Open AccessVibration Signal-based Structural Damage Detection through Deep Learning and Digital Image Correlation
Author(s) -
Gongfa Chen,
Gongfa Chen
Publication year - 2021
Publication title -
iop conference series. earth and environmental science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.179
H-Index - 26
eISSN - 1755-1307
pISSN - 1755-1315
DOI - 10.1088/1755-1315/719/2/022047
Subject(s) - digital image correlation , convolutional neural network , vibration , signal (programming language) , computer science , artificial intelligence , structural health monitoring , displacement (psychology) , artificial neural network , stiffness , frame (networking) , deep learning , pattern recognition (psychology) , structural engineering , computer vision , acoustics , engineering , telecommunications , physics , optics , psychology , psychotherapist , programming language
The methods that structural damage detection (SDD) based on vibration signal can effectively detect invisible structural damages. This paper presents a progressive SDD method using a deep learning algorithm and the digital image correlation (DIC) measurement technique. As structural damage will affect the structural mass, damping and stiffness, and then leads to changes in dynamic response, thus the vibration signal may be able to effectively reflect structural defects. A vibration signal database was established from the experimental tests of a steel frame under random excitations, a camera used to record the vibration responses of the structure. The DIC method was employed to obtain the dynamic displacements of the points of interest (POIs). The obtained POI displacement signal was employed as the training and testing data for a convolutional neural network (CNN) which was designed to classify vibration signals. The results confirm that it was feasible to employ a CNN to detect structural damage, the accuracy was nearly 100%, the computational performance and accuracy exceed back-propagation neural networks (BPNN); its uptime was only about 12% that of the BPNN. It has been demonstrated that: (1) the CNN was sensitive to the structural damage detection; (2) the computational performance of the CNN was superior to that of the BPNN.