A method and system for diagnosing bridge condition anomalies based on support vector machine and BWIM data

Abstract

This invention proposes a bridge anomaly diagnosis method and system based on support vector machine (SVM) and BWIM data. It selects common six-axle vehicles as indicators for bridge health monitoring, conducts load calibration and inspection tests on the bridge to obtain bridge influence surface and vehicle wheelbase information, and uses a multi-vehicle dynamic weighing algorithm for vehicle weight identification. A kernel distribution function is used to fit the weighing results of the six-axle vehicles, and feature indicators are proposed based on the fitted probability distribution curve. SVM is used for training to obtain the range of feature indicators under the initial bridge state, and outlier detection is used to determine the bridge's anomaly state. This invention does not require prior measurement of vehicle information and does not require traffic closure, thus avoiding excessive impact on busy traffic systems and significant economic losses from prolonged traffic disruptions.

Description

Technical Field

[0001] This invention belongs to the field of bridge structural health and safety monitoring technology, specifically, it relates to a bridge condition anomaly diagnosis method and system based on support vector machine and BWIM data. Background Technology

[0002] Currently, the main methods for bridge inspection in my country include manual visual inspection, image recognition technology, and load testing. Among these, dynamic and static load tests using vehicles are the most intuitive and effective methods for assessing the structural condition of bridges. Load testing methods have seen significant practical application both domestically and internationally, resulting in numerous standards and procedures. In engineering practice, static load testing can assess the load-bearing capacity of bridges, reflecting their stiffness and strength. However, it suffers from drawbacks such as high time and economic costs, and the need for prolonged traffic closures, making it difficult to widely apply to the routine inspection of numerous small and medium-sized bridges. In contrast, dynamic load testing offers greater efficiency and speed. Research on bridge condition assessment techniques and methods based on dynamic load testing has long been a research hotspot in the fields of bridge safety assessment and damage diagnosis.

[0003] Bridge-based vehicle dynamic weighing (BWIM) has yielded substantial research results and made some progress in practical applications. BWIM technology involves deploying a number of sensors on the bridge to monitor passing vehicles in real time, measuring and analyzing parameters such as wheel weight and position to achieve dynamic weighing and overload detection. BWIM technology enables online vehicle weighing and detection, avoiding traffic congestion and unnecessary vehicle wear caused by crowded roadside weigh stations, while simultaneously improving the safety and efficiency of highway traffic.

[0004] However, research on bridge condition assessment using BWIM data as an indicator has yet to consider the characteristics of highway traffic loads in China. Furthermore, current research lacks efficient methods for assessing bridge structural condition that fully consider the characteristics and patterns of large amounts of vehicle weight data. Traditional bridge structural condition assessment methods are mainly based on statistical analysis and empirical judgment, limiting the accuracy and efficiency of the algorithms. With the development of artificial intelligence and big data technologies, bridge structural condition assessment technologies based on machine learning and deep learning are rapidly emerging. These methods can automatically extract features and patterns, transforming large amounts of bridge monitoring data into quantitative structural condition assessment results, thereby improving the accuracy and efficiency of the assessment. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a bridge condition anomaly diagnosis method and system based on support vector machine and BWIM data. The method selects the six-axle trailer, the most common type of heavy-duty vehicle on highways, as the indicator of bridge condition based on traffic survey data. It calculates the total vehicle weight using a dynamic weighing algorithm, fits the vehicle weight probability distribution using a kernel function, and uses the two peaks of the probability distribution curve as feature indicators. Finally, it introduces a support vector machine (SVM) method to detect outliers, indicating abnormal bridge conditions.

[0006] This invention is achieved through the following technical solution:

[0007] A bridge condition anomaly diagnosis method based on support vector machine and BWIM data:

[0008] The method specifically includes the following steps:

[0009] Step 1: Develop a test plan for bridge dynamic load calibration.

[0010] In the initial state of the bridge, a fixed displacement and strain measurement point is selected to identify the influence surface of the bridge.

[0011] Step 2: Develop a bridge inspection and testing plan;

[0012] Based on the fixed displacement and strain measurement points determined in step 1, and in conjunction with computer vision or spatial integrated monitoring methods, multiple detection experiments are conducted on the bridge in its initial state to obtain information on the bridge's dynamic response, vehicle wheelbase, and lateral position on the bridge.

[0013] Step 3: Perform dynamic weighing and weight distribution fitting for the six-axle vehicle;

[0014] Based on the bridge influence surface and bridge dynamic response data obtained in steps 1 and 2, multi-vehicle dynamic weighing is performed to obtain the total weight and distribution of six-axle vehicles. The distribution curve of the total vehicle weight is fitted using kernel density to obtain and extract the corresponding detection feature indicators.

[0015] Step 4: Perform support vector machine training on a single-class distribution;

[0016] The multiple sets of feature indicators obtained in step 3 are trained using a support vector machine to form the corresponding feature indicator region in the initial state of the bridge.

[0017] Step 5: Perform dynamic weighing on the bridge under inspection.

[0018] The bridge under inspection is subjected to the inspection experiment described in step 2. The dynamic weighing results of multiple vehicles are subjected to kernel density fitting to extract feature indicators, which are then compared with the feature indicator regions obtained in step 4 to determine the bridge condition.

[0019] Further, in step 1, the dynamic load calibration test includes:

[0020] The initial testing must be conducted under healthy bridge conditions. The strain measurement points include important components or nodes at the piers, supports, mid-span, and side spans.

