(1999-Tang - Schmidt) GLU-Trnsient Dyanamics PDF

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Transient Dynamic Characteristics of the

Gas-Lift Unloading Process


Yula Tang, SPE, Zelimir Schmidt,* SPE, R.N. Blais, SPE, and D.R. Doty, U. of Tulsa

Summary Finally, their model considers only a single tubing and casing size
This paper presents a new comprehensive transient model and restricted to a vertical wellbore. Subsequently, Hall and Decker3
dynamic simulator that describe the complicated characteristics of applied the model of Capucci and Serra2 to investigate gas-lift
the gas-lift unloading process. The model uses conservation equa- wells in the North Sea.
tions of mass, momentum and energy, for both the cocurrent and Avest and Oudeman4 studied gas-lift instability with a transient
countercurrent multiphase flow, along with the necessary closure unloading simulator in 1994. Unfortunately, they did not report
relationships. The effect of liquid flowing back into the formation any details about the theoretical aspects and solution procedures
is also considered. The transient temperature changes in the gas- used in the development of their model.
lift unloading process are studied and the transient tubing pressure Since 1994, the Tulsa University Artificial Lift Projects
and two-phase flow parameters are incorporated into the tempera- 共TUALP兲 has been studying the transient aspects of gas-lift
ture calculation. Two heat transfer mechanisms within the well- unloading.5 In this paper, a new comprehensive model and simu-
bore annulus are considered in the temperature model. lator is presented that includes the features mentioned in the ab-
An explicit numerical scheme is used for the mass and momen- stract.
tum partial differential equations. A double iteration procedure on
both the temperature and pressure numerically solves the three
conservation equations simultaneously. Description of the Gas-Lift Unloading Process
Several example calculations are given to illustrate the charac- Gas-lift unloading may be defined as a transient process in which
teristics of the unloading process and to underscore the unreliabil- pressurized gas is injected into the well 共usually into the annulus
ity of conventional design methods. Gas-lift instability is also ana- between the tubing and the casing兲 to lift both the wellbore fluid
lyzed with the dynamic simulator. In addition, field data from 25 and the formation fluids to the surface. When the gas-lift unload-
wells support the accuracy of the transient model. Finally, the ing process reaches either steady production or cyclic unstable
simulator can be used to perform gas-lift optimal design, stability production, then unloading ends and production begins. In other
analysis, and gas-lift operation troubleshooting. words, the consequence of gas-lift unloading has two possibilities:
stable production or unstable production.
A typical single-well system for continuous gas lift consists of
Introduction the tubing, casing, packer, gas-lift valves, a surface injection
Gas lift is one of the most widely used artificial lift methods. A choke, and the reservoir. If it is possible for the kill fluid to dam-
continuous injection of high-pressure gas from the surface into the age the formation, an operator may install a check valve 共standing
production string lightens the tubing liquid, enabling a higher pro- valve兲 at the bottom of the tubing to prevent the introduction of
duction rate. Normally, gas-lift designs assume steady flow and kill fluid back into the formation. In offshore gas-lift operations, a
incorporate much guesswork. Often, the empirical design proce- popular practice is to place a subsurface safety valve 共SSSV兲 into
dures used to space and design the unloading and operating gas- the gas-lift tubing string to protect surface equipment and also to
lift valves result in multipoint gas injection, unstable production, shut-in the well in case of an emergency. A typical gas-lift well
and low lift efficiency. One important cause for these shortcom- installation is illustrated in Fig. 1.
ings is the inadequacy of using steady-state equations to approxi- The unloading process can be divided into two phases. In the
mate an inherently transient unloading process. People have long first phase, only kill fluid is pushed into the tubing from the an-
realized that a better understanding of the transient unloading pro- nulus by the high-pressure gas injected through the gas-lift valves
cess can provide a solid basis for improved design, optimal injec- or the sliding sleeve placed above the packer. In the second phase,
tion gas utilization and for gas-lift troubleshooting. gas enters the tubing from the annulus via the gas-lift valves. If no
check valve is installed in the tubing, then a portion of the kill
Background fluid will be displaced into the formation. Under this condition,
Shinta1 performed the first work on transient unloading simulation countercurrent two-phase flow 共rising gas and falling liquid兲 may
in 1989. Unfortunately, his model is a pseudo-transient model exist in the second phase of the unloading process.
because it uses steady-state equations for each finite time interval. The intent of a traditional unloading design is for all gas-lift
The calculated values at one time step are used as the starting valves to remain open while they are covered by the liquid, and
parameters for the next time step. then close one by one from the top to the bottom as they are
Capucci and Serra2 developed a true transient unloading model uncovered. In fact, this is untrue in many cases, regardless of
in 1991. However, their model assumes a check valve at the bot- whether a graphical design or computer design is used. A design
tom of the tubing string thereby suppressing back flow. Therefore, procedure based on steady-state flow may lead to unsatisfactory
no countercurrent flow was considered in their model. In addition, performance characteristics like multipoint injection, unstable
they neglected mass exchange between the liquid and gas phases. production, or inefficient production.

*Deceased.
Copyright © 1999 Society of Petroleum Engineers
Mathematical Model for the Transient Pressure,
This paper (SPE 57659) was revised for publication from paper SPE 38814, presented at
Velocity, and Closure Relationships
the 1997 SPE Annual Technical Conference and Exhibition held in San Antonio, Texas, The transient characteristics of an unloading system are described
5–8 October. Original manuscript received for review 21 October 1997. Revised manu-
script received 7 April 1999. Manuscript peer approved 14 April 1999. using the conservation equations for mass, momentum, and en-

268 SPE Journal 4 共3兲, September 1999 1086-055X/99/4共3兲/268/11/$3.50⫹0.15


where q v l,i is in bbl/d.
The gas injection flow rate through the surface choke, q g,inj at
standard conditions, Mscf/d, can be calculated from the energy
balance equation for compressible flow 共the Thornhill-Craver
equation兲:

q g,ing⫽155.5C d A ch p inj 冉 共 2 兲共 322兲共 k/ 共 k⫺1 兲兲共 ⑀ 2/k ⫺ ⑀ 共 1⫹1/k 兲 兲


␥ gi 共 460⫹T inj兲 Z 冊 1/2
,
共6兲
where C d 共the discharge coefficient兲 is approximately equal to
0.82. For the above equation, if ⑀ ⬍ ⑀ c , then let ⑀ ⫽ ⑀ c , where

⑀ c⫽ 冉 冊
2
1⫹k
k/ 共 k⫺1 兲
. 共7兲

In this study, when a gas-lift valve is not covered by liquid,


Fig. 1–Typical single-well system for continuous gas lift. then q v g,i 共gas throughput at standard conditions via the ith gas-
lift valve兲 was calculated using TUALP’s latest gas-lift valve dy-
namic performance models for injection pressure operated 共IPO兲
ergy along with closure relationships. For simplicity, the energy or production pressure operated 共PPO兲 valves. For orifice valves,
equation describing the heat transfer is discussed in a subsequent the Thornhill-Craver equation was used because there is neither a
section. spring nor a nitrogen dome loading element. For a convergent/
divergent nozzle-venturi valve, a special equation is needed which
Tubing Mass Balance Equations. The mass balance equation for can be derived from compressible flow theory and corrected by
the liquid phase is experimental tests for a specific design geometry. The upstream
pressures of the gas-lift valves were calculated by incorporating
⳵ ⳵ both the gas column weight and kill fluid weight.
共 ␳ y 兲⫹ 共 ␳ v 兲 ⫽w v l , 共1兲
⳵t l l ⳵ L l Sl Finally, the volume change of kill liquid within the annulus is
where the mass balance equation for the gas phase is the solution to the following equation:
dV l

⳵t
共 ␳gy g兲⫹

共 ␳ v 兲 ⫽w v g .
⳵ L g Sg
共2兲 dt
⫽⫺ 兺q v l,i . 共8兲

If no gas-lift valve is included within the tubing segment, both w v l Liquid Holdup Equation. During the unloading process, two
and w v g are zero. The mass flux between the liquid phase and the kinds of flow may exist: cocurrent and countercurrent two-phase
gas phase may be neglected, or alternately, the mass exchange is flow. Countercurrent flow occurs when liquid falls back toward to
taken into account by calculating the solution gas in the oil using the formation but gas rises up the well.
a black oil model. The liquid density, ␳ l , and gas density, ␳ g , Cocurrent Two-Phase Flow. For cocurrent two-phase flow, ei-
will change both with time, t, and position, L. ther the empirical slug flow correlations of Schmidt6 or the
mechanistic model of Chokshi7 were used in this simulator. From
Tubing Mixture Momentum Equation. The mixture momentum the input data: d ti , v Sl , v Sg , using Schmidt’s correlation allows
equation can be expressed in the following form: the calculation of ␤ ⫽L S /L U 共the ratio of liquid slug length over

