Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

PETROLEUM SOCIETY PAPER 2002-111

CANADIAN INSTITUTE OF MINING, METALLURGY & PETROLEUM

A Methodology to Determine the Liquid


Column Height of Intermittent Gas Lift Wells
S.C. Moy Cheung, S. Gasbarri
PDVSA Intevep

This paper is to be presented at the Petroleum Society’s Canadian International Petroleum Conference 2002, Calgary, Alberta,
Canada, June 11 – 13, 2002. Discussion of this paper is invited and may be presented at the meeting if filed in writing with the
technical program chairman prior to the conclusion of the meeting. This paper and any discussion filed will be considered for
publication in Petroleum Society journals. Publication rights are reserved. This is a pre-print and subject to correction.

ABSTRACT wells in Lake Maracaibo. The best solution was based on


The estimation of the liquid column height of a well is Aziz’s and Wallis’ correlations as higher accuracies were
one of the most complex problems in the petroleum achieved in reproducing the experimental data from the
industry. This complexity is mainly due to the behaviour field. With this new methodology, the knowledge of liquid
of the wells’ two-phase flow, which varies with pressure, column heights will enable more efficient oil production
temperature, viscosity, and other factors. Knowing the in intermittent gas lift wells.
liquid column height of a well helps determine with
higher precision the amount of liquid holdup in the well INTRODUCTION
and to optimise the oil’s production per cycle. Although Many wells do not have enough energy to lift all the
this height can currently be estimated for wells produced fluids from the bottomhole to the surface. Artificial lift
with various lift methods, it still cannot be estimated for methods are required to supply hydraulic energy to the
intermittent gas lift wells. fluids. Gas lift is one of the most reliable methods,
This paper describes a new methodology to determine particularly in offshore fields. Continuous gas lift is used
the liquid column height of wells produced using on first stages after natural flow.
intermittent gas lift. Two pressure drop correlations were Intermittent gas lift has proven to be more efficient
studied and modified to compute liquid column heights; when production declines below certain level. It works as
Hagendorn and Brown’s, and a combination of Aziz’s a positive displacement mechanism where the gas pushes
and Wallis’. These correlations were programmed and rapidly the fluid column accumulated after some shut-in
tested using experimental data obtained from twelve time. To determine the best performance for this method

1
it is necessary to know all the wells’ characteristics as The flow through a static liquid column creates a
well as to have an appropriate designing tool. Knowing special type of multiphase flow, termed ZNLF (Zero Net
the liquid column height is one of most important aspect Liquid Flow). Most models are based on holdup
for designing the method in order to inject the correct calculations, using different velocities between gas and
amount of gas required for lifting the column, increasing liquid phases. The gas velocity is based on modelling the
oil production, and saving gas lift pressure and flowrate. movement of the Taylor bubble [2].

Assuming a theoretical height of the accumulated


liquid column might as well be not precise for the fluid vg = Co * Vm + V∞T .......................................................(1)
column composed by liquid and gas. After a literature
review and a development of a predicting model, this This equation is valid from bubble flow to churn.
paper describes a methodology to determine with better Considering the liquid velocities equal zero, the equation
precision the height of the fluid column in an intermittent for calculating liquid hold-up is the following [2]:
gas lift well. Experimental data from several wells
located in Lake Maracaibo were used for tune the model. vsg vsg
Hlo = 1 − = ...................................(2)
vs Co * vsg + V∞T
LITERATURE REVIEW
In case of having a liquid velocity different from zero,
Several multiphase flow models have been developed
the equation for the liquid hold-up is given by the
to predict pressure gradients in different problems found
following equation:
in the oil industry. In almost all-vertical wells, gas flows
along the producing liquid, and different flow patterns are
encountered as the fluids density changes with pressure Qg
Hlo = ..................................(3)
and temperature throughout the well. The purpose of the Co * (Qg + Ql ) + A * V∞T
optimisation in these cases is to obtain the bottomhole
flowing pressure, given the wellhead pressure or Considering slug flow, Wallis described that the
viceversa. Although in this study the objective is to calculation of the rising velocity for a bubble inside of the
obtain the column height, pressure gradient methods used vertical well with static liquid is governed by the
in oil well production may help achieve the goal. Even interaction between its buoyancy and other forces that act
though it does not include flow pattern predictions, one of on the bubble, originating its form and movement. If the
the most used empirical method for the theoretical gas viscosity in the bubble is negligible, there are three
pressure gradient calculation is the Hagendorn and forces that are important for the study of the rising
Brown correlation [1]. This method was included in this velocity of the bubble, i.e., inertia, liquid viscosity and
study since it is believed to be more accurate. liquid surface tension [3].

