Bridge damage positioning method and system based on acoustic emission signal and graph neural network

Abstract

The invention relates to the field of bridge damage positioning, in particular to a bridge damage positioning method and system based on acoustic emission signals and a graph neural network, and the method comprises the steps: constructing a bridge damage positioning data set, and training the graph neural network through the data set; acquiring deviations between measured values and true values of the acoustic emission sensors under different historical temperatures and humidity to obtain error distribution of the acoustic emission sensors under different temperatures and humidity; the method comprises the following steps: acquiring a measurement value of a real-time bridge acoustic emission sensor and temperature and humidity of an environment where the sensor is located, obtaining error distribution of the sensor according to the temperature and the humidity, obtaining true value distribution of the sensor according to the error distribution of the sensor and the measurement value, and inputting the true value into a graph neural network to obtain a detection result of bridge damage. And weighting according to the distribution of the true values and the detection result to obtain a final detection result. The method has the effect of improving the bridge damage positioning accuracy.

Description

Technical Field

[0001] This application relates to the field of bridge damage localization, and in particular to a bridge damage localization method and system based on acoustic emission signals and graph neural networks. Background Technology

[0002] In recent years, with a large number of bridges in my country entering their aging stage and the increasing traffic load, bridge structural health monitoring has become a critical issue that urgently needs to be addressed in the field of civil engineering. Developing advanced, intelligent, and precise damage identification and location technologies to achieve a shift from "passive maintenance" to "active early warning" is of great practical significance for ensuring the long-term safety of infrastructure. Traditional bridge inspection methods mainly rely on manual inspections, which suffer from low efficiency, strong subjectivity, and difficulty in detecting internal damage and early micro-damage. Among non-destructive testing technologies, acoustic emission technology, as a dynamic and active monitoring method, exhibits unique advantages: when stress waves are generated inside the material due to damage (such as crack propagation or fiber breakage), acoustic emission sensors can capture these "sound signals" in real time, thereby achieving online and dynamic monitoring of damage activity, and are particularly sensitive to sudden damage.

[0003] However, the classic time-of-flight positioning method relies heavily on the propagation speed of sound waves in materials. This speed is not constant in anisotropic concrete, composite materials, and complex steel structures. It is greatly affected by material inhomogeneity, internal defects, and ambient temperature and humidity, leading to positioning errors. Summary of the Invention

[0004] To address the challenges of temperature and humidity in bridge damage detection, this application provides a bridge damage localization method and system based on acoustic emission signals and graph neural networks.

[0005] Firstly, this application provides a bridge damage localization method based on acoustic emission signals and graph neural networks, employing the following technical solution: A bridge damage localization method based on acoustic emission signals and graph neural networks includes constructing a bridge damage localization dataset. A sample in the dataset consists of the acoustic signals collected once by all acoustic emission sensors and the corresponding damage location. A graph neural network is constructed, where nodes represent acoustic emission sensors, and the attributes of the nodes represent the characteristics of the acoustic emission sensors. By obtaining the deviation between the measured values ​​and the true values ​​of the acoustic emission sensor under different historical temperatures and humidity levels, the error distribution of the acoustic emission sensor under different temperatures and humidity levels can be obtained. The system acquires real-time measurements from bridge acoustic emission sensors, as well as the temperature and humidity of the sensor environment. Based on the temperature and humidity, it obtains the sensor error distribution. Based on the sensor error distribution and the measured values, it obtains the sensor's true value distribution. Based on the true value distribution of each sensor, it calculates the overall sensor true value distribution. The overall sensor true values ​​are then input into a trained graph neural network to obtain the bridge damage detection results. Finally, based on the overall sensor true value distribution and the corresponding detection results, the system obtains the damage location.

[0006] Optionally, in the graph neural network, the attributes of a node are the amplitude, energy, arrival time, and duration of the node's sound signal. If two nodes are on the same bridge structure, their connection weight is 1, and otherwise it is 0.

