271284 (2012)
Received: 1 Sep. 2011; Revised: 4 Jan. 2012; Accepted: 16 Jan. 2012
271
METHODOLOGY FOR THE RESIDUAL AXIAL THRUST EVALUATION
IN MULTISTAGE CENTRIFUGAL PUMPS
Simone Salvadori *, Alessandro Marini and Francesco Martelli
Energy Engineering Department Sergio Stecco, University of Firenze, Via di S. Marta, 3
50139 Firenze, Italy
* EMail: simone.salvadori@unifi.it (Corresponding Author)
ABSTRACT: One of the most challenging aspects in horizontal pumps design is the evaluation of the residual
axial thrust acting on the rotating shaft. The thrust is affected by pump characteristics and working conditions.
Solving this problem is easier for a single stage pump than for multistage pumps, even in partially selfbalancing
opposite impeller configuration. The challenge is then to individuate a procedure that will provide the residual thrust
value with a moderate computational effort, dealing with the industrial requests of accuracy and reduced time
consumption. A procedure is proposed, which consists in the numerical simulation of each pump component. For
each component, the obtained massflow/thrust correlations are coupled by using a momentum balance equation
used to calculate the axial thrust as a function of the working conditions. The main topic in multistage pump
modeling is the leakage flows characterization by means of accurate numerical analysis. Therefore, the cavity flows
behavior is investigated and the flow structures individuated. The numerical investigation of the pumps components
provides also a thorough knowledge of fluid dynamic fields. The proposed procedure is able to predict both the
direction and the variation of the thrust in a selected range of flow rates, while the value of the thrust is affected by a
nonnegligible error generated by real machine effects.
Keywords: axial thrust, cavity flows, centrifugal pumps, CFD, multistage, momentum balance
1. INTRODUCTION
In a multistage horizontal centrifugal pump, the
residual axial thrust is balanced by bearings that
can guarantee the mechanical reliability (when
properly chosen). Over the pump operating range
the main contribution to the axial thrust is due to
the impellers flow fields, the leakage flows
through wear rings and the pressure distribution
that occurs inside the gaps between impeller
shrouds and pump stationary walls. A detailed
analysis of the origins of multistage pump
unbalance has been proposed by Gantar et al.
(2002). They underlined the effect of cavity flows
and offdesign conditions but did not propose a
procedure to couple the separated contributions
coming from the pump components.
The cavity flow behavior is a key parameter for
the pressure field evaluation and then for the
thrust calculation. From a physical point of view,
the flow filed inside the impeller side chambers
has been extensively described by the classical
contributions by Batchelor (1951) and Stewartson
(1953). Many contributions are also available
where the leakage flow has been analyzed to
provide correlations between the leakage mass
flow, the impeller head, the friction losses and the
geometrical parameters. Amongst them it is
worthwhile to remember the works of Denny
(1954), Traupel (1958), Worster and Thorne
(1959), Daily and Nece (1960), Utz (1972) and
more recently Tamm and Stoffel (2002) and
Gulich (2003a and b).
The main limit of these correlations is that they
refer to an ideal configuration, which is
somehow representative of cavity flows, but
neglect two fundamental aspects. The first one is
that the geometrical characteristics of side
chambers are usually decided by compactness and
reliability criteria instead of analyzing disk
friction losses. Then, their shape could be very
different from the one represented by the
correlations, especially considering the front
cavity. The second issue is related to the
boundary conditions: since cavity and mainflow
are coupled by clearances and there are nonlinear
effects to be considered. Baskharone and Wyman
(1999) and Adami et al. (2005) proposed methods
based on the numerical modeling to manage with
this kind of problem.
Thamsen and Bubelach (2011) suggested that the
main problem in residual axial thrust evaluation
was the nonuniformity of the cavity inlet
conditions generated by the impeller/diffuser
interaction. The experimental analysis of a single
stage pump demonstrated that the flow field and
Engineering Applications of Computational Fluid Mechanics Vol. 6, No. 2 (2012)
272
the pressure distributions were not axisymmetric.
This phenomenon is even more important when
radial gaps are non negligible. Furthermore, flow
recirculations occurring at off design conditions at
the impeller exit section are responsible for
increased nonuniformities.
To overcome all of these problems, the numerical
simulation of each pump component has been
individuated as a possible solution to the complex
problem of axial thrust evaluation. In fact,
Computational Fluid Dynamics (CFD) has been
demonstrated to be accurate enough to study real
cases in the turbo machinery field of interest.
