Investigations on the effect of tip clearance gap and inducer on the

transport of air-water two-phase flow by centrifugal pumps

Trupen Parikha , Michael Mansoura,b,∗, Dominique Thévenina

a Lab. of Fluid Dynamics & Technical Flows, University of Magdeburg “Otto Von Guericke”, 39106 Magdeburg, Germany
b Mechanical Power Engineering Department, Faculty of Engineering - Mattaria, Helwan University, 11718 Cairo, Egypt

Abstract

This study presents numerical and experimental investigations on air-water flows in a centrifugal pump.
The main objectives are to examine the accuracy of common numerical models, and to explain the influence
of installing an upstream inducer or increasing the tip clearance gap on the performance. 3D transient
simulations were performed using the Volume of Fluid (VOF) method for two-phase modeling, and the k−ω
SST model for turbulence modeling. However, the simulations could not predict the sudden head drop
occurring at 5% gas content. Overall, the averaged relative errors between the simulations and the experi-
ments for part-load, optimal, and overload flow are 3.8% and 10.4% and 45%, respectively. The use of the
inducer or increasing the gap leads to improvements in performance and phase mixing. Finally, bubble size
distributions were analyzed using shadowgraphy, confirming that the inducer and the increased gap provide
finer bubbles with narrower distributions.

Keywords: Centrifugal pumps, Pump inducer, Gas-liquid two-phase flow patterns, Computational Fluid
Dynamics (CFD), Volume of Fluid model (VOF)

1. Introduction
Centrifugal pumps transporting two-phase flows are broadly employed in several engineering and indus-
trial applications [1, 2]. For instance, pumping of two-phase flows appears in the petroleum industry for
artificial lifting [3–6], nuclear power stations [7–9], geothermal power stations [10], agriculture systems for
irrigation purposes [11], coolant pumping for refrigeration [12], multiphase pumping in chemical industries,
medical treatment, paper industry, shipbuilding industry, food production, oil industry and wastewater
treatments [13, 14] and mini hydropower plants [15].
The transport of two-phase gas-liquid flows by centrifugal pumps can cause several problems starting
from evident performance deterioration as compared to the reliable single-phase flow performance. In most
cases, at gas volume fractions (ε) between 5% and 10%, malfunction and pump ”breakdown” is unavoidable,

∗ Correspondingauthor
Email address: [Link]@[Link] (Michael Mansour )

Preprint submitted to Chemical Engineering Science January 2, 2020

 Nomenclature

Roman symbols R Gas constant of air [J/kg K]

2
A Cross-sectional area, m S Impeller tip clearance gap [m]
AD Cross-sectional area of discharge line [m2 ] Si Inducer tip clearance gap [m]
2
Af Grid face area, m Su Surface uniformity
2
AS Cross-sectional area of suction line [m ] T Flow temperature [K]
b1 Blade inlet width [m] V1 Suction pipe superficial flow velocity [m/s]
b2 Blade outlet width [m] V2 Discharge pipe superficial flow velocity [m/s]
c Local grid face volume fraction Va Inlet velocity of air, m/s
c̄ Surface-averaged volume fraction Vw Inlet velocity of water, m/s
+
D32 Sauter mean bubble diameter y Dimensionless wall distance
Dmedian Median bubble diameter z1 Suction elevation [m]
2
g Gravitational acceleration [m/s ] z2 Delivery elevation [m]
H Pump head [m]
Greek Symbols
ṁa Air mass flow rate [kg/s]
∆P Static pressure difference (P2 -P1 ) [Pa]
ṁt Total mass flow rate (ṁa +m˙w ) [kg/s]
∆Popt Static pressure difference at optimal flow [Pa]
m˙w Water mass flow rate [kg/s]
ε (Inlet) Gas volume fraction [%]
Mc Mixing coefficient
η Pump efficiency [%]
n Rotational speed [min−1 ]
µ̇ Gas mass fraction [%]
ns Specific speed [min−1 ]
ω Angular speed [rad/s]
P1 Suction static pressure [Pa]
ρa Air density [kg/m3 ]
P2 Discharge static pressure [Pa]
ρw Water density [kg/m3 ]
PP Pump power [W]
σ Standard deviation of size distributions
PSh Shaft power [W]
τ Torque [N m]
Q Volume flow rate [m3 /s]
τopt Torque at optimal flow [N m]
Qa Air volume flow rate [m3 /s]
Υ Specific delivery work [m2 /s2 ]
Qcr Q-criterion, s−2
Υmax Maximum specific delivery work [m2 /s2 ]
Qopt Optimal (nominal) flow rate [m3 /s]
Υopt Optimal specific delivery work [m2 /s2 ]
Qt Total volume flow rate (Qa +Qw ) [m3 /s]
Qw Water volume flow rate [m3 /s]

particularly in part-load flow conditions, where the pump is completely unable to convey the mixture [1, 16].
The main reason for that is the fast accumulation of the gas phase on the blades within the impeller
[11, 17, 18]. When the gas starts to accumulate on the blades, the impeller rapidly becomes unable to
provide kinetic momentum to the mixture, which is known as ”gas locking”. The performance degradation is
influenced by several parameters, including the physical properties of the fluids, the gas volume fraction, the
inlet pressure, and the flow temperature or the thermal conditions. Additionally, the design parameters like
the impeller geometry and rotational speeds are also quite influential [1, 5, 16, 17, 19, 20]. The performance
degradation is accompanied of course by a reduction in all pump parameters such as the efficiency, the flow

2
rate, and the shaft torque [1, 8]. These problems have motivated substantial experimental and numerical
research by numerous authors to explain the flow features and corresponding two-phase patterns within the
impeller, considering the effect of various parameters on the pump performance.
In recent studies [1, 5, 16, 21], the flow within the impeller was observed by high-speed imaging for
increasing gas volume fraction. The two-phase flow patterns were commonly classified as bubbly flow,
agglomerated bubbles flow, gas pocket flow, alternating pocket flow, and segregated flow. The bubbly flow
is only observable at very low gas volume fractions (mostly lower than 1%) where the bubbles do not interact
significantly, causing a limited decrease in pump performance. At slightly higher gas volume fractions the
bubbles tend to agglomerate forming bigger bubbles within the impeller, leading to a further reduction in
the performance. At gas volume fractions around 3% [16], a big gas pocket shows up mostly on the blade
suction side. This leads to sudden and steep deterioration in pump performance curves. For some specific
flow conditions, the size and the location of gas pockets become very unstable [1, 22], which causes severe
flow instabilities in pump operation and strong cyclic pressure fluctuations; a phenomenon commonly known
as ”pump surging” [23, 24]. At high enough gas contents (about 9%), the segregated flow regime occurs in
closed impellers, at which the pump shows very low performance. The segregated flow regime could not be
observed in semi-open impellers even at very high gas volume fractions, due to the strong bubble break-up
that occurs in these impellers near the blade trailing edges [1, 16].
The effect of fluid viscosity was considered in some studies, showing that the increase in fluid viscosity
results mainly in increased performance degradation [24–26]. This is due to the reduced turbulence levels,
which allows an easier formation of large bubbles and eventually pump performance degradation. Never-
theless, in a study of oil-gas mixtures, the increase of oil viscosity resulted in an improvement of pumping,
which was more evident in overload operation, since the drag exerted on the bubbles increases, reducing
phase separation [27]. However, another study showed that the viscosity effects are becoming negligible
for rotational speeds higher than 600 rpm as the centrifugal forces dominate the flow [28]. Generally, the
increase of the rotational speed delays the performance degradation as it churns big bubbles into smaller
bubbles, allowing the pump to handle higher gas volume fractions [5, 6, 29]. The relationship between
the pump performance and the flow patterns was studied recently in [5] for different rotational speeds. It
was shown that the performance of the pump is improved and the flow regime transitions are delayed at
higher rotational speeds. Further, it was also shown that higher gas density postpones the drop of the
pump performance, while higher surface tensions were found to be negative since bubble breakup is less in
this case [6]. Another interesting study involved introducing surfactants in the air-water flow, helping to
reduce the surface tension and as a consequence reducing head degradation [30]. Recently, it was shown that
the occurrence of large separation zones is the main reason for the development of big gas accumulations
in two-phase flows [31]. Additionally, it was also shown that the gas accumulation can be decreased by
providing high turbulence levels to disturb the accumulated pockets.

