Influence of Engineered Roughness on

the Flow Instabilities in a Centrifugal

Compressor
Model Development & Output Analysis

Prakhar Kapoor
Master of Science Thesis
Report Number: 2641

Mechanical, Maritime and Materials Engineering

Influence of Engineered Roughness on the Flow Instabilities in a
Centrifugal Compressor
Model Development & Output Analysis

Master of Science Thesis

For the degree of Master of Science in Mechanical Engineering:

Sustainable Process and Energy Technology at Delft University of
Technology

Prakhar Kapoor

August 29, 2014

Faculty of Mechanical, Maritime and Maritime Engineering (3ME) · Delft University of

Technology
The work in this thesis was supported by Mitsubishi Turbocharger and Engine Europe B.V
(MTEE). Their cooperation is hereby gratefully acknowledged.

Copyright c Energy Technology

All rights reserved.
 Delft University of Technology
Department of
Energy Technology

The undersigned hereby certify that they have read and recommend to the Faculty of
Mechanical, Maritime and Maritime Engineering (3ME) for acceptance a thesis
entitled
Influence of Engineered Roughness on the Flow Instabilities in a
Centrifugal Compressor
by
Prakhar Kapoor
in partial fulfillment of the requirements for the degree of
Master of Science Mechanical Engineering: Sustainable Process and
Energy Technology

Dated: August 29, 2014

Supervisor(s):
Dr. Rene Pecnik

Dr. Adeel Javed

Reader(s):
Dr. Mathieu Pourquie

Prof. Dr. Bendiks Jan Boersma

 Abstract

Centrifugal compressors form an integral part of automotive turbochargers. Strict emission

regulations and increased engine downsizing in the automotive industry are pushing the tur-
bocharger centrifugal compressors to provide the charged air at very lower mass flow rates.
However, at low mass flow rates, the operation and application of centrifugal compressors
for a given turbocharger is limited by the fluid dynamic instabilities. These flow instabilities
cause the compressor to enter the state of stall and/or surge. Stall defines the lower limit of
stable operating range for the compression system. The suppression of this flow instability
is a key research focus of turbomachinery aerodynamics. In this thesis, an attempt has been
made to reduce the compressor instability and subsequently increase the surge margin by the
application of engineered roughness on different compressor parts.
The analysis has been split into two parts. Firstly, CFD analysis have been carried out on sev-
eral test cases to validate the reliability of the commercial solver. The comparison of the CFD
results with the available reference data shows a reasonable agreement in terms of prediction
of laminar to turbulent transition location and influence of wall roughness on transition onset
location. Secondly, a detailed CFD analysis of the centrifugal compressor has been made in
two phases. In the first phase, the overall compressor performance has been simulated from
stall to choke over a specific turbocharger rotational speed. In the second phase, a parametric
roughness study has been carried out by considering the effects of predefined wall roughness
and localized roughness strips on the unstable flow in the compressor at low mass flow rates.
For the wall roughness evaluation, the roughness has been applied on the impeller blades,
impeller shroud, diffuser and inlet shroud. On the other hand, for the localized roughness
strips, roughness have been defined at specific locations at the main blade suction side and
the diffuser shroud. The analysis reveals a reduction in flow instability in the compressor
domain. The steady state simulations show a significant improvement in the flow structure
of the diffuser in terms of the reduction in flow reversal. Furthermore, an improvement in the
impeller is observed by reduction in the low momentum wake region.

Master of Science Thesis Prakhar Kapoor

ii

Prakhar Kapoor Master of Science Thesis

 Table of Contents

Acknowledgments ix

Nomenclature xiii

1 Introduction 1
1-1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1-2 Compressor Stall and Surge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1-2-1 Steady Stall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1-2-2 Dynamic Stall and Surge . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1-3 Flow Separation and Boundary Layer Transition . . . . . . . . . . . . . . . . . . 5
1-4 Computational Fluid Dynamics and Transition Modeling . . . . . . . . . . . . . 6
1-5 Surface Roughness, Boundary Layer Transition and Applicability to Turbomachin-
ery Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1-6 Research Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1-7 Outline of the Thesis Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Theoretical Background 11
2-1 RANS Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2-1-1 k-ω-SST Turbulence Model . . . . . . . . . . . . . . . . . . . . . . . . 13
2-2 Boundary Layer Transition and Transition Modeling . . . . . . . . . . . . . . . . 14
2-2-1 γ-Reθt Transition Model . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2-3 Effect of Surface Roughness on Turbulent Flow Fields . . . . . . . . . . . . . . . 20
2-4 Admissible Roughness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3 Validation Cases 23
3-1 Flat Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3-1-1 Smooth Flat Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3-1-2 Roughened Flat Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3-1-3 Flat Plat with Imposed Pressure Gradient . . . . . . . . . . . . . . . . . 25
3-2 Von Karman Institute Transonic Turbine Guide Vane . . . . . . . . . . . . . . . 28

Master of Science Thesis Prakhar Kapoor

iv Table of Contents

4 Compressor Performance Evaluation 33

4-1 Compressor Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4-2 Preprocessing and CFD Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4-2-1 Geometric Parametrization . . . . . . . . . . . . . . . . . . . . . . . . . 34
4-2-2 Computational Mesh and Numerical setup . . . . . . . . . . . . . . . . . 35
4-2-3 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4-3 Compressor Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4-3-1 Volute Performance Model . . . . . . . . . . . . . . . . . . . . . . . . . 37
4-3-2 Performance Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4-4 Impeller Flow Field Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4-5 Flow Phenomenon Responsible for Surge . . . . . . . . . . . . . . . . . . . . . . 41
4-5-1 Inlet Recirculation Region . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4-5-2 Flow Separation within the Impeller . . . . . . . . . . . . . . . . . . . . 42
4-5-3 Flow Separation within the Vaneless Diffuser . . . . . . . . . . . . . . . 43

5 Roughness Application and Evaluation 47

5-1 Wall Roughness Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5-1-1 Effect of Surface Roughness on Diffuser Flow Field . . . . . . . . . . . . 48
5-1-2 Effect of Surface Roughness on Impeller Flow Field . . . . . . . . . . . . 50
5-2 Localized Roughness Strips and their Effect on Flow Field . . . . . . . . . . . . 53
5-2-1 Modeling of Localized Roughness Strips . . . . . . . . . . . . . . . . . . 55
5-2-2 Roughness Parametrization . . . . . . . . . . . . . . . . . . . . . . . . . 56
5-2-3 Effects of Roughness Location and Magnitude . . . . . . . . . . . . . . . 56
5-2-4 Effect of Roughness Strips on Diffuser Flow . . . . . . . . . . . . . . . . 57
5-2-5 Effect of Roughness Strips on Impeller Flow Field . . . . . . . . . . . . . 58

6 Summary and Conclusions 65

A Reference Data for the Test Cases 69

B Matlab Code for the Volute Performance Estimation 73

C Experimental Compressor Map 75

D Component Wise Mesh Distribution 77

Bibliography 79

Prakhar Kapoor Master of Science Thesis

 List of Figures

1-1 Inducer velocity triangle and inducer stall [1] . . . . . . . . . . . . . . . . . . . . 3

1-2 Diffuser velocity triangle [1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2-1 Typical mean velocity distributions, normalized using wall variables, over smooth
and rough walls [2] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3-1 Variation of skin friction coefficient with Reynolds number for a smooth flat plate 24
3-2 Variation of skin friction coefficient with Reynolds Number for a rough flat plate 25
3-3 Computational domain for flat plate with imposed pressure gradient . . . . . . . 26
3-4 Comparison of surface static pressure distribution carried out to validate the com-
puted domain geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3-5 Effect of trip height on losses at Tu=0.5 % . . . . . . . . . . . . . . . . . . . . 27
3-6 Coefficient of pressure distribution along the plate length . . . . . . . . . . . . . 28
3-7 Computational mesh for VKI turbine guide vane . . . . . . . . . . . . . . . . . 29
3-8 Distribution of heat transfer coefficient over the VKI blade profile. . . . . . . . . 30
3-9 Skin fiction distribution over the VKI blade profile. . . . . . . . . . . . . . . . . 31

4-1 The modeled impeller geometry for the computational fluid dynamic analysis . . 34
4-2 Computational domain for the test compressor . . . . . . . . . . . . . . . . . . . 35
4-3 The mesh for computational fluid dynamic analysis . . . . . . . . . . . . . . . . 36
4-4 Comparison of the experimental and simulated performance curves for the test
compressor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4-5 Relative Mach number, meridional velocity and static entropy contours at different
streamwise direction at the best efficiency point . . . . . . . . . . . . . . . . . . 41
4-6 Relative Mach number contour at a section located at 80% of impeller spanwise
direction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4-7 Meridional velocity contours at the diffuser midplane for different flow rates . . 43
4-8 Relative velocity contours at the midspan of the impeller domain . . . . . . . . 44

Master of Science Thesis Prakhar Kapoor

vi List of Figures

4-9 Relative velocity streamlines depicting the secondary flow structure at different
operating points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4-10 Meridional velocity contours at the diffuser mid-plane for different flow rates . . 46

5-1 Meridional velocity streamlines along the diffuser meridional plane for different
surface roughness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5-2 Streamwise locations used to study the spanwise distribution of radial and tangen-
tial velocities in the diffuser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5-3 Variation of radial velocity from hub to shroud at different streamwise locations
(Fig 5-2) along the diffuser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5-4 Variation of tangential velocity from hub to shroud at different streamwise locations
(Fig 5-2) along the diffuser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5-5 Normalized total pressure contours on a section located at 70 % of the impeller
spanwise direction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5-6 Variation of meridional velocity from the inlet to the outlet of the impeller domain
mid span. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5-7 Meridional velocity contours ot a location close to the trailing edge . . . . . . . 54
5-8 Circumferential mass flux distribution at the impeller trailing edge for the different
roughness cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5-9 Modified impeller and diffuser domain for the parametric roughness study . . . . 56
5-10 Modification of the radial velocity distribution as a result of the localized roughness
applied to strip I at the diffuser . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5-11 Modification of the radial velocity distribution as a result of the localized roughness
applied to strip II at the diffuser . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5-12 Modification of the mass flux distribution as a result of the localized roughness
applied to strip 1 at the impeller . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5-13 Meridional velocity contours depicting the influence of roughness strip 1 on the
meridional velocity distribution at the impeller trailing edge . . . . . . . . . . . . 61
5-14 Modification of the mass flux distribution as a result of the localized roughness
applied to strip 2 at the impeller . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5-15 Meridional velocity contours location depicting the influence of roughness strip 2
on the meridional velocity distribution at the impeller trailing edge . . . . . . . . 62
5-16 Modification of the mass flux distribution as a result of the localized roughness
applied to strip 3 at the impeller . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5-17 Meridional velocity contours depicting the influence of roughness strip 3 on the
meridional velocity distribution at the impeller trailing edge . . . . . . . . . . . . 63

C-1 Experimental map of the test compressor . . . . . . . . . . . . . . . . . . . . . . 75

Prakhar Kapoor Master of Science Thesis

 List of Tables

3-1 Surface trip parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3-2 VKI transonic blade geometric parameters . . . . . . . . . . . . . . . . . . . . . 29
3-3 Data pertaining to the VKI turbine guide vane test cases. Mis,out and Rec,out are
the outlet Reynolds number and Mach number based on the blades chord. ReM
is the Reynolds number based on the velocity and mesh spacing of the turbulence
grid in the experimental arrangement, used to obtain specific dissipation rate ω [3] 29

4-1 Compressor specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

5-1 Magnitudes of considered roughness elements . . . . . . . . . . . . . . . . . . . 48

5-2 Roughness magnitude and location . . . . . . . . . . . . . . . . . . . . . . . . . 57

A-1 Boundary data corresponding to the smooth flat plate . . . . . . . . . . . . . . . 69

A-2 Boundary data corresponding to the roughened flat plate . . . . . . . . . . . . . 70
A-3 Boundary data corresponding to the roughened flat plate with imposed pressure
gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
A-4 Boundary data corresponding to the turbomachinery test case-VKI transonic tur-
bine guide vane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

D-1 Component wise mesh in the compressor domain . . . . . . . . . . . . . . . . . 77

Master of Science Thesis Prakhar Kapoor

viii List of Tables

Prakhar Kapoor Master of Science Thesis

 Acknowledgments

The completion of the Master thesis would not have been possible without the able assistance
of some people, whom I would like to thank.
Firstly, I would like to acknowledge my University supervisor Dr. Ir. Rene Pecnik. I consider
myself fortunate to have you as my supervisor. Without your diligence, enthusiasm and
direction, my work would not have been productive. Your approach towards problem solving
and defining the project scope helped me to streamline the work and the report itself. Thank
you very much for your support throughout the study.
Further, I am grateful to MTEE for providing me the opportunity to experience their company.
Their cooperation and assistance during the project period is truly appreciated. But, I would
like to convey a special thanks to my supervisor Dr. Adeel Javed, who was always available to
help me out during the 1 year period of mine at MTEE. I honestly appreciate your willingness
to discuss even the smallest of problems with me during the course of this work despite being
engaged with your Ph.D. defense. Moreover, a special thanks to all my colleagues at MTEE
for being generous enough to a non-Dutch speaker.
I want to express my deepest gratitude to the examination committee. I appreciate their time
devoted to reading and evaluating this document.
Also, I am grateful to Ir. Enrico Ranaldi, Ir. Uttiya Sengupta and Ir. Hassan Nemati for
their participation and out of the way help for reviewing my work. I would also take this
opportunity to thank the staff of department of Process and Energy for arranging everything
that I required during the period of my thesis work.
My special gratitude to my fellow friends and colleagues at TU Delft for their enthusiasm and
help during the testing periods during the project.
Finally, I would like to thank my parents for their selfless love and support in the completion
of the studies and the project in particular. They have been a continuous source of motivation
that has propelled me over the time. I owe everything that I am today to them. Thank you
for making it all possible

Delft, University of Technology Prakhar Kapoor

August 29, 2014

Master of Science Thesis Prakhar Kapoor

x Acknowledgments

Prakhar Kapoor Master of Science Thesis

"The only way of discovering the limits of the possible is to venture a
little way past them into the impossible."
— Arthur C. Clarke
 Nomenclature

Symbol Description Units

A Area mm2
b Width m
Cf Skin friction coefficient -
Cp Coefficient of pressure -
D Diameter mm
h Heat transfer coefficient W/m2 K
i incidence angle o

K Loss coefficient m/s

k Turbulent kinetic energy m2 /s2
Geometric roughness µm
ks Equivalent sand grain roughness mm
l∗ Length scale m
M Mach number, -
Mass flux kg/m2 /s
N Rotational Speed r.p.m
p Instantaneous pressure Pa
q Heat flux W/m2
Ra Arithmetic meanline roughness µm
Re Reynolds number -
ReT Turbulent Reynolds number -
Rev Vorticity Reynolds number -
Reθ Momentum thickness Reynolds number -
Reθt Transition onset momentum thickness -
Reynolds number
Reθt Local transition onset momentum thick- -
ness Reynolds number
r radius mm
T Temperature K
Tu Turbulence intensity %
U Blade Peripheral Velocity m/s
u Instantaneous velocity m/s

Master of Science Thesis Prakhar Kapoor

xiv Nomenclature

uτ Wall friction velocity -

u∗ Velocity scale m/s
V Absolute velocity m/s
W Relative velocity m/s
y+ Non-dimensional wall distance -

Greek symbols
β Inlet flow angle o

βb Blade angle o

 Loss coefficient -
η Efficiency -
µ Dynamic viscosity P a.s
ν Kinematic viscosity m2 /s
νt Kinematic eddy viscosity m2 /s
π Pressure ratio -
θ Boundary layer thickness mm
Circumferential angle o

ρ Density kg/m3
τ Wall sheer stress Pa
ω Turbulence eddy frequency 1/s

Subscript
0 Total or stagnation state
1 Impeller inlet
2 Impeller outlet
5 Diffuser exit
6 Return bend exit
7 Volute exit
∞ Free stream
m Meridional component
r Radial component
t Tangential component
w wall

Abbreviations
CFD Computational fluid dynamics
MTEE Mitsubishi Turbocharger and Engine Eu-
rope B.V.
PS Pressure side
SS Suction side
SST Shear stress transport

Prakhar Kapoor Master of Science Thesis

 Chapter 1

Introduction

1-1 Motivation

The stability of the centrifugal compressors is limited at low mass flow rates by surge, ro-
tating stall or a combination of both. These instabilities have a detrimental effect on the
system and should be avoided. In the case of specific application of centrifugal compressors
in the automotive turbochargers, this aspect becomes even more important. The demand
for fuel economy and emission reduction calls for engine down sizing. The automotive tur-
bochargers compensates for the performance loss accounted by the downsizing requirements.
For a turbocharged engine, some of the engine operating points lie close to the surge limit
of the compressor. Operation of the compressor at these operating points results in large
flow separations in the centrifugal compressor and eventually leads to surge. Furthermore,
it causes vibratory stresses in the blading of the compressor. Thus, it becomes essential to
devise techniques that can delay flow separation and hence increase the stability range of
the centrifugal compressor by increasing the surge margin. The current work thus focuses
on studying the effect on the flow instabilities inside the centrifugal compressor due to the
addition of roughness patches on the impeller and the diffuser.

