You are on page 1of 8

IADC/SPE 81636

Improvements in Dynamic Modeling of Underbalanced Drilling
Kjell K. Fjelde, RF-Rogaland Research, Rolv Rommetveit, RF-Rogaland Research, Antonino Merlo,
Eni-Agip Division, and Antonio C.V.M Lage, Petrobras
Copyright 2003, IADC/SPE Underbalanced Technology Conference and Exhibition
This paper was prepared for presentation at the IADC/SPE Underbalanced Technology
Conference and Exhibition held in Houston, Texas, U.S.A., 25–26 March 2003.
This paper was selected for presentation by an IADC/SPE Program Committee following
review of information contained in an abstract submitted by the author(s). Contents of the
paper, as presented, have not been reviewed by the International Association of Drilling
Contractors or the Society of Petroleum Engineers and are subject to correction by the
author(s). The material, as presented, does not necessarily reflect any position of the IADC,
SPE, their officers, or members. Electronic reproduction, distribution, or storage of any part of
this paper for commercial purposes without the written consent of the International Association
of Drilling Contractors or the Society of Petroleum Engineers is prohibited. Permission to
reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not
be copied. The abstract must contain conspicuous acknowledgment of where and by whom the
paper was presented. Write Librarian, SPE, P.O. Box 833836, Richardson, TX 75083-3836
U.S.A., fax 01-972-952-9435.

Abstract
The present work discusses some improvements that have
been introduced in a dynamic model, which was developed for
simulating the two-phase flow transient phenomena associated
with underbalanced drilling operations. The model
enhancements are basically obtained by implementing
mechanistic closure relationships and more accurate numerical
schemes. This process of improvement is validated through
comparison to full-scale experimental data in transient
scenarios, showing that the gains in terms of increasing the
model accuracy are significant.
Introduction
Flow modelling has become more and more important in the
whole planning process of an UBD operation. Steady-state
models have been used for years for designing the operational
window. The only drawback here is that steady-state models
are not able to reproduce accurately the transient behaviour
that occurs during e.g. unloading, connections, and other
inevitable transient situations that occur while performing the
operation. On the other hand, dynamic models have this
capability.
Proper modelling can ensure that the operation can be
designed in an optimum manner, and predict the drawdown for
various conditions. It is of direct importance to maintain the
underbalanced conditions throughout the whole operation to
avoid formation damage. Previous experiences indicate that
even temporarily overbalanced conditions can reduce the
formation productivity. In that sense, both steady-state and
dynamic modelling can be of great importance and, in this
respect; reliable models are necessary. .
The present work is concerned with improvements in transient
modelling of underbalanced operations. The accuracy of the
model, which is an approximation of the reality, depends
heavily on using proper closure laws (mechanistic model) for

flow pattern description, pressure losses and gas slippage.
Another source of error is the basic numerical scheme that
solves the fundamental flow equations.
The process of improvement involves a new mechanistic
approach that has been implemented in a transient model. The
simulation results are compared with full-scale data in both
steady-state and transient conditions, with the main focus on
performing connections. The enhanced model not only matchs
up very well with the experimental data but also shows a
significant improvement compared to older models,
particularly, with regards to describing gas dominated systems
properly.
The paper also focuses on how numerical schemes can be
improved with regards to numerical diffusion. Schemes of
high accuracy are required for giving a correct description of
the maximum flowrates occurring at the separator (e.g. during
the liquid unloading). This is of great importance for sizing
properly the surface equipment, particularly the separator.
Results are presented showing how a numerical scheme with
reduced false diffusion differs from a conventional one that
greatly underestimates the maximum flowrates.
Constructing a Flow Model
In general, multi-phase flow can be described by the
fundamental two-fluid model1. It consists of separate
conservation equations for each of the phases with respect to
mass, momentum and energy. A simpler model can be
obtained by adding the momentum conservations equations
into a mixture momentum equation. This model is named drift
flux. In addition, if the temperature modeling is not of large
importance, it is also possible to neglect the energy equations
and assume a fixed temperature gradient. Based on this
assumption, a simplified version of the drift flux model is
presented bellow.
Conservation Equations. When considering flow of two
fluids e.g. oil and gas, the drift-flux formulation of the
conservation equations is given by

