Bridge support damage detection method based on axle contact point response

Abstract

The invention discloses a bridge support damage identification method based on axle contact point response, which comprises the following steps: 1, establishing an axle finite element model, adding axle uncertainty and support damage to simulate an axle coupling dynamic load test, and collecting vehicle response test data; 2, determining a contact point displacement response signal of the double-axle vehicle in the test scene according to the vehicle response, the driving speed of the double-axle vehicle and the vehicle parameters of the double-axle vehicle; 3, carrying out time shift subtraction on the contact point displacement response signals, and carrying out feature selection on the data by utilizing a Pearson's linear correlation coefficient; 4, training a CNN-BILSTMXGB neural network model by utilizing the contact point signals and corresponding support damage experiment data, and constructing improved parameters for optimizing the CNN-BILSTMXGB; and 5, substituting the contact point signal data of the prediction set into the improved network model to predict the damage of the bridge support. According to the invention, the problems of large sensor installation number, low detection speed, insufficient detection precision and the like in bridge support damage detection work can be effectively solved.

Description

Technical Field

[0001] The present invention belongs to the field of health monitoring of bridge bearings, and specifically relates to a method for detecting bridge bearing damage based on the response of the vehicle-bridge contact point. Background Art

[0002] As a key component connecting various parts of a bridge structure, when damage occurs to a bridge bearing, it will not only affect the bearing capacity and safety of the bridge, but may also accelerate the aging of the bridge, increase maintenance costs, and threaten traffic safety. Therefore, bridge bearing monitoring is of great significance in ensuring the safety of bridge structures, extending service life, and improving traffic safety. In the prior art, the recognition method based on visual detection has the advantages of being fast, non-contact, and low-cost in bridge bearing damage recognition, but it is greatly affected by environmental conditions and has insufficient accuracy; the bearing damage recognition method based on the dynamic response of the bridge is widely used because of its non-destructive detection and high sensitivity, but there are problems such as the use of more sensors, cumbersome steps in using detection equipment, and difficulties in damage location and quantification; the bearing damage recognition technology based on machine learning has the characteristics of automation and high efficiency, but there is still a problem of insufficient generalization ability of the model. Summary of the Invention

[0003] In order to overcome the deficiencies of the prior art, the present invention proposes a method for detecting bridge bearing damage based on the response of the vehicle-bridge contact point, in order to achieve high-precision recognition of bridge bearing damage location and quantification, and effectively overcome the deficiencies and drawbacks of high test costs, long time, and low accuracy in bridge bearing damage recognition.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] The method for detecting bridge bearing damage based on the response of the vehicle-bridge contact point of the present invention is characterized by including the following steps:

[0006] Step 1: Two displacement sensors are respectively arranged on the front axle and the bottom plate of the detection vehicle, and an angle sensor is arranged on the vehicle chassis;

[0007] Step 2: Use finite element software to model the bridge and the vehicle to obtain a vehicle-bridge finite element model, including: a bridge finite element model and a vehicle finite element model; among them, the bridge has bearings, and the vehicle finite element model is a double-axle vehicle model;

[0008] Step 3: Add uncertainties to the vehicle-bridge finite element model to simulate the real structures of the bridge and the vehicle;

[0009] Step 4: Use the uniform design method to select N kinds of bearing damage conditions to obtain the bridge bearing damage degree matrix , where is the i-th damage condition, and , is the damage degree of the j-th bearing under the i-th damage condition;

[0010] Use finite element software to perform dynamic load calculations on the bearing stiffness of the bridge under the i-th damage condition to obtain the vertical displacement responses of the centers of gravity of the two axles under the i-th damage condition , the vertical displacement response of the center of gravity of the chassis , the angular response of the center of gravity of the chassis ; among them, represents the bearing stiffness of the bridge model in the normal state, is the vertical displacement data of the center of gravity of the -th axle at the j-th moment under the i-th damage condition, ; is the vertical displacement data of the center of gravity of the chassis at the j-th moment under the i-th damage condition; is the rotational response signal of the center of gravity of the chassis at the j-th moment under the i-th damage condition;

