Skip to Main content Skip to Navigation

L'analyse de sensibilité globale sur matériaux composites vibroacoustiques avec la dépendance paramétrique

Abstract : With rapid development of mathematical models and simulation tools, the need of uncertainty quantification process has grown higher than ever before. Parametric uncertainties and overall decision stacks are nowadays the two main barriers in solving large scale systematic problem.Global Sensitivity Analysis (GSA) is one reliable solution for uncertainty quantification which is capable to assess the uncertainty of model output on its inputs’. Among several GSA algorithms, Fourier Amplitude Sensitivity Test (FAST) is one of the most popular choices of researchers. Based on ANOVA-HDMR (ANalysis Of VAriance - High Dimensional Model Representation), it is both mathematically solid and computationally efficient.One unfortunate fact is that the uniqueness of ANOVA-HDMR relies on the independency of input variables. It makes FAST unable to treat many industrial cases especially for those with only datasets but not distribution functions to be found. To answer the needs, two extended FAST methods with correlation design are proposed and further studied in this research. Among them FAST-c is distribution-based and FAST-orig is data-based.As a frame of validation and application, a number of vibroacoustic problems are dealt with in this research. Vibroacoustic materials with substructures, are perfect test candidates for FAST-c and FAST-orig. Two application cases are presented in the first part of this thesis, following the literature review. The models chosen here are poroelastic material and sandwich composite structures, both having their mechanical properties hugely influenced by their microscopic and mesoscopic geometric parameters. Getting the original FAST method compared to the two with correlation design, many different features on materials’ vibroacoustic performance are latter discovered.Having got an answer for GSA on models with dependent variables, the second part of this thesis contains more extended researches related to FAST. It is taken into comparison with Random Forest, a well-known data-mining algorithm. The potential error of both algorithms are analyzed and the possibility of joint application is discussed. In the following chapters, more applications of FAST-series methods are reported. They are applied under various conditions where another improved version named FAST-pe is developed to treat a model of periodic structures with correlation among each units. Upon these FAST application cases, the design of preliminary process and the sampling strategies is the core part to be introduced.
Document type :
Complete list of metadatas

Cited literature [63 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Wednesday, April 24, 2019 - 11:25:10 AM
Last modification on : Thursday, November 21, 2019 - 1:40:42 AM


Version validated by the jury (STAR)


  • HAL Id : tel-02108453, version 1



Wenqi Chai. L'analyse de sensibilité globale sur matériaux composites vibroacoustiques avec la dépendance paramétrique. Other. Université de Lyon, 2018. English. ⟨NNT : 2018LYSEC037⟩. ⟨tel-02108453⟩



Record views


Files downloads