THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS 93-GT-102

345 E. 47th St, New York, N.Y. 10017

The Society shall not be responsible for statements or opinions advanced in
papers or discussion at meetings of the Society or of its Divisions or Sections,
or printed in its publications. Discussion is printed only if the paper is pub-
lished in an ASME Journal. Papers are available from ASME for 15 months
after the meeting.
Printed in U.S.A.
Copyright © 1993 by ASME

BLADING DESIGN PROCEDURE FOR CENTRIFUGAL IMPELLERS

Sarim N. Al-Zubaidy
Department of Mechanical Engineering
United Arab Emirates University
Al-Ain, United Arab Emirates

ABSTRACT blade angle, relative flow angle

0 relative coordinate
The paper proposes a direct method for the aero- a
angle between absolute velocity components
thermodynamic design of centrifugal flow impellers (based on a
prescribed space rate of diffusion) that can be used to aid the design Subscripts
methodology of radial impellers. The design procedure employs the
basic laws of fluid motion to determine the required flow path and 0

stagnation condition
thus the detailed passage coordinates. Numerical optimization
impeller inlet
1
techniques were used to control the manner by which the relative
2 impeller exit
velocity components were allowed to vary so that a simple loss
bb blade to blade
criteria (entropy generation) was kept minimum throghout the
curvature
solution.
e inducer tip condition
The development of the method was undertaken with the idea of
h hydraulic
obtaining a self sufficient design method which would be influenced meridional component, meridional
m
as little as possible by the preconceived notions of what the optimum pressure side
p
impeller blading should look like. radial
r
suction side
NOMENCLATTTRE:
tangential

q tangential
b tip width
z axial
C absolute velocity

Cd dissipation coefficient
INTRODUCTION
Cn laoding coefficient

d diameter As a result of continuing interest in the development of highly

e inducer to tip diameter ratio efficient centrifugal compressors for variety of applications, many

F(x) objective function (defined in equation 15) studies on the exact nature of flow in impellers and other components
unit reactor have been reported in the last three decades, and considerable
path length
m progress has been made in refining the analysis models. However,
inducer tip Mach number
Me the actual design of centrifugal compressor stages is still far from
Pressure being clear cut. Open literature dealing with the methodology of

r radius design, particularly from industrial sources is still very scarce.

S entropy Wallace (1975) and Whitfield (1978) proposed systematic design

T temperature procedures for impellers. In these methods, the principal dimensions

U tangential velocity would be obtained by using one-dimensional calculations, and the
relative flow velocity hub, shroud and blade profiles would be defined analytically, e.g.
independent design variables Lame Ovals (Wallace (1975)). This preliminary shape would then be
number of baldes, axial length of impeller. analyzed with the help of a three-dimensional (or quasi-three-

p density dimensional) flow analysis program. The hub, shroud and blade
rotational speed profiles would be modified successively until the velocity

Presented at the International Gas Turbine and Aeroengine Congress and Exposition
Cincinnati, Ohio — May 24-27, 1993

Downloaded From: https://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/asmep/82379/ on 04/07/2017 Terms of Use: http://www.asme.org/ab

