Journal of Membrane Science 287 (2007) 219229

Optimum design of reverse osmosis system under different feed concentration and product specication
Yan-Yue Lu a , Yang-Dong Hu a, , Xiu-Ling Zhang b , Lian-Ying Wu a , Qing-Zhi Liu a
a

Key Laboratory of Marine Chemistry Theory and Technology, Ministry of Education, College of Chemistry and Chemical Engineering, Ocean University of China, Qingdao 266003, China b Department of Chemistry, Dezhou University, Dezhou 253011, China Received 7 July 2006; received in revised form 16 October 2006; accepted 20 October 2006 Available online 25 October 2006

Abstract The design of various multistage RO systems under different feed concentration and product specication is presented in this work. An optimization method using the process synthesis approach to design an RO system has been developed. First, a simplied superstructure that contains all the feasible design in present desalination process has been presented. It offers extensive exibility towards optimizing various types of RO system and thus may be used for the selection of the optimal structural and operating schemes. A pressure vessel model that takes into account the pressure drop and concentration changes in the membrane channel has also been given to simulate multi-element performance in the pressure vessel. Then the cost equation relating the capital and operating cost to the design variables, as well as the structural variables of the designed system have been introduced in the objective function. Finally the optimum design problem can be formulated as a mixed-integer nonlinear programming (MINLP) problem, which minimizes the total annualized cost. The solution to the problem includes optimal arrangement of the RO modules, pumps, energy recovery devices, the optimal operating conditions, and the optimal selection of types and number of membrane elements. The effectiveness of this design methodology has been demonstrated by solving several seawater desalination cases. Some of the trends of the optimum RO system design have been presented. 2006 Elsevier B.V. All rights reserved.
Keywords: Reverse osmosis; Seawater desalination; Optimum design; Structure; Module model

1. Introduction Desalination of sea and brackish waters is the main source for supplying fresh water in the regions suffering from the scarcity of natural fresh water supplies. Reverse osmosis (RO) is the major technologies for large-size plants for desalinating water. Since 1960s, due to the development of new RO modules and membranes, RO has become an attractive process for both seawater and brackish water desalination [15]. The rapid growth of RO process is because it is able to produce fresh water with lower cost [3]. The other attractive feature of RO process is that the plant design and operation is simplicity and modularity. Membrane plants are often more compact, can be scaled up easily and installed more quickly than thermal separations plants. Also, it makes the maintenance of RO systems

Corresponding author. Tel.: +86 53266781875. E-mail address: ydhuhd@ouc.edu.cn (Y.-D. Hu).

easier [6,7]. Another advantage of the RO process is that it is able to meet varying feed water concentration and varying production water quantity and quality requirement through change system construction and operation condition. RO membrane manufacturers have developed various membrane types to precisely meet the varying need of a wide range of industrial, municipal, commercial and drinking water application, such as high ux, high rejection membranes and fouling resistant membranes, low pressure, high rejection membrane etc. [8,9,30]. All this advantages have made the design of RO process more exible. Considerable efforts for the research of the optimum RO system design have been made [6,1013]. Based on the state-space approach, El-Halwagi [14] developed a structural representation of RO networks. The RO networks were described using four boxes: a pressurization/depressurization streamdistribution box, a pressurization/depressurization matching box, a reverse-osmosis stream-distribution box, a reverseosmosis matching box. The function of the distribution boxes was to represent all possible combinations of stream splitting,

0376-7388/$ see front matter 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.memsci.2006.10.037

220

Y.-Y. Lu et al. / Journal of Membrane Science 287 (2007) 219229

mixing, bypass and recycle. The matching boxes locate all possible stream assignments to units. With this formulation, all possible structure arrangements could be represented. The mathematical model was formulated as a mixed integer nonlinear programming (MINLP). In the further work, Zhu and El-Halwagi [15] used El-Halwagis representation and included a factor for ux decline over time. Voros and Maroulis [16,17] simplied the El-Halwagis representation by reducing the distribution boxes to junctions. Consequently, the model was formulated as a nonlinear programming (NLP) by using a variable split ratio. They have considered the effect of various feed conditions and product requirements. Maskan and Wiley [18] used a directed graph and connectivity matrix to represent the RO networks superstructure. In the mathematical model of the superstructure, a variable reduction technique is performed to accelerate the computational process. Nemeth [19] studied the behavior of the ultra-low pressure RO membranes in the fullscale system and presented recommendations to improve system design. Van der Meer [20,21], and Wessels and Van der Meer [22] have developed a simplied mathematical model to optimize the performance of NF and RO membrane ltration plants. The study showed that the productivity of nanoltration plants can be signicantly improved by installing a reduced number of membrane elements serially in pressure vessels (PV) and changing system conguration. See et al. [23] used a systematic method to calculate the optimal cleaning schedule for a given RO network. Malek and Hawlader [24] provided the realistic economic model that relates the various operational and capital cost elements to the design variable values. Most of previous studies presented the general representation of RO networks, which contains all the possible designs that are to be considered as candidates for the optimal solution. Following their research efforts, this study presented a simplied superstructure representation that contains all the feasible design in present desalination process. The model of superstructure could be easier solved with less calculation time. A pressure vessel model has also been given. The model could be used for the optimum selection of types and number of membrane element, according to its performance characteristics, the prices, and the design requirements of a specic desalination application. Therefore the optimal design of RO systems was formulated as a mixed integer nonlinear programming problem (MINLP). The objective is to determine the optimal system structure and operating conditions for a given feed water concentration and water production requirement. The solution to the problem also includes the most appropriately choice of the types of membrane elements in each stages and the optimal number of membrane elements in each PV. 2. RO unit model 2.1. The mass transfer model of RO membrane There are numerous membrane performance models that have been introduced [7,25,26]. They are derived from different theories and all of them can be simplied to the solution diffusion model, as shown in Eqs. (1) and (2). For the RO system design

and optimization, it is necessary to adopt the appropriate modeling equations that can satisfactorily predict the membrane performance with reasonable computational complexity. Therefore the solution diffusion model is the one most commonly used in RO system design. The model is mainly based on two parameters: the water permeability, A, and the solute transport parameter, B. The values for these two parameters are usually specied by membrane manufactures. According to the model, the pure water ux, Jw (kg/m2 s), and the salt ux, Js (kg/m2 s), are given as follows: Jw = A Pf Pp Pf 2 (w p ) 106 (1) (2) (3) (4) (5)

