You are on page 1of 13

PO172501 DOI: 10.

2118/172501-PA Date: 14-November-14 Stage: Page: 337 Total Pages: 13

A New Comprehensive Model for


Predicting Liquid Loading in Gas Wells
Shu Luo, Mohan Kelkar, Eduardo Pereyra, and Cem Sarica, University of Tulsa

Summary push the liquid to the surface. Ultimately, if the liquid is not
Liquid loading, which can lead to rapid gas-rate decline and can removed promptly, the well will fail to produce gas, which would
even cease gas production, is a common phenomenon found in require it to be abandoned. Although the well will continue to pro-
most mature gas wells. An accurate prediction of the inception of duce for a period of time even after the liquid loading starts, the
liquid loading is of great interest to operators, for the reason that production is unstable, and it could cause damage to the surface
remedial measures can be applied in a timely manner to prevent equipment. Well production under stable conditions is desirable,
such conditions from being realized, thereby extending the pro- with relatively small changes in the gas rate as a function of time.
duction life of the gas well. However, the mechanism that is re- A relatively simple way to detect the onset of liquid loading is
sponsible for liquid loading still remains controversial. In the to observe the change in wellhead pressure and temperature. The
literature, at least three different definitions of liquid loading bottomhole pressure increases because of the accumulated liquid
exist. The first definition is based on the intersection of inflow and in the wellbore, and results in a rise in the wellhead pressure.
outflow curves, the second definition is based on the reversal of Wellhead temperature, on the other hand, drops because of the
entrained liquid droplets, and the third definition is based on the lower volume of high-temperature gas produced from the forma-
reversal of liquid film. These definitions yield different results tion. In addition, nodal analysis is another widely used method to
when predicting the inception of liquid loading. In this paper, a determine liquid loading. As gas rate decreases, the tubing curve
new definition of liquid loading is introduced. This new definition of a gas well exhibits a minimum value, which can be viewed as
is based on the relative contributions of gravity and residual pres- the transition point from annular flow to intermittent flow. The
sure drop, and it is validated by its agreement with air/water ex- intersection of inflow-performance-relationship and outflow-per-
perimental data. formance-relationship curves is often used to check the stability
A new comprehensive model is developed that is based on the of a gas well.
Barnea (1986, 1987) model. For vertical wells, the new model can Many researchers (Duggan 1961; Dousi et al. 2006; Guo et al.
better predict the inception of liquid loading than the widely used 2006; Chupin et al. 2007) have developed techniques to predict
Turner et al. (1969) equation. For deviated wells, it is observed in the inception of liquid loading. Two leading theories have been
the field and in laboratories that liquid loading starts much earlier adopted to predict the inception of liquid loading: liquid-droplet
than in vertical wells, and most liquid-loading equations are not reversal and liquid-film reversal.
appropriate for deviated wells. The new model takes into account
the nonuniform film thickness around the circumferential position Liquid-Droplet Reversal. The droplet-reversal model suggests
of the pipe, and, thus, it improves the prediction of liquid loading that the falling of liquid droplets in the gas core triggers the onset
in deviated wells. The new model is validated through the use of of liquid loading. The widely used droplet model is a correlation
field data in the literature and experimental data obtained at the developed by Turner et al. (1969). The equation suggests that the
University of Tulsa. In addition to the literature data, a new set of inception of liquid loading is related to the minimum gas velocity
field data is reported and used to validate the new model, which to lift the largest liquid droplet in the gas core. The critical gas ve-
shows a significant improvement over the droplet model as well locity can be calculated by Eq. 1, which is corrected by 20% to
as other film models. match their field data.
 
Introduction rðqL  qG Þ 0:25
 G;T ¼ 1:9 : . . . . . . . . . . . . . . . . . . . ð1Þ
As gas wells become marginal, the gas rate drops below a certain q2G
threshold, resulting in liquid accumulation in the wellbore. This is
Later, Coleman et al. (1991a, 1991b) examined the Turner
often termed liquid loading. In the early stages of production, gas
et al. (1969) method with low wellhead pressure data. They con-
and liquid are produced together under annular flow conditions.
cluded that the original Turner et al. derivation (without the 20%
The liquid carried by the gas phase is generally in two forms—liq-
upward adjustment) matches better with their own data. Veeken
uid droplets entrained in the gas core and liquid film attached to
et al. (2010) also compared prediction results from the Turner
the pipe wall. As the reservoir pressure declines, the gas phase is
et al. equation with their offshore data. They proposed a Turner
incapable of lifting sufficient amounts of liquid out of the well-
ratio, which is the ratio of observed critical rate to the Turner
bore. As a result, liquid starts accumulating at the bottom of the
et al. critical rate, to examine the validity of the model. The ratio
well, which creates an increase in backpressure on the formation.
has a value between 1.0 and 2.0, suggesting that the method used
Once the energy builds up in the formation, the increased reser-
by Turner et al. (1969) underestimates the critical velocity.
voir pressure pushes the accumulated liquid to the surface. At this
Veeken et al. (2010) believed that liquid loading occurs because
point, the flow pattern in the production tubing is intermittent
of liquid-film reversal instead of liquid-droplet reversal.
flow, or slug flow, which bears the characteristics of a series of
Several subsequent papers (Belfroid et al. 2008; Veeken et al.
liquid slugs separated by gas pockets. This is reflected in a well
2010) have shown that the droplet model may not be accurate
producing under unstable conditions, with a sharp increase in liq-
enough to predict liquid loading. van’t Westende et al. (2007) car-
uid production followed by a jump in the gas production. Subse-
ried out upward annular air/water experiments to study droplet
quently, the well experiences a sharp decline in gas production,
behavior. It was shown that the droplet size used in the study of
followed by a decrease in liquid production. The cycle continues
Turner et al. (1969) is too large and is unrealistic under air/water
until the buildup in the reservoir pressure is not large enough to
annular-flow condition (van’t Westende et al. 2007). The actual
droplet in annular flow should have a Weber number smaller than
Copyright V
C 2014 Society of Petroleum Engineers 30, whereas the Weber number used in the Turner et al. (1969)
Original SPE manuscript received for review 25 September 2013. Revised manuscript equation, after the adjustment, is 60 (Sutton et al. 2010). They also
received for review 8 July 2014. Paper (SPE 172501) peer approved 14 July 2014. reported that only 0.4% of the droplets have an axial velocity close

November 2014 SPE Production & Operations 337

ID: jaganm Time: 16:13 I Path: S:/3B2/PO##/Vol00000/140020/APPFile/SA-PO##140020


PO172501 DOI: 10.2118/172501-PA Date: 14-November-14 Stage: Page: 338 Total Pages: 13

10 B
Barnea
Superficial Liquid Velocity, Vsl (m/s) Zhang et al. IPR
Turner et al.

1
Stable OPR

Pressure
0.1
Slug Annular
Flow Flow
Unstable
0.01
Liquid loading

0
0.001 0 Flow Rate A
1 10 100
Superficial Gas Velocity, Vsg (m/s)
Fig. 2—Nodal analysis on liquid loading.

Fig. 1—Transition boundaries of different models.


