基于贝叶斯模型更新的结构损伤识别方法改进及应用

发布时间：2018-05-19 20:09

本文选题：最优提议分布 + 马尔可夫链蒙特卡罗抽样　；参考：《中国地震局工程力学研究所》2015年硕士论文

【摘要】：结构地震损伤识别是土木工程领域的重要研究方向。受各种不确定性因素影响,结构损伤识别是高度不确定性问题,发展可靠、高效的结构地震损伤识别概率方法与技术成为当前亟待解决的关键科学问题。本文针对目前贝叶斯方法在实际结构损伤概率识别中存在的突出问题,采用理论分析、数值模拟和试验相结合的手段,从结构响应出发,改进随机抽样技术,提出了基于贝叶斯模型更新的结构物理参数识别和损伤诊断改进方法,并将Park-Ang双参数损伤模型与该方法相结合,实现了结构地震损伤水平概率识别与评估。本文主要研究工作如下:1、马尔可夫链蒙特卡罗(MCMC)抽样方法改进MCMC抽样是实现贝叶斯模型更新的最重要技术手段之一,目前该抽样技术在实际工程应用中还存在计算效率低、收敛速度慢甚至不收敛、求解维度低等问题,针对这些问题,本文提出了基于最优提议分布的逐分量自适应Metropolis-Hastings(MH)抽样技术与改进算法,提高了算法的计算效率、收敛速度和稳定性,并通过一个数值算例,验证了所提抽样算法的有效性与可靠性。2、基于结构时域响应的物理参数识别贝叶斯方法针对结构损伤识别非确定性问题及目前传统贝叶斯模型更新两阶段识别方法依赖于模态参数识别问题,本文将结构时域响应作为观测量,根据贝叶斯模型更新方法确定结构物理参数的后验联合分布,并采用所提基于最优提议分布的逐分量自适应MH抽样算法进行分析,得到各物理参数的后验边缘概率分布和最优估计值,从而给出了基于结构时域响应的物理参数识别贝叶斯方法。对某五层剪切型数值模拟结构进行参数识别,结果表明:所提方法可以准确的识别出结构物理参数及其变化,并可有效降低物理参数的不确定性。3、钢筋混凝土(RC)框架结构振动台试验损伤识别利用所提的基于结构时域响应的物理参数识别贝叶斯方法对一个3层RC框架结构试验模型在多次地震动作用下的物理参数及累积损伤进行了识别,验证了方法的有效性、可靠性与实用性。进一步与传统基于模态参数的结构物理参数识别方法分析结果进行对比,结果表明:基于结构时域响应的物理参数识别贝叶斯方法得到的结果更为可靠。4、结构地震损伤水平概率评估将Park-Ang双参数损伤模型与贝叶斯模型更新方法相结合,综合结构非线性地震反应分析与改进的抽样算法,确定结构定量化损伤指标的概率分布,从而给出了一种结构地震损伤水平概率识别与评估新方法。对上述3层框架结构试验模型进行损伤识别与评估,并与试验现象进行对比,验证了方法的有效性与合理性。
[Abstract]:Structural seismic damage identification is an important research direction in the field of civil engineering. Under the influence of various uncertain factors, structural damage identification is a highly uncertain problem. The development of reliable and efficient probabilistic methods and techniques for structural seismic damage identification has become a key scientific problem to be solved. In this paper, aiming at the outstanding problems of Bayesian method in actual structural damage probability identification, the random sampling technique is improved by combining theoretical analysis, numerical simulation and experiment. An improved method of structural physical parameter identification and damage diagnosis based on Bayesian model updating is proposed, and the probability identification and evaluation of seismic damage level of structure is realized by combining Park-Ang two-parameter damage model with this method. The main research work of this paper is as follows: 1) the Markov chain Monte Carlo (MCM) sampling method is one of the most important technical means to realize Bayesian model updating. At present, the sampling technique has low computational efficiency in practical engineering application. The convergence rate is slow or not, and the dimension is low. In order to solve these problems, an adaptive Metropolis-HastingsMH sampling technique and an improved algorithm based on the optimal proposed distribution are proposed in this paper, which improves the computational efficiency of the algorithm. Convergence rate and stability, and through a numerical example, The validity and reliability of the proposed sampling algorithm are verified. The Bayesian method of physical parameter identification based on the time-domain response of structure is used to solve the non-deterministic problem of structural damage identification and the traditional Bayesian model updating two-stage identification method. The method depends on the identification of modal parameters. In this paper, the time-domain response of the structure is taken as the observation, and the posteriori joint distribution of the physical parameters of the structure is determined according to the Bayesian model updating method, and the proposed component-by-component adaptive MH sampling algorithm based on the optimal proposed distribution is used to analyze it. The posterior edge probability distribution and optimal estimation of each physical parameter are obtained, and a Bayesian method for identifying physical parameters based on structural time domain response is presented. The parameters of a five-layer shear numerical simulation structure are identified. The results show that the proposed method can accurately identify the physical parameters of the structure and its changes. And can effectively reduce the uncertainty of physical parameters .3. reinforced concrete frame structure vibration table test damage identification using the proposed time domain response based on the physical parameters identification Bayesian method for a three-story RC frame structure The physical parameters and cumulative damage of the experimental model under repeated ground motions are identified. The validity, reliability and practicability of the method are verified. Compared with the traditional structural physical parameter identification method based on modal parameters, The results show that the Bayesian method based on the time-domain response of structure is more reliable. The probability evaluation of structural damage level combines Park-Ang two-parameter damage model with Bayesian model updating method. The probability distribution of the quantitative damage index is determined by combining the nonlinear seismic response analysis and the improved sampling algorithm, and a new method to identify and evaluate the seismic damage level is presented. The damage identification and evaluation of the above three story frame structure test model are carried out, and compared with the experimental phenomena, the validity and rationality of the method are verified.
【学位授予单位】：中国地震局工程力学研究所
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TU317

【参考文献】