Prediction of Dispersed Phase Holdup in the Kühni Extraction Column Using a New Experimental Correlation and Artificial Neural Network

In this work, the dispersed phase holdup in a Kuhni extraction column is predicted using intelligent methods and a new empirical correlation. Intelligent techniques, including multilayer perceptron and radial basis functions network are used in the prediction of the dispersed phase holdup. To design the network structure and train and test the networks, 174 sets of experimental data are used. The effects of rotor speed and the flow rates of the dispersed and continuous phases on the dispersed phase holdup are experimentally investigated, and then the artificial neural networks are designed. Performance evaluation criteria consisting of R2, RMSE, and AARE are used for the models. The RBF method with R2, RMSE, and AARE respectively equal to 0.9992, 0.0012, and 0.9795 is the best model. The results show that the RBF method well matches the experimental data with the lowest absolute percentage error (2.1917%). The rotor speed has the most significant effect on the dispersed phase holdup comparing to the flow rates of the continuous and dispersed phases.