3
 The effect of geometrical parameters on the performance and the flow patterns was also considered in
the literature. Based on the study of Murakami and Minemura [32], an impeller with 5 to 7 blades is
effective concerning single and two-phase pumping. Sato et al. (1996) [17] showed that the pump head can
keep higher head under gas accumulations when larger outlet blade angles are used. Furukawa et al. [33]
used a combination of rotating and stationary cascades to delay gas accumulation by crushing the bubbles.
Mustafa et al. [34] showed that the energy efficiency of the pump can be increased by using splitter blades
with specific designs.
The use of a split vane impeller pump (SVIP) was also found beneficial for achieving higher phase
mixing and finer bubbles across the pump [35, 36]. In one of our previous studies [16], it was shown that
a semi-open impeller has much higher resistance against gas accumulations in the range of 1% ≥ ε ≥ 3%,
compared to a geometrically similar closed impeller. The improved gas accumulation resistance of semi-
open impellers is due to the tip-clearance flow, which helps to disturb the big bubbles and improve the
phases mixing. Additionally, the flow instabilities and pump surging are more limited and weaker in the
semi-open impellers. The size of the tip clearance of a semi-open impeller has also a major effect on pump
performance. Although the increase of the tip clearance gap leads normally to a decrease of the single-phase
flow performance, it provides much better mixing for two-phase pumping together with higher resistance
against gas accumulation [1, 37].
Though numerous experimental investigations are available in the literature, detailed and minute flow
features are quite complex and/or expensive to be delivered experimentally. Accordingly, investigations
based on computational fluid dynamics (CFD) have become very common. There are mainly two standard
approaches used in CFD studies to model the gas-liquid interactions, i.e. the two-fluid (Euler-Euler) model
[38, 39] and the Volume of Fluid (VOF) model [40]. The two-fluid model solves two different sets of
conservation equations for each phase separately along with models to couple the two sets (inter-phase
models), while in VOF model, one single set of conservation equations is solved together with an additional
transport equation for the phase volume fraction.
Pineda et al. [41] investigated the phase distribution numerically in an Electrical Submersible Pump
(ESP) impeller using the Volume of Fluid (VOF) model. The analysis was able to capture the basic essence
of head degradation with the increase of the gas volume fraction, showing also better mixing at higher
rotational speeds. Another study [42] compared Euler-Euler model with the VOF model, revealing that
the Euler-Euler model is superior in some cases, since it is able to capture the phase velocity slip more
accurately. Huang et al. [43] successfully studied the self priming process of a centrifugal pump using the
Eulerian multiphase model. Nevertheless, the Euler-Euler approach is always very sensitive to the inter-
phase models and the bubble size used in the simulations. Strong deviations of the results occur when the
inter-phase models or the bubble size are inappropriately selected [18, 44–46]. On top of that, there is no
standard combination of inter-phase models that can be generally applied for different flow conditions [47].

4
 This confirms that most numerical models are still in the developing phase, and need detailed exper-
imental characterizations. For these reasons, CFD results are only efficient if they are validated against
corresponding measurements. Thus, this present work involves detailed experimental and numerical analy-
ses of two-phase air-water flows in a centrifugal pump. In our previous work [1], performance enhancement
was investigated experimentally by installing an upstream inducer or increasing the tip clearance gap of the
impeller. Additionally, in [48, 49], we performed numerical investigations on the single and two-phase flows
across a pump inducer, revealing improved mixing of the two phases downstream of the inducer, particularly
in part-load conditions. The experimental and numerical analyses are extended here by considering three
(full) pump configurations, i.e the pump with a standard gap, increased gap, and a standard gap together
with an inducer. Accordingly, additional details about the effect of the pump inducer and tip clearance
gap on the two-phase flow behaviour are obtained. The pump performance, the internal flow details, the
two-phase mixing in the impeller, and the bubble size distributions are investigated for the three pump
configurations. The study involves also comparisons of the predicted two-phase flow patterns against exper-
imentally recorded high-speed images of the flow, discussing the prediction accuracy and the limitations of
the CFD simulations.

2. Experimental Set-up

Figure 1 shows a 3D view of the test-rig of the pump with the inducer installed. The inlet pipe to the
pump is connected to a 6.3 m3 tank to provide a closed-loop operation. The water flow rate Qw is measured
before being mixed with air by an electro-magnetic flow meter (Endress+Hauser Promag 30F, with ± 0.5 %
RD accuracy), and controlled by a motorized gate valve installed on the discharge line. Air is injected at the
center of the inlet pipe via a gas distribution nozzle as indicated on Figure 1. The air flow rate is measured
by a mass-flow meter (Bronkhorst F-113AC-HD-55-V, with ± 0.5 % RD plus ± 0.1 % FS accuracy), and
controlled by a throttle valve installed on the air inlet pipe. Two absolute pressure sensors (Sensotec Z, with
± 0.25 % FS accuracy) P1a and P1b are installed on the inlet pipe to measure the suction pressure of the
pump while running without and with the inducer, respectively. Another similar absolute pressure sensor
P2 is placed downstream on the outlet pipe to measure the discharge pressure change after the pump.
A temperature sensor (Pt100 Industrial Sensor Probe, Class B, with ± 0.3 K maximum absolute error)
placed near the injected air is used to measure the flow temperature T , which is used together with the
suction pressure (P1 ) to calculate the air volume flow rate (Qa ) and the air volume fraction (ε). The pump is
driven by a variable speed electric motor (5.7 kW, with 3000 rpm maximum rotational speed). The rotational
speed is further controlled by a frequency controller and measured by an analog tachometer (TDP 0.2 LT -
4, with 1% RD accuracy). A torque sensor (Model: HBM T1, with an accuracy of 0.4% FS) is installed on
the motor shaft to measure the shaft torque and obtain the pump efficiency. Further details regarding the
entire pump test-rig and flow loop can be found in our previous publications [1, 16].

5
 The whole pump casing and the impeller were made from acrylic glass to maximize optical accessibility.
The two-phase flow patterns were recorded by using a high-speed camera (Model: Imager pro HS 4M CCD)
with 2016 x 2016 pixels resolution. To illuminate the flow inside the pump, 4 LED lamps were distributed
around the pump casing as shown in Figure 1, so that the air-water interfaces can be clearly seen. The
windows where the two-phase flow images and the bubble size distribution (BSD) were observed are indicated
by dashed rectangles in Figure 1.

Outlet P2
flow
Pressure
Acrylic pump sensor
casing
LED
lamps
Shadowgraphy
window

Impeller

Suction
pressure P1a
sensors
P1b
Inducer

Inlet flow

Air injection Flow visualization

window

Figure 1: Details of the experimental set-up.

Figures 2a and 2b show cross-sectional views for the impeller without the inducer and with the inducer,
receptively. As shown, the inducer can be easily installed by replacing the impeller nut, while all other pump
geometrical parameters remain identical. The tip clearance gap can be set by adjustable rings installed
behind the impeller as shown in Figure 2a. The impeller was designed with 6 non-twisted blades to allow
maximum flow visualization, while the inducer has 3 helical blades to provide low inlet solidity [19]. The
front shroud of the impeller is fixed on the pump body, and the impeller is installed with a standard tip-
clearance gap (S/b2 = 2.5%). The “radial” tip clearance gap of the inducer is Si /b2 = 12.5%, where b1 and
b2 are the impeller blade inlet and outlet widths. Further geometrical details of the impeller and inducer are

6
 (a) Sectional view for the pump without inducer. (b) Sectional view for the pump with inducer

Figure 2: Details of the impeller and the inducer.

listed in Table 1, in which some parameters are given as a ratio to the blade outlet width b2 or the suction
pipe diameter DS . Since the impeller components are made of acrylic glass, a low rotational speed of 650
rpm was kept for the present investigations, avoiding strong vibrations occurring at higher speeds. In this
case, the specific speed of the pump is approximately nq = 21 min−1 , calculated by Equation (1):

√
n Q
nq = 3/4 (1)
H
where n is the rotational speed of the pump in (rpm), Q is the volume flow rate in (m3 /s) and H is the pump
head in (m). The suction pressure of the experiments was always between 1.28 and 1.14 bar, corresponding
to the minimum and the maximum flow rates, respectively.

Table 1: Impeller geometrical dimensions.

