22-1

PERFORMANCE ANALYSIS OF CENTRIFUGAL COMPRESSOR STAGE, BY

MEANS OF NUMERICAL AND EXPERIMENTAL INVESTIGATION OF
ROTOR-STATOR INTERACTIONS.

O.DOMERCQ R. THOMAS A. CARRERE

CIFRE Engineer, Head of the turbine designteam, Professor at SUPAERO,
Centrifugal compressorsdesign, Aerothermodynamic (Ecole Nationale Superieure de
Aerothermodynamic department, department, 1ACronautiqueet de 1Espace)
Turbomeca - Boite 17 Turbomeca, Boite 17 Thesis director,
65411 Bordes, FRANCE. 65411 Bordes, FRANCE. 3 1400 Toulouse, France.

1. ABSTRACT u, v, w Absolute velocity Cartesian

This paper deals with numerical and experimental coordinates
investigations of rotor-stator interactions between a x, YI z Cartesiancoordinates
backswept centrifugal impeller and its associated
vaned diffuser. Experimental data were obtained by Greek symbols :
laser two focus velocimetry and fast response rl Efficiency
Kulite transducers. Computations were carried out P Density
thanks to a three-dimensionalNavier-Stokes solver, T Shearstresstensor
customisedby the authors, for the current purpose.
Time-resolved simulations of the full stage with Superscripts :
passages number reduction but respect to real T Transposedmatrix or vector
geometry of components are then presented.
Comparisonswith experimental data lead to a code Subscripts :
validation phase and critic investigations of rotor- a Absolute frame of reference
stator interaction phenomena. Evidence of the i Total conditions
existence of a strong interaction between the rotor r Relative frame
and the stator flow fields are pointed out. In S Static conditions
particular, an intense upstream influence of the V Viscous terms
vaned diffuser was observed. Finally, steady stage
calculations, coupling the components by a mean
interfacial treatment, are examined. The reasonable 2. INTRODUCTION
computational cost of this method now allows such Advances made in the design of centrifugal
numerical simulations of centrifugal stagesto be compressorsin the past thirty years, elevated that
part of design cycles. device to a key position in small and medium gas
The numerical part of the study was performed at turbine engines. Besides,aeroenginesare required
Turbomeca, using the local software and hardware to be lighter and to involve reduced cost of
facilities, whereas the experimental campaign took ownership and maintenance. Considering
place in the Propulsion Laboratory of SUPAERO, compressors,those trends lead to more compact
part of the LAMEP (Laboratoire Mixte en components with high level of aerodynamic
Energetique et Propulsion), which recently loading, particularly for military applications. Then,
designeda test rig devoted to compressors. the aim of first-attempt design with adequate
performance and stability margin, becomes of
Nomenclature : rough achievement.
CP Diffuser static pressurerecovery In the last decade, thanks to detailed flow
coefficient investigations, the extensive use of 3D steady
Q,E,F,G Elements of Navier-Stokes (Euler or Navier-Stokes) solvers and experience of
equationsin vectors form designers gave birth to a high performance
G Corrected massflow generation of centrifugal impellers. But, however
J Jacobian efficient may be the rotor, the matching with its
M Mach number associated vaned diffuser remains a hard task,
u,v,w Contravariant velocity components almost always obtained through mono-dimensional
PYPs Static pressure considerations.The flow generatedby the impeller,
Pi Total pressure entering the vaneless space and later the radial

Paper presented at the RTO AVT Symposium on Design Principles and Methods for Aircraji Gas Turbine Engines,
held in Toulouse, France, 11-15 May 1998, and published in RTO MP-8.
22-2

