(641b) Model Predictive Control of Feed Flow Reversal In a Reverse Osmosis Desalination Process

Reverse osmosis (RO) has emerged as one of the leading methods for water desalination due to the low cost and energy efficiency of the process [1]. Lack of fresh water sources has necessitated further development of these desalination plants, especially in areas with dry climates. In many reverse osmosis processes, particularly with brackish water feeds or systems running at a high level of recovery, dissolved ions can precipitate out of solution and crystallize on the membrane surface in a process called scaling. Scale formation on the membrane surface will lead to decreased permeate productivity, as well as permanent membrane damage if scaling is allowed to progress past its initial stages [2]. Several methods are currently used to prevent scale formation; addition of anti-scalant chemicals to the feed or flushing the membrane units with low-TDS (total dissolved solids) permeate water are two common procedures to accomplish this.

These current methods of scale mitigation have several disadvantages, anti-scalants can be expensive, and the addition of excess amounts can actually promote membrane scaling [3]. In the case of the permeate flush, this process will require the reverse osmosis operation to stop for a substantial amount of time to allow for the flushing cycle, eliminating any permeate production (even using up some of the previously produced permeate water). A novel technique called feed flow reversal has been developed which can prevent scale without the addition of expensive chemicals or extensive periods of system down-time [4]. This technique uses a system of actuated valves around the membrane modules configured specifically so that the direction of the feed flow through the membrane units can be reversed. This reversal of the feed flow also reverses the axial salt concentration profile at the surface of the membrane, effectively "resetting the induction clock" [4]. The reversal, if activated after crystals have already formed, also allows a substantial portion of scale deposited on the membrane surface to re-dissolve into solution. Model-predictive control algorithms are applied to a high capacity reverse osmosis system that utilizes feed flow-reversal in order to prevent and/or reverse scale crystal formation on the membrane surface. A dynamic non-linear model which incorporates feed concentration and membrane properties is used for simulation and control purposes [5]. Before flow reversal can take place on a high capacity RO plant, the flow into the membrane unit must be reduced to eliminate the risk of water hammer. A cost-function is formulated for the transition between the normal high flow steady-state operating point to a low flow steady-state operating point where it is safe to reverse the flow direction. Open-loop and closed-loop simulations demonstrate non-linear model-predictive control strategies that transition from the high-flow to low-flow steady-states in an optimal way while subjected to plant-model mismatch on the feed concentration, actuator constraints, and sampled measurements. These model-predictive control (MPC) strategies are compared to proportional integral (PI) control as well as a manually controlled transition.

References:

[1] A. Rahardianto, J. Gao, C. J. Gabelich, M. D. Williams, and Y. Cohen. High recovery mem-

brane desalting of low-salinity brackish water: Integration of accelerated precipitation softening with membrane RO. Journal of Membrane Science, 289:123-137, 2007.

[2] D. Hasson, A. Drak, and R. Semiat. Inception of CaSO4 scaling on RO membranes at various

water recovery levels. Desalination, 139:73-81, 2001.

[3] A. Rahardianto, W. Shih, R. Lee, and Y. Cohen. Diagnostic characterization of gypsum scale

formation and control in RO membrane desalination of brackish water. Journal of Membrane

Science, 279:655-668, 2006.

[4] N. Pomerantz, Y. Ladizhansky, E. Korin, M.Waisman, N. Daltrophe, and J. Gilron. Prevention

of scaling of reverse osmosis membranes by "zeroing" the elapsed nucleation time. part i.

calcium sulfate. Ind. & Eng. Chem. Res., 45:2008-2016, 2006.

[5] C. W. McFall, P. D. Christofides, Y. Cohen, and J. F. Davis. Fault-tolerant control of a reverse osmosis desalination process. In Proceedings of 8th IFAC Symposium on Dynamics and Control of Process Systems, volume 3, pages 163-168, Cancun, Mexico, 2007.