You are on page 1of 8

Suppression of Secondary Flows

in a Mixed-Flow Pump Impeller


by Application of Three-
M. Zangeneh
Department of Mechanical Engineering,
Dimensional Inverse Design
University College of London,
London, United Kingdom Method: Part 1—Design and
A. Goto Numerical Validation
This paper describes the design of the blade geometry of a medium specific speed
T. Takemura mixed flow pump impeller by using a three-dimensional inverse design method in
which the blade circulation (or rV$) is specified. The design objective is the reduction
Ebara Research Company, Limited, of impeller exit flow nonuniformity by reducing the secondary flows on the blade
Fujisawa-shi, Japan suction surface. The paper describes in detail the aerodynamic criteria used for the
suppression of secondary flows with reference to the loading distribution and blade
stacking condition used in the design. The flow through the designed impeller is
computed by Dawes' viscous code, which indicates that the secondary flows are well
suppressed on the suction surface. Comparison between the predicted exit flow field
of the inverse designed impeller and a corresponding conventional impeller indicates
that the suppression of secondary flows has resulted in substantial improvement in
the exit flow field. Experimental comparison of the flow fields inside and at exit from
the conventional and the inverse designed impeller is made in Part 2 of the paper.

Introduction this secondary flow can be minimized by using backsweep (now


commonly used), which reduces the tangential component of
The efficiency and stability of mixed and radial flow turboma-
Corriolis acceleration (and therefore the blade loading). The
chines are adversely affected by the presence of impeller exit
meridional component of secondary flow, which is mainly due
flow nonuniformity. In recent years, as a result of improvements
in experimental techniques and numerical methods, it has been to the combined action of radial component of Corriolis acceler-
possible to obtain a better understanding of the basic mechanism ation (2toWe) and streamline curvature on the blade surface
behind this phenomenon; see Eckardt (1976, 1980), Zangeneh boundary layers, can be observed on both the pressure and
et al. (1988), and Goto (1992a). It is now a well-established suction surfaces; see Goto (1992a) and Casey et al. (1992).
fact that the flow in these impellers is dominated by secondary The control and possible suppression of this meridional compo-
flows, which tend to move low-momentum fluids present on nent of secondary flow, however, are more difficult as it requires
the blade surface and endwall boundary layers toward the loca- careful design of the complicated three-dimensional blade ge-
tion of minimum reduced pressure (or relative Mach number ometry of the impeller.
in compressible flow) on the shroud suction surface corner. In this paper, the design of the blade geometry of a mixed
The generation of secondary flows in the impeller can be flow pump impeller is described in which meridional secondary
explained by reference to the following equation derived from flows have been suppressed. To suppress secondary flows, the
classical secondary flow theory (see Zangeneh et al, 1988): impeller blades were designed by using a three-dimensional
inverse design method, in which the blade circulation (or load-
w-v(w-o) = 2n-(w-v)w + n-(2w x w) (i) ing) distribution is specified; see Zangeneh (1991). In Part 1
of the paper we shall consider in detail the approach used to
where (W • ft) represents the streamwise component of absolute
suppress the secondary flows on the blades by reference to the
vorticity. According to this equation streamwise components of
vorticity (and therefore secondary flows) are generated when inputs of the design method. We shall also compare the flow
there is a component of acceleration (streamline curvature or field in the inverse designed impeller with a conventional impel-
Coriolis) in the direction of the absolute vorticity. As a result ler by using the incompressible version of Dawes' code (1988).
there are two types of secondary flow in the impellers of mixed In Part 2 of the paper results of detailed experimental measure-
flow and radial turbomachines. A blade-to-blade secondary flow ments for the conventional and inverse designed impeller will
(moving low-momentum fluid from pressure to suction surface) be presented.
is mainly due to the tangential component of Coriolis accelera-
tion (2uiWr), which is in the direction of vorticity generated by
endwall boundary layers. In the case of compressors and pumps, Inverse Design Method
In this design method the blades are represented by sheets
Contributed by the International Gas Turbine Institute and presented at the 39th of vorticity whose strength is determined by a specified distribu-
International Gas Turbine and Aeroengine Congress and Exposition, The Hague,
The Netherlands, June 13-16, 1994. Manuscript received by the International
tion of circumferentially averaged swirl velocity, or rV0 (di-
Gas Turbine Institute February 4, 1994. Paper No. 94-GT-45. Associate Technical rectly related to the bound circulation litrVg and the blade
Editor: E. M. Greitzer. loading) defined as:

536 / Vol. 118, JULY 1996 Transactions of the ASME


Copyright © 1996 by ASME
Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 01/18/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
Table 1 Design specification

"-££ rVed9,
Hub/Mid- span/S hroud Design flow rate: 9.2 m3/min
where B is the number of blades. To simplify the problem the Design Pump head: 6.5 1T1H2O
blades are represented by a single sheet of vorticity, representing ri (mm) 58.0/102.3/132.5
the mean camber line of the blades. The blade blockage effects X2 (mm) 142.5/157.8/171.7 Rotational Speed: 800 rpm
are then accounted for by using a mean streamsurface thickness
parameter in the continuity equation of the mean flow. It is
therefore possible to express the bound vorticity on the blades
by conditions on the flow is discussed in detail by Zangeneh (1992)
and will be discussed further later on in this paper.
Ub = V X V = (VrV0 X Va)8Ja), (2) Once a value for the blade shape has been obtained, it is then
possible to use Eqs. (2) and (3) to determine the vorticity
where throughout the blade region. This in turn can be used to compute
the velocity field throughout the flow; see Hawthorne et al.
-f(r, z) = n — , (3) (1984) and Zangeneh (1991). The flow field is decomposed
n
into tangentially mean and periodic components. A governing
represents the blade surfaces, 9 is the tangential coordinate of equation for the mean flow is obtained by using the Stokes
a cylindrical polar coordinate system, a n d / ( r , z) is the angular stream function together with tangential component of Eq. (4).
coordinate of the point on the thin blade surface, or the mean The governing equation of the periodic flow is obtained by
camber angle (wrap angle in turbocharger terminology). In Eq. applying the continuity equation of the periodic flow to the
(3), B denotes the number of blades and n is an integer, which Clebsch formulation for the velocity.
takes values between 1 and B. Sp(a) is the periodic delta func- The flow field is computed by solving the governing equa-
tion, whose pitchwise mean is unity, so that the tangentially tions of the mean and periodic flow (two-dimensional and three-
mean bound vorticity is given by dimensional Poisson's partial differential equations, respec-
tively ). For the solution of the governing equation of the peri-
fl = V X V = (VrV 9 X V a ) . (4) odic flow it is possible to express all three-dimensional flow
variables in terms of a Fourier series in the tangential direction.
The method starts by using one-dimensional estimates of the As a result the three-dimensional Poisson's equation is reduced
meridional velocity obtained from the specified mass flow rate to a two-dimensional Helmholtz equation, which is then solved
and meridional geometry together with the specified mean tan- for each hamonic of the Fourier coefficients in the frequency
gential velocity to compute an initial blade shape. The blade domain. The use of a Fourier expansion in the tangential direc-
shape is computed by using the inviscid slip condition, which tion results in substantial savings in computational time and
implies that the flow must be aligned to the blade surfaces. This memory as the three-dimensional flow field is solved without
condition can be expressed mathematically as: the need for a grid in the tangential direction.
W„,-Va = 0. (5) The partial differential equations for the flow field are solved
numerically by using a finite difference approximation on a
where Va is a vector normal to the blade surface and W« is body-fitted coordinate system; see Zangeneh (1991) for more
the relative velocity at the blade surface (W w = (W + + W " ) / detail. Equation (6) is also solved numerically by using a
2, where W + and W are the velocities on the upper and lower Crank-Nicholson type finite difference discretization.
surface of the blades). Expanding Eq. (5) we obtain Once a new estimate for the velocity field has been obtained,
it is then possible to compute the new blade shape by using Eq.
Q± + Vrl df_ (6) (6). This iterative process is then repeated until changes in
-2
dz dr blade shapes between two iterations fall below a certain given
tolerance, usually taken as 10~5 radians.
where / i s the wrap angle and w is the rotational speed. Equation
(6) is a first-order hyperbolic partial differential equation, which
has to be integrated along the meridional projections of stream- Impeller Design
lines on the blade surface in order to find the blade shape. The The design conditions used for the design of the mixed flow
integration, as in the case of other initial value problems, cannot pump were based on the medium specific speed impeller, which
be completed without some initial condition on / . This initial has been studied extensively by Goto (1992a, b). The basic
value will be called the stacking condition of the blade. In this design conditions for the impeller are presented in Table 1.
method the stacking condition is implemented by giving as This impeller was designed by using conventional techniques
input, the values of blade wrap angle /along a quasi-orthogonal, involving the use of curve fits to connect the blade angles
for example at the leading edge. The effect of different stacking smoothly between the leading and trailing edges.

