Vous êtes sur la page 1sur 11

A Model for Evaluating Inlet Systems

to Gas/Liquid Separators
Logan Grim, Wood

Background This paper provides some guidelines for early stages of projects
Gas/liquid separators are a basic unit operation used to segregate and a method for evaluating the effect of inlet piping and inlet de-
phases, commonly found in process plants (Fig. 1). In many situ- vices on gas/liquid separator performance. The effects on the ve-
ations, high-efficiency segregation of the phases is not required as locity profile due to the piping elements will be reviewed as well
the downstream equipment can tolerate carry-over. In other cases, as the impact on the droplet size distribution. With this informa-
poor removal efficiency results in off-specification products, give- tion, the efficacy of gravity separation, and therefore liquid load/
away of product quality and/or suboptimal operation of equipment, challenge to the mist eliminator, can be evaluated using estab-
and thus is critical. It is therefore important to be able to quantify lished methods, such those described by Bothamley (2013). These
the gas/liquid separation performance. methods give designers useful tools to evaluate the adequacy of a
Most vapour/liquid separators are sized using a gravity-based proposed design.
criterion and assume plug flow of the gas and liquid. In reality, the
phases are not evenly distributed across the cross section, resulting Outline of the Procedure
in a degree of underperformance. In certain situations, the piping The methodology is carried out in several steps, illustrated in Fig. 3.
configuration can result in a coherent rotational motion. Centri­
fugal motion forces the liquid to the outside of the pipe forming Flow Pattern
an annulus, even if the flow pattern isn’t annular. If this motion Knowledge of the flow pattern is important to the design of sepa-
persists to the inlet of the separator, liquid will jet to the top of the rators, as this tells the designer if there will be intermittent forces,
drum and this may result in bulk liquid carry-over (Fig. 2). such as those occurring in slug flow or high levels of entrainment
A major factor contributing to these effects is the inlet piping such as in annular/mist flow. The author recommends the Taitel and
configuration. However, few guidelines are available. The Amer- Dukler (1976) model, as this has been proven to be reasonably reli-
ican Petroleum Institute (API) provides guidelines (API 686, 2009) able even at larger pipe sizes (Shoham 2006).
for the inlets to pumps, and numerous vendors have specific re- The Taitel and Dukler model calculates the dimensionless liquid
quirements for instruments; however similar criteria are not widely holdup and tests this against stability parameters to determine the
used for separators. Some company standards mandate 10 pipe di- flow pattern. In the original Taitel and Dukler article, for horizontal
ameters (10D) of straight pipe immediately upstream of the sepa- flow, it is shown that the dimensionless liquid height is a function
rator while others have no requirements at all. of the Martinelli parameter only. This allowed the development of
In many situations, however, 10D of straight length is either a graph with the abscissa of the Martinelli parameter which is very
not followed or becomes impractical, especially for large-diameter convenient for hand calculations.
piping. Heijckers (2012) has shown that bends and other pipe fit- For computer applications, this would require the regression of
tings can lead to velocity maldistribution of more than twice the the transition lines for each of the regions. Therefore, it is more
mean velocity (i.e., the maximum face velocity on the mesh pad is useful to calculate the liquid holdup and directly use the transition
twice the average) or flooding of the mist eliminator by bulk liquid. criteria. Since rigorous solution of the liquid holdup is non-trivial,

Plug Flow Reality

Inlet Mist
Inlet flow device Eliminator
Swirl
Gas
Distribution H

Velocity
Distortion

Fig. 2—Gas distribution and rotational motion on the inlet in


Fig. 1—Gas scrubber and piping schematic. a separator.

This version of this article was updated on 22 May 2018 and contains a number of corrections. | 2018  •   Oil and Gas Facilities 1
the author has fit a curve, Eq. 2, based on the Martinelli param-
eter. For turbulent flow in smooth pipes, the Martinelli parameter Flowrate Physical Properties Inlet Piping
(­Chisholm 1983) is:

Eq. 1—Martinelli parameter


Evaluate Flow Pattern (Section 3)

Characterize Droplet Distribution (Section 4 & 5)

The dimensionless liquid holdup can be estimated as:


Evaluate the Flow Distortion and Rotation (Section 6)
Eq. 2—Dimensionless liquid holdup.

Translate Effects to the Inlet Device (Section 7)

Fig. 3—Calculation steps.


