Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

78-GT-110

The Society shall not be responsible for statements or opinions


advanced in papers or in discussion at meetings of the Society or of its
Copyright © 1978 by ASME Divisions or Sections, or printed in its publications. Discussion is printed
only if the paper is published in an ASME journal or Proceedings.
Released for general publication upon presentation.

Downloaded from http://asmedigitalcollection.asme.org/GT/proceedings-pdf/GT1978/79726/V01BT02A010/2392402/v01bt02a010-78-gt-110.pdf by guest on 27 May 2022


Full credit should be given to ASME, the Technical Division, and the
$3.00 PER COPY author(s):
1.50 TO ASME MEMBERS
1.00 £ at Wembley

Through Flow Analysis of Centrifugal


Compressors

A. GOULAS
Research Student

R. C. BAKER*
Lecturer

Mechanical Engineering Dept.,


Imperial College of Science and Technology,
London, England

The matrix through-flow analysis is applied to the Si and S2 surfaces of a centri-


fugal compressor impeller. The dissipation force is expressed in terms of the
stress tensor and this allows the K - model of turbulence to be applied. The limita-
tions on the use of this model are discussed. Solutions are presented for both
Si and S2 surfaces. For the S2 surface, the streamline pattern for the K - e model
is compared with the isentropic solution. Values of meridional velocity are also
plotted. For the Si surface, as well as the K - e model, a simple turbulence model
using the logarithmic law of the wall is used. The simple model confirms that
large entropy gradients are responsible for the wake formation and the K - e model
predicts the wake formation.

*Present address: Professor of Fluids Engineering, Cranfield Institute of Technology, Cranfield, Bedford,
England.

Contributed by the Gas Turbine Division of The American Society of Mechanical Engineers for
presentation at the Gas Turbine Conference & Products Show, London, England, April 9-13, 1978.
Manuscript received at ASME Headquarters December 22, 1977.

Copies will be available until January 1, 1979.

THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS, UNITED ENGINEERING CENTER, 345 EAST 47th STREET, NEW YORK, N.Y. 10017
hrough Flow Analysis of Centrifugal
Compressors

A. GOULAS R. C. BAKER

Downloaded from http://asmedigitalcollection.asme.org/GT/proceedings-pdf/GT1978/79726/V01BT02A010/2392402/v01bt02a010-78-gt-110.pdf by guest on 27 May 2022


■l1
1:RETJCLATURB ITITRODUCTI015

B = integration factor for the con- Matrix through-flow analysis is widely


tinuity equation used for calculation of the flow in turbcma-
= dissipation force Per unit mass chines. It is based on the theoretical work
I = rothalpy, I = H - wr V^ of Wu ( 1k ) . l Marsh ( 5 ) presented a finite
0
H o = stagnation enthalpy difference method to solve the flow on a merid-
S = streamwise direction ional stream surface. In order to simulate the
Ti = normal to the stream line flow of a viscous fluid, a loss model was used
n = normal to the stream surface in the form of a local pclytropic efficiency.
n = unit vector normal to a stream Bosnian and marsh (1) then derived a set of
surface equations which govern the flow and are con-
s = entropy sistent with a loss model by introducing a
p = density dissipation force opposing the velocity
T = temperature vector. The loss model used consisted of
V = absolute velocity either experimental data or equations relating
T = relative velocity the change of entropy to the fluid, the flow
^,Ft = angles defined by equation (1) and the blade geometry. Goulas and Baker (3),
in Appendix by relating the dissipation force to the stress
r = radius tensor, have been able to calculate the entropy
z = axial direction field without having to rely upon experimental
0 = circumferential direction data for the distribution of entropy. This
T = stress tensor method was used to obtain solutions of the flow
K = kinetic energy of turbulence field within a centrifugal compressor impeller
e = rate of dissipation of turbulence which are presented here.
µ t = turbulence viscosity The Prediction of flow within an impeller
xi = Cartesian space co-ordinate (S, passage attracted the attention of a number of
TT, n) research workers. The publications of Stanitz
C l ,C 2' Cµ = constant in turbulence model (12), Hamrick, et al. (4), Katsanis (1),
= angular velocity Senoo, et al. (11), McDonald, et al. (10),
= distance from the hub for S2 and Sturge, et al. (13) and Howard and Osborne (5),
from the Pressure side for Si have to be mentioned here. Almost all of them
= distance between hub and shroud implemented an inviscid flow approach, over
for S2 and pressure and suction which a jet-wake model or some sort of effi-
side for Si stream surface ciency was imposed.
i/ = stream function. In the present study, two different sets
of solutions will be presented. The first,
Subscripts known as the Sl surface, extends from blade to
i, j, k subscripts denoting Cartesian co- blade; the second, the 52 surface, extends from
ordinate directions hub to shroud.
r radial direction
u circumferential direction Underlined numbers in parentheses
z axial direction designate References at end of paper.

