Acoustic emission localization model of concrete material based on convolutional neural network
ObjectiveHydraulic concrete will produce acoustic emission phenomenon due to cracking and other damages. It is important to quickly and accurately locate the damage source according to the acoustic emission signals for real-time monitoring of the health status of hydraulic buildings. The traditional...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Editorial Department of Journal of Sichuan University (Engineering Science Edition)
2024-01-01
|
| Series: | 工程科学与技术 |
| Subjects: | |
| Online Access: | http://jsuese.scu.edu.cn/thesisDetails#10.12454/j.jsuese.202400554 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850037840305455104 |
|---|---|
| author | DENG Yongdong ZHOU Jingren LU Xiang CHEN Jiangkang |
| author_facet | DENG Yongdong ZHOU Jingren LU Xiang CHEN Jiangkang |
| author_sort | DENG Yongdong |
| collection | DOAJ |
| description | ObjectiveHydraulic concrete will produce acoustic emission phenomenon due to cracking and other damages. It is important to quickly and accurately locate the damage source according to the acoustic emission signals for real-time monitoring of the health status of hydraulic buildings. The traditional iterative localization method is greatly affected by the initial value of iteration and the iteration method. Inappropriate initialvalue of iteration may lead to unstable or divergent iteration processes, ultimately resulting in poor localization accuracy. Therefore, the selection of the initial value of iteration is particularly important for the traditional iterative localization method. Moreover, the traditional iterative localization methods are greatly affected by the number of sensors and environmental noise, resulting in unstable localization efficiency. The rapid development of deep learning in recent years has provided new ideas for acoustic emission localization. Deep learning has a strong feature extraction capability and generalization ability. In response to the problems of traditional iterative localization methods, a convolutional neural network-based acoustic emission localization model is constructed, which can improve the efficiency stability and accuracy of acoustic emission localization to a certain extent.MethodsThe absolute propagation time of acoustic emission is generally difficult to obtain, but the arrival time difference between the sensors contains enough information that can be used to localize the acoustic emission source position. In this paper, a convolutional neural network-based acoustic emission localization model is constructed by taking the cylindrical concrete specimen as the experimental object, taking the 3D coordinates of eight sensors and the propagation time difference as input, and the 3D coordinate position of the acoustic emission source as output. In addition, the effect of the number of convolutional layers and the size of convolutional kernel on the localization accuracy is analyzed, and the optimal convolutional neural network structure is obtained. At the same time, in order to verify the localization performance of the constructed localization model, it is compared with the traditional iterative localization method, and the advantages of the constructed localization model in terms of localization accuracy and efficiency are analyzed.Results and Discussions For the case of 8 sensors, the optimal number of convolutional layers is 4, and the optimal convolutional kernel size is 3×1. In the X, Y and Z directions, the root mean square errors (<italic>R</italic><sub>MSE</sub>) of the localization model are 0.865 2, 0.826 6 and 0.722 1 respectively, and the mean absolute errors (<italic>M</italic><sub>AE</sub>) are 0.532 2, 0.617 3 and 0.473 3 respectively, and the mean absolute percentage errors (<italic>M</italic><sub>APE</sub>) are 0.222 0, 0.510 1 and 0.051 0 respectively, and the coefficients of determination (<italic>R</italic><sup>2</sup>) are 0.994 2, 0.993 8 and 0.999 3 respectively, which are close to 1. The vast majority of the localization errors are distributed near 0, which basically conforms to the standard normal distribution. The localization accuracy of the localization model in the depth direction is higher than that in the horizontal direction. Compared with the traditional iterative localization method, the localization efficiency is stable, and it has obvious advantages in processing a large number of localization tasks, and the localization error is reduced by about 5%. The damage location detected by the localization model basically matches the real crack location.ConclusionsThe proposed localization model shows good localization efficiency and localization accuracy. Compared with the traditional iterative localization method, it is not affected by the initial value of iteration and the iterative method, and has the advantages of stable efficiency and high efficiency. Meanwhile, it will have stable performance for the new data points and strong applicability, which can be used as a reference for early warning of damage evolution based on nondestructive testing and is expected to be applied to the damage detection of other materials in the future. |
| format | Article |
| id | doaj-art-a5ff59c7356740c0a2ac2fe3240fdc49 |
| institution | DOAJ |
| issn | 2096-3246 |
| language | English |
| publishDate | 2024-01-01 |
| publisher | Editorial Department of Journal of Sichuan University (Engineering Science Edition) |
| record_format | Article |
| series | 工程科学与技术 |
