Vous êtes sur la page 1sur 16

SPE 56645

Optimal Design of Two- and Three-Phase Separators: A Mathematical Programming


Formulation
Evert O. Grødal, SPE, Kværner ASA Technology Development, Georgia Institute of Technology, and
Matthew J. Realff, Georgia Institute of Technology.

Copyright 1999, Society of Petroleum Engineers Inc.


of the nature of multi-variable manual trial and error
This paper was prepared for presentation at the 1999 SPE Annual Technical Conference and procedures. The embedded nature of the solution procedure
Exhibition held in Houston, Texas, 3–6 October 1999.
within the model makes modification of either model or
This paper was selected for presentation by an SPE Program Committee following review of procedure difficult. Moreover, such a program does not
information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to represent an efficient analyzing interface for the assessment of
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at the optimality and flexibility of its design, as many operational
SPE meetings are subject to publication review by Editorial Committees of the Society of different cases must normally be analyzed to establish a single
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is separator design.
prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300
words; illustrations may not be copied. The abstract must contain conspicuous
The separator models proposed in this paper, and their
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O. solution algorithms, determine the optimal horizontal
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
separator design for a nominal set of design parameters. The
optimal solution is found using the objective of minimizing
Abstract the separator vessel manufacturing cost. The design is
Oilfield primary separator units are subject to nonlinear fluid constrained by a set of fluid dynamic and mechanical
dynamic and mechanical constraints. Their design represents a relationships formulated from gravity settling theory, re-
challenging and important problem. In this paper we present entrainment models, controllability issues and includes a
separator design problems for which the lowest cost, optimal, simple model for separators mounted on moving vessels. We
solution is found using sequential quadratic programming believe, for several reasons, that the efficiency and
techniques. The optimization models utilize existing separator effectiveness of the separator design procedure are improved
design theories and relationships. They consist of a declarative significantly if formulated and solved as a mathematical
set of equations and inequalities that is independent of the program. First, using mathematical programming techniques,
solution method and hence it is easy to make modifications the optimum design is obtained through a simultaneous search
and the assumptions are clear. The models are outlined and over all design variables. Second, the solution provides the
discussed in the context of horizontal separators. The designer with insight into what constrains the design. Third,
efficiency and effectiveness of the suggested systematic the solution is obtained in mere seconds in an automated
design method are exemplified by two illustrative design procedure, which reduces the risk of human error. Finally, the
cases. model is declarative and thus its components can be modified
with little effort. For example, the objective could be to
Introduction minimize operational weight, length or footprint area and this
The basics of horizontal separator design methods have been would require minimal change to the model.
described elsewhere1-4. Typically, these methods assume half- In this paper we focus on the design of single separator
filled liquid vessels without level control issues and appear to units with a given operating point (i.e., fixed physical
be somewhat simplified versions of industrial design properties). However, the presented formulation forms a basis
procedures. Two articles by Monnery and Svrcek5,6 outline for several more advanced design problems. Some of these
design procedures in a more comprehensive manner and are studies are design for reuse, flexibility and of product lines7
claimed to be close to what is practiced. However, their (standardization issues), and the design for lifetime application
procedures are manual trial and error with extensive table of more complete primary production separation systems8.
look-ups and require the application of many rules of thumb.
The procedures are not impossible to computerize, but to Process Unit Design and Mathematical Programming
establish a feasible and acceptable design still requires Algorithmic mathematical optimization techniques have been
significant manual iteration on the design variables values. used extensively in systematic methods for chemical process
Such programs provide limited tools for the designer because design. Mathematical programming formulations are used for
2 E.O. GRØDAL AND M.J. REALFF SPE 56645

the design of units, systems, and are even used at the perceived that the phase change is nearly complete as the fluid
conceptual design level9. The power of a mathematical enters the separators13. The two theories are often used in a
program, if properly formulated, is that it can integrate and coupled fashion where settling theory applies to the separation
simultaneously solve the traditional chemical process design in the gas phase and retention time theory applies to the
activities such as synthesis, analysis, evaluation and separation in the liquid phases. Arnold and Koszela14 present a
optimization. comparison of the two methods.
To cast the separator design problem into a mathematical In the following sections the settling theory and the
program it must be formulated in the following general form, retention time theory are presented along with their following
max (or min) z = f (x1 , x 2 ,.....x n ) , ………………….(1)
three major underlying assumptions related to the design of
horizontal separators:
• The assumption that potential problems due to waxing,
subject to g1 ( x1 , x 2 ,.....x n ) (≤, =, or ≥ ) b1 foaming and emulsions are solved successfully through
g 2 ( x1 , x 2 ,.....x n )(≤, =, or ≥ ) b2 chemical injections.
M • The plug flow assumption.
• The assumption of linear movements of droplets and
g m ( x1 , x 2 ,.....x n ) (≤, =, or ≥ ) bm
bubbles in the settling theory.
The objective function (z), which is a function of the design
variables (xi ), is to be minimized or maximized subject to a set Settling Theory. Settling theory is an abstraction of the real
of constraints (gi). The constraints are functions of the design process that considers the velocities of the phases through the
variables and describe physical design relationships expressed separator and the settling velocity of the various bubbles and
as inequalities or equalities. A feasible solution of a droplets. Settling theory has been successfully applied to the
mathematical program satisfies all the constraints. A subset of design of many other units, e.g., settling tanks for the
the inequality constraints will be active, that is satisfied at separation of formation particles from the drilling mud or the
equality. The active constraints will inform the designer which many applications of skimmers15. To apply settling theory in
physical relationships are constraining the design. The status separator design, it is assumed that the droplets and bubbles
of the inactive constraints can be used to quantify the acts as spherical particles, and that they will settle in a
flexibility of the design. This analysis is illustrated in the case stagnant continuous phase because of gravity forces. For a
studies. three-phase separator, the free settling velocity for 6 different
Many algorithms exist to solve mathematical programs10. cases must be considered. These are:
Because of nonlinearity in the separator design problem (both • Water droplet settling in continuous gas phase
in objective function and in constraints) we have chosen to use • Oil droplet settling in continuous gas phase
sequential quadratic programming9,11,12 (SQP) to find the • Gas bubble settling in continuous oil phase
optimal separator design. A globally optimal design is one for • Water droplet settling in continuous oil phase
which no other design have a higher (lower) objective function • Oil droplet settling in continuous water phase
value. A locally optimal design is one where within a certain • Gas bubble settling in continuous water phase
neighborhood no better solution exists. Although an explicit
proof of mathematical optimality is omitted here, we address In general, the net vertical force acting on a perfect
the optimality of the solutions in the case studies. spherical bubble/droplet immersed in a stagnant phase is the
force resulting from the gravity force downwards and the
Foundations in Horizontal Separator Design Theory buoyancy force upwards. For oil droplets in continuous gas
Many geometrically different units for the separation of fluids phase, the following expression for free settling velocity is
produced from hydrocarbon reservoirs have appeared in obtained by equating the two forces3,
industry throughout its history. There are several designs of
0.5
two- and three-phase horizontal separators. The major designs  4 d og ρo − ρ g 
include5,6; interface control, interface control with weir, u og = g  . …………………….....(2)
 3 C D,og ρ g 
interface control with bucket, and bucket and weir.
The function of a separator is to separate the input, The spherical drag coefficient for both laminar and turbulent
containing water, oil, and gas, into phases that are relatively flow is given by
free from each other at given operating points. Separator
sizing techniques are based upon both theoretical and 24 3
C D ,og = + + 0.34 , ……………….…...(3)
empirical formulas, which have worked quite well in practice. N Re,og N Re,og
Separators are designed to yield certain simultaneous gas, oil
and water capacities. These capacities are commonly where the droplet Reynold’s number is given by
determined based on settling theory or on retention time
ρ g uog d og
theory. Both theories have weaknesses and different literature N Re,og = . ………………………………..…(4)
concludes in favor of either theory. In both theories it is µg
SPE 56645 OPTIMAL DESIGN OF TWO- AND THREE-PHASE SEPARATORS: A MATHEMATICAL PROGRAMMING FORMULATION 3

