Abstract
We have developed ANFIS (adaptive neuro-fuzzy inference system) modeling strategy for investigations on molecular separation using membrane systems. The results can be useful for purification and isolation of pharmaceutical molecules during manufacturing, and also for purification of water effluents. The developed methodology is based on combination of machine learning model which is ANFIS in this case and also computational fluid dynamics (CFD). The integration between two models is built through prediction of drug concentration in the system via both method in which the CFD results are considered as inputs to the ANFIS model. In order to analyze the existing data set, which has 2 inputs r and z and one output C, we used the optimized ANFIS model. A total of over 8 thousand data rows are included in this data set which are produced using the CFD simulation of membrane separation. The ANFIS model has been optimized using three population-based methods, namely Genetic Algorithm (GA), Firefly Algorithm (FA), and Bat Algorithm (BA), in order to achieve desired results. The optimization of hyper-parameters with all three techniques has had significant results and all have shown an R2 score higher than 0.99. In terms of MSE, the error rate decreased to 3.907 and the RMSE optimized to 1.02E+01.
[Display omitted]
•Computational simulation of drug separation using membranes.•ANFIS simulation of molecular separation via polymeric membranes.•Optimization of ANFIS parameters using three optimization methods.•Combination of CFD and ANFIS in membrane simulation.