(ρ G α G ) + ∂ (ρ G α G v G ) = m ,…………………… (1)
∂t
∂z

(ρ Lα L ) + ∂ (ρ Lα L v L ) = −m ,…………………. (2)
∂t
∂z

(ρ Lα L v L + ρ Gα G v G ) + ∂ ρ Lα L v 2L + ρ Gα G v G2 + p
∂t
∂z
= −Fw − (ρ Lα L + ρ Gα G ) gsinθ .…………………… (3)

(

)

the model has to be extended with conservation equations for other phases like formation fluids and cuttings. The way this flux is .i . fluid properties and superficial velocities. Finally.……….…………………. gravity and acceleration. In the standard drift-flux approach.i  .i . The flow equations are then solved in time in certain time steps.. µ G. the flow equations across box i in the well can be discretized in an upwind manner. v G. First one has to decide whether to use implicit or explicit numerical schemes for discretizing the flow equations.6.i . slug flow. . ρ L. which is hyperbolic in an ample region of physical parameters2.. the slip relation presents the following form: v G = C 0 (α G v G + α L v L ) + v 0 . Especially.10 presented recently a new model for upward twophase flow in concentric annuli. In addition. implicit schemes have been most popular for multiphase applications 4.….i . σ i . Traditionally. Generally. v Ls. It takes into account bubble flow. the simple slip relation shown above must be replaced with a set of mechanistic procedures that properly models the phase velocities and pressure drops in the various flow patterns that appear. some examples are presented where the mechanistic model is used in a transient model for predicting the dynamic effects taking place during connections. Equation (5) is a symbolic and synthesized representation of this calculation process: ( ) calculate F d1 ..9. d 2 . but the model has been updated with the most recent developments within mechanistic modeling. The model has the original work presented by Caetano5 as basis. One example can be: ∆x ( Fin−1 − Fin ) ………………. One should be aware that the discretization process and the way the equations are discretized represent an approximation of the original flow model. For instance. The idea is to discretize the well into a number of boxes and discretize the flow equations in time and space according to the chosen numerical scheme. in some of the examples presented later. If the focus is only on the mass transport equations and a situation is considered where the flow is upward. Improvements in Numerical Modeling In order to solve the fundamental flow equations. The quality of the mechanistic model is very important because it is strongly associated with the accuracy of the predictions obtained from the transient tool. ρ G. stating that the total pressure gradient depends on friction..i . the mechanistic model calculates the phase volume fractions. Below.. explicit schemes will have a time step that is limited by ∆x /( v G + c) where c is the speed of sound that can range from around 25 – 1500 m/s depending on the flow conditions.g.. .2 IADC/SPE 81636 The two first equations represent the mass transport of gas and liquid.i . the development of an appropriate model for annular flow is important for describing gas dominated systems properly. some kind of numerical strategy must be chosen.12.i . However.7 in addition to several other investigations. For further details about the models consult the references. µ L. dispersed bubble and annular flow. The next step is to decide for a solution strategy of the flow equations. In both cases. water and air) and full-scale experiments from Taquipe. This transient drift-flux model is a system of non-linear partial differential equations. In general. the proposed model performed better. the absolute average errors were lower than 10%. ( ) a.