To solve for the free settling velocity, Eqs. 2 to 4 must be used that injection systems for production chemicals often are
in an iterative manner. We have found that a Newton-Raphson added to the process system on a “just in case” basis.
iterative search10 is fast and works well on this problem. In
case of a laminar flow a simpler expression exists for the drag Plug Flow Assumption. As depicted in Fig. 1, the plug flow
coefficient, which neglects the two last terms in Eq. 3. This assumption is clearly an idealized picture of the complex flow
allows an explicit solution for the settling velocity. pattern in large diameter separators, especially since it violates
The challenge in the application of settling theory is to the no-slip flow condition at the wall and interfaces. Internals,
determine which minimum droplet/bubble diameter will yield like baffle plates, are normally installed to improve this
the desired separation efficiency. Arnold and Stewart3 reports assumption. As discussed by Asheim18 a more realistic
that from field experience it appears that if 100-micron velocity profile in horizontal separators are as depicted in Fig.
droplets are removed by settling in the gas continuous phase, 1.
then the mist extractors will not become flooded and will be Realistic velocity pattern Assumed velocity pattern
Gas outlet

able to remove those drops between 10- and 100-micron in


diameter. They also report that a removal criterion of 500-
Gas
micron for both water droplets in oil and oil droplets in water phase
ug ug
Inlet
provides good overall separation. The separator design is
fairly sensitive to the selected diameters of the settled droplets
and bubbles, and thus some sort of experimental knowledge is Oil uo uo
preferable, either on the actual oil or from experience about phase
similar oils from neighboring reservoirs.
Water
phase
uw uw
Retention Time Theory. Retention time theory uses
experimental settling and coalescence data to approximate the Water outlet Oil outlet
time that the various phases must spend in the vessel to enable
separation to occur. A weakness in the retention time theory is Fig. 1 – Plug flow assumption visualized in three-phase separator.
that the retention time tests will limit the freedom of the
separator design, unless a quite extensive number of tests are Linear Movement Assumption. As a consequence of the
performed. First, the tests must be performed on the actual plug flow assumption and settling theory the bubbles and
fluids at the separator operating pressure and temperature. droplets have linear settling paths. However, as illustrated by
Second, since the retention time tests are performed in a tank numerical simulations of the flow in separators19, the linear
with certain oil and water pad thickness, the actual oil and motion assumption is far from true. Because of the “random”
water pad thickness in the separator design needs to be similar. motion of droplets/bubbles in a turbulent flow, there is no
Typical retention times range in the area 30 sec to 3 min3. assurance that a larger droplet will settle before a smaller one,
In case of foaming crudes and severe emulsions the retention even though this case is more likely. The real picture can
time might be many times higher. However, these situations be explained by considering droplet/bubble diameter
call for the use of chemicals. Hafskjold et al.16 is a reference distributions. The number of droplets/bubbles and their
clearly in favor of the retention time theory. diameter distributions that enters the separators are complex
functions of the fluids upstream path through chokes, valves
Chemical Injections. All natural occurring hydrocarbon and bends that creates chaotic flow. The numerical studies by
mixtures exhibit various degrees of waxing, foaming and are Grødal19 demonstrated that for a certain entering size
subject to the creation of both oil-in-water and water-in-oil distribution of particles, the shape of the distribution at the
emulsions. These problems are normally solved through the outlet would be skewed towards smaller diameter droplets.
injection of chemicals. However, providing higher retention Thus, reducing the mean of all diameters, but not creating a
times of the oil phase or the water phase can solve small sharp cut at a certain diameter size. This observation has also
emulsion and foaming problems. In the settling theory, this been found experimentally20.
could be achieved by simply reducing the minimum diameter Attempts to include information about the droplet
of the various settled bubbles or droplets to allow higher distributions in the separator design procedures have been
residence times. made16. However, in practice the minimum diameter size
Even though research is underway to better understand the parameter (settling theory) is viewed as an adjustable
mechanisms behind the above problems17, little is found in parameter dependent on the droplet and bubble distributions at
open literature that easily converts into mathematical design the inlet of the separator and is determined based on
relationships. In this paper it is assumed that the above experience. Thus, settling theory is used to capture the
problems are successfully solved through the use of chemicals. mathematical trend (i.e., a design with use of smaller
Kaasa17 is a good reference for insight into the problem. He droplet/bubble diameter gives better separation), rather than
discusses the synergistic effects that can be achieved if a better believing that all droplets or bubbles down to a certain
understanding is reached. An interesting note Kassa makes is diameter are separated.
4 E.O. GRØDAL AND M.J. REALFF SPE 56645

Physical Separator Design Model enhances the linear momentum of the fluid along the
In this work settling theory is applied for separation in all longitudinal axis of the separator. At the entrance to the
phases for two reasons. First, we believe that a general settling section the various phases will contain oil droplet,
optimization model should be based on first principles as water droplet and gas bubble distributions. In order for these
much as possible and that field specific experimental data bubbles and droplets to be separated they must travel to their
should be used within this framework rather than to usurp it. respective continuous phases. This situation is analyzed by
Second, cyclones have been applied as inlet internals in using critical settling paths as will be introduced in next
separators and have replaced the many designs of momentum section. The separator will be designed with perforated baffle
breakers or deviators21. It has been found that the cyclones plates transversely mounted along the separator for moving
perform a better coarse separation by reducing both the offshore vessel applications. The settling section contains a
number of droplets and their size. The creation of foam and water jetting system so that BS&W can be flushed out of the
emulsions is also reduced by cyclones. The experimental bottom of the separator. The end section includes the outlet
coalescence tests, performed for the retention time theory, nozzles, weir and a vane demister. The water and oil outlet
might not simulate the actual separation mechanism for the nozzles are equipped with vortex breakers. A weir separates
flow of small and scattered droplets coming out of cyclones. the water outlet/vent and the oil outlet to prevent water
The retention time theory can be applied by applying the escaping out of the oil outlet. A vane demister that removes
simple modifications presented in Appendix A. The detailed the very small droplets of liquid in one final separation
horizontal separator configuration and an argument for a process is mounted ahead of the gas outlet nozzle. The level
simplified application of the settling theory are presented in controllers operate in conjunction with liquid discharge valves
the following sections. to maintain the appropriate levels in the vessel. A pressure
controller, whose output is fed back to a gas outlet valve,
Specific Design, Internals and Controls. A horizontal maintains the appropriate separator operating pressure.
separator with interface control and weir, which is widely used
for offshore applications, has been chosen as the basis for the Critical Settling Paths and Phase Levels. The critical
models. A three-phase separator of this kind is depicted in Fig. settling paths in a three-phase separator are illustrated in Fig.
2. A two-phase separator of this design will be similarly 3. The thinking behind these critical paths is that if the droplet
configured without the normal interface level. The water outlet furthest away from its destination level is settled, all equal or
in the two-phase separator is a water vent for occasional larger sized particles, having initially equal or shorter paths to
flushing of the separator to remove basic sediment and water travel, will also reach their destination level. Thus, in order to
(BS&W). The weir in a two-phase separator avoids BS&W settle all 100-micron and larger water droplets, one 100-
reaching the oil outlet during flushing operations. Thus micron water droplet must travel from the top of the separator
converting the two-phase separator into a three-phase through the gas phase, cross the gas-oil interface, travel trough
separator, or vice versa, requires only simple modifications. the oil phase before it dissolves into the oil-water interface
(path 2 in Fig. 3). Similarly, to settle all 100-micron and larger
Lc Le Gas outlet gas bubbles, one 100-micron gas bubble must travel from the
bottom of the separator through the water phase, cross the
Vane
Cyclone Baffle
demister water-oil interface, travel through the oil phase before it
plates
L vx escapes into the gas phase (path 3 in Fig. 3).
Inlet
Normal Liquid Level Lb Ln
1 Settling oil droplets 3 Rising gas bubbles. Gas outlet
2 Settling water droplets 4 Rising oil droplets.
Weir

Normal Interface Level Vortex D


breaker 1
Inlet

Lh NLL
Water jetting system Water outlet Oil outlet
2
Fig. 2 – Three-phase separator with interface control and weir.
3
NIL
The separators consist of three sections; initial separation
4
section, gravity settling section and the end section. The fluids
enter the separator in the initial separation section. This Water outlet Oil outlet
section contains a cyclone that absorbs the fluid momentum
and directs the gas to the top and the liquid to the bottom of Fig. 3 – Critical settling paths in three-phase horizontal separator.
the vessel. The fluids are directed into the gravity settling
section through a perforated baffle plate. The baffle plate As visualized in Fig. 3, the critical paths are related to the
covers the whole cross-section of the cylindrical shell and normal liquid levels. However, two- and three-phase flows
through pipes and risers will often exhibit slugging and
SPE 56645 OPTIMAL DESIGN OF TWO- AND THREE-PHASE SEPARATORS: A MATHEMATICAL PROGRAMMING FORMULATION 5

surging. Thus, first stage (i.e., three-phase) separators should situation in the two-phase separator is mathematically simple,
be designed to handle these transient flow rates and yield the the changing angle of the critical paths in the three-phase
specified separation efficiency. There are two approaches to separator (paths 2 and 3 in Fig. 3) requires simplification. This
handle unsteady flows that do not include a sophisticated simplification is achieved, with minor errors, by considering
transient model for the control and discharge valve systems. the relative magnitudes of the settling velocities. The
Either the separator can be designed for highest instantaneous simplified critical paths are illustrated in Fig. 6 and explained
rates4,13 or the separator can be designed with slug and surge below.
volumes5,6. From an operational and economic perspective we
believe that the latter approach is most appropriate. First, Gas outlet

