Swiss Federal Institute of Technology, ETH, Zurich

Turbomachinery Laboratory
Prof. Dr. R. S. Abhari
CFD Investigation on
Inlet Flow Distortion in a Centrifugal Compressor
Masters Thesis 06/07
Armin Zemp
Supervisor:
Dipl. Ing. Albert Kammerer
Inlet Flow Distortion in a Centrifugal Compressor Proposal
Armin Zemp 00-914-663 1
I. Proposal
In turbomachinery, rotating components are subject to vibration. This in particular applies to rotating blades
with vibration being one of the main contributors to failure during operation. Blade vibration is caused by a
complex interaction between the fluid and the structure. Forced response originates from unsteady fluid
structures as conditioned at compressor inlet by pipe bendings, struts or inlet guide vanes. With respect to this
the proposed project addresses the problem of flow distortion at compressor inlet.
The centrifugal compressor test facility RIGI at LSM is modified such that blade vibration measurements with
strain gauges can be carried out. Major requirements of the test facility encompass the utilization of a suitable
inlet flow distortion mechanism with the capability to generate specific flow patterns. This depends on the
design of distortion elements used and their installation upstream of the impeller.
According to the described problem the master project addresses the following problems:
1. Simulation and validation of the CFD model on the basis of steady state computational fluid dynamics
2. Implementation of the FRAP- measured inlet flow field with different distortion patterns as inlet boundary
condition for the computational model
3. Qualitative description of flow patterns within the impeller
4. Time resolved CFD to get unsteady pressure distribution on impeller blades
Supervisor and Contact:
Dipl. Ing. Albert Kammerer
Institut fr Energietechnik
Labor fr Strmungsmaschinen
ETH Zentrum, ML J32
Sonneggstrasse 3
CH- 8092 Zrich
Phone: 044/632 26 94
Email: [email protected]
Proposal Inlet Flow Distortion in a Centrifugal Compressor
2 00-914-663 Armin Zemp
Inlet Flow Distortion in a Centrifugal Compressor Abstract
Armin Zemp 00-914-663 3
II. Abstract
In turbomachinery rotating components are subject to vibration. This in particular applies to rotating blades
being one of the main contributors to failure during operation. Forced response of the blades originates from
unsteady fluid structures as conditioned in the inlet section by duct bends, struts or inlet guide vanes. Most of
the pressure- increasing turbomachines in use are of the radial flow type. The development of turbomachines
using centrifugal effects for increasing fluid pressure have been in use for more than a century. The develop-
ment of centrifugal compressors continued into the 1950s. The enquiry for advanced military helicopters
powered by small gas turbine engines in the 1960s caused a rapid development of the centrifugal compres-
sor. In industry especially in automotive applications the inlet section due to the reduced amount of space con-
tains often bends. These bent ducts create a non- uniform inlet flow field entering the impeller eye of a
centrifugal compressor. Such non- uniformities may have disadvantageous effects on impeller performance
and can excite the impeller blades. Cumpsty (Cumpsty, 2004) presents an investigation of a multistage axial
compressor and remarked that the performance can be seriously affected by inlet distortion. The turboma-
chinery laboratory of ETH Zurich runs a centrifugal compressor test rig. To create inlet distortion in the com-
pressor rig screens with metal grid baffles are installed in the inlet section. Different baffle geometries and
grids of variable porosities are used to generate different distortion patterns. The amplitude of the distortion
is controlled by the porosity of the applied grids. To quantify the flow properties entering the impeller eye the
data acquisition in the inlet flow field is done with the so called fast response aerodynamic probe (FRAP) mea-
surement technique. An extended traversing system allows to move the probe in a plane perpendicular to the
main flow direction. The measured inlet flow field is applied as inlet boundary condition for the numerical in-
vestigation of the centrifugal compressor.
A detailed validation of the CFD model on the base of steady state computations over the entire compressor
map showed good consistence with the test rig performance. The qualitative description of the flow patterns
within the impeller provides information about the fluid dynamics taking place for several operating points.
A main focus lies on the development of the shroud gap vortex. The investigation shows the trend to increased
total pressure loss due to the tip leakage vortex with increased mass flow rate at constant blade speed. In con-
trast the total pressure loss due to the tip leakage vortex in the inter blade channel decreases with increased
impeller speed at constant mass flow rate due to the fact that the amount of main blade leakage flow passing
in front of the splitter blade leading edge increases. The time resolved CFD investigation provides information
about the unsteady pressure distribution on the main blade surface. The spectral analysis of the time resolved
blade pressure distribution at several operating points with different distortion screen geometries shows the
temporal evolution of the dynamic load on the blade caused by the inlet distortion. The transformation of the
unsteady blade pressure into the frequency domain provides information about the frequency and the ampli-
tude of the excitation over the whole blade surface. The comparison of different inlet boundary conditions
such as turbulence intensity profile measured with FRAP probes versus constant turbulence intensity level, or
idealized total pressure profile at the inlet of the CFD model versus measured total pressure profile containing
all rudiments of upstream installations (struts for the slip ring support for example), shows the insensitivity
of the model concerning the unsteady pressure distribution on the main blade surface.
Abstract Inlet Flow Distortion in a Centrifugal Compressor
4 00-914-663 Armin Zemp
Inlet Flow Distortion in a Centrifugal Compressor Acknowledgement
Armin Zemp 00-914-663 5
III. Acknowledgement
I would like to thank Prof. Dr. R. S. Abhari for the possibility to make the present work at Turbomachinery Lab-
oratory at ETH Zurich. The present work is part of real research of Mr. Kammerers Ph.D.- Thesis at the institute
and allowed to have a detailed insight in current research activities at the Turbomachinery Laboratory. The
masters thesis together with the foregoing semester thesis caused great fascination for centrifugal compres-
sors.
I also would like to express my sincere thanks to my supervisor Dipl. Ing. Albert Kammerer. His exemplary way
of assistance and his friendly nature were the base to render possible such a salutary time at the institute. His
expert knowledge in the centrifugal compressor area and his cooperativeness combined with a scientific way
of solving disparate problems allowed me to learn a lot not only on radial compressors but also on scientific
methodology.
Acknowledgement Inlet Flow Distortion in a Centrifugal Compressor
6 00-914-663 Armin Zemp
Inlet Flow Distortion in a Centrifugal Compressor Table of Contents
Armin Zemp 00-914-663 7
IV. Table of Contents
I. Proposal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
II. Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
III. Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
IV. Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1 General Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Centrifugal Compressors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Internal Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2. Objectives and Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11
2.2 Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11
3. Theoretical Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .13
3.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2 Non- dimensional Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Work Input Coefficient 14
Polytropic Head Coefficient 14
Polytropic Efficiency 14
Flow Coefficient 15
Mach Number 15
Stage Mach Number 15
Turbine Euler Equation 15
3.3 Theoretical Analysis of a Centrifugal Compressor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Inlet Casing 17
Impeller 17
Conservation of Rothalpy 18
Diffuser 19
Inlet Velocity Limitations 19
3.4 Performance of Centrifugal Compressors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Determing the Pressure Ratio 20
Effect of Backswept Vanes 21
3.5 The Diffuser System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Vaneless Diffusers 22
Vaned Diffusers 23
3.6 Stall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.7 Surging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.8 Choking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.9 Internal Flow Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .26
Boundary Layer in a Bend 26
Inlet Flow Structure and its Effect on the Compressor Performance 27
3.10 Vibration and Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Vibration 27
Forced Vibration 28
3.11 The Rigi Test Rig at Turbomachinery Laboratory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .29
Test Rig Features 30
3.12 Measurement Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.13 Measurement Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Distortion Screens 33
3.14 Computational Fluid Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
History of CFD 35
Mathematics of CFD 35
CFD Methodology 36
Governing Equations 38
Turbulence Models 39
4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.1 Stationary CFD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .49
Grid 49
Boundary Conditions 50
Turbulence Model and Solver Setup 51
Convergence Quantification 52
Validation 55
Flow through Impeller 61
4.2 Transient CFD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Grid 74
Boundary Conditions 75
Turbulence Measurement 75
Table of Contents Inlet Flow Distortion in a Centrifugal Compressor
8 00-914-663 Armin Zemp
Turbulence Model and Solver Setup 79
Operating Points for Transient CFD 79
Convergence Quantification 80
Validation 83
Influence of Turbulence Intensity Boundary Condition 83
Flow through Impeller 84
Unsteady Blade Pressure 87
5. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6. Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7. References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Inlet Flow Distortion in a Centrifugal Compressor Introduction
Armin Zemp 00-914-663 9
1. Introduction
1.1 General Background
The centrifugal or radial compressor finds the most widespread use of any compressor type. At one extreme
there are machines producing pressure ratios of more than 8:1 from one stage. These impellers used for avia-
tion appliances have titanium alloy impellers. At the other end of the range there are very low cost devices in
domestic appliances fabricated out of sheet metal where the overriding concern is to keep the cost low. Re-
search and development activity has concentrated on those areas where the performance is high.
1.2 Centrifugal Compressors
The development of turbomachines using centrifugal effects for increasing fluid pressure have been in use for
more than a century. The earliest machines using this principle were hydraulic pumps, followed later by ven-
tilating fans and blowers.
The development of centrifugal compressors continued into the 1950s but it had become clear that for the
larger engines required for aircraft propulsion the axial flow compressor was preferred. Not only due to the
smaller frontal area (and therefore smaller drag) needed for an axial flow compressor but also due to the high-
er efficiency for the same duty (about 3% to 4%). But at very low air mass flow rates the efficiency of axial com-
pressors drops dramatically. The blading is small and difficult to fabricate. Especially the narrow range of
tolerance in such small machines is difficult to keep.
The enquiry for advanced military helicopters powered by small gas turbine engines in the 1960s caused a
rapid development of the centrifugal compressor.
Fig. 1.2.0.1: Multi Stage Centrifugal Compressor (ABB)
Introduction Inlet Flow Distortion in a Centrifugal Compressor
10 00-914-663 Armin Zemp
1.3 Internal Flows
It is often the fluid motion which set the performance of devices such as propulsion systems and their com-
ponents, fluid machinery, ducts and channels. There are often large changes in direction and in velocity. For
example, deflections of over 90 are common in fluid machinery. Deflection of the non- uniform flows might
create three dimensional motions normal to the mean flow direction. Additionally, there is often strong swirl,
with consequent phenomena that are different than for flow without swirl. For example, static pressure rise
can be associated almost entirely with the circumferential swirl velocity component. The motions are often
unsteady. Unsteadiness is necessary for work exchange in turbomachines. Waves, oscillations and self- excit-
ed unsteadiness not only affect system behavior, but can sometimes be a limiting factor for operational re-
gimes.
Fig. 1.3.0.1: Flow Field after 90 Bend (Engeda et al., 2003)
Inlet Flow Distortion in a Centrifugal Compressor Objectives and Approach
Armin Zemp 00-914-663 11
2. Objectives and Approach
2.1 Objectives
The present work focuses on inlet flow distortion in centrifugal compressors. The excitation of the impeller
blades due to the harmonic load generated by the inlet distortion in the flow in front of the impeller blades is
subject of research and part of Mr. Albert Kammerers PhD- Thesis. The present work contains the numerical
simulation of the centrifugal impeller and the tracking of the distortion convecting through the compressor.
The FRAP measurement results of the foregoing semester thesis are used as boundary condition at the inlet
of the computational domain.
A main focus lies on the unsteady pressure distribution on the main blade surface due to the fact that the dis-
tortion convecting through the device may excite the impeller blades if the excitation is in resonance with the
blades.
2.2 Approach
A series of steady state computations at operating points over the whole compressor map allow a detailed val-
idation of the performance of the numerical model. In a second step unsteady CFD simulations provide time
resolved results of the distortion convecting through the impeller. The unsteady pressure distribution on the
surface of the blades is used for a spectral analysis.
Objectives and Approach Inlet Flow Distortion in a Centrifugal Compressor
12 00-914-663 Armin Zemp
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 13
3. Theoretical Background
3.1 Definitions
Most of the pressure- increasing turbomachines in use are of the radial- flow type and vary from fans that pro-
duce pressure rises equivalent to a few millimeters of water to pumps producing heads of many hundreds of
metres of water. The term pump is used when referring to machines that increase the pressure of a flowing
liquid. The term fan is used for machines imparting only a small increase in pressure to a flowing gas. In this
case the pressure rise is usually so small that the gas can be considered as being incompressible.
A compressor gives a substantial rise in pressure to a flowing gas. For purposes of definitions, the boundary
between fans and compressors is often taken as that where the density ratio across the machine is 1.05. Some-
times, but more rarely, the term blower is used instead of fan.
A centrifugal compressor or pump consists essentially of a rotating impeller followed by a diffuser. Figure
3.1.0.1 shows the various elements of a centrifugal compressor. Fluid is drawn in through the inlet casing into
the eye of the impeller. The function of the impeller is to increase the energy level of the fluid by whirling it
outwards, thereby increasing the angular momentum of the fluid. Both the static pressure and the velocity
are increased within the impeller. The purpose of the diffuser is to convert the kinetic energy of the fluid leav-
ing the impeller into pressure energy. This process can be accomplished by free diffusion in the annular space
surrounding the impeller or by incorporating a row of fixed diffuser vanes which allows the diffuser to be
made very much smaller.
Outside the diffuser is a scroll or volute whose function is to collect the flow from the diffuser and deliver it to
the outlet pipe. In low- speed compressors and pump applications where simplicity and low cost count for
more than efficiency, the volute follows immediately after the impeller.
Fig. 3.1.0.1 Centrifugal Compressor Stage (Dixon, 1998)
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
14 00-914-663 Armin Zemp
3.2 Non- dimensional Parameters
With the assumption of a polytropic process, , and neglecting inlet guide vanes, return channel or
scroll the following parameters are used tho characterize a centrifugal compressor system:
Fig. 3.2.0.1 Enthalpy- Entropy Diagram (Abhari, WS 05/06)
3.2.1 Work Input Coefficient
Equation 3.2.1.1
3.2.2 Polytropic Head Coefficient
Equation 3.2.2.1
with and
3.2.3 Polytropic Efficiency
Equation 3.2.3.1
v p d
h d
--------- const =

P
m

----
u
2
2
-----
h
01 03
u
2
2
---------------------
h
0
u
2
2
--------- = = =

p
v p d
01
03

u
2
2
-------------- =
v p d
01
03

RT
01
n
n 1
------------
\
| |
p
03
p
01
-------
\
| |
n 1
n
------------
\
| |
1
)

`


=
n 1
n
------------
\
| |
p
03
p
01
-------
\
| |
ln
T
03
T
01
--------
\
| |
ln
------------------- =

p
v p d
01
03

h
01 03
---------------------

-------
1

------------
p
03
p
01
-------
\
| |
ln
T
03
T
01
--------
\
| |
ln
------------------- = = =
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 15
3.2.4 Flow Coefficient
Equation 3.2.4.1
3.2.5 Mach Number
Equation 3.2.5.1
This is the most useful aerodynamic parameter in compressible flows. It is named after the nineteenth- cen-
tury austrian physicist Ernst Mach but was first used by Professor Ackeret at the ETH Zurich.
3.2.6 Stage Mach Number
Equation 3.2.6.1
This definition is used as a non- dimensional speed. Values above 1 do not imply that the flow is choked! The
significance of the stage Mach number can best be appreciated in combination with the definition of the work
input coefficient assuming a perfect gas:
Equation 3.2.6.2
with and . The temperature ratio is and the pressure ratio
is with . The stage Mach number is therefore a useful
non- dimensional parameter that relates the work input coefficient to the pressure and temperature rise
across the stage.
3.2.7 Turbine Euler Equation
With the Turbine Euler Equation the work input coefficient can be given as a function of velocity components:
Equation 3.2.7.1
and with (no inlet swirl)
Equation 3.2.7.2

01
D
2
2
u
2

4
---
-------------------------
V

D
2
2
u
2

4
---
---------------- = =
Ma
w
a
----
w
local
RT
local
-------------------------- = =
Mu
2
u
2
a
2
-----
u
2
RT
01
-------------------- = =
h
0
u
2
2
c
p
T
03
T
01
( )
T
03
T
01
-------- 1
u
2
2
c
p
T
01
------------- + = = =
c
p

1
------------R = a
1
RT
01
=
T
03
T
01
-------- 1 1 ( )Mu
2
2
+ =

0
p
03
p
01
------- 1 1 ( )Mu
2
2
+ { }

1
------------
= =
n
n 1
------------
p

1
------------ =
h
0
u
2
c
2
u
1
c
1
=
c
1
0 =

h
0
u
2
2
---------
u
2
c
2
u
1
c
1

u
2
2
-----------------------------------
c
2
u
2
--------- = = =
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
16 00-914-663 Armin Zemp
An alternative form of the Euler equation sheds considerable light on the power and the pressure rise in a com-
pressor stage:
Equation 3.2.7.3
Then the Euler equation can also be expressed as:
Equation 3.2.7.4
or
Equation 3.2.7.5
In a typical radial compressor stage the centrifugal effect accounts for about one half of the work input. Since
the centrifugal effect is free of losses radial stages have higher pressure rise than axial ones and still attain
relatively high efficiencies despite their very complicated flow paths. The second term in the equation ad-
dresses the diffusion in the impeller. This effect accounts for about 20% of the work input. The last term in the
equation shows the diffusion in the non rotating part and accounts for about 30% of the work input.
3.3 Theoretical Analysis of a Centrifugal Compressor
Fig. 3.3.0.1 Enthalpy- Entropy Diagram for Complete Centrifugal Compressor Stage (Dixon, 1998)
c
r2
2
c
2
2
c
2
2
w
2
2
u
2
c
2
( )
2
u
2
c
2

