Aungier - 2003 - Axial-Flow Compressors A Strategy For Aerodynamic

AXIAL-FLOW
COMPRESSORS
A STRATEGY FOR AERODYNAMIC DESIGN AND ANALYSIS
Ronald H. Aungier
NEW YORK ASME PRESS 2003
Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

© 2003 by The American Society of Mechanical Engineers
Three Park Avenue, New York, NY 10016
ISBN: 0-7918-0192-6
Co-published in the UK by Professional Engineering Publishing Limited,

Northgate Avenue, Bury St Edmunds, Suffolk, IP32 6BW, UK
ISBN: 1-86058-422-5
All rights reserved. Printed in the United States of America. Except as permitted under the
United States Copyright Act of 1976, no part of this publication may be reproduced or dis-
tributed in any form or by any means, or stored in a database or retrieval system, without
the prior written permission of the publisher.
Statement from By-Laws: The Society shall not be responsible for statements or opinions
advanced in papers . . . or printed in its publications (B7.1.3)
INFORMATION CONTAINED IN THIS WORK HAS BEEN OBTAINED BY THE AMERI-

CAN SOCIETY OF MECHANICAL ENGINEERS FROM SOURCES BELIEVED TO BE
RELIABLE. HOWEVER, NEITHER ASME NOR ITS AUTHORS OR EDITORS GUARAN-
TEE THE ACCURACY OR COMPLETENESS OF ANY INFORMATION PUBLISHED IN
THIS WORK. NEITHER ASME NOR ITS AUTHORS AND EDITORS SHALL BE RESPON-
SIBLE FOR ANY ERRORS, OMISSIONS, OR DAMAGES ARISING OUT OF THE USE OF
THIS INFORMATION. THE WORK IS PUBLISHED WITH THE UNDERSTANDING THAT
ASME AND ITS AUTHORS AND EDITORS ARE SUPPLYING INFORMATION BUT ARE
NOT ATTEMPTING TO RENDER ENGINEERING OR OTHER PROFESSIONAL SER-
VICES. IF SUCH ENGINEERING OR PROFESSIONAL SERVICES ARE REQUIRED,
THE ASSISTANCE OF AN APPROPRIATE PROFESSIONAL SHOULD BE SOUGHT.
For authorization to photocopy material for internal or personal use under circumstances
not falling within the fair use provisions of the Copyright Act, contact the Copyright Clear-
ance Center (CCC), 222 Rosewood Drive, Danvers, MA 01923, Tel: 978-750-8400,
www.copyright.com.
Library of Congress Cataloging-in-Publication Data
Aungier, Ronald H.
Axial-Flow compressors : a strategy for aerodynamic design and analysis / Ronald
Aungier.
p. cm.
ISBN 0-7918-0192-6
1. Compressors—Aerodynamics. 2. Compressors—Design and construction. I. Title.
TJ267.5.C5 A94 2003

621.5’1—dc21 20020385720
Cover photo: Courtesy of Elliott Turbomachinery Co., Inc., Ebara Group

To Anne

TABLE OF CONTENTS
Preface xi
1 Introduction 1
1.1 Axial-Flow Compressor Basics .........................................................3
1.2 Basic Velocity Diagrams for a Stage................................................5
1.3 Similitude and Performance Characteristics...................................7
1.4 Stage Matching and Stability........................................................11
1.5 Dimensionless Parameters.............................................................13
1.6 Units and Conventions ..................................................................14
2 Thermodynamics 17
2.1 First and Second Laws of Thermodynamics................................18
2.2 Efficiency.......................................................................................20
2.3 Fluid Equation-of-State Fundamentals.......................................22
2.4 The Caloric Equation of State .....................................................24
2.5 Entropy and the Speed of Sound................................................25
2.6 The Thermal Equation of State for Real Gases ..........................26
2.7 Thermodynamic Properties of Real Gases ..................................30
2.8 Thermally and Calorically Perfect Gases .....................................31
2.9 The Pseudo-Perfect Gas Model ...................................................32
2.10 Component Performance Parameters ........................................33
2.11 Gas Viscosity .................................................................................37
2.12 A Computerized Equation-of-State Package .............................37

vi • Table of Contents
3 Fluid Mechanics 41
3.1 Flow in a Rotating Coordinate System .........................................43
3.2 Adiabatic Inviscid Compressible Flow...........................................46
3.3 Adiabatic Inviscid Compressible Flow Applications .....................48
3.4 Boundary Layer Analysis................................................................50
3.5 Two-Dimensional Boundary Layer Analysis..................................51
3.6 Axisymmetric Three-Dimensional Boundary Layer Analysis........54
3.7 Vector Operators in Natural Coordinates.....................................57
4 Axial-Flow Compressor Blade Profiles 59

4.1 Cascade Nomenclature ................................................................60
4.2 NACA 65-Series Profile.................................................................62
4.3 Circular-Arc Camberline...............................................................65
4.4 Parabolic-Arc Camberline ............................................................66
4.5 British C.4 Profile..........................................................................68
4.6 Double-Circular-Arc Profile..........................................................69
4.7 NACA A4K6 63-Series Guide Vane Profile ...................................70
4.8 Controlled–Diffusion Airfoils.......................................................71
4.9 Blade Throat Opening .................................................................73
4.10 Staggered Blade Geometry .........................................................75
5 Two-Dimensional Blade-to-Blade Flow Through

Cascades of Blades 77
5.1 The Blade-to-Blade Flow Problem ................................................79
5.2 Coordinate System and Velocity Components .............................81
5.3 Potential Flow in the Blade-to-Blade Plane .................................83
5.4 Linearized Potential Flow Analysis ...............................................92
5.5 The Time-Marching Method .........................................................96
5.6 Blade Surface Boundary Layer Analysis......................................107
5.7 Summary.......................................................................................113
6 Empirical Performance Models Based On

Two-Dimensional Cascade Tests 117
6.1 Cascade Geometry and Performance Parameters....................119
6.2 Design Angle of Attack or Incidence Angle .............................121

Table of Contents • vii
6.3 Design Deviation Angle.............................................................125

6.4 Design Loss Coefficient and Diffusion Factors .........................128
6.5 Positive and Negative Stall Incidence Angles...........................134
6.6 Mach Number Effects.................................................................136
6.7 Shock Wave Loss for Supersonic Cascades................................138
6.8 Off-Design Cascade Performance Correlations ........................141
6.9 Blade Tip Clearance Loss............................................................146
6.10 Shroud Seal Leakage Loss..........................................................147
6.11 Implementation, Extensions and Alternate Methods .............149
7 Meridional Through-Flow Analysis 153

7.1 Meridional Coordinate System ...................................................154
7.2 Inviscid, Adiabatic Flow on a Quasi-Normal...............................157
7.3 Linking Quasi-Normals ................................................................162
7.4 Repositioning the Stream Surfaces.............................................164
7.5 Full Normal Equilibrium Solution ...............................................165
7.6 Simplified Forms of the Through-Flow Analysis ........................167
7.7 Annulus Sizing .............................................................................169
7.8 Numerical Approximations .........................................................171
8 End-Wall Boundary Layer Analysis 175

8.1 Historical Development of End-Wall Boundary Layer Theory .....177
8.2 The End-Wall Boundary Layer Equations ...................................181
8.3 The Boundary Layer Velocity Profile Assumptions ....................183
8.4 Empirical Models for Entrainment and Wall Shear Stress .........184
8.5 The Blade Force Defect Thicknesses ...........................................187
8.6 Seal Leakage Effects for Shrouded Blades .................................191
8.7 Boundary Layer Jump Conditions ...............................................193
8.8 Solution Procedure ......................................................................194
8.9 Typical Results ..............................................................................195
9 Aerodynamic Performance Analysis 199

9.1 Geometry Considerations............................................................200

viii • Table of Contents
9.2 Cascade Performance Considerations.........................................203

9.3 Stall and Compressor Surge Considerations...............................204
9.4 Approximate Normal Equilibrium Results..................................207
9.5 Full Normal Equilibrium Results..................................................211
9.6 Concluding Remarks ....................................................................213
10 Compressor Stage Aerodynamic Design 215

10.1 Dimensionless Performance Parameters.................................217
10.2 Application to Stage Design....................................................219
10.3 Blade Design.............................................................................221
10.4 Selecting the Stage Performance Parameters ........................222
10.5 Selecting the Swirl Vortex Type...............................................229
10.6 Free Vortex Flow ......................................................................230
10.7 Constant Reaction Vortex Flow...............................................235
10.8 Constant Swirl and Exponential Vortex Flow.........................242
10.9 Assigned Flow Angle Vortex Flows .........................................245
10.10 Application to a Practical Stage Design .................................245
10.11 A Repeating Stage Axial-Flow Compressor............................251
10.12 A Computerized Stage Design System ...................................257
11 Multistage Axial-Flow Compressor Aerodynamic

Design 259
11.1 The Basic Compressor Design Approach ..................................261
11.2 Aerodynamic Performance Specifications................................262
11.3 Blade Design ..............................................................................264
11.4 Refining the Compressor Design ..............................................266
11.5 An Axial-Flow Compressor Design Example.............................268
11.6 The Distribution of Stage Performance Parameters................272
11.7 The Swirl Vortex Type ................................................................280
11.8 Risks and Benefits ......................................................................284
12 Quasi-Three-Dimensional Blade Passage Flow

Field Analysis 287
12.1 Quasi-Three-Dimensional Flow .................................................289
12.2 Hub-to-Shroud Flow Governing Equations ..............................291

Table of Contents • ix
12.3 Numerical Integration of the Governing Equations................294

12.4 Repositioning Stream Surfaces .................................................297
12.5 The Hub-to-Shroud Flow Analysis.............................................298
12.6 Coupling the Two Basic Flow Analyses .....................................299
12.7 Boundary Layer Analysis............................................................302
13 Other Components and Variations 309

13.1 Adjustable Blade Rows ..............................................................311
13.2 The Exhaust Diffuser..................................................................316
13.3 The Scroll or Collector ...............................................................322
13.4 Reynolds Number and Surface Roughness Effects...................328
13.5 The Axial-Centrifugal Compressor............................................328
Answers to the Exercises 333
References 349
About the Author 355
Index 357

PREFACE
Books on compressor aerodynamics approach the subject in various ways

depending on the intended audience. Introductions to the fundamentals are
available to students and newcomers to the field. The experienced aerodynami-
cist can find more comprehensive overviews of the core technologies that review
alternate approaches and summarize important contributions that have signifi-
cantly advanced the state of the art. Users of compressors can find books that
address the alternatives and critical issues of particular interest to the applica-
tion, selection, procurement and operation of compressors. But in recent years,
the aerodynamicist seeking detailed information and practical guidance on the
effective application of basic technology to compressor aerodynamic design and
analysis has been less fortunate.
When I started working in this field, books of that type were available,
although they were somewhat dated. The Centrifugal Compressor Stage (Fergu-
son, 1963), Axial Flow Compressors (Horlock, 1958) and Aero-Thermodynamics
and Flow in Turbomachines (Vavra, 1960) were my particular favorites, and they
provided valuable guidance during my early efforts to formulate compressor
aerodynamic design and analysis systems. Unfortunately, their modern equiva-
lents have simply not appeared. With his usual insight, Cumpsty (1989) recog-
nizes and discusses this change in emphasis in recent books in his preface to
Compressor Aerodynamics. He attributes it to the expanded roles of the computer
and of proprietary industrial research and development, which have combined to
make it rather impractical for authors to provide a general and detailed descrip-
tion of modern compressor aerodynamic design and analysis methods. Several
years ago, similar reasoning led me to consider a different type of book to meet
this need. If a general description of aerodynamic design and analysis methods is
no longer practical, a detailed description of a specific aerodynamic design and
analysis system might be a viable alternative. Although unsure of how it might be
received, I decided to write a book on centrifugal compressor aerodynamics and
managed to find a publisher willing to try something different (Aungier, 2000).
The feedback received since my book was published confirms that there is inter-
est in this type of book as well as a significant audience.
Aerodynamic design and analysis systems evolve through a trial-and-error
process while we painfully learn what capabilities should be included, how to use
them effectively and how they can interact efficiently to support the overall
process. The conversion of basic technology into a working design and analysis

xii • AXIAL-FLOW COMPRESSORS
system is a very complex process involving many decisions and false starts. Along
the way, all developers expend a lot of time and effort formulating virtually iden-
tical models for phenomena fundamental to the process. These critical details are
quite important to aerodynamicists who must develop, maintain or improve an
aerodynamic design and analysis system. The present book approaches axial-
flow compressor aerodynamics in a manner similar to that of my previous book
on centrifugal compressors. The approach used is a description of a comprehen-
sive aerodynamic design and analysis system in sufficient detail so that readers
can readily implement the complete system or any of its components.
This proved to be more difficult for axial-flow compressors than was the case
for centrifugal compressors. Centrifugal compressor stage designs tend to be
fairly unique and specific to the design objectives. Most current design and analy-
sis systems share many common features and are used in a wide range of appli-
cations. The basis for axial-flow compressor design is far more varied and
application-dependent, often based on proprietary information that is not avail-
able in the open literature. Indeed, it is now quite common to find designs based
on proprietary and customized airfoil families, such as the popular controlled
diffusion airfoil. The geometry and performance characteristics of these propri-
etary airfoils are well known only to the organizations that developed them.
There are also inherent differences in the technology used on industrial axial-
flow and centrifugal compressors. The development of a completely original
industrial axial-flow compressor design is relatively rare. These compressors are
almost always unique, one-of-a-kind designs that must rely on variations of stan-
dard components to minimize risks while maintaining acceptable development
and manufacturing costs. Hence, the variety of application experience is far more
limited for the axial-flow compressor design and analysis system than was the
case for the centrifugal compressor system.
I decided that the desired objectives could still be achieved by adopting the
classical design approach based on the systematic application of standard airfoil
families to develop the blade geometry used in the compressors. The basic prin-
ciples of the design process described here remain applicable when proprietary
airfoil families are in use. But it is likely to be necessary to adapt them to reflect
the specific geometry and performance characteristics of those airfoils. Aerody-
namic performance prediction accuracy is established by comparing predictions
with experimental data for several axial-flow compressors. That established per-
formance prediction accuracy is then used to demonstrate the effectiveness of
the overall design and analysis system. A substantial number of design examples
are included to illustrate the use of this design and analysis system, as well as to
provide some evaluation of alternate design approaches suggested in the litera-
ture, or which I have found to be effective.
Considerable care is taken to provide complete and detailed descriptions of
this comprehensive aerodynamic design and analysis system for axial-flow com-
pressors. The basic principles of thermodynamics and fluid mechanics required
are presented in a form particularly well-suited to the axial-flow compressor
application. Well-defined empirical models are used to augment these basic prin-
ciples to address the essential problem areas of performance analysis, stage
design, compressor design and internal flow analysis. Descriptions of numerical
methods used are included as well as other critical considerations important to

Preface • xiii
readers who may wish to apply these methods. In a few cases where components
are common to both my centrifugal and axial-flow compressor design and analy-
sis systems, I refer to Aungier (2000) rather than repeat some rather lengthy and
detailed descriptions that will be of interest only to readers who choose to imple-
ment those specific methods.
Some important topics have received inadequate treatment or have been com-
pletely omitted. Surge and stall are discussed only in the context of estimating
the expected limits of stable operation, while noise and blade vibration are not
discussed at all. I prefer to limit myself to topics on which I can offer at least
some original ideas. I always feel uncomfortable when presenting ideas obtained
almost entirely from others, even when I have considerable confidence in the
sources. I much prefer that readers obtain such information from qualified
authors, even if it does display my own limitations.

Chapter 1
INTRODUCTION
Compressors are commonly classified as either positive displacement or dynamic

compressors. The positive displacement compressor achieves its pressure rise by
trapping fluid in a confined space and transporting it to the region of higher pres-
sure. The dynamic compressor develops its increase in pressure by a dynamic
transfer of energy to a continuously flowing fluid stream. There are two basic
types of dynamic compressors: axial-flow compressors and centrifugal (radial-
flow) compressors. The flow streamlines through rotating rows in an axial-flow
compressor have a radius that is almost constant, whereas they undergo a sub-
stantial increase in radius in a centrifugal compressor. For this reason, the cen-
trifugal compressor can achieve a much greater pressure ratio per stage than the
axial-flow compressor. But the axial-flow compressor can achieve a significantly
greater mass flow rate per unit frontal area. Figure 1-1 compares normalized dis-
charge pressure, P, versus flow rate, Q, for these two compressor types to illus-
trate the differences in their performance characteristics. The axial-flow
compressor approximates a variable pressure ratio—constant flow machine,
whereas the centrifugal compressor is closer to a constant pressure ratio—vari-
able flow machine. The performance data displayed in Fig. 1-1 are for a single-
stage centrifugal compressor and a five-stage axial-flow compressor, both of
which have about the same design pressure ratio. This demonstrates the superior
pressure ratio-per-stage capability of the centrifugal compressor. Traditionally,
the centrifugal compressor has been the more rugged and lower-cost type, while
the axial-flow compressor has offered better efficiency. Those differences have
become much less significant in recent years due to advances in technology, par-
ticularly with regard to efficiency. Presently, the compressor type selected is more
likely to be based on the performance characteristics, size and cost that is best
suited to the application.
NOMENCLATURE
a = sound speed
C = absolute velocity
cp = specific heat at constant pressure
cv = specific heat at constant volume
H = total enthalpy and compressor head

2 • AXIAL-FLOW COMPRESSORS
h = static enthalpy
k = ratio of specific heats = cp / cv
M = Mach number = C / a
ṁ = mass flow rate
N = rotation speed (rpm)
P = pressure
Q = volume flow rate = ṁ / ρt
R = Gas constant and stage reaction
Re = Reynolds number
r = radius
T = temperature
U = local blade speed, ωr
W = relative velocity
β = flow angle
γ = stagger angle
δ = Pt0 / Pref
η = efficiency
θ = polar (tangential) coordinate and Tt0 / Tref
µ = viscosity
ρ = density
φ = stage flow coefficient
ψ = stage work coefficient
ω = rotation speed (radians/second)
FIGURE 1-1 Performance of Dynamic Compressors

Introduction • 3
Subscripts
d = parameter at compressor discharge

ref = reference thermodynamic conditions
rev = reversible thermodynamic process
t = total thermodynamic condition
z = axial component
θ = swirl (tangential) component
0 = parameter at guide vane inlet or compressor inlet
1 = parameter at guide vane exit and rotor inlet
2 = parameter at rotor exit and stator inlet
3 = parameter at stator exit
Superscripts
′ = relative condition
1.1 AXIAL-FLOW COMPRESSOR BASICS

Figure 1-2 illustrates the basic configuration of an axial-flow compressor. The first
blade row shown is an inlet guide vane (IGV) to develop the swirl (tangential)
velocity for which the first rotating row (R1) was designed. If the first rotating row
is designed for no inlet swirl, the inlet guide vane will normally be omitted. This is
followed by a series of stages (two in this illustration), where a stage refers to a
rotating row, or rotor, in combination with its downstream stationary row, or sta-
tor (e.g., R1 and S1). The rotor row imparts energy to the fluid by increasing the
swirl velocity. The stator row removes the swirl developed by the rotor to convert
kinetic energy to static pressure and to establish the proper swirl velocity for the
flow to enter the next rotor. Typically, an exit guide vane (EGV) follows the last
stage to remove any residual swirl velocity to convert that kinetic energy to static
pressure. Like the inlet guide vane, this row may be omitted if the last stator is
designed to remove all of the swirl velocity. Although not shown on the figure, a
diffuser-combustor (gas turbine) or diffuser-collector (industrial compressor) will
follow the exit guide vane to recover as much kinetic energy as possible, as well as
to direct the flow to its intended destination. Similarly, an inlet passage will pre-
cede the inlet guide vane. This can range from a smooth axial bell-mouth inlet to
a complex side inlet, depending on the compressor’s application.
Figure 1-3 illustrates the blade profiles for a stage viewed on a polar stream
surface between adjacent blades. The rotor row is rotating with a velocity, U = ωr,
where ω is the rotation speed and r is the radius. Viewed in a frame of reference
rotating with the rotor, the upstream velocity, W, is referred to as the relative
velocity. The rotor deflects the flow such that the velocity in the stationary frame
of reference of the stator (the absolute velocity), C, is properly aligned to enter
the stator row. This process repeats in subsequent stages, with each stage adding
energy to the fluid to achieve the overall pressure ratio required.
Axial-flow compressor design strategies are quite varied. Gas turbine com-
pressors are normally intended for use in many identical units. Extensive design

FIGURE 1-2 Axial-Flow Compressor Configuration
FIGURE 1-3 Polar Surface View of a Stage

Introduction • 5
and development programs are common and may include unique blade designs
for all blade rows. By contrast, industrial compressors are usually designed
specifically for a unique application, rarely involving any duplicate units. Here, a
repeating stage approach is more common, where one or more basic stage
designs are used for all compressors. Aerodynamic and mechanical flexibility are
obtained by minor adjustments that do not compromise the basic stage’s per-
formance. Blades may be scaled to longer and thicker blades for mechanical
integrity with a corresponding adjustment of the number of blades per row to
preserve aerodynamic similarity. Blades usually must be restaggered, i.e., rotated
on their base to change the stagger angle, γ, to achieve different performance lev-
els. Otherwise, the intended duty would normally require a non-integer number
of stages. Often the inlet guide vane and some of the stator blades may be
adjustable so they can be restaggered by a control system while the machine is in
operation to broaden the compressor’s application range. This approach allows
these “one-of-a-kind” compressors to be designed within practical cost. It also
allows each compressor’s design to be based on a well-established performance
history. This is important, since these compressors cannot be confirmed by per-
formance testing until after they are manufactured.
Figure 1-2 illustrates normal cantilevered blades that are attached at the root,
with a clearance between the blade tip and the adjacent end-wall. Figure 1-4
shows a different style often used for stator blades. Here, a shroud band is
attached to the blade tips to connect them together. This is often done for reasons
of mechanical integrity. To reduce fluid leakage from the blade discharge back to
the blade inlet, seal fins are normally attached to the shroud band. These provide
a reduced clearance to retard leakage, yet are thin enough to minimize damage in
the event that a rotor shaft excursion or “rub” causes the seals and shaft to come
into contact. To minimize damage to the shaft, the stator blades and stator
shrouds, the seal fins will be sacrificed in the event of a rub.
1.2 BASIC VELOCITY DIAGRAMS FOR A STAGE

The construction of velocity diagrams is a very useful concept for axial-flow com-
pressor design. Here they will be used to illustrate the velocity vectors entering
and leaving blade rows in a stage. It will be necessary to use both absolute and
relative velocities, where relative velocities are viewed in a frame of reference
rotating with the compressor’s rotation speed, ω. Designating the relative and
absolute tangential velocities as Wθ and Cθ, respectively, the two are related by
Wθ = Cθ − ω r (1-1)
where r is the local radius. The axial components of velocity are identical in both
frames of reference, i.e.,
Wz = Cz (1-2)
Therefore, the absolute and relative velocities are

FIGURE 1-4 Shrouded Stator Blades
C = Cz2 + Cθ2
(1-3)
W = Cz2 + Wθ2
(1-4)
The absolute and relative flow angles are designated as β and β′, respectively.
They are defined by
tan β = Cθ / Cz (1-5)
tan β ′ = Wθ / Cz
(1-6)
Figure 1-5 illustrates the velocity diagrams for an inlet guide vane. The flow
enters the guide vane with no swirl, i.e., Cθ 0 = 0, C0 = Cz0. The guide vane deflects
the flow by an angle, β1. If Cz1 is known, this defines the swirl velocity compo-
nent, Cθ1. Then Eq. (1-1) is applied in vector form to subtract ωr from Cθ1 to
define the swirl velocity component in the relative frame, Wθ1 and the relative
flow angle, β′1. Hence, the complete velocity diagram for the entrance to the
downstream (rotating) rotor blade row is known. Figure 1-6 shows the velocity
diagram construction for the rotor blade row. The inlet velocity diagram is the
same as that determined for the guide vane exit. The rotor blade deflects the flow
in the relative frame of reference from β′1 to β′2 to produce the discharge swirl
velocity, Wθ2. If Cz2 is known, Wθ 2 can be computed. Then vector addition of ωr to

Introduction • 7
FIGURE 1-5 Guide Vane Velocity Diagrams
Wθ2 yields the absolute swirl component Cθ 2. Hence, all velocity components and
the flow angle in the absolute frame of reference can be computed to define the
inlet conditions for the downstream (stationary) stator blade row. Construction
of the stator and exit guide vane blade row velocity diagrams is accomplished in
a similar fashion and will be left as an exercise for the reader. The important
thing to note is that construction of this simple velocity diagram is a fundamen-
tal technique commonly used by turbomachinery aerodynamicists to convert
between absolute and relative flow conditions. Here, Cz has been treated as
known. In practice, values of Cz may be specified design conditions from which
the flow passage areas will be computed to conserve mass. This will be referred to
as the design mode. Alternatively, Cz may be computed from basic mass and
momentum conservation for specified passage areas and the mass flow rate. This
will be referred to as the analysis mode.
1.3 SIMILITUDE AND PERFORMANCE CHARACTERISTICS
Similitude or similarity is one of the most useful concepts in turbomachinery

aerodynamics. Two turbomachines are completely similar if the ratios of all

FIGURE 1-6 Rotor Blade Velocity Diagrams
corresponding length dimensions, velocity components and forces are equal

(Sheppard, 1956). If two turbomachines are completely similar, it is possible to
present their performance on a common performance map by selecting appropri-
ate equivalent performance parameters. Equivalent performance requires that the
two compressors have similar velocity diagrams throughout. To maintain similar
velocity diagrams while conserving the mass flow rate, ṁ, the ratio of gas density-
to-inlet density must also be similar throughout. This means that inlet volume
flow, Q0, is the relevant equivalent flow rate parameter, where
˙ / ρt
Q=m (1-7)
The local axial velocity is actually given by
˙ / ( A0ρ0 )
Cz0 = m (1-8)
where A is the passage area and ρ0 is the inlet gas density, which is unknown.
But the exercises in Chapter 2 will show that the ratio ρ0 / ρt0 is a function of Cz0,

Introduction • 9
where ρt0 is the known inlet total gas density. Hence, unique velocity diagrams
are associated with a unique Q0 / A0, but can correspond to many values of ṁ/A0
simply by altering ρt0. For this reason, all dynamic compressors are commonly
referred to as volume flow machines. The exercises in Chapter 2 will show that
the requirement for density ratio similarity requires that the Mach numbers be
similar throughout, where the Mach number is the ratio of fluid velocity to the
local sound speed, a. It will be shown that the ratio of a0 / at0 is, itself, a function
of the Mach number, so the unknown a0 can be replaced by the known inlet total
sound speed, at0. Figure 1-7 shows an equivalent performance map based on
these requirements. The flow parameter used is volume flow normalized by the
inlet total sound speed, at0, and the inlet area, A0. This ensures that the inlet
axial Mach numbers will be similar. Three performance characteristics, or speed
lines, are shown for three different rotation speeds: N, multiplied by a charac-
teristic diameter, D, and normalized by the inlet total sound speed. This will
ensure similarity of the tangential Mach numbers. If two axial-flow compressors
are geometrically similar, and use the same working fluid, this performance map
will apply to both machines. This, in turn, ensures that the pressure ratios will
be the same for both, so pressure ratio is a reasonable choice for the other per-
formance parameter. The situation becomes more complicated if the two com-
pressors use different working fluids. In that case, complete similarity usually
cannot be achieved, since different working fluids may produce different gas
density (or specific volume) ratios for the same blade row velocity diagram. This
“volume-ratio” effect will compromise similarity after the first rotor row, since
FIGURE 1-7 An Equivalent Performance Map

the differences in density will necessitate differences in axial velocity to conserve

mass. Indeed, even with identical working fluids, non-ideal gas behavior may
compromise similarity. True equivalent performance is assured only for working
fluids that obey the perfect gas equation of state discussed in Chapter 2, which
can be expressed as:
P = ρRT (1-9)
where P is pressure, T is temperature and R is the gas constant. An exercise in

Chapter 2 will also show that achieving equivalent density ratios requires that the
two working fluids have the same ratio of specific heats, k = cp / cv. Since axial-
flow compressors are widely used in aircraft engine gas turbines, discussions on
this topic often overlook these more subtle effects. That application of compres-
sors deals exclusively with air as the working fluid, which is very nearly a perfect
gas, offering little chance of any volume-ratio effect. Consequently, equivalent
performance maps must be used with caution when a perfect gas model with a
constant value of k cannot approximate the working fluid.
Similarity is also compromised when the two compressors operate at sub-
stantially different Reynolds number, Re. Reynolds number is a measure of the
inertia forces to viscous forces, Re = ρCL/µ, where L is a characteristic length
and µ is the gas viscosity. The Reynolds number directly affects wall friction,
which can alter the compressor’s performance. In most cases, effects of the
Reynolds number are small enough to be neglected. When that is not the case,
suitable Reynolds number corrections (e.g., Wassell, 1968) may be applied to
adjust the performance.
A common use of equivalent performance maps is to define performance of a
specific compressor that is operated at various inlet total thermodynamic condi-
tions. For that case, D and A0 are constant and can be omitted on the map.
Another common use is to relate performance of different compressor frame
sizes, derived by directly scaling the geometry. Then, D and A0 are included so the
map defines the change in speed, N, needed to preserve Mach number equiva-
lence and the flow rate supplied for each pressure ratio on a speed line.
The equivalent flow rate used in Fig. 1-7 is the true similarity parameter. When
the application is to a perfect gas with constant R and k, Eq. (1-9) can be used to
derive alternate equivalent flow parameters, i.e.,
Q / at0 ∝ m ˙ θ /δ
˙ Tt0 / Pt0 ∝ m (1-10)
where the sound speed has been replaced by a perfect gas relation from chapter 2,
a = kRT (1-11)
and θ and δ relate inlet total conditions to reference conditions (Tref, Pref), such as
standard atmosphere conditions, i.e.,
θ = Tt0 / Tref (1-12)

δ = Pt0 / Pref (1-13)

Introduction • 11
Similarly, the equivalent speed parameter can be replaced, using
N / at0 ∝ N / Tt0 ∝ N / θ (1-14)
The alternate flow rate and speed parameters in Eqs. (1-10) and (1-14) are com-
monly used, but have less fundamental significance than Q0 / at0 and N / at0.
Similarly, compressor head, ∆Hrev, can be used in place of pressure ratio,
where head is defined as the total enthalpy increase required to produce the
actual pressure rise by an ideal, reversible process, i.e.,
dp
∆Hrev = ∫ ρ
(1-15)
rev
It can be shown that the appropriate equivalent head parameter is ∆Hrev / (at0)2.
The use of an equivalent head is common practice for centrifugal compressors,
but is much less common for axial-flow compressors.
Figure 1-7 supplies only part of the performance information required. In
addition to the pressure ratio and flow produced, it is necessary to know the
work required to drive the compressor. Hence, a second equivalent perform-
ance map is required to completely define the compressor’s performance. The
most common parameter for this purpose is efficiency, η, defined as the com-
pressor head or ideal (no loss) total enthalpy rise divided by the actual total
enthalpy rise, i.e.,
η = ∆Hrev / ∆H (1-16)
Alternate reversible processes that can be used to define ∆Hrev and η are dis-
cussed in Chapter 2. Figure 1-8 shows an equivalent efficiency map to be used in
conjunction with Fig. 1-7. In some cases, it may be appropriate to use the exit
static thermodynamic conditions rather than total values to define η and PR. This
is appropriate when the kinetic energy available at the compressor discharge
serves no useful purpose for the specific application to which the compressor will
be applied.
1.4 STAGE MATCHING AND STABILITY
Each blade row in a compressor will achieve its best performance for a specific
inlet flow angle where losses are minimum. Basically, the designer seeks to
“match” succeeding blade rows such that all operate close to their optimum inlet
flow angles at a specific operating condition, commonly called the compressor’s
design point or match point, defined by the design flow rate and design speed.
Hence, at design speed, losses can be expected to increase and performance to
deteriorate as the compressor operates farther from its design flow rate. At flow
rates less than the design flow rate, losses will increase to a point that the pres-
sure-flow rate characteristic reaches a maximum. At lower flow rates, the char-
acteristic will have a positive slope, which is theoretically unstable. The onset of

FIGURE 1-8 An Equivalent Efficiency Map
this severe unstable operation is commonly called surge. The limit of stable oper-
ation is referred to as the surge line as illustrated in Fig. 1-7. Surge is a very com-
plex phenomenon, which depends on the entire system, not just the compressor.
So associating it with a maximum on the pressure–flow rate characteristic is an
oversimplification, but a useful one. In some cases, an approach to zero-slope
near surge is evident, such as for speed line N3 in Fig. 1-7. In other cases, the
compressor may experience an abrupt stall, such that the characteristic appears
to be quite steep at surge, similar to speed line N1 in Fig. 1-7. This is mainly
because the drop in pressure with reduced flow is so abrupt that it cannot be
resolved in a performance test. Indeed, estimation of the onset of surge during
the design phase is based more on the expected blade loading limits at the onset
of stall than on the predicted shape of the pressure-flow characteristics. Simi-
larly, at flow rates greater than the design flow rate, the increase in loss will even-
tually result in no rise in pressure. This condition is commonly referred to as
choke, although it may be caused by large losses due to off-design operation
rather than a true aerodynamic choke condition.
When the compressor is operated at off-design speeds, operation at different
Mach number levels will compromise the stage matching, similar to the volume-
ratio effect mentioned previously. Consequently, it is unlikely that all stages will
be close to their optimum operating conditions at any flow rate for off-design
speeds. Rather, optimum performance will occur at the flow rate offering the
best compromise on stage matching. Performance will deteriorate for flows dif-
ferent from this optimum, much as described for the design speed performance.

Introduction • 13
The more speed deviates from the design speed, the greater the compromise of
the stage matching. In general, at speeds lower than the design speed, the front
stages are required to supply a greater portion of the rise in pressure while the
rear stages become less effective. The inverse is true for speeds greater than the
design speed. This stage mismatching can be alleviated to some degree if some
of the stationary blade rows are adjustable during operation. Closing some of
the stationary rows (i.e., increasing their stagger angles) in a controlled fashion
will shift the optimum matching condition to lower speeds to reduce the mis-
match at low speeds.
The Mach number level has a definite influence above and beyond its pro-
nounced effect on stage matching. As a blade row’s inlet Mach number increases,
its low-loss operating range will decrease. At sufficiently high values, the blade
row will start to experience aerodynamic choke in the blade row to significantly
reduce its maximum flow capacity. Even the minimum loss levels will increase
when the inlet Mach number becomes high enough to produce shock waves that
are strong enough to induce boundary layer separation or to produce significant
bow shock losses. Consequently, as the equivalent speed increases, pressure-flow
characteristics become steeper, with less flow range from surge to choke as illus-
trated in Fig. 1-7, and the maximum achievable efficiency can be limited by Mach
number levels, similar to speed line N1 in Fig. 1-8.
1.5 DIMENSIONLESS PARAMETERS
In addition to the dimensionless parameters associated with similitude, axial-

flow compressor design is often based on a number of other useful aerodynamic
dimensionless parameters. To introduce these parameters, it is necessary to
anticipate some results that are developed in more detail in Chapters 2 and 3. In
particular, the total enthalpy rise for simple axial flow through a rotor blade row
is expressed by the well-known Euler turbine equation, i.e.,
∆H = U (Cθ 2 − Cθ1) (1-17)
where U = ω r is the local blade speed and H is the total enthalpy. Similarly, it will
be shown in Chapter 2 that the static enthalpy, h, is related to H by
H = h + 12 C 2 (1-18)
It is often useful to introduce dimensionless stage performance parameters

expressed for a “repeating” stage, i.e., a stage designed to be followed by another
identical stage. This means that the velocity diagrams for the rotor inlet (station
1) and the stator exit (station 3) must be identical. Then, the stage work coeffi-
cient, ψ, can be defined as
ψ = ∆H / U 2 = (Cθ 2 − Cθ1) / U (1-19)
where all data correspond to a constant, mean radius, or “pitch line” for the
stage. The stage flow coefficient, φ, is defined by

φ = Cz1 / U (1-20)
The stage reaction, R, is defined as the fraction of the stage static enthalpy rise
that occurs in the rotor, i.e.,
R = ( h2 − h1) / ( h3 − h1) (1-21)
Substituting P for h in Eq. (1-21) yields an alternate definition of reaction, in

terms of static pressures, that is sometimes used. Requiring Cz to be constant
through the stage, Eqs. (1-17), (1-18) and (1-21) can be combined to yield
R = 1 − (Cθ 2 + Cθ1) / (2U ) (1-22)
In Chapter 10, it will be seen that parameters φ, ψ and R provide useful guidance
for stage design. Stage design involves defining blade geometry that will produce
the desired performance. These dimensionless performance parameters define
performance in a form general to any stage design problem. They are normally
used to specify the performance objectives the stage should achieve at its mean
radius or pitch line. While there are no fixed rules for selecting values for them,
preferred values can normally be established based on the design goals most
important to the designer, supported to some degree by simple logic. For exam-
ple, 50% reaction stages (R = 0.5) are quite common, prompted mainly by the
intuitive judgment that it is best to share the flow diffusion load equally between
the rotor and the stator. Once specified, these parameters can be used to define
the stage velocity diagrams from which the blades can be designed. For example,
Eqs. (1-19) through (1-22) can be combined to yield
tan β1′ = −(ψ / 2 + R) / φ (1-23)

tan β2′ = (ψ / 2 − R)/φ (1-24)
tan β1 = (1 − R − ψ / 2)/φ (1-25)
tan β2 = (1 − R + ψ / 2)/φ (1-26)
and the velocity diagrams for the stator exit and rotor inlet are identical for a
repeating stage. These parameters have defined the velocity diagrams at the
pitch line only. It is necessary to supply additional design specifications and use
fundamental fluid dynamics relations to generate the velocity diagrams at other
radial locations.
1.6 UNITS AND CONVENTIONS
This book assumes consistent units throughout, such that the reader may use any
set of consistent units preferred. For historical reasons, many turbomachinery
organizations do not use consistent units, often using different units for different
disciplines such as aerodynamics and thermodynamics. For example, it is not

Introduction • 15
uncommon to find energy terms, fluid velocity and equation-of-state parameters

expressed in inconsistent units, necessitating conversion factors in expressions
such as Eq. (1-18). It will be left to the reader to recognize the need for those con-
version factors. Flow angles and blade angles will be measured from the merid-
ional plane, i.e., a plane of constant polar angle, θ, in a cylindrical coordinate
system. These angles, and the associated swirl velocity components Cθ and Wθ,
are considered to be positive in the direction of rotation. While the nomenclature
is reasonably consistent throughout, the wide range of topics covered does not
permit unique symbols for every parameter. Consequently, each chapter will
include its own list of nomenclature to avoid confusion.
EXERCISES
1.1 An axial-flow compressor is to be operated with a different working

fluid, which can be modeled as a perfect gas, but has values of gas
constant, R, and ratio of specific heats, k, that are different from the
normal working fluid. Develop new equivalent speed and flow rate
parameters, in terms of ṁ, Tt0 and Pt0 that will ensure Mach number
equivalence at the compressor inlet.
1.2 Free vortex stages with β1 = β3 = 0 are often used for axial-flow com-
pressors. Derive an expression for work coefficient as a function of
reaction for this type of stage. If the stage is also to have 50% reaction,
specify the range of values for ψ and β′1 that can be used. If the result-
ing stage is to be used as a repeating stage in a multistage compressor,
what type of inlet and exit guide vanes will be needed?
1.3 All dimensions of the compressor producing the performance map
shown in Fig. 1-7 are scaled by a factor of 1.2 and both compressors are
operated with the same inlet conditions and working fluid. If operating
points for the original compressor are denoted as NA and QA, develop
expressions for equivalent operating conditions NB and QB for the
scaled compressor. If the original compressor operates at a speed of
3,600 rpm, what speed must be used for the scaled compressor? How
much additional flow capacity will the scaled compressor have?
1.4 What scale factor should be applied to the compressor producing the
performance map shown in Fig. 1-7 to increase the compressor’s flow
capacity by 20% for the same inlet conditions and working fluid?
What adjustment in speed will be needed?

Chapter 2
THERMODYNAMICS
This chapter highlights some of the fundamental concepts from thermodynamics

that are essential to the aerodynamic design and analysis of axial-flow compres-
sors. These concepts provide the basis for defining and evaluating the energy
transfer processes and for modeling the state properties of the working fluid.
Unlike the centrifugal compressor, axial-flow compressors are most often applied
to rather ideal working fluids, the most common being air. But the application of
axial-flow compressors to non-ideal working fluids is becoming more common.
Consequently, a basic description of non-ideal gas modeling is included. This
description is sufficient for application of the techniques, but is less detailed than
this author’s previous description for centrifugal compressors (Aungier, 2000).
Readers interested in more detail are referred to that earlier reference or to
Aungier (1995, 1998).
NOMENCLATURE
A = Helmholtz energy, dA = -PdV

a = sound speed and gas constant defined in Eq. (2-32)
b = gas constant defined in Eq. (2-33)
C = velocity
c = gas constant defined in Eq. (2-34)
cp = specific heat at constant pressure and pressure recovery coefficient
cv = specific heat at constant volume
H = total enthalpy
h = enthalpy
k = ratio of specific heats = cp / cv
M = molecular weight
n = exponent in Eq. (2-34), defined in Eq. (2-35)
P = pressure
q = specific heat transfer
q̇ = heat transfer rate
R = gas constant
RU = universal gas constant
s = specific entropy

T = temperature
u = specific internal energy
V = specific volume
w = specific work input
ẇ = power input rate
z = compressibility factor
η = efficiency
µ = gas viscosity
ρ = gas density
ω = acentric factor
ξ = parameter defined in Eq. (2-69)
–
ω = loss coefficient
Subscripts
ad = adiabatic-reversible (isentropic) process

c = critical point parameter
d = discharge condition
i = inlet condition
p = polytropic process
r = reduced parameter (normalized by its critical point value)
ref = reference condition
t = total condition
0 = condition ahead of an inlet guide vane
1 = condition ahead of a rotor blade row
2 = condition following a rotor blade row
Superscripts
0 = condition where the ideal gas model applies
2.1 FIRST AND SECOND LAWS OF THERMODYNAMICS
The first law of thermodynamics covers the basic principle of conservation of

energy. The first law can be applied to compressors with one restriction. Since a
compressor is an open system, steady flow is the only case to which the first law
is applicable. If ẇ is the power input to the compressor and q̇ is the heat transfer
between the compressor and its surroundings, the first law of thermodynamics
can be written
q˙ + w ˙ ∆[u + 1 C 2 + P / ρ]
˙ =m (2-1)
2
where ṁ is the mass flow rate, u is the specific internal energy, C is velocity, P is
pressure, ρ is density and any change in potential energy due to gravitational

Thermodynamics • 19
force is neglected. The term “specific” designates a parameter measured per unit
mass of fluid. The terms in brackets specify the internal energy, kinetic energy
and the flow work, P/ρ. The term flow work refers to the work necessary to move
the fluid across the boundaries of the system. Only when steady flow crosses the
system’s boundaries does P/ρ provide a direct measure of the flow work, which
restricts Eq. (2-1) to the steady flow case. The normal practice in fluid dynamics
applications is to combine the internal energy and flow work terms into a single
parameter called the enthalpy, h.
h = u+ P / ρ (2-2)
The above thermodynamic conditions are simple state variables or static condi-
tions. It is useful to introduce a special class of thermodynamic conditions
referred to as total or stagnation conditions. A total thermodynamic condition is
defined as the value of a parameter that will exist if the fluid is brought to rest
with no transfer of heat or external work, i.e., all kinetic energy is completely
recovered. Total conditions will generally be designated with a subscript, t. Total
enthalpy occurs so often that an exception will be made and H will be used
instead of ht. Total enthalpy is given by
H = h + 12 C 2 (2-3)
Heat transfer can normally be neglected for flow through a compressor. There are
obvious exceptions, such as when a heat exchanger is included in the system. If
heat transfer is neglected, the flow is called adiabatic and Eqs. (2-1) through (2-3)
combine to yield
˙ =m
w ˙ ( Hd − Hi ) (2-4)
where the subscripts i and d refer to the compressor’s inlet and discharge condi-
tions, respectively. Equation (2-4) is the basic energy equation for steady, adia-
batic flow through a compressor.
The second law of thermodynamics introduces the concept of a reversible
process. A process is referred to as reversible if the system and its surroundings
can be returned to their original states after the process has occurred. If that is
not the case, the process is called irreversible. Processes influenced by heat trans-
fer or friction effects are common examples of irreversible processes. The specific
entropy, s, is defined as
dqrev
ds = (2-5)
T
where T is the temperature, q is the specific heat transfer and the subscript, rev,
designates a reversible process. The second law of thermodynamics can be
expressed as

∆s ≥ 0 (2-6)
Hence, an adiabatic, reversible process is also a constant entropy or isentropic

process. Indeed, it is fairly common practice in turbomachinery to use the terms
adiabatic and isentropic interchangeably, although the latter is clearly the more
precise term. Entropy provides a fundamental measure of the irreversibility of a
process, i.e., the inefficiency or losses associated with the process. If the first law
of thermodynamics is applied to a fluid element in a closed system,
( dq)rev = Tds = du + dw = du + PdV (2-7)
where w is the specific work and V = 1/ρ is the specific volume. Equations (2-2)
and (2-7) provide a fundamental thermodynamic equation for entropy that is
valid for any process, i.e.,
Tds = dh − VdP (2-8)
2.2 EFFICIENCY
The aerodynamic quality of a compressor or a component of a compressor is
commonly measured in terms of efficiency, which is a measure of actual per-
formance relative to an ideal performance that would be achieved by some
reversible process. Figure 2-1 illustrates a typical enthalpy-entropy diagram for a
compressor or a portion of a compressor that includes at least one rotor row,
such that work has been done on the fluid. The inlet conditions are designated as
Pti, Tti, si and Hi. The compressor does work on the fluid to produce discharge
conditions Ptd, Ttd, sd and Hd. Note that static and total conditions are, by defini-
tion, related by a reversible process. Hence, there is no difference between total
and static entropy and the subscript, t, can be omitted for s. One measure of effi-
ciency is to compare the actual process to an ideal adiabatic (isentropic) process.
As seen in Fig. 2-1, an isentropic process could produce the change in total pres-
sure with an enthalpy rise ∆Had, commonly referred to as the adiabatic head. The
actual process required an enthalpy rise of ∆H. Since lines of constant pressure
always diverge on an h-s diagram, ∆H is always larger than ∆Had for a non-isen-
tropic process. Hence, the adiabatic or isentropic efficiency, ηad, is defined as
∆Had
ηad = (2-9)
∆H
where ∆Had is given by
Ptd dP
∆Had = ∫ ; (s = const) (2-10)
Pti ρ
This development of efficiency has considered a compressor operating between

inlet and discharge total conditions. Hence it is often called the total-to-total

adiabatic efficiency. It is often the case that the fluid kinetic energy at the com-
pressor discharge serves no useful purpose to a specific application. In that case,
Ptd may be replaced by the static pressure, Pd, in Eq. (2-10) to yield the total-to-
static adiabatic efficiency. The additional substitution of Pi for Pti yields the
static-to-static adiabatic efficiency. Some care is required to understand which
basic definition is being used when interpreting efficiency data. It is not uncom-
mon for the term adiabatic efficiency to be used for any of the above three types
without qualification.
Adiabatic efficiency is the most common definition used for axial-flow com-
pressors. But it has a definite weakness as a means of evaluating the aerody-
namic quality of a design. As illustrated in Fig. 2-1, constant pressure lines
diverge on an h-s diagram. This means that two compressors having the same
basic aerodynamic design quality, but operating at different pressure ratios,
will have different adiabatic efficiencies. Hence, adiabatic efficiency is not par-
ticularly useful to an aerodynamic designer seeking to evaluate the true aero-
dynamic quality of a compressor or a stage design. Another consequence of
this thermodynamic effect is that the adiabatic head of a multistage compres-
sor is not equal to the sum of the stage adiabatic heads. Polytropic efficiency is
a more useful definition, which eliminates this undesirable thermodynamic
effect. Polytropic efficiency is sometimes referred to as the “small-stage” or
“true aerodynamic” efficiency. Instead of using a path of constant entropy as
the reversible path, polytropic efficiency uses a path of constant efficiency
defined by
FIGURE 2-1 An Enthalpy-Entropy Diagram

1 dP
ηp = (2-11)
ρ dh
where ηp is defined such that the path passes through the two end points of the
process, e.g., (Pti, Hi) and (Ptd, Hd). This is straightforward in principle, but less so
in practice. For many years, common practice was to approximate a polytrope by
a path defined by P/ρ e = constant. This approximation is appropriate for ideal
gases, but can introduce large errors for non-ideal gases. Models that use poly-
tropic efficiency with non-ideal gases have been reported by Shultz (1962) and
Mallen and Saville (1977). Huntington (1985) studied the problem in detail and
showed that the Mallen-Saville model yields excellent accuracy. Huntington pro-
posed a slightly better model by including an intermediate point on the path, but,
for convenient use, this had the disadvantage of requiring a numerical method.
Hence, this writer employs the Mallen-Saville model in all cases. This model uses
an empirical path defined by
ds
T = constant (2-12)
dT
Equations (2-8) and (2-12) can be combined to yield the polytropic head, ∆Hp, as
∆H p = ∆H − ( sd − si )(Ttd − Tti ) / ln(Ttd / Tti ) (2-13)
to provide the total-to-total polytropic efficiency, ηp, by
∆H p
ηp = (2-14)
∆H
Extension of Eqs. (2-13) and (2-14) to total-to-static and static-to-static efficiency

is analogous to adiabatic efficiency as previously discussed. Basically, it is now
possible to employ polytropic efficiency with no more difficulty than adiabatic
efficiency. It is only necessary to have an appropriate equation of state to use
either model.
2.3 FLUID EQUATION-OF-STATE FUNDAMENTALS
Thermodynamics contributes one of the fundamental governing equations for

compressor aerodynamic design and analysis, commonly called the equation of
state. To be more precise, there are actually two equations of state required. The
first is the thermal equation of state, which supplies a relationship among the
fundamental state variables, typically in the functional form P = P(ρ, T). The sec-
ond is the caloric equation of state, which relates the energy content of the fluid
to state variables, typically in the functional form h = h(T, P) or u = u(T, P). These
equations of state may be derived from kinetic theory or statistical mechanics, or
they may be developed empirically from experimental data.

Axial-flow compressors are usually applied to rather simple fluids, the most
common being air. Consequently, books on axial-flow compressors usually pro-
vide a very limited discussion of the equation of state applicable only to very ideal
working fluids. But the application of axial-flow compressors to very non-ideal
fluids is becoming more common, so this chapter provides a broader discussion.
A fluid is considered to be an ideal or thermally perfect gas if P, T and ρ can be
related by the simple linear thermal equation of state
P = ρRT (2-15)
where R is a constant for the fluid. The gas constant, R, is related to the universal
gas constant, RU, and the fluid’s molecular weight, M
R = RU / M (2-16)
where RU = 8314 Pa-m3/(kmole-°K) in metric units. All working fluids exhibit

non-ideal behavior under appropriate conditions. Figure 2-2 is a schematic of a
pressure-enthalpy diagram for any working fluid. It is seen that the fluid may be
liquid, vapor or both at various state points. Clearly, it will not be possible for a
thermally perfect gas equation of state to model all possible state points. This is
true even when the equation is restricted to the vapor phase. Figure 2-2 illus-
trates the location of the fluid’s critical point, which is defined as the highest
FIGURE 2-2 A Pressure-Enthalpy Diagram Schematic

temperature at which liquid and vapor can coexist. Experimental measurements

of critical-point properties (Tc, Pc, ρc) are commonly made and almost never
conform to Eq. (2-15). Normally, the thermally perfect gas equation of state can
provide a reasonable approximation if T is much greater than Tc and P is much
less than Pc. At pressures well above Pc, the fluid is said to be in the supercritical
regime. Again, it is known from experiment that supercritical fluids almost
always show non-ideal gas behavior.
Yet the role of the thermally perfect gas equation of state is far more significant
than just providing a useful approximation. Indeed, at sufficiently low values of
density, all fluids follow the perfect gas law. This provides a dramatic simplifica-
tion to the process of modeling non-ideal gases. Under conditions where the gas is
thermally perfect, it can be shown that the energy content of a fluid is independ-
ent of pressure, i.e., h0 = h0(T) and u0 = u0(T), where the superscript, 0, designates
conditions where the fluid is thermally perfect. This means it is always possible to
define the caloric equation of state as a function of only one state variable.
2.4 THE CALORIC EQUATION OF STATE
From the previous discussion, it is seen that the caloric equation of state can be
specified in a general form for use in either an ideal or non-ideal gas model.
Under low-density conditions where the gas is thermally perfect, the specific
heats at constant pressure and at constant volume are defined as
 ∂h0 
c0p (T ) =   (2-17)
 ∂T  P
 ∂u0 
c0v (T ) =   (2-18)
 ∂T  V
For a thermally perfect gas, it can be shown that
c0p (T ) − c0v (T ) = R (2-19)
The caloric equation of state can be specified by supplying either c0p(T) or c0v(T)
and using Eqs. (2-17) through (2-19), i.e.,
T
h0 (T ) = h0 (Tref ) = ∫ c0p (T )dT (2-20)
Tref
T
u0 (T ) = u0 (Tref ) = ∫ c0v (T )dT (2-21)
Tref
where h0 and u0 can be assigned any desired values at a reference state point (Tref,
Pref). Specific heat correlations as a function of temperature are readily available
for most fluids of interest (e.g., Ried, et al., 1977, 1987; and Yaws, 1999). For the

special case where the specific heats are constant, the gas is referred to as a calor-
ically perfect gas. In that case, Eqs. (2-20) and (2-21) can be directly integrated to
yield analytical equations, i.e.,
h0 (T ) = h0 (Tref ) + c0p (T − Tref ) (2-22)

u0 (T ) = u0 (Tref ) + c0v (T − Tref ) (2-23)
Most books on axial-flow compressors limit attention to thermally and calori-

cally perfect gases. This simplifies the writer’s task by reducing some concepts to
simple, analytical forms. This practice is avoided in this book because it results in
some unfortunate consequences. Even experienced turbomachinery engineers
have occasionally applied these simplified concepts beyond their limits of valid-
ity, resulting in some unfortunate conclusions. It has often encouraged develop-
ers of computerized aerodynamic analyses to incorporate special thermodynamic
relations that are limited to this special class of fluids. A numerical analysis
rarely benefits from these simplified methods, and it can be very difficult to gen-
eralize such an analysis later, should that become necessary. It is quite possible to
have hundreds of such relations scattered through an analysis, presenting a for-
midable problem when attempting to generalize it.
2.5 ENTROPY AND THE SPEED OF SOUND

An entropy equation is essential to all aspects of compressor aerodynamic design
and analysis. The efficiency definitions introduced previously in this chapter all
require an entropy equation. Even the simplest aerodynamic analyses require an
entropy equation to convert between total and static thermodynamic conditions.
This conversion is an isentropic process, accomplished by imposing a change in
enthalpy (the kinetic energy) at constant entropy. An entropy equation is also
required to impose a non-isentropic loss mechanism such as a total pressure loss.
It is useful to develop the entropy equation consistent with the caloric equation
of state defined for a thermally perfect gas. This will be needed for the non-ideal
gas model described later in this chapter. For any pressure where the gas is ther-
mally perfect, the specific entropy is given by Eq. (2-8), i.e.,
T
c0p (T )
s0 (T , P ) = s0 (Tref , Pref ) + ∫ T
dT − R ln( P / Pref ) (2-24)
Tref
where s0 can be assigned any desired value at any reference state point (Tref, Pref).
For a calorically perfect gas, this simplifies to
s0 (T , P ) = s0 (Tref , Pref ) + c0p ln(T / Tref ) − R ln( P / Pref ) (2-25)
The speed of sound is also an essential parameter governing the performance

of a compressor. From fundamental thermodynamics, the speed of sound, a, for
any gas is given by

 ∂P   ∂P 
a 2 =   = k  (2-26)
 ∂ρ  s  ∂ρ  T
where k is the ratio of specific heats.
k = cp / cv
(2-27)
For a thermally perfect gas, this yields
a0 = kRT (2-28)
2.6 THE THERMAL EQUATION OF STATE FOR REAL GASES
The general thermal equation of state for a real gas is
P / ( ρRT ) = z(T , P ) (2-29)
where z is the compressibility factor. This equation applies to any fluid, with z = 1
for the special case of a thermally perfect gas. The compressibility factor can be
obtained from generalized tabular data (e.g., Nelson and Obert, 1954; and Pitzer
et al., 1955). But that is rarely done today since many excellent real gas equations
of state are available. The real gas equation of state directly provides the many
other thermodynamic parameters required and yields much better computational
speed when used in numerical methods. The simple two-parameter equations-of-
state are a good choice for general aerodynamic design and analysis. They offer
good accuracy, excellent computational speed and easy access to the required gas
property data for almost any working fluid or fluid mixture. The Redlich-Kwong
equation (Redlich and Kwong, 1949) and various modifications to it (Aungier,
1994, 1995; Barnes, 1973; Soave, 1972; and Wilson, 1966) are recognized as being
among the most accurate of the two-parameter equations. Aungier (1994, 1995)
evaluated these five equations of state in considerable detail for twelve different
compounds over a wide range of temperatures and pressures. Specific emphasis
was placed on covering a wide range of accentric factors, ω, since that parameter
is used by the various modified Redlich-Kwong equations to improve the predic-
tion accuracy. Accentric factor (Pitzer et al., 1955) is defined as
ω = − log10 ( Pv / Pc ) − 1 ; T / Tc = 0.7 (2-30)
where Pv is pressure on the vapor saturation line and the subscript c designates a
critical point property (Fig. 2-2). Table 2-1 lists the compounds investigated,
together with their accentric factors. Based on this evaluation, it was concluded
that only the original Redlich-Kwong equation and Aungier’s modified Redlich-
Kwong are suitable for general turbomachinery aerodynamic design and analy-
sis. The modified equations of Barnes, Soave and Wilson all showed improved

Table 2-1 List of Compounds
Compound ω
Ammonia 0.2550
Carbon Dioxide 0.2250
Ethylene 0.0868
Helium –0.464
Hydrogen –0.220
I-Butane 0.1848
Methane 0.0080
N-Pentane 0.2539
Nitrogen 0.0400
Propane 0.1520
Refrigerant R134a 0.3254
Steam 0.3440
accuracy over the original equation for the types of compounds and thermody-
namic property ranges for which they were developed. But they were found to
have serious deficiencies for ω < 0, and they exhibited reduced accuracy for
supercritical fluids (i.e., P > Pc, Fig. 2-2). Consequently, only the original Redlich-
Kwong equation and Aungier’s modified form will be described here. The original
Redlich-Kwong equation is
RT a
P= − (2-31)
V − b V (V + b) Tr
where Tr = T / Tc is the reduced temperature, and
a = 0.42747 R2Tc2 / Pc (2-32)

b = 0.08664 RTc / Pc (2-33)
Equations (2-32) and (2-33) are derived from the thermodynamic stability condi-
tion, which requires that the first and second partial derivatives of P with respect
to V must both equal zero at the critical point. For its application, this equation
requires only the critical temperature and critical pressure for the fluid (and the
caloric equation of state data). It offers very good accuracy over a wide range of
thermodynamic conditions. It does have one well-known deficiency near the crit-
ical point for the fluid. For any fluid, this equation yields z = 1/3 at the critical
point, which is not typical of most fluids. Basically, this equation should never be
used for points close to the critical point.
Aungier’s modified Redlich-Kwong equation of state is
RT a
P= − (2-34)
V − b + c V (V + b)Trn
where a and b are given by Eqs. (2-32) and (2-33), respectively, n is given by

n = 0.4986 + 1.1735ω + 0.4754ω 2 (2-35)
and the constant, c, is calculated directly by applying Eq. (2-34) at the critical
point with all critical point properties specified. It is included to remove the defi-
ciency in the original equation for points near the critical point. Aungier’s equa-
tion requires two more specified parameters than does the original equation, i.e.,
ω and either Vc or zc. Note that if c = 0 and n = 0.5, Aungier’s equation reduces to
the original equation. If a = b = c = 0, the thermally perfect gas equation of state
is obtained. This makes it very easy to develop a computerized equation-of-state
package for use in any aerodynamic analysis, which can offer all three equations
of state.
Figure 2-3 illustrates the benefits of Aungier’s modified Redlich-Kwong equa-
tion over the original form. This figure shows a comparison of the prediction
accuracy of the two equations for about 25% of the data points considered in the
evaluation in Aungier (1994, 1995). Tabular (P, V, T) data from the literature were
used for that purpose. The two equations of state were used to predict P from the
tabular values of T and V, which were compared to the tabular value of P to com-
pute the error. It is seen that Aungier’s model provides a significant improvement
in prediction accuracy for a vast majority of the points considered. Aungier
(1994, 1995) notes that Aungier’s model resulted in about a 50% reduction in the
root-mean-square error for the complete set of data considered. This writer uses
Aungier’s model for all real gas problems, but maintains the original model as a
FIGURE 2-3 State-Point Prediction Accuracy

viable alternative. No equation-of-state is infallible, so it is a good idea to have an

alternative available.
Either of the above equations of state can be used if the accentric factor and
the critical point properties are known. This is a real advantage offered by these
equations, since those data are readily available for virtually any compound
likely to be encountered in turbomachinery applications. Ried and Sherwood
(1966), and Reid et al. (1977, 1987) and Yaws (1999) are good sources for these
gas property data.
Equations (2-31) and (2-34) can also be applied to gas mixtures. If the sub-
script, i, designates the ith compound in a mixture of N compounds and xi is its
mole fraction, the constants for the mixture are given by
N
a = ∑ xi ai (2-36)
i =1
N
b = ∑ xi bi (2-37)
i =1
N
zc = ∑ xi zci (2-38)
i =1
N
ω = ∑ xiω i (2-39)
i =1
N
M = ∑ xi Mi (2-40)
i =1
N
cp = ∑ xi cpi (2-41)
i =1
The gas constant for the mixture is given by R = RU / M. The effective values of Tc
and Pc for the mixture can be computed from a, b and R, using the definitions in
Eqs. (2-32) and (2-33). Then c can be computed from Eq. (2-34). Note that Eq.
(2-41) requires that cp be the specific heat per mole rather than per unit mass. For
the original Redlich-Kwong equation, c = 0, and Eqs. (2-38) and (2-39) are not used.
A mixture equation of state formed in this manner is applicable only to the
vapor phase of the fluid. Although the above real-gas models apply to the more
general two-phase flow problem, the composition of the two phases may be dif-
ferent and vary with state-point conditions, requiring special treatment. Axial-
flow compressors are not intentionally applied to two-phase flows, so the
restriction to the vapor phase is not a concern in this application. Centrifugal
compressors often encounter a special case of two-phase flow when intercoolers
are located between stages. These intercoolers may cause some components to
liquefy and drop out of the mixture. This is commonly referred to as liquid-
knockout, and can be modeled by an equilibrium flash calculation. This is not
commonly needed for axial-flow compressors and will not be discussed in this
book. Should this capability be needed, Aungier (2000) describes the equilib-
rium-flash calculation using the equations of state presented here.

When the real-gas equation of state is restricted to the vapor phase, it is prudent
to avoid accidental excursions into the “wet region” where liquid may form. In a
numerical analysis, this can easily occur due to numerical errors in iterative solu-
tions that are far from convergence, often terminating the analysis with a fatal
error. The vapor saturation pressure, Pv, is known to vary proportional to 1/T (Ried
and Sherwood, 1966). The gas property specifications required for Aungier’s model
include two points on the vapor saturation line shown in Fig. 2-2. One point is the
critical point, while the other follows from the definition of the accentric factor in
Eq. (2-30). Thus, the vapor saturation pressure can be approximated by
log10 ( Pv / Pc ) = 7(1 + ω )(1 − Tc / T ) / 3 (2-42)
This equation can easily be inverted to predict the vapor saturation temperature
as a function of pressure. These relations can be used to limit the relevant inde-
pendent thermodynamic conditions in all calculations intended only for the
vapor phase to prevent fatal errors from these accidental excursions into the wet
region. In some aerodynamic analyses, this simple precaution can yield a dra-
matic improvement in the reliability of the analysis.
2.7 THERMODYNAMIC PROPERTIES OF REAL GASES
Specification of the caloric equation of state, h0 or u0, has been limited to state
points where the fluid is thermally perfect. For non-ideal fluids, h and u are func-
tions of pressure as well as temperature. The calculation of the thermodynamic
properties of a non-ideal fluid is best accomplished using departure functions
(Ried et al., 1977). Departure functions are defined as the difference between the
actual value of a parameter and its value under conditions where the fluid is ther-
mally perfect, e.g., the quantity h – h0 is the enthalpy departure function. To
employ this process, the pressure at which the fluid can be considered to be ther-
mally perfect, P0, must be specified. The precise value is not important, but it will
be a low (but non-zero) value of pressure where the thermally perfect gas approx-
imation can be considered to be valid. Equation (2-15) supplies the correspon-
ding specific volume, i.e., V0 = RT / P0. If A = the Helmholtz energy, Ried et al.
(1977) shows that the relevant departure functions are
V
A − A0 = − ∫ ( P − RT / V )dV − RT ln(V / V 0 ) (2-43)
∞
 ∂( A − A0 ) 
s − s0 = −   (2-44)
 ∂T V
h − h0 = ( A − A0 ) + T ( s − s0 ) + RT ( z − 1) (2-45)
0 0 0
u − u = ( A − A ) + T (s − s ) (2-46)
For the Redlich-Kwong equation of state, the departure functions are

a V + b
h − h0 = PV − RT − ( n + 1)Tr− n ln   (2-47)
b  V 
 V V − b + c  na − n  V + b 
s − s0 = R ln  0  − bT Tr ln  V  (2-48)
V V   
where c = 0 and n = 0.5 for the original Redlich-Kwong equation of state. Basic
thermodynamics provides the other parameters commonly required for aerody-
namic analysis, i.e.,
 ∂h 
cp =   (2-49)
 ∂T  P
2
 ∂u    ∂z    
cv =   = cp − R  z + T 
 ∂T  V 
 
 ∂T  P  / z − P ∂∂Pz  
T

(2-50)
 ∂P  kzRT
a 2 = k  =
 ∂ρ  T P  ∂z  (2-51)
1−  
z  ∂P  T
2.8 THERMALLY AND CALORICALLY PERFECT GASES
When the fluid can be considered thermally perfect (z = 1) and calorically perfect
(cp, cv, and k are constants), equation-of-state calculations are greatly simplified.
If the subscript, ref, designates conditions at an arbitrary reference state point
(Tref, Pref) where h and s can be assigned arbitrary values, Eqs. (2-22) and (2-25)
can be written
h = href + cp (T − Tref )
(2-52)
s = sref + cp ln(T / Tref ) − R ln( P / Pref ) (2-53)
Combining Eqs. (2-3) and (2-52), the total and static temperatures are related by
cp (Tt − T ) = 1 C2
2 (2-54)
For an isentropic or adiabatic-reversible process Eq. (2-53) requires
k −1
T / Tref = ( P / Pref ) k = ( ρ / ρ ref )k −1 (2-55)
Equations (2-54) and (2-55) provide simple analytical expressions to relate total
and static conditions. Similarly, the efficiency calculations in Eqs. (2-9) and
(2-14) simplify to

k −1
(P / P ) k − 1 (2-56)
ηad = td ti
Ttd / Tti − 1
k −1 ln( Ptd / Pti )
ηp = (2-57)
k ln(Ttd / Tti )
Reduction of these thermodynamic relations to simple analytical form can yield

substantial reductions in computation time for computerized aerodynamic
analyses. If the fluid is calorically imperfect, numerical integration of Eqs. (2-20)
and (2-21) will normally be necessary. If the fluid is thermally imperfect, the
departure functions of Eqs. (2-47) and (2-48) must also be applied. In either case,
numerical iteration is required for isentropic processes to converge to the
required constant entropy condition. This is the primary source of increased
computation time. Typically, an analysis will perform a massive number of isen-
tropic calculations to relate total and static conditions and to relate conditions in
rotating and stationary coordinate systems. A calorically imperfect, thermally
perfect fluid requires nearly as much computation time as a general non-ideal
gas. Hence, the thermally and calorically perfect gas model offers significant
advantages when it can be employed.
2.9 THE PSEUDO-PERFECT GAS MODEL

The pseudo-perfect gas model (Aungier, 1998, 2000) is a very useful concept that
often provides all of the benefits of the thermally and calorically perfect gas model
for cases where that model is not adequate. Its use requires some care in formulat-
ing an aerodynamic analysis. It is necessary to avoid use of any state relations that
assume the relationships between cp, cv, R and k expressed in Eqs. (2-19) and (2-
27). This really does not complicate the analysis, but the advantages of the pseudo-
perfect gas model cannot be exploited unless this is done. The concept is to use
fictitious values of cp, cv, R and k in an otherwise standard thermally and calori-
cally perfect gas model. If the fictitious constants are represented with an overbar,
they can be calculated from any two state points, designated by subscripts 1 and 2.
R = R z1z2 (2-58)
cp = ( h2 − h1) / (T2 − T1) (2-59)
cv = (u2 − u1) / (T2 − T1) (2-60)
k = ln( P2 / P1) / ln( ρ2 / ρ1) (2-61)
Equation (2-61) yields the isentropic exponent, which can be quite different from
the ratio of specific heats for a real gas. The pseudo-perfect gas model is generally
adequate as long as the values of z1 and z2 are not too different. This is usually
true for compressor components operating on a non-ideal fluid. It can also be
applied to applications such as multistage compressor performance analysis,

simply by recomputing the fictitious constants between stages or between blade

rows. The modest stage pressure ratios in axial-flow compressors almost always
ensure that the model is sufficient. Indeed, this writer has used it extensively for
industrial centrifugal compressor performance analysis, where stage pressure
ratios are much higher and non-ideal working fluids are common. No case has
been encountered to date where the pseudo-perfect gas model yields results sig-
nificantly different from a full non-ideal gas model.
This writer’s normal practice is to select the inlet total conditions and the con-
ditions corresponding to sonic flow as the two state points to define the fictitious
constants—that includes most conditions likely to be encountered in a compres-
sor blade row. The full real gas model is used to compute the constants. Then the
pseudo-perfect gas model is used in the actual analysis. This writer uses a com-
puterized equation-of-state module that contains the pseudo-perfect gas model
as well as the other equations of state described in this chapter. This module is
supported by a gas property database from which the equation-of-state for the
fluid or fluid mixture is easily formulated. All aerodynamic design and analysis
computer programs use this module to perform all equation-of-state calcula-
tions. Even for thermally perfect fluids, the pseudo-perfect gas model yields sub-
stantial reductions in computation time by using a calorically perfect gas
approximation. Its adequacy for those cases is a virtual certainty.
2.10 COMPONENT PERFORMANCE PARAMETERS
The adiabatic and polytropic efficiencies introduced previously in this chapter can
be used to evaluate the performance of a compressor, a compressor stage or even
a single rotor blade row. But they have no meaning for a stationary compressor
component, since total enthalpy is constant for those components. Yet, it is useful
to be able to evaluate a stationary component’s performance against some ideal,
reversible process. Most components in a compressor serve to diffuse the flow to
convert kinetic energy into static pressure. The most important of these is the
exhaust diffuser, although stator and exit guide vane rows are also stationary, dif-
fusing components. One method of evaluating a diffusing component is known as
the diffuser efficiency. Its definition is really quite similar to the adiabatic effi-
ciency introduced previously, except that static enthalpy is used as its basis. Figure
2-4 presents a schematic of an h-s diagram for a diffuser. Flow enters the diffuser
with a velocity Ci and exits the diffuser with a lower velocity Cd. Since the total
enthalpy, H, is constant, Eq. (2-3) can be used to compute the static enthalpy at
the inlet, hi, and at the discharge, hd. It is seen from Fig. 2-4 that this process
results in an increase in static enthalpy of ∆h. Since irreversible losses such as fric-
tion occur, the entropy increases from si to sd. Now, trace the line of constant pres-
sure for the discharge pressure back to the inlet entropy. This shows that the static
enthalpy increase required for an adiabatic, reversible process to produce the
same pressure increase is ∆had. Therefore, a diffuser efficiency can be defined as
∆had
ηdiff = (2-62)
∆h

FIGURE 2-4 A Diffuser h-s Diagram
A more common parameter used to evaluate the performance of a diffuser is the

static pressure recovery coefficient, cp. This expresses the ratio of the static pres-
sure rise to the available kinetic energy at the inlet. Usually, the available kinetic
energy is expressed as a “velocity pressure,” which is the difference between the
total and static pressure.
Pd − Pi
cp = (2-63)
Pti − Pi
Hence, cp is the fraction of the available kinetic energy that has been recovered as
static pressure by the diffuser.
Although compressors are basically diffusing machines, there are some com-
pressor components that accelerate the flow. Inlet passages and inlet guide vanes
are typical examples. In those cases, the component can be evaluated using the
nozzle efficiency. Figure 2-5 illustrates the parameters used to define it. The flow
accelerates from the inlet velocity, Ci, to a higher discharge velocity, Cd. Due to irre-
versible processes such as wall friction, the entropy increases from si to sd and the
total pressure decreases from Pti to Ptd. If the flow is adiabatic (no heat transfer),
the total enthalpy, H, is constant for this process. Equation (2-3) relates H, h and C
at the inlet and the discharge. This is illustrated in Fig. 2-5, along with the ideal dis-
charge velocity, Cad, which would be produced by an isentropic process between

FIGURE 2-5 A Nozzle h-s Diagram
the actual inlet pressure, Pi, and discharge pressure, Pd. A nozzle efficiency can be
defined as the ratio of the actual increase in kinetic energy to the increase in kinetic
energy corresponding to the isentropic process, i.e.,
Cd2 − Ci2
ηnoz = 2 (2-64)
Cad − Ci2
Occasionally, a simpler definition is used: It employs the velocity pressure, Pt – P,

in place of C2 as the measure of kinetic energy and uses Pti – Pd as the ideal (no
loss in total pressure) discharge kinetic energy. Simple substitution of these alter-
nate kinetic energy terms into Eq. (2-64) yields
Pti − Ptd
ηnoz = 1 − (2-65)
Pi − Pd
The most fundamental measure of irreversibility is the increase in entropy. But

an entropy increase is difficult to interpret, so total pressure loss is more com-
monly used. If Eq. (2-8) is applied to total conditions, with total enthalpy constant,
∆Pt Pti − Ptd

= = T∆s (2-66)
ρ ρ

It is known from observation that total pressure loss is usually proportional to

kinetic energy, so a logical definition of loss coefficient can be obtained by
dividing by 1/2C2, i.e., the loss coefficient can be defined as
∆Pt
ω = (2-67)
1
2
ρ C2
Experience has shown that use of velocity pressure as a measure of kinetic

energy provides a better definition, i.e.,
∆Pt
ω = (2-68)
Pt − P
It is found that the second definition is much less sensitive to changes in the
Mach number. This is very important in axial-flow compressor performance
analysis. Typically, blade row performance is predicted with empirical loss coef-
ficient models derived from low-speed cascade tests, but applied to blade rows
with relatively high inlet Mach numbers. The flow is essentially incompressible
(Mach number essentially zero) in the low-speed cascade tests, so the two defi-
nitions of loss coefficient are basically identical. But when empirical loss coef-
ficient models are applied to predict losses in higher Mach number blade rows,
Eq. (2-68) is found to yield substantially better results.
Equations (2-67) and (2-68) are ambiguous with regard to what kinetic
energy is to be used in the denominator. The usual practice is use of the largest
kinetic energy relevant to the component. For diffusing components, such as
diffusers and compressor blade rows, the inlet kinetic energy is usually
employed. For accelerating components, such as turbine blade rows, the dis-
charge kinetic energy is usually employed. The formulation for loss coefficient
assumed constant total enthalpy, which is far from true for rotor blade rows.
But chapter 3 will show that total enthalpy viewed in a frame of reference
rotating with the blades (the relative total enthalpy) is constant for axial flow.
So, the above definitions can be used for rotors in that context. There is a sub-
tle thermodynamic effect present when applying empirical loss coefficient
models to rotating blade rows. In fact, the flow is not usually precisely axial,
i.e., there is usually some change in radius as a streamline passes through a
rotor. Consequently, even the relative total enthalpy is not usually constant
through the blade row. For axial-flow compressors, this effect is so small that
it can be neglected in virtually all cases. The situation is quite different in cen-
trifugal and mixed-flow compressors, where consideration of this thermody-
namic effect is essential. Aungier (2000) develops a correction procedure to
apply empirical loss coefficients to rotating blade rows with a significant
change in radius from inlet to discharge. Since it is not necessary for axial-
flow compressors, it will not be included here. But investigators attempting to
combine axial-flow compressor stages with mixed-flow or radial-flow stages
should consult that reference.