The form of the above equation has been selected to ensure that
the limits of Xtt tend to the correct values (i.e., as Xtt→0 then hL→0
and as Xtt→∞ then hL→1). treated in the same fashion. Dispersed bubble flow occurs at very
With the dimensionless liquid holdup calculated, the stability high mass fluxes, would result in very poor separation, and is very
criteria for stratified to nonstratified flow can be tested. The re- rarely encountered in practice.
quired geometric terms are as follows: The original Taitel and Dukler article indicates the transition
from annular to slug flow occurs when the pipe is more than half
Eq. 3—Dimensionless gas area. full. However, this has been found not to match the data (Shoham).
Therefore, the following is used as per the recommendations
of Shoham:

Eq. 8—Annular stability parameter.

Shear Droplets
To quantify separation performance of gravity separation or mist
Eq. 4—Dimensionless interface area. eliminators, the distribution of the entrained droplets must be
known. The assumption of droplet entrainment preassumes that the
continuous phase is gas and therefore the flow must be annular. The
author has opted to use a two-parameter model for the droplet dis-
tribution based on the recommendations of Kouba (2014). Many
maximum droplet size correlations are available; the author has
Eq. 5—Dimensionless gas velocity. opted to use the modification of Hinze’s equation (1955) by Calder­
bank (1958):

Eq. 9—Calderbank dmax.

This is checked against the stability parameter for stratified


flow:
The Weber number and the Reynolds number are evaluated using
Eq. 6—Gas stability parameter. the gas density and viscosity at the velocity of the two phases, as-
suming homogenous flow. Note that the viscosity is averaged based
on the volumetric flow (i.e., the fraction in contact with the wall).

Eq. 10—Weber number.

Eq. 7—Stratified stability limit.

Eq. 11—Reynolds number.

For the purposes of separators, the key flow patterns are annular,
slug, and stratified. Stratified wavy is a subset of stratified and it is

2 Oil and Gas Facilities  •   2018 | This version of this article was updated on 22 May 2018 and contains a number of corrections.
Few minimum droplet size correlations are available and Subsequent fittings will result in higher levels of shear and thus
­Kouba’s is recommended: smaller droplets.
Droplet coalescence is an ongoing area of research and is not
Eq. 12—Kouba’s dmin. well understood. For very high liquid fractions with significant
mass flux, the flow pattern will be slug flow. For slug flow any
liquid droplets will rapidly collide and coalesce, thus breakup due
to fittings has little effect.
For systems with relatively small liquid fractions, some form of
a collision model is required to give a realistic basis for the coales-
With the minimum and maximum droplet size determined, cence of droplets. The following equation based on the model of
the system can be fitted to an assumed distribution such as a Log Radovcich and Moissis (1962) is used to determine the maximum
Normal Distribution, details of which can be found in Kouba. droplet size after a given length of pipe:
Normally, the gas density is used in the minimum droplet
size equation. However, for systems where droplet shattering is Eq. 15—Droplet diameter due to collisions.
likely, such as with inlet deflectors and halfpipes, the gas den-
sity (ρg) is substituted for the liquid density (ρl) as recommended
by Kouba.
For situations where the flow is not annular, alternative methods
are applied. For stratified flow, the flow is likely to separate in the
pipe, and the assumption of no entrainment is reasonable.
For slug flow, the system is characterized by intermittent gas
and liquid flow which will have differing levels of entrainment.
For flow-induced slugging the slug velocity can be estimated
by Shoham:

Eq. 13—Slug translational velocity. Where P is the collision efficiency and c is the characteristic
velocity
The characteristic velocity is assumed to be the turbulent eddy
velocity given by:

Eq. 16—Characteristic velocity of gas.


The minimum droplet diameter is estimated using Eq. 12,
using the slug velocity, (VTB) and the liquid density (ρl). The max-
imum droplet diameter is estimated using Eq. 9, using the average
voidage based on a suitable slip correlation and the slug velocity.
Since this is higher than the two-phase mixture velocity, the drop-
lets will be smaller than those calculated in an annular flow model To simplify the above equation the superficial gas velocity has
and is therefore conservative. been used in place of the actual gas velocity, which avoids the need
to compute actual velocities and entrainment.
The Effect of Fittings The probability of a successful collision P is given by Chester
The droplet size correlations apply to equilibrium flow, where the (1991):
effects of interfacial shear, mass transfer with the annulus, and the
droplet collision have balanced. In real piping systems, fittings re- Eq. 17—Droplet collision efficiency.
sult in increased shear and smaller droplets. The rate of shear is
primarily driven by the pressure drop of the component, and there-
fore is primarily a function of the frictional losses in fittings such
as valves.
Fittings also tend to result in increased turbulence in the flow,
which results in greater shear and finer droplets. Computational where the Weber number is given by
fluid dynamics (CFD) analysis can be used to model the fitting and
evaluate the turbulent energy dissipation and the droplet size distri- Eq. 18—Collision Weber number.
bution using a model such as Hinze (Walsh 2016). The following
model is practical for simpler spreadsheet applications and is based
on the frictional resistance coefficient of a fitting. The effective ve-
locity of fluid in the fitting is approximated by:

Eq. 14—Effective diameter of fittings. The coalescence model is tentative at best, with both the char-
acteristic velocity and the collision efficiency models having little
empirical data behind them. The model also ignores the influence
of peripheral liquid film mass transfer, which may be significant.
However, this exhibits the correct trends (low collision rates at low
liquid fractions and rapid increase in collision rate at high liquid
With the effective diameter calculated, the new effective ve- fractions) and the main sensitivity of the model is the void fraction.
locity can be determined. Using Eqs. 9 and 12, the droplet size dis- Therefore, the result should be representative of the magnitude of
tribution can be determined. coalescence in the pipe.
In the piping, downstream of the fitting, some of the small drop-
lets will begin to coalesce into larger droplets and, given sufficient
straight length, will return to the equilibrium pipe droplet pattern.

This version of this article was updated on 22 May 2018 and contains a number of corrections. | 2018  •   Oil and Gas Facilities 3
High Pressure Region

Secondary Flows

Fig. 4—Secondary flow in bends (after Miller 1978).

Maldistribution of the Feed


Heijckers has shown that, even when designed with 10D straight
length, separators can fail to perform due to maldistribution of the Fig. 5—Swirl due to bend interactions (after Miller).
gas and liquid in the pipe. Maldistribution of the flow is primarily
caused by bends in the pipes.
When two bends are positioned in series orthogonally to one A fraction of the energy lost in the bend will result in swirling
­another, fluid flows through the shortest path that results in a flow, also known as Dean vortices. If it is assumed that all the en-
swirling flow (Fig. 4): ergy lost is converted to swirl, the resulting value is overly con-
Only some of the pressure losses due to a bend occur directly servative and results in unrealistic swirl rates. Based on the data
in the bend, with the remaining dissipating in the straight length of Dutta and Nandi (2015) for single-phase flow, the effective
downstream (Miller 1978). These losses are due to the dissipation ­velocity head for swirl generation is:
of the secondary flows generated in the bend and reestablishment
of the velocity profile. The main factor which influences the frac- Eq. 21—Velocity head converted to swirl due to a bend.
tion of the losses occurring downstream of the bend is the radius
of the bend.
Based on the results of Miller, all secondary flows are dissipated
within 30 pipe diameters and the peak velocity is about two times
the minimum. An exponential equation with the limit of 99.8% of
flow dissipated by 30 pipe diameters is presented: This curve fit by the author, while empirical, is a good fit to
Dutta & Nandi’s data and has the correct limits tending to 0 as the
Eq. 19—Approximated velocity profile distortion downstream of r/D is large and resulting in a value less than the K for a mitre joint
a bend. (r/D=0.5).

Eq. 22—Energy loss due to a bend.

Normally only horizontal flow distortion is of concern as the


vertical distortion will be dissipated in the upturn of a vertical sepa-
rator and flow reversal in a horizontal separator. If the inlet nozzle Miller has stated that if bends are located close enough then the
is in the end of a horizontal separator, the vertical distortion should static pressures interfere with the flow as illustrated in Fig. 5. If the
also be included. bends are separated by a reasonable distance, the velocity profiles
In addition to the distortion of the velocity profile, bends can will re-establish and the swirl will be greatly reduced. For the pur-
result in a “swirl” of the fluid. This occurs as the low energy fluid poses of this method, it is assumed that the length over which the
moves to the inside of successive bends that are not oriented inline dissipation occurs is the sum of the length between the bends and
(Fig. 5). the piping downstream of the last bend as per Eq. 19.
Heijckers showed via CFD that this is also the case in two-phase The angular velocity of the swirl can be determined by assuming
flow. On a similar basis to the single-phase flow in bends, it is as- the motion results in an annulus around a core of gas. If no entrain-
sumed by the author that the swirl and velocity profile distortion ment in the gaseous core is assumed then the diameter of the core
result from secondary flows. For two-phase swirling flow Liu and can be calculated from the void fraction. If it is also assumed that
Bai (2015) have shown experimentally similar exponential dissipa- there is no slip between the gas and liquid the following diameter
tion over 30 pipe diameters after deliberately inducing a swirl. This results for the gaseous core:
is a good fit to the decay proposed in Eq. 19; however, the equilib-
rium condition is no rotation, thus the swirl must decay to zero and Eq. 23—Diameter of nonslip gas core.
therefore the –1 term is not required.