⳵t
共 ␳ l v Sl ⫹ ␳ g v Sg 兲 ⫹
⳵L yl 冉
⳵ ␳ l v Sl 兩 v Sl 兩 ␳ g v Sg 兩 v Sg 兩

yg 冊 the total slug unit length兲; y g , the average void fraction of the
slug unit; and y gLS , the void fraction within the liquid slug. Using
Chokshi’s model, the gas void fraction, y g , for bubbly flow and

⫹ 共 144g c 兲
⳵p
⳵L
⫹g ␳ m sin共 ␪ 兲 ⫹
⳵p
⳵L
冏 fric
⫽0. 共3兲
slug flow can be calculated, respectively 共concentration coeffi-
cient, C 0 , is 1.08兲.
Chokshi’s data show that the upper limit for this condition is
Annulus Mass Conservation Equation. The previous equations y g ⭐0.8, which is also consistent with Schmidt’s procedure. Com-
describe the unloading performance for the tubing. An additional parison of Schmidt’s correlation and Chokshi’s mechanistic
equation is needed to couple the tubing behavior to the casing model produced similar results.
behavior. This equation incorporates both the dynamic gas-lift Countercurrent Two-Phase Flow. Hasan et al.8 presented a
valve performance, q v g,i or q v l,i , and the gas injection through drift-flux method based on their experiments to calculate void
the surface choke, q g,inj . Capucci and Serra2 derived an equation fraction in countercurrent bubble-slug flow. For bubbly flow, they
for single tubing and casing strings without consideration of the used C 0 ⫽2. For slug flow, they recommended two kinds of ap-
gas column weight within the annulus: proaches: a cellular approach and an average rise velocity ap-
proach. Because we found that the latter produces a smooth tran-
dp
冏 共 ␳ gsc q g,inj⫺ ␳ gsc 兺 q v g,i ⫺ ␳ g 兺 q v l,i 兲 sition from bubbly flow to slug flow 共and is close to the results


⫽ . 共4兲
dt surface
d␳ g predicted by Chokshi’s mechanistic model兲, we used the average
Vg rise velocity approach in the dynamic model.
dp surface

During the unloading process, when a gas-lift valve is covered Density and Frictional Pressure Gradients. For cocurrent flow,
by liquid, the annulus liquid flow rate through the ith gas-lift the mixture density was calculated using the liquid slug length
valve, q v l,i , can be derived using an energy balance: ratio, ␤:

q v l,i ⫽8048d 2p,i C d 冋 共 p 3i ⫺p 4i 兲


␳L 册 0.5
, 共5兲
␳ m ⫽ ␤ 共 y lLS ␳ l ⫹y gLS ␳ g 兲 .
For countercurrent flow, we have
共9兲

Tang et al.: Characteristics of the Gas-Lift Process SPE Journal, Vol. 4, No. 3, September 1999 269
␳ m ⫽y l ␳ l ⫹y g ␳ g . 共10兲 ever, in this study we have proved that these assumptions are valid
only for a pure gas or a pure liquid. For a two-phase mixture
The friction pressure gradient is expressed as inside the tubing, it is not true.
⳵p
⳵L
冏 fric

f m␳ mv m兩 v m兩
2d ti
. 共11兲
Lower Wellbore Region–Modified Natural Convection
Model. When we consider the heat transfer within the wellbore
below some gas-lift valve, where the annulus fluid remains at rest,
This study uses a black oil model to calculate the physical the radial-direction equal-heat-flow-rate model can be used. This
properties of the fluids. Any changes for the liquids are related to region is referred to as the Lower Region.
the mass transfers between the two phases. We assume that the heat flux from the tubing into the cement is
equal to the heat flux from the cement into the formation. From
the energy equation for the tubing fluids, the outlet temperature
Mathematical Model of Heat Transmission for each small segment can be expressed with inlet temperatures
The model that follows for heat transmission was originally de- as follows.11
veloped and presented in full by the authors in Ref. 9. The tem-
perature model, which is only one component of the gas-lift dy- g sin ␪ F c g T sin ␪
T t f ,out⫽T ei,out⫺ ⫹ ⫹
namic simulator, is summarized here for the sake of completeness. g c JC p,m A A A

Formation Heat Transfer. The formation heat transfer can be


described using the diffusivity equation along with the appropriate

⫹e ⫺A⌬L T t f ,in⫺T ei,in⫹
g sin ␪ F c g T sin ␪
⫺ ⫺
g c JC p,m A A A
, 冊
boundary conditions. The solution is often expressed in terms of 共14兲
the dimensionless temperature, T D , where T D is defined as
where
2 ␲ k h,e
T D⫽ 共 T ⫺T ei 兲 . 共12兲
w 共 dQ/dL 兲 wb dp vm dv m
F c ⫽C J ⫺ , 共15a兲
Because T D is a function of time, it is necessary to explicitly dL g c JC p,m dL

冉 冊
express it as a time function, f (t). In this study, we will use a
modification of Remey’s original model developed by 2␲ r ti Uk h,e
A⫽ . 共15b兲
Hasan-Kabir,10 which considers the existence of a real wellbore C p,m W k h,e ⫹ f 共 t 兲 r ti U
size.
As the unloading time increases, the heat transfer rate, Note that the tubing pressure, p, and fluid velocity, v m , are func-
(dQ/dL) * w, changes with time, invalidating the constant bound- tions of both space and time, so the tubing fluid temperature is
ary condition described in the Hasan-Kabir model. The superpo- related to the tubing fluid pressure and velocity profiles during the
sition principle is able to account for this changing heat flux. unloading process. In this study, Fc will be strictly calculated by
Thus, the time function, f (t), for the variable heat flux can be combining conservation equations of mass and momentum as dis-
expressed as cussed in the previous paragraphs, instead of using empirical
correlations.11

再冋冉 冊 冉 冊 册 冎
n
dQ dQ In Eqs. 共15a兲 and 共15b兲, C p,m and C J are calculated by Alves’s

i⫽1
w
dL i
⫺ w
dL
T D 共 t D ⫺t D,i⫺1 兲 ‘‘black oil’’ model.14 The overall heat transfer coefficient between

冉 冊
i⫺1
f 共 t 兲⫽ . 共13兲 the fluid and the cement-earth interface can be expressed as
dQ

冉 冊 冉 冊 冉 冊
w
dL r to r co r wb
n r ti ln r ti ln r ti ln
1 1 r ti r ti r ci r co
Wellbore Heat Transfer. Heat transfer in the radial direction ⫽ ⫹ ⫹ ⫹ ⫹ ,
U h tf k h,t r ci h an k h,c k h,cem
from the well axis to the outer surface of the cement is referred to 共16兲
as the wellbore heat transfer. In this study, we use the general
outline of the Hasan-Kabir wellbore heat transfer models11,12 but where the heat transfer between the tubing fluids and tubing wall
with significant modifications. The modifications are based on the can be considered as forced convection. Brill and Beggs15 recom-
following considerations. mend the use of the Nusselt number, N nu , to calculate h tf for
First, Hasan-Kabir originally assumed that the heat transfer turbulent flow. The natural convection coefficient in the annulus,
mechanism within the annulus is either by natural convection h an , is approximated with the correlation for natural convection
共1991兲,11 or by forced convection 共1993兲.12 However, both heat between two vertical plates as used by Hasan and Kabir.11
transfer mechanisms exist within the annulus: forced convection Upper Wellbore Region—Forced Convection Model. For the
above the gas injection valve and natural convection below the annulus region above some gas-lift valve, both the injection gas
gas injection valve. Hasan-Kabir never combined the two mecha- and possibly the annulus liquid are in motion. The heat transmis-
nisms together to describe the real situation of heat transfer within sion is by forced convection instead of natural convection. We
a gas-lift well. Second, the original Hasan-Kabir model is valid call this wellbore region the Upper Region. In this region, even
only for steady-state production. It is not valid for the transient when the annulus liquid level remains constant for a relatively
process. For example, the Hasan-Kabir natural convection model long time interval, the gas within the annulus is still in motion.
uses the empirical correlation presented by Sagar et al.13 to ap- For simplicity, we assume an average sectional temperature for
proximate the contribution of the pressure and velocity into the the annulus fluids. The energy equation for the annulus fluids is
temperature calculation. For the unloading process, the pressure the same as the equation used by Hasan-Kabir,12 where the
and velocity change dramatically, so that the empirical correlation pressure-gradient term, the gravity-gradient term, and the kine-
derived from steady-state conditions is no longer appropriate. matic term approximately cancel out for a pure fluid. Thus,
Third, Hasan-Kabir’s treatment of the forced convection model
assumes that the pressure-gradient term balances the gravity term, dT an 共 T ei ⫺T an兲 共 T t f ⫺T an兲
⫽ ⫹ , 共17兲
and the kinematic energy change may also be neglected. How- dL A1 B1