[0021] Step 1.1, conduct experimental data acquisition: Under the initial state, collect displacement and strain data, and obtain the displacement and strain data of the measuring point under different loads through experimental measurement;

[0022] Step 1.2, Data Processing: Process and analyze the collected displacement and strain data, and plot the corresponding load response curves;

[0023] Step 1.3, Influence Surface Identification: By comparing the changes in the load response curve, determine the change in the response of the measuring point to the load, so as to further identify whether it is an influence surface or a non-influence surface;

[0024] Step 1.4, Influence Surface Determination: Based on the influence surface information obtained from all measuring points, and combined with the bridge's design parameters and structural characteristics, determine the distribution of the bridge's influence surface.

[0025] Furthermore, in step 2,

[0026] Based on the fixed displacement and strain measurement points determined in step 1, sensors are arranged at the same location to record the bridge's dynamic response. The wheelbase and lateral position information of the six-axle vehicle are obtained by combining computer vision and spatial integrated monitoring methods.

[0027] Furthermore, in step 3,

[0028] After obtaining the influence surface of the bridge, multiple bridge inspection experiments were conducted under initial conditions. A six-axle trailer was selected as the main research object for multi-vehicle dynamic weighing.

[0029] The constraint inequality in the dynamic weighing calculation for a vehicle crossing a bridge is:

[0030]

[0031] In the formula, f is the objective function; g is the gradient of the objective function; x is the target object, an n-order vector; H is an n×n symmetric matrix, whose non-zero terms are h. ij A is an m×n matrix with nonzero terms a. ij ; l and u are the lower and upper bounds of the solution x, respectively.

[0032] Furthermore, after calculating the total weight of the six-axle vehicle using a multi-vehicle dynamic weighing algorithm, kernel density estimation is used to fit the weight distribution of the six-axle vehicle:

[0033] For a series of independent observation data samples x1, x2, ..., xn in a random variable X n Assuming the existence of a cumulative distribution function F(x) and a probability density function f(x), whose functional expressions can be estimated using sample data;

[0034]

[0035] In the formula, n represents the number of observed samples; h represents the bandwidth, and the value is determined as shown in formula (4); K(·) represents the kernel smoothing function;

[0036] Based on the probability density function, the cumulative distribution function can be calculated using indefinite integrals:

[0037]

[0038] In the formula,

[0039] The optimal bandwidth is obtained by finding the minimum mean integrated square error (MISE), and its expression is as follows:

[0040]

[0041] Expanding the above equation using Taylor series yields:

[0042]

[0043] Ignoring infinitesimals, the derivative yields the optimal bandwidth as follows:

[0044]

[0045] The abscissas of the two peak points of the fitted probability density curve are selected as feature indices, and the calculation method is as follows:

[0046]

[0047] In the formula, (x f ,y f — Feature index; x1 is the x-coordinate of the first peak point; x2 is the x-coordinate of the second peak point.

[0048] Furthermore, in step 4,

[0049] The distribution of the weight of the six-axle vehicle over multiple time periods under the initial condition of the bridge and the corresponding x-coordinates of two feature points in each group can be fitted. By using a support vector machine to perform single-class classification training on several groups of feature indicators, the range of indicators for the initial state of the bridge can be obtained. The specific method is as follows:

[0050] The v-SVM method, which selects support vector machines for one-class classification, is based on the following principle:

[0051] For a series of training sample datasets {(x1,y1),(x2,y2),...,(x l ,y l )}∈χ, where l is the number of training samples and χ is the space R N A subset of χ; let Φ be the feature map from χ to F, i.e., the map of the inner product space F, so that the inner product in the image of χ can be calculated using a kernel function:

[0052] k(x,y)=(Φ(x)·Φ(y)) (12)

[0053] In the formula, k(x,y) represents the kernel function, and Φ(·) is the eigenmap of χ→F;

[0054] To separate the dataset from the origin, the following quadratic problem needs to be solved:

[0055]

[0056] In the formula, w is a variable in the inner product space, and ξ i represents a non-zero slack variable; v represents a parameter that is cleared in subsequent calculations, v∈(0,1]; ρ is the offset;

[0057] Due to ξ i The objective function is a penalty function, and this problem can be solved using w and ρ. The decision function is set as follows:

[0058] f(x)=sgn((w·Φ(x))-ρ) (14)

[0059] In the formula, sgn(·) is the step function;

[0060] Based on variable α i ,β i ≥0, construct the Lagrange function:

[0061]

[0062] To make the derivatives of the original variables w, ξ, ρ zero, let:

[0063] w = ∑ i α i Φ(x i (16)

[0064]

[0065] In the formula, x i Let x represent support vectors that satisfy {x} i :i∈[l],αi >0}.

[0066] Equation (13) is converted into a kernel function expression:

[0067] f(x)=sgn(∑ i α i k(x i ,x)-ρ) (18)

[0068] Using equation (12), we obtain the dual problem:

[0069]

[0070] If, after optimization, the optimal solution is achieved, then α i and β i Not zero, i.e., 0 < α i If <1 / vl, then the two inequality conditions in equation (6) will be converted into equations.

[0071] Furthermore, in step 5,

[0072] Step 5.1, conduct bridge dynamic response monitoring: select a bridge under test for a period of time, and conduct bridge dynamic response monitoring according to the method in Step 2 to obtain the vibration characteristic parameters of the bridge.