Several examples are available in the recent
literature for compressors (Zachos et al., 2011),
combustion chambers (Marzouk and Huckaby,
2010), turbines (Zhou et al., 2011; Liu and Wang,
2011; Montomoli et al., 2011) and pumps
(Sifikhani et al., 2011; Koombua and Pidaparti,
2010).
Each contribution to the residual thrust intensity
is calculated taking into account the local pressure
distribution on the rotating walls. Concerning
cavity flows, all of them are analyzed considering
the actual geometry and working range with a
loosely coupling with the impeller/diffuser stages.
The latter are analyzed considering realistic inlet
conditions coming from the numerical analysis of
the upstream component. The obtained
correlations are matched together by means of a
momentum balance equation obtained by
studying a pump stage. Furthermore, an indepth
study of the fluid dynamic fields inside the single
pump components can help to obtain more
knowledge of the pump characteristics during the
design phase to avoid offdesign issues.
The proposed approach represents a possible
answer to the industrial requests of accuracy and
smartness in axial thrust evaluation. The known
limits of this method are a simplified approach to
the leakage flow evaluation and the loosely
coupling of some components. Nevertheless,
those approximations will not affect the accuracy
of the analysis while the method is applied to a
range of working conditions close to the Best
Efficiency Point (BEP) (15%) for centrifugal
pumps with small gaps at the impeller outlet
diameter.
2. THRUST EVALUATION IN
MULTISTAGE PUMPS
Present paper describes the development of a
numerical method for the residual axial thrust
calculation. The evaluation of the resultant of the
forces acting on each single stage is the starting
point to determine the thrust. With reference to
the schematic distribution of forces and control
volume shown in Fig. 1, this balance can be
expressed by Eq. 1:
F
fs
+
F
bs
+
F
inlet
+
F
mom
+
F
bus
=
T
ax
(1)
The balance is referred to a control volume
containing all the rotating walls. The term F
fs
is
the global force acting on the front shroud walls
while F
bs
is relative to the back shroud: both of
them are evaluated considering the static pressure
distribution along the rotating walls. F
inlet
is
generated by the pressure field at the impeller
inlet section and is a function of the local static
pressure value. F
mom
is the momentum
contribution along axial direction and is
calculated considering the mean value of inlet
velocity. The term F
bus
is the pressure integral on
the bushing walls. All these terms can be
evaluated considering pressure levels, pump
geometry and massflow conditions. For each
pump (at a chosen flow rate) the axial thrust can
be calculated once the pressure field inside its
components, including leakage cavities, is known.
Therefore the multistage pump axial load can be
obtained algebraically by adding the single stage
contributions, which have been calculated by
applying Eq. 2.
2
fs bs inlet inlet
A A
bus bus ax
inlet
p ndA p ndA p A n
Q
n p A n T
A
+ +
+ =
} }
(2)
In the pump shown in Fig. 2, the flow enters the
suction nozzle and passes through the two stages
of the first bench. Then a crossover device leads it
to the third stage inlet through an annular
chamber, thus being responsible for its turning of
direction. The fluid evolves in the three stages of
the second bench and reaches the discharge
nozzle. All the diffusers are characterized by the
same geometry except for the ones facing the
Fig. 1 Forces balance for a single stage (the dotted
lines indicate the control volume).
Engineering Applications of Computational Fluid Mechanics Vol. 6, No. 2 (2012)
273
Fig. 2 Multistage centrifugal pump (courtesy of
WEIRGabbioneta SRL).
Fig. 3 Impeller side chambers.
Fig. 4 Central balancing drum.
crossover. All the impellers are geometrically
identical but with different inlet conditions, due to
the presence of different upstream components.
The opposite impeller configuration helps to
balance the axial thrust, but does not guarantee it
also in pumps with an even impellers number
because of the effect of leakage flows and shroud
stationary walls gaps.
To properly choose the thrust bearings, the
residual axial load intensity and direction must be
evaluated taking into account the contribution of
the impeller shroud chambers. The leakage flows
and the presence of gaps modify the pump flow
rate (and, consequently, the total head) and affect
the pressure distribution on the rotating walls. In
such a complex machine, three kinds of cavities
are present: front and back impeller shroud
cavities and a central drum. Fig. 3 shows the main
and leakage flows in the front and back shroud
chambers. While the flow inside the front shroud
chamber always turns inward in the radial
direction, the one inside the back shroud cavity
turns according to the local pressure gradient.
