Beruflich Dokumente
Kultur Dokumente
497-510
R.G.D. Teixeira et al., published by IFP Energies nouvelles, 2014
DOI: 10.2516/ogst/2013191
IFP Energies nouvelles International Conference
Rencontres Scientiques d'IFP Energies nouvelles
NEXTLAB 2014 - Advances in Innovative Experimental Methodology or Simulation Tools used
to Create, Test, Control and Analyse Systems, Materials and Molecules
NEXTLAB 2014 - Innover dans le domaine de la mthodologie exprimentale et des outils de simulation
pour crer, tester, contrler et analyser des systmes, matriaux et molcules
* Corresponding author
Abstract Two-phase flow in pipes occurs frequently in refineries, oil and gas production facilities and
petrochemical units. The accurate design of such processing plants requires that numerical algorithms be
combined with suitable models for predicting expected pressure drops. In performing such calculations,
pressure gradients may be obtained from empirical correlations such as Beggs and Brill, and they must be
integrated over the total length of the pipe segment, simultaneously with the enthalpy-gradient equation
when the temperature profile is unknown. This paper proposes that the set of differential and algebraic
equations involved should be solved as a Differential Algebraic Equations (DAE) System, which poses a
more CPU-efficient alternative to the marching algorithm employed by most related work. Demon-
strating the use of specific regularization functions in preventing convergence failure in calculations due
to discontinuities inherent to such empirical correlations is also a key feature of this study. The developed
numerical techniques are then employed to examine the sensitivity to heat-transfer parameters of the
results obtained for a typical refinery two-phase flow design problem.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0),
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
498 Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 70 (2015), No. 3
Oil and gas production problems have motivated techniques [9-11]. In particular, the correlation of Beggs
most of the work done on the subject of two-phase flow and Brill [5] stands out for being able to predict pressure
in pipes so far, including development of pressure gradi- drops for angles other than vertical upward flow. Inter-
ent correlations. Bibliographical references which prior- estingly enough, the pressure drops calculated by this
itize this type of problem are abundant, and can be easily method in recent studies have been found to show
found [1, 4]. On the other hand, process engineers acceptable agreement with corresponding CFD results
involved in simulation and design of refineries and petro- [12, 13].
chemical process units are faced with a lack of studies on The Beggs and Brill correlation was adopted for liquid
two-phase flow in pipes when it comes to petroleum frac- holdup and pressure gradient calculations in this work
tions and petroleum downstream industry. As a result, for the reasons outlined above, and also because it shows
correlations originally developed for oil and gas two- the discontinuities that can lead to convergence failure,
phase flow have also been widely adopted in downstream which this study intends to demonstrate how to avoid.
design calculations. A brief summary of this method is presented in this sec-
Given the above scenario, all calculations performed tion, as some of its equations and coefficients need to be
in this work consider a typical refinery vapor-liquid mix- referenced in subsequent discussions.
ture. Obtained results are analyzed in a first attempt to
determine which heat-transfer considerations are critical
1.1 The Correlation of Beggs and Brill
(and which are not) in this petroleum downstream indus-
try design problem. In spite of this, it must be empha- The velocities at which both phases would flow individ-
sized that the numerical techniques presented here are ually through the pipe cross section (superficial veloci-
relevant for oil and gas production calculations as well. ties) can be calculated from the internal diameter of
the pipe (d) and their volumetric flow rates (qL and qV ):
qL
1 PRESSURE-GRADIENT CORRELATIONS vsL 1a
At
The first correlations ever developed for prediction of
qV
pressure gradients in two-phase flow in pipes are empir- vsV 1b
At
ical in nature, often based on dimensionless groups and
data from laboratory test facilities (with a few using field
In which At pd 2 =4 is the pipe cross-sectional area.
data) [5-7].
