J Braz. Soc. Mech. Sci. Eng.

DOI 10.1007/s40430-015-0331-7

TECHNICAL PAPER

Analysis of slug tracking model for gas–liquid flows in a pipe

E. S. Rosa1 · R. A. Mazza1 · R. E. Morales2 · H. T. Rodrigues3 · C. Cozin2 

Received: 20 December 2014 / Accepted: 25 February 2015

© The Brazilian Society of Mechanical Sciences and Engineering 2015

Abstract  This work focuses on the physical representa- List of symbols

tiveness of slug tracking models. The numerical analyses A Pipe cross-sectional area
are supported by a detailed experimental campaign in a Cav Average coalescence rate
horizontal test section with 900 pipe diameters long. The C0 Drift flux parameter
analyses address: (a) the model’s performance in predict- C∞ Drift flux parameter
ing averaged slug flow properties, (b) the bubble overtaking f Slug frequency
mechanism and the slug flow sensitivity to the wake func- Fr Froude number
tion, (c) the slug flow sensitivity to the slug insertion pro- g Gravity acceleration
cess, (d) the capture of the slug’s evolution in flows with hf Liquid film thickness
distinct slug formation processes and (e) the inclusion of j Cell number
the advection term and the induced slug oscillations. The J Mixture superficial velocity
work presents a compressible, one-dimensional slug track- JL Liquid superficial velocity
ing model satisfying the mass and momentum equations. JG Gas superficial velocity
The equations are presented in a unified fashion to han- L Pipe length
dle horizontal and inclined gas–liquid flows. The model Lf Liquid film length
embodies all terms introduced by the previous models and LS Liquid slug length
includes the advection term, so far neglected. The analyses, ṁ Mass flow rate
despite of being supported by the present numerical model, Rf Liquid film holdup
are not constrained to this model, but apply to slug tracking RS Liquid slug holdup
models in general. ud Dispersed bubble drift velocity
Ub Gas-phase velocity within the slug
Keywords  Mathematical modeling · Multiphase flow · Uf Liquid-phase velocity within the film
Fluid mechanics · Slug flow · Slug tracking UG Gas velocity above the film
US Liquid-phase velocity within slug
UM Mixture velocity
x Liquid slug front
y Bubble nose front
Technical Editor: Francisco Ricardo Cunha. z Pipe axial coordinate
* E. S. Rosa
[email protected] Greek symbols
λ Liquid film interfacial angle
1
Mechanical Engineering Faculty, State University μ Dynamic viscosity
of Campinas, São Paulo, Brazil ξ Centroid coordinates
2
Federal Technological University of Paraná, Paraná, Brazil ρ Density
3
PETROBRAS, Rio de Janeiro, Brazil θ Pipe angle with the horizontal

13
 J Braz. Soc. Mech. Sci. Eng.

σ Surface tension Inspired by the wave-like behavior of the translational

τ Wall shear stress velocity of the elongated bubble nose, Barnea and Taitel
[3] proposed a kinematic model applied to hydrodynamic
Subscripts slugs to predict the liquid slug length distribution based
G Gas phase on the bubble overtaking mechanism [22]. At the entrance,
L Liquid phase they estimate a relationship between the film and the slug
M Gas–liquid mixture lengths satisfying the volumetric gas flow rate considering
non-aerated slugs and no liquid film thickness. Compari-
sons against experimental data shown that slug length dis-
1 Introduction tribution at the entrance does not affect the outlet distribu-
tions at 400D downstream the inlet. Zheng et al. [33] adds
The slug flow regime is characterized by a succession of the liquid-phase mass balance on this model to track indi-
aerated liquid slugs trailed by elongated gas bubbles, here- vidual slugs. The basic mechanism of slug growth, genera-
after denominated as cells, which are not periodic in time tion and dissipation, taking place on hilly terrains, was cap-
and in space. This intrinsically unsteady flow regime is due tured by the model but the authors admit that the proposed
to the interaction between neighboring cells, which creates model is very approximate. Cook and Behnia [5] compared
or destructs the liquid slugs, changes the cells’ frequencies, Barnea and Taitel model against experimental slug length
hold up, lengths, and the pressure drop as they move down- distributions taken at +5° inclined line.
stream the line. These flow disturbances may be significant Nydal and Banerjee [23] propose a Lagrangian track
in long pipelines usually found in the transport of crude oil algorithm to the liquid slug and to the elongated gas bub-
from the well to the oil rig. Large slugs exiting the pipeline ble. The bubble–slug interface displacements are evaluated
are a potential source of damage to the downstream plant satisfying the phases’ mass and the momentum balances
processing facilities. The pipeline operational conditions applied to the liquid slug and to the liquid film. They arrive
and the safety of the downstream equipment have been into a set of equations which define the pressures and the
motivating the development of slug flow models. The liter- slug velocities. Unfortunately the presentation is brief and
ature distinguishes the slugs as hydrodynamic and terrain- many important details were omitted. The model accounts
induced. The former is associated with the development for gas compressibility, but neglects the slugs’ aeration. The
of hydrodynamic gas–liquid instabilities which lead to the pressure equation arises from the conservation of the gas
slug formation and the latter is linked to the pipeline topog- phase inside the pipe, but does not account to the gas fluxes
raphy. The time scale of the hydrodynamic slugs is smaller at the pipe inlet and outlet. The model includes the acceler-
than that of the terrain-induced slugs due to transient liquid ation terms, but the presented numerical calculations were
accumulation in the lowest points of the line. performed without the acceleration terms. The numerical
Despite the existence of previous models on slug flow, integration routine is explicit in time. Two slug insertion
this work takes for referencing purposes the first hydrody- methods were tested: a constant slug length and randomly
namic slug flow model as the one developed by Dukler and generated slug length. The model is tested employing two
Hubbard [6]. The model is based on a periodic cell or unit cases: a hypothetical flow of a train of hydrodynamic slugs
cell and predicts the average liquid fractions and pressure in a straight pipe and a experimental terrain-slugging flow
drop for given flow rates, pipe inclination and fluid trans- data. Based on the hypothetical flow, the work explored the
port properties. Choosing a frame of reference moving with wake effect influence on the formation of liquid slug popu-
the bubble nose velocity, the space coordinates are frozen lation and other flow features associated with the slug ter-
in time and the unit cell becomes stationary or in steady- rain data.
state regime. A review of hydrodynamic steady-state slug Taitel and Barnea [26] enforce the model presented
flow models is found in Taitel and Barnea [27] and Fabre in Barnea and Taitel [3] to satisfy the phases’ mass and
and Liné [7]. momentum balances. This hydrodynamic slug tracking
An advance to the steady-state slug flow models came model neglects the bubble overtaking mechanism and the
along the 90s with the development of the slug track- advection term, while the momentum equations assume a
ing models applied to hydrodynamic and terrain-slugging local equilibrium force balance for each cell. The evalu-
types. This class of model employs a Lagrangian approach ated variables are the pressure and the liquid slug veloc-
to track the bubble–slug boundaries as they evolve down- ity, both of them positioned at the mid-length of the liquid
stream the pipe. The models range from simple kinematic slug. The system’s discrete form is a bidiagonal matrix.
expressions to complex models where slug boundary dis- The analysis is based on the numerical results taken in a
placements are constrained by mass and momentum bal- 100-m-long horizontal line with one end open to atmos-
ances. A brief review of slug tracking models follows. phere for an air–water flow. At the inlet, the liquid slugs

13
J Braz. Soc. Mech. Sci. Eng.

and the elongated bubbles have constant sizes. The peri- is called slug capturing methods. A big issue still to be
odic mass insertion on the inlet causes a periodic fluctua- surpassed by these models is to account the 3D gas–liq-
tion on the pressure and on the liquid slug velocity. The uid mixer details into a 1D model once entrance flows are
results disclose that the gas compressibility effect causes strongly influenced by the mixer discharge conditions.
an increase in the gas bubble lengths and has a minor Also, there are no capture models applied to upward verti-
effect on the growth of the liquid slug lengths as they cal flows. In addition, this new class of methods demands
flow downstream. Al-Safram et al. [1] extended Taitel large computational effort. These models are promising in
and Barnea [26] model to hilly terrain applications using capturing the slug initiation in horizontal and near hori-
Zheng et al. [33] model and included liquid slug holdup zontal lines, but likewise slug tracking models are still in
and the overtaking mechanism. The liquid and gas phases development.
are injected into the inlet in specific time intervals; but no This work concerns to hydrodynamic slug tracking mod-
further details are given about the inlet process. The model els whose numerical advantage consists of a simple grid
is validated against two hilly terrain experimental cases with elements encompassing a liquid slug or an elongated
with reported accuracy of 7.7 % off to the maximum slug bubble which usually are -several pipe diameters long. This
length. The authors pointed a need of improvement in the class of method captures the evolution of hydrodynamic
following topics: slug length distribution at the entrance, slug properties and is the foundation to develop terrain slug
the initiation at the bottom elbow, a better closure rela- models.
tionship for the liquid slug holdup and the slug dissipation The review work on slug tracking models shows that
mechanism for downhill sections. More recently, Wang the numerical results are compared against the experimen-
et al. [34] included in this model the wake effect and the tal averaged slug length and its distribution [3, 5, 23, 26,
pressure drop term due to liquid film acceleration to study 34]. Probably, the developers’ focus was on the applica-
the experimental liquid slug length distributions in a hori- tion where the maximum slug length is a target to design
zontal line with 2660 pipe diameters long. Accordingly to industrial facilities. Other slug flow properties such as the
Wang et al. [34], the mean slug length at z/D = 2609 was elongated bubble length, the slug frequencies, the bubble
underestimated due to the assumption of a short minimum coalescence rate and pressure drop were neglected in the
stable slug length assumed in the model. analyses. To improve the understanding of how this class of
Ujang et al. [29] use the liquid slug and the elongated model captures the flow the authors consider necessary an
gas bubble as separate objects to propose a new hydrody- analyses of the additional slug flow properties in conjunc-
namic slug tracking model. The solved variables are the tion with the slug length.
time rate of the film length and the time rate of gas mass This paper focuses on the slug tracking model capability
inside the elongated gas bubble. The mass conservation to capture the physical flow features of the intermittent slug
equations of the phases are applied to each object, but the regime. The numerical results are supported by a detailed
gas is considered incompressible. The liquid slugs are aer- experimental campaign which evolved in parallel to the
ated. A quasi-steady momentum balance is used to get the numerical development. The analyses, despite being drawn
pressure. The insertion process has fixed slug length and to a specific slug tracking model, are not constrained to a
occurs at time intervals taken from an exponential distribu- specific model, their results apply to slug tracking models
tion based on experimental data. The model was originally in general. The analyses address:
developed to horizontal flows and validated against the
experimentally determined slug frequencies in an air–water 1. The model’s performance to predict averaged slug flow
flow at near atmospheric pressure. The model is sensitive to properties.
the slug length distribution at the entrance and the authors 2. The process of bubble overtaking mechanism and the
recommend a predictive model to this parameter. sensitivity of the slug flow properties on the wake
A limitation of the slug tracking models is the need function.
of the slug properties (velocities, lengths, holdups and 3. The sensitivity of the slug flow properties on the slugs
frequencies) at the inlet or at an elbow where the flow is distributions at the inlet.
restarted due to the new pipe inclination. This process is 4. The capture of slug flow development with distinct for-
called slug initiation where the flow instabilities evolve to mation processes.
hydrodynamic slugs. Issa and Kempf [14] demonstrated 5. The inclusion of advection term and the induced slug
the capability of the two-fluid model to capture the slug oscillations.
initiation process to horizontal and near horizontal flows.
An alternative approach is the development of hybrid two- The next sections introduce the physical model, the
fluid and slug tracking models; see Larsen et al. [17] and numerical model, the experimental techniques and the
Kjeldby et al. [15] for example. This new class of models results.