2.11 GAS VISCOSITY
Gas viscosity is often required for aerodynamic analysis. A generalized estima-

tion method compatible with the previous equation-of-state models is provided
by Dean and Stiel (1965), and is reviewed by Ried et al. (1966). It is necessary to
depart from this book’s practice of maintaining independence from specific units
in this case. The Dean and Stiel model predicts viscosity in centipoises and
employs a dimensional parameter defined by
ξ = Tc 6 /  MPc 3 
1 2
(2-69)
where Tc is in degrees Kelvin and Pc is in atmospheres. The low-pressure fluid vis-

cosity is given by
8
µ 0ξ = (3.4 ⋅10−4 )Tr 9 ; Tr ≤ 1.5 (2-70)
µ 0ξ = 0.001668(0.1338Tr − 0.0932) 9 ; Tr > 1.5

5
(2-71)
Then, the viscosity at any pressure is defined by a departure function
( µ − µ 0 )ξ = (1.08 ⋅10−4 )[exp(1.439ρ r ) − exp(−1.111ρ1r .858 ) (2-72)
Ried et al. (1966) indicate that accuracy within 5% can be expected for non-polar
molecules, with somewhat larger errors possible for polar molecules. This viscos-
ity estimation method is quite compatible with the equation-of-state models pre-
viously recommended in this chapter as long as the fluid is in the vapor phase. In
principle, the real gas equations of state apply to the liquid or two-phase fluid
also, but this viscosity model does not. The critical point data needed for
Aungier’s modified Redlich-Kwong equation of state are sufficient for this gas
viscosity model also, but appropriate care is required to adjust the units of Tc and
Pc in Eq. (2-69) and the µ predicted.
2.12 A COMPUTERIZED EQUATION-OF-STATE PACKAGE

One of the most useful tools for the developer of axial-flow compressor design
and analysis software is a computerized equation-of-state package that can be
used as a module in any other analysis to handle all equation-of-state input,
output and thermodynamic calculations. Some useful features to include are
the following.
• Provide the non-ideal gas, ideal gas (with variable specific heats) and
pseudo-perfect gas as options always available to the user. Be sure to
take advantage of opportunities for reduced computation time for the
simpler models. For a basic ideal gas, the pseudo-perfect gas model

functions directly as a thermally and calorically perfect gas model.

Include the capability to specify pseudo-perfect gas data directly, or to
have these data computed from the equation of state and the specified
gas mixture.
• Maintain a database of gas property data for compounds you need.
Establish the capability to set up any of the equation-of-state models
from this data simply by selecting compounds in the mixture and sup-
plying their mole fractions. For the models described in this chapter,
the data needed are M, Pc, Tc, Vc (or zc), ω and cp(T). This writer uses a
third-order polynomial fit of cp as a function of T, consistent with Ried
et al. (1977), which is a good source of all the data required for a very
wide range of compounds. Note that setup of the pseudo-perfect gas
model from the equation of state must be requested by the main analy-
sis when appropriate thermodynamic conditions are known, as
described previously in this chapter. Include a routine to accomplish
this on demand.
• Include logic to save and recover gas property data so that input files of
other programs maintain a common format for all aerodynamic soft-
ware. It is a good idea to include the logic for basic checks on data valid-
ity to be sure the equation of state is complete. Reserve a special code
for the equation-of-state model to signal to the program using this mod-
ule that data is lacking or invalid.
• It is useful to include a standard output routine that can be used to
insert a description of the mixture and the equation of state in use in
output files for other programs.
• Include routines for the basic thermodynamic calculations needed in
aerodynamic analysis. Standard functions likely to be needed include
V(T, P), P(T, V), T(V, P), a(T, P), h(T, P), s(T, P), ηad(Ti, Pi, Td, Pd) and
ηp(Ti, Pi, Td, Pd). A routine to compute the isentropic change in P and T
for a specified change in h is essential for conversion between static and
total thermodynamic conditions and between rotating and stationary
coordinate systems. A routine to compute the speed of sound at sonic
flow conditions for known total conditions is often useful.
• Include the capability to edit the thermodynamic model to modify the
compounds in the mixture, their mole fractions and the equation-of-
state model to be used.
• The Dean and Stiel (1965) viscosity estimation model can be included
to treat all gases and gas mixtures formed from the gas property data-
base. If pseudo-perfect gas data are specified rather than calculated
from the equation of state, viscosity data will also need to be specified.
Using this approach, development of aerodynamic design and analysis software

will involve minimal logic related to the equation of state. In general, axial-flow
compressor state calculations should be limited to the vapor phase using the
constraint expressed in Eq. (2-42). If special consideration involving liquid
phases is necessary, the approximate liquid phase models described in Aungier
(1998, 2000) and the equilibrium flash calculation described in Aungier (2000)
may be useful.

EXERCISES
2.1 Consider uniform axial flow at the inlet of a compressor with no inlet
guide vanes (i.e., stations 0 and 1 are identical) with velocity Cz1 and
Cθ1 = 0. For a thermally and calorically perfect gas, show that ratios
ρ1/ρt1 and a1/at1 are functions of Mach number M1 = Cz1 / a1 (these
results were used in Chapter 1 without derivation to develop the
equivalent flow and speed parameters).
2.2 The flow at station 1 of Exercise 2.1 passes through a rotor exiting at
station 2 with Cz2 = Cz1 and with a swirl velocity, Cθ2, at the mean
radius where the blade speed is ωr = U. The total pressure loss coef-
ficient across the blade row, based on the inlet velocity pressure, is
–. For a thermally and calorically perfect gas, derive expressions for
ω
the ideal (no loss) discharge total temperature and pressure, Tt2id
and Pt2id, and the actual values, Tt2 and Pt2, at the mean radius in
terms of the inlet parameters Pt1, Tt1, H1 and P1. Use the Euler tur-
bine equation, Eq. (1-17), to compute the change in total enthalpy
across the rotor.
2.3 For the rotor row of Exercise 2.2, extend the Mach number equiva-
lence parameters of Chapter 1 to include an expression for the equiv-
alent total enthalpy rise across the rotor. Show that if the flow is
– = 0), Mach number equivalence at the inlet of
isentropic (i.e., that ω
the blade row will produce Mach number equivalence at the dis-
charge.
2.4 The rotor row of Exercises 2.1 and 2.2 is to be operated with a new
working fluid. The operation with the new working fluid will start
with the same inlet conditions (Pt1, Tt1, ρt1) as for the original working
fluid. Its operating speed and mass flow rate will be determined by
requiring Mach number equivalence at station 1. The two fluids have
the same gas constant, R, but the original working fluid has k = 1.4
and the new working fluid has k = 1.38. Assuming that the flow is
isentropic, show that both the equivalent speed and flow parameters
cannot be satisfied at station 2 (the rotor exit). Is complete Mach
number equivalence achieved in this case?
2.5 An axial-flow compressor is to be operated with propane as the work-
ing fluid with an inlet pressure of 200 kPa. Fluid property data for
propane are M = 44.1, Tc = 369.83° K, Pc = 4249.6 kPa and ω = 0.152.
Estimate the lowest inlet temperature that can be used to avoid the
risk of liquid erosion due to two-phase flow.
2.6 Two axial-flow compressors operate on a thermally and calorically
perfect gas with k = 1.4. Both compressors have an adiabatic effi-
ciency of 85%. The pressure ratios of the two compressors are 3.0 and
5.0. Calculate the polytropic efficiencies of the two compressors.
2.7 An axial-flow compressor consists of three repeating stages, each with
a stage pressure ratio of 1.1 and a stage adiabatic efficiency of 85%.
The working fluid is a thermally and calorically perfect gas with k =
1.4. Compute the overall adiabatic efficiency of the three-stage com-
pressor. Repeat this problem assuming all efficiencies are polytropic.

2.8 Consider adiabatic flow of a calorically and thermally perfect gas with
k = 1.4 through a diffuser. The inlet total temperature is 300° K, the
inlet total pressure is 230 kPa, the inlet static pressure is 200 kPa, the
static pressure recovery coefficient is 0.6 and the total pressure loss
coefficient based on the inlet velocity pressure, (Pti – Pi), is 0.1. Com-
pute the diffuser efficiency, ηdif.

Chapter 3
FLUID MECHANICS
Fluid mechanics and thermodynamics are the fundamental sciences used for the
aerodynamic design and analysis of axial-flow compressors. This chapter high-
lights some fundamental concepts from fluid mechanics to complement the con-
cepts from thermodynamics covered in Chapter 2. The governing equations will
be developed in forms suitable for the various aerodynamic analyses commonly
employed for axial-flow compressors. Detailed solution procedures will be cov-
ered in subsequent chapters.
Several types of fluid dynamic analysis are useful for this purpose. The
through-flow analysis is widely used in both design and performance analysis.
This involves solving the governing equations in the hub-to-shroud plane at sta-
tions located between blade rows. The flow is normally considered to be axisym-
metric at these locations, but still three-dimensional because of the existence of a
tangential velocity component. Empirical models are employed to account for
the fluid turning and losses that occur when the flow passes through the blade
rows. A simplification of this analysis is the “pitch-line” or “mean-line” one-
dimensional flow model, which ignores the hub-to-shroud variations. These were
very common for many years, but are no longer particularly relevant to the prob-
lem. Computers are sufficiently powerful today that there is really no need to
simplify the problem that much. The through flow in an axial-flow compressor is
strongly influenced by viscous effects near the end walls. The primary influence
from these end-wall boundary layers is commonly described as end-wall block-
age. An inviscid through-flow analysis ignores the low momentum fluid in the
boundary layers and will overestimate the mass flow that the passage can accom-
modate for a given flow field solution. To compensate the common practice is to
impose a blockage factor to effectively reduce the passage area. This requires
consideration of boundary layer analysis to estimate the appropriate blockage
factors to be used. More fundamental internal flow analyses are often useful for
specific components, particularly blade rows. These include two-dimensional
flow analyses in either the blade-to-blade or hub-to-shroud direction, and quasi-
three-dimensional flow analyses developed by combining and interacting these
two-dimensional analyses. Again, wall boundary layer analysis is often used to
evaluate viscous effects. Any of these analyses may be used in a design mode as
well as an analysis mode. A design mode seeks to define the gas path geometry
(end-wall contours and blades) to produce the desired flow field, while an analy-
sis mode seeks to predict the flow field from specified geometry.

Viscous computational fluid dynamics (CFD) solutions are also in use for axial-
flow compressors. These are typically three-dimensional flow analyses, which
consider the effects of viscosity, thermal conductivity and turbulence. In most
cases, commercial viscous CFD codes are used although some in-house codes are
in use within the larger companies. Most design organizations cannot commit the
dedicated effort required to develop these highly sophisticated codes, particularly
since viscous CFD technology is changing so rapidly that any code developed will
soon be obsolete unless its development continues as an ongoing activity. Conse-
quently, viscous CFD is not covered in this book beyond recognizing it as an essen-
tial technology and pointing out some applications for which it can be effectively
used to supplement conventional aerodynamic analysis techniques.
NOMENCLATURE
a = sound speed
b = stream sheet thickness
E = entrainment function
→
e = unit vector
f = body force
H = total enthalpy
h = static enthalpy
I = rothalpy
m = meridional coordinate
n = normal coordinate
P = pressure
r = radius
→
r = position vector in space
s = entropy
T = temperature
u = velocity in x direction
→
V = general vector
v = velocity in y direction
ẇ = power
x = coordinate along a wall
y = coordinate normal to a wall
z = axial coordinate
δ = boundary layer thickness
δ* = displacement thickness
θ = tangential coordinate and momentum thickness
κ = curvature
ν = force defect thickness
ρ = gas density
τ = torque and shear stress
φ = streamline slope with axis and a general function
ω = rotation speed

Fluid Mechanics • 43
Subscripts
e = boundary layer edge condition

m = meridional component
n = normal component
r = radial component
w = parameter at a wall
1 = blade inlet parameter and meridional defect parameter
11 = meridional defect parameter
12 = tangential flux defect parameter
2 = blade exit parameter and tangential defect parameter
22 = tangential defect parameter
θ = tangential component
Superscripts
′ = a relative value in the rotating coordinate system
3.1 FLOW IN A ROTATING COORDINATE SYSTEM

The analysis of the flow in rotor blade rows is accomplished in a coordinate sys-
tem, which rotates with the blade. The flow conditions in a rotating coordinate
system are referred to as the relative conditions. If a blade row is rotating with an
angular velocity, ω, the relative tangential velocity in a coordinate system rotating
with the blade, Wθ, is related to the absolute tangential velocity, Cθ, by
Wθ = Cθ − ω r (3-1)
The axial and radial velocity components are independent of the rotation, i.e.,
Wz = Cz (3-2)
Wr = Cr (3-3)
It will be more convenient to work with the meridional velocity component, Wm,
defined as
Wm = Wz2 + Wr2 = Cm (3-4)
Wm is the velocity component lying in the meridional (constant θ) plane and in a

stream surface. A stream surface is defined as a surface having no fluid velocity
component normal to it and, therefore, no mass flow across it. Thus, the defining
characteristic of a stream surface is that the mass flow rate between it and the
hub contour surface is constant everywhere. The meridional coordinate, m, is
measured along the stream surface and in a meridional plane, i.e.,

( dm)2 = ( dr )2 + ( dz)2 (3-5)
For axial-flow compressors, it is almost always reasonable to assume stream sur-

faces are axisymmetric. That assumption will be used throughout this book. Fig-
ure 3-1 illustrates a schematic of a stream surface and unit vectors for the
meridional and polar coordinates. Normal to these unit vectors, and to the
stream surface, is the third coordinate of interest, the normal coordinate, n, as
shown in Fig. 3-2. It is convenient to develop the governing equations of fluid
mechanics in this “natural” coordinate system (θ, m, n), where by definition
Wn = Cn = 0 (3-6)
Now consider the flow through a thin stream sheet, i.e., a thin annular passage
bounded by two stream surfaces. The torque, τ, acting on the fluid between
meridional stations 1 and 2 is given by conservation of angular momentum.
τ =m
˙ ( r2Cθ 2 − r1Cθ1) (3-7)
This torque must balance the power input, i.e.,
˙ = ωτ = ω m
w ˙ ( r2Cθ 2 − r1Cθ1)
(3-8)
Combining Eq. (3-8) with Eq. (2-4) yields the well-known Euler turbine equation
FIGURE 3-1 Schematic of a Stream Surface

FIGURE 3-2 Natural Coordinate System
H2 − H1 = ω ( r2Cθ 2 − r1Cθ1) (3-9)
This is the general energy equation relating the total enthalpy change produced
by a transfer of mechanical energy between the fluid and a rotating blade row. It
is convenient to introduce the rothalpy, I, defined by
I = H − ω rCθ (3-10)
On introducing Eq. (3-10) into Eq. (3-9), it can be seen that rothalpy is constant
on a stream surface. Hence, rothalpy is the basic parameter expressing energy
conservation for a rotating blade row. It is also valid for a stationary blade row,
since I = H in that case, and Eq. (3-9) requires that H be constant in the absence
of energy transfer with a rotating blade row.
Aerodynamic analysis of axial-flow compressors involves alternately solving
the governing equations in rotating coordinates (rotors) and stationary coordi-
nates (stators). Hence, we need to relate the relative total enthalpy, H′, in a rotat-
ing coordinate system to the absolute total enthalpy, H, in a stationary coordinate
system. Noting that static thermodynamic conditions are identical for either
coordinate system,
h = H ′ − 12 W 2 = H − 12 C 2 (3-11)
The relative velocity, W, follows from Eqs. (3-1), (3-4) and (3-6)

W = Wm2 + Wθ2 (3-12)
Equations (3-1), (3-10), (3-11) and (3-12) combine to yield
H ′ = H − ω rCθ + 12 (ω r )2 = I + 12 (ω r )2 (3-13)
Noting that entropy is constant between total and static conditions, and therefore
between a rotating and a stationary coordinate system, Eq. (3-13) can be used to
relate the two coordinate systems. For example, the change in all other relative
total thermodynamic conditions between the two coordinate systems can be cal-
culated from an appropriate equation of state as a function of (H, H′, s). This
requires calculation of the isentropic change in the parameter of interest for a
specified change in enthalpy. Hence, Eq. (3-13) is an important relation that
allows us to relate all thermodynamic parameters between the stationary and the
rotating coordinate systems. Also, since I is constant on the stream surface, Eq.
(3-13) allows calculation of H′ at all points on a stream surface when one value is
known, e.g., at the inlet.
3.2 ADIABATIC INVISCID COMPRESSIBLE FLOW

Adiabatic compressible inviscid flow analysis is commonly used in turbomachin-
ery. This flow model assumes that fluid viscosity and thermal conductivity can be
neglected. Basic conservation of mass, momentum and energy, supported by a
suitable equation of state, govern the flow. It is useful to derive the governing
equations in a rotating coordinate system, noting that these equations will be
valid for a stationary coordinate system if ω is set to zero. The vector form of the
momentum equation can be written (Novak, 1967; Vavra, 1960; Wu, 1952)
r r
dC 1 r dW r r r r r
= − ∇P = + 2(ω × W ) + ω × (ω × r ) (3-14)
dt ρ dt
where the last two terms in Eq. (3-14) are the Coriolis and centrifugal accelera-
tions imposed by the rotating coordinate system, and the time derivative is the
substantial derivative, i.e.,
r r
dW ∂W r r r
= + (W ⋅ ∇)W (3-15)
dt ∂t
Hence, the momentum equation in rotating coordinates is

r r
∂W r r r r r r r r ∇P
+ (W ⋅ ∇) W + 2(ω × W ) + ω × (ω × r ) = − (3-16)
∂t ρ
Using standard vector identities, this equation can also be written as

r r
r r r r r
∂W 2r 1 2 ∇P
− W × (∇ × W + 2ω ) − rω er + 2 ∇W = − (3-17)
∂t ρ
where →er is a unit vector in the radial direction. On introducing Eqs. (2-3), (2-8),
(3-10) and (3-11), an alternate form of Eq. (3-17) is obtained.
r
∂W r r r r r r
− W × (∇ × W + 2ω ) = T ∇s −∇I (3-18)
∂t
The continuity and energy equations in vector form are
∂ρ r r
+ ∇ ⋅ (ρ W ) = 0 (3-19)
∂t
∂I 1 ∂P r r
− + (W ⋅ ∇)I = 0 (3-20)
∂t ρ ∂t
Equations (3-16) through (3-20) are vector equations, which are valid in any
coordinate system. To express the equations in the natural coordinates (θ, m, n),
standard curvilinear coordinate transformations are used. These can be found in
most advanced calculus books, which cover vector field theory. Vavra (1960,
Appendix A) provides specific and detailed derivations of the vector operators
and governing equations in natural coordinates. For general reference, the
important vector operators are provided without derivation at the end of this
chapter. The resulting governing equations are
∂ρ 1  ∂ rρ Wm ∂ρ Wθ 
+ + + κ n ρ Wm = 0 (3-21)
∂t r  ∂m ∂θ 
∂Wm ∂Wm Wθ ∂Wm sin φ 1 ∂P
+ Wm + − [Wθ + ω r]2 = − (3-22)
∂t ∂m r ∂θ r ρ ∂m
∂Wθ ∂Wθ Wθ ∂Wθ Wm sin φ 1 ∂P
+ Wm + + [Wθ + 2ω r] = − (3-23)
∂t ∂m r ∂θ r rρ ∂θ
cos φ 1 ∂P
κ mWm2 + [Wθ + ω r]2 = (3-24)
r ρ ∂n
∂I 1 ∂P ∂I Wθ ∂I
− + Wm + =0 (3-25)
∂t ρ ∂t ∂m r ∂θ
The curvature of the stream sheet, κm, and of the normal surface, κn, are related
to the angle φ shown in Fig. 3-2.
∂φ
κm = − (3-26)
∂m
∂φ 1 ∂b
κn = = (3-27)
∂n b ∂m

Parameter b in Eq. (3-27) is the thickness of a stream sheet bounded by two

stream surfaces, as shown in Fig. 3-2. Hence, the curvature κn is related to the
meridional divergence or convergence of the stream surfaces. That form is useful
in some applications of these governing equations, particularly when analyzing
the two-dimensional flow in a blade-to-blade stream surface. Equations (3-22)
through (3-24) can be expressed differently using Eq. (3-18).
∂Wm Wθ  ∂Wm ∂( rWθ + ω r 2 )  ∂s ∂I

+  − =T − (3-28)
∂t r  ∂θ ∂m  ∂m ∂m
∂Wθ Wm  ∂Wm ∂( rWθ + ω r 2 )  ∂s ∂I

−  − =T − (3-29)
∂t r  ∂θ ∂m  ∂θ ∂θ
Wθ ∂( rWθ + ω r 2 ) ∂Wm ∂I ∂s
κ mWm2 + + Wm = −T (3-30)
r ∂n ∂n ∂n ∂n
Since there are only two velocity components (i.e., Wn = 0), one of the three
momentum equations is redundant. The redundant equation has been replaced
by the assumption that the stream surfaces are known or can somehow be deter-
mined as part of the solution. If the meridional surfaces are not stream surfaces,
the governing equations must be modified to include a normal velocity compo-
nent, Wn. This will not be required for analyses described in this book, although
there is no reason why a flow analysis could not be accomplished in an arbitrary
(θ, m, n) coordinate system. Aungier (2000) includes the more general form of the
governing equations appropriate for that type of analysis.
3.3 ADIABATIC INVISCID COMPRESSIBLE FLOW

APPLICATIONS
The governing equations are applied in a variety of analyses in the aerodynamic

design and analysis of axial-flow compressors. Most analyses employ the time-
steady form of the governing equations, although the unsteady form does find
application when the fluid velocity exceeds the sonic flow velocity. One of the
most common applications is to determine the flow in the meridional plane. This
application normally restricts the solution to stations outside of the blade rows,
using empirical models to impose the influence of blade rows between stations in
terms of fluid turning and total pressure loss correlations. These analyses treat the
flow as axisymmetric to require conservation of mass, normal momentum and
energy. Early practice was to develop these analyses in a simple cylindrical coor-
dinate system (θ, m, r) such that the normal coordinate is replaced by the radial
coordinate. Hence, the normal momentum equation was commonly referred to as
the radial equilibrium equation in that context. Various simplifications may be
employed in these analyses. If stream surface curvature is neglected (κm = 0) and
the gradient of entropy across the passage is assumed to be zero, the term “simple
radial equilibrium” has been used to describe the analysis. If the gradient of
entropy is included, the solution is often referred to as “simple non-isentropic

radial equilibrium.” When curvature and entropy are both included, the term “full
radial equilibrium” is often used. Advances in computer technology and numeri-
cal analysis techniques have reduced the role of simple radial equilibrium solu-
tions to cases where the entropy gradient cannot be properly defined, such as the
general-purpose stage design described in Chapter 10. Simple non-isentropic
radial equilibrium continues to be useful in basic blade row or stage design, where
the streamline curvatures to be encountered in the actual compressor are not
known. Indeed, simple non-isentropic radial equilibrium analysis is often quite
sufficient for actual axial-flow compressors, where stream surface curvatures may
be negligible. This is often true for industrial axial-flow compressors. When the
stream surface curvature can be ignored, a dramatic reduction in computation
time is realized, since the analysis becomes a simple marching solution. This fol-
lows from the fact that the flow at any axial station is not dependent on the flow at
downstream stations. A simple variant used by this writer is to approximate
stream surface curvatures from end-wall contours by simple linear interpolation.
This allows the advantages of a simple non-isentropic radial equilibrium analysis,
yet can approximate curvatures imposed by end-wall contour design.
Solutions for the two-dimensional flow in the meridional plane within blade
passages are also fairly common. These usually seek to predict the average flow
in the passage from the hub to the shroud as a two-dimensional flow problem. In
the more general case, these hub-to-shroud analyses may solve for the two-
dimensional flow on specific stream surfaces from hub to shroud. In both cases,
either the flow angle or Wθ distributions throughout the passage must be sup-
plied to replace solution of the tangential momentum equation.
Analysis of the flow passing through a blade row and lying on a stream surface
is also common in axial-flow compressor design and analysis. These two-dimen-
sional flow analyses are commonly called blade-to-blade flow analyses. Typically,
the stream surface geometry is specified along with the distribution of the stream
sheet thickness, b. Then, conservation of mass, energy, tangential momentum
and meridional momentum can describe the flow. If the flow is assumed to be
isentropic, Eqs. (3-28) and (3-29) show that one of the momentum equations is
redundant, resulting in a simpler problem. This is fairly common practice for
subsonic flow problems and is referred to as potential flow or irrotational flow.
This results in a classical boundary value problem of an elliptic equation. Indeed,
the governing equations for inviscid flow are elliptic in form as long as W < a
throughout the flow field. When supersonic flow is encountered (W > a), the gov-
erning equations become hyperbolic in mathematical form, which requires a
marching type solution—such as the method of characteristics—rather than a
boundary value problem solution. Cases where the flow is supersonic throughout
are rare. Usually mixed subsonic-supersonic flow is involved. Then the time-
steady governing equations are elliptic in some regions and hyperbolic in others,
requiring two different solution techniques that must be matched together in
some fashion. It is now fairly common practice to employ the time-unsteady
equations for these cases. The advantage of that approach is that the unsteady
equations of motion are hyperbolic in form for both subsonic and supersonic
flow. This allows a single numerical method to be used for the mixed subsonic-
supersonic flow case. This approach is commonly called the “time-dependent” or
“time-marching” method of solution.

Hub-to-shroud and blade-to-blade flow analyses may also be combined to

form a “quasi-three-dimensional” flow analysis within blade passages. A hub-
to-shroud analysis can supply the stream surface geometry and stream sheet
thickness distribution required for a blade-to-blade flow analysis. Similarly, a
blade-to-blade flow analysis can supply the distribution of flow angle or tan-
gential velocity required for a hub-to-shroud flow analysis. Thus an iterative
solution solving these two two-dimensional flow problems with interaction
between them can provide an approximate three-dimensional flow analysis.
This approach was originally suggested by Wu (1952) and is a commonly used
analysis technique.
Analysis of the complete three-dimensional inviscid flow problem is seldom
used today. The additional information supplied relative to the simpler and faster
quasi-three-dimensional flow analysis is relatively minor. Also, the evolution of
three-dimensional viscous CFD analysis techniques has discouraged use of a
three-dimensional inviscid flow analysis. As discussed at the beginning of this
chapter, most turbomachinery design groups use one of the many excellent com-
mercially available viscous CFD codes when a more detailed and fundamental
analysis is needed.
3.4 BOUNDARY LAYER ANALYSIS
Adiabatic inviscid flow analyses, such as those described in the previous sections,
are commonly augmented by boundary layer analysis techniques to evaluate vis-
cous effects that are not considered by those analyses. The basic premise of
boundary layer theory is that viscous effects are confined to a thin layer close to
the physical surfaces bounding the flow passages (Schlichting, 1968, 1979). This
is by no means always the case in axial-flow compressors, but selective use of
boundary layer analysis has been found to be very effective in many applications.
As with the discussion of three-dimensional inviscid flow analysis in the previous
section, practical trade-offs with fully viscous flow analyses must be considered.
Boundary layers in axial-flow compressors always involve significant three-
dimensional character. Yet, there is little merit to a fully three-dimensional
boundary layer analysis today, when commercially available viscous CFD codes
can treat the problem much more accurately. Rather, it is the simplified bound-
ary layer analysis techniques that are most effective in augmenting the inviscid
flow analyses discussed in the previous section. Indeed, it is little short of
remarkable that boundary layer analysis has been used so effectively in axial-flow
compressor design and analysis, considering the fact that the fundamental
boundary layer approximations are almost always violated to some degree and
very often to a substantial degree. There are two important types of boundary
layer analysis commonly used in axial-flow compressor aerodynamics. These
involve boundary layers on the blade surfaces and those on the compressor end-
wall contours. Blade surface boundary layers are of interest since they play a key
role in viscous losses and stall or boundary layer separation. End-wall boundary
layers are extremely important in performance analysis, since they can produce
substantial viscous blockage effects that have significant impact on a compres-
sor’s performance.

Two-dimensional boundary layer analysis is a useful approximation for blade

surface boundary layers, particularly in blade design. The primary goals of blade
design are really governed by viscous effects, namely minimizing viscous losses
and avoiding or delaying flow separation. Two-dimensional boundary layer
analysis provides at least a qualitative assessment of these effects, yet adds very
little complexity or computational time to the overall analysis. Indeed, conven-
tional practice for blade design is to design blade sections in the context of sim-
ple two-dimensional cascades, where two-dimensional boundary layer analysis is
directly applicable. These blade sections are then “stacked” to create the actual
three-dimensional compressor blade. When applied to the actual compressor
blade surface boundary layers, two-dimensional boundary layer analysis pro-
vides only qualitative results, since three-dimensional effects that are not consid-
ered by the analysis often become significant.
By contrast, end-wall boundary layers are necessarily three-dimensional, due
to the presence of the swirl velocity component and tangential blade forces. Sim-
ilar to inviscid flow analysis in the meridional plane, it is fairly common practice
to conduct the boundary layer analysis for stations between blades—where an
axisymmetric, three-dimensional boundary layer approximation can be used—
while relying on empirical models to impose the blade row influence. This
approach is a common basis for end-wall boundary layer blockage calculations
for aerodynamic performance analysis (e.g., Balsa and Mellor, 1975). This
axisymmetric three-dimensional boundary layer model has also been used within
blade passages to provide an approximation to the boundary layer averaged
between blade passages (Horlock, 1970, Aungier, 2000). Again, empirical models
are required to model the influence of blade forces.
The basic boundary layer equations relevant to these two types of boundary
layers will be developed in the remainder of this chapter. Specific applications of
these governing equations will be discussed, as required, in subsequent chapters
to support the various inviscid flow analyses.
3.5 TWO-DIMENSIONAL BOUNDARY LAYER ANALYSIS
Basic conservation of mass and momentum provide the governing equations for
two-dimensional boundary layer flow over an adiabatic wall.
∂ρ bu ∂ρ bv
+ =0 (3-31)
∂x ∂y
∂u ∂u 1 ∂P 1 ∂τ
u +v + = (3-32)
∂x ∂y ρ ∂x ρ ∂y
where τ is the shear stress. The coordinates (x, y) and velocity components
(u, v) are illustrated in Fig. 3-3, along with a typical boundary layer velocity
profile. The stream sheet thickness, b, has been included in Eq. (3-31) since it
is often a function of x in turbomachinery applications, i.e., streamlines often
converge or diverge. This directly affects conservation of mass. The basic
assumption of boundary layer theory is that the pressure is constant across the

FIGURE 3-3 Boundary Layer Nomenclature
boundary layer, i.e., P is a function of x only. Boundary layer analysis in turbo-

machinery is most conveniently accomplished by applying the governing equa-
tions in integral form. Equation (3-31) can be integrated across the boundary
layer, using the Liebnitz rule to interchange the order of integration and dif-
ferentiation, to yield
δ
∂ ∂δ ∂
∂x ∫
bρ udy = bρ eue − bρ e ve = [bρ eue (δ − δ * )] (3-33)
∂x ∂x
0
The subscript e denotes inviscid flow conditions at the boundary layer edge, δ is
the boundary layer thickness and δ* is called the displacement thickness or mass
defect thickness, defined as
δ
ρ eueδ * = ∫ [ρ eue − ρ u]dy (3-34)
0
Equation (3-34) can be rewritten as
∫ ρ udy = ρeue [δ − δ
*
] (3-35)
0

The displacement thickness is a fictitious thickness that can be used to correct the
mass balance relative to the inviscid flow solution. As seen from Eq. (3-35), if the
inviscid boundary layer edge or “free stream” conditions are applied within the
boundary layer and the thickness δ* is assumed to have zero mass flow, mass con-
servation will be corrected for viscous effects. Equations (3-31) and (3-32) can be
combined to express the momentum equation in conservation form. This yields
1 ∂bρ u2 ∂ρ uv ∂P ∂τ
+ + = (3-36)
b ∂x ∂y ∂x ∂y
Analogous to the displacement thickness, the momentum thickness or momen-

tum defect thickness is defined as
ρ eue2θ = ∫ ρ u[ue − u]dy (3-37)
Equations (3-35) and (3-37) combine to yield
∫ ρ u dy = ρeue [δ − δ
2 2 *
− θ] (3-38)
0
If the free stream conditions are applied within the boundary layer with no flow
in the thickness δ* and, in addition, no momentum in the thickness θ, momentum
conservation will be corrected for viscous effects. Hence, if δ*and θ can be pre-
dicted, we have a simple method to correct the known inviscid free stream mass
and momentum flux for viscous effects. This is really the basis of integral bound-
ary layer analysis methods. Integrating Eq. (3-36) across the boundary layer,
again using the Liebnitz rule, and noting that P = Pe is constant across the bound-
ary layer, yields
δ
∂ ∂δ ∂P
∂x ∫
bρ u2dy − ρ eue2 + ρ eue ve + δ e = −τ w (3-39)
∂x ∂x
0
Combining Eqs. (3-33), (3-38) and (3-39) yields
1 ∂ u ∂ ∂P
[bρ eue2 (δ − δ * − θ )] − e [bρ eue (δ − δ * )] + δ e = −τ w (3-40)
b ∂x b ∂x ∂x
Equation (3-40) can be rearranged to yield
1 ∂bρ eue2θ ∂u  ∂P ∂u 
+ δ * ρ eue e − τ w = δ  e + ρ eue e  (3-41)
b ∂x ∂x  ∂x ∂x 
By applying Eq. (3-32) at the boundary layer edge, where the gradients of u and τ
in the y direction are zero, it is easily seen that the right-hand side of Eq. (3-41) is

identically zero. Hence, Eq. (3-41) reduces to the well-known momentum inte-
gral equation.
1 ∂bρ eue2θ ∂u
+ δ * ρ eue e = τ w (3-42)
b ∂x ∂x
In the special case of two-dimensional, axisymmetric flow, it can be noted that b

is proportional to radius, r, and the momentum integral equation becomes
1 ∂rρ eue2θ ∂u
+ δ * ρ eue e = τ w (3-43)
r ∂x ∂x
The momentum integral equation is valid for both laminar and turbulent bound-
ary layers. Laminar boundary layer analysis usually employs specific boundary
layer flow profile assumptions to permit direct integration of the momentum
integral equation. Turbulent boundary layer analysis usually employs several
empirical models for solution, which may include specific boundary layer flow
profile assumptions. Usually, turbulent boundary layer analysis employs a sec-
ond conservation equation, such as conservation of mass, energy or moment of
momentum (Rotta, 1966). This writer prefers conservation of mass as the second
equation, commonly called the entrainment equation. In this case, Eq. (3-33) is
written in the form
∂
[bρ eue (δ − δ * )] = bρ eue E (3-44)
∂x
The parameter E is called the entrainment function, which specifies the rate at
which free stream fluid is entrained into the boundary layer at the boundary
layer edge. To employ this model, an empirical correlation for E is required,
which must be derived from experiment. Combining Eqs. (3-33) and (3-44), it is
seen that the entrainment function is given by
∂δ ve
E= − (3-45)
∂x ue
Indeed, entrainment is governed by the gradient of the shear stress at the bound-
ary layer edge. Hence, entrainment should depend on the shape of the boundary
layer profiles, which is the usual basis for empirical models.
3.6 AXISYMMETRIC THREE-DIMENSIONAL BOUNDARY

LAYER ANALYSIS
Equation (3-43) describes axisymmetric boundary layers where the flow field
is two-dimensional, i.e., there is no tangential velocity component. When a

tangential velocity component is present in an otherwise axisymmetric flow

field, the meridional and tangential components of the boundary layer profiles
develop differently, resulting in a three-dimensional flow problem. The analysis
of these problems is referred to as axisymmetric three-dimensional boundary
layer analysis. This type of analysis is directly applicable to swirling flows in
annular passages with no blades or vanes present, such as inlets and diffusers
[e.g., Aungier 1988(b); Davis, 1976; Senoo et al., 1977]. This model has also
been used effectively in axial-flow compressor performance analysis with appli-
cation to stations between blade rows (e.g., Balsa and Mellor, 1975). When a
blade row lies between successive computing stations, the axisymmetric
assumption will have been violated within the blade passage. This requires use
of empirical models to address the influence of the blade rows. Horlock (1970)
also reports some success while applying this model within blade row passages.
While these flows are far from axisymmetric, this model is used to provide an
evaluation of the average or mean boundary layer behavior between the blades
on the end-wall contours. Aungier (2000) uses this approach for quasi-three-
dimensional flow analysis, a practice also followed in this book.
The governing equations for axisymmetric three-dimensional boundary layer
flow in a rotating coordinate system in natural coordinates (θ, m, y) are
1 ∂ρ Wm ∂ρ Wy
+ =0 (3-46)
r ∂m ∂y
∂Wm ∂Wm sin φ 1 ∂P ∂τ 
Wm + Wy − (Wθ + ω r )2 =  fm − e − m  (3-47)
∂m ∂y r ρ ∂m ∂y 
∂Wθ ∂Wθ sin φ 1 ∂τ 
Wm + Wy + Wm (Wθ + 2ω r ) =  fθ − θ  (3-48)
∂m ∂y r ρ ∂y 
A rotating coordinate system is needed for turbomachinery applications, since

end-walls may be either rotating or stationary. The coordinate system should
rotate with the end-wall to simplify imposing the boundary condition that all
velocity components are identically zero at the wall. The terms fm and fθ in Eqs.
(3-47) and (3-48) are body force terms used to account for blade forces acting on
the flow. The blade forces at the boundary layer edge can be evaluated directly by
applying Eqs. (3-47) and (3-48) to the inviscid flow at the boundary layer edge,
where all inviscid free stream conditions are known and where Wn and the shear
stress terms are identically zero. Hence,
∂Wme ∂Pe sin φ

fme = ρ eWme + − ρ e (Wθe + ω r )2 (3-49)
∂m ∂m r
∂Wθe sin φ ρ W ∂rCθe
fθe = ρ eWme + ρ eWme (Wθe + 2ω r ) = e me (3-50)
∂m r r ∂m
The boundary layer equations are converted to integral form in the same fashion
as described earlier. The algebra is more tedious and several additional defect
thicknesses are required. The resulting integral equations are

∂
[rρ eWme (δ − δ1* )] = rρ eWe E (3-51)
∂m
∂ 2 ∂Wme
[rρ eWmeθ11] + δ1* rρ eWme − ρ eWθe sin φ [Wθe (δ 2* + θ 22 ) + 2ω rδ 2* ]
∂m ∂m (3-52)
= r[τ mw + fmeν m ]
∂ 2  ∂Wθe 
[r ρ eWmeWθeθ12 ] + rδ1* ρ eWme  r + sin φ (Wθe + 2rω )
∂m  ∂m  (3-53)
= r 2 [τ θw + fθeνθ ]
The various mass, momentum and force defects used in these equations are
defined as
δ
ρ eWmeδ1* = ∫ ( ρ eWme − ρ Wm )dy (3-54)
0
δ
2
ρ eWmeθ11 = ∫ ρ Wm (Wme − Wm )dy (3-55)
0
δ
ρ eWmeWθeθ12 = ∫ ρ Wm (Wθe − Wθ )dy (3-56)
0
δ
ρ eWθeδ 2* = ∫ ( ρ eWθe − ρ Wθ )dy (3-57)
0
δ
ρ eWθ2eθ 22 = ∫ ρ Wθ (Wθe − Wθ )dy (3-58)
0
δ
ν m fme = ∫ ( fme − fm )dy (3-59)
0
δ
νθ fθe = ∫ ( fθe − fθ )dy (3-60)
0
The momentum integral equations can be simplified by assuming that the

blade forces are constant through the boundary layer. This would certainly be
consistent with the usual boundary layer approximations, but it is now
accepted that this is often not true for end-wall boundary layers of axial-flow
compressors. Mellor and Wood (1971) advanced compelling arguments for the
existence of force defects. Smith (1970) and Hunter and Cumpsty (1982) meas-
ured force defects experimentally. For vaneless annular passages, these body
force terms will normally vanish, but they can be used to advantage when
boundary layers merge to form fully developed viscous flow profiles [Aungier,
1988(b)].

3.7 Vector Operators in Natural Coordinates
The development of the governing equations in an axisymmetric natural coordi-

nate system presented in this chapter requires use of several standard vector
operators. Appendix A of Vavra (1960) provides detailed derivations of these
operators. Here, the various operators are presented for reference purposes,
without derivation. The gradient of any function φ is given by
r ∂φ r ∂φ r 1 ∂φ r
∇φ = em + en + eθ (3-61)
∂m ∂n r ∂θ
→
where →
e is a unit vector. The divergence of any vector V is given by
r 1  ∂rV ∂rVn ∂Vθ 

∇ ⋅V =  m
+ + + κ nVm + κ mVn
∂θ 
(3-62)
r  ∂m ∂n
→
The curl of any vector V is given by
r r  ∂V ∂V r
∇ × V =  n − m + κ nVn − κ mVm  eθ
 ∂ m ∂n 
1  ∂rVθ ∂Vn  r 1  ∂Vm ∂rVθ  r (3-63)
+  − em +  − en
r  ∂n ∂θ  r  ∂θ ∂m 
The Laplacian of any function φ is given by
1 ∂2φ ∂2φ ∂2φ  1 ∂r  ∂φ  1 ∂r  ∂φ

∇2φ = + + + +κ n + +κm (3-64)
 r ∂m  ∂m  r ∂n  ∂n
2 2 2
r ∂θ ∂m ∂n2
In evaluating the convective derivative in Eq (3-15), the following vector identity

is useful.
r r r r r r r r r
(V ⋅ ∇)V = 1 ∇(V ⋅ V ) − V × (∇ × V )
2 (3-65)
Equation (3-65) is usually written as

r r r r r r r
(V ⋅ ∇)V = 1 ∇V 2 − V × (∇ × V ) (3-66)
2
→
where V is the magnitude of the vector V as given by the dot product.
EXERCISES
3.1 Consider time-steady flow in a rotating coordinate system. Use Eqs.

(3-28) through (3-30) to analyze the flow at stations outside of the

blade row passages where the flow can be considered to be axisym-

metric, but may have a tangential velocity component. Develop equa-
tions governing the variation of Cθ, s and I on stream surfaces.
3.2 For the flow analysis in Exercise 3.1, modify Eq. (3-30) to consider
cases where the relative flow angle, β′, is known for all stream sur-
faces, where tanβ′ = Wθ / Wm. Repeat for cases where the absolute
flow angle, β, is known for all stream surfaces where tanβ = Cθ / Cm.
3.3 For Eq. (3-66), express V in terms of its components in the three coor-
→
dinate directions. Derive an equation for –21 ∇V2 in terms of V and its
derivatives.
3.4 Consider one-dimensional, time-steady flow in a simple annular pas-
sage, i.e., κm = 0 and all gradients with respect to n and θ are identi-
cally zero. The passage width, b(m), is a function of the meridional
coordinate. Derive a set of governing equations for this problem from
Eqs. (3-21) through (3-27).
3.5 Consider time-steady one-dimensional flow at the exit of a simple
annular passage, with two identical boundary layers on the end-walls.
The boundary layer parameters θ, δ* and δ and the inviscid core flow
data ρ, u and P at the passage exit are known. The flow is incompress-
ible, i.e., ρ is constant and Pt = P + 1/2ρu2. Develop expressions for the
exit mass and momentum flow in terms of the boundary layer and
inviscid core flow parameters.
3.6 Assume that the boundary layer flow and inviscid core flow in Exer-
cise 3.5 mix instantaneously into a uniform flow with no change in
static pressure. By requiring conservation of mass and momentum,
show that the total pressure loss between the inviscid core flow and
the fully mixed flow is given by
∆Pt = 1
2
ρ eue2 [(2δ * / b)2 + 4θ / b]

Chapter 4
AXIAL-FLOW COMPRESSOR
BLADE PROFILES
The traditional approach to axial-flow compressor aerodynamic design was to

use various families of airfoils as the basis for blade design. American practice
was based on various families defined by the National Advisory Committee for
Aeronautics (NACA), the most popular being the 65-series family. British practice
often centered about the C-series families, using circular-arc or parabolic-arc
camberlines. As design requirements began to favor transonic operation, double-
circular-arc blades became popular. The performance characteristics of these air-
foil families are well understood due to extensive experimental cascade testing,
much of which is available in the literature.
In recent years, use of blades designed for a prescribed surface velocity distri-
bution or blade loading style, instead of for predefined airfoil families has
become popular. Often, inverse design methods that predict the blade shape
required for the desired blade loading are used. As the relation between blade
shape and preferred loading styles became better understood, it also became
common to use conventional or direct analysis methods in a trial-and-error mode
to arrive at the same result. These airfoils have been referred to as prescribed
velocity distribution (PDF) blades (Cumpsty, 1989), even though the term con-
trolled diffusion airfoils is probably more common today. Although the literature
offers general guidelines for these designs, the actual airfoil designs in use are
proprietary. In general, the performance characteristics of these airfoils are well
known only to the organizations that developed them.
As discussed in the preface to this book, this situation posed a significant com-
plication to the goal of providing a complete description of the working design
and analysis system. It was quickly recognized that it is no longer possible to
write a book that can be directly applied to all of the many proprietary designs in
use today. But this is not considered to be a serious limitation. In this writer’s
experience, the process of adapting classical blade performance prediction mod-
els to a more modern controlled diffusion airfoil design is not particularly diffi-
cult, assuming the performance characteristics of the airfoil are known.
This chapter provides a complete description of the more commonly used tra-
ditional airfoil families, and Chapter 6 provides a detailed description of the per-
formance modeling for these same airfoil families. This ensures that this book is
at least complete in the context of classical axial-flow compressor technology.

General concepts from the literature used to guide the development of controlled
diffusion airfoils are also briefly reviewed in this chapter.
NOMENCLATURE
a = distance along chord to the point of maximum camber

b = distance normal to the chord line to the point of maximum camber
Cl0 = isolated airfoil lift coefficient
c = chord length
d = length defined in Eq. (4-24)
i = incidence angle
o = blade throat opening
R = circular-arc radius of curvature
s = blade pitch (spacing)
tb = blade maximum thickness
x = coordinate along the chord
y = coordinate normal to the chord
yC = y coordinate at the origin of RC
α = angle of attack
β = flow angle relative to axial direction
χ = blade angle relative to the chord line
δ = deviation angle
γ = stagger (setting) angle
κ = blade angle relative to the axial direction
θ = camber angle
σ = solidity
φ = parameter defined in Eq. (4-30)
Subscripts
C = camberline parameter
L = blade lower or pressure surface parameter
U = blade upper or suction surface parameter
1 = blade leading edge parameter
2 = blade trailing edge parameter
4.1 CASCADE NOMENCLATURE

Figures 4-1 and 4-2 illustrate the basic parameters used to describe axial-flow com-
pressor blades and cascades. Blades are defined by a mean camberline, y(x), upon
which a profile or thickness distribution, tb(x), is imposed. The angles between
slopes to the camberline and the chord line at the leading and the trailing edges are
designated as χ1 and χ2, respectively. The blade camber angle is defined as
θ = χ1 + χ 2 (4-1)

Axial-Flow Compressor Blade Profiles • 61
FIGURE 4-1 Basic Airfoil Geometry
FIGURE 4-2 Basic Cascade Geometry

The pitch or the spacing between adjacent blades, s, and the chord length, c,
define the cascade solidity, σ, by
σ = c/ s (4-2)
The angle between the chord line and the axial direction is referred to as the stag-
ger angle, or setting angle, γ. The angle between the inlet velocity vector, W1, and
the chord line of the staggered blade is called the angle of attack, α. The flow
angle with respect to the axial direction will be designated as β1. The angles
between slopes to the camberline and the axial direction at the leading and the
trailing edges will be designated as κ1 and κ2, respectively. Similarly, the flow
angle at the blade trailing edge will be designated as β2. The flow incidence angle,
i, the deviation angle, δ, and the angle of attack, α, are defined as
i = β1 − κ1 (4-3)
δ = β2 − κ 2 (4-4)
α = β1 − γ (4-5)
This nomenclature is directly applicable to blades based on well-defined camber-

lines such as the circular-arc and parabolic-arc camberlines typical of British
practice. American practice has often been based on blades derived from NACA
aircraft wing airfoils, which typically have infinite camberline slopes at the lead-
ing and trailing edges. In those cases, parameters such as κ, χ, θ, i and δ lose sig-
nificance unless a suitable approximate or reference camberline is used to define
them. Common practice has been to use an equivalent circular-arc camberline as
a reference for the NACA 65-series blades (Johnsen and Bullock, 1965). This
writer has made similar use of an equivalent parabolic-arc camberline for the
NACA A4K6 63-series guide vanes (Dunavant, 1957), where the point of maxi-
mum camber is not at mid-chord.
Construction of blades from the base camberline and profile has occasionally
been a source of confusion. Profile thickness distributions are normally supplied
for zero-camber blades as a function of distance along the chord, which also is
the camberline for that case. When imposing a profile on a blade with camber,
the thickness distribution data should be interpreted in terms of dimensionless
distance along the camberline rather than along the chord line.
4.2 NACA 65-SERIES PROFILE
The NACA 65-series blades are derived from NACA aircraft wing airfoils designed
for approximately uniform loading. The original aircraft wing airfoil was not
structurally suitable for the compressor cascade application. There are a number
of adaptations of the original profile thickness distribution in use. The profile
reported by Emery et al. (1958) is representative of the basic NACA 65-series cas-
cade profile. Table 4-1 provides the thickness distribution from that reference.
Even that profile has been structurally suspect due to its sharp trailing edge, as

Table 4-1 Dimensionless Data for Standard Axial-Flow Compressor Blade Types
NACA 65-(10)10 Series NACA A4K6 63 Series C.4 Series
x/c ~ % y/c ~ % tb/c ~ % y/c ~% tb/c ~ % tb/c ~ %

0 0 0 0 0 0
0.5 0.250 1.544 0.376 —— ——
0.75 0.350 1.864 —— —— ——
1.25 0.535 2.338 0.792 1.542 3.30
2.5 0.930 3.480 1.357 2.114 4.54
5.0 1.580 4.354 2.248 2.924 6.16
7.5 2.120 5.294 —— —— 7.24
10 2.585 6.080 3.531 4.020 8.04
15 3.365 7.332 4.420 4.772 9.10
20 3.980 8.286 5.040 5.312 9.66
25 4.475 9.006 5.458 5.682 ——
30 4.860 9.520 5.710 5.908 10.0
35 5.150 9.848 5.824 6.000 ——
40 5.355 9.992 5.820 5.942 9.78
45 5.475 9.926 5.713 5.754 ——
50 5.515 9.624 5.516 5.446 9.14
55 5.475 9.060 5.239 5.034 ——
60 5.355 8.292 4.891 4.602 8.10
65 5.150 7.364 4.479 4.170 ——
70 4.860 6.312 4.011 3.740 6.74
75 4.475 5.168 3.492 3.308 ——
80 3.980 3.974 2.922 2.876 5.08
85 3.365 2.770 2.308 2.444 ——
90 2.585 1.620 1.642 2.014 3.20
95 1.580 0.612 0.912 1.582 2.12
100 0 0 0 0 0
RLE/c ~ % —— 0.687 —— 0.297 1.20
RTE/c ~ % —— 0 —— 0.600 0.60
well as it being very thin toward the trailing edge. Kovach and Sandercock (1961)
describe a more satisfactory modification for use in compressors. They use the
basic distribution from Table 4-1 up to 60% of the chord. Then the thickness is
varied linearly to match a trailing edge radius equal to 0.8% of chord. It is likely
there are many other variants on the 65-series profile that are in use for reasons of
structural integrity. The base profile has its maximum thickness at 40% of chord.
NACA 65-series airfoils are designated by their lift coefficient and maximum
thickness-to-chord ratio. The lift coefficient in tenths appears first in parentheses,
followed by the thickness-to-chord ratio as a percentage. Hence a 10% thick airfoil
with a lift coefficient of 1.5 is designated as NACA 65-(15)10. For lift coefficients
less than one, the parentheses may be omitted. The base camberline is defined for
a lift coefficient of 1.0 and is supplied in Table 4-1 (Emery et al., 1958). Simply mul-
tiply these coordinates by the lift coefficient to create other camberlines. Hence,
the base airfoil for the NACA 65-series is the NACA 65-(10)10. Figure 4-3 illustrates
this airfoil, using the base thickness of the NACA 65-010 from Table 4-1.

FIGURE 4-3 The NACA 65-(10)10 Airfoil
As noted, the slope of the NACA 65-series camberline becomes infinite at the
leading and trailing edges. For this reason, experimental data from cascade testing
are normally expressed in terms of angle of attack and fluid turning instead of inci-
dence angle and deviation angle. It is now accepted practice to define effective inlet
and discharge blade angles using an effective circular-arc camberline. The circular-
arc is defined as that which passes through the end points and the point of maxi-
mum camber at mid-chord (Johnsen and Bullock, 1965). Figure 4-4 shows a
comparison of the equivalent circular-arc camberline with the NACA 65-(12) cam-
berline. The construction of the circular-arc camberline reviewed in the next section
can be used for that purpose. Figure 126 of Johnsen and Bullock (1965) provides a
FIGURE 4-4 Equivalent Circular-Arc Camberline

graphical relation between the effective camber angle and the lift coefficient. As will
be seen in the next section, Clo and the equivalent θ can be related analytically by
tan(θ / 4) = 0.1103 Clo (4-6)
4.3 CIRCULAR-ARC CAMBERLINE
The circular-arc camberline is commonly used in conjunction with the British

C.4 series blade profile. It is also the camberline used for the double-circular-arc
profile, and is reported to be used in place of the NACA 65-series camberline,
when using the NACA 65-series profile (Cumpsty, 1989). As mentioned in the pre-
vious section, it is commonly used as an effective camberline for the NACA 65-
series blades to provide a meaningful definition of the leading and trailing edge
blade angles. This camberline is completely defined by the camber angle, θ, and
chord length, c. Figure 4-5 illustrates its construction. The radius of curvature,
RC, of the camberline is given by
FIGURE 4-5 Circular-Arc Camberline Construction

c / 2 = RC sin(θ / 2) (4-7)
The coordinates of the origin of the radius of curvature are (0, yC), where
yC = − RC cos(θ / 2) (4-8)
Then, for any x from – c / 2 to c / 2,
y = yC + RC2 − x2 (4-9)
Using Eqs. (4-7) and (4-8), the camberline coordinate, y(0), at mid-chord can be
expressed as
2 y(0) / c = [1 − cos(θ / 2)] / sin(θ / 2) = tan(θ / 4) (4-10)
which is the basis for Eq. (4-6), where y(0) / c = 0.05515 is given in Table 4-1 and
multiplied by Cl0 to obtain the value for any lift coefficient.
4.4 PARABOLIC-ARC CAMBERLINE

The parabolic-arc camberline is also used with the British C.4 profile and can be
used with other profiles as well. This writer has used it as an equivalent camber-
line to define effective blade angles for the NACA A4K6 63-series guide vane cam-
berline. The parabolic-arc allows a more general blade loading style than the
circular-arc. Front-loaded, mid-loaded and rear-loaded blades are all possible,
depending on where the point of maximum camber is located. Figure 4-6 illus-
trates this blade style. The point of maximum camber is located at x = a and y = b,
which provides the basic definition of the camberline. The basic constraints to be
satisfied are
y(0) = 0 (4-11)
y( c) = 0 (4-12)
y( a) = b (4-13)
y'( a) = 0 (4-14)
The camberline is generated using the general second-order equation for a

parabola that is given in many standard mathematics references. It can be
written as
Ax2 + 2 AE xy + By2 + Cx + Dy + E = 0 (4-15)
There appears to be a problem since we have five coefficients but only four con-
straints. However, one of the coefficients is arbitrary, e.g., we can divide through

FIGURE 4-6 The Parabolic-Arc Camberline
the equation by B and eliminate it. Hence Eqs. (4-11) through (4-14) are suffi-
cient to determine all coefficients in Eq. (4-15). The algebra is very tedious, but
the result can be shown to be
c − 2a ( c − 2a)2 2 c2 − 4ac
x2 + xy + 2
y − cx − y=0 (4-16)
b 4b 4b
Normally, specification of the blade camber angle, θ, or the blade angles, χ1 and
χ2, is preferred. Differentiating Eq. (4-16) and evaluating the derivatives at x = 0
and x = c yields
tanχ1 = 4b / (4a − c) (4-17)

tanχ 2 = 4b / (3c − 4a)
(4-18)
Equations (4-1) and (4-16) through (4-18) can be combined to yield.
b / c = { 1 + (4 tanθ )2 [a / c − ( a / c)2 − 3 / 16] − 1} / (4 tanθ ) (4-19)
This defines the parabolic-arc camberline in terms of camber and the ratio, a/c.
For compressor blades, it is reasonable to restrict the leading and trailing edge
angles to be less than 90°, i.e.,

FIGURE 4-7 The C.4 and 65-Series Profiles
0.25 < a / c < 0.75s (4-20)
For any value of x, Eq. (4-16) can be solved for y as a standard quadratic equa-
tion. This approach will be singular for a/c = 0.5, since two of the terms drop out.
But that special case is a simple and direct solution. Both cases can be treated by
a numerical recursion equation of the form
( c − 2a)2 c − 2a c2 − 4ac
y = x( c − x) / [ y+ x− ]
4 b2 b 4b (4-21)
Simply start with y = 0 and repeatedly solve this equation to converge on the cor-
rect value of y.
4.5 BRITISH C.4 PROFILE

The C.4 profile is one of several profiles in the British C-series (Howell, 1942).
The C.4 series received the most attention in the literature relative to its perform-
ance characteristics (e.g., Johnsen and Bullock, 1965). Its base thickness distri-
bution is tabulated in Table 4-1. Figure 4-7 shows an overlay of the C.4 and the
NACA 65-series profiles. The C.4 profile is thicker toward the leading edge and
has its maximum thickness at 30% of chord, compared to 40% for the 65-series
profile. This would be expected to make it less effective for higher Mach number
applications, but it would normally offer advantages relative to structural
integrity. Similar comments apply to the trailing edge region. But, as noted, the
65-series profile is usually modified in that region. This profile is normally
applied to circular-arc or parabolic-arc camberlines. It is reported that a later
series, the C.7 profile, has seen more use in compressors and has many features
in common with the 65-series profile (Cumpsty, 1989). C-series profiles are desig-
nated by a code giving the tb, profile, θ, camberline and a/c. Hence, a 10C4/20P40
blade is a 10% thick C.4 profile with a 20° camber angle using a parabolic-arc
camberline with a/c = 0.4. A 10C4/20C50 blade would be similar, but with a cir-
cular-arc camberline. This writer has not had direct experience with this profile,

but it is well-established and empirical performance prediction models do exist

for its application. On the other hand, there really is no reason why the 65-series
profile cannot be applied to the circular-arc and parabolic-arc camberlines. That
approach would be expected to yield better performance, particularly at higher
Mach numbers.
4.6 DOUBLE-CIRCULAR-ARC PROFILE

The double-circular-arc profile is constructed with both surfaces formed by cir-
cular-arcs, which blend with a nose radius, r0, applied at both the leading and
trailing edges. Designate the lower and upper surface arc radii of curvature as RL
and RU, respectively, as illustrated in Fig. 4-8. The construction of the camber-
line, with a radius of curvature, RC, has been described in Section 4.3. Here, con-
struction of the upper surface of the profile will be illustrated. Construction of
the lower surface is quite similar, except that certain parameters, such as tb and
r0, are assigned negative values. The distance, ∆xU, from mid-chord to the center
of the nose radius at the trailing edge is given by
∆xU = ( RU − r0 ) sin(θU / 2) = c / 2 − r0 cos(θ / 2) (4-22)
where θ is the blade camber angle and θU is shown in Fig. 4-8. The distance, ∆yU,
is given by
FIGURE 4-8 Double-Circular-Arc Profile

∆yU = RU − y(0) − tb / 2 + r0 sin(θ / 2) = RU − d (4-23)
where the camberline coordinate, y(0), is given by Eq. (4-10) and d is defined as
d = y(0) + tb / 2 − r0 sin(θ / 2) (4-24)
The Pythagorean theorem applied to the right triangle in Fig. 4-8 requires
[ RU − r0 ]2 = [ RU − d]2 + [c / 2 − r0 cos(θ / 2)]2 (4-25)
After some tedious algebra, this yields
d2 − r02 + [c / 2 − r0 cos(θ / 2)]2 (4-26)

RU =
2( d − r0 )
The upper surface circular-arc extends through polar angles from -θU / 2 to θU / 2,
constructed using the radius of curvature, RU, and the location of its origin at
x = 0 and y = y(0) + tb / 2 – RU. The leading and trailing edge radii are constructed
about their centers at y = r0 sin(θ / 2) and x = ± [c / 2 – r0 cos(θ / 2)] to blend with
the circular-arc.
4.7 NACA A4K6 63-SERIES GUIDE VANE PROFILE
Dunavant (1957) provides design and application data for a very effective vane pro-
file for use as inlet guide vanes. This vane has excellent flow guidance and a wide
incidence operating range. The camberline is developed by combining a front-
loaded (A) profile with Cl0 = 0.4 and a uniform-loaded (K) profile with Cl0 = 0.6,
which is designated as the A4K6 camberline corresponding to Cl0 = 1. This is com-
bined with the 6% thick NACA 63-series profile as the base guide vane geometry.
The base camberline coordinates and thickness distribution are listed in Table 4-1
and illustrated in Fig. 4-9. Similar to the 65-series blades, the camberline coordi-
nates can be scaled directly by lift coefficient to alternate camberlines. Similarly,
the thickness distribution can be scaled to other values from the base 6% thick pro-
file. The general vane designation is 63-(Cl0 A4K6)nn, where nn is the maximum
thickness as percent of chord. The maximum thickness of this vane is at 35% of
chord. The location of the point of maximum camber can be estimated by interpo-
lation to be at approximately a / c = 0.375 with b / c = 0.0583 for Cl0 = 1. As is the
case with the NACA 65-series camberline, the leading and trailing edge camberline
slopes are infinite. Here, an equivalent parabolic-arc camberline can be used to
allow viable definitions of leading and trailing edge blade angles, incidence angle
and deviation angle. Equation (4-19) can be solved for camber angle to yield
b/c
tan θ = (4-27)
2[a / c − ( a / c) − 3 / 16 − ( b / c)2 ]
2

FIGURE 4-9 NACA 63-(10A4K6)06
Hence, for the A4K6 camberline,
291.5Cl0
tan θ =
468.75 − (5.83Cl0 )2 (4-28)
which yields θ = 33.8° at Cl0 = 1. A simpler approximation is obtained by general-

izing the equivalent circular-arc camberline conversion of Eq. (4-6) by dividing
the right-hand side by 2a / c to yield.
tan(θ / 4) = 0.05515Cl0 / ( a / c) (4-29)
which yields θ = 33.5° at Cl0 = 1. As long as only the NACA 65-series and A4K6
camberlines require conversion between θ and Cl0, this equation can be applied
to either one. It also can be easily inverted for circular-arc and parabolic-arc cam-
berlines to permit application of empirical blade performance correlations given
as a function of Cl0 to those camberline types.
4.8 CONTROLLED-DIFFUSION AIRFOILS
The standard blade profiles described in this chapter have been used extensively
for axial-flow compressors. They are well understood, reliable and can yield
excellent performance when properly applied. But, in recent years, many investi-
gators have explored alternatives offering better Mach number range and higher
efficiency. These are often referred to as controlled diffusion airfoils, since the
design of the profiles is based on producing carefully controlled blade surface
Mach number distributions.
Hobbs and Weingold (1984) and Dunker et al. (1984) have reviewed the basic
design strategy. They indicate that the key features are:
• A continuous acceleration along the suction surface and near the

leading edge to avoid laminar boundary layer separation or prema-
ture separation.

• The peak Mach number should not exceed 1.3 to avoid shock-wave-
induced separation.
• Carefully controlled deceleration along the suction surface from the
peak Mach number to avoid turbulent boundary separation ahead of
the trailing edge.
• A nearly constant subsonic Mach number distribution on the pressure
surface.
Figure 4-10 is a qualitative schematic of the type of Mach number distribution

employed. A key feature is to avoid shock wave–boundary layer interaction, such
that boundary layer analysis can be used effectively in establishing the desired
controlled diffusion characteristic along the suction surface. Initially, inverse
techniques, which compute blade geometry from specified Mach number distri-
butions, were used. Once the basic concept was clarified, it became possible to
use direct methods to iteratively refine the geometry to achieve the desired aero-
dynamic characteristics. Typical controlled diffusion profiles (Hobbs and Wein-
gold, 1984) appear to be more robust than the 65-series in the forward portion of
the profile, followed by a relatively thin aft region of almost constant thickness.
But that is not necessarily a general conclusion. This is really a design concept to
produce specific, proprietary profile designs. It has not resulted in a standard air-
foil family that can be employed and analyzed in a general sense. Some stan-
dardization is clearly possible as evidenced by the fact that controlled diffusion
FIGURE 4-10 Controlled Diffusion Airfoil

airfoils have found favor for industrial axial-flow compressors, where unique
profile development for each compressor is not economically feasible. But to
employ this concept, specific profiles of the desired characteristics must be at
least initially designed. And performance prediction models must also be estab-
lished, since those used for standard profiles are unlikely to be adequate.
4.9 BLADE THROAT OPENING

The blade throat opening, o, is the minimum distance between adjacent blades,
as illustrated in Fig. 4-11. It is an important parameter when conducting an aero-
dynamic performance analysis. The throat opening governs the onset of local
flow choking within the blade passage. At sufficiently high inlet Mach number
levels, this will define the maximum flow capacity that the compressor can pass.
The best approach to determine the throat opening is to define adjacent blades
using the stagger angle, camberline and profile coordinates to locate the mini-
mum distance between the blades. For that purpose, this writer uses a computer
database containing the information in Table 4-1. Then, before an actual per-
formance analysis, the throat openings along the span of all blade rows are com-
puted by a simple trial-and-error process. Typically, about 50 points are
distributed along both blade surfaces and the minimum distance is easily
located. This process is carried out in the context of a two-dimensional cascade
with the correct solidity. The ratio of throat-opening-to-pitch, o / s, is computed.
Conformal mapping shows that this ratio is applicable to the annular cascade of
a compressor (e.g., Aungier, 2000, Chapter 7). There is no need to attempt some-
thing more sophisticated, since this is a rather trivial problem for a computer.
FIGURE 4-11 Throat Geometry

Plus, these data are added to the basic problem input file, so throat openings are
only computed once for each compressor to be analyzed. The same process can
be used for any blade type, such as the controlled diffusion airfoils discussed in
the previous section. It is only necessary to have the camberline and profile coor-
dinates available from a database.
It is also possible to approximate throat openings with reasonable accuracy
using an empirical correlation. This writer developed a throat-opening correla-
tion for NACA 65-series blades from the carpet plots of Dunavant et al. (1955). A
modified stagger angle parameter, φ, is defined as
φ = γ (1 − 0.05C1l0.5 ) + 5C1l0.5 − 2 (4-30)
and the ratio of throat opening-to-pitch, o / s, is estimated from
σ (4-31)
o / s = [(1 − tb σ / c) cos φ ]
Results from this correlation are compared to actual values of o/s for typical
NACA 65-series blades in Fig. 4-12. This correlation has been found to be reason-
ably accurate for other blade types also. As a somewhat extreme example, it is
compared to actual throat openings for a C4-series profile that is imposed on a
parabolic-arc camberline with a/c = 0.4 in Fig. 4-13. Equation (4-29) was used to
FIGURE 4-12 Throat-Opening Correlation

FIGURE 4-13 Extended Throat-Opening Correlation
relate lift coefficient and camber angle for this comparison. While both the pro-
file and camberline are quite different from the NACA 65-series blades for which
the correlation was developed, accuracy is still quite good except at very low stag-
ger angles in combination with high camber angles. That small region of inaccu-
racy is unlikely to be encountered in a real compressor application. Nevertheless,
it is prudent to use the actual throat openings for performance analysis applica-
tions. This removes the uncertainty as to applicability of the empirical correla-
tion to the specific blades used. It also ensures accurate treatment of special
blade types for which the correlation is not likely to apply. A common example is
an inlet guide vane row, where the blade throat will normally be located at the
trailing edge, similar to a turbine blade row. A compressor blade throat correla-
tion is unlikely to handle this situation accurately. While inlet guide vane choking
is not common, it could be a factor if stagger angles are set too high. This could
produce an unexpected limit on the compressor’s flow capacity unless the prob-
lem is handled correctly.
4.10 STAGGERED BLADE GEOMETRY
Once the camberline and profile coordinates for any of the proceeding blade
types have been generated along the chord, the geometry of the staggered blade
in the cascade is obtained by a simple rotation of coordinates to the stagger
angle, γ. The staggered blade inlet and discharge angles are given by

κ1 = χ1 + γ (4-32)
κ 2 = γ − χ2 (4-33)
From Eqs. (4-1), (4-32) and (4-33), it follows that
θ = κ1 − κ 2 (4-34)
For the circular-arc or the NACA 65-series equivalent circular-arc approximation,

it is easily shown that
χ1 = χ 2 = θ / 2 (4-35)
γ = (κ1 + κ 2 ) / 2 (4-36)
For the parabolic-arc or equivalent parabolic-arc camberlines, Eqs. (4-17)

through (4-19) can be used to compute χ1 and χ2 as a function of θ and a/c for use
in Eqs. (4-32) through (4-34). Indeed, it is easily shown that the same approach
can be used for circular-arc and equivalent circular-arc camberlines, since Eqs.
(4-17) through (4-19) are equivalent to Eqs. (4-35) and (4-36) when a/c = 0.5.

