International Journal of Multiphase Flow 149 (2022) 103972

Contents lists available at ScienceDirect

International Journal of Multiphase Flow

journal homepage: www.elsevier.com/locate/ijmulflow

A transient analysis of slug flow in a horizontal pipe using slug tracking

model: Void and pressure wave
Jéssica Leonel Gonçalves a, *, Ricardo Augusto Mazza b
a
CEPETRO, State University of Campinas, Brazil
b
Mechanical Engineering Faculty, State University of Campinas, Brazil

A R T I C L E I N F O A B S T R A C T

Keywords: In the oil and gas industry, the transient flow arises in daily situations, such as the start-up and shutdown of lines,
Two-phase flow pigging, artificial lifting, and drilling operations. These disturbances affect the temporal behavior of pressure and
Slug flow void fraction, which could cause separator flooding, system vibration, and production line facilities damage.
Slug tracking
Nevertheless, there are few numerical studies to describe this phenomenon. In this article, the transient slug flow
Pressure wave
Void fraction wave
is studied using the slug tracking model, which predicts the transient behavior and the evolution of slug flow
properties. This study demonstrates the slug tracking model’s capability to reproduce the transient flow as well
as the pressure and void fraction wave’s behavior and velocity. Experimental data presented by Maria and Rosa,
2016 were used to assess the model performance. The experimental data show a 6.0 s time delay between the two
steady states. This time delay does not influence the void fraction wave, and its numerical maximum deviation is
9.0%. On the other hand, the numerical performance for the pressure wave velocity depends on the time delay.
Using the experimental time delay, the maximum deviation for the pressure wave velocity is 6.8%. The slug
tracking model is combined with a mass–dashpot–spring analogy to avoid the experimental time delay de­
pendency. Using this methodology, the maximum deviation for pressure wave velocity is 8.2%. Some slug flow
properties, such as the bubble nose translational velocity, the bubble and liquid slug lengths, are also compared
to experimental data. The comparison is for averaged values and statistical distribution. The bubble nose ve­
locities and pressure deviations are less than 1.5%, while the bubble and liquid slug lengths deviations are 4.1%
and 5.3%, respectively. Most slug flow parameters have normal statistical distribution, but slug length follows a
close to log-normal distribution. The model reproduces this behavior.

Introduction pressure and void fraction, named as pressure and void fraction wave.
Transient flow is a daily phenomenon in several petroleum industry
Slug flow is present in several chemicals, petroleum, and nuclear procedures such as start-up and shutdown of two-phase lines, pigging
industry applications. The slug flow regime occurs over a wide range of process, artificial lift and drilling operation. According to Lin et al.
liquid and gas flow rates. Its main characteristic is the intermittent (2013) pressure wave propagation is crucial for gas influx detection,
passage of aerated liquid slug followed by elongated gas bubbles, which avoids severe financial losses, environmental contamination and
denominated as cells. The interaction with neighboring cells results in potential loss of human lives. Additionally, the comprehension of how
the creation or destruction of liquid slugs. Even when the steady-state the transient modifies hydrodynamic slug flow parameters such as slug
regime is considered, the slug flow’s key parameters follow a normal lengths, pressure peak, and wave velocities is essential to avoid flooding
distribution, except for slug length. It has a close to log-normal distri­ of separators, high backpressure, system vibration, and production line
bution (Bernicot and Drouffe, 1991; Nydal et al., 1992; Al-Safran et al., facilities damage.
2005; Nieckele et al., 2013). This feature makes the modeling complex Several researchers have made efforts to predict slug flow patterns’
in both transient and steady-state flow. parameters since the 20th century, and the authors had developed
Transient slug flow occurs due to modification of operating condi­ models capable of capturing the main slug flow parameters. However,
tions such as changes in the inlet flow rate or outlet pressure (Vigneron only few computationally demanding models can capture slug flow
et al., 1995). These disturbances change the temporal behavior of transient characteristics. Santim et al. (2017) compared the results of the

* Corresponding author.

https://doi.org/10.1016/j.ijmultiphaseflow.2022.103972
Received 29 January 2021; Received in revised form 22 December 2021; Accepted 7 January 2022
Available online 11 January 2022
0301-9322/© 2022 Elsevier Ltd. All rights reserved.
J.L. Gonçalves and R.A. Mazza International Journal of Multiphase Flow 149 (2022) 103972

