Font Size: a A A

The Research On Acoustic Emission Test Of Concrete Bridge Based On Deep Learning Algorithm

Posted on:2023-07-18Degree:MasterType:Thesis
Country:ChinaCandidate:S WangFull Text:PDF
GTID:2532306914455214Subject:Engineering
Abstract/Summary:PDF Full Text Request
This paper explores the possibility of deep learning algorithms to analyze the acoustic emission signals of concrete damage and conducts an experimental study on the localization of the acoustic emission source as well as the damage level identification task,respectively.The aim is to investigate whether this novel artificial intelligence algorithm has the ability to locate the location of concrete cracking and further investigate whether it can predict the actual damage level at the moment of concrete cracking on a macroscopic scale.A large number of lead break tests were conducted on concrete T-beams as well as steel core slabs to train the deep learning algorithm for the localization task and to give the accuracy of its localization;multiple loading tests were conducted on a prestressed concrete cable-stayed bridge model to train the algorithm for the damage extent prediction task.The two intelligent algorithmic frameworks for localization and damage level identification both showed fairly good performance,and the corresponding research conclusions were obtained.(1)Deep learning algorithms based on acoustic emission techniques can locate acoustic emission sources on concrete structures very well regionally and the error mostly stays within 2 cm,and the performance of the method is found to depend on the complexity of the localization task.Although the proposed method does not successfully locate all signals,its misclassified locations are not very far from the real locations,and in practical engineering applications,absolute positioning accuracy is often not required and a certain amount of positioning deviation is allowed,so the error of the method does not affect practical engineering applications too much.(2)Based on deep learning algorithm and acoustic emission technology can effectively identify three damage levels of prestressed concrete cable-stayed bridges.The prediction accuracies of 95.18%,97.99%and 96.45%were achieved for mild,moderate and severe,respectively.The proposed method is 5%-10%more accurate than the traditional BP neural network,convolutional neural network,and is able to train and identify waveform signals with multiple dimensions and does not depend on the extraction of acoustic emission signal feature parameters.(3)The architecture study for deep learning algorithms found that simple convolutional neural networks could instead achieve good performance for acoustic emission source localization and damage level identification tasks.This study finds that deep networks with better fitting performance are not necessarily suitable for prediction and localization of acoustic emission signals,so researchers need to optimize the neural network framework for specific tasks.(4)The size of the requirement of the algorithm for the training set of acoustic emission signals was investigated.Increasing the number of acoustic emission signals in the training set can effectively improve the localization performance of the convolutional neural network,but in practice,the acquisition cost of damaged acoustic emission signals is often high,and this paper finds that more than 1000 signals as the training set quantity is more appropriate.
Keywords/Search Tags:Concrete bridge, source localization, degree of damage, acoustic emission techniques, deep learning algorithms
PDF Full Text Request
Related items