FLOW ON STREAM SURFACES


_ _ [aU; a .1 ] 2 a (7)
Tip _ - P ^i°. "t ax. ` a s ° di
The complete theoretical analysis can be
found in References ( 1 ) and ( 3 ), Here only the
basic equations will be set for the complete- The scaler turbulent viscosity, µt, is given as
ness of the presentation.
For a curvilinear orthogonal system of KZ (8)
µt = c u ° c
co-ordinates on the stream surface with axes
S along and N across a streamline on the stream
surface and a third axis n normal to the surface and the kinetic energy K and the rate of dissi-

Downloaded from http://asmedigitalcollection.asme.org/GT/proceedings-pdf/GT1978/79726/V01BT02A010/2392402/v01bt02a010-78-gt-110.pdf by guest on 27 May 2022


the equations governing the flow on the stream pation c of turbulence are obtained by solving
surface are: the following system of differential equations:

(a) Continuity DK 1 a µ t In (9)


+ G - e
(b) Equation of motion for the N direction Dt P IX k (0 k ax k

On 1 a ^µ t ar c1 I E2

N.W x (9 x V) = N. ( IV- TV) (1) Dt


__ ]*__0c2-_
P ax k o f x k (1G )

µ t 90 i 1Uk 9U i
(c) Equation of motion for the S-direction and G = P ax k + ax i axk (11)

(2) is the generation term.


The constants involved are given as
(d) A geometrical condition derived from the
stream surface C = 0.09 , C1 = 1.44 , C2 = 1.92

a = 1.3
W.i = 0 (3)

(e) The energy equation for steady flow ON THE USE OF K - £ MODEL

DI (L)
=0
There are two assumptions made when
Dt
writing equation (7) for the "Reynolds" stress-
(f) Equation of state. es. The first is that the turbulence is iso-
tropic. As was pointed out by Launder and
The combination of the first two equations will Spalding ( 8 ), this assumption is valid for two-
yield an expression for the stream function ciof dimensional flows without swirl. In three-di-
the form mensional flows, as in the case of the flow
within an impeller passage, a different level
V2' = 4(r, 4, z) (5 of turbulent viscosity may be needed for each
active stress component. The second assumption
The form which q takes depends on the set of is that the "Reynolds" stress tensor and the
the independent variables used, and, therefore, strain tensor have the same sign. In this case
on the geometry of the stream surface under the generation term G is always positive. This
consideration. The complete expressions for q is true for most flows. One exception is flow
are given in the appendix. Goulas and Baker in rotating channels and centrifugal turboma-
( 3 ) have shown that the dissipation force can chinery where experiments by Johnston, et al.
be expressed as ( 6 ) show that there is an area of negative pro-
duction of turbulence energy near the suction
1 (6) side. Therefore the results obtained using the
O = - tv.t)
K - e model should be treated with caution.

To obtain the stress tensor the K -e model of THE SOLUTION PROCEDURE


turbulence as described by Launder and Spalding
(8) is used. The stresses are given by the The flow field on a stream surface can be
equation obtained by solving simultaneously equations
(2) and (5) using a finite difference method.
The flow domain is covered by two sets of grid

2
lines. One set consists of lines oorhich follow, upstream y - line is found by fitting a cubic-
aprroxi:^iately, the neap floor (called the x - spline polynomial to the stream function dis-
limes) and the other set of lines, trhich are tribution along the y line.
az^,proxirmtely noric_al to the x lines ( called the Finally with velocity, entropy and density
- lis) . known, function q(r, o , z) can be calculated and
The partial derivatives present in the a banded matrix set up to solve equation (5).
equations are replaced at each grid node by The resulting distribution of has to be
flltc difference forms obtained by using the compared with the initial one and the ,whole Pro-
fauloriz sealer. A nine point lattice is used cedure repeated until a convergence criterion,
aro -_tnc each grid node. The discretization er- defined as

Downloaded from http://asmedigitalcollection.asme.org/GT/proceedings-pdf/GT1978/79726/V01BT02A010/2392402/v01bt02a010-78-gt-110.pdf by guest on 27 May 2022