literature models reported to study transient scenarios, which are the velocity depends only on the final state properties. The pressure wave
Two-Fluid Model, Drift-Flux Model, and the Non-Pressure Wave prin­ velocity depends on whether the perturbation increases or decreases the
ciple. All these models solve the transient transport equations for mo­ gas or the liquid flow rates.
mentum and mass conservation, considering only the main flow Henry et al. (1971), Martin and Padmanabhan (1979), Nguyen et al.
direction. Therefore, it is a one-dimensional solution. Additionally, this (1981), Xu and Gong (2008), Lin et al. (2013) and Li et al. (2016)
kind of model cannot capture the interaction between neighboring cells developed models to determine the velocity of pressure wave using
or the slug’s main flow parameters such as the bubble and the liquid slug properties such as the density of each phase, unit void fraction, bubble
lengths distributions, bubble nose velocity, among others. and slug length. Wallis (1969) proposed a model for the void fraction
Due to the restrictions mentioned above, the slug tracking model wave based on the drift flux kinematic relationship developed by Zuber
present itself as a good option. This model retains the simplicity of in­ and Findlay (1965).
tegrals formulation, with the ability to predict the evolution of flow The first numerical study was presented by Caussade et al. (1989).
parameters and adjacent cell interaction. The slug tracking model em­ They solved the transient drift-flux model numerically using an explicit
ploys a Lagrangian approach for tracking the bubble and liquid slug finite difference method. The authors compared their result with
boundaries as they travel through the pipe. The mass and momentum experimental data carried out in a horizontal line with a 50 mm diam­
conservation equations are applied to a control volume, where volume eter and 90 m long. Fabre et al. (1995) used a generalized drift-flux
and boundaries change in time and space. Barnea and Taitel (1993) model to study simultaneously void fraction and characteristics of
proposed the first model of this class, but the authors did not present any pressure wave in transient slug flow. The model underestimates the
model details. Other authors (Zheng et al., 1994; Grenier, 1997; Taitel pressure wave velocity.
and Barnea, 1998,2000; Franklin, 2004; Rodrigues, 2009) introduced More recently, Santim et al. (2017) compared the performance of
new mechanisms, such as liquid slug aeration, overtaking, gas three different models against Maria and Rosa’s experimental data. Two
compressibility, and hilly terrain in the following years. were models embedded in software used by the petroleum industry. The
More recently, Rosa et al. (2015) proposed a new slug tracking model other was an approximate Riemann solver based on the Drift-Flux Model
with all terms introduced by the previous models plus the inclusion of an proposed in Santim and Rosa (2016). Maria and Rosa’s experimental
advection term in the slug momentum and two-phase momentum fluxes data showed that even with the transient being introduced by a sudden
through the boundaries. The advection term allows the accelerating change in inlet flow rate, the flow parameters varies slightly between the
liquid slugs to compress or expand the elongated gas bubbles inducing two steady states due to inertia. There is a time delay between the end of
oscillations along the pipe. Furthermore, this model applies a control the first steady state and beginning of the second. Santim et al. (2017)
volume that encompassing the liquid slug and elongated bubble. The pointed out the importance of avoiding steep functions as boundaries
model presents the equations in a unified way and handles horizontal conditions. Their study used a polynomial adjustment using the
and inclined gas–liquid flows. This feature makes the model complete, measured inlet flow rate by Maria and Rosa (2016). The curve starts
including all one-dimensional flow mechanisms and makes it suitable to when the transient begins and ends when the flow rate reaches its new
investigate the pressure and void waves numerically in a new fashion value. The experimental data determined the elapsed time. All the
way. models captured the pressure and the void wave’s main characteristics,
Many authors have used experimental approaches to study transient such as its velocities, the pressure under and overshoot, the dispersive
slug flow, but only few have used analytical or numerical approaches. nature of the pressure, and the conservative nature of the void fraction
Experimental studies used mostly horizontal rigs with air and water as wave. However, none of the tested models captured the slug flow
working fluids. Analytical approaches simplified two-phase models to intermittency reflected by the void fraction’s oscillation during the
obtain pressure wave velocities. Numerical studies used two-fluid transients.
models (TFM), drift-flux models (DFM), and non-pressure wave (NPW) Combining the lack of studies on pressure and void fraction wave in
formulation to determine pressure and void wave velocities. From the slug flow, their broad applications with the strength and simplicity of
author’s best knowledge, there are no studies using slug tracking Rosa et al.’s model, this paper aims to demonstrate the slug tracking
models. The majority of two-phase transient studies considered model’s capability to capture the transient flow and the pressure and
dispersed flow, with only few having slug flow as the objective. Below, a void fraction wave’s behavior and velocity. Furthermore, to improve the
brief review of these studies. model’s independence of experimental inlet data, this work also aims to
Fabre et al. (1995) suggested that the lack of pressure wave studies in present a method to calculate the time delay between the steady-state
slug flow is due to the complexity of the flow itself and the absence of a regimes. The numerical results are compared to the Maria and Rosa
theoretical framework. Other researchers point out that the complexity (2016) experimental campaign. The model performance analyses are
of this phenomenon is related to the various coexisting processes such as based on:
gas compressibility, the changes of the interface between the gas and
liquid, the momentum and energy transfer between two phases, wall • Prediction of pressure and void fraction wave velocities.
friction, slip between the phases, and heat transfer (Miyazaki et al., • The sensitivity of the velocities of the waves to different time delays.
1971; Huang et al., 2005; Xu and Gong, 2008; Lin et al., 2013). • Temporal profile of pressure, void fraction and slug flow parameters.
Three experimental studies presented key concepts on slug flow • The statistical distribution of the slug flow parameters as well as its
transients (Fabre et al., 1995; King et al., 1998; Maria and Rosa, 2016). averaged values.
The campaigns acquired transient flow data in a horizontal line with air
and water. The flow rate perturbations were introduced at the inlet by
sudden changes in the flow rates. Fabre et al. (1995) and Maria and Rosa The physical model
(2016) identified that the void wave velocity is constant and close to the
average bubble nose translational velocity at the second steady state. The slug flow is a succession of elongated bubbles and liquid slugs,
Furthermore, the void fraction wave is conservative and slower than the exchanging mass and momentum while flowing through the pipe. Fig. 1
pressure wave. King et al. (1998) and Maria and Rosa (2016) pointed out shows the jth cell control volume encompassing the liquid slug and film
an overshoot and undershoot of the pressure wave when the gas inlet or the elongated gas bubble region. Furthermore, Fig. 1 presents the
velocity increases and decreases. When it refers to the liquid transient, nomenclature used for the liquid slug and elongated gas bubble prop­
the pressure change occurs gradually from the initial to the final states. erties. The cell has length LU and is a sum of bubble (LB) and slug (LS)
Furthermore, Maria and Rosa (2016) calculated the velocity of pressure length. At a given instant time, the elongated bubble properties are
and void fraction waves. The authors observed that the void wave liquid holdup (Rf), the void fraction (αf), liquid film thickness (hf), and

2
J.L. Gonçalves and R.A. Mazza International Journal of Multiphase Flow 149 (2022) 103972

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

also the liquid film (Uf), gas phase (UG) and bubble nose (UT) velocities. is the gas density temporal variation in the cell, and it is an unknown
For the liquid slug region, the parameters are liquid holdup (RS), the parameter. The consecutive cells’ liquid slug velocities (US) are the other
void fraction (αS), and also liquid (US) and dispersed bubbles (Ub) ve­ two. Eq. (2) is the liquid’s momentum balance, and it is not easy to
locities. Fig. 1 also represents jth cell identification and its neighboring disclose each term. However, the RHS represents the exchanging of
cells at a pipe with length L, diameter D, cross-sectional area A and momentum, the pressure difference between two neighboring cells, the
inclination θ with the horizontal. The liquid and gas phases have the wall friction at the liquid slug, and the hydrostatic forces due to gravity.
density and dynamic viscosity defined by ρL, ρG, μL and μG, respectively. The LHS of Eq. (2) represents the time momentum ratio of the liquid slug
Slug tracking uses a Lagrangian approach to track bubble and slug velocity. The liquid slug velocity (US) is one of the unknown parameters,
boundaries throughout the pipe, capturing its properties for each time and the other is the pressure of two consecutive cells. The model also
step. Several authors had proposed models using this approach, but this uses the following hypothesis:
work uses the model proposed by Rosa et al. (2015). The model includes
the advection term plus all other terms of previous models and captures 1 The flow is isothermal and the liquid phase is incompressible.
the liquid slug acceleration due to the elongated bubbles compression or 2 The gas density is much smaller than the liquid density
expansion inducing oscillations along the pipe. The model applies (ρG /ρL << 1), constant within the cell, but changes from cell to cell
one-dimensional mass and momentum balance to the cell-based control as the pressure changes.
volume and obtains two time-dependent ODE equations related to the 3 There are no surface tension effects, and the liquid and the gas phases
time evolution from a previous state for the gas density and liquid slug share the same pressure.
velocity for each jth cell: 4 The liquid film is non-aerated.
5 The pressure along the bubble is constant, and the liquid slug wall
1 dρjG shear stress is:
K1 = K2 Uj−S 1 − K3 UjS + K4 , (1)
ρjG dt {
1 Cf = 0.079Re−M0.25 ReM > 2000
τjS = Cf ρM UM |UM |where withReM
dUjS ( )2 ( ) 2 Cf = 16.Re−M1 ReM ≤ 2000
= − UjS Q1 − UjS Q2 + Q3 PjG − Pj+1
G − Q4 . (2)
dt ρM UM D
= .
Table 1 shows K’s and Q’s coefficients. Eq. (1) is the gas and liquid μM
mass balance over the cell, and the RHS represents the gas crossing
neighbor cells due to liquid slug and drifts velocities. The LHS of Eq. (1)

