Open AccessDeep neural network for detecting earthquake damage to brace members installed in a steel frame
Author(s) -
Yamashita Takuzo,
Kohiyama Masayuki,
Watanabe Kenta
Publication year - 2021
Publication title -
japan architectural review
Language(s) - English
Resource type - Journals
ISSN - 2475-8876
DOI - 10.1002/2475-8876.12188
Subject(s) - earthquake shaking table , artificial neural network , classifier (uml) , brace , structural engineering , artificial intelligence , computer science , frame (networking) , deep learning , structural health monitoring , engineering , machine learning , telecommunications
We are developing an artificial intelligence system for structural health monitoring that can detect local damage in a building structure by using the E‐Simulator numerical simulation system that is being developed by the Japanese National Research Institute for Earth Science and Disaster Resilience. In this study, we confirmed the applicability of a multiclass classifier using a deep neural network to address the problem of identifying damage patterns in braces installed in a steel frame. Experimental data obtained from shaking table tests were used for training and testing. Cross‐validation tests were conducted for several cases with different numbers of sensors, sensor degrees of freedom, and nodes in the hidden layers of the network. The results demonstrated that the accuracy of the damage pattern detection from the constructed classifier exceeded 77% when the appropriate hidden layers were selected and reached 87.9% for the best case.