^_
°•o° _o of the order of 0 (o h /lxi ).
Ter the blade-to-blade solution and at 5new g old
0.001
the ex-t cf the flow field a "parabolic" type ^o1d

lattice i<; used. This lattice consists of six


ints u,;hich lie ur^strean of the exit node. is satisfied at each Point.
The discretization error in this case is of the To obtain the three-dir,:ensional flow field
order of 0 (a 3 ^/axi 3 ). For a rectangular grid an iterative sequence between the Sl and S2
this is equivalent to an ''upwind-difference." solutions is possible. This has not been tried,
The first step for the solution is to as a different approach which is thought to be
assume a distribution of the stream function . superior is being developed and it is hoped to
The corresponding two components of the velocity report on this in the near future.
vector are then calculated from equations (3),
(j), or (7) cf the Appendix depending upon which THE BOUNDARY CONDITIONS
set of coordinates is used. The third component
of velocity can be calculated from the geometry The distribution of the stream function
of the stream surface, i.e. at the inlet can be obtained by integrating one
of the equations defining . Along solid bounda-
d .n = 0 ries 1' remains constant. The exit condition for
Li differs for the two surfaces. For the Si sur-
The second step is to solve equations (9) face it was taken as
and (10) to obtain the kinetic energy and the
a^ = o
rate of dissipation of turbulence. Guessed
values for the first iteration or those obtained
from a Previous iteration for K and E are used which is equivalent to
to calculate the right hand side of the equa-
tions. A set of linear equations is formed and wz = 0

solved simultaneously for each y - line. Start-


ing from the entrance to the flow field, where at the exit. This seemed appropriate for a
K and e are given as initial conditions, the radial impeller.
solution proceeds along x - lines to the exit. For the Si stream surface the radial
The third step is to calculate the tur- velocity at the nearest exit grid point to the
bulent viscosity µ t from equation (8), the re- pressure side was set equal to half the initial
sulting stresses fro r.. equation (7), and finally axial velocity. For the rest of the grid points
the dissipation force from equation (6). at the exit, the parabolic type lattice was used,
The fourth step is to integrate equation thus avoiding the need for prescribing a distri-
(2) along a stream line assuming that the ex- bution for the stream function at the exit. This
nression boundary condition is a compromise between in-
terfering with the development of the flow as
(T.D) /(T W) little as possible and ensuring that there is a
high velocity at the pressure side as found ex-
is constant along a stream line between two perimentally. The value of half the initial
consecutive y - lines. This integration yields velocity is only tentative and still under study.
an entropy field in accordance with the velocity The logarithmic law of the wall was used
field assumed. The exact position of the Point to obtain values for the stresses at the wall
of intersection between the stream line passing boundaries and values for the kinetic energy
through the point under consideration and the and rate of dissipation of turbulence at the
0

Downloaded from http://asmedigitalcollection.asme.org/GT/proceedings-pdf/GT1978/79726/V01BT02A010/2392402/v01bt02a010-78-gt-110.pdf by guest on 27 May 2022


Fig. 2 Streamline patterns for S2 stream surface
Fig. 1 Meridional section of impeller
60,000 rpm, and the corresponding mass flow 0.14
nearest internal point of the flow field. De- kg/s.
tails can be found in Launder and Spalding (8).
S2 SOLUTION
THE RESULTS
The geometry of the surface considered is
The impeller considered in the present the same as the blade.
work is a small radial compressor with an in- Fig. 2 shows the pattern of the streamlines
ducer, similar to those commercially available for both the isentropic and K -s case. The
for turbochargers. Fig. 1 gives the meridional
cross-section and the camber line of the blade
in the cylindrical section A-A. The initial
velocity prescribed is uniform axial velocity

0.6
W = const , W = 0 , W = - wr.
z r u
V
Ur
tip
The rothalpy I is constant at the eye of the
0.4
impeller and remains so throughout the impeller
passage. Because of lack of either accurate
measurements of the flow field or detailed 0.2

geometry of a compressor, the results obtained


were compared only to the isentropic solution
for the same case thus permitting indirect com- 0 0.5454 1
Qjeo
parison with Eckardt's data (2). hub shroud

The tip speed is 228 m/s, rotational speed Fig. 3(a) Meridional velocity profile at station
1
4
a

grou-:th of the boundary layers at the hub dis-


places the streaelines right fror.: the eye of the
impeller. At the shroud, the displacement is
restricted by the prese e of a pressure gradient
0.6
V normal to the shroud which is due to the cen-
C rp
trinetal forces at the inducer.
0.4 Pik . 3 shows the see idicnil velocity pro-
,

files for three different stations. At the


exit the maximum of the meridional velocity
0.2
shifts towards the hub as vckhardt's (2) meas-