Nomenclature
B = number of blades a = angular coordinate of blade sur- 9 = tangential component
/ = blade wrap angle (9 value at faces (Eq. (3))
the blade) Sp(a) = periodic delta function Superscripts
/• = radius LO = rotational speed = pitchwise mean value
(r, 9, z) = cylindrical-polar coordinate SI = vorticity + = relative to upper blade surface;
system suction surface in pump
S(a) = sawtooth function Subscripts — = relative to lower blade surface;
v = periodic velocity bl = at the blade pressure surface in pump
V = velocity r = radial component
W = relative velocity z = axial component (three-
dimensional)

Journal of Turbomachinery JULY 1996, Vol. 1 1 8 / 5 3 7

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 01/18/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


i—i—i—i—i—i—i—i—i—i—I—'—i—•-—r
0.0 0.1 0.2 0.3 0.4 0.3 0.0 0.7 0.B 0.0

Fig. 1 Computational grid on meridional plane; grid size: 145 x 49


Percentage Meridional Distance
Fig. 3 Meridional derivative of rVe
The computational mesh used to represent the meridional
geometry of the pump impeller is presented in Fig. 1. This no-incidence and the Kutta conditions. Equation (7) also indi-
grid consists of 145 quasi-orthogonals and 49 uniformly spaced cates that the pressure distribution onjhe blades is directly
quasi-streamlines, with 30 quasi-orthogonals in the upstream related to the meridional derivative of rVe (or loading distribu-
region and 81 quasi-orthogonals inside the blade region. The tion) and by using a smooth loading distribution we should be
meridional geometry of the impeller is the same as the conven- able to ensure a smooth pressure distribution.
tional impeller, apart from the curved leading edge, which in The easiest way to satisfy all these criteria is to specify the
this case has been replaced by a straight line connecting the loading distribution rather than the rVe distribution. This is
hub and shroud. As the tangential variations in flow variables done by assuming the following distribution for the meridional
are represented by a Fourier series, no blade-to-blade grid is derivative of rVe:
required. For the thickness distribution, the normal thickness of
the conventional impeller was specified.
rVe
Specified Circulation Distribution. The specified circula-
tion (or rVe) distribution is presented in Fig. 2. A number of
important criteria were used to arrive at this rV% distribution.
The basic approach used is to select an optimum distribution
of rVn at the hub and shroud and then interpolate linearly be-
tween these values to obtain the overall distribution of rVg on
the meridional plane. The values of rVs at the leading and
trailing edges are obtained from Euler's pump equation. In this
case the nondimensional (by rmUm) exit value of rVe was set
to 0.437 while the inlet rVe was set to zero. Since the jump in
pressure across the blades is given by:
2TT
• P = (WwVrVi) (7)
B
the derivative of rVe in the meridional direction, at the leading
and trailing edges, must be set to zero in order to satisfy the NC ND
Percentage Meridional distance