Js = B(Cw Cp ) = 0.2641C(T + 273) 1.0 106 C Jw + J s p Js 1000 Vw

Vw = Cp =

where Pf , Pp (MPa) denote the feed pressure and the permeate pressure, respectively. Pf is the pressure drop in the membrane channel. w (MPa) is the osmotic pressure of the brine at the membrane wall concentration Cw (ppm), and p and Cp are corresponding variables for the permeate. p denotes the density of the permeate. Vw (m/s) is the permeate velocity. 2.2. The model for RO module In the practical RO process, a PV with up to eight membrane elements that are connected in series consists of a RO module. The concentrate of the rst element becomes the feed to the second, and so on. The product tubes of all elements are coupled and connected to the module permeate port [2022]. For every application a suitable hydraulic design can be made (28 serial elements), based on the actual situation. Maskan and Wiley [18] developed a model of the tubular membrane elements. Al-Bastaki and Abbas [25,26] have given the models of the spiral-wound and hollow-ber membrane elements, which took into account the pressure drop and concentration changes in the membrane channel. The PV performance can be approximately simplied to the performance of the membrane elements connected in series. Therefore, base on the models mentioned above, we have presented a pressure vessel model with spiralwound membrane element. The model could be used for the optimal selection of the types and numbers of membrane element. The equations of the model are shown as follow. Fig. 1

Fig. 1. Schematic diagram of a RO unit.

Y.-Y. Lu et al. / Journal of Membrane Science 287 (2007) 219229

221

indicates the schematic representation of a RO unit and the variable: Cw = Cp + C f + Cb Cp 2 eVw /K (6)

K (m/s) is the mass transfer coefcient, which can be calculated from empirical relations such as K = 0.04 Re0.75 Sc0.33 Re = Vd Ds d (7) (8)

where Re and Sc are the Reynolds and the Schmidt numbers and Ds is the solute diffusivity. d is the feed spacer thickness, the feed side solution density and is the water viscosity. V denotes the ow velocity that was calculated using the averaged values of the inlet and outlet ow rates in the membrane channel. For a spiral-wound membrane module, each of the permeate and feed side ows can be considered as a ow between two parallel plates with a length L, a width W and a spacing d. hence the pressure drop on the feed side can be calculated as follows: Pf = 0.0033QLpv Wd 3 (9) (10) (11)

Fig. 2. The process ow diagram of typical RO system.

Lpv = mLm Pf 0.35

where Q (m3 /h) is the ow rate that was calculated using the averaged values of the inlet and outlet ow rates in the membrane channel. Lpv and Lm denote the length of the PV and the length of a membrane element, respectively. The m is the number of the membrane elements in each PV. It determines the optimum selection of number of membrane element. The maximum allowable pressure drop of the pressure vessel is 0.35 MPa. The technical constraint is usually specied by membrane manufacturers. For the spiral-wound membrane element, the membrane width (W), can be calculated by the membrane area (Sm ) and the number of leaves (Nl ): Sm = WLm Nl (12)

operating conditions, and the optimal arrangement of the membrane elements. Several RO system congurations commonly used in seawater and brackish desalination applications have been investigated (shown as Fig. 2). In the single stage RO system with pressure exchanger (PX) type energy recovery device, the main high pressure pump is sized to approximately equal the RO permeate ow, not the full system feed ow. Therefore, the PX signicantly reduces the capital and operating cost. The multiple PX-arrays are similar to membranes arrays providing the operator with benecial redundancy [2729]. El-Halwagi [14,15] and Voros and Maroulis [16,17] have used the state space approach to develop the structural representation of RO network. Here we adopt and properly simplify the approach according to the practical RO process. Compared to the previous representation the characteristic differences in this paper are: 1. It was assumed that the RO network has only one input. The fresh-feed supply to stage 1. 2. The distribution boxes were reduced to junctions. The stream mixed at the pressurization stages and split at the RO stages. 3. The stream split ratio and isobaric-mixing constraints were used in the model of superstructure. These constraints simply the mathematical description of stream mixing and splitting. 4. The structural representation has included the pressure exchanger (PX) type energy recovery device. As shown in Fig. 3, a RO network consists of Nps pressurization stages and NRO reverse osmosis stages. The number of stream junctions employed is Nps + 2, where 2 indicated the brine and product streams leaving the network. Each stream node among the Nps nodes represents a stream connected to a pump. The streams pressurized by high pressure (HP) pump or not are connected to a corresponding RO stages. The RO stages consist of multiple parallel reverse osmosis pressure vessels operating at the same conditions. Each stream of the network (the brine and permeate streams leaving all reverse osmosis stages) may be linked to all the Nps + 2 nodes.