A mixture velocity (vm ) and corresponding Froude
It has long been recognized that the accuracy of
number (N Fr ) are given by:
pressure-gradient predictions cannot be improved with-
out the introduction of basic physical mechanisms, vm vsL vsV 2
which is the main limiting factor of the empirical meth-
ods [1]. For this reason, over the past decades, efforts vm 2
have been directed to the development of phenomeno- N Fr 3
gd
logical or mechanistic correlations, i.e., correlations in
which fundamental laws are used to model the most If the liquid and vapor phases both travelled at the
important flow phenomena and less empiricism is same velocity (no-slippage), the liquid holdup along
required [3, 8]. the entire pipeline would equal the input liquid content.
Empirical correlations were readily adopted in the Therefore, no-slip liquid holdup (kL ) can be calculated
petroleum industry, as they were the first to be con- from:
ceived. Although greater accuracy is expected when
using a theoretical correlation instead of an empirical qL
kL 4
one, selecting a mechanistic method for use in design cal- qL qV
culations is not a trivial task, specially when it comes to
refining or petrochemical processes, for some of these The flow pattern which would exist if the pipe were
methods only apply to certain pipe inclinations [8] while horizontal can be determined by the location of the pair
others [3] have not been validated enough to persuade (kL , N Fr ) in the correlation flow pattern map. The chart
engineers and companies into updating long-experienced area is divided in four regions each corresponding to
design practices. It follows that empirical methods still one of four possible flow regimes: Segregated, Transi-
enjoy widespread use in industry and also for compari- tion, Intermittent and Distributed by the following
son purposes in the development of new predictive transition boundaries:
500 Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 70 (2015), No. 3
L1 316 k0:302
L 5a This correction factor is calculated from the liquid
velocity number (N Lv ), the pipe inclination angle from
L2 0:0009252 k2:4684
L 5b horizontal (h; kL and N Fr :
s
L3 0:10 k1:4516
L 5c N Lv vsL
4 qL
8a
grL
L4 0:50 k6:738
L 5d
w 1:0 C sin 1:8h 0:333 sin3 1:8h 8b
Thus, each flow pattern is associated with a set of
inequalities, as shown in Table 1.
C 1:0 kL ln e kfL N gLv N hFr ; C 0 8c
In two-phase flow in pipes, liquid and vapor phases
travelling at different velocities (slippage) result in
in-situ liquid volume fractions which differ from the ori- In which qL and rL are the density and the surface ten-
ginal liquid input content. Liquid holdup is defined as sion of the liquid phase. The coefficients e, f , g and h are
the fraction of the volume of a segment of the pipeline given in Table 3 for each horizontal-flow regime (except
which is occupied by liquid. Transition).
The holdup which would exist at the given conditions When the horizontal-flow pattern is found to be Tran-
if the pipe were horizontal can be determined from: sition, the liquid holdup should be interpolated between
the values calculated for the Segregated (HLSeg ) and
a kbL Intermittent (HLInt ) regimes:
HL0 6
N cFr
L3 N Fr
A 9a
with the restriction that: HL0 kL . L 3 L2
The coefficients a, b and c are given in Table 2 for all
flow patterns, except for Transition, which is discussed HL A HLSeg 1 A HLInt 9b
afterwards.
The horizontal-position liquid holdup must be cor- A normalizing friction factor fn can be obtained from
rected for the effect of pipe inclination by a multiplica- the smooth pipe curve on a Moody diagram or from:
tive factor w:
2
Re
fn 2 log 10
HL w HL0 7 4:5223 log Re 3:8215
TABLE 1
Flow regime limits in Beggs and Brill horizontal-flow-pattern map TABLE 2
Empirical coefficients for horizontal-flow liquid holdup
Segregated kL < 0.01 and NFr <L1 or kL 0.01 and NFr < L2
a b c
Transition kL 0.01 and L2 NFr L3
Segregated 0.980 0.4846 0.0868
Intermittent 0.01 kL < 0.4 and L3 < NFr L1 or
kL 0.4 and L3 < NFr L4 Intermittent 0.845 0.5351 0.0173
Distributed kL < 0.4 and NFr L1 or kL 0.4 and NFr > L4 Distributed 1.065 0.5824 0.0609
TABLE 3
Empirical coefficients for calculation of factor w
e f g h
The Reynolds number in Equation (10) is given by: Payne et al. [14] proposed two well accepted modifica-
tions to improve liquid holdup and pressure-drop pre-
qn vm d dictions by this correlation. They found that the Beggs
Re 11
ln and Brill method overpredicted liquid holdup in uphill
and downhill flow and recommended correction factors
qn qL kL 1 kL qV 12 of 0.924 and 0.685, respectively. They also recommended
that the normalizing friction factor fn be obtained from a
ln lL kL 1 kL lV 13 Moody diagram, as the use of the smooth-pipe equation
results in underpredicted values of two-phase friction
The two-phase friction factor, ftp , is determined from factors. All calculations performed in this study include
the relation: these modifications.
