Grant Musgrove Thesis

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/261098624
Computational Design of a Louver Particle Separator for Gas Turbine Engines
Conference Paper · June 2009

DOI: 10.1115/GT2009-60199
CITATIONS READS
15 144
5 authors, including:
Grant Musgrove Karen A. Thole

Southwest Research Institute Pennsylvania State University
28 PUBLICATIONS 357 CITATIONS 290 PUBLICATIONS 6,571 CITATIONS
SEE PROFILE SEE PROFILE
Eric Andrew Grover

Pratt & Whitney
16 PUBLICATIONS 404 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Pin Fin Heat Exchangers View project
Heat Exchanger Development View project
All content following this page was uploaded by Grant Musgrove on 04 January 2023.
The user has requested enhancement of the downloaded file.

The Pennsylvania State University
The Graduate School
Department of Mechanical and Nuclear Engineering
COMPUTATIONAL PREDICTIONS AND
EXPERIMENTAL MEASUREMENTS OF THE PERFORMANCE OF A
LOUVER PARTICLE SEPARATOR FOR USE IN GAS TURBINE ENGINES
A Thesis in
Mechanical Engineering
by
Grant O. Musgrove
© 2009 Grant O. Musgrove
Submitted in Partial Fulfillment

of the Requirements
for the Degree of
Master of Science
August 2009
The thesis of Grant O. Musgrove was reviewed and approved* by the following:
Karen A. Thole
Head of the Department of Mechanical and Nuclear Engineering
Professor of Mechanical Engineering
Thesis Advisor
H. Joseph Sommer III

Professor-In-Charge of MNE Graduate Programs
Professor of Mechanical and Nuclear Engineering
Horacio Perez-Blanco
Professor of Mechanical Engineering
*Signatures are on file in the Graduate School.
ii
Abstract
Gas turbine engines that power aircraft operate in harsh environments where solid
particles, such as sand, are ingested into the engine. Solid particles damage aircraft engines by
eroding and depositing on turbine components, resulting in decreased engine efficiency.
Particles in the engine are also present in flow that passes through coolant channels. In these
channels, particles deposit to create blockages. Blocked cooling channels in the turbine reduce
the flow resulting in increased component temperatures. High temperatures reduce component
life and lead to erosion on airfoils.
This study presents computational predictions and experimental measurements of the
performance of an inertial separator that removes solid particles from gas turbines. The
separator is intended to be placed in the secondary coolant flow path of a gas turbine engine,
located immediately downstream of the combustor. The separator studied here is made up of an
array of louvers followed by a static collector. The separator performance is determined from
the predicted and measured values of pressure loss and particle collection efficiency. The
effectiveness of different louver and collector configurations are studied using a two-dimensional
computational model matched to engine conditions at constant Reynolds number for a range of
sand sizes. An experimental method is developed to quantify the performance of two separator
configurations that differ only in louver geometry. Two- and three-dimensional computational
models matched to lab experiment conditions predict the separator performance over a range of
Reynolds numbers and sand sizes.
Two-dimensional predictions matched to engine conditions indicated that the orientation
of the flow circulation in the collector was critical to successfully capturing the particles. Two-
and three-dimensional computational predictions matched to lab experiment conditions were
within 10% of measured collection efficiencies. The separator performance was measured to be
consistent with predicted low pressure losses. Predicted and measured collection efficiencies
were highest at low Reynolds number with a maximum measured value of 35%.
iii
Table of Contents
List of Tables ................................................................................................................................. vi
List of Figures ............................................................................................................................... vii
Preface.......................................................................................................................................... xiv
Acknowledgments......................................................................................................................... xv
Paper 1: Computational Design of a Louver Particle Separator for Gas Turbine Engines............. 1
Abstract ............................................................................................................................... 1
Nomenclature ...................................................................................................................... 1
Introduction ......................................................................................................................... 2
Review of Relevant Literature ............................................................................................ 3
Louver Separator Geometry and Design Approach ............................................................ 6
Computational Methodology .............................................................................................. 8
Discussion of Results ........................................................................................................ 13
Conclusions ....................................................................................................................... 23
Acknowledgments............................................................................................................. 24
References ......................................................................................................................... 24
Paper 2: Performance Measurements of a Unique Louver Particle Separator for Gas Turbine
Engines .......................................................................................................................................... 27
Abstract ............................................................................................................................. 27
Nomenclature .................................................................................................................... 27
Introduction ....................................................................................................................... 28
Review of Relevant Literature .......................................................................................... 29
Experimental Facility and Methodology .......................................................................... 30
Computational Methodology ............................................................................................ 36
Measurement Uncertainty and Repeatability .................................................................... 37
Conclusions ....................................................................................................................... 51
Acknowledgments............................................................................................................. 52
References ......................................................................................................................... 52
iv
Conclusions and Summary ........................................................................................................... 55
Appendix A: Computational Study of the Louver Separator........................................................ 57

Nomenclature .................................................................................................................... 57
Introduction ....................................................................................................................... 57
Computational Methodology ............................................................................................ 58
References ......................................................................................................................... 97
Appendix B: Uncertainty Analysis ............................................................................................... 98

Nomenclature .................................................................................................................... 98
Precision Error .................................................................................................................. 99
Pressure Loss Coefficient Bias Error ................................................................................ 99
Collection Efficiency Bias Error ..................................................................................... 101
References ....................................................................................................................... 102
Appendix C: Description of the Test Facility ............................................................................. 103
Appendix D: Engineering Drawings ........................................................................................... 110
v
List of Tables
Table 1.1. Result of Changing Separator Parameters .................................................................... 7
Table 1.2. Parameter Range of Values ........................................................................................... 7
Table 1.3. Shape Factor Effect on Overall Collection ................................................................. 12
Table 2.1. Area Ratio and Louver Geometry ............................................................................... 35
Table 2.2. Effect of Collector Modifications ............................................................................... 36
Table A.1. 2D and 3D Computational Models............................................................................. 58
Table A.2. Boundary Conditions for Each Computational Model ............................................... 61
Table A.3. Operating Conditions for Each Computational Model .............................................. 61
Table A.4. Cell Skewness for Variable Louver Configuration .................................................... 62
Table A.5. Scaled Particle Sizes................................................................................................... 80
Table B.1. Constant Louver Uncertainty Values for Cp at Re = 46,400 ................................... 100
Table B.2. Constant Louver Uncertainty Values for Cp Re = 8,000 .......................................... 101
Table B.3. Constant Louver Uncertainty Values for η at Re = 46,400 ...................................... 101
Table B.4. Constant Louver Uncertainty Values for η at Re = 8,000 ........................................ 102
vi
List of Figures
Figure 1.1. The separator is located within the engine in an empty cavity downstream of the
combustor. ....................................................................................................................................... 6
Figure 1.2. Drawing showing the a) separator design space, b) primary louver parameters, and c)
louver shapes including straight, bent tip, concave, and convex. ................................................... 7
Figure 1.3. Two-dimensional louver model showing a) design space without collector and b)
design space with collector. ............................................................................................................ 8
Figure 1.4. Effect of restitution coefficient on collection efficiency for the configuration:
AR=1.50, 8 louvers, straight louvers, φ=30° to 45° to 30°, with collector. .................................. 11
Figure 1.5. Effect of shape factor on collection efficiency for the configuration: AR=1.50, 8
louvers, straight louvers, φ=30° to 45° to 30°, with collector. ..................................................... 11
Figure 1.6. Size distribution of particles used in the computational domain representing those of
interest [23]. .................................................................................................................................. 13
Figure 1.7. Pressure loss is shown to decrease with increasing area ratio for configurations with
straight louvers at φ=30°, without collector. ................................................................................. 14
Figure 1.8. Collection efficiency for area ratios of 1.25 and 1.50, straight louvers at φ=30°,
without collector. .......................................................................................................................... 15
Figure 1.9. Collection efficiency for varying louver angle and shape, AR=1.50, without
collector......................................................................................................................................... 17
Figure 1.10. Particle trajectories for 5µm particles, showing an increase of curvature with louver
length for AR=1.50, φ=30° configurations, without collector, with a) 8, b) 11, and c) 19 louvers.
....................................................................................................................................................... 18
Figure 1.11. Particle trajectories for 40µm particles, showing increasing surface deflections with
increased louver lengths for AR=1.50, φ=30° configurations, without collector, with a) 8, b) 11,
and c) 19 louvers. .......................................................................................................................... 18
Figure 1.12. Particle trajectories through an 8 louver, AR=1.50, bent tip φ=30° louver
configuration, without collector, for 5µm (upper) and 40 µm (lower) particle sizes. .................. 19
Figure 1.13. Collection efficiency is affected by varying louver angle and louver shape along the
louver array. .................................................................................................................................. 20
Figure 1.14. The best performing louver configuration varies the louver angle from φ = 30° to
45° to 30° along the array, resulting in higher small particle collection with 5µm particles shown.
....................................................................................................................................................... 20
vii
Figure 1.15. Key geometry features of the collector are shown to affect collection efficiency for
the configuration: AR=1.50, 8 louvers, straight louvers, φ = 30° to 45° to 30°. .......................... 22
Figure 1.16. Collector circulation patterns showing (upper) clockwise rotation with an inlet
vortex, and (lower) counter-clockwise circulation without an inlet vortex .................................. 22
Figure 1.17. Particle trajectory plot showing the collecting pattern for 5µm (upper) and 40µm
(lower) particles in the best separator design. ............................................................................... 23
Figure 2.1. The (a,b) open loop test facility is shown with the sand injection system and (c)
transparent test section. ................................................................................................................. 33
Figure 2.2. Sand size distribution by percent volume of ISO 12103-1 A4 test sand [16]. .......... 34
Figure 2.3. Both the variable louver (top) and constant louver (bottom) configurations
incorporated structural supports and a wedge and baffle at the collector inlet. ............................ 35
Figure 2.4. Test repeatability was shown by similar sand distribution through the test facility for
varying mass loading and sand amount: variable louver angle, Re = 25,000, 0 < Ds < 200µm. .. 39
Figure 2.5. Collection efficiency was independent of mass loading and sand amount for the
variable louver configuration at Re = 25,000 for sand sizes 0 < Ds < 200µm. ............................. 40
Figure 2.6. Pressure loss across both separator configurations increased for decreasing Reynolds
numbers. ........................................................................................................................................ 41
Figure 2.7. Collection efficiency for both configurations increased as Reynolds number
decreased for 0 < Ds < 200µm....................................................................................................... 42
Figure 2.8. Sand was uniformly collected across the collector span in increasing amounts with
increasing Reynolds number for 0 < Ds < 200µm......................................................................... 43
Figure 2.9. Collector circulation size decreased was observed to decrease with Reynolds number
for the (a) constant louver and (b) variable louver configurations for sizes 0 < Ds < 200µm. ..... 44
Figure 2.10. Distribution of sand through the test facility for the variable louver configuration
when varying Reynolds number for 0 < Ds < 200µm. .................................................................. 44
Figure 2.11. Distribution of sand through the test facility for the constant louver angle
configuration when varying Reynolds number for 0 < Ds < 200µm............................................. 45
Figure 2.12. Collection efficiency was dependent on particle size and louver configuration for
Re = 46,400. .................................................................................................................................. 47
Figure 2.13. Particle sizes Ds < 20µm were distributed evenly across the collector, however,
sizes Ds > 20µm deposited only at the collector ends for Re = 46,400. ........................................ 48
viii
Figure 2.14. Particle trajectories through the variable louver configuration were drawn from
observations for minimum and maximum sand size intervals at Re = 46,400. ............................ 48
Figure 2.15. Particle trajectories through the constant louver configuration were drawn from
observation for minimum and maximum sand size intervals at Re = 46,400. .............................. 49
Figure 2.16. Distribution of sand through the test facility for variable louver angle when varying
sand size at Re = 46,400. .............................................................................................................. 49
Figure 2.17. Distribution of sand through the test facility for constant louver angle when varying
sand size at Re = 46,400. .............................................................................................................. 50
Figure A.1. The 2D engine model did not include part thickness or the wedge and baffle
modification, variable louver configuration shown. ..................................................................... 58
Figure A.2. The 2D experiment matched model included part thickness and the wedge and baffle
modification, (upper) variable and (lower) constant louver configurations. ................................ 59
Figure A.3. The 3D experiment matched model was created by extruding the 2D experiment
matched geometry into the spanwise direction, (upper) variable and (lower) constant louver
configurations. .............................................................................................................................. 59
Figure A.4. The domain for the 2D engine model used an (a) unstructured mesh in the design
space and (b) structured mesh outside of the design space. .......................................................... 60
Figure A.5. The 2D and 3D experiment matched domains matched the test facility from the sand
injection location to the downstream channel exit, using a combination of (a) unstructured and
(b) structured mesh regions. .......................................................................................................... 61
Figure A.6. Collection efficiency did not significantly change for the 2D experiment matched
grid size greater than 6.4 x 104 cells for the variable louver configuration at Re = 46,400. ........ 62
Figure A.7. The collection of particle sizes less than 20 µm varied with the 2D experiment
matched grid size more than other sizes for the variable louver configuration at Re = 46,400. .. 63
Figure A.8. For most particles sizes, a 3D grid size of 3.0 x 106 did not significantly affect the
collection efficiency of the constant louver configuration at Re = 46,400. .................................. 63
Figure A.9. The 3.0 x 106 cell grid did not significantly affect collection efficiency for each
particle size range injected compared to the 1.6 x 106 cell grid for the constant louver
configuration at Re = 46,400 ........................................................................................................ 64
Figure A.10. One hundred particles was sufficient for independent collection efficiency for the
2D engine model of the variable louver configuration at Re = 135,500....................................... 66
Figure A.11. One thousand particles were injected in the 2D experiment matched model to
obtain independent collection efficiency for the variable louver configuration at Re = 46,400. . 66
ix
Figure A.12. The number of particles and spacing of particle injections were investigated to
determine an independent particle injection method. ................................................................... 67
Figure A.13. A total of eleven spanwise injections of 1,000 particles each with spanwise spacing
H/2 were required for independent collection efficiency of the constant louver configuration at
Re = 46,400. .................................................................................................................................. 67
Figure A.14. Spanwise injection spacing of H/8 was required for independent collection
efficiency results for the constant louver configuration at Re = 46,400. ...................................... 68
Figure A.15. Different louver shapes and angles were simulated to improve particle collection.
....................................................................................................................................................... 69
Figure A.16. Curved louvers did not improve collection efficiency relative to straight louvers
and bent tip louvers over a range of louver angles; AR=1.50, 11 louvers, no collector. .............. 70
Figure A.17. The sensitivity of the louver tip was investigated by varying tip angle (α) and tip
length............................................................................................................................................. 71
Figure A.18. Varying the louver tip angle and length did not affect collection efficiency;
AR=1.50, 11 louvers, no collector. ............................................................................................... 71
Figure A.19. Louver configurations incremented the louver angle along the array for straight and
bent tip louvers. ............................................................................................................................. 72
Figure A.20. Bent tip louvers did not increase the collection efficiency of variable louver
configurations; AR=1.50, 11 louvers, no collector. ...................................................................... 72
Figure A.21. Final modifications were confined to individual louvers in the best performing
louver configurations to attempt to increase small particle collection. ........................................ 73
Figure A.22. Modification of a few louvers in a variable louver configuration reduced particle
collection; AR=1.50, 8 louvers, no collector. ............................................................................... 74
Figure A.23. The collector inlet design was investigated to increase particle collection. ........... 75
Figure A.24. The collector with a large inlet and inlet lip performed better than a larger inlet or
without an inlet lip. ....................................................................................................................... 75
Figure A.25. The collector inlet lip direction was investigated for similar collector shape and
size. ............................................................................................................................................... 76
Figure A.26. The inward inlet lip had better mid size particle collection than the outward inlet
lip. ................................................................................................................................................. 76
Figure A.27. Collector shape modifications included changes to the back wall of the collector. 77
Figure A.28. The back wall of the collector affected the collection of large particles. ............... 77
x
Figure A.29. Two collector configurations were axially shortened to determine the effects of
collector length.............................................................................................................................. 78
Figure A.30. Collection efficiency is not affected by axial length for the curved collector. ...... 78
Figure A.31. The effect of the curvature of the bottom wall on the collection efficiency was
investigated. .................................................................................................................................. 79
Figure A.32. Modifying the contour of the bottom wall of the curved collector did not improve
collection efficiency over the original design. .............................................................................. 79
Figure A.33. Collection efficiency predictions were not significantly affected by the thickness of
the separator for the 2D engine model: Re=135,500, No wedge and baffle modification. .......... 81
Figure A.34. The collection efficiency of configurations without the wedge and baffle
modification were affected by Reynolds number when part thickness was not included compared
to when part thickness was included. ............................................................................................ 82
Figure A.35. The wedge and baffle lowered the predicted overall collection efficiency for both
the 2D engine and 2D experiment matched models at Re=46,400. .............................................. 82
Figure A.36. Pressure loss across both separator configurations increased for decreasing
Reynolds numbers. ........................................................................................................................ 84
Figure A.37. 2D and 3D Cp predictions were similar for the variable louver (left) and constant
louver (right) configurations at Re = 46,400. ............................................................................... 84
louver (right) configurations at Re = 8,000. ................................................................................. 85
Figure A.39. Collection efficiency predictions of sand size range 0-200µm were within 10% of
measurements over the range of Reynolds numbers for the variable and constant louver
configurations. .............................................................................................................................. 86
Figure A.40. The 3D model predicted that the favorable circulation in the constant louver
configuration was disrupted due to spanwise flow in the collector at Re = 46,400, Z=1.25H (left),
Z=2.5H (right). .............................................................................................................................. 86
Figure A.41. Favorable circulation in the collector was uniform across the span for the variable
louver (left) and constant louver (right) configurations at Re = 8,000. ........................................ 87
louver (left) and constant louver (right) configurations at Re = 46,400. ...................................... 88
Figure A.43. Collector pathlines predicted by the 3D models indicated that the constant louver
configuration (left) had a weaker inlet vortex than the variable louver configuration (right). ..... 89
xi
Figure A.44. Stagnation location was higher on the wedge face for the variable louver
configuration (left) than the constant louver configuration (right) at Re = 46,400. ..................... 90
Figure A.45. Computational results over-predicted collection efficiency for each sand size
interval at Re = 46,400. ................................................................................................................. 91
Figure A.46. Particle trajectories of 10µm (blue) and 30µm (red) particles for the 2D variable
louver (top) and 2D constant louver (bottom) configurations at Re = 46,400.............................. 92
louver (top) and 2D constant louver (bottom) configurations at Re = 46,400.............................. 93
Figure A.48. 2D predictions showed that collection efficiency was dependent on both particle
size and Reynolds number for the variable louver configuration. ................................................ 94
size and Reynolds number for the constant louver configuration................................................. 94
size and Reynolds number for the variable louver configuration. ................................................ 95
size and Reynolds number for the constant louver configuration................................................. 95
Figure A.52. The collector circulation zones are independent of Reynolds number for the (left)
variable louver and dependent on Reynolds number for the (right) constant louver configuration.
....................................................................................................................................................... 96
Figure C.1. Upstream flow uniformity was achieved using a 48.3mm x 178mm (1.9″ x 7″)
splash plate with center located 51mm (2″) left of the channel midspan. .................................. 104
Figure C.2. A splash plate was placed 51mm (2″) left of midspan at the upstream channel inlet
to produce uniform flow at the test section inlet......................................................................... 104
Figure C.3. The a) feeder conveyed sand from the b) hopper to the mixing manifold by way of
an b) auger screw. ....................................................................................................................... 106
Figure C.4. The feed rate of the sand feeder was calibrated over the controller speed range. .. 106
Figure C.5. The sand manifold used three compressed air ports (not shown) to mix the sand
before passing into the upstream channel. .................................................................................. 107
Figure C.6. A section of the upstream channel between the test section and sand injection slot
was removable for sand retrieval. ............................................................................................... 107
Figure C.7. The test section had a removable lid to allow the interchanging of separator
configurations. ............................................................................................................................ 108
xii
Figure C.8. The tested separators included structural reinforcements, guide tracks, and endcaps
(variable louver configuration shown). ....................................................................................... 108
Figure C.9. The filter box removed sand that was not collected by the separator using three
filters placed in series.................................................................................................................. 109
xiii
Preface
This thesis is written in manuscript form consisting of two papers. The first paper,
entitled “Computational Design of a Louver Particle Separator for Gas Turbine Engines,” is a
two-dimensional computational study of the louver separator that was presented at the 2009
ASME International Gas Turbine Institute Conference in Orlando, FL. The effects of area ratio,
louver geometry, and collector geometry on separator performance are investigated at engine
scale. Separator performance is determined from the predicted pressure loss and collection
efficiency of the separator configuration.
The second paper, entitled “Performance Measurements of a Unique Louver Particle
Separator for Gas Turbine Engines,” is an experimental study of the performance of two louver
separator configurations. The effects of Reynolds number and sand size on collection efficiency
were investigated. The louver configurations, experimental test facility, and experimental
method are described. Experimental measurements of collection efficiency and pressure loss
coefficient are compared to two-and three-dimensional predictions that were matched to the
experimental conditions. This paper will be submitted to the 2010 ASME International Gas
Turbine Institute Conference in Glasgow, Scotland.
Four appendices are included that describe in detail topics that are presented in both
papers. Appendix A gives details of the computational study of the louver separator. The
computational study includes configurations not presented in the first paper as well as detailed
explanations of the results of the two- and three-dimensional models presented in the second
paper. Appendix B describes the uncertainty calculations of pressure loss coefficient and
collection efficiency. Appendix C presents the experimental measurement method in detail as
well as a description of the disassembly of the test facility for the removal of deposited sand.
Appendix D contains engineering drawings of the experimental test section and louver
separators.
xiv
Acknowledgments
I would like to thank Pratt & Whitney for sponsoring and funding this research work.
Specifically Eric Grover, Joe Barker, and Ryan Levy who all helped to guide the research work
over the past two years.
At Penn State there are many people that deserve my appreciation. My advisor Karen
Thole, who always pushed me one step further in everything I did. Without her guidance I
would not be the engineer that I am today. The people I worked with in the lab every day were
an incredible group that I’ll never forget (Mike Barringer, Steve Lynch, Gaelyn Neely, Seth
Lawson, Jason Ostanek, Steve Weaver, Gina Casadei, Alan Thrift). They are some of the best
friends I have ever had. I want to especially thank Mike Barringer and Steve Lynch who always
made time to help me with anything, no matter what they were doing. Thank you, Gaelyn for
always being there to talk to about work, school, and life.
I could not have survived in Pennsylvania if it were not for the support of my family and
friends living in Texas, Oklahoma, and New Mexico. Thank you Mom and Dad for always
being there for me, and raising me to be the person that I am. You are the best parents anyone
could have. If it wasn’t for my brother I wouldn’t have become an engineer. Thank you
Cameron for your advice, you always reminded me of the bigger picture in life.
The person that I owe my thanks to the most is my fiancée Julie. She is my best friend,
and makes me laugh every day. Julie has supported me in everything that I have done. She was
always there to listen to me talk about work and school, although she sometimes didn’t
understand what I was talking about. Graduate school would have been much more difficult
without her in my life.
xv
Paper 1: Computational Design of a Louver Particle
Separator for Gas Turbine Engines
Presented at the 2009 IGTI Conference (GT2009-60199)*
Abstract
The extreme temperatures in a jet engine require the use of thermal barrier coatings and
internal cooling channels to keep the components in the turbine section below their melting
temperature. The presence of solid particles in the engine's gas path can erode thermal coatings
and clog cooling channels, thereby reducing part life and engine performance. This study uses
computational fluid dynamics to design the geometry of a static, inertial particle separator to
remove small particles, such as sand, from the engine flow. The concept for the inertial
separator includes the usage of a multiple louver array followed by a particle collector. The
results of the study show a louver design can separate particles while not incurring large pressure
loss.
Nomenclature
A louver anchor length
AR area ratio, AR=δ(# Louvers)/H
AsR aspect ratio of design space, AsR=X/H
CD particle drag coefficient
Dc Stokes characteristic dimension
dp particle diameter
∆P pressure loss, ∆P=(Pin-Pex)total/(Pin)total
H height of design space
L louver length, L=δ/sinφ
Re inlet Reynolds number Re=ρΗUin/µ
Rep particle Reynolds number Rep=ρdp|Up-U|/µ
Resph particle Reynolds number using the dia. of a sphere with equivalent particle volume
s surface area of a sphere with equivalent particle volume
S surface area of the actual particle
U local fluid velocity
Up local absolute particle velocity
w distance between last louver tip and top domain wall
X axial length of design space
Z particle shape factor, Z=s/S
*Musgrove, G.O., Barringer, M.D., Thole, K.A., Grover, E., Barker, J., 2009, “Computational Design of a Louver
Particle Separator for Gas Turbine Engines,” International Gas Turbine and Aeroengine Congress and Exposition,
Orlando, FL, ASME Paper GT2009-60199
1
Greek:
δ louver gap, δ=AR·H/(# Louvers)
η collection efficiency, η=1-(Nescaped/Ninjected)particles
θ louver axis angle , θ=tan-1(1/AsR)
µ fluid dynamic viscosity
ρf fluid density
ρp particle density
τrea particle reaction time, τres=ρpdp2/18µ
τres particle residence time, τres=Dc/Up
φ louver angle with respect to θ
Introduction
Ingestion of solid particles into gas turbine engines occurs during takeoff, landing, and in
flight. On the ground, helicopters stir up dust with their rotors that result in severe ingestion.
Airplanes, on the other hand, create an intake vortex between the engine inlet and the ground that
lifts debris and sand into the engine [1]. During flight, particles can be ingested when flying low
in dusty environments, such as deserts, or when flying through volcanic ash clouds [2]. Ingested
particles reduce the performance and life of an engine through surface erosion, deposition, and
the blocking of cooling channels in the turbine. Cooling channel blockage is increased by
particle size [3,4] and amount of particles [3,4,5] ingested.
While there are many particle separation methods, inertial separators are best applicable
to aircraft engines because of their ability to handle high flow rates while not significantly
increasing pressure losses in the engine. Two commonly applied inertial separators are cyclones
and particle deflectors combined with some type of particle trap. Both types make use of inertial
forces to separate particles from the fluid flow. Variations of the particle deflectors and cyclones
have been applied to gas turbines [6-10] and are generally integrated into various locations
within the engine. A deflector design that is often used in helicopters is integrated into the inlet
of the engine and is known as an inlet particle separator. An inlet particle separator contains a
secondary flow system (scavenge flow) that is separate from the engine flow. The scavenge flow
convects particles away from the main gas path through a separate duct. Another common
deflector is a louver design, which also uses a scavenge flow duct similar to the inlet particle
separator, but the louver design has yet to be applied to aircraft engines. Neither the inlet
particle separator nor louver separator designs have been applied in a location other than the inlet
to gas turbines.
2
The particle separator concept presented in this paper is a louver separator, but is one that
does not require a scavenge flow. The separator presented in this paper is one that uses a static
collection bin (collector) to capture the separated particles, located downstream of a series of
louvers. This particular separator concept is intended for placement within the combustor bypass
flow. The particle separator is intended to remove particles that could potentially block cooling
channels in the turbine section of the engine. This paper presents a computational study of a
particle separator containing both a louver section and a collector section. The focus of the study
is to minimize the pressure loss and maximize the particle collection. This study evaluates
various louver arrays and collector geometries while maintaining a fixed channel size (design
space) and set of flow boundary conditions. The remainder of the paper includes a brief
presentation of relevant literature, explanation of the geometry and computational model of the
separator concept, and presents particle collection and pressure loss results for various simulated
geometry configurations.
Review of Relevant Literature