Table 1 Closure equations

The K’s and Q’s parameters for Eqs. (1) and (2).
Parameters Expressions To solve Eqs. (1) and (2) closure equations for fluid and mixture
K1 j j j j properties, slug dispersed bubbles and bubble nose velocities, and
LB αf − LS αs
[ ] bubble nose and slug front positions are necessary. The gas density is
K2 j− 1 j− 1
cb αS
RS
j− 1
1+ correlated to pressure by some thermodynamic equation of state. The
1− cj−b ]
1 j− 1

K3
[ j j
αS
experimental gas phase is air near atmospheric pressure and at room
j cb αS
RS 1 + j j temperature. Therefore, it is considered that the relationship of gas
[ 1 − cb αS ]
density and pressure is that of an ideal gas for an isothermal process:
[ ]
K4 j− 1 αj−S 1
j αjS
ud j− 1 j− 1
− ud j j
⎧(1 − cb αS 1− cb α⎫
)2 S 1 dρG 1 dPG
Q1 j
)(
(3)
1 ⎨ RS dxj /dt ⎬ = .
j j+1
− 1 1− j
ρG dt PG dt
LS ⎩ Rf US ⎭
The liquid slug is similar to a dispersed bubble flow, and the ho­
⎛ ⎞⎛ ⎞
Q2 j j j j
ρ
j M |U |
M ⎝ 1 j ud (1 − RS )⎠
2Cf ⎠⎝ RS + mogenous model is used in this region to determine the mixture
ρL D.RjS 1 − cjb (1 − RjS ) US
j

Q3 1 properties:
ρjL RjS LjS
Q4 gsin(θ) χ M = χ G (1 − RS ) + χ L RS , (4)

3
J.L. Gonçalves and R.A. Mazza International Journal of Multiphase Flow 149 (2022) 103972

where χ is either the density or the dynamic viscosity, RS is the slug’s dyj
liquid holdup, and the subscript G and L refer to the liquid and gas = UTj . (11)
dt
phases. The mixture velocity is:
UM = US RS + Ub (1 − RS ), (5) Numerical model

where US is the liquid slug velocity, and the velocity of the dispersed The above equations are solved numerically using the fourth-order
bubble is: Runge-Kutta method due to its higher order and stability. The cells’
Ub = cb UM + uD . (6) properties are updated each sub-step. The slug tracking models can only
be applied when the slug regime is already present. They do not
The distribution parameter (cb) represents the nonuniformity of flow reproduce the slug formation process. Therefore, these models depend
and dispersed bubbles concentration. In the horizontal flow, bubble on the slug’s cell properties definition at outlet and inlet. As an outlet
concentration is not uniform due to the buoyancy. However, cb was boundary condition, the pressure must be known in all time steps. In this
taken as unit (cb = 1) as recommended by Taitel and Barnea (1990). The paper, all the cells leave the pipe at atmospheric pressure. As an inlet
authors claimed that the liquid slug dispersed bubbles have the same boundary condition, the cell’s properties, such as superficial velocities
average velocity as the liquid slug velocity in the slug mixture to support (JL and JG), bubble and slug lengths (LB and LS), and film and slug
the recommendation. According to Harmathy (1960), the drift velocity holdup (Rf and RS), are set when each cell enters the numerical domain.
for relatively large and deformable bubbles is: Rodrigues (2009) presents a detailed description of the cell insertion and
( ) removal process. The inlet data’s representativeness will affect the
σ × g(ρL − ρG ) 0.25
uD = 1.54 2
(RS )1.75 Sinθ, (7) simulation’s result. Therefore, estimating the cell’s properties distribu­
ρL
tion as close as possible to experimental data are essential. This paper
uses two different inlet boundary conditions based on the experimental
where σ is the surface tension and g is the gravity acceleration. The
results: random and periodic distribution. Section 4.1 will present it in
translational velocity of the elongated bubble nose is defined as:
details. Furthermore, as the numerical procedure is a time march, all the
√̅̅̅̅̅̅̅̅̅̅̅̅̅ )
(
Δρ cells’ properties (bubble and slug lengths, void fractions, bubble nose
UT = C0 UM + C∞ gD (1 + h(LS )), (8) velocities and pressure gradient) inside the numerical domain at the
ρL
initial time (t = 0 s) must be known. For this purpose, the unit cell-based
where the constants, C0 and C∞ depend on Reynolds, Froude, Eötvos and model proposed by Taitel et al. (1990) is used. The cell’s properties
pipe inclination. The C0 and C∞ values result from several authors’ along the pipe are affected by the gas expansion due to the pressure
contributions, including Bendiksen (1984) and Viana et al. (2003), and gradient.
Table 2 gives their values. The acceleration of the trailing bubble due to
the wake behind the leading bubbles changes the nose velocity. Several Simulation and data processing
authors (Moissis and Griffith, 1962; Barnea and Taitel, 1993; Grenier,
1997; Cook and Behnia, 2000; van Hout et al., 2003) studied this phe­ The tests section is a horizontal pipeline with a 26 mm ID and 22.256
nomenon and proposed a wake function as: m long (L/D ~x223C 856), and is similar to Maria and Rosa (2016).
Three probes located at 4 m (S#1), 10.35 m (S#2) and 18.64 m (S#3)
h(LS ) = aw exp(bw ⋅LS /D) + cw , (9) downstream the pipeline inlet register the slug flow parameters evolu­
tion – see Fig. 2. These probes’ positions are coincident with Maria and
where aw and bw are constants and define the wake intensity; cw is a Rosa’s measuring stations. Tap water is the liquid working fluid with the
correction for aw and bw. density and dynamic viscosity of 999 kg/m3 and 8.55 × 10− 4 Pa.s,
Applying the mass conservation to control volume that encompass­ respectively. Compressed air is the gas fluid with the density related to
ing only the bubble region limited by the boundaries Yj and Xj-1, the slug the pressure by Eq. (3) and viscosity of 1.7 × 10− 5 Pa.s. The surface
front position can be determined by: tension between air and water is 0.075 N/m. The time step [Δt] is 10− 3
( )
[ ]
( ) ( s, and it is following the mesh grid test presented by Rodrigues (2009)
j ) ( )
1 dPG
for similar simulations. Additionally, the default configuration sets the
j
1 − Rf LB Pj dt + dydt RjS − Rjf − 1 − Rj−S 1 Ubj− 1 + 1 − RjS Ubj
j j

local atmospheric pressure of 95 kPa as the outlet pressure.