(4) where C0 and v0 are flow dependent parameters. The model we have presented here is quite simplified and is not sufficient for describing the underbalanced conditions properly. It describes the fully transient behavior of both pressure pulse propagation and mass transport. the modeling approach they adopted for gas slippage and pressure losses was not fully mechanistic. their model is refered to as a “non mechanistic” one. several mechanistic approaches have been investigated5.i . v Gs. Rommetveit et al4 presented a model that takes more complex issues into account. where Petrobras has a research facility with an instrumented well (water and nitrogen).. The proposed model was also compared against the mechanistic model developed by Caetano5 and the Beggs and Brill11 correlation. v L. Additional error sources are introduced especially when it comes to the description of sharp transients like for instance the liquid peak following a gas pulse at surface. it is necessary to provide an appropriate model for the source term in the momentum equation that corresponds to the frictional pressure losses. The advantage of the first class of scheme is that they are faster since the time steps ∆t will be only be limited by ∆x /( v G ) where ∆x is the length of the boxes in the discretized well and v G is the maximum gas velocity. the closure of the system is achieved by specifying density models for each of the phases and a slip relation between the phase velocities. Next sessions are presenting some improvements that have been achieved with respect to mechanistic modeling and to enhance the numerical strategies. On the other hand.14. After the discretization process. Lage8.i  → dp dp dp    α G. Because of that. Model predictions were compared to both small-scale data5 (kerosene. explicit schemes are much simpler to use and treats flow networks very easily.i . the well is divided in N boxes of a certain length. The third one is the total momentum balance. Due to the complexity of the equations a numerical strategy has to be chosen. phase velocities and pressure gradient.3.i .(6) ∆t n n where the numerical flux is given by Fi = (ρ g α g v g ) i and (ρ g α g ) in +1 − (ρ g α g ) in = represents gas flow out of the box. but also explicit formulations have been considered e. ( ) g. On the other hand. ( ) f. 13. Advanced PVT models have to be included for proper modeling of the phase volumes and possible phase transitions.(5) dz dz dz   Throughout times.. Based on the input of geometry. which is expected to match up with the reality. An Improved Mechanistic Model for Upward Flow in Concentric Annuli A mechanistic model is composed of a procedure for flow pattern identification and separate models for each specific flow pattern.