[0073] Step 5.2, Perform multi-vehicle dynamic weighing: Use the multi-vehicle dynamic weighing method to calculate the weight of the vehicle by utilizing the vibration and deformation of the bridge as the vehicle travels on the bridge.

[0074] Step 5.3, extract feature indicators: For the weighing data obtained from multi-vehicle dynamic weighing, perform kernel density fitting to extract feature indicators;

[0075] Step 5.4, compare the feature index area: compare the feature index area obtained in step 4 with the feature index under the test state to determine whether the bridge has been damaged or deformed. If the feature index under the test state exceeds the feature index range under the initial state, it indicates that the bridge has been damaged or changed.

[0076] Step 5.5, Determine the bridge status: Based on the comparison results, determine the bridge status and take appropriate action; if it is determined to be normal, continue to monitor its status during subsequent use; if a problem is confirmed, it needs to be repaired or replaced.

[0077] A bridge condition anomaly diagnosis system based on support vector machine and BWIM data:

[0078] The system includes a calibration test module, a bridge inspection test module, a data processing module, and a bridge inspection module;

[0079] The calibration test module is used to develop a calibration test plan for bridge dynamic loads;

[0080] In the initial state of the bridge, a fixed displacement and strain measurement point is selected to identify the influence surface of the bridge.

[0081] The bridge inspection and testing module is used to develop bridge inspection and testing plans;

[0082] Based on the fixed displacement and strain measurement points determined by the calibration test module, and in conjunction with computer vision or spatial integrated monitoring methods, multiple test experiments are conducted on the bridge in its initial state to obtain information on the bridge's dynamic response, vehicle wheelbase, and lateral position on the bridge.

[0083] The data processing module is used to perform dynamic weighing and weight distribution fitting of six-axle vehicles.

[0084] Based on the bridge influence surface and bridge dynamic response data obtained from the calibration test module and the bridge inspection test module, multi-vehicle dynamic weighing is performed to obtain the total weight and distribution of the six-axle vehicle. The distribution curve of the total vehicle weight is fitted using kernel density to obtain and extract the corresponding detection characteristic indicators.

[0085] The data processing module is also used to perform support vector machine training for a single distribution.

[0086] The multiple sets of feature indicators obtained by the data processing module are trained using a support vector machine to form the corresponding feature indicator region in the initial state of the bridge.

[0087] The bridge inspection module is used to dynamically weigh the bridge under inspection.

[0088] The bridge inspection test module describes the inspection experiment performed on the bridge under inspection. The dynamic weighing results of multiple vehicles are subjected to kernel density fitting to extract feature indicators, which are then compared with the feature indicator regions obtained by the data processing module to determine the bridge condition.

[0089] An electronic device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the above method.

[0090] A computer-readable storage medium for storing computer instructions that, when executed by a processor, implement the steps of the above-described method.

[0091] Beneficial effects of the invention

[0092] This invention uses dynamic weighing of a six-axle truck to extract two peak parameters with obvious characteristics from the kernel density fitting curve for algorithm training, forming a feature index region for bridge condition judgment to diagnose whether the bridge condition is abnormal.

[0093] This invention eliminates the need for prior vehicle measurements and traffic closures, meaning it minimizes disruption to busy traffic systems and avoids significant economic losses from prolonged traffic disruptions, particularly for urban and highway bridge condition monitoring. Furthermore, the practical bridge test requires only a single displacement measurement point, greatly reducing the manpower and time costs associated with bridge inspection.

[0094] More importantly, for the same bridge, only the initial bridge condition anomaly diagnosis requires bridge calibration tests and dynamic vehicle weighing over multiple time periods. The feature index region is constructed using support vector machine single-class classification training. Subsequent bridge detection tests can then focus solely on outlier detection. Furthermore, the non-outlier features extracted from subsequent bridge condition diagnoses can be used as criteria for updating the feature index region in the next diagnosis. This translates to a more reliable diagnostic indicator region and lower time costs. Attached Figure Description

[0095] Figure 1 This is a flowchart of the method of the present invention;

[0096] Figure 2 This is a schematic diagram of the kernel density fitting probability curve for a six-axle vehicle.

[0097] Figure 3 This is a cross-sectional view of the mid-span of a 30m simply supported T-beam;

[0098] Figure 4 This is a finite element model diagram of a 30m simply supported T-beam;

[0099] Figure 5 This is a probability model diagram of random traffic flow vehicle types;

[0100] Figure 6 This is a simulation result of the random distribution of the weight of a six-axle vehicle;

[0101] Figure 7 This is a simulation result diagram of the random distribution of vehicle spacing;

[0102] Figure 8 This is a diagram showing the SVM classification results for one class.

[0103] Figure 9 This is a Class I classification result diagram of SVM based on the vehicle weight index of 300 six-axle vehicles;

[0104] Figure 10 The SVM results considering noise and vehicle position error are shown in the figure. Detailed Implementation

[0105] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0106] Combination Figures 1 to 10 .

[0107] A bridge condition anomaly diagnosis method based on support vector machine and BWIM data, such as Figure 1 As shown, the method includes the following steps:

[0108] Step 1: Develop a test plan for bridge dynamic load calibration.

[0109] Bridge dynamic load calibration tests are designed to obtain the bridge's influence surface and must be conducted under the bridge's initial conditions. Based on the bridge type and critical section, a fixed displacement and strain measurement point needs to be determined, and bridge load tests are performed to obtain the bridge's influence surface at that point. Displacement and strain data are collected under the initial conditions, and experimental measurements are used to obtain the displacement and strain response data of this measurement point under different loads. Bridge load tests are then conducted to obtain the bridge's influence surface at that point. However, when the bridge conditions change, because the calculated influence surface remains unchanged, the vehicle weight calculated by dynamic weighing will deviate from the true value.

