Journal of Petroleum Science and Engineering 119 (2014) 2835

Contents lists available at ScienceDirect

Journal of Petroleum Science and Engineering

journal homepage: www.elsevier.com/locate/petrol

Application of evolutionary computational approach in design of

horizontal three-phase gravity separators
Mehdi Mostafaiyan a, Mohammad Reza Saeb b,n, Alireza Emami Alorizi c, Maysam Farahani c
a

Department of Polymer Engineering and Color Technology, Amirkabir University of Technology, P.O. Box 15875-4411, Tehran, Iran
Department of Resin and Additives, Institute for Color Science and Technology, P.O. Box 16765-654, Tehran, Iran
c
Department of Polymer Engineering, Mahshahr Branch, Islamic Azad University, Mahshahr, Iran
b

art ic l e i nf o

a b s t r a c t

Article history:
Received 6 August 2013
Accepted 5 April 2014
Available online 26 April 2014

We introduce in this work an evolutionary computational algorithm with sophisticated procedures for
optimization of three-phase gravity separators. A model horizontal gas/oil/water separator, considered
as target design, having well-dened length, diameter, effective length, bafe height, and water channel
length, is hereby presented that features optimal design corresponding to minimum volume. The
developed computer code is executed to monitor and update evolution in the volume of model separator
towards target design. In this way, the implemented evolutionary algorithm manipulates volume
genotype to create new alternatives by recombination and mutation of previous generations. This ended
in near-to-optimal solutions irrespective of the properties of the rst generation adapted to the model
separator. The outputs from this theoretical investigation provide a broader image on the level of
separation in gravity separators that can be utilized for actual design.
Published by Elsevier B.V.

Keywords:
gravity separator
optimization
multiphase ow
target design
computational method

1. Introduction
Identication of phase separation in multi-component systems
was the subject of different investigations in the past, so that the
industry and academia were alike keen on developing new
techniques to achieve an optimal separation level under specied
conditions. In this regard, varieties of strategies have ever examined to gain desired extent of separation in different processes. A
major part of such investigations is aimed to separate oil from
water or other aqueous uids (Auem et al., 2001; Jaworski and
Meng, 2009; Kang et al., 2012).
The gravity separators are obviously the most important
type among this family, which are widely served to sever immiscible phases with different densities. The literature provides
multifarious reports on the application of gravity separators
(Zeevalkink and Brunsmann, 1983; Fossen et al., 2006; Oh et al.,
2012; Bazin et al., 2012). In this way, the kinetics of separation of
model oil-in-water emulsions under enhanced gravity is considered as a key enabling interface evaluation (Krebs et al., 2012). In a
gravity separator, the dispersed droplets of the dense component
move downward through the continuous light phase to join the

Corresponding author.
E-mail address: saeb-mr@icrc.ac.ir (M.R. Saeb).

http://dx.doi.org/10.1016/j.petrol.2014.04.003
0920-4105/Published by Elsevier B.V.

heavy continuous stream, while an inverse situation is the case for

the light phase (API Spec. 12J, 1989; Ludwig, 1994; Datta, 2008).
Despite the fact that separation mechanism in a gravity separator seems quite simple, in reality the design of such apparatus
confronts one with obvious difculties. The complexity of the
design procedure arises from the fact that there are varieties of
parameters, which affect the performance of separator, namely the
separator length and its diameter, the bafe position, bafe height,
residence times, surge times, and nally the extent of the separation (Walas, 1998; Smith, 2005). Thus, proposing a proper design
via conventional methods needs a great deal of experience and
barely leads to a unique response. The efciency of gas cyclone
separators has been intensively considered through modeling and
optimizing by using different mathematical approaches such as
genetic and neural network algorithms (Rosa et al., 2001; Elsayed
and Lacor, 2012, 2010). In addition, there exist a number of papers
on dealing with optimization and characterization of gravity
separators in terms of ow characteristics (Auem et al., 2001;
Zhang et al., 2007; Plasencia et al., 2013).
Normally, one can design a gravity separator as horizontal or
vertical vessel depending on the platform requirements and
practical limitations. Horizontal separators provide larger liquid
surface areas compared to vertical types leading to better settling
and gas breakout. Besides, the longer distance from the input to
outputs of the vessel in horizontal arrangement improves the
efciency of separation (Skeie and Halstensen, 2010). From a

