Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Both radial and axial hydraulic forces act on the impeller of a centrifugal pump. The radial forces are due to three different mecha-
nisms: (a) radial thrust due to non-uniform pressure distribution around the impeller, (b) labyrinth forces and (c) impeller-diffuser
interaction forces. The nature, size and origin of these forces are discussed. Steady radial thrust depends on the specijic speed of the
pump, type of casing (single or double volute, diffuser), geometric parameters and operation conditions including cavitation.
The impact of these factors on radial loads is demonstrated and quantitative data are presented. Four direrent methods for radial
thrust measurements are presented.
Axial thrust on pump impellers is extremely sensitive to tolerances and to the labyrinthflowrate. Two methods for predicting axial
thrust are discussed and predictions are compared to measurements on a multi-stage boiler feed pump. A number of parameters
influencing axial thrust are treated.
AKR= AH*
~-
Force Type Mechanism/origin/remarks QIQm H
Steady radial thrust D Rotation unsymmetries 0.02
V at impeller inlet, >0.5 to
AFR increases with Q/Q, 0.05
D Rotational unsymmetries 0.01
caused by pump outlet, 20.5 to
AFRincreases with Q/Q, 0.02
D Diffuser tolerances, 20.5 0.01
AFR increases with 010,
Part load flow separation in diffuser <0.5 0.01
to
D * Caused by Q tolerances (?) 0.02
* The equilibrium of a system
of Z , parallel channels
is very sensitive to
small perturbations
D Static eccentricity of 0.01
impeller with respect to
V to diffuser? 0.02
SV Large-scale non-uniform <0.5 0.2
pressure distribution to
0.4
DV Outer volute has higher resistance 0.05
than inner volute which to
gives non-uniform pressure 0.15
distribution around impeller
Labyrinth force All Steady pressure distribution All --o.2ot
in eccentric labyrinth, force: to
* Increases with E , ~ J -0.05
* Decreases with serrations
Deadweight To be considered in test data
evaluation (mind also buoyancy - -
-
0)
-
Q AK, = 2
AH
Force Type Mechanism/origin/remarks R Q, H
Mechanical unbalance To be considered in test data 1 -
evaluation (determined
in air)
Unsteady radial thrust All Geometric tolerances of impeller: 1 0.04
* AB2 to
* AB2 0.08
a Mi, AAi ~
occur with a basic mode (vD= v , = 1) or at higher force F,, . The observed rotor position would therefore
modes if be O,, and the measured radial thrust FR, differs by
magnitude and direction from the true hydraulic thrust
VD 2, - v,z, = k1 F R T . A non-vibrating rotor under the assumed radial
Both steady and unsteady radial thrust data given in load would rotate around 0, and a vibrating rotor
this paper are normalized according to would show some orbit around 0,. In practice every
rotor will be subject to some vibrations.
The labyrinth forces discussed above can be esti-
mated from numerous publications, for example refer-
ence (5), but reliable data for high Reynolds numbers
Unsteady radial thrust values given in this paper are and different types of labyrinths are still scarce. In the
to be understood as peak amplitudes in a frequency present context it is sufficient to note that the radial and
range between zero and impeller blade passing fre- tangential forces increase with the head of the pump or
quency. the square of the speed and the damping forces increase
with the speed of the pump. There is also an apparent
mass associated with the acceleration of liquid in the
2.2 Labyrinth forces gap of the labyrinth, but this effect can be ignored for
the discussion of radial thrust. In the literature the laby-
In a labyrinth there are radial and tangential forces that rinth forces are treated as stiffness, damping and mass
increase with the static eccentricity and the vibration coefficients. The order of magnitude of the stiffness
orbit of the shaft. While these forces are negligible for effect (calculated with half the labyrinth clearance) can
radial labyrinths, they are strong for axial flow laby- be seen from Table 1, where the labyrinth forces were
rinths as used in most pumps. Therefore these forces normalized according to equation (2) in order to
should not be ignored when interpreting radial thrust provide a comparison with other radial forces. The
measurements. negative sign indicates that it is a restoring force acting
Figure 1 shows how radial thrust and labyrinth forces in opposite direction of the radial displacement of the
determine the rotor position and influence the apparent rotor.
