Abstract

Terminal velocity (TV) is one of the important aerodynamic properties of materials, including seeds of agricultural crops that are necessary to design of pneumatic conveying systems, fluidized bed dryer and cleaning the product from foreign materials. Prior attempts to predict TV utilized various physical and empirical models with various degrees of success. In this study, supervised artificial neural networks (ANN) were used for predicting TV. Experimentally, the TV of rice, chickpea, and lentil seeds were obtained as a function of moisture content and seed size. TV was significantly influenced by seed type, moisture content and seed size. Using a combination of input variables, a database of 54 patterns was obtained for training, verification and testing of ANN models. The results obtained from this study showed that the ANN models learned the relationship between the three input factors (seed type, moisture content and seed size) and output (TV) successfully, and described the TV of seeds with different shapes extremely well. The best 4-layer ANN model produced a correlation coefficient of 0.997 between the actual and predicted TV. The ANN models compared to mathematical models were able to learn the relationship between dependent and independent variables through the data itself without producing a formula. These benefits significantly reduce the complexity of modeling for TV.

Key words: Artificial neural networks, terminal velocity, prediction, back-propagation.