M. Mostafaiyan et al. / Journal of Petroleum Science and Engineering 119 (2014) 2835

Nomenclature
Le
Lt
Lw
Desp
Lw/Le
Vf
Dp
h
l
c
g
g
c
Re
tres,w
tsurge,w

separator effective length (m)

separator total length (m)
water channel length (m)
seperator diameter (m)
bafe position (dimensionless)
dispersed phase droplet velocity (m/s)
droplet diameter of dispersed phase (mm)
heavy phase density (kg/m3)
light phase density (kg/m3)
continuous phase density (kg/m3)
gas phase density (kg/m3)
acceleration rate (m/s2)
continuous phase viscosity (mPa s)
Reynolds number (dimensionless)
residence time for water(min)
surge time for water (min)

practical standpoint, oil removal efciency is strongly dependent

on the conguration of internal devices and the layout. The
minimum required volume of the owing gravity separators is
mostly governed by the residence time (Lopez-Vazquez and Fall,
2004). The design of such vessels with minimum volume is always
associated with a trial and error process and strongly dependent to
the skill of designer. Accordingly, the lack of information would
necessitate optimizing the size of gravity separators varying their
congurational parameters. In this regard, genetic algorithm might
be a potential tool to get started in estimating the variation
interval for changing parameters to meet a near-to-optimal solution. This benecial method takes genotype variables, instead of
variables themselves, by seeking the population of solutions based
on probabilistic principles. Interestingly, the global optimum is
achievable whether the optimization problem is discrete or nonlinear (Michalewicz, 1996). However, there are a few reports, to
our knowledge, on the application and implementation of computational methodologies in design of gravity separators.
The present work aims to introduce an optimal design based on
evolutionary computational strategy that ultimately approaches a
unique solution. In this way, the efciency of a model horizontal
gas/oil/water separator is investigated by altering the design
criteria. To make a better sense of separation level, an optimal
design corresponding to the case that the volume of separator
takes the minimum value is reported. In this regard, a computer
code is created that takes an initial design criteria as the rst
generation of separator design procedure and propose a situation
in which next generation is sampled from previous genotypes by
recombination and mutation with respect to the population of
previous generations. The evolutionary approach implemented in
this work allows searching in a broad database of solutions leading
to a unique solution that is applicable to an industrial plant. This
algorithm is exible in application with regard to the target
design, by which a separator giving minimum weight or design
cost.

tres,o
tsurge,o
tr
ta
HHLL
HLL
NLL
LLL
LLLL
HILL
HIL
NIL
LIL
LLIL
Top
Btm
Vsep

29

residence time for oil (min)

surge time for oil (min)
required time (min)
available time (min)
high high liquid level (min)
high liquid level (min)
normal liquid level (min)
low liquid level (min)
low low liquid level (min)
high high interface level (min)
high interface level (min)
normal interface level (min)
low interface level (min)
low low interface level (min)
top of the vessel (min)
bottom of the vessel (min)
separator volume (m3)

that the heaviest phase departs the vessel from the water outlet,
while gas phase, as the lightest phase, leaves the vessel from top
gas outlet. The oil phase departs the vessel through oil outlet gate.
The corresponding parameters of different parts of a typical
separator are listed in Table 1.
To put the suggested design procedure into practice and dene
a clear objective, the target is set to calculate the optimal size of a
three-phase separator, which is installed in the downstream of a
presumed wellhead. The typical ow condition and properties are
chosen as summarized in Table 2 (API Spec. 12J, 1989).
In reality, the plant and equipment requirements impose some
considerations on design procedure toward favored separation,
which are recognized as design criteria. The limitation to the
length of a separator and maximum allowable hydrocarbon content in downstream of a are are examples of drawbacks, which
must be considered due to the plant and equipment, respectively.
To meet a proper design, all of the criteria must be taken into
account. In the current work, to achieve a well-dened design, a

Fig. 1. The longitudinal section of a typical three-phase gravity separator.

