Alex Diss PDF

COMPUTATIONAL ANALYSIS OF STALL AND SEPARATION CONTROL
IN CENTRIFUGAL COMPRESSORS
A Thesis
Presented to
The Academic Faculty
by
Alexander Stein
In Partial Fulfillment
of the Requirements for the Degree
Doctor of Philosophy in Aerospace Engineering
Georgia Institute of Technology
May 2000
ii
COMPUTATIONAL ANALYSIS OF STALL AND SEPARATION CONTROL
IN CENTRIFUGAL COMPRESSORS
Approved:
______________________________
Lakshmi N. Sankar, Chairman
______________________________
Suresh Menon
______________________________
J.V.R. Prasad
Date Approved__________________
iii
Dedicated to
my Lord Jesus Christ
But seek first His kingdom and
His righteousness, and all these
things will be given to you as well.
TheGospel of Matthew, Chapter 6, Verse33
iv
ACKNOWLEDGEMENTS
I would like to express my sincere gratitude to Dr. Lakshmi N. Sankar, my thesis
advisor, whose delightful personality and detailed knowledge about this research topic
has guided me along the way. His encouragement, patience and helpful suggestions
during times of little progress have left a deep impression on my life.
I would also like to thank Dr. J. V. R. Prasad, Dr. S. Menon, Dr. B. T. Zinn and
Dr. A. Glezer for serving as thesis committee members and for their valuable comments.
I have greatly benefited from our monthly meetings as part of the MITE program and
have gained a better understanding of gas turbine technology.
I would like to thank Dr. M. Hathaway from NASA Glenn Research Center for
providing the LSCC computational grid and experimental data. I also wish to thank Dr.
H. Krain from Deutsche Luft- und Raumfahrt for making his experimental results
available to me.
I would like to acknowledge the U.S. Army Research Office under the
Multidisciplinary University Research Initiative (MURI) on Intelligent Turbine Engines
as the sponsor of this research. High-performance computer time was provided by the
Major Shared Research Center of the U.S. Army Engineer Research and Development
Center (ERDC MSRC).
I would also like to thank my friend and co-worker Saeid Niazi for his great effort
during code development and his continued support throughout my studies.
v
I am especially thankful to my wife Jane Carole for her unconditional love and
selflessness. Without her encouragement and prayers, this work would not have been
possible. She also spent countless hours proofreading and revising this thesis. In
addition, I would like to thank my father-in-law, James C. Meredith, for taking the time
to read this thesis and make helpful suggestions.
I am also greatly thankful to my parents, Lothar and Christel Stein, who have
generously supported my undergraduate studies in Germany and loved me all these years.
Finally, I thank my Lord and Savior Jesus Christ who has given me the strength
and joy to persevere in this work. He has not only provided for all of my needs but has
given me everlasting life through His redemptive sacrifice on the cross.
vi
TABLE OF CONTENTS
DEDICATION III
ACKNOWLEDGEMENTS IV
TABLE OF CONTENTS VI
LIST OF TABLES IX
LIST OF FIGURES X
NOMENCLATURE XVII
SUMMARY XXV
1 INTRODUCTION 1
1.1 BACKGROUND............................................................................................................................. 1
1.2 OPERATION AND PERFORMANCE .................................................................................................. 2
1.3 ROTATING STALL ........................................................................................................................ 5
1.4 SURGE......................................................................................................................................... 6
1.4.1 Mild Surge.......................................................................................................................... 6
1.4.2 Deep Surge......................................................................................................................... 7
1.5 STABILITY OF COMPRESSION SYSTEMS......................................................................................... 8
1.6 COMPRESSOR CONTROL............................................................................................................. 11
1.7 ANALYTICAL AND COMPUTATIONAL MODELING OF COMPRESSOR CONTROL ............................... 16
vii
1.8 GOALS OF COMPRESSOR CONTROL RESEARCH............................................................................ 19
2 OVERVIEW OF THE PRESENT WORK 27
2.1 MAJOR STEPS IN THE PRESENT RESEARCH.................................................................................. 28
2.2 POTENTIAL AND LIMITATIONS.................................................................................................... 30
3 NUMERICAL ANALYSIS TOOLS 33
3.1 THREE-DIMENSIONAL GOVERNING EQUATIONS.......................................................................... 34
3.2 DISCRETIZATION OF NAVIER-STOKES EQUATIONS ...................................................................... 37
3.2.1 Roes Flux Difference Splitting ......................................................................................... 39
3.3 LINEARIZATION AND APPROXIMATE FACTORIZATION ................................................................. 42
3.4 TURBULENCE MODELING........................................................................................................... 46
3.5 INITIAL AND BOUNDARY CONDITIONS........................................................................................ 50
3.5.1 Initial Conditions.............................................................................................................. 50
3.5.2 Boundary Conditions ........................................................................................................ 51
3.6 LOGICAL STRUCTURE OF FLOW SOLVER..................................................................................... 58
4 CODE VALIDATION STUDIES 63
4.1 COMPRESSOR CONFIGURATIONS AND GRID GENERATION............................................................ 64
4.1.1 NASA Low-Speed Centrifugal Compressor (LSCC) ........................................................... 64
4.1.2 DLR Centrifugal Compressor (DLRCC)............................................................................ 65
4.2 VALIDATION OF THE NASA LOW-SPEED CENTRIFUGAL COMPRESSOR (LSCC)............................ 68
4.2.1 Validation Results at Design Operating Conditions........................................................... 69
4.2.2 Validation Results at Off-Design Operating Conditions..................................................... 73
4.3 VALIDATION OF THE DLR CENTRIFUGAL COMPRESSOR (DLRCC) .............................................. 74
5 COMPUTATIONAL RESULTS AT STALL CONDITIONS 90
5.1 HIGH-SPEED COMPRESSOR RESULTS AT OFF-DESIGN CONDITIONS.............................................. 91
viii
5.2 HIGH-SPEED COMPRESSOR RESULTS DURING STALL CONDITIONS............................................... 93
5.3 LOW-SPEED COMPRESSOR RESULTS AT OFF-DESIGN CONDITIONS............................................... 96
6 AIR INJECTION COMPRESSOR CONTROL 109
6.1 COMPUTATIONAL MODELING OF AIR INJECTION CONTROL ....................................................... 112
6.2 NUMERICAL RESULTS OF STEADY AIR INJECTION ..................................................................... 113
6.3 NUMERICAL RESULTS OF PULSED AIR INJECTION...................................................................... 123
7 CONCLUSIONS AND RECOMMENDATIONS 145
7.1 CONCLUSIONS ......................................................................................................................... 146
7.2 RECOMMENDATIONS................................................................................................................ 147
REFERENCES 150
VITA 163
ix
LIST OF TABLES
Table 3.1 Non-Dimensionalization Factors in the Flow Solver.....................................37
Table 3.2 Constants in Spalart-Allmaras Turbulence Model.........................................50
Table 4.1 Comparison of Centrifugal Compressors......................................................68
Table 5.1 Comparison of Relative Mach Numbers and B-Parameters for the Operating
Conditions Considered at 100 Percent Design Speed and 200 Percent Design
Speed......................................................................................................... 100
Table 6.1 Overview of Experimental Injection Schemes ............................................ 111
Table 6.2 Comparison Between Results of Yaw Angle Criterion and CFD Results .... 119
Table 6.3 Comparison Between Neural Network Predicted and CFD Predicted Surge
Amplitudes................................................................................................. 123
x
LIST OF FIGURES
Figure 1.1 Volumetric Size Range for Different Categories of Compressors ..............21
Figure 1.2 Cutaway View of Allied Signal TPE331-14 Turboprop Gas Engine with
Two Centrifugal Compressor Stages .........................................................21
Figure 1.3 Schematic of Compressor Performance Characteristic Map (source:
Hunziker
64
) ...............................................................................................22
Figure 1.4 Centrifugal Impeller Damage Due to Compressor Instabilities (source:
NASA)......................................................................................................22
Figure 1.5 Impeller Damage Due to Compressor Instabilities (source: NASA)...........23
Figure 1.6 Rotating Stall Pattern ................................................................................23
Figure 1.7 Schematic of Mild and Deep Surge Cycle .................................................24
Figure 1.8 Helmholtz-Resonator-Model (source: Greitzer
11
) ......................................24
Figure 1.9 Schematic of Static and Dynamic Instability (source: Greitzer
11
)...............25
Figure 1.10 Active and Passive Compressor Control Schemes .....................................25
Figure 1.11 Development of Computational Resources ................................................26
Figure 2.1 Centrifugal Compressor Configurations Used in This Work......................32
Figure 3.1 Cell-Centered Finite Volume Formulation and Four-Point Stencil .............59
Figure 3.2 Boundary Conditions for Compressor Single Flow Passage.......................59
Figure 3.3 Coupling Between Diffuser and Plenum at Outflow Boundary ..................60
Figure 3.4 Zonal and Periodic Boundaries..................................................................60
xi
Figure 3.5 Injection Boundary....................................................................................61
Figure 3.6 GTTURBO3D Flow Chart ........................................................................62
Figure 4.1 Low-Speed Centrifugal Compressor (LSCC) in NASA Test Rig (source:
NASA)......................................................................................................77
Figure 4.2 Schematic of Low-Speed Centrifugal Compressor (LSCC) .......................77
Figure 4.3 Single Flow Passage Computational Grid for LSCC (129x61x41) .............78
Figure 4.4 DLR Centrifugal Compressor (DLRCC), (source: Krain
97
)........................78
Figure 4.5 Single Flow Passage Computational Grid for DLRCC (141x49x33)..........79
Figure 4.6 Computed and Measured
85
Blade Surface Pressure, p/p
, at 5 Percent Span,
LSCC Operating Design Conditions (30kg/sec).........................................79
85
85
85
85
85
85
Axial Velocity, u/U
tip
, at 25 Percent Chord,
xii
85
Axial Velocity, u/U
tip
85
Axial Velocity, u/U
tip
Figure 4.15 Velocity Field 4%, 50% and 97% Away from the Pressure Surface, LSCC
Operating Design Conditions (30kg/sec) ...................................................84
85
LSCC Operating Off-Design Conditions (23.5kg/sec) ...............................84
85
85
85
85
85
98
Static Pressure Along Shroud (Circumferentially
Averaged), p/p
0
,, DLRCC Operating Design Conditions (4.0 kg/sec) ......87
94
Meridional Velocity, c
mer
/U
tip
, at 20% Chord,
DLRCC Operating Design Conditions (4.0 kg/sec) ...................................88
xiii
94
mer
/U
tip
, at 60% Chord,
94
mer
/U
tip
, at 99% Chord,
Figure 4.26 Computed Meridional Velocity Vectors Colored by Total Pressure, DLRCC
Operating Design Conditions (4.0 kg/sec) .................................................89
94
DLRCC Performance Map With the Amplitude
of the Fluctuations Denoted by Horizontal and Vertical Bars................... 102
Figure 5.2 Computed Time History for Selected Points A-D on DLRCC Performance
Map, A: sec / kg 6 . 4 m & , B: sec / kg 8 . 3 m & , C: sec / kg 4 . 3 m & , D:
sec / kg 2 . 3 m & ....................................................................................... 102
Figure 5.3 Computed Transient Response of the Mass Flow Rate in the DLRCC During
Limit Cycles (Point D in Figure 5.2) ....................................................... 103
Figure 5.4 Growth of Reversed Flow Regions in the DLRCC During Limit Cycles
(Point D in Figure 5.2) ............................................................................ 103
Figure 5.5 Relative Velocity Vectors Near Blade Leading Edge During DLRCC Limit
Cycles (Single Flow Passage Top View V-V at 99 Percent Span)............ 104
85
LSCC Performance Map With the Amplitude of
the Fluctuations Denoted by Horizontal and Vertical Bars....................... 105
Figure 5.7 Growth of Shroud Separation Zone in LSCC at Design Speed and Different
Operating Conditions, Impeller Regions are Shaded................................ 105
xiv
Figure 5.8 Velocity Vectors Near Leading Edge During LSCC Limit Cycles at 200
Percent Design Speed and at 50 Percent Pitch; Shading Indicates the
Impeller................................................................................................... 106
Figure 5.9 Unsteady Pressure Rise During LSCC Limit Cycles at 200 Percent Design
Speed...................................................................................................... 107
Figure 5.10 Computed Frequencies During LSCC Limit Cycles at 200 Percent Design
Speed...................................................................................................... 107
Figure 5.11 Computed Wave Numbers During LSCC Limit Cycles at 200 Percent
Design Speed .......................................................................................... 108
Figure 6.1 Schematic and Nomenclature of Injected Fluid Sheet .............................. 131
Figure 6.2 Local Incidence Angle Versus Yaw Angle in the DLRCC at 3.2 kg/sec
Mean Mass Flow Rate, 3.2% Injection Rate ............................................ 131
Figure 6.3 Effect of Varying Yaw Angles, , on Flow Rate Fluctuations in the DLRCC
at 3.2 kg/sec Mean Mass Flow Rate, 3.2% Injection Rate........................ 132
Figure 6.4 Effect of Air Injection on Incidence Angles in the DLRCC at 3.2 kg/sec
Mean Mass Flow Rate, 3.2% Injection Rate and 7.5 Degrees Yaw Angle 132
Figure 6.5 Local Incidence Angle Versus Yaw Angle in the LSCC at 3.2 kg/sec Mean
Mass Flow Rate, 200% Design Speed and 5% Injection Rate .................. 133
Figure 6.6 Effect of Varying Yaw Angles, , on Flow Rate Fluctuations in the LSCC at
3.2 kg/sec Mean Mass Flow Rate, 200% Design Speed and 5% Injection
Rate ........................................................................................................ 133
xv
Figure 6.7 Schematic and Nomenclature of Velocity Triangles in Simplified Injection
Model ..................................................................................................... 134
Figure 6.8 Nondimensional Surge Amplitude for Varying Yaw Angles and Varying
Injection Rates in the DLRCC at 3.2 kg/sec Mean Mass Flow Rate......... 134
Injection Rates in the LSCC at 200% Design Speed, 30 kg/sec Mean Mass
Flow Rate................................................................................................ 135
Figure 6.10 Neural Network with Two Hidden Layers and Log-Sigmoid Transfer
Functions ................................................................................................ 135
Figure 6.11 Neural Network Surface Approximation of DLRCC Injection Performance
Map ........................................................................................................ 136
Figure 6.12 Neural Network Surface Approximation of LSCC Injection Performance
Map ........................................................................................................ 136
Figure 6.13 Nondimensional Surge Fluctuations, Incidence Angle and Injection Rate in
the DLRCC at 3.2 kg/sec Mean Mass Flow Rate, 7.5 Degrees Yaw Angle
and 0.023 + 0.007sin(
stall
t) Pulsed Injection Rate................................... 137
and 0.015 + 0.015sin(
stall
t) Pulsed Injection Rate................................... 138
Figure 6.15 Mach Contour Snapshots at Midpassage in the DLRCC at 3.2 kg/sec Mean
Mass Flow Rate, 7.5 Degrees Yaw Angle and 0.015 + 0.015sin(
stall
t)
Pulsed Injection Rate............................................................................... 139
xvi
and 0.023 + 0.007sin(2
stall
t) Pulsed Injection Rate................................. 140
and 0.023 + 0.007sin(2.5
stall
t) Pulsed Injection Rate.............................. 141
and 0.023 + 0.007sin(4
stall
and 0.015 + 0.015sin(4
stall
Figure 6.20 Comparison of Vorticity Magnitudes Using No Jets, Low-Frequency Jets
(
inj
=
stall
) and High-Frequency Jets (
inj
= 4
stall
); the Data was Collected
at Midpassage, 99 Percent Span and 0.025R
inlet
....................................... 144
Figure 6.21 Comparison of Shear Stresses,
xy
, Using Low-Frequency Jets (
inj
=
stall
)
and High-Frequency Jets (
inj
= 4
stall
); the Data was Ensemble-Averaged
over the Region Indicated in the Schematic ............................................. 144
xvii
NOMENCLATURE
a speed of sound
a
p
plenum speed of sound
a
J
free-stream speed of sound
A
c
flow-through area in Helmholtz-Resonator model
A
inj
amplitude of pulsed injection
C
, B
, A
transformed Jacobian matrices

t t t
C
, B
, A
positive and negative transformed Jacobian matrices

B non-dimensional B-parameter
C compressor characteristic
c
b1
, c
b2
constants in Spalart-Allmaras turbulence model
c
mer
meridional velocity
c
p
specific heat at constant pressure
c
v
specific heat at constant volume
c
v1
constant in Spalart-Allmaras turbulence model
c
w1
, c
w2,
c
w3
c
t1
, c
t2
, c
t3
, c
t4
d distance to closest wall in Spalart-Allmaras turbulence model
D
trailing edge
trailing edge diameter
xviii
e internal energy per unit volume
G , F , E
r r r
inviscid flux vectors
G
, F
, E
transformed inviscid fluxes

f
t1
, f
t2
functions in Spalart-Allmaras turbulence model
f
v1
, f
v2
, f
w
functions in Spalart-Allmaras turbulence model
g, g
t
, r functions in Spalart-Allmaras turbulence model
h static enthalpy
h
in
static enthalpy at compressor inlet
h
out
static enthalpy at compressor exit
I identity matrix
I
inj
mean of pulsed injection rate
k , j , i
r r r
cartesian unit vectors
k thermal conductivity
L
c
length of pipe in Helmholtz-Resonator model
n
i
n
M , M matrices in approximate factorization scheme
m& mass flow rate
c
m& mass flow rate at diffuser-plenum interface
inj
m& injection rate
t
m& mass flow rate at plenum throttle
n used as superscript indicating variable at n-th time step
n
r
unit normal vector
xix
N
n
unit normal in Roes flux splitting scheme
p pressure
p
p
plenum pressure
p
ref
reference pressure
p
J
free-stream pressure
q
r
vector of conservative flow variables
0
q
r
nominally steady flow variables
q
i
heat flux vector
r
in
blade radius at compressor inlet
r
out
blade radius at compressor exit
T , S , R
r r r
viscous flux vectors
T
, S
, R
transformed viscous fluxes

R
n
right-hand side in approximate factorization scheme
R
inlet
blade leading edge radius
S vorticity magnitude in Spalart-Allmaras turbulence model
S
~
modified vorticity magnitude in Spalart-Allmaras turbulence
model
s
mer
meridional blade chord
t time
T throttle characteristic or temperature
T
stall
period of limit cycle
xx
T
0
stagnation temperature
u, v, w cartesian velocity components
u
in
circumferential velocity at compressor inlet
u
n
normal velocity component
u
out
circumferential velocity at compressor exit
U contravariant velocity
U
tip
impeller exit tip speed
U
*
conservative flow variables in Roes flux splitting scheme
v
n
normal injection velocity
v
rot
rotor velocity
v
t1
, v
t2
tangential injection velocities
V volume
V
r
velocity vector
Grid
V
r
vector of grid velocity
Solid
V
r
velocity vector of solid wall
v
inj,abs
injection velocity in absolute reference frame
v
inj,rel
injection velocity in relative reference frame
V
ref
reference velocity
V
p
plenum volume
w
in
relative fluid velocity at compressor inlet
w
out
relative fluid velocity at compressor exit
x, y, z cartesian coordinates
xxi
x
ref
reference length
Greek Symbols
q change in conservative flow variables
q
*
, q
**
intermediate changes in conservative flow variables during
approximate factorization scheme
s change in entropy
U velocity at trip point in Spalart-Allmaras turbulence model
t discrete time step
x
t
grid spacing at wall trip point in Spalart-Allmaras turbulence
model
injection angle
yaw angle
abs
yaw angle in absolute reference frame
rel
yaw angle in relative reference frame
difference operator
1
,
2
terms in Roes flux splittling scheme
ij
Kronecker-Delta
q
r
vector of unsteady disturbances
specific heat ratio
flux limiter
xxii
von Krman constant
i
~
characteristic velocities
diagonal matrix in Roes flux splitting scheme

thermodynamic viscosity
kinematic viscosity
t
turbulent viscosity
~
working variable in Spalart-Allmaras turbulence model
total to total pressure ratio
density
p
plenum density
ref
reference density
J
free-stream density
constant in Spalart-Allmaras turbulence model
h
Helmholtz frequency
in
rotational shaft speed at compressor inlet
out
rotational shaft speed at compressor exit
inj
injection frequency
stall
stall frequency
t
wall vorticity at trip point in Spalart-Allmaras turbulence model
time
xxiii
ij
viscous stress tensor
, , curvilinear coordinates
x
,
y
,
z
,
t
coordinate transformation metrics in
Subscripts
0 total (stagnation) quantity
abs quantity in absolute reference frame
c quantity denoting the compressor
i, j, k discrete curvilinear indices
in quantity at compressor inlet
inj injection quantity
inlet quantity at blade leading edge
L quantity to the left of a cell face
out quantity at compressor exit
mer quantity in meridional plane
n normal quantity
p plenum quantity
R quantity to the right of a cell face
ref reference quantity
rel quantity in relative reference frame
stall quantity at stall conditions
t turbulence quantity
xxiv
x, y, z, t cartesian derivatives of subscripted variable w.r.t subscript variable
, , , curvilinear derivatives of subscripted variable w.r.t subscript
variable
J free-stream quantity
fluctuating quantity
Superscripts
n time level
*, ** intermediate quantities in approximate factorization scheme
xxv
SUMMARY
A numerical technique for simulating unsteady viscous fluid flow in
turbomachinery components has been developed. In this technique, the three-
dimensional form of the Reynolds averaged Navier-Stokes equations is solved in a time-
accurate manner. The flow solver is used to study fluid dynamic phenomena that lead to
instabilities in centrifugal compressors.
The results indicate that large flow incidence angles, at reduced flow rates, can
cause boundary layer separation near the blade leading edge. This mechanism is
identified as the primary factor in the stall inception process. High-pressure jets upstream
of the compressor face are studied as a means of controlling compressor instabilities.
Steady jets are found to alter the leading edge flow pattern and effectively suppress
compressor instabilities. Yawed jets are more effective than parallel jets and an optimum
yaw angle exists for each compression system.
Numerical simulations utilizing pulsed jets have also been done. Pulsed jets are
found to yield additional performance enhancements and lead to a reduction in external
air requirements for operating the jets. Jets pulsed at higher frequencies perform better
than low-frequency jets. These findings suggest that air injection is a viable means of
alleviating compressor instabilities and could impact gas turbine technology.
Results concerning the optimization of practical air injection systems and
implications for future research are discussed. The flow solver developed in this work,
xxvi
along with the postprocessing tools developed to interpret the results, provide a rational
framework for analyzing and controlling current and next generation compression
systems.
1
CHAPTER I
1 INTRODUCTION
1.1 Background
A centrifugal compressor is a mechanical device used to increase the static
pressure of a compressible fluid. Some uses of centrifugal compression systems include
gas turbine engines powering tanks and rotorcraft, reaction processes in chemical
industries, compression and transportation of hydrogen in petrochemical refineries, and
refrigeration systems that contain heat pumps powered by small centrifugal compressors.
Compressors are generally categorized by their volumetric size which, in turn,
determines the area of application. Figure 1.1 shows the typical volumetric range for
different categories of compressors. Centrifugal compressors are widely used for
volumetric flow rates of 1,000 to 10,000 ft
3
/min. Multi-stage axial compressors are often
preferred for larger volumetric sizes. Centrifugal compressors are used for low-flow rate
applications because of their ability to achieve higher pressure-rise-to-weight ratios than
axial compression systems. Pressure ratios as high as 12 have been recorded for a single
centrifugal stage
1
, whereas axial systems generally reach pressure ratios up to two.
In gas turbine engines, multiple centrifugal compressors are used, sometimes in
combination with multiple axial stages, to yield high compression ratios before the fluid
2
enters the combustor. Figure 1.2 shows the cutaway view of an Allied Signal TPE331-14
turboprop gas engine with two centrifugal compressor stages.
Isentropic stage efficiencies for modern centrifugal compressors with vaned
diffusers range from 82 to 87 percent. If vaneless diffusers are operated according to
manufacturing constraints, peak efficiencies may only reach up to 80 percent. Polytrop
rotor efficiencies for unshrouded impellers with 25 to 50 degrees backsweep may reach
up to 93 percent
2
.
1.2 Operation and Performance
The purpose of a centrifugal compressor is to produce a distinct increase in static
pressure, thus effectively increasing the static enthalpy. The increase in static enthalpy,
h, across a compressor may be written as
) u u (
2
1
) w w (
2
1
h h
2
in
2
out
2
out
2
in in out
+ (1.1)
where w
in
and w
out
are the fluid velocities measured in the relative reference frame and u
in
= r
in

