Prediction of performance indices and optimal parameters of rough rice drying using neural networks

Abstract An artificial neural network was developed for rough rice drying to predict six performance indices: energy consumption, kernel cracking, final moisture content, moisture removal rate, drying intensity and water mass removal rate. Four drying parameters: rice layer thickness, hot airflow rate, hot-air temperature and drying time were the inputs of the neural network. After evaluating a large number of trials with various neural network architectures, the optimal model is a four-layered back-propagation neural network, with 8 and 5 neurons in the first and the second hidden layers, respectively. The effectiveness of the proposed model is demonstrated using experimental data. The mean relative error varied from 2·0 to 8·3% for six predictions with an average of 4·4%. Using a multiple-objective programming for optimisation of the drying parameters, the optimal values are rice layer thickness of 66 cm, hot airflow rate of 0·30 m s −1 , hot-air temperature of 93°C and drying time of 23 min.