Parameter Symbol Value

Impeller blade inlet angle (see [48] for details) β1 24°
Impeller blade outlet angle β2 24°
Inducer hub to tip diameter ratio Dh /Dt 41.85 %
Inducer hub solidity (Blade chord to pitch at hub) σh = Ch /Ph 2.836
Inducer tip solidity (Blade chord to pitch at tip) σt = Ct /Pt 1.807
Inducer area solidity σa 23.326 %
Inducer hub blade angle (see [49] for details) θh 34°
Inducer tip blade angle θt 58°
Inducer blade axial length Lb /DS 1.0
Impeller tip clearance gap S/b2 2.5%
Inducer tip clearance gap Si /b2 12.5% mm

7
 To provide further details about the two-phase mixing offered when installing the inducer or increasing
the gap, the bubble size distributions were measured in the impeller for the three pump configurations by
using shadowgraphy measurements as presented in Figure 3a. The LED light source has been installed
behind the pump opposite to the camera. In this way, the interface between the water phase and the air
bubbles appears dark, so that bubble size can be clearly determined. The shadowgraphy images have been
obtained by using the same high-speed camera as employed for the flow regimes measurements. A specific
window in the impeller channels has been selected and considered as a representative area for the BSD.
The location of the window, where the bubbles were observed is shown in Figure 3b. The bubbles moving
through the area marked by the red border in Figure 3b were then masked and analyzed.

LED
Light
Source

Test Section

CCD
Camera
(a) Shadowgraphy set-up (top view). (b) Shadowgraphy Window (front view).

Figure 3: Explanation of the shadowgraphy measurements for the bubble size distribution.

3. Numerical Modeling

To perform the 3D numerical simulations, the commercial CFD code Siemens STAR-CCM+ V13.02 [50]
was used. A transient set-up was always applied together with a moving-mesh approach to effectively model
the impeller and inducer rotation. The k −ω shear stress transport (SST) model [51] was used for turbulence
modeling. This model is often used in the literature for pump simulations since it can predict strong pressure
gradients and flow separations appearing in the impeller [46, 52–55]. Additionally, the k − ω SST model
delivered in some configurations the best agreement with experiments concerning air-water two-phase flows
in pipes compared to other turbulence models, see e.g., [56]. Further, the k −  model was not preferred
here as it cannot accurately predict strong streamlines curvatures, which results in big deviations from the
experiments in centrifugal pumps [57]. A second-order upwind convection scheme was used together with
a segregated solver. The continuity, the momentum, and the (isothermal) energy balances are described by
Equations (2), (3), and (4), respectively.

8
 ∂ρ ~)=0
+ ∇ · (ρV (2)
∂t
~
DV
ρ ~ · p + µ∇
= ρ~g − ∇ ~2·V
~ (3)
Dt

∂(ρE) ~ ~)=V
~ · ρ~g − ∇
~ · (pV
~)+∇
~ · (τ · V
~)
+ ∇ · (ρE V (4)
∂t

~ is the continuum velocity, ~g is the acceleration, p is the pressure, µ is the

where ρ is the fluid density, V
fluid viscosity, E is the total energy per unit mass, and τ is the viscous stress tensor. Further details can
be found in [50].
As mentioned in the introduction, the Volume of Fluid (VOF) method was used to model the air-water
phase interactions. It was also shown recently that the combination of the VOF and the k − ω SST models
leads to very good agreement with the experiments concerning air-water two-phase regimes and interactions
in helical pipes [58]. The VOF model solves a single set of conservation equations (mass, momentum
and energy equations) for both phases, assuming common velocity, pressure and temperature fields. An
additional equation is solved for phase volume fraction. The transport equation of the volume fraction Cm
of the mth phase can be given by Equation (5) for a flow of n phases (here, n = 2), where the summation
of volume fractions Cm is always equal to 1, as given by Equation (6).

∂Cm ~
+ V · (∇Cm ) = 0 (5)
∂t
Xn
Cm = 1 (6)
m=1
Primarily being an interface modeling technique, the VOF model demands the use of second-order
advection scheme to avoid smearing of the interface and excessive false diffusion. Thus, this was considered
for all the simulations. In all two-phase simulations, water was modeled as an incompressible fluid with
a density of ρw = 997.56 kg/m3 , a molecular weight of 18.015 kg/kmol, a specific heat of 4181.7 J/kgK,
a thermal conductivity of 0.62027 W/mK, and a dynamic viscosity of µw = 8.887 · 10−4 Pa · s. Air with
a molecular weight of 28.966 kg/kmol, a specific heat of 1003.6 J/kgK, a thermal conductivity of 0.02603
W/mK, and a dynamic viscosity of µa = 1.855 · 10−5 Pa · s was modeled as a compressible fluid using
the universal gas law, i.e. ρa = P/RT , where ρa is the air density, P is the local absolute pressure, R is
the gas constant of air, and T is the flow temperature (300 K). In the present investigations, the suction
pressure was limited in the range of 1.14–1.28 bar, and the maximum discharge pressure was always limited
to about 2 bar. Note that, air can be modeled as an ideal gas up to a pressure of 5 bar without involving
any significant error [59]. The surface tension between the two phases was taken into account using a value
of 0.072 N/m for the surface tension between air and water.

9
3.1. Simulation domain
Three pump configurations were considered in the present study, i.e. 1) pump with standard tip clearance
gap, 2) pump with standard tip clearance gap and inducer, and 3) pump with increased tip clearance gap.
The three cases are compared in detail to examine the improved flow mixing provided by the inducer and
the increased gap clearance. The details of the simulation domain are shown in Figure 4. As shown, the
simulation domain was basically divided into 3 parts; (1) the inlet pipe, (2) the rotating zone surrounding
either the impeller or the impeller and the inducer, and (3) the volute with the exit pipe, considering in-
place interfaces between them. The length to diameter ratios of the inlet and outlet pipes are approximately
4 and 5, respectively. Similar to the experiments, three pressure probes P1a , P1b , and P2 were placed in
corresponding locations upstream and downstream of the impeller.
For two-phase simulations, the inlet was divided into 2 surfaces; one for each phase, using a center circle
for the air inlet as can be seen in Figure 4. Additionally, in a comparable way to the experiments, the
radius of the air inlet surface was regulated based on the inlet gas volume fraction value, so that the inlet
velocities of air and water remain equal (Va = Vw ) for all simulations. To achieve that, the radius of the
√
air inlet surface was set to DS ε. Here, Qt is the total volume flow rate, representing the sum of air (Qa )
and water (Qw ) volume flow rates, respectively. The (inlet) gas volume fraction is calculated by Equation 7.
Concerning the boundary conditions of the simulation, a mass flow inlet and a pressure outlet condition
were applied. In the simulations, the outlet pressure value was set similar to the corresponding experimental
value to ensure that the air density is always comparable between the simulations and the experiments.
Qa Qa
ε= = (7)
Qt Qa + Qw

Pressure outlet

Rotating zone In-place interface

Ds

Inducer

Mass flow inlet In-place interface Impeller Volute

Figure 4: Simulation domain.

10
3.2. Grid-independence test

For the numerical grid, polyhedral elements were generated for the whole simulation domain, together
with 8 prism layers along all walls to ensure accurate modeling of the boundary layer flow. To eliminate the
use of wall models, the thickness of the first layer was kept sufficiently small in order to keep the average
value of the dimensionless wall distance (y + ) always below 1. Additional mesh refinement was applied
in the rotating zone, i.e. around the impeller and the inducer to improve the prediction in this zone of
central interest. A mesh-independence test was carried out for single-phase flow, considering different mesh
resolutions to minimize the numerical errors of space discretization.
Based on preliminary simulations, it was found that the mesh resolution is generally more influential
at higher flow rates due the increased turbulence, while a grid independence solution is achieved earlier at
lower flow. Accordingly, the maximum flow rate considered in the present study (overload conditions) was
used in the test, i.e. Qw /Qopt = 1.6, where Qopt is the optimal flow rate of the pump. In the experiments,
the cavitation onset was observed starting from Qw /Qopt = 1.65. Accordingly, a cavitation-free condition is
always confirmed. The details of the meshes used in the test are shown in Table 2.

Table 2: Details of different grids used for the mesh-independence test.