ftp fn exp S 14
subscript i) have already been determined and that Marching algorithm steps for one calculation increment.
downstream pressure and temperature (subscript i 1)
are to be calculated. The PVT-calculation (Pressure/
Volume/Temperature) box involves phase-equilibria cal-
culations as well as those of physical and thermody-
namic properties (densities, viscosities, surface tension equations can be expressed as functions of them. Thus,
and enthalpies). In short, this strategy, which is very sim- a minimum state vector would be:
ilar to a forward Euler integration, implies solving for
the unknown variables using two separate, nested loops X P h T 26
with the algebraic thermodynamic and heat-transfer
constraints being solved in each iteration of the inner Three equations must be solved along the pipe for
loop, rather than numerically solving all the equations these variables:
in a true simultaneous fashion.
It is one goal of this study to emphasize and demon- dP
strate that the marching algorithm is not the most effi- f 1 X 27
dx
cient technique available for coupling the relevant
differential equations (pressure gradient and enthalpy dh
gradient). These equations and those required for PVT f2 X 28
dx
and heat-transfer calculations all form a Differential
Algebraic Equations (DAE) system, and they must be hX g1 X 29
represented as such in order to be simultaneously solved
by more efficient, suitable numerical methods. where f1 is a two-phase pressure-gradient correlation (in
Pressure, temperature and enthalpy can be considered this study, Eq. 22, i.e., the Beggs and Brill correlation),
the main variables of this problem, since all relevant f2 is the right-hand side of Equation (23) and g1 is a
R.G.D. Teixeira et al. / Two-Phase Flow in Pipes: Numerical Improvements and Qualitative 503
Analysis for a Refining Process
n-Pentane 4.03
Equations (27, 28) and (30) can be combined by use of
a mass matrix M [16]: n-Heptane 13.55
M X_ F X
2-Methyl-Heptane 16.98
31
Cyclohexane 17.28
in which: 2 3 Benzene 3.72
1 0 0
6 7
M 40 1 05 32 Toluene 25.26
0 0 0 Ethylbenzene 4.57
p-Xylene 10.96
2 3
dP=dx o-Xylene 3.65
_X 6
4 dh=dx 5
7
33
dT =dx
at constant pressure, which was done using equations
and: and constants given in [19]. Liquid-phase density, viscos-
2 3 ities of liquid and vapor phases and surface tensions were
f 1 X
6 7 all calculated using methods presented in [18] and [20]
F X 4 f2 X 5 34
for petroleum fractions of defined composition.
f 3 X In two-phase flow, radial convective heat transfer
between the flowing fluids and the internal wall of the
pipe depends on flow pattern, which makes it more com-
MATLABs ODE15S built-in stiff solver was used in plex than for single-phase flow. But according to [4], in
this work for solving the DAE System (Eq. 31). This most two-phase flow problems, flow is turbulent enough
approach is expected to prove more efficient in that all dif- that convective heat transfer becomes significant and
ferential equations and algebraic constraints will be simul- temperature differences between the flowing mixture
taneously processed, instead of sequentially being iterated and the wall are negligible. This last assumption was
for in a nested structure. CPU-cost gains are typical bene- adopted in all reported calculations, which corresponds
fits of not separately solving algebraic equations in every to assigning a zero value to the internal convection resis-
step, as verified by [17] in a flash-drum DAE problem. tance in the bracketed term in Equation (25).
