Você está na página 1de 11

-

PRODUCTION ENGINEERING FUNDAMENTALS I 111

vertical flow-mass balance equations result in a set of eight equations to be solved for eight unknowns. Churn
(transition) flow is treated as part of slug flow. In mist flow, a Newton-Raphson iterative procedure is used to find the
thickness of the falling liquid film present on the inside pipe wall.

Chokshi, Schmidt, and Doty [62J report on an extensive experimental project in a vertical, l,348-ft-long flow pipe
of 31f2 in. diameter. In addition to downhole pressure transducers, a gamma-ray densitometer provided accurate data on
liquid holdup. Measurement data were used for comparison purposes only, and the pressure gradient formulas
developed by the authors reflect a fully mechanistic approach. Only three flow patterns-bubble, slug, and annular
mist-are consideredf and the flow is assumed to be vertical. In bubble flowf the drift-flux model is foIlowed; and a set
of sevenequations is to be solved for slug flow. For mist flow, the separate flow of a gas core and a liquid film is
describedf with liquid film thickness as the main variable.

Kaya, Sarica, and Brill [34J presented a calculation mode! for deviated wells and have included the effect of pipe
inclination on flow pattern transitions using the Barnea [32J equations. Bubble, dispersed bubble, slug, churn, and
annular mist flow patterns are distinguished. For bubble flow, the drift-flux mode! is usedi for dispersed bubble flow, the
no-slip mode! is used. They use, with significant modifications, the Chokshi, Schmidt, and Doty [62J model in slug
flow. Churo flow is treated similar to slug flow, and the Ansari et aI. [61] model is adopted for annular mist flow.

The pressure drop calculation model of Gomez et aI. [63J represents a new approach to solving multiphase flow
problems because it treats horizontal, inclined, and vertical flows with one single unified mechanistic model. This
approach has definite advantages when a simultaneous solution of weIlbore and surface flow problems is sought, like in
No daI Analyses.

2.5.3.5 Calculation of pressure traverses. Previous sections described the numerous ways pressure
gradients in multiphase oil weIl flow can be determined. Although this is the most difficult part of multiphase flow
calculations, it is the pressure drop in the weIl or the pressure distribution along the tubing that production engineers
are mostly interested in. The present section details how the basic pressure gradient equations are solved to reach that
goal.

The general pressure gradient equations for vertical or inclined multiphase flow (see Section 2.5.2.3) are ordinary,
non-algebraic differential equations that can be written in the foIlowing symbolic form:

~ =J (p, T) =J (p, h) 2.245


dh

As seen, pressure gradient is basicaIly a function of the prevailing pressure and temperature. However, flowing
remperature depends on weIl depth (this function is assumed to be known in our treatment) and this is the reason why
me right-hand side of the equation can be expressed as a function of pressure and flow length. The analytical solution
Df the previous differential equation is impossible because the right-hand side contains many empirical functions
deviation factorf friction factor, etc.) and further it cannot normaIly be expressed in an analytical formo

Thus, some numerical method must be employed for the solution, i.e. for the calculation of pressure distribution
i:Iong the flow pipe. The method most often used involves a trial-and-error (or iterative) procedure ideaIly suited to
cornputer calculations. It is based on the conversion of the basic differential equation into a difference equation using
ir-jte differences. After conversion, the right-hand side of Equation 2.245 is evaluated at the average conditions of the
~-en increment as shown here. Provided these increments are smaIl enough, then the change of the flowing gradient
:s negligible and the new equation close!y approximates the original one:

2.246

This equation, in the knowledge of a previously set pressure or depth increment, aIlows the calculation of the
_-'.:-.own incremento The only complication is that the pressure gradient at average conditions (the right-hand side)
- :::lds on average temperature, which in turn is a function of weIl depth. Thus an iterative scheme must be used to
= =-

. - :::le increment from the other. The details of a complete pressure distribution calculation are given as foIlows.
=-
112 I GAS LIFr MANUAL

The calculation of a flowing pressure


traverse in a multiphase oil well starts with
dividing the total well into pressure or depth
increments, depending on the approach
desired. Figure 2-40 shows a flowchart for
calculating the bottomhole pressure based on
set pressure increments. Calculations start
from the known wellhead conditions P j = Pwh
and hj = O. First, the actual pressure increment
,1p is set, whose proper se!ection will be
discussed later. Then, in order to calculate an
average temperature, which in turn is a
function of the depth increment to be sought,
an initial value of ,1h needs to be estimated. ITER = O

After average pressure in the actual calculation Pavg = Pi + Ap/2


step is found, average temperature is calculated
from the temperature-depth function that is
assumed to be known. Since average
conditions in the actual calculation step are
known, the pVT properties (densities, Calculate pVT properties Ah = Ahcalc
viscosities, etc.) of the flowing fluids can be at Pavg, T avg ITER = ITER + 1
calculated. The next step is to evaluate the
pressure gradient using the appropriate Calculate dp/dh
procedure prescribed in the utilized pressure Ahcalc = Ap/(dp/dh)
drop calculation mode!.

