Professional Documents
Culture Documents
Aungier - 2003 - Axial-Flow Compressors A Strategy For Aerodynamic
Aungier - 2003 - Axial-Flow Compressors A Strategy For Aerodynamic
Aungier - 2003 - Axial-Flow Compressors A Strategy For Aerodynamic
COMPRESSORS
A STRATEGY FOR AERODYNAMIC DESIGN AND ANALYSIS
Ronald H. Aungier
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)
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.
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.
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
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
References 349
Index 357
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
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.
INTRODUCTION
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
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)
Subscripts
Superscripts
′ = relative condition
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.
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)
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
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.
˙ / ρt
Q=m (1-7)
˙ / ( 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,
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
P = ρRT (1-9)
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.,
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.
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
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.
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.
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)
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.,
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
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.
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
EXERCISES
THERMODYNAMICS
NOMENCLATURE
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
Superscripts
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
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)
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.,
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
Ptd dP
∆Had = ∫ ; (s = const) (2-10)
Pti ρ
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
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
ηp = (2-14)
∆H
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)
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
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.,
where s0 can be assigned any desired value at any reference state point (Tref, Pref).
For a calorically perfect gas, this simplifies to
∂P ∂P
a 2 = = k (2-26)
∂ρ s ∂ρ T
k = cp / cv
(2-27)
a0 = kRT (2-28)
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
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
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
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
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
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
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.
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)
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
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)
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 )
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,
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
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
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
Pti − Ptd
ηnoz = 1 − (2-65)
Pi − Pd
∆Pt
ω = (2-67)
1
2
ρ C2
∆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.
ξ = Tc 6 / MPc 3
1 2
(2-69)
8
µ 0ξ = (3.4 ⋅10−4 )Tr 9 ; Tr ≤ 1.5 (2-70)
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.
• 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
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.
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
C = absolute velocity
E = entrainment function
→
e = unit vector
f = body force
H = total enthalpy
h = static enthalpy
I = rothalpy
m = meridional coordinate
ṁ = mass flow rate
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
W = relative velocity
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
Subscripts
Superscripts
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
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)
˙ = ωτ = ω m
w ˙ ( r2Cθ 2 − r1Cθ1)
(3-8)
Combining Eq. (3-8) with Eq. (2-4) yields the well-known Euler turbine equation
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)
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.
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
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
∂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
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.
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.
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.
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
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
∫ 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
∫ 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
1 ∂ u ∂ ∂P
[bρ eue2 (δ − δ * − θ )] − e [bρ eue (δ − δ * )] + δ e = −τ w (3-40)
b ∂x b ∂x ∂x
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
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.
Equation (3-43) describes axisymmetric boundary layers where the flow field
is two-dimensional, i.e., there is no tangential velocity component. When a
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
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
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 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
→
where V is the magnitude of the vector V as given by the dot product.
EXERCISES
∆Pt = 1
2
ρ eue2 [(2δ * / b)2 + 4θ / b]
AXIAL-FLOW COMPRESSOR
BLADE PROFILES
General concepts from the literature used to guide the development of controlled
diffusion airfoils are also briefly reviewed in this chapter.
NOMENCLATURE
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
θ = χ1 + χ 2 (4-1)
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)
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
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.
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
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
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)
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
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.
y(0) = 0 (4-11)
y( c) = 0 (4-12)
y( a) = b (4-13)
y'( a) = 0 (4-14)
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
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
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.,
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.
where θ is the blade camber angle and θU is shown in Fig. 4-8. The distance, ∆yU,
is given by
where the camberline coordinate, y(0), is given by Eq. (4-10) and d is defined as
The Pythagorean theorem applied to the right triangle in Fig. 4-8 requires
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.
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
291.5Cl0
tan θ =
468.75 − (5.83Cl0 )2 (4-28)
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.
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:
• 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.
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.
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
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
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.
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)
θ = κ1 − κ 2 (4-34)
χ1 = χ 2 = θ / 2 (4-35)
γ = (κ1 + κ 2 ) / 2 (4-36)
TWO-DIMENSIONAL
BLADE-TO-BLADE FLOW
THROUGH CASCADES OF BLADES
NOMENCLATURE
Subscripts
Superscripts
(η) = relative to the η direction
(ξ) = relative to the ξ direction
′ = relative condition and first derivative
″ = second derivative
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
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.
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)
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.
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)
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
∂ 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
∂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
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 ∂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
∂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
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)
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
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
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.
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
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
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)
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.
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.
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.
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
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
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
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
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).