The total permeate ow rate of the nth pressure vessel, Qp,n (m3 /h), the total brine ow rate and concentration, Qb,n (m3 /h) and Cb,n (ppm), can be calculated from the mass and salt balance equations: Qp,n = 3600Vw Sm m Qf,n = Qb,n + Qp,n Qf,n Cf,n = Qb,n Cb,n + Qp,n Cp,n . 3. RO network representation 3.1. Problem description For a given specication of feed water, the design objective is to identify the most cost effective RO network conguration, the (13) (14) (15)

222

Y.-Y. Lu et al. / Journal of Membrane Science 287 (2007) 219229

NRO NRO

Qps,i Cps,i =
j=1

Qb,j xb,i,j Cb,j +

j=1

Qp,j xp,i,j Cp,j , (22)

i = 2, 3, . . . , Nps
Nps +2

xb,px,j +
i=1 Nps +2

xb,i,j = 1,

j = 1, 2, . . . , NRO

(23)

xp,i,j = 1,
i=1

j = 1, 2, . . . , NRO

(24)

Fig. 3. representation of the RO network via the superstructure.

3.2. Mathematical formulation The complete mathematical model that describes the superstructure is presented as follow by means of the appropriate relationships between the variables (mass balance equations, technical and operational constraints):
NRO NRO

where Qps,1 , Cps,1 denote the ow rate and concentration of the rst pressurization stage, Qf , Cf denote the feed ow rate and feed concentration of the RO network, Qps,i , Cps,i denote the ow rate and concentration of the ith pressurization stage, respectively. Qb,j , Cb,j denote the brine ow rate and concentration of the jth RO stage, Qp,j , Cp,j denote the permeate ow rate and concentration of the jth RO stage, respectively. xb,i,j , xp,i,j indicate the stream split ratios of the brine and the permeate, respectively. The values determine the ow rates of brine and permeate leaving the jth RO stage and being linked to the ith pressurization stage. Pf , Pps,1 are the feed pressure of the RO network and the inlet pressure of the rst pressurization stage, respectively. Qpx , Cpx is the ow rate and concentration of high pressure brine entering pressure exchanger, respectively. Following are the stream-mixing constraints. Whenever mixing occurs, it is necessary that only streams with equal pressure can be mixed together: (Pps,i Pb,j )Qb,j xb,i,j = 0, j = 1, 2, . . . , NRO (Pps,i Pp,j )Qp,j xp,i,j = 0, j = 1, 2, . . . NRO i = 1, 2, 3, . . . , Nps , (26) i = 1, 2, 3, . . . , Nps , (25)

Qps,1 = Qf +
j=1

Qb,j xb,1,j +
j=1

Qp,j xp,1,j

(16)

NRO

NRO

Qps,1 Cps,1 = Qf Cf +
j=1

Qb,j xb,1,j Cb,j +

j=1

Qp,j xp,1,j Cp,j (17) (18)

where Pps,i denotes the inlet pressure of the ith pressurization stage. Pb,j , Pp,j denote the brine and permeate pressure of the jth RO stage, respectively. The streams leaving the ith pressurization stages are correspondingly connected to the jth RO stages. Therefore the following equation can be obtained: QRO,j = Qps,i , j = i, j = 1, 2, 3, . . . , NRO j = i, j = 1, 2, 3, . . . , NRO j = i, j = 1, 2, 3, . . . , NRO (27) (28) (29)

Pps,1 = Pf
NRO

CRO,j = Cps,i , PRO,j = Pps,i ,

Qpx =
j=1

Qb,j xb,px,j
NRO

(19)

Qpx Cpx =
j=1

Qb,j xb,px,j Cb,j

(20)

NRO

NRO

Qps,i =
j=1

Qb,j xb,i,j +
j=1

QP,j xp,i,j ,

i = 2, 3, . . . , Nps (21)

where QRO,j , CRO,j , PRO,j denote the feed ow rate, concentration, and operation pressure of the jth RO stage, respectively. Pps,i is the outlet pressure of the ith pressurization stage. The mathematical models that predict the performance of each RO stage have been presented in detail in the previous section. These model equations relate the ow rate and concentration of the brine and permeate leaving an RO stage to the ow rate, concentration, and pressure of the stream entering the stage. The arrays of pressure vessels (PV) with 28-membrane elements per PV consist of a RO stage. In this paper, the optimal

Y.-Y. Lu et al. / Journal of Membrane Science 287 (2007) 219229

223

PV structure has been researched. Four different types of spiral wound FilmTec reverse osmosis membrane elements have been considered. According to its performance characteristics and the design requirements of a specic desalination application, the optimum selection of types and numbers of the membrane element employed in each PV can be determined by the following equations:
4

4. Solution methodology The optimization design problem is formulated as a mixedinteger nonlinear programming (MINLP) for minimizing the total annualized cost subject to thermodynamic, technical, and exibility constraints. The total annualized cost (TAC) of the system consists of two terms: annual operating cost (OC) and annualized capital cost (CC). The annual operating cost includes the energy cost (E) necessary for pumps, the cost of membrane module maintenance (OCm ). The annualized capital cost is for the initial membrane module, pumps and pressure exchanger. The objective function is presented as follows: TAC = (CCin + CChpp + CCpx + CCbp + CCm )1.411 0.08 + OCin + OChpp + OCbp + OCm (41) (42) (43) (44)
NRO

Aj =
k=1 4

Zj,k Ak ,

j = 1, 2, 3, . . . , NRO

(30)

Bj =
k=1 4

Zj,k Bk ,

j = 1, 2, 3, . . . , NRO

(31)

Zj,k 1,
k=1

j = 1, 2, 3, . . . , NRO

(32)

