SCOPE
F’redicting the flow regime for concurrent gas liquid strongly on the particular data being used to prepare
flow in pipes has been a central unresolved problem in the map. For this reason extension to other conditions
twophase flow. The usual approach has been to collect of pipe size or inclination, fluid properties, and flow rates
data for flow rates and fluid properties and to visually are of uncertain reliability.
observe the flow pattern through a transparent test section This work has the objective of presenting a means for
window. Then a search is undertaken for a way to map unambiguous analytical prediction of the transition be
the data in a twodimensional plot by locating transition tween flow regimes based on physically realistic mecha
nisms for these transitions. The regimes considered are
boundaries between the regimes. This requires a decision intermittent (slug and plug), stratified smooth, stratified
to be made about the coordinates which are used. Be wavy, dispersed bubble, and annularannular dispersed
cause no theoretical basis for selection of coordinates has liquid flow. The theory predicts the effect on transition
existed in the past, this approach represents a coordina boundaries of pipe size, fluid properties, and angle of
tion of the data rather then a correlation and depends inclination.
The earliest and perhaps the most durable of regime Hubbard, 1975). Annular dispersed liquid flow represents
maps for twophase gasliquid flow was proposed by the condition described in the past of annular or semian
Baker (1954). Many have been suggested (White and nular with various degrees of liquid entrainment ranging
Huntington, 1955; Govier and Omer, 1962; Kosterin, between very small and large amounts of dispersed liquid.
1949). AlSheikh et al. ( 1970) defined a variety of dimen The process of analyzing the transitions between flow
sionless groups and using the Dukler twophase flow data regimes starts from the condition of stratified flow. The ap
bank evaluated the suitability of various pairs for mapping proach is to visualize a stratified liquid and then do deter
the flow regimes. They concluded that no two groups mine the mechanism by which a change from qtratified
characterized all of the transitions and all of the data. flow can be expected to take place, as well as the flow
Recently a mapping based on coordinates of superficial pattern that can be expected to result from the change.
gas and liquid velocity has been constructed from a larger In many cases stratified flow is seen to actually exist in the
data base (Mandhane et al., 1974). entry region of the pipe. However, the fact that stratified
Part of the problem arises from the lack of precision flow may not actually exist is not important, since it is well
in describing these visual observations. There have been established that the existence of a specific flow pattern at
innumerable classifications suggested, such as smooth strat specified gas and liquid rates is independent of the path
ified, wavy, semiannular, bubble, annular, froth, dispersed used to arrive at that state.
bubble, dispersed liquid, plug, and slug flow, among Since the condition of stratified flow is central to this
others. Hubbard and Dukler (1966), based on studies of analysis, the initial step is the development of a general
the spectral distribution of wall pressure fluctuations, sug ized relationship for stratified flows.
gested that each observation represents the superposition Equilibrium Stratified Flow
of three basic patterns: separated, intermittent, and dis
tributed flow. However, the concept does not discriminate Consider smooth, equilibrium stratified flow as shown
between stratified and annular flow, or between the dis in Figure 1. A momentum balance on each phase yieds
persed liquid or dispersed gas flow regimes, and these are
differences of considerable practical concern.
In this work a mechanistic model is developed for the
unambiguous analytical prediction of transition between
flow regimes. This approach also provides considerable in
sight into the mechanisms of the transitions.
THEORY
The analysis which follows considers the conditions for
transition between five basic flow regimes: (SS) smooth
stratified, (SW) wavy stratified, (I) intermittent (slug and
plug), (AD) annular with dispersed liquid, and (DB)dis
persed bubble. No distinction is made between slug, plug,
or elongated bubble flows, all being considered different
conditions of the intermittent flow regime (Dukler and Fig. 1. Equilibrium stratified flow.

hL 4Y=O (7)
D
where
x 2 =
D
(8)
.=[I (dP/dx)s
( dP/ dx :1
" Y= (PL  p c ) g sin = 
 (PL  P d g sin
I (dP/dX) G'1
Q
Fig. 2. Equilibrium liquid level for stratified flow (turbulent liquid, (9)
turbulent or laminar gas).