The convective heat transfer coefficient between the
outer surface of the pipe and the surroundings must be
4 METHODS calculated from a Nusselt number, which in turn should
be determined from equations for natural or forced con-
Unless otherwise indicated, all reported calculations vection, whichever be the case [15]. The correlations of
were performed considering 115 000 kg/h of naphtha Churchill and Chu [21] and Churchill and Bernstein
entering a horizontal, 30-meter long carbon steel pipe [22] for natural and forced convective heat transfer were
segment (k = 60.5 W/m/K, d = 202, 7 mm and used for estimating ho under stable and unstable
D = 219,1 mm) at 168C and 5.0 kgf/cm2. A value of (36 km/h wind velocity) atmospheric conditions, respec-
27C was assumed for ambient air temperature. These tively. Occurrence of wind was considered by default.
latter values represent typical refinery design conditions.
Table 4 gives the assumed naphtha composition,
which is very similar to that of the gas chromatography 5 RESULTS AND DISCUSSION
example presented in [18]. The Peng-Robinson equation
of state was used in calculations of Vapor-Liquid Equi- The two-phase flow problem must be solved both by the
librium (VLE), vapor-phase density and residual enthal- marching algorithm and the DAE approach if these
pies. Enthalpy calculations by the latter method also methods are to be compared in terms of results and
require determination of ideal gas state heat capacities CPU requirements.
504 Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 70 (2015), No. 3
Temperature (k)
0.190
Liquid holdup
0.185
4.8 440
0.180
0.175
4.7 Temperature 439 0.170
0.165
4.6 438 0.160
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Distance from inlet (m) Distance from inlet (m)
Figure 2 Figure 3
Marching algorithm results along the pipe (m 10 and Liquid holdup discontinuity due to change in flow regime.
m 100) for pressure and temperature.
Seeing as the marching algorithm was expectedly b, c, e, f , g and h to be used when calculating liquid
found to be somewhat intensive with respect to compu- holdup (Tab. 2, 3).
tational effort its execution took 206 seconds when Investigation of the results previously calculated by
the pipe was subdivided into 100 segments it was ini- the marching algorithm shows that from x 15 m to
tially sought to find a reasonable value of m (number x 18 m, kL changes from 0:0732 to 0:0723, causing L1
of pipe increments) representing a compromise between to vary from 143:47 to 142:94. N Fr changes from
calculated results and CPU cost, in order to ensure that 142:87 to 145:89, thus becoming greater than L1 . Conse-
comparisons of both methods were not biased by exces- quently, according to Table 1, the horizontal-flow
sive pipe subdivision here. regime changes from Intermittent to Distributed.
Changing m from 100 to 10 drastically reduced exe- Figure 3 shows the resulting abrupt change of liquid
cution time (to 25 s) at the cost of very slight changes holdup.
in the pressure at the outlet of the pipe (0.0033%)
and other calculated results, as shown in Figure 2.
Further reduction of m to 5 led to a smaller relative 5.2 Usage of Regularization Functions
change of execution time (to 14 s) while the difference
Large discontinuities in empirical models are known to
of downstream pressure to that of m 100, although
cause difficulties to integrator solvers indeed, and the
still small, increased approximately five times
use of specific interpolation functions for linking two
(to 0.0153%).
adjacent discontinuous domains in order to come
around this problem has been investigated in recent
studies [23, 24]. The regularization function technique
5.1 Discontinuities and Lack of Convergence
used in [25] when approximating discontinuous jumps
Attempting to solve the same problem by the DAE in consistent initializations of DAE systems has also
method failed as convergence was not achieved at a dis- been shown by [24] to successfully replace model discon-
tance close to 15 m from the pipe inlet. tinuities while even reducing computational effort. In
The results calculated from some two-phase flow cor- [24], discontinuities such as:
relations show rather severe discontinuities which often
lead to lack of convergence in simultaneous pressure
z1 X; pX pmax
and temperature calculations, as detailed in [2]. In the zX 35
Beggs and Brill method, this occurs when flow condi- z2 X; pX < pmax
tions change just enough that a different flow regime gets
predicted, which causes different values of coefficients a, are replaced by a single algebraic equation:
R.G.D. Teixeira et al. / Two-Phase Flow in Pipes: Numerical Improvements and Qualitative 505
Analysis for a Refining Process
1 tanh arg=d
where garg; d is a continuous regularization function garg; d 38
2
evaluated at scalar argument arg and parameter d. In
order for Equation (36) to closely reproduce 35, g must Figure 4 illustrates how the magnitude of n can be
be such that: adjusted through tuning of d when using regularization
function (38).