Table 1
Parameters devoted to the longitudinal section of the separator.

2. Problem description
Fig. 1 depicts an image of a typical horizontal three-phase
gravity separator. In this illustration, the inlet ow, specied as
mixture, comprises oil, water and gas phases.
When the mixture stream passes through the separator length,
it splits into three separate phases. One can see through the gure

Line

Denition

Parameter

AB
CD
EF
DG
CE

Separator length
Separator effective length
Bafe height
Separator diameter
Water channel length

Lt
Le
hb
Dsep
Lw

30

M. Mostafaiyan et al. / Journal of Petroleum Science and Engineering 119 (2014) 2835

Table 2
Inlet ow to the separator.

Table 3
Outlet ow criteria in a separator.

Physical properties

Unit

Value

Gas Phase

Flow rate
Density
Viscosity

l/s
kg/m3
mPa s

40
40
0.01

Oil or light liquid phase

Flow rate
Density
Viscosity

l/s
kg/m3
mPa s

15
600
0.7

Water or heavy liquid phase

Flow rate
Density
Viscosity

l/s
kg/m3
mPa s

5
1000
0.55

100 m
100 m
100 m

Light (hydrocarbon) phase droplet size in gas phase

Heavy (water) phase droplet size in light (hydrocarbon) phase
Light (hydrocarbon) phase droplet size in heavy (water) phase

Table 4
Criteria assumed for level setting in the separator.
Upper level

Lower level

(mm)

(min)

HHLL
HLL
NLL
LLL
LLLL
HHIL
HIL
NIL
LIL

HLL
NLL
LLL
LLLL
HHIL
HIL
NIL
LIL
LLIL

100
100
100
100
100
100
100
100
100

1
1
1
1
1
1
1
1
1

Up to

Table 5
Residence time and surge time in the separator.

Fig. 2. Schematic phase separation in a three-phase horizontal gravity separator.

set of typical criteria are introduced that can be easily changed

when required.
The rst criterion is the extent of the separation in the products
or outlet streams. Since a complete phase separation is a near-toimpossible task, it is customary to dene a separation degree on
the basis of plant requirements. The droplet size of the dispersed
phase in the continuous phase is an important factor in determining viscosity of the outlet streams (Kang et al., 2012; Plasencia et
al., 2013); hence, the degree of separation is routinely dened as
the maximum droplet size of the secondary phase within the
outlet stream.
Whenever the mixture stream passes through the vessel,
different phases need the chance to be separated one from the
other. In a horizontal three-phase separator, various separation
phenomena must be considered. The oil droplets (liquid light
phase droplets) dispersed in the gas continuous phase move
towards the oil continuous phase. At the same time, water droplets
(liquid heavy phase droplets) have the tendency to be separated
from light oil phase to join water continuous phase. Moreover, the
oil droplets trapped into the water phase rise up for being involved
in the oil continuous phase. Fig. 2 shows the described phase
separation, schematically.
Table 3 illustrates the maximum droplet size of the secondary
phase (the separation degree), as the rst criteria (API Spec. 12J,
1989; Jaubert et al., 2002).
The obligations of the liquid level controlling system, which are
important for a safe and practical operation, are introduced as
another set of design criterion. The distance between various
levels ought to be so large to provide sufcient space for installation of required equipment. In addition, level up time must be

Required time

Denition

Minimum
required time
(min)

