Computational Fluid Dynamics-Based

Study of an Oilfield Separator—

Part I: A Realistic Simulation
Ali Pourahmadi Laleh and William Y. Svrcek, University of Calgary, and Wayne D. Monnery, Chem-Pet Process Tech Limited

Summary scale three-phase separator. The separator of interest, with diameter

A realistic computational fluid dynamics (CFD) simulation of a of 3.33 m and length of 16.30 m, was the first stage of the three-
field three-phase separator has been developed. This realistic CFD stage-dual-train production process installed on Gullfaks-A off-
simulation provides an understanding of both the microscopic and shore platform. The production on the Gullfaks-A platform started
macroscopic features of the three-phase separation phenomenon. successfully in 1986–87. However, because of a projected increase
For simulation purposes, an efficient combination of two mul- in the water production, several separation inefficiencies such as
tiphase models of the commercial CFD software, Fluent 6.3.26 water-level control, emulsion problems, and increased impurities
(ANSYS 2006a), was implemented. The flow-distributing baffles were experienced in the following years. In order to develop an in-
and wire mesh demister were also modeled using the porous media depth understanding of this complex three-phase separation pro-
model. Furthermore, a useful approach to estimating the particle cess, Hansen et al. (1993) developed a CFD model of the separator.
size distribution in oilfield separators was developed. The simu- Because of the problem scale and importance, and also because
lated fluid-flow profiles are realistic and the predicted separation almost all of the operating and physical parameters required for
efficiencies are consistent with oilfield experience. CFD simulations have been provided by Hansen et al. (1993), this
significant case was selected for comprehensive CFD studies in
Introduction the present work. Hallanger et al. (1996) developed a CFD model
Once a crude oil has reached the surface, it must be processed for a three-phase (free gas, oil+dispersed water, and free water)
so that it can be sent either to storage or to a refinery for fur- separator by extension of the two-fluid model. The mixture model
ther processing. In fact, the main purpose of the surface facilities was used for modeling the oil phase while the water droplets were
is to separate the produced multiphase stream into its vapor and distributed through different droplet size classes. The momentum
liquid fractions. On production platforms, a multiphase separator equation for the mixture phase together with the continuity equa-
is usually the first equipment through which the well fluid flows, tions for each class was solved. Interaction between dispersed
followed by other equipment such as heaters, exchangers, and dis- droplets, such as coalescence and breakup, was neglected. The
tillation columns. Consequently, a properly sized primary multi- pressure-correction approach with some adjustments for the mix-
phase separator can increase the capacity of the entire facility. ture phase was used to obtain the numerical solution of the system.
CFD simulation is routinely used to modify the design and to The model was used to simulate a first-stage separator equipped
improve the operation of most types of chemical process equip- with a deflector baffle, two perforated baffles, a demister, and a
ment, combustion systems, flow measurement and control sys- weir plate. The CFD results indicated that most of the smaller water
tems, material handling equipment, and pollution control systems droplets would remain in the oil phase. The CFD results in terms of
(Shelley 2007). There are two approaches to developing CFD concentrations of water droplets in the oil outlet vs. oil-residence
models of a multiphase flow: the Euler-Lagrange approach and the time compared well with empirical data. The effects of inlet dis-
Euler-Euler approach. In the Euler-Lagrange approach, the contin- tributors and distributing baffles on reducing the size and weight of
uous fluid phase is modeled by solving the time-averaged Navier- separation trains were reviewed by Frankiewicz et al. (2001). The
Stokes equations, and the dispersed phase is simulated by tracking sensitivity of the installed vortex cluster to the inlet flow rate has
a large number of droplets through the flow field based on New- also been demonstrated. This paper did not include any informa-
ton’s second law. The Euler-Euler approach, on the other hand, tion on the developed CFD models. Frankiewicz and Lee (2002)
deals with the multiple phases as continuous phases that interact studied the influence of inlet nozzle configuration, flow distribu-
with each other. Because the volume of a phase cannot be occu- tors, perforated plates, and outlet nozzles on the flow pattern in two
pied by the other phases, phase-volume fractions are assumed to be and three-phase oilfield separators. They used Fluent 6.0 software
continuous functions of space and time, and their sum is equal to 1. and took the inlet nozzle, the inlet momentum breaker, perforated
The literature on the critical unit operation of multiphase separa- plates, weir or bucket plates, and outlet nozzles as the key affecting
tors abounds with macro studies and design methodologies for two- components. To increase the effective liquid retention time, CFD
and three-phase vertical and horizontal separators. However, there studies indicated that a second perforated plate just upstream of the
are very few papers that focus on the micro details of the actual sep- outlet nozzle was required. Moreover, the use of composite plate
aration process. The most important features of these studies have electrodes and the shrouded pipe distributors increased the capacity
been reviewed as follows. Hansen et al. (1993) presented the simu- of the operating separator to some 67% owing to an improved plug-
lation results of the developed CFD code, FLOSS, for an industrial flow regime. Lee et al. (2004) evaluated the design of internals for
a three-phase separator to mitigate sloshing of liquid phases caused
by the offshore platform location. In order to prevent the water
phase from being pulled up toward the oil weir, on the basis of the
Copyright © 2012 Society of Petroleum Engineers performed CFD simulations, a perforated baffle was designed and
Original SPE manuscript received for review 29 November 2010. Revised manuscript
placed near the oil weir, the open areas of two preceding baffles
received for review 10 February 2011. Paper (SPE 161212) peer approved 22 March 2011. were decreased, and the configuration of the oil weir was modified.

December 2012 • Oil and Gas Facilities 57

 1630.1 cm

Gas
295.05 cm 380 cm 303 cm 338.6 cm Oulet

Inlet z

332.8 cm
y
x

123.4 cm 60 cm 92.5 cm 87.5 cm Water Oil

Oulet Oulet

Fig. 1––Geometrical specifications of the Gullfaks-A separator (Hansen et al. 1993).