1; arg > n
garg; d 37
0; arg < n
5.3 Regularization Functions and Discontinuities
where 0 < n << 1. When n arg n, g assumes an at Flow-Regime Boundaries
intermediate value in the 0; 1 interval. n is a function
The horizontal-flow regime is a function of kL and N Fr ,
of user-supplied parameter d, which must be adjusted
as can be seen from Table 1. Any flow-regime inequali-
to each particular application so that the interval
ties can be easily combined with Equation (38) into a
n; n (inside which Eq. 36 does not exactly reproduce
continuous function of these variables which yields a
Eq. 35) is small enough that deviations between both
value of 1 when the given regime is to be predicted,
equations can be neglected.
and 0 when it is not. Such a function for Segregated
This scheme ensures a continuous and smooth transi-
flow-regime would be:
tion from z1 X to z2 X while closely resembling the
0.2 L3 N Fr
AkL ; N Fr 1 1 TrankL ; N Fr 41
L3 L2
0
In Equation (40), HL1 and HL2 are given by
-5 -4 -3 -2 -1 0 1 2 3 4 5 Equation (7), and subscript 2 always denotes Intermit-
Arg x 10-3 tent flow-regime holdup calculation. Subscript 1, in turn,
denotes Segregated holdup calculation when Transition
Figure 4 flow-regime gets predicted, and whichever flow regime
Tuning of regularization function d . is detected when it is not Transition.
TABLE 5
Flow-regimes detection inequalities in terms of regularization function g
Segregated Seg(kL; NFr) = g(0.01 kL; d) g(L1 NFr; d) + g(kL 0.01; d) g(L2 NFr; d)
Intermittent Int(kL; NFr) = g(kL 0.01; d) g(0.4 kL; d) g(NFr L3; d) g(L1 NFr; d) + g(kL 0.4; d) g(NFr L3; d) g(L4 NFr; d)
Distributed Dist(kL; NFr) = g(0.4 kL; d) g(NFr L1; d) + g(kL 0.4; d) g(NFr L4; d)
506 Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 70 (2015), No. 3
The empirical coefficient a that is used in calculation angles in degrees and calculated values of L1 , L2 , L3
of HL1 should also be determined from its correspond- and L4 for kL close to 0:07.
ing continuous function: Figure 5 shows empirical coefficients a1 and b1 calcu-
lated by this method for kL 0:073 (L1 143:35, also
a1 kL ; N Fr 0:98 Seg kL ; N Fr TrankL ; N Fr shown in this figure as a continuous function) and N Fr
between 142 and 145, where the horizontal-flow regime
0:845 IntkL ; N Fr 1:065 DistkL ; N Fr 42
changes from Intermittent to Distributed, as previously
discussed. It can be clearly seen that even though the
and for calculation of HL2:
a2 kL ; N Fr 1 0:155 TrankL ; N Fr 43
1.10 0.61
Similar equations for all empirical coefficients of the
Beggs and Brill liquid holdup correlation are given in L1
1.04 a1 0.59
Table 6, where checking for downhill flow is also done b1
through the regularization function approach, so that
0.98 0.57
discontinuities of e, f , g and h with respect to h are dealt
a1
b1
with as well. This is accomplished by defining a function
which yields 1 for downhill flow, and 0 for horizontal 0.92 0.55
Dowh g0:01 h; d 44
0.80 0.51
142.0 142.5 143.0 143.5 144.0 144.5 145.0
with h given in degrees. The 0.01 offset in Equation (44)
ensures that Dow0 is always zero, and not 0:5 (because NFr
g0; d 0:5).
A value of d 107 was used throughout this study, Figure 5
as its associated value of n was shown to be much smaller Discontinuities in empirical coefficients a1 and b1 captured
than 103 (Fig. 4), which is negligible enough next to by the regularization function approach.
