Mathematical and Optimization Modelling in Desalination: State-Of-The-Art and Future Direction
Mathematical and Optimization Modelling in Desalination: State-Of-The-Art and Future Direction
Mathematical and Optimization Modelling in Desalination: State-Of-The-Art and Future Direction
Emirates
2
Division of Engineering, New York University Abu Dhabi, Saadiyat Island, 129188, Abu Dhabi,
The growing water demand across the world necessitates the need for new and improved processes
as well as for a better understanding of existing processes. This level of understanding includes
processes. Through mathematically modelling real life systems, we gain a deeper understanding
of processes while being able to predict performance more effectively. Advances in computational
capacity and the ease of assessing systems allow researchers to study the feasibility of various
minimizing energy requirements and, as such, have been an active area of research in desalination.
In this review, the most recent developments in mathematical and optimization modelling in
desalination are discussed with respect to transport phenomena, energy consumption, fouling
predictions, and the integration of multiple scaling evolution on heat transfer surfaces has been
have been analyzed from an energy consumption perspective. Transport models for membrane-
based desalination processes, including relatively less understood processes such as nanofiltration
and forward osmosis are presented, with recent modifications to allow for different solutes and
solutions. Mathematical modelling of hybrid systems integrated with RO has also been reviewed.
A survey of the literature shows that mathematical and optimization modelling of desalination
processes is an exciting area for researchers in which future scholarship includes coupling of
renewable energy systems with desalination technologies, as well as more advanced descriptions
1. Introduction ............................................................................................................................. 6
4. Conclusion ............................................................................................................................ 70
5. References ................................................................................................................................. 72
6. Abbreviations ........................................................................................................................ 81
7. Symbols................................................................................................................................. 82
1. Introduction
Between 2015 and 2050, the world population is projected to increase by 35% causing an even
more rapid increase in the demand for water [1]. Despite being an abundant resource, fresh water
needed for human consumption makes up a tiny fraction of the water on earth [2]. In addition, the
amount of fresh water stays constant while our water demands are on the rise [3]. Millions of
people die annually due to the lack of adequate water supply and proper sanitation. Many solutions
have been proposed to help meet growing water demands, including better water resource
management, increased water reuse and desalination. Many of the world’s communities affected
by drought have access to brackish ground water and seawater that can be converted to fresh water
through desalination to help meet their water demands [2]. Consequently, the global demand for
desalination is increasing [4], as is evident by the rapid desalination market growth in recent years.
Globally, desalination plants produce around 95 million m3/day. The Middle East and North Africa
(MENA) regions are responsible for 48% of the global installed desalination capacity [5, 6].
technologies. Although membrane technology, specifically reverse osmosis (RO), makes up the
majority of installed desalination capacity worldwide, thermal desalination remains the dominant
technology in the Middle East. The two major thermal desalination technologies employed are
multi-evaporation distillation (MED) and multistage flash (MSF), both of which are in an
advanced growth phase, likely to reach saturation before 2050, as identified by Mayor [7]. On the
other hand, most of the newly installed desalination capacity around the world operates on RO
membranes, which not only are more energy efficient than thermal desalination processes, but also
Although surpassed by reverse osmosis in terms of global installed capacity, thermal desalination
plants are still dominant in the Middle East, where they are often integrated with power plants and
are known to have long lifetimes of up to 30 years [8, 9]. Thermal desalination technologies such
as MED and MSF rely on the principle of evaporation, which is the creation of a hot surface with
heating steam, which then condenses on one side of the surface, allowing vapor to form on the
other side [10]. Evaporator surfaces are of many types: submerged tube, falling films and plates,
etc. [11]. Applications and limitations of each of these are discussed thoroughly in [12].
Separation in MSF occurs when some of the feed is evaporated in consequent stages by flashing.
Figure 1 shows a schematic of a single flashing stage in MSF. The hot feed water is met with a
lower pressure than its vapor pressure in each subsequent stage, causing some of the feed to flash.
The vapor formed in each stage passes through a demister and condenses on the external surface
of the tube bundle [12]. The simplest design, known as once-through MSF (OT-MSF), involves
returning the brine leaving the last stage back to the sea as brine blow down. The brine leaving the
last stage of the MSF can be returned to the sea as brine blow down, a configuration known as
once through MSF (MSF-OT). Another configuration which is known as brine mixing MSF (BM-
MSF) involves mixing a portion of the brine from the last stage with the incoming feed.
In each subsequent stage, the temperature is reduced by flashing, boiling point elevation and non-
equilibrium losses. The highest temperature at the inlet of the first stage is known as top brine
temperature (TBT), while the difference between the TBT and the brine temperature in the final
Modelling an MSF system requires formulation of material balance, energy balance and
momentum balances, such that the model predicts output stream variables for each stage, given
Figure 2: Depiction of a process model as a set of relationships between the input and output streams for an MSF stage [14]
The independent variables for each stream include the mass flow rate, temperature, salt
concentration and pressure. Material balance and energy balance for thermal systems have been
used to understand the transport phenomena in thermal desalination over a period of several
Various membrane-based separation processes exist, each one distinct in the size of particles or
solutes it can retain, as shown in Figure 3. Only nanofiltration (NF) and RO are used for removing
dissolved ions from aqueous streams, while others, such as ultrafiltration and microfiltration, are
often used as pretreatment to RO as they remove larger particles. The most widely employed
membrane-based desalination technology is RO, as mentioned above. The membrane material and
structure play a critical role in transport properties and hence in membrane performance [17].
Membrane technology has the advantage of being modular, and attention to new configurations of
the membrane module as well as the flow streams have enabled reduction in energy consumption.
More than 60% of the world’s installed desalination capacity operate with reverse osmosis (RO).
passage of water molecules through the membrane. In reverse osmosis, the phenomenon of natural
osmosis, in which the solvent will flow from the region of low solute concentration to high solute
osmotic pressures between the feed side and permeate side, forcing the solvent to move from the
region of high salt concentration to low salt concentration (Figure 4) [18].
In osmosis, water spontaneously passes from the low-salt concentration side to the high-salt
concentration side until an osmotic equilibrium is reached between both sides. However if a
pressure greater than the osmotic pressure is applied as is the case in reverse osmosis, the flow of
water molecules is reversed and water will pass through the membrane from the high-salt
concentration side to low-salt concentration side [20] Thus the effective pressure that drives water
through the membrane is the difference between the applied pressure and the osmotic pressure.
For simple systems, the osmotic pressure is calculated using the van’t Hoff’s equation described
below:
𝑅𝑅𝑅𝑅
∆𝜋𝜋 = 𝑣𝑣𝑖𝑖 𝑐𝑐𝑖𝑖
𝑀𝑀𝑀𝑀
where 𝑣𝑣𝑖𝑖 is the number of ions in the dissociated salt, 𝑐𝑐𝑖𝑖 is the concentration of salt in g/L, MW is
the molar mass of the ion. As the expression for osmotic pressure contains molecular weight in the
denominator, it can be seen that the osmotic pressure only comes into play for retention of very
small solutes as is the case in RO and NF, but not for UF or MF.
Although many models have been developed for transport across RO membranes, the solution
diffusion model for non-porous membranes remains the most widely accepted. It assumes that
both solvent and solute dissolve in and diffuse across the membrane down a concentration gradient
and this diffusion depends on the chemical potential of each, which is a function of the
concentration and pressure gradients across the membrane (Figure 5) [21]. In other words,
Brownian diffusion, flush and jump diffusion allow water to permeate through the membrane. The
interactions of water and ions with the membrane depend strongly on the membrane structure [22,
23].
Figure 5: Assumptions of solution-diffusion model showing chemical potential (μ), pressure (P), and activity gradients across
membrane (a).
where αi is the activity of species i (solute or solvent), Vi is the molar volume, and Δp is the
pressure differential across the membrane [21]. RO is used to remove dissolved substances,
including single charged ions from the aqueous feed streams [24]. Transport mechanisms for RO
Reverse osmosis was first commercialized by Loeb and Sourirajan [25], who developed the first
cellulose acetate membranes for RO. Ever since, RO has gained much commercial success and is
currently the dominant desalination technology. New membrane materials, improved pretreatment
methods and novel process design have enabled the technology to operate close to theoretical
energetic limit. Innovations in system configuration such as the use of multiple stages and/or
passes have been incorporated in largescale RO plants [26], mainly to overcome drawbacks of the
single-stage RO process in which the large applied pressure results in avoidable energy dissipation
and high initial permeate flux. Current developments focus on configuration improvements as well
as hybridization of RO with other technologies with the aim of further reducing energy costs.