Mesh # Total No. of cells (millions) Average y +

1 0.7 0.97
2 1.5 0.69
3 2.5 0.53
4 5.5 0.35
5 9.0 0.27

Figure 5 shows the numerical results of the 5 meshes considered in terms of the normalized pressure
change across the pump (∆P/∆Popt ). All optimal values used for normalization in this paper correspond
to single-phase flow values of the one pump configuration. As shown, the pressure change shows a constant
behaviour between the fourth and fifth mesh with an error of less than 0.18%. Accordingly, the fourth
mesh with 5.5 million cells was used for all single-phase simulations. However, an additional test was done
to examine the meshes under two-phase conditions in the inlet pipe. Again, the maximum flow rate was
considered (Qw /Qopt = 1.6), together with a gas volume fraction of 3% as a sample value. The numerical
results of all meshes are presented in Figure 6 in terms of the gas volume fraction for a section located at
a distance of 2DS from the inlet surface. As can be seen, the air volume fraction is mostly smeared for
all low-resolution meshes (mesh 1 to mesh 3), and even – to some slight extent – for mesh 4; the air-water
interface is best resolved using mesh 5. Therefore, mesh 5 with 9.0 million cells was finally employed for all
two-phase simulations. Details of this mesh are illustrated in Figure 7.

11
 0.6

0.5

ΔP/ΔPopt
0.4

0.3

0.2
0 2 4 6 8 10
No. of cells (in millions)
Figure 5: Normalized pressure change for different grid resolutions.

Mesh 1 Mesh 2 Mesh 3 Mesh 4 Mesh 5

Figure 6: Comparison of the air volume fraction resolved by different mesh resolutions.

Inducer blade
clearance
Impeller tip
clearance

Figure 7: Mesh resolution used for two-phase simulations (9.0 million 3D cells).

12
3.3. Timestep independence test

The timestep of the simulation solver should be also assessed carefully. Thus, a timestep-independence
test was performed taking into consideration 3 different timesteps, namely 0.5° (0.128 ms), 1.0° (0.256 ms),
and 2.0° (0.512 ms) angle rotations per timestep, respectively. Similar to the mesh-independence test, the
maximum flow rate of Qw /Qopt = 1.6 was considered, ensuring timestep-independence at all other, lower
flow rates as well.
Figure 8 shows the plot for the normalized pressure change across the pump for different timesteps. It
can be seen that the difference in the results between 1.0° and 0.5° is very small. Therefore, the angular
rotation of 1.0° per timestep was considered to have a negligible influence on the numerical results and kept
for all further simulations. Additionally, a number of 25 inner iterations were set to confirm convergence
of all residuals at each time-step by a drop of approximately three magnitudes. The total simulation time
considered in single-phase simulations is 0.5 s, which corresponds to more than 5 complete impeller rota-
tions, while the two-phase simulations were stopped after at least 1.0 s of simulation time. For two-phase
simulations at part-load condition (Qw /Qopt = 0.5), 1.5 s (more than 16 impeller rotations) were required
for the simulation time in order to ensure that the air phase reaches the domain outlet in quasi steady-state.

Figure 8: Normalized pressure change for different timesteps considered.

All the numerical results shown later were averaged over one complete rotation of the pump (approxi-
mately 0.1 seconds) to obtain statistically stable values for all cases considered. Figure 9 shows pressure and
torque results, normalized by optimal values, for different flow conditions (part-load, optimal, and overload)
and ε = 3%, together with their cumulative averaged values over the last 0.5 seconds of the simulation time.
Here, the instantaneous fluctuations of the pressure typically exhibit an increasing behaviour with the flow
rate, while the torque fluctuations are higher for part-load and overload than those of optimal flow due to
the increased shock losses. As shown, the cumulative averaged properties reach constant values for all flow
cases after approximately 0.1 seconds, confirming proper averaging.

13
 1.20 0.70
1.10 0.65
1.00 0.60
opt

τ τopt
0.90 0.55
0.80 Instantaneous 0.50 Instantaneous
Cumulative averaged Cumulative averaged
0.70 0.45
1 1.1 1.2 1.3 1.4 1.5 1 1.1 1.2 1.3 1.4 1.5
Time (s) Time (s)
(a) Normalized pressure - Part-load (b) Normalized torque - Part-load

1.05 1.00
0.95 0.95
0.85 0.90

τ τopt
opt

0.75 0.85
0.65 Instantaneous 0.80 Instantaneous
Cumulative averaged Cumulative averaged
0.55 0.75
0.5 0.6 0.7 0.8 0.9 1 0.5 0.6 0.7 0.8 0.9 1
Time (s) Time (s)
(c) Normalized pressure - Optimal (d) Normalized torque - Optimal

0.50 1.15
0.40 1.10
0.30 1.05
τ τopt
opt

0.20 1.00
0.10 Instantaneous 0.95 Instantaneous
Cumulative averaged Cumulative averaged
0.00 0.90
0.5 0.6 0.7 0.8 0.9 1 0.5 0.6 0.7 0.8 0.9 1
Time (s) Time (s)
(e) Normalized pressure - Overload (f) Normalized torque - Overload

Figure 9: Instantaneous and cumulative averaged values over the last 0.5 seconds of simulation time for the pump with standard
gap.

4. Results and Discussion

4.1. Performance analysis

In this section, the numerical and experimental pump performances are compared in terms of the nor-
malized specific delivery work (Υ) and pump efficiency (η). The specific delivery work can be calculated
by Equation (8), considering the specific work of the liquid phase and the isentropic compression work of
the gas phase [1]. In Equation (8), µ̇ is the mass fraction of air (Equation 9), V1 is the mixture superficial
velocity at pressure sensor 1, V2 is the mixture superficial velocity at pressure sensor 2, g is the gravitational

14
acceleration, z1 is the elevation at pressure sensor 1, and z2 is the elevation at pressure sensor 2. The
complete derivation of Equation 8 can be found in [1]. In Equation (9), ṁa , ṁw , and ṁt are the air, the
water, and the total mass flow rates, respectively. The superficial velocities (V1 and V2 ) are calculated based
on the continuity equation as given by Equations (10) and (11), dividing the total flow rate (Qt ) by the
suction and discharge cross-sectional areas. The efficiency (η) is calculated by Equation (12), where PP is
the pump power, PSh is the shaft power, and ω is the angular speed (2πn/60).
 
1 − µ̇ P2 1
V2 2 − V1 2 + g (z2 − z1 )


Υ= (P2 − P1 ) + µ̇RT ln + (8)
ρw P1 2
ṁa ṁa
µ̇ = = (9)
ṁt ṁa + ṁw
Qt
V1 = (10)
AS
Qt
V2 = (11)
AD
PP ṁt Υ
η= = (12)
PSh τ ω
Figures 10 and 11 show the normalized specific delivery work (Υ/Υopt ) and the efficiency (η) of the
pump, respectively, with standard gap (left column), increased gap (middle column), and standard gap with
inducer (right column) as a function of the normalized volume flow rate (Qw /Qopt ) for different gas volume
fractions. Here, Qopt and Υopt represent the optimal flow rate and the optimal specific work of the pump,
respectively. All the simulation results were averaged over the last 0.1 seconds as discussed in the previous
section, while the experimental results were obtained from 200 instantaneous measurements recorded over a
period of 50 seconds (4 Hz acquisition frequency). The simulations show generally very good agreement with
the experiments in the range of ε = 0% to ε = 3%, particularly for part-load and optimal conditions. The
simulation points deviate more noticeably at overload conditions, which can be justified by the limitation
of numerical models to accurately predict the strong flow separation at the volute nose occurring at these
conditions. This will be further explained by visualizing the flow field in the following sections. Furthermore,
there is a slight overestimation of the performance in the simulations especially for single-phase flow with
the increased gap. This could be due to the inaccurate prediction of the increased secondary flow by the
simulations across the tip clearance. For ε = 5%, the simulation results slightly over-predict the performance
at part-load conditions, while the sudden performance drop corresponding to a flow regime transition in the
impeller occurring at higher flow rates could not be predicted by the simulation. In case of the increased
gap, the pump does not experience the performance drop for ε = 5% in the experiments, as the increased
gap positively delays the performance drop to ε = 7% [1]. Thus, the simulated performance curve for the
increased gap is still comparable with the experimental results at ε = 5%. Overall, the averaged relative
errors in the normalized specific delivery work between the simulations and the experiments for part-load,
optimal, and overload flow are 3.8% and 10.4% and 45%, respectively.