Even though the massflow through the back
shroud chamber has a slight influence on the
whole pump performance, its contribution to the
axial thrust cannot be neglected. The second and
the fifth impeller back shroud cavities correspond
to the central drum ones in which the leakage
flow goes from the second to the first group of
impellers (Fig. 4). A lateral balancing drum is
located before the first impeller of the second
bench and is connected to the suction volute by
means of a balancing duct thus keeping the
stuffing box pressure very close to the suction
pressure.
3. CFD PROCEDURE DESCRIPTION
The full procedure is applied to the described
pump designed by WEIRGabbioneta SRL (Fig.
2). The evaluation of the components
performances requires the development of a
strategy to manage the following tasks:
The single stage analysis (both impeller and
diffuser hydraulic channels);
The simulation of stationary components
(volutes and crossover);
The study of the flow conditions inside the
front and back shroud impeller side chambers;
The study of the flow conditions inside the
central balancing drums;
The final data collection for the residual
hydraulic axial thrust calculation.
Several second order accurate 2D/3D steady
simulations solving Reynolds Averaged Navier
Stokes (RANS) equations have been carried out
using the ANSYS Fluent commercial code.
Hybrid meshes have been realized using the
commercial tool Centaur
by Centaur soft. A
second order accurate finitevolume pressure
correction procedure for incompressible flows has
been employed with a twoequation k
turbulence model with standard wall functions.
The choice of the turbulence model is coherent
with the near wall mesh quality, which ensures a
y
+
value higher than 30.The pump has been
divided into its main components, which have
been analyzed separately to reduce computational
Engineering Applications of Computational Fluid Mechanics Vol. 6, No. 2 (2012)
274
costs. The following elements or group of
elements have been simulated:
Suction volute;
First impeller;
Each diffuser not facing the crossover device
coupled with its downstream stage impeller;
Second diffuser together to crossover device
and annular chamber;
Third impeller;
Fifth diffuser and discharge volute;
Impeller side chambers (front and back
shrouds);
Central and lateral balancing drums.
The sequence of simulations has been chosen
considering each result as a boundary condition
for the following CFD analysis. For instance, the
impeller boundary conditions have been provided
by its upstream stationary component simulation
(either diffuser or annular chamber). Finally, all
the results have to be combined as indicated in Eq.
2, in order to obtain the axial thrust of each of the
two groups of stages, thus helping to understand
their reciprocal balancing effects.
To study the shroud chambers behavior, the
relation between the head across the impeller and
the leakage flows should be known. The impeller
shroud chambers have been analyzed with 2D
axisymmetric CFD simulations while a 3D
model has been used for the central balancing
drum. This choice is supported by the results
shown by Gantar et al. (2002) for geometries
similar to the present one. Flow recirculation in
the impeller exit area should not affect leakage
flow rate in this case due to the small clearance
that separates the main impeller passages from
their own shroud chambers as reported by Gantar
et al. (2002). As already reported, recently
Thamsen and Bubelach (2011) denied this
assumption in case of large gaps. In the latter case
the proposed approach must be changed to
include impeller/cavity interaction and a fully
coupled approach under unsteady conditions is
suggested.
Grid dependence analysis has been performed for
all the components. The procedure used to define
the optimal mesh is here described for the front
shroud only and has been repeated for the other
components. The sensibility analysis of the
leakage flow and the axial thrust to the spatial
resolution has been performed. Grids
characterized by a different resolution both
nearby and far from the walls have been realized
Fig. 5 Front shroud computational hybrid grids.
(a)
(b)
Fig. 6 Leakage flow (a) and thrust (b) as a function
of impeller head for front shroud.
(Fig. 5). Grid 1 contains around 27300 elements,
Grid 2 70400 (+258%) and Grid 3 89800
(+379%). In Fig. 6 the leakage mass flow and
axial thrust value as a function of the head across
the cavity for the front shroud are reported. The
shown results are non dimensional with respect to
the BEP values. It should be noticed that, despite
of the complex geometry of the front shroud
Engineering Applications of Computational Fluid Mechanics Vol. 6, No. 2 (2012)
275
cavity, both the leakage mass flow and the axial
thrust acting on the rotating wall are a linear
function of the impeller head. It can be seen that
leakage flow and thrust are underestimated,
respectively, by about 22% and 16% by Grid 1
relative to Grid 2, while for Grid 2 and Grid 3 the
mismatches are negligible. Therefore, Grid 2
strikes a balance between computational costs and
accuracy of results.