| spelling | doaj-art-a5ff59c7356740c0a2ac2fe3240fdc492025-08-20T02:56:45ZengEditorial Department of Journal of Sichuan University (Engineering Science Edition)工程科学与技术2096-32462024-01-0111176207809Acoustic emission localization model of concrete material based on convolutional neural networkDENG YongdongZHOU JingrenLU XiangCHEN JiangkangObjectiveHydraulic concrete will produce acoustic emission phenomenon due to cracking and other damages. It is important to quickly and accurately locate the damage source according to the acoustic emission signals for real-time monitoring of the health status of hydraulic buildings. The traditional iterative localization method is greatly affected by the initial value of iteration and the iteration method. Inappropriate initialvalue of iteration may lead to unstable or divergent iteration processes, ultimately resulting in poor localization accuracy. Therefore, the selection of the initial value of iteration is particularly important for the traditional iterative localization method. Moreover, the traditional iterative localization methods are greatly affected by the number of sensors and environmental noise, resulting in unstable localization efficiency. The rapid development of deep learning in recent years has provided new ideas for acoustic emission localization. Deep learning has a strong feature extraction capability and generalization ability. In response to the problems of traditional iterative localization methods, a convolutional neural network-based acoustic emission localization model is constructed, which can improve the efficiency stability and accuracy of acoustic emission localization to a certain extent.MethodsThe absolute propagation time of acoustic emission is generally difficult to obtain, but the arrival time difference between the sensors contains enough information that can be used to localize the acoustic emission source position. In this paper, a convolutional neural network-based acoustic emission localization model is constructed by taking the cylindrical concrete specimen as the experimental object, taking the 3D coordinates of eight sensors and the propagation time difference as input, and the 3D coordinate position of the acoustic emission source as output. In addition, the effect of the number of convolutional layers and the size of convolutional kernel on the localization accuracy is analyzed, and the optimal convolutional neural network structure is obtained. At the same time, in order to verify the localization performance of the constructed localization model, it is compared with the traditional iterative localization method, and the advantages of the constructed localization model in terms of localization accuracy and efficiency are analyzed.Results and Discussions For the case of 8 sensors, the optimal number of convolutional layers is 4, and the optimal convolutional kernel size is 3×1. In the X, Y and Z directions, the root mean square errors (<italic>R</italic><sub>MSE</sub>) of the localization model are 0.865 2, 0.826 6 and 0.722 1 respectively, and the mean absolute errors (<italic>M</italic><sub>AE</sub>) are 0.532 2, 0.617 3 and 0.473 3 respectively, and the mean absolute percentage errors (<italic>M</italic><sub>APE</sub>) are 0.222 0, 0.510 1 and 0.051 0 respectively, and the coefficients of determination (<italic>R</italic><sup>2</sup>) are 0.994 2, 0.993 8 and 0.999 3 respectively, which are close to 1. The vast majority of the localization errors are distributed near 0, which basically conforms to the standard normal distribution. The localization accuracy of the localization model in the depth direction is higher than that in the horizontal direction. Compared with the traditional iterative localization method, the localization efficiency is stable, and it has obvious advantages in processing a large number of localization tasks, and the localization error is reduced by about 5%. The damage location detected by the localization model basically matches the real crack location.ConclusionsThe proposed localization model shows good localization efficiency and localization accuracy. Compared with the traditional iterative localization method, it is not affected by the initial value of iteration and the iterative method, and has the advantages of stable efficiency and high efficiency. Meanwhile, it will have stable performance for the new data points and strong applicability, which can be used as a reference for early warning of damage evolution based on nondestructive testing and is expected to be applied to the damage detection of other materials in the future.http://jsuese.scu.edu.cn/thesisDetails#10.12454/j.jsuese.202400554acoustic emissionlocalization methodconvolutional neural networkdamage detection |
| spellingShingle | DENG Yongdong ZHOU Jingren LU Xiang CHEN Jiangkang Acoustic emission localization model of concrete material based on convolutional neural network 工程科学与技术 acoustic emission localization method convolutional neural network damage detection |
| title | Acoustic emission localization model of concrete material based on convolutional neural network |
| title_full | Acoustic emission localization model of concrete material based on convolutional neural network |
| title_fullStr | Acoustic emission localization model of concrete material based on convolutional neural network |
| title_full_unstemmed | Acoustic emission localization model of concrete material based on convolutional neural network |
| title_short | Acoustic emission localization model of concrete material based on convolutional neural network |
| title_sort | acoustic emission localization model of concrete material based on convolutional neural network |
| topic | acoustic emission localization method convolutional neural network damage detection |
| url | http://jsuese.scu.edu.cn/thesisDetails#10.12454/j.jsuese.202400554 |
| work_keys_str_mv | AT dengyongdong acousticemissionlocalizationmodelofconcretematerialbasedonconvolutionalneuralnetwork AT zhoujingren acousticemissionlocalizationmodelofconcretematerialbasedonconvolutionalneuralnetwork AT luxiang acousticemissionlocalizationmodelofconcretematerialbasedonconvolutionalneuralnetwork AT chenjiangkang acousticemissionlocalizationmodelofconcretematerialbasedonconvolutionalneuralnetwork |