in
and u
out
= r
out

out
are the circumferential velocities at the compressor inlet and
exit, respectively
3
. As the impeller rotates, gas is moved between the rotating blades
from near the hub radially outward to discharge into a non-rotating section, called the
diffuser. Energy is transferred to the fluid particles as they travel through the rotor. A
portion of the energy is directly converted to pressure while another part is transformed
into kinetic energy. As the fluid is slowed in the diffuser, a second increase in static
3
pressure is achieved. The fraction of the pressure rise that takes place in the rotor is a
function of the backward sweep of the blades. The more radial the blades, the less
pressure conversion is experienced in the impeller, and therefore more conversion occurs
in the diffuser.
While the pressure rise in axial compressors relies solely on the deceleration of
the fluid particles through the stator and work performed by the rotor, centrifugal
compressors experience additional pressure conversion due to the centrifugal pumping
effect. This additional pressure rise takes place isentropically. In a typical centrifugal
compressor stage with perfectly radial trailing edges, about 40 percent of the entire
pressure rise is due to the centrifugal pumping effect, 20 percent is due to the deceleration
of fluid particles in the rotor, and the remaining 40 percent is attributable to the
deceleration of the absolute fluid velocity in the diffuser. Thus, an isentropic increase of
the pressure rise can only be achieved by increasing the rotor exit radius. This increase,
however, is limited by large mechanical stresses in the rotating parts due to the
centrifugal forces. Since a significant portion of the pressure rise is due to the centrifugal
pumping effect, acceptable compressor efficiencies are found even in the case of an
aerodynamically poor rotor leading to local reversed flow. As a result, centrifugal
compressors achieve stability over a wide range of operating points.
A schematic of the centrifugal compressor performance map is shown in Figure
1.3. This diagram represents the variation of the total-to-total pressure rise with the flow
rate across the compressor for a fixed rotational speed. Stable operation of a compressor
is limited at both ends of the abscissa.
4
The choke limit is reached at high flow rates due to the occurrence of sonic
conditions in the throat of the rotor. Any attempt to raise the flow rate beyond this limit
will result in an increased shock strength, thus effectively reducing the compressor
efficiency.
As the flow rate through the system is reduced to small values, strong fluctuations
and limit cycle oscillations are observed. This limit is called the surge limit and is also
referred to as stall line or instability limit. Two types of flow phenomena, rotating stall
and surge, give rise to the surge limit. Near the surge limit, fluctuations can grow in
amplitude and cause considerable fatigue and damage to the entire compression system.
If the fluctuations are left unchecked, even a complete flow reversal is possible, a
situation that an engine cannot tolerate since hot combustion products would damage the
compressor blades and lead to a catastrophic failure of the entire system. Modern
centrifugal compressors with vaneless diffusers and pressure ratios between three and
five achieve a stable flow rate range of 30 to 40 percent based on the maximum flow
rate
2
.
Figures 1.4 and 1.5 show representative examples of compressor damage due to
rotating stall or surge. In Figure 1.4, a picture is shown of the shaft of a centrifugal
compressor with splitter blades that was damaged as a result of excessive loading from
unstable compressor operation. The blade tip damage observed in Figure 1.5 is typical of
adverse rotating stall effects.
5
1.3 Rotating Stall
Rotating stall is a flow phenomenon in which a circumferentially uniform flow
pattern is disturbed by local flow separation. It was first reported by Emmons et al.
4
.
Figure 1.6 schematically depicts the development of this instability. At low flow rates,
the compressor blades are subject to an increased incidence angle which may cause local
leading edge separation on one airfoil due to non-uniform inflow conditions, system
perturbations, or small manufacturing discrepancies. This results in a partial blockage of
the stalled Flow Passage 2 which leads to a deflection of the incoming flow. While Flow
Passage 1 experiences a smaller incidence angle which causes the flow to re-attach
locally, the inflow conditions for Flow Passage 3 produce separation. As a result, local
separation packets move from blade to blade, and appear to rotate about the shaft axis at
an angular velocity, which is one-third to one-half that of the shaft angular velocity.
Although rotating stall is often a precursor to surge, the global compressor flow
rate is largely unaffected by this type of instability. The principal difference between
rotating stall and surge is that the mean flow in rotating stall is steady in time with a
locally rotating mass deficit, while during surge the mean flow is unsteady.
Rotating stall phenomena have been measured in many axial compressors
5-7
.
Periodic blade loadings with oscillations near resonance frequencies can cause blade
damage or even blade fracture. Centrifugal compressor rotating stall is limited to low-
pressure systems and high-pressure systems at partial speed
8
. For high-pressure systems
operating at design speed, the most commonly used centrifugal configuration in industrial
6
applications, rotating stall has been reported to have little effect on pressure rise and flow
rate, and merely serves as a precursor to surge
9
.
1.4 Surge
While centrifugal compressors often show greater resistance toward rotating stall
instabilities than axial systems, large periodic fluctuations in the flow rate and pressure
develop as the system operates at conditions near or beyond the surge line. The
compressor is then referred to as surging. The described fluctuations may reach intensity
levels that jeopardize safe engine operation. As alluded to earlier, surge is a global one-
dimensional instability that can affect the whole compression system. In Reference 10,
the non-stationary flowfield during surge was experimentally captured using Digital
Particle Imaging Velocimetry (DPIV). To characterize surge, the combined system of
inlet, impeller, diffuser, connecting elements and plenum chamber should be viewed as a
damped mass-spring system
11
. In this analogy, the fluid compressibility represents the
potential energy, and the inertia of the particles is the kinetic energy. Limit-cycle
oscillations due to surge occur at frequencies that are equivalent or below the natural
system frequencies. Depending on the severity of the fluctuations, two types of surge
may develop, as illustrated in Figure 1.7.
1.4.1 Mild Surge
As the flow rate through the compressor is decreased, limit cycle oscillations in
virtually all flow properties grow stronger around a mean operating point. These
7
oscillations may be triggered by rotating stall or other non-uniformities in compression
systems that have a large downstream plenum. Operation of the compressor under these
mild surge conditions, however, can be considered stable as long as the dynamic throttle
characteristic of the system is more positively sloped than the compressor characteristic
11
.
Since the dynamic throttle characteristic in systems with a large plenum exhibits a small
slope, the point of dynamic instability is reached near the peak of the compressor
characteristic. The frequency of mild surge fluctuations is often close to the natural
frequency of the compression system
15
.
1.4.2 Deep Surge
Deep Surge is the most severe form of all of the instabilities in which even a
complete flow reversal over the entire annulus is possible. Figure 1.7b shows the
development of a deep surge cycle. Starting at a maximum pressure rise operating
condition on the performance characteristic map, even a small flow perturbation may be
sufficient to destabilize the peak operating point M. Due to the large volume, the plenum
pressure may not be affected at first; however, the compressor is no longer capable of
producing the desired peak pressure. This causes the flow rate to drop until complete
flow reversal at point A is encountered. Flow reversal is extremely dangerous in gas
turbines because of the possible back flow of hot combustion fluid particles into the
compression system. Compressor operation at point A leads to a gradual drop in pressure
throughout the whole system and to a change from point A to point B along the
compressor characteristic. A tendency to further decrease the system pressure causes the
8
operating condition to jump from point B to point C. After this jump, the compressor
resumes building up static pressure along the compressor characteristic until both
compressor and plenum operate again at a peak pressure point M. The surge cycle
repeats continually.
Amplitudes and frequencies during deep surge cycles depend upon the size of the
system, in particular, the plenum chamber. Compression systems with large plenum
volumes possess characteristic time scales for pressure drop and pressure recovery that
lead to surge frequencies well below mild surge frequencies. In centrifugal compression
systems with small plenums, the time needed for emptying and filling of the plenum is
greatly reduced, and the above-described oscillations at small flow rates are suppressed.
In such a situation, local flow separation may still occur.
Due to the fact that deep surge is almost always preceded by mild surge in
centrifugal compressors, most research efforts in compressor control are focused on
understanding and avoiding factors that lead to the development of mild surge.
1.5 Stability of Compression Systems
Two stability criteria for compression systems may be derived based upon the
damped mass-spring system analogy outlined in Section 1.4 to describe instabilities. This
idea of mapping a real axial or centrifugal compression system to a simple one-
dimensional non-linear model was pioneered by Greitzer
11,12
. The compressor is viewed
as an actuator disk, the fluid inertia is exclusively contained in the pipes of length L
c
with
flow-through area A
c
, and the spring-like system properties are confined to the plenum
9
with volume V
p
. A schematic of the Helmholtz-Resonator-Model is shown in Figure 1.8.
Assuming that the compressor characteristic, C, and the throttle characteristic, T, are
known, a non-linear analysis leads to two stability criteria:
(A) the compression system is statically stable if the slope, C, of the compressor
characteristic, C, is smaller than the slope of the throttle characteristic, T.
(B) the compression system is dynamically stable if the slope, C, of the compressor
characteristic, C, is smaller than 1/B
2
T where T is the slope of the throttle
characteristic and B is the B-parameter due to Greitzer
13,14
.
These two criteria are schematically illustrated in Figure 1.9. The dimensionless B-
parameter is defined as
c c
p
c h
A L
V
a 2
U
L 2
U
B
(1.2)
where
h
is the Helmholtz frequency, U is the rotor speed, L
c
is the effective length of the
compressor duct, A
c
is the compressor flow-through area, a is the speed of sound, and V
p
is the plenum volume.
Criterion (A), in the above equation, implies that a compression system becomes
unstable if the compressor characteristic, C, is steeper than the throttle characteristic, T.
Depending on the value of the B-parameter, the system may experience instabilities at
even higher flow rates. According to criterion (B), this can occur for very large B-
parameters, since in this case the criterion becomes C > 0. This situation is reached
slightly to the left of the compressor characteristic peak. Thus, stable operation of such a
10
system is restricted to points that lie to the right of the peak with inherently negative
slopes.
Greitzer further showed that for surge to occur, the system B-parameter must be
greater than a critical B-parameter. The definition of the B-parameter illustrates that
large values are achieved through high rotor speeds, U, or large plenums, V
p
. Operation
of a system with the B-parameter less than the critical B-parameter leads to the
development of rotating stall instabilities rather than surge. The critical B-parameter
depends on geometric and operational system parameters; its exact value, however, varies
from system to system.
Although it is necessary to look at each compression system individually,
additional understanding of the physical processes leading to instabilities may be gained
by re-writing the definition of the B-parameter (Equation (1.2)) as
forces inertia
forces pressure
A L U
A
2
U
B
c c h
c
2
,
_
(1.3)
Equation 1.3 illustrates that for large B-parameters, the pressure forces dominate over the
inertia forces and produce a driving force for the acceleration of the fluid in the system.
Therefore, global oscillations, in this context referred to as surge, grow stronger and
destabilize the system. Conversely, in systems with small B-parameters, the inertia
forces dominate over the pressure forces and cause the oscillations to decay.
Fink et al.
15
showed that the system response of a centrifugal compressor is
similar to that of an axial system. Both types of systems experience a greater tendency
11
toward surge at large rotor speeds of larger plenums, thus effectively increasing the B-
parameter.
1.6 Compressor Control
All types of the above described fluid dynamic instabilities limit the overall
compressor performance and jeopardize safe operation of the entire engine. Unsteady
fluctuations, caused by rotating stall or surge, may lead to excessive heating of the
impeller blades and to a greater compressor exit temperature. Large amplitude
fluctuations can cause additional periodic loads on the blades, which result in increased
operating noise levels as well as fatigue, or even fatal damage of the compression unit.
Due to the severity of these hazardous conditions, compressors are conservatively
designed to operate well below the peak pressure rise point. A safety margin of 10 to 20
percent is generally introduced between the surge line and the design operating condition.
In the past several years, the implementation of appropriate stall detection and stall
avoidance devices has significantly reduced this margin. These measures insure the safe
operation at conditions of higher compressor pressure ratios and extend the useful
operating range to smaller flow rates. In general, all studies aimed at controlling
compressor operation fall into one of two categories: (1) passive or open-loop control, or
(2) active or closed-loop control. References 9, 16 and 17 contain surveys that illustrate
many of these studies.
Open-loop control is achieved by changes in the compressor design and
construction such that the performance characteristic map is modified and the surge line
12
is shifted to smaller flow rates. This idea hinges on the assumption that the compression
system tends to become unstable near the point of maximum pressure rise due to zero
slope of the compressor characteristic (C = 0). Open-loop control strategies are aimed at
moving the pressure peak to smaller mass flow rates, even at the expense of a slight
decrease in adiabatic efficiency. Casing treatments
18-21
, stationary guide vanes
22,23
, and
steady air injection
24
are representative examples of open-loop control strategies.
Casing treatments seek to enhance compressor stability by modifications in the
design of the rotor casing. These treatments consist of various shaped grooves or
perforations over the impeller tip. The intended purpose of casing treatments is to
decrease the level of blockage in a flow passage, thus suppressing the development of
rotating stall. Since casing treatments are aimed at alleviating the onset of rotating stall,
their use is limited to axial configurations.
Inlet guide vanes and steady air injection are open-loop control strategies, proven
effective for both axial and centrifugal compressors. The schemes are comparable in that
they both seek to decrease flow incidence angles in compression systems at reduced mass
flow rates. In the same way a wing at large angles of attack experiences leading edge
stall, increased blade incidence angles often cause leading edge flow separation and lead
to rotating stall or surge. Inlet guide vanes effectively turn the incoming flow and
alleviate the danger of leading edge separation. A drawback of inlet guide vanes is the
wake that forms aft of the vane trailing edge and impinges on the rotor leading edge in an
unsteady fashion. Depending on the vane-rotor spacing and the severity of the
phenomenon, this wake may lead to high-cycle fatigue of the rotor.
13
Air injectors mounted upstream of the compressor face have also proven
successful in decreasing incidence angles, thus suppressing leading edge separation. Air
injectors, however, rely on an external supply of high-pressure air that must be taken
from a downstream stage. It is, therefore, imperative to properly tailor injection schemes
to the individual compressor and minimize the injected amount of air needed to stabilize
the compressor.
Most open-loop control schemes are relatively easy to implement and give
immediate performance improvements. However, any scheme is limited in the extent to
which real operating range extensions are feasible, and should be tailored to the
individual compressor. A detailed knowledge of the fluid dynamic processes that lead to
the development of rotating stall and surge in each compressor is necessary for applying
open-loop control schemes properly.
Over the past five decades, active or closed-loop control strategies have been
extensively explored. An overview of closed-loop compressor control is schematically
shown in Figure 1.10. Fast-response-stall-detection devices, in most compressors
circumferentially distributed over the casing walls, measure flowfield data and send
filtered signals to a controller unit. When a compressor approaches a stall-like condition,
the presence of growing instabilities (stall precursor waves) is measured and signaled
back to the controller unit before the actual stall cycle begins. The control mechanism is
then activated by a direct link between the controller unit and a set of actuation devices.
A wide range of actuators can be found in the literature. The type of active
control scheme is often categorized by the choice of a particular actuator. Pinsley et al.
25
14
adopted a plenum gate valve to regulate the flow leaving the compression system. The
valve was operated at frequencies tailored to damp out all potential disturbances that
would lead to the onset of surge. Other examples of closed-loop compressor control
utilizing bleed valves are documented in References 26 and 27.
Gysling
28
achieved up to 25 percent surge margin extension by implementing a
moveable plenum wall. Arnulfi et al.
29
adopted the same strategy and extended the
scheme to a four-stage system. Other types of compressor stabilization schemes include
high-fidelity plenum loudspeakers
30
, moveable inlet guide vanes (IGV)
31
, tip clearance
control
32
, synthetic jets
33
, and air injectors upstream of the compressor face
8,25,34-44
. The
latter scheme produces the best results because most instabilities leading to the
development of rotating stall and surge originate near the impeller leading edge. Large
leading edge incidence angles and vortices that form near the blade tip are potential
sources of instabilities. Air injection, at times referred to as blowing, directly energizes
regions of low-momentum fluid and reduces the risk of leading edge separation. The
advantage of active air injection over a passive injection scheme is that the jets may be
turned on only as needed, thus saving useful high-pressure air. In addition, the jets may
be pulsed sinusoidally using measured stall-precursor frequencies to cancel out a desired
number of unstable modes.
Although closed-loop compressor control is an ongoing area of research with
advancements bringing significant improvements in compressor performance, several
challenges must be addressed before the realization of closed-loop compressor control in
real engines. These challenges are listed below:
15
1) Due to the small time scales that characterize the development of rotating stall and
surge, it is imperative for the designer to employ fast-acting components within the
closed control loop. This represents a need for high-performance sensors, controller
units and actuators. Many high-speed compressors used today in industrial
applications encounter saturation phenomena, i.e. a point where further actuation
leads to no improvement, if operated with the current generation active control
components. References 45 and 46 evaluate the effectiveness of different sensors and
actuators.
2) Mandatory in active compressor control is the availability of a simple model for the
onset of rotating stall and surge. However, the fluid dynamic phenomena leading to
compressor instabilities are complex and not yet well understood. Many models that
were developed from phenomenological considerations require prior knowledge of
the compressor performance map or the availability of flow properties that are
difficult to measure. Despite these challenges, several compressor control models
have been developed and tested in real engines. In particular, the Moore-Greitzer
model
47,48
and extensions of this model
49,50
have been extensively explored. More
experimental and computational studies that can validate and improve these models
are urgently needed.
3) Many active control schemes were developed and tested in experimental facilities
under highly controlled ambient conditions. It is not certain if these devices would be
successful in general situations. Some parametric studies
39
were carried out. More
research on physical mechanisms leading to active compressor control is necessary.
16
1.7 Analytical and Computational Modeling of Compressor Control
Most of the control schemes described in the previous sections fail to explain the
physical origin of rotating stall and surge. In experiments, control actuators are often
used in an ad hoc fashion where significant results are obtained, but the flowfield details
remain unknown. An important first step in controlling compressor instabilities is to
understand the underlying physical phenomena in the system and to develop adequate
mathematical models that describe the most significant phenomena.
Various models of different types have been proposed in the literature for the
description of rotating stall and surge. The categorization of such models is often based
on the applied flow description: one-dimensional lumped-parameter models can only
describe axisymmetric flow phenomena, whereas two-dimensional models can include
variations in the azimuthal direction. The type of mathematical approach used further
classifies the proposed models. Early models, such as that derived by Emmons et al.
4
use
a linear analysis to examine compression system instabilities. Since such approaches are
restricted to infinitesimal departures from an equilibrium situation, large amplitude
fluctuations, e.g. surge, may not be adequately described within such a framework.
In a pioneering effort, Moore and Greitzer
47,48
developed a non-linear,
phenomenological model for rotating stall and surge that is widely used in axial
compressors. Their model simulates the compression system with only three
components. The first component is the inlet duct that allows infinitesimally small
disturbances at the duct entrance to grow until they reach an appreciable magnitude at the
compressor face. The second component is the compressor itself, modeled as an actuator
17
disk, which raises the pressure ratio by performing work on the fluid. The third
component is the plenum chamber downstream of the diffuser, which acts as a large
reservoir and responds to fluctuations in mass flow with fluctuations in pressure behind
the actuator disk.
The Moore-Greitzer model gives rise to three ordinary differential equations: the
first for the amplitude of mass flow rate fluctuations, caused by rotating stall; the second
for the non-dimensional, circumferentially averaged mass flow rate through the
compressor; and the third for the non-dimensional total-to-static pressure rise across the
compression system. The most important approximations underlying the Moore-Greitzer
model are (1) it is valid under small perturbations, and (2) the time scale of the dynamics
governing the rotating stall amplitude is much faster than the time scale of the dynamics
governing the flow rate.
The Moore-Greitzer model does not attempt to explain what physical mechanism
triggers these instabilities. Rather, it attempts to determine the conditions under which
the disturbances will grow and what control measures can suppress the instabilities. It is
also used in surge control research based on the belief that rotating stall is a precursor to
surge, and therefore the elimination of rotating stall will also eliminate the development
of surge. This assumption has not found widespread acceptance among researchers. A
second drawback of the Moore-Greitzer model is its failure to work in centrifugal
compressors, since it was derived from two-dimensional theory. Hence, fluid dynamic
phenomena that lead to the development of surge and rotating stall in centrifugal
compressors may not be studied nor controlled using the model described above.
18
Only the flow visualization techniques and computational fluid dynamics (CFD)
modeling can clearly explain the physical origin of compressor instabilities. CFD
methods provide an efficient way to study complex flow phenomena
51
.
Figure 1.11 documents the rapid increase in computing power over the past five
decades, a development that now enables the designer to perform three-dimensional,
time-accurate CFD simulations of compression systems. With the advent of parallel
processing techniques and the establishment of high-performance computing centers,
CFD methods have proven to be powerful tools in modeling compressor aerodynamics.
A number of CFD codes for detailed modeling of turbomachinery flowfields
exist. Chima et al.
52
, Hall
53
, Dawes
54
, Hah et al.
55
, and Adamczyk et al.
56
, among others,
have developed three-dimensional codes that are capable of analyzing unsteady
turbomachinery flow with multiple blade passages and/or rotor-stator interaction. Most
of these applications, however, are limited to modeling the steady-state phenomena in
axial and centrifugal compressors or to modeling unsteady flow phenomena caused by
rotor-stator interactions. Casartelli et al.
57,58
performed a three-dimensional numerical
analysis of a centrifugal compressor at near-stall operating conditions and showed that a
re-design of diffuser guide vanes yielded an extension of the useful compressor operating
range. Many researchers have also simulated rotating stall and surge on axial
compressors using simple one-dimensional and two-dimensional codes
59-63
. Three-
dimensional simulations of stall and surge have not been extensively attempted due to the
computational resources required.
19
1.8 Goals of Compressor Control Research
The goals of the current research were to gain a fundamental understanding of
compressor instabilities, and to develop intelligent control systems for turbomachinery
components.
While much progress has been achieved in experimentally exploring and
developing compressor control schemes, most numerical efforts have been limited to
studying the steady-state behavior of compression systems. As a result, the present
generation of compressor designers still lacks understanding regarding the fundamental
mechanism that leads to the development of rotating stall and surge. Under what
circumstances and operating conditions will a certain centrifugal compressor exhibit stall-
like behavior? What type of compressor instability is to be expected in a certain system?
In which part of the compression system does stall first occur? What fluid dynamic
measures are necessary to suppress the onset of stall? The compressor industry is still
struggling to develop a comprehensive formula to answer these questions without
expensive test rig measurements of a particular compressor.
The current research attempts to provide some answers to the above questions. In
this work, CFD methods are employed to simulate centrifugal compressor flow at various
operating conditions and to closely examine the physical phenomena within the flowfield.
Once the eminent factors in stall inception are identified, CFD methods are utilized to test
appropriate stall control concepts that address the identified factors. Parametric CFD
studies are then carried out to find the optimum control setup.
20
The following chapters give a detailed description of the concepts, tools,
strategies and results of this research. This thesis is organized as follows:
1. Chapter II presents an overview of the CFD concepts that lead to effective centrifugal
compressor control. It outlines the specific objectives and the chosen approach.
2. Chapter III contains a description of GTTURBO3D, the three-dimensional Navier
Stokes solver.
3. Chapter IV presents validation results at compressor design operating conditions by
comparison against experimental data.
4. Chapter V discusses CFD results at off-design conditions that identify fluid dynamic
phenomena leading to the onset and development of instabilities in centrifugal
compressors.
5. Chapter VI shows simulations addressing the effectiveness of air injection as a means
to alleviate compressor instabilities. These simulations include steady air injection
and unsteady jets.
6. Chapter VII summarizes the principal conclusions drawn in this research and
provides recommendations on ways to build on the knowledge gained from this
research to further improve compressor control technology.
21
Figure 1.1 Volumetric Size Range for Different Categories of Compressors
Figure 1.2 Cutaway View of Allied Signal TPE331-14 Turboprop Gas Engine with Two
Centrifugal Compressor Stages
Rotary
Compressors
Single-stage
Centrifugal
Multi-stage
Centrifugal
Single-stage
Reciprocating
Multi-stage
Axial
Multi-stage
Reciprocating
Volumetric Flow Rate (ft
3
/min)
1E+4
200
2
1E+6 1E+5 1E+3 1E+2
20
P
r
e
s
s
u
r
e

R
a
t
i
o

22
Figure 1.3 Schematic of Compressor Performance Characteristic Map (source:
Hunziker
64
)
Figure 1.4 Centrifugal Impeller Damage Due to Compressor Instabilities (source: NASA)
Lines of Constant
Rotational Speed
Lines of Constant
Efficiency
Desired Extension of
Operating Range
T
o
t
a
l

P
r
e
s
s
u
r
e

R
i
s
e

Flow Rate
C
h
o
k
e

L
i
m
i
t
S
u
r
g
e

L
i
m
i
t
23
Figure 1.5 Impeller Damage Due to Compressor Instabilities (source: NASA)
L
o
c
a
l