Eq. 20—Residual swirl velocity.

4 Oil and Gas Facilities  •   2018 | This version of this article was updated on 22 May 2018 and contains a number of corrections.
Eq. 28—Maximum vertical rise of swirling liquid.
Swirling Liquid Film
Flung to Outside

In vertical separators, the spacing between the inlet nozzle and


mist eliminator should be greater than the vertical rise of the in-
di
Gas Core coming liquid. In horizontal separators, this rise increases the ef-
fective fall distance of a droplet and should be added to the droplet
dp settling distance.
For multiple bends in a system the swirl energy for each bend
before the final two is added to the energy of the total.

Eq. 29—Summation of swirl velocities.

Although the basis for this method is not particularly strong,


Fig. 6—Gas liquid swirling flow. it has several advantages. Ignoring the space between the last
two bends before the separator would cause errors if there was
an elbow immediately upstream of the inlet nozzle. Also, in-
Alternatively, a less conservative assumption is to consider slip cluding the energy contribution of the last two elbows ensures
between the gas and liquid, thus resulting in a thicker annular ring. that the most important contributors to the energy of the swirl
For the purposes of the worked example the thickness of the ring are included.
has been calculated using the Chisholm slip correlation:
Translation to the Inlet Device
Eq. 24—Chisholm slip correlation. Inlet devices diffuse the gas across the cross section of the separator
and facilitate the knockout of bulk liquids. Maldistribution results
in higher localized gas velocities which entrain finer droplets and
causes zones of higher velocity in the mist-handling equipment,
­resulting in premature flooding. Fig. 2 illustrates the problems.
Popular inlet devices include, half-pipe (1) and deflector plates
With the void fraction given by: (2) in Fig. 7. Both result in droplet impaction and shattering and
are not recommended (Rhyne 2015). For most applications, an inlet
Eq. 25—Void fraction equation. vane device (3) is suitable, as it is effective, robust, and inexpen-
sive. The curved blades progressively divide the gas, distributing
it evenly, and smoothly remove the liquid minimizing shear. Cy-
clones (4) are also commonly used and have the highest capacity,
and are usually reserved for revamp situations.
Bothamley provided empirical curves (Fig. 8) which provide an
The liquid is modeled as a spinning annulus around the inside estimate of the maldistribution of gas downstream of an inlet de-
pipe wall as shown in Fig. 6. vice, for the purposes of design and rating.
Using this model, and the appropriate moment of inertia equa- The curves in Bothamley’s article also appear to conform to the
tions for cylinders and hollow cylinders, it can be shown that the dissipation equation presented earlier resulting in the following
angular velocity is given by modified form of Eq. 19:

Eq. 26—Angular velocity of the annular liquid ring and gas core. Eq. 30—Velocity distortion due to the inlet device.

Where E is the number of bends being modeled and ω is the an- The term A is a constant for each type of internal and is given as:
gular velocity. The tangential velocity (Vtan) at the inside edge is
used to estimate the maximum vertical travel of the bulk liquid in
the separator. The maximum tangential velocity is given by:
Device Type Constant A
Eq. 27—Maximum tangential velocity of the liquid annulus.
No device 3

Deflector 2

Half pipe 1.6


This is then used to determine the effective upwards velocity Vane device 1.3
and the kinetic energy can be equated to the increase in potential
energy accounting for buoyancy Cyclone 1.45

This version of this article was updated on 22 May 2018 and contains a number of corrections. | 2018  •   Oil and Gas Facilities 5
(1) Half Pipe (2) Deflector Plate (Frequent in Vertical Drums)

Some gas overflows

Smaller droplets
“bouncing” off
of the end plate Smaller droplets
“bouncing” off
High velocity over the of the end plate
Most gas “hits” surface re-entrains liquid
liquid surface

High velocity over the


surface re-entrains liquid

(3) Inlet Vane Device (4) Inlet Cyclone


Centrifugal motion forces
Gas “sliced” off of the main stream droplets to the outside
of the cyclone

Vane device Cyclonic inlet


device

Liquid drop deflected by the vanes


and fall downwards

Fig. 7—Schematics of common inlet devices.

