Fluid Dynamics Research 28 (2001) 267–280

Experimental and theoretical study of the ow in the volute

Abstract
The ow in the volute of a low speci#c-speed pump was studied both experimentally and numerically near its design
point. Measurements included time-averaged values of velocity and static pressure at a large number of locations in the
volute. The numerical computations were based on the unsteady three-dimensional potential ow model for the core ow.
Viscous losses were quanti#ed using additional models that use the potential ow as input. It is shown that near the
design point of this pump, the core ow behaves like a potential ow, provided that no boundary layer separation occurs.
Explanations are given for the presence of local deviations due to secondary ow. These local deviations do not in uence
the overall potential ow characteristics signi#cantly.
c 2001 Published by The Japan Society of Fluid Mechanics and
Elsevier Science B.V. All rights reserved.

Keywords: Centrifugal pump; Volute; Velocity measurement; Potential ow

1. Introduction

In order to improve the design of centrifugal pumps and compressors, a better understanding of
the ow of such machines is required. This paper deals with an experimental and theoretical study of
the ow in the volute of a low speci#c-speed centrifugal pump. Extensive measurements have been
performed of both velocities and pressures. The unsteady ow in the impeller–volute combination is
also computed, using the potential ow model. Additional models are used to quantify viscous losses.
For pumps operating near their design point the in uence of viscosity is restricted to thin boundary
layers and wakes, provided no massive boundary layer separation occurs. The core of the ow
∗
Corresponding author. Tel.: +31-53-489-2528; fax: +31-53-489-3695.
E-mail address: [Link]@[Link] (N.P. Kruyt).
1
Current address. Faculteit Scheikundige Technologie, Technische Universiteit Eindhoven, P.O. Box 513, 5600 MB
Eindhoven, Netherlands.
2
Current address. Flowserve Corporation, P.O. Box 55, 7550 AB Hengelo, Netherlands.
3
Current address. Faculteit Werktuigbouwkunde, Technische Universiteit Eindhoven, P.O. Box 513, 5600 MB Eindhoven,
Netherlands.

0169-5983/01/$20.00
c 2001 Published by The Japan Society of Fluid Mechanics and Elsevier Science B.V.
All rights reserved.
PII: S 0 1 6 9 - 5 9 8 3 ( 0 0 ) 0 0 0 3 2 - 0
268 J.D.H. Kelder et al. / Fluid Dynamics Research 28 (2001) 267–280

can then be predicted fairly accurately by means of three-dimensional inviscid methods. If one
further assumes the incoming ow to be irrotational, the core of the ow can be modelled as an
incompressible potential ow.
Contrary to methods based on the potential ow model, current methods based on the Navier–
Stokes equations with turbulence models (for example, Dawes, 1995; Croba and Kueny, 1996)
are of limited suitability as part of a design tool given their extreme requirements in terms of
computer resources. In addition, other open problems in such computations are (see Guelich, 1999):
(i) accurate modelling of boundary layer transition and separation (Casey et al., 1995); (ii) the
choice of an appropriate turbulence model that accounts for the eGects of three-dimensional boundary
layers, curvature and rotation (Speziale, 1985; Rodi, 1986; Speziale, 1989; Lakshminarayana, 1991;
Schilling, 1994); (iii) interaction between impeller and diGusor (Guelich and Egger, 1992); and
(iv) in uence of mesh size near solid walls on predictions (Guelich et al., 1997). Therefore the
potential ow model with viscous corrections is suitable for conditions near the design point, while
the turbulent viscous models are suitable for a somewhat wider range of ow conditions, although at
a much higher cost. Examples of the capabilities of the potential ow model with viscous corrections
as a design tool for complex three-dimensional impellers are given by Dijkers et al. (2000).
Previous studies on the ow in volutes dealing with the eGect of volute geometry on overall
characteristics (head and eJciency versus capacity curves) were performed by Bowerman and Acosta
(1957), RKutschi (1961), Worster (1963) and Decker (1990).
Measurements of velocities in a volute were presented by Sideris and van den Braembussche
(1987), Paone et al. (1989), Miner et al. (1989) and Flack et al. (1992). The pressure distribution
around the impeller is of importance for the radial forces acting on the impeller (Iversen et al., 1960;
Lorett and Gopalakrishnan, 1986; Sideris and van den Braembussche, 1987; Ojeda et al., 1992).
Design methods for volutes are usually based either on the assumption of constant average velocity
(StepanoG, 1948) or on the assumption of constant angular momentum (P eiderer, 1949).
From these design methods and from experiments it follows (for example Worster, 1963) that at
a single ow rate the pressure along the periphery of the impeller is approximately constant. For
lower ow rates the pressure increases from volute inlet (tongue) to volute outlet: the volute acts
like a diGuser. For higher ow rates the pressure decreases from volute inlet to volute outlet: the
volute acts like a converging nozzle.
The outline of this paper is as follows. Firstly, the test set-up is described, followed by an outline
of the computational method. Then the models are described that are used to account for losses
due to viscous eGects. Finally, the results of both measurements and computations are given and
discussed.