TABLE 1—QUALITY OF THE MESH PRODUCED FOR THE GULLFAKS-A SEPARATOR

IN THE GAMBIT ENVIRONMENT
Number of Cells Maximum Squish Maximum Skewness Maximum Aspect Ratio

884847 0.748182 0.895873 44.1708

Skewness of the Produced Mesh
Skewness range 0–0.20 0.20–0.40 0.40–0.60 0.60–0.80 0.80–1.0
Density of cells 79.0416% 15.4785% 3.8489% 1.6285% 0.0025%

These improvements led to eliminating the water spillover problem. Developed CFD Model
Unfortunately, the details of the CFD simulations and the obtained Physical Model. Fig. 1 provides the geometrical specifications of
solutions were not presented in the paper. the Gullfaks-A separator as provided by Hansen et al. (1993). As
A vertical two-phase separator equipped with a deflector baffle Fig. 1 shows, a spherical deflector baffle was used to break the
and a vane-type demister was modeled by Swartzendruber et al. momentum of the inlet three-phase fluid flow entering the vessel as
(2005) using Fluent software. They focused on the quality of gas a high-momentum jet. The upper part of the vessel was equipped
flow distribution through the demister. Again, unfortunately, de- with internals, including flow-distribution baffles and a demister, to
tails of the developed CFD model are missing from the paper. On enhance the separation of liquid droplets from gas.
the basis of the resulting fluid-flow streamlines, two changes were In this study, building the physical model and generating the
devised to mitigate the uneven flow distribution in the vane dem- corresponding mesh system were performed in the Gambit 2.4.6
ister. Thus, the deflector baffle was moved away from the inlet and (ANSYS 2006b) environment. In order to have a discretized model
installed parallel to vane demister, and a 90° elbow with turning with “good” grid quality, the mesh-generation process was com-
vanes was installed between the inlet and the deflector baffle. Lu pleted in a step-by-step sequence. The vessel was split into areas
et al. (2007) studied the effectiveness of perforated plate baffles for and the inlet nozzle, deflector baffle, splash plate, weir, and outlet
improving the separation performance of a FWKO separator. The nozzles were first discretized. In doing so, the edges of nozzles and
Fluent 6.2 software was used for simulation, but the multiphase other internals were discretized before the mesh generation for the
modeling was based only on a balance between available compu- separator surfaces and volumes. Then, the cylindrical part of the
tational resources and model capabilities. The velocity contours of vessel was discretized such that some cells in this part were sepa-
fluid flows visually confirmed that the previous large flow circula- rated and referred to as the porous media, and did include mesh
tions were broken into small ones by installing the perforated plate for distribution baffles and the demister pad. The horizontal sur-
baffles. Furthermore, the mean residence time of fluid particles was rounding surfaces of each baffle (with thickness of 0.02 m) and
calculated and showed an increase from 630 to 980 seconds for the those of demister pad (with thickness of 0.15 m) were assumed to
water phase and from 520 to 745 seconds for the oil phase because be flat surfaces. Therefore, in the cylindrical part of the vessel, the
of the installed distributing baffles. Lee et al. (2009) discussed sev- grids must be fine enough and arranged horizontally in regular and
eral engineering judgments and the corresponding CFD verifica- constant intervals. After generating the mesh for the cylindrical part
tions to revamp the phase-separation inefficiencies experienced in of the vessel, the remaining parts of the vessel were “swept” by the
a major oil production facility. Their debottlenecking studies led Gambit mesh generation tool. Mesh elements were generally hexa-
to some suggestions for the weir height, liquid levels, and config- hedral. However, for regions with complex geometry (i.e., the inlet
uration and position of distribution baffles. Again, details of the and outlet nozzles), tetrahedral elements were necessarily used.
CFD models, developed through the Fluent 6.3.26 software, have The global quality of the produced mesh in terms of number of
not been provided in the paper. The simulation results showed that cells, maximum cell squish, maximum skewness, and maximum
the applied improvements mainly influenced the water phase, and aspect ratio are presented in Table 1, and Fig. 2 includes screen
the fluid-flow streamlines visually confirmed that the large flow shots of the generated model in the Gambit environment. Further-
circulations were broken into small weak ones by implementing the more, to ascertain the quality of the generated mesh, the cell skew-
suggested modifications. ness was evaluated and, as shown by the mesh results of Table 1,
only a negligible fraction of cells (0.0025%) was of poor quality.
However, the grids with a cell skewness factor greater than 0.8 were

58 Oil and Gas Facilities • December 2012

Fig. 2––Physical model and mesh generated for the Gullfaks-A separator in the Gambit environment.

TABLE 2—PHYSICAL PARAMETERS OF FLUIDS IN GULLFAKS-A SEPARATOR

PROVIDED BY HANSEN ET AL. (1983)

1988 Production Rate Future Production Rate Density Viscosity

(m3/h) (m3/h) (kg/m3) (Pa·s)

Gas 1640 1640 49.7 1.30e–5

Oil 1840 1381 831.5 5.25e–3
Water 287 1244 1030 4.30e–4
Operating Conditions Temperature=55.4°C; pressure=6870 kPa

converted to polyhedral grids. Although this minor modification ting the composition of the mixture was the accuracy of the mix-
did not reduce the maximum values reported in Table 1, the number ture density and viscosity at operating temperature and pressure
of cells was reduced from 884,847 to 884,805. compared to the values given in the original study. Using the PR
equation of state and the TRAPP model, the density and viscosity
Material Definition. The physical parameters for the fluids in of the mixture were estimated to be 783.59 kg/m3, and 0.005296
the Gullfaks-A separator are taken from Hansen et al. (1993) and Pa·s, respectively, with an estimation error for the oil density and
presented in Table 2. Because interfacial surface tensions were viscosity of 5.76 and 0.88%, respectively. Thus, it was assumed that
not given in the original paper, estimated values were used. For the oil/gas surface tension estimated by HYSYS was reasonable,
this purpose, a hydrocarbon mixture was defined in HYSYS 3.2 and a surface tension of 0.0238 N/m was assumed for the oil/gas
(AspenTech 2003) to simulate the oil phase. The criterion for set- interface. The assumed value compared well with the oil surface