13
 J Braz. Soc. Mech. Sci. Eng.

Fig. 1  Physical representation of the slug flow cell properties, the cell numbering and the pipe cross section

2 The physical model associates the gas density to the jth cell due to the com-
pressible effects.
The identification of the liquid slug and elongated gas bub- Slug tracking models view the intermittent flow as a
ble properties within the jth cell is done using the nomen- succession of individual cells flowing downstream the
clature depicted in Fig. 1 for a pipe length L, whose diam- pipe, exchanging mass and momentum with the neighbor-
eter and cross-sectional area are D and A and it is inclined ing cells. Concisely, the slug tracking models can be rep-
with respect to the horizontal by θ. The liquid slug and bub- resented in two steps: (a) a pair of ODEs which define the
ble front positions at the jth cell are xj and yj, relative to an time evolution from state ‘old’ to ‘new’ of P and US at the
inertial and stationary frame of reference displayed in the jth cell:
figure. The liquid slug properties of the jth cell at a given
instant of time are the length, holdup and the velocities of dPj j−1 j
K1 − K2 US + K3 US = K4 , (1)
the liquid and dispersed bubbles represented, respectively, dt
j j j j
by LS , RS , US , and Ub . The liquid slug wall shear stress j
and the wetted perimeter, not shown in the figure, are rep- dUS j
j K5 + K6 US + Pj+1 − Pj = K7 , (2)
resented by τs and S j , respectively. The liquid film region dt
is composed by a non-aerated liquid film and an elongated and (b) the displacement of the jth bubble nose and slug
gas bubble. The properties of the jth liquid film region at front:
a given instant of time are the length, the holdup, the liq-
uid film thickness and the velocities belonging to the liq- dyj j
j j j j j = UT , (3)
uid film, gas phase and the bubble nose: Lf , Rf , hf , Uf , UG dt
j
and UT , respectively. In addition, Pj is the jth elongated j
bubble pressure and it is considered uniform within the jth dx j dL j
= S +UT . (4)
film region. Lastly, it is necessary to define the wall shear dt dt
j j
stresses due to the liquid and gas phases as τf and τG and Equations (1) and (2) represent the mass and momentum
j j
the associated wetted perimeters, Sf and SG , not represented balance as shown in Eqs. (13) and (17) to be seen in Sects.
in the figure. A special notation is employed at the bubble– 2.2 and 2.3. Furthermore, the coefficients ‘K1’ through ‘K7’
slug interface to distinguish the liquid and the gas veloci- are dependent of the slug flow as defined in Table 4.
ties: Uf,xj , Uf,yj and UG,xj , UG,yj, see representation in Fig. 1. The nonsimilarities among the slug tracking models arise
The liquid and gas phases have the density and dynamic in step (a) due to the approximations carried on by the models
j
viscosity defined by ρL, ρG , μL and μG where the j label to get the mass and momentum balances. These differences

13
J Braz. Soc. Mech. Sci. Eng.

lead to distinct definitions of coefficients ‘K1’ through ‘K7’ in sections at xj−1 and xj, see Fig. 1. The control surface tracks
Eqs. (1) and (2). To date the slug tracking algorithms neglect the jth cell boundaries. The lumped parameters associated
the coefficients ‘K5’ and ‘K6’ simplifying the momentum with the aerated liquid piston, RS, US and Ub are uniform
equation to a balance between pressure difference and friction throughout the liquid slug. Due to the uniformity hypoth-
forces. The time–space advance of the interfaces is a straight esis the following identity holds at the boundaries xj and yj:
forward integration stated in step (ii) provided dLS/dt and UT
are known. The bubble nose advance procedure done in step Rs,xj = Rf,yj ≡ Rsj ; US,xj = Uf,yj ≡ Usj and
(ii) is the same for all slug tracking models, but some differ- UG,yj = Ub,yj ≡
j
Ub . (5)
ences among the models arise at the slug front position, dxj/dt
due to the used approximations to estimate dLS/dt. Employing the definitions given in Eq. (5), the phase’s
Equations (1) and (2) combine a continuous time deriva- mass flow rates crossing xj and yj at the boundaries are con-
tive with a spatial difference due to the finite lengths of the veniently expressed in terms of the slug properties as:
liquid slugs and film regions. It is noteworthy to observe
dx j
 
that expressing the time derivatives by finite differences, ṁL,xj = ρL −
j
US
j
RS A and
the system’s final form has a remarkable similarity with the dt
first-order discretization of a quasi-linear hyperbolic sys- dx j 
  
j j j
tem evoking the wave behavior to be seen in the solutions. ṁG,xj = ρG Ub − 1 − RS A, (6)
dt
The slug tracking model advances each cell in space and in
time allowing interaction between neighboring cells through
dyj
 
the wake law and enforcing the mass and momentum balances. j
ṁL,yj = ρL US −
j
RS A and
The degree of approximation on the mass and momentum var- dt
ies from one model to the other. This paper presents a com- dyj 
  
j j j
pressible, one-dimensional and transient slug tracking model ṁG,yj = ρG Ub − 1 − RS A (7)
dt
satisfying the mass and momentum equations in a unified fash-
ion to handle horizontal and inclined gas–liquid flows. Each Uniform values of RS, US and Ub do not accurately esti-
tracked cell has distinct gas and liquid velocities, lengths and mate the mass flow rate crossing the xj boundary due to
holdups. The present model embodies all terms of the previ- the wake effect induced by the leading bubble. The liquid
ous models and includes the advection term, so far neglected. film associated with the jth +1 bubble sheds liquid into the
It allows the accelerating liquid slugs to compress or expand jth liquid slug creating a 3D turbulent recirculation region
the elongated gas bubbles inducing oscillations along the pipe. responsible for the liquid slug aeration mechanism. Mod-
This feature makes the model complete with regard to the eling the slug aeration processes and the near wake esti-
inclusion of all one-dimensional flow mechanisms. mates for RS, US and Ub are out of the scope of this work.
The development of the proposed slug tracking model
has its origin in Grenier’s [11] work on non-aerated, com- 2.2 The phases’ mass balance at the jth cell
pressible, horizontal slug tracking model. The present
model is based on the following hypothesis: The liquid and gas mass balances at the jth cell are:
d j j

i. The flow is isothermal, the liquid phase is incompress- A ρL Rf Lf + ρL Rsj Lsj + ṁL,xj − ṁL,xj−1 = 0, (8)
ible and the gas phase behaves as an ideal gas. dt
ii. The gas density is constant within the cell, but it may and
change from cell to cell as the pressure changes. Fur-
d j    
thermore, the gas density is much smaller than the liq- A
j j j
ρG 1 − Rf Lf + ρG 1 − Rsj Lsj + ṁG,xj − ṁG,xj−1 = 0. (9)
dt
uid density; ρG/ρL ≪ 1.
iii. The liquid and the gas phases share the same pressure; Substituting the definitions given by Eq. (6) into Eq. (8)
surface tension effects are neglected. and simplifying, one gets:
iv. The pressure within the liquid film region is constant.
j j
v. The liquid film is non-aerated. 
j
 dyj

j dx
 j−1
j dR dRs
Rf − Rsj + Rsj−1 − Rf + Lf f + Lsj
dt dt dt dt
2.1 Control volume and mass fluxes definitions j−1 j−1 j j
= Rs Us − Rs Us . (10)

The mass and momentum analyses are carried out on a con- Similarly, substituting Eq. (6) into Eq. (9) and consid-
trol volume encompassing the jth cell. The control volume ering the gas density between neighboring bubbles nearly
boundaries coincide with the pipe wall and the pipe cross equal, i.e., ρj−1 j
G  ≅ ρG, one gets:

13
 J Braz. Soc. Mech. Sci. Eng.


j
 dyj   j−1
j dx j dR
j
dRs
j pressure difference between two neighboring gas bubbles.
− Rf − Rsj − Rsj−1 − Rf − Lf f − Lsj This result arises naturally because the pressure difference
dt dt dt dt

j−1 j−1

j j acting on xj−1 and xj boundaries corresponds to the pressure
= 1 − Rs Ub − 1 − Rs Ub
of the (j) and (j + 1) cells. The second term is the mixture

j

j
   1 dρ j weight followed by the wall friction at the liquid slug and at
− 1 − Rf Lf + 1 − Rsj Lsj j G
. the liquid film region, the 3rd and 4th terms. The 5th term
ρG dt
(11) represents the hydrostatic pressure forces which depend on
the centroid coordinates for the gas and liquid phases, ξG
2.3 The pressure equation at the jth bubble and ξL, and are a function of the film thickness to the pipe
diameter ratio, hf/D.
For an isothermal flow of an ideal gas, the pressure and the The time rate of momentum inside the film region is
density are related through: troublesome to evaluate. The film region has a constant
pressure and the liquid film behaves as a free surface sub-
1 dρG 1 dPG
= . (12) ject to interfacial waves, see Madani et al. [18]. Due to the
ρG dt PG dt
nature of the applied lumped analysis it is not possible to
Using the definition given in Eq. (12) and adding Eqs. capture wave features, but the time rate of momentum at
(10) and (11), one gets the pressure time ratio within the jth the film region can be estimated performing a momentum
bubble as: balance on the film region only, i.e., at xj−1 and yj bounda-
ries yielding:
dPj Pj
=     .
dt j j j
1−Rf Lf + 1−Rs Ls
j
(13) Lf 
d
   
j−1 j−1 j j j−1 j−1 j j
    
Rs Us −Rs Us + 1−Rs Ub − 1−Rs Ub j j j j j
ρL Rf Uf + ρG 1 − RS UG Adzf + U j ṁ j + U j ṁ j
dt f,y L,y G,y G,y
0
Equation (13) defines the pressure time ratio propor-  
− Uf,xj−1 ṁL,xj−1 + UG,xj−1 ṁG,xj−1
tional to the pressure itself divided by a time scale given      
j j j j j j j j j
by the ratio between the product of cell’s gas fraction times = −g sin (θ ) ρL Rf + ρG 1 − Rf Lf A − τf Sf + τG SG · Lf
the cell’s length (Lf, LS) to the net slug volumetric fluxes 
j j
+ g · cos (θ ) · D · ρL Rf ξL −
ρM j
 
j j

+ ρG 1 − Rf ξG A.
between neighboring cells. 2
(15)
2.4 The mixture momentum equation at the jth cell Inserting the   definition of
 Lf  j j j j j

