Desalination 554 (2023) 116519

Contents lists available at ScienceDirect

Desalination
journal homepage: www.elsevier.com/locate/desal

An improved Closed-Circuit RO (CCRO) system: Design and

cyclic simulation
Mingheng Li *
Department of Chemical and Materials Engineering, California State Polytechnic University, Pomona, CA 91768, USA

H I G H L I G H T S

• An improved CCRO design is proposed to achieve the energetics of BRO.

• Continuous permeate production is possible in the proposed system.
• Cyclic dynamics are investigated using spatiotemporal models.
• Approximate analytical solutions are provided to analyze energy breakdowns.
• Advantages and limitations of original and improved CCROs are discussed.

A R T I C L E I N F O A B S T R A C T

Keywords: A volume-varying stirred tank with a free-piston are placed in front of DuPont's Closed-Circuit RO (CCRO) system
Batch RO to achieve improved energetics. Valves are used to direct fluid flow and piston motion to enable cyclic operations
Process design consisting of CC filtration and flushing steps. Similar to CCRO, overall system recovery in the proposed system is
Process dynamics
controlled by the CC/flushing time ratio, and uninterrupted permeate production is possible if the flushing is
System analysis
Cyclic steady state
done at high pressures. Spatiotemporal system-level models are developed and formulated as a set of coupled
Partial differential equation model partial differential equations (PDEs) and ordinary differential equations (ODEs). Detailed numerical simulations
and parametric studies are carried out to explore the flow and mass transfer characteristics and to compare its
performance at the cyclic steady-state (CSS) with the original CCRO and the conventional steady-state RO.

1. Introduction to evaluate the process performance at the cyclic steady state (CSS),
which is not the same as the one in the first cycle.
In recent years, dynamic and cyclic design concepts have been a In an effort to advance fundamental understanding of transient and
research hotspot in the RO desalination community. Some examples cyclic RO processes, the author developed a spatiotemporal RO model
include semi-batch RO (SBRO) or closed-circuit RO (CCRO) that took into account of pressure drop, concentration polarization and
[10,13,26,28,30,31], batch RO (BRO) [12,17,18,26], flow reversal RO axial dispersion [20]. The model was applied to flow reversal and
(FRRO) [2,8,9,27], pulse feed RO (PFRO) [25], stage rotation RO membrane flushing [20,23]. More recently, it has been extended to
(SRRO) [23] and periodic flow reversal and retentate recycle (FRRR) cyclic simulation of CCRO, which has a long CC filtration period fol­
with a time-varying ratio [23]. Different from traditional designs, where lowed by a short flushing period in each cycle [21]. The computational
pressure, flow and concentration conditions are functions of spatial framework enables detailed studies of cyclic transport phenomena and
location, these novel designs feature spatiotemporal dynamics. Similar parametric analyses that reveal thorough energy consumption break­
cyclic design concepts have been extensively studied in other engi­ downs in CCRO. Contrary to conjectures in literature, it suggests that
neering fields, such as pressure swing adsorption for gas separation [29] CCRO may not be claimed as an “energy-efficient” technology for high-
and adsorption enhanced reforming for hydrogen generation [7]. The recovery brackish water RO (BWRO), despite the relatively uniform flow
key technical challenges for analyzing and optimizing such cyclic pro­ and flux conditions. As pointed out in a few papers [5,18,21,26], the
cesses are to solve the coupled spatiotemporal transport phenomena and mixing between the low-concentration fresh feed and the increasingly

* Corresponding author.
E-mail address: [email protected].

https://doi.org/10.1016/j.desal.2023.116519
Received 21 January 2023; Received in revised form 18 February 2023; Accepted 23 February 2023
Available online 5 March 2023
0011-9164/© 2023 Elsevier B.V. All rights reserved.
M. Li Desalination 554 (2023) 116519

concentrated brine in CCRO causes undesired entropy generation, which BRO. However, its operation differs from the “batch” concept typically
adversely impacts its efficiency. If the mixing occurs in a volume-varying referred in literature where permeation is non-continuous. Instead, un­
tank, where the rising salt concentration results naturally from the interrupted water production is possible in the proposed system
permeate leaving the system, the energy efficiency can be enhanced. In employing the same operating strategy in CCRO. Therefore, it may be
this work, the CCRO design is modified to approach the energetics of a called a BRO following the convention in literature or an improved

Fig. 1. Schematic of (a) CC/refilling mode (piston moving forward), (b) flushing mode, (c) CC/refilling mode (piston moving backward), (d) flushing mode in BRO.
Red line: refilling, Blue line: filtration, Green line: flushing. Black line: idle. r = (1 − YSP)/YSP. (For interpretation of the references to colour in this figure legend, the
reader is referred to the web version of this article.)

2
M. Li Desalination 554 (2023) 116519

CCRO. Because of additional components used in the new system, the 2. Cyclic design
cyclic model developed for CCRO [21] must be significantly enhanced.
A schematic of the proposed BRO system is shown in Fig. 1. A
continuous stirred tank reactor (CSTR) or perfect mixer with a free-
piston is placed in front of a pressure vessel, housing one or more

Fig. 2. Schematic of (a) CC/refilling mode (piston moving forward), (b) flushing mode, (c) CC/refilling mode (piston moving backward), (d) flushing mode in BRO
with flow reversal. Red line: refilling, Blue line: filtration, Green line: flushing. Black line: idle. (For interpretation of the references to colour in this figure legend, the
reader is referred to the web version of this article.)