j− 1
dx G
= ( ) .
dt Rj − Rj S f

(10) Grid test – transient flow

The bubble nose front travels with the bubble nose velocity, and its Maria and Rosa’s experimental campaign performed four tests aim­
positions can be determined by: ing to study transient behavior in slug flow. Changing the liquid or gas
superficial velocities generates the transients. In Test #1, the gas flow
rate reduces, while in Test #2, it increases. The liquid flow rate is con­
stant for both tests. Regarding Test #3 and #4, the gas flow rate remains
Table 2 constant while the liquid flow rate is reducing in Test#3 and increasing
C0 e C∞ parameters as a function of Reynolds, Froude, Eötvos and pipe in Test#4. Table 3 shows the inlet condition to gas and liquid superficial
inclination. velocities. For each test, the authors performed a series of experimental
ReM FrM C0 C∞ for ReB > 200 runs to determine all slug parameters’ statistical distributions and
average values. The experimental data also showed that the liquid and
≥ 2000 ≥ 3.5 1.2 CV∞ Sinθ
gas superficial velocities are low enough for the elongated bubble to
< 3.5 2
1.0 + 0.2Sin θ CH
∞ Cosθ + CV∞ Sinθ
carry all the gas, i.e., the liquid slug is non-aerated (RS = 1).
< 2000 - 2.0 CH V
∞ Cosθ + C∞ Sinθ
√̅̅̅̅̅̅̅̅̅̅̅̅̅ Test #1 and #2 are antisymmetric, as well as Test #3 and #4, as
ρM UM D UM ρ2L gD2 (ρL − ρG )gD2 shown in Table 3. Initial steady state refers to the period between the
where, ReM = , Fr = √̅̅̅̅̅̅, ReB = , Eo = , C∞H
μM gD μL σ start of the data recording and the beginning of the transient. The nu­
merical data recording starts after the 600th cell exits the pipe to ensure
( )
1.76 0.34
= 0.542 − and CV∞ =
Eo0.56 (1 + 3805/Eo3.06 )0.58 the initial steady state. The final steady state refers to the period

4
J.L. Gonçalves and R.A. Mazza International Journal of Multiphase Flow 149 (2022) 103972

Fig. 2. Geometrical configuration of the pipeline.

function to represent the transition. The time delay used in Fig. 3 is

Table 3 experimental. The inverse tangent represents the transitions closely,
Liquid and gas superficial velocities at the initial and final states: Test #1 to #4.
with an RMS deviation of 1.6%.
Steady Test #1 Test #2 Test #3 Test #4
state
JL JG JL JG JL JG JL JG Characteristic time delay
[m/ [m/ [m/ [m/ [m/ [m/ [m/ [m/
s] s] s] s] s] s] s] s] Maria and Rosa (2016) show a 6 s time delay for the superficial ve­
Initial 0.60 0.54 0.60 0.28 0.60 0.54 0.30 0.57 locity to reach the second steady state. Santim et al. (2017) used this
Final 0.60 0.27 0.59 0.53 0.30 0.57 0.60 0.56 experimental time delay to capture the wave velocities, but they could
not reproduce the time delay using the proposed models. This paper uses
an analogy between Rosa et al.’s model and the mass–dashpot–spring to
between the flow properties re-establishment and the end of the simu­
study the time delay influence combining Eqs. (1), (2) and (3) with αS =
lation. The simulation stop condition is the moment of the 600th cell
0 and θ = 0:
reaching the exit the pipeline. Because of the two steady-state regimes,
the list of cell proprieties must have two independent stages: initial and j j
mÜ S + cU̇ S + kUSj = F(t), (12)
final states’ flow parameters.
The list of cell properties of the inlet boundary condition can be j j j j j j j j j j
where,m = ρL LS LB αf , c = (4Cf /D)pL LS LB αf US and k = PG , represent ting
j

random or periodic. Both conditions were defined based on Maria and

Rosa’s experimental data. In periodic conditions, all cells entering the mass, damping coefficient and spring constant, respectively; F(t) =
j+1
numerical domain are identical and their values are the average j
PG U S
j− 1 j j
− LB αf ṖG represents the exciting force. Using the experimental
measured data at inlet. The random condition has each cell with a data to analyse Eq.(12), the system behavior is underdamped with the
different value for the parameters. This condition uses the measured damping time scale for the liquid slug velocity oscillation given by:
distributions for all properties.
Maria and Rosa’s data show a gradual velocities profile modification t=
D
, (13)
between the initial and final states, with a time delay before the flow 2Cfj USj
rate reaches the second steady state. As pointed out by Santim et al.
j
(2017), the right time delay is crucial to obtaining a good agreement where D is the pipe diameter, Cf is the friction factor and US liquid slug
between experimental and numerical results for the wave velocities. velocity of the second steady state. The damping time scale represents
This paper studies the time delay influences using two different transi­ the minimum time at which the system reaches the new velocity. The
tions between the two steady states: maximum time is the void fraction wave residence time (tR), which is
defined by the ratio of void fraction wave velocity and pipe length.
• Sudden change (SC): The velocity immediately assumes the value of
the second steady state.
• Gradual change (GC): The velocity spans slightly from the first Determination of pressure and void fraction wave velocities
steady state to the second, and the model calculates the time delay.
The pressure and void fraction temporal profiles obtained by the slug
Fig. 3 shows the transition for Test #1 where the superficial gas tracking model allow us to calculate the waves’ propagation velocities.
velocity changes suddenly or gradually, using inverse tangent as time The travel time between two virtual probes and their distances de­
termines the wave velocities. Three different combinations between
probes were used to capture the wave behavior throughout the pipe, the
distances were: S#1 to S#2 with 244D; S#2 to S#3 with 319D; S#1 to
S#3 with 563D. The final steady state defines the criterion for deter­
mining the instant when the wave reaches the probe. If initial state
values are higher than the final state, search for the minimum value on
the time series before the jump of PG or α as the y coordinate value. The
time is the value of the x-axis when the minimum value at the y-axis
crosses the wave’s jump, as shown in Fig. 4. If the first steady state is
lower than in the second, the search is for the maximum value on the
time series before the jump.

Data analysis

This section investigates the slug tracking performance against Maria

Fig. 3. The flow transition considering a sudden change (SC) and gradual and Rosa ’s experimental data. Four different setups explore the inlet
change (GC) using inverse tangent and experimental time delay. condition, the transition’s type and the wake’s effect influence. The

5
J.L. Gonçalves and R.A. Mazza International Journal of Multiphase Flow 149 (2022) 103972

Fig. 4. Criterion to determine the instant the wave’s front reaches the virtual probe for Test #1.

