M. Asafi; R. Meamar Dastjerdi; M. Noshad
Abstract
Introduction In recent years, with increasing population growth and improving livelihoods, the consumption of vegetable oils has been increasing and has led to an increase in the level of oilseed cultivation. Sesame (Sesamum indicum L.) is an economically important crop which is widely cultivated all ...
Read More
Introduction In recent years, with increasing population growth and improving livelihoods, the consumption of vegetable oils has been increasing and has led to an increase in the level of oilseed cultivation. Sesame (Sesamum indicum L.) is an economically important crop which is widely cultivated all over the world. Sesame has been considered as an oil plant for cultivation in Iran's climatic conditions recently. Sesame contains about 58-44% oil, 18-25% protein and 13.5% carbohydrate. Sesame is grown mainly in the developing tropical and subtropical areas of Asia, Africa. The three countries of China, India and Myanmar are accounted as the largest producers of this product in the world. Screw pressing is the most reliable method for extracting oil from oilseed grains. This method is simpler than others and is more efficient in terms of cost and food security. The general objective of this research was to investigate the effects of rotational speed, temperature, type of screwing and die diameter on the amount of oil extraction from sesame oil and prediction of oil extraction using artificial neural network and compare to regression models. Materials and Methods In this research, a sesame oil extractor machine was designed and manufactured. Various experiments were carried out to determine the amount of oil extracted based on variable parameters such as the shape of the press screw, the rotational speed, the temperature and the diameter of the die. The experiment was performed at three levels of press screw type (constant pitch, variable pitch and conical), temperature (30, 60, 90), three levels of rotational speed (20, 50, 80 rpm) and three level of die diameter (6, 8, 10mm). The experimental design was factorial based on completely randomized design with three replications. The mathematical software (Matlab, 2012b) was used to determine the optimal neural network. The type of network was Multi-Layer Perceptron (MLP). In order to design this network, there were 3 neurons in the first layer (input), which was equal to the number of studied variable parameters (type of screw, rotational speed and temperature), the second layer was hidden layer, and the last layer (the output) had a neuron for the extracted oil) was equal to the number of outputs examined in this network. The Levenberg-Marquardt algorithm (LM) was used to train it, which is one of the fastest neural network training methods. The Second-order polynomial regressions were performed based on the step-by-step method and non-meaningful sentences were eliminated from the model. The accuracy of the models was determined by calculating the correlation coefficient and root mean square error (RMSE) indices. Results and Discussion The results of the experiments showed that the effect of type of press screw, rotational speed, extraction temperature and die diameter on the amount of oil extraction was significant (p≤0.01). The highest amount of extracted oil was obtained at conical press screw , rotational speed of 50 rpm, temperature of 60 °C and die diamter of 6 mm. An artificial neural network of three-layer perceptron and regression models were used to predict the amount of sesame oil extracted. The results showed that the artificial neural network model (1-8-3) with a correlation coefficient of 97.47% and a RMSE of 0.65 compared to linear regression and quadratic regression models had the higher efficiency in predicting the amount of extracted oil. Conclusion In this study, the effect of temperature, rotational speed, press screw type and die diameter on the amount of extracted oil were investigated. The results of this study showed that the change in the type of screw, rotational speed, diameter of die and temperature on the amount of extracted oil was significant at 1% level. Results also showed that the artificial neural network method was more efficient than linear and second order regression methods.