span concrete bridges in cold regions. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the working principle of the present invention;

[0037] Figure 2 A flowchart for bridge parameter identification based on star-vehicle heterogeneous test information;

[0038] Figure 3 This is a schematic diagram of a deep learning network structure;

[0039] Figure 4 The flowchart shows the process of using the VMD-wavelet threshold joint denoising method to separate noise components in the multi-source data preprocessing module. Detailed Implementation

[0040] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0041] Specific implementation method one: Combining Figures 1 to 4 This embodiment describes the method for identifying star-vehicle collaborative constraint-driven damage in cold-region heavy-load bridges. The method involves sequentially preprocessing the constructed star-vehicle dual-layer collaborative sensing system with data and performing full-process physical constraint-driven damage inversion processing. Then, an L1-L2 hybrid regularized model for collaborative inversion of star-vehicle dual-source data is established. Based on this model, damage assessment and uncertainty quantification are completed.

[0042] The spaceborne-vehicle dual-layer collaborative sensing system established in this invention serves as the data foundation. It utilizes data collected from actual satellites and heavy-duty vehicles to provide accurate data for subsequent evaluation. Specifically, the spaceborne-vehicle dual-layer collaborative sensing system uses multi-source data. During the preprocessing of the multi-source data, a variational mode decomposition and wavelet thresholding joint denoising method is used to separate environmental noise and the vehicle's inherent vibration components. A physically constrained variational autoencoder is constructed to reconstruct the wheel axle contact point response. After extracting a multi-dimensional damage-sensitive feature set, transfer learning is performed using the multi-dimensional damage-sensitive feature set to complete the adaptive compensation process for cold-region temperature-road roughness coupling interference. The multi-dimensional damage-sensitive feature set includes frequency domain modal features, time domain statistical features, and time-frequency domain energy features. The frequency domain modal features include frequency shift and mode shape curvature; the time domain statistical features include root mean square response and peak factor; and the time-frequency domain energy features include wavelet energy entropy and Hilbert-Huang transform marginal spectrum.

[0043] This invention is suitable for use in the process of bridge structural health monitoring. Specifically, it is adapted to the intersection of integrated air-space-ground sensing, structural dynamics inversion, and cold-region civil engineering. It is particularly suitable for low-cost, high-precision early damage identification and safety assessment of a large number of small- and medium-span concrete bridges on heavy-load trunk highways in cold regions such as Northeast and Northwest China.

[0044] In this embodiment, L1-L2 in the L1-L2 hybrid regularization model of satellite-vehicle dual-source data collaborative inversion refers to two objectives, that is, L1 is one objective and L2 is another objective. L1-L2 is a dual objective, that is, the L1-L2 hybrid regularization model of satellite-vehicle dual-source data collaborative inversion is a dual-objective hybrid regularization model of satellite-vehicle dual-source data collaborative inversion.

[0045] Specific Implementation Method Two: This implementation method is a further limitation of Specific Implementation Method One. In this implementation method, the construction process of the space-vehicle dual-layer collaborative perception system is a data system established based on space-vehicle data, vehicle data, and space-vehicle deformation data and vehicle dynamic response data after spatiotemporal matching processing.

[0046] The satellite data acquisition process involves acquiring one year of satellite C-band SAR data, with a revisit period of 6 to 12 days and a spatial resolution of 5 to 20 meters. After inverting the long-term temporal deformation data of the bridge using PS-InSAR technology, multiple effective deformation observations are obtained.

[0047] The vehicle-mounted data collection process involves deploying vehicle-mounted terminals on multiple vehicles with a speed of 40-60 km / h. The vehicle-mounted terminals include a three-axis accelerometer, a Beidou positioning module, and an environmental sensor. They collect bridge crossing data in two periods, winter and summer, with 40-60 valid bridge crossing samples collected in each period.

[0048] The spatiotemporal matching processing is based on the spatiotemporal reference module, and uses timestamp alignment and spatial coordinate transformation algorithms to obtain spaceborne deformation data and vehicle-mounted dynamic response data. The time synchronization error between the spaceborne deformation data and the vehicle-mounted dynamic response data is less than or equal to 1 second, and the spatial registration error is less than or equal to 0.5 m.