Past literature indicates that inertial separators have potential for turbine applications to
remove particles from engine flow. This section documents previous methods used for
computationally modeling two-phase flow in particle separators. The summary of the literature
also provides insights that helped guide the work completed in the current study.
Many of the studies presented in the literature have focused on inlet particle separators
applied to helicopters. Breitman et al. [11] performed a numerical study, similar to Vittal et al.
[12] and Shieh et al. [13], in which they found that large particles are more resistant than small
particles to changes in flow trajectory due to their high inertia. Stokes theory states that at low
Stokes numbers (small particle diameters) particles follow the fluid flow whereas at high Stokes
numbers (large particle diameters) the particles do not follow the flow.
Zedan et al. [16] expanded the work of others [11-13] to show computational agreement
of overall particle separation with measurements for two distinct inlet particle separators. Zedan
et al. reported particle collection efficiencies between 82-85%. Using a similar conceptual
design and modeling approach, Ghenaiet and Tan [15] reported separation efficiencies greater
than 80%, depending on sand size distribution and scavenge flow rates. Modifying the separator
geometry to control the particle trajectories and surface impacts, Ghenaiet and Tan increased
3
overall particle separation efficiency by almost 10%. Collection of small particles was increased
by as much as 80%. Ghenaiet and Tan’s analysis agreed with Breitman et al. in that the small
particle trajectories are dominated by drag forces. Conversely, large particle trajectories are
more resistive than small particles to changes in flow direction.
A number of investigators have evaluated design features associated with louver
separators. These design features include the louver angle, louver length, louver spacing, louver
overlap, scavenge flow rate, and Reynolds number. Past studies show effects of these design
parameters on pressure losses and particle separation.
One of the earliest studies of a louver separator was performed by Zverev [16].
Separation efficiency was experimentally measured and found to increase with increased
scavenge flow rate, increased Reynolds number, and decreased louver spacing. Zverev found
that increasing the louver angle reduced the pressure loss of the separator. The highest particle
separation occurred for a 30° louver angle. The trends in separation efficiency found by Zverev
were experimentally confirmed for other louver separators by Smith and Goglia [17], Gee and
Cole [18], and Poulton et al. [19].
Smith and Goglia [17] found that particle separation was highly dependent upon the
location of the particle’s impact on the louver. They found that separation efficiency could be
increased by modifying the louver array to eliminate the louver overlap and to include a louver
anchor. Furthermore, their work defined that equal flow rates through each louver passage was
necessary for good collection efficiency, confirmed by Gee and Cole [18]. Smith and Goglia
found that dust loading has little effect on particle collection, which was also confirmed
experimentally by Gee and Cole [18]. Dust loading refers to the mass flow rate of particles
relative to the flow stream. Gee and Cole showed that a V-shaped louver did not improve
particle separation over a straight louver although V-shaped louvers reduced the pressure loss.
Poulton et al. [19] used flow visualization on a louver separator and found a significant
separation region on the backside of the louvers. This region was found to affect the particle
trajectories near the louvers by changing the radius of curvature of the flow around the louvers.
The separation region also increased the pressure loss of the separator by reducing the effective
flow area between the louvers. Poulton et al. found that pressure loss could be reduced by
spacing the louvers such that the flow along the backside of the louver reattached to the louver.
Additional experiments by Poulton et al. indicated that separation efficiency increased with
4
louver thickness and reduced louver spacing. The effect of Reynolds number on separation
efficiency agreed with previous studies, but Poulton et al. showed that there is an upper limit.
Pressure loss was reported to increase with louver length, and it was found that there is a
favorable combination of louver overlap and louver spacing.
Past studies have modeled two-phase flow in inertial separators by first solving for the
flow field in the separator, and then calculating particle trajectories by solving a force balance on
each particle. This method assumes that the particle concentration in the fluid is small enough
that particle-particle interactions are negligible. This method also assumes that there is no effect
of the particles on the flow field. Methods for solving the flow field in the separator have
included both ideal flow [11,12 ,18,19,20] and fully turbulent viscous models [14,15].
The force balance used to calculate a particle trajectory includes a drag force term. Most
investigators [11,14,18-20] have used a modified form of the Stokes’ drag coefficient on a
sphere. To correct for the irregular shape of solid particles, Breitman et al. [11] and Zedan et al.
[14] made use of empirical shape factor data to modify the drag force term. Particle shape factor
is used to quantify how spherical a particle is. Shape factor ranges from a value of zero (a highly
irregularly shaped particle) to a value of one (a perfectly shaped sphere).
Another important modeling consideration is the use of a particle restitution coefficient,
which is the ratio of particle velocity after and before an impact. A value of one indicates that
velocity magnitude is unchanged by the impact, and zero indicates the particle has no velocity
after impact. Restitution coefficients can vary with impact angle and material. Previous studies
[11,12,14 ,19] have found agreement with experimental results when using particle restitution
coefficients. Breitman et al. [11] and Poulton et al. [19] used constant values for restitution
coefficients, where Breitman et al. found that using values of 1 and 0.85 for the tangential and
normal restitution coefficients, respectively, improved agreement with experimental data.
Presented in this paper is a computational design of the louver-type separator using a
two-dimensional model of the louver array and collector. The computational model includes the
use of a k-ω turbulence model with enhanced wall functions for the near wall region. Collection
efficiency and pressure loss results are presented for a number of design parameters including
area ratio, louver angle, louver shape, number of louvers, and collector geometry. Unique about
this study is that the evaluated louver separator uses a static collector instead of scavenge flow.
5
Louver Separator Geometry and Design Approach
For this particular application, a louver separator with a fixed design space is considered
for use in an aircraft gas turbine. The intended location of the separator is downstream of the
combustor, as shown in Figure 1.1, along the outer casing of the first stage turbine. Here, the
particle separator is intended to remove particles from the engine’s bypass flow before the
bypass flow enters turbine components, such as the blade outer air seals and airfoils. The
specifics of the louver design are illustrated in Figure 1.2. The objective of the louver design is
to direct particles to a downstream collector while passing particle-free flow through the louver
passages. As can be seen there are several parameters that can be evaluated to achieve good
particle collection with minimal pressure losses. These parameters include louver gap, length,
angle, and shape. Louver shapes considered include straight, bent-tip, concave, and convex
surfaces.
The design space considered for the overall louver design was held constant throughout
this study as shown in Figure 1.2a. This constraint required that some of the louver parameters
remain constant. The louver array angle, θ, was held constant, but louver gap and length were
varied, as shown in Figure 1.2b and summarized in Table 1.1. Previous studies [16,18,19] have
shown the effects of louver length and spacing on particle separation and pressure loss.
Table 1.2 shows the range of values for the louver parameters investigated over the
course of the current study.
Design Space
Combustor Vane Blade Vane
Figure 1.1. The separator is located within the engine in an empty cavity downstream of the
combustor.
6
a) Louvers Collector
H
b) w
δ L
φ A
θ
c)
Figure 1.2. Drawing showing the a) separator design space, b) primary louver parameters, and c)
louver shapes including straight, bent tip, concave, and convex.
Table 1.1. Result of Changing Separator Parameters
Parameter Change Affected Parameter Effect
Number of Louvers ↑ Louver Gap, δ ↓
Area Ratio, AR ↑ Louver Gap, δ ↓
Louver Angle, φ ↑ Louver Length, L ↓
Table 1.2. Parameter Range of Values

Parameter Min Max
Louver Angle, φ 30°, 45°, 60°
Louver Gap, δ/X 0.010 0.038
Louver Length, L/X 0.012 0.075
Louver Anchor, A/L 0.021 6.449
Dimension w, w/H 0.126 0.257
Area Ratio, AR 0.5 1.50
Louver thickness was neglected during this study since it has been shown to affect
separation efficiency by no more than 10% [19]. Other computational studies [17-19] have also
neglected louver thickness. Furthermore, the louver length, louver gap, louver anchor length and
the dimension w, shown in Figure 1.2b, are not controlled parameters. Instead, their values are
determined from the relations presented in the nomenclature. The controlled parameters are the
number of louvers, area ratio, louver angle, and louver shape. However, dimension w was
7
required to be larger than the louver gap of the last louver row so as to not influence the area
ratio of the separator.
The collector, shown in Figure 1.2a, has key features that were modified, including the
collector inlet and the surface shape. Different overall collector shapes were investigated to find
a design that trapped particles using surface deflections and by entraining particles in a favorable
circulation pattern within the collector.
To design the particle separator for minimal pressure losses and good particle collection,
the design parameters were independently investigated. A methodical approach was taken by
considering each of the following three features of the louver separator: the area ratio, the louver
geometry, and the collector design. More specifically, the best area ratio, AR, was first
determined by evaluating pressure loss across the louver array until the pressure loss was
reduced to its lowest value. This best area ratio was then held fixed and the most effective
louver geometry without a collector was determined by evaluating collection efficiency until the
collection efficiency was increased to its maximum value. The louver geometry without the
collector is shown in Figure 1.3a. After having determined the best louver geometry and area
ratio, a collector was added to the configuration, as shown in Figure 1.3b. An effective collector
design was defined by one that increased particle collection while not significantly increasing the
pressure loss. Results indicated that the pressure loss was not significantly affected by the
collector and, as such, the particle collection efficiency was the focus.
a)
b)
H
H X X
Figure 1.3. Two-dimensional louver model showing a) design space without collector and b)
design space with collector.
Computational Methodology
All computational work was performed using FLUENT 6.3.26 [21], a commercially
available CFD solver, to solve for the flow field and particle trajectories. A Lagrangian particle
tracking model was chosen, which uncoupled the particle trajectory calculations from the flow
field calculations, similar to previous studies [11,12,14 ,15 ,18-20].
8
The separator flow was solved using RANS equations with the k-ω turbulence model, as
well as enhanced wall functions that blend the linear and logarithmic laws of the wall. The
computational domain, illustrated in Figure 1.3b, has upstream and downstream sections added
to the design space each with axial lengths of H and X. The downstream section converges to an
exit height 60% of the inlet height to remove reversed flow at the exit boundary. Temperatures
in the computational domain were considered constant. Mass flow inlet and pressure outlet
boundary conditions used engine matched values.
The remaining domain boundaries and separator surfaces were modeled as no-slip walls.
The flow was assumed to be steady, incompressible, and an ideal gas. Flow solutions were 2nd
order accurate and generally took, on average, 5,000 iterations to meet the convergence criteria
of 10-6 for all variables.
To ensure grid insensitivity, the computational grid was refined until the calculated
pressure loss did not change with grid refinement. The grid was refined in regions where the
local total pressure gradient was less than 10% of the maximum gradient. A grid size of
approximately 45,000 cells was required for grid independent flow field results, including
approximately 4,000 cells inside the collector. It was found that grid resolution in the collector
affected particle trajectory calculations. The grid for the louver separator consisted of
quadrilateral cells. Grid quality was evaluated by cell skewness, in which 94% of all cells had
skewness less than 0.60, and an average skewness of 0.15.
The modeling of particle trajectories made use of FLUENT’s discrete phase model,
which calculates trajectories using flow velocity from the solved flow field. The discrete phase
model is applicable when the volume ratio of particles to fluid is much less than one [22]. The
discrete phase model neglects particle-particle interactions and assumes that the presence of the
particles does not change the flow field. Similar to previous works [11,12,14,15,18-20], a force
balance equation, which is shown per unit mass in Equation 1.1 [22], was used to calculate
particle trajectories.
dU p 18µ C D Re p g (ρ p − ρ f )
= 2
(U − U p ) + + Fx (1.1)
dt ρpd p 24 ρf
CD =
24
(1 + Re sph b 2 ) + b 3 Re sph (1.2)
Re sph b 4 + Re sph
9
The equation is reduced to only a drag force term by neglecting gravity and the term Fx.
The variable Fx is an additional force term, per unit mass, that describes the force to accelerate
the fluid surrounding the particle and the pressure gradient of the fluid acting on the particle.
This Fx term can be neglected when the fluid to particle density ratio is much less than one [22].
In this study the ratio of fluid to particle density is on the order of 10-3.
Particle trajectories were calculated from the integration of Equation 1.1. Turbulent
dispersion of particle trajectories was not accounted for; instead, mean velocity values were used
in Equation 1.1. Previous studies [11,12] using the mean velocity to calculate particle
trajectories found good agreement with experimental data. The number of integration steps and
length factor were chosen to be 20,000 and 6 mm, respectively. The length factor is used to
determine the integration time step for solving equation 1. These values of integration steps and
length factor allow sufficient calculation of particle trajectories to allow all particles not trapped
in the collector to escape the domain. The domain exit was set to an escape boundary condition
to allow uncollected particles to leave the domain.
Previous studies [11,12,14,15,19] have used a range of values for restitution coefficient
and shape factor. Throughout this study the particles were assumed to be spherical and to have
perfectly elastic impacts with the wall boundaries in the separator. Shape factor is used to
modify the drag coefficient, CD, shown in Equation 1.2 [22], where bn represents polynomial and
exponential functions of shape factor. Restitution coefficient is used to determine the normal
and tangential particle velocities after impact with a wall boundary. In this study, a value of one
was used for both the restitution coefficient and shape factor. However, changes to the
restitution coefficient and shape factor can slightly influence the collection efficiency results,
shown in Figure 1.4 and Figure 1.5.
Figure 1.4 shows the dependence of collection efficiency on restitution coefficient for the
best separator design found in this study, as will be described in a later section. Restitution
coefficient is shown to have little effect on small particles, whereas low restitution coefficients
increase collection efficiency for the large particles. An accurate restitution coefficient should be
chosen based on the particle and surface materials as well as the impact angle. Previous studies
[11,12,14,15] have used empirical relations taken from particle impact studies. Shape factor,
defined in the nomenclature [22], is shown in Figure 1.5 for the best separator design found in
this study. Shape factor affects small particle collection more than large particle collection
10
because low shape factor values increase particle drag. Large particle drag influences small
particle trajectories more than large particle trajectories. Large particles are not affected due to
their high particle inertia, which makes them less susceptible than small particles to directional
changes in the flow.
100%
80%
Restitution Coefficient
60%
1.0
η 0.50
0.25
40% 0.10
20%
0%
0 10 20 30 40 50
µ m)
Particle Size (µ
Figure 1.4. Effect of restitution coefficient on collection efficiency for the configuration:
AR=1.50, 8 louvers, straight louvers, φ=30° to 45° to 30°, with collector.
100%
80%
60% Shape Factor

η 1.0
0.82
40% 0.70
0.50
20%
0%
0 10 20 30 40 50
µ m)
Particle Size (µ
Figure 1.5. Effect of shape factor on collection efficiency for the configuration: AR=1.50, 8
louvers, straight louvers, φ=30° to 45° to 30°, with collector.
11
A shape factor of 0.82 is consistent with previous studies [11,15] and is shown in Figure
1.5 to give collection efficiency results within 20% of a shape factor value of 1.0 for most
particle sizes. It must also be pointed out that previous studies [11,15] use a different equation to
define shape factor. An approximated relationship between the shape factor equation used in this
study and previous studies [11,15] was found to indicate a shape factor of 0.82 was consistent
with the previous studies [11,15]. In addition to the shape factor dependency shown in Figure
1.5, overall particle collection decreases with shape factor, shown in Table 1.3. Overall particle
collection values are calculated by integrating the collection efficiency curves shown in Figure
1.5.
Table 1.3. Shape Factor Effect on Overall Collection
Shape Factor 1.0 0.82 0.70 0.50