3
M. Li Desalination 554 (2023) 116519

membrane elements. Multiple valves are used to direct fluid flow and
piston motion in order for the system to run in a dynamic and cyclic NE Qf (tFT + tFL )
Q0 =
manner. The reciprocating-piston chamber is not a new idea; it is seen in YSP tFT + tFL
the commercial Dual Work Exchanger Energy Recovery (DWEER) de­ NE Qf Ytot + (1 − δ − Ytot )YSP
vices in seawater RO (SWRO) plants [1]. A similar double-acting piston =
YSP 1 − δYSP
is used to reduce down time in batch-type operations [3,4]. In Fig. 1, a ⎧ (3)
full cycle consists of two CC/refilling and two flushing steps. The CC and ⎪
⎪ NE Qf Ytot , for HPF
⎪
⎪
⎨ Y
refilling occur at the same time; when one side of the CSTR undergoes SP
=
the CC filtration step, the other is being refilled with fresh feed. ⎪
⎪
⎪ N E Qf
⎪
Initially, the piston is near the left of the CSTR (shown in Fig. 1(a)). ⎩
YSP
[Ytot + (1 − Ytot )YSP ], for LPF
The piston divides the CSTR into two compartments whose initial vol­
umes are Vf on the left and V0 on the right. As the fresh feed is pumped where Qf is the feed rate divided by the total number of elements in a
into the left compartment of the CSTR, the piston gradually moves to the conventional RO plant, and NE is the number of elements per vessel in
right. The RO concentrate is completely recycled in the closed circuit BRO. It is important to check that the flow per vessel is within the typical
(represented by the blue line) and mixed with the solution in the right range recommended by membrane manufacturers. For steady-state
compartment of the CSTR before entering to the RO unit. Some me­ design using 8” FilmTec membranes, the feed flow per vessel is about
chanical agitation (for example, by installing rotating blades in the 8–12 m3/h (35–55 gpm). The minimum concentrate flow is 3.6 m3/h (16
CSTR) may be induced to enhance mixing. How to implement this in a gpm) [6].
practical and efficient way is yet to be addressed in future studies.
The average water flux during CC filtration Jw,FT is dependent on Q0
Similar to the CCRO system studied previously [21], a time-invariant
[21]:
combined feed rate per vessel (Q0) and a constant single pass recovery
(YSP) are maintained in the CC filtration step. The refilling and permeate Q0 YSP
J w,FT = (4)
rates are both Q0YSP. The recycle to fresh feed ratio in the CC filtration Am
mode is r = (1 − YSP)/YSP.
Once the right compartment of the CSTR reaches a final volume of Vf, where Am is the membrane area per vessel.
or the cumulative volume of permeate production is V0 − Vf, the piston In BRO-LPF, both Q0 and Jw,FT should be slightly higher than those in
pauses and the flushing step begins (shown in Fig. 1(b)). Similar to the BRO-HPF (by a factor of 1 + YSP(1 − Ytot)/Ytot) to maintain the same
original CCRO system, the same time-invariant Q0 at the RO entrance is cycle-average flux.
maintained during flushing. Both high pressure flushing (HPF) and low
pressure flushing (LPF) schemes are available. In the HPF, the same YSP 3. Mathematical model
may be used so permeate production is not interrupted. In the LPF,
permeation is turned off (or YSP = 0) temporarily. In either flushing The BRO system shown in Fig. 1 or 2 consists of two sub-systems: the
scheme, the total system recovery (Ytot) is controlled by the ratio of CC RO and the CSTR, which are coupled and should be solved simulta­
filtration time (tFT) to flushing time (tFL), which resembles the operation neously. This is one major difference between the current work and the
strategy in the original CCRO [13,21]: previous CCRO work [21].
The dimensionless spatiotemporal RO model is the same as the one
YSP (tFT /tFL ) + δYSP
Ytot = (1) used in CCRO [21]:
YSP (tFT /tFL ) + 1
∂q* γ0
where δ = 1 for HPF and 0 for LPF. Alternatively, the time ratio 0= + (θ − c* )
∂x* 1 + Lp π0 c*
tFT Ytot − δYSP km
= (2)
tFL YSP (1 − Ytot ) ∂c*
∂c *
∂q* 1 ∂2 c* (5)
*
= − q* * − c* * +
Steps shown in Fig. 1(c)-(d) mirror those in Fig. 1(a)-(b), whereas the ∂t ∂x ∂x PeD ∂x*2
piston in the CSTR moves to the left. The four steps constitute one ∂θ a2
0 = * + q*n2
complete cycle (or two half-cycles) that can be repeated over and over. ∂x π0
It is worth pointing out that if the CSTR unit and associated valves
are removed from Fig. 1, the system becomes essentially a CCRO. where q* is flow rate in the feed channel (Q) divided by its value at the
Moreover, if the CSTR is replaced by a Plug Flow Reactor (PFR), or RO entrance (Q0). c* is salt concentration (C) normalized by the con­
mixing of the RO concentrate and the tank solution occurs only at the RO centration of the fresh feed (C0). θ is the transmembrane pressure (ΔP)
entrance, it also behaves similarly to a CCRO. divided by the osmotic pressure of the fresh feed (π0). t* is the actual
It is reported in literature that feed flow reversal may mitigate time t divided by the space time τ (τ = Q0/Vc, where Vc is the void
scaling formation in RO [27]. A cyclic design for BRO with flow reversal volume in the RO feed channel). x* = x/L is the dimensionless length
is shown in Fig. 2. Every time when the piston changes its direction of with 0 and 1 representing entrance and outlet of the RO stage respec­
movement, the feed to the RO unit is reversed. tively. L is the length of a pressure vessel, which is about 1 m for each RO
For a fair comparison among conventional steady-state RO, CCRO element. γ0 is a dimensionless parameter defined based on the combined
and BRO, the total raw intake rate, the total recovery rate and the total feed rate and the osmotic pressure of the fresh feed, or γ0 = AmLpπ0/Q0
number of RO elements are all fixed. Under these conditions, the cycle- (Am is membrane area per vessel and Lp is membrane hydraulic
average flux (for a time period lasting one CC filtration and one flushing permeability). km = a1q*n1 is the mass transfer coefficient, which varies
cycle) is the same in all designs [21]. The feed rate per vessel Q0, which as a function of cross velocity. PeD is the dispersive Peclet number. a1, n1,
is an important design parameter in CCRO and BRO, may be determined a2 and n2 are parameters that characterize mass transfer and pressure
from mass balance as below [21]: drop [19]. The model is based on the following assumptions: (1) the salt
rejection of the membrane is 100 %, (2) the concentration polarization
factor CPF = exp(Jw/km) ≈ 1 + Jw/km, and (3) the residence time dis­
tribution can be reasonably described by the dispersion model [19].
The boundary conditions for Eq. (5) is listed in Table 1, where cT* is
the salt concentration in the tank divided by the fresh feed concentration

4
M. Li Desalination 554 (2023) 116519

Table 1
dV
Boundary conditions in BRO simulations. = Q0 YSP
dt
Closed-circuit mode High-pressure flushing Low-pressure flushing ( *) (12)
mode mode d VcT
= Q0 YSP ⋅1
)⃒ )⃒ )⃒ dt
( 1 ∂c* ⃒⃒ ( 1 ∂c* ⃒⃒ ( 1 ∂c* ⃒⃒
c* − c* − c* −
The initial value for cT* in Eq. (12) is based on its end value in the
= = =
PeD ∂x* ⃒x* =0+ PeD ∂x* ⃒x* =0+ PeD ∂x* ⃒x* =0+
c*T
⃒
c*T
⃒
c*T
⃒
prior CC step. At the CSS, the refilling dynamics of right compartment
∂c* ⃒⃒ ∂c* ⃒⃒ ∂c* ⃒⃒ (during time period (2n − 1)(tFT + tFL) to tFT + (2n − 1)(tFT + tFL)) du­
=0 =0 =0
∂x* ⃒ ∂x * ⃒ ∂x* ⃒
x* =1 x* =1 x* =1 plicates that of the left (during time period 2n(tFT + tFL) to tFT + 2n(tFT +
q*(x*, t*)|x*=0 = 1 q*(x*, t*)|x*=0 = 1 q*(x*, t*)|x*=0 = 1
q*(x*, t*)|x*=1 = 1 − YSP q*(x*, t*)|x*=1 = 1 − YSP q*(x*, t*)|x*=1 = 1
tFL)), where n is the cycle number.
a2 a2 a2 It is not necessary to solve Eq. (12) by integration if one is only
θ (x* , t* ) |x* =0 = − θ (x* , t* ) |x* =0 = − θ (x* , t* ) |x* =0 = −
′ ′ ′

