See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/316292628

Experimental and CFD investigation of a multistage centrifugal pump

including leakages and balance holes

Article in Desalination and Water Treatment · January 2017

DOI: 10.5004/dwt.2017.20153

CITATIONS READS

32 5,548

4 authors:

Osman Babayigit Muammer Ozgoren

Karamanoglu Mehmetbey Üniversitesi Selcuk University
13 PUBLICATIONS 145 CITATIONS 89 PUBLICATIONS 1,341 CITATIONS

SEE PROFILE SEE PROFILE

Muharrem Hilmi Aksoy Osman Kocaaslan

Konya Technical University Selcuk University
56 PUBLICATIONS 438 CITATIONS 10 PUBLICATIONS 79 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Muharrem Hilmi Aksoy on 09 February 2018.

The user has requested enhancement of the downloaded file.

 Desalination and Water Treatment 67 (2017) 28–40
March
www.deswater.com
doi: 10.5004/dwt.2017.20153

Experimental and CFD investigation of a multistage centrifugal

pump including leakages and balance holes

Osman Babayigita, Muammer Ozgorenb,*, Muharrem Hilmi Aksoyb, Osman Kocaaslanc

a
Selcuk University Hadim Vocational School, Konya, Turkey, email: [email protected]
b
Selcuk University, Mechanical Engineering Department, Konya, Turkey, emails: [email protected] (M. Ozgoren),
[email protected] (M.H. Aksoy)
c
Selcuk University Huglu Vocational School Konya, Turkey, email: [email protected]

Received 1 March 2016; Accepted 13 August 2016

abstract
The aim of this study is to analyze numerically a multistage pump with some new approaches and
compare the results with experimental data. A centrifugal pump, consisting of six backward curved
blades, ten vane diffusers and two stages, was used. First, models of an impeller, a diffuser, suction
and discharge sections of the centrifugal pump were separately designed by empirical equations. The
flow volume of the pump was numerically solved by different turbulence models. Thus, the most
accurate results in comparison with experimental data were obtained for the realizable k-ε turbulence
model. For the purposes of this study, the effect of balance holes and leakages (clearances) on the
performance was scrutinized. In most studies, it is not taken into consideration in computational fluid
dynamics (CFD) analyses due to the difficulty of meshing these regions. The experimental studies
were performed at a computer controlled pump test rig that was established according to the standard
TS EN ISO 9906. The results of CFD/experimental analyses for the pump head, hydraulic efficiency
and hydraulic power on the design point, having the flow rate of 80 m3/h and the revolution of
2,975 rpm, were found to be 81.47/80.70 m, 51.98%/51.42% and 17.94/17.77 kW, respectively. The
obtained CFD and experimental results were found to be close agreement for the design flow rate
as well as for all tested flow rates. Other characteristics of the pump such as velocity, static pressure,
streamline patterns and turbulence kinetic energy were also investigated. In addition, the leakages and
balance holes have significantly affected the pump characteristics, which must be considered in CFD
analyses in order to find more precise results for true-to-life simulations.
Keywords: Balance hole; CFD; Clearance; Hydraulic efficiency; Leakages; Multistage centrifugal pump

1. Introduction The number of blades is one of the design parameters

and is determined depending on the other parameters. There
Centrifugal pumps are widely used for the purpose of
are many pumps impeller, which have different number of
water supply, industrial applications, waste pump, water
blades in the industry. Especially on the radial flow pumps,
treatment, irrigation systems and much more [1,2]. In order to
most of the impellers have six blades. In addition, radial flow
decrease the present 20% electricity consumption ratio of the
pumps are one of the commonly used types in the industry.
pumps in the world, even a small improvement of efficiency
So that six-blade numbers were used in this study owing to
of pumps enable to save a significant amount of energy [3,4].
the providing satisfactorily necessary head and volume flow
Therefore, detailed analyses should be carried out on pumps
rate.
including all physical, geometrical and material variables.
In recent years, with the rapid development of the com-
puter technology and computational fluid dynamics (CFD),
* Corresponding author. a numerical simulation has become an important tool to
study detailed flow field in pumps and also to predict pump

1944-3994/1944-3986 © 2017 Desalination Publications. All rights reserved.