default setup employs: time step of 0.001 s, no wake effect (h(LS ) = 0), approximately 1/3 of the measured time delay. The exception is Test #3
RS = 1, periodic insertion process based on experimental data – see using the experimental friction factor, but the uncertainty for this test is
Section 4.1. Table 4 shows all setup configurations, highlighting the the highest one. Test #3 has the most extended bubbles and liquid slugs,
differences. The analysed parameters are void and pressure wave ve­ and the lowest measured pressure gradient, which turns difficult a
locity, pressure and void wave time evolution, and slug flow parameters correct friction factor estimation. The maximum time delay uses the
evolution. void wave velocity. The void wave residence’s time reveals a maximum
time delay of 20 s for Test #1 and #3, and 16 s for Test #2 and #4.
Void and pressure wave velocity Nevertheless, the time delays showed in Table 7 are not the measured
ones. They represent a time scale for them.
Several numerical simulations show that the time delay does not Fig. 5 and Fig. 6 show measured and numerical average pressure
affect the void fraction wave velocity, and Table 5 shows the numerical wave velocities for a characteristic time spanning from 0 to 18 s and all
and Maria and Rosa’s void fraction velocity. The numerical simulation tested cases. The range was chosen based on the time analysis shown in
uses the default setup with the SC inlet transition. The void wave ve­ Table 7. The experimental pressure wave velocity is the horizontal
locity is constant along the pipe, and the model underestimated the continuous black line, and the numerical data are the dashed blue line
parameter with a maximum deviation spanning from 6.4% to 9.0%. Both with symbols. The symbol represents the numerical pressure velocity
void velocities are close to the elongated bubble nose’s translational determined using the time delay showed on the x-axis. With no time
velocity at the second state, which agrees with Vigneron et al. (1995). delay, the deviation from experimental data is maximum. The deviation
Therefore, the closure equation for the nose’s translational velocity for Test #3 and #2 is 81% and 33%, respectively, with the two other
defines the void wave velocity. Santim et al. (2017) also underestimates tested cases between them. Increasing the time delay further than three
the void wave velocities, and the deviation spans from 0% to 12%. The times the minimum time scale showed in Table 7, the deviation for all
best results were using the two-fluid model. tested cases increases again. The numerical results confirm the influence
Table 6 shows the pressure wave velocities for all tests. As reported of the time delay on the determination of the pressure wave velocity.
by Santim et al. (2017), the time delay measured by Maria and Rosa The numerical pressure wave velocity (CP) has a relative minimum
(2016) is 6 s. Thus, the numerical pressure wave shown in Table 6 used value, and it is coincident with the measured time delay for Test #3 and
the 6 s time delay and default setup. The maximum deviation is 6.8% for #4, as shown in Fig. 6. Additionally, it represents the minimum devia­
Test #4, and the minimum is 1.4% for Test #2. Using a transient drift tion, as shown in Table 6. Test #1 and #2 deviations on the relative
flux model, Santim et al. (2017) had deviations spanning from 0% to minimum value are 8.2% and 3.8%, respectively. The time delay on this
21.1% and from 1.9% to 28.9% using a two-fluid model. All the models point is 9 s and 5 s for Test #1 and #2, respectively. These deviations are
agree with the experimental data for the pressure wave velocities when higher than 6 s time delay (see Table 6), but the highest deviations are
the time delay is already known, including the slug tracking model. still lower than those found by Santim et al. (2017). These results show
However, if there are no experimental data to express the time delay, the that it is possible to find the pressure wave velocity using a relative
models’ performances are poor. minimum value search as the time delay increases, and, therefore, with
The time delay has a significant influence on the pressure wave ve­ no experimental data. The time delay defined through the analogy be­
locity. To assess how the time delay influences the pressure wave ve­ tween the slug tracking model and mass–dashpot–spring can be used as
locity, several numerical simulations were performed with different the minimum limit of scale, even calculated by the Blasius correlation,
time delays spanning from no time delay (SC condition) to a maximum and the maximum time limit is defined by the time residence of void
value based on Table 7. The minimum time delay scale showed in the fraction wave inside the pipe.
table uses the mass–dashpot–spring analogy presented in Section 4.2.
The error analysis shows a 12% maximum error when the friction factor Temporal evolution of the continuous variables: void fraction
uses the measured pressure gradient, and 135% if it uses the Blasius and pressure
correlation. All tested cases have similar damping times, and it is
Fig. 7 and Fig. 8 show the superficial velocities, void fraction and
pressure time evolution at two pipe positions, probe S#1 and S#3. The
Table 4 numerical simulations used a 6 s time delay for all tests. All tests are
Slug tracking setup configurations. antisymmetric, with the final steady state of one being the initial steady
(i) Default Periodic insertion; aw = cw = 0 state of the other. Fig. 7(a) and (b) is when the gas superficial decreases
(ii) Periodic insertion – aw and bw in Eq. (9) given by Grenier (1997)
(Test #1) and increases (Test #2), respectively. The numerical void
(iii) Random insertion – aw and bw in Eq. (9) given by Grenier (1997) fraction abruptly changes to the new steady state while the experimental
(iv) Random insertion – aw and bw experimentally adjusted changes more gradually. The numerical result is from one single run and

6
J.L. Gonçalves and R.A. Mazza International Journal of Multiphase Flow 149 (2022) 103972

Table 5
Void fraction velocities comparisons among experimental (Cα_EXP) and numerical data (Cα_NUM) for Test #1 to #4.
Test #1 Test #2 Test #3 Test #4

Stations Cα_EXP [m/s] Cα_NUM [m/s] Cα_EXP [m/s] Cα_NUM [m/s] Cα_EXP [m/s] Cα_NUM [m/s] Cα_EXP [m/s] Cα_NUM [m/s]

S#1 – S#2 1.16 1.08 1.46 1.33 1.16 1.07 1.45 1.36
S#1 – S#3 1.16 1.09 1.46 1.34 1.16 1.08 1.45 1.36
S#2 – S#3 1.16 1.09 1.47 1.35 1.17 1.09 1.45 1.37
Average velocities 1.16 1.09 1.46 1.34 1.17 1.08 1.45 1.36
Relative Deviation (%) 6.4% 9.0% 8.3% 6.6%

The experimental uncertainty is ±0.01 m/s for all void wave velocities.

Table 6
Pressure wave velocities comparisons among experimental (Cp_EXP) and numerical data (Cp_NUM) for Test #1 to #4.
Test #1 Test #2 Test #3 Test #4

Stations Cp_EXP [m/s] Cp_NUM [m/s] Cp_EXP [m/s] Cp_NUM [m/s] Cp_EXP [m/s] Cp_NUM [m/s] Cp_EXP [m/s] Cp_NUM [m/s]

S#1 – S#2 22.0 20.5 26.4 23.5 18.6 19.8 20.9 20.5
S#1 – S#3 22.8 22.5 25.0 24.8 18.3 17.4 20.7 22.2
S#2 – S#3 23.5 24.4 24.0 25.9 18.1 16.0 20.5 23.7
Average velocities 22.8±1.3 22.5 25.1±1.4 24.7 18.3±1.0 17.7 20.7±1.1 22.1
Relative Deviation (%) 1.5% 1.4% 3.1% 6.8%

