Korean J. Chem. Eng.

, 38(2), 366-379 (2021) pISSN: 0256-1115

DOI: 10.1007/s11814-020-0697-9 eISSN: 1975-7220
INVITED REVIEW PAPER INVITED REVIEW PAPER

Mathematical and experimental modeling of reverse osmosis (RO) process

Zeinab Hadadian*,†, Sina Zahmatkesh*, Mostafa Ansari*, Ali Haghighi*, and Eskandar Moghimipour**

*Faculty of Civil Engineering and Architecture, Shahid Chamran University of Ahvaz, Ahvaz, Iran
**Nanotechnology Research Center, Ahvaz Jundishapur University of Medical Sciences, Ahvaz, Iran
(Received 13 May 2020 • Revised 26 September 2020 • Accepted 15 October 2020)

AbstractThis paper provides a mathematical simulation model for the reverse osmosis (RO) process with series ele-
ments. A mathematical simulation model was developed based on the mass, material and energy balances considering
the concentration polarization. The simulation model is open-source and easy to couple with other computational tools
like optimization algorithms and SCADA1 applications. An RO laboratory pilot was also set up in the Hydraulic Lab of
Shahid Chamran University of Ahvaz to validate the simulation results. Comparing the results of the simulation model
with the experiments and ROSA commercial software, the proposed simulation model functions well and is reliable. The
comparisons indicate that the simulation results are over 96% close to ROSA and over 80% close to experimental results.
Keywords: Desalination, Reverse Osmosis, Simulation, Experimental, Mathematical Modeling

INTRODUCTION energy recovery systems [5]. Geraldes et al. optimized the simula-
tion of a two-stage desalination system and the considered spiral
The reverse osmosis (RO) method for desalination is currently wound membrane (the most common type). They also used the
the most popular technology for seawater and brackish water desali- experimental data to identify specific parameters in design equa-
nation [1]. The RO process has several advantages over the other tions. That paper focused on optimizing the pressure and flow of
desalination methods, especially in terms of energy consumption feed water for various recovery rates [6]. Guria et al. presented a mul-
and efficiency. For each RO system, the membrane performance tiobjective optimization problem for seawater desalination, using
should be evaluated to determine the type of membrane, recovery, spiral wound and tubular membranes. That paper does not include
and number of membrane elements. The construction and opera- experimental data [7]. An optimization model for different feed
tion costs, as well as the permeate concentration and the permeate water concentrations was presented by Lu et al. in 2007 [8]. Choi
flux, are critical factors to determine for the design of an RO sys- et al. presented an RO and forward osmosis (FO) computer pro-
tem. Commercial software, like ROSA and IMSDesign, has been gram; the model was developed only for a one-stage system [9]. Du
widely used in many investigations and industrial projects for pre- et al. presented a simulation-optimization model for seawater and
dicting the performance of the RO systems [2]. These commercial brackish water, using spiral wound membranes. They considered
models have been mostly released by the membrane manufacturers the permeate flux as a constraint in addition to the constraints on
and therefore are useful for their productions. For research objec- the membrane manufacturer. The model also included a maximum
tives, one needs to have an open-source model to extend the sim- of two-stage membrane and considered energy recovery systems
ulations and couple the RO models with the optimization algorithms [10-12]. Altaee estimated the performance of RO systems with sev-
and other mathematical techniques. eral elements based on the solution-diffusion model, and most of
Villafafila and Mujtaba developed a simulation and optimization the equations used in that paper were from the empirical equations
model for seawater and brackish water. They optimized energy con- presented in the DOW design guide, which included single-stage
sumption, recovery, and the number of tubular membranes. They systems and has no experimental data. It was a model that was
considered the number of pressure vessels, pressure values, mem- unable to establish new constraints or to optimize the system [2].
brane diameter, and feed water flux as the constraints [3]. Barello Saavedra et al. presented a design method for RO brackish water
et al. provided experimental data that examined the pure water per- plants, which was based on the application of maximum available
meability and salt permeability constant for this type of membrane recovery without scaling of any inorganic compounds presented in
element (tubular) [4]. Marcovecchio et al. optimized the simula- water [13,14]. Choi and Kim presented a simulation and optimiza-
tion of seawater reverse osmosis (SWRO) system with a two-stage tion model for RO systems with two stages, including spiral wound
arrangement. That work included hollow fiber membranes and membranes, but did not provide experimental work [15]. Kotb et
al. examined an SWRO system with a maximum of three-stages
1
Supervisory Control And Data Acquisition [16]. Haluch et al. evaluated the experimental and semi-empirical
model of a small-capacity reverse osmosis desalination unit. They
†
To whom correspondence should be addressed. found that the semi-empirical model predictions agreed with their
E-mail: z-hadadian@phdstu.scu.ac.ir experimental counterparts within the measurement uncertainty
Copyright by The Korean Institute of Chemical Engineers. threshold [17]. Chee et al. investigated the performance evaluation

