Bridge expansion joint state identification and early warning method based on bayesian data fusion

Abstract

This invention discloses a method for bridge expansion joint status identification and early warning based on Bayesian data fusion, comprising: S1, constructing a disaster-causing factor system and a failure cause database for expansion joints; S2, collecting multi-dimensional data of expansion joints; S3, preprocessing and extracting features from the collected multi-dimensional data; S4, constructing a Bayesian network model based on a fault tree; S5, fusing multi-dimensional data to determine the prior probability of the Bayesian network model; S6, constructing a conditional probability table for each node of the Bayesian network based on maximum a posteriori estimation and the analytic hierarchy process; S7, quantifying the failure probability of expansion joints through forward inference using the Bayesian network, classifying risk levels and risk warnings; and screening the main disaster-causing factors through backward inference using the Bayesian network. This invention integrates multi-dimensional data and multi-disciplinary theoretical methods, solving the problems of single data dimensions, low fusion efficiency, and insufficient early warning accuracy in existing technologies, and significantly improving the accuracy of expansion joint damage identification and the reliability of risk assessment.

Description

Technical Field

[0001] This invention belongs to the field of bridge structural health monitoring technology, specifically relating to a method for identifying and warning of the condition of bridge expansion joints based on Bayesian data fusion. Background Technology

[0002] Bridge expansion joints are core load-bearing components connecting bridge beams, and their functional integrity directly determines the overall structural integrity, driving comfort, and service safety of the bridge. These joints are subjected to the combined effects of vehicle load impacts, temperature cycles, rain and snow erosion, and vibration fatigue over extended periods. Key connecting components are prone to loosening, slippage, and even detachment, gradually leading to the failure of the expansion joint. This can trigger a chain reaction, such as stress concentration and concrete damage at the beam connection points, ultimately threatening the safety of the bridge structure and potentially causing traffic accidents. Therefore, achieving accurate condition monitoring and early warning for modular bridge expansion joints is a crucial requirement for ensuring long-term safe service and reducing maintenance costs.

[0003] Current technologies for detecting and warning of bridge expansion joints still have significant limitations: They rely primarily on manual inspections, which are not only inefficient, labor-intensive, and costly, but also highly susceptible to subjective factors. This results in low sensitivity and a high rate of missed detections for early-stage minor faults such as loose bolts, failing to meet the routine monitoring needs of large-scale bridges. Existing automated detection technologies largely depend on single-sensor data collection, providing limited data dimensions that fail to comprehensively reflect the complex service conditions of expansion joints and are easily affected by environmental noise, leading to insufficient detection accuracy. Furthermore, some multi-source data fusion methods suffer from low fusion efficiency and a lack of theoretical support and expert experience in parameter learning, failing to effectively integrate multi-dimensional disaster-causing factors such as visual, acoustic, and vibration data, resulting in insufficient accuracy and timeliness in condition assessment and early warning. Summary of the Invention

[0004] The main objective of this invention is to overcome the shortcomings and deficiencies of the prior art and propose a method for identifying and warning the state of bridge expansion joints based on Bayesian data fusion.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for bridge expansion joint status identification and early warning based on Bayesian data fusion includes the following steps:

[0007] S1. Construct a disaster-causing factor system and a failure cause database for bridge expansion joints;

[0008] S2. Collect multi-dimensional data of bridge expansion joints;

[0009] S3. Preprocess and extract features from the collected multi-dimensional data;

[0010] S4. Construct a Bayesian network model based on fault trees;

[0011] S5. Multi-dimensional data fusion to determine the prior probability of the Bayesian network model;

[0012] S6. Construct conditional probability tables for each node of a Bayesian network based on maximum a posteriori estimation and analytic hierarchy process.

[0013] S7. Quantify the failure probability of expansion joints through forward inference using Bayesian networks, classify risk levels and risk warnings; screen the main disaster-causing factors through backward inference using Bayesian networks, providing a decision-making basis for targeted maintenance of bridge expansion joints.

[0014] This invention also includes a bridge expansion joint status identification and early warning system based on Bayesian data fusion. The system adopts the bridge expansion joint status identification and early warning method based on Bayesian data fusion provided by this invention. The system includes: a disaster-causing factor system and fault cause database construction module, a multi-dimensional data acquisition module, a data preprocessing and feature extraction module, a Bayesian network construction module, a conditional probability table construction module, and a risk early warning module.

[0015] The module for constructing the disaster-causing factor system and failure cause database is used to build the disaster-causing factor system and failure cause database for bridge expansion joints.

[0016] A multi-dimensional data acquisition module is used to collect multi-dimensional data of bridge expansion joints;

[0017] The data preprocessing and feature extraction module is used to preprocess and extract features from the collected multi-dimensional data;

[0018] The Bayesian network building block is used to build Bayesian network models based on fault trees.

[0019] The conditional probability table construction module is used to perform multi-dimensional data fusion and determine the prior probability of the Bayesian network model; it constructs the conditional probability table of each node of the Bayesian network based on maximum a posteriori estimation and the analytic hierarchy process.

[0020] The risk warning module quantifies the failure probability of expansion joints through forward inference using Bayesian networks, classifies risk levels and provides risk warnings; it also uses backward inference using Bayesian networks to screen the main disaster-causing factors, providing a basis for decision-making for targeted maintenance of bridge expansion joints.

[0021] The present invention also includes a computer-readable storage medium storing a program that, when executed by a processor, implements the bridge expansion joint status identification and early warning method based on Bayesian data fusion provided by the present invention.