When the gas velocity increases, numerical and experimental reach the
Table 7
second steady state 22 seconds after the velocity changes at S#1. When
Time delay scale.
the gas velocity decreases, the numerical procedure reaches the second
Cf Test Test Test Test steady-state pressure level 14 seconds faster than the experimental.
#1 #2 #3 #4
Fig. 8 shows the void and pressure evolution when liquid velocity
Damping time tC [s] Experimental 2.4 ± 2.5 ± 5.7 ± 2.6 ± changes. Fig. 8(a) and (b) refer to the liquid velocity increases and de­
0.3 0.3 0.7 0.3
creases, Tests #3 and #4, respectively. There is a slight change in gas
Blasius 2.4 ± 1.9 ± 2.4 ± 1.9 ±
0.2 0.6 3.3 0.7
velocity, less than 4%, when the liquid velocity changes. There is no
Void wave residence’s - 20.5 16.6 20.6 16.3 change in the liquid velocity when the gas velocity changes, see Fig. 7.
time tR [s] The numerical void fraction is also conservative with constant ampli­
tude. The experimental void fraction has oscillations, despite the
average process, which can be associated with the alternate passage of
the experimental is the average of 100 runs. The deviation of the nu­
long bubbles and slugs in a typical slug formation process. The model is
merical and experimental void fraction is minimum, less than RMS 1%.
restricted to slug flow pattern and cannot capture slug formation and the
The void wave shape is conservative with a constant amplitude and
oscillations mentioned above. The numerical deviation is minimum, less
agrees with the experimental data.
than RMS 1%, and the main difference is the absence of void oscillations
The numerical procedure underestimates the pressure on S#1, and
on numerical solutions.
Section 5.3 will present a detailed analysis of pressure gradients. The
Numerical and experimental results show that the pressure changes
pressure level on S#3 for all tests is the same and for both steady state.
abruptly and then reaches the second steady state by a damped oscil­
This station is near the pipe exit, and the pressure is near the local at­
lation region. It takes 20 s to reach the new steady-state pressure level,
mospheric pressure. The numerical procedure fixes the outlet pressure
but the pressure changes 80% in the first 4 seconds. As for Test#1 and
by a boundary condition. The pressure wave presents an undershoot
#2, the model underestimates the pressure levels, but the model
before the second steady state, when the gas velocity decreases (Test
captured the pressure evolution. Considering station S#1, the RMS de­
#1), as shown in Fig. 7(a). There is a pressure overshoot when the gas
viation between the numerical and experimental pressure is a maximum
velocity increases (Test #2), as shown in Fig. 7(b). In both cases, the
of 4% for Test #2 and a minimum of 0.8% for Test #3.
pressure reaches the new steady-state level by damped oscillations.

Fig. 5. Average pressure wave velocity comparison between the experimental and numerical data for Test #1 (a) and Test #2 (b). Numerical is the dashed blue line
with symbols, and the continuous black line is the experimental.

7
J.L. Gonçalves and R.A. Mazza International Journal of Multiphase Flow 149 (2022) 103972

Fig. 6. Average pressure wave velocity comparison between the experimental and numerical data for Test #3 (a) and Test #4 (b). Numerical is the dashed blue line
with symbols, and the continuous black line is the experimental.

Fig. 7. Comparison between experimental (dashed line) and numerical (continuous line) for gas and liquid superficial velocity, void fraction and pressure time
evolution for Test #1 (a) and Test #2 (b) at the measuring stations S#1 and S#3.

Discrete variables analyses: the slug flow properties setups, as shown in Table 4. Setup (i) is the simplest, and it is the same
for the transient analyses. Setups (ii) and (iii) analyse the inlet condi­
This section aims to disclose the model’s capacity to predict slug flow tion’s influence, periodic or random, on the flow parameters. Both use
properties such as pressure, bubble nose velocity, liquid slug and bubble wake coefficients as proposed by Grenier (1997). Finally, the wake co­
lengths. The analyses compare the averaged values and distribution efficients are Grenier’s in Setup (iv) with the bW adjusted to minimize
model against the experimental data. The analyses use four different the deviations. The simulations used 0.8 and 0.3 for initial and final

8
J.L. Gonçalves and R.A. Mazza International Journal of Multiphase Flow 149 (2022) 103972

Fig. 8. Comparison between experimental (dashed line) and numerical (continuous line) for gas and liquid superficial velocity, void fraction and pressure time
evolution for Test #3 (a) and Test #4 (b) at the measuring stations S#1 and S#3.

state, respectively. All the simulations use Test #3 steady-states regimes residues for the second steady-state. However, Setup (iv) underestimates
as inlet flow rates. Setup (iii) and (iv) use the experimental data as inlet the pressure gradient, while Setup (ii) and (iii) overestimate it with 19%
conditions for each cell that enters the pipe, while Setup (ii) uses the residue. Despite high residues on pressure gradients for all Setups, the
averaged values. model’s performance is satisfactory. The pressure gradient is highly
Fig. 9 shows the averaged values for at least 600 cells passing affected by the friction factor correlations and the model uses Blasius,
through each station. Setup (iv) has the minimum deviation for exper­ which presents a good accuracy for smooth pipes, and monophasic flows
imental data, with less than 2.9% deviation on average considering all with Re < 105. However, the use of Blasius correlation is wide because of
stations and parameters. Bubble and liquid lengths have the most sig­ its simplicity.
nificant deviations for all variables spanning between 4.1% and 5.3% on Only Setup (iv) presents good bubble and liquid slug lengths’ results
average. The maximum deviation is at station #4 for the bubble length. for the two steady-state regimes, as shown in Fig. 9(a.2), (b.2), (a.3) and
The pressure has the minimum residue, followed by the elongated nose (b.3). The Setup (iv) average residue is 5%, while the other two Setups
velocities for all setups. Setup (ii) and (iii) have a similar performance have 19%. The adjusted wake coefficients represent the overtaking
with similar residues for all variables. The averaged residue, considering mechanism properly. All setups demonstrate that the liquid slugs are
all analysed slug parameters, is less than 10.3% for both setups. Setup smaller at the inlet pipe for the first steady state. For Setup (ii), the
(iv) has the best result for the translational bubble nose velocity, and the bubbles and liquid slug length only grow from station S#2 to S#3 for the
residue is 1.3% on average. Setup (ii) and (iii) have 2% and 3% of res­ initial steady state. However, in the final steady state, the lengths remain
idue, respectively. constant throughout the pipe, as demonstrated in Fig. 9(a.2) and (a.3).
All setups underestimated the first steady-state pressure and over­ The wake law does not significantly influence the bubble and liquid slug
estimated the second one at the inlet, as shown in Fig. 9(a.4) and (b.4). lengths with the periodic inlet. In Setup (iii) and (iv), the wake accel­
For the first steady state, setup (iv) has 1.3% of residue on average, and erates the trailing bubbles, and the liquid slugs grow. In the random
for Setup (ii) and (iii) both have 2.2%. For the second, Setup (iii) and (iv) inlet, each cell that enters the numerical domain has a different liquid
have less than 0.4% of residue, while Setup (ii) and (iii) have 1.0% and slug length, and the wake accelerates the trailing bubbles on the smallest
0.6%, respectively. All setups underestimate the first steady-state pres­ liquid slug. Therefore, the averaged bubble and liquid slug lengths in­
sure gradient. Setup (iii) and (iv) present similar performances with 30% crease along the pipe, as shown in Fig. 9(a.2), (a.3), (b.2) and (b.3),
of residue, while Setup (ii) has 40%. Setup (iii) and (iv) have 5% of Setup (iii) and (iv).

9
J.L. Gonçalves and R.A. Mazza International Journal of Multiphase Flow 149 (2022) 103972

Fig. 9. Slug flow parameters for the first steady state (a) and second steady state (b). From top to bottom are the translational elongated bubble nose velocity, bubble
length, liquid slug length and pressure.

