Bridge damage identification method under moving load

Abstract

The invention discloses a bridge damage identification method under a moving load. The method comprises the following steps: 1, substituting bridge damage displacement data obtained through finite element simulation into a residual neural network to obtain a predicted damage reduction coefficient; 2, based on a Newmark-beta method, embedding available physical knowledge of a bridge structure into a feature learning process, and substituting a damage reduction coefficient obtained by network prediction into a physical equation to obtain predicted damage displacement data; 3, substituting the solved damage reduction coefficient and displacement damage data into a loss function to obtain a loss value, and updating network parameters; and 4, predicting bridge damage by using the trained physical guidance residual neural network. The method has high damage identification precision under the condition of considering the uncertainty of the vehicle weight and the uncertainty of the bridge structure parameters, and can improve the accuracy of damage identification of the network when the actually measured data is less.

Description

Technical Field

[0001] in the field of bridge inspection, specifically a method for bridge damage identification under moving loads. Background Art

[0002] During their service, a large number of bridges will suffer from diseases and damage due to various factors such as environmental erosion, long-term loads and material aging, resulting in degradation of structural performance and reduced safety. Accurate monitoring of bridge damage is helpful to evaluate the safety performance and usage status of the structure, and help managers operate the bridges in service. In the past decade, people have conducted a lot of research on vibration-based structural damage identification technology and have achieved substantial results. These results are based on the principle that changes caused by damage to structural characteristics will lead to detectable changes in structural vibration characteristics. Therefore, the current health status of the structure can be determined by extracting damage-sensitive features and patterns from the measured vibration characteristics.

[0003] In practice, physical and deep learning methods are often used for damage identification. Physics-based methods directly or indirectly use the physical laws that control the behavior of the structure in order to extract meaningful information about the damage and its evolution from the measured vibration response. Physics-based methods require the establishment of an accurate finite element model of the bridge. However, in practical applications, even with the knowledge of the entire model, it is difficult to establish a complete mechanism model through this knowledge. In addition, the uncertainty of the bridge structure may seriously affect the performance of structural model updating and damage identification. Deep learning-based methods use a large amount of training data without the need for an accurate initial structural finite element model to learn abstract features from training data under different damage conditions and make predictions based on the learned relationships. These training data sets can take into account the influence of parameter uncertainty of the bridge structure and output more accurate damage identification results under the influence of uncertainty. However, training a reliable machine learning model usually requires sufficient data, which is not available in most bridge damage identification engineering tasks. Another problem with deep learning-based damage identification methods is that deep learning models are only trained on data and lack physical interpretability for many engineering problems. Summary of the invention

[0004] The present invention aims to solve the shortcomings of existing damage identification methods and provide a bridge damage identification method under moving load, so as to have higher damage identification accuracy while considering the uncertainty of vehicle weight and bridge structure parameters, and to improve the accuracy of damage identification when the network has less measured data.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme.

[0006] The present invention provides a bridge damage identification method under a moving load, which comprises the following steps:

[0007] Step 1: Establish the vehicle-bridge coupling model:

[0008] The bridge model is constructed using finite element software, and a moving load is applied to the bridge model for simulation, so as to determine the length of the bridge model. , the elastic modulus is ;

[0009] The vehicle model is constructed using finite element software, and the parameters of the vehicle model are determined, including: the weight of the vehicle model is , the driving speed of the vehicle model ;

[0010] The vehicle model moves at a constant speed After passing the bridge model, the number of sampling points is calculated as ;in, is the sampling frequency, is the bridge crossing time of the vehicle model, and ;

[0011] Step 2: Set the damage condition and obtain the displacement response data of the bridge model

[0012] Step 2.0: Define and initialize the type of working condition ;

[0013] Step 2.1: Define and initialize the sequence number of the current unit point ;

[0014] Step 2.2: Define the i-th damage reduction factor as , and initialize ;

[0015] Step 2.3: Add the i-th damage reduction factor to each element point , forming the first Damage Conditions ;in, Indicates The i-th damage reduction factor is applied to each element point The damage conditions; p is the total number of unit points on the bridge model;

[0016] Step 2.4: Randomly initialize the variables , so that the vehicle model is Uncertain vehicle weight Through the bridge model, and use Method to obtain the bridge model in Damage Conditions Displacement response data under ,in, is the peak value of change;

[0017] Step 2.5: Assign to Then return to step 2.4 and execute sequentially until So far, we can get The i-th damage reduction factor is applied to each element point of The displacement response data of the bridge model under various damage conditions; Indicated in The i-th damage reduction factor is applied to each element point The number of uncertain vehicle weight types set when

