基于压缩感知的EMI信号与神经网络的结构损伤识别研究  

Structural Damage Identification Method by Compression Sensing-based Signal EMI and Neural Networks

作　　者：彭必全 严蔚[1] 王骥[2] 王炜[1] 

机构地区：[1]宁波大学建筑工程与环境学院,浙江宁波315211 [2]宁波大学机械工程与力学学院,浙江宁波315211

出　　处：《防灾减灾工程学报》2015年第2期173-179,共7页Journal of Disaster Prevention and Mitigation Engineering

基　　金：国家自然科学基金重点项目(10932004);浙江省自然科学基金项目(LY12E08005);宁波大学重中之重学科项目(zj1216)资助

摘　　要：针对当前结构健康监测系统中信号数据量过大、所占内存过高所导致的难以传输和储存的问题,提出将压缩感知方法运用于神经网络系统与压电阻抗(Electro-mechanical impedance,简称EMI)技术进行结构的损伤识别。以压缩后的EMI数据代替原始数据来表征原始损伤信号,用离散余弦变换分析了EMI信号的稀疏度,并以高斯随机矩阵作为观测矩阵,满足了压缩感知理论的限制性等距条件。以齿轮结构为例,采用有限元技术构建了EMI模型,得到了不同裂纹深度下的电导(电阻抗的倒数)信号。采用压缩感知方法先对电导信号进行压缩,再进行主成分分析,以降低BP神经网络输入参数的维度。结果表明:使用压缩感知技术之后,数据传输带宽和储存空间只需为原来的44%;将压缩后的电导数据的主成分作为神经网络的输入参数,该系统能够检测到结构裂纹的存在,并能对裂纹的损伤程度进行有效的定量归类。The transmission and storage of the measured data were excessive in the present struc- tural health monitoring system because of the overloaded signal. Thus, it is proposed to apply compression sensing method to the neural network system and electro--mechanical impedance （EMI） technology for structural damage identification. The compressed EMI data was used to depict the structural damages instead of the original ones. The discrete cosine transform （DCT） was employed to analyze the sparseness degree of the EMI signal. Furthermore, to satisfy the re- stricted isometry conditions of the compression perception theory, the Gauss random matrix was used as the observation matrix. The finite element EMI model was first established for a gear structure and the conductance data were accordingly obtained for various crack depths. The data compressed by the compression perception method was investigated using the principal compo- nent analysis and the obtained principal components were then used to be the input parameters of the BP neural network. Results show that the transmission bandwidth and storage space of the data are only 44% of the original ones after using the compression perception theory, and the neural network system could detect the existence of cracks and could further classify the crack se- verities quantitatively using the principal components of the compressed conductance.

关 键 词：压缩感知 压电阻抗(EMI) BP神经网络 损伤识别 

分 类 号：TH132.417[机械工程—机械制造及自动化] TU317[建筑科学—结构工程]