[0022] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0023] 1. Comprehensive coverage of disaster-causing factors and data dimensions: The present invention systematically constructs a multi-dimensional disaster-causing factor system encompassing environment, load, and structure, integrating multi-source detection data from vision, acoustics, and vibration to comprehensively capture the coupled causes and multi-dimensional characterization of damage to expansion joints, thus solving the problems of single data and incomplete information coverage in existing technologies.

[0024] 2. High efficiency and accuracy in data processing and fusion; This invention optimizes the Gold-YOLO model through an innovative aggregation-distribution mechanism, and combines targeted data processing techniques such as inverse perspective mapping, wavelet denoising, and Hilbert-Huang transform to effectively eliminate interference factors and improve feature extraction accuracy; At the same time, it achieves deep fusion of multi-source data based on Bayesian networks, taking into account both data statistical regularities and expert experience, and improving the reliability of fusion results.

[0025] 3. The early warning model combines theoretical rigor with practicality. This invention adopts a fault tree-Bayes network transformation model, combined with MAP estimation and AHP parameter learning methods. This ensures the theoretical rigor of the model and adapts it to engineering practice through expert experience, thus solving the defects of existing models with strong subjectivity or detachment from engineering scenarios.

[0026] 4. Achieving risk quantification and main control factor localization: This invention quantifies the failure probability and classifies the risk level through forward reasoning using Bayesian networks, and screens the main control disaster-causing factors through reverse reasoning. This not only enables early warning, but also provides precise targeting for maintenance decisions, solving the problem that existing technologies can only make qualitative judgments and lack targeted maintenance.

[0027] 5. Strong engineering adaptability and wide application scenarios: This invention adopts a vehicle-mounted detection system combined with a variable frequency acquisition strategy to adapt to the needs of large-scale and routine monitoring; the method is applicable to mainstream modular expansion joints such as comb-tooth plate type, and can be widely used in the health monitoring of various bridges, with significant engineering practical value and promotion prospects. Attached Figure Description

[0028] Figure 1 This is a flowchart of the method of the present invention.

[0029] Figure 2 This is a flowchart of the multi-dimensional data fusion and Bayesian network parameter learning of the present invention.

[0030] Figure 3 This is a schematic diagram of the fault tree model of the modular telescopic device in the embodiment.

[0031] Figure 4 This is a schematic diagram of the Bayesian network model of the modular scaling device in the embodiment.

[0032] Figure 5 This is a schematic diagram of inference warning and maintenance decision-making in the embodiment.

[0033] Figure 6 This is a schematic diagram of the system according to an embodiment of the present invention. Detailed Implementation

[0034] like Figure 1 As shown, the bridge expansion joint status identification and early warning method based on Bayesian data fusion includes the following steps:

[0035] S1. Construct a disaster-causing factor system and a failure cause database for bridge expansion joints; specifically:

[0036] S11. Review the service environment, stress characteristics and structural composition of bridge expansion joints, identify disaster-causing factors and classify them to construct a multi-dimensional disaster-causing factor system. Disaster-causing factors include environmental disaster-causing factors, load disaster-causing factors and structural disaster-causing factors. Among them, environmental disaster-causing factors include temperature cycling, rain and snow erosion and humidity changes.

[0037] Load-related disaster factors include: vehicle impact load, traffic flow, and load frequency;

[0038] Structural disaster-causing factors include: material fatigue, component aging, and installation deviations;

[0039] S12. Based on the constructed multi-dimensional disaster-causing factor system, collect the types of bridge expansion joint failures and their corresponding causes, and establish a failure cause database. The failure types include bolt loosening, bolt breakage, comb plate deformation, and anchorage structure failure. The methods for collecting typical failure types and their corresponding causes of bridge expansion joint failures include literature review, engineering disease statistics, and expert interviews.

[0040] S2. Collect multi-dimensional data on bridge expansion joints; specifically:

[0041] Based on the constructed disaster-causing factor system, multi-dimensional data acquisition equipment was deployed to carry out on-site data collection and form a multi-dimensional raw dataset;

[0042] The multi-dimensional raw dataset includes multi-dimensional data on the condition of bridge expansion joints under various conditions, including intact condition, slightly loose bolts, severely loose bolts, broken bolts, and deformed comb plates. The multi-dimensional data is specifically divided into visual data, acoustic data, and vibration data, and the acquisition methods are as follows:

[0043] Visual data acquisition: A vehicle-mounted inspection system is used, with an industrial camera obliquely fixed to the top of the inspection vehicle and angled downwards towards the bridge expansion joint area. A variable frequency image acquisition strategy is adopted, adjusting the acquisition frequency to 1-5 frames / second according to traffic flow to obtain visual images of the expansion joint surface and bolts.

[0044] Acoustic data acquisition: At key stress points of the telescopic device, such as the joint of the comb plate and the bolt fixing end, acoustic acquisition equipment is deployed. Microphone arrays or acoustic sensors are used to collect the acoustic signals generated by the telescopic device when the vehicle passes through the gap. The sampling frequency is set to 2kHz-5kHz.

[0045] Vibration data acquisition: Accelerometers are installed at the anchoring structure of the expansion joint and the beam connection to collect vibration signals generated by vehicle impact and structural vibration. The sampling frequency is set to 1kHz-2kHz.

[0046] S3. Preprocess and extract features from the collected multi-dimensional data; specifically:

[0047] For the multi-dimensional data collected in step S2, targeted preprocessing and feature extraction methods are used to obtain the damage-sensitive features of each dimension. Based on the damage-sensitive features, the prior probability parameters of the subsequent Bayesian network model are calculated. The preprocessing and feature extraction of the multi-dimensional data are as follows:

[0048] Visual data: Inverse perspective mapping (IPM) technology is used to convert oblique views into bird's-eye views, eliminating shooting perspective bias; bolt target detection is achieved using the Gold-YOLO model, and the corner coordinates and center point offset of the bolts are extracted using the Harris corner detection algorithm; the center point coordinates of the bolts in the time-series images are defined as follows: (Image of frame i) and (For the j-th frame image), the offset d of the bolt center point is calculated according to formula (1):

[0049] (1);

[0050] in, , , , These are the pixel coordinates of the bolt center point in images at different times, which are converted into physical coordinates using a pixel-actual size calibration coefficient.