There are several studies developed to predict The system is dominated by inertia when viscosity and
bottomhole pressures in the tubing casing annulus for surface tension are negligible. The rising velocity is given
sucker rod and progressive cavity pumps. In these cases, by the equation of the Taylor bubble velocity [3]:
while some of the producing gas flows through the
annulus bypassing a pseudo static liquid, the fluid level
g * D * ( ρL − ρg )
might be determined using acoustic devices. Although the V ∞T = C * .....................................(4)
geometry is different, comparing the tubing with the ρL
annular section described before, this phenomenon is
The value of the parameter C may be estimated using
somehow similar to our liquid accumulation in
the following equations 5-7.
intermittent gas lift.

2
2 3/ 2
g * D * ( ρL − ρg ) σ * ρL
N = ............................................. (5) NAr = .......................... (11)
E σL 2 1/ 2 1/ 2
µL * g * ( ρL − ρg )

Where C is defined as a function of Ne and Nv.


3
g * D * ρL * ( ρL − ρg )
N = ................................ (6) Co is the flow distribution coefficient and has an
V µL approximated value to 1,2 for flowpatterns from bubble
to churn [3] . While Zuber and Findlay define this
parameter to be 1.2 for typical properties (σ=20dynes/cm,
0.5
C = Fr = 0.345 * (1 − exp(−0.029 * N )) * ρl=1gr/cm3) from bubble flow to slug flow [4].
V ............ (7)
* (1 − exp((3.37 − N ) / m))) V∞T, is the Taylor bubble velocity, considered for the
E
hold-up calculation in the liquid column. The
Additionally, Table 1 shows an estimation of the dimensionless numbers such as Eo (Eotvos), M (Mach),
parameter m for different values of parameter Nv. Fr (Froud), etc., describe the phenomenon of the rising
velocity of the Taylor bubble [2].
Experiments obtained by Dumitrescu gives a
somewhat different result, C = 0.346. After different Aziz et al. [5] stated that Taylor bubble rise velocity in
experiments, a similar value was obtained by White and a stagnant column is [2]:
Beardmore [3].

If the system is dominated by viscosity, the values for V ∞ T = f (C , g , D , ρL , ρg )


C for circular vertical pipes given by Wallis is 0.010,
Wallis studied the behaviour of the rising slug for
while White and Beardmore estimated 0.0096.
different fluids in different diameters for vertical pipes [3].
Finally, if the system is dominated by surface tension Where NE, called by Harmathy as dimensionless number
the bubble does not move at all. The static interface Eotvos, and NV is another dimensionless number. It is
adopts an individual form, in such a way that the important to emphasise that An et al. defines a more
hydrostatic forces are completely balanced with the accurate equation for the gas superficial velocity in the
superficial forces. For circular vertical pipes, the value of transition regime from bubble to slug under their
C depends on two dimensionless numbers, the Eotvos experimental condition [2].
and the Bond numbers [3].

There is a generic solution that is governed under the g * σ * ( ρL − ρg )


three parameters, which is presented by dimensionless vsg = 0.3825 * 4 ft/s ...................... (12)
2
numbers. The result obtained from experimental data ρL
shows the following solutions [3].

