基于时间序列模型和主成分分析的结构损伤识别研究

发布时间：2018-01-24 17:54

本文关键词： 损伤识别时间序列主成分分析平稳时间序列条件异方差　出处：《烟台大学》2013年硕士论文　论文类型：学位论文

【摘要】：基于时间序列的损伤识别作为一种全新的研究方法，近年来得到了不断的完善和发展，属于结构损伤诊断的前沿领域。但目前时间序列分析的应用存在一些问题：对于简支梁、连续梁等简单结构应用较多，对于框架、网架等复杂结构的应用研究还比较少；基于时序分析的研究方法以往大都应用于结构的状态预测，对于结构的损伤定位和程度的判断还处于发展和完善阶段；线性平稳时间序列的应用广泛，对于非线性时间序列的应用很少。针对以上问题，本文拟结合框架和网架结构的自身特点，对于时间序列分析法在结构损伤识别中的应用作了进一步研究。主要工作包括： (1)综述了结构损伤识别的当前国内外研究情况和未来的发展趋势。首先系统总结了损伤识别领域内不同识别方法的优缺点。并根据时间序列模型和结构系统之间的联系，重点阐明了时间序列分析在结构损伤识别中的基本原理。 (2)对时间序列模型的建模过程作了深入研究，包括对系统激励源的分析、结构响应数据的采集和预处理、模型识别、模型参数的计算和模型定阶等问题。详细阐述了主成分分析法的基本原理，对其在结构损伤识别中的应用背景作了介绍。 (3)提出了基于时间序列和主成分分析的结构损伤识别方法。分别采用时间序列模型参数和均方根误差作为损伤敏感因子进行结构损伤识别。由于损伤识别矩阵维数较高，信息出现重叠现象，采用主成分分析法对原始矩阵进行压缩和解释，得到损伤敏感因子主成分矩阵，保留了原始矩阵中的有效信息，简化了计算过程。针对AR、MA模型自身的局限性，本文采用ARMA模型对结构的动力响应数据进行建模，，采用美国Los Alamos实验室框架结构模型的损伤实验数据验证了基于时间序列和主成分分析法的有效性。 (4)利用时间序列和主成分分析法对一个典型的双层网架结构进行了损伤识别数值模拟。首先基于ANSYS进行有限元建模，对结构进行模态分析。然后在不同工况下分别采用高斯白噪声作为激励源，对结构进行动力时程响应分析，获取各节点的加速度响应作为损伤识别原始数据。最后通过MATLAB健康监测工具箱建立ARMA模型进行损伤识别，并考虑了结构响应在不同噪声水平干扰下的变化。研究结果表明网架结构损伤单元节点处的损伤敏感因子DSF值远大于未损伤单元节点，定位效果显著，进一步验证了该方法的可行性和有效性。 (5)针对ARMA模型在非线性损伤识别中的所存在的问题，基于美国Los Alamos实验室框架结构模型的损伤实验数据，采用条件异方差GARCH模型进行建模识别。将两种模型的识别结果进行对比，研究结果表明GARCH模型在非线性损伤识别中受外界环境干扰较小，计算方便，对于结构的损伤定位和程度的判断能力高于线性ARMA模型，进一步提高了损伤识别精度。本文提出的基于时间序列和主成分分析的结构损伤识别方法适用于框架、网架等较为复杂的结构，进一步提高了结构损伤的定位和程度判别效果，具有一定的实践意义。
[Abstract]:As a new research method of damage identification based on time series in recent years has been the forefront of continuous improvement and development, which belongs to the field of structural damage diagnosis. But the application of time series analysis has some problems: for more, simply supported beam, continuous beam and simple structure for application framework, application of grid complex structure is still relatively small; research methods of time series analysis is usually applied to the state prediction based on the damage localization of the structure and the degree of judgment is still in the stage of development and perfection; the application of linear stationary time series, for the application of nonlinear time series. To solve the above problems, this paper combines the framework and network structure itself the characteristics of time series analysis method for the further study of structural damage identification. The main work includes:
(1) review of structural damage recognition at home and abroad research situation and development trend in the future. Firstly, summarizes the advantages and disadvantages of damage identification in different recognition methods. And according to the time sequence between the model and the structure of the system, focusing on the time series analysis in the basic principle of structural damage recognition.
(2) the process of modeling the time series model is researched, including the analysis of the excitation of the system, the response data acquisition and preprocessing, model identification, order such as calculating the model parameters and model. Elaborated the basic principle of principal component analysis, introduces the structure of the application background of damage identification.
(3) put forward the structural damage identification of time series analysis and principal component method based on time series model parameters and the RMS error are used as damage sensitive factor for structural damage identification. The damage identification matrix of high dimension, information overlap phenomenon, using principal component analysis to compress and interpretation of the original matrix method, get the damage sensitive factor of the principal component matrix, retaining the effective information in the original matrix, simplify the calculation process. According to AR, the limitations of the MA model, this paper uses ARMA model to model the structural dynamic response data, verify the validity of the time series and the principal component analysis method based on the experimental data of American Los laboratory Alamos damage the frame structure of the model.
(4) analysis was carried out on a numerical simulation of damage identification of double grid structure by using the method of typical time series and the first principal component. Based on ANSYS finite element model, the modal analysis of the structure. Then under different conditions using Gauss white noise as the excitation source, the response analysis of the structure of power, access the node acceleration response as the damage identification of the original data. Finally through the MATLAB toolbox to establish the ARMA model for health monitoring and damage identification, and considering the structural response changes at different noise levels under the interference. The results showed that the damage sensitive factor DSF grid structure damage element node is greater than the undamaged unit node positioning effect further, to verify the feasibility and validity of this method.
(5) for the ARMA model on nonlinear damage identification of the problems, the United States Los Alamos laboratory injury experimental data frame structure model based on the conditional heteroskedasticity GARCH model identification. Will identify the results of the two models were compared, the results showed that the GARCH model in nonlinear damage recognition by small interference the external environment for the convenience of calculation, the linear ARMA model to determine the ability of structure damage location and degree of above, to further improve the accuracy of damage identification.
The structural damage identification method based on time series and principal component analysis is suitable for complex structures such as frame and grid. It further improves the location and degree discrimination of structural damage, and has certain practical significance.

【学位授予单位】：烟台大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TU317

【相似文献】