TY  - RPRT
PB  - University of Trento
A1  - Salucci, M.
A1  - Anselmi, N.
A1  - Oliveri, G.
A1  - Massa, A.
M1  - technical_report
KW  - Eddy current testing
KW  -  inverse scattering
KW  -  nondestructive testing and evaluation
KW  -  statistical learning
KW  -  learning-by-examples
KW  -  support vector regression
KW  -  output space filling
KW  -  partial least squares
KW  -  adaptive sampling
N2  - This document presents an innovative adaptive learning-by-examples (LBE) strategy for accurate crack localization in planar conductive specimens. The developed approach exploits a Partial Least Squares (PLS) feature extraction technique and an adaptive sampling strategy in order to build optimal training sets of input/output (I/O) pairs. Such information is then used to train, during a preliminary off-line phase, a Support Vector Regressor (SVR) in order to build a fast surrogate model of the inverse operator linking ECT data and crack position. Finally, during the on-line test phase previously-unseen ECT measurements are given as input to the trained SVR in order to retrieve an estimation of the defect coordinates. Some numerical results are shown, in order to verify the effectiveness of the proposed LBE inversion methodology.
Y1  - 2016///
UR  - http://www.eledia.org/students-reports/713/
ID  - elediasc12713
AV  - public
TI  - An Innovative Learning-by-Examples Approach for Crack Localization Based on Partial Least Squares and Adaptive Sampling
ER  -