366
 Mathematical and experimental modeling of reverse osmosis (RO) process 367

of RO desalination pilot plants using the ROSA Simulation soft- each element. Unlike other commercial software (ROSA, IMSDe-
ware. They found that in terms of flux and recovery ratio, the sim- sign, etc.), the proposed model is open-source and easy to couple
ulated results and the experimental data showed a marginal dis- with other computational tools like optimization algorithms, and
crepancy with deviations <2% and <8%, respectively. Their find- since it has been based on the mathematical equations of the RO
ings also confirmed the feasibility of adopting ROSA software to process (the ROSA software utilizes experimental equations that
verify the performance of a pilot plant with all operational param- control Filmtec membranes), it is capable of using all types of mem-
eters being ideally optimized [18]. Al-Obaidi et al. evaluated the branes manufactured in different companies. In existing commer-
performance of a medium-sized industrial BWRO desalination plant cial software, the total system recovery is considered as input, while
of the Arab Potash Company using the mathematical model and the present model considers the recovery of each stage separately.
the real data [19]. Chen and Qin developed mathematical model- An RO laboratory pilot with 50 m3/day production capacity has
ing of glucose-water separation through reverse osmosis (RO) mem- been also set up in the Hydraulic Lab of Shahid Chamran Univ-
brane to research the membrane’s performance during the mass ersity of Ahwaz to validate the simulation results. Two single-ele-
transfer process. They validated the model using experimental results ment and multi-element cases were simulated using the proposed
and found that the calculated results were consistent with the experi- simulator model and the results were compared with the ROSA
mental data [20]. Maure and Mungkasi obtained a mathematical software simulation and the experimental results which are dis-
model using numerical integration for the reverse osmosis system cussed.
[21]. Li proposed a predictive mathematical model based on the
solution-diffusion theory for a commercial spiral wound SWRO MATERIALS AND METHODS
module. They concluded that the mathematical model with the
parameters obtained from the experimental data can predict the 1. Model Development
flow of water and salt as well as the pressures under different feed An RO mathematical simulation model was developed based
conditions of temperature, flow, and pressure with a mean error on mass, material, and energy balances for a given configuration.
4% [22]. Gaublomme et al. developed a generic steady-state model For more realistic modeling, the concentration polarization is also
for RO and applied it to a unique three-year data set of a full-scale taken into account. The related equations used to simulate the RO
RO process. They validated the model with online conductivity data process are reported in Table 1.
as input taking into account the uncertainty originating from online 2. Membrane Characteristics
sensors and compared to the commercial software Winflows. They The full characteristics of the membranes provided by the man-
found that the model has satisfactory results, i.e., an average devia- ufacturers are critical parameters for the simulation of the RO sys-
tion from the data at 2.7%, 12.7%, 34.1% and 18.7%, respectively, tems. The required data include the active membrane area, maximum
for the recovery, the concentrate pressure, the permeate, and con- operating pressure, salt rejection, pure water permeability constant
centrate solute concentration [23]. Siegel et al. developed a mathe- (Am), feed spacer, pressure drop in element, length of the element,
matical model describing the RO enrichment process using a novel diameter of the element, spacer diameter and thickness.
device. They created it in MATLAB Simulink software and vali- 3. Arrangements of a Multi-stage RO System
dated it with experimental results. Using the calculation of the mean The schematic of a multi-stage RO unit is shown in Fig. 1.
relative error between the model and the experimental results, they For a three-stage system, there are three splitter points (F, R1,
concluded that the model is useful for describing the RO system and R2) with three split ratios ,  and  between 0 and 1 so that
and the RO device is suitable for the enrichment of estrogens prior their summation at each point is 1. The flow rate of the feed solu-
to instrumental or in vitro analysis [24]. Ligaray et al. presented a tion to each stage can be calculated by Eq. (16) [16].
novel energy self-sufficient desalination system design that incor-
porates rechargeable seawater batteries as an additional energy stor-  Qf1 fQf

age system. They predicted the experimental data using the ROSA  Qf2  fQf  R1QR1 (16)
model to determine the configuration of the lowest energy con- 
 Qf3  fQf  R1QR1 R2QR2
sumption and highest charging rate. The results showed that the
seawater battery achieved satisfactory desalination performance where, QR1 is retentate flow rate from the first stage (m3/s), QR2
[25]. Mansour et al. focused on employing an energy recovery sys- is retentate flow rate from the second stage (m3/s), QF1 is feed flow
tem (ERS) to enhance the performance of the small RO plant for rate to the first stage (m3/s), QF2 is feed flow rate to the second stage
remote areas using an experimental pilot and simulation model. (m3/s), QF3 is feed flow rate to the third stage (m3/s),  is the frac-
Their obtained results showed good agreement between experi- tion of stream branching to the left from a split point,  is the frac-
mental and simulation model values. They evaluated the cost anal- tion of stream branching to the right from a split point, and  is the
ysis of the small RO desalination plant with and without ERS and fraction of stream branching straight forward from a split point.
showed a significant reduction in total cost [26]. 4. The Calculation of Permeate Water at Each Stage
In this paper, we present a mathematical simulation program The purpose of the RO simulation is to determine the pressure
for a multi-element RO system and consider multiple elements in required at each stage to obtain the required permeate water. Each
each pressure vessel. The output parameters of the program include stage may include several pressure vessels in parallel, and each pres-
the feed pressure, the permeate concentration, the recovery of each sure vessel includes several membranes in a series. The number of
element in the pressure vessel, and the permeate flow generated by pressure vessels in each stage depends on the total number of the

Korean J. Chem. Eng.(Vol. 38, No. 2)

368 Z. Hadadian et al.

Table 1. Equations for the RO process simulation

Meaning Equation No. Ref.
Feed flow rate QF  Q R  QP (1) [1]
Permeate flow rate QP  TCFjwA  (2) [1]
 o
 1, TF  25 C

  25,000
---------------  -------------------------
1 1
- ------------------------- - o
Temperature correction factor TCF =  exp  R  T0  273.15  TF  273.15 , TF  25 C (3) [10]

  20,000-  ------------------------- - , TF  25 oC
1 1
 exp  --------------
R
-  -------------------------
 T0  273.15 TF  273.15

Material balance QF  CF   QP  CP    QR  CR  (4) [1]
Feed concentration CF  CP/1 rejection (5) [2]
Permeate water flux jw  Am P     AmPeff (6) [1]
Residual transmembrane pressure Peff  Pf  Pp  Pin  Pf/2    w  p (7) [1]
3
  nMR T  273.1510
Van’t hoff’s equation 3
(8) [2]
For NaCl   2C/0.0585 RT  273.15 10
Cw  Cp J w d Jw
-----------------  exp  -------   ---
-
Concentration on the feed side membrane wall Cf  Cp  DS  exp  k  (9) [1]
Cw  CP   Cf  Cp exp Jw/k
0.4 0.17 0.77 0.77
Mass transfer coefficient k  0.5510 Re Sc  Cf/ DS/d  (10) [9]
2
  498.4m  248,400m  752.4mCf
Density (11) [1]
4
m 1.0069  2.757  10 Tf
6
Viscosity  1.234  10 exp 0.00212Cf 1,965/Tf  273.15 (12) [1]
6 3
Diffusivity DS  6.725  10 exp 0.1546  10 Cf  2,513/Tf  273.15 (13) [1]
Aeff  VT/l*
Effective membrane area (14) [23]
 1 VSP/VT
N
Number of pressure vessels in the ith stage NPVi  -------i (15)
Nei

Fig. 1. Schematic of multi-stage RO unit [16].