π0 π0 π0 interested in the final value of cT*. Note that the solution in the CSTR
a2
θ (x* , t* ) |x* =1 = − θ (x* , t* ) |x* =1 = − θ (x* , t* ) |x* =1 = − compartment at the end of the refilling step is a mixture of the fresh feed
′ ′ ′

π0
a2
(1 − YSP )n2
a2
(1 − YSP )n2 (with a volume of V0 − Vf) and the solution at the end of the flushing step
π0 π0 (with a volume of Vf). Using mixing rule,
( ) Vf
(cT* = CT/C0), which varies temporally according to the dynamics of the c*,ar
T = 1 + c*,br
T − 1 (13)
V0
CSTR. Danckwerts boundary conditions are used for solute transport.
The dynamic CSTR model has different forms, depending on the where cT*, br is cT* before refilling, or the final solution to Eq. (11). cT*, ar
operation mode of BRO. In the CC step (e.g., 0 to tFT), the total mass is cT* after refilling, which is the initial condition for the next CC
balance and salt mass balance in the CSTR compartment located in the filtration step.
closed circuit are: To find the CSS in BRO shown in Fig. 1, the following procedure is
used:
dV
= − Q0 YSP
dt
(6) 1. Assume C = C0 everywhere in the system. Note that the CSS is largely
d(VCT )
= Q0 (1 − YSP )Cc − Q0 CT independent of the initialconditions [21].
dt
2. Solve Eqs. (5) and (9) simultaneously for the CC filtration step. The
where V is volume of the mixing tank that is connected to the RO, Cc and end conditions are used as initial conditions in the next step.
CT are salt concentrations in the RO concentrate and in the tank, 3. Solve Eqs. (5) and (11) simultaneously for the flushing step.
respectively. 4. Update the end value of cT* in the step 3 using Eq. (13) for the
Using the initial condition V(0) = V0, it is determined from Eq. (6) refilling step. The end conditions are used as initial conditions in the
that: next step.
5. Repeat steps 2–4.
V = V0 − Q0 YSP t (7)
Therefore, the salt balance in Eq. (6) becomes: For BRO with flow reversal shown in Fig. 2, the concentration profile
is flipped before being used as initial condition for the next half-cycle.
(V0 − Q0 YSP t)
dCT
= Q0 (1 − YSP )(Cc − CT ) (8) Orthogonal collocation is used as the discretization method [20,21].
dt The mathematical model is converted to a set of ordinary differential
equations (ODEs) with a singular mass matrix and is solved using
which can be converted to a dimensionless form shown below:
ode15s in Matlab. Specifically, the spatial domain of RO (from 0 to L) is
dc*T 1 − YSP ( * ) discretized using N collocation points (N = 50 in this work). The points
= c − c*T (9) of collocation are chosen to be the roots of the orthogonal Jacobi
dt* V0 /Vc − YSP t* c
polynomials. The boundary conditions in Table 1 are then used to
where cc* = Cc/C0, which varies temporally according to the RO dy­ eliminate variables at the boundary points to yield 3(N − 2) equations
namics. The initial value for cT* in Eq. (9) is from its end value at the for the RO. The total number of equations for the coupled RO and CSTR
previous refilling step. system is 3N − 5. The CPU time for solving these equations is no more
In the flushing step (e.g., tFT to tFT + tFL), the CSTR has a constant than 0.4 s per half-cycle in a Dell Precision 7560 mobile workstation
volume Vf and a constant inlet concentration C0. The salt balance equipped with Intel Core i9-11950H processor and 64 GB RAM,
becomes: demonstrating its potential for real-time digital twin and control
applications.
dCT
Vf = Q0 (C0 − CT ) (10)
dt
4. Results and discussion
which may be converted to a dimensionless form shown below:
Simulation case studies are conducted for a high-recovery BWRO
dc*T 1 ( ) process to compare the performance among BRO, CCRO and conven­
*
= / 1 − c*T (11)
dt Vf Vc tional multi-stage RO [21,23]. The application of BRO to SWRO will
then be discussed in a qualitative way.
The initial value for cT* in Eq. (11) is based on its end value in the
preceding CC step.
The flushing step is followed by refilling before the next CC filtration. 4.1. Baseline case
In the refilling step (e.g., 0 to tFT for the compartment on the left or (tFT
+ tFL) to tFT + (tFT + tFL) for the compartment on the right), the CSTR In the baseline case, Qf = 1.01 m3/h (4.446 gpm), design flux = 24.5
dynamics may be described by: lmh, π0 = 0.62 bar, Ytot = 90 %, NE = 1, YSP = 10 %, tFL = (1 + Vf/Vc)τ,
and Vf/Vc = 0. To avoid singularity in the mathematical model, Vf/Vc is
set to be slightly above 0 (e.g., 0.001). BRO with flow reversal is chosen
for the study. The cycle-average flow, flux and recovery conditions

5
M. Li Desalination 554 (2023) 116519

match those in a conventional three-stage BWRO employing FilmTec filtration step. Such a trend is attributed to the transient behavior of salt
BW30–400 elements [23]. For BRO-HPF, Jw,FT = 24.5 lmh, Q0 = 9.1 m3/ concentration presented in Fig. 7. Before the CC filtration step, the RO
h (40.0 gpm), and tFT/tFL = 80. For BRO-LPF, Jw,FT = 24.8 lmh, Q0 = 9.2 membrane is still loaded with salt because it cannot be completely
m3/h (40.5 gpm), and tFT/tFL = 90. These parameters are the same as purged out in the flushing step. As the feed to the RO is reversed, the salt
those used in the CCRO study [21]. is concentrated near the entrance. It takes at least one space time to
The parameters for BW30–400 elements based on CFD simulations sweep out most of the brine in the RO channel. During this period, the
validated against plant experiments in the recovery range of 78–90 % spatial average salt concentration in the RO may reduce despite
are below [22]: permeate production. As a result, the applied pressure reduces tempo­
( )0.40 rarily. Subsequently, the average salt concentration in the RO increases
28 as permeate leaves the system at a constant rate. At the end of CC
a1 = 2.27 × 10− 2 Q0.40
0
NPV flushing step, the spatial salt concentration profile is completely flipped
n1 = 0.40 backwards in both BRO-HPF and BRO-LPF.
( )1.67 ( ) (14) Fig. 8 shows the spatial profiles of concentration at the CSS. They are
28 NE
a2 = 1.75 × 10− 4 Q1.67
0
switched back and forth between two consecutive half-cycles. Because
NPV 7
LPF requires a larger flux to maintain the same cycle-average water
n2 = 1.67 production, a higher salt concentration is observed at the end of the CC
filtration cycle [21]. This in turn, leads to a higher applied pressure in
where NPV is number of vessels in parallel. PeD is about 40 for one RO Fig. 6.
element [19]. Fig. 9 shows the temporal profile of concentrate concentration Cc and
Five full cycles (or ten half-cycles) were simulated to reach the CSS. ∫L
spatial average concentration (Cave = C = 0 Cdx/L) in 10 half-cycles.
The results, unless specifically stated, are based on the last cycle. The
spatiotemporal profiles of Q/Q0 are shown in Fig. 3. Because of the small Different from CCRO, the minima of C in BRO did not occur at the end
spatial variations in flow and flux, the CPF is approximately constant of flushing cycles. For LPF and Vf = 0, it is observed that the same
(1.12 in both cases), as shown in Fig. 4. In the continuous three-stage flushing efficacy model derived from CCRO can be used for BRO [21]:
design with inter-stage booster pumps, the CPF predicted by the [ ]
Ytot
model ranges from 1.07 to 1.14 [23]. The driving forces at five different c*i+1 = (1 − f ) + c*i + f (15)
1 − Ytot
progression times (t1 − t5) are shown in Fig. 5, indicating that the CC
filtration is operated far above the thermodynamic equilibrium. In other where c* is the dimensionless salt concentration after flushing and
words, the design flux necessitates an applied pressure much higher than before refilling, f is the flushing efficacy of the RO channel [19], and i
the osmotic pressure. During LPF, the concentrate flow is Q0, or there is and i + 1 represent two consecutive cycles. At the CSS, c*i = c*i+1 , or:
no net permeate production. These are similar to those in CCRO [21].
The spatiotemporal profiles of the dimensionless transmembrane Ytot (1 − f )
c*CSS = 1 + (16)
pressure (θ = ΔP/π0) are shown in Fig. 6. Different from the f (1 − Ytot )
quadrilateral-like shape in CCRO, θ in the CC filtration step decreases
Eq. (16) with f = 0.912 matches the cyclic data very well, as shown in
initially, then rises at an accelerated rate at all locations in the CC

