Energy 158 (2018) 373e382

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Practical limit of energy production from seawater by full-scale

pressure retarded osmosis
Minseok Kim, Suhan Kim*
Department of Civil Engineering, Pukyong National University, 45 Yongso-ro, Nam-gu, Busan, 48513, South Korea

a r t i c l e i n f o a b s t r a c t

Article history: Pressure retarded osmosis (PRO) produces energy using the salinity gradient between two solutions
Received 26 December 2017 (draw solution (DS) and feed solution (FS)). Net energy production (NEP) of PRO was analyzed using a
Received in revised form module-scale model developed in this work. The NEP analysis determines net energy from PRO by the
3 May 2018
difference between energy production by turbine and energy consumption by DS, FS, and booster pumps.
Accepted 8 June 2018
Especially, the effects of system capacity and membrane fouling on NEP are investigated using a module-
Available online 14 June 2018
scale modeling approach for the first time. The maximum net specific energy (NSE) per PRO system
capacity (sum of DS and FS flow rates) is close to 0.1 kWh/m3 without pretreatments. The maximum NSE
Keywords:
Pressure retarded osmosis (PRO)
decreases at smaller system capacities, and it becomes around 0.03 kWh/m3 from a PRO system with
Module-scale model 520 m3/d as capacity. NSE from seawater decreases in the presence of membrane fouling, but it remains
Seawater positive under the severe fouling condition where water flux decreases by 32% if the system capacity is
Net specific energy (NSE) large enough to have efficient pumps and turbines.
System capacity © 2018 Elsevier Ltd. All rights reserved.
Fouling

1. Introduction
The salinity gradient is one of emerging renewable energy re-
sources, which are intensively investigated recently [1e3]. The
global energy potential by mixing river water and seawater is
estimated to be 2 TW [4]. Among the technologies to harvest en-
ergy from the salinity gradient, pressure retarded osmosis (PRO) is
the most rigorously studied [5] since it was first proposed by Sidney
Loeb in 1975 [6]. In a PRO process, a highly concentrated solution
(called draw solution (DS)) draws water across a semi-permeable
membrane from a less concentrated feed solution (FS) when a
hydraulic pressure lower than the osmotic pressure difference be-
tween DS and FS is applied on the DS side. The volume-expanded
(and diluted) DS with the hydraulic pressure flows through a
hydro-turbine to produce energy. PRO can be used not only to
produce energy as a power plant, but also to decrease energy
consumption in reverse osmosis process [7e11].
Fundamentals of PRO are well studied by previous researches
focused on lab-scale PRO experiments and modeling [12e14]. PRO
membrane coupons were tested to find out the effects of the active
and support layer characteristics, hydrodynamic conditions, and
fouling on the performance of lab-scale PRO processes. Power
* Corresponding author.
E-mail address: suhankim@pknu.ac.kr (S. Kim).
density, which is defined as the produced hydraulic power per unit
PRO membrane area, is one of the key characteristics obtained from
a lab-scale PRO test.
However, power density obtained from a lab-scale PRO test
cannot be used to calculate the actual energy production from a
full-scale PRO system, where the actual salinity gradient decreases.
This is because DS is diluted by gaining permeate from FS, which is
concentrated by losing its volume by the amount of the permeate.
In addition, the inefficient hydraulic devices (e.g., pump, motor,
energy recovery device (ERD), turbine, and generator), and pres-
sure loss along the channels (e.g., the DS and FS channels, and
pipelines) makes the net energy smaller [15,16].
Thus, the net energy produced from full-scale PRO processes
should be carefully calculated in consideration of the decreased
salinity gradient, inefficient hydraulic devices, and pressure loss
along the channel. Straub et al. (2014) analyzed the module-scale
performance of PRO based on the modeling approach [17]. In this
work, ‘performance down’ of module-scale PRO processes
(compared to lab-scale) is well described by accounting for changes
in flow rate, pressure, and concentration inside the module. How-
ever, the energy production is rather overestimated due to
simplified assumptions in the modeling work (e.g., no pressure loss
in the DS and FS channels, and the hydraulic devices with perfect
efficiencies). He et al. (2016) evaluated the performance of a scaled-
up PRO process considering the various efficiencies of hydraulic
devices, but pressure loss along the channel length was not

