Fast Eddy Current Forward Models Using Artificial Neural Networks

Eddy current testing is a widely used nondestructive evaluation (NDE) technique in which flaw information is extracted from the impedance change of a coil placed above a metal testpiece. Typical applications of eddy current NDE are the inspection of heat-exchanger tubes in steam generators of nuclear power plants and detection of hidden corrosion in the lap-splices of aircraft skins. To obtain quantitative information about flaw size and shape, we would like to have a forward model which is able to predict the impedance change of a coil for different flaws in the test geometry. Analytical solutions exist for simple test geometry and flaws with good symmetry properties. However, for flaws with irregular shapes in complex geometry, an analytical solution usually is not available so we must find a numerical solution. There have been several numerical models in the literature, e.g., the finite element method [1], the boundary element method [2], and the volume integral method [3–5]. Those numerical models can be used in a wide range of applications with moderately complex geometry. However, numerical models are inherently computational intensive and thus are not suitable for applications in which modeling speed has the first priority. One application of a fast forward model is to build fast eddy current simulators which can be used for educational purpose. Another application of the fast forward model is in the solution of the nonlinear inverse problem in which a large number of forward solutions must be compute