Department of Mechanical and Aerospace Engineering

CARLETON UNIVERSITY

MECH 5401

TURBOMACHINERY

SUPPLEMENTARY COURSE NOTES

S.A. Sjolander
January 2010
 CARLETON UNIVERSITY
Department of Mechanical & Aerospace Engineering

MECH 5401 - Turbomachinery

COURSE CONTENTS

Week

1 Introduction. Review of similarity and non-dimensional parameters. Ideal versus non-ideal gases.
Velocity triangles.

2 Energy considerations and Steady Flow Energy Equation. Angular momentum equation. Euler
pump and turbine equation. Definitions of efficiency.

3 Preliminary design: meanline analysis at design point. Stage loading considerations. Blade
loading and choice of solidity. Degree of reaction.

4 Correlations for performance estimation at the design point for: axial compressors, axial turbines
and centrifugal compressors. Approximate off-design performance: compressor maps and turbine
characteristics.

5 Two-dimensional flow in turbomachinery. Spanwise flow effects. Simple radial equilibrium. Free-
vortex and forced-vortex analysis.

6 Actuator disc concept. Application to blade-row interactions. Through-flow analysis: governing

equations and computational implementation; role in design.

7 Blade-to-blade flow. Blade profile design considerations: boundary layer behaviour and diffusion
limits; significance of laminar- to turbulent-flow transition.

8 Three-dimensional flows in turbomachinery. Governing equations. Role of Computational Fluid

Dynamics (CFD) in turbomachinery design and analysis. Limitations of CFD.

9 Compressible flow effects: choking in turbomachinery blade rows; shock waves in transonic
compressors and turbine; shock-induced boundary layer separation; limit load in axial turbines.
Effects of compressibility on losses and other flow aspects.

10 Unsteady flows in turbomachinery. Fundamental role of unsteadiness. Significance of wake-

blade interaction. Approximate analysis of unsteady behaviour of compression systems: dynamic
system instability (surge); factors affecting compressor surge.

11 Current issues in turbomachinery aerodynamics. Very high loading for weight and blade-count
reduction. Effects of gaps, steps, relative wall motion and purge flow on blade passage flows.

12 Passive and active flow control to extend range of performance. Aero-thermal interactions. Multi-
disciplinary optimization.

S.A. Sjolander
January 2010
 Department of Mechanical and Aerospace Engineering
CARLETON UNIVERSITY

MECH 4305 - Fluid Machinery

TABLE OF CONTENTS

Page

1.0 INTRODUCTION

1.1 Course Objectives

1.2 Positive-Displacement Machines vs Turbomachines
1.3 Types of Turbomachines

2.0 NON-DIMENSIONAL PARAMETERS AND SIMILARITY

2.1 Dimensional Analysis - Review

2.2 Application to Turbomachinery

1. Non-Dimensional Parameters for Incompressible-Flow Machines

2. Effect of Reynolds Number
3. Performance Curves for Incompressible-Flow Turbomachines
4. Non-Dimensional Parameters for Compressible Flow Machines
5. Performance Curves for Compressible-Flow Turbomachines

2.3 Load Line and Operating Point

2.4 Classification of Turbomachines - Specific Speed
2.5 Selection of Machine for a Given Application - Specific
2.6 Size Cavitation

3.0 FUNDAMENTALS OF TURBOMACHINERY FLUID MECHANICS AND

THERMODYNAMICS

3.1 Steady-Flow Energy Equation

3.2 Angular Momentum Equation
3.3 Euler Pump and Turbine Equation
3.4 Components of Energy Transfer
3.5 Velocity Diagrams and Stage
Performance Parameters
1. Simple Velocity Diagrams for Axial Stages
2. Degree of Reaction
3. de Haller Number
4. Work Coefficient
5. Flow Coefficient
6. Choice of Stage Performance Parameters for
Design
 3.6 Efficiency of Turbomachines

1. Incompressible-Flow Machines
2. Compressible-Flow Machines

4.0 AXIAL-FLOW COMPRESSORS, FANS AND PUMPS

4.1 Introduction
4.2 Control Volume Analysis for Axial-Compressor Blade Section

1. Force Components
2. Circulation

4.3 Idealized Stage Geometry and Aerodynamic Performance

1. Meanline Analysis
2. Blade Geometries Based on Euler Approximation
3. Off-Design Performance of the Stage
4. Spanwise Blade Geometry

4.4 Choice of Solidity - Blade Loading Limits

4.5 Empirical Performance Predictions

1. Introduction
2. Blade Design and Analysis Using Howell’s Correlations
3. Blade Design and Analysis Using NASA SP-36 Correlations

4.6 Loss Estimation for Axial-Flow Compressors

1. Blade Passage Flow and Loss Components

2. Loss Estimation Using Howell’s Correlations
3. Loss Estimation Using NASA SP-36 Correlations
4. Effects of Incidence and Compressibility
5. Relationship Between Losses and Efficiency

4.7 Compressor Stall and Surge

1. Blade Stall and Rotating Stall

2. Surge

4.8 Aerodynamic Behaviour of Multi-Stage Axial Compressors

4.9 Analysis and Design of Low-Solidity Stages - Blade-Element Methods
5.0 AXIAL-FLOW TURBINES

5.1 Introduction
5.2 Idealized Stage Geometry and Aerodynamic Performance
5.3 Empirical Performance Predictions

1. Flow Outlet Angle

2. Choice of Solidity - Blade Loading
5.3.2.1 Zweifel Coefficient
5.3.2.2 Ainley & Mathieson Correlation
5.3.3 Losses

6.0 CENTRIFUGAL COMPRESSORS, FANS AND PUMPS

6.1 Introduction
6.2 Idealized Stage Characteristics
6.3 Empirical Performance Predictions

1. Rotor Speed and Tip Diameter

2. Rotor Inlet Geometry
3. Rotor Outlet Width
4. Rotor Outlet Metal Angle - Slip
5. Choice of Number of Vanes - Vane Loading
6. Losses

7.0 STATIC AND DYNAMIC STABILITY OF COMPRESSION SYSTEMS

7.1 Introduction
7.2 Static Stability
7.3 Dynamic
Stability -
Surge

Appendix A: Curve and Surface Fits for Howell’s Correlations for Axial Compressor
Blades Appendix B: C4 Compressor Blade Profiles
Appendix C: Curve and Surface Fits for NASA SP-36 Correlations for Axial Compressor
Blades Appendix D: NACA 65-Series Compressor Blade Profiles
Appendix E: Curve and Surface Fits for Kacker & Okapuu Loss System for Axial Turbines
Appendix F: Centrifugal Stresses in Axial Turbomachinery Blades
 Department of Mechanical and Aerospace Engineering
CARLETON UNIVERSITY

MECH 4305 - Fluid Machinery

Recommended Texts

S.L. Dixon, Fluid Mechanics, Thermodynamics of Turbomachinery, 5th ed., Elsevier Butterworth-
Heineman, 2005.

A short, inexpensive book which covers all the major topics, but sometimes a little too briefly.
Somewhat short on design information and data. Clearly written.

H.I.H. Saravanamuttoo, G.F.C.Rogers, H. Cohen, and P.V. Straznicky, Gas Turbine Theory, 6th ed.,
Pearson Education, London, 2008.

About gas turbine engines generally, but there are useful chapters on the three types of
turbomachines which are used most often in these engines: axial and centrifugal compressors and
axial turbines. These chapters contain methods and correlations which can be used in
preliminary aerodynamic design.

D. Japikse and N.C. Baines, Introduction to Turbomachinery, Concepts-NREC Inc./Oxford University

Press, 1994.

A recent book published for use with a short course offered by Concepts-NREC, a company in
Vermont which develops courses on various turbomachinery topics for industry. Reasonably
good. One of the few books on turbomachinery fluid mechanics which also addresses mechanical
design aspects (centrifugal stress, creep, durability, vibrations etc.).

B. Lakshminarayana, Fluid Dynamics and Heat Transfer of Turbomachinery, Wiley, New York, 1996.

A hefty, recent book written by the head (recently deceased) of turbomachinery research at Penn
State University. The emphasis is on more advanced topics, particularly computational
techniques. Brief and somewhat weak on fundamentals and the concepts used in preliminary
design. For these reasons, not well suited as a companion to this course. However, someone
continuing in turbomachinery aerodynamic design will probably want to have a copy of the book
in his/her personal library.

Additional Readings

The Library has a number of older textbooks on turbomachinery in which you may find material
of interest: see for example the books by Vavra, Csanady and Balje. The following books are ones I have
found particularly useful over the years. Some of them cover topics discussed in the present course
while others extend the material to topics which are beyond its scope.

D.G. Shepherd, Principles of Turbomachinery, Macmillan, Toronto, 1956.

A deservedly popular text book in its day. Now out of print, as well as somewhat
out-of-date. Nevertheless, it contains a lot of useful material and very lucid discussions on
most topics it covers.
 The following two, relatively short books were written by the man who subsequently helped to
found the Whittle Turbomachinery Laboratory at Cambridge University. He spent a number of
years as its Director. Good discussion of the design techniques which were current at the time
(and which still play a part in the early stages of design). Lots of practical engineering
information. They remain in-print thanks to an American publisher who specializes in
reprinting classic technical books which remain of value.

J.H. Horlock, Axial Flow Compressors, Fluid Mechanics and Thermodynamics, Butterworth, London,
1958, (reprinted by Krieger).

J.H. Horlock, Axial Flow Turbines, Fluid Mechanics and Thermodynamics, Butterworth, 1966, (reprinted
by Krieger).

The next book is by a more recent Director of the Whittle Laboratory. In the Preface he explicitly
disclaims any intention to present design information. However, it presents a detailed, relatively
up-to-date discussion of the physics of the flow in axial compressors, which is still very useful.

N.A. Cumpsty, Compressor Aerodynamics, Longman, Harlow, 1989.

The following book on radial machines (both compressors and turbines) is also published
published by Longman, like Cumpsty and Cohen, Rodgers & Saravanamuttoo. It is the least
satisfactory of the three, and is apparently going out of print. Nevertheless, worth being aware
of since most other available books on radial turbomachinery are quite old and rather out-of-date.

A. Whitfield and N.C. Baines, Design of Radial Turbomachines, Longman, Harlow, 1990.

To the extent that they present design information, the books by Horlock and Cumpsty reflect
largely British practice. The North American approach to axial compressor design was
developed by NASA (then called NACA) through the 1940's and 50's. The results are
summarized in the famous SP-36, and many axial compressors continue to be designed according
to it.

NASA SP-36, “Aerodynamic Design of Axial Compressors,” 1956.

AGARD, the scientific arm of NATO, organizes conferences, lecture series and specialist
courses on many aerospace engineering topics, including turbomachinery aerodynamics. The
following are two particularly useful publications which have come out of this activity.

A.S. Ucer, P. Stow and Ch. Hirsch eds., Thermodynamics and Fluid Mechanics of Turbomachinery,
Martinus Nijhoff, Dordrecht, Vol. I and II, 1985.

AGARD-LS-167, Blading Design for Axial Turbomachines, 1989.

 NOMENCLATURE FOR TURBOMACHINES

ENERGY TRANSFER TO THE FLUID

Fans
Blowers Gases
Incompressible flow
Pumps Liquids

Compressors Gases Compressible flow

Propellers Both Both

ENERGY TRANSFER FROM THE FLUID

Turbines
Turbo-expanders
Wind mills/Wind turbines
 1. NON-DIMENSIONAL PARAMETERS AND SIMILARITY

2. DIMENSIONAL ANALYSIS - REVIEW

Non-dimensional parameters allow performance data to be presented more compactly. They can also
be used to identify the connections between related flows, such as the flow around a scale “model” and that
around the corresponding full-scale device (sometimes called the “prototype”).

Two flows are completely similar (“dynamically similar”) if all non-dimensional ratios are equal for
the two flows. This includes geometric ratios, which are needed for “geometric similarity”. For example, if
the flows around two geometrically-similar airfoils are dynamically similar, then

 lift force   L

 drag force  prototype
 model

Similarly for other force ratios, velocity ratios, etc. D

For a given case there is only a limited number of independent non-dimensional ratios: these are the
“criteria of similarity”. If the criteria of similarity are equal for two flows, all other non-dimensional ratios
will also be equal, since they are dependent on the criteria of similarity.

Finding Criteria of Similarity:

(1) List all the independent physical variables that control the flow of interest (based on experience,
judgment, physical insight etc.). For example, consider again the airfoil flow. Assume that the flow
is compressible and the working fluid is a perfect gas.

For a particular airfoil shape, the flow is completely determined by:

c - chord
α - angle of attack
U - freestream velocity L
ρ - fluid density μ - fluid
viscosity R - gas
constant
a - speed of sound
γ - specific heat ratio 
D
Note that the pressure and temperature are not quoted. U
M
For a perfect gas,
c
P   RT a   RT

Thus, by specifying a, γ and R, we have implicitly specified T. Similarly, with ρ and R specified, and
T implicitly specified, then P is implicitly specified through the perfect gas law. Therefore, for our
particular choice of independent variables, P and T are just dependent variables. All other quantities,
such as the lift, L, and drag, D, likewise depend uniquely on the values of the independent variables.
(2) Form non-dimensional groups from the independent variables.

Buckingham’s Π Theorem gives the number of independent non-dimensional ratios which exist:

If n = no. of independent physical variables

r = no. of basic dimensions (eg. Mass, Length, Time, Temp. (θ), etc.)

Then (n - r) criteria of similarity exist

eg. for the airfoil n=8

r = 4 (M, L, T, θ)
ˆ (n - r) = 4

ie. there are 4 criteria of similarity

Form the criteria of similarity by inspection, or using dimensional analysis.

eg. for the airfoil, we can non-dimensionalize the density as follows:

1  U  Uc
 c 
  
M LT  L
 
3
L
L
M
which is clearly the Reynolds number, Re
T
α is already non-dimensional and can be used directly as a criterion of
similarity
γ is also already non-dimensional
U
 Mach number, M
a

Thus, for the airfoil 4 suitable criteria of similarity are: Re, M, α, and γ. If these are matched between
two geometrically similar airfoils, the two flows will be dynamically similar.

(3) All other non-dimensional ratios are then functions of the criteria of similarity.

Take each dependent variable in turn and non-dimensionalize it using the independent variables.
eg. for the drag of airfoil (per unit span), D
D
D 1

1

1  or  CD
 1 U2c
 U2 c D 2 
L
1
3 1 U 2 c
M L T2 
 T2 M LL
L2

then CD  f Re, M ,  , 


 Similarly for all other dependent non-dimensional ratios (CL, Cm, etc.).

Any non-dimensional ratios we develop could also be combined, by multiplication, division etc., to
form other valid non-dimensional ratios. This does not provide any new information, simply a rearrangement
of known information. However, the resulting ratios may be useful alternative ways of looking at the
information. For example, for the airfoil, having derived C D and CL then

CL L

CD D

is another valid (and in fact useful) non-dimensional parameter.

2.2 APPLICATION TO TURBOMACHINERY

2.2.1 Non-Dimensional Parameters for Incompressible-Flow Machines

For now, consider just pumps, fans, and blowers. Hydraulic turbines will be discussed briefly in
Section 2.4.

W˙
Q

For a given geometry, the independent variables that determine

performance
D - are usually
machinetaken as
size (usually rotor outside diameter)
.ρ - fluid density
μ - fluid viscosity
N (or ω) - machine speed; revs or rads per unit time
Q- volume flow rate through the machine

Note that the choice of independent variables is somewhat arbitrary. One way to visualize what are
possible independent variables and what are dependent variables is to imagine a test being conducted on the
machine in the laboratory. The variables which, when set, fully determine the operating point of the machine
is then one possible set of independent variables. In the laboratory test, one might set the rotational speed (by
controlling the drive motor) and the flow rate (by throttling at the inlet or outlet ducts). With N and Q set, the
head or pressure rise produced or power absorbed are then dependent functions of the characteristics of the
machine. Alternatively, if the throttling valve is adjusted to produce a particular pressure rise, then we lose
control over the flow rate and it becomes a dependent variable. The independent variables listed above are the
most common choices for incompressible flow machines that raise the pressure of the fluid. All other
variables are then dependent. For example
 ΔH - total head rise across machine (or sometimes, total pressure rise)
W˙ shaft power absorbed by the machine
- torque absorbed by the machine
T - efficiency of the machine
η -
H  f1 D, N , Q, , 

W˙  f 2 D, N ,Q, , etc.


Applying Buckingham Π Theorem:

n=5 n - r = 2 (ie. are 2 criteria of similarity)

r = 3 (M, L, T)

(1)
Form theFlow rate:
criteria of similarity:
1 1 Q
Q  D3

 N N D3
T 1
L3   L3
1
T
This is known as the flow coefficient, capacity coefficient or flow number

(2) Fluid properties (specifically, viscous effects):

 N DD  N D2
 ie. the Reynolds number
 

All other non-dimensional ratios or coefficients then depend on these two criteria of similarity.

For power coefficient (non-dimensional work per unit time)

1 1 1
   D5  W
W N3   N˙ 3 D5
˙
 M L2 
 T2  
1
3
 T  L3 L5
T
1 M

then W˙  f  Q ,  N D 
2

 N 3 D5  
 N D3

Obviously, μ rather than ρ could have been used to cancel the M appearing in W˙ . It can easily be shown
that the resultant power coefficient would be the one derived here multiplied by the Reynolds number.
 Next consider the total head rise, ΔH, across the machine. By definition, the total head H is given by

P V2
H  z
g 2g

 static head  dynamic head  elevation

and H can be interpreted physically ashead
the mechanical energy content per unit weight. However, the energy
content is more commonly expressed on a per unit mass basis:

g H  mechanical energy per unit mass

We therefore create a non-dimensional head coefficient as follows:

1 1 gH
g H N2
 D2 
 N 2 D2
L T2
1
T2
L  L2
 1

Sometimes the head rise ΔH is simply written H. As with the power coefficient, the head coefficient is a
dependent function of the two criteria of similarity:

gH   
 f  Q ,  N D 
2
 or gH 
N 2 D2
 
 N 2 D2   N D3

The g is also sometimes dropped to give H/N2D2, but the head coefficient is then dimensional and will take
different values in different systems of units.

A corresponding total pressure coefficient can be obtained from

gH  gH  P0
 
N 2 D2  N 2 D2  N 2 D2

since ρgΔH has units of pressure.

Using the conventional definitions, efficiency is already non-dimensional. For pumps, fan and
blowers, the efficiency is usually defined as:

useful power transferred to fluid fluid power

 pump  
input power shaft power
and
fluid power  mass flow rate  mechanical energy change per unit
mass

 m˙  gH
 QgH
Thus  QgH
 pump 
W
 ˙
 Q 3   gH
2 2

 ND  N D 

 
 W3 5 
  N˙ D 
Flow Coefficient


shaftCoefficient
Head power
turb   W˙
fluid power Power Coefficient
 QgH
Power Coefficient
Similarly, for turbines: 
Flow Coefficient  Head Coefficient
2.2.2 Effect of Reynolds Number

We have shown that in general for incompressible flow:

gH W˙
, ,  , etc. ⎛ Q ,  N D 2 |⎞
fns|  ⎠
N 2 D2 5  ⎝ N D3
N D
3

⎛ ⎞
| , Re|
⎝ N D3
Q ⎠
 fns

The flow in most turbomachines is highly turbulent. Therefore, most frictional effects are due to
turbulent mixing. Viscosity has a minor direct effect and losses tend to vary slowly with Re: recall from the
Moody chart that in pipe flow the friction factor varies much more slowly with Re for turbulent flow than for
laminar flow. Thus, if the Reynolds numbers are high and the differences in Re are not too large between the
machines being compared, Re is often neglected as a criterion of similarity. We can then use, as an
approximation

⎞
gH W˙ ,  , etc. ⎛ Q only|
, fns|
N 2 D2 5  ⎝ N D3 ⎠
N D
3

Where Re variations can not be neglected, a number of empirical relations have been proposed for
correcting for the effect of Re on efficiency. These corrections typically take the form

1 ⎛ ⎞
n
P
 M (1)
1Re  M | |
⎝ Re P ⎠

where ReM is the smaller of the two values of the Reynolds number and n varies with the type of machine and
Reynolds number level. For example, the ASME Power Test Code (PTC-10, 1965) suggests the following
values:

n = 0.1 for centrifugal compressors

n = 0.2 for axial compressors

if ReM≤ 105, where Re = ND2/v (ie. the tip Reynolds number). Note that (1) indicates that efficiency improves
with increasing Re.
Reynolds Number Based on Blade Chord

Taken from: AGARD-LS-167

2.2.3 Performance Curves for Incompressible-Flow Turbomachines

Relationships such as

gH ⎛ ⎞
2 2 Q | (neglecting Re)
f ⎝| N D 3 ⎠
N D 

imply that if we test a family of geometrically-similar, incompressible-flow machines (different sizes, different
speeds etc.), the resulting data will fall on a single line if expressed in non-dimensional form. For example, the
non-dimensional coefficients for a pump of fan might appear as follows (we will discuss later why the curves
will have the particular trends shown):

Likely "Design Point"

Coefficients

W˙

N 3 D 5

gH
N2 D2

Q
ND3

The thick curves are used to suggest variations which could be due to the neglected Re effects, and perhaps
some secondary effects which were not included in the original list of independent parameters (e.g. mild
compressibility effects for a fan or blower). The dashed line indicates the likely "design point": the preferred
operating point, since the efficiency is best there.

Because of the universality of the performance curves, the tests could be conducted for a single
machine and the results used to predict the performance of geometrically similar machines of different sizes,
different operating speeds, and even with different working fluids.

Note again that there is flexibility in the choice of dependent and independent parameters. See P.S. #1
Q 1 for the form of non-dimensional parameters which are often used for hydraulic turbines.
2.2.4 Non-Dimensional Parameters for Compressible-Flow Turbomachines

We now develop the criteria of similarity for compressible-flow turbomachines. Assuming the
working fluid is a perfect gas, a suitable list of independent variables which control performance is as follows:

 a01   P01 
N , D, m˙ ,  or  ,  or  , , R,
  01   01 
T   

where 

m˙ = mass flow rate (rather than Q as measure of flow

rate)

a01   RT01 (stagnation speed of sound)

ˆ could (perfect
use T01 rather
gas) than a01
P01  01 RT01

ˆ can use ρ01 or P01, as convenient

(N.B. temperatures and pressures must be absolute values)

Then from the Buckingham Π Theorem:

n=8 r=4 (M, L, n - r = 4 (4 criteria of similarity)

T, θ)

By inspection, the 4 independent coefficients are:

ND
(1) speed parameter (effectively the tip Mach number)
a01
(2) flow parameter (effectively the axial Mach number)
 01 Dm2˙a01

 01 N D2
(3)  D  1 or we could use again
m˙ Re


Cp specific heat ratio (which is already non-dimensional)

(4)   Cv

All other performance coefficients are then functions of these four coefficients (as always, geometrical
similarity is assumed).
Dependent performance coefficients:

The main change from incompressible-flow machines is in the form of the pressure change
coefficient. Instead of the head or total pressure coefficient, we conventionally use the pressure ratio:

P02 P02  machine outlet total pressure

P
01

Then


 ND m , Re,  (1)
P02 W 
P01 , , , etc.  fns a ,  a 
˙ D5
01 N 3 01 01 ˙ 01 D 2


The form of the independent coefficients used here is very general. The main assumption that has
been made is that the working fluid is a perfect gas. We can make use of some of the perfect gas expressions
to rewrite the independent parameters in a somewhat more convenient form:

(1) Speed coefficient:

ND ND N
 
a01  RT01
(2) Flow coefficient:
D T01 
m R 1
m˙  m˙ R  T01
01 a 01 D 2 ˙ P01 
P01  RT01 D2
D2
RT01

Then (1) can be written

 N D m R 1 
P02 , W
, , etc.  fns T , P
T01 , Re,  (2)
P01 ˙ 3 D5
01 N  01  ˙ 01  D2 
 
R

This is the form of the parameters that is appropriate for the most general case, where we are relating the
performance of geometrically-similar, compressible-flow turbomachines of different sizes and operating with
different working fluids (both of which are perfect gases).

In practice, the parameters are often simplified somewhat according to specific circumstances.

In many cases, the same working fluid (eg. air) will be used for both the model and prototype. Thus,
R and γ are often known constants and it is somewhat tedious continually to have to include them in the
calculation of the coefficients. If we then omit the known, constant fluid properties we can write:
 
P02 , W  ND m T01 (3)
P01 , , etc.  fns T , ˙P 2 , Re
˙
01 N 3 D5  01 01 D 


This form of the coefficients is suitable for relating geometrically-similar machines with different sizes but
with the same working fluid. Note that by assuming the same working fluid, we have reduced the number of
criteria of similarity by one. The main disadvantage to this form of the coefficients is that the speed and flow
coefficients are now dimensional and we must specify what system of units we are working in.

If the performance curves are intended to represent the performance of a particular machine operating
at different inlet conditions, then D is a known constant and is often omitted:

N m 
P02 , W˙ T01 (4)
P01 , , etc.  fns T , ˙ P , Re

01 N 3 D5  01 01 

This is the form of the independent coefficients typically used to present the performance characteristics of the
compressors and turbines for gas turbine engines.

As with incompressible-flow machines, it is sometimes possible to neglect Re as a criterion of

similarity (by the same arguments used in Section 2.2.2). Note that the speed and flow coefficients are again
dimensional.
2.2.5 Performance Curves for Compressible-Flow Turbomachines

If we can neglect the Reynolds number effects, Eqns. (3) and (4) indicate that our performance curves
will take the form:

m
P02  f1 ⎛| ND , T01 ⎞| etc.
| T ˙P D 2 |
⎝ 01 01 ⎠
P01

Thus, whereas our performance tests for the incompressible-flow machines led to a single curve for each
dependent performance coefficient, for compressible-flow machines we will obtain a family of curves.

The resulting performance diagrams for compressible-flow compressors and turbines would then look
as follows (again, we will discuss the reasons for the detailed shape of the characteristics later in the course):

(a) Compressor ("Compressor Map")

LINE OF CONSTANT
P02
ND
SURGE LINE
P01 (UPPER LIMIT OF
T01
STABLE
OPERATION)

CHOKING

ND
INCREASING

T01

m˙
TP0101 D
2

Implicitly, this map applies for one value of some reference Reynolds number. If the effects of Re can not be
neglected, then we would have to generate a series of such graphs, each one containing the performance data
for a different value of the reference Re.
(b) Turbine Characteristic:

P02 STATORS CHOKED

P
01

ND
LINES OF CONSTANT

T01

m˙ T01
P01 D2

In a gas turbine engine, the pressure ratio developed by the compressor is applied across the turbine at
the hot end of the engine. The mass flow rate swallowed by the turbine and its power output are then
dependent functions of the turbine characteristics. That is, as far as the turbine is concerned the pressure ratio
is imposed and is effectively an independent parameter. When presenting performance data, we generally plot
independent parameters on the “x axis” and dependent parameters on the “y axis”, as was done on the
compressor map. By this argument, the turbine characteristic should be presented as:

m˙
TP0101 D
2

ND
CONSTANT

T01

P02

P01

and this is in fact the way turbine characteristics are generally

presented in the gas turbine business.
NASA 8-Stage Research Axial Compressor
2.3 LOAD LINE AND OPERATING POINT

The performance diagrams discussed in the earlier sections present a wide range of conditions at
which the machine can operate. For example, the compressor in the last section can operate stably at any point
to the right of the surge line. The precise point at which a turbomachine actually operates depends on the load
to which it is connected.