1
2
--- c
2
2
u
2
2
w
2
2
+ ( ) = = =
h
0
u
2
c
2
u
1
c
1

1
2
--- c
2
2
u
2
2
w
2
2
+ ( )
1
2
--- c
1
2
u
1
2
w
1
2
+ ( ) = =
h
0
u
2
2
u
1
2

2
-----------------
w
2
2
w
1
2

2
-------------------
c
2
2
c
1
2

2
---------------- + =
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 17
The flow through a compressor stage is a highly complicated, three- dimensional motion and a full analysis
presents many problems of the highest order of difficulty. However, we can obtain approximate solutions
quite readily by simplifying the flow model. We adopt the so- called one-dimensional approach which as-
sumes that the fluid conditions are uniform over certain flow cross- sections. These cross- sections are conve-
niently taken immediately before and after the impeller as well as at inlet and exit of the entire machine.
Where inlet vanes are used to give prerotation to the fluid entering the impeller, the one- dimensional treat-
ment is no longer valid and an extension of the analysis is then required.
3.3.1 Inlet Casing
The fluid is accelerated from velocity c
0
to velocity c
1
and the static pressure falls from p
0
to p
1
. Since the stag-
nation enthalpy is constant in steady, adiabatic flow without shaft work then h
00
= h
01
or,
Equation 3.3.1.1
3.3.2 Impeller
The general three- dimensional motion has components of velocity c
r
, and c
x
respectively in the radial, tan-
gential and axial directions and
Equation 3.3.2.1
Thus, the rothalpy is
Equation 3.3.2.2
Adding and subtracting this becomes
Equation 3.3.2.3
From the velocity triangle, , and together with :
Equation 3.3.2.4
or
Equation 3.3.2.5
since
Equation 3.3.2.6
h
0
1
2
--c
0
2
+ h
1
1
2
--c
1
2
+ =
c

c
2
c
r
2
c

2
c
x
2
+ + =
I h
1
2
-- c
r
2
c

2
c
x
2
2Uc

+ + ( ) + =
1
2
--U
2
I h
1
2
-- U c

( )
2
c
r
2
c
x
2
U
2
+ + { } + =
U c

= w
2
c
r
2
w

2
c
x
2
+ + =
I h
1
2
--- w
2
U
2
( ) + =
I h
0 rel ,
1
2
---U
2
=
h
0 rel ,
h
1
2
---w
2
+ =
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
18 00-914-663 Armin Zemp
Since across the impeller,
Equation 3.3.2.7
The above expression provides the reason why the static enthalpy rise in a centrifugal compressor is so large
compared with a single stage axial compressor. On the right hand side of the above expression, the second
term is the contribution from the diffusion of relative velocity.
The first term, , is the contribution due to the centrifugal action which is zero if the streamlines re-
main at the same radii before and after the impeller.
In centrifugal compressors and pumps, the absolute flow normally has no whirl component or angular mo-
mentum and . This is the normal situation where the flow is free to enter axially. For such a flow the
specific work done on the fluid is written as
Equation 3.3.2.8
In high pressure ratio compressors it may be necessary to impart prerotation to the flow entering the impeller
as a means of reducing a high relative inlet velocity. The effects of high relative velocity at the impeller inlet
are experienced as Mach number effects in compressors and cavitation effects in pumps. The usual method
of establishing prerotation requires the installation of a row of inlet guide vanes upstream of the impeller.
3.3.3 Conservation of Rothalpy
A cornerstone of the analysis of steady, relative flows in rotating systems has, for many years, been the immu-
table nature of the fluid mechanical property rothalpy.
The conditions under which the rothalpy of a fluid is conserved in the flow through impellers and rotors have
been closely scrutinized by several researchers. Lyman (Lyman, 1993) reviewed the equations and physics gov-
erning the constancy of rothalpy in turbomachine fluid flows and found that an increase in rothalpy was pos-
sible for steady, viscous flow without heat transfer or body forces. He proved mathematically that the rothalpy
increase was generated mainly by the fluid friction acting on the stationary shroud of the compressor consid-
ered. From his analysis, and put in the simplest terms, he deduced that:
Equation 3.3.3.1
where is the power loss due to fluid friction on the stationary shroud.
I
1
I
2
=
h
2
h
1

1
2
--- U
2
2
U
1
2
( )
1
2
--- w
1
2
w
2
2
( ) + =
1
2
--- w
2
2
w
1
2
( )
1
2
--- U
2
2
U
1
2
( )
c

0 =
W U
2
c
2
h
02
h
01
= =
h
02
h
01
Uc

( )
2
Uc

( )
1

W
f
m

------ + =
W
f
m

I
2
I
1
( ) =
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 19
Lyman did not give any numerical values in support of his analysis. In the discussion of Lymans paper, Moore
disclosed that earlier viscous flow calculations of the flow in centrifugal flow compressors (Moore et al., 1984)
of the power loss in a centrifugal compressor had shown a rothalpy production amounting to 1.2 per cent of
the total work input. This was due to the shear work done at the impeller shroud and it was acknowledged to
be of the same order of magnitude as the work done overcoming disc friction on the back face of the impeller.
Often disc friction is ignored in preliminary design calculations. A later, careful, order- of- magnitude investi-
gation by Bosman and Jadayel (Bosman et al., 1996) showed that the change in rothalpy through a centrifugal
compressor impeller would be negligible under typical operating conditions. They also believed that it was not
possible to accurately calculate the change in rothalpy because the effects due to inexact turbulence model-
ling and truncation error in computation would far exceed those due to non- conservation of rothalpy.
3.3.4 Diffuser
The fluid is decelerated adiabatically and the static pressure rising. The volute and outlet diffuser involve some
further deceleration. As the stagnation enthalpy in steady adiabatic flow without shaft work is constant,
or .
3.3.5 Inlet Velocity Limitations
The inlet eye is an important critical region in centrifugal pumps and compressors requiring careful consider-
ation at the design stage. If the relative velocity of the inlet flow is too large in pumps, cavitation may result
with consequent blade erosion or even reduced performance. In compressors large relative velocities can
cause an increase in the impeller total pressure losses. In high-speed centrifugal compressors Mach number
effects may become important with high relative velocities in the inlet. By suitable sizing of the eye the max-
imum relative velocity, or some related parameter, can be minimized to give the optimum inlet flow condi-
tions.
3.4 Performance of Centrifugal Compressors
The performance of compressible flow machines is usually described in terms of the groups of variables pre-
sented in chapter 3.2. These characteristics are dependent on other variables such as the conditions of pres-
sure and temperature at the compressor inlet and physical properties of the working fluid. To study the
performance of a compressor completely, it is necessary to plot the total pressure coefficient against
the mass flow parameter for fixed speed intervals of . Figure 3.4.0.1 shows an idealized fixed speed
characteristic. Consider a valve placed in the delivery line of a compressor running at constant speed. First, it
is supposed that the valve is fully closed. Then the pressure ratio will have some value as indicated by Point A.
This pressure ratio is available from vanes moving the air about in the impeller.
Now, it is supposed that the valve is opened and airflow begins. The diffuser contributes to the pressure rise,
the pressure ratio increases, and at Point B, the maximum pressure occurs. But the compressor efficiency at
this pressure ratio will be below the maximum efficiency. Point C indicates the further increase in mass flow,
h
02
h
03
= h
2
1
2
---c
2
2
+ h
3
1
2
---c
3
2
+ =

p
03
p
01
------- =
m

T
01
p
01
-----------------
N
T
01
------------
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
20 00-914-663 Armin Zemp
but the pressure has dropped slightly from the maximum possible value. This is the design mass flow rate
pressure ratio. Further increases in mass flow will increase the slope of the curve until point D. Point D indi-
cates that the pressure rise is zero. However, as mentioned this is an idealized characteristic.
Fig. 3.4.0.1: Fixed Speed Characteristic (Dixon, 1998)
3.4.1 Determing the Pressure Ratio
Considering a centrifugal compressor having zero inlet swirl, compressing a perfect gas. With the usual nota-
tion the energy transfer is
Equation 3.4.1.1
The overall or total to total efficiency is
Equation 3.4.1.2
Now the overall pressure ratio is
Equation 3.4.1.3
W
W
c
m

------- h
02
h
01
U
2
c
2
= = =

c
h
03 s ,
h
01

h
03
h
01

-------------------------
c
p
T
01
T
03 s ,
T
01
------------ 1
\
| |
h
02
h
01

----------------------------------------- c
p
T
01
T
03 s ,
T
01
------------ 1
\
| |
U
2
c
2
--------------------------- = = =
p
03
p
01
-------
T
03 s ,
T
01
------------
\
| |

1
------------
=
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 21
In all compressors the basic flow process is one of diffusion. Boundary layers are prone to separate and the
flow is extremely complex. With separated wakes in the flow, unsteady flow downstream of the impeller can
occur. It must be stressed that a broad understanding of the flow processes within a centrifugal compressor
is still a vital requirement for the further progress of new design methods. A characteristic of all high perfor-
mance compressors is that as the design pressure ratio has increased, so the range of mass flow between
surge and choking has diminished. In the case of the centrifugal compressor, choking can occur when the
mach number entering the diffuser passages is just in excess of unity. This is a severe problem which is aggra-
vated by shock- induced separation of the boundary layers on the vanes which worsens the problem of flow
blockage.
3.4.2 Effect of Backswept Vanes
Came (Came, 1978), Whitfield and Baines (Whitfield et al., 1990) have commented upon the trend towards the
use of higher pressure ratios from single- stage compressors leading to more highly stressed impellers. The
increasing use of back swept vanes and higher blade tip speeds result in higher direct stress in the impeller
and bending stress in the non- radial vanes.
Fig. 3.4.2.1 Pressure Ratio vs Stage Mach Number for Different Backsweep Angles (Dixon, 1998)
The effect of using back swept impeller vanes on the pressure ratio is shown in figure 3.4.2.1 for a range of
blade mach number. It is evident that the use of back sweep of the vanes at a given blade speed causes a loss
in pressure ratio. In order to maintain a given pressure ratio it would be necessary to increase the design speed
which, it has been noted already, increases the blade stresses. With high blade tip speeds the absolute flow
leaving the impeller may have a mach number well in excess of unity. As this mach number can be related to
the mach number at entry to the diffuser vanes, it is of some advantage to be able to calculate the former.
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
22 00-914-663 Armin Zemp
According to Whitfield and Baines (Withfield et al., 1990) the two most important aerodynamic parameters at
impeller exit are the magnitude and direction of the absolute mach number . If has a too high value,
the process of efficient flow deceleration within the diffuser itself is made more difficult leading to high fric-
tion losses as well as the increased possibility of shock losses. If the flow angle is large the flow path in the
vaneless diffuser will be excessively long resulting in high friction losses and possible stall and flow instability.
Several researchers, e.g. Rodgers and Sapiro (Rodgers et al., 1972) have shown that the optimum flow angle is
in the range 60 < < 70. Backswept vanes give a reduction of the impeller discharge Mach number at any
given tip speed. A designer making the change from radial vanes to back swept vanes will incur a reduction in
the design pressure ratio if the vane tip speed remains the same. To recover the original pressure ratio the de-
signer is forced to increase the blade tip speed which increases the discharge Mach number. Fortunately, it
turns out that this increase in is rather less than the reduction obtained by the use of backsweep.
3.5 The Diffuser System
Centrifugal compressors and pumps are, in general, fitted with either a vaneless or a vaned diffuser to trans-
form the kinetic energy at impeller outlet into static pressure.
3.5.1 Vaneless Diffusers
The simplest concept of diffusion in a radial flow machine is one where the swirl velocity is reduced by an in-
crease in radius (conservation of angular momentum) and the radial velocity component is controlled by the
radial flow area. From continuity, since , where b is the width of passage, then
Equation 3.5.1.1
Assuming the flow is frictionless in the diffuser, the angular momentum is constant and . Now the
tangential velocity component is usually very much larger than the radial velocity component . There-
fore, the ratio of inlet to outlet diffuser velocities is approximately . Clearly, to obtain useful reductions
in velocity, vaneless diffusers must be large. This may not be a disadvantage in industrial applications where
weight and size may be of secondary importance compared with the cost of a vaned diffuser. A factor in favour
of vaneless diffusers is the wide operating range obtainable, vaned diffusers being more sensitive to flow vari-
ation because of incidence effects.
For a parallel- walled radial diffuser in incompressible flow, the continuity of mass flow equation requires that
is constant. Assuming that remains constant, then the absolute flow angle is also constant as the
fluid is diffused outwards. Under these conditions the flow path is a logarithmic spiral.
M
2
M
2

2
M
2
m

Ac
r
2rbc
r
= =
c
r
r
2
b
2

2
c
r2
rb
------------------------ =
c

c
2
r
2
r
---- =
c

c
r
c
2
c
3
-----
r
3
r
2
----
rc
1
rc

2
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 23
3.5.2 Vaned Diffusers
In the vaned diffuser the vanes are used to remove the swirl of the fluid at a higher rate than is possible by a
simple increase in radius, thereby reducing the length of flow path and diameter. The vaned diffuser is advan-
tageous where small size is important.
The flow follows an approximately logarithmic spiral path to the vanes after which it is constrained by the dif-
fuser channels. For rapid diffusion the axis of the channel is straight and tangential to the spiral as shown. The
passages are generally designed on the basis of simple channel theory with an equivalent angle of divergence
of between 8 and 12 to control separation.
In many applications of the centrifugal compressor, size is important and the outside diameter must be min-
imized. With a vaned diffuser the channel length can be crucial when considering the final size of the com-
pressor. Clements and Artt (Clements et al., 1988) considered this and performed a series of experiments
aimed at determining the optimum diffuser channel length to width ratio, L/W. They found that, on the com-
pressor they tested, increasing L/W beyond 3.7 did not produce any improvement in the performance, the
pressure gradient at that point having reached zero.
Another significant result found by them was that the pressure gradient in the diffuser channel when
L/W > 2.13 was not greater than that which could be obtained in a vaneless diffuser. Hence, removing com-
pletely that portion of the diffuser after this point would yield the same pressure recovery as with the full dif-
fuser.
The number of diffuser vanes can also have a direct bearing on the efficiency and surge margin of the com-
pressor. It is now widely accepted that surge occurs at higher flow rates when vaned diffusers are used than
when a simple vaneless diffuser design is adopted. Came and Herbert (Came et al., 1980) quoted an example
where a reduction of the number of diffuser vanes from 29 to 13 caused a significant improvement in the surge
margin. Generally, it is accepted that it is better to have fewer diffuser vanes than impeller vanes in order to
achieve a wide range of surge- free flow. With several adjacent diffuser passages sharing the gas from one im-
peller passage, the uneven velocity distribution from that passage results in alternate diffuser passages being
either starved or choked. This is an unstable situation leading to flow reversal in the passages and to surge of
the compressor. When the number of diffuser passages is less than the number of impeller passages a more
uniform total flow results.
3.6 Stall
Stalling of a stage will be defined as the aerodynamic stall, or the breakaway of the flow from the suction side
of the blade airfoil. A multistage compressor may operate stably in the unsurged region with one or more of
the stages stalled, and the rest of the stages unstalled. Stall, in general, is characterized by reverse flow near
the blade tip, which disrupts the velocity distribution and hence adversely affects the performance of the suc-
ceeding stages.
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
24 00-914-663 Armin Zemp
Referring to the cascade of fig. 3.6.0.1, it is supposed that some non- uniformity in the approaching flow or in
a blade profile causes blade B to stall. The air now flows onto blade A at an increased angle of incidence due
to blockage of channel AB. The blade A then stalls, but the flow on blade C is now at a lower incidence, and
blade C may unstall. Therefore the stall may pass along the cascade in the direction of lift on the blades. Ro-
tating stall may lead to vibrations resulting in fatigue failure in other parts of the system.
Fig. 3.6.0.1 Mechanism of Stall Propagation (Gorla et al., 2003)
3.7 Surging
Surging is marked by a complete breakdown of the continuous steady flow throughout the whole compressor,
resulting in large fluctuations of flow with time and also in subsequent mechanical damage to the compres-
sor. The phenomenon of surging should not be confused with the stalling of a compressor stage. Figure 3.7.0.1
shows typical overall pressure ratios and efficiencies of a centrifugal compressor stage. The pressure ratio for
a given speed, unlike the temperature ratio, is strongly dependent on mass flow rate, since the machine is usu-
ally at its peak value for a narrow range of mass flows. When the compressor is running at a particular speed
and the discharge is gradually reduced, the pressure ratio will first increase, peaks at a maximum value, and
then decreased. The pressure ratio is maximized when the isentropic efficiency has the maximum value.
When the discharge is further reduced, the pressure ratio drops due to fall in the isentropic efficiency. If the
downstream pressure does not drop quickly there will be backflow accompanied by further decrease in mass
flow. In the mean time, if the downstream pressure drops below the compressor outlet pressure, there will be
increase in mass flow. This phenomenon of sudden drop in delivery pressure accompanied by pulsating flow
is called surging. The point on the curve where surging starts is called the surge point. When the discharge
pipe of the compressor is completely choked (mass flow is zero) the pressure ratio will have some value due to
the centrifugal head produced by the impeller. Between the zero mass flow and the surge point mass flow, the
operation of the compressor will be unstable. The line joining the surge points at different speeds gives the
surge line.
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 25
Fig. 3.7.0.1 Centrifugal Compressor Characteristic (Gorla et al., 2003)
3.8 Choking
When the velocity of fluid in a passage reaches the speed of sound at any cross section, the flow becomes
choked (air ceases to flow). In the case of inlet flow passages, mass flow is constant. The choking behavior of
rotating passages differs from that of the stationary passages, and therefore it is necessary to make separate
analysis for impeller and diffuser, assuming one dimensional, adiabatic flow, and that the fluid is a perfect gas.
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
26 00-914-663 Armin Zemp
3.9 Internal Flow Fundamentals
3.9.1 Boundary Layer in a Bend
When a flow that is parallel but non- uniform in velocity or density is made to follow a curved path, the result
is a three- dimensional motion with velocity components normal to the overall flow direction. Cross- flow of
this type is associated with the generation of a streamwise component of vorticity and commonly referred to
as secondary flow. The name derives from the view that one can identify a primary flow direction along a pas-
sage or bend and hence also specify the departures from this primary direction. Although the term secondary
is in common use, it can be a misnomer because the cross- flow velocities are often a substantial fraction of
the primary velocity.
The flow in a boundary layer on the bottom of a curved passage such as a rectangular bend furnishes an illus-
tration of the type of motions to be addressed. Generation of secondary flow can be viewed in terms of the
differential convection of boundary layer vorticity through the bend. The flow outside of the boundary layer
can be considered as a two- dimensional irrotational stream.
Fig. 3.9.1.1 Elementary Contour of Two Streamlines (Greitzer et al., 2004)
Figure 3.9.1.1 shows an elementary contour formed by two streamlines and two normals between the stream-
lines in a two- dimensional flow. From evaluation of the circulation around this contour, the component of
vorticity perpendicular to the page can be written in terms of the rate of variation of velocity in the normal
direction and the local streamline radius of curvature as
Equation 3.9.1.2
For a flow which is irrotational outside of the boundary layer, , states that particles on the outside of a bend
have a lower velocity than particles on the inside. Particles on the outside of the bend travel a longer distance
than those on the inside. If boundary layer vortex lines are convected with a velocity proportional to the local
free- stream velocity, vortex lines initially normal to the flow will be tipped into the streamwise direction as
they traverse the bend with a resulting cross- flow. In a plane perpendicular to the free stream one can see the
boundary layer fluid migrating towards the inside of the bend.