the masses in each box are averaged. tend to be smeared out quite extensively. Front tracking techniques 4. the simulated bottomhole pressure is compared against the experimental data for three different situations. In the next section.e. Again a good agreement between experimental and simulated data can be observed. The results are shown in Figure 8. Figure 7 compares the simulated and experimental bottomhole pressure development. quite recently these techniques have also been introduced within multiphase flow modeling. The drawback is that one has to use fronts all places where sharp gradients occur. the well is initially in a steady state situation where the gas rate is 800 scfm and the liquid rate is 160 gpm. This effect is called numerical diffusion. In the fourth example. After the connection. they will be distributed as piecewise linear functions. After the steady state conditions are reached. We will try to demonstrate this mechanism with an example. the situation is somewhat special. Using n or n+1 in defining the numerical scheme decides whether our scheme is explicit or implicit.12 can be used where an additional scheme is used to track and follow transition regions like e. Then during the next time step. The bottomhole pressure decreases until a new steady state situation is obtained. Another alternative is to introduce so called slope limiter or MUSCL techniques15. see Figure 3. There are now large volumes of gas present in the well and the drawdown is nearly 700 psi compared to the original liquid filled well. while n+1 represent the new time level that will be reached. we present the steady-state results for a situation where the gas rates are very large compared to the liquid rates which lead to a highly gas dominated system. The well is unloaded with a gas rate of 300 scfm and a liquid rate of 160 gpm. The well pressures increases due to migration of free gas in a closed environment. Again a reasonable agreement between experimental and simulated data can be seen. In addition. After 10 min. Transient Modeling of Connections. Four pressure-temperature sensors were disposed along the casing string logging experimental data for the comparison. we notice that the amount of gas that leaves the box is lower than that seen in Figure 3. In addition. In addition. instead of treating masses as averaged values i. focus will be on the numerical issues where we show how the slope limiter technique can be used for minimizing numerical diffusion. In that sense. Figure 6 shows the geometry of the research well. The results are compared against full-scale data from a research well in the Taquipe Field. Several remedies exist. we have compared the new mechanistic approach with the model presented in (Ref. During the next time step of size ∆x/(2v g ) the gas bubble will migrate to the end of this box as shown in Figure 2. Originally. 25 % of the gas will enter the next box. Then the gas rate is increased to 600 gpm while the liquid rate is reduced to 80 gpm. In the last example. After a while the new steady state situation is reached. the choke is opened and the steady state condition is achieved after a while. Figure 1 displays a situation in a discretized well where a pure gas bubble has migrated half way through one of the boxes. The concept is quite simple. a gas slug. The results are shown in Figure 9. Both models give very similar results. Example 1. In the second session. a connection is performed where both liquid and gas supply is stopped.IADC/SPE 81636 approximated is very important for the results and different alternatives exist. liquid injection was resumed. All first order schemes have this problem. Example 4. However. The new mechanistic model gives a better agreement with the experimental data compared to other model. introducing one front only solves the diffusion problem locally. use of front tracking techniques can make model implementation quite complex. Example 3. constant values in each box. Then the gas rate is increased to 1000 scfm and the liquid rate is reduced to 80 gpm. They reduce numerical diffusion automatically without the need to have exact knowledge about where the sharp transients are. Note that Equation (6) has to be solved for each box in the flow network.g. The fluids considered were water and nitrogen. It can be observed that the gas volume fraction profile has been smeared out compared to the real situation shown in Figure 2. First order scheme have problems with so called numerical diffusion. Then a 10 min connection is simulated where the choke is closed. The slope limiter approach reduces numerical diffusion and a second order or higher order numerical scheme is obtained. the averaged gas volume in the box which contains the gas front is 50 %. Results First some examples will be shown where the new mechanistic approach is used in a transient model for predicting the dynamic effects taking place during connections. Since this model was not fully 3 mechanistic we will refer to this as the non mechanistic model. The difference is less than 10 %. Hence. The calculated gas volume profile at the new time level is shown in Figure 4. In this case. these were developed for use within aerodynamics. First of all drillstring gas injection was used instead of parasite gas injection. Both models give quite good agreement but the new mechanistic model is slightly better in this case. In a standard numerical scheme. In the first three examples. 4). The scheme given here is an example of a first order upwind scheme. Brazil. They reduce numerical diffusion to a minimum. Initially. The examples given there are of more academic nature but clearly reveal the significance of having high accuracy numerical schemes for predicting the flow. During the next time step. A very good agreement was observed. The way the numerical flux is approximated is very important for the results and different alternatives exist. An example is shown in Figure 5. Both models tend to underestimate the bottomhole pressure. Sharp gradients like for instance regions where the gas volume fraction changes rapidly. Subscript n refers to values from the last calculated time level. After steady-state conditions are achieved a connection is performed where the liquid rate was turned to zero. the initial conditions in the well prior to the experimental session were not known due to a . some comparisons will be presented. Example 2. Inclusion of these techniques is quite simple. The slopes of these functions can be constructed using a so called slopelimiter15 which takes into account the profile of the sharp transition zone region. the well is a steady state situation with a gas rate of 300 scfm and a liquid rate of 160 gpm.

