Measurements of near-surface radial profiles of electrophysical characteristics of cylindrical objects by the eddy current method using a priori data

Abstract

A new multiparameter express method for eddy-current measurement of radial near-surface profiles of electrophysical parameters of cylindrical control objects with a priori accumulation of information about them is proposed. The method combines in-situ measurements and model calculations using high-performance computing technologies of artificial intelligence based on neural networks, carried out both in advance in order to obtain specific information about objects, and directly in the process of performing measurements to quickly obtain a result. Mathematically, the method is based on the unique ability to quickly solve Maxwell's equations as a result of its approximation by deep neural networks without actually explicitly executing this solution. This allows deep learning to be used not only in the forward direction, but also in the opposite direction, i.e. apply to solve inverse measuring problems. The method is universal and can be extended to multiparameter measurement control with simultaneous additional determination of the diameter of a cylindrical object. The adequacy of the proposed method by numerical experiments is proved; examples of the implementation of all stages of its application are given. Algorithms and a complex of programs in the Python 3 environment have been created, which make it possible to practically implement the method. The profile measurement accuracy established on model calculations is characterized by maximum relative errors not exceeding 0.5%, provided that the probe signal is perfectly fixed. It is possible to generalize the use of the proposed method to similar eddy current measurements with surface probes of profiles of material parameters of flat objects.