S
e
p
a
r
a
t
i
o
n

L
o
c
a
l

S
e
p
a
r
a
t
i
o
n

1
2 3

Figure 1.6 Rotating Stall Pattern
24
Figure 1.7 Schematic of Mild and Deep Surge Cycle
Figure 1.8 Helmholtz-Resonator-Model (source: Greitzer
11
)
Throttle
A
c
Length L
c
Compressor
Plenum
V
p
Flow Reversal
Time
Flow
Rate
Period of Deep
Surge Cycle
Pressure
Rise
Flow
Rate
Mean
Operating
Point
Limit Cycle
Oscillations
a) Mild Surge b) Deep Surge
Pressure
Rise
Flow
Rate
M A
B C
Peak
Performance
Time
Flow
Rate
Period of
Mild Surge
Cycle
25
Figure 1.9 Schematic of Static and Dynamic Instability (source: Greitzer
11
)
Figure 1.10 Active and Passive Compressor Control Schemes
Static Instability
C > T
C
T
Pressure
Rise
Mass Flow
Dynamic Instability
C > 1/B
2
T
C
T
Pressure
Rise
Mass Flow
Controller Unit
Bleed
Air
Pressure
Sensors
Air
Injection
Bleed Valves
Movable Plenum Walls Steady Blowing
Guide Vanes
26
Figure 1.11 Development of Computational Resources
Flops
1950 1960 1970 1980 1990 2000
Relays Vacuum
Tubes
Transistors Integrated
Circuits
Microprocessors
ENIAC
UNIVAC
IBM 704 CDC 1604
LARC
Stretch
CDC 6600
CDC 7600
ILLIAC IV
Cray-1
Cray X-MP
Cray Y-MP
Cray-2
CM-2
Delta
Intel Paragon
Cray C-90
Teraflop
Scalar
Vector
Multiprocessors
Massively Parallel
Mark I
10
1
10
2
10
3
10
4
10
5
10
6
10
7
10
8
10
9
10
10
10
11
10
12
1
0.1
27
CHAPTER II
2 OVERVIEW OF THE PRESENT WORK
This chapter gives an overview of the concepts studied in this research.
Compressor control is a multi-faceted field, therefore present centrifugal compressor
control studies focus on only three major aspects of flow control. These aspects are as
follows:
1. Investigation of the system stability
What happens when the compressor performance map peaks at the point of the
maximum pressure rise? This operating condition is of great interest to the designer,
but it usually lies dangerously close to the surge line. Even small perturbations can
cause a system that operates at the maximum pressure point to become dynamically
unstable. As outlined in Section 1.5, the criterion for dynamic stability depends on
the value of the B-parameter. Compression systems with large B-parameters
experience a greater tendency towards dynamic instability near the maximum
pressure point than systems with small B-parameters. This B-parameter theory was
investigated in this research using CFD methods.
2. Application of steady air injection as a means to control centrifugal compressors
As outlined in Chapter I, air injection is widely accepted as a viable compressor
control scheme. Based on a series of computational results, the air injection model is
examined in this work. A parametric study is then employed to develop criteria that
28
allow a prediction of the optimum steady injection setup and the minimum steady
flow rate needed to suppress instabilities in a given centrifugal compressor. CFD
simulations with low-speed and high-speed centrifugal compressors serve as
validation tools for these injection criteria.
3. Study of unsteady air injection as a stall avoidance scheme
As discussed in Chapter I, studies have shown that compressor instabilities originate
near the impeller blade leading edges. These instabilities consist of two frequencies:
a low frequency related to the compressor-plenum interactions (onset of limit cycle
instabilities), and a high frequency characteristic of the leading edge vortex shedding
that occurs at reduced flow rates. CFD simulations are performed to determine these
frequencies. Unsteady leading edge air injection is then utilized to counter the
observed flow phenomena and estimate the effectiveness of pulsed jet control for
low-speed and high-speed centrifugal compressors.
2.1 Major Steps in the Present Research
The three aspects presented in the previous section are the research topics chosen
for detailed study in the current work. Figure 2.1 schematically shows some of the main
steps involved in this study. The flow chart is divided into two main trees, one for each
test configuration considered in this research. The tree on the left represents a NASA
Lewis low-speed centrifugal compressor while the right side deals with a high-speed
impeller built at Deutsche Luft- und Raumfahrt (DLR) in Cologne, Germany. The
specific steps used to accomplish the goals can be outlined as follows:
29
1) Grid generation was performed for each test configuration. In the case of the high-
speed compressor, this was accomplished using the commercial software Gridgen
65
.
Dr. Michael Hathaway (NASA Glenn) supplied the computational mesh for the low-
speed compressor.
2) The computational grids were used to simulate compressor operation at the design
point for which experimental measurements were available. In the case of the low-
speed compressor, the flow solver was additionally validated against a limited
number of off-design data. The purpose of this validation step was to ensure the
stability of the flow solver, the accuracy of the computational grids, and the well-
posedness of boundary conditions and input data.
3) The viscous flow solver was used to explore the flowfield dynamics as the
compressor flow rate was reduced and the surge line was approached. The main
phenomena that led to the onset of stall were identified, and the behavior of the two
different compressors was compared. The flowfield during the bifurcation process
was investigated.
4) Based on the conclusions from step 3 and from a literature survey of several air
injection control studies by other researchers, the critical parameters were identified
and compressor stall avoidance criteria with steady air injection were developed. The
validity of the findings was tested on the two test configurations and optimized.
5) Using the optimized steady air injection setup from step 4, pulsed air injection at
varying injection frequencies and injection rates was evaluated.
30
These steps produced noteworthy results. However, the parameters utilized
limited the scope of the research. A summary of significant contributions as well as the
limitations of this work is presented in the next section.
2.2 Potential and Limitations
The current research used an unsteady three-dimensional flow solver that solves
the flow governing equations in conservation form from first principles. This ensured
that all flow phenomena were modeled within the accuracy of the employed computer,
flow solver, computational grid and boundary conditions.
Due to limitations in computational resources, single flow passage grids were
used with appropriate periodic boundary conditions. This approach assumes that the
stable operation of centrifugal compressors is limited by instabilities that develop within
a single flow passage. Instabilities that form and grow across several compressor flow
passages in the circumferential direction have been neglected within the framework of
this analysis.
The instability frequencies examined in this study were between 70 and 90 Hz for
the plenum dimensions chosen. Surge tends to occur at lower frequencies
11
although
Fink et al.
15
measured mild surge frequencies of 99 Hz in a centrifugal compressor with a
small plenum volume. This system, however, had a small B-parameter and may not be
comparable to the high-B compression systems studied in the present work. Thus, the
instabilities observed in this work may be a combination of rotating stall and surge. Due
to the limiting assumption of blade-to-blade periodicity, the type of fluid dynamic
31
instability that occurs in the present compressors could not be established. For brevity
sake, the terms surge and stall are used interchangeably in the rest of this work,
although it is understood that surge and rotating stall are two different phenomena.
Separating these two effects will require modeling flow through several flow passages, a
formidable task beyond the scope of this study.
Other limiting assumptions are:
1. The fluid flow within the compressor plenum was modeled in a phenomenological
manner and coupled with the flow solver through interface boundary conditions.
2. No reversed flow across the compressor-plenum interface boundary was permitted.
As a result, a complete deep surge cycle with flow reversal throughout the whole
compression system could not be simulated. Since deep surge cycles are always
preceded by large limit cycle oscillations, the control scheme focused on the detection
and avoidance of fluid dynamic phenomena that lead to these limit cycle oscillations.
3. Turbulent flow phenomena were modeled with a one-equation Spalart-Allmaras
turbulence model
66
. Since no information regarding transition locations in the two
compressors considered was available, the flow was assumed to be fully turbulent
throughout the entire computational domain. Experiments
67
with real compressors
have shown that significant portions of the impeller boundary layers remain laminar
at various operating conditions. Hence, the fully turbulent assumption led to an over-
prediction of dissipation and diffusion processes as well as viscous losses, thus
underpredicting the total-to-total pressure rise across the compressor.
32
Air Injection Control Scheme
- Derivation of Criterion for Steady Blowing
Identification of Parameters that
Lead to Instabilities
- at Design Speed
- at Off-Design Speed
Code Validation vs. Experimental Data
and Grid Sensitivity Study
- Design Operation
- Off-Design Operation
Grid Generation
(NASA Glenn)
322,629 Grid Points
NASA Low-Speed
Centrifugal Compressor
Air Injection Control Scheme
- Derivation of Criterion for Steady Blowing
- Demonstration of Unsteady Blowing
Identification of Parameters that
Lead to Instabilities
- at Design Speed
Code Validation vs. Experimental Data
and Grid Sensitivity Study
- Design Operation
Grid Generation
(Gridgen)
230,000 Grid Points
DLR High-Speed
Centrifugal Compressor
Centrifugal Compressor CFD Configurations
Figure 2.1 Centrifugal Compressor Configurations Used in This Work
33
CHAPTER III
3 NUMERICAL ANALYSIS TOOLS
This chapter is devoted to a description of the computational tool incorporated in
studying and simulating three-dimensional compressor flow, namely the viscous flow
solver. John D. Anderson, Jr., one of the pioneers in fluid dynamics, defines CFD
methods as follows:
Computational fluid dynamics is the art of replacing the governing partial
differential equations of fluid flow with numbers, and advancing these numbers in
space and/or time to obtain a final numerical description of the complete flow-field
of interest. The end product of computational fluid dynamics is indeed a collection
of numbers, in contrast to a closed-form analytical solution
68
.
The following sections highlight the computational methods used to replace the
governing equations with numerical approximations in the present research. The
presentation of detailed derivations of the employed solution schemes can be found in
other publications. For a more comprehensive treatment of viscous CFD methods, the
works of Anderson et al.
69
and Hirsch
70
serve as excellent references.
In Section 3.1, the governing equations for unsteady compressible fluid motion
are presented in Cartesian coordinates. Section 3.2 outlines the discretization methods,
and Section 3.3 illustrates the approximate factorization solution algorithm used in the
flow solver. Sections 3.4 and 3.5 conclude the mathematical description of the viscous
flow solver with a review of the turbulence model, the initial conditions and boundary
conditions. In Section 3.6, the code structure and implementation are discussed.
34
3.1 Three-Dimensional Governing Equations
The set of five coupled partial differential equations for the conservation of mass,
momentum and energy in fluid flows is known as the Navier-Stokes equations. These
equations describe continuum fluid flows from first principles. A good derivation of the
Navier-Stokes equations is given in the classic work of Schlichting
71
. Schlichting also
points out that these equations can be presented in both differential and integral forms.
Some terms of the full Navier-Stokes equations may be simplified or ignored if certain
assumptions are made. In this work, the fluid was treated as a calorically perfect gas with
a constant ratio of specific heats, . Under the additional assumptions of fluid
compressibility, no body forces and external heat addition, the set of three-dimensional
equations may be cast into an inertial Cartesian coordinate system
T S R G F E q
z y x z y x t
r r r r r r
r
+ + + + + (3.1)
where
( ) ( ) ( )
1
1
1
1
1
1
]
1
+
+
1
1
1
1
1
1
]
1
1
1
1
1
1
1
]
1
1
1
1
1
1
1
]
1
w p e
p w
wv
wu
w
G ,
v p e
vw
p v
vu
v
F ,
u p e
uw
uv
p u
u
E ,
e
w
v
u
q
2
2
2
r r r
r
(3.2)
35
1
1
1
1
1
1
]
1
+ +
1
1
1
1
1
1
]
1
+ +
1
1
1
1
1
1
]
1
+ +
z zz zy zx
zz
zy
zx
y yz yy yx
yz
yy
yx
x xz xy xx
xz
xy
xx
q w v u
0
T
q w v u
0
S ,
q w v u
0
R
r
r r
(3.3)
The first row of the vector Equations (3.1), (3.2) and (3.3) corresponds to the continuity
equation; rows two, three and four are the momentum equations; and the fifth row
corresponds to the conservation of energy equation. The vector q
r
is the state vector with
the unknown variables: density, , Cartesian velocities, u, v, w, and total energy, e.
G and F , E
r r r
are the inviscid flux vectors, and T and S , R
r r r
are the viscous flux vectors. In
addition to the unknown variables of the state vector, the inviscid flux vectors also
contain the unknown static pressure, p. The viscous fluxes contain the components of the
viscous stress tensor,
ij
, and the heat flux vector, q
i
. These quantities may be written in a
compact form using tensor notation
ij
k
k
i
j
j
i
ij
x
u
3
2
x
u
x
u
,
_
, i, j, k = 1, 2, 3 (3.4)
i
i
x
T
k q
, i = 1, 2, 3 (3.5)
36
In these equations, is the molecular viscosity, k is the thermal conductivity of the fluid
and
ij
represents the Kronecker-Delta. All quantities in the Navier-Stokes equations
were scaled by appropriate reference values to yield numerical values verging on unity.
Table 3.1 summarizes the reference quantities used in the flow solver. In addition to
Equations (3.1) through (3.5), the equation of state for perfect gases was used to close the
system
( )
1
]
1
+ + ) w v u (
2
1
e 1 p
2 2 2
(3.6)
where is the ratio of specific heats. This quantity was given a value of 1.4 assuming
calorically perfect gas. The relationships for the internal energy, e, and the static
enthalpy, h, are
e = c
V
T and h = c
p
T (3.7)
where T is the static temperature, c
V
is the specific heat at constant volume and c
P
is the
specific heat at constant pressure, respectively.
37
Table 3.1 Non-Dimensionalization Factors in the Flow Solver
Variable Explanation Comment
x
ref
D
trailing edge
Trailing edge tip diameter
V
ref
a
J Free-stream speed of sound
*
ref

J Free-stream density
*
p
ref
p
J =
J
a
J
2
/
Free-stream pressure
*
*
Free-stream quantities denote flow conditions upstream of the compressor inlet (e.g. reservoir properties)
3.2 Discretization of Navier-Stokes Equations
The major challenges in solving the system of governing Equations (3.1) through
(3.7) are their non-linear nature and the complexity of shapes in the geometries. These
challenges make analytical solutions in closed form impossible for turbomachinery
components. The most common strategy for solving viscous flow is breaking down the
computational domain surrounding the geometry into small cells. Integration of the
Navier-Stokes equations across each computational cell solves for the flowfield at time
level t
n
. The purpose of the numerical procedure is to advance the solution to a new time
level, t
n+1
, using a discrete time step, t. Integration of the Navier-Stokes Equations (3.1)
gives
[ ] [ ] dS n T S R dS n G F E dV q
t
S S V
r
r r r
r
r r r
r
+ + + + +

(3.8)
38
where n
r
is a unit normal vector pointing outward from the surface, S, bounding the cell
volume, V. The advantage of the integral form is that it remains valid in the presence of
discontinuities such as shock waves and contact surfaces. Temporal changes in the
conserved quantities mass, momentum and energy, modeled by the unsteady term in
Equation (3.8), are directly proportional to the convective fluxes across the six
boundaries of the computational cell.
Figure 3.1 shows the discretization of the cell-centered finite volume scheme.
The state vector, q
r
, was evaluated at the cell centers, whereas the surface integrals were
written as the sum of contributions from the six faces of the hexagonal cell
[ ] ( ) [ ] ( ) [ ]
( ) [ ] ( ) [ ]
( ) [ ] ( ) [ ]
2
1
2
1
2
1
2
1
2
1
2
1
k k
j j
i i
S
S n G S n G
S n F S n F
S n E S n E dS n G F E
+
+
+
+
+ +
+ + + +
r
r
r
r
r
r
r
r
r
r
r
r
r
r r r
(3.9)
These may be viewed as finite differences analogs. For example,
( )
( ) [ ] ( ) [ ]
2
1
2
1
2
1
2
1 i i
i i
E
S n E S n E
E
S n E
+
+
+

r
r
r
r
r
r
(3.10)
Note that the minus sign appears on the right-hand side because the unit normal vectors
on faces i + and i - point in opposite directions.
39
A four-point stencil, shown in Figure 3.1, was used to compute the inviscid fluxes
to the left and right of each cell face, a technique called flux difference splitting. The
following section outlines the flux difference splitting approach attributed to Roe that was
implemented in the flow solver.
Discretization of the viscous fluxes was performed by using central difference
formulations.
3.2.1 Roes Flux Difference Splitting
Roes flux splitting scheme is extensively documented in References 72 through
76. The method is based on a characteristic decomposition of the flux differences while
ensuring the conservation properties of the scheme. Knowing the initial uniform left and
right states, the interface flux may be calculated by solving for the difference in the flux
across the individual waves obtained from an underlying set of Rieman problems. The
interface fluxes were computed by treating each direction in a locally one-dimensional
manner. As such, the numerical flux,
2
1
i
E
+
, may be written as
( ) ( ) { } { }
2
1
2
1 2
1
i
L R
i
L L R R i
q q A
~
q E
q E
5 . 0 E
+ +
+
+
r r r r
(3.11)
The first term in this equation is the physical flux contribution while the second term
controls the amount of artificial viscosity that is implicitly added to damp out high-
frequency spatial oscillations. The vector of primitive variables to the right,
40
[ ]
T
R R R R R R
p , w , v , u , q
r
, and to the left, [ ]
T
L L L L L L
p , w , v , u , q
r
, of the
interface i + is written as
( ) ( )
( ) ( )
1 i i i i 1 i i i L
i 1 i i 1 i 2 i i 1 i R
q q
6
1
q q
3
1
q q
q q
3
1
q q
6
1
q q
2
1
2
1
2
1
2
3
+
+
+
+
+
+ + +
+ +
+ +

r r r r r r
r r r r r r
(3.12)
using third order upwind biased interpolation. At cell interfaces near boundaries, this
interpolation scheme was relaxed to first order accuracy. The coefficient in Equation
(3.12) serves to limit the high-frequency numerical oscillations near shock waves and is
patterned after Roes Superbee limiter
77
.
( ) ( )
( ) )] 2 , r ( min ), 1 , r 2 ( min , 0 [ max r
q q
q q
r ,
q q
q q
r
r , r
1 i i
i 1 i
i
i 1 i
1 i i
i
i i i i
2
1
2
1
2
1
2
1
2
1
2
1

+ +
+

+
+
+
(3.13)
Since the compressor flowfields studied in this work contain only weak shocks, or are
void of shocks, the limiter formulation summarized in Equation (3.13) was not used, and
the coefficients
were set to unity.

According to the Roe decomposition in Equation (3.11), the two physical
contributions to the flux may be expressed as
41
( ) ( )
1
1
1
1
1
1
]
1

+
+
+
1
1
1
1
1
1
]
1

+
+
+
t L oL L L
z L L L L
y L L L L
x L L L L
L L
L L
t R oR R R
z R R R R
y R R R R
x R R R R
R R
R R
n p h U
n p w U
n p v U
n p u U
U
q E
,
n p h U
n p w U
n p v U
n p u U
U
q E
r r
(3.14)
where the contravariant velocity, U, and the stagnation enthalpy, h
o
, were defined as
( )

p e
h
S n V V U
o
Grid
r
r r
(3.15)
and
( ) S n V n
Grid t

r
r
(3.16)
where V
r
is the fluid velocity at the surface (i + , j, k). The grid velocity,
Grid
V
r
, and the
unit normal vector, n
r
, were calculated at the grid interface (i + , j, k). The second term
in Equation (3.11) was expanded by Liu et al.
78
to give
{ }
n 2
*
1 1
i
L R
N
~
U
~
q
~
q q q A
~
2
1
+ +
+
r r r r
(3.17)
where
1
1
1
1
1
1
]
1
1
1
1
1
1
1
]
1
U
~
n
n
n
0
N
~
,
h
~
~
w
~ ~
v
~ ~
u
~ ~
~
U
~
z
y
x
n
o
*
(3.18)
42
a
~
p
2
~ ~
U
~
2
~ ~
~
a
~
U
2
~ ~
a
~ ~
p
2
~ ~
~
3 2 3 2
1 2
3 2
2
3 2
1 1
,
_
+
+
,
_
+
+
(3.19)
In this formulation, the
i
~
are the characteristic velocities
a
~
U
~ ~
, a
~
U
~ ~
, U
~ ~
3 2 1
+ (3.20)
and variables that have a ~denote that these quantities must be Roe-averaged according
to
quantities other all for
1 1
1 ~
densities for
~
L R
L R
R
L R
L
L R
,
_
+

+
,
_
+

(3.21)
In summary, Equations (3.11) through (3.21) were solved to find the flux,
2
1
i
E
+
, at a
typical cell interface (i + , j, k). The fluxes entering and leaving the other five cell faces
(i - , j, k), (i, j , k) and (i, j, k ) were similarly handled.
3.3 Linearization and Approximate Factorization
The combination of Equations (3.8) and (3.9) represents a non-linear algebraic
vector equation for the unknown flow variables. Several mathematical procedures exist
in the literature for the solution of such systems: Newtons method, Gauss-Seidel
43
iterations, successive relaxation methods and alternate direction implicit (ADI) schemes
are several examples. The flow solver utilizes a diagonal form of an implicit three-factor
approximate factorization procedure that is subsequently described. For more detailed
references, the works of Pulliam et al.
79
and Beam et al.
80
are excellent sources.
First order temporal discretization was applied to the unsteady term in Equation
(3.8)
1 n
k
1 n
k
1 n
j
1 n
j
1 n
i
1 n
i
n
2
1
2
1
2
1
2
1
2
1
2
1
G
t
q
+
+
+
+
+
+
+
+
+
+ +
(3.22)
where
( ) V q q q
n 1 n n

+
(3.23)
In this expression, V is the volume of the computational cell. To simplify these
equations, the inviscid flux terms,
1 n 1 n 1 n
G
and F
, E
+ + +
, were linearized about the previous
time level, t
n
n n n 1 n
n n n 1 n
n n n 1 n
q C
q B
q A
+
+
+
+
+
+
(3.24)
where
n n n
C
and B
, A
are the approximate Jacobian matrices according to

n
n
n
n
n
n
q
G
,
q
F
,
q
E
1
]
1
1
]
1
1
]
1
(3.25)
44
A detailed description of the process of obtaining analytical expressions for the flux
Jacobians is included in Reference 60. Substitution of (3.24) into (3.8), and some
rearranging yields
n n n
tR q M (3.26)
The expressions for the matrices M
n
and R
n
are
T
n n n n n n n
2
1
k
2
1
k
2
1
j
2
1
j
2
1
i
2
1
i
G
R
1
]
1
+ +
+ + +
(3.27)
where I is the (5x5) identity matrix. The terms in the expressions for M
n
and R
n
represent
numerical fluxes crossing each of the six cell faces. Therefore, the individual elements of
M
n
are (5x5) matrices. For large systems, the direct inversion of Equation (3.26) is
computationally expensive. The basic premise behind ADI methods is to separate the
matrix on the left-hand side into an approximate product of three smaller matrices as
shown
1
1
1
1
1
1
1
1
1
1
1
1
1
]
1
+ + +
n n n n n n
n
2
1
k
2
1
j
2
1
i
2
1
i
2
1
j
2
1
k
C
t B
t A
t
I
A
t B
t C
t
M
45
n
3
n
2
n
1
n
M M M M (3.28)
where, for example
(3.29)
Similarly,
n
3
n
2
M and M contain the Jacobian matrices
n n
C
and B
, respectively. Equations
(3.28) and (3.29) were substituted into Equation (3.26), and the system was solved in
three sweeps to get an expression for the change in the flow variables, q
* * n
3
* * * n
2
n * n
1
q q M : 3 Sweep
q q M : 2 Sweep
tR q M : 1 Sweep

(3.30)
If further reduction in CPU time is desired, one can approximate
n
2
1
i
A
+
and
n
2
1
i
A
by
simply
n
i
A
. This matrix has a set of eigenvalues with distinct eigenvectors, a property

that was used to apply a similarity transformation and diagonalize the matrix according to
( )
i
1 n
i
T
T A

(3.31)
1
1
1
1
1
1
1
1
1
]
1

+
n n n
1
2
1
i
2
1
i
A
t
I
A
t M
46
This similarity transformation is described by Warming et al.
81
and Turkel
82
who also
give analytical expressions for T, T
-1
and
is a diagonal matrix with the

characteristic eigenvalues as diagonal elements and T, T
-1
represent (5x5) matrices with
the corresponding eigenvectors as columns. Substitution of Equation (3.31) into
Equation (3.30) yields
( ) ( )
n *
1 i i
1 *
i
*
1 i i
1
tR q T
T q I q T
T + +
+
(3.32)
for the first sweep. After pre-multiplying by T
-1
and rearranging, Equation (3.32) may be
written
( ) ( ) ( )
n 1 *
1 i
1
i
*
i
1 *
1 i
1
i
R tT q T
q T I q T
+ + (3.33)
This equation represents a tridiagonal system for the unknown
* 1
q T
; it was efficiently
solved using a scalar Thomas algorithm
69
. Similar procedures were utilized during
Sweeps 2 and 3 to avoid block tridiagonal matrix inversions.
3.4 Turbulence Modeling
Many turbomachinery flows give rise to boundary layers and wakes that exhibit
turbulent flow structures. With the advent of high-speed compressors, average Reynolds
numbers have increased, thus effectively amplifying turbulent and transitional flow
phenomena. These flow phenomena can be directly modeled and captured by Navier-
Stokes solvers only if extremely fine grids are used, a strategy called Direct Numerical
47
Simulation (DNS) that is physically valid but leads to unreasonably large computational
expenses. To fully capture the turbulent dynamics of only 1 cm
3
of the flowfield, 10
5
grid points would be required. For this reason, the present work utilized the time-
averaged form of the Navier-Stokes equations called the Reynolds-averaged Navier-
Stokes (RANS) equations. Time-averaging the equations gives rise to new terms that can
be interpreted as apparent stress gradients, similar to the components of the viscous stress
tensor on the right-hand side of Equation (3.8). These new terms, often referred to as the
components of the Reynolds stress tensor, are additional unknowns and must be
appropriately modeled to close the system of governing equations. A wide range of
turbulence closure models exists in CFD related publications, most of which differ in the
type and number of additional equations to be solved.
The turbulence closure model used in the flow solver was patterned after the
Boussinesq assumption
71
,
_

i
j
j
i
t j i
x
u
x
u
u u (3.34)
The left-hand side of this equation contains the unknown Reynolds stress tensor, while
t
is the turbulent viscosity or eddy viscosity. In this form, the effect of turbulent
phenomena on the flowfield was evaluated by computing a net viscosity consisting of a
laminar viscosity, , and a turbulent viscosity,
t
. The laminar viscosity is a fluid
property, while the turbulent viscosity is a flow property that is determined by solving
additional equations.
48
In this work, a partial differential transport equation for the turbulent viscosity
was solved to update
t
at each time step. This approach, known as the Spalart-Allmaras
turbulence model
66
, has found widespread popularity among CFD researchers. In the
Spalart-Allmaras model, the turbulent viscosity is given by