[0018] Step 2.6: Assign to Then return to step 2.3 and execute sequentially until So far, we can get Apply to each unit point The damage reduction factor The displacement response data of the bridge model under various damage conditions; Indicates the number of damage reduction factor types set;

[0019] Step 2.7: Assign to Then return to step 2.2 and execute sequentially until Until, under the uncertainty of vehicle weight of the vehicle model, the p unit points are applied successively The damage reduction factor Damage Conditions Displacement response data under ; n represents the total number of damage conditions, and ;

[0020] Step 3: Input the residual neural network for processing to obtain the bridge model Prediction value of damage condition ;

[0021] Step 4: Construct a physics-guided residual neural network and use it to obtain the bridge model in the first Prediction value of damage condition Prediction of displacement response of bridge model under ;

[0022] Step 4.1: Use equation (1) to get the bridge model in the Prediction value of damage condition The global stiffness matrix under ;

[0023] (1)

[0024] In formula (1), Represents the overall stiffness matrix of the bridge model in the undamaged state;

[0025] Step 4.2: Use equation (2) to construct the Newmark-β method structural dynamics equation:

[0026] (2)

[0027] In formula (2), M represents the overall mass matrix of the bridge model, C represents the overall damping matrix of the bridge model, The representative bridge model is Prediction value of damage condition The overall stiffness matrix under The representative bridge model is Prediction value of damage condition The acceleration response data under The representative bridge model is Prediction value of damage condition The speed response data under The representative bridge model is Prediction value of damage condition The displacement response data under The representative bridge model is Prediction value of damage condition Moving load under

[0028] Step 4.3: Use the Newmark-β method to solve the structural dynamics equation and obtain the bridge model in the first Prediction value of damage condition Prediction of bridge displacement response under ;

[0029] Step 5: Use formula (3) to construct the total loss function :

[0030] (3)

[0031] In formula (3), is the weight to adjust the data loss, is the weight to adjust the physical loss; represents the data loss function, represents the physical loss function, and has:

[0032] (4)

[0033] (5)

[0034] Step 6: Based on and , use the Adam optimizer to train the physically guided residual neural network and calculate the total loss function Update network parameters until Until convergence, the optimal physical-guided bridge damage identification model under the uncertainty vehicle weight of the vehicle model is obtained, which is used to realize the damage identification of a certain part of the bridge.

[0035] The bridge damage identification method under moving load described in the present invention is also characterized in that:

[0036] Replace the vehicle model uncertainty in step 2.4 with the parameter uncertainty of the bridge model, thus Change the moment of inertia of the bridge model, and follow the process from step 2.0 to step 2.7 to obtain the uncertainty moment of inertia of the bridge model. The damage reduction factor Damage Conditions Displacement response data under ; And according to the process of step 3 to step 6, the optimal physical-guided bridge damage identification model of the bridge model under the uncertainty moment of inertia is obtained; wherein, represents the bridge model under uncertainty moment of inertia damage condition, represents the bridge model under uncertainty moment of inertia Damage Conditions Displacement response data under .

[0037] The electronic device of the present invention includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the bridge damage identification method, and the processor is configured to execute the program stored in the memory.

[0038] The present invention provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the bridge damage identification method when the computer program is executed by a processor.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. Compared with the traditional model correction method, the present invention does not require an accurate finite element model. The network learns the characteristics of the damage under different working conditions and makes predictions based on the learned relationship. At the same time, the data can take into account the influence of vehicle weight uncertainty and bridge structure parameter uncertainty, thereby enabling accurate prediction of bridge damage in complex situations.

[0041] 2. Compared with the bridge damage identification method based on deep learning, the present invention adds the structural dynamics equation as physical knowledge to the network training process based on deep learning. This process similar to finite element solution implemented in network training narrows the gap between input response and predicted response, thereby further optimizing the bridge damage coefficient. At the same time, in the case of limited data samples, the embedded physical knowledge realizes the constraints on the network input and output results, retaining its physical meaning. This method based on physical guidance can ensure that the network obtains more accurate identification results with limited data volume. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of the calculation process of the present invention;

[0043] Figure 2 It is a schematic diagram of a numerical simulation uniform cross-section simply supported beam model of the present invention;

[0044] Figure 3 It is a schematic diagram of the network structure of the present invention;

[0045] Figure 4 A comparison diagram of the damage reduction coefficient prediction value and the actual value of the uniform cross-section simply supported beam bridge under moving load excitation according to the present invention;

[0046] Figure 5 This is a comparison chart between the predicted value and the actual value of the damage reduction coefficient of a simply supported beam bridge of equal cross-section under the condition of a small data set.

