2011 Book UnsteadyEffectsOfShockWaveIndu

114 Notes on Numerical Fluid Mechanics and Multidisciplinary Design (NNFM)
Editors
W. Schröder/Aachen
B.J. Boersma/Delft
K. Fujii/Kanagawa
W. Haase/München
M.A. Leschziner/London
J. Periaux/Paris
S. Pirozzoli/Rome
A. Rizzi/Stockholm
B. Roux/Marseille
Y. Shokin/Novosibirsk
Unsteady Effects of Shock
Wave Induced Separation
Piotr Doerffer, Charles Hirsch,

Jean-Paul Dussauge, Holger Babinsky,
George N. Barakos (Eds.)
ABC
Piotr Doerffer Holger Babinsky
Institute of Fluid Flow Machinery Engineering Department
Polish Academy of Sciences University of Cambridge
Fiszera 14 Cambridge, CB2 1PZ, GB
PL 80-952 Gdansk United Kingdom
Poland
E-mail: doerffer@karol.imp.gda.pl George N. Barakos
CFD Laboratory
Charles Hirsch Department of Engineering
President The University of Liverpool
Numerical Mechanics Application Room UG31, Harrison Hughes Bld.
International The Quadrangle - Liverpool L693GH
5 Avenue Franklin Roosevelt, United Kingdom
1050 Brussels
Belgium
Jean-Paul Dussauge
Directeur de Recherche au CNRS
IUSTI, Supersonic Group
5, Rue Enrico Fermi
13453 Marseille Cedex 13
France
E-mail: jean-paul.dussage@
polytech.univ-mrs.fr
ISBN 978-3-642-03003-1 e-ISBN 978-3-642-03004-8

DOI 10.1007/978-3-642-03004-8
Notes on Numerical Fluid Mechanics

and Multidisciplinary Design ISSN 1612-2909
Library of Congress Control Number: 2010937341

c 2010 Springer-Verlag Berlin Heidelberg
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,
reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication
or parts thereof is permitted only under the provisions of the German Copyright Law of September 9,
1965, in its current version, and permission for use must always be obtained from Springer. Violations
are liable for prosecution under the German Copyright Law.
The use of general descriptive names, registered names, trademarks, etc. in this publication does not
imply, even in the absence of a specific statement, that such names are exempt from the relevant protective
laws and regulations and therefore free for general use.
Typeset & Cover Design: Scientific Publishing Services Pvt. Ltd., Chennai, India.
Printed on acid-free paper

543210
springer.com
NNFM Editor Addresses
Prof. Dr. Wolfgang Schröder Prof. Dr. Sergio Pirozzoli

(General Editor) Università di Roma “La Sapienza”
RWTH Aachen Dipartimento di Meccanica e Aeronautica
Lehrstuhl für Strömungslehre und Via Eudossiana 18
Aerodynamisches Institut 00184, Roma, Italy
Wüllnerstr. 5a E-mail: sergio.pirozzoli@uniroma1.it
52062 Aachen
Germany Prof. Dr. Jacques Periaux
E-mail: office@aia.rwth-aachen.de 38, Boulevard de Reuilly
F-75012 Paris
Prof. Dr. Kozo Fujii France
Space Transportation Research Division E-mail: jperiaux@free.fr
The Institute of Space
and Astronautical Science Prof. Dr. Arthur Rizzi
3-1-1, Yoshinodai, Sagamihara Department of Aeronautics
Kanagawa, 229-8510 KTH Royal Institute of Technology
Japan Teknikringen 8
E-mail: fujii@flab.eng.isas.jaxa.jp S-10044 Stockholm
Sweden
Dr. Werner Haase E-mail: rizzi@aero.kth.se
Höhenkirchener Str. 19d
D-85662 Hohenbrunn Dr. Bernard Roux
Germany L3M – IMT La Jetée
E-mail: office@haa.se Technopole de Chateau-Gombert
F-13451 Marseille Cedex 20
Prof. Dr. Ernst Heinrich Hirschel France
(Former General Editor) E-mail: broux@l3m.univ-mrs.fr
Herzog-Heinrich-Weg 6
D-85604 Zorneding Prof. Dr. Yurii I. Shokin
Germany Siberian Branch of the
E-mail: e.h.hirschel@t-online.de Russian Academy of Sciences
Institute of Computational
Prof. Dr. Ir. Bendiks Jan Boersma Technologies
Chair of Energytechnology Ac. Lavrentyeva Ave. 6
Delft University of Technology 630090 Novosibirsk
Leeghwaterstraat 44 Russia
2628 CA Delft E-mail: shokin@ict.nsc.ru
The Netherlands
E-mail: b.j.boersma@tudelft.nl
Prof. Dr. Michael A. Leschziner

Imperial College of Science
Technology and Medicine
Aeronautics Department
Prince Consort Road
London SW7 2BY
U.K.
E-mail: mike.leschziner@ic.ac.uk
Notes on Numerical Fluid Mechanics and Multidisciplinary Design
Available Volumes
Volume 114: Piotr Doerffer, Charles Hirsch, Jean-Paul Dussauge, Holger Babinsky, and George N.
Barakos (eds.): Unsteady Effects of Shock Wave Induced Separation. ISBN 978-3-642-03003-1
Volume 112: Andreas Dillmann, Gerd Heller, Michael Klaas, Hans-Peter Kreplin, Wolfgang Nitsche,
and Wolfgang Schröder (eds.): New Results in Numerical and Experimental Fluid Mechanics VII –
Contributions to the 16th STAB/DGLR Symposium Aachen, Germany 2008. ISBN 978-3-642-14242-0
Volume 111: Shia-Hui Peng, Piotr Doerffer, and Werner Haase (eds.): Progress in Hybrid RANS-LES
Modelling – Papers Contributed to the 3rd Symposium on Hybrid RANS-LES Methods, Gdansk, Poland,
June 2009. ISBN 978-3-642-14167-6
Volume 110: Michel Deville, Thien-Hiep Lê, and Pierre Sagaut (eds.): Turbulence and Interactions –
Proceedings the TI 2009 Conference. ISBN 978-3-642-14138-6
Volume 109: Wolfgang Schröder (ed.): Summary of Flow Modulation and Fluid-Structure Interaction
Findings – Results of the Collaborative Research Center SFB 401 at the RWTH Aachen University,
Aachen, Germany, 1997-2008. ISBN 978-3-642-04087-0
Volume 108: Rudibert King (ed.): Active Flow Control II – Papers Contributed to the Conference “Active
Flow Control II 2010”, Berlin, Germany, May 26–28, 2010. ISBN 978-3-642-11734-3
Volume 107: Norbert Kroll, Dieter Schwamborn, Klaus Becker, Herbert Rieger, Frank Thiele (eds.):
MEGADESIGN and MegaOpt – German Initiatives for Aerodynamic Simulation and Optimization in
Aircraft Design. ISBN 978-3-642-04092-4
Volume 106: Wolfgang Nitsche, Christoph Dobriloff (eds.): Imaging Measurement Methods for Flow
Analysis - Results of the DFG Priority Programme 1147 “Imaging Measurement Methods for Flow
Analysis” 2003–2009. ISBN 978-3-642-01105-4
Volume 105: Michel Deville, Thien-Hiep Lê, Pierre Sagaut (eds.): Turbulence and Interactions - Keynote
Lectures of the TI 2006 Conference. ISBN 978-3-642-00261-8
Volume 104: Christophe Brun, Daniel Juvé, Michael Manhart, Claus-Dieter Munz: Numerical Simula-
tion of Turbulent Flows and Noise Generation - Results of the DFG/CNRS Research Groups FOR 507
and FOR 508. ISBN 978-3-540-89955-6
Volume 103: Werner Haase, Marianna Braza, Alistair Revell (eds.): DESider – A European Effort on
Hybrid RANS-LES Modelling - Results of the European-Union Funded Project, 2004–2007. ISBN 978-
3-540-92772-3
Volume 102: Rolf Radespiel, Cord-Christian Rossow, Benjamin Winfried Brinkmann (eds.):
Hermann Schlichting – 100 Years - Scientific Colloquium Celebrating the Anniversary of His Birthday,
Braunschweig, Germany 2007. ISBN 978-3-540-95997-7
Volume 101: Egon Krause, Yurii I. Shokin, Michael Resch, Nina Shokina (eds.): Computational Sci-
ence and High Performance Computing III - The 3rd Russian-German Advanced Research Workshop,
Novosibirsk, Russia, 23–27 July 2007. ISBN 978-3-540-69008-5
Volume 100: Ernst Heinrich Hirschel, Egon Krause (eds.): 100 Volumes of ’Notes on Numerical Fluid
Mechanics’ - 40 Years of Numerical Fluid Mechanics and Aerodynamics in Retrospect. ISBN 978-3-
540-70804-9
Volume 99: Burkhard Schulte-Werning, David Thompson, Pierre-Etienne Gautier, Carl Hanson, Brian
Hemsworth, James Nelson, Tatsuo Maeda, Paul de Vos (eds.): Noise and Vibration Mitigation for Rail
Transportation Systems - Proceedings of the 9th International Workshop on Railway Noise, Munich,
Germany, 4–8 September 2007. ISBN 978-3-540-74892-2
Volume 98: Ali Gülhan (ed.): RESPACE – Key Technologies for Reusable Space Systems - Results
of a Virtual Institute Programme of the German Helmholtz-Association, 2003–2007. ISBN 978-3-540-
77818-9
Volume 97: Shia-Hui Peng, Werner Haase (eds.): Advances in Hybrid RANS-LES Modelling - Papers
contributed to the 2007 Symposium of Hybrid RANS-LES Methods, Corfu, Greece, 17–18 June 2007.
ISBN 978-3-540-77813-4
Volume 96: C. Tropea, S. Jakirlic, H.-J. Heinemann, R. Henke, H. Hönlinger (eds.): New Results in
Numerical and Experimental Fluid Mechanics VI - Contributions to the 15th STAB/DGLR Symposium
Darmstadt, Germany, 2006. ISBN 978-3-540-74458-0
Volume 95: R. King (ed.): Active Flow Control - Papers contributed to the Conference “Active Flow
Control 2006”, Berlin, Germany, September 27 to 29, 2006. ISBN 978-3-540-71438-5
Volume 94: W. Haase, B. Aupoix, U. Bunge, D. Schwamborn (eds.): FLOMANIA - A European Ini-
tiative on Flow Physics Modelling - Results of the European-Union funded project 2002 - 2004. ISBN
978-3-540-28786-5
Volume 93: Yu. Shokin, M. Resch, N. Danaev, M. Orunkhanov, N. Shokina (eds.): Advances in High
Performance Computing and Computational Sciences - The Ith Khazakh-German Advanced Research
Workshop, Almaty, Kazakhstan, September 25 to October 1, 2005. ISBN 978-3-540-33864-2
Volume 92: H.J. Rath, C. Holze, H.-J. Heinemann, R. Henke, H. Hönlinger (eds.): New Results in Nu-
merical and Experimental Fluid Mechanics V - Contributions to the 14th STAB/DGLR Symposium
Bremen, Germany 2004. ISBN 978-3-540-33286-2
Volume 91: E. Krause, Yu. Shokin, M. Resch, N. Shokina (eds.): Computational Science and High Per-
formance Computing II - The 2nd Russian-German Advanced Research Workshop, Stuttgart, Germany,
March 14 to 16, 2005. ISBN 978-3-540-31767-8
Volume 87: Ch. Breitsamter, B. Laschka, H.-J. Heinemann, R. Hilbig (eds.): New Results in Numerical
and Experimental Fluid Mechanics IV. ISBN 978-3-540-20258-5
Volume 86: S. Wagner, M. Kloker, U. Rist (eds.): Recent Results in Laminar-Turbulent Transition -
Selected numerical and experimental contributions from the DFG priority programme ’Transition’ in
Germany. ISBN 978-3-540-40490-3
Volume 85: N.G. Barton, J. Periaux (eds.): Coupling of Fluids, Structures and Waves in Aeronautics -
Proceedings of a French-Australian Workshop in Melbourne, Australia 3-6 December 2001. ISBN 978-
3-540-40222-0
Volume 83: L. Davidson, D. Cokljat, J. Fröhlich, M.A. Leschziner, C. Mellen, W. Rodi (eds.): LESFOIL:
Large Eddy Simulation of Flow around a High Lift Airfoil - Results of the Project LESFOIL supported
by the European Union 1998 - 2001. ISBN 978-3-540-00533-9
Volume 82: E.H. Hirschel (ed.): Numerical Flow Simulation III - CNRS-DFG Collaborative Research
Programme, Results 2000-2002. ISBN 978-3-540-44130-4
Volume 81: W. Haase, V. Selmin, B. Winzell (eds.): Progress in Computational Flow Structure Inter-
action - Results of the Project UNSI, supported by the European Union 1998-2000. ISBN 978-3-540-
43902-8
Volume 80: E. Stanewsky, J. Delery, J. Fulker, P. de Matteis (eds.): Drag Reduction by Shock and
Boundary Layer Control - Results of the Project EUROSHOCK II, supported by the European Union
1996-1999. ISBN 978-3-540-43317-0
Volume 79: B. Schulte-Werning, R. Gregoire, A. Malfatti, G. Matschke (eds.): TRANSAERO - A Euro-