 O. Babayigit et al. / Desalination and Water Treatment 67 (2017) 28–40 29

performance. CFD provides a cost-effective and accurate alter-

native to scale model testing with variations on the simulation
being performed quickly [5–7]. Development in CFD makes
available to analyze detailed flow structures in pumps [8].
A centrifugal pump consists of a set of rotation vanes
enclosed within housing or casing that is used to impart
energy to a fluid through centrifugal force [9]. There are
several parameters that affect the pump efficiency and life-
time. The life and size of the bearings in centrifugal pumps
are greatly influenced by the axial thrust generated by fluid
flow. High axial thrust loads can cause rapid thrust bearing
wear and either subsequent pump failure or frequent over-
hauls [10]. This axial force increases with the increasing stage
number in multistage pumps. The impeller must be axially
balanced if it is not possible to absorb the entire axial thrust
in the axial bearing. There are several methods to reduce the
thrust on the shaft but they result in hydraulic losses. One of
the ways to balance this axial thrust is using balance holes.
Balance holes in the impeller of the centrifugal pumps are
mainly drilled into the hub surface. The balance holes con-
nect the back of the impeller with the inlet area, where pres-
sure is low. The leak flow through the holes influences the
1- suction 2- impeller 3- diffuser
flow in the cavities above the impeller and thereby reduces
the axial force but it results in leakage and reduces the pump Fig. 1. Clearances and balance hole on a partial cross section
efficiency [11]. view at one stage of the pump.
There are also some other losses on the pump that
should be considered. Some liquid is passed through the
gap between rotor and stator in order to remove the heat cre- in the centrifugal pump. Numerical simulation based on the
ated by the motor losses and fluid friction. To let the impel- Reynolds-averaged Navier–Stokes (RANS) combined with
ler rotate freely within the pump casing, small clearances the Shear Stress Transport (SST) k-ω turbulence model was
should be designed between the impeller and pump casing. used to obtain hydrodynamic performance of the centrifugal
Therefore, a leak flow will occur in the gap between the rotat- pump. The numerical results of the hydrodynamic perfor-
ing impeller and stationary pump housing. The rate of leak mance have coincided with the experimental data. It was also
flow depends mainly on the design of the gap, the impeller proven that above or below the design flow rate the reason
pressure rise and impeller rotating speed. Clearances and bal- of the errors between experimental and numerical results in
ance hole causing leak flows are shown in Fig. 1. To minimize pump may occur due to neglecting of leakage loss through
the amount of leak fluid through these clearances to increase balancing holes. It was pointed out that the impact of balance
the efficiency of the centrifugal pumps, some wearing ring or holes and leakages cannot be neglected. Stel et al. [21] carried
impeller seal can be used [12]. However, these precautions do out a numerical study on the flow in a multistage, mixed-
not prevent these leakages completely. type electric submersible pump. They used a CFD program
Especially in the regions of leakages, it has been extremely based on the finite volume method to find the numerical
challenging to create grids and analyses for simulation due to solution of the flow inside a multistage pump. It was con-
the very small size of the flow volume. Therefore, in many cluded that the results provided by the numerical model are
CFD studies, the effect of leakages and balance holes in a accurate enough and acceptable for the purposes of the study.
multistage centrifugal pump has not been investigated. This They inferred that some differences in numerical and exper-
approach causes a deviation between the experimental and imental study due to the leakage flows and gaps were not
numerical results. Complicated leak flows, clearances and considered. The study about the impact of centrifugal pump
any balance holes make the hydrodynamic analyses very side clearance on pump performance was carried out numer-
challenging without performing three-dimensional CFD ically by Ayad et al. [22]. The influence of the variation of the
analyses. centrifugal pump side clearance width (0:0.2 impeller width)
There are many studies related to the pump performance on the centrifugal pump performance parameters at differ-
improvement in the literature. Some of them focuses on ent flow rates (0:5 L/s) was investigated. It was proven that
the detailed investigation of geometrical modifications and the impeller side clearance had a great regression effect on
elaborates experimental tests. Three-dimensional turbulence the centrifugal pump performance. Engeda and Rautenburg
flow simulation for centrifugal impellers was performed [23] performed a s­ystematic study on semi-open impellers
numerically in many studies [13–19]. In most of these studies, and yielded a ­conclusion that the tip clearance had an evi-
balance holes and leak flows are not considered. dent effect on pump performance. Liu et al. [24] conducted
There are few studies that express the importance of the experimental and numerical investigations for tip clearance
clearances and effect of balance holes. Jiang et al. [20] car- effect on impeller-geometry of mini turbo-pump. In the
ried out a numerical investigation of influence of the clocking performance tests, rotational speed varied between 3,000
effect on the unsteady pressure fluctuation and radial forces and 10,000 rpm, and the axial clearance at the blade tip of
30 O. Babayigit et al. / Desalination and Water Treatment 67 (2017) 28–40

impeller exit was between 0.2 and 0.8 mm. They came to the Table 1
conclusion that the clearance has affected the pump perfor- Specifications of the multistage pump
mance significantly and its effect could be attenuated by the
impeller geometry such as larger outlet blade angle. Description Parameter Value
Although some studies are done on the analyses of cen- Design flow rate (m³/h) Qd 80
trifugal pumps as aforementioned above, a study on a two- Single-stage head (m) hs 40
stage centrifugal pump performance including the effect of Rotating speed (rpm) n 2,975
leakage flow losses and balance holes has not been encoun- Stage number N 2
tered. The motivation of this paper is to perform both numer-
Specific speed nq 28
ical and experimental studies on a multistage centrifugal
pump in order to provide the lacking data. Moreover, the
effect on the flow structure of the leakage losses and the bal- Table 2
ance holes was examined in detail. Geometric parameters of the impeller and diffuser

Description Parameter Value

2. Material and methods
Impeller
In this study, a volume flow rate, head and revolution
Inlet diameter (mm) D1 100
values of a two-stage centrifugal pump with a vertical shaft
Inlet blade angle (°) β1 27
were selected. The impeller, a diffuser, suction and discharge
sections of the centrifugal pump were separately designed Inlet blade width (mm) b1 23
using the empirical equations and were drawn using ANSYS- Outlet diameter (mm) D2 205
BladeGen and Solidworks software [12]. Then, the numerical Outlet blade angle (°) β2 20
analyses of the designed pump were made in ANSYS-Fluent Outlet blade width (mm) b2 14
software to determine the optimum geometry of impeller Blade wrap angle (°) ϕ 125
and diffuser to be manufactured. Finally, performance test Blade number z 6
of the manufactured pump was performed to compare the
Diffuser
numerical results with experimental ones.
Inlet diameter (mm) D3 206
Inlet blade angle (°) β3 9
2.1. Numerical study
Vane number zv 10
ANSYS-Fluent program for the numerical analyses of the Outlet diameter (mm) D4 274
selected pump operations were performed in the following Return vane number zr 8
order:
Vane wrap angle (°) ϕv 50
• designing and drawing of the pump impeller, the dif- Return vane wrap angle (°) ϕr 25
fuser, the suction and the discharge sections;
• creating the flow volume and mesh structure;
• performing the CFD analyses under proper boundary
conditions; and the number of grids/grid size on the computational results as
• obtaining the pump flow characteristics as head, hydrau- shown in Fig. 5. As the mesh size increases up to 38 million,
lic power and hydraulic efficiency. the hydraulic efficiency and pump head reach to asymptotic
value at 25 × 106 and became independent of the mesh size
The main design parameters and geometric specification and a mesh number. Therefore, this mesh size was applied to
of the examined pump are given in Tables 1 and 2. The cross all analyses to save the computation time. A relativity exam-
sections of the impeller and the diffuser are presented in ination of the grid number was done to obtain grid indepen-
Figs. 2 and 3. dency. The effect of grid number on pump characteristics less
than 1% is acceptable [25].
2.1.1. The solid model and flow volume of the pump The occurrence of leak flow regions has 0.3 mm gap width.
It is observed a decrease in mesh quality when mesh transi-
First, models of the diffuser, the suction and discharge sec- tion from large surfaces to very small leakage surfaces has
tions of the pump are separately designed using Solidworks been realized. Therefore, in this study before forming mesh
software as shown in Fig. 4(a). After designing of the pump structure, flow volumes of leakages are separately defined
sections, a solid model of the pump was completed, and flow and meshed as shown in Fig. 6. In addition, eight layers of
volume at inlet, within the pump, and outlet were formed by the inflation were used for the impeller volume by creating
closing the suction and discharge region of the pump as it can thin elements that can correctly capture the velocity and
be seen in Fig. 4(b). pressure gradients near the no-slip walls. The flow domain
of the diffuser, the impeller, suction and the discharge section
of the pump volume were meshed with tetrahedral cells, and
2.1.2. Creating mesh structure
small leakage flow regions of the pump volume were meshed
CFD studies are performed in seven different grid sizes with hexagonal cells as it is shown in Fig. 6. Hexagonal mesh
for the grid independency test to diminish the influence of increases the mesh quality especially in small volumes and
 O. Babayigit et al. / Desalination and Water Treatment 67 (2017) 28–40 31