https://doi.org/10.1016/j.energy.2018.06.042
0360-5442/© 2018 Elsevier Ltd. All rights reserved.
374 M. Kim, S. Kim / Energy 158 (2018) 373e382
considered [18]. Since pressure loss along the channel plays an
important role to evaluate the energy consumption in a full-scale
PRO process, a net energy production (NEP) analysis should ac-
count for this viscous dissipation.
Recently, the modeling approaches for full-scale PRO simulta-
neously accounted for the changes in parameters (e.g., pressure loss
and the inefficient hydraulic devices) along the channel length
[19e21]. In these works, the NEP analyses are supported by a
fundamental modeling approach to predict the full-scale perfor-
mance of a PRO process, and account for the energy production by
turbine (plus effect), the energy consumption by pumps (minus
effect), and the energy saving by ERD (plus effect) with assumed
efficiencies of hydraulic devices, respectively. These NEP analyses
found that the membrane power density was not coupled with NEP
(e.g., the operation condition towards the maximum power density
cannot accomplish the maximum NEP.). Instead, the effects of FS
and DS flow rates are very important parameters to determine NEP
and a wrong selection of FS and DS flow rates may result in a
negative NEP.
It is quite interesting that a full-scale PRO cannot always pro-
duce a net positive energy from the salinity gradient, and thus it is
of paramount importance to select a proper design option (e.g.,
system capacity, membrane area (or module length), the DS and FS
flow rates, and the mechanical pressure on the DS side) to produce
a positive energy. The relationships between hydraulic parameters
(e.g., flow rates and pressures) and NEP are studied in the previous
researches [17e21], and some of these works warn that full-scale
PRO may not extract net positive energy from the salinity
gradient between seawater and river water because of the pressure
loss and the efficiencies of hydraulic devices. According to a
recently published review paper [22], net energy cannot be
extracted from PRO using the salinity gradient between seawater
and river water mostly due to energy consumption for
pretreatments.
While reviewing literature, we have set two questions about
NEP of PRO. First, can net energy be extractable if pretreating
seawater and river water is not necessary? The module-scale
simulation results from the literature have answered this ques-
tion with yes. However, most of those works did not consider the
fact that the efficiencies of hydraulic devices such as turbine and
pump are dependent upon their capacities. Selection of hydraulic
devices with higher capacities results in the higher efficiencies of
these devices [17,23], and thus a smaller PRO system with less
efficient hydraulic devices produces less energy from the same
salinity gradient compared to a larger one. It may be impossible to
extract net energy from clean seawater and river water if the sys-
tem capacity is too small to have efficient hydraulic devices. To our
best knowledge, the effect of system capacity on NEP in a full-scale
PRO process has not been reported so far. In our work, the effect of
system capacity on NEP of full-scale PRO was systemically inves-
tigated by developing and simulating a module-scale PRO model.
While most of previous researches [17,19e21,24] simply calculated
the energy consumption of the membrane system with pumps and
ERD using assumed constant efficiencies of the hydraulic devices,
this work considered them as a function of pressure and flow rate.
The second question is whether the net energy from PRO using
seawater and river water becomes negative in the presence of
fouling. Pretreating DS and FS requires 0.1e0.4 kWh/m3 of specific
energy per treated volume and net specific energy (NSE) per PRO
system capacity (sum of DS and FS flow rates) should be less than
0.156 kWh/m3 with consideration of inefficiencies of hydraulic
devices [22]. The pretreatment is believed to be essential to keep
the efficiency of energy production from PRO, but the energy
requirement to pretreat DS and FS may exceed NSE of PRO. What if
we would remove the pretreatment processes in a PRO system? Of
course, it is difficult to control fouling and the performance of PRO
membranes becomes poorer. However, if NSE of a PRO system re-
mains positive even in the presence of severe fouling, we may take
into consideration of removing the pretreatment processes from
the system in order to obtain net energy from seawater and river
water. Thus, we investigated the effect of fouling on the NSE of a
full-scale PRO using the module-scale model developed in this
work. We defined levels of fouling by controlling the values for
water permeability of membrane (A), solute permeability of
membrane (B), structural parameter of membrane support layer (S),
and friction factors of DS and FS channels (fd and ff). The objective of
this work is to answer the two questions discussed above on net
energy extractable from PRO using seawater and river water.
2. Methods
2.1. The decreased salinity gradient in a module-scale PRO
The actual salinity gradient in a module-scale PRO decreases
because the volume of permeate is enough to change the concen-
trations of DS and FS. The decreased salinity gradient can be esti-
mated by the mass balance in a module-scale PRO as shown in Fig. 1
using length-averaged parameters (e.g., water flux, Jw, and reverse
solute flux, Js) inside the PRO module. The PRO module in the
system is considered as a black-box containing the intrinsic mem-
brane parameters (e.g., water permeability, A, solute permeability,
B, and structural parameter of the support layer, S), and module
parameters (e.g., width (w), length (l), the heights (Hd and Hf) of the
DS and FS channels, and the membrane area (Am ¼ wl); the module
is assumed to be an ideal flat sheet module with no dead zone.).
This type of approach has been applied to accurately predict the
performance of a commercial spiral wound forward osmosis
module [25].
DS with a pressure (Pd,in) flows into the DS side of the PRO
module with a flow rate (Qd) and a concentration (Cd). Inside the
PRO module, the DS is diluted by gaining fresh water from the feed
side with a water flux, Jw (i.e. the permeate flow rate, Qp, is JwAm.)
and loses some pressure by viscous dissipation along the DS
channel length. Thus, the DS is changed into the diluted DS with a
concentration, Cdd (<Cd), a flow rate, Qdd (>Qd), and a pressure, Pd,out
(<Pd,in) when it comes out of the module. On the other side (i.e, the
feed side), FS with a pressure (Pf,in) enters the module with a flow
rate (Qf) and a concentration (Cf). Inside the module, the FS is
concentrated by losing fresh water to the DS side with the same
water flux, Jw and the solutes coming from the DS side with a solute
flux, Js, and loses some pressure by viscous dissipation along the FS
channel length. Therefore, the FS turns into the concentrate with a
concentration, Cc (>Cf), a flow rate, Qc (<Qf), a pressure, Pf,out (<Pf,in).
Water and solute mass balances on the DS and FS sides are
described as [16]:
Qd þ Jw Am ¼ Qdd ðWater mass balance on the DS sideÞ (1)
Cd Qd  Js Am ¼ Cdd Qdd ðSolute mass balance on the DS sideÞ
(2)
Qc þ Jw Am ¼ Qf ðWater mass balance on the FS sideÞ (3)
Cf Qf  Js Am ¼ Cc Qc ðSolute mass balance on the FS sideÞ (4)
Before entering the PRO module, the DS flows into the ERD by
the DS supply pump with flow rate, Qd, and pressure, Pd, and it is
pressurized by the diluted DS, which partly returns to ERD with
flow rate, Qd, and pressure, Pd,out (if the head loss along the pipeline
 M. Kim, S. Kim / Energy 158 (2018) 373e382 375

Fig. 1. A schematic of a module-scale PRO process (modified from Ref. [20]).

is ignored) to transfer the hydraulic energy to the DS as shown in

Fig. 1. If the ERD is perfectly efficient, the DS is pressurized up to
Pd,out by the ERD before flowing into the booster pump. Since a real
ERD is inefficient, however, pressure loss (DPERD ) occurs when the
returning diluted DS transfers its hydraulic energy to the DS flow-
ing into the ERD. Thus, the pressure of the DS before the booster
pump becomesPd;out  DPERD , and the pressure of the booster pump
(Pb) is determined by:


Pb ¼ Pd;in  Pd;out  DPERD ¼ DPERD þ DPd (5)

where DPd is the pressure loss along the DS channel, and Eq. (5)
means the required booster pump pressure is the sum of the two
head losses by the ERD and the DS channel. In this work, DPERD is
assumed to be 5% of the pressure given to the ERD by the diluted DS
(i.e., Pd,out), which is equivalent to 95% of the ERD efficiency.
Fig. 2. Pressure drops along the feed channel in the 8 inch RO modules (SW30XHR-
400i, Dow chemical Co.); The length, width, and height of the module are 1.0 m, 37.0 m
2.2. Pressure loss along the DS and FS channels and 0.71 mm, respectively [29] and the spacer porosity is assumed to be 0.72 [21].

