Intelligent Real Time Prediction of Moisture Content using Artificial Neural Network in Pistachio Thin Layer Drying

In this paper, real time prediction of pistachio nuts moisture content (MC) during drying, based on previous moisture contents using artificial neural network have been studied. In order to obtain experimental data, thin layer drying experiments were conducted at four air temperatures (25, 40, 55, 70ºC), three air velocities (0.5, 1 and 1.5 m/s) and two relative humidities (5, 20%). Moisture contents of pistachio nuts (Ohadi variety) were measured during drying in these conditions. The drying air temperature had the greatest effect and air velocity and relative humidity had a small effect on the drying kinetics of pistachio nuts. Effective diffusivity of water varied from 5.42×10-11 to 9.29×10-10 m2/s over the temperature range studied. Moisture content signals were considered as time series to predict MC based on special combinations of previous times MCs using three modified neural networks. The neural network with the best performance to predict MCs, which has 11 nerons in the hidden layer for every learning rate, was obtained. The best obtained neural network can be trained during drying in real time. Maximum root of mean squared error (RMSE) 0.23 and maximum variance of prediction errors 0.13 were obtained. This neural network can predict the moisture content of fourth point in drying process based on three previous points and when the process reaches to the predicted point and the real value of moisture content is measured, the difference between predicted and real value is calculated and according to the neural network structure and its output and the defined error function the neural network is trained and the training process will be continued as mentioned above until the drying process will finish. The results show that the obtained neural network is a very good predictor that learns the process of drying and its parameter will be approximately fixed after a few times the process begins. The derived predictor can predict moisture contents with high accuracy except at a few primary times of drying process in which the predictor is trained and no control is applied to the drying process. The predictor can be used either as a part of nonlinear real time control systems or as a differentiator of MCs signal.