[0007] Optionally, in the graph neural network, the attribute of a node is the original signal collected by the sensor within a preset time period, and the inverse of the distance between nodes is normalized as the connection weight between two nodes.

[0008] Optionally, the error distribution of the sensor is the probability of different errors occurring under the target temperature and humidity.

[0009] Optionally, the error distribution of the sensor is calculated as follows: divide the temperature and humidity into optimal intervals, and calculate the probability of different errors occurring in each interval.

[0010] Optionally, the calculation process of the optimal interval is as follows: construct an interval division evaluation index, which is directly proportional to the amount of data in each interval and inversely proportional to the dispersion of sensor error in each interval.

[0011] Optionally, the bridge damage detection results include whether the bridge is damaged and the location of the damage.

[0012] Optionally, the final damage location calculation is as follows: First, the detection results of whether the bridge is damaged are weighted according to the distribution of the overall sensor true values. If the weighted result indicates that the bridge is damaged, the location of the bridge damage is weighted, and the system outputs the weighted damage location. If the weighted result indicates that the bridge is not damaged, the system outputs that the bridge is not damaged.

[0013] Secondly, this application provides a bridge damage localization system based on acoustic emission signals and graph neural networks, employing the following technical solution: A bridge damage localization system based on acoustic emission signals and graph neural networks includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a vehicle-charging pile data communication transmission method according to the above-mentioned AC charging pile.

[0014] The beneficial effect is that the above-mentioned bridge damage localization method based on acoustic emission signals and graph neural networks is generated into a computer program and stored in the memory so that it can be loaded and executed by the processor. Thus, the system can be made according to the memory and processor, which is convenient to use.

[0015] This application has the following technical effects: 1. By calculating the error distribution of acoustic emission sensors under different historical temperatures and humidity levels, when predicting bridge damage and damage location using a graph neural network, the error of the acoustic emission sensors in real time is considered to obtain the distribution of the real values ​​of the acoustic emission sensors in real time, thereby obtaining the distribution of the real values ​​of the overall acoustic emission sensors. The damage status and damage location of the bridge are obtained by weighting the detection results input into the graph neural network based on the characteristics of the overall acoustic emission sensors and the distribution of the real values ​​of the overall acoustic emission sensors, thus improving the accuracy of damage detection and damage location detection.

[0016] 2. When calculating the error distribution of acoustic emission sensors under different temperatures and humidity, the temperature and humidity are divided into intervals, and evaluation indicators are constructed to obtain the optimal interval division. The evaluation indicators take into account the amount of data within the interval and the concentration of sensor error. The optimal division avoids the influence of randomness on the error distribution, and at the same time makes the sensor error distribution within the interval more concentrated, providing more reliable and accurate data for subsequent bridge damage detection. Attached Figure Description

[0017] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. Several embodiments of this application are illustrated in the drawings by way of example and not limitation, and the same or corresponding reference numerals denote the same or corresponding parts.

[0018] Figure 1 This is a flowchart illustrating steps S1-S3 of the bridge damage localization method based on acoustic emission signals and graph neural networks in this application.

[0019] Figure 2 This is a structural block diagram of the bridge damage localization system based on acoustic emission signals and graph neural networks in this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] It should be understood that when the terms "first," "second," etc., are used in the claims, description, and drawings of this application, they are only used to distinguish different objects and not to describe a specific order. The terms "comprising" and "including" used in the description and claims of this application indicate the presence of the described features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.

[0022] This application discloses a bridge damage localization method based on acoustic emission signals and graph neural networks, referring to... Figure 1 This includes the following steps: S1: Construct a bridge damage localization dataset. A single sample in the dataset consists of the sound signals collected by all acoustic emission sensors in one go and the corresponding damage location. Construct a graph neural network, where nodes represent acoustic emission sensors, and the attributes of each node represent the characteristics of the acoustic emission sensor.