Fig. 3. Spatiotemporal profiles of Q/Q0 at the CSS in (a) CC mode of BRO-HPF, (b) CC mode of BRO-LPF, (c) flushing mode of BRO-HPF, and (d) flushing mode of
BRO-LPF.

6
M. Li Desalination 554 (2023) 116519

Fig. 4. Spatiotemporal profiles of CPF at the CSS in (a) CC mode of BRO-HPF, (b) CC mode of BRO-LPF.

Fig. 5. Dimensionless driving force ((ΔP − CPF ⋅ π )/π0) at five progression times in (a) CC mode of BRO-HPF, (b) CC mode of BRO-LPF, (c) flushing mode of BRO-
HPF, and (d) flushing mode of BRO-LPF.

Fig. 9(d). In this case, c*CSS = 1.87, which is consistent with CCRO. concentration in the RO unit during the CC filtration step in the first
However, after the refilling step, c* will be different. cycle. If it is multiplied by 1.09, the profile matches the one at the CSS
When Vf = 0, V0/(V0 + Vc) = Ytot and Vc/(V0 + Vc) = 1 − Ytot. If one obtained from the mathematical model except the very beginning when
assumes that the solutions in the CSTR and in the RO are instantaneously the brine residue is being purged out from the RO channel and mixed
mixed at the beginning of the CC filtration step, the average salt con­ with the solution in the CSTR. Therefore, it is reasonable to use Eq. (17)
for the salt retention factor in BRO.
centration in the RO after the refilling step (c*,ar
CSS ) is:
An alternative way to derive Eq. (17) is as follows. Let c*,ar
CSS be the c
*

c* Vc + V0 in the closed-circuit right after the refilling step at the CSS, its values
c*,ar ≈ CSS
after the subsequent CC, flushing and refilling steps are c*,ar
CSS
V0 + Vc /(1 − Ytot ),
[ *,ar ] {[ *,ar ] CSS }
= c*CSS (1 − Ytot ) + (Ytot ) (17) cCSS /(1 − Ytot ) (1 − f) + f and cCSS /(1 − Ytot ) (1 − f) + f (1 −
{[ *,ar ] }
Ytot Ytot ) + Ytot respectively. From c*,ar
CSS = cCSS /(1 − Ytot ) (1 − f) + f (1 −
=1+ − Ytot
f Ytot ) + Ytot , it is determined that c*,ar
CSS = 1 + Ytot /f − Ytot .
After the results at the CSS are obtained, the specific energy con­
which is evaluated to be 1.09. sumption (SEC) normalized by the feed osmotic pressure (NSEC) can be
Fig. 10 shows the temporal profile of the spatial-average salt

7
M. Li Desalination 554 (2023) 116519

Fig. 6. Spatiotemporal profiles of ΔP/π 0 at the CSS in (a) CC mode of BRO-HPF, (b) CC mode of BRO-LPF, (c) flushing mode of BRO-HPF, and (d) flushing mode of
BRO-LPF.

Fig. 7. Spatiotemporal profiles of C/C0 at the CSS in (a) CC mode of BRO-HPF, (b) CC mode of BRO-LPF, (c) flushing mode of BRO-HPF, and (d) flushing mode of
BRO-LPF.

calculated. There are four terms in the energy calculations: (1) pump fresh feed in the flushing mode, and (4) hydraulic energy of the
energy to drive the fresh feed in the CC/refilling mode, (2) pump energy concentrate that may be recovered by an Energy Recovery Device (ERD).
to recycle the concentrate to the CSTR, (3) pump energy to drive the The theoretical NSEC can be calculated as follows [21]:

8
M. Li Desalination 554 (2023) 116519

Fig. 8. Spatial profiles of C/C0 at the CSS in (a) BRO-HPF and (b) BRO-LPF.

Fig. 9. Temporal profiles of (a) Cc/C0 in BRO-HPF, (b) Cc/C0 in BRO-LPF, (c) spatial average C/C0 in BRO-HPF, (d) spatial average C/C0 in BRO-LPF.