u
n
-----
u
r
c
---- + =

Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background

Armin Zemp 00-914-663 27
3.9.2 Inlet Flow Structure and its Effect on the Compressor Performance
The performance of centrifugal compressors can be seriously degraded by inlet flow distortions that result
from an unsatisfactory inlet configuration. The distortion can be in static pressure or stagnation temperature,
but the most common distortion is stagnation pressure. Such distortions often occur naturally because of the
unsatisfactory nature of the inlet or because of operational effects. Very often the distortion is transient and
it is generally recognized that the distortion will have little or no effect unless it persists for at least one revo-
lution of the impeller. The distortion pattern is normally nonuniform in the circumferential and the radial
sense. Circumferential distortion seems to be the most serious.
Ariga et al. (Ariga et al., 1982) investigated experimentally the influence of inlet distortion on the performance
of a low- speed centrifugal compressor with vaneless diffuser, mainly in the impeller with artificially created
radial and circumferential distortion generators by locating multiple layers of honeycomb at upstream of im-
peller and compared the result with the case of no distortion. According to his results, the distorted inlet pro-
file degrades the impeller efficiency significantly by changing the incidence angle, especially in case of tip
distortion. He observed the tendency that the performance degenerating effect due to the distortion grows
as the rotational speed and the flow rate increases. The highest pressure and the lowest stable flow are
achieved with inlets that have no distortion. Circumferential distortions create the largest loss of pressure ra-
tio and flow angle. Circumferential distortions can be generated by non- axisymmetric obstacles such as
struts or a bending duct.
3.10 Vibration and Noise
Blade vibration and noise are conveniently considered together for two reasons. Both are unsteady processes
and each can be a serious nuisance. Of the two the problem of blade vibration is likely to be far more worrying
to the designer because high levels of vibration can lead to part or all of the blades breaking off. Enormous
amounts of downstream damage are then likely to occur possibly causing total destruction of the compressor.
Noise is a less serious threat to the compressor itself and only in exceptional cases is the level of pressure fluc-
tuation high enough to cause mechanical damage.
3.10.1 Vibration
The vibrations of blades in compressors and turbines have many features in common and in axial turbines vi-
bration was a problem before the first axial compressor was built. Campbells (Campbell, 1924) work on axial
turbines carries over to todays compressors and is remembered in the Campbell diagram. The vulnerability of
turbomachines to vibration, particularly axial machines, is not surprising in view of the use of long slender
blades, the large gas loads, the proximity of moving and stationary components and the small amount of me-
chanical damping.
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
28 00-914-663 Armin Zemp
There are two quite different types of vibration to be considered. Forced vibration and flutter. The former,
which is easier to understand and to obviate, arises from the movement of the rotor through disturbances
which are usually stationary, for example wakes or potential field of an upstream stator, the wake of some up-
stream element such as a strut or the inlet distortion. It can also arise from the excitation produced by passing
through rotating stall cells. Forced vibration becomes a problem when the excitation frequency coincides
with a natural frequency of the blades. Flutter, on the other hand, is a self- excited oscillation or instability at
or close to the natural frequency of the mechanical system which does not require any disturbance of finite
amplitude to excite it.
3.10.2 Forced Vibration
Fig. 3.10.2.1 Campbell Diagram for a Rotor Blade (circles indicate for forced resonances, crosses show either rotational stall or flutter)
(Cumpsty, 2004)
The Campbell diagram (see fig. 3.10.2.1) is a convenient way to view the possible coincidence of blade natural
frequency with excitation of the rotor blades by non- uniformities of the flow. Some of the engine orders are
more important than others. The lowest engine orders tend to produce quite high excitation because the flow
around the annulus is never truly uniform. There are several reasons for this. The inlet flow has non- uniform
stagnation pressure because of ingested distortion or intake boundary layer effects or has static pressure dis-
tortion due to such effects as upstream bends. Inlet distortion tends to produce excitations at the low engine
orders, the precise excitation being obtained from a Fourier analysis of the circumferential pattern. Struts up-
stream of downstream excite the blades at engine orders corresponding to their number and harmonics
thereof. Upstream obstructions cause excitation by their potential flow effect and by their wakes. Down-
stream bodies can only affect the rotor by the potential flow effect. The potential flow effect decreases ap-
proximately exponentially with distance in the flow direction. The effect of the wake decreases more slowly
than the potential effect and downstream of an obstruction the wake is usually more important.
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 29
In a radial compressor the vanes in a vaned diffuser are a powerful source of excitation of the impeller and one
of the principal reasons for mounting the vanes some way out from the impeller is to reduce the strength of
the excitation of the impeller by the static pressure field of the vanes.
The force and moment imparted to a blade as it experiences a non- uniform flow can be calculated in the case
of inviscid two- dimensional flow. To be useful, such calculations need there to be an accurate input of the
flow disturbance. To know whether the excitation will be a problem in turn requires an estimate of the me-
chanical damping.
3.11 The Rigi Test Rig at Turbomachinery Laboratory
The Turbomachinery Laboratory of ETH Zurich runs a centrifugal compressor test rig named after a famous
mountain near Lucerne. In the past and today the test facility provides a solid basis to undertake research in
the field of impeller aerodynamics, system instabilities, aerodynamic probe development and application of
LDA measurement techniques. Currently the test facility is being redesigned aiming to extend its research ca-
pabilities into the field of forced response of radial compressors.
Fig. 3.11.0.1: RIGI Test Rig at Turbomachinery Laboratory ETH Zurich
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
30 00-914-663 Armin Zemp
3.11.1 Test Rig Features
RIGI is a closed loop facility. As such within the test section the pressure is adjusted independently from am-
bient conditions. During operation pressure and temperature at the inlet of the impeller are controlled for
constant values. Herein the facility can be evacuated to 0.25 bar or pressurized to a maximum value of 4 bar.
The maximum deliverable power to the rotating components is limited by the gearbox to 370kW and is pro-
vided by an electrical DC motor. RIGI allows maximum rotational speeds of 22000rpm. The mass flow is con-
trolled through an orifice. Downstream of the impeller a water cooled heat exchanger reduces the fluid
temperature to the required impeller inlet temperature.
Fig. 3.11.1.1 Schematic of RIGI Test Rig Closed Loop
A number of impeller configurations have been investigated in the test facility with typical outer diameters of
280mm and pressure ratios around 2. In a recent research project an impeller was installed measuring
400mm on the outer diameter generating a maximum pressure ratio in the excess of 3. The diffuser section
makes allowance to modify the diffuser configuration.
The impeller used in the measurements has an outer diameter of 400mm containing 7 main blades and 7
splitter blades. The impeller used for the forced vibration analysis has the same number but very thin blades.
In the map in figure 3.11.1.2 the performance of both impeller blades is compared.
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 31
Fig. 3.11.1.2 Performance Map for A8C and A8C41 Impeller (A. Kammerer)
The A8C41 impeller is used for the forced vibration investigations. The blue line in figure 3.11.1.2 is the perfor-
mance of the impeller used in the inlet flow field measurements. As mentioned the thin bladed impeller used
for vibration analysis is populated with several strain gages. The inlet distortion may excite the blades of the
impeller. A characteristic plot for the vibration analysis is the Campbell diagram:
Fig. 3.11.1.3 Campbell Diagram for A8C41 Impeller (A. Kammerer)
The eigenfrequencies of the impeller are plotted against the shaft speed. Where the line of the engine order
cross the line of the eigenfrequency the blades get into forced resonance. In the measurements the distortion
screens to excite the second, third, fourth and fifth engine order are available.
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
32 00-914-663 Armin Zemp
3.12 Measurement Techniques
During operation two separate acquisition systems record data from sensors installed in the facility. On the
one hand performance data is acquired in order to monitor and to control the operating conditions of the fa-
cility and on the other hand data is acquired from a set of fast response measurement techniques i.e. pressure
transducers, the fast aerodynamic probe FRAP or the LDA system. In order to perform on impeller measure-
ments, RIGI is currently being extended by Mr. Albert Kammerer by a rotating transmission system. This will
allow to perform strain gauge and pressure sensor measurements on the impeller surface.
3.13 Measurement Setup
The flow properties are measured with a fast response aerodynamic probe (FRAP). To traverse an area of the
inlet flow cross section a traversing system with the possibility to move the probe in radial as well as in the
circumferential direction is needed. The simplest way of realizing it is to simply put the existing system on a
ring rotating in circumferential direction driven by an additional electrical motor. The following photograph
shows the huge gear and down right the gear of the motor. Top left the tower of the traversing system is vis-
ible. Also the probe stem of the FRAP probe can be seen looking very closely.
Fig. 3.13.0.1 Third Axis of Traversing System
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 33
To ensure the same inlet flow conditions as they will be in Albert Kammerers investigations the whole appa-
ratus therefore is installed. Figure 3.13.0.2 shows a sectional view of the inlet section of the RIGI test rig.
Fig. 3.13.0.2: Sectional View of Test Rig Inlet (A. Kammerer)
Plain 1 represents the location where the distortion will be generated and at position 2 the FRAP probe mea-
sures the inlet flow field. The tube in the center of the inlet channel is used for data transmission in Albert
Kammerers experiments using a slip ring.
3.13.1 Distortion Screens
The distortion is generated by screens with different baffle geometries and different metal mains. The aim is
to control the amplitude of the introduced distortions. The distortion as it passes through the impeller eye will
excite the impeller blades. The layout of the baffles depends on the engine order to be excited. The engineer-
ing drawings show the screen layout for the engine orders two, four and five:
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
34 00-914-663 Armin Zemp
The parameters of the metal mains set the amplitude of the velocity variation over the circumference. The
frame of the screens as well as the calculation of the grid properties was done by Albert Kammerer. With the
required relationship the velocity variation depending on the grid parameters and the flow rate can be pre-
dicted.
Fig. 3.13.1.1: Characteristics for Specified Grid Geometry (A. Kammerer)
The uniform flow field upstream the distortion screen redistributes due to the blockage effect of the distortion
grid. The line named free stream velocity describes the value of the velocity passing the screen without block-
age. Thus the free stream velocity is higher than the velocity of the uniform flow upstream the screen. In con-
trast to the undistorted flow the velocity of the portion passing the grid is lower.
If one knows the corrected mass flow rate the velocity of the flow after the grid and the velocity of the flow
through the undistorted section of the screen can be reed off. The blue curve indicates the pressure drop co-
efficient.
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 35
3.14 Computational Fluid Dynamics
Computational Fluid Dynamics (CFD) is a computer- based tool for simulating the behavior of systems involv-
ing fluid flow, heat transfer and other related physical processes. It works by solving the equations of fluid flow
in a special form over a region of interest, with specified known conditions on the boundary of that region.
3.14.1 History of CFD
Computers have been used to solve fluid flow problems for many years. Numerous programs have been writ-
ten to solve either specific problems, or specific classes of problems. From the mid-1970's, the complex math-
ematics required to generalise the algorithms began to be understood, and general purpose CFD solvers were
developed. These began to appear in the early 1980's and required what were then very powerful computers,
as well as an in- depth knowledge of fluid dynamics, and large amounts of time to set up simulations. Conse-
quently, CFD was a tool used almost exclusively in research. Recent advances in computing power, together
with powerful graphics and interactive 3D- manipulation of models have made the process of creating a CFD
model and analysing results much less labour intensive, reducing time and, hence, cost. Advanced solvers con-
tain algorithms which enable robust solutions of the flow field in a reasonable time. As a result of these fac-
tors, Computational Fluid Dynamics is now an established industrial design tool, helping to reduce design
timescales and improve processes throughout the engineering world. CFD provides a cost- effective and accu-
rate alternative to scale model testing, with variations on the simulation being performed quickly, offering ob-
vious advantages.
3.14.2 Mathematics of CFD
The set of equations which describe the processes of momentum, heat and mass transfer are known as the
Navier- Stokes equations. These partial differential equations were derived in the early nineteenth century
and have no known general analytical solution but can be discretised and solved numerically. Equations de-
scribing other processes, such as combustion, can also be solved in conjunction with the Navier- Stokes equa-
tions. Often, an approximating model is used to derive these additional equations, turbulence models being a
particularly important example. There are a number of different solution methods which are used in CFD
codes. The most common is known as the finite volume technique. In this technique, the region of interest is
divided into small subregions, called control volumes. The equations are discretised and solved iteratively for
each control volume. As a result, an approximation of the value of each variable at specific points throughout
the domain can be obtained. In this way, one derives a full picture of the behaviour of the flow.
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
36 00-914-663 Armin Zemp
3.14.3 CFD Methodology
CFD may be used to determine the performance of a component at the design stage, or it can be used to anal-
yse difficulties with an existing component and lead to its improved design. For example, the pressure drop
through a component may be considered excessive:
Fig. 3.14.3.1 Example Component (ANSYS, 2007)
The first step is to identify the region of interest:
Fig. 3.14.3.2 Region of Interest (ANSYS, 2007)
The geometry of the region of interest is then defined and the mesh is created. After importing the mesh into
the pre- processor, other elements of the simulation including the boundary conditions (inlets, outlets, etc.)
and fluid properties are defined:
Fig. 3.14.3.3 Region of Interst with applied Boundary Conditions (ANSYS, 2007)
The flow solver is run to produce a file of results which contain the variation of velocity, pressure and any other
variables throughout the region of interest. The results can be visualised and can provide an understanding of
the behaviour of the fluid throughout the region of interest:
Fig. 3.14.3.4 Flow Field through Region of Interest (ANSYS, 2007)
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 37
This may lead to design modifications which can be tested by changing the geometry of the CFD model and
seeing the effect. The process of performing a single CFD simulation is split into four components:
1. Geometry / Mesh
2. Physics Definition
3. Solver
4. Post- Processor
Geometry / Mesh:
This interactive process is the first pre- processing stage. The objective is to produce a mesh for input to the
physics pre- processor. Before a mesh can be produced, a closed geometric solid is required. The geometry and
mesh can be created in CAD2Mesh or any of the other geometry/mesh creation tools. The basic steps involve:
1. Defining the geometry of the region of interest
2. Creating regions of fluid flow, solid regions and surface boundary names
3. Setting properties for the mesh
Physics Definition:
This interactive process is the second pre- processing stage and is used to create input required by the Solver.
The mesh files are loaded into the physics pre- processor. The physical models that are to be included in the
simulation are selected. Fluid properties and boundary conditions are specified.
Solver:
The Solver produces the required results in a non- interactive/batch process. The problem is solved as follows:
1. The partial differential equations are integrated over all the control volumes in the region of interest. This is
equivalent to applying a basic conservation law (e.g. for mass or momentum) to each control volume
2. These integral equations are converted to a system of algebraic equations by generating a set of approxi-
mations for the terms in the integral equations
3. The algebraic equations are solved iteratively
An iterative approach is required because of the non- linear nature of the equations, and as the solution ap-
proaches the exact solution, it is said to converge. For each iteration, an error, or residual, is reported as a mea-
sure of the overall conservation of the flow properties. How close the final solution is to the exact solution
depends on a number of factors, including the size and shape of the control volumes and the size of the final
residuals. Complex physical processes, such as combustion and turbulence, are often modeled using empirical
relationships. The approximations inherent in these models also contribute to differences between the CFD
solution and the real flow. The solution process requires no user interaction and is, therefore, usually carried
out as a batch process. The solver produces a results file which is then passed to the post- processor.
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
38 00-914-663 Armin Zemp
Post- Processor:
The post- processor is the component used to analyse, visualise and present the results interactively. Post- pro-
cessing includes anything from obtaining point values to complex animated sequences. Examples of some
important features of post- processors are:
- Visualisation of the geometry and control volumes
- Vector plots showing the direction and magnitude of the flow
- Visualisation of the variation of scalar variables (variables which have only magnitude, not direction, such as
temperature, pressure, ...) through the domain
- Quantitative numerical calculations
- Animation Charts showing graphical plots of variables
- Hardcopy output
3.14.4 Governing Equations
The software in use for the present work is the commercial ANSYS CFX 10.0 SP1 software containing the pre-
processor, the solver and a post- processing tool. The governing equations are implemented in the code as
they are described on the following pages.
Transport Equations:
The instantaneous equations of mass, momentum and energy conservation can be written as follows in a sta-
tionary frame of reference:
Continuity:
Equation 3.14.4.1
Momentum Equations:
Equation 3.14.4.2
where the stress tensor, , is related to the strain rate by
Equation 3.14.4.3
Total Energy Equation:
Equation 3.14.4.4
where is the total Enthalpy. The term represents the work due to viscous stresses and is called the
viscous work term. The term represents the work due to external momentum sources and is currently
neglected.