CChpp = 52( PQhpp )0.96 CCpx = 3134.7Q0.58 px Qps,1 = Qhpp + Qpx

NRO 4

In the present work, it was assumed that the membrane element employed in the PV in the same RO stage are the same. The membrane intrinsic properties (A, B) are assumed to be constant. Zj,k is the binary variable. It takes the value of 1 when the kth element type is utilized in the PV in the jth RO stage. Otherwise, it takes the value of 0. The overall material balances for the RO network and a set of product quantity and quality constraints concerning the minimum desirable product ow rate, and the maximum allowable product concentration are presented as follows: Qf = Qb + Qp Qf Cf = Qb Cb + Qp Cp
NRO

Cm =
j=1 k=1

Zj,k Ck mj,k nj +
j=1

Cpv nj

(45)

OChpp = PX =

PQCe fc 3.6hpp motor (PQ)out 100% (PQ)in

(46) (47)

(33) (34) (35)

Qb =
j=1

Qb,j xb,j + Qpx

NRO

Qb Cb =
j=1 NRO

Qb,j xb,j Cb,j + Qpx Cpx

(36)

Qp =
j=1

Qp,j xp,j
NRO

(37)

Qp Cp =
j=1

Qp,j xp,j Cp,j

(38) (39) (40)

Qp Qp min Cp Cp max

where Qb , Cb are the ow rate and concentration of the brine leaving the RO network, respectively. Qp , Cp are the ow rate and concentration of the product water, respectively. xb,j , xp,j are the outlet stream split ratios. Qp min refers to the minimum desirable product ow rate, Cp max refers to the maximum allowable product concentration.

where CCin , CChpp , CCbp , CCpx is the capital cost of the seawater intake pump, the high pressure pump, the booster pump, and the pressure exchanger (PX), respectively. OCin , OChpp , OCbp is the energy cost necessary for these pumps. Their functions refer to the papers [2,6,15,24]. The main high pressure pump capacity equals the full feed ow of the RO system minus the brine ow entering the PX. The required capacity of the PX and the booster pump should equal the brine ow entering the PX. Cm denotes the total membrane module cost. Ck is the price of the kth membrane element and Cpv is the price of the pressure vessel. The nj is the number of pressure vessel employed in the jth RO stage. 1.411 is the coefcient that used to calculate the practical investment. 0.08 is the capital charge rate. Ce is the cost of electricity and fc is the load factor. hpp , motor , px are the efciency of the HP pump, the electric motor, and the PX. The terms of capital cost items take into account the number and type of RO modules utilized for each RO stage, and the number and capacity of each pump and PX. The existence of the specic device may be determined indirectly by operational variables, such as the input ow rate to the unit, the pressure of the pressurization stage or RO stage. The structural optimization may take place in terms of eliminating all unnecessary pressurization stages or RO stages. This procedure is carried out by introducing an excessive number of units as an initial guess, while at the optimum certain design variables, such as stream split ratios, are either set to zero or to a value that indicate the absence or presence of the specic stage.

224

Y.-Y. Lu et al. / Journal of Membrane Science 287 (2007) 219229 Table 2 The parameters for calculation Feed water temperature, T ( C) Average brine density, (kg/m3 ) The brine viscosity, (kg/m s) Diffusion coefcient, Ds (m2 /s) High pressure pump efciency, hpp Pressure exchanger efciency, px Electric motor efciency, motor The cost of electricity, Ce ($ (kWh)1 ) The pressure vessel cost (estimation) ($) 25 1020 1.09 103 1.35 109 75% 90% 98% 0.08 1000

The MINLP can be solved using the software GAMS. It solves the problem by decomposing it into a series of nonlinear (NLP) sub-problems and mixed integer (MIP) master problems. Several starting points are used to obtain the best possible solution. 5. Case study The proposed methodology for RO system optimization design has been applied to the investigation of the operational and structural characteristics of multi-stage RO system. Four different types of FilmTec reverse osmosis membrane elements from DOW have been included in the design studies of the current work. They are the low energy, high productivity element SW30XLE-400, the high rejection, high productivity element SW30HR-380, the high rejection, fouling resistant element SW30HR-320, and the high productivity, high rejection brackish water RO element BW30-400. The product specications and the membrane transport properties (membrane pure water permeability, solute transport parameter) of these elements are given in Table 1 [30]. The parameters for calculation, which are given based on Refs. [6,15,24], are listed in Table 2. In general, solving the mathematic programming problem take between 0.27 and 0.84 CPU time (s), 134 and 856 iterations to nish. 5.1. The study of varying feed composition Several cases were solved, in which the feed water concentration vary while the product demand and quality constraints maintain at same levels. The total production required is 120 m3 /h, and the maximum allowed salt concentration is 500 ppm. For the relatively higher feed concentrations varying from 38,000 to 48,000 ppm, the results of the RO system optimization design were presented in Table 3. As indicated in the table, the simple one stage structure is favored (shown as Fig. 4) in the range of concentration. The number of elements in each PV is a continuous variable in the present work. With increasing feed concentration the values increase lightly. Therefore, they are all take the value of 5 after rounding. In stage 1, the number of PV is 40 after rounding. The selected type of membrane element is the SW30XLE-400 that offers very high productivity and rejection, enabling the lowest total cost of water for seaTable 1 Characteristics of FilmTec spiral wound reverse osmosis membrane elements Element type Active area Length of the element (in.) (mm) Diameter of the element (in.) (mm) Feed space (mil)a Feed ow rate range (m3 /h) Permeate ow (gpd) (m3 /d) Stabilized salt rejection (%) Maximum operating pressure (psig) (MPa) Pure water permeability constant, A (kg/m2 s Pa) Salt permeability constant, B (kg/m2 s) Membrane element cost ($) (estimation)
a

Fig. 4. The optimum RO system for feed concentration varying from 38,000 to 48,000 ppm.

water desalination. In this range of concentration, increasing the operating pressure is the main method to meet the product constraints. For the medium feed concentrations varying from 20,000 to 35,000 ppm, the design results were presented in Table 4. The two-stage structure in which the brine coming from stage 1 enter fully to stage 2 (xb,1,2 = 1) have been identied in this range of concentration (shown as Fig. 5). Because the feed con-

Fig. 5. The optimum RO system for feed concentration varying from 20,000 to 35,000 ppm.

