Improving the reliability of POD curves in NDI methods using a Bayesian inversion approach for uncertainty quantification - Archive ouverte HAL Access content directly
Conference Papers Year : 2016

Improving the reliability of POD curves in NDI methods using a Bayesian inversion approach for uncertainty quantification

(1) , (1) , (1)
1

Abstract

This contribution provides an example of the possible advantages of adopting a Bayesian inversion approach to uncertainty quantification in nondestructive inspection methods. In such problem, the uncertainty associated to the random parameters is not always known and needs to be characterised from scattering signal measurements. The uncertainties may then correctly propagated in order to determine a reliable probability of detection curve. To this end, we establish a general Bayesian framework based on a non-parametric maximum likelihood function formulation and some priors from expert knowledge. However, the presented inverse problem is time-consuming and computationally intensive. To cope with this difficulty, we replace the real model by a surrogate one in order to speed-up the model evaluation and to make the problem to be computationally feasible for implementation. The least squares support vector regression is adopted as metamodelling technique due to its robustness to deal with non-linear problems. We illustrate the usefulness of this methodology through the control of tube with enclosed defect using ultrasonic inspection method.

Dates and versions

cea-01827753 , version 1 (02-07-2018)

Identifiers

Cite

A. Ben Abdessalem, F. Jenson, P. Calmon. Improving the reliability of POD curves in NDI methods using a Bayesian inversion approach for uncertainty quantification. 42nd annual review of progress in quantitative nondestructive evaluation: Incorporating the 6th European-American Workshop on Reliability of NDE, Jul 2015, Minneapolis, United States. ⟨10.1063/1.4940653⟩. ⟨cea-01827753⟩
28 View
0 Download

Altmetric

Share

Gmail Facebook Twitter LinkedIn More