diffuser appears highly distorted in both The rotor has an equal number of full blades and
circumferential and spanwise directions and subject splitter blades.
to severe unsteady conditions. Among many Close to the maximum efficiency point on the
aerodynamic causes, we can put emphasis on design speed curve, the impeller blade passing
curvature, viscosity and compressibility effects as frequency is superior to 5000 Hz. The compressor
well as Coriolis force influence. Then the vaned stage pressure ratio and the mass flow rate are
diffuser aimed at static pressure recovery from the respectively around 4.0 and 2.0 kg/s. The impeller
flow kinetic energy, receives strongly time- isentropic efficiency is then over 90%. This
dependant inlet conditions, whose influence on its compressor and its experimental mounting were
performances and stability is not yet accurately designedby Turbomeca company. In particular, the
known but undoubtedly has undesirable adverse optical access is provided by a shroud inserted
effects. Those assertions are indeed derived from window allowing measurements situated in the
experimental studies and industrial experience23456. vaneless diffuser and in the captation area (i.e.
Therefore future design methods will need to take semi-vanelessspace).
into account aspects of the flow field currently Experimental studies were performed in the
ignored. In particular, the previously mentioned Propulsion laboratory of the SLJPAERO (part of the
effects of the blade row proximity and LAMEP, Toulouse, France). The compressors
unsteadiness. Some 3D Navier-Stokes codes, either devoted test rig used is powered by an electric
in the form of commercial packages or research motor delivering 400kW. The air intake was
derived software, are now available to model those precalibrated to ensure accurate mass flow
features. But the designer needs an estimation of measurement. The outflow enters a plenum
the relevance of numerical results. Then, each chamber and the desired pressureratio is reached
constructor must assess his numerical design tools by means of a butterfly valve motion. The
on unsteady data, obtained by detailed operating point of the compressoris controlled with
investigations on engine representative machines, 42 pressure probes, inlet and outlet temperature
by carrying out a code validation phase. probes. A computer linked to the rig allows on-the-
TURBOMECA initiated such actions several years fly storageof those data with a view to taking into
ago and recently enhanced its data bases within a account small operating point disturbances in
joint research program with the LAMEP (Toulouse, experimental post-processing ( for instance mass
France) among others. flow recovery calculation). Figure 1, shows the
The present article is intended to give an outline of SUPAERO facility.
this process from the experimental work to the code
validation phase and the modifications of the
numerical package necessary to allow a design
relevant use of such tools, regarding current
hardware capabilities. The first part of the text will
briefly describe the test compressor and the test rig,
before examining the experimental data acquisition
carried out by means of a Laser Two Focus
velocimeter and fast responsepressuretransducers.
In a secondpart, after a brief numerical background
explanation, we will focus our attention on full
unsteady simulations of the centrifugal stage flow
field. Then, a steady mode of stage calculation,
involving Riemann invariant theory, will be
presented and assessedas an available design tool, Figure l- SUPAERO compressor test rig.
already in useat TURBOMECA.

3.2 Laser Two Focus velocimeter.

3. TEST STAGE AND MEANS OF
INVESTIGATION. 3.2. I General description.
The experimental approach had been carried out
3.1 Compressor and test rig. with a Laser Two Focus velocimeter. The present
The centrifugal compressor stage investigated in device is derived from DLR works and its main
the present work is composed of a transonic characteristics were given by Schold*. Briefly, the
backswept impeller matched with a vaned diffuser. principle of L2F velocimetry consists in the
measurement of the transit time of seeding
particles, included in the flow, between two highly
 22-3

focused laser spots. Histograms of particles absolute speed and over-deflected flow feature.
detection as a function of spot alignment direction Velocity fluctuations up to 20% and flow angle
and time of flight ( i.e. transit time) lead to flow variations of 15 to 20 are measured. However, the
velocity determination. The projection of the flow delivered by the rotor exhibits a relative
velocity vector on the plane perpendicular to the smoothness, i.e. no jet/wake pattern in the sense of
laser beams is obtained through its modulus, angle Dean and Krain* is observed. Turbulent rates reach
and turbulent rates by statistical calculations as 10 to 15% within the wake regions and are lower to
described by Schold. The measurement accuracy 5% elsewhere.
depends on the flow turbulence intensity. The question then arise, to know to which extend
Considering our experimental facilities, the these flow heterogeneities mix out before the radial
precision can be evaluated to +I- 1% on the diffuser leading edge as currently always assumed
absolute velocity and +I- 1 on the absolute flow in design process. Considering the second
angle, when turbulent rates are less than 5%. measurement section, in terms of velocity
magnitude, both the peak induced by the blade
Those measurements are intended to provide data wake passing and the outer gradients have been
for rotor-stator interaction mechanisms reduced. However, large level of fluctuations,
comprehension. Then, they have to be taken for around 15% remain. The mixing hypothesis is then
various relative locations of the impeller main blade contradicted in our particular case. Similar
in front of the diffuser passage. A magnetic pick-up observations can be drawn from the flow angle
situated on the machine rotation axis allow the fluctuations remaining around 10 to 15. Besides
necessary data storage triggering. the discrepancies between turbulent rates obviously
suggest that the mixing process is far to be
The choice of L2F velocimeter instead of LDV ( completed at the radial diffuser leading edge.
Laser Doppler) device was linked to the narrowness At the throat section, the flow is still highly time-
of the flow path and the restriction of flow dependant and spatially heterogeneous. However,
accessibility to a single window. The need of high due to the differences in streampathes, adverse
signal to noise ratio necessary for wall approaching pressure gradients and in the influence of the
investigations claimed in favour of the time of leading edge stagnation area across the diffuser
flight solution. Discussions on the technical aspects pitch, the time-organized blade passing feature has
of this choice, as well as interesting developments vanished.
on advantages and disadvantages of both devices
applied to small high speed turbomachinery should 3.3 Unsteady pressure measurement.
be found in Fagan and Fleeter and Elder, Forster
and Gill publications. 3.3.1 Transducers and measurements locations.
The transducers used for this application were
3.2.2 Measurement positions.
Kulite pressure probes of the XCQ-093 serie. They
Rotor-stator interaction influence on flow evolution fit the temperature and pressure conditions
has been investigated in three main test sections: encountered in that flow region and present a
. the mean radius of the vaneless diffuser. bandwidth suitable for high frequency
. the vaned diffuser leading edge. measurements. Each probe was calibrated in terms
l the throat of the vaned diffuser. of transient response in a shock tube with its own
Each of the three main sections contains ten protection screen, flush mounting device and
measurement locations. Several relative depths acquisition facility used during later compressor
were investigated. runs. The measurement configuration is presented
on Figure 2.
NB: Some results taken from a previous
experimental campaign from the ONERA and
conducted by Fradin are added to our currently
performed measurements with a view to increasing
our validation data base.