February, 2021
 Mathematical and experimental modeling of reverse osmosis (RO) process 369

membranes and the number of membranes in each pressure ves- P

QPz  A*TCF*AmPf  Ppz  Pinz  ---------fz-
sel, which is obtained from Eq. (15). The system recovery depends 2
on the permeate water (Qp) and the feed flow (Qf ) as follows: 2RT  273.15
  ----------------------------------- *10  Cwz  CPz
3
0.0585
Q
Recovery  ------p (17) P
Qf QPz  A*TCF*AmPf  Ppz  Pinz  ---------fz- (22)
2
Since the amount of permeate water should be constant, by 2RT  273.15 J
  ----------------------------------- *10  CPz  Cfz  Cpz exp  ---w-  CPz 
3
determining the system recovery the amount of feed flow is deter- 0.0585 kz
mined. If we assume the recovery in the first, second, and third
P
stages, respectively, rec1, rec2, and rec3, the amount of permeate QPz  A*TCF*AmPf  Ppz  Pinz  ---------fz-
2
water at each stage can be calculated according to Eq. (16) as fol-
2RT  273.15 QPz  
  ----------------------------------- *10  *CPz ---------------------------- 1 * exp  -----------
3 1
lows: -
0.0585 1 rejection A*kz 
Q
Qf1 fQf  Qf  ------f1- In each pressure vessel, the feed flow rate and feed pressure inlet
f
Q
to each element are equal to the retentate flow rate and retentate
Qf  ------f1- , QR1   1 rec1 Qf1 pressure outlet from the previous element.
f
Qf2  R1QR1 fQf
QFz1  CFz1 QFz  CFz  QPz  CPz (23)

Qf2  Qf11 rec1R1 ----f  (18)
 f QFz1 QFz  QPz (24)
Q
Qf  ------p Qp Qp1   By using Eqs. (5), (23), and (24), the permeate concentration at
rec
-  ---------- 1 rec1R1 ----f 
---------
rec2 rec1  f the z+1th element is obtained as follows:
CPz
Qf3  R2QR2  fQf  R1QR1 QFz  --------------------------- -  QPz  CPz
1 rejection
 CPz1 1 rejection ------------------------------------------------------------------- (25)
 R21 rec2 Qf2  R11 rec1Qf1 ----f Qf1 (19) QFz  QPz
f
Qp3 R21 rec2  R11 rec1 f  For each pressure vessel, the summation of permeate flux of all
-  ------------------------------ Qp2   ------------------------------
---------  --------------
- Q
rec3 rec2  rec1 frec1 p1 elements is equal to the required permeate flux of the pressure vessel.
By using Eqs. (18) and (19) along with the continuity equation, Nei
QPpvj   QPz (26)
the values of productive permeate water are obtained at each stage z1
as follows:
Then, for all elements in the pressure vessel, a set of nonlinear
 equations consisting of permeate flow rate (Eq. (22)) and concen-
 1 rec1  ---- f
tration (Eq. (25)) as variables with Eq. (26) is formed, which is solved
 R1
f Qp2
 --------------------------------------- Qp1 ---------- 0 by the Newton-Raphson method and obtained feed water pressure
 rec1 rec2 of that stage and permeate flow rate and concentration in each ele-
  1 rec1  f  R21 rec2 Qp3
  ------------------------------
R1
- Qp1 ------------------------------
 -------------- Qp2  ---------
- 0 (20) ment.
 rec1 frec1  rec2 rec3 Finally, the permeate concentration of the jth pressure vessel (Cppvj)

 3 is calculated as follows:
  Qi  Qp
 i1 Nei
  QPz*CPz
------------------------------
CPpvj  z1 (27)
Finally, the feed rate intake at each stage can be determined by QPpvj
having Qp1, Qp2, Qp3, the recovery, and the coefficients of the split-
Total permeate concentration and recovery of the system are cal-
ter points.
culated as follows:
5. The Calculation of Operating Parameters for Each Element
of Pressure Vessel 3

Given that there is uniformity of the pressure vessels at each stage,  QPi*CPi
----------------------------
Cptotal  i1 (28)
the feed flow and permeate flow rates are split equally among the QP
pressure vessels and are obtained for each pressure vessel (QFPVj,
Q QP QP
QPPVj). In each pressure vessel permeate water produced by the rectotal  -----P-  --------------  ---------------------------------
- (29)
Qf Qf1/f Qp1/rec1*f
elements is calculated using Eqs. (2), (6), and (7) as in the follow-
ing equation: 6. The Simulation Algorithm
The solving procedure of the RO equations is illustrated in Fig.
P
QPz  A*TCF*AmPf  Ppz  Pinz  ---------fz-   wz   pz (21) 2. The simulation model is mathematically nonlinear and implicit
 2  
and should be iteratively solved. The proposed simulation algorithm
By placing Eqs. (8), (9), and (5) in Eq. (21): consists of the following steps:

Korean J. Chem. Eng.(Vol. 38, No. 2)

370 Z. Hadadian et al.

Fig. 2. Flowchart of RO simulation model.

Fig. 3. A graphical user interface (GUI) window.

