基于混合高斯分布的联合鲁棒稀疏贝叶斯学习结构损伤识别模型
A Joint Robust Sparse Bayesian Learning Model with the Mixture of Gaussians for the Structural Damage Identification
李荣鹏 1宋学力 1易稳 1肖玉柱 1邓庆田 1李新波1
作者信息
- 1. 长安大学理学院,陕西西安 710064;西安市结构损伤数字化检测技术重点实验室,陕西西安 710064
- 折叠
摘要
损伤识别是结构健康监测的核心问题之一.基于混合高斯分布的鲁棒稀疏贝叶斯学习损伤识别模型(RSBLM-MoG),因利用混合高斯分布准确地量化损伤识别中的不确定性而得到了较好的损伤识别结果.然而,该模型在损伤识别中忽略了多组测试数据之间相似性信息的利用.为了进一步提高损伤识别精度,通过贝叶斯联合学习技术实现多组测试数据之间相似性信息的利用,提出基于混合高斯分布的联合鲁棒稀疏贝叶斯学习损伤识别模型(JRSBLM-MoG)和其相应的基于Laplace近似的期望最大化算法.平面桁架结构数值算例和固支梁试验研究结果表明:JRSBLM-MoG的损伤识别精度比RSBLM-MoG分别平均提高了7.61％和3.54％;JRSBLM-MoG可有效利用多组测试数据之间相似性信息提高损伤识别精度.
Abstract
Damage identification is one of the core components in structural health monitoring.The robust sparse Bayesian learning model with the mixture of Gaussians(RSBLM-MoG)has been illustrated to be effective in the field of damage identification because of the accurate quantification of the mixture of Gaussians for the uncertainties of damage identification.However,RSBLM-MoG ignores the similarity information among the multiple groups of measurements in damage identification.In order to further improve the accuracy of damage identification,this paper employs joint learning technology to utilize the similarity information among these measurements.As a result,a joint robust sparse Bayesian learning model with the mixture of Gaussians(JRSBLM-MoG)and its corresponding expectation-maximization algorithm based on the Laplace approximation technique are proposed for damage identification.A numerical example on a truss structure and an experimental validation on a fixed-end beam structure have verified that JRSBLM-MoG improves the accuracy of damage identification by 7.61％and 3.54％than RSBLM-MoG,respectively.This study shows that JRSBLM-MoG can effectively employ the similarity information among multiple measurements to improve the accuracy of damage identification.
关键词
桥梁工程/损伤识别/稀疏贝叶斯学习/联合学习/期望最大化/Laplace近似Key words
bridge engineering/damage identification/sparse Bayesian learning/joint learning/expectation-maximization/Laplace approximation引用本文复制引用
基金项目
陕西省自然科学基础研究计划(2024JC-ZDXM-23)
出版年
2024
NETL
NSTL
万方数据