270 Tang et al.: Characteristics of the Gas-Lift Process SPE Journal, Vol. 4, No. 3, September 1999
TABLE 1– INPUT DATA FOR UNLOADING ANALYSIS

D ti ⫽2.441 in. D ci ⫽7.5 in. Depth⫽8,000 ft API⫽35° g t ⫽7.625


D to ⫽2.875 in. D co ⫽8.5 in. ␥ g ⫽0.7 10⫺3 (°F/ft)
k h , e ⫽1 k h , cm ⫽0.6 C p , e ⫽0.17 ␤ kf ⫽4.7e ⫺4 (1/°F) R gl ⫽400
k h , t ⫽15 k h , l ⫽0.07 k h , g ⫽0.025 C p , l ⫽0.55 C p , g ⫽0.44 ␦ cem⫽1.5 in. ␳ e ⫽140 (lbm/ft3) f w ⫽0.5
p wh ⫽120 psi T sur⫽109°F PI ⫽5 b/d/psi q l , M ⫽3,600 b/d
p res⫽1,920 psi T bot⫽170°F p b ⫽1,950 psi p bh, M ⫽1,200 psi
Camco 1.0 in. BK valve d port⫽0.3125 in. T inj⫽100°F d choke⫽0.1875 in.
R ⫽0.2474 p inj⫽1,050 psi
Valve No. Depth (ft) P VC . (psig) Valve No. Depth (ft) P VC . (psig)

1 2,350 800 5 5,250 845


2 3,350 820 6 5,650 855
3 4,150 830 7 5,850 860
4 4,750 840 8 5,900 860

A 1⫽
2␲ 冉
C p,anw an k h,e ⫹r ci U an f 共 t 兲
r ci U ank h,e
, 冊 共18兲
annulus and tubing fluids. By using the heat boundary conditions
on the outer wall of the tubing, these two equations are coupled
together to produce a second-order ODE. Solving these equations

1

r ci ln 冉 冊
r co
r ci

r ci ln 冉 冊
r wb
r co
, 共19兲
yields an analytical expression for the tubing temperature and an-
nulus temperature as follows:
U an k h,c k h,cem T tf⫽ ␣ 1 e ␭ 1 L ⫹ ␣ 2 e ␭ 2 L ⫺g T L sin共 ␪ 兲 ⫹C 2 g T sin共 ␪ 兲 ⫺C 3 , 共24a兲
␭1L ␭2L
C p,anw an T an⫽C 4 ␣ 1 e ⫹C 5 ␣ 2 e ⫺g T L sin共 ␪ 兲 ⫹C 6 , 共24b兲
B 1⫽ , 共20兲
2 ␲ r ti U t where C 0 ,...,C 6 and ␭ 1 ,␭ 2 are given in Appendix A. Note that
r to the expressions for the coefficients are different from those used
r ti ln in the Hasan-Kabir model.
1 1 r ti
⫽ ⫹ . 共21兲 After T t f and T an have been calculated, it is possible to use this
Ut htf k h,t information to calculate the corresponding temperatures of the
The energy equation for the tubing fluids is other components within the system, especially the gas-lift valve
dome temperatures. In many field applications, the gas-lift valves
dT tf g sin共 ␪ 兲 共 T tf⫺T an兲
⫺F c ⫹ ⫽⫺ , 共22兲 are installed in a side pocket mandrel. Using the results from both
dL g c JC p,m B2 experimental and numerical simulation 共FEM兲 given by
C p,m w t Faustinelli,16 the dome temperature can be expressed approxi-
B 2⫽ . 共23兲 mately as
2 ␲ r ti U t
T d ⬵0.3T tf⫹0.7T an . 共25兲
Here, we can no longer prove that the assumption for the pure
annulus fluids is valid for a two-phase mixture of the tubing fluids. Initial Conditions
Thus, the term F c on the left-hand side of Eq. 共22兲 must be cal- The model assumes that the tubing and casing are initially filled
culated. Therefore, the tubing and the annulus temperatures are with kill fluid to a certain height. The bottomhole pressure caused
again related to the transient pressure and velocity distributions by the kill fluid equals the static reservoir pressure. No flow exists
along the length of the tubing as in the case of Lower Region. through the tubing, the annulus, the surface choke or the gas-lift
Eqs. 共17兲 and 共22兲 are the energy equations, respectively, for the valves. Liquid holdup equals one in the region below the static
liquid level within the tubing, while it is zero above the static
liquid level.

Fig. 2–Bottomhole pressure, p bh , wellhead casing pressure,


p cs , and q g ,inj vs. time. Fig. 3–Liquid level depths and liquid flow rate vs. time.

Tang et al.: Characteristics of the Gas-Lift Process SPE Journal, Vol. 4, No. 3, September 1999 271
Fig. 4–Gas injection via gas-lift valves 3, 4, and 5. Fig. 5–Gas injection via gas-lift valves 5, 6, 7, and 8.

The initial wellhead casing pressure behind the surface choke is ing liquid level has reached the wellhead. When the temperature
assumed to be zero. The upstream pressure (p 3 ) on a gas-lift calculation is initiated, a numerical algorithm initiates a double
valve equals the gas column weight plus the static liquid pressure. iteration process to solve for the transient pressure and tempera-
The downstream pressure (p 4 ) on a gas-lift valve equals the well- ture distributions, which are interrelated. The full iteration solu-
head tubing pressure plus static liquid pressure above the valve. tion procedure calculates the initial temperature distribution using
The initial wellbore temperature is assumed to be the steady geo- the preliminarily fluid pressure and velocity fields; calculates the
thermal gradient distribution. gas-lift valve dome temperature and pressure for each valve using
the new tubing and annulus temperatures; and finally recalculates
Boundary Conditions the pressure and velocity fields again with the new temperature
This study assumes that the wellhead tubing pressure and the up- distribution. The double iteration process is continued until con-
stream surface gas injection choke pressure remain constant vergence is achieved.
throughout the unloading process.
The bottom hole is considered as a mass source 共fluids out兲 or
sink 共fluids in兲. When the bottomhole pressure is less than the Development of the Dynamic Simulator
reservoir pressure, then the fluids flow out of the formation. A The following practical aspects have been taken into account:
simplified IPR method is to use a straight line IPR 共if p bh⬎p b 兲 or 1. inclined or vertical wellbore;
Vogel’s IPR 共if p bh⬍p b 兲. Thus, the superficial liquid velocity can 2. both tapered tubing and casing;
be calculated and the superficial gas velocity can be calculated 3. tubing run to any depth within the casing;
using R gl , R s , and B g . 4. different injection gas and formation gas gravities;
For greater accuracy, the transient IPR should be taken into 5. tubing production and annulus injection, or tubing injection
account. For single-phase constant-production liquid flow, an ex- and annulus production;
pression relating pressure and flow rate can easily be found. For 6. any type of gas-lift valve 共IPO, PPO, orifice, nozzle-
variable flow, the superposition principle has to be used. How- venturi兲;
ever, for most gas-lift wells, the reservoir is under solution gas 7. one or more valves at a single location;
drive. Unfortunately, no simple expression exists for transient 8. with or without a subsurface safety valve 共SSSV兲;
two-phase flow within the reservoir. Observing that the transient 9. with or without a check valve; and
wellbore response is much faster than the transient reservoir re- 10. different kill fluid and formation fluid gravities.
sponse, then using a steady-state IPR should be reasonable. There-
fore, the Vogel IPR will be used to predict the two-phase reservoir
response.
When the bottomhole pressure is larger than the reservoir pres-
sure, liquid backflow into the reservoir occurs if there is no check
valve at the bottom of the tubing. However, if there is a check
valve, no backflow occurs. Again, lacking a simple relationship
that describes this behavior, then the formation in-take capacity is
assumed to be half of the productivity.