when a slug enters the separator the liquid level will rise
because of the time constant of the control and discharge valve TV
systems. Thus, by including multiple levels the true critical
Inlet HHLL
paths can be defined. Since the slug can be bled off over a HLL
NLL
higher flow area, the horizontal flow velocity is marginally LLL
LLLL
altered from its value at the average rate. Second, the
instantaneous flow rate during slugging can be in excess of the HHIL WH
average flow rate, but not necessarily large in volume. For NIL
HIL
example, if the average rate consists of equal volumes of gas BV
LIL
LLIL
and liquids, the liquid instantaneous rate might be twice as
high when the slug enters the separator. Thus, potentially Water outlet Oil outlet
creating a quite expensive overdesign. This argument also
applies to surges when the gas rate is higher than the average Fig. 4 – Control levels in a three-phase separator with weir.
rate. In general, horizontal separators are not desired in cases
Gas outlet
of severe slugging and surging. In these cases it will be more 1 Settling oil droplets 3 Rising gas bubbles.
economical to install a slug catcher upstream of the inlet
separator.
In the case of two-phase separators, which receive a stable Inlet 1
oil from the upstream separator, we do not expect slugs and NLL
HLL

surges where the ratio of gas and liquid rates are altered from LLL

the ratio at the average rates. However, a two-phase separator 3


must be designed to separate the higher rates that the upstream
separator bleeds off while recovering from a slug.
Alternatively, one can let the first stage separator choke the
flow so that the outlet flow of the separator always has a
constant outlet rate, i.e., having the option to design the Water vent Oil outlet
succeeding separators in the train based on average rates. Fig. 5 – Critical settling paths in two-phase horizontal separator.
The models in this work are designed for average rates and
include slug and surge capacities for the three-phase separator 1 Settling oil droplets Gas outlet
model, and a set of level constraints based on operational 2 Settling water droplets 4 Rising oil droplets.
considerations from the NORSOK standards22. As depicted in
Fig. 4, 10 levels and the weir height have to be determined for 1
successful operation of a three-phase separator. The Inlet
abbreviations used are explained in the list of subscripts. In a HLL
NLL
two-phase separator the 5 levels related to the liquid level LLL
2
have to be determined.
The low and high levels are alarm levels that, if reached,
HIL
will notify the operator of some possible malfunction. The NIL
LIL
low-low and high-high levels are trip levels that, if reached, 4

will automatically shut down the separation system. Thus,


specified separation should take place at any time when the Water outlet Oil outlet

levels are between their low and high levels. The critical Fig. 6 – Selected critical settling paths in three-phase separator.
settling paths are therefore related to the maximum travel
distance in the gravity settling section during normal transient The gas compartments must be designed so that most of
flow operations. The continuous phases will change their flow the oil and water droplets will settle and a relatively liquid free
directions in the end section and no effective gravity settling gas will flow out of the separator to be further processed in the
can be expected. The critical settling paths used in the two- compressors and scrubbers. However, since water is heavier
phase separator model are depicted in Fig. 5. While the
6 E.O. GRØDAL AND M.J. REALFF SPE 56645

than oil and if we expect their same particle size to be The condition required to complete the settling of an oil
separated from the gas phase, only the settling of oil droplets droplet of a certain size, is given by
will be critical. Thus, the procedure for designing the gas
t g ≥ t og . ……………………………………………….(7)
capacity will be identical for the two-phase separator and the
three-phase separator. Substituting Eqs. 5 and 6 into Eq. 7 and rearranging yields
The oil compartment must be designed so that most gas
and water will leave the phase and the phase can continue to u g (Di − h LLL )
Le ≥ . ………………………………….(8)
flash in a potential downstream separator. As the settling u og
velocity of water droplets in a continuous oil phase is slow, we
do not expect gravity settling of same particle size as in the Expressing the gas velocity in terms of average volumetric
gas compartment. Since these water droplets have a much flow rate and the cross-sectional area for gas flow we obtain
q g (Di − hLLL )
higher settling velocity in the gas phase than in the oil phase,
the time it takes for these to fall through the gas phase is Le ≥ . ….……………..……...……...(9)
comparably negligible. Thus, the critical path 2 in Fig. 3 is u og ( ATV − ANLL )
simplified to path 2 in Fig. 6.
The water compartment must be designed so that most gas Inequality 9 is the gas capacity constraint and applies to both
and oil will leave the phase and the phase can continue to the two- and three-phase separators.
water cleaning system. Since the gas bubbles has much higher
settling velocity than oil droplets in the continuous water Oil Capacity Constraint. The design of oil capacity in the
three-phase separator is based on the gravity settling of water
phase, and than the water droplets in the continuous oil phase,
droplets according to path 2 in Fig. 6. The same derivation as
the settling of gas bubbles in the three-phase separator is never
the gas capacity case results in
considered critical. Thus, path 3 in Fig. 3 can be omitted as a
critical design path. However, in the two-phase separator q o (hHLL − hLIL )
Le ≥ , ….…………………...……..(10)
(where the water phase is not present) the oil compartment is u wo ( ANLL − ANIL )
designed based on the critical settling velocity of a certain size
gas bubble (normally in the range 100 to 500 microns4). as the oil capacity constraint for the three-phase separator.
The design of oil capacity for a two-phase separator is
Mathematical Formulation of Separator Models based on the gravity settling of gas bubbles according to path
In a mathematical program, the physical design relationships 3 in Fig. 5. Following the same derivation as for Eq. 9, yields
must be expressed as inequalities or equalities. The industry
qo hHLL
practice of using rules of thumb, individual experience and Le ≥ , ….…………………………...………(11)
table look-ups fit poorly in a mathematical program. In the u go ANLL
following sections, mathematical formulations of physical
separator design relationships are presented. Opposed to the as the oil capacity constraint for the two-phase separator.
traditional unit system of the oil industry, the variables in all
equations have a self-consistent set of units, unless otherwise Water Capacity Constraint. The design of water capacity in
noted. Many of the equations presented contain the variable A the three-phase separator is based on oil droplet gravity
with a subscript that express a part of a separators cross- settling according to path 4 in Fig. 6. This yields
sectional area. This relationship is often given as graphical q w hHIL
correlations3,4 or as a ratio of two fitted polynomials5,6. Le ≥ , ...……………………………………..(12)
u ow ANIL
However, an exact equation for this area and its use is
presented in Appendix B. The volume in the separator heads as the water capacity constraint for the three-phase separator,
are conservatively neglected for the many volume expressions using the same arguments presented for Eqs. 9-11.
presented below.
Slug and Surge Capacity Constraints. As discussed, the
Gas Capacity Constraint. The design of gas capacity in both separator should be designed to handle a certain slug and
three-phase and two-phase separators is based on oil droplet surge flow without having the alarm to go off. The desired
gravity settling. The settling direction for the oil droplets will slug and surge capacity can be determined based on
be perpendicular to the gas velocity direction, thus, creating simulations of the flow in the risers and flowlines. The slug
settling paths 1 in Fig. 5 and Fig. 6. The travel time for the gas and surge capacity constraints related to the normal liquid
through the settling section and the settling time for an oil level for both two- and three-phase separators are formulated
droplet may respectively be stated as: as:
t g = Le u g ….….…………...………….………….…(5) Vslug ≤ (Lc + Le + Ln )( AHLL − ANLL ) .………………....(13)

and t og = (Di − h LLL ) u og . ……...…..…..….……………..(6) and Vsurge ≤ (Lc + Le + Ln )( ANLL − ALLL ) . ……………….(14)
SPE 56645 OPTIMAL DESIGN OF TWO- AND THREE-PHASE SEPARATORS: A MATHEMATICAL PROGRAMMING FORMULATION 7