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.
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
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
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
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
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)
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
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
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
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)
ρ eWeθ
Reθ = > 250 (5-126)
µ
∂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)
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:
The skin friction coefficient correlation of Ludwieg and Tillmann (1950) is com-
monly used for incompressible turbulent boundary layer analysis.
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
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.
5.7 SUMMARY
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).
EMPIRICAL PERFORMANCE
MODELS BASED ON
TWO-DIMENSIONAL
CASCADE TESTS
NOMENCLATURE
a = distance to point of maximum camber along chord
Bwake = blade wake blockage
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
t = total thermodynamic 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
θ = tangential component
* = sonic flow condition
Superscripts
* = design condition
′ = relative condition
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
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)
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.,
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.
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
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
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
κ1 = γ + 1
2θ (6-16)
α * = κ1 + i * − γ + ( a / c − 0.5)θ (6-17)
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.
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.,
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:
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.
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
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,
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).
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.,
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).
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
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
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
where h is the blade height. Howell’s secondary flow drag coefficient is based on
the blade lift coefficient, CL
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
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-
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.
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:
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
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.,
It follows directly that the minimum correction required for the minimum loss
coefficient is
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
When the inlet Mach number exceeds the critical Mach number, the minimum
loss coefficient is estimated from
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.
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
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
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)
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
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.
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)
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
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
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.
The average pressure difference across each blade in the blade row is
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
(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)
˙ 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.
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
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
ω SF ∝ cf c / ( s cos β2 ) (6-97)
ω SF ∝ cf [c / ( s cos β2 ) + c / h] (6-98)
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
[ ]
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.
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)?
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
C = absolute velocity
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
m = meridional coordinate
ṁ = mass flow rate
n = normal coordinate
P = pressure
r = radius
s = entropy
T = temperature
W = relative velocity
y = coordinate along a quasi-normal
z = axial coordinate
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
m = meridional component
s = shroud parameter
t = total thermodynamic condition
θ = tangential component
1 = condition at point preceding point being considered
2 = condition at point being considered
3 = condition at point following point being considered
Superscripts
′ = relative condition
* = parameter on stream surface where Wm is specified for annulus
sizing
Figure 7-1 illustrates the basic meridional coordinate system used in a merid-
ional through-flow analysis. A series of meridional computing stations or
∆z zh − zs
tan λ = = (7-1)
∆r rs − rh
∂r (7-2)
sin φ =
∂m
ε =φ−λ (7-3)
∂ 1 ∂ ∂
= − sin ε (7-4)
∂n cos ε ∂y ∂m
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)
∂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.
Equation (7-12) can also be expressed in terms of the relative flow angle, noting
that Wθ = Wm tanβ′.
After some basic algebra and trigonometry, Eq. (7-13) simplifies to the form
∂Wm f ( y)
= f1( y)Wm + f2 ( y) + 3 (7-15)
∂y Wm
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
∂( 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
1 ∂ρ 1 C 2 sin φ ∂Wm
= 2 θ − Wm (7-25)
ρ ∂m a r ∂m
1 ∂Wm 2 sin φ ∂φ
(1 − Mm ) = −(1 + Mθ2 ) − (7-26)
Wm ∂m r ∂n
Mm = Wm / a (7-27)
Mθ = Cθ / a (7-28)
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)
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 ˙
∂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
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)
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
• 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.
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)
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.
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
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)
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.
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
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)
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
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.
In this way, the same through-flow analysis can be applied to a variety of axial-
flow compressor aerodynamic design and analysis functions.
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.
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
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.
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
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
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?
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
C = absolute velocity
cf = skin friction coefficient
E = entrainment function
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
ṁ = mass flow rate
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
δ = boundary layer thickness
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
Subscripts
Superscripts
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.
∆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),
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
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
∂ 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.
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.
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
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.
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.
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
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.
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.
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 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
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:
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.,
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.,
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
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)
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
m˙ leak
∆+ = (8-59)
(2π rρ eVme )in
+
θ11 = θ11 + ∆+ (8-60)
in
Uleak = 1 rω c + Vθ − Cθ
2 (8-61)
The tangential momentum balance, for either positive or negative leakage mass
flow rate, yields
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
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)
Cθ2eθ 22 = Wθ2eθ 22
′ + ω rWθeδ 2′ * (8-70)
Cθ2eθ 22 (1 + H2 ) = Wθ2eθ 22
′ (1 + H2′ ) + 2ω rWθeδ 2′ * (8-71)
equations can proceed through the compressor without any complex logic at
transitions between rotating and stationary walls.