(a) The simplest case is a compressor or pump connected to a passive load (e.g. pipe line with valves,
elbows etc.). At the steady-state operating point we must have:

(1) Qmachine  Qload (or, for compressible flow, m˙ machine  m˙

load )
(2)
Hmachine  Hload (or P0,machine  P0,load )
Thus, the operating point is where the machine and load H vs Q , or P0 vs m˙ ,
characteristics intersect.

e.g. Suppose a pump is supplying flow to a pipe line. The head drop along the pipe varies with V2
(or Q2), as determined from the friction factor (e.g. Moody chart) and the loss coefficients of any other
components in the pipe system. The resulting ΔH vs Q variation is known as the load line for the
system. The head rise produce by the pump is a function of the flow rate and the rotational speed.
Then if the pump is run at N1, the operating point will be A, etc.

H LOAD
LINE

C
PUMP CHARACTERISTICS
AT CONSTANT SPEED

N3
A

N2

N1

(b) For a gas turbine engine, the operating points of the compressor and turbine are determined by
compressor/turbine matching conditions (a propulsion nozzle will also influence operating points - see
Saravanamuttoo et al., Ch. 8 & 9).

m˙ fuel

COMBUSTOR

W˙ C W˙out
COMPRESSOR TURBINE

m˙
C m˙
T
For the simple shaft-power engine shown, the matching conditions would be:

m˙ T  m˙ C  m˙
fuel

NC  NT

W˙ T  W˙ C 
W˙ o u t

(c) In hydro-power installations, total head across the turbine is imposed by the difference in
elevation between reservoir and tailwater pond (minus any losses in the penstock). Since

W˙T  T  gQH

to produce varying power (according to electrical demand), it is necessary to vary the equilibrium Q, at fixed
ΔH. Furthermore, since the electricity must be generated at fixed frequency, we do not have the option of
varying N to achieve different operating points. The solution to this is to vary the geometry of the machine.
This can be done with variable inlet guide vanes or withCONSTANT
variable rotor blade pitch.
SPEED LINES -
SAME SPEED,
H DIFFERENT BLADE
SETTINGS
1 2 3

LOAD LINE
NEGLECTING FRICTION

LOAD LINE
INCLUDING FRICTION

Q
EXAMPLE (Section 2.3):

A pump is connected to the piping system shown. What flow rate of water will be pumped for
the two valve settings?

VALVE K = 1, K=10

K = 0.9 K = 0.9

K = 0.9 K = 1 (EXIT LO SS)

K = 0.9
6 m.

WATER K = 0.9

PUMP
K = 0.5 (ENTRY LOSS)

Pipe diameter: mm (smooth)

dpipe  50

Pipe length: m
L  125
Viscosity (water): m2/s
6
  10
The pump has the characteristics shown in the plot, and the following information applies to the
pump:

Pump Characteristics
Pump diameter: D  30 cm 4

Pump speed: N  1750 RPM

Q 3
Flow coefficient:
3
ND
Head Coefficient

Head coefficient: 2
gH
2 2
N D

(with N in revs/s in the coeffcients)

1

0
0 0.002 0.004 0.006 0.008 0.01
Flow Coefficient
2.4 CLASSIFICATION OF TURBOMACHINES - SPECIFIC SPEED

Neglecting Reynolds number effects, for a given family of geometrically-similar incompressible-flow

turbomachines the efficiency is a function of one criterion of similarity only. Normally we use the flow
coefficient as the independent parameter. That is

FAMILY B

FAMILY A
 Q 
  f only
 3

 ND

Q
ND3

Thus, the maximum 0 will occur for this family (say family A) at some particular value of Q/ND3. For
another family of machines, the maximum 0 might occur at a different value of Q/ND3. We could therefore
classify turbomachines according to the value of Q/ND3 at which they produce the best efficiency. Then if we
knew the value of Q/ND3 that we required in a given application, we would choose the machine that gives the
best value of efficiency at that value of Q/ND3. Unfortunately, this idea presupposes that we know the
diameter of the machine. In general, this will not be the case. We therefore look for an alternative parameter
to Q/ND3 that does not involve the size of the machine to use as a basis for classifying families of
turbomachines.

We can always form valid new non-dimensional parameters by combining existing ones. Combine
the flow and head coefficients to eliminate D:
1
 Q 2
 3
 1

 ND  NQ 2
3
 3

 g H  4 gH 4

 
 N 2 D2 

Following convention, we then define

1
 Q2
 3
 gH 4

where T is in radians/s so that S is truly non-dimensional. Conceptually, we could then plot the efficiencies
of various families of turbomachines against S (rather than Q/ND3) and note the value of S at which each
family achieves its best 0. This value of S is known as the specific speed for that family of machines. The
next figure (taken from Csanady) shows the values of specific speed that are observed for various types of
turbomachines:
 A number of more detailed summaries of specific speed have been presented over the years.
Unfortunately, the non-dimensional form of the specific speed has not been used consistently. The following
table can be used to convert between the various definitions used:

AREA OF APPLICATION SPECIFIC SPEED EQUIVALENT S

FANS, BLOWERS AND RPM cfs

COMPRESSORS N S1
NS 1  3 
(BRITISH UNITS)
ft 4 129

PUMPS RPM USgpm NS 2

N  
(AMERICAN
MANUFACTURERS) S2
2730
3

ft 4

HYDRAULIC TURBINES RPM HP NS 3

N  
(BRITISH UNITS)
S3 5
42
(IF WORKING FLUID IS WATER)
ft 4

HYDRAULIC TURBINES RPM metric HP NS 4

N  
(METRIC UNITS)
S4 5 187
(IF WORKING FLUID IS WATER)
m4
FANS, BLOWERS AND
COMPRESSORS
RPM m3 s NS 5
(METRIC UNITS)
NS 5  3

53
m4
Several plots showing the specific speeds for various classes of machines are given on the next pages.
In addition to giving the values of specific speed, the plots can also be used for initial estimates of the
efficiencies that can be expected. These efficiencies apply for machines that are well-designed, correctly sized
for their applications, and operating at their design points.
 Hydraulic turbines are usually characterized according to their output power rather than the flow rate.
Since shaft power output is related to the flow rate by

W˙ t  t  QgH

we can rewrite the specific speed as

 Q 
  W˙
3 5
 g H 4   gH
4

In practice, 0, D and g are usually dropped, and T is replaced by N (usually in RPM). Thus, the
"power specific speed" normally used with hydraulic turbines is

N S  N W5
˙
H 4

The following figure (from Shepherd, 1956) shows the variation of the power specific speed for hydraulic
turbines of different geometries.
 The plots shown above were based on data that is as much as 50 years old. One might expect that
over time the efficiency of all types of machines would improve as a result of the application improved design
tools such as computational fluid dynamics. This is illustrated in the following figure which shows the
variation of efficiency with specific speed for compressors. The baseline data, taken from Shepherd (1956),
dates from 1948 or earlier. Japikse & Baines (1994) compared more recent compressor data with the plot from
Shepherd and concluded that efficiencies had improved noticeably since Shepherd’s time. They also projected
that there would be further improvements by 2000, as shown in the figure.

0.9

Axial-Flow
0.8 Machines
Efficiency, 

0.7 Centrifugal
Machines

0.6 Positive-Displacement
Machines

0.5 Shepherd (1956): 1948 Data

Japikse & Baines (1994): 1990 Data
Japikse & Baines (1994): 2000 Projected

0.4 1
10 20 40 60 80 1002 400 600 1000
200
Specific Speed, NS
2.5 SELECTION OF MACHINE FOR A GIVEN APPLICATION - SPECIFIC SIZE

The selection starts from the required “duty”: the conditions at which it is intended to operate:

For pumps, compressors N, Q and ∆H (or ∆P0) are typically specified.

For turbines
N, W˙ and ∆H (or ∆P0) are typically
In practice, a precise value of N may not bespecified.
known, but it is often constrained to specific values by the fact
that, for example, electrical motors come with certain maximum speeds according to the number of poles.
There may also be mechanical constraints (e.g. maximum tip speed, because of centrifugal stress
considerations). Often the selection process will involve varying the speed to get a specific speed which
results in good efficiency.

From the duty, one can work out the specific speed and then use the figures in Sec. 2.4 to select an
appropriate type of machine. However, the efficiencies shown on the figures will be achieved only if the
machine is well-designed and correctly sized. Size is important because:

(a) if machine is too small: high flow velocities, and since frictional losses vary as 0.5pV2
(and with gases, shocks can occur), the efficiency will be poor;
(b) if machine is too big: low velocities, low Reynolds numbers, boundary layers will be thick and
may separate, again reducing the efficiency; also, machine will be expensive.

In Sect 2.4, we noted that for a given family of machines the peak occurs for a particular Q/ND3. In effect,
having chosen a suitable machine, knowing Q and N, we want to pick D to get the appropriate Q/ND3.
However, efficiency data for turbomachines has not in fact been correlated in this form. Instead of using
Q/ND3, we define a new parameter, the "specific size" ∆:

1
DgH 4

Q

The specific size for a given machine is then the value of ∆ at which it achieves its best efficiency. The value
of ∆ depends on the machine type (i.e. fi ) and to some degree on its detailed design. However, in the early
1950s Cordier examined the data for a wide range of well-designed, actual machines, and found that ∆
correlated quite well with fi alone: the correlation is summarized in the Cordier diagram (see over).
Summarizing:

To get best efficiency for a specified duty:

(1) Select the machine type such that its fi is

「
|  Q |
 | 3 |
 |Lg H 4
duty

|]
(2) From fi , read ∆ from the Cordier diagram and size the machine such
that
「
| DgH 4 |
1
| | 
| Q 
|]
L
duty
2.5 SELECTION OF MACHINE FOR A GIVEN APPLICATION - SPECIFIC SIZE

The selection starts from the required “duty”: the conditions at which it is intended to operate:

For pumps, compressors N, Q and ) H (or )P 0 ) are typically specified.

For turbines
N, W˙ and ) H (or )P 0 ) are typically
In practice, a precise value of N may not bespecified.
known, but it is often constrained to specific values by the fact
that, for example, electrical motors come with certain maximum speeds according to the number of poles.
There may also be mechanical constraints (e.g. maximum tip speed, because of centrifugal stress
considerations). Often the selection process will involve varying the speed to get a specific speed which
results in good efficiency.

From the duty, one can work out the specific speed and then use the figures in Section 2.4 to select an
appropriate type of machine. However, the efficiencies shown on the figures will be achieved only if the
machine is well-designed and correctly sized. Size is important because:

(a) if machine is too small: there will be high flow velocities, and since frictional losses
vary as 0.5DV2 (and with gases, shocks can occur), the efficiency will be poor;
(b) if machine is too big: there will be low flow velocities, low Reynolds numbers, boundary
layers will be thick and may separate, again reducing the efficiency; also, the machine will be
expensive.

In Section 2.4, we noted that for a given family of machines the peak 0 occurs for a particular Q/ND3. In
effect, having chosen a suitable machine, knowing Q and N, we want to pick D to get the appropriate Q/ND3.
However, efficiency data for turbomachines has not in fact been correlated in this form. Instead of using
Q/ND3, we define a new parameter, the "specific size" ) :
1
DgH 4

Q

The specific size for a given machine is then the value of ) at which it achieves its best efficiency. The value
of ) depends on the machine type (i.e. S) and to some degree on its detailed design. However, in the early
1950s Cordier examined the data for a wide range of well-designed, actual machines, and found that )
correlated quite well with S alone: the correlation is summarized in the Cordier diagram (see over).
Summarizing:

To get best efficiency for a specified duty:

(1) Select the machine type such that its S is



  Q 
  3 
g H 4 

duty
(2) From S, read ) from the Cordier diagram and size the machine such that

 1
 DgH

4  
 
 Q 

duty
Example (Section 2.5):

A small hydraulic turbine is to deliver a power of 1000 kW. The total head available is 6 m. and the
turbine is directly connected to an electrical generator which is to deliver power at 60 Hz.
(a) What is the required flow rate?
(b) Determine a suitable type, size and speed for the turbine.
2.6 CAVITATION

If the local absolute static pressure falls below the vapour pressure of a liquid, it will boil, forming
vapour cavities or bubbles. This is known as cavitation. When the bubbles collapse, brief, very high forces
are created which can cause rapid erosion of metal surfaces. Cavitation will also cause significant
performance deterioration. Thus, cavitation should be avoided.

Cavitation is a danger on the low-pressure ("suction") side of the machine: the inlet for pumps, the
outlet for turbines.

Define the Net Positive Suction Head (NPSH):

Hsv  Habs  hv

where Habs is the absolute total head at the suction side of the machine, defined as

2
H abs  Pabs V 
 
  g 2g  suction side

where Pabs is the absolute value of the static pressure and V is the fluid velocity, both on the lower pressure or
suction side of the machine. hv is the head corresponding to the vapour pressure of the liquid,
P vap

hv
g

Note: Habs is not the usual total head H since it does not include the elevation term. In fact H abs = P0/ρg.
At the minimum pressure point on the suction side of the machine, the local static head will be less than the
total head, Habs, but directly related to it. Thus, the onset of cavitation will occur for some critical, positive
value of Hsv.

P01
P 1
 1
2

2 V 1
 V2 2 P01
P1 2
1
 gH S V  2
2

2 V
P2  gH SV critical
P1
P2
Pv Pv
f T

o
 We non-dimensionalize Hsv to obtain the "suction specific speed", S

 Q
S 3
 gHsv  4
For a given machine there will then be some critical value of S ( = Si, “i” for cavitation “inception”),
corresponding to the critical value of Hsv, at which cavitation will start. If

S < Si

then there is no cavitation. The higher the value of Si, the more resistant the machine is to cavitation.

The value of Si can be found experimentally by holding Q and N constant (i.e. Q/ND3 constant) while
reducing the pressure on the suction side of the machine and observing the ΔH or η behaviour. For example,
for a pump a valve in the intake pipe can be used to reduce gradually the inlet total head while an outlet valve
can be used to maintain the constant the flow rate. Plot the results versus the resulting values of S:

H

DATA FOR CONSTANT

Q
ND3

INCEPTION

Si S

At cavitation inception, the blade passages fill with vapour and ΔH and η drop drastically.

The value of Si depends in the detailed design of the machine (e.g. surface curvatures in the low-
pressure section of the blade passage). However, for machines which have been properly designed to avoid
cavitation it has been found that the values of S i are fairly similar:

For pumps: Si . 2.5 - 3.5 N.B.: near the design

For turbines: point Si . 3.5 - 5.0

Recall that a higher value of Si means a machine more resistant to cavitation.

The Thoma Cavitation Parameter, σ, is also sometimes used:

Hsv crit
 
H
where Hsv crit is the critical value of Hsv : that is, the value at cavitation inception. However, the value of
σ will vary with the details of the design of the machine. This can be illustrated by considering two pump
impellers that have identical inlet geometries:

2
1

D2
D1

If the pumps are run at the same rotational speeds and flow rates, the flow in the inlet region will be identical.
Thus, they should cavitate at the same values of Hsv. Then since

 Q
S 3
 gHsv  4
it follows that the two machines have the same critical value of S: Si1 = Si2. However, the two rotors do not
have the same value of ΔH. In fact, the larger rotor will produce a significantly larger ΔH because of its
higher tip speed (ΔH varies as (ND)2, as implied by the form of the head coefficient; see also later sections).
Thus, at cavitation
H sv H sv crit ,2

 1 crit ,1  2 
H1 H 2

since ΔH1 < ΔH2. Consequently, the Thoma parameter should be used only within a geometrically-similar
family of machines. For example, a critical value of σ determined from model tests can be used to predict the
conditions for the onset of cavitation in another member of the same family.

Since cavitation is a significant danger to the machine, checking for cavitation should be a normal
part of selecting a hydraulic machine for a particular duty.
EXAMPLE (Section 2.6): In Section 2.5 we selected a hydraulic turbine for the following service: W = 1000kW, H =
6 m. An axial-flow (propeller or Kaplan) turbine was chosen, with a diameter of 2.7 m, a flow rate of 18.9
m3/sec and running at 180 RPM. What is the maximum height above the tailwater level that thi turbine can
be installed if cavitation is to be avoided? The draft tube is a length of diffusing duct at the exit of the
turbine. Assume that the draft tube has an outlet area of 6 m2 and the outlet is 3 m below the turbine. The
water is at 20 oC for which Pv = 2.3 kPa. Patm = 101.3 kPa. Assume that the tailpond is large compared
with the draft tube outlet so that the flow is effectively being dumped into a very large reservoir at the draft
tube outlet.

6m

h
3m

TAIL POND

DRAFT TUBE OUTLET

1. FUNDAMENTALS OF TURBOMACHINERY FLUID MECHANICS
AND THERMODYNAMICS

2. STEADY-FLOW ENERGY EQUATION

Consider a control volume containing a turbomachine:

1 2

m m
˙ ˙
W˙ shaft

Q˙

For steady flow, conservation of energy can be written

Rate of energy flow into CV + Rate of energy addition inside = Rate of energy
flow out of CV

E 1 dm˙   d Q˙  W˙ shaft  E 2 d m˙

If the energy content is the same for all fluid entering or leaving the CV (or using mean values) SFEE can be
m˙ E1  Q˙  W˙ shaft  m˙ E2 (1)
written
where m
˙ = mass flow rate of fluid
E = energy per unit mass for fluid
Q = rate of heat transfer to the machine
W˙ ˙ sha = shaft power into the machine
ft
The energy content of the fluid includes thermal and mechanical components:

P C2
E  u    2  gz (2)
 
thermal  mechanical

 h  C 2  gz
2
where u = internal thermal energy per unit mass (=
CvT) P/ρ = flow work (“pressure energy”) per unit mass
C = absolute velocity of fluid
C2/2 = kinetic energy per unit mass
gz = potential energy per unit
mass
h = P/ρ + u = enthalpy per unit
mass
 For a turbomachine at steady state, the flow is essentially adiabatic, Q˙  0 . For gases, we
usually neglect potential energy changes. Then SFEE can be written

 C2   C2
W˙ sha  m˙  h2  2    h1  1  
ft   2   2 
 m˙ h02  h01  (3a)


where 2
h0  h C  stagnation
2 enthalpy

C P T0 for perfect gases
For general non-uniform flows, we would write

W˙ shaft   h0 dm˙   (3b)

h0 d m˙
2
For incompressible flow , temperature (i.e. internal energy, u) changes only due to frictional heating,
1
since ρ is constant and we have already assumed the process is adiabatic. In order to separate the frictional
effects from other effects, we retain the internal energy separate from the flow work:

 m˙  P   P
    
2
 2  C 2  gz    C2 
(4)
 or  u  21  gz1  
1
W shaft 
˙  
2
u  2 2
 
  
  Q 1

It is also common to write 

u2  u1
g  H L  " total head loss" due to friction inside the
machine

The total head is a measure of the total mechanical energy content of the fluid

P C2
H  total head
 z
 g 2g

Then for an incompressible-flow compression machine (eg. a pump or blower) (4) can be written

W˙ shaf   Q g H2  H1  H L (5)

t   Q g H   Q g L
H
ΔH = H2 - H1 is the total head rise that appears in the fluid between the inlet and outlet of the machine. It is
the ΔH which was used in the head coefficient, (gΔH/N 2D2), and ρQgΔH is what was referred to earlier as the
“fluid power”.
 We defined the efficiency of a pump or blower as

 pump fluid power

 shaft power

then
pump QgH
 QgH  QgH
L
1 (6)

H
1  HL

As shown later, we have ways to estimate the various contributions to HL (eg. frictional losses at the walls vary
as V2). We can then use (6) to estimate the resulting efficiency of the machine.

For incompressible-flow expansion machines (i.e. turbines),

  Q g H   Q g H L
W˙ shaf
t
since the friction inside the machine now reduces the shaft power output compared with the fluid power
released by the fluid, as given by ρQgΔH. We then define turbine efficiency

shaft power out

turbine
 fluid power

Efficiency is discussed further in Section 3.6.

3.2 ANGULAR-MOMENTUM EQUATION

The energy transfer between the fluid and the machine occurs by tangential forces exerted on the fluid
as it interacts with the rotor blades. Although forces are also exerted between the fluid and the stators
(stationary blades), no energy transfer occurs since there is no displacement associated with the forces - thus,
stators can only redistribute energy among its components.

The angular form of Newton’s second law (the angular-momentum equation) governs the interaction
(see earlier courses for derivation):

Torque applied to fluid in CV = outflow of angular momentum - inflow of angular momentum

T0

∫out r  C dm˙  in∫ r  C

d m˙
The torque about the axis of rotation of the machine is then

(7)
T  ∫ rC w dm˙  ∫ rCw

d m˙
where r = radial distance
out from the
in axis
Cw = tangential component of absolute
velocity
Or using mean values

T m˙rCw out  m˙rCw in


3.3 EULER PUMP AND TURBINE EQUATION

We will use the following nomenclature in this and the subsequent sections:

 

ROTOR
U

   

C  

W STATORS

 

C = absolute velocity
W = relative velocity (as seen in the rotating frame of reference)
U = blade circumferential speed ( = ωr)

Subscripts:

a = axial component (of velocity) (subscript x also used)

r = radial component
w = "whirl" (circumferential or tangential) component (subscripts t and θ also used)

Angles:

α = absolute velocity
α = stator blade metal angles
N = relative velocity
β = rotor blade metal angles
β
N
The datum for all angles is the main flow direction: axial in axial-flow machines, radial in radial-flow
machines.

Sign conventions: The question of signs only arises with reference to velocity components and
angles in the tangential direction. Unfortunately, there is not much consistency in the use of signs in the
turbomachinery literature. When needed, we will use the following conventions:
(i) Tangential components of velocity are positive if they are in the same direction as the blade speed,
U.
(ii) The signs of angles are consistent with the sign convention for the tangential velocity components.
 Consider again the general turbomachinery rotor

The torque applied to the fluid as it passes through the rotor is given by (7):

(7)
T   rC w dm˙   rCw

d m˙
The torque is supplied at the shaft, transmitted2 through the 1disk and blades, and applied by the blades to the
fluid in the form of a tangential force. The corresponding shaft power is

W˙ shaft  T

and multiplying through by ω in (7)

W˙ shaft   Cw dm
 ˙ 
rC dm˙ w
r 2 1
(8)

UC2 
˙ 
w dm
dm
UC ˙
1
w

where U = rω is the blade speed.

But the SFEE also relates the shaft power, W˙ shaft , to the energy changes in the fluid. Equating
the shaft powers from Eqns. (3) and (8)

(9)
 h dm˙   h dm˙   UC
0 0 w dm˙   UCw

d m˙
2 1 2
1
 (10)
h02  h01 U 2 Cw2  U1Cw1
If we approximate the flow quantities by their mean values, then we can write
 For an incompressible-flow compression machine (from eqn. (5))

gH2  H1  H L   U 2 Cw2  U 2 Cw2

and letting ΔH = H2 - H1 (the total head rise seen across the machine) and ΔHE = H2 - H1 + HL = ΔH + HL
(the "Euler head") then
g HE  U 2 Cw2  U1 Cw1 (11)

Eqns. 9-11 are versions of the famous Euler Pump and Turbine Equation (or Euler Equation). The
Euler equation is the fundamental equation of turbomachinery design. It relates the specification (for example,
the head rise required) to the blade speed of the machine and the changes in flow velocity that it must produce
to achieve the required performance. As described later, these changes in flow velocity are directly related to
the rotational speed and geometry (eg. blade shapes, etc.) of the machine.

Note that the Euler equation involves the full energy transfer between the machine and the fluid,
including the energy that will be dissipated in overcoming friction. For a pump

H
H E
  pump

ΔH will be specified to the designer. But from eqn. (11), ΔHE is needed to determine the flow turning
(change in UCw) which will achieve the required ΔH. Thus, to design the machine we need to know its
efficiency. As a result, the design process becomes iterative.
3.4 COMPONENTS OF ENERGY TRANSFER

We now examine in more detail the process of energy transfer within the rotor. Recall that

absolute velocity = relative velocity + velocity of moving reference frame

C W U

The drawing shows a hypothetical velocity diagram at outlet (station 2) for the generalized rotor (a
similar diagram could be drawn for station 1)

From the Euler Equation

W˙ shaft (12)
m  gHE  h0  U2 Cw2  U 1C w1

˙
We then rewrite the velocity terms on the RHS in terms of the velocity vectors in the drawing

C2 C2 C2 C2 (a)

2 a2 w2 r2

and similarly for the relative velocity (the components are not labelled on the figure to
avoid clutter)
W22  Wa2 2  W w2
2
r
W2 2 (b)
 C
2
 Ca2  2  C w2 2
r2
2

Solve (a) and (b) for Ca22 + Cr22 and equate

U

C 22 2 C 2  W 2 w2U 2  2U C 2 C
w2 2 2
w2
Then
1 2
U 2 C w2 
2

C2 U 22  W 22 
Similarly for the velocity triangles at the inlet, station1,

1

U 1 C w1  C12  U12 
2
1
2

W
Substituting into (12)

W˙ shaft 1
 h  C 2  C12  U22  U 12  W12  W 2
m g
E 0 2 2    2   (13)

H˙
(1) (2) (3)


Note that (13) is another (and useful) version of the Euler Equation.

Now consider the physical interpretation of the three terms on the RHS of (13).

1 2
 2
Term (1), 2 C 2  C 1 is clearly the kinetic energy change of the fluid across the rotor. In a pump,

blower or compressor, the kinetic energy of the fluid normally increases across the rotor. Some of this kinetic
energy can be converted to static pressure rise in a subsequent diffuser or set of stators.

To see the physical meaning of the other two terms, apply the SFEE between the inlet and outlet of
the rotor again, assuming adiabatic flow and neglecting potential energy changes:


 P1 C2  ˙ P C2
m˙

 1
2
 u1  Wshaft  m˙ 

2

2
2

  u2  

Substitute for W˙ shaft from the Euler Eqn., (13), and solve for the static pressure rise through the rotor passage

1 1
P2  P
1
   
2
 U 222  U 12   W12  W
2
  u  u 
2
(14) 2
1

Equation (14) shows that there is some direct compression (or expansion) work done inside the rotor blade
passage and it is associated with the changes in U and W that the fluid experiences as it passes through the
rotor. Note that if there is friction present, u2 > u1, and this reduces the pressure rise that would be achieved by
a compression machine, as one would expect.

Term (2),
1 2
2
 2
U 2  U 1 is then energy transfer to the fluid due to the centrifugal compression (or


expansion) of the fluid as it passes through the rotor ("centrifugal energy" change). The rotation of the fluid
imposed by the rotor results in a radial pressure gradient to balance the centrifugal forces on the fluid particles.

For example, consider a centrifugal pump or compressor rotor for the limiting case where there is no
flow (say that a valve has been closed in the discharge duct). The fluid particles trapped inside the rotor travel
in circular paths. The force required to give the corresponding acceleration towards the axis of rotation is
supplied by the radial pressure gradient that is set up in the rotor.

(+)

F
2
(-)

1 

For this case, W1 = W2 = 0, and from (14) then

1
P2  P1  2
 U 22  U 2
1


Thus, a radial machine will produce a pressure rise even for no flow. The delivery pressure for this case is
sometimes known as the “shut-off head”.

When there is flow, the fluid particles that move through the radial pressure field will likewise be
compressed (or expanded) and the corresponding work per unit mass is accounted for by term (2) in Eqn. (13).