t
------ U ( ) + 0 =
U ( )
t
--------------- U U ( ) + p SM + + =

U U ( )
T
2
3
--- U +
\
| |
=
h
0
( )
t
-----------------
p
t
----- Uh
0
( ) + T ( ) U ( ) U SM SE + + + =
h
0
U ( )
U SM
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 39
Equations of State:
The flow solver calculates pressure and static enthalpy. Finding density requires that we select the thermal
equation of state and finding temperature requires that we select the constitutive relation. The selection of
these two relationships is not necessarily independent and is also a modelling choice. The thermal equation
of state is described as a function of both temperature and pressure:
Equation 3.14.4.5
The specific heat capacity, , may also be described as a function of temperature and pressure:
Equation 3.14.4.6
For an ideal gas, the density is defined by the ideal gas law and, in this case, can be a function of only tem-
perature:
Equation 3.14.4.7
Ideal Gas Equation of State:
For an ideal gas, the relationship is described by the ideal gas law:
Equation 3.14.4.8
where is the molecular weight of the gas and is the universal gas constant.
3.14.5 Turbulence Models
Turbulence consists of fluctuations in the flow field in time and space. It is a complex process, mainly because
it is three dimensional, unsteady and consists of many scales. It can have a significant effect on the character-
istics of the flow. Turbulence occurs when the inertia forces in the fluid become significant compared to vis-
cous forces, and is characterised by a high Reynolds Number. In principle, the Navier- Stokes equations
describe both laminar and turbulent flows without the need for additional information. However, turbulent
flows at realistic Reynolds numbers span a large range of turbulent length and time scales, and would gener-
ally involve length scales much smaller than the smallest finite volume mesh, which can be practically used in
a numerical analysis. The Direct Numerical Simulation (DNS) of these flows would require computing power
which is many orders of magnitude higher than available in the foreseeable future.
p T , ( ) =
c
p
c
p
c
p
p T , ( ) =
c
p
c
p
c
p
T ( ) =

w p p
ref
+ ( )
R
0
T
--------------------------- =
w R
0
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
40 00-914-663 Armin Zemp
To enable the effects of turbulence to be predicted, a large amount of CFD research has concentrated on meth-
ods which make use of turbulence models. Turbulence models have been specifically developed to account for
the effects of turbulence without recourse to a prohibitively fine mesh and Direct Numerical Simulation. Most
turbulence models are statistical turbulence model. The two exceptions to this are the Large Eddy Simulation
model and the Detached Eddy Simulation model.
Statistical Turbulence Models and the Closure Problem:
When looking at time scales much larger than the time scales of turbulent fluctuations, turbulent flow could
be said to exhibit average characteristics, with an additional time- varying, fluctuating component. For exam-
ple, a velocity component may be divided into an average component and a time varying component. In gen-
eral, turbulence models seek to modify the original unsteady Navier- Stokes equations by the introduction of
averaged and fluctuating quantities to produce the Reynolds Averaged Navier- Stokes (RANS) equations.
These equations represent the mean flow quantities only, while modelling turbulence effects without a need
for the resolution of the turbulent fluctuations. All scales of the turbulence field are being modelled. Turbu-
lence models based on the RANS equations are known as Statistical Turbulence Models due to the statistical
averaging procedure employed to obtain the equations. Simulation of the RANS equations greatly reduces the
computational effort compared to a Direct Numerical Simulation and is generally adopted for practical engi-
neering calculations. However, the averaging procedure introduces additional unknown terms containing
products of the fluctuating quantities, which act like additional stresses in the fluid. These terms, called tur-
bulent' or Reynolds' stresses, are difficult to determine directly and so become further unknowns. The Rey-
nolds (turbulent) stresses need to be modelled by additional equations of known quantities in order to achieve
closure. Closure implies that there is a sufficient number of equations for all the unknowns, including the
Reynolds- Stress tensor resulting from the averaging procedure. The equations used to close the system define
the type of turbulence model.
Reynolds Averaged Navier Stokes (RANS) Equations:
As described above, turbulence models seek to solve a modified set of transport equations by introducing av-
eraged and fluctuating components. For example, a velocity may be divided into an average component and
a time varying component:
Equation 3.14.5.1
The averaged component is given by:
Equation 3.14.5.2
where is a time scale that is large relative to the turbulent fluctuations, but small relative to the time scale
to which the equations are solved.
U
U t ( ) U u' t ( ) + =
U
1
t
----- U t ( ) t d
t
t t + ( )

=
t
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 41
Substituting the time averaged quantities into the original transport equations results in the Reynolds- aver-
aged equations given below:
Continuity:
Equation 3.14.5.3
Momentum:
Equation 3.14.5.4
Scalar Transport:
Equation 3.14.5.5
with as a non reacting scalar.
The continuity equation has not been altered but the momentum and scalar transport equations contain tur-
bulent flux terms additional to the molecular diffusive fluxes. These are the Reynolds stress, , and the
Reynolds flux, . These terms arise from the non- linear convective term in the un- averaged equations.
They reflect the fact that convective transport due to turbulent velocity fluctuations will act to enhance mix-
ing over and above that caused by thermal fluctuations at the molecular level. At high Reynolds numbers, tur-
bulent velocity fluctuations occur over a length scale much larger than the mean free path of thermal
fluctuations, so that the turbulent fluxes are much larger than the molecular fluxes.
The Reynolds- averaged energy equation is:
Equation 3.14.5.6
where the mean total enthalpy is given by:
Equation 3.14.5.7
In addition to the mean flow kinetic energy, the total enthalpy now contains a contribution from the turbulent
kinetic energy, , given by:
Equation 3.14.5.8
Turbulence models close the Reynolds- averaged equations by providing models for the computation of the
Reynolds stresses and Reynolds fluxes.

t
------ U ( ) + 0 =
U ( )
t
--------------- U U ( ) + u' u' ( ) SM + =
( )
t
-------------- U ( ) + u' ( ) SE + =

u' u'
u'
h
0
( )
t
----------------- Uh
0
u' h T + ( ) +
p
t
----- =
h
0
h
1
2
---U
2
k + + =
k
k
1
2
---u'
2
=
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
42 00-914-663 Armin Zemp
Eddy Viscosity Turbulence Models:
One proposal suggests that turbulence consists of small eddies which are continuously forming and dissipat-
ing, and in which the Reynolds stresses are assumed to be proportional to mean velocity gradients. This de-
fines an eddy viscosity model.' The eddy viscosity hypothesis assumes that the Reynolds stresses can be
related to the mean velocity gradients and Eddy (turbulent) Viscosity by the gradient diffusion hypothesis, in
a manner analogous to the relationship between the stress and strain tensors in laminar Newtonian flow:
Equation 3.14.5.9
here, is the Eddy Viscosity or Turbulent Viscosity. This has to be prescribed. Analogous to the eddy viscosity
hypothesis is the eddy diffusivity hypothesis, which states that the Reynolds fluxes of a scalar are linearly re-
lated to the mean scalar gradient:
Equation 3.14.5.10
here, is the Eddy Diffusivity and has to be prescribed. The Eddy Diffusivity can be written as:
Equation 3.14.5.11
where is the turbulent Prandtl number. Eddy diffusivities are then prescribed using the turbulent Prandtl
number.
The above equations can only express the turbulent fluctuation terms of functions of the mean variables if the
turbulent viscosity is known. Both the and two- equation turbulence models provide this variable.
Subject to these hypotheses, the Reynolds averaged momentum and scalar transport equations become:
Equation 3.14.5.12
Equation 3.14.5.13
where is the sum of the body forces, is the effective viscosity and is the effective diffusivity, defined
by:
Equation 3.14.5.14
u' u'
2
3
---k
2
3
---
t
U
t
U U ( )
T
+ ( ) + =

t
u'
t
=

t
Pr
t
------- =
Pr
t
k k
U ( )
t
--------------- U U ( ) + B p'
ef f
U U ( )
T
+ ( ) ( ) + =
( )
t
-------------- U
ef f
( ) + S =
B
ef f

ef f

ef f

t
+ =
Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background
Armin Zemp 00-914-663 43
and:
Equation 3.14.5.15
is a modified pressure, defined by:
Equation 3.14.5.16
where is the bulk viscosity.
The Reynolds averaged energy equation becomes:
Equation 3.14.5.17
It has to be remarked that although the transformation of the molecular diffusion term may be inexact if en-
thalpy depends on variables other than temperature, the turbulent diffusion term is correct, subject to the
eddy diffusivity hypothesis. Moreover, as turbulent diffusion is usually much larger than molecular diffusion,
small errors in the latter can be ignored.
Two Equation Turbulence Models:
Two- equation turbulence models are very widely used, as they offer a good compromise between numerical
effort and computational accuracy. Two- equation models are much more sophisticated than the zero equa-
tion models. Both the velocity and length scale are solved using separate transport equations (hence the term
two- equation'). The and two- equation models use the gradient diffusion hypothesis to relate the
Reynolds stresses to the mean velocity gradients and the turbulent viscosity. The turbulent viscosity is mod-
elled as the product of a turbulent velocity and turbulent length scale. In two- equation models, the turbu-
lence velocity scale is computed from the turbulent kinetic energy, which is provided from the solution of its
transport equation. The turbulent length scale is estimated from two properties of the turbulence field, usu-
ally the turbulent kinetic energy and its dissipation rate. The dissipation rate of the turbulent kinetic energy is
provided from the solution of its transport equation.
In all the solved cases in the present work, the two- equation model is used due to the fact that the other
turbulence models (namely and ) did not result in a stable convergence history.

ef f

t
+ =
p'
p' p
2
3
---k U
2
3
---
ef f

\
| |
+ + =

h
0
( )
t
-----------------
p
t
----- Uh
0
( ) + T

t
Pr
t
------- h +
\
| |
SE + =
k k
k
k SST
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
44 00-914-663 Armin Zemp
The Turbulence Model:
is the turbulence kinetic energy and is defined as the variance of the fluctuations in velocity. It has dimen-
sions of ( ). is the turbulence eddy dissipation (the rate at which the velocity fluctuations dissipate), and
has dimensions of per unit time ( ). The model introduces two new variables into the system of
equations.
The continuity equation is then:
Equation 3.14.5.18
and the momentum equation becomes:
Equation 3.14.5.19
where is the sum of body forces, is the effective viscosity accounting for turbulence and is the modi-
fied pressure given by:
Equation 3.14.5.20
The model is based on the eddy viscosity concept, so that:
Equation 3.14.5.21
where is the turbulence viscosity. The model assumes that the turbulence viscosity is linked to the tur-
bulence kinetic energy and dissipation via the relation:
Equation 3.14.5.22
where is a constant.
The values of and come directly from the differential transport equations for the turbulence kinetic energy
and turbulence dissipation rate:
Equation 3.14.5.23
Equation 3.14.5.24
where , , and are constants.
k
k
L
2
T
2

k L
2
T
3
k

t
------ U ( ) + 0 =
U ( )
t
--------------- U U ( )
ef f
U ( ) + p'
ef f
U ( )
T
B + + =
B
ef f
p'
p' p
2
3
---k + =
k

ef f

t
+ =

t
k

t
C

k
2

----- =
C

k
k ( )
t
-------------- Uk ( ) +

k
----- +
\
| |
k P
k
+ =
( )
t
-------------- U ( ) +

----- +
\
| |


k
-- C
1
P
k
C
2
( ) + =
C
1
C
2

k

Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background

Armin Zemp 00-914-663 45
is the turbulence production due to viscous and buoyancy forces, which is modelled using:
Equation 3.14.5.25
Modelling Flow Near the Wall:
The wall distance is used in various functions that control the transition between near- wall and free- stream
models. The wall- function approach in the commercial code used for the present work, ANSYS CFX, is an ex-
tension of the method of Launder and Spalding (Launder et al., 1974). In the log- law region, the near wall tan-
gential velocity is related to the wall- shear- stress, , by means of a logarithmic relation. In the wall- function
approach, the viscosity affected sublayer region is bridged by employing empirical formulas to provide near-
wall boundary conditions for the mean flow and turbulence transport equations. These formulas connect the
wall conditions (e.g., the wall- shear- stress) to the dependent variables at the near- wall mesh node which is
presumed to lie in the fully- turbulent region of the boundary layer. The logarithmic relation for the near wall
velocity is given by:
Equation 3.14.5.26
where:
Equation 3.14.5.27
and:
Equation 3.14.5.28
is the near wall velocity, is the friction velocity, is the known velocity tangent to the wall at a distance
of from the wall, is the dimensionless distance from the wall, is the wall shear stress, is the von
Karman constant and is a log- layer constant depending on wall roughness.
The equation for the near wall velocity has the problem that it becomes singular at separation points where
the tangential velocity approaches zero. In the logarithmic region, an alternative velocity scale, can be used
instead of :
Equation 3.14.5.29
This scale has the useful property that it does not go to zero if goes to zero. In turbulent flow is never com-
pletely zero. Based on this definition, the following explicit equation for can be obtained:
Equation 3.14.5.30
P
k
P
k

t
U U U
T
+ ( )
2
3
--- U 3
t
U k + ( ) P
k0
+ =

u
+
U
t
u

-----
1

--- y
+
( ) C + ln = =
y
+
yu

--------------- =
u

------
\
| |
1 2
=
u
+
u

U
t
y y
+

C
u
+
u

u
+
u

1 4
k
1 2
=
U
t
k
u

U
t
1

--- y

( ) C + ln
------------------------------ =
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
46 00-914-663 Armin Zemp
The absolute value of the wall shear stress is then obtained from:
Equation 3.14.5.31
where:
Equation 3.14.5.32
One of the major drawbacks of the wall- function approach is that the predictions depend on the location of
the point nearest to the wall and are sensitive to the near- wall meshing; refining the mesh does not neces-
sarily give a unique solution of increasing accuracy (Grotjans et al., 1998). The problem of inconsistencies in
the wall- function, in the case of fine meshes, can be overcome with the use of the Scalable Wall Function for-
mulation developed by CFX. It can be applied on arbitrarily fine meshes and allows to perform a consistent
mesh refinement independent of the Reynolds number of the application. The basic idea behind the scalable
wall- function approach is to limit the value used in the logarithmic formulation by a lower value of
. 11.06 is the intersection between the logarithmic and the linear near wall profile. The com-
puted is therefore not allowed to fall below this limit. Therefore, all mesh points are outside the viscous
sublayer and all fine mesh inconsistencies are avoided.
Numerical Discretisation:
Analytical solutions to the Navier Stokes equations exist for only the simplest of flows under ideal conditions.
To obtain solutions for real flows, a numerical approach must be adopted whereby the equations are replaced
by algebraic approximations which may be solved using a numerical method.
Segregated solvers employ a solution strategy where the momentum equations are first solved, using a
guessed pressure, and an equation for a pressure correction is obtained. Because of the guess- and- correct'
nature of the linear system, a large number of iterations are typically required in addition to the need for judi-
ciously selecting relaxation parameters for the variables. ANSYS CFX uses a coupled solver, which solves the
hydrodynamic equations as a single system. This solution approach uses a fully implicit discretisation of the
equations at any given time step. For steady state problems, the time- step behaves like an acceleration pa-
rameter', to guide the approximate solutions in a physically based manner to a steady- state solution. This re-
duces the number of iterations required for convergence to a steady state, or to calculate the solution for each
time step in a time dependent analysis.
All numerical approximation schemes are prone to a degree of error. Some errors are a result of truncation of
additional terms in series expansions. Others are a result of the order of the differencing scheme used for the
approximation.