The model is able to predict the transient behavior taking place during connections. Especially in example 4. The examples here show that a transient tool can be valuable in evaluating different procedures and make the right choices. the model under predicted pressures quite extensively. A pipe of 1000 m depth and 0. In the following. Improved Numerical Modeling. The tendency was explained by the lack of a churn flow model in the new mechanistic approach. the time for gas entrance at surface is predicted at a later stage when using the second order scheme. The numerical solution of the transient flow problem is shown in Figures 11-14. In general. the gas breaks through at surface. During underbalanced operations. The gas rate is then increased quite significantly to 1000 scfm. Gas and liquid is then injected at the bottom of the pipe rates. the expected drop in bottomhole pressure can be observed as gas migrates upwards the pipe. where the difference between collapse and pore pressure are low. In Figure 11. This is probably caused by insufficient annular flow modeling. On the other hand the maximum flow rate peaks are much better resolved by the second order scheme. The pipe is initially filled with liquid. In IADC/SPE 81636 addition. it is of importance that one can rely on the predicted results and to have an idea of where the models are reliable. Earlier comparison of this new model against experimental . As seen in examples 5 and 6. a further decrease in bottomhole pressure can be seen until gas enters at surface around 1000 sec. The pipe has been discretized in 25 boxes where box 1 is at the bottom and box 25 is at top of the pipe. see Figure 16. After this the total gas fraction in the well will decrease and the bottomhole pressure will start to increase. The model used does not treat annular flow properly which is clearly seen from the results. Liquid injection is resumed after 1200 sec. The “non mechanistic model” underestimates bottomhole pressure quite significantly. A traditional first order implicit upwind scheme is compared against an upgraded second order version of the same scheme by use of slope limiter techniques. On the other hand.1 m inner diameter is considered. The reason for this is that the first order scheme diffuses the sharp transition zone between the one-phase flow region and the two-phase flow region. see Figures 15 and 17. As the gas expands. the two numerical schemes predict quite different values for the maximum peak rates. the simulated scenarios are initiated from a steady-state situation with a gas rate and liquid rate of 300 scfm and 40 gpm. A good agreement could be observed. the transition zone between one-phase and two-phase flow region) numerical diffusion can smear out these transition zones quite extensively. The pipe is initially filled with stagnant liquid. In Figure 10. A 1000 m vertical pipe with a diameter of 0.4 period prior to the experiment where the injection was stopped and phase segregation took place. Another issue that could be commented is the negative liquid flowrates observed in Figure 16. different examples are given where we demonstrate the effect of introducing anti-diffusion mechanisms in the numerical schemes.0 kg/s. it is of importance to have an appropriate numerical scheme that gives reliable results. The injection rates are increased to 7. A gas pulse is now migrating upwards towards surface. Numerical diffusion in the first order scheme tends to make the V-shape somewhat wider compared to the second order scheme although the difference is not significantly large. In some cases. This example focuses on the typical transient effects taking place during unloading of a well.5 kg/s for the liquid and 0. The second order scheme predicts a much larger maximum value compared to that given by the first order scheme. This leads to a highly gas dominated well and the bottomhole pressure decreases quite significantly before the steady state is reached. the pressure stabilises at the original level. This is followed by a peak in the liquid rate.g. In this case. Use of a slug flow model in churn flow regions leads to pressure predictions lower than measured values.15 kg/s for the gas. But at the same time. Conclusion A new mechanistic approach has been implemented in a transient model and evaluated against experimental data. This is clearly seen in Figure 14. The other model shown in the examples gave also satisfactory results although it behaved slightly worse than the new model. This leads to a drop in bottomhole pressure due to reduced friction pressure losses. After a while. In regions where the gas volume fraction changes rapidly (e. After the gas has left the well. if separator design and transient surface flow rates are to be evaluated. On the other hand. The rates are kept constant in 200 sec before the injection stops. This effect is typically seen in the prediction of the maximum surface flowrates where numerical diffusion leads to under prediction of the maximum rates. Example 5. which leads to negative liquid flowrates in the upper parts of the well Discussion The first four examples show how transient modeling can be a valuable tool for predicting the dynamics throughout connections. As seen in Figures 12 and 13.2 m is considered. The liquid and gas rates are 150 kg/s and 1. this effect is reduced quite extensively. it is of importance to maintain wellbore pressures below formation pressures to avoid formation damage. However. it is of importance to minimize the pressure fluctuations that appear throughout the connections. the experimental steady state bottomhole pressure is compared against the results given by the two different models. This tendency was more profound during parasite gas injection compared to drillstring injection. In this respect. the new mechanistic approach seemed to give quite satisfactory results. see Figure 15. the new mechanistic model gives quite good results. by introducing higher order schemes. the well was dominated by large gas rates and annular flow was present in large parts of the well. We notice the typical V shape curve of the bottomhole pressure as the gas pulse migrates upwards. the choice of an appropriate numerical scheme can be of importance when evaluating transients during underbalanced drilling. it had a tendency for slightly underestimating pressure which was also the conclusion in reference 10 where a wider range of data was analyzed. wellbore pressures must not be lower than collapse pressure. Example 6. In cases. The reason for this is that the upward movement of gas forces the liquid downwards. In the examples shown here.