 distributions in the two main planes are satisfactory. Came (1978)
also described a method for defining the hub, shroud and blade
surfaces analytically, to aid the flow analysis, and subsequently
manufacture by numerically controlled machines. The procedure for
the aero-thermodynamic design was similar to that described above.
From the forgoing, the following problem area can be directly
identified: For a given set of performance requirements, i.e. mass flow
rate, pressure ratio and in some cases rotational speed, several
impeller shapes may be drawn which would appear to meet the
specifications. Figure 1 shows four groups of impellers which have
been designed for zero pre-whirl, same axial velocity at inlet, and
identical velocity triangles at outlet. Impellers in group 1 have the
same tip diameter but the hub diameter, inducer diameter and blade
angle from hub to shroud are varied in such a manner that the
projected inlet area and the axial velocity are constant. In the case of
impellers shown in group 2, the inducer and hub diameters are held
constant. The tip diameter and rotational speed are changed, so that
the tip velocity is also constant. The blade angle at inlet will have
to be adjusted to achieve a constant axial velocity. The principal
dimensions of group 3 impellers are identical, but the blade curvature
in the axial direction is changed in accordance with prescribed
schedules, e.g. circular, parabolic, elliptical etc.. Impellers in group 4
have identical overall dimensions, except the axial length which is
different in each case. From the point of view of simple one-
dimensional considerations, the twelve impellers should meet the
same design requirements. However, in practice, it would be
surprising indeed if they reached the same performance levels.
In this paper an impeller design method is proposed which can
be used to aid detailed geometry optimization of centrifugal
impellers (for a specified performance requirements) from an initially
prescribed deceleration schedule along the mean streamline.
PRELIMINARY AERODYNAMIC CALCULATION
The region in which the impeller principal dimension should
be selected from was first accomplished by solving the basic governing
equations at the impeller inlet and outlet, assuming the flow to be
steady, one-dimensional and the working fluid to be a perfect gas.
Figure 2 , shows an impeller design chart from which the practical
range of variation of inlet and outlet geometry and flow condition
0. 056
0 OM 0.55 0.60
0.45
INDUCER TO TIP DIAMETER RATIO- de/d 2
FIG.2. DESIGN CHART FOR THE CHOICE
OF PRINCIPAL DIMENSIONS.
could be easily established. An optimization process was then
carried out by means of a simple one-dimensional procedure.
The procedue used is based on the assumption that the flow
inside the impeller is controlled by the maximum achievable
diffusion which for a given geometry defines the separation point
location. Downstream of this point the flow splits into a jet and
wake. It is further assumed that the jet flow has a constant Mach
number in the flow direction and the only losses are due to friction
along the passage walls. The wake static pressure is fixed by a
tangential equilibrium between the jet and wake taking into account
centrifugal and coriolis forces. This non-unformity of the relative
velocity at the impeller exit gives rise to mixing losses at the diffuser
system inlet. Clearance losses are assumed to increase linearly with
relative clearance. During the calculation procedure the impeller
efficiency is closely related to the impeller geometry which allows
examination of the influence of change in dimensions. A modified
computer program which incorporated the above mentioned
procedure, that was first proposed by Frigne (1978), was used to
evaluate the influence of variation in the impeller outlet width
which allows the definition of the optimum value of tip width to tip

CONSTANT CONSTANT INLET VELOCITY OUTLET VELOCITY

GROUP MERIDIONAL PROFILE DESIGN FEATURES DESIGN FEATURES TRIANGLE-ZERO SWIRL TRIANGLE-CONSTANT SLIP

I fol if) TIP VELOCITY, HUB DIAMETER,

TIP DIAMETER, EYE DIAMETER,
A
? 44 sC,
TIP WIDTH, BLADE INLET 1

ir AC,, AXIAL LENGTH, ANGLE. Ci

INLET AREA. i al
• uic Us UIC U2
TIP VELOCITY, ROTATIONAL SPEED,

.41A
INLET AREA,
A; TIP DIAMETER,
B j, AXIAL LENGTH. TIP WIDTH, Cz

Mitj
el W
BLADE INLET
W2
ANGLE. 2
_....A Oa 4101 1
iv or ilv U1F UIE UID U2
TIP VELOCITY, BLADE-TO-BLADE

C 4,,,/
Ji _AA., Ng, ,,
, a
TIP DIAMETER,
TIP WIDTH,
INLET AREA.
BLADE INLET
PROFILE IN THE
INDUCER SECTION.

W2
Cz

, ■.7 A • - r -, ANGLE,
AXIAL LENGTH.
1

I I U7

iv ;
TIP VELOCITY,

Ai t; TIP DIAMETER,

A
$ TIP WIDTH, AXIAL LENGTH.
D P INLET AREA.
BLADE INLET
AI 1r ANGLE. U2
1

FIG.1. IMPELLER CONFIGURATION FOR THE SAME

NOMINAL DESIGN REQUIREMENTS.