1.1.2.2 Nanofiltration
On the separation spectrum, nanofiltration falls between ultrafiltration and reverse osmosis, and is
thus a unique filtration process in which large salts and low molecular weight cut-off (MWCO)
solutes can be separated from the feed stream. Also known as ‘loose RO’ membranes, NF
membranes can be used to reduce salinity and are often used as pretreatment to other desalination
processes such as RO, MSF, and MED. Ionic transport in NF membranes is still not fully
topic of great interest among researchers. What is known is that ionic transport through NF
membranes depends on charge, steric and dielectric effects [27, 28]. The first of these is a
consequence of the charge polarities between the membrane and solutes, while the second effect
is due to the size of ions relative to membrane pores, while the third results from differences in
dielectric constant between bulk and membrane pores [27]. Transport of solutes through NF is
most widely modelled using the extended Nernst Plank (ENP) differential equations, which
evolution of these models to better predict NF performance to account for different solutions as
Rejection of divalent ions by NF is typically in the range of 75-99%, whereas monovalent ions are
only rejected at 30-50% by NF membranes [29]. NF is used in several industries for various
Table 1: Commercial applications of NF membranes for aqueous and nonaqueous processes [30]
micropollutants elimination,
pretreatment to RO
Oil and gas Solvent recovery from lube oil and Aqueous and
Forward osmosis, as the name suggests, refers to the movement of molecules across a
semipermeable membrane due to difference in osmotic pressure. This osmotic pressure difference
is brought about using a concentrated draw solution on the permeate side, that ‘draws’ the water
from the feed [31]. Due to this transport of water molecules across the membrane, the feed solution
becomes more concentrated while the draw solution is diluted. As FO is an osmotically driven
process that does not need external hydraulic pressure (as is the case for NF and RO), the energy
requirements can be significantly lower than in RO. Transport of water in FO results from a water
chemical potential gradient driven by a difference in the solute concentration on either side of the
membrane [32]. This difference in solute concentration causes an osmotic pressure differential
across the membrane, which results in a more concentrated feed stream and a more dilute permeate
Membrane materials and choice of draw solution have been an active area of research in FO.
Desalination using FO is a two-stage process, as shown in Figure 6: (1) FO which results in water
permeating from feed to draw solution, and (2) regeneration of the dilute draw solution to recover
pure water.
Figure 6: Schematic of the two stages of fresh water production with FO [33]
Membrane fouling is the accumulation of undesired substances either on the surface of the
membrane, or inside its pores. As it reduces the effective surface area for desalination, fouling
leads to undesirable consequences such as decrease in membrane flux (or increase in hydraulic
pressure to maintain the same flux) and reduction in salt rejection [23]. Fouling mechanisms in
high-pressure membrane processes differ from those in MF and UF in that surface fouling is the
predominant fouling mechanism on the more compact and non-porous RO and NF membranes
[34]. As fouling depends strongly on the physical and chemical interactions between foulants and
membrane surface, the extent of fouling, or degree of attachment, is a function of feed composition,
membrane properties, hydrodynamic conditions, cleaning strength and frequency. Fouling can be
classified as colloidal, inorganic, organic and biofouling [35]. Fouling can also be aggravated by
the phenomenon of concentration polarization (CP). CP refers to the increase in salt concentration
at the membrane surface as compared to the bulk salt concentration on the feed side of the
membrane. It results from rejection of salt ions at the membrane surface as water passes through.
surface, increased salt passage through the membrane, increased potential of salt precipitation i.e.
scaling at the membrane surface and increased fouling [36]. External CP described above has been
phenomenon that occurs only in osmotically driven processes such as FO. ICP is a reduction in
osmotic pressure gradient across the active layer resulting from a sharp concentration gradient
formed within the support layer of the membrane [37, 38]. It results in a sharp concentration
equations and using some approaches to solve the mathematical equations as a guide to
deconstructing and solving the original problem [39]. One of the most commonly applied types of
choosing input values from within a set that stratifies some constraints and computing the value of
the function. Although real world problems cannot always be explained entirely by mathematical
equations, mathematical solutions alone are also not practical as they take into account several
simplifying assumptions. Optimization is the process of finding the best possible solution to a
given problem by examining several alternatives [40]. In recent years, multiscale modelling and
optimization has gained significant interest with potential for better prediction and understanding
of systems in material science, fluid mechanics, biology, chemistry, and physics [41].
complexity of systems, mathematical modelling and optimization are now considered essential
components of the design process. Today, we have access to powerful software tools that enable
real world system can be understood more sufficiently and predictions can be made through
enables manipulation of design parameters to meet certain objectives and/or helps predict system
key components under focus [43]. For any multifaceted process, accurate models with realistic
assumptions that are not too complicated to solve are a challenge. The presence of uncertainties in
real-life systems as well as the high costs associated with building pilot plants render the modelling
approach all the more valuable, but also more complex with several constraints that need to be
satisfied and development of models that match the real-world system as closely as possible [44].
Modelling enables better prediction and control of system performance and helps our
understanding of everyday processes. However, developing suitable models requires a certain level
of understanding of the mechanism(s) being studied. It can be argued that the modelling approach
strongly complements experimental research and forms a critical component of research in any
field.
150
100
50
0
98
99
00
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
19
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
Year
Figure 7: Publications with topic keywords: optimization modelling , desalination from 1998 to 2018 (Web of Science)
Interest in optimization modelling techniques in desalination has increased dramatically in the last
decade, as shown in Figure 7. Van der Bruggen cites three critical benefits of process modelling
specifically in pressure-driven membrane separation [45]. First, models help predict expected
provides a deeper understanding of the mechanisms responsible for permeation and separation,
which is of particular value in newer, less understood processes. Finally, modelling allows for
process monitoring and a study of the factors that affect performance characteristics for each
process, helping us find configurations. The type of material and structure will determine the kind
of mass transport through the membrane (solution, diffusion, Knudsen diffusion, convection, etc.)
and, therefore, the mathematical model to be applied to describe the mass transfer (solution-
diffusion model, pore-flow model, etc.). Modelling and control of RO desalination systems was
previously reviewed by Sobana and Panda in 2011 [46]. Blanco-Marigorta reviewed differing
forward osmosis (FO), modelling techniques in FO have not been discussed on their own [48].
In 2007, Weijuan et al. reviewed modelling techniques in nanofiltration membranes [49]. Later,
Oatley-Radcliffe et al. highlighted the need for a reevaluation of NF modelling, especially for
complex feeds in their review of existing modelling strategies [50]. Recently, Yaroshchuk et al.
reviewed existing models for NF of electrolyte solutions in which they derive equations for ion
transfer from linear irreversible thermodynamics and identify membrane properties that control
membrane performance for NF of multi-ion solutions [51]. They also included the development of
an advanced engineering model for NF of multi-ion solutions which relies on a solution diffusion
Radcliffe et al. [52] linked a decline in the topic of nanofiltration modelling between 2009 and
2016 to limited practical understanding of the process, owing to the lack of drastic advances in
separation mechanisms stems from a lack of in-depth knowledge of the physical and electrical
was indeed stagnant or moving towards decline during this period, the last three years have shown
a resurgence of interest in this area with more than a two-fold increase in the number of
20
Number of publications
16
12
0
98
99
00
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
19
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
Year
Figure 8: Number of publications with ‘nanofiltration’ and ‘modelling’ in title from 1998 to 2018 (Web of Science)
Apart from traditional mathematical modelling of processes, artificial intelligence (AI) techniques
such as artificial neural network (ANN) are gaining attention in desalination, although the advent
of AI in desalination has recently been reviewed by two different groups in the last year [53, 54].
As such, these have been left out of this review. In this review, developments of the last five years
in modelling and optimization in desalination have been critically reviewed with respect to
such as MSF and RO from an energy consumption perspective. For relatively less understood
processes such as NF and FO, simple transport models have been described, with a review of recent
modifications to allow for different types and concentrations of solutes. Gaps in literature are
identified, paving the way for future areas of research in modelling approaches to desalination.
2. Optimization modelling in desalination
Thermal desalination processes such as MSF and MED are mature technologies, whose lower
competitiveness and energy efficiency in comparison to RO have limited the market to the Middle
East, where the cost of energy is relatively low and the feed seawater is of higher, often aggressive,
salinity [7, 55]. Fouling, which is the unwanted accumulation of solid materials on the heat transfer
surface, increases the thermal resistance and leads to performance deterioration [56]. Scale
formation or precipitation of certain salts on the heat transfer surfaces impedes the rate of heat
transfer and reduces the efficiency of the heat transfer process, thus leading to increased specific
energy consumption as well as the need for frequent cleaning. Scale formation can also lead to
clogging and corrosion of heat exchangers and evaporators. Many recent modelling studies in
thermal desalination focus on better understanding the formation of scale on heat transfer surfaces
as well as modelling scale control strategies, both of which are highlighted in this review.
Various models have been developed to correlate the formation of scale on heat transfer surfaces
over the last sixty years, and these models are not limited to desalination units. The earliest of
these, proposed by Kern and Seaten, was based on a diffusion model in which the net rate of
deposition is the difference between the rate of deposition and rate of removal at any given time,
correlating an increase in fluid velocity with a reduction in fouling layer thickness [57, 58]. Over
the years, further modifications were incorporated into ionic diffusion models as well as kinetic
models to predict the rate of scale deposition as a function of operating parameters and the use of
antiscalants, including the use of computational fluid dynamics (CFD) [59-65]. For a more
comprehensive insight into the description of fouling of heat exchanger surfaces, including fouling
mechanisms, related mathematical modelling, and control of fouling, the reader is guided to other
literature [66, 67]. This section will focus exclusively on developments between 2014 and 2019 in
the modelling and optimization of fouling in MSF and MED systems. The presence of a series of
connected heat exchangers in these systems means that the fouling behavior is often complex.
Only recently have models been developed to account for the changing temperature and feed
salinity from one stage to another, as well as variation of the fouling factors over time [68].
scales such as calcium carbonate (CaCO3) and magnesium hydroxide (Mg(OH)2) [69-72]. The rate
rate of CO2 release, concentration of Ca2+ and Mg2+ ions, and total dissolved solids (TDS) [69,
73]. The rate of fouling is defined as the average deposit surface loading per unit surface area in a
unit of time. Often, the thickness of the deposited layer and porosity are used to describe the extent
of fouling on a heat exchange surface [67]. Effective scale control requires accurate calculations
of the amount of scale formation [74]. Alsadaie and Mujtaba developed a dynamic model to predict
scaling with CaCO3 and Mg(OH)2 in the MSF condensing tubes and with increasing cooling water
temperatures in subsequent stages [56]. Their model was based on a dynamic fouling model for
heat transfer surfaces, coupled with an MSF dynamic model to predict the scaling behavior in
condensing tubes in the heat recovery section of an MSF plant. First, the deposition rate of either
CaCO3 and Mg(OH)2 was found using the diffusion transport rate and surface reaction rate of ions
(Figure 7). Total deposition rate is calculated by considering transport of species towards the
surface followed by accumulation at the surface, i.e. a combination of diffusion and reaction
mechanisms [56]. Transport towards the surface is caused by a concentration gradient between the
bulk phase and solid-liquid interface, while accumulation is a result of the concentration difference
between the solid-liquid interface and saturation concentration.
Figure 9: Radial concentration and temperature profiles at heat transfer surfaces [56, 60]; Cb, Ci and Cs are the concentration of
ions in the bulk phase, at the solid-liquid surface and the saturation concentration respectively.
They found that lower temperatures favored the deposition of CaCO3, but the slightly increased
surface temperature in the middle stages reduced the ratio of Ca to Mg, which in turn favored the
formation of Mg(OH)2 in these stages. They found that the resulting varied fouling factor from
their dynamic model could lead to a higher performance ratio (Figure 10) and cost reductions, in
Figure 10: MSF performance ratio for fixed and varied fouling factor [56]
pretreatment of the feed to remove scale forming agents, and/or 2) the use of antiscalants to inhibit
the formation of scale [75].
Most pretreatment processes do not involve removal of dissolved solids such as multivalent ions
that can later precipitate and cause scaling on heat exchangers, which reduces heat transfer
efficiency in both MSF and MED systems. In addition to scale formation, heat transfer surfaces
are also affected by the release of non-condensable gases such as carbon dioxide, oxygen, and
nitrogen. The presence of CO2 in the feed stream alters the concentration of HCO3-, CO32-, CO2,
H+ and OH- ions in the brine stream, which in turn affects alkaline scale formation [74].
Nanofiltration has gained importance in the last decade as pretreatment to prevent scale deposition
in MSF desalination plants. Nanofiltration can be used to lower the concentration of scale-forming
elements in seawater, such as Ca2+, Mg2+, SO42- and HCO3-. After pretreatment with NF, the feed
has a significantly lower salinity as NF membranes are capable of not only removing multivalent
ions with high efficiency, but can lower the concentration of monovalent ions such as Na+, Cl- and
K+. The lower feed salinity resulting from NF pretreatment allows the TBT to increase to 130 °C
[76]. The use of NF as pretreatment for thermal desalination systems is not new; both mathematical
models [55, 77] and experimental studies [78, 79] on NF/MSF pilot plants shown have shown the
efficiency of NF as pretreatment for MSF over the last two decades, demonstrating their ability to
reduce scaling and in turn operational costs of the thermal desalination system. Typical models are
based on mass transfer and chemical reaction of solutes in the brine, which can be used to calculate
scale formation in MSF evaporator tubes. Al-Rawajfeh investigated NF as CO2 deaerator for
thermal desalination systems [74]. Using the mathematical model of mass transfer, they found the
molar release rate of CO2. They simulated the desorption-deposition of CO2-CaCO3 by mass
balance of a differential volume element of liquid at the gas-liquid phase interface. They found
that NF pretreatment significantly enhances deaeration (Figure 11) and this decrease in CO2 release
rates is correlated with lower heat transfer resistances in MSF plants. In addition to reducing
scaling potential, the negative effects of CO2 and other non-condensable gases on heat transfer and
Figure 11: Effect of NF pretreatment on CO2 release rates in brine recirculation MSF (BR-MSF) [74]
Rawajfeh et al. found that incorporating NF pretreatment with traditional salt precipitation
Hanshik et al. used theoretical calculations to indicate that increased TBT led to increased water
production which in turn lowers the specific energy consumption (SEC) for a once-through
multistage flash (OT-MSF) plant [80]. Figure 12 shows the effect of TBT on the production of
desalinated water and on the cooling seawater quality, temperature. However, they did not
incorporate any experimental correlation or adjustment for fouling, seawater quality and/or
chemical dosing.
Figure 12: Effect of TBT on freshwater production (solid black lines) and cooling seawater temperature (dotted blue lines) [80]
Roy et al. similarly developed a mathematical model to investigate the effect of increasing TBT
on the performance of an OT-MSF system, including the required specific area [8]. Unlike
Hanshik’s work in which a fixed number of stages was studied, they considered a fixed inter-stage
temperature drop ΔT and varied the number of stages to increase the TBT. The effect of increasing
TBT of up to 160 °C on PR and required specific area is shown in Figure 13. They found that the
performance ratio (PR) increases with TBT while the required specific area decreases but then
slightly increases beyond a certain TBT. As mentioned in the previous section, a high TBT of 160+
Mabrouk et al. demonstrated a 25% reduction in heat transfer surface area by using long tube
evaporator bundles as compared to traditional brine recycle and cross tube bundle configurations
[55]. In a long tube configuration, the tubes are parallel to the direction of brine flow, whereas in
a cross tube configuration, the tubes are perpendicular to the direction of brine flow in the stages
(Figure 14).
Figure 14: MSF a) cross tube configuration and b) long tube configuration [55]
Their study focused on further scaling up MSF to systems greater than 20 MIGD. They found that
although the operating cost is similar, an MSF system with long tube bundles allows for 15% lower
capital cost than its cross tube counterpart, for a large scale project of 100 MIGD.
Ben Ali and Kairouani optimized operating parameters of a brine recirculation MSF plant [81],
considering changes in brine heater fouling factor and the seasonal variation in seawater
temperature using genetic alogrithms used to solve multi-objective optimization problems. The
operating parameters that were optimized were heating steam temperature (Ths), recycled brine
flow rate (MR), cooling seawater flowrate (MCW) and make-up seawater flow rate (Mf). The plant
data used includes 16 flashing stages and a nominal production capacity of 26,700 m3/day. The
objectives were to maximize fresh water production capacity, minimize thermal energy
consumption by reducing heating steam flow rate and minimize electrical energy consumption by
minimizing flow rates from pumps. They obtained a set of Pareto optimal solutions in which
combinations of optimal operating parameters were defined. To solve the optimization problem,
they used a steady-state process model of the plant, which consisted of mass and energy balances
They found that, for a constant Ths, Mcw, MR and Mf, fresh water production decreases as seawater
temperature rises and although the fouling factor decreases continuously, its effect on water
production capacity is less pronounced. This can be observed in Figure 15, where between April
and November, the fouling factor increased by 90.7% but fresh water production declined only by
2.2%.
Figure 15: Variation of plant production capacity as a function of seawater temperature and brine heating fouling factor
(Ths = 93 °C, Mcw = MR = Mf = 1500 kg/s) [81]
Improvements in configuration have led to significant cost reductions in MSF. The simplest
design, known as once-through MSF (OT-MSF), involves returning the brine leaving the last stage
back to the sea as brine blow down. The brine leaving the last stage of the MSF can be returned to
the sea as brine blow down, a configuration known as once through MSF (MSF-OT). Another
configuration which is known as brine mixing MSF (BM-MSF) involves mixing a portion of the
brine from the last stage with the incoming feed. Dahdah and Mitsos sought to optimize brine and
feed routing by developing a superstructure representing thermal desalination structures [82]. This
Bandi et al. [83] adopted a differential evolution (DE) algorithm to address the global optimal
design of MSF systems. They use non-linear programming (NLP) based process models together
M) and MSF-brine recycle (MSF-BR) configurations, and compared obtained solutions with those
obtained with MATLAB. A non-deterministic algorithm differs from traditional algorithms in that
it can arrive at outcomes using various routes, or that, even for the same input, can exhibit different
behaviors for different runs [84]. Bandi et al. use freshwater production cost as the objective
function for minimization, constraints are imposed by mass, energy and enthalpy balances. They
found that the obtained global solution from DE is >2% better than from other deterministic
optimization algorithms such as SQP, MS-SQP and DE-SQP. In the latter, the optimal variable
value set and objective function depend on the initial gas value, whereas DE provides better
initialization strategies and is more suitable for complex problems in terms of decision variables.
Figure 16 shows how the different optimization methods differ in terms of optimal cost obtained
Figure 16: Comparison of different optimization methods used to minimize fresh water production costs in MSF-OT, MSF-M
[83]
Selected design parameters and their effect on MSF performance are shown in Table 2.
Table 2: Selected operation and design parameters and their effect on MSF performance
temperature decreases
The solution diffusion model, developed in the 1960s, remains the most commonly used model to
describe transport through an RO membrane is the solution diffusion model (Figure 17). In this
model, transfer of the solvent (water) and the solute (salt) through a non-porous membrane occurs
in three steps: absorption to the membrane, diffusion through the membrane and desorption from
the membrane [18]. The driving force is the chemical potential gradient such that when the applied
hydrostatic pressure is greater than the difference in osmotic pressure between both sides of the
membrane, water is transported through against its natural flow due to difference in chemical
potential.
Figure 17: Solution-diffusion model for RO membrane [18]
In the solution diffusion model, salt and water flux are given by:
Although there has traditionally been little evidence on the presence of pores in RO membranes
due to measurement limitations, gradually strengthening the support for the solution-diffusion
model over the decades, some pore-based models also emerged [85-87], which are now being
supported by experimental data as measurement tools for sub-nanometer pores become advanced,
Early on, Starov’s group developed a model to investigate RO for multicomponent electrolyte
solutions [89-91], in which they applied extended Nernst-Plank equations to include diffusion,
confection and electromigration of ions. In their model, the boundary conditions included both (i)
distribution coefficients for individual ions, determined by specific interaction of ions and
membrane material, and (ii) electric potential jump at the feed solution-active layer interface,
known as Donnan potential. The model incorporates a mechanism for varying membrane fixed
charge as a function of ion concentration and pH inside the active layer of the membrane. In
addition to sodium and chloride ions, hydrogen and hydroxide ions are also also taken into
solutions was developed, which allowed prediction of rejection coefficients of all ions in the
mixture as functions of both salt concentrations and pH based on experiments with individual salts
(Figure 15), which was then verified in [89], showing a reasonable agreement between theoretical
Figure 18: Rejection vs. pH for a feed concentration of 6 x 10-4 M NaCl solution. Solid line according to the theory predictions
[89]. The membrane used was Osmonics SS10 cellulose acetate membrane.
Recently, Shen et al. apply non-equilibrium molecular dynamics to relate water transport to the
membrane structure for RO, arguing that existing models rely on macroscale assumptions and do
they found that membranes with similar density and tortuous paths differed in water transport,
which correlates with the percolated free volume through the membrane thickness. Molecular
collisions alter the structure of the membrane under hydration which also has an effect on the
transport of water molecules. They suggested that solute transport could correspond to its bonding
with the functional groups of the membrane and/or its hydrating solvation shell.
Energy consumption, which represents more than 50-60% of total costs in desalination [93], is the
key determining factor in the widespread employment of any technology. On the energy front,
and energy recovery devices (ERDs) are employed and lead to significantly lower energy
In the 1970s, RO consumption was over 15 kWh/m3 of water produced. Currently, RO consumes
as low as 2 kWh/m3. As a whole, RO plants today consume 2.5-5 kWh/m3 of water produced [93].
This drastic reduction in overall energy consumption is a result of lower energy consumption in
each of the components making up the RO plant. These include the pretreatment system, high
pressure pumps, membrane material, membrane configuration, energy recovery devices (ERDs),
and post-treatment. In a recent review, Zarzo and Prats discuss strategies for minimizing energy
As RO technology is already running very close to its theoretical energy consumption, research
focus has shifted to improving system design, optimizing pre- and post-treatment, integrating RO
with other desalination processes and/or renewable energy sources [95, 96].
SWRO specific energy consumption can be further reduced through improvements in RO design
before lab-scale experiments are carried out to validate results. As such, modelling tools are crucial
to the development of energy-efficient process designs, because they provide a facile tool for
process optimization without the need for pilot testing. The configuration of RO membranes inside
pressure vessels has been the focus of recent studies aimed at reducing energy consumption,
especially for seawater reverse osmosis (SWRO). Part of the driving force is the tradeoff between
membrane selectivity and permeability [97]. Membranes with high selectivity may separate salt
efficiently, but the low water transport results in high SEC. On the other hand, high-flux
membranes have higher water production, but also a greater tendency to foul [94]. Lin and
Elimelech compared SSRO, two-stage RO and CC-RO in terms of specific energy consumption
and average water flux [98] as an indication of RO mass transfer energetics and kinetics,
respectively. In the single stage process, the feed stream is split into brine and permeate streams
as it passes through the RO membrane. In a two-stage RO process, the brine stream of the first
stage becomes the feed to the second stage and permeate streams of both stages are collected. In a
CC RO system, the brine is mixed with the feed solution and sent through the membrane (Figure
15).
Figure 19: Schematic showing A) Single stage RO (SSRO), B) Two-stage RO and C) Closed-circuit (CC) RO [98]
Their results show that a CC-RO is less energy-efficient than a two-stage RO process due to the
extra energy required to reduce the entropy generated by the mixing of the feed and retentate [26].
A two-stage RO also yields a higher water flux for brackish water desalination, where the recovery
rate is typically high (90%) [98].
Although energy efficiency can be enhanced by adding more stages, the additional capital costs
associated with adding a stage outweigh the reduction in energy costs [99]. Other configurations
and routing of the brine and permeate are also necessary to bring about improvements in energy
efficiency and costs. Recently, Chong and Krantz [100] developed an energy-efficient reverse
osmosis (EERO) system in which they sought to increase overall water recovery by sending the
retentate from one or more SSRO as feed to a countercurrent membrane cascade with recycle
(CMCR), consisting of one or more low salt-rejection RO stages (Stage 1) and a high salt-rejection
stages (Stage 2) (Figure 16). In EERO, the retentate from an SSRO is sent to a two-stage CMCR.
Along with retentate reflux in Stage 1, the countercurrent flow of the retentate and permeate
streams result in lower osmotic pressure differential and therefore lower net specific energy
consumption [100]. By using EERO, an overall water recovery of 75% can be attained at a cost
Figure 20: Schematic of EERO process in which retentate from SSRO is sent to a two-stage CMCR [100]
In a later study focusing on the numerical model-based analysis of the EERO system [99], Chong’s
group evaluated the fouling potential of the EERO system. As shown in Figure 17, the elements
in the EERO modules maintained a lower flux. This is especially true for the lead elements and
can be significant in mitigating the effects of membrane fouling in these elements, as well as
Figure 21: Permeate flux Jw of an RO stage in conventional SSRO and EERO processes (simulated at 60% overall water
recovery) [99]
Kim and Hong introduced split partial single pass RO (SSP-RO) configuration in which the
permeate from only the back RO elements in a pressure vessel is blended with the feed to RO in
order to dilute the feed [101]. This results in high-quality permeate with lower energy demand.
They modelled the process and found that energy efficiency is maximized for the SSP RO process
when the permeate from the last element is blended with the feed. Compared to conventional
single-pass RO, the permeate from the modified process was 15% greater in purity and more
Typical configuration of a single stage RO applies the same membrane type throughout a pressure
vessel. This causes the front elements to be exposed to the feed seawater, resulting in a higher net
pressure difference and higher flux across these elements in comparison to subsequent elements.
This uneven distribution of net driving force also results in greater propensity to foul for the front
elements. An improvement of the process design aimed at making the flux distribution along a
pressure vessel more uniform, is to use a hybrid membrane configuration, known as internally
staged design (ISD). ISD involves using lower flux membranes at the front and high flux
membranes in subsequent elements (Figure 18) [102, 103]. It has been shown that such a hybrid
membrane inter-stage design has the potential for significant reductions in permeate costs by
requiring fewer pressure vessels and membranes [104, 105]. This section covers optimization of
ISD and other such developments in RO membrane configurations using modelling and simulation
Figure 22: Schematic showing a hybrid membrane configuration, or internally staged design (ISD) [102]
Jeong et al. developed a finite difference model to numerically optimize ISD in the presence of
membrane process and investigated the impact of the membrane element arrangement on long-
term operation (Figure 19) [106]. Compared to conventional designs where the same membrane is
incorporated throughout the vessel, the ISD resulted in greater water flux and higher energy
efficiency for long-term operation, without compromising on the quality of the permeate (< 400
mg/L). They applied finite difference approximation to numerically calculate the spatial and
temporal distribution of water and solute transport. This is done by discretizing the spatial domain
x into 100 finite elements and time on a daily basis over a period of 90 days. Salt and water fluxes
over each segment were calculated from the first to the last membrane element at a given time
step. Model parameters for calculation of the cake layer growth are updated based on those of the
previous time step and then applied to the equations for steady-state mass transfer. Similar
recursive algorithms have been used to predict fouling in other studies as well [107-111].
In a full-scale RO process, four to eight RO elements are connected in series in a single pressure
vessel and each membrane’s performance varies depending on the temporal and spatial variation
in local fluid conditions. For an accurate calculation of local water and solute fluxes, they
considered the spatial distribution of cross-flow velocity, solute concentration, and trans-
Figure 23: Schematic of RO process illustrating discrete spatial and time domains for numerical calculations [106]
Han et al. improved vessel design by implementing ISD on a single-pass SWRO design and
evaluated the effect of configuration on SEC, permeate water quality, and boron rejection [103]
using ROSA9, the commercial simulation program provided by Dow Water and Process Solutions.
They used three types of RO membranes: high rejection, standard, and high flux membranes (Table
3) in standard configuration and six ISD configurations (Table 4) in a single stage single pass RO
system. They found that using an HID configuration with 3 standard membranes in the front and
4 high flux membranes in the tail saves 0.41 kWh/m3 for the same recovery rate and feed conditions
[103].
Table 3: Specifications of selected SWRO membranes used by Han et al. [103] (32,000 ppm NaCl, 800 psi, 25 °C).
400i rejection
400i
400i
XHR SW30XHR-400i
HRLE SW30HRLE-400i
ULE SW30ULE-400i
Optimal design of RO units has been the focus of considerable research. According to Kotb et al.¸
most of these optimization studies involve complex models or highly nonlinear equations with
many constraints [112]. In their recent study, they implemented a simple transport model to
determine the operating parameters corresponding to optimum RO system structure i.e. single,
two, and three-stage arrangements with respect to minimum permeate production cost for a given
permeate flow rate with defined maximum total dissolved solids (TDS) [112]. They found that the
minimum overall cost per unit permeate for a three-stage system is 0.91 $/m3 produced at a rate of
17 m3/h.
Figure 24: Effect of permeate flow rate on minimum cost per unit permeate $/m3 for single module, two-stage and three-stage
modules [112]
Figure 20 suggest that the optimum permeate flow rate increases with the number of stages,
indicating that while a single-stage RO system is suitable for up to 6 m3/h, three-stage modules are
Obaidi et al. [113] optimized a two-stage/two-pass RO process for chlorophenol removal from
wastewater and found that they could increase rejection by 12.4% compared to SSRO, for a 40%
The cake filtration mechanism used to describe particulate fouling on the surface of NF and RO
membranes is extensively covered in literature [114]. Cake filtration models are often used when
scaling, pore blocking, and biofouling are not major contributors to fouling.
Tomaszewska et al. [115] used empirical equations and numerical modelling to formulate trends
on the membrane surface and, thus, to predict membrane scaling during RO. Numerical modelling
takes into account operating conditions as well as physicochemical properties of the feedwater
with and without antiscalants. In comparison to traditionally used methods to predict scaling such
as RSI and LSI, the model suggested in this study predicts scaling phenomena as well as reactions
between water and antiscalants. Table 5 shows the expressions for calculation of water flux and
Table 5: Expressions for average water flux and net SEC for selected RO configurations
System Ref.
����
Average water flux 𝑱𝑱𝑱𝑱 Specific energy consumption
configuration
Closed- ∅ 𝟏𝟏 𝑵𝑵 + 𝟏𝟏 [98]
𝟏𝟏 + 𝜺𝜺 +
𝑵𝑵 ∅ 𝟐𝟐𝟐𝟐
circuit RO 𝒍𝒍𝒍𝒍(𝟏𝟏 + )
𝑵𝑵𝑵𝑵
(CCRO)
stage) 100]
𝟏𝟏 − 𝑹𝑹𝑹𝑹𝑺𝑺𝑺𝑺𝑺𝑺𝑺𝑺 𝚫𝚫𝝅𝝅 𝟏𝟏 − 𝑹𝑹𝑹𝑹𝑺𝑺𝑺𝑺𝑺𝑺𝑺𝑺
+ � � � 𝜼𝜼 𝚫𝚫𝝅𝝅
𝑹𝑹𝑹𝑹𝟐𝟐 (𝟏𝟏 − 𝑹𝑹𝑹𝑹𝟐𝟐 + 𝑹𝑹𝑹𝑹𝑺𝑺𝑺𝑺𝑺𝑺𝑺𝑺 𝜼𝜼𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑 − 𝑹𝑹𝑹𝑹𝑺𝑺𝑺𝑺𝑺𝑺𝑺𝑺 + (𝟏𝟏 − 𝑹𝑹𝑹𝑹𝟐𝟐 ) 𝑬𝑬𝑬𝑬𝑬𝑬
𝑒𝑒 1 1
*TCF = exp �𝑅𝑅 (298 − 273+𝑇𝑇)�
Khayet et al. [116] used response surface methodology (RSM) and ANN to develop predictive
models for RO desalination. They used a polyamide thin film composite membrane in spiral wound
configuration, with aqueous NaCl solutions as model feed solutions and developed the models
based on experimental designs. The input variables for their model were NaCl concentration in
feed, feed temperature T, feed flow-rate Q and operating hydrostatic pressure P. RSM is an
optimization approach that allows combination of several input factors to optimize a given
objective function [117], which in this case is the RO performance index given by salt rejection
factor x permeate flux. Response surface models involve fitting a polynomial regression model
using the input variables to determine the critical point, i.e. maximum performance index. On the
other hand, ANN is a non-linear processing system in which neurons, or nodes, and connections
between them are used for mapping input and output data [118]. A neuron is a computational
processor which operates in conjunction to other neurons such that the connections are
characterized by weights and biases. It was discovered that a single RSM model could not
adequately predict the performance index over a wide range of salt concentration (for both brackish
and seawater desalination conditions). They found that pressure played an important role at higher
feed temperatures and the feed temperature was an important factor at higher operating pressures.
The optimal conditions found through ANN were better than those by RSM [116] and the predicted
values from ANN closely fit experimental data with a correlation coefficient close to unity, as
shown in Figure 25. This was attributed to selection of optimal ANN architecture and other factors.
Figure 25: RO performance index predicted by ANN vs. experimental values [116]
Due to the complexity of modelling transport and separation, especially of charged solutes, many
studies have focused on NF modelling in the last two decades [50, 119]. In this section, theories
pertaining to ion transport and fouling in NF are considered, and existing models are connected
There are several models describing transport through NF membranes, including irreversible
thermodynamic models, pore models, space charge models, electrostatic, and steric-hindrance
models [49, 120]. A summary of commonly used ion transport models in NF is shown in Table
76, and described in detail in [30, 49]. Modified versions of these models are used to describe
separation of more complex solutions with NF, such as mixed salt solution [49]. Physical sieving
is the dominant mechanism for rejection of large molecules and solutes, whereas the chemistries
of solute and membrane take precedence for ions and low molecular weight organics. In most
models, the membrane is assumed to be a bundle of charged capillaries with specific structural
parameters, namely pore radius rp, with a ratio of porosity to membrane thickness (ε/l) and surface
charge density.
models, in which the membrane structure is not considered and transport is described as an
irreversible process which continuously produces entropy and dissipates free energy, or models in
which the structural and physiochemical properties of the membrane dictate solute transport such
which the solute flux is a function of solute permeability coefficient 𝑃𝑃𝑖𝑖 , average solute
concentration in the membrane 𝑐𝑐̅𝑖𝑖 , solvent permeability coefficient 𝐿𝐿𝑝𝑝 , permeation pressure ∆𝑝𝑝,
reflection coefficient 𝜎𝜎𝑖𝑖 (corresponding to the solute fraction rejected by the membrane), and the
model, the solvent flux 𝐽𝐽𝑗𝑗 and solute flux 𝐽𝐽𝑖𝑖 of aqueous solutions containing a single solute are
expressed by:
Both diffusion and convection contribute to solute transport. Diffusion depends on solute
concentration while convection depends on the applied pressure. Retention then is dependent not
only on the flux but also on the solute concentration. The Spiegler-Kedem model expresses solute
flux in a differential form, when the concentration difference between retentate and permeate is
high.
This model was modified by Spiegler and Kedem, who expressed the solute flux Ji as:
𝑑𝑑𝑐𝑐𝑖𝑖
𝐽𝐽𝑖𝑖 = −𝑃𝑃′( ) + (1 − 𝜎𝜎𝑖𝑖 )𝐽𝐽𝑗𝑗 𝑐𝑐̅𝑖𝑖 )
𝑑𝑑𝑑𝑑
where P’ is the local solute permeability (𝑃𝑃’ = 𝑃𝑃𝑖𝑖 ∆𝑥𝑥). In this model, the solute permeability
coefficient and the reflection coefficient are obtained by fitting observed solute rejection R vs. flux
F, according to:
The retention coefficient which corresponds to the maximum rejection at infinite volume flux can
thermodynamic models require less information to set up and have been used to describe the
rejection behavior of many solutes. The limitation of using thermodynamic models is that less
Another kind of model involving porous membranes is the steric-hindrance pore (SHP) model. In
this model, separation is described in terms of membrane pore radius and the ratio of membrane
porosity to thickness (ε/l). For a system containing a single uncharged solute, the reflection
coefficient and the solute permeability coefficient can be expressed in terms of steric parameters
related to the wall correction factors in the convection and diffusion coefficients, as well as to the
Where 𝐻𝐻𝐹𝐹 and 𝐻𝐻𝐷𝐷 are steric parameters related to the wall correction factors in the convection and
diffusion coefficients, respectively, and 𝑆𝑆𝐹𝐹 and 𝑆𝑆𝐷𝐷 are the distribution coefficients of solute in the
convection and diffusion conditions, respectively. Another equation widely applied to describe
pure solvent flux through uniform cylindrical pores without any concentration gradient across the
𝜀𝜀𝑟𝑟𝑝𝑝2 ∆𝑃𝑃
𝐽𝐽𝑗𝑗 =
8𝜇𝜇𝜇𝜇𝜇𝜇
where 𝜇𝜇 is the solvent viscosity and 𝜀𝜀, 𝑟𝑟𝑝𝑝 , 𝑙𝑙 and 𝜏𝜏 are membrane porosity, pore radius, membrane
The Teorell-Meyer-Sievers (TMS) model, also known as the fixed-charge model, relates salt
rejection by charged membranes to the ratio of membrane effective charge density to feed
concentration. The TMS model, which is based on the Donnan equilibrium theory and the extended
potential and concentration [119, 121]. Donnan equilibrium theory describes the behavior of
charged particles near a semi-permeable membrane wherein some ions are retained by the
Built on the extended Nernst-Planck equation, the Donnan steric pore-flow model (DSPM) is
widely used to describe ion transport in NF. In addition to the sieving effect, this model also takes
into consideration ion transport in terms of diffusion and migration, resulting from concentration
and electrical potential gradients, as well as convection due to the pressure difference across the
membrane [30]. In the DSPM, the membrane is considered a charged porous layer in which the
pore radius, volumetric charge density, and effective membrane thickness are controllable
parameters [123]. It describes partitioning through steric hindrance and the Donnan equilibrium
theory.
(DSPM) interactions
Now that key transport models for NF have been discussed, the present section reviews new
models as well as improvements of existing NF transport models that have been developed in the
Kowalik-Klimczak [124] evaluated the DSPM model for analyzing NF for the removal of
chromium(III) ions from an acidic salt solution. The pore dielectric constant was identified by
testing at different pressures and feed compositions. The permeate flux values obtained from the
Kong et al. employed the DSPM and dielectric exclusion model to predict the performance of
two NF membranes for the rejection of six haloacetic acids (HAA) and six pharmaceuticals
(PhACs) with different molecular weight, hydrophobicity, and charge [125]. Their model could
predict the rejection of HAAs with a mean standard error less than 5%. However, the model
overestimates the rejection of PhACs. This is because even though the model cites diffusion as the
predominant mass transport mechanism, experiments show that diffusion had a much smaller
contribution. The disagreement of model predictions and experimental values is possibly due to
inappropriate quantification of the hindrance factors for convection and diffusion, i.e., HD and HF.
Abdellah et al. applied NF for the recovery of bio-derived solvents from binary mixtures with
canola oil [126]. They used Maxwell-Stefan formulations together with Flory-Huggins solubility
model to describe the flux data as a function of concentration, feed temperature, transmembrane
Labban et al. applied the Donnan-Steric pore model with dielectric exclusion (DSPM-DE) to
describe membrane performance for a low pressure NF softening process [119]. They validated
the model by comparing with experiments of various feed chemistries including uncharged solutes,
single salts, salt mixtures and artificial seawater to characterize and predict its performance. Using
the model, they found that the high rejection of NF membranes to multivalent ions results from
primarily the membrane pore dielectric constant, followed by pore size (Figure 28). Membrane
charge density and membrane thickness were not as crucial in determining rejection for softening
applications.
Figure 28: Sensitivity of NF salt rejection for low pressure softening to intrinsic membrane properties [119]
Ochando-Pulido applied a boundary flux theory to model the performance of and predict the
fouling in NF for purification of olive mill wastewater after pretreatment [127]. The boundary flux
concept is a combination of critical and threshold flux and is a direct function of time [128]. It
separates membrane operation into two regions: one in which the impact of fouling is very low
and another in which fouling builds up exponentially [127]. Critical flux theory has been used to
describe the maximum permeate flux before fouling occurs, and it has been a crucial component
in membrane process design for all processes [129, 130]. An alternative concept, the threshold
flux, then emerged as the flux that separates a low fouling region from a high fouling region [130,
131], and was subsequently used to model fouling where critical flux was not applicable. In their
work, Ochando-Pulido estimate the boundary flux through by determining the fouling parameter
in each region. The experimental flux decline was in agreement with the boundary flux model
developed. They found that a high permeate productivity of up to 68.2 L m-2 h-1 could be reached
The pore hindrance transport model was initially developed for neutral solutes in pressure-driven
nanofiltration [132]. Recently, Xie et al. applied the pore hindrance transport model for the first
time to forward osmosis, to describe the rejection of trace organic contaminants (TrOCs) as a
function of permeate water flux. The pore hindrance transport model relies on the steric hindrance
to the entrance of a molecule into the pore as well as viscous resistance inside the pore [133]. In
this model, the membrane is considered a bundle of cylindrical tubes of the same radius through
which spherical solute particles enter randomly. The ratio of solute radius to the membrane pore
radius is related by the distribution coefficient when steric interactions are considered. The size of
uncharged solutes and permeation experiments are used to obtain retention and/or reflection
coefficients which are then used to determine the average membrane pore radius and the ratio of
𝑟𝑟𝑠𝑠
solute radius to pore radius, λ = 𝑟𝑟𝑝𝑝
. Another method to determine λ is from the Hagen-Poiseuille
equation, in which experimental values of pure water permeability and the average pore radius
obtained by steric hindrance pore model are input. Silva et al. compared several correlations from
literature for the pore hindrance model and found that the most suitable correlation had been
proposed by Bungay and Brenner (Figure 28) [134], who provided a complete correlation for 0 <
λ ≤ 1 [135].
Figure 29: Diffusive pore hindrance factors from literature as a function of λ [134]
For both cellulose acetate and TFC polyamide membranes, the rejection of charged TrOCs was
higher as they are rejected by both size exclusion and electrostatic repulsion (Figure 29). For
neutral TrOCs, rejection was greater through the TFC membranes although they have the larger
pore size, which the authors attributed to greater pore hydration. Pore hydration is the attachment
of a layer of water molecules to the negatively charged membrane surface via hydrogen bonding.