The right-hand side of Equation 2.246 >e


being known the depth increment ,1hcalo
corresponding to the previously set pressure
increment is easily calculated from that
formula. This increment is compared to its
hi+ 1 = hi + Ahcalc
assumed value and if they differ by more than
the previously set accuracy ê a new iteration Pi+1 = Pi - Ap
step is performed with an assumed depth
increment equal to the value just calculated. No
This iterative process is continued until the last
two values of the depth increment match with
the required accuracy. At this point, the depth
increment belonging to the actual pressure
increment is known, specifying the pressure
and depth at a lower point in the well. If the
well bottom is not yet reached, a new step is
processed and the calculation of a new depth
increment follows by repeating the previous Fig.2--40 Flowchart of pressure traverse calculation for iterating on depth increment.
procedure. When the well bottom is reached,
bottomhole pressure is easily calculated by
interpolation.

In case the wellhead pressure is to be calculated from a known bottomhole pressure, the previously detailed process
is slightly modified. In this case, calculations are started from the well bottom with the average pressures and
temperatures calculated accordingly.

The other approach to pressure traverse calculations is to divide the flow pipe into a series of depth increments
and calculate the corresponding pressure increments. Again, Equation 2.246 is used but now it is solved for the pressure
increment ,1Pcalc The flowchart of this calculation mode! is given in Figure 2--41 for a case where wellhead pressure is
sought. Calculations follow the one described previously and are started from the conditions at the well bottom. Again,
PRODUCTION ENGINEERING FUNDAMENTALS I 113

calculation of bottomhole pressures requires


only minor changes in the procedure given.

The two basic solutions (iterating on depth


or on pressure increment) detailed previously
offer different advantages and drawbacks.
Setting the pressure increment and then finding
the corresponding depth increment may not
converge when the pressure gradient
approaches zero, a condition often occurring in
downward flow. In such cases, the solution of
Equation 2.246 has pressure gradient in the
denominator and may thus result in an infinite
depth incremento On the other hand, the
advantage of this approach is that no iterations ITER = O
at alI are required if flowing temperature is Pavg = Pi - Llp/2
constant along the pipe. The other approach
iterating on pressure) alIows one to exactly
folIow the well's inclination profile since depth
increments can be selected at wil1. In addition,
no interpolation is required at the end of the Calculate pVT properties Llp = LlPcalc
calculation processo Further, the convergence at Pavg, Tavg ITER = ITER + 1
problem mentioned previously does not exist
because the solution of Equation 2.246 for L1p
involves multiplication only. Calculate dp/dh
LlPcalc = Llh/(dp/dh)

As with every numerical solution, the


dlCcuracyof the previous procedures relies on the
>E
proper selection of step sizes and iteration
1Dlerances.NaturalIy, the smaller the pressure or
depth increment, the more accurate the solution.
Small increments, on the other hand, require
more steps to be taken and increase calculation hj+1 = hj - Llh
time. Consequently, the proper selection of Pi+ 1 = Pi - LlPcalc
íncrement sizes is a matter of finding a balance
between the required accuracy and computation No
time. Some guidelines are given as follows:

• For obtaining sufficient accuracy, the basic


mIe is that increments should be selected
50 that pressure gradient is essentialIy
constant inside the incremento

• Pressure increments should vary with welI


depth 50 that they decrease at smaller
depths. This is recommended because Fig. 2-41 Flowchart of pressure traverse calculation for iterating on pressure
increment.
fIowing gradients decrease in the same
direction. UsualIy, increments of about
10% of the actual pressure are used.

• Selection of depth increments should consider the welI's deviation profile by varying the increment sizes.