(15) into Eq. (14) and considering ρG  ≪  ρL, the mixture
d
dt 0 ρL Rf Uf + ρG 1 − RS UG Adzf given in Eq.
The transient gas–liquid momentum equation to the jth cell
is written as:
momentum equation reduces to a liquid-phase momentum
Lf  balance:
d j j j

j
 
j
ρL Rf Uf + ρG 1 − RS UG Adzf
dt
0 d   
LS 
A· ρL Rsj Usj Lsj + Uf,xj ṁL,xj − Uf,yj ṁL,yj
d j
  
j
dt 
+ ρL Rsj Usj + ρG 1 − Rsj Ub AdzS

j j j
dt = Pj − Pj+1 A − ρL RS LS g sin (θ )A − τsj S · LS
0
     j 
+ Uf,xj ṁL,xj + Ub,xj ṁG,xj − Uf,xj−1 ṁL,xj−1 + Ub,xj−1 ṁG,xj−1 Rs j+1 j+1
+ ρL gDcos(θ) · − ξf Rf A. (16)
  
j j

j

j 2
= Pj − Pj+1 A − g sin (θ ) ρL Rf + ρG 1 − Rf Lf
   
j j j
+ ρL RS + ρG 1 − Rsj LS A Equation (16) consistency is checked consider-
j

j j j j

j ing a steady-state model, i.e., with d/dt(ρLRsj Usj Lsj )  = 0,
− τsj S · LS − τf Sf + τG SG · Lf
   ṁL,xj+1 ≡ ṁL,yj and dxj/dt = dyj/dt = Ut. At this condition,
j j j+1 j+1 j j j  
+ g · cos (θ ) · D · ρL ξL Rf − ρL ξL Rf + ρG ξG 1 − Rf
the momentum flux, Uf,xj+1 ṁL,xj+1 − Uf,yj ṁL,yj , becomes
 
j+1 j+1 j+1
−ρG ξG 1 − Rf A  
equal to ρL Rs (Ut − Us ) Uf,yj − Uf,xj A and Eq. (16)
(14)
matches Eq. (29) in Taitel and Barnea [28].
where the LHS represents the transient and the convec- For numerical matters, it is necessary to express
tive momentum terms associated with the jth cell. The Eq. (16) in terms of Us. Substituting the definitions
RHS terms are identified as follows. The first term is the given in Eqs. (5) and (6) into Eq. (16), expressing

13
J Braz. Soc. Mech. Sci. Eng.

Table 1  C0 and C∞ parameters as a function of Re, Fr, Eo and pipe

  
j j+1 j
Uf,xj = dx j dt + (Rs /Rf ) Us − dx j dt and considering

inclination θ
the ratio (Usj   − dyj/dt)/(Usj   − dxj/dt)  ≅ 1, the momentum
C0 values for ReM > 2000 C∞ values for ReB > 200
equation further simplifies to:   
Re = ρM UM D µM
ReB = ρL2 D3 g�ρ µL
2  j 
dx j

d j j j

j j Rs
ρL Rs Us Ls + ρL Rs Us − j+1
−1 Valid for C0 = 1.2 CV∞sen(θ)
dt dt Rf Fr ≥ 3.5
S Valid for C0 = 1 + 0.2·Sin2θ H
C∞ cos(θ) + CV∞sen(θ)
 
j j j
= Pj − Pj|+1 − ρL RS LS g sin (θ) − τsj · LS Fr < 3.5
 j  A
Rs
  
+ ρL gDcos(θ ) ·
j+1 j+1
− ξ f Rf . (17) where Fr = UM gD; H
C∞ = 0.542 − 1.76Eo−0.56 ;
2
V 0.34 �ρgD2
C∞ =  0.58 and Eo =
3805 σ
2.5 Closure equations 1+ Eo3.06

The variables ρG, Ub, τS, dy/dt, dx/dt, RS, Rf, and ξL have to
be determined by closure equations to solve Eqs. (13) and   
(17) in terms of P and US. �ρ
UT = C0 UM + C∞ gD (1 + h(LS )), (21)
ρL
2.5.1 The gas density, the dispersed bubble velocity
and the wall shear stress where C0 and C∞ depend on Reynolds, Froude, Eötvos and
pipe inclination [7] and h(LS/D) is the wake effect func-
The gas density follows the ideal gas law. For isother- tion due to the bubble overtaking mechanism. For turbulent
mal flow ρG = P/RT0 where R and T0 are the gas constant regime, C0 and C∞ result from the contribution of a series
and its temperature. The velocities of the dispersed bub- of authors including Bendiksen [4] and Viana et al. [32];
bles inside the liquid slug, Ub, and the wall shear stress at their values are given in Table 1 as a function of ReM, Fr,
the slug region, τS, are estimated using a mixture model Eo and pipe inclination. The mean C0 values span from 1 to
H
for bubbly flow. Defining the mixture properties, density, 1.2 while the drift parameters, C∞ and CV∞, have an upper
velocity and viscosity as: bound of 0.54 and 0.34, respectively.
In continuous slug flow, the elongated bubbles are sepa-
ρM = ρL RS + ρG (1 − RS ); UM = US RS + Ub (1 − RS ) and
rated by the liquid slugs. If the liquid slug length is short
µM = µL RS + µG (1 − RS ), (18)
enough, the liquid shedding from the leading bubble causes
furthermore, the velocity of the dispersed bubbles comes a wake which induces an acceleration on the trailing bub-
from the drift model, Ub = cbUM + ud which in terms of Us ble. This phenomenon rules the bubbles’ interactions and is
turns to be (apud [27]): responsible for changing the lengths, velocities and frequen-
cies of the elements due to the bubble coalescence. The pio-
cb US RS + ud
Ub =   where cb = 1 and neering study in this area is due to Moissis and Griffith [22]
1 − cb 1 − RS followed by Barnea and Taitel [3], Grenier [11], Fagundes
 1/4
Netto et al. [8], Cook and Behnia [5] and van Hout et al. [31].
 
σ g ρf − ρ G
ud = 1.54 (Rs )1.75 sin θ. (19) The wake function proposed by these authors is expressed as:
ρf2
  
h(LS ) = aw Exp −bw · LS D + cw . (22)
The liquid slug wall shear stress for turbulent regime is
estimated as: The dimensionless parameters aw and bw define, respec-
tively, the intensity of wake and its decaying rate, and the
1 
τS = Cf ρm UM |UM | where Cf = 0.079 Re0.25 M cw is a correction for aw and bw. The coefficients are sensi-
2
tive if the flow is vertical, horizontal or inclined and are not
ρM UM
if ReM > 2,000 and ReM = , (20) universal among the authors, see Table 2.
µM
2.5.2 The bubble nose velocity 2.5.3 The hydrostatic force

The translational velocity of the elongated bubble nose is Considering slug flow in inclined pipes with a plane inter-
defined as: face at the pipe cross section, see Fig. 1, it is possible to

13
 J Braz. Soc. Mech. Sci. Eng.

Table 2  Wake function parameters between the pipe’s cross-sectional area taken by the liquid
Authors aw bw cw Pipe phase and the pressure evaluated at the centroid coordi-
inclination nates, CG, which is expressed as [21]:
Moissis and 8 1.06 0 +90°
2 3 j
   j
Griffith
1 
ξL ≡ CGf = sin − cos , (24)
[22] 3 2 2 2
Barnea and 5.5 6/(Lstab/D) 0 +90°
Taitel [3] 2.5.4 The film holdup, the slug holdup and the
Grenier [11] 0.4 0.5 0 0° displacement of the xj boundary
Fagundes 0.22(1 − LS/Lcrit) 0.16 0 0°
Netto The film profile and the film holdup are determined
et al. [8]
employing the film equation proposed in Taitel and Barnea
Cook and 0.56 0.46 0 +5° [27]. Knowing the film length is possible to estimate the
Behnia [5]
averaged film holdup and the averaged film thickness.
van Hout 0.27 −0.51 (LS/D)−1.4 +10°
et al. [31] The liquid slug holdup is estimated through experimen-
0.75 −0.47 (LS/D)−1 +30°
tal correlations shown in Table 3 for horizontal and vertical
where 10 < Lstab/D < 15 and Lcrit/D = 6.3 pipe flows. As pointed by Al-Safram et al. [1], the accuracy
of the correlations is not good unless the application sce-
nario is close to the experimental condition where the cor-
define an internal angle λ which is related to the liquid film relation came from. More recently, a series of mechanistic
thickness hf through the trigonometric relation: models to predict the liquid slug holdup for vertical flows
  are being developed, for example Guet et al. [13], but none
 = 2cos−1 1 − 2 hDf , (23) of these models were tested in the present model.
Lastly, the term dLS/dt, appearing in Eq. (4), is estimated
and, the liquid film holdup is defined as Rf = (λ − sinλ)/2π. through a gas-phase mass balance at the liquid film region.
The hydrostatic pressure force is determined by the product Combining Eqs. (9) and (12) one gets:

Table 3  Correlations for the slug holdup, RS

Author RS Pipe inclination

Gregory et al. [12] [1 + (UM/8.66)1.39]−1 0°

Malnes [19] 1 − [1 + 83(EoΔρ/ρL)−1/4/Fr]−1 0°
Barnea and Brauner [2] 1 − (0.402Fr6/5Eo0.1C2/5
f,S  − 0.175)
2
90°
Marcano et al. [20] (1.001 + 0.0179UM + 0.0011U2M)−1 0°
Gomez et al. [10] exp[−(2.48 × 10−6Re + 0.45θ)] 0°–90°

where Fr = UM/(Δρ g D/ρL)1/2, Eo = Δρ g D2/σ, Cf,S is the liquid slug friction factor, Re = ρL UM D/μL and 0 ≤ θ ≤ π/2