[0011] Step 5: Use Equation (4) to obtain the displacement response signal of the contact point corresponding to the -th axle at the j-th moment under the i-th damage condition , so as to obtain the displacement response signals of the contact points corresponding to the -th axle at all moments under the i-th damage condition ;

[0012] (4)

[0013] In Equation (4), is the mass of the -th axle; is the second derivative with respect to time; is the stiffness of the -th axle; is the wheelbase of the -th axle; is the damping of the -th axle; is the first derivative with respect to time; is the first derivative with respect to time; is the first derivative with respect to time; is the tire stiffness corresponding to the -th axle;

[0014] Step 6: Obtain the displacement response signals of the corresponding contact points of the first axle and the second axle at all moments under the i-th damage condition , and after performing time shift subtraction on the displacement responses of the two axle contact points, obtain the time shift difference data set of the contact points under the i-th damage condition ;

[0015] Step 7: Calculate the Pearson linear correlation coefficients between each time shift difference data of the contact points in and the damage degree data of M supports respectively, take the absolute values of the Pearson linear correlation coefficients, and then sort them in descending order, so as to perform standardization processing on the characteristic data corresponding to the first B Pearson linear correlation coefficients, and obtain the processed time shift difference data set of the contact points under the i-th damage condition , where B represents the number of feature selections set;

[0016] Step 8: Create a neural network model, including: an input layer, a pattern layer, a summation layer and an output layer, using the processed time shift difference data of the contact points under the i-th damage condition as the input of the neural network model, using as the output of the neural network model, and train the neural network model, and obtain the trained bridge support damage prediction model under the i-th damage condition, which is used to identify the damage degree of the bridge support.

[0017] The characteristics of a bridge support damage detection method based on the response of the vehicle-axle contact point described in the present invention also lie in that step 3 includes the following steps:

[0018] Step 3.1: Simulate the bridge deck roughness by setting the grade of the pavement roughness of the bridge finite element model ;

[0019] Step 3.2: Simulate the uncertainty of the vehicle by changing the suspension stiffness , suspension damping and vehicle speed of the double-axle vehicle model, so as to simulate the uncertainty of the vehicle by using formula (1);

[0020] (1)

[0021] In formula (1), represents the suspension stiffness of the double-axle vehicle model input in the test; represents the damping of the double-axle vehicle model input in the test, represents the vehicle speed of the double-axle vehicle model input in the test; , represents 2 random numbers, and , , are two change amplitudes; represents the loading time;

[0022] Step 3.3: By changing the elastic modulus of the bridge finite element model, the influence of temperature on the bridge is simulated using Equation (2);

[0023] (2)

[0024] In Equation (2), represents the elastic modulus of the bridge model input in the test, represents the third random number, and ;

[0025] Step 3.4: By changing the density of the bridge finite element model , the influence of temperature on the bridge is simulated using Equation (3); (3)

[0026] In Equation (3), represents the density of the bridge model input in the test, represents the fourth random number, and .

[0027] Furthermore, Step 8 includes the following steps:

[0028] Step 8.1: The input layer transmits to the pattern layer for convolution operation to extract local features , and the pattern layer then uses activation function to process the local features to obtain non-linearly processed data , and then performs max pooling on the non-linearly processed data to obtain the feature map data after dimensionality reduction ;

[0029] Step 8.2: The summation layer includes: model and model, and processes respectively, and correspondingly obtains the predicted value of the damage degree of the bridge bearing in the i-th damage case predicted by the model and its weight and the predicted value of the damage degree of the bridge bearing in the i-th damage case predicted by the model and its weight ;

[0030] Step 8.3: The output layer uses Equation (12) to obtain the predicted result of the damage degree of the bridge bearing under the i-th damage condition. ;

[0031] (12)

[0032] Step 8.4: Based on and construct a loss function for the damage degree of the bridge bearing and use it to train the neural network model to update the model parameters, so as to obtain the trained bridge bearing damage prediction model under the i-th damage condition after training.

