Wear 476 (2021) 203654

Contents lists available at ScienceDirect

Wear
journal homepage: http://www.elsevier.com/locate/wear

A mechanistic model for predicting erosion in churn flow

Ronald E. Vieira *, Siamack A. Shirazi
The Erosion/Corrosion Research Center, Department of Mechanical Engineering, The University of Tulsa, Tulsa, OK, 74104, USA

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

Keywords: Sand particles, which are regularly entrained with the production fluids while transporting from a reservoir to
Solid particle erosion surface facilities, can cause severe erosion and erosion-corrosion damage to production equipment. A crucial
Erosion modeling parameter for studying erosion in multiphase flows is the flow pattern. The particle motion characteristics will
Multiphase flow
change drastically with changes in the flow pattern in the pipelines. In vertical upward gas-liquid flow, a widely
Churn flow
accepted flow pattern classification is bubble, slug, churn and annular flow. Churn flow is a chaotic flow pattern
consisting of Taylor bubbles and liquid slugs that are distorted in shape and it is characterized by the presence of
periodic large interfacial waves called flooding-type waves. Studying the erosion due to sand particles entrained
in churn flows is extremely difficult, since the solid particle erosion magnitudes are coupled with several
multiphase flow parameters such as phase distribution and velocities. This work focuses on improvement of a
multiphase churn flow erosion model for elbows in vertical pipes. In this investigation, a drift-flux model is
proposed to better predict the flow behavior and the characteristic initial particle velocity for churn gas-liquid
flow. Based on the drift-flux model, the actual gas velocity is calculated as a function of mixture and drift ve­
locities. The coefficients of the drift flux expression under churn flow conditions are determined from datasets
available in literature. The model predictions are compared with the erosion data bank at the Erosion/Corrosion
Research Center and other published papers. The data comprises effects of pipe diameter, particle characteristics,
flow velocities, and liquid viscosity. The results show that the new approach provides much better predictions for
the maximum erosion in bends when compared with the original mixture based model and other models
available in literature.

1. Introduction
Predicting erosion is well recognized by many researchers and the
authors as a very complex problem. Current models for erosion calcu­
lation that are presented in the literature are based on semi-empirical
correlations that relate solid particle impact velocity to pipe wall ma­
terial loss rate [2]. Consequently, erosion estimation can be even more
complicated in multiphase flow systems where solid particles are
entrained in a complex fluid structure and the particle impingement
velocity and impingement location on the pipe wall must be determined.
A more complete approach to erosion modeling involves flow field so­
lution, particle tracking, and erosion equations. A flow field solution is
used to determine the distribution of the density and velocity of a fluid
over space and time for a given geometry, and then the particle tracking
model is used to determine the particle trajectories for particles
entrained in the flow. The particle impingement data is extracted from
the trajectories and it is used along with empirical formulas to predict
the erosion rates. This approach has been studied and applied previously
[3,4]. This procedure requires expertise in computational fluid dy­
namics as well as a great deal of computational effort and time. More­
over, flow modeling in multiphase flow is still under development and
modifications, and this does not even consider covering the particle
tracking models for multiphase flows. For the reasons stated, this
comprehensive methodology is not a suitable alternative for design and
production engineers, especially if multiphase flow is present. Thus, a
good alternative would be to develop simpler models that are based on
the comprehensive erosion model and empirical information supported
by experimental data.
When gas and liquid are simultaneously introduced in a pipe, they
can distribute themselves into different flow patterns/regimes. In ver­
tical upward gas-liquid flow, a widely accepted flow pattern classifica­
tion is bubble, slug, churn and annular flow. Churn flow regime is
present for a wide range of gas-liquid ratios for two-phase flows, espe­
cially when the pipes have large diameters [5]. It can be defined as

* Corresponding author.
E-mail address: rev87@utulsa.edu (R.E. Vieira).