Pressure loss occurs along the DS and FS channels due to the

frictions between the fluids and spacers filled channel, and it
significantly increases the energy consumption by pumps in a
module-scale PRO [19,20]. The pressure drops on the DS and FS
sides (DPd andDPf ) can be expressed by the Darcy-Weisbach
equation [20]:
rd Ud2
DPd ¼ Pd;in  Pd;out ¼ fd L (6)
2dH;d
rf Uf2
DPf ¼ Pf ;in  Pf ;out ¼ ff L (7)
2dH;f
where f is the Darcy friction factor; r is the density of solution; U is
the length-averaged velocity in the channel
(Ud ¼ ðQd þ Qdd Þ=ð24wHd Þ and Uf ¼ ðQf þ Qc Þ=ð24wHf Þ, where 4 is
the spacer porosity); dH is the channel hydraulic diameter (dH ¼
2wH=ðw þ HÞ); and the subscripts d and f denote the DS and FS
channels, respectively. In order to obtain the Darcy friction factor, f,
we assumed that the flow regimes of the DS and FS channels are
identical to that of the feed channel of an existing reverse osmosis
(RO) module. Fig. 2 shows the pressure drops on the feed channel of
the commercially available 8 inch RO modules (SW30XHR-400i,
Dow chemical Co.) according to the feed crossflow velocity and the
channel length (i.e., the number of the modules in series; for
example, ‘L ¼ 3 m’ means three modules are connected in series.).
The pressure drop data were obtained using a free RO simulation
program, ROSA 9.0 from the manufacturer [29] and the Darcy-
Weisbach equation (Eq. (6)) with f (¼4.69) as a fitting parameter.
The pressure drop on the FS side (DPf ) can be converted into the
minimum required feed pump pressure (Pf) such as:
 
Pf ¼ Pf ;in ¼ DPf þ Pf ;out ¼ DPf Pf ;out ¼ 0 (8)
where the minimum value of Pf is obtained when the FS outlet
pressure, Pf,out, approaches zero.
2.3. Net energy production (NEP) analysis
In a general PRO process like Fig. 1, the three pumps (DS, FS, and
booster pumps) consume energy and the turbine produces energy.
The power produced or consumed by these hydraulic devices can
be calculated by multiplication between pressure and flow rate.
Thus, the net power (Wnet) produced from a module-scale PRO
process can be expressed as:
Wnet ¼ Wt  Wd  Wf  Wb (9)


Wt ¼ Pd;out Qp ht hg ¼ Pd;in  DPd Qp ht hg (10)
376 M. Kim, S. Kim / Energy 158 (2018) 373e382
. 
Wd ¼ Pd Qd hp;d hm;d
.  . 
Wf ¼ Pf Qf hp;f hm;f ¼ DPf Qf hp;f hm;f
.  . 
Wb ¼ Pb Qd hp;b hm;b ¼ ðDPERD þ DPd ÞQd hp;b hm;b
where W is the power produced or consumed by a hydraulic device;
Qp is the permeate flow rate; ht , hg , hp , and hm are the efficiencies of
turbines, generators, pumps, and motors; and the subscripts d, f,
and b mean DS supply pump, FS supply pump, and booster pump,
respectively. In general, the efficiency of a hydraulic device is a
function of the capacity of the device and the efficiencies becomes
higher at larger capacities [17,23]. The efficiencies of pumps and
motors can be determined by the calculation using Power Model,
Energy Recovery, Inc. by inputting a flow rate and a pressure for a
pump [24]. The efficiencies of turbines and generators are assumed
to be the same as those of pumps and motors if the capacity of the
pump and the turbine are identical.
Using Eqs. (9)e(13), the net power can be rearranged into:
 . 
Wnet ¼ Pd;in Qp ht hg  DPd Qp ht hg þ Qd hp;b hm;b
.  .  c5 ¼
DPf Qf hp;f hm;f  DPERD Qd hp;b hm;b
. 
Pd Qd hp;d hm;d
The only positive term in Eq. (14) is determined by the DS inlet
pressure, the permeate flow rate, and the efficiencies of turbine and
generator. If the efficiencies are unity, this positive term is equiv-
alent to membrane power density MPD ¼ Pd;in Qp =Am ¼ Pd;in Jw,
when pressure losses and FS pressure are ignored) multiplied by
the membrane area (Am). The MPD is the most frequently used term
to evaluate the performance of a PRO membrane, but the net power
is significantly affected by pressure drops and the hydraulic devices
efficiencies as shown in Eq. (14). In order to quantify NEP, the net
power is divided by the sum of DS and FS flow rates for normali-
zation in terms of a module-scale PRO system capacity, which re-
sults in net specific energy production per unit system capacity
(NSE) expressed as [18]:
Wnet
NSE ¼
Qd þ Qf
where NSE means the net energy extracted per total volumes of the
DS and FS [16].
In order to calculate NSE, all the equations discussed so far (i.e.,
Eqs. (1)e(15)) should be solved simultaneously, but the length-
averaged water and solute flux (Jw and Js) in the PRO module
should be obtained by modeling the internal and external con-
centration polarization phenomenon (ICP and ECP). The procedure
to model ICP and ECP in the coupon-size PRO membrane are
explained specifically in the literature published by our research
group [27,30,31]. The length-averaged ICP and ECP phenomena in a
module-scale PRO module are assumed to be identical to those in a
PRO membrane coupon as shown in Fig. 3, where c1 and c5 are the
solute concentrations of the bulk FS and DS, respectively;c2 and c4
are the solute concentrations at the feedeactive layer and the
DSesupport layer interfaces, respectively (resulting from ECP); and
c3 is the solute concentration at the interface between the active
and support layers (resulting from ICP within the support layer)
[30].
The boundary conditions to model the length-averaged ICP and
ECP can be obtained from mass balance in a module-scale PRO
(11)
(12)
(13)
Fig. 3. The solute concentration profile across PRO membrane (adopted from Ref. [30]).
process discussed in Section 2.1 such as:
Cd þ Cdd
(16)
(14) 2
Cf þ Cc
c1 ¼ (17)
2
Pd; in þ Pd; out Pf ; in þ Pf; out
DPm ¼  (18)
2 2
where, DPm is the transmembrane pressure. The water and solute
flux (Jw and Js ) are determined by the ICP and ECP models in a PRO
membrane coupon described in our previous work [30,31] using
these length-averaged parameters with the length-averaged ve-
locity in the channel (Ud ¼ ðQd þ Qdd Þ=ð24wHd Þ and Uf ¼ ðQf þ
Qc Þ=ð24wHf Þ, where 4 is the spacer porosity). Since the water and
solute flux affect the boundary condition (pressures, flow rates, and
concentrations) in the module-scale PRO model, all the equations
described so far should be solved together with the equations for
the ICP and ECP model in a PRO coupon (e.g., more than 30 equa-
(15) tions are simultaneously solved in this work.). In order to solve this
problem, we used the evolutionary algorithm in Microsoft Excel
Solver, which is one of the most accessible problem solving tools to
general researchers [30,31].
2.4. Simulation conditions for the NEP analysis
Table 1 shows the modeling parameters used in this work and
they cover all parameters generally considered in previous works
[17e21] dealing with a module-scale PRO modeling approach. The
intrinsic membrane characteristics are obtained from the best
available PRO membranes [19] and the module dimensions are
assumed from those of the DS channel of an existing spiral-wound
PRO module [32]. Both DS and FS are assumed to be a pure sodium
chloride solution, and 600 mol/m3 and 10 mol/m3 of solute con-
centrations to simulate the osmotic pressure of seawater and river
water, respectively. In addition, the module-scale PRO model
developed in this work covers two more parameters than previous
works, system capacity and membrane fouling. Increasing system
capacity is accomplished by increasing the number of PRO modules,
which means the modules are arranged in parallel. The DS, FS, and
booster pumps, and turbine are assumed to cover the whole system
 M. Kim, S. Kim / Energy 158 (2018) 373e382 377