[0033] Furthermore, Step 8.2 includes the following steps:

[0034] Step 8.2.1: The model performs forward and reverse processing on respectively, and correspondingly obtains the forward predicted value of the damage degree of the bridge bearing and the reverse predicted value of the damage degree of the bridge bearing , so as to obtain the predicted value of the damage degree of the bridge bearing under the i-th damage condition by using Equation (5);

[0035] (5)

[0036] In Equation (5), , represent the forward weight and the reverse weight respectively;

[0037] Step 8.2.2: The model processes by using Equation (6) to obtain the predicted value of the damage degree of the bridge bearing under the i-th damage condition:

[0038] (6)

[0039] In Equation (6), is the learning rate; represents the tree model at the t-th iteration; n represents the total number of iterations;

[0040] Step 8.2.3: Calculate the error of the model and the error of the model respectively by using Equation (7) and Equation (8):

[0041] (7)

[0042] (8)

[0043] Step 8.2.4: Determine whether formula (9) holds. If it holds, execute Step 8.2.5; otherwise, execute Step 8.2.6;

[0044] (9)

[0045] In formula (9), represents the set weight threshold;

[0046] Step 8.2.5: Use formula (10) and formula (11) to calculate the weights of the model and weights of the model :

[0047] (10)

[0048] (11)

[0049] Step 8.2.6: If , then let , let ;

[0050] If , then let , let .

[0051] An electronic device according to the present invention includes a memory and a processor, characterized in that the memory is used to store a program for supporting the processor to execute the damage identification of the bridge bearing, and the processor is configured to execute the program stored in the memory.

[0052] A computer-readable storage medium according to the present invention, characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is run by a processor, it executes the steps of the bridge bearing damage identification prediction method described above.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0054] 1. Compared with damage identification based on visual inspection, the present invention has the advantages of low operation difficulty, small environmental impact, and high accuracy.

[0055] 2. Compared with the identification method based on the dynamic response of the bridge, the present invention does not require installing sensors on the bridge body, is easy to operate with low cost, and only requires a single train movement during the detection process, etc.

[0056] 3. Compared with the recognition method based on the traditional neural network, the present invention solves the problems of insufficient generalization, robustness, and accuracy of a single neural network algorithm by combining the test results of the EMU with the neural network algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic diagram of the specific process of the method of the present invention;

[0058] Figure 2 It is a front view schematic diagram of the simply supported beam bridge and vehicle finite element model of the present invention;

[0059] Figure 3 For the present invention Schematic diagram of the network structure;

[0060] Figure 4 It is a comparison diagram of the predicted value and the true value of the damage of the bridge bearing of the simply supported beam bridge of the invention. DETAILED DESCRIPTION OF THE INVENTION

[0061] In this embodiment, a bridge damage identification method based on the contact point response is to establish the relationship between the contact point response on the vehicle-bridge system and the damage of the bridge bearing, and use the on-bridge EMU test data to identify the location and size of the bridge bearing damage. Specifically, as Figure 1 shown, this method is carried out according to the following steps:

[0062] Step 1: Two displacement sensors are respectively arranged on the front axle and the bottom plate of the front of the detection vehicle, and an angle sensor is arranged on the vehicle chassis;

[0063] Step 2: Use finite element software to model the bridge and the vehicle to obtain a vehicle-bridge finite element model, including: a bridge finite element model and a vehicle finite element model; among them, the bridge has bearings, each of which is simulated by four three-way springs, and the vehicle finite element model is a double-axle vehicle; the bridge section is a box section, 40 m long, and the front view schematic diagram of the simply supported beam bridge and vehicle finite element model is as Figure 2 shown;

[0064] Step 3: Add uncertainties to the vehicle-bridge finite element model to simulate the real structures of the bridge and the vehicle;

[0065] Step 3.1: By setting the level of the road surface roughness of the bridge finite element model to simulate the roughness of the bridge deck;

[0066] Step 3.2: By changing the suspension stiffness , suspension damping and vehicle speed of the double-axle vehicle model, thereby simulating the uncertainty of the vehicle by using Equation (1);

[0067] (1)

[0068] In formula (1), represents the suspension stiffness of the double-axle vehicle model for test input; represents the damping of the double-axle vehicle model for test input, represents the vehicle speed of the double-axle vehicle model for test input; , represents two random numbers, and , ; are two change amplitudes, is the change amplitude; represents the loading time, and .