radial thrust. We assume that a radial thrust FRTacts
on the impeller and pushes the impeller from its concen-
tric position M to the eccentric position 0 which would 2.3 Impeller diffuser interaction
be observed with a radial labyrinth. An axial flow laby-
rinth, however, would give a restoring radial force F,, , If a rotor vibrates with respect to the casing (volute or
pushing the rotor somewhat back, and a tangential diffuser) the pressure distribution around the impeller
Q IMechE 1987 Proc lnstn Mech Engrs Vol 201 No A3
166 J GUELICH, W JUD AND S F HUGHES
tX Advantages
1. This method is simple, since no sophisticated
equipment is needed.
2. No resonance with flexible measuring elements
endangers the accuracy of the results.
Disadvantages
1. The measurement of a multitude of pressure tappings
and the integration is cumbersome.
2. The accuracy is low because only the pressure dis-
tribution on the wall is recorded.
3. In stationary thrust components cannot be deter-
mined.
The interpretation of the results is difficult because
the pressure distribution on the impeller shroud and
hub and their possible contribution to the force on the
impeller are unknown. It is therefore not surprising to
find that radial forces integrated from the measured
pressure distribution in reference (10) are lower than the
forces on the bearing since the integration was only
done over the impeller outlet width.
(a) Volute flow at part load (c) Combination of concentric casing and volute
as described in reference (6). As with measurement of flow deceleration takes place in the second half of the
the bearing forces, resonances must be avoided and the volute and the radial thrust is directed to the stalled
measured forces include also the labyrinth forces unless region, that is somewhat downstream of the tongue.
a radial labyrinth (see Fig. 1) is employed. This tech- Putting in an additional rib opposite the cutwater
nique clearly is much more complex than the measure- (that is part of a double volute) produces a second
ment of the bearing forces and is only justified in special stalled region and increases the symmetry of flow and
cases. pressure distribution considerably (see Fig. 2b). Tests
If resonance problems cannot be avoided the test reported in reference (ll),which demonstrate this effect
pump can be treated as a vibrating system consisting of very clearly, are reproduced in Fig. 3.
one mass, stiffness and damping (8). If the tongue is cut back to form a concentric casing
As in the methods of Sections 3.2 and 3.3, the system (for example over 90" of the circumference), as shown in
must be checked for resonances since the shaft must be Fig. 2c, the stalled region is eliminated since the geo-
flexible enough to permit accurate readings. metric tongue angle has been decreased and radial
thrust at part load is reduced accordingly. The radial
4 RADIAL THRUST thrust measurements of references (l),(12) and (13) show
clearly this reduction of thrust at part load.
4.1 Radial thrust in volute pumps At flowrates in excess of the design point the flow
separation is (due to the high flow angle) on the pres-
4.1.1 Volute pressure distribution and radial thrust
sure side of the tongue (that is in the pressure nozzle).
At part load there is flow separation at the suction side Consequently there is a pressure build-up on the
of the cutwater since the flow angle is smaller than the suction side of the tongue (see Fig. 2d) and the radial
geometric angle of the tongue (see Fig. 2a). Consequent- thrust is directed more or less opposite to the direction
ly in the stalled region there is little pressure recovery, at part load. Since the volute throat velocity is much
0 IMechE 1987 Proc lnstn Mech Engrs Vol 201 No A3
168 J GUELICH, W JUD AND S F HUGHES
10 20 30 40 50 60 70 80 90
Y
Fig. 4 Maximum steady radial thrust coefficient for single
volutes: curve 1 according to reference (1); curve 2 for
large clearances
y=a.nq (3)
where
a=l for single-entry impellers
a= 42 for double-entry impellers
Figure 4 is based on the data of references (1) and (2),
the band of scatter being estimated from available test
data. Curve 2 is a rough estimate from tests with large
clearances, which are confirmed by reference (17).