55 84
Efficiency head

54 83

82
53

ηh (%)
81

h (m)
52
Fig. 2. Cross section view and solid model of the impeller. 80
51 79

50 78
0 5 10 15 20 25 30 35 40
Grid number (106)

Fig. 5. Performance characteristics of the pump for different grid

numbers.

Fig. 3. Cross section view and solid models of the diffuser.

(a)

Fig. 6. Mesh structures of the fluid domain of the pump.

decreases the number of grids but it is not possible to apply

it to complex geometries such as an impeller and a diffuser.
Average and minimum orthogonal values of the mesh qual-
ity were found to be 0.85 and 0.15, respectively.

2.1.3. Numerical analysis

To represent the flow phenomena of the pump, some
of the common methods for the steady analyses in the lit-
erature SST k-ω, standard k-ε and realizable k-ε turbulence
(b) models were examined; thus, realizable k-ε turbulence model
1- suction section 2- first impeller 3- first diffuser 4- second impeller was found to be appropriate for CFD analysis of the present
5- second diffuser 6- discharge section 7- leakage losses 8- balance hole
centrifugal pump. The boundary conditions were taken for
the real operation conditions as given in Table 3. The real-
Fig. 4. Pump appearance: (a) demounted solid model and izable k-ε turbulence model and SIMPLEC algorithm were
(b) flow volume of the pump. applied to solve the equations. The simulation was assumed
32 O. Babayigit et al. / Desalination and Water Treatment 67 (2017) 28–40

Table 3
Boundary conditions for the present study S = 2S ij S ij (7)

Position Boundary condition In the given equations, Gkε represents the generation of
Inlet Mass flow inlet
turbulence kinetic energy due to the mean velocity gradi-
ents; Gb is the generation of turbulence kinetic energy due to
Outlet Pressure outlet
buoyancy; YM represents the contribution of the fluctuating
Walls at hub and shroud Non-slip wall dilatation incompressible turbulence to the overall dissipa-
Impeller Rotating reference frame tion rate; C2 and C1ε are constants and σk and σε are the tur-
Interface General connection bulent Prandtl numbers for k and ε, respectively. Sk and Sε
Wall function Standard wall function are user-defined source terms [30]. The standard values of
different constants appearing in equations (C1ε = 1.44, C2 = 1.9,
σk = 1.0, σε = 1.2) were used in the present computations [31].
The solution of Eqs. (3) and (4) gives spatial variation of k
to be steady flow, and moving reference frame was applied and ε, which, in turn, can be used to find out spatial variation
to the impeller-diffuser interaction surfaces. Mass flow rate of turbulent viscosity or eddy viscosity µt using the Prandtl–
was specified in the pump inlet, and static pressure was used Kolmogorov relation [28,32]:
at the pump outlet. Average height of surface roughness
was measured to be approximately 0.035 mm for all non-­ k2
machined surfaces as the property of the sand casting pro- µt = ρc µ (8)
ε
duction with the sanding process and structure properties of
the cast iron material. Non-slip wall conditions were applied After performed the numerical analysis, hydraulic effi-
on all the physical surfaces except the interfaces between dif- ciency of the pump was calculated using pump head in m
ferent parts of the flow surfaces of the multistage blades and unit (h), volume flow rate in m3/s (Q ), angular velocity in
diffuser connections. Residuals of the numerical calculation rad/s (w), revolution in rpm (ṅ) and torque value in Joule (T)
were set to 10–5 for the static pressure value at the pump out- as follows [33]:
let. When the overall imbalance of the calculated parameters
was less than 0.1% or the minimum residual was reached, the 
ρgQh
simulation was assumed to be converged for final computa- Ph = (9)
1000
tion. If the value of y+ is found to be between 30 and 100, it
shows the results of the grid structure can be acceptable [26]. Tω
In this study, the overall value of y+ was found to be between Ps = (10)
1000
30 and 80.
The ANSYS FLUENT software was used to solve includ- π n
ing the centrifugal force source in the impeller and the steady ω= (11)
30
terms. Turbulent flow in the pump was simulated using the
realizable k-ε model. For three-dimensional incompressible, Ph
steady flow, the continuity and momentum equations are ηh = (12)
written as follows [27–29]: Ps


∇V = 0 (1)
2.2. Experimental setup

 
 
  
ρ∇(V V ) = − ∇P + µ∇ 2 V + ρg + F (2) The elements of the set up are as follows:

Transport equations for the realizable k-ε model are as • a differential pressure transmitter, which was located
follows: between the pump inlet and outlet,
• a magnetic flow meter, which was used for measuring the
flow rate by a throttle valve to regulate it manually, and
∂ ∂  µ  ∂k 
• an electrical control panel to measure values of voltage
(ρKu j ) =  µ + t   + G k + G b − ρε −YM + S k (3)
∂x i ∂x i  σk  ∂x j  and current.

∂ ∂  µ  ∂ε  ε2 ε (4) Experimental setup is schematically depicted in Fig. 7.

(ρεuj ) =  µ + t   + ρC1Sε − ρC 2 + C1ε C3Gb + Sε The experiment was repeated for the different flow rates by
∂xi ∂xi  σε  ∂x j  k + vε k
adjusting the throttle valve at the revolution of 2,975 rpm.
where: The values measured by the volume flow meter, differen-
tial pressure transmitter and electrical control panel were
 η  recorded and processed.
C1 = max 0.43  (5) Hydraulic power and hydraulic efficiency were calcu-
 η + 5
lated using the experimental data from the Eqs. (9) and (12)
k while Eqs. (13) and (14) were used to calculate electrical and
η=S (6) shaft power, respectively.
ε
 O. Babayigit et al. / Desalination and Water Treatment 67 (2017) 28–40 33

0.8
0.7
0.6
0.5

|wηh (%)|
0.4
0.3
0.2
0.1
0.0
10 20 30 40 50 60 70 80 90 100
Q ̇ (m³/h)
Fig. 7. Schematic view of the test rig.
Fig. 8. Variation of uncertainty against volume flow rate.

