Dynamic hybrid model for ultrafiltration membrane processes

Membrane processes are an interesting alternative for water treatment because these are compact and they require low energy consumption and low or no chemical load. However, the performance of membrane processes are sensitive to water conditions and the operation deteriorates with time. For this reason, operational strategies for physical or chemical membrane cleaning are applied in order to prolong system operation. For control and design purposes, mathematical models that describe the impact of these inputs on the membrane performance are required. In this work, a hybrid model was developed for this purpose, combining Darcy's law and a neural network. The experimental data used are from the wastewater treatment of a petrochemical process. Two strategies of physical cleaning, back-shock and backwashing, are considered in the hybrid model. Additionally, the model can predict the behavior of the ultrafiltration process without cleaning. Hence, the hybrid model can be used for the optimization and the control of ultrafiltration processes under several strategies of operation.4