~
,
c
1 f , f
~
3
1 v
3
3
1 v 1 v t
(3.35)
where is the molecular viscosity. The working variable,
~
, obeys the transport
equation
( ) ( ) ( ) ( )
3 2 1
4 4 4 3 4 4 4 2 1
4 4 4 4 4 3 4 4 4 4 4 2 1
4 43 4 42 1
ition Trip/Trans
2
1 t
n/Wall Destructio
2
2 t
2
1 b
w 1 w
Diffusion
2
2 b
Production
2 t 1 b
U f
d
~
f
c
f c
]
~
c
~ ~
[
1
~
S
~
f 1 c
Dt
~
D
+
1
]
1

1
]
1

+ +
(3.36)
Here,
1 v
2 v 2 v
2 2
f 1
1 f , f
d
~
S S
~
+
+ (3.37)
where S is the vorticity magnitude, and d is the distance to the closest wall. The non-
dimensional function f
w
is
( )
2 2
6
2 w
6 / 1
6
3 w
6
6
3 w
w
d S
~
~
r , r r c r g ,
c g
c 1
g f
+
1
1
]
1
+
+
(3.38)
49
To avoid computer overflow for large r, r was truncated to 10. The functions used in the
destruction and the transition term, respectively, were
( )
( )
2
4 t 3 t 2 t
2
t
2
t
2
2
2
t
2 t t 1 t 1 t
c exp c f
d g d
U
w
c exp g c f

,
_

(3.39)
where
,
_
t
t
t
x
w
U
, 1 . 0 min g (3.40)
U is the difference between the velocity at the field point and the trip, and x
t
is the grid
spacing along the wall at the trip.
The Spalart-Allmaras model was originally developed to solve viscous flows
around isolated wings. References 66 and 83 contain a number of test cases that
demonstrate the models improved capability for predicting reversed flow regions. The
constants shown in Table 3.2 were tailored to fine-tune such flow phenomena.
The flow solver solves the transport Equation (3.36) by an implicit three-sweep
ADI method, similar to the procedure outlined in Section 3.3. The terms on the right-
hand side were evaluated explicitly, while the convective forces on the left-hand side
were factored in the -, -, and -directions.
50
Table 3.2 Constants in Spalart-Allmaras Turbulence Model
c
b1
= 0.135
+
+
2 b
2
1 b
1 w
c 1 c
c c
t1
= 1
= 2/3 c
w2
= 0.3 c
t2
= 2
c
b2
= 0.622 c
w3
= 2 c
t3
= 1.2
= 0.41 c
v1
= 7.1 c
t4
= 0.5
3.5 Initial and Boundary Conditions
Due to the parabolic (in time) and elliptic (in space) nature of the three-
dimensional Navier-Stokes equations, a set of initial and boundary conditions are
imposed on the flow. While the choice of initial conditions is somewhat arbitrary and
serves the sole purpose of starting the iterative solution process, the proper choice of
boundary conditions is critical. Although the compressor boundary condition treatment is
not the central focus of this research, it is nevertheless an important ingredient for an
accurate simulation of compressor instabilities. The following sections describe the
implementation of initial and boundary conditions in the viscous flow solver.
3.5.1 Initial Conditions
The flow solver has two options for the specification of initial conditions. If
nothing is known about the compressor flowfield, the user can choose to initialize all
51
flow variables to the given free stream values. This approach is referred to as compressor
cold start. It is similar to the case of a real compressor that undergoes some transients
when first started until a steady-state operation is reached. Cold start initialization does
not ensure fast convergence rates.
As a second choice of initial conditions, a previous flowfield solution may be read
into the code, a strategy often called restart. Restart is the preferred type of initialization
if a previous solution, even at a different operating condition, is available.
3.5.2 Boundary Conditions
Numerical values for the flowfield variables were specified on all six faces of the
computational domain. Figure 3.2 shows a centrifugal compressor single flow passage
with the different boundary conditions highlighted. Additional information regarding
each boundary condition is presented in this section.
3.5.2.1 Solid Wall Boundary Conditions
For viscous flows, the non-slip condition dictates that the relative flow velocities
of surface grid points with respect to the solid wall be equal to zero. Thus, in the case of
a rotating solid wall, the inertial flow velocities were set equal to the surface velocities
Solid
V V
r r
(3.41)
This boundary condition was applied at the impeller blades and the hub. Since the
vaneless diffuser hub is stationary in most centrifugal compressors, the absolute flow
52
velocities were set to zero along this wall and along the entire compressor casing wall. In
some instances, e.g. spooled compressors, part of the compressor shaft may be stationary.
An equation for the surface pressure on -constant solid walls can be obtained
from a normal momentum relation. Pulliam et al.
79
outline the details of this derivation
and give the resulting expression for the surface pressure
( ) ( ) ( )
,
_

+

+

+

+ + + + + + + +
+ +

z
y
x t
2
z
2
y
2
x z z y y x x z z y y x x
2
z
2
y
2
x
n
w v u
p p p
p
(3.42)
This equation was implicitly solved by decoupling the -, - and -derivatives and
independently sweeping in each of the three directions, similar to the approximate
factorization scheme outlined in Section 3.3. The pressure equations for -constant and
-constant walls were constructed following this scheme. Equation (3.42) reduces to
0
n
p
(3.43)
on an orthogonal grid. Since computational meshes for turbomachinery applications
generally possess non-orthogonal cells around the impeller surfaces due to blade twisting,
backsweep and finite edge radii, Equation (3.42) was only used in the -direction. The
surface pressures along hub and casing walls (-direction) were extrapolated from the
interior using Equation (3.43).
53
Finally, the flow solver computes the surface density from one-sided
extrapolations. This is equivalent to setting /n 0.
3.5.2.2 Inflow Boundary Conditions
At this boundary, the stagnation temperature, T
0
, and the total pressure, p
0
, were
specified from ambient or reservoir conditions. In a real compressor, stagnation
properties can be measured or approximated from isentropic relations. The tangential
velocity components at this boundary were set to zero assuming no swirl inflow, although
it was possible within the scope of this study to prescribe swirl or non-uniform stagnation
properties at the inlet. This would, for example, be the case for an analysis that
investigates the adverse effect of inlet distortions on the onset and development of
compressor instabilities.
At every time step, the speed of sound, a, and the normal velocity component, u
n
,
were determined from a one-dimensional wave equation to account for acoustic and
numerical waves leaving the computational domain
0 u
1
a 2
n
) a u ( u
1
a 2
t
n n

,
_
,
_
(3.44)
which was approximately solved as
2 i
n
1 i
n
u
1
a 2
u
1
a 2

(3.45)
54
with the auxiliary constraint
0 p
2
n
2
T c
2
u
1
a
+

(3.46)
After computing the speed of sound, a, the inflow velocity, u
n
, and the temperature, T,
from Equations (3.44) through (3.46), isentropic relations were used to evaluate the inlet
density, , and inlet static pressure, p.
3.5.2.3 Outflow Boundary Conditions
In this study, the foundational assumption was made that the flow from the
vaneless diffuser exhausts into a plenum chamber or large reservoir. As described in
Section 1.4, the coupling between compressor and plenum is essential for the
development of instabilities and cannot be neglected. Conversely, limitations in
computational resources often do not allow modeling of flow inside the plenum. The
approach implemented in the flow solver was a pure outflow boundary condition with
respect to the computational domain but took into account flow phenomena within the
plenum. Figure 3.3 shows a schematic of this strategy with the following assumptions in
modeling the plenum:
1. Negligible fluid velocity and acceleration
2. Spatially uniform plenum pressure
3. Isentropy
4. Constant volume, V
p
, and speed of sound, a
p
55
Conservation of mass in the plenum demands that
( )
t c p p
m m V
t
& &
(3.47)
where
p
is the spatially uniform plenum density,
c
m& is the mass flow rate at the diffuser
plenum interface, and
t
m& is the mass flow rate through the plenum throttle. Since the
plenum volume is not a function of time, Equation (3.47) may be re-written using the
chain rule
t c
p
p
p
p
m m
t
p
p
V & &

(3.48)
Considering the isentropic expression for the speed of sound in the plenum, a
p
,
p
p
2
p
p
a

(3.49)
Substituting Equation (3.49) into Equation (3.48) yields
( )
t c
p
2
p p
m m
V
a
t
p
& &
(3.50)
where p
p
is the pressure at the diffuser plenum. First order temporal discretization of
Equation (3.50) gives
( ) t m m C p p
t c
n
p
1 n
p
+
+
& & (3.51)
56
Here, the constant C was substituted for a
p
2
/V
p
. Numerical values between 0.05 and 0.2
were used for C within the framework of this study. Updated values for the
backpressure, p
p
n+1
, were calculated at each time step using the known properties p
p
n
, the
mass flow rate at the outflow boundary,
c
m& , and the desired mass flow rate through the
plenum throttle,
t
m& .
All other flow variables, namely the density and the three flow velocity
components, were extrapolated from the interior domain at the outflow boundary.
3.5.2.4 Periodic and Zonal Boundary Conditions
Figure 3.4 illustrates the approach for blade-to-blade periodic and zonal
boundaries in the compressor single flow passage. Both types of boundaries were similar
in the sense that the flow properties were averaged on either side of the boundary to
compute the interface value. The only difference was that density, momentum and
energy were averaged directly along zonal interfaces while along periodic boundaries
direct averaging only held for density, energy and axial velocity. The remaining two
velocity components in y- and z-direction were converted into radial and azimuthal
components on either side of the interface before they were averaged. Once the interface
values were computed, they were converted back to Cartesian velocity components.
3.5.2.5 Injection Boundary Conditions
In this study, both steady and unsteady air injection upstream of the compressor
face were tested as open-loop compressor control schemes. This type of injection is not
57
efficient for use in a real compressor system because of the continuous high-pressure air
supply needed to operate the injection system. It would be more effective to incorporate
the injection actuators in a closed-loop with appropriate stall sensors and a controller unit
to activate the injectors only if the compression system experiences stall onset. Air
injectors can be designed in many ways. A number of injection valves may be placed
upstream of the compressor face on the circumference of the compressor casing. The
optimum configuration of injector characteristics, e.g. injected mass flow rate, injection
angle, yaw angle, varies from system to system.
Figure 3.5 shows a schematic of the injectors used in this study. It was assumed
that a ring of jets continually injects fluid into the tip region of the rotor, a method that
requires a significant amount of external high-pressure fluid. Experiments
84
have shown
that by using only four sheet injectors, the same effect of stabilizing the compressor may
be achieved as with a greater number of injectors or with the ring injector employed in
this study. Thus, the requirements for effective stall control may be much smaller than
outlined in this study, a fact that could make air injection control even more desirable to
engine designers.
The numerical implementation of the injection boundary condition into the flow
solver required user-specified inputs for the injection location, injection angle and
injection mass flow rate. These inputs were converted into Cartesian velocity
components along the computational boundary. Since these jets were subsonic, pressure
was extrapolated from the interior and density was computed from the state equation.
58
3.6 Logical Structure of Flow Solver
Sections 3.1 through Section 3.5 describe the mathematical details of the flow
solver used in this study. This section depicts the process of coding and the functioning
of the flow solver. Figure 3.6 shows a logic flow chart of the main modules that were
developed under this effort. The entire solver package consists of:
1. The main program in modular form coded in FORTRAN 77/90
2. Input files in ASCII format
3. Grid and solution files in binary format
The flow chart illustrates that GTTURBO3D consists of five main modules of
which four modules are grouped in the main iterative loop that is repeatedly executed
during each computational time step.
After all the input data are loaded into the code at the beginning of program execution,
the computational mesh is rotated about the x-axis by a small angle, typically between
one-fiftieth and one-fifth of a degree, depending on the desired time step. At the new
grid location, the multi-block-solver module computes the fluxes and solves for the
updated flowfield in each of the computational blocks. This procedure does most of the
computational work and demands the bulk of the CPU time. Next, the boundary
condition module invokes the various boundary condition routines to specify numerical
values at all six faces of each computational domain, according to the specified input. A
variety of boundary subroutines are available for both viscous and inviscid flowfield
simulations. Finally, the diagnostics and output module checks for convergence,
calculates and prints out the performance properties.
59
Figure 3.1 Cell-Centered Finite Volume Formulation and Four-Point Stencil
Figure 3.2 Boundary Conditions for Compressor Single Flow Passage
Outflow Boundary
(Coupling with Plenum)
Periodic Boundary
at Compressor Inlet
Solid Wall Boundary
at Compressor Casing
Periodic Boundary
at Diffuser
Solid Wall Boundary
at Impeller Blades
Periodic Boundary
at Clearance Gap
Solid Wall Boundary
at Compressor Hub
Inflow
Boundary
Zonal Boundary
i i-1 i+1 i+2
Cell Face
for i+
Stencil for
qright
Stencil for
qleft
Left Right
i,j,k
i
j
k
Ei+
Ei-
i- Face i+ Face
60
Figure 3.3 Coupling Between Diffuser and Plenum at Outflow Boundary
Figure 3.4 Zonal and Periodic Boundaries
Periodic
Boundaries
Block 1 Block 2
Zonal
Boundary
Dashed
Gridlines
are Used
for Averaging
mt
.
mc
.
Plenum Chamber
u(x,y,z) = 0
pp(x,y,z) = const.
s = 0
ap, Vp
Outflow
Boundary
61
Figure 3.5 Injection Boundary
5
0.1RInlet
Casing
Rotation Axis
Impeller
RInlet
62
Figure 3.6 GTTURBO3D Flow Chart
GTTURBO3D
Input
Computational Grid
Previous Solution
(if Restart Run)
Initialize Solution
(if Coldstart Run)
Main Input Data
Boundary Condition
Data
GTTURBO3D
Multi-Block-Solver
Compute Inviscid Fluxes
Compute Viscous Fluxes
Compute Eddy Viscosity
(if Turbulent Run)
Add Artificial Viscosity
Approximate Factorization
(Matrix Inversion in 3
Sweeps)
GTTURBO3D
Boundary
Conditions
Inflow/Outflow BC
Solid Surface BC
Periodic BC
Zonal BC
Injection BC
GTTURBO3D
Grid Rotation
GTTURBO3D
Diagnostics and
Output
Check For Convergence
Output Performance Data
(Flow Rate, Pressure)
Output Solution
63
CHAPTER IV
4 CODE VALIDATION STUDIES
The following three chapters are devoted to the presentation of centrifugal
compressor simulation results using the viscous flow solver. Flowfield visualization in
the form of vector and contour plots, graphs of the compressor characteristic maps and
comparisons with experimental data allowed new insight and an increased understanding
of the unsteady flow phenomena that lead to instabilities in centrifugal compressors.
Since the flow solver was applied to both low-speed and high-speed centrifugal
configurations, a second objective of this study was to investigate the differences in the
system dynamics with respect to the onset of instabilities. Finally, this study sought to
gain a better understanding of the mechanism of compressor control by air injection.
This chapter is organized as follows: First, the two centrifugal compression
systems used in this study to simulate compressor instabilities are introduced in Section
4.1. This includes their geometric models, computational grids, and characteristic
performance data at compressor design conditions. Section 4.2 and Section 4.3 compare
CFD flow predictions with experimental data at the design point for each configuration.
This validation step was intended to produce confidence in the codes and models prior to
further use in the present research. A limited number of grid sensitivity studies are also
presented for the two configurations.
64
4.1 Compressor Configurations and Grid Generation
The flow solver was validated through an analysis of two rotors, a low-speed
centrifugal compressor tested at NASA, and a high-speed compressor tested at DLR.
Their geometric dimensions along with characteristic performance data are summarized
in Table 4.1. The performance data compare these two compressors at the design point
and consider the total-to-total pressure rise across the impeller, adiabatic efficiencies, and
rotational speeds. The generation of three-dimensional grids is described for one of the
systems. Figures of the single flow passage meshes illustrate the clustering and block
structure for the two systems.
4.1.1 NASA Low-Speed Centrifugal Compressor (LSCC)
The LSCC was designed and tested at the NASA John Glenn Research Center
(formerly NASA Lewis Research Center) for fundamental compressor flow physics
research. Its specific purpose was to mimic the flowfield of large transonic impellers in a
low-speed environment. The test facilities at NASA include sophisticated laser
anemometer systems to provide CFD viscous flow solvers with a detailed experimental
test bed for validation purposes
85
. References 86 through 92 contain additional
representative studies carried out with the LSCC.
The LSCC rotor is a 55-degree backsweep impeller with 20 full blades. Its
geometric dimensions are relatively large with a 0.87 m inlet diameter and a 1.524 m exit
diameter, respectively, to ensure optimal access for optical velocity measurements.
Figure 4.1 shows a picture of the large impeller and the NASA compressor rig. The tip
65
clearance between the blade tip and the compressor casing was designed to be 2.54 mm,
1.2 to 1.8 percent of the impeller blade height.
Figure 4.2 shows a schematic of the LSCC with the geometrical characteristics
outlined above, in particular the large backsweep. This was designed to obtain spatially
constant blade loading and smooth velocity distribution while maintaining minimum
blade lean. At design operating conditions of 1,862 rpm rotational shaft speed, the LSCC
total pressure ratio was measured at 1.17 with a mass flow rate of 30 kg/sec. Additional
design parameters are given in Table 4.1.
Figure 4.3 shows the computational grid used for single flow passage simulations
in this study. The H-H-type grid has 129 grid points in the streamwise, 61 grid points in
the spanwise and 49 grid points in the pitchwise direction, respectively. Grid lines are
clustered near boundary layers and regions with large gradients in the flowfield (impeller
leading edge and trailing edge). Dr. Michael Hathaway, a researcher with NASA, who
supplied the grid, placed four points between the blade tip and the compressor casing to
resolve clearance gap effects. According to Hathaway et al.s report
85
, the original
vaneless diffuser was modified to eliminate a reversed flow region on the diffuser
backwall. This ensures favorable outflow boundary conditions for CFD analysis codes at
design conditions.
4.1.2 DLR Centrifugal Compressor (DLRCC)
In contrast to the NASA LSCC, the DLR Centrifugal Compressor (DLRCC)
shown in Figure 4.4 was designed to operate in the high subsonic regime. The
66
corresponding relative Mach numbers range from 0.92 at the rotor inlet to 0.96 at the
rotor exit. These values were measured at a design rotational shaft speed of 22,360 rpm,
at a mass flow rate of 4.0 kg/sec and a pressure rise of 4.7. The geometric dimensions
given in the photograph of the rotor indicate that the DLRCC is almost four times smaller
than the LSCC and has only a 30 degree backsweep. Additional design operating data
are summarized in Table 4.1. Since the DLRCC was designed by a CAD/CAM
procedure developed at DLR
93
, the impeller was manufactured on a five-axis NC-milling
machine. Therefore, the blade surfaces were generated by straight lines that run from hub
to tip.
The DLRCC was built to study the effect of compressor geometry on the
development of secondary flow structures that lead to increased losses and reduced
impeller efficiencies at design operating conditions. In Reference 94, detailed
experimental and numerical flowfield studies were performed for two versions of the
DLRCC having the same blade geometry but different shroud contours. The researchers
determined that by a careful re-design of the compressor casing, a low kinetic energy area
could be suppressed resulting in a smoother exit-velocity profile and better efficiencies.
Even though a re-design of the rotor was not attempted, the authors argue that additional
performance enhancements could be expected. Internal flowfield measurements in this
study include optical L2F velocity data as well as pressure data from 24 circumferentially
averaged pressure tappings. The described experimental investigations were paralleled
by three-dimensional numerical studies at design operating conditions using the Dawes
code
94,95
and the STAR-CD code
96
. While the first study on a relatively coarse
67
computational mesh only provided adequate code validation, the analysis with the STAR-
CD code agreed more favorably with experimental data. In addition, this study assessed
the effect of different inlet geometries on compressor performance and showed that the
inclusion of spinner geometry in the CFD analysis had little effect on the main
performance parameters.
References 97 and 98 contain additional measurements and performance data
collected from the DLRCC. As a result of this comprehensive experimental database, the
DLRCC is classified as an AGARD standard test case and is a suitable configuration for
code validation studies.
The computational mesh employed in the DLRCC study is shown in Figure 4.5.
It was generated by loading the geometrical data provided in Reference 94 into the
meshing package Gridgen
55
. Algebraic surface grids and H-H-type volume grids were
then generated using a hyperbolic tangent (tanh) node point distribution function near
surfaces to resolve viscous boundary layers. The resulting single flow passage grid had
the dimensions 141x49x33 in the streamwise, spanwise and pitchwise directions,
respectively. Eight computational cells were placed in the clearance gap. To ensure
sufficient grid resolution, a second grid with twice the number of grid points in each of
the three directions was generated. The computational results obtained with the two grids
were then compared against each other in a special grid sensitivity analysis. The results
from this study are discussed in Section 4.3.
68
In the experimental studies performed at DLR, the rotor was coupled to a vaneless
constant area diffuser. The diffuser hub wall was aligned perpendicularly to the machine
axis while the shroud wall was inclined against the hub wall at a small angle.
Table 4.1 Comparison of Centrifugal Compressors
Parameter LSCC DLRCC
Rotor
Number of Blades 20 24
Rotor Inlet Tip Diameter, cm 87 22
Rotor Exit Tip Diameter, cm 152.4 40
Blade Backsweep, degrees 55 30
Clearance Gap, mm 2.54 0.5 (Inlet)0.2 (Exit)
Inlet Tip Relative Mach Number 0.31 0.92
Exit Tip Absolute Mach Number 0.29 0.96
Design Conditions
Rotational Speed, rpm 1,862 22,360
Flow Rate, kg/sec 30 4
Pressure Ratio 1.17 4.7
Adiabatic Efficiency, % 92.2 83
4.2 Validation of the NASA Low-Speed Centrifugal Compressor (LSCC)
As a first step in using the numerical code to model compressor instabilities, the
viscous flow solver was validated against experimental measurements. The availability
of performance data was a prerequisite for the computational analysis of the two test
configurations described in the previous section. Most experimental turbomachinery
studies record velocity, pressure and performance measurements at operating design
69
conditions. Flowfield velocities are usually obtained using optical methods, such as laser
anemometry. Pressure and performance measurements rely on static or total pressure
tappings that are evenly distributed around the annulus. Surface blade pressure
measurements require special treatment for tap mounting and wiring due to the impeller
rotation. Such measurements are difficult and more expensive. In this study, only one
test configuration database (LSCC) contains blade surface pressure measurements at
operating design conditions and at a single off-design condition. The results of these
validation studies are discussed in the following sections. Additional results from grid
sensitivity studies are included to illustrate that the baseline computational mesh
described in the previous section was sufficient in resolving the main fluid flow
phenomena in the LSCC.
4.2.1 Validation Results at Design Operating Conditions
The flowfield was advanced by several thousand time steps in a time-accurate
mode until a steady-state design mass flow rate of 30 kg/sec was achieved. All numerical
simulations with the LSCC mesh were computed on a high-performance Silicon Graphics
Origin 2000 with multiple processors, using either four or six processors per run. During
the early stages of this research, the flow solver code was optimized for execution of
turbomachinery components on shared-memory parallel computers such as the Silicon
Graphics Origin 2000. Using the auto-parallelization option of the Origins FORTRAN
compiler and employing four parallel processors, the code executed at about 1x10
-5
sec
per point per time step while running in a time-accurate mode. Although the present
70
implicit scheme can use large time steps, the CFL number was maintained at or below
two for resolving the unsteady flow in a time-accurate manner. For the LSCC
calculations at design speed, the computational time step was determined to yield
approximately one degree of rotor revolution per 25 iterations. At this rate, the flowfield
simulation of one rotor revolution required about 9,000 computational time steps which
equated to approximately five to six hours of elapsed wall clock time on the Origin 2000
using four processors. Engaging six processors increased the program execution by
about 10 percent. However, since a large number of unsteady runs were required during
this study, the maximum number of processors was not increased beyond six. The
additional processors available on the computer were used to execute multiple runs
simultaneously.
The LSCC calculations at design conditions were generally run for three to four
compressor revolutions before a steady-state flowfield was obtained. The mass flow rate
across several streamwise planes was monitored and used as a convergence criterion.
Figures 4.6 through Figure 4.11 show comparisons between the computed and
measured static pressures at the design mass flow rate of 30 kg/sec. Experimental and
computational pressure data are each plotted without the contribution from the centrifugal
pumping effect (V
2
), thus containing only the pressure rise due to the deceleration of
fluid particles in the rotor. Since the LSCC operates at a low shaft rotational speed, the
increase in static pressure as plotted in Figures 4.6 through 4.11 was small.
Comparisons are given at several span stations. Within this context, a span
station is defined as a line on the compressor blade that runs from the blade leading edge
71
to the trailing edge along the flow direction. The 100 percent span station corresponds to
the blade tip, while zero percent span corresponds to the line where the blade is
connected to the hub. The overall agreement between experiments and computational
data is reasonable. Local differences remain well below one percent. Slight
discrepancies in predicted static pressures near the compressor casing (93 percent span
and 97 percent span) can be attributed to the tip clearance effect. Both the CFD
predictions and the experimental data show a region of minimum pressure near the
leading edge blade tip on the suction side. As the fluid enters the impeller region, it
immediately accelerates, similar to the case of an airfoil at an angle of attack. This
increase in speed is accompanied by a decrease in static pressure.
In addition to the comparison between experimental and computational data in
Figures 4.6 through 4.11, the results of the grid sensitivity study are included in these
graphs. CFD calculations were performed with two different computational meshes. The
results labeled as CFD Baseline Grid were obtained using the mesh described in Section
4.1.1 and graphed in Figure 4.3 (dimensions 129x61x49; 322,000 grid points). To ensure
grid insensitivity of the CFD results, additional calculations were performed on a fine
grid with twice the number of grid points in each of the three directions. The fine grid
had dimensions of 259x119x97 in the streamwise, spanwise and pitchwise direction,
respectively, totaling almost three million nodes. However, the computed pressures
along the two sides of the compressor blades showed only a slight difference between the
baseline grid solution and the fine grid solution. Additional investigations of the two
flowfields also demonstrated that the baseline grid was sufficiently dense in boundary
72
layer regions and other regions of potentially large gradients. Therefore, all subsequent
computational results were obtained using the baseline mesh.
Figures 4.12 through 4.14 compare computed velocity data to velocity
measurements obtained using a two-component laser fringe anemometer operating in on-
axis backscatter mode. The uncertainty of these measurements was approximately two
percent. Plots of the axial velocity component referenced by the rotor exit tip speed,
u/U
tip
, are given for three chordwise planes: 25 percent, 40 percent and 90 percent chord.
At design operating conditions, the rotor exit tip speed, U
tip
, was 153 m/sec. At each of
the three chordwise stations, numerical and experimental data were compared at two
spanwise planes. The results show that the velocity profiles increased almost linearly
from the pressure side (0 percent pitch) to the suction side (100 percent pitch) of the
compressor blades. Thus, regions of higher pressure experienced lower fluid velocities.
The axial velocity profiles at 90 percent chord showed negligible values indicating that
the flow near the rotor trailing edge was almost purely radial except for two small
separation regions near the blades. The agreement between computational and
experimental results can be considered reasonable for all velocity profiles. The largest
discrepancy between CFD and experiments was 14 percent for the case at 25 percent
chord and 60 percent span.
Figure 4.15 illustrates the development of the impeller throughflow in the form of
vector plots. In these plots, the velocity vectors were obtained by transforming the three-
dimensional velocity field to the x-r-meridional plane. Three such planes with projected
velocity vectors are shown: 4, 50 and 97 percent away from the pressure surface. On
73
each plane, the overall flowfield was well behaved and attached. Small regions of
reversed flow were present near the compressor shroud. This separation zone was part of
a throughflow momentum deficit that existed in the outer 10 to 20 percent of the span.
The same wake-like phenomenon was reported by Hathaway et al.
85
who experimentally
studied the NASA LSCC. They concluded that this momentum deficit is generated as a
result of the tip clearance flow. A similar phenomenon has been observed in several
other centrifugal compressor facilities by various researchers
97,99
.
4.2.2 Validation Results at Off-Design Operating Conditions
The extensive experimental database for the NASA Low-Speed Centrifugal
Compressor (LSCC) includes a limited number of off-design blade pressure data. These
measurements were taken at a mass flow rate of 23.5 kg/sec, approximately 75 percent of
the design mass flow rate. This off-design operating condition is stable as is the design
operating point. According to the researchers at NASA, audible compressor instabilities
were recorded for operating conditions below 65 percent of the design mass flow rate.
Figure 4.16 through Figure 4.21 show a comparison between off-design predicted
and measured blade pressures at the same span stations described in Figure 4.6 through
Figure 4.11. In comparison to these design blade pressures, the values at off-design
condition were approximately two percent higher and the suction peak was slightly
stronger. The latter was a result of the reduced flow rate which, in turn, produced larger
flow incidence angles, thus causing the flow to accelerate more rapidly near the blade
74
leading edge. The agreement between measured and computed results lies within one
percent.
4.3 Validation of the DLR Centrifugal Compressor (DLRCC)
After validation of the flow solver at nearly incompressible flow conditions in the
LSCC, the code was used to model transonic flow in the DLR centrifugal compressor. At
design conditions (4.0 kg/sec mass flow rate), the predicted pressure rise across the
impeller was 4.5. This computed value underpredicted the measured result ( = 4.7) by
approximately 4 percent. The researchers at DLR in Cologne, Germany did not measure
any DLRCC blade surface pressures; only static shroud pressures and meridional
velocities were recorded in the DLRCC database
94
for comparison with CFD solvers. In
contrast to the LSCC experimental results discussed in the previous section, experimental
DLRCC measurements were limited to the design operating condition. The internal
flowfield of the DLRCC was analyzed with a Laser-Two-Focus Velocimetry developed
at DLR. This technique measures the magnitude and orientation of the mean absolute
flow vector. The shroud static pressures were measured at several locations along the
shroud meridional chord and were then time-averaged over the duration of the
measurement cycle.
Figure 4.22 shows a comparison between the measured (time-averaged) pressure
and the computed static pressure on the shroud along the blade meridional chord, s/s
mer
.
As the fluid particles travel from the blade leading edge (s/s
mer
= 0) to the blade trailing
edge (s/s
mer
= 1), the shroud static pressure nearly triples due to the transference of energy
75
from the impeller to the flow. Although the tip clearance effect was expected to produce
complex flow structures near the shroud, the agreement between measurements and the
numerical predictions exceeded expectations resulting in excellent correlation.
The CFD calculations at design operating conditions were performed using two
different computational grids. The results labeled CFD Baseline Grid were produced
with the mesh described in Section 4.1.2 and plotted in Figure 4.5. To demonstrate grid
insensitivity of these results, a second grid with twice the number of grid points in each
direction was tested. The dimensions of the grid labeled CFD Fine Grid in Figure 4.22
were 281x97x65. The computed static shroud pressures from this analysis, however,
showed nearly the same result regardless of the grid used. Consequently, the baseline
grid was assumed to be sufficient for capturing the main flow phenomena in the DLRCC
with reasonable accuracy. Due to limitations in computational resources, this grid was
utilized in all subsequent discussions.
Figure 4.23 through Figure 4.25 show comparisons between measured and CFD-
predicted meridional velocities at three different chordwise locations. The meridional
velocity was obtained by projecting the relative velocity vector into the meridional plane
and calculating the component that was orthogonal to the shroud contour at each
respective chordwise location. Therefore, it may be viewed as a close approximation to
the throughflow velocity that crosses a quasi-orthogonal plane. At each of the three
chordwise locations, experiments and computations were compared at 10 percent and 70
percent depth relative to the shroud contour. The meridional velocity profiles at 60
percent and 99 percent chord show a large velocity deficit near the compressor casing.
76
This wake-like flow phenomenon is a result of the tip clearance vortex and extended from
about 40 percent chord into the impeller exit plane at 99 percent chord. This trend was
captured by the flow solver with reasonable accuracy. The flow velocities at 70 percent
depth are virtually unaffected by this throughflow deficit.
Figure 4.26 depicts the DLRCC flowfield in the form of velocity vectors colored
by total pressure plotted in meridional planes near the pressure side, near midpassage and
near the blade suction side. Both the velocity vectors and the total pressure distribution
revealed no separation regions throughout the entire flowfield. Such behavior was
expected for compressor flow at operating design conditions and was in agreement with
the DLR measurements. The velocity deficit described in the previous paragraph was
again visible in the meridional plane near the midpassage.
77
Figure 4.1 Low-Speed Centrifugal Compressor (LSCC) in NASA Test Rig (source:
NASA)
Figure 4.2 Schematic of Low-Speed Centrifugal Compressor (LSCC)
78
Figure 4.3 Single Flow Passage Computational Grid for LSCC (129x61x41)
Figure 4.4 DLR Centrifugal Compressor (DLRCC), (source: Krain
97
)
40 cm
79
Figure 4.5 Single Flow Passage Computational Grid for DLRCC (141x49x33)
85
LSCC Operating Design Conditions (30kg/sec)
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 0.2 0.4 0.6 0.8 1
Meridional Distance, s/s
mer
S
t
a
t
i
c