3 I cosϕV
Pe = (13) Continuing decreasing of uncertainty values occurs due to
1000
the fact that volume flow rate increment is more dominant
on the calculation of hydraulic efficiency than the other mea-
Ps = Pe ηe ηm (14)
surement parameters including even high magnitudes of the
pressure difference between the inlet and outlet of the pump.
The efficiency of the electric motor and all mechanic loses
were taken from the manufacturer of the motor according
to loading percentage and also taken into account in exper- 3. Results and discussion
iments. In the experimental studies, uncertainties occurred In this study, a multistage centrifugal pump was exper-
because of measuring equipment, the surrounding environ- imentally and numerically examined. Parameters chosen
ment and human errors made by the person doing the exper- according to the industrial application were Q = 80 m3/h,
iment. In each experiment, the efficiency uncertainty, whh, ṅ = 2,975 rpm and h = 80 m, which were provided by Sempa
caused by different independent variables can be expressed Pump Company in Turkey. These parameters were used as
as in Eq. (15) [34,35]. The uncertainty values were calculated initial values for CFD analysis. For CFD analysis of the pump
with respect to the specifying effects of total errors for the performance, revolution was kept constant while the volume
measured quantities, achieving more accurate measurements flow rate was changed from 16 to 98 m3/h. After the iterations
and assisting in reducing the uncertainties. converged, the parameters of mass flow rate, turbulence
kinetic energy, turbulence distribution rate and velocity
1
 ∂η 2
  ∂η
2
  ∂η
2
  ∂η
2
  ∂η  
2 2 values were controlled to check if the convergence is in the
 h ωP  +  h ωQ  +  h ωI  +  h ωV  +  h ωϕ  
 range of acceptable level of 10–5.
 ∂P   ∂Q   ∂I   ∂V   ∂ϕ  
wηh = ±  100 In flow analysis for the design flow rate, SST k-ω, stan-
ηh
dard k-ε and realizable k-ε turbulence models were applied,
(15) and the results were presented in Table 4. The realizable k-ε
turbulence model has given the best result when compared
Here, hh presents the calculated hydraulic efficiency using with the obtained experimental data. The results of CFD/
Eqs. (9)–(14). Variables wQ. , wP, wI and wV show, respectively, experimental analysis for the pump head and hydraulic effi-
accuracy values of the measurement devices for volume ciency at the design flow rate of 80 m3/h were found to be
flow rate, differential pressure transmitter, ampere meter 81.47/80.70 m and 51.98%/51.42%, respectively.
and voltmeter while whh is relative uncertainty amount (±%) According to the CFD results, the maximum hydraulic
of dependent variable efficiency for the calculation derived efficiency and the head values were calculated as 51.98% and
from the measured and taken values. Accuracy values of the 81.47 m, respectively. A comparison of the hydraulic efficiency
measurement parameters for pressure transmitter (Pa), flow and head value with the volume flow rate for the experimen-
rate (m3/h), electric current (A), voltage (V) and cosϕ value tal and numerical results of the pump systems is presented
were taken as ±0.05%, ±0.04%, ±0.1%, ±0.1% and ±0.1%, in in Figs. 9(a) and (b). The hydraulic efficiency is increasing
sequence. with increasing volume flow rate up to 80 m3/h for the exper-
Other parameters in Eqs. (9)–(14) were assumed to be imental and CFD results while decreasing head occurs at the
constant. Variation of relative uncertainty against the vol- exit of the pump as expected for the general pump charac-
ume flow rate is given in Fig. 8. Uncertainty values of the teristics curve. The difference between the experimental and
efficiency in Eq. (12) depending on the measurement devices numerical simulation results is decreasing as the flow rate
and parameters such as pump inlet and outlet pressures, approaches to the design point value of 80 m3/h, and then, it
volume flow rate, water density, electric current, voltage is increasing with higher flow rate and decreasing in the head
and electric motor efficiency were found to be less than ±1%. due to the complex flow physics.
34 O. Babayigit et al. / Desalination and Water Treatment 67 (2017) 28–40

Comparison of numerical and experimental hydrau- the deviation between the experimental and CFD results for
lic power vs. the flow rate is demonstrated in Fig. 9(c). The head, hydraulic efficiency and hydraulic power values were
hydraulic moment increases until the best efficient point found to be in the range of less than 10% for all flow vol-
value of 80 m3/h is reached, and then, it decreases slightly ume rates and those values were, respectively, determined as
because of having more effective head decrement than vol- 0.95%, 1.1%, and 0.96% at design flow volume rate. Provided
ume flow rate increment as commented in Eq. (9). Variation of results demonstrate that the applied CFD simulations enable
relative difference between the CFD and experimental results to predict the pump performance within an acceptable accu-
is shown in Fig. 9(d). It is seen that relative difference is the racy [26,36].
smallest for the best efficiency point, and it increases with The pumps are designed according to head, flow rate and
increasing or decreasing volume flow rate values. That is, the revolution values. Depending on the distance to the design
relative differences were found to be in the range of 1%–10% flow rate by varying pressure and the velocity of fluid in the
while it was only 1% at the best efficient point. Furthermore, pump may occur some disorders such as: flow separations,
reverse flows, circulating flows and cavitation. Hence, the
turbulence effects on the off-design flow rates are increased.
Table 4 The accuracy of the results, obtained by the turbulence
Comparison of numerical results with different turbulence model used in CFD, depends on its representation ratio of
­models on design flow rate the occurred turbulence effect. Therefore, the relative differ-
ences between the experimental and CFD values are higher at
Model Head (m) Hydraulic efficiency (%) off-design flow rates as seen in Fig. 9.
k-ω SST 77.64 50.74 Numerical flow visualization studies on a separate plane
k-ε standard 79.83 50.35 formed on the pump flow rate were conducted. The YZ plane
was generated from the center of a balance hole of the pump
k-ε realizable 81.47 51.98
flow volume. The cross-sectional view of the flow volume on
Experimental data 80.70 51.42 this plane is shown in Fig. 10. The flow characteristics such