The main characteristic of slug flow is intermittence, with the flow states. The most probable (mode) values for all parameters are slightly
parameters distributed around the average value. Therefore, a complete different from the average values. The S#2 distributions are almost
flow characterization requires statistical analysis of the flow parameters. coincident with the experimental data due to the inlet numerical
The analysed flow parameters are the translational elongated bubble boundary condition, which uses the experimental distribution. The
nose velocity and bubble and slug lengths. Numerical and experimental exception is the elongated bubble nose velocity, as shown in Fig. 10
parameters comparison is at stations S#2 and S#4 to capture the slug’s (a.1). The nose velocity uses Eq. (8) instead of the experimental data.
flow parameter evolution. The numerical results use only Setup (iv) due The results are not better than the correlation used, which has a large
to its better performance, as shown above. spread compared to available literature data (Bendiksen, 1984). The
Fig. 10 shows the probability density function (PDF) at both steady distribution at S#4 is not different for the same reasons, as shown in

10
J.L. Gonçalves and R.A. Mazza International Journal of Multiphase Flow 149 (2022) 103972

Fig. 10. Distributions of the experimental and numerical values of elongated bubble nose velocity and bubble and slug lengths at S#2 (top row) and S#4 (bot­
tom row).

Fig. 11. Experimental (blue line) and numerical (red line) time evolution during the transient. From top to bottom are the translational elongated bubble nose
velocity, bubble length, liquid slug length and slug frequency.

11
J.L. Gonçalves and R.A. Mazza International Journal of Multiphase Flow 149 (2022) 103972

Fig. 10(b.1), (a.2) and (b.2). The experimental bimodal bubble length CRediT authorship contribution statement
distribution observed for the second steady state in both measuring
stations indicates a non-fully developed flow. The model does not cap­ Jéssica Leonel Gonçalves: Writing – review & editing, Conceptu­
ture bimodal behavior but captures the spreading downstream. The alization, Methodology, Software, Validation, Formal analysis. Ricardo
model predicts the maximum bubble length accurately but with a larger Augusto Mazza: Writing – review & editing, Conceptualization, Meth­
number of long bubbles. The first steady state has a normal distribution, odology, Funding acquisition, Supervision, Project administration.
and the model accurately predicts the distribution. The liquid slug
length follows a close to log-normal distribution at S#2 and S#4 for both Declaration of Competing Interest
steady states, as shown in Fig. 10(a.3) and (b.3). The model captures the
distribution accurately. The authors declare that they have no known competing financial
Fig. 11 shows translational elongated bubble nose velocity, and interests or personal relationships that could have appeared to influence
bubble and slug lengths time evolution at S#2 and S#4 during the the work reported in this paper.
transient, similar to pressure and void fraction in Fig. 7 and Fig. 8. Once
again, the numerical procedure uses Setup (iv) and Test #3 in this Acknowledgments
analysis. The transient occurs at 30 s due to the superficial liquid ve­
locity decreasing at the inlet. All the parameters oscillate around their The authors would like to thank PETROBRAS S.A. for the financial
average values. The numerical oscillations are higher than the experi­ support of this research under the grant no. 0050.0022719.06.4-BR.
mental data due to Maria and Rosa’s, 2016 employment of a centered Gonçalves, J.L. would also like to thank the Brazilian National Council
moving average to draw a trend line. This process is needed because for Scientific and Technological Development (CNPq) for their contri­
these properties are outcome by an ensemble of the discrete variables of bution through the award of a MSc scholarship.
all experimental runs. Nevertheless, it does not necessarily apply this
process in numerical results, once these properties are sampled for each References
time step. Thus, one simulation reflects well the behavior of slug flow
parameters evolution. Bubble and liquid slug lengths increase signifi­ Al-Safran, E.M., Sarica, C., Zhang, H.Q., Brill, J.P., 2005. Probabilistic/mechanistic
modeling of slug-length distribution in a horizontal pipeline. SPE Prod. Facil. 20,
cantly during the transient but decrease to the second steady-state value 160–172. https://doi.org/10.2118/84230-PA.
rapidly. The numerical results captures this behavior. Barnea, D., Taitel, Y., 1993. A model for slug length distribution in gas-liquid slug flow.
Int. J. Multiph. Flow 19, 829–838. https://doi.org/10.1016/0301-9322(93)90046-
W.
Conclusions Bendiksen, K.H., 1984. An experimental investigation of the motion of long bubbles in
inclined tubes. Int. J. Multiph. Flow 10, 467–483. https://doi.org/10.1016/0301-
This paper uses a slug tracking model to study the transient slug flow. 9322(84)90057-0.
Bernicot, M.F., Drouffe, J.M., 1991. Slug-length distribution law for multiphase
The numerical results are compared to Maria and Rosa’s experimental transportation systems. SPE Prod. Eng. 6, 166–170. https://doi.org/10.2118/17864-
data to assess the numerical performance. Increasing or decreasing the pa.
liquid or gas superficial velocities generates the transients, and three Caussade, B., Fabre, J., Jean, C., Ozon, P., Théron, B., 1989. Unsteady phenomena in
horizontal gas-liquid slug flow. In: 4th International Conference on Multi-Phase
virtual stations sample the main slug flow properties. The horizontal test
Flow. Nice, France, p. 17.
section geometry and virtual sampling station coincide with Maria and Cook, M., Behnia, M., 2000. Slug length prediction in near horizontal gas–liquid
Rosa’s experimental campaign. The experimental data show that the intermittent flow. Chem. Eng. Sci. 55, 2009–2018. https://doi.org/10.1016/S0009-
flow needs time to reach the final state velocity, despite altering the 2509(99)00485-6.
Fabre, J., Liné, A., Gadoin, E., 1995. Void and pressure waves in slug flow. In: IUTAM
velocity abruptly. Thus, this paper proposes a method using the slug Symposium on Waves in Liquid/Gas and Liquid/Vapor Two-Phase Systems,
tracking model to achieve this time, which minimizes the experimental pp. 25–44.
data dependency. Franklin, E., de, M., 2004. Modelagem Numérica para Seguimento Dinamico de Bolhas
em Escoamento Intermitente Horizontal Gás-Líquido. Universidade Estadual de
The time delay between the two steady states significantly influences Campinas.
the determination of pressure wave velocities. The influence of the time Grenier, P., 1997. Evolution des Longueurs de Bouchons en Écoulement Intermittent
delay on the determination of the pressure wave velocity is the same for Horizontal. Institut de Mécanique des Fluides de Toulose, France.
Harmathy, T.Z., 1960. Velocity of large drops and bubbles in media of infinite or
all tested cases, with a local minimum value as the time delay increases. restricted extent. AIChE J. 6, 281–288. https://doi.org/10.1002/aic.690060222.
The minimum value was the closest to experimental data for Test #3 and Henry, R.E., Grolmes, M.A., Fauske, H.K., 1971. Pressure-pulse propagation in two-phase
#4. The error at minimum value was lower than 8.2% for Test #1 and one and two component mixtures.
Huang, F., Takahashi, M., Guo, L., 2005. Pressure wave propagation in air-water bubbly
#2. The time delay did not influence the void fraction wave velocity. It and slug flow. Prog. Nucl. Energy 47, 648–655. https://doi.org/10.1016/j.
was close to the elongated bubble nose’s translational velocity at the pnucene.2005.05.068.
second steady-state, as also reported by Vigneron et al. (1995). King, M.J.., Hale, C.P., Lawrence, C.J., Hewitt, G.F., 1998. Characteristics of flowrate
transients in slug flow. Int. J. Multiph. Flow 24, 825–854. https://doi.org/10.1016/
The numerical simulations of the transient induced by the change of
S0301-9322(97)00088-8.
gas velocity are in accordance with King et al. (1998) and Maria and Li, H., Chen, R., Li, X., Meng, Y., Zhu, L., Zhao, J., 2016. Investigation of pressure wave
Rosa (2016) with either an undershoot or overshoot in pressure wave propagation and attenuation characteristics in wellbore gas-liquid two-phase flow.
profile. The transient introduced by liquid velocity changes the pressure, J. Nat. Gas Sci. Eng. 35, 1088–1100. https://doi.org/10.1016/j.jngse.2016.09.020.
Lin, Y., Kong, X., Qiu, Y., Yuan, Q., 2013. Calculation analysis of pressure wave velocity
with an abrupt change followed by a gradual evolution until reaching in gas and drilling mud two-phase fluid in annulus during drilling operations. Math.
the final state, as demonstrated by Maria and Rosa (2016). Probl. Eng. 2013, 1–17. https://doi.org/10.1155/2013/318912.
Besides the transient, the model is also used to assess the steady-state Maria, L.D., Rosa, E.S., 2016. Void fraction and pressure waves in a transient horizontal
slug flow. Int. J. Multiph. Flow 84, 229–244. https://doi.org/10.1016/j.
slug flow parameters. Comparing the numerical to experimental data ijmultiphaseflow.2016.05.008.
shows that the model accurately captured the main slug flow parame­ Martin, C.S., Padmanabhan, M., 1979. Pressure pulse propagation in two-component
ters. These analyses compare both the average and statistical distribu­ slug flow. J. Fluids Eng. 101, 44–52. https://doi.org/10.1115/1.3448733.
Miyazaki, K., Fujii-E, Y., Suita, T., 1971. Propagation of pressure wave in air-water two-
tions at virtual stations. The use of the wake coefficient does not phase system, (I). J. Nucl. Sci. Technol. 8, 606–613. https://doi.org/10.1080/
guarantee sufficient cell interaction to promote bubble and liquid slug 18811248.1971.9733007.
length growth. Only by adding a randomized process of entering cells, Moissis, R., Griffith, P., 1962. Entrance effects in a two-phase slug flow. J. Heat Transf.
84, 29. https://doi.org/10.1115/1.3684284.
these lengths properly grow downstream. In this process, all the bubbles Nguyen, D.L., Winter, E.R.F., Greiner, M., 1981. Sonic velocity in two-phase systems. Int.
and liquid slugs have different sizes, which increases the interactions J. Multiph. Flow 7, 311–320. https://doi.org/10.1016/0301-9322(81)90024-0.
between (unit) cells.