[0110] When selecting bridge locations, it is necessary to choose representative locations as strain gauges based on the bridge's structure and design parameters to collect experimental data. These locations include important components or nodes such as piers, supports, mid-span, and side spans. Then, appropriate measuring instruments, such as displacement gauges, strain gauges, and accelerometers, should be selected. During the experiment, the data acquisition frequency and duration, load type and magnitude, and changes in displacement, strain, acceleration, and other indicators at each monitoring point should be recorded, along with any potential interference factors during the experiment.

[0111] A fixed displacement and strain measurement point is selected in the initial state of the bridge to identify the bridge's influence surface.

[0112] Step 1.1, conduct experimental data acquisition: Under the initial state, collect displacement and strain data, and obtain the displacement and strain data of the measuring point under different loads through experimental measurement.

[0113] Step 1.2, Data Processing: Process and analyze the collected displacement and strain data, such as smoothing and fitting, and plot the corresponding load-response curves.

[0114] Step 1.3, Influence Surface Identification: By comparing the changes in the load-response curves, determine the change in the response of the measuring point to the load, in order to further identify whether it is an influence surface or a non-influence surface.

[0115] Step 1.4, Influence Surface Determination: Based on the influence surface information obtained from all measuring points and combined with the bridge's design parameters and structural characteristics, determine the distribution of the bridge's influence surface to provide a scientific basis for subsequent load identification and bearing capacity assessment.

[0116] Step 2: Develop a bridge inspection and testing plan;

[0117] Based on the fixed displacement and strain measurement points determined in step 1, sensors are arranged at the same location. Combined with computer vision or spatial integrated monitoring methods, multiple test experiments are conducted on the bridge in its initial state to obtain information on the bridge's dynamic response, vehicle wheelbase, and lateral position on the bridge, which can then be used to calculate the bridge's service status.

[0118] Step 3: Perform dynamic weighing and weight distribution fitting for the six-axle vehicle;

[0119] A six-axle vehicle is a vehicle with six axles (two in the front and two in the rear, and one in the left and one in the right). It is used for dynamic weighing of important projects such as bridges to detect data such as vehicle weight and wheelbase when the vehicle passes over the bridge.

[0120] Based on the bridge influence surface and bridge dynamic response data obtained in steps 1 and 2, multi-vehicle dynamic weighing is performed to obtain the total weight and distribution of six-axle vehicles. The distribution curve of the total vehicle weight is fitted using kernel density to obtain and extract the corresponding detection feature indicators.

[0121] Kernel density fitting of the vehicle gross weight distribution curve is a fitting method based on probability density functions. By superimposing the probability density functions of multiple sets of vehicle weight data, a continuous probability density curve is obtained, and then relevant statistical analysis is performed.

[0122] After obtaining the influence surface of the bridge, multiple bridge inspection experiments were conducted under initial conditions. A six-axle trailer was selected as the main research object for multi-vehicle dynamic weighing. The principle of the multi-vehicle dynamic weighing method is as follows:

[0123] The constraint inequality in the dynamic weighing calculation for a vehicle crossing a bridge is:

[0124]

[0125] In the formula, f is the objective function; g is the gradient of the objective function; x is the solution objective, an n-order vector; b is the constraint condition; H is an n×n symmetric matrix, whose non-zero terms are h. ij A is an m×n matrix with nonzero terms a. ij; l and u are the lower and upper bounds of the solution x, respectively.

[0126] The preliminary vehicle axle load was obtained by iterative calculation using the linearly constrained least squares method.

[0127] Update the constraint boundaries, gradients, and constant terms of the objective function when variable x j When it appears within the i-th constraint:

[0128] b i + =b i -a ij x j (2)

[0129] The constant term of the objective function is updated as follows:

[0130]

[0131] Gradient update is as follows:

[0132]

[0133] The prediction-correction algorithm is continuously iterated until the following condition is met:

[0134]

[0135] In the formula, It is an extended linear matrix containing inequalities and equality constraints; The vector corresponding to the extended linear constraint equation.

[0136] After calculating the total weight of a six-axle vehicle using a multi-vehicle dynamic weighing algorithm, the distribution of the total vehicle weight is fitted using kernel density. The principle of kernel density estimation for fitting the probability distribution can be expressed as follows.

[0137] For a series of independent observation data samples x1, x2, ..., xn in a random variable X n Assume that there exists a cumulative distribution function F(x) and a probability density function f(x), whose functional expressions can be estimated using sample data.

[0138]

[0139] In the formula, n represents the number of observed samples; h represents the bandwidth, and the value is determined as in formula (15); K(·) represents the kernel smoothing function. For the vehicle weight probability distribution fitting in this invention, the Gaussian kernel function is selected as the kernel smoothing function.

[0140] Based on the probability density function, the cumulative distribution function can be calculated using indefinite integrals:

[0141]

[0142] In the formula,

[0143] This invention obtains the optimal bandwidth by finding the minimum value of the Mean Integrated Square Error (MISE), the expression of which is:

[0144]

[0145] Expanding the above equation using Taylor series yields:

[0146]

[0147] Ignoring infinitesimals, the derivative yields the optimal bandwidth as follows:

[0148]

[0149] The kernel density function was used to fit the gross vehicle weight of a six-axle vehicle, and the results are as follows: Figure 2 As shown, the weight distribution of the six-axle vehicle exhibits a bimodal shape. Unlike previous studies that used the average vehicle weight as an indicator, this invention selects the abscissas of the two peak points of the fitted probability density curve as characteristic indicators to better reflect the actual weight distribution.

