Neuro-fuzzy modeling and solution of magnetic field inverse problem

Abstract

The present paper intended to identify the locations of piecewise linearly connected shape of a carrying current wire and its current magnitude based on data of three-axis magnetic sensor array mounted either on a surface or inside a volume. At first, we tested a few classical numerical computation methods and the least square method. The results indicated that these methods were incommensurate for this application. Then, we applied a locally linear neuro-fuzzy (LLNF) model with tree learning for modelling and identification of a linear piece of current. Based on sensor array locations; we proposed nine scenarios, a few of which include superficial array, volumetric array and a mix of superficial and volumetric arrays. For each sensor array model, a neuro-fuzzy structure identification based on the mean square criterion was carried out to select the best number of locally linear models. We found that LLNF model with volumetric and mix sensor array identified the position coordination and magnitude of a linearly shaped current very well. Finally, we applied one of the best arrays and locally linear modelling to solve this magnetic field inverse problem for piecewise linearly connected shapes of current. Our method for solving this inverse problem proved successful.