Ancient bridge damage identification method and system based on improved directed graph convolutional network, medium and computer program

Abstract

The invention relates to an ancient bridge damage identification method and system based on an improved directed graph convolutional network, a medium and a computer program. The damage identification method comprises the following steps: constructing a training data set according to ancient bridge monitoring and detection information; and a kinetic equation considering damage is obtained in combination with possible damage conditions of the ancient bridge, and a target solution set is solved through integration. A damage sensitivity matrix is calculated according to the target solution set, and the sensitivity degree relation between damage and test response is reflected. And correcting the sensitivity matrix according to the sensor type, and taking the corrected sensitivity matrix as a weight factor of the directed graph. Then, carrying out difference on the training data in a time domain so as to expand features of the data in a time domain dimension; and finally, a graph neural network is established for training and is used for ancient bridge damage identification. Compared with the prior art, the method can effectively identify the damage mode of the ancient bridge, has high accuracy, precision, recall rate and F1 score value, and can provide scientific support for preventive protection of bridge cultural relics.

Description

Technical Field

[0001] The present invention relates to the field of civil engineering, and in particular to an ancient bridge damage identification method based on an improved directed graph convolutional network. Background Art

[0002] As witnesses of history and cultural heritage, ancient bridges carry a wealth of historical information and artistic value. However, due to long-term exposure to the elements, such as sun and rain, these ancient bridges suffer significant damage, such as cracking and deformation. Therefore, damage identification can help protect and pass on this precious cultural heritage.

[0003] However, current damage identification based on structural health monitoring has the following main problems: (1) Some damage characteristic tests of ancient bridges, such as natural vibration frequency, may not be sensitive to local or small damage, and more sensitive feature quantities need to be extracted from the dynamic response signal. In fact, many monitoring projects of ancient bridges often only have low-frequency static data. (2) Measured data are greatly affected by environmental noise, and a lot of effort needs to be invested in decoupling test information from environmental factors. Summary of the Invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method, system, medium, and computer program for identifying damage to ancient bridges based on an improved directed graph convolutional network. Based on data from multiple different types of sensors, it can realize bridge damage identification that takes into account environmental impact under conditions of low sampling frequency and long monitoring period.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A first aspect of the present invention provides a method for identifying damage to an ancient bridge based on an improved directed graph convolutional network, comprising the following steps:

[0007] S1: A training dataset is constructed based on the information obtained from continuous monitoring and on-site testing of the ancient bridge. In the training dataset, the input data is collected from different types of detection sensors, and the output data is a binary array label;

[0008] S2: Based on the linear elastic dynamic equation and the possible damage conditions in the ancient bridge structure, the corresponding dynamic equation considering the damage is derived. Then, the target solution set is solved by integrating the dynamic equation.

[0009] S3: Based on the target solution set, further calculating a damage sensitivity matrix, wherein the damage sensitivity matrix is ​​used to reflect the sensitivity relationship between damage and monitoring response;

[0010] S4: Modify the sensitivity matrix based on the actual test content of different sensor types. Take the maximum value of the updated sensitivity matrix in the time dimension. Then, calculate the Pearson matrix coefficient matrix of the sensitivity matrix in the damage dimension. The calculated Pearson matrix coefficient matrix is ​​the weight factor matrix of the directed graph;

[0011] S5: Establish a graph neural network, train the graph neural network, and use the graph neural network to identify damage to ancient bridges.

[0012] Furthermore, in S1, in the training data set, the input of the training set is the monitoring data of different types of sensors within a preset time length, and the output label data is a series of binary arrays with values ​​of 0 or 1, and the dimension of the label array is the number of damage categories found in the detection.

[0013] Furthermore, in S2, the linear elastic dynamics equation is shown in formula (1):

[0014]

[0015] Where M, C, and K represent the mass matrix, damping matrix, and stiffness matrix respectively;

[0016] x is the acceleration, velocity and displacement response array respectively;

[0017] F represents the input stimulus.

[0018] Furthermore, in S2, when there are n e When the finite element of units is used, the parameter reduction matrix is ​​recorded as {γ k}, where k∈[1,n e ], and there is γ k ∈[0,1],γ k represents the damage of unit k. The dynamic equation considering the damage is formula (2):

[0019]

[0020] Among them, M k,0 and K k,0 They represent the mass and stiffness matrices of element k when it is undamaged;

[0021] α and β are Rayleigh damping constants;

[0022] Target solution set of current damage By integrally solving formula (2), we can get: A represents the selection matrix, which represents the degree of freedom of the corresponding sensor layout position.