Table 4  Parameters belonging Parameters Expressions

to the pressure and momentum,
Eqs. (1) and (2) K1 [(1 − Rfj )Lfj  + (1 − Rsj )Lsj ]/Pj
K2 Rj−1 j−1 j−1
S /[1 − cb (1 − RS )]
K3 RSj /[1 − cbj (1 − RSj )]
K4
   
j−1 j
1−Rs 1−Rs
 j−1 −  j
j−1 ud j ud
 
j−1 j
1−cb 1−RS 1−cb 1−RS

K5 ρLRsj Lsj
K6
    2     
 dLsj dt    j  j
dx j dt j
j j
 UM  L
j j
Rs ρL Us j + 1− j
Rs
j+1 − 1 + 2Cf ρρmL j
S 1 
Us Us Rf U D 1−cj 1−Rj
s b s

K7
   
  j+1
 j
    j j

j j+1 Rf LS j j  j  j L 1−Rs
Rs ρL gD cos(θ) 21 − ξf j − D sin (θ) − 2Cf ρm UM ud DS j

j

Rs 1−cb 1−Rs

13
J Braz. Soc. Mech. Sci. Eng.

the interfaces, Eqs. (3) and (4), constitute the set of ODEs
� �� �
j j j j+1
dLS
�
dx j dyj
� 1 − Rs Ub − Ub � �
j+1 j
dt
≡
dt
−
dt
= �
j
� + UT − UT to be solved. Replacing the closure relationships defined in
Rf,xj − Rs
Sect. 2.5 into Eqs. (13) and (17), and comparing their final
form against Eqs. (1) and (2) it is possible to identify the
� � � �
j+1 j+1 j j+1 �
 
j+1
�
1 − Rf L f + 1 − Rs Ls 1 dPG
−  �
j
� 
j+1 dt
. parameters ‘K1’ through ‘K7’ as shown in Table 4.
Rf,xj − Rs PG
The set of Eqs. (1) and (2) is discretized using a Crank–
(25)
Nicolson scheme rendering a tridiagonal system in terms
Equation (25) is complex but it is better appraised con- of P and US for each cell within the pipe. As long as the
sidering RS = 1 and replacing dy/dt by UT in Eq. (25) to get: slug flow pattern endures the evaluated values of P and US
 j+1 j+1  reflects the bubbles–slugs interactions due to changes in the
dx j j+1 Lf dPG pipe inclination and flow transients.
= UT + j+1 dt
. (26) The time–space displacement of yj and xj given in Eqs.
dt P G
(3) and (4) is evaluated using the closure relationships
Equation (26) shows that the bubble tail velocity lags the defined in Sect. 2.5 and a first-order integration routine.
bubble nose velocity due to the gas expansion. The expan-
sion effect on the bubble tail velocity becomes negligible 3.1 The numerical algorithm
as the pressure increases. Identical expression is obtained
observing that the mass of gas within the elongated bubble The j index is used to track each individual cell as they
is constant because RS = 1. flow downstream the pipe by a FORTRAN object-oriented
program. The cell index range is 0 ≤  j  ≤  N where j  = 0
2.6 Model’s closing remarks applies to a cell which is being inserted into the pipe,
1  ≤  j  ≤  N  − 1 to the cells inside the pipe and j  =  N to
The proposed slug tracking model embodies the features the cell which exits the pipe. The numerical procedure
of the previous models and adds the advection terms. The has three stages. The first evaluates the parameters ‘Κ1’
influence of the advection term is postponed to Sect. 5.6. through ‘K7’ using the available slug and bubble properties
The inclusion of the advection term is an attempt to enhance at the present time step. The second solves the tridiagonal
the model’s flow representativeness. But, the use of a more matrix to get P and US at the next time step. Finally the
sophisticated model does not mean that it becomes less third advances the boundaries xj and yj to the next time step
dependent on closure laws and on the slug insertion pro- updating the positions of the liquid slugs and elongated
cess at the inlet. The present model, likewise the previous bubbles.
models, relies on experimental input, such as the liquid slug
holdup or the cell’s frequency, the bubble nose velocity, the 3.2 Initial and boundary conditions
liquid slug hold up and the wake law. The one-dimensional
approach allows interactions along pipe axial direction only The processes of insertion and removal of bubbles and
and misses important 3D flow effects mainly present on the slugs at the pipe define the boundary conditions. Before
wake law, on the aeration processes at the liquid slug and at inserting the first cell, the following conditions apply: the
the bubble nose region. The 3D flow field effects are trans- averaged phases’ flow rates are known, the outlet pressure
mitted to model through experimental correlations. is known and the pipe is full of liquid, as if the entire pipe-
The major simplification done on the present model was line length is taken by a single non-aerated slug, flowing at

on the hypothesis of ρG  ≪  ρL. This simplification, also

to neglect the gas momentum equation on Eq. (16) based the given liquid-phase averaged velocity.
The insertion of numerical cells occurs at the numerical
used by the previous models, limits the model application domain inlet which is coincident with the first experimental
for high pressure or dense gas scenarios. To overcome this measuring station positioned at 127 D downstream of the
limitation, one can work with Eq. (15) but the algebraic air–water mixer, see Fig. 2. At the initial time, the first cell
expressions will increase in complexity. Complementary, to be inserted has its front positioned at z = 0, and labeled
the model application to a large diameter pipe is natural as as (j = 0), see Fig. 1. The liquid slug insertion occurs at a
long as the flow is in the slug regime. constant velocity UM with US and RS given. It lasts until the
whole liquid slug is inside the pipe and the gas bubble nose
is at yj = 0 = 0. The gas bubble insertion occurs at P and UT
3 The numerical model constants. The pressure is defined at the slug tail with Uf, Rf
already defined. It lasts until the whole bubble is inside the
The pressure and momentum equations represented by Eqs. pipe. At this instant, the cell is labeled as (j = 1), its proper-
(1) and (2) together with the time–space displacement of ties are evaluated by the tridiagonal system and a new cell

13
 J Braz. Soc. Mech. Sci. Eng.

(a) experimental setup (b) parallel stream mixer

measurement stations
i
air
air line air
S1 S2 S3 S4
Patm vent
water
127 D 267 D 494 D 777 D 4D
water
t liline mixer
i D
900 D
liquid
Corioli storage (c) concentric stream mixer
flow meter tank
D
laminar flow air .....
.....
.....
T .....
element compressed
∆P P T air supply porous
water stone

Fig. 2  Schematic of the experimental facility with a horizontal test section

insertion process begins. The cells are removed at the pipe described as normal or log-normal, or even worse, the
outlet. When the slug front of the (j = N) cell touches the distributions are bimodal.
pipe outlet, xj = N = L, this cell is removed from the matrix Very often the estimated slug flow distributions at the
system and its properties frozen. A detailed description of inlet do not accurately represent the experimental distribu-
the cell insertion and removal is in Rodrigues [25]. tions. Inaccurate inlet distributions propagated by the slug
The slug tracking models intrinsically require the defi- tracking model may lead to erroneous downstream distri-
nition of LS, Lf, f, RS, Rf and UT at the inlet. In addition, butions. The question is: the predicted slug population bias
it is desirable that the defined inlet properties satisfy the is due to the model’s lack of physical representativeness or
averaged inlet mass flow rates. Unfortunately, the accom- to the inaccuracy of the inlet slug populations? To answer
plishment of these estimates constitutes a difficult task. this question, this model uses experimental slug properties
Barnea and Taitel [3] proposed an approximated relation- at the inlet. The individual values of Lf, LS, f and UT are
ship between LS and LF as a function of JG and UT. Cook experimentally determined for a sample of 300–600 cells.
and Behnia [5] add to Barnea and Taitel [3] relationship a This information is used with a unit-cell model to deter-
dependence on the liquid film holdup. This work employs mine the liquid film holdup and the liquid slug velocity for
the Taitel and Barnea [27] steady-state model to get the each cell. The estimated cell’s properties: Lf, LS, UM, RS
inlet slug properties satisfying the averaged mass flow and Rf are stored in a matrix and used the inlet data. This
rates. method is called random insertion process based on experi-
The variability of the slug flow properties at the mental data or simply random process.
numerical inlet is estimated by a random method to gen-
erate populations of lengths, holdups and velocities [3, 3.3 Bubble coalescence
11, 23, 25, 26, 34]. Frequently, the used distributions are
the normal and the log-normal based on fully developed The bubble coalescence is a phenomenon when the trail-
slug data [24, 30], but also are employed ad hoc distribu- ing bubble merges with the leading bubble. The bubble
tions such as the exponential [29], slug lengths uniformly overtaking mechanism is due to the wake effect of the
distributed [23] or a simple constant inlet value referred leading bubble on the trailing one. After coalescence, the
as a periodic insertion [23]. At the inlet, the authors mass of the new bubble is equal to the sum of the masses
adopt ad hoc slug lengths’ intervals spanning from 2D of the original bubbles. The existing liquid slug between
to 8D [5, 23, 34] and use a random generator, bounded the bubbles disappears due to the liquid-phase transfer to
by a maximum and minimum LS interval, to estimate the the upstream slug. When coalescence occurs the program
LS population. The other variables’ populations are esti- is halted, the cell numbering is updated and the program
mated after the LS population. The flaw of this procedure resumes.
consists that actual slug flow properties, near the inlet,
are dependent on the discharge conditions of gas–liq-
uid mixer. The slug regime forms within a 60 D–100 D 4 Experimental apparatus
downstream the mixer, the neighboring bubbles have a
strong coupling because the slugs are short, the coales- A series of experimental runs were performed to determine
cence rate is high and the distributions are not exactly the statistical distributions of Lf, LS and UT as well as its