4. STAGE COUPLING APPROACH
The axial thrust on pump shaft is due to the forces
balance on the impeller walls; however the correct
pressure field evaluation requires also the
stationary components simulation to provide the
right boundary conditions for the rotating parts
analysis. The presence of both stationary and
rotating parts in each stage suggests a coupled
approach for the diffuser/impeller interaction
evaluation, namely to deal with the computation
of domains characterized by different periodic
conditions because of the different number of
vanes/blades. A mixing plane technique has been
applied in present study. Diffuser inlet and
impeller outlet have been extended in the
computational models to avoid errors induced by
imposing the boundary conditions on an interface
plane exactly corresponding to the real
components inlet and outlet sections.
The mixing plane approach allows solving
separately the rotating and the stationary domains
in steady conditions. Data from adjacent zones are
tangentially averaged and then imposed as mixed
boundary conditions at the interface. This
approach removes any unsteadiness deriving from
the circumferential variations in the interface
plane, thus yielding a steady state result.
Nevertheless, this method allows to maintain a
nonuniform radial distribution of the variables
and then a swirled inlet velocity distribution.
Since the impeller eye has a large inlet area and
the tangential nonuniformity of the flow at the
impeller inlet is negligible with respect to the
axial component of the flow, this method provides
a realistic evaluation of the stage performance.
Also Adami et al. (2005) demonstrated that the
approximation of the timeaveraged results was
quite reasonable especially for the performance
parameters evaluation. It must be pointed out that
this kind of approach is not accurate for volute
pumps. In that case unsteady calculations would
be appropriate, although the computational time
increase would be a nonnegligible factor
depending on the flow field complexity and the
mesh characteristics.
In the present method, a further hypothesis of the
stage kinematics repeatability has been assumed.
Therefore, the velocity profile at the impeller exit
section has been employed to update the inlet
boundary conditions of the diffuser model. The
whole process is iterative and can be synthesized
in the following main steps:
(a) Imposition of the initial boundary conditions
on the diffuser inlet (flow rate and velocity
direction) and on the impeller inlet (flow rate)
and front shroud leakage inlet (flow rate);
(b) Solution of the RANS equations in both
domains;
(c) During the simulation, the CFD solver
updates the conditions on the mixing plane
averaging the flow field and the static
pressure values in tangential direction;
(d) Once the convergence has been reached, the
boundary conditions are updated at the
interfaces;
(e) The steps from (a) to (d) have to be repeated
until convergence is achieved both for
velocity profiles and pressure values.
The stage computational hybrid grid is reported in
Fig. 7. The spatial discretization consists
approximately of 500000 elements both for
impeller and diffuser. Inlet and outlet sections of
the model are shown as well as the mixing planes.
The mean values for the updating of the boundary
conditions on the diffuser inlet are obtained on the
real interface. Then the velocity vectors are scaled
assuming a free vortex distribution and finally
imposed at the diffuser inlet.
Impellerdiffuser interaction is not limited to the
main flow, but also involves the leakage flow that
considerably affects both the axial thrust and the
efficiency of the whole pump. A complete
description of the coupling procedure can be
Fig. 7 Computational grid for steady stage
simulation with mixing plane approach.
Engineering Applications of Computational Fluid Mechanics Vol. 6, No. 2 (2012)
276
IMPELLER
Boundary Conditions:
Inlet velocity direction
(from DIFFUSER EXIT because of repeatability)
Balancing flow rate
(from SIDE CHAMBER simulation)
DIFFUSER
Boundary Condition:
Inlet velocity direction
(from IMPELLER EXIT because of repeatability)
Outlet static pressure
(from IMPELLER INLET by mixing plane)
Balancing flow rate
(from SIDE CHAMBER simulation)
Fig.8 Scheme of complete coupling procedure.