[0049] Specific Implementation Method 3: This implementation method is a further limitation of Specific Implementation Method 1. In this implementation method, the data preprocessing process of the spaceborne-vehicle dual-layer collaborative perception system is to sequentially perform noise suppression processing, contact point response reconstruction processing, and environmental interference compensation processing on the spaceborne-vehicle dual-layer collaborative perception system.

[0050] The noise suppression process involves using variational mode decomposition (VMD) to decompose the vehicle acceleration signal into 6 to 12 intrinsic mode functions (IMFs). The IMF components containing the bridge response are selected by using the kurtosis criterion. Then, the residual noise is removed by wavelet thresholding. After the signal-to-noise ratio is improved by more than 12 dB, the first-level processed data is obtained.

[0051] The process of reconstructing the contact point response of the primary processing data is as follows: a physical constraint variational autoencoder PC-VAE is constructed through the physical constraint response reconstruction module. The vehicle-axle coupling dynamic equation is used as the network regularization term. The vehicle body acceleration signal is input and the wheel axle contact point response is output. After eliminating the interference of vehicle inherent vibration and road surface roughness, the secondary processing data is formed.

[0052] The process of environmental interference compensation for secondary processed data involves establishing a temperature-road roughness coupled correction model through transfer learning using the environmental interference compensation module, inputting environmental temperature and road surface condition data, standardizing damage features, and eliminating the impact of winter and summer environmental differences in cold regions on the identification results to form tertiary processed data.

[0053] Specific Implementation Method Four: This implementation method is a further limitation of Specific Implementation Method One, Two, or Three. In this implementation method, the process of establishing an L1-L2 hybrid regularized model for collaborative inversion of satellite-vehicle dual-source data after performing full-process physical constraint-driven damage inversion processing on the three-level processed data is based on the maximum deformation value at the mid-span of the bridge inverted by InSAR, establishing the feasible region of structural stiffness degradation parameters, limiting the stiffness reduction coefficient range to 0.8~1.2, using the dynamic differential equation of the vehicle-bridge coupled system as the core control equation for inversion, constraining the physical consistency of the inversion process, using the energy conservation integral method to discretize the nonlinear dynamic equation, ensuring the energy conservation and numerical stability of the solution process, and then establishing an L1-L2 hybrid regularized model for collaborative inversion of satellite-vehicle dual-source data.

[0054] The process of establishing the L1-L2 hybrid regularization model for the collaborative inversion of satellite-vehicle dual-source data is as follows: The objective function corresponding to the L1-L2 hybrid regularization model for the collaborative inversion of satellite-vehicle dual-source data is established through an L1-L2 hybrid regularization unit:

[0055]

[0056] In the above formula, θ is the structural stiffness degradation parameter vector, α is the first regularization coefficient, β is the second regularization coefficient, and the first regularization coefficient α and the second regularization coefficient β are determined according to the local heavy-duty vehicle coefficient; A is the system matrix after linearization of the vehicle-bridge coupling system; y is the observation vector containing the spaceborne deformation and vehicle response; m represents the test value.

[0057] The L1-L2 hybrid regularized model is solved by using the adaptive constrained unscented Kalman filter (UKF) solution unit. An adaptive forgetting factor is introduced to dynamically adjust the weights of historical data, thereby completing the process of obtaining the stiffness degradation parameters of each bridge element after solving the regularized model.

[0058] Specific Implementation Method 5: This implementation method is a further limitation of Specific Implementation Method 1, 2, 3 or 4. In this implementation method, during the process of solving the formula corresponding to the L1-L2 hybrid regularization model of the completed star-rocket dual-source data collaborative inversion, the Alternating Direction Multiplier Method (ADMM) is used to solve the global optimal solution of the objective function.

[0059] Specific Implementation Method Six: This implementation method is a further limitation of Specific Implementation Method One, Two, Three, Four or Five. In this implementation method, the damage assessment and uncertainty quantification process is completed by the L1-L2 hybrid regularization model based on the collaborative inversion of star-vehicle dual-source data. The damage assessment and output module completes the damage identification and uncertainty quantification assessment process.

