Indirect simultaneous eddy-current measurements of subsurface profiles of electrophysical properties of planar objects using apriori knowledge about them

Abstract

An eddy-current method of simultaneous indirect measurements of distributions of electrical conductivity and magnetic permeability in the subsurface zone of planar objects is proposed, based on a surrogate optimization algorithm using neural network metamodels of reduced dimensionality. Reduction of their dimensions and the space for finding an extremum is performed using the Kernel PCA method, which involves nonlinear transformations as a result of computational operations with the Gaussian kernel function. The construction of metamodels involved the use of deep learning methods. The peculiarities of metamodels include the performance of two functions, in particular, providing low-cost efficient computing and accumulating additional a priori information about the measurement process, which is digitally entered into the design of the experiment determining the training samples for training of deep neural networks. Taken as a whole, it made it possible to achieve higher accuracy characteristics of indirect measurements.