Influence of partially known parameter on flaw characterization in Eddy Current Testing by using a random walk MCMC method based on metamodeling

4th International Workshop on New Computational Methods for Inverse Problems

物理学;计算机科学

Cai, Caifang^1 ; Rodet, Thomas^2 ; Lambert, Marc^1

L2S-SUPELEC, 3 Rue Joliot-Curie, Gif sur Yvette

91192, France^1

SATIE, ENS-Cachan, 61 Avenue du Président Wilson, Cachan

94230, France^2

关键词: Flaw characterization;    Forward model calculations;    Markov chain Monte Carlo;    Markov chain Monte Carlo method;    Metamodeling;    Metamodeling methods;    Parameter dependence;    Random Walk;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/542/1/012009/pdf
DOI  :  10.1088/1742-6596/542/1/012009

学科分类：计算机科学（综合）

来源: IOP

【 摘 要 】

First, we present the implementation of a random walk Metropolis-within-Gibbs (MWG) sampling method in flaw characterization based on a metamodeling method. The role of metamodeling is to reduce the computational time cost in Eddy Current Testing (ECT) forward model calculation. In such a way, the use of Markov Chain Monte Carlo (MCMC) methods becomes possible. Secondly, we analyze the influence of partially known parameters in Bayesian estimation. The objective is to evaluate the importance of providing more specific prior information. Simulation results show that even partially known information has great interest in providing more accurate flaw parameter estimations. The improvement ratio depends on the parameter dependence and the interest shows only when the provided information is specific enough.