摘要
提出了基于全局灵敏度分析的有限元模型修正参数选择方法,考虑了参数整个变化空间的作用及参数间的相互作用,具有适用于参数不确定性大和无模型限制等优势。全局灵敏度分析常采用蒙托卡罗方法计算灵敏度指标,因此足够大的采样次数是获得可靠灵敏度指标的前提,但是同时会造成计算成本的增加。为此,采用高斯过程模型取代耗时的有限元模型用于降低计算成本,同时探讨了拉丁超立方抽样、Halton序列和Sobol序列3种空间采样方法用于全局灵敏度分析的计算效率,旨在选择一种高效的采样方法。最后,一桁架人行桥实例验证了有限元模型修正参数选择和采样选择方法。
This paper proposes a global sensitivity analysis（GSA）approach for parameter selection in finite element modal updating.The GSA has the capability to quantify the effects of individual parameters over their entire space and interaction effects among parameters.This approach has a couple of advantages,such as capability to high degree of parameter uncertainty and model independence.However,it is still computationally intensive because a large number of model evaluations for Monte Carlo simulation（MCS）are needed to obtain a confident estimate of the sensitivity indices（SIs）.Therefore,the fast-running Gaussian process model is adopted to replace the time-consuming finite element model to alleviate the computational burden.In addition,this study explores the efficiency of three space-filling sampling methods,i.e.,Latin hypercube sampling,Halton sequence and Sobol sequence,in the SIs evaluation,which has great value for choosing an efficient sampling method.At last,a real-world truss pedestrian bridge is used to demonstrate the proposed framework.
出处
《振动工程学报》
EI
CSCD
北大核心
2015年第5期714-720,共7页
Journal of Vibration Engineering
基金
国家自然科学基金资助项目(51508144)
中央高校基本科研业务费专项资金资助项目(JZ2015HGBZ0098
JZ2015HGQC0215)
关键词
参数选择
有限元模型修正
全局灵敏度分析
高斯过程模型
拉丁超立方抽样
parameter selection
finite element modal updating
global sensitivity analysis
Gaussian process model
Latin hypercube sampling
