Desalination 394 (2016) 3043

Contents lists available at ScienceDirect

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

Global optimization of MSF seawater desalination processes

Chandra Sekhar Bandi, R. Uppaluri , Amit Kumar
Department of Chemical Engineering, Indian Institute of Technology Guwahati, North Guwahati, Assam 781039, India

H I G H L I G H T S
For MSF-BR and MSF-M processes, DE provided global optimal solutions.
Cost based MSF process ranking is MSF-BR > MSF-M > MSF-OT* (* refers solution with penalty).
For important MSF process parameters, obtained solutions improved by 2.31%, 3.9%, 2.92%, 20.24%, 3.53% and 5.2%.

a r t i c l e

i n f o

Article history:
Received 2 February 2016
Received in revised form 5 March 2016
Accepted 6 April 2016
Available online 12 May 2016
Keywords:
Differential evolution algorithm
Global optimization
Seawater desalination
Multi stage ash (MSF)
MSF-BR
MSF-OT
MSF-M
SQP
Modeling
Optimization

a b s t r a c t
This article addresses the global optimal design of multi-stage ash desalination processes. The mathematical formulation accounts for non-linear programming (NLP) based process models that are supplemented with the
non-deterministic optimization algorithm. MSF-once through, -simple mixture (MSF-M) and -brine recycle
(MSF-BR) process congurations have been evaluated for their optimality. While freshwater production cost
has been set as the objective function for minimization, mass, energy and enthalpy balances with relevant supplementary equations constitute the equality constraints. Differential evolution algorithm (DE/rand/bin) was
adopted to evaluate the global optimal solutions. Further, obtained solutions have been compared with those
obtained with MATLAB optimization toolbox solvers such as SQP and MS-SQP. The global optimal solution corresponds to a variable value set of [2794.4 m3/h, 1.0499, 7.62 m, 3.359 kW/m2 K, 3.297 kW/m2 K, 3.042 kW/m2 K
and 22] for decision variables [WM, RH, LT, UB, UR, Uj, NR] in the MSF-BR process to yield an optimal freshwater
production cost of 1.0785 $/m3. Compared to the literature, the obtained global solution from DE is 2.31% better.
Further, inequality constraint resolution has been excellent for DE but not other methods such as MS-SQP, SQP
and DE-SQP.
2016 Elsevier B.V. All rights reserved.

1. Introduction
Among several technologies viable for potable water production,
the desalination of sea and brackish water is an established technology in several countries including the USA, Persian Gulf and
European countries [15,35]. Based on the working energy principle,
desalination processes are further classied primarily into two classes namely thermal processes that involve phase change due to addition of heat and membrane processes that involve pressure energy.
While thermal processes are primarily classied into multi-effect
evaporation (MEE), MSF and vapor-compression (VC) processes,
membrane processes are primarily classied into RO and electrodialysis (ED) processes. Among various alternate technologies for sea
water desalination, MSF processes have the promising features of

Corresponding author.
E-mail address: ramgopalu@iitg.ernet.in (R. Uppaluri).

http://dx.doi.org/10.1016/j.desal.2016.04.012
0011-9164/ 2016 Elsevier B.V. All rights reserved.

large scale operation and ability to deliver good quality potable

water (550 ppm total dissolved solids).
A typical MSF process involves brine heating followed with ash
distillation in multiple stages and subsequent heat recovery. Thereby,
a MSF process plant has three important sections namely brine heater,
heat rejection and heat recovery sections. Design variations in the
MSF process systems refer to either once through (OT) or simple
mixer (M) or brine recirculation (BR) process congurations to yield
MSF-OT or MSF-M or MSF-BR processes respectively. Among these,
while MSF-OT is the simplest in design, it is not as efcient as the
MSF-BR system.
The design of efcient MSF processes invariably requires simulation
and optimization studies. Several researchers have conducted simulation studies to obtain insights upon the process performance of MSF
processes. These have been contributed by Mandil and Abdel Ghafour
[19], Helal et al. [2], Al-Mutaz and Soliman [14], Rossol et al. [26], Thomas et al. [29], Abdel-Jabbar et al. [28], Hawaidi and Mujtaba [6], and
Tayyebi and Alishiri [34]. Many of these literatures emphasized upon

C.S. Bandi et al. / Desalination 394 (2016) 3043

stage-to-stage calculations and deployed NewtonRaphson method or

tridiagonal matrix (TDM) formulations solved with Thomas algorithm
(TA) for evaluation of MSF process performance.
Further, optimization studies have also been conducted by several
researchers. These include MSF-OT processes [35,2123]; MSF-M processes [20] and MSF-BR processes [1,35,6,7,8,9,16,18,24,25,31,32,33].
Considering the minimization of water production cost as objective
function, the literature refer to the deployment of either one of the following methods: genetic algorithm (GA) [24,32]; sequential quadratic
programming (SQP) method [22,23], deterministic optimization
methods built in gPROMS [6,25], generalized reduced gradient (GRG)
[1,35,20] and in DICOPT++ [22,23]. Further, MATLAB programming
environment has also been used in several engineering applications as
a competent modeling tool for simulation and optimization studies
[36,37,38,39].
A critical analysis of the available literatures in optimization studies
refers to the following. Firstly, earlier research works mostly addressed
either MSF-BR or MSF-M or MSF-OT processes for process optimization
based insights. Only [35] addressed MSF-OT and MSF-BR process optimization but not the MSF-M process. The authors adopted GRG optimization method which is a local optimization tool. It is well known that
GRG might provide local solutions whose quality could not be judged
in conjunction with the global optimality. Further, GRG is well known
to be non-rigorous and fails to solve problems with larger number of inequality constraints, as the method needs the satisfaction of all inequality constraints in each iteration. While SQP method foregoes such
limitation, the SQP also could not provide insights upon the quality of
generated optimal solutions. On the other hand, non-deterministic
models such as GA were only investigated for the MSF-BR but not
MSF-OT and MSF-M processes. Thus, it is apparent that global optimization methods have not been applied till date for the comparative assessment of MSF-BR, MSF-M and MSF-OT processes.
Secondly, a critical issue with respect to alternate optimization
methods such as GRG, SQP, and GA, is with respect to the satisfaction
of inequality constraints. The traditional approach to couple a penalty
function with cost function may or may not yield feasible solutions
using GRG and SQP methods, given the fact that these algorithms may
require additional ne tuning of optimization algorithm parameters
such as maximum number of iterations, maximum function evaluations, and penalty parameters, to obtain feasible solutions. Thus, it
might be the case that an engineer may have to spend a signicant
amount of time in ne tuning these parameters for the deterministic optimization methods. On the other hand, such insights may not be applicable for the non-deterministic optimization methods due to random
nature of solution search. Therefore, an important issue that also
needs to be addressed is the ability to fetch feasible solutions with similar penalty function parameters for both deterministic and nondeterministic optimization methods.
A third and essential insight is to visualize upon the sensitivity of
process and operating parameters using global optimization approaches. While such sensitivity analysis might be possible with local
optimization methods, they may not provide the most stringent sensitivity analysis. Therefore, the sensitivity analysis conducted with nondeterministic methods needs to be judged with that conducted with deterministic methods.
In summary, this work addresses three major objectives. The rst objective refers to comparative assessment of MSF-M, MSF-OT and MSF-BR
processes using non-deterministic optimization. The second objective refers to the evaluation of inequality constraint resolution ability for both
deterministic and non-deterministic methods. The nal objective is to
conduct sensitivity analysis of all MSF processes in the light of global
and local optimization. Differential evolution (DE) has been chosen as
the global optimization tool as it has not been studied for MSF process optimization despite being proven effective for other engineering optimization problems. Thereby, suitable benchmarks are expected to be set for
the engineering optimization of MSF processes.

31

2. Process congurations
A schematic representation of the MSF-OT, MSF-M and MSF-BR process congurations is presented in Fig. 1(a)(c). Among these processes,
while MSF-OT limits the temperature of the last stage to 3040 C for
winter and summer operations, the ashing operation on several ash
stages requires vacuum pressure conditions to achieve operating temperatures below 100 C. As indicated in the gure, the common features
of these process congurations are briey summarized as follows:
The feed seawater (WMF) at temperature TSea, is de-aerated and
chemically treated before being introduced into the condenser/preheater tubes of the last ashing stage in the heat recovery section.
The preheated feed seawater at temperature T2 enters the brine
heater tubes, where the heating steam (WS) is condensed on the
outside surface of the tubes. Eventually, the seawater reaches the
maximum design temperature value also known as the top brine
temperature (T3).
The feed seawater nally enters the ashing stages, where a small
amount of fresh water vapor is generated by brine ashing in each
stage. In each stage, the ashed off vapor condenses on the outside
surface of the condenser tubes, where the feed seawater (WMF)
ows inside the tubes from the cold to the hot side of the plant.
Thereby, the heat recovery process enables an increase in the feed
seawater temperature. The condensed fresh water vapor outside
the condenser tubes accumulates across the stages and forms the
distillate product stream (WMD).

Fig. 1(b) illustrates that the MSF-M process essentially consists of a

brine heater, heat recovery section and brine recycle mixing tank.
Hence, the MSF-M process conguration facilitates a brine recycle
stream to reduce fresh seawater requirements and associated chemical
pretreatment costs. This is achieved by mixing part of the blowdown
brine stream (WMR) with the feed stream (WMSC), thereby generating
a mixed stream (WMF) with higher salinity than that of the fresh seawater (set as 70,000 ppm for the upper bound according to El-Dessouky
et al. [11].
It can be further observed in Fig. 1(c) that the MSF-BR desalination
plant has heat rejection, recovery section and brine heater section. The
nal reject stream from the heat recovery section is being split into
two streams which serve as cooling seawater stream (WMCW) and
makeup stream (WM). The makeup stream is further chemically treated
and mixed in the brine pool of the last ashing stage in the heat rejection section. The mixed stream is sent to blowdown splitter S2 from
which the brine recycle stream (WMR) is introduced into the condenser
tubes of the last stage in the heat recovery section. The stream after absorbing the latent heat of condensation from ashing vapor in several
stages leaves the last stage and enters the brine heater, where its temperature is enhanced to saturation temperature (i.e., top brine temperature) at the prevalent system pressure.
3. Methodology
Process optimization of alternate MSF congurations has been
targeted by coding a competent simulation model that is supplemented
with a non-deterministic optimization algorithm. For comparison purposes, deterministic optimization algorithms have also been considered
to evaluate upon the efcacy of the non-deterministic optimization algorithm. The following sub-sections summarize the simulation and optimization models.
3.1. Simulation model
The simulation models for MSF-OT and MSF-BR processes were
adopted from Helal et al. [3]. For the MSF-M process, the simulation

32

C.S. Bandi et al. / Desalination 394 (2016) 3043

model was taken from Abduljawad and Ezzeghni [20]. Other than these
literatures, a comprehensive account of the models for MSF processes is
not apparent elsewhere in other literature.
The process simulation model accounts for the mass and energy balance equations applied for each stage including the brine heater, recovery and rejection sections. The process models are non-linear in nature
and consist only algebraic expressions. Further details of the process
models can be obtained from the cited literature and this section briey
outlines various important features of the same. As outlined in the literatures, the following assumptions are usually applicable for the process
models on a theoretical basis [3,10]:

d) The specic heat capacity of brine solution is a weak function of salt

concentration.
e) Distillate product is salt free; non-condensable gases have negligible
effect on the heat transfer process.
f) Effect of the boiling point rise and non-equilibrium losses on the stage
energy balances is negligible.
g) The average specic heat capacity of brine solution is equal to that of
the distillate.
h) The boiling point rise at the exit from the last recovery for MSF is negligible.
i) Heat loss is negligible.

a) Temperature proles of all streams owing within the plant are linear.
b) Each section has a constant value for heat transfer coefcient, heat
transfer area, boiling point rise and specic heat capacity (CP) of
brine solution.
c) The latent heat of vaporization of water () is constant, and independent of temperature.

The non-linear system of algebraic equations involving mass and energy balances has been deduced by carrying out overall mass balances
across blowdown splitter, rejected sea water splitter (for the MSF-BR
process), salt balances across mixer, energy balances across brine heater
and condenser, enthalpy balances on the heat recovery section, heat

Fig. 1. Schematic of (a) once-through MSF (MSF-OT), (b) brine-mixing MSF (MSF-M), and (c) brine recycle MSF (MSF-BR) seawater desalination processes.

C.S. Bandi et al. / Desalination 394 (2016) 3043

33

Fig. 1 (continued).

rejection section (MSF-BR process), brine heater, mixers and ashing

brine. Supplementary expressions for the process model include expressions for distillate product temperature (equilibrium correlation),
overall heat transfer coefcients in various process sections, inside
lm resistance, average tube side brine temperature and thermal resistance. The thermal resistance is accounted as a function of steam-side
condensing lm, steam-side fouling, tube metal and brine side fouling
resistance components.
Table 1(a) summarizes various parametric and design specications
required for the process simulation model for cross ow type MSF processes. Table 1(b) presents the process cost parameter data, which is
adopted from Helal et al. [3]. For various process models, the independent simulation model variables that need to be specied are (WMF, LT,
UB, UR and NR) for MSF-OT; (LT , UB , UR , WMSC , CMSMF and NR) for MSFM and (WM , LT , UB , UR , NR , RH and Uj) for the MSF-BR process. Henceforth, these variables are treated as independent (decision) variables
during process optimization.
3.2. Optimization model
The MSF process simulation model consisting of non-linear set of algebraic equations is formulated as a non-linear programming (NLP) optimization model dened as:
Min OF
Subjected to
f x 0
gx0

where refers to the objective function dened as the sum of annualized freshwater production cost and associated penalties. The process
simulation model is specied as f(x) = 0 for a specic MSF processes
and g(x) refers to the set of inequality constraints.
The total annualized freshwater production cost function is evaluated using relevant expressions for direct capital less intake investment
(CDCLIC), intake-outfall cost (CIC) (evaluated as a sum of costs of electrochemical equipment, civil work, electro chlorination, brine disposal cost
and annual plant intake-outfall), direct capital investment cost (CDCC),
indirect capital investment (CICC) and operating and maintenance cost
(COMC) (evaluated as a sum of costs of steam, chemical treatment,
power, labor and spares costs). Relevant expressions have been adopted
from Helal et al. [3]. Thereby, the objective function for MSF is modied
and expressed as:

MSF

C IC C DCLIC C ICC C OMC

 3
m
WY
y

 

$
y

penalty of gx:

Inequality constraints refer to lower bound and upper bound specications for makeup ow rate for MSF-BR (WM), feed ow rate for MSFOT (WMF) and MSF-M (WMSC), tube length (LT), number of recovery
stages (NR), heat transfer coefcients in various sections, brine loading,
brine velocities in various sections, rejected brine concentration values
(CMBD, Clast) and absolute values of various heat transfer coefcients.
The inequality constraint parameters and values have been presented
in Table 2(ab). The penalty function has been evaluated using large

34

C.S. Bandi et al. / Desalination 394 (2016) 3043

Table 1
(a) Parametric and design specications data for cross ow type MSF processes. (b) A
summary of MSF process cost parameter data.
(a)
Variable
a
CMF
CPR
Cpj
Cp
ID
OD
Nj
T1
T3
T6
TD
TS
TSea
WMD
R
j

B
W

Table 2
(a) Summary of lower and upper bound values for the optimal design of MSF processes.
(b) A summary of inequality constraint parameters for MSF optimization model.
(a)

Unit

ppm
kJ/kgK
kJ/kgK
kJ/kgK
m
m

C
C
C
C
C
C
m3/h
C
C
J/kg
kg/m3
kg/m3

MSF-OT
0.88
42,000
4.18513
4.17658
4.18513
0.02199
0.024069

110

30
114
25
1122
1.4

2346.3
1060
1000

MSF-M
0.88
42,000
4.18513
4.17658
4.18513
0.02199
0.024069

110

30
114
25
1122
1.4

2346.3
1060
1000

MSF-BR
0.88
42,000
4.18513
4.17658
4.18513
0.02199
0.024069
3
34
110
28
30
114
25
1122
1.4
1.78
2346.3
1060
1000

(b)
Unit

Value

Current work capacities

MSF plant section

m3/d

26,928

Reference capacities and ow rates

Ref. feed rate for intake calculation
Ref. blowdown rate for outfall calculations

m3/d
m3/d

1000
750

MSF plant cost data

Capital recovery factor
Parameter in xed capital cost term
Chemicals
Energy (pumping)
Spares
Labor

$/m prod
$/m3 prod
$/m3 prod
$/m3 prod

0.0963
5500
0.024
0.03
0.082
0.1

MSF-BR
Variable

Unit

WM
WMF
WMSC
RH
LT
NR
UB
Uj
UR

m3/h
m3/h
m3/h

kW/m2 K
kW/m2 K
kW/m2 K

MSF-OT

MSF-M

Lower

Upper

Lower

Upper

Lower

Upper

2000

0.8
7
18
2.7
2.7
2.7

3000

2
10
30
3.7
3.7
3.7

5500

7
15
2.7

2.7

10,000

15
50
3.7

3.7

5500

7
15
2.7

2.7

10,500

15
50
3.7

3.7

(b)
Constraints

Unit

Lower

Upper

Brine loading
VB
Vj
VR
CMBD or Clast
ABS(1 UBcal / UB)
ABS(1 Ujcal / Uj)
ABS(1 URcal / UR)

m3/h m width
m/s
m/s
m/s
ppm

900
0.9144
0.9144
0.9144

1200
1.8288
1.8288
1.8288
80,000
0.001
0.001
0.001

parametric settings that were dened for the initialization, propagation

and termination strategies of various optimization methods for the MSF
optimization problem. Further, it shall be noted that for several methods
including SQP, MS-SQP and DE-SQP, initial vector values (independent
variables) had to be dened for MSF-OT, MSF-BR and MSF-M processes.
Table 3(b) summarizes the initial vector values set for the optimization
studies. All simulations were conducted in MATLAB programming environment. SQP has been implemented from MATLAB optimization toolbox [27,36].
4. Model validation

positive penalty parameters that are set to realize the satisfaction of all
inequality constraints.
3.3. Optimization algorithm
Differential evolution (DE) algorithm has been applied for the nondeterministic optimization of NLP process models. Introduced by Storn
and Price [30], the DE optimization algorithm is a stochastic population
based direct search optimization method that essentially involves the
generation of new candidate solutions by combining the parent individual and several other individuals of the same population. This is facilitated by adding the weighted difference between any two population
vectors to a third population vector [13]. Further, the parent vector is replaced with the mutant vector only when the mutant vector provides a
better tness value [17]. Thus, DE is an effective, fast, simple, robust, and
inherently parallel technique that has few control parameters and
needs less effort to tune and adopt optimization parameters. The DE
has been applied as a sequence of mutation, cross-over and selection
operations for all populations. The maximum number of permitted generations has been set as the termination criteria.
The efcacy of the DE algorithm has been evaluated by comparing
the results obtained from sequential quadratic programming (SQP)
method with and without multi-start (MS) approach. Further, hybrid
optimization approach involving the combination of both DE and SQP
was also considered. For this case, the solution generated from DE is
being provided as an initial guess value for the SQP method. Thus, the
optimization methodology involves the application of either one of
DE, SQP, MS-SQP and DE-SQP methods. Table 3(a) summarizes the

Model validation precisely refers to the validation of developed code

for simulation and optimization models. Since appropriate data has
been available only for the MSF-BR system, the simulation model code
validation has been carried out using the data provided by Helal et al.
[3]. Based on the input simulation variable data set of [2790 m3/h
(2,790,000 kg/h), 1.0129, 18, 3.26 kW/m2 K, 3.443 kW/m2 K,
2.864 kW/m2 K, 7.62 m] for [WM, RH, NR, UB, UR, Uj, and LT], the obtained
results of dependent variables are [101.7 C, 32.41 C, 70,178 ppm,
135.5 m3/h (135,446 kg/h), 3662.8 m2, 2515.5 m2, 3355.4 m2,
9040 m3/h (9,040,000 kg/h)] for variables [T2, TBD, CMBD, WS, AB, AR, Aj,
and WMF] respectively. Precisely, the same values have been reported
in the literature [3] and hence the model code validation is inferred to
be successful.
The DE algorithm code has been tested for standard optimization
model test function such as Rosenbrock Banana Function. For the algorithm parameters [F, CR, NG, and NP] specied as [0.8, 0.8, 100 and
100], the DE provided optimal solution of [1.000, 1.000] with a standard
deviation of 106. The obtained optimal solution is in complete agreement with that available as the global optimal solution for Rosenbrook
(Banana) function [12]. Hence, the DE algorithm is inferred to be effective for the optimization of alternate MSF process congurations.
5. Results and discussions
5.1. Efcacy of the DE algorithm
Fig. 2 depicts the comparative performance of various optimization
methods to obtain optimal solutions for the MSF-M, MSF-OT and MSF-

C.S. Bandi et al. / Desalination 394 (2016) 3043

Table 3
(a) Parameters for various optimization techniques. (b) Initial independent variable set
data for MSF processes and alternative optimization methods.
(a)
Parameter

DE

SQP

MS-SQP

NP/no. of starting points

NG max/function evaluations
DE-step-size, F
CR
Termination criteria

350
400
500
1000
700
1000
0.9

0.9

NG max (DE) or 106 (SQP)

DE-SQP
500
1000
0.9
0.9

(b)
Optimization MSF-BR
method
[WM, RH, NR, UB, UR, Uj,
LT]

MSF-M
[WMSC, NR, UB, UR, LT]

MSF-OT
[WMF, NR, UB, UR,
LT]

SQP

[6168, 21, 3.4599,

3.5266, 7.4, 99.21,
69,944]
[6014, 23, 3.1432,
3.5221, 8.12, 103.23,
69,021]
[6546, 22, 3.3565,
3.4526, 8.85, 101.45,
69,584]
[6059, 25, 3.2452,
3.4411, 7.85, 102.56,
69,885]
[6356, 25, 3.5232,
3.6221, 8.25, 101.23,
69,981]
[5857, 25, 3.7, 3.6719,
9.74, 101.48,70,772]

[9543, 26,
3.4517, 3.6275,
8.12]
[9561, 28,
3.5454,
3.5514.7.65]
[9756, 27,
3.4624,
3.6454.7.69]
[9628, 26,
3.3515, 3.5554,
7.18]
[9643, 28,
3.3217, 3.3275,
7.98]
[9500, 23,
3.4108, 3.7398,
8.62]

[2449, 1.02, 19,

3.5124, 3.1937,
2.9223, 8.51]
[2767, 1.17, 18,
3.1673, 3.4619,
3.4454, 8.21]
[2439, 1.16,21, 3.3351,
3.2680, 3.5171, 7.12]

MS-SQP

DE-SQP

[2566, 1.03, 23, 3.

3513, 3.4259,
3.6637.7.51]
[2787, 1.07, 21,
3.4673, 3.4119,
3.4454, 7.61]
[2790, 1.05, 22,
3.3624, 3.3003,
3.0448, 7.62]

BR plant congurations. It has been evaluated that the optimal fresh

water production cost using DE, SQP, MS-SQP and DE-SQP methods
are 1.2251, 1.2856, 1.2785 and 1.2534 $/m3 for MSF-OT, 1.198, 1.2445,
1.22 and 1.2135 $/m3 for MSF-M and 1.0785, 1.1, 1.0852 and 1.0843
for MSF-BR respectively. Hence, it can be concluded that the DE provides
the lowest fresh water production cost for MSF-BR. Based on the obtained results, the optimal cost based ranking is MSF-BR b MSF-M b MSFOT*. For the MSF-BR plant, the efcacy of various methods for the optimization is in the order of DE N SQP N MS-SQP* N DE-SQP*. The solution
with asterisk (*) corresponds to solution with penalty (and hence
infeasible solution). An infeasible solution for the DE-SQP is due to
rounding of solutions generated from DE, which were then supplied to
SQP as initial guess.
A summary of the important independent and dependent variable
values obtained after optimization of MSF-BR, MSF-M and MSF-OT

35

using various methods is presented in Tables 4, 5 and 6 respectively.

In these methods, optimization methods with asterisk indicate the existence of penalties in the nal solutions obtained and hence these solutions cannot be recommended as appropriate design solutions for the
respective MSF process congurations. For several cases, UB inequality
constraint could not be resolved without penalty.
Further, the literature data for MSF-BR has also been presented in
Table 4 to reect upon the comparative optimality of obtained variable
values with those presented in the literature. As outlined in Table 4, in
comparison with the available literature data for the MSF-BR process,
the optimal solution obtained with DE refers to marginal combinations
of higher feed ow rate (2794.4 but not 2790.0 m3/h), higher specic
heat ratio (1.0499 but not 1.0001), lower tube length (7.62 m but not
7.92 m), higher overall heat transfer coefcient in brine heater section
(3.359 kW/m2 K but not 3.260 kW/m2 K), lower overall heat transfer
coefcient in recovery section (3.297 kW/m2 K but not 3.443 kW/
m2 K), higher overall heat transfer coefcient in brine heater
(3.042 kW/m2 K but not 2.864 kW/m2 K), higher number of recovery
stages (22 but not 18), lower steam ow rate (130.365 m3/h but not
135.446 m3/h), lower feed ow rate, cooling water and cost. It is further
interesting to note that the optimal concentration of rejected stream
leaving the heat rejection section is precisely the same for this case
and literature data. Compared to the literature reported optimal
cost of 1.104 $/m3 , the DE generated solutions for the MSF-BR
process to be 2.31% lower (1.0785 $/m3), MSF-M process to be
7.85% higher (1.198 $/m3) and MSF-OT process to be 9.88% higher
(1.2251 $/m3).
No relevant data was available in the same literature to compare the
solutions obtained for the MSF-M and MSF-OT process congurations. It
can be observed that marginal improvement in solutions can be obtained for the MSF-BR plant conguration in comparison with the literature
using the DE algorithm. For the MSF-BR process congurations, solutions with penalty were obtained using the DE-SQP method. For the
MSF-M process model, only DE provided solutions without penalty. Further, all optimization methods can be observed to provide solutions
with penalty for the MSF-OT congurations. This conveys the efcacy
of the DE algorithm to obtain solutions without penalty (feasible solutions) for both MSF-BR and MSF-M congurations. Hence, it is important to envisage that DE has potential to obtain high quality solutions
even for the MSF-M process. The best set of optimal decision variable
values are [WM, RH, LT, UB, UR, Uj and NR] for [2794.4 m3/h, 1.0499,
7.62 m, 3.359 kW/m2 K, 3.297 kW/m2 K, 3.042 kW/m2 K and 22].
For the MSF-BR, the optimal variable values corresponded to production cost, thermal performance, specic heat transfer area and
plant recovery of 1.0785 $/m3, 8.61, 0.009 m2/h/kg and 0.1287 respectively. These corresponded to an improvement of about 2.31, 3.90,
2.92 and 20.24% with respect to those reported in the literature [3]
(1.104 $/m3, 8.283, 0.008487 m2/h/kg and 0.107 respectively). For the
same case, the optimal feed ow rate and cooling water ow rate
were evaluated as 8720 m3/h and 5930 m3/h respectively, which correspond to a reduction of about 3.53% and 5.2% with respect to the best
known optimal values in the literature (9040 m3/h and 6250 m3/h
respectively). These results indicate marginal improvement in the
optimal solutions obtained and thereby convey the competence
and efcacy of the DE algorithm for desalination process design
and analysis.

5.2. Optimality of other dependent variables

Fig. 2. Bar chart depicting the performance of optimization methods for MSF desalination
processes.

Based on thermodynamics, heat transfer and process economics

based insights, several dependent variables have been dened to indicate upon the optimality of various MSF processes. These refer to specific heat transfer area, specic feed ow rate, specic cooling water rate
and overall plant recovery. Denitions and formulae to evaluate the
same are presented as follows:

36

C.S. Bandi et al. / Desalination 394 (2016) 3043

Table 4
Optimal independent and dependent variable values for the MSF-BR process and alternate optimization methods.

Independent variables

Dependent variables

Variable

Unit

DE

SQP

MS-SQP*

DE-SQP*

Literature [3]

WM
RH
LT
UB
UR
Uj
NR
T2
TBD
CMBD
WS
AB
AR
Aj
WMF
WMCW
WMBD
AT
Cost

m3/h

m
kW/m2 K
kW/m2 K
kW/m2 K

C
C
ppm
m3/h
m2
m2
m2
m3/h
m3/h
m3/h
m2
$/m3

2794.4
1.0499
7.62
3.359
3.297
3.042
22
102.00
32.35
70,178
130.365
3470.5
3297.1
3044.7
8721.0
5926.6
1672.4
85140.8
1.0785

2809.8
1.0001
7.92
3.362
3.161
3.114
19
102.03
32.38
67,092
139.860
3165.7
3137.9
2992.8
8732.6
6257.5
1353.1
74902.1
1.1000

2617.8
1.0339
7.98
3.7
3.294
3.687
20
102.03
32.34
67,093
129.900
3304.4
2588.7
2868.4
8856.5
6238.8
1495.8
63683.6
1.0852

2784.4
1.0149
7.61
3.254
3.436
2.858
18
101.50
32.34
70,038
135.175
3655.47
2510.47
3348.69
9021.9
6237.5
1662.4
58799.61
1.0843

2790.0
1.0129
7.62
3.260
3.443
2.864
18
101.70
32.41
70,178
135.446
3662.8
2515.5
3355.4
9040.0
6250.0
1668.0
59008.8
1.104

The specic heat transfer area is dened as the ratio of total heat
transfer area to the total amount of fresh water produced i.e.
Specific heat transfer area
total heat transfer area
A

total amount of fresh water produced W MD

total seawater intake fed to plant

W MF

total amount of fresh water produced W MD

The specic cooling water ow rate is dened as the ratio of total
cooling water ow rate of the plant to the total amount of fresh water
produced i.e.
Specific cooling water flow rate
total cooling water flow rate of plant W MCW
:

W MD
total amount of fresh water produced

Overall plant recovery

The specic ow rate is expressed as the ratio of total seawater fed to

plant to the total amount of fresh water produced i.e.
Specific feed flow rate

The overall plant recovery is dened as the percentage of total seawater intake that gets converted to the fresh water i.e.

total amount of fresh water produced

W MD
 100:
 100
W MF
total seawater intake fed to plant

Along with thermal performance, Table 7 summarizes the optimal

dependent variable values for various MSF processes. The thermal performance values of MSF-OT, MSF-M, MSF-BR and MSF-BR literature
data are evaluated to be 10.3457, 7.5918, 8.6066 and 8.2837 respectively. Fresh water production cost is lowest for MSF-BR plant (1.0785 $/
m3) and the corresponding water recovery is highest (12.28). For
these dependent variables, the DE approach used in this work provided
better results than those reported in literature [3] using the Newton
Raphson method (1.10 $/m3 and 10.7 respectively). The specic feed
ow rate value is also the lowest for MSF-BR case (7.77) and this indicates lower processing cost involved with the feed pre-treatment plant.
5.3. Contributions of various cost functions
Fig. 3(a), (b), (c) and (d) presents the pie charts that depict the percentage cost contributions of ICM, DCLIC, ICC, FCC and OMC to the total
cost for MSF processes. For the literature data, the pie chart depicting

Table 5
Optimal independent and dependent variable values for the MSF-M process and optimization methods.

Independent variables

Dependent variables

Variable

Unit

DE

SQP*

MS-SQP*

DE-SQP*

WMSC
NR
LT
UB
UR
T2
CMF
TBD
CMBD
WS
AB
AR
WMF
WMBD
AT
Cost

m3/h

m
kW/m2 K
kW/m2 K
C
ppm
C
ppm
m3/h
m2
m2
m3/h
m3/h
m2
$/m3

5852.7
26
8.32
3.699
3.671
101.48
70,772
36.52
70,803
147.790
3661.3
1964.7
9727.1
8605.1
54743.5
1.198

5475.1
22
9.99
3.626
3.665
100.28
56,342
37.18
56,399
171.429
3907.2
1992.9
9745.3
8623.3
47751.0
1.2445

5617.8
22
8.15
3.7
3.499
101.42
75,829
36.57
75,905
148.434
3496.5
2700.0
9755.3
8633.3
62896.5
1.22

5784.4
22
9.98
3.625
3.664
100.23
56,314
37.16
56,370
171.343
3905.2
1991.9
9740.4
8618.5
47705.2
1.2135

C.S. Bandi et al. / Desalination 394 (2016) 3043

37

Table 6
Optimal independent and dependent variable values for the MSF-OT process and optimization methods.

Independent variables

Dependent variables

Variable

Unit

DE*

SQP*

MS-SQP*

DE-SQP*

WMF
LT
UB
UR
NR
T2
TBD
CMBD
WS
AB
AR
WMBD
AT
Cost

m3/h
m
kW/m2 K
kW/m2 K

C
C
ppm
m3/h
m2
m2
m3/h
m2
$/m3

10043.2
10.18
3.7
3.456
23
101.48
31.4
48,035
153.450
3053.3
3530
1672.4
84243.3
1.2251

10061.9
8.0
3.410
3.468
21
103.6
31.39
47,624
153.490
3094.4
4397.0
1353.1
95431.4
1.2856

10072.1
11.8
3.70
3.498
21
103.6
31.4
47,623
158.450
3285.1
3596.9
1495.8
78820.0
1.2785

10056.9
11.79
3.691
3.495
21
103.51
31.37
47584.9
157.363
3249.8
3594.0
1662.4
78663.9
1.2534

the cost contributions of these components is presented in Fig. 3(d).