found in Fig. 8. The leakage flow passing through
the back shroud chamber is negligible with
respect to the main flow rate (less than 0.5% in
the range of interest), while the flow passing in
the front shroud cavities has been considered
(about 1.5%). Therefore, the impeller geometric
domain has been modeled including the stage
interface with front shroud chamber. Concerning
the back shroud it must be underlined that the
effect of the leakage flow on the stage
performance is neglected while the contribution
provided by the pressure field to the axial thrust is
very important and is included in the F
bs
term of
Eq. 1. The numerical study of the impeller shroud
chambers has been carried out before the stage
analysis, assessing a correlation between the
impeller head and the front shroud leakage flow
to update the boundary conditions. An iterative
cycle on the leakage flow rate has to be performed
because the impeller head decreases when the
main flow rate increases, while the leakage flow
increases with the impeller head. The CFD
procedure has also been repeated for different
capacities. Stage efficiency and characteristic
curve have been obtained and their dimensionless
version is shown in Fig. 9. The reference value
corresponds to the result obtained at the BEP. The
static head has been defined as reported in Eq. 3.
outlet inlet
p p
H
g
=
(3)
As observed in the stage simulation, the impeller
performances strongly depend on the inlet
conditions. To simulate the first and the third
impeller, the suction volute and the crossover
device have been studied by a standalone
approach as well. For these components,
correlations between the evolving flow rate and
the pressure variation have been obtained.
Furthermore, velocity distributions of the suction
volute and crossover device have been
tangentially averaged at the interface plane and
then imposed on the downstream impeller inlet.
This method is similar to mixing plane one, but
neglects both the steady and unsteady interactions.
The importance of considering realistic boundary
conditions for the impeller inlet is evidenced by
Fig. 10, where an example of the pressure field
and streamlines in the crossover are reported for
the BEP. As can be seen, mechanical support
devices provide nonuniform conditions at the
crossover exit/impeller inlet section, which can be
estimated in yaw angle values up to 15 in off
design conditions.
Non Dimensional Stage Head and Efficiency
0.0
0.2
0.4
0.6
0.8
1.0
1.2
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
flow rate / flow rate @ BEP []
h
e
a
d
/
h
e
a
d
@
B
E
P
[

]
0.0
0.2
0.4
0.6
0.8
1.0
1.2
e
f
f
/
e
f
f
@
B
E
P
[

]
Stage Head
Stage Efficiency
Fig.9 Non dimensional stage head and efficiency.
Engineering Applications of Computational Fluid Mechanics Vol. 6, No. 2 (2012)
277
Fig. 10 Pressure field and streamlines in crossover.
Non Di mensi onal Impel l er Head
0.70
0.75
0.80
0.85
0.90
0.95
1.00
1.05
1.10
0.70 0.80 0.90 1.00 1.10 1.20 1.30 1.40
fl ow rate / fl ow rate @ BEP []
h
e
a
d
/
h
e
a
d
o
f
f
i
r
s
t
i
m
p
e
l
l
e
r
@
B
E
P
[

]
First Impeller
Third Impeller
Stage Impeller
Fig. 11 Dimensionless head for simulated impellers.
Non Di mensi onal Impel l er Effi ci ency
0.90
0.95
1.00
1.05
1.10
0.70 0.80 0.90 1.00 1.10 1.20 1.30 1.40
fl ow rate / fl ow rate @ BEP []
h
e
a
d
/
h
e
a
d
o
f
f
i
r
s
t
i
m
p
e
l
l
e
r
@
B
E
P
[

]
First Impeller
Third Impeller
Stage Impeller
Fig. 12 Dimensionless efficiency for simulated
impellers.
Impellers performance and efficiency curves have
been also plotted (Figs. 11 and 12). The shown
results are non dimensional with respect to the
value obtained for the first impeller at the pump
BEP. The first impeller shows the highest head
for every flow rate due to the fact that the flow
enters the first impeller axially, that is at the
design conditions. Instead, the third impeller has
lower values of head that get further and further
from the first impeller ones since the flow rate
grows. In fact, the flow entering the third impeller
has not been straightened by the diffuser but
directly comes from the annular chamber where
only structural elements are present (Fig. 10). The
other impellers head is a linear function of the
flow rate and has halfway values between the first
and the third ones. The efficiency curves confirm
good performances for the first impeller and
worst values for the third one. However the third
impeller efficiency is higher than expected at low
flow rates, probably due to the leakage flow effect.
The results confirm a high dependence of
impeller performance on its inlet flow field
distribution.