[0023] In one embodiment, an array of acoustic emission sensors is deployed on the bridge to collect sound signals from the bridge. Those skilled in the art determine the location of the damage by combining the actual damage to the bridge. The sound signals collected by all sensors at one time and the corresponding damage locations form a sample, and the samples collected multiple times form a bridge damage location dataset.

[0024] Construct a graph neural network where nodes represent acoustic emission sensors. The features of each node include the amplitude, energy, arrival time, and duration of a single acquired sound signal. The adjacency matrix is ​​set as follows: if two sensors are directly connected on the bridge structure (e.g., on the same bridge deck or the same beam), an edge is established between them. A[i, j] = 1 indicates that sensors i and j are connected, and 0 indicates they are not connected.

[0025] In another embodiment, in the graph neural network, nodes represent acoustic emission sensors, and the characteristics of a node are the raw signals collected by the sensor within a preset time period. The adjacency matrix is ​​defined based on the Euclidean distance between the sensors. Specifically, the connection weight between any two sensors is the normalized result of the inverse of the Euclidean distance between the two sensors. Using the Euclidean distance between the sensors as the connection weight takes into account the attenuation characteristics of sound waves with distance.

[0026] A graph neural network (Graph Neural Network) is trained using a bridge damage localization dataset. The input to the Graph Neural Network consists of the features of each node, and the output indicates whether the bridge is damaged and the location of the damage. When the bridge is not damaged, there is no damage location; the damage location output by the model is set to 0 when there is no damage. The loss function of the Graph Neural Network is the sum of cross-entropy loss and mean squared error loss. Cross-entropy loss helps the model learn whether the bridge is damaged, while mean squared error loss helps the model learn the specific location of the damage. Since learning whether the bridge is damaged is more important than knowing where the damage is, in one embodiment, the weight of cross-entropy loss is 0.7, and the weight of mean squared error loss is 0.3. Gradient descent is used to update the model parameters. The model stops training when it reaches the maximum number of training iterations or when the model's loss is less than a set threshold. An exemplary maximum number of training iterations is 1000, and the model's loss threshold is 0.001.

[0027] S2: Obtain the deviation between the measured value and the true value of the acoustic emission sensor under different historical temperatures and humidity levels, and obtain the error distribution of the acoustic emission sensor under different temperatures and humidity levels.

[0028] In one embodiment, since bridges are often located in water, the humidity near bridges is high. Prolonged operation in high humidity can affect the accuracy of the sensor. To ensure more accurate bridge damage detection results, this application considers sensor errors under different environmental conditions. Specifically, it collects historical data on sensor operating temperature and humidity, as well as the sensor's measured and actual values. The difference between the measured and actual values ​​is the sensor error. The probability of different errors occurring under different temperatures and humidity levels is calculated to obtain the sensor's error distribution. The probability of different errors occurring is calculated as follows: the number of times the target error occurs and the number of times historical data is collected at the target temperature and humidity are statistically analyzed. The ratio of the number of times the target error occurs to the number of times historical data is collected at the target temperature and humidity is the probability of the target error occurring.

[0029] In another embodiment, temperature and humidity are divided into intervals, and the distribution of sensor errors within each interval is calculated to construct an evaluation index for the temperature and humidity intervals. The reason for dividing temperature and humidity into intervals is that when the amount of historical data is small, the calculation of the sensor error distribution at each temperature and humidity level will have a certain degree of randomness, which will lead to inaccurate final bridge detection results. Therefore, temperature and humidity are divided into intervals, and the distribution of sensor errors within each interval is calculated. To make the sensor error distribution more beneficial for subsequent detection, the evaluation index for the interval division is constructed from two aspects: First, the amount of data in each interval after division should be reasonable, because randomness will lead to inaccurate results when the amount of data is small; second, the error distribution within each interval after division should be close. The closer the error distribution, the more accurate the final judgment result of bridge damage. Specifically, the calculation formula for the evaluation index is as follows:

[0030] in, Evaluation indicators representing the division of temperature and humidity ranges. This represents the amount of data in the i-th interval of temperature and humidity. Represents the maximum value function. The base of the exponential function is... , This represents the variance of the sensor error within the i-th interval of temperature and humidity. The larger the variance of the sensor error within the divided interval, the more discrete the error distribution within the interval, resulting in less accurate judgment of bridge damage based on this error distribution, and a worse division result. Conversely, the larger the amount of data within the divided interval, the less the sensor error distribution within the divided interval is affected by randomness, and a better division result.

[0031] The optimal interval division for temperature and humidity is obtained using an intelligent optimization algorithm. The error distribution of the sensor in each interval is calculated based on the optimal interval division result. The intelligent optimization algorithm is either a particle swarm optimization algorithm or a genetic algorithm.

[0032] S3: Obtain the real-time measurement values ​​of the bridge acoustic emission sensors, the temperature and humidity of the sensor environment, obtain the sensor error distribution based on the temperature and humidity, obtain the sensor true value distribution based on the sensor error distribution and measurement values, calculate the overall sensor true value distribution based on the true value distribution of each sensor, input the overall sensor true value into the trained graph neural network to obtain the bridge damage detection results, and obtain the final damage location based on the overall sensor true value distribution and the corresponding detection results.

[0033] In one embodiment, the measured values ​​of the real-time bridge acoustic emission sensor, the temperature and humidity of the sensor's environment, are acquired. The error distribution of the sensor under the real-time temperature and humidity is calculated. Specifically, the distance between the real-time temperature and humidity and the midpoint of each interval under the optimal interval division is calculated. The sensor error distribution corresponding to the interval with the smallest distance is taken as the real-time sensor error distribution. The distribution of the sensor's true value is obtained based on the sensor error distribution and the sensor's measured value. For example, when the sensor error distribution is: the probability of an error of 0.1 is 0.3, the probability of an error of 0 is 0.4, and the probability of an error of 0.2 is 0.3, and the sensor's measured value is 1.0, then the distribution of the sensor's true value is: the probability of a true value of 1.1 is 0.3, the probability of a true value of 1.0 is 0.4, and the probability of a true value of 1.2 is 0.3. The distribution of the true values ​​of all sensors is obtained through the above operations for all sensors.

[0034] Based on the distribution of the true values ​​of the bridge sensors, the overall distribution of the true values ​​of the bridge sensors is calculated, and the probability of the overall true values ​​of the bridge sensors is calculated as follows:

[0035] in, This represents the probability that the overall combination of true values ​​from the bridge sensors is A. This indicates that the true value of the j-th sensor is in combination A. This represents the probability of the true value of the j-th sensor appearing in combination A.

[0036] The probability of different combinations of overall sensor true values ​​is calculated to obtain the distribution of overall bridge sensor true values. These combinations are then input into a trained graph neural network to output whether the bridge is damaged and the location of any damage. The final detection result is obtained by weighted summation of the output results based on the probability of each input combination. The formula for calculating whether the bridge is damaged is as follows:

[0037] in, This indicates whether the bridge was ultimately damaged; a value of 1 indicates damage, and a value of 0 indicates no damage. This represents the probability that the overall combination of true values ​​from the bridge sensors is A. This represents the output of the neural network based on input graph A, indicating bridge damage. A value of 1 indicates bridge damage, while a value of 0 indicates no damage. For a vector function, when When true, the function value is 1; when true... If the condition is not met, the function value is 0.

[0038] When the probability of damage occurring is greater than the probability of no damage occurring, the bridge is considered damaged. The final damage location is calculated based on the probability of the input combination occurring and the corresponding damage location in the output. The specific calculation formula is as follows:

[0039] in, Indicates the final location of the damage. The output is the probability of the B-th true value combination of bridge damage occurring. This indicates that the output result is the location of the bridge damage, which is the result of the Bth combination of the true values ​​of the bridge damage.