pean Initiative on Transient Aerodynamics for Railway System Optimisation. ISBN 978-3-540-43316-3
Volume 78: M. Hafez, K. Morinishi, J. Periaux (eds.): Computational Fluid Dynamics for the 21st Cen-
tury. Proceedings of a Symposium Honoring Prof. Satofuka on the Occasion of his 60th Birthday, Kyoto,
Japan, 15-17 July 2000. ISBN 978-3-540-42053-8
Contents
1 Introduction – UFAST Project Overview . . . . . . . . . . . . . . . . 1

1.1 Project Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Project Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 List of Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Project Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5 Project Organisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Part I Transonic Interactions
2 Bump at a Wall (George Barakos) . . . . . . . . . . . . . . . . . . . . . . . 13

2.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Introduction – The QUB Tunnel Setup . . . . . . . . . . . . . . . . . . . 13
2.3 Contoured Upper Wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4 CFD Tools and Mesh Generation . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.6 Experimental Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.7 URANS Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.8 Flow Control Attempts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.9 LES and OES Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.10 Summary and Suggestions for Future Work . . . . . . . . . . . . . . . 51
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3 Biconvex Aerofoil (Stefan Leicher) . . . . . . . . . . . . . . . . . . . . . . . 55

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.2 Buffeting Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3 Flow Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.4 Comparisons with Experiments and Cross Plotting . . . . . . . . 90
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
X Contents
4 NACA0012 with Aileron (Marianna Braza) . . . . . . . . . . . . . . 101

4.1 The IoA Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.2 Numerical Simulation of the IoA Test-Case . . . . . . . . . . . . . . . 114
4.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Part II Nozzle Flows
5 Nozzle Forced Shock Oscillations with Wall Bump

(Reynald Bur) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.2 Presentation of the Experiments . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.3 Numerical Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
5.4 Comparison between Experimental and Numerical
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
6 Nozzle Forced Shock Oscillations (Holger Babinsky) . . . . . 163

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.2 Experimental Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.3 Numerical Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6.4 Flow Case Results: Steady Tests . . . . . . . . . . . . . . . . . . . . . . . . . 164
6.5 Flow Case Results Part II: Unsteady Tests . . . . . . . . . . . . . . . . 168
6.6 Comparison with CFD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
6.7 Other Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.8 Conclusions and Further Work . . . . . . . . . . . . . . . . . . . . . . . . . . 181
7 Natural Shock Unsteadiness in Nozzle

(Piotr Doerffer) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
7.2 Basic Flow Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
7.3 Flow Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
7.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Part III Shock Reflection
8 Oblique Shock Reflection at M = 1.7

(Sergio Pirozzoli) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
8.1 Presentation of Flow Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
8.2 Experimental Methodology (Source TUD
Deliverable 2.3.10) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
8.3 Overview of RANS Simulations . . . . . . . . . . . . . . . . . . . . . . . . . 228
Contents XI
8.4 Overview of LES Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . 238

8.5 Comparison between Experiment and CFD . . . . . . . . . . . . . . . 242
8.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
9 Oblique Shock Reflection at M = 2.0 (Neil Sandham) . . . 263

9.1 Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
9.2 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
9.3 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
9.4 Description of the Flow Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
9.5 Brief Summary of Methods Used . . . . . . . . . . . . . . . . . . . . . . . . 270
9.6 Steady Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
9.7 Unsteady Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
9.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
10 Oblique Shock Reflection at M = 2.25 (Eric Garnier) . . . . 287

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
10.2 Flow Case Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
10.3 Important Flow Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
10.4 Numerical Investigations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
10.5 Controlled Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
10.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
Part IV Summary by Work Packages
11 WP-2 Basic Experiments (Jean-Paul Dussauge) . . . . . . . . . 315

11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
11.2 An Overview of the Main Results of WP2 . . . . . . . . . . . . . . . . 315
11.3 Lessons Learned and Open Issues . . . . . . . . . . . . . . . . . . . . . . . . 317
12 WP-3 Flow Control Application (Holger Babinsky) . . . . . . 321

12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
12.2 Main Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
12.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
13 WP-4 RANS/URANS Simulations (Charles Hirsch) . . . . . 327

13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
13.2 Transonic Flow Around a Profile . . . . . . . . . . . . . . . . . . . . . . . . 327
13.3 The Normal Shock Experiments . . . . . . . . . . . . . . . . . . . . . . . . . 329
13.4 Oblique Shock Reflection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
13.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
XII Contents
14 WP-5 LES and Hybrid RANS/LES (George Barakos) . . . 339

14.1 Motivation, Objectives and Work Share . . . . . . . . . . . . . . . . . . 339
14.2 Summary of Observations – Transonic Flow Cases . . . . . . . . . 340
14.3 Summary of Observations – Normal Shock Cases . . . . . . . . . . 342
14.4 Summary of Observations – Reflected Shock Cases . . . . . . . . . 345
14.5 Summary of Conclusions and Suggestions for
Further Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
From the Coordinator
One of the main goals of the UFAST project was the creation of a compre-
hensive Data Bank on shock wave boundary layer interaction, and one which
for the first time includes measurements of the unsteady effects. The analysis
of the flow control devices also includes the unsteady characteristics of the
interaction.
This book provides a comparison between experimental results and cor-
responding numerical simulations. Numerical simulations were carried out
with the newest methods and their usefulness to these very complicated flow
cases is discussed. Range of applicability of RANS, URANS, LES and Hybrid
methods gathers some new light in the presented here analysis. The experi-
mental results alone are assembled in a separate book: “UFAST Experiments
– Data Bank”, IMP PAN Publishers, ISBN 978-83-88237-46-1, 2009.
Piotr Doerffer
Partners Responsible for the Experiments
Flow
case Partner Name
Emmanuel Benard
1.1 QUB Juliana Early, Jui-Che Huang
Catalin Nae
1.2 INCAS Florin Munteanu
Wojciech Kania
1.3 IoA Marek Miller, Andrzej Krzysiak
ONERA Reynald Bur
2.1 (DAFE) Didier Coponet
Holger Babinsky
2.2 UCAM-DENG Paul J.K. Bruce
Piotr Doerffer
2.3 IMP PAN Ryszard Szwaba
Bas van Oudheusden
3.1 TUD Louis J. Souverein, Ray Humble Fulvio Scarno
Anatoly Maslov
3.2 ITAM Andrey Sidorenko, Pavel Polivanov
Jean-Paul Dussauge
3.3 IUSTI Pierre Dupont, Jean-François Debiève
Sébastien Piponniau, Louis. J. Souverein
Chapter 1
Introduction – UFAST Project
Overview
1.1 Project Summary
Aligned with the needs of the aeronautics industry, the general aim of the
UFAST project was to foster experimental and theoretical research in the
highly non-linear area of unsteady shock wave boundary layer interaction
(SWBLI). Although previous EU projects concentrated on transonic/
supersonic flows, they did not examine unsteady shock wave boundary layer
interaction. Important developments in experimental and numerical methods
in recent years have now made such research possible.
The main cases of study, shock waves on wings/profiles, nozzle flows and
inlet flows, provide a sound basis for open questions posed by the aeronautics
industry and can easily be exploited to enable more complex applications to
be tackled. In addition to basic flow configurations, control methods (syn-
thetic jets, electro-hydrodynamic actuators, stream-wise vortex generators
and transpiration flow) have be investigated for controlling both interaction
and inherent flow unsteadiness.
The interaction unsteadiness is initiated and/or generated by SWBLI itself
but it is often destabilised by the outer/downstream flow field. Therefore, the
response of shock wave and separation to periodic excitations is of utmost
importance and has been included in the research program.
Thus emphasis is focused on closely linked experiments and numerical
investigations to allow the application of numerical results in the experiments
and vice versa for the sake of identifying and overcoming weaknesses in both
approaches.
Using RANS/URANS and hybrid RANS-LES methods, UFAST has cast
new light on turbulence modelling in unsteady, shock dominated flows. More-
over, LES methods were applied to resolve the large coherent structures that
govern SWBLI. This way, UFAST provided the “range of applicability” be-
tween RANS/URANS and LES.
P. Doerffer et al. (Eds.): Unsteady Eff. of Shock Wave Induced Separation, NNFM 114, pp. 1–10.
springerlink.com c Springer-Verlag Berlin Heidelberg 2010
2 1 Introduction – UFAST Project Overview
1.2 Project Objectives