tiple inflow-performance-relationship (IPR) curves are plotted for
a gas well. The change of IPR curves represents the decline in res-
to zero when transition of churn-annular flow occurs, which sug-
ervoir pressure. The OPR curve exhibits a minimum value that
gests that the droplet reversal is not responsible for liquid loading
can be treated as the transition point from stable to unstable flow.
(van’t Westende 2008). In addition, the droplet model is not a func-
On the right of the minimum point, where the gas rate is high, the
tion of inclination angle or the pipe diameter, which is known to
flow is stable, and is usually annular flow. As long as the intersec-
influence the onset of liquid loading (Coleman et al. 1991a).
tion point of IPR and OPR is on the right of the minimum, the
well produces under stable conditions (black points in Fig. 2).
Liquid-Film Reversal. Unlike the Turner et al. (1969) equation,
When the intersection point moves to the left of the minimum
film-reversal models predict the inception of liquid loading by
(red point in Fig. 2), the well produces under unstable conditions
analyzing liquid-film instability. The unified model of Zhang
(liquid-loading conditions).
et al. (2003a, 2003b) is developed on the basis of slug dynamics.
Skopich (2012) collected air/water experimental data in 2- and
The model uses slug flow, which is closely related to all other
4-in.-diameter pipes. The pressure drops were measured at differ-
flow patterns, as the starting point. They constructed the momen-
ent gas velocities by pressure transducers. The data were collected
tum and continuity equations for slug flow by considering the
over several minutes and then converted to time-averaged values.
entire liquid-film region as the control volume. With continuity
The calculation of pressure gradient follows this procedure:
equations and proper closure relationships, such as interfacial-
• Total pressure drop is measured, and the gradient is calcu-
friction factor and liquid entrainment in the gas core, the critical
lated by dividing the pressure drop by the distance between
velocity corresponding to the transition from slug to annular flow
the pressure transducers.
can be obtained.
• Liquid holdup (HL) is measured in the trap section, and the
Barnea (1986, 1987) also developed a unified model to predict
gravitational gradient is calculated by Eq. 2.
the flow-pattern transition. There are two mechanisms that trigger
the transition from annular to intermittent flow in Barnea’s model.  
Dp
The first mechanism is the instability of the liquid film causing ¼ qm g: . . . . . . . . . . . . . . . . . . . . . . . . . . ð2Þ
Dz G
flow reversal near the pipe wall. The second mechanism is a gas-
core spontaneous blockage caused by the large liquid supply. Dif-
Here, qm can be calculated by qLHL þ qG(1 – HL).
ferent from the Zhang et al. (2003a, 2003b) model, Barnea con-
• The time-averaged residual-pressure gradient can be calcu-
structs force balance for annular flow and predicts the transition
lated by subtracting the gravitational pressure gradient from
from annular to intermittent flow by analyzing the interfacial shear-
the total pressure gradient,
stress change in the liquid film. In annular flow, the velocity profile
in the liquid film is parabolic and is directed upward. As the gas ve-      
Dp Dp Dp
locity decreases, the interfacial shear stress between the gas and liq- ¼  : . . . . . . . . . . . . . . . . . ð3Þ
Dz R Dz T Dz G
uid film also decreases. Because of a lack of drag force at the
interface, liquid closer to the pipe wall starts falling back, and the A high-speed camera was used to observe the liquid-film flow-
velocity profile in the liquid film becomes partially directed down- ing direction in the pipe. Inception of liquid loading was then
ward. The film thickness at the transition point can be obtained determined, as the reversal of liquid film was observed. The tran-
when the interfacial shear stress reaches a minimum. Then, the crit- sition points determined from the film reversal (experimental ob-
ical velocity can be calculated by solving the momentum equation servation) and from the minimum pressure drop (when the gravity
and the Wallis (1969) interfacial-friction-factor correlation. and frictional gradients are equal) are quite different and show
The transition boundaries calculated from different models are distinct behaviors.
compared in Fig. 1. It can be seen that the Turner et al. (1969) The pressure drop in the tubing comprises the gravitational
equation gives the smallest critical velocity and the Barnea (1986, and the frictional gradient (assuming that the kinetic-energy gra-
1987) model has the most-conservative results, while the bound- dient is negligible at low gas rates, in which case the frictional
ary from Zhang et al. (2003a, 2003b) is somewhere in between. pressure gradient is equal to the residual pressure gradient in Eq.
The difference in the predicted critical velocity is large between 3). For a given liquid fraction, as the gas velocity increases, the
droplet and film models. The critical velocity in the Barnea model gravitational pressure gradient decreases and the frictional pres-
can be twice that of the Turner et al. equation. sure gradient increases. If the decrease in the gravitational gradi-
ent is larger than the increase in the frictional gradient, the overall
Liquid-Loading Definition pressure drop in the tubing decreases. This continues until the
There are several ways to define the inception of liquid loading. pressure drop reaches a minimum, after which the increase in the
The traditional definition suggests that liquid loading starts when frictional gradient will be greater than the decrease in the gravita-
the minimum pressure drop in the tubing is reached (nodal analy- tional gradient; hence, the overall pressure drop increases with
sis). In Fig. 2, the outflow-performance-relationship (OPR) curve gas rate. The minimum on the pressure-drop curve corresponds to
[also referred to as the tubing-performance relationship] and mul- the gas velocity at which the decrease in the gravitational gradient

338 November 2014 SPE Production & Operations

ID: jaganm Time: 16:13 I Path: S:/3B2/PO##/Vol00000/140020/APPFile/SA-PO##140020


PO172501 DOI: 10.2118/172501-PA Date: 14-November-14 Stage: Page: 339 Total Pages: 13

100 1200 mum pressure drop occurs when the gas velocity is approximately

Change or Pressure Gradient, d(Δp)/d(vsg)


15 m/s. From the experimental results, the liquid-film reversal is
observed when the gas velocity equals 21 m/s (liquid-film flowing

Pressure Gradient, Δp/Δz (Pa/m)


50 1000
direction changes), which is different from the result with the
minimum-pressure-drop definition (15 m/s), whereas the residual
0 800
pressure gradient reaches zero when the gas velocity is approxi-
mately 21 m/s, which is again the same as the film-reversal
–50 600 definition.
Air/water experimental data in the 3-in. pipe (Yuan 2011) are
–100 400 also examined by use of the same method. The high-speed video
is not available for these data to determine the film flowing direc-
–150 d(Δp)/d(vsg)—Gravity 200
tion, but the loading status at different gas velocities is determined
d(Δp)/d(vsg)—Friction with a bubble-tracking technique in the liquid film. In Fig. 4, the
Δp/Δz pressure gradient and liquid-loading status are shown for the 3-in.
–200 0 pipe. Similarly, the down arrows mean that liquid loading is
0 5 10 15 20 25 30 35
Superficial Gas Velocity, vsg (m/s)
observed in the pipe, and the up arrows mean stable annular flow.
It can be seen that the pressure gradient reaches the minimum
value when the gas velocity equals 15 m/s, which is the critical
Fig. 3—Gravitational and frictional pressure gradients.
gas velocity by the traditional definition. However, the inception
of liquid loading on the basis of the Yuan (2011) observation
starts as the gas velocity is between 20 and 22.5 m/s, and thus it
is equal to the increase in the frictional gradient. Fig. 3 (Skopich can be assumed that the liquid-film reversal occurs when the gas
2012) shows an example of changes in the gravitational and fric- velocity is approximately 21 m/s. From the plot of gravitational
tional pressure gradients with respect to gas velocity for air/water and residual pressure gradients for the 3-in. pipe, the residual
flow (this is the derivative of gradient with respect to velocity). pressure gradient becomes negative when the gas velocity equals
The data were collected from experiments conducted in a 40-ft- 21 m/s, which is close to the critical gas velocity by the film-re-
long, 2-in.-diameter pipe. The frictional and gravitational gra- versal definition.
dients were calculated by following the preceding procedure. By These observations are important to understand the differences
examining the incremental change in both the gravitational and between the various definitions. The film-reversal definition indi-
frictional terms against changes in superficial gas velocity, the cates that the liquid loading starts earlier in the larger-diameter
values of the y-axis (on the left-hand side) in Fig. 3 could be pipe, whereas the minimum-pressure-drop definition shows that
determined. The change in gravitational pressure gradient (plotted the liquid loading starts earlier in the smaller-diameter pipe. In
in blue) is negative, and the change in frictional pressure gradient Fig. 6, the critical gas velocity is plotted as a function of pipe di-
(plotted in green) is positive. This means that, as the gas velocity ameter on the basis of experimental data in 2-, 3-, and 4-in. pipes.
increases, the gravitational gradient decreases and vice versa for It can be seen that the trends of critical gas velocity are opposite
frictional gradient. When the absolute values of the two gradients between the two different definitions. The data from Veeken et al.
are equal (at approximately 20 m/s), the overall pressure drop (2010), which are collected primarily in the larger-diameter pipe,
reaches a minimum. show that liquid loading starts earlier in the larger-diameter pipe.
In Fig. 4, pressure gradients at different superficial gas velocities This may prove the validity of the liquid-film definition. It will
are plotted. For the 2-in. pipe, the minimum pressure drop occurs as also be shown later that the liquid-film-reversal model is more
the gas velocity equals 21 m/s. The flowing direction of the liquid appropriate to define liquid loading because it matches better with
film determined from the high-speed video is represented by the up the field data.
and down arrows. It can be seen that the film reversal starts when
the gas velocity reaches approximately 16 m/s, which is different
from the minimum-pressure-drop definition (21 m/s). On the right New Model for Predicting Liquid Loading
of the total pressure-gradient plots in Fig. 4, the gravitational pres- The film-reversal model will be focused on for predicting the
sure gradient calculated from Eq. 2 is plotted in blue and the resid- inception of liquid loading because liquid loading is more likely
ual pressure gradient calculated from Eq. 3 is plotted in green. The caused by liquid-film reversal than liquid-droplet reversal. There
residual pressure gradient is negative at low gas velocity and are several reasons by which we are able to reach this conclusion:
increases as the gas velocity increases. The reason for the negative • The liquid-droplet size required in the Turner et al. (1969)
residual gradient is that the liquid film is flowing in the direction op- equation is too large and is not seen in actual wells (van’t
posite that of the gas at low gas velocity. Zabaras et al. (1986) also Westende et al. 2007).
investigated upward concurrent gas/liquid annular flow in a 2-in. • The percentage of liquid flowing as liquid droplets is very
pipe. Similar experimental results were shown in their paper, and small compared with the percentage of liquid flowing as a
the negative residual-pressure gradient was explained as a result of film under annular-flow condition (Alamu 2012).
the negative wall shear. In Fig. 5, the residual pressure loss is plot- • It is shown that the Turner et al. (1969) equation underpre-
ted against time at different gas velocities. At high gas velocity, the dicts the inception of liquid loading (Veeken et al. 2010).
residual pressure loss is positive and remains almost constant. At • In addition, from our experimental observations, the liquid-
this point, liquid film is flowing upward and the velocity profile in film-reversal definition consistently matches with the experi-
the liquid film is in the same direction as that of the gas. As the gas mental data.
velocity decreases, the residual pressure loss approaches negative In this work, a new model is proposed that is based on the
value and starts fluctuating. This suggests that the liquid film close Barnea (1986, 1987) model, which shows better prediction
to the wall reverses in direction, while the liquid next to the gas results than the droplet model in vertical wells; however, it can-
core is still moving upward. It can be seen that the residual pressure not correctly predict the liquid loading in deviated wells because
loss has an average value less than zero when the gas velocity is of the assumption that the film thickness is constant for all devia-
below 16 m/s. In Fig. 4, it is shown that the residual pressure gradi- tion angles. With a deviation angle, the liquid film at the bottom
ent transitions from negative to positive also at approximately 16 m/ of the pipe is thicker than the film at the top, which is caused by
s. This is consistent with the film-reversal definition, whereas the gravitational force, as shown in Fig. 7. Previous studies by
traditional definition (minimum pressure drop occurs at 21 m/s) Hewitt et al. (1990), Paz and Shoham (1999), and Geraci et al.
overpredicts the critical gas velocity. (2007) investigated the liquid-film structure and the film thick-
For the 4-in. data, similar results can be observed. The pres- ness in the inclined pipe. It is shown that the deviation angle has
sure gradients of the 4-in. pipe can be seen in Fig. 4. The mini- great impact on the thickness distribution of the liquid film.