[0060] The damage identification process involves determining the degree of damage based on the stiffness degradation parameters of each bridge unit obtained from the data. At the same time, the calculated stiffness degradation parameters of each bridge unit are compared with the measured results of the static load test to determine the error rate, thereby verifying the accuracy of the damage assessment.

[0061] Combination Figure 2 As shown, the uncertainty quantification process in this embodiment is to quantify the uncertainty of the inversion results using the Monte Carlo-response surface methodology, classify the damage level based on the degree of stiffness degradation, and output a bridge damage assessment report that includes damage location, damage degree, confidence interval, and maintenance recommendations.

[0062] The Monte Carlo random field generation process is a joint distribution of simulation process noise and observation noise. A surrogate model is established to quickly assess the sensitivity of parameter identification results to noise and output confidence intervals. Simulation process noise refers to model error, and observation noise refers to sensor error.

[0063] An adaptive forgetting factor mechanism for the identification process is established, specifically by setting a damage mutation rate index based on a sensitive parameter threshold, dynamically adjusting the weight of historical data, and enhancing the tracking capability of time-varying parameters.

[0064] In the above processing, the formula for calculating the forgetting factor is:

[0065]

[0066] In the above formula, the sensitivity parameter α∈(0,1), and tr represents the trace of the matrix. Let P be the observation error at time step k. yy,k Let be the measurement prediction covariance at time step k.

[0067] To enhance the recognition of time-varying structural parameters, the innovation covariance, cross covariance, and posterior state covariance in the UKF algorithm's recursive recognition process are simultaneously corrected based on the forgetting factor. The specific calculation process is as follows:

[0068] ;

[0069] ;

[0070] ;

[0071] ;

[0072] .

[0073] In the above formula, where For the new information covariance at step k, For cross covariance, For the posterior state covariance, W c i The covariance weighting coefficients are... For the observations after the k-th step of the unscented transformation, Let be the estimated value of the observations at k steps. The values ​​are estimates obtained after the unscented transformation of the state variables. + represents the posterior estimate and - represents the prior estimate.

[0074] Specific Implementation Method Seven: This implementation method is a further limitation of Specific Implementation Methods One, Two, Three, Four, Five, or Six, combined with... Figure 1 and Figure 2 As shown, the improved unscented Kalman filter (UKF) algorithm involved in this embodiment has branched processing for time-varying / non-time-varying parameter scenarios. The overall process steps are as follows: starting from the initialization stage, firstly, the iteration counter k=0 is set, and then the core initialization step is entered: establishing the state equation and determining the initial state estimate mean based on the initial conditions. Covariance P K This step is constrained and optimized by improving the state variable distribution model to enhance the rationality of the initial state estimation; then, state Sigma points are generated based on the initial state mean and covariance. and their corresponding weights Scheme Two imposes constraints on the selection method and weights of the Sigma points, optimizing the distribution characteristics of the sampling points and providing a more reliable sample basis for subsequent nonlinear propagation. After generating the Sigma points, the process enters the prediction stage: first, the state Sigma points of the prediction step are calculated using the state equation. Then, based on these predicted Sigma points, the mean state estimate of the prediction step is calculated. Covariance Subsequently, the observation Sigma points for the prediction step are calculated using the observation equation. And calculate the mean of the observation estimates for the prediction step based on the observed Sigma points. Observation covariance This prepares data for subsequent Kalman gain calculations.

[0075] After the prediction phase is completed, the process enters the parameter judgment branch: determining whether the system parameters are time-varying. If the corresponding system parameters are time-varying, then the mean of the observation estimates from the prediction step is calculated. Observation covariance and mutual covariance Adaptive adjustments are made; when the corresponding system parameters are time-invariant, the original prediction step observation estimate is directly used. Observation covariance and mutual covariance Both branches will calculate the Kalman gain based on the corresponding data. .