=
y

---------------- =
y

max y

11.06 , ( ) =
y

Inlet Flow Distortion in a Centrifugal Compressor Theoretical Background

Armin Zemp 00-914-663 47
Numerical Diffusion:
Numerical diffusion is usually exhibited by difference equations where the advection term has been approxi-
mated using an odd- order scheme which is first order accurate. The effect of this over a whole flow domain
is that the features of the flow are smeared out. This phenomenon is sometimes called gradient smearing.'
The effect varies according to the alignment of the mesh with the flow direction. It is therefore relatively
straightforward to achieve highly accurate solutions to simple flow problems, such as flow in a duct where
alignment of the mesh with the predominant flow is relatively simple. However, for situations in which the
flow is predominantly not aligned with the mesh, numerical diffusion effects limit the accuracy of the solu-
tion.
Numerical Dispersion:
Numerical dispersion is usually exhibited by discretised equations whose advection term has been approxi-
mated using schemes that are even- order accurate. Dispersion results in oscillations or wiggles' in the solu-
tion particularly where there are steep flow gradients (e.g., across a shock).
Theoretical Background Inlet Flow Distortion in a Centrifugal Compressor
48 00-914-663 Armin Zemp
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 49
4. Results
4.1 Stationary CFD
The stationary CFD investigation of the centrifugal compressor provides a basis to validate the numerical
model via comparison with the measured performance of the test rig. In addition to the performance of the
machine (pressure ratio versus mass flow), the total pressure distribution and the absolute flow angle profile
from hub to shroud in the diffuser section of the compressor provides the possibility for a detailed validation
of the model. The simulation over a broad range of operating points allows a characterization of the flow
through the impeller.
4.1.1 Grid
The grid used for the CFD investigation consists of one main blade and one splitter blade representing the sev-
enth part of the entire impeller. The grid is from ABB Turbo Systems made by Dr. Matthias Schleer and can es-
sentially be divided in four subdomains shown in the figure below:
Fig. 4.1.1.1: CFD Grid (Seventh Part of Impeller)
The illustrated part of the grid consists of 571600 nodes building 509184 elements (hexahedra) and corre-
sponds to the geometry of the A8C41 impeller used for the strain gage measurements of Albert Kammerer.
The values are in the required range for the solver used in the present work. y
+
Results Inlet Flow Distortion in a Centrifugal Compressor
50 00-914-663 Armin Zemp
The number and distribution of nodes and associated elements leads to the required grid quality showed in
the following illustration:
Fig. 4.1.1.2: Grid Quality Check (ANSYS ICEM CFD, Dr. M. Schleer)
Skewness:
This histogram is based upon calculations of the maximum skewness of a hexahedral or quadrilateral ele-
ment. The skewness is defined differently for volume and surface elements. For a volume element, it is ob-
tained by taking all pairs of adjacent faces and computing the normals.
Aspect ratio:
For hexahedral and quadrilateral elements, the aspect ratio is defined as the ratio of the distances between
diagonally opposite vertices (shorter diagonal/longer diagonal). For tetrahedral elements, MED calculates the
ratio between the radii of an inscribed sphere to a circumscribed sphere for each element. For triangular ele-
ments, this operation is done using circles.
4.1.2 Boundary Conditions
The common setup for the boundary conditions leading to a stable convergence of the solution is to set the
total pressure and total temperature at the inlet of the model and the static pressure at the outlet. Both the
total pressure and total temperature are available from performance measurements of the test rig at the cor-
responding operating point and are held constant by the system during the measurement campaign.
The outlet static pressure is adjusted to match the measured mass flow due to the fact that the outlet static
pressure is not measured at the same location in the test rig compared to the location of the outlet in the com-
putational domain.
An additional remark concerns the diffuser part of the centrifugal compressor. As illustrated, the model con-
sists of four subdomains. The diffuser is not implemented as separate domain. To make allowance to zero wall
velocity of the diffuser part a velocity contrary to the rotational direction of the impeller is impressed on each
node modelling the diffuser wall.
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 51
Boundary Conditions for the stationary CFD investigation:
Inlet: constant total pressure and total temperature (performance measurement)
Outlet: static pressure (average static pressure, varied until mass flow matches measurement)
Shroud: counter rotating wall, no slip, adiabatic
Impeller: constant rotational speed, no slip, adiabatic
Fluid Boundaries: periodic interfaces (rotational)
4.1.3 Turbulence Model and Solver Setup
The software used for the CFD investigation is the commercial ANSYS CFX software. The ANSYS Solver is a cou-
pled solver which solves the hydrodynamic equations as a single system. This solution approach uses a fully
implicit discretisation of the equations at any time step. For stationary problems the time step behaves like an
acceleration parameter to guide the approximate solutions in a physically based manner to a steady state so-
lution.
Several turbulence models (namely , and ) have been applied to compare the stability of the
solver process. Only the turbulence model resulted in a stable solution of the set of equations. Both the
and turbulence model showed bad convergence compared to the model and strong recircula-
tion (up to 15% of the outlet area) at the outlet of the domain. An explanation of the stable solution with the
model might be the diffusive characteristic of this turbulence model. For a detailed description of the
turbulence model the reader is referred to chapter 3.14.5.
For the steady state investigations a constant turbulence intensity of 5% which is a reasonable value for nom-
inal turbulence through a circular inlet is applied as inlet boundary condition. Comparison of the effect of dif-
ferent turbulence intensities (1%, 5%, 10% and 15%) at the inlet on the resulting solution showed negligible
importance for the stationary CFD investigation.
k k SST
k
k SST k
k
k
Results Inlet Flow Distortion in a Centrifugal Compressor
52 00-914-663 Armin Zemp
4.1.4 Convergence Quantification
Solver Residual:
The residual is a measure of the local imbalance of each conservative control volume equation. It is the most
important measure of convergence as it relates directly to whether the equations have been solved. The resid-
ual is calculated using only the spatial flux terms and essentially represents a discrete conservation balance.
The following residual plot representatively shows the convergence history of the stationary simulations:
Fig. 4.1.4.1: RMS Residuals
According to the CFX Solver Guidelines a decrease of the RMS residuals of the main kinematic properties of at
least three orders of magnitudes is required.
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 53
A second information about the solution process is the value and location of the maximum residual. This in-
formation provides the place in the computational domain where the set of equation causes numerical insta-
bilities. The plot of the maximum residuals of the main kinematic properties below indicates what one
expects. The values of the maximum residuals are, as a matter of course, higher compared to the root mean
square values. A decrease of the maximum residuals of the order of about three orders of magnitudes implies
acceptable convergence.
Fig. 4.1.4.2: Maximum Residuals
The location of the maximum residuals is illustrated below. It is, unsurprisingly, at the exit of the domain in
the diffuser part of the model:
Fig. 4.1.4.3: Location of MAX residual
Results Inlet Flow Distortion in a Centrifugal Compressor
54 00-914-663 Armin Zemp
It is not a surprise due to the fact that the boundary surface is part of a rotating domain. Therefore the whole
set of equations is solved in the relative frame of reference leading to extremely small angles between the rel-
ative flow direction and the surface of the domain boundary:
Fig. 4.1.4.4: Streamlines in Relative Frame of Reference (50% Span)
To avoid numerical instabilities at the outlet of the impeller an additional domain would be helpful. With such
a domain the diffuser section of the compressor is then modeled and implemented in the stationary frame of
reference to avoid small angles causing numerical uncertainty.
Mass Imbalance:
A very important information about convergence of the numerical solution is the mass imbalance between
inlet and outlet of the domain. In industrial environment a mass imbalance of more than 0.1% is inacceptable.
The mass imbalance of the stationary solutions in the present work is representatively illustrated below and
is of the order of :
Fig. 4.1.4.5: Mass Flow Imbalance
10
6
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 55
4.1.5 Validation
A common method to illustrate the performance of a compressor is the well known compressor map. Plotting
the pressure ratio versus mass flow for both the test rig and the computational model allows a validation of
the CFD results:
Fig. 4.1.5.1: Compressor Map
The compressor map above shows a good match of the numerical results compared to the measurements
over the whole operating range. The values along the resistance line are to handle with care due to the fact
that the CFD model doesnt reproduce the whole system.
A short explanation might be necessary concerning the pressure ratio in the compressor map. The schemat-
ic below explains where the different pressures are measured in the rig and where the computational domain
ends:
Fig. 4.1.5.2: Schematic of RIGI Test Rig

Results Inlet Flow Distortion in a Centrifugal Compressor

56 00-914-663 Armin Zemp
It is obvious that the total pressure ratios of test rig and CFD model can not be compared because of the dif-
ferent location of the exit total pressures (CFD) and (performance measurement of the rig). To get a
proper pressure ratio to compare the two systems the values of static pressure, , at radius 250mm, are used
to build the pressure ratios both for the measurements and the CFD results.
The total pressure ratio of the performance measurement of the test rig is scaled as follows:
Equation 4.1.5.3
The same procedure is applied to the total pressure ratio of the CFD model:
Equation 4.1.5.4
The measurement of static wall pressure was done by Matthias Schleer. Values are available for several
stage Mach numbers but not for Mu=1.26 and Mu=1.33. By plotting the pressure correction coefficient
for both the measurements and the CFD results an estimate can be done for Mu=1.26 and Mu=1.33:
Fig. 4.1.5.5: Pressure Correction Coefficient
Due to the almost linear trend the values for Mu=1.26 and Mu=1.33 are extrapolated and assumed to be a good
approximation.
The reference operating point is set at a corrected mass flow of Q=82.5 at Mu=1.33. The corresponding pressure
ratio is 2.85.
p
03
p
04
p
2

p
04
p
01
-------
\
| |
p
2
p
04
-------
\
| |

p
2
p
01
-------
\
| |
measurement
=
p
03
p
01
-------
\
| |
p
2
p
03
-------
\
| |

p
2
p
01
-------
\
| |
CFD
=
p
2
p
2
p
0 exi t ,
---------------
\
| |

ref
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 57
Hub to Shroud Profiles:
The accomplishment of FRAP measurements in the diffuser section of the test rig at radius 210mm allows a
comparison of the hub to shroud profiles for both the total pressure and absolute flow angle and offers a
detailed validation of the numerical results. The operating points which are used to compare the profiles are
indicated in the map below:
Fig. 4.1.5.6: Compressor Map: FRAP measurement points in Diffuser
The attentive reader might ask why there are no points to compare the profiles above a stage Mach number
of Mu= 0.8. The reason is the missing probe calibration to evaluate the measured data at Mach numbers of
about 0.8 at the measurement location. Furthermore the measurements have to be handled carefully at an
operating point with a stage Mach number of Mu=1.o. The highly fluctuating flow conditions and tempera-
tures of about 90C which influence accuracy of the measurement.

2
Results Inlet Flow Distortion in a Centrifugal Compressor
58 00-914-663 Armin Zemp
The absolute total pressure ratio profiles for the indicated operating points confirm the consistence in perfor-
mance of the numerical model with the test rig:
Fig. 4.1.5.7: Total Pressure Profiles at R=210mm
For the absolute flow angle comparison the definition of the angle is explained below:
Fig. 4.1.5.8: Definition of Absolute Flow Angle at Exit
FRAP 1: Mu=0.6, Q/Q
ref
=0.27 FRAP 3: Mu=0.8, Q/Q
ref
=0.40
FRAP 2: Mu=0.6, Q/Q
ref
=0.50 FRAP 4: Mu=0.8, Q/Q
ref
=0.58
shroud
hub

2
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 59
Fig. 4.1.5.9: Absolute Flow Angle Profiles at R=210mm
The profiles, both the total pressure profile as well as the absolute flow angle distribution, show good consis-
tence of the CFD results compared to the measurements. The relative deviation of the total pressure profile of
the order of 2% corresponds to the difference in performance (compressor map).
FRAP 1: Mu=0.6, Q/Q
ref
=0.27 FRAP 3: Mu=0.8, Q/Q
ref
=0.40
FRAP 2: Mu=0.6, Q/Q
ref
=0.50 FRAP 4: Mu=0.8, Q/Q
ref
=0.58
shroud
hub
Results Inlet Flow Distortion in a Centrifugal Compressor
60 00-914-663 Armin Zemp
For both the stage Mach numbers Mu=0.6 and Mu=0.8 the total pressure ratio profiles at radius R=210mm
over one pitch (main blade to main blade) are compared with FRAP measurements:
Fig. 4.1.5.10: Total Pressure Ratio at R=210mm (Mu=0.6, Q=29)
Fig. 4.1.5.11: Total Pressure Ratio at R=210mm (Mu=0.8, Q=45)
The contours show the total pressure ratio on a plane at constant radius in the diffuser part of the compressor.
On the left the main blade jet is clearly visible. The jet region of the splitter blade can be identified at the right.
The low total pressure ratios at the bottom of the contours show the boundary layer at the shroud.
The total pressure ratios show good consistence between CFD model and FRAP measurement. A major differ-
ence in both CFD profiles compared to the measured values is the vortical structure caused by the tip leakage
fluid flow. The regions with very low total pressure (blue zones) develop stronger in the CFD contour compared
to the FRAP data. On the one hand it might be that the numerical solution overpredicts the total pressure loss
in the tip leakage vortex. On the other hand it is possible that the FRAP measurement underpredicts the total
pressure loss in such highly fluctuating regions due to the strong variation in the fluid flow angle.
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 61
The detailed validation of the CFD results on the basis of steady state computations is required for the appli-
cation of the model in transient investigations where measurement data is not available to compare the per-
formance with the test rig behavior.
4.1.6 Flow through Impeller
Fig. 4.1.6.1: Streamlines through Impeller (Velocity in rel. Frame of Reference)
The intention of this chapter is to characterize the flow through the centrifugal impeller and to compare sev-
eral parameters at different operating points. As representatively illustrated in the figure above the flow
through a centrifugal compressor is highly complex and it is therefore difficult to present an all- embracing
characterization of the fluid dynamic phenomena taking place in such a device. In a first part the focus of the
description concentrates on the compressor blade as it is one of the principal items of the impeller. In a second
part the flow field is visualized by absolute and relative Mach numbers. A separate view addresses the tip leak-
age vortex.
Results Inlet Flow Distortion in a Centrifugal Compressor
62 00-914-663 Armin Zemp
Pressure Distribution on Main Blade Surface:
The figures below show the variation of the blade pressure distribution for both the pressure and suction side
at a constant mass flow but for different impeller speeds. The pressures are nondimensionalized by the inlet
static pressure :
Fig. 4.1.6.2: Pressure Distribution on Impeller Blade
Q/Q
ref
=0.93 pressure side suction side
Mu=1.10
Mu=1.20
Mu=1.26
Mu=1.33
p
1
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 63
The comparison shows the effect of incidence at the leading edge of the blades. By increasing the blade speed
at a constant mass flow the incidence increases. Therefore the region of low static pressure coefficient visible
on the suction side of the mainblade travels towards leading edge. The loading increases with rising blade
speed. To get a more detailed picture of the static pressure coefficient the values at mid height of the main
blade from leading to trailing edge are compared in the following plot:
Fig. 4.1.6.3: Static Pressure Coefficient at 50% Span of Mainblade (Q/Q
ref
=0.93)
In order to prove the statement that the loading increases with higher blade speed the loading coefficient
according to Schleer (Schleer et al., 2003) is plotted, again at mid height on the main blade
from leading to trailing edge:
Fig. 4.1.6.4: Blade Loading Coefficient at 50% Span of Mainblade (Q/Q
ref
=0.93)
C
n
p
PS
p
SS
( )
w
2
mean
--------------------------- =
Results Inlet Flow Distortion in a Centrifugal Compressor
64 00-914-663 Armin Zemp
An other approach to characterize the pressure distribution is to compare operating points along constant
speed lines in the compressor map:
Mu=0.60:
Pressure Distribution:
Fig. 4.1.6.5: Pressure Distribution on Mainblade at 50% Span (Mu=0.60)
Loading:
Fig. 4.1.6.6: Loading Coefficient on Mainblade at 50% Span (Mu=0.60)
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 65
Mu=0.80:
Pressure Distribution:
Fig. 4.1.6.7: Pressure Distribution on Mainblade at 50% Span
Loading:
Fig. 4.1.6.8: Loading Coefficient on Mainblade at 50% Span (Mu=0.80)
Results Inlet Flow Distortion in a Centrifugal Compressor
66 00-914-663 Armin Zemp
Mu=1.20:
Pressure Distribution:
Fig. 4.1.6.9: Pressure Distribution on Mainblade at 50% Span (Mu=1.20)
Loading:
Fig. 4.1.6.10: Loading Coefficient on Mainblade at 50% Span (Mu=1.20)
For all the tree speed lines the pressure coefficient drops with increasing mass flow, especially on the suction
side of the mainblade in the rear part. The loading coefficient therefore increases for higher mass flow rates.
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 67
If we now plot the inlet flow angle and compare the value with the metal angle of the mainblade one can see
that for most of the corrected mass flow range there is negative incidence, excepting corrected mass flow
rates below about Q=64, which reduces the risk of separation on the suction side:
Fig. 4.1.6.11: Absolute Inlet Flow Angle (Mu=1.20)
An important information for the currently planned measurements of the time resolved blade pressure with
pressure transducers on the mainblade of the impeller in the RIGI test rig is the pressure gradient. The decision
for the positioning of the transducers might be affected by the local pressure gradient on the blade surface.
The following curves representatively show the pressure gradient in streamwise direction on the mainblade
surface at 50% span from leading to trailing edge for a stage Mach number of Mu=1.00 and a mass flow rate
of Q/Q
ref
=0.75:
pressure side:
Fig. 4.1.6.12: Pressure Gradient on Pressure Side of Mainblade at 50% Span
Results Inlet Flow Distortion in a Centrifugal Compressor
68 00-914-663 Armin Zemp
suction side:
Fig. 4.1.6.13: Pressure Gradient on Suction Side of Mainblade at 50% Span
The run of the curves show that one is almost free concerning the choice of location for the pressure probes
along the blade surface (except regions near the leading and trailing edge) due to the nearly constant gradient
in streamwise direction.
Mach Numbers:
In order to get a general description of the flow conditions through the impeller the Mach numbers, both the
absolute and relative, are plotted on a surface at constant 50% span. In the diffuser section a plane at constant
radius of R=210mm is displayed to indicate the Mach numbers to determine the required calibration range for
the probes for intended FRAP measurements in the near future in the diffuser section.
Mu=0.60, Q/Q
ref
=0.36:
Ma
abs
: Ma
rel
:
Fig. 4.1.6.14: Absolute Ma (Mu=0.60, Q/Q
ref
=0.36) Fig. 4.1.6.15: Relative Ma (Mu=0.60, Q/Q
ref
=0.36)
p
s
-----
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 69
Mu=0.80, Q/Q
ref
=0.53:
Mu=1.00, Q/Q
ref
=0.79:
Ma
abs
: Ma
rel
:
Fig. 4.1.6.16: Absolute Ma (Mu=0.80, Q/Q
ref
=0.53) Fig. 4.1.6.17: Relative Ma (Mu=0.80, Q/Q
ref
=0.53)
Ma
abs
: Ma
rel
:
Fig. 4.1.6.18: Absolute Ma (Mu=1.00, Q/Q
ref
=0.79) Fig. 4.1.6.19: Relative Ma (Mu=1.00, Q/Q
ref
=0.79)
Results Inlet Flow Distortion in a Centrifugal Compressor
70 00-914-663 Armin Zemp
Mu=1.10, Q/Q
ref
=0.93:
The table below shows the mass averaged quantities at the R210 surface:
Due to the fact that the calibration facility for the FRAP probes is limited to a Mach number calibration of 0.7
measurements in the diffuser section can only be done up to a stage Mach number of about Mu=0.90. The
temperature limit of the probes lies in the region of about 390K to 400K and is therefore not the limiting fac-
tor.
A noticeable feature visible in all plots for the Mach number in the relative frame of reference is a zone of lower
relative Mach number at the exit of the blading. The origin of this feature can be found by tracking the tip leak-
age vortex.
Ma
abs
: Ma
rel
:
Fig. 4.1.6.20: Absolute Ma (Mu=1.10, Q/Q
ref
=0.93) Fig. 4.1.6.21: Relative Ma (Mu=1.10, Q/Q
ref
=0.93)

