Sensitivity analysis of inverse problems in EM non‐destructive testing. Issue 5 (1st July 2020)

• Record Type: Journal Article
• Authors: Bilicz, Sándor
• Is Part Of: IET science, measurement & technology. Volume 14:Issue 5(2020)
• Page Start: 543
• Page End: 551
• Publication Date: 2020-07-01
• Subjects: polynomials -- sensitivity analysis -- eddy current testing -- nondestructive testing -- inverse problems -- chaos -- stochastic processes; Sobol indices -- polynomial chaos expansion surrogate model -- entire inversion scheme -- eddy‐current NdT -- sensitivity analysis -- inverse problem -- EM‐NdT -- electromagnetic nondestructive testing -- material defect parameters -- EM field measurements -- reconstructed defect parameters -- model‐based inversion -- EM simulation -- computational complexity -- configuration uncertainties; Measurement -- Periodicals; Electrical engineering -- Periodicals; Electronics -- Periodicals; Nanotechnology -- Periodicals; Electromagnetism -- Periodicals; Medical instruments and apparatus -- Periodicals; 621.3
• DOI: 10.1049/iet-smt.2019.0370
• Languages: English
• ISSNs: 1751-8822

Abstract

Abstract : The inverse problem of electromagnetic (EM) non‐destructive testing (NdT) consists of reconstructing material defect parameters invoking EM field measurements. Uncertainties of the configuration (e.g. imprecise constitutive and geometrical parameters) are inevitably present; hence, the reconstructed defect parameters are also uncertain. In this study, the different sources of uncertainty are ranked by means of sensitivity analysis. The model‐based inversion (involving EM simulation) is computationally demanding; moreover, sensitivity analysis usually requires a vast number of repeated runs of the inversion. To overcome the computational complexity, surrogate models are applied at different levels. Interpolation on a sparse grid is used as a surrogate model of the EM simulation. The sensitivity of the reconstructed defect parameters concerning configuration uncertainties is characterised by means of Sobol indices. The Sobol indices are obtained from a polynomial chaos expansion surrogate model of the entire inversion scheme. A numerical example drawn from eddy‐current NdT is thoroughly analysed to illustrate the proposed methodology and to demonstrate its performance.