[0076] The final step of the process is the calibration phase: introducing system observations. Combined with Kalman gain Calculate the mean state estimate of the correction step Covariance The covariance update for time-varying parameter scenarios is as follows: In non-time-varying scenarios, the update is directly performed. After the correction is completed, the iteration counter is updated to k=k+1, and the updated state mean and covariance are used as the initial conditions for the next iteration. The process returns to the step of generating the state Sigma point and repeats the complete process of prediction, parameter judgment and correction to achieve a continuous and high-precision evaluation of the system state.

[0077] The improved unscented Kalman filter (UKF) algorithm in this embodiment improves the constraints on the state distribution model. To address the issues of initial state estimation error and model mismatch, it enhances stability by adding constraints and correcting the initial covariance values ​​to constrain the state distribution.

[0078] Specific Implementation Method Eight: This implementation method is a further limitation of Specific Implementation Methods One, Two, Three, Four, Five, Six, or Seven, combined with... Figure 3As shown, preprocessing of satellite displacement signals, vibration signals from vehicle-mounted intelligent equipment, and position signals was performed to extract features. Specifically, the vehicle response signal was decomposed into the bridge dominant mode component (IMF), filtering out high-frequency noise and road surface excitation interference. Time-frequency analysis was conducted on the IMF components to extract the wavelet coefficient matrix of the damage-sensitive frequency band. Then, damage localization and verification were achieved. The energy difference of the wavelet coefficients before and after damage was calculated based on the residual energy focusing algorithm, and the damage area was determined by detecting local energy peaks. Finally, a scaled-down steel bridge-vehicle test platform was built to simulate damage such as cracks and support failures. The robustness of the algorithm to temperature drift and noise SNR=10dB was verified in the laboratory. Temperature drift was achieved by setting multiple temperature differences.

[0079] By combining the adaptive frequency band characteristics of VMD with the time-frequency localization capabilities of CWT, the ambiguity of traditional time-domain methods is overcome. A closed-loop verification system ensures the reliability of the method through a full-chain verification process from simulation to laboratory to field. The constructed PC-VAE network structure includes an encoder network, a decoder network, and a discriminator network.

[0080] Example 1: This example is described in conjunction with the beneficial effects of the present invention. In this example, the long-term temporal deformation of the bridge is inverted using PS-InSAR technology based on satellite SAR data from the space-based sensing unit; vehicle-mounted terminals are deployed on heavy-duty trucks using ground sensing units to collect effective bridge crossing data for both winter and summer; and the BeiDou spatiotemporal reference module completes the time synchronization and spatial registration of cross-scale data, with errors controlled within 1s and 0.5m, respectively.

[0081] Combination Figure 4 As shown, in the data preprocessing and environmental interference compensation processing stages, the multi-source data preprocessing module uses the VMD-wavelet threshold joint denoising method to separate noise components, improving the signal-to-noise ratio by more than 12dB; the physical constraint response reconstruction unit constructs a PC-VAE network, uses the vehicle-bridge coupling equation as a regularization term, reconstructs the wheel-bridge contact point response, and eliminates the interference of vehicle inherent vibration and road surface roughness; the environmental interference adaptive compensation unit establishes a temperature-road coupling correction model based on transfer learning to standardize the damage characteristics under extreme cold environment.

[0082] In the process of full-process physical constraint-driven damage inversion processing, combined with Figure 1As shown in the central dashed box, the full-process physical constraint inversion module integrates three types of constraints: boundary constraint submodule 71, equation constraint submodule 72, and energy constraint submodule 73. Specifically, boundary constraint submodule 71, based on InSAR inversion, limits the stiffness reduction factor range to [0.6, 1.0] for the maximum deformation at mid-span -12.3 mm; equation constraint submodule 72 embeds the vehicle-bridge coupled dynamic differential equations; energy constraint submodule 73 uses energy conservation integrals to ensure numerical stability; and finally, the L1-L2 hybrid regularized solver 74 uses the adaptive constraint UKF algorithm to solve the problem, introducing a forgetting factor of 0.98 to track time-varying stiffness degradation.

[0083] During the damage assessment and result output processing, the damage assessment and uncertainty quantification unit calculated the stiffness reduction factor at the mid-span of the bridge to be 0.82, with a 95% confidence interval of [0.79, 0.85], which is only 2.5% different from the measured value of the static load test. Finally, the output terminal generates an assessment report containing the damage level, location, degree and maintenance recommendations.