November 2014 SPE Production & Operations 339

ID: jaganm Time: 16:13 I Path: S:/3B2/PO##/Vol00000/140020/APPFile/SA-PO##140020


PO172501 DOI: 10.2118/172501-PA Date: 14-November-14 Stage: Page: 340 Total Pages: 13

1200 1400 400


2-in. pipe 2-in. pipe

Gravitational Pressure Gradient (Pa/m)

Residual Pressure Gradient (Pa/m)


1200 300
Pressure Gradient, Δp/Δz (Pa/m)
1000

1000 200
800
800 100
600
600 0
400
Film reversal 400 –100

200 200 –200


Minimum pressure drop
Gravitational pressure gradient
Residual pressure gradient
0 0 –300
0 5 10 15 20 25 30 0 5 10 15 20 25 30 35
(a) Superficial Gas Velocity, vsg (m/s) (d) Superficial Gas Velocity, vsg (m/s)

700 700 250


3-in. pipe 3-in. pipe Gravitational pressure gradient

Gravitational Pressure Gradient (Pa/m)


Residual pressure gradient

Residual Pressure Gradient (Pa/m)


600 600 200
Pressure Gradient, Δp/Δz (Pa/m)

500 500 150

400 400 100

300 300 50

200 Liquid loading 200 0


Minimum pressure drop
100 100 –50

0 0 –100
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
(b) Superficial Gas Velocity, vsg (m/s)
(e) Superficial Gas Velocity, vsg (m/s)

700 800 150


4-in. pipe 4-in. pipe
Gravitational Pressure Gradient (Pa/m)

Residual Pressure Gradient (Pa/m)


600 700 100
Pressure Gradient, Δp/Δz (Pa/m)

600
500 50
500
400 0
400
300 –50
300
200 –100
200

100 Film reversal 100 –150


Minimum pressure drop Gravitational pressure gradient
Residual pressure gradient
0 0 –200
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
(c) Superficial Gas Velocity, vsg (m/s) (f) Superficial Gas Velocity, vsg (m/s)

Fig. 4—Total pressure gradient, gravitational pressure gradient, residual pressure gradient, and film-reversal observations in 2-, 3-,
and 4-in. pipes.

Because the thicker film requires more energy to be carried in The experiment of van’t Westende (2008) showed that the crit-
the pipe, it tends to fall back earlier when the gas velocity is ical gas velocity reaches a maximum value at an approximately
insufficient, which means a higher critical velocity should be 30 deviation angle, as shown in Fig. 9. As the deviation angle
observed for deviated wells. changes from 0 to 30 , the critical gas velocity increases and liq-
As shown in Fig. 8, the transition boundaries of the Barnea uid loading occurs earlier. As the angle deviates, yielding a
(1986, 1987) model shift slightly to the left as the deviation angle thicker liquid film on the low end of the pipe, higher gas velocities
increases from the vertical position (0 ) to 30 , while the Turner are required to prevent the falling of the liquid film. On the other
et al. (1969) boundary is a straight line on the left of the Barnea hand, gravitational force in the flow direction decreases as the
boundaries (independent of deviation angle and liquid velocity). pipe deviates, which reduces the critical gas velocity. The com-
This decreasing critical velocity in the Barnea model contradicts bined effect of a thicker film and reduced gravitational gradient
the experimental observations by van’t Westende (2008), who determines the change in the critical gas velocity. When the well
observed an earlier annular- to intermittent-flow transition by is deviated from the vertical position, the influence of the thicker
increasing the deviation angle. Later, Sarica et al. (2013) con- film is greater than the gravitational-gradient change, and, thus,
ducted air/water experiments in 15 and 30 deviated pipes, which the critical gas velocity increases. At approximately 30 , the criti-
also showed that the critical velocity of a deviated pipe is greater cal gas velocity reaches a maximum value. At deviation angles
than that of a vertical pipe. greater than 30 , the critical gas velocity decreases as the

340 November 2014 SPE Production & Operations

ID: jaganm Time: 16:13 I Path: S:/3B2/PO##/Vol00000/140020/APPFile/SA-PO##140020


PO172501 DOI: 10.2118/172501-PA Date: 14-November-14 Stage: Page: 341 Total Pages: 13

1000 25
800

Pressure-Drop Fluctuation (Pa)


600 22

Critical Gas Velocity (m/s)


400
19
200
0
16
–200
–400
13
–600 Film reversal

–800 Minimum pressure drop


0 20 40 60 80 100 120 140 160 180 200 10
1 2 3 4 5
Time (seconds)
Pipe Size (in.)
vsg = 10.57 m/s vsg = 14.65 m/s vsg = 17.14 m/s
vsg = 19.77 m/s vsg = 22.45 m/s vsg = 25.05 m/s
Fig. 6—Effect of pipe diameter on liquid loading.
Fig. 5—Residual pressure-loss fluctuation in 2-in. pipe.
10

Superficial Liquid Velocity, vsl (m/s)


Turner et al.
1

Liquid Gas Gas Barnea


Liquid
0.1

From 0 to 30°, boundry


shifts to the left.
Critical gas velocity
0.01 decreases.

Vertical Well Inclined Well

Fig. 7—Liquid-film thickness in vertical and deviated wells.


0.001
1 10 100
Superficial Gas Velocity, vsg (m/s)
deviation angle increases. In this range, decreasing gravitational Turner et al. Barnea (0°) Barnea (10°)
gradient in the flow direction exceeds the influence of thicker film Barnea (20°) Barnea (30°)
on liquid loading. Accounting for the nonuniform film thickness,
the new model is able to capture this trend so that liquid loading Fig. 8—Boundaries of the Barnea (1986, 1987) model at differ-
in deviated wells can be predicted correctly. ent deviation angles.
Comparison between uniform-thickness film and nonuniform
film is presented in Fig. 10. Material balance is considered to tion angle of the pipe (h). Assuming that the film thickness
ensure the amount of liquid carried in both cases would be the changes linearly with the pipe circumferential position, the area
same. For a constant film thickness, the area of the film is approxi- of the film can be approximated with a trapezoid, and thus can be
mately equal to the area of an expanded rectangle, which can be calculated with Eq. 5.
calculated with Eq. 4. For the nonuniform case, the film thickness
is a function of both the circular pipe position (U) and the devia-
Uniform film

δc
20
δc

Maximum critical
18 gas velocity
Expand to a rectangle
A1 = πD δc
Critical Gas Velocity (m/s)

16 Nonuniform film

δ(Φ, θ)
14 δ(π, θ)
δ(0, θ)

12
Expand to a trapezoid
10
A2 = ½ [δ(0, θ) + δ(π, θ)] πD

Φ – Pipe circumferential position (0–360°)


8 θ – Pipe deviation angle (0–90°)
–10 0 10 20 30 40 50 60 70 80 90
Deviation angle (degrees) (0° = vertical)
Fig. 10—Schematics of uniform and nonuniform liquid-film
Fig. 9—Critical gas velocity at different deviation angles. thicknesses.