[0150]

[0151] In the formula, (x f ,y f — Feature index; x1 is the x-coordinate of the first peak point; x2 is the x-coordinate of the second peak point.

[0152] Step 4: Perform support vector machine training for one-class classification;

[0153] The multiple sets of feature indicators obtained in step 3 are trained by support vector machine for one-class classification to form the corresponding feature indicator region in the initial state of the bridge.

[0154] First, define the input features and the output model: use a support vector machine (SVM) as the classification model, where the model uses a kernel function to perform a non-linear mapping on the input features, mapping the original data to a high-dimensional space for classification.

[0155] Refit the dataset: Use the training dataset to fit and train the SVM model to maximize the margin of the classification boundary and make the samples as separate as possible.

[0156] Model evaluation: The trained SVM model is evaluated using a test dataset. Metrics such as precision and recall are calculated on the test dataset to assess the model's classification performance.

[0157] The feature index region is derived as follows: When the model performs well on the test dataset, it is used to predict the distribution of the weight of the six-axle vehicle over multiple time periods under the initial condition of the bridge. The prediction results can obtain the abscissa of each set of two feature points, and each feature point corresponds to a specific set of feature index values. The combination of these feature index values ​​forms the feature index region under the initial condition.

[0158] The distribution of the weight of the six-axle vehicle and the corresponding x-coordinates of two feature points in each group under the initial condition of the bridge are fitted. Then, a support vector machine is used to perform single-class classification training on several groups of feature indicators to obtain the index range of the bridge's initial state features. The specific method is as follows:

[0159] The v-SVM method, which selects support vector machines for one-class classification, is based on the following principle:

[0160] For a series of training sample datasets {(x1,y1),(x2,y2),...,(x l ,y l )}∈χ, where l is the number of training samples and χ is the space R N A subset of χ. Let Φ be the feature map from χ to F, i.e., the map of the inner product space F, so that the inner product in the image of χ can be calculated using a kernel function:

[0161] k(x,y)=(Φ(x)·Φ(y)) (12)

[0162] In the formula, k(x,y) represents the kernel function, and Φ(·) is the eigenmap of χ→F.

[0163] By freely using different types of kernel functions, this simple geometric problem can be mapped to various nonlinear estimation algorithms in the input space; to separate the dataset from the origin, the following quadratic problem needs to be solved:

[0164]

[0165] In the formula, w is a variable in the inner product space, and ξ i ρ represents a non-zero slack variable; v represents a parameter that is cleared in subsequent calculations, v∈(0,1]; ρ is the offset.

[0166] Due to ξ i The objective function is a penalty function, and this problem can be solved using w and ρ. The decision function is set as follows:

[0167] f(x)=sgn((w·Φ(x))-ρ) (14)

[0168] In the formula, sgn(·) is the step function.

[0169] Based on variable α i ,β i ≥0, construct the Lagrange function:

[0170]

[0171] To make the derivatives of the original variables w, ξ, ρ zero, let:

[0172] w = ∑ i α i Φ(x i (16)

[0173]

[0174] In the formula, x i Let x represent support vectors that satisfy {x} i :i∈[l],α i >0}.

[0175] Equation (13) is converted into a kernel function expression:

[0176] f(x)=sgn(∑ i α i k(x i ,x)-ρ) (18)

[0177] Using equation (12), we obtain the dual problem:

[0178]

[0179] If, after optimization, the optimal solution is achieved, then α i and β i Not zero, i.e., 0 < α i If <1 / vl, then the two inequality conditions in equation (13) will be converted into equations.

[0180] By using the aforementioned support vector machine training method, the classification results of the feature value index of a six-axle vehicle can be obtained.

[0181] Step 5: Perform dynamic weighing on the bridge under inspection.

[0182] After obtaining the characteristic index region of the bridge in its initial state, the bridge is inspected in the state to be inspected. The bridge dynamic response monitoring described in step 2 is performed on the bridge in the state to be inspected. Multi-vehicle dynamic weighing is performed on the bridge dynamic response data. The weighing results are kernel density fitted to extract characteristic indexes, which are compared with the characteristic index region obtained in step 4 to determine the bridge state.

[0183] Step 5.1, conduct bridge dynamic response monitoring: First, select a bridge under test for a period of time, and conduct bridge dynamic response monitoring according to the method in Step 2 to obtain the vibration characteristic parameters of the bridge and save them to the computer.

[0184] Step 5.2, Perform multi-vehicle dynamic weighing: Next, the multi-vehicle dynamic weighing method is used to calculate the vehicle's weight by utilizing the vibration and deformation of the bridge as the vehicle travels on it. The weighing data for each vehicle is saved.

[0185] Step 5.3, Extract Feature Indicators: For the weighing data obtained from multi-vehicle dynamic weighing, kernel density fitting is performed to extract feature indicators, such as vehicle weight and vehicle position. These feature indicators will be used for subsequent judgment.

[0186] Step 5.4, compare the feature index area: compare the feature index area obtained in step 4 with the feature index under the test state to determine whether the bridge has been damaged or deformed. If the feature index under the test state exceeds the feature index range under the initial state, it indicates that the bridge has been damaged or changed.

