Partial least square regression: an analysis tool for quantitative non-destructive testing

¹ Laboratoire de Génie Electrique de Paris (LGEP), CNRS UMR8507, Supelec, Université Pierre et Marie Curie-P6, Université Paris Sud-11, 91192 Gif-sur-Yvette, France
² Department of Broadband Infocommunications and Electromagnetic Theory, Budapest University of Technology and Economics, 1521 Budapest, Egry J. u. 18., Hungary

^a e-mail: yann.le-bihan@lgep.supelec.fr

Received: 23 October 2013
Revised: 22 February 2014
Accepted: 27 February 2014
Published online: 13 August 2014

Abstract

The scanning of parts in eddy current testing can lead to a large amount of measurement data (predictors). Partial least square (PLS) regression is a mean to reduce the dimensionality of the subsequent inverse problem by projecting the predictors in a latent subspace of reduced dimension maximizing the covariance between the projection and the responses which have to be estimated. In a second step, a regression model is elaborated linking the responses to the latent variables. PLS was originally developed in the field of chemical analysis. In this paper, the PLS method is applied in the field of eddy current testing for the characterization of minute cracks. It is tested firstly on simulated data and then on experimental data. It is found that the reconstruction of the area of minute cracks is made possible by PLS.