P
r
e
s
s
u
r
e
,

p
/
p
CFD Fine Grid

CFD Baseline Grid
Exp
80
85
85
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 0.2 0.4 0.6 0.8 1
mer
S
t
a
t
i
c

P
r
e
s
s
u
r
e
,

p
/
p
CFD Fine Grid

CFD Baseline grid
Exp
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 0.2 0.4 0.6 0.8 1
mer
S
t
a
t
i
c

P
r
e
s
s
u
r
e
,

p
/
p
CFD Fine Grid

CFD Baseline Grid
Exp
81
85
85
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 0.2 0.4 0.6 0.8 1
mer
S
t
a
t
i
c

P
r
e
s
s
u
r
e
,

p
/
p
CFD Fine Grid

CFD Baseline Grid
Exp
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 0.2 0.4 0.6 0.8 1
mer
S
t
a
t
i
c

P
r
e
s
s
u
r
e
,

p
/
p
CFD Fine Grid

CFD Baseline Grid
Exp
82
85
85
Axial Velocity, u/U
tip
, at 25 Percent Chord, LSCC
Operating Design Conditions (30kg/sec)
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 0.2 0.4 0.6 0.8 1
mer
S
t
a
t
i
c

P
r
e
s
s
u
r
e
,

p
/
p
CFD Fine Grid

CFD Baseline Grid
Exp
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0% 20% 40% 60% 80% 100%
Percent Pitch From Pressure Surface
u
/
U
t
i
p
CFD 80% Span
CFD 60% Span
Exp 80% Span
Exp 60% Span
83
85
Axial Velocity, u/U
tip
85
Axial Velocity, u/U
tip
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0% 20% 40% 60% 80% 100%
Percent Pitch from Pressure Surface
u
/
U
t
i
p
CFD 70% Span
CFD 50% Span
Exp 70% Span
Exp 50% Span
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0% 20% 40% 60% 80% 100%
u
/
U
t
i
p
CFD 60% Span
CFD 40% Span
Exp 60% Span
Exp 40% Span
84
4% Away from
Pressure Side
50% Away from
Pressure Side
97% Away from
Pressure Side
Wake-like
Momentum
Deficit
Figure 4.15 Velocity Field 4%, 50% and 97% Away from the Pressure Surface, LSCC
85
LSCC Operating Off-Design Conditions (23.5kg/sec)
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 0.2 0.4 0.6 0.8 1
mer
S
t
a
t
i
c

P
r
e
s
s
u
r
e
,

p
/
p
CFD
Exp
85
85
85
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 0.2 0.4 0.6 0.8 1
mer
S
t
a
t
i
c

P
r
e
s
s
u
r
e
,

p
/
p
CFD
Exp
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 0.2 0.4 0.6 0.8 1
mer
S
t
a
t
i
c

P
r
e
s
s
u
r
e
,

p
/
p
CFD
Exp
86
85
85
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 0.2 0.4 0.6 0.8 1
mer
S
t
a
t
i
c

P
r
e
s
s
u
r
e
,

p
/
p
CFD
Exp
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 0.2 0.4 0.6 0.8 1
mer
S
t
a
t
i
c

P
r
e
s
s
u
r
e
,

p
/
p
CFD
Exp
87
85
98
Static Pressure Along Shroud (Circumferentially
Averaged), p/p
0
,, DLRCC Operating Design Conditions (4.0 kg/sec)
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 0.2 0.4 0.6 0.8 1
mer
S
t
a
t
i
c

P
r
e
s
s
u
r
e
,

p
/
p
CFD Exp
0
0.5
1
1.5
2
2.5
3
0 0.2 0.4 0.6 0.8 1
mer
S
t
a
t
i
c

P
r
e
s
s
u
r
e
/
T
o
t
a
l

P
r
e
s
s
u
r
e
,

p
/
p
0
,
CFD Fine Grid
CFD Baseline Grid
Exp (Time Mean)
88
94
mer
/U
tip
, at 20% Chord,
DLRCC Operating Design Conditions (4.0 kg/sec)
94
mer
/U
tip
, at 60% Chord,
0
0.1
0.2
0.3
0.4
0.5
0% 20% 40% 60% 80% 100%
c
m
e
r
/
U
t
i
p
Exp 10% Depth
Exp 70% Depth
CFD 10% Depth
CFD 70% Depth
0
0.1
0.2
0.3
0.4
0.5
0% 20% 40% 60% 80% 100%
c
m
e
r
/
U
t
i
p
Exp 10% Depth
Exp 70% Depth
CFD 10% Depth
CFD 70% Depth
89
94
mer
/U
tip
, at 99% Chord,
Figure 4.26 Computed Meridional Velocity Vectors Colored by Total Pressure, DLRCC
Operating Design Conditions (4.0 kg/sec)
Near
Suction
Side
Midpassage
Near
Pressure
Side
0
0.1
0.2
0.3
0.4
0.5
0% 20% 40% 60% 80% 100%
c
m
e
r
/
U
t
i
p
Exp 10% Depth
Exp 70% Depth
CFD 10% Depth
CFD 70% Depth
90
CHAPTER V
5 COMPUTATIONAL RESULTS AT STALL CONDITIONS
The results discussed in the previous chapter focused on the modeling of fluid
flow in centrifugal compressors near design conditions. At these conditions, stable
compressor operation is ensured because of the surge margin between the design point
and the unstable region in the compressor performance map. Most compressor control
systems currently employed in industry are based on this strategy, also called avoidance
control.
However, as the flow through the system is decreased from the design point to the
surge limit, the steady compressor flow pattern becomes unstable and leads to the
development of rotating stall and surge, fluid dynamic phenomena that must be
suppressed. Thus, over the course of the past few decades, much scientific effort has
been concentrated on analyzing, understanding and ultimately alleviating these
instabilities.
This chapter discusses computational results from unsteady CFD simulations that
modeled a centrifugal compression system approaching the surge limit from first
principles. The detection of flowfield phenomena that lead to the onset of instabilities as
well as a detailed analysis of limit cycle oscillations were the main emphases. Results are
shown in the form of performance maps with unsteady time histories for selected points
and vector/contour plots of the flowfield identifying separation zones. Once the
91
important factors in the development of instabilities were identified, compressor control
was directed toward the successful suppression of these phenomena. Results of this
second step are discussed in Chapter VI.
5.1 High-Speed Compressor Results at Off-Design Conditions
The performance of a centrifugal compressor is typically evaluated by mapping
the pressure rise against the flow rate for fixed inlet conditions, rotational shaft speed and
guide vane angles. Such performance characteristic maps are sometimes referred to as
head-flow curves. Figure 1.3 schematically illustrates the shape of a compressor
characteristic map. By allowing the rotational speed to take a series of discrete values, a
family of performance maps can be generated. Compressor manufacturers
experimentally measure the pressure rise across a single impeller, a compressor stage or
an entire multi-stage compression system at various shaft speeds. Common flow
measuring instruments include orifice plates and venturi meters.
Similarly, constant torque curves can be used to plot the amount of energy that
must be added to the fluid to sustain a given flow rate. Each of these torque curves
represents a combination of gas properties, inlet and outlet piping, valve positions,
backpressures and operating devices. The intersection of a constant rotational shaft speed
curve with a constant torque curve determines the operating point. By parametrically
varying the backpressure with a plenum throttle valve, the operating condition can be
adjusted to a specified flow rate and pressure rise.
92
In this study, the performance map was constructed in a similar fashion. Equation
(3.51) in Section 3.5.2.3 illustrates how the backpressure was calculated from a user-
specified mass flow rate at the plenum throttle. In turn, the pressure difference between
the compressor inlet and the diffuser exit determined the mass flow rate at the operating
point. For the DLRCC, the flow solver was used to simulate compressor flow at various
operating conditions.
Figure 5.1 shows a comparison between the calculated and the experimental
DLRCC performance map. The choke limit (at high mass flow rates) and the surge limit
(at low mass flow rates) frame the stable operating range of the DLR centrifugal
compressor on both ends of the abscissa, while the design operating point (4.0 kg/sec
mass flow rate) was designed to be sufficiently separated from both limits. In an attempt
to quantify the flow unsteadiness at each computed operating point, horizontal and
vertical bars were added to the time-averaged mean flow data. The size of each bar
depicts the magnitude of the fluctuations in the flow rate and pressure rise. At high mass
flow rates, e. g. 4.6 kg/sec, the fluctuations in flow variables stayed well below one
percent, and the flowfield remained attached and well aligned throughout the entire
domain. As the flow rate through the compressor was decreased, the performance curve
reached a maximum pressure rise at approximately 3.4 kg/sec. Although detailed
experimental off-design measurements were not available for the DLRCC, Krain et al.
94
reported the onset of instabilities for the same operating condition (3.4 kg/sec). At a
lower mass flow rate of 3.2 kg/sec, the fluctuations in the mass flow rate and pressure rise
increased by a factor of approximately 25 compared to the operating point at 4.6 kg/sec.
93
The fluctuations remained bounded around the time-averaged mean value. This means
that the observed limit cycle oscillations did not grow into deep surge.
In addition to the performance map data, Figure 5.2 shows the time history for
four selected points, A through D, on the performance map. The four points chosen were:
A: Mass flow rate = 4.6 kg/sec (stable operation)
B: Mass flow rate = 3.75 kg/sec (stable operation with increased fluctuations)
C: Mass flow rate = 3.4 kg/sec (peak pressure point, onset of stall)
D: Mass flow rate = 3.2 kg/sec (unstable operation)
Plots A through D show the percent fluctuations in mass flow rate and pressure rise at
each selected operating condition, respectively. These plots are similar to the Poincare
maps used to study chaos in non-linear systems. At the stable operating point, referred to
as an attractor, A, fluctuations in mass flow rate and pressure rise were less than one
percent. As the mass flow rate was decreased, the fluctuations increased to two to three
percent (point B) and 10 percent (point C) near the point of maximum pressure rise. A
further reduction in mass flow caused the performance map to drop off which was again
accompanied by a large increase in fluctuations, namely 20 to 30 percent at point D. In
this case, much of the flow through the impeller was reversed.
5.2 High-Speed Compressor Results During Stall Conditions
Figure 5.3 shows the transient response of the computed compressor mass flow
rate during stall conditions. This data corresponds to a time-accurate CFD simulation at a
specified plenum throttle mass flow of 3.2 kg/sec. After each computational time step,
94
the momentum-averaged mass flow rate was evaluated at 73 stations along the blade
chord and ensemble-averaged, to obtain the mean mass flow rate across the impeller.
The diagram illustrates how the limit cycles develop over time. After 30 rotor
revolutions, the amplitude grew to 1.7 kg/sec. The stall mechanism was affected by the
interaction of the compressor with the plenum chamber. Therefore, the stall frequency
was largely a function of the plenum volume. The greater the volume, the lower the stall
frequency. Using the numerical values for the plenum volume described in Chapter IV,
the computed stall frequency was approximately 90 Hz, one-hundredth of the blade
passing frequency.
As outlined in Section 1.5, the B-parameter represents a criterion to determine
whether surge or rotating stall will occur as the compressor is throttled towards the surge
line. For surge to occur, the system B-parameter must exceed a critical value that varies
from system to system
11
. Using V
p
/a
2
= 5 and the DLRCC exit tip speed of 468.31 m/sec
yielded a B-parameter of B = 4.3 for the current DLR compression system. This value
was large in comparison to critical B-parameters reported by other researchers
13,14
.
Compression systems in these studies exhibited critical B-parameters between one and
two. Based on this observation, it seems that the instabilities encountered in Figure 5.2
and Figure 5.3 were indicative of surge-like behavior. However, surge tends to occur at
frequencies well below the computed stall frequency of 90 Hz. The particular type of
fluid dynamic instability observed in Figure 5.2 and Figure 5.3 could therefore not be
readily determined.
95
Figure 5.4 depicts the flowfield phenomena that contributed to the development of
a limit cycle in the DLRCC. These snapshots, taken at various time intervals over the
period of one limit cycle, T
stall
, reveal the growth of two large separation regions within
the DLR centrifugal compressor. A separation zone originates at the impeller leading
edge (at t = 0.25T
stall
) and extends to approximately 75 percent chord length. Since this
reversed flow region is located near the compressor casing, it was concluded that the
vortex flow pattern that forms in compressors as a result of tip clearance leakage likely
contributed to this instability.
A second separation zone appears near the diffuser hub. The first separation zone
most likely triggered this separation. As a result of the shroud separation, the flow
leaving the rotor was deflected and directed away from the casing which, in turn, changed
the local flow angles. The diffuser hub was not designed for such flow angles.
Therefore, the hub boundary layer separated and gave rise to diffuser stall. This
condition occurred when the instantaneous flow rate through the compressor was smallest
during the limit cycle (at t = 0.5T
stall
), due to the partial blockage of the compressor flow
passage. At this condition, the two separation zones reached their largest dimensions and
enveloped significant parts of the flowfield. Over a period of time, the leading edge flow
recovered from the separation, producing more favorable flow conditions in the diffuser.
Both separation zones disappeared (at t = 0.75T
stall
), and the limit cycle was repeated.
Figure 5.5 shows a more detailed analysis of the phenomena that gave rise to the
large leading edge separation zone described above. The relative velocity vector plots are
shown in planes of constant 99 percent span at five instances in time during the limit
96
cycle. The two-dimensional flowfield vectors were obtained by projecting the three-
dimensional velocity vector onto the plane of constant span. At the beginning of the limit
cycle, all boundary layers were attached, and the flow was well behaved. A low-pressure
region extending downstream from the leading edge into the flow channel was caused by
fluid particles rapidly accelerating away from the leading edge stagnation point, similar
to the case of an airfoil at an angle of attack. However, due to a decrease in local mass
flow and the attendant adverse pressure gradient, the suction side boundary layer
separated and gave rise to unsteady vortex shedding at t = 0.25T
stall
. At t = 0.38T
stall
, the
flow structure broke down and at t = 0.5T
stall
the flow direction was completely reversed.
The reversed flow caused the downstream plenum pressure to drop, which allowed the
flow to gradually recover and the blade boundary layer to re-attach (at t = 0.87T
stall
). The
limit cycle was then repeated.
5.3 Low-Speed Compressor Results at Off-Design Conditions
As in the case of the high-speed compressor, the compressor performance
characteristic map served as a starting point for analyzing flow instabilities in the NASA
Low-Speed Centrifugal Compressor (LSCC). After validation of the flow solver, the
applied boundary conditions and the computational mesh at operating design conditions
in Section 4.2, the compressor flowfield was evaluated at different flow rates. Figure 5.8
illustrates the result of this analysis for two different rotational shaft speeds.
Experimental data are compared with the CFD results at design speed. Both
experimental and computational data show the same quantitative behavior. As the flow
97
rate decreased, the pressure ratio slightly increased. The reason the slope of the design
speed curve stayed rather small can be attributed to the low rotational shaft speed of the
compressor which resulted in mild pressure ratios. The predicted performance map was
in good agreement with measurements performed by Hathaway et al.
85
.
In contrast to the behavior of the high-speed compressor, discussed in Section 5.1
and Section 5.2, the fluctuations of flowfield properties in the LSCC were mild even
when the flow rate through the compressor was decreased. The horizontal and vertical
bars added to the design speed performance curve illustrate that the fluctuations at 50
percent of the design mass flow only increased by a factor of approximately five, relative
to predicted values at design conditions. This suggests that a much weaker form of limit
cycles developed in the LSCC at design speed.
Figure 5.9 depicts the flowfield phenomena that cause compressor stall in the
LSCC. The growth of a shroud separation zone is illustrated in the form of relative
velocity vectors near 50 percent pitch. The three snapshots correspond to flowfields near
midpassage at three different off-design operating conditions. At a mean mass flow rate
of 19 kg/sec through the compressor, two small regions experienced reversed flow, one
region in the vicinity of the compressor casing and another region near the diffuser hub.
The remaining flowfield, however, was well aligned. This operating condition marked
the onset of instabilities in the LSCC, a phenomenon that was confirmed by Hathaway et
al.
85
who reported audible instabilities at operating conditions below 20 kg/sec. Even
though the shroud separation grew in size as the flow rate was throttled to 13 kg/sec, it
did not affect the flowfield upstream of the compressor face. A visualization of the
98
unsteady flowfield at this operating condition revealed that large incidence angles
produced vortex shedding near the LSCC leading edge which explains the increase in
flow rate fluctuations in Figure 5.8. However, these phenomena were not severe enough
to cause a cyclic breakdown of the flow pattern as was observed in the high-speed
compressor shown in Figure 5.5.
Three main questions demand answers: Did this flow phenomenon occur in the
LSCC because of its low-speed and nearly incompressible flowfield? Was the overall
energy content of the inflowing particles too low to cause local perturbations or
instabilities? Conversely, do high-speed centrifugal compressors with an inherently
higher influx of kinetic energy exhibit a greater tendency to experience limit cycle
oscillations? To address these questions, the off-design behavior of the LSCC low-speed
system was examined at a rotational shaft speed of twice the design speed.
The LSCC performance map graphed in Figure 5.8 shows the results of
simulations at twice the design speed. While the predicted operating points illustrate that
the pressure ratio significantly increased as the rotational shaft speed doubled, the curve
reached a peak pressure point at a mass flow rate of 30 kg/sec. This was accompanied by
large limit cycle oscillations in observed flow properties. Mass flow rate fluctuations at
the peak pressure point increased by a factor of approximately 15 compared to the stable
operating point at 45 kg/sec. Figure 5.10 shows the flowfield details that reveal the
source of these instabilities. The four snapshots correspond to flowfields near midpassage
at different instances during the LSCC limit cycle. A growing separation zone originated
from the impeller leading edge and enveloped much of the flow passage near the
99
compressor casing. The initial flow separation near the blade leading edge at t = 0.25T
stall
was caused by an increased incidence angle at the reduced mass flow rate. The azimuthal
flow component remained unaffected because the shaft rotational speed was held
constant. The axial flow component, however, decreased due to the smaller flow rate.
As a result, the net angle of attack increased, making the flow more susceptible to
separation. The snapshot taken at t = 0.5T
stall
illustrates that approximately 25 percent of
the blade span was covered by the separation zone that extended downstream into the
impeller region. In effect, the LSCC started to develop limit cycle oscillations. At this
operating condition, the compressor could not provide adequate energy to the fluid to
sustain the desired flow rate.
In an attempt to identify the causes of the increased instabilities at 200 percent
design speed compared to the considerably milder fluctuations at design speed, an
analysis of compressibility effects and the B-parameters was performed. Table 5.1 shows
a comparison of relative Mach numbers and B-parameters for the two cases considered.
The relative Mach numbers were calculated at two different locations of the compressor
blade leading edge: close to the compressor hub and in the vicinity of the blade tip. The
results indicate that the flowfield at 100 percent speed was nearly incompressible while
the relative Mach numbers at twice the design speed varied from 0.26 to 0.475, low
compressible flow conditions. Section 1.4 outlines the damped mass-spring system used
to characterize compressor instabilities. In this model, the fluid compressibility
represents the potential energy, and the inertia of the particles gives rise to the kinetic
energy. From this model, it is evident that compressibility effects in the plenum are
100
fundamental in the development of instabilities. A comparison of the B-parameters in
Table 5.1 reveals that due to the increased tip velocity at 200 percent design speed, the B-
parameter doubled. These two observations seem to support the conclusion that low-
speed centrifugal compressors exhibit a greater resistance towards the development of
limit cycle oscillations than high-speed systems.
Table 5.1 Comparison of Relative Mach Numbers and B-Parameters for the Operating
Conditions Considered at 100 Percent Design Speed and 200 Percent Design Speed
Operating
Condition
Relative Mach Number Near
Leading Edge Hub
Relative Mach Number Near
Leading Edge Tip
B-Parameter
100% Design
Speed
0.13 0.23 1.95
200% Design
Speed
0.26 0.475 3.9
Figure 5.11 through Figure 5.13 show the results of a spectral analysis using the
computed pressure rise fluctuations during limit cycles in the LSCC at 200 percent design
speed. The time-averaged pressure rise of 1.98 corresponds to a mean mass flow rate of
30.1 kg/sec at this operating condition. The unsteady response plotted in Figure 5.11
shows five limit cycles of nearly equal amplitude. Fluctuations of higher frequency and
smaller amplitude were superimposed to the computed stall frequency of approximately
70 Hz. While the predicted stall frequency was largely a function of the fixed plenum
volume, the frequency spectrum plotted in Figure 5.12 shows that the small-scale
fluctuations varied in frequency between 350 and 400 Hz. These fluctuations exhibited
101
wave numbers of 25 and higher as shown in Figure 5.13. There was not any dominant
frequency associated with these small-scale oscillations. Based on this observation, it
was concluded that they were caused by unsteady leading edge vortex shedding.
Depending on the instantaneous mass flow rate, the incidence angle varied with time,
thus producing different leading edge flow conditions at different span stations.
The conclusions drawn from the above-described phenomena provided the
starting point for the implementation of improved compressor control strategies.
Compressor control tailored to both the high-speed and the low-speed compressor must
address the development and growth of the described separation zones in order to extend
the useful compressor operating range. The knowledge gained from the computed
compressor flowfields at different operating conditions provided valuable information on
altering the flow pattern in order to suppress undesirable flow recirculation zones.
The diffuser flow may be improved by means of diffuser guide vanes or hub
suction. The shroud separation may be addressed by casing treatments, inlet guide vanes
or air injection. Since air injection directly affects the flowfield near the compressor
leading edge, such active control strategies are most suitable in alleviating leading edge
vortex shedding, thus suppressing the development of the described separation zones.
Control of centrifugal compressors using airjets in a steady and an unsteady manner is
discussed in the following chapter.
102
94
DLRCC Performance Map With the Amplitude of
the Fluctuations Denoted by Horizontal and Vertical Bars
Figure 5.2 Computed Time History for Selected Points A-D on DLRCC Performance
Map, A: sec / kg 6 . 4 m & , B: sec / kg 8 . 3 m & , C: sec / kg 4 . 3 m & , D: sec / kg 2 . 3 m &
3
3.5
4
4.5
5
5.5
2 2.5 3 3.5 4 4.5 5
Mass Flow (kg/sec)
T
o
t
a
l