3.2.3 Data analysis.

The flow in the mean vaneless space section
appears to be dominated by a high level of time
dependant heterogeneities. More precisely, a strong
wake influence of the impeller passing blades is
clearly observed. The wakes take the form of a high
Figure 2 - Kulite flush mounting facility.
22-4

Sixteen measurement positions were distributed on sufficient number of cycles is described. In

the external wall of the radial diffuser (see later particular, no difference is observed between a five
figures 5) : rotor rotations or ten rotor rotations acquisition.
l Four positions at the mean radius The accuracy in terms of reduced time dependency
between impeller trailing edge and on synchronisation device can be evaluated to 2%
vaned diffuser leading edge at the of the cycle duration.
reduced pitch ratios of O., 0.25, 0.5 and Nota : The unsteady pressures on the mean
0.75. vaneless space radius and the throat section were
l Three positions at the vaned diffuser treated to extract the time mean pressure
leading edge, at reduced pitch ratios of
(p,(t) = -&)dd which can be compared to static
0.25, 0.5 and 0.75. 0
l Three positions in the captation area. pressure tappings originally implemented on the
l Three positions at the diffuser throat test compressor. The comparisons showed some
section. discrepancies between pressure probes. Correcting
l Two positions downstream the diffuser factors correlatedfor difSerent operating points and
throat situated on a mid-pitch line. assessed by swapping probe positions were then
taken into account.
3.3.2 Data acquisition and processing.
The transducer signal is amplified via 4 channels of
ANS-E300F4 and converted from analogic to
digital through an OX2000 Metrix Oscilloscope at a
5OOkHz rate. A pentium 90MHz PC is used to
command the A/D converter and store the files via
an RS232 serial link.
The acquired data consists in the raw signal
obtained during five to ten impeller rotations. The
synchronisation signal is given by a one per blade
impulse allowing the initial time to be converted in
a reduced time of the impeller blade passing in Figure 3 - Raw Data
front of the diffuser pitch but preventing distinction
between impeller passages. Then manufacturing
differences between rotor passages will not be
taken into account in our current study.
Those data are processed by a phase-locked
averaging process, as described by Ainsworth or
Cicatelli and Sieverding13. Figure 3 represents the
raw data obtained from position 1 for instance. The
averaging consists in the following :

a) Ensemble averaged signal :

Figure 4 - Phase-meaned data
P(t) = $$P(n,t)
n-l 3.3.3 Data presentation.
b) Average random unsteadiness :
Figure 5 represents the phase averaged period for
P(n, t) = P(t) + p(n, t) the 4 experimental positions. The x-coordinate
corresponds to the reduced period of rotor blade
passing in front of the diffuser pitch, where 0.
corresponds to a situation where the impeller
trailing edge and the stator leading edge stands at
where P(n,t) is the raw signal, N is the number oi the same azimuth.
cycles acquired, t is the time, n is the index withir
the N cycles. A cycle corresponds to the rotation 01
an imueller passage in front of the probing position.

Figure 4 give the results of this procedure for

position 1 for two different numbers of cycles, thus
showing the independence of the solution when a
 22-5

Q=(P.P.P.PW.)7
E=(pu,pu* +p,puv,puw,~(e+/7))~

I;=(pv,puv,Pv2+p,PVW,V(e+P))l

G=(pu,puw,pvw,pw+p,w(e+/~))~

E,, = (0. r, r,r,ur,+vr,iwr, 1

I;, = (0 7rw.~,.r, TX+ +vryxT + wr,, 1

equations can be transformed to an arbitrary

curvilinear space &q,<, r *., while retaining strong
conservation law form as:

ai, a(E-E,)+a(G-d.,+a(%-F.)=O
Figure 5 - Pressure evolutions (1). at+ ag aq at;
4. THREE DIMENSIONAL NAVIER-STOKES
SOLVER. where the vectors are given by :
A
Q = J-(p,pu,pv,pw,e)T
4.1 Foreword. b = J-(PU,PUU + S,P,PVU + <!P,PwU + 5,p,(e + PYJ - LP)
The NREC commercial software package
VISIUNTM constituted the basis of this numerical i = J-(PV,PUV + *I,P,PVV + rl,p,pwV + rl,p,(e + P)V - ?,P)~

study. This code, written for turbomachinery

6 = J-'(pW,puW +i,p,pvW +<!P,PwW +<,,p,(e + PY -C.,P)
purpose solves the three-dimensional Navier-Stokes
system of equations4567. Pulliam and Steger
works give an overview of the implemented
method. Developments were carried out by the
authors wishing to closely examine rotor-stator
interaction features. In particular, the treatment of
interfacial areas and multi-channels domains had
been customised, as well as the possibility to
perform steady stage calculations.
the viscous terms are transformed in the same way,
4.2 Numerical background.
their expression is given by Pulliam I8