Overall Collection 7% 8% 1% 4%
The particle size distribution of interest ranged from 1-50 µm. This size range makes up
75% of sand sizes found in atmosphere, as shown in Figure 1.6. Walsh et al. [4] found that
blockage of turbine cooling channels occurs for sand with particle sizes smaller than 100 µm.
For simulations in FLUENT, ten discrete particle sizes were injected with the same
number of particles per size injected, as shown in Figure 1.6. Each of the ten particle sizes was
injected equally spaced along the inlet of the computational domain in groups of 100. The
number of particles injected was determined to be sufficient for collection efficiency results to be
independent of the number of injected particles.
Particle trajectories predicted by FLUENT were checked using the Stokes number, which
is a dimensionless ratio of particle reaction time, τrea, to residence time, τres [24]. A Stokes
number much less than one indicates that the particle will follow the flow field and a value much
greater than one indicates that the particle will deviate from the flow. Stokes numbers were
calculated for various particle sizes along their trajectories, particularly near the louvers. Stokes
number predictions of whether or not particles separated from the flow direction were found to
be consistent with the computational predictions.
Collection efficiency, η, was used as the variable to define how successful the louver
separator removed particles from the flow. The efficiency, defined in the nomenclature, is
calculated from the number of particles injected at the domain inlet and the number of particles
12
that escape the domain exit. Particles that do not escape the domain exit were assumed to be
collected. This differs from the reviewed numerical studies [11,12,14,15,18,19] that base
separation efficiency on mass rather than on number of particles.
100 100
Number Injected
% Passing 80
60
40
20
0
0.1 1 10 100 1000
µ m)
Particle Size (µ
Figure 1.6. Size distribution of particles used in the computational domain representing those of
interest [23].
In this study the efficiency was determined for each particle diameter injected. As such,
the collection efficiency reported in this study is for a given particle size. Pressure loss across
the separator was evaluated by integrating mass-weighted total pressure values at the inlet and
exit of the computational domain.
Discussion of Results
A methodical approach was used in evaluating three primary components of the
separator: the area ratio; the louver design; and the collector design. The inlet Reynolds number
remained constant for all studies at 135,500. Pressure loss and collection efficiency predictions
were considered for all calculations. The next sections of the paper present a summary of what
was learned for each of the steps. Results of previous studies [11,15] have shown that small
particles (less than 7µm) were more difficult to separate from the flowstream than large particles
(greater than 40µm). Our study agreed with those findings. In this paper, particles less than
7µm are considered small particles and particles greater than 40µm are considered large
particles. The focus in designing the area ratio and louver geometry was to increase the
13
collection of small particles, whereas the focus for the collector design was to increase the
collection of large particles.
Selection of the Area Ratio

The best area ratio for the separator was determined using an initial louver geometry of
straight louvers with φ=30° for different numbers of louvers. The focus was to ensure particle
collection while minimizing the pressure loss.
Figure 1.7 shows that increasing the area ratio from 0.50 to 1.50 decreases the pressure
loss by approximately 0.5%. The reduction of flow restriction through the louvers causes the
decrease in pressure loss. Although area ratios of 1.25 and 1.50 result in the lowest pressure loss
through the array, the predicted pressure losses of all of the area ratios is considered to be
insignificant to overall engine performance.
2%
Area Ratio
1.5% 0.50
0.75
1.00
1.25
∆P 1% 1.50
0.5%
0%
0 5 10 15 20 25
Number of Louvers
Figure 1.7. Pressure loss is shown to decrease with increasing area ratio for configurations with
straight louvers at φ=30°, without collector.
With respect to the collection efficiency, Figure 1.8 indicates the efficiency for each
particle diameter for 1.25 and 1.50 area ratios. Each of the two area ratios has a different number
of louvers. Figure 1.8 shows both area ratios have better small particle collection with fewer
louvers, and better large particle collection with many louvers. Each louver is longer for a
configuration with few louvers compared to a configuration with many louvers. The relationship
14
between louver length and the particle sizes collected will be described in more detail in the
following section.
An area ratio of 1.50 was chosen for the remainder of the study because it resulted in low
pressure loss and high small particle collection. Compared to the other parameters that were
investigated (louver shape and collector geometry), the area ratio was found to have the largest
effect on pressure loss. Therefore, the remainder of the study focused on particle collection
without attempting to further reduce the pressure losses.
100%
80%
60%
η
40%
AR=1.50 AR=1.25
20% 8 Louvers 7 Louvers
11 Louvers 11 Louvers
19 Louvers 18 Louvers
0%
0 10 20 30 40 50
µ m)
Particle Size (µ
Figure 1.8. Collection efficiency for area ratios of 1.25 and 1.50, straight louvers at φ=30°,
without collector.
Design of the Louvers

The area ratio investigation performed in this study confirmed from previous studies
[11,15] that trajectories for small particles (less than 7µm) were dominated by their reaction to
the local flow field in comparison to trajectories for large particles (greater than 40µm). It was
also seen that large particles were easily separated such that they were directed to the region
where the collector was to be placed. However, small particles, due to their small inertia, were
shown to be difficult to separate. Because of the difficulty in separating small particles, the
geometry of the louver array was designed to increase small particle collection. All louver
geometries investigated considered the area ratio of 1.50. Three different numbers of louvers
including 8, 11, and 19 were investigated where each number resulted in a different louver gap
15
and louver length. Comparisons of simulation results using a baseline configuration of 11
louvers were made between more than 20 unique configurations. Louver geometries with the
best performance were further evaluated in an array containing 8 and 19 louvers.
Collection efficiency was predicted to be dependent on the number of louvers as well as
the louver angle and shape. Pressure loss was not significantly affected by the louver geometry.
Figure 1.8 shows that small particle collection efficiency is greater when a small number of
louvers are used, compared to a large number of louvers. Adversely, small particle collection is
predicted to decrease as the louver angle is increased, as shown in Figure 1.9. In addition, small
particle collection is also predicted to decrease for concave or convex louvers as shown in Figure
1.9.
Recall, Figure 1.2 illustrates the convex and concave louver geometry. Increasing the
louver angle and using concave or convex louvers decreases the collection of small particles
because these changes decrease the length of the louvers. A larger number of louvers in the
configuration also decrease the length of each louver.
Long louvers cause an increase in flow velocity and particle momentum by converging
the flow since the top domain wall remains fixed. The high momentum results in the small
particles being less susceptible to directional changes of the flow and bypassing the louver
passages, as illustrated in Figure 1.10. Large particle collection is shown to decrease with
increased louver lengths. Increasing the large particle momentum beyond the already high value
causes the tendency for large particles to have many deflections from the separator walls and
louvers. As such, the large particles ultimately pass through the last louver passage, as shown in
Figure 1.11.
16
30° 60° Concave
45° Bent Tip Convex
100%
11 Louvers
80%
60%
η
40%
20%
0%
0 10 20 30 40 50
µ m)
Particle Size (µ
Figure 1.9. Collection efficiency for varying louver angle and shape, AR=1.50, without
collector.
Bent tip louvers, also illustrated in Figure 1.2, were shown to increase small particle
collection over straight louvers, as shown in Figure 1.9. This increase in collection occurs
because the louver length of the bent tip louvers is comparable to straight louvers, while
additionally reducing the effective flow area between louvers. The reduction in effective flow
area causes an abrupt change in flow direction around the louvers, allowing the momentum of
the small particles to carry the particles away from the louver passages as shown in Figure 1.12.
This effect has also been found by Poulton et al. [19] where the reduction in flow area was
caused by the separation region behind straight louvers.
17
a)
b)
c)
Figure 1.10. Particle trajectories for 5µm particles, showing an increase of curvature with louver
length for AR=1.50, φ=30° configurations, without collector, with a) 8, b) 11, and c) 19 louvers.
a)
b)
c)
Figure 1.11. Particle trajectories for 40µm particles, showing increasing surface deflections with
increased louver lengths for AR=1.50, φ=30° configurations, without collector, with a) 8, b) 11,
and c) 19 louvers.
18
Figure 1.12. Particle trajectories through an 8 louver, AR=1.50, bent tip φ=30° louver
configuration, without collector, for 5µm (upper) and 40 µm (lower) particle sizes.
Investigations were also carried out by varying the louver geometry along the louver
array by changing individual louvers. Results from representative configurations are shown in
Figure 1.13 with 8 louvers in each configuration. Louver angles are varied along the array by
incrementally changing the louver angle, such as configuration B. Configuration B varies the
louver angle from φ = 60° to 30° and was predicted to perform slightly better than the baseline φ
= 30° configuration (configuration A). Furthermore, some louver configurations combine both
straight and bent tip louvers. It was thought that the success of the bent tip louver would be
additive to the success of the straight louver configurations. Figure 1.13 shows that combining
the bent tip and straight louvers into one configuration (configuration D) does not result in better
performance than either louver configuration alone. Figure 1.13 shows the best performing
louver configuration was found to have straight louvers at varying angle from φ = 30° to 45° to
30° uniformly incremented along the 8 louver array. The best design geometry is shown in
Figure 1.14. Eight louvers were chosen for the best design because, as explained earlier, fewer
louvers in the configuration result in higher small particle collection. Although the best design
(configuration E) shown in Figure 1.13 has low efficiency in collecting large particles, the
collector itself still needed to be modified to improve the large particle collection.
19
100%
8 Louvers A
B
80% C
D
E
60%
η
40%
20%
0%
0 10 20 30 40 50
µm)
Particle Size (µ
configuration A 30°, straight louvers

configuration B 60° - 30°, straight louvers
configuration C 30°, all bent tips
configuration D 60° - 30°, bent last three tips
configuration E 30°- 45°- 30°, straight louvers
Figure 1.13. Collection efficiency is affected by varying louver angle and louver shape along the
louver array.
Figure 1.14. The best performing louver configuration varies the louver angle from φ = 30° to
45° to 30° along the array, resulting in higher small particle collection with 55µm
m particles shown.
Design of the Collector

The remainder of the study focused on the design of the collector using the previously
determined area ratio of 1.50 and tthe
he louver configuration of 8 louvers with varying louver angle
from φ = 30° to 45° to 30° along the array, as shown in Figure 1.14.. The purpose of the
th collector
is to provide a place for particles to remain after they have been separated from the flow. In the
actual gas turbine application, this collector will need to be cleaned periodically to remove the
20
collected particles. The collector design focused on increasing large particle collection with the
goal that the small particle collection should not decrease.
The contour shape of the collector walls and the collector inlet were modified and the
effects are shown in Figure 1.15. Curved collector walls show better large particle collection
than straight walls with no apparent reduction of small particle collection. The straight wall
designs that were investigated were not able to reflect the large particles such that they remained
in the collector.
To improve the large particle collection, it is desirable to have a collector inlet section to
restrict particles from leaving the collector after having been captured. It was found that a lip
added to the collector inlet helped to deflect particles away from the collector inlet and keep
them within the collector region. Predictions also indicate that the lip created an inlet vortex, as
shown in the upper image of Figure 1.16. The louver geometry also helps to establish this vortex
by increasing the flow velocity passing over the last louver tip. If the louver exit velocity is not
large enough, the flow will not impinge on the inlet lip in such a way to develop the vortex as
shown in the lower image of Figure 1.16.
Figure 1.15 indicates the improved collection efficiency when a lip with a vortex at the
inlet to the collector is present. The circulation direction of the vortex is essential to set up a
favorable circulation pattern within the collector. Figure 1.16 compares circulation patterns with
and without a vortex at the inlet to the collector. The favorable circulation pattern set up in the
collector results in particle trajectories that are not directed back towards the collector inlet.
Figure 1.17 shows representative particle paths for 40µm and 5µm particles for the best design
configuration that has the favorable collector circulation. Alternatively, without the inlet vortex,
particles are directed towards the inlet by the unfavorable circulation, thereby exiting the
collector.
21
100%
80%
60%
η
Straight Walls
40% Curved Walls
Without Vortex
20% With Vortex
0%
0 10 20 30 40 50
µm)
Particle Size (µ
Figure 1.15. Key geometry features of the collector are shown to affect collection efficiency for
the configuration: AR=1.50, 8 louvers, straight louvers, φ = 30° to 45° to 30°.
Inlet Vortex
Inlet Lip
U/Uinlet
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Figure 1.16. Collector circulation patterns showing (upper) clockwise rotation with an inlet
vortex, and (lower) counter-clockwise circulation without an inlet vortex
22
Figure 1.17. Particle trajectory plot showing the collecting pattern for 55µm
m (upper) and 40µm
40
(lower) particles in the best separator design.
Conclusions
A methodical design of an inertial particle separator that does not utilize scavenge flow to
aid particle separation has been presented. The separator is intended for placement downstream
of the combustor section of a gas turbine. The successful design removes particles from the
coolant stream and traps the particles in a static collector that can be periodically cleaned. The
separator consisted of an array of louvers followed by a static collector.
The results indicated that the pressure losses across the separator louver array were
primarily dictated by the flow area ratio. An area ratio of 1.50 was found to have minimal (<
1%) impact on the overall engine performance in terms of pressure losses. Predicted results for
an area ratio of 1.50 indicated that particle diameters les
less than 7µm
m were better collected with a
smaller number of louvers than a larger number of louvers. The design of the louver geometry
indicated that small particle collection was increased for louver configurations with long louver
lengths. Straight louvers
rs at varying louver angles were found to have better collection than
curved louvers or bent tip louvers. Predicted results indicated that varying the angle of straight
louvers along the louver array resulted in better collection efficiency compared to a constant
louver angle.
To collect the separated particles, a collector was designed to trap particles by entraining
them in a clockwise fluid rotation. The geometric features that were important for the collector
included curved walls and a lip placed aatt the inlet to the collector. The key flow feature that was
found to best collect the particles was the presence of an inlet vortex, caused by flow impinging
on the collector inlet lip. This vortex at the inlet to the collector was found necessary to set up a
favorable circulation pattern in the collector. The circulation pattern directed particle trajectories
away from the collector inlet so the particles could not escape. The vortex phenomenon was
23
important because it can be utilized in the design of any device that requires particles having a
range of sizes to be separated from the working fluid where a scavenge flow is not possible.
Acknowledgments
The authors would like to thank Pratt & Whitney for funding and sponsoring this research
effort including Joel Wagner, Ryan Levy, and Takao Fukuda.
References
[1] Moroianu, D., Karlsson, A., Fuchs, L., 2004, “LES of the Flow and Particle Ingestion
into an Air Intake of a Jet Engine Running on the Ground,” GT2004-53762.
[2] Hamed, A., Tabakoff, W., 2006, “Erosion and Deposition in Turbomachinery,” J.
Propulsion and Power, 22 (2), pp. 350-360.
[3] Land, C.L., Thole, K.A., Joe, C., 2008, “Considerations of a Double Wall Cooling
Design to Reduce Sand Blockage,” GT2008-50160.
[4] Walsh, W.S., Thole, K.A., Joe, C., “Effects of Sand Ingestion on the Blockage of Film-
Cooling Holes,” International Gas Turbine and Aeroengine Congress and Exposition,
Barcelona, Spain, ASME Paper GT2006-90067.
[5] Cardwell, N.D., Thole, K.A., Burd, S.W., 2008, “Investigation of Sand Blocking Within
Impingement and Film-Cooling Holes,” GT2008-51351.
[6] Schneider, O., Dohmen, H.J., Reichert, A.W., 2002, “Experimental Analysis of Dust
Separation in the Internal Cooling Air System of Gas Turbines,” GT2002-30240.
[7] Schneider, O., Dohmen, H.J., Benra, F.-K., Brillert, D., 2003, “Investigations of Dust
[8] Schneider, O., Dohmen, H.J., Benra, F.-K., Brillert, D., 2004, “Dust Separation in a Gas
Turbine Pre-Swirl Cooling Air System, a Parameter Variation,” GT2004-53048.
24
[9] Schneider, O., Benra, F.-K., Dohmen, H.J., Jarzombek, K., 2005, “A Contribution to the
Abrassive Effect of Particles in a Gas Turbine Pre-Swirl Cooling Air System,” GT2005-
68188.
[10] Syred, C., Griffiths, A., Syred, N., 2004, “Gas Turbine combustor with Integrated Ash
Removal for Fine Particulates,” GT2004-53270.
[11] Breitman, D.S., Duek, E.G., Habashi, W.G., 1985, “Analysis of a Split-Flow Inertial
Particle Separator by Fineie Elements,” J. Aircraft, 22 (2), pp 135-140.
[12] Vittal, B.V.R., Tipton, D.L., Bennett, W.A., 1985, “Development of an Advanced
Vaneless Inlet Particle Separator for Helicopter Engines,” AIAA/SAE/ASME/ASEE 21st
Joint Propulsion Conference.
[13] Shieh, C.F., Delaney, R.A., 1984, “Analysis of the Flow Field in an Engine Inlet Particle
Separator,” ASME Fluids Engineering Division (Publication), 14, pp 23-28.
[14] Zedan, M., Mostafa, A., Hartman, P., 1992, “Viscous Flow Analysis of Advanced
Particle Separators,” J. Propulsion and Power, 8 (4), pp 843-848.
[15] Ghenaiet, A., Tan, S.C., 2004, “Numerical Study of an Inlet Particle Separator,” GT2004-
54168.
[16] Zverev, N.I., 1946, “Shutter-Type Dust Collector of Small Dimensions,” Engineer’s
Digest, 7 (11), pp. 353-355.
[17] Smith, Jr., J.L., Goglia, M.J., 1956, “Mechanism of Separation in Louver-Type Dust
Separator,” ASME Transactions, 78 (2), pp. 389-399.
25
[18] Gee, D.E., Cole, B.N., 1969, “A Study of the Performance of Inertia Air Filters,” Inst.
Mech. Engineers - Symposium on Fluid Mechanics and Measurements in Two-Phase
Systems, University of Leeds, pp. 167-176.
[19] Poulton, P., Cole, B.N., 1981, “An Experimental and Numerical Investigation of Louvred
Inertia Air Filter Performance,” Conference on Gas Born Particles, Institution of
Mechanical Engineers, pp. 161-170.
[20] Jones, G.J., Mobbs, F.R., Cole, B.N., 1971, “Development of a Theoretical Model for an
Inertial Filter,” Pneumotransport 1 – 1st Int’l. Conf. on Pneumatic Transport of Solids in
Pipes, Paper B1.
[21] ANSYS, Inc., Version 6.3.26, 2008 (ANSYS Inc.: New Hampshire)
[22] Fluent User Guide, ANSYS, Inc., Version 6.3, 2008 (ANSYS, Inc.: New Hampshire).
[23] Particle size distribution provided by Pratt & Whitney, 2007.
[24] Schetz, J., Fuhs, A., 1996, Handbook of Fluid Dynamics and Fluid Machinery, Vol 1,
John Wiley & Sons, Inc., New York, pp 905-906, Chap. 14.
26
Paper 2: Performance Measurements of a Unique Louver
Particle Separator for Gas Turbine Engines
To be submitted to the 2010 IGTI Conference* and the Journal of Turbomachinery*
Abstract
Solid particles, such as sand, ingested into gas turbine engines reduce coolant flow in the
turbine by blocking cooling channels in the secondary flow path. Suitable methods to remove
solid particles in high flow rate applications with minimal pressure losses include inertial particle
separators. In this paper, an inertial separator was studied that used an array of louvers followed
by a static collector. A unique test facility was designed to measure the pressure loss and sand
collection efficiency of two louver configurations. Measurements were taken over a range of
Reynolds numbers and sand sizes. To complement the measurements, two- and three-
dimensional computational results were presented for comparison. Computational predictions of
sand collection efficiency agreed within 10% of experimental measurements over the range of
Reynolds numbers for a sand size range 0-200µm. Collection efficiency was over-predicted for
discrete sand sizes within the 0-200µm size range. Pressure loss predictions agreed with
measurements at high Reynolds numbers. Collection efficiency values were measured to be as
high as 35%, and pressure loss measurements were equivalent to less than 1% pressure loss in an
engine application.
Nomenclature
AR area ratio AR = δ(# Louvers)/H