Chapter 5
TWO-DIMENSIONAL
BLADE-TO-BLADE FLOW
THROUGH CASCADES OF BLADES
Prediction of the flow through cascades of blades is fundamental to all aspects of

axial-flow compressor aerodynamic design and analysis. Although the flow
through the annular cascades of blades in a compressor is really a three-dimen-
sional flow problem, there are many advantages to considering the simpler prob-
lem of two-dimensional flow in cascades. It offers a very natural view of cascade
fluid dynamics to make it easier for designers to develop an understanding of the
basic flow processes involved. Indeed, very simple two-dimensional cascade flow
models were used in this educational role long before computational methods
and computers had evolved enough to produce useful design results. Today,
blade-to-blade flow analysis is a useful design and analysis tool that provides rea-
sonable approximations to many problems of interest. Inviscid blade-to-blade
flow analysis addresses the general problem of two-dimensional flow on a stream
surface in an annular cascade, as discussed in Chapter 3. Two-dimensional
boundary layer analysis can be used to approximate viscous effects. Although
very useful, this approach does have limitations. It ignores secondary flows that
develop from the migration of low momentum boundary layer fluid across the
stream sheet. It also loses accuracy when significant flow separation is present.
This chapter considers theoretical methods to model two-dimensional blade-to-
blade flow. The next chapter considers a different approach to the problem,
where empirical models are derived from two-dimensional cascade test data. The
methods presented in this chapter are basically the same as those presented in
Aungier (2000), but are adapted to conventions more commonly used for axial-
flow compressors. Aungier (2000) provides detailed guidance relative to imple-
mentation of the methods in numerical analyses that is not repeated here.
Readers interested in those details should consult the original reference.
NOMENCLATURE
A = area and a function used in Eq. (5-22)

a = sound speed and a parameter in Eq. (5-52)

B = function in Eq. (5-23)

C = absolute velocity and a function in Eq. (5-24)
cf = skin friction coefficient
cp = specific heat at constant pressure
D = function in Eq. (5-25)
E = function in Eq. (5-26) and the entrainment function
→
n = a unit vector
H = total enthalpy and shape factor of Eq. (5-129)
Hk = kinematic shape factor, Eq. (5-130)
H1 = shape factor, Eq. (5-128)
h = enthalpy
I = rothalpy
K = shape factor of Eq. (5-122)
M = Mach number = C/a
→
r = unit vector normal to an area element
P = pressure
Q = velocity component of Eq. (5-76)
Reθ = momentum thickness Reynolds number
r = radius
S = blade pitch = r (θ1 – θ0)
T = temperature
t = time
u = general function
V = volume
v = general function
x = distance along a wall
y = distance normal to a wall
α = angle of attack = βin – γ
β = flow angle and coordinate angle (Fig 5-4)
γ = blade stagger angle
δ′ = density thickness
δ* = displacement thickness
δE = energy thickness
δh = enthalpy thickness
δu = velocity thickness
η = dimensionless tangential coordinate (Fig. 5-3)
θ = polar angle and momentum thickness
Λ = shape factor of Eq. (5-113)
µ = stabilizing term and viscosity
ξ = streamwise coordinate (Fig. 5-3)
ρ = density
Φ = stabilizing term

Two-Dimensional Blade-to-Blade Flow Through Cascades of Blades • 79
φ = stream surface slope angle with axis

ψ = stream function
τw = wall shear stress
ω = rotation speed
ω– = total pressure loss coefficient
Subscripts
in = cascade inlet condition

out = cascade discharge condition
q = component normal to ξ coordinate
ξ = component tangent to ξ coordinate
0 = parameter on blade surface θ0 (Fig. 5-3)
1 = parameter on blade surface θ1 (Fig. 5-3)
Superscripts
(η) = relative to the η direction
(ξ) = relative to the ξ direction
′ = relative condition and first derivative
″ = second derivative
5.1 THE BLADE-TO-BLADE FLOW PROBLEM
Figures 5-1 and 5-2 illustrate the basic problem to be considered. Figure 5-1
shows a schematic of a streamline pattern on a stream sheet between adjacent
blades in a cascade of blades. It will be sufficient to consider a single passage
between two blades, since the flow in all blade passages will be assumed to be
identical. As discussed in Chapter 3, a stream sheet is a thin annular passage
bounded by two stream surfaces, where a stream surface has no velocity compo-
nent normal to it, i.e., it has no mass flow across it. A schematic of a stream sheet
that might be used is illustrated in Fig. 5-2. Stream sheets will be assumed to be
axisymmetric. Although that is not a necessary assumption, it does greatly sim-
plify the problem. This chapter considers the stream sheet geometry to be known.
In later chapters, techniques to define the stream sheet geometry to support these
blade-to-blade flow methods will be covered, including generalization of the
process into a quasi-three-dimensional flow analysis.
The approach to be used is to generate a time-steady inviscid flow analysis fol-
lowed by a blade surface boundary layer analysis to approximate viscous effects.
It is assumed that the rothalpy and entropy are both constant at the upstream
boundary. From the time-steady form of Eqs. (3-25), (3-28) and (3-29), it can be

FIGURE 5-1 Blade-to-Blade Plane Flow
FIGURE 5-2 Blade-to-Blade Stream Sheet

seen that these assumptions require rothalpy and entropy to be constant over the
entire stream sheet. The steady form of Eqs. (3-28) and (3-29) governs conserva-
tion of momentum in the stream sheet. It is seen that these two equations are
identical for the present problem. Indeed, from Eqs. (3-1), (3-29) and (3-63) it is
easily shown that the component of the absolute vorticity normal to the stream
sheet must be zero. This reduces the problem to a classical potential flow prob-
lem governed by conservation of mass, Eq. (3-21), and the irrotational flow con-
dition, Eq. (3-29).
But a major complication encountered in blade-to-blade flow analysis arises
from the mathematical character of the governing equations of motion for time-
steady inviscid flow. When the flow is everywhere subsonic (W < a), the governing
equations are elliptic in form. This presents a classical boundary value problem,
where the solution is completely determined by conditions imposed on the
boundaries of the solution domain. But when the flow is supersonic (W > a), the
governing equations are hyperbolic in form. This type of problem requires some
type of marching solution, such as the method of characteristics. When super-
sonic flow is present, there is usually subsonic flow present in the solution
domain. The mixed subsonic-supersonic flow problem requires two solution
techniques that must be matched in some fashion. It is now common practice to
consider time-unsteady flow in those cases, since the governing equations for
unsteady inviscid flow are hyperbolic for both subsonic and supersonic flow. This
solution technique is commonly called the time-marching or time-dependent
technique. The solution is simply advanced in time until it has reached essen-
tially a steady flow prediction. When the time-marching approach is used, the
simplifications that lead to the potential flow model are no longer present. For
example, Eq. (3-25) must be solved, since rothalpy can no longer be treated as a
constant on the stream sheet. Nor are the two momentum equations for the flow
in the stream sheet identical for the time-unsteady case. Thus, a potential flow
analysis can be used for subsonic flow, but a more general analysis will be needed
for a mixed subsonic-supersonic flow case. These more general solutions are
commonly referred to as Euler techniques, which include the time-marching
method. The basic characteristic of an Euler method is that conservation of
mass, energy and all relevant momentum equations are solved without simplifi-
cation, such as assuming isentropic flow.
5.2 COORDINATE SYSTEM AND VELOCITY COMPONENTS
Figure 5-3 illustrates the solution domain to be considered for the blade-to-blade
flow solution. It is convenient to use a coordinate transformation to define new
coordinates (ξ, η) such that the blade surfaces correspond to lines of constant η.
The new coordinates are given by
m
dm
ξ= ∫ cos β (5-1)
0
η = [θ − θ0 ] / [θ1 − θ0 ] (5-2)

FIGURE 5-3 Blade-to-Blade Solution Domain
This transforms the complex solution domain of Fig. 5-3 into a simple rectangu-
lar domain in (ξ, η) space where η varies from 0 to 1. The blade surface tangential
coordinates, θ0 and θ1, are illustrated in Fig. 5-3, and β is the angle between a tan-
gent to a constant η line and the meridional direction, i.e.,
 r∂θ 
tan β =   = tan β0 + [tan β1 − tan β0 ] η (5-3)
 ∂m η
Outside the blade passage, θ0 and θ1 are somewhat arbitrary, except that θ1 = θ0 +
2π / N, where N is the number of blades. Figure 5-4 shows an expanded view of a
basic control volume within the solution domain from Fig. 5-3. It also illustrates
special velocity components useful for developing the governing equations in (ξ,
η) space. Wξ and Wq are simply the velocity components parallel to and normal to
a constant η line, respectively. The following equations relate these velocity com-
ponents to the usual Wm and Wθ velocity components.
Wξ = Wm cos β + Wθ sin β (5-4)

Wq = Wθ cos β − Wm sin β (5-5)
Wm = Wξ cos β − Wq sin β (5-6)
Wθ = Wq cos β + Wξ sin β (5-7)

FIGURE 5-4 A Control Volume
5.3 POTENTIAL FLOW IN THE BLADE-TO-BLADE PLANE

As discussed previously, potential flow analysis on a blade-to-blade stream sur-
face involves solution under the assumption that the absolute flow is time-steady,
inviscid and adiabatic, with rothalpy and entropy both constant on this surface.
No restrictions are imposed relative to variation of rothalpy and entropy normal
to the stream surface. In general, rothalpy and entropy will vary from stream sur-
face to stream surface, which is quite compatible with the present analysis. The
governing equations are conservation of mass, Eq. (3-21), and the irrotationality
condition, Eq. (3-29). The steady form of these equations could be transformed
directly into the (ξ, η) space by developing transformations for the derivatives
from Eqs. (5-1) and (5-2). But a more accurate numerical analysis is obtained by
a more fundamental development using the basic control volume shown in Fig.
5-4. Conservation of mass for this control volume can be stated as
 ρ bW   ρ bWq  
2∆m  
q
 − 
 cos β   cos β  m,η + ∆η  (5-8)
 m ,η − ∆ η 
[ ]
+2∆η ( Sρ bWm )m − ∆m,η − ( Sρ bWm )m + ∆m,η = 0
where the subscripts identify specific grid points on the control volume of Fig.
5-4, and the tangential spacing, S(m), is defined by

S = r (θ1 − θ0 ) (5-9)
Taking the limit as ∆m and ∆η approach zero, Eq. (5-8) reduces to the following
continuity equation.
∂  ρ bWq  ∂( Sρ bWm )
 + =0 (5-10)
∂η  cos β  ∂m
Developed in this fashion, it can be noted that the continuity equation contains
coordinates m and η and velocity components Wq and Wm from two different
coordinate systems. While perhaps a little unusual, it does result in a more pre-
cise statement of conservation of mass for use in a numerical analysis. Numerical
approximations to the governing equations will apply them to a finite control vol-
ume. More accurate numerical approximations will result from using the control
volume approach to develop the equations instead of using a mathematical trans-
formation of the derivatives in Eq. (3-21). The condition for irrotational absolute
flow in the stream surface requires that the component of the absolute vorticity
normal to the stream sheet be zero, i.e.,
r r r r r r r
en ⋅ (∇ × C ) = en ⋅ [∇ × (W + rω eθ )] = 0 (5-11)
Stokes theorem is a convenient method to impose this condition on the control

volume. Stokes theorem relates the line integral of the velocity about any closed
path to the integral of the normal component vorticity over the area enclosed by
the path. In the rotating coordinate system, Stokes theorem can be applied as
r r r r r r r r
∫ W ⋅ dr = ∫ [en ⋅ (∇ × W )da = − ∫ en ⋅ (∇ × rω eθ )da (5-12)
C A A
When applied to the control volume, this yields
 W   Wξ  
ξ
2∆m   − 

 cos β   cos β  m,η + ∆η 
 m ,η − ∆η  (5-13)
S ∂r 2ω
[
+2∆η ( SWθ )m − ∆m,η − ( SWθ )m + ∆m,η ] = 4∆η ∆m
r ∂r
Taking the limit as ∆m and ∆η approach zero, Eq. (5-13) reduces to
∂  Wξ  ∂( SWθ )
 = + 2Sω sin φ
∂η  cos β  ∂m (5-14)
where φ is the stream sheet angle with the axial direction as shown in Fig. 5-2
Hence,

∂r
sin φ = (5-15)
∂m
A stream function, ψ, is defined by
∂ψ
˙
m = − ρ b(Wθ − Wm tan β ) (5-16)
∂m
∂ψ
ṁ = Sρ bWm (5-17)
∂η
·
where m is the stream sheet mass flow rate. Hence, the velocity components are
given by
˙ ∂ψ
m
Wm = (5-18)
Sbρ ∂η
m˙  tan β ∂ψ ∂ψ 
Wθ =  S ∂η − ∂m  (5-19)
bρ  
By substituting Eqs. (5-5), (5-18) and (5-19) into Eq. (5-10), it is easily shown that
the definition of ψ identically satisfies the continuity equation. Note that ψ varies
from 0 to 1 as θ varies from θ0 to θ1 or as η varies from 0 to 1. Thus, both conser-
vation of mass and energy are satisfied, requiring solution of the irrotationality
condition only to predict the inviscid flow field. Introducing the stream function
into Eq. (5-14) yields the required equation.
∂ m ˙ (1 + tan 2 β ) ∂ψ m ˙ tan β ∂ψ 
 − 
∂η  Sbρ ∂η bρ ∂m 
(5-20)
∂ m ˙ ∂ψ 
˙ tan β ∂ψ mS
=  −  + 2Sω sin φ
∂m  bρ ∂η bρ ∂m 
This equation can be simplified to the form
∂2ψ ∂2ψ ∂2ψ ∂ψ ∂ψ

A − 2B +C +D +E = 2Sω sin φ (5-21)
∂η 2 ∂η∂m ∂m 2 ∂η ∂m
where
˙ / ( Sbρ cos 2β )
A(m, η) = m (5-22)
B(m, η) = m
˙ tan β / ( bρ ) (5-23)
C(m, η) = mS
˙ / ( bρ ) (5-24)

∂A ∂B
D(m, η) = − (5-25)
∂η ∂m
∂C ∂B
E(m, η) = − (5-26)
∂m ∂η
Boundary conditions for ψ(m, η) on the solution domain shown in Fig. 5-3 are
reasonably straightforward. On the blade surfaces,
ψ (m, 0) = 0 (5-27)
ψ (m,1) = 1 (5-28)
For the side boundaries outside of the blade passage, the periodicity condition is
used. Since the flow is identical in all blade passages, the flow field must repeat
in the tangential direction with a period of ∆η = 1. The periodic nature of the
flow can be used to extend the solution into adjacent passages such that points
on these side boundaries can be treated the same as interior points in the solu-
tion domain.
ψ (m, η + 1) = ψ (m, η) + 1 (5-29)

ρ(m, η + 1) = ρ(m, η) (5-30)
Wm (m, η + 1) = Wm (m, η) (5-31)
Wθ (m, η + 1) = Wθ (m, η) (5-32)
Conditions on the upstream and downstream boundaries may be assigned as uni-

form flow, basically requiring that ψ vary linearly in the tangential direction. But
reliability of the numerical analysis is improved if a less stringent boundary con-
dition is imposed. A good choice is to require that the flow angle be constant on
these boundaries. Then, if the geometry of the side boundaries is defined such
that β equals the local flow angle on the upstream and downstream boundaries,
the appropriate upstream and downstream boundary condition is
∂ψ ∂ψ
= cos β =0 (5-33)
∂ξ ∂m
as can be seen from Eq. (5-16). Typically, β will be assigned to vary uniformly
along the side boundary from the upstream boundary flow angle to the blade
leading edge blade angle, and analogously for the downstream boundary. Once
this distribution of β along the side boundaries is specified, simple integration of
Eq. (5-3) yields θ0(m) and θ1(m), noting that β0 = β1 outside the blade passage. A
potential flow analysis can be accomplished for virtually any specified flow
angles at the upstream and downstream boundary. But, in reality, these two flow
angles are not independent. A prediction of the downstream flow angle for any

upstream flow conditions is to be preferred. To accomplish this, some additional

constraint is required. Typically, the well-known Kutta condition is applied at the
trailing edge. This simply requires that the pressures on the two sides of the
blades must be equal at the trailing edge. This follows from the fact that there is
no longer a tangential blade force to sustain a pressure difference. It is equivalent
to imposing the following trailing edge condition.
[W (m, 0)]te = [W (m,1)]te (5-34)
Typically, iterative adjustment of the discharge flow angle is accomplished as part

of the solution process until a value that satisfies Eq. (5-34) is found.
Equation (5-21) is solved subject to the above boundary conditions while
treating the gas density field throughout the solution domain as constant,
starting with an initial guess for the density field. After the stream function is
predicted, Eqs. (5-18) and (5-19) are used to compute the velocity field. Then,
the density field is updated using an appropriate equation of state from Chap-
ter 2. The stream function is then recalculated, and the process continued until
convergence is achieved. A grid structure is developed over the solution
domain, as illustrated in Fig. 5-3. The spacing between nodes ∆m and ∆η is
constant in each direction. Unequal spacing can be used, but this writer’s expe-
rience has shown that the benefits of unequal spacing do not justify the added
complexity and reduced computational speed. Equation (5-21) is reduced to
linear form using standard finite–difference approximations for the deriva-
tives. If subscripts i and j are used to designate the ith meridional node and the
jth tangential node, the relevant finite-difference approximations for interior
points are
∂ψ ψ i +1, j − ψ i −1, j
=
∂m 2∆m (5-35)
∂ψ ψ i , j +1 − ψ i , j −1
=
∂η 2∆η (5-36)
2
∂ψ ψ i +1, j − 2ψ i , j + ψ i −1, j
=
∂m2 ( ∆m)2 (5-37)
2
∂ψ ψ i , j +1 − 2ψ i , j + ψ i , j −1
=
∂η 2 ( ∆η)2 (5-38)
∂ψ2 ψ i +1, j +1 − ψ i −1, j +1 − ψ i +1, j −1 + ψ i −1, j −1
=
∂m∂η 4∆m∆η (5-39)
Equations (5-35) through (5-39) are easily derived using truncated Taylor series
expansions of ψ as functions of m and η (Aungier, 2000). The terms D and E in
Eqs. (5-25) and (5-26) require approximations for the first derivatives at nodes on
the boundary. These can be developed in the same way as for interior points.
They are

∂ψ 4ψ i +1, j − 3ψ i , j − ψ i + 2 , j (5-40)
=
∂m 2∆m
∂ψ 4ψ i , j +1 − 3ψ i , j − ψ i , j + 2
= (5-41)
∂η 2∆η
∂ψ 3ψ i , j − 4ψ i −1, j + ψ i − 2 , j
= (5-42)
∂m 2∆m
∂ψ 3ψ i , j − 4ψ i , j −1 + ψ i , j − 2
= (5-43)
∂η 2∆η
Substitution of the finite-difference approximations into Eq. (5-21) for any inte-
rior node (i, j) yields a simple linear equation for ψ
ψ i , j + A˜ i , jψ i −1, j + B˜ i , jψ i +1, j + C˜ i , jψ i , j −1 + D˜ i , jψ i , j +1
(5-44)
+ E˜ i , j [ψ i +1, j +1 − ψ i +1, j −1 − ψ i −1. j +1 + ψ i −1, j −1] = Q˜ i , j
The coefficients in Eq. (5-44) are given by
 Ci , j Ei , j   2Ai , j 2Ci , j 
A˜ i , j = −  −   +  (5-45)
 ( ∆m)
2 2∆m   ( ∆η)
2
( ∆m)2 
 Ci , j Ei , j   2Ai , j 2Ci , j 
B˜ i , j = −  +   +  (5-46)
 ( ∆m)
2 2∆m   ( ∆η)
2
( ∆m)2 
 Ai , j Di , j   2Ai , j 2Ci , j 
C˜ i , j = −  −   +  (5-47)
 ( ∆η )2 2∆η   ( ∆η )2
( ∆m)2 
 Ai , j Di , j   2Ai , j 2Ci , j 
D˜ i , j = −  +   +  (5-48)
 ( ∆η)
2 2∆η   ( ∆η)
2
( ∆m)2 
 Bi , j   2Ai , j 2Ci , j 
E˜ i , j =    +  (5-49)
 2∆m∆η 
2
 ( ∆η) ( ∆m)2 
 2Ai , j 2Ci , j 
[
Q˜ i , j = − 2Sω sin φ ] 
 ( ∆η)
2
+
( ∆m)2 
 (5-50)
Numerical solution of the linearized stream function equation can be accom-

plished with a relaxation technique (Katsanis 1968, 1969) or a matrix method
(Smith and Frost, 1969; Aungier, 2000). This writer has used both methods and
has found the matrix method to be superior in both computational speed and
reliability. In the matrix method, Eq. (5-44) is used to develop equations for all
nodes, resulting in a matrix equation, where the major matrix is a square matrix
with the number of rows and columns equal to the number of nodes. It is a very
sparse matrix, having non-zero values only in a band about its diagonal, so that a

rather small number of values actually need to be stored in the computer’s mem-
ory. Aungier (2000) provides a rather detailed description of a very efficient
matrix solution for this problem.
After each solution for the stream function, the density field must be updated
using a new velocity field estimate from Eqs. (5-18) and (5-19). Since rothalpy is
constant on the stream sheet, Eqs. (3-11) and (3-13) require
h = H ′ − 12 W 2 = I + 12 ( rω )2 − 12 W 2 (5-51)
And, since entropy is also constant on the stream sheet, all thermodynamic prop-
erties can be calculated as a function of (h, s), using an appropriate equation of
state from Chapter 2. While conducting iterations to converge on a density field,
convergence on a discharge flow angle to satisfy the Kutta condition, Eq. (5-34), is
also accomplished. As long as the flow is subsonic, this procedure of lagging the
density solution one iteration behind the stream function solution offers excellent
numerical stability and rapid convergence. Once velocities greater than sonic
velocity are encountered, the solution will become unstable, and will almost
always diverge. Blade-to-blade flow problems involving local patches of super-
sonic flow are often encountered in axial-flow compressor analysis. The useful-
ness of a potential flow analysis is greatly increased if it is extended to be capable
of addressing these transonic flow cases. This can be accomplished by readjusting
the mass flow rate and speed, or the inlet total sound speed, such that the inlet
velocity triangle is the same as the actual problem but all velocities are subsonic.
After solving the subsonic flow problem, the streamline pattern is assumed to be
correct, and some type of streamline curvature numerical technique can be used
to calculate the flow for the actual inlet conditions. Katsanis (1969) is a good
example of this type of extension of a potential flow solution. The weakness in this
approach is that the resulting flow field will no longer satisfy the irrotationality
condition. Aungier (2000) presents a better technique, which can predict an irro-
tational flow field that conserves mass and involves local patches of supersonic
flow. The inlet total sound speed is adjusted to reduce all velocities to subsonic
values. At the same time, the stream sheet thickness distribution is adjusted such
that the resident velocity field, which conserves mass for the fictitious subsonic
problem, will also conserve mass for the actual transonic flow problem when the
actual inlet conditions and steam sheet thickness distribution are used. Thus,
when the subsonic flow solution is obtained, it is only necessary to accomplish a
final update of all thermodynamic data with the predicted velocity field to satisfy
all governing equations for the transonic flow case. Implementation of this exten-
sion to the analysis is straightforward, but depends to some degree on the equa-
tion of state used in the analysis. Aungier (2000) provides an illustration of its
implementation for the special case of a calorically and thermally perfect gas.
The stability of a numerical solution of the blade-to-blade flow problem can be
significantly influenced by the manner in which the grid structure is established
near the leading and trailing edges of the blades. Figure 5-5 illustrates two meth-
ods of locating the grid near the leading edge. In one case, the first nodes on the
blade surface are outside the passage, touching the blade at a single point. The
other case locates the first nodes on the blade surfaces inside the passage with a

FIGURE 5-5 Leading Edge Grid Structure
node on each blade surface. Experience has shown that the first method can
result in local numerical instability, while the second method almost never expe-
riences that problem. The cause of instability has been traced to the behavior of
the stagnation streamline, coupled with the finite-difference approximations for
derivatives with respect to η used at blade surface points. In the illustration in
Fig. 5-5, it is seen that the stagnation streamline passes between the blade surface
and the node next to the surface when the leading edge nodes are outside the pas-
sage. Since ψ has the same value on the stagnation streamline and the blade sur-
face, a difference approximation to the tangential derivative will provide a poor
estimate. And, without special logic in the solution, it is not obvious which direc-
tion the tangential difference approximation for the node on the blade should
use, i.e., Eq. (5-41) or (5-43). In the case illustrated, nodes to the right of the lead-
ing edge node should be used, i.e., Eq. (5-41). Minor changes in gas density at the
leading edge can induce the stagnation streamline to move, possibly even making
it shift to the opposite blade surface. It is not uncommon for this to result in an
oscillation on successive iterations, all occurring very local to this region, such
that a converged solution is never realized. Numerical damping, refining the grid
near the leading edge, etc., can alleviate this problem. But a simpler and more
effective approach is to move the leading edge nodes into the passage as illus-
trated in Fig. 5-5. This removes the ambiguity regarding the direction to be used
for the surface derivatives, and the local oscillation problem almost never occurs
when this is done. It is a very simple method to avoid the tendency toward local
instability near the leading and trailing edges.

Figure 5-6 illustrates typical results from this potential flow solution. Pre-
dicted blade surface Mach numbers are compared to experimental results
reported by Dunavant et al. (1955). For the purpose of a blade loading diagram
comparison, the experimental surface pressure coefficient data were converted to
Mach numbers using standard compressible flow relations. This is really the
same example used in Aungier (2000), but the analysis is now capable of a more
precise treatment of the blade geometry. Basically, the methods described in
Chapter 4 are now included in the analysis to very precisely define the blade
geometry for standard axial-flow compressor blade camberlines and profiles.
Overall, rather good agreement is achieved between predictions and experiment.
Near the trailing edge, the boundary layer analysis described later in this chapter
predicts boundary layer separation as noted in Fig. 5-6. This would be expected
to suppress further diffusion of the velocity, much as seems to have occurred in
the experiment. Some additional insight into the quality of the predicted results
is provided from empirical blade performance correlations that are described in
the next chapter. For this particular cascade test, those correlations indicate that
the angle of attack is within 0.5° of the optimum value. Noting that the predicted
Mach numbers on the two sides of the blade are nearly equal at the leading edge,
the analysis indicates that the flow enters the blade smoothly, which is often used
as an indication of an optimum inlet angle. Indeed, these predicted Mach num-
bers should not be expected to be exactly equal under optimum inlet conditions,
since the leading edge node structure lies inside the passage as described previ-
ously. In effect, the leading edge predicted values are slightly downstream of the
true entrance point for the flow. The potential flow analysis predicts a discharge
FIGURE 5-6 Potential Flow Results

flow angle of 24.4°, which is in reasonably good agreement with the experimental
value of 25.0°. Hence, this potential flow analysis provides very useful informa-
tion about the performance of this cascade that is consistent with the cascade
test results and with empirical cascade performance models from Chapter 6.
Figure 5-7 illustrates the transonic capability of this extended potential flow
analysis. The case shown is the same as that used for Fig. 5-6, except that the
higher inlet Mach number results in a substantial supersonic patch within the
flow field. The agreement between the predictions and the experimental data of
Dunavant et al. (1955) is reasonably good, both in terms of the blade loading dia-
gram and the discharge flow angle. The procedure used is very robust, providing
rapid and reliable convergence on these transonic flow problems. It is a very
valuable extension for axial-flow compressor analysis, permitting much wider
use of the potential flow method and resulting in far fewer solution failures. As
long as Mach numbers do not become so large that imbedded shock waves sig-
nificantly influence the flow, the predictions are generally quite accurate as well,
as evidenced by this comparison with experimental data.
5.4 LINEARIZED POTENTIAL FLOW ANALYSIS
The two-dimensional potential flow analysis of the previous section can be sim-
plified to provide an exceptionally fast blade-to-blade flow analysis, yet provide
surprisingly good prediction accuracy [Aungier, 1988(a), 2000]. If the sole
FIGURE 5-7 Transonic Potential Flow Solution

purpose of an analysis is to accomplish stand-alone blade-to-blade flow predic-

tion, there is really little reason to consider this simplified model. The two-
dimensional potential flow analysis described above is so fast that it is not
necessary to accept any compromise in accuracy for increased computation
speed. The real purpose of the linearized method is its use in a quasi-three-
dimensional flow analysis, where blade-to-blade analyses are conducted on sev-
eral stream sheets and must be repeated many times. It is described here to
prepare for its later use in that more general application.
The linearized blade-to-blade model is nothing more than a simplification of
the potential flow model of the previous section. The stream function is approxi-
mated by
ψ (m, η) = a(m)[η − η 2 ] + η 2 (5-52)
From Eq. (5-17) it is easily shown that this is equivalent to assuming that the
quantity ρbWm varies linearly with η. Equation (5-14) will be solved in integral
form, noting that W = Wξ on the blade surfaces, i.e.,
1
W1 W0 ∂SWθ
cos β1 cos β0 ∫ ∂m
− = dη + 2Sω sin φ (5-53)
0
The velocity normal to the blade surfaces must be zero, which requires that
Wm 1 ˙ ∂ψ
m
W= = (5-54)
cos β cos β Sbρ ∂η
Combining Eqs. (5-52), (5-53) and (5-54) yields
W1 m˙ (2 − a)
= (5-55)
cos β1 Sbρ cos2β1
W0 ˙
ma
= (5-56)
cos β0 Sbρ cos2β 0
which supply the terms on the left-hand side of Eq. (5-53). From Eqs. (5-18),
(5-19) and (5-52)
˙ [tan β ( a − 2aη + 2η) − a′S(η − η 2 )] / ( bρ )

SWθ = m (5-57)
where the prime notation denotes the total derivative with respect to m. As was
done with the two-dimensional potential flow analysis, the stream function solu-
tion will be accomplished with the density held constant. To simplify the equa-
tions to follow, define

u(m, η) = m
˙ tan β / ( bρ ) (5-58)
v(m, η) = mS
˙ / ( bρ ) (5-59)
Differentiating Eq. (5-57) and introducing Eqs. (5-58) and (5-59) yields
∂SWθ ∂u ∂v
= ( a − 2aη + 2η) + (1 − 2η)ua′ − ( va′′ + a′ )(η − η 2 ) (5-60)
∂m ∂m ∂m
Using truncated Taylor series expansions for any function, F(η), for values at η =
0, 0.5 and 1, a three-point difference approximation to the integral is obtained.
∫ F(η)dη = (F0 + 4F + F1) / 6 (5-61)

0
where the overbar designates a value at η = 0.5. With the above equations and
some tedious algebra, the integral term in Eq. (5-53) is given by
1
∂SWθ
∫ ∂m
dη = [au0′ + u0a′ + 4u ′ − va′′ − v ′a′ + u1′(2 − a) − u1a′] / 6 (5-62)
0
Combining Eqs. (5-53), (5-55), (5-56) and (5-62) yields a simple linear differential
equation.
a′′ + Aa′ + Ba = C (5-63)
where A, B and C are functions of m only
A (m) = [ v ′ − u0 + u1] / v (5-64)

u′ − u′ 6  v1 v0 
B(m) = 1 0 −  +  (5-65)
v vS2  cos2β1 cos2β0 
2u1′ + 4u ′ + 12ω sin φ 12v1

C ( m) = − (5-66)
v vS2 cos2β1
Equation (5-63) is solved from the blade leading edge to the trailing edge. The lead-
ing edge boundary condition follows from the known inlet angular momentum
supplied by the upstream flow, Wθ,in. Integrating Wθ across the passage at the lead-
ing edge using Eq. (5-61) yields the following leading edge boundary condition.
a′ + a[u1 − u0 ] / v = [4u + 2u1 − 6SWθ ,in ] / v (5-67)
The Kutta condition is again used as the trailing edge boundary condition, i.e.,
W0 = W1 at the trailing edge. From Eqs. (5-55) and (5-56), the trailing edge
boundary condition is

a = 2 cos β0 / [cos β1 + cos β0 ] (5-68)
The governing equation is cast in finite-difference form using the difference

approximations previously introduced to develop a matrix equation for solution.
In this case, a simple tri-diagonal matrix is obtained, except for the equation at
the leading edge point, which contains one extra term. Inversion of this matrix is
a rather trivial problem for the numerical analysis, which results in an exception-
ally efficient blade-to-blade flow analysis. The iteration process to update the gas
density field follows the same process as that for the two-dimensional potential
flow, and the same transonic flow extension is incorporated. Aungier (2000) pro-
vides a fairly detailed description of the numerical analysis used to implement
this model.
Figure 5-8 compares results from the linearized and two-dimensional potential
flow analyses for the same test case used earlier. The linearized method does not
resolve the blade loading too well near the leading edge, but does reasonably well
over most of the blade. The linearized method predicts a discharge flow angle for
this case of 25.8°, which is in reasonably good agreement with the experimental
value of 25.0°. The blade surface flow predictions at the leading edge are generally
ignored, since the solution imposes the upstream boundary condition at that
point. It will, by definition, correctly match the average angular momentum at the
leading edge, but not necessarily the blade surface data, particularly on relatively
blunt blades such as this one. Later in this book, it will be seen that the role of this
FIGURE 5-8 Linearized Potential Flow Results

analysis in a quasi-three-dimensional flow analysis is to define the streamwise

variation of the average angular momentum, which it does rather well. Typical
practice is to provide a direct and automated interface to the two-dimensional
method for a detailed blade loading analysis after the quasi-three-dimensional
flow analysis is completed. At that point, the stream sheet geometry and flow con-
ditions are well defined for input to the two-dimensional method. The advantage
offered by the linearized method for the quasi-three-dimensional flow application
is exceptional computation speed while accurately estimating the average angular
momentum distribution through the blade passage.
5.5 THE TIME-MARCHING METHOD
The potential flow method can be applied to a very wide range of axial-flow com-
pressor blade-to-blade flow problems, particularly with the transonic flow exten-
sion. But when Mach numbers become too high, a more general analysis
technique is needed. The time-marching method provides a more general solu-
tion capability that is suitable for subsonic, supersonic or mixed subsonic-super-
sonic flow problems. Von Neumann and Richtmyer (1950) suggested this method
for treating flows with imbedded shock waves. Except for some interest from
mathematicians (Lax, 1954; Lax and Wendroff, 1964), the method received little
attention until computers evolved enough for it to be used on practical problems.
This writer participated in the development of this technique for application to
hypersonic reentry vehicles in the late 1960s [Aungier, 1968, 1970, 1971(a),
1971(b)]. Soon after its successful application to the reentry problem, the method
became popular for the blade-to-blade flow problem (e.g., Gopolakrishnan and
Bozzola, 1973; Denton, 1982). The time-marching technique of Aungier [1970,
1971(a), 1971(b)] has also been adapted to this application and later reported in
Aungier (2000). An abbreviated description of this technique that is sufficient to
understand the fluid dynamics of the problem is provided in this chapter. Read-
ers interested in developing a numerical analysis to implement the procedure
may find the expanded description in Aungier (2000) helpful.
The time-marching solution will be accomplished using the same solution
domain, velocity components and coordinate system as those used for the
potential flow analysis, i.e., Fig. 5-3 and Eqs. (5-1) through (5-7). The governing
equations are Eqs. (3-21), (3-22), (3-23) and (3-25), which are to be solved over
the stream sheet in their full time-unsteady form. The solution is advanced in
time until the flow becomes approximately steady with time. The solution
approaches the steady state asymptotically, so a true steady-state solution is
never actually achieved. Rather, the solution is advanced in time until variations
with time are considered negligible. As was discussed previously for the poten-
tial flow analysis, it is not advisable to simply transform the derivatives in the
governing equations to solve them in the (ξ, η) coordinate system. A more pre-
cise numerical analysis will be achieved by developing the equations for the con-
trol volume to be used in the solution. This is accomplished by applying the
integral form of the equations of motion to the control volume in Fig. 5-4. The
integral form of the continuity, momentum and energy equations for inviscid,
time-unsteady flow are

∂ρ r r
∫ ∂t
dV + ∫ ρ (W ⋅ n)dA = 0 (5-69)
V A
r
∂ρ W r r r r r r r
∫ ∂t
dV + ∫ ρ W (W ⋅ n)dA + ∫ P e ( e ⋅ n)dA = ∫ fdV (5-70)
V A A V
 ∂H ′ ∂P  r r r r
∫ ρ ∂t
− 
∂t 
dV + ∫ ρ H ′( W ⋅ n)dA = ∫ ρ ( f ⋅ W )dV (5-71)
V A V
where V and A denote volume and area integrals, respectively, → n is a unit vector
normal to the area and directed out of the control volume, →e is a unit vector along
→ →
w and f is a body force. The body force is used to account for the Coriolis and
centrifugal acceleration terms in the rotating curvilinear coordinate system. After
some tedious algebra, application of these integral equations to the control vol-
ume yield:
∂ρ ∂ ∂
Sb + [Sbρ Wm ] + [bρ Q] = 0 (5-72)
∂t ∂m ∂η
∂ρ Wm ∂ ∂bP ∂
Sb + [Sb( ρ Wm2 + P )] − tan β + [bρ QWm ]
∂t ∂m ∂η ∂η
1 ∂Sb (5-73)
= SBρ sin φ (Wθ + rω )2 + P
r ∂m
∂ρ Wθ 1 ∂ ∂ ∂
Sb + [rSbρ Wm (Wθ + rω )] + [b( ρ QWθ + P )] = rω [Sbρ Wm ] (5-74)
∂t r ∂m ∂η ∂m
∂( ρ I − P ) ∂ ∂
Sb + [Sbρ Wm I ] + [bρ QI ] = 0 (5-75)
∂t ∂m ∂η
These equations have been written in conservation form, such that they will be
valid when applied across a shock wave as suggested by Lax (1954). Since shock
waves can form when the flow is supersonic, this is an important consideration
in the time-marching method. The parameter Q is a special velocity component
to conserve properties at the constant η boundaries of the control cell, defined by
Q = Wq / cos β = Wθ − Wm tan β (5-76)
Note that Q = Wq = 0 for points on the blade surface. For these points, there is only
one velocity component, Wξ. Applying the integral momentum equation in the ξ
direction yields a special momentum equation for points on the blade surfaces.
∂ρ Wξ ∂ ∂
Sb + [Sb( ρ WmWξ + P cos β )] + [bρ QWξ ]
∂t ∂m ∂η
∂ (5-77)
=P [Sb cos β ] + Sbρ sin φ cos β rω 2
∂m

Wm = Wξ cos β (5-78)
Wθ = Wξ sin β (5-79)
Equations (5-77) through (5-79) replace Eqs. (5-73) and (5-74) for points on the
blade surfaces.
The boundary condition for blade surface points is that the velocity normal to
the surface is zero, which is satisfied by Eqs. (5-78) and (5-79). For the side
boundaries outside of the blade passage, the procedure used is the same as that
for the potential flow problem, i.e., the solution is extended into adjacent pas-
sages using the periodicity condition so these points can be treated in the same
fashion as any interior point. The upstream and downstream boundaries require
more care. The number and type of boundary conditions depend on how these
boundaries are influenced by the flow inside the solution domain. A fundamental
property of hyperbolic differential equations is that there are certain characteris-
tic directions along which derivatives of the dependent variables normal to these
“characteristics” can be discontinuous. For each characteristic direction, certain
dependent variables can be determined by integration along them. Aungier
(2000) derives the characteristic directions for the unsteady flow problem for the
upstream and downstream boundaries. Since Wm is normal to these boundaries,
these unsteady characteristics are defined by
dm
= Wm + a (5-80)
dt
dm
= Wm − a (5-81)
dt
dm
= Wm (5-82)
dt
This simply shows that information can be transmitted within the flow field by
waves traveling at the fluid velocity and waves where the fluid velocity is aug-
mented or opposed by the local acoustic velocity. Aungier (2000) also shows that
the characteristics of Eqs. (5-80) and (5-81) determine Wm and P, while the char-
acteristic of Eq. (5-82) determines Wθ and I. These characteristics can be used to
define the number and type of boundary conditions needed for the upstream and
downstream boundaries. Figure 5-9 shows these three characteristics drawn on
an m-t diagram for an upstream boundary with a subsonic meridional velocity
component. Since one of the characteristics that determines the flow at time t +
∆t lies within the solution domain, one dependent variable on the boundary can
be computed as part of the solution, while the other three must be assigned as
boundary conditions. The characteristic for Wθ and I lies outside the solution
domain, so both must be assigned as boundary conditions, with one more bound-
ary condition to be supplied. A logical choice for the computed dependent vari-
able is density. Then, P and Wm follow directly from the equation of state and the
definition of rothalpy if entropy is known. The most logical upstream boundary
conditions for this case are Wθ, Pt, and Tt. They are usually known conditions for

FIGURE 5-9 Subsonic Upstream Boundary
a blade-to-blade flow analysis and they specify rothalpy and entropy through the
equation of state and Eq. (3-10). So, for an upstream boundary with Wm < a, the
continuity equation will be solved for density, with all other dependent variables
computed from the boundary conditions. Figure 5-10 shows the m-t diagram for
an upstream boundary where the meridional velocity component is supersonic.
Here, all of the characteristics that determine the flow at t + ∆t lie outside of the
solution domain. Consequently, none of the dependent variables on this bound-
ary can be computed from the solution. All dependent variables must be assigned
as boundary conditions when Wm > a on an upstream boundary. Figure 5-11
shows the m-t diagram for a downstream boundary where the meridional veloc-
ity component is subsonic. Here, one of the characteristics that determine the
flow at t + ∆t lies outside the solution domain. This means that one dependent
variable must be specified as a boundary condition, while the other three are
computed as part of the solution. The usual practice is to specify the discharge
static pressure as the boundary condition. It can be convenient to specify the
mass flow rate instead, but then the solution procedure must compute the dis-
charge pressure needed to produce that mass flow rate. Figure 5-12 shows the m-
t diagram for a downstream boundary where the meridional velocity component
is supersonic. Here, all of the characteristics that determine the flow at t + ∆t lie
inside the solution domain. This means that all dependent variables can be com-
puted from the solution, with no boundary conditions required. Usually, it is not
possible to state that Wm is supersonic. That is a very unique case for each set of

FIGURE 5-10 Supersonic Upstream Boundary
FIGURE 5-11 Subsonic Downstream Boundary

FIGURE 5-12 Supersonic Downstream Boundary
specified upstream boundary conditions. Rather, the discharge static pressure is

normally specified in all cases. If the solution requires Wm > a to achieve that
pressure, the solution procedure must ignore the boundary condition and com-
pute all discharge conditions. In contrast to the potential flow analysis, it is noted
that no empirical Kutta condition has been needed. Indeed, there would be no
way to actually use it, since Wθ is always computed at the downstream boundary
as part of the solution. If Wθ were specified as a boundary condition, a Kutta con-
dition would be needed to select a value. Indirectly, the Kutta condition is applied
by virtue of the periodicity condition for all solution points downstream of the
trailing edge. It is important to note that the boundary condition requirements
are not optional. For example, specifying more boundary conditions than out-
lined here will not produce a valid solution, and normally will cause the solution
to diverge. Some investigators use the unsteady characteristics directly to solve
for independent variables on the boundaries, e.g., Gopolakrishnan and Bozzola
(1973). In most cases this just complicates the analysis for no good reason. The
same thing is accomplished by simple upstream or downstream differences in
the governing equations using Eqs. (5-40) and (5-42) with far less numerical
logic. On occasion, unsteady characteristics can be used to advantage at bound-
ary points. For example, this writer used them for the upstream bow shock
boundary in the hypersonic reentry problem [Aungier, 1970, 1971(a), 1971(b)].
But there really is no merit to using them in the present application. The impor-
tant role of unsteady characteristics is to identify the type of and the number of
boundary conditions required.

A major issue to be addressed in a time-marching solution is numerical stabil-

ity. It is well known that a finite-difference solution of the governing equations
that is explicit in time is unconditionally unstable. The term explicit in time
refers to predicting flow data at time t + ∆t from conditions at time t. In contrast,
implicit solutions seek to use conditions at both t and t + ∆t for this purpose.
Those solutions also have critical numerical stability issues, but this book will
deal with only the explicit solution procedures. Von Neumann and Richtmyer
(1950) developed a stability analysis procedure for this problem. They used it to
achieve a stable numerical solution by including additional stabilizing terms in
the governing equations similar in form to viscous terms. This approach is often
referred to as including artificial viscosity terms. Lax (1954) used an averaging
scheme for parameters at time t to project data at t + ∆t, which is equivalent to
introducing artificial viscosity terms. Another approach that has been used is to
apply a Taylor series expansion in time to the governing equations to extend them
to second (or higher) order accuracy in time (e.g., Lax and Wendroff, 1964).
Regardless of the approach used, stabilizing terms must always be added to the
governing equations to achieve a stable explicit finite-difference solution. Unless
extreme care is taken, these stabilizing terms can significantly influence the solu-
tion, possibly producing very unsatisfactory results. A method developed by this
writer [Aungier 1970, 1971(a), 1971(b), 2000] has a definite advantage in that
regard. It always permits the user of the analysis to reduce the influence of the
stabilizing terms on the solution as much as necessary. But reduction of these
effects is accomplished at the cost of requiring longer computation times for a
steady-state solution. The method was developed by conducting a Von Neumann
stability analysis on a simplified one-dimensional momentum equation to deter-
mine the minimum allowable magnitude of the stabilizing terms for a specified
time step. Aungier (2000) describes this stability analysis in considerable detail.
Here, only the results will be reviewed. It is convenient to represent any of the
governing equations in the general form
ut = v(ξ , η, t) + µ (ξ )uξξ + µ (η )uηη (5-83)
where the subscript notation identifies first and second partial derivatives, and
the last two terms are the stabilizing terms. For a stable solution for any time
step, ∆t, the stability analysis shows that the coefficients of the stabilizing terms
must satisfy the following conditions.
µ (ξ ) ≥ 12 ( Wm + a)2 ∆t (5-84)
µ (η ) ≥ 12 ( Wθ + a)2 ∆t (5-85)
µ (ξ ) ≥ 12 [( Wξ + a) cos β ]2 ∆t (5-86)
µ (η ) ≥ 12 [( Wq + a) / cos β ]2 ∆t (5-87)
This is satisfactory for nearly any blade-to-blade flow problem, but there are
occasional exceptions when the grid structure is highly skewed and the node

spacing is much smaller in the tangential direction than in the meridional direc-
tion. The following empirical correction is applied after the meridional stabiliz-
ing term is established from Eqs. (5-84) and (5-86).
µ (ξ ) → µ (ξ ) + 12 [( Wθ + a) sin 2β ( ∆m) / ( S∆η)]2 ∆t (5-88)
So far, ∆t has been treated as arbitrary, but that is really not the case. The sta-
bility analysis shows that the well-known Courant-Friedricks-Lewy (CFL) limit
(Courant et al., 1928) must always be satisfied. This basically limits ∆t to the time
it takes for the fastest relevant characteristic wave to travel between adjacent
nodes in the solution field. Thus, the maximum value of the time step allowed by
the CFL limit is given by
∆m
∆tmax ≤ (5-89)
Wm + a
S ∆η
∆tmax ≤ (5-90)
Wθ + a
The maximum time step is computed on all time iterations by applying Eqs. (5-
89) and (5-90) at all nodes in the solution field. The actual time step used is some
fraction of this maximum, specified by the user.
∆t = µ0∆tmax (5-91)
Experience has shown that the following limits should be observed to avoid
numerical stability problems.
0.1 ≤ µ0 ≤ 0.9 (5-92)
If the time derivative in Eq. (5-83) is approximated by a forward finite-difference

approximation, the general governing equation is
u(ξ , η, t + ∆t) = u(ξ , η, t) + [ v(ξ , η, t) + µ (ξ )uξξ + µ (η )uηη ]∆t (5-93)
But since the coefficients of the stabilizing terms are also proportional to ∆t, as
seen from Eqs. (5-84) through (5-88), the stabilizing terms are second order with
respect to ∆t, while the dynamic terms represented by the general function v are
first order with respect to ∆t. Hence, by simply using smaller values of ∆t the
influence of the stabilizing terms can be reduced. Of course, that will mean
more time iterations must be processed for the solution to approach a steady
state. The approach used by this writer is to start the analysis with a fairly large
value of µ0 (typically, 0.75) and steadily reduce it to some smaller value (typi-
cally, 0.25) as the solution approaches a steady state. This allows fairly large
time steps to be used in the early iterations to accelerate the approach to a

steady state, but relatively small time steps when the solution is close to a steady
state to reduce the effect of the stabilizing terms.
The stability analysis of Aungier (2000) produced other useful results. It shows
that for points on the blade surfaces, no stabilizing term normal to the surface is
required, i.e., the last term in Eq. (5-83) is omitted for blade surface points. It also
shows that for Wm > a, and backward difference approximation for meridional
partial derivatives
∂u
= [ui, j − ui −1, j ] / ∆m (5-94)
∂m
no stabilizing term in the ξ direction is required, i.e., the second term on the
right-hand side of Eq. (5-83) can be omitted. A similar result is obtained for neg-
ative Wm, except a forward difference approximation is used
∂u
= [ui +1, j − ui, j ] / ∆m (5-95)
∂m
Hence, supersonic Wm with “upwind” differences requires no stabilizing term

in the ξ direction. This permits special procedures to be used on partial deriva-
tives with respect to m to improve accuracy for very high Mach number flows.
This is not usually necessary for compressor applications, but can significantly
improve results for the high Mach numbers often encountered in turbine blades.
Basically a weighted average of forward and backward differences are employed,
as follows:
um = F[ui +1, j − ui, j ] / ∆m + (1 − F )[ui, j − ui −1, j ] / ∆m (5-96)

 2Wm Wn 
1
F= −  (5-97)
2
 ( Wm + a + Wm + a )2 
Wm = 1
4[Wm (m + ∆m,η) + 2 Wm (m,η) + Wm (m − ∆m,η)] (5-98)
am = 14 [ am (m + ∆m, η) + 2 am (m, η) + am (m − ∆m, η)] (5-99)
µ (ξ ) → 4µ (ξ )F(1 − F ) (5-100)
This procedure uses basic central-difference approximations and the basic stabi-
lizing term form outlined previously when Wm = 0. As Wm→ a, F → 0. Hence,
for Wm ≥ a, the solution will use an upwind difference approximation for the
partial derivative, and the ξ stabilizing term will be zero. The result is that the
minimum magnitude in the stabilizing terms allowed for stability is always used.
It also results in faster convergence and sharper ‘shock capturing” when imbed-
ded shock waves form in the flow field. For a period of time, a similar procedure
was used relative to partial derivatives and stabilizing terms relative to the η
direction. It was found that no significant benefit resulted for high Mach number

turbine blade problems, and the procedure occasionally slowed convergence and
sometimes produced mild numerical instability. Interaction with the side bound-
aries and the higher probability that Wq and Q at a given node may change sign
during the solution are believed to be the source of the problem. Presently, this
writer uses standard central difference approximations and the basic stabilizing
terms relative to the η direction.
The basic form of the stabilizing terms significantly affects their influence on
the solution. The stabilizing terms should be formulated such that their magni-
tude is expected to be small when the flow approaches a steady state. The recom-
mended stabilizing terms to be added to the right-hand side of Eqs. (5-72)
through (5-75) and Eq. (5-77), in that order, are
Φ ρ = Sb[µ (ξ )ρ mm + µ (η )ρηη ] (5-101)
Φ m = µ (ξ ) ( Sbρ Wm )mm + µ (η ) ( Sbρ Wm )ηη + Φ ρ Wm (5-102)
Φη = Sbρ [µ (ξ ) ( rCθ )mm + µ (η ) ( rCθ )ηη ] / r + Φ ρ Wθ (5-103)
Φ I = Sbρ [µ (ξ )Imm + µ (η )Iηη ] + Φ ρ I (5-104)
Φξ = µ (ξ ) ( SbρWξ )mm + Φ ρ Wξ (5-105)
The terms involving Φρ in Eqs. (5-102) through (5-105) can be shown to be direct
corrections for the known error in mass conservation caused by Φρ when solving
Eq. (5-72).
The numerical analysis used is a fairly straightforward finite-difference analy-
sis applied to the grid structure illustrated in Fig. 5-3. The only subtle feature
required relates to defining the side boundaries outside of the blade passages.
When Eq. (5-77) is solved at the blade leading and trailing edges, the finite-differ-
ence approximation to the ξ derivative will involve the closest node outside the
blade passage. Equation (5-77) is derived for nodes where Wq = 0. As the solution
proceeds, the side boundaries outside the blade are continually readjusted such
that β is equal to the local flow angle at the closest node outside the blade pas-
sage. This ensures that Wq will be zero at this node so that the difference approx-
imation used in Eq. (5-77) will be valid. Aungier (2000) provides additional
details on the numerical analysis that may be of interest to readers considering
implementation of this method in a numerical analysis.
Figures 5-13 and 5-14 illustrate typical results from this time-marching
method for the problems discussed previously for the potential flow solution,
including comparison with that method. For the subsonic case, the time-march-
ing solution generally shows better agreement with the experiments than does
the potential flow method, particularly near the leading and trailing edges. It
does show somewhat higher pressure-surface Mach numbers than either the
experiment or the potential flow analysis, but agreement is considered very satis-
factory. The predicted discharge flow angle is 21.1°, which is not in as good
agreement with the experimental value of 25.0° as was achieved by the potential
flow prediction of 24.4°. The transonic flow case shows similar trends, except
that the potential flow analysis more accurately locates the point of maximum

FIGURE 5-13 Time-Marching Solution Results
FIGURE 5-14 Transonic Time-Marching Results

Mach number. Comparison of the time-marching method with this experimental

data is particularly difficult and a little misleading. This case is very close to the
passage choke flow limit. The time-marching method predicts choke at a slightly
lower mass flow rate than indicated by the potential flow analysis and the exper-
iment, no doubt due to the stabilizing term influence. Indeed, the mass flow rate
used in the time-marching solution is slightly lower than for the potential flow
solution and the experiment. The time-marching results correspond to the lowest
discharge pressure that could be assigned without causing the flow to choke,
expand to supersonic velocities near the discharge and then form shock waves to
adjust to the discharge pressure.
Both the potential flow and time-marching blade-to-blade flow analyses are
seen to provide very useful guidance as aerodynamic design and analysis tools.
For the sample problems considered, the potential flow method would be pre-
ferred due to the absence of stabilizing term effects and due to the much shorter
computation time required for a solution. But the time-marching method pro-
vides a capability for solution at Mach number levels that are beyond the capa-
bility of potential flow methods.
5.6 BLADE SURFACE BOUNDARY LAYER ANALYSIS

Two-dimensional boundary layer analysis is a valuable supplement to the inviscid
blade-to-blade flow analyses discussed in this chapter. This type of analysis can
be added to an inviscid flow analysis with very little increase in computation
time. The basic assumption of boundary layer theory is that viscous effects are
confined to a thin layer close to the walls. This is usually a valid assumption for
the flow in a blade-to-blade stream sheet, where the boundary layers form on the
blade surfaces. However, there are usually significant pressure gradients normal
to the stream sheet. Unlike the inviscid flow, the low momentum of the boundary
layer fluid will be unable to balance these normal pressure gradients, causing the
boundary layer fluid to migrate across stream sheets. Two-dimensional boundary
layer analyses must usually be considered to be an approximation, since they
neglect this effect. Yet they provide a practical and useful method to quantify vis-
cous effects and to identify likely boundary layer separation problems. These
phenomena are very important since they govern the level of total pressure loss
that will occur as the flow passes through the blade row. Boundary layer analysis
in turbomachinery is most conveniently accomplished by applying the equations
in integral form. The basic governing equation is the momentum-integral equa-
tion derived in Chapter 3 as Eq. (3-42). In the context of the present problem, this
equation can be written
1 ∂bρ eWe2θ ∂We

+ δ * ρ eWe = τw (5-106)
b ∂x ∂x
The subscript e designates inviscid, boundary layer edge parameters, x is the dis-
tance along the blade surfaces and τw is the wall shear stress. The momentum
and displacement thicknesses are defined as

δ
ρ eWe2θ = ∫ ρ W (We − W )dy (5-107)
0
δ
ρ eWeδ * = ∫ ( ρ eWe − ρ W )dy (5-108)
0
where y is the distance normal to the wall and δ is the boundary layer thickness.
Usually the wall shear stress is expressed in terms of the skin friction coefficient,
cf, where
τw
cf = 2
(5-109)
2 ρ eWe
1
When the boundary layer initially forms on the blade, the flow will be laminar.
Typically, the boundary layers on axial-flow compressor blades soon transitions
to turbulent flow. Hence, both laminar and turbulent boundary layer analyses
are needed.
This writer prefers the laminar boundary layer analysis of Gruschwitz (1950)
as reviewed by Schlichting (1968). It is a generalization of the classical incom-
pressible Karmen-Pohlhausen solution (Pohlhausen, 1921) to compressible flow.
Among other advantages, this solution allows a very direct treatment of transi-
tion to turbulence by simple application of conservation of mass and momen-
tum. This method employs a universal boundary layer velocity profile in the form
W = C1η + C2η 2 + C3η3 + C4η 4 (5-110)

We
y
1 ρ
δ ′ ∫ ρe
η= dy (5-111)
0
δ
ρ
δ′ = ∫ dy (5-112)
ρe
0
Denoting the fluid viscosity by µ, a shape factor, Λ, is defined as
ρ e2 (δ ′ )2 dWe (5-113)
Λ=
ρ w µ dx
Then, by matching the boundary layer edge conditions, the coefficients in Eq.
(5-110) are
C1 = 2 + Λ / 6, C2 = − Λ / 2, C3 = Λ / 2 − 2, C4 = 1 − Λ / 6 (5-114)
and the momentum thickness is given by

θ 37 Λ Λ2
= − − (5-115)
δ ′ 315 945 9072
The boundary layer energy thickness, δE, and velocity thickness, δW, are given by
δ
δE ρ W  W2  798048 − 4656Λ − 758Λ2 − 7 Λ3
=∫ 1 − 2  dy = (5-116)
δ′ ρ W  We  4324320
0 e e 
δ
δW  W 3 Λ FWe2
= ∫ 1 −  dy = − + (5-117)
δ′ We  10 120 2cpTe
0
where the function F is given by
2 3
Λ  Λ   Λ 
F = 0.232912 − 0.831483 + 0.650584  + 17.8063  (5-118)
100 100  100 
For the case of adiabatic walls with the Prandtl number equal unity, the bound-
ary layer enthalpy thickness is
δ
δh ρW  h  We2δ E
=∫  − 1 dy = (5-119)
δ′ ρW h 2cpTeδ ′
0 e e  e 
and the displacement thickness is given by
δ * = δ h + δW (5-120)
Again considering adiabatic walls with the Prandtl number equal unity, the fol-
lowing parameters are introduced for convenience
ρe T′
b0 = = t (5-121)
ρ w Te
2
θ  ρ θ 2 dWe (5-122)
K = Λ   = b0 e
δ ′  µ dx
and Gruschweitz (1950) shows that
1c µ  Λ
= 1+ (5-123)
2 f ρ eWeδ ′  6 
2
 37 Λ Λ2 
K= − −  Λ (5-124)
 315 945 9072 

The above equations are sufficient to permit numerical integration of Eq. (5-106)
along x, starting at the leading edge, where θ = 0, to predict θ, δ* and δ at all other
stations. In the process Gruschwitz limits the shape factor, Λ to lie in the range
−12 ≤ Λ ≤ 12 (5-125)
where Λ = -12 corresponds to boundary layer separation. Normally, the boundary

layer can be expected to undergo transition to turbulent flow. Many criteria for
boundary layer transition have been proposed. This writer recommends use of
the momentum thickness Reynolds number in the following transition criterion:
ρ eWeθ
Reθ = > 250 (5-126)
µ
Once the boundary layer undergoes transition to turbulent flow, different

empirical relations are required to integrate Eq. (5-106). This writer prefers the
entrainment method of Head (1958, 1968) as adapted by Green (1968) to com-
pressible flows. This requires integration of the entrainment equation, presented
previously as Eq. (3-44). In context of the present application, this is written as
∂bρ eWe (δ − δ * )
= bρ eWe E (5-127)
∂x
The entrainment function, E, specifies the rate at which mass is entrained into
the boundary layer at the boundary layer edge. To integrate Eqs. (5-106) and
(5-127), empirical relations are required for E, cf and for the various boundary
layer thicknesses as a function of θ and (δ – δ*). Head’s entrainment method was
developed for incompressible boundary layers using the following shape factors
as its basis:
H1 ≡ (δ − δ * ) / θ (5-128)
*
H = δ /θ (5-129)
Green (1968) recommends using a kinematic shape factor Hk in place of H when

generalizing an incompressible boundary layer model to compressible flow.
δ
1 ρ  W
Hk = ∫ 1−
θ ρ e  We 
 dy (5-130)
0
Green shows that for adiabatic walls, with the Prandtl number equal unity, Hk
can be related to H by
H = ( Hk + 1)Tt′ / Te − 1 (5-131)

Solution of the governing equations requires empirical models for E and cf and
to relate Hk and H1. Many alternate empirical models have been suggested for
this purpose. This writer has found the following relations suitable for the pres-
ent application:
Hk = 1 + [0.9 / ( H1 − 3.3)]0.75 (5-132)

E = 0.025( Hk − 1) (5-133)
The skin friction coefficient correlation of Ludwieg and Tillmann (1950) is com-
monly used for incompressible turbulent boundary layer analysis.
cf , inc = 0.246 exp(−1.561Hk ) Reθ−0.268 (5-134)
Green (1968) recommends correcting this incompressible flow correlation to

compressible flow by
cf = cf , incHk ( Hk + 1) / [2Hk + H ( Hk − 1)] (5-135)
These empirical relations are sufficient to integrate Eqs. (5-106) and (5-127),
starting at the transition point. At the transition point, the mass and momentum
flow in the turbulent boundary layer must match the values computed for the
laminar boundary layer. From Eqs. (3-35) and (3-38), this requires
(δ − δ * )turb = (δ − δ * )lam (5-136)

θ turb = θ lam (5-137)
From the Gruschwitz (1950) profiles, it can easily be shown that
7 Λ 
δ −δ* = δ′ +  (5-138)
10 120 
Equation (5-137) is normally applied at the transition point, along with some
assumption on how H changes during transition from laminar to turbulent flow.
Use of the Gruschwitz laminar boundary layer model permits a more fundamen-
tal method based on conservation of mass and momentum. This writer uses Hk ≥
2.4 as a separation criterion for turbulent boundary layers. During the analysis,
the kinematic shape factor is limited to this value to avoid solution divergence
and permit the analysis to continue through a separation zone. This is necessary,
since it is not uncommon for the boundary layer to reattach, particularly on the
pressure surface of the blade.
When boundary layers on the blade surfaces have been computed, it is useful
to predict the total pressure loss coefficient for the cascade from the boundary
layer data at the trailing edge. Following Lieblein and Roudebush (1956), the

total pressure loss coefficient based on the cascade inlet velocity pressure can be
approximated by
2
∆Pt  cos βin  2Θ + ( ∆* )2
ω = = (5-139)
( Pt − P )in  cos β out  (1 − ∆* )2
Θ=
∑θ (5-140)
S cos β out
∆* =
∑δ * (5-141)
S cos β out
where the summations are carried out for the boundary layers on both blade sur-
faces at the blade trailing edge. In the case of rotor blade rows, this total pressure
loss coefficient can also be used to estimate the rotor efficiency via the methods
described in Chapter 2. Equation (5-139) is derived directly from conservation of
mass and momentum, at the blade trailing edge, while assuming that the low
momentum boundary layer fluid instantly mixes with the inviscid free stream
fluid while the static pressure remains constant. Those readers who completed
Exercise 3.6 have, in fact, already derived a simple incompressible form of this
equation under the same assumptions.
Loss coefficients computed from two-dimensional boundary layer predictions
should be regarded as quite approximate. The analysis ignores secondary flow
effects associated with boundary layer migration normal to the stream sheets,
which are often quite significant in an annular cascade within an axial flow com-
pressor. Also, the boundary layers are often predicted to separate at some point
along the blade surface. The basic assumptions of boundary layer theory are not
satisfied in separation zones, causing the predictions to be of questionable accu-
racy. It can be expected that the loss coefficient in an annular cascade of an axial-
flow compressor will be significantly higher than is predicted by Eq. (5-139).
Loss coefficients calculated in this fashion do have qualitative significance to
guide the designer in evaluation of relative differences in loss for alternative cas-
cade designs.
Figure 5-15 shows predicted boundary layer shape factors, H, of Eq. (5-129)
that are generated by a two-dimensional boundary layer analysis conducted
using results from the inviscid potential flow analysis results as the boundary
layer edge conditions. The inviscid flow blade-loading diagram for the case con-
sidered is shown in Fig. 5-6. Transition from laminar to turbulent flow occurs
close to the leading edge for both blade surfaces. The deceleration of the inviscid
velocity on the suction surface is severe enough to lead to a significant boundary
layer separation zone starting well upstream of the blade trailing edge. As noted
in the discussion of Fig. 5-6, the experimental data appears consistent with the
premise that boundary layer separation may have occurred in this region.
Numerical analysis for the two-dimensional boundary layer is relatively
straightforward. The governing equations are parabolic in mathematical form.
Hence, a simple marching type solution is needed, since the solution at each
streamwise station depends only on the upstream boundary layer parameters.

FIGURE 5-15 Boundary Layer Analysis Result
Streamwise derivatives in the governing equations are replaced with backward-

difference approximations of the form of Eq. (5-94). An iterative solution proce-
dure is used to complete the integration of the governing equations at each
streamwise station before proceeding to the next station. Aungier (2000) outlines
the procedure used by this writer. In this book, the outline of a numerical proce-
dure is left as an exercise for the interested reader, with appropriate guidance
provided in the exercises.
5.7 SUMMARY
The analysis methods described in this chapter are representative of available

theoretical methods for predicting the two-dimensional flow in cascades. There
are many alternate solution techniques in use, both for the inviscid flow and the
boundary layer. But the precise technique used is less significant than the
approximations that are inherent in the two-dimensional flow model. It has been
seen that reasonable estimates of the fluid turning can be achieved, but reason-
able estimates are not good enough when errors are compounded in a multistage
compressor analysis. The limitations relative to loss prediction have been dis-
cussed in the previous section. These theoretical methods suffer from a more fun-
damental limitation that has not yet been discussed. Both the inviscid flow and
the boundary layer analyses suffer a dramatic loss of accuracy when applied
under far off-design conditions, where severe flow separation may exist. Indeed,

there is very little merit in even applying them to such cases, yet these situations
must be addressed when analyzing a compressor.
Faced with this problem, design engineers normally turn to empirical models
or seek a more fundamental analysis technique. Empirical modeling of cascade
performance has a very long history in axial-flow compressor aerodynamic tech-
nology. Supported by extensive cascade testing, particularly by the NACA, rather
accurate empirical models are available. These empirical models play such an
essential role in axial-flow compressor aerodynamic design and analysis that
they are covered in some detail in Chapter 6. Use of more fundamental theoreti-
cal methods is also receiving much attention. One well-established extension of
the methods in this chapter is to combine them with a hub-to-shroud flow two-
dimensional flow analysis to form a quasi-three-dimensional flow analysis.
Through interaction between the two analyses, it is possible to identify the
stream sheet geometry, which has simply been treated as a known quantity in this
chapter. In the case of centrifugal compressor aerodynamic design and analysis,
the quasi-three-dimensional flow model plays an essential role (Aungier, 2000).
Its advantages for axial-flow compressors are far less dramatic, but sufficient to
warrant presenting the technique in Chapter 12 of this book. The more funda-
mental advance in theoretical analysis is the use of viscous flow computational
fluid dynamics (CFD) codes. The design engineer will typically employ one of sev-
eral commercially available CFD codes that are well suited to turbomachinery
applications. Viscous CFD offers considerable promise for removing most of the
limitations of the methods described in this chapter. Indeed, it has already
greatly alleviated many of them, although much remains to be done in the areas
of turbulence modeling, numerical methods and computational speed before
these methods can be considered exact. Viscous CFD is occasionally applied to
the two-dimensional blade-to-blade flow problem, but its real merit lies in treat-
ing the fully three-dimensional flow problem, where the important secondary
flow patterns are also modeled. At present, the primary role of viscous CFD is in
the area of advanced blade or stage design where its more fundamental fluid
dynamics models can be used to advantage for reducing losses and increasing the
operating range within practical computer running times.
EXERCISES
5.1 Derive an alternate expression for Eq. (5-10) by expressing the partial
derivatives with respect to m and θ in terms partial derivatives with
respect to ξ and η and substituting them into Eq. (3-21). Reduce this
alternate continuity equation to a finite-difference form using central-
difference approximations similar in form to Eqs. (5-40) and (5-41).
Simplify this difference equation to a mass balance equation for the
control volume of Fig. 5-4, and show that the mass balance achieved
for a numerical analysis by this alternate derivation is less accurate
than that given by Eq. (5-8).
5.2 Develop Taylor series expansions about the meridional coordinate
position, m, for ψ(m+∆m) and ψ(m-∆m) in terms of ψ(m) and its
derivatives. Use these relations to derive Eqs. (5-35) and (5-37).

Express the order of accuracy of these finite-difference approxima-

tions in terms of ∆m.
5.3 Develop Taylor series expansions about the meridional coordinate
position, m, for ψ(m+∆m) and ψ(m+2∆m) in terms of ψ(m) and its
derivatives. Use these relations to derive Eq. (5-40). Express the order
of accuracy of this finite-difference approximation in terms of ∆m.
5.4 From a truncated Taylor series expansion about the meridional coor-
dinate position, m, for u(m+∆m), derive Eq. (5-95). Express the order
of accuracy of this finite-difference approximation in terms of ∆m.
5.5 The paragraph preceding Eq. (5-101) notes that the form of the stabi-
lizing terms used in a time-marching solution is an important consid-
eration. Consider flow in the portion of the solution domain upstream
of the blade where the flow is uniform on the upstream boundary.
Comment on the expected magnitudes of the stabilizing terms associ-
ated with µ(ξ) in Eqs. (5-102) through (5-104) when the flow is approx-
imately steady state. Why is the term Sbρ outside of the partial
derivative in Eqs. (5-103) and (5-104) but inside the partial derivative
in Eq. (5-102)? If the form of the stabilizing terms causes them to be
very small in magnitude near a steady state, how can they stabilize
the numerical analysis?
5.6 Consider the numerical solution of the laminar boundary layer equa-
tions at a specific streamwise station where all boundary layer edge
conditions and all boundary layer parameters at the upstream station
are known.
(a) Define a procedure to estimate a “safe” initial guess for θ at the cur-
rent station from the upstream station, which accounts for change
in just boundary layer edge conditions. Suggest a “safe” procedure
for the case where the upstream station is the leading edge.
(b) Outline a procedure that might be followed to predict the laminar
boundary at this station in terms of specific steps [parameter to
predict and equation number(s) to be used]. It is understood that
these steps will be repeated, in order, until the process converges.
5.7 Repeat Exercise 5.6 for turbulent boundary layers, except that in this
case the upstream station cannot be the leading edge, so that issue is
not relevant.

Chapter 6
EMPIRICAL PERFORMANCE
MODELS BASED ON
TWO-DIMENSIONAL
CASCADE TESTS
The theoretical methods to analyze the flow in cascades presented in Chapter 5

are useful design and analysis methods, but they are not sufficient for day-to-day
axial-flow compressor design activity. The many limitations restricting the accu-
racy and range of application of those methods have been discussed in Chapter 5.
Consequently, empirical models are commonly used as a means of predicting the
basic performance of cascades in axial-flow compressor design and analysis.
These empirical models are derived from experimental data obtained from two-
dimensional cascade testing. Extensive testing of this type has been accomplished,
particularly by the NACA. Rather sophisticated empirical methods are available
for the standard blade profiles discussed in Chapter 4. In the case of advanced
blade types, such as the controlled diffusion airfoils discussed in Chapter 4, alter-
nate empirical models may be required. Typically these will be an adaptation of
the methods used for the standard airfoil profiles, specifically addressing the
improved performance characteristics achieved by the new profile design.
The basic objective of the empirical modeling process is to predict the fluid
turning and total pressure loss for a cascade under fairly general operating con-
ditions. Also, operating conditions where near-optimum performance can be
expected need to be identified by the empirical models. These are often referred
to as design conditions, since they are appropriate for use under the compressor’s
design operating condition where optimum performance is usually desired.
Indeed, it will be seen in Chapters 10 and 11 that these can be used as a basis for
selecting blade geometry to match the desired flow field through the compressor.
Special empirical models to address blade tip leakage and seal leakage through
stator shroud seals will also be covered in this chapter.
NOMENCLATURE
a = distance to point of maximum camber along chord
Bwake = blade wake blockage

b = exponent factor in Eq. (6-21); also distance from chord line to

the point of maximum camber
Cc = seal carryover coefficient
Cr = seal contraction ratio
Ct = seal throttling coefficient
Cl0 = lift coefficient
c = blade chord
D = diffusion factor in Eq. (6-29)
Deq = equivalent diffusion factor
h = blade height
i = incidence angle
Ksh = blade shape parameter
Kt,i = design incidence angle thickness correction factor
Kt,δ = design deviation angle thickness correction factor
M = Mach number
Mc = critical Mach number
m = slope parameter in Eq. (6-19)
N = number of seal fins
Nrow = blade row number (sequential through the compressor)
n = slope parameter in Eq. (6-12)
o = throat opening
P = pressure
p = seal pitch, Fig. 6-22
R = range from design inlet angle to stall angle, gas constant or
radius of curvature
Rec = blade chord Reynolds number
r = radius
s = blade pitch
T = temperature
t = seal point thickness, figure 6-22
Uc = leakage velocity
Z = number of blades in a blade row
α = angle of attack; also parameter in Eq. (6-38)
β = flow angle
χ = blade angle with chord
δc = blade tip or seal clearance
γ = stagger angle
κ = blade angle with the meridional direction
θ = camber angle
θw = blade wake momentum thickness
ρ = gas density
φ = angle defined in Fig. 6-17
ψ = angle defined in Fig. 6-17

Empirical Performance Models Based On Two-Dimensional Cascade Tests • 119
σ = solidity
τ = torque
ω = rotation speed
–
ω = total pressure loss coefficient
Subscripts
c = negative stall angle parameter or leakage parameter
m = meridional component; also minimum loss condition
max = maximum value
s = positive stall angle or bow shock wave condition
u = parameter on upper (suction) blade surface
0 = zero-camber condition
1 = parameter at blade inlet
2 = parameter at blade exit
10 = parameter for 10% thick profile
* = sonic flow condition
Superscripts
* = design condition
6.1 CASCADE GEOMETRY AND PERFORMANCE

PARAMETERS
Figure 6-1 illustrates the basic parameters and nomenclature used to describe the
cascade flow. Subscripts 1 and 2 are used to designate conditions at the inlet and
discharge of the blade, respectively. Blades have a chord length, c, and a tangen-
tial spacing between adjacent blade camberlines or pitch, s. The stagger or set-
ting angle, γ, is the angle between the chord line and the axial direction. The flow
enters the cascade with velocity, W1, and the flow angle with the axial direction is
β1. The angle between the inlet velocity vector and the chord line is called the
angle of attack, α, i.e.,
α = β1 − γ (6-1)
The flow exits the cascade with velocity, W2, and the flow angle with the axial
direction is β2. The blade angles κ1 and κ2 are the angles between the camberline
and the axial direction at the leading and trailing edges, respectively. The inci-
dence angle, i, and the deviation angle, δ, are defined as

FIGURE 6-1 Cascade Nomenclature
i = β1 − κ1 (6-2)
δ = β2 − κ 2 (6-3)
As discussed in Chapter 4, the angles κ1 and κ2 are not well defined for the NACA
families of blade camberlines. For this reason, NACA cascade data is usually pre-
sented in terms of the angle of attack and the fluid turning, ε.
ε = β1 − β2 (6-4)
It has become fairly common practice to use an equivalent circular-arc camber-

line for the NACA 65-series camberline as a basis for defining i, δ, κ1 and κ2, fol-
lowing the procedure described in Chapter 4. Several other parameters will be
used in the empirical correlations reviewed in this chapter. The minimum passage
opening or throat width, o, is illustrated in Fig. 6-1. The location of the point of
maximum camber, a / c, is illustrated in Fig. 4-6. The thickness-to-chord ratio, tb /
c, is illustrated in Fig. 4-1. The solidity, σ, and the camber angle, θ, are defined by
σ = c/ s (6-5)
θ = κ1 − κ 2 (6-6)
Equation (4-29) will be used to relate lift coefficient, Cl0, of the NACA camber-
lines to camber angle, i.e.,
tan(θ / 4) = 0.05515Cl0 / ( a / c) (6-7)

6.2 DESIGN ANGLE OF ATTACK OR INCIDENCE ANGLE
The design incidence angle, i*, or design angle of attack, α*, define a near-opti-
mum or minimum-loss inlet angle for the cascade. Figure 6-2 shows a chart of
the design angle attack by Herrig et al. (1957) for NACA 65-series blades with
solidity and lift coefficient as parameters. The selection of α* was based on
achieving smooth blade surface pressure distributions, particularly on the suc-
tion surface. This writer has formulated the following empirical model based on
that design chart.
α * = [3.6K sh Kt,i + 0.3532θ ( a / c)0.25 ]σ e (6-8)

e = 0.65 − 0.002θ (6-9)
The data in Fig. 6-2 applies to NACA 65-series blades with tb / c = 0.1, so Kt, i =
Ksh = 1 and a / c = 0.5. The parameters Ksh and Kt,i are adapted from the design
incidence correlation of Lieblein (1960), which is also presented in NASA SP-36
(Johnsen and Bullock, 1965). Lieblein shows that design incidence angle corre-
lations for NACA 65-series blades with tb / c = 0.1 can be extended to other pro-
file types and other thickness-to-chord ratios by applying the correction factors
to the design incidence (or design angle of attack) for camber angle equal zero.
Ksh assumes values of 1.0 for NACA profiles, 1.1 for the C4-series profile and 0.7
for the double-circular-arc profile. Figure 6-3 shows the correction term Kt, i
FIGURE 6-2 The Design Angle of Attack

FIGURE 6-3 Thickness Correction for α*
provided graphically in Johnsen and Bullock (1965), compared to the following

empirical equation:
Kt, i = (10tb / c)q (6-10)

0.3
q = 0.28 / [0.1 + (tb / c) ] (6-11)
The parameter a/c introduced into Eq. (6-8) serves to extend the correlation to
the NACA A4K6 inlet guide vane camberline, and is presumed to be applicable to
the parabolic-arc camberline also, although experimental data isn’t available for
the writer to actually confirm that.
Lieblein’s (1960) design incidence angle correlation is developed from basi-
cally the same data as that in Fig. 6-3, but with the intent to identify the mini-
mum loss incidence angle. The equivalent circular-arc camberline is used for the
NACA 65-series blades as the basis for defining the incidence angle. The form of
Lieblein’s correlation is
i* = K sh Kt, i (i0* )10 + nθ (6-12)
The first term on the right-hand side of Eq. (6-12) is the design incidence angle
for a camber angle of zero. It is computed from a correlation for NACA 65-series
blades with tb / c = 0.1, corrected by Ksh and Kt,i. The base zero-camber incidence

angle graphical correlation from Johnsen and Bullock (1965) is shown in Fig. 6-4,
along with predictions from the following empirical equations developed from
the graphical correlation.
β1p
(i0* )10 = − 0.1σ 3 exp[(β1 − 70) / 4] (6-13)
5 + 46 exp(−2.3σ )
p = 0.914 + σ 3 / 160 (6-14)
Figure 6-5 shows the graphical correlation for the slope factor, n, from
Johnsen and Bullock (1965) along with predictions from the following empiri-
cal equation:
(β1 / 90)(1+1.2σ )
n = 0.025σ − 0.06 − (6-15)
1.5 + 0.43σ
It can be noted that both Eqs. (6-13) and (6-15) contain the flow angle, which is,
itself, a function of i*. Hence, an iterative solution is needed to compute the
design incidence angle using Lieblein’s (1960) correlation. Liebleins’s model
applies only to blades where a/c = 0.5. To treat the NACA A4K6 guide vanes and
parabolic-arc blades, this writer has used the following procedure. A pseudo-
blade inlet angle is computed as
FIGURE 6-4 Zero-Camber Design Incidence Angle

FIGURE 6-5 Design Incidence Angle Slope Factor
κ1 = γ + 1
2θ (6-16)
This is used to compute a pseudo-design incidence angle using Lieblein’s correla-

tions. Then, the design angle of attack for the actual blade is adjusted by
α * = κ1 + i * − γ + ( a / c − 0.5)θ (6-17)
The actual design incidence angle is given by
i* = α * + γ − κ1 (6-18)
This writer has used both the design angle of attack correlation of Eq. (6-8)
and Lieblein’s design incidence models rather extensively in axial-flow compres-
sor aerodynamic performance analysis. Clearly, Lieblein’s model is based on a
less subjective criterion than α* and includes more parameters in the correlation.
But, in practice, the design angle of attack model has consistently resulted in
more accurate compressor performance predictions. That is not necessarily a sig-
nificant evaluation of the two models. A performance analysis uses many other
empirical models, which are discussed in the remainder of this chapter. Perfor-
mance prediction accuracy depends on the complete set of models, more than on

any single model. So, all that really can be said is that this writer has had better
success using the design angle of attack model.
6.3 DESIGN DEVIATION ANGLE
Lieblein (1960) also supplies an empirical model for the design deviation angle,
δ*, corresponding to operation at the design incidence angle. This model is
reviewed in somewhat greater detail in NASA SP-36 (Johnsen and Bullock, 1965).
The model is similar in form to Lieblein’s design incidence angle model, i.e.,
δ * = K sh Kt,δ (δ0* )10 + mθ (6-19)
Ksh is the same as for the design incidence angle model. Figure 6-6 shows the
graphical form of the base zero-camber deviation angle from Johnsen and Bul-
lock (1965) along with predictions from the following empirical equation:
(δ 0* )10 = 0.01σβ1 + [0.74σ 1.9 + 3σ ](β1 / 90)(1.67 +1.09σ ) (6-20)
The slope parameter, m, is expressed as a function of its value for a solidity of 1.0,
m1.0, corrected for other values of solidity in the following form.
FIGURE 6-6 Zero-Camber Deviation Angle

m = m1.0 / σ b (6-21)
The graphical form of the parameters m1.0, b and Kt, δ from NASA SP-36 (Johnsen
and Bullock, 1965) are shown in Fig. 6-7, 6-8 and 6-9, along with predictions
from empirical equations used by this writer. Defining x = β1 / 100, the slope fac-
tor for the NACA 65-series camberline is modeled as
m1.0 = 0.17 − 0.0333 x + 0.333 x2 (6-22)
For circular-arc camberlines,
m1.0 = 0.249 + 0.074 x − 0.132 x2 + 0.316 x3 (6-23)
The exponent, b, is modeled by
b = 0.9625 − 0.17 x − 0.85 x3 (6-24)

Kt,δ = 6.25(tb / c) + 37.5(tb / c)2 (6-25)
It can be seen that Lieblein’s design deviation angle model does not apply to the
NACA A4K6 inlet guide vane or parabolic-arc camberlines. For those blade types,
FIGURE 6-7 Design Deviation Angle Slope Factor

FIGURE 6-8 Design Deviation Angle Exponent
FIGURE 6-9 Thickness Correction for *

a modified design deviation angle correlation by Howell (1942, 1945) can be

used. Howell’s correlation is expressed in terms of the discharge flow angle,
which is a function of the deviation angle, but that problem is easily remedied by
substituting for β2 using Eq. (6-3). Howell also does not account for blade thick-
ness-to-chord and profile shape effects, so it is adjusted using Lieblein’s correc-
tions for those effects. The modified form used is
0.92( a / c)2 + 0.002κ 2 θ

δ* = + ( K sh Kt,δ − 1)(δ 0* )10 (6-26)
1 − 0.002θ / σ σ
Equation (6-26) is also used for all inlet guide vanes, regardless of camberline or
profile type. Inlet guide vanes are unique in the sense that they are normally the
only cascades in an axial-flow compressor that accelerate the flow. Indeed, at suf-
ficiently large stagger angles, such that the blade passage throat is located at the
discharge, they function just like turbine nozzle blade rows. In those cases, an
axial-flow turbine deviation angle correlation would be more appropriate. But
that is not really a viable approach, since the throat of inlet guide vanes also may
be located at the inlet. In the case of adjustable inlet guide vanes, both situations
may be encountered in the same compressor. Basically, the function of inlet guide
vanes is to turn the flow from an inlet flow angle of zero to some larger flow
angle. In the context of a compressor analysis, this means an inlet guide vane will
have a negative camber angle. The preceding empirical equations are all applica-
ble to positive camber angles only. They can be applied to a cascade with negative
camber simply by changing the signs of all blade angles (θ, γ, κ1 and κ2) and flow
angles (β1 and β2), applying the correlations, and then changing all signs again to
cast the results in the sign convention used for the compressor analysis. When
these sign corrections are applied, inlet guide vanes have the unique feature that
the corrected γ is negative. Also, the corrected κ2 will have a relatively large nega-
tive value. Howell’s design deviation angle correlation is the only one known to
this writer that can properly handle this situation. Indeed, deviation angles pre-
dicted by Eq. (6-26) for the NACA A4K6 inlet guide vane camberline are in rather
good agreement with the design charts of Dunavant (1957).
6.4 DESIGN LOSS COEFFICIENT AND DIFFUSION FACTORS
Now that a reference or design incidence angle and the corresponding fluid turn-
ing or deviation angle have been established, to completely characterize the per-
formance of the cascade at the design operating condition, the corresponding
design total pressure loss coefficient has to be predicted. In the context of two-
dimensional cascade test results, the loss coefficient involved is referred to as the
profile loss coefficient. It is approximately related to the wake momentum thick-
ness, θw (Lieblein, 1959) by
2
θ w σ  W2 
ω =2 (6-27)
c cos β2  W1 

Equation (6-27) is similar to Eq. (5-139), except that wake properties rather than
boundary layer properties are used. Terms involving the wake shape factor, H, are
neglected, since it is close to unity at typical measuring stations in a cascade test.
Velocity relative to the blade, W, is used for generality, although W = C for a sta-
tionary cascade test. The wake momentum thickness is primarily a function of
blade surface skin friction and the blade surface velocity distributions. In partic-
ular, the amount of deceleration or diffusion of the surface velocity is a primary
factor in determining the wake momentum thickness. In the context of a general
blade-loading diagram, this is characterized by the magnitude of the maximum
velocity, Wmax, relative to the discharge velocity, W2. With reference to Fig. 6-10,
it can be seen that Wmax is primarily a function of W1, W2 and the blade loading
distribution, ∆W. For irrotational flow, Stokes’ theorem of vector field theory, Eq.
(5-12), can be used to show that ∆W is a function of the change in the tangential
velocity across the blade row, i.e.,
∆W ∝ W1 + f [(Wθ 2 − Wθ1) / σ ] (6-28)
This fact has been employed to develop various diffusion factors for use as corre-
lating factors for loss coefficient and to estimate the diffusion or loading limits
where boundary layer separation leads to an abrupt increase in loss. Two popular
diffusion factors are the so-called D-factor, D, of Lieblein et al. (1953) and the
equivalent diffusion factor, Deq, of Lieblein (1959).
FIGURE 6-10 Typical Blade Loading Diagram

D ≈ (Wmax − W2 ) / W1 (6-29)
Deq ≈ Wmax / W2 (6-30)
Equation (6-29) is replaced with a very specific approximation for use as a corre-
lating parameter.
W2 Wθ1 − Wθ 2
D = 1− + (6-31)
W1 2σW1
Johnsen and Bullock (1965) provide a good review of the development of the D-
factor as a correlating parameter. Through comparison with extensive two-
dimensional cascade data, it is shown that θw / c can be well represented by a
single curve as a function of D. From Eq. (6-27) it is expected that loss coefficient
should also correlate with D in the form
2
ω cos β2  W1 
 W  = f ( D) (6-32)
2σ  2
That has been confirmed by comparison with cascade test data. It is also shown
that a correlation can be developed in the form
ω cos β2 (6-33)
= f ( D)
2σ
The correlation based on Eq. (6-33) offers less resolution than a correlation based
on Eq. (6-32), and is far less useful for identifying the loading limit where an
abrupt increase in loss is observed. Despite these weaknesses, the correlation
form of Eq. (6-33) was adopted. After a correlation of loss coefficient as a func-
tion of D was developed from two-dimensional cascade data, it was evaluated
against loss data from annular compressor cascade data. It is found that losses in
a compressor cascade are significantly higher than for a simple two-dimensional
cascade test. Even at mid-span, where end-wall boundary layer and tip clearance
effects should be minimal, three-dimensional effects were found to be signifi-
cant. Based on the annular cascade data, the two-dimensional cascade correla-
tion was revised to the form shown in Fig. 6-11, which is well approximated by
the following empirical equation:
ω * cos β2*
= 0.0035[1 + 3.5D* + 37( D* )4 ] (6-34)
2σ
D* rather than D is used to emphasize that the correlation only applies for
operation at the design incidence angle. For values of D* > 0.6, an abrupt increase
in loss is observed, so D* = 0.6 is adopted as the diffusion limit. Equation (6-34)
should not be applied in those cases. The premise of this design loss coefficient

FIGURE 6-11 Loss Coefficient as a Function of D*
correlation is that well-designed compressor cascades will not operate beyond

the diffusion limit at the design incidence condition.
Lieblein (1959) develops the equivalent diffusion factor of Eq. (6-30). From
experimental cascade data for NACA 65-series and C.4 circular-arc blades,
Lieblein developed the following empirical equation for Wmax / W1 for operation
at minimum loss, i.e., for operation at the design incidence angle:
*
 Wmax  cos2 β1*
  = 1.12 + 0.61 [tan β1* − tan β2* ] (6-35)
 W1  σ
Hence, the design equivalent diffusion factor is

* *
* W  W1*  Wmax  cos β2* Wm1
Deq =  max  =  (6-36)
 W1  W2*  W1  cos β1* Wm2
Lieblein considered only two-dimensional cascades where the axial velocity is

constant through the cascade, so he could express Eq. (6-36) in terms of inlet and
discharge flow angles only. From two-dimensional cascade data, Lieblein devel-
oped an empirical correlation for θw / c as a function of the equivalent diffusion
factor. This can be converted to a correlation of design loss coefficient by apply-
ing Eq. (6-27). The result is shown in Fig. 6-12, along with predictions from the
following empirical equation.