The instabilities at low mass flow rates essentially initiate as a result of the stall. Stall can be
defined as a phenomenon, wherein the flow is separated from its path walls. A diffusing flow
along a surface or wall might retard so severely that it can no longer follow the surface. In
this case, the streamlines close to the wall will leave the wall and reverse flow regions develop
from that point along the wall surface. In other words, momentum in the streamlines close
to the wall is insufficient to overcome the adverse pressure gradient and viscous shear stress
along the wall. When the viscous shear effect and the adverse pressure gradient force the
streamlines to deviate from the surface, the flow is said to be stalled. Stage stall is said to
occur when one or more of the elements in a compressor stage gets stalled. In this case, the
overall pressure ratio vs. flow characteristic is no longer stable [1]. The subsequent literature
gives a brief introduction to the phenomenon of stall and surge.

Master of Science Thesis Prakhar Kapoor

2 Introduction

1-2 Compressor Stall and Surge

The compressor stall and surge is one of the most-significant criterion that defines the limit of
operation of the machine. The current section of the chapter tries to introduce and elaborate
upon compressor stall which forms a precursor to the surge. The aspect of the steady state
and dynamic stall has been elaborated and presented in the following sections.

1-2-1 Steady Stall

The phenomenon of stall involves flow separation along the compressor wall. Stall occurs as
a result of singular separation and/or three-dimensional separation. Singular separation can
be defined as a phenomenon where the fluid flow is affected as a result of a adverse pressure
gradient. As a consequence of the adverse pressure gradient, the fluid element close to the wall
separates resulting in the formation of a recirculating fluid flow region (separation bubble)
near the surface. The three-dimensional or ordinary separation is attributed to the presence
of secondary flow. The secondary flow results as a result of the pressure gradient in cross-
channel direction. If a secondary flow is present, it redirects the boundary layer flow such
that it cannot have a positive through-flow component. In the case of centrifugal compressors,
the ordinary separation occurs, for instance, in the vaneless diffuser where the streamlines
in proximity of the wall do not have a positive radial component, but have a strong velocity
component in the tangential direction. As a result, the dividing streamlines are directed back
towards the tip of the impeller and a back-flow condition results.
Depending on the flow conditions and geometrical features of a compressor the separation
can occur in the inducer, impeller, vaneless diffuser or the volute; which will be discussed in
more detail below.

Inducer Stall and Recirculation

The velocity triangle at the impeller inlet is shown in the Fig.1-1. According to Japikse [1], at
a given rotational speed as the mass flow rate is decreased, the inlet flow angle (β) increases
thus causing an increase in the incidence angle. As the incidence angle increases, the flow is
required to accelerate more around the leading edge. This causes a strong diffusion in the
flow in order to bring surface streamlines into balance with adjacent streamlines. This strong
diffusion causes the flow to separate from the surface at a high level of incidence.

Vaneless Diffuser Stall

The vaneless diffusers can be subjected to either singular or ordinary separation in addition to
the rotating stall (1-2-2). Singular separation is rarely encountered while ordinary separation
is very common in vaneless diffuser [1].
The velocity triangle at the diffuser is presented in Fig. 1-2. The performance of the vaneless
diffuser is given by the following equation:

rVt = constant (1-1)

Prakhar Kapoor Master of Science Thesis

1-2 Compressor Stall and Surge 3

Figure 1-1: Inducer velocity triangle and inducer stall [1]

ρ · Vm · 2πrb = m (1-2)

Vt
tanα = (1-3)
Vm

The above equations represent the conservation of mass and momentum through the vaneless
diffuser. The equations also show that the flow angle in the vaneless diffuser depends on
density and passage depth. Thus, assuming a constant density fluid the flow angle is constant
at a given flow rate resulting in a logarithmic spiral path through the vaneless diffuser. The
through-flow component of the flow tends to follow the log spiral closely. However, the
flow streamlines close to the wall have low kinetic energy. As a consequence, they have a
lower magnitude of the meridional component of the flow, while subjected to the same radial
pressure gradient. Thus, they can no longer follow the constant flow angle characteristics. In
this case, the flow separates. [1].

Impeller Stall

When the flow enters the impeller passage, boundary layer development begins at the impeller
blades, shroud and hub surfaces. As the flow proceeds through the impeller passage, both
the core flow and the boundary layers are subjected to a complex force field and a complex
flow pattern develops [4]. As the flow turns from the axial to the radial direction within a
compressors impeller, it experiences strong cross-passage force field or the Coriolis force field.
Since the Coriolis force is a function of the rotational speed and the velocity of fluid element,
it tends to separate the high-velocity and the low-velocity fluid elements. This results in
a secondary flow around the core flow with the low momentum fluid collected in proximity
of the shrouds’ suction surface while, the high momentum fluid is accumulated close to the
hub pressure surface [4]. Furthermore, the strong cross-channel Coriolis forces results in a

Master of Science Thesis Prakhar Kapoor

4 Introduction

Figure 1-2: Diffuser velocity triangle [1]

skewed boundary layer. The skewed boundary layers are likely to separate in the passage [1].
Additionally, the core flow is affected by the tip clearance flows.

Volute Stall

The volute also forms a source of the stall in centrifugal compressors. The performance of
the volute varies along a speed-line as the flow rate varies. The volute is designed so that the
flow accelerates through it for the right portion of the compressor map [1]. However, as the
mass-flow rate is reduced at a given operating speed, the flow conditions change. The flow
becomes non-accelerating and diffusing with a further reduction in mass flow rate due to the
increase in the tangential velocity [1]. Thus, the volute functions as a conical diffuser and
might enter into a stall mode.

1-2-2 Dynamic Stall and Surge

With the reduction in the mass flow rate through the compressor, two types of unsteady flow
phenomenon can be observed in a centrifugal compressor, namely dynamic stall and surge.
Dynamic stall or rotating stall is characterized by periodic flow and pressure variations in one
or more different components of the compressor [5]. Stationary stalled cells travel around the
compressor annulus in a circumferential direction. The angular velocity of the disturbance
propagation ranges from 20 % of the impeller rotational speed to a velocity larger than the
impellers’ angular velocity [6]. Rotating stall is frequently observed at the negative slope
side of the compressor performance map. The rotating stall is observed most often at low
rotational speeds. At high rotational speed, it triggers surge [4]. Dynamic stall results due
to the destabilization of the impeller flow, diffuser flow or because of an unsteady interaction
between the diffuser and impeller flow.
The most-common type of rotating stall in the case of a centrifugal compressor is observed
in a vaneless diffuser. The literature review reveals various analytical and experimental ap-
proaches to investigate the phenomenon of rotating stall in the case of diffuser of a centrifugal

Prakhar Kapoor Master of Science Thesis

1-3 Flow Separation and Boundary Layer Transition 5

compressor. According to Jansen [5], the three-dimensional boundary layer associated with
the diffuser explains the phenomenon of the rotating stall. A necessary (but not sufficient
condition) for occurrence of the rotating stall in diffuser is the three-dimensional flow sepa-
ration [5]. Similar theory is presented in the works of Senoo and Kinoshita [7], Fringe and
Van den Braembussche [8]. On the other hand, Tsujimoto et al. [9] suggested the existence
of a two-dimensional core flow instability at the instigation of rotating stall in a centrifugal
compressor.
Another form of the unsteady flow behavior that results in a centrifugal compressor is known
as surge. Surge consists of strong mass flow and pressure oscillations. It is a system phe-
nomenon and involves periodic breakdown of the flow through the compressor with periods
of complete backflow [1]. As a result, the average mass flow rate at a cross section changes
with time. In addition to the compressor fluid dynamics, surge also depends on the dynamic
properties of the system (throttle resistance, inlet and outlet ducting) in which the compres-
sor operates [6]. Surge can be further classified as mild, classic, modified and deep surge [10].
In the case of mild and classic surge, no flow reversals occur. However, periodic pressure
fluctuations are high for the classic surge whereas they are small for mild surge. Modified
surge is a combination of the classic surge and rotating stall. The deep surge is a stronger
version of the classic surge and involves flow reversals [10].

1-3 Flow Separation and Boundary Layer Transition

As pointed out in the previous section, one of the primary causes of flow instabilities in a
centrifugal compressor is separation of flow along the walls of the compressor. This section
of the chapter describes in detail about the basic phenomenon behind flow separation and
boundary layer transition.
According to Simpson [11], the term separation refers to the entire process of "departure"
or "breakaway" or "the breakdown" of the boundary layer flow. The basic phenomenon re-
sponsible for the separation of boundary layer is excessive momentum loss near the wall.
This happens in a boundary layer trying to move downstream against an increasing pressure
dp
gradient, > 0.
dx
The change in state of boundary layer from laminar to turbulent is termed as boundary layer
transition. The major factors affecting the transition process are adverse pressure gradient,
surface roughness, and free stream turbulence [12]. According to Mayle [13], there are three
important modes of transition namely: natural transition, bypass transition and separated
flow transition. Natural transition occurs at low free stream turbulence level and begins with
weak instability in the laminar boundary layer (Tollmien-Schlichting waves). These weak
instabilities are amplified at various stages till the flow becomes fully turbulent. The second
mode of transition or the bypass transition is seen in the flows involving high free stream
turbulence levels. The third mode also referred as separated flow transition occurs when
a laminar boundary layer separates under the influence of adverse pressure gradient. The
flow separation can cause a transition in the free shear layer like flow close to the surface
and subsequent reattachment [13]. A volume of recirculating fluid also known as Laminar
separation bubble (LSB) develops between the separation and the reattachment point at the

Master of Science Thesis Prakhar Kapoor

6 Introduction

surface [13]. This recirculating region of fluid might cause adverse effects such as an increase
in drag, decrease of lift force, noise and vibrations [12].

1-4 Computational Fluid Dynamics and Transition Modeling

Over the years, significant progress has been made to develop turbulence models that al-
low to simulate fully turbulent engineering flows accurately. A large domain of models has
been developed which can be employed for different applications. However, majority of the
available turbulence models do not include capabilities to model laminar-turbulent transition
accurately. The primary reason for this is that transition modeling does not provide the same
gamut of CFD-compatible models as available for turbulent flows [14].
Currently, there are methodologies available for modeling and predicting transition in the
industry. These are correlation-based models, low Reynolds number (Re) based models and
models based on stability analysis [14]. The low-Re based models make use of wall damping
terms in order to model the behavior of the viscous sublayer. These models do not require any
wall functions [3]. To model transition, they rely on the diffusion of free stream turbulence
into the boundary layer and interaction of this free stream turbulence with the model source
term. Some of the low-Re number models have been capable of predicting the transition fairly
accurately at reasonable Reynolds number. However, their applicability to predict transition
is coincidental due to the similarities between the viscous sublayer and developing boundary
layer where turbulence production is damped [15].
An alternate approach to predict transition is the en method. The model is based on the local
linear stability theory and parallel flow assumption. It computes the growth of disturbance
amplitudes from the boundary layer neutral point to the transition location [15]. One of the
inherent problems with the en method is that n-factor does not represent the disturbance am-
plitude in the boundary layer, but instead is the amplification factor from an initial unknown
amplitude. The initial amplitude of the disturbance is related to the external disturbance
environment through unknown receptivity process. Thus, the n-factor is not universal at the
transition onset and must be determined through calibration to wind tunnel or flight tests.
Furthermore, en method does not predict transition accurately in case of separated flow tran-
sition. In this case, laminar solution is separated and exhibits unsteady vortex shedding.
Thus, it is not possible in this case to compute local growth rates. Also, another drawback
of this method is its inability to predict bypass transition and surface roughness induced
transition [15].
The third methodology employed for transition modeling is the empirical correlation-based
models. The empirical correlations provide a relation between the free stream turbulence in-
tensity and transition Reynolds number based on the momentum thickness Reynolds number
(Reθt ). Inorder to employ empirical correlations for transition onset, the laminar solution
around the body of interest must be first determined. After this, the boundary layer quan-
tities are then integrated to obtain the momentum thickness Reynolds number (Reθ ). The
transition onset is thus assumed to occur when the local value of the momentum thickness
exceeds the one predicted by the correlation. Once the transition onset location is estimated,
a turbulence model is turned on, and the subsequent flow development is calculated [15].
In the current work a correlation based transition model (γ − Reθ ) has been used. The

Prakhar Kapoor Master of Science Thesis

1-5 Surface Roughness, Boundary Layer Transition and Applicability to Turbomachinery Flow 7

model is available within the commercial software package ANSYS CFX, which will be used
to perform the current study and research in perspective. The model as documented in
refs. [16] and [17], consists of two aspects. The first is a generic infrastructure that provides
two transport equations which link the CFD codes to experimental correlations. The first
transport equation is for intermittency, which is used to trigger the transition process. The
second transport equation is solved for the transition onset momentum thickness Reynolds
number. This second transport equation is an integral part of the model as it links the
empirical correlation to onset criterion in the intermittency equation. The second integral
aspect of the model are the empirical correlations.

1-5 Surface Roughness, Boundary Layer Transition and Applicabil-

ity to Turbomachinery Flow

Surface roughness has a profound impact on the boundary layer transition. The existence of
roughness causes an increase in the flow instability and accelerates the onset of transition.
Boundary layer tripping using surface roughnesses is employed in various engineering appli-
cations. For instance, boundary layer tripping is desired in scramjet as well as heat exchanger
design to enhance the mixing and heat transfer rates. Laminar separation on aerodynamic
surfaces can be avoided by increasing the surface roughness to cause an early transition.
Separation can occur from localized adverse pressure gradients and results in a reduction
in aerodynamic efficiencies of these surfaces as a consequence of the increase in the pressure
drag. In the situations, where the laminar separation is about to occur, tripping the boundary
layer so that it remains attached is preferable. In the case of the separated flows, transition
of laminar boundary layer is desired because of the ability of the turbulent boundary layer
to remain attached as a result of the higher momentum as a consequence of the turbulent
mixing [18]. In the view of Jonas et al. [19] the flow structure at rough surfaces assumes
local flow separations. Wake regions composed of counter-rotating vortex pairs originating at
the surface of roughness elements attached to the basic surface. These separations result in
local pressure distributions, which cause a local form of drag acting on the surface together
with the viscous wall shear stress [19]. It also results in a roughness sublayer. Vortices’ that
have a length scale proportional to the length scale of roughness elements are shed into the
flow above the crest of the elements. At a distance, the roughness induced sublayer structure
blends into the flow. Furthermore, the vortices’ are suppressed by viscosity at low Reynolds
number, and they accelerate the transition process with the increasing Reynolds number.
A lot of work has been done in investigating the effect of surface roughness on turbulent flows.
However, the influence of roughness on transition has been primarily limited to the effect of
isolated roughness elements in tripping the boundary layer.
The effect of surface roughness on the turbulent flow structure was closely studied by Nikurasde
[20]. His pressure loss data obtained with sand roughened pipe walls revealed the dependen-
cies of different flow regimes on different surface roughness and Reynolds number. In his
work, Nikuradse defined a dimensionless parameter, k + = ksνv∗ , based on the diameter of the
sand grain (ks ), wall friction velocity (v∗ ) and dynamic viscosity (ν) [21]. For the value of this
parameter greater than 70, the loss coefficient is only a function of ks . Where as for values of
ks between 5 and 70, both Re and k + form important parameters [21]. Based on his results,

Master of Science Thesis Prakhar Kapoor

8 Introduction

Nikuradsi [20] defined three flow regimes. In the Hydraulically smooth regime(0 < ks < 5),
size of the roughness is so small that it does not have a considerable impact on pressure loss.
In the transition regime (5 < k + < 70), the resistance elements protrude out of the viscous
sublayer and additional resistance is experienced due to the additional drag. In the completely
rough regime k + > 70, all the roughness elements protrude outside the viscous sublayer and
largest part of resistance to flow is in the form of a drag. Furthermore, Schlichting formulated
the concept of equivalent sand grain roughness as a means of converting the roughness data
and measurements corresponding to other flow profiles into a roughness profile equivalent to
the roughness data used by Nikuradse [22]. Colman et al. [23] provided further corrections
to the assumptions made by Schlichting in his work and correlations have been developed by
Simpson [24].
The literature pertaining to influence of surface roughness on laminar to turbulent boundary
layer is limited. Doenhoff and Horton [25] reported that in case of an aerofoil, distributed
roughness at a specific location downstream the stagnation point resulted in an earlier tran-
sition. However, it was also apparent that a similar effect could also be achieved through
small strips of roughness. Pinson and Wang [26] showed that the roughness height and not
the roughness geometry is the primary parameter in case of the roughness induced effects.
With the increase in the velocity, the earlier transition onset and heat transfer enhancement
in case of a turbulent boundary layer are roughness dependent.
Zhang and Hudson [27] conducted a parametric study of roughness elements including rough-
ness type size and location under steady and unsteady flow conditions on a flat plate with
adverse pressure gradient. The boundary layer was subjected to the pressure distribution
representative of an ultra high lift low-pressure turbine. In their work, they reported that
the combined effect of the surface roughness and unsteady wakes can reduce the profile losses
of ultra highly loaded LP turbine blades. For these blades, a large separation is formed due
to high-pressure gradient and low wake passing frequency. There is an optimum height for
roughness elements. These elements do not induce transition immediately after itself but only
aids in hastening transition in the separated shear layer. In addition to this, they reported
that the step type roughness elements more effective at inducing boundary layer transition
as compared to wire-type roughness elements.

1-6 Research Questions

As introduced in the previous section, the operational range of the centrifugal compressors is
limited at low mass flow rates by surge. The phenomenon of surge essentially stems from the
flow instabilities in the form of separation and reversals that occur within the compressors
flow domain. In the scope of this thesis, the aerodynamic phenomenon resulting in a re-
duction in stable operating range of the centrifugal compressor are identified and quantified.
Furthermore, the following research topics will be addressed:

• Validation of transition.

• Validation of transition with surface roughness.

• Performance prediction of a smooth walled compressor.