Before UFAST, not enough had been done to accurately predict and control
flows dominated by unsteady shock wave boundary layer interaction. Even
where advanced CFD techniques were applied to predict the flow around
full aircraft configurations, they only dealt with the steady flow features
and often only extrapolated from incompressible/subsonic domains to tran-
sonic/supersonic flow regimes. It is obvious that there was a lack of under-
standing of the flow-physics involved in unsteady SWBLI phenomena. There
was also clearly a need for appropriate modelling and – even more impor-
tantly – for a control of the flows in order to minimalise the physical risks for
aircraft. This is the clear objective of the UFAST project. The few attempts
of applying compressibility flow corrections in turbulence modelling proved
insufficient in respect to the predictive capabilities of unsteady transonic
flows. Moreover, attempts of ‘transposing’ turbulence modelling from incom-
pressible flows also proved insufficient for the accurate prediction of buffeting
and wing-flutter (examples: research program ETMA, Vieweg, Vol. 65; Eu-
ropean research program UNSI, final report published by Springer, Vol. 81)
with regard to the simulation of buffeting phenomenon and shock unsteady
motion. More recently a hybrid DES (Detached Eddy Simulation) approach,
i.e. an inherently 3D approach, was applied to the transonic flows around
airfoils, indicating the crucial need for improvement of our understanding of
flow-physics in order to modify the turbulence scales caused by unsteadiness
and compressibility. Another objective of the UFAST project is to increase
the efficiency of prediction methods.
There was a pertinent need to improve the predictive capability of
CFD methodologies, such as URANS, (Unsteady Reynolds Averages Navier-
Stokes), LES, (Large-Eddy Simulation), and hybrid RANS-LES approaches.
The present UFAST project has delivered a deeper insight into the physics
governing the unsteadiness of the shock, the shock/boundary layer interac-
tion, the development of buffeting, together with a study on efficient methods
for controlling these phenomena.
As the mentioned physical phenomena occur in high speed, i.e. transonic
and supersonic flows both in external and internal aerodynamics, these lead to
boundary layer separation which can cause structural damage and in all cases
downgrades the efficiency of the aircraft or propulsion system. SWBLI can
occur in supersonic air intakes and reduce their efficiency because the induced
separation becomes strongly unsteady and can induce serious damage in the
engine.
The interaction of turbulent eddies with shock waves causes the forma-
tion of very large eddies which propagate downstream of the interaction and
become yet another source of broadband frequency noise. The simulation
of such off-similarity and off-equilibrium phenomena delivers new informa-
tion, particularly in cases where the scales of unsteadiness and/or of large
coherent eddies play a dominant role. Furthermore, the modification of the
1.2 Project Objectives 3
turbulent scales in respect of the unsteadiness and compressibility effects

within strongly separated regions is now better understood. UFAST has
delivered an in-depth analysis of the aforementioned compressible flow phe-
nomena using a triple approach (theoretical, experimental and numeri-
cal) helping to supply efficient, robust and easy-to-implement methods, for
available prediction tools.
For these reasons, the UFAST project has delivered a set of well focused
experiments, relevant to the above mentioned flow-physics phenomena and
the Data Bank of both experimental and numerical results. This provides
a sound basis for work to be carried out in the future. It is accessible to
other interested groups in Europe, but primarily of course to the aeronautics
industry.
To summarise, the treatment of shock wave/boundary layer interaction
involves joining together a number of different physical aspects:
1. High frequency unsteadiness occurring in the incoming boundary layer
which is not clearly related to shock unsteadiness.
2. Unsteadiness of the whole flow field induced by a forced shock oscillation.
3. Unsteadiness of the separation bubble, which may be due to a flow field
forced pulsation or may result from vortex shedding. In the latter case, the
vortices produced in the separated zone are convected downstream often
over large distances.
4. Turbulence production (including strong compressibility effects) caused by
the shock wave itself.
5. Formation of a new boundary layer downstream of the separation, more
precisely downstream of re-attachment which is characterized by vortex
interactions, but also by low-frequency unsteadiness that might be induced
by the shock motion.
6. Strong coupling through acoustic waves between the different phenomena.
This UFAST project has improved knowledge and expertise by delivering:
• Reference experiments focused on unsteady effects.
• Improvement of existing numerical modelling methods.
• Enhanced understanding of complicated physical phenomena.
To arrive at general conclusions regarding these complex and challenging
issues most of the flow configurations in which shocks play a key role have
been addressed. These could only be met by a sufficiently large consortium of
organisations with appropriate skills and expertise in both experimental and
theoretical research. And this was achieved in the UFAST project, where
experimentalists and theoreticians collaborated closely to improve our un-
derstanding of SWBLI. This pooling of resources allowed us to achieve the
upstream goals of the UFAST project. In the work there was a small degree of
overlapping, which allowed for cross validation. The simultaneous gathering
and categorising new results produced a reliable knowledge base.
The first objective of the UFAST project was to provide a comprehensive

experimental Data Bank documenting both low frequency events and the
properties of the large scale coherent structures in the context of SWBLI.
It should again be stressed that before the project almost no experimental
information had been available, especially in industrially relevant flow cases.
Therefore flows in the important Mach number range from transonic condi-
tions to Mach number 2.25 were investigated. The measured flow configura-
tions correspond to generic geometries that can be easily exploited in more
complex geometries, such as airfoils/wings, nozzles, curved ducts/inlets, in
other words, all important flow cases governed by normal and oblique shocks.
This wide shock configuration platform was necessary to identify general in-
teraction unsteady features. And it should be repeated that the realisation
of this objective in a short space of time could only be achieved by involving
a sufficiently large number of laboratories sharing an enormous amount of
crucial experimental work.
The work was split into so-called “basic” (WP-2) and “control” (WP-3)
cases, the latter carried out to provide a means for industry to reduce the risk
of damage caused by flow dynamics, in particular by reducing flow unsteadi-
ness, noise and even material fatigue. Control devices were used to control
large eddies and included: perforated walls, a number of stream-wise vortex
generators, synthetic jets and electro-hydrodynamic actuators EHD/MHD.
As mentioned above, in the UFAST project great emphasis was placed on
the close connection between experimental and theoretical work. The experi-
ments were modified according to the geometry or flow parameters whenever
numerical results indicated a need for it. The UFAST structure involved CFD
groups in the design of experiments.
The second objective concerned the application of theoretical methods to
improve the understanding of unsteady SWBLI as well as the modelling of
such flows. The methods used were, RANS/URANS (WP-4), hybrid RANS-
LES and LES (WP-5). This investigation included advanced numerics, as
well as advanced modelling strategies and investigations on the “range of
applicability” for the different methods involved. The outcome of UFAST in
this respect provides “best-practice guidelines” for the simulation of SWBLI
problems.
From the application of CFD to SWBLI it becomes evident that there is
a strong need for high accuracy schemes, applied to the main three categories
of numerical tools. This requirement stems from the necessity to accurately
capture and resolve spontaneous unsteadiness, such as shear layer instabilities
generated by shock interactions. With the numerical work carried out before
the project it was obvious that shock wave/boundary layer modelling had to
be improved.
The third objective of the UFAST project was to improve our understand-
ing of all physical phenomena governing shock wave/boundary layer inter-
action. New knowledge has been acquired concerning unsteady interaction
phenomena, such as coupling between low frequency vortex shedding and
1.3 List of Participants 5
shock movement and turbulence amplification/decay at the shock wave. This

raised a number of important questions which the UFAST project could not
answer with a sufficient degree of generality. They included:
• what is the nature of the perturbations?
• what are the links and the possible couplings between them?
• what is the role of compressible and subsonic turbulence in the evolution
of relevant mechanisms?
• is it certain that the very low frequencies found close to the foot of the
mean shock are produced by the oscillation of the shock wave?
The project has improved our understanding of the investigated phenom-
ena in specific cases but has not allowed us to produce general conclusions
that would concern all the considered Flow Cases.
To conclude, it is evident that by working together, all the UFAST project
partners, with their excellent expertise in the field of interest, have con-
tributed to closing the knowledge gap on unsteady shock wave/boundary
layer interaction. In other words, a high interactivity between experimental
and theoretical work, resulted in a transparent set of deliverables summarized
in two books:
1. “UFAST Experiments – Data Bank”, IMP PAN Publishers, ISBN 978-83-
88237-46-1, 2009
2. “Unsteady effects in shock wave induced separation”, Springer series –
Notes on Numerical Fluid Mechanics and Multidisciplinary Design
(NNFM), ISBN 978-3-642-03003-1, 2009.
1.3 List of Participants

The consortium overview is given in the Table below. It shows that 18 organ-
isations from ten European countries participated (eight from EU member
states, one from an EU candidate state and one from Russia). In addition
to the partner consortium an “Observer Group” was established, consisting
of four industrial partners (Rolls Royce Germany, Dassault Aviation, Alenia
and ANSYS group) who attended UFAST meetings and took part in the
exploitation of the results.
6
Table List of partner organisations in UFAST project
Part. no. Participant name Participant short name Country

The Szewalski Institute of Fluid Flow Machninery
1 Polish Academy of Sciences IMP PAN Poland
2 CNRS Lab. IUSTI, UMR 6595, Marseille IUSTI France
3 ONERA: ( DAFE, DAAP) ONERA France
4 University of Cambridge, Dept. of Engineering UCAM-DENG Great Britain
5 Queens University Belfast, School of Aero. Eng. QUB Great Britain
Russian Academy of Science, Siberian Branch,
6 Novosibirsk, Inst. of Theor. App. Mech. ITAM Russia
7 Delft Univeristy of Technology, Aerodyn. Lab. TUD Netherlands
8 INCAS, Romanian Institute for Aeronautics INCAS Romania
9 University of Southampton, (SES) SOTON Great Britain
10 University of Rome “La Sapienza” URMLS Italy
11 University of Liverpool, Dept. of Engineering LIV Great Britain
12 NUMECA, Belgium, SME NUMECA Belgium
14 Institute Mécanique des Fluides de Toulouse IMFT France
16 FORTH/IACM, Found. for Res. and Techn.-Hellas FORTH Greece
17 Ecole Centrale de Lyon LMFA France
19 EADS-M, Deutschland GmbH Military Aircraft EADS-M Germany
20 Institute of Aviation, Warsaw IoA Poland
Podgorny Institute for Mechanical Engineering
21 Problems NASU UAN Ukraine
1 Introduction – UFAST Project Overview
1.4 Project Structure 7
1.4 Project Structure

The UFAST project divided its funds and research work into two main areas.
The one concerned experiments which delivered the Data Bank on SWBLI
and its control. The other area concerned numerical simulations, including the
modelling of SWBLI, using URANS, hybrid RANS/LES and LES methods,
and delivered an assessment of their applicability to the problem.
To consider the general features of unsteady SWBLI, most of the typical
flow configurations with shock waves had to be included in the investigation.
Three flow configurations were selected, as shown in Fig. 1.1.
Transonic interaction Nozzle flow Oblique shock reflection
Fig. 1.1 UFAST configurations of flow with shock waves
The selection of three configurations implied a high number of flow cases.