• The well should be divided into segments with a new segment wherever a significant change in inclination angle or fluid
rates occurs. The latter happens at gas lift injection points or in case of commingled production.
114 I GAS Lwr MANUAL

2.5.3.6 Accuracy and selection of pressure drop calculation models.

2.5.3.6.1 Introduction. Previous parts of Section 2.5.3 described in detail the many calculation models available to
present-day petroleum engineers dealing with multiphase flow problems. As seen from the long list of procedures, there
is a big choice available today. This section investigates the accuracies obtainable from the use of the different models
and presents the right philosophy to be adopted by the production engineer.

Before the detailed treatment of the objectives set forth previously, let us examine the behavior of empirical models.
As discussed before, empirical correlations were original1y developed under widely different physical circumstances. This
fact could readily be seen in Table 2-8 where the main parameters of experimental data used by the different authors are
shown. Grouping of the models is done according to the cri teria described in Section 2.5.3.3.1. Experimental data may
originate from laboratory measurements, special field tests, or routine field measurements. The pipe lengths, sizes, and
liquids investigated are listed along with the total number of measured pressure traverse curves. The table proves that
data of the different authors represent very different conditions.

field
test
1000
AUTHORIIII
III
Data
field
test
labData
test
#49
106
139
25
584 af
4000
148
581
175
Group I
field 1964
1967
1965
1958
1985
1963
1954
1961
1952
Publ. 1973 100
0&100
95
Water
0&100
OO
0&100
0-98 50 Cut, %

labia 2-8 Summary of experimental data sets used by the authors of empirical multiphase pressure drop correlations.

The difference in original conditions is shown for two further aspects in Figure 2-42 and Figure 2-43, where the
liquid flow rate and GLR ranges of the basic data are displayed for each correlation.

POETTMANN-CARPENTER
GILBERT
BAXENDELL
-- -
BAXENDELL-THOMAS

•-
DUNS-ROS

FANCHER - BROWN
GAITHER ET AL.

HAGEDORN - BROWN I
HAGEDORN - BROWN 11

ORKISZEWSKI
BEGGS - BRILL

MUKHERJEE - BRILL

o 1000 2000 3000 4000 5000 6000 7000 8000


Liquid Production Rate, BPD

Fig. 2-42 Liquíd flow rate ranges of different empirícal multiphase pressure drop correlations.
PRODUCTION ENGINEERING FUNDAMENTALS I 115

POETTMANN-CARPENTER

GILBERT
BAXENDELL

BAXENDELL-THOMAS
DUNS - ROS
I•
FANCHER - BROWN

GAITHER ET AL.
HAGEDORN - BROWN I

HAGEDORN - BROWN 1I

ORKISZEWSKI
BEGGS - BRILL

MUKHERJEE - BRILL

o 2000 4000 6000 8000 10000

Gas-Liquid Ratio, eu ft/bbl

Fig. 2-43 GLR ranges of different empirical multiphase pressure drop correlations.

The table and the two figures given previously indicate that authors used data from wide!y different physical
conditions (pipe sizes, flow rate, GLR-ranges, etc.) to develop their respective pressure drop prediction methods.
Consequently, if one correlation is used outside its original conditions, predictions with heavy errors are likely to occur.
Although this is quite obvious, in practice all correlations are being used irrespective of the flow conditions at hand.
2.5.3.6.2 Possible sources of prediction errors. As it is generally known, calculated pressure drops usually deviate
irom those actually measured in a wel!. Prediction errors can be attributed to a host of known and unknown issues. The
classification presented as follows is an approach to sort the possible sources of calculation errors in some larger
categories. A detailed analysis of the described sources of error would allow one to draw important conclusions for the
iurther improvement of accuracy in vertical two-phase pressure drop calculations.

Characteristics af the Physical Madel

When deve!oping a calculation mode! for multiphase pressure drop determination, it is never possible to include
all conceivable features actually affecting the pressure drop. Therefore, the behavior of the real and thus extremely
complex multiphase flow must be approximated by a properly chosen simplified physical mode!. The physical mode!
adapted has a decisive effect on the accuracy and behavior of the calculation mode! based on it. The more
comprehensive the physical mode! applied, the more like!y it is that flow conditions are properly described.

It was shown before that available calculation methods utilize different physical models for the description of
:nultiphase flow phenomena. Out of the many possible simplifications and approximations used to build the physical
:node!, two criteria are found to play an important role in the determination of calculation accuracy.