Numerical Solution Procedure


The fluid pressure and velocity distributions are calculated from
the mass and momentum conservation equations using a numeri-
cal method to solve the complicated system of equations. The
numerical scheme is presented in Appendix B.
The geothermal temperature gradient is used as the initial tem-
perature distribution within the wellbore. Calculation of the heat
transfer will be initiated only after: 共1兲 the first gas-lift valve is
uncovered, 共2兲 the formation is producing fluids, 共3兲 and the tub- Fig. 6–Tubing fluid temperature distribution.

272 Tang et al.: Characteristics of the Gas-Lift Process SPE Journal, Vol. 4, No. 3, September 1999
Fig. 7–Stabilized radial temperature distribution. Fig. 8–Stabilized temperature distribution along the radial di-
rection using the natural convection model.

Analysis of the Model’s Predictions valves 5 through 8 remain open resulting in multipoint gas injec-
Transient Unloading Process and Its Characteristics. A well tion even after the unloading process has been completed.
designed using a steady-state conventional design scheme is in-
vestigated with the dynamic simulator. The input data are listed in Transient Temperature Characteristics. The following obser-
Table 1 while Figs. 2 and 3 show the bottomhole pressure, p bh , vation was first presented in Ref. 9, but is presented here for the
wellhead casing pressure, p cs , tubing liquid level depth 共from the sake of completeness. Please refer to that paper for the complete
surface兲, D L,t , liquid level depth in the annulus, D L,an , liquid analysis and summary.
production rate, q l , and gas injection flow rate through the sur- Fig. 6 displays the tubing fluid temperature profile versus time
face choke, q g,inj , as functions of time. and depth for the same input data given in Table 1. The initial
During unloading Phase 1, when the injection gas has not en- temperature distribution is assumed to be the geothermal gradient
tered the tubing, the wellhead casing pressure increases rapidly 共straight line兲. The transient heat transfer calculation starts at t
from an initial value of zero because no gas is exiting the annulus ⫽3.85 hours. Fig. 6 shows the rapid rise of the tubing temperature
共Fig. 2兲. Correspondingly, the gas injection through the surface from t⫽3.85 to 4.45 hours. With increasing time, the temperature
choke decreases with time because the pressure differential de- variation eventually slows and finally stabilizes. Note that the
creases 共i.e., the downstream pressure, p cs , increases, while injec- temperature profile of the tubing fluids eventually becomes a con-
tion pressure holds constant兲. The high-pressure gas pushes the vex curve instead of a straight line. Fig. 7 illustrates the radial
annulus liquid into the tubing 关i.e., the liquid level in the annulus temperature distribution around the wellbore after the heat transfer
drops 共D L,an increases兲 and liquid level in the tubing rises 共D L,t stabilizes. For the Lower Region, the temperature on the casing
decreases兲 during the early part of Phase 1兴. However, an increas- wall is several degrees below the tubing temperature. This is
ing liquid column within the tubing causes the bottomhole pres- caused by the higher heat resistance of the annulus liquid, which
sure to exceed the reservoir pressure (p res⫽1920 psig). For this is at rest. For the Upper Region, the casing temperature is close to
example, there is no check valve at the bottom, therefore backflow the tubing temperature because the downward flowing annulus
into the reservoir occurs. After about an hour, backflow into the fluids influence the radial heat transfer and thereby create a more
reservoir outstrips the liquid flow supplied from the annulus. From homogeneous temperature profile from the outer tubing wall to
this point, the liquid level in the tubing descends 共Fig. 3兲 and the the inner casing wall. Note that the input injection gas temperature
bottomhole pressure reduces with time 共Fig. 2兲. No liquid is pro- at the wellhead is 100°F, which is lower than the undisturbed
duced during Phase 1 because the liquid level cannot reach the earth temperature at the surface 共109°F兲. Thus, the injection tem-
surface. perature hardly affects the wellbore temperature profiles except at
Phase 2 unloading starts when a gas-lift valve is uncovered and the inlet point. This is because the injection gas mass flow rate is
is also open. From Fig. 2, note that the bottomhole pressure drops much less than the mass flow rate of the tubing fluids. In addition,
because the injection gas lightens the tubing liquid. Figs. 4 and 5 the temperature at the outer interface of the cement is much lower
show the gas throughput via each valve. We note that valves 1 and than the tubing fluid temperature at the center of the wellbore
2 remain closed all of the time, contrary to the conventional de- because of the very high heat resistance of the cement material.
sign. Valves 3 and 4 open and close one after another. However, For comparison purposes, Fig. 8 presents the corresponding

TABLE 2– INPUT DATA FOR OPERATION INSTABILITY ANALYSIS

D ti ⫽1.995 in. D ci ⫽6.969 in. Depth⫽8,000 ft API⫽40°


D to ⫽2.375 in. D co ⫽7.625 in. p res⫽2,900 psig ␥ G ⫽0.65
R go ⫽1,000 ␥ kf ⫽0.826 p wh ⫽100 psi T sur⫽150°F
f w ⫽0.95 ␥ g ,inj⫽0.65 d choke⫽0.1 in. T bot⫽210°F
q l , M ⫽2,100 b/d PI ⫽10 b/d/psi T inj⫽100°F Vertical well
p bh, M ⫽2,600 psig p b ⫽2,550 psig p inj⫽550 psi Single valve
Camco 1.0 in. d port⫽0.25 in. Dep,val⫽2,250 (ft) P vct⫽300 psi
BK valve (IPO) R ⫽0.1583

Tang et al.: Characteristics of the Gas-Lift Process SPE Journal, Vol. 4, No. 3, September 1999 273
Fig. 9–Unstable production with serious fluctuations of p bh and
q g „q g ,avgⴝ90 Mscf/d, Case 1…. Fig. 11–Stable production with nozzle-venturi valve, q g
ⴝ90 Mscf/d „Case 3….

results calculated with a traditional temperature model that con-


siders only natural convection within the annulus. The natural
the tubing 共measured from the bottom兲 fluctuates between 7,500
convection model also unrealistically assumes that the flowing gas
and 8,000 ft 共which corresponds to the wellhead兲. Consequently
within the annulus is static with a large heat resistance. This is the
the liquid production, q l , behaves more like an intermittent gas-
reason that the traditional natural convection model tends to pre-
lift well instead of a continuous gas-lift well. The average liquid
dict unrealistic temperature distributions. By comparing Fig. 8
production rate is about 500 bbl/d and the average gas consump-
with Fig. 7, we find that the traditional natural convection model
tion is about 90 Mscf/d.
overpredicts the temperature of the tubing fluids by as much as
One method to stabilize this well is to increase the injection gas
10°F, which would produce significant errors in predicting the full
flow rate. In Case 2, the injection pressure is increased from 550
unloading process. Fig. 8 also indicates that the traditional natural
to 850 psig, and the surface choke size is enlarged from 0.1 to
convection model overpredicts the temperature differential be-
0.1875 in. In this way stable production is obtained. Under these
tween the tubing fluids and the inner-casing wall by as much as
conditions, the liquid production is improved from 500 to 2,900
15°F. The transient heat transfer also affects the accuracy of both
bbl/d, but the cost of obtaining the stable condition is the use of
gas-lift optimization and gas allocation. The simulator run without
five times as much gas 共450 Mscf/d兲.
using the transient temperature model predicts the following re-
Next, find a better way to reach stable flow and high production
sults for an optimal design: p inj⫽950 psi, d choke⫽0.15 in.,
without an increase in the gas consumption. The instability is
p v c (No. 8)⫽845 psi, and the gas injection requirement is 153
actually a form of casing heading. Under the condition of unstable
Mscf/d. When the simulator was run using the transient tempera-
production, the throughput of the gas-lift valve fluctuates with
ture model to predict an optimal design, quite different results
variations in the tubing pressure 共the valve operates in the sub-
were produced: p inj⫽950 psi, d choke⫽0.1875 in., p v c (No. 6)
critical flow regime兲. However, if the critical flow regime can be
⫽865 psi, p v c (No. 7)⫽870 psi, p v c (No. 8)⫽860 psi, and the
reached, the gas injection through the valve will not be affected by
gas injection requirement is only 44 Mscf/d.
the fluctuations of the tubing pressure. Fortunately, a nozzle-
Instability Analysis. Unstable production is a common occur- venturi valve17 is a good candidate for stabilizing this well be-
rence in the field. In this example, the ability of the dynamic cause it is designed to operate easily and efficiently in the critical
simulator to perform instability analysis is displayed. Here, three flow regime, often with a critical pressure ratio as high as 0.9.
cases are discussed. The primary input data for Cases 1, 2, and 3 Therefore, in Case 3, a nozzle-venturi valve 共NOVA兲 is used
are given in Table 2. with a throat diameter of 0.08 in., and an injection pressure of 650
The results from the dynamic simulator are shown in Figs. 9 psig. The results from the simulator are presented in Figs. 11 and
and 10 for Case 1. The production is unstable with all flow pa- 12. The figures show that the production becomes stable; the gas
rameters fluctuating seriously. For instance, the liquid height in

Fig. 10–Unstable production with serious fluctuations of H t Fig. 12–Stable production with nozzle-venturi valve, q l
and q l „q L avgⴝ500 bbl/d, Case 1…. ⴝ2,900 bbl/d „Case 3….

