Comparison of mathematical models and artificial neural network for prediction of moisture ration of orange slices during drying process

In present study, the thin- layer drying of orange slices in a laboratory scale hot-air dryer has been modeled. Drying experiments were conducted at three different temperatures of 50, 60 and 70°C, and two air velocities of 1.0 and 2.0 m/s. The statistical results of data showed the change of drying temperature and air velocity had significant effects on moisture ratio (p<0.05) but interaction effect of air velocity and temperature had insignificant effect on moisture ratio. Based on the results, the minimum moisture ratio of dried orange slices was obtained 5.3% when the dryer air temperature and velocity were 70°C and 2.0 m/s, respectively. After the end of experiments, the experimental data were fitted to the 7 well-known drying models. According to fitting results, Page’s model with determination coefficient R2-3 showed better performance to predict the moisture ratio. Also, this study used a feed forward back propagation neural network in order to estimate orange slices moisture ratio, based on the temperature, air velocity and time as input variables. In order to design this model, two main activation functions called tanh and purlin, widely used in engineering calculations, were applied in hidden and output layer, respectively. The artificial neural network with 3-20-1 topology and Levenberg-Marquardt training algorithm provided best results with the maximum determination coefficient (0.9994) and minimum Root of Mean Square Error (1.009×10-3) values. The results indicated the artificial neural network model was more accurate than Page’s model for prediction of moisture ratio of orange slices during drying process.