120 Numerical Experimental 60 Numerical Experimental

110 55
100 50
90 45
80 40
70 35
h (m)

ηh (%)

60 30
50 25
40 20
30 15
20 10
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
̇ ̇
Q (m³/h) Q (m³/h)
(a) (b)
20 Numerical Experimental
18 11 h η P
16 10
relative difference (%)

9
14 8
12 7
6
Ph (kW)

10
5
8 4
6 3
2
4 1
2 0
0 10 20 30 40 10 20 30 40 50 60 70 80 90 100
̇ 50 60 70 80 90 100
Q (m³/h) Q (m³/h)
(c) (d)

Fig. 9. Comparison of the numerical and experimental: (a) head, (b) hydraulic efficiency, (c) hydraulic power of the pump vs. the
volume flow rate and (d) the variation of relative difference between the CFD and experimental results.
 O. Babayigit et al. / Desalination and Water Treatment 67 (2017) 28–40 35

with balance hole without balance hole

(a)

Fig. 10. CFD flow visualization plane on YZ.

as velocity vectors, velocity contours, streamline patterns and with balance hole without balance hole
static pressure for the flow rates of 55, 80 and 93 m³/h for (b)

the impeller models with and without balance holes are pre-
pared at this plane as shown in Figs. 11–13.
Two-dimensional vector fields and corresponding veloc-
ity contours in YZ cross-section plane are indicated in Fig. 11,
in which acceleration and deceleration of the flow ways can
be seen clearly. From the exit of the impeller to the entrance of
the diffuser, flow becomes slower due to the increasing flow
area. The reverse flows and increasing of flow velocity in the with balance hole without balance hole
(c)
region, where the clearances and balance holes were formed,
can be seen in Figs. 11 and 12. When the impeller model Fig. 11. Tangential velocity vector and contours on YZ plane on
without balance holes compared with the balance holes in all flow rates of: (a) 55 m³/h, (b) 80 m³/h and (c) 93 m³/h.
three flow rates; lower velocity values were seen at the leak-
age flow region in front of the impellers; on the other hand,
a quite low velocity occurred at the leak flow region behind
impellers. It is seen in Fig. 11 that the velocity is getting lower
by the decreasing flow rate in the pump inlet region. A swirl-
ing flow occurs due to the large velocity gradient, and thus,
focus forms around tip of the blade as seen from streamline
topology in Fig. 12. A slight increase for the flow velocity and
reverse flow in the entrance region of the discharge section
was also noticed. Curved and recirculated streamline pat-
terns demonstrate the acceleration, deceleration and flow with balance hole
(a)
without balance hole

separation phenomena.
Fig. 13 displays the static pressure contours at the same
plane in Figs. 11 and 12. As it can be seen from the all col-
ored legends, the static pressure increases gradually from the
pump suction to the discharge side. In the absence of bal-
ance holes, higher static pressure distributions were occurred
toward the pump outlet compared with the ones with bal-
ance holes in all three flow rates. In addition, when flow rate with balance hole without balance hole
increase, a remarkable drop in static pressure was obtained (b)

from the second stage to the pump outlet for all three flow
rates with and without balance holes.
The effects of balance holes on the axial force on the
pump are also investigated. All forces in the axial direc-
tion are m­ easured and compared at the design flow rate of
80 m3/h for the impeller models with and without balance
holes. The CFD analyses show that the magnitude of the total
axial forces was found to be 875 N without balance holes; with balance hole
(c)
without balance hole

whereas, it was reduced up to 3 N when balance holes are

applied. It can be concluded that the balance holes have the Fig. 12. Streamline patterns on YZ plane on flow rates of:
great effect on reducing the total axial forces. (a) 55 m³/h, (b) 80 m³/h and (c) 93 m³/h.
36 O. Babayigit et al. / Desalination and Water Treatment 67 (2017) 28–40

with balance hole without balance hole

(a)

with balance hole without balance hole

(b)

with balance hole without balance hole

(c)

Fig. 13. Static pressure contours on YZ plane on flow rates of:

(a) 55 m³/h, (b) 80 m³/h and (c) 93 m³/h.

All leakages passing through the balance holes and clear-

ances between the impeller, the diffuser and suction side are
shown in Fig. 14. The total leak flow is determined as follows: Fig. 14. Representation of the flow directions in the first stage of
the pump.
Q L = Q L1 + Q L 2 + Q L 3 (16)

Here, Q L1 is the leakage occurring from outlet to the

inlet side of the impeller, and it is included to the main flow
entering into the inside of the impeller. Q L2 is the leakage
existing from outlet to the back size of the impeller. Q L3 is the
leakage happening from diffuser outlet to the back size of
the impeller in a similar manner, and Q L represents the total
leakages. Both Q L2 and Q L3 are back into the impeller inside
from balance holes located at the rear side of the impeller.
It is evident that all the leakage stream flow toward the
low-pressure region from the high-pressure side is shown
in Figs. 13(b) and 14.
Variation of the total leakages and the ratio of leak
flows against the main flow is shown in Fig. 15. As seen, the
total leak flow is almost unchanged at all main flow rates.
However, it is observed that the ratio of the total leak flow Fig. 15. Variations of the total leakages and the ratio of leak flows
rate is increasing with the decrease of the main flow rate in depending on the flow rate.
a significant way. This ratio is about 9% at the design flow
rate while rising up to about 50% at low flow rates. Leak flow
mainly depends on the clearance geometry, impeller rota- same in different flow rates as seen in Fig. 13. Therefore, leak
tional speed and head [37,38]. In case of fixed geometrical flow rates were found to be close to each other.
features and constant rotating speed as this study, leak flows All leakage flow rates detected for design flow rate of
therefore depend on the impeller static pressure mainly. The 80 m3/h are given in Table 5. It is seen from the table that Q L1,
static pressures at the exit of the first impeller were nearly the Q L2 and Q L3 are approximately 41%, 42% and 17% of the total
 O. Babayigit et al. / Desalination and Water Treatment 67 (2017) 28–40 37

Table 5
Main and leakage flow rates on the design point

Parameter Volume flow rate (m³/h)