https://doi.org/10.1016/j.wear.2021.203654
Received 10 September 2020; Received in revised form 16 November 2020; Accepted 30 November 2020
Available online 11 February 2021
0043-1648/© 2021 Elsevier B.V. All rights reserved.
R.E. Vieira and S.A. Shirazi
highly aerated slugs with repeated destruction of liquid continuity in the
slug during an oscillatory motion of the liquid slug.
Churn flow contains a high fraction of gas, and consequently, it is an
erosive flow pattern amid different multiphase flow regimes. Churn flow
exhibits an intermittent behavior, indicating similarities with slug flow.
Therefore, slug flow models such drift-flux [7] with proper closure re­
lationships can be used to model phase velocities and erosion in churn
flow.
In churn flow, similar to other flow regimes such as slug and annular,
the current E/CRC erosion model assumes that the sand particles are
uniformly distributed in the liquid phase [9]. Furthermore, the model
considers that the velocities of the liquid and sand are the same as the
mixture velocity. Therefore, the characteristic initial sand particle ve­
locity for churn flow is assumed to be the mixture velocity. It has been
recognized that this model is overly conservative and significantly over
predicts the erosion data [10].
The goal of this work is to improve the current mechanistic modeling
of erosion in churn flow. In this study, a drift-flux model is applied to
calculate the initial particle velocity used by one-dimensional (1-D) and
two-dimensional (2-D) erosion models developed by E/CRC.
In this paper, firstly, the background literature showing previous and
current progress in the area of erosion in churn flow is presented. Sec­
ondly, the churn flow erosion models which have been developed by E/
CRC investigators are described. Thirdly, the major modifications made
to the existing models are explained, namely the drift flux approach.
Finally, validation results obtained with the modified churn flow 1-D
and 2-D models are compared with experimental data and existing
models in the literature.
2. Background
2.1. Erosion modeling in churn flow
Many aspects of erosion in multiphase flow have been studied for
decades. However, models developed specifically for multiphase pro­
duction are few, and models for erosion prediction in churn flow are
even scarcer [10]. Recognizing these deficiencies, researchers began
developing more appropriate procedures.
Salama [11] proposed a simple empirical equation based on exper­
imental data on small I.D pipes, whereas few factors were taken into
account in the empirical equation:
1 Ẇ P vMIX 2 dP
E= (1)
sM D2 ρMIX
where E is the erosion rate in mm/yr, ẆP is the sand flow rate in kg/day,
vMIX is the mixture in m/s, D is the pipe diameter in mm, dP is the particle
diameter in μm, ρMIX is the fluid mixture density in kg/m3, and sM is a
geometry dependent coefficient which is 5.5 for elbows. Det Norske
Veritas [12] developed a more conservative assessment procedure to
estimate the erosion in various fittings merging with numerical simu­
lations, model equations and erosion data. The characteristic impact
angle was calculated by respectively tracking the solid particles in
straight pipes, welded joints, elbows, blinded tees and reducers. This
model is semi-empirical, and tries to predict the whole erosion process
by using macroscopic parameters such as sand flow rate, pipe size, su­
perficial velocities and the mixture fluid properties of the phases.
What the empirical models of Salama and DNV have in common is
that they do not distinguish the particle motion characteristics in
different flow patterns. Therefore, there is no modification of erosion
mechanisms in different flow patterns in their models. Mechanistic
models have overcome this limitation.
Mazumder et al. [13] proposed methods for calculating the initial
sand velocities in different flow patterns, and then predicted the
maximum erosion rates by combining previous E/CRC work [14,15].
The mechanistic model of Mazumder et al. [13] is based on the concept
2
Wear 476 (2021) 203654
Fig. 1. Ultrasonic transducers in a 76.2 mm in standard elbow [2].
of “stagnation zone” which was originally proposed by Shirazi et al. [14]
to predict sand erosion in single-phase flow. For multiphase flow,
McLaury and Shirazi [15] adopted the weighted average viscosity and
density to modify the description of the fluid in “stagnation zone”.
Recently, Shirazi et al. [16] incorporated the effects of turbulent eddies
on particle impact velocities with their previous work to update the
erosion prediction model in multiphase flow. However, the Shirazi et al.
[16] model uses an equation to determine a characteristic velocity for
multiphase flow based on empirical data.
Different erosion multiphase models have been proposed to take
more into account the fluid characteristics of various flow patterns. After
performing several sand sampling experiments and pseudo slug flow
velocity measurements, Kesana et al. [8] proposed a modified 1D model
for horizontal pseudo-slug flows that determined the mass fractions of
sand particles in the liquid slug body and the gas pocket (Taylor bubble)
regions using an energy dissipation concept. These mass fractions
enabled the erosion occurring in both regions to be calculated. Based on
erosion data and Wire-Mesh Sensor (WMS) characterization studies,
Vieira [18] proposed an initial characteristic particle velocity for erosion
calculations for annular-churn transition. This velocity was assumed to
be 1.8 times the gas core velocity. Also, a transition criterion between
annular and unstable annular flow regimes was identified based on
experimental data and implemented into the program for the prediction
of erosion magnitudes. More recently, Kang and Liu [19] proposed an
integrated model to predict sand erosion in elbows for multiphase flows.
In this model, the physical model for slug flow was employed to describe
characteristics in churn flow. Sedrez [20] successfully predicted the
erosion trend for dispersed bubble flows in vertical-horizontal 50.8 mm
I.D. elbows using mixture properties for the initial characteristic particle
velocity.
It is important to note that the characteristic parameters in slug flow,
annular and dispersed bubble flows are relatively clear and precise.
However, for churn flow, the flow field is more complex, resulting in
difficulty in describing the fluid characteristics theoretically.
2.2. Experimental database for churn flows
Previous experimental investigations concerning sand particle
erosion in churn flow are presented here.
The erosion tests for 26.5 mm and 49 mm I.D elbows were conducted
by Kvernvold and Sandberg [21] and Birchenough et al. [22], respec­
tively. Kvernvold and Sandberg [21] carried out erosion tests in duplex
stainless steel elbows with nitrogen and water as fluids at a pressure of 7
bar. Birchenough et al. [22] investigated the erosion in carbon steel
elbows for air–water multiphase flow at an operating pressure of 2 bar
absolute. Few points in these investigations were carried out for churn
flow conditions.
Several experiments were carried out during the past years to
investigate sand particle erosion in 3-inch I.D. (76.2 mm) standard el­
bows in vertical air-water flow conditions. Parsi et al. [2] and Vieira
et al. [23] used a non-intrusive ultrasonic measurement (UT) technique
R.E. Vieira and S.A. Shirazi
Fig. 2. Erosion contours in churn flow. Data from Parsi et al. [2].
a) UT erosion in churn flow; vSG = 18 m/s; vSL = 0.37 m/s; μL = 1 cP; 300 μm sand
b) UT erosion in churn flow; vSG = 27 m/s; vSL = 0.32 m/s ; μL = 10 cP; 150 μm sand;
to evaluate erosion rate in vertical annular and churn flows. They
employed sixteen (16) transducers to obtain erosion patterns on the
elbow outer wall (Fig. 1).
Superficial liquid and gas velocities ranged from 0.3 m/s to 0.70 m/s
and from 9.8 m/s to 27.2 m/s, respectively. Two different liquid vis­
cosities of 1 cP (water-only) and 10 cP (water + Carboxymethyl cellu­
lose or CMC) were used in these studies. Three different abrasive
particles with different mean diameters (dp) were considered in the
experiments: silica flour, dp = 20 μm; Oklahoma#1, dp = 150 μm; and
California 60, dp = 300 μm. A database of 33 points was obtained for
churn flow. Regardless of particle size and flow velocities, transducers 1,
2, and 3 in Fig. 1 showed the highest erosion rates compared to other
transducers. Fig. 2 shows two erosion contours for different particle sizes
and superficial velocities.
Vieira et al. [24] also conducted a series of experiments to investigate
the effect of superficial velocities, viscosity and sand size on erosion rate
in churn flow in a standard 76.2 mm elbow using an Electrical Resis­
tance (ER) probe. The ER probe was flush with the outer radius of the
bend at 45◦ . A database of 22 points was obtained for churn flow.
3. Erosion model description
Since the current models for churn flows described in this paper
utilize aspects of the original model, brief descriptions of the one-
dimensional (1-D) and two-dimensional (2-D) models are provided
next, but more details can be found in other sources [3,14,16].
3.1. One-dimensional model
The key to this model is to predict a representative particle impacting
velocity, and then apply an erosion equation to quantify the penetration
rate from the particle impact information. This model uses a semi-
mechanistic correlation [16] to calculate penetration rate:
WS ( )
h = FM FS FP Fr/D VL 2.41 + FC e− βSt
Vm 2.41 (2)
(D/D0 )2
where h is the penetration rate or thickness loss per unit time in m/s, FM
is the material constant, FS is the sand sharpness factor, FP is the pene­
tration factor in m/kg, Fr/D is the penetration factor for elbow radius of
curvature, WS is the sand production rate in kg/s, VL is the particle
impact velocity in m/s, (D /D0 ) is the ratio of pipe diameter to 25.4 mm
3
Wear 476 (2021) 203654
pipe, Vm is the characteristic impact speed contributed by turbulence, FC
is the fraction or erosion caused by turbulence and β is the Stokes
number damping coefficient. Currently, FC = 0.2, β = 0.01 and Vm is the
mixture velocity. In this correlation, mechanistic modeling is applied to
determine particle impact speed. In this work, the erosion rate is the
thickness loss per unit sand throughput and it is obtained by dividing Eq.
(2) by the sand rate.
A simplified 1-D particle tracking method based on a stagnation
length concept [14] is used. One representative particle is tracked along
the stagnation length to characterize particle behavior and determine
impact speed as it approaches the wall. The latest near wall velocity
profile adopts a jet velocity profile. The mathematical representation is
given in the following equation.
( )2.5
Vf = V 0 − V 0
x
(3)
Lstag
where Vf is the fluid velocity when the particle advances to a position x
in m/s, Lstag is the stagnation length and V0 is the characteristic particle
initial velocity or the particle initial velocity before the particle reaches
the stagnation zone.
In the original model for churn flow, the velocities of the liquid and
sand are assumed to be the same as the mixture velocity. Therefore, the
characteristic initial sand particle velocity for churn flow is assumed to
be the mixture velocity and can be calculated as:
V0 = vMIX = vSL + vSG (4)
Parsi et al. [10] showed that this model is overly conservative and
over predicts the erosion data by a factor of approximately 6.0.
3.2. Two-dimensional model
The 1D model assumes that the particle is moving in a straight di­
rection which may not be correct under high density conditions where
particles tend to follow the flow streamlines. Moreover, calculations
were based on one-dimensional particle tracking, and turbulence effects
are not considered in the one-dimensional particle tracking. Therefore
after extensive CFD-based erosion modeling, Zhang et al. [3] considered
both normal and tangential impact velocities as well as turbulent effects
to develop a model for predicting erosion.
The 2-D erosion prediction method was first developed for predicting
single-phase flow erosion with great advantage over predicting small
R.E. Vieira and S.A. Shirazi Wear 476 (2021) 203654