:”A Mechanistic Model for Predicting Annulus Bottomhole Pressures in Pumping Wells”. 279-293. J. for design and evaluation of separator equipment it is of importance to use schemes of high accuracy. and Time. K t = time.R. Pa (psi) T = Temperature. We have demonstrated a simple technique (slope limiter) that can be used for these purposes. Dissertation. INP. E. 6.D. France (1991). Norway (2000). Ishii. kg/(m3⋅s) p = pressure. l = liquid ls = liquid superficial Superscripts n = time step level 5 References 1. Nakagawa and A. 7. Hence.P.D. Papadimitriou. mL-2t-2. L. T.” SPE Production Engineering. The new model seems to give satisfactory results also in these situations. paper SPE 63127 presented at 2000 Annual Technical Conference and Exhibition held in Dallas. Lage. H. Houston. 1999. mL-3. Nomenclature d1 = outer diameter of drillstring. Duchet-Suchaux and C. 13.: “An Experimental and Theoretical Investigation of Upward Two-Phase Flow in Annuli”.V. (1975). 14. m d2 = inner diameter of casing.C.: “Ecoulements Diphasiques Instationnaires en Conduite Horizontale”. D.54 psix 6. Rommetveit.F. A. Since control of bottomhole pressure is the main issue during underbalanced drilling. Stavanger College.H. L. 10. K. O. SPEJ (September 2002). Lage. Agip. A.: “A Study of Two-Phase Flow in Inclined Pipes”.S. mL-1t-2. Sævareid. 2.K: “Numerical Schemes for Complex Nonlinear Hyperbolic Systems of Equations”.W. 1-4 October. C. Georges. t.: “Upward Two-Phase Flow Through an Annulus”.785 412 in x 2. TX. Texas.V. France (1989). R. Caetano. C. Int. Pa/m F = Numerical flux function or physical flux g = gravity acceleration. Multiphase Flow (1992) 18. P. Lt-1. and Kabir. The University of Tulsa (1985). Acknowledgements We would like to thank Agip and Pemex/IMP for funding parts of this work.: “Numerical computation of internal and external flows” Volume 1 & 2.: “Mechanistic Model for Upward Two-Phase Flow in Annuli”. A. JPT (May 1973). 4.W. A.: Thermo-Fluid Dynamic Theory of Two-Phase Flow. Norway. m Fw = frictional pressure loss.C. C. S. 7-9 Sept. PhD dissertation.F.048 gal x 3. mL-3t-1. R. PhD dissertation. E. Lt-2. ”The dynamic two-fluid modell OLGA: Theory and application. 325. J. Malnes. 5. 2. Bijleveld.C. Moe. m α = volumetric fraction. dimensionless θ = angle with the horizontal.: “Two-Phase Flow in Vertical and Inclined Annuli”. 607-617. Thèron.: ”Dynamic Underbalanced Drilling Effects are Predicted by Design Modell. L. and Time. 3. and S. 5-6 May.A. The other issue that has been discussed is the effect of numerical diffusion on the transient results. May 1991.” Paper SPE 56920 presented at the 1999 Offshore Europe Conference held in Aberdeen. Most models have problems with describing the annular flow pattern problem which typically occurs in case of gas dominated systems. D. 8. 12.M. B.V.C. Benzoni-Gavage. kg/m3 (lbm/gal) σ = interfacial tension. m/s2 m = mass transfer between phases. 6. Oklahoma City. 9. Lage. it is of direct importance to have models that reveals the reality as best as possible. s v = velocity. and Brill. Nuland. Hasan. A. Lyon. paper SPE 21669 presented at 1991 Production Operations Symposium.: “Analyse Numérique des Modèles Hydrodynamiques D`écoulements Diphasiques Instationnaires dans le Réseaux de Production Pétrolière”.M. Pauchon. C.: “Two-phase Flow models and Experiments for Low-Head and Underbalanced Drilling”.V. 15. SI Metric Conversion Factors ft x 3. Fjelde. In that sense. N/m ∆t = discrete timestep ∆x = length of discrete boxes Subscripts a = acceleration f = friction g = gas or gravitational gs = gas superficial i = related to the box no.M Lage. Johh Wiley & Sons 1990. O. Guarneri. mL-1t-2. Ph. m/s z = spatial coordinate. PhD dissertation. R. Univeristy of Bergen. 11. M. No. April 7-9. and Shoham. Beggs. we believe that the new model can give a positive contribution. We would also like to thank Petrobras. Hirch.894 757 E-01 = m E-03 = m3 E+00 = cm E+00 = kPa . R. PhD dissertation. TotalFinaElf and Shell for funding the fullscale UBD experiments in Taquipe. Henriot. Bendiksen.IADC/SPE 81636 steady-state data as well as other models has indicated that the new model gives quite good results with an average error less than 10 %. Eyrolles. ENS. Leibovici: TACITE: Contribution of fluid composition tracking on transient multiphase flow simulation” Paper OTC 8563 presented at the 1997 Offshore Technology Conference. K. 171-180. radians ρ = density. A. Toulouse. We have shown that classical first order schemes are not reliable in predicting maximum flowrate peaks that can occur at surface.M. V.