∫ *
tFT ∫ *
tFT solution to break down the NSEC in terms of thermodynamic restriction,
YSP θ(0, t* )dt* + (1 − YSP ) [θ(0, t* ) − θ(1, t* ) ]dt* design flux, friction resistance, concentration polarization, and salt
0 0
∫ * +t*
tFT ∫ * +t*
tFT retention, following the approach used previously [21]. The NSEC
imposed by thermodynamics is − ln (1 − Ytot)/Ytot = 2.56. The NSEC
FL FL
+ θ(0, t* )dt* − (1 − δYSP ) θ(1, t* )dt*
* * ( )
NSEC =
tFT tFT
(18) required by design flux is Ytot /γLPF = Jw, FT / Lp π0 = 14.49, where γLPF
* *
YSP tFT + δYSP tFL
= (AmLpπ0)/[(Q0YSPtFT + Q0tFL)/tFT] = 0.0621 [21]. Note that γ LPF is
The inlet pressures are shown in Fig. 11. Similar to those in CCRO, defined based on the total intake volume in one cycle divided by the CC
the temporal profile is discontinuous in LPF because the pressure serves step time, which is different from γ 0 used in Eq. (5). The NSEC due to salt
( )
to overcome the flow resistance only. The pressure drops are relatively retention is [ − ln(1 − Ytot )/Ytot ] c*,ar
CSS − 1 = [ − ln(1 − Ytot ) ](1/f − 1) =
constant in both cases. The NSEC for BRO-LPF is determined to be 21.5 0.22. The sum of these three terms is 17.3. When the effects of con­
with ERD (=17.91 + 3.54 + 0.69 − 0.64) and 22.1 without ERD centration polarization and friction loss are mostly eliminated in the
(=17.91 + 3.54 + 0.69). For BRO-HPF, it is 21.3 with ERD (=17.55 + mathematical model (by arbitrarily increasing the mass transfer coeffi­
3.43 + 2.62 − 2.32) and 23.6 without ERD (=17.55 + 3.43 + 2.62). All cient 100 times and decreasing the friction coefficient 100 times), the
are at lower levels than those in CCRO under the same design conditions NSEC with ERD for BRO-LPF also yields 17.3.
[21], implying that the proposed system is more energy-efficient. The pressure drop (ΔPL) assuming spatially constant flux along the
For BRO-LPF, it is possible to provide an approximate analytical RO (which holds reasonably well if YSP is small) may be calculated by

9
M. Li Desalination 554 (2023) 116519

CPF calculated from 1 + Jw,FT /km is 1.12 at both ends of the RO,
which is consistent with results shown in Fig. 4. The concentration po­
larization adds 0.76 to NSEC according to Eq. (21). When the mass
transfer coefficient is arbitrarily increased by 100 times in the mathe­
matical model, the NSEC reduces by 0.75. The numerical result and the
approximate analytical solution appear to be consistent.
The detailed energy consumption breakdowns for both BRO and
CCRO with LPF are shown in Table 2 for a comparison. It appears that
BRO and CCRO have the same friction loss and the same energy
requirement for the design flux. The CPF is also the same. However, BRO
outperforms CCRO in terms of energy efficiency in that (i) it is a ther­
modynamically favorable design as it mimics a dead-end membrane, (ii)
the effect of salt retention is smaller ([− ln (1 − Ytot)(1/f − 1)]/[Ytot(1 −
f)/f(1 − Ytot)] = [− (1 − Ytot) ln (1 − Ytot)]/Ytot, which is always less than
1 for any Ytot), (iii) the effect of concentration polarization is smaller
∫ given that the CPF functions as a multiplying factor on energy required
Fig. 10. Spatial average of salt concentration during CC filtration ( L0c*dx/L) in
the first cycle and at the CSS. The model is based on the concentration in the for overcoming thermodynamic restriction and salt retention.
first cycle multiplied by a constant factor (1 + Ytot/f − Ytot). When hydraulic energy of the brine leaving the system is fully
recovered, BRO-HPF is slightly more efficient than BRO-LPF because of a
[16]: smaller Jw,FT and a lower pressure drop. However, ERD is rarely installed
in high-recovery BWRO processes. In such a case, the brine energy is
ΔPL a2 1 1 − (1 − YSP )n2 +1 dissipated via the concentrate valve, and BRO-LPF is more efficient.
= (19)
π0 π 0 n2 + 1 YSP Similar trends are also observed in CCRO [21].
In LPF, YSP = 0, ΔPL/π 0 = a2/π0 using L'Hopital's rule. Despite that BRO outperforms CCRO, it still consumes more energy
Therefore, NSEC due to friction loss in BRO-LPF is estimated by: than the three-stage RO under the same intake, recovery and design flux
conditions (20.7 without booster pumps and 20.9 with booster pumps
1 1 − (1 − YSP )n2 +1 [23]). In the three-stage RO design with inter-stage booster pumps, the
Q0 tFT + Q0 tFL
a2 n2 + 1 YSP element-level recovery varies in the range of 8–14 %, and the averaged
NSECfriction,LPF =
π0 Q0 YSP tFT element-level recovery is 10 % [23]). If flow resistance is removed from
the three-stage model (by reducing the friction coefficient by 100 times),
1 1 − (1 − YSP )n2 +1 tFL
+ (20) the NSEC reduces by 3.1. It appears that the BRO designed with YSP =
a2 n2 + 1 YSP tFT
=
π0 YSP
1 1 − (1 − YSP )n2 +1 YSP (1 − Ytot ) Table 2
+
a2 n2 + 1 YSP Ytot Comparison of energy consumption breakdowns between CCRO-LPF and BRO-
=
π0 YSP LPF. Jw = 24.5 lmh, Ytot = 90%, YSP = 10%, π0 = 0.62 bar.

Using the baseline conditions, NSECfriction, LPF is calculated to be 3.99, Factors CCRO BRO
which differs 4 % from the value derived from the mathematical model Numerical Analytical Numerical Analytical
when the friction coefficient is reduced by 100 times (3.84). Thermodynamics 5.50 1 + Ytot/[2(1 2.56 − ln (1 − Ytot)/
The concentration polarization is expected to have a multiplying − Ytot)] = 5.50 Ytot = 2.56
( ) ( )
effect on the energy required to overcome both the thermodynamic re­ Design flux 14.49 Jw, FT / Lp π0 = 14.49 Jw, FT / Lp π0 =
striction and the salt retention, or 14.49 14.49
Flow resistance 3.84 Eq. (20) = 3.99 3.84 Eq. (20) = 3.99
⎧ [ ( )]
Concentration 0.75 Eq. (21) = 0.76 0.34 Eq. (21) = 0.33
⎪ (CPF − 1) − ln(1 − Ytot ) − ln(1 − Ytot ) 1 − 1 , BRO
⎪
⎪ polarization
⎨ Ytot f
NSECCP = [ ] Salt retention 0.87 Ytot(1 − f)/[f(1 0.22 − ln (1 − Ytot)
⎪
⎪ Ytot Ytot (1 − f ) − Ytot)] = 0.87 (1/f − 1) =
⎪
⎩ (CPF − 1) 1 + + , CCRO
2(1 − Ytot ) f (1 − Ytot ) 0.22
NSEC 25.4 25.6 21.5 21.6
(21)

Fig. 11. Temporal profiles of ΔP0/π0 in (a) BRO-HPF and (b) BRO-LPF.

