Large-scale parallel computation for the reconstruction of natural stress corrosion cracks from eddy current testing signals

An inversion algorithm for the reconstruction of cracks from eddy current signals is developed in this study and applied to the profile evaluation of natural stress corrosion cracks that were found in steam generator tubes of a nuclear power plant. A crack is modeled as an assembly of small regions having conductivities inside so that eddy currents that flow across the cracks are considered. The conductivity of each region, which is assumed to be a discrete value, is reconstructed by means of the algorithm. Since the algorithm is based upon a tabu search that usually requires a large number of evaluating solution candidates, simulations are carried out on a supercomputer with the use of parallel computation using up to 128 CPUs so as to reconstruct the crack profiles within a reasonable computational time. It is demonstrated that the algorithm can estimate the profiles of the natural cracks with sufficient accuracy. The simulations also show that the algorithm is highly compatible with parallel computation. Additional simulations using other models of natural cracks are performed. Reconstructed profiles of the natural cracks, as a notch with zero conductivity, are very different from the true profiles, even though the reconstructed signals agree well with the measured values. This reveals that it is necessary to take the internal conductivity into consideration when dealing with natural cracks. (C) 2003 Elsevier Ltd. All rights reserved.