Slug and surge capacity constraints are also appropriate for where the wetted perimeter is given by
the liquid interface level in the three-phase separator to avoid
water from flowing over the weir and avoiding oil escaping PHLL = Di cos −1 (1 − 2hHLL Di ) . ……………………..(21)
out the water outlet nozzle. Assuming that the liquid flow Viles23 suggest using the same formulation for both two- and
contains a constant ratio of water and oil during slugging and three-phase separators. The expression for urmax is different for
surging, the slug/surge capacity constraints can be formulated the various flow regimes and is for completeness presented
as: below (here presented in the units of m/s opposed to the
Vslug (q w qo ) ≤ (Lc + Le + 2d n ,w )( AHIL − ANIL ) ….…...(15) original units of ft/s).
For the low Reynolds number regime (NRef <160),
and Vsurge (q w qo ) ≤ (Lc + Le + 2d n , w )( ANIL − ALIL ) , ….….(16) (
u r max = 0.4572 σ og µ o ) (ρ o ρ g ) 0.5 N Re f
−0. 5
. ……...(22a)
where dn,w is defined in Fig. 8. For the transition regime (160 ≤ NRef ≤ 1635),
u r max = 3.5905(σ og µ o ) (ρ o ρ g ) N µ N Re f −
0.5 0.8 1/ 3
Slenderness and Liquid Re-Entrainment Constraint. ,
Slenderness is a structural relationship defined as the ratio
if N µ ≤ 1 15 , ……………………………………..(22b)
between the internal length and the internal diameter of the
and u r max = 0.4115(σ og µ o ) (ρ o ρ g ) N Re f
0. 5 −1 / 3
separator. Since an infinite set of separator length and ,
diameters yields feasible designs3,4, it is common to determine if N µ > 1 15 . ……………………………………..(22c)
the slenderness ratio based on experience and economic
considerations. The slenderness ratio is upper bounded by high For the rough turbulent regime (NRef >1635),
gas velocities that may induce liquid re-entrainment, i.e.,
u r max = 0.3048(σ og µ o ) (ρ o ρ g )
0.5 0.8
droplets breaking away from an unstable gas-oil interface. The Nµ ,
slenderness ratio is lower bounded by the geometry that if N µ ≤ 1 15 , ……………………………………..(22d)
promotes plug flow. Arnold and Steward3 reports that
and u r max = 0.03493(σ og µ o ) (ρ o ρ g )
0.5
slenderness ratios on the order of 3 to 5 are common, while ,
Svrcek and Monnery5,6 constrain their design to slenderness if N µ > 1 15 . ……………………………………..(22e)
ratios between 1.5 and 6. In their design procedures the
slenderness criteria is checked in the final step. If the The expressions for urmax will impose a set of 5 non-linear
slenderness criteria are violated the diameter is disjunctive constraints on the overall separator mathematical
increased/decreased and the design is recalculated until the program.
ratio is within the bounds. The need for a lower bound on the slenderness is
Ideally, the upper slenderness ratio should be a function of inherently rare. This is because the economic factors that drive
the separator operating point since re-entrainment forces vary the design will select the design with the smallest diameter
greatly with the physical properties of the fluids. Viles23 that will not support re-entrainment.
presents a procedure for predicting when re-entrainment is
possible in horizontal separators. His procedure is established Oil-Water Interface Control Constraint. The oil and water
based on analysis of experimental data24 and is translated into phases will normally flow with different velocities. This
a relationship that constrains the maximum horizontal velocity imposes a drag force on their interface. If this drag force is
difference between the gas and the oil phases in the gravity large the interface will be unstable and re-entrainment might
settling section according to occur. Circulation in either phase might also appear in case of
very large difference in velocity. However, in general their
qg qo effects are not considered in the three-phase separator design.
u r max ≥ u g − u o = − . ……...(17)
ATV − ANLL ANLL − ANIL Instead, the downstream oil and water cleaning systems are
relied upon to perform the finer separation. Ideally this
The urmax is defined through the Reynolds film number and an situation should be investigated and controlled as it might
interfacial viscosity number that characterize the two-phase reduce the sizes of the downstream units.
flow. These are respectively defined as:
N Re f = ρ o u o D H µ o .…………………………….…(18) Alarm and Trip Level Constraints. As mentioned earlier the
model will use the operational constraints from the
NORSOK22 standards. There are two constraints that must be
[
and N µ = µ o ρ oσ og (σ og (g (ρ o − ρ g ))) 0.5 ] 0.5
. ………...(19) simultaneously satisfied. The first constraint states that the
equivalent residence time between normal level and alarm
The general hydraulic diameter is most critical (gives lowest
level and between alarm level and trip level should not be less
urmax) if related to the high liquid interface and is defined as
than 30 seconds (∆tNOR). Thus, requiring the following 4
DH = 4 AHLL PHLL , ……………………...…………..(20) constraints related to the liquid level:
8 E.O. GRØDAL AND M.J. REALFF SPE 56645

(qo + q w )∆t NOR ≤ (Lc + Le + Ln ) (Ai − A j ) , ………(23a-d) hu − hv ≥ tan(α ) Lb 2 , …………………………...(27a-d)


where i and j are the sets {HHLL, HLL, NLL, LLL} and {HLL, where u and v are the sets {HLL, NLL, HIL, NIL} and {NLL,
NLL, LLL, LLLL}, respectively. The following 4 constraints LLL, NIL, LIL}, respectively.
related to the normal interface level in the three-phase
separator model are required: Inlet and End Section Design. We have chosen a cyclone as
( )
qw ∆t NOR ≤ (Lc + Le + 2d n ,w ) A p − Aq , …….…....(24a-d)
the momentum breaker and for coarse separation in the inlet
section. Many cyclones are available for this task. They often
where p and q are the sets {HHIL, HIL, NIL, LIL} and {HIL, consist of multiple cyclones each designed to take a fraction of
NIL, LIL, LLIL}, respectively. the inlet flow. The cyclone design will in general not rule the
The second constraint states that the height between optimal size of separators and we have chosen not to present a
normal level and alarm level and between alarm level and trip detail design procedure for this internal. In the design cases a
typical length, Lc, will be selected.
level should not be less than 0.08 m (∆hNOR). This can be
formulated as 4 constraints as follows The end section and its internals are depicted in Fig. 8. The
minimum length of the end section can be formulated as
( )
∆hNOR ≤ hi − h j , …………………….…..……...(25a-d)
Ln ≥ max[ Lvx , d n , g , 2d n, w + 2d n ,o + Lweir ] , ……….….(28)
and for the normal interface level in the three-phase model:
The minimum nozzle size is given by5 (here presented in
( )
∆hNOR ≤ h p − hq . …….…………………..…….(26a-d) SI-units, i.e., the coefficient has units: m0.25s0.5/kg0.25),
where i ,j, p and q are as defined in Eq. 23 and 24. d n ≥ 0.161 qρ 0.5 . ………………………………….(29)
Moving Process Plant Constraint. When separators are to be The water and oil outlets are mounted with vortex
placed on floating vessels, a major concern is the effect of breakers. These are designed according to the suggestions by
motions on the separation performance. Rice25 discusses the Patterson26 of being twice the size of the inner diameter of the
effects of the various degrees of movements. He states that the nozzles.
most critical movement is the separator pitch. Piching causes dn,g
the spirit level effect by which interfaces move to be Gas outlet
perpendicular to the gravitational force. This may greatly alter
the cross-sectional area for flow of the various phases.
Lvx
Installing perforated baffle plates may successfully control this Baffle plate
effect and others described by Rice.
In our model the inter-baffle baffle distance will be Lweir
constrained according to the geometry in Fig. 7 for the normal
liquid level and a similar geometry for the normal interface
level. The simplified design criterion is that no part of the 2dn,w 2dn,o
interfaces shall reach their high or low levels during maximum
pitch angle from initial normal level setting. We assume that Water Oil
the pitching does not disturb the instantaneous rates through outlet outlet
the baffle plates. Conservatively, the maximum tilted NLL is dn,w dn,o

simplified to be linear from its low level to its high level.


Fig. 8 – Details in three-phase separator end section.

Carpentier27 and Powers28 discuss the importance of mist


HLL Lb
eliminators, and agree that they greatly improve the separation
NLL and therefore should always be installed ahead of the gas
LLL Lb
α outlet. They also acknowledge that the superficial gas
velocities must be maintained between appropriate levels. At
α low gas velocities the impingement might be too low and at
high gas velocities re-entrainment is possible. However, they
g
do not quantify what appropriate levels are. Verlaan29 has
g looked at the effectiveness of vane demisters. We have chosen
to include and model a vane demister from his work (type
Fig. 7 – Model of interbaffle spaces at maximum pitch angle.
IIIa). This demister is designed for a top mounted gas outlet
nozzle as depicted in Fig. 9.
Thus, the effect of a pitching vessel is described by the The gas flows vertically upwards through a set of parallel
following 4 constraints: vanes of total height of 0.20 m (hvdo-hvdi). Verlaan found that
SPE 56645 OPTIMAL DESIGN OF TWO- AND THREE-PHASE SEPARATORS: A MATHEMATICAL PROGRAMMING FORMULATION 9

this specific design is effective for a wide range in gas velocity gas escaping out the oil outlet. This scenario is avoided with
and that both flooding and re-entrainment is controlled by the the following constraint,
maximum gas velocity defined by
hLLLL − hWH ≥ ∆hs , …………………………………..(33)
0.5
 ρo − ρg 
0.25
qg  σ og g 
u g max = ≤ 2.38    , ……(30) for both two- and three-phase separators.
  ρ 
Lvx Lvz  ρo   g  To avoid the water phase from flowing over the weir and
escaping out the oil outlet before shut down, there must be a
where Lvz for our horizontal separator is given by safety margin between the high-high interface level and the
weir height,
Lvz = 2 Di hvdm − hvdm . …………………………….(31)
2

hWH − hHHIL ≥ ∆hs , …………………………………..(34)

Gas outlet for the three-phase separator model.


Finally, to avoid the scenario of oil being sucked out the
water outlet a safety margin constraint can be written as
vdm
vdo hLLIL ≥ ∆hs , ……………………...…………………..(35)
vdi
Lvz for the three-phase separator model.