particle erosion, erosion in viscous and high-pressure flows compared
with its 1-D erosion prediction method counterpart [3]. It can be
considered as a hybrid CFD erosion prediction tool which utilizes flow
solutions from CFD codes. The flow field needed for particle tracking is
obtained by linear interpolation of stored flow solutions based on Rey­
nolds number. The stored meshes are also interpolated with the change
of pipe diameter. The in-house 2-D particle tracking was developed by
Zhang et al. [3] which considers turbulence fluctuations and offers the
user versatile control over eddy scales and near wall treatment.
In the 2-D method, particles are tracked in a 2-D domain with flow
field information assigned to the attached mesh. The particles are
released from the inlet at characteristic initial sand particle velocity and
tracked along the flow domain. Once a particle hits the wall, the
impacting speed and angle are recorded and averaged over each cell to
serve as input for the erosion calculation module. The erosion calcula­
tion module utilizes an erosion ratio equation to convert the impacting
parameters to erosion damage. The original extension of the 2-D model
to multiphase flow also assumes mixture velocity for the characteristic
initial sand particle velocity. Thus, it has similar limitations to the 1-D
model even though a CFD solution based on mixture properties of gas-
liquid is used.
4. Model modification
To extend the previous 1-D and 2-D models to this complex churn-
slug flow regime, a brief description of drift flux approach used in this
work to calculate the characteristic particle initial velocity is provided
below.
4.1. Drift-flux approach
Churn flow exhibits intermittent behavior, similar to slug flow. There
is not available mechanistic model in the literature to predict
hydrodynamic behavior of churn flow due to its highly disordered
chaotic nature. A recent comprehensive study of Woldesemayat and
Ghajar [25] identified over sixty-eight (68) void fraction correlations
available in the literature based on slip, k–α and drift flux models. They
found the void fraction correlations based on drift flux model to be more
flexible and fairly accurate as compared to other types of correlations for
void fraction in vertical upward pipe orientations. Other studies [1,26]
have also recommended the use of drift flux model based correlations for
vertical upward and downward pipe orientations.
A drift-flux model describes the flow in terms of a distribution
parameter and an averaged local velocity difference between the phases;
this implies that the flow can still be considered as a single phase flow
moving at a steady velocity while still accounting for the velocity dif­
ference between the two phases [27]. The first principal analysis to the
drift flux model is due to Zuber and Findlay [7]. Its application has
proved very successful in several engineering problems related with
transport systems involving two-phase flow dynamics.
In the drift flux model, the void fraction (α) is expressed as a function
of the distribution parameter (C0 ), gas-phase drift velocity (vD ), and
superficial velocities of gas (vSG ) and liquid (vSL ): α =
vSG (C0 (vSG + vSL ) + vD )− 1 . Considering a fully developed velocity pro­
file, C0 is defined as the approximate ratio between the maximum and
the average velocity of the slug unit. This ratio is determined from the
assumption that the propagation velocity of the gas bubble follows the
maximum local liquid velocity in front of the nose tip. The drift flux
velocity vD refers to the velocity at which the gaseous phase travels and
penetrates through the stagnant liquid phase within the pipe.
The drift flux model is much simpler in its formulation in comparison
to other two-fluid models due to several considerable assumptions that
must be considered, one of them being the pseudo-fluid treatment for
the two-phase mixture [28]. On the other hand, in the application of
engineering problems, these assumptions become very useful as they
allow detailed analysis of two-phase flow behavior to be carried out with