From Fig. 3(c) and (d), for the MSF-BR process conguration, it can be
observed that DCLIC is 2% lower for the DE based optimal solution in
comparison with that reported in the literature. For the same case, the
DE based optimal solution indicated the DCLIC, ICC, OMC and optimal
water production cost to be 5.44, 0.70, 22.40 M$/y and 1.0785 $/m3 respectively. Compared to the literature data [3], these values correspond
to a reduction of 12.71, 7.29, 3.22 and 2.31% respectively. Further, with
respect to the literature data, it can be also analyzed that the areas of
heat rejection and brine heater sections are respectively marginally
lower and marginally higher for the heat recovery section for the
MSF-BR process conguration.

5.4. Optimality of DE algorithm parameters

The effect of DE algorithm parameters (F and CR) on the solution
quality was investigated. According to Storn and Price [30], among F
and CR, DE is much more sensitive to F. For optimal DE algorithm performance, they further suggested that F, CR and NP can be set as F [0, 2],
CR [0, 1] and NP = 10 D, where D is dimensionally of the problem.
Typically, F and CR are specied as 0.9 and 0.8 respectively for engineering optimization problems. Thereby, algorithm optimality is critically
investigated for the optimality of NG and NP. Signicant tradeoffs
exist for the optimality of NG and NP. A very low value of NG and NP
may terminate the algorithm before it could reach the global optimal
and a very high value of NG and NP may take signicantly a long time
to achieve all solutions very close to the global optimal solution. Solution clustering phenomena are typically addressed as the criteria to
set optimal values of NP and NG. In this work, NP is varied from 25 to
340 and NG is varied from 50 to 2000 for the optimization of MSF
processes.
Fig. 4 panels (a) and (b) respectively summarize the variation of
total optimal objective function value (including penalty) and total
penalty function value with population size (NP = 35 to 340) and generation size. As shown, signicant penalties existed for lower combinations of NG and NP for the MSF-BR process. Solutions without penalty
have been achieved using a critical specication of NG and NP, over

Table 7
Optimality of thermodynamic, heat transfer and cost function variable values for MSF desalination processes.
Performance model

TPR

sA, m2/(kg/h)

sF

MSF-OT*
MSF-M
MSF-BR
MSF-BR literature [3]

10.34
7.59
8.61
8.28

0.005867
0.005014
0.008745
0.008497

8.95
8.66
7.77
8.06

sWCW

5.28
5.57

OPR
11.17
11.78
12.87
10.70

and above which solutions were found to exclusively cluster around

the global optimal domain. For all cases of NP, a critical value of NG =
200 has been evaluated to be relevant to yield optimal solutions without
penalty. Further, above an NG value of 800, almost all solutions
remained fairly constant, thus indicating that the solutions generated
are very close to the global optimal domain. Further, for few cases of
NP and generation size where few best solutions were obtained, additional investigations were carried out to evaluate upon the solution
quality in the context of the global optimal domain. For this purpose,
standard deviations were evaluated for the obtained solutions. The
standard deviation of the solutions obtained for a population and generation size of 270 and 800 respectively is 105 for the best 100 solutions.
For this case, the lowest optimal solution of 1.0785 $/m3 has been obtained. However, for the case of the population and generation size of
100 and 500 respectively, a standard deviation of 103 was obtained
for the best 100 solutions. The achievement of lower standard deviation
for lower combinations of NG and NP in comparison with the higher
combinations of NG and NP is expected, given the fact that higher NG
and NP combinations facilitate better search of the solution space. A
similar explanation could be provided for the results indicated in
Fig. 4(c)(f) for MSF-M and MSF-OT processes. Thus, the optimal DE algorithm parameter combinations for MSF-BR refer to F, CR, NG and NP
values of 0.9, 0.8, 800 and 270 respectively.
For the MSF-OT process conguration, Fig. 4 panels (c) and
(d) respectively depict the variation of optimal total objection function
value and optimal total penalty value with NP and generation size. The
obtained trends are similar to those obtained for MSF-BR case and the
solutions converged to optimal value for all populations (NPs) for a generation size of 800. The results reported for the MSF-OT system refer to
the existence of penalties for few constraints and the reported values
indicate solutions with lowest penalties. A careful analysis of the optimization results for MSF-OT indicated that brine velocity constraint
was the violated beyond the specied upper bound value (6 ft/s). Eventually, with higher brine velocity, the inequality constraint presented as
0:001 did not get satised, as UBcal is a function of brine
abs1 UUBcal
B
density, which is in turn a function of brine velocity. The penalties for
the MSF-OT have been successfully eliminated by targeting the following alterations in the parametric and design specications:
a) The brine velocity is xed as 6 ft/s for heat recovery section and the
inequality constraint has been specied to have a constraint parameter value of 102 but not 103. For such a scenario, the minimal
fresh production cost has been evaluated using the DE algorithm as
1.1249 $/m3.
b) The brine velocity is xed as 6 ft/s and the brine heater width is reduced to 30 ft from 35 ft. No additional changes have been carried
out for the abovementioned inequality constraint specication. For
such a scenario, solutions without penalty were obtained for the

38

C.S. Bandi et al. / Desalination 394 (2016) 3043

Fig. 3. Pie-charts representing the cost contributions of various cost functions to optimal freshwater production cost; (a) MSF-OT, (b) MSF-M, (c) MSF-BR and (d) MSF-BR literature
data [3].

MSF-OT system to indicate the optimal fresh water production cost

of 1.1154 $/m3.

Fig. 4 panels (e) and (f) respectively illustrate the variation of total
optimal objective function value (inclusive of penalties) and total optimal penalty function values for various cases of NP and generation
values. The obtained trends are similar to those presented for MSF-BR
case. Also, it can be observed that for MSF-M, solutions without penalty
(feasible solutions) have been obtained using the DE algorithm. The
MSF-OT conguration provided infeasible solutions among the three
different process congurations. Using DE, the optimal water production cost for the MSF-M process is 1.1980 $/m3 respectively.
5.5. Sensitivity analysis
5.5.1. Effect of feed concentration
Sensitivity analysis involves the evaluation the optimal freshwater
production cost critical dependence on with various operating parameters of the MSF process. Typically, feed concentration is varied from
20,00050,000 ppm for MSF processes [3]. Varying feed concentration
in this range, the optimal freshwater production cost was evaluated
using DE for MSF processes. Fig. 5(a) summarizes the results obtained
for the feed concentration effect on optimal fresh water production
cost. It can be observed that the optimal water production cost increased from 1.2125 to 1.2455, from 1.1876 to 1.2132, from 1.0656 to
1.1051 and from 1.09 to 1.13 $/m3 respectively for MSF-OT, MSF-M,
MSF-BR and MSF-BR literature processes. From a variation in feed concentration 20,000 ppm to 40,000 ppm, the optimal water production
cost varied from 1.0656 to 1.0708 for the MSF-BR process which afrms
that the cost remained fairly constant. From 40,00050,000 ppm variation in feed concentration, water product cost increased linearly from
1.0708 to 1.1051 for the MSF-BR process. The slope of the graph is
about 3.433 10 6. The insensitivity of the water production cost

with feed concentration up to 40,000 ppm is due to the insignicant effect of feed concentration in inuencing the product ow rates, concentrations and temperature. This might not be the case for reverse osmosis
process where feed concentration will have a signicant effect on the
water production cost. Above 40,000 ppm, the feed concentration can
be observed to have a signicant effect on the water production cost.
The DE based optimal cost proles lowered by 2.232.65% than
those reported in the literature for the variation in feed seawater concentration. In comparison with the literature data, it can be observed
that the lowest cost trends have been obtained for the MSF-BR process.
This once again conrms the efcacy of DE to obtain high quality solutions for MSF process optimization.
5.5.2. Thermal performance
The thermal performance is dened as the ratio of total amount of
fresh water produced to total steam intake of the MSF process:
Thermal performance

total amount of fresh water produced W MD

:

W MS
total steam intake fed to plant

In general, the thermal performance ratio varies from 6 to 12 for the

MSF desalination system and below 1 for the single stage ash desalination system [10]. Similarly, the top brine temperature (TBT) is varied
from 90 to 100 C for a variation in total number and thermodynamic
loss range of 1829 and 0.52 C [10].
The variation of thermal performance with TBT for MSF processes is
presented in Fig. 5(b). For a variation in TBT from 90 to 110 C, the thermal performance ratio varied from 6.66 to 7.31, from 6.75 to 7.59, from
7.26 to 8.61 and from 7.02 to 8.28 for MSF-OT, MSF-M, MSF-BR and
MSF-BR literature data. It can be observed that the thermal performance
is marginally sensitive with TBT for MSF-OT and MSF-M processes but
not MSF-BR. Further, it can be observed that the thermal performance

