PREDICTION OF PROCESS AND PRODUCT PARAMETRERS IN DEEP BED DRYING OF ROUGH RICE USING ARTIFICIAL NEURAL NETWORK

The objective of this study was to predict the performance indices in deep bed drying of Isfahan rough rice (Sazandegi variety) using artificial neural networks (ANNs). In our experiments, the effects of air temperature, air velocity, and air relative humidity on product output rate (POR) as an indicator of work capacity dryer, evaporation rate (ER) as a quality index of drying kinetics, and kernel cracking (KC) percentage as a criterion of dried product quality were investigated. To create training and test patterns, drying experiments were conducted using a laboratory dryer in deep bed mode. The desired parameters for various input variables were calculated using physical and thermodynamics relations. To predict the dependent parameters, three well-known networks namely multi layer perceptron (MLP), generalized feed forward (GFF), and modular neural network (MNN) were examined. Four learning algorithms consisted of step, momentum, conjugate gradient, and Levenberg-Marquardt (LM) were also used for the training purpose of the networks. The GFF network provided superior results than those achieved by MLP and MNN networks. Among several examined topologies and activation functions for the GFF network, the 3-8-7-3 topology and the hyperbolic tangent function revealed the best results. A remarkably high degree of prediction accuracy was achieved by the resulting GFF neural network, with a normalized mean square error (NMSE) of only 0.00865, mean absolute error (MAE) of 0.97514 and Spearman correlation coefficients (r) of 0.9912. It was concluded that the ANN could be an effective method to model Isfahan rough rice drying process.