TABLE 6
Empirical coefficients expressed as continuous functions of kL, NFr and h
a1 (kL; NFr) = 0.98 [Seg(kL; NFr) + Tran(kL; NFr)] + 0.845 Int(kL; NFr) + 1.065 Dist(kL; NFr)
b1(kL; NFr) = 0.4846 [Seg(kL; NFr) + Tran(kL; NFr)] + 0.5351 Int(kL; NFr) + 0.5824 Dist(kL; NFr)
c1(kL; NFr) = 0.0868 [Seg(kL; NFr) + Tran(kL; NFr)] + 0.0173 Int(kL; NFr) + 0.0609 Dist(kL; NFr)
e1(kL; NFr; h) = 4.7 Dow(h) + [1 Dow(h)] {0.011 [Seg(kL; NFr) + Tran(kL; NFr)] + 2.96 Int(kL; NFr) + Dist(kL; NFr)}
f1(kL; NFr; h) = 0.3692 Dow(h) + [1 Dow(h)] { 3.768 [Seg(kL; NFr) + Tran(kL; NFr)] + 0.305 Int(kL; NFr)}
g1(kL; NFr; h) = 0.1244 Dow(h) + [1 Dow(h)] {3.539 [Seg(kL; NFr) + Tran(kL; NFr)] 0.4473 Int(kL; NFr)}
h1(kL; NFr; h) = 0.5056 Dow(h) + [1 Dow(h)] { 1.614 [Seg(kL; NFr) + Tran(kL; NFr)] + 0.0978 Int(kL; NFr)}
442 x 105
5.0 2.415 22.6
Pressure Original correlation Original correlation
Reg. fct. (marching alg.) 2.410 Reg. fct. (marching alg.) 22.4
2.405 22.2
Temperaure (k)
Enthalpy
2.400 22.0
4.8 440
2.395 21.8
Vapor fraction
4.7 Temperature 439 2.390 21.6
2.385 21.4
4.6 438 2.380 21.2
0 5 10 15 20 25 30 0 5 10 15 20 25 30
a) Distance from inlet (m) b) Distance from inlet (m)
Figure 6
Results along the pipe for a) pressure and temperature, and b) specific enthalpy and vapor mass fraction, calculated by the original cor-
relation (marching algorithm with m 10) and with regularization functions (marching algorithm and DAE approaches).
0.190
Liquid holdup
4.9
Accurate temperature prediction is known to be very
important in oil and gas production two-phase flow
Pressure (kgf/cm2)
4.8
problems [4], which often involve transportation of
two-phase mixtures over long distances and high temper-
4.7
ature gradients. Approximate analytical solutions for
calculation of temperatures in wells have also been pub-
4.6
lished [1], which provides an alternative to performing Natural convertion, external T = 40C
rigorous heat transfer calculations. In short, there is a Forced convection, external T = 15C
4.5 Isothermal flow
reasonable number of studies on this matter which engi- Original parameters
neers engaged in well design can call upon. 4.4
On the other hand, little work on this subject has been 0 5 10 15 20 25 30
published when it comes to refining and petrochemical a) Distance from inlet (m)
two-phase flow. Consequently, process engineers
involved in downstream design are often dubious on
how strictly they should consider heat transfer calcula-
tions, or even if they need to be considered at all for cal- 441.5
culating pressure drops with acceptable engineering
441.0
accuracy. In order to develop a brief first insight into this
issue, the numerical techniques discussed above were 440.5
Temperature (K)
fractions, even though predicted pressure drops were analyse the influence of different heat-transfer
acceptably close to heat-transfer associated results. approaches on calculated results. The goal here was to
Therefore, further investigation should be conducted shed some light on the actual need for rigorous
before isothermal-flow assumption can be said to ade- heat-transfer calculations in downstream industry
quately dismiss rigorous enthalpy balances. design, as engineers are frequently faced with this issue.