Downloaded from http://asmedigitalcollection.asme.org/GT/proceedings-pdf/GT1978/79726/V01BT02A010/2392402/v01bt02a010-78-gt-110.pdf by guest on 27 May 2022


urements suggest.
Fig. 4 gives the meridional velocity dis-
0 0.5454
1 eleo tribution along the hub and along the shroud of
h W ro—I
the impeller.
I(1) I-Ieridionil vet city profile at station
2 Sl Solution
The St surface considered is an axisyra-
metric surface obtained by rctat ; a strea°:,line
of the 22 sues-ice. The inteTration fact=r L
n:ras obtained from the same S2 solution.
Trc models were use for the turbulence
.e,
The first .ms a : ir~_nle r. del. In this 1. ml the
t. stresses due to turbulence or the boundaries
C.4 ;rere set equal to those c iven by the lc":•arithr.!ic
-

latr cf the gall. At the rest of tI)e ft c : field


,

only the viscous stresti=es -:;ere consit-el.


The second _nedel was the K - e i:_,edel. In
the light of t discussion cf the section
"The Solution Procedure," the Ieneration tern_
0.5454 1 fle, ;as set equal to zero near the suction side.
a t shroud
=^i shorts the t:-elccity profiles for
-o. (c) :.eridional velocity profile at station ttiro cases. One is obtained by using the simple
3 model, and the other by assu=nino isentropic

V C 4

m
wr
tip

0.5 1
to

^iel city b tribution . the hub ana shroud, ta=, at the nearest 7rid roL.^_to

5
am

0.6

v 0.4

wr tiP 0. 5

0.3

0.4
w
wr r
tip

Downloaded from http://asmedigitalcollection.asme.org/GT/proceedings-pdf/GT1978/79726/V01BT02A010/2392402/v01bt02a010-78-gt-110.pdf by guest on 27 May 2022


0 0.5454 1 pie0
0.3
P - s s - s

Fig. 5(a) Axial velocity profile at station 1.


Si stream surface 0.2

flow. The jet-wake pattern of flow at the exit


is evident and it is a result of large gradients 0.1

of the entropy. This is in agreement with the


interpretation given by Eckardt (2) of his meas-
urements although the mechanism of entropy in- 0 0.5454 1 Q^^

s - s
crease at the suction side is different. Ac- P - s

cording to Eckhardt (2) the high entropy fluid Fig. 6 Radial velocity profiles at station 3.
is deposited at the suction side by the secondary S1 stream surface, K - E turbulence model
flow. As this is basically a three-dimensional
flow, a two-dimensional prediction method is in- two cases where the entropy at the points near-
adequate to predict the wake. The results ob- est to the suction side was increased. In the
tained by using the K - e model show an isen- first case the increase in entropy calculated
tropic type of flow at the core of the passage. by the K - F model was doubled, in the second
The profiles are flat near the boundaries and case the increase was quadrupled.
there is no formation of wake, due to the fact As the entropy gradients increase a wake
that the entropy gradients calculated are not starts forming at the suction side.
enough to create a wake. Fig. 6 shows the
velocity profiles for the K - e model and for CONCLUSIONS

The results obtained by solving for a hub-


to-shroud, S2 stream surface, are in qualitative
agreement with experimental results for the jet
0.6
side obtained by Eckardt.
For the Si surface, the use of the "simple"
0.5
turbulence model shows theoretically that large
entropy gradients are responsible for the forma-
tion of the wake. These gradients cannot be
0.4
Vr
predicted by the K - e model as they result from
r tip the secondary flow which a two-dimensional method
cannot predict. The 51 solution can, however,
0.3
predict the development of wake using the K - E
model, provided that the transfer of entropy to
0.2 the suction side by the secondary flow can be
included in the solution through a suitable
correlation.
0.1

REFERENCES

1 Bosman, C., and Marsh, H., An Improved


P - S s - s
Method for Calculating the Flow in Turbomachines
Fig. 5(b) Radial velocity profile at station 3. Including a Consistent Loss Model," Journal of
Si stream surface. Mechanical Engineering Science, Vol. 16, 1974,
p. 25.
6
2 Eckardt, D., "Detailed Flow Investiga- APPENDIX
tion Within a High Speed Centrifugal Compressor
Impeller," ASME, Journal of Fluid Engineering, THE EQUATIONS
Vol. 98, 197 6 , p. 390.
3 Goulas, A., and Baker, R. C., "Through The geometry of the stream function is
Flow Analysis of Viscous and Turbulent Flows," given by two angles
ARC 37017, Ae.Th 49, 1976.
tan 4 Hamrick, J. T.,
a Ginsburg, A.,
= and
n3 (1)