SW30XLE-400 (m2 ) 400 (37.2) 40 (1016) 7.9 (201) 28 0.816 9000 (34.1) 99.70 1,200 (8.3) 3.5 109 3.2 105 1200

SW30HR-380 380 (35.3) 40 (1016) 7.9 (201) 28 0.816 6000 (22.7) 99.70 1,200 (8.3) 2.7 109 2.3 105 1000

SW30HR-320 320 (29.7) 40 (1016) 7.9 (201) 34 0.814 6000 (22.7) 99.75 1,200 (8.3) 3.1 109 2.2 105 1400

BW30-400 400 (37) 40 (1016) 7.9 (201) 34 0.819 10,500 (40) 99.5 600 (4.5) 7.5 109 6.2 105 900

(ft2 )

1 mil = 0.0254 mm.

Y.-Y. Lu et al. / Journal of Membrane Science 287 (2007) 219229 Table 3 Design and optimization results for the feed concentrations varying from 38,000 to 48,000 ppm Feed concentration, Cf (ppm) 48,000 Process ow System feed ow, Qf (m3 /h) System overall recovery Product concentration, Cp (ppm) Membrane type in stage 1 Number of elements per PV in stage 1 Number of PV in stage 1 Operating pressure in stage 1, P1 (MPa) Unit product cost ($/m3 ) One stage system, Fig. 4 264 45% 380 SW30XLE-400 5 40 8.1 0.58 45,000 One stage system, Fig. 4 264 45% 360 SW30XLE-400 5 40 7.7 0.56 42,000 One stage system, Fig. 4 264 45% 330 SW30XLE-400 5 40 7.3 0.54 38,000

225

One stage system, Fig. 4 264 45% 300 SW30XLE-400 5 40 6.7 0.52

Table 4 Design and optimization results for the feed concentrations varying from 20,000 to 35,000 ppm Feed concentration, Cf (ppm) 35,000 Process ow System feed ow, Qf (m3 /h) System overall recovery Product concentration, Cp (ppm) Membrane type in stage 1 Membrane type in stage 2 Number of elements per PV in stage 1 Number of elements per PV in stage 2 Number of PV in stage 1 Number of PV in stage 2 Operating pressure in stage 1, P1 (MPa) Operating pressure in stage 2, P2 (MPa) Unit product cost ($/m3 ) Two stage system, Fig. 5 191 63% 300 SW30XLE-400 SW30XLE-400 2 5 29 20 7.3 8.3 0.5 30,000 Two stage system, Fig. 5 185 65% 270 SW30XLE-400 SW30XLE-400 3 5 29 18 6.7 7.7 0.47 25,000 Two stage system, Fig. 5 176 68% 240 SW30XLE-400 SW30XLE-400 3 6 27 16 5.9 6.8 0.43 20,000 Two stage system, Fig. 5 170 71% 200 SW30XLE-400 SW30XLE-400 3 6 26 15 5.3 6.2 0.4

centration of stage 2 increases, In order to meet the product quantity requirement, it is necessary to pressurize the brine coming from stage 1 and to increase the number of elements in each PV in stage 2. The selected type of membrane element is the SW30XLE-400 in both stages. For the lower feed concentrations (300016,000 ppm), the design results were presented in Table 5. In the range of concentration, the multi-stage RO systems with brine re-processing are favored and the selected types of membrane element are the BW30-400. With decreasing feed concentration, the structures employed during the design vary from two-stage to three-stage with brine recycle. It leads to the result that the overall recovery ratio increase and the unit product cost decrease. When the concentration of feed stream is 12,000 ppm, there is a bypass into the second stage for the brine coming from the rst unit of stage 1 (shown as Fig. 6), while the others enter the second unit of

stage 1 straightly (xb,1,2 = 0.526). The brine bypass is useful to dilute the feed stream of stage 2. As indicated in Figs. 6 and 7, the stage 1 can be made of more unit blocks in series. The feed pumps between units were omitted. The costs for the high pressure pump are not smooth functions of the ow rate and operating pressure. In the present study, the pump cost is about a given range of ow rate and operating pressure. If a different pump cost model is used, some pumps could be eliminated from the optimal structure. For the feed stream of concentration 3000 ppm, the optimal RO structure has been shown as Fig. 8. The brine leaving stage 3 partly recycle (xb,3,3 = 0.233), while the others come into the PX. In this case, there is another solution closing to the optimal one. The calculation results indicated that the structure employed in the suboptimum solution is two-stage arrangement (shown as Fig. 7), its unit product cost is $0.23 and the overall recovery is 83%. The unit cost is a little larger than that of the optimal one,

Fig. 6. The optimum RO system for feed concentration 12,000 ppm.

Fig. 7. The optimum RO system for feed concentration 6000 ppm.