274 Tang et al.: Characteristics of the Gas-Lift Process SPE Journal, Vol. 4, No. 3, September 1999
TABLE 3– COMPARISON FOR THREE CASES ON GAS-LIFT STABILITY

Pl p t at Result Result
d port (stb/ p inj p vct d ch ql qg valve p bh of of
Case (in.) psi/d) (psia) (psia) (in.) (bbl/d) (Mscf/d) (psia) (psia) TUALP ASHEIM

1 0.25 10 550 300 0.1 360 90 490 2,850 U U


2 0.25 10 850 300 0.2 2,900 460 476 2,620 S U
3 0.08 10 650 Vent 0.1 2,850 90 502 2,640 S S

consumption is only 90 Mscf/d—the same as in Case 1; and the Fig. 13 illustrates a typical case for a single well, which shows
liquid production is as high as 2,900 bbl/d—the same as in Case 2. how the predicted behavior 共dotted line兲 follows the trend of the
Thus, more efficient production is obtained using this type of real behavior 共continuous line兲. Observe that both the cycle time
valve. Finally, a comparison of the transient simulator and and pressure fluctuations are reasonably predicted by the dynamic
Asheim’s18 analytical gas-lift stability criterion was made. The model. The overall results from running the dynamic simulator for
predicted results for Cases 1, 2, and 3 are summarized in Table 3, the 25 unstable wells are as follows. For the wells where the
where ‘‘U’’ refers to unstable and ‘‘S’’ refers to stable. For Case gas-lift valve operates in the throttling critical regime, the simu-
1 and Case 3, both this simulator and Asheim’s criterion reached lator correctly predicted unstable behavior in 85% of the cases.
the same conclusion. Nevertheless, the conclusions from this For the wells where the gas-lift valve operates in the throttling
simulator and Asheim’s criterion for Case 2 are different. Gener- subcritical regime, the simulator correctly predicted unstable be-
ally speaking, the result from the simulator is expected to be more havior in 80% of the cases. Overall, 84% of the wells are correctly
reliable because it includes all of the mechanisms involved in the predicted to be unstable. By comparison, Alhanati’s stability
process of instability that cannot be included in Asheim’s crite- model generated errors as high as 88% in predicting the stability
rion. This observation is further supported by the following field for these 25 wells.19 In addition, the average percentage error in
data. predicting the bottomhole pressure is 17%, and the average per-
centage error in predicting the liquid flow rate is 5.1%. In general,
Model Validation with Field Data the simulator performed quite well.
Field data for 25 wells from the Tia Juana field, Maracaibo basin,
Venezuela collected by Tinoco19 were used to validate the simu- Conclusions
lator. All of the gas-lift wells were operating under unstable con- A comprehensive transient mechanistic model for gas-lift unload-
ditions. Most of the wells have less than 50% water, and a static ing has been developed which can be used to perform gas-lift
pressure between 800 and 1,500 psig. 43% of the wells produce optimal design, instability analysis and operation troubleshooting.
more than 300 bbl/d, 33% produce between 100 and 300 bbl/d. The dynamic model has the following advantages over previous
The API of the oil is between 20 and 25. Well depths are between models: 共1兲 the transient heat transfer is properly incorporated into
5,000 and 7,000 ft for most of the wells. Most have a 7 in. casing the model during the unloading process; 共2兲 the new model real-
diameter and 50% of the wells have a 3 21 in. tubing diameter. 85% istically incorporates all components of a gas-lift system, includ-
of the wells have 1 21 in. valves installed, 72% have a port size of ing liquid backflow into the formation and countercurrent two-
16/64 in., and 52% have an injection rate between 300 and 600 phase flow; 共3兲 the dynamic gas-lift valve performance is properly
Mscf/d. incorporated into the new model for all types of valves; 共4兲 the
Due to the unstable behavior of the wells under study, the new model is designed to be both versatile and accurate in prac-
pressure and temperature sensors were set at the perforation depth tical field situations.
to avoid operational problems. The instruments accurately re- It was found that a conventional gas-lift unloading design
corded downhole pressures and temperatures every 30 seconds for based on steady-state flow cannot guarantee the desired results.
a period of 2 hours, with an accuracy of ⫾3 psi and ⫾0.5°C, Some incorrect concepts of the steady-state model were corrected
respectively. As a result, several heading cycles were recorded for in this study.
each well. The new transient gas-lift unloading model can also be used to
study gas-lift instability. The new model was shown to be vastly
superior to existing stability models, which do not properly con-
sider all the components of a gas-lift system.
The transient temperature distribution may significantly impact
the unloading process, which can not be properly predicted by
using a traditional steady-state temperature model. The new tran-
sient temperature model corrects this situation.
Measured field data validate both the accuracy and the utility of
the new gas-lift unloading model.

Nomenclature
A ⫽ cross-sectional area, ft2
A ch ⫽ cross-sectional area of surface choke, ft2
Bg ⫽ volume factor of formation gas, ft3/scf
CJ ⫽ the Joule-Thompson coefficient, °F/psf
Cp ⫽ heat capacity at constant pressure, Btu/lbm °F
Cd ⫽ discharge coefficient
Fig. 13–Comparison of modeling results with field data. D L,an ⫽ liquid level depth in the annulus, ft