Osborn, W. M., "Method of Analysis of Compres-


sible Flow Through Mixed-Flow Centrifugal Im-

Downloaded from http://asmedigitalcollection.asme.org/GT/proceedings-pdf/GT1978/79726/V01BT02A010/2392402/v01bt02a010-78-gt-110.pdf by guest on 27 May 2022


n2
pellers of Arbitrary Design,"
tar) =
NACA-TM-2165,
(2)

1950.
5 Howard, J. H. G., and Osborn, C., The continuity equation is satisfied through
"A Centrifugal Compressor Flow Analysis Em- an integration factor B which is proportional
ploying a Jet-Wave Passage Flow Model," ASME to the thickness of the stream surface. If
Journal of Fluid Engineering, Vol. 99, 1 977, the independent variables used are r - z the
p. 141. equations will be
6 Johnston, J. P., Halleen, R. M., and
Lesius, D. K., "Spanwise Rotation of Turbulent F a ay = (3)
z - P Br V.a - c Br y r
Channel Flow," Journal of Fluid Mechanics,
Vol. 56, 1972, p. 553.
7 Katsanis, T. H., "Computer Program for and the final equation
Calculating Velocities and Streamlines on a
Blade-to-Blade Stream Surface of a Turbomachine," —a (P Br) - V, 2e (P Br) + (P B=) 2 d^ -
ae 2 * a = V,
NASA-TN-D-4225 , 1968.
- 8 Launder, B., and Spalding, D. B., "The w
P z^ [ - (v
z - w„ tar - oz (v r -w^ tar

Numerical Computation of Turbulent Flows," Com- * (4)


puter Methods in Applied Mechanics and Engi- a (r v n )- tan T a `^ (t V a ))
+ P B [tam s z

neering, Vol. 3, 1974, p. 269.


= q (r, P, z)
9 Marsh, H., "A Digital Computer Program
for the Through Flow Fluid Mechanics in an
Arbitrary Turbomachine Using a Matrix Method," If the independent variables are r - the
ARC R & M 3509, 1968. equations will be
10 McDonald, G. B., Lenneman, E., and
Howard, J. H. G., "Measured and Predicted Flow (5)
az = - BPwu • i a^ = BPw =
Near the Exit of a Radial Flow Impeller," ASME
Journal of Engineering for Power, Vol. 93,
1971, p. 441. and the final equation
11 Senoo, Y., and Nakase, Y., "An Analy-
sis of Flow Through Mixed Flow Impeller," ASME ar (B2) + (B2)2 dp
ar t + r2 p 2B r r a^ (B2) -
Journal of Engineering for Power, Vol. 94, 1972,
p. i3. + LM ar ^w -w z tan u) z a4 (w r - w^ tan al ] + (6)
w
12 Stanitz, J. D., "Some Theoretical
Aerodynamic Investigations of Impellers in aa - tan µ -p—) + r + 2w
Radial and Mixed Flow Centrifugal Compressors,"
+ PB ]tan A r @
1BP=
= q )_, A, z)
Transactions of ASME, Vol. 74, 1952, p. 473.
13 Sturge, D. P., and Cumpsty, N. A.,
"Two Dimensional Method for Calculating Sep- If the independent variables are z - 95 the
arated Flow in Centrifugal Compressor," ASME equations will be
Journal of Fluid Engineering, Vol. 97, 1975,
p. 581.
14 Wu, C. H., "A General Theory of Three
Dimensional Flow in Subsonic and Supersonic * n i (i = 1, 3) are components of the
Turbomachines of Axial, Radial and Mixed Flow unit vector normal to the stream surface. Values
Types," NACA-TN-2604, 1952. of ni for each set of co-ordinates are given in
Table 1.

7
Table 1

Set of
Co-ordinates n1 n2 n3

r - z n nu
nz

r - nr nU nz

Downloaded from http://asmedigitalcollection.asme.org/GT/proceedings-pdf/GT1978/79726/V01BT02A010/2392402/v01bt02a010-78-gt-110.pdf by guest on 27 May 2022


z - nz nU nr

-Bp w^ (7)
aP = BPWZ

and the final equation

azz + -- a = W . r 3^ (pB) - W u adz (PB) + (pB) 2 a-

+ W .P ^ az (W a - W r tan µ) - z a^ (W y - W r tan A)
1 + (8)
r
+ pB tan A r - tan µ ai
j - Bp [s + 2w J tan A

= q (r, •, z)

The overbars denote the special derivatives as


defined by Wu (14)

You might also like