[0051] Acoustic data: Wavelet denoising technology is used to eliminate environmental noise interference. The discrete wavelet transform is defined as:

[0052] (2);

[0053] in, Here, represents the wavelet coefficients, and 'a' represents the scaling factor. The translation factor is... The original acoustic signal, wavelet basis functions The conjugate function;

[0054] The denoised acoustic signal is decomposed into time and frequency domains using the Hilbert-Huang Transform (HHT). First, the signal is decomposed into n Intrinsic Mode Functions (IMFs) using Empirical Mode Decomposition (EMD), satisfying the following IMF selection criteria:

[0055] The number of local extrema of the signal over the entire time series is equal to or differs from the number of zero crossings by no more than 1, and the average value of the upper and lower envelopes of the signal at any time is 0.

[0056] The Hilbert transform of each IMF component is defined as follows:

[0057] (3);

[0058] in, For signal The Hilbert transform result, where PV represents the Cauchy principal value, is used to handle the integrand in... The singularity at the point guarantees the convergence of the integral;

[0059] Sensitive feature parameters, including peak value P and root mean square value, are extracted based on the Hilbert transform results. and clock speed Root mean square value Represented as:

[0060] (4);

[0061] Where N is the number of signal sampling points, Let the signal amplitude be the value at the k-th sampling point;

[0062] Vibration data: Wavelet packet decomposition technology was used to decompose the vibration signal at multiple scales, and effective frequency band components were selected. Peak values ​​were extracted by combining time-domain analysis. Kurtosis K and skewness S are used to extract the power spectral density through frequency domain analysis. :

[0063] (5);

[0064] (6);

[0065] (7);

[0066] Where μ is the mean of the vibration signal, and σ is the standard deviation of the signal. This is the Fourier transform result for a signal duration of T, where f is the frequency.

[0067] S4. Construct a Bayesian network model based on fault trees; specifically:

[0068] Taking the failure warning of the expansion joint as the top event, based on the disaster-causing factor system and failure cause library constructed in step S1, and combined with the structural composition and damage evolution mechanism of the expansion joint, a fault tree model is established, specifically as follows:

[0069] Fault tree top event: Bridge expansion joint failure;

[0070] Intermediate events in the fault tree: bolt damage, comb plate damage, anchoring structure failure, etc.

[0071] Fault tree base events: environmental disaster-causing factors, load-causing disaster-causing factors, and structural disaster-causing factors;

[0072] The fault tree model is transformed into a Bayesian network model, clarifying the top, middle, and bottom events of the network nodes and the causal relationships and logical connections between nodes. In the Bayesian network, the nodes correspond to the events in the fault tree, and the directed edges between the nodes correspond to the causal relationships between the events.

[0073] S5. Multi-dimensional data fusion to determine the prior probabilities of the Bayesian network model; specifically including:

[0074] Multi-dimensional feature standardization: For the three-dimensional sensitive features extracted in step S3, Z-score standardization is used to eliminate dimensional differences. The standardization formula is:

[0075] (8);

[0076] in, σ is the dimensionless value of the feature after standardization; x is the original value of the feature; μ is the mean of the feature in the full dataset; σ is the standard deviation of the feature in the full dataset.

[0077] The three-dimensional sensitive features are as follows:

[0078] Visual feature: Bolt center point offset d;

[0079] Vibrational characteristics: kurtosis K and peak power spectral density ;

[0080] Acoustic characteristics: root mean square and clock speed ;

[0081] Feature layer dimensionality reduction and redundancy removal: For the standardized multi-dimensional feature vectors, the dimensions are 5: d, K, ... , , Principal component analysis was used for dimensionality reduction to retain core information;

[0082] The single-dimensional damage recognition probability calculation is based on the dimensionality-reduced features of each dimension: visually independent features, acoustically independent features, and vibration-independent features. Three Support Vector Machine (SVM) damage recognition models are trained respectively, outputting the independent damage recognition probability for each dimension. The SVM decision function is defined as follows:

[0083] (9);

[0084] in, For the Lagrange multipliers corresponding to the support vectors, For sample labels; The radial basis function kernel is calculated using the following formula: ,in γ represents the support vector samples in the SVM training set, x represents the multi-source data feature samples of the bridge expansion joint to be identified, and γ represents the kernel function bandwidth parameter. For classification threshold, It is a symbolic function;

[0085] Bayesian probabilistic fusion is used to calculate prior probabilities. Using single-dimensional damage recognition probabilities as observational evidence, Bayesian probabilistic fusion rules are employed to integrate multi-source information, yielding the final prior probabilities of intermediate events in the Bayesian network. The fusion formula is:

[0086] (10);

[0087] in, The initial prior probability of intermediate event A is obtained through statistical data on engineering defects, such as the historical probability of bolt damage being 5%.

[0088] Let A be the initial prior probability that intermediate event A does not occur.

[0089] The probability that the model misclassifies the m-th dimension as damage when intermediate event A does not occur is calculated using the validation set and is usually taken as 5% to 10%, while this invention takes 8%.