FIGURE 6-12 Loss Coefficient as a Function of Deq
2
ω * cos β2*  W1*  * 2 * 8
 *  = 0.004[1 + 3.1( Deq − 1) + 0.4( Deq − 1) ] (6-37)
2σ  W2 
It is seen that losses increase rather abruptly for equivalent diffusion factors
greater than about 2. Lieblein suggests that an equivalent diffusion factor of 2.0 at
design incidence should be considered a diffusion limit, beyond which an abrupt
increase in loss can be expected. To permit use of the equivalent diffusion factor as
an indicator of the off-design diffusion limit, Lieblein extended Eq. (6-35) to
include operation at incidence angles greater than the design incidence angle. The
off-design equivalent diffusion factor for i ≥ i* is computed from Eq. (6-30), using
Wmax cos2 β1
= 1.12 + 0.61 [tan β1 − tan β2 ] + α (i − i* )1.43 (6-38)
W1 σ
The parameter α = 0.0117 for NACA 65-series blades and α = 0.007 for C4 circu-
lar-arc blades. J. Klapproth, in a discussion included in Lieblein (1959), extended
these results to include general annular compressor cascades where the axial
velocity and radius are not constant, and the cascade may be rotating. The more
general form is
Wmax cos2 β1 r1Cθ 1 − r2Cθ 2

= 1.12 + 0.61 + α (i − i* )1.43 (6-39)
W1 σ r1Cm1

Equation (6-39) accounts for changes in the meridional velocity and in the blade
circulation due to a change in radius and due to blade row rotation. By using
this form, the equivalent diffusion factor can be applied to the more typical
cases encountered in axial-flow compressor cascades for both design and off-
design operation.
In principle, Fig. 6-11 or Fig. 6-12 can be used to estimate the loss coefficient at
the design incidence angle. But, in practice, the blade profile loss is only part of the
loss that occurs in an axial-flow compressor blade row. Other important sources of
the overall total pressure loss include effects due to tip clearance, stator shroud
leakage, end-wall boundary layers, Mach number, Reynolds number and secondary
flows. Howell (1942, 1945) recognized this fact many years ago and included addi-
tional loss models to account for the effects of end-wall and secondary flow losses.
Howell expresses these additional losses in terms of drag coefficient, CD, where
ω = CDσ cos2 β1 / cos3 β (6-40)

tan β = (tan β1 + tan β2 ) / 2 (6-41)
The end-wall drag coefficient used by Howell is
CDa = 0.02s / h (6-42)
where h is the blade height. Howell’s secondary flow drag coefficient is based on
the blade lift coefficient, CL
CDs = 0.18CL2 (6-43)
CL = 2 cos β (tan β1 − tan β2 ) / σ (6-44)
The loss coefficient correlation of Fig. 6-11 presumably includes at least the sec-
ondary flow loss, since it is adjusted to reflect the higher losses seen in compres-
sor cascades relative to simple two-dimensional cascades. That is not the case for
the correlation in Fig. 6-12. Neither correlation is expected to account for other
loss sources listed above. Where possible, these additional losses should be
approximated by specific empirical models. Approximate methods to model Mach
number effects and tip clearance effects are discussed later in this chapter. Some
loss sources can often be neglected, such as Reynolds number effects. Attempts to
model end-wall boundary layer losses have been reported in the literature, but
none that appear to be particularly general or reliable. It is often more effective to
adjust the profile loss models to reflect the higher losses expected in a compressor
cascade. Equations (6-34) and (6-37) can be written in the more general form of
ω * cos β2*
= K1[K2 + 3.5D* + 37( D* )4 ] (6-45)
2σ
2
ω * cos β2*  W*  * 2 * 8
1
 *  = K1[K2 + 3.1( Deq − 1) + 0.4( Deq − 1) ] (6-46)
2σ  2
W

The parameters, K1 and K2, are empirical constants or functions to be determined

by comparing performance predictions with experiment for a suitable number
and variety of axial-flow compressors. Their precise form depends on how the per-
formance analysis is formulated, particularly with respect to which loss sources
are modeled explicitly and which sources are lumped into K1 and K2. The
approach adopted for this writer’s performance analysis is discussed at the end of
this chapter, along with some variations and alternatives from the literature.
The more general loss coefficient model of Fig. 6-12 or Eq. (6-46) appears to
be the better choice. There seems to be some resistance to that choice due to its
development from specific blade loading distribution information. It is some-
times argued that its application to other blade camberlines and profiles requires
specific blade loading information for development of a new empirical correla-
tion. But that really doesn’t justify choosing the model of Fig. 6-11 and Eq. (6-45)
instead. In truth, the definition of D given in Eq. (6-31) has an inherent assump-
tion regarding the blade loading distribution, even though specific blade loading
information was not used to define it. There is no reason to believe that D is a
more general parameter than Deq, when it is applied to blade types different from
those for which its loss coefficient empirical model was developed.
6.5 POSITIVE AND NEGATIVE STALL INCIDENCE ANGLES

Figure 6-13 is a schematic showing the typical variation of loss coefficient with
incidence angle. The loss coefficient is fairly constant over a range of incidence
FIGURE 6-13 Off-Design Loss Coefficient

angles near the design incidence angle, but increases rapidly when the cascade
is operated too far from the design incidence angle. It is conventional practice to
define the limits of low-loss operation by the positive and negative stall inci-
dence angles, is and ic, where the loss coefficient becomes twice the minimum
loss coefficient, as shown in Fig. 6-13. Herrig et al. (1957) developed an approx-
imate correlation of the positive and negative stall angles of attack for NACA 65-
series blades from low Mach number two-dimensional cascade test data. Figure
6-14 shows their correlation along with data computed from the following
empirical equations:
  0.48 
30   θ
α c − α * = −9 + 1 −  (6-47)
  β1c   4.176
 
 β1 s  θ
α s − α * = 10.3 + 2.92 −  (6-48)
 15.6  8.2
where α, β1 and θ are all expressed in degrees. Note that Eq. (6-47) is singular if
β1c = 0. This writer’s practice is to limit β1c ≥ 20°. Since α is a function of β1, these
equations are not directly usable in a performance analysis. But since β1 = α + γ,
they can be applied by a simple iterative solution. Since α – α* is simply the inci-
dence angle range to stall, these empirical models would be expected to be rea-
FIGURE 6-14 Stall Angles of Attack

sonable approximations for blade types other than the NACA 65-series. Indeed,
they have proven to be effective when used for performance analysis of double-
circular-arc blades, which suggests they should be applicable to the C4 circular-
arc blades as well. One exception is the NACA A4K6 camberline where cascade
test data shows a nearly constant range from the design angle of attack, given
approximately by αs – α* = 10° and αc – α* = –10° (Dunavant, 1957). This may sug-
gest that a correction is needed for blades in which the point of maximum cam-
ber is not at mid-chord, but it is more likely that the constant incidence ranges to
stall is due to the fact that the inlet flow angle is constant for inlet guide vanes.
Nevertheless, it should be recognized that the applicability of these empirical
models to the parabolic-arc camberline has not been established.
6.6 MACH NUMBER EFFECTS
When applying the low-speed cascade empirical correlations to an actual com-

pressor blade row, it is necessary to apply corrections to account for Mach num-
ber effects. As Mach number increases, the low-loss working range for the
cascade, αs – α* and α* – αc, is reduced relative to the low-speed cascade correla-
tions of Eqs. (6-47) and (6-48). These low-loss working ranges for low-speed cas-
cades will be designated as Rs and Rc, i.e.,
R c = α * − α c = i* − ic (6-49)
R s = α s − α * = is − i* (6-50)
Johnsen and Bullock (1965) note that as Mach number increases (i* – ic) and
(is – i*) are reduced by approximately the same amount for moderate Mach
number blade profiles such as the NACA 65-series and C4-series blades. But
for high Mach number blade profiles such as the double-circular-arc profile,
(i* – ic) is reduced much faster than (is – i*) as Mach number increases. This
writer computes the negative and positive stall incidence angles for all blade
sections by
ic = i* − R c/ [1 + 0.5M1′3 ] (6-51)
is = i* + R s/ [1 + 0.5( K sh M1′)3 ] (6-52)
but with the constraint that Ksh ≤ 1. The negative stall incidence angle is also
required to be no less than the value corresponding to an inlet flow angle for
which the mass flow rate is 2% from the blade choke mass flow. An appropri-
ate equation of state from Chapter 2 is used to compute the sonic flow gas den-
sity, ρ*, and velocity, W*, for the local inlet relative total thermodynamic
conditions. Assuming the stream sheet thickness is constant between the inlet
and the throat, basic conservation of mass yields the inlet flow angle corre-
sponding to choke:

ρ1W1s cos β1Choke = oρ*W* (6-53)
This imposes a lower limit on the negative stall incidence angle given by
ic ≥ β1Choke − κ1 + 1o (6-54)
Hence, the approach to choked flow, where i will be less than ic , will be associated
with large and rapidly increasing loss coefficients as illustrated in Fig. 6-13. In prin-
ciple, the loss should become infinite when the passage is choked. But the milder
choke condition used here is more appropriate for an axial-flow compressor per-
formance analysis. Blade passage choke is often a local condition along the blade
height, which causes a redistribution of the mass flow toward unchoked sections of
the blade. If the loss increase near choke is too severe, a numerical solution will
often diverge before this redistribution can occur. Beyond that, this simple one-
dimensional flow choke calculation is not precise enough to treat it as an absolute
limit. The procedure suggested here has been found to be an effective compromise
for use in an axial-flow compressor aerodynamic performance analysis.
When these Mach number adjustments are imposed, it may also be necessary
to readjust the minimum loss incidence angle and the minimum loss coefficient.
Indeed, it is quite possible for the design incidence angle, i*, to be less than ic. Fig-
ure 6-15 illustrates the type of corrections that may be imposed on the off-design
loss coefficient at elevated Mach number levels. A minimum loss incidence angle,
im, is defined by
FIGURE 6-15 Mach Number Corrections

im = ic + (is − ic ) R c / ( R c + R s ) (6-55)
For moderate Mach number blade profiles far from the choked flow limit, it can
be seen that im = i*. But for higher Mach number profiles, or cases where limit in
Eq. (6-54) becomes active, Eq. (6-55) results in im > i*, similar to the schematic
shown in Fig. 6-15. As a minimum, the minimum loss coefficient must be
adjusted for the difference between im and i*, using the low Mach number loss
curve directly. For incidence angles between ic and is, this writer uses a simple
second-order power law relation for the off-design loss coefficient, i.e.,
ω = ω m + ω m [(i − im ) / (ic − im )]2 ; i ≤ im (6-56)

2
ω = ω m + ω m [(i − im ) / (is − im )] ; i ≥ im (6-57)
It follows directly that the minimum correction required for the minimum loss
coefficient is
ω m = ω * [1 + (im − i* )2 / R2s ] (6-58)
Except for this fairly minor correction, loss coefficients at the minimum loss inci-
dence show little variation with Mach number until the fluid velocities on the blade
surfaces become supersonic. At that point, shock waves can form locally, eventu-
ally causing boundary layer separation to significantly increase the minimum loss
coefficient. Equation (6-39), applied with i = i* and δ = δ*, defines the maximum
velocity, Wmax, as illustrated in Fig. 6-10. It follows that the critical Mach number,
M′c, where the flow first becomes supersonic on the blade surfaces, is given by
Mc′ = M1′ W* / Wmax (6-59)
When the inlet Mach number exceeds the critical Mach number, the minimum
loss coefficient is estimated from
ω m = ω * [1 + (im − i* )2 / R2s ] + K sh [( M′ / Mc′ − 1)W* / W1]2 (6-60)
When applying Eq. (6-60), the limit M′ ≤ 1 is imposed since bow shock wave
losses are handled separately, as will be seen in the next section.
6.7 SHOCK WAVE LOSS FOR SUPERSONIC CASCADES
The Mach number effects described in the previous section do not account for
the additional loss caused by the upstream shock wave when the flow entering
the blade row is supersonic. Semi-empirical models to correct for bow shock
losses have been reported by Swan (1961) and Miller et al. (1961), both of which

employ essentially identical models. Swan’s description is perhaps a little more

complete and will be used as the basis for the present discussion. The physics of
the problem considered is illustrated in Fig. 6-16. The flow approaches the blade
row at a supersonic velocity, passes through an oblique bow shock wave and
undergoes a supersonic expansion over the suction surface of the blade. Finally,
the expanded, higher Mach number flow passes through the stronger normal
shock wave denoted as the passage shock wave in Fig. 6-16. The approach used
is to compute the blade surface Mach number modeled as a simple Prandtl-
Meyer isentropic expansion. The passage shock loss is then calculated using an
upstream Mach number that is the average of the blade inlet Mach number and
the expanded flow blade surface Mach number. Here, Swan’s (1961) ideal gas
model will be generalized somewhat so that it can be used for non-ideal gas
equations of state as well. The Prandtl-Meyer expansion is accomplished for a
basic double-circular-arc blade profile. The flow is assumed to approach the
blade row at a flow angle tangent to the blade suction surface. This is considered
fairly representative of the geometry for any profile likely to be used for super-
sonic inlet flow. And the rather narrow incidence angle operating range expected
for supersonic inlet flow also makes the suction surface tangency assumption
reasonable. From Figs. 6-16 and 6-17, it is seen that the flow must expand
through an angle φ along the suction surface. If the blade profile is assumed to
have a sharp leading edge, Eqs. (4-7) and (4-10) can be used to define the upper
or suction surface of the blade as
2Ru / c = sin(θ u / 2) (6-61)

2bu / c = tan(θ u / 4) (6-62)
FIGURE 6-16 Shock Wave Loss Model

FIGURE 6-17 Expansion Angle Geometry
where θu is the upper surface “camber angle” and Ru is the upper surface radius
of curvature. Similarly, the mean camberline circular-arc can be expressed by
2R / c = sin(θ / 2) (6-63)
2b / c = tan(θ / 4) (6-64)
where θ is the true camber angle, R is the camberline radius of curvature, and b
is analogous to bu of Fig. 6-17, but for the mean camberline. Since bu – b = tb / 2,
as can be seen in Fig. 4-8, it follows that
tan(θ u / 4) = tan(θ / 4) + tb / c (6-65)
For a circular-arc, the angle of the arc with the chord line at the inlet is given by
Eq. (4-35), i.e.,
χ u1 = θ u / 2 (6-66)
Since the flow angle is assumed to be equal to the suction surface inlet angle,
Eqs. (4-32) and (6-66) yield
β1 = 90o − ψ = θ u / 2 + γ (6-67)
ψ = 90o − θ u / 2 − γ = 90o − θ u / 2 − κ1 + θ / 2 (6-68)
where γ is the blade stagger angle and κ1 is the blade inlet angle, as shown in Fig.
6-1. The law of sines and basic trigonometry applied to the triangle shown in Fig.
6-17 yields
s cos ψ
tan φ = (6-69)
s sin ψ + Ru

Swan (1961) develops Eq. (6-69) in a more general form by including the leading
edge nose radius. That is a relatively insignificant effect, considering the approx-
imate nature of the model. Hence, the assumption of a sharp leading edge is used
here to avoid the need to specify the blade nose radius for an aerodynamic per-
formance analysis.
The blade surface Mach number, M′S, and velocity, WS, entering the passage
shock wave is given by the well-known Prandtl-Meyer isentropic expansion
through the expansion angle φ. In general form, this can be expressed as
WS
dW
φ= ∫ M′ 2 − 1
W
(6-70)
W1
For thermally and calorically perfect gases, this equation can be integrated ana-
lytically and expressed in terms of the Prandtl-Meyer angle, which is usually des-
ignated as ν(M). To permit use of the model for any appropriate equation of state,
Eq. (6-70) must be integrated numerically, noting that the Prandtl-Meyer expan-
sion is an isentropic process.
The shock wave total pressure loss is calculated for a normal shock wave with
an inlet Mach number of
′ = M1′MS′
Min (6-71)
Again, for thermally and calorically perfect gases, the total pressure loss across a
normal shock wave can be expressed analytically. To generalize the calculation
for any appropriate equation of state, a numerical solution is required. This
involves simple conservation of mass, momentum and energy across the shock
wave in the form
( ρ W )in = ( ρ W )out (6-72)

2 2
( P + ρ W )in = ( P + ρ W )out (6-73)
Hin = Hout (6-74)
sout ≥ sin (6-75)
Equations (6-72) through (6-74) can be solved by simple iteration, noting that the
flow downstream of the shock wave is subsonic. Equation (6-75) requires that the
second law of thermodynamics be satisfied as stated in Eq. (2-6). Then the down-
stream total pressure is calculated to yield the desired total pressure loss across
the shock wave.
6.8 OFF-DESIGN CASCADE PERFORMANCE CORRELATIONS
The calculation of loss and fluid turning at off-design incidence angles is more
complex than for the design incidence angle. Although two-dimensional cascade

test data provide some useful guidance, the off-design performance of an annular
cascade in an axial-flow compressor is far more complex. Successive blade rows
in a compressor are normally closely spaced, resulting in significant interaction
between them. Two-dimensional cascade test data is based on measurements
rather far downstream where blade wakes and flow distortion is minimal. That is
far from the case encountered in a compressor. Rotating stall can significantly
influence the flow in a compressor cascade, particularly when the compressor
operates at rotation speeds well below its design speed. Two-dimensional cascade
test data provides no insight into this phenomenon. A very fundamental differ-
ence between compressor cascades and cascade testing is the fact that the flow is
far from two-dimensional in the compressor. In a compressor, significant
changes in axial velocity and radius across the cascade are common and end-wall
boundary layers often influence a substantial portion of the flow field. By con-
trast, substantial effort is made to minimize these effects in two-dimensional cas-
cade tests, typically by use of boundary layer suction.
Consequently, off-design blade row performance empirical model develop-
ment requires calibration against axial-flow compressor performance data,
including a range of compressor types and operating conditions. It follows that
these empirical models are substantially influenced by the overall strategy used
for the compressor performance analysis. In effect, it soon becomes impossible
to separate the parts from the whole. Numerous approximations are required to
model the through-flow, end-wall boundary layers, tip clearance, shroud leakage,
etc., in a performance analysis. It is important to recognize that the off-design
empirical correlations suggested here are simply methods that have been found
effective in the context of this writer’s performance analysis as it is described in
this book.
The blade row incidence angle and the axial velocity ratio across the cascade
significantly influence the off-design deviation angle. Johnsen and Bullock (1965)
provide an empirical model for variation of the deviation angle with incidence
angle at the design incidence angle. This graphical correlation is shown in Fig.
6-18, along with predictions from the following empirical equation:
*
 ∂δ  4 2.5
 ∂i  = [1 + (σ + 0.25σ )(β1 / 53) ] / exp(3.1σ ) (6-76)
 
G. Mellor has developed plots of β2 as a function of β1 from NACA cascade data

(see Fig. 3.9, Horlock, 1958). These are equivalent to plots of δ versus i. A
schematic typical of Mellor’s constant-stagger angle characteristics is shown in
Fig. 6-19. It can generally be concluded that the deviation angle approaches a
constant value near the negative stall incidence angle. At the positive stall inci-
dence angle, the slope of the β2 versus β1 curves approaches unity. For some time,
this writer used these two observations and Eq. (6-76) in several unsuccessful
attempts to construct a general deviation angle correlation as a function of inci-
dence angle to match the observed behavior at ic, i* and is, as suggested by Novak
(1973). While this produced reasonable results for certain applications and oper-
ating conditions, definite exceptions seemed to always exist where this type of
model could not match actual compressor performance data. The requirement

FIGURE 6-18 Off-Design Deviation Slope
FIGURE 6-19 Schematic of 2 Versus 1 in Cascades

that the fluid turning does not increase with incidence beyond the positive stall
incidence angle is particularly troublesome when analyzing the performance of
compressors. While this appears to be the case in cascade test results, it does not
permit predicting compressor performance at some of the very severe incidence
angles at which they are often capable of operating. After experimenting with
various modifications and limits, a somewhat arbitrary model was tried, assum-
ing a linear variation of deviation angle with incidence angle with a slope given
by Eq. (6-76). This produced such a dramatic improvement in the accuracy and
versatility of the compressor performance analysis that the attempt to define a
more general empirical correlation was abandoned. The axial velocity ratio
across the cascade also has a definite influence on the deviation angle. The cor-
rection developed by Pollard and Gostelow (1967) from cascade test results is
about as good as any, although it is based on rather limited evidence. Combining
their axial velocity correction with the assumed linear variation of deviation
angle with incidence angle results in the expression used by this writer for off-
design deviation angle prediction.
*
 ∂δ 
δ = δ * +   (i − i* ) + 10(1 − Wm2 / Wm1) (6-77)
 ∂i 
where i and δ are expressed in degrees. Inlet guide vanes require special treat-
ment. It was noted previously in this chapter that inlet guide vanes usually func-
tion like turbine nozzle blade rows, where the throat area at the blade discharge
largely dictates the discharge flow angle. Hence, the second term on the right-
hand side of Eq. (6-77) is omitted for inlet guide vanes, such that deviation angle
is not influenced by incidence angle. This is an important feature when compres-
sors employ adjustable inlet guide vanes. In those cases, the vanes may be oper-
ated at extremely large incidence angles. Comparison of predicted and measured
performance of those types of axial-flow compressors shows no evidence of any
incidence angle influence on the deviation angle of the inlet guide vanes.
Figure 6-20 illustrates the model used for off-design loss coefficient. Define a
normalized incidence angle parameter as
ξ = (i − im ) / (is − im ); i ≥ im (6-78)
ξ = (i − im ) / (im − ic ); i < im (6-79)
– , the off-
Designating the upstream shock wave loss coefficient (if any) by ω s
design loss coefficient is given by
ω = ω s + ω m [1 + ξ 2 ]; − 2 ≤ ξ ≤ 1 (6-80)
ω = ω s + ω m [5 − 4(ξ + 2)]; ξ < −2 (6-81)
ω = ω s + ω m [2 + 2(ξ − 1)]; ξ > 1 (6-82)
As noted in Fig. 6-20, Eqs. (6-81) and (6-82) are simple linear extrapolations of
Eq. (6-80) outside of its designated range of application. The values of the loss

FIGURE 6-20 Off-Design Loss Coefficient
coefficients at ic, im and is are consistent with the definitions of those incidence
angles established earlier in this chapter. The remainder of the correlation is
completely empirical, based on optimizing the performance prediction accuracy
for a wide range of axial-flow compressor types and operating conditions. The
lower limit imposed on Eq. (6-80) is largely a safety feature, since operation at
such large negative incidence angles is almost never encountered. By contrast,
the upper limit imposed on Eq. (6-80) and the extrapolation defined by Eq. (6-82)
are essential, and are commonly encountered, particularly when an axial-flow
compressor operates at speeds well below its design speed. Limiting the rate of
increase in loss coefficient with incidence angle in these conditions really com-
pensates for the fact that the compressor is likely to operate in rotating stall.
These stall zones effectively block some of the blade passages such that the inci-
dence angles are not as large as those indicated by an ideal, axisymmetric flow
analysis. This writer’s performance analysis makes further provision for this type
of operation by imposing an area blockage, Bwake, due to the blade wakes. This is
based on the blade loading level indicated by Lieblein’s equivalent diffusion fac-
tor evaluated at off-design conditions, i.e., using the off-design incidence and
deviation angles in Eq. (6-39) to compute the off-design equivalent diffusion fac-
tor using Eq. (6-30). The wake blockage is given by
Bwake = 0; Deq ≤ 2 (6-83)
Bwake = 1 − (2 / Deq )0.9 ; Deq > 2 (6-84)

It will be seen later in this book that the through-flow analysis interprets Bwake as
the fraction of the stream sheet area that is unavailable for through flow.
6.9 BLADE TIP CLEARANCE LOSS

The calculation of the blade tip clearance loss is based on the same semi-empiri-
cal model used for centrifugal compressor impellers in Aungier (2000). Figure
6-21 shows the tip clearance geometry for a typical rotor blade. The situation for
unshrouded stator blades is similar, except that the clearance lies along the hub
contour. The pressure difference on the two sides of the blade produces a leakage
flow through the clearance gap, basically dissipating the pressure difference. The
pressure difference across the blade must balance the blade torque as given in
Eq. (3-7). For the clearance gap, this can be expressed as
τ = πδ c [( rρ Cm )1 + ( rρ Cm )2 ][r2Cθ 2 − r1Cθ1] (6-85)
The average pressure difference across each blade in the blade row is
∆P = τ / ( Zrtipδ cc cos γ ) (6-86)
where Z is the number of blades in the blade row. The fluid velocity of the leakage
flow is estimated from ∆P and the assumed throttling coefficient of Aungier
FIGURE 6-21 Blade Tip Clearance Geometry

(2000) for the first blade row, but is reduced as the blade row number, Nrow,
increases, i.e.,
0.2
Uc = 0.816 2∆P / ρ / Nrow (6-87)
As will be seen in Chapter 8, end-wall boundary layer growth in multistage

axial-flow compressors results in a substantial tangential blade force defect,
which is expected to reduce the tip leakage flow. The dependence on Nrow was
determined empirically from comparison of predicted and measured perform-
ance of several multistage axial-flow compressors. The leakage mass flow rate
is given by
˙ c = ρ UcZδ cc cos γ
m (6-88)
The clearance gap total pressure loss for the entire blade row is
∆Pt = ∆P m
˙c/m
˙ (6-89)
This total pressure loss is clearly concentrated in the clearance gap region of the
flow field. But if it is applied in that fashion, losses along the end-wall will accu-
mulate from each blade row, eventually causing the through-flow analysis to
diverge. The basic problem is that the flow in a multistage compressor undergoes
considerable fluid mixing at each blade row due to secondary flows. But a con-
ventional through-flow analysis does not account for this. Hence, once a loss is
imposed on a stream sheet in the analysis, it stays on that stream sheet through
the remainder of the compressor. To avoid solution divergence, this writer
imposes the leakage total pressure loss as a linear distribution across the annu-
lus, such that the integrated ∆Pt is equal to the value given by Eq. (6-89), but the
total pressure loss is zero at the wall opposite from the clearance gap. Since suc-
cessive blades encounter clearance (or shroud seal leakage) losses on alternate
end-wall contours, this produces a mild concentration of these losses near the
end-walls, but with about half of the loss applied at mid-passage. Experience has
shown that this yields a stable performance analysis that correlates well with
overall compressor experimental performance data.
6.10 SHROUD SEAL LEAKAGE LOSS

Figure 6-22 illustrates an alternate style of blade tip geometry that is often used
on stator blade rows. Here a shroud band is attached integral with the blades.
Typically seal fins are attached to the shroud band to reduce the clearance gap
leakage flow, yet protect the compressor from serious damage in the event that a
rotor excursion should cause the shaft to come in contact with the shroud. The
seal allows larger clearances between the shroud band and the shaft to be used
without causing excessive leakage. If contact occurs, the seal strips will be sacri-
ficed, but the more expensive components should be undamaged.

FIGURE 6-22 Shroud Seal Geometry
The total pressure loss due to shroud leakage is that given by the leakage mass
flow and the pressure difference across the blade row, i.e.,
∆Pt = ∆P m
˙ leak / m
˙ (6-90)
This total pressure loss is distributed across the annulus in exactly the same fash-
ion as the tip clearance loss. The leakage mass flow through the labyrinth seal
can be computed by the method of Egli (1935), who expresses the leakage mass
flow in the form
ṁleak = 2π rsealδ cCtCcCr ρ RT (6-91)
Aungier (2000) provides general empirical equations to approximate the coeffi-

cients in Egli’s model. Here, a simplified model can be used, since the number of
fins, N, in a shroud seal is always relatively small. The contraction ratio is
approximated by
1
Cr = 1 − 3.45
 54.3  (6-92)
3+  
1 + 100δ c / t 

If PR designates the ratio of the lower pressure to higher pressure across the seal,
the throttling coefficient is approximated by
2.143 ln( N ) − 1.464

Ct = [1 − PR ](0.375PR ) (6-93)
N − 4.322
If N ≤ 12, the carryover coefficient is approximated as
X1[δ c / p − X2 ln(1 + δ c / p)]

Cc = 1 + (6-94)
1 − X2
X1 = 15.1 − 0.05255 exp[0.507(12 − N )] (6-95)
X2 = 1.058 + 0.0218N (6-96)
Equation (6-94) yields a maximum when δc / p = X2 – 1. This represents a poor seal

design, which exceeds the range of Egli’s model. It is rarely encountered, but it is
wise to require δc / p ≤ X2 – 1 when applying Eq. (6-94). Aungier (2000) provides a
more detailed discussion and the extension of the empirical equations to N > 12.
6.11 IMPLEMENTATION, EXTENSIONS AND ALTERNATE

METHODS
Integration of empirical performance models into an axial-flow compressor per-

formance analysis is by no means a trivial process. Chapters 7 and 8 discuss the
through-flow analysis and end-wall boundary layer analysis, which are the other
two important components of an aerodynamic performance analysis. It is con-
venient to discuss the component parts of the performance analysis individually,
but the parts cannot really be separated from the whole. An objective of this book
is to provide a complete description of the methods used by the author. Comple-
tion of the description of the empirical loss models requires a definite qualifica-
tion. The final adjustments suggested here depend on the other components of
the performance analysis to be discussed in the next two chapters, as well as on
the use of the complete set of empirical models reviewed in this chapter. The final
adjustments may be ineffective if used in a different context.
To complete this discussion, the parameter, K1 and K2, in Eq. (6-46) must be
provided. These parameters adjust the profile loss model to compensate for loss
sources not specifically modeled here, such as end-wall and secondary flow
losses. In the simplest form, K1 and K2 may be considered as empirical constants.
That is actually not a bad assumption. Rather good performance prediction accu-
racy has been achieved simply using K1 = 0.004 and K2 ≈ 4. That is possible
because extremely low values of the aspect ratio are not normally encountered in
typical compressors. It is more prudent to include specific provision for aspect
ratio effects as suggested by Howell (1942, 1945) in Eq. (6-42). Consider a simple
model for skin friction loss in a cascade. If cf is the skin friction coefficient, this
can be represented by

ω SF ∝ cf c / ( s cos β2 ) (6-97)
This can be extended to include an approximate end-wall loss estimate by adding

the blade chord-to-height ratio, i.e.,
ω SF ∝ cf [c / ( s cos β2 ) + c / h] (6-98)
This simple argument leads to the following expression for K2:
K2 = 1 + ( s / h) cos β2 (6-99)
The correction is applied to K2 rather than K1, since blade loading is not expected
to significantly influence the end-wall loss. By comparison of performance pre-
dictions with experiment for a several axial-flow compressors, K1 = 0.0073, com-
bined with Eq. (6-99), was found to yield good results, as will be demonstrated in
Chapter 9.
It is also useful to include an approximate correction for Reynolds number.
Although not normally used in the author’s performance analysis, there are
extreme cases where a correction may be necessary. The Reynolds number cor-
rection is applied only to the skin friction portion of the simple cascade profile
loss, as represented by Eq. (6-37). For application to Eq. (6-46), Eq. (6-99) is gen-
eralized to the form
K2 = 1 + ( s / h ) cos β2 + 0.004KRe / K1 (6-100)
For laminar flow (Rec < 2.5 × 105),
KRe = 2.5 x 105 / Rec − 1 (6-101)
For turbulent flow (Rec > 2.5 × 105),
[ ]
2.58
KRe = log(2.5 x 105 ) / log(Rec ) −1 (6-102)
These corrections are based on the blade chord Reynolds number, Rec, derived
from classical Reynolds number formulations for boundary layer skin friction
coefficients. The skin friction models used can be found in a variety of books on
boundary layer theory, e.g., Schlichting (1968, 1979). These corrections are
consistent with the correlation in Fig. 6-12, which is based on data with
Rec ≈ 2.5 × 105. Fig. 6-23 illustrates the basis of this Reynolds number correction.
Cascade loss coefficient data from Johnsen and Bullock (1965) are shown as a
function of blade chord Reynolds number. The empirical correction curves
shown are obtained by normalizing Eqs. (6-101) and (6-102) to the data point
closest to Rec = 2.5 × 105 for each data set.

FIGURE 6-23 Reynolds Number Effect
It is appropriate to close this chapter with a reference to an excellent technical

paper by Koch and Smith (1976). They present a very careful development of a
significant alternate approach to the present methods. Koch and Smith rely on
an independent empirical model for end-wall loss and end-wall work formulated
from compressor test data. They developed a profile loss model using a com-
pressible boundary layer analysis, including corrections for Mach number,
Reynolds number and stream sheet contraction. An alternate model for bow
shock loss is also included. Attempts by this author to employ their approach
many years ago were unsuccessful, due to the fact that their end-wall loss model
simply did not correlate with experimental data for the specific class of axial-flow
compressors of interest at the time. But Koch and Smith have shown good agree-
ment with experiment for a significant number of cases, suggesting that this
author’s experience may not be typical. Indeed, this reference has substantially
influenced this author’s performance analysis, even though its empirical per-
formance models are not used directly. This will become apparent in Chapter 8 of
this book. It is a very significant reference that is highly recommended.
EXERCISES
6.1 Consider NACA 65-series blades with tb / c = 0.1. Hence, a / c = 0.5 and
Ksh = Kt,i = 1. Based on Eq. (6-8), what are the independent variables
defining the design incidence angle? Repeat for Eq. (6-12).

6.2 Repeat Exercise 6.1, but consider the design inlet angle, β*1, as the
dependent variable, instead of i*.
6.3 A performance analysis has been developed following the procedures
of this chapter, but using Eq. (6-12) as the basis for the design inci-
dence angle. Is there any reason to compute α* using Eq. (6-8)?

Chapter 7
MERIDIONAL THROUGH-FLOW
ANALYSIS
The term meridional through-flow analysis refers to an analysis of the flow in the
meridional plane, i.e., a plane defined by a constant polar angle of cylindrical
coordinates. A solution in the meridional plane can completely characterize the
flow field if the flow is locally axisymmetric. This is usually considered to be a
reasonable approximation for hub-to-shroud computing stations located outside
of the blade rows. Hence, common practice is to locate all hub-to-shroud com-
puting stations in a meridional through-flow analysis before, between or after the
blade rows. This requires a means to define the influence of the blade rows in a
form that can be imposed on the solution. Typically, this is accomplished by spec-
ifying the flow angle or swirl velocity and the entropy rise or total pressure loss
associated with flow passing through the blade row. In the case of performance
analysis of an existing axial-flow compressor design, the empirical models of
Chapter 6 can be used. When designing an axial-flow compressor, the influence
of the blade rows is specified directly. When the meridional through-flow analy-
sis is completed, the geometry of the blade rows is selected to produce the speci-
fied influence.
A properly formulated meridional through-flow analysis is a very powerful
technique that can be used to support a variety of axial-flow compressor aerody-
namic design and analysis functions. In this book, this technique will be used for
aerodynamic performance analysis, general stage design and complete axial-flow
compressor design. This chapter develops the governing equations and describes
methods of solution appropriate to these various applications.
NOMENCLATURE
A = annulus area
a = sound speed
Bwake = wake blockage
B* = stream surface repositioning damping factor
F = general function
f = general function

H = total enthalpy
h = static enthalpy
I = rothalpy
KB = boundary layer blockage factor
KW = wake blockage factor
M = Mach number
P = pressure
r = radius
s = entropy
T = temperature
y = coordinate along a quasi-normal
β = flow angle with m-direction
ε = φ – λ; deviation of quasi-normal from a true stream surface normal
κm = stream surface curvature
λ = quasi-normal angle, Eq. (7-1)
φ = stream surface angle with axial direction
θ = polar angle
ρ = gas density
ω = rotation speed (radians/second)
Subscripts
h = hub parameter
s = shroud parameter
1 = condition at point preceding point being considered
2 = condition at point being considered
3 = condition at point following point being considered
Superscripts
* = parameter on stream surface where Wm is specified for annulus
sizing
7.1 Meridional Coordinate System
Figure 7-1 illustrates the basic meridional coordinate system used in a merid-
ional through-flow analysis. A series of meridional computing stations or

Meridional Through-Flow Analysis • 155
FIGURE 7-1 Definition of the Problem
straight-line quasi-normals are distributed through the compressor, all of which

lie outside of the blade rows. The annulus is divided into a series of stream tubes
separated by stream surfaces. A stream surface is defined as a surface having no
mass flow across it, or equivalently, no velocity component normal to it. Hence,
the hub and shroud contours always define stream surfaces. The intermediate
stream surfaces are normally defined by requiring that the mass flow between
each stream surface and the hub stream surface be constant through the solution
domain. Quasi-normals in axial-flow compressors are often simple radial lines,
although provision should be made for non-radial quasi-normals to treat cases
such as that of the inlet section illustrated in Fig. 7-1.
Figure 7-2 illustrates the quasi-normal construction in further detail. As the
name implies, the intention is to construct the quasi-normal to be approximately
normal to the stream surfaces. The quasi-normal is defined by the (z, r) coordi-
nates of both end points or by one end point plus the quasi-normal angle, λ. As
illustrated in Fig. 7-2, the quasi-normal angle is defined by
∆z zh − zs
tan λ = = (7-1)
∆r rs − rh
The stream surface slope angle, φ, is given by
∂r (7-2)
sin φ =
∂m

FIGURE 7-2 Quasi-Normal Construction
where m is the meridional coordinate, measured along a stream surface. Noting

that λ is also the angle between a normal to the quasi-normal and the axial direc-
tion, the angle, ε, between a quasi-normal and a true stream-surface normal is
given by
ε =φ−λ (7-3)
Quasi-normals do not have to be defined with extreme precision, but it is pre-

ferred that ε be relatively small if possible to achieve better numerical accuracy.
This means that ε on the hub-and-shroud contours should be approximately
equal in magnitude, but opposite in sign. In axial-flow compressors, this is usu-
ally not an important consideration since a simple radial quasi-normal normally
results in very small values of ε. But for cases like that of the inlet passage illus-
trated in Fig. 7-1, a little more care may be needed.
In Chapter 3, the governing equations for adiabatic inviscid flow were devel-
oped in the natural coordinate system (θ, m, n). This coordinate system is conven-
ient for derivation of the equations, but less convenient for solution. The reason is
that two of the coordinates, m and n, must be determined as part of the solution.
It is advantageous to maintain the stream surface coordinate, m, despite the need
to compute it in the solution, since many of the basic conservation equations
explicitly apply along m. But the determination of n is an unnecessary complica-
tion offering no benefit to a numerical analysis. Katsanis (1964) suggested the use

of the quasi-normal coordinate, y, shown in Fig. 7-2, to avoid that complication. It

is easily shown that derivatives with respect to n can be expressed in the form
∂ 1 ∂ ∂ 
=  − sin ε  (7-4)
∂n cos ε  ∂y ∂m 
Since y is a fixed coordinate at all computing stations, the numerical analysis is

greatly simplified.
7.2 INVISCID ADIABATIC FLOW ON A QUASI-NORMAL
The relevant equations for adiabatic inviscid flow are developed in general form
in Chapter 3. For the present application, Eqs. (3-21), (3-25), (3-29) and (3-30)
can be used after they are simplified to their axisymmetric, time-steady form. It
will be convenient to satisfy conservation of mass in integral form instead of
using Eq. (3-21). Conservation of mass along a quasi-normal can be expressed in
the form
ys
˙ = 2πK B ∫ KW rρ Wm cos ε dy
m (7-5)
0
KB is the end-wall boundary layer blockage factor, which corrects the area avail-
able for through-flow for viscous blockage effects. It is the fraction of the total
area available for through-flow after subtracting the hub-and-shroud boundary
layer displacement thicknesses from the overall quasi-normal length. Methods to
estimate KB will be described in Chapter 8. For now, it is simply recognized that
it must be specified in some manner. KW is the blade wake blockage factor, which
serves a similar purpose in correcting for local wake blockage. From Eqs. (6-83)
and (6-84), it is given by
KW = 1 − Bwake (7-6)
The axisymmetric, time-steady tangential and normal momentum equations fol-

low directly from Eqs. (3-29) and (3-30).
∂( rWθ + ω r 2 ) ∂rCθ (7-7)

= =0
∂m ∂m
Wθ ∂( rWθ + ω r 2 ) ∂Wm ∂I ∂s
κ mWm2 + + Wm = −T (7-8)
r ∂n ∂n ∂n ∂n
where κm is the stream surface curvature given by Eq. (3-26), i.e.,
∂φ
κm = − (7-9)
∂m

The axisymmetric, time-steady energy equation follows directly from Eq. (3-25).
∂I
=0 (7-10)
∂m
Substitution of Eqs. (7-7) and (7-10) into Eq. (3-28) shows that for axisymmetric
flow, entropy must also be conserved in the meridional direction, i.e.,
∂s
=0 (7-11)
∂m
At first glance, Eqs. (7-7) and (7-11) would appear to preclude use of this model
for axial-flow compressors. It is known that entropy and angular momentum
change along stream surfaces when the flow passes through blade rows. But
these equations do not preclude changes between quasi-normals. They only
require that the local gradients be zero for the flow to be locally axisymmetric.
Introducing Eq. (7-4) into Eq. (7-8), and simplifying the result with Eqs. (7-7),
(7-10) and (7-11), yields the following expression for the normal momentum
equation.
∂Wm ∂Wm Wθ ∂( rCθ ) ∂I ∂s

Wm + κ mWm2 cos ε − Wm sin ε + = −T (7-12)
∂y ∂m r ∂y ∂y ∂y
Equation (7-12) can also be expressed in terms of the relative flow angle, noting
that Wθ = Wm tanβ′.
∂Wm ∂Wm Wm tan β ′ ∂( rWm tan β ′ + ω r 2 )

Wm + κ mWm2 cos ε − Wm sin ε +
∂y ∂m r ∂y
∂I ∂s (7-13)
= −T
∂y ∂y
After some basic algebra and trigonometry, Eq. (7-13) simplifies to the form
Wm ∂Wm 2 ∂Wm Wm2 tan β ′ ∂( r tan β ′ )

+ κ m Wm cos ε − Wm sin ε +
cos2 β ′ ∂y ∂m r ∂y
(7-14)
∂I ∂s
+2Wmω tan β ′ cos λ = −T
∂y ∂y
It is convenient to express the normal momentum equation in the following gen-

eral form:
∂Wm f ( y)
= f1( y)Wm + f2 ( y) + 3 (7-15)
∂y Wm

When Cθ and Wθ are known, the functions in Eq. (7-15) are
sin ε ∂Wm
f1( y) = −κ m cos ε + (7-16)
Wm ∂m
f2 ( y) = 0 (7-17)
∂I ∂s Wθ ∂( rCθ )
f3 ( y) = −T − (7-18)
∂y ∂y r ∂y
When β′ is known, the functions are
 tan β ′ ∂( r tan β ′ ) sin ε ∂Wm 

f1( y) = cos2 β ′ −κ m cos ε − +  (7-19)
 r ∂y Wm ∂m 
f2 ( y) = −2ω cos β ′ sin β ′ cos λ (7-20)

 ∂I ∂s 
f3 ( y) = cos2 β ′  − T  (7-21)
 ∂y ∂y
Numerical integration of Eq. (7-15) is straightforward if the functions f1, f2 and f3

are truly functions of y only. But it can be seen that f3 includes a term involving
the meridional gradient of Wm, which appears to depend on the flow solution on
other quasi-normals. Novak (1967) shows that this difficulty can be removed by
use of the differential form of the continuity equation, i.e., Eq. (3-21). For time-
steady, axisymmetric flow, this reduces to
∂( rρ Wm )
+ κ n rρ Wm = 0 (7-22)
∂m
Combining Eqs. (3-27) and (7-21) and expanding the result yields
1 ∂Wm 1 ∂ρ sin φ ∂φ
+ + + =0 (7-23)
Wm ∂m ρ ∂m r ∂n
Since entropy is conserved along stream surfaces, the term involving the gradient
of ρ can be expanded using Eq. (2-26) to yield
1 ∂ρ 1 ∂P  ∂ρ  1 ∂P
=   = (7-24)
ρ ∂m ρ ∂m  ∂P  s ρ a2 ∂m
Finally, introducing Eq. (3-22) for time-steady axisymmetric flow yields
1 ∂ρ 1  C 2 sin φ ∂Wm 
= 2 θ − Wm  (7-25)
ρ ∂m a  r ∂m 

Combining Eqs. (7-23) and (7-25) yields
1 ∂Wm 2 sin φ ∂φ
(1 − Mm ) = −(1 + Mθ2 ) − (7-26)
Wm ∂m r ∂n
where the meridional and tangential Mach numbers are given by
Mm = Wm / a (7-27)
Mθ = Cθ / a (7-28)
Introducing Eqs. (7-4) and (7-9) into Eq. (7-26) yields
1 ∂Wm 2 sin φ 1 ∂φ
(1 − Mm ) = −(1 + Mθ2 ) − − κ m tan ε (7-29)
Wm ∂m r cos ε ∂y
Hence, Eq. (7-29) can be used to evaluate the meridional gradient of Wm on any
quasi-normal, independent of the solution on other quasi-normals. It is necessary
to take the precaution of avoiding a singularity only if Mm = 1 should occur. This
writer imposes the following constraint when applying Eq. (7-29).
2
1 − Mm ≥ 0.1 (7-30)
Equation (7-15) can also become singular if Wm = 0. Aungier (2000) avoids that
problem by using conservation of mass in a stream tube in the form
∆ṁ = ρ Wm ∆A (7-31)
The stream tube area term is given by
∆A = 2π rKW cos ε ∆y (7-32)
Typically, all stream tubes are assumed to contain equal mass flows, although alter-
nate definitions can certainly be used. Now introduce the function, f4, given by
ρ ∆A (7-33)
f4 ( y) = f2 ( y) + f3 ( y)
∆m ˙
Then, Eq. (7-15) can be written as
∂Wm
= f1( y)Wm + f4 ( y) (7-34)
∂y
The solution of this linear differential equation can be found in almost any text-
book on differential equations as

y
f4 ( y)
Wm ( y) = Wm (0)F( y) + F( y)∫ dy (7-35)
F ( y)
0
where the function F(y) is given by
y 
F( y) = exp  ∫ f1( y)dy (7-36)
 
0 
The meridional velocity on the hub contour, Wm(0), is the constant of integration
for Eq. (7-34). It is determined from conservation of mass through Eq. (7-5).
Equations (7-34) and (7-5) are solved in an iterative numerical scheme, succes-
sively improving the estimate of Wm(0) until mass is conserved and the normal
momentum equation is satisfied. This requires calculation of thermodynamic
properties such as ρ and a, using an appropriate equation of state from Chapter
2. At any point, the relative total enthalpy is given by Eq. (3-13), i.e.,
H ′ = I + 12 ( rω )2 (7-37)
The local static enthalpy is given by
h = H ′ − 12 W 2 (7-38)
Then static thermodynamic conditions are computed from relative total thermo-
dynamic conditions for the change in enthalpy, (h – H′), while holding entropy
constant.
When computing the flow profile on a quasi-normal, it is also necessary that
the numerical analysis be able to recognize choked flow. The choke condition
corresponds to the maximum mass flow rate that can pass through the annulus
for the specified total thermodynamic conditions and swirl velocity or flow angle.
One way to identify choke is to compare mass flow rates calculated from Eq. (7-
5) for two different values of Wm(0). If the calculated mass flow and Wm(0) vary
in opposite directions, the higher value of Wm(0) is beyond the choke limit. An
iteration scheme can be used to converge on the actual choking value of Wm(0). It
has been found to be simpler, and equally effective, to monitor the average merid-
ional Mach number, Mm, of Eq. (7-27). For uniform, swirling flow in an annulus,
it can be shown that the condition for choke is Mm = 1. For the more general case
considered here, a reasonable criterion for choke is
ys
1
ys ∫ Mmdy ≥ 1 (7-39)
0
Indeed, this parameter is easily employed to limit the value of Wm(0) used
while seeking to converge on the mass flow. The existence of a choke condition at

a quasi-normal is not necessarily an indication of choke for the compressor. It

can be caused by incorrect stream sheet geometry in the early iterations. Hence,
the overall numerical scheme must include provision to continue the analysis
until it is clear that the choke condition is real. But during the process of calcu-
lating the flow profiles on a specific quasi-normal, it does represent the maxi-
mum mass flow rate that can pass through the annulus.
The procedure and equations outlined in this section can be used to compute
the flow along a quasi-normal under the following conditions:
• The stream surface coordinates, slopes and curvatures along the quasi-
normal are specified.
• The total thermodynamic conditions along the quasi-normal are specified.
• The flow angle or the swirl velocity distribution along the quasi-normal
is specified.
What remains is description of how these conditions are to be established in the

overall numerical analysis. For generality, the procedures described are pre-
sented for the rotating frame of reference. They are equally valid for the station-
ary frame of reference if ω = 0 and C, β and H are substituted for W, β′ and I,
respectively.
7.3 LINKING QUASI-NORMALS

Next, the determination of the total thermodynamic conditions and swirl or flow
angle profiles on a quasi-normal will be considered. It will continue to be
assumed that the stream surface coordinates, slopes and curvatures are all
known. This section describes techniques needed to link successive quasi-nor-
mals together so that the quasi-normal flow analysis of the previous section can
be conducted on all quasi-normals for a specified stream surface pattern.
To start the process, boundary conditions are needed for the first quasi-nor-
mal. The usual process is to supply specifications of the distributions of the inlet
flow angle or absolute tangential velocity and inlet total thermodynamic condi-
tions (e.g., Pt and Tt) along the first quasi-normal. Values of these parameters on
the stream surfaces of the first quasi-normal are determined by interpolation
from these profile specifications. Then the procedures of the previous section can
be used to compute the inlet flow profiles.
For all quasi-normals after the first one, the process required is to link the
quasi-normal being analyzed to the upstream quasi-normal, where all flow data
are known. Subscript 1 will be used to designate known data on the upstream
quasi-normal, and subscript 2 will be used to designate conditions on the quasi-
normal being analyzed, where both apply to the same stream surface. The sim-
plest case is successive quasi-normals in a simple annular passage with no blade
row between them. In this case, Eqs. (7-7), (7-10) and (7-11) provide the linking
relations, i.e.,
( rCθ )2 = ( rCθ )1 (7-40)

I2 = H2 − ω ( rCθ )2 = I1 = H1 − ω ( rCθ )1 (7-41)

s2 = s1 (7-42)
From Eqs. (7-40) and (7-41), it is seen that H2 = H1 is the true linking condition
implied by Eq. (7-41). Hence, all data needed for the quasi-normal flow analysis
of the previous section are available.
When a blade row lies between the two quasi-normals, the empirical models
of Chapter 6 are used to estimate the influence of the blade row. The process will
be described in the rotating frame of reference, recognizing that it is applied to
a stationary blade by simply setting ω = 0. The empirical models of Chapter 6
supply the blade row total pressure loss coefficient and discharge relative flow
angle. But the empirical models require knowledge of the discharge meridional
velocity, which is not yet known. Hence, an iterative solution procedure is
required, typically starting with the assumption that Wm2 = Wm1 on all stream
surfaces. The estimate of the discharge meridional velocity profile is improved
by successive application of the empirical models of Chapter 6 and the quasi-
normal flow analysis of the previous section until the process converges. First,
the inlet relative conditions are computed from the known upstream absolute
flow conditions.
I1 = H1 − ω ( rCθ )1 (7-43)
Wθ1 = Cθ1 − r1ω (7-44)
W1 = Wm21 + Wθ21 (7-45)
H1′ = h1 + 12 W12 (7-46)
Other relative total conditions (e.g., P′t1 and T′t1) can be computed from the equa-
tion of state and the known values of entropy and relative total enthalpy. At the
discharge station, conservation of rothalpy requires
I2 = I1 (7-47)
H2′ = I2 + 12 ( r2ω )2 (7-48)
The ideal (no loss) discharge total pressure is computed from the equation of
state, using the known discharge relative total enthalpy and the inlet entropy.
Then the actual discharge relative total pressure is computed from the total pres-
sure loss coefficient.
Pt′2 = Pt′2id − ω ( Pt1′ − P1) (7-49)
All other relative total thermodynamic conditions and the entropy at the dis-
charge are computed using the equation of state and the known relative total
pressure and relative total enthalpy. Hence, all data required for the quasi-normal

flow analysis of the previous section are known. When the blade performance
and discharge flow profile iteration process is converged, absolute discharge con-
ditions are computed as
Cθ 2 = Wθ 2 + r2ω = Wm2 tan β2′ + r2ω (7-50)

H2 = I2 + ω ( rCθ )2 (7-51)
Cm2 = Wm2 (7-52)
2 2 (7-53)
C = Cm 2 + Cθ 2
7.4 REPOSITIONING THE STREAM SURFACES
After solving the equations for conservation of mass and momentum to deter-
mine all flow field data throughout the solution domain, the new data will gener-
ally not be consistent with the stream surface geometry used in the process. The
variation of mass flow along any quasi-normal can be easily determined in func-
tional form using a modified form of Eq. (7-5).
y
˙ ( y) = 2π ∫ KW rρ Wm cos ε dy
m (7-54)
0
Note that the boundary layer blockage factor, KB, has been omitted in Eq. (7-54).
This writer prefers to treat the blockage factor as a simple area correction applied
to conservation of mass. In this approach, the hub-and-shroud stream surfaces
are always positioned on the corresponding end-wall contours. An alternate
approach is to reposition the hub-and-shroud stream surfaces from the end-wall
contours by the end-wall boundary layer displacement thicknesses described in
Chapter 8. In that case, the lower limit of integration in Eq. (7-54) will be yh
rather than zero. That added sophistication has not resulted in any observable
improvement to performance prediction accuracy, yet it can often complicate
convergence and numerical stability. In either case, the interior stream surfaces
are to be repositioned to yield the correct fraction of the mass flow calculated for
the shroud stream surface using Eq. (7-54). The actual mass flow is not used so as
to avoid any influence from numerical errors in conservation of mass or bound-
ary layer analysis. Both of these numerical calculations are governed by specified
convergence tolerances, but neither will yield an exact result. The correct loca-
tions of the interior stream surfaces for the computed flow field can be obtained
by interpolation from this function to yield the values of y that correspond to the
correct fraction of the mass flow function at ys. These computed values of y yield
the stream surface coordinates (z, r) for all stream surfaces on all quasi-normals.
z( y) = zh + ( zs − zh )( y − yh ) / ( ys − yh ) (7-55)
r( y) = rh + ( rs − rh )( y − yh ) / ( ys − yh ) (7-56)

These equations are written in general form, but are simplified in this writer’s
approach, where yh = 0. If the stream surfaces are simply repositioned to these
new positions, the streamline curvature numerical technique is known to be
numerically unstable. Normal practice is to reposition the stream surfaces to loca-
tions that are some fraction, F, of the distance between the old positions and the
new positions calculated for the current flow field data. This writer uses the
numerical damping procedure suggested by Novak (1973). For quasi-normals out-
side of the blade passages, such as the present application, Novak recommends
1 (1 − M2 ) y2
= 1 + * m 2s (7-57)
F B ( ∆m)
B* is an empirical constant, typically about 8, and ∆m is the minimum merid-

ional spacing with the adjacent quasi-normals. When applying Eq. (7-57), this
writer imposes the constraint that Mm ≤ 0.95. Once the stream surfaces have been
repositioned, new meridional coordinates are computed for all stream surfaces.
z   ∂r  2 
m= ∫ 1 +  ∂z   dz (7-58)
z1  
where the partial derivative in Eq. (7-58) is evaluated numerically from the
stream surface (z, r) coordinates and z1 is the value of z at the first quasi-normal.
Other stream surface geometry data follow directly from Eqs. (7-1), (7-2), (7-3)
and (7-9). Equation (7-58) sets m = 0 at the first quasi-normal, but that is arbi-
trary, since only relative values of m along a stream surface are significant for the
present solution procedure.
7.5 FULL NORMAL EQUILIBRIUM SOLUTION
There are several useful solution procedures that can be applied for a through-
flow analysis. The most general method is to solve the complete normal momen-
tum equation as given in Eqs. (7-12) and (7-14). From the common practice of
using radial lines as quasi-normals, this is often referred to as a full radial equi-
librium solution. In the more general form used in this chapter, full normal equi-
librium solution is a more appropriate term. Figure 7-3 shows a flow chart of a
typical full normal-equilibrium solution procedure referenced to the methods
outlined in Sections 7.2 through 7.4. The process starts by initializing the stream
surfaces throughout the solution domain. This is usually accomplished by
applying Eqs. (7-54) through (7-56), while assuming that the flow is uniform
from the hub to the shroud and ε = 0. The inlet boundary conditions are
imposed, and the flow is computed for the first quasi-normal using the iterative
procedure described in Section 7.2. The process then involves solving of the flow
field at all other quasi-normals. In this case, it is necessary to impose the linking
calculations of Section 7.3 as well as the iterative procedure of Section 7.2. In

FIGURE 7-3 Full Normal Equilibrium Solution Flow Chart
general, the linking calculations depend on the flow field being calculated, so a
second iteration loop is needed to converge on those calculations. When flow
field on all quasi-normals has been treated, the stream surfaces can be reposi-
tioned using the procedures of Section 7.4. This requires a third or outer itera-
tion loop to converge on the stream surface positions. When the flow field,
linking calculations and stream surface positions are all self-consistent within
an acceptable tolerance, the through-flow solution is complete. As shown in the
flow chart, the end-wall boundary layer calculations described in Chapter 8 are
normally carried out during this process, to include the end-wall boundary layer
blockage effect.
The procedure is quite simple in concept, but a number of complications may
be encountered. One common problem is that the flow is choked at a quasi-nor-
mal. It is not immediately obvious whether the choke condition is real, since
incorrect stream surface positions might cause a false choke indication. The flow
chart shows logic to require at least one completed stream surface reposition
operation before the choke is considered valid. It may be desirable to require
more than one completed reposition option. When the through-flow analysis is
conducted interactively on personal computers, the simplest approach is to
switch from an automatic iteration procedure to a manual one when choke
occurs, such that the user can decide whether to continue for another outer iter-
ation or terminate the solution. Another common complication is convergence
problems with the outer iteration loop on stream surface positions. The damping
procedure described in Section 7.4 is quite effective, but there is some uncer-
tainty with regard to B* in Eq. (7-57). B* = 8 is a good choice for most problems,

but exceptions do occur. It is good practice to monitor the maximum stream sur-
face position errors on successive iterations. If that error is increasing, B* can be
decreased to impose more damping on the numerical analysis. A good way to
make this adjustment is to impose the following correction:
B* → ( B* + 1) / 2 (7-59)
7.6 SIMPLIFIED FORMS OF THE THROUGH-FLOW

ANALYSIS
The full normal equilibrium through-flow analysis outlined in the previous sec-
tions is commonly referred to as the streamline curvature technique. It is well-
suited for implementation in a relatively robust and reliable numerical analysis.
But the streamline curvature technique is by no means totally free of numerical
stability and convergence problems. The process of repositioning stream surfaces
is, by far, the major source of these problems. It is also the process responsible
for most of the computer time required for a through-flow analysis. The main
purpose served by this relatively complex process is to determine φ, κm and ε. If
these terms are neglected or approximated in some fashion, the entire outer iter-
ation loop of the streamline curvature technique can be eliminated. The solution
then becomes a simple marching solution, where the analysis proceeds through
the solution domain, treating the quasi-normals in sequence. This is possible
because the flow analysis on a quasi-normal becomes totally independent of con-
ditions on downstream quasi-normals. The locations of the interior stream sur-
faces must still be established using Eqs. (7-54) through (7-56), but now the
process is numerically stable and requires no numerical damping procedures.
The process of relocating the stream surfaces can then be accomplished as a nor-
mal part of the process of solution at each quasi-normal.
In principle, this can be accomplished by solving Eqs. (7-2) and (7-9) using
upstream finite-difference approximations. However, this really provides an esti-
mate of the stream surface curvature within the blade passage rather than at the
blade passage exit where the solution is to be accomplished. It is not at all
uncommon for the stream surface curvature to be dramatically different at these
two locations, often even having opposite signs. The difficulty arises from the fact
that Eq. (7-9) is really equivalent to determining the second derivative of r as a
function of z along the stream surface. Upstream finite-difference approxima-
tions to second derivatives are often seriously in error. At best, this approach may
be capable of accounting for large curvature effects associated with passages hav-
ing large end-wall contour curvatures. But in the majority of situations encoun-
tered in axial-flow compressors, it is better to ignore stream surface curvature
entirely than to use upstream finite-difference approximations.
There certainly is merit to providing a through-flow analysis with the capa-
bility to approximate large stream surface curvatures present when passage cur-
vatures are large. The inlet portion of the flow passage illustrated in Fig. 7-1 is a
good example of a situation where stream surface curvature cannot be ignored.
Since the end-wall contours are normally completely specified in advance, the

hub-and-shroud stream surface curvatures can be accurately estimated. A good

way to extend a simplified through-flow analysis for it to be applicable to prob-
lems having significant passage curvature effects, is to assume that φ and κm
vary linearly with y from the hub to the shroud. This approach can recognize
significant passage-induced stream surface curvatures without the risk of pro-
ducing the highly erroneous curvatures that can follow from an upstream finite-
difference approach. In the context of axial-flow compressor performance
analysis, performance predictions using this simplified model and the full nor-
mal equilibrium model are consistently found to be virtually identical. And the
computer time required for solution is dramatically reduced when the simpli-
fied model is used.
The situation is quite different when a through-flow analysis is used as part of
an aerodynamic design procedure. In these cases, typical practice is to specify
one of the end-wall contours and to calculate the other from conservation of
mass. This process of sizing the annulus may also include design of the blade
rows, or it may employ a standard stage design. In either case, the slopes and cur-
vatures of one of the end-walls cannot be well approximated in a marching type
solution. In these cases, it is best to simply neglect stream surface curvature
effects entirely. Since design-mode applications are normally restricted to sta-
tions before and after blade rows, this is generally a reasonable approximation.
Usually any fine-tuning of the design that may be required is easily accomplished
with a normal performance analysis. To apply this approximation to the through-
flow analysis procedure outlined in the previous sections, it is only necessary to
set ε = κm = 0 in Eqs. (7-12) through (7-19). This type of analysis has commonly
been called a simple non-isentropic radial equilibrium solution, although in the
more general quasi-normal structure used here, it is better described as a simple
non-isentropic normal equilibrium solution.
There are also applications where the through-flow analysis must be further
simplified by ignoring the entropy gradients along the quasi-normal. This occurs
in industrial axial-flow compressors when designing a standard stage to be used
throughout the compressor. This is often done as a means of minimizing the cost
of manufacturing the compressor. Typically, the stagger angles of the blades in
this standard stage will be adjusted somewhat through the machine to fine-tune
the performance to the customer’s requirements. Since the standard stage may be
used in a variety of applications, it is not possible to compute the blade row per-
formance in any general context. Nor do entropy gradient effects have much sig-
nificance in this case, since they depend on all of the blade rows preceding the
location of a specific stage. In this case, it is best to simply ignore entropy gradi-
ents as well as the curvature effects. This type of analysis is commonly referred to
as a simple radial equilibrium solution. The term simple normal equilibrium is
perhaps more appropriate, although generally this type of analysis will employ
radial quasi-normals in any case.
Figure 7-4 illustrates a typical flow chart for any of these simplified forms of
the through-flow analysis. On comparing this flow chart with the one shown in
Fig. 7-3, it is seen that the entire outer iteration loop has been eliminated. In this
case, repositioning of stream surfaces will be accomplished while computing the
flow field at each quasi-normal. The procedure of Section 7.4 is used for this pur-
pose, but without any numerical damping. This type of analysis is extremely fast

FIGURE 7-4 Simplified Through-Flow Analysis Flow Chart
and numerically very stable. The full normal equilibrium model certainly should
be included for generality. But it is truly remarkable how seldom it is really
required for typical axial-flow compressor applications. In addition, the simpli-
fied forms allow application of the through-flow analysis to a number of aerody-
namic design functions where use of the full normal equilibrium model would be
totally impractical.
In summary, a through-flow analysis should normally be developed in a fairly
general form to include the capability of employing any of the following aerody-
namic models.
• Full normal equilibrium.

• Approximate (linear, hub to shroud) stream surface slope and curvature.
• Simple non-isentropic normal equilibrium.
• Simple (isentropic) normal equilibrium.
In this way, the same through-flow analysis can be applied to a variety of axial-
flow compressor aerodynamic design and analysis functions.
7.7 ANNULUS SIZING
The annulus sizing process mentioned in the previous section is one useful aero-
dynamic design function that is conveniently incorporated directly into the

through-flow analysis. In this mode of solution, only one of the end-wall contours
is specified. The analysis is used to determine the other end-wall contour, while
conserving mass and matching a desired distribution of Wm through the com-
pressor. Typically, the Wm distribution is specified for the mean stream surface,
although any other stream surface can be used as well. For generality, a super-
script, *, will be used to designate parameters on the stream surface for which
Wm will be specified. The special requirements for annulus sizing include
• Specify z and r for one end-wall contour and the angle, λ, of Fig. 7-2 for
all quasi-normals.
• Specify values of Wm for the selected stream surface for all quasi-normals.
• Neglect stream surface curvature effects, typically using the simple non-
isentropic normal equilibrium model.
The annulus area, A, at any quasi-normal is given by
A = π ( rs + rh ) ys (7-60)
As an initial estimate for the annulus area, use the specified meridional velocity,
the total density and conservation of mass
˙ / ( ρt*Wm* )
A=m (7-61)
where the inlet total density is used for the first quasi-normal, and its value at the
upstream quasi-normal is used for all others. From Eq. (7-60) and the specified
value of λ, it is easily shown that
rs2 = rh2 + ( A cos λ ) / π (7-62)

ys = A / [π ( rs + rh )] (7-63)
zs = zh − ys sin λ (7-64)
Equations (7-62) through (7-64) yield the coordinates of the unknown end-wall
contour from those of the known contour and the passage area. Next, the usual
through-flow analysis is conducted while using the specified meridional velocity
as a constant of integration in Eq. (7-35). Although Eq. (7-35) is written with the
hub meridional velocity (at y = 0) as the constant of integration, the value at y*
can be used with a simple substitution. From Eq. (7-35) it is easily shown that
y*
* / F( y* ) − f4 ( y)
Wm (0) = Wm ∫ F ( y)
dy (7-65)
0
The annulus sizing differs from the procedures presented for the analysis mode
only with regard to the application of Eq. (7-5). Here, it is used to compute the

mass flow rate, ṁ c, for the estimated annulus area. Then the annulus area esti-
mate is improved by
A → Am
˙ /m
˙c (7-66)
and the process is repeated until acceptable convergence on the mass flow rate is
achieved.
7.8 NUMERICAL APPROXIMATIONS
A numerical analysis based on the procedures described in this chapter is rela-

tively straightforward, except as it relates to approximations for the stream sur-
face curvature terms. When the streamline curvature technique was introduced,
common practice was to employ spline-connected cubic approximations for this
purpose (e.g, see Walsh et al., 1962). Although the spline fit seems almost ideal
for the smooth curves expected, experience eventually convinced most investiga-
tors that it is not a particularly good choice. The spline fit has a definite tendency
to destabilize the analysis, to increase the demands on the numerical damping
procedures.
The simple three-point finite-difference approximation for the partial deriva-
tives in Eqs. (7-2) and (7-9) is a much better choice for this application. These are
derived from truncated Taylor series approximations for the central point in a
series of three points, similar to the derivation of Eqs. (5-35) and (5-36). In the
present case, the points are not likely to have equally spacing, as was the case in
Chapter 5. A more general approximation can be shown to be
 ∂f  1  (m3 − m2 )( f2 − f1) (m2 − m1)( f3 − f2 ) 

 ∂m  = m − m  m2 − m1
+
m3 − m2
 (7-67)
 2 3 1 
where the subscripts 1, 2 and 3 designate any three successive points along the
curve. Similarly, three-point difference approximations can also be derived for
end points on the curve to yield
 ∂f  1  (m3 − m1)( f2 − f1) (m2 − m1)( f3 − f1) 

 ∂m  = m − m 
m2 − m1
−
m3 − m1
 (7-68)
 1 3 2  
 ∂f  1  (m3 − m1)( f3 − f2 ) (m3 − m2 )( f3 − f1) 
 ∂m  = m − m  m3 − m2
−
m3 − m1
 (7-69)
 3 2 1 
For this application, which can often induce numerical stability problems, a sim-
ple two-point difference approximation is a better choice for end points, i.e.,
 ∂f   ∂f  f2 − f1
 ∂m  =  ∂m  = m − m (7-70)
 1  2 2 1

Equation (7-70) will also be required for all quasi-normals when the simplified
through-flow models of Section 7.6, since data at m = m3 will not be known when
solving at m = m2, in those simple marching type analyses.
Solution of Eq. (7-35) requires numerical approximations for partial deriva-
tives with respect to y and for integrals with respect to y. Equations (7-67)
through (7-69) have been found to be good choices for the derivative approxima-
tions at interior and end points, where y is substituted for m. For numerical inte-
gration, an approximation for the second derivative at interior values of y is also
required. A suitable three-point approximation can be derived from truncated
Taylor series to yield
 ∂2 f  2  f3 − f2 f −f 
 2 =  − 2 1 (7-71)
 ∂y 2 y3 − y y
1 3 − y2 y 2 − y1 
A Taylor series approximation to the integral between points 1 and 2, where point
2 is an interior point, is easily shown to be
y2   ∂f  ( y2 − y1)  ∂2 f  ( y2 − y1)2 
∫ f ( y)dy =  f ( y2 ) −  
  ∂y 2 2
+ 2
 ∂y  2 6
 ( y2 − y1) (7-72)
y1  
Similarly, the integral between points 2 and 3, where point 3 is the last point, is
given by
y3   ∂f  ( y3 − y2 )  ∂2 f  ( y3 − y2 )2 
∫ f ( y)dy =  f ( y2 ) +  
  ∂y 2 2
+ 2
 ∂y  2 6
 ( y3 − y2 )

(7-73)
y2  
These two equations can be used to compute all integrals with respect to y from 0
to ys for any number of stream surfaces by simple summation of results between
successive points. Equations (7-67) and (7-71) provide the approximations for
the derivative terms in Eqs. (7-72) and (7-73).
EXERCISES
7.1 Geometry data for an axial-flow compressor blade is normally spec-

ified as a function of radius. Data sufficient to define the blade
might include the chord, c, the thickness-to-chord ratio, tb / c, and
any two of the following: camber angle, θ, stagger angle, γ, inlet
angle, κ1, or exit angle, κ2. The number of blades, Z, and the location
of maximum camber, a/c, will be constant for each blade row. A
through-flow analysis is to be applied in an aerodynamic perform-
ance analysis such that the blade geometry on each stream surface
is required so that the empirical models of Chapter 6 can be used to

estimate the blade row loss and fluid turning. Consider a stream sur-
face passing through a blade row with inlet coordinates (z1, r1) and
exit coordinates (z2, r2). Develop a method to estimate the blade
geometry on the stream surface from the known blade geometry as a
function of radius.
7.2 Equation (7-29) is recommended for evaluating the meridional gra-
dient of the meridional velocity component. Alternatively, this gradi-
ent might be evaluated using the finite-difference approximations of
Section 7.8. Discuss the relative merits of these two alternative
approximations.
7.3 Equation (7-39) has been recommended as an approximate criterion
for choked flow in the annular passage, outside of the blade rows. A
rigorous calculation of the choked flow limit could be accomplished
by determining the constant of integration, Wm(0), for Eq. (7-35) that
yields the maximum mass flow rate. Give two reasons why the
approximate criterion should be adequate for a through-flow analysis
in an axial-flow compressor. Under what conditions might the more
rigorous method be preferred?
7.4 Discuss the advantages and disadvantages of using Eqs. (7-31)
through (7-33) as a means of avoiding a singularity in Eq. (7-34). Con-
sider the accuracy of the approximation used for both interior and
end-wall stream surfaces. For a fully converged inviscid through-flow
solution, where can such a singularity occur? Is the approximation
more acceptable if an end-wall boundary layer analysis is conducted
as part of the overall solution?