[0023] Furthermore, in S3, the damage sensitivity matrix is ​​Y with respect to damage γ kThe sensitivity matrix of is calculated by formula (3):

[0024]

[0025] Among them, S l (γ) represents the response Y with degrees of freedom l l , response Y l Total t τ discrete values;

[0026] The central difference method shown in formula (4) is used to obtain:

[0027]

[0028] Furthermore, in S4, the specific process includes:

[0029] Based on the sensitivity matrix, the solution set is divided into Convert into corresponding test content At this time, the updated sensitivity matrix (5) is:

[0030]

[0031] in Indicates finding the maximum value of the array in the time dimension, Ns represents the number of sensor channels;

[0032] Using formula (6) to find the coefficient of the Pearson matrix in the γ dimension of formula (5) is Indicates that the correlation between sensors is determined based on the strength of the correlation of the damage sensitivity matrix of each sensor:

[0033]

[0034] Furthermore, in S5, the graph neural network is a graph neural network improved by using a time-domain difference operator, and the improvement process includes:

[0035] A difference layer is added to the graph neural network, and the data sampled by the sensor is regarded as a graph. The difference layer differs the sensor data on the time axis to expand the features of the graph in the channel dimension. The difference used is shown in formula (7):

[0036]

[0037] in Indicates that sensor q is at t i The test data at the moment, Represents the m-order difference, and fills the empty positions in the array with zeros.

[0038] A second aspect of the present invention provides an ancient bridge damage identification system based on an improved directed graph convolutional network, comprising:

[0039] Data collection and labeling module: This module uses the information obtained from continuous monitoring of the ancient bridge and on-site testing to build a training data set. In this training data set, the input data is collected from different types of detection sensors, and the output data is a binary array label.

[0040] Damage dynamics equation solving module: Based on the linear elastic dynamics equation and the possible damage conditions in the ancient bridge structure, the corresponding damage-considered dynamics equation is derived. The target solution set is then solved by integrating the dynamics equation.

[0041] Damage sensitivity calculation module: Based on the target solution set, further calculate the damage sensitivity matrix, which is used to reflect the sensitivity relationship between damage and other related factors;

[0042] The directed graph matrix solver module modifies the sensitivity matrix based on the actual test content of different sensor types. The updated sensitivity matrix is ​​minimized in the time dimension. The Pearson matrix coefficient matrix is ​​then calculated for the sensitivity matrix in the damage dimension. The calculated Pearson matrix coefficient matrix is ​​the weight factor matrix of the directed graph.

[0043] The ancient bridge damage identification module uses the data set to train the graph neural network and uses the graph neural network to identify the damage of ancient bridges.

[0044] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned ancient bridge damage identification method based on the improved directed graph convolutional network.

[0045] A fourth aspect of the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the above-mentioned ancient bridge damage identification method based on the improved directed graph convolutional network.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1) The present invention proposes an ancient bridge damage identification method based on an improved directed graph convolutional network. The Pearson coefficient matrix of the maximum damage sensitivity in the time domain is used to replace the Pearson coefficient matrix of the sensor raw data, which can effectively eliminate the influence of environmental factors on the construction of the directed graph matrix.

[0048] 2) Using differential layers, we construct high-dimensional features in the sensor's temporal dimension, expanding its temporal characteristics. Using directed graphs and graph convolutional networks, we construct high-dimensional features in the sensor's spatial and mechanical dimensions, enabling the network to fully learn the underlying patterns in the data. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flow chart of the ancient bridge damage identification method based on the improved directed graph convolutional network in the present invention. DETAILED DESCRIPTION

[0050] Overall, the present invention discloses a method for identifying damage to ancient bridges based on an improved directed graph convolutional network. First, an improved directed graph matrix construction method is established to avoid the environmental impact problems faced when constructing the directed graph matrix. Then, a graph convolutional neural network improved by time domain difference is established to identify damage in order to expand the features of the data in the time domain dimension. The method proposed in the present invention can effectively identify the damage pattern of ancient bridges with high accuracy, precision, recall rate and F1 score. The present invention can provide scientific support for the preventive protection of bridge cultural relics.

[0051] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Component models, material names, connection structures, control methods, algorithms, and other features not explicitly described in this technical solution are considered common technical features disclosed in the prior art.