15
 Concerning the effect of the inducer, in the range of 0% ≥ ε ≥ 3%, it improves the experimental
specific delivery work slightly by about 1% at part-load conditions, and reduces it by about 5% in overload
conditions due to the strong flow separation that occurs on the inducer blades at such high flow rates [48, 49].
However, the use of the inducer leads to a significant improvement in the pump performance in the ranges
of 4% ≥ ε ≥ 7% and 4% ≥ ε ≥ 5% at part-load and overload conditions, respectively [1]. Further details
about the effect of the inducer on the flow inside the pump are analyzed later by the numerical simulations.
Overall, the comparison of the performance curves confirms generally that the simulation results are accurate
and reliable for part-load and optimal conditions as well as for low gas volume fractions (ε ≤ 3%)

4.2. Pressure contours

Figures 12a, 12b and 12c illustrate the normalized static pressure contours, surface streamlines, and
Q-criterion (black iso-surface) at midspan of the impeller with the standard gap (top rows), the increased
gap (middle rows), and the standard gap with inducer (bottom rows), and for various gas volume fractions
and flow rates. The Q-criterion is used here as an indicator to visualize the vortex and wake regions. A
threshold value of Qcr = 20000 s−2 for the iso-surfaces was found proper to represent the wakes in the
impeller, highlighting also the differences among all cases. The pressure contours are very comparable for
the standard gap with and without inducer since the pump performances are comparable; however, for the
increased gap, a significant drop in the pressure can be seen for all flow conditions.
For part-load conditions as shown in Figure 12a, flow separations occur on the blade suction sides near
the leading edge together with some weaker separations on the blade pressure sides near the trailing edge,
as indicated by separation 1 and separation 2 in the Figure 12a. Separations are more significant and
the streamlines show more recirculation zones for the standard gap as compared to the other two cases.
The streamlines show, however, a smoother behaviour in the impeller for optimal and overload conditions
(Figures 12b and 12c), with minimum vortices at optimal conditions. Concerning the Q-criterion, the iso-
surface becomes more visible within elongated streamwise structures when the gap is increased, especially at
optimal conditions. This is due to the increased interaction between the impeller rotation and the secondary
flow as a result of increasing the gap. On the contrary, the Q-criterion isosurfaces are slightly reduced for
the pump with inducer as the inducer allows a better incidence of the flow on the impeller blades, also
enhancing mixing as later discussed in the Section 4.4.
Nevertheless, concerning overload conditions, significant flow separation occurs at the volute nose for all
the considered cases. The wrong prediction of flow separation is assumed to be a reason for the worse per-
formance prediction at overload conditions. Lastly, it can be seen that as the gas volume fraction increases,
the pressure values are getting lower, while the wake regions are getting moderately larger.

16
 Semi-open impeller Semi-open impeller Semi-open impeller
Standard gap Increased gap Standard gap and inducer

1.4 1.4 1.4

1.2 1.2 1.2
1.0 1.0 1.0

ϒ/ϒopt
ε = 0%
ϒ/ϒopt

ϒ/ϒopt
0.8 0.8 0.8
0.6 0.6 0.6
Simulations Simulations Simulations
0.4 0.4 Experiments
0.4 Experiments
Experiments
0.2 0.2 0.2
0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2
Q/Qopt Q/Qopt Q/Qopt

1.4 1.4 1.4

1.2 1.2 1.2
1.0 1.0 1.0
ε = 1%

ϒ/ϒopt
ϒ/ϒopt

ϒ/ϒopt

0.8 0.8 0.8

0.6 0.6 0.6
Simulations Simulations Simulations
0.4 Experiments 0.4 0.4
Experiments Experiments
0.2 0.2 0.2
0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2
Q/Qopt Q/Qopt Q/Qopt

1.4 1.4 1.4

1.2 1.2 1.2
1.0 1.0 1.0
ε = 3%
ϒ/ϒopt

ϒ/ϒopt

ϒ/ϒopt

0.8 0.8 0.8

0.6 0.6 0.6
Simulations Simulations Simulations
0.4 Experiments 0.4 Experiments 0.4
Experimants
0.2 0.2 0.2
0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2
Q/Qopt Q/Qopt Q/Qopt

1.4 1.4 1.4

Simulations Simulations Simulations
1.2 Experiments 1.2 Experiments 1.2 Experiments
1.0 1.0 1.0
ε = 5%

ϒ/ϒopt
ϒ/ϒopt

ϒ/ϒopt

0.8 0.8 0.8

0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2
Q/Qopt Q/Qopt Q/Qopt

Figure 10: Comparison of experimental and numerical specific delivery work of the pump for various gas volume fractions.

17
 Semi-open impeller Semi-open impeller Semi-open impeller
Standard gap Increased gap Standard gap and inducer

1 1 1
0.8 0.8 0.8
ε = 0%

0.6 0.6 0.6

η
η

η
0.4 0.4 0.4
0.2 Simulations 0.2 Simulations 0.2 Simulations
Experiments Experiments Experiments
0 0 0
0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2
Q/Qopt Q/Qopt Q/Qopt

1 1 1
0.8 0.8 0.8
ε = 1%

0.6 0.6 0.6

η

η
η

0.4 0.4 0.4

Simulations Simulations Simulations
0.2 0.2 0.2
Experiments Experiments Experiments
0 0 0
0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2
Q/Qopt Q/Qopt Q/Qopt

1 1
0.8
0.8 0.8
ε = 3%

0.6
0.6 0.6
η

η
η

0.4 0.4 0.4

0.2 Simulations 0.2 Simulations 0.2 Simulations
Experiments Experiments Experiments
0 0 0
0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2
Q/Qopt Q/Qopt Q/Qopt

1 1 1
0.8 0.8 0.8
ε = 5%

0.6 0.6 0.6

η
η
η

0.4 0.4 0.4

Simulations
0.2 Simulations 0.2 0.2 Simulations
Experiments Experiments
Experiments
0 0 0
0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2 1.6 2
Q/Qopt Q/Qopt Q/Qopt

Figure 11: Comparison of experimental and numerical efficiency of the pump for various gas volume fractions.

18
 Semi-open impeller Semi-open impeller Semi-open impeller
with standard gap and inducer with increased gap with standard gap

Separation 2
Separation 1
= 0% ,,

Separation 2
Separation 1

19
= 1% ,,

(a) Part-load conditions

= 3% ,,

Normalzied static pressure (P/ΔPopt)

 Semi-open impeller Semi-open impeller Semi-open impeller
with standard gap and inducer with increased gap with standard gap = 0% ,,

20
= 1% ,,

(b) Optimal conditions

= 3% ,,

Normalzied static pressure (P/ΔPopt)

 Semi-open impeller
Semi-open impeller with increased gap Semi-open impeller
with standard gap and inducer ( / 1.6 for = 0% and 1%) with standard gap
( / 1.5 for = 3%)
= 0% ,,

21
= 1% ,,

(c) Overload conditions

= 3% ,,

Normalzied static pressure (P/ΔPopt)

Figure 12: Normalized static pressure contours, surface streamlines and Q-criterion within the impeller at different conditions.
4.3. Visualization of two-phase flow patterns

Figures 13a, 13c and 13b show sample instantaneous flow fields obtained by the high-resolution camera
from the experiments in comparison with the corresponding instantaneous flow patterns obtained in sim-
ulations using the VOF model. The experimental and the simulation images have been compared for the
standard gap, the increased gap, and the standard gap with inducer for gas volume fractions ε = 1%, 3%
and 5%. The air phase is represented in the simulation images by volume fraction iso-surfaces, which show
all cells with air volume fraction higher than 5%, while cells involving lower values than this threshold were
excluded from the representation. This value was found suitable to highlight the air phase and reveal the
differences among various cases. To be comparable with the experimental images, the numerical results are
shown for the whole impeller depth. This information is now mentioned in the revised version of the paper..
Considering the fact that the VOF model is an interface-tracking technique, very fine bubbles could only be
resolved using an extremely high mesh resolution, leading to unacceptably long computation times. Never-
theless, the agreement of the predicted gas pockets and flow patterns between experiments and simulations
is already remarkable.
For ε = 1%, the experimental flow regime corresponds to a bubbly flow. As a consequence the simulation
shows correspondingly dispersed and disconnected air iso-surfaces. Additionally, for the flow patterns at
ε = 3% and 5%, simulations predict the right location of gas-phase accumulation. The blade curvature causes
a recirculation on the concave side (suction side), leading there to large gas pockets. The accumulation of
gas pockets on the suction side of the blades observed in the experiments is very well predicted by the VOF
model as well.
Considering the case of an increased gap, the gas phase distribution becomes more uniform due to the
interaction between the secondary flow and the impeller rotation. Similarly, for the pump with inducer, the
swirling effect combined with the inducer wake causes turbulence near the impeller eye; this, together with
the blade leading edge breaking the gas pockets leads to a spatially more uniform distribution. Hence, the
pump with inducer shows improved performance at part-load conditions due to the positive head generated
by the inducer and the enhanced two-phase mixing.
Comparing now the different flow conditions, the simulations show at overload conditions a non-uniform
distribution of the gas-phase within the impeller; some impeller channels have a considerably lower gas
content compared to other channels. This is more obvious at ε = 1% and 3%, indicating reduced mixing.
On the contrary, it can be seen that the air is more dispersed for part-load conditions, due to the increased
residence time at lower flow rates. This also helps to enhance mixing between the two phases, as discussed
further in the next section.