Prakhar Kapoor Master of Science Thesis

1-7 Outline of the Thesis Report 9

• Effect of surface roughness on the flow instability in a centrifugal compressor.

These research questions will be addressed by performing steady state simulations on a com-
mercial CFD tool ANSYS CFX.

1-7 Outline of the Thesis Report

Chapter 2 presents the theoretical background pertaining to turbulence modeling, boundary

layer transition, numerical modeling of the transition and effect of surface roughness on flow
fields. Brief overview is presented related to Reynolds averaged Navier-Stokes equation, the
formulation of low Reynolds number k-ω SST turbulence model, the γ − Reθ transition model
and admissible surface roughness. Chapter 3 discusses and validates the ability of the
transition model formulation available in CFX by validating the results of various test cases,
Chapter 4 deals with the computational set up in ANSYS CFX to simulate the performance
of the test compressor. Furthermore, it provides a brief description about the inception of
flow instabilities in a centrifugal compressor. The report proceeds to Chapter 5 where the
effect of surface roughness on the flow instabilities in a centrifugal compressor is investigated.
A parametric roughness study is presented in this chapter to gain further insights on the
effect of roughness location and height. The report is completed with Chapter 6 providing
the conclusions and recommendations.

Master of Science Thesis Prakhar Kapoor

10 Introduction

Prakhar Kapoor Master of Science Thesis

 Chapter 2

Theoretical Background

This chapter reviews the literature pertaining to turbulence modeling, boundary layer tran-
sition and influence of surface roughness on transition. Brief discussion of the turbulence
modeling and the model (k-ω SST) is presented in section 2-1. Furthermore, a study of the
phenomenon of laminar to turbulent transition and its aspect of numerical modeling is pre-
sented in section 2-2. Next, the effect of roughness on turbulent flow field is discussed in
section 2-3 Finally, the chapter ends by a review of the influence of surface roughness on the
laminar to turbulent transition in 2-4.

2-1 RANS Modeling

The motion of a fluid can be completely described in terms of its continuity and momentum
equations. These equations collectively are referred as the Navier-Stokes equation. Neglecting
the effects of body forces, frame rotation and denoting density, kinematic viscosity, pressure
and velocity by ρ, ν, p, v respectively, the Navier-Stokes equation for an incompressible fluid
is given by:

∂ui
=0 (2-1)
∂xi

∂ui ∂ui 1 ∂p ∂ 2 ui
+ uj =− +ν 2 (2-2)
∂t ∂xj ρ ∂xi ∂ xj

The equation (2-1) represents the conservation of mass, while the equation (2-2) relates to
the momentum conservation of the fluid volume. Despite the availability of sophisticated and
powerful computational resources, the estimation and resolution of all the scales of turbulent
flows is a challenging affair. The Reynolds-averaged Navier-Stokes equations or simply the
RANS equation offer a solution to the above problem. The RANS equation is based on the
concept of Reynolds decomposition, which involves the decomposition of an instantaneous

Master of Science Thesis Prakhar Kapoor

12 Theoretical Background

quantity into a mean and fluctuating component. Following the Reynolds decomposition the
instantaneous velocity u and the instantaneous pressure p can be expressed as:

0 0
ui = ui + u p=p+p (2-3)
0 0
where, ui is the mean velocity, u is the fluctuating velocity , p is the mean pressure and p is
the pressure fluctuation. Substituting the above decomposition in the Navier Stokes equation
(equations 2-1 and 2-2) results in the following equation:

∂ 0
(ui + ui ) = 0
∂xi
∂ 0 0 ∂ 0 1 ∂ 0 ∂2 0
(ui + ui ) + (uj + uj ) (ui + ui ) = − (p + p ) + ν 2 (ui + ui ) (2-4)
∂t ∂xj ρ ∂xi ∂ xj

Applying the Reynolds decomposition followed by statistical averaging to equation 2-4, the
above equation reduces to
∂ui
=0
∂xi
0 0
∂ui ∂ui 1 ∂p ∂ 2 ui ∂ui uj
+ uj =− +ν 2 − (2-5)
∂t ∂xj ρ ∂xi ∂ xj ∂xj

The above set of equations is referred as Reynolds-averaged Navier-Stokes Equation. The

obtained equation is similar to the original Navier-Stokes equation, except for an additional
last term. The extra term designates Reynolds stress. For a statistically three-dimensional
flow, four independent equations govern the mean velocity field, the three components of the
Reynolds equation and along with the mean continuity equation [28]. However, these four
equations contain more than four unknowns. Apart from u and p, there are also the Reynolds
stresses. Such set of equations with more unknowns than the equations are referred as un-
closed. This formulates the closure problem of the Reynolds equations i.e, these equations
cannot be solved unless the Reynolds stresses are somehow determined [28].
The Reynolds stress terms directly arise from the flow and the length scales of the perturba-
tions, which are the typical scales we are dealing. Thus, it becomes impossible to assume a
universal value for all the cases and thus they need to be modeled accurately [28]. In order
to solve the problem, the turbulent viscosity hypothesis proposed by Boussinesq is used. The
Boussinesq hypothesis relates the deviatoric Reynolds stress to the mean rate of flow.

0 0 2 ∂ui ∂uj
ui uj = k − νt ( + ) (2-6)
3 ∂xj ∂xi

where, νt is the turbulent viscosity or eddy viscosity and k is the kinetic energy of fluctuations.
The above equation (2-6) provides a closure to the Reynolds equations (2-5) provided that
an appropriate specification of the turbulent viscosity (νt ) is determined. Furthermore, the
turbulent viscosity (νt ) can be expressed as a function of the length scale (l∗ ) and velocity
scale (u∗ ), according to the following expression

νt = u∗ l∗ (2-7)

Prakhar Kapoor Master of Science Thesis

2-1 RANS Modeling 13

Thus, to provide an estimate of the turbulence viscosity, the length scale and the velocity scale
must be specified. Different turbulent models achieve this in different ways. For instance,
in case of the mixing-length model, the length scale is defined on the basis of the geometry
of the flow. Whereas, in case of the two-equation k- model, the length scale and velocity
scale are related to the turbulence kinetic energy (k) and dissipation rate () for which the
transport equations are solved.
The k-ω-SST turbulence model is used for the current work. The model is described in detail
in the subsequent section.

2-1-1 k-ω-SST Turbulence Model

The k-ω-SST as proposed by Menter [29] is a two-equation eddy viscosity model that effec-
tively blends the robust formulation of k-ω model in the near-wall region with the far stream
independence of k- model. To achieve this, the k- model is converted to a k-ω formula-
tion. The standard k-ω model and the altered k- model are first multiplied by a blending
function and then added together. The blending function takes a value of one in the near
wall region, which switches k-ω formulation. Away from the wall, the value of the function
is zero, which activates the k- formulation. Furthermore, in the low Reynolds number SST
model the definition of turbulent viscosity is modified to account the turbulent shear stress
transport.
The original k-ω model is given by:
" #
Dρk ∂ui ∂ ∂k
= τij − β ∗ ρkω + (µ + σk1 µt ) (2-8)
Dt ∂xj ∂xj ∂xj

" #
Dρω γ1 ∂ui ∂ ∂ω
= τij − β1 ρω 2 + (µ + σω1 µt ) (2-9)
Dt νt ∂xj ∂xj ∂xj
The modified k- model is given by:
" #
Dρk ∂ui ∂ ∂k
= τij − β ∗ ρkω + (µ + σk2 µt ) (2-10)
Dt ∂xj ∂xj ∂xj

" #
Dρω γ2 ∂ui ∂ ∂ω 1 ∂k ∂ω
= τij − β2 ρω 2 + (µ + σω2 µt ) + 2ρσω2 (2-11)
Dt νt ∂xj ∂xj ∂xj ω ∂xj ∂t
In order obtain the shear stress formulation, the transport equations corresponding to the
original k-ω formulation, (2-8) and (2-9), are multiplied by a blending function F1 , while
the equation (2-10) and (2-11) are multiplied by (1-F1 ) and the corresponding equations are
summed up.
The new model obtained as a consequence of above transformation is represented by the
following set of equations:
" #
Dρk ∂ui ∂ ∂k
= τij − β ∗ kρω + (µ + σk µt ) (2-12)
Dt ∂xj ∂xj ∂xj

Master of Science Thesis Prakhar Kapoor

14 Theoretical Background

" #
Dρω γ ∂ui ∂ ∂ω 1 ∂k ∂ω
= τij − βρω 2 + (µ + σω µt ) + 2ρ(1 − F1 )σω2 (2-13)
Dt νt ∂xj ∂xj ∂xj ω ∂xj ∂xj

Any constant, φ is a blend of inner and outer constants, and is given by φ = F1 φ1 +(1−F2 )φ2 .
Furthermore, the turbulent eddy viscosity for the model is computed from

a1 k
νt = (2-14)
max(a1 ω, ΩF2 )
with a1 being a constant, Ω being the absolute vorticity and F2 is a function that takes a
value of zero for the free sheer flows and one for boundary layer flows.
The constants used in the k-ω-SST model are:
Inner constants (SST inner)

β∗ = 9
100 , σω1 = 0.5, β1 = 0.0750, a1 = 0.31, σk1 = 0.85, k = 0.41
Outer constants (standard k-)

β∗ = 9
100 , σω2 = 0.856, β2 = 0.0828, σk2 = 1, k = 0.41
In addition to the above constants, the following auxiliary relations are used in the model
formulation: (  √ 4 )
k 500ν 4σω2 k

F1 = tanh min max ∗ , 2 , (2-15)
β ωy y ω CDkω y 2

!
1 ∂k ∂ω
CDkω = max 2ρσω2 , 10−20 (2-16)
ω ∂xj ∂xj

(  √ 2 )
2 k 500ν
F2 = tanh max ∗ , 2 (2-17)
β ωy y ω

β1 σω1 k 2
γ1 = − √ (2-18)
β∗ β∗

β2 σω2 k 2
γ2 = − √ (2-19)
β∗ β∗

2-2 Boundary Layer Transition and Transition Modeling

The prediction of boundary layer transition is an aspect in the development of computational

fluid dynamics [14]. Transition can be defined as the change in state of a laminar boundary
layer to turbulent. As mentioned in section 1-3, boundary layer flow over different objects
may transition through different mechanisms in different applications. Despite the availability
of a gamut of the turbulence models that allow a precise simulation of fully turbulent flows,

Prakhar Kapoor Master of Science Thesis

2-2 Boundary Layer Transition and Transition Modeling 15

not many CFD compatible models are available to predict the aspect of laminar to turbulent
transition.
Transitional boundary layer flows are important in many applications of engineering interests,
such as turbine blades, aircraft wings, ship hulls, etc. Thus, modeling of transitional flows
becomes a critical and integral aspect of computational fluid dynamics. The γ−Reθt transition
model formulated by Menter et al. [14] provides a generic infrastructure that allows the
coupling of the general purpose CFD with experimental transition data. A detailed description
about the model is provided in the subsequent literature.

2-2-1 γ-Reθt Transition Model

The transition model as formulated by Menter et al. [14] consists of two components. The first
is a universal framework provided by two transport equations that allow experimental data to
be linked with the CFD code. The second integral component consists of the correlations. The
model formulation involves two transport equations: an equation for intermittency (γ) and a
second transport equation which is solved in terms of transition onset momentum-thickness
Reynolds number (Reθt ). In this work, the transport equation formulation is described in five
parts. First part deals with the description about the intermittency transport equation. The
second part pertains to the transport equation for the transition onset momentum thickness
Reynolds number. The part section explains the modification that is used to improve the
separated flow transition prediction. The fourth part gives a brief overview of the correlations
that are employed along with the model. Lastly, the fifth part provides a link between the
transition model and SST model.

Transport equation for intermittency

The transport equation for intermittency is used to trigger the transition locally. In the
current model formulation, the intermittency function is coupled with k-ω SST model [14].
The transport equation essentially is used to activate the production term of turbulent kinetic
energy downstream of the transition point in the boundary layer.
The intermittency equation is formulated as follows:
" ! #
∂ ∂ ∂ µt ∂γ
(ργ) + (ρuj γ) = Pγ − Eγ + µ+ (2-20)
∂t ∂xj ∂xj σf ∂xj

The following equation describes the source term for transition:

p
Pγ1 = Flength · ca1 ρS · γFonset · (1 − γ) (2-21)

where, S is the strain rate magnitude. The transition source term is devised to be zero in the
laminar boundary layer, upstream of the transition point and is active wherever the value of
local strain rate Reynolds number exceeds the local transition onset criterion. The magnitude
of the source term is controlled by transition length function i.e. Flength . This term governs
the length of the transition region. The source term limits the value of intermittency, so that
it does not exceed unity.

Master of Science Thesis Prakhar Kapoor

16 Theoretical Background

The Fonset function is used to trigger the intermittency production in the equation (2-21).
This function is designed to change from the value of zero in the laminar boundary layer to a
value of one downstream of the transition onset location. Fonset is a function of the turbulent
Reynolds number (ReT ) and the local strain rate Reynolds number and is expressed as:

Rev
Fonset1 = (2-22)
2.193Reθc

4
Fonset2 = min(max(Fonset1 , Fonset1 ), 2.0) (2-23)
3 !
ReT

Fonset3 = max 1 − ,0 (2-24)
2.5

Fonset = max(Fonset2 − Fonset3,0 ) (2-25)

ρy 2 S ρk
where, Rev = µ and ReT = µω
Reθc in the equation (2-22) is defined as the critical Reynolds number where the intermittancy
first increases in the boundary layer. This happens upstream of the transition Reynolds
number(Reθt ). There is a delay because turbulence must increase to a substantial level before
laminar profile shows a change. Thus, Reθc can be considered as a location where turbulence
starts to grow, while Reθt can be considered as the location where a deviation from the
laminar velocity profile occurs. The relation between these two is obtained from empirical
correlations where, Reθc = f (Reθt ). Reθt is obtained from the transport equation (2-2).
Furthermore, the destruction/relaminarization term (Eγ ) in equation (2-20) acts as a sink
term. The term ensures that the intermittency remains close to zero in the laminar boundary
layer and also enables the model to predict relaminarization. The term is defined by the
following equation:
Eγ = ρca2 Ωγ · Fturb · (γce2 − 1) (2-26)
where, Ω is the vorticity magnitude, ca2 is a constant that controls the strength of the de-
struction term and ensures that the entire term is smaller than the transition source term.
The constant ce2 controls the lower limit of intermittency where the term changes sign [14].
In order to disable the destruction/relaminarization source term in-case of a fully turbulent
regime, Fturb is used. This term is defined as :

ReT 4
Fturb = e−( 4
)
(2-27)

For intermittency (γ), the boundary condition at the wall is zero normal flux. At the inlet, γ
is assigned a value 1. The constants for intermittency equations are:
ca1 =2.0; ce2 =50; ca2 =0.06; σf =1.0;
The subsequent section describes the formulation of the transportation equation for the tran-
sition momentum thickness Reynolds number used in the transition model formulation. The
equation forms an integral part of the model as it ties the empirical correlations to the onset
criterion defined in the intermittency equation.

Prakhar Kapoor Master of Science Thesis

2-2 Boundary Layer Transition and Transition Modeling 17

Transport equation for transition onset momentum thickness Reynolds number

The second transport equation in the model formulation is solved in terms of the transition
onset momentum thickness Reynolds number (Reθt ). This equation is required to capture
the non-local influence of turbulence intensity, which changes as a consequence of decay of
turbulent kinetic energy in the free stream and also due to the changes in free stream velocity
outside the boundary layer [14].
The transport equation for the transition momentum thickness Reynolds number is given by:

" #
∂ ∂ ∂ ∂Reθt
(ρReθt ) + (ρuj Reθt ) = Pθt + σθt (µ + µt ) (2-28)
∂t ∂xj ∂xj ∂xj

where, Pθt is the production term and is defined by the equation:

ρ
Pθt = cθt · · (Reθt − Reθt ) · (1 − Fθt ) (2-29)
t
where t = 500µ
ρu2
, is a time scale. The blending function Fθt in the above equation is used to
deactivate the production term in the boundary layer. The blending function is assigned a
value of one in the boundary layer and zero in the free stream.

The production term is designed to force the transported scalar (Reθt ) to match the local
value of Reθt outside the boundary layer. The local value of Reθt is computed from empirical
correlation [14].
The blending function (Fθt ) is given by the following set of equation

 
1
!2  
y 4 γ−
Fθt = min max Fwake · e−( ) , 1.0 −δ
Ce2
1
 , 1.0 (2-30)
1− Ce2

Reθt µ 15 50ωy
θBL = δBL = θBL δ= δBL (2-31)
ρu 2 u

ρωy 2
Reω = (2-32)
µ

Reω

2
−
Fwake = e 10−5 (2-33)

The function Fwake is used to ensure that the blending function is not activated in the wake
regions of a blade or an aerofoil. cθt and σθt are the model constants, where cθt = 0.03 controls
the magnitude of source term and σθt = 2.0 controls the diffusion coefficient.

Master of Science Thesis Prakhar Kapoor

18 Theoretical Background

Separated Flow Prediction

For the laminar boundary layer separation, the model formulation discussed above predicts
the turbulent reattachment at a downstream location. In order to improve this inaccuracy,
a modification is proposed by to the transition model that allows turbulent kinetic energy to
grow rapidly once the laminar boundary layer separates [14]. The modification has negligible
effect on the prediction for attached transition or fully turbulent flow. The modification
is incorporated into the blending function, Fθt from the transport equation for momentum
thickness Reynolds number (2-28). The following modification is introduced into the model,

Reν
     
γsep = min s1 max 0, − 1 Freattach , 2 Fθt (2-34)
3.235Reθc
 4
ReT
− 20
Freattach = e (2-35)
γef f = max(γ, γsep ) (2-36)

where, s1 = 2
The basic point behind the above formulation is to allow the local intermittancy value to
exceed above one whenever laminar boundary layer separation occurs. This causes a sudden
increment in the production of turbulent kinetic energy (k) which in turn affects earlier
transition. As seen in equation (2-34), this is achieved when the strain rate Reynolds number
(Reν ) exceeds the critical momentum thickness Reynolds number (Reθc ).
In the modified equation (2-34) for separated flow prediction, the constant s1 controls the
size of the separation bubble. Furthermore, once the viscosity ratio is large enough to cause
reattachment, the function Freattach deactivates the modification.