The effect of a poorly established velocity profile upstream •  Using Eq. 14, approximately 90% of secondary flows are
of the drum needs to be added to the distortion due to inlet de- likely to be dissipated within 10dp of inlet straight pipe
vice. Since the cumulative effect of the maldistribution upstream °° There should be no bends in the horizontal plane within
of the 30dp of straight pipe (not equivalent length) will be 0.002 10dp of the inlet.
(by Eq. 19), this can be ignored. The cumulative distortion factor °° Bends local to the separator within 20dp must be in the same
(Fi & Ftot) is calculated stepwise for each fitting and subsequent plane to avoid bulk liquid swirl.
straight  length: •  High-shear fittings, such as globe valves, should be positioned
as far as possible from the separator, at least 30dp.
Eq. 31—Dissipation of the velocity profile distortion. •  Liquid droplets in systems with high liquid fractions will rap-
idly coalesce while higher void fractions, often found in gas scrub-
bers, will take much longer to reach equilibrium.
•  If the above rules cannot be followed, the effectiveness of gas/
liquid separation of a system can be checked using the method-
ology presented.
Actual Velocity/Plug Flow Velocity

3.5
3
2.5
No Device
2
Deflector Plate
1.5
Halfpipe
Refer to the worked example for better understanding of the ad- 1 Vane Device
dition of the distortion factors. 0.5 Cyclone
0
Conclusions and Recommendations 0 2 4 6 8 10 12
A model is presented that enables a designer to evaluate the impact L/Di
of the inlet piping on gas liquid separators. Based on this model the
following observations and recommendations are made. Fig. 8—Effect of inlet device on gas liquid distribution.

6 Oil and Gas Facilities  •   2018 | This version of this article was updated on 22 May 2018 and contains a number of corrections.
Symbology Worked Example
A = Distortion constant for an inlet device, dimensionless Below is a worked example of how to use the methods described
A g̃ = Dimensionless gas area above. (The specifics of the example are taken from a problem pre-
ˉc = Characteristic velocity of a particle, m/s sented by Rhyne). A three-phase mixture is passed to a drum. In
df = Effective diameter of a fitting this example, the properties are:
di = Inner diameter of rotating annulus, m
dini = Initial diameter of a droplet entering a pipe Property Gas Liquid
segment, m Flow (kg/s) 8.829 0.435
dmax = Maximum stable droplet diameter, m
dmin = Minimum droplet diameter due to pipe flow, m Density (kg/m3) 15 630
dp = Pipe internal diameter, m
Viscosity (pa.s) 1.2×10–5 2.5×10–4
E = Number of bends effecting rotational motion,
dimensionless Surface tension (N/m) 0.014
En = Energy lost per fitting
F = Taitel and Dukler stability parameter
The inlet pipe is an 8-in. schedule standard pipe (0.20274m ID)
Fi = Cumulative velocity distortion due to fittings
as per Rhyne’s example.
Ftot = Summation of velocity distortion Rhyne does not indicate the fraction of the liquid that is water,
g = Acceleration due to gravity, (9.80665m/s2) therefore the entire stream is assumed to be hydrocarbon.
H = Maximum liquid jet height due to rotation, m The flow pattern first needs to be determined. The mass quality
hL̃ = Dimensionless liquid holdup is calculated from
K = Resistance coefficient of a fitting, dimensionless
Keff = Effective resistance coefficient of a bend,
dimensionless
L = Length of pipe, m
mg = Mass flow of gas, kg/s Next the Martinelli parameter is computed from Eq. 1:
ml = Mass flow of liquid, kg/s
P = Probability of a successful collision, dimensionless
Qg = Gas volumetric flow, m3/s
Ql = Liquid volumetric flow, m3/s
r = Average radius of a bend, m Using Eq. 2 the dimensionless liquid holdup can be determined:
Rens = No-slip Reynolds number, dimensionless
S = Slip ratio, dimensionless
S̃i = Dimensionless interfacial area gradient
V = Velocity, m/s
Vave = Average plug flow velocity, m/s Using the geometric parameters from the original Taitel and
Dukler paper (Eqs. 3, 4, and 5):
Vm = Two-phase homogenous velocity, m/s
Vmax = Maximum velocity of pipe flow, m/s
VSG = Superficial gas velocity, m/s (gas volume flow over
pipe area)
Vtan = Maximum tangential velocity, m/s
VTB = Taylor bubble velocity, m/s
ṼG = Dimensionless gas velocity
We = Pipe Weber number, dimensionless
Wed = Droplet collision Weber number, dimensionless
x = Two-phase vapor mass quality, dimensionless
Xtt = Martinelli parameter for turbulent liquid, turbulent
gas flow, dimensionless
ε = Void fraction, dimensionless This is checked against the stability parameter for stratified flow
λ = Volumetric quality, dimensionless from the original Taitel and Dukler paper (Eqs. 6 and 7). The super-
μL = Liquid dynamic viscosity, Pa.s ficial gas velocity is calculated by:
μG = Gas dynamic viscosity, Pa.s
μNS = No-slip two-phase dynamic viscosity. Pa.s
ρ = Density, kg/m3
ρg = Gas density, kg/m3
ρNS = Homogenous density, kg/m3
The stability parameters are
ρl = Liquid density, kg/m3
σ = Surface tension, N/m
τ = Shear stress, Pa
ω = Angular velocity, rad/s
ωi = Initial angular velocity, rad/s
ωtot = Total angular velocity, rad/s