[0187] Step 5.5, Determine Bridge Status: Finally, based on the comparison results, determine the bridge's status and take appropriate action. If it is determined to be normal, continue monitoring its status during subsequent use. If a problem is confirmed, repair or replacement is required. Simultaneously, the monitoring and comparison process needs to be reviewed and adjusted to improve the accuracy and reliability of the detection.

[0188] A bridge condition anomaly diagnosis system based on support vector machine and BWIM data:

[0189] The system includes a calibration test module, a bridge inspection test module, a data processing module, and a bridge inspection module;

[0190] The calibration test module is used to develop a calibration test plan for bridge dynamic loads;

[0191] In the initial state of the bridge, a fixed displacement and strain measurement point is selected to identify the influence surface of the bridge.

[0192] The bridge inspection and testing module is used to develop bridge inspection and testing plans;

[0193] Based on the fixed displacement and strain measurement points determined by the calibration test module, and in conjunction with computer vision or spatial integrated monitoring methods, multiple test experiments are conducted on the bridge in its initial state to obtain information on the bridge's dynamic response, vehicle wheelbase, and lateral position on the bridge.

[0194] The data processing module is used to perform dynamic weighing and weight distribution fitting of six-axle vehicles.

[0195] Based on the bridge influence surface and bridge dynamic response data obtained from the calibration test module and the bridge inspection test module, multi-vehicle dynamic weighing is performed to obtain the total weight and distribution of the six-axle vehicle. The distribution curve of the total vehicle weight is fitted using kernel density to obtain and extract the corresponding detection characteristic indicators.

[0196] The data processing module is also used to perform support vector machine training for a single distribution.

[0197] The multiple sets of feature indicators obtained by the data processing module are trained using a support vector machine to form the corresponding feature indicator region in the initial state of the bridge.

[0198] The bridge inspection module is used to dynamically weigh the bridge under inspection.

[0199] The bridge inspection test module describes the inspection experiment performed on the bridge under inspection. The dynamic weighing results of multiple vehicles are subjected to kernel density fitting to extract feature indicators, which are then compared with the feature indicator regions obtained by the data processing module to determine the bridge condition.

[0200] The following is a diagnostic analysis of abnormal bridge conditions based on actual circumstances:

[0201] Taking a 30m simply supported T-beam bridge as the analysis object, using C50 concrete, the main beam consists of 6 T-beams connected laterally by cast-in-place wet joints. A finite element model is established using the beam grid method in Midas / civil. The bridge deck is 13.5m wide, with a deck layout of: 0.75m (crash barrier) + 12.25m (carriageway) + 0.5m (crash barrier). The mid-span cross-section of the bridge is shown below. Figure 3 As shown, a finite element model is established based on Midas / civil using the beam grid method. Figure 4 As shown, the bridge damping ratio is taken as 0.05, and the transverse connection is simulated by a virtual beam that only considers stiffness and not weight. It is assumed that only shear force is transmitted at the hinge joint, which is achieved by releasing the beam end moment constraint at the hinge joint position.

[0202] This invention uses random traffic flow instead of actual traffic flow for testing and dynamic weighing of bridges. For example... Figure 5The diagram shows traffic statistics for each lane. A probabilistic model of vehicle type for each lane is established based on a uniform distribution. In actual random traffic, the weight of large trailers often exhibits a multi-peak distribution. A log-normal distribution is used to simulate the weight of whole trucks, and a mixed normal distribution is used to simulate the weight of trailer-trailers. After selecting lanes, seven types of traffic flow are distributed according to vehicle type ratios. Vehicle weight is allocated to axle loads according to the axle load ratio of each vehicle type, generating random vehicle weights for both whole trucks and trailers based on their respective probability distributions. Then, the vehicle spacing for each lane is generated based on a log-normal distribution, grouping vehicles appearing on the bridge sequentially into a group. Finally, the random traffic flow of each lane is substituted into a vehicle-bridge coupled random vibration program for numerical simulation.

[0203] The simulation results of the random distribution of vehicle weight and the random distribution of vehicle spacing for a six-axle vehicle are as follows: Figure 6 and Figure 7 As shown.

[0204] The dynamic displacement response at mid-span of beam #3, obtained through simulation analysis, was used to simulate the measured response. The displacement influence surface at mid-span of beam #3 was then used for dynamic weighing identification of vehicles crossing the bridge. Using a developed vehicle-bridge coupled vibration analysis and random traffic flow simulation program, the dynamic response data of the bridge when random traffic flow passes over the bridge was simulated, and the displacement response at mid-span of beam #3 was also output. Kernel density estimation was performed on the vehicle weight identification results of all six-axle vehicles, and the probability distribution curve fitting results were obtained using 300 vehicles as an example.

[0205] The weighing results of every 300 six-axle vehicles are output, and a set of characteristic indicators (x) are obtained by fitting a probability density curve. f ,y f Five sets of indicators, representing 1500 six-axle vehicles in the bridge's intact state, were taken as training samples for the support vector machine. A single-class classification learning process was then performed to form the feature indicator regions for the bridge's initial state. The classification learning results are as follows: Figure 8 In the figure, the contour lines represent the classification scores of the support vector machine, and the contour lines with a score of 0 are the classification boundaries.

[0206] By changing the elastic modulus E of the bridge finite element model to simulate the overall stiffness reduction of the bridge, a random traffic flow containing 300 six-axle trailers was regenerated. The traffic flow was then substituted into the vehicle-bridge vibration analysis program for calculation, and the mid-span displacement response of beam #3 was output. The weight of the six-axle trailers was identified based on the dynamic weighing of the healthy bridge influence surface.