December 2012 • Oil and Gas Facilities 59

tension range of 0.023 to 0.038 N/m at 20°C proposed by Streeter ing from atmospheric pressure to 9.2 MPa), using three different
and Wylie (1985). HYSYS 3.2 was also used for estimation of wa- fluids (air, nitrogen, and natural gas). The data have been fit to a
ter/gas surface tension of 0.0668 N/m. Using the chart provided by Hazen-Dupuit-Darcy type equation (Eq. 2) for calculation of the
Heidemann et al. (1987), the surface tension of pure water is 0.067 pressure drop caused by demister:
N/m at 55.4°C, which is in agreement with the HYSYS estimate.
Finally, the empirical study of Kim and Burgess (2001) was used to ∆P µ
= V + C ρV 2 , .................................................................. (2)
estimate the oil/water surface tension of 0.052 N/m at 25°C. They ∆x α
noted that although oil is a mixture of various hydrocarbons, each
one constituting hydrocarbon in contact with water has almost the where m is the fluid viscosity in Pa·s, α is permeability factor in m2,
same interfacial surface tension. Thus, it was assumed that the sur- V is flow velocity in m/s, C is the plate discharge coefficient, and ρ
face tension for oil/water interface at 25°C would be almost the is the fluid density in kg/m3.
same as reported by Kim and Burgess (2001). Furthermore, in Helsør and Svendsen (2007) have provided the correlating pa-
order to account for surface tension temperature functionality as rameters (a and C) for the different mesh pad types. As noted by
proposed by Poling et al. (2001), the reported value was modified Pourahmadi Laleh (2010), provided that the type and characteris-
and a surface tension of 0.0486 N/m was used for the oil/water tics of the mesh pad are available, these correlating parameters can
interface at 55.4°C. The estimated value is in agreement with the be used to calculate the parameters required for the porous media
Antonoff’s rule in that the oil/water surface tension should be ap- model. Because the specifications of the demister were not pro-
proximately equal to the absolute difference between oil and water vided in the original paper of Hansen et al. (1993), the most com-
surface tensions (Antonoff 1907), which are 0.0238 and 0.0668 monly used wire mesh properties were assumed for calculation of
N/m, respectively. these constants. A wire mesh pad with a thickness of 0.15 m is
commonly used in separators (Walas 1990; Lyons and Plisga 2005;
Modeling the Distribution Baffles. The porous media model with Coker 2007); hence, a type E demister as specified by Helsør and
appropriate modifications was used to model the flow through baf- Svendsen (2007) was selected for simulation purposes. The viscous
fles and demisters. This approach is based on the evaluation of the resistance factor was calculated to be 3.85e6 m–2, and the inertial
momentum source term in the porous media model by correlations resistance factor was calculated to be 126 m–1.
developed by Kolodzie and Van Winkle (1957) for the fluid flow
through a perforated plate (Pourahmadi Laleh 2010). Eq. 1 was de- Multiphase Models Incorporated. In order to develop a visual
veloped for calculation of the inertial resistance factor C2 in the understanding of the complex three-phase separation process,
direction normal to the baffle plate: Hansen et al. (1993) modeled the overall fluid-flow regimes inside
the separator. To simplify this complicated simulation task, they fo-
1  A  2  cused on two zones: the inlet and momentum breaker zone and the
C2 = 2  p  − 1 , ............................................................ (1) bulk liquid flow zone. In the current study, however, all the separa-
Cδ  A f   tion zones of the separator were simulated. Therefore, the results
 
should provide an overall picture of separation quality not only in
where C is the plate discharge coefficient, δ is the baffle thickness the inlet and bulk liquid zones, but also in the gas and interface
in m, and Ap and Af are total area and open area of the baffle in m2, zones. Exploiting the various multiphase models available in the
respectively. Therefore, if the configuration and dimensions of the Fluent 6.3.26 software, an efficient combination of two multiphase
flow-distributing baffles are provided, C2 and the porosity of baf- modeling approaches is used for modeling both the macroscopic
fles, which are necessary for the porous media model, can be cal- and microscopic features of this three-phase separator. Therefore,
culated. Although these crucial specifications were not provided the Euler-Lagrange approach is used for simulation of the move-
in the original paper of Hansen et al. (1993), Hansen et al. (1991) ment of fluid droplets that are injected at the separator inlet, and
performed several experiments to obtain data in order to validate the Euler-Euler approach is used for simulation of fluid-flow pat-
the results of their developed computer code, FLOSS, which was terns in the immiscible three-phase flows. Implementation of the
used for the simulation of the Gullfaks-A separator. Inside their Euler-Lagrange approach leads to the discrete phase model (DPM),
experimental model, with dimensions of 0.46×0.46×1.83 m, the which works well for flow regimes in which the discrete phase is
flow-distributing baffle was specified as a perforated plate with 173 of less than 12% volume fraction. In addition to the gravity and
holes, each with diameter of 6.4 mm and distance between centers drag forces, which are the most affecting forces in the phase-sepa-
of 25 mm. Because the model has been used for validation of the ration phenomenon, all other relatively effective forces, such as the
computer code, and the computer code was used to simulate the virtual mass force, the Brownian force, and the lift force, are also
Gullfaks-A separator, it can be expected that the model baffle had taken into account while tracking the droplets in the DPM model.
the same overall configuration (hole pattern) as the baffles of the Coalescence of particles and their breakups are also modeled by
Gullfaks-A separator. Therefore, with the further assumption of a DPM. For this purpose, the collision model is used for modeling
baffle thickness of 20 mm, the baffle porosity ε was calculated to be droplet coalescence, and on the basis of the particle Weber number,
0.05, and constant C2 was calculated to be 29240 m–1. a proper model within the spray model theory [i.e., the Taylor anal-
ogy breakup (TAB) model or wave model] is used for modeling
Modeling the Wire Mesh Demister. Wire mesh demisters were droplet breakup. As the surface tracking model of the Euler-Euler
also modeled using the porous media model. For this purpose, the approach, the volume of fluid (VOF) is used for simulation of the
porous media parameters, which are used for pressure drop calcula- fluid-flow patterns. The VOF model is designed for the simulation
tions in the media, need to be set. Because the mesh pad demisters of immiscible multiphase flows where the position of the interface
generally result in very low pressure drops, their pressure drop was between any two adjacent different phases is of interest. In this
assumed to be negligible. Fortunately, quite recently, a comprehen- model, a single set of momentum equations is shared by the fluids,
sive and practical research study has appeared in the public litera- and the volume fraction of each phase in the computational cells
ture that deals with the characterization of pressure drop in knitted is tracked throughout the domain. Note that the VOF is not per-
wire mesh demisters. Helsør and Svendsen (2007) have reviewed fect for modeling the inlet zone of a separator because a complex
the two other relevant studies in this field and presented their model momentum exchange occurs in this zone and fluid phases are fully
for pressure drop calculation in mesh pads. In their experimental interpenetrating. However, the VOF model can effectively capture
studies, the data have been collected and analyzed for seven differ- the macroscopic aspects of the major part of a multiphase separa-
ent wire mesh demisters, at four different system pressures (rang-

