A new predictive neural architecture for modelling electric field patterns in microwave-heating processes

Byline: J.L. Pedreno-Molina, J. Monzo-Cabrera, M. Pinzolas, M.E. Requena Perez In this work, a learning architecture based on neural networks has been employed for modelling the electric field pattern along an axis of a multimode microwave-heating cavity that contains dielectric materials. The multilevel configuration of this architecture, based on Radial Basis Functions (RBF) and polynomial structures, allows the fitting of the electric field as a function of the dielectric parameters (i.e. I[micro]*=I[micro]′−jI[micro]″) along one axis (x) of the cavity as well as inside the sample. In the learning stage, different samples have trained the neural architecture, by means of the mapping between (I[micro]′, I[micro]″) and the absolute value of the electric field pattern, generated with a 2D simulation platform based on the Finite Elements Method (FEM). The results obtained with conventional samples, such as polyester, epoxy, silicon crystal or beef steak, show that the proposed neural model is able to accurately predict the electric field spatial distribution under appropriate training processes.