Chapter 8
END-WALL BOUNDARY
LAYER ANALYSIS
The flow in the end-wall regions has a substantial influence on the aerodynamic
performance of axial-flow compressors. As noted in Chapter 6, a significant por-
tion of the losses in axial-flow compressors is directly associated with the end-
wall flow. The through-flow analysis of Chapter 7 requires some external
specification of the viscous end-wall blockage factor for solution of Eq. (7-5). In
addition, individual stage loading limits and the compressor surge flow limit are
often associated with end-wall stall. Unfortunately, there is no available theoreti-
cal aerodynamic model capable of predicting the detailed behavior of these
highly complex end-wall flows. Indeed, even modern computational fluid dynam-
ics (CFD) viscous flow solvers are found to be incapable of resolving many of the
important flow patterns that are observed in the end-wall regions of axial-flow
compressors. When fundamental analysis techniques are not sufficient to treat a
problem of interest, engineers commonly resort to a combination of theoretical
and empirical models. That approach is always used when formulating an aero-
dynamic performance analysis for axial-flow compressors. The role of end-wall
boundary layer models used within specific performance analyses varies consid-
erably. It is always necessary to address the problem of end-wall blockage effects
to effectively apply an inviscid through-flow analysis to the problem. Attempts to
model end-wall work and loss effects from boundary layer analysis results will be
briefly discussed in this chapter. But, in this writer’s experience, none of the avail-
able end-wall boundary layer models is sufficiently accurate and reliable for that
purpose. Chapter 6 has already described empirical models used to extend cas-
cade loss models to account for clearance and end-wall loss effects.
This chapter presents an end-wall boundary layer analysis used to account for
end-wall boundary layer blockage effects. The blade row performance models of
Chapter 6, the through-flow analysis of Chapter 7 and this end-wall boundary
layer analysis are the basic components of an aerodynamic performance analysis.
Chapter 9 describes the performance analysis and qualifies it by comparing per-
formance predictions with experimental data. In keeping with the stated objective
of this book, Chapters 6 through 9 provide a detailed description of the aerody-
namic performance analysis. But it should be emphasized that qualification of the
performance analysis evaluates its basic components in combination. That type of
qualification does not separate the parts from the whole. In Chapters 6 and 7, it
was possible to discuss the assumptions, approximations and limitations of the

models. That is not the case in the present chapter. The merits of the present end-
wall boundary layer analysis cannot be established beyond demonstrating its
effectiveness in supporting the methods of Chapters 6 and 7 to predict the overall
performance of axial-flow compressors.
NOMENCLATURE
B = fractional area blockage
f = blade force
g = blade row staggered spacing
H = boundary layer streamwise shape factor
H1 = boundary layer meridional shape factor
H2 = boundary layer tangential shape factor
KB = blockage factor
m = meridional coordinate and tangential velocity profile exponent
n = meridional velocity profile exponent
P = pressure
q = inlet dynamic head
Reθ = momentum thickness Reynolds number
r = radius
s = blade pitch
Uleak = leakage flow tangential velocity
V = velocity relative to the wall
W = velocity relative to the blade row
y = distance normal to the wall
β = flow angle
γ = blade stagger angle
∆ = leakage flow correction parameter
δc = blade clearance
δ* = boundary layer streamwise displacement thickness
δ 1* = boundary layer meridional displacement thickness
δ*2 = boundary layer tangential displacement thickness
θ = tangential coordinate and streamwise momentum thickness
θ11 = meridional momentum thickness
θ12 = tangential momentum flux thickness
θ22 = tangential momentum thickness
µ = fluid viscosity
ν = blade force defect thickness
ρ = fluid density
τ = shear stress
φ = contour angle with the axial direction
ψ = pressure coefficient
ω = rotation speed

End-Wall Boundary Layer Analysis • 177
Subscripts
e = boundary layer edge condition

in = blade inlet condition
leak = seal leakage parameter
out = blade discharge condition
s = streamwise component
w = wall condition
Superscripts
′ = parameter in rotating coordinates

 = average of blade inlet and discharge values
+ = upstream leakage flow correction
- = downstream leakage flow correction
8.1 HISTORICAL DEVELOPMENT OF END-WALL

BOUNDARY LAYER THEORY
Early attempts to account for the effects of end-wall boundary layer blockage
relied on assigned blockage factors. Typical practice was to assign the fractional
area blockage to vary linearly through the compressor (e.g., Sandercock et al.,
1954). Alternatively, the boundary layer displacement thickness or the blockage
was assumed to vary linearly through the front stages and remain constant in the
rear stages (e.g., Voit, 1953; Jansen and Moffet, 1967). These blockage allowances
were rather arbitrary, based largely on the investigator’s experience from previ-
ous compressors. Indeed, they really were rather arbitrary corrections that
appeared to explain differences between calculated and measured compressor
performance. Considering the relatively crude calculation methods used, these
corrections probably compensated for many weaknesses and omissions in the
analyses in addition to the end-wall boundary layer blockage effects.
Early attempts to compute end-wall boundary layer blockage in multistage
compressors using boundary layer analysis techniques were published simultane-
ously by Stratford (1967) and Jansen (1967). Both methods seek to predict the
average boundary layer growth assuming that the blade forces are conserved inside
the boundary layer and that the stream surface slope can be neglected. This
reduces the problem to consideration of a simplified axial momentum-integral
equation similar to Eq. (3-52) but with sinφ = νm = 0. These assumptions uncouple
the axial momentum-integral equation from the tangential momentum-integral
equation, so that Eq. (3-53) is unnecessary. Stratford employed flat-plate approxi-
mations for the boundary layer shape factor, H, and the wall shear stress. Jansen
used an approximate integral solution (Schlicting, 1968 and 1979). Subsequent
investigations showed that these early analyses are overly simplified. Nevertheless,
they introduced the important concept of analyzing the gap-averaged or pitch-
averaged boundary layer flow using integral boundary layer analysis techniques.

Smith (1970) reported detailed end-wall boundary layer data measured in a

multistage axial-flow compressor. His measurements clearly demonstrate that
blade forces are not constant across the boundary layer. He also provides experi-
mental evidence to support the “repeating stage model.” Basically this model
assumes that the end-wall boundary layers achieve an equilibrium condition
after passing through several stages, such that the blade row exit and inlet flow
profiles are essentially identical. Once the equilibrium condition is reached,
boundary layer growth is viewed as primarily a function of aerodynamic loading.
Smith suggests that the repeating stage concept has been recognized for many
years (e.g., Howell, 1947), but has not been used effectively. Smith provides
experimental boundary layer data that may provide at least a preliminary basis
for using this model. In a subsequent publication (Koch and Smith, 1976), the
same basic boundary layer data was reworked as a sum of the hub-and-shroud
boundary layers, as shown in Figs. 8-1 and 8-2. The combined data show less
data scatter as compared to Smith (1970). The pressure coefficient, ψ, is a sum of
the rotor and stator pressure coefficients for the stage, specifically defined by
∆protor + ∆pstator
ψ = (8-1)
qrotor + qstator
where q is the inlet dynamic head. The meridional displacement thickness, δ 1*,
and the tangential force defect thickness, νθ, are defined in Eqs. (3-54) and (3-60),
FIGURE 8-1 Displacement Thickness Data

FIGURE 8-2 Tangential Force Defect Data
and δc is the blade tip clearance. The staggered spacing, g, is a function of the
blade pitch, s, and stagger angle, γ, where average values are used,
g = s cos γ (8-2)
Hunter and Cumpsty (1982) report similar results obtained with an isolated
rotor. Koch and Smith use the empirical curves shown in Fig. 8-1 to estimate the
sum of the displacement thicknesses or blockage. This requires that ψmax values
be supplied by some unspecified stall criterion. The displacement thicknesses
plus the tangential force defect are then used to estimate the efficiency reduction
due to end-wall losses. There is no doubt that Smith made a substantial contri-
bution to our knowledge of end-wall boundary layers. But there is little reason to
believe that the empirical models outlined can be used for general application. To
recognize that the empirical curves shown are far from correlations of experi-
mental results, one need only note that the tip clearance-to-staggered spacing
ratios for all experimental data in Fig. 8-1 lie between 0.028 and 0.062. Careful
study of the original reference shows that the experimental data trends contra-
dict the empirical curves about as often as they are in agreement. And the exces-
sive data scatter in Fig. 8-2 provides no real basis for any empirical correlation of
the tangential force defect. Although far from a complete end-wall flow model,
these references provide important insight into the end-wall boundary layer
problem. It is clear that any end-wall boundary layer theory must address the

problem of blade force defects. Substantial tangential force defects clearly do

exist. It must be presumed that the same is likely to be true for meridional force
defects. But the repeating stage model poses special constraints. For an equilib-
rium blockage to be established, it is clear that meridional blade force defects
must become small, or possibly even negative. Unfortunately, none of the above
experimental investigations provided any direct insight relative to the meridional
force defect behavior. It is apparent that our knowledge of the mechanisms
behind the blade force defects is far from adequate to support an accurate end-
wall loss calculation method.
There have been numerous attempts to formulate end-wall boundary layer
analyses since the original attempts by Stratford and Jansen. One of most signif-
icant was the work of Mellor and Wood (1971), who provided a rather rigorous
development of the basic gap-averaged governing equations. Their development
includes effects previously neglected, such as blade force defects and “jump” con-
ditions, to treat boundary layers moving between rotating and stationary frames
of reference. They conducted numerous analytical studies to show how these
phenomena can be used to explain observed flow behavior not modeled by the
simpler methods. Like Smith (1970), Mellor and Wood considered the need to
explain the existence of a repeating stage condition to be important. Balsa and
Mellor (1975) continued this development by attempting to formulate a usable
end-wall boundary layer analysis. In a serious attempt to address the blade force
defect problem, the difference in the tangential and meridional momentum
defect thicknesses was approximated with a secondary flow model. It was also
assumed that the overall blade force remains approximately normal to the mean
mainstream velocity vector. These conditions are sufficient to solve the governing
equations and do offer a mechanism that can produce the repeating stage condi-
tion. This development includes techniques to convert boundary layer data into
an end-wall loss prediction. Although undoubtedly the most complete theoretical
development up to that time, several important features were largely ignored.
The critically important boundary layer shape factor relating the displacement
and momentum thicknesses was somewhat arbitrarily assigned to match predic-
tions to experiment in sample overall compressor performance predictions. The
boundary layer skin friction coefficient was treated in a similar fashion. The cou-
pling between the tangential and meridional momentum-integral equations was
ignored, even though it can be very significant when φ is not zero in Eqs. (3-52)
and (3-53).
An important series of publications on this subject by Professor Hirsch and
associates also should be mentioned. These include Hirsch (1974 and 1976) and
De Ruyck et al. (1979 and 1980). These references provide a very detailed discus-
sion of the important force defect terms and investigate various alternative blade
force defect models derived from secondary flow theory. They also attempt to
model some of the features that were handled rather arbitrarily by Balsa and
Mellor. Although the results fall short of a complete end-wall boundary layer the-
ory, these references provide considerable insight into the problem and are defi-
nitely recommended.
Basically, the current state of the art of end-wall boundary layer theory is
insufficient to analyze the complete end-wall flow problem, including the asso-
ciated end-wall losses. The prediction of end-wall boundary layer blockage to

support the aerodynamic performance analysis of axial-flow compressors is

probably the best that can be expected. Nevertheless, the experimental and ana-
lytical studies discussed above have clearly identified several important features
that strongly influence the end-wall flow problem. Even the more modest goal of
predicting end-wall blockage effects requires consideration of the blade force
defect terms and should include a mechanism that can produce the repeating
stage condition.
8.2 THE END-WALL BOUNDARY LAYER EQUATIONS
End-wall boundary layer analysis in axial-flow compressors considers equa-

tions for the flow averaged over the pitch between adjacent blades. The govern-
ing equations are basically identical to the axisymmetric three-dimensional
boundary layer equations presented in Chapter 3. Most investigators have
ignored the streamline slope terms in the governing equations, basically assum-
ing that φ = 0. Also, it has been common practice to develop the governing equa-
tions in either stationary coordinates or in a coordinate system fixed to the
blade rows. In the real problem, the wall and blades rotate at different speeds at
the blade tips, unless an attached shroud is present. This effect is usually
ignored, except perhaps as a correction to the wall shear stress. Aungier (2000)
develops the end-wall boundary layer equations for use in centrifugal compres-
sors, where these effects cannot be neglected. This more general form of the
governing equations will be used here also. The absolute velocity will continue
to be designated as C and the velocity relative to the blade by W. But the veloc-
ity relative to the wall is also needed, and will be designated by V. If the rotation
speed of the blade is ω and the rotation speed of the wall is ωw, the three veloc-
ities are related by
Cm = Wm = Vm (8-3)
Cθ = Wθ + rω = Vθ + rω w (8-4)
The no-slip condition requires that the fluid velocity at the wall must vanish in
the coordinate system fixed to the wall. So the governing equations should be
written in that coordinate system. The axisymmetric three-dimensional
boundary layer equations presented in Chapter 3 are valid for any rotating
coordinate system. So they are easily transformed to the coordinate system
fixed to the wall by substituting V for W and ωw for ω. Consequently, the gov-
erning equations for the end-wall boundary layer flow problem are easily
shown to be
∂
[rρ eVme (δ − δ1* )] = rρ eVe E (8-5)
∂m
∂ 2 ∂V
[rρ eVmeθ11] + δ1* rρ eVme me − ρ eVθe sin φ [Vθe (δ 2* + θ 22 ) + 2ω w rδ 2* ]
∂m ∂m (8-6)
= r[τ mw + fmeν m ]

∂ 2  ∂V 
[r ρ eVmeVθeθ12 ] + rδ1* ρ eVme  r θe + sin φ (Vθe + 2rω w ) = r 2 [τ θw + fθeνθ ] (8-7)
∂m  ∂m 
The blade forces at the boundary layer edge are
∂Vme ∂Pe sin φ

fme = ρ eVme + − ρ e (Vθe + ω w r )2 (8-8)
∂m ∂m r
ρ eVme  ∂Vθe  ρ V ∂rCθe
fθe = r + sin φ (Vθe + 2rω w ) = e me (8-9)
r  ∂m  r ∂m
Equations (8-7) and (8-9) can be combined to yield
∂ 2
[r ρ eVmeVθeθ12 ] + r 2δ1* fθe = r 2 [τ θw + fθeνθ ] (8-10)
∂m
If y is the distance normal to the wall and δ is the boundary layer thickness, the
various defect thicknesses in the boundary layer equations are defined as
δ
ρ eVmeδ1* = ∫ ( ρ eVme − ρ Vm )dy (8-11)
0
δ
2
ρ eVmeθ11 = ∫ ρ Vm (Vme − Vm )dy (8-12)
0
δ
ρ eVmeVθeθ12 = ∫ ρ Vm (Vθe − Vθ )dy (8-13)
0
δ
ρ eVθeδ 2* = ∫ ( ρ eVθe − ρ Vθ )dy (8-14)
0
δ
ρ eVθ2eθ 22 = ∫ ρ Vθ (Vθe − Vθ )dy (8-15)
0
δ
ν m fme = ∫ ( fme − fm )dy (8-16)
0
δ
νθ fθe = ∫ ( fθe − fθ )dy (8-17)
0
Solution of this set of equations requires empirical correlations for the entrain-
ment function, E, the wall shear stresses, τwm and τwθ, and the force defect thick-
nesses, νm and νθ. It also requires empirical relationships between the mass and
momentum defect thicknesses, typically derived from some assumed form of the
boundary layer velocity profiles.

8.3 THE BOUNDARY LAYER VELOCITY PROFILE

ASSUMPTIONS
Aungier [1988(b)] obtained good agreement with experimental data for axisym-
metric swirling flow in vaneless annular passages by solving the axisymmetric
three-dimensional boundary layer equations using simple power-law velocity
profile assumptions.
n
Vm  y 
=  (8-18)
Vme  δ 
m
Vθ  y 
=  (8-19)
Vθe  δ 
Substitution of these profile assumptions into the definitions of the various mass
and momentum defect thicknesses, assuming density can be regarded as essen-
tially constant, yields the following relationships.
n = θ11 / (δ − δ1* − 2θ11) (8-20)
m = θ12 ( n + 1)2 / [δ − θ12 ( n + 1)] (8-21)

H1 = δ1* / θ11 = 2n + 1 (8-22)
δ − δ1* = 2H1θ11 / ( H1 − 1) (8-23)

δ1* / δ = n / ( n + 1) (8-24)
H2 = δ 2* / θ 22 = 2m + 1 (8-25)
δ 2* / δ = m / (m + 1) (8-26)
These profile assumptions and empirical relations require some modification

when applied to flows through blade rows. This can be illustrated by considering
the flow through an axial-flow compressor stage on the hub contour. For the
rotor, the blades and the hub end-wall rotate at the same speed, so V = W in the
boundary layer equations. If the downstream stator is shrouded, both the hub
end-wall and the blades are stationary, so V = C. From Eqs. (8-4) and (8-13), it is
easily shown that the jump condition between the rotor exit and the stator inlet is
Cθ eθ12 = Wθ eθ12
′ (8-27)
The prime designates the tangential momentum defect thickness viewed in the
rotating coordinate system. Since Wθ and Cθ normally have opposite signs, it can
be seen that the tangential momentum defect thickness will change sign at any
meridional station where the wall rotation speed changes from rotating to sta-
tionary, or inversely. Hence, the boundary layer velocity profile assumption must
include cases where the θ12, θ22 and δ 2* are negative and H2 < 1. In this simple

illustration, a deficit in tangential momentum in the rotating coordinate system

corresponds to excess tangential momentum in the stationary coordinate system.
The boundary layer equations can accommodate this jump condition, but the
power-law profile assumption of Eq. (8-19) cannot. Aungier (2000) extended the
power-law profile model to accommodate this situation. When m < 0.05, Eq. (8-
19) is replaced by
0.05 2 0.1
Vθ  y   y  y
=  + 0.1705(1 − 20m)1 −    (8-28)
Vθe  δ   δ  δ 
Figure 8-3 shows typical velocity profiles from Eqs. (8-19) and (8-28). Both equa-
tions yield identical results at m = 0.05. Equation (8-28) simply extends the
power-law profiles in a plausible fashion to accommodate cases where the inte-
grated tangential momentum in the boundary layer exceeds the boundary layer
edge value, i.e., where m becomes negative. From Eqs. (8-18) and (8-28) and the
various boundary layer defect thickness definitions,
δ 2* 20m
= (8-29)
δ 21
θ 22
= 0.95m − 1.684m2 (8-30)
δ
θ 0.05  ( n + 1.1)( n + 2.1)( n + 3.1)
m = 0.05 +  12 − (8-31)
 δ ( n + 1)( n + 1.05)  6.82
δ 2* 20 (8-32)
H2 = =
θ 22 21(0.95 − 1.684m)
Figure 8-4 presents the functional relationship between m and H2 from Eqs.
(8-25) and (8-32).
If the entrainment equation is solved for (δ – δ 1*) and the momentum integral
equations are solved for θ11 and θ12, n and m can be computed from Eqs. (8-20)
and (8-21) or (8-31). Then the boundary layer profile assumptions provide all
other boundary layer data. This writer’s analysis limits H1 and H2 to a maximum
of 2.4, which is used as the boundary layer profile separation limit. Empirical
models to compute the entrainment function, wall shear stresses and blade force
defect thicknesses are required to complete the formulation of the analysis.
8.4 EMPIRICAL MODELS FOR ENTRAINMENT AND WALL

SHEAR STRESS
The wall shear stress terms can be approximated using a suitable skin friction
coefficient model. Assume that the shear stress is directed along the boundary
layer edge streamline. Then the shear stress components are related to the skin
friction coefficient, cf.

FIGURE 8-3 Extended Boundary Velocity Profiles
FIGURE 8-4 The Boundary Layer Shape Factor

τw
cf = (8-33)
1
2
ρ eVe2
τ mw = 1c ρ VV
2 f e e me (8-34)
τ θw = 1c ρ VV
2 f e e θe
(8-35)
The well-known skin friction coefficient model by Ludwieg and Tillmann (1950)
is one of the most accurate methods available for turbulent boundary layers. It
can be expressed as
cf = 0.246 exp(−1.561H )( ρ eVeθ / µ )−0.268 (8-36)
where θ and H are the momentum thickness and shape factor in the free stream
direction, and µ is the fluid viscosity. Designate the free stream component of the
fluid velocity within the boundary layer by Vs.
Vs = Vm cos β e + Vθ sin β e (8-37)
Noting that Vse = Ve, the defect thicknesses in the free stream direction are
given by
δ
ρ eVe2θ = ∫ ρVs (Ve − Vs )dy (8-38)
0
δ
ρ eVeδ * = ∫ ( ρ eVe − ρVs )dy (8-39)
0
and H = δ* / θ. Substituting Eq. (8-37) into Eqs. (8-38) and (8-39) yields expres-
sions for these defect thicknesses in terms of the axisymmetric three-dimensional
boundary layer defect thicknesses.
δ * = δ1* cos2 β e + δ 2* sin 2 β e (8-40)

θ = (θ11 + δ1* ) cos4 β e + (θ 22 + δ 2* ) sin 4 β e + 2(θ12 + δ1* ) sin 2 β e cos2 β e −δ *
(8-41)
Some care is required to avoid values of these defect thicknesses that will invali-
date Eq. (8-36). It is recommended that all defect thicknesses on the right-hand
side of Eqs. (8-40) and (8-41) be limited to be no less than zero. In addition, when
solving Eq. (8-36), it is recommended that H ≤ 2.4 be required, where H = 2.4 is
regarded as a boundary layer separation limit. The momentum thickness
Reynolds number in Eq. (8-36) should be limited to a value to ensure that transi-
tion to turbulent flow has occurred. A reasonable transition limit for this purpose
is to require ρeVeθ / µ ≥ 250.

When solving the boundary layer equations in blade free passages, this writer
has observed that the entrainment function is best approximated by
E = 0.025( H1 − 1) (8-42)
Since gradients in the tangential direction all vanish for these cases, it is not too
surprising that the shape of the meridional flow profile controls the fluid entrain-
ment at the boundary layer edge. This approach was adopted by Davis (1976) at
this writer’s suggestion and has also been shown to be effective by Aungier
[1988(b)] and Schumann (1985). When solving the boundary layer equations
within a blade passage, where the flow is guided by the blades, the free stream
shape factor, H, is more relevant, much like the case in classical two-dimensional
boundary layer theory, i.e.,
E = 0.025( H − 1) (8-43)
For the present application to axial-flow compressor performance analysis, the

boundary layer equations are normally solved across blade rows. Equation (8-43)
is also used in this case, even though the solution is carried out using data out-
side the blade passage, where the flow is considered to be axisymmetric. In this
case, the boundary layer development occurs primarily within the blade passage
where the flow is guided by the blades. Hence the entrainment process should be
governed in the same way as when solving within a blade passage.
8.5 THE BLADE FORCE DEFECT THICKNESSES

The important blade force defect thicknesses need to be specified. The best avail-
able experimental data is the tangential blade force defect data by Koch and
Smith (1976) shown in Fig. 8-2, along with an empirical curve used by them.
Koch and Smith used the empirical curve to estimate end-wall losses. They pro-
posed a separate correlation for end-wall blockage as illustrated in Fig. 8-1. It is
evident that the empirical curve in Fig. 8-2 does not correlate to the experimental
data with sufficient accuracy for it to be used in predicting end-wall blockage.
This writer has reworked the data of Koch and Smith in the form shown in Fig.
8-5. Staggered spacing is used to normalize νθ instead of the displacement thick-
ness. Since νθ is known to increase with blade clearance (Hunter and Cumpsty,
1982), it was somewhat arbitrarily corrected by subtracting half of the average
blade clearance. Data scatter remains significant, but the result does appear to
suggest trends that might be used in an empirical correlation, such as the empir-
ical curve shown in Fig. 8-5. In the rare cases where the boundary layer thickness
is less than the blade clearance, this model yields very questionable results. It is
unlikely that νθ can become significant until the flow about the blade profiles
comes under the influence of the distorted boundary layer flow profiles. To avoid
that weakness, the data from Fig. 8-5 was expressed in the form shown in Fig.
8-6, which seems equally satisfactory as the basis for an empirical correlation.

FIGURE 8-5 Tangential Force Defect Model
FIGURE 8-6 Adjusted Tangential Force Defect Model

For the purpose of end-wall blockage prediction, the more critical parameter
is the meridional blade force defect thickness, νm. There are diverse and contra-
dictory opinions concerning νm and no experimental data available for guidance.
Smith (1970) and Koch and Smith (1976) argue that νm must be small based on
the existence of the repeating stage condition. Balsa and Mellor (1975) argue that
the blade force should remain approximately normal to the free streamline, sim-
ilar to the blade force components at the boundary layer edge. This assumes that
νm = νθ, since both are governed by the same fundamental fluid dynamics. They
also consider the need to explain the repeating stage condition to be a basic
requirement. They assume that the blades guide the flow toward a collateral flow
condition such that θ11 and θ12 approach a common value at the blade discharge,
but with a correction for blade clearance effects. In this way, small or even nega-
tive values of νm are possible to provide the mechanism for the repeating stage
condition. De Ruyck and Hirsch (1980) propose a correlation for νm with second-
ary flow parameters, but impose a transverse force defect correction similar to
that of Balsa and Mellor. None of these models has been found to be particularly
effective for end-wall blockage prediction, although the model of Balsa and Mel-
lor provided the most promising results. Hence, this writer’s blade force defect
model is an adaptation of that method.
Following Balsa and Mellor (1975) it is assumed that the blade force remains
oriented in the same direction as the free stream blade force such that νm = νθ.
Hence the subscript can be omitted and the blade force defect simply designated
as ν. A base defect thickness is estimated as
(0.12 g + δ c / 2)(8θ11 / g)3

ν0 = (8-44)
1 + (8θ11 / g)3
This equation is similar to the empirical curve shown in Fig. 8-6. The average
meridional momentum thickness is used instead of the displacement thickness
because it is less likely to be subject to abrupt changes during the analysis. It had
been expected that the equation would require adjustment by some multiplying
factor to compensate for the change in the independent variable relative to the
correlation in Fig. 8-6. It was somewhat of a surprise to find that qualification of
the performance analysis against experimental performance indicated that no
adjustment is necessary. Indeed, for most of the axial-flow compressors analyzed
in the qualification study, Eq. (8-44) was quite sufficient to provide very good per-
formance prediction accuracy using the models presented in this Chapter and in
chapters 6 and 7. But a few cases were encountered where the blade force defects
from Eq. (8-44) appeared to overestimate the blockage. Based on the previous dis-
cussions, this is to be expected. It is clear that the blade force defect must become
small or even negative if the flow in the compressor approaches the repeating
stage condition. A correction procedure not unlike that of Balsa and Mellor was
found effective in correcting the blade force defect from Eq. (8-44) in those cases.
If the blades provide sufficient guidance to the flow to force the boundary layer to
be completely collateral at the discharge, θ11 and θ12 will approach a common
value when viewed in a coordinate system fixed to the blades. Regarding W as the
velocity relative to the blade row being considered, this requires that the boundary
layer at the blade discharge satisfy the following condition:

Vθθ12 − Wθθ11 = 0 (8-45)
As noted in Eq. (8-27), the first term in Eq. (8-45) is identical to the tangential
momentum defect in the coordinate system fixed to the blade row. If we use a
prime to designate the tangential momentum thickness in the coordinate system
fixed to the blade row, Eq. (8-45) can be written as
Wθθ12
′ − Wθθ11 = 0 (8-46)
Hence Eq. (8-45) is the condition for collateral flow. If Eq. (8-10) is integrated
across the blade row of interest, the meridional momentum thickness at the dis-
charge can be represented by
r 2 fθe∆m
Vθeθ12 = A + ν (8-47)
r 2ρ eVme
Similarly, integrating Eq. (8-6) across the blade row and multiplying by Wθe yields
rfme∆m
Wθeθ11 = B + ν Wθe 2
(8-48)
rρ eVme
For convenience, assume that the blade force is normal to the mean free stream
velocity, i.e.,
fme = − Wθe fθe / Vme (8-49)

rfθe∆m Wθe
Wθeθ11 = B − ν Wθe 2 V (8-50)
rρ eVme me
If a blade force defect increment ∆ν is imposed, Eqs. (8-47) and (8-50) yield
 r W W  rf ∆m
∆(Vθeθ12 − Wθeθ11) = ∆ν  + θe θe  θe (8-51)
 r VmeVme  rρ eVme
This can be used to calculate the increment in the blade force defect thickness
needed to impose the collateral flow condition, i.e.,
∆(Vθeθ12 − Wθeθ11) = ξ∆ν = − (Vθeθ12 − Wθeθ11) (8-52)
where
 r W W  rf ∆m
ξ =  + θe θe  θe (8-53)
 r VmeVme  rρ eVme

In practice, the solidity would have to be infinite for the blade guidance to be suf-
ficient to force collateral flow in the boundary layer. Hence the condition
imposed in the analysis is
∆ν = − (Vθeθ12 − Wθeθ11)F / ξ (8-54)
Since a correction appeared to be required only in cases where the blockage and
defect force thicknesses become relatively large, it was reasonable to expect F to
be similar in form to the correlation for ν0, which proved to be true.
1 (8θ11 / g)3
F= (8-55)
2 1 + (8θ11 / g)3
Note that F is similar in form to the function illustrated in Fig. 8-5, but
approaches one-half as the quantity (8θ11/g) becomes large. The overall blade
force defect thickness is given by
ν = ν0 + ∆ν (8-56)
Equation (8-54) is to be solved using boundary layer data obtained with ∆ν =

0. In practice, it is preferable to correct the blade force defect thickness continu-
ally as the governing equations are being solved in an iterative fashion. That can
be done as long as the correction process accounts for any non-zero value of ∆ν
used in the previous iteration. It can easily be shown that updates for ∆ν on all
iterations can be imposed as
∆ν → [∆ν − (Vθeθ12 − Wθeθ11) / ξ ]F (8-57)
8.6 SEAL LEAKAGE EFFECTS FOR SHROUDED BLADES
When the blade row is shrouded at the end-wall being analyzed, the boundary
layer data must be corrected for the shroud seal leakage. The shroud seal leak-
age mass flow calculation has been described in Chapter 6. The sign convention
used for the leakage mass flow is positive when the leakage is directed from the
blade row discharge to the inlet, as illustrated in Fig. 8-7. The shroud seal leak-
age will change the mass flow rate in the boundary layer. The leakage flow is
expected to have no meridional velocity component, so the boundary layer
meridional momentum flow rate should be unchanged. The absolute tangential
velocity of the leakage flow entering the boundary layer is assumed to be half of
the local speed of the rotating wall. If the leakage flow is out of the boundary
layer, its tangential velocity relative to the wall is assumed to be essentially zero,
due to the no-slip condition. The details of the mass and momentum balances to
account for the leakage flow will be left to the exercises. Here, only the results
will be given. Using the nomenclature defined in Fig. 8-7, the blade row inlet
boundary layer mass flow is corrected by

FIGURE 8-7 A Shrouded Blade Row
(δ − δ1* )+ = (δ − δ1* )in + ∆+ (8-58)
m˙ leak
∆+ = (8-59)
(2π rρ eVme )in
The meridional momentum balance yields
+
θ11 = θ11 + ∆+ (8-60)
in
If ωc is the compressor rotation speed, the tangential velocity of leakage flow

entering the boundary layer, relative to the blade shroud, is
Uleak = 1 rω c + Vθ − Cθ
2 (8-61)
The tangential momentum balance, for either positive or negative leakage mass
flow rate, yields
(Vθθ12 )+ = (Vθθ12 )in + ∆+Vθ − 12 ( ∆+ + | ∆+ |)Uleak (8-62)

in

These leakage corrected boundary layer data are sufficient to define the power-
law profile and correct all other boundary layer parameters using Eqs. (8-20)
through (8-26). The boundary layer analysis is conducted across the blade row,
with the wall rotation speed set equal to the blade rotation speed. Then mass and
momentum balances and Eqs. (8-20) through (8-26) are used to correct the cal-
culated discharge boundary layer data for the seal leakage to obtain the final
blade discharge boundary layer data.
˙ leak
m
∆− = (8-63)
(2π rρ eVme )out
(δ − δ1* )out = (δ − δ1* )− − ∆− (8-64)

− −
θ11 = θ11 − ∆ (8-65)
out
− − − −
(Vθθ12 )out = (Vθθ12 ) − ∆ Vθ + 12 ( ∆ − | ∆ |)Uleak (8-66)
out
8.7 BOUNDARY LAYER JUMP CONDITIONS

The boundary layer analysis has been developed in terms of the fluid velocity rel-
ative to the wall, where the wall may be stationary or rotating. It is necessary to
consider jump conditions for cases where the boundary layer moves from a sta-
tionary wall to a rotating wall, or inversely. Indeed, the boundary layers along the
hub contour in axial-flow compressors with shrouded stator blades must make
this transition at every blade row. For clarity, designate the velocity relative to the
wall rotating at rotation speed ω as W and C to be the velocity relative to the sta-
tionary wall. Boundary layer parameters relative to the rotating wall will be des-
ignated with a prime. Hence,
Wθ = Cθ − ω r (8-67)
Substituting Eq. (8-67) into the definitions of the various boundary layer defect
thicknesses yields
Cθeθ12 = Wθeθ12
′ (8-68)
Cθeδ 2* = Wθeδ 2′ * (8-69)
Cθ2eθ 22 = Wθ2eθ 22
′ + ω rWθeδ 2′ * (8-70)
Cθ2eθ 22 (1 + H2 ) = Wθ2eθ 22
′ (1 + H2′ ) + 2ω rWθeδ 2′ * (8-71)
Equations (8-68) and (8-71) contain basic terms in the momentum-integral

equations. Note that they have identical values in either rotating or stationary
coordinates, so no special logic is required to convert them at a transition
between a rotating and stationary wall. Consequently, integration of the governing

equations can proceed through the compressor without any complex logic at
transitions between rotating and stationary walls.
8.8 SOLUTION PROCEDURE
Since the end-wall boundary layer analysis presented in this chapter considers
only turbulent boundary layers, boundary layer data must be specified at the
compressor inlet. The data needed are θ11, θ12, H1 and H2. It is reasonable to start
the analysis with H1 = 1.4, which is a typical value for simple flat-plate boundary
layer flow. The classical one-seventh power-law profile is a reasonable choice for
the tangential velocity profile, i.e., m = 1 / 7 or H2 = 1.286. The boundary layer
fractional area blockage, B, is assigned at the inlet. Assume the blockage is
equally split between the hub-and-shroud contours and compute the meridional
displacement thickness for either end-wall by
2π rρ eVmeδ1* = 1
2
˙
Bm (8-72)
Equations (8-22) through (8-26) provide all other initial boundary layer data. If
the analysis is to be started immediately upstream of the first blade row, experi-
ence has shown that B = 0.02 is a good choice. If the analysis can be started well
upstream of the first blade row, a very small value can be entered for B so that the
boundary layer development ahead of the first blade row is predicted by the
analysis. Throughout the analysis, the boundary layer is always constrained to be
turbulent by imposing a limit on the Reynolds number based on θ11, i.e.,
Reθ = ρ eVeθ11 / µ ≥ 250 (8-73)
Hence, specifying a very small value of B at the inlet station forces the analysis to
start with a boundary layer essentially at the transition point from laminar to tur-
bulent flow.
The analysis on each end-wall can be accomplished with a simple marching
technique, starting at the second meridional station, and completing the solution
at each station before proceeding to the next. The basic procedure is
1. Impose the inlet shroud seal leakage corrections of Section 8.6, if

applicable. Estimate initial values of the momentum thicknesses from
the upstream station using simplified forms of Eqs. (8-6) and (8-7).
∂ 2
[rρ eVmeθ11] = 0 (8-74)
∂m
∂ 2 (8-75)
[r ρ eVmeVθeθ12 ] = 0
∂m
Assume H1 and H2 are equal to the upstream values and initialize all
other parameters using Eqs. (8-22) through (8-26).

2. Compute the entrainment function and the wall shear stresses as

described in Section 8.4. If applicable, compute the blade force defect
using the methods described in Section 8.5.
3. Integrate Eq. (8-5) for (δ – δ 1*), Eq. (8-6) for θ11, and Eq. (8-7) for θ12.
4. Compute n and δ from Eqs. (8-20) and (8-24). As a precaution, require
θ12 ≤ 0.99δ / (n + 1). Compute m from Eq. (8-21).
5. Compute the other boundary layer parameters from Eqs. (8-22)
through (8-26).
6. Check for convergence on θ11, H1 and ν. If not converged, return to
Step 2 and repeat.
7. Impose the discharge shroud seal leakage corrections of Section 8.6, if
applicable.
8. Return to Step 1 and treat the next meridional station until all stations
have been analyzed.
When the boundary layer analysis is completed at all meridional stations on both
end-walls, the end-wall boundary layer blockage and the blockage factor can be
computed at all meridional stations for use in the meridional through-flow
analysis of Chapter 7. If ṁ is the compressor mass flow rate, they are given by
Shroud
B= ∑ 2π rρ eVmeδ1* / m
˙ (8-76)
Hub
KB = 1− B (8-77)
Since the boundary layer and through-flow analyses are conducted sequentially
in an iterative fashion, some limits and numerical damping are normally
required. The boundary layer edge parameters may be quite unrealistic in the
early iterations, resulting in rapid changes in the predicted blockage values.
8.9 TYPICAL RESULTS
Figures 8-8 through 8-10 illustrate end-wall blockage predictions from this end-
wall boundary layer analysis for three different styles of axial-flow compressors.
The NACA 8-stage compressor (Voit, 1953) is designed with two inlet transonic
stages. The NACA 10-stage compressor (Johnsen, 1952) is a conservative design
using all subsonic stages. The NACA 5-Stage compressor (Kovach and Sander-
cock, 1961) is designed with all transonic stages. The analysis of the NACA 10-
stage compressor had to be started at the exit of an undefined inlet guide vane
using measured vane discharge flow angles. The inlet blockage is set to 2% of the
annulus area as recommended above. The analyses of the other two compres-
sors started well up in the inlet passage with minimum (transition point) block-
age levels to let the boundary layer develop naturally based on the analysis. It is
seen that blockage levels and blockage distributions predicted for these three
compressors are quite different, and show considerable variation of blockage
level with overall pressure ratio, PR. It will be seen in Chapter 9 that the overall

FIGURE 8-8 NACA 8-Stage Compressor Blockage

performance predictions obtained using these end-wall boundary layer blockage

predictions are quite accurate for all of these compressors. Hence, the present
end-wall boundary layer analysis is quite effective in its intended role of sup-
porting the aerodynamic performance analysis.
EXERCISES
8.1 Consider the inlet of the shrouded stator blade shown in Fig. 8-7.
From conservation of mass for the upstream boundary layer and the
leakage mass flow, derive Eq. (8-58) for the boundary layer down-
stream of the injected leakage flow. The velocity, V, is relative to the
blade shroud,
8.2 The leakage mass flow in Exercise 8.1 enters with Vm = 0. From con-
servation of meridional momentum for the upstream boundary layer
and the leakage mass flow, derive Eq. (8-60) for the boundary layer
downstream of the injected leakage flow.
8.3 The leakage mass flow in Exercise 8.1 enters with tangential velocity
Uleak, relative to the blade shroud. From conservation of tangential
momentum for the upstream boundary layer and the leakage mass
flow, derive Eq. (8-62) for the boundary layer downstream of the
injected leakage flow.

8.4 Repeat Exercises 8.1 through 8.3 for the discharge of the stator blade,
noting that the tangential velocity of the leakage flow leaving the
boundary layer, relative to the blade shroud, is zero.
8.5 (δ – δ 1*), θ11 and θ12 are the boundary layer parameters obtained from
the basic conservation equations. Develop expressions for these
parameters as functions of n, m and δ, using Eqs. (8-11), (8-12), (8-13),
(8-18) and (8-19). Use the results to derive Eqs. (8-20) and (8-21).

Chapter 9
AERODYNAMIC PERFORMANCE
ANALYSIS
An aerodynamic performance analysis is essential for nearly all aspects of axial-

flow compressor aerodynamic design and application. There are basically two
types of performance analysis techniques in common use: One-dimensional or
mean-line methods analyze the performance along a mean stream surface. If
applied to well-designed stages, the mean-line performance may be considered to
be representative of the overall performance, at least near the compressor’s design
operating conditions. This approach is more questionable for off-design perform-
ance prediction. In those cases, blade incidence angle matching, blade loading lev-
els, etc., may vary dramatically at different locations along the blade span, such
that the mean-line performance is no longer representative of the overall perform-
ance. This is particularly true with respect to blade loading limits, blade stall and
end-wall stall, which establish the compressor’s surge limit. In off-design opera-
tion, the extremes in incidence angle and flow diffusion will almost always occur
on the hub or the shroud contours. Some one-dimensional methods include
approximate calculations at the hub-and-shroud contours to provide additional
guidance to the user and to better evaluate tip clearance losses, shroud leakage
effects and end-wall boundary layer blockage. The more general approach is to
conduct the performance analysis for a series of stream surfaces from hub to
shroud. These methods are referred to by various names, such as streamline
methods, through-flow methods, streamline-curvature methods or three-dimen-
sional methods. Hub-to-shroud performance analysis is a more accurate term,
which is used in this book. It has been common practice to conduct hub-to-
shroud performance analysis using the full normal equilibrium equation as
described in Chapter 7. The longer computer times required and reduced reliabil-
ity due to numerical instability for these techniques is probably the main reason
that one-dimensional methods have continued to be used. But if the through-flow
analysis offers the approximate normal equilibrium models described in section
7.6, the advantages of a hub-to-shroud performance analysis can be obtained with
computation speed and reliability comparable to those of a mean-line method.
Consequently, there is really no longer a need for one-dimensional performance
analysis methods for axial-flow compressors.
The component parts of a hub-to-shroud performance analysis have been pre-
sented in Chapters 6 to 8. This chapter describes methods to integrate those com-
ponent analyses into a hub-to-shroud performance analysis and suggests some

useful features that can make it more effective. It also compares performance
predictions from this performance analysis against experimental data for axial-
flow compressors to demonstrate the merits of the procedures presented in
Chapters 6 to 8.
NOMENCLATURE
AR = diffuser area ratio

b = diffuser passage width
c = blade chord length
Deq = equivalent diffusion factor
g = staggered spacing
i = stream surface number
L = diffuser or camberline length
N = number of stream surfaces
P = pressure
r = radius
s = cascade pitch
VR = diffuser velocity ratio
WRE = equivalent velocity ratio
Z = number of blades
γ = stagger angle
θ = camber angle
κ = blade camberline angle
σ = solidity
φ = stream surface slope angle
–
ω = total pressure loss coefficient
Subscripts
c = corrected (smoothed) parameter

1 = inlet condition
2 = discharge condition
Superscripts
9.1 GEOMETRY CONSIDERATIONS
Application of procedures from previous chapters requires specific procedures

for specifying the geometry of the end-wall contours, quasi-normals and blades

Aerodynamic Performance Analysis • 201
FIGURE 9-1 Basic Gas Path Geometry for an Axial-Flow Compressor
in a form suitable for the performance analysis. Figure 9-1 illustrates the basic
gas path geometry for a typical axial-flow compressor. The hub-and-shroud con-
tour coordinates are entered at the end points of a series of quasi-normals that
are used to conduct the analysis. In this example, all quasi-normals are simple
radial lines, although that is certainly not a requirement. This analysis used a sin-
gle quasi-normal between successive blade rows, although more could have been
used if they were considered necessary. For each quasi-normal after the first one,
it is necessary to specify what type of blade row lies upstream. Some choices for
this specification are rotor, stator, guide vane or none. It is useful to distinguish
between stators and guide vanes so that the analysis can correctly distinguish
stages by stage number to permit individual stage performance data to be output.
The end-wall contour specification should also identify the first and last quasi-
normals for which the hub wall is rotating, so that the end-wall boundary layer
analysis can use the correct wall rotation speed. In this case, that is simply the
first and last quasi-normal. It is also useful to provide for alternate input to size
the annulus. In that case, coordinates are specified for only one end-wall contour.
The angle between the quasi-normal and a radial line and the meridional velocity
for the mean streamline are the other data required. The program will then com-
pute the annulus area required and the coordinates of the other end-wall as out-
lined in Chapter 7.
The blade geometry also needs to be specified. The standard blade profile types
described in Chapter 4 should be available to make full use of the empirical blade
performance models of Chapter 6. The blade construction logic of Chapter 4 is
recommended so that the blade throat openings can be computed accurately.
Alternatively, the empirical approximation of Eq. (4-31) can be used, although the
more precise calculation is preferred. When the latter approach is used, provision
should be made to compute the throat openings one time and save the computa-
tion in the program’s input file once all other blade geometry is available. It is
inefficient to have the program perform these calculations every time a perform-
ance analysis is conducted. If a special blade profile such as a controlled diffusion
airfoil is in use, it will be necessary to extend the methods of Chapters 4 and 6, or
to confirm that use of one of the standard profiles can adequately approximate
the performance of the profile. Usual practice is to specify the geometry of blade
sections at a series of radii that extend over a range at least as large as the range
of end-wall radii at the quasi-normals before and after the blade row. Usually it is
reasonable to specify a single value of the location of the point of maximum cam-
ber, a / c, for each blade row. Similarly, a single specification per blade row can be

supplied for the number of blades in the blade row, Z, and the blade tip clearance,
δc. The most useful data to supply for the radial sections are the chord, c, the
thickness-to-chord ratio, tb / c, and the blade inlet and discharge angles, κ1 and κ2.
For double-circular-arc blades, a specification for the leading and trailing edge
radii may also be necessary, unless a standard specification is considered accept-
able. The blade angles are the most useful specifications, but may not be the most
convenient, so there is provision to opt for choices such as camber angle, θ, stag-
ger angle, γ or lift coefficient, Cl0. The program can then compute κ1 and κ2 from
the data actually supplied, as described in Chapter 4. In the case of shrouded sta-
tor blades, the shroud seal clearance and the number of seal fins will also be
required instead of the blade tip clearance. There are a number of features that
can easily be incorporated to greatly simplify the blade geometry specification
process, particularly for industrial axial-flow compressors. These include:
• The capability to copy the geometry of an already specified blade row.

• The capability to apply a geometrical scale factor to the data.
• The capability to impose a radial shift on the data.
• The capability to impose a change in stagger angle, i.e., to rotate the blade.
• The capability to import geometry from a file for a commonly used
standard profile.
• The capability to export geometry to a file for use as a standard profile
for future analyses.
• The capability to import geometry designed by the stage design proce-
dure described in Chapter 10.
Often application of one or more of these procedures can greatly simplify the
blade geometry input process so it requires at most, only minor editing, avoids
the need to enter all data.
When conducting the actual performance analysis, the blade geometry on a
stream surface is required. If the stream surface coordinates at the blade row
inlet and discharge are (z1, r1) and (z2, r2), respectively, the stream surface angle,
φ, is estimated from
r2 − r1
tan φ = (9-1)
z2 − z1
Then the corrected blade geometry on the stream surface is given by
c → c / cos φ (9-2)
tb / c → cos φ tb / c (9-3)
tan κ1 → cos φ tan κ1 (9-4)
tan κ 2 → cos φ tan κ 2 (9-5)
s = π ( r1 + r2 ) / Z
(9-6)
Where base values are obtained from the input data by interpolation with
respect to radius to obtain κ1 at r1, κ2 at r2, and other data at the mean radius.
Then all other blade geometry data can be computed as described in Chapter 4.

9.2 CASCADE PERFORMANCE CONSIDERATIONS
Some care is required in applying and interpreting the empirical cascade per-
formance models of Chapter 6 when predicting the performance of an axial-flow
compressor. When analyzing a compressor at far off-design conditions, specific
blade rows may operate locally in deep stall or in choke while other blade rows
operate under near optimum conditions. If the empirical models of Chapter 6 are
applied directly, the overall performance analysis will be very unreliable and
incapable of analyzing many cases where the compressor is capable of operation.
The major reason for this is that the meridional through-flow analysis of Chapter
7 does not permit any fluid mixing between stream sheets, such that entropy can
build up locally on a stream surface to cause the solution to diverge. In the actual
compressor blade rows, there are substantial secondary flow patterns or bound-
ary layer migration, which does result in fluid mixing between stream sheets.
Thus, it is necessary to impose limits on the loss coefficient models and apply
some artificial smoothing procedures to simulate the fluid mixing processes.
This writer imposes the following limit on all loss coefficients:
ω ≤ 0.5 (9-7)
This will rarely be required, except when analyzing flow points where local
choke is encountered in the blade passages. This can be encountered when the
compressor is operating close to its choke limit and flow profiles are highly dis-
torted. Smoothing of the total pressure loss is a more important consideration.
Without it, the performance analysis can be very unreliable for predicting per-
formance close to the surge line. This is particularly true at low off-design
speeds, where severe local blade stall is commonly encountered. If N stream sur-
faces are used from hub to shroud, the total pressure loss on stream surface
number i is given by
( ∆Pt′)i = ω i ( Pti′ − Pi )in (9-8)
The prime designates the total pressure in a frame of reference relative to the
blade row, and the subscript i designates the stream surface, numbered
sequentially from the hub contour. Smoothed total pressure loss values are
computed from
[ ]
( ∆Pt′)i, c = ( ∆Pt′)i −1 + 2( ∆Pt′)i + ( ∆Pt′)i +1 / 4 ; 1 < i < N (9-9)
( ∆Pt′)1, c = 2( ∆Pt′)2, c − ( ∆Pt′)3, c (9-10)
( ∆Pt′)N , c = 2( ∆Pt′)N −1, c − ( ∆Pt′)N −3, c (9-11)
Equations (9-10) and (9-11) are derived from a simple trapezoidal-rule numerical
approximation for the integral of the total pressure loss over the two stream
sheets adjacent to the walls. They result in those integrals being identical when
approximated using either the smoothed or the uncorrected total pressure loss
values at those three points.

9.3 STALL AND COMPRESSOR SURGE CONSIDERATIONS
An important consideration in performance analysis is estimating the compres-

sor’s surge limit from the detailed performance data. This is rather subjective
and imprecise, but there are some useful indicators that have been found to be
helpful in this process. It is well known that a pressure versus flow characteris-
tic at constant speed, which has a positive slope, is theoretically unstable.
Hence, a predicted characteristic that approaches a slope of zero can be consid-
ered to be an obvious indication of probable surge. But in most cases, the per-
formance analysis will not be capable of resolving the characteristic’s slope
accurately enough to rely on this. Indeed, the occurrence of a positive slope is
often not even obvious from experimental performance data. In the presence of
blade or end-wall boundary layer stall, there may be a very abrupt drop in com-
pressor discharge pressure that is not apparent from either the predictions or
experiment. Of course, in the case of the predictions, this lack of resolution is
really deliberate. Practices such as the loss smoothing previously discussed are
employed to achieve acceptable reliability, which will preclude resolving any
abrupt changes in performance due to the onset of local stall. The best we can do
is attempt to identify stall parameters that may indicate such an abrupt change
is likely to occur.
Blade stall is expected to be a function of the limit loading diffusion factors,
such as D or Deq of Chapter 6. In this writer’s experience, those parameters are of
limited value as a means of estimating stall limits likely to be associated with the
compressor’s surge limit. It is quite common to observe values of these diffusion
factors well in excess of typical recommended loading limits at operating points
well removed from the observed surge line. A stall criterion proposed by de
Haller (1953) is given by
W2 / W1 < 0.72 (9-12)
The generally accepted opinion today is that de Haller’s criterion is associated

with an end-wall stall rather than a blade stall. In any event, this criterion has
been found to be more effective than the diffusion factors as an indicator of the
onset of surge. But, again, a significant number of exceptions are encountered
where velocity ratios well below this limit are predicted at operating points that
are quite far from a compressor’s surge limit. Koch (1981) improved on the sim-
ple de Haller criterion by including blade geometry parameters. Koch approxi-
mated the blade passages, averaged over a stage, as a simple diffuser. He was able
to achieve reasonable agreement between observed stage pressure coefficients at
stall with the two-dimensional diffuser data of Sovran and Klomp (1967) for a
specific inlet blockage level. To obtain this correlation, the compressor stage data
was extensively corrected to adjust for differences in blade clearance, Reynolds
number and axial spacing between blade rows. This correlation is intended as a
means to estimate the maximum achievable stage pressure coefficient. It is of
limited value as a stall criterion for use in an axial-flow compressor performance
analysis. Here, a stall criterion for a blade row instead of a stage is needed. The
use of pressure coefficient is also undesirable, since it results in a stall criterion
that is a strong function of the blade performance models, which are by no

FIGURE 9-2 Two-Dimensional Diffuser Correlation
means universal methods. Finally, the Koch correlation requires a number of cor-
rections for it to be applicable to a specific axial-flow compressor.
This writer has developed a correlation for the onset of stall that is more appro-
priate to the present application. Reneau et al. (1967) developed a graphical cor-
relation for the geometry of simple two-dimensional diffusers operating at their
peak static pressure recovery. Aungier (2000) correlated their data in the form
AR = 1 + 0.4( L / b1)0.65 (9-13)
Where AR is the diffuser area ratio, L is the diffuser length and b1 is the diffuser
width at the inlet. Since the diffuser test data correspond to basically incom-
pressible flow, this can also be related to the ideal diffuser velocity ratio, VR.
VR = 1 / [1 + 0.4( L / b1)0.65 ] (9-14)
Equations (9-13) and (9-14) are presented in Fig. 9-2. Note that these equations
identify peak pressure recovery condition, but are not dependent on the value of
the peak pressure recovery. It is reasonable to expect the onset of stall in any dif-
fusing passage to closely correspond to this peak pressure recovery condition. To
apply this concept to a blade passage, it is assumed that L should be the camberline

length and b1 should be the staggered spacing. If the camberline is approximated

as a circular-arc, this can be written as
L / b1 → L / g = L / ( s cos γ ) = θσ / [2 sin(θ / 2) cos γ ] (9-15)
When applying an incompressible flow model to compressible flow through a

blade row, it is always preferable to employ velocity pressures rather than veloc-
ity to minimize the influence of Mach number effects, i.e.,
Pt′2 − P2
VR → WRE = (9-16)
Pt1′ − P1
WRE will be called the equivalent relative velocity ratio across the blade row.
Hence the expected stall criterion is
(0.15 + 11tb / c) / (0.25 + 10tb / c)

WRE < (9-17)
1 + 0.4[θσ / {2 sin(θ / 2) cos γ }]0.65
The numerator on the right-hand side of this equation reflects a very weak
dependence on the ratio of tb/c observed while comparing stall estimates from
Eq. (9-17) with observed axial-flow compressor surge limits. These comparisons
have also suggested two other modifications: First, the application to blade pas-
sages requires extrapolating Eq. (9-14) to values of L / b1 that are considerably
lower than those covered by the test data and correlation of Reneau et al. (1967).
Thus, a limit is imposed on the effective value of L/b1 used on the right-hand side
of Eq. (9-17) by requiring
θσ / [2 sin(θ / 2) cos γ ] ≥ 1.1 (9-18)
Second, Eq. (9-17) can be too pessimistic for highly diffusing blades, where blade
wake blockage can be expected to be significant. It is rather subjective and very
tedious to attempt to estimate stall in those cases. It has been observed that Eq. (9-
17) becomes pessimistic whenever Deq is greater than about 2.2. When Deq > 2.2, an
empirical correction to Eq. (9-17) has been found to be reasonably effective, i.e.,
[(2.2 / Deq )0.6 ](0.15 + 11tb / c) / (0.25 + 10tb / c)

WRE <
1 + 0.4[θσ / {2 sin(θ / 2) cos γ }]0.65 (9-19)
The stall criterion for Deq< 2.2 and tb / c = 0.1 is illustrated in Fig. 9-3.
In summary, this writer employs the following three basic criteria as a basis
for estimating the onset of compressor surge:
1. When the discharge pressure versus flow characteristic approaches a

slope of zero.
2. When end-wall boundary layer stall is predicted (H1 = 2.4).
3. When stall is indicated by Eqs. (9-17) through (9-19).

FIGURE 9-3 The Blade Row Stall Criterion
The first two criteria are employed at all speeds, while the third criterion is most
reliable for speeds greater than about 85% of design speed. At lower speeds, the
flow profiles typically become so distorted that the accuracy of the profile pre-
dictions with the inviscid through-flow analysis of Chapter 7 becomes too ques-
tionable for it to rely heavily on the third criterion. This will be illustrated in the
presentation of results from the performance analysis.
9.4 APPROXIMATE NORMAL EQUILIBRIUM RESULTS
The approximate normal equilibrium model described in Section 7.6 offers sub-
stantial advantages in computation speed and reliability over the full normal
equilibrium method. This approximation assumes that the stream surface slope
and curvature vary linearly between the two known end-wall contours. In this
section, typical results from the performance prediction procedures described in
this book using the approximate normal equilibrium model will be reviewed. In
the next section, results for the same problems using the full normal equilibrium
model will be reviewed for comparison.
Figure 9-4 compares performance predictions with experiment for the NACA
10-stage subsonic axial-flow compressor. The compressor design is described in
Johnsen (1952). The performance data are provided in Budinger and Thomson

FIGURE 9-4 The NACA 10-Stage Compressor
(1952). All blade rows in this compressor use fairly conventional NACA 65-series
blades. Design data for the inlet guide vane are not provided in the references, so
the performance analysis was started at a station downstream of the inlet guide
vane using experimental measurements of the flow angle distribution, with an
assumed end-wall boundary layer blockage of 2%. Typical end-wall boundary
layer blockage predictions for this case were shown in Fig. 8-8. The estimated
stall line based on stall criterion #3 is also shown in Fig. 9-4. The experimental
compressor surge limit is well approximated by the lowest flow data point for
each speed line. It is seen that stall criterion #3 is a good indicator of surge at the
design speed, but is pessimistic at lower speeds, where stall criterion #1 is the
best indicator. This compressor has the unusual characteristic that stall crite-
rion #3 is first encountered in the front stages at the rotor tips and that the limit
expressed in Eq. (9-18) is active at these locations. Under these conditions, stall
criterion #3 is of questionable validity and it is far too insensitive to variations in
mass flow rate to be very useful. At the tips of the front stage rotors of an axial-
flow compressor, the variation of WRE with mass flow rate is far too weak to
resolve a meaningful stall limit. For example, the difference in WRE between the
experimental and predicted surge limits at 90% speed is just 0.015. This is the
only axial-flow compressor encountered by this writer with this unusual charac-
teristic. It is normally the case that blade diffusion limits are first encountered
along the hub contour, where WRE shows the strongest variation with mass flow
rate. Flow diffusion limits are easier to avoid at the rotor tips, and designers

have strong incentive to do so due to the higher Mach number levels encoun-
tered there.
5-stage transonic axial-flow compressor (Kovach and Sandercock, 1961). Sander-
cock et al. (1954) describes the detailed design. Kovach and Sandercock (1954)
provide the experimental performance data. The first two stator rows and all
rotor rows in this compressor use double-circular-arc blades, while the last three
stator rows use NACA 65-series blades. This case provides a fairly significant test
of all high Mach number performance models used, including bow-shock losses,
supercritical Mach number effects and blade passage choking. This analysis was
started well up in the inlet passage with negligible inlet end-wall boundary layer
blockage. Figure 9-1 illustrates the compressor cross-section and the computa-
tional stations used in this analysis. Typical end-wall boundary layer blockage
predictions for this case were shown in Fig. 8-9. The estimated stall line from
stall criterion #3 is also shown in Fig. 9-5. The experimental compressor surge
limit is not well defined in the references, but presumably can be approximated
by the lowest flow data point on a speed line. At 90% and 100% speed, there is
substantial scatter in the experimental data, so the surge limit is less obvious and
possibly better indicated by the highest pressure-ratio point achieved on the
speed line. It is seen that stall criterion #3 is reasonably significant as an indica-
tor of surge at all speeds. This case provides a fairly dramatic illustration of the
merits of stall criterion #3, as shown in Fig. 9-6. The predicted equivalent velocity

FIGURE 9-6 Illustration of Stall Criterion 3
ratios defined by Eq. (9-16) along the hub contour are shown for all blade rows,
along with the stall limit values computed from Eqs. (9-17) through (9-19). The
simple de Haller limit of Eq. (9-12) is also shown for reference. It is apparent that
the simple de Haller limit would predict a substantially lower pressure ratio at
stall. Indeed, the de Haller criterion predicts that blade stall is present for the
entire predicted characteristic shown in Fig. 9-5. For the specific blade geometry
used in this compressor, Eqs. (9-17) through (9-19) predict a stall limit well below
the de Haller limit. Although an unusually extreme example, this case serves to
demonstrate the importance of a more fundamental stall criterion than that pro-
posed by de Haller.
8-stage transonic axial-flow compressor (Voit, 1953; Geye et al., 1953). The first
two rotor blade rows use double-circular-arc blades. All other blade rows use
NACA 65-series blades. The performance analysis for this case is a little less pre-
cise than the previous examples. The hub contour around the first stator row is
not well defined in Voit (1953) and the exit guide vane row geometry is not sup-
plied. The exit guide vane is not expected to have a significant influence on the
total-to-total pressure ratio prediction. So an exit guide vane was simply designed
for this compressor for use in the performance analysis. The exit guide vane
design used a blade configuration similar to that of the last stator row, with cam-
ber and stagger angles defined to effectively remove the swirl from the flow exit-
ing the last stator row. The analysis was started well upstream in the inlet passage

with negligible inlet end-wall boundary layer blockage. Typical end-wall bound-
ary layer blockage predictions for this case were shown in Fig. 8-7. The estimated
stall line from stall criterion #3 is also shown in Fig. 9-7. The experimental com-
pressor surge limit is well approximated by the lowest flow data point for each
speed line (Geye et al., 1953). Stall criterion #3 is a good indicator of the surge
limit at the 90% and 100% speed lines, but is somewhat conservative at the 70%
and 80% speed lines, where stall criterion #1 is a better indicator.
9.5 FULL NORMAL EQUILIBRIUM RESULTS
Figures 9-8 through 9-10 compare performance predictions using the full normal
equilibrium model with the approximate normal equilibrium results from the
previous section. Except for some minor differences in the stall lines estimated
from the two models, it is seen that the performance predictions from the two
models are essentially identical. It is quite evident that there is very little loss in
performance prediction accuracy when the approximate normal equilibrium
model is used for these three compressors. This is quite typical of this writer’s
experience on other axial-flow compressor performance analyses. There is
almost never a need to employ the full normal equilibrium model to obtain accu-
rate overall compressor performance predictions.


9.6 CONCLUDING REMARKS
This chapter provides supplemental procedures to apply the analyses described in

Chapters 6 through 8 to the prediction of the aerodynamic performance of axial-
flow compressors. The resulting performance analysis has been applied to three
NACA axial-flow compressors to demonstrate the level of prediction accuracy to
be expected. The compressors considered cover a significant variety of design
styles, ranging from the conservative subsonic NACA 10-stage compressor to the
more aggressive transonic designs used in the NACA 5-stage and 8-stage compres-
sors. But it must be recognized that the empirical models of Chapter 6 specifically
cover only the convention blade profiles of Chapter 4. Application of these proce-
dures to other blade profiles, such as proprietary controlled diffusion airfoils, may
require supplemental empirical performance models to achieve the same level of
prediction accuracy as demonstrated by the results provided in this chapter.
The performance predictions reviewed in this chapter show that the approxi-
mate normal equilibrium model of Section 7.6 provides excellent prediction
accuracy as well as dramatic improvements in computation speed and reliability.
This supports an earlier comment in this chapter suggesting that one-dimen-
sional mean-line performance analysis methods no longer offer any significant
advantage over the more fundamental hub-to-shroud flow performance analyses.
Some general guidelines have been suggested for estimating a compressor’s
surge limit based on the performance analysis results. These guidelines can

normally be refined and improved through comparison with experimental

results when the analysis is applied to user-specific design styles and blade pro-
file types. Surge is a transient phenomenon dependent upon the complete sys-
tem in which the compressor operates. Consequently, the surge limit cannot be
predicted by a compressor performance analysis. A performance analysis can
only provide fluid dynamic data that may be used as a guide to estimate the
probable surge limit. It should be apparent from the results presented in this
chapter that this process remains quite subjective, relying heavily on the expe-
rience and judgment of the aerodynamicist to interpret results from the per-
formance analysis. At speeds close to the design speed, the stages are
reasonably well matched. In these cases, stall criterion #3 is usually the best
indicator. At speeds much less than the design speed, excessive front stage load-
ing produces severe flow profile distortion, making the predicted values of WRE
too inaccurate for stall criterion #3 to be very useful. There, stall criterion #1 is
the most useful indicator. Although stall criterion #2 isn’t a factor in the cases
considered in this chapter, it is occasionally encountered. There is evidence to
suggest that stable operation is very unlikely when significant end-wall bound-
ary layer separation is predicted. In the few cases encountered, stall criterion
#2 provided a good estimate of the surge limit and none of those compressors
actually operated beyond that limit.
EXERCISES
9.1 Show that the integration of the corrected total pressure loss data from
Eqs. (9-9) and (9-10) with respect to mass flow rate between stream sur-
faces 1 and 3 yields the same result as integration of the uncorrected total
pressure loss data. Assume the mass flow rate is identical in all stream
sheets and equal to ∆ṁ. Use the simple trapezoidal rule approximation for
numerical integration, e.g.,
∆m
˙
∫ ˙ ≈ 1 [( ∆Pt′ )1 + ( ∆Pt′ )2 ]∆m

∆Pt′dm 2
˙
0
9.2 Derive an expression for the ratio of the blade camberline length-to-the
staggered spacing, s cosγ, for a circular-arc camberline to confirm Eq.
(9-15).

Chapter 10
COMPRESSOR STAGE
AERODYNAMIC DESIGN
The aerodynamic design of an axial-flow compressor stage involves definition of

the rotor and stator velocity triangles and blade geometry that will produce
desired stage performance characteristics. When appropriate, this may include
the design of inlet and exit guide vanes as well. In the context used in this book,
there are subtle but important differences between stage design and compressor
design. Stage design will involve the selection of basic dimensionless perform-
ance parameters and the design of appropriate blade rows to produce them. No
specific reference is made to the precise application of the stage in an actual
compressor. Without knowledge of the specific working fluid, Mach number lev-
els, matching with adjacent stages, etc., stage design is a rather idealized process.
It is not possible to establish precise end-wall contours and compute the losses in
this context. The purpose of this chapter is to introduce basic dimensionless per-
formance parameters and to discuss their application to the design process. Rep-
resentative applications of the stage designs will be used to indicate some of the
consequences of the choices made. Chapter 11 extends these basic design con-
cepts to the design of blade rows for complete axial-flow compressors with spe-
cific operating conditions and working fluids.
The stage-design procedures described in this chapter offer far more than
abstract educational value. They certainly can be used to design stages for actual
applications within a compressor, although the procedures in chapter 11 are
more efficient. They often play a key role in the design of industrial axial-flow
compressors. Each industrial axial-flow compressor usually has a unique design.
Development and manufacturing costs become primary considerations when
there are no duplicate machines to share them. It is fairly common practice to
employ a standard repeating stage design for these compressors. Blade stagger
angles are commonly adjusted to accommodate the specific applications. Scaling
of the blades with corresponding modification of the number of blades per row
may be used to satisfy mechanical strength requirements. Since the precise appli-
cation is not predefined and unique designs for all stages are not acceptable, the
design of a standard repeating stage will precisely follow the process described in
this chapter. The specific axial-flow compressor design will involve application of
this standard stage design with appropriate sizing of the annulus and various
minor adjustments to accommodate the application. An example of this use of
stage design procedures is included at the end of this chapter.

NOMENCLATURE
c = chord
H = total enthalpy
h = static enthalpy
i = incidence angle
K = ψ / ψlim
m = vortex exponent
n = vortex exponent
P = pressure
PR = pressure ratio
R = reaction
RR = recovery ratio
r = radius
tb = maximum blade thickness
U = local blade speed = ωr
β = flow angle
γ = stagger angle
η = efficiency
θ = camber angle
κ = blade camberline angle
ρ = gas density
σ = solidity
φ = flow coefficient
ψ = work coefficient
ψlim = value of ψ yielding W2 / W1 = 0.7
ω = rotation speed, radians/sec.
Subscripts
c = parameter on the reference radius

h = hub condition
R = rotor parameter
S = stator parameter
s = shroud condition
z = axial component
0 = inlet guide vane inlet condition
1 = rotor inlet condition
2 = stator inlet condition
3 = stator exit condition
4 = exit guide vane exit condition

Compressor Stage Aerodynamic Design • 217
Superscripts
* = minimum loss condition

10.1 DIMENSIONLESS PERFORMANCE PARAMETERS
In the design of an axial-flow compressor stage, the designer normally wants to

control the flow capacity, the work done per stage and the distribution of the flow
diffusion load between the rotor and stator. The specifications should be dimen-
sionless to enable application to any specific stage inlet thermodynamic condi-
tions, rotation speed and geometrical scale. This leads quite naturally to the
dimensionless performance parameters introduced in Chapter 1, Section 1.5. If ω
is the rotation speed, the local blade speed is given by U = ωr. The dimensionless
flow capacity of the stage will be governed by the dimensionless axial velocity
component or flow coefficient, defined by
φ = Cz1 / U (10-1)
The flow coefficient has been defined at the rotor inlet, since an inlet guide vane
is not necessarily defined as part of the stage design process. The dimensionless
work per stage follows directly from the Euler turbine equation, Eq. (3-9). The
work coefficient is defined as
ψ = ( H2 − H1) / U 2 = (Cθ 2 − Cθ1) / U (10-2)
where the subscripts 1 and 2 refer to the rotor inlet and exit stations, respectively,
using the station nomenclature illustrated in Fig. 10-1. The distribution of the
flow diffusion between the rotor and stator can be expressed in terms of the frac-
tion of the stage static enthalpy rise that occurs in the rotor, i.e., the stage reac-
tion, defined as
R = ( h2 − h1) / ( h3 − h1) (10-3)
Alternatively, reaction may be defined in terms of static pressures, but that is less
convenient for the design problem. Since Cr = 0 for this case, Eq. (3-11) and the
basic kinematics of the velocity components yield,
H = h + 12 C 2 (10-4)
C 2 = Cz2 + Cθ2 (10-5)
Noting that H3 = H2, Eqs. (10-2) and (10-4) yield
h2 − h1 = U (Cθ 2 − Cθ1) − (C22 − C12 ) / 2 (10-6)

FIGURE 10-1 Stage Station Number Nomenclature
h3 − h1 = U (Cθ 2 − Cθ1) − (C32 − C12 ) / 2 (10-7)
Now consider a repeating stage design, i.e., a stage that is designed assuming it
will be followed by another identical stage. This means that all velocity compo-
nents at the stator exit and rotor inlet must be identical. Let us further require
that Cz2 = Cz1. Then Eqs. (10-3) through (10-7) yield
R = 1 − (Cθ 2 + Cθ1) / (2U ) (10-8)
It follows that for a repeating stage design with Cz2 = Cz1, all relevant stage veloc-
ity triangles can be specified in terms of the dimensionless performance parame-
ters φ, ψ and R. For example, from Eqs. (3-1), (10-1), (10-2) and (10-8), it is easily
shown that
Cθ1 / U = Cθ 3 / U = 1 − R − ψ / 2 (10-9)
Cθ 2 / U = 1 − R + ψ / 2 (10-10)
Wθ1 / U = Cθ1 / U − 1 = − R − ψ / 2 (10-11)
Wθ 2 / U = Cθ 2 / U − 1 = ψ / 2 − R (10-12)
tan β1 = (1 − R − ψ / 2) / φ (10-13)

tan β2 = (1 − R + ψ / 2) / φ (10-14)
tan β1′ = −( R + ψ / 2) / φ (10-15)
tan β2′ = (ψ / 2 − R) / φ (10-16)
Specifications for the inlet and exit guide vanes have not been discussed yet. The
simplest specification is to require Cz to be constant across these vanes and Cθ0 =
Cθ4 = 0.
10.2 APPLICATION TO STAGE DESIGN
Development of the dimensionless performance parameters in the previous sec-

tion assumed constant Cz (or φ) throughout the stage. But Cz is also governed by
basic fluid dynamics, as has been described in Chapter 7. In particular, Cz must
satisfy some form of the normal equilibrium equation. This means that the stage
dimensionless performance can only be specified at one radius for any axial sta-
tion. The radius at which the dimensionless performance is specified will be des-
ignated as rc, and a subscript, c, will be used to identify all parameters on that
radius. Hence, the stage dimensionless performance will be specified by
φ c = Czc / Uc (10-17)
ψ c = (Cθ 2c − Cθ1c ) / Uc (10-18)
Rc = 1 − (Cθ1c + Cθ 2c ) / (2Uc ) (10-19)
In Chapter 7, normal equilibrium and conservation of mass were combined to

complete the solution. That requires that the stage design be specific to the work-
ing fluid and Mach number level for which it is to be applied. In this chapter, it is
useful to avoid imposing those constraints in the interest of a more general inves-
tigation of design alternatives. This will also accommodate the general industrial
repeating stage design application discussed earlier in this chapter. The applica-
tion-specific design problem will be considered in chapter 11.
Hence the problem to be addressed in this chapter is illustrated in Fig.10-2.
The hub, shroud and reference radii will all be held constant through the stage.
The stage will be designed as a repeating stage with φc constant throughout the
stage, i.e., Czc is constant on the reference radius. The absence of an equation of
state and Mach number dependence precludes a meaningful calculation of
losses from the models presented in Chapter 6, so the flow will be considered to
be isentropic. Common design practice is to require constant total enthalpy
from hub to shroud, which will be assumed to be the case in stage design stud-
ies considered in this chapter. It will be shown later in this chapter that it is rel-
atively easy to include provisions to relax many of these constraints in an actual
computerized stage design system to obtain a more general design capability.
But that does not add any benefits to developing an understanding of the merits
of alternate dimensionless performance specifications. The flow profiles at other
radii will be supplied by applying the simple, isentropic normal equilibrium

FIGURE 10-2 The Stage Design Application
equation to a prescribed tangential velocity distribution. A general vortex type

equation is applied at the rotor inlet.
Cθ1 / Uc = (1 − Rc )( rc / r )n − (ψ c / 2)( rc / r )m (10-20)
where the exponents n and m are specified. From Eq. (10-9), it can be seen that
this vortex equation is consistent with the dimensionless performance specifica-
tions at r = rc. It will be seen later in this chapter that this vortex equation
includes a wide variety of design styles, including the styles most commonly
used. At the rotor exit, the constant total enthalpy assumption is used.
Cθ 2 / Uc = Cθ1 / Uc + ψ c ( rc / r ) (10-21)
For our repeating stage design problem, the other tangential velocity distribu-
tions are given by
Cθ 3 / Uc = Cθ1 / Uc (10-22)
Cθ 0 / Uc = Cθ 4 / Uc = 0 (10-23)
The basic normal equilibrium equation is obtained from Eq. (3-30) by expressing it
in stationary coordinates, neglecting streamline curvature and entropy gradients.