22
 Overload ( / 1.6 for = 1% and 3%)
Optimal ( / 1.0) Part-load ( / 0.5)
wwwwww( / 1.2 for = 5%)
= 1%

23
(a) Standard gap
= 3%
= 5%
 Overload ( / 1.6 for = 1%)
wwwwww( / 1.5 for = 3%) Optimal ( / 1.0) Part-load ( / 0.5)
wwwwww( / 1.2 for = 5%)
= 1%

24
(b) Increased gap
= 3%
= 5%
 = 1% = 3% = 5%

0.5)
/
Part-load (
1.0)
/
Optimal (
1.6 for = 1% and 3%)
1.2 for = 5%)
/
Overload ( /
wwwwww(

(c) Standard gap with inducer

Figure 13: Comparison between experimental images and numerical results for the accumulated gas volume fraction.

25
4.4. Mixing Performance

In order to understand the mixing behaviour in detail, mixing between the two phases is analyzed at the
midspan of the impeller for a value of ε = 3%. This is done by calculating the surface uniformity in terms
of the gas volume fraction to quantify the mixing of the two phases. The surface uniformity Su , defined by
Equation (13), determines how homogeneous the gas is distributed on a specific surface based on the air
volume fraction. In Equation (13), c is the local gas volume fraction, c̄ is the average gas volume fraction
over the surface area A, and Af is the area of an elementary cell face. The average gas volume fraction c̄
can be defined by Equation (14). The maximum value of Su is 1.0 and its minimum value (Su,min ) occurs
at the simulation domain inlet, where air and water are completely separated.

P
f |c − c̄| Af
Su = 1 − P (13)
2 |c̄| f Af

Z
1
c̄ = c dA (14)
A

Now, the mixing coefficient Mc can be defined by normalizing the surface uniformity leading to Equa-
tion (15), with a range of 0 ≤ Mc ≤ 1, where 0 (or 0 %) indicates no mixing, and 1 (or 100 % mixing)
indicates complete mixing, see also [60, 61].

Su − Su,min
Mc = (15)
1 − Su,min

Figure 14 compares the mixing coefficient for part-load, optimal and overload conditions for the pump
with the standard gap, the increased gap and the standard gap with inducer. Here, the average values
of the mixing coefficient over one rotation are shown together with the pseudo error-bars representing the
fluctuation range of the mixing coefficient in each case. As expected, flow mixing decreases monotonically
with the increase in the flow rate (from bottom to top in Fig. 14) due to the reduced residence time.
Furthermore, the mixing of the two phases within the impeller is always better when the pump clearance
gap is increased or the inducer is installed as compared to the standard case. The inducer significantly
improves the mixing at part-load conditions, reaching values higher than 37%, while the improvements are
lower at optimal and overload conditions due to the decreased residence time. Concerning the increased gap,
mixing is significantly improved at overload conditions, justifying the enhanced performance of the pump at
overload conditions even up to ε = 5%. This is a result of stronger interactions between the pump impeller
and the secondary flows, thus enhancing mixing.

26
 Standard gap and inducer
Overl-load
conditions Increased gap
Standard gap
conditions
Optimal
conditions
Part-load

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4

Mixing coefficient, Mc

Figure 14: Mixing coefficient for different pump configurations and flow conditions at ε = 3%.

4.5. Bubble size distribution

In this last section, the bubble size distributions (BSD) in the impeller are experimentally measured and
compared for a low gas volume fraction (bubbly flow regime) to check further the influence of the inducer
and of the increased gap clearance on the flow properties. The bubble sizes were determined by using
shadowgraphy measurements as presented in Figure 3a. The shadowgraphy images were obtained by using
the same high-speed camera as employed for the flow regime evaluation. In this comparison, two values of
gas volume fraction (ε = 0.25% and ε = 0.5%) were considered at overload conditions (Qw /Qopt = 1.5),
since the bubbles are more dispersed in this case, allowing an accurate post-processing. Figure 15 shows
the bubble size distributions obtained for all cases with the median diameter Dmedian , the Sauter mean
diameter D32 , and the standard deviation of the distributions σ given on each plot. Comparisons between
the six cases are shown in Figure 16. As shown, the median diameter and the standard deviation of the BSD
are generally increasing by increasing the gas volume fraction from ε = 0.25% to ε = 0.5%. Installing the
inducer results in slightly lower median diameters and more homogeneous distributions, i.e. lower standard
deviations. The reason is that the rotary action of the inducer boosts the phase mixing and makes the BSD
more homogeneous. Similarly, the increased gap clearance leads to lower median diameters compared to the
standard gap case. These results support the previous observations concerning the influence of both inducer
and clearance gap.

27
 100 100
90 Dmedian = 0.443 mm 90
Dmedian = 0.492 mm

Percentage of maximum
Percentage of maximum

80 80
70 D32 = 1.553 mm 70 D32 = 2.223 mm
60 σ = 0.491 mm 60 σ = 0.649 mm
50 50
40 Bubble size distribution 40
Bubble size distribution
30 30
20 Cumulative summation 20 Cumulative summation
10 10
0 0
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
2.2
2.4
2.6
2.8
3.0
3.2
3.4
3.6
3.8
4.0

0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
2.2
2.4
2.6
2.8
3.0
3.2
3.4
3.6
3.8
4.0
Bubble diameter (mm) Bubble diameter (mm)
(a) Standard gap - ε = 0.25% (b) Standard gap - ε = 0.50%
100 100
90 90
Dmedian = 0.338 mm

Percentage of maximum
Percentage of maximum

80 Dmedian = 0.361 mm
80
70 D32 = 1.451 mm 70 D32 = 1.734 mm
60 σ = 0.418 mm 60 σ = 0.489 mm
50 50
40 Bubble size distribution 40 Bubble size distribution
30 30
20 Cumulative summation 20 Cumulative summation
10 10
0 0
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
2.2
2.4
2.6
2.8
3.0
3.2
3.4
3.6
3.8
4.0
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
2.2
2.4
2.6
2.8
3.0
3.2
3.4
3.6
3.8
4.0

Bubble diameter (mm) Bubble diameter (mm)

(c) Increased gap - ε = 0.25% (d) Increased gap - ε = 0.50%
100 100
90 90
Dmedian = 0.349 mm Dmedian = 0.486 mm
Percentage of maximum

Percentage of maximum

80 80
70 D32 = 0.920 mm 70 D32 = 1.583 mm
60 σ = 0.283 mm 60 σ = 0.452 mm
50 50
40 40
Bubble size distribution Bubble size distribution
30 30
20 Cumulative summation 20 Cumulative summation
10 10
0 0
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
2.2
2.4
2.6
2.8
3.0
3.2
3.4
3.6
3.8
4.0
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
2.2
2.4
2.6
2.8
3.0
3.2
3.4
3.6
3.8
4.0

Bubble diameter (mm) Bubble diameter (mm)

(e) Standard gap with inducer - ε = 0.25% (f) Standard gap with inducer - ε = 0.50%

Figure 15: Bubble size distributions.

0.6 1.0
Median bubble diamter (mm)

BSD standard deviation (mm)

Standard gap Standard gap

Increased gap Increased gap
Standard gap with inducer 0.8 Standard gap with inducer
0.5
0.6
0.4
0.4

0.3 0.2
0 0.25 0.5 0.75 0 0.25 0.5 0.75
Gas volume fraction, ε % Gas volume fraction, ε %
(a) Median diameter of the BSD (b) Standard deviation of the BSD

Figure 16: Comparison of bubble size distributions.