I (dP/dx) designates the pressure drop of one phase flow
 AG (9)
 T ~  ~iSj
~ +
~sina = 0
SPGAG (2)
ing alone in the pipe. Thus, X is recognized as the parame
ter introduced by Lockhart and Martinelli (1949) and can
be calculated unambiguously with the knowledge of flow
rates, fluid properties, and tube diameter. Y is zero for
Equating pressure drop in the two phases and assuming
horizontal tubes and represents the relative forces acting
that at transition conditions the hydraulic gradient in the
on the liquid in the flow direction due to gravity and pres
liquid is negligible, gives the following results
sure drop. It too can be calculated directly. All dimension
less variables with the superscript
hL/D, as can be seen from
 depend only on hL =
CI
2
Ti = fi
PC (UG
2 =  (10)
AG 0.25 [COS' ( 2 h ~ 1)
(4)
with the liquid and gas friction factors evaluated from
CI
 ( 2 h
~ 1) 4 (2zL 1)2] (11)
c CI
= 7r  cos1 (2hL  1 )
where DL and DG are the hydraulic diameter evaluated in
SL
CI
SG = cos1 (2hL
 1)
(12)
(13)
the manner as suggested by Agrawal et al. (1973) :
A/AG (16)
closedduct flow. It has been established that for smooth
stratified flow, fi N f~ (Gazley, 1949). Even though many Thus, each XY pair corresponds to a unique value of hL/D
of the transitions considered here take place in stratified for all conditions of pipe size, fluid properties, flow rate,
flow with a wavy interface, the error incurred by making and pipe inclinations for which stratified flow exists. The
this assumption is small. At flow rate conditions, where solution of Equation (7) has been executed for turbulent
transitions are observed to take place, u G >> ui. Thus, the flow of both phases, which is clearly the case of greatest
gas side interfacial shear stress is evaluated with the same practical interest ( n = m = 0.2, CG = CL = 0.046). The
equation as the gas wall shear. In this work the following results are shown as the solid curves in Figure 2 . Other
coefficients were utilized: CG = C L = 0.046, n = m = 0.2 situations may be readily solved from Equation (7) by
for the turbulent flow and CG = CL = 16, n = m = 1.0 utilizing the applicable coefficients. The case for turbulent
for laminar flow. liquid with laminar gas flow can occur in practice for
I t is useful to transform these equations to dimensionless transitions which take place at low gas rates. The solutions
form. The reference variables are: D for length, D 2 for for n = 0.2, m = 1, CL = 0.046, CG = 16 is shown dotted
area, the superficial velocities, u L S and u G S for the liquid in Figure 2 and is remarkably close to the turbulent/turbu
and gas velocities, respectively. By designating the dimen lent case, It should be noted that the decision on whether
sionless quantities by a tilde ( w ) , Equation (3) with (4) laminar or turbulent flow takes place in each phase should
and ( 5 ) takes the form be based on the Reynolds number calculated by using the
where C1 depends on the size of the wave: where F is a Froude number modified by the density ratio:
i lo
(dP/ dx :)
r
F= .& U i
(ppG)=
The region designated above as a stratified regime in In dimensionless form this can be expressed as
cludes two subregions: stratified smooth (SS) and strati
fied wavy (SW) . These waves are caused by the gas flow
under conditions where the velocity of the gas is suffi
cient to cause waves to form but slower than that needed where K is the product of themodified Froude number and
for the rapid wave growth which causes transition to in the square root of the superficial Reynolds number of the
termittent or annular flow. liquid:
Since U L and

UG depend only on hL/D [see Equations ( 15)
dimensionless coordinates. It is possible to map this transi
tion on Figure 4 by using the common X abscissa and T
and (16) 1, they are determined once X and Y are speci as an ordinate as shown by curve D calculated for Y = 0.