Curve 1 in Fig. 4 is valid for clearances according to
reference (18).
The excessive leakage associated with large clearances
t I
0 01
:
' -T
0 02
t I
0 03
-- -I I
004
(that is clearance above two to three times the API 610
values) has an impact on the pressure distribution on
Flow
___ the impeller shrouds. Nevertheless it is suggested that
m'/s curve 2 should be used too for radial labyrinths with
any type of clearance.
Fig. 3 Influence of volute rib length on radial thrust (11)
The maximum steady thrust given by Fig. 4 occurs
mostly at Q = 0; in some cases for 0, Q = 0.5Q,.
larger than the impeller outlet velocity, there is an accel- Having determined K,, from Fig. 4 the thrust curve
eration and an associated drop in static pressure near KR = f (Q) can be estimated from the formula proposed
the cutwater. by Stepanoff (19):
The mechanisms described above are confirmed by
measured pressure distributions (9, 14-16) and direc-
tions of the radial thrust. However, non-uniform pres- (4)
sure distribution at off-design conditions occurs already
without flow separation. The direction of the radial Contrary to the above formula, the steady radial
thrust in single volutes is given in Fig. 2a and d. thrust is not zero at BEP but has a finite value in the
For double volutes the steady radial thrust at part range of 0.03 < K , < 0.08 which was found to be inde-
load can be very sensitive to small geometrical varia- pendent of the specific speed in the range of
tions, which impair the symmetry of the pressure dis- 10 < nq < 100. This force is caused mainly by the finite
tribution in the inner and outer volute. This is thickness of the cutwater.
demonstrated also by Fig. 6 which shows that small Because of the high number of variables, it seems so
deviations from a 180" volute cause an appreciable rise far impossible to establish a correlation with less scatter
in radial thrust. than in Fig. 4 by introduction of specific geometric
It appears that the low radial thrust of a double design parameters.
volute depends on a rather delicate equilibrium of pres- Figure 5 gives steady radial thrust coefficients for 50
sures in the inner and outer volute that can be per cent of BEP flow. The upper scatter band is even
destroyed easily by unsymmetries of the cutwaters and larger than in Fig. 4 since often maximum thrust does
the shape of the volute. In this respect radial thrust in not occur at shut-off but at part load up to 50 per cent.
single volutes is is probably less sensitive to small modi- The maximum thrust occurs when the impeller blade
fications. loading becomes excessive (that is the maximum local
pressure in the volute is limited by the impeller blade
loading)
4.1.2 Design data for single and double volutes
For double volutes Fig. 6 together with Fig. 4 can be
In the absence of specific test data Fig. 4 can be used to used to determine the maximum steady radial thrust.
estimate the maximum steady radial thrust for single The maximum steady radial thrust of a double volute is
Proc Instn Mech Engrs Vol 201 No A3 @ IMechE 1987
PARAMETERS INFLUENCING HYDRAULIC FORCES ON CENTRIFUGAL IMPELLERS 169
0 1 ' 1 1 1 1 1 1 1 1 ~ 7 ' 1 -
'a-i
entire load range, whereas components with other fre-
quencies were not changed. The reduction in static
R 0 05 - thrust came rather unexpectedly.
h The variation of 15" is much higher than any manu-
facturing tolerances would be. A first test. which was
done with an increase of 4" of the impeller blade outlet
angle, showed virtually no influence on radial thrust.
0 0
I t I 1 -
This finding is confirmed by reference (23). From the
tests so far available it appears that geometric toler-
ances of the impeller outlet do not contribute signifi-
cantly to the (unsteady) radial thrust.
