Modelling and analysis of supercritical fluid extraction using soft computing based approaches
Supercritical fluid extraction is an environmental friendly separation technique which exploits the solvent properties of fluid near the critical point, and is used in food science, chemical industry and petroleum industry. Modelling of the solubility of biomaterials is an essential issue in supercritical fluid extraction, and can help to develop a simulation and auto control system for quality control in industrial production to combine technical, economic and environmental aspects. In this thesis, soft computing approaches are used to develop the simple and proper models for supercritical fluid extraction. Firstly, a radial basis function networks model and a hybrid model that combines the radial basis function networks and Peng-Robinson equation of state model are proposed. Secondly, a hybrid dynamic genetic algorithms and Peng-Robinson equation of state model for supercritical fluid extraction is designed to recognize the change of temperature during the extraction process. Thirdly, a novel neuro-fuzzy model for supercritical fluid extraction is proposed as an extension to the neural networks models. This neuro-fuzzy model can provide accurate prediction of the extraction process under a wide range in pressure and temperature. Simulation studies show that results using the proposed models are generally better than those using the conventional Peng-Robinson equation of state method. Finally, based on the studies above, a novel neuro-fuzzy approach is proposed to model an industrial process, extraction of the [beta]-carotene and lycopene by supercritical CO2 extraction from tomato waste, which can help to achieve predictive control and improve productivity.