C.S. Bandi et al. / Desalination 394 (2016) 3043

39

Fig. 4. Effect of DE algorithm parameters (NG and NP) on the solution optimality for various MSF processes (a) total optimal objective function value for MSF-BR, (b) total penalty
function value for MSF-BR, (c) total optimal objective function value for MSF-M, (d) total penalty function value for MSF-M, (e) total optimal objective function value for MSF-OT and
(f) total penalty function value for MSF-OT.

increased linearly with increasing TBT and the corresponding slope

values are 3.2 102, 4.0 102, 6.7 102, and 6.4 102 respectively for the said process sequence. Also, in comparison with the literature data, it is apparent from the gure that the highest thermal
performance values were obtained for the MSF-BR process using the
DE algorithm. The thermal performance of the MSF system has been

varied from 6 to 9, where the DE based optimal costs have been evaluated to be 3.423.99% lower than those reported in the literature.
5.5.3. Chemical cost multiplier
Typically, cost multipliers are varied from 0.25 to 2 for MSF desalination processes [3]. Fig. 5(c) presents the variation of optimal water

40

C.S. Bandi et al. / Desalination 394 (2016) 3043

Fig. 5. Sensitivity of MSF processes with respect to various process and operating parameters; (a) feed concentration, (b) TBT on TPR, (c) chemical cost multiplier, (d) steam cost multiplier.
(e) Labor cost multiplier, (f) power cost multiplier, (g) spares cost multiplier, (h) TBT on cost (+[3]).

C.S. Bandi et al. / Desalination 394 (2016) 3043

production cost as a function of chemical cost multiplier for MSF processes. It can be observed that the water production cost increased
non-linearly from 1.1345 to 1.2958, from 1.0866 to 1.12798, from
1.0015 to 1.1365 and from 1.022 to 1.161 $/m3 with increasing chemical
cost multiplier (0.252) for MSF-OT, MSF-M, MSF-BR and MSF-BR literature data. As it is apparent in other sensitivity analyses, the lowest
water production cost trends have been obtained in this work in comparison with the existing data trends of the MSF-BR process in the literature. The freshwater production cost varied linearly with chemical cost
multiplier in the range of 0.252. For such a case, the corresponding
slopes of the trend lines are 0.09, 0.11, 0.071, and 0.072. These slope
values are indicative towards water production cost sensitivity with respect to chemical cost multiplier. Thus the water cost ($/m3) will increase by 9 /m3, 11 /m3, 7.1 /m3, and 7.2 /m3 respectively for a
unit increase in chemical cost multiplier for MSF-OT, MSF-M, MSF-BR
and MSF-BR literature data [3]. For a variation in the cost multiplier
from 0.25 to 0.5, the corresponding optimal water cost varied nonlinearly from 1.1348 to 1.1752, from 1.0866 to 1.1271, from 1.0015 to
1.0495 and from 1.022 to 1.076 $/m3. A further increase in the chemical
cost multiplier from 0.5 to 2.0 enabled a linear enhancement for the optimal water cost from 1.1752 to 1.2958, from 1.1271 to 1.2798, from
1.0495 to 1.1365 and from 1.076 to 1.161 $/m3 for the said sequence
of processes. Compared to the literature, the obtained optimal cost is
22.46% better for variation in the chemical cost multiplier.
5.5.4. Steam cost multiplier
For the steam cost multiplier sensitivity analysis, the steam cost was
varied from 0.00104 to 0.00832 $/kg of steam (corresponding to a variation of steam cost multiplier from 0.25 to 2.0) at a constant TBT value of
110 C [3]. The obtained sensitivity analysis based cost trends with respect to steam cost multiplier are presented in Fig. 5(d). As shown, for
a variation in steam cost multiplier from 0.25 to 2.0, the minimal
water cost varied from 0.8656 to 1.7123, from 0.8202 to 1.687, from
0.7284 to 1.4876 and from 0.739 to 1.5212 $/m3 for MSF-OT, MSF-M,
MSF-BR and MSF-BR literature data [3]. The obtained cost trends with
respect to steam multiplier indicate a linear variation in water cost
with steam multiplier. Such a trend is expected, given the fact that the
MSF is a thermal process and its performance is a strong function of
steam as a heat source for ash operation. The sensitivity parameters
can be obtained from the slope of the obtained data trends. These values
have been evaluated correspondingly as 0.48, 0.49, 0.43 and 0.44. Thus,
the water cost can be evaluated to increase by 48 /m3, 49 /m3, 43 /m3
and 44 /m3 respectively for a unit increase in steam cost for the said sequence of processes. The obtained simulation based trends indicate that
MSF-BR provides the lowest water cost trends which are placed marginally below the cost trends reported in the literature [3]. For a variation in
steam cost multiplier from 0.2 to 2.0, the DE based optimal cost is 1.43
2.21% lower than that reported in the literature.
5.5.5. Labor cost multiplier
Fig. 5(e) shows that the optimal water production cost evaluated for
MSF processes is sensitive with respect to labor cost multiplier for all
processes. As shown, for a variation in labor cost multiplier from 0.25
to 2.0, the minimal water cost varied from 1.1395 to 1.3396, from
1.209 to 1.3008, from 1.0014 to 1.1813 and from 1.0239 to 1.2001 $/
m3 for MSF-OT, MSF-M, MSF-BR and MSF-BR literature data [3].
Hence, signicantly higher cost trends can be observed for MSF-OT
and MSF-M processes but not for MSF-BR and MSF-BR literature data.
The obtained cost trends with respect to labor cost multiplier indicate
that the variations are linear with slopes of 0.11, 0.10, 0.10 and 0.10
for MSF-OT, MSF-M, MSF-BR and MSF-BR literature [3] respectively.
Thus the water cost ($/m3) will increase by 11 /m3, 10.3 /m3, 10 /
m3, and 10 /m3 respectively for a unit increase in labor cost multiplier
for the said sequence of processes. The lowest data trends refer to the
data obtained with DE for the MSF-BR process. For a variation in labor

41

cost multiplier from 0.2 to 2.0, the DE based optimal cost is 1.572.2%
better than that reported in the literature.
5.5.6. Power cost multiplier
For a variation in power cost multiplier from 0.25 to 2.0,
Fig. 5(f) presents the variation of optimal water production cost trends
for various MSF desalination processes. The observed trends are similar
to those obtained for labor cost multiplier. As shown, for a variation in
power cost multiplier from 0.25 to 2.0, the minimal water cost varied
from 1.1286 to 1.3589, from 1.1139 to 1.3101, from 0.9915 to 1.1921
and from 1.02 to 1.213 $/m3 respectively for MSF-OT, MSF-M, MSF-BR
and MSF-BR literature data [3]. The corresponding slopes of the obtained cost trends are 0.13, 0.11, 0.11 and 0.11. Thus the water cost ($/m3)
will increase by 12.9 /m3, 11.2 /m3, 10.9 /m3, and 10.9 /m3 respectively for a unit increase in labor cost multiplier for the said sequence
of processes. The lowest cost trend corresponds to that obtained with
MSF-BR and the DE algorithm in this work. For a variation in power
cost multiplier from 0.2 to 2.0, the DE based optimal cost is 1.72
2.79% better than that reported in the literature.
5.5.7. Spares cost multiplier
Fig. 5(g) illustrates the variation of minimal water production cost as
a function of spares cost multiplier for alternate MSF processes. As
shown, for the variation in spares cost multiplier from 0.25 to 2.0, the
minimal water cost varied from 1.1456 to 1.3102, from 1.1348 to
1.2823, from 1.0203 to 1.1628 and from 1.0375 to 1.1825 $/m3 for
MSF-OT, MSF-M, MSF-BR and MSF-BR literature data [3]. The cost trends
are similar to those obtained for labor and power cost multipliers. The
slopes of the linearized trends are 0.09, 0.08, 0.08 and 0.08 respectively
for the said sequence of processes. Thereby, the sensitivity of the spares
cost multiplier has been evaluated in terms of an increase by 9.2 /m3,
8.4 /m3, 8.2 /m3, and 8.2 /m3 respectively for a unit increase in
labor cost multiplier for the said sequence of processes. For a variation
in spares cost multiplier, the DE based optimal cost is 1.66% lower
than that reported in the literature.
5.5.8. Effect of top brine temperature
Fig. 5(h) presents the variation of optimal water production cost
with variation in top brine temperature for various cases. As shown,
for a variation in TBT from 90 to 110 C, the costs varied linearly from
1.1051 to 1.2251 and from 1.0845 to 1.1980 $/m3 for MSF-OT and
MSF-M processes. However, for the MSF-BR processes (reported in
this work and in literature), up to a temperature of 100 C, the optimal
water production cost was not at all affected with variation in TBT.
Above 100 C, the optimal water production cost increased with increasing TBT. Based on these increasing trends, the slopes of the various
plots have been evaluated (5.9 103, 5.6 10 3, 2.5 103, and
2.8 103 respectively for the said sequence of processes). Thus, for a
unit increase in top brine temperature, the water cost will increase by
0.59 /m3, 0.56 /m3, 0.25 /m3, and 0.28 /m3 for the said sequence
of processes. Overall, the lowest data trends have been obtained for
the MSF-BR and DE algorithm case. For a variation in TBT from 95 to
110 C, the DE based optimal cost is 1.66% lower than that reported in
the literature.
In summary, the DE based sensitivity analysis of MSF-BR enables
one to infer that the slope based ranking of various process and
operating parameters is as per the following order: steam cost multiplier (43 /m3) N labor cost multiplier (10 /m3) N power cost multiplier
(8.2 /m3) N spares cost multiplier (8.2 /m3) N chemical cost multiplier
(7.1 /m3) N TBT (0.25 /m3) N feed concentration (fairly constant). In
other words, the optimal freshwater production cost for the MSF processes is highly sensitive to steam cost multiplier, marginally sensitive with all
other process parameters but not feed concentration and TBT. Since the
MSF process is highly energy intensive, the highest sensitivity of the
MSF-BR process optimal cost with steam multiplier is expected.