28
5. Conclusions

Experimental and numerical investigations were performed for gas-liquid two-phase flows in a centrifugal
pump. The results were compared for three configurations, i.e. 1) semi-open impeller with a standard gap,
2) semi-open impeller with an increased gap, 3) semi-open impeller with a standard gap and inducer. In
the simulations, a moving-mesh approach was used to model the rotation together with k-ω SST turbulence
model and the VOF method for modeling a turbulent two-phase flow. The discretization in space and time
was selected after performing a corresponding dependence study. The simulations show very good agree-
ment with the experiments for gas volume fractions from ε = 0% to 3%. Higher deviations occur at overload
conditions due to the inaccurate prediction of the flow separation at the volute nose. For ε = 5%, the simu-
lations correctly predict the location of gas phase accumulation in the impeller; however, it fails to predict
the sudden performance drop at overload conditions. Accordingly, the numerical results are comparable to
the experiments only at part-load and near optimal conditions. Generally, the flow regimes obtained by the
simulations and the experiments are in good agreement, and the gas accumulations are mostly predicted on
the suction side of the blades. However, fine bubbles and minute two-phase interactions cannot be resolved
by the VOF simulations as it would require an extremely fine mesh, leading to unacceptable computing
times. The homogeneity of both phases is quantified along the midspan section of the impeller by com-
puting the mixing coefficient, revealing the improved mixing when the gap is increased or the inducer is
installed. Finally, the bubble size distributions have been measured by high-speed shadowgraphy measure-
ments for low gas volume fractions (ε = 0.25% and ε = 0.5%). All results confirm that installing the inducer
or increasing the gap result in more homogeneous bubble size distributions with lower median diameters.
Overall, the use of inducer is more effective in part-load and optimal conditions, while the increased gap is
preferred at overload.

Acknowledgments

The authors gratefully acknowledge the financial support of this project by the Verband Deutscher
Maschinen- und Anlagenbau e.V. (VDMA), as part of a collaboration with Professor R. Skoda from Ruhr
University in Bochum. The Ph.D. work of Mr. Mansour was partially supported by a scholarship from the
Egyptian government.

References

[1] M. Mansour, B. Wunderlich, D. Thévenin, Effect of tip clearance gap and inducer on the transport of
two-phase air-water flows by centrifugal pumps, Exp. Therm. Fluid Sci 99 (2018) 487–509.

[2] J. Zhang, S. Cai, Y. Li, H. Zhu, Y. Zhang, Visualization study of gas-liquid two-phase flow patterns
inside a three-stage rotodynamic multiphase pump, Exp. Therm. Fluid Sci. 70 (2016) 125–138.

29
 [3] J. Caridad, M. Asuaje, F. Kenyery, A. Tremante, O. Aguillón, Characterization of a centrifugal pump
impeller under two-phase flow conditions, J. Pet. Sci. Eng. 63 (1-4) (2008) 18–22.

[4] Z. Zhu, P. Xie, G. Ou, B. Cui, Y. Li, Design and experimental analyses of small-flow high-head
centrifugal-vortex pump for gas-liquid two-phase mixture, Chin. J. Chem. Eng. 16 (4) (2008) 528–534.

[5] W. M. Verde, J. L. Biazussi, N. A. Sassim, A. C. Bannwart, Experimental study of gas-liquid two-phase

flow patterns within centrifugal pumps impellers, Exp. Therm. Fluid Sci. 85 (2017) 37–51.

[6] J. Zhu, X. Guo, F. Liang, H. Q. Zhang, Experimental study and mechanistic modeling of pressure
surging in electrical submersible pump, J. Nat. Gas Sci. Eng. 45 (2017) 625–636.

[7] T. Schäfer, M. Neumann, A. Bieberle, U. Hampel, Experimental investigations on a common centrifugal

pump operating under gas entrainment conditions, Nucl. Eng. Des. 316 (2017) 1–8.

[8] A. M. C. Chan, M. Kawaji, H. Nakamura, Y. Kukita, Experimental study of two-phase pump perfor-
mance using a full size nuclear reactor pump, Nucl. Eng. Des. 193 (1-2) (1999) 159–172.

[9] A. Poullikkas, Effects of two-phase liquid-gas flow on the performance of nuclear reactor cooling pumps,
Prog. Nucl. Energy 42 (1) (2003) 3–10.

[10] A. Amoresano, G. Langella, V. Niola, G. Quaremba, Advanced image analysis of two-phase flow inside
a centrifugal pump, Adv. Mech. Eng. 6 (2014) 1–11.

[11] Q. Jiang, Y. Heng, X. Liu, W. Zhang, G. Bois, Q. Si, A review of design considerations of centrifugal
pump capability for handling inlet gas-liquid two-phase flows, Energies 12 (6) (2019) 1078.

[12] Q. Si, G. Bois, Q. Jiang, W. He, A. Ali, S. Yuan, Investigation on the handling ability of centrifugal
pumps under air–water two–phase inflow: model and experimental validation, Energies 11 (11) (2018)
3048.

[13] M. Neumann, T. Schäfer, A. Bieberle, U. Hampel, An experimental study on the gas entrainment in
horizontally and vertically installed centrifugal pumps, J. Fluids Eng. 138 (9) (2016) 91301.

[14] C. A. Cappellino, D. R. Roll, G. Wilson, Design considerations and application guidelines for pumping
liquids with entrained gas using open impeller centrifugal pumps, in: Proceedings of the 9th Interna-
tional Pump Users Symposium, Turbomachinery Laboratories, Department of Mechanical Engineering,
Texas A&M University, 1992, pp. 262–264.

[15] S. V. Jain, A. Swarnkar, K. H. Motwani, R. N. Patel, Effects of impeller diameter and rotational speed
on performance of pump running in turbine mode, Energy Convers. Manage. 89 (2015) 808–824.

30
[16] M. Mansour, B. Wunderlich, D. Thévenin, Experimental study of two-phase air/water flow in a cen-
trifugal pump working with a closed or a semi-open impeller, in: ASME Turbo Expo 2018: Turbo-
machinery Technical Conference and Exposition, American Society of Mechanical Engineers, 2018, p.
V009T27A012.

[17] S. Sato, A. Furukawa, Y. Takamatsu, Air-water two-phase flow performance of centrifugal pump im-
pellers with various blade angles, JSME Int. J. Ser. B Fluids Therm. Eng. 39 (2) (1996) 223–229.

[18] J. Caridad, F. Kenyery, CFD analysis of electric submersible pumps (ESP) handling two-phase mixtures,
J. Energy Resour. Technol. 126 (2) (2004) 99–104.

[19] J. F. Gülich, Centrifugal pumps, Springer, 2008.

[20] J. Zhang, H. Zhu, C. Yang, Y. Li, H. Wei, Multi-objective shape optimization of helico-axial multiphase
pump impeller based on nsga-ii and ann, Energy Convers. Manage. 52 (1) (2011) 538–546.

[21] C. Shao, C. Li, J. Zhou, Experimental investigation of flow patterns and external performance of a
centrifugal pump that transports gas-liquid two-phase mixtures, Int. J. Heat Fluid Flow 71 (2018)
460–469.

[22] A. Serena, A multiphase pump experimental analysis, Ph.D. thesis, Norwegian University of Science
and Technology (2016).

[23] L. Barrios, M. G. Prado, Modeling two-phase flow inside an electrical submersible pump stage, J.
Energy Resour. Technol. 133 (4) (2011) 042902.

[24] G. M. Paternost, A. C. Bannwart, V. Estevam, Experimental study of a centrifugal pump handling

viscous fluid and two-phase flow, SPE Prod. Oper. 30 (02) (2015) 146–155.

[25] J. Zhu, H. Banjar, Z. Xia, H. Q. Zhang, CFD simulation and experimental study of oil viscosity effect
on multi-stage electrical submersible pump (ESP) performance, J. Pet. Sci. Eng. 146 (2016) 735–745.

[26] J. L. Biazussi, W. M. Verde, M. P. Varón, A. C. Bannwart, V. Estevam, Comparison of experimental

curves of ESP with a simple single-phase approach, in: 22nd International Congress of Mechanical
Engineering (COBEM 2013). Ribeirão Preto, 2013.