[0084] Example 2: This example uses a 20-meter span prestressed concrete simply supported beam bridge as the test object. This bridge is a 20-meter span prestressed concrete simply supported T-beam bridge, in service for 18 years, located on a core heavy-duty freight transport route in a cold region, with an average daily heavy-duty truck traffic exceeding 1200 vehicles. It has already shown defects such as concrete freeze-thaw spalling and micro-cracks in the web. The specific implementation steps of this invention are as follows:

[0085] Step 1: Satellite-vehicle collaborative sensing data acquisition and spatiotemporal matching to construct a two-layer collaborative sensing system for satellite-borne InSAR and heavy-duty commercial vehicles:

[0086] The satellite data acquisition process involves acquiring nearly one year of Sentinel-1A satellite C-band SAR data with a revisit period of 12 days and a spatial resolution of 10m×10m. PS-InSAR technology is used to retrieve long-term temporal deformation data of the bridge, and a total of 31 valid deformation observations are obtained.

[0087] The vehicle-mounted data acquisition process involves deploying vehicle-mounted terminals on three heavy-duty commercial trucks with an axle load of 32t and a speed of 40~60km / h. These terminals include a triaxial accelerometer with a sampling frequency of 200Hz and a range of ±10g, a Beidou positioning module with a positioning accuracy of ±5cm, and an environmental sensor with a temperature measurement range of -40℃~80℃. Two periods of bridge crossing data are collected, one in January (winter) and one in July (summer), with 50 valid bridge crossing samples collected in each period.

[0088] The spatiotemporal matching process is based on the BeiDou spatiotemporal reference module and uses timestamp alignment and spatial coordinate transformation algorithms to achieve a time synchronization error of ≤1s and a spatial registration error of ≤0.5m between satellite deformation data and vehicle power response data.

[0089] Step Two: Data Preprocessing and Environmental Interference Compensation, which involves the following operations performed by the data preprocessing module:

[0090] For noise suppression, variational mode decomposition (VMD) is used to decompose the vehicle acceleration signal into eight intrinsic mode functions (IMFs). The IMF components containing the bridge response are selected by kurtosis criterion. Then, wavelet thresholding is used to remove residual noise, improving the signal-to-noise ratio by more than 12 dB.

[0091] The contact point response reconstruction process constructs a physical constraint variational autoencoder (PC-VAE) through a physical constraint response reconstruction module. The vehicle-axle coupled dynamic equation is used as the network regularization term. The vehicle body acceleration signal is input, and the wheel axle contact point response is output, eliminating the interference of vehicle inherent vibration and road surface roughness.

[0092] Environmental interference compensation processing involves establishing a temperature-road roughness coupled correction model based on transfer learning through an environmental interference compensation module. By inputting environmental temperature and road surface condition data, the damage characteristics are standardized to eliminate the impact of the difference between winter and summer environments in cold regions on the identification results.

[0093] The above three processes can be performed sequentially.

[0094] Step 3: Damage inversion driven by full-process physical constraints, that is, damage inversion is completed through the full-process physical constraint inversion module. Specifically, this involves constructing a full-process physical constraint framework, which corresponds to three types of constraints:

[0095] Boundary constraints: Based on the maximum deformation value of the bridge mid-span at -12.3 mm obtained by InSAR inversion, a feasible region for structural stiffness degradation parameters is established, and the stiffness reduction factor is limited to the range of [0.8, 1.2].

[0096] Equation constraints: The dynamic differential equations of the vehicle-bridge coupled system are used as the core control equations for the inversion process to constrain the physical consistency of the inversion process;

[0097] Energy constraints: The nonlinear dynamic equations are discretized using the energy conservation integral method to ensure energy conservation and numerical stability during the solution process.

[0098] Establish an L1-L2 hybrid regularization model: Construct an optimization function using L1-L2 hybrid regularization units.