[0207] The above method was used to obtain characteristic indicators under bridge damage conditions. Figure 9The figure shows the location of each set of indicators and the indicator area of ​​the bridge in its initial state. As can be seen from the figure, the characteristic indicators based on the weight of the six-axle vehicle are highly sensitive to changes in the bridge's state. Using a single-class classification method with support vector machines, the abnormal state of the bridge when its overall stiffness is reduced by 3% and 5% can be accurately determined.

[0208] To verify the effectiveness of this method under the conditions of measurement errors and noise in actual engineering, 10 dB of Gaussian white noise was added to the numerical simulation results in the aforementioned vehicle-axle vibration analysis program. Simultaneously, using a uniformly distributed simulation method, the longitudinal and transverse errors of the vehicle position were incorporated into the vehicle weight calculation for each vehicle. The results are as follows: Figure 10 As shown.

[0209] It is evident that, even after considering test noise and vehicle position recognition errors, the bridge condition assessment method based on dynamic weighing vehicle weight probability density proposed in this paper still possesses excellent accuracy and applicability.

[0210] Analysis shows that the large sample size of six-axle vehicle weights and the machine learning method can effectively overcome random errors caused by noise and inaccurate vehicle positioning. Therefore, the bridge condition assessment method based on vehicle weight probability density has good robustness. Furthermore, this study selected the weights of 1500 six-axle vehicles as training samples, a number far smaller than the average daily number of six-axle vehicles on actual bridges. Therefore, it will not consume excessive time in bridge displacement measurement, and the determination of the characteristic index regions of the initial bridge state can be completed within one day. The subsequent weighing of 300 six-axle vehicles in each bridge inspection requires even less time; during periods of high six-axle vehicle traffic, the actual data collection for bridge condition assessment can be completed in only 1-2 hours.

[0211] An electronic device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the above method.

[0212] A computer-readable storage medium for storing computer instructions that, when executed by a processor, implement the steps of the above-described method.

[0213] The above provides a detailed description of the bridge state anomaly diagnosis method and system based on support vector machine and BWIM data proposed in this invention, and elucidates the principles and implementation methods of this invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A bridge condition anomaly diagnosis method based on support vector machine and BWIM data, characterized in that: The method specifically includes the following steps: Step 1: Develop a test plan for bridge dynamic load calibration. In the initial state of the bridge, a fixed displacement and strain measurement point is selected to identify the influence surface of the bridge. Step 2: Develop a bridge inspection and testing plan; Based on the fixed displacement and strain measurement points determined in step 1, and in conjunction with computer vision or spatial integrated monitoring methods, multiple detection experiments are conducted on the bridge in its initial state to obtain information on the bridge's dynamic response, vehicle wheelbase, and lateral position on the bridge. Step 3: Perform dynamic weighing and weight distribution fitting for the six-axle vehicle; Based on the bridge influence surface and bridge dynamic response data obtained in steps 1 and 2, multi-vehicle dynamic weighing is performed to obtain the total weight and distribution of six-axle vehicles. The distribution curve of the total vehicle weight is fitted using kernel density to obtain and extract the corresponding detection feature indicators. After obtaining the influence surface of the bridge, multiple bridge inspection experiments were conducted under initial conditions. A six-axle trailer was selected as the main research object for multi-vehicle dynamic weighing. The constraint inequality in the dynamic weighing calculation for a vehicle crossing a bridge is: where f is the objective function; g is the gradient of the objective function; x is the solution to the objective, n is the vector of order; H is n is n is the symmetric matrix of order whose nonzero entries are h ij ; A is m is n is the matrix of order whose nonzero entries are a ij ; l 1, u are the lower and upper bounds of the solution x , respectively. Step 4: Perform support vector machine training on a single-class distribution; The multiple sets of feature indicators obtained in step 3 are trained using a support vector machine to form the corresponding feature indicator region in the initial state of the bridge. Step 5: Perform dynamic weighing on the bridge under inspection. The bridge under inspection is subjected to the inspection experiment described in step 2. The dynamic weighing results of multiple vehicles are subjected to kernel density fitting to extract feature indicators, which are then compared with the feature indicator regions obtained in step 4 to determine the bridge condition.

2. The method according to claim 1, characterized in that: In step 1, the dynamic load calibration test includes: The initial testing must be conducted under healthy bridge conditions. The strain measurement points include important components or nodes at the piers, supports, mid-span, and side spans. Step 1.1, conduct experimental data acquisition: Under the initial state, collect displacement and strain data, and obtain the displacement and strain data of the measuring point under different loads through experimental measurement; Step 1.2, Data Processing: Process and analyze the collected displacement and strain data, and plot the corresponding load response curves; Step 1.3, Influence Surface Identification: By comparing the changes in the load response curve, determine the change in the response of the measuring point to the load, so as to further identify whether it is an influence surface or a non-influence surface; Step 1.4, Influence Surface Determination: Based on the influence surface information obtained from all measuring points, and combined with the bridge's design parameters and structural characteristics, determine the distribution of the bridge's influence surface.

3. The method according to claim 2, characterized in that: In step 2, Based on the fixed displacement and strain measurement points determined in step 1, sensors are arranged at the same location to record the bridge's dynamic response. The wheelbase and lateral position information of the six-axle vehicle are obtained by combining computer vision and spatial integrated monitoring methods.