Greater pore hydration results in reduced effective membrane pore size which translates to
membrane module [136]. They studied the effect of various operating parameters on water flux,
feed recovery rate, and the final concentration of the diluted draw solution. They found that the
counter-current crossflow mode of operation leads to greater water flux, higher recovery, as well
as higher DS final concentration, all of which are indicative of improved performance. From their
analysis, they developed a modified equation for the water extraction capacity of a draw solute,
Figure 30: Real rejection of neutral TrOCs vs. permeate water flux by TFC and CTA membranes; solid lines represent
predictions from the membrane pore hindrance transport model [137]
approach with a single experimental FO test [138]. The experimental component was carried out
to measure the water and reverse solute flux in the feed solution where DI water was used as the
feed and NaCl as the draw solute. They used a statistical approach to find the optimal water
permeability, salt permeability and resistance to salt diffusion in the support layer to predict the
water and reverse solute flux using ICP and ECP models [138]. Results from the model were in
close agreement with experimental values and can be used to predict experimental water and
Attarde et al. also used the ECP and ICP models and combined them with the Spiegler-Kedem
model to also allow description of mass transport through the active layer of a spiral wound FO
module [139]. They applied a nonlinear constrained optimization technique, together with
experimental FO data, to predict the unknown parameters and minimize the error function. To
minimize the error function, they used a hybrid function technique which includes a genetic
algorithm technique and Fmincon from MATLAB. First, the genetic algorithm reaches the region
close to the optimum point and uses that point as the initialization point for the Fmincon [139].
Figure 30 shows the simple algorithm used by Attarde et al. to determine unknown model
parameters. Comparing the SK model with the traditional solution diffusion model, they found
that the FO performance predicted in terms of permeate flux, solute flux as well as power density
In FO, permeation of the draw solute through the membrane in reverse lowers the driving force
for water flux, adversely affects the feedwater quality, and is also met with resistance from the
foulant cake layer on the membrane surface. The foulant layer increases CP and cake-enhanced
osmotic pressure (CEOP) and reduces water flux. Modelling can help overcome limitations of
experimental instruments to study the various aspects of fouling in FO, allowing researchers to
evaluate the effect of changing physical and chemical parameters on FO fouling without an
experimental setup. Given the potential of modelling in this area, and the rapid growth of research
in FO, the limited number of studies carried out to investigate fouling in FO processes is startling.
Park et al. developed a numerical model to predict the flux decline due to colloidal fouling in an
FO membrane system [107]. They used a control volume approach and assumed that the cake layer
growth is based on a first-order reaction to derive the kinetic equation; see the schematic shown in
Figure 31. They found that the resistance of the cake layer is a major contributor to flux decline in
the beginning, but increased reverse draw solute permeation through the membrane had little effect
on flux decline. However, flux decline depends on the diffusivity, and hence the selection of the
They used the resistance-in-series model used to express the flux of an osmotically driven process
as:
𝐽𝐽𝑤𝑤 is the permeate flux; 𝜋𝜋𝑖𝑖 and 𝜋𝜋𝑎𝑎 are the osmotic pressures of the active layer-support layer
interface and active layer surface, respectively; 𝜇𝜇 is the dynamic viscosity; R, 𝑅𝑅𝑚𝑚 and 𝑅𝑅𝑐𝑐 refer to
the membrane resistance, intrinsic membrane resistance, and cake layer resistance, respectively.
Recently, Wang et al. modified the solution-diffusion model to incorporate draw solution
concentration and operating temperature and focused on maximizing FO water flux with respect
to these parameters for a commercial thin-film composite membrane [140]. They quantified the
effect of each parameter using a concentration-induced flux increment (CIE) and a temperature-
induced flux increment efficiency (TIE). Interestingly, CIE increased with an increase of draw
solution concentration and increased with an increase in temperature. On the other hand, the TIE
increased when the temperature was raised from 25 to 47 °C and also increased for increased draw
solution strength. Figure 32 shows that values projected theoretically were aligned with those
regions (Figure 33), which can be of significance for optimizing FO water flux.
Figure 33: Predicted and experimental values of water flux vs. draw solution concentration at different temperatures [140]
Figure 34: Temperature and draw solution concentration-sensitive zone as determined by Wang et al.’s modified solution
diffusion model. Red line indicates where water flux is equally sensitive to temperature and concentration [140]
Modelling approaches for FO fouling hold strong potential for growth and development of such
techniques will help not only in predicting fouling in FO, but also in optimizing operational
Table 7 shows process parameters and intrinsic membrane properties that can be optimized for
Table 7: Selected variables and objective functions for membrane-based desalination processes [116, 119, 140]
Filippini et al. analyzed a hybrid MED and RO system for seawater desalination by developing
models for each of the two systems and an integrated model to evaluate several configurations of
the two processes in the hybrid system [141]. Y. Chan modelled mass transport through
nanomaterials, in particular carbon nanotubes and graphene sheets, as membrane materials for RO
[142]. Shahzad et al. [143] proposed a tri-hybrid system consisting of RO and a multi-evaporator
adsorption system (ME-AD), arranged in series for maximum recovery from pretreated feed.
Theoretical results show that the overall recovery rate for seawater desalination on this hybrid
system can be as high as 81%. Coupling of RO with thermal desalination systems as well as with
renewable energy systems has been briefly reviewed by Qasim et al. [144] in a recent review.
Pretreatment is usually carried out to reduce the natural organic matter and suspended solids in the
feed to minimize fouling in the desalination unit. However, most pretreatment processes do not
involve removal of dissolved solids such as multivalent ions that can later precipitate and cause
scaling on heat exchangers, which reduces the heat transfer efficiency in both MSF and MED
systems [145]. Although antiscalants are used to prevent scaling, pretreatment needs to be able to
remove multivalent ions. Nanofiltration and forward osmosis have been considered as
pretreatment alternatives for MSF and MED. Due to nanofiltration being an energy-intensive
conditions for pretreatment to MSF and found that increasing the brine temperature from 25 °C to
osmosis (FO) offers a low-energy alternative. FO depends on the osmotic pressure difference
between a concentrated draw solution and a feed stream across a semi-permeable membrane [146].
Apart from draw solute recovery and internal concentration polarization, FO membranes are still
susceptible to fouling, although to a lesser degree than membranes in hydraulic pressure-driven
processes such as RO [107]. Fouling in FO has been found to be more reversible and less
compacted as compared to RO [147]. Altaee and Zaragoza have developed a model to estimate
power consumption in FO for seawater softening in an FO-MSF plant [148, 149]. Figure 35 shows
the steps used to predict the concentration of diluted draw solution and estimated permeate flow
Comparing the water flux and power consumption in FO to that in NF for pretreatment of MSF,
Altaee and Zaragoza found that although water flux in NF is higher than in FO, specific power
consumption and, hence, operation cost is higher for NF than for FO. However, feed salinity did
not affect the power consumption in FO whereas power consumption for NF increased with
salinity. In another study, Altaee et al. applied the same model to evaluate the effectiveness of FO
pretreatment in the removal of divalent ions for a high temperature FO-MSF/MED hybrid system
[150]. They simulated MSF at 130 °C using FO as pretreatment and calculated the concentration
of Ca2+ and SO42 – ions in each stage. They found that FO pretreatment increased MED TBT to 85
°C, which led to a distillate flow rate 1.8 times higher than a TBT of 65 °C. They also developed
an FO pretreatment-MED Scale Index (FMSI) to determine the required FO recovery rate to avoid
scale problems. This scale index was based on calculating the Ryznar Scale Index (RSI) with
different MED operating temperatures and FO recovery rates. They then used the FMSI to
determine the required mixing ratio for FO and NF feed as a pretreatment for MED and found that
Pal et al. recently modelled an integrated FO-NF system for the treatment of hazardous wastewater
[151]. They applied a flat sheet cross flow FO membrane module coupled with an NF system to
recover the draw solute. The developed model is based on solution-diffusion mechanism for FO
while the DSPM with dielectric exclusion phenomenon is used to describe the NF component. The
transport mechanisms that dominate NF are diffusion, convection, and electromigration while only
the latter two exist in the case of FO. As mentioned earlier, ionic separation at the solution-
membrane interface is described by Donnan equilibrium and steric effects. Dielectric exclusion
phenomenon is included because DSPM on its own is suitable for uncharged solutes, but not in
the case of mixed electrolytes solution and/or multivalent ions [152]. They found that performance
predicted by the model was in agreement with experimental data with a low relative error of < 0.1
and a high correlation coefficient; these results suggested that their model will help scale-up of
this hybrid system. Zaviska and Zou modelled a bench scale FO process as pretreatment for RO
and found that FO can help avoid RO fouling while achieving higher permeate recovery [153]. In
this simulation, the diluted draw solution becomes the RO feed and is then re-concentrated for
reuse in FO. The model took into account flux, water recovery, and the final draw solution. They
assumed 1000 m3/day of RO feed (i.e., diluted draw solution outlet from FO process) which is
treated using 1000 m2 RO membrane, together with a pressure exchanger as ERD. They used their
model to determine the operating pressure and energy consumption required for RO.