226

Y.-Y. Lu et al. / Journal of Membrane Science 287 (2007) 219229

Table 5 Design and optimization results for the feed concentrations varying from 3000 to 16,000 ppm Feed concentration, Cf (ppm) 16,000 Process ow System feed ow, Qf (m3 /h) System overall recovery Product concentration, Cp (ppm) Membrane type in stage 1 Membrane type in stage 2 Membrane type in stage 3 Number of elements per PV in stage 1 Number of elements per PV in stage 2 Number of elements per PV in stage 3 Number of PV in stage 1 Number of PV in stage 2 Number of PV in stage 3 Operating pressure in stage 1, P1 (MPa) Operating pressure in stage 2, P2 (MPa) Operating pressure in stage 3, P3 (MPa) Unit product cost ($/m3 ) Two stage system, Fig. 5 171 70% 250 BW30-400 BW30-400 3 5 26 15 3.9 4.5 0.32 12,000 Two stage system, Fig. 6 160 75% 200 BW30-400 BW30-400 2 6 34 12 3.6 4.2 0.29 6000 Two stage system, Fig. 7 145 83% 120 BW30-400 BW30-400 2 6 35 7 2.7 3.3 0.24 3000 Three stage system, Fig. 8 140 86% 60 BW30-400 BW30-400 BW30-400 3 3 5 22 12 8 2.0 2.3 2.4 0.22

but its structure is relatively simple. Therefore, the suboptimum solution may be better for the case from operative point of view. In general, for different feed concentration (varying from 3000 to 48,000 ppm), the design results (shown as Tables 35) indicate that the unit product cost is proportional to the feed concentration. With increasing raw feed concentration the overall recovery ratio decrease and the product water quality deteriorate. For processing higher feed concentration, the high operating pressure is necessary. However, the number of elements and pressure vessels employed in RO system is less than that of lower feed concentration. It indicated that the arrangement and operation of RO system are more exible when the operating pressure is relatively lower. When the high operating pressure is necessary, the simple one stage structure is more favored. The optimization of RO networks were affected by feed concentration and permeate recovery per module. As shown in the Figs. 48, the transition from one structure to the other is continuous when the feed concentration changes. With decreasing feed concentration, the structural schemes vary from single to three-stage with brine bypass and recycle. For the feed concentration of 12,000 ppm, there is a brine bypass in the optimum structure (shown as Fig. 6). The feed stream of the stage 2 could be diluted with the brine bypass stream. With decreasing feed concentration, the brine bypass ratio decreases constantly. In the vicinity of 6000 ppm, the bypass disappeared. The feed concentration of stage 2 still meets the requirement although without

the bypass stream. For lower feed concentration (3000 ppm), a brine recycle around stage 3 that leads to higher overall recovery appears in the optimum structure (shown as Fig. 8). Although the brine recycle will increase the feed concentration of stage 3, the brine concentration coming from stage 2 is relatively lower, and the third stage can still produce permeate required. 5.2. The study of varying product composition For different values of the product quality constraints varying between 300 and 50 ppm, the optimization of the superstructure has been carried out at a constant raw feed stream of concentration 35,000 ppm and the product requirement of 120 m3 /h. The design results are presented in Table 6. The optimal structure employed in the design is strongly dependent on the required product concentration. For the looser permeate concentration requirement (300 ppm), two-brine staging conguration (shown as Fig. 5) is favored that increase the system recovery ratio. For the product quality requirement of concentration 200 ppm, the optimal RO structure changes to the single stage (shown as Fig. 4). For the lower permeate quality requirement of concentration 100 ppm, the three-stage structure with permeate re-processing and brine recycle have been identied (shown as Fig. 9). The permeate leaving stage 1 mostly split into the stage 2 (xp,1,2 = 0.852), the other come to the product output. The feed stream of stage 3 include the full brine coming from stage 2 and

Fig. 8. The optimum RO system for feed concentration 3000 ppm.

Fig. 9. The optimum RO system for feed concentration 35,000 ppm, the product concentration 100 ppm.

Y.-Y. Lu et al. / Journal of Membrane Science 287 (2007) 219229 Table 6 Design and optimization results for the product concentrations varying from 300 to 50 ppm Feed concentration, Cf (ppm) 35,000 Process ow System feed ow, Qf (m3 /h) System overall recovery Product concentration, Cp (ppm) Membrane type in stage 1 Membrane type in stage 2 Membrane type in stage 3 Number of elements per PV in stage 1 Number of elements per PV in stage 2 Number of elements per PV in stage 3 Number of PV in stage 1 Number of PV in stage 2 Number of PV in stage 3 Operating pressure in stage 1, P1 (MPa) Operating pressure in stage 2, P2 (MPa) Operating pressure in stage 3, P3 (MPa) Unit product cost ($/m3 ) Two stage system, Fig. 5 191 63% 300 SW30XLE-400 SW30XLE-400 2 5 29 20 7.3 8.3 0.5 35,000 One stage system, Fig. 4 264 45% 200 SW30XLE-400 35,000 Three stage system, Fig. 9 295 41% 100 SW30XLE-400 BW30-400 BW30-400 7 8 7 45 19 29 5.8 0.84 0.84 0.74 35,000

227

40

7.1

0.52

Three stage system, Fig. 10 300 40% 50 SW30XLE-400 BW30-400 BW30-400 7 8 7 46 22 34 5.8 0.84 0.84 0.78

the partly recycle brine leaving stage 3 (xb,3,3 = 0.867). For the tighter permeate quality requirement of concentration 50 ppm, the structure employed in the design is similar to the previous three-stage structure (shown as Fig. 10). The split ratio of permeate entering to the stage 2 (xp,1,2 ) has increased up to 0.978, and the recycle ratio of brine entering to the stage 3 (xb,3,3 ) has slightly decreased. With decreasing concentration of the product, the overall system recovery ratio decrease, the unit cost of the product increase, and the feed stream of stage 2 varies from brine to permeate. Therefore the selected types of membrane element also vary from the SW30XLE-400 to BW30-400. Consequently, the required operating pressure has decreased. For the product requirement of concentration 100 ppm, the calculation results indicated that the permeate coming from the stage 1 (concentration 400 ppm) is re-processed in stage 2 (shown as Fig. 9). The permeate concentration and the brine concentration leaving stage 2 are 10 ppm and 670 ppm, respectively. Then, the permeate concentration leaving stage 3 increase up to 70 ppm because of a brine re-circulation around stage 3. The product water consist of the permeate coming from each stage. The proportion of the permeate coming from stage 3 is the largest, while that coming from stage 1 is the smallest. Therefore, the brine re-circulation around stage 3 is useful in increasing the overall recovery ratio and making the product concentration meet to the requirement.