P
r
e
s
s
u
r
e

R
a
t
i
o
Exp
CFD
Surge Limit Choke Limit
A B
C D
103
Figure 5.3 Computed Transient Response of the Mass Flow Rate in the DLRCC During
Limit Cycles (Point D in Figure 5.2)
Figure 5.4 Growth of Reversed Flow Regions in the DLRCC During Limit Cycles (Point
D in Figure 5.2)
t/2,
t = 0 t = 0.25T
stall
t = 0.5T
stall
t = 0.75T
stall
104
A: t = 0
B: t = 0.25T
stall
C: t = 0.38T
stall
D: t = 0.50T
stall
E: t = 0.87T
stall
Figure 5.5 Relative Velocity Vectors Near Blade Leading Edge During DLRCC Limit
Cycles (Single Flow Passage Top View V-V at 99 Percent Span)
D E
A B C
V V
105
85
LSCC Performance Map With the Amplitude of
the Fluctuations Denoted by Horizontal and Vertical Bars
Figure 5.7 Growth of Shroud Separation Zone in LSCC at Design Speed and Different
Operating Conditions, Impeller Regions are Shaded
1
1.2
1.4
1.6
1.8
2
2.2
10 20 30 40 50
Mass Flow Rate (kg/s)
T
o
t
a
l

P
r
e
s
s
u
r
e

R
a
t
i
o
Exp
CFD
100% Design Speed
200% Design Speed
sec / kg 13 m &
Velocity
Vectors at
Midpassage
sec / kg 19 m &
sec / kg 15 m &
106
Figure 5.8 Velocity Vectors Near Leading Edge During LSCC Limit Cycles at 200
Percent Design Speed and at 50 Percent Pitch; Shading Indicates the Impeller
t = 0 t = 0.25T
stall
t = 0.5T
stall
t = 0.75T
stall
107
Figure 5.9 Unsteady Pressure Rise During LSCC Limit Cycles at 200 Percent Design
Speed
Figure 5.10 Computed Frequencies During LSCC Limit Cycles at 200 Percent Design
Speed
Frequency (Hz)
P
o
w
e
r

108
Figure 5.11 Computed Wave Numbers During LSCC Limit Cycles at 200 Percent Design
Speed
109
CHAPTER VI
6 AIR INJECTION COMPRESSOR CONTROL
Considerable experimental evidence exists proving that air injected in a steady or
unsteady fashion improves the overall operability of the compressor
34-44
. By properly
designing fast injection valves and placing them upstream of the compressor face,
researchers were able to achieve stable operation and a significant increase in the stall
margin. However, to date, a certain degree of confusion exists among compressor
researchers as to the reason air injection is successful and how it is best applied. Day
34
proposed the suppression of stalling disturbances and the removal of stalling cells to be
the primary mechanism. Weigl et al.
35
attributed the increase in operability to the
additional momentum, added by the jets, that increases the total pressure at the rotor inlet
and suppresses separated and/or reversed flow. Yeung et al.
41
and Weigl et al.
35
concluded that air injection inherently modulates the shape of the performance map and
shifts the last stable operating point to lower mass flow rates.
Table 6.1 presents an overview of several experimental studies employing air
injection compressor control. Although most of the studies were performed at different
operating conditions and with varying compression systems, this literature survey is
intended to identify possible trends or commonalties. Almost all air injection schemes
listed utilize injection mass flow rates up to approximately five percent of the compressor
mass flow rate. Compression systems with higher injection rates may still yield operating
110
enhancements but may not be efficient for use in a real compressor because of the high-
pressure air supply needed to operate the injection system. A second significant point is
that three of the four injection systems consist of twelve evenly distributed injection
valves around the compressor annulus. These parameters were chosen to produce a
spatially uniform sheet of high-momentum fluid near the leading edge tip, therefore
postponing the growth of the clearance gap separation zone to smaller flow rates.
Bright
84
, however, showed that four half-height valves injecting a total of 0.9 percent of
the main flow rate into the blade tip region yields the same operating enhancements as
eight full-height valves using 3.6 percent of the main flow rate. This reflects a significant
reduction in external high-pressure supply needed. Furthermore, the same authors found
that these results were independent of the azimuthal arrangements of these injection
valves.
Freeman et al.
43
performed a study with varying injection angles and determined
that injecting high-pressure air into the blade tip region of the compressor gave
significantly better results than injecting the air into the core of the main compressor
flow. This finding supports the conclusion made in Chapter V that compressor
instabilities originate near the compressor blade tip, and therefore injection control must
be focused on injecting high-momentum fluid into this region. Of all studies listed in
Table 6.1, only the works by Yeung et al.
24
, Behnken et al.
39
and DAndrea et al.
44
, who
performed injection experiments using the axial compressor rig facility at California
Institute of Technology, report results of varying injection yaw angles.
111
Table 6.1 Overview of Experimental Injection Schemes
R
e
s
u
l
t
s

(
D
e
n
o
t
e
s

I
n
j
e
c
t
i
o
n

R
a
t
e

i
n

%
C
o
m
p
r
e
s
s
o
r

F
l
o
w

R
a
t
e

a
n
d

%

I
n
c
r
e
a
s
e
i
n

S
u
r
g
e

M
a
r
g
i
n
)
7
0
%

S
p
e
e
d
:

5
.
8
%

i
n
j

=
>

2
5
.
5
%

b
e
t
t
e
r

3
.
6
%

i
n
j

=
>

1
7
.
4
%

b
e
t
t
e
r

1
.
5
%

i
n
j

=
>

2
.
8
%

b
e
t
t
e
r
1
0
0
%

S
p
e
e
d
:

5
.
8
%

i
n
j

=
>

1
0
%

b
e
t
t
e
r

3
.
6
%

i
n
j

=
>

4
.
3
%

b
e
t
t
e
r
1
0
0
%

S
p
e
e
d
:

4
%

i
n
j

=
>

9
.
7
%

b
e
t
t
e
r
7
0
%

S
p
e
e
d
:
S
t
a
b
i
l
i
t
y

E
n
h
a
n
c
e
m
e
n
t

I
n
d
e
p
e
n
d
e
n
t

o
f
A
z
i
m
u
t
h
a
l

I
n
j
e
c
t
o
r

A
r
r
a
n
g
e
m
e
n
t
,
4

H
a
l
f
-
h
e
i
g
h
t

I
n
j
e
c
t
o
r
s

(
0
.
9
%

i
n
j
)

a
s
E
f
f
e
c
t
i
v
e

a
s

8

F
u
l
l
-
h
e
i
g
h
t

(
3
.
6
%

i
n
j
)
N
/
A
R
e
d
u
c
e
s

R
e
q
u
i
r
e
m
e
n
t
s

f
o
r

B
l
e
e
d

V
a
l
v
e
R
a
t
e

f
r
o
m

1
4
5

H
z

t
o

1
0

H
z
5
%

i
n
j

(
-
3
5

Y
a
w
)

=
>

1
0
.
5
%

b
e
t
t
e
r
5
%

i
n
j

(
3
0

Y
a
w
)

=
>

1
3
.
1
%

b
e
t
t
e
r
N
/
A
U
n
s
t
e
a
d
y

0
.
4
%

i
n
j

=
>

6
%

b
e
t
t
e
r
8
%

i
n
j

=
>

9
%

b
e
t
t
e
r
S
t
e
a
d
y

a
t

T
i
p

=
>

3
8
%

b
e
t
t
e
r
I
n
t
o

C
o
r
e

=
>

2
2
.
6
%

b
e
t
t
e
r
R
P
M
1
7
,
1
6
0
1
7
,
1
6
0
1
7
,
1
6
0
7
,
5
5
0
6
,
0
0
0
6
,
0
0
0
6
,
0
0
0
1
,
7
9
0
3
,
0
0
0
N
/
A
D
e
s
i
g
n
C
o
n
d
i
t
i
o
n
s
2
0
.
2

k
g
/
s
=
1
.
5
2
0
.
2

k
g
/
s
=
1
.
5
2
0
.
2

k
g
/
s
=
1
.
5
2
2
.
9
6

k
g
/
s
=
1
.
4
7
0
.
1
9

m
3
/
s
=
9
6
0

P
a
0
.
1
9

m
3
/
s
=
9
6
0

P
a
0
.
1
9

m
3
/
s
=
9
6
0

P
a
N
/
A
N
/
A
2
3

k
g
/
s
=
5
.
2
5
I
n
j
e
c
t
i
o
n
L
o
c
a
t
i
o
n
1
.
1

R
o
t
o
r
C
h
o
r
d
s
1
.
1

R
o
t
o
r
C
h
o
r
d
s
1
.
1

R
o
t
o
r
C
h
o
r
d
s
N
/
A
1

T
i
p
R
a
d
i
u
s
1

T
i
p
R
a
d
i
u
s
1

T
i
p
R
a
d
i
u
s
N
/
A
N
/
A
N
/
A
I
n
j
e
c
t
o
r
T
y
p
e
1
2

S
h
e
e
t

o
r
3
-
H
o
l
e
1
2

S
h
e
e
t

o
r
3
-
H
o
l
e
4
-
8

S
h
e
e
t
N
/
A
3

V
a
l
v
e
s
3

V
a
l
v
e
s
3

V
a
l
v
e
s
1
2

V
a
l
v
e
s
1
2

V
a
l
v
e
s
1
2

V
a
l
v
e
s
Y
a
w
A
n
g
l
e
1
5
1
5
1
5
2
0
-
3
5
2
7
-
4
0
2
7
-
4
0
2
7
-
4
0
N
/
A
N
/
A
N
/
A
I
n
j
e
c
t
i
o
n
A
n
g
l
e
N
/
A
N
/
A
N
/
A
N
/
A
N
/
A
N
/
A
N
/
A
3
0
N
/
A
N
/
A
S
t
e
a
d
y
/
U
n
s
t
e
a
d
y
U