[0052] Example 1

[0053] This embodiment discloses a method for identifying damage to an ancient bridge based on an improved directed graph convolutional network. Figure 1 , including the following steps:

[0054] Step 1: Establish a structural health monitoring system for the historic bridge, conduct long-term monitoring, and conduct monthly on-site damage inspections during the initial phase of the monitoring project. Based on the monitoring and inspection results, construct a training dataset. The training dataset input consists of sensor data of different types over a specific time period. The output label data is a series of binary arrays with values ​​of 0 or 1. The dimension of the label array is the number of damage categories detected.

[0055] Step 2: The linear elastic dynamics equation is shown in formula (1):

[0056]

[0057] Where M, C, and K represent the mass matrix, damping matrix, and stiffness matrix, respectively. x is the acceleration, velocity and displacement response array respectively. F is the input excitation. e The parameter reduction matrix can be written as {γk}, where k∈[1,n e ], and there is γ k ∈[0,1]. γ k represents the damage of unit k. Then the dynamic equation considering the damage can be changed to equation (2).

[0058]

[0059] Among them, M k,0 and K k,0 They represent the mass and stiffness matrices of element k when it is undamaged. α and β are Rayleigh damping constants. At this time, the target solution set for the current damage is It can be solved by the integral of formula (2). Where A represents the selection matrix, which represents the degree of freedom of the corresponding sensor layout position. Step 3: Y is about the damage γ k The sensitivity matrix can be calculated by formula (3).

[0060]

[0061] Among them, S l (γ) represents the response Y with degrees of freedom l l (The response has a total of t τ discrete values), about the damage γ k sensitivity value. It can be obtained by the central difference method shown in formula (4).

[0062]

[0063] Step 4: For the sensitivity matrix obtained by equations (3) and (4), the solution set is divided into Convert into corresponding test content For example, the displacement is eliminated to calculate the height, and the inverse triangle is taken to obtain the inclination. For the updated sensitivity matrix at this time, the maximum value is taken in the time dimension, and equation (5) can be obtained.

[0064]

[0065] in Indicates the maximum value of the array in the time dimension, and Ns represents the number of sensor channels. At this time, the coefficient of the Pearson matrix in the γ dimension of formula (5) is obtained by using formula (6): The correlation between sensors is determined based on the strength of the correlation between the damage sensitivity matrices of each sensor. Compared with directly calculating the Pearson matrix between the raw sensor data, it has two advantages: (1) It is calculated based on finite element theory, which can avoid the confusion of environmental factors on the correlation between signals; (2) The correlation coefficient obtained from the damage sensitivity value can better reflect the correlation behavior between different sensors during the evolution of structural damage.

[0066]

[0067] Step 5: The established graph neural network can be improved using a time-domain difference operator. The sensor data after sampling is considered as a graph. The purpose of the difference layer is to expand the features of the graph in the channel dimension by differentiating the sensor data along the time axis. The difference used is shown in Equation (7).

[0068]

[0069] in represents the test data of sensor q at time ti, represents the mth-order difference. Because differences can cause data misalignment, missing positions are padded with zeros. This converts the original single-channel sensor graph data into m-channel graph data. After completing training on the training set and reaching the maximum number of training epochs, the network with the optimal epoch is tested on the test set to verify network performance.

[0070] This embodiment also provides an ancient bridge damage identification system based on an improved directed graph convolutional network, including:

[0071] Data collection and labeling module: This module uses the information obtained from continuous monitoring of the ancient bridge and on-site testing to build a training data set. In this training data set, the input data is collected from different types of detection sensors, and the output data is a binary array label.

[0072] Damage dynamics equation solving module: Based on the linear elastic dynamics equation and the possible damage conditions in the ancient bridge structure, the corresponding damage-considered dynamics equation is derived. The target solution set is then solved by integrating the dynamics equation.

[0073] Damage sensitivity calculation module: Based on the target solution set, further calculate the damage sensitivity matrix, which is used to reflect the sensitivity relationship between damage and other related factors;

[0074] The directed graph matrix solver module modifies the sensitivity matrix based on the actual test content of different sensor types. The updated sensitivity matrix is ​​minimized in the time dimension. The Pearson matrix coefficient matrix is ​​then calculated for the sensitivity matrix in the damage dimension. The calculated Pearson matrix coefficient matrix is ​​the weight factor matrix of the directed graph.

[0075] The ancient bridge damage identification module uses the data set to train the graph neural network and uses the graph neural network to identify the damage of ancient bridges.

[0076] The functional modules in various embodiments of the present invention may be integrated into a single processing module, or each module may exist physically separately, or two or more modules may be integrated into a single module. The integrated modules may be implemented in the form of hardware or software functional units.