November 2014 SPE Production & Operations 341

ID: jaganm Time: 16:13 I Path: S:/3B2/PO##/Vol00000/140020/APPFile/SA-PO##140020


PO172501 DOI: 10.2118/172501-PA Date: 14-November-14 Stage: Page: 342 Total Pages: 13

δ(0, 90) ≈ 0 2
1.8
1.6
Deviation angle θ = 90°

Film Thickness (δc)


1.4
1.2
1
Maximum δ(π, 90) = 2 δc 0.8
0.6
Fig. 11—Liquid-film thickness at horizontal position. 0.4
0.2
0
0 30 60 90 120 150 180 210 240 270 300 330 360
A1 ¼ pDdc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð4Þ Circular Pipe Position (Φ)
90° 60° 45° 30° 20° 15° 0°
1
A2 ¼ ½dð0; hÞ þ dðp; hÞpD; . . . . . . . . . . . . . . . . . . . ð5Þ
2 Fig. 12—Liquid-film-thickness distribution around the pipe cir-
where D is the pipe diameter, dc is the constant film thickness, cumference (U 5 180º at the bottom of the pipe).
d(0,h) is the film thickness at the top of the pipe, and d(p,h) is the
film thickness at the bottom of the pipe.
For both cases, the liquid mass-flow rates in the pipe are iden- Here, U is the pipe circumferential position (ranging from 0 to
tical, and, thus, the area of the liquid film should also be equiva- 360 (0 is at the top and 180 is at the bottom) and h is the pipe
lent. Because A1 equals A2, the relationship between nonuniform deviation angle (ranging from 0 to 90 ). The constant value of a
film thickness and constant film thickness can be expressed as from 0 to 30 is calculated by the equation: 0.5530–0.868, which
is approximately 0.0287.
1 Plotting the film thickness as a function of the pipe circumfer-
dc ¼ ½dð0; hÞ þ dðp; hÞ: . . . . . . . . . . . . . . . . . . . . . ð6Þ ential position, the film distribution at different deviation angles
2
can be seen in Fig. 12. The film thickness is constant at the verti-
Then, the thickest liquid film, which will be adopted to predict the cal position. Then, the film distribution changes rapidly from 0 to
inception of liquid loading in deviated wells, can be calculated by 30 . After 30 , the shape of the film distribution does not vary as
Eq. 6, if the distribution of the film thickness is known. On the ba- much.
sis of the experimental data, the maximum critical gas velocity is Paz and Shoham (1999) conducted two-phase annular-flow
observed at a deviation angle of approximately 30 . From 0 to 30 experiments in deviated pipes and measured the liquid-film thick-
deviation angle, the film thickness on the low end of the pipe ness around the pipe at different deviation angles. A probe was
grows rapidly, as does the critical velocity. After 30 , the film used to measure the local liquid-film thickness (at circumferential
thickness increases much slower, and the film thickness on the positions of 0, 45, 90, and 180 ) around the pipe for deviation
high end of the pipe asymptotically approaches to zero. At hori- angles of 0, 15, 30 and 45 . Only one gas velocity (approximately
zontal position (deviation angle equals 90 ), the top film thickness 18.28 m/s) was used in the experiment. The liquid velocities range
becomes almost zero, which corresponds to d(0,90) in Fig. 11, from 0.006 to 0.06 m/s. The ratio of minimum to maximum film
and the bottom film thickness d(p,90) is the maximum thickness it thickness is calculated to compare with the experimental results at
could reach, which equals 2dc. different liquid velocities. The reason for using this ratio was to
For film distribution along the pipe circumferential position, eliminate the difference between liquid velocities and pipe sizes.
the following empirical equation is used in the new model: In Fig. 13, the measured film thickness is compared with the
model prediction. The solid line is the prediction of the ratio from
dðU; hÞ ¼ ð1  ahcosUÞdc ; . . . . . . . . . . . . . . . . . . . . ð7Þ our model, and the points are the ratio of measured film thick-
nesses at different liquid velocities. It can be seen that the ratios
where are fairly close at different liquid velocities, especially for the
 case of the 45 deviation angle. This indicates that the liquid-film-
0:0287; 0  h < 30 thickness distribution is somewhat independent of the liquid ve-
a¼ : . . . . . . . . . . . . . . . . ð8Þ
0:55h0:868 ; 30  h  90 locity. In the new model shown in Eq. 7, the film thickness is not
a function of liquid velocity. Examining the model’s prediction in
Fig. 13, the liquid film at the 0 deviation angle is constant around
the pipe, so that the ratio of minimum/maximum is unity. From 0
Minimum/Maximum Film-Thickness Ratio

1
to 30 , the ratio decreases to approximately 0.1. Beyond 30 , the
0.9
ratio decreases slowly, and eventually the ratio drops to almost
0.8 zero at 90 .
0.7 In Fig. 14, the film thicknesses predicted by the model are
0.6 compared with the measured local film thickness at 0.03-m/s liq-
0.5
uid velocity (similar results can be obtained from the other five
liquid velocities). The equivalent film thickness (dc) is calculated
0.4
by averaging the film thickness measured at different pipe-cir-
0.3 cumferential positions. The shape of the film distribution around
0.2 the pipe can be seen in solid lines, and the experimental data are
0.1 plotted with solid points. Except for some small discrepancies, the
0
predicted film thickness matches well with the measured liquid-
0 10 20 30 40 50 60 70 80 90 100 film thickness at different deviation angles.
Deviation Angle (Degrees) The proposed model of film-thickness as a function of devia-
Model vsl = 0.006 m/s vsl = 0.012 m/s vsl = 0.024 m/s tion angle can be incorporated into the Barnea (1986, 1987)
vsl = 0.03 m/s vsl = 0.046 m/s vsl = 0.06 m/s model. Because the liquid rate in the gas well is relatively low,
only the liquid-film-instability mechanism will be used to predict
Fig. 13—Minimum/maximum-film-thickness ratio vs. deviation liquid loading. First, the critical film thickness dT (the film thick-
angle. ness when liquid loading starts) can be calculated by

342 November 2014 SPE Production & Operations

ID: jaganm Time: 16:13 I Path: S:/3B2/PO##/Vol00000/140020/APPFile/SA-PO##140020


PO172501 DOI: 10.2118/172501-PA Date: 14-November-14 Stage: Page: 343 Total Pages: 13

2.5 (a) 10

Superficial Liquid Velocity, vsg (m/s)


2 Turner et al.
1
Film Thicknes/δc
1.5 New
model
0.1
1
From 0 to 30°, boundry
shifts to the right.
0.5 Critical gas velocity
0.01
increases.

0
0 30 60 90 120 150 180 210 240 270 300 330 360
Circular Pipe Position (Φ) 0.001
0.1 1 10 100 1000
15° (Model) 15° (Measurement) 30° (Model)
Superficial Gas Velocity, vsg (m/s)
30° (Measurement) 45° (Model) 45° (Measurement)
Turner et al. 0° 10° 20° 30°

Fig. 14—Liquid-film-thickness distribution comparison (vsl 5 (b) 10


0.03 m/s).

Superficial Liquid Velocity, vsgl (m/s)