fied. Thus this transition between smooth and wavy annular
flow depends on the three parameters K, X , and Y. For RESULTS
any fixed inclination, this becomes a twoparameter de The Generalized Flow Regime Map
pendence on X and K . The relationship which satisfies the
equality of Equation (30) can conveniently be mapped in The generalized flow regime map for the case of hori
Figure 4 by designating a lfferently scaled ordinate than zontal tubes ( Y = 0) appears in Figure 4. Curbe A repre
that which applies to the two transitions discussed previ sents the transition from stratified (S) to intermittent ( I )
ously. Curve C shows the results for Y = 0 (s = 0.01). or annulardispersed liquid ( A D ) flows, with the coordi
While the location of this transition curve is approxi nates for curve A being F vs. X . The curve gives the locus
mate, it is important to note that it is based on a physi of the FX pairs which satisfies Equation (25) And results
cally realistic model. Should it be necessary to locate the from the argument that waves of finite size will grow and
curve more accurately, this would be possible once addi tend to block or sweep around the pipe when the force due
tional data on c/uL and s are available. However, the re to the Bernoulli effect above the wave is greater than
sult depends on each of these quantities to the onehalf gravity force acting on the wave. Thus, all values of X to
power and thus is relatively insensitive to changes. the left of the curve represent conltions under which
stratified flow will exist.
Transition Between Intermittent ( I ) and Dispersed Bubble Curve B locates the transition between intennittent ( I )
(DB) Regimes or dispersed bubble ( D B ) and annulardispersed liquid
For values of X in Figure 4 to the right of boundaries A ( A D ) flow. This occurs at a constant value of X resulting
and B, waves will tend to bridge the pipe forming a liquid from the argument that the growing waves will have sufE
slug and an adjacent gas bubble. At high liquid rates and cient liquid supply to form a slug only when h L / D 0.5,
low gas rates, the equilibrium liquid level approaches the and below that value they will be swept around the pipe
top of the pipe, as is apparent from Figure 2. With such into an annular configuration.
a fast running liquid stream the gas tends to mix with the Curve C represents the transition between stratified
liquid, and it is suggested that the transition to dispersed smooth (SS) and stratified wavy (SW) flow. It is plotted
bubble flow takes place when the turbulent fluctuations are in the KX plane and locates the KX pairs which satisfy
strong enough to overcome the buoyant forces tending to Equation (30). The model is based on the assumption that
keep the gas at the top of the pipe. the Jeffreys model is valid for describing the condition for
The force of buoyancy per unit length of the gas region transfer of energy to the liquid in order to create waves,
is with the wave velocity estimated from the mean velocity
FB = g cos a ( P L  ~ G ) A G (32) of the liquid film and the sheltering coefficient determined
from an analysis of Benjamin. Any value of K lower than
In a manner used by Levich (1962), the force acting be
curve C in the KX plane will provide insufficimt gas flow
cause of turbulence is estimated to be
to cause waves to form.
FT = M P L P Si (33) Curve D indicates the transition between intermittent
and dispersed bubble flow. It represents an identification
where u' is the radial velocity fluctuation whose rootmean of conditions where the turbulent fluctuations in the liq
square is estimated to be approximately equal to the fric uid become equal to the buoyant forces which tend to
tion velocity. Thus make the gas rlse to the top of the pipe. This curve gives
the locus of the TX pairs which satisfies Equations (36).
(34) All values of T below the curve represent conditions where
turbulence is insufficient to keep the gas mihed, and the
Dispersion of the gas is visualized as taking place when elongated gas bubbles characteristic of intermittent flow
FT 2 Fs, or will form. The set of transition curves for other values of
Y can easily be calculated from Figure 2 and the defining
transition equations.
The effect of pipe roughness on these transitions is not
specifically considered in the development. However, sub
In a dimensionless form, Equation (35) takes the form ject to experimental demonstration, it is suggested that if
the (dP/dx) values are calculated by using known rough
ness parameters, the transition boundaries of Figure 4 will
continue to apply.
