MODELING AND SOLVING THE INVERSE PROBLEM OF EDDY CURRENT TESTING BY THE OPTIMIZATION METHOD USING A SURROGATE MODEL OF REDUCED DIMENSIONALITY

A number of computer experiments have investigated the effectiveness in terms of accuracy of the method for simultaneously determining the distributions of electrical conductivity and magnetic permeability in the subsurface zone of planar conductive objects when modeling the process of eddy current measurement testing with surface probes. The method is based on the use of surrogate optimization, which involves the application of a high-performance neural network proxy model of a deep learning probe as part of the target quadratic function. The surrogate model acts as a carrier and storage of apriori information about the object and takes into account the influence of all the main factors that are significant in the formation of the probe output signal. To build the surrogate model, a computer-generated homogeneous design of experiment was created on quasi-Sobol`s sequences with low discrepancy rates, both for the three-dimensional case and for two-dimensional projections. The problems of the surrogate model's bulkiness and mitigation of the “curse of dimensionality” effect are solved by applying techniques for reducing the dimensionality of the design space based on the PCA algorithm using a singular value decomposition to the matrix obtained by drawing up a homogeneous design of experiment on modified LPτ quasi-Sobol`s sequences. The paper investigates the options for compromise solutions regarding the dimensionality of the PCA space and the accuracy of obtaining the desired material characteristic profiles by the optimization method. The main emphasis in the numerical modeling is placed on the study of the relationship between the dimensionality of the new PCA space and the accuracy of obtaining the desired parameters of the inverse problem by the optimization method. Acceptable options for making appropriate compromise decisions are determined. The results of modeling the inverse measurement problem indicate a sufficiently high accuracy of profile reconstruction.