Table 1
The classified modeling parameters in a module-scale PRO process.

Class Parameters Symbol (unit) Value(s)

Water & solution characteristics Diffusion coefficient of solute D (109 m2 s1) 1.48
Viscosity of the solution m (103 kg m1s1) 0.933
Temperature of the solution T (K) 296
Density of the solution r (kg/m3) 1000
van't Hoff coefficient b 2
Membrane and module characteristics Water permeability A (1012 m s1Pa1) 9.56
Solute permeability B (109 m s1) 131
Structural parameter S (106 m) 434
Width of the both channels w (m) 29
Height of the DS channel Hd (103 m) 0.8
Height of the FS channel Hf (103 m) 0.4
Length of the both channels L (m) 1e8
Porosity of the spacer f 0.72
Boundary condition FS flow rate per module Qf (m3 d1) 10e100
DS flow rate per module Qd (m3 d1) 10e100
DS inlet pressure Pd, in (bar) 1e25
Solute concentration in DS Cd (mol m3) 600
Solute concentration in FS Cf (mol m3) 10
Hydraulic devices efficiencies Pump efficiency hp 0.44e0.90
Motor efficiency hm 0.87e0.97
Turbine efficiency ht 0.43e0.88
Generator efficiency hg 0.89e0.96

capacity with a single unit. The efficiencies of pump, motor, turbine,
and generators are obtained from the calculations by Power Model,
Energy Recovery, Inc. as a function of system capacity (i.e., flow
rates and pressures of DS, FS, and permeate).
The membrane fouling parameter is considered by inputting
adjusted modeling parameters according to fouling levels defined
in Table 2. When fouling occurs, water permeability (A) decreases
and solute permeability (B), and structural parameters of the sup-
port layer (S) increase. We defined one fouling level by 10%
decrease or increase of these parameters compared to those at the
state of no fouling. For example, water permeability decreases by
50% compared to that of the clean membrane (i.e., no fouling) at LV-
5 of PRO membrane fouling as shown in Table 2. In addition, the
pressure losses along the DS and FS channels become increased due
to the shrunken channel heights and the increased roughness of
spacers and membrane surfaces when fouling layer grows inside
the channels. For simplification, we simulated the pressure losses
along both channels by increasing the friction factors (fd and ff). In
each fouling level the friction factors are determined by multiplying
the number of level, the friction factors of the clean membrane, and
two. For example, the friction factors at LV-5 of PRO membrane
fouling are 10 (¼ 5  2) times as large as those of the clean mem-
brane as shown in Table 2.
clearly shows the decoupling of MPD and NSE as the module length
(L) increases. MPD decreases as module length increases because of
the two following reasons. First, MPD will increase at the higher
average fluxes (Jw) in the module. Thus, it decreases at longer
module lengths because the average flux decreases due to the effect
of diluted DS and concentrated FS along the channels. Second, MPD
is inversely proportional to the membrane area. Since the mem-
brane area increases as the module length becomes longer, MPD
should decrease at longer module lengths. By the way, NSE in-
creases as the module length increases. The power gained by PRO is
the product of the DS outlet pressure (Pd,out) and the permeate flow
rate (Qp). Because the permeate flow rate increases at larger
membrane areas, the gained power will increase at longer module
lengths. However, the increasing rate of NSE decreases as module
length increases (Fig. 4a) because pressure losses and the decrease
in average permeate flux retard the increase in NSE.
Fig. 4b, c, and d show the specific energy (SE) distribution of a
module-scale PRO as functions of the DS inlet pressure, DS flow
rate, and FS flow rate, respectively. In a PRO process, energy is
produced by the pressure (Pd;in ) and the part of the flow rate of
diluted DS (i.e., the permeate flow rate, Qp ) out of the module. Here
the maximum specific energy (MSE) per the unit PRO system ca-
pacity is defined if all the losses are ignored such as:

3. Result and discussion

Pd;in Qp
MSE ¼ (19)
3.1. Effect of hydrodynamic parameters Qd þ Qf

The actual specific energy gained by PRO should be smaller than

As discussed in literature [17e22], the key parameter to assess
the performance of a full-scale PRO is not MPD but NSE. Fig. 4a
Table 2
The definition of fouling level of PRO membrane.
Fouling level A (1012 m s1Pa1) B (109 m s1) S (106 m) fd
MSE because of the turbine is not perfectly efficient. Other losses
from the DS channel pressure drop, FS channel pressure drop (i.e.,
energy consumption by FS supply pump), inefficient ERD, and en-
ergy consumption by DS supply pump are calculated using each
negative term in Eq. (14) such as:
ff

No 9.56 131 434 4.69 4.69 SE loss from the DS channel pressure drop
LV-1 8.60 157 521 9.38 9.38  . 
LV-2 7.65 183 608 18.76 18.76 DPd Qp ht hg þ Qd hp;b hm;b
LV-3 6.69 210 694 28.14 28.14 ¼ (20)
Qd þ Qf
LV-4 5.74 236 781 37.52 37.52
LV-5 4.78 262 868 46.90 46.90
378 M. Kim, S. Kim / Energy 158 (2018) 373e382

. 
DPf Qf hp;f hm;f
SE loss from the FS channel pressure drop ¼
Qd þ Qf
(21)

. 
DPERD Qd hp;b hm;b
SE loss from the inefficient ERD ¼ (22)
Qd þ Qf

. 
Pd Qd hp;d hm;d
SE loss from the DS supply pump ¼ (23)
Qd þ Qf

In PRO, water permeates through membrane from FS to DS side,

which is the opposite direction of the pressure difference between
DS and FS (i.e., the pressure in DS side is higher than that in FS
side.). Hence, the permeate flow rate decreases as the DS inlet
pressure increases. If permeate flow rate decreases, the DS channel
pressure drop decreases because the actual DS flow rate (by addi-
tion of the permeation volume to the DS inlet flow) becomes
smaller than that in a higher permeate flow rate. Thus, the SE loss
from the DS channel pressure drop decreases as the DS inlet
pressure increases (Fig. 4b) because the DS channel pressure drop
and the permeate flow rate decrease (Eq. (20)).
Meanwhile, the FS channel pressure drop increases at higher DS
inlet pressures because the actual FS flow rate (by losing the
permeation volume from the FS inlet flow) becomes larger as the
permeate flow rate decreases. Therefore, the SE loss from the FS
channel pressure drop increases as the DS inlet pressure increases
(Fig. 4b and Eq. (21)). The SE loss from the inefficient ERD is a
function of the pressure loss of ERD (DPERD ) as shown in Eq. (22).
Since DPERD is assumed to be 5% of the DS outlet pressure as dis-
cussed in section 2.1, the SE loss from the inefficient ERD is almost
proportional to the DS inlet pressure (Fig. 4b). The SE loss from the
DS supply pump remains constant because the pressure acquired
by DS supply pump is assumed to be 1 bar, which is necessary to
push the DS to the ERD [26].
When DS or FS flow rate increases, the permeate flow rate be-
comes increased because (1) the effect of diluted DS or concen-
trated FS is decreased and (2) the effect of ECP is decreased. As
discussed earlier, the DS channel pressure drop increases and the FS
channel pressure drop decreases as the permeate flow rate in-
creases. Hence, as the DS inlet flow rate increases (Fig. 4c), the SE
loss from the DS channel pressure drop increases (Eq. (20)), the SE
loss from the FS channel pressure drop decreases (Eq. (21)), the SE
loss from the inefficient ERD increases (Eq. (22)), and the SE loss
from the DS supply pump increases (Eq. (23)). According to Fig. 4d,
as the FS inlet flow rate increases, the SE loss from the DS channel
pressure drop does not change very much because the increase in
the numerator is similar to that in the denominator in Eq. (20), the
SE loss from the FS channel pressure drop increases (Eq. (21)), the
SE loss from the inefficient ERD decreases (Eq. (22)), and the SE loss
from the DS supply pump decreases (Eq. (23)).
As the operation parameters (Pd,in, Qd, and Qf) increase, NSE in-
creases until it reaches the maximum value and then decreases,
(Fig. 4b, c, and d). If the DS inlet pressure increases, the permeate
flow rate decreases, which retards the increase in MSE and finally
makes MSE decreased (Fig. 4b). If the DS or FS inlet flow rate