Proc Instn Mech Engrs Vol 201 No A3
170 J GUELICH, W JUD AND S F HUGHES
0.004- ' 1 \
/
i
'w' \ /
0.002- \
\+,'
0 I
An offset of the hydraulic from the geometric centre (b) Influence of difiser tolerances If one channel of a
or blade inlet angle tolerances are other factors that can vaned diffuser differs in geometry from the other chan-
contribute to the unsteady radial thrust at the frequency nels a static radial thrust is expected. Figures 9 and 10
of rotation. show thrust measurements where one of nine diffuser
Total
1 I I
0 0.5 1.0 I .5
Q
Qo 0
1 +Original design
Reduced cross-section area by 30 per cent
-0-
at one channel
0.08 -
/
P
0.07 - /
/
0.06- d
0
/
0.05-
P'
0.04 -
/
#
P#
0.03 - /
+ Original design
0.20 -*--
Reduced cross-section area by 30 per cent
at one channel
0.5 I .o 1.5
e
Qo 0
channels had a throat area reduced by about 30 per inlet and/or outlet. Some test data are given in reference
cent from the design value (same test pump as in the (12). Although these data were derived for a pump with
previous section). As expected, the steady thrust one bend downstream of the diffuser and one bend
increases whereas the unsteady thrust is not much shortly upstream of the impeller, they might be useful
affected, except near shut-off. (in the absence of other data) to estimate the radial
thrust :
4.3 Influence of cavitation on radial thrust
Data in references (2) and (17) showed no increase of
steady or unsteady radial thrust with increasing cavi-
Static thrust: K & = 0.02 for
Unsteady thrust: Kgun = 0.01 Q = Q
, I
For Q > Q,, static and unsteady thrust increase. Note
tation. Tests of reference (10) showed an increase of that KZ is defined differently from radial impellers:
radial thrust only at flowrates above BEP and at high
degrees of cavitation. Figure 11 confirms this for the
total (that is steady and unsteady) radial thrust of a dif-
fuser pump (nq = 47). It is concluded that cavitation up
to 3 per cent head drop will not cause, in general, an It is also possible to estimate the radial thrust caused
increase in steady or unsteady radial thrust-neither in by an uneven inlet velocity distribution by calculating
volute nor in diffuser pumps. the differences in head produced by the differences in
local incidence.
/
,
0 0.5 I .o I .5
Q
z
Fig. 11 Influence of cavitation on radial thrust
Q IMechE 1987 Proc lnstn Mech Engrs Vol 201 No A3
172 J GUELICH, W JUD AND S F HUGHES
Radial thrust coefficient K, The empirical method as discussed in reference (27) can
Type of -
Q
pump Q, Steady Unsteady be characterized as follows:
Single W.5 Fig. 4 0.07 A ‘theoretical axial thrust‘ is calculated on the basis
volute 0 12 of more or less crude assumptions as to the rotative
1.0 0.03 to 0.01 to velocity of the liquid between the casing and impeller
0.08 0.05 and as to the static pressure rise of the impeller.
~ ~~
A s single
From model or prototype tests with the hydraulic
volute equation (5) volute components in question the difference between actual
thrust and theoretical thrust is determined and an
DiKuser Table 3 appropriate dimensionless corrective term is derived.
This corrective term can subsequently be used to
predict the thrust of pumps of different size, speed
with geometric and operational data. Amplitudes and and number of stages but with geometrically similar
frequencies depend on flow separation and geometric hydraulic components.
tolerances which disturb the often delicate flow sym- The corrective term must be known as a a function
rnetry in the pump. of Q/Qoo,Reynolds number and labyrinth clearance.
From the authors’ experience the following assess-
ment can be made: Since the labyrinth leakage increases with Reynolds
number and the leakage has an important impact on
Subsynchronous thrust components, which show the pressure distribution on the impeller shrouds, the
broadband spectrum behaviour, amount normally influence of Re is strong. For this reason prototype tests
up to 20 per cent of the total unsteady thrust. Any are preferred over low-speed model tests.
clearly defined frequencies in the subsynchronous If test data of high quality are available the accuracy
range, such as reported in reference (25), could lead of prediction is entirely satisfactory. For new configu-
to a vibration problem in high-speed machinery. rations or for the investigation of the influence of indi-
The synchronous thrust component is expected to vidual parameters a finer calculation model is required.
amount to 1&20 per cent of the unsteady thrust.