1 2
Term (3),
2

W1  W 22 represents the change in pressure energy due to the change in fluid velocity


relative the rotor. Consider the flow in a the rotor-blade passage of an axial compressor. Neglecting
friction (u2 = u1) and if the stream tube is at constant radius (so that U1 = U2) then from Eqn. (14)

1
P P   W2  W 2 (15)
2 1
2 
1 2

As shown in the sketch, a typical compressor rotor passage increases in


cross-sectional area as the relative flow is turned towards the axial
(which is necessary in order to increase the C w in the absolute frame).
W2

From continuity, W2 < W1 and from (15) there is a corresponding

pressure rise. The passage is thus a diffuser. The forces exerted on the
fluid by the blade surfaces cause the static pressure to rise between inlet
U
W1
and outlet, and since there is also displacement associated with these
forces (since the rotor is moving) work is being done on the fluid.
 Note that the pressure rise along the rotor blade passage can cause separation of the blade boundary
layers and therefore stalling of the airfoils. We therefore find it necessary to limit the change in W that we
permit in a given blade passage.

Summarizing:

(a) Term (1) in Eqn. (13) represents the change in kinetic energy (dynamic pressure) of the fluid
due to the work done on it in the rotor.
(b) Terms (2) and (3) represent the direct static pressure changes (compression or expansion
work) which occur inside the rotor.

In general, all three components of energy transfer will tend to be present in all rotors. However, for
axial rotors the centrifugal compression tends to be small (since U1 – U2 for every streamtube that passes
through the rotor), whereas it is large in radial rotors.
3.5 VELOCITY DIAGRAMS AND STAGE PERFORMANCE PARAMETERS

3.5.1 Simple Velocity Diagrams for Axial Stages

A turbomachinery stage generally consists of two blade rows, a rotor and a set of stators:
• A compressor stage normally has a rotor followed by a row of stators. As noted in 3.4, some
static pressure rise can occur inside the rotor. The stators can produce a further static
pressure rise by reducing the fluid velocity.
• A turbine stage normally has a row of stators ("inlet guide vanes" or "nozzles") followed by a
rotor. The nozzles impart swirl to the flow, accelerating it and thus causing a static pressure
drop. The rotor then extracts energy from the fluid by removing the swirl. This may be
accompanied by a further static pressure drop inside the rotor.

Consider a thin streamtube passing through an axial compressor stage (say near the mean radius):

We then draw a hypothetical set of velocity vectors as they might appear in the axial plane:
 Note that the inlet flow has been assumed to have some swirl (α1 … 0.0). Therefore, there must be
another stage or a set of inlet guide vanes ahead of the present stage. The stators have also been shaped to
give a stage outlet flow vector equal to the inlet vector (C 3 = C1). This is sometimes referred to as a “normal
stage”.

Even for an axial stage, as the flow passes through the stage, the streamtube may vary slightly in
radius. Thus, in general U1 … U2. Also, due to the density changes and changes in the cross-sectional area
of the annulus, the axial velocity at different locations may vary (Ca1 … Ca2). However, across a given axial
rotor blade, the radial shift in any given streamline tends to be quite small. For reasons discussed later, it is
also undesirable to have the axial velocity change significantly along the machine. The latter is the reason for
the tapering of the annulus which is seen in most multistage compressors and turbines.

For discussion purposes only, we may therefore make the following simplifying assumptions
for axial
stages:

(i) Assume the streamline radius is constant through a rotor: U1 = U2.

(ii) Assume constant axial velocity through a given stage: Ca1 = Ca2 = Ca3.

The resulting velocity diagrams are sometimes known as the “simple” velocity diagrams (or velocity
triangles). For actual design calculations, we would not make these simplifications: we would use the true,
general velocity diagrams. But in practice most axial stages come close to satisfying the simplifying
assumptions and therefore the conclusions which we will draw about the stage behaviour, based on the simple
velocity triangles, will be quite realistic.

One convenient feature of the simple velocity triangles is that we can combine the inlet and outlet
triangles because of the common blade speed vector U. We can therefore draw the velocity triangles for the
axial compressor stage as follows:
 3.5.2 Degree of Reaction

If the pressure is rising in the direction of the flow (ie. if there is “diffusion”), then there is a danger of
the boundary layers on the walls separating. When this happens on a turbomachinery blade, there is generally
a large reduction in the efficiency of the machine and an impairment of its ability to transfer energy to or from
the fluid. In the case of compressors, boundary layer separation can lead to the very serious phenomena of
stall and surge which will be discussed later.

Diffusion is present most obviously in compressors since they are specifically intended to raise the
pressure of the fluid. While overall the pressure drops through a turbine stage, diffusion may still be present
locally on the blade surfaces. Thus, the possibility of boundary layer separation is a concern in the design of
both compressors and turbines.

As evident from the velocity triangles, pressure rise can occur in both blade rows of a compressor
stage. Intuitively, it would seem beneficial to divide the diffusion fairly evenly between the blade rows.
Similarly, in a turbine stage both blade rows can benefit from the expansion. The choice of the split in
pressure rise or drop between the two blade rows is one of the considerations for the designer of a
turbomachinery stage.

We define the degree of reaction, Λ

rate of energy transfer by pressure change inside the rotor


 total rate of energy transfer
1
 2
 
U 22  U 12  W12  W 22
(16)
 h02  h01 

which can also be written

h2  h1
 (17)
 h02  h01

where h = static enthalpy, h 0 = total enthalpy. Using the Steady Flow Energy Equation or Euler Equation,
there are several alternative ways of expressing the denominator in (16) and (17).

If the flow is assumed incompressible and isentropic, and the stage inlet and outlet velocities are the
same (ie. if is a “normal stage”), (17) reduces to

 Protor (18)
 Pstage

Thus, (16) and (17) are also approximate measures of the fraction of the static pressure change which occurs
across the rotor.

A well-designed pump, fan or compressor will then have Λ > 0 in order to spread the diffusion
between the blade rows. A value of Λ . 0.5 has often been used. In an open machine, such as a Pelton wheel
turbine, P1 = P2 = Patm and Λ = 0. A machine with Λ = 0 is known as an impulse machine. Impulse wheels are
sometimes used for axial turbines, particularly steam turbines.
 The effect of the choice of Λ on the machine geometry can be seen by examining the velocity
diagrams for a few examples.

Axial-Flow Impulse Turbine (Λ = 0):

Consider the mean radius. Assume incompressible flow, constant annulus area and no radial shift in
the streamlines. Thus U1 = U2 = U and from continuity, Ca0 = Ca1 = Ca2 since m˙  Ca Aannulus . We
therefore have the conditions for simple velocity triangles. The turbine stage will look as follows:

The basis for the stage geometry is as follows:

Nozzles: We must accelerate the flow through the nozzles, since all expansion is to occur in
here (Λ = 0): ie. we want C1 > C0. This can be done by turning the flow since this
will reduce the area of the flow passage from A0 to A1noz (for the constant height,
A1noz = A0 cosα1 N). Bear in mind that Ca0 = Ca1 from continuity.

Rotor Blades: For Λ = 0, we need W1 = W2 (since U1 = U2). Thus we need A2rot = A1rot, which is
obtained with β1N = β 2 N. Therefore, the impulse turbine will have equal inlet and
outlet metal angles.
 What determines the value of α1N which is chosen? From the Euler Equation:


W˙  m˙ U 2 Cw2  U1Cw1   m˙UCw
Redraw the velocity triangles with the common blade speeds U superimposed. Note that ΔCw = Cw2 - Cw1 will
be negative, consistent with our sign convention that power in is
positive. The magnitude of ΔCw (for a given U) is clearly related to C =Ca1 a2

α1. Thus, the required W˙ plays a direct role in determining the

velocity triangles, and ultimately the metal angles. C2 Cw2 (+)
1 (+)
Note also that to sketch the blade shapes we assumed that
the fluid leaves a blade row at the metal angle: U W2

1  1, 2  Cw1 (+) Cw

C1
2
This is not strictly true, as will be discussed later, but is often a
reasonable first approximation. It is sometimes known as the "Euler
W1
Approximation".

Axial-Flow Turbine with Λ>0 (Reaction Turbine):

Again assume constant streamline radius, constant annulus area and incompressible flow. Then U1 =
U2 and Ca1 = Ca2 as before. The nozzles will again impart swirl to obtain some expansion. To get expansion in
the rotor, need W2 > W1 and thus *β2N* > *β1N*. An example of the geometry of a reaction turbine is then as
follows:
Axial-Flow Compressor with Λ>0:

Again, assume U1 = U2 and Ca1 = Ca2. To get static pressure rise across the rotor we need W2 < W1.
Examining the compressor used as an example in Section 3.5.1, it is evident that this compressor meets this
requirement:
 3.5.3 de Haller Number

The importance of diffusion in compressor blade rows was discussed in Section 3.5.2. By selecting a
degree of reaction close to 50%, the diffusion is shared roughly equally between the rotor and the stators.
However, this does not address the question of whether the blade rows will be able to sustain the level
of diffusion which is being asked of them. We will later examine diffusion limits which are used in the
detailed design of the blade rows. However, it is useful to have a simple approximate criterion for diffusion
which can be applied at the point in the design where we are taking basic decisions about the velocity
triangles.

An axial compressor blade row in effect forms a rectangular diffusing duct. Based on various
compressor designs of the time, de Haller in the mid 1950’s suggested that the maximum static pressure rise
which could be achieved in axial compressor blade passages is given by
P
C p,max 
(a)
 2
1 V 0.44
2 

where ΔP = static pressure rise between inlet and outlet of the blade row
V = velocity at the inlet to the passage (relative velocity for rotors, absolute for stators).

Taking a rotor blade passage and assuming no change in radius of the streamlines (so that there is no
centrifugal compression) and neglecting friction, from Section 3.4 the static pressure rise is

1 1
P2  P 1 W12   W22 .
2 2

Substituting into (a) and simplifying,

 W2  
 W1  min 0.75.

The ratio W2/W1 (or Cout/Cin for a row of stators) is known as the de Haller number.

The de Haller limit should be used as a rough guide only. It does not take into account details of the
blade passage design which can improve the diffusion capability of the passage. Successful modern
compressor designs have used values of the de Haller number as low as 0.65. The de Haller number should be
used mainly to alert the designer to the fact that the level of diffusion in a particular compressor blade row
may present a design challenge.
 3.5.4 Work Coefficient

From the Euler Equation

h0  U 2 Cw2  U1Cw1

 UCw 

and for an axial machine with simple velocity triangles (so that U1≈ U2 = U)

h0  UC w .

From the velocity triangles, if we vary U, adjusting Ca to maintain geometrically similar triangles, then
C w  U
and h0  U
2
.
Thus, the power transfer varies as U2. The head or enthalpy change "per unit U2" is a useful measure of the
stage loading and is known as the work coefficient, †, where

h0 UCw  gHE

   
2 2
U U U2

For “high” †, we are taking full advantage of the blade speed and we have “high stage loading”: we will
specify what constitutes “high” † for different types of machines in Section 3.5.6.

For a centrifugal machine, tip speed, U2, would be used in †.

For an axial machine with simple velocity triangles (so that U1≈ U2 = U)

UCwU Cw
U2 

Normally, † is taken as positive. For our sign convention, ∆h0 and ∆Cw are negative for turbines.
Therefore, we use absolute values in †
 3.5.5 Flow Coefficient

Consider two compressor rotors designed for the same service (same Q, ΔP0 and N):

The same mean radii have been used so that the rotors have the same blade speeds U. From the Euler
equation, h0  UCw , and to achieve the same h0 , and thus the same pressure rise, they must
therefore have the same change in swirl velocity, ΔCw. As a result, the rotors have the same work
coefficient
(ψ = ΔCw/U) and thus the same loading. However, rotor B has twice the axial velocity of rotor A: this is
achieved by reducing the cross-sectional area of the machine. This change obviously has a significant effect
on the rotor blade geometry. It also has aerodynamic consequences:

(i) For rotor B, both the absolute and relative velocities have been increased. Since losses generally
vary as 0.5ρV2 (where V = W for the rotor), rotor B will, all other things being equal, have poorer
efficiency than rotor A.
(ii) All other things are not equal. Note that the increase in Ca in rotor B has had the effect of increasing
the de Haller number (W2/W1). Thus, the diffusion has been reduced in rotor B, which is
aerodynamically favourable.

We can thus identify an additional important parameter which must be chosen by the designer, the flow
coefficient, φ:
C
  Ua

For a centrifugal compressor, we would use Cr2/U2, where Cr2 is the radial component of velocity at the rotor
outlet.

Note that for the compressors shown, the change in flow coefficient did not in fact change the degree
of reaction. As you will show in Problem Set 3, the symmetry of the velocity triangles for both machines
implies that they both have 50% reaction.
 3.5.6 Choice of Stage Performance Parameters for Design

We have identified four useful performance parameters: the degree of reaction, the de Haller number,
the work coefficient and the flow coefficient. Experience shows that to design a stage with good efficiency, φ,
ψ and Λ, and for fans and compressors, the de Haller number, should be kept within certain ranges.

Design Fans, Pumps, Compressors Axial Turbines

Parameter
Axial Centrifugal

φ 0.2 6 0.7 . 1 (at outlet) 0.4 6 1.2

ψ 0.3 6 0.6 0.6 6 1.0 (see later) 0.3 6 3.0
<0.5 - “Lightly Loaded”
>1.5 - “Highly Loaded”
Λ 0.3 6 0.7 (Not much used) 06 1
de Haller >0.65 (well-designed machines See Section 6.4.3 N/A
with clean inlet flow)
>0.80 (simple design, poor
inlet flow uniformity)

For compressible-flow axial turbines, Smith ( S.F. Smith, "A Simple Correlation of Turbine
Efficiency," J. Royal Aero. Soc., Vol. 49, July 1965, pp. 467-470.) developed a very useful figure (the “Smith
chart”) which summarizes the influence of φ and ψ on the efficiency of the stage:

Variation of Stage Efficiency with φ and ψ (for Zero Clearance).

 The "Smith Chart" or "Smith Diagram" presents the results for a large number of turbine tests (for
both model and full-scale machines) conducted at Rolls-Royce from 1945 to 1965. Over that period, the flow
over the tip of the rotor blades ("tip leakage") was considerably reduced. The tip-leakage flow is an important
source of losses and as a result there was significant improvement in efficiency. To isolate the influence of the
stage loading and shape of the velocity triangles, the efficiencies were corrected back to their zero-clearance
equivalents. Thus, efficiencies for actual machines can be expected to be lower than those shown by a couple
of percentage points. Note that the degree of reaction is not mentioned on the Smith chart. The turbines used
to generate the chart had a range of degrees of reaction. However, the performance of turbines is not strongly
dependent on the degree of reaction, provided reasonable values are used.

The Smith chart is well known and is widely used by axial turbine designers during the preliminary
stages of design. The usefulness of the Smith chart makes it surprising that comparable charts are not more
widely used by axial and centrifugal compressor designers. Part of the reason lies in the important role played
by diffusion (expressed through both the degree of reaction and the de Haller number) in compressor
performance. Thus a single “Smith chart” for compressors is not feasible. However, it is possible to generate
a small number of charts, each for a different value of degree of reaction say, and then use these in design. In
the late 1980's Casey (M.V. Casey, “A Mean Line Prediction Method for Estimating the Performance
Characteristics of an Axial Compressor Stage,” Proceedings, I.Mech.Eng., C264/87, 1987, pp. 273-285.)
calculated compressor stage performance for a wide range of conditions. In a recent textbook, Lewis (R.I.
Lewis, “Turbomachinery Performance Analysis”, Arnold, London, 1996) took this data to generate “Smith
charts” for axial compressors for three values of degree of reaction: 50, 70 and 90%. Note the rapid
deterioration in efficiency when the de Haller number is less than about 0.7.

“Smith” Charts for Axial Compressors: (a) Λ = 0.5, (b) Λ = 0.7, (c) Λ = 0.9.

The use of the guidelines presented in this section will be illustrated in the next chapter.
3.6 EFFICIENCY OF TURBOMACHINES

3.6.1 Incompressible-Flow Machines

The definitions of efficiency used for incompressible-flow machines have been discussed briefly in
earlier sections. The definitions are repeated here for completeness.

Fundamentally, the efficiency of a turbomachine is defined in terms of a comparison with a related

“ideal” machine in which there are no losses. However, there are small conceptual differences between the
definitions of efficiency used for incompressible- and compressible-flow machines. These will therefore be
clarified now.

(a) Pumps, Fans and Blowers

From the steady flow energy equation,

W˙ shaft  m˙gH E 

where ΔHE m˙=

Euler head hhead
0 equivalent of the shaft power input to the machine
= head rise that would be achieved in the ideal (no losses) machine with the
= same shaft power input as the actual machine.

The fluid power is defined as the useful, mechanical power that actually appears in the fluid across the
machine

W˙ fluid  m˙gH

where ΔH = total head rise that is actually observed across the

machine. The Euler head and the actual total head are related by

H  HE  HL

where HL is the head loss due to friction inside the machine. Neglecting elevation changes,
we can also write

P0,actual  gH

P0,ideal  gHE

We then define the efficiency for a pump, fan or blower as

 pump  Fluid power  QgH  H  P0,actual
Shaft power W˙
H P0,ideal
shaft E

To help visualize the significance of this definition, and for comparison with the definition of efficiency used
for compressors, we represent the processes on the Δh0 versus s diagram.
 The specification calls for the machine to raise the P02
fluid head by ΔH, or the total pressure by
P0,actual  P02  P01  gH . With the same h0
P02
input
shaftper unit mass flow (Δh 0), the ideal machine would raise
power
the pressure by ΔP0,ideal = PN02 - P01. Thus, the efficiency
for
pumps, fans and blowers is defined by comparing the head or
total pressure rises for the actual and an ideal machine that
have the same shaft power input. As described below, the h0 ACTUAL P01
definition of efficiency for compressors is slightly different. IDEAL

(b) Turbines

For turbines, the head drop, ΔH, or pressure drop ΔP0,actual that is available is normally specified.
However, some of the fluid power released by the fluid is used in overcoming friction inside the machine and
is therefore not available to be extracted as shaft power output, W˙ shaft  m˙gH E . That is,

H  HE  HL

The turbine efficiency is then defined as

turbine Shaft power QgH E HE


Fluid power  QgH H 
P0,actual
P0,ideal
The physical interpretation can again be seen in terms of the h0
h0 versus s diagram. The actual pressure drop is ΔP 0,actual = P01 - P02 P01

and the shaft power extracted is W˙ shaft  m˙gH E  m˙h0 .

ideal machine, a smaller pressure drop, ΔP0,ideal = P01 - PN02, would be
In an
needed to produce the same shaft power output. Thus, the turbine ACTUAL

efficiency is defined in terms of two machines that have same shaft h0
IDEAL
power output. The comparison is between the head or total pressure P02
drops required to obtain that shaft power output in the ideal and
actual machines. The similarity with the definition used for pumps, P02
fans and blowers is evident.

s
 3.6.2 Compressible-Flow Machines

The efficiency of compressible flow machines is defined slightly differently. The comparison is again
between ideal and actual machines. However, instead of the shaft power input or output, the common basis is
the pressure rise or drop across the machines.

(a) Compressors

The h0-s diagram is again used to compare

the processes used to define the efficiency. For h0
P02
compressible-flow machines, the pressure rise or
drop across the machine is generally expressed in
terms of the total pressure ratio. The h0-s diagram
shows the ideal and actual compression processes h 0,actual
needed to obtain the same pressure ratio, P02/P01. For P01
ACTUAL
the ideal machine, the shaft power required is h0,ideal
IDEAL

W˙ideal  m˙h0, ideal

while for the actual machine

s

W˙ actual  m˙h0, actual .

The compressor efficiency is then defined as the ratio of the shaft powers required to produce the same
pressure ratio in the ideal and actual machines:
m˙h0, ideal h0,
c  ideal
W˙actual  m˙h0, actual h 0, actual
W˙ideal
If we assume that the working fluid is a perfect gas, then h 0 = CpT0, and it is common to present the
processes on a T0-s diagram, rather than the h 0-s diagram. The efficiency can then be written
C pT 0, ideal
  T  T
0 2 01 T0
C p T 0, actual  T02  T 01
c
P02

For any isentropic process involving a perfect gas,

T02

P T02

 const. ACTUAL
P01

IDEAL

where γ = Cp/Cv, the specific heat ratio. Then using the perfect gas T01
law, P = ρRT, we can write

 1 s
T02  P02  
 

T01  P01 
Then
C T  T
h0, actual  p 02 01
 c
  1 
C p T01 
  P02  
c    1
  P 01  


and this expression allows the shaft power required to drive the actual machine, W˙ actual  m˙h0, actual , to
be related to the specified pressure ratio.

(b) Turbines

The efficiency of compressible-flow turbines is similarly defined by comparing the shaft power
produced by the expansion through the same pressure ratio for an ideal and the actual machine. Following the
same procedure as for the compressor, we obtain

W˙ actual h 0, actual T02  T 01

t   T0
 W˙ideal
h 0,ideal T02  T01 P01

and T01
ACTUAL

  1  IDEAL
 P02
h0, actual  Cp T01   P 
t 
02  T02
 1
   P 01   T02


Note the expression for Δh0, actual will be negative, consistent with
our sign convention that power into a machine is positive. s

3.6.3 Polytropic Efficiency

Consider a multi-stage axial compressor consisting of a number of stages with equal stage pressure
ratios. If the stages are designed using the same technology, it is reasonable that they will each have the same
stage isentropic efficiency. It is then possible to calculate the overall pressure ratio and isentropic efficiency
for the machine as a whole.

Let PRs = stage total-pressure ratio

ηs = stage isentropic efficiency

It can then be shown that the

actual temperature at the outlet of the
Nth stage is   1 N

T T 1
 PR s  
 1
0 N 1 01
  s 

The overall pressure ratio for the N stages is

 PRc  s
N

PR

and the isentropic temperature rise for the whole compressor is then

  1 
T0N 1  T01  T01 PRc 
 1
 
  1 
 T01  PR sN

 
 1


and thus the overall isentropic efficiency is


N  1 
T0 N 1  T 01 PRs 
1
c 
 T T  1 N
0 N 1 01  
 PR s 
 1  1
1  s 
 

For example, if PRs = 1.2 and ηs = 0.9, the resulting variation of the overall pressure ratio and overall
isentropic efficiency with the number
of stages is shown in the figure. As
EFFECT OF PRESSURE RATIO ON OVERALL ISENTROPIC EFFICIENCY
seen, the overall efficiency decreases as
the pressure ratio increases.
0.9 1
2
Overall Compressor Isentropic Efficiency

3
When cycles for gas turbine 4 5
6 Number of Stages
7
engines are being investigated, it is 0.88 8
9
10
normal to examine the effect of varying 11
12
pressure ratio. It is evident that
0.86
assuming a constant value of the
overall compressor isentropic
efficiency is not valid for such 0.84
investigations. To account for the
effect of the pressure ratio on the Stage PR = 1.2
0.82 Stage isen = 0.9
isentropic efficiency, the concept of the
small-stage or polytropic efficiency has
been introduced. 0.8 2 3 4 5 6 7 8 9 10
1
Overall Compressor Pressure Ratio

From the Second Law of Thermodynmaics, for a general infinitesimal process

dh0  dP0  T0 ds
0

Then for an isentropic process (ds = 0)

 dP0
dh0
 0

Define the polytropic efficiency, ηp, as the isentropic efficiency for the infinitesimal process

dh0   pdh0

dP
 0  p 0
dh 0

R
Then assuming a perfect gas, h0 = CpT0 and P0 = ρ0RT0. Also Cp - Cv = R, or C p  , and the fluid
 1
properties are assumed constant through the process. Then

dP
0
  d CCpTT0 
p 0
0 C p T0
p

dP0 dT0
 p
T

0 RT0  0
 1

or dT0   1 dP0
T0   p
P0
Integrating this between the start and end of a finite process

 T 02  
 1
 P02 
ln   p ln 
 T01   P01 

 1
T02  
 p
or
P02   
 P01 
T01

For a compression process, the isentropic efficiency is defined as

c  h
0
h0
 1   1 
 
where
0
h  C T  PR 
 1 and h0  Cp  PR  pc
 1 where η pcis the polytropic efficiency for
01 
  
T
p 01 


the compressor. Then

  1


c PR 1
  1
 pc
PR
1

For a turbine,

h0
t
 h0

and, with inlet at 3 and outlet at 4, it can then be shown that

 pt  1
 P04  
1 
t  P03 
 1
 
 P04 
1 
 P03 

The following figure shows the resulting variation of isentropic efficiency with pressure ratio for an assumed
polytropic efficiency of 0.9 and γ = 1.4, for both a compressor and a turbine. Also shown are the earlier
results for the multistage compressor with stage pressure ratio of 1.2.

VARIATION OF ISENTROPIC EFFICIENCY WITH PRESSURE RATIO

Polytropic Efficiency, p = 0.9,  = 1.4

0.94

Turbine
0.92
Isentropic Efficiency

0.9 1
2

3 5
4 6
7
0.88 8
9
10
11
Compressor 12

0.86

1 2 3 4 5 6 7 8 9 10
Pressure Ratio

The concept of polytropic efficiency should be used with caution. It is only valid if the machine can
be considered to employ comparable technology and produce comparable performance as the pressure ratio is
varied. For this reason, it should be applied only to explore the influence of pressure ratio on performance for
multistage machines. It is assumed that the pressure ratio is varied by adding or removing comparable stages.
Polytropic efficiency should not be used to predict how the efficiency of a single stage will vary as its design
pressure ratio is changed. As will be shown later, stage performance is closely related to its tip speed. For
example, to increase the design pressure ratio of a compressor stage, the tip speed must normally be increased.
This in turn results in higher flow velocities generally. As these velocities reach and exceed the speed of
sound, shock waves will appear, providing a source of additional losses that is not present at lower speeds.
Thus, as the stage pressure ratio is changed, the technology cannot be considered to remain unchanged.
2. CONTROL VOLUME ANALYSIS FOR AXIAL-COMPRESSOR BLADE SECTION

1. Force Components

Consider the control volume for the flow through one blade passage:

P1
y
A B

x C1
1 Ca1

Cw1

m Y
Cm Ca m
F
Cwm X L D

D 2 C
C2
P2
Ca2

Cw2

Take unit depth in the z direction. Also, make the following simplifying assumptions
(i) Incompressible flow
(ii) Constant axial velocity through the passage: Ca1 = Ca2 = Ca.

The blade exerts a force F on the flow thought the passage. This is divided into axial and tangential
components X and Y. By definition, the lift generated by a turbomachinery blade L is the component of the blade
force normal to the vector mean flow direction through the blade row. The drag D is the component of the blade
force parallel to the vector mean flow direction.

