基于贝叶斯估计的RC框架结构地震损伤概率识别方法改进与验证

The structural seismic damage identification is one of the most important research area in civil engineering. The structural damage identification is a high indeterminate problem under the influences of the several uncertainty factors. Therefore, the structural damage identification based on probabilistic analysis needs to be developed and improved urgently. In this project, the damage identification method for RC frame structure based on Bayesian estimation will be presented and improved. According to the outstanding problems of the application of Bayesian estimation in the damage identification, the improved single component Metropolis-Hasting (MH) sampler technique and algorithm will be presented based on the optimal proposal distribution with changing variance, to enhance the sampling efficiency, stability and reliability. The marginal probability distribution and optimal estimation of the structural physical parameters will be identified using the improved MH sampler algorithm, and the damage location and extent will be detected and estimated. Furthermore, the cumulative seismic damage model based on displacement and hysteretic energy proposed by Park and Ang will be introduced into the Bayesian estimation theory to compute the probability distribution of the quantitative damage index, by synthesizing structural nonlinear response and the new sampler algorithm. Then, the damage will be identified and estimated quantitatively in term of probability meanings. At last, the efficiency, robustness and accuracy of the new method will be verified by using numerical simulation combined with shaking table test, while the practicability, reliability and validity will be tested by using the seismic response records of the existing structures damaged in earthquakes. The most important significance is that the improved method and technique can not only detect the damage locations but also identify the extent of damage quantitatively. The results will provide scientific and technical support for structural health monitoring, safety assessment, rehabilitation and strengthening, and related work.

结构地震损伤识别与评估是土木工程领域重要研究方向。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。采用理论分析和试验模拟相结合的手段，针对RC框架结构，根据贝叶斯估计方法在实际结构损伤识别中存在的突出问题，从结构响应出发，改进随机抽样技术，提出基于最优提议分布的变方差逐分量MH抽样技术与算法，提高算法计算效率、稳定性及可靠性，从而识别结构参数边缘概率分布和最优估计值，并根据其变化探测结构损伤位置及程度。将Park & Ang双参数破坏准则与贝叶斯估计理论相结合，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概率分布，实现RC框架结构地震损伤概率识别与评估。利用结构数值模拟、振动台试验及实际结构震害，验证方法有效性、鲁棒性及实用性。成果可供结构健康诊断、安全鉴定及修复加固等工作参考。

结构损伤识别与评估是土木工程领域重要研究方向，可以大大减轻地震灾害损失。鉴于结构损伤识别的不确定性特征，发展可靠、高效的结构地震损伤识别概率方法是当前地震工程领域亟待解决的重要问题。基于此背景，本项目严格按照任务书开展研究，全部按计划完成了预定研究内容和考核指标，解决了关键科学技术问题，取得了预期重要成果。.（1）提出了基于结构响应的物理参数识别贝叶斯方法并进行了算例验证。在全面调研研究现状的基础上，针对结构运动微分方程，提出了基于结构地震响应与环境激励响应的物理参数识别贝叶斯方法。所提方法省去了识别模态参数的过程，避免了引入额外误差，只需要布置较少的测点或根据有限的反应记录即可以获得较好的识别效果。.（2）提出了改进的逐分量自适应MH抽样算法并进行了验证。对传统的SCAM算法中提议分布方差的构造方法进行了改进，提出了基于最优提议分布、逐分量变方差MH抽样算法，给出了改进算法抽样的具体步骤，编制了计算分析程序。改进算法可以减少抽样过程中病态样本的数量，显著提高了结构物理参数抽样效率和收敛速度。.（3）提出了基于贝叶斯估计的结构地震损伤定量指标概率评估方法。将Park-Ang双参数损伤指数模型引入到了贝叶斯框架中，综合结构非线性地震反应分析与所改进的抽样算法，提出了一种基于定量损伤指标的结构地震损伤概率评估新方法，实现了结构地震定量损伤水平的概率评估。.（4）开展了典型RC框架结构模拟地震振动台试验。建立了2个典型RC框架结构振动台试验模型，开展不同水平地震动作用下的结构模型破坏试验，获得了结构模型的各类反应数据、损伤数据及破坏现象，为验证本项目所提方法提供了可靠依据和科学数据。.（5）利用结构振动台试验数据及实际结构地震反应记录对改进的结构参数概率识别方法以及地震损伤概率评估方法进行了验证。将所提方法应用于振动台试验结构模型及实际建筑结构，对结构模型及实际结构各类参数进行了识别和损伤状态进行了评估，给出了结构损伤位置识别及地震损伤评估结果，并与实际情况进行对比，全面验证了所提方法有效性、准确性、可靠性和实用性。.项目研究成果及技术方法可以为工程结构地震反应监测、结构地震损伤评估、结构地震安全鉴定等提供重要技术支撑。