12
J.L. Gonçalves and R.A. Mazza International Journal of Multiphase Flow 149 (2022) 103972

Nieckele, A.O., Carneiro, J.N.E., Chucuya, R.C., Azevedo, J.H.P., 2013. Initiation and Taitel, Y., Barnea, D., 1998. Effect of gas compressibility on a slug tracking model. Chem.
statistical evolution of horizontal slug flow with a two-fluid model. J. Fluids Eng. Eng. Sci. 53, 2089–2097. https://doi.org/10.1016/S0009-2509(98)00007-4.
Trans. ASME 135, 1–11. https://doi.org/10.1115/1.4025223. Taitel, Y., Barnea, D., 1990. Two-phase slug flow. Adv. Heat Transf. 83–132. https://doi.
Nydal, O.J., Pintus, S., Andreussi, P., 1992. Statistical characterization of slug flow in org/10.1016/S0065-2717(08)70026-1.
horizontal pipes. Int. J. Multiph. Flow 18, 439–453. https://doi.org/10.1016/0301- van Hout, R., Shemer, L., Barnea, D., 2003. Evolution of hydrodynamic and statistical
9322(92)90027-E. parameters of gas–liquid slug flow along inclined pipes. Chem. Eng. Sci. 58,
Rodrigues, H.T., 2009. Simulação Numérica do Escoamento Bifásico Gás-Líquido no 115–133. https://doi.org/10.1016/S0009-2509(02)00441-4.
Padrão de Golfadas Utilizando um Modelo Lagrangeano de Seguimento de Pistões. Viana, F., Pardo, R., Yánez, R., Trallero, J., Joseph, D., 2003. Universal correlation for
Universidade Tecnológica Federal do Paraná. the rise velocity of long gas bubbles in round pipes. J. Fluid Mech. 494 https://doi.
Rosa, E.S., Mazza, R.A., Morales, R.E., Rodrigues, H.T., Cozin, C., 2015. Analysis of slug org/10.1017/S0022112003006165. S0022112003006165.
tracking model for gas–liquid flows in a pipe. J. Braz. Soc. Mech. Sci. Eng. 37, Vigneron, F., Sarica, C., Brill, J.., 1995. Experimental analysis of imposed two-phase flow
1665–1686. https://doi.org/10.1007/s40430-015-0331-7. transients in horizontal pipelines. In: Proceedings of the 7th International
Santim, C.G., Rosa, E.S., 2016. Roe-type Riemann solver for gas-liquid flows using drift- Conference. Cannes, França, pp. 199–217.
flux model with an approximate form of the Jacobian matrix. Int. J. Numer. Methods Wallis, G.B., 1969. One Dimensional Two-Phase flow. McGraw-Hil, New York ed.
Fluids 80, 536–568. https://doi.org/10.1002/fld.4165. Xu, X.-X., Gong, J., 2008. A united model for predicting pressure wave speeds in oil and
Santim, C.G.S., Gaspari, E.F., Paternost, G.M., 2017. A transient analysis of gas-liquid gas two-phase pipeflows. J. Pet. Sci. Eng. 60, 150–160. https://doi.org/10.1016/j.
slug flow inside a horizontal pipe using different models. J. Pet. Sci. Eng. 151, 62–76. petrol.2007.05.012.
https://doi.org/10.1016/j.petrol.2017.01.042. Zheng, G., Brill, J.P., Taitel, Y., 1994. Slug flow behavior in a hilly terrain pipeline. Int. J.
Taitel, Y., Barnea, D., 2000. Slug-tracking model for hilly terrain pipelines. SPE J. 5, Multiph. Flow 20, 63–79. https://doi.org/10.1016/0301-9322(94)90006-X.
102–109. https://doi.org/10.2118/61445-PA. Zuber, N., Findlay, J.A., 1965. Average volumetric concentration in two-phase flow
systems. J. Heat Transf. 87, 453–468. https://doi.org/10.1115/1.3689137.

13