Characterization of subsurface cracks in eddy current testing using machine learning methods

Abstract

The exploitation of nondestructive eddy current testing (NDT‐EC) has become a capital necessity. Therefore, the development of a fast tool for the eddy‐current signal inversion is necessary. This paper proposes an inversion of signals coming from the NDT‐EC sensor response using machine learning methods, to reconstruct the length and depth of the defect and obtain its geometric characterization by solving the inverse problem. In this context, a database comprising the impedance of the sensor‐cracked part system (constituting the crack signature) was constructed from a 3D finite element simulation and validated by an experimental companion. The machine learning algorithms were trained using this database. The results show that the defect can be quantified using these developed approaches. The numerical approach can replace the expensive experimental investigation or the optimization algorithm that has a prohibitive computing time. The results show that the approaches developed estimated the desired parameters of the crack with good precision. These different machine learning methods for solving the inverse problem were implemented using the MATLAB software.

Numerical Model of Eddy Current Inspection with DC Magnetic Field Associated

“…09) [19]. For the linear case, when the coil is excited by a sinusoidal wave, the equation 09 can be reduced to: (10) Where:…”

“…From the industry point of view, it is relevant to have a dedicated simulation tool to reliably represent real inspection situations in a virtual environment. In this way, several works have addressed methodologies of computational analysis aiming at the validation of experiments and customization of ECT sensors [8][9][10][11], resulting though in a few FEM commercial software for the standard ECT application. However, conventional ECT generally works with low-intensity magnetic fields and for advanced variations of the technique, where external DC magnetic fields can be applied to locally decrease the magnetic permeability, there are no FEM packages available to deal with such a nonstandard model.…”

Fast and efficient inversion methods for crack sizing using Bz magnetic signature

“…The rotating uniform eddy current (RUEC) probe, in Reference 9, has proven its reliability to address the NDT practical problem of unknown or unpredictable defect orientation by producing a rotating electromagnetic field on the workpiece. Referring to the experimental results aimed to detect artificial cracks at arbitrary angles in underwater structures using the rotating ACFM in Reference 10, the validation of the RUEC probe is carried out.…”

“…Helifa et al 21 have proposed a methodology that couples a 3D finite element method (FEM) with neural network data inversion to achieve a full characterization of the surface crack. in Reference 22, a comparative study presents different machine learning methods, namely: ANN, deep learning and fuzzy inference systems to estimate the subsurface cracks of an aluminum specimen. The double gradient fusion algorithm (DGFA) is developed, in Reference 23, to reconstruct the irregular crack morphology using simulation and experimental results.…”

Enhancement of crack reconstruction through inversion of eddy current testing signals with a new crack model and a deterministic optimization method

“…For crack sizing based on inverse analysis of ECT signals, a suitable crack model properly fitting the real crack with less crack parameter is the foundation in the crack reconstruction. Up to date, the crack models of rectangle [6][7][8] and semi-elliptical shape [9][10][11] are adopted to parameterize the crack to be reconstructed with ECT signals. However, these simplified crack models have a shortage of not exactly matching the profile of actual cracks such as fatigue cracks and stress corrosion cracks.…”