Water residence time

Water surge time
Hydrocarbon residence time
Hydrocarbon surge time

NIL to btm
HIL to LIL
NLL to NIL
HLL to LLL

4
4
4
4

sufciently large to provide the operator with enough time to take

necessary action in case of emergencies. Based upon this fact,
some typical levels, are specied as the second set of design
criteria (Table 4). This leveling system has customarily been
considered by the designers of gravity separators (Skeie and
Halstensen, 2010). The assigned parameters are schematically
described in Figs. 8 and 10 when model outputs are presented.
The reader is referred to Nomenclature section for denitions of
employed parameters.
To guarantee a practical separation through safe operations,
some additional criteria like residence time and surge time have to
be taken into consideration (Table 5).
It is recommended to set the highest liquid level at least
300 mm or 20 percent of vessel diameter under the vessel top to
make the installation of the mesh frames possible. It is also
advised to set the lowest liquid level at least 150 mm above
the vessel bottom to provide required space for installation of
the vortex breaker (Ludwig, 1994; Datta, 2008). It should be
noted once again that the data chosen for design of separator
(Tables 25) are typical values and can be easily changed and
adopted for other cases.
Settling resistance is another factor that affects the settlement
rate. Accordingly, as the velocity of gas phase increases, resistance
to settlement of the oil phase droplets increases too. It is worthy to
mention that when the velocity of gas phase goes beyond the
critical velocity, the settlement rate is ignorable. This is an
unfavorable happening for the separation phenomena. The critical
gas velocity can be calculated employing (Datta, 2008)
V c 0:135 

s
l  g
g

where l and g are the densities of liquid and gas phase,

respectively.

M. Mostafaiyan et al. / Journal of Petroleum Science and Engineering 119 (2014) 2835

As a design assumption for a separator, the gas velocity is lower

than the critical value. In a real separation practice, the levels of
different phases can differ from what are shown in Fig. 1. For
example, the heavy liquid phase level, which is dened as oil
water interface, may rise up and down between HHIL and LLIL. The
separation phenomena shall not be corrupted in analogous cases.
To achieve this goal, the required time (tr) for the droplet of the
light hydrocarbon phase to rise upward the light continuous phase
must be lower than available residence time (ta). This is the case
when oilgas interface changes between HHLL and LLLL. In other
words, the droplets of heavy phase (water) must have enough
time for settlement (tr) toward the water continuous phase,
and this is attainable when tr is smaller than available residence
time (ta).

31

Fig. 3. Arrangement of seven bites in an individual vector.

Table 6
The allowable range of the design parameters.
Parameter

Minimum

Maximum

Dsep
Le/Dsep
Lw/Le
tres,w
tsurge,w
tres,o
tsurge,o

1
3
0.6
4
4
4
4

5
6
0.9
15
15
15
15

3. Separation in theory
When two immiscible uids, gasliquid or liquidliquid systems, are kept in contact for a sufciently enough time, the denser
phase will settle down whilst the other rises upwards. This
phenomenon is called phase separation. Depending on the
Reynolds number of the ow, hereafter referred to as Re, the nal
velocity in which the heavy phase settles down and light phase
rises up can be easily calculated employing the following formulas
(Datta, 2008):
s
gDp h  l
Newton's law V f 1:74
Re Z 500
2  a
c
Stokes' law

Vf

Intermediate law

1000gD2p h  l
18c
Vf

Fig. 5. Mutation arrangement of one seven-bite in a cell.

Re r 2

3:54g 0:71 D1:14

h  l 0:71
p
0:29
0:43
c
c

2  b

2 o Re o 500
2  c

where Vf is the nal velocity, Dp is the particle diameter, is the

viscosity, is the density, and g is the gravity constant. In addition,
the subscripts h, c, and l denote the heavy, light, and continuous
phases, respectively. Before utilizing all above equations, Re should
be calculated via
Re

Fig. 4. Recombination of two seven-bit cells.

1000Dp c
Vf
c

Since Re itself is a function of settling velocity, an iterative

method must be employed to simultaneously calculate the Re and
Vf. It is also assumed that nal velocity increases up to a limiting
value of Vf 2 mm/s. Based upon Eqs. (1) and (2),settling velocity
of liquid droplets in a continuous vapor or liquid phase is
dependent on the droplet size (Walas, 1998).
It is clear that increasing the diameter or length of vessel causes
diminishing the velocity as well as increasing the residence time
(Lopez-Vazquez and Fall, 2004). In this situation, a better separation process is expectable. Nonetheless, this status is not often a
favored action, since in this way the vessel will be very massive.
This necessitates the importance of choosing a powerful method
to achieve an optimal design.