Cp pressure loss coefficient Cp = (Pinlet-Pexit)static / (Pinlet)dyn
Ds sand particle diameter
H height of test section and test facility channel
m mass of sand
ML mass loading, ML= (sand mass flow rate) / (fluid mass flow rate)
Re test section inlet Reynolds number Re = ρHUin/µ
Uin test section inlet flow velocity
Greek:
δ louver gap
η collection efficiency, η = mcol / min
*Co-authors:
Dr. Michael Barringer, Dr. Karen Thole – Mechanical and Nuclear Engineering Department, The Pennsylvania State
University
Eric Grover, Joe Barker – Pratt & Whitney, United Technologies Corporation
27
µ fluid dynamic viscosity
ρ fluid density
φ louver angle
Subscripts:
0 sand initially placed in sand feeder
col sand collected in the collector after a test
chan sand deposited in the upstream channel
feed sand remaining in the feeder
in sand entering the test section, min = m0-mfeed-mman-mchan
man sand deposited in the mixing manifold
Introduction
Aircraft gas turbines ingest solid particles during takeoff, landing, and while flying in
dusty environments. The presence of solid particles in the engine flow path has been shown to
reduce turbine power and efficiency [1]. Particle ingestion also reduces performance over the
life of the engine through erosion, deposition, and blockage of cooling channels. Ingested
particles pass through both primary and secondary flow paths in the turbine. In the primary flow
path particles erode and deposit on airfoil surfaces resulting in rough airfoil surfaces that reduce
engine efficiency and increase heat transfer to the airfoil. Deposition on turbine airfoils results
in blocked film cooling holes and can cause engine surge when particulates build up on first
stage turbine vanes [2,3]. Ingested particles that pass through the secondary flow path of the
engine will reduce coolant flow by blocking cooling channels [4,5]. When the coolant flow is
reduced, the temperature of turbine components increases thereby reducing component life.
This paper presents performance measurements of a louver separator placed within the
secondary flow path to remove solid particles, such as sand from secondary coolant flow.
Performance measurements over a range of Reynolds numbers and sand sizes include pressure
loss and particle collection efficiency. A unique method is developed to quantify particle
collection efficiency in a laboratory setting. To compliment the data reported in this paper, two-
and three-dimensional computational results are presented for direct comparison. This paper
presents a review of relevant studies, methodology, measurement uncertainty and repeatability,
and a discussion of results.
28
Review of Relevant Literature
Methods to remove particles from gas turbine flows have included inertial particle
separators because of their favorable characteristics to handle high flow rates at low pressure
losses. The operating principle of inertial separators is to cause the particle-laden flow to
abruptly change direction, whereby the particles do not react to the change in flow direction due
to their larger inertia as compared to the flow. Typically, inertial particle separators in literature
are placed at engine inlets exhausting the separated particles through a scavenge duct. Such a
separator is one that uses an array of louvers to change the flow direction.
The pressure loss and particle separation efficiency of louver separators are affected by
Reynolds number, particle size, and louver geometry. Previous studies have reported that
particle separation efficiency increased with inlet and scavenge Reynolds numbers [6-9].
Poulton et al. [9] found that separation efficiency had an upper limit and leveled off at high inlet
Reynolds numbers. Additionally, high Reynolds numbers were reported by Smith and Goglia
[7] and Poulton et al. [9] to increase pressure losses resulting from the separator. Increased
separation efficiency was obtained for larger particle sizes as shown by Gee and Cole [8] and
Poulton et al. [9]. Particle mass loading (ratio of particle mass flow to air mass flow) values less
than 0.05 were found to not affect separation efficiency [7,8].
The effect of the louver geometry on separation efficiency was studied with regard to
louver angle, length, and spacing. Studies by Zverev [6] and Poulton et al. [9] reported that
particle separation efficiency was affected by louver angle by as much as 10%. Zverev [6] found
that increasing the louver angle decreased the pressure loss of the separator. Louver thickness
was reported by Poulton et al. [9] to affect separation efficiency by no more than 10%. Louver
length was reported by Gee and Cole [8] to increase separator pressure loss by as much as 40%
when the louver length was doubled. Increased louver spacing was reported by Zverev [6] and
Poulton et al. [9] to reduce separation efficiency.
Louver separator designs using scavenge flow are the most common geometries reported
in the literature. A scavenge flow design, however, is not practical to remove solid particles
from the secondary flow path within gas turbines because of flow path considerations. Others
have investigated unique inertial separation methods that can be located within gas turbine flow
paths.
29
Schneider et al. [10-13] did extensive experimental and computational investigations into
the effectiveness of using a pre-swirl cavity to divert particles from internal cooling flow to the
main flow path. Syred et al. [14] computationally investigated a combustor design that integrated
a particle removal system that removes both large and small particles. The can-type combustor
removed large particles by acting as a cyclone separator. Small particles were separated by a
vortex collector pocket located near the combustor exit. A louver separator using a static
collection bin (collector) instead of a scavenge flow duct was studied by Musgrove et al. [15].
The louver separator trapped the separated particles in the collector and was intended for
placement in the combustor-bypass flow. Musgrove et al. [15] computationally investigated the
performance effects of area ratio (AR), louver geometry, and collector geometry. The best
design consisted of AR = 1.50 and eight, straight louvers at varying angle (φ = 30° to 45° to 30°)
followed by a contoured collector that produced a favorable circulation. Musgrove et al. [15]
reported that the favorable circulation was necessary for particle collection by directing trapped
particles away from the collector inlet. The presence of a vortex at the collector inlet (inlet
vortex) was found to set up the favorable circulation in the collector.
This study presents measurements of two louver separator designs to complement the
computational study presented by Musgrove et al. [15]. Both configurations have the same area
ratio (AR = 1.50), number of louvers (eight), and identical collectors. The louver configurations
include one that increments each louver angle along the array (φ = 30° to 45° to 30°) while the
second configuration maintains constant louver angle (φ = 45°). Performance measurements
consist of pressure loss across the separator and sand collection efficiency. Good performance of
the separator was considered to be at low pressure loss with high sand collection efficiency. In
addition to experimental measurements, results from both two- and three-dimensional
computational models are compared.
Experimental Facility and Methodology

The performance of a louver separator was evaluated from measured sand collection
efficiency (η) and pressure loss coefficient (Cp). To evaluate the separator performance, an open
loop test facility was constructed to make the measurements. The collection efficiency was
calculated based on the mass of sand captured in the collector relative to the mass of sand
passing through the louvers. The pressure loss coefficient was calculated from the static pressure
30
loss across the separator relative to the dynamic pressure at the inlet to the test section enclosing
the separator. In this section of the paper, the test facility, sand characterization, and louver
separator configurations are explained in detail.
Test Facility
Testing was conducted in an open loop test facility that circulated in a clockwise
direction, as indicated by the arrows in Figure 2.1. Flow through the test facility was driven by a
0.66 m3/s (1400 CFM) blower powered by an 11 kW variable speed electric motor. Downstream
of the blower, the flow passed through a pipe containing a venturi meter that was isolated
upstream and downstream by 20 pipe diameters. Downstream of the venturi meter, the flow
from the lower section of the test facility expanded through a transition into a rectangular
channel. At the inlet to the rectangular channel, a honeycomb laminar flow element and splash
plate were placed to produce spanwise uniform flow at the test section inlet. Additionally, the
flow was tripped downstream of the channel inlet with sand paper placed along the top and
bottom of the channel. Further downstream sand was injected across the span of the channel.
The sand injection system was made up of a sand feeder, a mixing manifold, and an
injection slot. The sand feeder, modified from commercial use as a plastics feeder, utilized a
helical auger screw to convey sand from the feeder hopper into the mixing manifold at a constant
flow rate. The flow rate was controlled by a variable speed auger screw, where the maximum
flow rate was determined by the size and number of threads of the auger screw. The maximum
feed rate for the auger screw used in this study was 8 g/sec (ML = 0.1 at Re = 8,000). In the
mixing manifold, sand from the feeder was dispersed using compressed air injected through three
ports. The objective of the compressed air was to both mix the sand and maintain slightly higher
pressure in the manifold relative to the channel. The sand-air mixture in the manifold passed
into the channel through a slot extending across the channel width. The slot was angled 45°
upstream relative to the flow whereby the sand-air mixture was injected. Injecting the sand in
the upstream direction ensured that sand particles were well mixed with the channel flow.
Downstream of the sand injection slot the louver separator was enclosed in a transparent
test section, as shown in Figure 2.1. The test section was constructed of transparent material to
allow visual observation of sand trajectories through the louver array and in the collector. After
passing through the separator, air flow and any uncollected sand passed through a downstream
rectangular channel followed by a pipe and finally into a filter box. Any remaining sand in the
31
flow was removed by three air filters placed in series in the filter box. As sand collected in the
filters during each test, the increasing pressure loss across the filters reduced the test facility flow
rate. The flow rate through the test facility, however, was decreased by no more than 2%.
The pressure loss coefficient (Cp) was calculated from the dynamic pressure at the test
section inlet, and the static pressure loss across the test section. The dynamic pressure was
calculated based on the measured mass flow. Static pressures were measured at the inlet and exit
of the test section. Spanwise uniformity of the flow through the louver separator was checked by
normalizing the static pressure loss measurements. The pressure measurement at each tap was
normalized from the difference of the tap pressure and the span wise average pressure divided by
the inlet dynamic pressure. Flow uniformity was achieved when normalized values deviated by
less than ±5%.
After each test, the test facility was disassembled and all deposited sand mass was
removed and weighed. To give an accurate account of the collection efficiency (η), a
methodology was developed that allowed a true accounting of all of the sand placed in the
feeder. This accounting was needed to include any sand that remained in the feeder, deposited
on the channel walls, and was collected in the downstream filters. For most tests, between 90-
95% of the injected sand was accounted for, as will be shown in the results. To calculate the
collection efficiency, the sand trapped in the collector (mcol) was divided by the sand that passed
through the louver test section inlet (min). The sand that was considered to be collected was that
amount accumulated in the collector. The sand that passed through the louver separator was
considered to be the sand placed in the feeder minus the sand that deposited on any upstream
walls (mchan) or was left in the manifold (mman). The sand that collected in the filters was
considered to be the amount that was not successfully collected by the louver separator. The
filters were weighed before and after each test to determine the amount of sand that the collector
did not capture. The largest amount of sand recovered from the test facility was in the filters and
feeder. After the deposited sand was removed, all surfaces were wiped clean to remove sand
residue. It is important to point out that prior to each test, the injected sand was heated at 100°C
for at least three hours to remove any moisture.
32
Sand
feeder Filter box
Sand manifold
(a) Venturi meter
Splash plate
Sand injection
Honeycomb slot
Trip strips
(b) H
8H 9H 10H 18H
36H
10H
(c)
Flow
Figure 2.1. The (a,b) open loop test facility is shown with the sand injection system and (c)
transparent test section.
Sand Characterization
The coarse sand used in this study was ISO 12103-1 A4 certified sand with a bulk density
of 1201 kg/m3, purchased from Powder Technology Inc. Five size ranges of the test sand were
studied to determine the effect of sand size on collection efficiency: 0-200µm (entire sample
size), 0-20µm, 20-40µm, 38-63µm, and 63-200µm. Size ranges purchased from Powder
Technology Inc. included 0-200µm, 0-20µm, and 20-40µm with size distributions shown in
Figure 2.2 [16]. Sand size ranges of 38-63µm and 63-200µm were obtained by filtering the 0-
200 µm size distribution through sieves. The size ranges of the sieved sand were obtained by
sieving the sand through mesh screens with screen opening sizes corresponding to the maximum
or minimum desired sand size. For example, the 63-200µm size range was obtained by filtering
the 0-200µm range through a sieve with mesh screen opening sizes of 63µm.
33
100 20
0 - 20 µm
µ
20 - 40 µ m
80
15 0 - 200 µ m
60
Under Volume 10
Sand [%]
Diameter 40
[%]
0 - 20 µm 5
20 20 - 40 µm
0 - 200 µm
0 0
0 50 100 150 200 2 3 6 9 12 18 28 60 200
µm)
Sand Diameter, Ds (µ µm)
Sand Diameter, Ds (µ
Figure 2.2. Sand size distribution by percent volume of ISO 12103-1 A4 test sand [16].
Louver Separator Geometry

The two louver separators studied in this paper were similar to configurations presented
in a previous computational study [15]. Each louver separator was made up of an array of
straight louvers followed by a collector. One configuration had a varying louver angle and the
second configuration had constant louver angle, as shown in Figure 2.3. Details of each
configuration are listed in Table 2.1. In the previous computational study [15], the separator
geometries did not include part thickness or structural supports. With the inclusion of part
thickness for this study the separator geometry needed to be slightly modified to maintain the
design area ratio of 1.50. Modifications included the adjustment of louver spacing, louver
anchor length, and the axial position of the collector inlet. Structural supports were added to the
design to resist the flow drag and maintain correct louver spacing and angle. Three louver struts
were placed at intervals across the span, and seven supports were placed at spanwise intervals
between the last louver and the collector. Guide tracks were also included on either side of the
louver separators to hold them in place in the test section. The guide tracks fit flush with the
endwall surfaces and did not protrude into the test section flow.
34
Wedge
Baffle
Figure 2.3. Both the variable louver (top) and constant louver (bottom) configurations
incorporated structural supports and a wedge and baffle at the collector inlet.
Table 2.1. Area Ratio and Louver Geometry
Variable louver Constant louver
Area Ratio (AR) 1.50 1.50
Number of Louvers 8 8
φ)
Louver angle (φ 30° to 45° to 30° 45°
Testing the design specified through computational studies [15] resulted in negligible
collection due to the observed collector circulation directing sand towards the collector inlet
(unfavorable circulation). Therefore, the collector inlet was modified relative to the previous
computational study to improve sand collection efficiency [15]. Modifications that were made to
the collector inlet were guided by the experimentally observed sand circulation in the collector.
Table 2.2 illustrates different collector inlet geometries that were tested, and their resulting
collection efficiency. A baffle, shown in Table 2.2, slightly improved collection efficiency by
isolating a region of the collector from the unfavorable circulation. A second modification
placed a wedge at the collector inlet to converge the collector inlet flow against the top wall of
the test section. The wedge produced a favorable circulation in the collector that directed sand
away from the collector inlet. Use of the wedge or baffle modifications independently did not
significantly improve collection efficiency. Collection efficiency was improved, however, when
the wedge and baffle were used together. The favorable circulation set up by the wedge directed
35
sand away from the collector inlet and into the isolated region created by the baffle. There was
no notable increase in pressure loss as a result of the wedge and baffle.
Results reported in this paper are for both of the louver configurations shown in Figure
2.3 with the inclusion of the wedge and baffle modification. The performance of each separator
configuration is expressed in terms of collection efficiency (η) and pressure loss coefficient (Cp).
Table 2.2. Effect of Collector Modifications
No modification η = 0%
Baffle η = 3%
Wedge η = 0%
Wedge and baffle η = 20%
In addition to experimental measurements, two-dimensional and three-dimensional
computational models were simulated over a range of Reynolds numbers and sand sizes. The
computational study was motivated by the difference between the measurements and two-
dimensional predictions [15] of collection efficiencies of both separator configurations without
the wedge and baffle modification as well as part thickness. All computational models were
solved using FLUENT 6.3.26 [17], similar to the methodology followed Musgrove et al. [15] in
solving the flow field and particle trajectories.
Each simulation was matched to the experimental flow conditions and geometry
including the wedge and baffle modification. The length of the computational domain matched
the experiment from the sand injection slot to the exit of the downstream channel. Both two-
dimensional and three-dimensional models used boundary conditions of inlet velocity and exit
outflow with no-slip surfaces for all remaining surfaces. The three-dimensional model used a
symmetry boundary condition at a spanwise depth of 5H. Two-dimensional and three-
dimensional grid sizes of 6 x 104 and 3 x 106 cells were required for second order accurate grid
independent solutions that met a convergence criterion of 10-6. Solutions used steady RANS
equations and the k-ω turbulence model. Both computational models utilized a boundary layer
mesh along the top and bottom walls of the test section with y+ = 30.
36
Pressure loss coefficients (Cp) were calculated from integrating mass averaged total
pressure and area averaged dynamic pressure at locations consistent with the inlet and exit static
pressure taps in the test facility. Similar to Musgrove et al. [15], the collection efficiency was
calculated from the number of sand particles injected into the domain relative to those that
escaped the domain. Sand was injected along the channel height at an upstream location
consistent with the injection slot in the test facility. Sand sizes 1-200µm with bulk density 1201
kg/m3 were injected. To obtain collection efficiency values independent of the number of
particles injected, 1,000 particles were injected across the channel height. For the three-
dimensional simulation, 1,000 particles were injected at spanwise intervals of 0.125H for each
sand size. Because the computational predictions used numbers of sand particles instead of sand
mass to calculate collection efficiency, the computational results were converted to a mass
weighted equivalent. A mass weighted value that could be directly compared to experimental
data was obtained by integrating the predicted collection efficiency values with respect to sand
size and normalizing by the integrated size range.
Measurement Uncertainty and Repeatability

The uncertainty analysis for collection efficiency and pressure loss coefficient followed
the partial derivative method described by Moffat [18]. Overall uncertainty was calculated from
the root sum square of the precision and bias errors. The precision error was calculated from the
standard deviation of five sets of measurements, where each set consisted of three tests. Each
test measured pressure loss coefficient and collection efficiency for similar Reynolds number,
sand amount, and mass loading.
Precision uncertainty values of ±0.078 for Cp and ±0.026 for η were calculated using a
95% confidence interval on Student’s t-distribution. The calculated bias error due to the
measurement devices had maximum values of ±2.14 for Cp and ±0.02 for η. The bias and
precision errors were combined to determine the overall uncertainty of each calculation. The
constant louver configuration had an overall uncertainty of ±11% for Cp = 20.0 and ±3% for η =
35% at Re = 8,000. Overall uncertainty for the same configuration at Re = 46,400 was ±3% for
Cp = 12.1 and ±3% for η = 8%. Because the overall uncertainty values obtained in this manner
were acceptable values, the measurements reported in this paper were an average of three tests.
37
Experimental repeatability was confirmed for each test by recording the amount and
location of sand deposited in the test facility. The mass of deposited sand was dependent on the
amount of sand placed into the feeder (m0) and the sand mass loading (ML). Therefore, m0 and
ML were varied to determine the effect on collection efficiency, and to verify the repeatability of
the sand retrieval method. Sand amounts of 60g, 180g, and 240g were tested at constant ML =
0.002. Mass loading values of 0.0002, 0.002, and 0.02 were tested for constant m0 = 60g. Mass
loading was limited to values much less than one so that particle-particle interactions and particle
effects on the flow field could be ignored [19], which is representative of engine ingestion.
Reynolds number (Re = 25,000), sand size distribution (0 < Ds < 200µm), and separator
configuration (variable louver) were held constant for each sand amount and mass loading test.
Varying the mass loading and sand amount did not affect the collection efficiency of the variable
louver separator.
The repeatability of each sand test was confirmed by comparing the deposited sand
amounts among each set of three tests, as shown in Figure 2.4. Sand that was removed upstream
of the test section included sand that remained in the feeder and sand adhering to the manifold
and upstream channel walls. Deposited sand in the test section was adhered to the louvers and
trapped in the collector. Downstream of the test section sand adhered to the downstream channel
walls and collected in the filters. Sand deposition in the test facility is shown in Figure 2.4 as a
percentage of sand amount (m0) for varying ML and m0. Note that the collected sand in Figure
2.4 is less than the measured collection efficiency because collection efficiency was calculated
from the amount of sand entering the test section. Other sections of the test facility that are not
presented in Figure 2.4 were inspected after each test to confirm that sand was not deposited.
The similar amounts of sand retrieved from the sections of the test facility for each set of tests
indicated repeatable sand flow. For all tests, the missing sand amount was consistently less than
10% of the initial sand amount (m0). Sand deposition in the manifold and upstream channel was
higher for ML = 0.02 than lower mass loadings because the flow rate of the compressed air
injected into the manifold was not enough to disperse the high flow rate of sand from the feeder.
The high amount of deposited sand in the manifold and upstream channel reduced the amount of
sand that was collected by the filters, but did not affect the amount of sand captured in the
collector. The percentage of sand in the filter and collector, however, increased with m0 because
38
more sand passed through the test facility with an unchanged amount of sand remaining in the
feeder.
Missing sand Louver Surfaces

Feeder Filters
Manifold Downstream Channel
Upstream channel Collector
100%
80%
60%
40%
20%
0%
0.0002 0.002 0.02 60 180 240
Mass Loading, ML Sand Amount, m0 (g)
Figure 2.4. Test repeatability was shown by similar sand distribution through the test facility for
varying mass loading and sand amount: variable louver angle, Re = 25,000, 0 < Ds < 200µm.
Figure 2.5 shows that collection efficiency was independent of sand amount and mass
loading for sand sizes 0-200µm. Changing the mass loading and sand amount was shown in
Figure 2.4 to affect deposited sand amounts throughout the test facility. Therefore, collection
efficiency was independent of sand deposition in the test facility. The variable louver
configuration resulted in measured collection efficiency of η = 32% at Re = 25,000 for all tested
values of m0 and ML. Repeatable tests of varying mass loading (ML) and sand amount (m0) did
not affect sand collection efficiency of sand sizes 0-200µm for the variable louver separator at
Re = 25,000. Therefore, all tests used sand amount and mass loading values of m0 = 60g and
ML = 0.002.
39
Mass Loading, ML
0.0001 0.001 0.01 0.1
50%
40%
30%
η
20% Sand Amount
Mass Loading
10%
0%
0 50 100 150 200 250
Sand Amount, m0 (g)
Figure 2.5. Collection efficiency was independent of mass loading and sand amount for the
variable louver configuration at Re = 25,000 for sand sizes 0 < Ds < 200µm.
As was stated, each louver separator was tested over a range of Reynolds numbers and
sand sizes at a constant mass loading (ML = 0.002) and sand amount (m0 = 60g). Tests over the
range of Reynolds numbers injected the sand size distribution with size range 0-200µm. Effects
of sand diameter were tested at constant Reynolds number (Re = 46,400) by injecting discrete
size intervals. In addition to pressure loss and collection efficiency results, sand distribution
through the test facility and images of sand accumulation in the collector are presented. Results
from both the two- and three-dimensional models are presented along with measurements for
direct comparison.
Pressure Loss Results

Measured and predicted pressure loss coefficients (Cp) for both separator configurations
are shown in Figure 2.6 for a range of Reynolds numbers. The similar Cp values for both
configurations indicated that louver angle did not significantly affect pressure loss. These
measured pressure loss coefficients translate to losses that can occur in the engine. The
maximum measured pressure loss coefficient for the constant louver configuration (Cp = 20) was
40
equivalent to a 0.7% pressure loss in the engine. The minimum measured pressure loss
coefficient (Cp = 12.1) was equivalent to a 0.4% loss.
Computational predictions of Cp agreed with data for Reynolds numbers greater than
25,000; however, pressure loss coefficients were under-predicted at lower Reynolds numbers.
The disagreement between measurements and predictions at low Reynolds numbers was most
likely because the predictions did not show that the flow around the louvers changed with
Reynolds number. Computational simulations predicted similar separation regions on the
backside of the louvers for all Reynolds numbers. It was expected, however, that the measured
pressure loss increased for low Reynolds numbers because the size of the separation regions
increased. Furthermore, little difference was seen between the two- and three-dimensional
pressure loss predictions because spanwise flow effects were not apparent.
25
20
15
C
p
10
Experiment Experiment
5 2D CFD 2D CFD
3D CFD 3D CFD
0
0 1 2 3 4 5
4
Re [10 ]
Figure 2.6. Pressure loss across both separator configurations increased for decreasing Reynolds
numbers.
Reynolds Number Effects on Collection Efficiency

Contrary to similar pressure losses between both louver configurations, collection
efficiency was significantly affected by the two different louver arrays. The results reported in
Figure 2.7 include experimental and computational results for injecting the 0-200µm sand for
varying Reynolds numbers.
The variable louver configuration results in Figure 2.7 were found experimentally and
computationally to collect more sand than the constant louver configuration at all Reynolds
41
numbers. As Reynolds number decreased, the two louver configurations approached similar
collection efficiencies. At low Reynolds numbers, the variable louver configuration had a
maximum collection efficiency of 32%. The constant louver configuration had increasing
collection efficiency with decreasing Reynolds number to a maximum measured value of 35% at
Re = 8,000. Reynolds numbers less than 8,000 were not tested because sand fell out of the
channel flow. Collection efficiencies of both separator configurations were measured to increase
for decreasing Reynolds numbers because the momentum of the sand particles decreased with
Reynolds number. As the momentum of the sand particles was reduced the particles were more
susceptible to fall out of the collector circulation and remain in the collector.
50%
40% 2D CFD 2D CFD
3D CFD 3D CFD
30%
η
20%
10%
0%
0 1 2 3 4 5
4
Re [10 ]
Figure 2.7. Collection efficiency for both configurations increased as Reynolds number
decreased for 0 < Ds < 200µm.
To compliment the measured collection efficiency, photographs were taken of sand

accumulation in the collector for each Reynolds number tested. Sand collection was observed to
be repeatable and uniformly symmetric about the midspan of the test section. Figure 2.8 shows
half span images of sand accumulation at the minimum and maximum Reynolds numbers for
both louver designs. Comparison of the images confirmed more sand was collected at lower,
rather than higher, Reynolds numbers. For low Reynolds numbers sand accumulation was
concentrated on the bottom wall of the collector out-from-under the baffle. For higher Reynolds
numbers sand accumulated under the baffle and at the ends of the collector. The change in
accumulation location is explained by visual observations of sand trajectories in the collector.
42
The circulation size in the collector was observed to increase with Reynolds number, as shown in
Figure 2.9. The large circulation directed particles under the baffle resulting in more sand
accumulation under the baffle at high Reynolds numbers. The small circulation at low Reynolds
numbers resulted in sand depositing in the collector out-from-under the baffle.
To confirm the repeatability of each sand test the deposited sand mass throughout the test
facility was removed and measured, as shown in Figure 2.10 and Figure 2.11 as a percentage of
the total sand injected. Most of the sand retrieved from the test facility was located in the filters
and the feeder. The percentage of sand depositing upstream of the test section remained constant
over the range of Reynolds numbers for both louver configurations. The percentage of sand
depositing in the filters was dependent on the amount of sand captured by the collector. The
similar sand deposition between the tests for each Reynolds number indicated repeatable sand
flow through the test facility. For most tests, more than 90% of the injected sand amount was
retrieved.
(Variable louver) (Constant louver )
Re=46,400 Re=46,400
Re=15,000 Re=8,000
Figure 2.8. Sand was uniformly collected across the collector span in increasing amounts with
increasing Reynolds number for 0 < Ds < 200µm.
43
(a) (b)
M~0.001
Re=46,400 Re=46,400
Re=8,000 Re=15,000
Figure 2.9. Collector circulation size decreased was observed to decrease with Reynolds number
for the (a) constant louver and (b) variable louver configurations for sizes 0 < Ds < 200µm.