4.2. I System of Equations. The metrics and the jacobian are :

The numerical approach of VISIUNTM is based
upon a three dimensional Navier-Stokes code, 5, = J(Y,& -Ycz,) ll, =.(z,y, -y:z:)

written for turbomachinery purpose. The full 5, = J(Z& -x,zc) q, =J(x,z, -x$Q)

system of equations expressed in strong 5, = Jb,Y: -xzY,J 11 =J(y,x, -x-y;)

conservation form is used herein. Its compact form 6, = J(YlZ, -ZrY,) 5, = -G,-Y.L -a

can be expressed as follows in Cartesian 6, = J(xqq -xszq) 3, = -x,'1,-Yr'1, - =,'1,

6, = J(%Y, -Y& 5, = -vLY.L - ZL,
coordinates.

Q d(E-EJ + W-GJ +W-Fv) -o

-+
at ax ay az - 4.2.2 Turbulence modelling and solution
where vectors are given by : algorithm.
The algebraic eddy viscosity model of Baldwin and
Lomax is implemented to take into account
turbulence effects. Equations are solved
numerically using a Beam and Warming type of
algorithm8~2. At inflow boundary, total
temperature, total pressure and two flow angles are
22-6

specified ; while static pressure is given at the component so that Kl and K2 are small but Kl/K2
outflow boundary. stays as close as possible to Nl/N2. A periodic
azimuthal condition is implemented in each row
(Iper ad Iper. lines for the impeller and Dper and DperY
4.2.3 Two frames of reference lines for the diffuser on Figure 6). This results in a
The main difference between single blade row time lag cancelling at the extreme azimuthal
computations and stage simulations lies in the boundaries of each group of passages. The
existence of two frames of reference, one for each information transferred through the interfacial area
component, separated by an interfacial area, from one grid to the opposite must be compatible
through which numerical information has to be with those spatially periodic conditions. Then we
transferred in order to predict rotoristator have to use an expansion/contraction step at the
interaction effects. Several different approaches, interface. Let e,=2ITK,/N,, e,=2HK,/N, be the
either conservative or non conservative, have been respective azimuthal extension of each component
used. The matching of the two frames retained in grid on Figure 6 and ec= (e,+e,)/2. Fl is the
VISIUNTh is based on M. M. Rai 22 method. The downstream boundary of the first sub-domain
grid is split into a rotating zone and a non-rotating related to el, F2 is the upstream boundary of the
zone with an interfacial overlay. Flux vectors are second sub-domain related to e2. Fc is then a
interpolated from one grid to the other in this virtual frontier related to ec through which data are
overlaid area to obtain boundary conditions related transferred by interpolation on an equal azimuthal
to each component sub-domain. H-type meshes are extension ec thanks to previous
used to model the geometry. Two overlaid grids are expansion/contraction of the azimuthal gradient
generated, one for each component. The interface between Fl and Fc or between F2 and Fc as
area consists in a cylindrical surface situated at the represented on Figure 6.
ec=1/2(el+e2)
mean radius between the impeller trailing edge and
the vaned diffuser leading edge.

4.2.4 Unsteady stage calculations : a reduction of

the computational domain.
Cpu time and memory necessary for unsteady stage
numerical simulations make the use of a
Periodic boundaries
computational domain reduction unavoidable. A
common approach consists in geometry
modifications in order to obtain the same azimuthal Figure 6 - Reduction of the computational
extension in the rotor and in the stator modelled domain.
passages. This was basically used by VISIUNTM,
but also by Rai22 and Dawes6 among others. The comparison between numerical results and
In fact, geometrical transformations are, in the case experimental data presented in the following parts
of centrifugal compressors, mainly applied to the will give an outline of the level of description
vaned diffuser considered passages. Therefore, the obtained with this method.
hub and/or shroud countours have to be ajusted to
ensure the total design throat section conservation if 4.2.5 Steady stage calculations : an industrial
an accurate simulation of stage performance is necessity.
awaited. Then, in a code assessment contex, The previously described unsteady calculations
interpolations in the numerical flow field to probe involved a full time resolved interpolation process
the solution at the exact measurement locations at the rotor-stator interface and a slow convergence
have to bear reduced space considerations towards a periodic numerical solution. Considering
Moreover, some cases lead to a slightly diverging current hardware capabilities available to
vaneless space which may adversly affect a aeronautical design engineers, a time mean
commonly observed shroud recirculating region. interfacial treatment becomes an industrial
Therefore, an alternative method was preferred by necessity. Such a calculation tool, taking into
the authors with a view to computing the flow in account rotor-stator coupling mainly in terms of
the actual geometry thanks to minor numerical hub to shroud distortions, is already in use during
modifications. This reduction technique, developed design cycles. For instance, three dimensional
by the ONERA (see for instance Fourmaux23), may geometries of components, fitting intra-stage
be summarised as follows : let Nl and N2 be the interactions, can be studied within short term
real blade numbers of the rows to study, let Kl and development cycles.
K2 the number of passages considered in each stage
 22-l