Q 80
QL
7.38
Q L1 3.05
QL2
3.06
Q L3 1.27

leak flow rate of Q L = 7.38 m3/h, respectively. These different

ratios occur depending on the gap distance and the pressure
difference.
The mid-span planes on XY plane were generated from
the center of the impellers. A cross-sectional view of the flow
volume of the pump on these planes is shown in Fig. 16.
Tangential velocity vectors on the total velocity, static pres-
sure contours and turbulence kinetic energy contours for the
first and the second stages of the impeller models with and
without balance holes on the mentioned planes are presented Fig. 16. CFD flow visualization mid-span planes on XY for the
in Figs. 17 and 18. It can be interpreted from these images first and second stage.
that the impeller and geometry design of diffuser convert
smoothly the mechanical shaft power to the flow energy.
When the balance holes are included, flow separa-
tion, formed around positive pressure sides of the impeller
blades, decreases. Velocity distribution around balance holes
is higher than in the other regions for both stages as seen in
Figs. 17(a) and 18(a).
As indicated in Figs. 17(b) and 18(b), in the absence of
balance holes, regular increase of static pressure is seen
from the entrance to the exit of the impeller for both stages. with balance hole without balance hole
However, in the presence of balance holes, a sudden pressure (a)
drop through the six balance holes placed around the flow
entrance region between the blades occurs due to the large
pressure difference between the stages. Increments of the
static pressure for both stages are approximately the same
when the momentum effect is neglected because of the bal-
ance hole.
The turbulence kinetic energy dissipation for both stages
occurs in a similar manner as shown in Figs. 17(c) and 18(c).
When there aren’t any other balance holes, nearly the same
turbulence kinetic energy distributions occur through the
with balance hole without balance hole
planes for both stages. It can be seen that the turbulence (b)
kinetic energy values increase in regions where the balancing
holes and entrance of the diffuser vanes were located. The
distinguished increment of the turbulence kinetic energy
values takes place around the balance holes. Moreover, it is
demonstrated that the effect of the balance holes causes the
pressure magnitude decrement as presented in Figs. 17(b)
and 18(b).
To show the effect of balance hole on the pump stage, some
points are created crossing over the balance hole in mid-span
on XY plane in the first and the second impellers as shown in
with balance hole without balance hole
Fig. 19. The point indicated by number 3 is located in the cen- (c)
ter of the balance hole. Static pressure and turbulence kinetic
energy values on these points are measured for the impeller Fig. 17. (a) Tangential velocity vectors and contours, (b) static
models with and without balance holes. It can be seen in the pressure contours and (c) turbulence kinetic energy contours at
Fig. 20(a) that the static pressure of the impeller model without mid-span on XY for the first stage on the design flow rate.
38 O. Babayigit et al. / Desalination and Water Treatment 67 (2017) 28–40

920 impeller 1
impeller 2
820
impeller 1 without balance hole

Static Pressure (kPa)

720 impeller 2 without balance hole
620
520
420
320
220
120
with balance hole without balance hole
20
(a)
-80
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Point

(a)

55 impeller 1
50 impeller 2

Turb. Kin. Energy (J/kg)

45 impeller 1 without balance hole
40 impeller 2 without balance hole
35
with balance hole without balance hole 30
(b) 25
20
15
10
5
0
-5
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Point

(b)

with balance hole without balance hole Fig. 20. (a) Static pressure and (b) turbulence kinetic energy on
(c) defined points shown in Fig. 19 for the first and second stage
on design flow rate for the impeller models with and without
Fig. 18. (a) Tangential velocity vectors and contours, (b) static balance holes.
pressure contours and (c) turbulence kinetic energy contours at
mid-span on XY for the second stage on the design flow rate.
linearly increasing again until reaching the impeller exit to the
location point 12. The static pressure shows a small decrease
on the impeller exit while increasing turbulence kinetic energy
due to the mixing and sectional differences between the impel-
ler exit and diffuser entrance.
It can be seen from the Fig. 20(b) that the turbulence
kinetic energy of the impeller model without balance holes
stays almost constant up to the impeller exit. While approach-
ing the exit, the turbulence kinetic energy rises. While the
impeller model with balance holes is taken into consider-
ation, turbulence kinetic energy nearly does not change from
the impeller entrance to the balance hole, but it increases sig-
nificantly on the balance hole and takes its maximum value at
point 4. Then it starts decreasing for a while and nearly stays
unchanged to the impeller exit. It shows an increase on the
impeller exit due to sectional differences between the impel-
Fig. 19. The points following blade-profile and crossover balance ler exit and diffuser entrance. It can be inferred that balance
hole at mid-span plane of the impeller. holes are the major factor, which causes turbulence kinetic
energy in the impeller.
balance holes continuously increases up to region close to the
exit of the impeller. While approaching to the exit of the impel-
4. Conclusions
ler, the static pressure increases at slower rate. In case of the
impeller model with balance holes, static pressure increases on In this work, the CFD simulations and experimental
a regular basis from the impeller entrance to the balance hole, studies of a two-stage centrifugal pump were presented in
but it shows a sharp decrease on the balance hole, and then, it is detail including the effects of balance holes and leakages.
 O. Babayigit et al. / Desalination and Water Treatment 67 (2017) 28–40 39