Three different experiments were designed for each configuration. In order
to manage this in a three-year project, a number of experimental facilities
were engaged and various theoretical methods, e.g. CFD codes, were used.
That was the main reason why as many as 18 partners participated in the
realisation of the ambitious goals of the UFAST project.
The structure of the research part of the project including the work pro-
gram details is presented in Fig. 1.2. The Work Packages are presented hori-
zontally. The left column shows the WP number, the WP topic and the WP
leader’s name. The physical phenomena groups are presented in the remain-
ing three columns, numbered and using different colours for greater clarity.
In each WP row this division into three physical phenomena columns defines
three different Tasks.
In each Task of WP-2 different flow cases are labelled by a letter, whereas
in each Task of WP-3 each flow control method is labelled by a number.
Thus in each Flow Case a label consisting of a letter and a number allows to
identify the flow case together with the flow control device.
In WP-4 and WP-5 these labels are used to indicate the flow cases sim-
ulated numerically by each CFD partner. It is important to note that this
labelling is only relevant to its given physical phenomena column.
Experiments in WP2 are presented separately, depending on where they
were carried out, even if the nature of the experiments was similar. Thanks
to this valuable information is provided on the dependence of interaction
unsteadiness on flow constraints.
Fig. 1.2 Graphical presentation of work program
1.5 Project Organisation
In order to secure the exchange of information obtained from experiments

and numerical simulations for the formulation of the main UFAST project
conclusions, it was decided to group partners according to particular Flow
Cases, as shown below.
1.5 Project Organisation 9
Flow cases structure:

This structure of Flow Cases is the basis of the present book, which is
supplemented by the comments of Work Package Leaders.
Part I
Transonic Interactions
Chapter 2
Bump at a Wall (George Barakos)
2.1 Summary
Shock/boundary layer interaction at transonic flow conditions had been the

investigated using wind tunnel experimentation and computational fluid dy-
namics. The main objective of the investigation was to establish a database of
measurements to use for CFD validation and, subsequently, exploit the CFD
to extract further understanding about the flow mechanism and physics. Of
particular interest was the unsteady transonic interaction at conditions ap-
proaching aircraft flight. At first, a detailed study of the effect of the per-
forated and solid walls on the obtained shock configuration was carried out
using the URANS approach. This resulted in a decision to adopt a con-
figuration with a solid contoured upper wall. At the same time, issues re-
lated to the air humidity in the tunnel were investigated by QUB to ensure
that comparisons against the clean air computations of ULIV were possible.
Following this, computations with URANS and zonal LES based on wall-
functions were undertaken to resolve the flow along with Organised Eddy
Simulations (OES). For this test case, there was no low-frequency unsteadi-
ness as initially expected. Instead a rapid small-amplitude shock excursion
was observed. URANS was not able to resolve this phenomenon and the
employed turbulence simulation methods had moderate success as well. The
experiments were conducted at the Queens University of Belfast (QUB) with
simulations carried out by the Institut de Méchanique des Fluides de Toulouse
(IMFT) and the University of Liverpool (ULIV).
2.2 Introduction – The QUB Tunnel Setup

The initial objective of the experiment was to simulate low-frequency shock
oscillations resulting form unsteady shock/boundary layer interaction [1,2,3]
and control these using synthetic jets. The target configuration was a circular
arc aerofoil. Figure 2.1 presents results obtained prior to the UFAST project
springerlink.com
c Springer-Verlag Berlin Heidelberg 2010
14 2 Bump at a Wall (George Barakos)
Fig. 2.1 Unsteady shock/boundary layer interaction for an 18% circular arc case
at transonic Mach number of 0.78. The shock position as a function of time is also
shown (b) compared against experimental data [3]. URANS solution using the k-ω
turbulence model.
for this test case and highlights the good agreement against experiments for
the shock location as well as the complexity of this flow.
After initial considerations, the size of the wind tunnel section of QUB
was found to be small for a full aerofoil experiment and a compromise had
to me made. It was therefore decided to produce a half-model of the aerofoil
and mount it on the lower tunnel wall like a bump.
Experience with such flows was also available since wall-mounted bumps
were already used for experimentation in shock/boundary layer interaction
[4–6]. In their work, Delery and Marvin [4] note that SBLI can considerably
increase the size of the dissipative flow region which would affect the whole
flow field and generally lead to losses in terms of lift, drag increase, and buffet.
All these phenomena are of profound relevance to the aerospace industry.
Several works have so far provided good insight into steady-state interac-
tions and resulted in established databases for validation of CFD methods
and development of better turbulence models for complex transonic flows.
Amongst others, the works of Bachalo and Johnson [5] and Delery [4,6] are
the most popular examples of shock/boundary layer interaction. One of the
strengths of these works is that they provided detailed flow measurements
near the interaction region including surface pressure and boundary layer
data, that have been used as a benchmark against which many CFD codes
and turbulence models are assessed [7]. Figure 2.2 shows indicative result
from the case reported in [5] and one can see the formation of the shock and
the separation just downstream its foot.
The test section of the QUB tunnel is shown in Figure 2.3 with the wall-
mounted bump. Two initial configurations were proposed with solid and per-
forated upper walls. The perforated wall was made out of two slots covered
with porous material.
The cross section of the test section was 101.6 mm × 101.6 mm at the inlet
and has upper and lower plenum chambers each 101.6 mm wide and 24.7 mm
depth (approximately). The test section had a length of 979 mm and linked
2.2 Introduction – The QUB Tunnel Setup 15
Fig. 2.2 CFD predictions for the transonic SBLI case of Bachalo and Johnson [4]
the nozzle with the diffuser. Figure 2.3 shows that the test section consists
of two parallel sidewalls (solid with two port-holes for the schlieren/PIV
system), one top wall which could be contoured or perforated roof, and one
bottom wall with a bump. The top and the bottom walls of the test section
may not be parallel in order to compensate for the blockage effect due to
boundary layer growth.
The lower wall of the tunnel was occupied by the bump which is shown in
Figure 2.4. The model was a bi-convex semi profile of 9% thickness, 101.6 mm
long and the maximum thickness was 9.144 (e/2). The model was spanning
the test section width. The radius of the bump was 145.68 mm and the centre
of the bump was located at 600 mm from the entrance of the test section.
Therefore the bump leading edge was located at X = 549.2 mm.
The stagnation conditions for the experiment were near ambient pressure
and temperature: p0 ≈ 0.99 ± 0.04 × 105 Pa and T0 ≈ 290 ± 5 K. The rela-
tive humidity was kept below 18% and the free stream turbulence conditions
for the tunnel can be considered as isotropic, with a level of 0.35% velocity
fluctuations. The active part of a test run will be in the range of 8 to 10 sec-
onds. In the case of the porous wall configuration, the shock location on the
model is measured to be at approximately 65% chord length, corresponding
to a free-stream Mach number M ∞ = 0.78 (at station X = 382 mm) and
peak Mach number Mpk = 1.31 for 30% of humidity. For the contoured wall
case, the shock location on the model was measured to be at 65% of the
chord length, peak Mach number Mpk = 1.37 and free-stream Mach num-
ber M∞ = 0.785 (18% of humidity). The reattachment point was located
Fig. 2.3 (a) solid and (b) porous upper walls for the QUB wind tunnel. (c, d)
Bump model and co-ordinate system.
2.2 Introduction – The QUB Tunnel Setup 17
between X = 670 and 690 mm (or 120 to 140% of the chord). Preliminary
results gave an upstream momentum based Reynolds’ number (calculated at
X = 382 mm) approximately equal to Rθ ≈ 6400.
Due to problems with tunnel humidity, the first sets of data were obtained
with the porous tunnel walls and comparisons against CFD computations
using free-stream conditions were not encouraging. This prompted a detailed
study by QUB of the effect of humidity on the obtained results. This study
is reported in reference [8]. As observed, a level of humidity of less than
18% was required for direct comparisons against CFD obtained using dry air
conditions.
Subsequently, an effort was put in place to quantify the effect of the
top tunnel wall on the transonic interaction. Inviscid results (presented in
Figure 2.4) were obtained with far-field boundary condition for the top wall
and periodic condition for the side walls. The results indicate that there is
some interference between the shock and the top wall even for domains where
the top wall is placed at 2.5 bump-chords away from the bottom wall. The
distributions of the stream-wise and normal velocity components along the
top boundary are presented in Figure 2.5 for a Mach number of 0.745.
Viscous computations were also undertaken using the k-ε turbulence model
and results are shown in Figure 2.6. The conclusion from both inviscid and
viscous computations is that a top boundary at about three bump-chords
away from the bottom wall is needed for free-stream conditions to be ap-
proximated for inlet Mach numbers in the 0.7 to 0.76 region.
A further step was also taken to assess the potential use of a porous wall
boundary condition for CFD computations. Reference [9] presents a simple
model for porous walls according to which a small velocity component is
allowed in the direction normal to the solid wall. For viscous computations,
according to this model, the pressure and density are extrapolated on the
wall from interior points, the tangential velocity components are set to zero
and the velocity component normal to the wall (w) is given by:
Fig. 2.4 3D, steady, inviscid results for the QUB test cases. The mid-plane pressure
field is shown for three locations of the top boundary. Results were obtained for
a Mach number of 0.7 and on a grid of 290k cells.
Fig. 2.5 3D, steady, inviscid results for the QUB test cases. The distributions of
the steam-wise and normal velocity components are shown at 1, 2 and 3 chords
away from the bottom wall and for domains where the top boundary is placed at
1 (red), 2 (cyan) and 3 (green) chords away. Results were obtained for a Mach
number of 0.745 and on a grid of 290k cells.
Fig. 2.6 3D, steady, viscous results for the QUB test case. The distributions of
the steam-wise and normal velocity components are shown at 1, 2 and 3 chords
away from the bottom wall and for a domain where the top boundary is placed at
3 chords away. Results were obtained for a Mach number of 0.745 and on a grid of
420 k cells.
w p − pplenum
=σ
U∞ ρ∞ − U ∞2
where σ is the porosity of the wall and pplenum is the pressure at the plenum
around the porous wall.
Indicative results shown in Figure 2.6 suggest that the pressure coefficient
distribution along the top wall can be reduced (zero is the desired value).
The corresponding changes at the wall velocity components (obtained just
below the top wall) are also given in Figure 2.7. The values of the porosity
parameter are based on the area of the top wall only.
2.3 Contoured Upper Wall 19
Fig. 2.7 3D, steady, viscous results for the QUB test case. The distributions of the
wall pressure coefficient, steam-wise and normal velocity components are shown for
top-wall porosities of 0, 5, 10 and 40%. The top boundary was placed at 3 chords
away from the bottom wall and results were obtained for a Mach number of 0.74
and on a grid of 420k cells.
2.3 Contoured Upper Wall