February, 2021
 Mathematical and experimental modeling of reverse osmosis (RO) process 371

1. The input data include the feed water and membrane charac- 9. Permeate concentration of pressure vessel (Cppvj) is calculated
teristics, recovery of each stage, required permeate water, number using Eq. (27).
of elements, the number of series element in each pressure vessel, 10. Steps 4 to 9 are repeated for all stages.
and the split ratios for the membrane’s arrangement are introduced Based on the above algorithm, simulation software with a graph-
to the model. ical user interface (GUI) was designed in MATLAB. This algo-
2. QPi is calculated using Eq. (20) for each stage (i=1 : 3) then rithm makes it possible to do the RO simulation by entering the
QFi and QRi are obtained using Eqs. (17) and (1), respectively. initial data, feed water, and membrane characteristics to obtain the
3. Set i=1 for the first stage required feed pressures for stages, feed flow rate, retentate flow rate,
4. QFPVj, QPPVj are calculated for each pressure vessel (j=1: NPVi) and the permeate concentration as displayed in Fig. 3.
in ith stage and because of the uniformity of the pressure vessels at
each stage, QFi and QPi are split equally among the pressure vessels. EXPERIMENTAL
5. z, z, Dsz, Rez, Scz, and kz are calculated using Eqs. (10)-(13)
for each element. 1. Experimental Pilot
6. Each pressure vessel assumes an initial value for the feed water An RO experimental pilot with 50 m3/day production capacity
pressure (PFi) and permeate flow rate, and concentration in each was constructed at the Hydraulic Lab of Shahid Chamran University
element (z=1: Nei). of Ahwaz and used to validate the results of this simulation model.
7. Due to the similarity of the pressure vessels in a stage, the pres- This pilot is shown in Figs. 4-5 and includes both pretreatment and
sure in all pressure vessels is equal to the pressure applied by the desalination units. The pretreatment unit consists of a booster pump
pump to the stage. The input pressure to the first element has been (1.34 kW), a carbon filter, a sand filter, four micro-filters (5 microns),
already assumed in step 6. and a water tanker. The desalination unit consists of a water tanker
8. Solving Eqs. (22), (25), and (26) for all elements (z=1: Nei) in with a mixer, a booster pump (1.21 kW), a high-pressure pump (3.06
a pressure vessel in a set of nonlinear equations, which is solved by kW), and a BW30-400 membrane from DOW. The membrane’s
the Newton-Raphson method. characteristics are given in Table 2. The pressure and flow through

Fig. 4. Flowchart of the RO pilot plant at Hydraulic Lab of Shahid Chamran University of Ahvaz.

Korean J. Chem. Eng.(Vol. 38, No. 2)

372 Z. Hadadian et al.

oped and validated by the ROSA software and an experimental

pilot.
1. Performance of the Experimental Pilot
The performance evaluation of the laboratory scale experiments
was conducted with various operating conditions, including oper-
ating pressure (700 and 1,100 kPa) and feed concentration (2, 3, 4,
5 kg/m3). The experimental results are reported in Fig. 6. Our experi-
ments were limited to operating pressures below 1,100 kPa and re-
covery ratios below 50%.
As shown in Fig. 6(a), at constant pressure, the increase of the
feed concentration from 2 to 5 kg/m3 decreases the permeate flow
rate from 9.55 to 3.75 (m3/min)*103 and 15.29 to 10.55 (m3/min)*
103 for experimental pressure of 700 and 1,100 kPa, respectively.
In addition, the decreasing trend was almost linear with correla-
tion coefficients of 0.9888 and 0.9961 for experimental pressures
of 700 and 1,100 kPa, respectively. The highest permeate flow rate
(15.29 (m3/min)*103) was obtained at pressures of 1,100 kPa and
feed concentration of 2 kg/m3. A higher permeate flow rate means
that the membrane can produce a large amount of water per unit
area and time. At a constant feed concentration, with increasing
pressure from 700 to 1,100 kPa, the permeate flow rate also in-
creases by 5.74, 5.42, 6.64, and 6.8 (m3/min)*103) at feed concen-
tration of 2, 3, 4, and 5, respectively. This is an acceptable issue and
can be justified using Eq. (6). According to this equation at con-
stant pressure, the permeate flow rate decreases with increasing
osmotic pressure due to increasing feed concentration. At a con-
stant feed concentration, the permeate flow rate also increases when
increasing the pressure.
Fig. 6(b) indicates that the highest recovery was obtained by
48.58% at pressure of 1,100 kPa and feed concentration of 2 kg/m3.
Fig. 5. The RO pilot plant at Hydraulic Lab of Shahid Chamran
As shown in this figure, at constant pressure, the increase of the
University of Ahvaz.
feed concentration from 2 to 5 kg/m3 decreases the recovery from
41.54% to 20.67% and 48.58% to 37.06% for experimental pressure
the process are also measured by pressure gauges and flow meters of 700 and 1,100 kPa, respectively. At a constant feed concentra-
in different parts in the system. The value of pure water permea- tion, with increasing pressure from 700 to 1,100 kPa, the recovery
bility constant for the BW30-400 membrane was found 7.94e-9 also increases by 7.04%, 7.05%, 12.9%, and 16.39% at feed concen-
(m/s·kPa) through experimental data analysis and calibration. trations of 2, 3, 4, and 5, respectively, and the average percentage
2. Test Cases and Methodology of the Experiments of increased recovery was 10.85%. In other words, the higher the
To validate the simulation model, some RO experiments were concentration is the higher the percent recovery.
carried out at the pressures of 700 and 1,100 kPa for the brackish As shown in Fig. 6(c), at constant pressure, the increase in the
water as feed solution with a concentration of 2-5 kg/m3 by the RO feed concentration from 2 to 5 kg/m3 decreases the salt rejection
plant. For all tests feed water temperature was 15 oC and PH=7.3± from 96.6% to 94.58% and 97.45% to 96.52% for experimental pres-
0.1. Each RO experiment was performed by maintaining the feed sure of 700 and 1,100 kPa, respectively. The salt rejection also in-
pressures constantly and varying the concentration of the feed solu- creases by increasing the feed pressure at a constant feed concen-
tion from 2-5 kg/m3. The flow rate and concentration of permeate tration. It can also be justified using Eq. (5). The highest salt rejec-
and retentate water were measured, and then the recovery of each tion was obtained at 97.45% at pressures of 1,100 kPa and feed con-
case was calculated for the applied pressure. centration of 2 kg/m3 and the average salt rejection was 96.36%.
Since energy consumption is a key factor that affects the cost of
RESULTS AND DISCUSSION the RO system, the change of the specific energy consumption (SEC)
values versus the feed concentration for experimental pressure of
In this study, a simulator model for the RO process was devel- 700 and 1,100 kPa is shown in Fig. 6(d). The SEC is directly related