5. SIDE CHAMBERS AND BALANCING
DRUM ANALYSIS
Impeller shroud chambers have been studied by
means of a 2D axisymmetric approach. To get an
exhaustive leakage characterization, the CFD
analysis has been performed for different heads
across the cavity, calculating the corresponding
leakage massflow and the contribution to the
axial thrust. Cavity flows are therefore solved at
different operating points in a preliminary phase
and do not require to be calculated at every update
of the stage coupling boundary conditions. The
results have been employed to assess correlations
between the head across the chambers, the
Engineering Applications of Computational Fluid Mechanics Vol. 6, No. 2 (2012)
278
leakage flow rate and the axial thrust on the
rotating walls. The obtained equations are
consistent with the correlations proposed by
Stirling (1982), Denny (1954) and Worster and
Thorne (1959), currently in use at WEIR
Gabbioneta SRL.
The flow inside the cavity moves outward in
radial direction at the rotating wall and inwards at
the casing wall. The central core rotating speed is
lower than the impeller one (Batchelor, 1951;
Stewartson, 1953).The rotating velocity inside the
chambers depends on peripheral conditions,
namely on the impeller rotating speed and on the
head across the cavity itself. Due to pump
geometry, head across the front shroud cavity
corresponds to impeller head, while the leakage
flow in the back shroud chamber depends on the
pressure rise in the diffuser (Fig. 3). Therefore,
leakage massflow and its contribution to the
axial thrust depend on the pump working
conditions.
Results from the front shroud chamber analysis
have been discussed in a previous section. In Fig.
13a the back shroud impeller chamber grid is
shown. Its leakage flow rate and its contribution
to the residual axial thrust as a function of the
diffuser head are visible in Fig. 15. Both leakage
flow rate and axial load are a linear function of
the diffuser head. Considering the central
balancing drum, the head across the cavity
depends on the working conditions in the second
group of impellers and in the crossover device.
The computational grid is reported in Figs. 13b
and 14, where a section of the mesh is presented,
while the obtained results are visible in Fig. 16.
All the correlations can be expressed as linear
function of the head across the balancing drums.
The two contributions to the thrust have opposite
behaviors. The thrust on the second impeller is
nearly constant in the range of interest while high
variations on the fifth impeller can be observed.
In fact, for a variation of the head of 40% the
axial thrust almost decreases at the same
percentage. The most interesting conclusion is
that looking at the contributions of each chamber,
the central balancing drum is the key element for
residual axial thrust intensity. Changing this
component allows to control the thrust.
6. SIDE CHAMBERS FLOW FIELD
ANALYSIS
Head across the cavity results in a centripetal flow
with inwards structures at the stationary wall and
outwards ones at the rotating walls. Local
tangential velocity v
divided by the impeller
Fig. 13 Back shroud (a) and central drum (b)
computational hybrid grids.
Fig. 14 Detail of computational grid of central
balancing drum along section AA.
Di ffuser Head vs Leakage and Thrust for Back Shroud
0.6
0.7
0.8
0.9
1.0
1.1
1.2
0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1
di ffuser head / di ffuser head @ BEP []
l
e
a
k
a
g
e
/
l
e
a
k
a
g
e
@
B
E
P
[

]
0.90
0.95
1.00
1.05
1.10
t
h
r
u
s
t
/
t
h
r
u
s
t
@
B
E
P
[

]
Leakage Flow
Thrust
Fig. 15 D Leakage flow and thrust as a function of
impeller head for back shroud.
Head vs Leakage Flow and Thrusts for Central Drum
0.80
0.90
1.00
1.10
1.20
0.7 0.8 0.9 1.0 1.1
head (central drum) / head (central drum) @ BEP []
l
e
a
k
a
g
e
/
l
e
a
k
a
g
e
@
B
E
P
1.2
1.1
1.0
0.9
0.8
0.7
0.6
t
h
r
u
s
t
/
t
h
r
u
s
t
@
B
E
P
Leakage Flow
Thrust Fifth Impeller
Thrust Second Impeller
Fig. 16 Leakage flow and thrust as a function of head
across the central drum.
Engineering Applications of Computational Fluid Mechanics Vol. 6, No. 2 (2012)
279
velocity u (Eq. 4) has been reported in Fig. 17 for
the front and back shroud cavities together with
the streamlines. Repeating calculations showed
that peripheral velocity with respect to the
impeller one depends on the massflow condition
through the cavity.
loc loc
r
r
u u
u u
e e u u
O O
= = = (4)
The dimensionless rotating velocity distribution
in the middle of the back cavity is reported in Fig.