[0099]

[0100] Where θ is the structural stiffness degradation parameter vector, α and β are regularization coefficients determined according to the local heavy-duty vehicle coefficient, A is the system matrix after system linearization, y is the observation vector, and m represents the test value;

[0101] After establishing the model, the model is solved: the elements are solved by adaptive constraint UKF, an adaptive forgetting factor of 0.98 is introduced, the weight of historical data is dynamically adjusted, and the stiffness degradation parameters of each element of the bridge are obtained by solving the regularized model.

[0102] Step 4: Damage assessment and uncertainty quantification process involves the following operations completed through the damage assessment and output module:

[0103] The damage identification results show that the stiffness reduction factor at the mid-span of bridge unit 3 is 0.82, indicating moderate damage. The error between this and the static load test result of a stiffness reduction factor of 0.80 is only 2.5%.

[0104] The uncertainty quantification process involves using an uncertainty quantification unit and the Monte Carlo response surface methodology to quantify the influence of process noise and observation noise, resulting in a 95% confidence interval for the damage parameter of [0.79, 0.85].

[0105] The output assessment report is a bridge damage assessment report that classifies damage levels based on the degree of stiffness degradation as mild (0.9~1.0), moderate (0.7~0.9), and severe (<0.7). The output includes the location of damage, the degree of damage, the confidence interval, and maintenance recommendations.

[0106] In terms of implementation effect verification, it can be seen that the damage identification accuracy of the method of the present invention is significantly better than that of the traditional pure data-driven method under different noise levels: when the noise level is 10%, the damage localization accuracy of the method of the present invention is 96%, and the damage quantization error is 7.2%; while the damage localization accuracy of the traditional BP neural network method is only 68%, and the damage quantization error is 23.5%.

[0107] The dual-layer collaborative sensing system constructed by this invention, which combines spaceborne InSAR wide-area survey with vehicle-mounted sensing for precise detailed survey of heavy-duty operational vehicles, achieves high-precision spatiotemporal matching of cross-scale heterogeneous data based on the BeiDou spatiotemporal reference.

[0108] The physical constraint variational autoencoder PC-VAE established in this invention can realize the reconstruction of wheel axle contact point response and, combined with transfer learning, complete the adaptive compensation of cold region temperature-road roughness coupling interference.

[0109] The full-process physical constraint framework constructed in this invention is a composite physical constraint framework of boundary constraints, equation constraints, and energy constraints, and an L1-L2 hybrid regularization model for collaborative inversion of satellite-rocket dual-source data is established.

[0110] This invention employs an adaptive constrained unscented Kalman filter (AC-UKF) to solve the inversion model, achieving accurate identification of damage location and severity, as well as quantification of uncertainty.

[0111] Practical verification shows that this invention can significantly reduce the maintenance cost of heavy-load bridges: it eliminates the need to deploy fixed sensors on the bridge structure, and by relying on existing operating vehicles and commercial satellite resources, the monitoring cost of a single bridge is reduced by more than 90%. It can realize large-scale monitoring of bridge clusters in cold regions, and is particularly suitable for low-cost, wide-area intelligent operation and maintenance of heavy-load trunk highways in cold regions and a large number of small and medium-span bridges in urban road networks, as well as for improving the safety and resilience of transportation infrastructure, optimizing maintenance decisions, and controlling heavy-load traffic.

[0112] Furthermore, in practical use, this invention has high recognition accuracy, integrates spaceborne global deformation constraints and vehicle-mounted local dynamic response, achieves a damage location accuracy of ≥95% at a noise level of 10%, and a damage quantification error of ≤8%, which is more than 30% higher than the accuracy of traditional pure data-driven methods.

[0113] This invention exhibits strong physical consistency by embedding three types of physical constraints throughout the entire process: boundary constraints, equation constraints, and energy constraints. This fundamentally suppresses non-physical interpretations, and the inversion results possess clear mechanical significance.

[0114] This invention has good cold-region adaptability. Through adaptive compensation for environmental interference, it can effectively eliminate the effects of extreme environments such as large temperature differences, road surface frost heave and thaw settlement, and strong noise in cold regions, and is suitable for the time-varying deterioration scenarios of bridges in cold regions.