Table 1
Experimental and predicted erosion results for churn flow; UT data; μL = 1 cP and 10 cP.
μL (cP) dP (μm) vSG (m/s) vSL (m/s) Data (mm/ Original 1-D (mm/ Modified 1-D (mm/ Modified 2-D (mm/ Salama (mm/ DNV (mm/
kg) kg) kg) kg) kg) kg)

1 300 49.0 0.5 5.55E-04 4.97E-03 1.88E-03 1.58E-03 5.62E-03 1.78E-03

1 300 27.0 0.5 2.24E-04 1.16E-03 4.62E-04 3.27E-04 1.05E-03 3.85E-04
1 20 18.4 0.4 2.31E-05 1.19E-04 1.08E-04 1.31E-05 2.77E-05 4.31E-05
1 20 26.9 0.3 1.02E-04 3.91E-04 2.69E-04 3.95E-05 1.02E-04 1.79E-04
1 150 10.8 0.4 1.50E-05 4.81E-05 2.19E-05 1.11E-05 4.34E-05 3.69E-05
1 150 15.2 0.4 3.37E-05 1.24E-04 5.42E-05 2.88E-05 1.16E-04 8.90E-05
1 150 18.7 0.5 3.67E-05 2.08E-04 9.47E-05 5.25E-05 1.89E-04 1.51E-04
1 150 20.0 0.4 7.51E-05 2.65E-04 1.03E-04 4.71E-05 2.84E-04 1.76E-04
1 150 27.1 0.3 2.04E-04 5.87E-04 2.03E-04 7.88E-05 7.57E-04 3.82E-04
1 150 27.2 0.4 1.44E-04 5.68E-04 2.31E-04 1.03E-04 6.15E-04 3.89E-04
1 150 27.1 0.4 1.31E-04 5.60E-04 2.31E-04 1.04E-04 5.96E-04 3.86E-04
1 300 9.8 0.6 1.79E-05 9.00E-05 4.68E-05 2.78E-05 4.89E-05 3.06E-05
1 300 9.8 0.7 9.56E-06 8.60E-05 4.93E-05 3.45E-05 4.21E-05 3.15E-05
1 300 17.7 0.4 1.55E-04 4.27E-04 1.50E-04 7.95E-05 4.00E-04 1.29E-04
1 300 17.7 0.6 9.94E-05 4.05E-04 1.92E-04 1.09E-04 2.71E-04 1.33E-04
1 300 18.3 0.8 6.28E-05 4.19E-04 2.35E-04 1.37E-04 2.29E-04 1.49E-04
1 300 18.1 0.6 7.69E-05 4.26E-04 2.06E-04 1.19E-04 2.80E-04 1.41E-04
1 300 26.8 0.5 2.40E-04 1.14E-03 4.35E-04 2.89E-04 1.09E-03 3.76E-04
1 300 27.1 0.4 3.37E-04 1.19E-03 4.02E-04 2.70E-04 1.27E-03 3.85E-04
1 300 27.2 0.4 3.55E-04 1.21E-03 3.89E-04 2.75E-04 1.35E-03 3.88E-04
1 300 27.4 0.6 1.83E-04 1.18E-03 5.19E-04 3.57E-04 9.70E-04 4.02E-04
1 300 31.1 0.5 2.33E-04 1.64E-03 6.26E-04 4.39E-04 1.61E-03 5.52E-04
1 300 49.0 0.5 7.13E-04 5.00E-03 1.77E-03 1.53E-03 6.04E-03 1.77E-03
10 20 27.2 0.4 9.27E-06 2.74E-04 2.74E-04 1.48E-05 8.03E-05 1.05E-04
10 150 10.8 0.5 3.54E-06 3.43E-05 1.91E-05 4.81E-06 3.47E-05 2.82E-05
10 150 18.5 0.5 1.45E-05 1.85E-04 8.85E-05 3.91E-05 1.70E-04 1.47E-04
10 150 27.2 0.3 1.40E-04 6.15E-04 2.18E-04 8.26E-05 8.09E-04 3.85E-04
10 300 11.2 0.4 1.39E-05 1.34E-04 5.99E-05 2.72E-05 9.00E-05 4.12E-05
10 300 18.2 0.4 6.79E-05 4.94E-04 1.95E-04 7.09E-05 4.33E-04 1.39E-04
10 300 27.3 0.3 2.79E-04 1.36E-03 4.57E-04 1.80E-04 1.68E-03 3.89E-04
10 300 27.3 0.3 2.70E-04 1.35E-03 4.86E-04 1.83E-04 1.55E-03 3.90E-04
10 300 27.2 0.4 1.78E-04 1.33E-03 5.24E-04 2.25E-04 1.35E-03 3.88E-04