For the present case, the meridional coordinate is z, and the normal coordinate is
r. This yields
∂Cz Cθ ∂rCθ ∂H
Cz + = (10-24)
∂r r ∂r ∂r
In cases to be considered in this chapter, the total enthalpy gradient will also be
zero, but this term has been retained to accommodate provision for a more gen-
eralized work distribution. Equation (10-24) can be integrated numerically, with
the constant of integration supplied by the known value of φc. Hence, the dimen-
sionless velocity components and velocity triangles can be computed at all of the
axial stations on Fig. 10-1 and at all radii once values for φc, ψc, Rc, n and m are
supplied. Specified values of the hub, shroud and reference radii are also
required. The design of blades to produce these velocity triangles is discussed in
the following section.
Practical application or evaluation of these ideal stage designs is provided
through the performance analysis procedures described in Chapters 6 through 9.
The ideal stage design is accomplished over a range of radii sufficient to pass the
desired mass flow for the intended applications. The performance analysis then
selects the portion of the blades designed that are actually required, using the
annulus sizing capability described in Chapter 7, Section 7.7.
10.3 BLADE DESIGN
Once the velocity component and velocity triangle distributions have been com-
puted at the entrance and exit stations for all blade rows to be designed, the blade
geometry to be used can be computed. Some basic geometry data must be speci-
fied to carry out the blade design process. Several alternative specifications are
possible, but this writer finds the following specifications to be effective and rel-
atively easy to supply:
• Specify the blade camberline and profile type to be used.

• Specify the number of blades in the blade row, Z.
• Specify the chord, c, at the hub, reference and shroud radii.
• Specify the thickness-to-chord ratio, tb / c, at the hub, reference and
shroud radii.
• If required, specify the location of maximum camber, a / c.
• For double-circular-arc blades specify the leading and trailing edge
nose radii, r0. A convenient specification is the ratio, r0 / tb.
• Specify the difference between the incidence angle and the design
incidence angle, ∆i = i – i*, at the hub, reference and shroud radii.
Values of c, tb / c and ∆i at other radii are obtained by interpolation from the three
points supplied. The local solidity is given by
σ = cZ / (2πr ) (10-25)

The blade angle sign convention used in this book generally results in a negative
turning angle for rotors and inlet guide vanes. To apply the blade geometry pro-
cedures of Chapter 4 and the empirical cascade performance models of Chapter
6 in these cases, it is necessary to change the signs on all flow and blade angles.
After the blade design is complete, the signs of all angles are changed back to
obtain blade geometry consistent with the sign convention used. The blade
design process will be illustrated for the rotor blade, with the understanding that
this sign adjustment is required. Design of other blade rows is handled in the
same manner. The procedure is as follows:
1. Initialize the blade inlet and discharge angles equal to the relative flow
angles, i.e., κ1 = β 1′ , κ2 = β 2′ .
2. Compute all other cascade geometry from κ1 and κ2 and the above
geometry specifications, using the procedures of Chapter 4.
3. Compute i* and δ* using the procedures of Chapter 6.
4. Recompute the blade inlet and discharge angles from κ1 = β 1′ – i* – ∆i
and κ2 = β 2′ – δ*.
5. Repeat Steps 2 through 4 until convergence on κ1 and κ2 is achieved.
When the blades have been designed at all radii, all data required for the per-
formance analysis are available. Detailed camberline and blade profile data can
also be obtained as described in Chapter 4.
10.4 SELECTING THE STAGE PERFORMANCE PARAMETERS

It may appear that the designer has a vast array of choices for stage design
through the basic performance specifications of φc, ψc, Rc, n and m. This is cer-
tainly true to some degree, but practical considerations dramatically reduce the
designer’s choices. When recommendations from the literature, gained from past
stage design experience, are considered, the designer’s range of choices actually
become quite limited. Figures 10-3 through 10-5 illustrate practical constraints.
Here, it is assumed that the relative velocity ratio across any blade row should
not be less than 0.7. This is an adjusted de Haller stall limit obtained from Eq.
(9-17) using the limit expressed in Eq. (9-18) and with tb / c = 0.1. Noting that the
minimum relative velocity ratio will almost always occur at either the hub or
shroud radius, it is doubtful that designing for lower velocity ratios at the refer-
ence radius would be a wise choice. Eqs. (10-11) and (10-12) can be used to
express the rotor blade relative velocity ratio as
W2 φ c2 + (ψ c / 2 − Rc )2
= (10-26)
W1 φ c2 + (ψ c / 2 + Rc )2
Similarly, for the stator, Eqs. (10-9) and (10-10) yield
C3 C1 φ c2 + (1 − Rc − ψ c / 2)2 (10-27)
= =
C2 C2 φ c2 + (1 − Rc + ψ c / 2)2

FIGURE 10-3 ψ-φ Limits for 50% Reaction
FIGURE 10-4 φ-ψ Limits for 25% and 75% Reaction

FIGURE 10-5 φ-ψ Limits for 0% and 100% Reaction
Eqs. (10-26) and (10-27) yield identical results for 50% reaction, i.e., the velocity
triangles for the rotor and stator are identical (except for the sign convention). If
these equations are applied for 25% and 75% reaction, it can also be seen that the
minimum velocity ratio will be the same for both cases. The minimum simply
appears on different blade rows for these two cases. Thus, the restrictions
imposed by the velocity-ratio limit are symmetrical about Rc = 0.5. It can be use-
ful to specify the work coefficient as some fraction of the value at this velocity-
ratio limit to more easily choose practical values. This is easily done using the
following empirical correlation for ψc corresponding to the velocity ratio of 0.7.
ψ lim (φ , R) =
6 Rˆ  0.5 
+ 0.85 
1.18
φ
(2+0.1/ Rˆ ) (10-28)
17  Rˆ 
Rˆ = 0.5 + R − 0.5 (10-29)
Results from this empirical equation are illustrated in Figs. (10-3) through (10-5).
Another constraint used on Figs. 10-3 through 10-5 is that the absolute values
of all flow angles should not exceed 70°. Larger flow angles are likely to be imprac-
tical with respect to maintaining an acceptable throat area. Beyond that, many of
the empirical cascade performance correlations of Chapter 6 are limited to flow

angles less than 70°, which is sufficient reason to impose this limit. Noting that β3
= β1, Eqs. (10-14) and (10-15) can be used to express this limit as approximately
ψ c / 2 ≤ 2.75φ c − Rc (10-30)
ψ c / 2 ≤ 2.75φ c + Rc − 1 (10-31)
Note that the restrictions imposed by this constraint are also symmetrical with
respect to Rc = 0.5. Thus, these simple practical considerations have substantially
reduced the range of choices for φc and ψc. It can also be noted that 50% reaction
permits designing for the lowest values of φc and ψc. Indeed, as reaction increases
or decreases from 50%, the major impact seen in Figs. 10-3 through 10-5 is that
the acceptable range of design choices for φc moves toward higher values. There
is also a reduction in the range of acceptable values of ψc for a given value of φc,
but that effect is relatively minor.
Another important consideration is obtaining an acceptable surge margin, i.e.,
an acceptable flow range between the flow rate at the design point and the flow
rate at which surge occurs. Alternatively, surge margin may be expressed in terms
of the difference in discharge pressure between the design point and the surge
point. It is clear that improved surge margin can be obtained by designing for a
steeper slope on the constant speed pressure-flow characteristics. This resists the
trend for increased losses or abrupt stalls to force the characteristic toward an
unstable positive slope as the flow rate is reduced. Cumpsty (1989) notes that it is
useful to approximate the work input curve in the form
ψ = 1 − φ (tan β1 − tan β2′ ) (10-32)
This relation follows directly from Eqs. (10-13) and (10-16). The advantage of
this form of the work input equation is that the flow angles involved are both
discharge angles relative to the upstream blade row. It was shown in chapter 6
that the deviation angle is a very weak function of incidence angle. Hence, these
flow angles can be considered to be approximately constant over a speed line. As
illustrated in Fig. 10-6, this can be used to approximate Eq. (10-32) by
ψ ≈ 1 − (1 − ψ c )φ / φ c (10-33)
It follows that the surge margin will be improved by choosing lower values of
both φc and ψc. From Eqs. (10-13), (10-16) and (10-32) it is easily shown that the
slope of ψ as a function of φ will be positive if ψc > 1 is selected. This is certainly
inconsistent with a reasonable surge margin. Hence, ψlim > 1 is treated as a third
constraint on Figs. 10-3 through 10-5. Cumpsty (1989) also notes that the slope of
the work-input characteristic is important to resist the influence of local total
pressure distortions, even at operating conditions far from surge. He illustrates
this by approximating the flow as incompressible, such that
Pt = P + 12 ρ W 2 = P + ρ U 2φ 2 / cos2 β ′ (10-34)

FIGURE 10-6 Approximate Work-Input Characteristic
Hence a local perturbation in total pressure, within an essentially constant static

pressure flow field, can be approximated by
δPt = 2ρ U 2φδφ / cos2 β ′ (10-35)
It follows that a local deficit in total pressure will produce a local deficit in φ. If
the work-input characteristic is sufficiently steep, the corresponding increase in
ψ will reduce, and possibly eliminate, the total pressure deficit.
Smith (1958) provides a more detailed analysis of the capacity of a stage
design to resist local total pressure deficits. In a discussion at the end of the paper
by Smith, Ashley reports that the NACA had independently arrived at almost
exactly the same result. Using arguments analogous to Eq. (10-35), Smith devel-
oped an equation to express the magnitude of a pressure deficit, (δPt)out, at the
stage exit caused by an local inlet deficit, (δPt)in. He expressed this in terms of a
parameter, RR, which he called the recovery ratio.
(δPt )out
RR = 1 − (10-36)
(δPt )in
It follows that for RR = 1, the inlet total pressure deficit has been completely
removed at the stage exit. For the case corresponding to the present dimensionless

performance parameters at the reference radius, Smith’s recovery ratio can be

expressed as
RR = (cos2 β1 tan β1 − cos2 β2′ tan β2′ ) / φ + cos2 β1 cos2 β2′ tan β1 tan β2′ ) / φ 2 (10-37)
Once again, it can be shown that RR is symmetrical about R = 0.5. Figs. 10-7
through 10-9 show typical results for various values of reaction. To maintain rea-
sonable values of ψc in these predictions, φc was constrained to the limits shown
in Figs. (10-3) through (10-5), and Eq. (10-28) was used in the form
ψ c = K cψ lim (φ c , Rc ) (10-38)
While values of RR = 1 are certainly not required, it is expected that small or neg-
ative values should be avoided where possible. Based on results similar to those
shown in Figs. 10-7 through 10-9, Smith suggested that values of φc < 0.5 are pre-
ferred. Indeed, φ < 0.5 at all radii would be preferred, which is a much stronger
constraint. Clearly, if φc = 0.5 were imposed as an upper limit on Figs. 10-3
through 10-5, the range of choices available to the designer would be narrow.
This is not considered to be appropriate in light of the many successful axial-flow
compressor designs that have used significantly higher values of φc. But there
does appear to be a definite trend toward lower values of φc in more modern
FIGURE 10-7 RR for 50% Reaction

FIGURE 10-8 RR for 25% and 75% Reaction
FIGURE 10-9 RR for 0% and 100% Reaction

designs. Smith also concluded that the recovery ratio is not a strong function of
reaction, which may be somewhat questionable. It is evident that RR < 1 is to be
expected for any practical combination of φc and ψc when Rc is 0 or 1.
Other guidance available in selection of φc, ψc and Rc are more in the form of
opinions and rule-of-thumb practices. It has been common practice to favor 50%
reaction designs, based on the intuitive benefit expected from distributing the
flow diffusion load equally between the rotor and stator. Sometimes slightly
higher reaction is recommended based on the assumption that rotors are slightly
more efficient than stators. Similarly, designers seem to be most comfortable with
choosing ψc ≤ 0.4 in most cases. None of these opinions have been well substanti-
ated by either experiment or analysis. Indeed, such guidelines are probably rather
academic, since the selection of φc, ψc and Rc is dependent on many other factors.
The swirl vortex type discussed in the next section of this chapter is very impor-
tant. That determines the stage performance at other radii, and may have a signif-
icant influence on the choice of the reference radius performance parameters.
Also, the developments in this chapter do not consider the effect of the Mach num-
ber level, which can have important consequences and impose definite con-
straints. That can be illustrated by a design goal common to nearly all axial-flow
compressor designs, i.e., achieving the maximum mass flow per unit frontal area
and the minimum number of stages. For aircraft engine compressors this permits
essential size and weight reductions, while for industrial compressors the impor-
tant benefit is reduced size and cost. In either case, this immediately encourages
selection of larger values of ψc and φc, although lower values of φc are generally
preferred. This might be resolved by simply designing for a higher speed to satisfy
both needs. But Mach number effects impose definite limits on that approach as
well as result in performance penalties when the approach is used.
10.5 SELECTING THE SWIRL VORTEX TYPE
In addition to the reference radius performance parameters, it is necessary to

select the vortex type by specifying n and m in Eq. (10-20) to set the rotor inlet Cθ
distribution. The Cθ distributions at other stations are given by Eqs. (10-21)
through (10-23) for a constant-work, repeating stage. When the Cθ distributions
are known, the Cz distributions at all stations are obtained by solving Eq. (10-24).
Then the velocity triangles are known at all radii for all stations. If those velocity
triangles are acceptable, the blades can be designed as described in Section 10.3.
At this point in the design process, fluid dynamic data are available at all radii
and must be evaluated. The same design preferences and practical limits consid-
ered relative to the reference radius certainly are relevant at other radii as well. It
will be seen that problems associated with excessive flow diffusion or expected
Mach number limits almost always occur first on the end-walls, usually at the
rotor shroud and stator hub locations. It is not uncommon for an apparently very
conservative selection of φc, ψc and Rc to produce unacceptable performance
parameters at other radii. In the following sections, some commonly used vortex
types will be reviewed to illustrate some of the important considerations in
selecting the vortex type. To allow comparison of the various vortex types, a com-
mon design problem will be considered. All examples will be constant-work,

repeating stage designs. A hub-to-tip ratio of rh / rs = 0.5 will be considered, which

is typical of the front stages in an axial-flow compressor. The reference radius
will be set to rc / rs = 0.8, which approximately divides the annulus into two equal
area passages. In principle, the reference radius is an arbitrary design parameter,
but, in practice use of a typical mean-radius is usually necessary. Since Cz is con-
stant on rc, use of a mean-radius ensures approximate conservation of mass,
which will be required when the stage is applied in a compressor. If the stage
design deviates too far from this, the velocity triangles produced after sizing of
the compressor annulus area may be very different from those intended.
10.6 FREE VORTEX FLOW

It is instructive to start with the popular free vortex design style, which is given
by setting n = m = 1. From Eqs. (10-20) through (10-23) it can be shown that this
yields constant angular momentum (rCθ) profiles at all stations. If rCθ and H are
constant, Eq. (10-24) also yields constant Cz profiles. Indeed, since Cz is constant
on the reference radius, this means that Cz is constant everywhere. Let us design
a free vortex stage using very conservative dimensionless performance parame-
ters. Following Smith, (1958) set φc = 0.5. To provide significant margin from the
loading limit, set Kc = 0.8. Choose Rc = 0.5 to distribute the loading equally
between the rotor and stator. It would be expected that such conservative design
specifications and such a simple type of vortex would certainly result in an
acceptable design. Figures 10-10 and 10-11 show some of the important results
obtained with these design specifications. It is apparent from the flow angle dis-
tributions that this design will require a highly twisted rotor blade, which may
pose structural and manufacturing problems. The stator blade will be highly
staggered to match flow angles close to the 70° limit suggested earlier in this
chapter. But more serious is the fact that the design yields a substantial region of
negative reaction near the hub. This results in significant flow acceleration
across the rotor, followed by significant deceleration across the stator. The stator
is overloaded near the hub, and probably will stall. Since ψh > 1, stability is a con-
cern, as discussed in Section 10.4. The unacceptable features of this example
could have been anticipated. Since φ is constant across the blade rows at all radii,
Eq. (10-8) can be applied at any radius to predict the reaction. From Eq. (10-20),
free vortex flow yields
Cθ1 / U = (1 − Rc − ψ c / 2)( rc / r )2 (10-39)
Equation (10-21) requires
Cθ 2 / U = Cθ1 / U + ψ c ( rc / r )2 (10-40)
Substituting Eqs. (10-39) and (10-40) into Eq. (10-8) yields
R = 1 + ( Rc − 1)( rc / r )2 (10-41)

FIGURE 10-10 Flow Angle Distributions
FIGURE 10-11 Blade Loading Distributions

Hence the condition to avoid negative reaction at the hub for free vortex flow is
Rc ≥ 1 − ( rh / rc )2 (10-42)
For this example this requires Rc ≥ 0.609. Rc = 0.5 can be maintained by requiring
rh / rc ≥ 0.7071. Since Cz is constant everywhere, the choice of rc is quite arbitrary
for free vortex flow. It is likely that Rh significantly greater than zero would be pre-
ferred, which can be easily obtained using Eq. (10-41). This simple example illus-
trates that selection of the dimensionless stage performance parameters is strongly
influenced by the vortex type to be used and the radius-ratio range required.
Instead of refining the previous example, it is more useful to continue the dis-
cussion of free vortex flow with a more commonly used form obtained by adding
the requirement that Cθ1 = Cθ3 = 0. The advantage of this form is that no inlet or
exit guide vanes are required. From Eq. (10-9) this requires
ψ c = 2(1 − Rc ) (10-43)
To maintain a negative slope on the work-input characteristic, ψc < 1 is required.

That requires Rc > 0.5, which means that the rotor is more highly loaded than the
stator. To maintain a reasonable loading level in the rotor, it is desirable to spec-
ify the velocity ratio, W2c / W1c, as well. Equations (10-26) and (10-43) yield
(W2c / W1c )2 − (1 − 2Rc )2

φ c2 = (10-44)
1 − (W2c / W1c )2
Figure 10-12 is a (Rc, φc) design chart with W2c / W1c as a parameter, with the
usual practical limits imposed. Clearly, Smith’s recommendation of φc ≤ 0.5
requires rather high reaction. From Eq. (10-43) that also requires very low values
of ψc. Thus, it may be difficult to follow Smith’s recommendation while main-
taining a reasonable stage work input coefficient. For example, Fig. 10-12
includes the (Rc, φc) mean-radius design point for the first stage of the NACA 5-
stage axial-flow compressor (Sandercock et al., 1954). This is an actual example
of a free vortex design with Cθ1 = 0, which illustrates the type of compromise that
may be required in a practical stage design.
To illustrate features of the free-vortex design, the basic specifications for the
first stage of the NACA 5-stage axial-flow compressor were used to generate a stage
design using the procedures presented in this chapter. Figure 10-13 shows the rele-
vant flow angles for this design. A variable camber, highly twisted rotor blade will
be required, although not as extreme as the earlier example. Figure 10-14 shows
that the rotor tip inlet relative velocity is by far the largest velocity in the stage and
will be the value to be evaluated in terms of Mach number levels. No general con-
clusion can be reached without specifying the stage inlet conditions and rotation
speed, but the rotor tip will be the critical location. Indeed, as observed in chapter
9, the first stage of the NACA 5-stage compressor is a transonic design. The stator
inlet and exit velocity levels are quite modest by comparison, so Mach number level
should not be a concern. Figure 10-15 shows that the velocity ratios across the

FIGURE 10-12 Free Vortex Design Chart
FIGURE 10-13 Free Vortex Flow Angles

FIGURE 10-14 Free Vortex Velocities
FIGURE 10-15 Free Vortex Performance Parameters

blades are relatively close to the de Haller limit on both the rotor and the stator. As
seen in Figure 10-12, the design specifications correspond to a rotor relative veloc-
ity ratio at the reference radius of only 0.75, so this is really to be expected. The
reaction is significantly positive at the hub, which has avoided the velocity ratio
problems seen in Fig. 10-11 for the simple example considered earlier. A smaller
value of ψc probably should be used to avoid the ψh > 1 condition and to improve
the velocity ratios across the blade rows.
The blade design procedures described earlier in this chapter were used to
design double-circular-arc blades to match the velocity triangles. The blades were
designed with constant chord, and with tb / c = 0.1. The number of blades in each
row was set to yield σ = 1 at the reference radius. Figure 10-16 shows the camber
and stagger angle distributions for the rotor and the stator. As expected, a highly
twisted, variable camber rotor blade is required. Variations of camber and stag-
ger angles with radius on the stator are relatively modest. However, the radial
variation in β2, as seen in Fig. 10-13, is probably too great to seriously consider
using a constant camber, constant stagger stator blade for this stage.
10.7 CONSTANT REACTION VORTEX FLOW
Another vortex type occasionally suggested is obtained by setting n = -1 and

m = 1. Inserting these values into Eqs. (10-20) and (10-21) and substituting
FIGURE 10-16 Free-Vortex Blade Design

into Eq. (10-8) appears to yield a constant reaction design, which is the title
commonly applied to this vortex type. It must be recognized that, normally,
constant reaction will not be achieved, since φ is not generally constant across
the rotor as is assumed by Eq. (10-8). This is easily illustrated by repeating the
“conservative” design used as an example of free vortex flow in the previous
section. Figures (10-17) and (10-18) show results from that design with the con-
stant reaction type of vortex. It is seen that the design is by no means that of
constant reaction. Indeed, reaction is close to zero at the hub radius. Velocity
ratios across both the rotor and stator are well below the de Haller limit, which
is not very encouraging for operation at the stage’s design point. But the major
problem is that the axial velocity at the rotor discharge drops to zero below the
tip radius. The stage design software used restricts the axial velocity from
becoming negative to prevent fatal errors in the analysis. So the zone of zero
axial velocity is actually a reverse flow zone. Data predicted in this zone are not
accurate, as indicated by the dashed lines near the shroud radius. Once again, it
is seen that the specific vortex style can produce an unacceptable design, even
though the dimensionless performance parameters appear to be rather conser-
vative. It is quite evident that the large variation in axial velocity between the
rotor exit and inlet stations has completely nullified the intention of achieving
an essentially constant reaction.
One of the exercises at the end of this chapter will be to show that the radial
gradient of reaction for any constant work, repeating stage is given by
FIGURE 10-17 Axial Velocity Distributions

FIGURE 10-18 Performance Parameter Distributions
∂R
rc = 2(1 − Rc )( rc / r )(2+ n) + ψ c [( rc / r )3 − ( rc / r )(2+ m) ] (10-45)
∂r
With n = -1 and m = 1, this yields
∂R
rc = 2(1 − Rc )rc / r (10-46)
∂r
Equation (10-45) can be integrated to yield
R = Rc + 2(1 − Rc ) ln( r / rc ) (10-47)
Clearly, the only case for which this vortex type will yield constant reaction is
when Rc = 1. For other values of Rc, Eq. (10-47) can be used to compute a value of
Rc that will yield a desired value of reaction at the hub. It is clear that reaction is
completely independent of φc and ψc. However, it will be possible to eliminate the
reverse flow zone by increasing φc. Indeed, another exercise at the end of this
chapter will be to show that the condition to avoid reverse flow at the rotor exit
for any radius, r, for the constant reaction vortex style is
φ c2 ≥ 2(1 − Rc )2 [( r / rc )2 − 1] + 2(1 − Rc )ψ c ln( r / rc ) (10-48)

Reverse flow will first occur at the rotor exit on the shroud. Hence Eq. (10-48)
can be used to compute the minimum acceptable value of φc for any value of Rc
using r = rs. Alternatively, the value of Rc required to avoid reverse flow for any φc
can be estimated from Eq. (10-48) by trial and error. In this way, it can be shown
that values of φc down to 0.5 will be achievable for this case with no reverse flow if
Rc ≥ 0.6 is used. This limit was used to investigate the influence of φc and Rc on the
stage design for this vortex type. The influence of φc is shown in Fig. 10-19. It is
seen that φc has little influence on the velocity ratios across the blades. Increasing
φc increases the inlet velocity for both blade rows. Since the loss is proportional to
the inlet kinetic energy, this is expected to increase the losses. It may also lead to
other problems due to the increased Mach number levels. The only real benefit
from larger values of φc is the higher values of ψc obtained for the same value of
Kc. The influence of Rc is shown in Fig. 10-20. Increasing Rc will increase the rotor
inlet kinetic energy, raising concerns about loss and Mach number levels that are
similar to the concerns regarding Fig. 10-19. The reduction of the stator inlet
kinetic energy is some compensation for these concerns. The work coefficient is
nearly independent of Rc for a specified Kc. Hub reaction and the minimum veloc-
ity ratio across the stator both benefit significantly from increased values of Rc.
The minimum velocity ratio across the rotor appears a little erratic, which is
caused by the location of the minimum value switching from the hub radius to the
shroud radius. Probably the most notable observation is that there is an optimum
value of Rc at about 0.79 that yields the highest minimum velocity ratio across
FIGURE 10-19 Influence of Flow Coefficient

FIGURE 10-20 Influence of Reaction
either blade row. This would be expected to provide the best stall margin, assum-
ing the increased rotor inlet Mach number level does not become the limiting fac-
tor. To investigate this more closely, a constant reaction vortex stage was designed
based on this optimum. For the purpose of comparison, a free vortex stage was
designed for the same set of dimensionless performance parameters. Figures
10-21 through 10-23 compare some of the more significant results. It is seen that
the constant reaction vortex flow yields the lowest peak rotor inlet velocity and
offers greater margin from stall as indicated by the velocity ratios across the blade
rows. Both vortex types require a twisted, variable-camber rotor blade, but the
variations are less extreme for the constant reaction vortex style. Twisted, vari-
able-camber stator blades are required for both vortex types, with slightly more
extreme variations for the constant reaction type of vortex. The work input coeffi-
cient distributions for both vortex types are identical, with ψh = 0.913, so the work
input characteristics at all radii should have the desired negative slope. It can be
concluded that the constant reaction vortex style does offer some advantages over
the free vortex style for this specific set of dimensionless performance parameters,
particularly with regard to increased stall margin.
The most important conclusion with regard to the constant reaction vortex
style is that the usual claim that it yields blades of nearly constant reaction is
simply not correct, unless Rc values that are very close to 1.0 are used. In other
cases, it is necessary to choose a reaction at the reference radius that will yield an
acceptable reaction at the hub, much like the case of the free vortex style. This is

FIGURE 10-21 Comparison of Velocity Ratios
FIGURE 10-22 Comparison of Rotor Blade Geometry

FIGURE 10-23 Comparison of Stator Blade Geometry
easily accomplished with the aid of Eq. (10-47), so effective use of the constant
reaction vortex style is not too difficult. It is worth considering whether this vor-
tex style can be modified so that it will more closely approximate a constant reac-
tion stage. From Eq. (10-45) it is clear that there is really no alternate selections
for m and n that will yield constant reaction. One of the exercises at the end of
the chapter will be to show that the only possible constant reaction design is the
100% reaction case. A very modest reduction in the radial gradient of the reac-
tion can be achieved by choosing n = -2 and m = 1, which yields
∂R
rc = 2(1 − Rc ) (10-49)
∂r
R = Rc + 2(1 − Rc )( r / rc − 1) (10-50)
This has the undesirable effect of producing steeper gradients near the shroud
radius, to impose additional restrictions to avoid reverse flow. It is possible to
more closely approximate constant reaction by choosing n = -1 or -2 and select-
ing m > 1. From Eq. (10-45) it can be seen that this introduces a negative contri-
bution to the radial-gradient of R when r < rc to partially balance the usual
positive gradient. However, it has the reverse and unfavorable effect when r > rc.
This approach is generally not too effective, but it could be considered for very
specific applications.

10.8 CONSTANT SWIRL AND EXPONENTIAL VORTEX

FLOWS
Horlock (1958) suggests that the exponential vortex style is often a good choice,
and is obtained by setting n = 0 and m = -1. A very similar alternative is obtained
by setting n = m = 0, which will be called the constant swirl vortex style. As seen
from Eq. (10-20), the constant swirl vortex style yields constant Cθ1. With appro-
priate choices for φc, ψc, and Rc, these two vortex types may yield a constant-cam-
ber stator blade, possibly with a constant stagger angle as well. This is certainly
attractive for reducing manufacturing cost and may offer other advantages, such
as improved structural characteristics. From Eq. (10-45) it is easily shown that
the reaction at any radius for exponential vortex flow is given by
R = Rc − 2(1 − Rc )( rc / r − 1) (10-51)
whereas for constant swirl vortex flow,
R = Rc − [2(1 − Rc ) + ψ c ( rc / r − 1) / 2]( rc / r − 1) (10-52)
Hence the exponential vortex style has the advantage that the value of Rc required
to achieve any desired hub reaction is easily determined. It is also clear that the
exponential vortex flow will yield the higher hub reaction for any value of Rc.
While these advantages tend to favor the exponential vortex style, the constant
swirl vortex flow also has a definite advantage, as illustrated by Figs. 10-24 and
10-25. These figures show the difference in the stator camber angle and stagger
angle between the hub and the shroud for stages designed with these two types of
vortex. If the difference in camber angle is small enough, a constant camber sta-
tor blade can probably be used. Similarly, if the difference in stagger angle is
small enough, a constant stagger stator blade can probably be used. It is clear
that both vortex styles can yield a constant-camber stator blade if an appropriate
value of Rc is chosen. It can be seen that the precise value of Rc required varies
with φc. Indeed, it will also depend on Kc to some degree. Clearly, the prospects
for achieving a stator blade with both constant camber and constant stagger
angles are substantially better if the constant swirl vortex flow is employed.
To illustrate the advantages of constant swirl vortex flow, a stage was designed
specifically to achieve a constant-camber, constant-stagger stator blade. The
important stage performance data for this design are shown in Fig. 10-26. It will
be noticed that this design used a lower value of Kc than has been used in previ-
ous examples. This was necessary to maintain a reasonable velocity ratio across
the stator at the hub, which is typically the weakest area of a constant swirl or
exponential vortex design. The reduced work input capability is not as pro-
nounced as it may first appear. Indeed, Fig. 10-21 shows that, for similar reasons,
a similar reduction in Kc would have been beneficial for the free vortex design as
well. In comparison to the free vortex and constant reaction vortex designs in Fig.
10-21, it is seen that lower values of rotor-inlet relative velocity have been
achieved. The basic limitations for this vortex style are associated with low values
of R and C3 / C2 near the hub. Figure 10-27 shows the rotor and stator camber

FIGURE 10-24 Exponential Vortex Stators
FIGURE 10-25 Constant Swirl Vortex Stators

FIGURE 10-26 Constant Swirl Vortex Design Data
FIGURE 10-27 Constant Swirl Vortex Blade Angles

and stagger angles obtained using the blade design procedures of Section 10.3. In
this case, NACA 65-series blades were designed with tb / c = 0.1 and σ = 1 at r = rc.
The camber angles in Fig. 10-27 are the equivalent circular-arc camber angles
defined in chapter 4. Note that the result is basically a constant-camber, con-
stant-stagger stator blade. This is usually not too difficult to achieve. The con-
stant swirl vortex style yields flow angles that essentially produce a linear
variation of camber and stagger angle from hub to shroud. For any basic design
parameters, constant camber is achievable by selecting the proper value of Rc, as
illustrated in Fig. 10-25. By adjusting the other dimensionless performance
parameters, the variation in stagger angle can usually be reduced to a point
where constant stagger angles can also be used without significantly compromis-
ing performance. This would be a very cost-effective design for use as a standard
repeating stage for industrial axial-flow compressors.
10.9 ASSIGNED FLOW ANGLE VORTEX FLOWS
Occasionally, it is useful to assign the absolute flow angle distribution at the rotor
inlet or discharge rather than the swirl velocity distribution. Noting that Cθ = Cz
tanβ, it is easily shown that Eq. (10-24) can be replaced by
1 ∂Cz cos2 β ∂( r tan β )2 cos2 β ∂H

=− + (10-53)
Cz ∂r 2r 2 ∂r Cz2 ∂r
The specification of φc, ψc and Rc defines β1c and β2c. If it is assumed that β varies
linearly with r, and the difference between hub and shroud, βs – βh, is specified at
either the rotor inlet or discharge station, Eq. (10-53) can be solved for the corre-
sponding distribution of Cz. Since Cθ = Cz tanβ, the swirl velocity distribution is
also defined. For constant work stages, Eq. (10-21) defines the swirl distribution
at the other station. This vortex specification is a useful variant on the constant
swirl vortex type discussed in the previous section. Sometimes it is easier to fine-
tune the blade angle distributions by refining the flow angle distribution rather
than the swirl velocity distribution. Usually, it is more effective to specify the
rotor exit flow angle distribution for this purpose. This vortex type can be quite
useful, and it is easily incorporated into a computerized stage design system.
10.10 APPLICATION TO A PRACTICAL STAGE DESIGN
The stage design examples provided in the previous sections of this chapter have
been rather arbitrary in illustrating some of the features of the various types of
vortex flow styles. In practice, stage design is far more focused on specific
design objectives and constraints. Usually there is substantial conflict between
various desired design objectives in terms of establishing the specific stage
design parameters.
In almost any axial-flow compressor design, minimum size, weight and cost
will be important objectives. Clearly, these goals are best served by achieving a

large mass flow per unit frontal area, thus favoring a high flow coefficient stage.
Similarly, reducing the number of stages required will be important, thus favor-
ing a high work input coefficient design. As seen in Figs. 10-3 through 10-5, a
high work input coefficient requires a high flow coefficient to avoid exceeding
practical blade loading limits. So size, weight and cost considerations are definite
incentives for the high flow coefficient and work input coefficient stages.
Good aerodynamic performance will also be an important objective for any
axial-flow compressor design. Usually this is expressed in terms of efficiency
and surge margin requirements. Surge margin may be expressed in terms of
flow range or pressure-rise range between the design point and the surge limit.
Surge margin and efficiency are not really independent concepts, although spe-
cific applications may impose special surge margin requirements. Normally, the
stage design point is expected to be approximately its best operating condition.
It is quite obvious that there is little merit in achieving optimum performance
too close to the surge limit, where the stage will almost never operate in an
actual compressor. In Section 10.2 it was shown that improved surge margin can
be expected to favor low flow coefficients and low work input coefficients. In
Figs. 10-7 through 10-9, Smith’s (1958) recovery ratio parameter has strongly
suggested that flow coefficients around 0.5 should be preferred. Smith (1958)
seems to favor flow coefficient not exceeding 0.5 at any radius, although he
notes that it may be difficult to achieve. Indeed, that seems a little impractical
and perhaps not even desirable. Depending on the hub-to-shroud radius ratio,
achieving a hub flow coefficient of 0.5 may require very low shroud flow coeffi-
cients. That may be as harmful as using flow coefficients that are too high, as
can be seen in Fig. 10-7. Figs. 10-7 through 10-9 also show that low work coeffi-
cient (or K) is desirable and has the effect of reducing the significance of the
flow coefficient selection.
The role of stage efficiency in setting stage design parameters is less signifi-
cant than whether the surge margin is adequate to make use of the efficiency
achieved. Indeed, if maximum efficiency is used as the major design criterion,
the design point is very likely to be very close to the surge limit. This follows
from the fact that designing for high work input may be the best way to reduce
the impact of the losses on efficiency. Excellent stage efficiencies have been
achieved for a wide range of stage performance parameters and vortex types.
There is little credible evidence to associate any set of performance parameters
or vortex type to optimum and practical design-point efficiency levels. It is some-
times suggested that optimum efficiency is associated with 50% reaction
designs, although that is of questionable validity and of very little practical value.
The stage reaction, in terms of Rc, will usually be established by surge margin
considerations. Specifically, achieving satisfactory hub reaction and velocity
ratios across the blades at the hub and shroud will usually be the major factors
influencing the selection of Rc. Design approaches directed specifically toward
increasing the stage design-point efficiency are generally beyond the scope of the
methods described in this chapter. For example, special blade designs such as
the controlled diffusion airfoils mentioned in Chapter 4, may be used. Some
designers favor special features in the end-wall boundary layer regions, such as
increased end-wall work. With modern viscous computational fluid dynamics
codes, special designs to minimize secondary flow effects may be considered.

Clearly the designer is faced with conflicting priorities to trade off perform-
ance against size, weight and cost. Usually a parametric study will be required to
select the best compromise, including evaluation of performance with the meth-
ods of Chapter 9. Unless carefully structured, such a parametric study can be
very confusing and time-consuming. Usually the most efficient approach is to
investigate alternate values of φc, while specifying the minimum acceptable val-
ues of the blade row velocity ratios, W2/W1 and C3/C2. Generally, the designer will
be able to select an acceptable range of values for φc, while the minimum velocity
ratios ensure reasonable blade loading and surge margin capability. Other con-
straints will be specified and monitored. For example, reasonable limits on flow
angle, flow coefficient and work coefficient at all radii will normally be required.
To illustrate the process, a simple optimization study has been conducted for the
following constraints:
• φc = 0.5
• W2 / W1 and C3 / C2 ≥ 0.73
• β and |β′| < 70°
• ψ<1
• φ≤1
Designs were generated for constant swirl vortex, free vortex and constant reac-
tion vortex stages. The free vortex stage was not required to have Cθ1 = 0 for this
design. The design radius was chosen as the true mean stream surface, i.e., the
root-mean-square average of the hub-and-shroud radii was used. The only
parameters available to be selected are Rc and Kc (or ψc), which were chosen to
best satisfy the constraints. In all cases, the velocity ratio across the stator proved
to be controlling constraint, although the flow angle constraint was a significant
factor for the constant reaction vortex. Figures 10-28 through 10-30 show the
most important performance parameters to compare the three vortex types. It is
seen that the free vortex design resulted in the highest rotor inlet relative veloci-
ties, making it the most susceptible to Mach number effects. The constant reac-
tion vortex stage would appear to be the best choice for high Mach number
applications, but that is a little misleading. That design is very close to the flow
angle constraint near the shroud. It would be difficult to achieve adequate blade
passage throat widths while using practical blade thicknesses, so blade passage
choke could easily be a problem near the shroud. The constant reaction stage has
the largest flow coefficient at the hub, but it is within the limit set for these sam-
ple designs. In other respects, the performance parameter distributions for the
three designs are reasonably similar. Except for the rather large flow angles pro-
duced by the constant reaction vortex stage, the basic dimensionless perform-
ance and velocity triangle data provide little reason for preference of one design
over the others.
The more significant differences in these three designs are seen in Figs. 10-31
and 10-32, which compare the blade geometry required for the three stage
designs. In all cases, NACA 65-series blades were designed with constant chord,
constant tb / c = 0.1 and σ = 1 at r = rc. All rotor blades are variable camber,
twisted blades, but the constant reaction vortex stage requires the most extreme
variations. Indeed, a negative camber angle is required at the shroud for that

FIGURE 10-28 Constant Swirl Vortex Stage
FIGURE 10-29 Free Vortex Stage

FIGURE 10-30 Constant Reaction Vortex Stage
FIGURE 10-31 Rotor Blade Geometry

FIGURE 10-32 Stator Blade Geometry
design, with a very high stagger angle at the same location. The constant swirl
vortex design results in very simple stator geometry, which is easily approximated
by using constant camber and stagger angle blades. The free vortex stage requires
variable camber, twisted stator blades, but with very reasonable variations. The
constant reaction vortex stage requires highly twisted stator blades with a large
variation in the camber angle. The constant reaction stage blades would need
careful evaluation relative to its structural integrity and practicality for manufac-
turing. It is likely that application of this vortex style should be limited to higher
hub-to-shroud radius ratios than those in this case
The performance analysis described in Chapter 9 was applied to these three
stage designs to evaluate their expected performance. The annulus geometry for
all stages was obtained by assuming a constant hub radius and sizing the shroud
radius based on the same volume flow rate and rotation speed. This design vol-
ume flow rate is easily computed from the rotation speed and the free vortex
velocity triangles, where Cz is constant. The rotation speed and inlet thermody-
namic conditions were chosen to limit the maximum rotor inlet relative Mach
number to about 0.78 to avoid biasing the comparison with Mach number
effects. Once the approximate annulus sizing was accomplished, the shroud
radii were adjusted to produce a conical shroud wall to avoid the results being
biased by local contour angle variations that are not representative of a stage’s
true performance when used as a repeating stage in a multistage compressor.
Similarly, end-wall boundary layer blockage and stream surface curvature

effects were ignored to obtain as fair a comparison as possible. This process pro-
duced the expected repeating stage for the free vortex and constant swirl vortex
stage designs. That was not the case for the constant reaction vortex design,
where a repeating stage configuration could not be achieved in practice. The
annulus sizing to match the design velocity triangle for that case produced an
increase in shroud radius across the rotor, followed by a large decrease across
the stator. The most reasonable approximation to the annulus area schedule
required was to simply assign a constant shroud radius. The source of the prob-
lem is the axial velocity ratios across the blade rows, which alter the predicted
deviation angles as discussed in Chapter 6. As seen in Fig. 10-17, axial velocity
ratios can be quite significant for this vortex style. The computerized stage
design system used to generate these designs does not correct deviation angles
for axial velocity ratio effects. To do so is usually misleading, since axial velocity
ratios in an actual application may be quite different from those produced in the
ideal stage design calculations. Application of a standard industrial compressor
repeating stage involves use of various portions of the blade span in different
stages as required to conserve mass flow rate. Adjustments in stagger angles and
axial velocity levels through the multistage compressor are usually necessary to
produce the required performance with a specific number of stages. If it is not
possible to neglect the effect of axial velocity ratio on deviation angle in the
basic stage design, the stage is simply not a good candidate for a standard
repeating stage. That is really the case for this constant reaction vortex stage.
The hub-to-shroud ratio used in this example is simply too low for a constant
reaction vortex to produce a viable repeating stage. Of course, it is possible that
this stage might be used in a single, specific application. In that case, the appli-
cation-specific design procedures discussed in the next chapter should be used
to design the blades for the stage.
Figure 10-33 presents a comparison of the performance predictions for the
three stage designs. The lower limit of volume flow rate for each stage corre-
sponds to the predicted stall limit based on stall criterion #3 of Chapter 9. It can
be seen that the free vortex and constant swirl vortex stages produce essentially
the same performance characteristics. The free vortex stage has a slight advan-
tage in stable operating range, while the constant swirl vortex stage has a slight
advantage in efficiency. Neither of these differences can be considered significant
within the uncertainty of the performance analysis procedures. The constant
reaction vortex design produces lower pressure ratios and efficiencies than the
other two stages, even though it had the highest design work input coefficient, ψc.
Based on the expected stage performance, either the free vortex or constant swirl
vortex design could be selected. The simpler stator geometry of the constant swirl
vortex stage and the higher rotor inlet relative Mach number level of the free vor-
tex design would favor selection of the constant swirl vortex design.
10.11 A REPEATING STAGE AXIAL-FLOW COMPRESSOR
As mentioned at the beginning of this chapter, a repeating stage design can be used
as a standard stage in industrial axial-flow compressors. Each industrial axial-flow
compressor usually has a unique design. Pattern, tooling and development costs

FIGURE 10-33 Comparison of Predicted Performance
are critical issues, since there are seldom any duplicate machines to share those
costs. Similarly, experimental confirmation of the performance of the finished
compressor will be obtained before it is shipped to the customer. Failure to achieve
the guaranteed performance may require very expensive rework of the compressor.
Use of a standard repeating stage can alleviate these problems. Now pattern and
tooling costs can be shared among many industrial compressors, even though each
one has a unique design. The risk of a performance deficiency is also greatly
reduced, since experience with previous compressors can be used for guidance.
Application of a standard stage in a multistage compressor is reasonably
straightforward using an aerodynamic performance analysis that is capable of siz-
ing the annulus as described in Chapter 7. To illustrate the process, a ten-stage
axial-flow compressor will be configured using the constant swirl vortex repeating
stage designed in the previous section. The compressor will be designed with a
constant hub radius, so that the same rotor blade can be used throughout, by sim-
ply trimming the blades to the proper shroud contour. Since constant-camber,
constant-stagger stator blades can be used for the basic stage, a common stator
blade is also usable. Under these constraints, it is easily seen that the basic blade
geometry is identical for all stages. The simplest way to configure the basic com-
pressor is to assign the mean stream surface meridional velocity through the com-
pressor and size the annulus to define the approximate shroud contour. To
simplify this process, the performance analysis is set up with one computing sta-
tion between successive blade rows. Then the meridional velocity on the mean
stream surface is assigned at specific computing stations and obtained at other

stations by linear interpolation with the station number. For this example, it will
be assumed that the assigned meridional velocity must vary linearly with station
number from the entrance to the first rotor to the exit of the last rotor. The com-
pressor will be sized to use the full flow capacity of the standard stage, at the same
rotation speed as that used in Fig. 10-33. So the meridional velocity at the
entrance to the first rotor can be assigned immediately from the stage design data.
Some fine-tuning is required to match the known shroud radius at this point due
to the effect of gas density variations across the inlet guide vane and to the end-
wall boundary layer blockage assumed at the compressor inlet. This requires a
minor adjustment to the design inlet mass flow until the correct shroud radius is
obtained. The sizing of the annulus for the entrance to the inlet guide vane and the
exit to the last stator and the exit guide vane can be rather arbitrary at this point.
As long as the meridional velocities assigned at these stations are not unreason-
able, they have little influence on the performance. Manual sizing of the annulus
at those stations is rather trivial once the shroud contour is defined at the other
stations. So the task to be accomplished in this simple example is selection of the
meridional velocity to be assigned at the last rotor exit. The first priority is to
obtain a reasonable shroud contour. For example, assigning the meridional veloc-
ity on the mean stream surface to be constant through the compressor resulted in
the shroud radius increasing with axial distance in the rear stages. This could be
anticipated since the mean stream surface radius decreases through the machine.
Since the axial velocity for the constant swirl vortex increases as the radius is
reduced, it could be expected that the assigned meridional velocity should
increase through the compressor. The assigned meridional velocity to be used at
the last rotor discharge was easily determined by trial and error, using the per-
formance analysis. Figure 10-34 shows the Cm distribution used and Figure 10-35
shows the hub-and-shroud contours developed by the annulus sizing with that Cm
distribution. Figure 10-36 shows the predicted performance using the methods of
Chapter 9. The estimated surge limit is also shown, including the stall criterion
from Chapter 9 that was first encountered for each speed line. It can be seen that
a standard repeating stage designed by the methods of this chapter can be effec-
tively employed in a multistage compressor.
This application of a standard stage in a multistage compressor illustrates the
conflicting priorities that the designer regularly encounters. The stage design in
Section 10.10 followed practices expected to favor good efficiency and stable
operating range. This was reasonably successful, but the cost-effectiveness of a
ten-stage compressor that achieves a design pressure-ratio of about 2.2 is very
questionable. The design constraints selected in Section 10.10 do not permit
increasing the work coefficient to increase the pressure ratio. The compressor
might be operated at a higher rotation speed to increase the pressure ratio, but
only by exceeding the limit imposed on the rotor tip relative Mach number. It
might be concluded that a standard repeating stage design is simply impractical
for these design constraints. But that is really not the case. Rather, the stage hub-
to-shroud radius ratio of 0.5 used throughout this chapter is not a good choice
for an industrial axial-flow compressor using this as a standard repeating stage.
Figure 10-37 shows the important performance parameters for a constant-
swirl vortex stage designed to the same constraints as for the design in Fig. 10-28,
but with a hub-to-shroud radius ratio of 0.7. Note that a substantial increase in

FIGURE 10-34 The Assigned Cm Distribution
FIGURE 10-35 Hub-and-Shroud Contours

FIGURE 10-36 Aerodynamic Performance Prediction
FIGURE 10-37 Alternate Constant-Swirl Vortex Stage

work coefficient is achieved while satisfying the same design constraints. An

alternate 10-stage axial-flow compressor was configured using this alternate
stage design as a standard repeating stage. This results in the predicted perform-
ance map shown in Fig. 10-38. This compressor achieves a pressure ratio of
about 3.9 at the design mass flow rate. The larger hub-to-shroud radius ratio also
results in higher efficiency for both the stage and the ten-stage compressor. Of
course, the mass flow rate achieved is reduced relative to Fig. 10-36 due to the
larger hub-to-shroud radius ratio. But scaling the geometry to a larger hub radius
and reducing the rotation speed to obtain the desired mass flow rate, while main-
taining the same dimensionless performance parameters and Mach number lev-
els, could easily correct that.
In practice, application of a standard stage is more complicated than this sim-
ple illustration. It is unlikely that a direct application of the repeating stage will
produce the design point flow and pressure ratio required. For example, it might
be concluded that the application requires a 91/2-stage compressor with that con-
figuration. In that case, it will be necessary to modify the design to achieve the
required performance in either a 9-stage or a 10-stage compressor. This may
require adjusting some of the blade stagger angles and modifying the Cm distribu-
tion. The stagger angle adjustment modifies the velocity triangles, while the Cm
distribution can be used to expand or contract them. Both approaches can change
the work input per stage to refine the overall pressure ratio to a more appropriate
value. Both approaches will probably be needed to modify the performance while
maintaining a good match between stages. After some experience with a specific
FIGURE 10-38 Alternate Compressor Performance

standard stage, the designer usually develops a fairly systematic approach to

achieving the desired modifications. The same methods can be used to improve
the surge margin, to better optimize the efficiency at a specific operating point of
special interest, or even to achieve a flow capacity somewhat greater than the
value for which the stage was designed. Hence, there is considerable flexibility
when applying a standard stage, even though it was designed to be a simple
repeating stage.
10.12 A COMPUTERIZED STAGE DESIGN SYSTEM
It is relatively simple to computerize the procedures described in this chapter into

a systematic interactive design system. The most obvious use of this type of sys-
tem is the design of standard industrial compressor stages such as those illus-
trated in the previous section. For that application, it is logical to ignore Mach
number effects and losses, since they will vary with the specific application and
the location of the stage in the multistage compressor. In Chapter 11, a more gen-
eral approach is described where blades are designed specifically for each stage in
a multistage compressor, including the effects of Mach number and losses. A com-
puterized stage design system based on the methods from this chapter can be very
useful to support that approach as well. It can provide a convenient method to
investigate the alternate stage performance parameters and vortex types to be
employed in the more general design approach. It can be rather cumbersome and
confusing to attempt those investigations in the context of a multistage compres-
sor design. Another application of a stage design system is for field-service repairs
where a new stage design is required. This can occur if a stage is too damaged to
determine its original geometry through reverse engineering. When a stage design
system is available, it can often be a useful tool for a variety of simple and fast
investigations to support other design activity. The effort required to develop a
computerized system is so modest that it can be easily justified. It is advisable to
include some additional flexibility, which is not covered in this chapter, when cre-
ating the computerized system. In particular, it is quite simple to provide for a
variation of Czc through the stage. Similarly, provision for a radial variation in
work input is easily included for cases where additional end-wall work or imposed
radial energy gradients are considered desirable.
EXERCISES
10.1 Show that Eq. (10-45) is valid for any constant-work, repeating stage.
10.2 Show that Eq. (10-48) is the condition to avoid reverse flow at the
rotor exit for a constant reaction vortex flow in a constant-work,
repeating stage.
10.3 Show that a constant-work, repeating stage can have constant reac-
tion only for 100% reaction.

Chapter 11
MULTISTAGE
AXIAL-FLOW COMPRESSOR
AERODYNAMIC DESIGN
The basic concepts introduced in chapter 10 to design an ideal axial-flow com-

pressor stage are easily extended to the design of a complete axial-flow com-
pressor. When the specific working fluid, rotation speed, mass flow rate and
inlet thermodynamic conditions are known, the design process can include the
influence of total pressure loss and Mach number levels. This is possible as long
as the design is restricted to blade operating conditions reasonably close to the
design incidence angles, i*, such that the blades operate in the low-loss inci-
dence angle range where loss coefficients are reasonably constant. This is nor-
mal practice at the compressor’s design point, i.e., for the mass flow rate and
speed condition at which the stages are to be well matched throughout the
compressor. An axial-flow compressor design system of this type is unlikely to
produce all of the aerodynamic and structural design features desired and all of
the off-design operating characteristics required. But it can produce an excel-
lent preliminary configuration to be fine-tuned using the performance analysis
of Chapter 9.
The same procedures can be used to design individual stages or blade rows
intended for specific applications. For example, it might be used to design
replacements for blade rows that are too badly damaged to be reverse-engi-
neered. When this type of design system is available, the benefits of the ideal
stage design system of Chapter 10 may seem questionable. Indeed, if sufficient
care is taken, this more rigorous design approach may be used in place of the
ideal stage design system to generate a standard industrial compressor stage
design. It is necessary to avoid imposing restrictions that may preclude using the
standard stage for intended applications different from the design conditions
used. The procedures defined in this chapter size the compressor annulus to
match the specific operating conditions. This imposes a specific stream surface
pattern and limits the span of the blades designed to whatever is required to con-
serve mass. The influence of end-wall boundary layers is also significantly
affected by the specific design conditions. If this system is applied at the lowest
Mach number level expected and at a higher volume flow rate than is expected to
be required, a stage design may be sufficient for general use as a standard stage.

But that typically reduces the design procedure to a more complicated version of
the methods described in Chapter 10, effectively nullifying the benefits of the
more rigorous models used. Indeed, the design methods of Chapter 10 are often
useful for exploring alternative stage performance parameters before attempting
a design using the methods of this chapter.
NOMENCLATURE
c = chord
Fg = fractional axial gap between blades
Fsh = shroud-to-hub chord ratio
g = axial gap between blade rows
H = total enthalpy
h = blade height
i = incidence angle
i* = minimum loss incidence angle
m = vortex exponent
n = vortex exponent
PR = pressure ratio
R = reaction
r = radius
U = blade speed = ωr
WRE = equivalent velocity ratio
γ = stagger angle
η = efficiency
θ = camber angle
σ = solidity
φ = flow coefficient and stream surface slope angle
ψ = work coefficient
Subscripts
c = parameter on the mean stream surface
h = parameter on hub contour
i = stage number
s = parameter on shroud contour
1 = rotor inlet condition
2 = rotor exit condition
3 = stator exit condition

Multistage Axial-Flow Compressor Aerodynamic Design • 261
11.1 THE BASIC COMPRESSOR DESIGN APPROACH
Figure 11-1 illustrates the basic compressor design problem to be solved. The
geometry of the end-wall contours and of all blade rows is to be generated, based
on specified swirl vortex types and dimensionless performance parameters on
the mean stream surface for all stages. As usual, the mean stream surface is
defined by the requirement that equal mass flow rates are achieved on both sides
of it. To accomplish the design, one of the end-wall contours is defined, while the
other is computed from the annulus sizing procedure of Chapter 7. The case
illustrated in Fig. 11-1 is typical of an assigned shroud contour design. Since all
operating conditions are known, the hub contour can be computed as part of the
design solution from simple conservation of mass.
The design procedure to be used is a combination of the stage design method
presented in Chapter 10 and a simplified version of the performance analysis of
Chapter 9. The performance analysis is simplified by assuming that the loss coef-
ficients are given by the design loss coefficient models of Chapter 6. This is a rea-
sonable approximation as long as the incidence angle is not too different from the
design incidence angle. Chapter 6 shows that design loss coefficient is primarily a
function of the aerodynamic data. Indeed, from Eqs. (6-35) through (6-37), it is
seen that the only geometrical parameter required is the solidity. The flow field
throughout the compressor will be generated using specified swirl vortex types
and dimensionless stage performance parameters, similar to the approach used in
Chapter 10. The simple normal equilibrium equation, Eq. (10-24), is replaced by
the meridional through-flow analysis of Chapter 7, using the annulus sizing
FIGURE 11-1 Compressor Geometry

option. By considering the calculation of solidity as part of the design calcula-

tions, the influence of total pressure loss can be included in the flow field calcula-
tions. End-wall boundary layer blockage effects are included using the boundary
layer analysis of Chapter 8. As long as the design loss coefficient models are valid,
this provides a rather accurate and complete prediction of the flow field through-
out the compressor. The major approximation required is that streamline curva-
ture effects must be neglected to size the annulus. But once the annulus has been
sized, the design process can be repeated using the fixed end-wall geometry mode
of the meridional through-flow analysis to allow curvature effects to be included.
The development of this type of compressor aerodynamic design system is
much simpler than it may appear. Nearly all of the computational methods
required have been presented in previous chapters of this book. The design sys-
tem used to generate results presented later in this chapter is a good example. It
was rather easily formulated directly from the aerodynamic performance analy-
sis. All that was needed was simplification of the blade performance models and
revision of the meridional through-flow logic to include the stage performance
specifications and airfoil design logic of Chapter 10. Indeed, nearly all of the
computer codes needed for this design system already existed in the aerodynamic
performance analysis and the ideal stage aerodynamic design system.
11.2 AERODYNAMIC PERFORMANCE SPECIFICATIONS
The aerodynamic performance specifications used are similar to those employed

in Chapter 10 for a stage design. For each stage in the compressor, the flow coeffi-
cient, work coefficient and reaction on the mean stream surface will be supplied.
Parameters on the mean stream surface will be designated by a subscript c, and
subscripts 1 through 3 will designate successive stations in the stage from rotor
inlet to stator exit. Then the performance data for stage number i are given as
φ ci = (Cmc1 / Uc1)i (11-1)

ψ ci = ( Hc2 − Hc1) / Uc21 = [( rc2Cθc2 / rc1 − Cθc1) / Uc1 (11-2)
Rci = 1 − (Cθc1 + Cθc2rc2 / rc1) / (2Uc1) (11-3)
The definitions of flow coefficient and work coefficient are direct generalizations
of the definitions used in Chapter 10. The definition of reaction is chosen for con-
venience. It will be the true stage reaction only in the case of a repeating stage
with a constant-radius mean stream surface. The vortex exponents, ni and mi, are
also specified for each stage. Then the rotor inlet swirl velocity distribution for
any stage is given by
ni mi
Cθ1 = Uc1[(1 − Rci )( rc1 / r1) − (ψ ci / 2)( rc1 / r1) ] (11-4)
The swirl velocity distribution at the rotor exit or stator inlet results from the
assumption that the work input is constant from hub to shroud. From Eq. (11-2)
this requires

Cθ 2 = [r1Cθ1 + ψ ciUc1rc1] / r2 (11-5)
If there is another stage following this one, the stator exit swirl velocity distribu-
tion is given by Eq. (11-4), using performance parameters for the next stage, i.e.,
n i+1 m i+1
Cθ 3 = Uc3 [(1 − Rc,i +1)( rc3 / r3 ) − (ψ c,i +1 / 2)( rc3 / r3 ) ] (11-6)
For the last stator, a special specification is required. A convenient choice is to

specify the ratio of the exit to inlet swirl velocity as a constant for the last stator, i.e.,
Cθ 3 / Cθ 2 = constant (11-7)
The meridional velocity on the mean stream surface must also be specified as the
constant of integration for the normal momentum equation. The specifications
used are
Cmc1 = Uc1φ ci (11-8)

Cmc3 = Uc3φ c,i +1 (11-9)
Cmc2 = (Cmc1 + Cmc3 ) / 2 (11-10)
If an inlet guide vane is used, its discharge conditions are identical to the inlet
conditions for the first rotor. Similarly, the inlet conditions for an exit guide vane
are identical to the discharge conditions for the last stator. At the stations before
the inlet guide vane and after the exit guide vane, the swirl velocity is set to zero
and it is assumed that the meridional velocity is constant across the adjacent
blade row. If additional computing stations are inserted between blade rows, con-
servation of angular momentum supplies the required swirl velocity distribution
from the adjacent station where data has been assigned, and the meridional
velocity is set to the same value as that adjacent station. A useful variant on the
above equations is to replace Uc and rc in Eqs. (11-4) through (11-9) by their val-
ues at the entrance to the first rotor. This often makes it easier to select appropri-
ate values for the performance parameters by removing the influence of the
unknown variation of the mean stream surface radius through the compressor.
This is easily included as an option in the computerized design system.
Hence for any assumed annulus and stream sheet geometry, the normal
momentum equation, Eq. (7-12), can be integrated if the entropy distributions
are known. As discussed in Chapter 7, Section 7.6, the stream surface curvature
terms are neglected when sizing the annulus. Thus the compressor design solu-
tion is a simple marching process where the solution at each axial computing
station is completed before proceeding to the next station. The entropy distri-
butions are computed as part of the solution by computing the design total loss
coefficients. As mentioned, this requires knowledge of the blade row solidity,
which will be discussed in the next section. Then the annulus area is adjusted
as required to balance the known mass flow rate. During this process, the
stream sheet pattern is continually changing. Equations (11-4) through (11-10)

are continually applied to update the performance data as the stream sheet
geometry changes.
11.3 BLADE DESIGN
The basic process used to design the blades has been described in Chapter 10,
Section 10.3. Some minor variations on that process are more appropriate for the
present design problem. The following assumptions and specifications have been
found to be effective and relatively simple to employ:
• The basic airfoil camberline and profile types from Chapter 4 are
specified.
• If required, specify the location of the point of maximum camber, a / c.
• Specify the ratio of the chord at the shroud to its value at the hub and
require a linear variation from hub to shroud.
• Specify the thickness-to-chord ratio on the hub and shroud and require
a linear variation from hub to shroud.
• Specify the solidity on the mean stream surface.
• Specify the fraction of the axial spacing between computing stations,
Fg, that is reserved for the axial gap, g, between blade rows.
• Specify the difference between the actual incidence angle and the
design incidence angle, (i – i*), on the hub and shroud and require a lin-
ear variation from hub to shroud.
• Specify the blade tip clearance as a fraction of the shroud radius of the
first rotor row.
The blade design procedure is essentially the same as that described in Chap-
ter 10, Section 10.3. The major difference is that the blades must be inserted
within a predefined axial distance, since coordinates for one of the end-walls are
specified. Here it is necessary to select the number of blades in the blade row and
the chords such that the desired axial gap is maintained between blade rows. As
illustrated in Figure 11-2, the chord at the hub, which provides the specified axial
gap, is
ch = ∆zh (1 − Fg ) / cos γ h (11-11)
A similar calculation on the shroud contour yields
cs = ∆zs (1 − Fg ) / cos γ s (11-12)
But these two values are also constrained by the specified shroud to hub ratio,
Fsh = cs / ch. This can be used to determine which of the previous two equations
specifies a chord necessary to achieve the axial gap. Then the other chord is cal-
culated from Fsh. Once the correct chord at the hub is known, the chord at any
radius is given by

FIGURE 11-2 The Blade Row Axial Gap
c = ch [1 + ( r − rh )( Fsh − 1) / ( rs − rh )] (11-13)
If Z is the number of blades in the blade row, the solidity is given by
σ = [ch Z / (2πr )][1 + ( r − rh )( Fsh − 1) / ( rs − rh )] (11-14)
Solving Eq. (11-14) on the mean stream surface, and normalizing it by the result,
yields
r [1 + ( r − rh )( Fsh − 1) / ( rs − rh )]
σ = σc (11-15)
r[1 + ( r − rh )( Fsh − 1) / ( rs − rh )]
Hence the solidity at any radius can be computed from the specified solidity at the
mean stream surface before the blades are designed. Since that is the only blade
geometrical parameter needed to compute the design loss coefficient, the design
flow field analysis can be conducted prior to designing the blades. Then the pre-
cise chords and number of blades can be selected using Eqs. (11-11) through
(11-14) and the known solidity at the mean stream surface. The largest chords and
lowest number of blades permitted by the specified axial gap are selected.
If tip clearance losses are included in the analysis, there is a weak dependence
on chord and stagger angle, as can be seen from Eq. (6-86). The end-wall bound-
ary layer analysis also depends on the stagger angle, as seen from Eqs. (8-44) and
(8-55). Hence, initial guesses are made for the number of blades and the chord on

the mean stream surface that will yield the desired solidity. The stagger angles on
the end-wall contours are approximated from the local flow angles until the first
blade design is accomplished. The flow analysis and blade design is repeated
until convergence on the chord (or number of blades) is achieved. One repetition
is almost always sufficient. At least one repetition is necessary to adequately
account for the effect of stagger angle on the end-wall boundary layers.
Another minor difference relative to the procedure of Chapter 10 is that
stream surfaces are normally not cylindrical surfaces. The blade geometry used
in the flow field analysis needs to be adjusted for streamline slope, using the
angle, φ, illustrated in Fig. 11-2 for the hub contour. The required corrections
have been presented in Eqs. (9-1) through (9-6).
11.4 REFINING THE COMPRESSOR DESIGN
In general, an axial-flow compressor design accomplished by the procedures out-

lined above will prove a little disappointing when analyzed in the performance
analysis of Chapter 9. Even when directly derived from the performance analysis,
this design procedure does not normally result in a design that is simply confirmed
by it. Since the performance analysis works with assigned end-wall contours, it can
directly estimate stream surface slopes and curvatures. In contrast, the design pro-
cedure outlined above has only a crude estimate of the sized contour’s slope and
ignores curvature effects entirely. There are several simple steps that can be incor-
porated into the design procedure to minimize the differences seen when a per-
formance analysis is conducted on the resulting design.
First, it is very simple to include the capability to repeat the design using the
sized contour on the previous attempt as the basis for contour slope calculation.
This can be repeated until the contour slopes no longer show a significant change
between successive attempts. When that has been achieved, it is very useful to be
able to redesign the compressor with the resulting sized contour used directly,
instead of sizing the annulus. This will also permit inclusion of stream surface
curvature effects, while still generating a flow field consistent with the desired
performance parameters and blade geometry consistent with that flow field. It is
also useful to constrain the annulus sizing procedure to prohibit an increase in the
annulus area in the streamwise direction. When a passage area increase is
required, the annulus area can be set to the value at the upstream station and con-
servation of mass is used to override the local meridional velocity specification.
Resolving inconsistencies with the performance analysis is usually not totally
sufficient. When an end-wall contour is computed by the annulus sizing proce-
dures, the resulting contour is likely to be unrealistic, often resulting in erratic
variations in coordinates, slopes and curvatures. Refining the dimensionless per-
formance parameters can reduce this erratic behavior. But that is a rather
tedious process, and involves changes in the performance parameters that are
really physically insignificant. It is much more effective to include procedures to
permit smoothing and editing the contour coordinates. Then the design process
can again be repeated with the revised contour to produce the desired perform-
ance. A simple smoothing procedure that has been found effective is illustrated in
Fig. 11-3. In this case, the shroud has been sized with somewhat erratic results.

FIGURE 11-3 Contour Smoothing
The approach used is performance of a simple linear interpolation between adja-

cent axial computing stations to determine an alternate radius on the quasi-nor-
mal of interest. Then a smoothed point is defined as an average of the actual and
interpolated points. The smoothed points can be used directly for all interior
quasi-normals. Alternatively, smoothing can be omitted at certain quasi-normals,
typically either at the rotor inlet or rotor discharge quasi-normals. That may be
desirable to preserve key elements of the sizing process. It is best to implement
this contour smoothing as an interactive process where the smoothed contour is
displayed graphically, with the original contour points in the background. Then
the designer can try alternate smoothing strategies to select the most promising
method. This can also include provision to directly edit some coordinates that do
not adequately respond to the smoothing procedures alone. That is fairly com-
mon for quasi-normals upstream of the first rotor and downstream of the last
stator. The same contour smoothing procedures can be very useful in the per-
formance analysis, particularly when the annulus sizing procedure is used.
Indeed, these smoothing procedures were used for the two standard repeating
stage compressor designs described in Section 10.11
If properly implemented, all of these procedures for refining the design can
be employed with very little effort on the part of the designer. If they are not
accomplished here, they will almost certainly need to be accomplished manually
in the performance analysis. That will involve much more effort, particularly
since the blade designs will require modification during the process and the

desired performance parameters will almost certainly be compromised. By

incorporating them in the compressor aerodynamic design system, performance
goals and blade design are handled automatically. This results in a dramatic
reduction in fine-tuning the design using the performance analysis.
11.5 AN AXIAL-FLOW COMPRESSOR DESIGN EXAMPLE
The application of this axial-flow compressor aerodynamic design procedure will

be illustrated by developing various alternate designs for the second ten-stage
compressor designed in Chapter 10 using a standard repeating stage. The proce-
dures in this chapter will be employed to design the end-wall contours and the
geometry of the 22 blade rows required. NACA 63-series inlet guide vanes will be
used while NACA 65-series blades will be used for all other blade rows. All
designs will limit the rotor tip relative Mach numbers to a maximum value of 0.8.
All compressor designs will be sized to specified performance parameters with a
constant hub-contour radius. The shroud contours will be smoothed and refined
as described in the previous section of this chapter.
The first example is designated as Compressor A. It is based on stage perform-
ance parameters similar to those achieved in Chapter 10 with the standard
repeating stage. Figure 11-4 shows the stage mean stream surface performance
parameters specified. A simple linear variation with stage number was used,
FIGURE 11-4 Stage Design Data for Compressor A

where data for the first and last stages were obtained from the standard-stage
compressor performance analysis shown in Fig. 10-38. Then the work coefficient
for the last stage was adjusted by trial and error to produce approximately con-
stant-camber, untwisted stator blades throughout the compressor. After sizing
the annulus, the shroud contour was smoothed twice, each time holding the
rotor inlet shroud radii constant. Figure 11-5 shows the end-wall contours
obtained from that process. Then the flow field was predicted for these end-wall
contours using the approximate normal equilibrium model described in Section
7.6 for the meridional through-flow analysis to obtain the final blade row
designs. All blades were designed to operate at their design incidence angles,
while matching with the predicted flow field. The computerized design system
used makes this a rather simple process. Once the initial design specifications
had been entered, the entire design process for Compressor A was completed in
about five minutes.
Then Compressor A was analyzed using the performance prediction methods
of Chapter 9. Consistent with the aerodynamic design approach, the approximate
normal equilibrium model described in Section 7.6 was used for the performance
analysis. The predicted difference between the incidence angles and the design
incidence angles on the hub-and-shroud contours is shown in Fig. 11-6. The per-
formance analysis supplies incidence angles rounded off to the nearest 0.1°.
Hence, it can be seen that the deviation from the intended incidence angle match
was less than 0.15° in all cases. Comparison of the calculated incidence angle
FIGURE 11-5 End-Wall Contours of Compressor A

FIGURE 11-6 Predicted Incidence Angle Match
match with the design specifications is the best way to validate the aerodynamic
design system. Even minor differences between the design system and the per-
formance analysis will produce significant differences in the incidence angles
that cascade to rather large values in the rear stages of a multistage compressor.
The results in Fig. 11-6 are fairly typical of what can be achieved by a properly
formulated aerodynamic design system.
Figure 11-7 shows a predicted performance map for Compressor A, along with
results from Fig. 10-38 for the standard-stage design; The two designs are quite
similar. The present design achieves a larger pressure ratio and improved surge
margin at design speed, but at the expense of a reduced maximum flow capacity.
Careful review of the standard stage compressor performance analysis showed
that the stages exhibit slightly positive values of (i – i*) in the front stages,
whereas the nearly ideal matching shown in Fig. 11-6 was achieved for Compres-
sor A. Hence the differences in performance are not unexpected nor are they par-
ticularly significant. A minor re-matching of the stages in either of these designs
could easily result in performance nearly identical to that of the other design.
Clearly, the substantial cost advantage offered by the standard repeating stage
compressor with its constant camber, untwisted stator blade is achieved with lit-
tle compromise. The nearly ideal stage matching with completely arbitrary blade
geometry used in the design of Compressor A does not result in any significant
performance improvement. In this specific example, the most significant benefit

FIGURE 11-7 Performance of Compressor A
obtained from the present aerodynamic design system is that the design of Com-
pressor A was much easier to accomplish than the design of the standard repeat-
ing-stage compressor.
It should not be concluded that this more general aerodynamic design proce-
dure does not offer the potential for improved performance over a standard
stage compressor design. Since the design of Compressor A is based on specifi-
cations obtained from a performance analysis of a repeating stage compressor,
some similarity to a repeating stage compressor is to be expected. Figure 11-8
shows the basic blade geometry on the hub contour, where the radius is con-
stant. It can be seen that the design does approximate a repeating stage configu-
ration. The geometry for the last stator has been omitted, since it is rather
arbitrary, depending directly on the specification of how removal of the swirl
from the last rotor is to be split between the last stator and the exit guide vane.
Although not obvious from Fig. 11-8, it is simply stated that a standard constant-
camber, untwisted blade could easily be substituted for the stator blades
designed for Compressor A, but with stagger angles varying through the
machine. However, the rotor blade geometry varies too much through the
machine to permit substitution of a standard rotor blade. Hence the design of
Compressor A was really constrained to approximate the repeating-stage design
of Chapter 10. Although the blade geometry was not constrained, this added
flexibility was not used to much advantage.