[0047] Figure 6 This is a comparison chart between the predicted value and the actual value of the damage reduction coefficient of a simply supported beam bridge of uniform cross-section under the influence of bridge structure uncertainty of the present invention. DETAILED DESCRIPTION

[0048] In this embodiment, Figure 2 The elastic modulus of the bridge is , the density of bridge per linear meter is , bridge section height , the width is , finite element software was used for simulation.

[0049] A bridge damage identification method under moving load is to use a physics-guided residual neural network for damage identification, add the dynamic control equation to the feature learning as a physical constraint to guide the network to learn, and formulate the Newmark-β method as the available physical knowledge into the loss function to provide physical constraints for the network output, thereby achieving accurate identification of bridge damage under the condition of limited data samples and large influence of bridge structure parameter uncertainty, thereby improving the accuracy of bridge damage identification. Specifically, the principle of a bridge damage identification method under moving load, the simply supported beam model used is as follows Figure 1 As shown, the method comprises the following steps:

[0050] Step 1: Establish the vehicle-bridge coupling model:

[0051] The bridge model was constructed using finite element software, and moving loads were applied to the bridge model for simulation, so as to determine the length of the bridge model. , the elastic modulus is ;

[0052] There are p=10 unit points on the bridge model, and =3 unit points as observation points, and the The unit point number corresponding to each observation point is recorded as j, and the displacement response data of the bridge model under different damage conditions are recorded;

[0053] Use finite element software to build a vehicle model and determine the parameters of the vehicle model, including: the weight of the vehicle model is , front wheelbase is , rear wheelbase is , the driving speed of the vehicle model ;

[0054] The vehicle model moves at a constant speed After passing the bridge model, the number of sampling points is calculated as , the number of calculated sampling points is 100; among them, is the sampling frequency, is the bridge crossing time of the vehicle model, and ;

[0055] Step 2: Set the damage condition and obtain the displacement response data of the bridge model:

[0056] Step 2.0: Define and initialize the type of working condition ;

[0057] Step 2.1: Define and initialize the sequence number of the current unit point ;

[0058] Step 2.2: Define the i-th damage reduction factor as , and initialize , It represents the damage reduction factor of the bridge model, which is obtained by reducing the stiffness of the partition in the bridge finite element model. To simulate bridge damage;

[0059] Step 2.3: Add the i-th damage reduction factor to each element point , forming the first Damage Conditions ;in, Indicates The i-th damage reduction factor is applied to each element point The damage condition, Except The values ​​of all other positions except the current position are 0 to ensure that each set of damage conditions has only one damage position.

[0060] Step 2.4: Randomly initialize the variables , so that the vehicle model is Uncertain vehicle weight Through the bridge model, and use Method to obtain the bridge model in Damage Conditions The displacement response data of the bridge model under ,in, is the peak value of change;

[0061] Step 2.5: Assign to Then return to step 2.4 and execute sequentially until So far, we get The i-th damage reduction factor is applied to each element point of The displacement response data under various damage conditions; Indicated in The i-th damage reduction factor is applied to each element point of The number of uncertain vehicle weight types under the damage condition, .

[0062] Step 2.6: Assign to Then return to step 2.3 and execute sequentially until So far, we get Apply to each unit point The damage reduction factor The displacement response data under various damage conditions; represents the number of damage reduction coefficient types set in this method. In this embodiment, , There are 4 different values .

[0063] Step 2.7: Assign to Then return to step 2.2 and execute sequentially until Until, under the uncertainty of vehicle weight of the vehicle model, the p unit points are applied successively The damage reduction factor Damage Conditions Displacement response data under ; n = 2200 represents the total number of damage conditions, and .

[0064] Step 3: Construct a residual neural network, including: convolutional layer, residual block, pooling layer, global average pooling layer, fully connected layer, and RELU activation function;

[0065] Will In the residual neural network, the input layer is passed first, and then through the continuous The convolutional kernel size of the last convolutional layer is , the number of convolution kernels is Each convolutional layer is followed by an activation layer and a normalization layer. After the convolutional layer, the data enters The combined structure of residual blocks, each of which consists of After passing through the residual block, the data passes through the adaptive average pooling layer, and finally the data passes through the fully connected layer to obtain the first Prediction value of damage condition .