FIELD DATA
Inertia Nv > 300 and Ne > 100 , C = 0.345 ................ (8)
Around the year of 1996, there was a special need for
improving gas lift installation efficiency in Lake
Viscosity Nv < 2 and Ne > 100 , C = 0.01 * Nv .......... (9) Maracaibo. A number of intermittent gas lift tests were
performed in the experimental well, CEPRO, located in
Tía Juana in order to determine liquid fallback and gas
Surface Tension Ne = 3.37 consumption for different well characteristics [6]. The
2
Nv = 6.2 * NAr ........................................................... (10) dynamic behaviour of the reservoir when being exposed
to cyclical periods of production in intermittent lift was

3
theoretically analysed by Gasbarri et al. [7] . This • Liquid fallback is neglected during the
behaviour can also be studied with bottomhole pressure accumulation stage inside the production pipe
surveys, taken by means of a special procedure.
• The combination of liquids, water and oil, are
Bottomhole flowing pressure tests were made for several
considered homogenous
intermittent gas lift wells in La Salina, north of Lake
Maracaibo, Venezuela [8]. This study consisted of running The analysis of the column was considered as
surveys of bottomhole pressure and temperature with a described in the scheme shown in Fig. 2. Once entered all
new technique for intermittent gas lift [9]. the necessary data for the calculations, the following
procedure is used.
The wells were selected to help explain the dynamic
behaviour of different valve types, as well as the capacity a. Define the number of pressure segments to divide the
of contribution of several reservoirs. In order to whole liquid column within the production pipe. This
determine the fluids dynamic gradient for the different is necessary since fluids densities vary throughout
well characteristics, the plot of the pressure and the column and hence, flow pattern and liquid
temperature at two depths for the dynamic conditions of holdup.
three different cycles was considered, see Fig. 1. The The following equation is used to calculate the
value of pressure and temperature for each depth is pressure drop by segments. This value will be a constant
obtained from the average of the three cycles. in all the segments and iterations.
The real pressure gradient is obtained for twelve different
wells as follows, Pwf − Pwh
∆Pseg = ....................................................(14)
N
∆P P 2'− P1'
= .............................................................(13) b. The pressure average for each segment will be used
∆H H 2 − H1
to calculate the fluid properties.
The values of the real pressure gradient, obtained with For the first pressure segment, i = 1
equation 13, are compared with the values estimated by
the model developed on next chapter in this study.
Additionally to the values obtained with these dynamic ∆Pseg
P (i , j ) = Pwf + .................................................(15)
surveys, the main characteristics of each well such as: 2
water cut, gas liquid ratio, oil production and API, the
For the rest of the pressure segment, i ≠ 1
diameter of the pipe, among others, were also obtained.

MODEL DESCRIPTION P (i , j ) = Pwf − i * ∆Pseg + 0.5 * ∆Pseg .....................(16)

The following assumptions were considered to develop c. As a first approximation, assuming an average
the model. pressure gradient, the level of the fluids column may
• The producing liquid and gas flowrate are stables be estimated.
during in the period of time.

• The temperatures and pressures are considered  Pwf − Pwh 


Length = H −  ...........(17)
assume
constants throughout small segments of the  Gradreal _ assume 
column.
In order to estimate the average temperature for each
• The distance between the two depths measured
section, let’s assume equal pressure gradient for all
defines the column of fluid analysed.
segments and estimate the length of the segments (ft).

4
H − Lenght assume For i ≠ 1,
∆Zassume = .............................. (18)
N
Y (i, j ) = Y (i − 1, j ) + 0.5 * ∆Z (i − 1, j ) + 0.5 * ∆Z (i, j ) ... (25)
Where i = 1
h. The total fluids column can be obtained adding the
T 1(i , j ) = Twf ................................................................ (19) length of each segment.

This procedure was programmed in Visual BASIC for


Where i ≠ 1 Excel due to its flexibility for handling data and
equations.
T 1(i , j ) = T 2(i − 1, j ) ..................................................... (20)
RESULTS
The methodology to estimate the liquid column inside
∆Zassume
T 2(i , j ) = T 1(i , j ) − (Twf − Twh ) .............. (21) an Intermittent Gas Lift well was programmed. For the
H
pressure gradient calculation, two different models
described in the literature review were used: the empirical
T (i , j ) = 0.5 * (T 1(i , j ) + T 2(i , j )) ................................ (22) method of Hagendorn and Brown (H&B), Wallis and
Aziz et al correlation.
d. The fluids properties at pressure and temperature for
The model was first tested using the H&B option for a
each segment should be calculated.
wide variety of well characteristics. A comparison of
e. A pressure gradient model has to be applied. Two results between the model and a standard simulator
different models described in the literature were program for calculating bottomhole pressure gradient
evaluated, Hagendorn and Brown’s, and a using H&B was made. The simulator program has to be
combination of Aziz’s and Wallis’. run manually varying the length of the column until
obtaining the bottomhole pressures desired. By other
Calculation of the length for each segment is estimated
hand, the model designed in this paper directly calculates
as follows.
the length of the column of the two-phase mixture.