Model Correlations

In the γ-Reθt transition model formulation, three empirical correlations are defined for Reθt ,
Flength and Reθc . Reθt is used in equation 2-2 and establishes the transition onset. While
Flength is used in equation (2-2) and defines the length of the transition region, Reθc forms
the point where the transition model is activated. Flength is also a function of Reθt and is
given by the following correlation:

 2

 398.189 · 10−1 + (−119.270 · 10−4 )Reθt + (−132.567 · 10−6 )Reθt , Reθt < 400
2

 −2 −5
263.404 + (−123.939 · 10 )Reθt + (194.548 · 10 )Reθt



3
Flength = +(−101.695 · 10− 8)Re ,θt 400 ≤ Reθt < 596

− − · 3.0 · 10−4 , 596 ≤ Reθt < 1200




 0.5 (Reθt 596.00)

0.3188, Reθt ≥ 1200


As mentioned in Subsection (2-2-1), Reθc is a function of transition momentum thickness

Reynolds number (Reθt ). The exact relation between the two is given by the following em-
pirical correlation:

Prakhar Kapoor Master of Science Thesis

2-2 Boundary Layer Transition and Transition Modeling 19


2


 (−369.035 · 10−2 )Reθt + (120.656 · 10−4 )Reθt + (868.230 · 10−6 )Reθt
+(−696.506 · 10− 9)Re3 + (174.105 · 10− 12)Re4 ,

Reθt ≤ 1870

θt θt
Reθc =




Reθt − (593.11 + (Reθt − 1870) · 0.482), Reθt > 1870



Lastly, Reθt is also a function of the turbulence intensity (Tu), Thwaite’s pressure gradient
coefficient λθ and flow accelration parameter (K) [16]. The following empirical correlation
defines the relation between these quantities:
Reθt = 803.73[T u + 0.6067]−1.027 F (λθ , K) where,



 1 − (−10.32λθ − 89.47λ2θ − 265.51λ3θ )e(−T u/3.0) , λθ ≤ 0



F (λθ , K) =


 1 + (0.0962(K106 ) + 0.148(K106 )2 + 0.0141(K106 )3 )


 (−T u/1.5)
·(1 − e (−23.9λθ ) (−T u/1.5)
) + 0.556(1 − e )e , λθ > 0
√
2k/3
where, Tu is the local turbulence intensity and is defined as T u = 100 U and λθ is the
pressure gradient coefficient and is given as λθ = ρθ dU
µ ds .

Coupling with the SST model

The transition model is coupled to the SST turbulence model by making use of the effective
intermittency defined in equation 2-2-1. The production and dissipation terms in the k
equation are modified as follows:

P k = γef f Pk (2-37)

Dk = min(max(γef f , 0.1), 1.0)Dk (2-38)

where Pk and Dk are the production and destruction terms from the turbulent kinetic energy
equation in the original SST turbulence model and γef f is the effective intermittency obtained
from the transition formulation.
Another modification incorporated into the SST turbulence model includes a change in the
blending function F1 . The blending function F1 is redefined such that it always remains equal
to 1 in laminar boundary layer. The modified blending function is given as:
√
ρy k
Ry = (2-39)
µ
 8
Ry
− 120
F3 = e (2-40)
F1 = max(F1,orig , F3 ) (2-41)

where, F1,orig is the orignal blending function employed in the SST model.

Master of Science Thesis Prakhar Kapoor

20 Theoretical Background

2-3 Effect of Surface Roughness on Turbulent Flow Fields

The wall bounded flow is considered to be composed of an inner layer and an outer layer.
The inner layer is defined as the region where the mean velocity profile is determined by the
viscous scales and is independent of the flow length scale and bulk velocity. In case of a wall
bounded channel flow the viscous scales are defined as- friction Reynolds number, Reτ = uντ δ
and viscous length scale δv = uντ . In case of the outer layer, the direct effect of the viscosity
on the mean velocity profile is assumed to be negligible.
In general, for near wall turbulent flow theories, postulate an velocity profile characterized by
a overlap region where the mean flow is given by [30]:

1
u+ =ln(y + ) + B (2-42)
k
where u+ = u/uτ , y+ = y/δv and B is a constant.
However, in presence of roughness, particularly when the scale (s) of the roughness element
is large compared to the viscous scale δv , the local Reynolds number of the flow over the
roughness elements is large. The transfer of momentum from the fluid to the wall is accom-
plished by drag on roughness elements, which at high Reynolds number is predominantly due
to pressure forces rather than viscous stresses [28]. In such cases the velocity profile in the
overlap region is modified as;
1 y s
u+ = ln( ) + B( ) (2-43)
k s δv
In case of a fully rough wall bounded pipe flow the constant B takes a value of 8.5.
The difference between the mean velocity profile on a smooth surface and rough surface is
given by the following equation [30]:

1 s
∆u+ = ln( ) − 3.5 (2-44)
k δv
∆u+ represents the shift in the velocity profile as shown in figure 2-1.
Furthermore, in case of a boundary layer the mean velocity profile is given by the equation:

1 2π y
 
u+ = lny + + B − ∆u+ w (2-45)
k k δ
The above equation consists of the overlap region profile, the roughness shift ∆u+ and the
wake function w(y/δ). The parameter π in the equation (2-45) is defined as the wake strength
parameter and its value is flow dependent. While δ in the above equation is the boundary
layer thickness. The mean velocity profile can be more conveniently described in terms of its
deviation from the value at the edge of the boundary layer, i.e at y = δ,

2π y 1 y
    
u+ +
e −u = w(1) − w − ln (2-46)
k δ k δ
The mean velocity defect is dependent on two parameters for a given wake function- wake
strength parameter (π) and the friction velocity (uτ ). Both these parameters are influenced
by the surface roughness.

Prakhar Kapoor Master of Science Thesis

2-4 Admissible Roughness 21

Figure 2-1: Typical mean velocity distributions, normalized using wall variables, over smooth
and rough walls [2]

2-4 Admissible Roughness

The admissible roughness can be defined as the maximum height of the individual roughness
elements which causes no increase in the drag as compared to a smooth wall [22]. As men-
tioned in the previous chapter, in case of turbulent boundary layers, surface roughness has no
effect and the wall can be considered as hydraulically smooth, if the maximum height of the
roughness elements lies within the laminar sublayer. In case of a pipe flow the value of the
non dimensional parameter determines whether the wall is hydraulically smooth. In general,
if the non dimensional parameter, k + = ν∗νks < 5, the wall can be considered as hydraulically
smooth.
The above correlations can also be considered for a flat plate. For a flat plate however, the
following correlation can be used to estimate the admissible value of surface roughness [22]:
ν
kadm ≤ 100 (2-47)
U∞
where U∞ is the free stream velocity.
The above equation provides one value for the admissible roughness for the whole plate. But,
as the boundary layer thickness is smaller near the leading edge, the admissible value of the
roughness should be smaller near the leading edge as compared to the trailing edge. This
is taken into account by modifying the above equation (2-47) and introducing the local skin
0
friction coefficient (cf ). The modified equation is given as :

U∞ kadm 7
<q 0 (2-48)
ν cf

However, for practical applications the admissible value of roughness is directly related to the
length of the body (l), under consideration [22]. To achieve this the equation (2-47) is written
as:
100
kadm ≤ l (2-49)
Rl

Master of Science Thesis Prakhar Kapoor

22 Theoretical Background

U∞ l
where, Rl = ν
The height of the roughness element that causes a transition in a laminar boundary layer
is called critical height or critical roughness [22]. Surface roughness affects the resistance
offered by the wall by causing the transition point to move upstream. This causes the drag to
increase or decrease depending upon the shape of the body under consideration. In general,
the drag may be increased by such a shift in the transition point when the drag is due to the
skin friction, while it may be decreased when the drag of the body is due to form drag [22]
The critical value of roughness is defined by [51]:

v∗ kcrit
= 15 (2-50)
ν

Prakhar Kapoor Master of Science Thesis

 Chapter 3

Validation Cases

In the current chapter, the ability of the transition model available in ANSYS CFX to predict
onset of the transition is validated for different 2D test cases and their results have been
elaborated. The first test case involves modeling of flow over an adiabatic flat plate [31].
This is followed validating the results on flow over a flat plate with a distributed surface
roughness [32] and a plate with imposed pressure gradient [27]. The above cases are performed
in order to understand the implementation of the transition model and its ability to capture
roughness induced transition. Finally, a turbo-machinery test case has been evaluated wherein
the variation of heat transfer coefficient and skin friction coefficient over a VKI transonic guide
vane [3, 33] has been modeled and validated.

3-1 Flat Plate

In order to investigate the capability of the transition model available in ANSYS CFX, three
test cases have been simulated for the flat plate study. The first test case corresponds to the
flow over a smooth flat plate. The second test case involves the study of flow over a roughened
flat plate. This study has been carried out to investigate the capability of the commercial
CFD software to capture the effects of surface roughness on boundary layer transition. Finally,
the loss coefficients have been evaluated for the flow over a flat pate with imposed pressure
gradient and surface trips.

3-1-1 Smooth Flat Plate

The purpose of this test case is to compare the prediction of the transition (γ − Reθ ) model
formulation available in ANSYS CFX with the experimental results available for subsonic flow
over a smooth flat plate. The flow conditions compare with the test case T3A, documented
by Savill [31]. The boundary conditions for the test case are documented in Appendix A.
The computational grid is provided by Dr. Rene Pecnik and comprises of 5120 control volumes
divided by a H-type grid which ensures a resolution of y + < 0.3 at the walls [3]. The results

Master of Science Thesis Prakhar Kapoor

24 Validation Cases

Figure 3-1: Variation of skin friction coefficient with Reynolds number for a smooth flat plate

obtained are presented in terms of the distribution of the skin friction coefficient cf over the
plate surface as depicted in Fig. 3-1. The skin friction coefficient is defined as
τw
Cf = 2
(3-1)
0.5 ∗ ρ∞ ∗ U∞

where τw = µ dU
dy is the wall shear stress, U∞ and ρ∞ are the freestream density and velocity,
respectively.
The figure 3-1 depicts a sudden increase in the skin friction coefficient for a certain range of
Reynolds number (Rex =150000 to 270000). This corresponds to the transition from laminar
flow regime. Furthermore, the results obtained from CFD solver are in a close agreement
with the experiments which justifies the capability of the model to predict transition onset.

3-1-2 Roughened Flat Plate

In the present section, the results of a two dimensional flow over a roughened flat plate are
presented. The motivation behind this test was to determine the capability of the transition
model available in CFX to model roughness induced transition.
The test case corresponds to the work of Dassler et.al [32]. The computational domain consists
of 5120 elements and provides an average y + ≈ 0.95 at the wall. To validate the case, a sand
grain roughness height of 30 microns has been applied to the flat plate. The corresponding
flow conditions and boundary condition are documented in Appendix A.
In their, work Dassler et.al [32] have equivalent surface roughness (ks ) as the measure of
the roughness at the surface. However, since the transition model formulation in CFX also
requires the specification of the geometric roughness height the following correlation as men-
tioned in their work [32] has been used:

ks = 4.433k (3-2)

Prakhar Kapoor Master of Science Thesis

3-1 Flat Plate 25

(a) Smooth Wall Case (b) Wall Roughness =30 µm

Figure 3-2: Variation of skin friction coefficient with Reynolds Number for a rough flat plate

Fig.3-2 shows the comparison between the skin friction coefficient obtained from the current
simulations and the simulation results documented in the work of Dassler et.al [32]. It can be
seen that the presence of roughness causes a shift in the transition location. The transition
onset location shifts from Rex =750000 to Rex =385000. The simulations carried out in the
current work agree reasonably well for the smooth case,as depicted in Fig. 3-2(b). However,
in the case of the roughened flat (Fig. 3-2(a)) the simulations predict an early transition and
reattachment as compared with the baseline reference case. This deviation can be attributed
to the approximate correlation used to relate the equivalent surface roughness height to the
geometric roughness height for the roughness elements. It can be inferred that the presence
of surface roughness results in an earlier transition.
The results obtained in this section are in a agreement with the simulation results of the
reference case [32] and shows the influence of the surface roughness to cause the transition
from laminar to turbulent flow regime at a lower Reynolds Number. Thus, this gives a
validation to the ability of the model in CFX to predict roughness induced transition.

3-1-3 Flat Plat with Imposed Pressure Gradient

In the current test case, the effect of surface trips on the boundary layer development on an
ultra-high-lift low pressure LP turbine blade is considered. The objective of the study is to
model the capability of the CFD solver to predict transition under the combined effect of
the surface roughness and adverse pressure gradients encountered in typical turbomachinery
applications. The parametric study is conducted on a flat plate with the same surface pressure
distribution as encountered in case of a LP turbine blade. For this case, the experimental
results of Zhang and Hodson [27] have been validated.
To obtain the same surface distributions on the flat plate as encountered on the suction side
of the LP turbine blade, the geometry for the upper-wall of the simulation domain is obtained
using the experimental results of the inviscid coefficient of pressure distribution over the test
plate. This is done using the Bernoulli’s equation for incompressible fluid. The computational

Master of Science Thesis Prakhar Kapoor

26 Validation Cases

Figure 3-3: Computational domain for flat plate with imposed pressure gradient

Table 3-1: Surface trip parameters

S No Roughness Height k(mm) Normalized Height k/SO (%)

1 1.20 0.24
2 0.81 0.86
3 0.74 0.15
4 0.68 0.14
5 0.61 0.12

domain consists of 18000 elements and the domain for the current test case is presented in
Fig. 3-3 .
Fig.3-4 shows the comparison between the experiments and simulated coefficient of pressure
(Cp ) distribution for the in-viscid and the turbulent case. It can be seen that the simulated
results for the surface distribution of the coefficient of pressure agrees reasonably well with
the experimental results. This validates the computed domain geometry.
Cp is defined by the following equation:
P01 − Ps
Cp = (3-3)
P01 − Ps2
where P01 is the inlet total pressure, Ps2 is the exit static pressure and Ps is the measured
static pressure. The roughness height directly affects the boundary layer development and
therefore the profile loss. To study the effects of the roughness heights on the profile loss,
different roughness magnitudes have been considered at 50% of the plate length. Table 3-1
summarizes the parameters of the surface trips used in the parametric study on the flat plate.
The profile loss coefficient () is defined as:

2θ
= (3-4)
s · cos(α2 )
where θ is the trailing edge momentum thickness, s is the equivalent pitch of pressure dis-
tribution under investigation and α2 is the exit flow angle and takes a value of 63.2o . The
parameters associated with the test case are documented in Appendix A.

Prakhar Kapoor Master of Science Thesis

3-1 Flat Plate 27

(a) Inviscid Simulation (b) Turbulence Intensity= 0.5 %

Figure 3-4: Comparison of surface static pressure distribution carried out to validate the com-
puted domain geometry

Figure 3-5: Effect of trip height on losses at Tu=0.5 %

The effect of surface trip height on the profile losses at Re=174000 and Tu=0.5% is shown in
Fig. 3-5. It can be seen that the experimental data and the simulated results agree reasonably
well for the rough regime. However, for the smooth case, the CFD simulations under-predict
the loss coefficient. As a result of the separation bubble, the loss coefficients are higher in
the smooth case. The addition of the surface trips reduces the profile loss. Furthermore, the
losses continue to decrease as the heights of the surface elements increases. Fig. 3-6 presents
the variation of Cp along the blade surface. As a result of the adverse pressure gradient, the
boundary layer separates at a plate length of 0.35 mm for the smooth case and reattaches at
a length of 0.39 mm. The addition of the surface roughness causes the transition location to
shift upstream and also causes an early reattachment.