Fig. 4. (a) Effect of the module length (L) on NSE and MPD; SE diagram of a module-
scale PRO as a function of (b) DS inlet pressure (Pd,in), (c) DS flow rate (Qd), and (d) FS
flow rate (Qf). All the simulation parameters except the variable in the x axis are as
follows: Qd,in ¼ 50 m3/d, Qf,in ¼ 50 m3/d, Pd,in ¼ 13 bar, L ¼ 8 m, hp ¼ 0:80, hm ¼ 0:95,
ht ¼ 0:87, and hm ¼ 0:98. Parameters not mentioned here are taken from Table 1.
 M. Kim, S. Kim / Energy 158 (2018) 373e382 379

increases, the permeate flow rate increases, which increases power
gained by PRO. However, the increase in SE is slowing down and SE
turns around to decrease because the system capacity increases due
to the increase in the DS of FS inlet flow rate (Fig. 4c and d). In
addition, the sum of the SE losses (Eqs. (20)e(23)) increases as the
operation parameters increase, which makes the decreasing rate of
NSE higher. Therefore, NSE is highly dependent upon the selection
of operation parameters such as the DS inlet pressure, the DS inlet
flow rate, and the FS inlet flow rate. A wrong selection may result in
negative NSE.
3.2. Maximizing net specific energy
The primary objective of PRO is to produce positive net energy.
Thus, it is very important to find the optimal operation parameters
to maximize NSE. As shown in Fig. 4b, c, and d, the pattern of
change in NSE is all the same regardless of the operation parameter.
As one of any operation parameters increases, NSE increases until it
reaches its maximum, and then decreases, which means NSE has
only one local maximum as a function of one operation parameters.
Since the operation parameters controllable by PRO operators are
three (e.g., Pd,in, Qd, and Qf), the algorithm to find the optimal
operation conditions can be an iterative process to find each
optimal operation parameter with fixed other parameters until all
three values become identical (Fig. 5).
Table 3 shows some examples of changes in operation param-
eters while the algorithm is carried out. The flow rates change by
1 m3/d and the DS inlet pressure changes by 0.1 bar for each iter-
ative process. The maximum NSE extractable from PRO using
simulation condition listed in Table 1 and fixed hydraulic devices
efficiencies (hp ¼ 0:80, hm ¼ 0:95, ht ¼ 0:87, and hm ¼ 0:98) is
found to be 95.12 Wh/m3 with Pd,in ¼ 9.3 bar, Qd ¼ 6 m3/d, and
Qf ¼ 20 m3/d as the optimal operation condition as shown in
Table 3.
Fig. 6 shows the SE distribution as a function of the DS inlet
pressure with optimal DS and FS flow rates conditions, and total
loss means the sum of the SE losses (Eq. (20) to (23)). Compared to
Fig. 4b, the total loss becomes smaller, which means controlling SE
losses is essential to obtain the optimal operation conditions.

3.3. Effect of system capacity and fouling

According the module-scale simulation in this work, the

Fig. 5. The algorithm to find the optimal operation condition to maximize NSE.
maximum NSE from PRO using seawater and river water is less than
0.1 kWh/m3, which is a little smaller than that expected from a
literature [22]. If clean seawater and river water are available (i.e.,
pretreating DS and FS is not necessary), the practical limit of PRO
becomes around 0.1 kWh/m3. However, it only works with the
assumed efficiencies of hydraulic devices (e.g., hp ¼ 0:80, hm ¼
0:95, ht ¼ 0:87, and hm ¼ 0:98). Since the efficiency of hydraulic
device is highly dependent upon the capacity of the device, the
system capacity may limit extracting net energy from PRO.
In order to discuss the practical limit of a full-scale PRO with
relation to system capacity, the optimal operation parameters to
maximize NSE should be fixed while the hydraulic devices effi-
ciencies according to the system capacity are varied for the simu-
lation. Fig. 7 shows the effect of system capacity on the SE
distribution. According to the optimal operation condition derived
from section 3.2, the DS and FS flow rates are 6 and 20 m3/d,
respectively. Thus, the unit module capacity is 26 m3/d for the
optimal operation. The full-scale system consists of several mod-
ules with the unit capacity of 26 m3/d arranged in parallel. As dis-
cussed in section 2.4, the hydraulic devices are assumed to cover

Table 3
Examples of traces of operation parameters (Pd,in, Qd, and Qf) while carrying out the algorithm to maximize NSE described in Fig. 5

n 1st try 2nd try 3rd try

3 3 3 3 3 3
Qd (m /d) Qf (m /d) Pd,in (bar) NSE (Wh/m ) Qd (m /d) Qf (m /d) Pd,in (bar) NSE (Wh/m ) Qd (m3/d) Qf (m3/d) Pd,in (bar) NSE (Wh/m3)