where the distribution is parabolic up to point NC, then a linear


variation between point NC and ND with a specified slope,
followed by another parabolic distribution, which reduces the
loading to zero at the trailing edge (point NB) in order to
satisfy the Kutta condition. The rVt distribution is then found
by integrating these values of d{rVe) I dm. The specified loading
distribution used for the design is presented in Fig. 3. In this
case, the location of maximum loading at the hub and shroud
was moved away from the leading edge in order to minimize
possible cavitation problems.
Stacking Condition. Apart from the loading (or r Ve) distri-
bution, there is one other input specification, which has a very
important effect on the flow field. This is the stacking condition
mentioned earlier. To obtain radial stacking, the values of / ( o r
mean camber angle) can be set to a constant value (e.g., zero)
on the stacking quasi-orthogonal. However, it is also possible
Fig. 2 Specified rV„ distribution (contour interval = 0.05, normalized by
to specify a nonzero value for / in which the values of / at the
stacking quasi-orthogonal vary from hub to shroud. This type of

538 / Vol. 118, JULY 1996 Transactions of the ASME

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 01/18/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Solid Line
(a) Impeller A Dashed Line Impeller CII
Fig. 4 Contours of wrap angle f (contour interval = 0.25 rad)

stacking condition is equivalent to lean in axial turbomachinery impeller C) is radically different from the conventional
terminology, which results in a spanwise force that affects the impeller.
spanwise pressure distribution along the blade surfaces. The effect of the stacking condition used to design impeller
It is a well-known fact that secondary flows move low-mo- A is to produce a spanwise force, which increases the blade
mentum fluids toward the location of minimum reduced static loading at the tip (as compared to the corresponding radial
pressure (see Johnson, 1978). As a result of this fact the pres- stacking condition) at the expense of the loading at the hub;
ence of meridional secondary flows can be directly correlated while the effect of the stacking condition used for impeller C
with spanwise gradients of reduced static pressure (or relative is to reduce the blade loading at the tip (near the trailing edge)
Mach number); see Zangeneh et al. (1988). Therefore, to mini- and increase the loading at the hub. The effect of these different
mize meridional secondary flows, particularly on the suction stacking conditions on the pressure distribution can be seen
surface, where most of the low-momentum fluids are generated, clearly in Fig. 5, which compares the reduced pressure distribu-
our main objective was to reduce the spanwise gradients of tion on the hub and shroud for impellers A (Fig. 5(a)) and C
reduced static pressure. As the choice of loading distribution (Fig. 5(b)), as predicted by the inverse design method. By
was to some extent limited by the cavitation reduction criteria, comparing the reduced pressure distributions on the suction
we decided to use the stacking condition in order to control the surface of impellers A and C, it is possible to see that hub to
spanwise gradients of reduced static pressure. shroud difference in reduced pressure (and therefore the span-
In order to investigate the effect of the stacking conditions wise gradients of reduced pressure) on the suction surface have
on the impeller flow field two different impellers with the same been reduced considerably in impeller C as opposed to that on
loading distribution (as in Fig. 3) but different stacking condi- impeller A. The corresponding reduced static pressure distribu-
tions were designed. In the first impeller design (to be desig- tion for the conventional impeller is presented in Fig. 5(c).
nated impeller A) the stacking quasi-orthogonal was set at the The pressure distribution on the conventional impeller was com-
trailing edge. The stacking values of / w e r e set to vary linearly puted by using a three-dimensional finite element potential
from 0 rad at the hub to 0.15 rad at the tip. In the second impeller method; see Daiguji (1982). The hub-to-shroud difference in
design (to be designated impeller C) the stacking condition was reduced pressure (and therefore the spanwise gradients of re-
specified at quasi-orthogonal 78 (approximately 96 percent of duced pressure) on the suction surface in this case are consider-
meridional chord) rather than the trailing edge. In this case the ably higher than that in impellers A and C, particularly near
stacking values of / were set to vary linearly from 0.5 rad at the trailing edge. This is partly due to the fact that this impeller
the hub to 0 rad at the tip. To achieve further reduction in has five blades rather than seven blades used on the inverse
secondary flows, both impellers were designed with seven design impellers. However, and perhaps more importantly, the
blades as opposed to the five blades used on the conventional blade angle distribution used to design this impeller has resulted
impeller. The computed distributions of / for both impellers A in a leaned geometry near the trailing edge in the opposite
and C are presented in Fig. 4. The absolute difference between direction to that used to design impeller C—see Fig. 9 and
blade angles for the inverse designed impellers and the conven- Fig. 2 of Goto et al. (1994) —thereby increasing the spanwise
tional impeller at a few chordwise locations are presented in gradients of reduced static pressure.
Table 2. From the results in Table 2 we can see clearly that To confirm that the reduction in spanwise pressure gradients
the geometry of the inverse designed impellers (particularly in the case of impeller C as compared to the conventional impel-
ler does in fact help to diminish (or perhaps even eliminate)
secondary flows, the flow through the inverse designed impel-
Table 2 Absolute difference in blade angles = | /3i„ v „, e d.sig„ - /30on»ontion.i I lers A and C was computed by using a three-dimensional viscous
method.
Impeller 8% Chord 50%-Chord 92%-Chord
Shroud/Hub Shroud/Hub Shroud/Hub
A 1.4/2.5 7.2/3.3 5.1/8.1 Viscous Predictions
C 1.3/2.6 7.9/0.4 10.7/11.7 The flow fields in the two inverse designed impellers were
analyzed by the incompressible version of three-dimensional
ClI 1.7/2.4 7.3/1.6 10.3/17.8
Navier-Stokes solver developed by Dawes (1988) and Walker
and Dawes (1990). For both calculations a computational mesh
(degree) consisting of 29 pitchwise, 129 streamwise (with 21 points