For the case that the product water concentration is 50 ppm, there is another solution without recycle closing to the optimal one. The structure is shown as Fig. 11. The calculation results indicated that the unit product cost of the suboptimum solution is $0.86, the overall recovery is 32%. The cost is 10% higher than that of the optimal one. Mathematically, the optimum RO structure is the one shown in Fig. 10, of which the unit product cost is the lowest. However, current RO systems are normally operated at two-stage or three-stage structure without recycle stream because of the operational and engineering limit to the performance of multi-stage system. Therefore, the three-stage (shown as Fig. 11) or even two-stage structure may be better from operative point of view. 5.3. The effect of unit costs For a typical seawater desalination process, which has a xed feed concentration (35,000 ppm) and product concentration requirement (300 ppm), the effect of unit costs on the design results were shown in Table 7. Three types of membrane element (SW30XLE-400, SW30HR-380, SW30HR-320) were included in the present study, and the unit price of each was assumed to be 1200, 1000, 1400 $, respectively. The table shows that the unit product cost is very sensitive to electricity price. When the electricity price increase (0.080.12 $ (kWh)1 ), the unit product cost increase 28%.

Fig. 10. The optimum RO system for feed concentration 35,000 ppm, the product concentration 50 ppm.

Fig. 11. Another solution closing the optimal one for feed concentration 35,000 ppm, the product concentration 50 ppm.

228

Y.-Y. Lu et al. / Journal of Membrane Science 287 (2007) 219229

Table 7 Sensitivity analysis for electricity, membrane element unit prices Electricity price ($) (kWh)1 [membrane element price ($)] 0.05 (1200, 1000, 1400) Operating pressure in stage 1, P1 (MPa) Operating pressure in stage 2, P2 (MPa) The number and type of membrane employed in stage 1 The number and type of membrane employed in stage 2 Overall recovery Unit product cost ($/m3 ) 8.1 9.2 52, SW30XLE 82, SW30XLE 67% 0.4 0.12 (1200, 1000, 1400) 6.8 7.8 98, SW30XLE 92, SW30HR-320 59% 0.64 0.08 (1200, 1000, 1400) 7.3 8.3 58, SW30XLE 100, SW30XLE 163% 0.5 0.08 (1600, 1400, 1800) 7.6 8.6 54, SW30XLE 85, SW30XLE 64% 0.52 0.08 (800, 600, 1000) 7.0 8.1 68, SW30XLE 132, SW30HR-380 62% 0.48

While a 33% increase in membrane price results in only 4% increase in unit product cost. When the electricity price decrease (0.080.05 $ (kWh)1 ), or the membrane price decrease (1200, 1000, 1400800, 600, 1000 $), the lower product cost can be achieved by applying high operating pressure or increasing the membrane area. However, the product cost is more sensitive to changes in electricity price (operating cost) than membrane price (capital cost). Increasing operating pressure achieves higher overall recovery and lower unit product cost than increasing membrane area. Thus, reduction in operating costs is more important than reduction in capital cost in decreasing product cost. The electricity is the major cost component. It dominates because of the high pressure of the optimal designs. 6. Conclusion The optimum design of RO systems under different feed concentration and product specication has been studied in this work. A simplied superstructure that contains all the feasible design in present desalination process has been presented. It offers extensive exibility towards optimizing various types of RO system and thus may be used for the selection of the optimal structural and operating schemes. A pressure vessel model has also been given. It could be used for the optimal selection of types and number of membrane elements, according to its performance characteristics, the prices, and the design requirements of a specic desalination application. The design task has been formulated as an MINLP, which seeks to minimize the total annualized cost of the RO system while considering the thermodynamic, modeling, economic, and feasibility constraints. The objective is to determine the optimal system structure and operating conditions for a given feed water concentration and water production requirement. The solution to the problem also includes the most appropriately choice of the types of membrane elements in each stages and the optimal number of membrane elements in each PV. The effectiveness of this design methodology has been demonstrated by solving several seawater desalination cases. The result indicated that it is possible to obtain lower product water cost by using this method presented.

Nomenclature A B C Ce Ck Cm Cpv Cw CCbp CChpp CCin CCpx d Ds Js Jw K Lpv Lm m nj Nl OCin OChpp OCbp P Pf Q Qp,n Re Sc Sm T TAC Vw W water permeability (kg/m2 s Pa) solute transport parameter (kg/m2 s) concentration of solute (ppm) electricity cost ($/(kWh)) the price of the kth membrane element ($) membrane module cost ($) the price of the pressure vessel ($) concentration at the membrane wall (ppm) capital cost of the booster pump ($) capital cost of the high pressure pump ($) capital cost of the seawater intake pump ($) capital cost of the pressure exchanger ($) the feed spacer thickness (m) the solute diffusivity (m2 /s) salt ux (kg/m2 s) water ux (kg/m2 s) the mass transfer coefcient (m/s) length of the pressure vessel (m) the length of a element (m) the number of membrane elements in each PV the number of pressure vessel employed in the jth RO stage the number of leaves in a membrane element energy cost of the intake pump ($) energy cost of the high pressure pump ($) energy cost of the booster pump ($) operating pressure (MPa) the pressure drop in the membrane channel ow rate (m3 /h) the total permeate ow rate of the nth pressure vessel (m3 /h) Reynolds number Schmidt number the membrane area per element (m2 ) temperature ( C) total annualized cost ($) the permeate velocity (m/s) membrane width (m)

