University of Tehran Press Iranian Journal of Biosystem Engineering 2008-4803 47 2 2016 08 22 Neuro-fuzzy and response surface modeling of osmotic dehydration of pomegranate arils Neuro-fuzzy and response surface modeling of osmotic dehydration of pomegranate arils 243 255 58774 10.22059/ijbse.2016.58774 FA Mohammad Ganjeh Seyed Mahdi Jafari Sajjad Ghaderi Journal Article 2013 11 18 In this research, pomegranate arils are dehydrated by osmotic dehydration in 40, 50, and 60 % sucrose ‎solutions and at 45, 55 and 65 degrees C‎‏ ‏‎ and Weight Reduction, Solids grain and Water Loss of the products ‎were measured at 60, 120 and 180 minutes of process. Osmotic dehydration processes was modeled by ‎combination of neural network and fuzzy logic techniques (Neuro-fuzzy) and response surface methodology. ‎For modeling, interpolation and increase of the data’s, fuzzy logic was used. By entering the obtained results ‎from fuzzy model into the neural network tool, the Feed-Forward-Back-Propagation network with the ‎topology of 3-8-3 and the correlation coefficient of 0.98344‎‏ ‏and mean square error of 0.02278‎‏ ‏with ‎application of Log-sigmoid transfer function‏ ‏‎(logsig) and Levenberg–Marquardt learning algorithm was ‎determined as the best neural model. Regression models created by response surface methodology by ‎correlation coefficient of 0.90 were also capable for prediction of response factors but in comparison with ‎Neuro-fuzzy models have a lower accuracy.‎ In this research, pomegranate arils are dehydrated by osmotic dehydration in 40, 50, and 60 % sucrose ‎solutions and at 45, 55 and 65 degrees C‎‏ ‏‎ and Weight Reduction, Solids grain and Water Loss of the products ‎were measured at 60, 120 and 180 minutes of process. Osmotic dehydration processes was modeled by ‎combination of neural network and fuzzy logic techniques (Neuro-fuzzy) and response surface methodology. ‎For modeling, interpolation and increase of the data’s, fuzzy logic was used. By entering the obtained results ‎from fuzzy model into the neural network tool, the Feed-Forward-Back-Propagation network with the ‎topology of 3-8-3 and the correlation coefficient of 0.98344‎‏ ‏and mean square error of 0.02278‎‏ ‏with ‎application of Log-sigmoid transfer function‏ ‏‎(logsig) and Levenberg–Marquardt learning algorithm was ‎determined as the best neural model. Regression models created by response surface methodology by ‎correlation coefficient of 0.90 were also capable for prediction of response factors but in comparison with ‎Neuro-fuzzy models have a lower accuracy.‎ https://ijbse.ut.ac.ir/article_58774_f4e8226fb6b4d49a865d94fe87df2f44.pdf