Then apply the linear momentum equation to the control volume. In the x direction:

Fx  m˙Vx2  Vx1 

X  P1 s  1  P2 s  1  m˙Ca2  Ca1 

Note that the pressure forces along the left and right faces of the control volume exactly balance each
other in both the x and y directions. Then since we have assumed Ca1 = Ca2, the x-wise momentum equation
reduces to
X  P2  P1 s (1)

For the y direction:

Fy  m˙Vy2  Vy1 

and since the pressure forces on the control volume cancel each other in the y-direction, the only force in the
y-
direction is that due to blade, Y

Y  Ca s  1 Cw1    Cw2

(2a)

Since

 Ca sCw1  Cw2 
C tan C
tan 1  C w1 , 2  Cw2
a  a
we can also write

Y  Ca2 s tan 1  tan  

2 (2b)

From the definition total pressure for incompressible flow, the total pressure loss through the passage is
given by

P0  P01  P02   P  P2  1 C 2 C 2


1
2 1 2

From the velocity triangles,


C12  C 22  Cw1
2 2
 Ca1 2
 C w2 2

 Ca2 

C w1  Cw2 Cw1  Cw2 
since Ca1 = Ca2. Then

P0  P1  P2   2 Cw1 

1 w2  w1
 C w2 

C C
substituting for Cw1  Cw2 from (2a)

1 
 Y 
X 
P0     C w1  w2
s 2  Ca s
C
1
and using the vector mean flow direction through the passage, tan m 
2
tan  1  tan  2  , we can write


1
P 0 
s
 X  Y tan  m (3)

From the force vector triangles, the drag D can be expressed in terms of X and Y as follows

D  Y sinm  X cosm
(4)
 cosm  X  Y tanm 
and substituting from (3)

D  P0 s cos (5)

m
Then from the definition of the drag coefficient

D
CD
 1 Cm2c  1
2
P0 s cos
 1
m
Cm2 c
2

and finally, since σ = c/s is the solidity

1
P0  CD  1 2
m  (6a)
  cos m
C 2

or alternatively, using Cm = Ca/cosαm,

1
P0  CD  1 2
 (6b)
 a
 cos 3  m
C 2

As will be seen later, some axial fan and compressor prediction procedures use the airfoil drag coefficient to
express the loss performance for the blade row. Equation (6a) or (6b) can then be used to express this as a total
pressure loss.

Returning to the lift force, from the force triangles the lift L can be expressed as

L  X sin  m  Y cos (7)

m
Solving (4) for X
D
X  Y tan m 
cos m

and substituting into (7)


L  Y tan   D 
 m  sin  m  Y cos m
 cos m 
Y  D tan
 m
cos m 

Then substituting for Y from (2b)

C 2 s
L a
cos m tan  1  tan  2   D tan  m (8)

By definition
L L
CL 
1
 Cm2c 1 C 2 c 1 

2 2 a 2
 cos  
m
then
C 2a s
tan 1  tan  2 D
C L  cos m  tan
 1  1  1 2 
m

C 2 c  Cmc
2 a
2
 cos2  m 

or

C L  2  cos m tan 1  tan  2   C

s D
tan m (9a)
 

Since the drag force is normally much
c  smaller than the lift, the drag term is often omitted from
(9a),

C L  2  cos m tan 1  tan  2 

s (9b)


c

4.2.2 Circulation

Any lifting surface has circulation. By definition, the circulation Γ is

(10)
   VS
dS contour enclosing the lifting surface. VS is the tangential
where the integral is evaluated along any closed
component of the flow velocity along the enclosing curve and S is arc length. For the axial-compressor airfoil,
the curve A-B-C-D-A shown on the control volume in the last section is a convenient curve for use in (10):

B C D A
 V dS  V dS  V dS  V dS
A
S
B
S
C
S
D
S

 identical lengths and velocity distributions,

Since B-C and D-A are periodic surfaces with

C A
 V S dS  
dS 
D
V
S

B
and their contributions to Γ cancel. Along A-B and C-D, VS is simply Cy (= Cw) along the respective segments. The
direction of the integration changes so that the integrals will have opposite signs (since Cw1 and Cw2 have the same
sign for the control volume shown). Thus, we can write

  Cy1s  Cy2 s

We are assuming constant axial velocity, and since tan   Cy Cx (and Cx = Ca), we can write

  Ca tan 1  tan  2 s (11)

 From Eqn. (8), neglecting the drag term and substituting from (11) we can also write

L  Cm 

which is the expression given by the Kutta-Joukowski Theorem for an isolated airfoil. Note that in the case of the
blade row, the “undisturbed” velocity seen by the airfoil is in fact the vector mean velocity through the passage.
4.3 IDEALIZED STAGE GEOMETRY AND AERODYNAMIC PERFORMANCE

4.3.1 Meanline Analysis

rt

rm

1 2 3
rh

For preliminary design, we typically consider just the flow at the mean radius and treat the flow
through stage as one-dimensional. The mean radius is normally defined as the radius that divides the flow
area in half:


rm2  rh2  rt2
2

h t
 is known as meanline analysis.
where r = hub radius and r = tip radius. This approach

The first step is to define the meanline velocity triangles, starting from the specification ( m˙  P0
or Q  H ) and using the guidelines for φ, ψ etc. from Chapter 3. To illustrate the procedure, we will use a
semi-quantitative example. Assuming an incompressible flow machine, we will define the velocity triangles
for a stage, consisting of a rotor and a row of stators, that delivers a head rise ΔH at a volume flow rate Q.
From the general guidelines, we choose the following values for the mean radius:
C
Flow coefficient:  U a 
0.5

Work coefficient: gHE

   0.4
U2

Degree of reaction:   0.5

Note that if we were using the Lewis charts from Section 3.5.6, we would probably choose a slightly higher
value of φ for this value of ψ:

To proceed, we need the value of the Euler head rise, ΔHE = ΔH/η. Therefore, we need to guess a
value for the stage efficiency η. This can be done from experience, or from the specific speed plots in Chapter
2, or from the approximate correlations shown in Section 3.5.6. Later, we will see how to calculate the
efficiency of the stage we have designed. If this efficiency is different from the one we have guessed here, we
will have designed the stage with an incorrect value of the ΔH E and it will not match the required performance
ΔH. If this turns out to be the case, we will have to return to the beginning and revise the design. Thus, the
design of a turbomachine inherently tends to be iterative: to design the machine we need its efficiency, but we
do not know its efficiency until we have designed it.

Having estimated ΔHE we can then calculate the absolute blade speed (at the mean radius) from our
chosen value of the work coefficient ψ:
 gHE
U
 

With U determined, the axial velocity at mean radius follows from the chosen value of the flow coefficient φ:

Ca  U

The chosen value of φ also determines the relative magnitudes of Ca and U as they will appear in the velocity
diagram: in this case, Ca = 0.5U. Finally, having established Ca, the required annulus area for the stage
follows from one-dimensional continuity:

m˙ Q
A 
Ca Ca

We will assume “simple” velocity diagrams, as defined in Section 3.5.1. That is, we assume that the
annulus is shaped such that Ca and U remain constant through the stage: U1 = U2 = U, Ca1 = Ca2 = Ca3 = Ca.
Then from the Euler equation

gHE  U 2 Cw2  U1Cw1  UCw

Knowing U, we now know ΔCw. Note also that for the simple velocity diagrams, ψ can be written

Cw
 
U

and we therefore also know the relative magnitudes of U and ΔCw in the velocity triangles: ΔCw =
0.4U.
occurring in the rotor and the stators. Thus the de Haller C a1  a2
numbers for the rotor and stators must be the same: C
Finally, we make use of the degree of reaction Λ to completely define the velocity triangles. Since we
have chosen 50% reaction,Wequal C amounts diffusion are
2  3 2
 1 
W1 C2 C  w1

On Problem Set #3, you will show that for simple C1

velocity diagrams, this is achieved by making the
velocity triangles for the inlet and outlet of the rotor
symmetrical, and by designing the stators so that C3 =
C1. That is, we design the stators so that the flow at the U C Cw
w2
stage outlet is identical to the flow that entered the stage. C2 
The rotor velocity triangles will then look as shown.
W1
With the velocity triangles established, we can
determine the de Haller numbers:
W2

W2 C  2 
W1  C 2 
3
 1 
0.678
This is approaching the limit of about 0.65 that was
recommended in Section 3.5.6 and we will therefore have to monitor our design for the possibility of stall. As
noted in Section 3.5.5, we could reduce the diffusion levels by increasing the flow coefficient φ.

Note that to achieve 50% reaction in this stage, the

inlet flow must have a swirl angle α1. Thus, there must either Ca1  Ca2
be a stage ahead of the present one, or a set of inlet guide C1
vanes, that leave the required amount of swirl in the flow. The 2
flow from this stage will also leave with swirl α3 = α1, so that 
C3 = C1.
Cw 2  Cw

Suppose instead that the inlet swirl was specified. For C2

example, if this is the first stage in the machine then we will
normally have no swirl in the flow, α1 = 0. Using the same
U
values of φ and ψ, the velocity triangles will then look as
W1 1  0
shown. We can then show that the resulting degree of reaction
is Λ = 0.8. This means that the diffusion is much higher in the W2
Cw 1 
rotor than in the stators and this might at first be a matter for 0
concern. However, consider the values of the de Haller
numbers (we will assume that the flow leaves the stage with
no swirl, C3 = C1): 2 
1 
W2
Rotor:  0.699
W1

C3
Stators:  0.781
C2

As expected, the value is lower for the rotor than for the stators. However, the diffusion is actually less than
for the 50% reaction machine. As a result, this stage may be just as feasible as the earlier stage, despite the
high value of degree of reaction.

Having determined the velocity triangles, the next step is to define the blade geometries that will
produce the required velocities.
4.3.2 Blade Geometries Based on Euler Approximation

For the idealized analysis, we define the blade geometry using the assumption that the fluid leaves the
blade row parallel to the metal angle at the trailing edge of the blades: this is known as the Euler
Approximation. In a later section, we will develop the procedures for estimating the actual outlet flow angle,
which will turn out to be slightly different. To bring the flow smoothly into the blade passage, we will also
make the leading edge metal angle parallel to the inlet flow angle. We can then define the shapes of the blades
for the 50% reaction stage as follows:

As indicated on the drawing:

(i) 1  1 to bring the flow smoothly onto the leading edge of the rotor blades.
(ii)In the relative frame of reference, the flow must leave the rotor blade passage at  2 to produce the required

turning. Based on the Euler approximation, the flow will leave the trailing edge at the metal angle and
we therefore use 2  2 .
(iii) The stators see the flow in the absolute frame. To bring the flow smoothly onto the leading edge
of the stator blades we therefore make  2   2 .
(iv)Again, the flow is assumed to leave the stators at the metal angle, and we use  3   3   1 .
Note that with the assumptions made, the rotor and stator blade geometries are identical for the 50% reaction
stage.
4.3.3 Off-Design Performance of the Stage

The geometry of the idealized stage was defined to give the required performance at the design point:
that is, at the design flow rate and rotational speed. However, any turbomachine will often be operated away
from its design point. The idealized analysis can also be used to give reasonable predictions of how the stage
will perform for off-design operating points.

(a) Effect of Varying Flow Rate

Consider first the effect of a reduction in flow rate at fixed blade speed U (i.e. at constant RPM). The
resulting velocity triangles will look as follows:

The new velocity triangles were arrived at as follows:

(i) Based on the Euler Approximation, the flow will still leave the blade rows at the metal angle.
Therefore, α1, β2 (and α3) are unchanged. Recall that there must be a set of stators or inlet guide vanes
ahead of the rotor to account for the inlet swirl.
(ii) From continuity, Ca is reduced and thus so is C1. In a quantitative calculation, the new value of
Ca would just be obtained from Ca  Q A , where Q is the new volume flow rate, and A is the
annulus area as established at the design point.
(iii) The magnitude of W2 is also reduced, by continuity, but the direction is unchanged.

From the velocity triangles, Cw1 has decreased while Cw2 has increased. As a result, the change in
swirl velocity ΔCw has increased. From the Euler equation

gHE  UCw

and the head rise produced by the machine H  HE will be increased. Equivalently, for a
compressible flow machine, h0 , and the corresponding pressure ratio, P02 P01 , will be increased. Note
that this is consistent with the increase in incidence (“angle of attack”) at the leading edge of the rotor blade.
As a result
of this, the blade should develop greater lift, do more work on the fluid, and thus increase the head rise. On
the other hand, increasing the incidence will eventually lead to stalling of the blade. Thus, reducing the flow
rate through a compressor stage will move it towards stall. Note that the incidence was also increased for the
stators, bringing them closer to stall as well.

Clearly, we can use the velocity triangles and the Euler equation to predict the quantitative stage
characteristic for the idealized stage. It is convenient to express the characteristic in terms of the work and
flow coefficients. The flow turning is

Cw  Cw2  Cw1

and from the velocity triangles (noting that Ww2 is negative for the conventional compressor
velocity triangles)

Cw1  Ca tan 1 Cw2  U  Ww2  U  Ca tan 2

Then
Cw  U  Ca tan 2  tan 1 

and dividing by U Cw Ca

U
1
U
tan  2  tan  1
or
  1  m

Thus, the ψ versus φ curve (effectively, the head rise versus flow rate characteristic) is a straight line with
slope
m  tan 2  tan 1

For the present case, the symmetry of the velocity triangles implies that 2   1 and the slope is then
m   2 tan 1 . For α1 > 0, as is the case here, this gives a negative slope and an inverse relationship
between head rise and flow rate, as inferred above.
1.0
Alternatively, since the characteristic
passes through the design point (say, φD and ψD),
we can write

D 1
m
D
and the slope of the characteristic is seen to be 0.5 D3
determined by the choice of design point (note
D2
also that in all cases ψ = 1 at φ = 0 for the ideal
characteristic). Interestingly, the characteristic
D1
will be steeper for a more lightly-loaded stage
(lower design work coefficient ψD) as illustrated in
the plot. INCREASING
D DESIGN-POINT
LOADING

0.0 0.5  1.0

(b) Effect of Varying Blade Speed

It is also worth looking briefly at the effect of varying the blade speed at constant flow rate. Using the
Euler Approximation again, it can be shown that the change in the velocity triangles will look as follows:

From the triangles,   Ca U   D since U has decreased. For the work coefficient,   Cw
U , ΔCw has clearly decreased, but so has U. However, ΔCw has decreased more rapidly than U; as can be
seen, a small further decrease in U would reduce ΔCw to zero. We therefore conclude that   Cw U   D
and φ and ψ are again seen to vary inversely. In summary, any deviation from the design point will cause the a
given compressor to move along the same ψ versus φ characteristic.

It is also worth noting that the reduction in rotational speed has had a very strong effect on the
absolute work transfer:

gHE  UCw

Since ΔCw decreases directly with U (and in fact faster than U) the head rise varies
approximately as

gHE  kU 2

and the head rise delivered by the stage, at a fixed value of flow rate, will change strongly with the
rotational speed: for example, reducing the speed by a factor of 2 will reduce the head rise by about a factor
of 4. Thus, high rotational speed is essential to obtain high pressure rise from a compressor stage. This will
be illustrated further in later sections.
 As seen, the Euler Approximation results in an idealized ψ versus φ characteristic for the stage that is
a straight line with a negative slope.

We have already noted that some changes in operating point will result in positive values of the
incidence at the leading edge of the airfoils. If this incidence becomes too large, we would expect the airfoils
to stall. Also, we would expect the efficiency of the stage to be best when the rotor and stator blades are
operating at the design point. We can therefore project what the actual stage characteristic is likely to be
based on the idealized characteristic:

MAXIMUM

STALL

D
USING EULER
APPROXIMATION

LIKELY
ACTUAL

D 
The characteristic shown applies for all rotational speeds. As noted, there is a strong effect of
rotational speed on the absolute performance (say ΔH for a given Q). To emphasize this, the characteristics
are often plotted in absolute terms as variations of ΔH (or ΔP0) versus Q (or m˙ ) for constant values of
rotational speed N. The corresponding curves are easily calculated from the non-dimensional characteristic.
The resulting map will look as follows.

max
H
CONSTANT

N CONSTANT

Q
 On each of the constant speed lines, there will be a point that corresponds to the design point values
of φ and ψ on the non-dimensional characteristic. At each of those points, the velocity triangles will be
similar, as indicated in the drawing. In each case, the relative velocity vector at the rotor inlet is lined up with
the metal angle and the flow comes smoothly onto the leading edge. As shown, we would therefore expect
that the machine will operate at its maximum efficiency at each of those points, apart perhaps for some small
effect of differing Reynolds numbers. Also, as we will see later, frictional losses vary as V2 and thus the
higher flow velocities with increasing rotational speed will result in higher frictional losses. This effect will
be partly offset by the fact that the Reynolds number is also increasing.

Later in the chapter, we will examine to what degree actual machines match the performance
characteristics we have inferred from the velocity triangles in this section.
4.3.4 Spanwise Blade Geometry

Finally, we use the idealized stage analysis to give an example of how the blade shape will vary
across the span. For this example, we will take the case with no inlet swirl from Section 4.3.1. At the mean
radius, φ = 0.5 and ψ = 0.4. For discussion purposes, we will also take the hub-to-tip ratio, HTR = rh/rt as 0.5.
Note that since the cross-sectional area is determined by the flow rate and the choice of flow coefficient φ,
once we choose the HTR, we can calculate the various required radii, r h, rm and rt. Finally, with the mean
radius known and the mean blade speed U m fixed by the choice of work coefficient ψ, we have the rotational
speed, ω = Um/rm.

To define the resulting spanwise geometry, we assume that the inlet axial velocity Ca is constant across
gHE  UCw , at every spanwise section. This fixes the ΔCw as a
the span and that we want the same total head rise,
function of radius and allows us to draw the velocity triangles for each spanwise section. The drawing shows
the resulting velocity triangles and the blade geometry based on the Euler Approximation, for three spanwise
sections. The table on the next page summarizes the corresponding values of the performance parameters.

C1

rt
C2 W1

U W1
W2

C1 rm

C2
W1

W2

C1

rh
W1

C2

W2
 Parameter TIP MEAN ROOT

Flow Coefficient, φ 0.395 0.5 0.791

Work Coefficient, ψ 0.25 0.4 1.0

Degree of Reaction, Λ 0.875 0.8 0.5

de Haller Number (Rotor) 0.788 0.699 0.62

de Haller Number (Stators) 0.845 0.781 0.62

Note:

(i)This blade design is clearly not acceptable. The work coefficient is far too high at the root and the de
Haller numbers there also indicate too much diffusion. The blade will need to be redesigned. If the stage is
still to produce uniform pressure rise across the span, the mean line work coefficient will have to be reduced.
(ii)The blade exhibits considerable twist across the span. Both this and the large variation in the design
parameters is a function of the hub-to-tip ratio, HTR = rh/rt. Increasing the HTR will make the blade more
uniformly loaded across the span, but since the cross-sectional area is fixed (by the choice of φ), this has
consequences for the tip diameter of the machine and the rotational speed. This is demonstrated in the
following sketch, which shows three different blades with the same annulus cross-sectional area but different
values of HTR. In multi-stage compressors, the HTR will normally increase along the machine since the
cross-sectional area is decreased to keep the axial velocity high. This is illustrated by the cross-section of the
compressor from the GE LM2500+ gas turbine engine (17 stages, PR = 23.3).

From: Wadia et al., ASME 99-GT-210

HTR 0.3 0.5 0.8
RPM Higher Lower
4.4 CHOICE OF SOLIDITY - BLADE LOADING LIMITS

The design parameters introduced in the last chapter apply to a stage or a blade row. Experience has
shown that it is possible to design a stage of good efficiency if the guidelines for those design parameters are
followed. The parameters also fully define the velocity triangles and the corresponding airfoil geometries.
However, the guidelines give no information about the number and the spacing of those airfoils:
in other words, about the solidity σ = c/s of the blade rows.

For a blade row, the larger the spacing between the airfoils the larger the mass flow that each airfoil is
required to turn. From the control volume analysis in Section 4.2, the resulting lift coefficient was given by

C L  2 s  cos mtan 1 tan 2



 1 
 2c  cos
 m tan 1 tan 2


and it is seen to vary directly with spacing,or inversely with the solidity. Just as for an isolated airfoil, there is
an upper limit to the lift that a turbomachinery blade can develop before it stalls. For a given set of inlet and
outlet flow angles, it is possible to stay below the loading limit by making the solidity of the blade row large
enough. Thus, the solidity of the blade row is selected on the basis of a blade loading limit. This is in contrast
to the work coefficient, ψ, which was a stage loading limit.

In the past, loading limits for compressor blades have sometimes been expressed in terms of the lift
coefficient (Horlock, 1958). In the early 1950s, Howell suggested that a well-designed compressor airfoil will
stall at
3
C  
CL  1 
 C2  3.3

and designers of low-solidity fans have sometimes used the criterion

c
CL    1.1
 
s

However, expressing the loading limit simply in terms of CL has been found to be unreliable. Recent practice
has therefore taken a somewhat different approach.
Howell (British Practice)

In the 1950s, Howell conducted an extensive series of cascade measurements on the compressor
airfoils that were commonly used in British compressor design. The performance was measured for a wide
range of the design parameters, including the flow turning angle and solidity. Howell varied the amount of
flow turning up to the onset of stall. The corresponding total-pressure losses were also measured. Howell
suggested that a suitable design turning angle for a blade row was that which corresponded to about 80% of
the turning that would result in stall. He also found that the losses were close to a minimum at this condition.
He therefore presented a correlation that could be used to estimate the solidity that would result in the blade
row operating at 80% of the stalling turning angle. This correlation is shown in the next figure (taken from
Saravanamuttoo et al., 2001).
 Knowing the design deflection and outlet flow angle from the velocity triangles, Fig. 5.14 can be used
to select a suitable value of solidity (note that the plot is expressed in terms of s/c = 1/σ).

Lieblein (NASA Design Practice)

Like Howell in Britain, in the 1950s NACA (now NASA) conducted an extensive set of cascade
measurements to determine the performance of compressor airfoils for a wide range of geometric and
aerodynamic parameters. As described later, these results became the basis for a compressor design system
which is now widely used, both in North America and in Europe (including Britain).

The drawing shows the hypothetical velocity distribution around a compressor blade.

C C2
SUCTION SURFACE

Cmax

C1
C1

C2

0
0 x/c 1.0
 The performance of the blade is limited by the deceleration (that is, the diffusion or adverse pressure
gradient) on the suction surface of the airfoil. If the diffusion is too great, the boundary layer separates, the
blade stalls, and the losses increase significantly. Lieblein proposed a parameter to measure the severity of the
diffusion:
Cmax  C2 (1)
D C1

As usual, relative velocities W would be used for rotor blades.

Unfortunately, Cmax is a function of the detailed flow around the particular airfoil, which would not be
known early in design. However, the larger the lift (or circulation) being generated by the airfoil the larger
Cmax must be. From Section 4.2.2, the circulation is given by

  sCw1  Cw2   sCw

and thus we can write

Cmax  C1  f 

 C1  f sCw 

Substituting into (1),

1
D  1  C 2  C f sC w
C
1 1

Experiments showed that the following form for D correlates the loss and stalling behaviour of a wide range
of blade geometries:

 Cw
C (2)
D  1  C12 2
C1
This parameter is known as the diffusion factor.
Note that (2) depends only on the upstream and
downstream velocities, which are known once
the velocity triangles are established.

The figure (taken from NASA SP-36,

1966) shows the variation of the total pressure
loss coefficient, ω1, with D. As seen, the losses
rise sharply for D > 0.65, implying the onset of
stall. At the design point, the diffusion factor
should therefore be less than this. A suitable
value might be D = 0.3 - 0.4. With D chosen,
the only unknown in (2) is the solidity and it can
therefore be used to select the value of σ.
4.5 EMPIRICAL PERFORMANCE PREDICTIONS

1. Introduction

The idealized stage analysis used in Section 4.3 made a number of assumptions that are not fully
satisfied in practice. For example, the flow angle at the trailing edge does not precisely match the metal angle,
as assumed in the Euler Approximation. Nor does matching the inlet flow angle to the inlet metal angle
necessarily result in the lowest losses. Finally, we need methods for estimating the losses generally, in order
predict the efficiency of the stage and thus complete its design. To accomplish a more realistic stage analysis,
we need to draw on correlations for the behaviour of actual blade geometries, as determined experimentally.
Such empirical correlations were alluded to in the discussion of blade-loading limits in the last
section.

Two systems for empirical performance predictions of axial compressors have been used fairly
widely. The British system, connected mainly with the name of Howell, will be discussed since it is relatively
easy to apply in hand calculations. However, it omits the influence of a number of blade geometric
parameters, does not directly apply to all the families of blade geometries that are in common use, and has
somewhat limited ability to predict the influence of factors such as compressibility.

A more comprehensive, but less easily applied, prediction system was developed by NASA during the
1950s and 60s. This system is summarized in a famous document, NASA SP-36, “Aerodynamic Design of
Axial-Flow Compressors” published in 1965. SP-36 continues to form the basis for much practical axial-
compressor design, both in North America and outside. The correlations presented in SP-36 have also been
re-evaluated and updated from time to time so that the system continues to be applicable.

It should be mentioned the largest gas turbine engine companies (eg. Pratt & Whitney, General
Electric and Rolls-Royce) have to some extent developed their own compressor design systems that reflect
their in-house design philosophies and proprietary blade profile designs. However, these systems are often
structured in similar ways and strongly influenced by the design systems that are available in the open
literature.

2. Blade Design and Analysis Using Howell’s Correlations

The figure shows the nomenclature used by Howell:

Nomenclature:

s = blade spacing
c = blade chord (solidity σ = c/s)
ζ = stagger angle
θ = α1' - α2' = camber angle
a = distance of maximum camber aft of blade leading edge
t = maximum thickness of blade

i = incidence = α1 - α1'
δ = deviation = α2 - α2' = difference between outlet flow angle and metal angle
ε = flow turning = α1 - α2

The nomenclature applies for a stationary blade row. For a rotor, replace α by β and use the relative
components of velocity.

Typical results obtained by Howell for a particular cascade geometry are shown in the following
figure. The figure (taken from Horlock, 1958) shows the variation of flow turning, ε and the total pressure
loss as a function of the incidence, i.

The cascade performance should depend on the blade and cascade geometry as well as the flow
conditions. Howell suggested that:

 , , losses  f blade geometry, cascade geometry, flow conditions

 f a c, s c,
,
 i,  if
He also found that the results collapse well onto universal curves 2 they are normalized in terms of the results

at the "nominal" (or "design"or "reference") flow condition for each cascade. The nominal condition is
defined, somewhat arbitrarily, as the condition at which the flow turning, ε, is 0.8 of the value at stall. Stall is
the appearance of boundary layer separation, towards the trailing edge, on the low pressure side of the blade.
The appearance of stall manifests itself in a rise in the losses and an impairment of the ability of the blade to
turn the flow. For convenience, Howell defined the stalling incidence as the positive incidence at which the
losses have increased to twice their minimum value. This definition is fairly easy to apply to experimental
data. As the figure above indicates, it also seems to correspond fairly well to the point of maximum flow
turning. The latter point could perhaps have been use as an alternative for identifying the “stalling” incidence.
 The superscript * is used designate nominal values of the flow quantities. Thus

ε* = nominal deflection = 0.8 εstall

The corresponding values of i, δ and α2 are designated i*, δ* and α2*.

Howell’s correlations can be presented in a small number of formulae and

graphs.

(a) Deviation at the trailing edge: n

 *  m 1  (1)
 
where 


 a
2
* (2)
m  0.23   2
 c 500
2
with all angles are measured in degrees.

For normal compressor rotor and stator blades n = 0.5. For the inlet guide vanes (IGVs) ahead of a
compressor stage, Howell suggested using n = 1.0 and a constant value of m = 0.19. Unlike typical
compressor rotor and stator blades, IGVs form an accelerating flow passage. They therefore behave more like
a turbine blade row and this accounts for the difference in the behaviour of the deviation.

(b) Flow turning:

Howell found that the nominal flow turning, ε*, correlated quite well with just the flow outlet
angle,
α2*, and the solidity of the blade row, σ = c/s
 *  f 2* ,
s
 
 c 

The correlation is usually presented graphically ( Fig. 5.14 from Saravanamuttoo et al.) and was used in
Section 4.4 to select the solidity.