Claims

1. A method for damage identification of heavy-load bridges in cold regions using a star-vehicle cooperative physical constraint-driven approach, characterized in that: After the completed spaceborne-vehicle dual-layer collaborative perception system is preprocessed and subjected to full-process physical constraint-driven damage inversion processing, an L1-L2 hybrid regularized model for spaceborne-vehicle dual-source data collaborative inversion is established. Based on the L1-L2 hybrid regularized model for spaceborne-vehicle dual-source data collaborative inversion, the damage assessment and uncertainty quantification process is completed.

2. The method for identifying damage driven by star-vehicle cooperative physical constraints on heavy-load bridges in cold regions according to claim 1, characterized in that: The construction process of the spaceborne-vehicle dual-layer collaborative perception system is to establish a data system based on spaceborne data, vehicle data, and spaceborne deformation data and vehicle dynamic response data after spatiotemporal matching processing; The satellite data acquisition process involves acquiring one year of satellite C-band SAR data, with a revisit period of 6 to 12 days and a spatial resolution of 5 to 20 meters. After inverting the long-term temporal deformation data of the bridge using PS-InSAR technology, multiple effective deformation observations are obtained. The vehicle-mounted data collection process involves deploying vehicle-mounted terminals on multiple vehicles with a speed of 40-60 km / h. The vehicle-mounted terminals include a three-axis accelerometer, a Beidou positioning module, and an environmental sensor. They collect bridge crossing data in two periods, winter and summer, with 40-60 valid bridge crossing samples collected in each period. The spatiotemporal matching processing is based on the spatiotemporal reference module, and uses timestamp alignment and spatial coordinate transformation algorithms to obtain spaceborne deformation data and vehicle-mounted dynamic response data. The time synchronization error between the spaceborne deformation data and the vehicle-mounted dynamic response data is less than or equal to 1 second, and the spatial registration error is less than or equal to 0.5 m.

3. The method for identifying damage driven by star-vehicle cooperative physical constraints on heavy-load bridges in cold regions according to claim 1 or 2, characterized in that: The data preprocessing process of the spaceborne-vehicle dual-layer collaborative perception system is to sequentially perform noise suppression processing, contact point response reconstruction processing, and environmental interference compensation processing on the spaceborne-vehicle dual-layer collaborative perception system. The noise suppression process involves using variational mode decomposition (VMD) to decompose the vehicle acceleration signal into 6 to 12 intrinsic mode functions (IMFs). The IMF components containing the bridge response are selected by using the kurtosis criterion. Then, the residual noise is removed by wavelet thresholding. After the signal-to-noise ratio is improved by more than 12 dB, the first-level processed data is obtained. The process of reconstructing the contact point response of the primary processing data is as follows: a physical constraint variational autoencoder PC-VAE is constructed through the physical constraint response reconstruction module. The vehicle-axle coupling dynamic equation is used as the network regularization term. The vehicle body acceleration signal is input and the wheel axle contact point response is output. After eliminating the interference of vehicle inherent vibration and road surface roughness, the secondary processing data is formed. The process of environmental interference compensation for secondary processed data involves establishing a temperature-road roughness coupled correction model through transfer learning using the environmental interference compensation module, inputting environmental temperature and road surface condition data, standardizing damage features, and eliminating the impact of winter and summer environmental differences in cold regions on the identification results to form tertiary processed data.

4. The method for damage identification driven by star-vehicle cooperative physical constraints on heavy-load bridges in cold regions according to claim 3, characterized in that: The process of establishing an L1-L2 hybrid regularized model for collaborative inversion of satellite-vehicle dual-source data after performing full-process physical constraint-driven damage inversion on the three-level processed data is as follows: Based on the maximum deformation value of the bridge mid-span inversion using InSAR, a feasible region for structural stiffness degradation parameters is established, and the stiffness reduction coefficient is limited to the range of 0.8~1.