[0090] S6. Construct conditional probability tables for each node of the Bayesian network based on maximum a posteriori estimation and the analytic hierarchy process; specifically:

[0091] A parameter learning method combining Maximum A Posteriori (MAP) estimation and Analytic Hierarchy Process (AHP) is employed to construct a Conditional Probability Table (CPT) for each node of the Bayesian network. Specifically:

[0092] Quantification of expert experience: By inviting experts in bridge engineering and structural health monitoring, the Analytic Hierarchy Process (AHP) was used to score the correlation strength of each node under different combinations of parent node states, and the initial conditional probability was determined.

[0093] Statistical learning optimization: Based on the feature dataset and prior probabilities extracted in step S3, the initial conditional probabilities are iteratively optimized using maximum a posteriori estimation (MAP). The bias of expert experience is corrected by combining data statistical laws, resulting in a conditional probability table (CPT) that balances theoretical support and engineering practice, quantifying the conditional probability distribution of each node under different combinations of parent node states.

[0094] S7. Quantify the failure probability of bridge expansion joints through forward inference using Bayesian networks, classify risk levels and issue risk warnings; screen the main disaster-causing factors through backward inference using Bayesian networks, providing a decision-making basis for targeted maintenance of bridge expansion joints; specifically including:

[0095] Forward reasoning enables risk level classification and early warning, specifically as follows:

[0096] Inputting Bayesian network prior probability data calculated from real-time collected multi-dimensional monitoring data, the probability of top event expansion joint failure is calculated through forward inference using the Bayesian network. Combining this with the geographical spatial unit where the bridge is located, risk level thresholds are set to complete the risk level classification and early warning for each geographical spatial unit. The specific risk level thresholds are:

[0097] Low risk: Failure probability <10%;

[0098] Medium risk: 10% ≤ probability of failure < 30%;

[0099] High risk: 30% ≤ failure probability < 60%;

[0100] Extremely high risk: failure probability ≥ 60%;

[0101] The main control factor is selected through reverse reasoning, specifically as follows:

[0102] For high-risk and extremely high-risk geospatial units, the contribution of each bottom event (disaster-causing factor) to the top event (expansion joint failure) is calculated by Bayesian network inverse reasoning. The main disaster-causing factors are then sorted by contribution to provide a basis for targeted maintenance decisions.

[0103] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0104] Example; This example uses a modular expansion joint warning system on a highway bridge as an example, and includes the following steps:

[0105] Step 1: Construction of the disaster-causing factor system and failure cause database;

[0106] The service environment and structural characteristics of the modular expansion joint of the bridge were analyzed to identify multidimensional disaster-causing factors, as shown in Table 1 below.

[0107] Table 1 Multidimensional disaster-causing factors

[0108]

[0109] Step 2; On-site monitoring and multi-dimensional data collection, including:

[0110] Visual data acquisition: An industrial camera with a resolution of 1920×1080 is fixed at a 30° downward angle on the top of the inspection vehicle. The vehicle travels along the bridge lane at a speed of 30km / h and acquires visual images of the telescopic device at a frequency of 3 frames / second. A total of 500 images of each working condition, such as intact, loose bolts, and broken bolts, are acquired.

[0111] Acoustic data acquisition: One acoustic sensor is installed on each of the modular fixed ends on both sides of the telescopic device, with a sampling frequency of 3kHz, to collect acoustic signals when vehicles pass by, and the data is continuously collected for 3 hours to obtain acoustic data under different traffic loads.

[0112] Vibration data acquisition: A 3-axis accelerometer (sampling frequency 15kHz) is installed at the anchoring steel plate of the expansion joint to collect vibration signals when vehicles pass through the gap, and traffic flow and vehicle type information are recorded simultaneously.

[0113] Step 3; Multidimensional data preprocessing and feature extraction;

[0114] Visual data processing: Inverse perspective mapping (IPM) technology is used to transform the coordinates of the oblique angle images to eliminate the shooting angle deviation and obtain a bird's-eye view that meets the engineering inspection standards; the bolt target detection is achieved by improving the Gold-YOLO model, and the detection confidence is ≥0.85; the corner point detection algorithm of Harris is used to extract the bolt corner point coordinates, and the relative offset of the bolt center point in different time series images is calculated according to formula (1): the pixel coordinates of the bolt center point in the 100th frame image are known to be (x1=820, y1=645), and the pixel coordinates of the same bolt center point in the 200th frame image are (x2=823, y2=647). The pixel-to-actual size conversion coefficient is determined to be 0.1mm / pixel by field calibration. Substituting into formula (1), the relative offset of the bolt center point is 0.36mm.

[0115] Right now .

[0116] Acoustic data processing: The db4 wavelet basis was used for three-layer wavelet denoising. The wavelet coefficients were calculated according to formula (2) and the effective components were screened. After denoising, the signal-to-noise ratio was increased from 28dB to 42dB. The denoised acoustic signal was decomposed into 6 intrinsic mode functions by empirical mode decomposition. The Hilbert transform of each IMF component was performed according to formula (3) to extract the sensitive feature parameters: peak value 0.25Pa, root mean square value 0.08Pa, and main frequency 3.2kHz.

[0117] Vibration data processing: Wavelet packet decomposition technology was used to decompose the vibration signal into 8-level multi-scale components and screen the effective components in the 4-8kHz frequency band. Combined with time-domain and frequency-domain analysis, characteristic parameters were extracted: peak value 1.2m / s², kurtosis 3.8, and peak power spectral density 35dB / Hz.

[0118] Step 4; Bayesian network model construction;

[0119] The "failure warning" of the expansion joint is the top event. Based on the failure cause library and disaster factor system constructed in step 1, combined with the structural composition and damage evolution mechanism of the expansion joint, a fault tree model is established: Fault tree top event: failure of bridge modular expansion joint (E).

[0120] Intermediate events in the fault tree: bolt damage (A, including loosening and breakage), modular damage (B, including deformation and corrosion), anchorage structure failure (C), etc.