0 50 50 e e 75 75 e e 100 100 e e
1 28 37 13.0 75.22 45 47 12.9 60.45 64 55 12.8 43.98
2 18 30 12.2 84.92 26 35 12.9 77.08 33 39 13.5 70.18
3 13 26 11.5 89.82 17 29 12.2 85.70 20 31 12.5 82.77
4 10 23 10.9 92.52 13 26 11.4 89.97 14 27 11.7 88.80
5 9 23 10.5 93.47 10 23 10.9 92.52 11 24 11.0 91.82
6 8 21 10.1 94.20 9 23 10.5 93.47 9 22 10.6 93.37
7 7 20 10.0 94.64 8 21 10.1 94.20 8 21 10.3 94.09
8 6 19 9.8 94.86 7 20 10.0 94.64 7 20 10.0 94.64
9 6 20 9.4 95.11 6 19 9.8 94.86 6 19 9.8 94.86
10 6 20 9.3 95.12 6 20 9.4 95.11 6 20 9.4 95.11
11 e e e e 6 20 9.3 95.12 6 20 9.3 95.12
380 M. Kim, S. Kim / Energy 158 (2018) 373e382
Fig. 6. SE diagram of a module-scale PRO as a function of DS inlet pressure (Pd,in) with
optimal DS and FS flow rates: Qd,in ¼ 6 m3/d and Qf,in ¼ 20 m3/d. Parameters not
mentioned here are the same as those taken from the simulation conditions to derive
the results in Fig. 4.
Fig. 7. SE diagram of full-scale PRO systems as a function of system capacity. Effi-
ciencies of pump, motor, turbine, and generators are calculated based on the system
capacity using Power Model, Energy Recovery, Inc. All the simulation parameters
except the efficiencies are the same as those taken from the simulation conditions to
derive the results in Fig. 6.
the whole system capacity with a single unit. The minimum system
capacity simulated is 520 m3/d, which is one of the smallest ca-
pacities where hydraulic devices efficiencies are available in Power
Model, Energy Recovery, Inc. NSE available from the smallest PRO
system simulated (520 m3/d) is around 30 Wh/m3. Therefore, it can
be said that net energy is extractable even in a small system like
520 m3/d if clean seawater and river water are available. NSE
extracted from a full-scale PRO increases as the system capacity
increases (Fig. 7), because the efficiencies of hydraulic devices in-
crease at larger capacities.
However, clean seawater and river water for PRO are not avail-
able in most cases and pretreating DS and FS requires 0.1e0.4 kWh/
m3 of specific energy per treated volume [22]. Since the extractable
NSE is less than 0.1 kWh/m3 without pretreatments, net energy
cannot be extracted from PRO using seawater and river water with
pretreatments. If pretreatments are removed from a PRO system,
the membrane fouling is inevitable and the performance of the full-
scale PRO system must be degraded. Fig. 8 shows the effect of
membrane fouling on the SE distribution in cases of two systems
with different capacities, 1040 and 55,900 m3/d according to
fouling levels defined in Table 2 in section 2.4. In both cases, MSE
Fig. 8. The effect of membrane fouling on the SE distribution in cases of two systems
with different capacities, (a) 55,900 and (b) 1040 m3/d. The level of PRO membrane
fouling is defined in Table 2 and all other simulation conditions are the same as those
in Fig. 7.
decreases and total loss increases as fouling becomes severer,
which results in the decrease in NSE in the presence of fouling.
In the fouling simulation, water permeability (A) decreases, and
solute permeability (B) and structural parameters of the support
layer (S) increase as fouling becomes severer. The decrease in water
permeability and the increase in solute permeability simulate the
fouling of the active layer by seawater, and the increase in structural
parameter of the support layer simulates the fouling of the support
layer by river water. If water permeability decreases, the water flux
(per unit osmotic pressure difference) decreases. If solute perme-
ability increases, more solutes in DS side penetrates through
membrane and the actual osmotic pressure difference between DS
and FS decreases, which results in the flux decline. If structural
parameters of the support layer increases, the ICP becomes severer
and the water flux decreases. The decrease in water flux resulting
from the decreased water permeability, the increased solute
permeability and the increased structural parameter of the support
layer contributes the decrease in MSE, which is proportional to the
permeate flow rate. If fouling occurs on the both sides in a PRO
module, the pressure losses along the channels become increased
due to the shrunken channel heights and the increased roughness
by the growth of the fouling layer. The increased pressure losses
due to fouling contribute the increase in total loss as shown in
Fig. 8.
 M. Kim, S. Kim / Energy 158 (2018) 373e382
While flux decreases by 32% (LV-5 of PRO membrane fouling),
NSE becomes a third of that without fouling in the case of
55,900 m3/d of system capacity, and it becomes negative in the case
of 1040 m3/d of system capacity (Fig. 8). If the fouling level without
pretreatments is predictable, we may figure out if a full-scale PRO
can extract net energy in the presence of fouling. However, it is
impossible to expect the level of fouling without specific informa-
tion on seawater and river water quality, membrane characteristics,
and operation conditions, which is out of scope in this work.
By the way, water flux in the optimal operation condition is 2.3
LMH, which is extremely low even compared to those of anaerobic
membrane bioreactor (>3 LMH) operating with a huge amount of
foulants [33]. Hence, a severe fouling like LV-5 of PRO membrane
fouling (e.g., 50% decrease in water permeability, 50% increase in
solute permeability and structural parameter of the support layer,
and 10-folds increase in the friction factors of the DS and FS
channels) may not occur if this low flux (2.3 LMH) is below a critical
flux where water flux starts to decline [34].
It may be possible to extract net energy from PRO using
seawater and river water without pretreatments because of the two
following reasons. First, NSE remains positive even in a severe
fouling if system capacity is large enough to have efficient hydraulic
devices (Fig. 8a). Second, water flux in the optimal operation con-
dition to maximize NSE may be too small to experience severe
fouling. Since the required SE for pretreatments exceeds the SE loss
expected by fouling, it would be reasonable to consider removing
pretreatments from a full-scale PRO system to extract net energy
from PRO using seawater and river water.
4. Conclusions
Net energy extractable from a full-scale PRO using seawater and
river water is estimated using a module-scale PRO model devel-
oped in this work. The net energy is determined by the difference
between energy production by turbine and energy consumption by
FS, DS, and booster pumps. The energy consumption by pumps is
divided into energy losses by inefficient ERD, pressure drops along
the DS and FS channels, and required energy for DS supply to ERD.
According to the algorithm to find the optimal operation condition,
the maximum NSE available from seawater is close to 0.1 kWh/m3