The method implemented in VISIUNTM by the

authors and presented herein is based on Riemann For inlet flows, V. n > 0, one compatibility
invariants theory and the derived compatibility relation is applied (one negative eigenvalue) and
relations. This approach comes from two four boundary conditions are necessary. Then the
considerations : relations are :
l Compatibility relations are a way to take
into account the transport of flow IP - (pa) Vi, = P* - (pa)V,:
I I

l
properties by wave frontsZ4.
A grid point, steady in its own frame,
receives informations from opposite points
I P - (a)V;, = P* - (a2)V,: = I,

in the other frame points in the other frame

rotating in front of him.

Then, the approach chosen consists in imposing

P + (pa) Vi, = (P* + (pa)V,:), = I,
interfacial boundary conditions through azimuthaly where 12,13,14 and 15 are averaged by the same
meaned compatibility relations. The French previous process and result in hub to shroud
ONERA, first described this technique5.26. varying boundary conditions.

Let D be the computational domain and n an inward Numerical flow fields obtained through this
directed vector. The five compatibility relations approach will be compared to experiments in the
obtained from the five eigenvalues sixth part.

(V . n - a, V . n, V . n + a) of the hyperbolic 5. UNSTEADY FLOW CALCULATIONS AND

Euler system are : ASSESMENT.

5.1 Numerical simulations conducted.

The real ratio of blade numbers between the wheel
iii:;:~ and the stator is surrounded by I and l/2. Then,
two basic mesh configurations were built : the first
one contained one blade passage of each
where P is the static pressure, p the density, a the component and will be named case A, the second
sound speed, V the speed vector, V,, = V. n and one matched one blade passage of the impeller with
V, = V - (V . n) . n . The superscript * indicates a two passages of the vaned diffuser, named case B.
In both casesan approximate number of 125000
scheme value and indicates the flow properties at
nodes for each passagemeshedwas chosen which
the position receiving the corresponding
is quite low but planned to give an interesting
characteristic.
compromise between accuracy and solution time.
As will be stated later somefluctuations induced by
The implementation for the interface mean
impeller wakes were partly damped in those first
treatment can be summarised as follows. The
cases,a third configuration was then meshedwith
normal Mach number is supposed to be less than
an increased node number in the azimuthal
unity, inward and outward flow are distinguished.
direction. This casewill be referred as caseC.
For exiting flow, V on < 0, four compatibility
relations are applied (four negative eigenvalues)
and one boundary condition is necessary. This
I I Impeller 1Diffuser 1
I

CaseA 82;61x24 74x69~24

condition is similar to a non-reflection boundary
CaseB 82x61~24 2*(74x69x24)-1
condition. Then we have :
CaseC 82x82~24 2*(74x81x24)-1
P - (pa) Vi, = P* - (pa)V,t
Table I- Node distribution.
P -(a) Vi, = P* - (a*)V,:
For each configuration, the flow field of the peak
v; = VT efficiency point on the design speedcurve hasbeen
computed. A time-periodic solution was obtained in
P + (pa) Vl, = (P* + (pa)V,:), = I, all cases. Convergence was monitored by
where I1 is an azimuthaly averaged value examining static pressure evolution in the most
calculated on the opposite domain. The averaging sensitive parts of the flow field i.e. the impeller
process is surface weighted and results in hub to trailing edge and the diffuser leading edge. An
shroud profile of I,.
22-8

approximate number of 10 to 15 blade passing position of 0.3. More details are given by Domercq
period were necessary to obtain a periodic solution. and Thomas*.