[0121] Fault tree base events: environmental disaster factors (F1: temperature cycling, F2: rain and snow erosion), load disaster factors (F3: vehicle impact, F4: traffic flow), structural disaster factors (F5: material fatigue, F6: installation deviation), etc.

[0122] The fault tree model described above is transformed into a Bayesian network model, clarifying the network nodes (top event, intermediate events, and bottom events) and the causal relationships and logical connections between nodes. In this model, the nodes of the Bayesian network correspond to the events in the fault tree, and the directed edges between nodes correspond to the causal relationships between events. The network structure satisfies the following conditions: V is the set of nodes, and E is the set of directed edges.

[0123] like Figure 3 The diagram shown is a schematic of a fault tree model for a modular expansion joint. Figure 4 The figure shown is a schematic diagram of the Bayesian network model of the modular scaling device.

[0124] Step 5: Multidimensional data fusion, calculating the prior probability of the Bayesian network;

[0125] Feature standardization processing: Five sensitive features extracted from visual, vibration, and acoustic dimensions (bolt center point offset d, vibration kurtosis K, vibration power spectral density peak value) Root Mean Square Acoustics Acoustic main frequency Z-score standardization is performed according to formula (8) to eliminate dimensional differences:

[0126] Taking the bolt center point offset d as an example, the original value is 0.36mm, the mean of the entire dataset is μ=0.28mm, and the standard deviation is σ=0.15mm. After standardization: After standardization, all feature values ​​are mapped to the interval [-3,3], with mean ≈0 and standard deviation ≈1.

[0127] Feature layer dimensionality reduction and redundancy removal: Principal component analysis (PCA) is performed on the standardized 5-dimensional feature vectors to reduce dimensionality. The feature covariance matrix Σ is calculated, and the eigenvalues ​​λ1=2.86, λ2=1.75, λ3=0.92, λ4=0.38, and λ5=0.19 are obtained. Principal components are selected based on a cumulative contribution rate η≥90%, and the cumulative contribution rate of the first two principal components is:

[0128] ;

[0129] The requirements are met; construct the principal component matrix Φ=[φ1,φ2] (φ1 and φ2 are the eigenvectors corresponding to the first two eigenvalues), and calculate the fusion eigenvectors after dimensionality reduction. .

[0130] Single-dimensional damage recognition probability calculation: Three support vector machine damage recognition models are trained based on independent features of each dimension.

[0131] Visual Dimension Model: Input features are The training set to validation set ratio was 7:3. A radial basis function kernel was used, and the parameters γ=0.8 and the penalty coefficient C=10 were optimized through grid search. The validation set recognition accuracy was 93%.

[0132] Vibration Dimension Model: Input features are , The validation set recognition accuracy was 89%.

[0133] Acoustic dimensional model: Input features are , The validation set recognition accuracy is 87%; the trained model outputs the single-dimensional recognition probability of intermediate event A: , , .

[0134] Bayesian probability fusion calculates prior probabilities, specifically as follows:

[0135] Determine the initial prior probability (Based on historical damage statistics of the bridge's expansion joint). ; Calculate the false positive probability of the model for each dimension using the validation set. (i.e., the probability that the model misclassifies a bolt as damaged when it is undamaged); perform multi-source data fusion calculation according to formula (10):

[0136] ;

[0137] The prior probability of intermediate event A (bolt damage) in the Bayesian network is 89%. Similarly, following the above process, the prior probabilities of intermediate event B (modular damage) (12%), intermediate event C (anchoring structure failure) (8%), and the prior probabilities of basic events F1-F6 (e.g., F3 = high prior probability 35%, F5 = mild prior probability 60%) are calculated, forming a complete Bayesian network prior probability set.

[0138] like Figure 2 The diagram shows a flowchart of multi-dimensional data fusion and Bayesian network parameter learning.

[0139] Step 6; Construct the Bayesian network conditional probability table (CPT);

[0140] Expert experience quantification (AHP method): Five senior experts in bridge engineering and structural health monitoring were invited to score the correlation strength between each node and its parent node, construct a judgment matrix and calculate the weight vector;

[0141] Statistical learning optimization (MAP estimation): Based on the obtained feature dataset and prior probability set, the initial conditional probability is optimized using maximum a posteriori estimation (MAP).

[0142] The prior distribution of the parameters is set to a Dirichlet distribution with distribution parameter α = 1.0;

[0143] Using formula (10) as the objective function, the initial conditional probability is corrected through iterative calculations, with 100 iterations. For example, when the parent node state combination F3 = high and F5 = severe, the conditional probability of child node A = fracture is corrected to 82%.

[0144] Complete the conditional probability calculation for all nodes using the above process to construct the complete Bayesian network conditional probability table (CPT).

[0145] Step 7; Implementation of reasoning and early warning;

[0146] Forward reasoning: Input real-time data (F1=high, F2=medium, F3=medium, F4=severe), and calculate the probability of E (expansion device failure) through forward reasoning. The probability is 78%, which corresponds to an extremely high risk level, triggering an early warning.

[0147] Reverse reasoning: Reverse reasoning was performed on the geospatial unit with extremely high risk to calculate the contribution of each underlying event: F1 (vehicle impact load) contributed 42% and F4 (bolt material fatigue) contributed 35%. The main disaster-causing factors of this unit were determined to be vehicle impact load and bolt material fatigue. Maintenance measures such as strengthening bolt fastening and replacing bolts with high-strength bolts were recommended.

