Professional Documents
Culture Documents
Interact ions
Michael B. Giles
Department of Aeronautics and Astronautics
Massachusetts Institute of Technology
Cambridge, MA
For permission to copy or republish, contact the American Institute of AeroMPutict and Mmnautlct
1633 Broadway, Ntw York. NY 10019
Michael B. Giles
Department of Aeronautics and Astronautics
Massachusetts Institute of Technology
Cambridge, Massachusetts
Abstract
Subscripts
is
in1
iW
08
Variables
C
c
D
E
H
J*
WI,l,S.4
W
a
7
P
AT
x
w
Introduction
In the past the calculation of the 00w in turbomachinery has been largely based upon the approximation that
the Bow in each stator or rotor row is steady in the frame
of reference of the stator or rotor. In actuality the Eow
is inherently unsteady. In two dimensions there are three
dominant causes of unsteadiness. The first is wakelrotor
interaction in which the wakes produced by a stator row
are then swept downstream into the next rotor row. The
second is vortex shedding at trailing edges. The third is
the potential stator/rotor interaction in which the pressure
field associated with the leading edge of a rotor sweeps past
the trailing edge of an upstream stator, causing additional
unsteadiness at the trailing edge and possibly affecting the
vortex shedding mechanisms.
Nomenclature
4
speed of sound
axial chord length
maximum wake velocity defect
stagnation energy
stagnation enthalpy
Riemann invariants
preaaure
rotor and stator pitches
speed
velocity components
upstream axial velocity
wheel speed
characteristic variables
fractional wake width
coordinates
flow angle
ratio of specific heats
density
time lag in periodic b.c.s
ATIPr
frequency, %VIP,
Another important reason for calculating wakelrotor interaction is to be able to predict noise generation. One of
the questions in the development of counter-rotating p r o p
fans and ultra-high-bypass turbofans is the level of noise
which ia produced. Assuming that the fans are separated
sufficiently so that they are outside each others moving
potential Eeld, then one of the principal noise sources is
the wakelrotor interaction since the wakes decay very llt-
tle before entering the rotor passage. Because the fans are
transonic and highly loaded, simple linear theories are inadequate, and EO it is important to use the nonlinear, unsteady Euler equations to predict the noise generation, and
to check the accuracy on simple test cases for which linear
theory is d i d .
Experimental studies of wake/rotor interactions have
been performed by Kenebrock and Mkolajcaak 111 for compressors, and by Binder et d [2] and Hodson [3]for turbines.
These studies show that the dominant effects are inviscid,
with the leading order behavior being simply that the wake
vorticity and entropy is convected by the mean flow. The
next order effect is the wake migration, which is due to the
fact that the wake acts ad a 'negative jet' relative to the
mean Bow, and so afier the wake is cut inlo aegmenta Ly
the blade row the segments migrate towards the pressure
surface for compressors and towards the suction surface for
turbines.
On the theoretical side, Smith [4] developed a linear theory for subsonic flow, in which he considered small perturbations about uniform flow past a flat plate cascade at
zero incidence. The theory calculates the magnitudes of the
unsteady forces, moment, trailing edge vortex sheet and radiated pressure waves, in response to either incoming wakes
and pressure waves or blade flutter. The theory leads to an
integral equation which must be solved numerically, and
Whitehead has implemented this method in an extremely
fast program [5]. Unfortunately, Smith also showed that
in experiments the steady-state loading significantly affects
the results and EO his theory is best applied in only lightly
loaded cascades.
To calculate the nonlinear effects one must turn to more
recent computational methods. The most important work
in this area is by Rai [6],who has calculated the unsteady,
viscous rotor/stator interaction, using a sliding interface
between grids which are fixed in the rotor and stator frames
of reference. A limitation of his approach however is the
assumption that the rotor and stator have equal pitches, or
alternatively that there are an equal number of rotor blades
and stator blades. In practice the stator pitch is usually
greater than the rotor pitch, by a factor which can be up
to 1.5, or even larger. A numerical approach to treating
unequal pitches was developed by Erdos [7] for the analysis
of inviscid rotor/stator interaction. The same approach
was ala0 used by Hodson [SI in calculating a wakelrotor
interaction. In this paper we use a new approach which
requires fewer assumptions about the periodic nature of the
flow, and will allow the investigation of various interesting
phenomena which would not be possible using the Erdos
method.