• Physical phenomena are essentially different in the various flow patterns occurring in multiphase vertical flow,
considerations for this effect must be included in every computation method. Mechanistic mode!s, by their
nature, take care of this problem, but some of the earlier empirical methods lack this feature resulting in lower
calculation accuracy.

• Under certain conditions, the effect of gas slippage in liquid can be very important. ln such cases correlations
considering no slip between phases (e.g. Poettmann-Carpenter and its improvements) may perform poorly.
116 I GAS Lwr MANUAL

Errors in Fluid Properties

When performing the caIculation of a multiphase pressure traverse several thermodynamic parameters of the flowing
fluids (oil, water, natural gas) must be known in the wide ranges of pressure and temperature occurring in a well. But
experimental pVT data on these are usually available at reservo ir temperature and at projected reservoir pressures only.
A good example is oil volume factor Bo, commonly called FVF by reservoir engineers, usually measured at reservoir
conditions only. For production engineering caIculations, however, Bo must be known in the wide ranges of pressure and
temperature starting from wellhead to bottomhole conditions. This situation also holds for other thermodynamic
parameters like solution GOR, oil viscosity, bubblepoint pressure, etc. for which measured data are seldom available.

ln a lack of experimental data, which is the general case, the required thermodynamic properties of oil, water, and
gas have to be computed using standard petroleum industry correlations. Use of such correlations instead of
measurement data inevitably introduces errors and may heavily affect the accuracy of caIculated pressure drops.

Of the many fluid properties required for pressure drop caIculations, the value of the bubblepoint pressure plays
the most decisive role as shown by several authors. [55, 69J This is primarily due to the fact that the value of the
bubblepoint pressure determines the point in the well below which single-phase flow prevails and above which
multiphase flow starts. Since caIculation accuracy in single-phase flow is excellent but is much worse in multiphase flow,
a small error in the value of bubblepoint pressure can bring about great errors in the caIculated pressure drop. Standard
bubblepoint pressure correlations, however, may give 50% or higher prediction errors [36J, thus further increasing the
deviation of caIculated and measured pressure drops. Because of this, any multiphase pressure drop model may perform
differently if, in lack of experimental data, different correlations are used for the caIculation of bubblepoint pressure.
This effect is greatly affected by the direction of caIculation (from wellhead to bottom or vice versa) but that will be
discussed later. Consequently, use of measured bubblepoint pressure data may greatly improve caIculation accuracy, but
this holds for the rest of the fluid properties as well.

Most of the pressure drop caIculation models were developed for two discrete phases, a liquid and a gas one and
are therefore properly called two-phase methods. ln case the liquid phase is not pure oil or water but a mixture of
these, the properties of this phase have to be found based on the properties of oil and water. ln lack of sufficient
research results, the usual solution is to apply a mixing mIe based on the actual flow rates of the components. This
approach is at least questionable since it does not inc1ude the effects of slippage between oil and water, the occurrence
of emulsions, etc. Therefore, water-cut oil production usually increases the errors in caIculated pressure drops and is
a subject of future research.

Empirical Correlations

AlI empirical pressure drop correlations inc1ude several empirical functions applicable only in certain specified
ranges, outside which their performance is at least questionable. There is no use to list alI such limited correlations used
in alI pressure drop models because even a superficial study of the caIculation processes can reveal them. These are
responsible for the general observation that any correlation, if used outside its original ranges, will produce increased
caIculation errors. Mechanistic models, on the contrary, may exhibit much wider ranges of applicability, partly because
they almost completely eliminate the use of empirical correlations.

Calculation Direction

It is a widely observed fact that the accuracy of any multiphase flow model depends also on the direction the
caIculations are made as compared to the flow direction. If pressure drop is caIculated along the flow direction (wellhead
pressure is caIculated from a known bottomhole pressure), the error between measured and caIculated pressure drops is
always lower as compared to caIculations done in the other direction. Therefore, any multiphase flow model performs
better if the solution of the gradient equation is done starting from the bottom of the well; i.e. along the flow direction.
The explanation of this behavior lies in the great difference of caIculation accuracies valid in single-phase and multiphase
flows. As discussed before, single-phase liquid flow is relatively easy to describe, and pressure drop caIculations are quite
accurate. ln comparison, multiphase flow is much more complex and greater caIculation errors can be expected.