It is, of course, not necessary to use a flow regime map
where at all. Given any one set of flow conditions (rate, pressure,
line size, and inclination), the flow pattern that exists for
that condition can be determined rather simply by using
hand calculations from Equation (25), (303, and (36)
(PLPG)gCOSa I with the help of Figure 2.
Comparison with Doto
Mandhane, Gregory and Aziz (4974) have recently
(PL  PG)g cos a made a careful examination of flow regime data. They
T can be considered as the ratio of turbulent to gravity showed that over 1 0 0 0 data points for the airwater sys
forces acting on the gas. tem in horizontal pipes ranging in size from 1.3 to 15 cm
5.0 cm

.01 30 cm
U; Cm/sec~ U: Crn/sec I
Fig. 5. Comparison of theory and experiment. Waterair, 25"C, 1 atm,
2.5 cm. diam., horizontal.  theory;//////// Mandhane
et al. (1974). Regime descriptions a s in Mandhane.
25"C, 1 atm, horizontal. 
Fig. 6. Effect of pipe diameter on transition boundaries. Waterair,
theory; / /////Mandhow
(1974).
et al.
50 , 1 I I
1 their absolute locations. Thus it is possible to predict the
I
DISPERSED BUBBLE (DB)
/ boundaries as the uLs, u G S maps from this new theory.
Now that the theoretically based regime transition cal
culations have been shown to be in good agreement with
D
 data, it is possible to explore the effect of design variables
with some confidence. Again, by using airwater at 25°C
and 1 atm the theoretical predictions of Figure 4 have
I 5.0 crn dia. been recalculated to U L ~ ;U G coordinates
~ for 1.25, 5, and
u:  30 cm dio, 30 cm diameter horizontal pipes. These are shown in Fig
L m/sec 3 ure 6. Superimposed on each graph are the boundaries
ANNULAR
recommended by Mandhane which, as indicated above, are
.I DISPERSED based on data for 1.3 to 5 cm pipes. Note that the location
LIQUID of theoretical boundaries B and C are independent of line
(AD) size. It is seen that for a narrow range of line sizes (say 2
to 5 cm), the location of the boundaries are not very sensi
.01 tive to the size. However, for the larger sizes such as 30
cm, the displacement of the boundaries is significant. Con
.I I 10 100 500 siderable error will result if a single U G ~ uLs map is used.
As expected, for large pipe sizes, the theory predicts that
U; C m/sec I stratified flow will persist to much higher gas rates.
Fig. 7. Effect of fluid properties on transition boundaries. Crude oil Of considerable practical impoftance is the flow of oil/
natural gas, 38"C,68 atm, horizontal. natural gas at high pressures where the properties are
drastically different than for the air/water case. Flow re
in diameter could be coordinated on a map in which uLs gime transitions on U G ~ u, L s coordinates were calculated for
and U G are
~ the parameters. No theoretical basis was given a horizontal pipeline operating at 68 atm and 38°C with
for this method of mapping, and, in fact, the goodness of oil of density of 0.65 g/cm3 and natural gas with a density
fit did vary with pipe size and fluid properties. Because the of 0.05 g/cm3. Viscosities of oil and gas were set at 0.5
largest part of the data were for 1.3 to 5 cm pipe sizes, and 0.015 cp, respectively. The results for 5 and 30 cm
the location of the transition boundaries was strongly influ pipe sizes are shown in Figure 7. A comparison with Fig
enced by these small line size, airwater data. The location ures 5 and 6 drawn for airwater shows the inadequacy of
of these transition boundaries in the Mandhane map may assuming that the U G ~ U L ~coordinate maps are inde
then be considered a representation of that particular data. pendent of properties. The transition from smooth to wavy
Others have also suggested this method of mapping flow stratified flow and from stratified to annular flow shifts to
regimes (Govier and Omer, 1962). gas velocities an order of magnitude lower. This is, of
In order to compare the predictions of the theory given course, due to higher gas density. The new theory pre
in this paper with the data, the generalized flow bound sented here accounts for these conditions.