[0148] Targeted maintenance recommendations: Emergency measures: Tighten all fixing bolts of the expansion joint within 72 hours, replace 3 severely loose bolts with a center point offset >2mm, and after replacement, use a torque wrench to check the bolt tightening torque, which should be ≥300N·m;

[0149] Long-term optimization measures: Limit the speed of heavy vehicles on this section of the road to 80km / h to reduce the impact of vehicle impact loads on the expansion joints; replace all bolts on the expansion joints within one year, and select corrosion-resistant high-strength bolts with a strength grade of 8.8.

[0150] Monitoring scheme optimization: Increase the monitoring frequency of the telescopic device, adjust the visual data acquisition frequency to 5 frames / second, conduct a complete multi-source data acquisition and early warning assessment once per quarter, and collect core data monthly for other periods, combining vibration and acoustics for dynamic monitoring.

[0151] like Figure 5 The diagram shown illustrates the reasoning, early warning, and maintenance decision-making processes.

[0152] Verification through the above embodiments shows that the present invention can accurately identify the damage status of expansion joints, quantify failure risks, locate the main disaster-causing factors, and achieve an early warning accuracy rate of over 90%, providing effective technical support for the safe service and targeted maintenance of modular expansion joints for bridges.

[0153] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously.

[0154] Based on the same idea as the bridge expansion joint state identification and early warning method based on Bayesian data fusion in the above embodiments, this invention also provides a bridge expansion joint state identification and early warning system based on Bayesian data fusion. This system can be used to execute the aforementioned bridge expansion joint state identification and early warning method based on Bayesian data fusion. For ease of explanation, the structural diagram of the embodiment of the bridge expansion joint state identification and early warning system based on Bayesian data fusion only shows the parts related to the embodiments of this invention. Those skilled in the art will understand that the illustrated structure does not constitute a limitation on the device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0155] like Figure 6 As shown, the bridge expansion joint status identification and early warning system 100 based on Bayesian data fusion includes a disaster factor system and fault cause library construction module 101, a multi-dimensional data acquisition module 102, a data preprocessing and feature extraction module 103, a Bayesian network construction module 104, a conditional probability table construction module 105, and a risk early warning module 106.

[0156] The module for constructing the disaster-causing factor system and failure cause database is used to build the disaster-causing factor system and failure cause database for bridge expansion joints.

[0157] A multi-dimensional data acquisition module is used to collect multi-dimensional data of bridge expansion joints;

[0158] The data preprocessing and feature extraction module is used to preprocess and extract features from the collected multi-dimensional data;

[0159] The Bayesian network building block is used to build Bayesian network models based on fault trees.

[0160] The conditional probability table construction module is used to perform multi-dimensional data fusion and determine the prior probability of the Bayesian network model; it constructs the conditional probability table of each node of the Bayesian network based on maximum a posteriori estimation and the analytic hierarchy process.

[0161] The risk warning module quantifies the failure probability of expansion joints through forward inference using Bayesian networks, classifies risk levels and provides risk warnings; it also uses backward inference using Bayesian networks to screen the main disaster-causing factors, providing a basis for decision-making for targeted maintenance of bridge expansion joints.

[0162] It should be noted that the bridge expansion joint status identification and early warning system based on Bayesian data fusion of the present invention corresponds one-to-one with the bridge expansion joint status identification and early warning method based on Bayesian data fusion of the present invention. The technical features and beneficial effects described in the embodiments of the bridge expansion joint status identification and early warning method based on Bayesian data fusion are applicable to the embodiments of the bridge expansion joint status identification and early warning system based on Bayesian data fusion. For details, please refer to the description in the embodiments of the method of the present invention, which will not be repeated here.

[0163] Furthermore, in the above-described implementation of the bridge expansion joint status identification and early warning system based on Bayesian data fusion, the logical division of each program module is merely an example. In actual applications, the above functions can be assigned to different program modules as needed, for example, for the sake of corresponding hardware configuration requirements or the convenience of software implementation. That is, the internal structure of the bridge expansion joint status identification and early warning system based on Bayesian data fusion can be divided into different program modules to complete all or part of the functions described above.

[0164] In another embodiment, a computer-readable storage medium is provided, storing a program that, when executed by a processor, implements the bridge expansion joint state identification and early warning method based on Bayesian data fusion of the present invention, specifically as follows:

[0165] Construct a disaster-causing factor system and a failure cause database for bridge expansion joints;

[0166] Collect multi-dimensional data on bridge expansion joints;

[0167] Preprocessing and feature extraction are performed on the collected multi-dimensional data;

[0168] Construct a Bayesian network model based on fault trees;

[0169] Multi-dimensional data fusion to determine the prior probability of the Bayesian network model;

[0170] Conditional probability tables for each node of a Bayesian network are constructed based on maximum a posteriori estimation and the analytic hierarchy process.

[0171] By using Bayesian network forward reasoning to quantify the failure probability of expansion joints and classify risk levels and risk warnings, and by using Bayesian network backward reasoning to screen the main disaster-causing factors, a decision-making basis is provided for targeted maintenance of bridge expansion joints.

[0172] The computer-readable storage medium may be transient or non-transient. Exemplary examples include, but are not limited to, various media capable of storing computer program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0173] For example, the processor may be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0174] It should also be noted that, in this specification, terms such as "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0175] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for bridge expansion joint status identification and early warning based on Bayesian data fusion, characterized in that, Includes the following steps: S1. Construct a disaster-causing factor system and a failure cause database for bridge expansion joints; S2. Collect multi-dimensional data of bridge expansion joints; S3. Preprocess and extract features from the collected multi-dimensional data; S4. Construct a Bayesian network model based on fault trees; S5. Multi-dimensional data fusion to determine the prior probability of the Bayesian network model; S6. Construct conditional probability tables for each node of a Bayesian network based on maximum a posteriori estimation and analytic hierarchy process. S7. Quantify the failure probability of the scaling device through forward inference using Bayesian network, and classify the risk level and risk warning; By using Bayesian network reverse reasoning to screen the main disaster-causing factors, a decision-making basis is provided for the targeted maintenance of bridge expansion joints.