Since single-phase flow usually occurs at or above the well bottom, caIculations started from there involve the
determination of single-phase pressure drops first. Moving higher up the hole, at a depth corresponding to the local
------------------- --------------------------~---------------~-------------

PRODUCTION ENGINEERING FUNDAMENTALS I 117

bubblepoint, pressure flow type changes to a multiphase one and calculation accuracy rapidly deteriorates. Obviously,
the cumulative calculation error will depend on the reliability of the bubblepoint pressure value and the reiative depth
occupied by single-phase flow, as first shown by Gregory et a!. [70]. If, as is generally the case, a sufficient tubing length
is occupied by single-phase flow, the accuracy of the calculated total pressure drop will clearly be determined by the
excellent accuracy of single-phase flow description. Thus, pressure drop calculations done in the flow direction are
relatively more accurate.

If, on the contrary, calculations are started at the wellhead, initial calculation steps involve much greater calculation
errors due to the existence of multiphase flow at and below that point. Since these errors are cumulative and add up in
successive calculation steps, the greater accuracy achieved in the single-phase region below the depth corresponding
to the bubblepoint pressure does not prevent an excessive final total error.

Although the error in pressure drop calculations varies with the direction of calculations, the relative ranking of
the different calculation models is not affected by this phenomenon. This fact can be observed from the results of
those investigators who compared calculation accuracies for both cases, like Gregory et a!. [70], Chokshi et a!. [62],
andKaya et a!. [34J

Special Conditions

Most of the vertical pressure drop calculation models were developed for average oilfield fluids. This is why special
conditions like emulsions, non-Newtonian flow behavior, excessive scale or wax deposition on the tubing wall, etc. can
pose severe problems. Predictions in such cases could be doubtful.

Special treatment is required for two commonly occurring cases: flow in inclined wells and production through
casing-tubing annuli. Although there exist some approximate solutions to deal with such wells, calculation errors can
be high unless a specifically developed flow model is used.

Errors in Measured Data

Empirical and mechanistic pressure drop calculation models alike were developed based on accurately and reliably
measured flow parameters. This is why errors in the required input data can seriously affect the prediction accuracy of
any pressure drop calculation model. Evaluation of any flow model, therefore, must be made by using reliable measured
values of the actual flow parameters. Special care has to be taken to ensure that measurements in the well are carried out
under steady-state conditions. Also, the accuracy of gas flow rates has to be checked because much higher inaccuracies
can be expected in gas volume measurements than in liquid rates. In summary, pressure drop calculation errors stemming
trom poor input data will add to those of the calculation model itself and the model's true accuracy may be masked.

2.5.3.6.3 Results of published evaluations. Over the years, a great number of authors have investigated the accuracy
and applicability of the various pressure drop correlations for different conditions. Table 2-9 contains a compilation of
the results of such studies published so faroDue to insufficient data in some of the original papers, the table contains
the following statistical parameters only:
N
Ld.
. 1

d=~ average errar 2.247


N

2.248
cr = 1=1
.~ ~(di N-l
- d)2
standard deviation of errors

average abso/ute error 2.249

-ere N is the total number of cases considered.


118 I GAS Lwr MANUAL

Calculation error is defined as the reiative errar in calculated pressu~e drops, by


.ó..pcale - .ó..pmeas
x tOO 2.250
.ó..Pmeas

where i1Pcalc = calculated pressure drop in weIl

i1Pmeas = measured pressure drop for same conditions

Table 2-9 shows the previous statisticaI parameters as given by the various investigatars of the different
correlations. Negative average errors indicate calculated pressure drops Iower than measured ones, positive ones show