2. Test set-up

A detailed description of the test-rig that was built to study the two-dimensional ow in various
types of centrifugal impellers is given by Visser et al. (1999). In order to study the interaction
between the impeller and the volute of a pump, this test-rig has been adapted to include a volute. In
the study by Visser et al. (1999) of the ow in an impeller channel, the laser Doppler velocimetry
(LDV) measurement system rotated with the impeller, while in the current study of the ow in the
volute, the LDV system is stationary (non-rotating).
 J.D.H. Kelder et al. / Fluid Dynamics Research 28 (2001) 267–280 269

Fig. 1. Geometry of centrifugal impeller (cross-section) and volute (plane view and cross-section). All dimensions in mm.

Measurements were performed with a centrifugal pump (see Fig. 1) having a low speci#c-speed
◦
n! = (Q1=2 )=(gH )3=4 of 0.4. The impeller has seven blades with a constant blade angle of 70 with
respect to the radial direction and a constant blade thickness of 2 mm. The impeller blade inner
diameter is 320 mm, its outer diameter 640 mm, and the axial width is 25 mm. The volute has a
trapezoidal cross-section and is designed to approximately match the impeller at design condition
(design ow rate Qd = 0:008 m3 =s, angular velocity  = 4:2 rad=s), according to the method of
constant angular momentum (P eiderer, 1949). The Reynolds number (D2 )=
is 1:7 × 106 , where

is the kinematic viscosity of water. The volute tongue has a cylindrical shape with a diameter of 2
mm. During construction special attention was paid to the minimization of leakage ows. The pump
was operated at ow rates of 82.5 %, 100 % and 117.5% of the design ow rate.
Velocity measurements were performed using LDV. The LDV con#guration is described in detail
by Visser et al. (1999). It employs a dual reference beam forward scattering system, capable of
parallel detection of two perpendicular velocity components. Two Bragg cells were used to eGectuate
preshifts between main beam and the two reference beams, thus enabling the determination of the
direction of the velocity components. Two detectors measured the Doppler frequency. These signals
were sampled and stored on disc. Time-averages and RMS-values could be computed. Information
on the axial velocity component could not be obtained. U-tube manometers were used to obtain
values of the static pressure. Fig. 2 shows the locations in the volute where velocity and static
pressure measurements were obtained. At the locations just outside the impeller, the pressure was
measured at the shroud side, while at the volute outer wall it was measured at midheight.
The head-capacity curve was derived from the static pressure diGerence between inlet and outlet
of the pump and the assumption of uniform velocity in these regions. It is not possible to measure
the hydraulic eJciency of the pump with the current experimental set-up. Air-bubble visualisation
was used to investigate the ow near the tongue of the volute.
270 J.D.H. Kelder et al. / Fluid Dynamics Research 28 (2001) 267–280

Fig. 2. Measurement locations in the laboratory centrifugal pump. LDV measurements are performed along traverses A
to H at the volute centre line. Static pressure measurement locations are indicated with solid markers. The hatched area
shows the region which is visually inaccessible.

3. Computational method

In general centrifugal pumps will show unsteady ow behaviour, especially at oG-design conditions,
as a result of the interaction between impeller and volute. A proper time-dependent matching of the
respective computational domains is therefore necessary, which is one of the main objectives of the
current method.

3.1. Potential 5ow model

The Reynolds and Mach numbers for the ow in hydraulic turbomachines are usually such that
the core ow can be considered as inviscid and incompressible. Assuming the suction ow to be
irrotational, the problem can be formulated in terms of the velocity potential . The velocity vector
v can then be written as
C = ∇ (1)
and the continuity equation reduces to the Laplace equation
∇2  = 0: (2)
The pressure can be computed from the unsteady Bernoulli equation
@ 1 p
+ C · C + + gz = c(t); (3)
@t 2 
where t denotes the time, p the static pressure,  the density, g the gravitational acceleration, z the
height and c a constant which only depends on time t.