Tang et al.: Characteristics of the Gas-Lift Process SPE Journal, Vol. 4, No. 3, September 1999 275
D L,t ⫽ liquid level depth in the tubing, ft ⑀c ⫽
critical pressure ratio
d p,i ⫽ port diameter of ith valve, in. ␥ gi ⫽
specific gravity of injection gas
d ti ⫽ tubing inner diameter, ft ␭ L ,␭ g ⫽
no-slip hold-up and no-slip gas void fraction
f ⫽ friction factor ␪ ⫽
inclination angle of tubing segment from horizontal
f (t) ⫽ the time function of formation heat transfer (T D ) direction
g ⫽ gravity acceleration, ft/sec2 ␳ ⫽ density, lbm/ft3
gc ⫽ conversion factor, 32.2 poundal/bf ␳ n ⫽ no-slip density of the mixture, lbm/ft3
gT ⫽ geothermal gradient, °F/ft
h an ⫽ natural convection coefficient for annulus fluid, Subscripts
Btu/共°F ft2 hr兲
h tf ⫽ forced convection coefficient for tubing fluids, an ⫽ annulus
Btu/共°F ft2 hr兲 c ⫽ casing
H ct ⫽ liquid level height in the casing, ft cem ⫽ cement
Ht ⫽ liquid level height in the tubing, ft ch ⫽ choke
j ⫽ the sequence number of a space node e ⫽ earth
J ⫽ heat/energy conversion factor, 778 共ft.lbf兲/Btu g ⫽ gas
k ⫽ specific heat ratio of injection gas f ⫽ formation
kh ⫽ heat conductivity, Btu/共°F ft hr兲 i ⫽ inner
L ⫽ length position 共upward as positive兲, ft inj ⫽ injection
n ⫽ the sequence number of a time node kf ⫽ kill fluid
p ⫽ pressure, psig l ⫽ liquid
p cs ⫽ wellhead casing pressure, psig LS ⫽ liquid slug
p 3i ⫽ upstream pressure of ith valve, psig m ⫽ two-phase mixture
p 4i ⫽ downstream pressure of ith valve, psig M ⫽ measured values
p vc ⫽ valve closing pressure at 60°F, psig o ⫽ outer
p bh ⫽ bottomhole pressure, psig sc ⫽ standard condition
Q ⫽ heat transferred from the wellbore to the formation t ⫽ tubing
per unit mass of wellbore fluid, Btu/lbm tf ⫽ tubing fluid
Qf ⫽ heat per unit mass of annulus fluid transferred from wb ⫽ wellbore 共sandface兲
annulus to the formation, Btu/lbm
q g,inj ⫽ gas injection flow rate, scf/sec Acknowledgments
ql ⫽ liquid production rate, bbl/d The support provided by the Tulsa University Artificial Lift
Q ta ⫽ heat per unit mass of annulus fluids transferred from Projects member companies is gratefully appreciated. We also
tubing to annulus, Btu/lbm thank Maria Tinoco for supplying the test well data.
q v g,i ⫽ gas throughput via ith G-L valve, scf/sec
q v l,i ⫽ liquid flow rate via ith G-L valve ft3/sec
r ⫽ radius ft References
1. Shinta, A.A.: ‘‘Time Modeling of Unloading Process for Continuous
R gI ⫽ gas liquid ratio, scf/bbl
Gas-Lift Wells,’’ MS thesis, Colorado School of Mines 共1989兲.
Rs ⫽ solution gas oil ratio, scf/stbo 2. Capucci, E.C. and Serra, K.V.: ‘‘Transient Aspects of Unloading Oil
t ⫽ time, seconds Wells Through Gas-Lift Valves,’’ paper SPE 22791 presented at the
tD ⫽ dimensionless time 1991 SPE 66th Annual Technical Conference and Exhibition, Dallas,
T ⫽ temperature, °F 6–9 October.
T ei ⫽ undisturbed formation temperature, °F 3. Hall, J.W. and Decker, K.L. ‘‘Gas-Lift Unloading and Operating
TD ⫽ dimensionless temperature Simulation as Applied to Mandrel Spacing and Valve Design,’’ paper
T ei,bh ⫽ undisturbed earth temperature at bottomhole, °F SPE 29450 presented at the 1995 Production Operations Symposium,
U ⫽ the overall heat transfer coefficient Btu/共°F ft2 hr兲 Oklahoma City, 2–4 April.
4. Avest, D. and Oudeman, P.: ‘‘A Dynamic Simulator to Analyze and
v ⫽ flow velocity, ft/sec
Remedy Gas Lift Problems,’’ paper SPE 30639 presented at the 1995
Vg ⫽ annulus volume occupied by gas column, ft3
Annual Technical Conference and Exhibition, Dallas, 22–25 October.
Vl ⫽ annulus volume occupied by liquid, ft3 5. Tang, Y.L., Schmidt, Z., and Blais, R.N.: ‘‘Transient Dynamic Char-
vS ⫽ superficial velocity, ft/sec acteristics of Gas-Lift Unloading Process,’’ paper SPE 38814 pre-
w ⫽ mass flow rate, lbm/hr sented at the 1997 SPE Annual Technical Conference and Exhibition,
wv ⫽ mass flow rate per unit volume through valve, San Antonio, 5–8 October.
lbm/ft3/sec 6. Schmidt, Z.: ‘‘Experimental Study of Two-Phase-Flow Slug Flow in
x ⫽ quality of gas in the mixture of tubing a Pipeline-Riser Pipe System,’’ PhD dissertation, University of Tulsa,
yl ⫽ liquid holdup Oklahoma 共1977兲.
yg ⫽ gas void fraction 7. Chokshi, R.N.: ‘‘Experimental Study and the Development of a
Mechanistic Model for Two-Phase Flow Through Vertical Tubing,’’
Z ⫽ compressibility factor of gas
paper SPE 35676 presented at the 1996 SPE Western Regional Meet-
ing, Anchorage, Alaska, 22–24 May.
Greek 8. Hasan, A.R., Kabir, C.S., and Srinivasan, S.: ‘‘Countercurrent Bubble
and Slug Flows in a Vertical System,’’ Chem. Eng. Sci. 共1994兲 49,
␣ ⫽ thermal diffusivity, ft2/hr
No. 16, 2567.
␤ ⫽ ratio of liquid slug length (L S ) over the total slug 9. Tang, Y.L., Blais, R.N., Doty, D.R., and Schmidt, Z.: ‘‘A Study on
unit length (L U ) Transient Temperature Characteristics of Gas-Lift Unloading Pro-
⌬t ⫽ time step, seconds cess,’’ paper for the Petroleum Production Technology Symposium at
⌬L ⫽ length step, ft 1999 ASME Annual Energy Sources Technology Conference and Ex-
⑀ ⫽ ratio of downstream pressure to upstream pressure hibition 共ETCE兲, Houston, 1–3 February.

276 Tang et al.: Characteristics of the Gas-Lift Process SPE Journal, Vol. 4, No. 3, September 1999
10. Hasan, A.R. and Kabir, C.S.: ‘‘Heat Transfer During Two-Phase ⳵ V V nj ⫺V nj⫺1
Flow in Wellbores: Part I—Formation Temperature,’’ paper SPE ⫽ . 共B-1兲
22866 presented at the 1991 SPE 66th Annual Technical Conference ⳵L ⌬L
and Exhibition, Dallas, 6–9 October.
The time derivative is made discrete with centered space and
11. Hasan, A.R. and Kabir, C.S.: ‘‘Aspects of Wellbore Heat Transfer
During Two-Phase Flow,’’ SPEPF 共August 1994兲 211.
backward time differencing 共CSBT兲:
12. Hasan, A.R. and Kabir, C.S.: ‘‘Predicting Fluid Temperature Profiles
⳵ V V nj ⫹V nj⫺1 ⫺V n⫺1
j ⫺V n⫺1
j⫺1
in Gas-Lift Wells,’’ paper SPE 26098 presented at the 1993 SPE ⫽ . 共B-2兲
Western Regional Meeting, Anchorage, Alaska, 26–28 May. ⳵t 2⌬t
13. Sagar, R.K., Doty, D.R., and Schmidt, Z., ‘‘Predicting Temperature
Profiles in a Flowing Well,’’ SPEPE 共November 1991兲 441. Thus, the mass conservation equation for the liquid phase is ex-
14. Alves, I.N., Alhanati, F.J.S., and Shoham, O.: ‘‘A Unified Model for pressed as
Predicting Flowing Temperature Distribution in Wellbores and Pipe-
lines,’’ SPEPE 共November 1992兲 363. 共 ␳ l y l 兲 nj ⫹ 共 ␳ l y l 兲 nj⫺1 ⫺ 共 ␳ l y l 兲 n⫺1
j ⫺ 共 ␳ l y l 兲 n⫺1
j⫺1
15. Brill, J.P. and Beggs, H.D.: Two-Phase Flow in Pipes, The University 2⌬t
of Tulsa, Tulsa 共1988兲 1–33.
16. Faustinelli, J.G.: ‘‘Temperature and Flow Performance Modeling of 共 ␳ l v Sl 兲 nj ⫺ 共 ␳ l v Sl 兲 nj⫺1 共 ␳ l 兲 nj q v l, j
Gas-Lift Valves,’’ MS thesis, The University of Tulsa 共1997兲. ⫹ ⫽ . 共B-3兲
⌬L A⌬L
17. Tokar, T., Schmidt, Z., and Tuckness, C., ‘‘Stabilize Gas-Lift Injec-
tion Rates,’’ JPT 共January 1997兲 57. The mass conservation equation for the gas phase is expressed as
18. Asheim, H.: ‘‘Criteria for Gas-Lift Stability,’’ JPT 共November 1988兲
1452; Trans., AIME, 285. 共 ␳ g y g 兲 nj ⫹ 共 ␳ g y g 兲 nj⫺1 ⫺ 共 ␳ g y g 兲 n⫺1
j ⫺ 共 ␳ g y g 兲 n⫺1
j⫺1
19. Tinoco, M.: ‘‘Validate Gas-Lift Instability with Field data,’’ MS The- 2⌬t
sis, The University of Tulsa 共1998兲.
20. Nickens, H.V.: ‘‘A Dynamic Computer Model of a Kicking Well’’ 共 ␳ g v Sg 兲 nj ⫺ 共 ␳ g v Sg 兲 nj⫺1 ␳ g,sc q v g, j
SPEPE 共June 1987兲 159; Trans., AIME, 283. ⫹ ⫽ , 共B-4兲
⌬L A⌬L

where q v g j is in scf/sec.
Appendix A–Coefficients in Eqs. „24a… and „24b… for
the Upper Region Heat Transfer Difference Schemes for Momentum Conservation. The same

冉 冊
space and time difference schemes are used to make the momen-
g sin共 ␪ 兲 tum equation discrete. Therefore, the first two parts of the momen-
C 0 ⫽B 2 ⫺F c , 共A-1兲
g c JC p,m tum equation are

C 1 ⫽A 1 B 2 , 共A-2兲 ⳵
Part 1⫽ 共 ␳ v ⫹ ␳ g v Sg 兲
C 2 ⫽A 1 ⫹B 1 ⫹ 共 A 1 /B 1 兲 B 2 , 共A-3兲 ⳵ t l Sl