2. The method for bridge expansion joint status identification and early warning based on Bayesian data fusion according to claim 1, characterized in that, Step S1 is as follows: S11. Review the service environment, stress characteristics and structural composition of bridge expansion joints, identify disaster-causing factors and classify them to construct a multi-dimensional disaster-causing factor system. Disaster-causing factors include environmental disaster-causing factors, load disaster-causing factors and structural disaster-causing factors. Among them, environmental disaster-causing factors include temperature cycling, rain and snow erosion and humidity changes. Load-related disaster factors include: vehicle impact load, traffic flow, and load frequency; Structural disaster-causing factors include: material fatigue, component aging, and installation deviations; S12. Based on the constructed multi-dimensional disaster-causing factor system, collect the types of bridge expansion joint failures and their corresponding causes, and establish a failure cause database; failure types include, but are not limited to, bolt loosening, bolt breakage, comb plate deformation, and anchorage structure failure; methods for collecting typical failure types of bridge expansion joints and their corresponding causes include, but are not limited to, literature review, engineering disease statistics, and expert interviews.

3. The method for bridge expansion joint status identification and early warning based on Bayesian data fusion according to claim 2, characterized in that, Step S2 is as follows: Based on the constructed disaster-causing factor system, multi-dimensional data acquisition equipment was deployed to carry out on-site data collection and form a multi-dimensional raw dataset; The multi-dimensional raw dataset includes multi-dimensional data on the condition of bridge expansion joints under various conditions, including intact condition, slightly loose bolts, severely loose bolts, broken bolts, and deformed comb plates. The multi-dimensional data is specifically divided into visual data, acoustic data, and vibration data, and the acquisition methods are as follows: Visual data acquisition: A vehicle-mounted inspection system is used, with an industrial camera obliquely fixed to the top of the inspection vehicle and angled downwards towards the bridge expansion joint area. A variable frequency image acquisition strategy is used to obtain visual images of the expansion joint surface and bolts. Acoustic data acquisition: Acoustic acquisition equipment is installed at key stress points of the expansion joint to collect acoustic signals generated by the expansion joint when vehicles pass through the gap. The sampling frequency is set to 2kHz-5kHz. Vibration data acquisition: Accelerometers are installed at the anchoring structure of the expansion joint and the beam connection to collect vibration signals generated by vehicle impact and structural vibration. The sampling frequency is set to 1kHz-2kHz.

4. The method for bridge expansion joint status identification and early warning based on Bayesian data fusion according to claim 1, characterized in that, Step S3 is as follows: For the multi-dimensional data collected in step S2, targeted preprocessing and feature extraction methods are used to obtain the damage-sensitive features of each dimension. Based on the damage-sensitive features, the prior probability parameters of the subsequent Bayesian network model are calculated. The preprocessing and feature extraction of the multi-dimensional data are as follows: Visual data: Inverse perspective mapping (IPM) technology is used to convert oblique views into bird's-eye views, eliminating shooting perspective bias; bolt target detection is achieved using the Gold-YOLO model, and the corner coordinates and center point offset of the bolts are extracted using the Harris corner detection algorithm; the center point coordinates of the bolts in the time-series images are defined as follows: and Then the offset d of the bolt center point is calculated according to formula (1): (1)； in, , , , These are the pixel coordinates of the bolt center point in images at different times, which are converted into physical coordinates using a pixel-actual size calibration coefficient. Acoustic data: Wavelet denoising technology is used to eliminate environmental noise interference. The discrete wavelet transform is defined as: (2)； in, Here, represents the wavelet coefficients, and 'a' represents the scaling factor. The translation factor is... The original acoustic signal, wavelet basis functions The conjugate function; The denoised acoustic signal is decomposed into time and frequency domains using Hilbert-Huang Transform (HHT). First, Empirical Mode Decomposition (EMD) is used to decompose the signal into n Intrinsic Mode Functions (IMFs), satisfying the following IMF selection criteria: The number of local extrema of the signal over the entire time series is equal to or differs from the number of zero crossings by no more than 1, and the average value of the upper and lower envelopes of the signal at any time is 0. The Hilbert transform of each IMF component is defined as follows: (3)； in, For signal The Hilbert transform result, where PV represents the Cauchy principal value, is used to handle the integrand in... The singularity at the point guarantees the convergence of the integral; Sensitive feature parameters, including peak value P and root mean square value, are extracted based on the Hilbert transform results. and clock speed Root mean square value Represented as: (4)； Where N is the number of signal sampling points, Let the signal amplitude be the value at the k-th sampling point; Vibration data: Wavelet packet decomposition technology was used to decompose the vibration signal at multiple scales, and effective frequency band components were selected. Peak values ​​were extracted by combining time-domain analysis. Kurtosis K and skewness S are used to extract the power spectral density through frequency domain analysis. : (5)； (6)； (7)； Where μ is the mean of the vibration signal, and σ is the standard deviation of the signal. This is the Fourier transform result for a signal duration of T, where f is the frequency.

5. The method for bridge expansion joint status identification and early warning based on Bayesian data fusion according to claim 2, characterized in that, Step S4 is as follows: Taking the failure warning of the expansion joint as the top event, based on the disaster-causing factor system and failure cause library constructed in step S1, and combined with the structural composition and damage evolution mechanism of the expansion joint, a fault tree model is established, specifically as follows: Fault tree top event: Bridge expansion joint failure; Intermediate events in the fault tree: bolt damage, comb plate damage, anchoring structure failure; Fault tree base events: environmental disaster-causing factors, load-causing disaster-causing factors, and structural disaster-causing factors; The fault tree model is transformed into a Bayesian network model, clarifying the top, middle, and bottom events of the network nodes and the causal relationships and logical connections between nodes. In the Bayesian network, the nodes correspond to the events in the fault tree, and the directed edges between the nodes correspond to the causal relationships between the events.