The blade will often be used at other than the nominal (“design”) flow conditions. Howell was able to
correlate fairly successfully the “off-design” behaviour of the cascades by plotting the results against the non-
dimensional relative incidence, irel = (i - i*)/ε*. Figure 3.17 (taken from Dixon) shows the normalized flow
turning, ε/ε* as a function of i rel. The figure also shows the variation of the losses (expressed as a “drag
coefficient”) with relative incidence. As seen, the losses are close to a minimum at the nominal condition.
Loss estimates will be discussed separately later.
(c) Reynolds number effects:

Howell obtained most of his cascade data for a Reynolds number of 300,000 (based on blade chord
and upstream velocity). The resistance of the suction-surface boundary layer to separation is a function of the
thickness of the boundary layer and whether it is laminar or turbulent. Thus, the flow turning behaviour of the
blade row is a function of the Reynolds number, particularly at low values. Howell examined the dependence
of the flow turning on the Reynolds number. Figure 3.3 (taken from Horlock) shows the effect of Reynolds
number on the nominal turning.
 The correlations presented to this point can be used to predict the flow turning capability of a
compressor blade row. As mentioned, loss estimates will be considered later.

The correlations can be used in two ways: for analysis or for design.

Analysis: Predicting the performance of a blade row of specified geometry.

Design: Determining the geometry of a blade row which produces a specified
performance.

The approach is a little different for each case. Each will be described and the analysis mode
will then be illustrated with an example.

Analysis Mode Calculations:

In this case, the inlet flow direction (α 1 or β1) is specified and the blade row geometry is
known (α1',
α2', a/c, and σ = c/s). The goal is to predict the outlet flow angle, α2.

(i) The performance depends strongly on α2*. Since it is not known initially, it must be determined
(by iteration). Guess a value of α2*. *Use
 equations
  * (1) and (2) to calculate δ*. Then
2 2

Compare this value with the assumed α2*, revise as necessary and repeat until α2* and δ* are
consistent.
(ii) Read the value of ε* from Fig. 5.14. Then

1*  2 * 
 **
i  1*  1

The nominal conditions are now known.
(iii) If the actual i = α1* - α1' is different from i* then the blade row is operating "off-design". Fig.
3.17 would then be used to determine the actual flow turning. The Reynolds number correction
would be applied to the turning if appropriate.

Design Mode Calculations:

Again, the inlet flow direction (α 1 or β1) would be specified. Typically, the shape of the camber line
(ie. a/c) would also be selected. The goal is then to choose a blade row geometry (α1', α2', and σ = c/s) which
will give the desired outlet flow angle, α2. This application of the correlations is a little more complicated
since there is in fact a range of geometries which will satisfy the requirements.

One possible approach is to use the nominal values for the design point. This is reasonable since
nominal conditions give near-minimum losses and provide some stall margin. Then

*  *  and  *    
2 2 1 1 1 2

With α2* and ε* known, Fig. 5.14 is now used to choose the solidity, σ (this was the way that Fig 5.14
was used in Section 4.4). Since the blade row is operating at the nominal conditions, the deviation will also be
that given by Eqns. (1) and (2). However, δ* is also a function of the camber, θ. From the drawing of the
cascade, the flow turning is related to the camber by
 i (or in this case  *  i *     * )

Thus, the value of the camber will depend on the choice made for i*. Howell’s correlations indicate that there
is no unique choice for the design incidence, although he recommends that a value be chosen of a few degrees
at most. Reductions in camber can be compensated for by increases in incidence, and vice versa. Note that
these changes will also result in a change in the stagger of the blade row. In summary, according to the
Howell’s correlations a variety of blade geometries can produce identical aerodynamic performance. This
gives the designer some freedom to tailor the blade geometry to meet other possible requirements: eg. to
simplify the spanwise variation in the blade geometry, to alter a natural frequency, or to alter the stress level in
some region.

The Howell cascade measurements were made for the British C family of compressor blade profiles.
Therefore, a compressor designed according to the correlations is most likely to match the predicted
performance if the same blade profiles are used in the machine. The C4 profile, one of the most widely used
of the C-family profiles, is described in an appendix to these notes.

For use in computer programs or with analysis software (such as Mathcad or Matlab), the graphs for
the Howell’s correlations have been fitted by polynomials. These curve and surface fits are also given in an
appendix.
4.5.3 Blade Design and Analysis Using NASA SP-36 Correlations

The NASA correlations are based on a large body of cascade data collected for blades using the
NACA 65-series airfoil profile shape (Emery et al., "Systematic Two-Dimensional Cascade Tests of NACA
65-Series Compressor Blades at Low Speeds," NACA Report 1368, 1958). The results are correlated and
design procedures are summarized in NASA SP-36 ("Aerodynamic Design of Axial-Flow Compressors",
1965). SP-36 also includes data for double circular-arc (DCA) blades which have been used to design
transonic compressors.

As noted, Howell’s correlations do not give clear guidance for the choice of design incidence. While
the nominal incidence, i*, is a reasonable choice for the design point, it is also clear from Fig. 3.17 that using
i* does not in general minimize the profile losses. Howell’s correlations also do not take into account some
geometric parameters which are known to affect the blade performance, such the ratio of maximum-thickness-
to-chord, tmax/c. Finally, the Howell’s correlations are most suitable for analyzing the performance of a blade
row of specified geometry ("analysis mode") rather than determing a geometry which gives a desired
performance ("design mode").

By comparison, the NASA correlations are intended particularly for use in design mode, although
they can also be used for analysis. They guide the designer to a choice of design incidence which nominally
minimizes the profile losses. The correlations also account for more aspects of the blade geometry. The
drawback to using the NASA correlations is that reference must be made to more graphs than for the Howell’s
correlations.

For consistency with the SP-36 graphs, the procedures will be described in terms of the nomenclature
used by NASA.

As with the Howell correlations, the incidence and deviation are defined in terms of some reference
flow condition, although the definition of this condition is slightly different. Fig. 131 (from SP-36) shows the
definition of the reference incidence, iref. It is the
incidence half way between two off-design
values of incidence at which the losses are equal.
SP-36 usually refers to this as the “minimum-loss
incidence” although the losses will only be a
minimum if the loss “bucket” is symmetrical. As
evident from Fig. 3.17, this is not normally the case.
Nevertheless, the reference condition will be near
minimum loss and thus would be a reasonable choice
for the design point. The deviation produced at the
reference incidence is designated as δref.

For specified inlet and outlet flow angles, β1

and β2, the required flow turning, Δβ = β1 - β2, is
related to the camber, incidence and deviation by

    i  

If we use the reference values of incidence and

deviation then     iref
 ref (1)
It was found that the deviation angle and the minimum-loss incidence vary linearly with the blade camber:
 iref  i0  n
 ref   0  m

where i0 and δ0 are the values for the same blade when it has zero camber. Substituting into (1), the required
camber is given by

    i (2)
  1  n 0 m 0

The correlations are then used to find the values of the four unknowns on the right-hand side of (2).

The minimum-loss incidence at zero camber is written

i0  Ki  K  i 
sh
0 (3)
i t
where
10

= minimum-loss incidence for a blade with zero camber and 10% thickness
(i0)10

= shape correction to be applied when blades of other than the 65-series profile are
(Ki)sh
being used

(Ki)t thickness correction for blades with other than 10% thickness
=

For 65-A10 series blades, the correlations for the incidence related quantities are given on the
following graphs from NASA SP-36 (the graphs are reproduced at the end of the section):

(i0)10 = f1(β1,σ) Fig. 137

n = f2(β1,σ) Fig. 138

(Ki)t = f3(t/c) Fig. 142

For 65-series blades the shape correction, (K i)sh, is simply 1.0. However, it has been suggested that the same
correlations can be used to design C-series (C4 etc.) blades with circular-arc camber lines by setting
(Ki)sh = 1.1, and to design DCA blading by setting (Ki)sh = 0.7.

The zero-camber deviation, δ0, is obtained in a similar way:

 0  K  K   
0
(4)
sh
where t
10

= reference deviation for a blade with zero camber and 10% thickness
(δ0)10

= shape correction to be applied when blades of other than the 65-series profile are
(Kδ)sh being used

(Kδ)t = thickness correction for blades with other than 10%

thickness

For 65-A10 series blades, the correlations are given on the following graphs:
 (δ0)10 = f4(β1,σ) Fig. 161

(Kδ)t = f5(t/c) Fig. 172

As with the incidence, for 65-series blades the shape correction for deviation, (K δ)sh, is simply 1.0. For C4 and
DCA the same values of the shape correction as for incidence have been suggested: 1.1 and 0.7 respectively.

The deviation gradient, m, is also a function of β1 and σ. It is usually obtained using a deviation rule
similar to that used in the Howell’s correlations:

m
m 1.0 (5)
b
where

mσ=1.0
= value of m for a solidity σ = 1.0

= f6(β1) Fig. 163

= f7(β1) Fig. 164.

b

Eqn. (2) defines the camber required for the blade if the reference conditions are chosen as the design
point. However, there may be a variety of reasons to choose a different incidence at the design point, in the
same way that nominal conditions might not be used when designing a compressor using Howell’s
correlations. If i is different from iref then δ will also be different from δref. The resulting value of δ can be
predicted from

d (6)
    i  ref  di  ref
ref

i 
where (dδ/di)ref is given in Fig. 177 as a function of σ and β1.

The procedures just outlined can be used by the designer to obtain a blade row with a geometry which
will result in the required performance: that is, they are suitable for use in design mode. Of course, some
decisions must already have been made concerning the type of blading (C-series, 65-series, DCA etc.), the
camber line shape, if other than 65-series blades are used, and the maximum thickness.

Eqn. (6) also allows the correlations to be used in analysis mode. For analysis mode calculations the
following approach would be used:

(i) For the specified geometry and design inlet-flow direction, β1, the reference conditions are
first determined.
(ii) For an off-design inlet value of β 1, Eqn. (6) would then be used to predict the deviation.
This defines the outlet flow direction, β2, and the off-design velocity triangle is then known.

As with the Howell’s correlations, curve and surface fits for the SP-36 correlations are given in an
Appendix.
4.6 LOSS ESTIMATION FOR AXIAL-FLOW COMPRESSORS

4.6.1 Blade-Passage Flow and Loss Components

The drawing shows schematically the flow through the blade passage of a compressor rotor. In
addition to the frictional effects in the boundary layers on the surfaces of the rotor blades, there are a number
of other flow features that can generate losses. The losses due to each of these features are normally estimated
individually and then simply added to estimate the resultant losses through the blade passage.

For axial machines (both compressors and turbines), the losses are therefore subdivided into:

(i) Profile losses: These are the losses generated by friction in blade-surface boundary layers, by the
sudden expansion in area at the trailing edge, and by the mixing out of the wake
downstream of the blade.
(ii) Secondary losses: The slower-moving flow in endwall boundary layers is "over turned" by the
blade-to-
blade pressure field, as shown in the drawing. The fluid swept towards the low
pressure (“suction”) side of the passage is blocked by the blade surface and rolls up
into a "passage vortex" that generates additional losses through high shear stresses at
the endwalls and as it mixes with the downstream flow. The boundary-layer
(iii) Annulus losses: separation around the blade leading edge also results in a "horseshoe vortex".
These are generated by friction on the endwalls, mainly upstream and downstream of
the blade passage. The endwall losses inside the passage are normally assigned to
the secondary
(iv) Tip-leakage losses: There must be losses.
some clearance between the rotor blade tips and the compressor
casing. The flow that is driven through the tip gap rolls up into a "tip-leakage
vortex" as it interact with the main passage flow. There are viscous (frictional)
losses inside the gap, but most of the tip-leakage losses are generated through
downstream mixing with the surrounding fluid.

In transonic and supersonic compressors, there will be additional losses due to the presence of shock
waves.
4.6.2 Loss Estimation Using Howell’s Correlations

Howell gave simple correlations, expressed mostly in terms of drag coefficients, to estimate the losses:

(i) Profile Losses:

The profile losses were expressed as a function of both the incidence and the spacing-to-chord ratio,
s/l (Howell used the symbol l for chord length), as shown earlier in Dixon Fig. 3.17 (repeated here).

(ii) Secondary Losses:

Howell concluded that the secondary losses at the endwalls depended primarily on the lift being
generated by the airfoils, since this determined the pressure difference that drives the flow across the passage
to form the secondary flow. Thus

2
C DS  0.018 L
C
where from Section 4.2.1 the blade lift coefficient is given by

⎛
CL  2 tan
|  tan 2 
1 m
 |
⎝s⎞c⎠
cos
and

m  |⎛ tan   tan  |
1 2

arctan ⎝ ⎞ 2 ⎠

For the rotor flow, we would use the relative flow angles, β1 and β2, as usual.

(iii) Annulus Losses:

s sc
C DA  0.02h  0.02c h 
 AR
0.02
where h = blade height = rt - rh, AR = blade aspect ratio = h/c.

(iv) Tip-Clearance Losses:

The tip clearance loss is found to be a strong function of the height of the clearance gap τ compared
with the blade span h. Howell suggested that a 1% increase in the rotor clearance gap would reduce the stage
efficiency by 3%:

 clearance 

h
3

With the “drag coefficients” corresponding to the losses determined, the corresponding total-pressure
losses can be calculated from Eqn. (6a) or (6b) from Section 4.2.1. Equation (6b) is usually the most
convenient:

P0,loss  CD  |⎛ 1  2 ⎛ 1 | (6b)
|| 3
C ⎝2
⎞
a
⎞ cos 
⎝ m

⎠ ⎠

Section 4.6.5 explains how to use the estimated total-pressure losses to obtain the stage efficiency.

The NASA system for axial compressor loss prediction, described next, uses direct correlations for
total pressure loss coefficient, rather than for drag coefficient.
4.6.3 Loss Estimation Using NASA SP-36 Correlations

(i) Profile Losses

C C2

The profile loss system presented in SUCTION SURFACE

NASA SP-36 is associated with the name of Cmax

Lieblein, as were the blade loading limits

presented in Section 4.4. In that section, it was C1
C1
seen that the profile losses correlated quite
well with the diffusion factor defined by C2
Lieblein, which was defined as
0
0 x/c 1.0
Cmax  C2 (1)
D C1

However, Lieblein subsequently argued that the profile losses should depend primarily on the amount
of diffusion on the suction side of the blade. He therefore introduced an alternative parameter, known as the
equivalent diffusion ratio:

Deq  Cmax (2)

C2

Note that Deq resembles the deHaller number. Whereas the deHaller number defines the net diffusion between
the inlet and outlet of the blade row, Deq defines the local diffusion on the suction side of the airfoil. Lieblein
then correlated the profile losses with Deq and this approach has since been widely adopted.

As with the diffusion factor D, the exact value of the Deq is only known if the detailed flow around the
airfoil is known. For use in the early stages of design, an approximate value of Deq, estimated from the
circulation, is therefore used. The following correlation appears to be widely accepted:

cos 2 ⎛ cos 1  2
⎞
Deq |1.12  0.61  tan 1  tan 2 || (3)
 cos 1 | ⎠
⎝
y
The profile losses are reflected in a momentum deficit C2,ref
in the wake, as measured by the momentum thickness θ s
downstream of the airfoil:
C2 y 

C2 ⎞
s C2 ⎛
  ∫0 | 1  | dy
C2,ref ⎝ C2,ref ⎠
0
where C2,ref is the velocity outside the wake. The
corresponding total-pressure loss coefficient is then given by

2
⎛  ⎞   ⎛ cos ⎞
  2| | 1
|
(4)
|
⎝ c ⎠ cos 2 ⎝ cos 2 ⎠
where

P P
  101 02
(5)
C12
2

The loss correlation is then expressed in terms of the variation of the momentum thickness ratio, θ/c,
with equivalent diffusion ratio, Deq:


c
 f eq  (6)

Dobtained for NACA 65-series compressor airfoils, that was

The figure shows the original data set,
used by Lieblein. Also shown are various curve fits for the function in (6) that have been proposed over the
years. Note that losses begin to rise sharply at Deq 2.0 and this would be interpreted as the onset of stall. For
the original diffusion factor, Eqn (1), the corresponding value was D 0.6 (see Section 4.4)

AXIAL COMPRESSOR PROFILE LOSSES AT DESIGN INCIDENCE

Comparison of Correlations with Lieblein Data
0.12

0.1 Aungier
Wake Momentum Thickness Ratio, /c

Wilson & Korakianitis

Koch & Smith
Casey/Starke
Konig et al
0.08

0.06
Lieblein Data

0.04

0.02

0 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6

1
Equivalent Diffusion Ratio, Deq

Recently, Konig et al. (W.M. Konig, D.K. Hennecke & L. Fottner, “Improved Blade Profile Loss and
Deviation Models for Advanced Transonic Compressor Bladings: Part I - A Model for Subsonic Flow,”
ASME Journal of Turbomachinery, Vol. 118, January 1996, pp. 73-80.) investigated whether the Lieblein
correlation approach worked equally well for more recent compressor airfoil shapes. Their data are shown in
the next figure, along with the same curve fits.
 AXIAL COMPRESSOR PROFILE LOSSES AT DESIGN INCIDENCE
Comparison of Correlations with Konig et al. Data
0.12

Wake Momentum Thickness Ratio, /c

Aungier
0.1 Wilson & Korakianitis
Koch & Smith
Casey/Starke
Konig et al
0.08

0.06

0.04
Konig et al. Data

0.02
+
+ + + +

0 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6

1
Equivalent Diffusion Ratio, Deq

Although there is some evidence that more recent blade designs can tolerate somewhat higher values
of Deq before stalling, the curve fit suggested by Aungier (R.H. Aungier, Axial-Flow Compressors, ASME
Press, 2003) seems as reasonable as any, for both data sets:


 0.004「 
8

| 1.0  3.1 (7)

 1  0.4D  1 |]
eq 2 eq
c D
L 
Summarizing the procedure for estimating the profile losses:

(1) Deq is estimated from the velocity triangles and the blade row solidity using (3).
(2) From Deq obtain the momentum thickness ratio, θ/c, using (7).
(3) The total-pressure loss coefficient ω is then calculated from (4).

The method outlined here assumes that the blade is operating at its minimum-loss incidence, i* (see
Section 4.5.3). If i > i* then Lieblein suggested that (3) should be replaced by

cos 1 
2
cos 2 ⎛ tan 1  tan * 1.43 ⎞
|
Deq |1.12  0.61  2  a i |
 cos 1 |⎝ i
⎠ 
where a = 0.0117 for NACA 65-series blades and 0.007 for C4-series circular-arc blades.
(ii) Endwall Losses

NASA SP-36 does not provide clear guidance for estimating either the secondary losses or tip
clearance losses for the purposes of meanline analysis.

Instead, most recent text books (eg. Japikse & Baines) and papers seem to recommend a method
developed by Koch & Smith at General Electric (Koch, C.C. and Smith, L.H., “Loss Sources and Magnitudes
in Axial-Flow Compressors,” ASME J. Eng. for Power, Vol. 98, 1976, pp. 411-424). The method provides
combined estimates for the effects on stage efficiency of both secondary flows and tip leakage. This is
physically reasonable since, where both are present, the secondary and tip-leakage flows are in close proximity
and tend to interact significantly. Unfortunately, the method is somewhat difficult to apply since it requires a
fairly detailed knowledge of the stage geometry. It is also necessary to specify how close the stage is to stall
at the operating point for which the loss estimates are being made. Nevertheless, because of the importance of
endwall losses and the apparent widespread acceptance of the Koch & Smith method, it is worth examining.

The final output of the method is a correction to the stage efficiency, expressed in the form

 2 * 
 1 h
  (1)
P
 1  2
 h 

where ηP = stage efficiency as calculated from the profile losses only

δ* = average displacement thickness of the two endwall boundary layers
ν = average tangential force-deficit thickness for the two endwall boundary
layers

The tangential force-deficit thickness is a measure of the reduction in blade force near the endwalls due to the
lower fluid velocity present in the endwall boundary layers.

Koch & Smith provide correlations, derived from very wide-ranging tests conducted on a large, low-
speed compressor test rig, for estimating the values of δ* and ν. The drawing defines some of the geometric
parameters that appear in the correlations.

s = spacing
λ = stagger angle
g = staggered 
= spacing
s cos
In addition, the following are used
s
ε = tip clearance
g
h = blade span 
ξ = axial gap between rotor and stators

In the correlations, average values of the parameters are used. For

example, the staggered spacing used is the average value for the rotor and
stator blade passages. Similarly, the tip clearance would be the average of c
the values for the rotors and the stators. Normally, this would result in the
clearance value being half of that for the rotor blades, since the stator
clearance is usually zero. However, stators are sometimes cantilevered
from the casing wall and have a clearance at the hub wall. If the stators are variable pitch, they will also need
clearance.

The Koch & Smith correlation is embodied in three graphs.

(a) Displacement thickness. The first graph is used to estimate the displacement thickness as a function of
the clearance and the pressure rise ratio:

2 * C
 f  P
,  
g
 CP,max g
where

C P


with P the static pressure rise across the stage
P and q the average of the inlet dynamic pressures for the rotor
and stator rows. CP ,max is the maximum value
q of the static pressure rise coefficient for the same stage,
corresponding to the stalling of the stage.

The pressure rise ratio is probably the most difficult input to obtain. However, for preliminary design
it may be sufficient to choose a value that seems generally consistent with the stage and blade loading that has
been chosen. For example, if the deHaller numbers are low and the solidities have been selected to give
relatively high values of the diffusion factors, the pressure rise ratio would be expected to be towards the
higher end of the scale.

0.55

0.5

0.45

0.4

0.35

0.3
2*/g

/g = 0.10

0.25 0.075

0.050
0.2
0.025

0.15 /g = 0.0

0.1

0.05

0 0.75 0.8 0.85 0.9 0.95 1

0.7

CP/CP,max
(b) Effect of Axial Spacing. Koch & Smith concluded that the average displacement thickness of the endwall
boundary layer would vary with the axial spacing ξ between the rotor blade and the stators. If that spacing is
different from 0.35s, then the following correction is applied to the displacement thickness given by the
previous figure.

1.1

1.05

1
2*/(2*)ref

0.95

0.9

0.85

0.8 0.1 0.2 0.3 0.4 0.5 0.6 0.7

0
Axial Gap/Blade Spacing, /s

(c) Force Deficit Thickness. Finally the force-deficit thickness is correlated against the displacement
thickness as given in the following figure.

0.9

0.8

0.7

0.6

0.5
2/2*

0.4

0.3

0.2

0.1

0
0.75 0.8 0.85 0.9 0.95 1

CP/CP,max
 The staggered gap, g, is not a commonly-occurring variable in performance correlations. However,
the parameters in which it appears can be related to more familiar ones. For example,

 
hcs
g
 AR  (2)

h cos 
hcsg
Values of the tip clearance ε are often specified as a fraction of the blade span h. Therefore, reasonable values
of ε/h would be known early in design. Eqn. (2) also implies that the Koch & Smith correlation can be used to
conduct parametric studies to investigate the influence on the endwall losses of common design parameters
such as the solidity σ and the blade aspect ratio AR = h/c.

For use in Eqn. (1), note that

2 * 2 * g s

ch

g sch

2 * cos 

g AR
4.6.4 Effects of Incidence and Compressibility

The Howell correlation for the profile losses for C-series airfoils presented in Section 4.6.2 included
the influence of incidence. As seen, the losses rose more rapidly with positive incidence than with negative.
However, the precise behaviour of the losses with incidence is strongly influenced by the geometry of the
blade section.

In addition, the Howell results apply only for low subsonic values of the inlet Mach number. The loss
behaviour of the airfoil is also strongly influenced by the inlet Mach number.

We do not have time in this course to go into these issues in detail. Therefore, only some
representative results are presented to illustrate the complexities.

The figure (taken from SP-36) shows the variation of profile losses with both incidence and inlet
Mach number for four different airfoil and cascade geometries.

Note that the two examples of the British C4-series airfoils differ mainly in the shape of the camber
lines and yet their sensitivity to both the inlet Mach number and the incidence are significantly different.

The double circular arc (DCA) profiles were specifically developed by NACA for use in transonic
compressors. It is seen that their sensitivity to Mach number is delayed to a higher inlet Mach number than
some of the other shapes.
 The strong influence of the detailed airfoil geometry on the behaviour at both off-design incidence
and with increasing inlet Mach number obviously makes it more difficult to devise simple correlations for the
losses, liked those presented in Sections 4.6.2 and 4.6.3. For a recent attempt, see W.M. Konig, D.K.
Hennecke & L. Fottner, “Improved Blade Profile Loss and Deviation Models for Advanced Transonic
Compressor Bladings: Part II - A Model for Supersonicc Flow,” ASME Journal of Turbomachinery, Vol. 118,
January 1996, pp. 81-87.
4.6.5 Relationship Between Losses and Efficiency

With the total-pressure losses across the stage P0,loss determined from the correlations described in Section
4.6.2 or 4.6.3, the total-pressure rise across the stage is then:

P0  P0, ideal  P0, loss

where ∆P0,ideal is the pressure rise that would have been obtained in an ideal machine having the
same work input.

(i) For compressible flow of a perfect gas, the ideal pressure rise is


P0,ideal  P
03  P01 where P03 | ⎛ T03| ⎞ 
P01 ⎝ T 01
1
⎠
and T03 is the (actual) final T0 corresponding to the work input: ∆h0 = Cp(T03 - T01)

T0
P03

P03

T03

T03 P01

T01

Then the actual P03 is obtained from

P03  P03  P0,loss

where P0,loss is as predicted for the rotor and stators together.

The corresponding efficiency can be calculated by determining the power required to compress
through the same pressure ratio, P03/P01, in an isentropic machine:
 1

T03 ⎛P ⎞03
| |
T01 ⎝ P01 ⎠
Then

T03  T01


T03  T01

For hand calculations, the following approximation results in only a small error
P0

 P0, ideal

If any of the loss components is expressed in terms of an efficiency decrement, as is sometimes the case with
tip-leakage loss, its contribution to the total pressure losses can be estimated from

P0, TL  clearance P0, ideal

and this would be included in the earlier summation of losses.

(ii) For incompressible flow, we can use

P0, ideal   g  H E   U 2 C w 2  U 1 C w 1    U C w

and P0

 P0, ideal

where, as before
P0  P0,ideal  P0,loss
4.7 COMPRESSOR STALL AND SURGE

4.7.1 Blade Stall and Rotating Stall

As the flow rate through a compressor or fan is reduced at constant rotational speed, the velocity
triangles show that the incidence at the leading edge of the rotor blades is increased. If the incidence becomes
too large, the blades may stall.

The disorganized flow in the stalled region partially blocks the blade passage. As a result some of the
fluid that was previously passing through the stalled passage is diverted to the adjacent passages, as indicated
in the middle figure. This has the effect of increasing the incidence at the airfoil lying next to the stalled
region while reducing the incidence at the other adjacent airfoil. The increase in incidence on the one adjacent
airfoil may cause it to stall in turn. On the other hand, the airfoil which has its incidence reduced will move
further away from stall, or may unstall if it was previously stalled. Thus, there is a tendency for the stall cell
to migrate from blade passage to blade passage in the opposite direction to the rotation. This phenomenon is
known as rotating stall. The stall cells move in the opposite direction to the rotation at a relative speed which
is about half the rotational speed.