27.0
24.2
10.8
1042.4
1478
596
25.9
15.2
19.1
11.2892
65
10
-7.3
30.0
221.9
20.5
26.1
38.0
17.1
13.6
15.0
25.9
15.7
-1.2
1697
1079
1711
195.1
323
10.9
14.5
14.7
14.317.6
29.2
138021.6
21.0
-3.9
-0.1
14.7
11.4
010.1
14.5
18.6
-17.8
2.4
1710
14.4
12.9
48
49.8
8.8
63
72
62
61
78
-16.2
-4.3
3.8
8.0
6.6
19.9
1.0
14.2
-0.4
27.0
72627.2
27.3
0.1
-5.2
11.9
16.5
1.9
-0.82.6
1413 35
0.0
8.6
14844
148 -3.9
1.1
9.6
-8.5
8.9
750
44
7.6
3.6
3.3
1.3
.6
28.721
16.3
32.9
21.1
212
2.1
130
47
35.7
18.3
1423 7.8
9.9
90
30.5
46.7
39.5
56.4
42.6
29.1
26.6
-2.9
195.7
1703
1050
1712
53.8
27.5
107.3
0.7 45
22.9
1026
-25.7
-2.1
13.6
-10.5
11.7
23.4
1008
23.0
108.3 2.1 -10.5
12.3
12.1
-11.7
-8.9
3.0-11.8
15.5
19.4
4.4
9.7
6868.1
414
8.6
12.3
36.4
22.6
637
19.8
920
38
15.3
734
7.5
18.2
76
59
69
77
34
73
7075
31.3
22.8
21.7
-4.0
728
-3.6
-0.1
18.1
16.7
13.9
13.5
8.2
3.5
6.7
23.3
32.4
7.4
26.7
13.3
16.2
71,74685 -16.1
15.5 5.8
15.4 2.9d,d423.8
021
0.2
20.9
17.1
-5.1
14.619.6
427
1249
38.1
22.0
837
17.5
12.2
18.0
-19.2
14.4
27.4
104
323
32.7
1401
-8.2
1.5
726 990
14.0
50.2
36.4
6.9
10
19.4
-5.4
29.4
34.7 -1.1
32.7
.4
-2.5
21.6
48 9.6
6.4
34.6
0.5 -6.3
20.0
1710 323
25.4
31.7
27.2
23.0
-1.7
77 36.1
43.9
19.0 42.8
17.8
27.6
5.5
726 32.5
323
24.8
25.3 4.4
3.7
28.2 1.3
8.8
0.9
1.4
5.1
10 15.3
-3.3
130 References
30.9 20.4
21.6
21.9
30.5
28.9 11.8
2.5
7.6
13.6
-3.7
212 25.6
16.8
1.5
1711 14.7
13.1
-5.2 13.9
10.3
-2.3
21 1407
N C5

Table 2-9 Accuracies ofvertical multiphase pressure drop calculation models according to various literature sources.
PRODUCTION ENGINEERING FUNDAMENTALS I 119

that the particular correlation tends to overestimate actual pressure drops. The rows in the table refer to the different
vertical pressure drop calculation models and the columns correspond to the different published comparisons, headed
by their reference numbers.

In each column, i.e. for each investigation, the model found to give the least amount of error is indicated by a solid
border around the corresponding cell. As seen, the Orkiszewski [37J correlation collected the most number of first
places, five, followed by Hagedorn-Brown's [46J four. A more thorough study, however, reveals that Orkiszewski's wins
all come from investigations involving relatively low numbers of cases whereby the model shows higher errors in the
investigations involving greater numbers of cases. Therefore, the Orkiszewki correlation clearly cannot be declared as
an overall best achiever.

The Hagedorn-Brown [46J correlation, on the other hand, suffers from a basic bias that comes up with all
investigations using the Tulsa University Fluid Flow Projects (TUFFP) multiphase flow data bank [36]. The reason is
that the TUFFP data bank includes a great number of data originally taken from Hagedorn's doctorate thesis with
accurately measured pressure drop values that are definite favorites with other researchers as well. Consequentl)',
investigators using this data bank, perhaps unknowingly, give an unfair advantage to the Hagedorn-Brown mode!.
making their findings severely biased. This is the case with the results of Lawson-Brill [71 J whose 726 sets of well data
contained 346 sets taken from Hagedorn's thesis. Ansari et aI. [61], realizing this fact, have even published two sets of
statistical parameters in their evaluation (see the two columns labeled 61 in Table 2-9: one for all data, the other for
data excluding those of Hagedorn's. Based on these considerations, the Hagedorn-Brown model achieves only twO best
places and thus cannot be named a winner either.

Let us now find the extreme values of calculation error and investigate the absolute errors, where available. The
worst errors in the table (more than 100%) are given for the Poettmann-Carpenter and the Baxendell-Thomas
correlations, found by Lawson-Brill [71]. It is interesting to see that the next highest error in the whole table is around
38% only. The smallest errors, on the other hand, were all found by Cornish (Cornish 1.4%, Orkiszewski 3.7%) but
the low number of experimental data (only 10 sets) makes his results at least unreliable. The next best error was
achieved by the Beggs-Brill correlation, found as 6.7% for 414 data sets by Aggour [72].