13
J Braz. Soc. Mech. Sci. Eng.

Table 5  Experimental test grid Run JL (m/s) JG (m/s) UM (m/s) ReM (–) FrM (–) Mixer type
and reference values
#1 0.33 0.60 0.93 26,000 2.0 PS
#2 0.33 1.35 1.68 43,000 3.3 PS
#3 0.67 0.65 1.32 35,000 2.7 PS
#4 0.66 1.30 1.93 52,000 4.0 PS
#5 0.67 0.60 1.27 35,000 2.7 CS

averaged values, the average the pressure drop, the coales- air, respectively. The analogical signals of each measuring
cence rate and the cell frequencies. The data are taken in a station were sampled at 3 kHz, digitized and stored by a
test section, depicted in Fig. 2a, with a 26 mm ID straight, National Instruments data acquisition system. The bubble’s
horizontal and transparent Plexiglass pipe with 900 D long, nose velocity of each elongated bubble is determined by
i.e., 23.4 m. The working fluids were compressed air and the time interval required for the interface to move the dis-
ordinary tap water. Air compressors and a centrifugal pump tance S from one probe to the other. The liquid film and the
supply the air and the water to the mixer installed at the slug lengths were determined by multiplying the residence
entrance of the test section. Two types of air–water mixers time of the gas bubble or of the liquid slug by the bubble
are employed: a parallel stream and concentric stream mix- nose velocity. The frequency of each cell, f, is evaluated as
ers, hereafter named as PS and CS mixers to study the slug the reciprocal of the cell residence time. The signal post-
flow development at distinct slug formation processes, see processing gives out the LS, Lf, UT and f belong to each
Fig. 2b, c. At the end of the test section, the mixture is dis- cell. The number of coalesced bubbles per pipe length is
charged into a receiving tank open to the atmosphere. The assessed defining a local coalescence rate as:
ordinary tap water is the working liquid phase with den-  
1 Nb,d − Nb,u 1 dNb
sity and viscosity of 998 kg/m3 and 0.001 Pa s. The liquid C(z) = − Lim ≡ , (27)
Nb,u �x→0 �x Nb,u dx
flow rate was measured by a Metroval Coriolis mass-type
flow meter accurate within 1 %. A Merian laminar flow where Nb,d and Nb,u represent the total number of bubbles,
element is used to measure the air flow rate with reported at a given time interval, crossing two pipe cross sections
uncertainty of 1½ %. The range of water and air superficial spaced of Δx. Experimentally is not possible to determine
velocities spans, respectively, between 0.25–1.35 and 0.4– C(z) as Δx → 0, but its average value for pipe cross sec-
1.7 m/s at ambient pressure and temperature of 94.7 kPa tions spaced by a finite distance ΔL is evaluated from Eq.
and 23 °C. (27) as
The test section has four measuring stations located at   
Ln Nb,d Nb,u
127D, 267D, 494D and 777D downstream the air–water Cav (z̄) = − , (28)
mixer. Each measuring station has a pressure transducer, �L
Smar LD series, and a twin set of double wire conductive where z̄ represents the distance from the origin to the mid-
probes axially spaced of S  = 50 mm and in-house made. distance to the cross-sectional planes.
The pressure transducers measure a gauge pressure with Samples with 300–600 values of LS, Lf, UT and f are
one tap connected at the bottom of the test section pipe recorded employing a data acquisition for a period of 360 s.
while the other tap is open to the atmosphere. The double This sampling size was considered representative of the
wire probe consists of a pair of gold wires with 100 μm populations. The experimental data set consists of five runs
stretched along the pipe diameter and aligned vertically. with operational conditions shown in Table 5. The liquid
The probe picks up the variation of the air and water elec- and gas volumetric fluxes, JL and JG, are defined by the
trical conductance between the parallel wires [16]. The ratio of the volumetric flow rate of the respective phase by
circuit is driven by a 12-kHz oscillator. The output signal the pipe cross-sectional area. Runs#1 to #4 employ a gas–
is amplified and filtered with a cutoff frequency of 8 kHz liquid mixer with parallel streams, PS, and have JL and JG
to remove the carrier frequency. The liquid conductivity is progressively increasing. run#5 has the same operational
sensitive to changes on the temperature or on the water pH. conditions of run#3, but exhibits a distinct slug forma-
To exclude these undesirable changes from the measured tion process because employs a concentric stream mixer,
data, a slave probe is used to measure only the conductance CS. Flow visual inspection disclosed non-aerated liquid
changes due to temperature or water pH changes. slugs, for the given set of experimental velocities, therefore
The probes’ twin voltage signals resemble square waves RS = 1 for runs #1 thru #5.
shifted in time due to the probes’ spacing. The high and low The uncertainty to LS and Lf is estimated in ±1 D due to
values correspond to the occurrence of the water and the the interface shape separating the bubble tail from the slug

13
 J Braz. Soc. Mech. Sci. Eng.

Table 6  Uncertainty estimates Table 7  Slug tracking setup configurations

Run δLS/D (–) δLf/D (–) δUT (m/s) δf (Hz) δP (kPa) (i) Default
(ii) Experimentally adjusted aw and bw coefficients in Eq. (22)
#1 ±1 ±1 ±0.01 ±0.03 ±0.2
(iii) Periodic insertion uses constant values of Lf, LS, US and UT
#2 ±1 ±1 ±0.02 ±0.03 ±0.2
#3 ±1 ±1 ±0.01 ±0.50 ±0.2
#4 ±1 ±1 ±0.03 ±0.10 ±0.2
#5 ±1 ±1 ±0.01 ±0.50 ±0.2 1.E-02
ε L ~ ∆t1/2

RMS of ε L
head which may exhibit a stair case or a sharp jump shape
depending on Froude regime, Fagundes Netto et al. [9]. 1.E-03
The uncertainty on the bubble nose velocity depends on the
estimation of the elapsed time, Δt, during the passage of
the bubble nose by two successive wires axially space of
a distance S. The uncertainty of Δt is twice the reciprocal 1.E-04
of the signal’s sampling frequency, 1/fs. The uncertainty of 1.E-05 1.E-04 1.E-03
axial distance S is considered negligible. Thus, the uncer- ∆t (s)
tainty of the bubble nose velocity is:
Fig. 3  RMS of the relative mass imbalance as a function of the time
step for run#3
 
δUT = ± 2UT2 (S · fs). (29)

The cell’s frequency uncertainty, δf, depends on the εk = �ṁk ṁk , (31)
reciprocal of the cell’s residence time which is also equal
to the ratio between the bubble nose velocity, UT, to the cell where k is the phase indicator, k = L or k = G, ṁk is mean
length, LU  =  LS  +  Lf. The contribution of UT to the δf is inlet mass flow rate of phase k and ṁk represents the phase
considered much smaller than LU, therefore mass imbalance on the pipe total volume at the instant t:
2f M�t+1 − Mtk 1 1
δf = ±  . (30) k
+ ṁ�t+t + ṁtk out − ṁ�t+t + ṁtk in .
 
�ṁk (t) =
LU D �t 2 k 2 k

Lastly the gauge pressure uncertainty, δP, is considered (32)

as the sum of the instrument calibration procedure ±2 mm The subscripts ‘in’ and ‘out’ refer to the instantaneous
of water and the half of the pipe diameter water column, phases’ mass flow rates at the pipe inlet and outlet sections,
±13 mm, due to the intermittence of the liquid film and of Δt is the time step and Mk is the mass of phase k distrib-
the liquid slug; the result is δP = ±0.2 kPa. The absolute uted along the pipe. Mk is easily evaluated by summing the
uncertainty bounds are shown on Table 6 for convenience. phase k mass within each cell inside the pipe. The relative
mass imbalance for the liquid phase is shown in Fig. 3 as
a function of the time step for run#3. For example, a time
5 Data analysis step of 0.001 s the liquid mass imbalance within the pipe is
0.4 % of the inlet liquid mass flow rate. The convergence
The slug tracking response is analyzed along this section. rate is proportional to the square root of the time step. Sim-
The model default setup configuration employs: time step ilar results are found to the gas-phase mass imbalance.
of 0.001 s, Eq. (21) for UT, Eq. (22) for the wake function
with the coefficients given by Grenier [11], RS = 1 and ran- 5.2 Predicting averaged data
dom insertion process based on experimental data, see Sect.
3.2. Other setup configurations may also be used and their This section assesses the slug tracking performance by
distinct features from the default setup are listed in Table 7. comparing the numerical outputs against the averaged
experimental data for runs #1 thru #4. The water and the air
5.1 Mass imbalance and convergence rate of the slug superficial velocities span from 0.33 to 0.67 m/s and 0.60
tracking model to 1.35 m/s. The gas to liquid ratio changes from 1:1 to
4:1 and the mixture Re ranges from 26,000 to 52,000, see
The slug tracking relative mass imbalance, εk, within the Table 5. The slug tracking setup configuration is the default
pipe total volume is assessed by: configuration, see Table 7. The choice for setup (i) lays on

13
J Braz. Soc. Mech. Sci. Eng.

0.25 5 20 80

0.20 4
15 60
Cav (m-1)

0.15 3

LS/D
Hz)

Lf /D
D
10 40

f (H
0.10 2
5 20
0.05 1

0.00 0 0 0

0.25 5 20

120
0.20 4
15
m-1)

0.15 3 80
f (Hzz)
Cav (m

Lf /D
D
LS/D
10
0.10 2

5 40
00.05
05 1

0 00
0.00 0 0 0

0.25
0 25 5 20 80

0.20 4
15 60
0.15 3
Cav ((m-1)

Hz)

LSS/D

Lf /D
10 40
f (H

0.10 2

0.05 5 20
1

0.00 0 0 0

0.25 5 20 80

0 20
0.20 4
15 60
00.15
15
Cavv (m-1)

3
f (Hz)

LS/D

10 40

Lf /D
(

0.10 2

0.05 5 20
1

0.00 0 0 0
0 200 400 600 800 0 200 400 600 800 0 200 400 600 800 0 200 400 600 800
z/D z/D z/D
z/D

Fig. 4  Averaged experimental data (open symbols) and slug tracking (dashed line) comparison for runs #1 to #4 from top row to bottom row

Table 8  Experimental data for runs #1 to #4 open diamonds and the numerical predictions by a dashed
Run#1 Run#2 Run#3 Run#4 line.
The coalescence rate has a strong dependence on JL and
Cav-exp (1/m) 0.025 0.045 0.130 0.100 a weak dependence on JG. Doubling the liquid velocity
Lf inlet/D 45 70 10 25 more doubles the coalescence maximum. The experimen-
ΔLf-exp/Lfinlet +0.11 +0.36 +1.00 +0.60 tal Cav always decreases with the distance from the inlet
LS inlet/D 14 9 4 6 while the predicted Cav has a maximum except for run#3.
ΔLS-exp/LSinlet −0.29 −0.78 +2.25 +1.17 The experimental and numerical maximum value of Cav
finlet (Hz) 0.7 0.8 4.8 2.5 does not match. The use of Grenier [11] wake function
Δfexp/finlet −0.29 −0.25 −0.58 −0.60 tends to overestimate the Cav at large distances from the
inlet. Despite these differences, the model still captures Cav
Cav refer to the maximum coalescence rate. ΔL/L means the relative
difference between the output to input value of Lf, LS and f. The (inlet) experimental trend within ±5 % of bubbles per meter. The
means Lf, LS and f evaluated at z/D = 127 predicted slug frequency matches the experimental data for
runs #1 and #2 (low JL) and underpredicts the experimental
values for runs #3 and #4 at pipe axial locations far from
the use of wake function and UT correlations already avail- the inlet. The LS trend is captured within ±4D with a ten-
able in the literature. dency to underpredict the experimental values. Finally, the
The averaged data of Cav, f, LS, and Lf as a function of Lf growth tendency with the distance from the mixer is cap-
the pipe axial distance are shown in Fig. 4 along columns tured but at the z/D of 777 D the offset is over 100 % of the
which represent, in a sequence, run#1 (top line) thru run#4 experimental value. Surprisingly, a poor estimate of Lf does
(bottom line). The experimental data are represented by not imply in a poor estimate of LS.

