Abstract
Stirred rotating disk membrane modules (RDMM) have been found to provide maximum flux enhancement due to minimization of the concentration polarization layer by shear effect. But unfortunately rigorous theoretical modeling of RDMM has not been developed yet, which greatly affects the simulation and optimization of this device under applications in many fields. In this perspective, the work reported in this article was carried out to develop an efficient artificial neural network (ANN) capable of predicting permeate flux in case of RDMM. Different ANNs, like Feedforward, Elman, Hopfield and Radial Basis Network, were investigated to find the optimum network configuration and corresponding network parameters have been estimated. As a case study, the focus was given to ultrafiltration (UF) of casein whey, a complex protein mixture. For this purpose the UF of whey was carried out in RDMM at different pH and transmembrane pressures, with and without rotating the membrane. A detailed comparative study was made on the validation results between the predicted flux and corresponding experimental values for different ANNs investigated. Based on the results, it was concluded that the Radial Basis Network was most accurate in terms of flux prediction and stability for UF using RDMM.