Journal of Turbomachinery JULY 1996, Vol. 1 1 8 / 5 3 9

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 01/18/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


upstream of the blade and 97 points inside the blade region) Table 3 Predicted performance by Dawes code
and 29 spanwise grid points were used. The mesh points near
Impeller Euler Head Efficiency
the blade surfaces and endwalls were clustered in order to re- Difference %
solve the shear layers. Analysis of the flow field inside the Conventional 0.447 0.0
conventional impeller has already been reported in some detail
by Goto (1992a). The results reported by Goto (1992a) indicate A (7 blades) 0.355 -6.9
C (7 blades) 0.359 -7.1
CII (7blades) 0.401 -3.9
CII (5 blades) 0.392 -0.8
CIII (7 blades) 0.413 -2.2

clearly that the N - S method can predict the Euler head quite
accurately, although the actual head predicted by the method
is considerably higher than the measured head as the method
underestimates the amount of mixing inside and at the exit
from the impeller. The results also indicate that the method can
provide a good qualitative picture of the flow field at the exit
from the impeller. For easy comparison with the inverse de-
signed impellers, the flow field in the conventional impeller was
also analyzed by using a similar 29 X 129 X 29 computational
Percentage Meridional Distance mesh. All computations were performed for zero tip clearance.
(a) Impeller A Overall Performance. A summary of the overall perfor-
. , i — i — i — . — i — i — i I _ J . i . i , ' • ' • mance predictions for all three impellers is presented in Table
3. The predicted Euler head for both inverse designed impellers
is about 18 percent lower than the specified Euler head of 0.437
used to design the impellers. At the time and due to budgetary
time constraints, the investigation of the likely causes of the
lower Euler head was not possible and therefore to account to
some extent for the lower Euler head a new impeller was de-
signed with the same loading distribution and stacking condition
as impeller C, but with a higher nondimensional design Euler
head of 0.485. This was the impeller (to be called impeller CII)
that was manufactured and used for the experimental study
reported in Part 2 of the paper. The wrap angle distribution for
impeller CII is presented (by the dashed line) in Fig. 4(b) and
the corresponding reduced pressure distribution is presented in
Fig. 5(b). The flow through this impeller was also computed
by using the viscous code using the same computational mesh
as that used for impellers A and C. The overall performance
predictions for impeller CII are also presented in Table 3.
Percentage Meridional Distance
The efficiency presented in Table 3 was calculated by divid-
(b) Solid Line - Impeller C ing the mass-averaged head by the mass-averaged Euler head
Dashed Line - Impeller CII at the trailing edge plane of each impeller and therefore does
•• 1 • 1 1 l _ i 1 , I i I r__L
not include the effect of mixing losses occurring behind the
impeller trailing edge. The predicted efficiency for impellers A
and C is about 7 percent less than that of the conventional
impeller, while the efficiency of impeller CII is only 4 percent
less than the conventional. The main cause of the lower pre-
dicted efficiency of the inverse designed impellers was found
to be due to two main reasons. First, the inverse designed impel-
lers were designed with seven blades rather than five for the
conventional impeller (which later work showed to be the opti-
mum blade number for this configuration). Second, the inverse
designed impellers were designed with the same normal thick-
ness as that used for the conventional impeller. Due to the
difference in blade angles between the conventional and inverse
designed impellers, the specification of the normal thickness
resulted in higher tangential thickness particularly at the tip of
the inverse designed impeller (see Fig. 8). The combined effect
of higher blade number and tangential thickness is to increase
Percentage Meridional Distance the relative velocity in the blade passage and therefore losses.
(c) Conventional Impeller The higher losses may also be attributed to the lower predicted
Euler head in the inverse designed impellers. The lower Euler
Fig. S Reduced static pressure distribution (Cp = (Rothalpy - Reduced
static pressure)/0.5/)L/im) head implies slower rate of diffusion through the impeller and