Table 2. Characteristics of Filmtec spiral wound RO membrane element [27]

Element type Size (m) Active surface area (m2) Feed spacer thickness (mil) Applied pressure (kPa) Salt rejection
BW30-400 0.203*1.02 37.2 28 1,551.32 99.5%

February, 2021
 Mathematical and experimental modeling of reverse osmosis (RO) process 373

Fig. 6. Experimental result of (a): The permeate flow rate, (b) the recovery of system, (c) the salt rejection, and (d): The specific energy con-
sumption (SEC) versus feed concentration.

to the feed flow rate and pressure (pump power) and inversely sure of 700 and 1,100 kPa, respectively. These tables show how, under
related to the permeate flow rate. As shown in this figure, at con- the same conditions of feed flow rate and recovery, by increasing
stant pressure, the increase of the feed concentration from 2 to the feed concentration, the feed pressure decreases in both the
5 kg/m3 increases the SEC. This is because according to Fig. 6(b), ROSA software and simulator models and it can be justified using
the permeate flow rate decreases when increasing the feed concen- Eq. (6). The feed pressure estimated by the proposed model is also
tration and due to increases of SEC. The average SEC value was closer to the estimation by the ROSA software than the experi-
0.777 (kW hr/m3) and all SEC values were less than 1.04 (kW hr/m3). mental. Since the feed pressure has a direct effect on the calcula-
2. Performance of the Proposed Simulation Model tion of energy consumption, this is also true for SEC. The highest
2-1. Single Element Case SEC values were 1.115 and 1.047 (kW hr/m3) in the ROSA soft-
To validate the simulation model, the RO experiments (Fig. 6) ware and the proposed model, respectively.
were also simulated by the proposed simulation model as well as To better understand the feed pressure trend, the data sets are
by the ROSA software version 9.1. Note that the temperature of the shown in Figs. 7 and 8. As shown in Fig. 7, when the experimen-
laboratory was rectified using the temperature correction factor TFC tal pressure was 700 kPa, at the same recovery and the same feed
in the simulation model equations (Eq. (3)). Tables 3 and 4 com- flow rate, by increasing the feed concentration from 2 to 5 kg/m3,
pare the results of the RO experimental pilot, the ROSA software the feed pressure in both the ROSA software and simulator mod-
estimation, and estimated by the proposed simulator model for the els has a decreasing trend and as shown from 936 to 702 kPa and
feed concentration of 2, 3, 4, and 5 kg/m3 and experimental pres- 896 to 714 kPa in the proposed simulation model and the ROSA

Korean J. Chem. Eng.(Vol. 38, No. 2)

374 Z. Hadadian et al.

Table 3. Comparison of the results for a single element case (experimental pressure=700 kPa)
Experimental ROSA Proposed model
3
Feed concentration (kg/m ) 2 2 2
3 3
Q feed (m /min)*10 22.99 22.99 22.99
Recovery (%) 41.54 41.54 41.54
P feed (kPa) 700 896 936
SEC (kW hr/m3) 0.52 0.695 0.742
TDS permeate (kg/m3) 68 20.41 67.3
Salt rejection (%) 96.6 98.98 96.64
Feed concentration (kg/m3) 3 3 3
Q feed (m3/min)*10 3 22.39 22.39 22.39
Recovery (%) 35.82 35.82 35.82
P feed (kPa) 700 888 891
SEC (kW hr/m3) 0.603 0.768 0.794
TDS permeate (kg/m3) 118 36.89 116.75
Salt rejection (%) 96.07 98.77 96.11
Feed concentration (kg/m3) 4 4 4
3 3
Q feed (m /min)*10 19.75 19.75 19.75
Recovery (%) 27.24 27.24 27.24
P feed (kPa) 700 768 758
SEC (kW hr/m3) 0.793 0.859 0.909
TDS permeate (kg/m3) 181 68 179.12
Salt rejection (%) 95.48 98.3 95.52
Feed concentration (kg/m3) 5 5 5
3 3
Q feed (m /min)*10 18.12 18.12 18.12
Recovery (%) 20.67 20.67 20.67
P feed (kPa) 700 715 702
SEC (kW hr/m3) 1.044 1.047 1.108
TDS permeate (kg/m3) 271 112.09 268.6
Salt rejection (%) 94.58 97.76 94.63