4. The method according to claim 3, characterized in that: After calculating the total weight of the six-axle vehicle using a multi-vehicle dynamic weighing algorithm, kernel density estimation is used to fit the weight distribution of the six-axle vehicle. For a series of independent observation data samples in random variable X x 1, x 2, … , x n Assuming the existence of a cumulative distribution function F ( x ) and probability density function f ( x Its functional expression can be estimated using sample data; In the formula, n Indicates the number of observed samples; h The bandwidth is represented by the method of taking the value as shown in equation (4); K (·) denotes the kernel smoothing function; Based on the probability density function, the cumulative distribution function can be calculated using indefinite integrals: In the formula, ; The optimal bandwidth is obtained by finding the minimum mean integrated square error (MISE), and its expression is as follows: Expanding the above equation using Taylor series yields: Ignoring infinitesimals, the derivative yields the optimal bandwidth as follows: The abscissas of the two peak points of the fitted probability density curve are selected as feature indices, and the calculation method is as follows: In the formula, —Characteristic indicators; x 1 represents the x-coordinate of the first peak point; x 2 represents the x-coordinate of the second peak point.

5. The method according to claim 4, characterized in that: In step 4, The distribution of the weight of the six-axle vehicle over multiple time periods under the initial condition of the bridge and the corresponding x-coordinates of two feature points in each group can be fitted. By using a support vector machine to perform single-class classification training on several groups of feature indicators, the range of indicators for the initial state of the bridge can be obtained. The specific method is as follows: Select Support Vector Machine for Class 1 Class Classification Training v The SVM method works as follows: For a series of training sample datasets ,in l The number of training samples. For space R N A subset of; let for The feature mapping, i.e., the mapping of the inner product space F, makes... The inner product in the image can be calculated using a kernel function: In the formula, Represents the kernel function. for Feature mapping; To separate the dataset from the origin, the following quadratic problem needs to be solved: In the formula, w Let the inner product space be a variable. Represents a non-zero slack variable; v This indicates a parameter that is cleared in subsequent calculations. ; This is the offset; because The penalty function is defined in the objective function and can be obtained through... w and To solve this problem, the decision function is set as follows: In the formula, sgn(·) is the step function; Based on variables Construct the Lagrange function: Make the original variable Let the derivative be zero. In the formula, Represents support vectors that satisfy... ; Equation (13) is converted into a kernel function expression: Using equation (12), we obtain the dual problem: If the optimal solution is achieved after optimization, then... α i and β i Not zero, i.e., 0 < α i <1 / vl Then the two inequality conditions in equation (6) will be converted into equations.

6. The method according to claim 5, characterized in that: In step 5, Step 5.1, conduct bridge dynamic response monitoring: select a bridge under test for a period of time, and conduct bridge dynamic response monitoring according to the method in Step 2 to obtain the vibration characteristic parameters of the bridge. Step 5.2, Perform multi-vehicle dynamic weighing: Use the multi-vehicle dynamic weighing method to calculate the weight of the vehicle by utilizing the vibration and deformation of the bridge as the vehicle travels on the bridge. Step 5.3, extract feature indicators: For the weighing data obtained from multi-vehicle dynamic weighing, perform kernel density fitting to extract feature indicators; Step 5.4, compare the feature index area: compare the feature index area obtained in step 4 with the feature index under the test state to determine whether the bridge has been damaged or deformed. If the feature index under the test state exceeds the feature index range under the initial state, it indicates that the bridge has been damaged or changed. Step 5.5, Determine the bridge status: Based on the comparison results, determine the bridge status and take appropriate action. If it is determined to be normal, its status will continue to be monitored during subsequent use. If a problem is confirmed, it will need to be repaired or replaced.

7. A bridge condition anomaly diagnosis system based on support vector machine and BWIM data, characterized in that: The system is used to execute the bridge condition anomaly diagnosis method based on support vector machine and BWIM data as described in any one of claims 1 to 6; The system includes a calibration test module, a bridge inspection test module, a data processing module, and a bridge inspection module; The calibration test module is used to develop a calibration test plan for bridge dynamic loads; In the initial state of the bridge, a fixed displacement and strain measurement point is selected to identify the influence surface of the bridge. The bridge inspection and testing module is used to develop bridge inspection and testing plans; Based on the fixed displacement and strain measurement points determined by the calibration test module, and in conjunction with computer vision or spatial integrated monitoring methods, multiple test experiments are conducted on the bridge in its initial state to obtain information on the bridge's dynamic response, vehicle wheelbase, and lateral position on the bridge. The data processing module is used to perform dynamic weighing and weight distribution fitting of six-axle vehicles. Based on the bridge influence surface and bridge dynamic response data obtained from the calibration test module and the bridge inspection test module, multi-vehicle dynamic weighing is performed to obtain the total weight and distribution of the six-axle vehicle. The distribution curve of the total vehicle weight is fitted using kernel density to obtain and extract the corresponding detection characteristic indicators. The data processing module is also used for training a support vector machine with a first-class distribution. The multiple sets of feature indicators obtained by the data processing module are trained using a support vector machine to form the corresponding feature indicator region in the initial state of the bridge. The bridge inspection module is used to dynamically weigh the bridge under inspection. The bridge inspection test module describes the inspection experiment performed on the bridge under inspection. The dynamic weighing results of multiple vehicles are subjected to kernel density fitting to extract feature indicators, which are then compared with the feature indicator regions obtained by the data processing module to determine the bridge condition.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium for storing computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 6.

Images