Modeling Drying Kinetics of Pistachio Nuts with Multilayer Feed-Forward Neural Network

Drying kinetics of pistachio nuts (Akbari v.) was simulated using a multilayer feed-forward neural network (MFNN). Experiments were performed at five drying air temperatures (ranging from 40 to 80°C) and four input air flow velocities (ranging from 0.5 to 2 m/s) with three replicates in a thin-layer dryer. Initial moisture content in all experiments was held at about 0.3 kg/kg d.b. To find the optimum model, various multilayer perceptron (MLP) topologies, having one and/or two hidden layers of neurons, were investigated and their prediction performances were evaluated. The (3-8-5-1)-MLP, namely, a network having eight neurons in the first hidden layer and five neurons in the second hidden layer resulted in the best-suited model estimating the moisture content of the pistachio nuts at all drying runs. For this topology, R2 and MSE values were 0.9989 and 4.20E-06, respectively. A comparative study among MFNN and empirical models was also carried out. Among the empirical models, the logarithmic model, with MSE = 7.29E-6 and R2 = 0.9982, gave better predictions than the others. However, the MFNN model performed better than the Lewis, Henderson and Pabis, two-term, and Page models and was marginally better than the logarithmic model.