4
R.E. Vieira and S.A. Shirazi Wear 476 (2021) 203654

Table 2
Experimental and predicted erosion results for churn flow; ER Probe data; μL = 1 cP and 10 cP.
μL (cP) dP (μm) vSG (m/s) vSL (m/s) Data (mm/ Original 1-D (mm/ Modified 1-D (mm/ Modified 2-D (mm/ Salama (mm/ DNV (mm/
kg) kg) kg) kg) kg) kg)

1 300 10.7 0.24 8.30E-06 1.31E-04 3.82E-05 1.16E-05 1.36E-04 3.54E-05

1 300 22.3 0.13 1.05E-04 7.84E-04 9.67E-05 2.50E-05 1.87E-03 2.26E-04
1 300 27.9 0.10 1.20E-04 1.36E-03 1.02E-04 4.33E-05 4.33E-03 4.03E-04
1 150 12.0 0.23 3.56E-06 7.62E-05 2.46E-05 8.60E-06 9.46E-05 4.67E-05
1 150 30.0 0.09 1.76E-05 8.36E-04 1.26E-04 1.67E-05 2.76E-03 4.86E-04
1 150 22.1 0.14 1.67E-05 3.90E-04 8.36E-05 1.51E-05 8.38E-04 2.22E-04
10 150 27.6 0.13 2.01E-05 7.00E-04 1.58E-04 2.24E-05 1.70E-03 3.92E-04
10 150 21.2 0.19 2.31E-05 3.48E-04 9.67E-05 2.23E-05 5.86E-04 2.02E-04
10 150 11.1 0.27 6.43E-06 5.28E-05 1.89E-05 3.27E-06 6.80E-05 3.90E-05
10 300 10.7 0.27 1.44E-05 1.33E-04 4.45E-05 1.46E-05 1.23E-04 3.56E-05
10 300 21.5 0.19 1.02E-04 7.77E-04 2.06E-04 4.71E-05 1.22E-03 2.08E-04
10 300 27.4 0.13 9.86E-05 1.38E-03 2.74E-04 5.48E-05 3.37E-03 3.88E-04
10 300 50.3 0.46 1.97E-04 5.74E-03 2.43E-03 1.30E-03 6.54E-03 1.90E-03
10 300 10.7 0.57 8.62E-06 1.10E-04 5.64E-05 2.55E-05 6.30E-05 3.76E-05
10 300 18.0 0.39 6.12E-05 4.76E-04 1.96E-04 7.05E-05 3.91E-04 1.35E-04
10 300 27.0 0.31 1.70E-04 1.32E-03 4.65E-04 1.66E-04 1.54E-03 3.78E-04
10 20 15.2 0.10 7.48E-05 7.49E-05 6.64E-05 3.00E-07 6.72E-05 3.60E-05

less difficulty. In two-phase flow dynamics, information required for

ρM = ρL (1 − α) + ρG α (9)
engineering problems usually comes from the response of the mixture as
a whole, rather than two separate responses of each phase [6]. The drift
μM = μL (1 − α) + μG α (10)
flux model, despite being less rigorous than other more detailed
two-phase flow models, is extremely important since it allows proper In this work, the modified characteristic particle initial velocity ob­
prediction and identification of the physical structure of the flow in a tained from Eq. (8) is used for calculate erosion in churn flows using 1-D
relatively simple way. and 2-D models previously described in Section 3.

