Study of the influence of membrane structure and permeation conditions on the efficiency of separation of miscible liquid mixtures by pervaporation

Pervaporation separation of aqueous ethanol solution has, for the first time ever, been investigated with natural rubber latex (NRL) base membranes which contained a hydrocolloid as a blend ingredient. Three different hydrocolloids viz. methyl cellulose, carboxy methyl cellulose (sodium salt) and alginic acid (sodium salt) of low or medium or high molecular mass were used and tested. The weight percent of a hydrocolloid in the blended layer of a membrane has been varied from 1.25 to 20 on a dry rubber basis. The composition of ethanol in the aqueous feed solution was varied within the range of 5 to 96 weight percent of ethanol. The temperature of operation was fixed in the range of 20 °C to 75°C. Fourier Transform Infrared / Attenuated Total Reflectance (FT-IR/ ATR) spectra and Scanning Electron (SE) micrographs have been used to study the distribution of hydrocolloids within the membrane. Morphological features of the cross section of a blended layer have been used to develop a probable mechanism of water transfer through the membrane. Water selectivity has been found to depend on the type, the molecular mass and the weight percent of the hydrocolloid used in the membrane. Both FT-IR/ATR and SEM techniques have proved that a high molecular mass hydrocolloid distributes itself uniformly throughout the membrane. Both techniques have, independently, shown that a low molecular mass hydrocolloid will be situated at or near the top surface of the membrane. A very strong link between a good distribution of the polymer bridged clusters of rubber particles within the membrane and the maximum increase in water selectivity has been established. For the first time ever, artificial neural networks have been used for the modelling and prediction of the pervaporation separation performance of NRL base membranes. Quantitative data about the distribution of hydrocolloids within the membranes was needed in order to train the neural net models. The correlation coefficients and rms errors between the predicted and experimental results were found to be greater than 0.89 and less than 0.086 respectively