Preliminary results obtained during the first six months of the project indi-
cated that the flow experiments of QUB will probably result in steady-state
flow configurations due to the half-aerofoil model to be employed and the
difficulties to obtain high Mach number due to interactions taking place on
the upper wall of the tunnel. In fact, the requirement for high peak Mach
number in the tunnel in the range of 1.2 to 1.4 was put forward. This Mach
number would allow for flow separation which would perhaps lead to shock
oscillation, unsteadiness and buffet. To compromise between the requirement
for a transonic interaction with a well-identified supersonic pocket, the need
to reach high peak Mach number and the need for solid walls that could be
easily modelled within the URANS framework a decision was made to contour
the upper wall of the tunnel. For the solid wall configuration ULIV computed
a contoured wall shape which could minimise the influence of the upper wall
and result in a good transonic interaction. It has to be clarified, however,

that the issues of mean-flow unsteadiness for this test case still remained.
The one-step method of Amecke [10] has been used to this end. Results of
this effort can be seen in Figure 2.8. Several possible curves have been con-
sidered and a compromise had to be made between the obtained peak Mach
number and the required wall deflection.
Fig. 2.8 Contoured wall results for the QUB test case, curve 6 has been used for
3D computations
Fig. 2.9 Supersonic flow regions in the mid-plane of the channel. Results have
been obtained using curve 6 of Figure 2.4 to shape the upper wall of the tunnel.
2.3 Contoured Upper Wall 21
The obtained supersonic flow regions for curve 6 of Figure 2.8 are shown
in Figure 2.9 for a range of incoming flow Mach numbers. The results suggest
that a well-defined supersonic flow is possible for Mach numbers between 0.77
and 0.8. Based on this set of results, a decision was made for curve 6 to be
used and the upper wall of the tunnel was modified for this purpose.
Once a first round of test experiments was complete, a comparison between
the CFD predictions and the data was made for the peak Mach number.
This set of results is presented in Table 2.1 and as can be seen the CFD
predictions were accurate resulting in a peak Mach number of 1.365. The
shock was located at about 77.5% of the bump and the Mach number near
the top wall remained at about 0.88. This way, substantial flow separation
has been obtained and at the same time, the interaction remained transonic,
having some of the characteristics of the flow around an aerofoil at least as far
the existence of a subsonic flow around the shock and the provision of a clean
path for subsonic flow to mover around the supersonic pocked without the
formation of a normal shock between the upper and lower walls of the tunnel.
Table 2.1 Summary of obtained results for the contoured wall cases. Curve was
finally used.
M∞ Mmax % shock Max Mach Height of supersonic

location at top wall region (mm)
Curve5 0.77 1.32 73.5 0.83 43.6
Curve5 0.8 1.43 82.0 0.93 80.0
Curve6 0.77 1.365 77.4 0.877 53.7
Curve6 0.8 1.475 86.0 1.1 Supersonic region
reaches the top wall
Figure 2.10 presents the bump and the contoured wall inside the test sec-
tion of the QUB tunnel. On the same figure, the experiments for the contoured
and porous upper walls are compared. The isentropic Mach number along the
bump is very similar for the two cases and the schlieren images suggest that
the contoured wall is indeed effective for the selected set of flow conditions.
Fig. 2.10 (a) Geometry of the contoured wall and circular arc bump for the QUB
experiment. (b) comparison between isentropic Mach number on the bump for the
perforated and contoured upper wall cases.
2.4 CFD Tools and Mesh Generation
IMFT and QUB employed similar tools for the CFD analysis of this test
case. In particular, IMFT used the NSMB flow solver [11] which allows for
multi-block structured grids to be used. The solver uses MPI for parallel
execution. Time marching is done via the dual-time stepping method and
a third order Runge-Kutta scheme. The spatial discretisation employs an
implicit Roe scheme with the VanLeer limiter. Central differences are used
for the discrtetisation of the viscous fluxes. NSMB has a variety of turbulence
models including the option of performing Organised Eddy Simulations.
Liverpool used the Parallel Multi-Block (PMB) solver which also follows
the multi-block structured approach. The solver uses an implicit time march-
ing method [12], and the resulting linear system of equations is solved using
a pre-conditioned Generalised Conjugate Gradient (GCG) method. For un-
steady simulations, implicit dual-time stepping is used based on Jameson’s
[12] pseudo-time integration approach. From the beginning, the solver was
designed with parallel execution in mind and for this reason a divide-and-
conquer approach was used to allow for multi-block grids to be computed
on distributed-memory machines and especially low-cost Beowulf clusters of
personal computers.
For this work, a set of meshes was prepared by ULIV and was offered
to the rest of the partners computing this test case. The mesh density for
2.4 CFD Tools and Mesh Generation 23
each of these grids is shown in Table 2.2. As can be seen, two configurations
were considered with porous and contoured upper walls. All meshes had side
walls and provided local refinement near the shock/boundary layer interac-
tion region. The first cell above the lower wall was at y + of about 0.9 and
the streamwise and spanwise spacings in the near-bump region were of the
order of Δx+ and Δy + of 10 for the finest of the meshes.
Table 2.2 Summary of grids for the QUB test case
Control Domain Size QUB Porous QUB Contoured

No Full 0.9M X X
3,2M X X
7.9M X X
Reduced 0.5M X X
2.5M X X
7.0M X X
Yes Reduced 0.9M X
A simple multi-block topology was used since the geometry was smooth
and the number of the employed blocks was increased to allow for parallel ex-
ecution on about 64 processors with load differences of less than 5% between
the CPUs. A snapshot of the multi-block mesh topology used to model the
slotted upper wall of the QUB tunnel is shown in Figure 2.11 and one can
clearly seen that the exact geometry of the slots (where porous material was
placed) was preserved.
At the outset of the project, a proposal was tabled by QUB for the use
of synthetic jets for flow control. The key idea was to have a slot ahead of
the shock connected through a chamber inside the bump to a synthetic jet
mechanism [13]. A mesh was generated for this test case as well and the
details of the grid and the slot can be seen in Figure 2.12.
Fig. 2.11 (a) Multi-block mesh topology and (b) boundaries of the computational
domain for the reduced configuration of the QUB mesh. (c) View of the top con-
toured wall and the fine mesh near the bump region.
2.5 Boundary Conditions 25
Fig. 2.12 (a) Schematic, (b) mesh and (c) multi-block topology for the controlled
test case of QUB. The slot exit is shown in green and is covered by a separate layer
of blocks (b).
2.5 Boundary Conditions

An initial investigation of the flow has been undertaken aiming to identify
(a) the effect of the side walls and (b) assess the potential of URANS com-
putations for this flow case. For the contoured wall case an initial mesh has
been modified to allow for symmetry boundary conditions to be placed along
the side walls of the channel. The obtained results are shown in Figure 2.13
and they are compared against results obtained using wall-condition for the
sides of the tunnel. As can be seen, the obtained flow structure is quite dif-
ferent and based on preliminary flow visualisation [13], it was decided that
this configuration is not suitable for the continuation of the work.
At the same time, the effect of the flow condition was investigated trying
to adjust the exit flow condition to obtain good predictions for the position of
the shock in the tunnel. Figure 2.14 shows indicative results from this effort
compared against preliminary QUB measurements. Since the exit flow of the
QUB tunnel was not controlled by a double-throat mechanism, a wedge was
used to regulate the air downstream the test section. For the CFD conditions,
Fig. 2.13 Comparison of the flow recirculation region downstream the shock for (a)
side walls and (b) symmetry condition. Iso-surfaces of negative streamwise velocity
are shown for −0.01 of U∞ .
Fig. 2.14 Preliminary CFD results for the contoured upper wall case. The stan-
dard k-ω model was used. The obtained CFD predictions for the isentropic Mach
number in the centre of the test section and along the bump were used to adjust
the predicted shock position via the exit pressure.
the shock position had to be matched since the exact setting of the exit wedge
was not available. The obtained results show that overall the preliminary
URANS computations managed to capture most of the flow features related
to this flow, in a qualitative sense.
A final open question remained related to the upper porous wall configu-
ration and the porosity used during experiments. The first computations for
this test configuration revealed poor comparison against experiments. At the
same time QUB investigations revealed the presence of substantial humidity
in the tunnel for the cases where the upper porous wall was used. For this
reason, it was decided not to continue computations with this configuration
but focus on the upper contoured solid wall instead.
2.6 Experimental Findings

The surface pressure on the bump wall obtained at “dry” tunnel conditions
is shown in Figure 2.15 and suggests a peak Mach number around 1.38. This
is followed by a reversed flow region behind the shock.
The flow recirculation behind the shock was visualised using the oil flow
and china clay techniques and the results are shown in Figure 2.16. As can be
seen, the two methods yield similar results and suggest that corner vortices
are present near the side walls of the tunnel. On the same figure, a schematic
of the flow is compiled using the information provided by the experiments.
The position of the shock was tracked from schlieren images and was found
to oscillate about a mean position at relatively high frequency as indicated
by Figure 2.17. The same analysis was performed using the surface-mounted
hot-film data and the signals of the pressure transducers. The shock was
found to oscillate at a frequency of about 350 Hz.
2.6 Experimental Findings 27
Fig. 2.15 Isentropic Mach number measured at ‘dry’ conditions
Fig. 2.16 Visualisation of the separated flow region using two techniques.
A schematic of the flow configuration is also provided.
Fig. 2.17 Shock position as a function of time extracted from the schlieren images
Fig. 2.18 Spectral analysis of the shock wake unsteadiness at different locations
2.7 URANS Results 29
2.7 URANS Results

For the obtained URANS results a variety of turbulence models was em-
ployed. Although Liverpool used mainly the standard k-ω model [14], results
were also obtained using cubic Reynolds stress expansion [15] as well as the
new k-ω formulation of Wilcox [14]. Results are shown in Figure 2.19 where
comparison of the isentropic Mach number along the bump is made for two
turbulence models. The same figure shows the recirculation flow region and
its size. Overall the agreement for the shock position and the downstream
pressure is good and all models managed to predict the separated flow.
On the other hand, none of the URANS predictions provided any shock un-
steadiness and regardless of the employed time step and grid size, the models
failed to show the shock excursions reported by QUB. This was a disappoint-
ing result since the overall flow configuration is well-predicted.
Figure 2.20 shows the Mach number field in mid- and quarter-span planes
of the bump. Both flow regions are characterised by flow separation with the
Fig. 2.19 Isentropic Mach number along the mid-plane of the lower wall. Results
are shown for the NLEVM of [15] as well as for the new k-ω model of Wilcox [14].
Substantial flow re-circulation can be seen near the trailing edge of the bump.
Fig. 2.20 Mach number field and sonic surface obtained using the new k-ω model
of Wilcox [14]
Fig. 2.21 Comparison of the separated flow region near the trailing edge of the
QUB bump. The China Clay visualisation shows substantial flow recirculation near
the vertical walls which is partially captured by the CFD solution (iso-surface of
stream-wise velocity at −0.01 of the free-stream).
near-wall plane showing substantial flow distortion. As discussed in section 4,

near the vertical walls of the bump, a complex flow pattern emerges. This flow
region is harder to predict since it is dominated by the corner flow developed
along the length of the test section.
A comparison between the QUB flow visualisation and the obtained CFD
solutions is shown in Figure 2.21.
To further investigate the flow, a set of slices normal to the free-stream
were taken at three locations along the bump. The Mach number was vi-
sualised along with the in-plane velocity vectors (Figure 2.22). As can be
seen, the influence of the corner vortices begins early ahead of the bump
and progressively grows as the shock region is reached. As a result of this
flow, a more pronounced recirculation region is formed near the middle of
the bump. Table 2.3 presents a comparison between experiments and CFD
for the separation and re-attachment points along the bump. There are two
experimental values reported based on different techniques for measuring the
separation points.
Table 2.3 Comparison between experiments and CFD for the separation and re-
attachment location along the mind-plane of the bump (X ∗ of 1 corresponds to the
trailing edge of the bump)
Separation Re-attachment
point X ∗ point X ∗
Experiment Method 1 0.64 1.40
Experiment Method 2 0.63 1.32
URANS (k-ω) 0.65 1.48
LES (averaged) 0.62 1.49
(1.42 min 1.51 max)
More URANS results were obtained by IMFT using several turbulence