3.2. Computational domain and boundary conditions

In the multi-block approach adopted here, the ow region of interest is divided into subdomains,
all having a topologically cubic shape. The subdomains are non-overlapping, with nodal coincidence
at the interfaces.
 J.D.H. Kelder et al. / Fluid Dynamics Research 28 (2001) 267–280 271

Blocks in the impeller and the volute are separated by a cylindrical or conical surface. At the
interface, the values of the potential  at corresponding nodal points from impeller and volute are
matched. Note that this correspondence of nodal points is time-dependent. In this way the rotating
motion of the impeller relative to the volute is properly simulated without having to create a new
mesh for each timestep, as was done by Miner et al. (1992).
Appropriate uniform Neumann boundary conditions are applied at the inlet, outlet and blade sur-
faces. At block boundaries formed by the wake surfaces emanating from trailing edges, vortices are
shed downstream. These vortex sheets are a result of both non-uniform blade loading (variations
of the circulation along the blade’s span) and time-dependent variations of the blade circulations.
Vortex sheets are modelled by imposing discontinuities of the potential value across the wake surface
+ (r; s; t) = − (r; s; t) + (r; s; t); (4)
where the two sides of the wake are denoted by the ‘+’ and ‘−’, and r and s are local coordinates
in the wake. By allowing the potential jump  to vary along the vortex sheet, a discontinuity of the
tangential velocity across the wake is introduced. The value of the potential jump and its variation
is computed by determining the amount of circulation around the blade(s) that is needed to satisfy
Kutta’s condition of smooth ow from the trailing edges. In steady computations, the potential jumps
will be constant along streamlines, whereas in unsteady computations the time-dependent variation
of the blade circulation is convected downstream with the mean velocity relative to the rotating
impeller.

3.3. Numerical method

The numerical method is based on a fully three-dimensional #nite element method. Some special
techniques are employed in order to reduce computing time and memory requirements. These are
based on a superelement approach (Zienkiewicz and Taylor, 1989) in conjunction with implicitly
solved values of blade circulations. A fully three-dimensional method was used to account for the
three-dimensional parts of the geometry: the inlet and the volute region where the volute width varies
(see Fig. 1).
The superelement approach greatly facilitates the time-dependent coupling of the rotating impeller
to the stationary volute, which is essential in a correct description of the interaction between impeller
and volute. In the superelement approach internal degrees of freedom (DOFs) are expressed in terms
of DOFs that are located at the boundary of the superelements. The time-invariant nature of this
elimination process implies that unsteady computations can be performed with far fewer DOFs.
Furthermore, this elimination process is identical for geometrically similar superelements, such as
the impeller channels.
By considering the unknown circulations around the blades as additional DOFs in the #nite element
solution process, the Kutta conditions can be imposed implicitly. This avoids the need to compute a
large number of subpotentials, which is especially important in three-dimensional computations since
the number of unknown circulation values is much larger.
A detailed description of the numerical method is given in Kruyt et al. (1999). Here its main
advantages are summarized:
• Impeller symmetry is fully exploited.
• Superelement matrices have to be computed only once for unsteady computations.
272 J.D.H. Kelder et al. / Fluid Dynamics Research 28 (2001) 267–280

• Kutta conditions at the trailing edges are imposed implicitly.

4. Viscous loss models

Inviscid methods will always over-predict a pump’s head, while being unable to predict its eJ-
ciency. The obvious reason is that viscous losses are not taken into account. A very eJcient way
to incorporate the eGects of viscosity is to use an inviscid method along with additional models to
estimate the viscous losses. Many of these (well-established) models exist in the literature (see, for
example, Denton, 1993), for which most of the input can be supplied by potential ow analysis.
The models are adopted here without adjusting the empirical coeJcients.
Pump power losses are commonly classi#ed as:
• Hydraulic losses.
• Volumetric losses.
• Disc friction losses.
• Mechanical losses.
In the absence of leakage ows, the head is numerically determined by
M − PL; h
H= ; (5)
gQ
where  is the angular velocity, Q is the ow rate, PL; h is the hydraulic power loss, and M is
the moment exerted by the impeller blades on the uid. The latter can be computed by an inviscid
numerical method. External losses due to disc friction and mechanical losses need not be considered
here. All quantities are time-averaged. Existing models to quantify the above-mentioned losses will
now be presented.
Hydraulic power losses for this con#guration are composed of dissipative losses in the boundary
layers PL; diss along walls in both the impeller and the pump casing, and mixing losses PL; mix in wakes
PL; h = PL; diss + PL; mix : (6)
The dissipation power loss in attached boundary layers can be quanti#ed using a fairly simple method
based on dissipation coeJcients (Schlichting, 1979). The dissipation loss is written as


