The objective of this research was to predict head rice yield (HRY) in fluidized bed dryer using artificial neural network approaches. Several parameters considered here as input variables for artificial neural network affect operation of fluidized bed dryers. These variables include: air relative humidity, air temperature, inlet air velocity, bed depth, initial moisture content, final moisture content and inlet air temperature. In aggregate, 274 drying experiments were conducted for creating training and testing patterns by a laboratory dryer. Samples were collected from dryer, and then dehulling and polishing operations were done using laboratory apparatus. HRY was measured at several different depths , average of which was considered as HRY for each experiment. Three networks and two training algorithms were used for training presented patterns. Results showed that the cascade forward back propagation algorithm with topology of 7- 13-7-1 and Levenberg-Marquardt training algorithm and activation function of Sigmoid Tangent predicted HRY with determination coefficient of 95.48% and mean absolute error 0.019 in different conditions of fluidized bed paddy drying method. Results showed that the input air temperature and final moisture content has the most significant effect on HRY.

Keywords: Feed-forward back propagation network, Head rice yield, Levenberg-Marquardt, Rough rice.

Full-Text [PDF 533 kb]   (4201 Downloads)