Feeder Filters
100%
80%
60%
40%
20%
0%
1.5 2.5 3.5 4.6
4
Re [10 ]
Figure 2.10. Distribution of sand through the test facility for the variable louver configuration
when varying Reynolds number for 0 < Ds < 200µm.
44
Feeder Filters
100%
80%
60%
40%
20%
0%
0.8 1.5 2.5 3.5 4.6
4
Re [10 ]
Figure 2.11. Distribution of sand through the test facility for the constant louver angle
configuration when varying Reynolds number for 0 < Ds < 200µm.
Both two- and three-dimensional computational models predicted values of collection
efficiency within 10% of measured data. The predicted trends, however, disagreed with the
experimental trends. The two-dimensional computational models predicted that both louver
configurations had the same trend. In comparison, measurements indicated that the trend was
affected by the louver configuration. Measurements showed that the variable louver
configuration leveled off at low Reynolds numbers and the constant louver configuration
continued to increase for low Reynolds numbers. Computational models predicted that the
collection efficiency of both louver configurations increased for decreasing Reynolds numbers.
It is also of importance that the two- and three-dimensional predictions for the variable
louver configuration were nearly the same, whereas the two- and three-dimensional predictions
for the constant louver configuration were different. The reason the two-dimensional model for
the constant louver configuration predicted higher collection efficiency than the three-
dimensional model resulted from spanwise flow effects in the three-dimensional predictions.
45
The endwalls created a spanwise flow pattern that disrupted the favorable collector circulation,
thereby reducing the effectiveness of the collector. The favorable circulation was disrupted for
the constant louver configuration and not the variable louver configuration because the constant
louver configuration had a weaker collector circulation than the variable louver configuration.
The presence of the favorable circulation has been shown in a previous computational study to
occur with an inlet vortex [15]. The strength of the collector circulation and inlet vortex has
been shown to significantly affect the collection efficiency and to be dependent on louver
configuration [15]. Computational predictions of both louver configurations converged for low
Reynolds numbers because at low Reynolds numbers the effect of the louver configuration on
the inlet vortex was reduced.
Sand Size Effects on Collection Efficiency

In addition to Reynolds number effects, the influence of sand size on the collection
efficiency was also tested. Four discrete sand size intervals were independently injected, sizes:
0-20µm, 20-40µm, 38-63µm, and 63-200µm. For each sand size interval the Reynolds number,
mass loading, and injected sand amount were held constant (Re = 46,400, ML = 0.002, m0 =
60g).
Measured results indicated that the effect of sand size on collection efficiency was dependent
on the louver configuration, as shown in Figure 2.12. Specifically, the collection efficiency of
the variable louver configuration was dependent on sand size, while the constant louver
configuration was not. The variable louver configuration was best at collecting sand sizes less
than 20µm in diameter with reduced collection efficiencies for sizes larger than 20µm. Injecting
the entire sand size distribution (0-200µm) resulted in as much as 20% higher collection
efficiency for the variable louver collection efficiency compared to injecting only discrete sand
sizes. The collection efficiency of the constant louver configuration, however, was independent
of the injected sand sizes. Injecting the entire sand size range (0-200µm) resulted in the same
collection efficiency as injecting any of the discrete sand size intervals. Combining the
collection efficiency for each discrete sand size was not equivalent to injecting all sizes together;
therefore, it was likely particles agglomerated in the collector.
46
50%
40% 2D CFD 2D CFD
3D CFD 3D CFD
30%
η
20%
10%
0%
0-20 20-40 38-63 63-200 0-200
µm)
Sand Size, Ds (µ
Figure 2.12. Collection efficiency was dependent on particle size and louver configuration for
Re = 46,400.
Sand accumulation was observed to be dependent on particle size. Accumulation in the

collector was repeatable and symmetric about midspan with the half span shown in Figure 2.13.
Accumulation patterns were similar for both louver configurations for varying sand sizes. Sand
sizes less than 20µm in diameter were evenly distributed across the span of the collector for both
separator configurations where most accumulation occurred on the bottom of the collector out-
from-under the baffle. Sizes larger than 20µm accumulated in small amounts near the ends of
the collector. The increased sand accumulation for small sand sizes can be explained with
observations of sand trajectories through the separator. Sand trajectories are shown in Figure
2.14 and Figure 2.15 for minimum and maximum sand size intervals. For small sand sizes, sand
trajectories appeared to follow flow streamlines. Increasing the sand size resulted in increased
deflections from the louver and test section surfaces. In the collector, the circulation size and
rotation of sand appeared to be independent of sand size. The mechanism responsible for sand
trajectories in the collector was probably different for large sand sizes relative to small sand
sizes. Large sand sizes in the collector were likely conveyed by their deflection from the
collector walls while small sizes were likely convected by flow circulation in the collector.
47
To confirm test repeatability, the distribution of deposited sand throughout the test
facility was recorded for the different sand size intervals that were injected. The deposited sand
amounts that were retrieved after each test are shown in Figure 2.16 and Figure 2.17 as a
percentage of injected sand amount (m0). For the small size range (0-20µm), most of the
deposited sand was due to adherence to the test facility walls. In contrast, large sand sizes did
not deposit upstream of the test section, and were more easily caught in the filters than small
sand sizes. Therefore, tests with large sand sizes resulted in less sand amount missing.
(Variable louver) (Constant louver)
µm
0-20µ µm
0-20µ
µm
20-40µ µm
20-40µ
µm
38-63µ µm
38-63µ
µm
63-200µ µm
63-200µ
Figure 2.13. Particle sizes Ds < 20µm were distributed evenly across the collector, however,
sizes Ds > 20µm deposited only at the collector ends for Re = 46,400.
µm
0-20µ
µm
63-200µ
Figure 2.14. Particle trajectories through the variable louver configuration were drawn from
observations for minimum and maximum sand size intervals at Re = 46,400.
48
µm
0-20µ
µm
63-200µ
Figure 2.15. Particle trajectories through the constant louver configuration were drawn from
observation for minimum and maximum sand size intervals at Re = 46,400.
Feeder Filters
100%
80%
60%
40%
20%
0%
0-20 20-40 38-63 63-200
µ m)
Sand Size (µ
Figure 2.16. Distribution of sand through the test facility for variable louver angle when varying
sand size at Re = 46,400.
49
Feeder Filters
Upstream Channel Collector
100%
80%
60%
40%
20%
0%
0-20 20-40 38-63 63-200
µ m)
Sand Size (µ
Figure 2.17. Distribution of sand through the test facility for constant louver angle when varying
sand size at Re = 46,400.
Recall from Figure 2.12 that comparison of measurements to computational results showed
that computational models over-predicted collection efficiency for all size intervals. The over-
prediction indicated that the models did not completely agree with the measured data; however,
the three-dimensional predictions were lower than two-dimensional predictions. The trend in
predicted collection efficiency can be explained by considering the two-dimensional predictions.
The two-dimensional predictions indicated an increase followed by a decrease in collection
efficiency with increasing sand size. This change in efficiency was caused by the differences
between the particle momentum (increasing with sand size) and flow field velocities. Increasing
the size range from 0-20µm to 20-40µm increased particle momentum such that particles were
separated by the louver array and passed into the collector where they were captured. Further
increasing the size range to 38-63µm and 63-200µm imparted enough momentum such that the
sand particles deflected from the separator surfaces and were not separator by the louver array, or
trapped in the collector. The three-dimensional trend followed the increasing and decreasing
50
two-dimensional trend of collection efficiency for similar reasons. The spanwise flow in the
collector, however, complicated sand trajectories that resulted in the three-dimensional trend
deviating from the two-dimensional trend for some sand size intervals.
Conclusions
This paper has presented pressure loss and collection efficiency measurements for two
louver separator configurations. One separator had a varied louver angle along the array (φ = 30°
to 45° to 30°), and the second separator had constant louver angle along the array (φ = 45°).
Both configurations were tested in a unique test facility over a range of Reynolds numbers and
sand sizes. Both two- and three-dimensional computational predictions were presented along
with measurements.
The pressure loss coefficients of both separator configurations were shown to increase for
decreasing Reynolds numbers. The similar pressure loss coefficients between the two louver
configurations indicated that louver angle did not significantly affect separator pressure loss.
The louver angle, however, did significantly affect the collection efficiency of the separator over
the range of Reynolds numbers. The collection efficiency of both the variable and constant
louver configurations increased to similar values for decreasing Reynolds numbers. The variable
louver configuration leveled off at η = 32% for low Reynolds numbers, and the constant louver
configuration continued to increase to a maximum collection efficiency of 35% at Re = 8,000.
The sand circulation in the collector was observed to increase in size with Reynolds
number. The change in circulation size affected sand accumulation in the collector. Sand
accumulated under the baffle in the collector at high Reynolds numbers because the large
circulation size directed particles under the baffle. At low Reynolds numbers, however, the
small circulation size caused sand to accumulate on the bottom of the collector out-from-under
the baffle.
The effect of sand size on collection efficiency was investigated by injecting discrete size
intervals. The collection efficiency of the variable louver configuration was affected by sand
size; however, the collection efficiency of the constant louver configuration was independent of
sand size. The variable louver configuration was found to best collect sand sizes less than 20µm
compared to collecting larger particles. Sand collection was best, however, when the entire size
51
range (0-200µm) was injected. The collection efficiency was unchanged, however, for the
constant louver configuration for any sand size interval.
Both two- and three-dimensional computational models were used to predict collection
efficiency and pressure loss. For varying Reynolds numbers computational predictions were
within 10% of measured collection efficiency values, and agreed with pressure loss coefficient
data at high Reynolds numbers. The predicted trends, however, did not agree with
measurements over the range of Reynolds numbers. The collection efficiency trend for
increasing sand size did not agree with measurements. Collection efficiency was over-predicted
for all discrete sand size intervals.
In conclusion, this paper has presented results that indicate particle collection efficiencies
greater than 30% can be obtained for an inertial louver separator that uses a static collector.
Two- and three-dimensional computational predictions were in agreement with some measured
values; however, computational predictions did not match measured collection efficiency trends.
Acknowledgments
The authors would like to thank Pratt & Whitney for funding and sponsoring this research
effort including Joel Wagner, Ryan Levy, and Takao Fukuda.
References
[1] Tabakoff, W., Hosny, W., Hamed, A., 1976, “Effect of Solid Particles on Turbine
Performance,”J. Engineering for Power, pp. 47-52.
[2] Kim, J., Dunn, M.G., Baran, A.J., Wade, D.P., Tremba, E.L., 1993, “Deposition of
Volcanic Materials in the Hot Sections of Two Gas Turbine Engines,” J. of Engineering for
Gas Turbines and Power, 115 pp.641-651.
[3] Dunn, M.G., Baran, A.J., Miatah, J., 1994, “Operation of Gas Turbine Engines in Volcanic
Ash Clouds,” ASME, 94-GT-170.
[4] Walsh, W.S., Thole, K.A., Joe, C., 2006, “Effects of Sand Ingestion on the Blockage of
Film-Cooling Holes,” GT2006-90067.
52
[5] Land, C.C., Thole, K.A., Joe, C., 2008, “Considerations of a Double-Wall Cooling Design
to Reduce Sand Blockage,” GT2008-50160.
[6] Zverev, N.I., 1946, “Shutter-Type Dust Collector of Small Dimensions,” Engineer’s
Digest, 7 (11), pp. 353-355.
[7] Smith, Jr., J.L., Goglia, M.J., 1956, “Mechanism of Separation in Louver-Type Dust
Separator,” ASME Transactions, 78 (2), pp. 389-399.
[8] Gee, D.E., Cole, B.N., 1969, “A Study of the Performance of Inertia Air Filters,” Inst.
Mech. Engineers - Symposium on Fluid Mechanics and Measurements in Two-Phase
Systems, University of Leeds, pp. 167-176.
[9] Poulton, P., Cole, B.N., 1981, “An Experimental and Numerical Investigation of Louvred
Inertia Air Filter Performance,” Conference on Gas Born Particles, Institution of
Mechanical Engineers, pp. 161-170.
[10] Schneider, O., Dohmen, H.J., Reichert, A.W., 2002, “Experimental Analysis of Dust
[11] Schneider, O., Dohmen, H.J., Benra, F.-K., Brillert, D., 2003, “Investigations of Dust
[12] Schneider, O., Dohmen, H.J., Benra, F.-K., Brillert, D., 2004, “Dust Separation in a Gas
Turbine Pre-Swirl Cooling Air System, a Parameter Variation,” GT2004-53048.
[13] Schneider, O., Benra, F.-K., Dohmen, H.J., Jarzombek, K., 2005, “A Contribution to the
Abrassive Effect of Particles in a Gas Turbine Pre-Swirl Cooling Air System,” GT2005-
68188.
53
[14] Syred, C., Griffiths, A., Syred, N., 2004, “Gas Turbine combustor with Integrated Ash
Removal for Fine Particulates,” GT2004-53270.
[15] Musgrove, G.O., Barringer, M.D., Thole, K.A., Grover, E., Barker, J., “Computational
Design of a Louver Particle Separator for Gas Turbine Engines,” GT2009-60199.
[16] Powder Technology, Inc., 2009, Burnsville, MN.
[17] ANSYS, Inc., Version 6.3.26, 2009 (ANSYS Inc.: New Hampshire).
[18] Moffat, R.J., 1988, “Describing the Uncertainties in Experimental Results,” Experimental
Thermal and Fluid Science, 1, pp. 3-17.
[19] Schetz, J., Fuhs, A., 1996, Handbook of Fluid Dynamics and Fluid Machinery, Vol 1, John
Wiley & Sons, Inc., New York, pp 905, Chap. 14.
[20] Hamed, A., Tabakoff, W., Wenglarz, R., “Erosion and Deposition in Turbomachinery,” J.
Propulsion and Power, 22 (2), pp.350-360.
54
Conclusions and Summary
The design of an inertial separator that utilized an array of louvers followed by a static
collector was studied using a two-dimensional model that was matched to engine conditions.
The effects of area ratio, louver geometry, and collector geometry on particle collection and
pressure loss were predicted. Area ratio affected the pressure loss of the separator more than the
louver geometry or collector geometry; however, particle collection efficiency was not affected.
The louver configurations that were made up of straight louvers at varying louver angle were
more effective to separate small particles compared to other louver shapes and constant louver
angles. The best louver geometry was made up of eight straight louvers with varying louver
angle (φ = 30° to 45° to 30°). The collector geometry was required to set up a favorable
circulation in the collector. A favorable circulation was set up by an inlet vortex, and was
required for high particle collection efficiency. The inlet vortex was developed from the flow
passing over the last louver tip in the louver array and impinging on the inlet lip of the collector.
A unique test facility was design and constructed to measure the performance of two
louver separator configurations. Two- and three-dimensional computational models that
matched the experimental conditions were simulated to compare predictions to measurements.
Pressure loss coefficients were measured to increase for decreasing Reynolds numbers,
where all measured values were consistent with less than 1% pressure loss at the engine
conditions modeled in the first paper. The louver configuration did not significantly affect the
pressure loss of the separator. Two- and three-dimensional computational predictions of Cp were
constant for all Reynolds numbers, and agreed with measurements for high Reynolds numbers.
Two- and three-dimensional predictions were similar because spanwise flow was not predicted in
the louver array.
Collection efficiency measurements showed that the louver configuration affected the
collection efficiency of the separator. The collection efficiency of both louver configurations
increased for decreasing Reynolds numbers; however, the collection efficiency trends were
different. Collection efficiency was also dependent on sand size. The variable louver
configuration was most effective to collect sand sizes less than 20µm compared to sizes greater
than 20µm. The collection of the constant louver configuration, however, was independent of
sand size. Two- and three-dimensional computational models predicted sand collection
55
efficiency values within 10% of measured values over a range of Reynolds numbers. The
predicted trends, however, did not agree with measured trends. Collection efficiency was over-
predicted by both two- and three-dimensional computational models for all sand size intervals
compared to measured values. Over the range of Reynolds numbers, the two- and three-
dimensional computational models predicted similar collection efficiencies for the variable
louver configuration. The constant louver configuration, however, resulted in lower collection
efficiencies predicted by the three-dimensional model at high Reynolds numbers compared to the
two-dimensional model due to spanwise flow effects in the three-dimensional collector.
The computational studies of the louver separator revealed that particle momentum was
an important consideration for particle collection. Small particle trajectories were influenced by
the flow field resulting from the louver configuration. Large particle trajectories were influenced
by deflections from surfaces. Computational studies showed the importance of the collector
circulation on particle collection. The collector circulation was sensitive to the flow field
through the louver array as well as to the flow at the collector inlet.
Through experimental observations, the circulation in the collector was confirmed to
affect particle collection. Modifying the collector inlet to produce favorable circulation in the
collector improved particle collection. The effect of louver angle was measured to not
significantly affect the pressure loss of the separator. The collection efficiency, however, was
significantly affected by louver angle. Computational predictions agreed with some of the
measured values; however, complete agreement was not obtained.
Future work could use advanced flow measurement techniques to resolve the flow in the
collector. The measured flow field could be compared to the three-dimensional model for
validation. The comparison would indicate if the flow field prediction was inaccurate, or if the
particle trajectory model should be improved. The effects of the endwall on collection efficiency
could be investigated using a channel width larger than 10H, or an annular louver separator.
56
Appendix A: Computational Study of the Louver Separator
Nomenclature
AsR aspect ratio of the design space
AR area ratio, AR = δ(# louvers) / H)
CD particle drag coefficient
Dc characteristic dimension, Dc = H for particle scaling
Ds sand diameter
H height of the design space
I turbulence intensity, I = 0.16(ReH)-0.125
lt turbulent length scale
Stk Stokes number, Stk = ρpDs2Up/18µDc
Up particle velocity
Z spanwise distance measured from the endwall
Greek:
α louver tip angle with respect to the horizontal
δ louver gap
θ louver array angle with respect to the horizontal, θ=tan-1(1/AsR)
µ fluid viscosity
ρp particle density
φ louver angle with respect to θ
Introduction
The complete computational study of the separator included three different models: 2D
model matching engine conditions; 2D model matching lab experimental conditions; and 3D
model matching lab experimental conditions. The details of the three models are presented in
Table A.1. Computational models matched to experimental conditions were needed because
measurements of the variable louver configuration (φ=30° to 45° to 30°) resulted in negligible
collection efficiency compared to the predicted efficiency (η = 87%) in Paper 1. The results of
these computational studies and their methodologies have been presented in Paper 1 and Paper 2.
Appendix A describes for each model the computational methodology and results in detail.
Results for the 2D engine model of the variable louver configuration present separator
geometries not discussed in Paper 1. Results for the 2D and 3D experiment matched models of
the variable and constant louver configurations are presented and compared to experimental data.
57
Table A.1. 2D and 3D Computational Models
2D engine 2D experiment matched 3D experiment matched

φ=30° to 45° to 30° φ=30° to 45° to 30° φ=30° to 45° to 30°
Reynolds number 135,500 46,400 46,400