Mu=1.10, Q/Q
ref
=0.93 0.835 401.62 K 46.08
Mu=1.00, Q/Q
ref
=0.79 0.791 364.14 K 44.65
Mu=0.80, Q/Q
ref
=0.53 0.597 350.68 K 44.03
Mu=0.60, Q/Q
ref
=0.36 0.448 328.06 K 42.34
Ma
abs
T
0

2
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 71
Tip Leakage Vortex:
Fig. 4.1.6.22: Tip Gap Vortex (Streamlines, Mu=1.33, Q/Q
ref
=1.0)
Corresponding to Schleer (Schleer, 2006) the tip leakage vortex has a dominating effect on the formation of
secondary flows in the diffuser. Instead of the classical jet- wake pattern the proposed 3- zone flow model
holds for the investigated impeller with a tip clearance ratio of about 4.1%.
A series of illustrations characterizes in a qualitative point of view the modification of the vortical structure
for a stage Mach number of Mu=1.33 at different mass flow rates:
Fig. 4.1.6.23: Tip Gap Vortex (Streamlines, relative Total Pressure Ratio, Mu=1.33)
The location of the tip leakage vortex core does not change significantly by varying the mass flow rate but the
relative total pressure loss gets stronger with increased mass flow and the amount of flow passing in front of
the leading edge of the splitter blade decreases strongly for higher mass flow rates.
Q/Q
ref
=0.87 Q/Q
ref
=0.95 Q/Q
ref
=1.03
CR
t
b
2
----- =
Results Inlet Flow Distortion in a Centrifugal Compressor
72 00-914-663 Armin Zemp
To underline the statement of stronger vortical structure with increasing mass flow, profiles at the exit of the
impeller show relative total pressure ratio for the operating points illustrated previously:
Fig. 4.1.6.24: Relative Total Pressure Ratio (Mu=1.33, impeller exit, mainblade left, splitterblade right, turning: right to left)
To indicate the influence of blade speed on tip leakage vortex the streamlines are plotted for the same mass
flow rate of Q/Q
ref
=0.93 but for different impeller speeds:
Fig. 4.1.6.25: Tip Gap Vortex (Streamlines, relative Total Pressure Ratio, Q/Q
ref
=0.93)
With increased blade speed but constant corrected mass flow rate the amount of mass passing in front of the
leading edge of the splitter blade increases whereby the extension of the tip gap vortex through the inter
blade channel decreases.
Q/Q
ref
=0.87
Q/Q
ref
=0.95
Q/Q
ref
=1.03
Mu=1.20 Mu=1.26 Mu=1.33
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 73
A set of contour plots, again relative total pressure ratio at constant mass flow rate (Q/Q
ref
=0.93), supports
the statements above:
Fig. 4.1.6.26: Relative Total Pressure Ratio (Q/Q
ref
=0.93, impeller exit, mainblade left, splitterblade right, turning: right to left)
The trend from Mu=1.20 to Mu=1.33 is obvious. The region of low relative total pressure ratio in the inter blade
channel is decreasing. The vortex caused by fluid passing the shroud gap blocks a remarkable part of the chan-
nel. Due to the blockage effect of the vortical structure zones of higher relative total pressure ratios compen-
sate in terms of mass flow conservation.
Mu=1.2o
Mu=1.26
Mu=1.33
Results Inlet Flow Distortion in a Centrifugal Compressor
74 00-914-663 Armin Zemp
4.2 Transient CFD
In the second part of the present work the effect of inlet distortion in a centrifugal compressor is investigated.
The flow field downstream the distortion generated by grids described in chapter 3.13 is measured using FRAP
probes. The results of the measurement in the inlet section of the test rig are used as boundary condition for
the CFD model. The unsteady CFD investigation of the centrifugal compressor provides information about the
behavior of the distortion travelling through the impeller. As a source for the excitation of the impeller blades
the time resolved pressure distribution on the blade surface is in the focus of the transient simulations.
4.2.1 Grid
Due to the applied distortion patterns the entire impeller grid is needed for the unsteady investigation. The
impeller is implemented by simply completing the seventh part used for the steady state computations to the
full 360 degrees:
Fig. 4.2.1.1: CFD Grid (Entire Impeller)
The entire impeller grid consists of 4001200 nodes building 3564288 elements (hexahedra). A domain inter-
face is defined in the inlet section to enable a stationary and a rotating domain due to the fact that the FRAP
measured inlet boundary conditions are quantities in stationary frame of reference.
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 75
4.2.2 Boundary Conditions
The boundary conditions for the unsteady computations are equal to the steady state cases excepting the in-
let boundary condition:
Inlet: total pressure, total temperature and turbulence intensity profile (FRAP measurement)
Outlet: static pressure (average static pressure, varied until mass flow matches measurement)
Shroud: counter rotating wall, no slip, adiabatic
Impeller: constant rotational speed, no slip, adiabatic
Inlet Boundary Condition Profiles:
The flow field in the inlet section of the test rig was measured with FRAP probes in a foregoing semester thesis.
Due to the missing areas in the measurement plane because not accessible with the probe (white zones in the
illustration below) and rudiments in the flow field of upstream installations additionally to the distortion
screen (struts for the slip ring support needed for the strain gage signal transmission for example) the profiles
for the CFD model are applied to the model in an idealized form which also simplifies the achievment of peri-
odic convergency:
Fig. 4.2.2.1: Total Pressure Inlet Boundary Condition (Measured and Idealized) (A. Kammerer)
4.2.3 Turbulence Measurement
The performed measurements of the distorted inlet flow field with FRAP probes allow a Reynolds decomposi-
tion of the probe signal into a deterministic and a stochastic part:
Equation 4.2.3.1
The turbulent stresses are used to form the Reynolds stress tensor in terms of velocity fluctuations on the
three spatial directions. According to Porreca (Porreca et al., 2006) the turbulence intensity is extracted from
the time resolved measurements:
Equation 4.2.3.2
g t ( ) g

t ( ) g' t ( ) + =
Tu
u'
2
v'
2
w'
2
+ +
3c

2
--------------------------------- =
Results Inlet Flow Distortion in a Centrifugal Compressor
76 00-914-663 Armin Zemp
Two fundamental assumptions contained in the publication of Porreca (Porreca et al.) are not valid or can not
be proved currently. On the one hand the assumption of a much smaller gradient of the fluctuating velocity
component in streamwise direction compared to the other two directions to simplify the continuity equation
can not be proved. A series of measurements at different streamwise positions would be needed to check for
validity in an almost unaccelerated flow field.
On the other hand the continuity equation itself to integrate for the yaw component (only nonsimultanious
measurement) may not be applicable due to the coarse measurement grid. It is improper to integrate for
quantities with probably much smaller scales compared to the discretized spatial length scales of the mea-
surement. A validation of the proposed procedure of Porreca (Porreca et al.) is done in an undistorted inlet flow
field and points out the effect of the mentioned doubts:
Fig. 4.2.3.3: Turbulence Intensities in Undistorted Flow Field
The turbulence intensities in the circumferential (theta) and the radial direction should have the same order
of magnitude in the core of the flow field as a matter of course. The comparison of these two components
shows a smoothing of the circumferential component. The value of turbulence intensity in the boundary layer
is extremely high (might be an effect of probe geometry and/or pitch sensitivity) and therefore to handle with
care.
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 77
The smoothing effect is quantified statistically via mean turbulence intensity and standard deviation for the
present flow field:
The statistical analysis shows a smoothing of the mean turbulence intensity by a factor of about 4 in circum-
ferential direction compared to the radial component which is measured simultaneously and a reduction of
the standard deviation by a factor of almost 9.
A refinement of the measurement points to allow the discrete integration of the continuity equation for a
measurement plane of the order of m
2
is not recommended. The presented measurement of the un-
distorted flow field consists of about 800 measurement points. Therefore the Matlab post processing of the
time resolved FRAP data requires about 100 CPU hours on a PC with an Intel Conroe processor.
Nevertheless, the turbulence intensity can be evaluated due to the particular measurement setup. By travers-
ing the probe in circumferential direction in the measurement plane normal to the streamwise direction the
probe fixed cylindrical coordinate system allows a simplification of the equation for the turbulence intensity.
The equation for the turbulence intensity
Equation 4.2.3.4
with : aperiodic pressure (sensor 1)
: deterministic pressure
: deterministic absolute velocity
: aperiodic velocity component in circumferential direction
and : aperiodic pitch angle component
simplifies to
Equation 4.2.3.5
and is used to evaluate the FRAP measurements of the distorted inlet flow field of the compressor. The result-
ing turbulence intensity is then used as boundary condition at the inlet of the computational model addition-
ally to the total pressure and total temperature profile.
axial direction: 1.9682% 0.10677%
radial direction: 4.2675% 0.54887%
circumferential direction: 1.0785% 0.06243%
Tu Var
8 10
3

