Real-Time Diagnosis of Structural Damage Based on NARX Neural Network with Dynamic Response

In order to improve the applicability of the time series model for structural damage diagnosis, this article proposed a real-time structural damage diagnosis method based on structural dynamic response and a recurrent neural network model. Starting from the transfer rate function of linear structure...

Full description

Saved in:
Bibliographic Details
Published in:Mathematics (Basel) 2023-03, Vol.11 (6), p.1281
Main Authors: Xu, Yanxin, Zheng, Dongjian, Shao, Chenfei, Zheng, Sen, Gu, Hao, Chen, Huixiang
Format: Article
Language:English
Subjects:
Autoregressive models
Crack propagation
Degrees of freedom
Diagnosis
Dynamic response
factor certainty
Fault diagnosis
Marxian distance
Mathematics
Methods
NARX neural network
Neural networks
nonlinear system
Nonlinear systems
Real time
real-time damage diagnosis
Recurrent neural networks
Stiffness
Structural analysis (Engineering)
Structural damage
System effectiveness
Technology application
Time series
Vibration
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In order to improve the applicability of the time series model for structural damage diagnosis, this article proposed a real-time structural damage diagnosis method based on structural dynamic response and a recurrent neural network model. Starting from the transfer rate function of linear structure dynamic response, a generalized Auto-Regressive model with eXtra inputs (ARX) expression for a dynamic response under smooth excitation conditions was derived and extended to the case of nonlinear structure damage using a neural nonlinear ARX (NARX) network model. The method of NARX neural network construction and online parameter learning was studied to solve the definiteness of each factor in the network by applying unit input vectors to the model, and to construct diagnostic indices for structural nonlinear damage based on the Marxian distance (MD). Finally, the effectiveness of NARX damage diagnosis with neural network was verified by numerical arithmetic examples of stiffness loss in four-degree-of-freedom (4-DOF) nonlinear systems. The results showed that the NARX neural network can effectively describe the input-output relationship of the structural system under nonlinear damage. For dynamic neural networks, factor determination based on unit inputs has higher computational accuracy than that of the conventional method. The well-established MD damage index could effectively characterize the devolution of structural nonlinear damage.
ISSN:2227-7390
2227-7390
DOI:10.3390/math11061281