Master of Science Thesis Prakhar Kapoor

28 Validation Cases

Figure 3-6: Coefficient of pressure distribution along the plate length

3-2 Von Karman Institute Transonic Turbine Guide Vane

In this test case, computational aero-thermal validation of flow characteristics around a highly
loaded transonic turbine nozzle guide vane has been performed. The experimental data for
the validation corresponds to the measurements made by Arts [33] in the compression tube
facility. The experimental work involved the detailed analysis of the effect of free-stream
Mach and Reynolds numbers as well as turbulence intensity on the aerodynamic and heat
transfer over the turbine blade.
For the current computational analysis a finite volume method based fluid flow solver ANSYS
CFX has been used. The computational domain for the current work has been provided by
Dr. Rene Pecnik. The domain consists of 50728 elements and provides a resolution of y+ < 1
at the first cell row close to the blade surface [3]. The important geometric parameters of
the turbine cascade are summarized in the table 3-2. Fig. 3-7 shows the computational
mesh along with the imposed boundary condition used for the numerical simulations of the
VKI turbine blade. Pressure inlet boundary condition is imposed to define the fluid pressure
at the inlet to the flow domain. The total pressure and total temperature is specified in
accordance with the experimental data of Arts [33] to satisfy the flow calculations. At the
outlet boundary condition, static pressure value is specified in accordance to the experimental
case to represent the pressure outlet boundary condition. The suction and pressure surfaces
are specified as isothermal wall. Furthermore, the pitch-wise boundary ends are imposed as
periodic boundaries. Lastly, upper and lower side boundaries in the normal-wise direction
have been specified as free slip walls.
The numerical simulations have been performed for two different flow cases corresponding to
two different outlet- Reynolds number and Mach number. The corresponding flow conditions
for the two test cases, namely MUR-235 and MUR-241, are documented in the table 3-3.
Furthermore, the boundary conditions for the two test cases have been presented in Appendix
A . For the current case in perspective, transitional model formulation in ANSYS CFX has
been used.
The test case has been validated by comparing the heat transfer coefficient along the normal-
ized curvilinear coordinates of the turbine blade obtained by the numerical simulations with
the experimental data. The reference temperature for the simulation is prescribed as 420 K,

Prakhar Kapoor Master of Science Thesis

3-2 Von Karman Institute Transonic Turbine Guide Vane 29

Figure 3-7: Computational mesh for VKI turbine guide vane

Table 3-2: VKI transonic blade geometric parameters

S No Blade Geometric Parameter Value

1 Blade Chord 12.5 mm
2 Inlet Flow Angle 37.3o [-]
3 Stagger Angle 59.3o [-]
4 Pitch to Chord 0.850 mm

corresponding to the value used by Arts [33] in the experiments while the blade surface is
assumed to have a constant temperature of 298 K. The heat transfer coefficient(h) has been
modeled as

qw
h= (3-5)
T∞ − Tw

where qw is the wall heat flux, T∞ is the reference fluid temperature and Tw is the prescribed
wall temperature.
Figs. 3-8(a) and 3-8(b) show the distribution of the heat transfer coefficient along the nor-
malized curvilinear coordinates for the VKI-MUR 235 and VKI-MUR 241 cases respectively.
In the figures, the positive values of s/c correspond to the suction side while the negative

Table 3-3: Data pertaining to the VKI turbine guide vane test cases. Mis,out and Rec,out
are the outlet Reynolds number and Mach number based on the blades chord. ReM is the
Reynolds number based on the velocity and mesh spacing of the turbulence grid in the experimental
arrangement, used to obtain specific dissipation rate ω [3]

Case Tu[%] Mis,out Rec,out ReM

. MUR-235 6.0 0.927 1.15e6 61700
MUR-241 6.0 1.089 2.11e6 61700

Master of Science Thesis Prakhar Kapoor

30 Validation Cases

(a) MUR-235 case (b) MUR-241 case

Figure 3-8: Distribution of heat transfer coefficient over the VKI blade profile.

values correspond to the pressure side of the blade. Both the experimental data and numer-
ical simulation predict no transition at the pressure side of the blade. This is evident as a
smooth variation in the heat transfer distribution can be observed along the curvilinear co-
ordinates of the pressure side. While, for the suction side the experimental data reveals flow
transition. The increase in the heat transfer coefficient at s/c 0.8 and 0.6 for MUR-235 and
MUR-241 cases respectively shows the onset of laminar to turbulent transition at the suction
side of the investigated blade. The computational model as seen in Fig. 3-8(a) and 3-8(b)
predicts the transition over the suction side. However, there is a significant difference between
the experimental transition location and computed transition location. This is attributed to
the fact that in the γ − Reθt model the experimental correlations that affect the transition
prediction and modeling are based on the experiments conducted over an incompressible flat
plate. Furthermore, it can be observed that the numerical simulations under-predict the heat
transfer coefficient along the suction side and the pressure side of the turbine blade.
The value of skin friction coefficient along the blade surface has been compared with the
numerical results conducted by Raspopov [3]. Figs. 3-9(a) and 3-9(b) show the variation of
the skin friction coefficient along the pressure side and suction side of the turbine blade for
the two considered cases. The results obtained from the current simulations overpredict the
skin friction as compared to the reference data for the MUR-235 case while an appreciable
agreement is observed in case of the MUR-241 case. However, for both the cases the current
simulations predict a delayed transition close to the trailing edge of the blade.

Prakhar Kapoor Master of Science Thesis

3-2 Von Karman Institute Transonic Turbine Guide Vane 31

(a) MUR-235 case (b) MUR-241 case

Figure 3-9: Skin fiction distribution over the VKI blade profile.

Master of Science Thesis Prakhar Kapoor

32 Validation Cases

Prakhar Kapoor Master of Science Thesis

 Chapter 4

Compressor Performance Evaluation

The previous chapter provides a validation to the ability of the transition model formulation
available in CFX to accurately predict and model the transition onset. The current chapter
presents the numerical study performed on the centrifugal compressor. The compressor has
been adopted by the turbocharger developed by Mitsubishi Turbocharger and Engine Europe
B.V. Steady state simulations have been performed using ANSYS CFX to approximate the
unsteady flow field and the performance of the test compressor. The test compressor has been
simulated in accordance with a defined set of procedures and detailed evaluation of the flow
field, and performance has been made. The current analysis allowed to gain further insights
about the compressor flow structure, performance and loss mechanisms. The chapter provides
a detailed methodology employed for the compressor simulation and describes the outcome
of the analysis.
Firstly, in Section 4-1, the geometric specification of the test compressor is provided. The
compressor pre-processing and computational framework is described further in Section 4-
2. Next, detailed performance analysis of the test compressor is presented in Section 4-3.
Finally, the last Section 4-5, provides an overview of the phenomenon of flow instability
evolution within the centrifugal compressor.

4-1 Compressor Specification

The test compressor for the current work is provided by Mitsubishi Turbocharger and Engine
Europe B.V.(MTEE). The basic geometric parameters of the compressor are documented in
Table A-1
The modeled compressor consists of a total of 8 blades (4 splitter blades and 4 main blades).
The blades are leaned and have sufficient backsweep in the direction of impeller rotation. A
vaneless diffuser with a front pinch forms an integral part of the compressor domain. An
overhung volute with a 900 bend at the outlet finally receives the flow and completes the
compression stage.

Master of Science Thesis Prakhar Kapoor

34 Compressor Performance Evaluation

Table 4-1: Compressor specification

S No Compressor Domain Parameter Dimension

Tip radius, r2 20 mm.
1 Impeller Inlet shroud radius, r1s 12.5 mm
Backsweep angle, β2B 40o
2 Diffuser Outlet Radius, r5 12.5 mm
2 Volute Outer diameter, d7 38.4 mm

(a) Front view (b) Side view

Figure 4-1: The modeled impeller geometry for the computational fluid dynamic analysis

4-2 Preprocessing and CFD Setup

In order to obtain a CFD model for the test compressor, a sequential procedure is adopted.
This section presents the methodology adopted for development of a robust model for the nu-
merical simulations. The model development involves steps such as the geometric parametriza-
tion, grid processing and solver setup. A detailed description of all the steps employed for
the development of a robust model is presented in the subsequent literature.

4-2-1 Geometric Parametrization

The preliminary step in pre-processing is to obtain a precise geometric model of the compres-
sor. For the current compressor under investigation, the compressor geometric data available
in the casting and machine drawings have been transformed into a computer compatible 3D
model using the modeling tool ANSYS Bladegen. The tool allows rapid definition of radial
blade rows and renders an essential link between the blade design and advanced simulations
(Fig.4-1).
For the current work, the inlet duct, the impeller and the vaneless diffuser have been mod-
eled as a single domain. Only a single blade passage has been modeled for the impeller by

Prakhar Kapoor Master of Science Thesis

4-2 Preprocessing and CFD Setup 35

Figure 4-2: Computational domain for the test compressor

considering a rotational symmetry around the blade passages. This is done to reduce the
computational effort required for the simulations. Furthermore, appropriate interfaces have
been defined between the rotating and stationary parts of the domain. The single passage
fluid domain is represented in figure 4-2. The fluid enters the domain through the inlet duct
and leaves the domain through the vaneless diffuser. The impeller domain has been modeled
as a meridional curve with the help of coordinates as obtained from the drawings. The curve
represents surface of revolution. The impeller blades are modeled by defining the wrap angle
and thickness curves along the stream-wise direction. The angle and thickness curves have
been obtained from the part drawings of the impeller blade provided by MTEE. In the current
work, constant blade thickness has been used. The value at the midspan has been used as
the average thickness value for the impeller blades. Furthermore, to reduce the computa-
tional expense the volute has not been included in the computational domain. In order to
incorporate the effects of volute, a meanline code has been developed.

4-2-2 Computational Mesh and Numerical setup

Fig. 4-3(b) presents the structured grid of the impeller domain having a main blade and
a splitter blade. ANSYS TurboGrid has been used to obtain the mesh for the entire fluid
domain, i.e., inlet duct, impeller and diffuser. Turbogrid is a turbomachinery mesh creation
tool that allows the creation of high-quality hexahedral meshes. A structured mesh has been
used for the impeller and diffuser passages. An O type mesh has been employed near the
leading edge of the impeller blade whereas, an H type mesh has been adopted in the other
parts of the passages. In order to resolve the viscous sublayer, a very fine mesh has been
maintained near the walls by providing a sufficient number of nodes in the boundary layer
region. Thus, an overall y + ≈ 1 has been achieved at the walls. Furthermore, the tip clearance

Master of Science Thesis Prakhar Kapoor

36 Compressor Performance Evaluation

(a) Impeller blades and hub (b) Single passage mesh

Figure 4-3: The mesh for computational fluid dynamic analysis

of 1mm has been incorporated by trimming the blade shroud profile in turbo grid. An H type
grid has been used in the impeller blade tip clearance with no hexahedral elements.
In order to ensure that the solutions do not vary with grid size, a grid independence study has
been conducted using three computational grid sizes 0.2 million, 0.4 million and 0.8 million
for the entire compressor domain. To study the effect of grid on the solution, the pressure
ratio and compression efficiency at the peak efficiency operating point of 220000 rpm have
been compared for the three grid sizes. The variation in pressure ratio and efficiency has
been found to be 0.10% and 0.125 % for the medium grid (0.4 million) and the fine grid (0.8
million) respectively, in comparison to the coarse grid(0.2 million). Thus, the medium grid
size of 0.4 million elements was deemed sufficient for further modeling due to the marginal
variation in the monitored properties with increasing grid size.
To model the compressor performance, steady state analysis has been carried out in ANSYS
CFX. ANSYS CFX is a commercial CFD tool that solves 3D Reynolds averaged Navier-Stokes
equations using finite volume discretization in stationary or rotating frames. Furthermore,
of the available turbulence models in ANSYS CFX, k − ω SST model has been used. The
details about the model can be obtained from section 2-1-1 of the report.

4-2-3 Boundary Conditions

The compressor as shown in Fig. 4-2 has been simulated as a single passage fluid domain.
This is done to save the computational effort required to simulate the complete compressor.
Periodic boundary conditions are specified at the symmetric surfaces of the single passage
domain. The interfaces between the impeller, diffuser and inlet have been specified as mixing
plane or stage. In a mixing plane approach, each fluid domain is treated as a steady state
problem. The flow field data from adjacent zones are passed as boundary conditions that are
spatially averaged at the mixing plane interface [34]. The impeller shroud has been defined as

Prakhar Kapoor Master of Science Thesis

4-3 Compressor Performance Analysis 37

counter-rotating wall to allow simulation of the relative motion between the rotating impeller
and stationary shroud.
At the inlet boundary, total pressure of 1 atm and total temperature of 293.15K has been
specified in accordance with the ambient conditions used by MTEE. A medium turbulence
intensity of 5% is specified at the inlet. According to Javed et al. [35], variation of turbulence
intensity from 1- 10% does not cause a significant variation in the impeller performance.
The outlet boundary of the compressor is specified as mass flow outlet or static pressure
depending on operating region of the speed line. Near the choke limit, where a small change
in flow rate represents a considerable change in pressure, static pressure is specified at the
outlet. For the rest of the operating regions over the speedline, mass flow is specified at the
outlet boundary. Furthermore, in order to simulate the choke point the static pressure is
consistently reduced till the mass flow at the outlet no longer varies with the reduction in
static pressure.

4-3 Compressor Performance Analysis

This section of the chapter deals with predicting the compressor performance over a given op-
erating speed line and validation of simulation results with the experimental test performance
map. The section begins by introducing the fundamental equations that form the basis of
developing a performance model for the compressor volute. Subsequently, the steady state
analysis results for the test compressor has been discussed and elaborated in terms of the
compressor performance maps.

4-3-1 Volute Performance Model

In order to model the performance of the compressor, a one-dimensional modeling scheme

reported by Japikse [4] has been employed. The model is based on area ratio which is given
as :
A.R = A7 /A5 = (π/4)D72 /2πr5 b5 (4-1)

where A5 and A7 are the areas at the inlet and exit section of the volute respectively. The
flow is considered as incompressible for the model and a simplified relation for the inlet and
exit velocity at the scroll is given by the following equations:

V5 = (Q/A5 )(1 + λ2 )0.5 (4-2)

V7 = (Q/A7 ) (4-3)
λ = Vt5 /Vm5 (4-4)

where V is the absolute velocity, Q is the volumetric flow rate, Vm is the meridional velocity
and Vt is the tangential flow velocity.
The losses within the volute scroll can be estimated by very simple models. Firstly, it is
assumed that the meridional component of kinetic energy entering the scroll is completely
lost. Thus, the loss component can be evaluated as:

Master of Science Thesis Prakhar Kapoor

38 Compressor Performance Evaluation

∆po , meridional
Km =
0.5ρV52
2
0.5ρVm5
=
0.5ρV52

1
Km = F1 (4-5)
1 + λ2

The above equation represents a reasonable modeling assumption, but may be erroneous if
some of the meridional velocity can be used in the downstream element for flow stabilization
at the volute exit. In order to take this into account the correction term F1 is introduced [4].
The second loss component due to the tangential velocity can be modeled with two assump-
tions. If the core flow accelerates (Vt5 < Vt7 ), then it is valid to assume that no loss occurs.
On the other hand, if tangential flow decelerates in the volute (Vt5 > Vt7 ), then the flow
diffuses and it can be assumed that the pressure loss is equivalent to the sudden pressure loss
in a sudden expansion mixing process [4]. The above loss is modeled as:

∆po , expansion
Kt =
0.5ρV52
0.5ρVt52 (1 − A5 /A7 )2
=
0.5ρV52

(λ − 1/A.R)2
Kt = (4-6)
1 + λ2

A modification is applied to the equation(4-6) to account for the vortex diffusion. The mod-
ified equation is given by:

r5 2 (λ − 1/A.R)2
Kt = F2 ( ) (4-7)
r6 1 + λ2

where r5 and r6 are the diffuser inlet and return bend exit radius.
The pressure loss recovery coefficient is given by:

 2(λ−1/A.R)
2 for λA.R > 1 (Flow Diffuses)
A.R(1+λ )
Cp = (4-8)
 λ2 −1/A.R2 for λA.R < 1 (Flow Accelerates)
1+λ2

A detailed Matlab code is documented in Appendix B, based on the above equations, to model
the performance of the compressor volute. The subsequent subsection describes the results
obtained as a consequence of the steady state simulations and the volute performance code.
These simulations are necessary to validate the computational model with the experimental
results prior to performing further analysis.

Prakhar Kapoor Master of Science Thesis

4-3 Compressor Performance Analysis 39

(a) Variation of Pressure Ratio over a speed line (b) Variation of isentropic efficiency over a speed line

Figure 4-4: Comparison of the experimental and simulated performance curves for the test
compressor

4-3-2 Performance Curves

A steady state simulation of the test compressor has been performed at different operating
points at the 220,000 rpm operating speed line to evaluate the performance of the compressor.
The results have been compared with the experimental data available from the test maps.
Since the test compressors computational domain consists only of the impeller and vaneless
diffuser, a meanline code has been incorporated to predict the performance variation as a
result of volute. This is done in order to make realistic comparisons with the available test
data. The test data is obtained for the complete compressor with the fluid properties evaluated
at the compressor inlet and outlet.
The comparison between the simulated performance and test data has been made in terms of
compressor pressure ratio and isentropic efficiency. The overall compressor performance for
both the smooth case and the rough case, in terms of the speed lines representing the exper-
imental data and simulated results, is depicted Fig. 4-4. A reasonable agreement is observed
between the simulated performance and the test results. The numerical computations predict
the trend of pressure ratio reasonably for the simulated speed line. A kink is observed at
the low mass flow rates. This discrepancy in the pressure ratio prediction at low mass flow
rates can be attributed to the onset of rotating stall. The rotating stall is an unsteady phe-
nomenon and cannot be accurately accounted by steady state simulations. Furthermore, at
the low mass flow rate, a very small deviation in the pressure ratio is observed when compared
with experimental results. On the other hand, the simulations predict the choking condition
but the lowest pressure ratio is observed at a higher mass flow by the CFD computations.
As seen in Fig. 4-4, the simulations also capture the trend of the isentropic efficiency fairly
well. In this case also they predict the choking condition at a higher flow rate as compared
to the experiments.
It can thus be concluded from the current analysis that the CFD setup and simulation are
capable of predicting the characteristic curves of the compressor performance, despite some
deviations observed near the low flow rate region and choke point. This deviation can be

Master of Science Thesis Prakhar Kapoor

40 Compressor Performance Evaluation

attributed to the limitations of the turbulence model and the mixing frame approach applied
at the interfaces.