4.2. Characteristic particle initial velocity 5. Comparison with experimental data

In churn flow, it is assumed that the sand is uniformly distributed in In this section, the modified 1-D and 2-D models are first validated by
the liquid phase. Similarly, the velocities of the liquid and sand are the experimental database presented in Section 2.2. Then, the erosion
assumed to be the same. Therefore, the characteristic initial sand par­ models of other researchers [11,12] are compared to the modified in this
ticle velocity V0 can be estimated by finding the velocity of the liquid work. All flow conditions, measured experimental UT [2] and ER probe
phase in churn flow. The modification proposed here is to use a drift-flux [23] erosion data, and predictions from the models are listed in Tables 1
model to predict the liquid velocity vL . and 2. Liquid viscosity (μL) of 1 cP indicates water while that of 10 cP
Following the drift-flux approach, the actual gas velocity vG can be means water + CMC.
expressed with the following expression: A variety of statistical parameters could be used to evaluate the
model predictions using the entire data bank. Following are the defini­
vG = C0 (vSG + vSL ) + vD (5)
tions of the statistical parameters used in this study.
Using data from two-phase flow in a pipeline-riser system [29], The average percentage error is given by:
Tengesdal et al. [30] determined C0 to be 1.0. More recently, Sharaf
et al. [31] gathered data for C0 for churn flow from their WMS experi­ 100 ∑
N
E1 = eri (11)
ments in 127 mm I.D. pipes and seven datasets from literature also found N i=1
that most of the data are reasonably approximated by 1.0. On the whole,
they lay in a narrow band between 0.9 and 1.1 and showed no strong where the relative error, eri , is given by:
dependence with liquid superficial velocity, pipe diameter and physical (
xpred,i − xdata,i
)
properties. eri = (12)
xdata,i
In this work, the drift velocity vD is determined based on churn flow
data provided in the Tulsa University Fluid Flow Projects (TUFFP) and N is the number of cases. E1 indicates the overall trend of the per­
expanded well data bank [30]: formance relative to the measured erosion.
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ A deviation coefficient E2 is employed to estimate the systematic
gD(ρL − ρG )
vD = 0.28 (6) deviation between predictions and experimental data, given as:
ρL ⃒ ⃒
N ⃒
1 ∑ xdata − xpred ⃒
The overall void fraction can be expressed as: E2 = × 100% (13)
N i=1 xdata + xpred
vSG vSG
α= = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ (7) The effectiveness of each model can be quantified using the fraction
vG (v + v ) + 0.28 gD(ρρL − ρG )
SG SL
L of variance unexplained (FVU), which is given by:
Given that the sand particles are being carried by the liquid phase, ∑
[x(n) − ̂x (n)]2
the initial sand particle velocity can be expressed as: FVU = ∑ (14)
[x(n) − x]2
vSL
V0 = vL = (8)
1− α where x(n) is the experimental observation, ̂ x (n) is the predicted value
The mixture density and viscosity can be expressed as: from the correlation, and x is the mean of the experimental observations.
An FVU of zero indicates a perfect prediction.

5
R.E. Vieira and S.A. Shirazi Wear 476 (2021) 203654

Table 4
Performance of predicting in churn flows; UT data; μL = 10 cP.
Model E1 (%) E2 (%) FVU

Original 1-D 908.4 76.6 41.4

Modified 1-D 524.4 52.1 2.9
Modified 2-D 31.5 21.9 0.2

Fig. 3. Comparison between the original and modified models; UT data; μL =

1 cP.

Table 3
Performance of predicting in churn flows; UT data; μL = 1 cP.
Model E1 (%) E2 (%) FVU

Original 1-D 402.6 64.2 65.3

Modified 1-D 131.9 33.8 4.9
Fig. 5. Comparison between the original and modified models. ER Probe Data
Modified 2-D 30.7 25.2 2.7
μL = 1 cP and 10 cP.

Table 5
Performance of predicting in churn flows; ER Probe data; μL = 1 cP and 10 cP.
Model E1 (%) E2 (%) FVU

Original 1-D 1513.4 80.3 628.0

Modified 1-D 335.7 51.6 86.0
Modified 2-D 32.8 30.1 20.7

Fig. 4. Comparison between the original and modified models; UT data; μL =
10 cP.
5.1. Comparison between the original and modified models
The comparison between the original one-dimensional model
(Original 1-D) and modified models (Modified 1-D and Modified 2-D) is
shown in Fig. 3. The performances of these three models for churn flow
with UT data collected for μL = 1 cP are listed in Table 3. In Fig. 3, the
solid line is the perfect line where the predicted results equal the
experimental data. Moreover, if a symbol is above this line, it means that
for that specific case, the erosion model overpredicts the corresponding
experimental data point and vice versa.
As observed from Fig. 3 and Table 3, the accuracies of all cases are
improved after employing the new approach. In general, the modified
models show advantages in improving the accuracies of predictions,
compared with the original model. The Modified 2-D model has all the
lowest error values. The average deviation with the experiments de­
clines by 30.4% and 39% for 1-D and 2-D models, respectively. Despite
few underpredictions observed for the Modified 2-D model, the results
do nevertheless suggest that the model has the best accuracy. Under­
standing that underprediction of erosion is not tolerable for safety rea­
sons, the Modified 1-D erosion model is considered to provide the best
predictions of all three models.
The comparison between the models for churn flow with a liquid
viscosity of μL = 10 cP is shown in Fig. 4. The accuracies of the original
and modified models are listed in Table 4 for comparison.
As observed from Table 4, the accuracy of the 2-D Modified model for
μL = 10 cP is clearly the best, showing in these cases the lowest error
values.
The set of ER probe experimental results shown in Table 2 is also
used to evaluate the models. The comparison between the models for
churn flow is shown in Fig. 5. The accuracies of the original and
modified models are listed in Table 5 for comparison.
It is shown in Fig. 5 that the proposed model modification is able to
avoid a significant over prediction of erosion using the 1-D model and
works well for most of the cases for the 2-D model. According to the
statistical parameters, the modified models perform better than original
model which uses mixture velocities.
5.2. Comparison between the modified models and other models
The comparisons between existing erosion models [11,12] and the