540 / Vol. 118, JULY 1996 Transactions of the ASME

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 01/18/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


therefore higher relative velocity (and as a result losses) at
corresponding points in the impellers. The 3 percent improve-
ment in efficiency of impeller CII over that predicted for impel-
ler C is mainly due to the effect of increased diffusion rate in
impeller CII. The flow through impeller CII was also computed
using five blades instead of the seven that the impeller was
designed for and the results in Table 3 indicate clearly the
improvement in efficiency as a result of using five blades. The
results of flow analysis for impeller CII with five blades are
discussed in detail in Part 2 of the paper (Goto et al., 1994).
In order to investigate the causes of the lower predicted Euler
head for the inverse designed impellers, a number of studies
were undertaken. At first glance, the effect of hydrodynamic
blockage on the impeller flow field was thought to be insignifi-
cant, as the boundary layers are thin (because of higher Reyn-
olds numbers as compared to centrifugal compressors) and the
flow is not separated. So in order to establish the likely cause
of the lower Euler head, the flow through impeller CII was
computed by using Denton's (1983) inviscid time marching
code. As the convergence of this code slows down at low Mach
numbers the computation was performed at 10,000 rpm, with
scaled velocity triangles. The computed mass-averaged exit rVe
was found to be 0.4993, which is slightly higher than the speci-
fied value. As this computation was performed at a higher rota-
tional speed, the density increases toward the trailing edge as
a result of the compressibility effects. The effect of increase in
density on the predicted Euler head has been investigated by
using the compressible and incompressible version of Dawes
code, thereby relating the ratio of compressible Euler head over
incompressible Euler head to the density ratios between the
trailing edge and the inlet. In this case the density ratio was
1.086, which indicated that the ratio of Euler heads should be
about 1.029. Dividing the predicted Euler head (0.4993) by this
ratio (to allow for the change in volume flow rate) will give
0.485, which is the exact value specified to design impeller CII.
From this result we can therefore conclude that the predicted
lower Euler head is due to the hydrodynamic blockage effect,
which was not accounted for in the design.
In order to investigate the effect of hydrodynamic blockage
on the blade shape, a hydrodynamic blockage distribution was
assumed. This distribution was arrived at by assuming a linear
variation in blockage from zero at the leading edge to 15 percent
at the trailing edge and then linearly dropping to zero at the
far downstream boundary. The same blockage distribution was
applied in the spanwise direction. Using this distribution impel-
ler C was redesigned. The flow through this impeller (to be
designated impeller CHI) was computed by using the viscous
code and the results of the overall performance predictions are Fig. 6 Velocity vectors predicted by Dawes code near the pressure
surface
presented in Table 3. The results indicate clearly that by assum-
ing this aerodynamic blockage distribution the exit Euler head
has now increased to 0.413 (from 0.359) and the predicted
efficiency has increased by 5 percent. It must, however, be the pressure surface, but the intensity of the secondary velocities
emphasized that the above-mentioned distribution of aerody- on the suction surface has been markedly reduced as compared
namic blockage is based on empiricism and a more systematic to the conventional impeller. In the case of impeller C, the
approach is required in order to account for the blockage effects. secondary flow on the pressure surface is very similar to that
A viscous/inviscid interaction method such as that described observed for impeller A. On the suction surface, however, very
by Zangeneh (1994) should provide a more accurate means of little secondary flow can be observed. The secondary flows on
accounting for viscous effects in the impeller. impeller CII are identical to that observed for impeller C. The
results presented here provide further confirmation of the close
Internal Flow Predictions. The predicted velocity vectors correlation between gradients of reduced static pressure and the
on the pressure and suction surfaces of the conventional impeller presence of meridional secondary flows, as pointed out earlier
and impellers A, C, and CII are presented in Figs. 6 and 7. in the paper.
Strong secondary flows can be observed on both the pressure The development of the relative velocity field in the three
and suction surfaces of the conventional impeller, with the span- impellers is compared in Figs. 8 and 9. Figure 8 presents the
wise movement of low-momentum fluids starting at about 25 relative velocity contours at 50 percent chord. At this stage no
percent of meridional chord on both surfaces. As a result of the distinct wake region can be observed in any of the impellers.
secondary flows low-momentum fluid is accumulated on the One interesting feature of the results is the higher values of
shroud, creating the recirculation region, which can be clearly relative velocity predicted for impellers A and C and to a lesser
observed there, downstream of the trailing edge. In the case of extent CII. As explained above this is mainly due to the com-
impeller A, a relatively strong secondary flow can be seen on bined effect of higher blade numbers and tangential thickness.

