Real time groove characterization combining partial least squares and SVR strategies: application to eddy current testing

Abstract : A quasi real-time inversion strategy is presented for groove characterization of a conductive non-ferromagnetic tube structure by exploiting eddy current testing (ECT) signal. Inversion problem has been formulated by non-iterative Learning-by-Examples (LBE) strategy. Within the framework of LBE, an efficient training strategy has been adopted with the combination of feature extraction and a customized version of output space filling (OSF) adaptive sampling in order to get optimal training set during offline phase. Partial Least Squares (PLS) and Support Vector Regression (SVR) have been exploited for feature extraction and prediction technique respectively to have robust and accurate real time inversion during online phase.
Complete list of metadatas

https://hal.archives-ouvertes.fr/hal-01768554
Contributor : Andrea Massa <>
Submitted on : Tuesday, April 17, 2018 - 12:10:34 PM
Last modification on : Saturday, August 3, 2019 - 1:37:52 AM

Identifiers

  • HAL Id : hal-01768554, version 1

Citation

Shamim Ahmed, Marco Salucci, Roberto Miorelli, Nicola Anselmi, Giacomo Oliveri, et al.. Real time groove characterization combining partial least squares and SVR strategies: application to eddy current testing. International Workshop on New Computational Methods for Inverse Problems (NCMIP 2017), Jun 2017, Cachan, France. ⟨hal-01768554⟩

Share

Metrics

Record views

414