2. The dynamic differential equation of the vehicle-bridge coupled system is used as the core control equation for inversion, constraining the physical consistency of the inversion process. The nonlinear dynamic equation is discretized using the energy conservation integral method. After ensuring the energy conservation and numerical stability of the solution process, an L1-L2 hybrid regularized model for collaborative inversion of satellite-vehicle dual-source data is established. The process of establishing the L1-L2 hybrid regularization model for the collaborative inversion of satellite-vehicle dual-source data is as follows: The objective function corresponding to the L1-L2 hybrid regularization model for the collaborative inversion of satellite-vehicle dual-source data is established through an L1-L2 hybrid regularization unit: ； In the above formula, θ is the structural stiffness degradation parameter vector, α is the first regularization coefficient, β is the second regularization coefficient, and the first regularization coefficient α and the second regularization coefficient β are determined according to the local heavy-duty vehicle coefficient; A is the system matrix after linearization of the vehicle-bridge coupling system; y is the observation vector containing the spaceborne deformation and vehicle response. m represents the test value; The L1-L2 hybrid regularized model is solved by using the adaptive constrained unscented Kalman filter (UKF) solution unit. An adaptive forgetting factor is introduced to dynamically adjust the weights of historical data, thereby completing the process of obtaining the stiffness degradation parameters of each bridge element after solving the regularized model.

5. The method for identifying damage driven by star-vehicle cooperative physical constraints on heavy-load bridges in cold regions according to claim 4, characterized in that: In the process of solving the formula corresponding to the L1-L2 hybrid regularized model for the collaborative inversion of star-rocket dual-source data, the Alternating Direction Multiplier Method (ADMM) is used to solve the global optimal solution of the objective function.

6. A method for identifying damage driven by star-vehicle cooperative physical constraints on heavy-load bridges in cold regions according to claim 4 or 5, characterized in that: The damage assessment and uncertainty quantification process is completed by using the L1-L2 hybrid regularization model based on the collaborative inversion of star-vehicle dual-source data. This process involves damage identification and uncertainty quantification assessment through the damage assessment and output module. The damage identification process involves determining the degree of damage based on the stiffness degradation parameters of each bridge unit obtained from the data. At the same time, the calculated stiffness degradation parameters of each bridge unit are compared with the measured results of the static load test to determine the error rate, thereby verifying the accuracy of the damage assessment. The uncertainty quantification process involves quantifying the uncertainty of the inversion results using the Monte Carlo-response surface methodology, classifying the damage level based on the degree of stiffness degradation, and outputting a bridge damage assessment report that includes damage location, damage degree, confidence interval, and maintenance recommendations.

7. The method for identifying damage driven by star-vehicle cooperative physical constraints on heavy-load bridges in cold regions according to claim 6, characterized in that: In the uncertainty quantification process, the Monte Carlo random field generation process is a joint distribution of simulation process noise and observation noise. A surrogate model is established to quickly assess the sensitivity of parameter identification results to noise and output confidence intervals. Simulation process noise refers to model error, and observation noise refers to sensor error. Based on the sensitivity parameter threshold, a damage mutation rate exponent is set, and the weight of historical data is dynamically adjusted. The formula for calculating the forgetting factor, which enhances the tracking ability of time-varying parameters, is as follows: ； In the above formula, α∈(0,1) and tr represents the trace of the matrix.

8. The method for identifying damage driven by star-vehicle cooperative physical constraints on heavy-load bridges in cold regions according to claim 1, characterized in that: The spaceborne-vehicle dual-layer collaborative sensing system uses multi-source data. During the preprocessing of the multi-source data, a variational mode decomposition and wavelet thresholding joint denoising method is used to separate environmental noise and the vehicle's inherent vibration components. A physically constrained variational autoencoder is constructed to reconstruct the wheel axle contact point response. After extracting a multi-dimensional damage-sensitive feature set, the adaptive compensation process for cold-region temperature-road roughness coupling interference is completed through transfer learning using the multi-dimensional damage-sensitive feature set. The multi-dimensional damage-sensitive feature set includes frequency domain modal features, time domain statistical features, and time-frequency domain energy features. The frequency domain modal features include frequency shift and mode shape curvature, the time domain statistical features include root mean square response and peak factor, and the time-frequency domain energy features include wavelet energy entropy and Hilbert-Huang transform marginal spectrum.

Images