closures including the Spalart –Allmaras [16] model as well as the Non-Linear
Eddy Viscosity Model (NLEVM) of Abe, Jang and Leschziner [17].
The separated flow region, as well, as the sonic surfaces are shown in
Figure 2.23. It appears that the models predict well the development of the
separation between two corner flow regions adjacent to the side walls of the
tunnel. As was the case for results obtained by Liverpool, a tight sonic region
is present above the bump which extends up to the side walls of the tunnel
and its shape is influenced by the corner vortices.
The stream-wise development of the flow can be seen in Figure 2.24, where
results from the Abe et al. model [17] are shown. The vectors and the con-
tours of the velocity magnitude suggest an early development of the corner
vortices which grown in size at the separated flow region. As was the case for
Liverpool, a central separated flow region is present and large corner vortices
appear along the test section. These tend to dominate the flow as the bump
Fig. 2.22 Flow visualisation of the URANS results obtained using Wilcox’s new
k-ω model [14]. Mach number field is shown along with in-plane velocity vectors.
Fig. 2.23 Separated flow region and sonic surfaces predicted with the Abe-Jang-
Leschziner model (left) and the Spalart-Allmaras model (right)
is approached and near their peak they occupy more space than the central
reversed flow region.
Pressure iso-surfaces are shown in Figure 2.25 for the mid-span (50% of
bump length) and a station and quarter-span (25% of the bump width). The
results for the Spalart Allmaras model show a well-defined shock even at the
25% station in agreement with the sonic surface presented in Figure 2.19.
Figure 2.26 presents the same information, in the form of total pressure
for the results obtained using the Abe et al. model. No shock motion was
found in these computations as was also the case for the Liverpool URANS
results.
The surface pressure was extracted along the mid-plane of the bump and
Figure 2.27 presents the results obtained by IMFT. All models capture well
the development of the flow ahead of the shock and the differences are concen-
trated near the separation and re-attachment regions of the flow. The results
are overall encouraging showing that URANS is capable of capturing most
of the flow physics. On the other hand, the URANS computations could not
Fig. 2.24 Near-wall flow at two spanwise locations: Leading edge of the bump
(top) and trailing edge (bottom)
Fig. 2.25 URANS results obtained using the Spalart-Allmaras model at 50% (left)
and 25% (right) of the bump span
2.8 Flow Control Attempts 35
Fig. 2.26 Pressure field obtained using the Abe-Joung-Leschziner model at 50%
of the bump span (left) and 25% (right)
Fig. 2.27 Surface pressure coefficient along the centreline of the bump obtained
using the Abe-Jang-Leschziner model (left) and the Spalart-Allmaras model (right)
resolve the high frequencies of flow scales involved in the shock motion and
for this reason, attempts were made using a zonal LES method as well as
Organised Eddy Simulations.
2.8 Flow Control Attempts

Encouraged by the results obtained from the CFD and experimental efforts,
consideration was given to flow control aiming to alleviate the shock, reduce
the separated flow region and potentially affect the unsteadiness of the flow.
Due to the popularity of synthetic jets [18–21] as a flow control device, numer-
ical simulation has been used in an effort to quantify the effect of synthetic
jets ahead of the shock.
A slit was therefore placed upstream the foot of the shock and a synthetic
jet boundary condition was applied at its outlet. The origin of the slit was
placed on the circular-arc bump 60 degrees from the vertical so that the
jet fires towards the foot of the shock. The jet slot had a width of 0.5 mm
and was running along the span of the bump between the two side walls of
the tunnel. The jet condition was implemented as a plug velocity profile with
varying amplitude (between 0.5 and 2 percent of the free-stream velocity) and
frequency (ranging from 10 to 500 Hz). The flow at the jet outlet was assumed
to be at stagnation pressure and temperature. In contrast to previous CFD
results, the region near the jet had to be refined to allow adequate resolution
of the flow downstream the jet to capture the vortical structures formed just
after the jet exit.
For this test case, computations were undertaken to identify the opera-
tional parameters of a synthetic jet actuator placed ahead of the shock, on
the employed bump. After discussions with QUB, the configuration shown in
Figure 2.28 was considered as a starting point and computations have been
undertaken with the jet in a cross-flow.
Fig. 2.28 Example of synthetic jet actuator employed for preliminary

computations
The parameters initially varied were the amplitude and frequency of the
oscillation of the deforming surface below the plenum. These calculations
employed deforming grids and were quite demanding due to the large number
of parameters that had to be set. The grid deformation method in the Parallel
Multi-Block solver of Liverpool is based on the Trans-Finite Interpolation
(TFI) and performed well for this case.
2.8 Flow Control Attempts 37
It was found, that at the flow conditions upstream the shock, actuation
frequencies about 100 Hz were necessary in order to form vortices that persist
downstream the actuator. One such case is shown in Figure 2.29.
Fig. 2.29 Vortices generated by the synthetic jet
Based on this work, an approximate velocity profile has been derived for
the synthetic jet and it envisaged that this will be applied as a boundary
condition for the CFD computations thus avoiding the need to resolve the
details of the actuation mechanism as was the case for the computations
described above. Figure 2.30 presents the exit velocity shape employed for
computation. The streamwise (u) component was different from the vertical
one (v) and a constant profile was applied along the slit shown in Figure 2.12.
CFD results with frequencies lower than 100 Hz revealed little effect on
the base flow even when the maximum exit velocity was applied. Frequencies
between 350 and 500 Hz were more effective in changing the flow even when
a peak velocity equal to 1% of the free-stream was used.
Fig. 2.30 Exit velocity profiles selected for the QUB synthetic jet experiment
Figure 2.31, presents a comparison between the controlled case and pre-
viously obtained results without the jet. The k-ω model was used due to its
efficiency and no attempt was made to further adjust the back pressure to
obtain better agreement with experiments for the base flow. Once can see
that the peak Mach number is reduced and a smaller separation region is
observed. The experiments on Figure 2.31 correspond to the base flow since
no measurements for the controlled configuration were made by QUB. This
was due to difficulties with instrumentation and the high frequencies required
for computations.
Fig. 2.31 Comparison between CFD solutions for the base and controlled flows
(M∞ = 0.78 case, with square jet profile)
This first set of results suggests that synthetic jets may affect the shock
structure though further simulation and experimental work is needed before
conclusions can be drawn for the specific jet configuration and conditions
necessary for shock alleviation.
2.9 LES and OES Results

The results obtained using the URANS models presented in subsection 6 are
encouraging and show that most of the flow features were well-resolved. This
set of features lacked, however, unsteadiness and for this reason, turbulence
simulation was considered as a way to better study this flow. A preliminary
computation was carried out using wall-function-based LES of the QUB con-
figuration on a coarse CFD mesh. This model did not attempt to resolve the
turbulent flow near any of the tunnel walls. It was used only as a preliminary
investigation of the potential of LES for this flow case. The medium size mesh
2.9 LES and OES Results 39
of Table 2.2 was used for this test. Figure 2.32 presents two time-traces of the
unsteady pressure and fluctuating velocities just behind the shock and near
the trailing edge of the bump. The flow for this case appeared to be initially
steady. However, longer runs suggested that some small flow unsteadiness
may be present in this case. The information available at the time from the
experiments of QUB suggested a relatively high frequency of shock oscilla-
tion at between 325 and 400 Hz. This creates a problem for URANS and LES
simulations since not only fine meshes are required to resolve the walls of
the tunnel but also very fine time steps to allow for the resolution of this
frequency along with long computations so that a substantial length of the
time signal can be collected.
A literature survey [22,23,24,25,26] for unsteady shock/boundary layer in-
teraction, revealed a that the type of flow configuration obtained in the wind
tunnel of QUB was studied in the past [24,25] by Pearcey and his co-workers.
The flow unsteadiness was driven by an interaction between two separated
flow regions behind the shock as the schematic of Figure 2.33 shows.
To reveal the details of the flow near the interaction region, a more detailed
study was required. Since little has been published about LES for transonic
flow, the computations for this case were based on the few papers from the lit-
erature where steady shock-boundary layer interactions were simulated using
LES [26].
The first decision to be made was to isolate the interaction region near the
lower wall, using a zonal approach as shown in Figure 2.34. In addition a mesh
Fig. 2.32 Preliminary wall-function-based LES results for the QUB test case. The
Smagorinsky sub-grid scale model was used on the medium grid of Table 2.2. The
computation was run for 100 bump travel-times and the last third of the signal was
averaged and processed.
Fig. 2.33 Schematic of the interaction between shock and separated flow regions
suggested by Pearcey et al. [24,25]
Fig. 2.34 Zonal approach adopted for the QUB test case. A wall-function model
was used for the top and side walls around the bump.
of 432 × 135 × 120 cells in the steam-wise, normal and spanwise directions
was used which was load-balanced on 96 CPUs.
For this work, the compressible version of the Smagorinsky mode was con-
sidered according to Erlebacher et al. [27]:
The model uses the Favre-filtered viscous stress tensor:

2
τ̄ij = μ 2S̃ij − S̃kk δij
3
and the Sub-Grid Scale (SGS) viscous tensor model:

2 2
τ̄ij = μt 2S̃ij − S̃kk δij − ρ̄kSGS δij
SGS
3 3
where
1 ∂ ũi ∂ ũj
S̃ij = +
2 ∂xj ∂xi