Geometrical and Logical Constraints. Because of handling


Di and transportation separators have some practical upper
overall size limits:
Fig. 9 – Cross-section through vane demister at gas outlet.
L = Lc + Le + Ln + 2 Lh + 2t c ≤ 20 m .…….………..(36)
Safety Constraints. To be confident the separator shuts down
before certain undesirable states are reached safety margins and D = Di + 2t c ≤ 4.5 m , ………………………………(37)
are introduced. The safety margins are here conveniently set
equal to ∆ hs, as depicted in Fig. 10 for both two- and three- where the chosen elliptical heads gives Lh=Di /4. We must also
phase separators. specify that the normal liquid level should be larger than the
To avoid the oil phase from escaping out the gas outlet normal interface level. This can be stated with the absolute
before a shut down, there must exist a minimum distance from minimum distance between the two normal levels
the inlet of the vane demister to the trip level, formulated as hNLL − hNIL ≥ 4∆h NOR + 2∆hs . ……………………….(38)
hvdi − hHHLL ≥ ∆hs , …………………………………..(32)
The weir height is placed between the high-high interface
for both the two- and three phase separator models. level and the low-low liquid level according to:
hWH ≥ hHHIL + ∆hs , …………………………….…….(39)
Two-phase separator Three-phase separator
Gas outlet Gas outlet
and hWH ≤ hLLLL − ∆hs . ……………….…………...…….(40)
h vdi hvdi
∆ hs ∆ hs
HHLL HHLL
Mechanical Design Equations. The oil industry is very safety
NLL conscious and approaches the mechanical design of separators
NLL LLLL using codified design rules. The ASME Code for Unfired
∆hs
∆ hs Pressure Vessels30 is widely used. Using the ASME code the
HHIL
wall thickness in the cylindrical section is
LLLL
∆ hs NIL
LLIL p D Di
∆hs t cs ≥ + t c .……………………………...(41)
2σE − 1.2 p D
Water vent Oil outlet Water outlet Oil outlet
where the design pressure is the operating pressure with either
Fig. 10 – Detail of safety levels in separator end sections. 2 bar added to it or increased by 10%, whichever is greater:
The separator is to be shut down if the gas-oil interface p D ≥ max[ p + 200000, 1.1 p ] . ………………………(42)
reaches the low-low liquid level. The shutdown is to avoid gas
The allowable stress is a safety factor towards the material
phase escaping out the oil outlet. This might happen when the
tensile strength. ASME suggests a safety factor of 4. The
gas-oil interface approaches the weir height leaving marginal
tensile strength for the carbon steel commonly used for
flow over the weir. In this situation the liquid level on the oil
separators is 3800 bar (55000psi), i.e., σ = 950 bar. The joint
outlet side of the weir might fall rapidly and eventually having
10 E.O. GRØDAL AND M.J. REALFF SPE 56645

efficiency ranges from 0.6 to 1.0 for 100% X-rayed joints. The the stated constraints can by logical reasoning be found to
corrosion allowance typically ranges from 1.5 mm to 3.2 mm. always be active at the optimal solution for a specific
Many different designs of heads exist. Elliptical heads objective function (we have chosen Eq. 44). Thus, they can be
(2:1) provides an economic design with respect to material stated as equalities rather than inequalities. The equalities can
consumption because the code states that the overall vessel off course be kept inside the mathematical program (i.e., Eq.
wall thickness should be the larger of the cylindrical section 1), however, they can also be removed from the constraint list
and head. Ellipsoidal heads require the thickness and be used ahead of the mathematical program. This
approach will reduce the number of degrees of freedom, as it
p D Di
t eh ≥ + t c , ………………….….………(43) will fix one variable as a function of the remaining variables in
2σE − 0.2 p D that equation. This approach will speed up the computation for
two reasons. First, a reduction of the number of design
which is less than the wall thickness of the cylindrical section.
variables will reduce the search space. Second, removing
constraints will reduce the size of the Hessian matrix, which is
The Solution Algorithm and Mathematical Program
calculated for each SQP-iteration.
We propose the solution algorithm and mathematical program
The input to the algorithms has been divided into four
depicted in Figs. 11 and 12 for the design of a three- and two-
groups. The design parameters consist of the case specific data
phase separator, respectively. The algorithms consist of three
such as flow rates, physical properties and the specified
sections; input, SQP-loop and the output section. The SQP-
separation degree. The physical constants and fixed variables
loop consists of two segments that are constructed to reduce
consist of material, construction and safety constants and
the number of design variables, which is explained below. We
constants from the chosen standard. The fixed design variables
have reduced the problem to 4 and 3 design variables for
are the mounting point of the vane demister relative to the gas
three- and two-phase separators, respectively. These are: Di,
outlet, the thickness of the weir plate, and the length of the
Le, hNLL, and hNIL, without the latter for the two-phase
section containing the cyclone, as a detailed design of this
separator.
internal is omitted here. The third group of input consists of
parameters that are independent of the four selected design
INPUT SQP-LOOP variables, but dependent on the design parameters. The final
Design Parameters Calculate Design Variables group of inputs is the initial guesses of the four design
dog, dow, dwo, qg, qo, qw, hHLL (max) by Eqs. 13, 23b, 25b variables. In general, the closer the initial guess is to the
Vslug, Vsurge, α, σog, hHHLL (max) by Eqs. 23a, 25a solution, the faster a solution is found. However, the algorithm
µg, µo, µw, ρg, ρo, ρw. hLLL (min) by Eqs. 14, 23c, 25c will converge to the optimal design from any feasible guess.
hLLLL (min) by Eqs. 23d, 25d
Physical Constants hHIL (max) by Eqs. 15, 24b, 26b
and Fixed Variables hHHIL (max) by Eqs. 24a, 26a INPUT SQP-LOOP
E, Fa, Fc, Fh, hvdm, hvdo, hLIL (min) by Eqs. 16, 24c, 26c Design Parameters Calculate Design Variables
Lc, Lweir, tc, σ, ρs, hLLIL (min) by Eqs. 24d, 26d dog, dgo, qg, qo, hHLL (max) by Eqs. 23b, 25b
∆hNOR, ∆hs, ∆tNOR. hWH by Eq. 39, tcs by Eq. 41 α, σog, µg, µo, ρg, ρo, hHHLL (max) by Eqs. 23a, 25a
Lb (min) by Eqs. 27a-d hLLL (min) by Eqs. 23c, 25c
Calculated constants
Ln by Eq. 28, Lvz by Eq. 31 Physical Constants hLLLL (min) by Eqs. 23d, 25d
uog, uow, uwo by Eqs. 2-4
and Fixed Variables Lb (min) by Eqs. 27a-b
dn,g, dn,o, dn,w by Eq. 29
E, Fa, Fc, Fh, hvdm, hvdo, tcs by Eq. 41
LvxLvz by Eq. 30
hWH, Lc, Lweir, tc, σ, ρs, Ln by Eq. 28, Lvz by Eq. 31
PD by Eq. 42
Mathematical Program dn,w, ∆hNOR, ∆hs, ∆tNOR
Initial Guesses of min Eq. 44 (cost function)
Calculated constants
Design Variables subject to:
uog, ugo by Eqs. 2-4
Di, Le, hNLL, hNIL Eq. 9 (gas capacity)
dn,g, dn,o by Eq. 29,
Eq. 10 (oil capacity)
LvxLvz by Eq. 30 Mathematical Program
Eq. 12 (water capacity)
PD by Eq. 42 min Eq. 44 (cost function)
Eq. 17 (re-entrainment)
Eq. 32 (oil out gas outlet) subject to:
Initial Guesses of
Eq. 33 (gas out oil outlet) Eq. 9 (gas capacity)
Design Variables
Eq. 35 (oil out water outlet) Eq. 11 (oil capacity)
OUTPUT Di, Le, hNLL
Eq. 38 (normal levels apart) Eq. 17 (re-entrainment)
Eq. 36 (max separator length) Eq. 32 (oil out gas outlet)
OUTPUT Eq. 33 (gas out oil outlet)
Separator Design Eq. 37 (max separator diameter)
Eq. 46a-d (positive variables) Eq. 36 (max separator length)
Separator Design Eq. 37 (max separator diameter)
Eq. 46a-c (positive variables)
Fig. 11 – Solution algorithm for 3-phase separator model.

Fig. 12 – Solution algorithm for 2-phase separator model.


As formulated in the previous sections the separator
designs are functions of many variables. However, many of
SPE 56645 OPTIMAL DESIGN OF TWO- AND THREE-PHASE SEPARATORS: A MATHEMATICAL PROGRAMMING FORMULATION 11

