基于粒子滤波器的结构损伤识别及可靠度分析

发布时间：2018-05-22 19:26

本文选题：粒子滤波器 + 损伤识别　；参考：《东南大学》2016年硕士论文

【摘要】：众所周知,结构从投入使用开始,就面临着环境的侵蚀、材料的老化、荷载的长期作用、突变效应以及疲劳效应等因素的耦合作用,这将难以避免地引发结构的抗力衰减和损伤累积,最终导致结构损伤和破坏。中国建筑科学院关于我国建筑结构的调查研究表明,由于建筑结构的设计、施工和管理方面的原因,对于绝大多数的既有建筑结构来说均存在着不同程度的损伤。另外,我国早期的工业民用建筑、办公楼以及一些桥梁,已经接近设计基准年限。这些结构物在复杂的服役环境下,也已经受到了不同程度的损伤。准确判断结构的损伤部位和损伤程度,既可以确保结构的安全性和完整性,对结构的可靠度进行实时评定、避免灾难性的悲剧发生,也可以对既有建筑结构做出科学合理的维修和加固方案、减少维护费用、提高维护效率、对保障社会和人民的财产免受不必要的损失、加深对结构的性能的理解和研究,同时对促进工程结构实践领域的进一步发展研究均具有重要的现实意义。随着结构健康监测越来越受到人们的重视,对基于结构激励和响应识别结构参数的方法进行了深入研究。目前的结构损伤识别很多都是基于结构参数识别的基础上进行的,即工程结构发生损伤时,结构的参数不可避免地要发生改变。此时,如果能很好地识别结构的参数变化,便可以发现结构损伤的特征。贝叶斯模型修正法就是利用统计学里面的贝叶斯原理,将确定性的结构模型嵌入到一组可能的概率性模型中,使结构模型能够量化模型预测和观测的不确定性。贝叶斯识别方法的基本思路就是将所要估计的结构参数看作随机变量,通过观测和分析与该参数相关的其他变量,以此来推断这个参数值。对于线性高斯状态空间模型,卡尔曼滤波方法可以得到后验概率密度的解析表达式。而实际情况中的数据通常比较复杂,包含有非高斯和非线性的情况。对于这一问题的解决,学者们做了大量的研究,提出了扩展卡尔曼滤波和高斯求和滤波的方法。但是这两种方法均没有考虑过程中的全部统计特性,而产生较差的识别结果。而对于非线性非高斯的情况,问题的解决将变得更加复杂。粒子滤波算法以其对线性和非线性结构参数识别的适用性和有效性脱颖而出,得到人们的广泛关注。本文对粒子滤波在结构参数识别和结构损伤识别方面进行了分析研究,主要研究内容如下：(1)本文首先介绍了结构损伤识别的发展历程以及粒子滤波的主要应用领域和研究现状；其次,对粒子滤波在结构参数识别领域的优势进行了归纳性介绍,指出粒子滤波在结构参数识别领域具有突出的优势。再则,对目前人们在粒子滤波应用研究当中碰到的一些难点问题进行了总结归纳；最后阐明了本文的主要研究内容。(2)提出并介绍了一种基于最大似然值的结构参数识别方法。在不同高斯和非高斯噪声水平下,对单自由度结构的参数识别进行了数值仿真分析,并归纳总结了基于最大似然值的结构参数识别方法的优劣性。(3)介绍说明了粒子滤波算法在结构参数和损伤识别中的基本原理。同时,对粒子滤波算法在单自由度结构参数和损伤识别中的过程进行了详细的说明和系统的讨论。在不同高斯和非高斯噪声水平下,基于粒子滤波算法对单自由度结构的参数和损伤识别进行了数值仿真分析,并归纳总结基于粒子滤波器的结构参数和损伤识别方法的优劣性。另外,基于单自由度结构参数和损伤识别的数值仿真结果,对参数和损伤识别结果的可靠度进行了简要的分析。(4)在单自由度结构参数和损伤识别的研究基础上,将基于粒子滤波器的结构参数和损伤识别方法拓展至多自由度结构。分别对在不同高斯和非高斯噪声水平下,多自由度结构参数和损伤识别进行了数值仿真,分析结构参数和损伤的识别结果,总结归纳基于粒子滤波算法的多自由度结构参数和损伤识别的性能。并且通过一个4层铝框架的实验结构模型对该算法的参数和损伤识别性能进行了实验验证分析,对比实验识别结果与结构层间刚度的测量值,通过实验证实了粒子滤波算法在多自由度结构刚度损伤识别中的有效性。(5)总结了粒子滤波算法在结构参数和损伤识别领域的优越性,并对粒子滤波算法的应用前景进行展望。粒子滤波算法作为结构健康监测领域比较新颖的方法,作者根据自身的研究工作对粒子滤波算法中存在的问题和不足进行了归纳总结。最后,对粒子滤波算法下一步的研究工作进行了一些讨论和展望。
[Abstract]:As we all know, from the beginning of use, the structure is confronted with the coupling of environmental erosion, aging of materials, the long-term effect of load, catastrophe effect and fatigue effect, which will inevitably lead to the attenuation and damage accumulation of structural resistance and damage, and eventually lead to damage and damage to the structure. The structural investigation shows that, due to the design, construction and management of the architectural structure, there are different degrees of damage to most of the existing construction structures. In addition, the early industrial and civil buildings, office buildings and some bridges in our country are close to the design reference years. These structures are in complex service. In the environment, it has also been damaged in different degrees. The accurate judgment of the damage location and degree of the structure can not only ensure the safety and integrity of the structure, but also evaluate the reliability of the structure in real time, avoid the catastrophic tragedy, and make a scientific and rational maintenance and reinforcement scheme for the existing building structure, and reduce the dimension of the structure. To protect the cost, improve the efficiency of maintenance, to protect the society and the people's property from unnecessary losses, to deepen the understanding and research of the structural performance, and to promote the further development of the engineering structure, is of great practical significance. The method of excitation and response identification of structural parameters is studied deeply. Many of the current structural damage identification are based on structural parameter identification, that is, when the structure is damaged, the structural parameters will inevitably change. In this case, if the structure parameters can be identified well, the structure can be found. The Bias model correction method is to use the Bias principle in statistics to embed the deterministic structural model into a set of possible probabilistic models, so that the structural model can quantify the uncertainty of the model prediction and observation. The basic idea of the Bias recognition method is to consider the structural parameters to be estimated. Random variables, by observing and analyzing other variables related to the parameter, deduce the value of the parameter. For the linear Gauss state space model, the Calman filtering method can obtain the analytic expression of the posterior