C 3 ⫽C 0 ⫹ 共 A 1 /B 1 兲 C 0 ⫺T ei,bh , 共A-4兲 1
⫽ 关共 ␳ l v Sl ⫹ ␳ g v Sg 兲 nj
2⌬t
␭ 1 ⫽ 共 ⫺C 2 ⫹ 冑C 22 ⫺4C 1 兲 / 共 2C 1 兲 , 共A-5兲
⫹ 共 ␳ l v Sl ⫹ ␳ g v Sg 兲 nj⫺1 ⫺ 共 ␳ l v Sl ⫹ ␳ g v Sg 兲 n⫺1
j
␭ 2 ⫽ 共 ⫺C 2 ⫺ 冑 C 22 ⫺4C 1 兲 / 共 2C 1 兲 , 共A-6兲
⫺ 共 ␳ l v Sl ⫹ ␳ g v Sg 兲 n⫺1
j⫺1 兴 , 共B-5兲
C 4 ⫽1⫹␭ 1 B 2 , 共A-7兲
C 5 ⫽1⫹␭ 2 B 2 , 共A-8兲 Part 2⫽
⳵L 冋
⳵ ␳ l v Sl 兩 v Sl 兩 ␳ g v Sg 兩 v Sg 兩
yl

yg 册
冋冉 冊
C 6 ⫽g T 共 C 2 ⫺B 2 兲 sin共 ␪ 兲 ⫺C 3 ⫹C 0 . 共A-9兲
1 ␳ l v Sl 兩 v Sl 兩 ␳ g v Sg 兩 v Sg 兩 n
⫽ ⫹
Appendix B–Numerical Scheme ⌬L yl yg j
This work adopts with slight modification the numerical scheme
developed by Nickens20 for a dynamic gas kick simulator. The
conservation equations for mass and momentum are approximated
⫺ 冉 ␳ l v Sl 兩 v Sl 兩 ␳ g v Sg 兩 v Sg 兩
yl

yg 冊 册 n

j⫺1
. 共B-6兲

by explicit finite difference methods subject to the usual stability However, the gravity term and friction term are expressed as
restriction on the ratio of the maximum time step to the length
step. g c sin共 ␪ 兲
During unloading Phase 1, the liquid in both the tubing and Part 3⫽ 关共 ␳ l y l ⫹ ␳ g y g 兲 兩 nj ⫹ 共 ␳ l y l ⫹ ␳ g y g 兲 兩 nj⫺1 兴 , 共B-7兲
2
casing will have a uniform density, and therefore numerical itera-

再冉 冏 冊 冉 冏 冊 冎
tion is only required to solve the mass balance equation for the n n
1 dp dp
annulus. However, for unloading Phase 2, gas-liquid two-phase Part 4⫽ ⫹ . 共B-8兲
2 dL fric j
dL fric j⫺1
flow occurs within the tubing. Therefore, the pressure changes for
both time and space and a numerical finite difference scheme is The pressure gradient is expressed as a backward difference:
required to approximate the system of conservation equations.
Generate a space-time numerical mesh in the usual way so that dp p nj ⫺p nj⫺1
the solution of the explicit finite difference scheme can be calcu- ⫽ . 共B-9兲
dL ⌬L
lated at the nth time and jth node. The bottomhole corresponds to
j⫽0 and the initial time corresponds to n⫽0. Thus, the pressure at the nth time and jth node is

Difference Schemes for Mass Conservation. The space deriva- ⌬L 共 Part 1⫹Part 2⫹Part 3⫹Part 4兲
p nj ⫽p nj⫺1 ⫺ . 共B-10兲
tive is made discrete with backward Euler differencing. 32.2•144

Tang et al.: Characteristics of the Gas-Lift Process SPE Journal, Vol. 4, No. 3, September 1999 277
In the numerical simulator, the space step is automatically cho- Lift Projects (TUALP) and was an associate professor, produc-
sen to make ⌬L⬍100 ft for the tubing. The time step, ⌬t, is tion section head, a lecturer, and assistant at Southwest Pe-
chosen to guarantee numerical stability, and usually ranges from troleum Inst. (SWPI) in China. He holds a BS degree from SWPI
60 to 180 seconds. and a MS degree from the U. of Tulsa, both in petroleum en-
gineering. Zelimir Schmidt was Professor of petroleum engi-
neering at the U. of Tulsa and Director of TUALP. He spent 10
SI Metric Conversion Factors
years as a production engineer with INA-Naftaplin in the
bbl ⫻ 1.5899 E⫺01 ⫽ m3
former Yugoslavia and served as a consultant to various com-
Btu ⫻ 1.0549 E⫹03 ⫽ joules
panies before joining the U. of Tulsa faculty. He holds an engi-
(°F⫺32) ⫻ 5/9 E⫹00 ⫽ °C neering degree from the U. of Zagreb and MS and PhD de-
ft ⫻ 3.048 E⫺01 ⫽ m grees in petroleum engineering from the U. of Tulsa. A 1987-88
ft2 ⫻ 9.2903 E⫺02 ⫽ m2 Distinguished Lecturer, Schmidt was a 1994-95 Forum Series in
in ⫻ 2.54 E⫺02 ⫽ m South America and Caribbean Steering Committee member,
lbm ⫻ 4.5395 E⫺01 ⫽ kg a 1991-95 Editorial Review Committee member, and 1981-82
lbm/ft3 ⫻ 1.6018 E⫹01 ⫽ kg/m3 and 1994-96 U. of Tulsa Student Chapter Faculty Sponsor.
Mcf/d ⫻ 2.8317 E⫹01 ⫽ m3/d Roger N. Blais is Provost of the U. of Tulsa, Professor of Physics,
psi ⫻ 6.894 757 E⫹03 ⫽ KPa and Associate Director of TUALP. e-mail: roger-
blais@utulsa.edu. He joined the U. of Tulsa faculty in 1977 and
SPEJ has worked on several research and consulting projects in
fluid dynamics and petroleum production. Before that, he
taught at Old Dominion U. Blais holds a BA degree in physics
Yula Tang is a PhD candidate in the Dept. of Petroleum Engi- from the U. of Minnesota and a PhD degree in experimental
neering at the U. of Tulsa, Tulsa, Oklahoma, where he works as physics from the U. of Oklahoma. He served on an Annual
a research assistant in a project jointly sponsored by the U.S. Meeting Technical Committee during 1991-94. Dale R. Doty is
Dept. of Energy (DOE) and the petroleum industry studying Professor of mathematical science at the U. of Tulsa and As-
reservoir performance modeling and comprehensive models sociate Director of TUALP. e-mail: dale-doty@utulsa.edu. He
for horizontal-well completion and optimization. e-mail: joined the faculty in 1975 and has been involved in both re-
tangy@centum.utulsa.edu. His research interests include artifi- search and consulting in the area of petroleum production.
cial lift, well completion, horizontal-well performance, and res- Doty holds BS, MS, and PhD degrees in mathematics from
ervoir engineering. He previously worked with Tulsa U. Artificial Michigan State U.

278 Tang et al.: Characteristics of the Gas-Lift Process SPE Journal, Vol. 4, No. 3, September 1999
Discussion of Transient Dynamic
Characteristics of the Gas-Lift
Unloading Process
A.R. Hasan, SPE, U. of North Dakota, and C.S. Kabir, SPE, Chevron Overseas Production Co.