4. Evolutionary computation strategy

The sizing of a horizontal three-phase separator is to introduce
a set of design parameters to fulll the design criteria, which are
mentioned earlier; in a manner that the vessel volume attains its
minimum value. In the current work, the vessel diameter (Dsep),
length to diameter ratio (Le/Dsep), bafe position ( Lw/Le), water

residence time (tres,w), water surge time (tsurge,w), oil residence time
(tres,o) and oil surge time (tsurge,o) are selected as the most
inuential design parameters.
The relatively great number of effective parameters and design
criteria make it difcult to solve the design problem using
conventional methods. Therefore, an evolutionary computation
strategy is employed to nd a set of design parameters leading to
the optimal sizing. Thus, inspiring from evolutionary techniques, a
xed-length real-valued vector is considered to represent a
member or an individual of separator society, as shown in Fig. 3.
A bit string denes an individual, and each position in the
vector corresponds to a design parameter or behavioral characteristic of a member (Michalewicz, 1996). It can easily be seen that
using data stored in the seven-bit length vector; one can calculate
dimension and normal liquid levels in the separator. After dening
the structure of the individuals, populations of seven-bit vectors
that can pass design criteria are produced, whether or not their
volumes are minimum. The initial population or rst generation is
produced using the uniform design method (Fang, 1980) in the
region of accepted parameters, which are shown in Table 6.
The allowable range for different parameters is picked from
available database of similar design jobs (API Spec. 12J, 1989).
Although a very wide range of parameters could be applied on the
producing of the rst generation, this adjustment leads to a higher
quality of the rst generation and a faster solution as a result. The
number of rst generation population is set to be N 100. When a
generation of possible responses is available, the second step goes
to start. Here recombination and mutation, as the primary generation operation, are employed to produce offspring. The individuals selected from available population to produce an offspring
are called parents.
The main operator is intermediate recombination, in which
the average of two parents' vectors produce an offspring. In
the recombination operation, two random individuals or parents

32

M. Mostafaiyan et al. / Journal of Petroleum Science and Engineering 119 (2014) 2835

are selected from the former generation and the action of

recombination is performed on them. An example of recombination is shown in Fig. 4.

Another reproduction operator in this evolutionary programming is the mutation, in which a random value (positive or
negative) is added to each element of an individual's vector to
create a new offspring. Notably, none of the elements must go
beyond the limiting values mentioned through Table 6. Fig. 5
illustrates the mutation in an individual.
Since the possibility of mutation is always minuscule (Kaydani
et al., 2011), and in this investigation it is considered to be ve
percent. When a new offspring is produced, either by recombination or mutation, it must be checked for passing the design
criteria.
The quantity of offspring is selected to be as much as rst
generation (N). Starting from N population of the offspring, a
society with population of 2N is available at the end of the second
step. In the third stage, a population of N individuals having lower
volume must be chosen among the 2N individuals (parent and
offspring). These N individuals, which t the optimization criteria
and have lower volume, are called survivors. Likewise, steps two
and three must be followed using survivors of the third stage until
convergence is achieved. Arbitrarily, the convergence would be a
generation in which average volume of a generation is equal to the
minimum volume or their difference is lesser than 10  7 m3.

5. Results and discussion

To solve the design problem, an in-house computer code was
developed. Firstly, the ow rate of each phase and their physical
properties (densities and viscosities) as well as the degree of
separation, identied in Table 3, were applied on the computer
code as the input. Next, design criteria represented in Section 2
followed by the allowable range of the design parameters were
inserted into the program. The ow chart for the optimization
procedure is depicted in Fig. 6.
After execution of the computer code, two different parameters, namely the generation average volume and generation
minimum volume were illustrated to monitor the progress in
generation. The program has been run many times and very close
results obtained. Thus, two sets of data are typically depicted and
their outputs are compared and discussed.
5.1. First set of sizing results
Fig. 7 illustrates the mean value and minimum volume of separator population versus generation. As in this plot, a descending

Fig. 6. Outline of the principle of the optimization.

