A novel predictive architecture for microwave-assisted drying processes based on neural networks

Abstract

In this contribution, a novel learning architecture based on the interconnection of two different learning-based neural networks has been used to both predict temperature and drying curves and solve inverse modelling equations in microwave-assisted drying processes. In this way, a neural model that combines the accuracy of neural networks based on Radial Basis Functions (RBF) and the algebraic capabilities of the matrix polynomial structures is presented and validated. The architecture has been trained by temperature (Tc(t)) and moisture content (Xt(t)) curves, which have been generated by a previously validated drying model. The results show that the neural model is able to very accurately predict both kind of curves for any combination of the considered input variables (electric field and air temperature) provided that an appropriate training process is performed. The proposed configuration also permits the solution of the inverse problem in the drying process by finding the optimal value for the electric field. This provides Tc(t) or Xt(t) curves that fit to a desired drying condition in a specific time slot.