UTILIZATION OF STACKED NEURAL NETWORK FOR PORE SIZE PREDICTION OF ASYMMETRIC MEMBRANE

This study, investigates the possibility of applying stacked artificial neural network (ann) as an alternative method to estimate the pore size of the asymmetric hollow fiber membranes. ann, a connectionist-based (black box) model, consists of layers of nodes with nonlinear basis functions and weighted connections that link the nodes. using the nodes and weights, the inputs are mapped to the outputs after being trained with a set of training data. the input data needed for training the ann model, the solute rejection and the permeation rate, are obtained from permeation experiments. since the number of experimental data points needed for training the ann model is limited, stacked neural network is utilized instead of the more common and simple feedforward ann. with the development of this ann model, the procedure to estimate membrane pore size was found to be easier and faster with a testing error of less than 2% compared to the experimental data