This version of this article was updated on 22 May 2018 and contains a number of corrections. | 2018  •   Oil and Gas Facilities 7
Therefore, the flow is not stratified. This is then checked against A
Eq. 8, and as hL̃ is less than 0.35, the flow is annular. Thus the stan-
dard correlations for shear droplets can be applied.
The pipe routing in Fig. 9 has been followed between the con-
denser and the reflux drum. 4D
The initial conditions, upstream of the first elbow (A), are as-
sumed to be at equilibrium. To solve the Eq. 26 we need the ef-
fective velocity heads contribution to the swirl, the void fraction,
and the velocity and density of the mixed phase. There are three 6D
elbows in the system, each on alternate planes, and since they are B
4D
at right angles a swirl will be induced into the flow. The elbows in
this example are assumed to be long radius (r/D=1.5) therefore
(Eq. 21):
C

Fig. 9—Example pipe routing.

Using the Chisholm slip correlation Eq. 24: Now the energy from the first bend is added (Point A); first cal-
culate the swirl of the bend from Eq. 26:

This is dissipated through the inlet piping (Point A to the inlet


[14D]) Eq. 20:
The void fraction is given by Eq. 25:

The summation of the rotational energies is calculated from


The homogenous no-slip density is calculated by Eq. 25 with S Eq. 29:
set to 1:

This equates to 5.4 rotations per second and a maximum tan-


gential velocity of 3.45 m/s (Eq. 27), giving the vertical liquid rise.

The velocity (Vm) is 18.3 m/s


The resulting angular velocity is Eq. 26:

Therefore, the mist eliminator or gas exit nozzle should be


at least 0.62 m above the top of the inlet nozzle to prevent gross
liquid carry-over. This is less than typical design spacing require-
ments, therefore the drum is at low risk of bulk liquid entering
the mesh pad and therefore no need to modify the piping due to
this effect.
The total length, including the gap between the elbows, is 10D If the recommended 10D after the last elbow had been followed,
(4D+6D). The residual angular velocity is calculated from Eq. 20: the height would have been 0.12 m. If the 20D for the multiple
elbow in alternative planes had been followed, the estimated liquid
jet height would be just 0.01m. No quantitative data on the rota-
tional velocities were provided in Rhyne’s presentation; however,
the visualizations did depict a coherent rotational motion.
At this stage, the minimum droplet size can be determined using
Eq. 12:

8 Oil and Gas Facilities  •   2018 | This version of this article was updated on 22 May 2018 and contains a number of corrections.
Next, calculate the Reynolds number and the Weber number: Weber number is given by Eq. 18:

The probability of a successful collision P is given by Eq. 17:

Calculate the maximum droplet size (Eq. 9):

The maximum droplet size resulting from the collisions is Eq. 15:

