Sara Zalaghi1 
, Armin Aziminejad * 2, Hossein Rahami3 , Abdolreza Sarvghad moghadam4 , Mirhamid Hosseini1
, Armin Aziminejad * 2, Hossein Rahami3 , Abdolreza Sarvghad moghadam4 , Mirhamid Hosseini1
1- Department of Civil Engineering, Science and Research Branch, Islamic Azad University
2- Department of Civil Engineering, Science and Research Branch, Islamic Azad University ,armin.aziminejad@gmail.com
3- School of Engineering Science, College of Engineering, University of Tehran
4- International Institute of Earthquake Engineering and Seismology (IIEES), Tehran, Iran
2- Department of Civil Engineering, Science and Research Branch, Islamic Azad University ,
3- School of Engineering Science, College of Engineering, University of Tehran
4- International Institute of Earthquake Engineering and Seismology (IIEES), Tehran, Iran
Abstract: (359 Views)
Civil structures inevitably undergo damage over time due to various reasons such as environmental changes, material aging, load variations, and insufficient maintenance. Monitoring these structures, especially aging ones, is crucial to detect damage early on and implement suitable retrofitting measures, ensuring their continued safe and reliable operation without unexpected failures. Consequently, there has been significant research in this field, focusing on damage detection in both simple and complex structures. Health monitoring of highway bridges is essential for achieving a reliable transportation system. The vibration-based damage detection method uses changes in the vibrational properties of structures to detect damages and ensure a healthy state. In this study, the absolute value of the modal flexibility damage index and the modal strain energy damage index simultaneously are utilized to prevent unsafe decisions.
These absolute values of modal strain energy and flexibility damage indexes are utilized as the bases for training deep neural networks (DNNs). These indexes are applied to provide safe decisions and reliable damage evaluation in steel girder of the highway bridges. The convolution neural network (CNN) is utilized for damage quantification estimation. The CNN is one of the deep learning models that can currently be applied in 2D dominant approaches, such as pattern recognition and speech recognition. In addition, these networks can utilize the 1D time domain and vibrational signal data via the convolutional layer. The initial stage of CNN model comprises combined convolutional and pooling layers that apply different filters to extract features. Following this, fully connected layers, similar to a hidden layer of a multilayer perceptron are incorporated. Ultimately, these layers are classified together with a softmax layer. The convolution layer acts as a filter that convolutes the input layer with a set of weights, adding bias and applying an activation function to the outcome. Gradient descent momentum methods (SGDM) can be employed to optimize the parameters in CNN network architecture. SGDM estimates the gradient with high velocity in any dimension. This method mitigates issues such as jittering and saddle points by utilizing high-velocity inconsistent gradient dimensions and the SGD gradients, respectively. Additionally, when the Current gradient approaches zero, the SGDM provides some momentum.
The convolution neural network is trained to utilize damage indexes obtained from numerical simulation of the validated finite element model of the bridge. The damage indexes as the inputs for the neural network, which are achieved from different damage scenarios. Once network training and validation are completed, a well-trained neural network is used to detect, localize, and quantify the intensity of unknown damages. The proposed method overcomes previous damage detection problems such as false positive indications, the unreliability of a single damage index, and insufficient precision in determining the intensity. The results revealed that the presented method, based on the dual updated damage indexes and CNN, practically and accurately identified unspecified single damages' location and severity in multi-span beams. The new training method of deep neural network systems overcomes some shortcomings in ANN. Moreever, this deep neural network training scheme can reduce the need for huge amounts of input data and enhance the accuracy of network training. The method is capable in predicting single damage scenarios in steel beam.
These absolute values of modal strain energy and flexibility damage indexes are utilized as the bases for training deep neural networks (DNNs). These indexes are applied to provide safe decisions and reliable damage evaluation in steel girder of the highway bridges. The convolution neural network (CNN) is utilized for damage quantification estimation. The CNN is one of the deep learning models that can currently be applied in 2D dominant approaches, such as pattern recognition and speech recognition. In addition, these networks can utilize the 1D time domain and vibrational signal data via the convolutional layer. The initial stage of CNN model comprises combined convolutional and pooling layers that apply different filters to extract features. Following this, fully connected layers, similar to a hidden layer of a multilayer perceptron are incorporated. Ultimately, these layers are classified together with a softmax layer. The convolution layer acts as a filter that convolutes the input layer with a set of weights, adding bias and applying an activation function to the outcome. Gradient descent momentum methods (SGDM) can be employed to optimize the parameters in CNN network architecture. SGDM estimates the gradient with high velocity in any dimension. This method mitigates issues such as jittering and saddle points by utilizing high-velocity inconsistent gradient dimensions and the SGD gradients, respectively. Additionally, when the Current gradient approaches zero, the SGDM provides some momentum.
The convolution neural network is trained to utilize damage indexes obtained from numerical simulation of the validated finite element model of the bridge. The damage indexes as the inputs for the neural network, which are achieved from different damage scenarios. Once network training and validation are completed, a well-trained neural network is used to detect, localize, and quantify the intensity of unknown damages. The proposed method overcomes previous damage detection problems such as false positive indications, the unreliability of a single damage index, and insufficient precision in determining the intensity. The results revealed that the presented method, based on the dual updated damage indexes and CNN, practically and accurately identified unspecified single damages' location and severity in multi-span beams. The new training method of deep neural network systems overcomes some shortcomings in ANN. Moreever, this deep neural network training scheme can reduce the need for huge amounts of input data and enhance the accuracy of network training. The method is capable in predicting single damage scenarios in steel beam.
Keywords: Structural health monitoring, modal strain energy damage index, modal flexibility damage index, steel beam, deep learning.
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |