WSEAS TRANSACTIONS ON FLUID MECHANICS

Issue 7, Vol. 1, July 2006

ISSN: 1790-5087

763

CFD Analysis and Design Effects in a Radial Pump Impeller

JOHN S. ANAGNOSTOPOULOS
School of Mechanical Engineering / Fluids Section
National Technical University of Athens
9 Heroon Polytechniou ave., Zografou, 15780 Athens
GREECE
[email protected]
Abstract: The direct flow analysis in hydraulic turbomachines using Computational Fluid Dynamic (CFD)
methods can provide a quite illuminating picture of the developed flow field and its detailed characteristics. A
numerical model for the simulation of the 3-dimensional turbulent flow in centrifugal pump impellers is
developed in the present work, solving the Reynolds Averaged Navier Stokes equations with the control
volume approach and on Cartesian grids. The latter can be constructed by a very fast and fully automated
algorithm, whereas an effective numerical technique for solving the partially filled grid cells that are formed on
the irregular boundaries of the computational domain is also incorporated. The computations for the steady
flow field in a particular impeller are presented and analyzed, and the characteristic performance curves are
constructed. The impeller geometry is represented by a number of controllable design variables, providing the
capability of modifying the impeller shape and testing different configurations. This technique, combined with
the automated grid generation algorithm facilitates the investigation and assessment of the effects of impeller
design on its hydrodynamic behaviour. The results of such parametric studies conducted in the present work
show that a remarkable gain in hydraulic efficiency may be achieved by optimizing the impeller geometry.
Key-Words: Centrifugal pumps; Radial flow impeller; Numerical modelling; RANS equations; Cartesian grids;
Design optimization.

1 Introduction
A wide variety of centrifugal pump types have been
constructed and used in many different applications
in industry and other technical sectors. However,
their design and performance prediction process is
still a difficult task, mainly due to the great number
of free geometric parameters, the effect of which
cannot be directly evaluated. The significant cost
and time of the trial-and-error process by
constructing and testing physical prototypes reduces
the profit margins of the pump manufacturers. For
this reason CFD analysis is currently being used in
the design and construction stage of various pump
types [1-4]. Unsteady and dynamic phenomena, as
the rotor-stator interaction can be also studied with
the aid of CFD [5], the use of which reduces
significantly the new pump development costs. The
average reduction is estimated to 65% during 2005
[6].
The numerical simulation can provide quite
accurate information on the fluid behaviour in the
machine, and thus helps the engineer to obtain a
thorough performance evaluation of a particular

design. However, the challenge of improving the

hydraulic efficiency requires the inverse design
process, in which a significant number of alternative
designs must be evaluated. Despite the great
progress in the latest years, even CFD analysis
remains rather expensive for the industry, and the
need for faster mesh generators and solvers is
imperative [4].
The mesh generation process is a laborious task
for many CFD codes, and the quality of the final
mesh depends considerably on the users
experience. An alternative practice in complex
domains is the use of Cartesian grids that require a
much reduced construction effort. However such
grids cannot be everywhere body-fitted, and for this
reason various numerical techniques have been
developed to improve the accuracy in these regions
[7-8].
A Cartesian mesh approach is also followed in
the present work, where an advanced numerical
technique is incorporated in order to eliminate the
grid generation cost and to represent with adequate
accuracy the complex geometry of a centrifugal
pump impeller. The latter is parameterized using a

764

WSEAS TRANSACTIONS ON FLUID MECHANICS

reduced number of controlling geometric variables,

facilitating the investigation of their individual or
combined effects on the flow and the impeller
performance. The results of the predicted flow field
in the pump impeller illustrated in Fig. 1, as well as
of a parametric study concerning the variation of
several design variables are then presented and
discussed.

Issue 7, Vol. 1, July 2006

ISSN: 1790-5087

All the dependent variables are stored at the cell

centers of a uniform Cartesian mesh and the
governing equations are discretized with the finite
volume approach using a hybrid difference scheme
[10]. The system of the linearized form of the above
equations is numerically solved by a preconditioned
bi-conjugate gradient (Bi-CG) solver [11]. Fully
developed velocity profile is assumed for the flow at
the inlet section of the tube, where the reference
pressure is set to zero, and no-slip boundary
conditions are applied at all internal solid surfaces.
Due to the periodic symmetry of the impeller
geometry, the computational domain can be
restricted to a section of 2/z degrees, where z is the
number of blades, using periodicity boundary
conditions at the lateral planes.