13
 J Braz. Soc. Mech. Sci. Eng.

(a) (b)
50 16
40 12
30

LS /D
D
Lf /D
D

8
20
4
10
0 0
0 200 400 600 800 0 200 400 600 800
z/D z/D

Fig. 5  Track history of the 400th cell for run#3: film length (a) and slug length (b)

For referencing purposes, Table 8 displays the maximum its leading bubble twice, i.e., the bubbles belonging to the
Cav, the inlet, and the outlet to inlet relative difference of 399th and 398th cells. The overtaking process between the
Lf/D, LS/D, f regarding the experimental data. Runs hav- 400th bubble and its downstream neighbors makes the liq-
ing high coalescence peaks exhibit the largest frequencies uid slug separating the neighboring bubbles to diminish in
at the inlet and the largest ΔLS/LSinlet. The ΔLf/Lfinlet is not length due to the mass transfer to the upstream cell. After
sensitive to the coalescence peaks. the coalescence, the mass of liquid slug of the previous
It must be remembered that the slug tracking accuracy 399th cell is added to the liquid slug of the 400th cell. This
is equivalent to the accuracy of the steady-state slug mod- process is shown in Fig. 5b by the LS jumps occurring at
els because both have their roots in the same closure laws. the same z/D of 530 and 800. For computational reasons,
Lastly, it is observed that the slug tracking does capture the bubble coalescence is effective when the liquid slug
the axial evolution of the flow average properties, but length separating two bubbles is of 1 pipe diameter instead
its usefulness lays on the capture of the flow properties’ of shrinking its length to zero.
distribution. This subject is approached in the following The jumps in Lf and LS are a signature of coalescence.
sections. Considering the scenario where the cell is tracked along a
pipe with constant cross section with no bends nor changes
5.3 The overtaking mechanism and the model’s in its inclination, it is possible to say: there is no bub-
sensitivity to the wake function ble breakage, the liquid film length always increases and
the slug length may increase, decrease or even disappear.
This section describes the liquid piston length changes due Therefore, the liquid slug is subject to a greater variability
to the overtaking mechanism until the event of the bubble in length than the elongated bubble. This behavior explains
coalescence. Also, the section explores the numerical solu- why the slug length population is more quickly established
tion sensitivity to the wake function coefficients as com- than the elongated bubbles’ length population and helps to
pared against the experimental data. understand the additional LS growth and jumps displayed in
Fig. 5b at z/D of 190 and 425 not present on the Lf record.
5.3.1 The overtaking mechanism They are due to two coalescences happening to the 399th
bubble with its downstream neighbor, or the 398th bubble,
It is chosen to track the bubble nose front belonging to not shown in the figure. During this process, the mass of
the 400th cell entering the pipe out of a list of 600 cells in the liquid slug of the 399th was transferred to the upstream
run#3 employing setup (ii). Considering run#3, the test sec- slug (the 400th cell) making it grow. At the z/D positions
tion has on average 22 cells inside, the use of the 400th cell of 190 and 425, the observed jumps with amplitude of 1D
guaranties that slug regime is fully established, i.e., there is on the 400th LS record means that the 399th and the 398th
no effect of the flow start-up process. A frame of reference bubbles reached the minimum distance of 1D, the volume
moving with the bubble nose of the 400th cell identifies the of liquid separating the bubbles is instantaneously trans-
downstream neighbors’ cells as the 399th and 398th cells. ferred and the bubble coalescence took place.
The 400th cell’s film and slug lengths are shown in Fig. 5a, Not shown in Fig. 5, it must be in mind that the changes
b along the pipe axial distance. Along the analysis, the cell on the liquid slug length occur due the velocity difference
numbering will not change for convenience. Inspecting between neighboring bubbles, therefore the coalescence
Fig.  5a, it is observed two jumps on Lf at z/D of 530 and rate is also perceived by an increase of the bubble nose
800. These jumps occur because the 400th bubble overtakes velocity.

13
J Braz. Soc. Mech. Sci. Eng.

0.25 30 15
Exp
E
Exp setup (i)
0 20
0.20 setup (i) setup (ii)
setup (ii) setup (iii) 10
0 15
0.15 setup (iii) 20
Cav (1/m))

Lf/D

LS//D
0.10
10 5 Expp
setup (i)
0.05 setup(ii)
setup (iii)
0.00 0 0
0 200 400 600 800 0 200 400 600 800 0 200 400 600 800
z/D z/D z/D
5 1.6 110
Exp Exp
4 setup (i) setup (i)
setup(ii)
t (ii) setup(ii)
setup (iii) setup (iii)
3
f av (Hzz)

UT ((m/s)

P (kPa)
1.4 105
2 Exp
setup (i)
1 setup(ii)
set p (iii)
setup
0 1.2 100
0 200 400 600 800 0 200 400 600 800 0 200 400 600 800
/D
z/D z/D
/D /D
z/D

Fig. 6  Averaged slug flow properties compared against experimental on top row: coalescence rate, film length and slug length; bottom row
data for run#3 as function of the pipe axial distance. Numerical data cells’ frequency, bubble nose velocity and pressure drop
employs the setup configurations (i) (ii) and (iii). From left to right

5.3.2 The sensitivity of the slug flow properties on the Table 9  Differences between predicted and experimental values at
wake function z/D = 777D for setups (i) and (ii)

Setup (i) Setup (ii)

The influence of the wake law on the flow development is
Lf Overprediction 100 % Overprediction 33 %
assessed employing distinct coefficients aw and bw. While
LS Overprediction 17 % Overprediction 30 %
setup (i) employs aw and bw proposed by Grenier [11],
f Underprediction 33 % Match
setup (ii) uses aw and bw of (0.15, 0.45) based on a best fit
ΔP/L Underprediction 26 % Underprediction 18 %
against the experimental data. Figure 6 compares the pre-
dictions of the averaged slug flow properties against run#3
data. The x axis corresponds to pipe axial distance from the
air–water mixer in terms of multiples of the pipe diameter. not successful to match the experimental data from inlet to
The top row displays the coalescence rate, Cav, followed by outlet. This behavior suggests that a wake law with just two
the averaged film and slug lengths. The bottom row shows coefficients may not perform equally well to the regions
the averaged values of the cell frequency, the bubble nose near and far away from the mixer.
velocity and the pressure drop. Table 9 briefs the differences between setups (i) and (ii)
Focusing on the coalescence data, it is observed in Fig. 6 observed in Fig. 6. The numerical values express the differ-
that the setup (i) renders a monotonically decreasing value ences between the predicted and the experimental values at
for Cav as z/D increases, overpredicts the experimental Cav z/D = 777. The outcomes are overpredictions or underpre-
values at z/D = 200, but nearly matches experimental data dictions despite the use of distinct aw and bw. Apparently,
for z/D of 390 and 640. The setup (ii) gives a coalescence the overprediction of 100 % on Lf/D estimate employing
peak at nearly z/D = 200, and almost matches the Cav pro- setup (i) did not influence the outcome of LS/D. The aw and
file found in setup (i) for z/D of 390 and 640. The main bw coefficients employed on setup (ii) match the experi-
difference between setups on Cav is the fact that setup (i) mental frequency, improved the Lf/D and ΔP/L estimates
approaches the maximum experimental from above while in comparison to the result from setup (i) but got a worse
setup (ii) approaches from below. result than setup (ii) for LS/D. Surprisingly, the setup (i)
The use of setup (i) resulted in poor prediction of Lf has a mean pressure gradient lower than setup (ii) despite
and f but a good match of LS at stations far from the inlet. having larger slugs than setup (ii). This is due to the fact
Complementary, setup (ii) gave a better prediction to Lf and that setup (i) also overpredicts Lf. Therefore, the number of
f and captured the LS near the inlet. Additional trials, not cells inside the pipe is less than the one observed for setup
shown here, were performed seeking a better LS but were (ii).

13
 J Braz. Soc. Mech. Sci. Eng.

0.12 0.3 8
Exp Exp Exp
setup (i) setup (i) setup (i)
meaas. staationn #1

setupp ((ii)) setup (ii) 6 setup (ii)

0 08
0. setup (iii) 0.2
0 2 setup (iii)
setup (iii)
4
0 04
0.04 00.1
1
2

0.00 0.0 0
0 10 20 30 40 50 60 70 0 10 20 30 1.0 1.2 1.4 1.6 1.8 2.0
Lf/D LS/D UT (m/s)
0.24 0.3 16
Exp Exp Exp
setup
p ((i)) setupp ((i)) setup (i)
meas. statiion ##4

setup (ii) setup (ii) 12 setup (ii)

0.16 p ((iii))
setup 0.2 setup (iii) setup (iii)
8
0.08 0.1
4
m

0.00
0 00 00.00 0
0 10 20 30 40 50 60 70 0 10 20 30 1.0 1.2 1.4 1.6 1.8 2.0
Lf/D LS/D UT (m/s)

Fig. 7  Distributions of the experimental and numerical values of Lf, LS and UT for run#3. Top row distributions at the measuring station #1. Bot-
tom row distributions at the measuring station #4

Finally, the UT is not solved by the model but comes 22.8D, respectively. Surprisingly, the best prediction comes
from a kinematic relation given in Eq. (21). The predicted from setup (ii) which exhibits the poorest prediction of the
UT matches the experimental data, within uncertainty averaged slug length value!
bounds, z/D > 400, see Fig. 6. Specifically for setup (i), is The data comparison gives an appraisal of the slug flow
observed a peak on UT, at distances shorter than z/D = 400. sensitivity to the wake law coefficients. The predictions for
The offset is due to a high coalescence rate prediction the Cav employing both setups approach the experimental
which induces high velocities thus increasing the average. Cav peak from above and from below resulting in slug flows
The inlet distributions associated with setups (i) and with distinct bubble and slug lengths, frequencies and pres-
(ii) displayed at the top row of Fig. 7 are almost coinci- sure drop. Even though the model follows the trend of the
dent with the experimental distributions. This is a conse- experimental data the differences against experimental data
quence of a statement given in Sect. 3.2: the numerical inlet are still large. Also, some flow features still do not have an
conditions are set as close as possible to the experimental objective answer due to the multiple ways that the change
data from station #1. As the populations of Lf, LS and UT in the lengths and velocities takes place. For example, the
flow downstream they change in shape arriving at station relative success in predicting LS/D distributions at station
#4 with a shape shown at the bottom row of Fig. 7. At sta- #4 is not necessary tied to a good representativeness of the
tion #4, the predicted Lf/D distributions are at the right of Lf/D distributions as observed from the experimental data.
the experimental Lf/D distribution and have their mode Also a good match to the experimental frequency does not
and standard deviation greater than the experimental dis- mean a good match to LS/D prediction. In addition, was
tribution. The introduction to the adjusted coefficients aw tested but not shown here, that the absence of the wake
and bw in setup (ii) brought to the Lf/D distribution mode function there is no coalescence, i.e., the inlet population
closer to the experimental data and reduced the spread. The remains unchanged as the flow evolves downstream.
LS/D population from setup (i) almost matches the experi-
mental data while the population from setup (ii) keeps the 5.4 Slug flow sensitivity on the inlet distributions
same spread but has an offset to the left of the experimental
data. Finally, the velocity population from setup (i) and (ii) The cell insertion process disturbs the flow downstream
almost match each other but both have an offset from the due to the mass and momentum sources defined at the inlet.
experimental population due to the uncertainties associated The influence of the inlet distributions on the downstream
with the experimental data and to the use of Eq. (21). flow is assessed comparing random and periodic insertion
One piece of information of practical interest is the pre- processes embodied on setups (i) and (iii). As observed in
diction of the largest slug at z/D = 777D, here considered Fig. 6, the Cav profile in setup (iii) peaks is at z/D = 300,
the P(LSmax) = 0.990. The largest slug length for the exper- then reaches a minimum at z/D  = 450 to start increasing
imental setups (i) (ii) and (iii) were of 24.3D, 29.9D and again. It is noteworthy the difference on the Cav curve when