MIGRATION OF
STALL CELL

SEPARATION OF
BLADE BOUNDARY ROTATING STALL
NORMAL ATTACHED
LAYER - BLADE STALL
FLOW

The rotating stall can take a number of patterns. It may involve only one blade passage, or a large
number of adjacent blade passages around the annulus. If the rotor is at the front of a multi-stage compressor,
it will have a relatively low hub-to-tip ratio. As seen in Section 4.3.4, the loading will then vary considerably
across the span and it will be the hub region that will have the highest loading and therefore be the most likely
to stall. In that case, the may stall cells may only involve part of the span of the blade. If the blades have high
hub-to-tip ratio, the stall is more likely to extend across the full span.
 LOW HTR - HIGH HTR -
ROTATING STALL FULL SPAN STALL, ALL PASSAGES

Particularly if the rotating stall occurs at low speed and only involves part of the span, it may not be a
danger to the machine. It is nevertheless undesirable since:
(i) The stalled passages, and therefore the stage, produce less pressure rise.
(ii) The stage losses will be higher, leading to lower efficiency.
(iii) The fluctuating forces on the blades as they successively stall and unstall will be a source of
noise.

For reasons discussed in Section 4.8, it is fairly common for the early stages of multi-stage axial
compressors to experience some rotating stall at low rotational speeds. If it is present only during start-up and
shut-down of the machine, this may be acceptable.

4.7.2 Surge

If the stall is very extensive, the pressure rise may be affected to the point that the slope of the ΔP0
versus m˙ characteristic becomes positive. As will be shown in Chapter 7, if this occurs the system of
which the compressor is a part can become dynamically unstable. If this instability is triggered, the result is
known as surge. The peak point of the compressor characteristic is therefore often identified as the surge
point.

P0
SURGE

ROTATING
STALL

N const.

m
˙
 In Chapter 7, an approximate, unsteady-flow analysis is developed for a compression system. This
analysis identifies the various factors that will make the system more or less prone to surge. However, a
simple physical argument can illustrate the sequence of events that might occur during a surge event.

Consider a gas turbine engine that is operating at high speed and high power or thrust
output.

At steady state, the combustor is being “filled” with gas by the compressor at the same rate as it is
being “drained” through the turbine. At any instant in time, there is a fairly large mass of gas present in the
volume of the combustor. Now suppose that the last stage of the compressor suddenly stalls. This might be
due to some disturbance that causes a drop in the mass flow rate through the machine. Since the last stage
normally has a high hub-to-tip ratio, the stall may involve the full span of the blades and all of the passages, as
described in the last section.

If the stall is very extensive, there will be an abrupt drop in the pressure of the gas delivered by the
compressor. The flow area through the turbine is relatively small and this limits the outflow through the
turbine. Consequently, the pressure in the combustor will drop somewhat gradually. It the rate of pressure
drop is too slow, the situation can occur that the pressure in the combustor is higher than the pressure at the
compressor outlet. Since fluid tends to flow from a region of high pressure to one of low pressure, it is
therefore possible for the high pressure gases from the combustor to flow back upstream into the compressor.
The pressure in the combustor will eventually drop to below the compressor discharge pressure, at which point
the flow in the compressor may re-establish itself. However, if the conditions that led to the initial stall are still
present, the whole process can repeat. The cyclic flow reversal in the compressor can result in very large
fluctuating forces on the blades which can destroy the machine. In the gas turbine engine, the abrupt drop in
compressed air supplied to the combustor can also lead to over-temperatures and resultant serious damage to
the turbines.
4.8 MULTI-STAGE COMPRESSORS

We now examine the aerodynamic behaviour of multi-stage compressors.

For arguments sake, we will consider a hypothetical four-stage compressor made up of stages with
identical aerodynamic characteristics and thus identical stage design points. Therefore, at design point for the
machine as a whole, each of the stages will be running at their individual design points, which occur for the
same value of the flow coefficient φ= Ca/U for all of the stages. Assume also that the mean radius, and thus
the blade speed U, is the same for all four stages. Since the density of the gas increases across each successive
stage, to maintain the constant axial velocity Ca needed to keep φ constant it is necessary to reduce the annulus
area along the machine. This variation in the cross-sectional area would be determined at the design-point
flow conditions.

Ca

U
U

Ca
FOUR IDENTICAL STAGES


DESIGN N
1
 VARIATION
2 m˙  C
a

DUE TO
COMPRESSION
D 3 A
OR Ca
4
 m˙
A
Ca TARGET Ca
VARIATION

D 

A
AREA VARIATION
TO ACHIEVE
TARGET Ca

DESIGN POINT

Now consider what happens if the compressor is run at a rotational speed that is lower than the design
value. To see the effect, we will just consider the first two stages:

STAGE I STAGE II

1 2 3
 Since the annulus area has been adjusted such that at the design point the two stages have the same
flow coefficient:
D
 I D  II   D
For stage I
C aI
 D

I D
UD

and the corresponding pressure rise is

P 1 P I
 D

PI  P3  P 1 and the outlet density 3

D D
RT3 D

Then for stage II

Ca 3
m˙  D

Ca 3  II

D
D3 A 3
D
D
UD

where, as mentioned, A3 was adjusted to give  I   II .

D D

Now consider the effect of halving N (ie. halving U) while also halving m˙ (ie. halving Ca) to
keep Stage I operating at its design φ:

a1D 1
C
I  2  I
1U D

2
D 

Thus, Stage I will also be operating at its design ψ. However, the absolute Δh0 (and thus ΔP0} varies as U2 and
the pressure rise is therefore reduced to
1
P1  4 P D
1
P  4 PD and 3 RT3

Neglecting the changes in T3, which will be relatively much smaller than the changes in P 3, then

1
 3 P  4 PD
  1.0
3 P  PD
D

The corresponding change in Ca3 is then

m˙ 3
3 A3  D
m
Ca 3 
Ca 3 3

  1.0  1  k
D m˙D ˙D 
 3 D A3 m˙ 
2
where k > 1/2. Then the flow coefficient for Stage II
 C kC k II
 II  U a 3 1 U a 3 1 D
D
D
 2
2

and since k > 1/2,  II   II . Thus, the non-dimensional operating point for Stage II shifts to a lower value
of
D

ψ than the design value. Stage I undercompresses the fluid due to the reduction in U 2. But stage II
undercompresses the fluid even more than Stage I due to the reduction in both U2 and ψ. This effect only
increases in the subsequent stages. For the 4-stage compressor with four identical stages we would therefore
expect to see the following pattern of operating points:

Ca

U
U


Ca
LOW N, STAGE 1 AT DESIGN

1
2

D 3
4
m˙   Ca A  m˙

DES
 VARIATION
(UNDER-
Ca  ˙
m A COMPRESSION)

D 
Ca DESIGN C a
VARIATION

NEW Ca
VARIATION

A
AREA VARIATION
(FIXED)

REDUCED RPM, REDUCED

m˙


If we now reduce the mass flow rate at the low speed operating LOW N, THROTTLED
point, keeping N constant, the flow coefficient for Stage I will be 1
lowered. Stage I will then be producing slightly higher pressure rise. 2

However, the effect of the low U2 is much greater than the 3

D
small increase in ψ and Stage I will still be producing much lower 4
pressure rise than at the design N. Consequently, Stage I is still
under- compressing the fluid and the downstream stages will again
be at successively higher values of φ. We therefore conclude that if
we throttle the flow further, Stage I will be the first to reach its
stalling value of φ.
D 
 
HIGH N, STAGE 1 AT DESIGN

If we then consider operating points above the design N, similar 4

arguments will lead to the conclusion that each stage is over-compressing the 3

fluid. Consequently, the C a into successive stages decreases and so does the D
2

1
flow coefficient φ. We would therefore expect to see
the approximate pattern
of operating points shown. Note that if we throttle the flow further, it will now
be the last stage which stalls first.
D 

Combining these arguments, we can plot the expected map for the compressor as a whole.

1 2
3
4
P02
STAGE STALL LINE
P01

HIGH-SPEED
OPERATING POINT

max HYPOTHETICAL
STEADY-STATE
DESIGN OPERATING LINE
POINT

COMPRESSOR
SURGE LINE

CONSTANT
SPEED LINE

4 N

3 2
T01
1

m˙ T01

P01

Note that:
(i) Over most of the map, we assume that the stalling of any stage results in compressor surge. As a
result, when the onset of the stall switches from the front to the back of the machine (near the design
N), there is a discontinuity in the slope (or “knee”) in the surge line.
At low values of N, stall is expected to occur first in the first stage of the compressor.
However, since the early stages of the compressor have lower hub-to-tip ratios, the stall there is more
likely to be part-span, rotating stall (as discussed in Section 4.7). This, combined with the fact that
the absolute forces on the blades will be low at low N, means that some degree of rotating stall is
acceptable at low N. As a result, at the low end of the map the surge line has a “kink”, indicating that
some early-stage stall is allowed.
(ii) At the design point of the compressor, all of the individual stages are operating at their design points
and therefore have their maximum efficiencies. From the earlier discussion, it is evident that at any
other operating point at most one of the stages will be operating at best efficiency. Therefore, the
efficiency of the overall compressor will be less than its value at design. For this reason, the lines of
constant efficiency are shown as closed contours surrounding the design point.
 The compressor map shown is a hypothetical one. In practice, the individual stages in a multi-stage
machine will not all have identical characteristics. Nor are the stall lines for the individual stages likely to
cross at exactly the same point on the map, and as a result the knee in the surge line will probably not be as
well defined. Nevertheless, many of the features are reproduced by actual compressor maps, as shown on the
following:

NASA 8-Stage Research Compressor

Pratt & Whitney TF30 LP Compressor

4.9 ANALYSIS AND DESIGN OF LOW-SOLIDITY STAGES - BLADE-ELEMENT METHODS

For solidities, σ, less than about 0.4 each blade can be treated as an isolated airfoil. Note that σ = 0.4
was the lowest value of solidity that appeared on the NASA SP-36 correlations (Section 4.5.3). Usually, the
blade is divided into a series of spanwise segments or blade elements. Three-dimensional flow effects in the
form of spanwise flows are usually neglected, although the downwash induced by the trailing vortex system is
sometimes taken into account. This approach, known as the "blade-element method", is commonly used to
design propellers and low-performance axial fans.

Consider the flow relative to a blade element. The element behaves like an isolated airfoil in a stream
in the direction of the vector mean of the inlet and outlet flows:

ZLL = zero-lift line of blade element

Wm = vector mean velocity relative to blade element

Wm Ca ⎛ 1tan   2tan| ; C
⎞a
;  m  arctan|

Q

= angle of attack of blade element = angle between ZLL and Wm
∆L = lift force onblade
cos m ⎝
element (perpendicular 2 to W ) ⎠
m

= 1/2pWm 2c∆rC L A
where c = chord length of blade element
(1)
∆r = radial width of blade element
CL = lift coefficient of blade element (as obtained from airfoil
characteristics and )
∆D = drag force on blade element (parallel to Wm)
(normally ∆D ∆L)
∆X = axial component of force on blade element
∆X ≈ ∆L sinβm (see note at end of section)
∆Y = tangential component of force on blade element

CL, CD = fns [ , section shape, Re] (2)

- as obtained from airfoil data

The axial force is obtained from the momentum equation (with Ca = const.):
 Fx  N B X  AP  2rrP (3a)

where NB = no. of blades

∆P = static pressure difference across the blade row

Substituting for ∆X in terms of the lift coefficient

⎛ ⎞
NB| Wm2 crC L | sin m 
1⎝ ⎠
2rrP
2 ⎛ ⎞
N B | Wm2 cC L | sin m  (3b)
2rP
1⎝ 2
⎠
For the input power, from the energy equation

Win˙  T  
QP0
 0  QUC 
w (4a)
R
Qh
assuming ∆P0 is small so that ∆P 0 /p R ≈ ∆h0, and where

∆T = torque applied to flow through annulus width ∆r

∆Q = volume flow rate through annulus width ∆r
= rotor efficiency ( R =1 if CD = 0)
R

∆P0 = total pressure difference across rotor (usually ∆P0 ≈ ∆P since C1 ≈ C2)

Substituting for the torque in terms of the components of the lift and drag forces (∆T = NBr∆Y)

P0
Win˙  NB | WmcrC L cos m C Dsin 
「1 2

mr   Q  Q C w 
L2 |] R U (4b)
Rotor and stator blade rows can then be designed using Eqns. (1) - (4). Iteration will generally be
necessary since W2 is a function of ∆L, which is a function Wm, which in turn is a function of W2. The
analysis would be performed at enough spamwise sections to define the full blade geometry.

Propeller analysis usually takes into account the "downwash" induced along the blade span by the
trailing tip vortices from the blades. The downwash would slightly alter the effective flow incidence seen by
the blade and thus the lift it develops.

To make the velocity triangle diagram clearer, the blade was sketched with somewhat lower stagger
angle than would normally be found in practice. The diagram shows the force triangles for a more realistic
value of the stagger angle:
 y D

L
x
X F Wm

Y

Note that ∆D makes a noticeable contribution to the magnitude of ∆Y but has a much smaller influence on the
magnitude of ∆X. This is the reason that ∆D can be neglected when determining ∆X, but needs to included
when determining ∆Y.
2. IDEALIZED STAGE GEOMETRY AND AERODYNAMIC PERFORMANCE

The geometry of an axial-flow turbine blade is similar that of an axial-flow compressor blade, except
that camber is usually much larger. The stage consists of a set of stators ("nozzles") followed by a rotor. The
nozzles control the swirl in the flow entering the rotor and the rotor then extracts work from the fluid by
removing swirl. This arrangement of components results in stage aerodynamic characteristics that are very
different from those obtained for an axial compressor.

We begin again by estimating the stage performance based on an idealized stage:

(i) Simple velocity triangles are assumed: constant axial velocity through the stage and constant
mean radius, resulting in constant blade speed where the mean streamline enters and leaves the rotor.
(ii) Approximate blade geometries are obtained using the Euler Approximation.

Consider again the reaction turbine sketched in Section 3.5. The drawing shows the velocity
triangles:

Now reduce the mass flow at constant N, using the Euler Approximation to determine the outlet flow
angles. From the drawing shown over, the flow coefficient is reduced

C
  Ua  D


Clearly, ΔCw is smaller than at design. This is also consistent with the reduction in rotor blade incidence.
Thus

h0 Cw 
   U D
U 2

Therefore, ψ varies directly with φ. Compare this with the case of compressors where they varied inversely.

Next, consider reducing U while holding Ca constant:

From the velocity triangles

C C
  Ua  D   Uw  D

 
and the same trend is found as when the mass flow rate was changed.

Now consider the absolute output. From the Euler equation

h0  UCw
and from the velocity triangles, ΔCw increased as U decreased. It is not entirely clear whether the product
UΔCw has increased or decreased. However, it is clear that it, and therefore Δh0, has not changed very much.
Compare this with the compressor case, where a reduction in U resulted in a large reduction in UΔCw:

Summarizing, based on the velocity triangles, the aerodynamic performance characteristics of axial
compressors and turbines differ in two main ways:
(i) ψ versus φ, and therefore ΔP0 versus m˙ , is negative for compressors, positive for turbines.
(ii) The energy transfer Δh0 is a strong function of U for compressors, but only a weak function for
turbines.
 The following figures show the actual characteristics of the gas-generator turbine of the Orenda OT-2
gas turbine engine. Note that it is conventional to use the pressure ratio as the independent variable for
plotting turbine aerodynamic characteristics.

The characteristics confirm that the mass flow-pressure ratio characteristic is only a weak function of
the rotational speed. However, this does not mean that the rotational speed is not important in order to have a
high output of useful work. As seen from the velocity triangles, if the rotational speed is reduced below the
design value, the energy released by the fluid, Δh0 = UΔCw, may not be changed very much, but this is also
accompanied by high incidence on the rotor. This will lead to higher losses and therefore poor efficiency.
This is confirmed by the OT-2 efficiency curves. Thus, to have high energy release by the fluid and to
recover most of that energy as useful shaft power output, it is necessary to have high rotational speed.
3. EMPIRICAL PERFORMANCE PREDICTIONS

Cascade results are used for meanline analysis of turbines in much the same way as for axial
compressors. Again, primarily British results will be presented, but these are also widely used in North
America.

1. Flow Outlet Angle

Turbine blade rows, for gas turbine engines in particular, often operate at choked conditions or with
mildly supersonic outlet flow conditions. The correlations for outlet flow angles for such blade rows are
generally divided into two sections: one for low speeds (usually taken as M2 ≤ 0.5-0.7) and one for the sonic
condition (M2 = 1.0). For intermediate values of M2 the outlet angle is usually assumed to vary linearly
between the low-speed and the sonic values.

(i) Low Speed (M2 ≤ 0.5)

As mentioned earlier, the Carter & Hughes correlation for deviation (used by Howell for compressors)
has also been used for turbines:

  m⎛ s|⎞
n

⎝
|
c⎠
where θ = camber angle and the value of m is obtained from Fig. 3.6 (from Horlock).

For turbines, n is generally taken as 1.0, as used for compressor inlet guide vanes (as opposed to the value of
1/2 used for compressor rotor and stator blades). However, the Carter & Hughes correlation tends to over-
estimate the deviation for most modern turbine blades.

A more satisfactory (but less convenient) correlation is that due to Ainley & Mathieson (A-M). Their
correlation uses the so-called gauge angle θg as a reference angle to which the actual outlet angle is related:

⎛
g  1
| |
cos o⎞s
⎝

⎠
where o = throat opening and s = blade spacing. For an infinitesimally
thin blade which is straight from the throat to the trailing edge, the gauge
angle would define the direction normal to the throat line.

For low speed flow, A-M correlated the outlet flow angle a2
with the gauge angle. Fig. 7.13 (from Saravanamuttoo et al.) shows the
variation for a “straight-backed” blade: that is, a blade for which the
suction side is straight from the throat point to the trailing edge. The o g
curve in the figure can be approximated by
s

⎛
2  1
| |  12 g
1.1625cos o⎝⎞s

However, most turbine blades are not straight

backed. Instead they have a certain amount of “unguided
turning” as defined by the angle θu. In A-M’s day, if the
suction surface was not straight from the throat to the
trailing edge, it was usually defined by a circular arc. A-
M therefore corrected the outlet angle as follows:

 2  1.1625 1
⎛| o|⎞ 12  4 |⎛ |
cos ⎝s⎠
s⎞
⎝

where e is the suction side eradius

⎠ of curvature.
Unfortunately, modern turbine blades usually do not use
circular arcs to define their surface shapes. As a result, e
is not constant and generally not known. To use the A-
M correlation it is therefore necessary to obtain an
“equivalent” value of e. An approximate value can be
calculated from the unguided turning angle as follows:

s u
e 2
⎛ o⎞
180 1  | |
⎝s
for θu measured in degrees. ⎠
(ii) Sonic Condition

For M2 = 1.0 and a straight-backed blade, A-M indicated that the outlflow angle would be equal to the
gauge angle:

⎛
2  1
| |
cos o⎞s
⎝

⎠ suction side radius of curvature. The results were

For a curved-back blade, this was again corrected for the
presented graphically but can be approximated by the following curve fit:

⎛ 1.7874.128
s
| 1 ⎛
⎛ o ⎞ |⎞ ⎛ s
e
2  1
| sin | |
| | | |
cos ⎞⎝ s ⎠ | ⎝ ⎠ o⎝⎞s
⎝
⎠
e⎠

As mentioned, for 0.5 ≤ M2 ≤ 1.0 the value of a2 is obtained by linear interpolation:

2 2
M 2  0.5

 2M 2 1  2 2
2
M
2
 0.5

M 1.0
2. Choice of Solidity - Blade Loading

1. Zweifel Coefficient

In 1945, Zweifel introduced a tangential force coefficient to measure the loading of turbine blades.
Consider the control volume enclosing a single airfoil in a row of turbine blades. The CV extends unit depth
in the z direction.

Y
y

x
X

P1
C1

1 Cw1
s
Ca1

P2

Ca2

1
cx
Cw2

C2

Apply the linear momentum equation in the y direction:

(1)
Fy  m˙Vy 2 V y 1
Because the top and bottom faces of the CV are periodic boundaries, the pressure forces on them exactly

balance each other in both the x and y directions. Thus, the only contribution to ΣFy is the blade force Y.
Then

Y  m˙Cw2  Cw1  (2)

From the velocity triangles,

Cw1  Ca1 tan Cw2  Ca 2 tan

1 2

and for unit span, m˙  2 Ca 2 s1 . Note that we are using here a common convention in turbine
design practice that α1, α2, Cw1, and Cw2 are all taken to be positive: that is, we are not rigidly following
the sign conventions introduced earlier. Then (2) can be written
  C 
Y  2 sCa2  tan 2  a1 tan 2 
2   Ca 2 

or, since Ca 2  C2 cos

2 ,
1  Ca1 
Y  2 C22 2s cos2  tan 2 (3)
2    tan 2 
2   Ca 2 
The tangential force in (3) is just the integrated effect of the pressure distribution around the airfoil:

P
"IDEAL" DISTRIBUTION

P0

P1

1 2
PS P0  P 2 C 2
ACTUAL 2
DISTRIBUTION

P2

SS

x
0 cx

c
Y   PPS  PSS
x

dx
Zweifel then defined a reference, “ideal” loading distribution. This corresponds to the maximum loading that
could be achieved with the same inlet and outlet conditions while avoiding adverse pressure gradients on the
suction surface. This distribution, which is not physically realizable, corresponds to a pressure on the pressure
side of P0 and a pressure on the suction side equal to the discharge pressure P2. The resulting “ideal”
tangential force is then

1
Yideal   P
0  P2 x 1  2 C22 cx (4)
2
c
The Zweifel coefficient is then obtained by taking the ratio of the actual to the ideal tangential forces

Y
Z
 Yideal

Substituting from (3) and (4) then

  s  2  C 
Z  2 cos  2 tan 2  a1 tan
  1

 x
c C a2

(5)
Note that this definition neglects thesign convention for angles. For a typical turbine blade, α1 and α2 have
opposite signs. If the signs of α1 and α2 are taken into account then the coefficient becomes:

C 
Z  2 s  cos2  2 tan 2  a1 tan
  1
 cx  Ca2 

As usual, for rotor blades β replaces α. The normal definition of the solidity is σ = c/s. The way the Zweifel
coefficient is defined results in the “solidity” being expressed in terms in term of the axial chord length, cx,
rather than the true chord, c. The relationship between the true chord and axial chord can be seen from the
drawing, where ζ is the stagger angle: C
X

Zweifel (1945) concluded, based on European cascade data from the 1930s and 1940s, that Z .
8.gave minimum profile losses. Thus, for given velocity triangles, the “optimum” s/cx is that which gives the
value of Z which results in minimum profile losses:
• If s/cx is too high (which corresponds to low solidity), losses will be high due to separation,
• If s/cx is too low, profile losses are high because of excessive wetted area.

Using the Zweifel coefficient to choose s/c x is analogous to the use of the diffusion factor to select the solidity
for axial compressors. Since Zweifel’s time, profile design has improved and today turbines are often
designed with considerably higher values of Z ( Z = 1.00-1.05 is common).
 5.3.2.2 Ainley & Mathieson Correlation

Ainley & Mathieson developed a widely used loss system (see next section), based on British turbine
cascade data from the 1940s and 1950s. They likewise identified the geometries that gave minimum profile
losses for different combinations of inlet and outlet flow angles. These optimum geometries, expressed as
optimum s/c (“spacing-to-chord” ratio or “pitch-to-chord” ratio) were presented graphically as shown in Fig.
7.14 (from Saravanamuttoo et al.). This figure can therefore be used to choose solidity (as an
alternative to the Zweifel criterion).

Unfortunately, Zweifel and Ainley & Mathieson expressed

“solidity” differently: cx/s versus c/s. This makes it difficult to compare
the geometries that would be obtained using each approach, for the same
set of velocity triangles. The two ratios are related through the stagger
angle, ζ, of the blade row (see the figure on the previous page),
since cos  cx c . However, the value of the stagger angle is not
fixed by the inlet and outlet flow (or metal) angles. This is illustrated in
the
figure at the right, which shows two actual, very highly-loaded
(Z = 1.37) low pressure turbine blade rows that were designed for
identical inlet and outlet flow angles (α1 = 35o, α2 = 60o). The two blades
clearly have very different stagger angles. This is the result of different
decisions regarding the detailed pressure distributions around the blades.
The blade with the high stagger angle was designed to be “forward-
loaded”: that is, to develop most of its lift on the forward part of the
airfoil. The one with the lower stagger angle is much more “aft-loaded”.
The two airfoils have identical values of cx/s, and thus have the same
values of Z. However, they clearly have different values of s/c and
therefore cannot both have the “optimum” geometry according to Fig.
7.14. Despite these difficulties, it is possible to make an approximate
Z = 1.37
comparison between the results obtained by the two different approaches
to choosing the blade spacing. Kacker & Okapuu (KO; see Appendix E)
provided a correlation that gives the typical values of stagger angle that would be seen for different
combinations of α1 and α2:
 Using values of stagger angle obtained from K-O Fig. 5, the following figure shows the values of the
Zweifel coefficient for selected combinations of inlet and outlet flow angles. It is evident that the optimum
geometries based on the Ainley & Mathieson correlations lead to higher values of Z than Zweifel originally
recommended. Very high values of Z are obtained for impulse blades (α1 = α2). Since the Ainley &
Mathieson loss system was specifically based on loss measurements made for impulse blades, these results
suggest that relatively higher values of Zweifel coefficient can be tolerated in the rotor blades for stages with
low values of degree of reaction, especially if the total flow turning is low.

Z = 1.028

Z = 1.034

Z = 0.911

Z = 1.56

Z = 1.26
Z = 0.990
Z = 0.909
5.3.3 Losses

In both North America and Europe, most loss estimates for axial-flow turbines are based on a loss
system developed by Ainley & Mathieson (AM) in the UK in the early 1950s (ARC R&M 2974, 1957; see
also Saravanamuttoo et al.). The AM system has been updated a couple of times to reflect improvements in
blade design: for the design-point conditions, this was done most recently by Kacker & Okapuu (KO) of Pratt
& Whitney Canada (Kacker, S.C. and Okapuu, U., “A Mean Line Prediction Method for Axial Flow Turbine
Efficiency”, ASME J. Eng. for Power, Vol. 104, January 1982, pp. 111-119).

The KO system will be summarized here. The figures from the paper have also been fitted to curves
or surfaces and these fits are given in Appendix E.

For turbines, the total-pressure loss coefficient Y is defined as

Y P0,loss (1)
 P02  P2

Note that in this case, the loss is non-dimensionalized by the outlet dynamic pressure, whereas the inlet value
is used in the loss coefficients for axial compressors.

As for compressors, losses are again divided into components and these are then added linearly to
obtain the total losses:

YTotal  YP f Re  YS  YTET  YTC (2)

where the subscripts designate the components as follows: P = profile, S = secondary, TET = trailing-
edge thickness, TC = tip clearance. f(Re) represents a correction for the effects of Reynolds number on the
profile losses. The effect of Reynolds number on the other loss components is not well documented but it is
believed to be small.

The following figure shows the blade nomenclature used in the KO system. Note that they do not
follow the sign convention we defined earlier. Using that convention, the inlet and outlet flow and metal
angles will often have opposite signs because of the high turning that is normally present in turbine blade
rows. It becomes a nuisance to keep track of the signs and therefore it is common practice by turbine
designers to take both the inlet and outlet angles as positive, as shown in KO Fig. 3.
Profile Losses:

The profile loss is obtained as the weighted average of the losses for two extreme cases with the same
outlet flow angle: a nozzle blade (maximum blade-passage acceleration) and an impulse blade (zero
acceleration). In the original AM system, the expression took the form:

2 1
  2
YP , AM  P,nozzle  

  1  YP ,nozzle
P ,impulse
 Y   t max c

(3)
Y   2  0.2 

 of the blade. Note that KO use α for "air" angles, β for "blade" angles.
where tmax is the maximum thickness
The two reference loss coefficients were presented graphically by AM, as shown in Fig. 1 (for nozzles) and
Fig. 2 (for impulse blades). Note also that for a given value of the outlet angle α2 there is a value of solidity
σ = c/s that minimizes the profile losses. This was the origin of the "optimum σ" that is plotted on Fig. 7.14 in
Section 5.3.2.2.