6
R.E. Vieira and S.A. Shirazi Wear 476 (2021) 203654

of multiphase flow.

6. Conclusions

With the understanding obtained through experimental studies,

modifications to the 1D and 2-D multiphase flow models developed by
E/CRC have been made to predict erosion under churn flows. The
notable conclusions are provided below:

• In this investigation, a drift-flux model is proposed to better predict

the flow behavior and the characteristic initial particle velocity for
churn flow.
• Based on the drift-flux model, the actual gas velocity is calculated as
a function of mixture and drift velocities. The coefficients of the drift
flux expression under churn flow conditions are determined from an
expanded well data bank available in literature.
• The new approach was implemented in the current 1-D and 2-D
multiphase erosion models and it was validated against the E/CRC
experimental UT and ER Probe database for churn flows as well as
Fig. 6. Comparisons between Salama [11] and modified models. UT and ER two models from literature. The operating conditions include
probe data; μL = 1 and 10 cP.
changes in the superficial fluid velocities, sand sizes, and liquid
viscosity.
• The modified 1-D and 2-D models are successful in predicting the
erosion for a variety of operating conditions, being the 2-D model
less conservative than the 1-D model for all the cases studied.
Continued improvements of the proposed mechanistic modeling are
still needed to take into account the effects of radius of curvature,
pressure and pipe diameter in churn flows.

Declaration of competing interest

The authors declare that they have no known competing financial

interests or personal relationships that could have appeared to influence
the work reported in this paper.

Acknowledgements

This material is based upon work supported by the E/CRC member

companies.

Fig. 7. Comparisons between DNV [12] and modified models. UT and ER probe References
data; μL = 1 and 10 cP.
[1] S.M. Bhagwat, A.J. Ghajar, A flow pattern independent drift flux model based void
fraction correlation for a wide range of gas–liquid two phase flow, Int. J.
Multiphas. Flow 59 (2014) 186–205.
Table 6 [2] M. Parsi, R.E. Vieira, N. Kesana, B.S. McLaury, S.A. Shirazi, Ultrasonic
Performance of predicting in churn flows; UT and ER probe data; μL = 1 and 10 measurements of sand particle erosion in gas dominant multiphase churn flow in
cP. vertical pipes, Wear 328 (2015) 401–413.
[3] Y. Zhang, B.S. McLaury, S.A. Shirazi, E.F. Rybicki, A Two-Dimensional Mechanistic
Model E1 (%) E2 (%) FVU Model for Sand Erosion Prediction Including Particle Impact Characteristics, NACE
International Annual Conference, San Antonio, TX, 2010. Paper No. 14741.
Modified 1-D 274.7 43.4 9.0
[4] T.A. Sedrez, S.A. Shirazi, Y.R. Rajkumar, K. Sambath, H.J. Subramani, Experiments
Modified 2-D 31.6 26.3 3.2
and CFD simulations of erosion of a 90 elbow in liquid-dominated liquid-solid and
Salama [11] 1332.4 69.1 155.8 dispersed-bubble-solid flows, Wear 426 (2019) 570–580.
DNV [12] 344.1 46.4 6.9 [5] E. Pagan, W.C. Williams, S. Kam, P.J. Waltrich, A simplified model for churn and
annular flow regimes in small-and large-diameter pipes, Chem. Eng. Sci. 162
(2017) 309–321.
modified models using UT and ER probe data for μL = 1 cP and 10 cP are [6] M. Ishii, T. Hibiki, Thermo-fluid Dynamics of Two-phase Flow, Springer, 2006.
[7] N. Zuber, J. Findlay, Average volume concentration in two phase systems, ASME J.
shown in Figs. 6 and 7. The performances of the erosion models for
Heat Transf. 87 (1965) 453–468.
churn flows are summarized in Table 6. [8] N.R. Kesana, R.E. Vieira, M. Parsi, B.S. McLaury, S.A. Shirazi, A mechanistic model
As observed from Table 6, among the four models, the Modified 2-D for predicting erosion in pseudo slug flow, in: Proc. of: NACE 2014 Conference &
model presents the best accuracy of all, being less conservative than the Expo, 2014.
[9] R.E. Vieira, N.R. Kesana, S.A. Shirazi, B.S. McLaury, Experiments for Sand Erosion
others. Also, the erosion model of DNV has a similar accuracy compared Model Improvement for Elbows in Gas Production, Low-Liquid Loading and
with the Modified 1-D model, while the erosion model of Salama [11] Annular Flow Conditions. In CORROSION 2014, NACE International, 2014.
tends to overpredict the data. It should also be noted since the current [10] M. Parsi, M. Kara, P. Sharma, B.S. McLaury, S.A. Shirazi, Comparative Study of
Different Erosion Model Predictions for Single-phase and Multiphase Flow
model is based on mechanistic models for multiphase flow it is expected Conditions. In Offshore Technology Conference, May, Houston, Texas, USA, 2016,
that the range of its applicability to other conditions be significantly pp. 2–5.
better than other models in the literature that do not account for details [11] M.M. Salama, An alternative to API 14E erosional velocity limits for sand-laden
fluids, J. Energy Resour. Technol. 122 (2) (2000) 71–77.