18 at different flow rates. Fluid rotating speed
inside the cavity presents a typical value for
dimensionless radius ranging from 0.55 to 1. The
typical characteristics of the Batchelor flow
(Batchelor, 1951) can be observed, namely, the
fluid rotates as a solid body between two
boundary layers. For the internal regions, the
distribution called Stewartson flow (Stewartson,
1953) is encountered, with the classic vortex
distribution. Increasing the massflow through the
cavity, the presence of the Stewartson distribution
is increased, while the value of constant rotating
velocity of the core flow is reduced. This agrees
with the results of Debuchy et al. (1998). The
conclusion was that the Batchelor type flow can
be observed at low massflow rates, and far from
the periphery because of the influence of the inlet
conditions. It was also found that the flow
structure near the axis is strongly affected by a
(a)
(b)
Fig. 17 Nondimensional local rotational velocity
distribution inside front/back impeller
chamber.
weak stream, which enhances the level of the
coreswirl ratio near the axis. Pressure
distribution on the back and front shroud of the
impeller is qualitatively reported in Fig. 19. The
pressure distribution inside is highly influenced
by the flow conditions. In Fig. 20, the actual
radial distribution of static pressure inside the
back shroud cavity is reported. A different value
of leakage flow rate divided by the leakage flow
rate at BEP corresponds to each curve. All the
curves are plotted considering the same reference
pressure at the impeller exit.
Let us consider now the same distribution divided
by a term expressing the local centrifugal force
(Eq. 5), as reported for different massflow in Fig.
21. It should be underlined that e
loc
expresses the
local fluid rotational speed in front and back
impeller chambers and r is the local radial
coordinate. It can be observed that all curves are
coincident except between r/R=0.45 and r/R=0.55,
where the outlet of the back shroud chamber is
located.
2 2
'
loc
p
r
p
e
A
A =
(5)
Fig. 18 Dimensionless rotating velocity inside the
back impeller cavity at different massflow
(referred to the leakage massflow at BEP).
Fig. 19 Leakage contributions to axial thrust.
Engineering Applications of Computational Fluid Mechanics Vol. 6, No. 2 (2012)
280
Fig. 20 Radial distribution of static pressure inside
back shroud chamber at different massflow
rates.
Fig. 21 Radial distribution of dimensionless pressure
inside back shroud chamber at different mass
flow rates.
7. AXIAL THRUST: RESULTS AND
DISCUSSION
To calculate the axial thrust acting on the whole
pump, the contribution of every stage has been
separately evaluated according to Eq. 2.
Concerning the cavities (the F
fs
and F
bs
terms), the
contribution to the thrust is calculated by adding
the real corresponding reference pressure to the
pressure integral on the rotating wall:
, ,
,
ax cavity exit impeller
A A
exit impeller
A A
T p ndA p ndA
p ndA pndA
= + =
+
} }
} }
(6)
The area is referred to both the front and back
impeller shroud chambers. Similarly the axial
thrust is calculated for central and lateral
balancing drums. If directed from the second to
the first bench a force is considered positive. Each
stage contribution to the axial thrust for the design
flow rate is reported in Fig. 22, highlighting the
opposite effects of impeller shroud chambers. The
values are non dimensional with respect to the
absolute value of the axial thrust at BEP. It can be
noticed that, even for the opposite alignment of
the two groups of stages, the global thrust is not
null. The impeller back shroud contributions are
higher than the corresponding one from the
impeller front shroud. Furthermore, the second
bench has a higher number of stages and the
opposite configuration of the groups is not
sufficient to balance the global thrust. The
residual thrust is negative in the chosen reference
system and it means that the shaft is in
compression. It can be also verified that the
greater contribution is due to the central balancing
drum where the flow is driven by pump head and
by wear rings clearances eventually modified by
mechanical wear. Current result is consistent with
design assumptions usually considered by the
industry. In fact, when two benches are present,
the residual axial thrust is considered to be
generated by the uneven pump stage, when
existing. In our case the fifth stage provides a
nondimensional contribution of 0.98, which
means that the residual axial thrust is totally
generated by the last stage.
For a proper design of the thrust bearings, it is
still required to know the maximum thrust in
pump operating range. It can be also underlined
that the contribution of each component to the
residual thrust is one or two orders of magnitude
higher than the final value. Then, the use of the
proposed procedure for bearing dimensioning
requires a careful evaluation of the accuracy of
the CFD simulation and an experimental
validation of the procedure. Analyses have been
carried out between the 80% and the 115% of the
BEP flow rate as reported in Fig. 23.