Downloaded From: https://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/asmep/82379/ on 04/07/2017 Terms of Use: http://www.asme.org/ab

 diameter ratio of 0.079 (for a specified total - to - total pressure ratio
of 6:1 and a mass flow rate of 1 kg/s).
Lower values of the tip width ratio will result in lower separation
and mixing losses but higher friction and clearance losses. Larger
values will result in less friction losses for the impeller but early flow
separation and higher mixing losses will most likely result.

PRESCRIBED RELATIVE

The relative velocity vector that corresponds to the root mean

square diameter at the impeller inlet could be resolved into three
basic components. In this study, the relative inlet velocity was
assumed to have no component in the radial direction and that the
axial component was maximum. At outlet, the radial component of
relative velocity will attain its maximum value while the axial 0.2 0.4 0.6 0.8
component was assumed to diminish. The mean flow line is defined as PERCENT AXIAL LENGTH
the locus of points joining the root mean square of the radius at inlet FIG. 4. PRESCRIBED VELOCITY AND
with the point at outlet radius that bisect the impeller tip width. ASSUMED COMPONENT ALONG AXIAL
The detailed description of the mean streamline can now be obtained LENGTH
by prescribing desired velocity schedules between the inlet and
outlet stations. The general shape of the velocity vector diagram at 2irr
(P -P )brdm=p W (rC )dm (1)
any point along this line together with the relevant angles are shown P s m b Z dm U
in Figure 3. Overall impact of changes to relative velocity schedules
on the meridional view may be found in AL-Zubaidy (1988).
Simplifying;
The diffusion schedule of the relative velocity is known to be a
dominant factor in determining how efficiently the compression
2rt d (2)
process is being performed since most of impeller internal losses are
(P P -Ps)=E W sin 13 Z 711 (rc u) -
directly dependent on the nature of change. In reference to the jet and
wake model of Johnston (1966) it was decided to prescribe the
Assuming that the velocity varies linearly between the suction and
distribution shown in Figure 4 for the example design. Also shown on
pressure surfaces of the blade (Figure 5), then it can be written:
that Figure are the initial component distributions which are
mainly used to initialize computing procedure and were
S 2W=W +W p (3)
progressively modified during the the execution of the program .

Introducing the energy equation between suction and pressure surfaces

yields:

(P
p-PS )=
)= (W2s - W2P )=13W(W s - Wp ) (4)

Elimination of (P -P s) from (1) and (2) gives:

(W - W )= sin j3 2rc —
d (r C u )
s p Z m u

Defining C u = cor-W u yields:

(ws _wp) = sin 5 2zrc an 203 -rW cos 5) (5)

d (r

Upon simplification, the following general expression will result:

(wp _wp ) sin 27cr (20) dr 4. w sin § cos fi –cos w d r )

Z dm dm dm r an
ESTIMATION OF SURFACE VELOCITIES (6)

The blade surface velocities (both on pressure and suction
sides) could be evaluated using a relatively simple method described
as follows:
The flow is assumed to be axisymmetric and on the surface of
revolution in the direction of the flow varies smoothly from the inlet
relative flow angle to the impeller throat angle. Using this
assumption , the blade loading could be related to the change of
angular momentum as follows:
The above equation coupled with equation (3) can be used to calculate
the surface velocities in a step by step manner (along the mean flow
path) if the variation of relative velocity and blade angle along the
mean streamline are known. It should be mentioned here that the
meridional mean stream line is taken in the present study as located
in the middle of the flow channel. It lies in the center between two
blade elements and on the root mean square section between hub and
shroud. The flow velocity along this line is mass averaged.