Y.-Y. Lu et al. / Journal of Membrane Science 287 (2007) 219229

229

xb,i,j

xp,i,j

Z Greek w

the stream split ratio of the brine leaving the jth RO stage and being linked to the ith pressurization stage the stream split ratio of the permeate leaving the jth RO stage and being linked to the ith pressurization stage binary integer pump efciency the water viscosity (kg/m s) osmosis pressure (MPa) osmosis pressure of the brine at the membrane wall (MPa) density (kg/m3 )

Subscripts b brine stream b, j the brine stream of the jth RO stage bp booster pump f feed stream hpp high pressure pump in intake seawater k the kth element type p permeate stream p,j the permeate stream of the jth RO stage ps, i the ith pressurization stage px pressure exchanger RO, j the jth RO stage

References
[1] E. El-Zanati, S. Eissa, Development of a locally designed and manufactured small-scale reverse osmosis desalination system, Desalination 165 (2004) 133139. [2] A. Villafala, I.M. Mujtaba, Fresh water by reverse osmosis based desalination: simulation and optimization, Desalination 155 (2003) 113. [3] N.M. Wade, Distillation plant development and cost update, Desalination 136 (2001) 312. [4] M. Wilf, C. Bartels, Optimization of seawater RO systems design, Desalination 173 (2005) 112. [5] M. Wilf, Design consequences of recent improvements in membrane performance, Desalination 113 (1997) 157163. [6] M.G. Marcovecchio, P.A. Aguirre, N.J. Scenna, Global optimal design of reverse osmosis networks for seawater desalination: modeling and algorithm, Desalination 184 (2005) 259271.

[7] A. Abbas, N.M. Al-Bastaki, Modeling of an RO water desalination unit using neural networks, Chem. Eng. J. 114 (2005) 139143. [8] J.A. Redondo, A. Casanas, Designing seawater RO for clean and fouling RO feed. Desalination experiences with the FilmTec SW30HR-380 and SW30HR-320 elements technical economic review, Desalination 134 (2001) 8392. [9] M. Busch, W.E. Mickols, Reducing energy consumption in seawater desalination, Desalination 165 (2004) 299312. [10] Y.Y. Lu, Y.D. Hu, Optimum design of reverse osmosis seawater desalination system considering membrane cleaning and replacing, J. Membr. Sci., in press. [11] Y.D. Hu, Y.Y. Lu, Cleaning strategy of the membrane modules in the reverse osmosis seawater desalination system, J. Chem. Indus. Eng. (China) 56 (2005) 499505. [12] I.M. El-Azizi, Design criteria of 10,000 m3 /d SWRO desalination plant of Tajura, Libya, Desalination 153 (2002) 273279. [13] G. Al-Enezi, N. Fawzi, Design consideration of RO units: case studies, Desalination 153 (2002) 281286. [14] M.M. El-Halwagi, Synthesis of reverse osmosis networks for waste reduction, AICHE J. 38 (1992) 11851198. [15] M. Zhu, M.M. El-Halwagi, Optimal design and scheduling of exible reverse osmosis networks, J. Membr. Sci. 129 (1997) 161174. [16] N.G. Voros, Z.B. Maroulis, Optimization of reverse osmosis networks for seawater desalination, Comp. Chem. Eng. 20 (1996) 345350. [17] N.G. Voros, Z.B. Maroulis, Short-cut structural design of reverse osmosis desalination plants, J. Membr. Sci. 127 (1997) 4768. [18] F. Maskan, D.E. Wiley, Optimal design of reverse osmosis module networks, AICHE J. 46 (2000) 946954. [19] J.E. Nemeth, Innovative system designs to optimize performance of ultra-low pressure reverse osmosis membranes, Desalination 118 (1998) 6371. [20] W.G.J. Van der Meer, C.W.A. Averink, Mathematical model of nanoltration systems, Desalination 105 (1996) 2531. [21] W.G.J. Van der Meer, M. Riemersma, Only two membrane modules per pressure vessel hydraulic optimization of spiral-wound membrane ltration plants, Desalination 119 (1998) 5764. [22] L.P. Wessels, W.G.J. Van der Meer, Innovative design of nano- and ultraltration plants, Desalination 119 (1998) 341345. [23] H.J. See, V.S. Vassiliadis, D.I. Wilson, Optimisation of membrane regeneration scheduling in reverse osmosis network for seawater desalination, Desalination 125 (1999) 3754. [24] A. Malek, M.N.A. Hawlader, Design and economics of RO seawater desalination, Desalination 105 (1996) 245261. [25] N.M. Al-Bastaki, A. Abbas, Predicting the performance of RO membranes, Desalination 132 (2000) 181187. [26] N.M. Al-Bastaki, A. Abbas, Modeling an industrial reverse osmosis unit, Desalination 126 (1999) 3339. [27] PX, Energy Recovery, Inc., San Leandro. [28] R.L. Stover, Development of a fourth generation energy recovery device, Desalination 165 (2004) 313321. [29] G. Migliorini, E. Luzzo, Seawater reverse osmosis plant using the pressure exchanger for energy recovery, Desalination 165 (2004) 289298. [30] Membrane Technical Information, http://www.dow.com/liquidseps/ service/lm techinfo.htm.