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.,
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
∂ 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).
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.
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
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).
AERODYNAMIC PERFORMANCE
ANALYSIS
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
Subscripts
Superscripts
′ = relative condition
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:
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
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.
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
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.
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
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.
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
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
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
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.,
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:
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.
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
(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.
Figure 9-5 compares performance predictions with experiment for the NACA
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
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.
Figure 9-7 compares performance predictions with experiment for the NACA
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.
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.
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
˙
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).
COMPRESSOR STAGE
AERODYNAMIC DESIGN
NOMENCLATURE
C = absolute velocity
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
W = relative velocity
Z = number of blades
z = axial coordinate
β = flow angle
γ = stagger angle
δ = deviation 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
Superscripts
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
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
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)
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
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.
φ c = Czc / Uc (10-17)
ψ c = (Cθ 2c − Cθ1c ) / Uc (10-18)
Rc = 1 − (Cθ1c + Cθ 2c ) / (2Uc ) (10-19)
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.
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:
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.
W2 φ c2 + (ψ c / 2 − Rc )2
= (10-26)
W1 φ c2 + (ψ c / 2 + Rc )2
C3 C1 φ c2 + (1 − Rc − ψ c / 2)2 (10-27)
= =
C2 C2 φ c2 + (1 − Rc + ψ c / 2)2
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
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
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)
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
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
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.
Cθ 2 / U = Cθ1 / U + ψ c ( rc / r )2 (10-40)
R = 1 + ( Rc − 1)( rc / r )2 (10-41)
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)
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
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.
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
∂R
rc = 2(1 − Rc )( rc / r )(2+ n) + ψ c [( rc / r )3 − ( rc / r )(2+ m) ] (10-45)
∂r
∂R
rc = 2(1 − Rc )rc / r (10-46)
∂r
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
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
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
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.
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)
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
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.
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
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.
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
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.
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
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
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.
MULTISTAGE
AXIAL-FLOW COMPRESSOR
AERODYNAMIC DESIGN
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 = absolute velocity
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
W = relative velocity
WRE = equivalent velocity ratio
Z = number of blades
z = axial coordinate
γ = stagger angle
η = efficiency
θ = camber angle
σ = solidity
φ = flow coefficient and stream surface slope angle
ψ = work coefficient
ω = rotation speed, radians/sec.
Subscripts
c = parameter on the mean stream surface
h = parameter on hub contour
i = stage number
m = meridional component
s = parameter on shroud contour
θ = tangential component
1 = rotor inlet condition
2 = rotor exit condition
3 = stator exit condition
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
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
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)
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
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.
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
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
c = ch [1 + ( r − rh )( Fsh − 1) / ( rs − rh )] (11-13)
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).
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
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
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.
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
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
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
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
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.
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
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.
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,
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.
QUASI-THREE-DIMENSIONAL
BLADE PASSAGE FLOW
FIELD ANALYSIS
(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
tb = blade thickness
W = velocity relative to the blade row
y = distance along quasi-normal
Z = number of blades
z = axial coordinate
β = flow angle
∆A = stream sheet area
∆ν = increment in ν due to blade guidance
δ = boundary layer thickness
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
ω = rotation speed, radians/sec.
Subscripts
Superscripts
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.
∆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
∂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
∂( rWθ + ω r 2 ) ∂rCθ
= =0 (12-12)
∂m ∂m
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
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
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.
∂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,
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
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)
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 ˙
∂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)
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.
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.
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.,
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.
˙ / (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
∫ 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)
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
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)
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
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
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.
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
C = absolute velocity
cp = static pressure recovery coefficient
cf = skin friction 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
ω = rotation speed, radians/sec.
– = total pressure loss coefficient
ω
Subscripts
m = meridional component
r = fully rough wall condition
s = smooth wall value
sf = skin friction parameter
t = total thermodynamic condition or turbulent condition
tip = blade tip parameter
θ = tangential component
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)
Fi = [( N + 1 − i) / N]n (13-3)
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
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
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.
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
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
The experimental results in Fig. 8(b) of Reneau et al. (1967) show that diffusion
losses are low when D < Dm, where
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)
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)
dI D 1 dC
= −2( Pt − P )(1 − E) (13-18)
dm ρ C dm
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
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
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
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 )
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
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
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.
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 13-10 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)
C4 = Cm2 A2 / A4 (13-40)
The overall total pressure loss coefficient for the scroll or collector is defined as
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)
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
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
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)
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 dis = 1 − cp (13-61)
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
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).