The presmre and tot31 enthalpy are related to the density, velocities and total energy by the following two equ&
tions which assume a perfect gas with a constant specific
heat ratio.
p
(3)
H = E + -P
(4)
3
3.1
Boundary Conditions
Solid Wall Faces
Numerical Method
3.2
= ( 7 - 1 ) ( P E - i P ( U 1 + u',)
Periodic Nodes
For steady flows, and unsteady flows with equal wake and
rotor pitches, the periodic boundary condition is a s t a t e
ment that the 0ow on one periodic boundary is exactly the
U ( z , y , t )= U ( z , y + P , t )
(5)
3.3
A similar implementation is used at the outflow bouudary except that now w ~ , w aand wd are extrapolated from
an interior node and wg Is specified to be zero. The characteristic variables are now defined by,
w,
=
WJ
w,
- P o s ) - (P - P o s ) / 4 s
(u - U08) + (P - Pos)/PosCos
(P
= -(. - U O S )
=
(v - %a)
(8)
+ (P - P o s ) / P o s ~ o s
is extrapolated from an interior node just inside the boundary. The entire solution vector is then found in closed form
from these four conditions.
3.4
3.5
Wake Models
= Pis
= Pia (1 - / ( v ) ) cos(qs)
= Pia (1 - f ( v ) ) s i n ( q s )
(9)
(P
= (u - U i n l ) + (P - Pinl)/Piscis
= -(u - U i n l ) + (P -Pini)/piacis
w,
(V
- Pini) - (P - Pini)/cia
wa
w2
(7)
- Vini)
3
f(n)
= D sin('&$)
- 131a l a )) a
-4
4,
4
4.1
'4
(11)
AT = (Ps - P,)/V
Computational Approach
Computationally it is very easy to enforce the spatial periodicity for steady fiows, as described in an earlier section.
For unsteady Bows with the lagged periodicity condition
it was desired to have as simple an implementation. This
led to the following idea: suppose that instead of a computational %me level' being at a iixed time, it is sloped in
time such that if a node at y=O is at time t , then the corresponding periodic node at y =Pr is at time t+AT, and so
once again one has simple spatial periodicity in this inclined
computational plane. Fig. 2 illustrates this concept.
Mathematically this corresponds to the following coordinate transformation.
= z
Y) = Y
2)
(14)
(12)
U(z,y,f) = U(z,y+P,,f+AT).
As an example, consider vortex shedding from a turbine row. Imposition of spatially periodic boundary conditions forces the solution to exhibit synchronous shedding,
in which each blade sheds vortices of the same sign at the
same time. However it may be true that in actuality the
blades shed at the same time, but shed vortices of alternating sign, with one blade shedding a vortex of positive sign
at the same time that its two neighbors shed vortices of negative sign. This would be an example of a spatial subharmonic whose period is ZP,. Mathematically, the spatially
periodic solution produced by the program would be a valid
solxtion to the unsteady Eulerequationa, but it waul2 have
a linear, subharmonic instability which would grow into
the fully nonlinear, subharmonic shedding. In this case to
compute the true solution would require a computational
domain spanning two blade passages. .
4.2
timevarying coefficients) produce solutions with a subharmonic component, a component whose period is a multiple
of the original period.
(13)
.
)
[ Pitch/chord
I 0.57735
(PJC)
nola>
angle)
..o__
30"
30
..
0.7
1.0
-+,P(u
7-1
-+i(u
(77_p1
+ul)-Xw
+Ul)
5.1
Eliminating u and v using the laat three equations gives
a quadratic equation for p.
-Z B ~ +=
Co
(20)
(21)
where,
= (7+1)AP
B = p i -.\'IS
c = (7-1)(2919. - 9: - 9
Results
91
1 - xu
This same procedure can also be applied to any other numerical method for solving the Euler equations. A future
paper will discuss the numerical stability of these modifred
schemes and will show that for transonic 00ws the maximum time step can be increased (sometimea by a large
factor) leading to significant overall savings in CPU for the
calculation of steady and unsteady transonic Bows.
'.c/
The entropy plot shows there is a small amount of nnmerical diffusion which tend. to reduce the peak values, thereby
eliminating the peak contonn. Note the slight distortion of
the entropy contour linea in the blade passage, with the
maximum entropy l i n g spreading out on the pressure surface and the minimum entropy linea spreading out on the
suction surface. This is consistent with the observations
and theory of Kerrebrock and Mikolajczak [lI for a compressor. Alao, the shearing of the entropy lines behind the
blade is an indication of the unsteady vortex sheet which
extends from the trailing edge.
