Fundamental Study on Damage Classification Method for a Wooden Building Based on Hysteresis Loop Using Convolutional Neural Network

Abstract

Structural health monitoring, which evaluates building damage based on seismic response data, helps to quickly assess earthquake damage. When the target building is damaged by strong seismic motion, the curve shape of the restoring force–story drift angle relationship (hysteresis loop) changes. Therefore, this study proposes a method to determine the risk of collapse due to future seismic motion based on the image recognition of hysteresis loop using a convolutional neural network, targeting a wooden building. In the proposed method, two types of axis range settings for creating hysteresis loops are considered, and the accuracy is verified using test data generated by response analysis. The method in which the axis range is set based on the entire training data and fixed in all the hysteresis loop images showed 90-100% recall of the unsafe model and is confirmed to rarely misjudge the unsafe model.