7
R.E. Vieira and S.A. Shirazi
[12] Det Norske Veritas, Recommended Practice RP 0501: Erosive Wear in Piping
Systems, 2007.
[13] Q.H. Mazumder, S.A. Shirazi, B.S. McLaury, A Mechanistic Model to Predict
Erosion in Multiphase Flow in Elbows Downstream of Vertical Pipes. In
CORROSION 2004, NACE International, 2004.
[14] S.A. Shirazi, J.R. Shadley, B.S. McLaury, E.F. Rybicki, A procedure to predict solid
particle erosion in elbows and tees, J. Pressure Vessel Technol. 117 (1995) 45,
1995.
[15] B.S. McLaury, S.A. Shirazi, An alternate method to API RP 14E for predicting solids
erosion in multiphase flow, ASME J. Energy Resour. Technol. 122 (2000) 115–122.
[16] S.A. Shirazi, B.S. McLaury, H. Arabnejad, A Semi-mechanistic Model for Predicting
Sand Erosion Threshold Velocities in Gas and Multiphase Flow Production, in: SPE
Annual Technical Conference and Exhibition, vols. 26–28, UAE, Dubai, 2016.
September.
[18] R.E. Vieira, Sand Erosion Model Improvement for Elbows in Gas Production,
Multiphase Annular and Low-Liquid Flow, Ph.D. Dissertation, Department of
Mechanical Engineering, The University of Tulsa, Tulsa, Oklahoma, USA, 2014.
[19] R. Kang, H. Liu, An integrated model of predicting sand erosion in elbows for
multiphase flows, Powder Technol. 366 (2020) 508–519.
[20] T.A. Sedrez, A New Methodology to Predict Erosion in Liquid-Dominated Flows by
Computational Fluid Dynamics (CFD) Based on Experiments. Doctoral Dissertation,
The University of Tulsa, 2020.
[21] O. Kvernvold, R. Sandberg, Production Rate Limits in Two-phase Flow Suystems:
Erosion in Piping Systems for Production of Oil and Gas, Technical Report No. 93-
3252. Det Norske Veritas (DNV), Norway, 1993.
[22] P.M. Birchenough, S.G.B. Dawson, T.J. Lockett, P. McCarthy, Critical Flow Rates
Working Party. Report No. AEA-TSD-0348, AEA Technology, UK, 1995.
Wear 476 (2021) 203654
[23] R.E. Vieira, M. Parsi, P. Zahedi, B.S. McLaury, S.A. Shirazi, Ultrasonic
measurements of sand particle erosion under upward multiphase annular flow
conditions in a vertical-horizontal bend, Int. J. Multiphas. Flow 93 (2017) 48–62.
[24] R.E. Vieira, M. Parsi, P. Zahedi, B.S. McLaury, S.A. Shirazi, Electrical resistance
probe measurements of solid particle erosion in multiphase annular flow, Wear 382
(2017) 15–28.
[25] M.A. Woldesemayat, A.J. Ghajar, Comparison of void fraction correlations for
different flow patterns in horizontal and upward inclined pipes, Int. J. Multiphas.
Flow 33 (4) (2007) 347–370.
[26] P.V. Godbole, C.C. Tang, A.J. Ghajar, Comparison of void fraction correlations for
different flow patterns in upward vertical two-phase flow, Heat Tran. Eng. 32 (10)
(2011) 843–860.
[27] A.A. Gadgil, R.E. Randall, Two Phase Annular Flow Approximation Using 1-D Flow
Equations Coupled with a Drift Flux Model for Concurrent Flow in Vertical or Near
Vertical Channels, in: International Conference on Offshore Mechanics and Arctic
Engineering vol. 57694, ASME, 2017. V05AT04A048.
[28] Paula D. Pico, Juan P. Valdés, Nicolás Ratkovich, Eduardo Pereyra, Experimental
and CFD Modelling of the Drift Flux in Two-phase Air-(non) Newtonian Slug-Flow
Pattern Flow along Horizontal and Inclined Pipelines. Proceedings of the 3rd World
Congress on Momentum, Heat and Mass Transfer (MHMT’18), Budapest, 2018.
Hungary – April 12 - 14, 2018.
[29] Z. Schmidt, Experimental Study of Two-phase Flow in a Pipeline-Riser Pipe System,
PhD dissertation, University of Tulsa, Tulsa, Oklahoma, 1977.
[30] J.Ø. Tengesdal, A.S. Kaya, C. Sarica, Flow-pattern transition and hydrodynamic
modeling of churn flow, SPE J. 4 (1999) 342–348.
[31] S. Sharaf, G.P. van der Meulen, E.O. Agunlejika, B.J. Azzopardi, Structures in
gas–liquid churn flow in a large diameter vertical pipe, Int. J. Multiphas. Flow 78
(2016) 88–103.