[0069] Step 3.3: By changing the elastic modulus of the bridge finite element model, thus using formula (2) to simulate the influence of temperature on the bridge;

[0070] (2)

[0071] In formula (2), represents the elastic modulus of the bridge model for test input, represents the third random number, where , .

[0072] Step 3.4: By changing the density of the bridge finite element model , thus using formula (3) to simulate the influence of temperature on the bridge; (3)

[0073] In formula (3), represents the density of the bridge model for test input, represents the fourth random number, and ; where .

[0074] Step 4: Use the uniform design method to select N = 1000 cases of bearing damage, and obtain the bridge bearing damage degree matrix , where, is the i-th damage case, and , is the damage degree of the j-th bearing in the i-th damage case;

[0075] Use the finite element software to perform dynamic load operation on the bearing stiffness of the bridge in the i-th damage case, and obtain the vertical displacement response of the two-axle center of gravity, the vertical displacement response of the chassis center of gravity, and the angle response of the chassis center of gravity ; Among them, represents the stiffness of the bridge model support for each operation, represents the stiffness of the bridge model support under normal conditions, is the vertical displacement data of the center of gravity of the th axle at the jth moment under the ith damage, ; is the vertical displacement data of the center of gravity of the chassis at the jth moment under the ith damage; is the rotational response signal at the jth moment under the ith damage.

[0076] Step 5: Use Equation (4) to obtain the displacement response signal of the contact point corresponding to the th axle at the jth moment under the ith damage :

[0077] (4)

[0078] In Equation (4), is the mass of the th axle; is the second derivative with respect to time; is the stiffness of the th axle; is the wheelbase of the th axle; is the damping of the th axle; is the first derivative with respect to time; is the first derivative with respect to time; is the first derivative with respect to time; is the tire stiffness corresponding to the

[0079] th axle. Step 6: Obtain the displacement response signals of the contact points corresponding to the first axle and the second axle at all moments under the ith damage situation, .

[0080] Step 7: Calculate for each damage situation The Pearson linear correlation coefficients between each piece of feature data and the damage degree data of M = 1000 bearings are calculated. After taking the absolute values of the respective correlation coefficients and sorting them in descending order, the first B pieces of feature data corresponding to the B largest correlation coefficients are selected according to the number B of features to be selected, and then standardized to obtain the processed feature data, and the processed time shift difference data of the contact points are obtained. , where B represents the number of features to be selected.

[0081] Step 8: Create a neural network model; including: an input layer, a pattern layer, a summation layer, and an output layer. Using the processed time shift difference data of the contact points in the i-th damage case as the input of the neural network model, and using as the output of the neural network model, and training the neural network model to obtain the trained bridge bearing damage prediction model in the i-th damage case; The network structure diagram is as Figure 3 shown.

[0082] Step 8.1: The input layer transmits to the pattern layer for convolution operation to extract local features , and the first-step convolution operation is performed using Equation (5):

[0083] (5)

[0084] where k is the size of the convolution kernel, .

[0085] Step 8.2: Then, using Equation (6), is selected as the activation function to introduce non-linearity to obtain the non-linearly processed data , and then the non-linearly processed data is processed by max pooling to obtain the feature map data .

[0086] (6)

[0087] Step 8.3: The summation layer includes: a model and a model;

[0088] Step 8.3.1: The model performs forward and reverse processing on respectively, and correspondingly obtains the forward bridge bearing damage degree prediction value and the reverse bridge bearing damage degree prediction value , so as to obtain the predicted value of the damage degree of the bridge bearing under the i-th damage condition by using Equation (7). :

[0089] (7)

[0090] In Equation (7), , respectively represent the forward weight and the reverse weight;

[0091] Step 8.3.2: The model processes by using Equation (8) to obtain the predicted value of the damage degree of the bridge bearing under the i-th damage condition :

[0092] (8)

[0093] In Equation (8), is the learning rate; represents the tree model of the t-th iteration; n represents the total number of iterations.