Kacker & Okapuu compared the AM predictions of profile losses with those obtained from turbine
airfoils of more recent design. They concluded that the AM loss systems significantly over-estimates the
losses for modern turbine blades. The KO profile loss correlation therefore takes the form

YP, KO  2 P,  (4)
3 AM

where the factor of 0.914 was introduced tocorrect the AM loss estimate to that for zero trailing-edge
0.914Y
thickness (since KO handle trailing-edge losses separately) and the factor of 2/3 reflects the improvements in
profile design since Ainley & Mathiesons’ time.
 As seen, Eqn. (3) includes a correction for the maximum thickness of the airfoil: the data in Figs. 1
and 2 apply for a maximum thickness of 20% of the chord length. Decisions about the maximum airfoil
thicknesses would not normally be made at the stage of a meanline analysis for the blade row. However, KO
examined the range of maximum thicknesses observed for a number of recent actual designs and provided the
correlation shown in Fig. 4. Knowing the flow turning from the velocity triangles, this figure can then be used
to obtain a reasonable value for the thickness, ahead of the detailed design of the blade.

The estimates obtained from the correlations described above apply for low speed flows. The turbines
in gas turbine engines normally operate under compressible flow conditions. The Mach number levels
encountered depend to some degree on where the turbine is located in the engine:
High Pressure Turbine (HPT). The HPT is located immediately downstream of the
combustor and drive the high pressure compressor. To minimize the number of stages, HPTs are
typically designed to operate at transonic outlet flow conditions.
Low Pressure Turbine (LPT). The LPT drives the low pressure compressor, and the fan
stage in a turbofan engine. The fan has a large tip diameter and to keep the tip Mach numbers
acceptable, the fan shaft must rotate at a much lower speed than the high-pressure spool. The tip
diameter of the LPT is much smaller than that of the fan and as a result it runs at a relatively low blade
speed. This in turn results in lower flow velocities generally. It is therefore normal for the flow
around LPT airfoils to be subsonic everywhere.
As a result of these differences, the strongest effects of compressiblity are normally seen in HPTs.
The following Schlieren photos (taken from E. Detemple-Laake, “Measurement of the Flow Field in the
Blade Passage and Side Wall Region of a Plane Turbine Cascade,” AGARD-CP_469, 1989) show the flow
through an HPT blade passage with exit Mach numbers of 0.9 (left) and 1.25 (right):
 The profile losses can be affected by compressibility effects in at least two ways:

(i)Inlet Shock Losses. The high levels of curvature around the leading edges of turbine blades result
in high local velocities in this region. For inlet relative Mach numbers as low as 0.6, patches of
supersonic flow, terminating in a shock, can appear on the suction side of the airfoil.
(ii)Channel Acceleration and Outlet Shocks. A turbine blade passage is normally an accelerating
flow channel. As the outlet Mach number increases, there is a tendency for the blade surface
boundary layers to be thinned and their contribution to the losses actually decreases slightly. As the
outlet Mach number approaches 1.0, patches of supersonic flow, terminating in shocks, may begin to
appear on the aft suction surface. Finally, as the outlet Mach number becomes supersonic, expansion
waves and shocks appear in the trailing edge region. In addition to directly contributing additional
total pressure losses, it is common for one or more of the shocks to impinge on the surface of the
adjacent blade. This can cause boundary layer separation, which would further increase the losses.
This effect can be seen from the following figure, which shows Detemple-Laake’s cascade operating
at an outlet Mach number of 1.30.

The following figure shows the relative profile losses as a function of exit Mach number for another
HPT cascade (from Mee et al., “An Examination of the Contributions to Loss on a Transonic Turbine Blade in
Cascade,” ASME J. Turbomachinery, Vol. 114, January 1992, pp. 155-162).
 The complexity of the compressibility effects makes it difficult to predict their influence on
the losses. Kacker & Okapuu provide procedures for estimating the contributions to the profile losses;
see the paper for details.

Finally, KO give the following Reynolds number corrections for profile losses:

0.4
f Re  Re
c 
 for Rec  2 
  2  105  105
  for 2  105  Re 
106
1.0 c
0.2
 Re c for Rec  106
 
 106 

where the Reynolds number is based on the chord length and exit velocity.

Secondary Losses:

As in Howell’s correlations for compressors, the AM/KO loss systems indicate that the
secondary losses in axial turbines are a function of C 2:
2 2
YS  0.04 f  ARcos 2C L cos 2
(5)
 
 cos  1  s c  cos 
3 m

where
CL
 2tan  1  tan  2  cosm
sc

 m  tan 1 21 tan 2  tan 1


and as before, all angles are takenas positive.

The loss coefficients give the total-pressure losses as averaged over the total mass flow rate
through the blade passage. As the aspect ratio of the blade becomes larger, a smaller fraction of the span
is occupied by the secondary flow and the loss associated with it becomes averaged over an increasingly
larger mass flow rate. Consequently, the mass-averaged loss coefficient varies inversely with the aspect
ratio. This effect is embodied in the aspect ratio correction, f(AR) in Eqn. (5). Kacker & Okapuu found
that the AM loss system tended to over-estimate the effect of aspect ratio on blades of very low aspect
ratio (which are often used in modern HPTs). In the KO loss system, the correction for blade aspect
ratio therefore takes the following form:

1  0.25 2  h c
f  AR for h c
hc 2

(6)
1
 for h c
hc 2
 2.5

Aspect Ratio Correction, f(AR)

Ainley & Mathieson
1.5

Kacker & Okapuu (Eqn. (6))

0.5

0 0.5 1 1.5 2 2.5 3

0
Aspect Ratio, h/c

Kacker & Okapuu also provide a compressibility correction for the secondary losses (see the
paper).

Trailing-Edge Losses:

Due to the finite thickness of the trailing edge, the streamtube experiences a sudden increase in
area as it leaves the blade passage. The resulting sudden-expansion loss is correlated in terms of an
alternative form of loss coefficient, known as an energy loss coefficient, Δφ2, as a function of the ratio
of the trailing-edge thickess to the throat opening. KO correlated the values for nozzle blades and
impulse blades separately, as shown in Fig. 14.

In the same way as for the profile

losses, the
trailing-edge loss for an arbitrary blade is expressed as the weighted average of the values for nozzle
and impulse blades:


2
 TET
2
 TET    1  2
  2 TET   (7)
2 1   TET  1 2  1

 2 

0 0
The energy loss coefficient is then converted to the usual total pressure loss coefficient as
follows:


   1 M 2  1   1
1  2
2
  1  1
 TET 
YTET 1   2

 

   1 M 2   1
1  1 
 2 2


and for incompressible flow this reduces to

1
YTET 1
  TET
2

Tip-Clearance Loss:

For unshrouded blades, KO express the effects of tip-clearance losses as a correction to

the efficiency:

 k (8)
0  0.93
RTip
cosk 2is R
where η0 is the efficiency for zero tip clearanceh and the
Meantip clearance. Note that Eqn. (8) indicates
that a 1% increase in tip clearance, relative to blade span, will result in a 1% reduction in efficiency.
This is considerably lower sensitivity than the 3% reduction that is predicted by Howell’s correlation for
axial compressors. As seen, KO also found the loss to be a function of the hub-to-tip ratio of the blade,
since
21 1  HTR 2  , where HTR = R /R .
Mean Tip Hub Tip
R R

Low-pressure turbine blades are often shrouded to reduce the tip-leakage flow and losses.
KO recommend the following expression to estimate the tip-leakage losses for a shrouded rotor blade
row:
2 2
 C L  cos
0.78
YTC  c  k   2 (9)
 
0.37 h c   s c  cos3  m

where kN is the effective tip clearance

and
k
k
 Number of seals
0.42

To illustrate the relative magnitudes of the various components of loss, the predicted loss
components for two different turbine stages, one subsonic and one transonic, will be quoted (taken from
Moustapha et. al., Axial and Radial Turbines, Concepts NREC, 2003, pp. 89-90). The table summarizes
the design parameters for the two stages:
 Subsonic Turbine Transonic Turbine

Pressure Ratio 1.97 3.76

Work Coefficient, ψ 1.31 2.47

Flow Coefficient, φ 0.47 0.64

Reaction, Λ (%) 50 30

Stage Efficiency (%) 88 83.5

Stators Rotor Stators Rotor

Exit Mach Number 0.67 0.82 1.1 1.14

Total Flow Turning (o) 60 78 76 124

Blade Aspect Ratio, h/c 0.71 1.25 0.70 1.44

Tip Clearance, k/h (%) 1.5 1.5

Zweifel Coefficient 0.74 0.88 0.84 0.76

The figures show the resulting values of the loss coefficients:

Subsonic Turbine Transonic Turbine

0.4 0.4
0.35 Stators 0.35
Stators
0.3 Rotor
Loss Coefficient, Y
Loss Coefficient, Y

0.3 Rotor
0.25
0.25
0.2
0.2
0.15
0.15
0.1
0.1
0.05
0.05
0
Trailing Secondary Tip
0
Profile Total
Edge Clearance Profile Trailing Secondary Tip
Loss Component
Edge
Total Clearance
Loss Component
PW100 Turboprop
 100 2

95 1.8

90 1.6

85 , (Rotor Only) 1.4

Tip Mach Number

, Efficiency (%)

Diffuser
80 Loss 1.2

75 1
 (Rotor + Diffuser)
70 0.8

65 0.6

60 0.4

55 0.2

0
50 5 10 15
0
Compressor Pressure Ratio

Data for rotor only: Senoo, Y., Hayami, H., Kinoshita, Y. and Yamasaki, H., "Experimental Study on Flow in a Supersonic Centrifugal
Impeller," ASME J. Eng. for Power, Vol. 101, Jan. 1979, pp. 32-41.
Data for rotor with PWC pipe diffuser: Kenny, D.P., "A Comparison of the Predicted and Measured Performance of High Pressure Ratio
Centrifugal Compressor Diffusers," ASME Paper 72-GT-54, 1972.
Influence of diffuser design and diffuser pinch on pressure ratio, surge line and choking mass flow rate: Japikse, D. “Decisive Factors
in Advanced Centrifugal Compressor Design and Development,” I.MechE, Orlando, FL, November 2000.
6.2 IDEALIZED STAGE CHARACTERISTICS

Consider the outlet flow from a centrifugal rotor with backswept vanes. Assume that there is no swirl
in the rotor inlet flow and that the fluid is incompressible. Assume also the Euler Approximation so that the
flow leaves the rotor parallel to the metal angle at the vane trailing edge.

U2

W2
C2 2 (+) '2 (+)
Cr2

Cw2

From the Euler equation

h0  gHE  U 2 Cw2  U1Cw1

 U 2 Cw2

and tan  2  C w 2  U 2
Cr 2

or
Cw2  U 2  Cr 2 tan
2
Q
 U2  A tan 2
2

Then 
2
Q
gH E  U 2  U 2 tan 2
A2

or, dividing by N D
2 2

gH E K Q
 K1 2
(1)
2
N D 2 ND 3

where U 22 2
⎛  ⎞|  const.
K1 N 2 D 2 |
 ⎝ 60 ⎠
and

U D
K 2  NA2 tan 2
2

Equation (1) is the equation for the idealized head rise versus flow rate characteristic. Within the Euler
Approximation (β‘2 = β2), the slope of the characteristic, K2, is constant and has the same sign as β2, as shown
in the sketch:

Note that whereas the slope of the ∆H vs Q characteristic for axial machines was always negative
(assuming † < 1.0, which experience has shown is necessary), radial machines can have charateristics with
either positive of negative slopes, depending on the geometry of the vanes at the outlet:

Forward-swept vanes:
• Highest head rise.
• dHE/dQ > 0 is destabilizing, but losses can provide some stable operating range (see later
section).
• Very high C2: puts heavy demands on diffuser to recover pressure.
• Suitable where want to maximize head rise, efficiency is not a serious concern and surge is
not a problem.

Radial vanes:
• Simplest to manufacture.
• No bending stresses in vanes due to centrifugal effects (were therefore favoured in early gas
turbine engine applications of centrifugal compressors).

Backward-swept vanes:
• Lower head rise.
• Wide stable operating range (because dHE/dQ < 0).
• Lower C2: reduces diffuser losses.

Recall from Section 3.4 that there will be a pressure rise through a radial rotor due to centrifugal
compression, even with no flow. From Eqn. (1), for Q = 0, ∆HE = U22/g (or ∆h0 = U22). The head rise at zero
flow is known as the "shut-off" head.
3. EMPIRICAL PERFORMANCE PREDICTIONS

1. Rotor Speed and Tip Diameter

The rotor speed and size can be estimated from correlations using two different approaches: from
specific speed and specific diameter; or from the flow coefficient and work coefficient.

Specific Speed:

In Chapter 2, we have already used specific speed as a basis for selecting the type of turbomachine
that is suitable for a particular application. The following figures (from C. Rodgers, “Specific Speed and
Efficiency of Centrifugal Impellers” in "Performance Prediction of Centrifugal Pumps and Compressors" ed.
S. Gopalakrishnan et al., ASME International Gas Turbine Conference, New Orleans, March 1980,
pp.
191-200.) show more specific data for unshrouded centrifugal rotors, including the effect of several
geometric and aerodynamic parameters.
 The definition of specific speed used here is based average density (or average volume flow rate):

1
⎛  Q2 ⎞2
| 1
|
⎝
Q 2
NS 3
 gH⎠  4

Compare this with the usual non-dimensional specific speed:

Q12
 3
 gH 4

Note that for a typical case, fi will be slightly larger than N S .

If there are no constraints on the rotational speed, then one would normally choose the value of
N S that gives the highest ( N S  0.6  0.8 ). With N S chosen and fi estimated, the Cordier diagram can
be used to choose the diameter.

Work Coefficient and Flow Coefficient:

Aungier (R.H. Aungier, Centrifugal Compressors - A Strategy for Aerodynamic Design and Analysis,
ASME Press, New York, 2000) presents a convenient correlation of work coefficient versus flow coefficient
for industrial compressors of several configurations: with shrouded and unshrouded impellers; and with
vaneless and vaned diffusers. The correlations are based on results for compressors with pressure ratios up to
about 3.5, but can probably be extrapolated to somewhat higher values. Aungier defines the non-dimensional
parameters as follows:

m
Flow Coefficient:  Work Coefficient: P 
h0ref
  1 ˙r 2
2 U 2 U 22

where ∆h0ref is the total enthalpy rise for the reversible process with the same pressure ratio. P is the stage
polytropic efficiency. Aungier’s correlations are presented in the following two figures:
The figures are applied as follows:

(a) Select ф to give the best stage polytropic efficiency, P , and read the corresponding work
coefficient, µP. From

P h0ref

U 22

calculate U2.

(b) Then from the chosen ф

m

 1˙r2 U 2
2

calculate r2.

With the tip radius and tip blade speed defined, the rotational speed is known. If the rotational speed
is constrained (eg. driving motors are only available for certain speeds) then Fig. 6-1 or Fig. 6-2 can be used to
select a compromise size and rotational speed that minimizes the impact on the stage efficiency.
6.3.2 Rotor Inlet Geometry

From the Euler equation

h0  gHE  U 2 Cw2  U1Cw1

If there are no IGVs, Cw1 = 0 and the work transfer depends entirely on the rotor tip or outlet conditions. For
good efficiency, the impeller inlet must nevertheless be well designed (eg. the inducer inlet metal angle must
be matched to the inlet relative flow vector) and correctly sized.

Consider three rotors designed for the same m˙ , U2 and with the same outlet geometry so that all
three give the same Δh0. The critical region for frictional losses (which vary as V 2), cavitation and
compressibility effects is at the vane tip at the inlet, since that is where the relative velocity is the highest and
static pressure the lowest. The drawing shows the resulting inlet tip velocity triangles for three different inlet
sizes:
LARGE EYE SMALL EYE
HIGH U1t LOW U1t
LOW C1t HIGH C1t

r2
r1t

r1h

C1t C1t C1t

U1t
W 1t U1t W1t
W 1t U 1t

To allow room for a shaft, or for a nut to hold the rotor to the end of the shaft, typically r1h = 0.2r2 to 0.35r2.
For a given r1h, it is evident from the inlet velocity triangles that there is an optimum r1t that minimizes the
inlet relative velocity and Mach number:

OPTIMUM
W1t

M1t

r1t
6.3.3 Rotor Outlet Width

Consider the effect on the outlet velocity triangles of varying the rotor outlet width (or outlet vane
height) b2. The outlet metal angle is adjusted to maintain constant Cw2 and thus give the same pressure rise.
From continuity

m˙  2 Cr 2 A2  2 Cr 2 2r2 b2 

and thus for a fixed m˙ , the choice of b2 determines the radial component of velocity at the
rotor outlet:

Cw 2

U2
SMALL b 2
C2 W2
Cr 2
2
LARGE b2

C2 W2

b2
r2

Summarizing the effect of different choices of b2:

LARGE b2 SMALL b2

Lower Higher
C2 (Good) (Bad - Larger diffusion
required downstream)

W2 Lower Higher

Lower Higher
W2/W1 (Bad - Larger diffusion (Good)
required in rotor passage)

β2 Higher Lower

The value of b2 would thus be chosen to obtain a compromise between high diffusion inside the rotor
passage and high diffusion in the downstream diffuser (which serves the same function as the stators in an
axial compressor stage).

Note that W2/W1 is again the de Haller number. Various papers and textbooks provide
guidelines for
choosing the de Haller number for centrifugal fan, compressor, and pump rotor passages:

(1) Aungier (2000) Recommended: W2/W1 > 0.75

Never exceed: W2/W1 < 0.65

(2) Wilson & Korakianitis (1998) Recommended: W2/W1 > 0.8

(3) Rodgers (1978) Recommended: W2/W1 > 0.71

(4) Yoshinaga (PWC document, 1982) Low PR compressors and fans: W2/W1 > 0.8
High PR compressors (up to 8.0) W2/W1 > 0.6

where W1 = value of relative velocity at inlet mean radius.

6.3.4 Rotor Outlet Metal Angle - Slip

From Section 6.3.3, the required outlet flow angle β2 was seen to be related to the choice of b2. The
corresponding metal angle β2‘depends on the deviation, which is called “slip” in centrifugal machines. The
slip in turn depends on the rotor “solidity”: that is, the number of vanes, Z. Thus, the choices for β 2 ‘ and Z are
inter-related.

Consider a backswept rotor:

Because of slip, the rotor imparts less swirl to the flow than for the “ideal” case, for which β2 =β2‘ (that is, the
Euler approximation is taken to hold in the ideal case). Since Cw2 < Cw2‘, the ∆h0 is reduced by this effect. We
then define the slip factor σ as

Cw2
  
C w2
where σ ≤ 1.0.

A number of correlations have been proposed for σ. The one due to Stodola has been widely used:

 ⎛ 
cos ⎝| 2
  1 Z ⎠
⎞| ⎛
1  2 tan|⎝ 2 |⎞
⎠
where ф2 = Cr2/U2 and β 2 ‘ is the backsweep or forwardsweep angle (taken as positive in both cases). Stanitz
suggested a slightly simpler form:


0.63
  1 Z

1 2 tan 2

Wiesner (F.J. Wiesner, “A Review of Slip Factors for Centrifugal Impellers,” ASME Trans., J. Eng.
for Power, October 1967, pp. 558-572) reviewed the available slip factor correlations and pointed out that the
Stodola, Stanitz and similar correlations are only valid for impellers with long blades. Wiesner recommended
the Busemann correlation which takes into account the influence of r1/r2 and provided the following curve fit:

Letting e = r1/r2, and identifying a limiting value of e given by

1
 lim it
| 8.16cos  2 ⎞
 ⎛
| Z |
| ⎠
e⎝

then for e ≤ elimit (ie. longer vanes)


  1 sin  2
0.7
Z

and for e > elimit (ie. shorter vanes)

(
| cos 2  |⎞| 1 |(    limit ⎞| |
3
  1 |
| (Z
0.7
||
lim it
⎝ 1  ⎠ ⎞
⎝ ⎠
⎝ ⎠

The figure shows the predicted variation with Z and e

for an example backsweep angle of 45o (taken from
Aungier, 2000), who provides an equivalent but
slightly different curve fit:
6.3.5 Choice of Number of Vanes - Vane Loading

Wilson & Korakianitis (Design of High-Efficiency Turbomachinery and Gas Turbines, 2nd ed., Prentice-
Hall, 1998) provide a broad guideline for selecting the
number of blades, as function of the vane angle at the tip, as
shown in the figure at right.

More recently, Rodgers (2000) presented a

correlation for the number of vanes which, according to his
loss estimates, gives the best rotor efficiency:

25cos  2
Z  

where Ω is the usual non-dimensional specific speed.

Comparison with the Wilson & Korakianitis figure suggests
the Rodgers’ correlation is very conservative, leading to very
large numbers of vanes.

Aungier (2000) outlines a method of selecting the

number of vanes based directly on the vane loading. He
suggests the following limit: CENTRIFUGAL COMPRESSOR - NUMBER OF VANES

2W 45


W2  W1 0.9
40

where ΔW is the maximum relative velocity difference across

the vane. ΔW can be estimated from 35

2 D2 U 2  30
W 
Z LB
Number of Vanes, Z

25

where ψ = work coefficient = Δh0/U22 and L B is the length of the 20

vane along the mean camber line. A reasonable initial estimate
of LB can be obtained from
15

 b  1  D  D 1 10

L B  z I  2    2
 2  2  cos 2 
5


where ΔzI is the axial length of the rotor.
0
0 10 20 30 40 50 60 70
Backsweep Angle (Deg.)
Rodgers, Ns = 0.6
Rodgers, Ns = 0.7
Rodgers, Ns = 0.8
Wilson Zmax
Wilson Zmin
6.3.6 Losses

The actual stage characteristics are different from ideal due to slip and losses. Slip reduces output but
does not affect efficiency since the required input power is reduced along with the output.

Sources of losses:

(1) Disc friction: - friction on outer surface of impeller

-since this torque is not exerted on the through-flowing fluid, it does not appear in
the Euler work, pgQ∆HE
(2) Leakage: - fluid leaks through the tip gap leading to losses as in axial machines
-if the rotor is shrouded, compressed fluid can leak through the clearance back to the
inlet, to be recompressed over and over again
- thus, more fluid is compressed than is delivered by the machine, increasing the
power required and showing up as an apparent loss
(3) Inlet: -at other than design Q, flow angle and metal angle will be mismatched at the
leading edge, resulting in separation and additional losses
- a simple, inexpensive machine with no inducer will have significant inlet
losses at
(4) Impeller: all operating conditions
- frictional and separation losses inside the impeller channels
(5) Diffuser/Volute: - roughly a Q2
- frictional and separation losses roughly a Q2
- for vaned or pipe diffuser, additional leading-edge losses when Q Qdesign
(like (3))
- for volute,
The figure shows the approximate sudden
trend of theexpansion losses due
loss components withabrupt change in area
flow rate:

The next figures show the resulting stage characteristics, taking into account slip and losses, for
backward-swept and forward-swept vanes:
(i) Backward-swept vanes: (ii) Forward-swept vanes:

Note that due to the effects of the losses the machine with forward-swept vanes also has some stable
operating range (dHE/dQ < 0.0), although it tends to be narrower and does not include the design point.

Taking into account the losses, the required shaft power is

W˙  gQHE (th)  gQl HE (th)  Disc & Bearing Friction Power

where ∆HE(th) = theoretical Euler head (Euler head with slip but no losses)
Ql = leakage flow (volume flow which leaks from outlet back to inlet, to be
recompressed) for a shrouded rotor

The actual head delivered is

H  H E (th)  HL

where ∆HL = sum of losses (3) + (4) +

(5) and the corresponding efficiency is

overall gQH

W shaft
˙
As usual, for compressible flow substitute m˙ for pQ and ∆h0 for
g∆H.
 CHAPTER 7
Static and Dynamic Stability of Compression
Systems

7.1 INTRODUCTION The same argument can be made for points B and D.
Thus, operating points A, B and D are statically stable
It was mentioned in Chapter 4 that surge is very operating points.
dangerous to axial compressors. While centrifugal
compressors are more rugged than axial machines, surge is Point C is different. If m˙ is disturbed to a larger
still dangerous and should be avoided.
value, the machine delivers more )P 0 than the load
It was also noted that surge is a dynamic instability requires at the new m˙ . The flow in the load will
which depends on not just the characteristics of the therefore increase even further and the operating point
compressor but also on the aerodynamic moves further from the equilibrium point. Thus C is a
characteristics of the other components to which it is statically unstable operating point.
connected. It is possible to develop a simple lumped- Static stability does not guarantee that the system will
parameter analysis for a compression system. Such an finally settle at the original equilibrium operating point,
analysis can provide useful insights into which only that tend to move back towards the equilibrium point.
characteristics of the system encourage or delay the onset The system may overshoot and oscillate about the operating
of surge. For further information see Stenning (1980), point. If it eventually settles at the original operating
Greitzer (1980, 1981) and Cumpsty (1989). point, the system is dynamically stable.

7.2 STATIC STABILITY

7.3 DYNAMIC STABILITY - SURGE
A system is statically stable if, when it is disturbed by a
small amount from its equilibrium operating point, a A simple analysis can be developed to predict
reaction arises which tends to restore it to the equilibrium approximately the dynamic stability characteristics of a
condition. Static stability is normally a necessary, but compression system. A compressible flow system will be
not sufficient condition for dynamic stability examined. Only minor modifications are needed to make it
apply to an incompressible flow system.
Consider the compressor characteristic shown in Fig.
7.1. Points A - D are all equilibrium operating points Fig. 7.2 shows schematically a simple system consisting
()P0,load = of four components:
)P0,machine at the given m˙ ). Consider point A and (1) A compressor
suppose that a small disturbance causes an increase in m˙ : (2) A duct
(i) The machine delivers less )P 0 than required by the load (3) A plenum, in which mass can be stored.
at this m˙ . (4) A throttle, represented by a valve, which provides
(ii)The flow rate in the load must therefore decrease, the main pressure loss in the system. To a first
causing the system to move back towards A. approximation, the throttle could also represent the
turbine in a gas turbine engine.

Fig. 7.2 Four-component compression system.

Fig. 7.1 Compressor operating points.