42

C.S. Bandi et al. / Desalination 394 (2016) 3043

6. Conclusions
Based on the global optimization approach involving differential
evolution algorithm, this work provided signicant insights and inferences with respect to the comparative assessment of alternate MSF processes, non-deterministic/deterministic optimization methods and
pertinent sensitivity analysis. The modeling approach adopted in this
work might refer to design solutions under stringent uncertainty,
given the inability to generate feasible solutions with methods other
than DE. The following conclusions are applicable from the insights deduced in this work.
Firstly, DE has been proven to be effective to generate feasible
optimal design variable values for MSF-BR, MSF-M but not MSF-OT processes. Compared to the literature optimal value, DE provided a reduction of about 2.31% in the optimal freshwater production cost. This is
due to the identication of better optimal decision variable value set
of [2794.4 m3/h, 1.0499, 7.62 m, 3.359 kW/m2 K, 3.297 kW/m2 K,
3.042 kW/m2 K and 22] for variable set [WM, RH, LT, UB, UR, Uj and NR] respectively where the optimal freshwater production cost corresponds
to 1.0785 $/m3. Secondly, the deterministic optimization algorithms
such as SQP, MS-SQP and DE-SQP could not provide better solution
than the DE. This is primarily due to the dependence of the optimal variable value set and objective function on the initial guess values. Thus,
compared to other optimization methods, DE would provide better initialization strategies and is expected to serve better for problems with
greater complexity in terms of decision variables.
Another important insight that has been deduced in this work is that
the MATLAB based optimization toolbox uses default optimization algorithm parameters and they cannot be as such used for MSF optimization
problems studied in this work. Thirdly, the sensitivity analysis afrmed
that DE based analysis provided 1.413.99% better proles than those
available in the literature. While such improvement could be regarded
to be optimal, it is important to note that the freshwater production
cost related improvement is signicant, given the fact that optimization
studies that allow even 1% reduction in water production cost could
turn out in terms of a huge amount of savings. Fourthly, inequality constraint resolution appears to be better tackled by DE than any other optimization method. A further resolution of the generated solutions has
also been demonstrated in this work i.e., to alter certain design parameter value for chamber width. Fifthly, the chosen literature might be relatively old in the existing state-of-the-art, but the trends obtained in
this work appear to be generic to afrm upon the efcacy of DE as the
most versatile optimization method to yield feasible solutions under
strong conditions of uncertainty. In summary, it is inferred that DE
based optimization is highly effective to obtain feasible global optimization solutions in conjunction with SQP, MS-SQP and DE-SQP. It is anticipated that DE would be able to provide condence in the solutions
generated with complex and hybrid process congurations involving
MSF process congurations. This will be addressed in subsequent research articles.
Nomenclature
Abbreviations
CR
cross over ratio
DE
differential evolution
ED
electrodialysis
F
mutation factor
DICOPT++ DIscrete and Continuous OPTimizer
GA
Genetic algorithm
GAMS
general algebraic modeling system
GOR
gained output ratio
GRG
generalized reduced gradient
ID
inside diameter of condenser tubes, m
IDA
International Desalination Association

MEE
MS-SQP
MSF
MSF-BR
MSF-M
MSF-OT
NG
NLP
NP
OF
OPR
ppm
RO
sA
sF
sWCW
SQP
TA
TBT
TDM
TDS
TPR
USA
VC

multi-effect evaporator
multistart-sequential quadratic programming
multi-stage ash
brine recycle (BR) multistage ash system (MSF)
brine-mixing (M) multistage ash system (MSF)
once through (OT) multistage ash system (MSF)
maximum number of generations
non-linear programming
population size
objective function
overall plant recovery
parts per million
reverse osmosis
specic heat transfer area
specic feed ow rate
specic cooling feed ow rate
sequential quadratic programming
Thomas algorithm
top brine temperature
tridiagonal matrix
total dissolved solids
thermal performance
United States of America
vapor-compression

Symbols

a
AB
Aj
AR
AT
CDCC
CDCLIC
CIC
CICC
Clast
CMBD
CMF
CMR
CMSMF
COMC
Cp
CPR
Cpj
ID
LT
M1
Nj
NR
OD
RH
S12
sWCW
T1
T2
T3
T4

coefcient to account for using average latent heat of

vaporization
heat transfer area of brine heater, m2
heat transfer area of the rejection section, m2
heat transfer area of recovery section, m2
total heat transfer area of the MSF process, m2
direct capital investment, $
direct capital less intake investment, $
annual plant intake-outfall cost, $
indirect capital investment, $/y
concentration of brine stream from last stage of the heat rejection section, ppm
concentration of reject stream leaving the heat rejection section, ppm
feed (seawater) concentration, ppm
concentration of recycle stream (from splitter to heat recovery section), ppm
concentration of feed stream to the MSF-M process, ppm
operating and maintenance cost, $/y
heat capacity, kJ/kgK
avg. specic heat capacity, recovery section, kJ/kgK
avg. specic heat capacity, rejection section, kJ/kgK
inside diameter of condenser tubes, m
tube length, m
mixer 1
number of rejection stages
number of stages in heat recovery stages
outside diameter of condenser tubes, m
specic heat ratio (WMRCpR/WMFCpj)
splitter 12 respectively
specic cooling water ow rate
temperature of reject coolant stream in the MSF-BR process, C
temperature of brine stream entering the brine heater in the
MSF process, C
top brine temperature, oC
temperature of brine stream leaving last stage in the heat
recovery section of the MSF-BR process, C

C.S. Bandi et al. / Desalination 394 (2016) 3043

T5

temperature of brine stream leaving last stage in the heat

rejection section of the MSF-BR process, C
temperature of recycle stream in the MSF-BR process, C
T6
temperature of reject stream in the MSF-BR process, C
TBD
temperature of distillate product stream in the MSF process,
TD
C
steam temperature, C
TS
seawater temperature/feed temperature, C
TSea
overall heat transfer coefcient in brine heater, kW/m2 K
UB
overall heat transfer coefcient in the rejection section,
Uj
kW/m2 K
overall heat transfer coefcient in recovery section, kW/m2 K
UR
calculated overall heat transfer coefcient in brine heater,
UBcal
kW/m2 K
calculated overall heat transfer coefcient in the rejection
Ujcal
section, kW/m2 K
calculated overall heat transfer coefcient in recovery section,
URcal
kW/m2 K
brine velocity in brine heater, m/s
VB
brine velocity in the rejection section, m/s
Vj
brine velocity in recovery section, m/s
VR
ow rate of ashing brine stream leaving the last stage of the
Wlast
heat rejection section in the MSF-M and MSF processes, m3/h
ow rate of makeup stream in the MSF process, m3/h
WM
ow rate of rejected stream in the MSF process, m3/h
WMBD
ow rate of reject coolant stream in the MSF process, m3/h
WMCW
ow rate of total potable water product stream in the MSF
WMD
process, m3/h
feed ow rate in the MSF process, m3/h
WMF
feed ow rate of seawater (before mixing) of MSF-M, m3/h
WMSC
WMSMF feed ow rate of MSF-M, m3/h
ow rate of recycle stream in MSF-M and MSF-BR processes,
WMR
m3/h
WMRWBD ow rate of brine stream leaving mixer M1 and entering splitter S2 in the MSF-BR process, m3/h
ow rate of steam fed to brine heater, m3/h
WS
yearly capacity, m3/y
WY
Greek symbols