It was found that for the given problem, as long as
heat transfer was considered, it was not necessary to
SUMMARY AND CONCLUSIONS specify precise ambient air temperature or atmospheric
conditions (natural or forced convection), as oppositely
The reasons for choosing practical pressure-gradient extreme considerations for these parameters yielded vir-
correlations over more advanced techniques (such as tually indistinguishable results. Replacing the energy
CFD) in specific situations in industrial design for solv- balance with an isothermal-flow consideration resulted
ing two-phase flow problems nowadays have been dis- in small differences in predicted pressure drops, but dra-
cussed. These two-phase flow methods require values matically changed other results, such as vapor fractions
of physical properties of both flowing phases (liquid along the pipe. Hence, this assumption requires further
and vapor), which also depend on temperature in addi- testing before it can be reported as good design practice.
tion to pressure. Therefore, the pressure gradients Finally, it must be stressed that even though a typical
obtained from these correlations must be integrated over refinery problem was considered in all calculations per-
the pipe length simultaneously with an enthalpy-gradient formed in this study, the numerical techniques presented
equation if pressure drops are to be calculated. are just as relevant for oil and gas production design,
It has been proposed that the marching algorithm, where improvements in efficiency and convergence prop-
which some publications present for coupling the erties are also expected.
pressure-gradient and enthalpy-gradient equations, is
not the most efficient numerical method for solving this
two-phase flow problem, as it requires that pressures and
temperatures be separately iterated for in nested loops, ACKNOWLEDGMENTS
with the algebraic thermodynamic and heat-transfer
constraints being solved at each step of the inner loop. The authors would like to dedicate this paper to Profes-
Representing the relevant equations as a Differential sor Alberto Luiz Coimbra, in the 50th anniversary of
Algebraic Equations (DAE) System, thus enabling them COPPE (1963-2013), the Graduate School of Engineer-
to be simultaneously solved by more advanced, suitable ing of the Federal University of Rio de Janeiro. The
techniques, has been demonstrated. The proposed authors also wish to thank Frederico Wanderley Tavares
DAE approach expectedly performed faster than the and Victor Rolando Ruiz Ahon for their valuable eluci-
marching algorithm in all reported comparison dations on calculations of VLE and physical properties.
calculations.
Empirical two-phase flow pressure-gradient correla-
tions such as Beggs and Brills (which was adopted in this REFERENCES
study) have long been known to bear mathematical dis-
continuities which can lead to convergence failure of 1 Brill J.P., Mukherjee H. (1999) Multiphase Flow in Wells,
SPE, Richardson, Texas.
numerical integrators, and this did indeed occur when
2 Gregory G.A., Aziz K. (1973) Calculation of pressure and
using the DAE solver as first attempted. temperature profiles in multiphase pipelines and simple
The use of regularization functions in closely repro- pipeline networks, J. Petrol. Technol. 17, 373-384.
ducing discontinuous mathematical models was pre- 3 Petalas N., Aziz K. (2000) A mechanistic model for multi-
sented, and this was proposed as an effective technique phase flow in pipes, J. Can. Pet. Tech. 39, 43-55.
for coming around this lack of convergence problem. 4 Beggs H.D., Brill J.P. (1991) Two Phase Flow in Pipes, Uni-
The discontinous equations of the Beggs and Brill corre- versity of Tulsa, Tulsa, Oklahoma.
lation were rewritten in terms of the regularization func- 5 Beggs H.D., Brill J.P. (1973) A study of Two-Phase Flow in
tions, and this approach was successful in preventing Inclined Pipes, J. Petrol. Technol. 25, 607-617.
convergence failure while not noticeably altering the cor- 6 Duns H. Jr, Ros N.C.J. (1963) Vertical flow of gas and
liquid mixtures in wells, Proc. Sixth World Pet. Congress
relations original results. Frankfurt, Germany.
Seeing as little work has been published when it comes 7 Lockhart R.W., Martinelli R.C. (1949) Proposed correla-
to two-phase flow in downstream petroleum industry, tion of data for isothermal two-phase two-component flow
the numerical techniques discussed above were used to in pipes, Chem. Eng. Prog. 45, 39-48.