(a)
Figure 1. The examined radial impeller.

2 Numerical Method
2.1 Flow equations
The steady, incompressible, Reynolds Averaged
Navier-Stokes (RANS) equations are employed for
the flow calculations in polar coordinates and a
rotating with the impeller system of reference. The
governing conservation equations are expressed in
vectorized form as follows:
r
w = 0
(1)
Continuity:
Momentum:
r r
r r
r 1
r
1
w w = 2 w + 2 r p +
(2)

r
where w is the relative fluid velocity, is the
r
angular rotation speed of the impeller, r is the radial
distance and p, are the fluid pressure and density,
r
respectively. The viscous stress tensor includes
both the viscous and the turbulence viscosity terms:
ij = 2 sij wi wj

(b)

(3)

where is the fluid dynamic viscosity and sij is the

strain tensor. The second term on the right side of
the above relation represents the Reynolds stresses
due to turbulent motion. Since the Reynolds number
in a typical pump is high (Re > 104), the standard k-
model is adopted for the turbulence closure [9], by
solving additional conservation equations for the
turbulence kinetic energy production, k, and its
dissipation rate, .

Figure 2. Parameterization of the impeller

geometry: a) side view; b) top view.

2.2 Geometry parameterization

The geometry of the particular radial flow impeller
examined here (Fig. 1) corresponds to a model
impeller constructed in the Lab, and can be
represented using a relatively small number of
parameters; most of them are shown in Fig. 2 and

WSEAS TRANSACTIONS ON FLUID MECHANICS

their values are given in Table 1. The rotation speed

and the main impeller dimensions, namely the exit
diameter and width D2 and b2, as well as the blade
inlet and exit angles, 1 and 2, respectively,
determine the nominal head and volume flow rate of
the impeller. The computational domain is extended
to a certain radial distance beyond the impeller
outlet (here up to r = 125 mm, not shown in Fig. 2),
to prevent any backward influence of the free vortex
conditions set at the exit boundary.
The blades are constructed as circular arcs, and
they have constant width and both edges rounded,
allowing for both pump and turbine operation mode
of the impeller. The rest parameters constitute the
free design variables that can be modified in order to
improve the performance and hydraulic efficiency
of the impeller for this particular nominal operation
point.
Table 1. Impeller standard dimensions.
Parameter
Suction tube
Inlet diameter
Exit diameter
Exit width
Hub size
Hub height
Shroud inlet
Shroud inlet
Blades number
Inlet angle
Outlet angle
Blade thickness

Size
Di = 60 mm
D1 = 70 mm
D2 = 190 mm
b2 = 9 mm
Dh = 80 mm
Lh = 20 mm
Rs = 10 mm
Ls = 25 mm
9
1 = 26 deg
2 = 49 deg
s = 5 mm

Issue 7, Vol. 1, July 2006

( A P V S P ) P =
AP =

i = E , W , N , S ,U , D

(4)

where Ai are the linking coefficients between the

variable P and its neighboring values at the
adjacent grid cells, and SU, SP include the rest source
terms. The geometric coefficients i and v represent
the free portion (not blocked by the solid boundary)
of the cell faces and volume, respectively, and they
are automatically computed by a pre-processing
algorithm, along with the rest geometric quantities.
The fluid variables (velocities, pressures, etc) are
evaluated at the centroid of the Cartesian cells,
which may not coincide with the geometric centre of
a boundary cell (Fig. 3). The wall boundary
conditions are also set automatically to every
boundary cell (e.g. cells P1 to P4 in Fig. 3), with the
aid of a special algorithm that computes the normal
distance from the wall and the exact bounded area
within each grid cell.
In addition to its simplicity and generality, that
makes PFC method easily applicable to 3dimensional geometries of any complexity, the
conservation property is retained and the accuracy
of the boundary representation remains satisfactory
even with a coarse grid, and without affecting the
stability of the solver.
P2
E