aries of Figure 4 were recalculated into u L s  u G S co The effect of small degrees of inclination on the location
ordinates for the system airwater at 25°C and 1 atm pres of the transitions appears in Figures 8 and 9 as calculated
sure in a 2.5 cm diameter horizontal tube. Once these vari from the generalized theory. The case selected considers
ables are fixed, F , X , K , and T at each transition boundary airwater at low pressure in a 5 cm diameter pipe. The
are all expressible in terms of the two superficial velocities. effect of inclination is very pronounced. Downward in
The results are shown in Figure 5. The solid curves repre clinations cause the liquid to move more rapidly, have a
sent the prediction of the theory presented here. The bands lower level, and thus require higher gas and liquid flow
indicate the data (1.3 to 5 cm) represented by the Mand rate to cause a transition from stratified flows. As shown
hane boundaries. Very satisfactory agreement exists both by comparing Figure 6 with Figure 8, the intermittent
with respect to the significant trends of the curves and flow region shrinks substantially. Conversely, flow with
”: I
[ m/sec 1
.I .I
.01 .01
I I I I \ , , \ , I
.I I 10 100 500 I I 10 100 500
U: c m/sec I U: c m/sec I
Fig. 8. Effect of inclination on transition boundaries. Waterair, ZS”C, Fig. 9. Effect of inclination on transition boundaries. Waterair,
1 atrn, 5 crn. diarn., downflow. 25”C, I atrn, 5 crn. diam., upflow.
slight upward angles cause intermittent flow to take place x = coordinate in the downstream direction
over a much wider range of flow conditions as shown in X = Martinelli parameter, Equation (8)
Figure 9. I n fact, at an angle of 0.1 deg., intermittent Y = dimensionless inclination parameter, Equation (9)
flow is predicted to take place at extremely low liquid and a! = angle between the pipe axis and the horizontal,
gas rates. The peculiar shape of the intermittent to strati positive for downward flow
fied transition boundary for CY = 0.03 deg. is due to a p = density
change in the flow from turbulent to laminar. T = shear stress
It should be noted that the generalized map of Figure 4 Y = kinematic viscosity
has been calculated for turbulent flow of both phases. Subscripts and Superscripts
However, as shown in Figure 2, laminar flow of the liquid
has little effect on the result.
G = gas
i = liquid gas interface
L = liquid
ACKNOWLEDGMENT s = superficial, for single fluid flow
This paper is dedicated to Ovid Baker, who presented the
first useful flow regime map. He has been a continuing
influence on the research related to twophase flow problems

W
I
= pipe surface
= dimensionless variable
= disturbed variable
at the University of Houston. * = friction velocity
 = average
NOTATION
LITERATURE CITED
A = flow crosssectional area Agrawal, S. S., G. A. Gregory, and G. W. Govier, “An Analysis
AD = annularannular dispersed liquid flow of Horizontal Stratified TwoPhase Flow in Pipes,” Can.
c = wave velocity 1. Chem. Eng., 51,280286 ( 1973).
C = coefficient dependent on the size of disturbance, AlSheikh, J. N., D. E. Saunders, and R. S. Brodkey,
also constant in the friction factor correlation “Prediction of Flow Patterns in Horizontal TwoPhase Pipe
D = pipe diameter and hydraulic diameter Flow,” ibid., 48,21 (1970).
Baker, O., “Simultaneous Flow of Oil and Gas,” Oil Gas J.,
DB = dispersed bubble flow 53, 185 (July, 1954).
f = friction factor Benjamin, T. B., “Gravity Currents and Related Phenomena,”
F = modified Froude number, Equation (26) 1. Fluid Mech., 31,209248 (1968).
g = acceleration of gravity ., “Shearing Flow Over a Wavy Boundary,” ibid., 6,
h = liquid level or gas gap 161 (1959).
Z = intermittent (slug and plug) flow Brock, R. R., “Periodic Permanent Roll Waves,” Proc. Am.
