基于层合理论和误差反向传播神经网络的古建筑木结构损伤识别和量化研究  

STUDY ON DAMAGE DETECTION AND QUANTIFICATION OF ANCIENT BUILDING TIMBER STRUCTURES BASED ON LAMINATION THEORY AND BP NEURAL NETWORKS

作　　者：胡卫兵[1] 杨佳 王龙 侯艳芳 HU Weibing;YANG Jia;WANG Long;HOU Yanfang(Department of Civil Engineering,Xi’an University of Architecture and Technology,Xi’an 710055,China;China Jikan Research Institute of Engineering Investigation and Design Company,Xi’an 710043,China)

机构地区：[1]西安建筑科技大学土木工程学院,西安710055 [2]机械工业勘察设计研究院有限公司,西安710043

出　　处：《工业建筑》2020年第11期71-77,111,共8页Industrial Construction

基　　金：国家自然科学基金项目(51678475);陕西省自然科学基础研究计划重点项目(2018JZ5002)。

摘　　要：为解决环境激励下古建筑木结构的损伤量化问题,根据层合理论提出用有限元分层实体单元将木材分为损伤表层和中间无损伤层的方法。引入损伤深度作为新参数,以小波总能量变化率为损伤识别指标,共同作为特征参数输入反向传播(BP)神经网络,以弹性模量折减率为输出参数对损伤进行多维量化。分别将该方法应用于一榀木结构框架和西安钟楼模型,识别结果最大误差不超过2%,表明通过分层实体单元模拟木材损伤并引入损伤深度为参数量化木梁损伤是可行的,能够为环境激励下古建筑木结构梁损伤预警研究和复杂结构局部损伤分析提供参考理论。To quantify damage of ancient timber structures under ambient excitation,according to the lamination theory,a method was proposed to divide wood into damaged surface layers and undamaged middle layers by layered solid elements of the finite element method.The damage depth was taken as a new variable,combined with the total energy-change rates of wavelets as feature parameters and input into an improved BP neural network,and the reduction ratios of elastic moduli were considered as output parameters.Appling the method respectively to multi-scale models of a timber frame and Xi’an Bell Tower,the result showed that the maximum error of identification results was less than 2%,which meant it feasible to simulate wood damage by layered solid elements and adopt the damage depth as a parameter to quantify the damage of wood beams,which could provide reference for the study of damage early warning and local damage analysis of complex structures under ambient excitation。

关 键 词：古建筑木结构 损伤量化 有限元 层合结构 环境激励 

分 类 号：TU366.2[建筑科学—结构工程] TU-87[艺术—艺术设计]