FIGURE 11-8 Hub Contour Blade Geometry
11.6 THE DISTRIBUTION OF STAGE PERFORMANCE

PARAMETERS
More effective use of the general aerodynamic design procedure is possible by
simply applying a distribution of stage performance parameters through the
compressor. The variation in stage performance parameters shown in Fig. 11-4
for Compressor A is somewhat misleading, since it really only corrects the stage
performance parameters for the variation in mean stream surface radius. A com-
pressor design establishes the annulus area distribution and blade geometry suit-
able for the design rotation speed. When the compressor is operated at speeds
less than the design speed, the work per stage is reduced, such that the increase
in gas density is no longer sufficient for the stages to remain matched. The rear
stages will operate at flow coefficients higher than their design values, with a cor-
responding reduction in work per stage. The front stages will have the highest
work per stage and will be the first stages to stall. The rear stages contribute
much less work and will limit the compressor’s flow capacity. When the compres-
sor is operated at speeds greater than design speed, the opposite trend will occur.
The front stages produce more work and a larger increase in gas density than
intended. Downstream stages tend to operate at flow coefficients less than their
design values, with a corresponding increase in work coefficient. Generally, the
rear stages have the highest work per stage and will be the first stages to stall. The
front stages will contribute less work input and will limit the compressor’s flow

capacity. In effect, the annulus area distribution is too small for operation at low
speeds and too large for operation at high speeds.
It follows that the stages least affected by variations in rotation speed are the
mid-stages. In a properly designed compressor, the mid-stages will normally
operate closer to their design operating conditions than either the front or rear
stages, and are less likely to be responsible for the compressor’s stability or flow
capacity limits. Hence it is reasonable to employ more highly loaded mid-stages,
with more conservative loading of the front and rear stages. Stage loading is not
a well-defined term. Some investigators use it to refer to the stage work coeffi-
cient; others use it to refer to the flow diffusion across the blade rows. In the pres-
ent context, both aspects must be considered. Higher work coefficients will be
used for the mid-stages. Somewhat greater flow diffusion may be attempted,
since the mid-stages are expected to operate over a narrower flow coefficient
range than either the front or rear stages. However, excessive flow diffusion must
still be avoided. The evaluation of flow diffusion used here will be the relative
velocity ratio across the blade rows, W2 / W1, or the effective velocity ratio, WRE,
defined in Eq. (9-16). The two parameters are approximately the same, but the
latter has been used as the basis of the blade stall criterion developed in Chapter
9. In Chapter 10, it was shown that the work coefficient can be increased without
increased flow diffusion if the flow coefficient is also increased. That will require
a departure from using Smith’s (1958) recommended flow coefficient of 0.5
based on the recovery ratio of Eq. (10-36). From Figs. 10-7 through 10-9, it is
apparent that large positive values of the recovery ratio are achieved for a fairly
wide range of flow coefficients. Also, the risk of lower recovery ratios should be
much less at the mid-stages, since they are expected to experience less severe
operating conditions than the front and rear stages.
The present aerodynamic design system makes it relatively simple to develop a
suitable distribution of stage performance parameters. To illustrate the process,
Compressor B was designed for the same design mass flow, rotation speed and
hub contour as that used for Compressor A. The stage performance parameters
used are shown in Fig. 11-9. The design system used for this example permits
specification of stage performance data at an arbitrary number of stages, with
linear interpolation for the other stages. In this case, specifications were pro-
vided for stages 1, 2, 3, 4, 8 and 10. The distribution of φc starts with a value of
0.55 for the first stage and ramps up through the front stages to permit an
increase in ψc without unacceptable diffusion levels. Initial distributions of Kc
and Rc were assumed. Kc was assigned instead of ψc to provide more direct con-
trol over the diffusion levels in the stages. The distributions of Kc and Rc were
adjusted by trial and error until acceptable diffusion levels were achieved
through the compressor. Kc was adjusted to achieve acceptable levels of WRE, and
Rc was adjusted to approximately balance the diffusion load between the rotor
and stator for all stages. Fig. 11-9 shows the key values of WRE obtained on the
hub-and-shroud contours for the design mass flow and rotation speed. Stator
shroud values are omitted from figure 11-9 since they pose no problem for the
constant swirl vortex type. Similarly, the last stator is omitted since its diffusion
load can be controlled independent of the stage performance data. Hence the
mid-stages feature higher work and more diffusion than either the front or the
rear stages. Based on the previous discussion, it might be expected that the φc

FIGURE 11-9 Stage Design Data for Compressor B
distribution should ramp down to lower values in the rear stages. That is gener-
ally not advisable due to adverse effects on the end-wall boundary layers. The
end-wall boundary layers are relatively thick in the rear stages, such that a reduc-
tion in meridional velocity can easily lead to excessive boundary layer blockage.
Indeed, the interaction between the end-wall boundary layer analysis and the
annulus sizing calculations can become so severe that convergence is virtually
impossible. That is the main reason why the present design system constrains the
annulus sizing to preclude an increase in passage area in the streamwise direc-
tion. Had φc been ramped down through the rear stages, that constraint probably
would have caused the lower values to be ignored anyway. To further illustrate
the flexibility available for tailoring the stage performance distributions, Com-
pressor C was designed using the design data shown in Fig. 11-10. This design
followed the same strategy as that used for Compressor B, but uses higher stage
flow coefficients to permit higher work coefficients. Figure 11-10 also shows the
key values of WRE obtained. Note that compressor C achieves higher work per
stage than Compressor B, but has similar flow diffusion characteristics.
Figure 11-11 compares performance predictions for Compressors A, B and C.
Polytropic efficiency is shown for this comparison to avoid the thermodynamic
effect due to pressure ratio discussed in Chapter 2. Both the pressure ratio and the
stable operating flow range for Compressor B are generally improved relative to
Compressor A. There is a slight loss in stable operating range at the design speed,
but improved stability at other speeds. There is a slight reduction in efficiency due

FIGURE 11-10 Stage Design Data for Compressor C
FIGURE 11-11 Predicted Performance Maps

to the reduced aspect ratios that result from the higher values of φc and pressure
ratio. This requires use of lower values of blade height, h, in the rear stages, result-
ing in higher predicted loss coefficients as seen from Eq. (6-99). The estimated
surge limit for all three compressors is based on stall criterion #3 of Chapter 9. At
design speed, this initially occurs in the mid-stages, whereas at other speeds the
front stages stall first. The improved stable operating range is particularly notice-
able at the lowest speed. As expected, the reduced diffusion in the front stages has
permitted the compressor to operate more effectively at low speeds. Note that the
improved stable operating range is also present in Compressor C, even though its
design pressure ratio is higher than that of Compressor B. Clearly, careful control
of the blade row diffusion characteristics is a primary consideration with regard
to improved compressor performance. But there are certainly other considera-
tions that may not be reflected by current aerodynamic performance prediction
technology. All designers certainly have strong incentive to favor large values of φc
and ψc to increase the mass flow per unit frontal area and to reduce the number of
stages required. Whether the goal is reduced size and weight or reduced cost, the
incentive is always present. Nevertheless, recent design trends appear to favor
lower values of φc and ψc based primarily on perceived benefits to performance.
Possible explanations were discussed in Section 10.4 of the previous chapter.
Other investigators appear to believe that the explanation lies in the end-wall
boundary layer behavior.
While lower values of φc and ψc may be preferred, the designer has consider-
able flexibility in regard to stage performance parameters. Rather large values of
φc and ψc have been used with success. For example, Fig. 11-12 shows calculated
stage performance parameters obtained from the performance analysis of the
NACA 10-stage subsonic compressor of Fig. 9-4. This compressor achieved
respectable performance, considering the state-of-the-art in axial-flow compres-
sor aerodynamic technology fifty years ago when it was designed. Yet rather large
values of φc and ψc were used in most of the stages. For illustration, Compressor D
was designed using the highest values of φc and ψc allowed by the rotor tip relative
Mach number limit imposed on the previous designs. The same operating condi-
tions and design strategy as those for Compressors B and C were used. Figure
11-13 shows the design parameters used and the flow diffusion characteristics
obtained. It was not possible to balance the flow diffusion as well for this case.
Blade stall in a constant swirl vortex stage almost always occurs first on the hub
contour, usually in the stators. Hence, lower rotor shroud velocity ratios were
accepted so that the hub contour flow diffusion characteristics could be main-
tained in a manner similar to that of Compressors B and C. Indeed, for any com-
pressor, the blade sections at the hub must operate over a much wider incidence
angle range than the rotor shroud blade sections. Figure 11-14 compares the pre-
dicted performance maps for Compressors B, C and D. Again, polytropic effi-
ciency is shown for this comparison to avoid the thermodynamic effect due to
pressure ratio discussed in Chapter 2. It is seen that aspect ratio effects have an
adverse effect on efficiency, particularly for Compressor D. A special performance
analysis of Compressor D using a larger number of shorter chord length blades in
the rear stages produced essentially the same efficiency at design speed as for the
other two designs. This indicates that the reduction in design speed efficiency
seen in Fig. 11-14 is due to aspect ratio effects rather than being a direct result of

FIGURE 11-12 Actual Stage Performance Data
FIGURE 11-13 Stage Design Data for Compressor D

FIGURE 11-14 Effect of c on Performance
the values of φc and ψc used in the design process. At lower speeds, both efficiency
and stability appear to deteriorate as higher values of φc and ψc are used in the
design process. Hence the performance analysis of chapter 9 indicates some
adverse effects at off-design speeds from designing with higher values of φc and ψc.
But it really does not indicate any significant influence of φc and ψc on the per-
formance at design speed. It is now generally accepted that designing with lower
values of φc and ψc offers definite performance benefits. It is clear that those ben-
efits are not completely quantified by the performance analysis of Chapter 9.
The present design system makes it rather simple to tailor the distributions of
stage performance parameters and blade row diffusion characteristics to empha-
size aerodynamic features considered important to the design objectives. This is
more clearly illustrated by another simple example: Suppose that Compressor B
must operate on a load line that lies too close to the surge line at the lower rota-
tion speeds. Compressor E is to be designed with the objective of improving the
low-speed surge margin while maintaining the same basic design speed perform-
ance as Compressor B. The annulus contour and the number of stages are to be
identical to Compressor B. In effect, the design of Compressor E is constrained to
be identical to Compressor B except for the blade geometry. Figure 11-15 shows
the stage design parameters and flow diffusion characteristics used to achieve
this objective. Figure 11-16 illustrates a load line that might require this redesign
and compares the predicted performance maps for the two compressors. Com-
parison with Fig. 11-9 shows that the design of Compressor E differs from that of

FIGURE 11-15 Design Data for Compressor E
FIGURE 11-16 Performance of Compressors B and E

Compressor B by a fairly modest shifting of the stage loading from the front to
the rear. It can be seen that a substantial improvement in the low-speed surge
margin has been achieved with minimal effect on the design speed performance.
11.7 THE SWIRL VORTEX TYPE

Chapter 10 reviews the influence of swirl vortex in the context of a single, repeat-
ing stage design. The present aerodynamic design system can be used to evaluate
alternate vortex design types in the context of multistage compressors designed
with the tailored distributions of stage performance parameters and flow diffu-
sion characteristics. Compressor B is used as the constant swirl vortex design for
this comparison. Alternate compressor designs have been generated using the free
vortex and constant reaction vortex types. Although the constant reaction vortex
type was not very effective for the single stage designs in Chapter 10, the present
application is better suited to this type of vortex. The major weakness seen in
Chapter 10 is the highly distorted meridional velocity profiles. The higher hub-to-
shroud radius ratio used in the present case will reduce the influence of that pro-
file distortion. Figures 11-17 and 11-18 summarize the design data used for the
alternate designs. The alternate designs are based on the same operating condi-
tions and distribution of φc as Compressor B. The distributions of Rc and ψc are
adjusted to produce flow diffusion characteristics similar to those in Fig. 11-9.
The free vortex and constant swirl vortex types have very similar flow diffusion
FIGURE 11-17 Free Vortex Design Data

FIGURE 11-18 Constant Reaction Vortex Design Data
FIGURE 11-19 Effect of Vortex Type on Performance

characteristics. Only minor modifications to the data in Fig. 11-9 were needed to
obtain the results shown in Fig. 11-17. The flow diffusion characteristics of the
constant reaction vortex type are quite different. There the rotor velocity ratios on
the shroud had to be balanced against the stator hub values for most of the stages.
The stator hub flow diffusion was expected limit stability, which was subsequently
confirmed by the performance analysis. Hence emphasis was placed on achieving
similar stator hub velocity ratios as those used for the other two vortex types,
without excessive diffusion at other locations. Fig. 11-19 compares the predicted
performance maps for the three compressor designs. Note that the flow range
from the design point to surge at the design speed is essentially identical for all
three compressors. The performance predictions indicate that the stator hub flow
diffusion characteristics have much more influence on stability than the swirl vor-
tex type. It is also seen that the compressor efficiency level is rather insensitive to
the swirl vortex type used. The constant reaction vortex design low speed charac-
teristics are shifted in flow relative to the other two designs, but the efficiency lev-
els are comparable for all designs. The free vortex design produced a slightly lower
pressure ratio than the constant swirl vortex design, but otherwise yields perform-
ance very similar to Compressor B. The constant reaction vortex type yields a sig-
nificantly lower pressure ratio than Compressor B. Possibly this could be
improved somewhat with further refinement, but it seems clear that the constant
reaction vortex design does not offer any advantage over the other two designs.
The constant reaction vortex type is much harder to work with and has definite
limitations with regard to acceptable hub-to-shroud radius ratios. Hence there
seems to be little reason to consider using it for axial-flow compressor design.
The blade geometry resulting from the various swirl vortex types is also a sig-
nificant consideration. Figures 11-20 and 11-21 compare camber and stagger
angles for the rotor and stator blades for the first stage of the three compressor
designs. The first stage is selected since it has the longest blades, but compar-
isons of blades in other stages would be quite similar. The rotor blades for the
three designs are so similar that there is no clear preference. The stator blade
geometry is very dependent on the swirl vortex type. The constant swirl vortex
produces a stator that can be well approximated by a constant-camber, untwisted
blade. This will be the easiest to manufacture and costs the least. The free vortex
stator can be well approximated by a constant-camber blade with a linear twist.
This would be the second choice with regard to cost and ease of manufacturing.
The constant reaction vortex requires a more complicate camber angle distribu-
tion, but a linear twist might be used to approximate the stagger angle distribu-
tion. The characteristic increase in camber angle near the shroud is apparent.
This would be much more pronounced if a lower hub-to-shroud radius ratio were
used, as can be seen in Fig. 10-32. Thus, the complexity of the stator geometry is
another reason to avoid the constant reaction vortex type.
The present aerodynamic design system makes it relatively simple to investigate
the influence of alternate swirl vortex types. Here, three quite different vortex types
have been compared for a sample problem quite typical of an industrial axial-flow
compressor. Based on this comparison, it is reasonable to conclude that swirl vor-
tex type is not a major factor with regard to the aerodynamic performance of axial-
flow compressors of this type. If lower hub-to-shroud radius ratios are used, there
may be some influence on performance associated with the different maximum

FIGURE 11-20 Comparison of Rotor Blade Geometry
FIGURE 11-21 Comparison of Stator Blade Geometry

inlet relative Mach number levels or extremes in flow diffusion levels produced by
the alternate swirl vortex types. But, in general, suitability to the design strategy
and objectives, along with the cost and ease of manufacturing of the blades, are
more likely to be the basis for selecting the swirl vortex type.
11.8 RISKS AND BENEFITS
The benefits from the present axial-flow compressor aerodynamic design proce-
dure should be relatively clear from the examples presented in this chapter. The
input specifications required are rather modest considering the complex design
problem involved. The input specifications are relatively easy to supply, and pro-
vide the designer with substantial control over the aerodynamic features consid-
ered most critical to the design objectives. The tedious process of sizing the
annulus, selecting blades and fitting them into the available space is left to the
computer. The design process is guided by continual feedback from a perform-
ance analysis for the design mass flow rate and rotation speed. When a viable
candidate design is identified, the design system can create an input file for the
aerodynamic performance analysis to permit an evaluation of its off-design per-
formance characteristics.
To obtain full benefit from the design system, on-demand monitor screen dis-
plays of tabular and graphical data are essential. End-wall contour geometry,
blade geometry, relative and absolute flow data, blade row performance data and
end-wall boundary layer analysis data are essential to evaluate the designs.
Graphical display of the equivalent velocity ratios, WRE, is particularly useful
when attempting to tailor the flow diffusion distributions similar to the examples
presented in this chapter. It has been noted that the capability to specify stage
performance for a few key stages and to interpolate for intermediate stages can
simplify the input specifications. The same is true for specifying the fixed end-
wall contour data. Specifying axial spacing is often simpler than specifying axial
coordinates. Provision to enter radii at some stations and to interpolate for inter-
mediate stations is also a useful simplification. It is important to use a represen-
tative blade tip clearance, but it is usually sufficient to use a constant value for all
blade rows. Monitor screen displays of the actual blade profiles at any radius are
also useful, and the design system should have the capability to export blade pro-
file coordinates for the drafting and manufacturing of the blades. Figure 11-22 is
an example of these data obtained from the design of Compressor B.
It is important to recognize that the substantial benefits of this design system
also introduce a level of risk that requires designers to exercise judgment. Chap-
ters 6 through 9 present numerous procedures used in the process of predicting
the performance of axial-flow compressors. All of those methods involve a signif-
icant level of approximation. A performance analysis is normally qualified
against experimental data to fine-tune it until it offers sufficient prediction accu-
racy for the intended applications. In general, none of the cases considered in
such a qualification study will be evaluated by the performance analysis as com-
pletely optimized. But the present design system provides the unique situation of
producing an axial-flow compressor design that is precisely optimized to the
models used in the performance analysis. Since those models are far from exact,

FIGURE 11-22 Typical Blade Profiles
the present design approach will inevitably expose the weaknesses in those
models. The most obvious risk is that the predicted design point efficiency is
likely to be overestimated. In the examples presented in this chapter, every blade
row is operating precisely at its minimum loss condition on all stream surfaces. It
is highly unlikely that the empirical models of Chapter 6, the through-flow analy-
sis of Chapter 7 and the boundary layer analysis of chapter 8 are all sufficiently
accurate to actually produce that situation throughout a multistage compressor.
This design system should provide a very good design, but probably not as good
as will be indicated by the performance analysis. Indeed, experienced turboma-
chinery aerodynamic designers are quite aware of the risk involved with design-
ing directly with the performance analysis that will be used to evaluate the
design. The axial-flow compressor is somewhat unique in that it is rather easy to
convert a performance analysis to a design system of the type described in this
chapter. The substantial benefits provided certainly justify taking advantage of
this opportunity, but definitely require considerable judgement by the designer to
maintain realistic expectations for the resulting design.

Chapter 12
QUASI-THREE-DIMENSIONAL
BLADE PASSAGE FLOW
FIELD ANALYSIS
The aerodynamic design and analysis of multistage axial-flow compressors is usu-

ally based on empirical cascade performance models as described in Chapters 6
through 11. Although that approach is sufficient for most applications, designers
occasionally require the support offered by more fundamental internal flow analy-
ses. When a specific application requires operating a blade row at conditions more
extreme than justified by experience, supplemental analysis can greatly increase
the designer’s confidence or perhaps prevent a poor decision. In the case of indus-
trial axial-flow compressors, this usually relates to the blade row inlet Mach num-
ber level. The industrial compressor designer usually has a number of discrete
geometrical scales or frame sizes available. If a specific application slightly exceeds
the upper limit of flow capacity for a frame size, the designer must either accept
somewhat higher Mach number levels or choose a larger frame size with a sub-
stantial increase in cost. Alternatively, transonic blade sections such as double-cir-
cular-arc blades might be considered. Usually that also involves increased cost and
may be unattractive with respect to structural integrity. Even when the application
is well within the normal flow capacity limits of a frame size, an unusual working
fluid or a required rotation speed can result in Mach number levels beyond the
designer’s experience. In those cases, designers need a more fundamental internal
flow analysis to better evaluate probable impact of the higher Mach number levels.
Chapter 5 presented several useful techniques for the analysis of the internal
flow on stream surfaces within blade passages. Those blade-to-blade flow analy-
ses can provide the desired detailed evaluation, but all of them require that the
stream surface geometry and stream sheet thickness be specified in some fash-
ion. An approximate analysis can be conducted using the stream sheet geometry
before and after the blade row from the meridional through-flow analysis of
Chapter 7, while assuming a linear variation between those stations. That is cer-
tainly the fastest and simplest approach and often may be considered sufficient.
Alternatively, a more accurate analysis can be conducted using the quasi-
three-dimensional flow analysis technique mentioned in Chapters 3 and 5. That
is a very efficient analysis technique that provides a reasonable approximation to
the three-dimensional flow field through a blade row. Originally suggested by Wu

(1952), this technique relies on solving for the two-dimensional flow on the
blade-to-blade stream surfaces and on the hub-to-shroud stream surface, with
interaction between them until they are consistent with each other. This chapter
describes a quasi-three-dimensional flow analysis suitable for that purpose. This
analysis was originally developed for centrifugal compressors (Aungier, 2000),
where it plays an essential role in the aerodynamic design of impellers. It was
subsequently extended to treat arbitrary three-dimensional blade geometry for
additional flexibility in centrifugal impeller design. That extension made it possi-
ble to use the analysis directly for axial-flow compressor blade rows.
The only special extension added for axial-flow compressors is provision for
automatic generation of detailed blade geometry for all of the standard blade
profiles discussed in Chapter 4. This permits complete blade specification in
terms of the radial variation of the basic blade section parameters (e.g., θ, γ, c and
tb/c). The same capability is incorporated into the various blade-to-blade flow
analyses of Chapter 5 for a single blade section. Hence it is relatively simple to
provide the capability to automatically set up input files for either a simple blade-
to-blade analysis on a single stream surface or a quasi-three-dimensional flow
analysis for a blade passage directly from a completed meridional through-flow
analysis. In the author’s design and analysis system, both the performance analy-
sis of Chapter 9 and the design system of Chapter 11 contain that provision.
When a more detailed analysis is needed, it can be obtained in a few minutes
with very little additional effort.
NOMENCLATURE
B* = stream surface repositioning damping factor

D = stream surface curvature damping factor
f = blade force
g = 2πrcosβ′ /Z
H = total enthalpy
H1 = meridional shape factor = δ 1*/θ11
H2 = boundary layer tangential shape factor = δ 2*/θ22
h = static enthalpy
I = rothalpy
KB = blade blockage factor
M = Mach number
P = pressure
q = inlet dynamic head
r = radius
s = entropy
T = temperature

Quasi-Three-Dimensional Blade Passage Flow Field Analysis • 289
tb = blade thickness
W = velocity relative to the blade row
y = distance along quasi-normal
β = flow angle
∆A = stream sheet area
∆ν = increment in ν due to blade guidance
δc = blade clearance
δ 1* = boundary layer meridional displacement thickness
δ 2* = boundary layer tangential displacement thickness
ε = angle between quasi-normal and true normal
η = dimensionless tangential coordinate
θ = tangential coordinate and streamwise momentum thickness
θ11 = meridional momentum thickness
θ22 = tangential momentum thickness
κ = camberline angle with meridional direction
κm = stream surface curvature
λ = quasi-normal angle (Fig. 12-3)
ν = blade force defect thickness
ρ = fluid density
φ = stream surface angle with the axial direction
Subscripts
h = parameter at the hub contour

s = parameter at the shroud contour
Superscripts
′ = parameter in rotating coordinates

 = mass-averaged value
12.1 QUASI-THREE-DIMENSIONAL FLOW

Figure 12-1 shows a meridional plane view of a blade row to be analyzed. The
annular passage is divided into a series of stream sheets, separated by stream sur-
faces. All stream sheets are defined to contain equal mass flow. For simplicity, it is
assumed that all stream surfaces are axisymmetric, i.e., stream sheet twisting will
be ignored. The meridional coordinate, m, is measured along a stream sheet’s
mean surface, and the stream sheet thickness is designated as b. If the stream sur-

FIGURE 12-1 Hub-to-Shroud Flow Geometry
faces and the distribution of b are known, a complete two-dimensional blade-to-

blade flow analysis can be conducted using one of the methods described in Chap-
ter 5. Figure 12-2 shows a typical blade-to-blade surface view to illustrate the
solution domain for one of the blade-to-blade flow analyses to be conducted. Since
blade-to-blade flow analysis has been covered extensively in Chapter 5, it is only
necessary to identify the specific method to be used. The best choice for quasi-
three-dimensional flow analysis is the linearized potential flow analysis of Section
5.4, including the extension to transonic flow. That method offers exceptional com-
putation speed and reliability as well as excellent prediction accuracy for compres-
sor blade rows. To be sure, the two-dimensional blade-to-blade flow analyses offer
somewhat better accuracy. But it is still possible to take advantage of those more
general analyses when needed. The linearized method is quite adequate to provide
support to the hub-to-shroud flow analysis, such that accurate stream surface loca-
tions and stream sheet thicknesses are obtained. Once the quasi-three-dimensional
flow analysis is complete, all input data needed for the two-dimensional blade-to-
blade flow analyses are available. It is quite simple to make provision to generate
input files for those more exact methods for any or all stream surfaces—that is a

FIGURE 12-2 Blade-to-Blade Flow Geometry
more efficient approach than incorporation of the more exact methods directly
into the quasi-three-dimensional flow analysis. Indeed, it is not uncommon for a
complete quasi-three-dimensional flow analysis to require significantly less com-
puter time than a blade-to-blade flow analysis for the same problem on a single
stream sheet by the two-dimensional blade-to-blade flow analyses.
The hub-to-shroud flow analysis usually predicts the average meridional
through flow in the passage. Alternatively, the meridional flow field on a mean
hub-to-shroud stream surface is sometimes predicted. The prediction of the aver-
age through flow will be used in this chapter, since it is more consistent with the
assumption of axisymmetric blade-to-blade stream surfaces. The hub-to-shroud
flow analysis is similar to the meridional through-flow analysis of Chapter 7,
except that the tangential momentum equation is no longer satisfied by conser-
vation of angular momentum. Indeed, the distribution of angular momentum
through the blade row must be obtained from the blade-to-blade flow analyses
for all stream surfaces.
12.2 HUB-TO-SHROUD FLOW GOVERNING EQUATIONS
The hub-to-shroud flow analysis is conducted using the quasi-normal coordinate

system introduced in Chapter 7 for the meridional through-flow analysis. The
basic nomenclature and construction of that coordinate system developed in
Chapter 7 are summarized in the following equations and in Fig. 12-3.

FIGURE 12-3 Quasi-Normal Geometry
∆z zh − zs
tan λ = = (12-1)
∆r rs − rh
∂r
sin φ = (12-2)
∂m
ε =φ−λ (12-3)
∂ 1 ∂ ∂ 
=  − sin ε  (12-4)
∂n cos ε  ∂y ∂m 
∂φ
κm = − (12-5)
∂m
The relevant equations for adiabatic inviscid flow are developed in general
form in Chapter 3 in a natural coordinate system. For the present application,
Eqs. (3-21), (3-25) and (3-30) apply after they are simplified to their time-steady
form. It will be convenient to satisfy conservation of mass in integral form rather
than using Eq. (3-21). Conservation of mass along a quasi-normal can be
expressed in the form
ys
˙ = 2π ∫ K B rρ Wm cos ε dy
m (12-6)
0

KB is the blade blockage factor, which corrects the stream sheet area available for
through-flow to account for blade metal blockage. If the angle formed by a tan-
gent to the blade camberline with respect to the meridional direction is denoted
by κ, KB is given by
K B = 1 − tbZ / (2π r cos κ ) (12-7)
Consistent with the blade-to-blade flow analysis to be used, rothalpy and

entropy are assumed to be constant on stream surfaces, but may vary in the
direction normal to the stream surfaces. Using Eqs. (3-25) and (3-30), these
assumptions can be expressed by
∂I ∂s (12-8)
= =0
∂θ ∂θ
∂I ∂s
= =0 (12-9)
∂m ∂m
Wθ ∂( rWθ + ω r 2 ) ∂Wm ∂I ∂s
κ mWm2 + + Wm = −T (12-10)
r ∂n ∂n ∂n ∂n
Equation (12-10) can be solved for Wm if Wθ, s and I are known on all stream sur-
faces. I and s can be supplied through assigned upstream boundary conditions,
just as was done for the meridional through-flow analysis in Chapter 7. Equa-
tions (12-8) and (12-9) provide values at all other points in the flow field. The dis-
tribution of Wθ on the stream surfaces must be obtained from the blade-to-blade
flow analysis. Combining Eqs. (3-29) and (12-8),
∂( rWθ + ω r 2 ) ∂Wm
= (12-11)
∂m ∂θ
Chapter 5 develops the blade-to-blade flow analysis for axisymmetric stream

surfaces. Indeed, Eq. (12-11) is the principal momentum equation used in the
blade-to-blade flow analysis. Clearly, a hub-to-shroud flow analysis restricted to
axisymmetric stream surfaces can match only a single distribution of Wθ in the
meridional plane. The most logical choice is to use a mass-averaged distribution
of Wθ obtained from the blade-to-blade flow analyses on the various stream sur-
faces. In effect, the hub-to-shroud flow analysis will generate an axisymmetric
flow field that is consistent with the mass-averaged angular momentum distribu-
tion produced by the blade-to-blade flow analyses. Equation (12-11) is basically
ignored by this approximate hub-to-shroud analysis for all points within a blade
passage. This equation can be satisfied for points upstream and downstream of
the blade passage, where the conservation of the mass-averaged angular momen-
tum is required. In this case, Eq. (12-11) can be written as
∂( rWθ + ω r 2 ) ∂rCθ
= =0 (12-12)
∂m ∂m

Since the linearized blade-to-blade flow analysis of Chapter 5 applies only to

points within the blade passage, Eq. (12-12) is required to treat points outside the
blade passage. Combining Eqs. (3-1), (12-4), (12-9) and (12-10) yields the normal
momentum equation in the quasi-normal coordinate system.
∂Wm  sin ε ∂Wm  Wθ  ∂rCθ ∂rCθ 

+ κ m cos ε −  Wm +  − sin ε 
∂y  Wm ∂m  rWm  ∂y ∂m 
(12-13)
1  ∂I ∂s 
=  −T 
Wm  ∂y ∂y 
For points outside the blade passage where Eq. (12-12) can be applied, it is easily
seen that Eq. (12-13) is identical to Eq. (7-12), which is used for the meridional
through-flow analysis. For points within the blade passage it is more convenient
to express the angular momentum in terms of the relative flow angle, noting that
Wθ = Wm tanβ′. This yields
 sin ε ∂Wm tan β ′  ∂r tan β ′ ∂r tan β ′  

κ m cos ε − +  − sin ε   Wm
 Wm cos β ′ ∂m
2 r  ∂y ∂m  
(12-14)
1 ∂Wm 1  ∂I ∂s 
+ + 2ω tan β ′ (cos λ − sin φ sin ε ) =  −T 
2
cos β ′ ∂y W m  ∂y ∂y
For points outside the blade passage where Eq. (12-12) applies, Eq. (12-14)
should reduce to the form used in the meridional through-flow analysis of chap-
ter 7, i.e., to Eq. (7-14). Introducing Wθ = Wm tanβ′ into Eq. (12-12), it is easily
shown that
Wm tan β ′ ∂r tan β ′ ∂Wm

= − tan 2 β ′ − 2ω sin φ tan β ′ (12-15)
r ∂m ∂m
Substituting Eq. (12-15) into (12-14) does yield Eq. (7-14) as required. While the
two forms are mathematically equivalent, some care is recommended when for-
mulating a numerical analysis. Equations (12-13) and (12-14) are necessary when
the angular momentum distribution is supplied externally by the blade-to-blade
flow analyses. But outside of the blade passage where Eq. (12-12) is applicable, use
of Eqs. (7-12) and (7-14) is likely to result in a more accurate numerical solution.
12.3 NUMERICAL INTEGRATION OF THE GOVERNING

EQUATIONS
The mass and momentum conservation equations can be numerically integrated

for a specified stream surface geometry and flow angle distribution through the
blade passage. The numerical method used is basically identical to the method
used in chapter 7. The momentum equation is written in the form

∂Wm f ( y)
= f1( y)Wm + f2 ( y) + 3 (12-16)
∂y Wm
where f1, f2 and f3 are known functions of y. For quasi-normals within the
blade passage,
 tan β ′  ∂r tan β ′ ∂r tan β ′   sin ε ∂Wm

f1( y) = cos2 β ′ −κ m cos ε −  − sin ε  + (12-17)
 r  ∂y ∂m   Wm ∂m
f2 ( y) = −2ω cos β ′ sin β ′ [cos λ − sin φ sin ε ] (12-18)
 ∂I ∂s 
f3 ( y) = cos2 β ′  − T 
 ∂y ∂y (12-19)
For quasi-normals upstream and downstream of the blade passage,
sin ε ∂Wm
f1( y) = −κ m cos ε + (12-20)
Wm ∂m
f2 ( y) = 0 (12-21)
∂I ∂s Wθ ∂( rCθ )
f3 ( y) = −T − (12-22)
∂y ∂y r ∂y
At the upstream boundary, it is usually more convenient to supply boundary con-

ditions in the stationary frame of reference, in terms of distributions of H, s and
either Cθ or β. The first case is easily modeled by substituting Cθ for Wθ and H for
I in Eq. (12-22). When the distribution of β is known, Eqs. (7-19) through (7-21),
in the stationary frame of reference, yield
 tan β ∂( r tan β ) sin ε ∂Wm 

f1( y) = cos2 β −κ m cos ε − +  (12-23)
 r ∂y Wm ∂m 
f2 ( y) = 0 (12-24)
 ∂H ∂s 
f3 ( y ) = cos2 β  −T  (12-25)
 ∂ y ∂ y
For specified stream surface geometry and flow angle distributions through the
blade passage, f1, f2 and f3 will be evaluated at all grid points before integrating
the momentum equation. This process is straightforward with the exception of
the meridional gradient of Wm in f1, which depends on the solution. Usual prac-
tice is to use values of Wm from the previous iteration to evaluate this gradient.
That approach can result in numerical stability problems in cases where Wm is
changing rapidly between iterations. Since Wm appears in the denominator, it
can also produce singular results when Wm approaches zero. Aungier (2000) sug-
gests a better approach based on simple conservation of mass in all stream

sheets. If the local through-flow area of a stream sheet is designated by ∆A, con-
servation of mass in the stream sheet requires
∆ṁ = ρ Wm ∆A (12-26)
Then the meridional gradient in f1 can be replaced by
1 ∂Wm 1 ∂ρ ∆A
=− (12-27)
Wm ∂m ρ ∆A ∂m
This expresses the gradient in terms of the stream sheet geometry and gas
density. The stream sheet geometry is constant during the process of integrating
the mass and momentum conservation equations. It is still necessary to rely on
values of gas density from the previous iteration, but density normally does not
vary greatly between successive iterations. Since the mass flow rate in all stream
sheets is constant, the risk of singular results is also eliminated. Aungier (2000)
uses the same approach to remove the singularity in Eq. (12-16) when Wm
approaches zero by defining a new function, f4, as
ρ ∆A (12-28)
f4 ( y) = f2 ( y) + f3 ( y)
∆m ˙
Then Eq. (12-16) can be written as
∂Wm
= f1( y)Wm + f4 ( y) (12-29)
∂y
The solution of this linear differential equation has already been given in
Chapter 7 as
y
f4 ( y)
Wm ( y) = Wm (0)F( y) + F( y)∫ dy (12-30)
F ( y)
0
where
y 
F( y) = exp  ∫ f1( y)dy (12-31)
 
0 
The meridional velocity on the hub contour, Wm(0), is the constant of integration.
It is determined from conservation of mass through Eq. (12-6). Equations (12-29)
and (12-6) are solved in an iterative numerical scheme, successively improving
the estimate of Wm(0) until mass is conserved and the momentum equation is
satisfied. This requires calculation of thermodynamic properties such as ρ, using
an appropriate equation of state from Chapter 2. At any point, the relative total
enthalpy is given by Eq. (3-13), i.e.,

H ′ = I + 12 ( rω )2 (12-32)
The local static enthalpy is given by
h = H ′ − 12 W 2 (12-33)
Then static thermodynamic conditions are computed from relative total thermody-
namic conditions for the change in enthalpy, (h – H′), while holding entropy con-
stant. The entropy and rothalpy on each stream surface is known from the assigned
upstream boundary conditions. Similarly, the angular momentum, rCθ, upstream of
the blade on each stream surface is known from the assigned upstream boundary
conditions. The angular momentum downstream of the blade on each stream sur-
face is obtained from the solution at the blade trailing edge quasi-normal.
12.4 REPOSITIONING STREAM SURFACES
After integrating the mass and momentum conservation equations, the new flow
field data is generally not consistent with the resident stream sheet geometry.
The correct stream sheet geometry can be computed by integrating Eq. (12-6)
across the passage at each quasi-normal and interpolating for the new stream
sheet positions such that all stream sheets contain equal mass flow rates. In
practice, fairly sophisticated numerical damping procedures are required to
avoid numerical instability problems. This writer uses the numerical damping
procedure suggested by Novak (1973). For quasi-normals outside of the blade
passages Novak recommends
2
1 (1 − Mm )( ∆y)2
= 1+ (12-34)
F B* ( ∆m)2
F is the fraction of the distance between the new and old stream sheet positions
that will actually be used in repositioning the stream sheets, ∆y is the hub-to-
shroud quasi-normal length, ∆m is the minimum meridional spacing between
quasi-normals, Mm is the meridional Mach number and B* is an empirical con-
stant. For quasi-normals inside the blade passage, Novak suggests
1 (1 − M′ 2 )(cos β ′∆y)2
= 1+ (12-35)
F B* ( ∆m)2
This damping procedure has been found to be quite effective as long as Mm and
M′ are limited to be no greater than 0.9. B* values used are typically in the range
of 8 to 16, but the numerical procedure should adjust B* automatically based on
whether the stream surface position errors are increasing or decreasing on suc-
cessive iterations.

12.5 THE HUB-TO-SHROUD FLOW ANALYSIS
To start the solution, an initial guess must be supplied for all flow field data used
in the hub-to-shroud solution. Since no blade-to-blade flow analysis results are
available at this point, the initial guess must include an approximate blade-to-
blade flow definition. This writer uses the following procedure to initialize the
flow field data:
• Stream surfaces are positioned such that all stream sheets contain the
equal through-flow area on each quasi-normal.
• Entropy and rothalpy are computed for all stream surfaces from the
upstream boundary conditions.
• The meridional velocity is computed from local conservation of mass
assuming the local gas density is equal to the relative total density.
• Cθ upstream of the blade is obtained from conservation of angular
momentum, Eq. (12-12), using the upstream boundary conditions. This
includes the blade leading edge quasi-normal.
• The blade is assumed to provide perfect guidance to the flow over the
last 85% of the blade passage length based on the blade camberline
angle, κ.
• Cθ downstream of the blade is obtained from conservation of angular
momentum using the value at the trailing edge quasi-normal.
• An “inlet slip” condition is applied over the first 15% of the blade pas-
sage length based on the relative flow angle computed at leading edge
quasi-normal and the blade camberline angle, κ, i.e.,
tan β ′ = tan β LE ′ )[(m − mLE ) / (mTE − mLE ) / 0.15]2

′ + (tan κ − tan β LE (12-36)
This initialization procedure is very conservative and is virtually certain to

successfully start the analysis. But Eq. (12-36) is a rather crude approximation
for the blade-to-blade flow that often results in difficult and slow convergence of
the hub-to-shroud flow analysis. Once the first blade-to-blade flow analysis
results are obtained, convergence of the hub-to-shroud analysis consistently
improves dramatically. A much more efficient overall analysis is obtained by sim-
ply not requiring convergence on the first attempt at the hub-to-shroud flow
analysis. This writer simply limits the number of iterations on the hub-to-shroud
flow governing equations on this first attempt to, say 10-12, while ignoring the
convergence requirement. This is sufficient to obtain a reasonable stream sheet
pattern to run the first blade-to-blade flow analysis.
The general procedure used for the hub-to-shroud flow analysis is a direct
application of the well-known streamline curvature technique. The basic conser-
vation equations are solved at each quasi-normal using the procedures described
in Section 12.3. Then the stream surfaces are repositioned as described in Sec-
tion 12.4. The process is repeated until the stream surface positions and flow
field data have converged within an acceptable tolerance. This procedure is ade-
quate for most problems, but Aungier (2000) suggests two simple extensions that
can greatly improve the reliability of the analysis to the point that a converged

solution is almost always obtained: First, the hub-to-shroud flow analysis should
check for choked flow, and limit the overall mass flow rate to the choke value
when it occurs. In the nomenclature of Section 12.3, the flow on any quasi-nor-
mal is locally beyond the choke limit when
∂m
˙
≤0 (12-37)
∂Wm (0)
where ṁ is the calculated mass flow rate obtained from Eq. (12-6). When this
condition is encountered, the solution procedure should determine the mass
flow rate for which this gradient term is zero, which is the choke value. The
choked flow condition of Eq. (12-37) is not always an indication of true choked
flow. Often it is caused by numerical errors during the early iterations. The
solution should proceed using the largest acceptable mass flow rate, but not
exceeding the specified mass flow rate. In the case of a temporary choke due to
numerical error, the mass flow rate will gradually increase to the specified
value as the solution converges. This simple procedure avoids the most com-
mon cause of solution divergence. Another common cause of solution diver-
gence is numerical errors in the streamline curvature terms. The numerical
damping of the stream surface repositioning given in Section 12.4 avoids most
of these problems. But that damping procedure is indirect with respect to the
curvature terms. This writer also imposes a direct damping procedure that can
be expressed as
κ m → (κ mi + Dκ mi−1 ) / (1 + D) (12-38)
where the subscripts i and i-1 refer to the iteration number and D is a damping
factor. Typically, D = 1 is used, but D is increased if successive iterations show a
significant increase in convergence errors.
12.6 COUPLING THE TWO BASIC FLOW ANALYSES
The quasi-three-dimensional flow analysis is obtained by coupling the hub-to-

shroud flow analysis described in this chapter with the linearized blade-to-blade
flow analysis described in Chapter 5. The hub-to-shroud flow analysis defines the
stream sheet geometry and corresponding blade geometry for the blade-to-blade
flow analysis. The stream sheet thickness is also required, which is computed
from the hub-to-shroud flow data and the specified stream sheet mass flow by
simple conservation of mass.
˙ / (2π rK B ρ Wm )
b = ∆m (12-39)
Similarly, when the blade-to-blade flow analysis is completed, the results are
mass-averaged across the blade passage to provide the relative flow angle distri-
bution to be used on the next hub-to-shroud flow analysis. Recalling that the

dimensionless tangential coordinate, η, used in the blade-to-blade flow analysis

varies from 0 to 1, the data required are given by
∫ ρ bWmdη
2
Wm = 0
1
(12-40)
∫ ρ bWmdη
0
1
∫ ρ bWmWθ dη
Wθ = 0
1 (12-41)
∫ ρ bWmdη
0
tan β ′ = Wθ / Wm (12-42)
In addition to requiring convergence of the two component analyses, a conver-

gence criterion is needed for the overall quasi-three-dimensional flow analysis.
This writer requires that all relative flow angles predicted by Eq. (12-42) on suc-
cessive blade-to-blade flow analyses converge within an acceptable tolerance.
Figure 12-4 shows a basic flow chart for the quasi-three-dimensional flow
analysis. The combination of the present hub-to-shroud analysis with the lin-
earized blade-to-blade flow analysis of Chapter 5 provides a very reliable and effi-
FIGURE 12-4 Flow Chart of the Analysis

cient quasi-three-dimensional flow analysis for a wide range of axial-flow com-

pressor blade passage problems of interest. The major limitation is with regard to
Mach number levels, since the blade-to-blade flow analysis cannot account for
shock waves. That analysis does have the transonic extension described in Chap-
ter 5, so some supersonic flow is acceptable. The other major restriction follows
from the basic assumption of inviscid flow. This is a reasonable assumption as
long as the incidence angles are reasonable and flow diffusion is not excessive.
Extreme incidence angles or excessive diffusion will result in substantial flow
separation, which cannot be modeled by an inviscid flow analysis. To obtain
meaningful results, those cases require a more fundamental viscous computa-
tional fluid dynamics code.
This analysis requires a very large number of equation-of-state calculations.
The pseudo-perfect gas equation of state model described in Chapter 2 can
greatly reduce the computational time relative to real gas models, or even ideal
gas models with temperature-dependent specific heats. This model is almost
always adequate for blade passage flow analyses in compressors. This follows
from the fact that some care is normally taken to avoid any risk of operating com-
pressors with liquid present in the flow. Hence the operating conditions are nor-
mally safely in the super-heat zone, where gas properties do not show significant
variation over the limited range of conditions needed for a flow field analysis. In
contrast, turbine applications may involve two-phase flow, or inlet conditions
very close to the vapor saturation line. In either case, the pseudo-perfect gas
model is usually not adequate for those problems.
Quasi-three-dimensional flow analysis plays a central role in the aerodynamic
design of blades for centrifugal compressors, where each blade row is typically a
unique design for the specific application (Aungier, 2000). As discussed at the
beginning of this chapter, its role for axial-flow compressors is more likely to
involve an evaluation of a specific blade row operating under conditions more
extreme than justified by experience. To be effective in that role, the analysis
must be easily applied as well as fast and reliable. The author’s analysis includes
the methods described in Chapter 4 to very precisely define the blade geometry
for standard axial-flow compressor blade camberlines and profiles to minimize
the geometry data required. This, in turn, makes it quite simple to provide the
performance analysis of Chapter 9 and the aerodynamic design program of chap-
ter 11 with the capability to export a complete input file for the present analysis.
Hence a more detailed evaluation of any blade row analyzed in one of those pro-
grams can be accomplished with very little effort.
The performance analysis conducted on the NACA 10-stage compressor in
Chapter 9 will be used to illustrate a typical application of this analysis. With ref-
erence to Fig. 9-4, the performance analysis for a mass flow rate of 55 at 100%
speed predicts that the first stator operates at a supercritical Mach number along
the hub contour. Although the performance analysis does not indicate any signif-
icant adverse results, the designer might want to evaluate this situation closely.
The empirical models for estimating the critical Mach number are certainly quite
approximate, and the NACA 65-series blades used are not well suited to operating
with significant supersonic flow. To evaluate this situation, a quasi-three-dimen-
sional flow analysis input file for this stator was exported from the performance
analysis. After running the quasi-three-dimensional flow analysis, an input file

FIGURE 12-5 A Blade Loading Evaluation
for the two-dimensional blade-to-blade flow analysis of Chapter 5, Section 5.3,

for the hub surface was exported and processed for a more accurate analysis on
this critical stream surface. Figure 12-5 presents the results from these two
analyses. It is seen that the stator does operate at a supercritical Mach number.
As noted in Chapter 5, blade-loading data close to the leading edge predicted by
the linearized blade-to-blade flow analysis should generally be ignored. In that
context, the internal flow analyses indicate that the flow is only slightly super-
sonic and should not pose any significant risk of a performance penalty. The
entire process of completing two internal flow analyses to evaluate this situation
was accomplished in less than two minutes. That is an important advantage
offered by inclusion of efficient inviscid internal flow analysis techniques in a
well-formulated aerodynamic design and analysis system. Had a fundamental
viscous CFD code been used for this purpose, the evaluation might easily have
taken several days. This analysis really encourages designers to evaluate areas of
concern, which might otherwise be ignored due to the prohibitive effort required
for a more detailed evaluation.
12.7 BOUNDARY LAYER ANALYSIS
Boundary layer analysis provides a useful extension to an inviscid internal flow

analysis. It can provide an approximate evaluation of viscous effects with no sig-
nificant increase in computation time. Chapter 5 describes a two-dimensional

blade surface boundary layer analysis and loss coefficient model that can be
incorporated into the quasi-three-dimensional flow field analysis directly. Those
models are strictly applicable to just two-dimensional cascade flows. They ignore
the secondary flows that are always present in a real cascade. Nevertheless, they
do provide a reasonable qualitative evaluation of viscous effects that can provide
useful guidance to the designer. Unlike the case of blade surface boundary layers,
there is little merit in considering two-dimensional models for the highly three-
dimensional end-wall boundary layers. Nor is there much merit today in consid-
ering use of inviscid flow analysis coupled with three-dimensional boundary
layer analysis models. That approach would involve computation times and com-
plexity comparable to a viscous CFD code, but would offer less generality and
accuracy. With the many excellent commercially available viscous CFD codes
available today, there is no reason to consider more approximate methods unless
they offer substantial advantages in terms of computation speed and reliability.
Aungier (2000) uses an end-wall boundary layer analysis that is an early version
of the analysis described in Chapter 8. That end-wall boundary layer analysis has
since been upgraded to include some of the newer developments described in
Chapter 8. This type of axisymmetric, three-dimensional boundary layer analysis
is about the only method that can provide meaningful results and offer the com-
putation speed and reliability required. This approximation is known to yield
excellent results in vaneless passages [Aungier, 1988(b) and Davis, 1976]. Within
the blade passages, the boundary layers are certainly not axisymmetric. There the
model seeks to predict the gap-averaged boundary layer behavior (Horlock, 1970)
as discussed in Chapter 3.
The description of the end-wall boundary layer analysis provided in Chapter 8
is directly applicable here with two very specific exceptions: In Chapter 8, all
meridional computing stations lie outside of the blade passages, so when blade
rows are encountered, the meridional integration step is always across the blade
row. In the present application, there will be many meridional computing sta-
tions within the blade passage. The process of numerical integration remains the
same, but here the boundary layer equations are integrated over many spatial
steps within the blade passage instead of a single spatial step across it. That, in
turn, requires revision of the blade force defect model for it to be applicable
within the blade passage. Aungier (2000) suggests the following blade force
defect thickness model:
ν = Kθ11 (12-43)
Based on experimental data from Koch and Smith (1976) and Hunter and Cump-
sty (1982), values of K in the range of 0.7 to 1 were recommended. Since those sets
of experimental data are from axial-flow compressor tests, it is not too surprising
that this simple model yields reasonable results for axial-flow compressor appli-
cations. However, since Aungier (2000) was published, a number of centrifugal
compressor applications have been encountered where the results did not appear
realistic. The weakness in the model has been most evident during analysis of the
boundary layer within a blade passage where both the blades and the wall are
rotating. As the angle φ becomes large, the centrifugal forces in the boundary layer
also become large, tending to reduce θ11 and, therefore, ν. Under some conditions,

this can produce nearly uniform meridional velocity profiles and highly distorted
tangential velocity profiles. It is evident that such highly skewed boundary layer
profiles are not reasonable within the blade passages, where blades will tend to
guide the flow toward a collateral boundary layer condition. The analysis of
Aungier (2000) does not recognize that extreme skewing of the boundary layer
flow profiles will be constrained by the flow guidance imposed by the blades.
These observed weaknesses in the method described in Aungier (2000) have
recently been significantly improved by adapting the blade force defect model of
Chapter 8 to the present application. Figure 12-6 shows the same experimental
data as Fig. 8-5 and the simplified empirical correlation used in the present appli-
cation. Since the present analysis is intended for single blade passage analyses, the
larger values of displacement thickness observed in the rear stages of multistage
compressors will not be encountered. With little likelihood of approaching the
maximum postulated in Fig. 8-5, the simpler linear empirical correlation is a rea-
sonable approximation, which eliminates the staggered spacing term in the corre-
lation. Hence the basic blade force defect thickness of Eq. (8-44) is replaced by
ν0 = 0.4δ1* + δ c / 2 (12-44)
It can be seen that Eq. (12-44) is similar to Eq. (12-43), but is based on the dis-
placement thickness rather than the momentum thickness and includes the
influence of the blade clearance. Following the approach used in Chapter 8, the
FIGURE 12-6 Tangential Force Defect Thickness

blade force defect thickness is modified by a term ∆ν to account for the influ-
ence of the blade guidance
ν = ν0 + ∆ν (12-45)
This correction term is computed exactly as was done in Chapter 8, using Eqs.
(8-53) and (8-57), but with an alternate definition for the parameter, F, that is
more suitable to integration within the blade passage. The form used for F is
based on the following assumptions:
• The blades guide the flow toward collateral flow primarily close to the
blade surfaces. It is assumed that the gap-averaged deviation from col-
lateral flow imposed by integrating the boundary layer equations, using
only ν0, will be reduced by about 50% due to the blade guidance.
• The influence of blade guidance and clearance gap on the blade force
defect is negligible when the blade clearance gap is greater than the
boundary layer thickness.
• The influence of blade guidance and clearance gap on the blade force
defect is negligible at the blade row leading edge. This influence is
assumed to be proportional to the ratio of the meridional distance to
the blade spacing.
The second assumption simply recognizes the special case where the boundary
layer and blades are not really interacting. Although this situation is probably
very rare in a compressor, it can easily occur in the present analysis if the
entrance boundary layer thickness is specified small enough. Experience has
shown that this will lead to erratic and highly questionable results. The third
assumption recognizes that the influence of the blades will not extend across the
entire blade passage spacing until the flow penetrates well into the guided pas-
sage. This leads to the following approximation for F:
F = [1 − δ c / δ ]m / [( g + m] / 2 ; δ c ≤ δ (12-46)
F = 0; δ c > δ (12-47)
Equation (12-46) includes the influence of the ratio m/g, where m is the merid-
ional distance from the blade leading edge and g is analogous to the local stag-
gered spacing used in Chapter 8, i.e.,
g = 2π r cos β ′ / Z (12-48)
F is used directly in Eq. (8-57) and is also imposed as a correction applied to

Eq. (12-44).
ν0 = 0.4δ1* + δ cF (12-49)
Since there are now several spatial integration steps within the blade passage, the
blade force defect and the blade force direction given by Eq. (8-49) both vary

through the blade passage, but the process of numerical integration is the same
as that used in Chapter 8. However, the calculation of ∆ν requires more care since
it relates to a local blade force correction rather than an overall blade row cor-
rection. Locally, the ideal blade force may be extremely small, or even zero. This
can produce unrealistic values of ∆ν. To avoid such problems, a constraint is
imposed on ∆ν in the form
∆ν ≤ [0.05 + 0.2 ( H1 − 1)]δ (12-50)
This constrains the blade force correction to be less than about 75% of the merid-
ional displacement thickness, except at values of H1 very close to 1. The displace-
ment thickness will vanish when H1 → 1, so special care is required to avoid
suppressing the blade guidance effect entirely in that case.
Figures 12-7 and 12-8 show typical boundary layer parameters predicted by
this end-wall boundary layer analysis when applied to the axial-flow compressor
stator problem used as an example earlier in this chapter. This blade row oper-
ates at near optimum incidence angles at the operating conditions considered in
this sample case. Hence, the blade row pressure coefficient is relatively mild for
this case. Based on the experimental data shown in Fig. 8-2, the blade force
defect thickness, averaged over the blade row, is expected to be a fairly large frac-
tion of the meridional displacement thickness, which is at least consistent with
FIGURE 12-7 Hub Contour Boundary Layer

FIGURE 12-8 Shroud Contour Boundary Layer
the present predictions. That certainly does not provide a real validation of the
assumptions, nor is there any data available in the literature suitable for validat-
ing them. The assumptions are considered reasonable, and they produce results
consistent with overall blade row experimental data. Also, they produce no obvi-
ous deficiencies such as those seen from the simpler model of Aungier (2000).
That is about all that can be said to justify their use.

Chapter 13
OTHER COMPONENTS
AND VARIATIONS
The preceding chapters have considered the basic aerodynamic design and analy-
sis functions relative to typical axial-flow compressor configurations. Blade row,
stage and compressor design and analysis have been considered in the context of
the compressor flow path from the first blade row to the last, including operation
at off-design speeds and mass flow rates. This chapter addresses other design and
analysis functions that are often required by the specific application. One obvi-
ous consideration for industrial axial-flow compressors is dictated by the need to
supply the compressor discharge flow to the process through a discharge pipe,
typically exiting in the radial direction. This requires the design and analysis of
an exhaust diffuser and flow collection system, which can have significant influ-
ence on the overall performance of the compressor. Another important consider-
ation is the use of adjustable inlet guide vanes and stators as a means of
improving the surge margin. Normally this provides a substantially better surge
margin at mass flows that are significantly less than the design flow than can be
achieved with simple variable speed operation. Indeed, variable geometry is an
essential requirement for many industrial compressor applications. The influ-
ence of surface roughness on performance is discussed to provide a basis for its
evaluation when required. Finally, the axial-centrifugal compressor is briefly dis-
cussed. This configuration includes a centrifugal compressor stage following a
series of typical axial-flow compressor stages. Under appropriate operating con-
ditions, this can offer significant cost and performance benefits.
NOMENCLATURE
A = passage area
Am = maximum, stall-free passage area
B = fractional area blockage
BBL = boundary layer fractional area blockage
BSEP = minimum fractional area blockage due to flow separation
b = passage width
cp = static pressure recovery coefficient
D = 2tanθC

Dm =
limiting value of D for low diffusion losses
d =
characteristic diameter for Reynolds number definition
dH =
hydraulic diameter
E =
diffusion efficiency parameter
e =
peak-to-valley surface roughness
erms =
root-mean-square surface roughness
F =
normalized stagger angle adjustment
H =
total enthalpy
h =
static enthalpy
IC =
passage curvature loss term
ID =
passage diffusion loss term
i =
adjustable blade row number
L =
diffuser meridional length and scroll/collector flow path length
m =
meridional coordinate
ṁ =
mass flow rate
N =
total number of adjustable blade rows
n =
exponent in power-law stagger adjustment distribution
P =
pressure
R =
ratio of stagger angle adjustments on successive stationary rows
Re =
Reynolds number
r =
radius
SP =
scroll/collector sizing parameter
U local blade speed = ωr
=
γ =
stagger angle
δ =
boundary layer thickness
ε =
hub-to-shroud radius ratio
θ =
polar angle
θC =
diffuser divergence angle
κm =
stream surface curvature
µ =
gas viscosity
ρ =
gas density
τw =
wall shear stress
φ =
flow coefficient = Cm /U; also, slope angle of mean stream surface
with the axial direction
φ0 = characteristic inlet flow coefficient
– = total pressure loss coefficient
ω
Subscripts
0 = parameter at a stage inlet

1 = parameter at the diffuser inlet
2 = parameter at the diffuser exit
3 = parameter at the scroll/collector full-collection plane
4 = parameter at the exit cone discharge flange
l = laminar condition
ex = exit cone parameter

Other Components and Variations • 311
r = fully rough wall condition
s = smooth wall value
sf = skin friction parameter
t = total thermodynamic condition or turbulent condition
tip = blade tip parameter
13.1 ADJUSTABLE BLADE ROWS
There are two important considerations when using adjustable stationary blade
rows: The number of adjustable blade rows to be used must be selected and the
distribution of the stagger angle adjustments among those rows must be estab-
lished. These decisions can be conveniently made using a performance analysis
such as that described in Chapter 9. It is relatively easy to include the capability
to impose adjustments to the base stagger angles for specific blade rows. It is
most convenient to specify those adjustments as a fraction of a reference stagger
angle adjustment, e.g., the adjustment to be made to the first adjustable blade
row. If that is done, performance predictions for a series of different adjustable
blade row settings can be accomplished by simply changing the value of the ref-
erence stagger angle adjustment.
The most obvious design strategy would be to select one specific setting of the
adjustable blades, such as the most extreme adjustment to be used, and optimize
the distribution of the adjustments among all adjustable rows. In practice, that is
a rather ineffective approach. Adjustable blade rows can produce rather extreme
incidence angles on both rotors and stators to substantially increase the uncer-
tainty of the performance analysis and the reliability of any estimate of the surge
limit based on that analysis. If an adjustment strategy is established to optimize
the most extreme adjustment, prediction of an overall map of adjustments is
likely to be totally lacking in credibility. The most common result is to find the
estimated surge margin deteriorating at adjustments even slightly different from
the optimized adjustment. The real problem lies with the performance predic-
tions for the optimized adjustment. Off-design performance analysis at extreme
incidence angles involves too much uncertainty for the optimized adjustments to
really be believable. The very precise stage matching chosen to enhance surge
margin is very unlikely to be achieved in the actual compressor. Hence the
designer who attempts such an optimization strategy is really exceeding the
capabilities of off-design performance analysis technology. In effect, that is an
excellent way to expose all of the weaknesses of the performance analysis, which
is hardly a good basis for design. Even if the performance analysis could accu-
rately model this situation, a surge line that achieves enhanced surge margin very
local to one adjustment setting would not be very useful.
A better approach is to establish general stagger-angle adjustment distribu-
tions for this purpose and evaluate the results with the performance analysis.
These general distributions can offer considerable flexibility and yet be very
unlikely to produce a misleading “optimum stage matching” distribution. Three
general distributions found to be effective for this purpose will be presented and

demonstrated here. If there are N adjustable blades, let Fi represent the adjust-
ment to the ith adjustable blade row in the form
∆γ i = ∆γ ref Fi (13-1)
where ∆γref is the reference adjustment, which will be assumed to be the adjust-
ment to the first adjustable row, such that F1 = 1. One useful distribution is a sim-
ple linear variation such that the adjustment to the stator following the last
adjustable stator would be zero, i.e.,
Fi = ( N + 1 − i) / N (13-2)
A simple variation on the linear distribution will be referred to as the power-law

distribution, given by
Fi = [( N + 1 − i) / N]n (13-3)
Another useful distribution is the fractional distribution, which simply adjusts

each blade row as a specified ratio, R, of the adjustment on the preceding
adjustable blade, i.e.,
Fi = R (i −1) (13-4)
Figure 13-1 illustrates these three stagger angle distributions for a case where
N = 7. The linear and fractional distributions are adequate for most applications,
but the power-law distribution provides some additional flexibility if needed.
When these models are incorporated into the performance analysis, investigation
of alternatives for adjustable geometry is rather simple. It is only necessary to
specify the adjustment distribution type, the number of adjustable blade rows
and the stagger angle adjustment on the first row. Simply varying the last param-
eter permits analysis of alternate settings of the adjustable rows.
The ten-stage axial-flow compressor design of Fig.10-38 will be employed to
illustrate the use of adjustable stationary blades. This is a repeating stage design
with inlet guide vanes. Figure 13-2 compares the variable speed performance
map from Fig. 10-38 with a predicted performance map for a seven-row linear
stagger angle distribution (inlet guide vane and six stators). It can be seen that
variable geometry offers a significant improvement in surge margin compared to
variable speed operation at flow rates less than the design flow.
Figure 13-3 illustrates the influence of the number of adjustable blade rows
when the linear adjustment distribution is used for this compressor. Increasing N
has the expected effect of greater flow capacity control. It can be noted that the
best surge margin appears to be associated with N = 7. Generally, there is an opti-
mum value of N to enhance the surge margin, typically around one-half of the
total number of stationary rows available for adjustment, excluding exit guide
vanes, which are ineffective for this purpose. In this case, there are eleven rows
available (inlet guide vane and ten stators), so an optimum N around 6 might be

FIGURE 13-1 Alternate Adjustment Strategies
FIGURE 13-2 Variable Geometry and Variable Speed

FIGURE 13-3 Influence of N with Linear
expected. This really follows from the mechanism by which variable geometry
improves surge margin. Basically, the adjustable blades close down the front
stages, which would normally stall at the lower flows, to shift the load to the rear
stages. Stated differently, adjustable blades reduce the load on the front stages to
permit operation at lower volume flow rates where the rear stages will operate
closer to their design conditions. Hence variable geometry in the rear stages is
seldom effective for this purpose.
Figure 13-4 illustrates the fractional adjustment distribution for a series of val-
ues of R. It can be seen that increasing R results in increased capacity control,
while yielding essentially identical surge lines for all values considered. But there
is a slight adverse effect on surge margin and on efficiency with increasing R,
suggesting that larger values would not be good for this compressor. As a general
rule, R ≈ 0.8 is usually a good choice for the fractional adjustment distribution.
Figure 13-5 illustrates the influence of N with the fractional distribution with
R = 0.8. Increasing N from 5 to 7 increases both capacity control and surge mar-
gin. But only two values of N are shown, since the surge estimate for the higher
value is based on stall in one of the adjustable blade rows. Hence, adding more
adjustable blade rows with the fractional adjustment distribution on this com-
pressor cannot further improve the surge margin. Comparing Figs. 13-3 and 13-5
shows that the fractional adjustment distribution can achieve a surge margin
improvement comparable to the linear style for this compressor.
It can be concluded that the use of adjustable stationary blade rows is a very
powerful method for achieving increased surge margin at volume flow rates less

FIGURE 13-4 Influence of R with Fractional
FIGURE 13-5 Influence of N with Fractional

than the design value. By including the provision to impose these adjustments
within the aerodynamic performance analysis, it is relatively simple to arrive at
an appropriate adjustment distribution style and the appropriate number of
adjustable blade rows to be used. This also makes it rather easy to generate a
predicted performance map to evaluate a range of adjustable blade row settings.
The alternative of accomplishing these adjustments manually on each blade row
can make this a very tedious process, with little likelihood of arriving at an opti-
mum choice.
13.2 THE EXHAUST DIFFUSER

The industrial axial-flow compressor normally supplies the compressed gas to
some process through an exit pipe flange. Often, the final destination will involve
redirecting the flow from the axial flow leaving the compressor to a pipe oriented
normal to the compressor axis. Hence the overall performance of the compressor is
strongly influenced by the performance of the exhaust-end components. Generally,
the flow is first diffused in an exhaust diffuser to recover some of the kinetic energy
in the form of static pressure. This may be an annular, axial exhaust diffuser such
as that shown in figure 13-6. Alternatively, curved diffusers may be used to redirect
the flow in the radial direction for collection by a scroll or collector, such as illus-
trated in figure 13-7. Consequently, a very wide variety of exhaust diffusers may be
employed, requiring a very general performance analysis technique.
One aerodynamic performance analysis method that can handle these various
configurations is the vaneless passage analysis of Aungier (1993, 2000). That
analysis has been extensively qualified for centrifugal compressor performance
FIGURE 13-6 Axial Exhaust Diffuser Geometry

FIGURE 13-7 Curved Exhaust Diffuser
analysis, where it has been used to treat the swirling flows in vaneless diffusers
and 180° crossover bends. With one minor modification, it is directly applicable
to the axial-flow compressor exhaust diffuser problem. The area requiring modi-
fication is the modeling of the end-wall boundary layer growth and the associ-
ated viscous blockage effects. For axisymmetric flow in centrifugal compressors,
the rather ideal form of the tangential velocity profile in the boundary layer
makes that profile an obvious choice to estimate boundary layer growth. But that
choice is not appropriate for an axial-flow compressor exhaust diffuser, which is
likely to have little or no tangential velocity component. Similarly, viscous block-
age effects are less critical in the centrifugal compressor. Cases where substantial
blockage may be induced by excessive rates of diffusion generally occur only in
the crossover bend, where performance is primarily described by total thermody-
namic conditions rather than static. If the total pressure loss is predicted with
sufficient accuracy, the overall performance predictions are relatively insensitive
to the viscous blockage prediction. Again, the axial-flow compressor exhaust dif-
fuser often involves quite different requirements. A simple dump collector may
be used to collect the flow following an axial exhaust diffuser, such that little or

none of the kinetic energy is recovered. Thus, static conditions can be extremely
important, requiring a more careful consideration of viscous blockage effects.
Fortunately, generalization of the centrifugal compressor vaneless passage per-
formance analysis proved to be relatively straightforward, since most of the fun-
damental models were already included for the purpose of total pressure loss
prediction. The modified analysis used for axial-flow compressor exhaust dif-
fusers is presented in this section.
The analysis is a mean-streamline or one-dimensional flow analysis with
wall friction and empirical corrections for diffusion and curvature effects. Fig-
ure 13-6 illustrates the mean streamline in a typical axial exhaust diffuser con-
figuration and the nomenclature used in the analysis. The governing equations
for one-dimensional flow with skin friction are developed in Aungier (1993,
2000) as
2π rbρ Cm (1 − B) = m
˙ (13-5)
d( rCθ )
bCm = − rCCθ cf (13-6)
dm
1 dP Cθ2 sin φ dCm CCm cf dI D
= − Cm − − − IC (13-7)
ρ dm r dm b dm
H = h + 12 C 2 (13-8)
The blockage, B, specifies the fraction of the passage area unavailable for the
inviscid through flow, due to viscous effects. The skin friction coefficient, cf, sup-
plies a correction for the effect of the wall shear stress, τw.
τw
cf = (13-9)
1
2
ρ C2
The terms ID and IC are introduced in Aungier (1993, 2000) to account for losses
due to diffusion and passage curvature, respectively. In other respects, this set of
equations follows directly from the governing equations for inviscid flow pre-
sented in Chapter 3, when simplified to time-steady, axisymmetric, one-dimen-
sional flow in a stationary coordinate system.
From classical two-dimensional diffuser technology (e.g., Reneau et al., 1966)
it is known that the low loss regime is closely related to the well-known diffuser
divergence angle, 2θC. This parameter can be generalized to annular diffusers by
2θ C = 2 tan −1[b1( A2 / A1 − 1) / (2L)] (13-10)
The nomenclature shown in Fig. 13-6, where L is the length of the mean stream-
line and A = 2πrb is the passage area, is used. Aungier (1993, 2000) introduces a
divergence parameter, D, defined as
D = 2 tan θ C = b1( A2 / A1 − 1) / L (13-11)

The experimental results in Fig. 8(b) of Reneau et al. (1967) show that diffusion
losses are low when D < Dm, where
Dm = 0.4( b1 / L)0.35 (13-12)
Aungier (1993, 2000) uses an analogy with the classical diffuser parameters, D
and Dm, but defined as local diffusion parameters, i.e.,
b dC
D=− (13-13)
C dm
Dm = 0.4 cos β ( b1 / L)0.35 (13-14)
Based on detailed comparisons with predicted and measured losses in a substan-

tial number of centrifugal compressor vaneless diffusers, an empirical model was
developed for a diffusion efficiency parameter, E, in the form
E = 1; D ≤ 0 (13-15)
E = 1 − 0.2( D / Dm )2 ; 0 < D < Dm (13-16)
E = 0.8 Dm / D ; D ≥ Dm (13-17)
Then the streamwise diffusion term in Eq. (13-7) is given by
dI D 1 dC
= −2( Pt − P )(1 − E) (13-18)
dm ρ C dm
It was necessary to check for excessive diffusion in the meridional direction as

well as the streamwise direction, again using an analogy with classical diffuser
technology. The maximum stall-free local passage area, Am, is estimated from
Am = A1[1 + 0.1925m / b1] (13-19)
which corresponds to a local divergence angle limit of 2θC ≈ 11°. The highly
swirling flows in centrifugal compressors required a slightly more conservative
limit of 9°, but that is considered unnecessary for the exhaust diffuser applica-
tion. A minimum value of ID is estimated when A > Am
I D ≥ 0.65( Pt − P )[1 − Am / A] / ρ (13-20)
This value is imposed as a lower limit on ID obtained by integrating Eq. (13-18).

The passage curvature, κm, is used to estimate IC as
IC = κ m ( Pt − P )Cm / (13ρ C ) (13-21)

dφ
κm = − (13-22)
dm
Aungier (1993, 2000) employs simple one-seventh power law velocity profiles as
the boundary layer approximation. This continues to be used, since a one-dimen-
sional flow analysis really does not provide an adequate basis for modeling
boundary layer shape factors in any reasonable fashion. As noted, it is not possi-
ble to rely on the wall friction effects on the tangential momentum balance as a
basis for estimating boundary layer growth in an exhaust diffuser. Instead, the
meridional velocity profile is used with a simple flat-plat boundary layer growth
model, including adjustment for variations in radius and the boundary layer edge
meridional momentum. Turbulent boundary layer growth along a flat plate can
be estimated from (Pai, 1957)
dδ
≈ 5cf (13-23)
dm
This is generalized to the annular passage with two end-wall boundary layers in
the form
2
d[rbρ Cm (2δ / b)]
= 10cf ρ rCCm (13-24)
dm
The radius and the boundary layer edge meridional momentum corrections in
this equation are similar to momentum thickness corrections of Eq. (3-43). This
follows from the approximation that the boundary layer shape factor is constant,
which requires that the ratio of the boundary layer thickness to the momentum
thickness be constant. The boundary layer thicknesses estimated from Eq. (13-24)
must be limited by the fully developed viscous flow profile condition, i.e., 2δ ≤ b.
The fractional area blockage due to the two end-wall boundary layers will be
designated as BBL. The boundary layer blockage for one-seventh power law pro-
files can be shown to be
BBL = (2δ / b) / 8 (13-25)
For the exhaust diffuser, it must also be recognized that viscous blockage may
increase substantially if the rate of diffusion becomes excessive, i.e., if the maxi-
mum stall-free local passage area, Am, is exceeded. If that occurs, it is assumed
that further diffusion of Cm is suppressed by a minimum blockage, BSEP, where
BSEP = 0 ; A ≤ Am (13-26)
BSEP = 1 − Am / A ; A > Am (13-27)
BSEP can be imposed as a lower limit on the blockage estimated from Eq. (13-25),
but it has been found that a smooth transition between these two values is
obtained from the following empirical equation:

2
B = 1 − (1 − BSEP )(1 − BBL + BSEP ) (13-28)
Equations (13-24) through (13-28) contain all of the modifications to the original
analysis of Aungier (1993, 2000) developed specifically for the exhaust diffuser
application.
Finally, solution of the governing equations requires a model for the skin fric-
tion coefficient. Following Aungier (2000), a generalized pipe friction factor
model is employed for this purpose. While that models fully developed pipe or
channel flow, it is applicable to boundary layers by replacing the pipe diameter
(or channel width) with 2δ. For this problem, the relevant Reynolds number is
2ρ Cδ (13-29)
Re =
µ
The other parameter required is the peak-to-valley surface roughness height, e.

Usually surface roughness is measured as a root-mean-square value, erms. The
two values can be approximately related by assuming a sine-wave variation for
the roughness to yield
e = erms / 0.3535 (13-30)
If Re is less than 2000, the laminar skin friction coefficient applies.
cf = cfl = 16 / Re (13-31)
When Re > 2000 and the wall is smooth (e = 0), the well-known log-law profile yields
 
1 2.51
= −2 log10   (13-32)
4cfs  Re 4c 
 fs 
In the limit when the flow is turbulent and the wall is fully rough,
1  e  (13-33)
= −2 log10  
4cfr  3.71( 2δ ) 
The experimental results of Nikuradse (1930), show that surface roughness

becomes significant when
Ree = (Re− 2000)e / (2δ ) > 60 (13-34)
Hence, a general turbulent skin friction coefficient is given by
cft = cfs ; Ree < 60 (13-35)

cft = cfs + ( cfr − cfs )(1 − 60 / Ree ); Ree ≥ 60 (13-36)
If Re ≥ 4000, cf is defined by Eq. (13-35) or (13-36). For 2000 < Re < 4000, the flow
is in the transition zone and the skin friction coefficient is approximated by
cf = cfl + ( cft − cfl )(Re / 2000 − 1) (13-37)
These empirical equations yield values in very close agreement with the experi-
mental results of Nikuradse (1930).
The performance analysis of the exhaust diffuser involves integrating Eqs.
(13-5) through (13-8) along the diffuser length, supported by the supplemental
relations provided in this section. Input specifications required include the pas-
sage geometry, mass flow rate and inlet values of the total thermodynamic condi-
tions, tangential velocity and normalized boundary layer thickness, 2δ / b. To
show at least a qualitative evaluation of the method, the experimental data for
two-dimensional diffusers of Reneau et al (1967) will be used. In general, two-
dimensional and annular diffusers perform in a reasonably similar manner as
long as appropriate dimensionless parameters are used in any comparison. In
particular, the diffuser divergence angle definition of Eq. (13-10) should be used.
The highest inlet blockage level results in Fig. 4(d) of Reneau et al. (1967) are the
most representative of exhaust diffusers. These show an optimum static pressure
recovery coefficient, cp, around L / b1 ≈ 0.8 and A2 / A1 ≈ 2.4, as well as a strong
variation of cp with both diffuser divergence angle and area ratio. The static pres-
sure recovery coefficient is defined as
cp = ( P2 − P1) / ( Pt1 − P1) (13-38)
The exhaust diffusers for the analysis were specified with a constant radius hub
wall and a conical shroud wall. The inlet boundary layer thickness was defined as
2δ / b = 0.4 to match the experimental inlet blockage of B1 = 0.05. The inlet tan-
gential velocity was set to zero, and the other inlet data were selected to produce
the Mach number and Reynolds number levels indicate in the reference. Figures
13-8 and 13-9 show comparisons of results that pass close to the optimum per-
formance condition and illustrate the influence of variations in both 2θC and
A2/A1. Clearly the performance analysis shows reasonable agreement with the
experimental results. Although this is a rather qualitative evaluation, it seems evi-
dent that the analysis adapted from centrifugal compressor technology should be
applicable to the evaluation of exhaust diffusers for axial-flow compressors. From
its use in centrifugal compressors, it is also known to be applicable to more gen-
eral diffuser configurations, such as the curved diffuser illustrated in Fig. 13-7.
13.3 THE SCROLL OR COLLECTOR
For industrial axial-flow compressors, the flow from the diffuser may be collected
in a scroll or a collector and discharged into an exit pipe. Figure 13-7 presents the
side-view schematic of a curved diffuser discharging into a scroll or a collector.