5.2 Mono-dimensional results.

From the designer point of view, the overall
performance rendering capability of a code is of
major importance, allowing him to evaluate a new
designed release numerically without requiring
intensive test rig campaign. The comparisons
made on one-dimensional mean values can be
summarised in the following chart. The values
given are differences between experimental and
numerical results expressed in percent. Figure 7 - Time averaged spanwise evolutions.
The two calculations performed on the A and B
basic grids, restitute the same spanwise variation in
terms of velocity and angle, thus illustrating their
close one-dimensional behaviour. Besides, rather
accurate descriptions of the hub to shroud
experimental evolution are obtained. In particular,
the gradient of velocity is shown by both (A and B)
calculated profiles. The mean level difference is
linked to the slight mass-flow discrepancies in
calculations, see Table 2. Similarly, the calculated
and measured flow angles denote an interesting
Table 2- Mono-dimensionnal results. agreement. The shroud area subdeflexion , known
The operating point is then quite accurately to result from a complex interaction between tip
described in all cases. The efficiency lack generally leakage vortex, passage vortex and horseshoe
observed is similar to the difference experienced in vortex (Kang and Hirsch) is precisely given by
isolated wheel simulations. Moreover, it should be both calculations. Concerning. Case C, similar
noted that previous simulations obtained by the profiles are obtained with the increased node
sliding of impeller outflow conditions at the inlet of density. In particular, flow angle evolutions are
an isolated diffuser grid did not allow to reach such identical to the previous (A and B) results
a level of description of the diffuser operating
point. For instance the static pressure recovery 5.4 Unsteady assessment and unsteady flow
coefficient was underestimated by more than 20%, invetsigation.
thus suggesting, the importance of full stage
coupling in a transonic configuration. 5.4. I Unstea& assessment.

In the following part, we will focus our attention on

5.3 Two-dimensional results. case C results which, regarding the higher
The first validation step consists in comparisons azimuthal node density, gave suitable results to face
between measured and time averaged calculated experiments. Figure 8 and Figure 9 represent the
spanwise evolution of the absolute flow angle and absolute flow angle and velocity at midspan for
the absolute flow velocity. This phase will allow three reduced pitch positions of 0.3, 0.5 and 0.7.
the evalution of the code capability to reproduce the The measurement section considered is still section
hub-to-shroud distortions which, according to 1. i.e. the mean vaneless snace radius.
Dawes6 are a key factor of rotoristator interaction Rsduced pttch.0.5.

influence on stage performance. On Figure 7, the y-

axis coordinate represents the hub-to-shroud
reduced distance, where 0. corresponds to the hub.
Either the absolute flow angle (counted between the
velocity direction and the circumferential
direction), to which a reference angle is substracted Figure 8 - Midspan unsteady evolution of the
or the absolute velocity modulus (reduced by the absolute flow angle.
rotor tip speed) are given on the x-axis. Those
evolutions are extracted at the mean vaneless space
radius and here are given at a reduced pitch
 22-9

position. Small structure of the pressure field are

described with accuracy for locations 1 and 3. The
evolution of the signal between location 1 and 2,
consisting in the separation of the flat profile in a
main and secondary peak is also captured. The
description of this dual shape profile is confirmed
Figure 9 - Midspan unsteady evolution of the by location 3 comparison. However, an inadequate
reduced absolute velocity. numerical damping appears at location 4, since only
half of the signal amplitude has been simulated
numerically.
A strong blade wake influence characterises the
experimental results. This region is occupied by a
low velocity fluid attached to the trailing edge in
the rotating frame of reference. A speed triangle
analysis shows that this results in a low, i.e. rather
circumferential, absolute flow angle and high
speed pattern in the absolute frame of reference.
Those wakes are indeed larger near the hub, since
the trailing edge thickness increases from hub to
shroud. Moreover, a steady point ( i.e. a L2F
measurement point in the diffuser frame of
reference sees the pressure side of the backswept
blade before its suction side. The experimental
gradients observed thus suggest that the absolute
speed magnitude decreases from suction side to
pressure side while the absolute flow angle
increases. Then, in the relative frame, the impeller
delivers fluid of higher energy near the pressure Figure 10 - Confrontation of pressure evolutions.
side. Thus suggesting that the Coriolis effect that
generally drags low energy particles to the suction
side takes advantage, in our case, on the curvature As a conclusion, the unsteady assessment reveals to
effect, which on the contrary leads low energy fluid be quite satisfactory, since the global behaviour of
particles to the centre of curvature, that in our the matching between the impeller and its vaned
backswept case corresponds to the pressure side. diffuser is given by the code.
The comparison between experimental data and
computational results shows that the wakes 5.4.2 Unsteady flow investigation from impeller
extensions are underestimated by the numerical outlet to the captation area.
simulations. However, the magnitude of the Figure 11 and Figure 12 give the time-resolved
fluctuations is quite well estimated. A 10 to 12 evolution of the unsteady flow field through its
angle fluctuation is obtained numerically whereas static pressureand absolute Mach for six positions
around 15 were measured. In terms of velocity, the acrossthe blade passingperiod.
predicted order of variation is situated between 17
to 25% where the corresponding L2F results range
from 17 to 30 %. More precisely, despite the wake
extension reduction, the flow angle and speed
modulus blade-to-blade gradient, linked to
secondary flow behaviour, are accurately rendered.
Moreover, the speed modulus fluctuations are
reduced in the same way with increasing reduced
pitch position. Nonetheless, the downstream
propagation of the numerical wake, thinner than the
experimentally observed corresponding patterns
will also be subject to faster mixing in the captation
area.
Figure 10 presents the comparisons between case C
and unsteady pressure measurements obtained with Figure 11 - Unsteady flow field, static pressure.
Kulite transducers. A fair agreement is obtained.
The shape of the fluctuation is rendered for each
 The geometry of the compressor previously
investigated in an unsteady mode was submitted to
a steady calculation involving the coupling process
describedin paragraph4.2.5.
Table 3 sums up the discrepancies obtained
between the time averaged flow field of caseC and
the steady simulation name case D. They are
expressedin percent.