The high portion of blade passing frequency thrust
shown in Figs 8 and 10 is rather unusual when com- 5.2 Theoretical method
pared to the majority of the authors’ test data. Nor-
The theoretical approach is based on the following pro-
mally thrust components of multiples of the rotation
cedure:
frequency of the rotor are quite low and do not con-
tribute much to the vibration level of the rotor if 1. The static pressure rise of the impeller on the shroud
blade numbers of the impeller and diffuser are select- and hub streamline is calculated from a more or less
ed properly. sophisticated model (or taken from test data).
2. The circumferential absolute velocity at the impeller
4.6 Radial thrust design data summary outlet (C2J is calculated (or estimated from the Euler
equation).
For single- and double-entry impellers radial thrust can 3. The labyrinth flowrates are calculated.
be predicted according to Table 4. 4. Based on the above results the velocity and the pres-
Generally the measured unsteady thrust increases with sure distributions on the hub and shroud are calcu-
increasing labyrinth clearance because of the reduced lated as a function of geometrical data, labyrinth
damping. flowrate and direction and CZu.For this a procedure
The scatter of the radial thrust data of different similar to references (28) and (29) can be followed.
sources and for different pump types is caused by: Since the labyrinth Aowrate depends on the pressure
Different hydraulic designs of different manufac- distribution an iterative process is needed.
turers. 5. Finally, the pressure distribution is integrated to give
Influence of labyrinth forces (type of labyrinth, the forces on the hub and shroud and the resultant
length, clearance, Reynolds number). axial thrust.
Geometric tolerances of volutes, diffusers, impellers
(unsteady thrust). 5.3 Parameters influencing axial thrust
Differences in testing and data reduction procedures.
For example if a test is carried out with a very stiff To assess the merits and limitations of the methods dis-
shaft the bearing forces measured represent the true cussed above the parameters influencing axial thrust are
hydraulic thrust (little labyrinth forces since small examined.
deflection).
5.3.1 Main purumeters
5 COMMENTS ON AXIAL THRUST IN
MULTI-STAGE PUMPS The following parameters are correctly taken into
account by both methods:
Basically there is an empirical and a theoretical
approach to predict axial thrust. In practice, however, (a) diameters of impeller, labyrinths including balancing
both methods are distinguished only by the degree to device and steps of shaft subject to different pres-
which they rely on experimental data. sures,
Proc Instn Mech Engrs Vol 201 No A3 @ JMechE 1987
PARAMETERS INFLUENCING HYDRAULIC FORCES ON CENTRIFUGAL IMPELLERS 173
(b) head of pump and number of stages, These data show that Czuhas little effect on the pres-
(c) difference between normal and last stage of a pump, sure distribution on the impeller wall, if leakage is
(d) change of momentum of flow through the impeller. really outward or if no leakage is present. However,
with radially inward leakage there is a large impact
of C , , . Reference (31) shows, furthermore, that K
5.3.2 Static pressure rise of impeller depends strongly on whether a given impeller oper-
Although in general the hub and shroud streamline will ates in a volute or a diffuser, because the distribu-
be intended by design to give a constant static pressure tions of the circumferential velocity are very different
at the impeller outlet, this condition is not necessarily in both of these components.
always achievedGneither at part load nor even at the 7. Roughness of impeller ( K increases with increasing
BEP. It is obvious that a difference in static pressure roughness).
between hub and shroud has a direct impact on the 8. Roughness of casing ( K decreases with increasing
axial thrust. Test data have shown that a given impeller roughness).
can produce a constant pressure rise over the outlet 9. Reynolds number.