1
PL; diss = 
2
cD w3 dS; (7)
S

where S denotes the wall surface, cD is the dissipation coeJcient and w is the velocity of an inviscid
ow tangential to the wall. According to Denton (1993) the dissipation coeJcient for turbulent
boundary layers is relatively insensitive to the detailed state of the boundary layer (accelerating or
decelerating). Denton suggests an average value of 0.0038 for turbulent boundary layers with Re
(Reynolds number based on boundary layer momentum thickness) of order 1000, but it should be
stressed that actual values vary between 0.002 for accelerating ows and 0.005 for ows encountering
an adverse pressure gradient. As a #rst estimate boundary layers, both in the impeller and in the
volute, are assumed to be turbulent, with cD = 0:0038.
 J.D.H. Kelder et al. / Fluid Dynamics Research 28 (2001) 267–280 273

Fig. 3. Plane view of the computational mesh for the centrifugal pump with spiral volute. Only one of the seven impeller
channels is shown.

Mixing losses occur in wakes behind blunt trailing edges of impeller (or stator) blades. According
to Denton (1993) the loss can be quanti#ed using
  2 
1 2 Cpb b 2 ∗ + b
PL; mix = wref Q − + + ; (8)
2 W W W
with b the blade thickness and W the blade pitch. The sum of the momentum thicknesses at pressure
and suction side, at a station near the trailing edge, is given by . The same applies to ∗ , the sum
of the displacement thicknesses. The base pressure coeJcient Cpb is de#ned by
pb − pref
Cpb = 2
; (9)
1=2wref
with pb the static pressure just behind the trailing edge and pref and wref the pressure and velocity
at a reference position just before the trailing edge. In general, the value of Cpb will be negative,
with typical values close to −0:15.
The boundary layer displacement and momentum thickness at the blade trailing edge are computed
by a one-dimensional integral boundary layer method. The method of Thwaites (1949) is used for
the laminar part of the boundary layer, while Green’s “lag entrainment method” (Green et al., 1972)
is used for the turbulent part. Michel’s criterion (Michel, 1952) is used to determine the location
of transition. Thwaites’ criterion is implemented to predict laminar separation. The boundary layer
changes to a turbulent state either after laminar separation or transition. Turbulent separation is
assumed to occur when the shape factor (ratio of displacement thickness and momentum thickness)
exceeds the value of 2.8.
Of both hydraulic losses mentioned in this section, boundary layer dissipation is the most important
loss mechanism.

5. Results

In this section the results of measurements and computations are compared. These results deal
with velocities, pressures and overall characteristics. The computational mesh is shown in Fig. 3.
274 J.D.H. Kelder et al. / Fluid Dynamics Research 28 (2001) 267–280

It contains a total of 168,000 nodes, of which 14,500 are located in each of the seven impeller
channels. Computations are performed with 105 timesteps per shaft revolution.

5.1. Velocity in the volute

Radial and circumferential velocity components were measured in a plane at the centre of the
volute. The traverses A–H (see Fig. 2) were scanned for the low ow rate Q = 0:825Qd , the design
ow rate Qd and the high ow rate Q = 1:1175Qd .
Typical results are shown in Fig. 4, together with the time-averaged computed values. The full
set of measurements and computations is presented in Kelder (1996). In a large region of the volute
◦
(60–285 from the tongue) computed circumferential velocities agree very well with measurements
at design ow conditions. For the low ow rate this excellent agreement is restricted to a smaller
◦
region (150 –285 from the tongue). For high ow rate the agreement is not very good except for
◦
a small region 60 –150 from the tongue. The agreement between computed and measured radial
velocities is poor.
This discrepancy was further investigated by performing traverses over the width of the volute.
Results are given in Fig. 5 which shows the variation of normalised radial and circumferential
velocity over the width of the volute for two radial positions on cross-section F (see Fig. 2). The
circumferential velocity is practically constant over the width of the volute, while the (much smaller)
radial velocity shows severe secondary ow. Note that the circumferential velocity determines the
energy transfer by the impeller blades to the uid. The computed radial and circumferential velocities
were practically constant over the width of the volute.