fluid
P1

2.3 Numerical grid

P4
C
B

The important advantage from the use of Cartesian

grids for complex geometries is the elimination of
the computational effort and the automation of the
grid generation process. However the grid lines can
not in general fit in with the flow boundaries,
therefore there are cells partly filled by the fluid that
must be properly handled, as illustrated in the 2dimensional example of Fig. 3.
The Partly-Filled-Cells (PFC) method is
implemented for that purpose in the present work. It
is a cell-cut, sharp-interface grid technique
developed and successfully tested in various studies
in the past [12-13]. A desirable feature of the
method is that the same general form of the
linearized flow equations can be used for all the grid
cells, regardless of their blocked portion:

A + S
i

i Ai ,

765

ISSN: 1790-5087

free cell
partly filled
blocked

solid

P3
A

Figure 3. Example of the PFC grid method.

The numerical grid used for the present
calculations has a number of approximately 200,000
nodes, which was found to be a good compromise
between the results accuracy and the computation
speed. An indicative picture of the grid
configuration is given in Fig. 4 at the leading edge
region of a blade.

766

WSEAS TRANSACTIONS ON FLUID MECHANICS

Issue 7, Vol. 1, July 2006

Mu =

r2
r1

2.4 Hydraulic efficiency calculation

The hydraulic efficiency h of the impeller is
defined as the ratio of the net energy added to the
passing fluid, divided by the energy given at the
impeller shaft. The specific energy (energy per unit
weight of the fluid) can be expressed by the
corresponding heads, H and Hu, therefore the
efficiency is given by:
H
Hu

(5)

The net fluid head H is obtained by energy

balance at the inlet and the outlet of the impeller,
using the following flux-weighted relation (Fig. 2b):
= H 2 H1 =

Qu

p 2 p1 c 22 c12

g + 2g

dq (6)

where c is the absolute velocity of the fluid, Qu the

volume flow rate through the impeller and g the
gravity acceleration, while the subscripts 1 and 2
denote impeller inlet and exit conditions,
respectively. The right-hand side integral is
approximated by a summation over the radial flow
rates q at all grid cells facing the inlet or the exit
circumference of the impeller
The impeller head Hu can be calculated from the
definition equation of the absorbed power at the
shaft:

N u = g Qu H u

[(rr nr ) p + (rr rw ) cot ] b dr

(rr rw ) dA

(9)

r
where n is the local unit vector normal to the
r
surface, w the wall shear stress, the blade angle,
and b the impeller width. The first integral term on
the right represents the torque developed on the
impeller blades due to the pressure and the friction
forces, and includes both the pressure and the
suction side of the blades. The second integral
stands for the torque on the internal shroud and hub
surfaces, and contains only the friction forces, since
the pressure forces does not have a circumferential
component.

Figure 4. Detail of the grid lines arrangement.

h =

ISSN: 1790-5087

(7)

where Nu is analogous to the torque Mu developed

on the impeller:
Nu = M u
(8)
The latter can be computed based on the
conservation of angular momentum law, via the
following relation (Fig. 2b):

3 Results
3.1 Flow analysis
At first the numerical model is applied to calculate
the flow field developed in the standard design
impeller (Table 1), for various load conditions and
for a constant rotation speed of 1500 rpm. The
nominal volume flow rate is taken 31 m3/h to
comply with the corresponding laboratory model
pump design point.
Fig. 5a shows the resulting contours of pressure
and the velocity vectors at two grid levels that cross
the impeller normal to- and through its axis of
rotation. Increased flow velocity can be observed at
the blade inlet due to the blockage of the flow,
whereas on the contrary the pressure is reduced.
Further downstream the contours become smooth
between the blades and the pressure increases
continuously towards the exit of the computational
domain.
The minimum pressure appears, as expected, at
the suction side and near the leading edge of the
blade (Fig. 5b). The flow seems to enter almost
parallel to the blade (Fig. 5b) and the streamlines
follow a regular pathway between the blades, except
of the upper section near the shroud, where some
recirculation can be observed (Fig. 5c).
Fig. 6 illustrates the corresponding flow field for
a much reduced flow rate, equal to 20% of the
nominal. The different flow characteristics can be
clearly observed: The pressure gradients are lower
throughout the domain and the minimum values are
higher (Fig. 6a, 6b). However, a strong recirculation
is established within almost the entire blade-to-blade
region which, according to the theory, rotates in the
reverse direction than the impeller.