width with one diffuser but a non-uniform one with From the above discussion it is inferred that accurate
another diffuser of moderately different design. Con- prediction of axial thrust by analytical models, for the
sidering the dfliculties of predicting part load behav- time being, is only possible if supplemented by or
iour of pumps the above fact confirms that an accurate adapted to experimental data.
prediction of the static pressure at the impeller outlet is
not an easy matter. 5.3.4 Tolerances
The axial position of the rotor, which is known to have
5.3.3 Pressure distribution on impeller shroud and huh a significant influence on axial thrust, has an impact on
Often in pump technology, K = 0.5 was assumed ( K both the static pressure distribution at the impeller
being the ratio of fluid velocity in the space between the outlet (because the impelleridiffuser position is
impeller and casing to the circumferential velocity of the influenced) and on the velocity and pressure distribu-
impeller). The deviation of reality from the above tion between the casing and impeller.
assumption is the main reason for the failure to predict The same is true for the geometric tolerances of the
axial thrust in multi-stage pumps by any simple pro- impeller and diffuser.
cedure: in reality K is greater than 0.5 on the shroud
because of the inward leakage and less than 0.5 on the 5.4 Accuracy of thrust prediction
hub because of the outward leakage. Both deviations Naturally accuracy of thrust prediction would tend to
act in the same direction to increase the axial thrust to be assessed by comparing the calculated residual thrust
the suction side. to the measured one. To do this on a percentage basis
K and the resultant pressure distribution depend on a can be quite meaningless: for a highly balanced multi-
number of quantities: stage pump even a small discrepancy between predicted
1. Width of space between casing and impeller shroud. and measured thrust could show up as an error of 100
2. Magnitude and direction of labyrinth leakage and per cent or more, because the residual thrust is a small
hence Reynolds number as well as type and clearance difference of two large figures (that is the hydraulic
of labyrinth. thrust and the balance thrust).
3. On the (suction side) impeller shroud leakage is radi- Any deviation between predicted and measured
ally inward and K increases with leakage because the residual thrust should therefore be compared to the
flow at the impeller outlet has a circular momentum hydraulic thrust in order to get a meaningful assessment
(proportional to QL and C2,) which even increases of the accuracy of the prediction, which can be used
when flowing inwardly. subsequently to determine the necessary thrust bearing
4. On the (pressure side of a normal stage) impeller hub margin.
leakage is radially outward and K decreases with
increasing leakage. 5.5 Axial thrust tests of a boiler feed pump
5. The circular component of absolute velocity C,, at Figure 12 shows the residual thrust measured with a
the impeller outlet has an influence on the velocity four-stage boiler feed pump. The results of five different
and pressure distribution for radially inward leakage units lay within the scatter band of curve 1. The scatter
but negligible impact if leakage is radially outward. is due to unavoidable manufacturing and assembly tol-
At part load flow recirculation occurs at the impeller erances (presumably mainly labyrinth clearances). The
outlet. It starts mostly on the hub side where conse- predicted residual thrust for both clearances is also
quently C,, is much reduced. On the shroud side C,, given on Fig. 12 (prediction was based on prototype
increases, K increases accordingly and the axial tests according to the empirical method). The width of
thrust increases. The changing flow pattern at the the scatter band corresponds to about 8 per cent of the
impeller outlet at part load must therefore to a large hydraulic thrust at BEP and to about 11 per cent at 25
extent be responsible for the increase of axial thrust per cent load.
at part load (apart from the increase in leakage rate How some tolerances do affect the different axial
and head), any hysteresis or other unexpected forms forces is demonstrated by Table 5 for 25 per cent load.
of the axial thrust curve. The error of the residual thrust is roughly 15 times
6. The physical relationships described above are well greater than the error of the predicted force on the
demonstrated by test data given in reference (30). shroud (compare columns 2 and 4).
@ IMecbE 1987 Proc Instn Mech Engrs Vol 201 No A3
174 J GUELICH, W JUD AND S F HUGHES
A
/ Bandwidth of readings for 5 pumps
O ( I 1 I I - -