[0040] This application also discloses a bridge damage localization system based on acoustic emission signals and graph neural networks, referring to... Figure 2It includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the bridge damage localization method based on acoustic emission signals and graph neural networks according to this application.

[0041] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0042] In this application, the aforementioned memory can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store required information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device.

[0043] While this specification has shown and described numerous embodiments of the present application, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will arise for those skilled in the art without departing from the spirit and intent of the present application. It should be understood that various alternatives to the embodiments of the present application described herein may be employed in the practice of this application.

[0044] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A bridge damage localization method and system based on acoustic emission signals and graph neural networks, characterized in that, A bridge damage localization dataset is constructed, where a single sample in the dataset consists of the sound signals collected by all acoustic emission sensors in a single instance and the corresponding damage location. A graph neural network is constructed, where nodes represent acoustic emission sensors, and the attributes of the nodes represent the features of the acoustic emission sensors. By obtaining the deviation between the measured values ​​and the true values ​​of the acoustic emission sensor under different historical temperatures and humidity levels, the error distribution of the acoustic emission sensor under different temperatures and humidity levels can be obtained. The system acquires real-time measurements from bridge acoustic emission sensors, as well as the temperature and humidity of the sensor environment. Based on the temperature and humidity, it obtains the sensor error distribution. Based on the sensor error distribution and the measured values, it obtains the sensor's true value distribution. Based on the true value distribution of each sensor, it calculates the overall sensor true value distribution. The overall sensor true values ​​are then input into a trained graph neural network to obtain the bridge damage detection results. Finally, based on the overall sensor true value distribution and the corresponding detection results, the system obtains the damage location.

2. The bridge damage localization method based on acoustic emission signals and graph neural networks according to claim 1, characterized in that, In the graph neural network, the attributes of a node are the amplitude, energy, arrival time, and duration of the node's sound signal. If two nodes are on the same bridge structure, their connection weight is 1, and otherwise it is 0.

3. The bridge damage localization method based on acoustic emission signals and graph neural networks according to claim 1, characterized in that, In the graph neural network, the attribute of a node is the original signal collected by the sensor within a preset time period, and the inverse of the distance between nodes is normalized as the connection weight between two nodes.

4. The bridge damage localization method based on acoustic emission signals and graph neural networks according to claim 1, characterized in that, The error distribution of the sensor represents the probability of different errors occurring under target temperature and humidity conditions.

5. The bridge damage localization method based on acoustic emission signals and graph neural networks according to claim 1, characterized in that, The error distribution of the sensor is calculated as follows: the temperature and humidity are divided into optimal intervals, and the probability of different errors of the sensor occurring in each interval is calculated.

6. The bridge damage localization method based on acoustic emission signals and graph neural networks according to claim 5, characterized in that, The calculation process of the optimal interval is as follows: construct an evaluation index for interval division. The evaluation index is directly proportional to the amount of data in each interval and inversely proportional to the dispersion of sensor error in each interval.

7. The bridge damage localization method based on acoustic emission signals and graph neural networks according to claim 1, characterized in that, The bridge damage detection results include whether the bridge is damaged and the location of the damage.

8. The bridge damage localization method based on acoustic emission signals and graph neural networks according to claim 1, characterized in that, The final damage location calculation is as follows: First, the detection results of whether the bridge is damaged are weighted according to the distribution of the overall sensor true values. If the weighted result indicates that the bridge is damaged, the location of the bridge damage is weighted, and the system outputs the weighted damage location. If the weighted result indicates that the bridge is not damaged, the system outputs that the bridge is not damaged.

9. A bridge damage localization system based on acoustic emission signals and graph neural networks, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the bridge damage localization method based on acoustic emission signals and graph neural networks according to any one of claims 1-8.

Images