60 Oil and Gas Facilities • December 2012

 TABLE 3—DISCRETE PHASE PARAMETERS USED IN CFD SIMULATION OF GULLFAKS-A SEPARATOR
1988 Condition Future Condition

Discrete Phase Parameters Oil Drops Water Drops Oil Drops Water Drops

Maximum diameter (µm) 2,267 4,000 1,955 3,450

Mean diameter (µm) 907 1,600 780 1,380
Total mass flow rate (kg/s) 6.5e–4 4.4e–3 4.2e–4 2.8e–3
Number of tracked particles 1,000
Minimum diameter (µm) 100
Spread parameter 2.6

tor, the gravity separation zone. The necessary model settings will study in the field of droplet dispersions in the turbulent flow was
be described in the following sections. conducted independently by Kolmogorov (1949) and Hinze (1955).
They assumed that the maximum stable droplet or bubble size dmax
Definition of Droplet Size Distribution. In order to model the could be determined by the balance between the turbulent pressure
dispersion of oil and water droplets in the fluid-flow domain, the fluctuations, tending to deform or break the droplet or bubble, and
specification of the particle size distribution is a key step. However, the surface tension force resisting any deformation. The other im-
empirical data on the size of fluid particles was not available from portant theory for maximum stable bubble size was developed later
the Gullfaks-A separator (Hansen et al. 1993). Thus, a reliable by Levich (1962). He assumed that the maximum stable droplet or
method was required for prediction of particle size distribution for bubble size dmax could be determined by the balance between the
oil and water droplets entering the separator. There are numerous internal pressure of the droplet or bubble and the capillary pressure
research studies that predict the size distribution of fluid disper- of the deformed droplet or bubble. In a later study, Hesketh et al.
sions. However, most have focused on prediction of maximum (1987) modified the Levich theory to develop an equation that in-
stable droplet size because the other necessary size distribution cludes all the salient physical fluid properties required to describe
parameters such as spread parameter and minimum and mean droplet or bubble size in turbulent flow. Hesketh et al. (1987) con-
droplet size can be estimated based on the predicted (or measured) sidered both the Kolmogorov-Hinze and Levich theories and recog-
maximum stable droplet size and the nature of the fluid phases. In nized that in predicting maximum particle size for liquid/liquid and
the present study, the common particle size distribution function, gas/liquid dispersions, only the latter gives consistent results. By
the Rosin-Rammler (1933) equation, has been used. The Rosin- including a viscosity grouping term originally proposed by Hinze
Rammler equation contains two parameters: volume mean diam- (1955), Hesketh et al. (1987) have developed the following gener-
eter d¯¯ and spread parameter n: alized equation:
⎛ σ 0.6 ⎞ ⎛ D 0.5 ⎞
⎡ ⎛ d ⎞n⎤ dmax=1.38 ⎜ 0.3 0.2 0.1 ⎟ ⎜ 1.1 ⎟
Yd = exp ⎢ − ⎜ ⎟ ⎥ , ................................................................. (3) ⎝ ρc ρt μdc ⎠ ⎝ Vc ⎠
⎢⎣ ⎝ d ⎠ ⎥⎦ 0.6

(
 µ µ 0.25V 2.75 ρ −0.25D −1.25 d )  
1/3

where Yd is the mass (or volume) fraction of droplets with diameter × 1 + 0 .5975 
d c c c max
 ρc 
greater than d.   σ  ρd   ....... (5)
   
In Eq. 3, the volume mean diameter d¯¯ can be estimated from
maximum droplet diameter, dmax, through Eq. 4, proposed by
Green and Perry (2007): where rc and rd are the continuous and dispersed phase densities
(respectively) in kg/m3, mc and md are the continuous and dispersed
d¯¯ =0.4 dmax ............................................................................. (4) phase viscosities (respectively) in Pa·s, D is inside diameter of pipe
in m, and Vc is superficial velocity of continuous phase in m/s. Note
To specify the spread parameter n for the Gullfaks-A dispersions, that dmax should be calculated from Eq. 5 in an iterative manner,
two experimental studies, performed by Karabelas (1978) and because dmax is also present on the right side of Eq. 5 with an ex-
Angeli and Hewitt (2000), were used. The experiments of Kara- ponent of 0.2.
belas were carried out with kerosene (ρ=798 kg/m3, m=0.00182 Although estimation of dmax using Eq. 5 is tedious, the strong
Pa·s) and a more viscous transformer oil (ρ=892 kg/m3, m=0.0156 theoretical background and its very satisfactory representation of
Pa·s) as continuous phases and water as dispersed phase. The ex- empirical data provided confidence in using this method for predic-
periments of Angeli and Hewitt were performed with both water tion of the maximum droplet size. Note that Hesketh et al. (1987)
and the oil (ρ=801 kg/m3, m=0.0016 Pa·s) as dispersed and/or showed that this approach provided excellent results when dealing
continuous phases. The experimental distributions of Angeli and with experimental data that included a broad range of physical
Hewitt produced a value between 2.1 and 2.8 for the Rosin-Ram- properties: surface tension of 0.005 to 0.072 N/m, the continuous
mler spread parameter. This result agrees with the values of 2.13 to phase viscosity of 0.001 to 0.016 Pa.s, and the dispersed phase den-
3.30 reported by Karabelas (1978) for water dispersed in two dif- sity of 1 to 1000 kg/m3.
ferent oils. One of the most interesting experimental results of the The discrete phase parameters required for CFD simulation of
Karabelas (1978) study showed that the spread parameter can be the Gullfaks-A separator have been calculated and presented in
assumed to be constant and close to the measured average value for Table 3. The droplet size distributions for oil and water dispersions
either oil in water or water in oil dispersions. Therefore, the arith- are also represented in Fig. 3.
metic average value of 2.6, as reported by Karabelas (1978), was
used to set the particle size distribution. Setting the CFD Simulator Parameters. The Reynolds number
The next step involved finding a reliable method for prediction was much more than the transient value (Re=2,300) for all fluid
of maximum stable oil and water droplet sizes. The fundamental phases, and a suitable turbulence model was selected as the viscous

December 2012 • Oil and Gas Facilities 61

 7.0
6.5
Oil Droplets (Future)
6.0
5.5
5.0 Oil Droplets (1988)