K = wavy flow, dimensionless parameter, Equation SOC. Ciuil Engrs., 96, HYD 12, 25652580 ( 1970).
Butterworth, D., “A Visual Study of Mechanism, in Horizontal,
(31) Air Water Flow,” AERE Report M2556, Harwell, England
m trexponent, Equation ( 5 )
(1972).
n = exponent, Equation ( 5 ) Chu, K. T., “Statistical Characteristics and Modelling of
P = pressure Wavy Liquid Films in Vertical Two Phase Flow,” PhD
Re = Reynolds number thesis, Univ. Houston, Tex. ( 1973).
S = Jeffreys’ sheltering coefficient Dukler, A. E., and M. G. Hubbard, “A Model for GasLiquid
S = perimeter over which the stress acts, also stratified Slug Flow in Horizontal and Near Horizontal Tubes,”
flow Ind. Eng. Chem. Fundamentals ( 1975).
ss = stratified smooth flow Fulford, G. D., “The Flow of Liquids in Thin Films,”
Aduan. Chem. Eng., 5,151236 ( 1964).
sw = stratified wavy flow Gazley, C., “Intertacial Shear and Stability in TwoPhase
T = dispersed bubble flow dimensionless parameter, Flow,” PhD theses, Univ. Del., Newark ( 1949).
Equation (37) Govier, G. W., and M. M. Omer, “The Horizontal Pipeline
# = veiocity in the x direction Flow of AirWater Mixtures,” Can. J. Chem. Ehg., 40,
V = velocity normal to the x direction 93 (1962).
SCOPE
Typical systems of algebraic equations in steady state, has a onetoone solution sequence is said to contain
macroscopic design problems have a few characteristics persistent iteration.
which are dominating factors in the difficulty of obtain For systems of algebraic equations without persistent
ing solutions. Usually the equations contain few variables iteration, all onetoone solution sequences are considered
(are sparse) and are nonlinear. Solution strategies for as equal minimum difficulty solution strategies. This is a
sparse systems of equations are, in general, easily ob somewhat arbitrary objective function, since it might be
tained. Conversely, the nonlinearity of the equations in less difficult to solve a set of linear simultaneous equations
creases the difficulty of obtaining a solution. than a series of acyclic equations for nonlinear elements.
A general characteristic of a system of algebraic de Previous algorithms developed for the selection of de
sign equations is that the number of variables is equal sign variables in systems of equations without persistent
to or greater than the number of equations. When the iteration are based on the pioneering work of Steward
number of variables is greater than the number of equa (1962) which introduced the concept of admissable output
tions, a set of design variables must be assigned numerical sets. An admissable output set has two properties: each
values in order that the system be reduced to a determinant equation contains exactly one output variable, and each
system with an equal number of equations and variables. variable appears as the output element of exactly one
Judicious selection of design variables can lead to a set equation.
of equations whose solution sequence encounters a mini The work by Lee et al. (1966) was one of the earliest
mum of difficulty. attempts to base design variable selection on a mathemati
Solution sequences for systems of algebraic equations ical basis. This algorithm operates on a bipartite graph
can be separated into three classes: onetoone or acyclic, representation for a system of equations and operates
simultaneous, and iterative solution sequences. A oneto only on systems of equations without persistent iteration.
one, as opposed to simultaneous or iterative solution se The Lee et al. (1966) algorithm was adapted to operate
quences, solves each equation, whether linear or non on occurrence matrix representations as presented in
linear, in a one at a time technique and does not require Rudd and Watson (1968). This algorithm assigns an ad
any assumed solution points. Systems of equations which missable output set to a system of equations. The result
can be reduced to a determinant system with a onetoone of the algorithm is a single, optimal combination of de
solution sequence are said to be without persistent itera sign variables and a onetoone solution sequence for the
tion. Any system which cannot be reduced such that it system of equations.
Correspondence concerning this paper should be addressed to W. Fred
A system of equations can have many combinations of
Ramirez. design variables which will result in onetoone solution