1
The flow through the components is treated as one-
dimensional. Thus, the flow at any point is characterized dC 
dP2  P01  (2)
by a single value of P, T, C, m˙ etc. (if necessary, these 1 c 1
would be interpreted as the local average values). The dm dm
analysis will consider perturbations about an equilibrium
operating point and the perturbations will be assumed to be and since we are assuming that the perturbations from the
small. operating point are small, we can assume that c is constant
our analysis.
in That is, we linearize the compressor
The instantaneous value of any flow quantity is
characteristic at the operating point of interest.
represented by the sum of the mean value plus the
instantaneous (small) perturbation:
If we assume that P01 is constant (ie. that the compressor
draws fluid from a large, constant pressure reservoir) then
m2  m2  m2 P3  P3  P3  from (2)
where m2 isetc.
the mass flow rate at plane 2 (the dot is omitted
for convenience). The goal of the analysis is to determine
the behaviour of the perturbations over time after some dP2 dP01 dP2 (3)
  c
initial disturbance has occurred. If m2N, P3N etc. eventually
decrease to zero, the system is dynamically stable at the dm1 dm1
operating point in question. and integrating (3) for a small deviation
dm1 away from the
equilibrium point
The approach used is known as the lumped-parameter
method: equations for the behaviour of each component are P2 m1
developed separately and they are then linked by the flow
conditions at the interfaces between the components. dP 2 
cdm 1

Consider each component in turn: P2 m1

(1) Compressor

P2  P2  c m1  m1 
The pressure rise across the across the compressor,
represented by P2 - P01, is a function of the inlet mass flow:
From the definition of the perturbations, this can be written
(1)
P2  P01  Cm1 P2  cm1 (4)

where C is the function which defines the compressor This is then the perturbation equation for the compressor.
characteristic (see Fig. 7.3). The gradient at any operating
point along the characteristic is If we assume that the internal volume is small, so that
essentially no mass can be stored in the compressor, then
m1
= m2 at all times and an alternative to (4) is

P2   (5)
cm2

(2) Duct

We assume that the losses in the system occur primarily

in the throttle so that we can neglect the frictional losses in
the duct. We also neglect the volume of the duct relative to
the volume of the plenum. Therefore, the duct introduces
only inertia: a pressure difference is present between
stations 2 and 3 only when the fluid in the duct is being
accelerated or decelerated.

The equation governing the behaviour of the duct can

be obtained either by performing a force balance on the
Fig. 7.3 Compressor characteristics. free body consisting of the cylinder of fluid in the duct
or by

2
applying the unsteady momentum equation to a control
volume occupying the duct. For both analyses, we will L dm
2
neglect density changes along the duct. P2  P3 (7)
 A dt
(i) Force balance:
(3) Plenum
d The plenum can be a mass storage component.
 Fx 
dt Applying conservation of mass to the plenum:
mu d
P2 A  P3 A dt  d (8)
m2  m3  V dt 3
d
ALu  Ldt 
where m3 = mass flow rate through the valve. Changes in
Au
and DAu = m2 is the instantaneous mass flow rate at all the mass in the plenum will be reflected in the density of
points in the duct (since density changes are neglected), so the stored gas. In a pump system, a reservoir with a free
that surface or a surge tank would similarly act as a mass
L dm2 storage component.
P2  P 3 (6)
A dt
If the compression or expansion process is isentropic,
then
(ii) Control volume analysis:
P
For the control volume in the duct  const.

d
Fx  dt u  dV
V
muout muin Differentiating with respect to time and assuming a perfect
gas

Since the density is constant along the duct, the

instantaneous inflows and outflows of momentum must be d3 
 3
identical, and only the first term, the momentum dt P3 
accumulation term, remains on the right-hand side: dP3
dt
1

 RT3 dt
L dP3
d
u  Adx dt  m  12 dP3
L  dm2
d
Fx  dt  dx 
0 
2
 dt a3 dt
L 

 0 where a3 = speed of sound at the plenum conditions. Then

After substituting for EFx in terms of the inlet and outlet from (8)
pressures, (6) is again obtained.
V dP3
m2  m3 
We then substitute into (6) in terms of the perturbations a 32 dt

Substituting in terms of the perturbation quantities, the

P 2  P2    P 3 

L d m2 
perturbation equation for the plenum is obtained:

V (9)
m2  m3 
and since P 
3
there 2
 no losses in themduct,
are
A 
the mean inlet and
dP3 a 32 dt

outlet pressures must be the same. Thus,dtthe perturbation

equation for the duct becomes (4) Throttle

The throttle is handled in exactly the same way as the

compressor: the load line is linearized at the equilibrium
operating point. If the valve is choked, the mass flow rate

3
through it is a function of only the upstream pressure, P3. If
it is not choked, the pressure downstream is assumed to be dP3 dP2 Ld
constant. Then the perturbation equation for the throttle  
2
becomes: m2

P3  f m3 and from (11) dt dt A dt 2

(10)
dP2
c
where f is the local slope of the load line (note that f will dm2 dt
always be positive).
Characteristic Equation for the System Thus
dt
Summarizing, there are four perturbation equations for
the components in the system: dP dm L d
2 3 c 2 2
m
dt dt 
A dt 2
P2  (11)
or
cm2
L (12)
P2  P3 A dt
dm2 V dP3 f V dP3 c f V dm2 f V d (18)
m2  m3 2a 2
dt  32 dt  32 dt 2
(13) m32 a a
 a23 dt
P3  f (14)
Substituting (18) into (17) and rearranging
m3
These are four equations in the four unknowns m2N, m3N,
P2N and P3N. Solving for any one of the unknowns from f V L d 2m  L c f V  dm
2
(11) - A dt 2    2
  f  cm2  0
a 23  A a32 
(14) leads to a second-order ordinary differential equation
for dt (19)
For example,
the variation solving
in time m2N, substitute (13) and
forunknown.
for that
(14) into (11): This is seen to be a second-order ordinary differential
equation in m2N.

L (15) It can be shown that the corresponding equation for any

cm2  f m3 A dt
of the other three perturbations would have the save
dm2 coefficients as (19).

Multiply (13) by f (noting that f is non-zero and always Within the assumptions of the analysis, the coefficients
positive), are constant and, given initial conditions for m2N and
dm2N/dt,
V dP3 (19) can readily be solved to determine the response of
f m2  f m3  (16) the
f a 23 dt system. As noted earlier, if m2N tends to 0 with
increasing time,
A useful the system
analogy can isbedynamically stable.the present
drawn between
system and a mass-spring-damper system for which the
Then subtract (15) from (16) to eliminate governing equation is (for free vibrations)
m3N
2
V dP3 L dm m d x  s  kx  0 (20)
f m2  cm2   (17) dx
dt 2 dt
2
f a 23 dt A dt
where s = damping coefficient, k = spring constant. Two
conditions must be met for the system governed by (20) to
Differentiate (12) with respect to time and rearrange to be stable:
obtain an expression for dP3N/dt:
(i) k > 0 - that is, the spring constant must be positive

The equivalent condition in (19) is that f > c, which is

4
 precisely the requirement for static stability which we It is not immediately clear whether it requires a large
arrived at with qualitative arguments in Section 7.2. positive value of c (large in comparison to f, for example)
to destabilize the system, but note that as L tends to zero c
(ii) s > 0 - that is, the damping must be positive also tends to zero. Therefore, a compressor which is
connected to a plenum by a very short length of duct will
This is evident from the solution to the equation: the become unstable essentially at the peak of the compressor
system has a critical value of the damping coefficient, characteristic. That is why the latter is often used as a
sc, given by criterion for predicting surge. In general, we would
expect to encounter the condition for dynamic
instability near the peak of the characteristic and
sc  2 k m probably long before we reach the condition for
static instability (operating point C on the

If s < sc, the system is under-damped and the solution original )P 0 versus m˙ diagram) .
takes the form
The relationship between stall and surge now is a little
clearer. For a typical compressor characteristic, as the flow

s k  2 rate through the machine is reduced the output peaks
m t
x  Ae sc
sin 1  s   k
t  
 and eventually begins to reduce. This is generally the
  sc  result of increasingly extensive stall: perhaps an increasing
 m 
number of rotating stall cells and/or cells of increasing
spanwise extent as the flow rate is reduced. Stall thus
prepares the conditions for surge. Note that the appearance
Thus, the system oscillates sinusoidally in time, with of stall is a phenomenon of the compressor itself, not the
the magnitude of the peak displacement being system. On the other hand, surge is an unstable
controlled by the exponential term. condition in compression system in which flow
quantities, including the compressor mass flow and
Since m, k, and sc are all positive, if s > 0 delivery pressure, undergo oscillatory fluctuations which
the exponential term decreases in time, the magnitude grow over time. In systems such as gas turbine engines,
of the fluctuations decays, and the system is seen these fluctuations can reach destructive magnitudes in a
to be dynamically stable. If s > sc, the system is very small number of cycles.
References
over-damped and the solution is no longer
oscillatory but it again includes exponential terms Cumpsty, N.A., 1989, Compressor Aerodynamics, Longman,
which are a function of s. Again, if s is negative the Harlow.
exponential terms grow in time and the system moves
away from the equilibrium point in an unstable way. Greitzer, E.M., 1980, “Review - Axial Compressor Stall
Phenomena,” ASME J. Fluids Engineering, Vol. 102, June
Applying these ideas to the compressor system, it is seen 1980, pp. 134-151.
that there are two contributions to the system damping:
Greitzer, E.M., 1981, “The Stability of Pumping Systems,”
(a)positive (stabilizing) damping is supplied by the ASME J. Fluids Engineering, Vol. 103, June 1981, pp. 193-
inertia of the fluid in the duct (the L/A term), and 242.
(b) potentially negative damping is supplied by the
term Stenning, A.H., 1980, “Rotating Stall and Surge,” ASME J.
involving the slopes of the compressor and throttle Fluids Engineering, Vol. 102, March 1980, pp. 14-20.
characteristics.
Since f is always positive, the sign of the damping term is
controlled by the sign of the slope of the compressor
characteristic, c. If c < 0 (as it normally is at higher flow
rates) strong positive damping will be present and the
system will be stable. The condition for instability is then

L  cfV0 or La 2
c 3
A a 23 AV f

Thus, the system will become unstable for some positive

value of the slope of the compressor characteristic, the
precise magnitude being a function of a number of system
parameters.

5
 APPENDIX A:

Curve and Surface Fits for Howell’s Correlations for Axial Compressor Blades

(a) Design-Point flow Deflection, c* (C,R & S, Fig. 5.14)

c* is a function of inlet flow angle, a2 and s/c (=1/σ):

With: A = 33.5293 B = -0.530812 C = -15.2599

D = 0.00209610 E = -0.677212 F = 0.187148

c*(a2,s/c) = A + Ba2 + C ln(s/c) + Da22 + E(ln(s/c))2 + Fa2ln(s/c)

Applies for: 0 < a2 < 70o, 0.5 < s/c < 1.5 (or 0.666 < σ < 2.0).

(b) Reynolds Number Correction for Design-Point Deflection (Horlock Fig. 3.3)

With A = 0.664154 B = 22.1578 C = 1.03819 D = 4.71864

where Re is the Reynolds number based on inlet velocity and blade chord divided by 105.

(c) Off-Design Deflection (Dixon Fig. 3.17)

where

The curve fit is applicable for -0.8 < irel < 0.8.

(d) Profile Drag Coefficient, CDp (Dixon Fig. 3.17)

For values of irel from -0.7 to 0.3 the profile drag coefficient, CDp, is a function of solidity and irel:

1
CDp1A(irel,s/c) = -0.02842irel(s/c)2 + 0.004381(s/c)3 - 0.00788(s/c)2 - 0.003979(s/c) + 0.07753irel(s/c)

CDp1B(irel,s/c) = -0.01542irel2 (s/c) + 0.02277 - 0.04429irel + 0.05002irel2 + 0.009207irel3

CDp1(irel,s/c) = CDp1A(irel,s/c) + CDp1B(irel,s/c)

This curve fit is applicable for 0.5 < s/c < 1.5 (or 0.666 <
σ < 2.0).

For values of irel greater than 0.3, CDp is a function of the relative incidence only:

CDp2(irel) = 0.01665 - 0.004181irel - 0.01908irel2 + 0.06477irel3 + 0.3949irel4 + 0.3426irel5

2
 APPENDIX B:

C4 Compressor Blade Profiles

Like NACA 4-digit airfoils, the C-series compressor blades are defined by a symmetrical
thickness distribution which is superimposed on a specified mean, or camber, line. As indicated in the
Howell correlations, both circular arc and “parabolic” arc camber lines have been used with C-series
blades.

For the blade with a parabolic arc camber line, the point of maximum camber lies at other
than mid-chord. Typically, the point of maximum camber lies towards to leading edge; that is, a/c <
0.5.

The relationship between the camber angle 2 (= 21 + 22), a/c and b/c
is:

(1)

and

The term parabolic arc camber line is somewhat misleading. The mean line is not defined by
a single parabola, or even by two joined parabolas. For example, to define a polynomial which passes

1
through (0,0) with slope tan21 and through (a,b) with zero slope requires at least a cubic. The
following discussion will consider mainly the circular arc camber line.

Setting a/c = 0.5 in Eqn (1),

(2)

The equations of the camber line and its inclination, Nc, are then

(3)

and

(4)

The co-ordinates of the upper and lower sides of the blade are then

(5)

where yt is the local thickness of the blade. For the C4 profile, the blade thickness distribution is given by

where t is the maximum thickness of the blade as a fraction of the chord length.

The geometry of C-series blade is designated using a shorthand notation. For example, a blade
designated 10C4/30C50 refers to a blade with a C4 profile and: 10% maximum thickness, circular arc
camber, camber angle 30o and maximum camber at 50% chord (the last piece of information is
redundant
2
since circular arc camber has already been specified). The resultant geometry is
shown:

3
 APPENDIX C:

Curve and Surface Fits for NASA Correlations for Axial Compressor Blades

(a) Minimum-Loss Incidence (SP-36 Fig. 137)

The surface fit gives the minimum loss incidence for a blade of zero camber and 10% thickness
as a function of inlet flow angle, $1, and solidity, F:

With: A00 = -0.13571 A01 = 0.075795 A02 = 9.1315x10-4

A10 = 0.015986 A11 = 0.074959 A20 = -2.4954x10-4
i0(10)($1,F) = A00 + A01F + A02F2 + A10$1 + A11$1F + A20$1
2

Valid for: 0.4 < F < 2.0, 0.0 < $1 < 70.0.

(b) Slope Factor, n, for Minimum-Loss Incidence (SP-36 Fig. 138)

With: A00 = -0.066879 A01 = 0.05897 A02 = -0.054019

A03 = 0.033568 A04 = -7.1706x10-3 A10 = -6.0476x10-3
A11 = 7.402x10-3 A12 = -2.5749x10-3 A13 = 2.6067x10-4
A20 = -3.3001x10-5 A21 = -3.084x10-5 A22 = 1.3955x10-5
A30 = 8.0286x10-7 A31 = -1.2016x10-7 A40 = -9.1961x10-9

n1($1,F) = A00 + A01F + A02F2 + A03F3 + A04F4 + A10$1 + A11$1F + A12$1F2 + A13$1F3
n2($1,F) = A20$1 + A21$1 F + A22$1 F + A30$1 + A31$1 F + A40$1
2 2 2 2
3 3
4

n($1,F) = n1($1,F) +

n2($1,F) Valid for: 0.4 < F < 2.0, 0.0 < $1 < 70.0.

(c) Thickness Correction,

(Ki)t, for Minimum-Loss Incidence (SP-36 Fig. 142)

Valid for: 0.0 < t/c < 0.12, probably usable up to t/c = 0.15.

1
(d) Zero Camber Deviation Angle, * 0 (SP-36 Fig. 161)

With: A00 = 0.053535 A01 = -0.29275 A02 = 0.71879

A03 = -0.75902 A04 = 0.3706 A05 = -0.067233
A10 = -3.838x10-3 A11 = 0.02838 A12 = -0.02068
A13 = 3.4149x10-3 A14 = 5.8448x10-4 A20 = 3.5333x10-4
A21 = 2.0917x10-4 A22 = 3.0519x10-4 A23 = -1.2273x10-4
A30 = -1.3124x10-5 A31 = -1.0755x10-5 A32 = 1.7229x10-6
A40 = 2.3356x10-7 A41 = 1.1718x10-7 A50 = -1.4651x10-9

*o1($1,F) = A00 + A01F + A02F2 + A03F3 + A04F4 +A05F5 + A10$1 + A11$1F + A12$1F2 + A13$1F3 + A14$1F4

*o2($1,F) = A20$1 2 + A21 $1 2F + A22 $1 2F2 + A23 $

1 F + A
2 3
30 1$
3
A 1$ 3F + 32A 1 $
+ 31
F
3 2 + A 40$1 + A41$1 F + A50 $1
4 4

*o($1,F) = *o1($1,F) + *o2($1,F)

Valid for: 0.4 < F < 2.0, 0.0 < $1 < 70.0.

(e) Parameters for Deviation Rule (SP-36 Figs. 163,164)

The slope factor for the deviation rule is given by

mF=1($1) = 0.170 + 6.2698x10-5$1 + 1.4096x10-5$1 + 1.9823x10 $1

2 -7 3
where
b($1) = 0.965 - 2.5464x10-3$1 + 4.2695x10-5$1 - 1.3182x10 $1
2 -6 3

Valid for: 0.0 < $1 < 70.0.

(f) Thickness Correction, (K*)t, for Deviation (SP-36 Fig. 172)

Valid for: 0.0 < t/c < 0.12, probably usable up to t/c = 0.15.

(g) Gradient of Deviation Angle with Incidence, d*o/di (SP-36 Fig. 177)

2
Valid for: 0.4 < F < 1.8, 0.0 < $1 < 70.0.

3
 APPENDIX D:

NACA 65-Series Compressor Blade Profiles

The 65-series blade geometry is not represented by closed-form analytical expressions. Instead,
it is necessary to work with tabulated values:

x/c Thickness Camber Line (for CL = 1.0)

(for t = 0.10c)
yt/c yc/c dyc/dx

0.0 0.0 0.0 ---

0.005 0.00752 0.00250 0.42120

0.0075 0.00890 0.00350 0.38875

0.0125 0.01124 0.00535 0.34770

0.025 0.01571 0.00930 0.29155

0.050 0.02222 0.01580 0.23430

0.075 0.02709 0.02120 0.19995

0.10 0.03111 0.02585 0.17485

0.15 0.03746 0.03365 0.13805

0.20 0.04218 0.03980 0.11030

0.25 0.04570 0.04475 0.08745

0.30 0.04824 0.04860 0.06745

0.35 0.04982 0.05150 0.04925

0.40 0.05057 0.05355 0.03225

0.45 0.05029 0.05475 0.01595

0.50 0.04870 0.05515 0.0

0.55 0.04570 0.05475 -0.01595

0.60 0.04151 0.05355 -0.03225

0.65 0.03627 0.05150 -0.04925

0.70 0.03038 0.04860 -0.06745

0.75 0.02451 0.04475 -0.08745

0.80 0.01847 0.03980 -0.11030

0.85 0.01251 0.03365 -0.13805

0.90 0.00749 0.02585 -0.17485

0.95 0.00354 0.01580 -0.23430

1.00 0.00150 0.0 (-0.23430)

1
 The thickness distribution is given for a NACA 65-010 blade which has been modified to give a
finite trailing-edge thickness of 0.3% of the chord length. The baseline thickness distribution has zero
thickness at the trailing edge and therefore cannot be manufactured. The nominal maximum thickness
is 10% of chord. For blades with other values of maximum thickness, the tabulated distribution is
simply scaled accordingly.

The table indicates that maximum camber is at 50% of chord. However, the camber line is not a
simple circular arc. In fact, the slope of the camber line tends to infinity at the leading and trailing edges.
At the leading edge, this gives a "droop" to the nose of the blade which is believed to reduce its
sensitivity to incidence.

Because of the camber line shape, there is no simple relationship between the camber angle, as
defined earlier, and the magnitude of the maximum camber. Instead, the camber line shape is related to
the nominal maximum lift coefficient which the blade shape would achieve as an isolated airfoil. The
camber line shape quoted applies for a nominal lift coefficient CL = 1.0. To generate compressor blades
with a desired camber angle, the following can be used to relate an equivalent circular arc camber angle
to the nominal CL:

(1)

for 2 in degrees.

To generate the geometry for a 65-series compressor blade with a particular camber angle, 2:
(i) From (1), determine the nominal CL.
(ii) Scale the camber line co-ordinates and slope values by (CL/1.0).
(iii) Calculate the blade-surface co-ordinates by superimposing the tabulated thickness distribution
(scaled as necessary if the maximum thickness is to be different from 10% of chord) on the
camber line using Eqns. (5) from Appendix B.

The drawing compares the 10C4/30C50 blade with the 65-series which has the same maximum
thickness and the equivalent camber:

2
 APPENDIX E:

Curve and Surface Fits for Kacker & Okapuu Loss

System for Axial Turbines

Kacker & Okapuu ("A Mean Line Prediction Method for Axial Flow Turbine Efficiency," ASME
Trans., J. Eng. for Power, Vol. 104, January 1982, pp. 111-119) presented an updated version of the
Ainley & Mathieson loss system for axial turbines. The Kacker & Okapuu (KO) system presents a basis
for estimating the complete losses, and thus the efficiency, of an axial turbine at its design point. For a
complete outline of the loss system see the paper.

Some aspects of the loss system are presented only in graphical form in the paper. Therefore a
number of figures have been digitized and curves or surfaces fitted to the data. This appendix
documents the curve fits and, in some cases, demonstrates the quality of the fits graphically. The figure
numbers refer to the figures in the Kacker & Okapuu paper.

(a) Ainley & Mathieson (AMDC) Profile-Loss Coefficients (Figs. 1, 2)

KO use the AMDC correlation for profile loss coefficient, with corrections for Reynolds
number, exit Mach number, channel acceleration, and improvements in design. The AMDC loss
coefficient is obtained as a weighted average of the values for a nozzle blade ($1 = 0) and an impulse
blade. These values are obtained from the plots shown in Figures 1 and 2. The data in these figures
have been fitted to polynomial surfaces of the form:

The values of the coefficients follow:

(i) Nozzle (Fig. 1)

Blade, a0,0 = 0.358716
a0,1 = -1.43508
a0,2 = 1.57161
a0,3 = -0.496917
a1,0 = -0.0112815
a1,1 = 0.0548594
a1,2 = -0.0555387
a1,3 = 0.014165
a2,0 = 0.000175083
a2,1 = -0.000824937
a2,2 = 0.000652287
a2,3 = -7.30141E-05
a3,0 = -8.61323E-07
a3,1 = 3.95998E-06
a3,2 = -1.89698E-06
a3,3 = -4.9954E-07

1
(ii) Impulse Blade, (Fig. 2)

a0,0 = 0.0995503
a0,1 = 0.182837
a0,2 = 0.01603
a1,0 = 0.00621508
a1,1 = -0.0283658
a1,2 = 0.011249
a2,0 = -7.10628E-05
a2,1 = 0.000327648
a2,2 = -0.000122645

(b) Stagger Angle (Fig. 5)

In the early stages of design, axial chord rather than true chord of the blades is often specified.
However, the profile loss correlations require the solidity of the blade row, which is based on the true
chord. KO present an approximate correlation for the stagger angle as a function of the inlet and outlet
angles. The true chord can then be calculated from the axial chord. The graphical data are again fitted
to a surface, using a polynomial of the form:

with coefficients,

a0,0 = -2.90463 a0,1 = 0.307036 a0,2 = 0.370176E-02

a1,0 = 0.412797 a1,1 = -0.355369E-01 a1,2 = -0.194938E-03
a2,0 = 0.593956E-02 a2,1 = 0.389157E-03 a2,2 = 1.74147E-06

The surface fit and the digitized values are compared over.

2
(c) Inlet Mach Number Ratio (Fig. 6)

A correction is made for shock losses at the leading edge of the blade. Since the Mach
number tends to be higher at the hub than at midspan, KO present a correlation for the hub Mach
number as a function of the midspan value and the hub-to-tip ratio. The shock loss is then calculated
from the estimated hub Mach number. The following polynomials were fitted to the curves of Figure
6:

(i) Rotors

(ii) Nozzles

3
(d) Trailing-Edge Energy Coefficient (Fig. 14)

The trailing-edge losses are expressed in terms of the energy coefficient. This was correlated
with the ratio of the trailing-edge thickness to the throat opening. Curves were presented for nozzle and
impulse blades. The values from these curves are then averaged in a weighted way to give the
coefficient for a blade of arbitrary inlet and outlet flow angles.

(i) Impulse (Rotor) Blade

(ii) Nozzle

4
5
 APPENDIX F:

Centrifugal Stresses in Axial Turbomachinery Blades

1.0 Introduction

As briefly mentioned in lectures, the design of a turbomachine involves a trade-off between often
conflicting considerations: aerodynamics, heat transfer, materials, stresses, and vibrations (not to
mention cost). While our focus is on the aerodynamics, it is obviously wasteful to develop even a
preliminary aerodynamic design for a turbomachine which cannot be built for stress reasons.

Turbomachinery blades experience significant unsteady forces which lead to vibratory stresses,
and both low cycle and high cycle fatigue are important considerations. However, the level of the steady
stress determines the margin which is available for these unsteady stresses. In turbines, creep distortion
is an important consideration and the steady centrifugal stress is also the starting point for a creep
analysis. Thus, if the steady centrifugal stresses are kept within established limits, the design is likely to
be mechanically feasible. Fortunately, the steady centrifugal stresses in the rotor blades can be estimated
fairly easily in the early stages of the aerodynamic design.

A later section gives some criteria for judging whether the centrifugal stresses are
acceptable.
These criteria apply primarily to the high-performance machines used in gas turbine engines.
The stresses are particularly high in low hub-to-tip ratio fan blades and in turbine blades; they are much
lower in normal compressor blades. A survey of typical, industrial axial-flow fans from several
manufacturers shows that peak tip speeds are consistently below 120 m/s. It is believed that this limit is
related to the stresses which can be sustained by the rather simple blade attachments, rather than stresses
in the actual blades. Higher tips speeds can be used but these require a switch to a considerably more
expensive method of attachment.

2.0 Steady Centrifugal Stresses

1
 Consider the forces on the small blade element
shown.

Then
(1)

and this can then be integrated from radius R to the tip, RT, (with a specified blade area variation)
to obtain the centrifugal stress at R.

Constant Section Blade:

With dA = 0, integrating (1):

and the maximum stress occurs at the root:

(2)

Tapered Blades:

The cross-sectional area of turbomachinery blades often varies from hub to tip. If the area
decreases, the root stress will be reduced from the value given by (2). Taking into account the taper, the
hub stress can be written

(3)

where K depends on the nature of the taper in cross-sectional area:

(a) Blade with constant cross-section.

(b) Blade with linear taper.

2
 where

The cross-sectional area of the blade is roughly proportional to the product of the chord length (c) and
the maximum thickness (tmax). Thus, the area ratio can be approximated by

If both the chord length and the maximum thickness are tapered linearly from the hub to the tip, to
maintain constant maximum thickness-to-chord ratio, the cross-sectional area will in fact vary
parabolically. It can be shown that the resultant centrifugal stresses will be lower than for linear
taper. However, for HTR > 0.5 the stresses are very similar and the assumption of linear taper gives a
good, slightly conservative, estimate of the hub stress.

3.0 Allowable Stress Levels

From Eqn. (3)

where N = RPM and A = annulus area of the stage. Rearranging,

(4)

From the density and stress limits for currently available blade materials, values of the right-hand side of
(4) can defined by the structural engineer. The aerodynamicist can then use these to verify that the
proposed design is feasible mechanically. The following table gives values of KAN2 which are

3
reasonably representative of the current stress limits for axial
turbomachines:

MACHINE TYPE KAN2 KAN2

(A in inches2, (A in m2,
N in RPM) N in RPM)
Compressor 8-10 x 1010 5.2-6.5 x 107
High-Pressure Turbine (HPT) 4-5 x 1010 2.5-3.2 x 107
Shrouded Low-Pressure Turbine (LPT) 6-8 x 1010 3.8-5.2 x 107
Unshrouded LPT 8-10 x 1010 5.2-6.5 x 107