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:
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
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)?
N / at0 = N / kRTt0
Speed parameter:
ψ = Cθ 2 / U
R = 1 − Cθ 2 / (2U ) = 1 − ψ / 2
NB = NA / 1.2
QB = 1.44QA
2cpT1 = 2a12 / ( k − 1)
( a1 / at1)2 = 1 / (1 + k − 1 M12 )
2
( ρ1 / ρt1)k −1 = 1 / (1 + k −1 M2 )
2 1
H2 = H1 + UCθ 2
In the absence of losses, the flow is isentropic and Eq. (2-55) yields
k
Pt2id = Pt1(Tt2id / Tt1) k −1
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
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
k
Pt2 / Pt1 = (Tt2 / Tt1) k −1
1
ρt2 / ρt1 = (Tt2 / Tt1) k −1
˙ / ( ρt2at2 A2 ) = [m
m ˙ / ( ρt1at1A1)][ρt1at1A1 / ( ρt2at2 A2 )]
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
NN = NB (1.38Tt2N ) / (1.4Tt2 B )
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
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
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
2.8 From Eqs. (2-63) and (2-68), the discharge static and total pressures
are
For a thermally perfect gas with no work or heat transfer, Ttd = Tti =
300. Equation (2-55) yields the inlet and discharge static temperatures.
cp (295.44 − 288.26)
ηdiff = = 86.6%
cp (296.55 − 288.26)
∂( rWθ + ω r 2 ) ∂rCθ
= =0
∂m ∂m
∂I
=0
∂m
∂s
=0
∂m
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
∂Wm sin φ 1 ∂P
Wm − [Wθ + ω r]2 = −
∂m r ρ ∂m
∂( rWθ + ω r 2 )
=0
∂m
∂I
=0
∂m
3.5 From the uniform flow assumption, the total mass flow is
b δ b
b
˙ / (2π ) = ∫ rρ udy = r ρ eue ( b − 2δ * )
m
0
b δ b
∫ rρ u dy = r ρeue (b − 2δ ) + ∫ rρ u dy + ∫
2 2 2
rρ u2dy
0 0 b −δ
∫ rρ u dy = r ρeue (b − 2δ
2 2 *
− 2θ )
0
3.6 Conservation of mass for incompressible flow before and after mixing
requires
umix = ue (1 − 2δ * / b)
2
rb( Pe + ρ eumix ) = rbPe + r ρ eue2b[1 − 2(δ * + θ ) / b]
2
Pt,mix + 12 ρ eumix = Pt,e + ρ eue2 [ 12 − 2(δ * + θ ) / b]
Pt,e − Pt,mix = 1
2
ρ eue2 [(2δ * / b)2 + 4θ / b]
∂ ∂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
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.
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
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.
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.
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
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
8.1 From Eq. (8-11), the mass flow rate in a boundary layer is given by
∫ ρVmdy =ρeVme (δ − δ1 )
*
8.2 From Eqs. (8-11) and (8-12), the meridional momentum flux in a
boundary layer is given by
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
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
δ n n+ m
y y m
θ12 = ∫ − dy = δ
δ δ ( n + 1)( n + m + 1)
0
Ic = 12 [2( ∆pt′ )2,c − ( ∆pt′ )3,c + 2( ∆pt′ )2,c + ( ∆pt′ )3,c ]∆m
˙ = 2( ∆pt′ )2,c ∆m
˙
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)]
h3 − h1 = Uc2ψ c
∂R 1 ∂Cz1 ∂C ∂C ∂C
= Cz1 + Cθ1 θ1 − Cz2 z2 − Cθ 2 θ 2
∂r Uc2ψ c ∂r ∂r ∂r ∂r
∂Cz ∂C C2
Cz + Cθ θ = − θ
∂r ∂r r
Hence
∂R 1 rc Cθ22 Cθ21
rc = −
∂r ψ c r Uc2 Uc2
2
rc ∂(Cz2 / Uc ) r 2C ∂[rCθ 2 / ( rcUc )]
= − c θ2
2 ∂r rUc ∂r
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
∂R 1 rc Cθ22 Cθ21
rc = −
∂r ψ c r Uc2 Uc2
1 rc Cθ22 Cθ21
− =0
ψ c r Uc2 Uc2
Hence
(Cθ 2 + Cθ1)ψ c
=0
Uc
Cθ 2 = − Cθ1
2(1 − Rc) = 0
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.
13.3 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.
13.5 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.
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.
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.
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
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