The pressure plot shows that upstream of the blade row
there is an outgoing plane pressure wave, resulting from the
wake/rotor interaction, with a small amount of a pressure
wave of phase angle 0 = 380'. For sxbscnit 9 ~ a a, p r e s
snre wave of the form ezp(ikz i++y/P, - i w t ) will only
propagate for values of 4 in a flnite range. In this case the
relevant range is -220' < < 5 S 0 , and given the periodic
boundary conditions this further restricts to jnat the two
0 and 360' seen in the reanlts. Furthermore,
poaibilitiea '
)
the ratio of the amplitude of these two ( A ~ s o / A ogiven
by LINSUB is approximately 0.15 which is consistent with
the preasnre contonn produced by UNSFLO. On the downstream side the corresponding ratio is 1.04, and this is again
in qualitative agreement with the UNSFLO resnlta which
have a mnch more complicated contour pattern downstream
of the blade. There is some indication in the contour plot
of peculiarities in the neighborhood of the ontEow bonndary. This might be due to the trailing vortex sheet passing
through the boundary, or may be due to the quasi-normal
assumption used in formulating the non-reflecting bonndary conditions. However, as we will now aee, this has no
apparent effect on the blade pressures.
111 1 i
I
UNSFLO
LINsuB
200x25
400x50
extrap.
Lift
I
Real
hag
-0.480 -0.311
-0.497 -0.468
-0.503 -0.520
-0.501 -0.517
Moment
I
Real
hag
-0.296 -0.064
-0.337 -0.141
-0.351 -0.167
-0.347 -0.166
++
'w
To obtain a quantitative comparison the unsteady preasure8 from UNSFLO were Fourier transformed, and then
non-dimenaionalized in exactly the same manner aa LINSUB, using the product of the mean density, the mean
speed, and the maximum wake velocity normal to the blade.
To be able to compare phase results the wake definitiona at
the inflow are specifled such that the centeriine of the wake
pasaes through the blade leading edge at the beginning of
the period. Figs. 5 and 6 show the real and imaginary components of the complex amplitude of the first Fourier mode
of the pressure jump acroea the blade. The aymbola are
the resnits obtained from LINSUB, using sufficiently many
terms in the solution of the integral equation arising from
the analytic theory that the integral lift and moment are
correct to within 0.1%. The solid lines are the UNSFLO
results from the fine grid. Qualitatively the agreement is
very good although the l/\/j;singularity at the leading edge
cauaes some minor oscillations. Quantitatively the agreement is within about 5% over most of the blade, the excep
tiona being at the leading edge and near z = 0.25. which is
the point on the blade which is nearest to the leading edge
on the neighboring blade.
Table 2 compares the unsteady lift and moment coefficients. On the c o m e grid the error is approximately 30%,
while on the flne grid the error is approximately 7%. The
5.2
This paper has presented an analysis of wakejrotor interaction using Nis Lax-Wendroff timemarching method
to solve the unsteady, nonlinear Euler equations. The
boundary condition treatment splits the solution into a prescribed part, which defines the incoming wakes, and a perturbation part which corresponds to the outgoing pressure
waves and which is handled using simple quasi-normal onedimensional characteristic theory.
Inflow angle
0.15
Given a sufficiently fine grid the Euler calculation produces the correct unsteady lift and moment to within
1%,
.,
Conclusions
Acknowledgements
References
[l] J. L. Kerrebrock and A. A. Mikolajczak. Intra-Stator
l h w p o r t of Rotor Wakes and Its effect on Compressor
The results appear very similar although the entropy contours produced by U N S F M are rather more ragged because entropy is derived from the solution, whereaa in Hodsons calculation entropy is one of the primary variables
because he replaced the energy equation by the entropy
equation. The entropy results show the deformation of the
wake due to the faster flow in the center of the blade p w
sage, and the wake migration towards the suction surface.
The unsteady velocity vectors clearly show the negative
jet behavior of the wake and the generation of two counterrotating vortices due to the rolling up of the two halves of
the cut wake segment.
<*
[4]
[SI D.
[SI R.-H. Ni. A Multiple Grid Scheme for Soiving the Euler Equations. AIAA Journal, VOl. 2O:PP. 1565-1571,
Nov 1981.
ing Umteody Stator/Rotor Interaction Technical Report TR-858, MIT Computational Fluid Dynamics
Laboratory, 1986.
T. H.Pulliam, and J.L. Steger. A Gcnerd Perturbation Approach f o r the Equation6 of Fluid
Dynamics. AIAA Paper 83-1903, 1983.
[12] L. J. Chow,
Pr
1151 B. Engquiet and A. Majda. Absorbing Boundary Conditions for the Numerical Simulation of Waves. Mathematics of Computation, Vol. 31:pp. 629-651, July 1977.
il
4.
2.
-0.
APi
.-2.
3.5
3.0
-UNSFLO
-4.
LINSUB
2.5
-6.
(
2.0
0.2
0.4
0.0
1.o
0.8
1.5
Y
W
1.
8.
0.5
6.
0.0
-0.5
-UNSFLO
4.
LINSUB
AP~
-1.0
2.
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
0.
-2.
(
0.2
0.4
0.6
0.8
1.0
3.h
10