3
μt = CR ρ̄Δ2 S̃ij S̃ij , Δ= ΔxΔyΔz
A further modification to the eddy viscosity was employed following the
work of Nicoud and Ducros [28]:

d Sd 3
Sij
2 ij
μt = ρ̄ (Cw Δ)
d Sd 5 +
Sij d Sd 5
Sij
ij ij

1 ∂ ũi ∂ ũk ∂ ũj ∂ ũk 1 ∂ ũn ∂ ũk
d
Sij = + − δij
2 ∂xk ∂xj ∂xk ∂xi 3 ∂xk ∂xn

2 2 Cln μt 2
τ̄ijSGS = μt 2S̃ij − S̃kk δij − δij
3 3 ρ̄ Δ
μt ∂ T̃
kSGS = Cl Δ2 S̃ij S̃ij , qiSGS = −cp
Prt ∂xi
Cln = 45.8, CR = 0.012, Cl = 0.0066
The upstream boundary was imposed and white noise at the level of 0.2%
of the kinetic energy was added. The zone boundary followed the approach
described by Temmerman [29]. An attempt was also made to re-scale DNS
data from the ERCOFTAC database, but this was abandoned due to the need
to scale the DNS for time and space. The downstream condition followed the
approach described in reference [26].
From the LES results, the time histories of certain probes were kept. This
was not possible for the full domain, due to disk space limitations, and for
this reason, 100 POD modes were generated and used for the various flow
re-constructions. The POD was updated as the solution converged. A time
step of Δt = 10−5 seconds was used and just about 0.3 seconds of the flow
were captured. The last part of this signal (about 0.2 seconds) was kept for
comparisons. A fourth order central scheme was used for the spatial dis-
cretization.
A recent work [26] attempted to compute a similar test case studied in the
thesis by Bron [23] (Figure 2.35).
For the QUB investigation, some point-to-point correlation results are
shown in Figure 2.36. There correlation function for the w w component
reaches negative values quickly. The streamwise (u u ) component shows
better behaviour and perhaps grid-refinement and adjustment of the zone
boundary are necessary for better results for this case.
A first comparison against the few transducer signals available for the
QUB experiment is shown in Figure 2.35 and 2.38 for the time and frequency
domain, respectively. For the frequency comparison, the experiments and
CFD were processed in the same way. For this purpose, 0.1 seconds of signal
were used and sampled at 1 kHz. The results show some level of agreement
suggesting that the current wall-modelled LES has at least captured some of
the physics of the interaction.
Figure 2.39 presents a set of flow snapshots visualised using streamlines.
As can be seen, the instantaneous results reveal a complex flow with several
vortices present. On the average though the flow appears to be organised in
a well-identified recirculation area downstream the shock. This is similar to
what reference [26] suggests for a similar shock/boundary layer interaction.
Figure 2.40 presents the predicted isentropic Mach number and the results
are in good agreement with the experiments.
Fig. 2.35 (a) Shock/Boundary layer interaction base by Bron [23] and (b) numer-
ical results reported by Wolblad et al. [26]
Fig. 2.36 Point-to-point correlation results near the interaction region (streamwise
and spanwise velocity correlations are shown)
Using the unsteady flow snapshots re-constructed from POD Figure 2.41
and iso-surfaces of negative stream-wise velocity are shown. The size of the
central separated flow region changes with time and the corner flow appears
to be somehow less extended in comparison to the URANS results. It is not,
however, clear from this figure whether two or more separated flow regions
exist downstream the shock. A comparison between URANS and LES re-
sults is also shown in Figure 2.42. The averaged results show a similar shock
structure like the URANS and smaller corner vortices.
Figure 2.43 presents CFD visualisation of the re-constructed solution after
filtering with a band-pass filter between 200 and 500 Hz. One can see that
some flow structures near the trailing edge of the bump as well as further
upstream near the foot of the shock, survived the filtering. This suggests
that at least some of the structures in the separated flow region have the
same frequency as the shock oscillations reported in the experiments. As
shown, a small separated flow region appears just at the foot of the shock.
This structure appears to travel downstream towards the trailing edge and
eventually merge with the larger trailing edge separation.
The OES technique was pioneered by IMFT and several of the details
of the method are given in the literature [9]. This method allows for some
simulation of flow structures and presents a good alternative to URANS for
this test case.
Again, as was the case for the LES results presented above, the CPU cost
of the computations may be high and for this reason only one computation
was attempted.
Figure 2.44 presents some of the obtained results for the pressure field at
the mid-lane of the bump. The mean and RMS values are shown. The mean
value appears to be somehow different from the URANS results. A shock is
Fig. 2.37 Comparisons between CFD and experiments for the unsteady pressure
load at three stations along the bump
Fig. 2.38 Comparisons between CFD and experiments for the unsteady pressure
load at three stations along the bump
Fig. 2.39 Snapshots and averaged flow results for the QUB test case
Fig. 2.40 Comparison between CFD and experiment for the averaged isentropic
Mach number in the centre of the bump
present, but its strength appears to be high. The shock extends all the way
to the upper wall of the tunnel and this makes the comparisons somehow
difficult for this test case.
The ability of the OES method to resolve the flow unsteadiness is demon-
strated in Figure 2.45 where time histories of flow quantities are shown for
point 47 (in the interaction region). There is some frequency content in the
pressure signal and at about 0.08 units of dimensionless time, the signal
Fig. 2.41 Instantaneous flow visualisation for the QUB best case: The colour on
the surface represents pressure and the iso-surfaces negative streamwise velocity at
the level of 0.01 of the free-stream
Fig. 2.42 Comparison of the flow-field obtained with the coarse-grid LES of Fig-
ure 2.23 with the URANS results obtained with the k-ω model of Wilcox
Fig. 2.43 Filtered flow solution near the interaction region suggesting the presence
of a complex flow region similar to Percey’s observations [18,19]
Fig. 2.44 OES results for the pressure field at mid-span of the bump. Mean values
(left) and RMS values (right).
appears to be increase in amplitude. This is echoed the velocity and pressure

RMS values. Figure 2.46 presents the OES pressure signals and spectra at
0.66c and 0.81c locations. The FFT shows a peak at about 350 Hz.
Fig. 2.45 Time histories for point 47 (interaction region) for the pressure (top left),
kinetic energy of turbulence (top right), RMS pressure (bottom left) and RMS of
spanwise velocity(bottom right). Results were obtained using the OES approach.
Fig. 2.46 k-ε OES results of IMFT showing pressure signals and FFTs at two
locations over the bump. A frequency of about 250 Hz is shown in the spectra.
References 51
2.10 Summary and Suggestions for Future Work

The work carried out by ULIV for this project represents 12 person-months
of effort. The basic flow has been investigated and a substantial amount of
results are available for the database of UFAST. One of the key issues in
this investigation is that the low-speed shock oscillation was not achieved
and for this reason, the CFD task to simulate the detailed shock motion
at high frequencies was challenging. URANS provided adequate results for
most cases, when the separation length, shock strength and surface pressure
were compared with the experiments. It was not possible, at least with the
method employed at Liverpool and the available URANS models to resolve
the high-frequency shock oscillations. The presence of the corner flow and
the difficulty of URANS in dealing with this was also highlighted.
However URANS proved a good tool for computing the shape of the upper
wall of the tunnel as well and this is a good example of how CFD and tunnel
experimentation can be used in a synergetic way.
LES for this test case was also challenging since transonic LES is currently
in its infancy with few publications available in the literature. The QUB
configuration includes side walls with corner flow required a special technique.
For this purpose, a zonal method based on wall-modelled LES was adopted
and the obtained results, shed some further light in the fundamentals of the
interaction. In contrast to previous cases studied, the interaction for the QUB
test case appears to be driven by the complex unsteady flow pattern in the
recirculation region behind the shock. Along the same lines, the OES method
was used and again flow unsteadiness was predicted.
The flow control was only attempted using URANS since efforts to im-
plement synthetic jet for the experiment were abandoned and the first CFD
results suggest that high jet frequencies are required for the formation of
stable structures in the boundary layer that could possibly control this case.
The work for this test case should certainly be extended both at the CFD
and experimental front. Reducing the Reynolds number one could perhaps
obtain similar physics and make a full LES possible. This would certainly be
at the centre of any future investigations. The expansion of the comparisons
against experiments should also be pursued. At the moment, only surface
pressure and a few time series of pressure transducers are available and pres-
sure surveys normal to the wall along with entry and exit conditions could be
useful for CFD purposes. In addition, some better form of measuring the flow
unsteadiness could provide valuable data for CFD comparisons. This could
be in the form of PIV.
References
[1] Barakos, G., Drikakis, D.: Numerical Simulation of Buffeting Flows Using
Various Turbulence Closures. International Journal of Heat and Fluid Flow
21(5-6), 620–626 (2000)
52 References
[2] Chung, K.M.: Unsteadiness of Transonic Convex Flows. Exp. in Fluids 37,
917–922 (2004)
[3] Levy, L.L.: Experimental and Computational Steady and Unsteady Transonic
Flows about a Thick Airfoil. AIAA J. 16(6), 564–572
[4] Delery, J., Marvin, J.G.: Shock-Wave Boundary Layer Interactions, North At-
lantic Treaty Organization: Advisory Group for Aerospace Research and De-
velopment, AGARDograph No. 280 (1986)
[5] Johnson, D.A., Horstman, C.C., Bachalo, W.D.: Comparison between exper-
iment and prediction for a transonic turbulent separated flow. AIAA J. 20,
737–744 (1982)
[6] Delery, J.: Shock Wave / Turbulent Boundary Layer Interaction and its Con-
trol. Progress in Aerospace Science 22, 209–228 (1985)
[7] Barakos, G., Drikakis, D.: Assessment of Various Low-Reynolds Number Tur-
bulence Models in Shock-Boundary Layer Interaction. Comp. Meth. Appl.
Mech. Engng. 160(1-2), 155–174 (1998)
[8] Huang, J.C., Gault, R.I., Benard, E., Raghunathan, S.: Effect of humidity on
transonic bump flow. Journal of Aircraft 45(6) (November-December 2008)
[9] Castro, B.M., Jones, K.D., Ekaterinaris, J.A., Platzer, M.F.: Analysis of the
Effect of Porous Wall Interference on Transonic Airfoil Flutter. In: AIAA-
2001–2725, 31st AIAA Fluid Dynamics Conference & Exhibit, Anaheim, CA,
June 11-14 (2001)
[10] Amecke, J.: Direct calculation of wall interferences and wall adaption for two-
dimensional flow in wind tunnels with closed walls, NASA Technical Memo-
randum TM 88523 (1986)
[11] Martinat, G., Braza, M., Hoarau, Y., Harran, G.: Turbulence modelling of
the flow past a pitching NACA0012 airfoil at 105 and 106 Reynolds numbers.
Journal of Fluids and Structures 24(8), 1294–1303 (2008)
[12] Jameson, A.: Computational Algorithms for Aerodynamic Analysis and De-
sign. Applied Numerical Mathematics 13(5), 383–422 (1993)
[13] Huang, J.C., Benard, E.: QUB Report on WP-2 Basic Experiments Data-bank
input UFAST Deliverable D2.1.7 (2007)
[14] Wicox, D.C.: Turbulence Modelling for CFD, 3rd edn. DCW Industries (2006)
[15] Wallin, S., Johansson, A.: An explicit algebraic Reynolds stress model for
incompressible and compressible flows. J. Fluid Mech. 403 (2000)
[16] Spalart, P.R., Allmaras, S.R.: A One-Equation Turbulence Model for Aerody-
namic Flows. La Recherche Aerospatiale (1), 5–21 (1994)
[17] Abe, K., Yang, J., Leschziner, M.A.: An Investigation of wall anisotropy ex-
pressions and length scale equations for non-linear eddy viscosity models. J.
Heat, Fluid Flow 24 (2003)
[18] Kral, L.D., Donovan, J.F., Cain, A.B., Cary, A.W.: Numerical Simulation of
Synthetic Jet Actuators, AIAA Paper 97–1824
[19] Rizzetta, D.P., Visbal, M.R., Stanek, M.J.: Numerical Investigation of Syn-
thetic Jet Flowfields, AIAA Paper 98–2910
[20] Seifert, A., Greenblatt, D., Wygnanski, I.J.: Active separation control: an
overview of Reynolds and Mach numbers effects. Aerospace Science and Tech-
nology 8(7), 569–582 (2004)
[21] Glezer, A., Amitay, M.: Synthetic jets. Ann. Rev. Fluid Mechanics 34, 503–529
(2002)
References 53
[22] Liu, X., Squire, L.C.: An investigation of shock/boundary layer interactions