4.5
Mass Fraction (%)

4.0
Water Droplets (Future)
3.5
3.0
2.5 Water Droplets (1988)
2.0
2.5
2.0
1.5
1.0
0.5
0.0
0

0
0
50

50

50
50

50

50

50

50

50

50

50

50

50

50
50
25

45

65
85
10

12

14
16

18

20

22

24

26

28

30

32

34

36
38
Droplet Size (micron)

Fig. 3––Droplet size distributions for oil and water dispersions at 1988 and the future production conditions.

model. For this purpose, the standard k-ε (Launder and Spalding • Fluid-flow analysis was confined to the inlet zone and the bulk
1972) model was selected. This semi-empirical model has been se- liquid flow zone. Therefore, the interaction between multiple zones
lected as the default in most commercial packages and is accepted was ignored.
as the most cost-effective and widely applicable turbulence model • In both zones, the flow was considered to be symmetrical
(Sharratt 1990; Gosman 1998). around the vertical plane in the middle of the separator (xz-plane);
In order to input the boundary conditions for inlet, the ve- thus, only half of each zone volume was modeled. Apparently, this
locity and volume fractions of phases were set. For the gas-outlet is a questionable assumption, particularly when no plug flow re-
boundary, outlet pressure and volume fractions (as pure gas) were gime was established as shown by their results.
set while for the liquid-outlet boundaries, outlet velocities and • The inlet section of the separator in which all three phases are
volume fractions (as pure liquid) were set. For identifying the flow present was modeled as a two-phase gas/liquid flow, and the results
regimes in inlet and outlet nozzles, the turbulence intensity and did provide the boundary conditions for the distributed velocity
hydraulic diameter of flow through the nozzles were determined. field in the liquid pool. A two-phase simulation of a three-phase
The turbulence intensity in the inlet and outlet zones was estimated zone will reduce the accuracy of the results for the inlet zone, and
using an empirical correlation (Fluent 6.3 User’s Guide 2006): the incorrect boundary condition will also decrease the accuracy of
downstream bulk liquid solution flow.
I=0.16 Re–0.125 ...................................................................... (6) • The grid systems used for numerical simulations of the inlet
zone and the bulk liquid zone were 11×8×15 and 23×4×5, re-
In normal operation, the separator was half-filled with liquid spectively. Given the vessel dimensions, the generated grid systems
(Hansen et al. 1993). To set the position of interface between are rather coarse. Note that if the assumed grid system was devel-
phases, the volume fractions of phases above and below the as- oped such as to cover whole the vessel, the generated grid would in-
sumed interface were patched to the reasonable values. clude some 4,480 cells, which is 0.51% of the generated mesh cells
Solving a multiphase simulation problem is inherently subject of the developed CFD model. Therefore, the grid system of the cur-
to stability and convergence issues. Careful choice of the solution rent study is almost 200 times finer than that used in the original
method and under-relaxation factors markedly affects both the rate work of Hansen et al. (1993).
of convergence and the solution existence (Sharratt 1990; Anderson • The other major improvement when compared with not only
1995). Thus, in order to overcome the stability/convergence difficul- Hansen et al. (1993) but also other previous projects on the CFD-
ties, the pressure-implicit with splitting of operators (PISO) method based study of separator performance is the direct and quantitative
was used as the solver (Issa 1986), and the under-relaxation factor evaluation of the separator efficiency. For this purpose, five data-
for pressure, density, momentum, volume fraction, and turbulence recording planes were defined to record the characteristics of the
groups was set at 0.10, 0.90, 0.0005, 0.005, and 0.70, respectively. droplets passing through them. The recording surfaces of interest
were two vertical at the start and at the end of the gravity sepa-
Results and Discussion ration section, the gas outlet, the oil outlet, and the water outlet.
Before presenting the results of this case study, it is important to Computer codes were developed to analyze the data provided by
highlight the most important modifications of this study as com- the recording planes. These computer codes calculate the separa-
pared with the original research presented by Hansen et al. (1993). tion efficiencies (based on mass distribution of droplets among gas,
In their CFD simulation of this multiphase separator, Hansen et al. oil, and water outlets), Rosin-Rammler particle size distributions
(1993) made a number of simplifying assumptions: on the five capturing surfaces, and the number of droplet coales-
cence and breakup.

62 Oil and Gas Facilities • December 2012

 6.888e+06
6.887e+06
6.885e+06
6.884e+06
6.883e+06
6.881e+06
6.88e+06
6.879e+06
6.877e+06
6.876e+06
6.876e+06
6.873e+06
6.872e+06
6.87e+06
6.869e+06
6.868e+06
6.866e+06
6.865e+06 Z
6.863e+06
6.862e+06 Y X
(a)
6.861e+06

6.888e+06
6.887e+06
6.886e+06
6.885e+06
6.884e+06
6.882e+06
6.881e+06
6.88e+06
6.879e+06
6.878e+06
6.877e+06
6.876e+06
6.875e+06
6.874e+06
6.873e+06
6.872e+06
6.871e+06 Z
6.87e+06
Y X
6.869e+06
6.868e+06
(a)

Fig. 4––Contours of pressure (Pa) in the middle of the Gullfaks-A separator for (a) 1988 and (b) the future condition.

Having prepared the physical model and set all the CFD param- fore, a PC runtime of approximately 24 hours was required for
eters, some 4,000 iterations were required for the continuous-phase solution of continuous-phase fluid flows with a further PC runtime
solution convergence. Each iteration took approximately 22 sec- of approximately 3 hours required for the simulation of interactions
onds on a Pentium D (3.20 GHz) and 2.00 GB of RAM PC. There- among the dispersed droplets and the continuous phases.

December 2012 • Oil and Gas Facilities 63

 6.514
6.189
5.863
5.537
5.212
4.886
4.56
4.235
3.909
3.583
3.258
2.932
2.606
2.281
1.955
1.629
1.304
0.978
0.6523 Z
0.3266 Y
X
0.0009611
(a)
7.371
7.003
6.634
6.266
5.897
5.529
5.16
4.792
4.423
4.055
3.686
3.317
2.949
2.58
2.212
1.843
1.475
1.106
0.7377
Z
0.3692
Y
0.0006381 X (b)

Fig. 5––Vectors of velocity (m/s) in the middle of the Gullfaks-A separator for (a) 1988 and (b) the future condition.