Use of the slopelimiter approach.6 IADC/SPE 81636 Figure 1 . .The calculated situation after one timestep. Figure 4 . Figure 5 . Figure 3 .The gas is averaged acoss the whole box.The initial situation in the discretized well.The situation after one time step. Figure 2 .

Casing 1300 1250 1200 1150 1100 1050 1000 950 900 850 800 0. Gas rate = 1000 scfm.276 in. Comparison of predicted bottomhole pressure using a first and second order scheme. Figure 6 – Configuration of the Taquipe well. Liquid rate = 160 gpm. 500 1000 1500 2000 Time (sec) Figure 11 – Unloading scenario. Liquid rate = 80 gpm.00 Figure 9 – Parasite Injection. Bottomhole Pressure 100 1500 1450 1400 1350 1300 1250 1200 1150 1100 1050 1000 Experimental data Mechanistic Model Non Mechanistic Model Pressure (Bar) Pressure (psi) Bottomhole pressure 95 90 First order method 85 Second order method 80 75 70 0 0 20 40 60 80 Time (min) Figure 8 – Parasite Injection. .00 Logging tool at 1262 m ID=6. Gas rate = 1000 scfm. Bottomhole pressure Bottomhole pressure 1400 1900 1800 1700 1600 1500 1400 1300 1200 1100 1000 1200 Experimental data Mechanistic Model Non Mecanistic Model Pressure (psi) Pressure (psi) 40.00 Time (min) Experimental data 1000 800 Non Mechanistic Model 600 Mechanistic Model 400 200 0 20 40 60 80 0 0 Time (min) 10 20 30 40 50 60 70 Time (min) Figure 7 – Parasite Injection. Figure 10 – Drillstring Injection. Open Connection.00 Experimental data Mechanistic Model Non Mechanistic Model 20. Drill Pipe ID=2.764 in. Gas rate = 600 scfm. Liquid rate = 40 gpm – Steady State. Open Connection. Closed Connection. 60. Gas T P Parasite String P 2 3/8 in. Tubing String at 760 m Bottomhole pressure T Pressure (psi) Liquid 7 Gauge at 185 m Gauge at 605 m Gauge at 998 m Orifice and Check Valve 7 in. Gas rate = 300 scfm. Liquid rate = 40 gpm.IADC/SPE 81636 P T Separator 3 1/2 in.

04 First Order Method 0.00 First order method 0. 0 1000 2000 3000 Time (sec) Figure 17 – Comparison of predicted outlet gas rates using a first and second order scheme. Gas Volume Fraction 0.05 Flowrate (kg/s) Gas Fraction 0.02 0. Comparison of predicted bottomhole pressure using a first and second order scheme.00 0 500 1000 1500 2000 -2 Time (sec) 0 1000 2000 3000 Time (sec) Figure 13 – Unloading scenario. Figure 15 – Gas pulse example. Comparison of predicted outlet liquid rates using a first and second order scheme.25 Outlet Gas Flowrate 0. Figure 16 – Gas pulse example.20 Flowrate (kg/s) Rate (kg/s) First Order Method 100 6 First Order Method 4 Second order Method 2 0 0. Comparison of predicted outlet gas rates using a first and second order scheme.80 Second order method 0.15 Second order method First order method 0. Outlet Liquid Flowrate Outlet Gas Flowrate 10 1.60 0.60 1.40 8 1. .01 0 Figure 14 – Unloading scenario.8 IADC/SPE 81636 Bottomhole Pressure 110 Outlet Liquid Flowrate Pressure (Bar) 700 Rate (kg/s) 600 500 400 First order method 300 Second order method 105 Second order Method 95 90 200 100 85 0 0 500 1000 1500 0 2000 1000 2000 3000 Time (sec) Time (sec) Figure 12 – Unloading scenario.20 0.00 0 5 10 15 20 Number of Boxes 0.20 1.10 0.03 Second order Method 0.40 0. Comparison of predicted outlet liquid rates using a first and second order scheme.06 0.05 0. Comparison of predicted gas volume fraction versus depth using a first and second order scheme.