WSEAS TRANSACTIONS ON FLUID MECHANICS

(a)

Issue 7, Vol. 1, July 2006

767

ISSN: 1790-5087

(a)

(b)
(b)

(c)
(c)

Figure 6. Flow field analysis for Qu = 6,2 m3/h.

Legend as in Fig. 5.
Figure 5. Flow field analysis for Qu = 31 m3/h:
a) Pressure contours and velocity vectors; b) stream
lines and minimum pressure at the blade leading
edge; c) stream lines between the blades.
The results for the case of a double volume flow
rate (62 m3/h) were finally obtained and they are
shown in Fig. 7. Now the pressure and the velocity
fields exhibit higher gradients but no important
recirculation is observed in the impeller. However,
due to the higher radial flow velocity at the blades
inlet, the stagnation line is moved towards the
suction side of the blade leading edge, consequently

Figure 7. Stream lines and minimum pressure at the

blade leading edge for Qu = 62 m3/h.

768

WSEAS TRANSACTIONS ON FLUID MECHANICS

the minimum pressure is displaced to the opposite,

pressure side, while its value becomes even lower
than in nominal conditions (Fig. 5b).

3.2 Impeller performance

The numerical model is then applied to reproduce
the performance characteristic curves of the
standard impeller at 1500 rpm, covering the entire
load range from 20% to 200% of the nominal
value. The computed curves are drawn in Figs. 8a
and 8b and the first observation is that the maximum
hydraulic efficiency is achieved for Q 22,5 m3/h,
which is lower than the considered nominal value
of 31 m3/h. However, a better estimation of the latter
for the simulated geometry with the aid of the inlet
triangle of velocity vectors gives about 25 m3/h and
agrees better with the numerically obtained point.

84

14

81

Hydraulic efficiency
Net fluid head

12

10

20

30

40

78

50

Flow rate, Q (m3/h)

60

Nu (KW)

(b)

(a)

70

(b)

75

Minumum pressure
Impeller power
10

20

30

40

50

Flow rate, Q (m 3/h)

60

70

94

standard
design

16

91

15

88

Hydraulic efficiency
Net head

14
13

-0.6

17

-0.2

-0.4

Figure 9. Modified shroud geometry: a) Ls = 19

mm; b) Ls = 50 mm.

-0.8

Figure 8. Impeller performance curves.

The net fluid head and the impeller power
curves, plotted in Figs. 8a and 8b, respectively,
exhibit a smooth pattern and a reasonable variation,
typical for radial flow impellers, verifying thus the
reliability of the PFC grid technique. As expected,
an increase of the volume flow rate above a safety
limit deteriorates the suction characteristics of the
impeller. The minimum pressure values in Fig. 8b
are drastically reduced for Q > 50 m3/h, indicating
that intense cavitation may occur in the impeller.

20

30

40

Shroud geometry, Ls (mm)

85

50

Efficiency, h (%)

16

Efficiency, h (%)

87

Pmin (bar)

Head, H (m)

18

After the evaluation stage the numerical model is

applied to study and assess the effect of various
design parameters of the impeller on its
hydrodynamic performance and efficiency. The
blade inlet width, b1, is considered at first, which
can be modified by changing the value of the shroud
height Ls (Fig. 2). The geometry is progressively
varied between the two extremes shown in Fig. 9,
and the corresponding results for the net head and
the hydraulic efficiency are plotted in Fig. 10.

90

(a)

10

ISSN: 1790-5087

3.3 Parametric studies

Head, H (m)

20

Issue 7, Vol. 1, July 2006

82

Figure 10. Effect of shroud height modification.