without pretreatments. Since the required specific energy for pre-
treating DS and FS is 0.1e0.4 kWh/m3, it may be concluded that net
energy cannot be extracted from PRO using seawater and river
water. Moreover, the maximum NSE decreases as system capacity
becomes smaller and it becomes around 0.03 kWh/m3 from a PRO
system with 520 m3/d as capacity.
One thing to be hopeful is that net energy could be extracted
from PRO using seawater and river water in the presence of fouling
instead of adopting pretreatments which require more energy
(0.1e0.4 kWh/m3) than net energy extractable (0.1 kWh/m3). Ac-
cording to our simulation results, NSE remains positive (~0.03 kWh/
m3) even in a severe fouling where water flux decreases by 32% if
system capacity is large enough (e.g., 55,900 m3/d) to have efficient
hydraulic devices. In addition, water flux in the optimal operation
condition to maximize NSE may be too small (e.g., 2.3 LMH) to
experience severe fouling. Therefore, the idea to design and operate
PRO without pretreatments is well worth considering.
Acknowledgements
This work was supported by the National Research Foundation
of Korea grant funded by the Korea government (MSIP; Ministry of
Science, ICT & Future Planning) (No. 2017R1A2B4002990).
381
References
[1] Olabi AG. Energy quadrilemma and the future of renewable energy. Energy
2016;108:1e6.
[2] Olabi AG. Renewable energy and energy storage system. Energy 2017;136:
1e6.
[3] Foley A, Olabi AG. Renewable energy technology developments, trends and
policy implications that can underpin the drive for global climate change.
Renew Sustain Energy Rev 2017;68:1112e4.
[4] Mantia FL, Pasta M, Deshazer HD, Logan BE, Cui Y. Batteries for efficient energy
extraction from a water salinity difference. Nano Lett 2011;11:1810e3.
[5] Sarp S, Li Z, Saththasivam J. Pressure Retarded Osmosis (PRO): past experi-
ences, current developments, and future prospects. Desalination 2016;389:
2e14.
[6] Loeb S, Norman RS. Osmotic power plants. Science 1975;189:654e5.
[7] Qureshi BA, Zubair SM. Exergetic analysis of a brackish water reverse osmosis
desalination unit with various energy recovery system. Energy 2015;93:
256e65.
[8] Altaee A, Millar GJ, Zaragoza G. Integration and optimization of pressure
retarded osmosis with reverse osmosis for power generation and high effi-
ciency desalination. Energy 2016;103:110e8.
[9] Touati K, Tadeo F, Elfil H. Osmotic energy recovery from reverse osmosis using
two-stage pressure retarded osmosis. Energy 2017;132:213e24.
[10] Sharqawy MH, Zubair SM, Lienhard VJH. Second law analysis of reverse
osmosis desalination plants: an alternative design using pressure retarded
osmosis. Energy 2011;36:6617e26.
[11] He W, Wang Y, Shaheed MH. Stand-alone seawater RO (reverse osmosis)
desalination powered by PV (photovoltaic) and PRO (pressure retarded
osmosis). Energy 2015;86:423e35.
[12] Nagy E, Dudas J. Hegedus. Improvement of the energy generation by pressure
retarded osmosis. Energy 2016;116:1323e33.
[13] Kim YC, Elimelech M. Adverse impact of feed channel spacers on the per-
formance of pressure retarded osmosis. Environ Sci Technol 2012;46:
4673e81.
[14] Yip NY, Elimelech M. Influence of natural organic matter fouling and osmotic
backwash on pressure retarded osmosis energy production from natural
salinity gradients. Environ Sci Technol 2013;47:12607e16.
[15] Tran TTD, Park K, Smith AD. System scaling approach and thermoeconomic
analysis of a pressure retarded osmosis system for power production with
hypersaline draw solution: a Great Salt Lake case study. Energy 2017;126:
97e111.
[16] Kim M, Kim S. Can full-scale pressure retarded osmosis system derive energy
from the ocean. In: Proceeding of 10th International conference on sustainable
energy and environmental protection: marine and hydro power, Bled,
Slovenia; 2017. p. 35e44. June 27th e 30th.
[17] Starub AP, Lin S, Elimelech M. Module-scale analysis pressure retarded
osmosis: performance limitations and implications for full-scale operation.
Environ Sci Technol 2014;48:12435e44.
[18] He W, Wang Y, Mujtaba IM, Shaheed MH. An evaluation of membrane
properties and process characteristics of a scaled-up pressure retarded
osmosis (PRO) process. Desalination 2016;378:1e13.
[19] Feinberg BJ, Ramon GZ, Hoek EMV. Scale-up characteristics of membrane-
based salinity-gradient power production. J Membr Sci 2015;476:311e20.
[20] Wang Z, Hou D, Lin S. Gross vs. net energy: towards a rational framework for
assessing the practical viability of pressure retarded osmosis. J Membr Sci
2016;503:132e47.
[21] O'Toole G, Jones L, Coutinho C, Hayes C, Napoles M. River-to-sea pressure
retarded osmosis: resource utilization in a full-scale facility. Desalination
2016;389:39e51.
[22] Straub AP, Deshmukh A, Elimelech M. Pressure-retarded osmosis for power
generation from salinity gradients: is it viable? Energy Environ Sci 2016;9:
31e48.
[23] Karlis AD, Papadopoulos DP. A systematic assessment of the technical feasi-
bility and economic viability of small hydroelectric system installations.
Renew Energy 2000;20:253e62.
[24] Park SH, Park B, Shon HK, Kim S. Modeling full-scale osmotic membrane
bioreactor systems with high sludge retention and low salt concentration
factor for wastewater reclamation. Bioresour Technol 2015;190:508e15.
[25] Jeon J, Jung J, Lee S, Choi JY, Kim S. A simple modelling approach for a forward
osmosis system with a spiral wound module. Desalination 2018;433:120e31.
[26] Energy Recovery, Inc. Webpage, http://www.energyrecovery.com.
[27] Kim S. Scale-up of osmotic membrane bioreactors by modeling salt accumu-
lation and draw solution dilution using hollow-fiber membrane characteris-
tics and operation conditions. Bioresour Technol 2014;165:88e95.
[29] The Dow Chemical Co., Webpage, http://www.dow.com.
[30] Lee J, Kim S. Predicting power density of pressure retarded osmosis (PRO)
membranes using a new characterization method based on a single PRO test.
Desalination 2016;389:224e34.
[31] Lee J, Choi JY, Choi J-S, Chu KH, Yoon Y, Kim S. A statistics-based forward
osmosis membrane characterization method without pressurized reverse
osmosis experiment. Desalination 2017;403:36e45.
[32] Kim YC, Kim Y, Oh D, Lee KH. Experimental investigation of a spiral-wound
pressure-retarded osmosis membrane module for osmotic power genera-
tion. Environ Sci Technol 2013;47:2966e73.
382 M. Kim, S. Kim / Energy 158 (2018) 373e382

[33] Lin H, Zhang WPM, Chen J, Hong H, Zhang Y. A review on anaerobic mem- [34] Kim S, Park H. Applicability assessment of subcritical flux operation in
brane bioreactors: applications, membrane fouling and future perspectives. crossflow microfiltration with a concentration polarization model. J Environ
Desalination 2013;314:169e88. Eng 2002;128:335e40.