Experiments and numerical simulations of supercritical fluid extraction for Hippophae rhamnoides L seed oil based on artificial neural networks

In this paper, a supercritical fluid extraction setup with an extraction volume of one liter was established with which Hippophae rhamnoides L seed oil was extracted using supercritical CO(2.) The experiments show that many factors have an impact on the oil yield, such as extraction pressure, temperature, and fluid flow rate, as well as seed particle size and filling quantity. For the extraction process of H. rhamnoides L seed oil, the recommended conditions were as follows: extraction pressure of 20-30 MPa, extraction temperature of 35-40 degrees C, supercritical CO(2) flow rate of 0.15-0.3 m(3)/h, and extraction time of 4-5 h. Under such conditions, the oil obtained is very lucid and of good quality, and the yield can be greater than 90%. Gas chromatography analysis shows that the oil contains 12.3% saturated fatty acid and 87.7% unsaturated fatty acid. From the changes of oil yield with the extraction time, it can be concluded that the extraction process contains three stages: the fast extraction (line), transitional, and slow extraction stages. At the first stage, 75-80% of the oil has been extracted. On the basis of the experimental results, artificial neural network technology was applied to the simulation of the supercritical fluid extraction of vegetable oil. With a three-layer back-propagation network structure, the operation factors, such as pressure, temperature, and extraction time, were used as input variables for the network, and the oil yield was used as the output value. In the optimization of the topological structure of the net, six neurals from the middle hidden layer have been proven to be the optimum value, according to the minimum training and running time. With the normalization pretreatment of the initial input data, not only the convergence speed and accuracy has been improved greatly but also the problem of the derivative at zero has been solved. Therefore, the method is better than that of Fullana. On the basis of simplification of the extraction process, taking account of axial dispersion, a differential mass balance kinetic model has been proposed. With Matlab software as platform, an artificial neural networks-supercritical fluid extraction simulation system has been programmed. For the first time, the simulation for the supercritical fluid extraction process of H. rhamnoides L seed oil has been made, and the results show that the average absolute relative deviation is lower than 6%.