Particle size range 1-50µm 1-200µm 1-200µm
Channel height (H) 25.4 mm 48.3 mm 48.3 mm
Channel width 0 0 5H
Domain length 11H 39H 39H
Part thickness 0 2.36 mm 2.36 mm
Wedge and baffle No Yes Yes
Although each of the three computational models is different, the grid generation and
solution methodology are similar and presented together. The geometry of each separator is
shown in Figure A.1 through Figure A.3. The observable differences between the separator
geometries of the 2D engine and 2D experiment matched models are the presence of part
thickness, reduction in louver anchor lengths, collector size, and the wedge and baffle
modification. The collector size and louver anchor lengths were reduced when part thickness
was added to the separator design so that the correct separator area ratio (AR=1.50) was
maintained. The wedge and baffle was incorporated into the 2D and 3D experiment matched
models to match the experiment geometry. Not only is separator geometry different between the
engine and experiment matched models, but also the domain height and length, particle size
range, and Reynolds numbers. The effects of the model differences were investigated and are
presented later to compare the 2D engine model to the 2D and 3D experiment matched models.
Louvers Collector
Figure A.1. The 2D engine model did not include part thickness or the wedge and baffle
modification, variable louver configuration shown.
58
Wedge
Baffle
Figure A.2. The 2D experiment matched model included part thickness and the wedge and baffle
modification, (upper) variable and (lower) constant louver configurations.
Figure A.3. The 3D experiment matched model was created by extruding the 2D experiment
matched geometry into the spanwise direction, (upper) variable and (lower) constant louver
configurations.
For each model, the computational domain and separator geometry were constructed and
meshed using GAMBIT 2.4.6 [1]. The domain of the 2D engine model was axially shorter than
the experiment matched models; however, the design space of the 2D and 3D experiment
matched models were geometrically scaled to the 2D engine model using the channel height (H).
The domain and mesh of the 3D experiment matched model was constructed by extruding the 2D
experiment matched model to a length of 5H in the spanwise direction with 0.13H spanwise node
spacing. Recall that the test section in the experimental test facility had a span of 10H, thus a
symmetry boundary condition was placed at the 5H spanwise edge of the computational domain.
59
For the 2D engine model, the mass flow inlet was located 1H upstream of the separator,
and the pressure outlet was located 5H downstream of the separator, as shown in Figure A.4. In
comparison, the 2D and 3D experiment matched models incorporated the upstream and
downstream channel lengths of the experimental test facility into the computational domain, as
shown in Figure A.5. The velocity inlet location (9H upstream of the separator) was consistent
with the particle injection slot of the experimental test facility, and the domain outflow exit was
consistent with the downstream channel exit (23H downstream of the separator). Boundary
conditions and operating conditions for each computational model are presented in Table A.2
and Table A.3, respectively.
Each computational domain was meshed with a combination of unstructured and
structured quadrilateral cells, shown in Figure A.4 and Figure A.5. The unstructured mesh was
placed within the design space and the structured mesh was placed upstream and downstream of
the design space. The 2D and 3D experiment matched models used a boundary layer grid along
the top and bottom walls of the domain and along the outside wall of the collector. The
boundary layer mesh consisted of the first grid point placed a distance of 0.002H from the wall
(y+=30 for Re = 46,400), with sizing ratio 1.25 and 6 levels.
H 5H 5H
Mass flow inlet Design space Pressure exit
(a) (b)
Figure A.4. The domain for the 2D engine model used an (a) unstructured mesh in the design
space and (b) structured mesh outside of the design space.
60
Upstream channel Test section Downstream channel
9H 5H 5H 18H
Velocity inlet Design space H Outflow exit
(a) (b)
Figure A.5. The 2D and 3D experiment matched domains matched the test facility from the sand
injection location to the downstream channel exit, using a combination of (a) unstructured and
(b) structured mesh regions.
Table A.2. Boundary Conditions for Each Computational Model
Inlet Exit Walls Energy Equation

Mass flow Pressure outlet
2D engine No slip Yes
4233 kPa, 661 K, k=ω=1 4233 kPa, 661 K
2D experiment matched Velocity
Outflow No slip No
3D experiment matched I=4%, lt=48.3mm
Table A.3. Operating Conditions for Each Computational Model
Flow Flow Operating Gravity Y

X
density temperature pressure (gy) Z
2D engine Ideal gas 934 K 0 0
2D experiment matched 1.137 kg/m3 300 K 97.2 kPa -9.81 m/s2
3D experiment matched 1.137 kg/m3 300 K 97.2 kPa -9.81 m/s2
The mesh for each computational model was constructed with attention to grid size and
quality. Grid quality was maintained with low equiangle cell skewness values, as shown in
Table A.4. Appropriate grid sizes were determined through a grid independence evaluation,
where the grid was successively refined until collection efficiency did not change with grid size.
Pressure loss coefficient was not significantly affected by grid size and was not used to establish
grid independence. Collection efficiency predictions were found to be independent of grid size
when the grid was refined multiple times in regions of total pressure gradients less than 10% of
the maximum total pressure gradient in the domain. Collection efficiency predictions for three
grid sizes of the 2D experiment matched model are shown in Figure A.6 and Figure A.7. The
unrefined grid size of 3.0 x 104 cells was refined multiple times to obtain a size of 6.4 x 104 cells.
61
The design space was then further refined (12 x 105 cells) to show that a finer mesh did not affect
the computational predictions. The similar collection efficiency predictions between grid sizes
of 6.4 x 104 and 12 x 105 indicated that 6.4 x 104 was sufficient for grid independent results. A
similar grid study was also completed for the 2D engine and 3D experiment matched models.
Figure A.8 and Figure A.9 show that a grid size of 3.0 x 106 was required for grid independence
of the 3D model. All collection efficiency and pressure loss coefficient results presented in
following sections are for grid independent flow field solutions
Table A.4. Cell Skewness for Variable Louver Configuration
% cells with Average

Grid Size Cell Shape
skewness < 0.6 skewness
2D engine 4.5x103 Quadrilateral 94% 0.15
3
2D experiment matched 6.4x10 Quadrilateral 100% 0.06
3D experiment matched 3.3x106 Hexahedral 100% 0.09
100%
2D experiment matched
3.0 x 104 cells

80%
6.4 x 104 cells
12.0 x 105 cells
60%
η
40%
20%
0%
1 20 40 60 80 100 120 140 160 180 200
µm)
Particle Size (µ
Figure A.6. Collection efficiency did not significantly change for the 2D experiment matched
grid size greater than 6.4 x 104 cells for the variable louver configuration at Re = 46,400.
62
100%
4
3.0 x 10 cells
80%
6.4 x 104 cells
12 x 105 cells
60%
η
40%
20%
0%
1-20 20-40 40-60 60-200 1-200
µm)
Size Range (µ
Figure A.7. The collection of particle sizes less than 20 µm varied with the 2D experiment
matched grid size more than other sizes for the variable louver configuration at Re = 46,400.
100%
1.6 x 106 cells

80%
3.0 x 106 cells
60%
η
40%
20%
0%
1 20 40 60 80 100 120 140 160 180 200
µm)
Particle size (µ
Figure A.8. For most particles sizes, a 3D grid size of 3.0 x 106 did not significantly affect the
collection efficiency of the constant louver configuration at Re = 46,400.
63
100%
1.6 x 106 cells

80%
3.0 x 106 cells
60%
η
40%
20%
0%
1-20 20-40 40-60 60-200 1-200
µm)
Size range (µ
Figure A.9. The 3.0 x 106 cell grid did not significantly affect collection efficiency for each
particle size range injected compared to the 1.6 x 106 cell grid for the constant louver
configuration at Re = 46,400
The computational solution method followed the method described in Paper 1 where the
flow field was solved before particle trajectories were calculated by integrating a force balance
equation for each particle. The commercial computational software FLUENT 6.3.26 [2], was
used to solve the flow field using RANS equations and the k-ω turbulence model. Flow field
solutions were 2nd order accurate and met a convergence criterion of 10-6. Particle trajectories
were calculated by integrating a force balance on each particle, where particle drag and gravity
were the dominating forces, as shown in Equation A.1. Gravity affected collection efficiency
predictions as much as 15% for large particles, but did not affect the collection of small particles.
) ( )
dU p 18µ C D Re p g ρp − ρf (A.1)
dt
= 2
24
(
U − Up +
ρf
ρpd p
The drag coefficient (CD) in Equation A.1 is a Stokes drag coefficient that is a function of
shape factor. All computational models used a shape factor of one, indicating that all particles
were perfect spheres. The particle force balance equation was integrated to solve for the particle
trajectories through the domain, with a maximum of 50,000 integration steps and an integration
64
length scale of 0.06 mm. The integration steps and length scale were sufficient to allow all
particles not trapped in the collector to leave the domain exit. The domain exit was set to an
escape boundary condition and the walls in the domain used a reflected boundary condition
(restitution coefficient of one). Recall from Paper 1 that shape factor and restitution coefficient
influenced particle collection predictions.
In evaluating the performance of the separator, pressure loss coefficient and collection
efficiency were obtained from the simulation solutions. Pressure loss coefficient (Cp) was
computed from mass averaged total pressure and area averaged dynamic pressure. Pressure
values were obtained through integration along the channel height at the domain inlet and exit
(2D engine model) or at locations consistent with upstream and downstream static tap locations
in the test facility (2D and 3D experiment matched models). Collection efficiency (η) was
determined by recording the number of particles injected into and escaped from the domain.
Particle sizes 1 < Ds < 50µm (ρp = 2419 kg/m3) were injected into the 2D engine domain, and
particle sizes 1 < Ds < 200µm (ρp = 1201 kg/m3) were injected into the 2D and 3D experiment
matched domains. Collection efficiency was recorded for each particle size, and overall
collection efficiency was calculated by integrating the predicted collection efficiencies with
respect to particle size. Calculating overall collection efficiency this way allowed the
computational predictions to be weighted by particle size and therefore directly comparable to
measured collection efficiency results. The integration was performed using the trapezoidal
method to find the area under the predicted collection efficiency curve, then normalizing the
resultant area by the particle size range.
The number of particles that were injected into each computational model was
determined to provide collection efficiency results independent of the number of injected
particles. Particles were injected uniformly spaced across the domain height of the 2D engine
and 2D experiment matched. For the 2D engine model, one hundred particles for each diameter
were sufficient for independent collection efficiency, as shown in Figure A.10. One thousand
injected particles for each particle diameter was sufficient for the 2D experiment matched model,
as shown in Figure A.11. Particle injections for the 3D experiment matched model were
uniformly spaced across the domain height at spanwise intervals. To establish an independent
particle injection method for the 3D experiment matched model, the number of injected particles
as well as the spanwise spacing of the injections was determined, as illustrated in Figure A.12.
65
With an initial spanwise injection spacing of H/2, 1,000 particles were required for independent
collection efficiency, as shown in Figure A.13. A spanwise injection spacing of H/8 for
injections of 1,000 particles was required for independent collection efficiency, as shown in
Figure A.14.
10 particles 100 particles 1,000 particles
100%
2D engine
80%
60%
η
40%
20%
0%
1 5 7 10 15 20 25 30 40 50
µm)
Particle size (µ
Figure A.10. One hundred particles was sufficient for independent collection efficiency for the
2D engine model of the variable louver configuration at Re = 135,500.
100%
80%
10 Particles
100 Particles
1,000 Particles
60% 10,000 Particles
η
40%
20%
0%
1 20 40 60 80 100 120 140 160 180 200
µm)
Particle Size (µ
Figure A.11. One thousand particles were injected in the 2D experiment matched model to
obtain independent collection efficiency for the variable louver configuration at Re = 46,400.
66
Number of Particles
Spacing
Figure A.12. The number of particles and spacing of particle injections were investigated to
determine an independent particle injection method.
100%
80%
Total Particles Injection
particles per injection spacing
1,100 100 H/2
60% 11,000 1000 H/2
η 110000 10,000 H/2
40%
20%
0%
1 20 40 60 80 100 120 140 160 180 200
µm)
Particle size (µ
Figure A.13. A total of eleven spanwise injections of 1,000 particles each with spanwise spacing
H/2 were required for independent collection efficiency of the constant louver configuration at
Re = 46,400.
67
100%
80%
Total Particles Injection
particles per injection spacing
60% 11,000 1,000 H/2

21,000 1,000 H/4
η 41,000 1,000 H/8
81,000 1,000 H/16
40%
20%
0%
1 20 40 60 80 100 120 140 160 180 200
µm)
Particle size (µ
Figure A.14. Spanwise injection spacing of H/8 was required for independent collection
efficiency results for the constant louver configuration at Re = 46,400.
Collection efficiency and pressure loss coefficient predictions for all three computational
models are presented in this section. In addition, the differences in geometric and operating
conditions between the 2D engine and 2D experiment matched models are investigated. Results
are presented for separator configurations that were solved using the 2D engine model, but not
discussed in Paper 1. The 2D engine and 2D experiment matched models are compared, where
the differences between the models are described with regard to the effect on collection
efficiency. The results for the 2D and 3D experiment matched models are presented together and
discussed
2D Engine Model
While computational results for separator configurations of the 2D engine model were
presented in the Paper 1, not all of the simulated configurations were described. All of the area
ratios that were studied have been presented in the Paper 1. Louver configurations comprising
11 louvers with constant area ratio (AR = 1.50) and no collector are presented. Collector
configurations are presented using a constant louver configuration of 8 straight louvers at
constant area ratio (AR = 1.50) and variable louver angle (φ = 30° to 45° to 30°). Configurations
presented in Paper 1 are repeated here for comparison.
68
The louver configuration and collector geometry did not significantly affect the pressure
loss of the separator. All louver configurations resulted in pressure loss values between 0.62%
and 0.44% of the separator inlet total pressure. Pressure losses for all of the collector
configurations ranged between 0.47% and 0.58% for the best louver configuration (8 louvers, φ =
30° to 45° to 30°).
Louver Geometry
Straight, bent tip, convex, and concave louvers were investigated with different louver
angles (φ) for straight and bent tip louvers, as shown in Figure A.15. Bent tip louvers were
similar to straight louvers; however, the louver tips had a length 6% of the louver length and
were at angle α with respect to the horizontal. The concave and convex louvers were oriented
along angle θ and had arc angle π/2 with constant radius. Comparison of the four louver shapes
indicated that curved louvers did not improve collection efficiency over straight or bent tip
louvers, as shown in Figure A.16. Bent tip louvers were better than straight louvers to collect
small particles for low louver angles (φ = 30°, 45°). Bent tip louvers did not improve particle
collection, however, for high louver angles (φ = 60°, 75°).
α =θ
[α θ]
30°Straight 30°Bent tip
α =θ
[α θ]

α =θ
[α θ]

α =θ
[α θ]
Concave Convex
Figure A.15. Different louver shapes and angles were simulated to improve particle collection.
69
Straight α=θ
Bent tip [α θ] Curved
30° 60° 30° 60° Convex
45° 75° 45° 75° Concave
100%
80%
60%
η
40%
20%
0%
0 10 20 30 40 50
µ m)
Particle size (µ
Figure A.16. Curved louvers did not improve collection efficiency relative to straight louvers
and bent tip louvers over a range of louver angles; AR=1.50, 11 louvers, no collector.
Because bent tip louvers improved small particle collection, the sensitivity of tip angle
(α) and tip length was investigated. The tip angle was increased to α=θ+φ/2 for the φ=30°
louver configuration, and tip length was increased to 12% of the louver length for louver angle
φ= 60°, as shown in Figure A.17. The effects of varying the tip angle or tip length are shown in
Figure A.18 to not affect collection efficiency for two constant louver angle configurations.
Other louver designs incrementally changed the louver angle (φ) along the array for both
straight louver and bent tip louver configurations, as shown in Figure A.19. Straight louvers
with varying louver angle resulted in high small particle collection, where bent tip louvers did
not improve collection efficiency over straight louvers, as shown in Figure A.20. In fact, using
bent tip louvers for φ=30° to 45° to 30° dramatically reduced small particle collection compared
to φ=30° to 45° to 30° with straight louvers. On the other hand, using bent tip louvers (α=θ) for
φ= 60° to 30° did not affect collection efficiency compared to straight louvers of the same louver
angle. The effect of tip angle on varied louver designs is different that constant louver designs,
which showed no change in collection efficiency. Increasing the tip angle (α=θ+φ/2) for φ= 60°
to 30° reduced collection of mid size particles compared to α=θ.
70
30°Straight 60°Straight
α =θ
[α θ] α =θ
[α θ]
30°Bent tip 60°Bent tip
α =θ
[α θ+φ
φ/2] [tip length=(0.12)L]
α =θ
[α θ]
30°Bent tip 60°Bent tip
Figure A.17. The sensitivity of the louver tip was investigated by varying tip angle (α) and tip
length.
30° 60°
Straight Straight
α=θ
Bent tip [α θ] α=θ
Bent tip [α θ]
α=θ
Bent tip [α θ+φφ/2] α=θ
Bent tip [α θ, length=L*0.12]
100%
80%
60%
η
40%
20%
0%
0 10 20 30 40 50
µm)
Particle size (µ
Figure A.18. Varying the louver tip angle and length did not affect collection efficiency;
AR=1.50, 11 louvers, no collector.
71
60°to 30°Straight 30°to 45°to 30°Straight 75°to 30°Straight
α =θ
[α θ] α =θ
[α θ]
60°to 30°Bent tip 60°to 45°Bent tip
α =θ
[α θ+φ
φ/2] α =θ
[α θ+φ
φ/2]
60°to 30°Bent tip 30°to 45°to 30°Bent tip
Figure A.19. Louver configurations incremented the louver angle along the array for straight and
bent tip louvers.
Bent tip Bent tip

Straight α=θ
[α θ] α=θ
[α θ+φφ/2]
60° to 30° 60° to 30° 60° to 30°
75° to 30° 60° to 45° 30° to 45° to 30°
30° to 45° to 30° Blank
100%
80%
60%
η
40%
20%
0%
0 10 20 30 40 50
µ m)
Particle size (µ
Figure A.20. Bent tip louvers did not increase the collection efficiency of variable louver
configurations; AR=1.50, 11 louvers, no collector.
Other attempts to increase small particle collection efficiency included modifying

specific louvers within the array. Specific louvers were modified for the best performing
variable louver configurations of φ= 60° to 30° and φ=30° to 45° to 30°. The number of louvers
in the configuration was reduced from 11 to 8 louvers because fewer louvers resulted in higher
72
small particle collection compared to many louvers. The modified geometries used bent tip
louvers with tip angle α=θ for the last three louvers in the φ= 60° to 30° configuration, and the
last louver in the φ=30° to 45° to 30° configuration, as shown in Figure A.21. The specific
louvers were changed to bent tip louvers based on observations of predicted particle trajectories
near the last louvers in the array. The purpose of replacing only a few louvers was to influence
particle trajectories near those louvers such that the particles would be diverted into the area
where a collector would later be placed. Replacing the straight louvers with bent tip louvers
resulted in a different flow field through the entire louver array, which affected particle
trajectories upstream of the bent tip louvers. The new flow field resulted in lower particle
collection for both variable louver configurations, indicating that minor changes to the louver
configuration have major effects on separator performance, as shown in Figure A.22.
Straight Last 3 louvers bent,