The SQP-loop consists of two segments. The first segment Three-Phase Separator Case Study
calculates design variables that are fixed by the four chosen A three-phase separator is to be designed for a production rate
degrees of freedom. In this segment the control and alarm of 200 kg/s of hydrocarbons and 40 kg/s water. The
levels will be set as close as feasible to their normal levels hydrocarbon composition is taken from McCain’s31 black oil
along with the other inequalities that have been identified to reservoir fluid study. At stock tank conditions this will yield
always be active at optimal design. For example, to find hHLL about 110,000 bbl/day. The inlet three-phase separator is to
one must determine which of the Eqs. 13, 23b and 25b gives operate at 20 bara and 350 K. The flash is calculated by the
the highest value for the values of the design variables in each equilibrium constant (K-value) predictor of Varotsis32. The
SQP-iteration. For Eqs. 13 and 23b this requires a numerical physical properties of the fluids at the operating point are
scheme as hHLL must be calculated based on AHLL and Di. We calculated by correlations in McCain31. The separator shall be
accomplished this with a Newton-Raphson technique10 on Eq. designed to separate 100 micron oil droplets in the gas phase,
B.2. When hHLL is established one will find hHHLL with a 250 micron water droplets from the oil phase, and 250 micron
similar approach. The second segment in the SQP-loop oil droplets from the water phase. The separator shall be
consists of the mathematical program. As a mathematical baffled for 10 degree pitch angle and have capacity for slugs
program performs its search over an objective function, we and surges of 10 m3. The input values are summarized in Tab.
find it reasonable to let the solver minimize a manufacturing 1. The physical values and the fixed variables are listed
separator vessel cost function. Companies are likely to have alphabetically in Tab. 2.
different cost functions. However, applying the structure of a
cost function from Powers13 we obtain Tab. 1 – Design parameters to 3-phase separator case study.
3
ρg = 17.46 kg/m
-6 3
dog = 100×10 m Vslug = 10.0 m
C = t cs Fc ρ s [πDm (Lc + Le + Ln ) + 2Fa Fh Dm 2 ] , .….….(44)
3
ρo = 767.7 kg/m
-6 3
dwo = 250×10 m Vsurge = 10.0 m
α = 10.0 degrees ρw = 974.6 kg/m
-6 3
dow = 250×10 m
3
for our separator shell size where qg = 1.501 m /s µ g = 1.07×10 Pas
-5
σog = 1.78 × 10 kg/s
-2 2
3
µ o = 7.30×10 Pas
-4
qo = 0.226 m /s
Di + (Di + 2t cs )
2 2 3
µ w = 3.70×10 Pas
-4
qw = 0.041 m /s
Dm = . ……………………….....(45)
2 Tab. 2 – Physical constants and fixed variables for case studies.
The costs of the internals are only weak functions of the E = 1.0 hvdo = Di – 0.2 [m] ∆hNOR = 0.08 m
chosen design variables as their design and size is largely Fa = 1.09 Lc = 1.0 m ∆hs = 0.05 m
Fc =5.0 $/kg Lweir = 0.01 m ∆tNOR = 30 sec.
determined based on the various constant rates. Thus, we have
ρs = 7850 kg/m
3
Fh =3.0 tc = 0.0032 m
chosen to not consider the cost of the internals. This means hvdm = Di - 0.1 [m] σ = 950×10 Pa
5

that the internals are not directly subject to optimization. They


will merely adjust their shape according to what will yield the The algorithm used 8 seconds and 30 SQP-iterations to
less costly separator vessel. For example, the rectangular area find the optimal solution on a Pentium I, 200 MHz processor.
of the vane demister (LvxLvz) is fixed by parameters The solution converged to Di = 2.410 m, Le = 16.55 m, hNLL =
independent of the design variables (Eq. 30 as an equation). 1.386 m and hNIL = 0.378 m The initial guess was Di = 4.0 m,
However, the rectangular shape is a function of Ln and Di in Le = 15 m, hNLL = 2.0 m and hNIL = 1.0 m. The solution is
the objective function, i.e., our assumption is that it is the summarized alphabetically in Tab. 3 and the constraint status
demister area that bears the cost and not the shape of the area. is listed in Tab. 4.
The specified SQP-loop design has reduced the number of
constraints in the mathematical program according to Figs. 11 Tab. 3 – Overall solution to 3-phase separator design case.
and 12. Although minimum values do exist for the design C = 225480 $ hLLL = 1.164 m PD = 22.0×10 Pa
5

variables it is normally wise to include constraints that state Cog = 1.72 hLLLL = 0.985 m tcs = 0.032 m
that they are positive, thus: Cow = 3.54 hNIL = 0.378 m tr,g = 20.4 sec.
Cwo = 13.8 hNLL = 1.386 m tr,o = 166 sec.
Di , Le , hNLL , hNIL ≥ 0 . ……………………………(46a-d) D = 2.474 m hWH = 0.588 m tr,w = 185 sec.
DH = 2.812 m L = 20.0 m tog = 6.8 sec.
The SQP-solver will iterate in the SQP-loop by suggesting Di = 2.410 m Lb = 0.907 m tow = 36 sec.
new values for the design variables until convergence is Dmean= 2.440 m Le = 16.55 m two = 166 sec.
reached. dn,g = 0.403 m Lh = 0.635 m ug = 0.812 m/s
dn,o = 0.403 m Ln = 1.181 m uo = 0.100 m/s
The output section consist of the optimal separator design dn,w = 0.182 m Lvx = 0.815 m uw = 0.0895 m/s
including values of all the used parameters, the objective hHHIL = 0.538 m Lvz = 0.961 m uog = 0.182 m/s
function value and the status of the constraints at the solution. hHHLL = 1.811 m NRef = 2.97×10
5
uow = 0.0128 m/s
To illustrate the method and models, a three- and a two-phase hHIL = 0.458 m NRe,og = 29.7 uwo = 0.00798 m/s
separator design case study are presented in the following two hHLL = 1.614 m NRe,ow = 8.43 ugmax = 1.92 m/s
hLIL = 0.298 m NRe,wo = 2.10 urmax = 0.712 m/s
sections. The two separators are connected in series, i.e., the
hLLIL = 0.218 m Nµ = 0.0050
input to the two-phase separator is the oil output from the
three-phase separator.
12 E.O. GRØDAL AND M.J. REALFF SPE 56645

The SQP-software33 requires the constraints to be This observation is not universally true as it is dependent on
rearranged as follows: the specific design and of the normal levels’ positions. From
g i (Di , Le , hNLL , hNIL ) ≤ 0 . …………………………...(47)
this discussion it should be evident that there does not exist a
feasible separator design with a smaller diameter.
Thus, the constraint status (surplus) as listed in Tab. 4 is the The optimality of the solution can be further investigated
value of gi at the solution. We observe that constraint number by constructing the feasible design space (FDS) as depicted in
2, 4 and 9 are constraining the design. These are the oil Fig. 13. The lower bound in the FDS was constructed by
capacity, re-entrainment and maximum separator length letting the algorithm run in a loop over a set of fixed values of
constraints. We note that both the gas capacity and water Di while finding the optimal values for Le, hNLL and hNIL. The
capacity are far from constraining as both have a surplus in upper bounds were constructed by a similar procedure,
gravity settling length of more than 10 meters. This means that maximizing the objective function instead of minimizing it.
at this diameter the effective gravity settling section could The optimal point is indicated as the design with the smallest
have been reduced by at least 10 meters and the specified size diameter in the FDS as this is the least costly design in the
of oil droplets in gas and in water would have been separated FDS. The relative cost to the optimal design of the designs on
according to settling theory. The result is that the separation in the lower bound is shown graphically and increases with the
the gas and water phases will be better than specified at diameter (thick arrows). For example, as shown by the dotted
optimal solution , 46 micron and 95 micron oil droplets in gas lines, choosing a design of Di = 3.5 m and Le = 9.54 m results
phase and water phase respectively. However, we cannot infer in a separator that is 70% more expensive than the optimal
that the separator design can accept a higher gas rate. A higher design.
gas rate would increase the relative velocity between the gas
and the oil, which would violate the active re-entrainment Tab. 4 -- Constraint status at optimal 3-phase separator design.
Constraint # Status Constraint # Status
constraint. On the other hand, the water rate can be (1) Eq. 9 -10.98 m (8) Eq. 38 -0.585 m
significantly increased, as this will not violate any other (2) Eq. 10 0.000 m (9) Eq. 36 0.000 m
constraints. This result, a separator with higher rating than (3) Eq. 12 -13.36 m (10) Eq. 37 -2.029 m
was specified, is interesting: how can this be the optimal (4) Eq. 17 0.000 m/s (11) Eq. 46a -2.408 m
design? Why cannot the oil phase use some of the available (5) Eq. 32 -0.150 m (12) Eq. 46b -16.55 m
(6) Eq. 33 -0.344 m (13) Eq. 46c -1.383 m
flow area of the water phase? The explanation for this
(7) Eq. 35 -0.168 m (14) Eq. 46d -0.378 m
seemingly peculiar result is found by considering the
maximum settling heights3 in the settling theory. The equation 18 8
Gravity settling section length [ Le, m]

Relative Cost [rel. to optimal design]


for oil settling, Eq. 10, can be written as
hHLL − hLIL ≤ Le u wo ( ANLL − ANIL ) q o , ………………(48)
16 7
Optimal Feasible
14 6
Design Design
which illustrates that there exist a maximum height for the oil
12 Space 5
pad thickness at the active constraint. In our case, uwo and qo
are constants and Le is at its maximum value constrained by 10 4
the active constraint, Eq. 36. Thus, the only approach to Relative cost for
increase the maximum oil pad thickness is to adjust the levels. 8 design on lower 3
constraint
Increasing ANLL by raising hNLL is not a feasible option as this 6 2
is constrained by the active re-entrainment constraint. The
final option is to reduce ANIL by lowering hNIL, i.e., using some 4 1
of the excess flow area of the water phase. However, this will 2 2.5 3 3.5 4 4.5
Separator inner diameter [Di, m]
also fail. The extra area provided by lowering hNIL is not large
enough to reduce the oil velocity so that the water droplets can Fig. 13 – Feasible design space of 3-phase separator case study.
settle to the new hLIL, noting that hLIL is dependent on hNIL
through Eqs. 16, 24c or 26c, whichever gives the lowest value. It is interesting to note from Tab. 3 is that the settling of
Mathematically this can be proven by partially differentiating water or oil droplets in the liquid phases is not in the laminar
both sides of Eq. 48 with respect to hNIL at the solution. This region and thus Stokes law (applicable for Nre < 1) could not
will prove that the left-hand side grows faster than the right- have been used, which is often assumed for this settling. The
hand side at the solution, i.e., violating the active constraint. baffle plates have to be installed for every 0.907 m, which
Since Eq. 48 can be written as means that the separator would require a total of 19 baffle
plates. This is quite a large number and would represent a
hHLL − hLIL ≤ u wo t r ,o , ………………………………...(49) significant cost. This is the result of severe pitching and the
fact that the cost of the internals was not subject to
where tr,o is the oil retention time in the gravity settling optimization. However, including the baffle cost in the
section, one arrives at the conclusion that providing higher objective function and keeping Eq. 27 as an inequality and
retention times does not yield better separation. Which is in introducing Lb as a design freedom, the number of baffle
direct conflict with the retention time theory in Appendix A. plates can be optimized. If a lower number is found to be
SPE 56645 OPTIMAL DESIGN OF TWO- AND THREE-PHASE SEPARATORS: A MATHEMATICAL PROGRAMMING FORMULATION 13