[0066] Step 4: Figure 3 As shown in the figure, a physics-guided residual neural network is constructed and used to obtain the bridge model in the first Prediction value of damage condition Prediction of bridge displacement response under ;

[0067] Step 4.1: Use equation (1) to get the bridge model in the Prediction value of damage condition The global stiffness matrix under ;

[0068] (1)

[0069] In formula (1), Represents the overall stiffness matrix of the bridge model in the undamaged state.

[0070] Step 4.2: Use equation (2) to construct the structural dynamics equation:

[0071] (2)

[0072] In formula (2), M represents the mass matrix of the bridge model, C represents the damping matrix of the bridge model, The representative bridge model is Prediction value of damage condition The stiffness matrix under The representative bridge model is Prediction value of damage condition The acceleration under The representative bridge model is Prediction value of damage condition The speed of The representative bridge model is Prediction value of damage condition The displacement response data under The representative bridge model is Prediction value of damage condition Moving load under.

[0073] Step 4.3: Solve the structural dynamics equation using the Newmark-β method to obtain the bridge model in the first Prediction value of damage condition Prediction of bridge displacement response under .

[0074] Step 5: Use formula (3) to construct the total loss function :

[0075] (3)

[0076] In formula (3), is the weight to adjust the data loss, is the weight to adjust the physical loss; represents the data loss function, represents the physical loss function, and has:

[0077] (4)

[0078] (5)

[0079] Step 6: Based on and , use the Adam optimizer to train the physically guided residual neural network and calculate the total loss function Update network parameters until Until convergence, the optimal physical-guided bridge damage identification model under the uncertainty vehicle weight of the vehicle model is obtained, which is used to realize the damage identification of a certain part of the bridge.

[0080] In this embodiment, an electronic device includes a memory and a processor, wherein 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.

[0081] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium, and the computer program executes the steps of the above method when executed by a processor.

[0082] In this embodiment 1, Figure 2 The elastic modulus of the uniform cross-section simply supported beam bridge shown is , the density of bridge per linear meter is , bridge section height , the width is , using finite element software to establish a finite element model of the bridge. Considering the uncertainty of vehicle weight, select The data is used as the training set, The data is used as the validation set, The data is used as the test set. The actual values ​​and prediction results of the damage condition are shown in Figure 4 and Table 1.

[0083] In this embodiment 2, the basic parameters of the bridge remain unchanged. Considering the uncertainty of vehicle weight, The data is used as the training set, The data is used as the validation set, The data is used as the test set. The actual values ​​and prediction results of the damage condition are shown in Figure 5 and Table 1.

[0084] In this embodiment, the vehicle model uncertainty in step 2.4 is replaced by the parameter uncertainty of the bridge model, so Change the moment of inertia of the bridge model, and follow the process from step 2.0 to step 2.7 to obtain the uncertainty moment of inertia of the bridge model. The damage reduction factor Damage Conditions Displacement response data under ; And according to the process of step 3 to step 6, the optimal physical-guided bridge damage identification model of the bridge model under the uncertainty moment of inertia is obtained; wherein, represents the bridge model under uncertainty moment of inertia damage condition, represents the bridge model under uncertainty moment of inertia Damage Conditions Displacement response data under , n=2200.

[0085] In this embodiment 3, the basic parameters of the bridge remain unchanged. Considering the uncertainty of the bridge model structure, The data is used as the training set, The data is used as the validation set, The data is used as the test set. The actual values ​​and prediction results of the damage condition are shown in Figure 6 and Table 1.

[0086] The mean square error (MSE) and regression value (R1) are defined to evaluate the structural damage identification performance of the network under Examples 1, 2, and 3. MSE measures the distance between the true value and the predicted output of the training model. The regression value R1 is between 0 and 1 and is also used to evaluate the quality of the training model. Specifically, the R1 value represents the correlation coefficient, which quantifies the linear correlation between the predicted output and the true value. The comparison values ​​of the mean square error (MSE) and regression value (R1) of Examples 1, 2, and 3 are shown in Table 1.

[0087] (6)

[0088] (7)

[0089] Where n represents the number of total damage conditions, represents the set damage condition parameters, represents the damage condition parameters predicted by the residual neural network, , Indicates the total number of partition units and the total number of damage reduction factor types The product of . Represents the jth unit The actual damage index under the damage reduction factor is: Represents the Jth unit. The average damage index predicted under the damage reduction factor.

[0090] Table 1

[0091]

[0092] As shown in Table 1, by constructing a residual neural network model with physical information, the damage of the bridge model under the action of a moving vehicle is identified, and relatively accurate prediction results are obtained in Examples 1, 2, and 3. In summary, the present invention can better predict the damage of the bridge when the amount of data is insufficient and the structural uncertainty and vehicle weight uncertainty are taken into account.