[0077] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned method for identifying damage to an ancient bridge based on an improved directed graph convolutional network. The storage medium can be an electronic medium, a magnetic medium, an optical medium, an electromagnetic medium, an infrared medium, or a semiconductor system or a propagation medium. The storage medium can also include a semiconductor or solid-state memory, a magnetic tape, a removable computer disk, a random access memory (RAM), a read-only memory (ROM), a hard disk, and an optical disk. The optical disk can include a compact disk - read-only memory (CD-ROM), a compact disk - read / write (CD-RW), and a DVD.

[0078] This embodiment also provides a computer program product, including a computer program, which, when executed by a processor, implements the aforementioned method for identifying damage in ancient bridges based on an improved directed graph convolutional network. When executed on a processor, the computer program efficiently implements the aforementioned method for identifying damage in ancient bridges based on an improved directed graph convolutional network. This method begins by collecting continuous monitoring and on-site inspection information from ancient bridges to construct a training dataset. The method then derives and solves a dynamic equation that accounts for damage based on the linear elastic dynamics equation. The damage sensitivity matrix is ​​then calculated, and the resulting directed graph matrix is ​​used to solve the matrix. Finally, the dataset is used to train a graph neural network to accurately identify damage in ancient bridges.

[0079] Verification Example 1

[0080] To evaluate the benefits of the present invention, the following validation example can be used. In this case study, a structural health monitoring system was constructed for an ancient bridge. The system included two dual-channel accelerometers, two dual-channel inclination sensors, two settlement sensors, four displacement sensors, 10 strain sensors, one temperature sensor, and one humidity sensor, totaling 26 sensor channels. During testing, samples were collected once per hour, 24 times per day, for a total of 32 months. To expand the training data, the original data was first linearly interpolated. Then, based on 100 data points, the interpolated data was sampled with a 60% overlap. Each 100 consecutive data points was considered a sample, resulting in a total of 3,488 training data points. The 3,488 training data points were sequentially divided into 10 subsets, and 10-fold cross-validation was performed. Training was performed using a spatiotemporal difference graph convolutional neural network. The gated convolution module consisted of two 2D convolutional layers with 64-channel outputs (the final gated convolution module had 16-channel outputs) and a 1×3 convolution kernel. The output of the Gated Conv module enters the graph convolution layer. After passing through the ReLU activation function, the output is re-entered into the Gated Conv module, which then applies the output to the residual module (1×1 convolution kernel). This process is repeated once. The final output is passed through the Gated Conv module once, then adaptive average pooling is applied, and then passes through two fully connected layers (FC layers, with 256 hidden neurons). The result is sigmoid-encoded and multiple-hot encoded with a threshold of 0.5, converting the output into a binary matrix of 0 or 1. The learning rate is 0.001, the number of training epochs is 200, the total number of differencing orders is 4, and the batch size is 64.

[0081] The calculation results show that on the test set, the completely accurate prediction rate can reach 92.26%, the average precision rate is 93.5%, the average recall rate is 90.4%, the average false alarm rate is 3.0%, and the average F1 score is 0.917, indicating that the obtained model can accurately identify damage on the test set.

[0082] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0083] The above description of the embodiments is intended to facilitate understanding and use of the invention by those skilled in the art. It will be apparent that those skilled in the art can readily make various modifications to these embodiments and apply the general principles described herein to other embodiments without requiring inventive effort. Therefore, the present invention is not limited to the above-described embodiments. Improvements and modifications made by those skilled in the art based on the disclosure of the present invention, without departing from the scope of the present invention, should be within the scope of protection of the present invention.

Claims

1. A method for identifying damage of ancient bridges based on an improved directed graph convolutional network, characterized in that: The following steps are involved: S1: Based on the information obtained from continuous monitoring and on-site testing of the ancient bridge, a training data set is constructed. In the training data set, the input data is data collected from different types of detection sensors, and the output data is a binary array label; S2: Based on the linear elastic dynamics equation and the possible damage conditions in the ancient bridge structure, the corresponding dynamics equation considering the damage is obtained, and then the target solution set is solved by integrating the dynamics equation; S3: on the basis of obtaining the target solution set, further calculating a damage sensitivity matrix, wherein the damage sensitivity matrix is ​​used to reflect the sensitivity relationship between damage and monitoring response; S4: According to the actual test content of different types of sensors, the sensitivity matrix is ​​corrected, and the updated sensitivity matrix is ​​maximized in the time dimension. Then, the Pearson matrix coefficient matrix is ​​calculated for the sensitivity matrix in the damage dimension. The calculated Pearson matrix coefficient matrix is ​​the weight factor matrix of the directed graph. S5: Establish a graph neural network, train the graph neural network, and use the graph neural network to identify the damage of the ancient bridge.