R
j

B
W

1/100th $ (cent)
average boiling point rise, heat recovery section, C
average boiling point rise, heat rejection section, C
average latent heat of vaporization, kJ/kg
average brine density, kg/m3
average pure water density, kg/m3
objective function (cost), $/m3

Appendix A. Supplementary data

Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.desal.2016.04.012.
References
[1] A.K. Coleman, Optimization of a single effect, multi-stage ash distillation desalination system, Desalination 9 (1971) 315331.
[2] A.M. Helal, M.S. Medani, M.A. Soliman, J.R. Flower, A tridiagonal matrix model for
multi-stage ash desalination plants, Comput. Chem. Eng. 10 (1986) 327342.
[3] A.M. Helal, A.M. El-Nashar, E. Al-Katheeri, S. Al-Malek, Optimal design of hybrid RO/
MSF desalination plants part I: modeling and algorithms, Desalination 154 (2003)
4366.
[4] A.M. Helal, A.M. El-Nashar, E. Al-Katheeri, S. Al-Malek, Optimal design of hybrid RO/
MSF desalination plants part II: results and discussion, Desalination 160 (2004)
1327.
[5] A.M. Helal, A.M. El-Nashar, E. Al-Katheeri, S. Al-Malek, Optimal design of hybrid RO/
MSF desalination plants part III: sensitivity analysis, Desalination 169 (2004) 4360.

43

[6] E.A.M. Hawaidi, I.M. Mujtaba, Simulation and optimization of MSF desalination process for xed freshwater demand: impact of brine heater fouling, Chem. Eng. J. 165
(2010) 545553.
[7] E. Ali, Understanding the operation of industrial MSF plants part II: optimization and
dynamic analysis, Desalination 143 (2002) 7391.
[8] F. Mjallia, N. Abdel-Jabbar, H. Qiblawey, H. Ettouney, Neural and genetic based techniques for solving the MSF model as opposed to conventional numerical methods,
17th European Symposium on Computer Aided Process Engineering. ESCAPE17,
2007.
[9] H. El-Dessouky, S. Bingulac, Solving equations simulating the steady-state behavior
of the multi-stage ash desalination process, Desalination 107 (1996) 171193.
[10] H.T. El-Dessouky, H.M. Ettouney, Fundamentals of Salt Water Desalination, Elsevier
Science, USA, 2002.
[11] H. El-Dessouky, I. Alatiqi, H. Ettouney, Process synthesis: the multi-stage ash desalination system, Desalination 115 (1998) 155179.
[12] H.H. Rosenbrock, An automatic method for nding the greatest or least value of a
function, Comput. J. 3 (1960) (17518).
[13] H.I. Shaheen, G.I. Rashed, S.J. Cheng, Optimal location and parameter setting of UPFC
for enhancing power system security based on differential evolution algorithm, Int.
J. Electr. Power Energy Syst. 33 (2011) 94105.
[14] I.S. Al-Mutaz, M.A. Soliman, Simulation of MSF desalination plantsDesalination 74
(1989) 317326.
[15] IDA, The 19th IDA Worldwide Desalting Plant Inventory, International desalination
association, Topseld, MA, USA, 2006.
[16] J.H. Beamer, D.J. Wilde, The simulation and optimization of a single effect multistage ash desalination plant, Desalination 9 (1971) 259275.
[17] J.M. Ramirez, J.M. Gonzalez, T.O. Ruben, An investigation about the impact of the
optimal reactive power dispatch solved by DE, Int. J. Electr. Power Energy Syst. 33
(2011) 236244.
[18] K.A. Al-shayji, S. Al-wadyei, A. Elkamel, Modelling and optimization of a multistage
ash desalination process, Eng. Optim. 37 (2005) 591607.
[19] M.A. Mandil, E.E. Abdel Ghafour, Optimization of multi-stage ash evaporation
plants, Chem. Eng. Sci. 25 (1970) 611621.
[20] M. Abduljawad, U. Ezzeghni, Optimization of Tajoura MSF desalination plant, Desalination 254 (2010) 2328.
[21] M.C. Marcovecchio, S.F. Mussati, N.J. Scenna, P.A. Aguirre, in: M. Schorr (Ed.),
Optimization of Hybrid Desalination Processes Including Multi Stage Flash and
Reverse Osmosis Systems 2011, pp. 312334, http://dx.doi.org/10.5772/15048
(CC BY-NC-SA).
[22] M.C. Marcovecchio, P.A. Aguirre, N.J. Scenna, Global optimal design of reverse osmosis networks for seawater desalination: modelling and algorithm, Desalination 184
(2005) 259271.
[23] M.G. Marcovecchio, S.F. Mussati, P.A. Aguirre, N.J. Scenna, Optimization of hybrid desalination processes including multi stage ash and reverse osmosis systems,
Desalination 182 (2005) 111122.
[24] M.H.K. Manesh, H. Janalizadeh, A.M.B. Marigorta, M. Amidpour, M.H. Hamedi, Optimal design of integrated total site utility-multistage ash desalination plant, Desalin.
Water Treat. 52 (2014) 12871298.
[25] M.J. Tanvir, I.M. Mujtaba, Optimisation of design and operation of MSF desalination
process using MINLP technique in gPROMS, Desalination 222 (2008) 419430.
[26] M. Rossol, A. Beltramim, M. Mazzotti, M. Morbidelli, Modeling multistage ash
desalination plants, Desalination 108 (1996) 365374.
[27] MATLAB and Optimization Toolbox Release, 2013The MathWorks, Inc., Natick,
Massachusetts, United States, 2013.
[28] N.M. Abdel-Jabbar, H.M. Qiblawey, F.S. Mjalli, H. Ettouney, Simulation of large capacity MSF brine circulation plants, Desalination 204 (2007) 501514.
[29] P.J. Thomas, S. Bhattacharyya, A. Petra, G.P. Rao, Steady state and dynamic simulation of multi-stage ash desalination plants: a case study, Comput. Chem. Eng. 22
(1998) 15151529.
[30] R. Storn, K. Price, Differential evolutiona simple and efcient adaptive scheme for
global optimization over continuous spaces, ICSI Technical Report, No. TR-95-012
1995, pp. 15.
[31] S.A. Abdul-Wahab, J. Abdo, Optimization of multistage ash desalination process by
using a two-level factorial design, Appl. Therm. Eng. 27 (2007) 413421.
[32] S.G.N. Kumar, A.K. Mahendra, A. Sanyal, G. Gouthaman, Genetic algorithm-based optimization of multi-stage ash desalination plant, Desalin. Water Treat. 1 (2009)
86106.
[33] S. Mussati, P. Aguirre, N.J. Scenna, Optimal MSF plant design, Desalination 138
(2001) 341347.
[34] S. Tayyebi, M. Alishiri, The control of MSF desalination plants based on inverse
model control by neural network, Desalination 333 (1) (2014) 92100 (15).
[35] T.E. Hinkebein, M.K. Price, Progress with the desalination and water purication
technologies US roadmap, Desalination 182 (2005) 1928.
[36] M. Valipour, Determining possible optimal values of required ow, nozzle diameter,
and wetted area for linear traveling laterals, Int. J. Eng. 1 (1) (2012) 3743.
[37] M. Valipour, M.E. Banihabib, S.M.R. Behbahani, Monthly inow forecasting using
autoregressive articial neural network, J. Appl. Sci. 12 (20) (2012) 21392147.
[38] M. Valipour, M.E. Banihabib, S.M.R., Comparison of the ARMA, ARIMA, and the
autoregressive articial neural network models in forecasting the monthly inow
of Dez dam reservoir, J. Hydrol. 476 (2013) 433441.
[39] M. Valipour, A.A. Montazar, Optimize of all effective inltration parameters in
furrow irrigation using visual basic and genetic algorithm programming, Aust. J.
Basic Appl. Sci. 6 (6) (2012) 132.