[27] M. Sauer, Einfluss der Zuströmung auf das Förderverhalten von Kreiselpumpen radialer Bauart bei
Flüssigkeits-/Gasförderung, Ph.D. thesis, Technische Universität Kaiserslautern (2003).

[28] L. Barrios, M. G. Prado, Experimental visualization of two-phase flow inside an electrical submersible
pump stage, J. Energy Resour. Technol. 133 (4) (2011) 042901.

31
[29] Q. Si, G. Bois, K. Zhang, S. Yuan, Air-water two-phase flow experimental and numerical analysis in a
centrifugal pump, in: Proceedings of 12th European Conference on Turbomachinery Fluid Dynamics
and Thermodynamics, ETC, Vol. 12, 2017.

[30] J. Zhu, H. Zhu, J. Zhang, H. Q. Zhang, An experimental study of surfactant effect on gas tolerance
in electrical submersible pump (ESP), in: ASME 2017 International Mechnical Engineering Congress
Exposition, American Society of Mechanical Engineers, 2017, pp. V007T09A038–V007T09A038.

[31] M. Mansour, P. Kováts, B. Wunderlich, D. Thévenin, Experimental investigations of a two-phase

gas/liquid flow in a diverging horizontal channel, Exp. Therm. Fluid Sci 93 (2018) 210–217.

[32] M. Murakami, K. Minemura, Effects of entrained air on the performance of centrifugal pumps: 2nd
report, effects of number of blades, Bull. JSME 17 (112) (1974) 1286–1295.

[33] A. Furukawa, S. Shirasu, S. Sato, Experiments on air-water two-phase flow pump impeller with rotating-
stationary circular cascades and recirculating flow holes, JSME Int. J. Ser. B Fluids Therm. Eng. 39 (3)
(1996) 575–582.

[34] G. Mustafa, P. Yasar, S. Yakup, Energy saving in a deep well pump with splitter blade, Energy Convers.
Manage. 47 (5) (2006) 638–651.

[35] S. Pirouzpanah, S. R. Gudigopuram, G. L. Morrison, Two-phase flow characterization in a split vane

impeller electrical submersible pump, J. Pet. Sci. Eng. 148 (2017) 82–93.

[36] G. L. Morrison, S. Pirouspanah, K. Kirland, S. L. Scott, L. J. Barrios, Performance evaluation of

a multiphase electric submersible pump, in: Offshore Technology Conference, Offshore Technology
Conference, 2014.

[37] M. Murakami, K. Minemura, Effects of running clearance of semi-open impeller pumps under air ad-
mitting conditions, Bull. JSME 19 (136) (1976) 1141–1148.

[38] M. Ishii, M. A. Grolmes, Inception criteria for droplet entrainment in two-phase concurrent film flow,
AIChE J. 21 (2) (1975) 308–318.

[39] M. Ishii, K. Mishima, Two-fluid model and hydrodynamic constitutive relations, Nucl. Eng. Des. 82 (2-
3) (1984) 107–126.

[40] C. W. Hirt, B. D. Nichols, Volume of fluid (VOF) method for the dynamics of free boundaries, J.
Comput. phys. 39 (1) (1981) 201–225.

32
[41] H. Pineda, J. Biazussi, F. López, B. Oliveira, R. D. M. Carvalho, A. C. Bannwart, N. Ratkovich, Phase
distribution analysis in an electrical submersible pump (ESP) inlet handling water–air two-phase flow
using computational fluid dynamics (CFD), J. Petrol. Sci Eng. 139 (2016) 49–61.

[42] J. Zhu, H. Zhu, J. Zhang, H. Q. Zhang, A numerical study on flow patterns inside an electrical sub-
mersible pump (ESP) and comparison with visualization experiments, J. Petrol. Sci. Eng. 173 (2019)
339–350.

[43] S. Huang, X. Su, J. Guo, L. Yue, Unsteady numerical simulation for gas–liquid two-phase flow in
self-priming process of centrifugal pump, Energy Convers. Manage. 85 (2014) 694–700.

[44] S. Yamoah, R. Martı́nez-Cuenca, G. Monrós, S. Chiva, R. Macián-Juan, Numerical investigation of

models for drag, lift, wall lubrication and turbulent dispersion forces for the simulation of gas–liquid
two-phase flow, Chem. Eng. Res. Des. 98 (2015) 17–35.

[45] Q. Huang, C. Yang, G. Yu, Z. S. Mao, CFD simulation of hydrodynamics and mass transfer in an
internal airlift loop reactor using a steady two-fluid model, Chem. Eng. Sci. 65 (20) (2010) 5527–5536.

[46] M. Hundshagen, M. Mansour, D. Thévenin, R. Skoda, Numerical investigation of two-phase air-water

flow in a centrifugal pump with closed or semi-open impeller, in: Proceedings of 13th European Tur-
bomachinery Conference on Turbomachinery Fluid Dynamics and Thermodynamics, ETC, 2019.

[47] Q. Wang, W. Yao, Computation and validation of the interphase force models for bubbly flow, Int. J.
Heat Fluid Flow 98 (2016) 799–813.

[48] M. Mansour, T. Parikh, S. Engel, B. Wunderlich, D. Thévenin, Investigation on the influence of an in-
ducer on the transport of single and two-phase air-flows by centrifugal pumps, in: 48th Turbomachinery
& 35th Pump Symposia, 2019.

[49] M. Mansour, T. Parikh, S. Engel, B. Wunderlich, D. Thévenin, Numerical investigations of gas-liquid

two-phase flow in a pump inducer, J. Fluids Eng. 142 (2) (2020) 021302–1–021302–12.

[50] Siemens PLM Software Inc., STAR-CCM+ Version 13.02.013 User Guide, Plano, TX, USA (2018).

[51] F. R. Menter, Two-equation eddy-viscosity turbulence models for engineering applications, AIAA J.
32 (8) (1994) 1598–1605.

[52] W. Zhang, Z. Yu, B. Zhu, Numerical study of pressure fluctuation in a gas-liquid two-phase mixed-flow
pump, Energies 10 (5) (2017) 634.

[53] X. Li, S. Yuan, Z. Pan, Y. Li, W. Liu, Dynamic characteristics of rotating stall in mixed flow pump, J.
Appl. Math. 2013 (2013) 12.

33
[54] T. Müller, P. Limbach, R. Skoda, Numerical 3D RANS simulation of gas-liquid flow in a centrifugal
pump with an Euler-Euler two-phase model and a dispersed phase distribution, in: Proceedings of 11th
European Conference on Turbomachinery Fluid Dynamics and Thermodynamics, ETC, Madrid, 2015.

[55] T. Müller, P. Limbach, R. Skoda, Influence of geometry simplifications and numerical parameters in
3D URANS liquid-gas flow simulations of a radial pump with an Eulerian mono-dispersed two-phase
model, in: Proceedings of International Symposium on Transport Phenomena and Dynamics of Rotating
Machinery, 2016.

[56] T. Frank, P. J. Zwart, E. Krepper, H. M. Prasser, D. Lucas, Validation of CFD models for mono-and
polydisperse air–water two-phase flows in pipes, Nucl. Eng. Des. 238 (3) (2008) 647–659.

[57] B. Jafarzadeh, A. Hajari, M. Alishahi, M. Akbari, The flow simulation of a low-specific-speed high-speed
centrifugal pump, Appl. Math. Modell. 35 (1) (2011) 242–249.

[58] M. Mansour, A. Landage, P. Khot, K. D. Nigam, G. Janiga, D. Thévenin, K. Zähringer, Numerical

study of gas-liquid two-phase flow regimes for upward flow in a helical pipe, Ind. Eng. Chem. Res. In
Press, (2020).

[59] P. Stephan, S. Kabelac, M. Kind, H. Martin, D. Mewes, K. Schaber, VDI heat atlas (2010).

[60] M. Mansour, Z. Liu, G. Janiga, K. D. P. Nigam, K. Sundmacher, D. Thévenin, K. Zähringer, Numerical

study of liquid–liquid mixing in helical pipes, Chem. Eng. Sci. 172 (2017) 250–261.

[61] M. Mansour, P. Khot, D. Thévenin, K. D. P. Nigam, K. Zähringer, Optimal Reynolds number for
liquid–liquid mixing in helical pipes, Chem. Eng. Sci. In Press, (2018).

34