&

S
U

&

S
U

&

S
SSSUUUS
#

o
f
B
l
a
d
e
s
3
6
3
6
3
6
N
/
A
1
4
1
4
1
4
2
3
5
8
N
/
A
A
x
i
a
l
/
C
e
n
t
r
.
AAAAAAACAA
R
e
f
.
3
5
4
0
8
4
3
6
2
4
3
9
4
4
8
3
4
4
3
112
Their results indicate that the yaw angle is a critical parameter that must be
properly tailored to the individual compression system in order to achieve maximum
operating enhancements.
6.1 Computational Modeling of Air Injection Control
The CFD air injection simulations presented in this analysis were performed for
the high-speed compressor as well as the low-speed compressor. The results discussed in
this chapter include steady and unsteady air injection. Due to limitations in
computational resources, unsteady air injection was applied only to the high-speed
compressor, since this type of system is preferred in industrial applications. To achieve
best results, the injection system should be operated in an unsteady fashion, for example,
sinusoidally varying the injection flow rate. The frequency of the injected air flow rate
should be chosen close to the computed stall frequency to effectively cancel out limit
cycles. To prove these theories, results of CFD calculations with varying injection
frequencies are discussed in Section 6.3.
Figure 3.5 shows a schematic of the injectors used in this study. The injectors
were placed a short distance, ten percent of the inlet tip radius, upstream of the
compressor face. The objective of this was to achieve maximum control over the leading
edge flow by varying individual injection parameters. A limited number of preliminary
computations with twice the amount of spacing between the injectors and the compressor
face revealed no significant changes in the control effectiveness of the jets.
113
The injection angle, , was five degrees while the yaw angle, , schematically
shown in Figure 6.1, was parametrically varied. The injection angle was conservatively
chosen based on the experimental injection studies listed in Table 6.1, and with the
intention of generating a high-momentum fluid sheet to prevent the growth of the leading
edge separation near the blade tip. While the preliminary CFD simulations indicated that
the injection angle was also not a critical parameter, the yaw angle had a strong impact on
the effectiveness of compressor control by air injection. Parametric studies were
subsequently performed with varying yaw angles and with varying injection rates to find
the optimum injection configuration that gives maximum operating enhancements with
the least amount of external air supply. Results of steady air injection are discussed in
Section 6.2. Based on conclusions from these results, an unsteady injection scheme was
tested in the high-speed compressor. The results of this analysis are shown in Section
6.3.
6.2 Numerical Results of Steady Air Injection
The information from CFD simulations is extremely valuable in developing
robust and efficient compressor control schemes. In the present work, air injection
upstream of the leading edge was used to alter and energize the leading edge flow in
order to suppress vortex shedding and boundary layer separation that trigger compressor
instabilities. Parametric studies of the injection yaw angle and the injection rates were
performed to determine the configurations that provide the best steady results for both
compression systems studied in this work.
114
Figure 6.2 illustrates the effectiveness of steady air injection in directly altering
the local flow incidence angles as desired. In this figure, using the DLR high-speed
configuration, the injected flow rate was held constant at 3.2 percent of the mean flow
rate and the yaw angle was varied between -15 and 45 degrees relative to the compressor
face. In this analysis, positive yaw angles were measured relative to the compressor face
in opposite directions of rotational shaft speeds. In Figure 6.2, the relationship between
the local incidence angle and the yaw angle was nearly linear. This suggests that the
compressor inflow conditions were significantly affected by the injection scheme. This
gives the designer maximum control over the leading edge flow, an ability that can be
used to suppress separation, and alleviate centrifugal compressor instabilities.
Figure 6.3 and Figure 6.4 demonstrate that a careful choice of injection
parameters and a thorough understanding of their effects on the flowfield are necessary to
effectively apply compressor control. Figure 6.3 shows the transient response of the
DLRCC for various yaw angle settings at 3.2 kg/sec mean mass flow rate. The graph
compares CFD predictions using injection yaw angles of -15 degrees, 7.5 degrees and 45
degrees. Clearly, a yaw angle of -15 degrees triggered leading edge separation and
effectively amplified mass flow fluctuations. An investigation of the flowfield revealed
that this yaw angle caused flow separation of the blade pressure side. Conversely, a yaw
angle of 45 degrees relative to the compressor face produced favorable flow conditions
for the blade pressure side, but caused the boundary layer on the blade suction side to
separate. Figure 6.3 further indicates that for best injection results in the DLRCC, a
slightly positive yaw angle should be chosen. A value of = 7.5 degrees produced
115
favorable compressor inflow conditions on both sides of the blade, without vortex
shedding or boundary layer separation. As a result, the fluctuations in mass flow rate
decay and the compressor returns to a stable operating condition.
Figure 6.4 shows a comparison between DLRCC local incidence angles with and
without steady injection control. The case with injection control was performed at a yaw
angle setting of 7.5 degrees and an injection rate of 3.2 percent. Both CFD calculations
were done at a mean mass flow rate of 3.2 kg/sec and the data was collected over the
duration of one limit cycle. The figure illustrates that if no injection is applied, local
incidence angles near the blade leading edge grow from nearly zero degrees to values of
almost 120 degrees. At approximately 15 degrees incidence, the boundary layer
separates and gives rise to compressor instabilities. The boundary layer breakdown can
be traced by the onset of fluctuations at approximately 25 percent T
stall
. This critical
incidence angle agrees with measured critical angles of attack of common NACA airfoils
that undergo dynamic stall. If a small amount of air is injected into the main flow, as
shown in Figure 6.4, the incidence angle is held constant at small values to produce
favorable flow conditions and avoid boundary layer separation. In this case, instabilities
are successfully suppressed.
Figure 6.5 and Figure 6.6 demonstrate that the above mentioned flow control
mechanism of altering the blade incidence angles by means of upstream jets also works in
the low-speed compressor, LSCC. Figure 6.5 graphically presents the local incidence
angles for varying yaw angles and a fixed injection rate of five percent mass flow. The
data was computed while operating the LSCC at 200 percent rotational shaft speed, and
116
by throttling the compressor mass flow rate to 30 kg/sec. Similar to the high-speed
results discussed in Figure 6.2, a linear relationship occurs between the local flow
incidence angle and the user-defined yaw angle. Since the rotational direction of the
LSCC is opposite to the rotational direction of the DLRCC, the relationship between yaw
angles and incidence angles is reversed in the two configurations. In both systems,
positive yaw angles are measured relative to the compressor face in counter-rotating
directions. Due to significantly lower shaft speeds in the LSCC, large yaw angles of 30
degrees or more are necessary to decrease the flow incidence angle and suppress
boundary layer separation on either side of the blade surface.
Figure 6.6 shows the effect of varying yaw angles on computed flow rate
fluctuations in the LSCC using five percent injection. For this analysis, the compressor
was operated at 200 percent shaft speed with a mass flow rate of 30 kg/sec. The graph
illustrates the sensitivity of user-specified yaw angles on the development of instabilities.
While a yaw angle of 25 degrees caused boundary layer separation on the blade pressure
side, a setting of = 75 degrees led to the breakdown of the suction side boundary layer.
In either case, compressor instabilities were initiated by large incidence angles, resulting
in an amplification of mass flow rate fluctuations. A yaw angle of 45 degrees produced
favorable flow conditions for the boundary layers on both sides of the compressor blades.
As a result, the fluctuations in mass flow rate were decreased and injection compressor
control proved effective.
The observed linear relationship between user-specified yaw angles and resulting
blade incidence angles can be utilized in deriving a generic yaw angle criterion for the
117
suppression of compressor instabilities by air injection based on simple velocity triangles.
Such a criterion could be used as a preliminary tool during the design stage of a
compressor injection system. It assumes that the injectors are located close to the
compressor face, thus effectively neglecting differences in local flow angles. Figure 6.7
is a schematic drawing of the injection scheme and the employed nomenclature
considered in this analysis. The injection velocity component normal to the compressor
casing is v
n
and the two velocity components tangential to the casing are v
t1
and v
t2
. The
latter is oriented parallel to the compressor face. While is the injection angle,
abs
and
rel
are the yaw angles in the absolute and relative frame of reference, respectively. From
the velocity triangle in the absolute reference frame,
rel
can be found as
1 t
rot 2 t
rel
v
v v
tan
+
(6.1)
where v
rot
= r
tip
is the rotational velocity of the compressor. The ratio of both tangential
velocity components is defined in the absolute velocity triangle
1 t
2 t
abs
v
v
tan (6.2)
If the injection rate is given, a relationship for the normal velocity component, v
n
, can be
written
inj
inj
inj
inj
Area Injection
inj
n
A
m
) A (
m
dA
m
v
& & &
(6.3)
118
This equation assumes that the injection velocity is constant over the injection area, A
inj
,
and the density is close to unity. This is a reasonable approximation upstream of the
compressor face in single-stage compressors or in the first stage of a multi-stage system.
From the velocity triangle normal to the compressor casing
tan v v
1 t n
(6.4)
Combining Equation (6.1) through Equation (6.4) yields a relationship for the yaw angle
in the absolute reference frame
inj
inj rot
rel abs
m
tan A v
tan tan
&
(6.5)
All quantities on the right-hand side of this equation are known. The angle
rel
may be
viewed as the desired relative flow angle near the compressor leading edge. In order to
achieve small incidence angles,
rel
should be given a value that is less than the stall
angle, a parameter that is a function of the compressor blade geometry. A yaw angle,
abs
, that produces favorable flow conditions near the blade leading edge can then be
determined from the criterion in Equation (6.5).
Comparing the results of Equation (6.5) to CFD simulations tested the validity of
this derived yaw angle criterion. The results of this analysis are summarized in Table 6.2.
They indicate that Equation (6.5) proved adequate for the high-speed configuration.
While the error between the criterion and CFD simulations for the DLRCC was less than
119
10 percent, the error for the low-speed system was approximately 16 percent. Both
results suggest that the criterion may be useful during preliminary design stages of
compressor injection systems. For more accurate results, three-dimensional CFD
methods, as described in this work, should be utilized.
Table 6.2 Comparison Between Results of Yaw Angle Criterion and CFD Results
Operating Condition Yaw Angle,
abs
, for Zero
Incidence, from Eq. (6.5)
Yaw Angle,
abs
, for Zero
Incidence, from CFD
DLRCC at 3.2 kg/sec
Mass Flow Rate, 3.2%
Injection
27.21 Degrees 25 Degrees
LSCC at 200% Design
Speed, 3.2 kg/sec Mass
Flow Rate, 5.0%
Injection
45.7 Degrees 38 Degrees
The CFD injection results presented in Figure 6.2 through Figure 6.6 show
centrifugal compressor control at selected operating conditions. These results indicate
that the proper choice of individual injection parameters was necessary for best results.
Once a series of operating conditions are tested, a performance map may be constructed
that allows the identification of optimum control states. Current controller technologies
for compressor systems rely on such strategies. A number of operating conditions with
varying parameters can be tested experimentally to program the measured data on a
controller microchip. This controller always knows the optimum operating point and
seeks to drive the system back to this state of greatest efficiency. Due to high costs
associated with the performance of a large number of experiments, the development of
such controller hardware is an expensive methodology. Furthermore, experiments often
120
show insignificant insights into the physics behind a fluid dynamic system. CFD
methods provide a reasonable alternative. With a substantial increase in computational
resources over the past 10 years, a number of CFD simulations with varying operating
conditions may now be performed in relatively short time frames. Therefore, the
construction of computational performance maps leads to significant cost savings and
provides additional insights into the underlying flow phenomena. This strategy was
chosen in the current research.
Summaries of air injection parametric studies for the two centrifugal compressors
considered are shown in Figure 6.8 and Figure 6.9. A total of 60 unsteady CFD
simulations with five injection rates and six yaw angles were performed for the two
configurations. The results of this parametric study are plotted in three-dimensional
graphs where the non-dimensional surge amplitude on the vertical axis was defined as the
ratio of mass flow fluctuations with injection control and the computed mass flow
amplitude without injection control. The computed data for the high-speed compressor
illustrate that there is an optimum injection configuration for which the mass flow
fluctuations are decreased to 14.7 percent of the predicted computed mass flow amplitude
without injection control. This operating condition corresponds to a yaw angle setting
between 5 to 10 degrees and an injection rate of 3.2 percent. Clearly, for departures from
the optimum configuration in both yaw angle and injection rate, the fluctuations grow and
the compressor operation becomes less effective. The growing surge amplitude for
higher injection rates can be explained by the increased ratio of jet velocities to mean
flow velocities. In this case, little momentum transfer took place between injected fluid
121
particles and fluid particles of low momentum near the leading edge boundary layer. As
a result, higher injection ratios were less efficient and should be avoided for best results.
The three-dimensional graph for the low-speed compressor with injection control,
shown in Figure 6.9, indicates that its performance was similar to the high-speed
configuration. For large departures from optimum yaw angles that produce favorable
incidence angles, the predicted surge amplitude increased significantly. Since the low-
speed environment decreased the rotational velocity component, yaw angles between 30
to 40 degrees yielded best injection results. In comparing the high-speed and the low-
speed compressors, it was determined that the low-speed system needed slightly higher
injection rates to produce equivalent stall suppression. At the optimum operating point of
five percent injection and 35 degrees yaw angle, the predicted mass flow fluctuations are
almost nine times lower than in the case without control. In contrast to the high-speed
compressor, the DLRCC injection system also works well for higher injection rates of six
percent and more. Such high injection rates, however, require too much external air
supply and are, therefore, unacceptable.
Although the simulation of 30 operating conditions for each compressor
represents a significant computational task, the injection performance maps graphed in
Figure 6.8 and Figure 6.9 do not contain sufficient data points for use in real compressor
controllers. However, the surge amplitude at any combination of yaw angle and injection
rate can be determined by applying a three-dimensional surface approximation over the
entire domain of interest. Response surface methods, polynomial surface fits or neural
networks are viable mathematical methodologies that are often used to approximate non-
122
linear functions of multiple variables. Figure 6.10 shows the feedforward neural network
used in the present work to model the injection performance maps. The network contains
two hidden layers with 10 neurons and log-sigmoid transfer functions in each layer. By
training the network with the 30 known operating conditions, a reasonable non-linear
model was built that could be programmed into a real controller.
Figure 6.11 and Figure 6.12 show the neural network models of the injection
performance maps for the two compressor configurations used in the present analysis.
Both networks were observed to capture the peaks of the respective surfaces with good
accuracy. Due to large gradients in the LSCC injection performance map at yaw angle
settings of approximately 70 degrees, the neural network predicts some unphysical
oscillations at these locations. These phenomena can be attributed to the lack of data
points. If better accuracy in these regions is desired, more CFD calculations should be
performed and these results should be added to the data set utilized for training of the
neural network.
Besides visual observation of the neural network models to judge agreement with
the original injection performance maps, numerical tests were performed at additional
operating points that were not part of the original data set used for training the neural
networks. At these intermediate points, the non-dimensional surge amplitude was
calculated by applying the unsteady flow solver. Comparisons with values predicted by
the neural network models were then used for validation purposes. A summary of this
analysis is shown in Table 6.3. The results indicate that for both compressors the neural
network models underpredict the CFD computations by less than eight percent.
123
Therefore, the neural network surface fits can be regarded as good approximations of the
CFD computed injection performance maps and may be implemented into a real
hardware microchip for effective compressor control.
Table 6.3 Comparison Between Neural Network Predicted and CFD Predicted Surge
Amplitudes
Operating Condition CFD Predicted Non-
Dimensional Surge
Amplitude
Neural Network Predicted
Non-Dimensional Surge
Amplitude
DLRCC at 3.2 kg/sec
Flow Rate, 3.5% Injection
Rate, 7.5 Degrees Yaw
Angle
55.88% 52.85%
LSCC at 200% Design
Speed, 3.2 kg/sec Flow
Rate, 5.4% Injection Rate,
45 Degrees Yaw Angle
21.25% 19.56%
6.3 Numerical Results of Pulsed Air Injection
The results on steady compressor control discussed in the previous section proved
that air injection is a potentially practical technology for implementation in real engines.
The CFD simulations illustrate the ways flow pattern can be altered by jet actuation, the
optimization of the scheme using parametric studies and ways to build a real compressor
controller from this knowledge.
Table 6.1 shows that several researchers have obtained additional performance
enhancements by going from steady to pulsed air injection. The main advantage of
124
pulsed air injection is the significant reduction in external air supply requirements while
retaining the same favorable flow control characteristics. Secondly, a thorough
knowledge of the system dynamics and a careful choice of the actuation properties
amplitude, frequency and phase angle proved to effectively cancel out unstable modes
and suppress compressor instabilities. The results discussed in this section investigate
these issues by means of unsteady, time-accurate CFD simulations using the high-speed
compression system.
The type of pulsed air injection used in this study was of the form
) t sin( A I
m
) t ( m
inj inj inj
Flow
inj
+
&
&
(6.6)
where
inj
m& , I
inj
and A
inj
are percent fractions of the time-averaged mass flow rate through
the compressor, and
inj
is the user-specified injection frequency. In order to achieve
operating enhancements over steady air injection, the mean injection rate, I
inj
, should be
smaller than the optimum injection rate determined by the parametric studies in Section
6.2. Therefore, mean injection rates of 2.3 percent and 1.5 percent were computationally
tested by CFD methods. The optimum DLRCC yaw angle found from steady injection (
= 7.5 degrees) was used for the pulsed injection. The remaining parameters, injection
angle, injector spacing, injector arrangement and injection area, in the pulsed injection
setup were the same as in the steady injection.
Figure 6.13 shows a comparison between the non-dimensional surge fluctuations,
the local incidence angle and the injection rate in the DLRCC at a mean mass flow rate of
125
3.2 kg/sec and an injection rate of 0.023 + 0.007sin(
stall
t). The non-dimensional surge
amplitude is defined as the percent ratio of the predicted instantaneous mass flow rate and
the mass flow amplitude obtained without air injection control. Thus, a value of 50
percent denotes that the amplitude of the mass flow fluctuations was reduced by one-half
as a result of the pulsed actuation. The injection frequency chosen for this analysis was
stall
, the uncontrolled DLRCC stall frequency determined in Section 5.2.
A comparison between the mass flow fluctuations and the injection rate graphed
in Figure 6.13 reveals that the compressor instabilities decreased by a factor of five as
long as the injection phase angle was lagged 180 degrees behind the flow phase angle.
Due to small shifts of the flow frequency during injection, the injection phase angle had
to be adjusted several times during the calculations in order to remain exactly 180
degrees behind the flow phase angle. After approximately 22 rotor revolutions, the
injection phase was left frozen, thus resulting in increased mass flow fluctuations and
greater incidence angles. A comparison between the mass flow fluctuations and the
incidence angle shows that both signals are in phase. Whenever the mass flow
fluctuations were minimal, the incidence angle was greatest, thus illustrating that much of
the leading edge flow was separated. Injecting the maximum amount of air at this point
caused the incidence angle to decrease and the flow to recover. From this observation, it
can be concluded that during large portions of the limit cycle oscillations, the energizing
effect of the injected fluid was not needed to retain flow stability. Only when the leading
edge boundary layers broke down was a short boost from the injection valves sufficient
for flow recovery. To confirm this conclusion, the analysis from Figure 6.13 was
126
repeated with a smaller mean injection rate I
inj
= 0.015 and a larger amplitude A
inj
=
0.015 percent. The results are shown in Figure 6.14.
The three graphs in Figure 6.14 show a similar behavior compared to the results
obtained with the higher mean injection rate. During the first 13 rotor revolutions, the
phase angle of the pulsed injection was adjusted several times to lag 180 degrees behind
the flow phase angle. As a result, the incidence angles fluctuated between 20 and 40
degrees and the mass flow fluctuations decreased to approximately 20 percent, a
reduction in stall amplitude that is comparable to the results obtained with the higher
mean injection rate discussed in the previous paragraph. Therefore, Figure 6.14 confirms
the conclusion that the amplitude of the pulsed jet has a stronger impact on effective stall
suppression than the mean injection rate.
After these initial 13 rotor revolutions, the injection rate was left frozen, thus
causing the incidence angles to increase to almost 100 degrees and the fluctuations in
mass flow rate to reach values that were predicted during limit cycle oscillations without
injection control. This observation re-emphasizes the importance of phase angle
adjustments, a technique that can be automated in feedback compressor control.
Feedback control schemes measure the growth of stall-precursor signals at a limited
number of compressor locations
100
. This information is then sent to a controller unit that,
in turn, regulates the mean, the amplitude and the phase angle of the pulsed injection.
Figure 6.15 shows a series of Mach contour snapshots at midpassage and at
different instances in time over the period of one limit cycle, T
stall
. The label t = 0 refers
to the instance when the flow rate reached a maximum and the label t = 0.50T
stall
denotes
127
the instance of minimum flow rate. The corresponding injection signal was lagged 180
degrees behind the flow phase angle throughout the entire cycle. This data was computed
at a mean mass flow rate of 3.2 kg/sec and a yaw angle of 7.5 degrees. To illustrate the
effectiveness of the applied compressor control, pulsed air injection of the form 0.015 +
0.015sin(
stall
) was applied to the DLRCC compressor at the location depicted in
snapshot A. Vertical lines in each respective snapshot indicate the positions of the blade
leading edges. Picture A shows the flowfield at the instantaneous point of maximum
flow rate. At this point, the boundary layers remained attached even though no air was
injected. After 25 percent of the cycle (picture B), a small separation region developed
near the tip leading edge since the instantaneous injection rate (0.4 percent) at this point
was not strong enough to energize the fluid particles. Pictures C, D and E show
snapshots of the flowfield at minimum flow rate. These pictures illustrate how the
injected air mixed with low-momentum fluid in the tip recirculation regions, thus causing
the stalled flow to recover. The instantaneous injection rate corresponding to pictures C,
D and E was approximately three percent. The situation shown in picture F indicates that
after 75 percent of the cycle, the entire recirculation region has vanished and the process
is repeated.
The results discussed in Figure 6.13 through Figure 6.15 indicate that for best
injection results, the actuation phase angle should be lagged 180 degrees behind the flow
phase angle. Figure 6.16 through Figure 6.18 illustrate how the injection frequency
inj
should be specified to optimize pulsed injection. These calculations were performed with
injection frequencies of 2
stall
, 2.5
stall
and 4
stall
. The case of
inj
= 2.5
stall
was
128
included to rule out the possibility that only injection frequencies with multiples of the
uncontrolled stall frequency are effective in stabilizing the flow. The results from this
study show that pulsed injection with
inj
= 2
stall
reduces the surge fluctuations below
10 percent with maximum incidence angles of 20 percent. The two cases with higher
injection frequencies
inj
= 2.5
stall
and
inj
=
stall
yield even better results with much of
the non-dimensional surge fluctuations below five percent. This data support the
conclusion stated earlier that a short, temporary boost from the injection valves is
sufficient to stabilize the flow and suppress instabilities. Consequently, higher injection
frequencies should be utilized for best performance.
Figure 6.19 depicts that pulsed injection with a reduced mean injection rate of 1.5
percent also yields excellent results at a high injection frequency of
inj
= 4
stall
. Despite
predicted incidence angles between 50 and 60 degrees in this case, the mass flow
fluctuations remained well below 10 percent. Additional CFD simulations with injection
frequencies above
inj
= 4
stall
were not attempted, however, further improvements in
stall alleviation can be expected for these calculations. This was done with the impetus
that real compressor actuators are limited in bandwidth; therefore, compressor designers
prefer actuation systems that provide stability and robustness at low frequencies.
In an attempt to understand the underlying fluid dynamic phenomena that lead to
effective air injection control in centrifugal compressors, the flowfield near the leading
edge blade tip was investigated using no jets, low-frequency (
inj
=
stall
) and high-
frequency jets (
inj
= 4
stall
). Snapshots of the flowfields shown in Figures 5.4, 5.5, 5.8
and 6.15 illustrate the growth of a leading edge separation region. It is demonstrated that
129
the injected fluid improves the overall flow quality in centrifugal compressors by means
of energizing the low-kinetic energy particles near the blade leading edge, thus
effectively suppressing the growth of this separation region.
Figure 6.20 shows a comparison of vorticity magnitudes at a fixed location
downstream of the leading edge for the three cases considered: no jets, low-frequency
and high-frequency jets. This data was collected over the duration of approximately four
rotor revolutions at midpassage and 99 percent span. The streamwise location of the
probe was 0.025R
inlet
downstream of the leading edge. The graph illustrates that the
vorticity magnitudes without forcing are significantly lower than the computed vorticities
with forcing. The vorticity without forcing peaks at only two points: after approximately
one rotor revolution and after four rotor revolutions. These peaks correspond to
snapshots B and E in Figure 5.5 that depict local flow separation and recovery at the
probe location.
The time-averaged vorticity magnitudes with jet control are almost 250 percent of
the mean magnitudes in the unforced case. This indicates that during most of the sample
period increased amounts of mixing enhance the momentum transfer from the injected
fluid to the low-kinetic energy particles in the separation zone.
Figure 6.21 gives additional insight into the fluid dynamic phenomena that lead to
an enhanced performance of jets pulsed at higher frequencies than low-frequency jets.
The graph shows a comparison of shear stresses caused by jets pulsed at
inj
=
stall
and
jets pulsed at
inj
= 4
stall
. The CFD data was ensemble-averaged over the region
indicated in the schematic. This region extends from the leading edge to approximately
130
0.3R
inlet
downstream, from 85 percent span to the compressor casing and across the entire
azimuthal direction of the flow passage. The results indicate that high-frequency
actuation leads to significantly larger shear stress levels than low-frequency actuation.
When time-averaged over the entire sample period, this increase rises as high as 30
percent. Shear stress peaks at high-frequency forcing are almost twice as large as the
maximum stresses observed at low-frequency pulsing. Additional investigations reveal
that shear stresses in the core of the high-frequency jet are 75 percent larger than in the
core of the low-frequency jet.
These findings suggest that increased fluid dynamic stresses at high-level
excitation produce smaller but intense turbulent eddies. This mechanism enhances the
mixing of small length scales, thus improving the transfer of momentum and energy from
the jet to the low-momentum fluid. As a result, the separation region vanishes and the
onset of compressor instabilities is suppressed.
131
Figure 6.1 Schematic and Nomenclature of Injected Fluid Sheet
Figure 6.2 Local Incidence Angle Versus Yaw Angle in the DLRCC at 3.2 kg/sec Mean
Mass Flow Rate, 3.2% Injection Rate
Yaw Angle
Main Flow
Injected Fluid
Sheet
Compressor
Face
Compressor
Casing
-40
-30
-20
-10
0
10
20
-20 -10 0 10 20 30 40 50
Yaw Angle (Degree)
L
o
c
a
l

I
n
c
i
d
e
n
c
e

A
n
g
l
e

(
D
e
g
r
e
e
)
132
Figure 6.3 Effect of Varying Yaw Angles, , on Flow Rate Fluctuations in the DLRCC at
3.2 kg/sec Mean Mass Flow Rate, 3.2% Injection Rate
Figure 6.4 Effect of Air Injection on Incidence Angles in the DLRCC at 3.2 kg/sec Mean
Mass Flow Rate, 3.2% Injection Rate and 7.5 Degrees Yaw Angle
2
2.5
3
3.5
4
4.5
0 10 20 30 40
M
a
s
s

F
l
o
w

(
k
g
/
s
e
c
)
= 7.5 deg
= 45 deg
= -15 deg
t/2, Rotor Revolutions
-20
0
20
40
60
80
100
120
0% 20% 40% 60% 80% 100%
Time (in Percent of T
stall
)
I
n
c
i
d
e
n
c
e

A
n
g
l
e

(
D
e
g
r
e
e
)
No Injection
3.2% Injection
133
Figure 6.5 Local Incidence Angle Versus Yaw Angle in the LSCC at 3.2 kg/sec Mean
Mass Flow Rate, 200% Design Speed and 5% Injection Rate
Figure 6.6 Effect of Varying Yaw Angles, , on Flow Rate Fluctuations in the LSCC at
3.2 kg/sec Mean Mass Flow Rate, 200% Design Speed and 5% Injection Rate
22
24
26
28
30
32
34
36
38
0 5 10 15
M
a
s
s

F
l
o
w

(
k
g
/
s
e
c
)
= 45 deg
= 75 deg
= 25 deg
-35
-30
-25
-20
-15
-10
-5
0
5
10
15
0 20 40 60 80
Yaw Angle (Degree)
L
o
c
a
l

I
n
c
i
d
e
n
c
e

A
n
g
l
e

(
D
e
g
r
e
e
)
134
Figure 6.7 Schematic and Nomenclature of Velocity Triangles in Simplified Injection
Model
Injection Rates in the DLRCC at 3.2 kg/sec Mean Mass Flow Rate
Top View A-B, Absolute Frame
v
t1
v
n
Rotor
B A
v
t2
v
t1
abs
v
inj,abs
Top View A-B, Relative Frame
v
t2
v
t1
rel
v
inj,rel
v
rot
Optimum:
Nondim. Surge Amplitude = 14.7 %
Injection Rate = 3.2 %
Yaw Angle = 7.5 Degrees
Injection Rate (%)
Yaw Angle (Deg.)
135
Injection Rates in the LSCC at 200% Design Speed, 30 kg/sec Mean Mass Flow Rate
Figure 6.10 Neural Network with Two Hidden Layers and Log-Sigmoid Transfer
Functions
Optimum:
Nondim. Surge Amplitude = 12.5 %
Injection Rate = 5 %
Yaw Angle = 35 Degrees
Injection Rate (%)
Yaw Angle (Deg.)
Input Hidden Layer Hidden Layer Output Layer
Yaw Angle
Injection
Rate
W
+
b
W
b
+
Surge
Amplitude
W
b
+
136
Figure 6.11 Neural Network Surface Approximation of DLRCC Injection Performance
Map
Figure 6.12 Neural Network Surface Approximation of LSCC Injection Performance
Map
Injection Rate (%)
Yaw Angle (Deg.)
Injection Rate (%)
Yaw Angle (Deg.)
137
the DLRCC at 3.2 kg/sec Mean Mass Flow Rate, 7.5 Degrees Yaw Angle and 0.023 +
0.007sin(
stall
t) Pulsed Injection Rate
With Phase Angle
Adjustments
Without Phase
Angle Adjustments
138
0.015sin(
stall
With Phase Angle
Adjustments
Without Phase
Angle Adjustments
139
Figure 6.15 Mach Contour Snapshots at Midpassage in the DLRCC at 3.2 kg/sec Mean
Mass Flow Rate, 7.5 Degrees Yaw Angle and 0.015 + 0.015sin(
stall
t) Pulsed Injection
Rate
A
Injection
B
C D
E F
A: t = 0
B: t = 0.25T
stall
C: t = 0.49T
stall
D: t = 0.50T
stall
E: t = 0.51T
stall
F: t = 0.75T
stall
140
0.007sin(2
stall
141
0.007sin(2.5
stall
142
0.007sin(4
stall
143
0.015sin(4
stall
144
Figure 6.20 Comparison of Vorticity Magnitudes Using No Jets, Low-Frequency Jets
(
inj
=
stall
) and High-Frequency Jets (
inj
= 4
stall
); the Data was Collected at
Midpassage, 99 Percent Span and 0.025R
inlet
Figure 6.21 Comparison of Shear Stresses,
xy
, Using Low-Frequency Jets (
inj
=
stall
)
and High-Frequency Jets (
inj
= 4
stall
); the Data was Ensemble-Averaged over the
Region Indicated in the Schematic
0
100
200
300
400
500
600
0 1 2 3 4 5
V
o
r
t
i
c
i
t
y