[0094] Step 8.3.3: Calculate the error of the model and and the error of the model respectively by using Equation (9) and Equation (10):

[0095] (9)

[0096] (10)

[0097] Step 8.3.4: Judge whether Equation (11) holds. If it holds, execute Step 8.3.5; otherwise, execute Step 8.3.6;

[0098] (11)

[0099] In Equation (11), represents the set weight threshold.

[0100] Step 8.3.5: Calculate the weight of the model and and the weight of the model respectively by using Equation (12) and Equation (13):

[0101] (12)

[0102] (13)

[0103] Step 8.3.6: If , then let , let ;

[0104] If , then let , let .

[0105] Step 8.4: After the two are combined and the model is integrated, the prediction result obtained by using Equation (14) is output by the output layer:

[0106] (14)

[0107] Step 8.5: Based on and , construct a loss function for the damage degree of the bridge bearing and use it to train the neural network model to update the model parameters, so as to obtain the trained bridge bearing damage prediction model under the i-th damage condition for identifying the damage degree of the bridge bearing.

[0108] In this embodiment, an electronic device includes a memory and a processor. The memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0109] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium. When the computer program is run by a processor, it executes the steps of the above method.

[0110] Table 1

[0111]

[0112] As can be seen from Table 1, the error of the test prediction result is less than 1.5%, and the prediction result is accurate. The comparison diagram of the predicted value and the true value of the bridge bearing damage of the simply supported beam bridge is as Figure 4 shown.

Claims

1. A bridge bearing damage detection method based on vehicle-bridge contact point response, characterized in that: The steps include: Step 1: Two displacement sensors are respectively set on the front axle and the bottom plate of the test vehicle, and an angle sensor is set on the vehicle chassis; Step 2: Use finite element software to model the bridge and vehicle to obtain a vehicle-bridge finite element model, including: bridge finite element model and vehicle finite element model; the bridge has The vehicle finite element model is a two-axle vehicle model; Step 3: Add uncertainty to the vehicle-bridge finite element model to simulate the real structure of the bridge and vehicle; Step 4: Use the uniform design method to select N bearing damage conditions and obtain the bridge bearing damage degree matrix ,in, is the i-th damage case, and , is the damage degree of the jth support under the i-th damage condition; The stiffness of the bridge support under the i-th damage condition is calculated using finite element software. Perform dynamic load calculation to obtain the vertical displacement response of the center of gravity of the two axles under the i-th damage condition , vertical displacement response of chassis center of gravity 、Chassis center of gravity angle response ;in, represents the bridge model support stiffness under normal conditions, is the first time at the jth moment in the i-th damage case. The vertical displacement data of the center of gravity of each axle, ; is the vertical displacement data of the center of gravity of the chassis at the jth moment under the i-th damage condition; is the chassis center of gravity rotation response signal at the jth moment under the i-th damage condition; Step 5: Use formula (4) to obtain the first damage condition at the jth moment The displacement response signal of the contact point corresponding to each axle , thus obtaining the first The displacement response signal of the contact point corresponding to each axle ; (4) In formula (4), For the Axle mass; for The second derivative with respect to time; For the The stiffness of the axles; For the The wheelbase of the axles; For the Damping of each axle; for First derivative with respect to time; for First derivative with respect to time; for First derivative with respect to time; For the The tire stiffness corresponding to each axle; Step 6: Obtain the displacement response signals of the contact points corresponding to the first and second axles at all times under the i-th damage condition , and subtract the displacement responses of the two axle contact points in time, and obtain the contact point time displacement difference data set in the i-th damage case ; Step 7: Calculate the i-th damage case The Pearson linear correlation coefficient between the time displacement difference data of each contact point and the damage degree data of the M bearings is calculated, and the absolute value of each Pearson linear correlation coefficient is taken and sorted in descending order, so as to standardize the characteristic data corresponding to the first B Pearson linear correlation coefficients, and obtain the contact point time displacement difference data set after processing in the i-th damage case. , where B represents the number of feature selections set; Step 8: Create The neural network model includes: input layer, pattern layer, summation layer and output layer, with the contact point time displacement difference data processed under the i-th damage condition As The input of the neural network model is As The output of the neural network model and The neural network model is trained, and the trained bridge bearing damage prediction model under the i-th damage case is obtained to identify the damage degree of the bridge bearing.