Turner et al.
1
differentiating the combined momentum equation. This critical
film thickness dT is equal to the maximum film thickness along New
the pipe circumference because the thickest liquid film always ini- Model
tiates the liquid fallback and leads to liquid loading. To calculate 0.1
the critical velocity in deviated wells, the critical film thickness dT
From 30 to 90°,boundry
needs to be converted to equivalent film thickness dc. For exam- shifts to the left.
ple, if the deviation angle is 90 , the critical film thickness dT cal- 0.01 Critical gas velocity
culated from the combined momentum equation is equal to the decreases.
bottom-film thickness. By applying Eq. 7, the film thicknesses at
the top and bottom of the pipe can be estimated. Then, dc can be
calculated with Eq. 9, and this equivalent film thickness can be 0.001
0.1 1 10 100 1000
used to calculate the critical gas velocity. Superficial Gas Velocity, vsg (m/s)
Turner et al. 30° 40° 50°
1 1 60° 70° 80° 89°
dc ¼ ½0 þ dðp; 90Þ ¼ dT : . . . . . . . . . . . . . . . . . . . ð9Þ
2 2
Fig. 15—Transition boundaries of the new model at different
Using a similar procedure, the transition boundaries are calcu- deviation angles (0 to 90º ).
lated at different deviation angles. In Fig. 15, the transition boun-
daries of 0 to 30 and 30 to 90 deviation angles are plotted on the
basis of the new model. Different from Barnea’s original model, Liquid-Loading Prediction in Vertical Wells
the new model’s boundary shifts to the right from 0 to 30 , which A wide range of data from published literature, including field
means liquid loading starts earlier for these cases. From 30 to 90 , data from both onshore and offshore wells, is examined to vali-
the new model’s boundary shifts to the left, and the critical veloc- date the new model. In addition, a new set of data is obtained
ity decreases because of less gravitational force in the flowing from an operator. A comparison between the Turner et al. (1969)
direction. Then, the critical gas velocities predicted from the new equation and the new model will be presented in this work. All
model will exhibit a trend similar to that of the experimental data prediction results can be found in Table 1.
of van’t Westende (2008).
In addition to defining a nonuniform-film-thickness equation Field Data of Turner et al. (1969). Turner et al. (1969) reported
in the new model, a different interfacial-friction-factor equation is field data from gas wells producing liquid in their paper and used
used. The Wallis (1969) correlation of the interfacial friction fac- these data to examine the critical gas velocity. Some of the data
tor is used in the Barnea model, which gives conservative results were designed specifically to determine the critical flow rate and
(the predicted critical velocity is too high) when predicting the other data came from conventional-well-test data. Their data con-
inception of liquid loading. It is possible that interfacial shear sist of a total of 106 gas wells, which are treated as vertical wells
stress calculated with the Wallis (1969) correlation is too small at because no deviation angle is reported. Among all of the gas
the transition from annular to intermittent flow. Many other inter- wells, 37 wells were loaded and 53 wells were unloaded, and the
facial-friction-factor correlations have been developed to match rest of the wells were reported to be questionable. The predicted
the experimental data, such as Fore et al. (2000), Andritsos and critical rate from the Turner et al. (1969) equation is compared
Hanratty (1987), Whalley and Hewitt (1978), and Henstock and with the observed rate of the gas well to determine the liquid load-
Hanratty (1976). Among these correlations, the correlation of ing. If the observed rate is higher than the predicted critical rate,
Fore et al. (2000) exhibited the best match with field data. Fore the gas well is considered to be under stable condition. If the
et al. (2000) showed that the Wallis (1969) correlation is reasona- observed rate is inadequate (smaller than the predicted critical gas
ble for small values of film thickness and is not suitable for a rate), then the liquid should accumulate in this well. They then
larger liquid-film thickness. Because the transition to intermittent checked the predictions with the actual liquid-loading status.
flow is owing to the increasingly larger thickness of the liquid The gas wells reported in Turner et al. (1969) are flowing
film, the Wallis (1969) correlation may not be appropriate for pre- through either the standard production tubing or the annular area
dicting liquid loading. The following correlation of interfacial between the casing and the tubing. For wells producing from the
friction factor by Fore et al. (2000) is used in the new model: annular region, hydraulic diameter should be used to calculate the
    critical gas velocity. However, the Turner et al. (1969) equation,
17; 500 h as shown in Eq. 1, does not depend on the pipe diameter, and,
f1 ¼ 0:005 1 þ 300 1 þ  0:0015 : . . .ð10Þ thus, the pipe size will not affect the prediction results of the
ReG D
equation. In addition, liquid produced from these wells contains

November 2014 SPE Production & Operations 343

ID: jaganm Time: 16:13 I Path: S:/3B2/PO##/Vol00000/140020/APPFile/SA-PO##140020


PO172501 DOI: 10.2118/172501-PA Date: 14-November-14 Stage: Page: 344 Total Pages: 13

Loaded Unloaded Total


Correct Correct Correct
Source Model Forecast Accuracy Forecast Accuracy Forecast Accuracy
2
Turner et al. (1969) Turner (w/c ) 17/37 46% 53/53 100% 70/90 78%
3
Turner (w ) 32/37 86% 48/53 91% 80/90 89%
New Model 32/37 86% 37/53 70% 69/90 77%
Coleman et al. (1991b) Turner N/A N/A N/A N/A 28/56 50%
New Model N/A N/A N/A N/A 54/56 96%
Veeken et al. (2010) Turner N/A N/A N/A N/A 0/67 0%
Barnea N/A N/A N/A N/A 59/67 88%
New Model N/A N/A N/A N/A 61/67 91%
1
Luo et al. Turner 14/49 29% 12/13 92% 26/62 42%
New Model 40/49 82% 10/13 77% 50/62 81%
1
References the current work.
2
w/c = water or condensate.
3
w = water.

Table 1—Model prediction results.

both water and condensate. Turner et al. (1969) suggested that dif- line in the plot is the loading boundary, which represents the transi-
ferent equations should be used for water and condensate, and tion between loading and unloading conditions. This boundary
when both water and condensate exist in the well, only the water divides the plot into two regions, which are the loaded region
equation will be used. (upper region) and the unloaded region (lower region). If the pre-
The observed gas-flow rate is compared with the predictions dicted critical gas velocity from the Turner et al. equation is accu-
from the Turner et al. (1969) equation. The test-flow gas rate rate, loaded wells should fall into the loaded region and unloaded
reported in the paper is converted to the superficial gas velocity, wells should be plotted below the 45 line. It can be seen that all
which is then used to compare with the critical gas velocity they unloaded wells show a good match with predictions (53 out of 53
calculated for every individual well, as shown in Fig. 16. If the gas correct predictions, with 100% accuracy), but a great number of
well is producing under stable condition, the current gas velocity loaded data are in the unloaded region in the plot (17 out of 37 cor-
should be higher than the calculated critical gas velocity. On the rect predictions, with 46% accuracy). This indicates that the Turner
other hand, if the gas well is loaded up, the current gas velocity et al. (1969) equation underestimates the critical gas velocity.
should be lower. In Fig. 16, the loaded wells are plotted as blue In the bottom plot of Fig. 16, the prediction results from the
squares and the stable wells are plotted as red triangles. The 45 new model show great improvement over the Turner et al. predic-
tions. The predictions are correct for most of the loaded wells (32
out of 37 wells). Although some unloading wells are still overpre-
14
Turner et al. equation using water dicted (37 out of 53 correct predictions, with 70% accuracy), they
Calculated Critical Gas Velocity (m/s)

12 and condensate properties are very close to the loading boundary.


The prediction results are shown in Table 1. It is suggested in
10 some studies that only the water property should be used in the
Loaded up Turner et al. (1969) equation (Coleman et al. 1991a). Both results
8 are reported in Table 1 from using only the water property or ei-
ther water or condensate as appropriate. The Turner et al. equation
6 shows better total prediction accuracy; however, it clearly exhibits
Unloaded bias toward predicting the unloading condition. The new model
4 shows good accuracy for both loading and unloading wells, and
the results are better than the Turner et al. (1969) equation in
2
Loading wells terms of conservatism.
Unloading wells
0
0 2 4 6 8 10 12 14 Field Data of Coleman et al. (1991a). Coleman et al. (1991a)
Measured Gas Velocity (m/s) focused on wells experiencing liquid loading with lower reservoir
pressure, which normally are more likely to have a liquid-loading
14
New model problem. The wellhead flowing pressure for these data is below
Calculated Critical Gas Velocity (m/s)

12
500 psi, which is much smaller than the value in the wells
reported by Turner et al. (1969). Fifty-six gas wells, which are all
10 Loaded up vertical wells, are analyzed and all the wells produce through tub-
ing. For the data reported by Coleman et al. (1991a), both water
8 and condensate are produced in gas wells. After comparison, con-
densed water is the primary source of loading liquid and only the
6 water property is used to evaluate the critical gas velocity. Cole-
Unloaded man et al. (1991a) also reported that the liquid amount (gas/liquid
4 ratio less than 22.5 million scf/bbl) in the gas well has very little
impact on determining the critical rate.
2
Loading wells Instead of reporting current gas-flow rate as Turner et al. (1969)
Unloading wells did, Coleman et al. (1991a) reported gas-flow rate when liquid load-
0
0 2 4 6 8 10 12 14 ing was observed in gas wells. They increased wellhead pressure
Measured Gas Velocity (m/s) stepwise to force some of the gas wells into the loading condition.
When a typical exponential rate decline was observed, they reported
Fig. 16—The data of Turner et al. (1969) vs. the model predictions. this rate and compared it with prediction results.

344 November 2014 SPE Production & Operations

ID: jaganm Time: 16:13 I Path: S:/3B2/PO##/Vol00000/140020/APPFile/SA-PO##140020


PO172501 DOI: 10.2118/172501-PA Date: 14-November-14 Stage: Page: 345 Total Pages: 13

20 25
Turner et al. equation Turner et al. equation

Calculated Critical Gas Velocity (m/s)

Calculated Critical Gas Velocity (m/s)


18
16 20
14 Loaded up
12 15 Loaded up
10 Unloaded
8 10
6
Unloaded
4 5
2
Coleman et al. wells Veeken et al. wells
0 0
0 5 10 15 20 0 5 10 15 20 25
Measured Gas Velocity (m/s) Measured Gas Velocity (m/s)
20
New model Fig. 18—The data of Veeken et al. (2010) vs. the predictions of
Calculated Critical Gas Velocity (m/s)

18 the Turner et al. (1969) equation.