α =θ
θ
60°to 30° 60°to 30°
Straight Last louver bent,

α =θ
θ
30°to 45°to 30° 30°to 45°to 30°
Figure A.21. Final modifications were confined to individual louvers in the best performing
louver configurations to attempt to increase small particle collection.
73
60° to 30°
α=θ
60° to 30° Last 3 louvers have bent tip [α θ]
30° to 45° to 30°
α=θ
30° to 45° to 30° Last louver has bent tip [α θ]
100%
80%
60%
η
40%
20%
0%
0 10 20 30 40 50
µ m)
Particle size (µ
Figure A.22. Modification of a few louvers in a variable louver configuration reduced particle
collection; AR=1.50, 8 louvers, no collector.
Collector Geometry
The collector was added to the φ = 30° to 45° to 30° straight louver configuration with
eight louvers. Variations to the collector involved the inlet region, collector shape, and collector
size. The inlet region was investigated to identify the best inlet height and the effects of an inlet
lip. Modification of collector shape and size were based on predicted particle trajectories, where
wall shapes were changed to deflect particles away from the collector inlet. The goal of the
collector was to increase large particle collection without reducing small particle collection.
With a constant collector size and shape, the collector inlet was modified to determine the
effect of inlet opening size and the presence of an inlet lip, as shown in Figure A.23. The best
collector inlet had an inlet lip and did not have a small inlet height. The presence of the collector
inlet lip (collector 6) set up of a favorable circulation in the collector, as discussed in Paper 1.
Without the inlet lip (collector 7) the circulation direction was unfavorable (counter-clockwise)
and allowed particles entering the collector to leave through the inlet. The large collector inlet
(collector 6) resulted in better small particle collection than the small inlet (collector 5) because
more particles were able to enter into the collector through the large inlet.
74
Collector 5 Collector 6 Collector 7
Figure A.23. The collector inlet design was investigated to increase particle collection.
100%
80%
Collector 5
60% Collector 6
η Collector 7
40%
20%
0%
0 10 20 30 40 50
Particle Size (µ m)
Figure A.24. The collector with a large inlet and inlet lip performed better than a larger inlet or
without an inlet lip.
The inlet lip direction was investigated to determine if an outward lip would improve
collection efficiency over an inward lip, as shown in Figure A.25. Both inlet lips resulted in
similar collection efficiency because they both produced the inlet vortex and favorable
circulation in the collector. The inward lip (collector 10) was more effective to collect mid size
particles than the outward lip (collector 2), as shown in Figure A.26. Contrarily, the outward lip
(collector 2) was better to collect large particles than the inward lip (collector 10). The inward
inlet lip (collector 10) performed better than the outward lip (collector 2) for mid size particles
because the inward lip (collector 10) helped deflect particles circulating in the collector away
from the collector inlet. The outward lip (collector 2) was more effective to collect large
particles because the outward lip forced the flow around the last louver tip to turn abruptly so
that the flow passed between the last louver and the outward lip. The abrupt turn made it easier
for 30µm and 40µm particles to deviate from the turning flow and pass into the collector.
75
Collector 2 Collector 10
Figure A.25. The collector inlet lip direction was investigated for similar collector shape and
size.
100%
80%
60%
Collector 2
η Collector 10
40%
20%
0%
0 10 20 30 40 50
Figure A.26. The inward inlet lip had better mid size particle collection than the outward inlet
lip.
The collector wall was modified to influence particle deflections in the collector while
keeping the inlet region unchanged. Collector wall modifications shown in Figure A.27 were
confined to the back wall because the back wall affected most of the initial particle deflections in
the collector. Changes to the back wall of the collector primarily affected large particle
collection because large particle trajectories were influenced by deflection, as shown in Figure
A.28. Collector 8 had poor particle collection compared to the other collector designs because
the inlet lip of collector 8 was too short to set up the favorable circulation direction in the
collector. The curved wall of collector 9 resulted in better collection than collectors with straight
walls because the curved walls influenced large particle trajectories to not deflect towards the
collector inlet.
76
Figure A.27. Collector shape modifications included changes to the back wall of the collector.
100%
80%
60%
Collector 3
η Collector 4
Collector 6
40% Collector 8
Collector 9
Collector 10
20%
0%
0 10 20 30 40 50
Figure A.28. The back wall of the collector affected the collection of large particles.
Because the particle separator was to be placed within a confined engine space, the
collector length was investigated to determine the effect of collector length on collection
efficiency. Two of the best performing collector configurations were axially shortened and
compared to their longer counterparts, as shown in Figure A.29. Collection efficiency of the
curved collector was not affected by collector length, as shown in Figure A.30. The straight wall
collector, however, was affected by axial length because the shape of the back wall did not
deflect all impacting particles in the same direction. Because the curved wall collector had the
best overall collection efficiency and was not affected by collector length, the curved wall design
was chosen to be further modified to improve particle collection.
77
Collector 9 Collector 10
Collector 9 Short Collector 10 Short
Figure A.29. Two collector configurations were axially shortened to determine the effects of
collector length.
100%
80%
60%
Collector 9
η Collector 9 Short
Collector 10
40% Collector 10 Short
20%
0%
0 10 20 30 40 50
Figure A.30. Collection efficiency is not affected by axial length for the curved collector.
The compact, curved collector walls were modified to determine the effect of the contour
of the bottom wall on particle collection. The bottom wall was contoured outward (collector 11)
and inward (collector 12) from the design studied thus far, as shown in Figure A.31. The inward
and outward contours did not improve collection efficiency over the original contour design, as
shown in Figure A.32. Collection efficiency was reduced for the inward and outward contours
because the contours modified the circulation direction in the collector to have an unfavorable
circulation direction.
78
Collector 9 Short Collector 11 Collector 12
Figure A.31. The effect of the curvature of the bottom wall on the collection efficiency was
investigated.
100%
80%
60%
η Collector 9 Short
Collector 11
40% Collector 12
20%
0%
0 10 20 30 40 50
Figure A.32. Modifying the contour of the bottom wall of the curved collector did not improve
collection efficiency over the original design.

In summary, the pressure loss of the 2D engine model was significantly affected by area
ratio. The louver geometry and collector did not significantly affect pressure loss. Louver
geometry was most influential to affect small particle collection. Straight louvers at varying
louver angle were more effective to collect small particles than other louver shapes or constant
louver angles. Changes to a few louvers in the louver array affected the flow through the entire
array resulting in significant effects to collection efficiency predictions. The collector geometry
was most effective to increase large particle collection. The length of the collector did not affect
particle collection. Curved collector walls were better than straight walls to collect large
particles by controlling particle deflections in the collector. An inlet lip inside the collector set
up the favorable circulation and deflected particles into the collector.
79
Comparison of the 2D Engine and 2D Experiment Matched Models
The results of the 2D engine model and the 2D experiment matched model are compared
to discuss the effects of the model differences on collection efficiency. The model differences
that are independently investigated include Reynolds number, louver thickness, the wedge and
baffle, and particle size range. In comparing the two models, the same variable louver
configuration (φ = 30° to 45° to 30°) and collector geometry are simulated by both models.
The particle size range 1-50µm was injected in the 2D engine model; however, a particle
size range 1-200µm was injected in the 2D experiment matched model. The 1-200µm size range
was scaled to an equivalent size range for the 2Dengine model using Stokes number because of
the different Reynolds numbers and particle densities used in the two models, as shown in Table
A.5. Stokes numbers indicate how well particles follow the flow field. The inlet height (H),
inlet velocity, and particle density were used to scale the injected particle size range to match the
maximum stokes number in the experiment (Stk = 47), which was determined from the largest
particle size (200µm) at Re = 46,400. For comparison, the Stokes number corresponding to
50µm size particles in the 2D engine model at Re = 135,500 was Stk = 4.4.
Table A.5. Scaled Particle Sizes
Reynolds Inlet height Scaled particle ρp

number (H) size range (kg/m)3
2D engine 135,500 25.4mm 1-163 µm 2419
2D engine 46,400 25.4 mm 1-280 µm 2419
2D experiment matched 135,500 48.3 mm 1-117 µm 1201
2D experiment matched 46,400 48.3 mm 1-200 µm 1201
Recall that the variable louver configuration presented in the Paper 1 (no thickness, no
wedge and baffle) resulted in overall particle collection of η = 87% for sizes 1-50µm (Stk < 4.4).
The overall collection efficiency, however, decreased to η = 55% when injecting a size range
scaled to the experimental size range (Stk < 47). By injecting a larger particle size range, the
predicted collection efficiency was significantly decreased because of the low collection of
particles with high Stokes numbers, as shown in Figure A.33.
In comparison to the significant effect of particle size range, the presence of louver
thickness is shown in Figure A.33 to not affect collection efficiency for the 2D engine model.
The presence of louver thickness, however, did determine the effect of Reynolds number on
80
collection efficiency. Figure A.34 shows that Reynolds number significantly affected the
collection efficiency when no louver thickness was included in the model. Reynolds number had
little effect on particle collection when louver thickness was included in the model.
The wedge and baffle had the most significant effect on predicted collection efficiency
regardless of louver thickness, as shown in Figure A.35. For both thickness and no thickness
models the collection efficiency was predicted to decrease for low Stokes numbers and increase
for high Stokes numbers. The overall collection efficiency for the 2D engine model was more
significantly affected than the overall collection of the 2D experiment matched model.
Nevertheless, both of the 2D models predicted that the addition of a wedge and baffle to the
collector reduced collection efficiency. Measurements presented in the second paper, however,
indicated that the addition of the wedge and baffle increased collection efficiency. Therefore, the
predicted effect of the wedge and baffle was inconsistent with the measured effect.
100%
Re=135,500 No wedge & baffle
2D engine - No thickness
80% 2D engine - Thickness
60% η=55%
η η=50%
40%
20%
0%
0 10 20 30 40 50
Stokes number, Stk
Figure A.33. Collection efficiency predictions were not significantly affected by the thickness of
the separator for the 2D engine model: Re=135,500, No wedge and baffle modification.
81
2D engine - No thickness 2D experiment match - Thickness
Re=135,500 Re=135,500
Re=46,400 Re=46,400
100%
No wedge & baffle
80%
η=55%
η =32%
60%
η
η=42%
40%
20%
η=27%
0%
0 10 20 30 40 50
Stokes number, Stk
Figure A.34. The collection efficiency of configurations without the wedge and baffle
modification were affected by Reynolds number when part thickness was not included compared
to when part thickness was included.
.
100%
Re=46,400
2D engine - No thickness
80% η=42% No wedge & baffle
Wedge & baffle
60% 2D experiment match - Thickness
η=27% No wedge & baffle
η Wedge & baffle
40%
η=25%
20%
η=22%
0%
0 10 20 30 40 50
Stokes number, Stk
Figure A.35. The wedge and baffle lowered the predicted overall collection efficiency for both
the 2D engine and 2D experiment matched models at Re=46,400.
82
In summary, the collection efficiency of the 2D engine model decreased from 87% to
55% when including a Stokes range Stk < 47 instead of Stk < 4.4. The louver thickness did not
affect collection efficiency of the 2D engine model at Re = 135,500. Reynolds number did not
affect collection efficiency when louver thickness was included in the model. The wedge and
baffle decreased the collection efficiency of both 2D models for Stk < 47. The wedge and baffle
was predicted to reduce the collection efficiency of the separator to 25% for the 2D experiment
matched model and 22% for the 2D engine model.
2D and 3D Experiment Matched Models

The computational results for the 2D and 3D experiment matched models were presented
in the second paper. Recall that the predicted pressure losses and collection efficiencies agreed
with measured values for most Reynolds numbers. The effect of sand size on collection
efficiency, however, did not agree with measurements. Also, the predicted trends did not agree
with the measured trends. This section further explains the computational results reported in the
second paper.
Pressure Loss Results

The 2D and 3D predicted pressure loss coefficients agreed with measurements for most
Reynolds numbers, as shown in Figure A.36. The predictions agreed with data for Reynolds
numbers greater than 25,000; however, at lower Reynolds numbers predictions were less than the
measured values. Predicted pressure loss coefficients did not significantly change with
Reynolds number because the predicted separation regions in the separator did not change with
Reynolds number. Figure A.37 and Figure A.38 show similar contours of Cp in the separator for
Re = 46,400 and Re = 8,000, respectively, where high values of Cp indicate low pressure and low
values of Cp indicate high pressure regions. The Cp contours for the 2D and 3D models at both
high and low Reynolds numbers did not show a significant change in separation region size or Cp
magnitude, which indicates similar pressure losses at high and low Reynolds numbers.
Pressure loss coefficients were also predicted to be similar between the 2D and 3D
models at low and high Reynolds numbers because similar separation regions were predicted
between the 2D and 3D models, as shown in Figure A.37 and Figure A.38. Spanwise flow
effects through the louver array were not predicted by the 3D model. Therefore, Cp values were
uniform in the spanwise direction and similar to 2D predictions.
83
25
20
15
C
p
10
5 2D CFD 2D CFD
3D CFD 3D CFD
0
0 1 2 3 4 5
4
Re [10 ]
Figure A.36. Pressure loss across both separator configurations increased for decreasing
Reynolds numbers.
2D
Cp
3D
Z=0.125H
3D
Z=1.25H
3D
Z=2.5H
3D
Z=3.75H
3D
Z=4.875H
louver (right) configurations at Re = 46,400.
84
2D
Cp
3D
Z=0.125H
3D
Z=1.25H
3D
Z=2.5H
3D
Z=3.75H
3D
Z=4.875H
louver (right) configurations at Re = 8,000.
Reynolds Number Effects on Collection Efficiency

As was previously discussed in Paper 2, both 2D and 3D computational predictions
agreed within 10% of measured collection efficiency over the range of Reynolds numbers tested,
as shown in Figure A.39. The predicted trends, however, did not agree with measured trends.
Measurements showed that the collection efficiency trend was dependent on the louver
configuration. Both louver configurations resulted in increasing collection efficiency for
decreasing Reynolds numbers. The variable louver configuration was measured to level off at η
= 32% for low Reynolds numbers, while the constant louver configuration was measured to
increase to a maximum η = 35% at Re = 8,000 without leveling off. The 2D and 3D
computational predictions, on the other hand, predicted that both louver configurations had
similar trends, where neither configuration was predicted to level off at low Reynolds numbers.
Furthermore, the 2D and 3D predictions were similar for the variable louver configuration;
however, the 2D and 3D predictions for the constant louver configuration were not similar for
high Reynolds numbers.
85
50%
40% 2D CFD 2D CFD
3D CFD 3D CFD
30%
η
20%
10%
0%
0 1 2 3 4 5
4
Re [10 ]
Figure A.39. Collection efficiency predictions of sand size range 0-200µm were within 10% of
measurements over the range of Reynolds numbers for the variable and constant louver
configurations.
The 2D and 3D predictions agreed for the variable louver configuration because spanwise
flow in the 3D collector did not disrupt the favorable circulation. The favorable circulation of
the constant louver configuration, however, was disrupted by the spanwise flow in the 3D
collector. Recall from Figure 1.15 and Figure 1.16 that favorable circulation directed particles
away from the collector inlet, and was significantly affected by the louver geometry. Also recall
that an inlet vortex induced the favorable circulation, which was important to collect sand.
Disruption of the favorable circulation and the inlet vortex is shown in Figure A.40 for the
constant louver configuration at spanwise distance Z = 2.5H from the endwall relative to the
favorable circulation at Z = 1.25H.
Figure A.40. The 3D model predicted that the favorable circulation in the constant louver
configuration was disrupted due to spanwise flow in the collector at Re = 46,400, Z=1.25H (left),
Z=2.5H (right).
86
Disruption of the favorable circulation for the constant louver configuration increased
with Reynolds number. At low Reynolds numbers the circulation in the 3D collector was not
affected by spanwise flow, as shown in Figure A.41. The inlet vortex and favorable circulation
were present across the span for both louver configurations. Near the endwall, however, the
circulation was affected for the variable louver configuration (Z=0.125H) and the constant louver
configuration (Z=1.25H). The 3D predictions of collection efficiency were similar to 2D
predictions for both louver configurations at Re = 8,000 because both 2D and 3D models
predicted similar collector circulation, as shown in Figure A.41.
2D
3D
Z=0.125H
U/Uinlet
5
3D
4.5
4 Z=1.25H
3.5
3
2.5
2
1.5
1
0.5
0 3D
Z=2.5H
3D
Z=3.75H
3D
Z=4.875H
louver (left) and constant louver (right) configurations at Re = 8,000.
At high Reynolds numbers particle trajectories in the collector were influenced by the
circulation direction and the strength of the inlet vortex. Circulation was disrupted by spanwise
87
flow in the 3D collector at high Reynolds numbers, as shown in Figure A.42. At Re = 46,400 the
favorable circulation was present across the span for the variable louver configuration. Collector
circulation for the constant louver configuration, however, was disrupted far away from the wall.
The disrupted circulation directed sand towards the collector inlet at spanwise locations Z = 2.5H
and Z = 3.75H.
2D
3D
Z=0.125H
U/Uinlet
5
4.5
4
3.5 3D
3
2.5
2
Z=1.25H
1.5
1
0.5
0
3D
Z=2.5H
3D
Z=3.75H
3D
Z=4.875H
louver (left) and constant louver (right) configurations at Re = 46,400.
The favorable circulation was disrupted in the constant louver configuration and not the
variable louver configuration because the variable louver configuration had a stronger inlet
vortex than the constant louver configuration. The strong inlet vortex maintained the favorable
circulation when spanwise flow in the collector was present. The vortex strength is illustrated by
88
pathlines in Figure A.43. The pathlines exiting the collector of the variable louver configuration
were trapped in the inlet vortex. For the constant louver configuration, however, the exiting
pathlines passed through the inlet vortex and continued downstream. Pathlines were not trapped
in the inlet vortex of the constant louver configuration because the constant louver configuration
produced a weak inlet vortex relative to the variable louver configuration.
Endwall Endwall
U/Uinlet
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Figure A.43. Collector pathlines predicted by the 3D models indicated that the constant louver
configuration (left) had a weaker inlet vortex than the variable louver configuration (right).
The vortex strength was dependent on the stagnation location of the flow impinging on
the wedge face. Velocity vectors in Figure A.44 illustrate a higher stagnation location on the
wedge face for the variable louver configuration compared to the constant louver configuration.
The high stagnation location for the variable louver configuration indicated that more flow
passing over the last louver tip was interacting with the inlet vortex compared to a low stagnation
location where less flow interacted with the inlet vortex. The high stagnation location resulted in
a strong inlet vortex that was located nearer to the collector inlet compared to a low stagnation
location.
89
3D
Z=0.125H
3D
Z=1.25H
U/Uinlet
5
4.5
4
3.5
3
3D
2.5
2
Z=2.5H
1.5
1
0.5
0
3D
Z=3.75H
3D
Z=4.875H
Figure A.44. Stagnation location was higher on the wedge face for the variable louver
configuration (left) than the constant louver configuration (right) at Re = 46,400.
Collection Efficiency for Different Sand Sizes

Collection efficiency was calculated for four discrete size ranges by integrating predicted
collection efficiency results within the size range of interest. Values of collection efficiency are
reported in Figure A.45 for each of the four size intervals in comparison to the entire particle size
range (0-200µm). Collection efficiency values were predicted to be higher than measured values
for all particle size ranges. Peak collection efficiency was predicted by the 2D model for the size
range 40-63µm, indicated by the increasing and decreasing trend shown in Figure A.45.
90
50%
40% 2D CFD 2D CFD
3D CFD 3D CFD
30%
η
20%
10%
0%
0-20 20-40 38-63 0-200
µm)
Sand Size, Ds (µ
Figure A.45. Computational results over-predicted collection efficiency for each sand size
interval at Re = 46,400.
The increasing and decreasing trend of predicted collection efficiency with particle size
was produced by the effect of particle momentum on particle trajectories. Particle momentum
increased with particle size to cause particles to deviate from the direction of fluid flow.
Momentum was increased with particle size from 0-20µm to 20-40µm to improve particle
separation by the louver array. The increase in particle momentum did not affect collection of
the particles in the collector because the particles were small enough to be influenced by the
collector circulation. Large particles of increasing size ranges from 38-63µm to 63-200µm
caused increasing particle deflections from louver surfaces and walls resulting in particles
deflecting through the louver array or out of the collector. Therefore, increasing the sand sizes
reduced the collection efficiency. A similar effect of particle size on collection efficiency was
also observed for the 3D models; however, the spanwise flow in the 3D collector caused
predictions to deviate from the 2D predictions.
The effect of particle size on collection efficiency is illustrated with 2D particle
trajectories of 10µm and 30µm size particles in Figure A.46, and 55µm and 170µm particles in
Figure A.47. For small particles illustrated in Figure A.46 collection was influenced by the
91
effectiveness of the louver array to separate particles. Collection increased with particle size
because particle momentum increased such that the particles were not easily influenced by the
flow passing through the louver gaps. For example, 30µm particles were shown to only pass
through the last louver gap relative to 10µm particles that passed through louver gaps further
upstream. Large particle trajectories were influenced by deflections in the separator domain, as
shown by 55µm and 170µm trajectories in Figure A.47. Increasing the particles size (particle
momentum) resulted in reduced collection efficiency due to more particles deflecting through the
louver gaps, or deflecting from the wedge face. The primary mechanism for large particle
collection was particles depositing on the bottom of the collector wall or under the baffle.
Large particles were deposited on the bottom of the collector wall because particle momentum
was large enough such that the collector circulation did not affect particle trajectories.
louver (top) and 2D constant louver (bottom) configurations at Re = 46,400.
92
louver (top) and 2D constant louver (bottom) configurations at Re = 46,400.
The effect of particle size was also investigated for different Reynolds numbers. The 2D
predictions are shown in Figure A.48 and Figure A.49, and the 3D predictions are shown in
Figure A.50 and Figure A.51. Both variable louver and constant louver configurations resulted
in increased collection efficiency for particles sizes 40-160µm for decreasing Reynolds number.
These particle sizes were affected by Reynolds number more than other particle sizes because
they were dependent on the momentum imparted by the flow velocity to pass over the louver
gaps and into the collector. Therefore, at high Reynolds numbers 40-160µm size particles
passed into the collector, but deflected out of the collector due to high particle momentum.
Collection efficiency for 40-160µm particle sizes was high at low Reynolds numbers because the
particles had enough momentum to be separated by the louver array, but not enough momentum
to deflect out of the collector. The reduction in Reynolds number also reduced the circulation
size in the collector, which allowed particles to deposit on the bottom of the collector wall.
The effect of the spanwise flow in the 3D models affected collection predictions of the
constant louver configuration more than the variable louver configuration. For the variable
louver configuration, similar collection was predicted for varying Reynolds number between the
2D and 3D models. Predicted collection efficiency values for the 3D constant louver
configuration were significantly affected by spanwise flow relative to the 2D predictions.
93
100%
Re = 46,400
Re = 35,000
80% Re = 25,000
Re = 15,000
Re = 8,000
60%
η
40%
20%
0%
1 20 40 60 80 100 120 140 160 180 200
µm)
Particle Size (µ
size and Reynolds number for the variable louver configuration.
100%
Re = 35,000
Re = 46,400
Re = 25,000
80% Re = 15,000
Re = 8,000
60%
η
40%
20%
0%
1 20 40 60 80 100 120 140 160 180 200
µm)
Particle Size (µ
size and Reynolds number for the constant louver configuration.
94
100%
Re = 46,400
Re = 35,000
Re = 25,000
80% Re = 15,000
Re = 8,000
60%
η
40%
20%
0%
1 20 40 60 80 100 120 140 160 180 200
Particle Size (µm)
size and Reynolds number for the variable louver configuration.
100%
Re = 46,400
Re = 35,000
Re = 25,000
80% Re = 15,000
Re = 8,000
60%
η
40%
20%
0%
1 20 40 60 80 100 120 140 160 180 200
Particle Size (µm)
size and Reynolds number for the constant louver configuration.
95
The collector circulation pattern was shown in the Paper 1 and Paper 2 to be an important
flow feature, where a favorable circulation pattern was required for particle collection. The 2D
predictions of the collector circulation for both louver configurations are compared in Figure
A.52 to experimental observations for the maximum and minimum Reynolds numbers tested.
Predictions showed that the collector circulation was dependent on Reynolds number, which was
similar to experimental observations. Experimental observations showed that the size of the sand
circulation in the collector decreased with Reynolds number. Similarly, predicted circulation
zones in Figure A.52 show a dependence on Reynolds number for the constant louver
configuration. The predicted circulation of the variable louver configuration, however, was
independent of Reynolds number. Predictions revealed multiple circulation zones in the
collector that varied in size and location with Reynolds number. It is important to note that the
experimental observations were sketched from the observed flow direction of the sand in the
collector. Visual observations from the experiment did not indicate more than one circulation
zone; however, more than one circulation zone could exist and was not discernable.
Re=46,400 Re=46,400
Re=15,000 Re=8,000
Figure A.52. The collector circulation zones are independent of Reynolds number for the (left)
variable louver and dependent on Reynolds number for the (right) constant louver configuration.