Fig. 7. The alteration of separator volume in terms of generation (1st set).

M. Mostafaiyan et al. / Journal of Petroleum Science and Engineering 119 (2014) 2835

33

Fig. 8. The dimensions of the separator and the position of the levels.

Table 7
Quantities of parameters of the phase separator in
the rst set.
Parameter

Dimension

Dsep
Le
Lw/Le
hb
LLIL
LIL
NIL
HIL
HHIL
LLLL
LLL
NLL
HLL
HHLL

1.68
8.31
0.60
0.60
150
250
350
450
550
654
754
888
1024
1124

trend is obviously seen upon increasing the generation in both the

minimum and mean volumes. Accordingly, it is evident that
an optimum solution is attainable. Furthermore, the tendency
of mean volume toward minimum volume at higher generations
suggests that the whole populations of each generation are
acquiring the properties of optimum solution. In another words,
this case describes the situation that all individuals in the society
have a set of design data, which leads to minimum value. Thus, the
combination of members cannot always guarantee generation of a
member with better properties; however, the mutation may
propose a separator with lower possible volume.
What makes the results in Fig. 5 more signicant is concerned
with the ratio of mean volume in the rst generation to that
of separator after achieving convergence. Noticeably, the vessel
volume reduced 2.4 times after 159 generations.
When the solution is completed, the cell information of each
individual (separator vessel) having minimum attainable volume
is achieved. Fig. 8 illustrates the dimensions of the separator and
the position of the various levels in the separator. For convenience,
the exact quantities corresponding to dimensions of the separator
and those of positions of the levels are summarized in Table 7.
Table 8 compares required residence times with available ones,
for heavier phase, in each level. From this set of data it can be
realized that the residence times for all the levels are sufciently
higher than requirements, hence, the droplet of the light phase
(with diameter of more than 100 m) have enough time to rise up
and join the lighter continuous phase before leaving the vessel
with heavier phase. For example, when the heavy phase occupies

Table 8
The comparison between the available and required residence times in the
heavy phase.
Level up to btm

Available time
(min)

Required time
(min)

LLIL
LIL
NIL
HIL
HHIL

1.62
3.42
5.56
7.93
10.48

0.63
1.05
1.47
1.90
2.32

Table 9
Comparing the available and required residence times in the light phase.
Upper level

Lower level

Available time
(min)

Required time
(min)

LLLL
LLL
NLL
HLL
HHLL
LLLL
LLL
NLL
HLL
HHLL
LLLL
LLL
NLL
HLL
HHLL
LLLL
LLL
NLL
HLL
HHLL
LLLL
LLL

LLIL
LLIL
LLIL
LLIL
LLIL
LIL
LIL
LIL
LIL
LIL
NIL
NIL
NIL
NIL
NIL
HIL
HIL
HIL
HIL
HIL
HHIL
HHIL

7.29
8.18
3.28
4.2
5.44
6.69
7.58
2.57
3.49
4.73
5.98
6.87
1.78
2.69
3.94
5.19
6.08
0.93
1.84
3.09
4.34
5.23

4.69
5.22
2.17
2.7
3.42
4.15
4.69
1.63
2.17
2.89
3.61
4.15
1.09
1.63
2.35
3.08
3.61
0.56
1.09
1.81
2.54
3.08

the vessel bottom up to LIL level, it takes 3.42 min to leave the
vessel, while the light phase only needs just 1 min for moving
toward the hydrocarbon phase.
Table 9 compares required and available residence time, for
light liquid phase, in each level. The assessment of the available
and required times in Table 9 witnesses that the water droplets
wholly (with diameter of more than 100 m) settle down up to
required level. A suitable set of the required and the necessary
residence times for the gas phase to provide enough time for the
settlement of the hydrocarbon droplets are also compared through