∆Pseg Both programs were run under the same conditions.


∆ Z (i , j ) = ....................................................... (23) The results of this first test show that the values of
dP
calc bottomhole pressure obtained between the two programs
dZ
are similar. The average deviation is 3% over the 24
f. Check if the length of each segment has changed sample wells, only for high flowrate the deviation
during the iteration over a reasonable tolerance. In increases. These results show that the programming of
this case, another iteration has to be run. It takes less the model with the H&B option works properly. This
than ten iteration to match a sound value. includes model procedure and fluid properties
calculations.
g. Estimate the height of the centre of each segment to
be used for the average properties. Once verified the good operation of the program, the
twelve wells were run and compared with the
For i = 1, experimental data. Next, results for the two different
pressure gradient calculations are shown.
∆Z (i , j ) The following statements were observed in the details
Y (i , j ) = .......................................................... (24)
2 of the model runs for the experimental data. While the
pressure gradient decreases in the flow direction, the

5
volume of the fluids in the column increases. Down in the obtained in the field are different from the obtained with
bottom of the well, the temperature and the pressure are the model. Errors originated by the logging tool may be
highest, with lowest viscosity. The temperature and the possible as well as the not consideration of the transition
pressure are decreasing while viscosity increases, the patterns.
flow patterns may change. Depending on the amount of
It was observed that the values of the number of
gas liquid ratio present in the well, there are more or less
Eotvos and the dimensionless number of viscosity have
bubbles while the fluid is raising to the surface, creating
both relatively high values. This indicates that the
in most of the cases slug flow.
velocity of the Taylor bubble is governed by inertia,
Hagendorn And Brown where C is 0,345. According to the work made with water
Table 2 shows the results obtained from the empirical and air by Viana, Pardo and Rotzer [ 1 0 ], for great
equations of H&B, their comparison with the values diameters the velocity of the Taylor bubble tends to be
obtained in the field and the error obtained in each well. governed by the inertia of the fluid.
It is observed that the values of the real pressure gradient The Figs. 3 and 4 show the behaviour of the correction
obtained are much greater than the obtained with the factor for the Hagendorn and Brown Correlation and
model, therefore the errors in the pressure gradients are Wallis and Aziz Correlation, respectively. Both curves
all positive, and vary between 18% to 162%. This have an approximate bell shape performance. The H&B
originates an increase in the size of the liquid column curve tends to be wider than the one defined with Wallis
where the errors vary between 60% to 555%. The and Aziz Correlation. This last one has an approximate
differences between the values obtained in field and the average correction factor of 0.95 with a standard
calculated theoretically with this method show deviation of 0.524; values overcoming the results using
considerable errors. The most likely explanation is that H&B, with a average correction factor of 2.8 and a
the characteristics and conditions of our experimental standard deviation of 1.4.
data may not be similar to the values H&B used for
An attempt to identify the relationship of the main
developing the empirical correlation.
parameters and the fluid gradient was performed, i.e.,
In an attempt to characterise the deviation of the several plots were made isolating some of the
empirical correlation of H&B, a special analysis was fundamental variables. The Figs. 5 and 6 show this
performed identifying a correction factor to match the relationship for oil API and flowrate.
values obtained in the field for each well. If this curve
The Fig. 5 shows the dependency between the pressure
were acute, then a correction factor for all wells could
gradient and oil API. Increasing the oil API, the viscosity
have been used. Fig. 3 shows the behaviour of this
of the fluid decreases as well as it does the liquid density.
correction factor. Although an average correction factor
Based on the figure, the liquid density seems to control
can be found around 3, the curve tends to be wide, as the
the phenomena as it increases the pressure gradient for
standard deviation is high. An approximate bell shape
heavier oil. Although this fact is observed in both models
behaviour is observed.
as well as in the experimental data, for H&B model the
Wallis and Aziz et al. Correlation relationship is steeper.
Table 3 shows the results obtained from using Wallis
The Fig. 6 shows the relationship between the pressure
and Aziz et al, their comparison with the values obtained
gradient and liquid flowrate. Increasing the liquid
in the field and the error obtained for each well. Although
flowrate increases the pressure gradient, as the liquid
the values of the pressure gradient calculated
tends to flow as fast as the gas. This yields to higher
theoretically are not exactly the same respect to the
liquid hold-up factors for higher flowrates. The friction
obtained in the field, the deviation for many of wells are
effects were extremely low in all the cases studied. It was
low. Two exceptions from the evaluated wells were
noted that an increase in the water cut causes an increase
found. In these cases the values of pressure gradient
in the pressure gradient, since the water is heavier than