5.2. Static pressure in the volute

The static pressure diGerence Sp between the inlet of the pump and locations in the volute is
measured using U-tube manometers. It is made non-dimensional with the blade-tip speed according
to
pvol − pinlet
Sp = : (10)
(rTE )2

In Fig. 6 results of measurements and computations are shown for locations just outside the impeller
and along the volute outer wall, for three diGerent ow rates. The inviscid computations lead to
pressure values which, on average, are too high. However, in a large region, not too near to the
tongue, the qualitative agreement is quite good. A constant static pressure around the impeller can
be observed at design ow rate.
The computed static pressure values can be corrected for viscous losses in the impeller

Spcorr = Spinvisc − gSHL; h; imp ; (11)

where the hydraulic head loss in the impeller is denoted by SHL; h; imp . By doing so, the agreement
is improved, although deviations still occur at oG-design conditions.
 J.D.H. Kelder et al. / Fluid Dynamics Research 28 (2001) 267–280 275

Fig. 4. Non-dimensional radial and circumferential velocity in the volute along diGerent traverses for three ow rates.
Comparison between measurements (symbols) and computations (solid line). The scaled local coordinate along the traverse
is denoted by s, ranging from 0 at the impeller outer radius to 1 at the volute wall. Velocities are scaled with blade tip
speed.
276 J.D.H. Kelder et al. / Fluid Dynamics Research 28 (2001) 267–280

Fig. 5. Variation of radial and circumferential velocities with axial position for two radial positions on traverse F (blade
tip radius denoted by rTE ). Velocities are scaled with blade tip speed.

Fig. 6. Static pressure diGerence in the volute of the laboratory centrifugal pump as a function of orientation  (degrees),
for three ow rates. See Fig. 2 for locations at the impeller periphery and the volute wall. Pressure diGerence Sp is
de#ned in Eqs. (10) and (11).
 J.D.H. Kelder et al. / Fluid Dynamics Research 28 (2001) 267–280 277

Fig. 7. Head-capacity curve for the centrifugal pump, showing measured and computed values, both inviscid and corrected
for viscous losses.

Fig. 8. Photograph of the ow near the tongue of the volute at Q = 1:1175Qd showing boundary layer separation. Air
bubbles were inserted through the manometer for ow visualisation.

5.3. Head curve

The methods described in Section 4 are used to quantify the eGects of boundary layer dissipation
and wake mixing. The resulting head-capacity curve is shown in Fig. 7. The fraction of the total
head loss caused by wake mixing ranges from 10% at high mass ow to 25% at low mass ow.

5.4. Air bubble visualisation

The ow around the volute tongue was visualised using air bubbles that were injected into the
ow (see also Fig. 8).
For the low ow rate Q = 0:825Qd , the pressure around the impeller increases from tongue
towards the outlet (see Fig. 6). The stagnation point near the tongue will be located towards the
throat. Bubbles injected into the ow will therefore pass the tongue on the impeller side, consistent
with the visualisations.
For the design ow rate Qd , the stagnation point is (on average) located on the volute tongue, and
due to the time-dependent nature of the ow, half of the bubbles pass the tongue on the impeller
side and half pass on the outlet side.
278 J.D.H. Kelder et al. / Fluid Dynamics Research 28 (2001) 267–280

Fig. 9. Cross-section of the volute, showing measured radial velocity pro#les. Average velocities are indicated by dotted
lines.

For the high ow rate Q = 1:1175Qd , a recirculation zone is observed in the volute throat down-
stream of the volute tongue, indicating boundary layer separation. This phenomenon of boundary
layer separation near the tongue for large ow rates was also noted by Wo and Bons (1993).