If the ranges of observed errors are investigated for the two basic groups of pressure drop calculation models-the
empirical and the mechanistic ones-after exclusion of the unrealistic or unreliable values detailed previously, we get:

mino abs. error max. abs. error

Empirical models 6.7% 34.6%

Mechanistic models 7.6% 38%

The previous numbers indicate that the accuracy of mechanistic models still does not substantially exceed that of
me empirical ones. This observation is in great contradiction to the claims of several authors who vindicate substantially
better accuracies to mechanistic models. The older empirical correlations, therefore, cannot be ruled out when seeking
me most accurate vertical pressure drop prediction modeI.

Since the previous discussions have not resulted in a clear and reliable ranking of the available vertical multiphase
pressure drop calculation models, let us try a different approach. It is a well-known fact that statistical evaluations
become more reliable when the number of investigated cases is increased. Because of this, we should focus on the
:índings of those investigators who dealt with the greatest number of measured pressure drop data.

All the following three investigations utilized the very comprehensive TUFFP multiphase flow data bank: Ansari
eI aI. [61] used 755 sets of vertical well data (see the second column, 61) Chokshi et aI. [62J had 1712 vertical and
:nclined wells, and Kaya et aI. [34J used 1429 vertical wells. The numbers N in the table only show the number of
;-Kcessful calculations; the total number can be more, like in this case. Their findings on the accuracy of the Hagedorn-
:::~own, Duns-Ros, Orkiszewski, Hasan-Kabir, Ansari et aI., Chokshi et aI., and Kaya et aI. pressure drop calculation
-Jdels are shown in Table 2-10. From this table, it is clearly seen that the average absolute errors found for the different
~-essure drop prediction models by each investigator are quite similar. Based on the individual errors and the numbers
- cases involved, composite errors valid for the total number of investigated cases could be calculated for each modeI.
120 I GAS LIFr MANUAL

The results are shown in the bottom row of the table. The average errors thus received integrate the results of reliab ~
investigations and, therefore, can be considered as the most comprehensive indicators of the accuracy of verti c
multiphase pressure drop ca\culation models ..

Investigation1711
1413
27.3%
27.2%
1478
1407
25.7%
14.4%
11.4%
1697
1423
szewski
13.1%
12.3%
14.7%
11.4%
1380
728 734
38182446
17101712
27.2%
16.8%
14.7%
10.3%
10.1%
750
686
596
3638
2074
3124
1703
10.3%
38.0%
3870
19.8%
16.0%
13.6%
21.6%
15.0%
Hasan-Ndadada
Chokshi
Ansari
Orki-
et
et
Kabir
aI.
Ros
aI.
19.6%
Duns-
Kaya Hagedorn 15.5%
1249

labia 2-10 Accuracies of selected vertical pressure drop calculation models.

Although the total combined number of cases varies from 3870 to 1407, the ca\culated overall errors given in
Table 2-10 present a reliable comparison of the selected models. Basically,two conclusions can be drawn from these data:

(a) none of the models can be declared a clear winner

(b) mechanistic models do not seem to substantially improve ca\culation accuracy (as observed earlier)

2.5.3.6.4 Selection of an optimum modelo In preceding sections, based on the findings of previously published
evaluations, establishment of the ca\culation accuracies of the different pressure drop models was attempted. For this
reason, ali available data on ca\culation accuracies were collected in Table 2-9, which contains the most important
statistical parameters of the different investigations. As discussed previously, none of the ca\culation models was found
to be the most accurate one. Another, perhaps more important, conclusion is that the accuracy of the same model
substantially varies from one investigation to the other, i.e. with changes in the parameter ranges of measurement data
used for verification. Therefore, no multiphase pressure drop ca\culation model can achieve the same accuracy for ali
the possible conditions encountered in practice.

It follows from the previous that it does not make sense to ask which pressure drop model has the highest accuracy
or which is the best. There is no overall best method, and all efforts to find it are sure to fail. Nevertheless, production
engineers, in order to increase the accuracy and reliability of their designs and analyses, need to use the model giving
the least ca\culation error for the conditions of the problem at hand. This model can and must be found and it may
then be considered as the optimal one for the given case. The field engineer's philosophy on vertical pressure drop
ca\culations must always be based on this consideration.

Now that the right approach toward multiphase vertical pressure drop prediction models has been defined, the
main requirements for finding the proper one are discussed. First, the complex nature of pressure drop ca\culation
methods practically presupposes the use of high-speed digital computers. For conducting any analysis on the
application of the various models, one then needs appropriate computer programs. These can be found in several
software packages abundant\y available on the market.