Downloaded From: https://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/asmep/82379/ on 04/07/2017 Terms of Use: http://www.asme.org/ab

 form of discreet points, it was necessary to perform smoothing so that
the design procedure could be computerized and also to be able to
prepare instructions for numerically controlled machines. The
smoothing was accomplished in the following manner.
Using the conventional notation and geometry (Figures 6a and
6b) the first smoothing was done to r =f(Z), then values of a = tan-1
(dr/dZ) were numerically calculated. After applying a second
smoothing to a = f(m(Z)), the curvature of the meridional streamline
was calculated as 1/r, = da /dm. Similarly in the blade to blade
surface, the first smoothing was done to s=f(m) and then (1 /cosj3)
=ds/dm was calculated by taking the derivaitive numerically. The
FIG. 5. FLOW MODEL USED
mean streamline in the blade-to-blade surface can be found from 1/r
cbb = clPids= (c1f1/dm) cos 13. Smoothing in every instant (which helps
DESIGN THEORY speed up convergence) was accomplished by cubic spline fitting which
was provided in a separate subroutine.
The three equations of equilibrium of a compressible, ideal The differential pressure rise along the mean streamline could be
fluid in terms of relative cylindrical coordinates neglicting all blade written as follows:
body forces and assuming an axial symmetry, are written as follows:
ap ap aP
dz +— (10
W2 riP=dr
— dr+ dz
— dB
Wr 2
r =- 1 aP
(7) ap sp d.0
dr +ap — dz +— (12)
e p ar
— . -
d t r dr dz 88

1—
— d (rW ) + 2co W = -
(-1aP (8)
r d t a pr ae Subsituting for the values of partial pressures from equations (7)
to (9) and simplifying, the following expression results:
dw. ap (9)
d t p az
Using information of the mean streamline geometry, the equation of 6 + 2 W 0 co +o) 2 r) dr W I dW r + W Z dW Z +
dp = (p w2
(— s e dO
tangential equilibrium (2) can be rewritten in the following
(13)
convenient manner :

dW
sin a We W +W + 2wW = - aP
m m one (I
pr ae CENTRE OF
CURVAT - E
The above equation relates the effects of the impeller on the flowing
fluid in imparting a rotary motion by means of the tangential pressure
gradient between the blades. The above equation can be solved in a
step by step manner if it is reduced to an ordinary differential
equation as follows:

We
8P = - ( p r) (sin () W clWa FIG. 6a. CURVATURE IN THE MERIDIONAL
80 r
+W
m m chi ) + 2co W r )
— (
HUB - TO - SHROUD SURFACE

where it is assumed that:

ap 81,
ae 88

The assumption used above simply implies that the impeller flow
channel is regarded as lumped into a single flow line with a force
proportional to the quantity on the right hand side of the above
equation. CENTRE OF
CURVATURE
The pressure difference across the blade to blade angular
distance between adjacent blades can be evaluated at any point along FIG. 6b. CURVATURE IN THE BLADE
the streamline. Calculations have shown that the blade shape so TO-BLADE STREAM SURFACE.
determined is particularly sensitive to the variation of the radial
component of relative velocity. This finding allowed the use of the
radial-axial passage contour as a control to the blade shape for any
calculated pressure loading.
Since the data obtained for either the blade wrapping
curvature or the special geometry of the mean streamlines are in the

Downloaded From: https://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/asmep/82379/ on 04/07/2017 Terms of Use: http://www.asme.org/ab

 - -
••

It should be mentioned here, that ( do) in the above expression

1.0
represents the change in blade angle. Equation (13) can be evaluated
for each interval along the flow path, as all quantities are now ❑ OPTIMIZED DISTRIBUTION
known, that is, the assumed distribution of the relative velocity and 0 ■ A TRIAL DISTRIBUTIONS
components, pressure loading and known geometrical characteristics 0.828
of the mean streamline. The total pressure can be found by summation
of dp (i.e E dP) from equation (13). At each calculation station the 0

_4••
relative velocity components are allowed to take any value, subject to cc 0.62

/•
the condition that the following relation is satisfied:

0;/
O
W =W i + /111 i +Vor (14)
d O. 414
Z Z 0 0

An optimization algorithm for solving nonlinear programming 5 0. 207

problems (detailes may be found in Dixon (1978) and Biggs (1972) )
has been used to optimize an objective funciton given by:
i=n
Minimize F (X) = As (15)
i=1
where n = final station and As is defined by:
1w2 +w 2 + w 2 0.5
K C dr
0 z
AS — (sin y) dm (16)
dh T + w 21 0.5
m zj
The above expression incorporates the disipation loss described by
Traupel (1977); the entropy increase can thus be calculated by
integrating the above expression along the mean streamline. In order
to obtain a practical solution, another set of constraints (consisting
mainly of positivity constraints i.e. d e >0, dh >0 ,L > 0 etc.) has also
been imposed. Further details may be found in Al-Zubaidy (1989).
Figure 7 shows the initial trial distribution together with the
final optimized distribution of the radial velocity . It can be seen
that in order for velocity to satisfy the positivity geometrical
constraints, a sharp increase at entrance and a much slower increase at
exit (radial portion) is favoured.
It should however be mentioned here that on rare occasions it
was found that one or more constraints have not been obtained within
the desired tolerance, while the remaining constraints and the
optimization free gradient are small. To overcome this, the
constraints were rescaled to give the difficult ones greater weight in
the penalty function ( multiplying by a factor of 10 is usually
sufficient).
SEPARATION CRITERION
0
0 0.33.3 0.666 10
MERIDIONAL LENGTH RATIO
FIG. 7. OPTIMIZATION OF THE DISTRIBUTION OF
RADIAL RELATIVE VELOCITY COMPONENT.
W -W
s p
Cn— (18)
WP)
which equals to the difference of the relative velocities between the
suction and pressure sides of the blade, related to the mean velocity.
As a result of measurements in a rotating rectangular channel, Hill
(1962) found that separation begins at a critical value of the loading
coefficient of 0.7 (C 0 = 0.7). This results was confirmed by
measurements in a rotating conical channel (Murakami (1976)). As for
the velocity used in setting up the separation criterion, the relative
velocities (at the blade surface) are used in the normal plane. Figure
(8) shows the laoding distribution of two impellers, having the
same inlet and outlet dimensions and the formula which describes the
blade curvature. However, the axial length of the impeller along the
axis of rotation is different ( 41 vs. 46 mm). It can be seen with the
shorter impeller that the critical value of blading coefficient was
reached before the longer one. The procedure used in determining the
appropriate axial length for the given performance requirement was
that of Al-Zubaidy (1992).
1.20
2
e

The blade surface velocities calculated using equation (6) can z =.41mm 46mm
be used to give an indication to the likely separation. It is found that
CRITICAL VALUE OF REGION OF
the separation point is best defined by a criterion based on the PRESSURE LOADING
COEFFICIENT. SEPERATED
experimental investigations of Hill (1962) and Murakami (1976) and FLOW IN
IMPELLERS
a load coefficient defined by Morris (1971) which is given by the (1.2 F. MEASU-
following: REMENT )

1
P p Ps
-

C
n 2 W2 — (17) 0.
0 1.0
2
That is to say, it is equal to half the difference of the static pressures FIG. 8. LOADING DISTRIBUTION OF TWO IMPELLERS
on the pressure and suction sides of the blade, related to the mean HAVING DIFFERENT AXIAL LENGTH ALONG AXIS OF
dynamic pressure. It can also be written in terms of blade surface ROTATION.
velocities as follows:

Downloaded From: https://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/asmep/82379/ on 04/07/2017 Terms of Use: http://www.asme.org/ab