6. Discussion

In this section explanations are sought for the diGerences between measurements and computations.
An investigation of the axial distribution of radial velocities at a number of radial positions in the
impeller and the volute reveals that a region of severe secondary ow is located in the volute.
Typical radial velocity pro#les are sketched in Fig. 9. The observed convex radial velocity pro#le in
the volute region, with negative velocities near the upper and lower surfaces, can be explained by
an analysis of pressure forces and centrifugal forces (due to curvature) in the boundary layers and
the main ow. It is equivalent to the secondary ow encountered in the ow through a pipe bend. A
similar analysis for the ow through the impeller can be made (see, for example, Lakshminarayana,
1996), where the additional centrifugal force due to rotation and the Coriolis force have to be taken
into account. The equilibrium between pressure forces and Coriolis forces in the main ow is lost
in the boundary layers at hub and shroud surfaces, leading to a secondary ow in the boundary
layers directed from pressure to suction side. In the main ow a reverse secondary ow direction
is observed. This leads to the observed concave radial velocity pro#le for impellers with backward
curved blades.
The secondary ow does not seem to in uence the static pressure distribution. Except for the high
ow rate, the agreement between measurements and potential ow computations (after correcting for
viscous losses in the impeller) is quite good. It can be seen from the diGerence between both that
viscous losses build up as the uid is owing along the volute wall from the tongue to the volute
throat.
Similar results are obtained for the head-capacity curve. The good agreement at low and design
ow rate imply that other sources of viscous losses are not very important in this pump. At high- ow
rate, however, a larger deviation is observed between computations and experiments.
A possible cause for the disagreement between measurements and computations at the high- ow
rate is boundary layer separation at the exhaust pipe, downstream of the tongue, that was observed
by the air-bubble ow visualisation.
The blockage eGect resulting from the separated boundary layer could well be related to the
observed circumferential velocity pro#les at high ow rate, which deviate considerably from the
 J.D.H. Kelder et al. / Fluid Dynamics Research 28 (2001) 267–280 279

computations. A homogeneous distribution of tangential velocity was seen to cover the major part
of the volute extending from traverses B to G (Fig. 2). It is suggested by the authors that measured
velocity pro#les like these have inspired StepanoG (1957) to put forward his method of constant
mean velocity for constructing volutes. A similar ow #eld can emerge at design ow rate if the
volute is designed somewhat too small. It is known that volutes designed according to the method
of StepanoG are smaller than those designed according to the method of constant angular momentum
(P eiderer, 1949).

References

Bowerman, R.D., Acosta, A.C., 1957. EGect of the volute on performance of a centrifugal pump impeller. Trans. ASME
79, 1057–1069.
Casey, M.V., Eisele, K., Muggli, F.A., Guelich, J., Schachenmann, A., 1995. Flow analysis in a pump diGuser. Part 2:
validation of a CFD code for steady ow. ASME FED 227, pp. 135 –143.
Croba, D., Kueny, J.L., 1996. Numerical calculation of two-dimensional, unsteady ow in centrifugal pumps: impeller and
volute interaction. Int. J. Numer. Methods Fluids 22, 467–482.
Dawes, W.N., 1995. A simulation of the unsteady interaction of a centrifugal impeller with its vaned diGuser: ow analysis.
J. Turbomach. 117, 213–222.
Decker, H., 1990. Investigation of the performance of model pumps with semi-open impellers. Ph.D. Thesis, University
of Karlsruhe, Germany (in German).
Denton, J.D., 1993. Loss mechanisms in turbomachines. J. Turbomach. 115, 621 – 656.
Dijkers, R.J.H., Visser, F.C., op de Woerd, J.G.H., 2000. Redesign of a high-energy centrifugal pump #rst-stage impeller.
IAHR Conference 2000, Charlotte, NC, USA, Paper PD-01.
Flack, R.D., Miner, S.M., Beaudoin, R.J., 1992. Turbulence measurements in a centrifugal pump with a synchronously
orbiting impeller. J. Turbomach. 114, 350 –359.
Green, J.E., Weeks, D.J., Brooman, J.W.F., 1972. Prediction of turbulent boundary layers and wakes in compressible ow
by a lag-entrainment method. RAE Technical Report 72231.
Guelich, J.F., 1999. Centrifugal Pumps — A Handbook for Development, Design and Operation. Springer, Berlin, Germany
(in German).
Guelich, J.F., Egger, R., 1992. Part load ow and hydraulic stability of centrifugal pumps. EPRI Report TR-100219.
Guelich, J.F., Favre, J.N., Denus, K., 1997. An assessment of pump impeller performance predictions by 3D-Navier–Stokes
calculations. ASME FED-SM97, p. 3341.
Iversen, H., Rolling, R.E., Carlson, J.J., 1960. Volute pressure distribution, radial force on the impeller and volute mixing
losses of a radial ow impeller. J. Eng. Power 82, 136 –144.
Kelder, J.D.H., 1996. Measurements and computations of the ow in a quasi two-dimensional model pump. [Link]. Thesis,
Department of Mechanical Engineering, University of Twente, The Netherlands (in Dutch).
Kruyt, N.P., Esch, B.P.M., van Jonker, J.B., 1999. A superelement-based method for computing unsteady three-dimensional
potential ows in hydraulic turbomachines. Comm. Numer. Methods Eng. 15, 381–397.
Lakshminarayana, B., 1991. An assessment of computational uid dynamic techniques in the analysis of turbomachinery.
J. Fluids Eng. 113, 315 –352.
Lakshminarayana, B., 1996. Fluid Dynamics and Heat Transfer of Turbomachinery. Wiley, New York.
Lorett, J.A., Gopalakrishnan, S., 1986. Interaction between impeller and volute of pumps at oG-design conditions. J. Fluid
Eng. 108, 12–18.
Michel, R., 1952. Study of transition on wing sections; establishment of a criterion for the determination of point of
transition and calculation of the wake of an incompressible pro#le. ONERA Report 1=1578A (in French).
Miner, S.M., Beaudoin, R.J., Flack, R.D., 1989. Laser velocimeter measurements in a centrifugal ow pump. J. Turbomach.
111, 205–212.
Miner, S.M., Flack, R.D., Allaire, P.E., 1992. Two-dimensional ow analysis of a laboratory centrifugal pump. J.
Turbomach. 114, 333–339.
280 J.D.H. Kelder et al. / Fluid Dynamics Research 28 (2001) 267–280