The minimum droplets size is approximately 20 microns and


the maximum is approximately 2400 microns. This is somewhat
larger than the value calculated in Rhyne’s presentation using a
software package.
Next, we can compare this with the droplet size calculated due
to an elbow and the valve. Based on the Crane method in the sec- Due to the low liquid fraction (ε=0.99), the droplet size has
tion on fittings, the elbow resistance coefficient is: changed little to 471 microns from 467 microns. It is clear that the
globe valve has dramatically reduced the maximum droplet size
and that significant distance would be required to return to equilib-
rium. It is therefore recommended that any letdown valves are po-
The effective diameter is Eq. 14: sitioned away from the separator and that any local isolation valves
are fullbore gate or ball valves.
The dispersion calculated in the example with the valve will be
significantly more difficult to separate. Taking a vertical drum with
a K-value of 0.107 m/s, nearly 100% of the liquid would be en-
trained (this assumes a log normal droplet distribution) into the gas
If a piping globe valve were added, say as a stand-in for a back- and enter the mesh pad. With this level of entrainment, it is highly
pressure regulator, the resistance coefficient and effective diameter likely that the mesh pad would flood (York and Poppele 1963).
would be: Considering the piping in Fig. 9, the gas maldistribution can be
determined as follows:
Taking the initial elbow (A) to the second (B) (L/D=4) and dis-
sipating the distortion using Eq. 31:

Recalculating all parameter results in the following:

Adding the second elbow (B) and calculating the net to the third
Pipe Bend Globe Valve elbow (C) (L/dp=4):

d min 22 19 2

d max 2360 2090 467

Finally, from the third elbow (C) to the vessel (L/D=6):


Assuming that rather than equilibrium, just upstream of the
piping section is a globe valve, the coalescence model can be used
to estimate the droplet size entering the vessel.
The superficial velocity of the gas phase was calculated earlier;
the characteristic velocity is given by Eq. 16:

The vessel specifications given by Rhyne are as follows:

The L/D to the mesh pad is

This version of this article was updated on 22 May 2018 and contains a number of corrections. | 2018  •   Oil and Gas Facilities 9
Now the total residual distortion can be calculated from Eq. 30,
considering the effect of the inlet deflector:

0.15 m

0.15 m

Repeating this for the half pipe and inlet vane device results in
the following, which are compared to Rhyne’s data:

Calculated Rhyne
Device
Distortion Distortion
1.001 m
Deflector 2.19 2.55 1.032 m
Half pipe 1.86 1.86
Inlet Vane 1.62 1. 78

Note that Rhyne presented the distortion for most cases with ideal
pipework. It is assuming that the additive effect for the simple de-
flector is applicable to other cases. 0.203 m

In this example, there is a significant velocity distortion for all


the inlet devices. It is apparent that the distortion predictions are 0.25 m
not particularly accurate. For both the half pipe and the inlet vane
devices, the prediction of the distortion effect is good, while the de-
flector is underpredicted. On closer examination, it is apparent that,
for the deflector and half pipe, the majority of the distortion results 0.45 m
from the inlet device rather than the piping; therefore, it is worth
extracting the net contribution of the inlet to review the quality of
the prediction.
Eq. 30 can be rewritten as:

Fig. 10—Vessel schematic.

Acknowledgments
I would like to express my appreciation to Wood for assistance
The right-hand term is the contribution of the inlet piping, and in publishing this article. Special thanks to Chris Jones, Cam-
the left-hand is that resulting from the inlet device and the change eron ­McCormack, and Steve Harrow for their criticism and edito-
in direction in the separator. rial ­assistance.