The numerical results of the pump obtained by CFD were Symbols

compared with the test results of the pump manufactured
b1 — Entry width of the impeller, mm
by Sempa Company in Turkey. The following improvements
b2 — Exit width of the impeller, mm
and achievements were obtained:
C1 — k-ε model constant
• The CFD analyses of the pump were completed under C2 — k-ε model constant
different operational conditions including the balance C1ε — k-ε model constant
holes and the leakages providing a real case simulation. C2ε — k-ε model constant
Thus, the experimental and numerical results were nearly C3 — k-ε model constant
overlapped with each other. The best results of the CFD Cμ — k-ε model constant
and experimental analyses for the pump head, hydrau- D1 — Impeller inlet diameter, mm
lic efficiency and hydraulic power on the flow rate of D2 — Impeller outlet diameter, mm
80 m3/h were found to be 81.47/80.70 m, 51.98%/51.42%, D3 — Diffuser inlet diameter, mm
and 17.94/17.77 kW, respectively. D4 — Diffuser outlet diameter, mm
• Variation of relative difference between the CFD and F — External body force
experimental results shows that the relative difference is g — Acceleration of gravity, m/s²
the smallest for the best efficiency point and it increases Gk — Generation of turbulence kinetic energy due to the
with increasing or decreasing volume flow rate values. mean velocity gradients
That is, the relative differences were found to be in the Gb — Generation of turbulence kinetic energy due to
range of 1%–10% while it was only 1% at the best efficient buoyancy
point. h — Pump head, m
• It is also shown that the balance holes affect the flow I — Current, A
structures, pressure distributions and turbulence kinetic K — Turbulence kinetic energy, m2/s2
energy values significantly so that it should be consid- nq — Specific speed
ered in CFD studies. ṅ — Rotating speed, rpm
• Balance holes reduce the hydraulic efficiency of the pump N — Stage number
by increasing the leak flow, although it is important to P — Static pressure, Pa
reduce the axial forces of the pump shaft. Designing Pe — Electric power, kW
­double-suction pumps can be an alternative to balance Ph — Hydraulic power, kW
these axial forces. Ps — Shaft power, kW
• It is revealed that the obtained experimental results are Q — Volume flow rate, m3/h
repeatable and reliable because of having very low uncer- Q L — Total leak volume flow rate, m3/h
tainty values around ±1%. Q L1 — Leak volume flow rate, m3/h
• It is demonstrated as seen in Table 5 that Q L1, Q L2 and Q L3 Q L2 — Leak volume flow rate, m3/h
are approximately 41%, 42% and 17% of the total leak QL3
— Leak volume flow rate, m3/h
flow rate of Q L = 7.38 m3/h, respectively. These differ- S — Mean strain-rate of the flow
ent ratios occur depending on the gap distance and the Sij — Deformation tensor
pressure difference. The ratio of the total leak flow to the Sk — User defined source term
design flow rate is around 9% for the examined pump. Sε — User defined source term
• Parameters that determine the leak flow rate are the t — Time, s
size of the clearances between the fixed and rotating T — Torque, N.m
­components of the pump. It is not possible to prevent uj — y direction component of velocity vector
the ­leakage, but some solutions can be applied to reduce V — Voltage, volt

these clearances such as using seal and wear rings. v — Absolute velocity vector, m/s
• Achieving an optimal balance between leakage and fric- ω — Angular velocity, rad/s
tion can be an essential goal for the future studies as small w — Uncertainty
gap limits of the leakage flow; however, balance elements 
w — Relative velocity vector, m/s
increases the friction, risk of drag and noise. YM — Contribution of the fluctuating dilatation
• In future, some studies on the location and a size opti- y+ — Distance of the first cell center from the solid wall
mization of the balance holes for the pump performance z — Number of impeller blade
increment can be performed. zv — Number of diffuser blade
zr — Number of diffuser return blade
η — The ratio between the time scales of the turbulence
Acknowledgments
and the mean flow
This work is supported by SANTEZ program of the ηe — Electric motor efficiency, %
Ministry of Science, Industry and Technology of Turkey with ηh — Hydraulic efficiency, %
number 01076.STZ.2011-2 and Selcuk University’s Scientific ηm — Mechanical efficiency, %
Research Project Office Contract No.: 14401075. The authors µt — Coefficient of turbulence dynamic viscosity, kg/m.s
also would like to thank to the partner of the project Sempa σε — Prandtl number for ε
Co. board (http://www.sempaltd.com/en/) for their support. σij — Stress tensor
This study is a part of Osman Babayigit’s PhD Thesis. σk — Turbulent Prandtl number for k
40 O. Babayigit et al. / Desalination and Water Treatment 67 (2017) 28–40