16
Loaded up
14 Fig. 17 shows comparison between the critical gas velocities
12 of Turner et al. (1969) and the observed gas velocities of Coleman
et al (1991a). Because Coleman et al. reported critical rate after
10
liquid loading was observed, all the data should fall into the
8 loaded region, which is the region above the 45 line, and the data
6 points should also be close to the loading boundary. However,
Unloaded
4
nearly one-half of the data points are in the unloaded region (28
out of 56 correct predictions, with 50% accuracy). Similar to the
2 data reported by Turner et al., this result again indicates that the
Coleman et al. wells
0 Turner et al. equation underestimates the critical gas velocity. The
0 5 10 15 20
new model’s results are plotted in the bottom plot of Fig. 17 (54
Measured Gas Velocity (m/s) out of 56 correct predictions, with 96% accuracy), and the predic-
tions of the new model are much better than those of the Turner
Fig. 17—The data of Coleman et al. (1991a) vs. the model
et al. (1969) model.
predictions.
Field Data of Veeken et al. (2010). Veeken et al. (2010) exam-
25 ined liquid-loading data in North Sea wells. The field data are col-
Barnea et al. model lected from offshore gas wells with larger tubing sizes. On the
Calculated Critical Velocity (m/s)

basis of their observation, the critical rate exceeds Turner’s pre-


20 diction by an average of 40%.
Loaded up They reported field data from a total of 67 wells, including
15 both vertical and deviated wells. Tubing sizes ranged from 2 to 6
in. The liquids produced were water and condensate, and most of
the liquid came from the condensation of water. However, gas/
10 water ratio or gas/condensate ratio were not reported in the paper,
and a constant gas/water ratio of 5 million scf/STB was assumed
5 Unloaded in the calculations. Other liquid rates were also tested and the
results show insignificant influence on determining the critical gas
velocity, which is consistent with the observations of Turner et al.
0 (1969) and Coleman et al. (1991a, 1991b).
0 5 10 15 20 25
Similar to the data of Coleman et al. (1991a, 1991b), Veeken
Measured Gas Velocity (m/s)
et al. (2010) reported gas rate after liquid loading is observed in the
Vertical (5 wells) 0–20° (17 wells) 20–30° (24 wells) >30° (21 wells)
gas well. Thus, all the data should locate in the region above the 45
25 line if model predictions are correct. However, the Turner et al.
New model (1969) equation failed to predict liquid loading for the data of
Calculated Critical Velocity (m/s)

20 Veeken et al. (0% accuracy), as shown in Fig. 18. This underestima-


tion may be because the critical velocity of Turner et al. is not a func-
Loaded up tion of pipe diameter, and, thus, the Turner et al. equation can be
15 applied only to smaller-pipe-diameter wells, and does not work for
the larger-tubing wells offshore. Deviation angle might be another
10 reason for the poor prediction from the Turner et al. equation.
In Fig. 19, current gas velocities are compared with critical
gas velocities predicted by the Barnea (1986, 1987) model. It can
5 Unloaded be seen that most of the wells are in the loaded region (59 out of
67 correct predictions, with 88% accuracy), which is much better
0
than the results of Turner et al. (1969). To see the influence of the
0 5 10 15 20 25 deviation angle, gas wells are divided into four groups on the ba-
Measured Gas Velocity (m/s) sis of the deviation angles and plotted with different symbols. All
Vertical (5 wells) 0–20° (17 wells) 20–30° (24 wells) >30° (21 wells) the vertical wells are predicted correctly and fall in the upper
region. However, several wells with higher deviation angles
Fig. 19—The data of Veeken et al. (2010) at different deviation (greater than 20 ) are located in the wrong region. It is known
angles. that, on the basis of experimental results, the maximum critical

November 2014 SPE Production & Operations 345

ID: jaganm Time: 16:13 I Path: S:/3B2/PO##/Vol00000/140020/APPFile/SA-PO##140020


PO172501 DOI: 10.2118/172501-PA Date: 14-November-14 Stage: Page: 346 Total Pages: 13

1,200 350
Flow rate (Mcf/D)
Casing pressure (psia)
Tubing pressure (psia)
300
1,000

250

Gas-Production Rate (Mcf/D)


800

Pressure (psia)
200

600

150

400
100

200
50

0 0
0:00 2:24 4:48 7:12 9:36 12:00 14:24 16:48 19:12 21:36 0:00
Time (hours:minutes)

Fig. 20—Gas well without liquid loading at early time.

velocity occurs at approximately the 30 deviation angle (as dis- the gas well produces again with a high-gas-rate spike followed
cussed previously). These inaccurate predictions may result from by another accelerating rate decline. This cycle repeats continu-
the incorrect film thickness used in the Barnea model for a devi- ously with time spans ranging from hours to days. It is known that
ated well. By considering the nonuniform film thickness in devi- tubing and casing pressures diverge when liquid loading starts.
ated wells, the new model is able to improve the predictions for The rise of the liquid level in the tubing causes an additional
the data of Veeken et al. (2010) (61 out of 67 correct predictions, backpressure that will result in an increase in the casing pressure.
with 91% accuracy), as shown in the bottom plot of Fig. 19. It Fig. 20 provides an example of daily gas-well-production data,
should be noted that the wells with high deviation angles in the in which the blue line represents the gas rate, the red line repre-
lower region move upward to the loaded region or very close to sents the casing pressure, and the green line represents the tubing
the loading boundary. This improvement validates the nonuni- pressure. This well is producing without liquid accumulation at
form-film-thickness model. this point because the rate is not decreasing and tubing and casing
pressures are not diverging. At a later time, signs of liquid loading
New Field Data. Additional data are collected from an operator. are seen in this well, as shown in Fig. 21. It can be seen that gas
The data include a total of 62 wells with seven offshore wells rate has more fluctuations and declines at a faster rate. More
from the Gulf of Mexico. Tubing size is 1.995 or 2.441 in. for importantly, tubing and casing pressures show a clear divergence,
most of the wells. Minute-by-minute data are available for the which is denoted with a vertical line. The gas rate is picked at this
gas-production rate and tubing and casing pressures, so that the time as observed gas rate, which is approximately 400 Mcf/D, and
inception of liquid loading can be determined with better accu- it is compared with model predictions. For this set of data, some
racy. If liquid loading is observed in a gas well, gas rate would ex- gas wells are under stable flow, with production data similar to
perience a fast decline and then drop to zero. As pressure builds, those in Fig. 20. Out of all the wells, 49 wells are producing under

900 370 200


Flow rate (Mcf/D)
Casing pressure (psia)
800 Tubing pressure (psia) 180
360
160
700 Pc and Pt diverge
Gas-Production Rate (Mcf/D)

350 140
Casing Pressure (psia)

Tubing Pressure (psia)

600
120
500 340
100
400 330
80
300
320 60

200
40
310
100 20

0 300 0
0:00 2:24 4:48 7:12 9:36 12:00 14:24 16:48 19:12 21:36 0:00
Time (hours:minutes)

Fig. 21—Gas well with liquid loading at later time.

346 November 2014 SPE Production & Operations

ID: jaganm Time: 16:13 I Path: S:/3B2/PO##/Vol00000/140020/APPFile/SA-PO##140020


PO172501 DOI: 10.2118/172501-PA Date: 14-November-14 Stage: Page: 347 Total Pages: 13