Journal of Turbomachinery JULY 1996, VoL 118/541

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 01/18/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


same circulation distribution, but different stacking condition.
The flow through the inverse designed impellers was computed
by using a three-dimensional viscous code. The viscous predic-
tions showed clearly that it is possible to control meridional
secondary flows particularly on the suction surface, where
most of the low-momentum fluid in the impeller is generated,
by suitable choice of the inputs to the inverse design program.
In particular it was shown that the choice of stacking condition
and its location can have a very significant effect on the blade
reduced static pressure, whose spanwise gradients are directly
correlated to meridional secondary flows in the impeller. It
was also shown that the suppression of secondary flows on
the suction surface results in a uniform exit flow field from
the impeller, which should help to improve the stability range
of the pump as well as the performance of the diffuser down-
stream of the impeller.
The performance predictions indicated that it is necessary to
account for hydrodynamic blockage effects in the impeller in
order to achieve the specified Euler head when the inverse

Fig. 7 Velocity vectors predicted by Dawes code near the suction


surface

As a result of the higher overall relative velocity more low-


momentum fluid has been generated on the suction surface of
the three inverse designed impellers. The resulting relative ve-
locity distributions at the exit of the three impellers are com-
pared in Fig. 9. In the case of the conventional impeller a
distinct wake region can be observed at the shroud. The exit flow
from impeller CII, however, shows very little accumulation of
low-momentum fluids despite the presence of higher amount of
low-momentum fluid on the suction surface. This shows clearly
that the suppression of secondary flows on the suction surface
of the inverse designed impeller CII has resulted in marked
improvement in the uniformity of the impeller exit flow.

Conclusion
The impeller geometry of a medium specific speed mixed
flow pump was designed by using the three-dimensional in-
verse design method of Zangeneh (1991), in which the circu-
lation distribution is specified. The main objective of the de-
sign was to suppress secondary flows in order to obtain a
uniform exit flow field from the impeller. Two different impel-
lers were designed using the inverse design method with the Fig. 8 Relative velocity contours—50 percent chord (contour interval
= 0.05, normalized by U&n)

542 / Vol. 118, JULY 1996 Transactions of the ASME

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 01/18/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


design method is applied to incompressible turbomachines such
as pumps. A systematic way of accounting for viscous effects in
the impeller is to use the viscous/inviscid interaction approach
reported by Zangeneh (1993). Work is currently under way to
extend this method to incompressible flow.