2. The bridge support damage detection method based on vehicle-bridge contact point response according to claim 1 is characterized in that: Step 3 includes the following steps: Step 3.1: Set the road surface roughness level of the bridge finite element model To simulate the roughness of the bridge deck; Step 3.2: By changing the suspension stiffness of the two-axle vehicle model , suspension damping and vehicle speed , thus using formula (1) to simulate the uncertainty of the vehicle; (1) In formula (1), represents the suspension stiffness of the two-axle vehicle model of the test input; represents the damping of the two-axle vehicle model of the test input, Indicates the speed of the two-axle vehicle model input in the test; , represents 2 random numbers, and , , is 2 change amplitudes; Indicates loading time; Step 3.3: By changing the elastic modulus of the bridge finite element model, the influence of temperature on the bridge is simulated using equation (2); (2) In formula (2), represents the elastic modulus of the bridge model input by the test, represents the third random number, and ; Step 3.4: By changing the density of the bridge finite element model , thus using formula (3) to simulate the influence of temperature on the bridge; (3) In formula (3), represents the bridge model density of the test input, represents the fourth random number, and .

3. The bridge support damage detection method based on vehicle-bridge contact point response according to claim 2 is characterized in that: Step 8 includes the following steps: Step 8.1: The input layer will Transmitted to the pattern layer for convolution operation to extract Local features , the pattern layer reuses Activation function for local features Processing is performed to obtain nonlinear processing data , and then process the data nonlinearly Perform maximum pooling to obtain the feature map data after dimensionality reduction ; Step 8.2: The summation layer includes: Model and model, and respectively Processing, the corresponding Prediction value of bridge bearing damage degree under the i-th damage condition predicted by the model and its weight and Prediction of bridge bearing damage degree under the i-th damage condition predicted by the model and its rights ; Step 8.3: The output layer uses formula (12) to obtain the predicted value of the bridge bearing damage degree under the i-th damage condition: ; (12) Step 8.4: Based on and Construct the damage degree loss function of bridge bearings and use it to The neural network model is trained to update the model parameters, thereby obtaining the trained bridge bearing damage prediction model under the i-th damage case after training.

4. The bridge support damage detection method based on vehicle-bridge contact point response according to claim 3 is characterized in that: Step 8.2 includes the following steps: Step 8.2.1: Model pair Forward and reverse processing are performed respectively, and the forward bridge bearing damage degree prediction value is obtained accordingly and reverse bridge bearing damage degree prediction value , and thus the predicted value of the damage degree of the bridge bearing under the i-th damage condition is obtained using formula (5): ; (5) In formula (5), , Represent the positive weight and negative weight respectively; Step 8.2.2: The model uses formula (6) to After processing, the predicted value of bridge bearing damage degree under the i-th damage condition is obtained. : (6) In formula (6), is the learning rate; represents the tree model of the tth iteration; n represents the total number of iterations; Step 8.2.3: Use equations (7) and (8) to calculate Model Error and Model Error : (7) (8) Step 8.2.4: Determine whether equation (9) is true. If so, proceed to step 8.2.5; otherwise, proceed to step 8.2.6; (9) In formula (9), Indicates the set weight threshold; Step 8.2.5: Use equation (10) and equation (11) to calculate Model weights and Model weights : (10) (11) Step 8.2.6: If , then let ,make ; like , then let ,make .

5. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the bridge bearing damage identification as described in any one of claims 1-4, and the processor is configured to execute the program stored in the memory.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the bridge bearing damage identification and prediction method described in any one of claims 1 to 4 are executed.

Images