Ojeda, W., de Flack, R.D., Miner, S.M., 1992. Pressure distributions in a single and two versions of a double volute of
a centrifugal pump. Proceedings of International Gas Turbine and Aeroengine Congress and Exposition, 92-GT-20.
Paone, N., Riethmuller, M.L., Braembussche, R.A. van den, 1989. Experimental investigation of the ow in the vaneless
diGuser of a centrifugal pump by particle image displacement velocimetry. Exp. Fluids 7, 371–378.
P eiderer, C., 1949. Centrifugal Pumps for Liquids and Gases. Springer, Berlin, Germany (in German).
Rodi, W., 1986. Turbulence modelling for incompressible ows. Phys. Chem. Hydrodyn. 7, 297–324.
RKutschi, K., 1961. In uence of volutes on the performance of centrifugal pumps. Schweizerische Bauzeitung 79, 233–240
(in German).
Schlichting, H., 1979. Boundary Layer Theory. McGraw-Hill, New York.
Schilling, R., 1994. A critical review of numerical models predicting the ow through hydraulic machinery bladings.
Proceedings of the 17th IAHR Symposium, Beijing, GL2.
Sideris, M.T., van den Braembussche, R.A., 1987. In uence of a circumferential exit pressure distortion on the ow in
an impeller and diGuser. J. Turbomach. 109, 48–54.
Speziale, C.G., 1985. Modelling the pressure gradient-velocity correlation of turbulence. Phys. Fluids 28, 69–71.
Speziale, C.G., 1989. Turbulence modelling in noninertial frames of reference. Theoret. Comput. Fluid Dyn. 1, 3–19.
StepanoG, A.J., 1948. Centrifugal and Axial Flow Pumps: Theory, Design, and Application. Wiley, New York.
Thwaites, B., 1949. Approximate calculation of the laminar boundary layer. Aeronaut. Quart. 1, 245–280.
Visser, F.C., Brouwers, J.J.H., Jonker, J.B., 1999. Fluid ow in a rotating low speci#c-speed centrifugal impeller passage.
Fluid Dyn. Res. 24, 275–292.
Wo, A.M., Bons, J.P., 1993. Flow physics leading to system instability in a centrifugal pump. J. Turbomach. 116, 612–620.
Worster, R.C., 1963. The ow in volutes and its eGect on centrifugal pump performance. Proc. Inst. Mech. Engrs 177,
843–865.
Zienkiewicz, O.C., Taylor, R.L., 1989. The Finite Element Method. McGraw-Hill, Maidenhead, UK.