Fluid-Flow Profiles. Velocity vectors on four parallel horizontal tions predicted by Hansen et al. (1993) are a result of poor adjust-
planes and four parallel vertical planes for both the 1988 and the ments or assumptions (e.g., the poor setting of the over-relaxation
future production condition were obtained and compared with the parameters). Furthermore, the recent CFD-based study by Lu et al.
corresponding profiles from the original study. A complete set of (2007) does show that the distribution baffles generally improve
the profiles has been provided in Pourahmadi Laleh (2010), but the quality of liquid flow distribution in the vessel, break the large-
the most significant profiles will be presented in this paper. Using scale circulations into smaller ones, and reduce the short-circuiting
the original profiles, Hansen et al. (1993) addressed the rotation- flow streams.
al flow regimes established between any two internals. However, The pressure, velocity, and density profiles for both production
the large-scale fluid-flow circulations were not realized in current conditions are shown in Figs. 4 through 6. The simulated pressure
study, even though some minor flow circulations or backflows were profiles indicated that the pressure drops assigned to the baffles and
predicted. It was noted during the solution convergence trend that if demister are small (reasonable), and the velocity vectors were real-
the over-relaxation parameters are not adjusted correctly or the cor- istic. However, on the basis of the simulated density contours (Fig.
rect solver is not selected, large rotational flow patterns can be pro- 6), it would seem that the separator at both operating conditions
duced, and the solution fluctuates without approaching a realistic (particularly, at the future production condition) may suffer from
converged solution. In addition to this issue, the major simplifying foam and emulsion problems. The distortion of interfaces in the
assumptions used in the original work are another probable source inlet and outlet zones indicates a potential for foam and emulsion
of inaccuracy. Therefore, it would seem that the large flow circula- problems. The other detectable problem is the flow behavior near

64 Oil and Gas Facilities • December 2012

 1030
981
932
883
833.9
784.9
735.9
686.9
637.9
588.9
539.8
490.8
441.8
392.8
343.9
294.8
245.8
196.7 Z
147.7
98.72 Y X
49.7 (a)

1030
981
932
883
833.9
784.9
735.9
686.9
637.9
588.9
539.8
490.8
441.8
392.8
343.9
294.8
245.8
196.7 Z
147.7
98.72 Y X

49.7 (b)

Fig. 6––Contours of density (kg/m3) in the middle of the Gullfaks-A separator for (a) 1988 and (b) the future condition.

the water outlet predicted for the future production condition (Fig. to overcome this problem, one should minimize the risk of mixing
6b). With the large increase in the produced water flow rate, the liquid phases by improving the vessel design.
water phase must be pumped from the vessel at much higher rates.
Therefore, as indicated by the present CFD simulation results, there Separation Efficiencies. The analysis of the oil and water droplets
is an increasing tendency for the oil phase to be pushed toward the exiting at the separator outlets resulted in a predicted overall sepa-
water outlet. This, at least, will increase the risk of turbulence in the ration efficiency of 98.0% at the 1988 production conditions. The
water outlet zone and may lead to mixing of the phases. In order result is based on the mass distribution of injected oil and water

December 2012 • Oil and Gas Facilities 65

 TABLE 4—DROPLET SIZE DISTRIBUTION IN IMPORTANT ZONES OF THE GULLFAKS-A SEPARATOR

Before Gravity After Gravity

Discrete Phase Separation Separation Oil Water
Parameters Zone Zone Outlet Outlet

dmin (µm) 26 26 26 –
Oil Droplets dmax (µm) 1848 1830.3 1209 –
1988 640 627 419 –
Production d (µm)
Condition n 3.18 3.20 3.80 –
dmin (µm) 85 85 93 90
Water
Droplets dmax (µm) 2423 2338 438 2315
1077 1009 257 1008
d (µm)
n 2.81 2.79 6.79 3.05
dmin (µm) 37 37 47 25
Oil Droplets dmax (µm) 1009 1009 219 855
Future 422 423 154 300
Production d (µm)
Condition n 4.42 4.43 6.55 3.60
dmin (µm) 34 34 76 34
Water
Droplets dmax (µm) 1596 1596 148 1593
643 633 132 545
d (µm)
n 4.11 3.55 6.60 3.67

droplets at the separator outlets. This mass distribution analysis in- Conclusions
dicated that 100% of oil droplets and 96.9% of water droplets were A three-phase separator located in the Gullfaks oilfield in the
separated and exited through their corresponding outlets. Note that Norwegian sector of the North Sea was simulated. The combined
there was no droplet present in the gas phase outlet; hence, all the VOF-DPM approach was used to capture both macroscopic and
injected droplets came out in either the oil outlet or the water outlet. microscopic features of the phase-separation phenomenon. In this
As would be expected from practical field experience and also study, the installed distribution baffles and mist eliminator were
shown in the density contours of Fig. 6b, with an increase at the modeled using the porous media model, which required the detailed
future produced water flow rate, the separation efficiency for oil specifications and design information for the three-phase separator.
droplets was predicted to decrease to 1.3%. Again, there was no Using the available theoretical approaches and experimental corre-
predicted carryover in the gas outlet, and thus all the injected drop- lations, a useful methodology for estimation of droplet size distri-
lets came out with either the oil outlet or the water outlet. There- bution, which is necessary for implementing the DPM approach,
fore, the gas/liquid separation efficiency was still 100%. The very was developed. Compared with the original study of Hansen et
low separation efficiency for oil droplets indicates that the water al. (1993), the developed model did provide high-quality details
phase does not have sufficient residence time for oil droplets to rise of fluid-flow profiles, leading to a very realistic overall picture of
up and join the oil phase; therefore, almost all of the oil droplets are phase separation in all zones of the separator. The CFD simulations
carried by water phase to the water outlet. Although the separation demonstrated that major separation inefficiencies may be encoun-
efficiency was calculated to be 100% for water droplets, because of tered with the projected increase in the flow rate of produced water,
the difficulty in separating the oil droplets, the total separation ef- which compared well with the oilfield separator experience. The
ficiency has been reduced to 70.4%. CFD simulations showed that droplet breakage was common with
The result of droplet size distribution analysis on the selected an average rate of 76%, when dispersed droplets came into contact
surfaces of the separator is shown in Table 4. As the data of Table with the deflector baffle. Because of droplet breakup, the volume
4 shows, when compared with the initially defined droplet size dis- median diameter of droplets decreased to approximately 67% of
tribution (Table 3), droplets have become smaller. The reason is the initial value. However, the droplet size distribution remained al-
that droplet breakup occurs when the injected droplets strike the most the same while the droplets were traveling through the gravity
deflector baffle. Therefore, the volume median diameter has de- separation zone of the separator. Moreover, free coalescence of
creased to 70% of its initial value for oil droplets, and to 67% of its droplets was not a common phenomenon; hence, any positive ef-
initial value for water droplets for the 1988 production condition. fect of free coalescence on the separation efficiency was negligible.
With the projected increase in the inlet water flow rate, these values
change to 54% for oil droplets and to 46% for water droplets. The Nomenclature
CFD simulation results show that for the 2,000 injected droplets, Af = open area of a perforated plate, L2, m2
the number of breakups was predicted to be 1,590 for 1988 condi- Ap = total area of a perforated plate, L2, m2
tion and 1,543 at the future operating condition. Free coalescence C = discharge coefficient for a perforated plate
of droplets was not a common phenomenon in the three-phase sep- C2 = inertial resistance factor, L–1, m–1
arator. Droplet coalescence may happen at a very low rate of ap- ¯¯ = volume mean diameter in Rosin-Rammler equation,
d