13
J Braz. Soc. Mech. Sci. Eng.

Fig. 8  Averaged value comparisons for run#3 and run#5 employing parallel and concentric stream mixers, or PS and CS. Top row Cav, Lf, and
LS. Bottom row f, UT and P

using setup (i) or (ii) and setup (iii). While the formers fol- Complementary, the inlet distributions may alter signifi-
low the physical trend having a maximum and a decaying cantly the slug flow development distributions near the
rate the later does not exhibit this behavior. Nevertheless, inlet.
the model using a periodic inlet, almost matches the experi-
mental Cav at z/D of 380 and 636. It is considered a acci- 5.5 Comparing flows with distinct slug formation
dental coincidence. Complementary, the slug flow proper- process
ties resulted from this hypothetical inlet distribution evolve
downstream into a population with averaged values of Lf, This section compares two experimentally determined inlet
LS, f and UT at station #4 unexpectedly close to the values populations against the numerical predictions. The analy-
found in setup (ii). ses complement the analyses performed in the last section
The success of the averaged data prediction is attrib- where an hypothetical inlet distribution is compared against
uted to the fact that the bubbles just increase in size due the experimental data. Both analyses address to the flow
to coalescence but the slug length can decrease or increase loss of memory to the inlet conditions.
or even extinguish. As the flow evolves downstream it is The experimental setup employs a parallel and a con-
expected that a large variety of liquid slugs’ lengths but centric stream mixer as depicted in run#3 and run#5, see
the liquid film, with less degrees of freedom, is repre- Table 5. For the record, the numerical simulations in this
sented by a bimodal distribution. Once the LS/D pdf gets a section employ the setup configuration (ii). The averaged
shape approximately similar to the experimental data, the values of Cav, Lf, LS, f, UT and P are shown in Fig. 8 as a
frequencies and the pressure drop will follow, see Fig. 6 function of the inlet distance. Figure 9 displays the popu-
bottom row. Lastly, the bubble nose velocity UT is overpre- lations of Lf, LS and UT as they evolve from station #1
dicted only in a region centered at z/D  = 300, see Fig. 6 thru #4. The black and the gray lines represent data from
bottom row. The model predicts large coalescence rate hap- PS and CS mixers; the continuous and the dashed lines
pening at z/D = 300 which induces larger UT velocities that represent the experimental and the numerical data.
increase, locally, the averaged velocity. This a proof that The PS and CS experimental averaged values shown in
the coalescence rate is also perceived by an increase of the Fig. 8 display distinct values of Cav, LS and f at z/D = 196 but
bubble nose velocity as pointed in Sect. 5.3. nearly the same values at z/D of 380 and of 636. The Lf and
Figure  7 data suggest that even periodic insertion data P averaged values are coincident from station #1 thru #4. The
propagated downstream may create a population which differences in UT values result from the distinct gas super-
eventually matches the experimental population if there is ficial velocities employed; while run#3 has JG of 65 cm/s,
enough pipe length to allow the bubble to bubble interac- run#5 uses 60 cm/s. As before, the numerical predictions fol-
tions take place. In other words, it seems that the flow loses low the experimental data trend but, the effect of distinct slug
the influence of the inlet as the flow evolves downstream. formation is better observed analyzing the distributions.

13
 J Braz. Soc. Mech. Sci. Eng.

Fig. 9  Distribution comparisons among experimental (runs #3 and #5) and numerical data employing parallel stream and concentric stream
mixers, PS and CS

A glance at the first row of Fig. 9 discloses that at sta- experimentally determined LS/D distribution, the Lf/D dis-
tion #1 the LS and UT distributions arising from for PS and tribution follows the same reasoning. Comparing the LS
CS mixer are distinct. The lack of match between the inlet population at measuring stations #1 and #4 one observes
populations of UT to the numerical model is because the UT at station #1 an almost symmetric population with lengths
estimated by the model comes from published experimental spanning from 1D to 10D with mean in 5D, and at station
correlation, see Eq. (21). As the flow evolves to the station #4 a symmetric population spanning from 2D to 24D with
#4 the experimental Lf and LS distributions arising from the a mean in 13D. It is observed that the mean slug length of
PS and CS mixers become almost superposed. The distri- 5D at station #1 has to grow more than twice its length to
butions match at station #4 means that the influence of the get the mean slug length at station #4. Therefore, it is likely
inlet process ceased or the memory of the slug formation that most of bubbles merged more than once along the pipe
faded away. The slug tracking model equally predicts Lf to result in a slug population with large slugs at station #4.
and LS distribution match, is but with a shape close to the The slug length change from station #1 to #4 is briefed as
experimental distributions. follows. At the pipe entrance, between stations #1 and #2,
The changes of distributions’ shape of Lf/D and LS/D the coalescence rate extinguishes a fraction short slugs
observed between stations #1 to #4 are justified ana- which more likely have the faster trailing bubbles. This
lyzing Fig. 9. As an example it is chosen to follow the merging process creates population with intermediate

13
J Braz. Soc. Mech. Sci. Eng.

lengths which are subjected to subsequent coalescence to (i). Figure 10 shows (a) the time history of the bubble pres-
form large slugs at station #4. sure and liquid slug velocity, P and US (b) the time history
Inspecting the experimental UT populations shown in the of the bubble pressure and slug velocity fluctuations, P′ and
3rd column of Fig. 9 and the Cav in Fig. 8, an increase in the US′ (c) the P′ and US′ phase diagram and (d) the P′ and US′
UT standard deviation at station #2 as compared to ones in power spectral density (PSD). The symbols ‘A’ and ‘NA’
stations #1 and #3 is observed. This increase in the stand- stand for with and without advection term.
ard deviation is followed by the highest coalescence rates Despite the models with and without advection terms
which means that from station #1 to station #2 were created have the same inlet list of cells the history of each individ-
faster and slower bubbles due to the bubble to bubble inter- ual bubble is distinct due to the lack of the advection term.
action. Once again, the coalescence rate is also perceived Figure  10a shows P and US instantaneous values as the
as a change in the velocity population. It is also noted that 400th cell travels downstream in the pipe. The signals fluc-
after station #2, the UT standard deviation just decreases for tuate in time, P decreases due to the flow frictional pressure
both CS and PS mixers. The slug tracking model could not drop while the US slightly increases due to the gas expan-
capture the experimental velocity distribution increase of sion. The observed change in the residence time from 13 to
standard deviation observed at station #2. 10 s does not mean that 400th cell speed increased but it is
The loss of the memory of the slug formation is an due to multiple coalescence happening at the 400th cell.
important physical flow feature captured by the wake law. The P′ and US′ time fluctuations, see Fig. 10b, are larger
This property assures that even hypothetical slug distribu- when the advection term is present. A typical peak to peak
tions eventually converge to a developed slug distribution, values for P′ and US′ when the advection term is present
as observed in Sect. 5.5. On the other hand, the shape dif- are approximately ±1500 Pa and ±0.1 m/s while when
ferences between the numerical and experimental distri- the advection term is absent they reduce to ±200 Pa and
butions along stations #1 to #4 are attributed to the wake ±0.06 m/s. For reference, the RMS of P′ and US′ for the mod-
function coefficients which do not capture accurately the els with and without advection terms are, respectively, of
experimental coalescence rate. 440 Pa, 0.04 m/s and 90 Pa, 0.03 m/s. Nevertheless, the mag-
nitude of the fluctuations is rather small. The phase diagram
5.6 The advection term and the induced slug of P′ × 21 ρUS′2 in Fig. 10c shows a correlation between the
oscillations pressure and mean squared velocity fluctuation of +0.30 and
+0.67 corresponding to the model with and without advec-
The introduction of the advection term on the model causes tion terms. The power spectral density of P′ and US′ is shown
the accelerating liquid slugs to compress or to expand the in Fig. 10d. The analyzed spectrum extended up to 100 Hz,
elongated gas bubbles resulting in a dynamic behavior analo- above this frequency the power is 10−6 times smaller than
gous to a multiple mass–dashpot–spring oscillators in series. the more energetic signals. The band with energy is concen-
The analogy is not complete because the gas–liquid system is trated within 0.1–10 Hz, which is typical of the slug cells’
not linear. The system’s excitation source is the slug insertion frequency, with a decaying rate proportional to f−2.
process. The random insertion velocities cause accelerations Next, the influence of advection terms on the averaged
on the liquid slug which are transmitted to the downstream slug flow properties is assessed. Table 10 brings a com-
slugs due to the compressibility of the elongated bubbles. parison of averaged slug terms with and without advection
The equation set including the advection term is repre- terms. The differences between the models with and with-
sented by Eqs. (1) and (2). The slug tracking model without out advection are negligible, perhaps the greatest difference
advection keeps the same mass conservation, Eq. (1), but shows up on the pressure drop estimate which corresponds

For flows where ρG/ρL  ≪ 1, the inclusion of advection

Eq. (2) reduces to: to 3 % on the pressure gradient.

j
K6− US + Pj+1 − Pj = K7 , (33) terms is a minor effect and does not change the averaged
slug flow properties such as LS, Lf, UT and f. However, there
where the K− 6 represents a modification of the K6 param- are open issues regarding the role of the advection terms, for
eter without the first term between braces, see Table 4. This example in the analyses of transient slug flow and also in the
simplification reduces the matrix system to a bidiagonal influence of the pulsating flow on the slug wall friction factor.
system analogous to the one found in Taitel and Barnea
[26]. Equation (33) expresses a balance between the pres-
sure, friction and gravity forces only. 6 Conclusions
A comparison of the slug tracking model with and with-
out advection terms is developed exploring the time history A compressible, one-dimensional and transient slug track-
of the 400th cell for run#3 employing setup configuration ing model is presented satisfying the mass and momentum

13
 J Braz. Soc. Mech. Sci. Eng.

(a)

(b)

(c)

(d)

Fig. 10  The P and US values tracking the 400th cell of run#3 employing a model with and without advection terms. a P and US versus time, b
fluctuations of P′ and US′ versus time, c fluctuations’ phase diagram and d P′ and US′ power spectrum density

equations in a unified fashion to handle horizontal and other flow variables. The numerical data also revealed that a
inclined gas–liquid flows. Each tracked cell has distinct good variable to tune the slug tracking is the coalescence rate.
velocities, lengths and holdups. The present model embod- The numerical study on the bubble coalescence phe-
ies all terms of the previous models and includes the advec- nomenon disclosed: no elongated bubble breakage, the
tion term. liquid film always increases due to the gas expansion or to
The comparison of the numerical against experimen- the bubble coalescence and the slug length may increase,
tal data discloses that monitoring just one variable, the slug decrease or even disappear. Due to these distinct ways
length for example, does not assure that a good fit to the mon- to change the length the LS distributions are established
itored variable correspond, necessarily, to a good fit to the quicker than the Lf distributions.