510 Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 70 (2015), No. 3
8 Ansari A.M., Sylvester N.D., Sarica C., Shoham O., Brill 17 Lima E.R.A., Castier M., Biscaia E.C. Jr (2008) Differen-
J.P. (1994) A Comprehensive Mechanistic Model for tial-Algebraic Approach to Dynamic Simulations of Flash
Upward Two-Phase Flow in Wellbores, SPE Production Drums with Rigorous Evaluation of Physical Properties,
& Facilities 297, 143-152. Oil & Gas Science and Technology 63, 677-686.
9 Spedding P.L., Benard E., Donnelly G.F. (2006) Prediction 18 Riazi M.R. (2005) Characterization and Properties of Petro-
of pressure drop in multiphase horizontal pipe flow, Inter- leum Fractions, American Society for Testing and Materi-
national Communications in Heat and Mass Transfer 33, als, Philadelphia, Pennsylvania.
1053-1062. 19 Edmister W.C., Lee B.I. (1984) Applied Hydrocarbon Ther-
10 El-Sebakhy E.A. (2010) Flow regimes identification and modynamics - Volume 1, Gulf Publishing Company, Hous-
liquid-holdup prediction in horizontal multiphase flow ton, Texas.
based on neuro-fuzzy inference systems, Mathematics and 20 Daubert T.E., Danner R.P. (1997) API Technical Data
Computers in Simulation 80, 1854-1866. Book-Petroleum Refining, American Petroleum Institute
11 Garca F., Garca R., Joseph D.D. (2005) Composite power (API), Washington, DC.
law holdup correlations in horizontal pipes, International 21 Churchill S.W., Chu H.H.S. (1975) Correlating Equations
Journal of Multiphase Flow 31, 1276-1303. for Laminar and Turbulent Free Convection from a Hori-
12 Chang Y.S.H., Ganesan T., Lau K.K. (2008) Comparison zontal Cylinder, Int. J. Heat Mass Transfer 18, 1049-1053.
between Empirical Correlation and Computational Fluid 22 Churchill S.W., Bernstein M. (1977) A Correlating Equa-
Dynamics Simulation for the Pressure Gradient of Multi- tion for Forced Convection from Gases and Liquids to a
phase Flow, Proc. World Congress on Engineering, London, Circular Cylinder in Crossflow, J. Heat Transfer 99, 300-
UK, 2-4 July. 306.
13 Munkejord S.T., Mlnvik M.J., Melheim J.A., Gran I.R., 23 Alsoudani T.M., Bogle I.D.L. (2012) Linking Correlations
Olsen R. (2005) Prediction of Two-Phase Pipe Flows Using Spanning Adjacent Applicability Domains, Proc. 22nd
Simple Closure Relations in a 2D Two-Fluid Model, European Symposium on Computer Aided Process Engineer-
Fourth International Conference on CFD in the Oil and ing, London, England, 17-20 June.
Gas, Metallurgical & Process Industries, Trondheim,
Norway, 6-8 June. 24 Souza D.F.S. (2007) Abordagem Algebrico-Diferencial da
Otimizacao Dinamica de Processos, DSc Thesis, Universid-
14 Payne G.A., Palmer C.M., Brill J.P., Beggs H.D. (1979) ade Federal do Rio de Janeiro.
Evaluation of Inclined-Pipe Two-Phase Liquid Holdup
and Pressure-Loss Correlations Using Experimental Data, 25 Vieira R.C., Biscaia E.C. Jr (2001) Direct methods for con-
J. Petrol. Technol. 31, 1198-1208. sistent initialization of DAE systems, Computers and
Chemical Engineering 25, 1299-1311.
15 Incropera F.P., Dewitt D.P. (2007) Fundamentals of Heat
and Mass Transfer, John Wiley & Sons, New York, New
York.
16 Shampine L.F., Reichelt M.W. (1997) The MATLAB ODE Manuscript accepted in September 2013
Suite, Siam J. Scientific Computing 18, 1-22. Published online in April 2014
Cite this article as: R.G.D. Teixeira, A.R. Secchi and E. Biscaia (2015). Two-Phase Flow in Pipes: Numerical Improvements and
Qualitative Analysis for a Refining Process, Oil Gas Sci. Technol 70, 3, 497-510.