Senthil and Senthilmurugan simulated an integrated SWRO-PRO system for eliminating post-
treatment of brine from RO. By modelling six SWRO-PRO configurations using seawater as feed
solution, they found that direct mixing of diluted PRO draw solution with RO feed could reduce
Obaidi et al. [155] analyzed a multistage multi-pass medium-sized brackish water RO (BWRO)
desalination plant in Jordan, producing 1200 m3/day. For the spiral wound RO membranes, they
developed a model based on solution diffusion and employed it to simulate operation of low-
salinity BWRO. Plant data obtained experimentally was in line with results from the simulation,
as shown in Figure 36. When they carried out sensitivity analysis on their model, they found that
feed flow rate and operating pressure are the main factors affecting product salinity.
Figure 36: Effect of plant operating pressure on recovery rate of pass 1, pass 2, and total [155]
Malik et al. optimized MSF, RO and MSF-RO desalination systems for a total production capacity
alone, MSF alone and RO/MSF hybrid and additional equipment such as mixers and splitters, to
allow analysis on a single flowsheet (Figure 37). Using a feed separator eliminates the need for
multiple flowsheets. They optimized operating and design variables and found that the hybrid
system has a higher overall recovery and lower operating cost than the MSF system and better
Figure 37: Desalination superstructure schematic used to analyze various MSF/RO configurations [156]
Bartholomew [157] developed a cost optimization model for osmotically assisted RO, in which
they investigated the relationship between membrane stages, saline sweep cycles, and makeup,
purge and recycle streams for high-salinity feeds in the range of 50,000 to 125,000 ppm TDS. The
optimized design resulted in costs less than $6/m3 water with water recoveries between 30-70%.
They studied 3 cases: (1) feed TDS of 75 g/L and 50% water recovery, (2) feed TDS of 75 g/L and
70% water recovery, and (3) feed TDS of 125 g/L and 40% water recovery [157]. They found that
cost-optimal unit water cost was the lowest for case 1, as shown in Figure 38A. Figure 38B shows
the normalized costs of the various components (membrane capital and replacement costs, capital
costs of pumps, pressure exchanger, electricity costs and other operating expenses). They used a
nonlinear optimization model with the objective of minimizing the cost of the OARO system and
found that OARO can be economically feasible for feed salinities of up to 125 g/L and water
3. Future direction
Despite reduced energy consumption, energy still makes up the largest portion of overall costs in
desalination systems. With comparable advances in renewable energy systems such as solar and
wind energies [158], the use of renewable energy to power desalination is now being considered
an important alternative for providing fresh water, as is evident from the large number of reviews
on this topic [159-164]. Modelling the coupling of RE systems with desalination technologies is
an active area of research with tremendous potential. Several studies have recently been carried
out on the integration of renewable energy systems with desalination systems, including hybrid
particular interest and the most widely form of renewable coupled with desalination, as the most
water scarce regions are also those with the greatest solar energy abundance [170]. Kasaeian et al.
which they included several recent modelling and simulation studies. Often, each segment, the
renewable energy system and the desalination system, is optimized separately [172].
Mentis et al. developed a multiparameter dimensioning tool to evaluate the integration of
renewable energy with RO desalination on the islands of South Aegean Sea. The model includes
different desalination capacities depending on the size and water demand of the island, selection
of appropriate renewable energy technology for supplying electricity to the RO plant, energy
balances of the integrated system, and the cost of water production and electricity [173]. They
found that the price of water on the smaller island would be greater; however, RE would enable
both islands to sell water at a much lower cost than the current price.
Salehi et al. investigated the feasibility of producing distilled water with a geothermal power
system [174]. They applied a three-objective optimization procedure on a geothermal power plant
using a genetic algorithm, focused on optimizing electricity output, product unit cost, and distilled
water flowrate. In their study, which consisted of a double-flash geothermal power plant, the
decision parameters were the pressures of the two flash chambers and the temperatures of the
vaporator and generator. They compared two configurations of integrating a geothermal power
system with thermal distillation: one in which the reinjected geofluid temperature is assumed to
have a temperature of above 100 °C, and another in which the temperature is below 100 °C, but is
enhanced with an absorption heat transformer and the upgraded thermal energy is used to produce
purified water. They found that the first configuration yielded higher distilled water flow rates,but
Heidary et al. [175] recently designed a small scale MSF-RO desalination system producing 25
liters per hour and powered by hybridization of solar and wind. For the desalination system, they
studied six models and optimized air pressure, seawater temperature, seawater flow rate and
seawater salinity to minimize water production costs and maximize the volume of product water.
Figure 28 shows a schematic of the hybrid solar wind MSF-RO system. For the hybrid solar wind
energy system, they proposed an energy system composed of wind turbine, solar panel and solar
collector, electricity from all of which is saved in batteries. In the mathematical model, the weather
conditions, design parameters of the RO-MSF models and the energy demand of the desalination
The greatest water production was obtained with an integrated MSF-RO system, where part of the
brine from MSF goes through a single pass RO while the remaining is mixed with RO brine. The
system where the heat rejection of the MSF condenser liquid is used as feed for the RO was shown
to be the most energy efficient for large-scale production of > 1000 L/hour.
The cost of the hybrid system is the sum of the total cost for each subsystem i.e. wind, solar,
battery, RO and MSF system, which includes direct and indirect capital costs as well as operation
and maintenance costs. Economic optimization based on maximizing water production and
minimizing water cost for each of the configurations showed that hybridization of wind-solar and
RO-MSF were the optimal choices when compared to fossil fuel RO or MSF, fossil fuel RO-MSF,
Future direction involves lowering the energy consumption of newer membrane-based processes
such as FO through novel process design and configuration optimization, as has been the case for
RO recently. Among renewable energy systems for desalination, most of the modelling has
focused on photovoltaics coupled with reverse osmosis. More studies need to be carried out to
study the coupling of other desalination processes as well as hybrid RE systems. Additionally,
research in integration of new generation artificial intelligence algorithms into desalination is still
in its infancy and is expected to grow in coming years. As mentioned earlier, modelling and
experimental studies go hand in hand, and for many of the more complex or newer processes such
as nanofiltration, measuring tools are still lacking in providing a complete understanding of the
process. To design models that will reflect performance closer to real systems, concurrent
advancements in measurement methods and tools are necessary. Additionally, the same complex
processes may require hybridization of conventional and artificial intelligence models – an area in
which very limited work has been carried out to date. As has been shown in some of the studies
highlighted in this review, nondeterministic algorithms are becoming more relevant due to their
ability to model systems for a wider range of operating conditions that is not often possible with
conventional methods. Research will focus on benefitting from the strengths of the two types of
approaches, integrating them to achieve accurate solutions for complex systems using the least
amount of resources. Another area of particular interest is the model-based process control in
membrane-based technology.
4. Conclusion
As installed desalination capacity grows worldwide, there is an imminent need to reduce energy
consumption for desalination processes either through new configurations and process design for
parameters. Each of these solutions necessitates the need for model building to accurately describe
and analyze desalination processes. In the area of thermal desalination, although some new
configurations have been studied through modelling, the technology is reaching saturation and
recent studies are focused on understanding and controlling scale behavior on surfaces. For the
duration thermal desalination has been around, it is surprising that scale formation affecting MSF
performance was very little understood before recent years. For RO, modelling tools are being
used to assess the feasibility of new configurations of the membrane module, with much attention
on internally staged design modules to balance flux through the module and, hence, to minimize
energy consumption. For other pressure-driven membrane processes such as nanofiltration, simple
models are being modified to develop a deeper understanding of the transport mechanism, which
takes into account diffusion and convection through the pores and/or charged membrane.
Regarding the aspects discussed in this literature survey, there are a few gaps in literature we have
• Experimental studies in optimization of membrane materials and draw solutions for FO are
pretreatment technology for MSF or RO, either separately or in hybrid with other
desalination processes. Currently, the diluted draw solution needs to be further treated and
studies of such hybrid FO systems are still limited. This step would help determine the
• The use of models to predict specific kinds of fouling for all membrane-based desalination
processes, including newer processes such as forward osmosis and membrane distillation,
could open a whole new avenue for research. This means more models for validating the
5. References
6. Abbreviations
RO reverse osmosis
FO forward osmosis
NF nanofiltration
CP concentration polarization
TMS Teorell-Meyer-Sievers
DE differential evolution
SK Spiegler-Kedem model
7. Symbols
Js Salt flux
Jw Pure water flux
ε membrane porosity
l membrane thickness
rp pore radius
R membrane resistance
𝜇𝜇 dynamic viscosity
RR recovery rate
N Number of stages
𝜂𝜂𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 efficiency