Maximum efficiency is achieved for a higher Ls
value (35 mm compared to the standard value of 25
mm), namely for a greater blade inlet width. It
seems that this configuration results in a smoother
turning of the flow and it is associated with smaller
hydraulic losses at the entrance to the blade region.
If the inlet width is reduced quite below that
optimum value then the inlet section cannot any
longer guide properly the flow and the efficiency
drops (Fig. 10). On the other hand, in the case of a
very large inlet width the efficiency decreases again,
due mainly to the increased friction losses along the
wider blade surface (Fig. 9b). The net head in Fig.
10 increases continuously with the blade width,
since the acting blade surface becomes greater, and
hence the energy transfer rate increases.
The performance curves for the net head and the
hydraulic efficiency of the optimum inlet width

95

18

91

16

87

10

20

30

Flow rate, Q

79

40

50

60

70

75

16

In a second study the radius of the rounded

corner of the shroud (Rs, Fig. 2a) is varied from zero
(Fig. 12a) to a maximum value of 20 mm (Fig. 12b),
and the results obtained within that range are drawn
in Fig. 12c. As expected, the sharp corner is
associated with increased hydraulic losses that cause
a reduction in efficiency by about 3 percentage units
(Fig. 12c). The losses are quickly diminished with
the rounding and they are almost minimized above
Rs = 10 mm, showing that the standard impeller
design has a well-rounded shroud shape. Also, the
net head curve acquires its maximum values above
the same rounding radius.
The blade inlet diameter D1 (Fig. 2) is modified
next between a minimum and a maximum value, as
shown in Fig. 13. The blade inlet angle, 1, is also
properly varied in order to always retain similar
relative inflow conditions. As can be observed in
Fig. 13c, the standard design impeller operates at the
maximum efficiency region. However, the
efficiency becomes lower for greater D1 values
because the hydraulic losses increase at the blade
leading edge, where the blade peripheral velocity
and the relative fluid velocity acquire higher values.
For the same reason, the suction efficiency of the
impeller becomes worse and the minimum pressure
exhibits quite lower values (Fig. 13).
Finally, the hub shape is changed by regulating
its height and width (Lh and Dh, Fig. 2) between the

90

standard design

(c)

(m3/h)

Figure 11. Comparison between standard and

optimal impeller performance.

(b)

(a)

Head, H (m)

12
10

83

Head, optimum
Efficiency, optimum
Head, standard
Efficiency, standard

two extremes of Fig. 14, and using the impeller with

the previously found optimum shroud shape. The
particular hub design shows only a slight effect on
the hydraulic efficiency, less than one percentage
unit, throughout the entire variation range. It seems
that its role in guiding the flow is not so important,
and its contribution to the friction losses is small
and not strongly depended on its shape. However, a
more projected hub (Fig. 14a) causes greater
blockage and results in higher inlet flow velocities,
thus reducing the minimum pressure developed at
the blade inlet (Fig. 14c).

15.8

89

15.6

88

15.4

87

Hydraulic efficiency
Net head

15.2
15

12

16

86

20

Shroud rounded inlet, Rs (mm)

85

Figure 12. Shroud geometry modifications: a) Rs =

0; b) Rs = 20 mm; c) Performance curves.

(a)

(b)

-0.1

90

(c)
-0.2

Pmin (bar)

14

769

ISSN: 1790-5087

Efficiency, h (%)

20

Efficiency, h (%)

Head, H (m)

impeller were then computed and they are compared

with the standard design ones in Fig. 11. This graph
reveals that a remarkable enhancement of the
hydraulic efficiency can be achieved, and moreover
this gain is observable at all loading conditions. The
net head with the optimum shroud is also increased
at higher loads, whereas the minimum pressure
values do not drop below 0,4 bar, even for the
greatest flow rate tested (comp. Fig. 8b).

Issue 7, Vol. 1, July 2006

88

standard
design

-0.3

86

-0.4

84

Hydraulic efficiency
Minimum pressure

-0.5
-0.6

60

70

80

Blade inlet diameter, D1 (mm)

82

90

Efficiency, h (%)

WSEAS TRANSACTIONS ON FLUID MECHANICS

80

Figure 13. Blade inlet diameter variation: a) D1 =

62 mm; b) D1 = 94 mm; c) Performance curves.