4-4 Impeller Flow Field Analysis

The Fig. 4-5 shows the contours of the meridional velocity, relative Mach number and static
entropy contours at various streamwise locations along the compressor impeller blades. All
the results have been documented for the best efficiency case which corresponds to a mass
flow rate of 100 g/s at the rotational speed of 20000 rpm.
The relative Mach number contours are presented in Fig. 4-5(a). It can be observed from the
figure that at the suction side of the main blade leading edge, a region of supersonic Mach
number exists. The high relative Mach number region extends both along the circumferential
direction and the spanwise direction. The existence of this region can be attributed to the
high relative mass flow rate, which in turn results in high meridional velocity. As a result,
the relative flow angle is lower than the blade angle. This results in a positive incidence
angle which causes the flow to accelerate [36]. This region of high relative Mach number is
followed by an area of low relative Mach number, which increases in the both streamwise and
spanwise direction. It can be observed that the relative Mach number decreases along the
blade passages. Near the impeller outlet, a low Mach number region can be observed near the
shroud. This region can be attributed to a reduction in the flow velocities as a consequence
of the tip leakage flows.
Fig. 4-5(b) presents the meridional velocity contours at different streamwise location along
the compressor impeller. The meridional velocity is the absolute velocity component that
represents mass flow rate through continuity equations [36]. The variation of the meridional
velocity across the impeller also follows a similar trend as that of relative Mach number. A
low-velocity area develops on the shroud near the suction surfaces of both the main blade
and splitter blade. The largest magnitude of the meridional velocity is observed near the hub
of the impeller domain. At the trailing edge, a substantial region near the shroud can be
observed to have a low meridional velocity.
As a result of the secondary flows and wake development, losses are expected in the impeller
domain. The figure 4-5(c) presents the contours of entropy generation in the blade passage.
It can be observed from the figure that the loss production along the blade passage is con-
centrated near the impeller shroud. The largest loss region in located near the trailing edge
shroud. This can be attributed to large tip leakage flows which significantly affects the main
core flow.
In order to further analyse the flow field near the impeller shroud, the relative Mach number
contour at a location close to the impeller shroud have been shown in Fig. 4-6. It is evident
from the figure that close to the main blade leading edge, where a high-velocity region develops
near the suction side, the relative Mach number is supersonic. Also, high Mach number can
be observed close to the leading edge of the splitter blade. However, a decrease in the relative
Mach number can be observed in the streamwise direction. In the case of both main blade
and splitter blades, a low-velocity wake region develops further downstream.
The subsequent section deals with analyzing the changes in the centrifugal compressors flow
structure as a result of a reduction in the mass flow parameter at different operating points

Prakhar Kapoor Master of Science Thesis

4-5 Flow Phenomenon Responsible for Surge 41

(a) Relative Mach number (b) Meridional velocity

(c) Static entropy

Figure 4-5: Relative Mach number, meridional velocity and static entropy contours at different
streamwise direction at the best efficiency point

along a speedline. The analysis has been presented by analyzing the changes in the flow
structure within the impeller and diffuser.

4-5 Flow Phenomenon Responsible for Surge

In order to evaluate the performance of the compressor over a speedline, different operating
points ranging between the choke and the surge margin of the compressor have been simu-
lated as described in section 4-3. The current section deals with gaining greater insight into
the change in associated flow behavior as the mass flow is reduced within the centrifugal
compressor. The simulations at low mass flow rates explain the details of the phenomenon
responsible for the onset of the unstable compressor operations. The unstable flow regions
such as separation zones, inlet recirculation regions and other unstable flow phenomena which
cause the onset of surge have been identified and their growth with lower mass flow rates is
observed.

Master of Science Thesis Prakhar Kapoor

42 Compressor Performance Evaluation

Figure 4-6: Relative Mach number contour at a section located at 80% of impeller spanwise
direction

4-5-1 Inlet Recirculation Region

The inlet recirculation zone in a centrifugal compressor is a stationary phenomenon. The

inlet recirculation region is formed at the radial machines inlet with a pressure growth at
outlet [37]. In the view of Breugelmans and Sen [38], the inlet recirculation region forms
the first sign of unstable operation. Fig. 4-7 depicts the growth of the inlet recirculation
region in the test compressor with the mass flow rate in terms of the velocity vectors along
the meridional plane. According to Chodkiewicz et al. [37], the inlet recirculation effects
the compressors operating conditions and may not be equivalent to the appearance of surge.
However, in case of high-pressure ratio compressors, it might cause an intensive surge. It can
be seen that the inlet recirculation region develops even at high mass flow rates (4-7(b)). As
the flow rate is further reduced the recirculation region grows and effects a substantial portion
of the flow field near the impeller shroud as depicted in Figs. 4-7(c) and 4-7(d).

4-5-2 Flow Separation within the Impeller

As mentioned in Chapter 1, the flow separation effects the stable operation of the compressor.
The flow field within a centrifugal compressor is a complex viscous three-dimensional flow
wherein the curvature and rotational effects result in the development of non uniform flow.
The effect of mass flow reduction through the centrifugal compressor on the flow field within
the compressor is presented in terms of the velocity contours along a midplane of the impeller
section in Fig. 4-8. The operation of a centrifugal compressor at lower flow rates causes the
flow to separate near the trailing edge and along the suction side of the impeller and splitter
blades. This flow separation causes development of low energy wake region or separation
bubble. The flow separation in this region results from the combined effect of the secondary
flows (Fig. 4-9) and adverse pressure gradients. The Fig. 4-9 shows that at low mass flow

Prakhar Kapoor Master of Science Thesis

4-5 Flow Phenomenon Responsible for Surge 43

(a) Choke Flow (b) Best Efficiency ṁ= 100 g/s

(c) Near Stall ṁ= 75 g/s (d) Stall ṁ= 48 g/s

Figure 4-7: Meridional velocity contours at the diffuser midplane for different flow rates

rates, large secondary flow exists near the impeller shroud which affects the core flow. The
development of the secondary flow provides an indication regarding the onset of unstable
compressor operation.

4-5-3 Flow Separation within the Vaneless Diffuser

The flow separation within the centrifugal compressor forms a critical issue that affects the
stable operation. Rotating stall in diffusers is caused in most cases by reverse flow based on
three dimensional separation of boundary layer [39]. The separation in the vaneless diffuser
essentially occurs at low mass flow rates. This is because the flow in a vaneless diffuser is
skewed and at low mass flow rates, a local reverse flow region occurs. This is a precursor
for Rotating stall [39]. The experimental analysis performed by Kinoshita and Senoo [7],

Master of Science Thesis Prakhar Kapoor

44 Compressor Performance Evaluation

(a) Choke Flow (b) Best Efficiency ṁ= 100 g/s

(c) Near Stall ṁ= 75 g/s (d) Stall ṁ= 48 g/s

Figure 4-8: Relative velocity contours at the midspan of the impeller domain

Abdelhamid [40], Jaatien- Vaari [41] reveals that at low flow rates the efficiency of compression
system decreases because of flow separation and rotating stall inception, while at higher flow
rates, the compressor stage has a wider operating range and higher polytropic efficiency.
Fig.4-10 describes the evolution of the separation in a vaneless diffuser as a consequence
of the reduction in the flow rate from the choke to stall in terms of the contours of the
meridional velocity. It is observed that as the mass flow reduces, the separation and reversals
at the diffuser increase. The flow at the diffuser shroud separates and reverses back into the
impeller at the low mass flow operating points. Near the operating point close to the stall,
it can be observed that flow reversal occurs across the entire diffuser shroud(Fig. 4-10(c) &
4-10(d)). Thus, it can be concluded that at low flow rates, the flow instabilities in the diffuser
may form an essential cause of the unstable compressor operation.

Prakhar Kapoor Master of Science Thesis

4-5 Flow Phenomenon Responsible for Surge 45

(a) Choke (b) Best Efficiency ṁ= 100 g/s

(c) Near Stall ṁ= 70 g/s (d) Stall ṁ= 48 g/s

Figure 4-9: Relative velocity streamlines depicting the secondary flow structure at different
operating points

Master of Science Thesis Prakhar Kapoor

46 Compressor Performance Evaluation

(a) Choke (b) Best Efficiency ṁ= 100 g/s

(c) Near Stall ṁ= 70 g/s (d) Stall ṁ= 48 g/s

Figure 4-10: Meridional velocity contours at the diffuser mid-plane for different flow rates

Prakhar Kapoor Master of Science Thesis

 Chapter 5

Roughness Application and Evaluation

The previous chapter highlights the CFD modeling applied to the test case centrifugal com-
pressor to predict its performance. In addition to that, it describes the evolution of the flow
instabilities within the compressor with reduction in the mass flow rate. The current chapter
presents a parametric study on the effect of the roughness height and location on the flow
instabilities in the centrifugal compressor. The surface roughness has been introduced on
the compressor impeller and diffuser. In order to study the effect of roughness on the flow
behavior, simulations have been carried out at a flow rate lower than the stall point flow rate
as predicted by the test map. Firstly, the study has been carried out by analyzing the effect of
the predefined wall roughness applied on the impeller and diffuser domain. Lastly, localized
roughness strips have been modeled on the impeller blade and diffuser, and their effects on
the flow field have been studied.

5-1 Wall Roughness Analysis

The current section of the chapter deals with analyzing the effect of increasing the roughness
at the impeller blades, inlet shroud, impeller and diffuser shroud on the flow field within
the compressor. The rationale behind choosing the above locations in the compressor can
obtained from chapter 4. It can be observed from Fig 4-7 to 4-10 that at low mass flow rates,
considerable flow reversals prevail near the diffuser shroud. Furthermore, the flow field also
separates near the blades trailing edge. Lastly, an increase in the surface roughness at inlet
shroud is considered to reduce the extent of inlet recirculation that affects the inlet duct of
the compressor.
Table 5-1 documents the magnitude of the surface roughness applied to the compressor walls.
In order to relate the geometric roughness height to the equivalent roughness, the following
correlation as defined in the work of Simon and Bulskamper [42], has been used:

ks = 2Ra (5-1)

Master of Science Thesis Prakhar Kapoor

48 Roughness Application and Evaluation

Table 5-1: Magnitudes of considered roughness elements

S No Geometric Roughness Height k(µm)

1 50
2 100
3 200

where Ra is the centerline average roughness value.

It has been assumed that the geometric roughness height (k) is defined by centerline average
roughness value.
The effect of the engineered roughness on the flow structure has been studied separately for
the diffuser and the impeller. The following sections provide the results pertaining to the
analysis. To begin with, the effect of additional roughness on the flow through the vaneless
diffuser is elaborated. Next, a brief discussion about the effect on the impeller flow field and
reversals.

5-1-1 Effect of Surface Roughness on Diffuser Flow Field

Fig. 5-1 shows the streamlines along the meridional plane of the diffuser. The influence of
different magnitudes of roughness on flow reversals in the centrifugal compressor has been
highlighted in the figure. It also shows a comparison of the smooth case with the roughened
case. For all the cases studied, it can be seen that a region of reversed flow exists at the diffuser
inlet in proximity of the shroud. The existence of these flow reversals can be attributed the
fact that the flow at the diffuser inlet is highly unsteady and viscous [36]. At the impeller
outlet, a complex flow field exists and strong fluctuations in the velocity and flow angle are
observed [36]. It is believed that a jet-wake pattern exists at the impeller outlet. The jet flow
structure exists at the blade pressure side and is considered almost loss-free whereas the wake
forms the region of low flow velocity [43]. The wake carries losses, turbulence and accumulates
near the shroud or suction side edge [36]. This explains the existence of the region of reversed
flow at the inlet of the diffuser.
It can be seen that in case of the smooth-walled compressor, a reversed flow region exists
along the entire shroud wall. In addition, a flow reversal can be observed at the hub near the
diffuser exit. The addition of surface roughness has a considerable effect on the diffusers flow
field. The increase in roughness causes a reduction in reversals at the diffuser inlet. The most
prominent effect in this case is observed with the roughness magnitudes of 200 µm and 100
µm. The increase of roughness also causes an early reattachment of the flow and therefore
causes a reduction in the flow reversals along the diffuser shroud. Figs. 5-1(b), 5-1(c) and 5-
1(d) show the effect of increased roughness on the flow reversal at the diffuser. The increased
roughness height also causes an elimination of the reversed flow region at the diffuser hub.
Figs. 5-3 and 5-4 illustrate the effect of surface roughness on the spanwise distribution of radial
and tangential velocity respectively at the diffuser. The spanwise direction is based on the
diffuser width within the end walls (i.e., from the hub to shroud) and has been normalized such
that it ranges between 0 and 1. The distributions have been obtained at different streamwise

Prakhar Kapoor Master of Science Thesis

5-1 Wall Roughness Analysis 49

(a) Smooth Case (b) Roughness Height 50 µm

(c) Roughness Height 100 µm (d) Roughness Height 200 µm

Figure 5-1: Meridional velocity streamlines along the diffuser meridional plane for different
surface roughness

locations along the diffuser meridional plane. These locations are presented in Fig. 5-2. As
depicted in the Fig. 5-3(a), reverse flow zone occurs at the diffuser inlet for all the simulated
cases. However, an increase in the roughness results in a reduction in the flow reversals. Thus,
an increase in the wall roughness at the impeller shroud and blades results in a stabilization
of the outlet flow and results in reduced flow reversals at the diffuser inlet.
At streamwise location 2, it can be observed from the figure 5-3(b), that an increase in the
roughness at the diffuser shroud results in a decrease in reverse flow zone. As a result of
the increase in the magnitude of the surface roughness from 0-200 µm, it can be observed
that the flow reversals are completely eliminated. Similar results are observed at further
downstream locations. Fig. 5-3(c) shows complete elimination of the reverse flow zone as
a result of an increase in surface roughness. As compared to the baseline smooth case, the
cases with a roughness magnitude of 200 and 100 µm exhibit a complete elimination of the

Master of Science Thesis Prakhar Kapoor

50 Roughness Application and Evaluation

Figure 5-2: Streamwise locations used to study the spanwise distribution of radial and tangential
velocities in the diffuser

flow instability at the shroud wall. Although the case with a roughness magnitude of 50 µm
shows a reduction in the reverse flow zone as compared to the smooth case, a small region of
separated flow still exists in this case. Furthermore, it can be observed from the figure that
the peak of the radial velocity shifts towards the hub for the roughened cases. This shift in
the velocity peak can be attributed to the increase in the boundary layer thickness at the
shroud wall because of the increase in the roughness height.
It is seen in the Fig. 5-3(d) that a separated flow region exists close to the diffuser exit for
the smooth case. However, addition of the roughness to the diffuser shroud eliminates this
flow instability.
Fig. 5-4 elaborates upon the spanwise distribution of tangential velocity at different stream-
wise locations along the diffuser’s meridional plane. It can be seen that the addition of
roughness to the diffuser shroud results in a decrease in the tangential velocity. This trans-
lates to an overall performance reduction of the turbo-machine. However, a reduction in the
tangential velocity also causes a decrease in the friction losses which compensates for the
reduction in performance [44].

5-1-2 Effect of Surface Roughness on Impeller Flow Field

This section of the chapter presents the effect of increased surface roughness height on the flow
field of the centrifugal impeller. The nature of the impeller flow, especially the exit flow has a
significant effect on the performance of the compressor stage. A detailed investigation is thus
performed to investigate the effect of wall roughness on the flow structure in the impeller. The
analysis of the improvement in the flow structure is difficult using steady state simulations
because the stall and surge are unsteady phenomena. Nevertheless, a brief insight has been
provided by considering the flow behaviors at different locations in the impeller domain.
Fig. 5-5 shows the contours of total pressure at a section located at 70 % of the impeller
span. The distribution of total pressure gives an indication of the flow disturbances and
losses. Thus, the regions with a lower value of the total pressure correspond to the locations

Prakhar Kapoor Master of Science Thesis

5-1 Wall Roughness Analysis 51

(a) Streamwise location 1 (b) Streamwise location 2

(c) Streamwise location 3 (d) Streamwise location 4

Figure 5-3: Variation of radial velocity from hub to shroud at different streamwise locations (Fig
5-2) along the diffuser

where the generated losses are significantly greater. In the smooth case Fig. 5-5(a), it can be
observed that close to the main blade’s suction side, an area of low total pressure is present
(depicted by the blue region). This region can be attributed to the wake that develops as a
result of the three-dimensional separation due to the secondary flows. It can be inferred from
Figs.5-5(b), 5-5(c) and 5-5(d) that an increase in the magnitude of surface roughness at the
impeller blade, impeller and inlet shroud causes a reduction of the low total pressure region.
The effect of the increased roughness magnitude can also be interpreted in terms of the
variation of the mean meridional velocity from the inlet to the outlet of the impeller domain.
This variation is depicted in Fig. 5-6. It can be seen that addition of roughness causes
an increase in the magnitude of the meridional velocity. Since the meridional velocity is
influenced by the blockage of the flow passage, an increase in the meridional velocity can be
attributed to the additional blockage introduced by the thickening of the boundary layer.
Fig. 5-7 presents the contours of the meridional velocity at a streamwise location close to the
impeller blade trailing edge. It can be seen from the Fig. 5-7(a) that a low momentum wake
region exists near the main blade hub at a location close to the trailing edge. Furthermore,

Master of Science Thesis Prakhar Kapoor

52 Roughness Application and Evaluation

(a) Streamwise location 1 (b) Streamwise location 2

(c) Streamwise location 3 (d) Streamwise location 4

Figure 5-4: Variation of tangential velocity from hub to shroud at different streamwise locations
(Fig 5-2) along the diffuser

it can be clearly seen that an increase in the roughness height results in a reduction in the
separation zone (5-7(b), 5-7(b), 5-7(d)). The increase of the roughness height to 50 µm clearly
causes a reduction in the wake region in proximity of the main blade.