6
the oil and the gas accumulation decreases as it flows NOMENCLATURE
faster due to the lower liquid viscosity. A area (ft2)
API gravity of the crude oil
CONCLUSIONS AND RECOMMENDATIONS Co isothermal factor of compressibility of
1. This work presents the first attempt to determine the the crude oil
length of the multiphase behaviour of the fluid C Wallis coefficient
column accumulated in intermittent gas lift wells. D diameter of the tubing (ft)
g gravity
2. The tendencies between theoretical and experimental Grad gradient of pressure (psi/ft)
data are similar, according to most of the important Gradrealassume real pressure gradient assumed (psi/ft)
parameters. Nevertheless, none of the methods H height of the liquid column in the
examined reproduces the experimental data experimental data (ft)
accurately. H1 depth of the lowest downhole measure
3. In most cases, the experimental data show heavier (ft)
pressure gradients than the equivalent theoretically H2 depth of the highest downhole measure
estimates. The number of sample wells is not large (ft)
enough to determine with high precision the source H&B Hagendorn and Brown
of these deviations. The difference may be due to N number of liquid column segments
some of the following reasons. Length liquid column height (ft)
NE dimensionless Number Eotvos
• The liquid fallback is neglected during the
NV dimensionless number of viscosity
accumulation stage inside the production tubing
P pressure (psi)
• The producing liquid and gas flowrate are assumed P(i,j) pressure average of each segment (psia)
steady P1’ Pwf , bottomhole pressure in the
experimental data (psi)
• Inaccuracy of the subsurface logging tool and
P2’ Pwh, surface Pressure in the
visual appreciation of the graphs
experimental data (psi)
• Uncertainty in the production data of the wells Qg gas flowrate (scfd)
tested. Q flowrate (stkbpd)
4. The best method to reproduce the pressure gradient T temperature (F)
with the model is using Wallis and Aziz Correlation. T1 average temperature of the segment (F)
The values obtained from this theoretical method are T2 average temperature of next segment(F)
similar to the tested wells. V∞T Taylor bubble velocity (ft/s)
V velocity (ft/s)
5. Investigations about the C coefficient of correlation ρ density (lbm/ft3)
in Wallis for crude oil between 15-25 API may
µ viscosity (cp)
improve the accuracy of the estimates. Research
σ surface tension (dyne/cm)
about mechanistic models and flow patterns
∆P pressure variation (psia)
transition would probably yield better determination
of liquid column. Subscripts
assume assumed
ACKNOWLEDGEMENTS g gas
i local index of segment
The authors wish to acknowledge the permissions
j index of iterations made by segment
granted by PDVSA-Intevep to publish this paper.
L liquid

