基于向量自回归模型的结构损伤预警  

Structural damage identification based on vector auto-regressive model

作　　者：郑泓 段忠东 曾帝棋 ZHENG Hong;DUAN Zhongdong;ZENG Diqi(Civil and Environmental Engineering,Harbin Institute of Technology(Shenzhen),Shenzhen 518055,China)

机构地区：[1]哈尔滨工业大学(深圳)土木与环境工程学院,广东深圳518055

出　　处：《地震工程与工程振动》2021年第3期136-146,共11页Earthquake Engineering and Engineering Dynamics

基　　金：国家重点研发计划(2018YFC0705604);国家自然科学基金项目(51578189)。

摘　　要：基于向量自回归模型(VAR)的结构损伤识别方法缺少自回归系数与损伤的明确关系,在选择自回归系数作为损伤敏感特征时具有一定的盲目性。文中借助结构运动方程的二阶向量自回归模型(简写为VAR(2)模型)表达形式,论证了采用VAR(2)模型的一阶自回归系数作为损伤敏感特征的合理性,并通过灵敏度分析得到该系数变化与结构刚度变化的线性关系,据此提出了基于VAR(2)模型自回归系数异常检测的损伤识别策略。该方法首先对环境激励下的结构响应信号建立VAR(2)模型,提取第一阶自回归系数矩阵的对角线元素作为损伤敏感特征;然后以欧氏距离建立损伤指标,最后借助统计理论确定损伤阈值进行损伤预警。通过算例表明,该方法适合于长期在线预警。The disadvantage of vector auto-regressive model(VAR)-based structural damage identification methods is lack of both guideline for choosing the damage features and the clear physical meaning of the damage features.In this paper,according to the VAR formation of structural vibration equation,the first autoregressive coefficient of the VAR(2)is proven to be the damage sensitive feature,and a linear relationship between the coefficient and the stiffness is revealed by sensitive analysis.A damage identification method based on the autoregressive coefficients of VAR(2)model is proposed by using novelty detection.The procedure of the proposed method is to establish a VAR(2)model of the structural response under environmental excitation at first,then extract the diagonal elements of the first-order autoregressive coefficient matrix as damage sensitive features.The Euclidean distance of the coefficients between the reference and to-be-tested states is introduced,and the damage threshold for damage alert is determined using statistical theory.Results from a steel truss model test and Xihoumen bridge monitoring show that this method could be potentially used for online damage alert of structures.

关 键 词：结构健康监测 结构损伤识别 向量自回归模型 灵敏度分析 异常检测 

分 类 号：TU311[建筑科学—结构工程]