optimal from and overall cost perspective, it will result in a The feasible design space has been constructed in a similar
larger separator diameter as the distance between the normal procedure as for the three-phase separator and is depicted in
levels and the alarm level needs to be larger. Fig. 14. Calculating the manufacturing cost along the lower
A general comment on the solution found by the algorithm bound will again illustrate a significant increase as the
and the selected software is that the same solution is arrived at diameter increases (thick arrow). For example, as illustrated
for an extensive set of randomly chosen initial guesses. by dotted lines, selecting the least expensive separator with a
diameter of 3.0 meters results in a design that is 56% more
Two-Phase Separator Case Study expensive than the optimal design. The FDS illustrates that
The two-phase separator is fed the oil output from the three- there exists a minimum feasible diameter where an increase in
phase separator and is to operate at 1.5 bara and 340K. the gravity settling section does not deactivate the re-
Flashed at this operating point, the rates and physical entrainment constraint. In contrast to the three-phase
properties are as listed in Tab. 5, along with the separation separator, the two-phase separator has an interbaffle distance
degree of 100 micron oil droplets in the gas phase and 250 of 2.73 m, which results in 4 baffle plates for the separator
micron gas bubbles in the oil phase. The physical values and (same pitch design angle for both separators). This is because
fixed variables are as for the three-phase separator design case the minimum distances between the normal levels and the
(Tab. 2). alarm levels are larger for the two-phase separator in the
optimal design.
Tab. 5 – Design parameters to 2-phase separator case study.
α = 10.0 degrees ρg = 2.417 kg/m
-6 3
dog = 100×10 m 18

Relative Cost [rel. to optimal design]


Gravity settling section length [ Le, m]
µ g = 6.43×10 Pas ρo = 795.2 kg/m
-6 -6 3
dgo = 250×10 m 13
3 16
µ o = 1.31×10 Pas σog = 2.18 × 10 kg/s
-3 -2 2
qg = 3.997 m /s Optimal
3 14
qo = 0.206 m /s Design 11
Feasible
12 Design 9
The algorithm used 4 seconds and 30 SQP-iterations to 10 Space
converge to the solution; Di = 1.998 m, Le = 10.95 m and 7
8
hNLL = 0.806 m. The initial guess was; Di = 4.0 m, Le = 15 m
6
and hNLL = 2.0 m. The solution is summarized alphabetically in 5
Tab. 6 and the constraint status is listed in Tab. 7. 4 Relative cost for
design on lower 3
2
constraint
Tab. 6 – Overall solution to 2-phase separator design case. 0 1
C = 29146 $ hLLLL = 0.250 m Nµ = 0.0077 1 1.5 2 2.5 3 3.5 4 4.5 5
5
Cog = 2.715 hNLL = 0.806 m PD = 3.5×10 Pa
Separator inner diameter [Di, m]
Cgo = 133.0 hWH = 0.200 m tcs = 0.0069 m
D = 2.012 m L = 13.951 m tr,g = 5.3 sec. Fig. 14 – Feasible design space of 2-phase separator case study.
DH = 2.056 m Lb = 2.731 m tr,o = 62.9 sec.
Di = 1.998 m Le = 10.951 m tog = 3.6 sec. Conclusions
Dmean = 2.00 m Lh = 1.006 m tgo = 62.9 sec. A fully automated, systematic, design method for the physical
dn,g = 0.401 m Ln = 0.987 m ug = 2.048 m/s
design of two-phase and three-phase primary separators has
dn,o = 0.388 m Lvx = 0.871 m uo = 0.174 m/s
dn,w = 0.100 m Lvz = 0.831 m uog = 0.399 m/s been developed. The model is based on settling theory and is
hHHLL = 1.290 m NRef = 2.171×10
5
ugo = 0.0167 m/s solved through the application of mathematical programming
hHLL = 1.046 m NRe,og = 15.0 ugmax = 5.52 m/s techniques. The suggested algorithm and solver finds the
hLLL = 0.552 m NRe,go = 0.191 urmax = 1.87 m/s optimal design for a series of stated objectives.
The application of mathematical programming techniques
Tab. 7 – Constraint status at optimal 2-phase separator design.
Constraint # Status Constraint # Status
to the separator design problem results in a convenient
(1) Eq. 9 -3.519 m (6) Eq. 36 -6.049 m environment for analyzing the separator design problem. The
(2) Eq. 11 0.000 m (7) Eq. 37 -2.488 m declarative form of the model provides a forum in which to
(3) Eq. 17 0.000 m/s (8) Eq. 46a -1.998 m pose and discuss problems. The model solution times are fast,
(4) Eq. 32 -0.258 m (9) Eq. 46b -10.95 m on the order of seconds, and the abstraction of the solution
(5) Eq. 33 0.000 m (10) Eq. 46c -0.806 m
technology from the model means that advances in solution
methods can directly be used in the overall solution scheme.
We observe that constraints 2, 3 and 5 are constraining the
design. These are the oil capacity, the re-entrainment and the Acknowledgments
constraint that avoids having gas being sucked out the oil The authors gratefully acknowledge the financial support of
outlet. As for the three-phase separator, the gas capacity Kværner ASA, Technology Development and the Norwegian
constraint is not constraining, but the gas velocity is Research Council.
constrained by the re-entrainment constraint. Thus, a design
with a smaller diameter is infeasible even for designs with
longer gravity settling sections.
14 E.O. GRØDAL AND M.J. REALFF SPE 56645

Nomenclature ∆hNOR = height interval NORSOK standard, L, 0.08 m


A= cross-sectional area of separator between BV to ∆hs = safety margin height, L, m
level indicated by the subscript, L2, m2
∆t NOE = time interval NORSOK standard, t, 30 sec
C= separator manufacturing cost, M, $
CD= spherical drag coefficient, dimensionless
D= total separator diameter, L, m Subscripts
DH= hydraulic diameter, L, m BV= bottom vessel
Di = inner separator diameter, L, m g= gas
Dm= mean separator dimater, L, m go= gas bubble in oil phase
d= droplet/bubble diameter, L, m HHIL= high-high interface level
E= joint efficiency, dimensionless HHLL= high-high liquid level
Fa= factor for determining surface area of a vessel head HIL= high interface level
from vessel diameter squared ( ≈ 1.09 for 2:1 HLL= high liquid level
elliptical heads)30, dimensionless i= the set {HHLL, HLL, NLL, LLL}
Fc= cost factor per unit mass to manufacture a vessel j= the set {HLL, NLL, LLL, LLLL}
shell, M/m, $/kg LIL= low interface level
Fh= ratio of cost per unit mass to manufacture a vessel LLIL= low-low interface level
head compared with that of vessel shell (usually LLL= low liquid level
1.5 to 3.0)13, dimensionless LLLL= low-low liquid level
g= gravitational acceleration, L/t2, 9.81 m/sec2 NIL= normal interface level
h= height from bottom of separator, L, m NLL= normal liquid level
L= length, L, m o= oil
Lb= interbaffle distance, L, m og= oil droplet in gas phase
Lc= length of separator section with cyclone, L, m ow= oil droplet in water phase
Le= length of gravity separation section, L, m p= the set {HHIL, HIL, NIL, LIL}
Lh= separator head section length, L, m q= the set {HIL, NIL, LIL, LLIL}
Ln= length of separator end section, L, m s= steel
Lvx= length of vane demister, L, m TV= top vessel
Lvz= width of vane demister, L, m u= the set {HLL, NLL, HIL, NIL}
NRe= droplet or bubble Reynolds number, dimensionless v= the set {NLL, LLL, NIL, LIL}
NRef= Reynolds film number, dimensionless vdi= vane demister inlet
N µ = interfacial viscosity number, dimensionless vdm= vane demister mounting point
vdo= vane demister outlet
P= wetted perimeter, L, m WH= weir
p= operating pressure, m/Lt2, Pa w= water
pD= design pressure, m/Lt2, Pa wo= water droplet in oil phase
q= volumetric rate, L3/t, m3/sec
t= time, t, sec References
tc= corrosion allowance, L, m 1. Arnold, K. and Stewart, M. Jr.: “Designing Oil and Gas
tcs= wall thickness of cylindrical section, L, m Production Systems: How To Size and Select Two-Phase
teh= wall thickness of elliptical (2:1) heads, L, m Separators,” World Oil (Nov. 1984) 73.
tr= retention time in Le, t, sec 2. Arnold, K. and Stewart, M. Jr.: “Designing Oil and Gas
u= velocity, L/t, m/sec Production Systems: How To Size and Select Three-Phase
ugmax= maximum gas velocity ahead of vane demister, L/t, Separators,” World Oil (Dec. 1984) 87.
m/sec 3. Arnold K., and Stewart, M. Jr.: Surface Production Operations:
Design of Oil-Handling Systems and Facilities, Volume 1, Gulf
urmax= maximum relative velocity between gas and oil
Publishing Company, Houston, TX, USA (1995).
phase, L/t, m/sec 4. Bradley, H.B.; “Petroleum Engineering Handbook,” (Editor in
Vslug= slug volume, L3, m3 Chief), Third Printing, Society of Petroleum Engineers,
Vsurge= surge volume, L3, m3 Richardson, TX, USA (1992).
5. Svrcek, W.Y. and Monnery, W.D.: “Design Two-Phase
Greek Letters Separators Within the Right Limits,” Chemical Engineering
α = pitch degree, degrees Progress, (Oct. 1993) 53.
µ = dynamic viscosity, m/Lt, kg/msec 6. Monnery, W.D. and Svrcek, WY,: “Successfully Specify Three-
Phase Separators,” Chemical Engineering Progress, (Sep. 1994)
ρ = density, m/L3, kg/m3 29.
σ = tensile strenght, m/Lt2, Pa 7. Grødal, E.O., Realff, M.J. and Schug, B.: “A Product
σ og = surface tension between oil and gas, m/t2, kg/sec2 Positioning Approach for Optimal Product Lines in the
Hydrocarbon Production Separation Segment,” paper 217c
SPE 56645 OPTIMAL DESIGN OF TWO- AND THREE-PHASE SEPARATORS: A MATHEMATICAL PROGRAMMING FORMULATION 15

