Prediction of ternary gas permeation through synthesized PDMS membranes by using Principal Component Analysis (PCA) and fuzzy logic (FL)

Abstract(#br)Implementation of try and error method for membrane preparation procedures is a time and cost consuming technique. This study tries to present a novel idea to make membrane preparation procedure heuristic. Applying this method, helps researchers to predict performance of a membrane prior to its preparation . At first, a number of membranes are prepared and characterized. Then, their measured separation properties are used for prediction of performance of a membrane before its preparation. Furthermore, after preparation of each new membrane, its relevant data can be added to the data bank of the model to improve its capability for the next predictions. Therefore, this model will be improved step by step, after each new preparation. Fuzzy logic-based (FL) model and Principal Component Analysis (PCA) were employed to predict permeability of C 3 H 8 , CH 4 and H 2 in ternary gas mixtures using a membrane gas separation module. Based on Placket–Burman (P–B) design, eight different polydimethylsiloxane (PDMS) membranes were synthesized using different preparation conditions, solvent concentration, crosslinker concentration, catalyst concentration, type (or boiling point) of solvent, stirring time and synthesis times in ambient temperature and in an oven at 80\u003cce:hsp sp=\"0.25\"/\u003e°C, and their gas permeation properties were investigated. In an innovating procedure, effects of operating conditions, including feed temperature, pressure, flow rate, C 3 H 8 and H 2 concentration, as well as preparation conditions on the permeability of gasses through the synthesized membranes were investigated. Basically, in order to develop a FL model to predict permeability of gasses through all the synthesized membranes, synthesis and operating conditions should be considered, simultaneously, and this extends dimensionality of the problem. In all engineering problems, as the number of variables increases, the corresponding data matrix extends. To overcome the problem, PCA method was randomly used for seven of the prepared membranes from P–B design, and it was shown that the first four principal components could describe almost all of the variation in the data matrix. This means that the dimensionality of the problem reduced from 12 to 4. Using the first four principal components, a Sugeno type FL inference system was trained and applied to predict permeability of gasses. FL modeling results showed that there is an excellent agreement between the experimental data and the predicted values, with mean squared relative error (MSRE) of less than 0.0095.(#br)The developed model was used for the 8th membrane and its ability to predict separation parameters of this membrane was confirmed.