In summary, pressure loss and collection efficiency for both louver configurations were
predicted to agree with most measured values. Predicted trends, however, did not agree with
experimental trends.
Pressure loss predictions agreed with measured values for Reynolds numbers greater than
25,000. Predictions at low Reynolds numbers were lower than measured pressure loss values
because predicted separation regions were independent of Reynolds number. The 2D and 3D
96
experiment matched models predicted similar pressure losses for all Reynolds numbers because
spanwise flow effects were not predicted in the louver array.
Collection efficiency predictions were within 10% of measured values for all Reynolds
numbers. The predicted trend, however, was different than the measured trend. The 2D and 3D
experiment matched models predicted similar collection for the variable louver configuration
because spanwise flow effects in the collector did not disrupt the favorable circulation. The 2D
and 3D collection predictions for the constant louver configuration did not agree for high
Reynolds numbers. Collection efficiency predicted by the 3D model decreased with increasing
Reynolds number relative to the 2D model because particle momentum increased such that
particles could leave the collector through the weak inlet vortex of the constant louver
configuration.
The collection efficiencies for specific sand sizes were over-predicted for each size
interval. The 2D experiment matched model predicted a trend of increasing and decreasing
collection efficiency with increasing particle size. By increasing the particle size (particle
momentum) from 0-20µm to 20-40µm allowed particles to be separated by the louver array and
enter the collector. Increasing particle size from 38-63µm to 63-200µm resulted in decreased
collection efficiency because the increase in momentum caused increased particle deflections in
the separator. The increased deflections caused particles to deflect through the louver array and
not be collected. The collection efficiency trend for the 3D experiment matched model increased
and decreased with increasing particle size similar to the 2D experiment matched model.
Spanwise flow effects in the collector of the 3D experiment matched model caused the collection
efficiency trend to differ from the 2D predicted trend.
References
[1] ANSYS, Inc., Version 2.4.6 2008 (ANSYS Inc.: New Hampshire)
[2] ANSYS, Inc., Version 6.3.26 2008 (ANSYS Inc.: New Hampshire)
97
Appendix B: Uncertainty Analysis
Nomenclature
Cp pressure loss coefficient
H test section height
m sand mass
n number of samples
Patm atmospheric pressure
PE precision error
Pex exit pressure
Pin inlet pressure
Q volumetric flow rate
R gas constant of air
S standard deviation
T flow temperature
tα/2,ν t-statistic
Uin test section inlet velocity
uCp bias uncertainty
u' bias error
W test section width
Greek:
ρ fluid density
η collection efficiency
Subscripts:
0 sand initially placed in sand feeder
col sand collected in the collector
chan sand deposited in the upstream channel
feed sand remaining in the feeder
in sand entering the test section, min = m0-mfeed-mman-mchan
man sand deposited in the mixing manifold
Appendix B describes in detail the method used to obtain experimental uncertainty values
for pressure loss coefficient (Cp) and sand collection efficiency (η). The overall uncertainty of η
and Cp is calculated from the root-sum-square of the bias and precision errors. Bias error was
calculated using the partial derivative method described by Moffatt [1], and precision error was
calculated from the standard deviation of a set of measurements.
98
Precision Error
Precision error was calculated from the standard deviations of a set of measurements
taken under similar test conditions (Re=46,400, m0=60g, ML=0.002). Five sets of measurements
were made, where each set consisted of three tests. From each set of measurements, the average
pressure loss coefficient (Cp) and collection efficiency (η) was calculated to find the standard
deviation of the five sets of measurements. The precision error was calculated from the standard
deviation of a 95% confidence interval on Student’s t-distribution using Equation B.1. Precision
error is presented along with bias error in Table B.1and Table B.2 for Cp, and Table B.3 and
Table B.4 for η.
S
PE = t α / 2, υ (B.1)
n
Pressure Loss Coefficient Bias Error

The pressure loss coefficient was defined by the static pressure loss across the separator
and the inlet dynamic pressure, given in Equation B.2.
Cp =
(P in − Pex )static
(B.2)
(P )
in dyn
The inlet dynamic pressure was defined by the flow density and velocity at the test
section inlet, given in Equation B.3.

 P  Q  2
(Pin )dyn = 0 .5 atm 
2
 HW 
= 0 .5 ρ U in (B.3)
 RT
 flow 
Substitution of Equation B.3 into Equation B.2 results in the form used to calculate the
bias error of Cp, shown in Equation B.4.
(Pin − Pex )static
Cp = (B.4)
 P  Q  2
0.5 atm 
 RT  HW 
 flow 
The bias uncertainty of Cp was found from the root-sum-square of the partial derivatives
of Cp with respect to each measured value multiplied by the uncertainty of the measured value.
99
2 2 2 2
 ∂ Cp   ∂ Cp   ∂ Cp   ∂ Cp 
 u'  +  +  
Patm  +

∂P P   ∂ P u' Pex  ∂P u'

u'Tflow

  ∂ PTflow
in
 in   ex   atm 
u Cp = (B.5)
2 2 2
 ∂ Cp   ∂ Cp   ∂ Cp 
+ u'Q  +  u' H  +  u' W  .
 ∂Q   ∂H   ∂W 
     
The flow rate, Q, was determined by measuring the pressure loss in a calibrated venturi
meter. The bias uncertainty of the flow rate measurement was made up of the uncertainty from
the pressure measurement as well as the uncertainty of the venturi meter calibration, given by the
manufacturer. The calibration curve obtained from the manufacturer is given in Equation B.6,
where the venturi meter had a maximum uncertainty of 0.75% of the full scale flow, u'Q.
4 3 2
Q = −10.145∆Pf + 99.372∆Pf − 349.06∆Pf + 664.29∆Pf + 52.559 (B.6)
The bias uncertainty of the flow rate was found using the partial derivative method.
2
 ∂Q 
u Q =  u'∆Pf  ( )
 + u'Q 2 . (B.7)
 ∂ ∆Pf 
Table B.1 and Table B.2 show uncertainty values calculated at the high and low Reynolds
numbers tested for the constant louver configuration. Note that the inlet (Pin) and exit (Pex)
pressures are larger in Table B.2 compared to Table B.1 because a different pressure reference
was used than at high Reynolds numbers because the pressure loss from the original reference
location was out of the measurement range of the available pressure transducers.
Table B.1. Constant Louver Uncertainty Values for Cp at Re = 46,400

Variable Value Bias Precision Uncertainty % of Value uCp
Cp 12.1 0.308 0.078 0.318 2.6% 0.318
Pin (Pa) 18.0 0.1 0.1 0.56% 5.42x10-7
Pex (Pa) 1661 4.91 24.9 1.50% 0.034
Patm (Pa) 97675 22 22 0.02% 7.49x10-6
Tflow (°C) 27.4 1.0 1.0 3.65% 5.0x10-4
Q (m3/sec) 0.360 0.003 0.003 0.83% 0.050
W (mm) 483 0.588 1.59 0.33% 0.006
H (mm) 48.26 0.127 0.127 0.26% 0.004
100
Table B.2. Constant Louver Uncertainty Values for Cp Re = 8,000
Variable Value Bias Precision Unc % uCp
ertainty of Value
Cp 20.0 2.15 0.078 2.14 10.73% 2.147
Pin (Pa) 66.5 0.623 0.623 0.94% 0.027
Pex (Pa) 141 0.623 0.623 0.44% 0.027
Patm (Pa) 96986 22.0 22.0 0.02% 1.98x10-5
Tflow (°C) 22.0 1.0 1.0 4.55% 0.001
Q (m3/sec) 0.060 0.003 0.023 5.00% 4.520
W (mm) 483 1.59 1.59 0.33% 0.017
H (mm) 48.3 0.127 0.127 0.26% 0.011
Collection Efficiency Bias Error

The precision and bias errors of the collection efficiency measurement were calculated
similar to the pressure loss coefficient. Collection efficiency was calculated from the amount of
sand remaining in the collector, and the amount of sand that entered the test section. The amount
of sand entering the test section was calculated by subtracting from the initial sand amount (m0)
the sand that was retrieved from the sand feeder (mfeed), sand manifold (mman), and the upstream
channel (mchan) after each test, as shown in Equation B.8.
m col
η= (B.8)
m 0 − m feed − m man − m chan
The bias error of collection efficiency was calculated using the partial derivative method
shown in Equation B.9. Precision error of collection efficiency is reported along with bias error
and total uncertainty in Table B.3 and Table B.4.
2 2 2 2 2
 ∂η   ∂η   ∂η   ∂η   ∂η 
u η =  u′m  +  u′m  +  u ′m  +  u′m  +  u′m  (B.9)
 ∂m col   ∂m 0   ∂m feed   ∂m man   ∂m chan 
Table B.3. Constant Louver Uncertainty Values for η at Re = 46,400

Variable Value Bias Precision Uncertainty % of Value uη
η 0.08 0.013 0.026 0.029 36% 0.029
mcol (g) 3 0.5 0.5 16.7% 1.6 x 10-4
m0 (g) 60 0.5 0.5 0.8% 9.0 x 10-7
mfeed (g) 19 0.5 0.5 2.6% 9.0 x 10-7
mman (g) 1 0.5 0.5 50% 9.0 x 10-7
mchan (g) 0 0.5 0.5 9.0 x 10-7
101
Table B.4. Constant Louver Uncertainty Values for η at Re = 8,000
Variable Value Bias Precision Uncertainty % of Value uη
η 0.35 0.022 0.026 0.034 10.0% 0.034
mcol (g) 10 0.5 0.5 5.0% 3.2 x 10-4
m0 (g) 60 0.5 0.5 0.8% 4.1 x 10-5
mfeed (g) 24 0.5 0.5 2.1% 4.1 x 10-5
mman (g) 4 0.5 0.5 12.5% 4.1 x 10-5
mchan (g) 4 0.5 0.5 12.5% 4.1 x 10-5
References
[1] Moffat, R.J., 1988, “Describing the Uncertainties in Experimental Results,” Experimental
Thermal and Fluid Science, 1, pp. 3-17.
102
Appendix C: Description of the Test Facility
The measurement method used to take pressure loss and sand collection measurements
was constant for each test. After each test, sections of the test facility were disassembled to
remove deposited sand. The disassembled sections consisted of the test section, sand injection
system, the upstream channel, and the filter box. Appendix C describes the measurement
method and the disassembled sections of the test facility.
Prior to each test, the injected sand was heated at 100°C for at least three hours to remove
moisture and weighed before being placed in the particle feeder. The filters in the filter box were
weighed before and after each test, and cleaned between tests by blowing compressed air through
the filter media. Each test began after the sand injection system was sealed and the barrier filters
were in place. Pressure measurements were taken after a steady state flow condition was
obtained and before sand was injected. The steady state condition was achieved when a constant
flow stagnation temperature was measured. Flow temperature was measured from a single type
E thermocouple placed one pipe diameter upstream of the venturi meter.
Pressure loss across the separator was measured from seven static pressure taps equally
spaced across the span of the test section inlet and exit. Pressure measurements were averaged
over 20 seconds taken at a rate of 1 kHz with reference to a static pressure tap located 13H
upstream of the test section inlet. The static pressure loss measurement at each upstream and
downstream tap was normalized across the test section span to confirm flow uniformity. Flow
uniformity is expressed as a percentage pressure difference between the measured pressure loss
at a static pressure tap and the spanwise averaged pressure loss. For example, the upstream flow
uniformity was confirmed by measuring the pressure loss at each static pressure tap located at
the test section inlet. The percentage pressure difference was found by taking the difference
between each static tap measurement and the spanwise average of the static tap measurements,
and then normalizing the difference to the dynamic pressure at the test section inlet. Percentage
pressure differences within ±5% indicated uniform spanwise flow. Flow uniformity was
affected by the upstream and downstream channel transitions.
Upstream flow uniformity required that a 48.3mm x 178mm (1.9″ x 7″) splash plate be
placed at the inlet to the upstream channel, as shown in Figure C.1. Flow uniformity was
dependent on the location of the splash plate, where uniform flow was achieved when the splash
plate center was located 51mm (2″) left of the channel midspan, as shown in Figure C.2. The
103
downstream flow uniformity measured at the test section exit was dependent on the separator and
the distance from the downstream transition. The flow uniformity was made symmetric about
the channel midspan by reinforcing the separator relative to the test section at locations
symmetric about the test section midspan. Pressure measurements relative to the spanwise
average across the test section was reduced by as much as 10% by placing a downstream channel
of length 18H between the test section and the transition, as shown in Figure C.2.
51mm
(2″)
Splashplate
178mm (7″) 48.3mm (1.9″)
Figure C.1. Upstream flow uniformity was achieved using a 48.3mm x 178mm (1.9″ x 7″)
splash plate with center located 51mm (2″) left of the channel midspan.
20
Upstream Downstream
15 No splash plate No channel
Splash plate Channel
10
5
% 0
Pressure
Difference -5
-10
-15
-20
-0.5 -0.25 0 0.25 0.5
Spanwise location, Z/Span
Figure C.2. A splash plate was placed 51mm (2″) left of midspan at the upstream channel inlet
to produce uniform flow at the test section inlet.
After flow uniformity was confirmed, compressed air was fed into the mixing manifold at
a pressure slightly higher than the channel pressure. Then, sand was fed from the particle feeder
at a constant feed rate until all of the sand was conveyed out of the feeder. All tests were run for
104
a sufficient time after sand feeding stopped to confirm that the sand deposited in the collector
was steady. After each test, the test facility was disassembled and all deposited sand was
removed and measured.
Sand was retrieved from the sand injection system, the upstream channel, the test section,
and the filter box. The sand injection system consisted of a sand feeder and a mixing manifold.
The sand feeder was modified from use as plastic metering equipment by way of a cylindrical
hopper and a sand agitator that prevent sand from bridging in the hopper, as shown in Figure
C.3b. The sand agitator rotated to keep the sand in the hopper free-flowing as the sand was
conveyed by the auger screw. The sand feeder used an auger screw to convey sand from the
hopper, as shown in Figure C.3c. The feed rate was determined by the diameter and number of
threads of the auger screw. The feed rate of the auger screw was calibrated by measuring the
amount of sand conveyed by the feeder in a given amount of time. The calibration method
consisted of five different speed settings, where three tests were conducted at each speed setting.
The sand feeder was disassembled and cleaned between each calibration test. The calibration
curve is shown in Figure C.4 with maximum feed rate of 7 cm3/sec.
Sand that remained in the feeder after a test was located along the auger screw in the
auger screw insert. Figure C.3c shows the auger screw and screw insert that were removed after
each test to retrieve the deposited sand in the feeder. Sand was conveyed by the auger screw into
the sand manifold. The sand manifold in Figure C.5distributed sand from the feeder across the
span of the upstream channel using compressed air injected at three ports in the manifold. The
compressed air was throttled to maintain slightly higher pressure in the manifold relative to the
upstream channel. The manifold pressure was measured relative to the upstream channel using a
static pressure tap located in the manifold. The Sand was removed from the manifold by
separating the manifold from the sand feeder and the upstream channel, and then removing the
manifold lid to retrieve the sand adhered to the walls. Sand deposited in the upstream channel
was localized to the length of channel between the sand injection slot and the test section. The
top of the upstream channel within the region of sand deposition was removable for sand
retrieval, as shown in Figure C.6. The upstream channel design height H was confirmed each
time the channel was reassembled.
105
a) b)
Sand agitator
c)
Screw insert
Auger screw
Figure C.3. The a) feeder conveyed sand from the b) hopper to the mixing manifold by way of
an b) auger screw.
10
Calibration data
8 y = -0.37451 + 0.069762x R= 0.99714
6
Sand Flow Rate
(cm3/sec) 4
0
0 20 40 60 80 100
Controller Speed
Figure C.4. The feed rate of the sand feeder was calibrated over the controller speed range.
106
Figure C.5. The sand manifold used three compressed air ports (not shown) to mix the sand
before passing into the upstream channel.
Figure C.6. A section of the upstream channel between the test section and sand injection slot
was removable for sand retrieval.
The test section was constructed of transparent polycarbonate to observe sand passing
through the separator. Deposited sand in the test section was located in the collector and on the
louver surfaces. A negligible amount of sand deposited on the walls of the test section. Sand
was retrieved from the separator by way of a removable test section lid, as shown in Figure C.7.
The removable lid also allowed the separator to be replaced by sliding the separator in the guide
107
trenches cut into the test section endwalls. The guide tracks of the separator slid into the
trenches in the test section to maintain constant separator position and orientation to the channel
flow. The guide tracks were incorporated into the separator design, as shown in Figure C.8. The
guide tracks and structural supports were added to the separator design for experimental testing.
Endcaps were also added to the separator at the collector ends to prevent leakage flow between
the collector and the test section endwalls.
Removable lid
Separator slides out
Guide trench
Figure C.7. The test section had a removable lid to allow the interchanging of separator
configurations.
Louver support
Louver strut
Guide track Endcap
Figure C.8. The tested separators included structural reinforcements, guide tracks, and endcaps
(variable louver configuration shown).
108
Deposited sand in the filter box was captured from the outgoing air of the test facility by
three filters placed in series, as shown in Figure C.9. The filter sizes were 25.4 x 305 x 508mm
(1″ x 12″ x 20″) with efficiency ratings of MERV 11 and MERV 12, however. The middle filter
in the series had an efficiency rating of MERV 11, and the two other filters in series had a rating
of MERV 12. The MERV (Minimum Efficiency Reporting Value) rating is a measure of the
efficiency of a dust filter ranging from 1 to16, where a high rating indicates good particle
filtration. An efficiency of MERV 11 is rated to filter 90% of particle sizes 1-10µm, and MERV
12 is rated to filter 93% of 1-10µm particles. Both filters types were stated by the manufacturer
to remove particles as small as 0.3µm.
The filters were weighed before and after each test to determine the amount of sand
trapped by the filters. All of the filters were cleaned prior to each test by using compressed air to
blow sand out. The filters were cleaned until the filter weight was returned to a value similar to
an unused filter. Filters could be used as many as forty times depending on the location of the
filter in the series.
Figure C.9. The filter box removed sand that was not collected by the separator using three
filters placed in series.
109
Appendix D: Engineering Drawings
110
111
112
113
114
115
View publication stats

Grant Musgrove Thesis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Grant Musgrove Thesis

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Computational Design of a Louver Particle Separator for Gas Turbine Engines

Conference Paper · June 2009

Grant Musgrove Karen A. Thole

SEE PROFILE SEE PROFILE

Eric Andrew Grover

Pin Fin Heat Exchangers View project

Heat Exchanger Development View project

The user has requested enhancement of the downloaded file.

The Graduate School

Department of Mechanical and Nuclear Engineering

COMPUTATIONAL PREDICTIONS AND

EXPERIMENTAL MEASUREMENTS OF THE PERFORMANCE OF A

LOUVER PARTICLE SEPARATOR FOR USE IN GAS TURBINE ENGINES

© 2009 Grant O. Musgrove

Submitted in Partial Fulfillment

H. Joseph Sommer III

*Signatures are on file in the Graduate School.

Appendix A: Computational Study of the Louver Separator........................................................ 57

Appendix B: Uncertainty Analysis ............................................................................................... 98

Appendix C: Description of the Test Facility ............................................................................. 103

Appendix D: Engineering Drawings ........................................................................................... 110

Table 1.2. Parameter Range of Values ........................................................................................... 7

Table 1.3. Shape Factor Effect on Overall Collection ................................................................. 12

Table 2.1. Area Ratio and Louver Geometry ............................................................................... 35

Table 2.2. Effect of Collector Modifications ............................................................................... 36

Table A.1. 2D and 3D Computational Models............................................................................. 58

Table A.2. Boundary Conditions for Each Computational Model ............................................... 61

Table A.3. Operating Conditions for Each Computational Model .............................................. 61

Table A.4. Cell Skewness for Variable Louver Configuration .................................................... 62

Table A.5. Scaled Particle Sizes................................................................................................... 80

Review of Relevant Literature

Combustor Vane Blade Vane

Table 1.2. Parameter Range of Values

60% Shape Factor

Table 1.3. Shape Factor Effect on Overall Collection

Shape Factor 1.0 0.82 0.70 0.50

Selection of the Area Ratio

Design of the Louvers

configuration A 30°, straight louvers

Design of the Collector

[23] Particle size distribution provided by Pratt & Whitney, 2007.

AR area ratio AR = δ(# Louvers)/H

Experimental Facility and Methodology

(a) Venturi meter

Louver Separator Geometry

Table 2.2. Effect of Collector Modifications

Wedge and baffle η = 20%

Measurement Uncertainty and Repeatability

Missing sand Louver Surfaces

Pressure Loss Results

Reynolds Number Effects on Collection Efficiency

To compliment the measured collection efficiency, photographs were taken of sand

Missing sand Louver Surfaces

Sand Size Effects on Collection Efficiency

Sand accumulation was observed to be dependent on particle size. Accumulation in the

[16] Powder Technology, Inc., 2009, Burnsville, MN.

2D engine 2D experiment matched 3D experiment matched

Reynolds number 135,500 46,400 46,400

Mass flow inlet Design space Pressure exit

Velocity inlet Design space H Outflow exit

Inlet Exit Walls Energy Equation

Table A.3. Operating Conditions for Each Computational Model