10
M. Li Desalination 554 (2023) 116519

10% experiences a larger flow resistance than the three-stage design, as the one in three-stage RO (20.7 without booster pumps and 20.9 with
despite a much shorter flow length in the former. booster pumps [23]). When YSP is chosen to be 15 % or 20 %, BRO-LPF
Fig. 12 is a pie chart describing relative contributions of the five becomes 4 or 6 % energy efficient than three-stage RO, however, the CPF
factors to NSEC. For such a low-salinity BWRO process, design flux and increases to 1.15 or 1.17. The difference in NSEC between YSP = 20%
flow resistance are the two most significant factors. and YSP = 25% is fairly small because the uplifted CPF (1.19) offsets the
reduced friction loss. YSP = 25% may not be a good option for long-term
4.2. Effect of Vf/Vc operation of BRO because the concentrate flow Qc is out of the range
recommended by membrane manufacturers. Moreover, elevated con­
A simulation is done with Vf/Vc = 5 and tFL = (1 + Vf/Vc)τ for BRO- centration polarization may aggravate fouling [11]. To avoid over-
LPF. As shown in Fig. 13, it takes a few more cycles to reach the CSS. The fluxing, single element recovery rate is typically limited to 15 % or
salt concentration levels are elevated and a spatial variation is easily less [13].
observed. The applied pressure increases accordingly to maintain the Note that increasing YSP while keeping Q0 at its original value will
same recovery. not satisfy Eq. (3), or the comparison between BRO and steady-state RO
The NSEC of BRO at different values of Vf/Vc is shown in Fig. 14. The is not on the same footing.
lowest NSEC occurs at Vf/Vc = 0. However, the flushing fluid would be Adopting a YSP larger than 10 % and reducing Q0 accordingly in the
blocked from entering the CSTR. When Vf/Vc increases to 1, 5 and 10, original CCRO will reduce SEC as well. However, when YSP is varied in
the NSEC without ERD goes up by 3 %, 6 % and 7 %, respectively. the range of 10–25 %, the model shows that it always consumes more
It is speculated that the connection lines in the BRO system would energy than BRO or three-stage RO [21].
induce some effects similar to Vf. Reducing their volume relative to the Since the energy saving benefit of both the original and the improved
void channel space in RO may be desirable for the energetics of BRO and CCRO designs relative to conventional ROs is minimal or does not exist,
CCRO. For example, one piston-chamber assembly may be shared by one niche application is for treating feed water with high fouling po­
multiple pressure vessels connected in parallel in a scaled-up system. tential. By adopting a small YSP in CCROs, concentration polarization
The piston-chamber assembly used in the improved CCRO system may be mitigated at the expense of a higher friction loss. Moreover,
shown in Fig. 1 or 2 may be sized as follows. For LPF, the system re­ because of their relative compact sizes, CCROs may also be suitable for
covery Ytot = (V0 − Vf)/(V0 + Vc), or V0 = [Ytot/(1 − Ytot)]Vc + [1/(1 − decentralized high-recovery desalination applications.
Ytot)]Vf. Therefore, the total volume of the chamber is V0 + Vf = [Ytot/(1
− Ytot)]Vc + [(2 − Ytot)/(1 − Ytot)]Vf. Based on a user-specified Vf/Vc, a 4.4. Effect of extending flushing time tFL
total system recovery Ytot and an experimentally measured Vc (on the
order of 0.01 m3 per 8” RO element), the chamber size can be deter­ A simulation is done by doubling the flushing time in BRO-LPF (or tFL
mined. For a 90 % recovery, the volume of the reciprocating-piston = 2(1 + Vf/Vc)τ). To maintain the same recovery level, the CC filtration
chamber is estimated to be 150–200 l. The required volume may be time tFT must also be doubled according to Eq. (1). The simulation results
doubled if Ytot increases from 90 % to 95 %. To avoid a large CSTR for are shown in Fig. 16. It is found that the RO membrane is relatively clean
certain ultra-high recovery applications, one may consider a hybrid at the end of the flushing step, or c* = 1 everywhere. However, the salt
operation mode that begins with the improved CCRO and transitions to concentration level and the applied pressure are raised significantly at
the original CCRO near the end of the CC filtration step. This is some­ the end of the CC filtration step. In this particular case, the SEC increases
what analogous to the FRRR concept in the conventional design; about 2 %. Similar behaviors are observed in both numerical and
penalized energy consumption associated with undesirable entropy experimental studies of CCRO [13,21].
generation only occurs during part of the course of filtration [23]. Membranes must be operated below their recommended maximum
pressures to prevent mechanical damage. Therefore, it is important to
accurately determine the space time τ so that BRO and CCRO are
4.3. Effect of single-pass recovery YSP operated in the safe window.

Different values of YSP are used to study its effect on the performance 4.5. Effect of elements per vessel NE
of BRO-LPF and the results are shown in Fig. 15. When a larger YSP is
adopted in design, the recycle to raw feed ratio (r) and the time- To reduce capital expenditure in BRO and CCRO, it is common to
invariant feed rate per vessel (Q0) are both smaller. As a result, the house more than one element in each vessel [21]. As NE increases, a
friction loss reduces according to Eqs. (20) and (14). larger YSP may be adopted in process design while keeping the element-
When YSP increases to 12 %, the NSEC without ERD is about the same level recovery around 10 %. The pressure drop per vessel may increase
due to the extended length, however, the recycle to feed ratio (r) will
reduce and Q0 may also change. Its effect on SEC requires detailed
calculations.
A simulation study is done using NE = 3, YSP = 1 − (1 − 0.1)3 = 27.1
%. In this case tFT/tFL = 33.2, and Q0 = 10.4 m3/h. The results are shown
in Fig. 17. The concentration levels are similar to those in the baseline
case. However, the NSEC with/without ERD is about 0.8 higher than the
one in the baseline case. An estimation using Eq. (20) indicates that the
friction loss would lead to an increase about 0.7 in NSEC. Moreover,
because tFT/tFL is smaller, the average flux during CC filtration is higher,
resulting in additional 0.2 in NSEC.

4.6. Difference between BWRO and SWRO and potential applications of

BRO in SWRO

The proposed model can be applied to BRO of seawater if plant-

Fig. 12. Contributing factors to SEC in BRO-LPF. Feed osmotic pressure: 0.62 validated parameters (e.g., a1, a2, n1, n2 and Lp) are available. Some
bar. Cycle-average flux: 24.5 lmh. Recovery: 90 %. YSP = 10 %. qualitative discussions without detailed numerical simulations are

11
M. Li Desalination 554 (2023) 116519

Fig. 13. Effect of increasing Vf/Vc to 5 on (a) applied pressure, (b) spatial-average concentration and (c) spatial concentration profile at the CSS in BRO-LPF.