References
American Petroleum Institute. 2009. API 686: Recommended Practice for
Machinery Installation and Installation Design.
SPE 21506. 1991. Proseparator: A Novel Separator/Scrubber Design Pro-
The CFD analysis performed by Rhyne indicates that the dis- gram by M. Baker and D. Dick Baker, BP Engineering.
tortion resulting from ideal inlet piping is 2.0 and 2.55 from the Bothamley, M. 2013. Gas/Liquid Separators—Quantifying Separation Per-
non-ideal piping; therefore using the same analysis, the distortion formance—Part 1. Oil and Gas Facilities, August, pp. 21–29.
resulting from the inlet piping is 0.55. Thus, this method has under­ Calderbank, P.H. 1958. Physical Rate Processes in Industrial Fermentation.
estimated the effect but is the correct order of magnitude. With Chemical Engineering Research and Design, pp. 443-463.
more CFD data, it is likely that the dissipation rate, swirl, and the Chester, A.K. 1991. The Modelling of Coalescence Processes in Fluid-
distortion effect of the bends can be improved to more accurately Liquid Dispersions: A Review of Current Understanding. Transactions
reflect reality. of the IChemE, 69(A), pp. 259–270.
A possible explanation for the underprediction of the deflector Chin, R. 2015. The Savvy Separator Series. Oil and Gas Facilities.
is that the curves in Bothamley’s paper appear to be based on a Chisholm, D. 1983. Two Phase Flow in Pipelines and Heat Exchangers.
­downward-facing deflector while the deflector in Rhyne’s presen- Bath, UK: Pitman Press Ltd.
tation is a sideways deflection. Crane Technical Paper 410-2009.
The peak face velocity on the mesh pad is calculated to be Darby, R. 2001. Chemical Engineering Fluid Mechanics.
1.6–2.5 times the average. This is significant; nominally mesh pads Dutta, P. and Nandi, N. 2015. Effect of Reynolds Number and Curvature
are designed at 80% of flood, therefore face velocities more than Ratio on Single Phase Turbulent Flow in Pipe Bends. Mechanics and
1.2 times the average represent a risk of localized flooding. In this Mechanical Engineering, 19 (1), pp. 5–16.
worked example it would be prudent to use an inlet vane device GPSA. 2012. Gas Processing Suppliers Association Databook, 13th Edition.
and/or, if still in the design phase, modify the inlet piping. Heijckers, C. 2012. Flow Conditioning Impact on Separations. SPE Web
Event.

10 Oil and Gas Facilities  •   2018 | This version of this article was updated on 22 May 2018 and contains a number of corrections.
Hinze, J.O. 1955. Fundamentals of the Hydrodynamic Mechanism of Split- Svrcek, W.Y. and Monnery, W.D. 1993. Design Two-Phase Separators
ting In Dispersion Processes. American Institue of Chemical Engi- within the Right Limits. Chemical Engineering Progress, pp. 53–60.
neering Journal, pp. 289–295. Taitel, Y. and Dukler, A. 1976. A Model for Predicting Flow Regime Tran-
Kouba, G.E. 2014. Predicting Droplet Size Distributions. SPE Web Event. sitions in Horizontal and Near Horizontal Gas-Liquid Flows. AIChE J.
Liu, W. and Bai, B. 2015. Swirl Decay in the Gas–Liquid Two-Phase 22, pp. 47–55.
Swirling Flow Inside a Circular Straight Pipe. Experimental Thermal Walsh, J.M. 2016. Savvy Separator Series: The Effect of Shear on Produced
and Fluid Science, Vol. 68, pp. 187–195. Water Treatment. Oil and Gas Facilities.
Miller, D.S. 1978. Internal Flow Systems. Bedford: British Hydrome- York, O.H. 1954. Chemical Engineering Progress, 50, pp. 421.
chanics Research Association. York, O.H. and Poppele, E.W. 1963. Wire Mesh Mist Eliminators. Chem-
Moody, L. 1947. An Approximate Formula for Pipe Friction Factors. Trans. ical Engineering Progress, 59, pp. 45.
ASME, 69(12), pp. 1005–1011.
Radovcich, N.A. and Moissis, R. 1962. The Transition From Two Phase
Bubble Flow to Slug Flow. Massachusetts Institute of Technology. Logan Grim is a senior process engineer at Wood, based in Reading,
Rhyne, L.D. 2015. CFD Optimization of Scrubber Inlet Design. SPE Web UK. During his 9 years with Foster Wheeler, now Wood, he has worked
Event. on the process design of refining, LNG, and upstream facilities. He
Shoham, O. 2006. Mechanistic Modeling of Gas-Liquid Two-Phase Flow in has designed several applications for physical property prediction and
Pipes. Richardson, Texas: Society of Petroleum Engineers. equipment sizing and rating. Grim holds an MEng in chemical engi-
Souders, M. and Brown, G.G. 1934. Design of Fractionating Columns, En- neering from Imperial College London and is a chartered member of the
trainment and Capacity. Industrial & Engineering Chemistry, pp. 98–103. IChemE. He may be contacted at logan.grim@woodplc.com.

This version of this article was updated on 22 May 2018 and contains a number of corrections. | 2018  •   Oil and Gas Facilities 11

Vous aimerez peut-être aussi