Fig. 5-8 depicts the distribution of mass flux (mass flow per unit area) at the impeller trailing
edge. The plot obtained from the analysis represent the blade to blade variation of the mass
flux at three spanwise locations: hub, mid-span and shroud. At the shroud, the Fig. 5-8(a)
reveals an improvement in the mass flux distribution near the suction side of the main blade.
However, a significant improvement is not seen at the other blade to blade locations (Figs.
5-8(b) and 5-8(c)). This can be attributed to an increase in the boundary layer thickness close
to the blades and shroud as a result of the surface roughness. This increase in the boundary
layer thickness causes enhanced blockage effect. It can be further observed that the increase
in the roughness height results in an increase in the mass flux distributions at the hub and
mid-section of the impeller blade’s trailing edge. Although the improvement at the midspan
is not that significant, it gives an indication of the ability of the roughness elements to cause
a slight improvement in the flow structure. The relative increase in the mass flux distribution

Prakhar Kapoor Master of Science Thesis

5-2 Localized Roughness Strips and their Effect on Flow Field 53

(a) Smooth Case (b) Roughness Magnitude 50 µm

(c) Roughness Magnitude 100 µm (d) Roughness Magnitude 200 µm

Figure 5-5: Normalized total pressure contours on a section located at 70 % of the impeller
spanwise direction

as a result of the different roughness heights is observed to be the same. The increase in the
mass flux represents a reduction in the flow instabilities as a consequence of the secondary and
tip leakage flows. However, the gains achieved in terms of the reduction in low momentum
wakes due to the roughness elements is also hampered by an increase in the boundary layer
thickness.

5-2 Localized Roughness Strips and their Effect on Flow Field

The previous section of the chapter describes the effect of the wall roughness applied to
impeller blades, impeller shroud, diffuser and inlet duct’s shroud on the flow instabilities and
flow structure within the compressors flow domain. It can be concluded from the analysis that
the increase in the roughness height results in an improvement of the flow structure within
the compressor.

Master of Science Thesis Prakhar Kapoor

54 Roughness Application and Evaluation

Figure 5-6: Variation of meridional velocity from the inlet to the outlet of the impeller domain
mid span.

(a) Smooth Case (b) Roughness Magnitude 50 µm

(c) Roughness Magnitude 100 µm (d) Roughness Magnitude 200 µm

Figure 5-7: Meridional velocity contours ot a location close to the trailing edge

Prakhar Kapoor Master of Science Thesis

5-2 Localized Roughness Strips and their Effect on Flow Field 55

(a) Spanwise location: Shroud (b) Spanwise location: Mid-Span

(c) Spanwise location: Hub

Figure 5-8: Circumferential mass flux distribution at the impeller trailing edge for the different
roughness cases

The current section of this chapter focuses on analyzing the effects of localized roughness
strips applied to the diffuser shroud and impeller blade. This section is split into following
parts: firstly, it describes the methodology adopted for modeling and analyzing the localized
roughness strips. Next, it provides the details pertaining to the parametric roughness study.
Lastly, the results of the study are described.

5-2-1 Modeling of Localized Roughness Strips

In order to study the effects of localized surface roughness on the flow instabilities in the
compressor stage, the suction surface of the impeller blade has been divided into three sub
regions. In addition to this, the shroud of the compressor diffuser has been split into two
regions. The three patches on the impeller blade have been referred to as "I","II" and "III".
Those on the diffuser have been named as "1" and "2". The Fig. 5-9(a) and 5-9(b) presents
the modified domains for the diffuser shroud and the impeller blade.

Master of Science Thesis Prakhar Kapoor

56 Roughness Application and Evaluation

(a) Modified rotor domain

(b) Modified diffuser shroud

Figure 5-9: Modified impeller and diffuser domain for the parametric roughness study

5-2-2 Roughness Parametrization

This section presents the design of experiments applied for the parametric roughness study.
Table 5-2 provides the information pertaining to the location and magnitude of the roughness
strip. Numerical simulations have been carried out for all the cases documented in the table 5-
2. Furthermore, for all the simulations, the impeller shroud and inlet duct have been modeled
as a smooth wall. Thus, a roughness height of 50 µm at the diffuser inlet strip is sufficient to
cause a improvement in flow structure.

5-2-3 Effects of Roughness Location and Magnitude

The current section of the report deals with the parametric study about effect of the roughness
location and magnitude on the flow structure within the centrifugal compressor. The study
has been conducted by carrying out numerical simulations at a mass flow rate lower than the
last stable operating point defined by the experimental compressor map. Thus, the entire
analysis has been based on the flow conditions at a mass flow rate of 38g/s and 220000 rpm
rotational speed. In order to conduct a comparative study, all the results have been compared
with the smooth baseline case which forms the benchmark for the current analysis. The entire

Prakhar Kapoor Master of Science Thesis

5-2 Localized Roughness Strips and their Effect on Flow Field 57

Table 5-2: Roughness magnitude and location

S No Roughness Location Roughness Height Ra ,(µm)

50
1 I,1 100
200
50
2 II,2 100
200
50
3 3 100
200

study has been conducted by analyzing the effect of roughness location on the flow fields of
impeller and diffuser. The subsequent literature provides the results pertaining to the study.

5-2-4 Effect of Roughness Strips on Diffuser Flow

The current subsection presents the results concerning the effect of localized roughness strips
on the diffuser flow field. The results have been presented for Case 1,2 and 3 documented
in table 5-2 and have been compared with the results for the smooth case. The subsequent
literature presents the results for the three cases.

Roughness Case 1

The Fig.5-10 shows the radial velocity distribution at the diffuser mid and exit corresponding
to the roughness height documented in Case 1 of the table 5-2. The figure depicts that
the addition of localized roughness close to the diffuser inlet effects the flow profile. In
comparison with the smooth case, the roughness strip at the inlet can be seen to reduce the
magnitude of flow reversal at the mid-span of the diffuser. In addition to this, it causes a
complete elimination of the reverse flow region close to the diffuser outlet. Thus, this gives
an indication about the improvement in the flow structure brought about by the addition of
localized roughness. However, the strip close to the inlet does not cause a significant effect
on the flow reversal at the diffuser inlet. Furthermore, it can also be seen that different
magnitudes of the roughness height have a similar effect on the flow profile.

Roughness Case 2

Fig. 5-11 shows the effect of localized roughness on the diffuser spanwise distribution of the
radial velocity. The effect of the localized roughness has been considered in terms of the strip
2 defined in Fig. 5-9(b). The figure reveals that the addition of localized roughness strip
at a streamwise location away from the diffuser inlet has limited influence on the diffuser
flow instabilities. The increase in the surface roughness at the strip 2 does not cause a

Master of Science Thesis Prakhar Kapoor

58 Roughness Application and Evaluation

(a) Radial velocity distribution at the diffuser inlet (b) Radial velocity profile at the mid span

(c) Radial velocity profile at the diffuser exit

Figure 5-10: Modification of the radial velocity distribution as a result of the localized roughness
applied to strip I at the diffuser

significant improvement in the flow structure at the shroud of the diffuser (Fig. 5-11(b)).
However, roughness element causes a reduction of the reverse flow region at the hub close to
the diffuser exit as depicted in Fig. 5-11(c). However, it does not cause a complete elimination
of the flow reversals at the diffuser hub.
Thus, it can be concluded from the current analysis that addition of the localized roughness
elements close to the diffuser inlet results in a reduction in the flow instability within a
centrifugal compressor. In addition to this, it can be inferred from the current analysis that
a roughness of 50 µm is sufficient to induce an improvement in the flow characteristics within
the diffuser.

5-2-5 Effect of Roughness Strips on Impeller Flow Field

This section deals with analyzing the effect of roughness location and height on the flow in the
centrifugal compressor’s impeller. The study has been conducted by considering patches of
roughness at the main blade’s suction side as shown in Fig. 5-9(a). The roughness locations

Prakhar Kapoor Master of Science Thesis

5-2 Localized Roughness Strips and their Effect on Flow Field 59

(a) Radial velocity distribution at the diffuser inlet (b) Radial velocity profile at the mid span

(c) Radial velocity profile at the diffuser exit

Figure 5-11: Modification of the radial velocity distribution as a result of the localized roughness
applied to strip II at the diffuser

have been applied only to the suction side because a considerable wake affected flow can be
observed near the suction side of the main blade. Furthermore, in order to reduce the viscous
drag, sectioned roughness patches are considered instead of the overall wall roughness. The
subsequent literature describes in detail the effect of localized impeller roughness strips on
the flow behavior within the impeller domain. The study has been carried out by analyzing
the effect of different roughness magnitudes applied to the independent locations 1, 2 and 3
as shown in Fig. 5-9(a).

Impeller Strip 1

The current section of the report presents the results of adding the roughness strip close to
the leading edge of the impeller blade. The effect of adding roughness can be interpreted
in terms of the mass flux distribution close to the impeller shroud. The Fig. 5-12 shows
the effect of adding localized roughness strip close to the blades leading edge on the blade
to blade mass flux distribution at the trailing edge. It can be seen that in comparison to

Master of Science Thesis Prakhar Kapoor

60 Roughness Application and Evaluation

(a) Mass flux distribution at the shroud (b) Mass flux distribution at hub

Figure 5-12: Modification of the mass flux distribution as a result of the localized roughness
applied to strip 1 at the impeller

the overall roughness case, roughness at a strip 1 causes a significant improvement in mass
flux distribution close to the trailing edge. The effective gain over the overall wall roughness
case can be attributed to the reduced blockage accounted for by the boundary layer thickness
and reduction in the separated flow region. Thus, the localized strip presents itself as an
improvement over the wall roughness case. Fig. 5-13 presents the contours of the meridional
velocity at a streamwise location close to the blades trailing edge. The contour plots also
reveal a reduction in the wake region at the blade trailing edge. This explains the gain in the
mass flux observed in the Fig. 5-12.

Impeller Strip 2

The current section deals with analyzing the effect of adding localized roughness at the rotor
strip 2 defined in Fig. 5-9(a). The study has been carried out by considering roughness heights
and analyzing their effect on the impeller flow. Fig. 5-12 presents the effect of roughness strips
in terms of blade to blade variation of the mass flux at the blade’s trailing edge. It is evident
from the curves depicted in Fig. 5-12 that it causes an increase in the mass flux distribution
at the trailing edge of the impeller blade. This increase in the mass flux represents a reduction
in the low momentum flow region at the trailing edge. However, in comparison to the mass
flux gain achieved by adding roughness elements to the strip 1 (Fig.5-12), the gain achieved
in this case is lower. It is also observed that an increase in the roughness height from 50 to
200 µm does not cause a significant improvement in the mass flux distribution at the trailing
edge.

Further, the distribution of the meridional velocity at a streamwise location close to the blades
trailing edge is depicted in the contour plots in Fig. 5-15. The addition of the roughness at
the strip 2 causes a reduction in the low momentum wake region. However, the gain observed
in this case is lower than the case in which roughness is introduced at strip 1.

Prakhar Kapoor Master of Science Thesis

5-2 Localized Roughness Strips and their Effect on Flow Field 61

(a) Smooth Case (b) Roughness Magnitude 50 µm

(c) Roughness Magnitude 100 µm (d) Roughness Magnitude 200 µm

Figure 5-13: Meridional velocity contours depicting the influence of roughness strip 1 on the
meridional velocity distribution at the impeller trailing edge

Impeller Strip 3

Fig. 5-16 presents the effect of adding localized roughness at the rotor strip 3 (Fig. 5-9(a)) on
the impeller flow field. It can be inferred from the Fig. 5-16(a) that the roughened imposed
at strip 3 of the impeller results in a reduction in the mass flux distribution at the impeller
shroud close to the trailing edge in comparison with the roughness imposed at strip 1 and
strip 2 in the impeller domain. Thus, the location 3 does not offer significant improvement
in the flow stabilization. In addition to this, it can be seen that the increase in the roughness
height causes a reduction in the mass flux distribution.
In order to further investigate the effect of roughness strip 3 on the impeller flow instability,
meridional velocity contours have been plotted at a location close to the rotor’s trailing edge.
The plots reveal that the addition of roughness (Fig. 5-17(b) to 5-17(d)) results in a reduction
in the low momentum region in comparison with the smooth case 5-17(a). In addition to this,
it can also be inferred from the contour plots that, an increase in the roughness magnitude
at the strip 3 causes an increase in the low momentum region relative to each other.

Master of Science Thesis Prakhar Kapoor

62 Roughness Application and Evaluation

(a) Mass flux distribution at the shroud (b) Mass flux distribution at hub

Figure 5-14: Modification of the mass flux distribution as a result of the localized roughness
applied to strip 2 at the impeller

(a) Smooth Case (b) Roughness Magnitude 50 µm

(c) Roughness Magnitude 100 µm (d) Roughness Magnitude 200 µm

Figure 5-15: Meridional velocity contours location depicting the influence of roughness strip 2
on the meridional velocity distribution at the impeller trailing edge

Prakhar Kapoor Master of Science Thesis

5-2 Localized Roughness Strips and their Effect on Flow Field 63

(a) Mass flux distribution at the shroud (b) Mass flux distribution at hub

Figure 5-16: Modification of the mass flux distribution as a result of the localized roughness
applied to strip 3 at the impeller

(a) Smooth Case (b) Roughness Magnitude 50 µm

(c) Roughness Magnitude 100 µm (d) Roughness Magnitude 200 µm

Figure 5-17: Meridional velocity contours depicting the influence of roughness strip 3 on the
meridional velocity distribution at the impeller trailing edge

Master of Science Thesis Prakhar Kapoor

64 Roughness Application and Evaluation

Prakhar Kapoor Master of Science Thesis

 Chapter 6

Summary and Conclusions

In this work, general purpose finite volume based solver, ANSYS CFX, has been adopted
to perform flow and performance analysis of a centrifugal compressor that forms an integral
part of the automotive turbocharger. For this purpose, the transition model formulation
available in ANSYS CFX has been first used to evaluate and study the performance and flow
field of a centrifugal compressor along a constant speed line. Secondly, the work focuses on
evaluating the effect of surface roughness on the flow instabilities of a centrifugal compressor.
This investigation results from the desire to improve the stability range of the centrifugal
compressor.
Prior to evaluating the compressor performance and its analysis, the finite volume based
RANS solver has been validated on 2D baseline test cases. These include flow over an adiabatic
flat plate, roughened flat plate and a flat plate with imposed pressure gradient and roughness
trips. This analysis has been performed to validate the ability of transition model to predict
and model laminar to turbulent transition and also to validate its capability to predict the
effect of surface roughness. The results obtained by the CFD simulations agree reasonably
well with the test case reference data. Certain deviations are observed in the test cases
with imposed surface roughness. These deviations can be attributed to the approximate
correlations used for relating the equivalent surface roughness to the geometric roughness
height of the roughness elements.
Next, the transition model formulation available in CFX has been validated for a turboma-
chinery test case. The model has been adopted for predicting the heat transfer over a highly
loaded VKI transonic turbine guide vane. The turbine guide vane has been simulated for two
flow configurations (MUR-235 and MUR-241). The results reveal that the numerical simula-
tions underestimate the heat transfer coefficient over the blade surface. However, the model
predicts the laminar to turbulent transition location at the suction side of the blade and has a
good agreement with the reference test data. Thus, the model formulation available in CFX
predicts the transition location fairly accurately.
After the model validation, comprehensive performance analysis of the test compressor has
been carried out using the CFD tool. Firstly, the CFD setup of the compressor has been

Master of Science Thesis Prakhar Kapoor

66 Summary and Conclusions

validated by comparing the performance data obtained from numerical simulations with the
experimental data. Overall, a close agreement is seen between the simulated performance and
experimental data. However, some deviations in performance estimation are seen close to the
choke point. Next, a detailed flow field analysis of the test compressor has been carried out at
the peak efficiency point of the simulated speedline. Furthermore, a flow investigation has also
been carried out with a view of understanding the evolution of unsteady flow in compressor
with the reduction in the mass flow rate. The flow analysis at the peak efficiency point revealed
the existence of a low relative Mach number region on the main blade suction side. This region
grows from the shroud to the hub and is maximum near the blade trailing edge. In addition
to this, the splitter blade’s suction side exhibits a lower Mach number and also relatively
smaller low-velocity region. This results in a higher outlet wake due at the main blade in
comparison with that due to the splitter blade. The maximum loss production, observed in
terms of the static entropy rise, is found close to the shroud and extends in the spanwise
direction as a result of the low momentum region close to the main blade. Furthermore, the
flow analysis reveals that at low flow rate operations, a significant portion of the impeller
inlet flow is effected by the recirculating flow observed near the impeller shroud. Also, the
mass flow reduction corresponding to different operating points (choke to surge) results in an
increase in the low momentum wake region observed near the impeller blades. In addition to
this, close to the stall limit, a recirculating region is observed near the diffuser shroud wall.
All these flow phenomenon indicate unstable compressor operation.