In addition to the computer programs, a sufficiently large number of experimental pressure drop data has to be
established. These have to be taken from conditions similar to those of the future application, i.e. from the same field and
possibly the same wells. Care should be taken to ensure the reliability and accuracy of the measured data, because the
use of poor quality data can considerably distort the accuracy of apressure drop ca\culation model. If a sufficient number
of reasonably accurate data is used, the selection of the optimum pressure drop prediction method is straightforward.
First, pressure drops for ali data sets using ali ca\culation models investigated are ca\culated. Then, statistical parameters
of the ca\culation errors obtained for the various models are determined. Finally, the ca\culation model with the best
statistical parameters is chosen. This model is then considered the optimum one for the conditions at hand and must be
used in ali production engineering ca\culations involving the determination of vertical pressure drops.
PRODUCTION ENGINEERING FUNDAMENTALS I 121

2.5.3.6.5 Conclusions. By presenting an analysis of the findings of all previously published evaluations, Section 2.5. 3.6
provided the required insight and proper attitude to petroleum engineers who face the problem of predicting multiphase
pressure drops in oi! wells. Because of the great importance of the topic, it is useful to list the basic conclusions:

1. None of the available vertical multiphase pressure drop calculation models is generally applicable because
prediction errors considerably vary in different ranges of flow parameters.

2. There is no overall best calculation method, and all efforts to find one are bound to fail.

3. In spite of the claims found in the literature, the introduction of mechanistic models did not deliver a
breakthrough yetl because their accuracy does not substantially exceed that of the empirical models.

2.5.3.7 Gradient curves. Previous sections have shown that calculating pressure traverses in oil wells is a
cumbersome and tedious taskl and one usually has to rely on the use of computer programs. Before the advent of
computers, this was rarely feasible in the oilfield, and the practicing engineer had to find some simple solution to the
problem. This is the reason why pre-calculated pressure traverse sheets were prepared and used in engineering
calculations, eliminating the difficulty and time demand of multiphase flow calculations. Such sheets contain several
pressure traverses for selected combinations of flow parameters like pipe sizel liquid ratel etc. They are conventionally
called gradient curves I but a more precise definition would be multiphase pressure traverse curve sheets.

2.5.3.7.1 Gilbert's gradient curves. The first gradient curves were introduced by Gilbert [79J who in 1954 proposed
an alternative to the calculation of multiphase flow pressure drops. His objective was to provide the field engineer with
a simple tool that could be used to find wellhead or bottomhole pressures in flowing oil wells with a reasonable accuracy.
The model he proposed did not require detailed calculations, only the use of ready-made gradient curve sheets.

Based on field measurements and some theoretical background, Gilbert prepared several sheets with families of
arrves depicting the pressure distribution in vertical oil wells. The coordinate system he employed was flowing pressure
""5. well depthl and each family of curves was valid for a given liquid rate, tubing diameter, and fluid parameters. The
individual curves on any sheet have GLR as a parameter.

One gradient curve sheet from Gilbert's collection is shown in Figure 2-44, valid for a tubing 10 of 1.9 in. and an oil
rate of 600 bpd. The two axes correspond to flowing pressure and vertical depth, their use allows for an easily
understandable treatment of the flowing
pressure traverse. The family of curves shown 2000 2500
in the sheet has surface or production GLR as o
a parameter. At a GLR of zero, the pressure
aaverse is a straight line because single-phase
hydrostatic and mctional losse~ are present
only. Those being linear with pipe length, o 5
their sum must aIso be a linear function of
.:!l
<=
","éll depth. The other curves with increasing '"
'6'"
GlRs are no longer straight lines but show
:ncreased curvatures. til 10
-.::

.t
o
The sheet contains two sets of curves 8-
md set R should be shifted vertically on the .;
~th axis to the zero depth. The reason is fi15
1Il

.t::
~t curves in set R, if plotted among set A 'lS.
'"
::.l:-ves, would cross the original ones. The CI

:-••"0 sets of curves are divided by a curve with


20
::..-mow placed to its GLR. This curve is a
7-:." important one because it represents the
- -imum flowing pressures for the given set
. = Jnditions. For the given liquid rate and 20 25
- =:;g size, the GLR corresponding to this
. - -e is considered the optimum GLR
Fig. 2-44 Sample gradient curve sheet from Gilbert's collection. [79]

Você também pode gostar