7
m mixture 7. Gasbarri S., Gupta, A., Wiggins, M. Inflow
s slip Performance of Reservoirs Produced by Intermittent
seg segment Lift Methods. Paper 97-135 prepared for the 48th
sg superficial gas Annual Technical Meeting of the Petroleum Society
in Calgary, Alberta. June 8-11, 1997.
REFERENCES
8. Hernandez, A., Perez, C, Navarro, U, Lobo, W.
1. Hagedorn A.R., and Brown K.E.. Experimental study
Intermittent Gas Lift Optimization in Rosa Mediano
of pressure gradients occurring during continuous
Field. Paper SPE 53968 presented at the 1999 SPE
two-phase flow in small diameter vertical conduits,
LAPEC in Caracas, Venezuela, 21–23 April 1999
JPT, April, 1965, pg. 475-484.
9. Hernandez, A.; Garcia G.; Concho M.A. Navarro U.,
2. An, H.;Scott, S.L.; Langlinais J.P. Estimation of
Garcia R. Downhole Pressure and Temperature
Bottomhole Pressure in Pumping Oil Wells: Effect of
Survey Analysis for Wells on intermittent Gas Lift,
High-Viscosity Fluids and Casing head Pressure,
SPE 39853, prepared for the SPE International
SPE 63047, prepared for presentation at the 2000
Petroleum Conference and Exhibition held in
SPE Annual Technical Conference and Exhibition
Villahermosa, Mexico, 2-5 March 1998.
held in Dallas, Texas, 1-4 October 2000.
10. Viana, F.; Pardo, R.; Rotzer, I. Terminal Rise
3. Wallis, Graham B. One dimensional Two-Phase
Velocity of Various Bubble Sizes in a Tube and an
Flow. McGraw-Hill, Inc. 1969. Pags. 285-291.
Annular Space Through a Viscous Liquid. OAF-7.
4. Hasan, A. Rashid; Kabir, C. Shan; Rahman.. Prepared for the V Latin American and Caribbean
“Predicting Liquid Gradient in a Pumping Well Congress of Fluid Mechanics, May 14-17 2001.
Annulus, SPE 13638, prepared for the SPE Caracas, Venezuela.
Production Engineering, February 1988.
11. Moy, S. Diseño de Metodología para Estimar la
5. Aziz, K.;Govier, G. And Fogarasi, M. Pressure Drop longitud de la Columna Bifásica en pozos bajo el
in Wells Producing Oil and Gas, Journal of Método de Levantamiento Artificial por Gas
Canadian Petroleum, July-September 1972. Intermitente. Bachelor Thesis, Universidad Simón
Boíivar, Caracas 2001.
6. Hernandez, A., Gasbarri, S. Machado, M, Marcano,
L.; Manzanilla, Guevara. “Field-Scale Research On
Intermittent Gas Lift”. Paper SPE-52124 prepared
for the 1999 SPE Mid-Continent Operations
Symposium held in Oklahoma City, 28–31 March
1999.

8
NV m

≥250 10

250>NV>18 68(NV -0.35)

≤18 25

Table 1. Parameter m given the values of parameter Nv.

Fig. 1 Example of dynamic Pressure and Temperature

Wells Gradient Error Column Error (%)


(%)
Well 1 35 284
Well 2 18 555
Well 3 74 135
Well 4 21 467
Well 5 162 60
Well 6 30 340
Well 7 37 270
Well 8 54 186
Well 9 30 330
Well 10 48 210
Well 11 31 329
Well 12 83 120

Table 2. Results of the Hagendorn and Brown Model


Fig. 2. Diagram of the well analysis used in the
Wells Gradient Error Column Error programs.
(%) (%)
Well 1 12 -13 Frecuency vs. Correction Factor for Pressure
Well 2 27 -231 gradient in Hagendorn and Brown Model
Well 3 -127 56
Well 4 48 -93 0.3
Well 5 -389 80 0.25
Frecuency

Well 6 1 -3 0.2
Well 7 35 -52 0.15
Well 8 -33 23 0.1

Well 9 30 -41 0.05


0
Well 10 -76 44
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
Well 11 -17 11 Correction Factor
Well 12 -384 79
Fig. 3 Estimated correction factor for the pressure
Table 3. Results of the Wallis and Aziz Model
gradient (Hagendorn and Brown Model)

9
Frecuency vs. Correction Factor for the Pressure Ql (BPD) vs. DP/DZ Field
Gradient in Wallis and Aziz Model 160 Wallis&Aziz
140 H&B
0.80 120
0.70
100

Ql (BPD)
0.60
Frecuency

0.50 80
0.40
60
0.30
0.20 40
0.10 20
0.00
0.00 0.50 1.00 1.50 2.00 2.50 0
0 0.1 0.2 0.3 0.4 0.5
Correccion Factor
DP/DZ

Fig. 4 Estimated correction factor for the pressure


gradient (Wallis and Aziz Model) Fig. 6 Influence of Liquid Flowrate on Pressure Gradients

Fig. 5 Influence of API gravity on pressure gradient

10

You might also like