Tu
1
3
---
1
4
---
p
1
'
2
p

d
2
---------
\
|
|
| |
v'
2
c

2
------
\
|
| |
'
2
+ + =
p
1
'
p
d
c
v'
'
Tu
* 1
3
---
1
4
---
p
1
'
2
p

d
2
---------
\
|
|
| |
2 '
2
+ =
Results Inlet Flow Distortion in a Centrifugal Compressor
78 00-914-663 Armin Zemp
The evaluation for one third of the inlet measurement plane in the test rig is presented below for a distortion
screen to excite the 5. enginge order:
Fig. 4.2.3.6: Turbulence Intensity, Measurement Sector 1, 5. EO Screen
The idealized turbulence intensity profile (5. engine order excitation screen) is plotted below together with the
total pressure profile:
The values at the measurement points are interpolated onto the grid nodes at the corresponding inlet surface
of the computational domain. For the total temperature a constant value is used instead of a profile due to
the negligible variation in the measured flow field:
Fig. 4.2.3.7: Total Pressure Boundary Condition Profile (FRAP) (A.
Kammerer)
Fig. 4.2.3.8: Turbulence Intensity Boundary Condition Prof.
(FRAP) (A. Kammerer)
Fig. 4.2.3.9: Total Pressure Boundary Condition Profile (CFD) Fig. 4.2.3.10: Turbulence Intensity Boundary Condition Profile (CFD)
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 79
4.2.4 Turbulence Model and Solver Setup
As in the stationary case the turbulence model is in use. In a first step an initial solution with a frozen
rotor stator interface is required. The quasi steady state solution then accelerates the unsteady calculation
with a transient rotor stator interface at the frame change interface. The physical time step setup is a trade-
off between sampling rate of the unsteady flow properties and the computational effort. The size of the time
step has to ensure a proper resolution of the periodic distortion pattern and is affected by the rotational speed
of the impeller at the operating point under investigation due to a constant resolution of 77 physical impeller
positions over a period of one revolution leading to 11 points between two adjacent main blades. The highest
number of distortion baffles is five leading to a resolution of 15 to 16 points per distortion period at the inlet of
the impeller which is high enough to record the distortion.
4.2.5 Operating Points for Transient CFD
To determine the operating points of interest for the transient investigations the Campbell diagram of the im-
peller main blade is required. The investigated resonance points of the main blade are marked with blue cir-
cles:
Fig. 4.2.5.1: Campbell Diagram Impeller Main Blade (A. Kammerer)
CFD simulations of the following combinations of inlet boundary conditions and rotational speeds have been
done:
13360 rpm 16250 rpm 17340 rpm 20700 rpm
idealized p
01
boundary condition: X X X X
idealized p
01
& Tu boundary condition: X
real p
01
boundary condition: X
k
Results Inlet Flow Distortion in a Centrifugal Compressor
80 00-914-663 Armin Zemp
4.2.6 Convergence Quantification
Solver Residuals:
As for the steady state simulations the residuals are the most important measure of convergence. The RMS
residuals below show representatively the order of magnitude of the residuals for the transient cases:
Fig. 4.2.6.1: RMS Residuals (Transient CFD)
The level of the residuals remains constant over the whole range of iteration steps. At the beginning a slight
rise appears due to the fact that the initial solution was a quasi steady state case with a frozen rotor stator
interface.
Mass Imbalance:
The mass imbalances in the unsteady CFD simulations are of the order of which is extremely low for tran-
sient investigations:
Fig. 4.2.6.2: Mass Flow Imbalance
The red curve shows the imbalance of the inlet section which is a stationary domain. The blue curve is the
mass imbalance of the impeller. The imbalances as well as the residuals imply a certain periodicity. An other
convergence quantification for the transient investigations which is of the same importance as the solver re-
siduals is the periodicity of the solution due to the periodic distortion pattern at the inlet of the domain.
10
5
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 81
Periodic Convergence:
To record the periodicity of the solution several monitoring points are implemented on the surface of the main
blade:
Fig. 4.2.6.3: Monitoring Points on Main Blade
The monitoring points record the time resolved blade pressure. In the publication of Dickmann (Dickmann et
al., 2006) the periodicity is quantified graphically by connecting the peaks in the signal and looking for
straightness without a quantification of the periodicity. Dickmann (Dickmann et al., 2006) checked for period-
ic convergence by monitoring the unsteady blade power which causes a certain averaging (and therefore
smoothing) of the unsteady signal.
In contrast to Dickmann (Dickmann et al., 2006) the periodicity in the present work is quantified by statistical-
ly correlating the unsteady pressure signal of the monitoring points on the main blade surface of two consec-
utive impeller revolutions. This procedure provides a real quantification of the periodic c0nvergence by
correlating flow properties at discrete locations in the domain without any averaging or smoothing process.
The diagram below shows characteristic pressure signal development during a transient CFD run. The location
of the monitoring points corresponds to the figure one page before:
Fig. 4.2.6.4: Pressure Monitoring in Transient CFD Run
Results Inlet Flow Distortion in a Centrifugal Compressor
82 00-914-663 Armin Zemp
The resulting correlation coefficient of two consecutive impeller revolutions provides a convergence quantifi-
cation of the periodic pressure signal on the main blade surface:
Fig. 4.2.6.5: Correlation Coefficient for Transient CFD Run
The demanded correlation coefficient in the present work for periodic convergence must be at least 99%. The
values at the end of the diagram in the figure above are:
pressure side, 10% streamwise location: 99.9299%
pressure side, 50% streamwise location: 99.6755%
pressure side, 90% streamwise location: 99.9623%
suction side, 10% streamwise location: 99.9160%
suction side, 50% streamwise location: 99.9016%
suction side, 90% streamwise location: 99.8777%
If the required value for each correlation coefficient is reached the unsteady CFD results are recorded at each
physical impeller position over one revolution. The unsteady simulations with the mentioned convergence cri-
teria and the need for high quality initial solutions for the transient runs with a grid of about 4 million nodes
requires high computational power and causes long runtimes for each operating point under investigation. A
good initial solution needs about 15 CPU days and the transient run to reach periodic convergence requires
about 84 CPU days.
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 83
4.2.7 Validation
The validation of the transient investigation is not as simple as in the steady state cases. Once the mass flow
in the numerical simulation corresponds to the measured mass flow in the rig by adjusting the back pressure
in the CFD model the resulting velocity profile at the inlet boundary condition plane can be compared to the
measured FRAP profile:
The profiles look different in a first view due to the different color coding. The FRAP profile is made with Matlab
while the profile for the CFD case is made in the ANSYS CFX post processor. If one compares the numerical val-
ues the difference is in the range of 0.5% (maximum: 2.5%). The model therefore is assumed to be a good ap-
proximation of the reality.
4.2.8 Influence of Turbulence Intensity Boundary Condition
To quantify the influence of the turbulence intensity boundary condition at the inlet two simulations at
16250rpm are done. The only difference between the two cases is the turbulence intensity distribution. One
case uses an average turbulence intensity of 5% over the whole inlet area. The second case contains the tur-
bulence intensity distribution extracted from the FRAP measurements as explained in chapter 4.2.3. The sim-
ulations do not indicate any differences in the resulting flow field. To quantify the difference between the two
cases the same procedure used to quantify the periodic convergence is applied on the 6 blade pressure mon-
itoring points:
pressure side, 10% streamwise location: 99.979%
pressure side, 50% streamwise location: 99.993%
pressure side, 90% streamwise location: 99.909%
suction side, 10% streamwise location: 99.953%
suction side, 50% streamwise location: 99.986%
suction side, 90% streamwise location: 99.966%
Due to the high consistency of the compared results the transient investigations use an average turbulence
intensity of 5% as inlet boundary condition to simplify the whole procedure.
Fig. 4.2.7.1: Normalized Axial Velocity (FRAP) (A. Kammerer) Fig. 4.2.7.2: Normalized Axial Velocity (CFD)
Results Inlet Flow Distortion in a Centrifugal Compressor
84 00-914-663 Armin Zemp
4.2.9 Flow through Impeller
The intention of this chapter is to characterize the flow through the centrifugal impeller and to track the inlet
distortion convecting through the device. The figure below shows the trace of the distortion pattern through
the impeller. The color coding illustrates the ratio between relative total pressure divided by the relative total
pressure at the inlet. The red zones at the inlet are zones of high relative total pressure ratio and the orange
regions show the inlet flow distortion. The blue spots show the tip leakage vortex:
Fig. 4.2.9.1: Inlet Distortion Convecting through Impeller (Relative Total Pressure Ratio)
Due to the large pressure and velocity gradients through the impeller the tracking of the inlet distortion is not
trivial. The relative total pressure provides the possibility to visualize the pattern of low relative total pressure.
As a matter of course the relative total pressure in the distorted inlet flow is lower compared to the undistort-
ed flow. The relative total pressure of the fluid while convecting through the impeller is constant in an ideal
system. It is therefore possible to identify the zones of relative total pressure loss (boundary layer, vortical
structures, inlet distortion, etc.).
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 85
A descriptive possibility to visualize the trace of the distortion through the blading is a blade to blade view at
midspan:
Three different distortion screen geometries are in use for the unsteady analysis as illustrated in chapter 3.13.
The comparison above clarifies the varying spatial extension of the distortion pattern and explains the need
for the entire impeller in the transient CFD simulation due to the asymmetric allocation over the seven main
blade pitches. It is not possible to set up a model containing only a fraction of the whole impeller with periodic
boundary interfaces as in the steady state computations where only the seventh part of the impeller is imple-
mented.
The figure on the left shows the excitation of the first bending mode with the 5. engine order distortion at
16250 rpm. On the middle the 4. EO distortion screen excites the first bending mode at 20700 rpm. It is hard
to see differences in the illustrations except of the tip leakage vortex (blue zones) which is stronger at higher
rotational speed and higher mass flow rate. The blade to blade view on the right illustrates the relative total
pressure coefficient for a 3. engine order distortion screen that excites the 1st bending mode of the main blade
with the first harmonic (6. EO) at 13360rpm. An interesting feature is visible in the view at 13360 rpm on the
pressure side of the second main blade from the bottom. The blue zone shows a separation caused by the inlet
distortion and is then convected through the impeller.
Fig. 4.2.9.2: 16250rpm, 5.EO, Q/Q
ref
=0.73 Fig. 4.2.9.3: 20700rpm, 4.EO, Q/Q
ref
=0.98 Fig. 4.2.9.4: 13360rpm, 3.EO, Q/Q
ref
=0.58
Results Inlet Flow Distortion in a Centrifugal Compressor
86 00-914-663 Armin Zemp
As mentioned in chapter 4.2.2 the total pressure profile is applied in an idealized form to allow a simple iden-
tification of the periodicity to get a certain experience concerning unsteady simulations with the software in
use. About three months experience in setting up and monitoring time resolved simulations in ANSYS CFX al-
lowed to run a case with inlet total pressure profile as it is measured in the test rig with FRAP probes:
A certain idealization procedure still has to be done due to the fact that, as mentioned earlier, not the whole
measurement plane in the test rig is accessible because of struts supporting the inlet duct of the rig. The fol-
lowing comparison shows the relative total pressure coefficient for the same screen geometry (5. EO):
The contours on the left and in the middle illustrate exactly the same operating point. The only difference is
the inlet total pressure boundary condition. The blade to blade view on the left is the result of the idealized
inlet total pressure boundary condition profile. The contour in the middle shows the resulting relative total
pressure ratio for the real total pressure profile measured with FRAP probes. One can see that the idealized
case shows continuous distribution of the distortion pattern at the inlet compared to the real one. The dis-
torted regions at the inlet (yellow regions) have lower relative total pressure ratio niveau for the idealized case.
On the right the same distortion screen excites the second main blade mode via first harmonic (10. engine or-
der) again with an idealized total pressure profile at the inlet.
Fig. 4.2.9.5: Idealized Inlet Total Pressure Profile (A. Kammerer) Fig. 4.2.9.6: Real Inlet Total Pressure Profile (A. Kammerer)
Fig. 4.2.9.7: 16250rpm, 5.EO, Q/Q
ref
=0.73 Fig. 4.2.9.8: 16250rpm, 5.EO, Q/Q
ref
=0.73 Fig. 4.2.9.9: 17340rpm, 5.EO, Q/Q
ref
=0.80
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 87
4.2.10 Unsteady Blade Pressure
A major objective of the present work is the unsteady pressure distribution on the blade surface. The time
varying pressure distribution on the blade caused by the inlet distortion may excite modes of the blades. It is
an advantage of the numerical investigation compared to the test rig measurements that one can run the sys-
tem at resonance without the risk to destroy anything in the rig.
To understand the excitation of the blades it is important to know the magnitude and location of the unsteady
load. The first operating point under investigation is at 16250rpm and is the resonance point of the first bend-
ing mode of the main blade excited by the fifth engine order distortion screen.
16250 rpm, 5. EO:
To get a first idea of the pressure distribution on the suction and pressure side of the main blade the pressure
fluctuation is plotted on the main blade surface. The Reynolds decomposition of the unsteady pressure avoids
the difficulty to identify the distortion due to the comparatively strong pressure gradient through the blading
by just plotting the fluctuating part of the pressure signal. The figure below shows the unsteady pressure dis-
tribution for one particular time step:

Fig. 4.2.10.1: p(t) /p
01,rel
on Main Blade Surface (16250 rpm, 5. EO)
The orange zone on the pressure side of the blade shows a positive pressure fluctuation compared to the local
time mean pressure arising from an undistorted inlet flow field region. The blue areas on the suction side of
the main blade have negative pressure fluctuation resulting from a distorted zone in the inlet of the compres-
sor.
Results Inlet Flow Distortion in a Centrifugal Compressor
88 00-914-663 Armin Zemp
The path- time diagram offers the possibility to illustrate the distortion convecting along the blade surface by
plotting the fluctuating part of the pressure versus streamwise location at 50% span and time axis:
More information about the effect of the unsteady pressure field on the main blade can be gained by building
the difference pressure side minus suction side:
Fig. 4.2.10.4: Path- Time Pressure Side minus Suction Side (16250rpm)
The integral of the pressure fluctuation over the blade area equals the dynamic load caused by the distorted
inlet flow field. One can see that the distortion convects through the impeller with almost constant speed. The
amplitude gets weaker from the leading edge of the main blade to the leading edge of the splitter blade. The
alternating direction of the dynamic load at the leading edge may excite the blade if the frequency of the ex-
citation hits the natural frequency of the blade.
Fig. 4.2.10.2: Path- Time Suction Side (16250rpm) Fig. 4.2.10.3: Path- Time Pressure Side (16250rpm)
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 89
To illustrate the amplitude of the periodic excitation the pressure fluctuation is plotted over one impeller ro-
tation:
Fig. 4.2.10.5: p/p
01,rel
(16250rpm, 10% streamwise direction on mainblade)
The signal clearly shows the 5. EO excitation and the peak to peak amplitude is about 14% of the average rela-
tive total pressure at the inlet of the compressor. It is important to point the fact that there is no phase shift
between the fluctuation on the pressure and suction side.
The Fourier transform of the unsteady pressure fluctuation at midspan in streamwise direction on both the
pressure and suction side shows the development of the distortion convecting through the impeller:
As expected the excitation is very strong at the leading edge of the main blade. The first harmonic appears
only very weak at the front of the blade. The amplitude of the first harmonic is about 15% of the amplitude of
the fundamental frequency.
Fig. 4.2.10.6: FFT, 16250rpm, Pressure Side Main Blade, 50% Span Fig. 4.2.10.7: FFT, 16250rpm, Suction Side, Main Blade, 50% Span

Results Inlet Flow Distortion in a Centrifugal Compressor

90 00-914-663 Armin Zemp
The nonuniform development of the amplitude in the frequency domain at midspan from leading to trailing
edge demands for an illustration of the distribution of the amplitude on the blade surfaces:
The red zones are the regions of the highest amplitude of the signal in the frequency domain. But there are
huge areas on the surfaces with very low amplitude and therefore very weak dynamic load.
The next operating point under investigation is at 13360 rpm with a 3. EO distortion screen. The first harmonic
excites the first bending mode of the main blade.
13360 rpm, 3. EO:
The path- time diagrams offer a detailed view on the dynamics through the impeller:
[ ] [ ]
Fig. 4.2.10.8: FFT Amplitude, 126250rpm, Pressure Side Main Blade Fig. 4.2.10.9: FFT Amplitude, 126250rpm, Suction Side, Main Blade
Fig. 4.2.10.10: Path- Time Suction Side (13360rpm) Fig. 4.2.10.11: Path- Time Pressure Side (13360rpm)
p' /p
01 rel ,
p' /p
01 rel ,
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 91
The difference of pressure side minus suction side illustrates the dynamic load onto the main blade:
Fig. 4.2.10.12: Path- Time Pressure Side minus Suction Side (13360rpm)
The intention of the distortion screen with three distortion baffles is to excite the first main blade mode via
first harmonic. The signal at the leading edge indicates the content of the first harmonic quite clearly. To get
a better idea about the evolution of the fluctuating pressure difference at the leading edge the signal at 10%
streamwise direction is plotted over one impeller revolution:
Fig. 4.2.10.13: p/p
01,rel
(13360rpm, 10% streamwise direction on mainblade)
Results Inlet Flow Distortion in a Centrifugal Compressor
92 00-914-663 Armin Zemp
The Fourier transform of the unsteady pressure fluctuation at midspan in streamwise direction on both the
pressure and suction side shows the development of the distortion convecting through the impeller:
As expected the excitation is again very strong at the leading edge of the main blade. The first harmonic ap-
pears quite strong at the front of the blade. The FFT signal right at the leading edge shows the relation of the
3. EO compared to the harmonics:
Fig. 4.2.10.16: FFT, Leading Edge Main Blade (13360rpm)
The amplitude of the 6. EO excitation is about 3.7 times smaller than the 3. EO peak. The contour of the FFT
amplitude on the main blade surface of the fundamental frequency illustrates the distribution over the blade
surface:
The blue regions with very low amplitude on both sides of the blade are noticeable but the amplitudes at the
leading edge and in the rear part of the suction side are very high.
Fig. 4.2.10.14: FFT, 13360rpm, Pressure Side Main Blade, 50% Span Fig. 4.2.10.15: FFT, 13360rpm, Suction Side, Main Blade, 50% Span
[ ] [ ]
Fig. 4.2.10.17: FFT Amplitude, 13360rpm, Pressure Side Main Blade Fig. 4.2.10.18: FFT Amplitude, 13360rpm, Suction Side Main Blade
p' /p
01 rel ,
p' /p
01 rel ,
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 93
The next operating point under investigation is at 17340 rpm with a 5. EO distortion screen. The first harmonic
excites the second bending mode of the main blade.
17340 rpm, 5. EO:
The fluctuating part of the unsteady pressure distribution on the blade surface is plotted for one time step be-
low:
Fig. 4.2.10.19: p(t)/p
01,rel
on Main Blade Surface (17340 rpm, 5. EO)
The red zone on the pressure side is a rudiment of an undistorted inlet flow field. One can see that the leading
edge passes a distorted zone (blue leading edge on pressure side). On the suction side the blue area is again a
low relative total pressure pattern caused by a distorted inlet flow field. It is conspicuous that the distorted
area convects over the suction side at a niveau close to the hub. The reason might be the tip leakage vortex
pushing the flow field from the shroud to the hub.
The path- time diagrams offer a detailed view on the dynamics through the impeller:
Fig. 4.2.10.20: Path- Time Suction Side (17340rpm) Fig. 4.2.10.21: Path- Time Pressure Side (17340rpm)
Results Inlet Flow Distortion in a Centrifugal Compressor
94 00-914-663 Armin Zemp
The difference of pressure side minus suction side illustrates the dynamic load onto the main blade:
Fig. 4.2.10.22: Path- Time Pressure Side minus Suction Side (17340rpm)
The intention of the distortion screen with five distortion baffles is to excite the second main blade mode via
first harmonic. The signal at the leading edge does not indicates the content of the first harmonic.
To get a better idea about the evolution of the fluctuating pressure difference at the leading edge the signal
at 10% streamwise direction is plotted over one impeller revolution:
Fig. 4.2.10.23: p/p
01,rel
(17340rpm, 10% streamwise direction on mainblade)
The slight shift is the result of a not completely converged solution. An additional impeller revolution would
be sufficient to lose the drift. The relevant quantities are not affected by this effect.

