Machine Perception Based on Eddy Current for Physical Field Reconstruction of Conductivity and Hidden Geometrical Features

This paper presents a new machine perception method based on eddy-current (EC) effects to reconstruct physical fields (EC field, electrical-conductivity field, and hidden geometrical features) of a nonferrous material commonly encountered in intelligent manufacturing using one-time finite magnetic flux density (MFD) measurements. An analytical model for EC testing system with conductor discretization is established and expressed in state-space representation. Two improvements (physical constraints and adaptive element refinement) are developed and integrated into the system model. The measurement models of physical fields using discrete MFD measurements are linearly established, reducing the physical field reconstruction to a linear inverse problem for solving using the Tikhonov regularization method. The EC-based machine perception is numerically illustrated by reconstructing the EC density field, conductivity field, and hidden geometrical features of aluminum plates. Additionally, the effects of element refinement, physical constraints, and sensor configurations on the reconstruction results are analyzed numerically. Using an experimental prototype consisting of an air-cored electromagnet and a two-dimensional (2-D) array of magnetic sensors with associated electronics, the effectiveness of the machine perception method and the accuracy of the reconstructed physical field are demonstrated experimentally.