13
J Braz. Soc. Mech. Sci. Eng.

Table 10  Averaged slug flow z/D (–) Lf/D (–) LS/D (–) P (Bar) UT (m/s) f (Hz) Cav (m−1)
properties employing a model
with and without advection With advection terms
terms
 127 8.1 4.1 107 1.44 4.5
 265 10.8 5.7 105 1.47 3.4 0.118
 495 17.0 8.4 103 1.47 2.2 0.059
 777 22.2 10.3 100 1.49 1.8 0.024
Without advection terms
 127 8.1 4.1 104 1.44 4.5
 265 10.7 5.7 103 1.46 3.4 0.118
 495 16.6 8.4 102 1.46 2.2 0.058
 777 21.6 10.4 100 1.47 1.8 0.024

Small changes in the wake function parameter modify The accuracy of the slug tracking is limited due to the
the coalescence rate and also the distribution of the lengths, employed closure laws rather than a lack of physical rep-
velocity, frequency and pressure drop. A partial match with resentativeness of the 1D transient slug tracking concept.
the experimental data was achieved either at distances near The model proved to capture the averaged flow properties:
from the air–water mixer or farther. It was not possible to f, LS, Lf, P and UT with accuracy similar to other steady-
find wake law coefficients aw and bw capable of matching state slug flow models.
the experimental data from the inlet to the outlet suggest- The slug tracking major limitations are the lack of reli-
ing that the wake law cannot be represented by only two able slug inlet models and wake law. If the main interest
coefficients. Complementary, the suppression of the wake is on the flow property distributions far from the mixer
function results in no coalescence and the distributions of where the coalescence rate is lower than 1 % bubbles per
the flow properties at the pipe inlet remain unchanged up to meter then the inlet conditions can be relaxed. The use of
the pipe outlet. approximated inlet distributions together with the avail-
The use of the hypothetical periodic insertion defined by able wake laws coefficients will render approximated
constant values at the inlet showed that at the outlet the LS steady-state populations of lengths and velocities. On the
distribution shape approaches the experimental distribution other hand, if the goal is to predict accurately the evolution
indicating that the numerical predictions lose the influence of of the slug properties from the inlet up to a position down-
the inlet. The use of two distinct gas–liquid mixers allowed stream then it will be necessary accurate inlet and wake
to form slugs exhibiting different populations of lengths and law models.
velocities at the inlet. The experimental outcome showed that The wake law is the only model to represent the bubble
the populations of Lf and LS originating from the distinct mix- to bubble interaction, central to all slug tracking models.
ers become similar at z/D of 777D. The numerical and the The data comparison disclosed that the wake law coeffi-
experimental data showed that the slug flow properties lose cients are not universal. To enhance the slug tracking per-
the influence of the formation process. formance, further studies on the wake law are suggested.
The predicted distributions do follow the general trend
of the experimental data. As the flow evolves downstream, Acknowledgments  The authors gratefully acknowledge the receipt
of financial support from Petrobras. They also wish to thank to Dr
the distributions’ changes occur in small increments if not Fagundes Netto from Petrobras for bringing the theme under our
null, the numerical and experimental predictions become attention.
alike due to the loss of the influence of the formation pro-
cess. For this reason, the best matches were achieved at the
farthest measuring station. References
The gas compressibility in conjunction with the advec-
tion term makes the elongated gas bubbles oscillate like a 1. Al-Safram EM, Taitel Y, Brill JP (2004) Prediction of slug length
distribution along a hilly terrain pipeline using slug tracking
multiples oscillators in series. The pressure and velocity fre- model. J Energy Resour Technol 126:54–62
quency spectrum spans from 1 to 100 Hz, but frequencies 2. Barnea D, Brauner N (1985) Holdup of the liquid slug in two
above 10 Hz have negligible power. The high energy signals phase intermittent flow. Int J Multiph Flow 11:43–49
are within a band coincident with the slug cell’s frequency. 3. Barnea D, Taitel Y (1993) A model for slug length distribution in
a gas–liquid slug flow. Int J Multiph Flow 19(5):829–838
For flows where ρG/ρL ≪ 1, the introduction of the advec- 4. Bendiksen KH (1984) An experimental investigation of the
tion terms may affect the pressure gradient but constitutes a motion of long bubbles in inclined tubes. Int J Multiph Flow
minor effect to the averaged values of LS, Lf, f and UT. 10:467–483

13
 J Braz. Soc. Mech. Sci. Eng.

5. Cook M, Behnia M (2000) Slug length prediction in near hori- 20. Marcano R, Chen XT, Sarica C, Brill JP (1998) A study of slug char-
zontal gas liquid intermittent flow. Chem Eng Sci 55:2009–2018 acteristics for two-phase horizontal flow. International petroleum
6. Dukler AE, Hubbard MGA (1975) A model for liquid slug flow conference and exhibition of Mexico, Villahermosa, 3–5 March
in horizontal and near horizontal tubes. Ind Eng Chem Fundam 21. Mazza RA, Rosa ES, Yoshizawa CJ (2010) Analyses of liquid
14:337–347 film models applied to horizontal and near horizontal gas–liquid
7. Fabre J, Liné A (1992) Modeling of two-phase slug flow. Annu slug flows. Chem Eng Sci 65:3876–3892
Rev Fluid Mech 24:21–46 22. Moissis R, Griffith P (1962) Entrance effects in a two-phase slug
8. Fagundes Netto JR, Fabre J, Peresson L (1999) Shape of long flow. J Heat Transf 84:29–38
bubbles in horizontal slug flow. Int J Multiph Flow 25:1129–1160 23. Nydal OJ, Banerjee S (1996) Dynamic slug tracking simula-
9. Fagundes Netto JR et al (1999) Dynamique de poches de gaz tions for gas–liquid flow in pipelines. Chem Eng Commun
isolees en ecoulement permanent et non permanent horizontal. 141–142:13–39
PhD thesis, Institut National Polytechnique de Toulouse (INPT) 24. Nydal OJ, Pintus S, Andreussi P (1992) Statistical charac-

10. Gomez LE, Shoham O, Taitel Y (2000) Prediction of slug liquid terization of slug flow in horizontal pipes. Int J Multiph Flow
holdup: horizontal to upward vertical tubes. Int J Multiph Flow 18:439–453
26:517–521 25. Rodrigues HT (2009) Numerical simulation of slug flow employ-
11. Grenier P (1997) Evolution des longueurs de bouchons en
ing a lagrangian tracking algorithm (in portuguese). Master the-
écoulement intermittent horizontal. PhD thesis, Institut de Méca- sis, Federal Technological University at Curitiba, Brazil
nique des Fluides de Toulouse, France 26. Taitel Y, Barnea D (1998) Effect of gas compressibility on a slug
12. Gregory GA, Nicholson MK, Aziz K (1978) Correlation of the tracking model. Chem Eng Sci 53:2089–2097
liquid volume fraction in the slug for horizontal gas–liquid slug 27. Taitel Y, Barnea D (1990) Two-phase slug flow. In: Hartnett JP,
flow. Int J Multiph Flow 4:33–39 Irvine TF Jr (eds) Advances in heat transfer. Elsevier, 20, pp 83–132
13. Guet S, Decarre S, Henriot V, Liné A (2006) Void fraction in ver- 28. Taitel Y, Barnea D (1990) A consistent approach for calculating
tical gas–liquid slug flow: influence of liquid slug content. Chem pressure drop in inclined slug flow. Chem Eng Sci 45:1199–1206
Eng Sci 61:7336–7350 29. Ujang PM, Lawrence CJ, Hewitt GF (2006) Conservative incom-
14. Issa RI, Kempf MHW (2003) Simulation of slug flow in hori- pressible slug tracking model for gas–liquid flow in a pipe. 5th
zontal and nearly horizontal pipes with the two-fluid model. Int J BHRG North American conference on multiphase technology,
Multiph Flow 29:69–95 Banff, Canada, May 31
15. Kjeldby TK, Henkes RAWM, Nydal OJ (2013) Lagrangian slug 30. van Hout R, Shemer L, Barnea D (1992) Spatial distribution of
flow modeling and sensitivity on hydrodynamic slug initiation void fraction within the liquid slug and some other related slug
methods in a severe slugging case. Int J Multiph Flow 53:23–39 parameters. Int J Multiph Flow 18:831–845
16. Koskie JE, Mudawar I, Tiederman WG (1989) Parallel-wire
31. van Hout R, Shemer L, Barnea D (2003) Evolution of hydrody-
probes for measurement of thick liquid films. Int J Multiph Flow namic and statistical parameters of gas–liquid slug flow along
15:521–530 inclined pipes. Chem Eng Sci 58:115–133
17. Larsen M, Hustvedt E, Straume T (1997) PeTra: a novel com- 32. Viana F, Pardo R, Yánez R, Trallero JL, Joseph DD (2003) Uni-
puter code for simulation of slug flow. SPE annual technical con- versal correlation for the rise velocity of long gas bubbles in
ference and exhibition, San Antonio, Texas, 5–8 October round pipes. J Fluid Mech 494:379–398
18. Madani S, Caballina O, Souhar M (2009) Unsteady dynamics of 33. Zheng G, Brill JP, Taitel Y (1994) Slug flow behavior in a hilly
Taylor bubble rising in vertical oscillating tubes. Int J Multiph terrain pipeline. Int J Multiph Flow 20:63–79
Flow 35:363–375 34. Wang X, Guo L, Zhang X (2006) Development of liquid slug
19. Malnes D (1982) Slug flow in vertical, horizontal and inclined length in gas–liquid slug flow along horizontal pipeline: experi-
pipes. Report IFE/KR/E-83/002 V Inst. for Energy Technology, mental and simulation. Chinese J Chem Eng 14(5):626–633
Kjeller, Norway

13