presented at the 1998 AIChE Annual Meeting of American 28. Powers, M.L.: “Author’s Reply to Discussion of New
Institute of Chemical Engineers, Miami Beach, FL, USA, Nov. Perspective on Oil and Gas Separator Performance,” SPEPF,
15-20. (August 1993), 223.
8. Grødal, E.O. and Realff, M.J.: “Optimal Lifetime Design of 29. Verlaan. C.: “Performance of Novel Mist Eliminators,” PhD
Primary Hydrocarbon Separation Systems,” paper SPE 56639, dissertation, Delft University of Technology, Delft, (1991).
to be presented at the 1999 SPE Annual Technical Conference 30. American Society of Mechanical Engineers, ASME Boiler and
and Exhibition, Houston, TX, USA, Oct. 3-6. Pressure Vessel Code, Sec. VIII, Div.1 (1986).
9. Biegler, L.T., Grossmann, I.G. and Weterberg, A.W.: Systematic 31. McCain, Jr. W.D.: The Properties of Petroleum Fluids, second
Methods of Chemical Process Design, Prentice Hall PTR, Upper edition, PennWell Publishing Company, Tulsa, Oklahoma, USA
Saddle River, NJ, USA (1997). (1990).
10. Winston, W.L.: Operations Research: Applications and 32. Varotsis, N.: “A Robust Prediction Method for Rapid Phase-
Algorithms, third edition, Duxburry Press, International Behavior Calculations,” SPERE, (1989), 237.
Thompson Press, Belmont, CA USA (1994). 33. MATLAB with Optimization Toolbox, Version 5.2.0.3084, Jan.
11. Han, S.P.: “Superlinearly Convergent Variable Metric 17, 1998. The MathWorks, Inc., Natick, MA, USA.
Algorithm for General Nonlinear Programming Problems,” http://www.mathworks.com.
Math. Progr., (1976) 11, 263-282.
12. Powell, M.J.D.: “A Fast Algorithm for Nonlinearly Constrained Appendix A
Optimization Calculations,” In Numerical Analysis, Dundee, The solution algorithms in Fig. 11 and 12 apply settling theory
1977, G.A. Watson (Ed.), Lecture notes in Mathematics 630, for separation in all phases. These algorithm can be modified
Springer-Verlag, Berlin, Germany, (1978). to apply retention time theory given by the following three
13. Powers, M.L.: “New Perspective on Oil and Gas Separator
constraints:
Performance,” SPEPF, (May 1993) 77.
14. Arnold, K. and Koszela, P.L.: “Droplet-Settling vs. Retention- Le ≥ q g t r , g ( ATV − ANLL ) , ………………………….(A.1)
Time theories for Sizing Oil/Water Separator,” SPEPE, (Feb.
1990), 59.
15. Perry, R.H. and Green, D.W. (eds.): Perry’s Chemical Le ≥ qo t r ,o ( ANLL − ANIL ) , …………………………(A.2)
Engineers’ Handbook, seventh edition, McGraw Hill Book Co.
Inc., New York City, NY USA (1997). and Le ≥ q w t r ,w ANIL , …………………………….……(A.3)
16. Hafskjold, B., Morrow, T.B., Celius, H.K.B. and Johnson, D.R.,
“Drop-drop coalescence in oil/water separation,” paper SPE for the gas, oil and water phases, respectively. Thus, Eq. 9, 10
28536 presented at the 1994 SPE Annual Technical Conference and 12 in the constraint list of the mathematical program must
and Exhibition, New Orleans, LA, Sept. 25-28. be replaced by Eq. .A.1, A.2 and A.3, respectively for the
17. Kassa, O.: “SYNENERGI – Production Chemicals and Process
three-phase separator. For the two-phase separator, Eq. 9 and
Design,” paper SPE 30434 presented at the 1995 SPE Offshore
Europe Conference, Aberdeen, Scotland, Sept. 5-8. 11 must be replaced by A.1 and A.2, respectively, where ANIL
18. Asheim, H.: “Petroleumsproduksjon og Prosessering på in Eq. A.2 is set equal to zero. A mix of the two theories can
Plattformen, Tano A/S, Oslo, Norway, (1985). also be used, e.g., settling theory for the gas phase and
19. Grødal, E.O.: “Numerical Studies of Flow in Oilfield Separators retention time theory for the liquid phases.
and Their Internals,” unpublished, Georgia Institute of
Technology, June, 1998. Appendix B
20. Flanigan, D.A., Stolhand, J.E., Scribner, M.E. and Shimoda, E.: The functional relationship between the shaded area in Fig.
“Droplet Size Analysis: A New Tool for Improving Oilfield B.1, the diameter and an arbitrarily placed horizontal level h1
Separations,” paper SPE 18204 presented at the 1988 SPE
can be found by integration.
Annual Technical Conference and Exhibition, Houston, TX,
Oct. 2-5.
21. Kosberg, J.A. and Reid, D.: “Design of Primary Sepration h
Systems for FPSO’s,” presented at the 1996 FPSO-Norway, The
OCS Technology Group.
22. NORSOK standard,: “P-100 Process systems, 20 Separation and h1
Stabilization,” The competitive standing of the Norwegian Di
offshore sector, rev. 1, Aug. 1997. (http://www.nts.no/norsok/).
23. Viles, J.C.: “Predicting Liquid Re-Entrainment in Horizontal
Separators,” JPT, (May 1993) 405.
24. Ishii, M. and Grolmes, M.A.: “Inception Criteria for Droplet x
Entrainment in Two-Phase Concurrent Film Flow,” AIChE J. Fig. B.1 – Cross-section of horizontal separator.
(March 1975) 308.
25. Rice, C.L.: “Effects on Motion on Design of Process Facilities
for Floating Production Systems,” Production Facilities,
The function describing the circle in Fig. B.1 is given by
Reprint Series, SPE, Richardson, TX (1989) 25, 110-116
26. Patterson, F.M.: “Vortexing can be prevented,” The Oil and Gas x = ± Di h − h 2 . ……………………….………..…(B.1)
Journal, August 4, 1969.
27. Carpentier, P.L.: “Discussion of New Perspective on Oil and
Gas Separator Performance,” SPEPF, (August 1993) 222.
16 E.O. GRØDAL AND M.J. REALFF SPE 56645

Acknowledging the symmetry, twice the integrated positive


part of x from h=0 to h=h1 yields the following function for
the shaded area,

 D  D
2
 2h  πD 2
A =  h1 − i  Di h1 − h12 + i sin −1  1 − 1 + i .
 2  4  Di  8
………..…..(B.2)
For example, finding AHLL (A in Eq. B.2) requires the
substitution of h1 by hHLL. Finding h1 as a function of A and Di
requires a numerical scheme. (e.g., Newton-Raphson
technique10.

Conversion Factors
bar × 1.450 377 E+01 = psi
K × 1.8* E+00 = °R
kg × 2.204 623 E+00 = lbm
m × 3.280 840 E+00 = ft
m2 × 1.076 391 E+01 = ft2
m3 × 3.531 467 E+01 = ft3
Pa × 1.450 377 E-04 =psi
*
Conversion factor is exact

Vous aimerez peut-être aussi