software, respectively. When increasing the feed concentration, the kPa in the proposed simulation model and the ROSA software,
results of the model, ROSA, and the experimental become closer. respectively, and as the result of Fig. 7 shows by increasing the feed
To determine the accuracy of the obtained results, the relative concentration, the results of the model, ROSA, and the experimental
error (RE) value of the results relative to each other, which is defined become closer. The average RE of the model results compared to
as %RE (A_B)=|(AB)/B|*100, has been used. The average RE of the experimental results was 19.61%. The highest and lowest errors
the model results compared to the experimental results was 17.39%. were 26.72% and 13.81% at the feed concentration of 2 and 5 kg/
The highest and lowest errors were 33.71% and 0.28% at the feed m3, respectively. When increasing the feed concentration, the error
concentration of 2 and 5 kg/m3, respectively. When increasing the decreases. The average RE of the experimental results compared to
feed concentration, the error decreases. The average RE of the the ROSA results was 19.02%. The highest and lowest errors were
experimental results compared to the ROSA results was 13.46%. 19.94% and 18.51% at the feed concentration of 2 and 5 kg/m3,
The highest and lowest errors were 21.8% and 2.09% at the feed respectively. When increasing the feed concentration, the error de-
concentration of 2 and 5 kg/m3, respectively. When increasing the creases. The average RE of the model results compared to the ROSA
feed concentration, the error decreases. The average RE of the model results was 3.89%. The highest and lowest errors were 7.25% and
results compared to the ROSA results was 1.94%. The highest and 1.4% at the feed concentration of 5 and 2 kg/m3, respectively.
lowest errors were 4.4% and 0.33% at the feed concentration of 2 As found in Figs. 7 and 8, in the single element case for the feed
and 3 kg/m3, respectively. pressure result, the average RE was 18.50%, 16.24%, and 2.9% for
Similarly, for the experimental pressure of 1,100 kPa shown in the model compared to the experimental, experimental compared
Fig. 8, at the same recovery and the same feed flow rate, by increas- to the ROSA, and model compared to ROSA, respectively. It can
ing the feed concentration from 2 to 5 kg/m3, the feed pressure in be said that the simulation model is well matched with the ROSA
both the ROSA software and simulator models has a decreasing results with over 96% accuracy. However, the calculated pressures
trend, and as shown from 1,394 to 1,252 kPa and 1,374 to 1,350 by the simulation model and ROSA are over 80% accurately close

February, 2021
 Mathematical and experimental modeling of reverse osmosis (RO) process 375

Table 4. Comparison of the results for a single element case (experimental pressure=1,100 kPa)
Experimental ROSA Proposed model
3
Feed concentration (kg/m ) 2 2 2
3 3
Q feed (m /min)*10 31.47 31.47 31.47
Recovery (%) 48.58 48.58 48.58
P feed (kPa) 1100 1374 1394
SEC (kW hr/m3) 0.699 0.892 0.886
TDS permeate (kg/m3) 51 15 50.51
Salt rejection (%) 97.45 99.25 97.47
Feed concentration (kg/m3) 3 3 3
3 3
Q feed (m /min)*10 31.36 31.36 31.36
Recovery (%) 42.87 42.87 42.87
P feed (kPa) 1100 1350 1326
SEC (kW hr/m3) 0.792 0.992 0.955
TDS permeate (kg/m3) 80 25.7 79.32
Salt rejection (%) 97.33 99.14 97.36
Feed concentration (kg/m3) 4 4 4
3 3
Q feed (m /min)*10 29.94 29.94 29.94
Recovery (%) 40.14 40.14 40.14
P feed (kPa) 1100 1359 1291
SEC (kW hr/m3) 0.846 1.072 0.992
TDS permeate (kg/m3) 125 38.44 123.62
Salt rejection (%) 96.88 99.04 96.91
Feed concentration (kg/m3) 5 5 5
3 3
Q feed (m /min)*10 28.47 28.47 28.47
Recovery (%) 37.06 37.06 37.06
P feed (kPa) 1100 1350 1252
SEC (kW hr/m3) 0.916 1.151 1.043
TDS permeate (kg/m3) 174 52.17 172.46
Salt rejection (%) 96.52 98.96 96.55

Fig. 7. Variation of feed pressure in RO system with feed concentration (experimental pressure=700 kPa) (*: Relative Error).

to the laboratory results. This could be due to laboratory conditions, to the simulation equations. Since the equations used in ROSA and
measurement errors, and uncertainties and the simplifications applied the proposed simulation model are slightly different, this little dif-

Korean J. Chem. Eng.(Vol. 38, No. 2)

376 Z. Hadadian et al.

Fig. 8. Variation of Feed pressure in RO system with feed concentration (experimental pressure=1,100 kPa).

Table 5. Characteristics of case studies with two-pressure vessel and 5 elements in each pressure vessel
Number of elements in Number of Temp QPermeate Feed concentration
Element type Recovery
each pressure vessel pressure vessel (oC) (m3/h) (kg/m3)
5 BW30-400 2 25 0.5 8 2

ference in results was expected. The ROSA software utilizes exper-

imental equations that control Filmtec membranes, but in our
simulation model the mathematical equations that control the RO
process - based on the mass, material, and energy balance equa-
tions - are used. The equation that leads to significant differences
is the one used for calculating the pressure drop in the membranes.
The BW30-400 pure water permeability constant and the salt rejec-
tion used in the simulator model were obtained by laboratory data
analysis (7.94e-9 (m/s·kPa) and 94.58-97.45%) that have a differ-
ent quantity from the value set in ROSA (7.5e-9 (m/s·kPa) [8] and
97.76-99.25%).
It is observed in Tables 3 and 4 that by increasing the feed con-
Fig. 9. Feed pressure for each element in RO system.
centration, the permeate TDS estimated in both ROSA software
and simulator model has a decreasing trend. It can also be justified
using Eq. (5). The permeate TDS estimated by the proposed model several pressure vessels and several elements in each pressure vessel
is also closer (average ER: 1%) to the laboratory than estimated by according to Table 5. The simulated results, including the required
the ROSA software. This is because the salt rejection in the ROSA feed pressures, permeate flow rate and recovery of each element are
software (97.76-99.25%) is different from our salt rejection in the found in Figs. 9-11.
laboratory (94.58-97.45%) and the proposed simulator model (94.63- The calculated pressures are indicated in Fig. 9 where in each
97.47). Thus, for permeate TDS estimation the simulation model is element by the simulation model and the ROSA software, shows
accurate enough and well matched with the experimental results. that the pressure values in the first to fifth elements were 1,003,
2-2. Multi-element Case 988, 973, 958, and 943 in the proposed model and 912, 889, 869,
Since the experimental pilot has only one element in each pres- 853 and 840 in the Rosa model and have decreasing trends in both.
sure vessel, for the multi-pressure vessel and multi-element test cases, The average difference in the results was 100 kPa between the model
the comparisons were done between the simulation model and the and ROSA, which is according to the results reported by Altaee
ROSA software. For this purpose, a case study was simulated with [2]. The highest and lowest RE was 12.26% and 9.9% in the fifth

February, 2021
 Mathematical and experimental modeling of reverse osmosis (RO) process 377

software utilizes experimental equations that control Filmtec mem-

branes, but in our simulation model the mathematical equations
that control the RO process are used. The equation that leads to
significant differences is the one used for calculating the pressure
drop in the membranes.
Based on two single element and multi-element cases result and
RE value, the proposed simulation model functions well and is reli-
able. The calculated pressures by the simulation model are over 80%
accurately close to the laboratory results and 96% accurately close
to the results of the ROSA software.