Wellhead Tubing Gas Rate Liquid Rate Current Gas Turner et al. New Model
Pressure (psi) Diameter (in.) (Mscf/D) (B/D) Velocity (m/s) Equation (m/s) (m/s)
Unloading Wells
163 1.995 600 0 8.83 6.70 8.60
145 1.995 700 6 11.58 7.11 9.71
250 1.995 710 2.5 6.81 5.40 7.21
210 1.995 950 10 10.85 5.90 8.18
90 1.995 650 2 17.33 9.03 11.84
95 1.995 700 4 17.68 8.78 11.83
110 1.995 720 1 15.70 8.16 10.39
160 1.995 850 0.1 12.75 6.76 8.68
195 1.995 500 2 6.15 6.12 8.08
128 1.995 720 0.6 13.50 7.57 9.37
150 1.995 700 0.1 11.20 6.99 8.96
170 1.995 400 2 5.65 6.56 8.65
210 1.995 950 3 10.85 5.90 7.91
Loading Wells
50 2.441 450 5 14.42 12.12 17.67
80 1.751 250 5 9.73 9.57 12.22
90 1.751 250 5 8.65 9.03 11.54
55 1.995 220 5 9.60 11.55 15.55
60 1.995 190 5 7.60 11.06 14.90
82 1.995 400 5 11.70 9.46 12.80
52 1.995 300 5 13.84 11.88 15.98
170 1.995 580 5 8.19 6.56 8.93
190 1.995 500 3 6.31 6.20 8.32
160 1.995 570 2 8.55 6.76 8.92
215 1.995 550 0.1 6.14 5.83 7.49
130 1.995 400 1.5 7.38 7.51 9.76
130 1.995 380 0.5 7.01 7.51 9.20
102 1.995 400 0.1 9.41 8.48 9.20
100 1.995 300 3 7.20 8.56 11.42
205 1.906 500 0.15 6.41 5.97 6.67
235 2.375 380 1 2.74 5.58 7.50
170 1.995 720 0.1 10.16 6.56 8.42
170 1.995 440 0.5 6.21 6.56 8.05
72 1.995 300 0.1 10.00 10.09 10.93
160 1.995 520 6 7.80 6.76 9.25
130 1.995 500 3 9.23 7.51 10.04
140 1.995 480 0.08 8.23 7.23 7.72
180 1.995 500 0.1 6.66 6.38 8.18
175 1.995 600 2 8.23 6.47 8.53
195 1.995 600 2 7.38 6.12 8.08
130 1.995 390 0.1 7.20 7.51 8.16
130 1.995 450 0.1 8.31 7.51 8.16
170 1.995 450 0.2 6.35 6.56 7.56
150 1.995 400 2 6.40 6.99 9.21
145 1.995 550 2.5 9.10 7.11 9.45
160 1.995 500 1 7.50 6.76 8.63
170 1.995 400 1.5 5.65 6.56 8.54
155 1.995 510 1.1 7.89 6.87 8.81
90 1.995 280 0.5 7.46 9.03 11.03
130 1.995 450 0.5 8.31 7.51 9.20
140 1.995 520 0.1 8.91 7.23 9.27
148 1.995 520 0.1 8.43 7.03 9.02
165 1.995 630 1.4 9.16 6.66 8.65
155 1.995 500 0.5 7.74 6.87 8.43
Table 2—Field data and calculation results.

November 2014 SPE Production & Operations 347

ID: jaganm Time: 16:14 I Path: S:/3B2/PO##/Vol00000/140020/APPFile/SA-PO##140020


PO172501 DOI: 10.2118/172501-PA Date: 14-November-14 Stage: Page: 348 Total Pages: 13

Wellhead Tubing Gas Rate Liquid Rate Current Gas Turner et al. New Model
Pressure (psi) Diameter (in.) (Mscf/D) (B/D) Velocity (m/s) Equation (m/s) (m/s)
Loading Wells

125 1.995 440 0.25 8.45 7.66 8.95


120 1.867 450 0.09 10.27 7.81 8.32
120 1.995 400 0.75 8.00 7.81 9.80
183 1.995 750 0.1 9.83 6.32 8.12
140 1.995 480 1.5 8.23 7.23 9.41
165 1.995 590 3 8.58 6.66 8.92
160 1.995 600 2.5 9.00 6.76 9.00
110 1.867 490 0.1 12.20 8.16 10.14
115 1.995 480 0.1 10.01 7.98 10.21
Table 2 (Continued)—Field data and calculation results.

loading conditions and 13 wells are producing under stable condi- 1. Various definitions of liquid loading are examined. The most
tions. The field data are reported in Table 2. appropriate definition of liquid loading is to assume that the
The new data are similar to the data of Turner et al. (1969). If pre- liquid loading starts when the liquid-film reversal starts.
diction is accurate, blue squares should be seen in the upper region 2. By use of the experimental data, it is found that, depending on
and red triangles in the lower region. The results of the Turner et al. the definitions of liquid loading, different values of gas veloc-
equation are plotted in Fig. 22. Because of underestimation of the crit- ities can be determined as critical velocities at which liquid
ical gas velocity, most of the loading wells are located in the unload- loading starts. Interestingly, with the traditional definition of
ing region (14 out of 49 correct predictions, with 29% accuracy). The liquid loading, the liquid loading starts earlier in smaller tub-
new model’s prediction results can be seen in the bottom plot of Fig. ing. In contrast, with the film-reversal definition, the liquid
22 (with a total accuracy of 81%). Comparing the statistics in Table 1, loading starts earlier in the larger tubing.
the new model again outperforms the Turner et al. (1969) equation for 3. A relatively simple method for determining the transition point is
predicting the liquid loading. developed on the basis of the film-reversal definition. Because it
is difficult to observe the film reversal experimentally, a residual
Conclusions pressure-gradient calculation can be used to determine the transi-
In this paper, the liquid-loading problem in gas wells is investi- tion point. It is observed that the residual pressure gradient
gated. A new definition of the inception of liquid loading is dis- becomes negative when the film reversal starts.
cussed, and a new model is proposed to predict the onset of liquid 4. A nonuniform-film-thickness equation is developed to predict the
loading. The following conclusions can be reached: film thickness at different deviation angles. The predictions from
the equation agree with the thickness measurements. The film-
thickness distribution (or the ratio of minimum to maximum film
20 thickness) is a function of the deviation angle and the circumferen-
Turner et al. equation tial position of the pipe only, and is not a function of liquid velocity.
Calculated Critical Gas Velocity (m/s)

18
5. A new model, based on the film model of Barnea (1986, 1987),
16 is developed to predict the inception of liquid loading in both
14 vertical and deviated wells, and it accounts for variable film
Loaded up
thickness in deviated wells. A different interfacial-friction-fac-
12
tor correlation is used in the model.
10 6. The new model is compared with a large quantity of field data.
8 It can be consistently observed that the new model is able to
6 predict the transition point more accurately compared with the
Turner et al. (1969) equation.
4 Loading wells
Unloaded
2 Loading wells (Gulf) Nomenclature
0
Unloading wells D ¼ pipe diameter, in.
0 2 4 6 8 10 12 14 16 18 20 h ¼ mean film thickness, in.
Measured Gas Velocity (m/s) HL ¼ liquid holdup
20 ReG ¼ gas Reynold’s number
New model vG,T ¼ critical gas velocity, ft/sec
Calculated Critical Gas Velocity (m/s)

18
vsg ¼ superficial gas velocity, m/s
16 vsl ¼ superficial liquid velocity, m/s
14 dc ¼ constant film thickness, dimensionless
Loaded up
dT ¼ critical film thickness, dimensionless
12
h ¼ pipe deviation angle, degrees
10 qG ¼ gas-phase density, lbm/ft3
8 qL ¼ liquid-phase density, lbm/ft3
6 qm ¼ mixture density, lbm/ft3
r ¼ surface tension, dynes/cm
4
Unloaded
Loading wells U ¼ pipe circumferential position, degrees
2 Loading wells (Gulf)
Unloading wells Acknowlegments
0
0 2 4 6 8 10 12 14 16 18 20 Funding for this project is provided by the Research Partnership to
Measured Gas Velocity (m/s) Secure Energy for America, Chevron, Shell, ConocoPhillips, Mara-
thon, Nalco, and MultiChem. This research is also supported by the
Fig. 22—New field data vs. the model predictions. University of Tulsa.

348 November 2014 SPE Production & Operations

ID: jaganm Time: 16:14 I Path: S:/3B2/PO##/Vol00000/140020/APPFile/SA-PO##140020


PO172501 DOI: 10.2118/172501-PA Date: 14-November-14 Stage: Page: 349 Total Pages: 13

