结构健康监测的鲁棒性贝叶斯压缩采样和损伤识别方法研究

项目名称： 结构健康监测的鲁棒性贝叶斯压缩采样和损伤识别方法研究

项目编号： No.51308161

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 建筑科学

项目作者： 黄永

作者单位： 哈尔滨工业大学

项目金额： 25万元

中文摘要： 数据压缩和损伤识别是结构健康监测重要的科学问题。针对结构健康监测压缩采样数据解压缩和损伤定位对信息不完备程度和环境噪声比较敏感而可靠性不高的问题，并考虑其均具有稀疏性的数学特征，本项目研究结构健康监测数据压缩采样和损伤识别的鲁棒性贝叶斯方法。本项目以稀疏贝叶斯学习为理论基础，以鲁棒性为目标构造噪声方差的多层次贝叶斯模型，将该模型全概率积分应用于稀疏贝叶斯学习过程使方法具有鲁棒性。首先，研究健康监测数据的贝叶斯压缩采样的鲁棒性解压缩方法和数据解压缩精度诊断的鲁棒性方法，实现海量健康监测数据的高压缩效率和高解压精度；其次，研究基于损伤空间稀疏分布特征的结构损伤识别的鲁棒性方法，实现无临界值损伤定位，提高损伤定位精度。最后，通过模型试验和原型监测验证本项目方法。本项目研究将形成系统的结构健康监测贝叶斯统计反演鲁棒性算法，丰富和发展结构健康监测理论，具有重要的科学意义和实用价值。

中文关键词： 结构健康监测；压缩采样；损伤识别；贝叶斯反演；鲁棒性

英文摘要： Data compression and damage detection are important research topics in structural health monitoring. This project will research on the robust Bayesian method for compressive sampling and damage detection in structural health monitoring, motivated by the fact that both the compressive sampling reconstruction (decompression) and damage localization results are unreliable and sensitive to the incomplete information and environmental noise. Considering both the structural health monitoring signals and structural damage are sparse in some basis, the basic theory for this project relies on the sparse Bayesian learning framework. We will construct the hierarchical Bayesian model for noise variance and integrate it out to ensure the robustness of the algorithm. In the aspect of application, firstly, we will study robust decompression and diagnosis methods for compressive sampling of structural health monitoring signals. These two methods can achieve high compression efficiency and decompression accuracy for huge volume of data in structural health monitoring. Then we will propose a robust damage detection method considering the feature of spatially-sparse stiffness loss in a structure, which can localize damage without any specified threshold and also increase the resolution of the damage locations. Finally, the corresp

英文关键词： structural health monitoring；compressive sampling；damage detection；Bayesian inversion；robustness