【作者】 王步宇；

【导师】 俞亚南； 王柏生；

【作者基本信息】 浙江大学 ， 市政工程， 2014， 博士

【摘要】 梁型结构广泛应用于国民经济的各个方面,尤其是桥梁。桥梁作为交通运输的重要组成部分,是一个国家基础设施建设的重点,同时也是经济发展与技术进步的象征。近些年来,梁型结构的健康诊断技术已经成为工程界的研究热点,健康诊断系统及其理论研究也取得了很大进展。但是由于健康诊断系统本身的多学科交叉性、大型梁型结构及其环境的复杂性和不确定性,使得梁型结构的健康监测系统的许多关键技术从理论到实际应用还存在许多不足。本文采用理论与试验相结合的方法,针对梁型结构健康监测过程中的一些关键技术问题进行了研究。在基于结构振动特性的损伤识别技术的基础上,采用了信息理论、信号处理、时频分析、统计分析等领域内的先进方法,对梁型结构损伤识别的信号处理、特征参数识别等方面进行了探索。论文的主要工作内容体现在以下几个方面：(1)针对交通荷载的移动特性,通过研究和实践提出了用移动荷载作用于梁型结构上,对其产生的振动响应数据进行分析。当移动荷载作用于梁型结构上时,损伤引起的振动响应数据特征参数的变化将被放大,可以提高损伤检测特征参数的提取精度。(2)桥梁结构健康监测过程中获得的观测数据具有非线性、非平稳等复杂性,样本熵可以有效地表征信号的复杂性,估计信号的非线性程度,本文提出了使用样本熵来提取结构损伤信息,并且使用经验模式分解及神经网络对该方法进行了进一步的改进。(3)当移动荷载作用于有损伤的结构上时,产生的振动信号是非平稳信号,桥梁损伤结构振动响应的统计量将随着时间和载荷而变化。本文提出了梁型结构损伤分析的时频分析方法,并对时频分析中的交叉项问题进行了讨论,分析了抑制交叉项干扰的方法,并结合信息熵、神经网络进行结构损伤识别。(4)信号的高阶统计量具有良好的非高斯、非平稳信号处理能力,本文提出了损伤结构振动信号的高阶谱分析方法。由于高阶谱分析结果为二维甚至更高维的,包含的信息量大,因此本文提出了双谱的有效值熵分析方法,结合神经网络的模式识别能力进行结构的损伤识别。(5)为充分挖掘高阶谱分析结果中所包含的结构损伤信息,需要配合有效的降维分析方法。本文提出了一种改进的有监督保局投影数据降维方法,通过该方法对高阶谱分析结果进行特征向量的提取,并将损伤识别过程分为两个模块进行综合信息融合,运用神经网络进一步进行损伤识别。更多还原

【Abstract】 As an important part of transportation, Bridge is the focus of the national infrastructure construction, but also a symbol of economic development and technological progress. In recent years, the bridge structure health monitoring technology has become a research hotspot in engineering field. Health monitoring system and theoretical research has also made great progress. However, due to health monitoring system itself multidisciplinary nature, large bridge structure and the complexity and uncertainty of environment, many of the key technology of the bridge structures health monitoring system still exist many problems from theory to practical application.This paper combines theory with experiment; some key technical problems in beam structure health monitoring process were studied. Based on the technology of structural damage identification on the vibration characteristics, this paper has carried on the exploration of signal processing, feature parameter identification of bridge structural health monitoring system using the advanced method in the field of information theory, signal processing, time-frequency analysis, and statistical analysis. The major work of this paper can be presented in the following aspects:(1) According to the moving characteristics of traffic load, the moving load is put on beam structure, and the vibration response data were analyzed. When putting the moving load on the beam structure, characteristic parameters of vibration response data caused by damage will be amplified, and its extraction accuracy can be improved.(2) The observation data obtained in the process of bridge structure health monitoring is complex in nonlinear and non-stationary, sample entropy can effectively represent the signal complexity, estimate the signal nonlinear. This paper proposes using sample entropy to extract structural damage information, and improves this method using empirical mode decomposition and neural network.(3) When moving load is put on the damaged structure, vibration signal is non-stationary, and vibration response statistics of damage structure will change with time and the load. This paper presents the time-frequency analysis method for structural damage identification, discusses the cross-term problem of the time-frequency analysis, analyze the method of suppressing cross-term interference, and combine the information entropy and neural network for structural damage identification.(4) Higher-order statistics signal has a good non-gaussian, non-stationary signal processing ability. This paper proposes the high-order spectrum analysis method for damage structure vibration signal. Due to the high-order spectrum analysis results is two-dimensional or even more higher-dimensional, it contains a huge amount of information. This paper puts forward a bispectrum valid values entropy analysis method, combined with pattern recognition ability of neural network for structural damage identification.(5) In order to fully excavating the structural damage information contained in the higher-order spectrum analysis results, it is necessary to using the effective dimension reduction analysis method. This paper presents an improved supervised locality preserving projection dimension reduction method. Feature vectors can be extracted from the high-order spectrum analysis results. The damage identification process is divided into two information fusion modules, and then completed with neural network.更多还原