34

M. Mostafaiyan et al. / Journal of Petroleum Science and Engineering 119 (2014) 2835

the Table 10. It is also notable that there is enough space between
HHLL and top level to guarantee a gas velocity lower than its
critical value calculated by Eq. (1).
Evaluation of data in Tables 810 illustrates that the settlement
phenomena in the gas phase is more dominant in the sizing
procedure of the vessel, for the available residence times are very
close to those of required values.

descending trend, which is very similar to the previous case

(Fig. 5).
Fig. 9 shows that the average vessel volume in the rst
generation (42.25 m3) is reduced more than 2.3 times to a vessel
with volume of 18.17 m3. Fig. 10 depicts the dimensions and the
internal settings for the second case. In this case, also, all of the
available residence times are more than required residence time.
The detail of residence times are omitted for brevity.

5.2. Second set of sizing results

5.3. The comparison of the sizing results

Fig. 9 illustrates the mean value and minimum volume of

separator population versus generation. The curves undergo a

The design procedure of the horizontal three-phase separator

was applied to solve the problem for several times; however, two
typical sets were shown previously. In this section, the geometrical
information, stored in the seven-bit cells, is compared together for
six different program executions and the outputs are given in
Table 11.
The results from comparative evaluation indicate that the
optimal design is corresponding to the case that the separator
takes the volume of around 18.4 m3. Based on Table 11, one can
realize that a nearly unique solution is achievable, irrespective of
the quality of the rst generation.

Table 10
The comparison of the available and require residence times in the gas phase.
Top up to the level

Available time
(min)

Required time
(min)

LLLL
LLL
NLL
HLL
HHLL

4.89
4.31
3.54
2.76
2.2

4.06
3.66
3.13
2.59
2.19

6. Conclusion
This work provides an overview to design of three-phase
gravity separators by developing a computer code based on
evolutionary computational approach. A model horizontal gas/
oil/water separator is dened for which design parameters including separator length, diameter, water channel length, residence
and surge times are manipulated to meet a target design with
minimum attainable volume. In the procedure of computation,
each separator is considered as an individual in separator population and the design parameters are assumed to have a seven-bit
Table 11
The comparison of the separator dimensions applying different program
executions.
Set No.

Dsep

Le/Dsep

Lw/Le

Vsep

1
2
3
4
5

1.71
1.70
1.62
1.67
1.68

4.71
4.8
5.44
5.03
4.95

0.60
0.60
0.60
0.60
0.60

18.46
18.46
18.17
18.36
18.36

Fig. 9. The alteration of separator volume in terms of generation (2nd set).

Fig. 10. The dimensions of the separator and the position of the levels.

M. Mostafaiyan et al. / Journal of Petroleum Science and Engineering 119 (2014) 2835

representative. Through a consecutive procedure, new generations

are born by recombination and mutation of previous ones to
achieve new generation representing lower volume. This ended in
a separator with minimum possible volume. The utilized approach
yields a nearly unique and minimum volume for three-phase
separators, regardless of the rst generation. This has benecial
point of view for actual cases, in which engineers are encountered
with platform limitations. It is also believed that this methodology
can be extended to design separators with different geometrical
characteristics and optimization targets, such as separators with
minimum cost or minimum weight.
References
API Specication 12J, 1989J. Specication for Oil and Gas Separators, 7th ed.
American Petroleum Institute.
Auem, I.H, Kallevik, H., Westvik, A., Sjo
blom, A.J., 2001. Inuence of pressure and
solvency on the separation of water-in-crude-oil emulsions from the North Sea.
J. Pet. Sci. Eng. 31, 112.
Bazin, C., Payenzo, G., M., Desnoyers, M., et al., 2012. The use of simulation
for process diagnosis: application to a gravity separator. Int. J. Miner. Process.
104105, 1116.
Datta, A., 2008. Process engineering and design using visual basic. CRC Press, Taylor
& Francis Group 6000, Broken Sound Parkway, New York, pp. 207210
(Chapter 3).
Elsayed, Kh., Lacor, Ch., 2010. Optimization of the cyclone separator geometry for
minimum pressure drop using mathematical models and CFD simulations.
Chem. Eng. Sci. 65, 60486058.
Elsayed, Kh., Lacor, Ch., 2012. Modeling and Pareto optimization of gas cyclone
separator performance using RBF type articial neural networks and genetic
algorithms. Powder Tech. 217, 8499.
Fossen, M., Arntzen, R., Hemmingsen, P.V., et al., 2006. A laboratory-scale vertical
gravity separator for emulsion characterization. J. Dis. Sci. Tech. 27, 453461.
Fang, K.T., 1980. Uniform designapplication of number theory to experimental
design. Acta Math. Appl. Sin. 3, 363372.

