【作者】 蒋寅军；

【导师】 朱以文； 朱永华；

【作者基本信息】 武汉大学 ， 防灾减灾工程及防护工程， 2011， 博士

【摘要】 由于现场实测或模型试验的困难,在计算模型的基础上进行各种计算分析,是目前复杂结构设计、施工、维护过程中了解结构力学性能的重要手段。然而,由于计算模型隐含有各种假设,通常与实际结构存在差异,导致分析结果与实测结果不吻合。因此,要想获得符合实际的分析结果,首先要保证计算模型的正确性。工程领域的模型修正,就是指利用综合利用各种信息,使建立的结构分析模型与实际相吻合。本文通过回顾结构动力模型修正技术的发展历程和研究现状,指出传统的基于数值模型的修正方法对问题的复杂性考虑不够,将解决办法限制在经典力学原理的范围内,最终陷入数值计算的麻烦。近年来发展起来的基于统计分析技术的模型修正方法虽然注意到问题的复杂性,认为应当使用数值分析的结果作为样本,通过试验设计获取典型样本,使拟合(或回归)得到的结构动力模型尽量与测试结果一致,但没有充分利用结构分析所获得的其他先验信息,过于依赖统计分析技术,而且修正模型的物理意义不明确。本文认为,对于复杂结构,除了基本的经典力学原理外,还有复杂性规律在起作用,而复杂性规律隐藏在数据流中,使得问题变成面向数据的问题。在复杂结构的模型修正过程中,应当综合考虑数学物理关系、统计规律以及先(后)验信息,才有可能获得真实可信的修正结果。本文的主要工作和创新点集中在以下几个方面：(1)提出对复杂结构的动力修正问题应该针对于某一类具体结构进行,而不宜从一般意义上去寻找对任意结构都适用的动力修正技术,这样才能充分利用该类结构的先验信息,保证求解的可行性和稳定性。(2)提出对复杂结构应当建立一个等价的快速运行参数模型来进行动力修正,而不应当直接对有限元模型的质量矩阵、刚度矩阵等进行修正,然后利用响应面方法予以实施。(3)提出一种基于结构力学方程和量纲分析原理的响应面函数构造方法,用于改进传统响应面方法在复杂结构模型修正技术中的应用,力图利用先验信息指导响应面函数的选取,降低响应面回归的复杂度,并寻找响应与设计参数之间的机理模型。(4)提出一种基于统计学习理论的复杂结构模型修正方法。将统计学习理论的最新成果—支持向量机应用于复杂结构模型修正过程中的参数计算,避免了传统模型修正的数值计算困难,也避免了神经网络方法的维数灾难和过学习问题。(5)探索了支持向量机应用于复杂结构模型修正的可行性和途径,并在一个实际的复杂结构—抽水蓄能电站地下厂房结构模型修正过程中应用与实施。更多还原

【Abstract】 Due to the difficulty of prototype test or model test for complex structures during the period of design, construction and maintenance, anaylysis based on computing model is a common alternate method to achieve the mechanical properties of the complex structures. But differences are offen found between the initial computing model, which is built based on the specifications and the experiences of the engineers, and the prototype because of the suppositions in the modeling process. The differences may lead to analysis errors. It indicates that the initial computing model must be modified or updated to fit the analysis request, or make the computing model corresponding to the real conditions.Based on the review on the development and current research of structure dynamic model updating technology, the paper points out that traditional model based on numerical correction method did not take into account the complexity of the problem, limited solved method just in classical mechanics principle and finally entrapped in trouble of numerical calculation. Model updating based on statistical analysis has made a rapid progress in recent years. Although it has given into consideration the complexity of problem, obtained typical samples by means of experiment and employed analysis results as sample in order to make agreement between calculated results and actual ones, it did not make full use of prior infor-mation and excessively depended on analysis technology which leads to indefinite physical meaning of the modified model.This paper argues that not only basic classical mechanics but also complexity play roles in solving complicated structure issue. Complexity hidden in the stream of data, which makes model updating technology face to. The authentic, reliable update results can be achieved only if the relationship between mathematical and physical principles, the laws of statistics and the prior (posterior) information should be given into account during the process of modification model for complicated structure.(1) Put forward that dynamic correction for complex structures should be aimed at spe-cific structure rather than modification techniques which meet all structures, so as to make full use of prior information of this structure to reach and ensure solutions’feasibility and stability and ensure solution.(2) Propose that complex structure should be established on an equivalent fast runnig parameter model for dynamic model modification and use response surface method rather than base on finite element model of the mass matrix, stiffness matrix.(3) Come up with response surface methodology based on structure mechanics equa-tions and dimensional analysis principle to improve the application of traditional response surface method in model modification for complex structure and attempt to make the re-sponse surface function form closer to the real structure of the mechanism model.(4) Propose that complex structure model modification method based on statistical learn- ing theory. Apply the latest results-support vector machine (SVM) model into parameter calculation in the process of complex structure modification and avoid numerical calculation difficulty in traditional model and the dimension disasters and over-fitting problems in neural network method as well.(5) Explore the feasibility and approaches of support vector machine (SVM) in com-plex structure model modification. Apply and implement the theory into the model modifi-cation process of an actual complicated structure-an underground powerhouse structure in a pumped storage power station.更多还原