_.-._
Figure 12 - Unsteady flow field, Mach number.

Those representations are given at midspan where

experiment versus simulation agreement was
interesting. The rotor/stator interaction appears to
be reciprocal, the wheel indeed delivers pitch
dependant flow conditions as discussed previously
but the magnitude of stator vane counter effect
reaches the same level. In fact, the pressure field
Table 3 - Mono-dimensionnal comparison of
representation, for instance, underlines the diffuser
caseC (unsteady) and D (steady).
vane leading edge influence on the impeller
outflow pressure condition through the existence of Close performance description is given by the two
a high static pressure zone near the stagnation point coupling techniques. An increased level of loss is
situated on the leading edge. This leads to an predicted betweenthe impeller trailing edge and the
approximate variation of 20% of static pressureon radial diffuser leading edge by the steady coupling
the pressureside of the impeller blades. calculation. This mainly results from slight
Moreover, the blade passingwakes have an obvious discontinuities generatedat the averaging plane.
influence on the captation area flow field, in Finally, spanwiseevolutions of the absolute flow
particular in the vane suction side nearby. The angle and absolutereduced speedof casesC and D
impeller blade trailing edge imposesa high absolute are compared on Figure 13. The profiles are
Mach number and low absolute angle flow pattern extracted at the leading ec e radius of the stator.
to the diffuser inlet. This zone has hardly no effect
on the diffuser pressure side since its extend is
reduced by the leading edge high pressurefield,
when impeller and diffuser blades face each other
(t/T=O.).
Then, it affects the diffuser suction side by
dragging a low Mach number fluid into the channel
(t/T=0.2). The next step consistsin the creation of a --%
high Mach number flow region in the middle of the
diffuser pitch (t/T=0.4), which stands in place of a Figure 13 - Comparisons of spanwiseevolutions.
similar pattern appeared at t/T=0 and observed
whilst diffusing slightly downstreamat this reduced The two profiles exhibit a good agreement.The hub
time. Thus the captation area experiences a to shroud gradients are accurately rendered. The
successionof high speedand low speedbulbs that discrepancy in terms of absolute flow angle is less
mix downstream. The blade-to-blade flow than 1. Thus a steady coupling calculation proves
dependenceat the impeller outlet leadstherefore to to be able to simulate the correct evolution of
a fluctuation that follows the main flow direction. incidence angle and speedon the span of the radial
diffuser. Then it can be consideredas an interesting
6. STEADY SIMULATIONS OF BLADE ROW tool to take into account the coupling between the
INTERACTIONS. components of a centrifugal compressor stage
thanks to calculations of a design affordable cost.
This part of the study will give an overview of the
code capability to describe blade row interaction
within the frame of steady coupling.
 22-11

