Feasibility investigation of using artificial neural network in process monitoring of pumpkin air drying

In this paper, pumpkin cubes were dried by a laboratory scale convective hot air dryer. The drying process was carried out at four different temperatures (65, 75, 85 and 95 degrees C). After the end of drying process, initially, the experimental drying data were fitted with three well-known drying models. The results indicated that Newton model gave better results compared with other models to monitor the moisture ratio (MR) (with average correlation coefficient of determination, R-2=0.993). Also, this study used artificial neural network analysis (ANN) in order to feasibly predict dried pumpkins MR based on the time and temperature drying inputs. In order to do this project, two main activation functions including logsig and tanh that widely used in engineering calculations, were applied. Results showed that logsig activation function with 18 neurons in first hidden layer was selected as the best topology to predict MR. Comparison of the results obtained by ANN and classical modelling showed that artificial neural approach has a higher ability compared to classical modelling in predicting MR (R-2=0.9991 and 0.993, respectively).