References tiphase Flow 33 (6): 595–615. http://dx.doi.org/10.1016/j.ijmulti


Andritsos, N. and Hanratty, T.J. 1987. Influence of interfacial waves in phaseflow.2006.12.006.
stratified gas-liquid flows. AIChE J. 33 (3): 444–454. http://dx.doi.org/ van’t Westende, J.M.C. 2008. Droplets in annular-dispersed gas-liquid pipe-
10.1002/aic.690330310. flows. PhD thesis, Delft Technical University, Delft, The Netherlands.
Alamu, M.B. 2012. Gas-Well Liquid Loading Probed With Advanced Veeken, K., Hu, B., and Schiferli, W. 2010. Gas-Well Liquid-Loading-
Instrumentation. SPE J. 17 (1): 251–270. SPE-153724-PA. http://dx. Field-Data Analysis and Multiphase-Flow Modeling. SPE Prod & Oper
doi.org/10.2118/153724-PA. 25 (3): 275–284. SPE-123657-PA. http://dx.doi.org/10.2118/123657-PA.
Barnea, D. 1986. Transition from annular flow and from dispersed bubble Wallis, G.B. 1969. One Dimensional Two-Phase Flow, Vol 1. New York:
flow—unified models for the whole range of pipe inclinations. Int. J. McGraw-Hill.
Multiphase Flow 12 (5): 733–744. http://dx.doi.org/10.1016/0301-932 Whalley, P.B. and Hewitt, G.F. 1978. The Correlation of Liquid Entrain-
2(86)90048-0. ment Fraction and Entrainment Rate in Annular Two-Phase Flow. Har-
Barnea, D. 1987. A unified model for predicting flow-pattern transitions well, Oxfordshire: UKAEA Atomic Energy Research Establishment.
for the whole range of pipe inclinations. Int. J. Multiphase Flow 13 Yuan, G. 2011. Liquid loading problems of natural gas wells. MS thesis,
(1): 1–12. http://dx.doi.org/10.1016/0301-9322(87)90002-4. University of Tulsa, Tulsa, Oklahoma.
Belfroid, S.P.C., Schiferli, W., Alberts, G.J.N. et al. 2008. Prediction Onset Zabaras, G., Dukler, A.E., and Moalem-Maron, D. 1986. Vertical upward
and Dynamic Behaviour of Liquid Loading Gas Wells. Presented at the cocurrent gas-liquid annular flow. AIChE J. 32 (5): 829–843. http://
SPE Annual Technical Conference and Exhibition, Denver, 21–24 Sep- dx.doi.org/10.1002/aic.690320513.
tember. SPE-115567-MS. http://dx.doi.org/10.2118/115567-MS. Zhang, H.-Q., Wang, Q., Sarica, C. et al. 2003a. Unified Model for Gas-Liquid
Chupin, G., Hu, B., Haugset, T. et al. 2007. Integrated Wellbore/Reservoir Pipe Flow via Slug Dynamics—Part 1: Model Development. J. Energy
Model Predicts Flow Transients in Liquid-Loaded Gas Wells. Pre- Resour. Technol. 125 (4): 266–273. http://dx.doi.org/10.1115/1.1615246.
sented at the SPE Annual Technical Conference and Exhibition, Ana- Zhang, H.-Q., Wang, Q., Sarica, C. et al. 2003b. Unified Model for Gas-Liq-
heim, California, USA, 11–14 November. SPE-110461-MS. http:// uid Pipe Flow via Slug Dynamics—Part 2: Model Validation. J. Energy
dx.doi.org/10.2118/110461-MS. Resour. Technol. 125 (4): 274–283. http://dx.doi.org/10.1115/1.1615618.
Coleman, S.B., Clay, H.B., McCurdy, D.G. et al. 1991a. A New look at
predicting gas-well load-up. J Pet Technol 43 (3): 329–333. SPE- Shu Luo holds a BS degree in petroleum engineering from
20280-PA. http://dx.doi.org/10.2118/20280-PA. China University of Petroleum, Beijing, and MS and PhD
Coleman, S.B., Clay, H.B., McCurdy, D.G. et al. 1991b. Understanding degrees in petroleum engineering from the University of Tulsa.
gas-well load-up behavior. J Pet Technol 43 (3): 334–338. SPE- He is a member of SPE, and placed second in the doctoral di-
20281-PA. http://dx.doi.org/10.2118/20281-PA. vision of the SPE International Student Paper Contest at the
Dousi, N., Veeken, C., and Currie, P.K. 2006. Numerical and Analytical 2013 SPE Annual Technical Conference and Exhibition.
Modelling of the Gas Well Liquid Loading Process. SPE Prod & Oper Mohan Kelkar is the Williams Endowed Chair Professor and Chair
21 (4): 475–482. SPE-95282-PA. http://dx.doi.org/10.2118/95282-PA. of the Petroleum Engineering Department at the University of
Duggan, J. 1961. Estimating Flow Rates Required To Keep Gas Wells Tulsa. His research interests include reservoir characterization
Unloaded. J Pet Technol 13 (12): 1173–1176. SPE-32-PA. http://dx. and modeling and production optimization. Kelkar is a coauthor
doi.org/10.2118/32-PA. of the SPE book Applied Geostatistics for Reservoir Characteriza-
Fore, L.B., Beus, S.G., and Bauer, R.C. 2000. Interfacial friction in gas–- tion and the author of the book Natural Gas Production Engi-
liquid annular flow: analogies to full and transition roughness. Int. J. neering, published by PennWell. He holds an MS degree in
petroleum engineering and a PhD degree in chemical engi-
Multiphase Flow 26 (11): 1755–1769. http://dx.doi.org/10.1016/
neering, both from the University of Pittsburgh. Kelkar is an SPE
S0301-9322(99)00114-7. Distinguished Member and recipient of the 2009 SPE Distin-
Geraci, G., Azzopardi, B.J., and van Maanen, H.R.E. 2007. Effect of incli- guished Achievement Award for Petroleum Engineering Faculty.
nation on circumferential film thickness variation in annular gas/liquid
flow. Chem. Eng. Sci. 62 (11): 3032–3042. http://dx.doi.org/10.1016/ Eduardo Pereyra is an assistant professor at the McDougall
School of Petroleum Engineering, and the associate director
j.ces.2007.02.044.
of the Fluid Flow Project, at the University of Tulsa. His research
Guo, B., Ghalambor, A., and Xu, C. 2006. A Systematic Approach to Pre- interests are multiphase-flow systems and transport and flow-
dicting Liquid Loading in Gas Wells. SPE Prod & Oper 21 (1): 81–88. assurance and -separation technologies. Pereyra has auth-
SPE-94081-PA. http://dx.doi.org/10.2118/94081-PA. ored or coauthored several refereed journal and conference
Henstock, W.H. and Hanratty, T.J. 1976. The interfacial drag and the papers in his area of interest. He holds BE degrees in mechani-
height of the wall layer in annular flows. AIChE J. 22 (6): 990–1000. cal engineering and systems engineering from the University of
http://dx.doi.org/10.1002/aic.690220607. Los Andes, Venezuela, and MS and PhD degrees in petroleum
Hewitt, G.F., Jayanti, S., and Hope, C.B. 1990. Structure of thin liquid engineering from the University of Tulsa.
films in gas-liquid horizontal flow. Int. J. Multiphase Flow 16 (6): Cem Sarica, F.H. “Mick” Merelli/Cimarex Energy Professor of
951–957. http://dx.doi.org/10.1016/0301-9322(90)90100-W. Petroleum Engineering at the University of Tulsa, is currently
Paz, R.J. and Shoham, O. 1999. Film-Thickness Distribution for Annular serving as the director of three industry-supported consortia at
Flow in Directional Wells: Horizontal to Vertical. SPE J. 4 (2): 83– the university: fluid flow, paraffin deposition, and horizontal-
91. SPE-56233-PA. http://dx.doi.org/10.2118/56233-PA. well artificial-lift projects. His research interests are production
Sarica, C., Yuan, G., Sutton, R.P. et al. 2013. An Experimental Study on engineering, multiphase flow in pipes, flow assurance, and
Liquid Loading of Vertical and Deviated Gas Wells. Presented at the horizontal wells. Sarica has authored or coauthored more
than 150 publications. He holds BS and MS degrees in petro-
SPE Production and Operations Symposium, Oklahoma City, Okla-
leum engineering from Istanbul Technical University and a PhD
homa, USA, 23–26 March. SPE-164516-MS. http://dx.doi.org/10. degree in petroleum engineering from the University of Tulsa.
2118/164516-MS. Sarica currently serves as a member of the SPE Projects, Facili-
Skopich, A. 2012. Experimental Study of Surfactant Effect on Liquid ties, and Construction Advisory Committee; as a board mem-
Loading in 2-in and 4-in Diameter Vertical Pipes. MSE thesis, Univer- ber of the Flow Assurance Section; as chair of the SPE
sity of Tulsa, Tulsa, Oklahoma. International Production & Operations Award Committee;
Sutton, R., Cox, S., Lea, J. et al. 2010. Guidelines for the proper applica- and on the Projects, Facilities and Construction Program Sub-
tion of critical velocity calculations. SPE Prod & Oper 25 (2): committee for the SPE Annual Technical Conference and Exhi-
182–194. SPE-120625-PA. http://dx.doi.org/10.2118/120625-PA. bition. He has served previously as a member of the SPE
Turner, R.G., Hubbard, M.G., and Dukler, A.E. 1969. Analysis and Pre- Production & Operations and SPE Books committees and as
chair of the SPE International Projects, Facilities, and Construc-
diction of Minimum Flowrate for the Continuous Removal of Liquids
tion Award Committee, and he was a member of the SPE
from Gas Wells. J Pet Technol 21 (11): 1475–1482. SPE-2198-PA. Journal Editorial Board between 1999 and 2007. Sarica is the
http://dx.doi.org/10.2118/2198-PA. recipient of the 2010 SPE International Production and Opera-
van’t Westende, J.M.C., Kemp, H.K., Belt, R.J. et al. 2007. On the role of tions Award. He was also recognized as an SPE Distinguished
droplets in cocurrent annular and churn-annular pipe flow. Int. J. Mul- Member in 2012.

November 2014 SPE Production & Operations 349

ID: jaganm Time: 16:14 I Path: S:/3B2/PO##/Vol00000/140020/APPFile/SA-PO##140020

You might also like