Hidden security threat identification: Computing an ML library using FEM accelerated by a reduced order model

Abstract

This work presents computational results of a reduced order model for the rapid calculation of conducting object characterisations as a function of frequency in metal detection. We present a brief description of the eddy-current model and the magnetic polarizability tensor (MPT) used for our object characterisations. The transmission problem required for the computation of the MPT and its discretisation is then described followed by a summary of the reduced order model. As an illustration of the capabilities of the approach for characterising realistic objects, we show MPT spectral signatures of a kitchen knife along with examples of other threat and non-threat objects.