2. According to claim 1, a method for identifying damage of an ancient bridge based on an improved directed graph convolutional network is characterized in that: In S1, in the training data set, the input of the training set is the monitoring data of different types of sensors within a preset time length, and the output label data is a series of binary arrays with values ​​of 0 or 1, and the dimension of the label array is the number of damage categories found in the detection.

3. The ancient bridge damage identification method based on improved directed graph convolutional network according to claim 1 is characterized in that: In S2, the linear elastic dynamics equation is shown in formula (1): Where M, C, and K represent the mass matrix, damping matrix, and stiffness matrix, respectively; x are the acceleration, velocity and displacement response arrays respectively; F represents input stimulus.

4. The ancient bridge damage identification method based on improved directed graph convolutional network according to claim 3 is characterized in that: In S2, when there are ne finite elements, the parameter reduction matrix is ​​denoted as {γ k }, where k∈[1,n e ], and γ k ∈[0,1],γ k represents the damage of unit k, and the dynamic equation considering the damage is formula (2): Among them, M k,0 and K k,0 They represent the mass and stiffness matrices of element k when it is undamaged; α and β are Rayleigh damping constants; Target solution set for current damage By integrally solving formula (2), we can get: A represents the selection matrix, which represents the degree of freedom of the corresponding sensor layout position.

5. The ancient bridge damage identification method based on improved directed graph convolutional network according to claim 4 is characterized in that: In S3, the damage sensitivity matrix is ​​Y with respect to the damage γ k The sensitivity matrix of is calculated by formula (3): Among them, S l (γ) represents the response Y with degree of freedom l l , response Y l Total t τ discrete values; The central difference method shown in formula (4) is used to obtain:

6. The ancient bridge damage identification method based on improved directed graph convolutional network according to claim 5 is characterized in that: In S4, the specific process includes: Based on the sensitivity matrix, the solution set is divided into Convert to corresponding test content At this time, the updated sensitivity matrix (5) is: in It means to find the maximum value of the array in the time dimension, and Ns represents the number of sensor channels; Using formula (6), the coefficient of the Pearson matrix in the γ dimension of formula (5) is: Indicates that the correlation between sensors is determined based on the strength of the correlation of the damage sensitivity matrix of each sensor:

7. The ancient bridge damage identification method based on improved directed graph convolutional network according to claim 5 is characterized in that: In S5, the graph neural network is a graph neural network improved by using a time-domain difference operator, and the improvement process includes: A difference layer is added to the graph neural network, and the data after sensor sampling is regarded as a graph. The difference layer differs the sensor data on the time axis to expand the features of the graph in the channel dimension. The difference used is shown in formula (7): in Indicates that sensor q is at t i The test data at the moment, Represents the m-order difference, filling the empty positions in the array with zeros.

8. An ancient bridge damage identification system based on an improved directed graph convolutional network, comprising: Data collection and labeling module: Based on the information obtained from continuous monitoring of the ancient bridge and on-site testing, a training data set is constructed. In the training data set, the input data is data collected from different types of detection sensors, and the output data is a binary array label; Damage dynamic equation solving module: Based on the linear elastic dynamic equation and the possible damage conditions in the ancient bridge structure, the corresponding dynamic equation considering damage is obtained, and then the target solution set is solved by integrating the dynamic equation; Damage sensitivity calculation module: based on the target solution set, further calculates the damage sensitivity matrix, which is used to reflect the sensitivity relationship between damage and other related factors; The directed graph matrix solving module modifies the sensitivity matrix according to the actual test content of different types of sensors, takes the maximum value of the updated sensitivity matrix in the time dimension, and then calculates the Pearson matrix coefficient matrix of the sensitivity matrix in the damage dimension. The calculated Pearson matrix coefficient matrix is ​​the weight factor matrix of the directed graph; The ancient bridge damage identification module uses the data set to train the graph neural network and uses the graph neural network to identify the damage of the ancient bridge.

9. 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 ancient bridge damage identification method based on the improved directed graph convolutional network described in any one of claims 1 to 7 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the ancient bridge damage identification method based on the improved directed graph convolutional network as described in any one of claims 1 to 7.

Images