Inlet Flow Distortion in a Centrifugal Compressor Results

Armin Zemp 00-914-663 95
The Fourier transform of the unsteady pressure fluctuation at midspan in streamwise direction on both the
pressure and suction side shows the development of the distortion convecting through the impeller:
The excitation is very strong at the leading edge of the main blade. The first harmonic appears only very weak
at the front of the blade. The contour of the FFT amplitude on the main blade surface illustrates the distribu-
tion over the blade surface:
Fig. 4.2.10.24: FFT, 17340rpm, Pressure Side Main Blade, 50% Span Fig. 4.2.10.25: FFT, 17340rpm, Suction Side, Main Blade, 50% Span
[ ] [ ]
Fig. 4.2.10.26: FFT Amplitude, 17340rpm, Pressure Side Main Blade Fig. 4.2.10.27: FFT Amplitude, 17340rpm, Suction Side Main Blade
p' /p
01 rel ,
p' /p
01 rel ,
Results Inlet Flow Distortion in a Centrifugal Compressor
96 00-914-663 Armin Zemp
The last operating point under investigation is at 20700 rpm with a 4. EO distortion screen that excites the
first bending mode of the main blade.
20700 rpm, 4. EO:
The fluctuating part of the unsteady pressure distribution on the blade surface is plotted for one time step be-
low:
Fig. 4.2.10.28: p(t)/p
01,rel
on Main Blade Surface (20700 rpm, 4. EO)
The broad orange area on the pressure side is an undistorted zone of the inlet flow field. At the trailing edge
a distorted zone is leaving the impeller. On the suction side the blue area is again a low relative total pressure
pattern caused by a distorted inlet flow field.
The path- time diagrams offer a detailed view on the dynamics through the impeller:
Fig. 4.2.10.29: Path- Time Suction Side (20700rpm) Fig. 4.2.10.30: Path- Time Pressure Side (20700rpm)
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 97
The difference of pressure side minus suction side illustrates the dynamic load onto the main blade:
Fig. 4.2.10.31: Path- Time Pressure Side minus Suction Side (20700rpm)
The intention of the distortion screen with four distortion baffles is to excite the first main blade mode. To get
a better idea about the evolution of the fluctuating pressure difference at the leading edge the signal at 10%
streamwise direction is plotted over one impeller revolution:
Fig. 4.2.10.32: p/p
01,rel
(20700rpm, 10% streamwise direction on mainblade)
The slight shift is the result of a not completely converged solution. An additional impeller revolution would
be sufficient to lose the drift. The relevant quantities are not affected by this effect.

Results Inlet Flow Distortion in a Centrifugal Compressor

98 00-914-663 Armin Zemp
The Fourier transform of the unsteady pressure fluctuation at midspan in streamwise direction on both the
pressure and suction side shows the development of the distortion convecting through the impeller:
The excitation is very strong at the leading edge of the main blade. The high amplitude at the rear part of the
main blade on the suction side is conspicuous. The contour of the FFT amplitude on the main blade surface
illustrates the distribution over the blade surface:
The extension of the blue regions with very low amplitude on both sides of the blade is extremely large. The
reason for such a large extension might be the relatively high rotational speed. The high amplitudes not only
at the leading edge but also in the rear part of the blade are clearly visible as indicated in the FFT signal along
midspan.
Fig. 4.2.10.33: FFT, 20700rpm, Pressure Side Main Blade, 50% Span Fig. 4.2.10.34: FFT, 20700rpm, Suction Side, Main Blade, 50% Span
[ ] [ ]
Fig. 4.2.10.35: FFT Amplitude, 20700rpm, Pressure Side Main Blade Fig. 4.2.10.36: FFT Amplitude, 20700rpm, Suction Side Main Blade
p' /p
01 rel ,
p' /p
01 rel ,
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 99
The last investigation is done with non- idealized inlet total pressure profile at 16250rpm. The solver residuals
as well as the mass imbalance are of the same order of magnitude as in the investigations with idealized pro-
files.
The unsteady pressure signals at the solver monitoring points look different due to the slightly asymmetric
profile in circumferential direction:
Fig. 4.2.10.37: Pressure Monitoring in Transient CFD (16250rpm, two impeller revolutions, unidealized total pressure profile)
The path- time diagrams show the same trend as in the idealized case but with different distribution for the
distortion from screen to screen:
Fig. 4.2.10.38: Path- Time Suction Side (16250rpm, real) Fig. 4.2.10.39: Path- Time Pressure Side (16250rpm, real)
Results Inlet Flow Distortion in a Centrifugal Compressor
100 00-914-663 Armin Zemp
The difference of pressure side minus suction side illustrates the dynamic load onto the main blade:
Fig. 4.2.10.40: Path- Time Pressure Side minus Suction Side (16250rpm, real)
The development of the distortion convecting through the impeller shows good consistence compared to the
idealized case.
The evolution of the fluctuating pressure difference at the leading edge at 10% streamwise direction is plotted
over one impeller revolution:
Fig. 4.2.10.41: p/p
01,rel
(16250rpm, real, 10% streamwise direction on mainblade)
Inlet Flow Distortion in a Centrifugal Compressor Results
Armin Zemp 00-914-663 101
The Fourier transform of the unsteady pressure fluctuation at midspan in streamwise direction on both the
pressure and suction side shows the development of the distortion convecting through the impeller:
As expected the signal in the frequency domain is no more clean as in the idealized case but the excitation still
is very strong at the leading edge of the main blade. Lower engine orders appear quite strong at the leading
edge of the blade. The contour of the FFT amplitude on the main blade surface illustrates the distribution over
the blade surface:
The comparison with the idealized case show very good consistence of the resulting distribution of the FFT
amplitude on the main blade surface:
Fig. 4.2.10.42: FFT, 16250rpm, real, PS Main Blade, 50% Span Fig. 4.2.10.43: FFT, 16250rpm, real, SS, Main Blade, 50% Span
[ ] [ ]
Fig. 4.2.10.44: FFT Amplitude, 16250rpm, real, PS Main Blade Fig. 4.2.10.45: FFT Amplitude, 16250rpm, real, SS Main Blade
[ ] [ ]
Fig. 4.2.10.46: FFT Amplitude, 16250rpm, ideal, PS Main Blade Fig. 4.2.10.47: FFT Amplitude, 16250rpm, ideal, SS Main Blade
p' /p
01 rel ,
p' /p
01 rel ,
p' /p
01 rel ,
p' /p
01 rel ,
Results Inlet Flow Distortion in a Centrifugal Compressor
102 00-914-663 Armin Zemp
A comparison between the investigated cases shows the effect of the velocity difference at the inlet of the
computational domain (which accords to the FRAP measurement plain in the RIGI test rig) on the pressure
fluctuation on the main blade surface at 10% span:
The comparison is to handle with care due to the fact that all the parameters such as mass flow rate, pressure
ratio, etc. vary for all the cases. An investigation at constant rotational speed with variable mass flow rate
would be more meaningful.
case:
13360 rpm ~ 0.40 ~ 0.13
16250 rpm ~ 0.24 ~ 0.14
17340 rpm ~ 0.19 ~ 0.20
20700 rpm ~ 0.15 ~ 0.22
V
i nlet
V
i nlet
-----------------
p'
p
0 rel i nl et , ,
-------------------------
Inlet Flow Distortion in a Centrifugal Compressor Conclusions
Armin Zemp 00-914-663 103
5. Conclusions
The impeller grid used for the numerical investigation is of high quality. For future investigation one can think
of a reduction of the number of nodes without a significant loss in quality of the results to accelerate the solv-
er process and to reduce the amount of computational power. The detailed validation of the model on the
base of steady state simulations showed good consistence with the test rig performance. The deviations in
performance are of the order of 2% over the whole compressor map. The stationary computations showed
good convergence rates and extremely low mass imbalances of the order of 10
-6
. An additional stationary do-
main in the impeller model for the diffuser section would allow an amelioration of the convergence niveau by
reducing numerical instabilities due to the small angles between relative flow direction and domain boundary
in the present rotating diffuser exit. The comparison of the total pressure distribution at the exit of the impel-
ler from hub to shroud with FRAP measurements confirmed the prediction of the CFD results. The vortical
structure caused by tip leakage flow occurs stronger in the numerical solution than in the measurements. The
reason might be an overprediction of total pressure loss in the CFD model or a underprediction in the FRAP
measurements due to the highly fluctuating flow field at the exit of the impeller affecting accuracy of the
FRAP results. The qualitative description of the flow pattern through the impeller for several operating points
via relative and absolute Mach number provides information about the fluid dynamics within the centrifugal
compressor. A main focus concentrates on the tip leakage vortex. The results show the trend to stronger vor-
ticity with increased mass flow rate at constant blade speed. The comparison at constant mass flow rate but
increased blade speed indicated a higher amount of tip leakage fluid of the main blade shroud gap passing
the leading edge of the splitter blade. The vorticity in the inter blade channel therefore decreases with in-
creased rotational speed at constant mass flow rate. The FRAP measurements in the distorted inlet flow field
allowed the extraction of the turbulence intensity as additional inlet boundary condition profile for the CFD
model. To get information about accuracy of the predicted turbulence intensity with FRAP probes a validation
with alternative measurement techniques such as hot wire probes should be performed. The comparison of
the results with implemented turbulence intensity profile at the inlet to the case with average turbulence in-
tensity of 5% over the whole inlet of the model showed no significant change of the unsteady blade pressure
distribution. Six time resolved cases at four different operating points were performed. The transient CFD sim-
ulations show good periodic convergence quantified via correlation of discrete nodes on the main blade sur-
face monitoring unsteady blade pressure. The amount of computational power makes the unsteady solution
process extremely time consuming. Due to the fact that the simulation of the entire impeller is needed the
solver can only run in a parallel mode. At least four nodes on the Linux cluster are therefore necessary to han-
dle the large model. The tracking of the distortion through the impeller provides a basic understanding of the
interaction with the impeller blades. The spectral analysis of the unsteady pressure distribution on the main
blade provides information about the dynamic load caused by the inlet distortion. The comparison between
idealized total pressure profile as inlet boundary condition and the real FRAP measured profile at exactly the
same operating point showed good consistence and allows future investigation with idealized total pressure
profiles which accelerates the setup of transient CFD solver input files.
Conclusions Inlet Flow Distortion in a Centrifugal Compressor
104 00-914-663 Armin Zemp
Inlet Flow Distortion in a Centrifugal Compressor Future Work
Armin Zemp 00-914-663 105
6. Future Work
The steady state CFD results over the whole range of operating points provides a basis for future post process-
ing of the huge amount of data. For further investigation one may reduce the number of grid points due to
the fact that the y
+
values are not at the upper limit for the solver software in use to accelerate the process.
An additional stationary domain in the computational model for the diffuser section would reduce numerical
instability caused by small angles between relative flow direction and domain boundary using a rotating
frame of reference for the diffuser outlet.
The unsteady CFD investigation is done for only four operating points. It might be interesting to vary several
parameters such as mass flow rate and pressure ratio but the whole procedure is extremely time consuming
due to the large model and the relatively slow convergence rate of centrifugal compressor models.
The software used for the present work offers the possibility to couple the fluid model with the structure of
the impeller. The implementation of a model containing the fluid- structure interaction may provide interest-
ing and detailed knowledge about the excitation of the impeller blades due to inlet flow distortion. The result-
ing reaction of the impeller structure might be used to compare the strain gage measurements with the
numerical model.
The unsteady pressure distribution on the blades of the CFD model allows a comparison with the planned
pressure measurements on the main blades of the impeller in the test rig. The numerical results may simplify
Albert Kammerers task to find a model for the forcing function of the blade vibrations.
The turbulence intensity measured with FRAP probes and used as inlet boundary condition in the CFD model
has to be validated with other measurement techniques such as hot wires due to the fact that there is no in-
dication about the accuracy the FRAP results.
Future Work Inlet Flow Distortion in a Centrifugal Compressor
106 00-914-663 Armin Zemp
Inlet Flow Distortion in a Centrifugal Compressor References
Armin Zemp 00-914-663 107
7. References
Abhari, R.S., Prof. Dr., Lecture Notes Aerospace Propulsion, SS 2006 ETH Zurich
Abhari, R.S., Prof. Dr., Lecture Notes Turbomachinery Design, WS 2005/2006 ETH Zurich
Ainsworth, R.W., Miller, R.J., Moss, R.W., Thorpe, S.J., Unsteady pressure measurement, Measurement
Science and Technology, Vol. 11, 2000
ANSYS Europe Ltd., User Guide, 2007
Ariga, I., Kasai, N., Masuda, S., Watanabe, I., Watanabe, Y., The Effect of Inlet Distortion on the Perfor-
mance Characteristics of a Centrifugal Compressor, ASME paper No. 82-GT-92, 1982
Bosman, C., Jadayel, O.C., A quantified study of rothalpy conservation in turbomachines, Int. J. Heat
and Fluid Flow, 17, No. 4, 410-17, 1996
Brouchkaert, J.F., Sieverding, C.H., Manna, M., Development of a fast response 3- hole pressure probe,
The 14th Symposium on Measuring Techniques in Transonic and Supersonic Flows in Cascades and
Turbomachines, Limerick, 1998
Came, P., The development, application and experimental evaluation of a desing procedure for centrif-
ugal compressors, AGARD Conference Proc. No. 282, 1978
Came, P.M., Herbert, M.V., Design and experimental performance of some high pressure ratio centrifu-
gal compressors, AGARD Conference Proc. No. 282, 1980
Campbell, W., Protection of Steam Turbine Disk Wheels. ASME paper 120, 1924
Clements, W.W., Artt, D.W., The influence of diffuser channel length to width ratio on the efficiency of
a centrifugal compressor, Proc. Instn. Mech. Engrs, 202, No. A3, 163-9, 1988
Cumpsty, N. A., Compressor Aerodynamics, 2004
Dickmann, H.-P., Secall Wimmel, T., Szwedowicz, J., Filsinger, D., Roduner, C.H., Unsteady Flow in a
Turbocharger Centrifugal Compressor: Three- Dimensional Computational Fluid Dynamics Simulation
and Numerical and Experimental Analysis of Impeller Blade Vibration, ASME, 2006
Dixon, S.L., Fluid Mechanics, Thermodynamics of Turbomachinery, 1998
Emmons, H.W., Kronauer, R.E., Rocket, J.A., A survey of stall propagation- experiment and theory,
ASME, Series D, 81, 1959
Engeda, A., Kim, Y., Aungier, R., Direnzi, G., The Inlet Flow Structure of a Centrifugal Compressor Stage
and its Influence on the Compressor Performance, J. of Fluids Engineering, Vol. 125, 2003
Epstein, A.H., High frequency response measurements in turbomachniery, VKI Lecture Series, Rhode-
Saint- Gense, 1985
Gizzi, W.P., Dynamische Korrekturen fr schnelle Strmungssonden in hochfrequent fluktuierenden
Strmungen, PhD Thesis ETH Nr. 13482, 2000
Gorla, R.S.R., Khan, A.A., Turbomachinery Design and Theory, 2003
Gossweiler, C., On Probes and Measuring Techniques for Fast- Response Flow Measurement Using Pi-
ezo- Resistive Pressure Transducers, PhD Thesis ETH Nr. 10253, 1993
Greitzer, E.M., Nikkanen, J.P., Haddad, D.E., Mazzawy, R.S., Joslyn, H.D., A fundamental criterion for the
application of rotor casing treatment, J. Fluid Eng., 101, 204-262, 1979
Greitzer, E.M., Tan, C.S., Graf, M.B., Internal Flows - Concepts and Applications, 2004
Grotjans, H., Menter, F.R., Wall functions for general application CFD codes, 1998
References Inlet Flow Distortion in a Centrifugal Compressor
108 00-914-663 Armin Zemp
Horlock, J.H., Axial Flow Compressors, 1958
Kerrebrock, J.L., Aircraft Engines and Gas Turbines, 1992
Kleiser, L., Lecture Notes Turbulent Flows, WS 2005/2006 ETH Zurich
Kupferschmied, P., Kppel, P., Gizzi, W.P., Gyarmathy, G., Time- resolved flow measurements with fast-
response aerodynamic probes in turbomachines, Measurement Science an Technology, Vol. 11, Issue 7,
Article 318, 2000
Launder, B.E., Spalding, D.B., The numerical computation of turbulent flows, 1974
Lymann, F.A., On the conservation of rothalpy in turbomachines, J. of Turbomachinery, 115, 520-6, 1993
Moore, J. , Moore, J.G., Timmis, P.H., Performance evaluation of centrifugal compressor impellers using
three- dimensional viscous flow calculations, J. Eng. Gas. Turbines Power, 106, 475-81, 1984
Pfau, A., Schlienger, J., Kalfas, A.I., Abhari, R.S., Introducing the New Virtual Four Sensor Fast Response
Aerodynamic Probe (FRAP), The 16th Symposium on Measuring Techniques in Cascades and Turbo-
machines, Cambridge, Sept. 2002
Pfau, A., Schlienger, J., Kalfas, A.I., Abhari, R.S., Unsteady, 3-dimensional Flow Measurements Using a
Miniature Virtual 4- sensor Fast Response Aerodynamic Probe (FRAP), Draft for Proceedings of ASME
Turbo Expo, 16-19 June 2003, Atlanta, Georgia
Porreca, L., Hollenstein, M., Kalfas, A.I., Abhari, R.S., Turbulence Measurements and Analysis in a Multi-
stage Axial Turbine, 2006
Rodgers, C., Sapiro, L., Design considerations for high pressure ratio centrifugal compressors, ASME 72-
GT-91, 1972
Schleer, M., Mokulys, T., Abhari, R.S., Design of a high pressure- ratio centrifugal compressor for study-
ing Reynolds number effects, IMechE London, 2003
Schleer, M., PhD Thesis, Flow Structure and Stability of a Turbocharger Centrifugal Compressor, 2006
Senoo, Y., Kita, Y., Ookuma, K., Measruement of two dimensional periodic flow with a cobra probe, J.
Fluids Engineering, p295, 1973
Smith, G.D.J., Cumpsty, N.A., Flow phenomena in compressor casing treatment, J. Eng. Gas Turbines
and Power, 106, 532-41, 1984
Withfield, A., Baines, N.C., Design of Radial Turbomachines, 1990