Abstract
An adaptive sampling method based on simplex-mesh refinement is proposed for creating a database that facilitates the inverse problem solution in NDT. The resulting database is used as a generic interpolator of the forward problem. In order to obtain optimal sampling for the purpose of interpolation, physically based mesh generation technique relying on spring analogy is utilized. The method is tested with a four-parameter crack reconstruction problem. It is demonstrated that the optimized database provides high interpolation quality comparing to regular grid databases.