Artificial neural network for drying behavior prediction of paddy in developed chamber fluidized–bed dryer

After rice is harvested, it must be dried before its products can be stored. Therefore, this paper presented a novel and simple method for improving paddy drying process in a column fluidized bed dryer. Additionally, artificial neural network methods were applied to predict the drying behavior. The experiments were conducted under two different drying chamber characteristics (a conventional chamber and a chamber fitted with nozzle) and four different air flow velocities (3.03, 3.52, 4.12 and 4.85 m/s) at a drying air temperature of 60 °C. The results showed that the chamber fitted with the nozzle reduced the drying time by approximately 67, 52 and 38 % at the air velocity of 3.52, 4.12 and 4.85 m/s, respectively. The specified experimental conditions and the calculated moisture content of the paddy in this work were used as input and output data for the artificial neural network, respectively, to predict drying characteristics. The artificial neural networks were developed with various parameters, including activation function, sampling type, split ratio, number of neurons and epoch number. The optimal model provided the root mean squared error of 0.111 and the coefficient of determination of 0.999. The best prediction was observed in the model using rectifier activation function with a split ratio of 0.85, epoch number of 1800, and 50 and 90 neurons in the first and second hidden layers, respectively.