35

Jaubert, J., Avaullee, L., Souvay, J., 2002. A crude oil data bank containing more than
5000 PVT and gas injection data. J. Pet. Sci. Eng. 34, 65107.
Jaworski, A.J., Meng, G., 2009. On-line measurement of separation dynamics in
primary gas/oil/water separators: challenges and technical solutionsa review.
J. Pet. Sci. Eng. 68, 4759.
Kang, W., Guo, L., Fan, H., Meng, L., Li, Y., 2012. Flocculation, coalescence and
migration of dispersed phase droplets and oilwater separation in heavy oil
emulsion. J. Pet. Sci. Eng. 81, 177181.
Kaydani, H., Mohebbi, A., Baghaie, A., 2011. Permeability prediction based on
reservoir zonation by a hybrid neural-genetic algorithm in one of the Iranian
heterogeneous oil reservoirs. J. Pet. Sci. Eng. 78, 497504.
Krebs, T., Schroen, C.G.P.H., Boom, R.M., 2012. Separation kinetics of an oil-in-water
emulsion under enhanced gravity. Chem. Eng. Sci. 71, 118125.
Lopez-Vazquez, C.M., Fall, Ch., 2004. Improvment of a gravity oil separator using a
designed experiment. Water Air Soil Pollut. 157, 3352.
Ludwig, E.E., 1994. Applied Process Design for Chemical and Petrochemical Plans,
third ed. Gulf Publishing Company, Houston, TX.
Michalewicz, Z, 1996. Genetic algorithms data structures evolution programs,
third ed. Springer, Berlin.
Oh, A.G., Lee, D.W., Cho, Y., 2012. A continuous cell separator based on gravity
force and buoyant force in dissimilar density uids. Sens. Actuators B:Chem.
166167, 2430.
Plasencia, J., Pettersen, B., Nydal, O.J., 2013. Pipe ow of water-in-crude oil
emulsions: effective viscosity, inversion point and droplet size distribution.
J. Pet. Sci. Eng. 101, 3543.
Rosa, E.S, Frana, F.A, Ribeiro, G.S, 2001. The cyclone gasliquid separator: operation
and mechanistic modeling. J. Pet. Sci. Eng. 32, 87101.
Skeie, N., Halstensen, M., 2010. Level estimation in oil/water separators based on
multiple pressure sensors and multivariate calibration. J. Chemom. 24,
387398.
Smith, R., 2005. Chemical Process Design and Integration. John Wiley & Sons Ltd.,
the Atrium, Southern Gate, Chichester, England, pp. 143144.
Walas, S.M., 1998. Chemical Process Equipment Selection and Design. ButterworthHeinemann, USA, pp. 611618.
Zeevalkink, J.A., Brunsmann, J.J., 1983. Oil removal from water in parallel plate
gravity-type separators. Water Res. 17, 365373.
Zhang, Lu., Xiao, H., Zhang, H., et al., 2007. Optimal design of a novel oilwater
separator for raw oil produced from ASP ooding. J. Pet. Sci. Eng. 59, 213218.