7. CONCLUSIONS. 3 Krain H., Experimental observations of the flow

The flow patterns that govern the rotor/stator in impellers and diffusers
interaction in a centrifugal compresssor stage have VKI Lecture Series 1984-07, 1984.
been investigated by means of experimental and
numerical studies. The test stage representative of 4 Bois G., Duchemin J.M., Vouillarmet A.,
high speed and high performance turbomachinery Papaillou K.D.
devices, was known to be a rather severe Analyse expCrimentale de ICcoulement dans un
aerodynamic case. The present paper yields to the Ctage de compresseur centrifuge.
following conclusions concerning the unsteady part AGARD-CP-282. 1980.
of the study :
5 Hus H., Fradin Ch.
. An intense upstream influence of the vaned Influence de IhCt&o&nCitC de ICcoulement B la
diffuser on the impeller flow field is observed. sortie du rotor sur les performances du diffuseur
The performance of the compressor is then dun compresseur centrifuge.
properly caculated by a full time-resolved AGARD-CP-282, 1980.
simulation of the stage, whilst a method
consisting in using impeller outflow as upstream 6 Dawes W.
boundary condition of the vaned diffuser A simulation of the unsteady interaction of a
reveals unsatisfactory. centrifugal impeller with its vaned diffuser : flow
analysis.
l Unsteady variations measured thanks to a L2F ASME paper 94-GT- 105.
device were qualitatively fairly described. The
agreement in terms of blade to blade gradients, Bois G., Fradin Ch., Vignau H.
for instance, put emphasis on the code Probl6mes de validation des codes Euler 3D pour
capability to reproduce secondary flows compresseurs centrifuges.
influence on the stage flow field. Moreover this Third European Propulsion Forum , November 13-
gave an outline of the computational domain 15. 1991. AAAFIDGLRIRAeS.
reduction as an acceptable representation.
* Schold R..
l The interaction of the impeller blade leading Laser Two Focus Velocimetry.
edge and the vaned diffuser stagnation zone at AGARD-CP 399, 1986.
midspan appeared to drive both the impeller
outflow conditions and the unsteady flow 9 Fagan J.R., Fleeter S.
patterns in the captation area. L2F & LDV velocimetry measurement and analysis
of the 3D flow field in a centrifugal compressor.
Moreover, a method allowing design time- AIAA-89-2572.
affordable calculations, which couples components
thanks to a mean treatment of the interfacial area, Elder R.L., Forster C.P., Gill M.E.
using Riemann invariant theory, has been Application of Doppler and Transit Laser
implemented in the Navier-Stokes code. The Anemometry on small Turbomachines.
comparison made between the steady stage flow AGARD-CP-399.
field and the time-averaged flow field of unsteady
calculations underlines the capability of this tool to Fradin Ch.
integrate rotor-stator interaction features in the Constitution dune base de donnCes relative a
calculations of the performance of a transonic 1Ccoulement du fluide dans un compresseur
centrifugal stage. centrifuge.
ONERA Report.
References : I2 R. W. Ainsworth
P.Belaygue, H.Vignau,
Unsteady Measurements Techniques.
Le compresseur centrifuge, composant essentiel
VKI Lecture Series.
des turbomoteurs de petite et moyenne puissance.
AGARD-CP-537.
G. Cicatelli, C.H.Sieverding
The Effect of Vortex Shedding on the Unsteady
2 Krain H.,
Pressure Distribution Around the Trailing Edge of a
A study on centrifugal impeller and diffuser flow
Turbine Blade.
ASME paper 8 1-GT-9, 198 1.
22-12

Journal of Turbomachinery, Transaction of the 23Fourmaux A.,

ASME, Vol. 119, October 1997. Assessment of a Low storage Technique for Multi-
Stage Turbomachinery Navier-Stokes
I4Users Guide VISIUN-2FRTM Version 3.0, Computations.
Simultaneous three-dimensional CFD analysis of ASME Winter Annual Meeting, Chicago, USA,
rotating and stationary bladed passages in November 6- 11, 1994.
turbomachinery, a computational system.
Northern Research and Engineering Corporation, 24JP Veuillot, G. Meauze
Wobum, Massachusets, USA. A 3D Euler Method for Internal Transonic Flows
Computation with a Multi-domain Approach.
SAnalysis of three-dimensional unsteady flow in AGARD-LS- 140
centrifugal compressors.
Final report, phase one : diffuser performance. A. LEMEUR
Principal investigator : Oreper G. Calculs stationnaires et instationnaires dans un
Northern Research and Engineering Corporation, &age de turbine transsonique
Wobum, Massachusets, USA. AGARD - CP- 5 10.

I6 Analysis of three-dimensional unsteady flow in 26C. Toussaint, A. Fourmaux, G. Billonnet

centrifugal compressors. Comparison of Steady and Unsteady 3D Viscous
Final report, phase two : impeller performance. Flows Computations through a transsonic turbine
Principal investigator : Oreper G. stage.
Northern Research and Engineering Corporation, ISABE, November 1997.
Woburn, Massachusets, USA.
270. Domercq, R. Thomas.
Analysis of three-dimensional unsteady flow in Unsteady Flow Investigation in a Transonic
centrifugal compressors. Centrifugal Compressor Stage.
Final report, phase three : stage performance. AIAA-97-2877
Principal investigator : Oreper G.
Northern Research and Engineering Corporation, 28Hirsh Ch., Kang S., Pointel G.
Wobum, Massachusets, USA. A numerically supported investigation of the 3D
Flow in Centrifugal Impellers, Part II : Secondary
* Pulliam T.H., Steger J.L. Flow Structure
Implicit Finite-Difference Simulations of Three- ASME paper 96-GT- 152
Dimensional Compressible Flow.
AIAA Journal, Vol. 18, February 1980.

I9Anderson D.A., Tannehill J.C., Plechter R.H.

Computational Fluid Mechanics and Heat
Transfer-t.

2oBaldwin B.S., Lomax H.,

Thin layer approximation and Algebraic Model for
Separated Turbulent Flows.
AIAA paper 78-257, 1978.

Beam R.M., Warming R.F.

An implicit factored scheme for the compressible
Navier-Stokes Equations.
AIAA Journal, Vol. 16, April 1978.

22Rai M.M.
Navier-Stokes Simulations of Rotor-Stator
Interaction using Patched and Overlaid Grids.
AIAA paper 85-1519.