Lastly, the report concludes by a parametric study about the effect of surface roughness on
the flow instabilities in the centrifugal compressor. The roughness study has been split into
two parts: the first part deals with analyzing the effect of adding wall surface roughness at
the impellers’ blade, impeller shroud, inlet shroud and diffuser shroud on the flow instabilities
within the compressor’s diffuser and impeller. In the second part, a study has been carried
out by considering localized roughness strips on the main blade’s suction side and the diffuser
shroud. From the roughness analysis, it is observed that the surface roughness has a significant
effect on the flow reversals at the diffuser shroud. The addition of the roughness elements to
the diffuser shroud causes a significant reduction in flow reversals near the diffuser shroud.
The application of roughness also results in complete elimination of the flow reversal near the
diffuser hub. In the case of the impeller, the surface roughness causes a modification of the
flow structure. However, the visualization of the flow field at the impeller poses challenging
task for the steady state simulations as a result of the unsteady nature of the flow. Increase
in the roughness magnitude causes a reduction in the low momentum wake regions along the
impeller blade. In addition to this, it also causes an improvement in the mass flux distribution
close to the trailing edge. However, increased blockage due to an increase in boundary layer
thickness inhibits the gain achieved in terms of the reduction in the low momentum wake
regions. The analysis also shows that the increase of the roughness magnitude from 50-200 µm
does not cause a significant difference in terms of relative improvement of the flow structure.
Finally, the report discusses overcoming the shortcomings of the overall roughness analysis by
investigating the effect of adding localized roughness strips on the main blade’s suction side
and diffuser’s shroud. The analysis reveals that the addition of roughness strip close to the
diffuser shroud results in significant improvement in the flow structure. Furthermore, addition
of the roughness strip close to the impeller blades leading edge results in the reduction of low
momentum region. In this case, the increase in the roughness magnitude has a similar effect
on the flow structure.

Prakhar Kapoor Master of Science Thesis

 67

Recommendations and Future Work

The results obtained from the current analysis seem effective in terms of reducing the flow
instabilities in a centrifugal compressor and hence demand further investigation. However,
since the present work analyzes the improvement in flow instabilities using steady state sim-
ulations, the precise effect of the addition of roughness cannot be determined. Thus, in order
to further investigate the improvement in the stability range, time-dependent simulations
are proposed and recommended for the future evaluation. Furthermore, in order to validate
the results obtained from the simulations, comprehensive experimental evaluations have been
planned at MTEE.

Master of Science Thesis Prakhar Kapoor

68 Summary and Conclusions

Prakhar Kapoor Master of Science Thesis

 Appendix A

Reference Data for the Test Cases

The relevant geometric and boundary data pertaining to the test cases documented in chapter
2 is presented in this appendix. Table 1 presents the boundary condition data relevant to the
smooth flat plate, while Table A2 presents boundary layer data pertaining to the roughened
flat plate.

Table A-1: Boundary data corresponding to the smooth flat plate

Boundary Parameter Magnitude

ρ∞ , [kg/m3 ] 1.2
Reference T∞ , [K] 101634
Values P∞ , [P a] 293.1
µ∞ , [P a.s] 6.743e-05
Heat Transfer Adiabatic
Wall Uwall , [m/s] 0
kwall , [m2 /s2 ] 0
Pin , [P a] 101634
Inlet ωin , [1/s] 30
kin , [m2 /s2 ] 40000
Tin , [K] 293.1
Outlet Pout , [P a] 0.953e5

Master of Science Thesis Prakhar Kapoor

70 Reference Data for the Test Cases

Table A-2: Boundary data corresponding to the roughened flat plate

Boundary Parameter Magnitude

ρ∞ , [kg/m3 ] 1.2
Reference T∞ , [K] 101325
Values P∞ , [P a] 293.1
µ∞ , [P a.s] 17.952e-6
Heat Transfer Adiabatic
Wall Uwall , [m/s] 0
kwall , [m2 /s2 ] 0
Pin , [P a] 101325
Inlet Uin , [m/s] 100
T uin , [%] 1.1
Tin , [K] 293.1
Outlet Pout , [P a] 0.954e5

Table A-3: Boundary data corresponding to the roughened flat plate with imposed pressure
gradient

Boundary Parameter Magnitude

ρ∞ , [kg/m3 ] 1.2
Reference T∞ , [K] 101325
Values P∞ , [P a] 293.1
µ∞ , [P a.s] 2.65e-05
Heat Transfer Adiabatic
Wall Uwall , [m/s] 0
kwall , [m2 /s2 ] 0
Pin , [P a] 101325
Inlet T uin , [%] 0.5
Tin , [K] 293.1
Outlet Pout , [P a] 101287

Prakhar Kapoor Master of Science Thesis

 71

Table A-4: Boundary data corresponding to the turbomachinery test case-VKI transonic turbine
guide vane

Boundary Parameter VKI-MUR235 VKI-MUR241

ρ∞ , [kg/m3 ] 1.54 2.72
Reference T∞ , [K] 416.3 416.3
Values P∞ , [P a] 1.828e5 3.257e5
µ∞ , [P a.s] 1.716e-05
Twall 300 300
Wall Uwall , [m/s] 0 0
kwall , [m2 /s2 ] 0 0
Pin , [P a] 1.828e5 3.257e5
Inlet ωin , [1/s] 5e4 1.1e5
kin , [m2 /s2 ] 20.55 20.55
Tin , [K] 416.4 416.4
Outlet Pout , [P a] 1.049e5 1.547e5

Master of Science Thesis Prakhar Kapoor

72 Reference Data for the Test Cases

Prakhar Kapoor Master of Science Thesis

 Appendix B

Matlab Code for the Volute

Performance Estimation

clear all
clc
TT5 =377.155;
P5 =141297;

PT5 =189683 ;
VT5 =96.081;

VM5 = 215.089 ;
% Diffuser

R5 = 0.035; % Outlet Radius (Diffuser) [m]

B5 = 0.00237; %Diffuser Width at Outlet [m]

% Scroll
R6 = 0.03688; %Scroll Centre Radius (m)
D7 = 0.0384; % Exit Diameter (Scroll) (m)

% Input Thermodynamic Parameters

P01=101325; %Inlet Total Pressure (Pa)
T01=293.15; %Inlet Total Temperature (K)

% Constants
R=287; %Gas constant for air (j/kg/k)

Master of Science Thesis Prakhar Kapoor

74 Matlab Code for the Volute Performance Estimation

CP = 1005; %Specific Heat of air at constant pressure (j/kg/k)

GMA = 1.4; %Isentropic Coefficient

%********************Compressor Scroll Exit Parameters*********************%

F1=.6;
F2=.6;

AR_SCRL=((D7)2 )/(8 ∗ R5 ∗ B5); %Area Ratio (m2)

LMDA=VT5/VM5; %Inlet Swirl Parameter

KM = F 1 ∗ (1/(1 + (LM DA2 ))); %Meridional Loss Component

KT = F 2 ∗ ((R5/R6) ) ∗ ((LM DA − (1/AR_SCRL))2 )/(1 + (LM DA)2 ); %Tangential Loss
2

K_SCRL = KM + KT ; %Total Scroll Pressure Loss

if LM DA ∗ AR_SCRL > 1 %Flow Diffuses

CP _SCRL = (2 ∗ (LM DA − (1/AR_SCRL)))/(AR_SCRL ∗ (1 + (LM DA)2 )); % Pressure

Recovery
else
CP _SCRL = (LM DA2 − (1/ARSCRL
2 ))/(1 + LM DA2 );
end
T T 7 = T T 5; %Scroll Exit Total Temperature (K)
P T 7 = P T 5 − K_SCRL ∗ (P T 5 − P 5); %Scroll Exit Total Pressure (Pa)
P 7 = P 5 + CP _SCRL ∗ (P T 5 − P 5); %Scroll Exit Static Pressure (Pa)
T 7 = T T 7/((P T 7/P 7)( (GM A − 1)/GM A)); %Static Temperature at diffuser Exit (k)
M 7 = sqrt((2/(GM A − 1)) ∗ (((P T 7/P 7)( (GM A − 1)/GM A)) − 1));%Diffuser Exit Mach No
V 7 = M 7 ∗ sqrt(GM A ∗ R ∗ T 7); %Velocity at Scroll Exit (m/s)

%********************Stage Performance*******************%
P RT T = P T 7/P 01 %Total to Total Stage Pressure Ratio
ET AT T = ((((P RT T )( (GM A − 1)/GM A)) − 1)/((T T 7/T 01) − 1)) ∗ 100 %Total to Total Eff

Prakhar Kapoor Master of Science Thesis

 Appendix C

Experimental Compressor Map

Fig. C-1 below shows the experimental performance map of the test compressor in terms the
variation of the pressure ratio with the volumetric flow rate. The experimental map has been
obtained as a result of the experiments conducted at MTEE.

Figure C-1: Experimental map of the test compressor

Master of Science Thesis Prakhar Kapoor

76 Experimental Compressor Map

Prakhar Kapoor Master of Science Thesis

 Appendix D

Component Wise Mesh Distribution

This Appendix provides the component wise mesh distribution for the different components
within the centrifugal compressors domain. The mesh elements for the inlet duct, impeller
and the diffuser is documented in tableD-1.

Table D-1: Component wise mesh in the compressor domain

S.No Component Mesh Size

1 Inlet Duct 50344
2 Impeller 218844
3 Diffuser 126000

Master of Science Thesis Prakhar Kapoor

78 Component Wise Mesh Distribution

Prakhar Kapoor Master of Science Thesis

 Bibliography

[1] D. Japikse, “Stall, stage stall, and surge,” in Proceedings of the Tenth Turbomachinery
Symposium, pp. 1–13, 1981.

[2] R. Antonia and P.-Å. Krogstad, “Turbulence structure in boundary layers over different
types of surface roughness,” Fluid Dynamics Research, vol. 28, no. 2, pp. 139–157, 2001.

[3] R. Raspopov, “Modelling curvature effects on turbulence for turbomachinary flows,”

Master’s thesis, TU Delft, Nederlands, 2013.

[4] D. Japikse, “Centrifugal compressor design and performance(book),” Wilder, VT: Con-
cepts ETI, Inc, 1996., 1996.

[5] W. Jansen, “Rotating stall in a radial vaneless diffuser,” Journal of Fluids Engineering,
vol. 86, no. 4, pp. 750–758, 1964.

[6] J. Paduano, E. Greitzer, and A. Epstein, “Compression system stability and active con-
trol,” Annual Review of Fluid Mechanics, vol. 33, no. 1, pp. 491–517, 2001.

[7] Y. Senoo and Y. Kinoshita, “Influence of inlet flow conditions and geometries of cen-
trifugal vaneless diffusers on critical flow angle for reverse flow,” Journal of Fluids Engi-
neering, vol. 99, no. 1, pp. 98–102, 1977.

[8] P. Frigne and R. Van den Braembussche, “Distinction between different types of impeller
and diffuser rotating stall in a centrifugal compressor with vaneless diffuser,” Journal of
Engineering for Gas Turbines and Power, vol. 106, no. 2, pp. 468–474, 1984.

[9] Y. Tsujimoto, Y. Yoshida, and Y. Mori, “Study of vaneless diffuser rotating stall based
on two-dimensional inviscid flow analysis,” Journal of fluids engineering, vol. 118, no. 1,
pp. 123–127, 1996.

[10] S. Niazi, Numerical simulation of rotating stall and surge alleviation in axial compressors.
PhD thesis, Georgia Institute of Technology, 2000.

Master of Science Thesis Prakhar Kapoor

80 Bibliography

[11] R. L. Simpson, “Turbulent boundary-layer separation,” Annual Review of Fluid Mechan-

ics, vol. 21, no. 1, pp. 205–232, 1989.

[12] M. S. Genç, I. Karasu, H. H. Acikel, M. T. Akpolat, and M. Genc, “Low reynolds number
flows and transition,” Low Reynolds Number Aerodynamics and Transition, Genc, MS
Ed.; InTech: Rijeka, Croatia, pp. 1–28, 2012.

[13] R. E. Mayle, “The 1991 igti scholar lecture: The role of laminar-turbulent transition in
gas turbine engines,” Journal of Turbomachinery, vol. 113, no. 4, pp. 509–536, 1991.

[14] F. Menter, R. Langtry, and S. Völker, “Transition modelling for general purpose cfd
codes,” Flow, Turbulence and Combustion, vol. 77, no. 1-4, pp. 277–303, 2006.

[15] R. Langtry, “A correlation based transition model using local variables for unstructured
parallelized cfd codes,” 2006.

[16] S. Likki, S. Y.B., P. Huang, F. Menter, and R. Langtry, “A correlation-based transition

model using local variables -part 1: Model formulation,” Journal of Turbomachinery,
vol. 128, pp. 413–422, 2004.

[17] S. Likki, S. Y.B., P. Huang, F. Menter, and R. Langtry, “A correlation-based transition

model using local variables -part ii: Test cases and industrial applications,” Journal of
Turbomachinery, vol. 128, pp. 423–434, 2004.

[18] A. Jamal, “Boundary layer tripping by a roughness element,” Tech. Rep. 4, 1996.

[19] P. Jonáš, O. Hladík, O. Mazur, and V. Uruba, “By-pass transition of flat plate boundary
layers on the surfaces near the limit of admissible roughness,” in Journal of Physics:
Conference Series, vol. 318, p. 032030, IOP Publishing, 2011.

[20] J. Nikuradse, Laws of flow in rough pipes. National Advisory Committee for Aeronautics
Washington, 1950.

[21] J. P. Bons, “A review of surface roughness effects in gas turbines,” Journal of Turboma-
chinery, vol. 132, no. 2, p. 021004, 2010.

[22] H. Schlichting, “Boundary layer theory,” 1979.

[23] H. Coleman, B. Hodge, and R. Taylor, “A re-evaluation of schlichtingâĂŹs surface rough-

ness experiment,” Journal of fluids engineering, vol. 106, no. 1, pp. 60–65, 1984.

[24] R. L. Simpson, “A generalized correlation of roughness density effects on the turbulent

boundary layer.,” AIAA Journal, vol. 11, no. 2, pp. 242–244, 1973.

[25] A. E. Von Doenhoff and E. A. Horton, A low-speed experimental investigation of the

effect of a sandpaper type of roughness on boundary-layer transition. NACA, 1958.

[26] M. W. Pinson and T. Wang, “Effects of leading-edge roughness on fluid flow and heat
transfer in the transitional boundary layer over a flat plate,” International journal of
heat and mass transfer, vol. 40, no. 12, pp. 2813–2823, 1997.

Prakhar Kapoor Master of Science Thesis

 81

[27] X. F. Zhang and H. Hodson, “Combined effects of surface trips and unsteady wakes
on the boundary layer development of an ultra-high-lift lp turbine blade,” Journal of
turbomachinery, vol. 127, no. 3, pp. 479–488, 2005.

[28] S. B. Pope, Turbulent flows. Cambridge university press, 2000.

[29] F. R. Menter, “Two-equation eddy-viscosity turbulence models for engineering applica-

tions,” AIAA journal, vol. 32, no. 8, pp. 1598–1605, 1994.

[30] D. Bergstorm, N. Kotey, and M. Tachie, “The effects of surface roughness on the mean
velocity profile in a turbulent boundary layer,” Journal of Fluids Engineering, vol. 124,
pp. 664–370, 2002.

[31] A. Savill, “A synthesis of t3 test case predictions,” Numerical simulation of unsteady

flows and transition to turbulence, pp. 404–442, 1992.

[32] P. Dassler, D. Kožulović, and A. Fiala, “Modelling of roughness-induced transition using

local variables,” in V European Conference on CFD, ECCOMAS CFD, 2010.

[33] T. Arts and M. L. de Rouvroit, “Aero-thermal performance of a two-dimensional highly

loaded transonic turbine nozzle guide vane: A test case for inviscid and viscous flow
computations,” Journal of turbomachinery, vol. 114, no. 1, pp. 147–154, 1992.

[34] Fluent, “12.1 user manual,” ANSYS Inc, 2010.

[35] A. Javed and E. Kamphues, “Evaluation of the influence of volute roughness on tur-
bocharger compressor performance from a manufacturing perspective,” in Proceedings of
the ASME Turbo Expo 2014, 2014.

[36] M. Olivero, Evolution of a centrifugal compressor: From turbocharger to micro gas tur-
bine applications. PhD thesis, TU Delft, Delft University of Technology, 2012.

[37] R. Chodkiewicz, K. Sobczak, A. Papierski, and T. Borzęcki, “Cfd code-a useful tool for
the turbomachinery designer,” TASK Quarterly: scientific bulletin of Academic Com-
puter Centre in Gdansk, vol. 6, pp. 553–575, 2002.

[38] F. Breugelmans and M. Sen, “Prerotation and fluid recirculation in the suction pipe of
centrifugal pumps,” in Proc. 11th Int. Pump Symp., Texas A&M Univ, pp. 165–180,
1982.

[39] M. Ishida, D. Sakaguchi, and H. Ueki, “Suppression of rotating stall by wall roughness
control in vaneless diffusers of centrifugal blowers,” Journal of turbomachinery, vol. 123,
no. 1, pp. 64–72, 2001.

[40] A. Abdelhamid, “Effects of vaneless diffuser geometry on flow instability in centrifu-

gal compression systems,” in American Society of Mechanical Engineers, Gas Turbine
Conference and Products Show, Houston, Tex, 1981.

[41] A. Jaatinen-Värri, P. Röyttä, T. Turunen-Saaresti, and A. Grönman, “Experimental

study of centrifugal compressor vaneless diffuser width,” Journal of Mechanical Science
and Technology, vol. 27, no. 4, pp. 1011–1020, 2013.

Master of Science Thesis Prakhar Kapoor

82 Bibliography

[42] H. Simon and A. Bulskamper, “On the evaluation of reynolds number and relative surface
roughness effects on centrifugal compressor performance based on systematic experimen-
tal investigations,” Journal of Engineering for Gas Turbines and Power, vol. 106, no. 2,
pp. 489–498, 1984.

[43] R. C. Dean and Y. Senoo, “Rotating wakes in vaneless diffusers,” Journal of Fluids
Engineering, vol. 82, no. 3, pp. 563–570, 1960.

[44] J. Kurokawa, J. Matsui, T. Kitahora, S. L. Saha, et al., “A new passive device to control
rotating stall in vaneless and vaned diffusers by radial grooves,” 1997.

Prakhar Kapoor Master of Science Thesis