∇ — Laplace operator networks and genetic algorithms, Eng. Appl. Comp. Fluid,
β1 — Inlet blade angle of the impeller, 0 5 (2011) 37–48.
β2 — Outlet blade angle of the impeller, 0 [18] K.H. Wu, B.J. Lin, C.I. Hung, Novel design of centrifugal pump
impellers using generated machining method and CFD, Eng.
β3 — Outlet blade angle of the diffuser, 0 Appl. Comp. Fluid, 2 (2008) 195–207.
ε — Turbulence distribution rate [19] O. Kocaaslan, M.H. Aksoy, O. Babayigit, M. Ozgoren, An
εk — Turbulent Prandtl number for ε Experimental Investigation of Surface Roughness Influence
ϕ — Impeller wrap angle, 0 on Centrifugal Pump Characteristics, Proc. International
ϕv — Diffuser wrap angle, 0 Conference: EFM 2013, Kutna Hora, Czech Republic, 2013,
ϕr — Diffuser return wrap angle, 0 pp. 814–819.
[20] W. Jiang, G. Li, P.F. Liu, L. Fu, Numerical investigation of
cosϕ — Power factor influence of the clocking effect on the unsteady pressure
µ — Laminar dynamic viscosity coefficient, kg/m.s fluctuations and radial forces in the centrifugal pump with
r — density, kg/m3 vaned diffuser, Int. Commun. Heat Mass Transfer, 71 (2016)
τ — Shear stress, kPa 164–171.
[21] H. Stel, T. Sirino, F.J. Ponce, S. Chiva, R.E.M. Morales, Numerical
investigation of the flow in a multistage electric submersible
References pump, J. Pet. Sci. Eng., 136 (2015) 41–54.
[22] F. Ayad, H.M. Abdalla, A. Abou El-Azm Aly, Effect of semi-open
[1] J. Perez, S. Chiva, W. Segala, R. Morales, C. Negrao, E. Julia,
impeller side clearance on the centrifugal pump performance
L. Hernandez, Performance Analysis of Flow in a I­mpeller-
using CFD, Aerosp. Sci. Technol., 47 (2015) 247–255.
Diffuser Centrifugal Pumps Using CFD: Simulation and
Experimental Data Comparisons, ECCOMAS CFD 2010, [23] A. Engeda, M. Rautenberg, Partial Flow Performance
Lisbon, Portugal, 2010, pp. 1–18. Comparisons of Semi-open and Closed Centrifugal Impellers,
[2] M. Nataraj, R. Singh, Analyzing pump impeller for performance ImechE, C334/88 Edinburgh, U.K. Bury St. Edmunds, U.K.,
valuation using RSM and CFD, Desal. Wat. Treat., 52 (2014) Mech. Engng. Publications Ltd., 1988.
6822–6831. [24] S. Liu, N. Michihiro, K. Yoshida, Impeller geometry suitable for
[3] D. Kaya, E.A. Yagmur, K.S. Yigit, F.C. Kilic, A.S. Eren, C. Celik, mini turbo-pump, J. Fluids Eng., 123 (2001) 500–506.
Energy efficiency in pumps, Energy Convers. Manage., 49 (2008) [25] H. Liu, J. Wang, Y. Wang, H. Huang, L. Jiang, Partially-
1662–1673. averaged Navier-Stokes model for predicting cavitating flow in
[4] M.H. Aksoy, O. Babayigit, O. Kocaaslan, M. Ozgoren, Effect centrifugal pump, Eng. Appl. Comp. Fluid, 8 (2014) 319–329.
of Blade Wrap Angle to Centrifugal Pump Impeller Efficiency, [26] L. Zhou, W. Shi, W. Lu, B. Hu, S. Wu, Numerical investigations
Proc. International Conference: EFM 2013, Kutna Hora, Czech and performance experiments of a deep-well centrifugal pump
Republic, 2013, pp. 799–806. with different diffusers, J. Fluids Eng., 134 (2012) 1–8.
[5] S. Chakraborty, K.M. Pandey, Numerical studies on effects of [27] S.D. Kyparissis, E.C. Douvi, E.E. Panagiotopoulos, D.P.
blade number variations on performance of centrifugal pumps Margaris, A.E. Filios, CFD flow field analysis and hydrodynamic
at 4000 rpm, IJET, 3 (2011) 410–416. double-arc blade design effects for optimum centrifugal pump
[6] A. Arnone, P. Boncinelli, A. Munari, E. Spano, Application of performance, Int. Rev. Mech. Eng., 3 (2009) 284–294.
CFD Techniques to the Design of the Ariane 5 Turbopump, [28] Anonymous, Fluent 14.0 User Guide, Fluent Inc., 2015.
14th Computational Fluid Dynamics Conference, AIAA paper [29] H. Versteeg, W. Malalasekra, An Introduction to Computational
99-3380, Norfolk, VA, USA, 1999, pp. 1087–1097. Fluid Dynamics: The Finite Volume Method, John Wiley &
[7] O. Babayigit, O. Kocaaslan, M.H. Aksoy, K.M. Güleren, M. Sons, New York, 1995.
Ozgoren, Numerical identification of blade exit angle effect on [30] K.M. Pandey, S. Chakraborty, A.P. Singh, Numerical analysis of
the performance for a multistage centrifugal pump impeller, centrifugal pumps with fluent software, Int. J. Appl. Eng. Res.,
EPJ Web Conf., 92 (2015) 1–7. 6 (2011) 1117–1126.
[8] S. Yedidiah, A study in the use of CFD in the design of [31] W.B. Fan, W.G. Li, X.J. Gong, X.R. Zhang, Evaluation of
centrifugal pumps, Eng. Appl. Comp. Fluid, 2 (2008) 331–343. the effect of a hydraulic impeller in a flocculation basin on
[9] M.L. Hedia, K. Hatema, Z. Ridhaa, Numerical analysis of the hydrodynamic behavior using computational fluid dynamics,
flow through in centrifugal pumps, Int. J. Therm. Technol., Desal. Wat. Treat., 54 (2015) 1361–1374.
2277–4114 (2012) 216–221. [32] K.K. Singh, S.M. Mahajani, K.T. Shenoy, A.W. Patwardhan, S.K.
[10] V. Godbole, R. Patil, S.S. Gavade, Axial Thrust in Centrifugal Ghosh, CFD modeling of pilot-scale pump-mixer: single-phase
Pumps – Experimental Analysis, ICEM 15, Porto/Portugal, head and power characteristics, Chem. Eng. Sci., 62 (2007)
2012, pp. 1–14.
1308–1322.
[11] The Centrifugal Pump, Grundfos Research and Technology,
[33] B.K. Baysal, Centrifugal Pumps Account Drawings and
Chapter 5, Pump Losses, Grundfos Management A/S,
Construction Specifications, Istanbul Technical University
Bjerringbro, Denmark, 2012.
Publishing House, 1979. (in Turkish)
[12] J.F. Gulich, Centrifugal Pumps, Springer, Berlin, Germany,
2014. [34] J.P. Holman, J.G. Walter, Experimental Methods for Engineers,
[13] B. Cui, Z. Zhu, J. Zhang, Y. Chen, The flow simulation 2nd ed., McGraw-Hill, New York, 1994.
and experimental study of low-specific-speed high-speed [35] O. Kocaaslan, M. Ozgoren, M.H. Aksoy, O. Babayigit,
complex centrifugal impellers, Chin. J. Chem. Eng., 14 (2006) Experimental and numerical investigation of coating effect
435–441. on pump impeller and volute, J. Appl. Fluid Mech., 9 (2016)
[14] B. Jafarzadeh, A. Hajari, M.M. Alishahi, M.H. Akbari, The 2475–2487.
flow simulation of a low-specific-speed high-speed centrifugal [36] S.S. Yang, S. Derakshan, F.Y. Kong, Theoretical, numerical
pump, Appl. Math. Modell., 35 (2011) 242–249. and experimental prediction of pump as turbine performance,
[15] W.G. Li, Effects of viscosity on turbine mode performance and Renew. Energy, 48 (2012) 507–513.
flow of a low specific speed centrifugal pump, Appl. Math. [37] S. Salvodori, A. Marini, F. Martelli, Methodology for the residual
Modell., 40 (2016) 904–926. axial thrust evaluation in multistage centrifugal pumps, Eng.
[16] K.W. Cheah, T.S. Lee, S.H. Winoto, Z.M. Zhao, Numerical flow Appl. Comp. Fluid, 6 (2012) 271–284.
simulation in a centrifugal pump at design and off-design [38] P. Adami, S.D. Gatta, F. Martelli, L. Bertolazzi, D. Maestri, G.
conditions, Int. J. Rotating Mach., 2007 (2007) 1–8. Marenco, A. Piva, Multistage Centrifugal-Pumps: Assessment
[17] H. Safikhani, A. Khalkhali, M. Farajpoor, Pareto based multi- of a Mixing Plane Method for CFD Analysis, 60° Congresso
objective optimization of centrifugal pumps using CFD, neural Nazionale ATI, Rome, 2005.

View publication stats