CONCLUSION
Fig. 10. Permeate flow rate for each element in RO system.
A mathematical simulation model for the RO process, consider-
ing several series elements in each pressure vessel, was developed
based on the mass, material, and energy balances considering the
concentration polarization. Unlike other commercial software, the
proposed model is open-source and since it is based on the math-
ematical equations of the RO process, it is capable of using all types
of membranes manufactured by different companies. In the ROSA
software, the percentage of the feed flow cannot be assigned to stage
2 and stage 3, and the feed flow first enters stage 1, and after treat-
ment, the entire retentate from the first stage is considered as the
feed water for the second stage. This issue happens between stage
2 and stage 3. However, in the proposed simulation model, given
the three splitter points definition in the system, a certain percent-
age of the feed flow can be assigned to each stage. A percentage of
Fig. 11. Recovery for each element in RO system. the retentate from the first stage can also be removed from the
system and a percentage of it can be added to each of the second
element and the first element, respectively; the average RE was and third stages. The former stream also happens between stage 2
11.53%. The average pressure drop per element was 18 kPa and and stage 3. Accordingly, in the ROSA software, the feed pressure
15 kPa in ROSA and the simulation model, respectively, which are of high-pressure pumps is related to the first stage, and the feed
almost close to each other. pressure of the second stage is the summation of retentate pres-
From Fig. 10, which shows the permeate flow rate in each ele- sure from the first stage and the accelerator pump pressure. In the
ment in the model and ROSA, it can be seen that the permeate proposed simulation model, according to the three splitter point
flow rate value in the first to fifth elements was 0.93, 0.86, 0.8, 0.74, coefficients, the value of the high-pressure pumps is first determined
and 0.67 in ROSA and 0.88, 0.84, 0.8, 0.75, and 0.69 in the model in order to provide the pressure required for the first stage and ac-
where the results are very close [2]. The average RE was 2.4%, the cording to the energy equation. Then, for the second and third stage,
highest RE was 5.3% in the first element, and the lowest RE was the amount of feed pressure is calculated by the energy equation.
0% in the third element. On the other hand, the retentate pressure considered as the feed for
In Fig. 11, which shows the recovery in each element in the another stage is calculated concerning the energy equation, and
model and ROSA, the recovery value in the first to fifth elements the flow pressure prior to entering each stage is considered to be
was 0.12, 0.12, 0.13, 0.14, and 0.14 in ROSA, and 0.11, 0.12, 0.13, the minimum of the inflows to that stage. In case there is any need
0.14, and 0.15 in the model where the results are very close [2]. for a higher pressure, accelerator pumps are utilized.
The average RE was 3.09%. An RO experimental pilot was constructed at the Hydraulic Lab
Based on these obtained average RE values, it can be said that of Shahid Chamran University of Ahwaz and used to validate the
the pressure, the permeate flow rate, and the recovery calculated by results of this simulation model. The performance of the laboratory
the simulator model are (88.4%, 97.6%, 96.9%, respectively) accu- pilot was evaluated and the results were discussed in detail.
rately close to the calculated values by the ROSA software. Hence, The single element case was simulated using the proposed sim-
the proposed algorithm functions well in the multi-elements RO ulator model, and the results (including feed pressure, SEC, TDS
systems. The calculated pressures by the model are slightly higher permeate, and salt rejection) were compared with the ROSA soft-
than the calculated pressures by ROSA. This could be due to labo- ware simulation and the experimental results. The multi-element
ratory conditions, measurement errors, and uncertainties and the case was simulated using the proposed simulator model, and the
simplifications applied to the simulation equations. Since the equa- results (including feed pressure, permeate flow rate, and recovery
tions used in ROSA and the proposed simulation model are slightly in each element) were compared with ROSA.
different, this little difference in results was expected. The ROSA Based on two single element and multi-element cases result and

Korean J. Chem. Eng.(Vol. 38, No. 2)