pressure drop. When booster pumps are installed in a multi-stage sys­

tem, permeate production shifts more towards downstream. The flow
rate level in the feed channel is relatively higher, so is the friction loss
[23]. Multi-stage with booster pumps has the most balanced flux, fol­
lowed by multi-stage, then single-stage. To avoid inadequate fluxes in
rear elements, single-stage design is not recommended for high-recovery
BWRO applications. In such a case, the driving factor is the balanced
flux, not the SEC. By the same token, selecting too large a YSP in BRO or
CCRO is discouraged due to overfluxing and elevated concentration
polarization even though it may save energy consumption by reducing
the flow resistance.
For SWRO, the γ parameter in modern plants is large (around 1), the
pressure drop effect is relatively insignificant and the operation is
slightly above the thermodynamic equilibrium. The transmembrane
pressure mainly serves to overcome the transmembrane osmotic pres­
sure. In such a case, the flux uniformity is in accord with energy effi­
Fig. 14. Effect of Vf/Vc on NSEC in BRO-LPF. YSP = 10 %. ciency. The internal staging in BRO and CCRO could be beneficial,
however, the detrimental effect of salt retention should also be taken
provided below. into consideration.
The difference between BWRO and SWRO has been discussed in a On the basis of Ytot = 50%, YSP = 10%, γ = 1, CPF = 1.07 (based on
few papers published by the author (see, e.g., [15,17]). For BWRO, the γ exp(0.7 × 0.1) = 1.07 [6]) and f = 0.9 [19], it is estimated using
parameter in modern plants is small (about 0.05), the pressure drop equations in Table 1 that the contributions of thermodynamic restric­
effect is significant and the operation is far away from the thermody­ tion, design flux, salt retention, concentration polarization to NSEC in
namic equilibrium. The applied pressure mainly serves to meet the flux SWRO are 1.39, 0.55, 0.08 and 0.11. For CCRO, they are 1.50, 0.55, 0.11
requirement and to overcome the flow resistance. It is observed that the and 0.12, respectively. The contribution of flow resistance to NSEC
transmembrane hydraulic pressure at the end of the RO is still much estimated using plant conditions is 0.11 (assuming a pressure differen­
higher the transmembrane osmotic pressure, and the hydraulic energy of tial of 1.5 bars per pressure vessel and a π0 of 27 bar). Apparently,
the brine is dissipated via the concentrate valve [24]. From a viewpoint thermodynamic restriction and design flux are the top two contributing
of energy performance, single-stage design would be the most efficient, factors, which are drastically different from the BWRO case shown in
followed by multi-stage, then multi-stage with booster pumps [15,23]. Fig. 12. It is seen that BRO is about 7 % energy-efficient than CCRO.
The energy efficiency appears to be contradicting to the balanced design If the effects of concentration polarization and flow resistance are not
considerations in BWRO. In single-stage design, the cross velocity taken into consideration, the NSEC for BRO and CCRO are estimated to
quickly drops along the feed direction, and therefore, it has the least be 2.01, 2.16 for Ytot = 50% and γ = 1. Based on the same assumptions
and process conditions, NSEC is 2.08 and 1.93 for single-stage and two-

12
M. Li Desalination 554 (2023) 116519

Fig. 15. Effect of YSP on (a) Q0, Qc and recycle to feed ratio r = (1 − YSP)/YSP and (b) NSEC in BRO-LPF.

Fig. 16. Effect of doubling flushing time on (a) applied pressure, (b) spatial-average concentration and (c) spatial concentration profile at CSS in BRO-LPF.

stage ROs [18]. In such a case, the energy performance is ranked as: two- 5. Concluding remarks
stage RO > BRO > single-stage RO > CCRO for seawater desalination.
Note that the advantage of BRO and CCRO is manifested when a A piston-tank assembly and several valves are placed in front of the
larger γ is adopted in design [18]. As γ becomes sufficiently large, the CCRO system to improve its energetic performance. Similar to CCRO,
( )
contribution of design flux to NSEC (Ytot /γLPF = Jw,FT / Lp π0 ) becomes continuous permeate production is possible in the proposed system if the
small. In single- or multi-stage RO operated at steady state, a plateau in flushing step is done at high pressures.
the NSEC is quickly reached as γ increases [14]. When γ is quadrupled, The improved CCRO system is a thermodynamically favorable design
BRO may be similar to three-stage RO in terms of SEC [18]. Therefore, as it mimics a dead-end membrane, which reduces undesired entropy
the continuous development of highly-permeable membranes may favor generation. Moreover, the effect of salt retention is smaller than the one
BRO in future plant designs. However, it is noted that the maximum in the original CCRO under similar design conditions. The concentration
applied pressures at the end of the CC filtration step of BRO and CCRO polarization is at the same level, however, its penalized effect on the
are higher than what is observed in current plants operated at steady- energy consumption of the proposed system is smaller.
state. This calls a need for developing ultra-high pressure membranes The energy benefit of using BRO for treating low-salinity water is
[32] in order to advance the BRO and CCRO technologies. marginal on the basis of the same conditions used in a traditional plant.
Adopting a larger YSP may save energy, at the expense of increased
concentration polarization. Based on YSP = 15 %, it may consume 4 %

13
M. Li Desalination 554 (2023) 116519

Fig. 17. Effect of using 3 elements per vessel on (a) applied pressure, (b) spatial-average concentration and (c) spatial concentration profile at CSS in BRO-LPF.

less energy than three-stage RO with an averaged element-recovery of RO reverse osmosis

10 %. If Q0 and YSP are carefully chosen, BRO may have reduced fouling SEC specific energy consumption
risk due to a relatively uniform and controlled CPF, whereas the energy SWRO seawater reverse osmosis
consumption may be higher. Both the original and improved CCRO ΔP transmembrane hydraulic pressure
designs have a small footprint, which is advantageous for decentralized, ΔPL pressure drop
high-recovery desalination applications. When used for seawater desa­ δ 0 for LPF and 1 for HPF
lination, the energy performance of BRO is in between single-stage and γ membrane capacity intake ratio
two-stage ROs based on typical flux and recovery conditions used in π osmotic pressure
industry. Additionally, the availability of highly permeable membranes τ space time
will favor BRO in future plant designs. θ transmembrane pressure divided by feed osmotic pressure
The impact of salt retention on SEC in CCRO and BRO is only a few Am membrane area
percent if operated under optimal conditions. The applied pressure as­ C salt concentration
cends to a higher level and the energy performance degrades if the c* dimensionless salt concentration
flushing time goes beyond one space time. Therefore, it is essential to f flushing efficacy
accurately determine its value under plant operating conditions. Jw water flux across membrane
The peak pressures in both CCRO and BRO will be greater than those km mass transfer coefficient
observed in current plants operated at steady-state under similar design L total length of a pressure vessel
conditions. Lp hydraulic permeability
PeD dispersive Peclet number
Nomenclature Q volumetric flow rate
q* dimensionless volumetric flow rate
BRO batch reverse osmosis r recycle to fresh feed ratio
BWRO brackish water reverse osmosis t time
CCRO closed-circuit reverse osmosis t* dimensionless time
CPF concentration polarization factor V0 initial volume of CSTR compartment in the closed circuit
CSS cyclic steady state Vc volume of the void space in RO
CSTR continuous stirred tank reactor Vf final volume of CSTR compartment in the closed circuit
HPF high pressure flushing x length
LPF low pressure flushing x* dimensionless length
NSEC normalized specific energy consumption Y recovery
ODE ordinary differential equation 0 inlet
PDE partial differential equation ave average
PFR plug flow reactor c concentrate

14
M. Li Desalination 554 (2023) 116519

FL flushing [8] J. Gilron, M. Waisman, N. Daltrophe, N. Pomerantz, M. Milman, I. Ladizhansky,