Experimental study has been performed by
WEIRGabbioneta SRL using load cells. Axial
thrust values are nondimensional with respect to
the absolute value obtained experimentally for the
BEP condition.
Engineering Applications of Computational Fluid Mechanics Vol. 6, No. 2 (2012)
281
Fig. 22 Discrete residual thrust evaluation at BEP flow rate.
Fig. 23 Residual axial thrust between 80% and 115% of BEP flow rates.
Both experiments and CFD indicate that in the
range considered for the investigated pump, the
maximum thrust is obtained at the lowest flow
rates corresponding to the highest value of pump
head. Second order polynomials fit the data with
high accuracy: the value of R
2
is 1.000 for the
proposed method and 0.979 for the experiments.
These results suggest a linear relation between the
residual thrust and the pump head in the selected
range of flow rates. The numerical method also
succeed in the evaluation of both the residual
thrust direction and its reduction at the higher
flow rates.
The discrepancies between the experimental and
numerical data are not negligible. To quantify the
inaccuracy, the difference between the data with
respect to the experimental value is evaluated as
follows:
,
.100
ax
ax CFD
ax
T T
T
d
= (7)
The trend of this curve is linear, as demonstrated
by the R
2
value of 0.998 for a linear law. This
unexpected result suggests that a key
phenomenon is missing in the numerical model
and that this phenomenon can be represented by a
linear function of the flow rate. It must be pointed
out that no experimental uncertainties are
available and hence it is not possible to ascertain
the actual difference between the numerical data
and the experiments. Further studies are necessary
to individuate the flow feature to be accounted for.
Engineering Applications of Computational Fluid Mechanics Vol. 6, No. 2 (2012)
282
8. CONCLUSIONS
A computational study of a horizontal multistage
pump has been carried out, with the residual axial
thrust evaluation as the main target. The
contribution of each single pump component to
the axial load has been estimated by a CFD
investigation of its internal flow and pressure
field. Then, all the computational results all
collected and a methodology to get the residual
axial thrust of the whole pump is developed. The
suggested method is flexible and the obtained
results are consistent with the results given by
standard tools in use in industry. Furthermore,
each component is studied considering realistic
boundary conditions by means of a loosely
coupled approach with the upstream/downstream
components.
Impellers and diffusers have been coupled and
analyzed by applying a mixing plane approach.
The procedure described by the authors allows to
analyze also the interaction between the main
flow and the leakage flow, providing accurate
evaluation of stage performances. Comparisons
between stage curves evidenced the effect of
considering different inlet conditions and leakage
flows.
Leakage flows through wear rings of shroud
chambers have been simulated separately with 2D
axisymmetric models. The obtained results in
terms of correlations are consistent with the
literature information and represent the key
feature for the thrust evaluation, especially when
considering the central balancing drums. Flow
fields inside the cavities show the typical
structures of rotating cores. The main limit of this
approach is that the tangential nonuniformity
generated by the bladerow interaction and the
flow recirculation is neglected, thus providing
ideal curves only. It must be pointed out that
experimental correlations, which are often
obtained in controlled environments, also have
the same limits. A more detailed analysis of the
effect of nonuniform inlet conditions for cavity
flows is hence necessary.
Coming to the thrust curve, the selected method
suggests a proper dimensioning of bearings that
can guarantee mechanical reliability. The trend of
the curve is well recognized as well as the
direction of the thrust over the whole range,
which is one of the points of interest for the
industrial application of the selected method.
It could be concluded that the selected method
would be very useful to individuate trends when
modifying pumps crucial components, such as
the balancing drum or the cavity flows shape and
dimensions (during the design phase) or assessing
wear rings decreasing performances. The main
limit of the numerical approach is its poor
accuracy when applied to pumps with relatively
wide gaps for leakage flows. Furthermore, some
assumptions are invalid when evaluating off
design conditions. Nevertheless, the force balance
does not fail in the listed cases and it is possible
to perform fully coupled simulations of the
impeller/diffuser/ cavity region of flow at a higher
computational cost, if necessary. The leakage
flows should be studied by means of fully coupled
approach and the effect of the statistical
distribution of the tolerances could be considered
as well.
NOMENCLATURE
A passage area [m
2
]
BEP best efficiency point (pump design
point)
CFD computational fluid dynamics
d discrepancy between experimental and
numerical data [%]
F force [N]
g gravitational acceleration [m/s
2
]
H head [m]