proximately 0.1% without any noticeable trend. Table 4 also shows L, µm [m]
that the droplet size distribution before and after the gravity sepa- dmax = maximum droplet diameter, L, µm [m]
ration zone is almost the same. This implies that there should be dmin = minimum droplet diameter, L, µm
no further breakup while droplets are traveling through the main D = inside diameter of pipe, L, m
part of the separator; hence, droplet size distribution remains es- I = turbulence intensity
sentially constant. n = spread parameter in Rosin-Rammler equation

66 Oil and Gas Facilities • December 2012

 Re = Reynolds number Heidemann, R.A., Jeje, A.A., and Mohtadi, F. 1987. An Introduction to the
V = fluid velocity, L/t, m/s Properties of Fluids and Solids. Calgary, Alberta: The University of
Vc = superficial velocity of continuous phase, L/t, m/s Calgary Press.
Yd = the mass fraction of droplets with diameter greater Helsør, T. and Svendsen, H. 2007. Experimental Characterization of Pres-

than d sure Drop in Dry Demisters at Low and Elevated Pressures. Chem.
a = permeability factor, L2, m2 Eng. Res. Des. 85 (3): 377–385. https://dx.doi.org/10.1205/cherd06048.
δ = thickness of a perforated plate, L, m Hesketh, R.P., Fraser Russell, T.W., and Etchells, A.W. 1987. Bubble size
DP = pressure gradient, m/Lt2, Pa in horizontal pipelines. AIChE J. 33 (4): 663–667. https://dx.doi.
Dx = gradient along thickness of a perforated plate, L, m org/10.1002/aic.690330414.
m = fluid viscosity, m/Lt, Pa·s Hinze, J.O. 1955. Fundamentals of the hydrodynamic mechanism of split-

mc = continuous-phase viscosity, m/Lt, Pa·s ting in dispersion processes. AIChE J. 1 (3): 289–295. https://dx.doi.

md = dispersed-phase viscosity, m/Lt, Pa·s org/10.1002/aic.690010303.

ρ = fluid density, m/L3, kg/m3 Issa, R.I. 1986. Solution of the implicitly discretised fluid flow equations

ρc = continuous-phase density, m/L3, kg/m3 by operator-splitting. J. Comput. Phys. 62 (1): 40–65. https://dx.doi.

ρd = dispersed-phase density, m/L3, kg/m3 org/10.1016/0021-9991(86)90099-9.