6. The method for bridge expansion joint status identification and early warning based on Bayesian data fusion according to claim 1, characterized in that, Step S5 specifically includes: Multi-dimensional feature standardization: For the three-dimensional sensitive features extracted in step S3, Z-score standardization is used to eliminate dimensional differences. The standardization formula is: (8)； in, σ is the dimensionless value of the feature after standardization; x is the original value of the feature; μ is the mean of the feature in the full dataset; σ is the standard deviation of the feature in the full dataset. The three-dimensional sensitive features are as follows: Visual feature: Bolt center point offset d; Vibrational characteristics: kurtosis K and peak power spectral density ; Acoustic characteristics: root mean square and clock speed ; Feature layer dimensionality reduction and redundancy removal: Principal component analysis (PCA) is used to reduce the dimensionality of the standardized multi-dimensional feature vectors while retaining core information; For single-dimensional damage recognition probability calculation, three Support Vector Machine (SVM) recognition models are trained based on the reduced-dimensional features of each dimension, outputting independent damage recognition probabilities for each dimension. The decision function of the SVM is defined as follows: (9)； in, For the Lagrange multipliers corresponding to the support vectors, For sample labels; The radial basis function kernel is calculated using the following formula: ,in γ represents the support vector samples in the SVM training set, x represents the multi-source data feature samples of the bridge expansion joint to be identified, and γ represents the kernel function bandwidth parameter. For classification threshold, It is a symbolic function; Bayesian probabilistic fusion is used to calculate prior probabilities. Using single-dimensional damage recognition probabilities as observational evidence, Bayesian probabilistic fusion rules are employed to integrate multi-source information, yielding the final prior probabilities of intermediate events in the Bayesian network. The fusion formula is: (10)； in, Let A be the initial prior probability of intermediate event A; Let A be the initial prior probability that intermediate event A does not occur. Let m be the probability that the model misclassifies the intermediate event A as damage if the intermediate event A does not occur.

7. The method for bridge expansion joint status identification and early warning based on Bayesian data fusion according to claim 1, characterized in that, Step S6 is as follows: A parameter learning method combining Maximum A posteriori estimation (MAP) and Analytic Hierarchy Process (AHP) is used to construct the conditional probability table (CPT) for each node of the Bayesian network, specifically as follows: Quantification of expert experience: By inviting experts in bridge engineering and structural health monitoring, the Analytic Hierarchy Process (AHP) was used to score the correlation strength of each node under different combinations of parent node states, and the initial conditional probability was determined. Statistical learning optimization: Based on the feature dataset and prior probabilities extracted in step S3, the initial conditional probabilities are iteratively optimized using maximum a posteriori estimation (MAP). The bias of expert experience is corrected by combining data statistical laws, resulting in a conditional probability table (CPT) that balances theoretical support and engineering practice, quantifying the conditional probability distribution of each node under different combinations of parent node states.

8. The method for bridge expansion joint status identification and early warning based on Bayesian data fusion according to claim 5, characterized in that, Step S7 includes: Forward reasoning enables risk level classification and early warning, specifically as follows: Inputting prior probability data from a Bayesian network calculated using real-time multi-dimensional monitoring data, the probability of the top event is calculated through forward inference using the Bayesian network. Combining this with the geographical spatial unit where the bridge is located, risk level thresholds are set to complete the risk level classification and early warning for each geographical spatial unit. The specific risk level thresholds are: Low risk: Failure probability <10%; Medium risk: 10% ≤ probability of failure < 30%; High risk: 30% ≤ failure probability < 60%; Extremely high risk: failure probability ≥ 60%; The main control factor is selected through reverse reasoning, specifically as follows: For high-risk and extremely high-risk geospatial units, the contribution of each bottom event to the top event is calculated through Bayesian network inverse reasoning. The main disaster-causing factors are then sorted by contribution and selected to provide a basis for targeted maintenance decisions.

9. A bridge expansion joint status identification and early warning system based on Bayesian data fusion, characterized in that, The system adopts the bridge expansion joint status identification and early warning method based on Bayesian data fusion as described in any one of claims 1-8. The system includes: a disaster-causing factor system and fault cause library construction module, a multi-dimensional data acquisition module, a data preprocessing and feature extraction module, a Bayesian network construction module, a conditional probability table construction module, and a risk early warning module. The module for constructing the disaster-causing factor system and failure cause database is used to build the disaster-causing factor system and failure cause database for bridge expansion joints. A multi-dimensional data acquisition module is used to collect multi-dimensional data of bridge expansion joints; The data preprocessing and feature extraction module is used to preprocess and extract features from the collected multi-dimensional data; The Bayesian network building block is used to build Bayesian network models based on fault trees. The conditional probability table construction module is used to perform multi-dimensional data fusion and determine the prior probability of the Bayesian network model; it constructs the conditional probability table of each node of the Bayesian network based on maximum a posteriori estimation and the analytic hierarchy process. The risk warning module quantifies the failure probability of expansion joints through forward inference using Bayesian networks, classifies risk levels and provides risk warnings; it also uses backward inference using Bayesian networks to screen the main disaster-causing factors, providing a basis for decision-making for targeted maintenance of bridge expansion joints.

10. A computer-readable storage medium storing a program, characterized in that, When the program is executed by the processor, it implements the bridge expansion joint status identification and early warning method based on Bayesian data fusion as described in any one of claims 1-8.

Images