基于加速度二次协方差矩阵和神经网络的结构损伤识别  

Structural damage identification based on covariance of covariance of matrix of acceleration and neural network

作　　者：刘军香 王立新[2,3,4] 姜慧 朱嘉健[2,3] 卢滔 LIU Junxiang;WANG Lixin;JIANG Hui;ZHU Jiajian;LU Tao(Institute of Disaster Prevention,Sanhe 065201,China;CEA Key Laboratory of Earthquake Monitoring and Disaster Mitigation Technology,Guangdong Earthquake Agency,Guangzhou 510070,China;Guangdong Provincial Key Laboratory of Earthquake Early Warning and Safety Diagnosis of Major Projects,Guangdong Earthquake Agency,Guangzhou 510070,China;Shenzhen Academy of Disaster Prevention and Reduction,Shenzhen 518003,China)

机构地区：[1]防灾科技学院,河北三河065201 [2]广东省地震局中国地震局地震监测与减灾技术重点实验室,广东广州510070 [3]广东省地震局广东省地震预警与重大工程安全诊断重点实验室,广东广州510070 [4]深圳防灾减灾技术研究院,广东深圳518003

出　　处：《地震工程与工程振动》2019年第3期214-221,共8页Earthquake Engineering and Engineering Dynamics

基　　金：地震科技星火计划项目(XH16031);广东省科技计划项目(2015A020217007);中央高校基本科研业务费(ZY20180303)~~

摘　　要：为了能对结构早期损伤进行有效识别,本文提出了一种基于加速度响应二次协方差(CoC)矩阵和神经网络的结构损伤识别方法。首先通过数值模拟,以白噪声作为激励,获取结构在不同损伤位置和损伤程度下的加速度响应,并计算相应的二次协方差矩阵;然后,把二次协方差矩阵作为BP神经网络的输入特征向量,对网络进行训练并对损伤位置和损伤程度同时进行识别。本文以桁架为例,将二次协方差矩阵和BP神经网络结合,对结构单损伤和多损伤分别进行识别,同时采用模态频率和模态振型与BP神经网络结合作为对比指标。对比发现:相比于模态指标,基于加速度响应二次协方差矩阵和BP神经网络的损伤识别方法,能够较好的识别结构的单损伤和多损伤,且具有更好的稳定性和抗噪性。In order to identify structural early damage effectively,this paper proposed a method for structural damage identification based on the Covariance of Covariance (CoC) matrix of acceleration responses and neural network method.Firstly,under the white noise excitation,the acceleration responses of the structure with different damages is obtained and used to calculate the corresponding CoC matrix.Secondly,the covariance of covariance matrix was used as the input feature vector for back propagation (BP) neural network to train the network.Thirdly,the trained network is used to identify the structural damage state.In this paper,a truss structure is used as an example.The CoC matrix of acceleration responses was combined with the BP neural network to identify the damage.In the meanwhile,modal frequency and modal shape combined with BP neural network were used as the comparison index.Compared with the modal index,the damage identification method based on CoC matrix and BP network can better identify the damages,and has better stability and noise resistance ability.

关 键 词：加速度响应 白噪声激励 二次协方差矩阵 BP神经网络 损伤识别 

分 类 号：P315.9[天文地球—地震学]