3-D Eddy-Current Imaging of Metal Tubes by Gradient-Based, Controlled Evolution of Level Sets

Abstract

Abstract-Eddy-current non-destructive testing is widely used to detect defects within a metal structure. It is also useful to characterize their location and shape provided that proper maps of variations of impedance which the defects induce are available. Imaging of void defects in the wall of a hollow, nonmagnetic metal tube, is performed herein by controlled evolution of level sets. Such data are variations of impedance collected by a circular probe array close to the inner surface of the tube when a coil source operated at one single frequency is set along its axis at some distance from the array, both receiver and coil source being moved simultaneously. The defect zone is represented in implicit fashion as a zero level set, amenable to topological changes via a nonlinear iterative method that minimizes a least-square cost functional made of the difference between the measured (computer simulated) and model data. The procedure involves the rigorous calculation of the gradient of the variations of impedance, in the case of a multi-static configuration (driver and receiver coils are different), a vector domain integral field formulation being used to that effect. Numerical examples, via a dedicated extension of the generalpurpose CIVA platform, exhibit pros and cons of the approach for inner, outer, and through-wall void defects, with further comparisons to results provided by an independently-developed binary-specialized method.

Level Set Methods for Structural Inversion and Image Reconstruction

“…Three representative breast structures are displayed in the three images of the left column, where the value at each pixel of the images represents the physical parameter "static relative permittivity." 1 Three images from microwave breast screening. The three images are synthetically generated from MRI breast models.…”

“…It follows that a descent velocity is now given as v.x/ D F ' 1 .x/n.x/ with with grad b J being defined in (85). In (98), for each x 2 1 , two adjacent points of grad b J j S C and grad b J j S contribute to the value of F ' 1 .x/ which can be found in the normal direction to 1 …”

Metamodel-based Markov-Chain-Monte-Carlo parameter inversion applied in eddy current flaw characterization

“…Upon running a MCMC algorithm, a series of samples {x (1) , x (2) , · · · , x (K) } is available, K is the total number of iterations in Fig. 1.…”

“…Statistical methods are usually transformed into an optimization problem solved by a numerical optimization algorithm, as in [1,14,16,37,47]. Due to the complexity of the problem, statistical methods usually need to overcome two hurdles: high computational cost due to many evaluations of the forward model, non-convex objective function due to nonlinear parameter dependence.…”

“…Due to the complexity of the problem, statistical methods usually need to overcome two hurdles: high computational cost due to many evaluations of the forward model, non-convex objective function due to nonlinear parameter dependence. Gradient-based algorithms are widely used [1,14,16,37,47]. Yet, to calculate or approximate the gradient subject to the unknown parameters might not always be feasible.…”