基于计算机视觉的浮放式非结构构件地震损伤识别方法
Seismic damage identification of floating non-structural components based on computer vision techniques
王健泽 1杨雄磊 1江永清 1戴靠山 1徐军 2陈彬3
作者信息
- 1. 四川大学 土木工程系,四川 成都 610065
- 2. 成都设计咨询集团,四川 成都 610095
- 3. 成都市建筑设计研究院有限公司,四川 成都 610095
- 折叠
摘要
非结构构件是实现建筑使用功能的重要组成部分.非结构构件破坏常常会影响建筑震后正常使用功能,严重时造成震后建筑功能长时间中断,是影响建筑抗震韧性的重要因素.基于健康监测与遥感影像的建筑震后性态评估主要针对结构构件,未有关于非结构构件震损识别的相关研究.针对监控视频中的固定式、悬吊式和浮放式非结构构件地震破坏现象的差异,该研究基于YOLOF目标检测算法的分割检测功能和Slowfast视频理解算法的时间序列建模功能,提出了一种适用于浮放式非结构构件的响应与损伤识别方法.为验证该方法的可行性,通过开展小型振动台试验构建浮放式非结构构件地震响应视频数据库,基于数据库对浮放式非结构构件的运动响应影像进行了损伤识别.结果表明:该研究提出的浮放式非结构构件的运动损伤状态识别方法可行,识别准确度达到83.4%.
Abstract
Non-structural components are essential to building functions.Damage to non-structural components(NSCs)often affects the post-earthquake normal functionality of buildings,and in severe cases,can lead to prolonged functional paralysis.As a result,the seismic damage of NSCs is a critical factor affecting the seismic resilience of buildings.The existing approaches for post-earthquake performance assessment of buildings like health monitoring and remote sensing are only suitable for structural components.There has been limited research on non-structural component damage identification.This study proposes a damage identification method specific for freestanding non-structural components.Considering the characteristic differences in seismic response patterns of fixed,suspended,and freestanding NSCs,this study develops a method for seismic damage recognition of freestanding NSCs based on YOLOF object detection algorithm and the Slowfast video understanding model.The proposed method was validated based on a database of videos containing the seismic response and damage of floating NSCs which was collected in a series of scaled shake table tests.The results indicate that the proposed method is effective in identifying the response and damage patterns of freestanding non-structural components with an accuracy rate of 83.4%.
关键词
非结构构件/损伤现象/目标检测/视频理解/损伤识别Key words
non-structural components/damage phenomena/object detection/video understanding/damage recognition引用本文复制引用
基金项目
四川省自然科学基金(2022NSFSC0988)
四川大学博士后交叉学科创新启动基金(JCXK2240)
成都设计咨询集团有限公司重大科研课题(J2023-Z-05)
出版年
2024
NETL
NSTL
万方数据