Prediction of the overall sieve tray efficiency for a group of hydrocarbons, an artificial neural network approach

Mass transfer efficiency is of great importance in the design and analysis of the sieve tray columns, as it relates theoretical to the actual number of stages. However, the empirical models that have so far been presented for the estimation of the tray efficiency are either confined to a particular system or not sufficiently accurate. On the other hand, neural networks are utilized in cases where either mathematical modeling could not be applied or the relationships between the parameters are complex. Therefore, it is the aim of this research to utilize neural network in predicting the overall sieve tray efficiency. To obtain this objective, the overall sieve tray efficiency for eight different compositions (i.e., ethanol/water, acetone/water, methanol/water, acetic-acid/water, toluene/water, methyl-isobutyl-ketone (i.e., MIBK)/water, aniline/nitrobenzene and cyclo-hexane/n-heptane) as a single hydrocarbon system has been computed, using artificial neural network. To assess the performance of the technique, the predicted values of the neural networks have been compared with the experimental data and the correlation proposed by the Garcia and Fair (Garcia and Fair, 2000). The findings of this research reveal that there exist a mean absolute error of 1.21 percent which is negligible compared to the correlation presented by Garcia and Fair with absolute error of 18.22 percent. Therefore, the results of this work demonstrate that multi-layer perceptron neural network could provide a good practice of predicting the overall sieve tray efficiency and with a good degree of accuracy.