σ = surface tension, m/t2, N/m Karabelas, A.J. 1978. Droplet size spectra generated in turbulent pipe flow
of dilute liquid/liquid dispersions. AIChE J. 24 (2): 170–180. https://
Subscripts dx.doi.org/10.1002/aic.690240203.
c = continuous Kim, H. and Burgess, D.J. 2001. Prediction of Interfacial Tension between
d = dispersed, a diameter Oil Mixtures and Water. J. Colloid Interface Sci. 241 (2): 509–513.
max = maximum https://dx.doi.org/10.1006/jcis.2001.7655.
min = minimum Kolmogorov, A.N. 1949. On the Breaking of Drops in Turbulent Flow. Dokl.
r = reduced Akad. Nauk SSSR 66: 825–828.
Kolodzie, P.A. and Van Winkle, M. 1957. Discharge coefficients through
References perforated plates. AIChE J. 3 (3): 305–312. https://dx.doi.org/10.1002/
Anderson, J.D. Jr. 1995. Computational Fluid Dynamics: The Basics with aic.690030304.
Applications. New York: McGraw-Hill Science/Engineering/Math. Launder, B.E. and Spalding, D.B. 1972. Mathematical Models of Turbu-
Angeli, P. and Hewitt, G.F. 2000. Drop size distributions in horizontal oil- lence. New York: Academic Press.
water dispersed flows. Chem. Eng. Sci. 55 (16): 3133–3143. https:// Lee, C.-M., van Dijk, E., Legg, M. et al. 2004. Field Confirmation of CFD
dx.doi.org/10.1016/s0009-2509(99)00585-0. Design for FPSO-mounted Separator. Paper OTC 16137 presented
ANSYS. 2006a. Fluent, version 6.3.26 (commercial computational fluid at the Offshore Technology Conference, Houston, 3–6 May. https://
dynamics simulator). Canonsburg, Pennsylvania: ANSYS Inc. dx.doi.org/10.4043/16137-MS.
ANSYS. 2006b. Gambit, version 2.4.6 (primary pre-processor for Fluent). Lee, J.M., Khan, R.I., and Phelps, D.W. 2009. Debottlenecking and Com-
Canonsburg, Pennsylvania: ANSYS Inc. putational-Fluid-Dynamics Studies of High- and Low-Pressure
Antonoff, G.N. 1907. Surface Tension at the Boundary of Two Layers. J. Production Separators. SPE Proj Fac & Const 4 (4): 124–131. SPE-
Chim. Phys. Phys.-Chim. Biol. 5: 372–385. 115735-PA. https://dx.doi.org/10.2118/115735-PA.
AspenTech. 2003. Aspen HYSYS V3.2 Chemical Process Simulator. Burl- Levich, V.G. 1962. Physiochemical Hydrodynamics. Englewood Cliffs:
ington, Massachusetts: Aspen Technology, Inc. Prentice-Hall.
Coker, A.K. 2007. Ludwig’s Applied Process Design for Chemical and Pet- Lu, Y., Lee, J.M., Phelps, D. et al. 2007. Effect of Internal Baffles on Vol-
rochemical Plants, fourth edition. Burlington, Massachusetts: Gulf umetric Utilization of an FWKO--A CFD Evaluation. Paper SPE
Professional Publishing. 109944 presented at the SPE Annual Technical Conference and Ex-
Fluent 6.3.26 User Guide. 2006. Lebanon, New Hampshire: Fluent, Inc. hibition, Anaheim, California, USA, 11–14 November. https://dx.doi.
Frankiewicz, T. and Lee, C.-M. 2002. Using Computational Fluid Dynamics org/10.2118/109944-MS.
(CFD) Simulation to Model Fluid Motion in Process Vessels on Fixed Lyons, W.C. and Plisga, G.J. 2005. Standard Handbook of Petroleum and
and Floating Platforms. Paper SPE 77494 presented at the SPE An- Natural Gas Engineering, second edition. Burlington, Massachusetts:
nual Technical Conference and Exhibition, San Antonio, Texas, USA, Gulf Professional Publishing.
29 Septemper–2 October. https://dx.doi.org/10.2118/77494-MS. Poling, B.E., Prausnitz, J.M., and O’Connell, J.P. 2001. The Properties of
Frankiewicz, T., Browne, M.M., and Lee, C.-M. 2001. Reducing Sep- Gases and Liquids, fifth edition. New York: McGraw-Hill Profes-
aration Train Sizes and Increasing Capacity by Application of sional.
Emerging Technologies. Paper OTC 13215 presented at the Offshore Pourahmadi Laleh, A. 2010. CFD Simulation of Multiphase Separators.
Technology Conference, Houston, 30 April–3 May. https://dx.doi. PhD dissertation, University of Calgary, Calgary, Alberta, Canada
org/10.4043/13215-MS. (September 2010).
Gosman, A.D. 1998. Developments in Industrial Computational Fluid Rosin, P. and Rammler, E. 1933. The Laws Governing the Fitness of Pow-
Dynamics. Chem. Eng. Res. Des. 76 (2): 153–161. https://dx.doi. dered Coal. J Inst Fuel 7: 29–36.
org/10.1205/026387698524721. Sharratt, P.N. 1990. Computational Fluid Dynamics and its Application in
Green, D.W. and Perry, R.H. 2007. Perry’s Chemical Engineers’ Handbook, the Process Industries. Chemical Engineering Research Design 68
eighth edition. New York: McGraw-Hill. (1): 13–18.
Hallanger, A., Soenstaboe, F., and Knutsen, T. 1996. A Simulation Model Shelley, S. 2007. Computational Fluid Dynamics—Power to the People.
for Three-Phase Gravity Separators. Paper SPE 36644 presented at Chem. Eng. Prog. 103 (4): 10–13.
the SPE Annual Technical Conference and Exhibition, Denver, 6–9 Streeter, V.L. and Wylie, E.B. 1985. Fluid Mechanics, eighth edition. New
October. https://dx.doi.org/10.2118/36644-MS. York: McGraw-Hill.
Hansen, E.W.M., Heitmann, H., Lakså, B. et al. 1991. Fluid Flow Modeling Swartzendruber, J., Fadda, D., and Taylor, D. 2005. Accommodating Last
of Gravity Separators. Proc., 5th International Conference on Multi- Minute Changes: Two Phase Separation Performance Validated by
phase Production, Cannes, France, 19-21 June, 364–380. CFD. Proc., ASME Fluids Engineering Division Summer Meeting
Hansen, E.W.M., Heitmann, H., Laska, B. et al. 1993. Numerical Simula- and Exhibition, Houston, 19-23 June, 713–715.
tion of Fluid Flow Behavior Inside, and Redesign of a Field Sepa- Walas, S.M. 1990. Chemical Process Equipment Selection and Design.
rator. Proc., 6th International Conference on Multiphase Production, Newton, Massachusetts: Butterworth-Heinemann Series in Chemical
Cannes, France, 19–21 June, 117–129. Engineering, Butterworth-Heinemann.

December 2012 • Oil and Gas Facilities 67

Ali Pourahmadi Laleh is a research engineer with the Reservoir Simula- in establishing Hyprotech as a leading international process simulation
tion Group in Calgary. He has more than 11 years of experience as a software company. Svrcek’s teaching and research interests center on
research engineer, involved in rectifying industrial scale process ineffi- process-simulation control and design. He has been involved for many
ciencies and optimizing chemical plants. Pourahmadi Laleh has done years in teaching a continuing education course, Computer-Aided Pro-
research in separation technologies, coal liquefaction, and hot fluid- cess Design: Oil and Gas Processing, that has been presented world-
injection processes. He has proposed a novel approach for automatic wide. He has authored or coauthored over 200 technical articles/
design of the optimum distillation column sequence using genetic algo- reports and has supervised more than 50 graduate students. Svrcek
rithms. He has also developed an efficient strategy for realistic simula- holds BS and PhD degrees in chemical engineering from the Univer-
tion of oilfield separators and has provided improved design criteria for sity of Alberta.
these multiphase separators. Pourahmadi Laleh holds a BS degree from
Sahand University of Technology, an MSc degree from Sharif University Wayne D. Monnery is president of Chem-Pet Process Tech Limited in
of Technology, and a PhD degree from the University of Calgary, all in Calgary and an adjunct associate professor at the University of Calgary.
chemical engineering. He has 24 years of industrial experience as a process engineer, with
recognized expertise in applied thermodynamics, process simulation,
William Y. Svrcek is a professor emeritus at the University of Calgary and physical properties of petroleum systems, as well as in sweet gas
and president of Virtual Materials Group Inc. in Calgary. Prior to joining processing, sour gas treating, and sulfur recovery. Monnery has also
the University of Calgary, he worked for Monsanto Company as a senior worked on heavy oil and steam assisted gravity drainage (SAGD) facility
systems engineer and as an associate professor (1970–75) at the Uni- simulation and design. He has done research in sulfur plant kinetics, al-
versity of Western Ontario. Svrcek was also a senior partner in Hypro- ternative sour gas treating, water content of high-pressure acid gases for
tech, now part of Aspen Technology, from its incorporation in 1976. As acid gas injection, and phase separation. Monnery holds a PhD degree
a principal, director, and president (1981–1993), he was instrumental in chemical engineering from the University of Calgary.

68 Oil and Gas Facilities • December 2012