FIGURE 13-8 Influence of Area Ratio on cp
FIGURE 13-9 Influence of Divergence Angle on cp

Figure 13-10 shows the front view of a scroll, where the area increases in the cir-
cumferential direction. By contrast, a collector has a circumferentially uniform
passage area. Figure 13-11 is a schematic of an axial exhaust diffuser with a col-
lector similar to the arrangement used on the NASA 10-stage compressor
(Budinger and Thomson, 1952).
The performance of scrolls and collectors can be evaluated using models
developed for centrifugal compressors (Aungier, 2000). All flow conditions are
known at the entrance station, 2, from the exhaust diffuser exit conditions. The
next key location is designated as station 3, which is the cross-section where the
flow has been completely collected, as shown in Fig. 13-10. The flow is then dis-
charged through an exit cone to the exit flange, station 4. The velocity at stations
3 and 4 are computed from conservation of mass. For simplicity, the gas density
will be assumed to be constant in the scroll or collector. While that is not a nec-
essary assumption, Mach number levels are almost always sufficiently low to jus-
tify its use in estimating the loss coefficient. Hence
C3 = Cm2 A2 / A3 (13-39)
FIGURE 13-10 Side View of a Scroll

FIGURE 13-11 An Axial Diffuser and Collector
C4 = Cm2 A2 / A4 (13-40)
The overall total pressure loss coefficient for the scroll or collector is defined as
ω = ( Pt2 − Pt4 ) / ( Pt2 − P2 ) (13-41)
This loss coefficient is computed as the sum of four component loss coeffi-
cients. First, it is assumed that the meridional velocity head entering the scroll
or collector will be lost. The meridional component of the velocity will develop
into a secondary flow pattern in the scroll or collector, as illustrated in Fig.
13-11, and eventually dissipate as a loss. Hence the meridional loss coefficient
is given by
ω m = (Cm2 / C2 )2 (13-42)
The tangential component of velocity can be smoothly recovered in a properly

sized scroll. The ideal size is determined by requiring conservation of angular
momentum between stations 2 and 3, i.e.,
r3C3 = r2Cθ 2 (13-43)

A sizing parameter, SP, is defined as
SP = r2Cθ 2 / ( r3C3 ) (13-44)
SP = 1 is the ideal sizing condition. SP > 1 indicates an oversized scroll that is

attempting to overdiffuse the flow relative to the ideal condition. SP < 1 is an
undersized scroll that is attempting to accelerate the flow relative to the ideal
condition. The tangential loss coefficient is given by
1 r3C32
ωθ = [SP 2 − 1]; SP ≥ 1 (13-45)
2 r2C22
r3C32
ωθ = [SP − 1]2 ; SP < 1 (13-46)
r2C22
Equations (13-45) and (13-46) are the same as those used in Aungier (2000),
but Eq. (13-44) has been used to remove a singularity at SP = 0 in the original
reference. Unlike in the centrifugal compressor, Cθ2 = SP = 0 is a condition
likely to be encountered in an axial-flow compressor. In the case of a collector,
the equations for the tangential loss coefficient apply only to the full collection
station. When a constant passage area collector first starts to collect the flow,
the passage area is greatly oversized, such that the collector attempts to diffuse
the tangential velocity essentially to zero, which implies the local SP becomes
infinite. A corresponding tangential loss coefficient can be computed by com-
bining Eqs. (13-44) and 13-45) and taking the limit as SP approaches infinity.
The collector tangential loss coefficient is set to an average of the two extreme
values, i.e.,
ωθ r2Cθ22
(ωθ )coll = + (13-47)
2 2r3C22
A wall skin friction loss coefficient is given by
ω sf = 4 cf (C3 / C2 )2 L / dH (13-48)
L is the average path length of the flow in the scroll and dH is the mean hydraulic
diameter of the passage.
L = π ( r2 + r3 ) / 2 (13-49)
dH = 2A3 / π (13-50)
The conventional definition of hydraulic diameter is four times the passage area
divided by the wetted perimeter of the passage. That is used in Eq. (13-50),

assuming that the cross-section of the passage is circular and the mean pas-
sage area is one-half of A3. For a constant passage area collector, the relations
used are
L = π r3 (13-51)
dH = 4 A3 / π (13-52)
The skin friction coefficient is calculated in the same fashion as that for the
exhaust diffuser, but with the Reynolds number and surface roughness refer-
enced to dH rather than 2δ. Finally, an exit cone loss coefficient is given by
ω ex = [(C3 − C4 ) / C2 ]2 ; C3 > C4 (13-53)

ω ex = 0 ; C3 ≤ C4 (13-54)
The overall total pressure loss coefficient and exit total pressure for the scroll or
collector are given by
ω sc = ω m + ωθ + ω sf + ω ex (13-55)
Pt4 = Pt2 − ω sc ( Pt2 − P2 ) (13-56)
Conservation of energy and an appropriate equation of state from Chapter 2

yields H4 = H2 = H(Tt2, Pt2), from which Tt4 = T(Pt4, H2). Then a simple mass bal-
ance on A4 yields the discharge velocity and static conditions. The scroll or col-
lector static pressure recovery coefficient is given by
cpsc = ( P4 − P2 ) / ( Pt2 − P2 ) (13-57)
An overall exhaust system loss coefficient and static pressure recovery coeffi-
cient for the diffuser and the scroll or collector combined can also be calcu-
lated, i.e.,
ω = ( Pt1 − Pt4 ) / ( Pt1 − P1) (13-58)

cp = ( P4 − P1) / ( Pt1 − P1) (13-59)
One useful application of the performance analysis results is to define a dis-

charge loss coefficient for use by the axial-flow compressor performance analysis
of Chapter 9. That analysis basically predicts the compressor’s performance up to
the diffuser entrance, i.e., the station 1 data in the notation in this chapter. A dis-
charge pressure can be computed as
Pdis = Pt1 − ω dis ( Pt1 − P1) (13-60)

The appropriate definition of the discharge loss coefficient depends on whether

a total-to-total or a total-to-static evaluation is appropriate, i.e., whether or not
the kinetic energy at the discharge is valuable to the application served by the
compressor. In the case of total-to-total evaluation, Eq. (13-58) provides the
appropriate definition. In the case of total-to-static evaluation, the appropriate
discharge loss coefficient definition is
ω dis = 1 − cp (13-61)
where cp is given by Eq. (13-59).
13.4 REYNOLDS NUMBER AND SURFACE ROUGHNESS

EFFECTS
The exhaust diffuser performance analysis provides a good basis to discuss meth-
ods to include the effects of surface roughness on axial-flow compressor per-
formance. This is often an important consideration for industrial centrifugal
compressors, which accounts for its detailed treatment in the exhaust system
analyses adapted from centrifugal compressor technology. Surface finish effects
are much less likely to be important in typical axial-flow compressor applica-
tions. Hence, Reynolds number correction models from the literature seldom
considered it [e.g., the models in Chapter 6 and in Wassell (1968)]. But it can
become significant for applications involving very high Reynolds numbers or
very poor surface finishes. Fortunately, there is a fairly simple method to exter-
nally impose a correction for surface roughness on any Reynolds number correc-
tion model. This can be easily accomplished by limiting the Reynolds number
used in the correction to values for which the surfaces involved are hydraulically
smooth. A reasonable approximation for this purpose is to limit the Reynolds
number used in the correction by
Red ≤ 60d / e (13-62)
where d is the characteristic length dimension used to define the Reynolds num-
ber. Figure 13-12 applies this simple method to correct the skin friction coeffi-
cient from Eq. (13-32) and compares it to results from Eqs. (13-30) through
(13-37). Clearly, this simple correction is in rather good agreement with the
more precise empirical model derived from the pipe friction factor charts of
Nikuradse (1930).
13.5 THE AXIAL-CENTRIFUGAL COMPRESSOR
The axial-centrifugal compressor consists of a series of axial-flow compressor

stages followed by a centrifugal compressor stage. Under appropriate circum-
stances, this can offer definite advantages. Centrifugal compressor design and
analysis is covered in depth in Aungier (2000). The discussion here is limited to

FIGURE 13-12 Simple Surface Roughness Correction
the concept of combining the two compressor types. The relative merits and
characteristics of the axial-flow compressor and the centrifugal compressor were
discussed briefly in Chapter 1. To review, the following are key differences
between these two types of compressors:
• The axial-flow compressor can achieve significantly higher mass flow

rates per unit frontal area.
• The centrifugal compressor can achieve significantly larger pressure
ratios per stage.
• The centrifugal compressor is generally the more rugged and lower-cost
compressor type.
• The axial-flow compressor generally offers higher efficiency, although the
centrifugal compressor has become rather competitive in recent years.
• The centrifugal compressor offers stable operation over a much wider
mass flow range.
• The axial-flow compressor offers a much steeper pressure-mass flow
characteristic.
• The centrifugal compressor is much better suited to matching with a
scroll or collector at the discharge.
These differences need definite qualification when considering a combination

of the two types in a single compressor. Axial-flow and centrifugal compressor

stages achieve their optimum performance at quite different dimensionless per-

formance parameters. This difference is typically expressed in terms of specific
speed or the characteristic stage inlet flow coefficient, φ0. The latter should not be
confused with the definition of flow coefficient, φ, used earlier in this book. Here,
the characteristic stage inlet flow coefficient is defined as
2 (13-63)
φ0 = m
˙ / (π rtipUtipρt0 )
Figure 13-13 shows stage efficiency levels that should be achievable with opti-
mized centrifugal compressor stages as a function of φ0. It is adapted from gen-
eralized performance charts in Aungier (2000) to emphasize simple radial
discharge centrifugal stages. Operation at higher values of φ0 is certainly possi-
ble, but is likely to require a mixed-flow design to achieve adequate efficiency
levels. A mixed-flow design is a compromise between axial-flow and radial-flow
styles such that the flow exits the impeller with velocity components in both the
axial and radial directions. At sufficiently high values of φ0, an axial-flow stage
will be required. A meaningful extension of the chart in Fig. 13-13 to cover
mixed flow and axial-flow stages is not available. Balje (1981) presents a specific
speed chart intended to include the three compressor stage types. But his use of
total-to-static efficiency severely limits the value of his chart as a guide for
multi-stage compressor design. In Chapter 10, it was noted that many investiga-
tors now favor relatively low stage flow coefficients, φ, around 0.5. The flow
FIGURE 13-13 Centrifugal Stage Performance

coefficients φ0 and φ can be approximately related using the stage hub-to-shroud

radius ratio, ε.
φ0 ≈ 12 (1 − ε 2 )(1 + ε )φ (13-64)
ε = rh / rs (13-65)
Typical values of ε for front stages in a compressor are in the 0.4 to 0.6 range.
From Eq. (13-64) this suggests that φ0 in the range of about 0.25 to 0.3 are about
the lowest values to be expected for optimized axial-flow stages. The stage effi-
ciency is expected to be essentially unaffected by variation of ε from 0.4 to 0.6, yet
it results in substantial variation in φ0. Indeed, φ0 is rather ambiguous as a guide
to achievable efficiency levels for axial-flow stages. In the rear stages of an axial-
flow compressor, ε may become large enough to produce values of φ0 in the range
normal for centrifugal stages. But in that case, the performance of an axial-flow
stage is likely to be far from optimum due to aspect ratio effects associated with
the very short blades.
It is clear that axial-flow stages generally operate at values of φ0 that are con-
siderably higher than centrifugal stages. In addition, a centrifugal stage substi-
tuted for an axial-flow stage will normally operate at a lower value of φ0 than the
axial-flow stage it replaced. This follows from the fact that rtip and Utip will typi-
cally be greater for the centrifugal stage. Consequently, it may be beneficial to
replace some of the rear axial-flow stages in an axial-flow compressor with a cen-
trifugal stage. The higher stage pressure ratio offered by the centrifugal stage
may permit replacement of several axial-flow stages to substantially reduce cost.
If aspect ratio effects are significant in the replaced axial-flow stages, perform-
ance may also be improved. It is rather easy to evaluate the potential benefits
from this approach. Simply compute φ0 in the rear stages and estimate rtip of a
replacement centrifugal stage to compute φ0 for the centrifugal stage. If the
upstream axial-flow stages have reduced φ0 to a value suitable for a centrifugal
stage, the replacement may be appropriate. The expected centrifugal stage effi-
ciency can be estimated from Fig. 13-13, or by using the more refined charts or
the performance analysis in Aungier (2000).
Some industrial axial-flow compressor designers believe that a final centrifu-
gal stage is beneficial, even without obtaining a cost or performance benefit
directly from the substitution. This may involve using a rather modest pressure-
ratio centrifugal stage primarily to reduce the exhaust system losses. It is clear
that a scroll or collector following typical axial-flow stages cannot be well-
matched, since the sizing parameter of Eq. (13-44) will always be close to zero.
Hence the scroll or collector will always be grossly undersized. There is little that
can be done in terms of scroll or collector design except to minimize the harm
done to the performance by the tangential loss and exit cone loss. Hence the
designer’s ability to minimize the exhaust system loss is primarily limited to the
exhaust diffuser design. It can be argued that a centrifugal stage can efficiently
turn the flow to the radial direction to permit the design of a radial diffuser with
a well-matched scroll or collector. No doubt there is some merit to that approach,
although it may be difficult to quantify. If the axial length available is limited,
e.g., to favor rotor dynamics, the benefit may be more obvious. For example, if

space limitations require reducing the length of the axial diffuser of Fig. 13-11 or
replacing it with the curved diffuser of Fig. 13-7, the alternative of a final cen-
trifugal stage may be easier to justify.
EXERCISES
13.1 Flow exits an exhaust diffuser with Cθ = 0. What will be the mini-
mum possible value of the loss coefficient for a scroll or collector fol-
lowing the diffuser? What options do you have to minimize this loss
coefficient? Of the diffuser types shown in Figs. 13-7 and 13-11,
which one yields the lowest loss in the scroll/collector?
13.2 A curved diffuser similar to that of Fig. 13-7 is replaced by a very
low pressure-ratio centrifugal stage with a vaneless diffuser. The
diffuser has an exit passage area and radius identical to the original
diffuser and the scroll sizing parameter is 1. This substitution
reduces the scroll total pressure loss coefficient from 1.1 to 0.4.
What can you conclude about the overall exhaust system losses for
the two configurations? Which contributions to the overall loss are
likely to be reduced by this substitution? What steps are needed to
justify this substitution?
13.3 Assuming the diffusers of Figs. 13-7 and 13-11 have the same inlet
and discharge areas and Cθ = 0, which diffuser is likely to have the
lower loss? Under what circumstances would you choose the higher
loss configuration? Would your answers be different if Cθ > 0?
13.4 A general Reynolds number correction model is applied to correct
low Reynolds number test performance to high Reynolds number
operating conditions. Surface roughness is found to be significant,
so the approximate correction of Section 13.4 is imposed. How con-
fident can you be in the results? What areas of uncertainty need to
be considered?
13.5 Polishing blade and end-wall surfaces to improve surface finish can
add significant cost to the manufacturing process. How can you use
the skin friction coefficient model of Eqs. (13-30) through (13-37) to
determine whether the added cost is justified? How would you adapt
this procedure when the characteristic length is not a passage width
or diameter (e.g., blade chord Reynolds number)?

ANSWERS TO THE EXERCISES
1.1 Flow parameter:
˙ / ( ρt0 kRTt0 ) ∝ mRT

Q / at0 ∝ m ˙ t0 / ( Pt0 kRTt0 ) ∝ m
˙ RTt0 / k / Pt0
N / at0 = N / kRTt0
Speed parameter:
1.2 If β1 = 0, then Cθ1 = 0 and
ψ = Cθ 2 / U
R = 1 − Cθ 2 / (2U ) = 1 − ψ / 2
So if R = 0.5, then ψ = 1 is the only acceptable value. From Eq. (1-

24), the rotor discharge relative flow angle must be zero. Since the
absolute flow angles into and out of the stage are zero, no inlet or
exit guide vanes will be required.
1.3 DB = 1.2DA and AB = 1.44AA. Therefore,
NB = NA / 1.2
QB = 1.44QA
For equivalence with the original compressor at 3,600 rpm, the

scaled compressor must operate at 3,000 rpm and will supply 44%
more flow capacity than the original.
1.4 For a 20% increase in flow capacity, a 20% increase in A0 is needed,

which requires a scale factor of √1.2 . Hence the speed must be
reduced by a factor of √1.2 .
2.1 Equations (2-19), (2-27), (2-28) and (2-53) combine to yield
a12 = kRT1 = cp ( k − 1)T1

2cpT1 = 2a12 / ( k − 1)
Tt1 / T1 = 1 + ( k − 1)Cz21 / (2a12 ) = 1 + k −1 M2

2 1
Hence, from Eqs. (2-28) and (2-55),
( a1 / at1)2 = 1 / (1 + k − 1 M12 )
2
( ρ1 / ρt1)k −1 = 1 / (1 + k −1 M2 )
2 1
2.2 From Eq. (1-17),
H2 = H1 + UCθ 2
For a thermally and calorically perfect gas, Tt is a function of only H.

Hence Eq. (2-52) yields
Tt2id = Tt2 = Tt1 + ( H2 − H1) / cp
In the absence of losses, the flow is isentropic and Eq. (2-55) yields
k
Pt2id = Pt1(Tt2id / Tt1) k −1
Equation (2-68), using the inlet velocity pressure as relevant base

kinetic energy term, requires
Pt2 = Pt2id − ω ( Pt1 − P1)
2.3 For complete Mach number equivalence, the discharge Mach number
and flow angle must be equivalent, i.e., Cθ2/a2 and Cz2/a2 must be
equivalent. As shown in Exercise 2.1, when Mach number equivalence
exists, this requirement can be restated to require equivalence on
Cθ2/at2 and Cz2/at2. Mach number equivalence at the inlet requires
equivalence on U/at1. From Exercise 2.2,
∆H = H2 − H1 = UCθ 2
∆H / at21 = (U / at1)(Cθ 2 / at2 )at2 / at1
( at2 / at1)2 = Tt2 / Tt1 = 1 + ( kR / cp )( ∆H / a12 )
2
Hence equivalence on ∆H/at1 is required to produce equivalence on
Cθ2/at2, Tt2/Tt1 and at2/at1. Neglecting losses, Eq. (2-55) requires

Answers To the Exercises • 335
k
Pt2 / Pt1 = (Tt2 / Tt1) k −1
1
ρt2 / ρt1 = (Tt2 / Tt1) k −1
so equivalence on Pt2/Pt1 and ρt2/ρt1 is also achieved. Hence
˙ / ( ρt2at2 A2 ) = [m
m ˙ / ( ρt1at1A1)][ρt1at1A1 / ( ρt2at2 A2 )]
N / at2 = N / at1[at1 / at2 ]
Since the last terms on the right-hand side of the previous two equa-
tions satisfy equivalence, Mach number equivalence at the inlet of the
blade row will produce Mach number equivalence at the discharge.
2.4 Denote the base case with the original working fluid by subscript B,
and the new case by subscript N. Let Tt2B and ρt2B be the rotor exit
conditions for the base case. For mass flow and speed equivalence at
the rotor inlet,
˙ / ( ρt kRT ) = constant
m
N / kRT = constant
Since the inlet conditions and R are identical for the two fluids, the
mass flow and speed for the new case must be
˙N =m
m ˙ B 1.38 / 1.4
NN = NB 1.38 / 1.4
For speed equivalence at the rotor exit,
NN = NB (1.38Tt2N ) / (1.4Tt2 B )
Hence Tt2N = Tt2B is required to satisfy the equivalent speed condi-

tion at both the rotor inlet and exit. Mass flow equivalence at the
rotor exit with Tt2N = Tt2B requires
˙ B 1.38 / 1.4 ( ρt2N / ρt2 B )

˙N =m
m
Thus, to achieve mass flow equivalence at both locations, ρt2N = ρt2B

is required. But Eq. (2-55) with Tt2N = Tt2B requires
( ρt2 B / ρt1)0.4 = Tt2 B / Tt1 = ( ρt2N / ρt1)0.38

ρt2N / ρt1 = ( ρt2 B / ρt1)0.4 / 0.38

Hence the two conditions for Mach number equivalence cannot be

satisfied at both the rotor inlet and exit. Since Mach number equiva-
lence at both stations cannot be achieved for the same speed and
mass flow, complete Mach number equivalence through the com-
pressor is not achieved in this case.
2.5 Using Eq. (2-42), the temperature at which liquid will begin to form
can be estimated from
Tc 3  P
= 1− log10  
T 7(1 + ω )  Pc 
For P = 200 kPa, T > 247.6° K is required to avoid liquid phase

formation.
2.6 Use Eq. (2-56) to compute the temperature ratio, TR, from pressure
ratio, PR, and ηad. Use Eq. (2-57) to compute the polytropic efficiency.
Hence
PR = 3.0: TR = 1.4338, ηp = 87.11%

PR = 5.0: TR = 1.6868, ηp = 87.95%
2.7 Use Eq. (2-56) to compute the stage temperature ratio, TR, from the
stage pressure ratio and stage ηad. This yields a stage temperature
ratio of 1.032477. For the three-stage compressor, the pressure ratio is
(1.1)3 = 1.331 and the temperature ratio is (1.032477)3 = 1.10063.
Then Eq. (2-56) yields the overall adiabatic efficiency of 84.59%. If the
efficiencies are all polytropic, the stage temperature ratio, TR, can be
computed using Eq. (2-57). Then Eq. (2-57) can be used to compute
the overall compressor efficiency as
ηp = k −1 ln(1.13 ) / ln(T 3 ) = k −1 ln(1.1) / ln(T )

k R k R
So the overall compressor efficiency is identical to the individual

stage efficiency.
2.8 From Eqs. (2-63) and (2-68), the discharge static and total pressures
are
Pd = 200 + 0.6(30) = 218

Ptd = 230 − 0.1(30) = 227
For a thermally perfect gas with no work or heat transfer, Ttd = Tti =
300. Equation (2-55) yields the inlet and discharge static temperatures.
Ti = 300(200 / 230)(0.4 /1.4) = 288.26

Td = 300(218 / 227)(0.4 /1.4) = 296.55
For adiabatic reversible (isentropic) flow, the discharge temperature

would be
Td,rev = 300(218 / 230)(0.4 /1.4) = 295.44
Hence the diffuser efficiency is given by Eqs. (2-52) and (2-62)
cp (295.44 − 288.26)
ηdiff = = 86.6%
cp (296.55 − 288.26)
3.1 From Eq. (3-29) for axisymmetric, time-steady flow,
∂( rWθ + ω r 2 ) ∂rCθ
= =0
∂m ∂m
Similarly, Eq. (3-25) requires
∂I
=0
∂m
Inserting these results into Eq. (3-28) yields
∂s
=0
∂m
3.2 Direct substitution of Wθ = Wm tanβ′ into Eq. (3-30) yields
Wm ∂Wm Wm2 tan β ′ ∂r tan β ′

+
cos2 β ′ ∂n r ∂n
2 ∂I ∂s
+κ mWm + 2ω Wm tan β ′ cos φ = −T
∂n ∂n
For the stationary coordinate system, ω = 0, W → C, β′ → β and I →

H. Hence
2
Cm ∂Cm Cm tan β ∂r tan β 2 ∂H ∂s
+ + κ mCm = −T
cos β ∂n
2 r ∂n ∂n ∂n
3.3 From the definition of the dot product and Eq. (3-61),
r r
V = V ⋅ V = Vm2 + Vn2 + Vθ2

r ∂V r ∂V r V ∂V r
1 ∇V 2 =V em + V en + eθ
2 ∂m ∂n r ∂θ
3.4 For the stated conditions, Eqs. (3-21) and (3-27) combine to yield
∂brρ Wm
=0
∂m
Equation (3-22) simplifies to
∂Wm sin φ 1 ∂P
Wm − [Wθ + ω r]2 = −
∂m r ρ ∂m
Equation (3-23) and the definition of the angle φ yield
∂( rWθ + ω r 2 )
=0
∂m
and Eq. (3-25) simplifies to
∂I
=0
∂m
3.5 From the uniform flow assumption, the total mass flow is
b δ b
∫ rρ udy = r ρeue (b − 2δ ) + ∫ rρ udy + ∫ rρ udy

0 0 b −δ
where –r is the average radius. Using the standard boundary layer

approximation that r is constant for the last two terms and Eq.
(3-35), yields
b
˙ / (2π ) = ∫ rρ udy = r ρ eue ( b − 2δ * )
m
0
Similarly, the momentum flux is
b δ b
∫ rρ u dy = r ρeue (b − 2δ ) + ∫ rρ u dy + ∫
2 2 2
rρ u2dy
0 0 b −δ
and Eq. (3-38) yields

∫ rρ u dy = r ρeue (b − 2δ
2 2 *
− 2θ )
0
3.6 Conservation of mass for incompressible flow before and after mixing
requires
rbρ eumix = r ρ eue ( b − 2δ * )
umix = ue (1 − 2δ * / b)
Conservation of momentum with constant static pressure requires
2
rb( Pe + ρ eumix ) = rbPe + r ρ eue2b[1 − 2(δ * + θ ) / b]
Introducing the incompressible relation for total pressure,
2
Pt,mix + 12 ρ eumix = Pt,e + ρ eue2 [ 12 − 2(δ * + θ ) / b]
Introducing umix from the mass balance equation
Pt,mix + 12 ρ eue2 (1 − 2δ * / b)2 = Pt,e + ρ eue2 [ 12 − 2(δ * + θ ) / b]
Pt,e − Pt,mix = ρ eue2 [ 12 (1 − 2δ * / b)2 − 1

2
+ 2(δ * + θ ) / b]
Pt,e − Pt,mix = 1
2
ρ eue2 [(2δ * / b)2 + 4θ / b]
5.1 From Eqs. (5-1), (5-2), (5-3) and (5-9)
∂ ∂ξ ∂ ∂η ∂ 1 ∂ tan β ∂S ∂
= + = −
∂m ∂m ∂ξ ∂m ∂η cos β ∂ξ S ∂m ∂η
∂ ∂ξ ∂ ∂η ∂ r ∂
= + =
∂θ ∂θ ∂ξ ∂θ ∂η S ∂η
Substitution of these derivatives into the steady form of Eq. (3-21)

using Eq. (3-27) yields
1 ∂brρ Wm tan β ∂brρ Wm 1 ∂brρ Wθ

− + =0
cos β ∂ξ S ∂η S ∂η
Converting to finite-difference form, using the difference approxima-

tion form given in Eqs. (5-35) and (5-36) and multiplying through by
4∆ξ∆η, yields

2∆η
[( brρ Wm )ξ + ∆ξ ,η − ( brρ Wm )ξ − ∆ξ ,η ]
cos β
2∆ξ rb tan β
− [( ρ Wm )ξ ,η + ∆η − ( ρ Wm )ξ ,η − ∆η ]
S
2∆ξ rb
+ [( ρ Wθ )ξ ,η + ∆η − ( ρ Wθ )ξ ,η − ∆η ] = 0
S
where the overbar designates values at point (ξ, η). Noting that 2∆m
–
= 2∆ξcosβ ,
2∆η S
[( brρ Wm )ξ + ∆ξ ,η − ( brρ Wm )ξ − ∆ξ ,η ]
r
−2b ∆m tan β [( ρ Wm )ξ ,η + ∆η − ( ρ Wm )ξ ,η − ∆η ]
+2b ∆m[( ρ Wθ )ξ ,η + ∆η − ( ρ Wθ )ξ ,η − ∆η ] = 0
Checking the result for the control volume in Fig. 5-4, it can be seen
that the second term in this difference equation does not precisely
–
balance mass. The term tan β should be evaluated at each boundary,
instead of using a mean value. Thus there will be an inherent error in
the numerical approximation to the continuity equation.
5.2 The required Taylor series are
ψ (m + ∆m) = ψ (m) + ψ ′(m)∆m + 12 ψ ′′(m)( ∆m)2 + 16 ψ ′′′(m)( ∆m)3 + ⋅ ⋅ ⋅
ψ (m − ∆m) = ψ (m) − ψ ′(m)∆m + 12 ψ ′′(m)( ∆m)2 − 16 ψ ′′′(m)( ∆m)3 + ⋅ ⋅ ⋅
Subtract the second equation from the first and divide by 2∆m to
obtain
ψ (m + ∆m) − ψ (m − ∆m)
ψ ′( m ) = + O[( ∆m)2 ]
2∆m
Hence the difference approximation is of second-order accuracy in

∆m. Add the two Taylor series and divide by (∆m)2 to obtain the sec-
ond derivative difference approximation, which is also of second-
order accuracy in ∆m.
5.3 The required Taylor series are
ψ (m + 2∆m) = ψ (m) + 2ψ ′(m)∆m + 2ψ ′′(m)( ∆m)2 + 43 ψ ′′′(m)( ∆m)3 + ⋅ ⋅ ⋅
ψ (m + ∆m) = ψ (m) + ψ ′(m)∆m + 12 ψ ′′(m)( ∆m)2 + 16 ψ ′′′(m)( ∆m)3 + ⋅ ⋅ ⋅

Multiply the second equation by 4, subtract the first equation from it

and divide by 2∆m to obtain
4ψ (m + ∆m) − 3ψ (m) − ψ (m + 2∆m)

ψ ′( m ) = + O[( ∆m)2 ]
2∆m
which is also of second order accuracy in ∆m.
5.4 The required Taylor series is
u(m + ∆m) = u(m) + u′(m)∆m + 12 u′′(m)( ∆m)2 + 16 u′′′(m)( ∆m)3 + ⋅ ⋅ ⋅
Solve for the first derivative to obtain
u(m + ∆m) − u(m)

u′(m) = + O( ∆m)
∆m
Hence this difference approximation is of first-order accuracy in ∆m.
5.5 Under the conditions stated, the flow upstream of the blade at near
steady-state conditions will be approximately axisymmetric and time
steady. From Eq. (5-72), the quantity SbρWm will be approximately
constant, so its second derivative in Eq. (5-102) will be approximately
zero. Similarly, the quantities I and rCθ = rWθ +ωr2 will be approxi-
mately constant, as seen from Eqs. (3-25) and (3-29), so their second
derivatives in Eqs. (5-103) and (5-104) will be approximately zero.
Had the quantity Sbρ been moved outside of the second derivative in
Eq. (5-102), the stabilizing term would no longer be approximately
zero unless Sb is constant. In contrast, if Sbρ were moved inside of the
second derivatives in Eqs. (5-103) and (5-104), those stabilizing would
cease to be near zero unless Sb is constant. The derivative terms in the
stabilizing terms are chosen as the quantities most likely to be nearly
constant at near steady-state conditions. Although these terms should
normally be small, any numerical stability problem will cause them to
become large, basically introducing as much numerical damping as
needed for a stable solution.
5.6 (a) In general, estimate θ from the upstream value using Eq. (5-106)
with δ* and τw set to zero for the purpose of this initial guess. If the
previous station is the leading edge, where θ = 0, that procedure
won’t work. In that case, a reasonably safe initial guess could be
obtained by setting θ from an assumed value of Reθ that is well
below transition, e.g., 50.
(b) The steps in the iteration to solution at each station might be as

follows:

• Compute b0 from Eq. (5-121)

• Compute K from Eq. (5-122)
• Compute Λ from Eq. (5-124)
• Compute δ′ from Eq. (5-115)
• Compute δ* from Eqs. (5-116) through (5-120)
• Compute τw from Eqs. (5-109) and (5-123)
• Recompute θ from Eq. (5-106)
• Repeat above steps until converged
5.7 (a) Estimate θ from the upstream value using Eq. (5-106) with δ* and τw
set to zero for the purpose of this initial guess. Estimate (δ – δ*) from
Eq. (5-127) with E set to zero for the purpose of this initial guess.
(b) The steps in the iteration to solution at each station might be as

follows:
• Compute H1 from Eq. (5-128)

• Compute Hk from Eq. (5-132)
• Compute E from Eq. (5-133)
• Compute H and δ* from Eqs. (5-131) and (5-129)
• Compute cf from Eqs. (5-134) and (5-135)
• Recompute θ and (δ – δ*) from Eqs. (5-106) and (5-127)
• Repeat above steps until converged
6.1 For Eq. (6-8), i* = α* + γ – κ1 = α* – θ / 2 = f(σ, θ).

For Eq. (6-12), i* = f(β1*,σ,θ). But since β1* = κ1 + i*, the true independ-
ent variables can be expressed as i* = f(κ1,σ,θ)
6.2 For Eq. (6-8), β1* = α* + γ = f(σ,θ,γ).

For Eq. (6-12), β1* = i* + κ1 = f(σ,θ,κ1) = f(σ,θ,γ).
6.3 Both Fig. 6-2 and 6-14 are from the same reference, and use the same
definition of α*. Hence if the positive and negative stall incidence
angles are to be properly computed, Eq. (6-8) must be solved for α*.
7.1 Convert the data at base (constant radius) blade angle data to κ1 and
κ2. Interpolate for c and tb / c at the mean radius, κ1 at the inlet radius
and κ2 at the discharge radius. Estimate the stream surface angle from
tan φ = ( r2 − r1) / ( z2 − z1)
The effective geometry data in the stream surface are
c → c / cos φ
s = π ( r2 + r1) / Z
tb / c → cos φ tb / c
tan κ1 → cos φ tan κ1
tan κ 2 → cos φ tan κ 2

7.2 Calculation of the meridional gradient of Wm with simple finite differ-

ences will usually be based on data at points on opposite sides of
blade rows. This may cause the computed gradient to be meaningless,
since Wm may be strongly influenced by discontinuous changes in the
swirl velocity, etc., imposed by the blades.
7.3 A choke condition in an axial-flow compressor will almost always be

caused by choke in a blade row, with an associated abrupt increase in
loss. The approach to the Mach number limit in Eq. (7-29) is nor-
mally just an indication that blade row choke has occurred. In the
unlikely event of a true annulus choke, this should still be reasonably
accurate, since hub-to-shroud gradients in Wm are seldom extreme in
an axial-flow compressor. If the through-flow analysis were to be
applied within the blade passage, the more rigorous choke criterion
might be necessary.
7.4 The approximation should be quite accurate on interior stream sur-

faces, assuming a reasonable number of stream surfaces are used.
However, it is somewhat of an extrapolation for the end-wall surfaces.
Even there, it will offer acceptable accuracy unless the meridional
gradient of Wm becomes excessive near the end-walls. By the defini-
tion, a positive mass flow rate passes between adjacent stream sur-
faces, so the singularity will occur only on end-walls, unless it is
caused by numerical errors in the early iterations of the solution
process. If an end-wall boundary layer solution is included, a singu-
larity in the inviscid through-flow analysis should be suppressed by
the abrupt increase in end-wall blockage prediction when the end-
wall velocity becomes small.
8.1 From Eq. (8-11), the mass flow rate in a boundary layer is given by
∫ ρVmdy =ρeVme (δ − δ1 )
*
Hence a mass balance combining the two incoming flows yields
2πr( ρ eVme )in (δ − δ1* )+ = 2π ( rρ eVme )in (δ − δ1* )in + m

˙ leak
which results in Eq. (8-58).
8.2 From Eqs. (8-11) and (8-12), the meridional momentum flux in a
boundary layer is given by
∫ ρVmdy =ρeVme (δ − δ1 − θ11)

2 2 *
Hence the leakage flow contributes no meridional momentum, so

2
2πr( ρ eVme )in (δ − δ1* − θ11)+ = 2π ( rρ eVme
2
)in (δ − δ1* − θ11)in
Combining with the result in the previous exercise yields Eq. (8-60).
8.3 From Eqs. (8-12) and (8-13), the tangential momentum flux in a
boundary layer is given by
∫ ρVmVθ dy =ρeVmeVθe (δ − δ1 − θ12 )

*
Hence to balance tangential momentum of the two incoming flows,
2πr( ρ eVmeVθe )in (δ − δ1* − θ12 )+

= 2π ( rρ eVmeVθe )in (δ − δ1* − θ12 )in + (mU
˙ )leak
Combining with the result of Exercise 8.1 yields Eq. (8-62) for the
case of leakage flow entering the boundary layer.
8.4 The process is the same as for the previous three exercises, except for
the tangential momentum and direction of the leakage flow.
8.5 Substituting the power-law profiles into the designated equations
yields
δ
 y 
n
δ
δ1* = ∫ 1 −    dy = δ −
 δ   n +1
0 
δ
 y 
n 2n
 y n
θ11 = ∫   −    dy = δ
δ  δ   ( n + 1)(2n + 1)
0 
which combine to yield Eq. (8-20).
δ n n+ m 
 y  y m
θ12 = ∫   −    dy = δ
δ  δ   ( n + 1)( n + m + 1)
0 
which yields Eq. (8-21).
9.1 Trapezoidal-rule integration of the uncorrected data between stream

surfaces 1 and 3 yields
I = 12 [( ∆pt′ )1 + 2( ∆pt′ )2 + ( ∆pt′ )3 ]∆m

˙

Integration of the corrected data yields
Ic = 12 [( ∆pt′ )1,c + 2( ∆pt′ )2,c + ( ∆pt′ )3,c ]∆m

˙
Introducing Eq. (9-10),
Ic = 12 [2( ∆pt′ )2,c − ( ∆pt′ )3,c + 2( ∆pt′ )2,c + ( ∆pt′ )3,c ]∆m
˙ = 2( ∆pt′ )2,c ∆m
˙
From Eq. (9-9),
Ic = 12 [( ∆pt′ )1 + 2( ∆pt′ )2 + ( ∆pt′ )3 ]∆m

˙ =I
9.2 For a circular-arc camberline, Eq. (4-7) gives the arc radius of curva-
ture, RC, as
Rc = c/[2 sin(θ/2)]
Hence the camberline length, L, is given by
L = RCθ = cθ / [2 sin(θ / 2)]
Dividing by the staggered spacing, s cosγ, yields Eq. (9-15).
10.1 For a constant-work, repeating stage, C3 = C1 and U2ψ = constant.

Hence, Eqs. (10-6) and (10-7) yield
h2 − h1 = Uc2ψ c − 12 (C22 − C12 )
h3 − h1 = Uc2ψ c
Hence Eqs. (10-3) and (10-5) yield
Cz22 + Cθ22 − Cz21 − Cθ21

R = 1−
2Uc2ψ c
Differentiating with respect to r,
∂R 1  ∂Cz1 ∂C ∂C ∂C 
= Cz1 + Cθ1 θ1 − Cz2 z2 − Cθ 2 θ 2 
∂r Uc2ψ c  ∂r ∂r ∂r ∂r 
From Eq. (10-24),

∂Cz ∂C C2
Cz + Cθ θ = − θ
∂r ∂r r
Hence
∂R 1 rc  Cθ22 Cθ21 
rc =  − 
∂r ψ c r  Uc2 Uc2 
Substituting Eqs. (10-20) and (10-21) to eliminate the velocity terms

yields Eq. (10-45).
10.2 For a constant-work stage, Eq. (10-24) can be written as
2
rc ∂(Cz2 / Uc ) r 2C ∂[rCθ 2 / ( rcUc )]
= − c θ2
2 ∂r rUc ∂r
Equations (10-20) and (10-21), with n = -1 and m = 1, yield
Cθ 2 / Uc = (1 − Rc )( r / rc ) + 12 ψ c ( rc / r )
∂[rCθ 2 / ( rcUc )]
rc = 2(1 − Rc )( r / rc )
∂r
∂(Cz2 / Uc )2
rc = 4(1 − Rc )[(1 − Rc )( r / rc ) + 12 ψ c ( rc / r )]
∂r
Integration of the last equation yields Eq. (10-48).
10.3 In the answer to Exercise 10.1, it is shown that
∂R 1 rc  Cθ22 Cθ21 
rc =  − 
∂r ψ c r  Uc2 Uc2 
Hence constant reaction requires
1 rc  Cθ22 Cθ21 
 − =0
ψ c r  Uc2 Uc2 
Hence
Cθ22 − Cθ21 (Cθ 2 + Cθ1)(Cθ 2 − Cθ1)

= =0
Uc2 Uc2

For a constant-work stage, H2 = H1, so Eq. (10-2) can be used to yield
(Cθ 2 + Cθ1)ψ c
=0
Uc
Since reaction is undefined for the trivial case of ψc = 0, the

required condition is
Cθ 2 = − Cθ1
This condition also must be satisfied at the reference radius. Equa-

tions (10-9) and (10-10) are valid at the reference radius, so
2(1 − Rc) = 0
The only case where this is true is that of R = 1.
13.1 From Eq. (13-42) it is clear that the loss coefficient cannot be less
than one. About the only design option available is to minimize the
tangential loss, and possibly the exit cone loss, through the choice of
r3 and A3. If the diffusers achieve the same discharge flow condi-
tions, the diffuser type used has no effect on the scroll/collector loss.
13.2 From the information given, no firm conclusions can be reached

about overall exhaust loss. The reduced scroll loss coefficient is due
at least in part to the increase in kinetic energy supplied by the cen-
trifugal impeller. The absolute exhaust system loss could actually be
higher for the centrifugal stage configuration. The substitution is
almost certain to reduce the curvature loss of the original diffuser
and the tangential loss in the scroll. To justify the substitution, a per-
formance analysis of the centrifugal stage and its exhaust system
must be compared to the exhaust system analysis without the substi-
tution, including any benefits from the additional pressure rise sup-
plied by the centrifugal stage.
13.3 The axial diffuser will have the lower loss since there will be no cur-
vature loss contribution. The main reason a curved diffuser might be
chosen is to reduce the overall axial length of the compressor. If the
flow exiting the compressor has significant Cθ, the curved diffuser
could be more effective. The higher discharge radius will yield
greater diffusion of Cθ through conservation of angular momentum.
13.4 The results will basically be as dependable as the correction model

itself. While not all losses in a compressor are skin-friction-related,
that simply means not all losses are Reynolds-number-related. If the
correction model used accounts for that, the imposed roughness cor-

rection should be valid. If anything, it might be slightly conservative

if the Reynolds number lies in a zone of transition from smooth to
rough skin friction. The greatest uncertainty lies in the ability to
assign a surface roughness that is consistent with the characteristic
Reynolds number used in the correction model.
13.5 Calculate the value of surface roughness, e, that results in Ree = 60. It
is unnecessary to polish the surfaces to achieve a surface roughness
less than that value. The constant, 2,000, in the definition of Ree is
rather insignificant and can be omitted in Eq. (13-34) for more gen-
eral applications. Hence, the Reynolds number based on e is the sig-
nificant parameter.

REFERENCES
Aungier, R. H., 1968, “A Time-Dependent Numerical Method for Calculating the

Flow About Blunt Bodies,” Technical Report AFWL-TR-68-52, Air Force
Weapons Laboratory, Kirtland AFB, NM.
Aungier, R. H., 1970, “A Computational Method for Exact, Direct and Unified
Solutions for Axisymmetric Flow Over Blunt Bodies of Arbitrary Shape (Pro-
gram BLUNT),” Technical Report AFWL-TR-70-16, Air Force Weapons Labora-
tory, Kirtland AFB, NM.
Aungier, R. H., 1971(a), “A Computational Method for Unified Solutions to the
Inviscid Flow Field About Blunt Bodies,” The Entry Plasma Sheath and Its Effects
on Space Vehicle Electromagnetic Systems, (Proceedings of the Fourth Plasma
Sheath Symposium), NASA SP-252, NASA, Washington, DC, pp. 241–260.
Aungier, R. H., 1971(b), “A Computational Method for Two-Dimensional, Axisym-
metric and Three-Dimensional Blunt Body Flows (Program ATTACK),” Techni-
cal Report AFWL-TR-70-124, Air Force Weapons Laboratory, Kirtland AFB, NM.
Aungier, R. H., 1988(a), “A Systematic Procedure for the Aerodynamic Design of
Vaned Diffusers,” Flows In Non-Rotating Turbomachinery Components, FED-
Vol. 69, ASME, New York, NY, pp. 27–34.
Aungier, R. H., 1988(b), “A Performance Analysis for the Vaneless Components of
Centrifugal Compressors,” Flows in Non-Rotating Turbomachinery Compo-
nents, FED-Vol. 69, ASME, New York, NY, pp. 35–43.
Aungier, R. H., 1993, “Aerodynamic Design and Analysis of Vaneless Diffusers
and Return Channels,” Paper No. 93-GT-101, ASME, New York, NY.
Aungier, R. H., 1994, “A Fast, Accurate Real Gas Equation of State for Fluid
Dynamic Analysis Applications,” Contributed Papers In Fluids Engineering
1994, FED-Vol. 182, ASME, New York, NY, pp 1–6.
Aungier, R. H., 1995, “A Fast, Accurate Real Gas Equation of State for Fluid
Dynamic Analysis Applications,” Trans., Journal of Fluids Engineering, ASME,
June, pp. 277–281.
Aungier, R. H., 1998, “Thermodynamic State Relations,” The Handbook of
Fluid Dynamics (R. W. Johnson, editor), CRC Press LLC, Boca Raton, FL,
pp. 4-29–4-34.
Aungier, R. H., 2000, Centrifugal Compressors: A Strategy for Aerodynamic Design
and Analysis, ASME Press, New York, NY.
Balje, O. E., 1981, Turbomachines—A Guide to Design, Selection and Theory, Wiley
& Sons, New York, NY.

Balsa, T. F. and Mellor, G. L., 1975, “The Simulation of Axial Compressor Perfor-
mance Using an Annulus Wall Boundary Layer Theory,” Trans. Journal of
Engineering for Power, ASME, July, pp. 305–317.
Barnes, F. J., 1973, Ph.D. thesis, Department of Chemical Engineering, University
of California, Berkeley, CA.
Budinger, R. E. and Thomson, A. R., 1952, “Investigation of A 10-Stage Subsonic
Axial-Flow Compressor; II—Preliminary Analysis of Over-All Performance,”
NACA Research Memorandum RM E52C04, NACA, Washington, DC.
Courant, R., Friedricks, K. O. and Lewy, H., 1928, “Uber die Partiellen Differen-
zengleichungen der Mathematischen Physik,” Math. Ann., Vol. 100, p. 32.
Cumpsty, N. A., 1989, Compressor Aerodynamics, Longman Scientific and Techni-
cal, Essex, United Kingdom.
Davis, W. R., 1976, “Three-Dimensional Boundary-layer Computation on the Sta-
tionary End-Walls of Centrifugal Turbomachinery,” Trans. Journal of Fluids
Engineering, Sept., ASME, pp. 431–442.
Dean, D. E. and Stiel, L. I., 1965, AIChE Journal, Vol. 11, p. 526.
de Haller, P., 1953, “Das Verhalten von Tragflugelgittern in Axialverdichtern und
in Windkanal,” Brennstoff-Warme-Kraft 5, Heft 10.
Denton, J. D., 1982, “An Improved Time-Marching Method for Turbomachinery
Flow Calculation,” ASME Paper No., 82-GT-239, ASME, New York, NY
De Ruyck, J., Hirsch, C. and Kool, P., 1979, “An Axial Compressor End-Wall
Boundary Layer Calculation Method,” Trans. Journal of Engineering for Power,
April, ASME, Vol. 101, pp. 233–249.
De Ruyck, J. and Hirsch, C., 1980, “Investigations of an Axial Compressor End-
Wall Boundary Layer Prediction Method,” ASME Paper No. 80-GT-53, ASME,
New York, NY.
Dunavant, J. C., Emery, J. C., Walch, H. C. and Westphal, W. R., 1955, “High-
Speed Cascade Tests of the NACA 65-(12A10)-10 and NACA 65-(12A2I8b)-10
Compressor Blade Sections,” NACA Research Memorandum RM-L55I08,
NACA, Washington, DC.
Dunavant, J. C., 1957, “Cascade Investigation of a Related Series of 6-Percent
Thick Guide-Vane Profiles,” NACA TN 3959.
Dunker, R., Rechter, H., Starken, H. and Weyer, H., 1984, “Redesign and Perfor-
mance Analysis of a Transonic Axial Compressor Stator with Subsonic Con-
trolled Diffusion Airfoils,” Trans. Journal of Engineering for Gas Turbines and
Power, ASME, Vol. 106, April, pp. 279–287.
Egli, A., 1935, “The Leakage of Steam Through Labyrinth Glands,” Trans. ASME,
Vol. 57, pp. 115–122.
Emery, J. C., Herrig, L. J., Erwin, J. R. and Felix, A. R., 1958, “Systematic Two-
Dimensional Cascade Tests of NACA 65-Series Compressor Blades at Low
Speeds,” NACA Report 1368, NACA, Washington, DC.
Geye, R. P., Budinger, R. E. and Voit, C. H., 1953, “Investigation of a High-Pres-
sure-Ratio Eight-Stage Axial-Flow Research Compressor With Two Transonic
Inlet Stages; II—Preliminary Analysis of Overall Performance,” NACA
Research Memorandum RM E53J06, NACA, Washington, DC.
Gopalakrishnan, S. and Bozzola, R., 1973, “Numerical Representation of Inlet
and Exit Boundary Conditions in Transient Cascade Flow,” ASME Paper No.
73-GT-55, ASME, New York, NY.

References • 351
Green, J. E., 1968, “The Prediction of Turbulent Boundary Layer Development In

Compressible Flow,” Journal Of Fluid Mechanics, Vol. 31, p. 753.
Gruschwitz, E., 1950, “Calcul Approche de la Couche Limite Laminaire en
Ecoulement compressible Sur Une Paroi Non-conductrice de la Chaleur,”
Office National d’Etudes et de Recherche Aeronautiques (ONERA), Paris,
Publication No. 47.
Head, M. R., 1958, “Entrainment In the Turbulent Boundary Layer,” R&M 3152,
Aeronautical Research Council, London, United Kingdom.
Head, M. R., 1968, “Cambridge Work on Entrainment,” Proceedings of Computa-
tion of Turbulent Boundary Layers, Thermosciences Division, Stanford Univer-
sity, CA. pp. 188–194.
Herrig, L. J., Emery, J. C., and Erwin, J. R., 1957, “Systematic Two-Dimensional
Cascade Tests of NACA 65-Series Compressor Blades at Low Speeds,” NACA
TN 3916, NACA, Washington, DC.
Hirsch, C., 1974, “End-Wall Boundary Layers in Axial Compressors,” Trans. Jour-
nal of Engineering for Power, ASME, Vol. 96, Oct., pp. 413–426.
Hirsch, C., 1976, “Flow Prediction in Axial Flow Compressors Including End-
Wall Boundary layers,” ASME Paper No. 76-GT-72, ASME, New York, NY.
Hobbs, D. E. and Weingold, H. D., 1984, “Development of Controlled Diffusion
Airfoils for Multistage Compressor Application,” Trans. Journal of Engineering
for Gas Turbines and Power, ASME, Vol. 106, April, pp. 271–278.
Horlock, J. H., 1958, Axial Flow Compressors: Fluid Mechanics and Thermody-
namics, Butterworths Scientific Publications, London, United Kingdom.
Horlock, J. H., 1970, “Boundary Layer Problems In Axial Turbomachines,” Flow
Research on Blading (L. S., Dring, editor), Elsevier Publishing, Amsterdam, pp.
322–371.
Howell, A. R., 1942, “The Present Basis of Axial Flow Compressor Design; Part
I—Cascade Theory and Performance,” R&M 2095, British Aeronautical
Research Council, London, United Kingdom.
Howell, A. R., 1945, “Design of Axial Compressors,” Proceedings of the Institution
of Mechanical Engineers, Vol. 153, London, United Kingdom.
Howell, A. R., 1947, “Fluid Dynamics of Axial Compressors,” (Lectures on the
Development of the British Gas Turbine Jet Unit), War Emergency Proc. No. 12,
Institution of Mechanical Engineers, London, United Kingdom [American
Edition published by the American Society of Mechanical Engineers], pp.
441–452.
Hunter, I. H. and Cumpsty, N. A., 1982, “Casing Wall Boundary-Layer Develop-
ment Through an Isolated Compressor Rotor,” Trans. Journal of Engineering
for Power, ASME, Vol. 104, Oct., pp. 805-818.
Huntington, R. A., 1985, “Evaluation of Polytropic Calculation Methods for
Turbomachinery Performance,” ASME Paper no. 85-GT-13, ASME, New
York, NY.
Jansen, W., 1967, “The Application of End-Wall Boundary Layer Effects in the
Performance Analysis of Axial Compressors,” ASME Paper No. 67-WA/GT-11,
ASME, New York, NY.
Jansen, W. and Moffatt, W. C., 1967, “The Off-Design Analysis of Axial-Flow Com-
pressors,” Trans. ASME, Journal of Engineering for Power, Vol. 89, Oct., pp.
453–462.

Johnsen, I. A., 1952, “Investigation of a 10-Stage Subsonic Axial-Flow Compres-

sor; I—Aerodynamic Design,” NACA Research Memorandum RM E52B18
NACA, Washington, DC.
Johnsen, I. A. and Bullock, R. O., editors, 1965, “Aerodynamic Design Of Axial
Flow Compressors,” NASA SP-36, NASA, Washington, DC.
Katsanis, T., 1964, “Use of Arbitrary Quasi-Orthogonals for Calculating Flow Dis-
tribution in the Meridional Plane of a Turbomachine,” NASA TN D-2546,
NASA, Washington, DC.
Katsanis, T., 1968, “Computer Program for Calculating Velocities and Stream-
lines on a Blade-to-Blade Stream Surface of a Turbomachine,” NASA TN D-
4525, NASA, Washington, DC.
Katsanis, T., 1969, “Fortran Program for Calculating Transonic Velocities on a
Blade-to-Blade Stream Surface of a Turbomachine,” NASA TN D-5427, NASA,
Washington, DC.
Koch, C. C., 1981, “Stalling Pressure Rise Capability of Axial Flow Compressor
Stages,” Trans. Journal of Engineering for Power, ASME, Vol. 103, pp. 411–424.
Koch, C. C. and Smith, L. H. Jr., 1976, “Loss Sources and Magnitudes In Axial-
Flow Compressors, Trans. ASME, Journal of Engineering for Power, Vol. 98, pp.
411–424.
Kovach, K. and Sandercock, D. M., 1954, “Experimental Investigation of Five-
Stage Axial-Flow Research Compressor With Transonic Rotors In All stages;
II—Compressor Over-All Performance,” NACA Research Memorandum
RME54G01, NACA, Washington, DC.
Kovach, K. and Sandercock, D. M., 1961, “Aerodynamic Design and Performance
of Five-Stage Transonic Axial-Flow Compressor,” Trans. Journal of Engineering
for Power, ASME, Vol. 83, July, pp. 304-321.
Lax, P. D., 1954, “Weak Solutions of Nonlinear Hyperbolic Equations and Their
Numerical Computation,” Commun. Pure and Appl. Math., Vol. 7, pp. 159-193.
Lax, P. D., and Wendroff, B., 1964, “Differencing Schemes for Hyperbolic Equa-
tions With High Order of Accuracy,” Commun. Pure and Appl. Math., Vol. 17,
pp. 381–398.
Lieblein, S., Schwenk, F. C. and Broderick, R. L., 1953, “Diffusion Factor for Esti-
mating Losses and Limiting Blade Loadings in Axial-flow Compressor Blade
Elements,” NACA RM E53D01, NACA, Washington, DC.
Lieblein, S. and Roudebush, W. H., 1956, “Theoretical Loss Relations for Low-
Speed Two-Dimensional-Cascade Flow,” NACA TN 3662, NACA, Washington,
DC.
Lieblein, S., 1959, “Loss and Stall Analysis of Compressor Cascades,” Trans.,
Journal of Basic Engineering, ASME, Vol. 81, Sept., pp 387–400.
Lieblein, S., 1960, “Incidence and Deviation-Angle Correlations for Compressor
Cascades,” Trans. Journal of Basic Engineering, ASME, Vol. 82, Sept., pp 575–587.
Ludwieg, H. and Tillmann, W., 1950, “Investigations of the Wall-Shearing Stress
in Turbulent Boundary Layers,” NACA TM 1285, NACA, Washington, DC.
Mallen, M. and Saville, G., 1977, “Polytropic Processes in the Performance Pre-
diction of Centrifugal Compressors,” Paper No. C183/77, Institution of
Mechanical Engineers, London, United Kingdom, pp. 89–96.
Mellor, G. L., and Wood, G. M., 1971, “An Axial Compressor End-Wall Boundary
Layer Theory,” Trans. Journal of Basic Engineering, ASME, pp.300–316.

References • 353
Miller, G. R., Lewis, G. W. Jr. and Hartmann, M. J., 1961, “Shock Losses in Tran-
sonic Compressor Blade Rows,” Trans. Journal of Engineering for Power,
ASME, Vol. 83, July, pp. 235–242.
Moretti, G. and Abbett, M., 1966, “A Time-Dependent Computational Method for
Blunt Body Flows,” AIAA Journal, Vol. 4, pp. 2136–2141.
Nelson, L. C. and Obert, E. F., 1954, “Generalized pvT Properties of Gases,” Trans.
ASME, Vol. 76, pp. 1057–1066.
Nikuradse, J., 1930, “Laws of Resistance and Velocity Distributions for Turbulent
Flow of Water in Smooth and Rough Pipes,” Proceedings, 3rd International
Congress for Applied Mechanics, Stockholm, Sweden, pp. 239–248.
Novak, R. A., 1967, “Streamline Curvature Computing Procedures for Fluid-Flow
Problems,” Trans. Journal of Engineering for Power, ASME, Vol. 89, Oct., pp.
478–490.
Novak, R. A., 1973, “Axisymmetric Computing Systems for Axial Flow Turboma-
chinery,” Lecture 25, ASME Turbomachinery Institute Fluid Dynamics of Tur-
bomachinery, Iowa State University, Ames, Iowa.
Pai, S., 1957, Viscous Flow Theory, II-Turbulent Flow, Van Nostrand, Princeton, NJ.
Pitzer, K. S., Lippmann, D. Z., Curl, R. F., Huggins, C. M. and Peterson, D. E.,
1955, “The Volumetric and Thermodynamic Properties of Fluids. II Compress-
ibility Factor, Vapor Pressure and Entropy of Vaporization,” American Chemi-
cal Society, Vol. 77, pp. 3427–3440.
Pohlhausen, K., 1921, “Zur Naherungsweisen Integration der Differential-Gle-
ichung der Laminare Reibungsschicht,” ZAMM, Vol. 1, p. 235.
Pollard, D. and Gostelow, J. P., 1967, “Some Experiments At Low Speed on Com-
pressor Cascades,” Trans. Journal Of Engineering for Power, ASME, Vol. 89, pp.
427–436.
Redlich, O. and Kwong, J., 1949, “On the Thermodynamics of Solutions. V. An
Equation of State. Fugacities of Gaseous Solutions,” Chemical Review, Vol. 44,
pp. 233–244.
Reneau, L., Johnston, J. and Kline, S., 1967, “Performance and Design of Straight
Two-Dimensional Diffusers”, Trans. Journal of Basic Engineering, ASME, Vol.
89, pp. 141–150.
Ried, R. C. and Sherwood, T. K., 1966, The Properties Of Gases And Liquids,
McGraw-Hill, New York.
Ried, R. C., Prausnitz, J. M., and Sherwood, T. K., 1977, The Properties of Gases
and Liquids, McGraw-Hill, New York, NY.
Ried, R. C., Prausnitz, J. M. and Poling, B. E., 1987, The Properties of Gases and
Liquids, 4th Ed., McGraw-Hill, New York, NY.
Rotta, J. C., 1966, “Recent Developments in Calculation Methods for Turbulent
Boundary Layers With Pressure Gradients and Heat Transfer,” Trans. Journal
of Applied Mechanics, ASME, Vol. 88, p. 429.
Sandercock, D. M., Kovach, K. and Lieblein, S., 1954, “Experimental Investiga-
tion of a Five-Stage Axial-Flow Research Compressor With Transonic Rotors
In All stages; I – Compressor Design,” NACA Research Memorandum
RME54F24, NACA, Washington, DC.
Schlichting, H., 1968, Boundary-Layer Theory, 6th Ed., McGraw-Hill, New York,
NY.
Schlichting, H., 1979, Boundary-Layer Theory, McGraw-Hill, New York, NY.

Schultz, J. M., 1962, “The Polytropic Analysis of Centrifugal Compressors,”

Trans. Journal Of Engineering for Power, ASME, Vol. 84, Jan., pp. 69–82.
Schumann, L. F., 1985, “A Three-Dimensional Axisymmetric Calculation Proce-
dure for Turbulent Flows in a Radial Vaneless Diffuser,” ASME Paper No. 85-
GT-133, ASME, New York, NY.
Senoo, Y., Kinoshita, Y. and Ishida, M., 1977, “Axisymmetric Flow in Vaneless
Diffusers of Centrifugal Blowers,” Trans. Journal of Fluids Engineering, ASME,
Vol. 99, March, pp. 104–114.
Sheppard, D. G., 1956, Principles of Turbomachinery, Macmillan, New York, NY.
Smith, D. J. L. and Frost, D. H., 1969, “Calculation of the Flow Past Turbomachine
Blades,” Proc. Inst. Mech. Eng., Vol. 184, Paper 27, London, United Kingdom.
Smith, L. H.. Jr., 1958, “Recovery Ratio—A Measure of the Loss Recovery Poten-
tial of Compressor Stages,” Trans. ASME, Vol. 80, pp. 517–524.
Smith, L. H.. Jr., 1970, “Casing Boundary Layers in Multi-Stage Axial-Flow Com-
pressors,” Flow Research on Blading, (L.S. Dring, editor), Elsevier Publishing,
Amsterdam, pp. 275–304.
Soave, G., 1972, “Equilibrium Constants From a Modified Redlich-Kwong Equa-
tion of State,” Chemical Eng. Science, Vol. 27, 1197–1203.
Sovran, G. and Klomp, E. D., 1967, “Experimentally Determined Optimum
Geometries for Rectilinear Diffusers With Rectangular, Conical or Annular
Cross-Section,” Fluid Mechanics of Internal Flows, (G. Sovran, editor), Elsevier
Publishing, Amsterdam, pp. 270–319.
Stratford, B. S., 1967, “The Use of Boundary Layer Technique to Calculate the
Blockage From the Annulus Boundary Layer in a Compressor,” ASME Paper
No. 67-WA/GT-7, ASME, New York, NY.
Summer, W. J. and Shanebrook, J. R., 1971, “Entrainment Theory for Compress-
ible Turbulent Boundary Layers on Adiabatic Walls,” AIAA Journal, Vol. 9, p.
330–332.
Swan, W. C., 1961, “A Practical method of Predicting Transonic-Compressor Per-
formance,” Trans. Journal of Engineering for Power, Vol. 83, July, pp. 322-330.
Vavra, M. H., 1960, Aero-Thermodynamics And Flow In Turbomachines, Wiley,
New York.
Voit, C. H., 1953, “Investigation of a High-Pressure-Ratio Eight-Stage Axial-Flow
Research Compressor With Two Transonic Inlet Stages; I – Aerodynamic
Design,” NACA Research Memorandum RM E53I24, NACA, Washington, DC.
Von Neumann, J. and Richtmyer, R. D., 1950, “A Method for the Numerical Cal-
culation of Hydrodynamic Shocks,” Journal of Applied Physics, Vol. 21, pp.
232–237.
Walsh, J. L., Ahlberg, J. H., and Nilson, E. N., 1962, “Best Approximation Properties
of the Spline Fit,” Journal of Mathematics and Mechanics, Vol. 11, pp. 225–234.
Wassell, A. B., 1968, “Reynolds Number Effects in Axial Compressors,” Trans.
Journal of Engineering for Power, ASME, April, Vol. 90, pp. 149–156.
Wilson, G. M., 1966, “Calculation of Enthalpy Data From a Modified Redlich-
Kwong Equation of State,” Adv. Cryogenic Eng., Vol. 11, p. 392.
Wu, C. H., 1952, “A General Theory of Three-Dimensional Flow In Subsonic And
Supersonic Turbomachines of Axial-, Radial- and Mixed-Flow Types,” NACA
TN 2604, NACA, Washington, D.C.
Yaws, C. L., 1999, Chemical Properties Handbook, McGraw-Hill, New York, NY.

ABOUT THE AUTHOR
Mr. Aungier is the manager of Advanced Technology for Elliott Turbomachinery

Company Inc., Ebara Group in Jeannette, Pennsylvania. He has been active in
fluid mechanics research and development for more than 36 years, 32 of those in
turbomachinery aerodynamics, specializing in centrifugal compressors, axial-
flow compressors and radial-inflow turbines. He has numerous publications in
this field, primarily through the American Society of Mechanical Engineers. He
is a graduate of Cornell University, where he received a masters degree in Aero-
space Engineering and a bachelors degree in Engineering Physics.
Mr. Aungier started his career in 1966 as an officer in the U.S. Air Force, con-
ducting research in hypersonic re-entry vehicle aerodynamics at the Air Force
Weapons Laboratory in Albuquerque, New Mexico. He is the author of numerous
Air Force and NASA publications, some of which are the basis for one of the
analysis techniques described in this book. In 1970, Mr. Aungier joined the
Research Division of Carrier Corporation in Syracuse, New York, where he spent
11 years managing and conducting applied research on the fluid dynamics of tur-
bomachinery and air handling equipment. Most of his individual research was
focused on the interests of Elliott Company (then a division of Carrier), including
development of aerodynamic performance analysis techniques for axial-flow
compressors, centrifugal compressors and radial-inflow turbines. In 1981,
Mr. Aungier transferred to Elliott Company as manager of Compressor Develop-
ment, where his interests expanded to include the development of systematic and
efficient techniques for aerodynamic design of turbomachinery. His responsibili-
ties were extended to include turbine aerodynamic development in 1983 and
mechanical design and analysis in 1987. He continues to be an active contributor
to turbomachinery aerodynamic technology, specializing in comprehensive aero-
dynamic design and analysis systems. In 2000, ASME Press published his first
book, describing his centrifugal compressor aerodynamic design and analysis
system. The present book provides a similar treatment of his axial-flow compres-
sor aerodynamic design and analysis system.

INDEX
Accentric factor, 26 NACA 63–series, 70–71

Adiabatic efficiency; see efficiency, NACA 65–series, 62–65
adiabatic Blade row; also see blade and cascade
Adiabatic process; see isentropic Design, 221–222, 264–266
process Diffusion limit, 130, 132, 204–206
Adjustable blade rows, 311–316 Force defect; see boundary layer,
Angle of attack blade force defect
Defined, 62, 64, 119–120 Shrouded, 6, 147–149, 191–193
Design, 121,124 Throat, 73–75, 201
Stall, 135 Types, 3, 201
Angular momentum, conservation of, Blade-to-blade flow; see flow, blade-
157–158, 162, 293, 325 to-blade
Annulus sizing, 169–171, 263, Blockage, aerodynamic, 41, 145, 157,
266–268; also see end-wall 195, 318, 320–321
contours Blockage, blade, 293
Axial-centrifugal compressor, 328–332 Boundary layer
Axisymmetric, 54
Blade; also see blade camberline; Axisymmetric-three-dimensional,
blade profile and blade row 54–56, 175–197, 302–307
Angles, 60–62, 64–65, 70–71, 76, Blade force defect, 55–56, 179–180,
119–120 182, 187–191, 303–306
Construction, 62–71 Clearance gap, 179, 187, 305
Controlled diffusion airfoil, 59, Displacement thickness, 52, 56, 108,
71–73, 117 178, 182
Geometry on a stream surface, 202 Energy thickness, 109
Leading edge, 89–90 Enthalpy thickness, 109
Loading, 129, 222–224, 273 Entrainment, 54, 56, 181, 187
Blade camberline Laminar, 108–110
Circular-arc, 65–66 Momentum-integral equation, 54,
NACA A4K6, 70–71 56, 107, 181–182
NACA 65–series, 62–65 Momentum thickness, 53, 56,
Parabolic-arc, 66–68 108–109, 182
Blade profile Profiles, 108, 183–184
C-series, 68–69 Profile loss coefficient, 112
Double-circular-arc, 69–70 Separation, 110–111, 184

Boundary layer (continued) Conservation of mass, 157, 160, 292,

Shape factors, 108, 110–111, 296, 299; also see continuity
183–184 equation and boundary layer,
Shroud seal leakage effects, entrainment
191–193 Continuity equation, 47, 83–84, 97,
Transition, 110–111, 186, 194 159
Turbulent, 110–111, 175–197, Contours; see end-wall contours
302–307 Controlled diffusion airfoil; see blade,
Two-dimensional, 51–54, 107–113 controlled diffusion airfoil
Velocity thickness, 109 Critical Mach number, 138
Critical point, 23–24
Camber angle; see blade, angles Curvature effects, 47, 157, 292,
Camberline; see blade camberline 319–320
Cascade; see also blade row
Empirical performance models, Departure functions, 30–31, 37
121–151 Design, detailed
Geometry, 60–62, 75–76, 119–120 Adjustable blade rows; see
Centrifugal compressor; see adjustable blade rows
compressor types Multistage compressor, 251–257,
CFL stability criterion, 103 259–285
Characteristics, 98–101 Stage, 215–257
Choked flow, 161, 299 Deviation angle
Collector, 322–327 Defined, 62, 120
Compressibility factor, 26 Design, 125–128
Compressible flow analysis; see flow, Off-design, 142–144
compressible D-factor, see diffusion factors
Compressor design; see design, Diffuser
detailed, multistage compressor Divergence angle, 318–319
Compressor performance analysis; Divergence parameter, 318–319
see performance analysis, Exhaust, 316–322
compressor Two-dimensional, 204
Compressor types, 1, 328–332 Diffusion Factors, 129–132, 206
Computational fluid dynamics (CFD) Dimensionless parameters, 10–11,
Euler (inviscid flow) codes; see 13–14, 217, 262
flow, inviscid Displacement thickness; see
Viscous CFD codes, 42, 50, 114, 303 boundary layer, displacement
Computerized design system, useful thickness
features Divergence angle; see diffuser,
Aerodynamic performance analysis, divergence angle
200–202, 311, 327–328 Drag coefficient, 133
Blade geometry database, 73
Cascade performance models, 149 Efficiency
Equation-of-state package, 37–38 Adiabatic, 20–21, 32
Internal flow analysis, 288, Compressor, 11, 20–22, 32
290–291, 301 Diffuser, 33
Meridional through-flow, 167–171 Nozzle, 35
Multi-stage compressor, 261–268 Polytropic, 21–22, 32
Stage, 257 Elliptic equations, 49, 81

Index • 359
End-wall boundary layer; see Flow coefficient, 14, 217, 219, 262,
boundary layer, axisymmetric- 330–331
three-dimensional Flow work, 19
End-wall contours Fluid turning, 64, 120, 141
Designing, 169–171, 252, 263, 266
Smoothing, 266–267 Gas constant, 10, 23
Energy equation, 19, 47, 97, 158, 293 Gas mixtures, 29
Energy thickness; see boundary layer, Gas property data, 24, 29
energy thickness Gas viscosity; see viscosity
Enthalpy, 19, 25, 31, 45–46
Enthalpy thickness; see boundary Head
layer, enthalpy thickness Adiabatic, 20
Entrainment; see boundary layer, Defined, 11
entrainment Polytropic, 22
Entropy, 19, 25, 31 Helmholtz energy, 30
Equation of state Hub-to-shroud flow; see flow, hub-to-
Aungier’s modified Redlich-Kwong, shroud
26–29 Hydraulic diameter, 326–327
Caloric, 22, 24–25 Hyperbolic equations, 49, 81
Calorically perfect gas, 25, 31–32
Comparison of, 28
Incidence angle
Perfect gas, 23–26, 31–32
Choking, 136–137
Pseudo-perfect gas, 32–33
Defined, 62, 120
Real gas, 26, 30–31
Design, 122–124
Redlich-Kwong, 26–29
Minimum loss, 138
Thermal, 22, 26
Stall, 134–137
Thermally perfect gas, 23–26, 31–32
Equivalent diffusion factor, see Internal energy, 18, 24–25
diffusion factors Inviscid flow analysis; see flow,
Equivalent performance; see similitude inviscid
Euler turbine equation, 13, 44–45 Irrotational flow; see flow,
irrotational
Finite-difference approximations; see Isentropic efficiency, see efficiency,
numerical approximations adiabatic
Flow Isentropic process, 20, 46
Blade-to-blade, 41, 49, 77–107,
290–291 Kutta condition, 87, 94–95
Compressible, 316–322; see also
flow, inviscid Labyrinth seal, 148–149
Hub-to-shroud, 41, 49, 153–172, Lift coefficient, 63, 65–66, 70–71, 120,
291–299; also see normal 133
equilibrium Loss coefficient
Inviscid, 48–50, 77–107, 113–114, Blade tip clearance, 146–147
153–169, 288–302 Collector, 327
Irrotational, 49, 81, 83–84 Defined, 36
One-dimensional, 41, 318–322 Design, 128, 130, 132–134
Quasi-three-dimensional, 41, 50, Discharge, 327–328
287–302 End-wall, 133, 150
Transonic, 89 Minimum, 138

Loss coefficient (continued) Polytropic process, 21–22

Off-design, 138, 144–145, 150 Potential flow; see flow, irrotational
Profile, 112, 128, 130, 132 Power, 18–19, 44
Reynolds number effect, 150–151 Pressure recovery coefficient, 34, 322,
Scroll, 327 327
Shock wave, 138–141
Shroud seal leakage, 147–149 Quasi-normal, 155–157, 201, 291–292
Smoothing, 203 Quasi-three-dimensional flow; see
Supercritical Mach number, 138 flow, quasi-three-dimensional
Mach number, 9, 160; also, see Ratio of specific heats, 10, 26, 31
critical Mach number Reaction, 14, 217–219, 262
Mass conservation; see conservation Recovery ratio, 226–229
of mass Relative conditions, 3–6, 43–46
Matrix methods, 88–89 Repeating stage, 215, 219, 251–257
Meridional coordinate; see stream Reversible process, 11, 19
surface Reynolds number, 10, 110, 150–151,
Meridional through-flow; see flow, 321, 328
hub-to-shroud Rotating coordinate system, 43, 55
Momentum equations, 46–47, 97, Rotating stall, see stall
157, 293 Rothalpy, 45
Momentum-integral equation; see
boundary layer, momentum- Saturation line, 23, 30
integral equation Scroll, 322–327
Momentum thickness; see boundary Seal leakage, 147–149, 191–193
layer, Momentum thickness Shear stress, 51–54, 182; see also skin
friction coefficient
Natural coordinates, 44, 57, 156 Similitude, 7–11
Normal equilibrium Sizing parameter, 326
Approximate, 49, 168, 207–211 Skin friction coefficient
Simple, 48–49, 168, 219, 221 Calculation of, 109, 111, 186,
Full, 48–49, 165–167, 211–213 321–322
Numerical approximations, 87–88, Defined, 108, 318
94, 104, 171–172 Loss calculation from, 150, 326
Numerical stability, see stability, Solidity, 62, 120
numerical Sound speed, 10, 26, 31
Specific heat, 24, 31; also see ratio of
Passage curvature; see curvature specific heats
effects Stability, aerodynamic; see stall and
Performance analysis surge
Compressor, 199–214 Stability, numerical, 102–105, 165,
Diffuser, 316–322 297, 299
Volute and collector, 322–327 Stage design; see design, stage
Performance characteristics, 7–13 Stage loading, 273
Periodicity condition, 86 Stage matching, 11–13, 259, 270
Pitch, 62 Stagger angle; see blade, angles
Polytropic efficiency; see efficiency, Staggered spacing, 179, 206
polytropic Stall, 204–207, 222–224

Index • 361
Stall incidence angles; see incidence Torque, 44

angle, stall Transition; see boundary layer,
Stokes’ theorem, 84, 129; also, see transition
also flow, irrotational
Stream function, 85, 93 Vapor saturation conditions; see
Streamline curvature method; see saturation line
normal equilibrium, full Vector operators, 57
Stream surface curvature; see Velocity diagrams, 5–8
curvature effects Velocity thickness; see boundary
Stream surface, 43, 79–80, 155, 289 layer, velocity thickness
Surface roughness, 321–322, 328 Viscosity, 37
Surge, 11–12, 204–207 Viscous flow analysis; see boundary
Swirl vortex types; see vortex types layer and computational fluid
dynamics (CFD), viscous CFD
Thermodynamic properties codes
Real gas, 30–31 Volume ratio effects, 9, 12
Rotating-to-stationary coordinate Volute; see scroll
conversion, 45–46 Vortex types,
Thermally and calorically perfect Assigned flow angle, 245
gas, 25, 31 Comparison of, 247–251, 280–284
Thermally perfect gas, 24 Constant reaction vortex, 235–241,
Total-to-static conversion, 25, 31, 280
45–46 Constant swirl vortex, 242–245, 280
Thermodynamics, 17–40 Defining equation, 220, 262–263
First law of, 18 Exponential vortex, 242
Second law of, 19 Free vortex, 230–235, 280
Throat; see blade row, throat
Time-marching method, 49, 81, Work input, 13, 225
96–107 Work coefficient, 13, 217, 219, 262
Total thermodynamic conditions, 19,
45–46

Aungier - 2003 - Axial-Flow Compressors A Strategy For Aerodynamic

Uploaded by

Copyright:

Available Formats

You might also like

Aungier - 2003 - Axial-Flow Compressors A Strategy For Aerodynamic

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

Aungier - 2003 - Axial-Flow Compressors A Strategy For Aerodynamic

Uploaded by

Copyright:

Available Formats

AXIAL-FLOW

NEW YORK ASME PRESS 2003

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

Co-published in the UK by Professional Engineering Publishing Limited,

INFORMATION CONTAINED IN THIS WORK HAS BEEN OBTAINED BY THE AMERI-

Library of Congress Cataloging-in-Publication Data

TJ267.5.C5 A94 2003

Cover photo: Courtesy of Elliott Turbomachinery Co., Inc., Ebara Group

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

4 Axial-Flow Compressor Blade Profiles 59

5 Two-Dimensional Blade-to-Blade Flow Through

6 Empirical Performance Models Based On

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

6.3 Design Deviation Angle.............................................................125

7 Meridional Through-Flow Analysis 153

8 End-Wall Boundary Layer Analysis 175

9 Aerodynamic Performance Analysis 199

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

9.2 Cascade Performance Considerations.........................................203

10 Compressor Stage Aerodynamic Design 215

11 Multistage Axial-Flow Compressor Aerodynamic

12 Quasi-Three-Dimensional Blade Passage Flow

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

12.3 Numerical Integration of the Governing Equations................294

13 Other Components and Variations 309

Answers to the Exercises 333

About the Author 355

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

Books on compressor aerodynamics approach the subject in various ways

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

Compressors are commonly classified as either positive displacement or dynamic

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

FIGURE 1-1 Performance of Dynamic Compressors

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

d = parameter at compressor discharge

1.1 AXIAL-FLOW COMPRESSOR BASICS

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

FIGURE 1-2 Axial-Flow Compressor Configuration

FIGURE 1-3 Polar Surface View of a Stage

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

1.2 BASIC VELOCITY DIAGRAMS FOR A STAGE

Therefore, the absolute and relative velocities are

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

FIGURE 1-4 Shrouded Stator Blades

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

FIGURE 1-5 Guide Vane Velocity Diagrams

1.3 SIMILITUDE AND PERFORMANCE CHARACTERISTICS

Similitude or similarity is one of the most useful concepts in turbomachinery

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

FIGURE 1-6 Rotor Blade Velocity Diagrams

corresponding length dimensions, velocity components and forces are equal

The local axial velocity is actually given by

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/05/2016 Terms of Use: http://www.asme.org/about-a

FIGURE 1-7 An Equivalent Performance Map