 EVALUATION OF THE DESIGN METHOD
In order to validate the design method, the obtained detailed
geometry of the impeller has been used as an input to a quasi-three-
dimensional flow analysis program by Katsanis (1965a) and (1965b).
The program is based on the streamline curvature method. Two two-
dimensional solutions, one is the hub to shroud and the other is the
blade to balde plane, are assembled together to cosntruct a three
dimensional picture. The solution would be acceptable when the
mean stream values obtained by the two analyses are within the
specified tolerances. Figures 9 and 10 show the distribution of
velocity ratio and pressure ratio versus meridional coordenates. From
Figure 9 it can be observed that a smooth and continous deceleration
along the flow path is achieved. In addition, it can be seen that well
over half of the deceleration was achieved in the axial portion of
the impellers, while the radial portion of the blade was relatively
lightly loaded. Figure 10 also shows that a smooth rise in pressure
was achieved. The obtained distributions of both velocity and
pressure have the same general characteristics of existing designs
that have good performances (experimently confirmed).
2.0
x = LOWER
• = UPPER
1.5
0 Al-Zubaidy, S.N., 1989, "A Computer Aided Method for the
0 0 25 0 5 0.75 10
%MERIDIONAL COORDINATE
FIG. 9. VELOCITY DISTRIBUTION OF THE EXAMPLE
DESIGN.
1.0
0
0 025 0 5 0.75 10
% MERIDIONAL COORDINATE
FIG. 10. PRESSURE DISTRIBUTION OF THE EXAMPLE
DESIGN.
Downloaded From: https://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/asmep/82379/ on 04/07/2017 Terms of Use: http://www.asme.org/ab
CONCLUSIONSt
A blading design method for centrifugal impeller which is
based on a prescribed diffusion schedule has been presented. The
method sheds some light on correlations between overall passage
profile, velocity dsitributions and the extend of the blade wrapping
for a specified performance requirements.
RFZERENC F S;
Wallace, F.J., whitfield, A. and Atkey, R., 1975, "Computer
Aided Design of Radial and Mixed Flow Compressors", Journal of
Computer Aided Design, Vol. 7, No.3.
Whitefield, A.. and Wallace, F.J., 1978, "Design and Testing
of Two Radial Flow Backward Swept Turbocharger Compressors", I.
Mech. E., Paper No. C74/78.
Came, P.M., 1978, "The Development, Application and
Experimental Evaluation of a Design Procedure for Centrifugal
Compressors", Proc. of I.Mech. E., Vol. 192, No.5.
Frigne, P. and Van den Braembussche R.,1978, " One
Dimensional Design of Centrifugal Compressors Taking into Account
flow-Separation in the Impeller", Von Karman Inst.
(VKI), TN 129.
AL-Zubaidy, S., 1988, "Effect of Prescribed Velocity Schedules on
the Meridional Profiles of Centrifugal Impellers", Proceedings of
the 2nd European Turbornachinery Symposium, pp. 47-52.
Johnston, J.P and Dean, R.C., 1966, "Losses in Vaneless
Diffusors of Centrifugal Compressors and Pumps", Jour. of Eng. Power,
Vol. 88, No.1, Janaury.
Dixon, L.E.W. and Szego, G.P., 1978, "Towards Global
Optimization", North Holland Press.
Biggs, M.G., 1972, "Constrained Minimisation Using Recursive
Equality Quadratic Programming ", Numerical Methods for Non-
Linear Optimization, edited by F.A. Lootsma, Academic Press.
Traupel, W., 1977, "Thermische Turbomaschinen", Vol.1, 2nd
and 3rd Editions, Springer-Verlag, Berlin, 1966 and 1977.
Design of Centrifugal Impellers Using Optimization Techniques",
ASME Paper No. 89-GT-32.
Hill, P.G. and Moon, J.M., 1962, "Effects of Coriolis on the
Turbulent Boundary Layer in Rotating Fluid Machines", Report No.
69, MIT Gas Turbine Laboratory.
Murakami, M. and Kikuyama, K., 1976, "Effects of Rotation on
Diffuser Flow", Bulletin of the JSME, Vol. 19, No. 137.
Morris, R.E., and Kenny, D.P., 1971, "High Pressure
Centrifugal Compressors for Small Gas Turbines", 31st AGARD
meeting, Ottawa, June 1968 and ASME Advanced Centrifugal
Compressors.
Al-Zubaidy, S.N., 1992, "Axial Length Influence On
Cenbtrifugal Impellers", AIAA Journal of Propulsion and Power
November-Dec.Vol 8, No 6.
Katsanis, T., 1965a, "Use of Arbitrary Quasi-Orthogonals for
Calculating Flow Distribution in Meridional Plane of a
Turbomachine", NASA TN - D2546, Dec.
Katsanis, T., 1965b, "Use of Arbitrary Quasi-Orthogonals for
Calculating Flow Distribution an Blade to Blade Surfaces in a
Turbomachine", NASA TN-D2809, May.