WSEAS TRANSACTIONS ON FLUID MECHANICS

(a)

Pmin (bar)

-0.1

(b)

93

standard design

(c)

-0.2

92

-0.3

91

-0.4

90

Hydraulic efficiency
Minimum pressure

-0.5
-0.6

10

20

Hub height, Lh (mm)

89

30

40

Efficiency, h (%)

770

88

Figure 14. Modified hub geometry: a) Lh = 40 mm,

Dh = 60 mm; b) Lh = 0; c) Performance curves.

4 Conclusions
A numerical model is developed for the numerical
solution of the RANS equations in the impeller of a
centrifugal pump, and it is applied for direct flow
analysis and parametric investigation of the effect of
some impeller design details on its hydrodynamic
characteristics.
The use of properly defined geometric
parameters to describe the impeller shape, along
with the automated grid generation process
incorporated in the developed model, constitute
effective tools in order to be used for inverse design
and performance optimization in turbomachines.

Acknowledgement:
The project is co-funded by the European Social
Fund (75%) and National Resources (25%)
Operational Program for Educational and
Vocational Training II and particularly the Program
Pythagoras.

References:
[1] C. Hornsby, CFD Driving pump design
forward, World Pumps, Aug. 2002, pp. 18-22.
[2] S. Cao, G. Peng, and Z. Yu, Hydrodynamic
design of rotodynamic pump impeller for
multiphase pumping by combined approach of

Issue 7, Vol. 1, July 2006

ISSN: 1790-5087

inverse design and CFD analysis, ASME

Transactions, Journal of Fluids Engineering,
Vol.127, 2005, pp. 330-338.
[3] F.A. Muggli and P. Holbein, CFD calculation of
a mixed flow pump characteristic from shutoff
to maximum flow, ASME Transactions,
Journal of Fluids Engineering, Vol.124, 2002,
pp. 798-802.
[4] M. Asuaje, F. Bakir, S. Kouidri, and R. Rey,
Inverse design method for centrifugal impellers
and comparison with numerical simulation
tools, International Journal for Computational
Fluid Dynamics, Vol.18, no.2, 2004, pp. 101110.
[5] M. Zhang, and H. Tsukamoto, Unsteady
Hydrodynamic Forces due to Rotor-Stator
Interaction on a Diffuser Pump with Identical
Number of Vanes on the Impeller and Diffuser,
ASME Transactions, Journal of Fluids
Engineering, Vol.127, 2005, pp. 743-751.
[6] J. Spann, and S. Horgan, The evolution of pump
design simulation, World Pumps, September
2006, pp. 32-35.
[7] T. Ye, R. Mittal, H.S. Udaykumar, and W. Shyy,
An accurate Cartesian grid method for viscous
incompressible flows with complex immersed
boundaries, Journal of Computational Physics,
Vol.156, 1999, pp. 209-240.
[8] E.A. Fadlun, R. Verzicco, P. Orlandi, and J.
Mohd-Yusof, Combined immersed-boundary
finite-difference methods for three-dimensional
complex flow simulations, J. of Computational
Physics, Vol.161, 2000, pp. 35-60.
[9] B.E. Launder, and D.B. Spalding, The numerical
computation of turbulent flows, Computer
Methods in Applied. Mechanics and
Engineering, Vol.3, 1974, pp. 269-289.
[10] C. Rhie, and W. Chow, Numerical study of the
turbulent flow past an airfoil with trailing edge
separation, A.I.A.A. Journal, Vol.21, 1983, pp.
1525-1532.
[11] T.J. Chung, Computational Fluid Dynamics,
Cambridge University Press, UK, 2002.
[12] J. Anagnostopoulos, G. Bergeles, B. Epple, and
P. Stegelitz, Simulation of grinding and drying
performance of a fluid-energy lignite mill,
ASME Transactions, Journal of Fluids
Engineering, Vol.123, 2001, pp. 303-310.
[13] J. Anagnostopoulos and D. Mathioulakis, A
flow study around a time-dependent 3-D
asymmetric constriction, Journal of Fluids and
Structures, Vol.19, 2004, pp. 49-62.