probability density. The data in the actual situation are usually complex, including non Gauss and nonlinear cases. The solution of this problem, scholars have done a lot of research, and put forward the method of expanding Calman filtering and Gauss sum filtering. However, these two methods do not consider all the statistical characteristics in the process, and produce poor identification results. For nonlinear non Gauss, the solution of the problem will become more complex. Particle filtering is more complex. The algorithm is widely paid attention to the applicability and effectiveness of linear and nonlinear structural parameters identification. This paper studies the structure parameter identification and structural damage identification of particle filter. The main research contents are as follows: (1) this paper first introduces the development process of structural damage identification. The main application fields and research status of particle filtering are introduced. Secondly, the advantages of particle filtering in the field of structural parameter identification are introduced. It is pointed out that particle filtering has a prominent advantage in the field of structural parameter identification. Then, some difficult problems that people have encountered in the research of particle filtering are summarized. Finally, the main contents of this paper are clarified. (2) a structural parameter identification method based on maximum likelihood is proposed and introduced. The parameter identification of the single degree of freedom structure is numerically simulated under different Gauss and non Gauss noise levels, and the structural parameter identification based on the maximum likelihood is summed up. The advantages and disadvantages of the method. (3) the basic principle of particle filter algorithm in structural parameters and damage identification is introduced. At the same time, the process of particle filter algorithm in single degree of freedom structure parameters and damage identification is explained in detail and the system is discussed. Under different Gauss and non Gauss noise level, particle filter algorithm pair The parameters of the degree of freedom structure and the damage identification are numerically simulated, and the advantages and disadvantages of the structural parameters based on the particle filter and the damage identification method are summarized. In addition, the reliability of the parameters and the damage identification results is briefly analyzed based on the single degree of freedom structure parameters and the numerical simulation results of damage identification. (4) On the basis of single degree of freedom structural parameters and damage identification, the structure parameters and damage identification methods based on particle filter are extended to the most free degree structure. The number of parameters and damage identification of multi degree of freedom are simulated under different Gauss and non Gauss noise levels, and the structural parameters and the identification of damage are analyzed. As a result, the parameters of the multi degree of freedom structure and the performance of damage identification based on the particle filter algorithm are summarized, and the experimental structure model of a 4 layer aluminum frame is used to verify the parameters and the damage identification performance of the algorithm. The results of the experimental identification and the measurement of the stiffness between the structure layers are compared and the particles are verified by the experiment. The effectiveness of filtering algorithm in multi degree of freedom structural stiffness damage identification. (5) the superiority of particle filter algorithm in the field of structural parameters and damage identification is summarized, and the prospect of application of particle filtering algorithm is prospected. The particle filter algorithm is a more novel method in the field of structural health monitoring. The author is based on the research work of the author. The problems and shortcomings in the particle filtering algorithm are summarized. Finally, the research work of the next step of the particle filter algorithm is discussed and prospected.
【学位授予单位】：东南大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TU317

【参考文献】