The authors of this paper deserve credit for modeling a very dif- which the authors have done. That Tang et al. have been ex-
ficult flow problem in production operations. We are heartened by plicit about stating it and we did not is perhaps not a prudent point
their adaptation of our models for countercurrent two-phase flow to make.
and wellbore-heat transfer. However, while adapting our heat- We want to point out that we considered all the appropriate
transfer model, they made some remarks that we feel do not nec- heat-transfer mechanisms in our fully transient wellbore
essarily reflect the premise and usage of our model. We attempt to simulators5-7 for modeling transient wellbore phenomena. In this
clarify those points here. context, we believe the authors’ approach of the use of Vogel’s
They mention that ‘‘ . . . (dQ/dL) wchanges with time, invali- inflow performance relationship is an oversimplification of the
*
dating the constant boundary condition described in the Hasan- physical fact. For simplicity, one can use analytic reservoir flow
Kabir model. The superposition principle is able to account for solution for single-phase flow and subsequently two-phase flow in
this changing heat flux.’’ This statement implies that our model the wellbore.5 Alternatively, analytic pseudopressure formulation
does not mention the superposition principle. This implication is for two-phase flow in the reservoir can be used.
untrue; Ref. 1 contains an approximately half-page discussion on
superposition effects. Equally important, we have shown super-
position treatment for other heat-transfer problems involving fluid References
circulation2 and static-temperature estimation.3 1. Hasan, A.R. and Kabir, C.S.: ‘‘Aspects of Wellbore Heat Transfer
They go on to say that the ‘‘ . . . Hasan-Kabir natural- During Two-Phase Flow,’’ SPEPF 共August 1994兲 211.
convection model uses the empirical correlation proposed by Sa- 2. Hasan, A.R. et al.: ‘‘A Mechanistic Model for Circulating Fluid Tem-
perature,’’ SPEJ 共June 1996兲 133.
gar et al.’’ If they had examined Ref. 4, they would have found
3. Hasan, A.R. and Kabir, C.S.: ‘‘Static Reservoir Temperature Deter-
our statement 共after Eq. 12兲, ‘‘ . . . we used the empirical expres-
mination From Transient Data After Mud Circulation,’’ SPEDC
sion . . . developed by Sagar et al. shown in Appendix B. Alterna- 共March 1994兲 17.
tively, one could use the mechanistic approach proposed by Alves 4. Hasan, A.R. and Kabir, C.S.: ‘‘A Mechanistic Model for Computing
et al. to evaluate the J-T coefficient . . . ’’ Fluid Temperature Profiles in Gas-Lift Wells,’’ SPEPF 共August
They also state that we have assumed either natural or forced 1996兲 179.
convective heat-transfer mechanism for the annulur fluid. How- 5. Hasan, A.R., Kabir, C.S., and Wang, X.: ‘‘Wellbore Two-Phase Flow
ever, ‘‘Hasan-Kabir never combined the two mechanisms to- and Heat Transfer During Transient Testing,’’ SPEJ 共June 1998兲 174.
gether . . . ’’ Again, we feel that this is an unfortunate statement. 6. Hasan, A.R., Kabir, C.S., and Wang, X.: ‘‘Development and Appli-
We used natural convection below the gas-injection point and cation of a Wellbore/Reservoir Simulator for Testing Oil Wells,’’
SPEFE 共September 1997兲 182.
forced convection above it. This is the appropriate treatment,
7. Kabir, C.S. et al.: ‘‘A Wellbore/Reservoir Simulator for Testing Gas
(SPE 57662) Wells in High-Temperature Reservoirs,’’ SPEFE 共June 1996兲 128.

302 SPE Journal 4 共3兲, September 1999 1086-055X/99/4共3兲/302/1/$3.50⫹0.15


Authors’ Reply to Discussion of Transient
Dynamic Characteristics of the
Gas-Lift Unloading Process
Y.L. Tang, SPE, R.N. Blais, SPE, and D.R. Doty, SPE, U. of Tulsa

We are honored and pleased for this opportunity to respond to gravity-gradient term, and the kinematic term approximately can-
Hasan and Kabir’s discussion of our paper. We used Hasan and cel out 共even for the two-phase mixture兲 yielded Eq. 共13兲 for
Kabir’s mechanistic temperature model as the basis for modifica- tubing fluids. This simplification led to the final expressions of
tion in our study because of its excellence. We know they have Eqs. 共18兲 and 共19兲 for tubing and annulus temperature, respec-
applied their model to such drilling and production processes as tively. Thus, in their forced-convection model they used neither
flowing wells, gas-lift wells, and drill-mud circulation/workover Sagar et al.’s empirical correlation nor Alves et al.’s mechanistic
operations. Given the wide application of their model for heat- approach. As we emphasize in our paper and in Ref. 4, we can no
transfer calculations, our paper’s reference to their model is spe- longer prove that the assumption for the pure fluid is valid for a
cific to their papers on gas-lift wells.1-3 two-phase mixture of the tubing fluids. Because both their natural-
Previous researchers typically used geothermal gradient to cal- and forced-convection equations are approximate, we developed
culate fluid temperatures in modeling the unloading process; fluid our equations on the basis of mechanistic equations and fewer
temperatures are important because they influence valve perfor- assumptions.
mance, especially in gas-lift valves with nitrogen-charged domes. Their fourth paragraph states that they also combined natural
These researchers ignored even the steady-state-temperature and forced convection. We could not find this statement or a re-
model, let alone the transient-temperature behavior. Our study de- lated discussion in their papers on gas-lift wells.1-3 For differenti-
scribes unique behavior that differs from steady-state continuous ating the two mechanisms, the annulus-heat-transfer coefficient,
gas lift—the dynamic characteristics of gas-lift unloading. During U a , should be treated differently in the two regions. In their
unloading, not only do the pressure, velocity, and temperature of examples,3 however, U a is assumed to be a constant. By contrast,
the fluids inside the tubing change with time and location, but the the interface between the two regions is dynamic in our study.
location of the interface of injection-gas/kill-fluid in the annulus Tracking the interface location between the two regions for every
also varies with time. The opening and closing of the multistage timestep in the unloading process is more challenging than decid-
gas-lift valves further affect the boundary between the lower ing which equations should be used to calculate the temperature in
natural-convection region and the upper forced-convection region the two regions. The boundary of the two regions is located at the
of the annulus. This distinction must be kept in mind when read- lowest open valve, which is not necessarily the operating valve if
ing our clarifying comments. the operating valve has not been opened. The ideal design requires
We agree with Hasan and Kabir’s assertion in their second that the unloading gas-lift valves open and close sequentially from
paragraph that they already discussed the varying-heat-flux condi- the top to the bottom. As time passes, the boundary between the
tion and apologize for any implication that they had not. The two regions moves downward until the annulus liquid level finally
phrase ‘‘invalidating the boundary condition described,’’ from reaches the operating valve.
Ref. 4 共our first draft兲 unfortunately was lost in our final editing. We disagree with Hasan and Kabir’s last point on the transient
We intended to signify that the variable heat-flux invalidates the inflow performance relationship primarily because the transient
constant-heat-flux inner-boundary conditions 共wellbore兲 used for wellbore response is much faster than the transient reservoir re-
derivation of the dimensionless temperature T D or time function sponse. These minor disagreements, however, should not obscure
f (t). In Ref. 5, we do not say that Hasan and Kabir did not discuss the seminal contribution that Hasan and Kabir made to production
it. We included the equation of superposition to emphasize the engineering. We have merely developed this unloading model and
simulator as an extension of their pioneering work.
difference in temperature calculation between transient unloading
and the steady-state conditions of a flowing well or a stable gas- References
lift well. Hasan and Kabir1 discussed the varying-heat-flux prob- 1. Hasan, A.R. and Kabir, C.S.: ‘‘Heat Transfer During Two-Phase
lem, but their computational examples relate only to gas-lift wells Flow in Wellbores: Part I—Formation Temperature,’’ paper SPE
with constant heat flux. This is precisely the difference between 22866 presented at the 1991 SPE Annual Technical Conference and
Exhibition, Dallas, 6–9 October.
stable continuous-gas-lift and transient unloading.
2. Hasan, A.R. and Kabir, C.S.: ‘‘Aspects of Wellbore Heat Transfer
In their third paragraph, they note that, although they used Sa-
During Two-Phase Flow,’’ SPEPF 共August 1994兲 211.
gar et al.’s empirical expression, they also suggested ‘‘the mecha- 3. Hasan, A.R. and Kabir, C.S.: ‘‘A Mechanistic Model for Computing
nistic approach proposed by Alves et al. to evaluate the J-T coef- Fluid Temperature Profiles in Gas-Lift Wells,’’ SPEPF 共August
ficient . . . .’’ When they first presented their natural convection 1996兲 179.
model,2 they used Sagar et al.’s empirical expression for Fc and 4. Tang, Y.: ‘‘Transient Dynamic Characteristics of Gas-Lift Unload-
did not suggest using Alves et al.’s mechanistic approach. When ing,’’ MS thesis, U. of Tulsa, Tulsa, Oklahoma 共1998兲.
they published their forced-convection model,3 they gave energy 5. Tang, Y.L. et al.: ‘‘A Study on Transient Temperature Characteristics
equations for annulus fluids 共Eq. 9兲 and tubing fluids 共Eq. 11兲, of Gas-Lift Unloading Process,’’ paper presented at the Petroleum
respectively. Their assumption that the pressure-gradient term, the Production Technology Symposium of the 1999 ASME Annual En-
ergy Sources Technology Conference and Exhibition, Houston, 1–3
(SPE 57711) February.

302 SPE Journal 4 共3兲, September 1999 1086-055X/99/4共3兲/302/1/$3.50⫹0.15

You might also like