E. Korin, Prevention of precipitation fouling in NF/RO by reverse flow operation,
FT filtration
Desalination 199 (2006) 29–30.
i index [9] H. Gu, A.R. Bartman, M. Uchymiak, P.D. Christofides, Y. Cohen, Self-adaptive feed
T tank flow reversal operation of reverse osmosis desalination, Desalination 308 (2013)
tot tot 63–72.
[10] H. Gu, M.H. Plumlee, M. Boyd, M. Hwang, J.C. Lozier, Operational optimization of
closed-circuit reverse osmosis (CCRO) pilot to recover concentrate at an advanced
CRediT authorship contribution statement water purification facility for potable reuse, Desalination 518 (2021), 115300.
[11] E.M.V. Hoek, A.S. Kim, M. Elimelech, Influence of crossflow membrane filter
geometry and shear rate on colloidal fouling in reverse osmosis and nanofiltration
Mingheng Li: Conceptualization, Methodology, Software, Writing, separations, Environ. Eng. Sci. 19 (2002) 357–372.
Visualization, Funding acquisition. [12] E. Hosseinipour, K. Park, L. Burlace, T. Naughton, P.A. Davies, A free-piston batch
reverse osmosis (RO) system for brackish water desalination: experimental study
and model validation, Desalination 527 (2022), 115524.
Declaration of competing interest [13] T. Lee, A. Rahardianto, Y. Cohen, Multi-cycle operation of semi-batch reverse
osmosis (SBRO) desalination, J. Membr. Sci. 588 (2019), 117090.
[14] M. Li, Minimization of energy in reverse osmosis water desalination using
The authors declare that they have no known competing financial constrained nonlinear optimization, Ind. Eng. Chem. Res. 49 (2010) 1822–1831.
interests or personal relationships that could have appeared to influence [15] M. Li, A unified model-based analysis and optimization of specific energy
consumption in BWRO and SWRO, Ind. Eng. Chem. Res. 52 (2013) 17241–17248.
the work reported in this paper. [16] M. Li, Predictive modeling of a commercial spiral wound seawater reverse osmosis
module, Chem. Eng. Res. Des. 148 (2019) 440–450.
Data availability [17] M. Li, Dynamic operation of batch reverse osmosis and batch pressure retarded
osmosis, Ind. Eng. Chem. Res. 59 (2020) 3097–3108.
[18] M. Li, Effects of finite flux and flushing efficacy on specific energy consumption in
Data will be made available on request. semi-batch and batch reverse osmosis processes, Desalination 496 (2020), 114646.
[19] M. Li, Residence time distribution in RO channel, Desalination 506 (2021),
115000.
Acknowledgement [20] M. Li, A spatiotemporal model for dynamic RO simulations, Desalination 516
(2021), 115229.
The author would like to acknowledge financial support from the [21] M. Li, Cyclic simulation and energy assessment of closed-circuit RO (CCRO) of
brackish water, Desalination 545 (2023), 116149.
National Science Foundation (CBET-2140946). Any opinions, findings, [22] M. Li, T. Bui, S. Chao, Three-dimensional CFD analysis of hydrodynamics and
and conclusions or recommendations expressed in this material are concentration polarization in an industrial RO feed channel, Desalination 397
those of the author and do not necessarily reflect the views of the Na­ (2016) 194–204.
[23] M. Li, N. Chan, J. Li, Novel dynamic and cyclic designs for ultra-high recovery
tional Science Foundation.
waste and brackish water RO desalination, Chem. Eng. Res. Des. 179 (2022)
473–483.
References [24] M. Li, B. Noh, Validation of model-based optimization of brackish water reverse
osmosis (BWRO) plant operation, Desalination 304 (2012) 20–24.
[25] B. Liberman, L. Eshed, G. Greenberg, Pulse flow RO - the new RO technology for
[1] W.T. Andrews, D.S. Laker, A twelve-year history of large scale application of work-
waste and brackish water applications, Desalination 479 (2020), 114336.
exchanger energy recovery technology, Desalination 138 (2001) 201–206.
[26] S. Lin, M. Elimelech, Staged reverse osmosis operation: configurations, energy
[2] A.R. Bartman, C.W. McFall, P.D. Christofides, Y. Cohen, Model-predictive control
efficiency, and application potential, Desalination 366 (2015) 9–14.
of feed flow reversal in a reverse osmosis desalination process, J. Proc. Contr. 19
[27] N. Pomerantz, Y. Ladizhansky, E. Korin, M. Waisman, N. Daltrophe, J. Gilron,
(2009) 433–442.
Prevention of scaling of reverse osmosis membranes by “zeroing” the elapsed
[3] S. Cordoba, A. Das, J. Leon, J.M. Garcia, D.M. Warsinger, Double-acting batch
nucleation time. Part I. Calcium sulfate, Ind. Eng. Chem. Res. 45 (2006)
reverse osmosis configuration for best-in-class efficiency and low downtime,
2008–2016.
Desalination 506 (2021), 114959.
[28] S.M. Riley, D.C. Ahoor, K. Oetjen, T.Y. Cath, Closed circuit desalination of O&G
[4] P.A. Davies, A. Afifi, F. Khatoon, G. Kuldip, S. Javed, S.J. Khan, Double–acting
produced water: an evaluation of NF/RO performance and integrity, Desalination
batch–RO system for desalination of brackish water with high efficiency and high
442 (2018) 51–61.
recovery, in: Desalination for the Environment: Clean Water and Energy, European
[29] S. Sircar, Pressure swing adsorption, Ind. Eng. Chem. Res. 41 (2002) 1389–1392.
Desalination Society, Rome, Itay, 2016.
[30] R.L. Stover, Industrial and brackish water treatment with closed circuit reverse
[5] P.A. Davies, J. Wayman, C. Alatta, K. Nguyen, J. Orfi, A desalination system with
osmosis, Desalin. Water Treat. 51 (2013) 1124–1130.
efficiency approaching the theoretical limits, Desalin. Water Treat. 57 (2016)
[31] A. Waite, W. Broley, M. Ingalsbe, S. Wang, B. Trussell, Can closed circuit
23206–23216.
desalination boost brackish groundwater and recycled water recovery by squeezing
[6] Dupont, FilmTec reverse Osmosis membranes technical manual, Online, https://
concentrate?, in: AWWA California-Nevada Section Annual Fall Conference,
www.dupont.com/content/dam/dupont/amer/us/en/water-solutions/public/do
Rancho Mirage, CA, 2018.
cuments/en/45-D01504-en.pdf, 2020. (Accessed 25 April 2020).
[32] J. Wu, B. Jung, A. Anvari, S. Im, M. Anderson, X. Zheng, D. Jassby, R.B. Kaner,
[7] K. Duraiswamy, A. Chellappa, G. Smith, Y. Liu, M. Li, Development of a high-
D. Dlamini, A. Edalat, E.M.V. Hoek, Reverse osmosis membrane compaction and
efficiency hydrogen generator for fuel cells for distributed power generation, Int. J.
embossing at ultra-high pressure operation, Desalination 537 (2022), 115875.
Hydrog. Energy 35 (2010) 8962–8969.

15