M
a
g
n
i
t
u
d
e
Low-Frequency Jets
No Jets
0.0001
0.0002
0.0004
0.0005
0.0007
0.0008
0.001
0 2 3 5
, Rotor Revolution
x
y
High-Frequency Jets
145
CHAPTER VII
7 CONCLUSIONS AND RECOMMENDATIONS
In this research, a numerical technique has been developed for simulating fully
three-dimensional viscous fluid flow in turbomachinery components. The method solves
the unsteady Reynolds averaged Navier-Stokes equations from first principles and
advances the flowfield in a time-accurate manner. The flow equations are solved in a
body-fitted rotating coordinate system using an approximate factorization scheme. To
account for turbulence effects, the Spalart-Allmaras one-equation model is solved in
conjunction with the Navier-Stokes equations. The code is third-order accurate in space
and first-order or second-order accurate in time.
Two test cases, involving different fluid dynamics, were simulated at various
operating conditions. These test cases included a NASA low-speed centrifugal
compressor and a DLR high-speed centrifugal compressor. The flow solver was
validated by comparison against experimental data at operating design conditions.
Additional grid sensitivity studies were performed. A new unsteady downstream
boundary was implemented in the viscous flow solver to model diffuser-plenum
interactions that lead to compressor instabilities at off-design operating conditions. This
boundary condition allowed the simulation of compressor configurations with varying
plenum volumes, backpressures and rotational shaft speeds. A particular combination of
these parameters, known as the B-parameter, was varied and the effects studied.
146
Several limit cycles were simulated using single flow passage grids of the two
compression systems. These simulations revealed the fluid dynamic phenomena leading
to the onset of instabilities in centrifugal compressors.
Air injection upstream of the compressor face was numerically studied and
analyzed as a compressor control scheme. These analyses included air injection in a
steady fashion, and air injection in a pulsed, unsteady pattern. Parametric studies of
varying yaw angles and injection rates were used to develop performance maps for both
compressors in the presence of steady jets. These performance maps served as the
starting point for pulsed jet simulations of varying injection rates and pulse frequencies.
7.1 Conclusions
The investigations in this research lead to the following conclusions:
1. Full Navier-Stokes simulations are a viable means of studying stall and surge in
centrifugal compressors. This approach captures complex flow structures that lead to
the development of fluid dynamic instabilities.
2. It is extremely important to model the effects of the plenum chamber, at least as a
boundary condition, if stall phenomena are to be effectively modeled.
3. The B-parameter coined by Greitzer is a useful way of determining a priori what type
of fluid dynamic instability occurs in a centrifugal compressor. The present studies
confirm that limit cycle oscillations are suppressed below a critical B-value. This
value was found to be configuration dependent.
147
4. Steady jets are effective in controlling compressor instabilities. These jets were
found to alter the local incidence angles and suppress boundary layer separation.
Yawed jets are more effective than parallel jets. An optimum yaw angle exists for
each configuration.
5. Based on the numerical simulations, a criterion was found for the optimum yaw
angle. This criterion is given in Equation (6.5).
6. The behavior of the system under steady jet control can be studied using a neural
network. Once a neural network model is established, it may be utilized
inexpensively to determine optimum yaw angles and jet velocities for new operating
conditions. Such a model, after further studies and consideration of more parameters,
can even be implemented electronically.
7. Pulsed jets are more effective than steady jets for suppressing instabilities. Jets
pulsed at higher harmonics of the stall frequency were more effective than low-
frequency jets. However, due to saturation, there is a practical limitation on the
highest possible frequency.
7.2 Recommendations
The results and conclusions reached from the present work have prompted the
following recommendations for future research in the area of numerical modeling of
instabilities in compression systems:
1. The present research applied air injection upstream of the compressor face as a means
to alleviate instabilities. The air was injected after identifying the leading edge tip
148
and the shroud endwall regions to be the origin of compressor instabilities. As
outlined in Section 1.6, several other control strategies for axial and centrifugal
compressors exist that have proven effective in experiments. These strategies should
be systematically tested and optimized using the present flow solver and the
configurations employed in this research. Control schemes that are aimed at
energizing the low-momentum fluid in the described tip and shroud regions include,
among others, inlet guide vanes, synthetic jets and casing treatments. Numerical
modeling of inlet guide vanes should include the generation of a second, non-rotating
grid that must be coupled with the rotor-grid by a sliding interface. Casing
treatments, a strategy that has worked well in high-speed compressors, should require
only minor modifications of the shroud geometry. Other compressor control schemes
that may be worthwhile exploring are vaned diffusers and diffuser bleed valves.
2. The air injection compressor control scheme employed in the present research can be
categorized as an open-loop control strategy due to the lack of a feedback signal.
Open-loop systems are inefficient because the jet velocity is not linked to the flow
signal, which results in a tendency to use a higher velocity than needed. State-of-the-
art active compressor control utilizes a number of sensors located around the annulus
that detect stall precursors if the operating condition approaches the surge line. This
information is sent to a controller unit, which, in turn, operates the actuators. Such
closed-loop control strategies should be tested with the present flow solver.
Additionally, the effectiveness of various control laws should be systematically
evaluated.
149
3. The present research focused on the detection and analysis of fluid dynamic
phenomena leading to instabilities in a single compressor flow passage. Due to this
limiting assumption, the particular type of instability could not be determined.
Utilizing CFD simulations of multiple flow passages, flow phenomena leading to
rotating stall and appropriate control strategies could be analyzed. Such research is
currently in progress at Georgia Institute of Technology using a similar version of the
present flow solver
101
.
4. Experimental evidence exists that shows inflow disturbances contribute to the onset
of stall and surge in real compressors. Such disturbances may appear as local
stagnation pressure deficits and could be modeled using the present flow solver. An
analysis of different kinds of disturbances, their frequencies and magnitudes, and
their effect on stall inception would be a valuable contribution to the current
compressor control technology.
5. Current generation of turbomachinery flow solvers suffer from a lack of adequate
turbulence and transition models. With the advent of faster and more powerful
computers, large eddy simulations (LES) may soon be possible for compressors. This
will lead to improved flow models and increased accuracy in modeling compressor
instabilities.
It is hoped that this work will serve as a useful stepping stone for pursuing the
above recommendations.
150
REFERENCES
1. Yoshinaka, T., Surge Responsibility and Range Characteristics of Centrifugal
Compressors, Tokyo Joint Gas Turbine, 1977
2. Ribi, B., Centrifugal Compressors in the Unstable Flow Regime, Ph.D. Thesis,
Eidgenssische Technische Hochschule Zrich, 1996
3. Block, H. P., A Practical Guide to Compressor Technology, First Edition,
McGraw-Hill, New York, 1996
4. Emmons, H. W., Pearsen, C. E. and Grant, H. P., Compressor Surge and Stall
Propagation, Trans. ASME, Vol. 27, pp. 455-469, 1955
5. Saxer-Felici, H. M., Saxer, A. P., Inderbitzin, A. and Gyarmathy, G., Numerical
and Experimental Study of Rotating Stall in an Axial Compressor Stage, AIAA
Paper 98-3298, July 1998
6. Day, I. J., Stall Inception in Axial Flow Compressors, J. Turbomachinery, Vol.
115, pp. 1-9, Jan. 1993
7. Copenhaver, W. W. and Okiishi, T.H., Rotating Stall Performance and
Recoverability of a High-Speed 10-Stage Axial Flow Compressor, J. Propulsion
and Power, Vol. 9, pp. 281-292, Mar.-Apr. 1993
8. Lawless, P. B. and Fleeter, S., Active Control of Rotating Stall in a Low-Speed
Centrifugal Compressor, J. Propulsion and Power, Vol. 15, pp. 38-44, Jan.-Feb.
1999
151
9. de Jager, B., Rotating Stall and Surge Control, Proceedings of the 34
th
Conference on Decision & Control, New Orleans, LA, Dec. 1995
10. Wernet, M. P. and Bright, M. M., Dissection of Surge in a High-Speed Centrifugal
Compressor Using Digital PIV, AIAA Paper 99-0270, 1999
11. Greitzer, E. M., The Stability of Pumping Systems, J. Fluids Engineering,
Vol.103, pp. 193-242, June 1981
12. Greitzer, E. M., Review Axial Compressor Stall Phenomena, J. Fluids
Engineering, Vol. 102, pp. 134-151, June 1980
13. Greitzer, E. M., Surge and Rotating Stall in Axial Flow Compressors, Part I:
Theoretical Compression System Model, J. Engineering for Power, Vol. 98, pp.
190-198, Apr. 1976
14. Greitzer, E. M., Surge and Rotating Stall in Axial Flow Compressors, Part II:
Experimental Results and Comparison with Theory, J. Engineering for Power,
Vol. 98, pp. 199-217, Apr. 1976
15. Fink, D. A., Cumpsty, N. A. and Greitzer, E. M., Surge Dynamics in a Free-Spool
Centrifugal Compressor System, ASME Paper 91-GT-31, 1991
16. Botros, K. K. and Henderson, J. F., Developments in Centrifugal Compressor
Surge Control A Technology Assessment, J. Turbomachinery, Vol. 116, pp. 240-
249, Apr. 1994
17. Willems, F. and de Jager, B., Modeling and Control of Compressor Flow
Instabilities, IEEE Control Systems, Vol. 19, pp. 8-18, 1999
152
18. Khalid, S., A Practical Compressor Casing Treatment, ASME Paper 97-GT-375,
Apr. 1997
19. Lee, N. K. W. and Greitzer, E. M., Effects of Endwall Suction and Blowing on
Compressor Stability Enhancement, J. Turbomachinery, Vol. 112, pp. 133-144,
Jan. 1990
20. Bierbaum, K., Operation of a Centrifugal Compressor with Casing Treatments,
Ph.D. Thesis, Universitt Hannover, 1988
21. Holman, F. F. and Kidwell, J. R., Effects of Casing Treatments on a Small,
Transonic Axial-Flow Compressor, ASME Paper 75-GT-5, 1975
22. Dussourd, J. L., Pfannebecker, G. W. and Singhania, S. K., An Experimental
Investigation of the Control of Surge in Radial Compressor Using Close Coupled
Resistances, J. Fluids Engineering, Vol. 99, pp. 64-75, Mar. 1977
23. Whitfield, A., Wallace, F. J. and Atkey, R. C., The Effect of Variable Geometry on
the Operating Range and Surge Margin of a Centrifugal Compressor, ASME Paper
76-GT-98, Jan. 1976
24. Yeung, S. and Murray, R. M., Reduction of Bleed Valve Requirements for Control
of Rotating Stall Using Continuous Air Injection, Proceedings of the 1997 IEEE
International Conference on Control Applications, Hartford, CT, Oct. 1997
25. Pinsley, J. E., Guenette, G. R., Epstein, A. H. and Greitzer, E. M., Active
Stabilization of Centrifugal Compressor Surge, J. Turbomachinery, Vol. 113, pp.
723-732, Nov. 1991
153
26. Harris, L. P. and Spang III, H. A., Compressor Modeling and Active Control of
Stall/Surge, Proceedings of the 1991 American Control Conference, Vol. 3, pp.
2392-2397, June 1991
27. Leonessa, A., Haddad, W. M. and Li, H., Global Stabilization of Centrifugal
Compressors via Stability-Based Switching Controller, Proceedings of the 1999
IEEE International Conference Control Applications (CCA) and IEEE
International Symposium on Computer Aided Control System Design (CACSD),
Vol. 2, pp 1383-1388, Aug. 1999
28. Gysling, D.L., Dugundji, J., Greitzer, E. M. and Epstein, A. H., Dynamic Control
of Centrifugal Compressor Surge Using Tailored Structures, ASME Paper 90-GT-
122, June 1990
29. Arnulfi, G. L., Giamattasio, P., Massado, A. F., Micheli, D. and Pinamonti, P.,
Multistage Centrifugal Compressor Surge Analysis: Part II Numerical
Simulation and Dynamic Control Parameters Evaluation, J. Turbomachinery, Vol.
121, pp. 312-320, Feb. 1999
30. Ffowcs Williams, J. E. and Huang, X. Y., Active Stabilization of Compressor
Surge, J. Fluid Mechanics, Vol. 204, pp. 245-262, July 1989
31. Paduano, J. D., Epstein, A. H., Valvani, L., Langley, J. P., Greitzer, E. M. and
Guenette, G. R., Active Control of Rotating Stall in a Low-Speed Axial
Compressor, J. Turbomachinery, Vol. 115, pp. 48-56, Jan. 1993
154
32. Eisenlohr, G. and Chladek, H., Thermal Tip Clearance Control for Centrifugal
Compressor of an APU Engine, J. Turbomachinery, Vol. 116, pp. 629-634, Oct.
1994
33. Breuer, K., Schmidt, M. and Epstein, A.H., Active Control of Air-Breathing
Propulsion Using MEMS, MIT Year-End Report, July 1998
34. Day, I. J., Active Suppression of Rotating Stall and Surge in Axial Compressors,
J. Turbomachinery, Vol. 115, pp. 40-47, Jan. 1993
35. Weigl, H. J., Paduano, J. D., Frechette, L. G., Epstein, A. H., Greitzer, E. M.,
Bright, M. M. and Strazisar, A. J., Active Stabilization of Rotating Stall and Surge
in a Transonic Single Stage Axial Compressor, ASME Paper 97-GT-411, June
1997
36. Giffin, R. G. and Smith Jr., L. H., Experimental Evaluation of Outer Case Blowing
or Bleeding of Single Stage Axial Flow Compressor Part I, NASA Contract
Report CR-54587, Apr. 1966
37. Goto, A., Suppression of Mixed-Flow Pump Instability and Surge by the Active
Alteration of Impeller Secondary Flows, J. Turbomachinery, Vol. 116, pp. 621-
628, Oct. 1994
38. Weigl, H. J., Paduano, J. D., and E. M., Bright, Application of H
:
-Control With
Eigenvalue Perturbations to Stabilize a Transonic Compressor, Proceedings of the
1997 IEEE International Conference on Control Application, Hartford, CT, Oct.
1997
155
39. Behnken, R. L., Leung, H. and Murray, R. M., Characterizing the Effects of Air
Injection on Compressor Performance for Use in Active Control of Rotating Stall,
ASME Paper 97-GT-316, June 1997
40. Spakovszky, Z. S., Weigl, H. J., Paduano, J. D., von Schalkwyk, C. M., Suder, K.
L. and Bright, M. M., Rotating Stall in a High-Speed Stage With Inlet Distortion,
Part I Radial Distortion, ASME Paper 98-GT-264, June 1998
41. Yeung, S., Nonlinear Control of Rotating Stall and Surge With Axisymmetric
Bleed and Air Injection on Axial Flow Compressors, Ph.D. Thesis, California
Institute of Technology, Aug. 1998
42. Behnken, R. L. and Murray, R. M., Combined Air Injection Control of Rotating
Stall and Bleed Valve Control of Surge, Proceedings of the American Control
Conference, Albuquerque, NM, June 1997
43. Freeman, C., Wilson, A. G., Day, I. J. and Swinbanks, M. A., Experiments in
Active Control of Stall on an Aeroengine Gas Turbine, ASME Paper 97-GT-280,
June 1997
44. DAndrea, R., Behnken, R. L. and Murray, R. M., Rotating Stall Control of an
Axial Flow Compressor Using Pulsed Air Injection, J. Turbomachinery, Vol. 119,
pp. 742-752, Oct. 1997
45. Simon, J.S., Valavani, L., Epstein, A.H. and Greitzer, E.M., Evaluation of
Approaches to Active Compressor Surge Stabilization, J. Turbomachinery, Vol.
115, pp. 57-67, Oct. 1993
156
46. Hendrichs, G. J. and Gysling, D. L., Theoretical Study of Sensor-Actuator
Schemes for Rotating Stall Control, J. Propulsion and Power, Vol. 10, pp. 101-
109, 1994
47. Moore, F. K. and Greitzer, E. M., Theory of Post-Stall Transients in Axial
Compressor Systems: Part I - Development of Equations, ASME Paper 85-GT-
171, 1985
48. Moore, F. K. and Greitzer, E. M., Theory of Post-Stall Transients in Axial
Compressor Systems: Part II - Application, ASME Paper 85-GT-172, 1985
49. Gravdahl, J. T. and Egeland, O., Moore-Greitzer Axial Compressor Model with
Spool Dynamics, Proceedings of the IEEE Conference on Decision and Control,
Vol. 5, pp. 4714-4719, 1997
50. Longley, J.P., Review of Nonsteady Flow Models for Compressor Stability, J.
Turbomachinery, Vol. 116, pp 202-215, Apr 1994
51. Lakshminarayana, B., An Assessment of Computational Fluid Dynamic
Techniques in the Analysis and Design of Turbomachinery, J. Fluids Engineering,
Vol. 113, pp. 315-352, Sept. 1991
52. Chima, R. V. and Yokota, J. W., Numerical Analysis of Three-Dimensional
Viscous Internal Flow, AIAA Journal, Vol. 28, pp. 798-806, May 1990
53. Hall, E. J., Aerodynamic Modeling of Multistage Compressor Flowfields - Part I:
Analysis of Rotor/Stator/Rotor Aerodynamic Interaction, ASME Paper 97-GT-344,
June 1997
157
54. Dawes, W. N., Numerical Study of the 3D Flowfield in a Transonic Compressor
Rotor With a Modeling of the Tip Clearance Flow, AGARD Conference
Proceedings n401, Neuilly Sur Seine, Fr, Mar. 1987
55. Hah, C. and Wennerstrom, A. J., Three-Dimensional Flowfields Inside a Transonic
Compressor With Swept Blades, J. Turbomachinery, Vol. 113, pp. 241-251, Apr.
1991
56. Adamczyk, J. J., Mulac, R. A. and Celestina, M. L., Model for Closing the Inviscid
Form of the Average Passage Equations, J. Turbomachinery, Vol. 108, pp. 180-
186, Oct. 1986
57. Casartelli, E., Saxer, A. P. and Gyarmathy, G., Performance Analysis in a
Subsonic Radial Diffuser, ASME Paper 98-GT-153, June 1998
58. Casartelli, E., Saxer, A. P. and Gyarmathy, G., Numerical Flow Analysis in a
Subsonic Vaned Radial Diffuser With Leading Edge Redesign, ASME Paper 97-
GT-185, June 1997
59. Escuret, J. F. and Garnier, V., Numerical Simulations of Surge and Rotating Stall
in Multi-Stage Axial-Flow Compressors, AIAA Paper 94-3202, July 1994
60. Rivera, C. J., Numerical Simulation of Dynamic Stall Phenomena in Axial Flow
Compressor Blade Rows, Ph.D. Thesis, Georgia Institute of Technology, Aug.
1998
61. Neuhoff, H. G. and Grahl, K. G., Numerical Simulation of Rotating Stall in Axial
Compressor Blade Rows and Stages, ASME Paper 86-GT-27
158
62. Takata, H. and Nagano, S., Nonlinear Analysis of Rotating Stall, ASME Paper
72-GT-3, 1972
63. He, L., Computational Study of Rotating Stall Inception in Axial Compressors, J.
Propulsion and Power, Vol. 13, pp. 31-38, 1997
64. Hunziker, R., The Influence of the Diffuser geometry on the Instability Limit of
Centrifugal Compressors, Ph.D. Thesis, Eidgenssische Technische Hochschule
Zrich, 1993
65. Gridgen User Manual, Version 12, Pointwise, Inc., Bedford, TX, 1997
66. Spalart, P.R and Allmaras, S. R., A One-Equation Turbulence Model for
Aerodynamic Flows, AIAA Paper 92-0439, Jan. 1992
67. Halstead, D.E., Wisler, D.C., Okiishi, T.H., Walker, G.J., Hodson, H.P. and Shin,
H.-W., Boundary Layer Development in Axial Compressors and Turbines, J.
Turbomachinery, Vol. 119, pp. 426-444, July 1997
68. Anderson, J. D., Jr., Modern Compressible Flow, Second Edition, McGraw-Hill,
New York, 1990
69. Anderson, D. A., Tannehill, J. C. and Pletcher, R. H., Computational Fluid
Mechanics and Heat Transfer, Second Edition, McGraw-Hill, New York, 1984
70. Hirsch, C., Numerical Computation of Internal and External Flows, Vol. I and II,
First Edition, Wiley, New York, 1988
71. Schlichting, H., Boundary Layer Theory, Seventh Edition, McGraw-Hill, New
York, 1979
159
72. Roe, P. L., Approximate Riemann Solvers, Parameter Vectors, and Difference
Schemes, J. of Computational Physics, Vol. 135, pp. 250-258, Aug. 1997
73. Roe, P. L. and Pike, J., Efficient Construction and Utilization of Approximate
Riemann Solutions, Computing Methods in Applied Sciences and Engineering,
Vol. 6, Elsevier Science Publishers, INRIA, pp. 499-518, 1984
74. Roe, P. L., Discrete Models for the Numerical Analysis of Time-Dependent
Multidimensional Gas Dynamics, NASA Contract Report CR-172574, Mar. 1985
75. Roe, P. L., Some Contributions to the Modeling of Discontinuous Flows,
Lectures in Applied Mathematics, Vol. 22, pp. 163-193, 1985
76. Roe, P. L., Characteristic-Based Schemes for the Euler Equations, Annual Review
of Fluid Mechanics, Vol. 18, pp. 337-365, 1986
77. Roe, P. L., Some Contributions to the Modeling of Discontinuous Flows, Large-
Scale Computations in Fluid Mechanics, Edited by B. E. Engquist, S. Osher and R.
C. J. Somerville, Vol. 22, Pt. 2, Lectures in Applied Mathematics, ASME,
Providence, RI, pp. 163-193, 1985
78. Liu, Y. and Vinokur, M., Upwind Algorithms for General Thermo-Chemical
Nonequilibrium Flows, AIAA Paper 89-0201, Jan. 1989
79. Pulliam, H. P. and Steger, J. L., Implicit Finite-Difference Simulations of Three-
Dimensional Compressible Flow, AIAA Journal, Vol. 18, pp. 159-167, Feb. 1980
80. Beam, R. M. and Warming, R. F., An Implicit Finite-Difference Algorithm for
Hyperbolic Systems in Conservation-Law-Form, J. Computational Physics,
Vol.22, pp. 87-110, 1976
160
81. Warming, R. F., Beam, R. M. and Hyett, B. J., Diagonalization and Simultaneous
Symmetrization of the Gas-Dynamic Matrices, Math. Comp., Vol. 29, pp. 1037-
1045, 1975
82. Turkel, E., Symmetrization of the Fluid Dynamic Matrices with Applications,
Math. Comp., Vol. 27, pp. 729-736, 1973
83. Ekaterinaris. J. A. and Menter, F. R., Computation of Separated and Unsteady
Flows With One- and Two-Equation Turbulence Models, AIAA Paper 94-0190,
Jan. 1994
84. Private Conversation with M. M. Bright during Georgia Institute of Technology
Briefing with Researchers from NASA Glenn Research Center, Feb. 2000
85. Hathaway, M. D., Chriss, R. M., Wood, J. R. and Strazisar, A. J., Laser
Anemometer Measurements of the 3-D Rotor Flow Field in the NASA Low-Speed
Centrifugal Compressor, NASA TP-3527, June 1995
86. Hathaway, M. D., Criss, R. M., Wood, J. R. and Strazisar, A. J., Experimental and
Computational Investigation of the NASA Low-Speed Centrifugal Compressor
Flow Field, J. Turbomachinery, Vol. 115, pp. 527-542, July 1993
87. Hathaway, M. D., Wood, J. R. and Wasserbauer, C. A., NASA Low-Speed
Centrifugal Compressor for Three-Dimensional Viscous Code Assessment and
Fundamental Flow Physics Research, J. Turbomachinery, Vol. 114, pp. 295-303,
Apr. 1992
161
88. Criss, R. M., Hathaway, M. D. and Wood, J. R., Experimental and Computational
Results from the NASA Low-Speed Centrifugal Impeller at Design and Part Flow
Conditions, ASME-Paper 94-GT-213, June 1994
89. Hathaway, M. D. and Wood, J. R., Application of a Multi-Block CFD Code to
Investigate the Impact of Geometry Modeling on Centrifugal Compressor Flow
Field Predictions, J. Turbomachinery, Vol. 119, pp. 820-830, Oct. 1997
90. Moore, J. and Moore, J. G., A Prediction of 3-D Viscous Flow and Performance of
the NASA-Low-Speed Centrifugal Compressor, ASME-Paper 90-GT-234, June
1990
91. Tweedt, D. L., Chima, R. V. and Turkel, E., Preconditioning for Numerical
Simulation of Low Mach Number Three-Dimensional Viscous Turbomachinery
Flows, NASA TM 113120, Oct. 1997
92. Wood, J. R., Adam, P. W. and Buggele, A. E., NASA Low-Speed Centrifugal
Compressor for Fundamental Research, NASA TM 83398, June 1983
93. Krain, H., A CAD-Method for Centrifugal Impellers, J. Engineering for Power,
Vol. 106, pp. 482-488, 1984
94. Krain, H. and Hoffmann, W., Centrifugal Impeller Geometry and its Influence on
Secondary Flows, AGARD CP 469, Feb. 1990
95. Krain, H. and Hoffmann, W., Verification of an Impeller Design by Laser
Measurements and 3D-Viscous Flow Calculations, ASME-Paper 89-GT-159, June
1989
162
96. Clayton, R. P., Leong, W. U. A. and Sanation, R., A Numerical Study of the
Three-Dimensional Turbulent Flow in the Impeller of a High-Speed Centrifugal
Compressor, ASME Paper 98-GT-49, 1998
97. Krain, H., Swirling Impeller Flow, J. Turbomachinery, Vol. 110, pp. 122-128,
Jan. 1988
98. Krain, H., Test Case 2, Unpublished Results, 1998
99. Eckardt, D., Detailed Flow Investigations Within a High-Speed Centrifugal
Compressor Impeller, J. Fluids Engineering, Vol. 98, pp. 390-402, 1976
100. Prasad, J. V. R., Neumeier, Y. and Krichene, A., Active Control of Compressor
Surge Using a Real Time Observer, Proceedings of the NATO Applied Vehicle
Technology Conference, Braunschweig, Germany, May 2000
101. Niazi, S., Stein, A. and Sankar, L. N., Numerical Studies of Stall and Surge
Alleviation in a High-Speed Transonic Fan Rotor, AIAA Paper 2000-0225, Jan.
2000
163
VITA
Alexander Stein, son of Lothar and Christel Stein, was born in Offenbach,
Germany on January 10, 1969.
During high school, he participated in summer exchange programs in America
and England. Mr. Stein graduated with the Master of Science in Mechanical Engineering
from Ohio University in 1993. In 1995, he received the degree of Diplom-Ingenieur
Maschinenbau (equivalent to a Master of Science in Mechanical Engineering) from
Technical University Darmstadt in Germany. After graduation, Mr. Stein worked for one
year with a volunteer organization in the United States focusing on Christian education.
He joined the School of Aerospace Engineering at Georgia Institute of
Technology as a Graduate Research Assistant in January 1997. He presented scientific
papers at several American Institute of Aeronautics and Astronautics (AIAA)
conferences. He is a member of AIAA. Mr. Stein was published in the Journal of
Propulsion and Power.
In November 1999, Mr. Stein married the former Jane Carole Meredith in Atlanta,
Georgia.

Alex Diss PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Alex Diss PDF

Uploaded by

Copyright:

Available Formats

COMPUTATIONAL ANALYSIS OF STALL AND SEPARATION CONTROL

transformed Jacobian matrices

positive and negative transformed Jacobian matrices

transformed inviscid fluxes

transformed viscous fluxes

diagonal matrix in Roes flux splitting scheme

were set to unity.

are the approximate Jacobian matrices according to

. This matrix has a set of eigenvalues with distinct eigenvectors, a property

is a diagonal matrix with the

CFD Fine Grid

CFD Fine Grid

CFD Fine Grid

CFD Fine Grid

CFD Fine Grid

CFD Fine Grid

You might also like