Acknowledgments
The authors would like to thank Ebara Research Co. Ltd. for
permission to publish this paper.

References
Casey, M. V., Dalbert, P., and Roth, P., 1992, "The Use of 3D Viscous Flow
Calculations in the Design and Analysis of Industrial Centrifugal Compressors,"
ASME JOURNAL OF TURBOMACHINERY, Vol. 114, pp. 27-37.
Daiguji, H., 1982, "A Finite Element Method for Analysing the 3D Flow in
Turbomachinery," Proc. 4th International Symposium on Finite Element Methods
in Flow Problems, University of Tokyo Press, pp. 751-758.
Dawes, W. N„ 1988, "The Development of a 3D Navier-Stokes Solver
for Application to all Types of Turbomachinery," ASME Paper No.
88-GT-70.
Denton, J. D., 1983, "An Improved Time Marching Method for Turboma-
(b) Impeller A chinery Flow Calculations," ASME Journal of Engineering for Power, Vol.
105, p. 514.
Eckardt, D., 1976, "Detailed Flow Investigation Within a High Speed Centrifu-
0.55 gal Compressor Impeller," ASME Journal of Fluids Engineering, Vol. 98, pp.
390-402.
Eckardt, D., 1980, "Flow Field Analysis of Radial and Backswept Centrifugal
Compressor Impellers—Part I. Flow Measurements Using a Laser Velocimeter,"
Performance Prediction of Centrifugal Pumps and Compressors, Gopalakrishnan,
ed., ASME, New York, pp. 77-86.
Goto, A., 1992a, "Study of Internal Flow in a Mixed Flow Pump Impeller at
Various Tip Clearances Using 3D Viscous Flow Calculations,'' ASME JOURNAL
OF TURBOMACHINERY, Vol. 114, pp. 373-382.
Goto, A., 1992b, "The Effect of Tip Leakage Flow on Part-Load Performance
of a Mixed-Flow Pump Impeller," ASME JOURNAL OF TURBOMACHINERY, Vol.
(c) Impeller C 114, pp. 383-391.
Goto, A., Takemura, T., and Zangeneh, M., 1994, "Suppression of Secondary
Flows in a Mixed Flow Pump Impeller by Application of Three-Dimensional
0.55 Inverse Design Method: Part 2—Experimental Validation," ASME Paper No.
94-GT-46: ASME JOURNAL OF TURBOMACHINERY, Vol. 118, 1996, this issue, pp.
544-551.
Johnson, M. W., 1978, "Secondary Flows in Rotating Bends," ASME Journal
of Engineering for Power, Vol. 100, pp. 553-560.
Walker, P. J., and Dawes, W. N., 1990, "The Extension and Application of
Three-Dimensional Time-Marching Analysis to Incompressible Turbomachinery
Flows," ASME JOURNAL OF TURBOMACHINERY, Vol. 112, pp. 385-390.
Zangeneh, M., Dawes, W. N., and Hawthorne, W. R., 1988, "Three-Di-
mensional Flow in Radial-Inflow Turbines," ASME Paper No. 88-GT-103.
Zangeneh, M., 1991, "A Compressible Three Dimensional Blade Design
(d) Impeller CII Method for Radial and Mixed Flow Turbomachinery Blades," Int. J. Numerical
Methods in Fluids, Vol. 13, pp. 599-624.
Zangeneh, M., 1992, "An Inverse Design Method for Radial Turbomachines,"
VKI Lecture series, Radial Turbines, 1992-5, Belgium.
Zangeneh, M., 1994, "Inviscid/Viscous Interaction Method for Three-Dimen-
Fig. 9 Relative velocity contours—trailing edge (contour interval = 0.05, sional Inverse Design of Centrifugal Impellers," ASME JOURNAL OF TURBOMA-
normalized by Um,) CHINERY, Vol. 116, pp. 280-291.

Journal of Turbomachinery JULY 1996, Vol. 118/543

Downloaded From: http://turbomachinery.asmedigitalcollection.asme.org/ on 01/18/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

You might also like