Claims

1. A bridge damage identification method under moving load, characterized in that: The following steps are involved: Step 1: Establish the vehicle-bridge coupling model: The bridge model is constructed using finite element software, and a moving load is applied to the bridge model for simulation, so as to determine the length of the bridge model. , the elastic modulus is ; The vehicle model is constructed using finite element software, and the parameters of the vehicle model are determined, including: the weight of the vehicle model is , the speed of the vehicle model ; The vehicle model moves at a constant speed After passing the bridge model, the number of sampling points is calculated as ;in, is the sampling frequency, is the bridge crossing time of the vehicle model, and ; Step 2: Set the damage condition and obtain the displacement response data of the bridge model Step 2.0: Define and initialize the type of working condition ; Step 2.1: Define and initialize the sequence number of the current unit point ; Step 2.2: Define the i-th damage reduction factor as , and initialize ; Step 2.3: Add the i-th damage reduction factor to each element point , forming the first Damage Conditions ;in, Indicates The i-th damage reduction factor is applied to each element point The damage conditions; p is the total number of unit points on the bridge model; Step 2.4: Randomly initialize the variables , so that the vehicle model is Uncertain vehicle weight Through the bridge model, and use Method to obtain the bridge model in Damage Conditions Displacement response data under ,in, is the peak value of change; Step 2.5: Assign to Then return to step 2.4 and execute sequentially until So far, we get The i-th damage reduction factor is applied to each element point of The displacement response data of the bridge model under various damage conditions; Indicated in The i-th damage reduction factor is applied to each element point The number of uncertain vehicle weight types set when Step 2.6: Assign to Then return to step 2.3 and execute sequentially until So far, we get Apply on each unit point The damage reduction factor The displacement response data of the bridge model under various damage conditions; Indicates the number of damage reduction factor types set; Step 2.7: Assign to Then return to step 2.2 and execute sequentially until Until, under the uncertainty of vehicle weight of the vehicle model, the p unit points are applied successively The damage reduction factor Damage Conditions Displacement response data under ; n represents the total number of damage conditions, and ; Step 3: Input the residual neural network for processing to obtain the bridge model Prediction value of damage condition ; Step 4: Construct a physics-guided residual neural network and use it to obtain the bridge model in the first Prediction value of damage condition Prediction of displacement response of bridge model under ; Step 4.1: Use equation (1) to get the bridge model in the Prediction value of damage condition The global stiffness matrix under ; (1) In formula (1), Represents the overall stiffness matrix of the bridge model in the undamaged state; Step 4.2: Use equation (2) to construct the Newmark-β method structural dynamics equation: (2) In formula (2), M represents the overall mass matrix of the bridge model, C represents the overall damping matrix of the bridge model, The representative bridge model is Prediction value of damage condition The overall stiffness matrix under The representative bridge model is Prediction value of damage condition The acceleration response data under The representative bridge model is Prediction value of damage condition The speed response data under The representative bridge model is Prediction value of damage condition The displacement response data under The representative bridge model is Prediction value of damage condition Moving load under Step 4.3: Use the Newmark-β method to solve the structural dynamics equation and obtain the bridge model in the first Prediction value of damage condition Prediction of bridge displacement response under ; Step 5: Use formula (3) to construct the total loss function : (3) In formula (3), is the weight to adjust the data loss, is the weight to adjust the physical loss; represents the data loss function, represents the physical loss function, and has: (4) (5) Step 6: Based on and , use the Adam optimizer to train the physically guided residual neural network and calculate the total loss function Update network parameters until Until convergence, the optimal physical-guided bridge damage identification model under the uncertainty vehicle weight of the vehicle model is obtained, which is used to realize the damage identification of a certain part of the bridge.

2. The bridge damage identification method under moving load according to claim 1 is characterized in that: Replace the vehicle model uncertainty in step 2.4 with the parameter uncertainty of the bridge model, thus Change the moment of inertia of the bridge model, and follow the process from step 2.0 to step 2.7 to obtain the uncertainty moment of inertia of the bridge model. The damage reduction factor Damage Conditions Displacement response data under ; And according to the process of step 3 to step 6, the optimal physical-guided bridge damage identification model of the bridge model under the uncertainty moment of inertia is obtained; wherein, represents the bridge model under uncertainty moment of inertia damage condition, represents the bridge model under uncertainty moment of inertia Damage Conditions Displacement response data under .

3. 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 damage identification method according to claim 1 or 2, and the processor is configured to execute the program stored in the memory.

4. 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 damage identification method according to claim 1 or 2 are executed.

Images