378 Z. Hadadian et al.

RE value, it can be said that the comparison among the simula- St : spacer thickness [m]
tion model, the experiments, and the ROSA results indicates the T : temperature [oC]
reliability of the proposed simulation model. The calculated results T0 : reference temperature [25 oC]
by the simulation model are over 80% accurately close to the labo- TCF : temperature's correction factor
ratory results and 96% accurately close to the ROSA software. TF : feed water temperature [oC]
VSP : volume of the spacer [m3]
ACKNOWLEDGEMENTS VT : total volume [m3]
z : element counter in the Pressure vessel
The authors would like to thank Ghadir Khuzestan Water Com-  : feed water viscosity [Pa·s]
pany and Khuzestan Water and Power Authority for their finan-  : fraction of the stream branching to the left from a split point
cial and technical support.  : fraction of the stream branching to the right from a split point
 : fraction of the stream branching straight forward from a
NOMENCLATURE split point
Pf : pressure drop due to friction [kPa]
A : membrane surface area [m2] Pin : pressure drop at inlet [kPa]
Aeff : effective surface area of membrane [m2]  : osmotic pressure [kPa]
Am : pure water permeability constant [m/s·kPa] p : osmotic pressure on the permeate side [kPa]
C : solution concentration [kg/m3] w : osmotic pressure on the feed side of membrane wall [kPa]
CF : feed concentration [kg/m3]  : density [kg/m3]
CP : permeate concentration [kg/m3]  : Porosity
CR : retentate concentration [kg/m3]
Cw : concentration on the feed side of membrane wall [kg/m3] Subscripts
d : membrane channel diameter [m] f : feed
Ds : solute diffusivity [m2/s] F : feed split ratio
dsp : spacer diameter [m] p : permeate
i : stage counter r : retentate
j : pressure vessel counter in stage R1 : retentate split ratio for first module
jw : water flux [m3/m2·s] R2 : retentate split ratio for second module
k : mass transfer coefficient [m/s]
lch : length of the membrane channel [m] REFERENCES
M : molar concentration [mol/m3]
n : van’t Hoff factor 1. H. Kotb, E. Amer and K. Ibrahim, Desalination, 357, 246 (2015).
Nei : number of the membrane element in pressure vessel of the 2. A. Altaee, Desalination, 291, 101 (2012).
ith stage 3. A. Villafafila and I. Mujtaba, Desalination, 155, 1 (2003).
Ni : number of membrane element in the ith stage 4. M. Barello, D. Manca, R. Patel and I. M. Mujtaba, Comput. Chem.
NPVi : number of pressure vessel in the ith stage Eng., 83, 139 (2015).
Peff : residual transmembrane pressure [kPa] 5. M. G. Marcovecchio, P. A. Aguirre and N. J. Scenna, Desalination,
Pf : feed pressure [kPa] 184, 259 (2005).
Pp : permeate pressure [kPa] 6. V. Geraldes, N. E. Pereira and M. Norberta de Pinho, Ind. Eng.
QF : feed flow rate [m3/s] Chem. Res., 44, 1897 (2005).
QFPV : feed flow rate in the pressure vessel [m3/s] 7. C. Guria, P. K. Bhattacharya and S. K. Gupta, Comput. Chem. Eng.,
QF1 : feed flow rate to the first stage [m3/s] 29, 1977 (2005).
QF2 : feed flow rate to the second stage [m3/s] 8. Y.-Y. Lu, Y.-D. Hu, X.-L. Zhang, L.-Y. Wu and Q.-Z. Liu, J. Membr.
QF3 : feed flow rate to the third stage [m3/s] Sci., 287, 219 (2007).
QP : permeate flow rate [m3/s] 9. Y.-J. Choi, T.-M. Hwang, H. Oh, S.-H. Nam, S. Lee, J.-c. Jeon, S. J.
QPPV : permeate flow rate in the pressure vessel [m3/s] Han and Y. Chung, Desalination and Water Treatment, 33, 273
QR : retentate flow rate [m3/s] (2011).
QR1 : retentate flow rate from the first stage [m3/s] 10. Y. Du, L. Xie, Y. Wang, Y. Xu and S. Wang, Ind. Eng. Chem. Res.,
QR2 : retentate flow rate from the second stage [m3/s] 51, 11764 (2012).
R : universal gases constant [8.314 m3·Pa/mol·oK] 11. Y. Du, L. Xie, J. Liu, Y. Wang, Y. Xu and S. Wang, Desalination, 333,
Re : Reynold’s number 66 (2014).
RE : relative error 12. Y. Du, L. Xie, Y. Liu, S. Zhang and Y. Xu, Desalination, 365, 365
rec1 : recovery in the first stage (2015).
rec2 : recovery in the second stage 13. E. R. Saavedra, A. G. Gotor, S. O. Pérez Báez, A. R. Martín, A. Ruiz-
rec3 : recovery in the third stage García and A. C. González, Desalination and Water Treatment, 51,
Sc : schmidt number 4790 (2013).

February, 2021
 Mathematical and experimental modeling of reverse osmosis (RO) process 379

14. E. Ruiz-Saavedra, A. Ruiz-García and A. Ramos-Martín, Desalina- 22. M. Li, Chem. Eng. Res. Des., 148, 440 (2019).
tion and Water Treatment, 55, 2562 (2015). 23. D. Gaublomme, L. Strubbe, M. Vanoppen, E. Torfs, S. Mortier, E.
15. J.-S. Choi and J.-T. Kim, J. Ind. Eng. Chem., 21, 261 (2015). Cornelissen, B. De Gusseme, A. Verliefde and I. Nopens, Desali-
16. H. Kotb, E. Amer and K. Ibrahim, Energy, 103, 127 (2016). nation, 490, 114509 (2020).
17. V. Haluch, E. F. Zanoelo and C. J. Hermes, Chem. Eng. Res. Des., 24. J. Siegel, C. Wangmo, J. Cuhorka, A. Otoupalíková and M. Bittner,
122, 243 (2017). Environ. Technol. Innovation, 17, 100584 (2020).
18. K. P. Chee, K. P. Wai, C. H. Koo and W. C. Chong, EDP Sciences, 25. M. Ligaray, N. Kim, S. Park, J.-S. Park, J. Park, Y. Kim and K. H.
E3S Web of Conferences, 65, 05022 (2018). Cho, Chem. Eng. J., 395, 125082 (2020).
19. M. Al-Obaidi, A. Alsarayreh, A. Al-Hroub, S. Alsadaie and I. M. 26. T. M. Mansour, T. M. Ismail, K. Ramzy and M. Abd El-Salam,
Mujtaba, Desalination, 443, 272 (2018). Alexandria Eng. J., 59, 3741 (2020).
20. C. Chen and H. Qin, Processes, 7, 271 (2019). 27. DOW, FILMTEC Membranes, product information catalog, http://
21. O. P. Maure and S. Mungkasi, AIP Publishing LLC, AIP Conference www.lenntech.com/feedback/feedback_uk.htm?ref_title=Filmtec/
Proceedings, 2202, 020043 (2019). Filmtec-Reverse-Osmosis-Product-Catalog-L.pdf (2006).

Korean J. Chem. Eng.(Vol. 38, No. 2)