on curved surfaces at transonic speeds. J. Fluid Mech. 187, 467–486 (1987)
[23] Bron, O.: Numerical and Experimental Study of the shock-Boundary layer In-
teraction in Transonic Unsteady Flow, Ph.D. thesis, Royal Institute of Tech-
nology, Sweden (2003)
[24] Pearcey, H.H.: Some effects of shock-induced separation of turbulent boundary
layers in transonic flow past aerofoils, Aeronautical research council reports
and memoranda, No.3180, London (1959)
[25] Pearcey, H.H., Osborne, J., Haines, A.B.: The interaction between local ef-
fects at the shock and rear separation — a source of significant scale effects
in wind-tunnel tests on aerofoils and wings. In: AGARD CP-35, Transonic
aerodynamics, Paris, France, September 18-20, pp. 11.1–23 (1968)
[26] Wollblad, C., Davidson, L., Eriksson, L.E.: AIAA J., vol. 44(10) (2006)
[27] Erlebacher, G., et al.: Toward the large-eddy simulation of compressible tur-
bulent flows. J. Fluid Mechanics 238 (1992)
[28] Nicoud, F., Ducros, F.: Subgrid scale stress modelling based on the square of
the velocity gradient tensor. Flow Turbulence and Combustion 62 (1999)
[29] Temmerman, L., Leschziner, M.A., Hanjalic, K.: A-priori studies of a near-
wall RANS model within a hybrid LES/RANS scheme. In: Rodi, W., Fueyo,
N. (eds.) Engineering Turbulence Modelling and Experiment – 5, pp. 317–326
(2002)
Chapter 3
Biconvex Aerofoil (Stefan Leicher)
Abstract. Aligned with the needs of the aeronautics industry the general
aim of the UFAST project is to foster experimental and theoretical work
in the highly non-linear area of unsteady shock wave boundary layer inter-
action (SWBLI). Although in the past several EU projects were aiming at
transonic/supersonic flows, the area of unsteady shock wave boundary layer
interaction has not yet been treated. Moreover, experimental methods as well
as numerical approaches have been improved considerably.
The current test case is that of a biconvex airfoil in wind tunnel at buf-
feting flow conditions (McDevit et al, Levy). Three partners, EADS-MAS,
IMFT and INCAS have participated in this task. All three partners have
performed flow simulations while INCAS was responsible for the new experi-
ments as well. Structured and unstructured grids and codes were used as well
as URANS; DES and LES models were applied.
Contributors: S. Leicher, G. Barbut, M. Braza, C. Nae, F. Munteanu and
M.V. Pricop.
3.1 Introduction
Buffeting is a severe problem in airplane service which can lead to reduced
comfort, life cycle limitations or even structural fatigue. Therefore it is very
important during design to predict accurately the buffet onset.
The present test case is that of a biconvex symmetric airfoil in a wind tun-
nel. The experiments are performed by INCAS in their 1.2 m × 1.2 m Trisonic
Wind Tunnel at shock induced buffeting conditions. A strong shock in the
rear part of the profile leads to shock induced boundary layer separation.
The increasing boundary layer thickness and separation force the shock to
move upstream and it becomes weaker till the separation vanishes and the
flow reattaches. Then the movement starts from the beginning again. For the
comparison with the simulations the test run at Mach number of 0.76 and
an angle of attack of 1.0◦ and a Reynolds number of 6.78 × 106 were chosen.
springerlink.com c Springer-Verlag Berlin Heidelberg 2010
56 3 Biconvex Aerofoil (Stefan Leicher)
While INCAS performed the experiments all three partners did flow sim-
ulations using their own code, grids and different URANS, DES and LES
models.
Table 3.1 Grids, codes and turbulence models used by the partners
EADS X TAU SAE SAE-DDES

IMFT X NSMB SA
INCAS X DxUNsP 2 eq. k-ε LES
3.2 Buffeting Experiments

The experiments were performed by INCAS in the 1.2 m×1.2 m Trisonic Wind
Tunnel in buffeting using solid walls section, for a similar case as described
by McDevitt and Tijdeman. This was a decision taken so that CFD analysis
could be performed for the test conditions where boundary conditions have
a direct numerical representation. INCAS has also performed equivalent tests
for the buffeting case where transonic porous section was used (similar to
Mabey test cases), but CFD analysis for these experiments requested special
treatment for the boundary conditions at the perforated walls and this type
of simulation was considered out of the scope of the UFAST project.
Description of the test section and measurement setup
The flow field configuration is based on a classical biconvex aerofoil (18%)

at transonic speeds in the range of Mach 0.7 to 0.9. The experiments were
designed in order to enable buffeting on the model.
The rectangular model having an 800 mm span, a 400 mm chord and a bi-
convex profile of 18% relative thickness was attached on its lower side to
a rigid 72 mm diameter sting. The sting was installed in the model support
pitch system so that its angle of attack could be varied during the run.
Pressure holes, Kulite positions (blue) and synthetic jets (SJ1–5). The
exact dimensions of the model are given in Figure 3.1. The model is designed
to accommodate pressure scanning devices, Kulite pressure transducers and
SJ actuators.
The tests were performed in the 1.2 m × 1.2 m solid wall test section of the
INCAS Supersonic Wind Tunnel, see Figure 3.2. The geometry of the solid
wall test section is defined by the last portion of the nozzle contour which for
the subsonic regime is slightly divergent and the exact dimensions are given
in the UFAST Deliverable D2.1.2 revision 2.
3.2 Buffeting Experiments 57
Fig. 3.1 The geometry of the UFAST model
Fig. 3.2 The model inside the solid wall test section
Test program
All the experiments performed with the UFAST model are included in
Table 3.2.
Table 3.2 Experiments with the UFAST model
No. Period Runs Purpose Equipment

1. Feb.–May 2007 7406–7427 Schlieren visualizations Schlieren
2. June 2007 7428–7431 Pressure distr. on Flexible Nozzle Schlieren
and Variable Diffuser walls + Scanivalves
3. Oct. 2007 7432–7444 Pressure distribution on model Schlieren
top surface + ZOC
4. May 2008 7455–7462 Instantaneous pressures on model KULITE
top surface
5. June–Nov. 2008 7463–7481 Flow control experiments SJA, KULITE
The first series of tests – from #7406 to #7444 – were performed in order
to identify the regimes of buffeting, to determine the flow conditions up-
stream as well as downstream of the test section and to measure the pressure
distributions on the model in the 7 sections shown in Figure 3.1.
During these first experiments, the following data were recorded:
• The stagnation pressure P0 from 2 total pressure probes inside the settling
chamber, downstream of the last screen was measured using a 3.5 Bar
Druck pressure transducer of ±0.05% accuracy. The stagnation pressure
for all tests was kept at 1.3 Bar.
• The static pressure Ps from two reference holes on the test section walls,
2108.2 mm upstream of the schlieren axis, was measured using a 4 Bar
Druck pressure transducer of ±0.05% accuracy;
• The total temperature T0 was measured with a special thermoresistance
inside the settling chamber;
• The axial pressure distributions on the Flexible Nozzle side walls were
measured by means of a scanivalve using the existing static pressure holes
along the axis of each side wall (see Figure 3.4);
• The pressure distributions along the Variable Diffuser walls downstream
of the Second Throat were measured by means of a scanivalve (see
Figure 3.5);
• The pressure at location Pp situated at the schlieren vertical axis position
(X = 0), 473 mm below the horizontal tunnel axis, was measured by means
of a ±1 Bar DRUCK pressure transducer (see Figure 3.6).
The Mach number was determined from the stagnation pressure and the
static pressure using the established calibration relations. However, in this
case the calibration was not considered reliable due to the extremely high
blockage of the test section.
On the model, the pressure distributions were recorded at 217 holes in 7

sections as shown in Figure 3.1 above. The data was recorded using a 32 port
electronic pressure scanner type ZOC 23B/32Px from Scanivalve Corpora-
tion. The device has 32 transducers of ±50 psid capacity.
The tunnel parameters (P0 , Ps , T0 etc.) were passed through a 4 Hz low-
pass filter before being converted to digital values.
All the pressure transducers were calibrated before the test program by
using the existing calibration equipment – the Bell and Howell primary stan-
dard (dead weight tester) which has a precision of ±0.015% and the Texas
Instruments secondary standard with an accuracy of ±0.05%. The model
pitch mechanism was checked using the existing Wyler precision inclinome-
ter (±1’ accuracy).
Schlieren visualizations
The first phase of testing (from #7406 to #7427) was intended for the iden-
tification of buffet onset regimes by schlieren visualizations.
In order to identify the buffeting on the model and also to avoid flow block-
age/interaction caused by the solid walls, several Mach regimes and incidences
were tested and detailed global schlieren images were recorded. The schlieren
pictures provided qualitative information on the flowfield configuration and
proper identification of buffeting phenomena.
The active part of a test run was in the range of 20 to 35 seconds, depending
on the regime. The model was moved to the pre-established angles of attack
in step mode in order to allow the pressure scanners and schlieren system to
operate at a constant incidence.
The schlieren pictures were taken using a Nikon D200 camera equipped
with a Tamron lens having a maximum focal distance of 300 mm. The expo-
sure time was 1/1250 seconds, and color or graded black and white schlieren
filters were used.
Fig. 3.3 Schlieren pictures for 18% biconvex, 1 deg., Mach 0.762, Reynolds
6.78 mil
The buffeting conditions were established for Mach = 0.76, Re = 6.78×106

and 1◦ of incidence (Figure 3.3) and this was defined as the reference test
case for the biconvex symmetric profile.
Shock location on the model was varying from 55% to 85% chord length
on both upper and lower surfaces, alternating in the buffeting phenomena.
Pressure distributions on the tunnel walls
The objective of this phase of the test program (runs #7428 to #7431) was
to determine the pressure distributions on the tunnel walls upstream and
downstream of the test section at Mach numbers between 0.75 and 0.77.
The geometry of the solid wall test section is defined by the last por-
tion of the nozzle contour which for the subsonic regime is slightly divergent
with a halfangle of 0.1576◦ from a point situated at X = 1524 (total height
Z = 1186 mm) to the nozzle exit (X = −812.8) where the total height is
Z = 1199 mm.
Static pressure holes are provided along the axis of both vertical walls of
the flexible nozzle.
For subsonic operation the pressure holes no. 49 (manifolded together from
both walls) are used to measure the reference static pressure used to deter-
mine the Mach number. The holes are situated at 2108.2 mm upstream from
the schlieren axis. The reference Mach number was computed using the ref-
erence static pressure Ps and the stagnation pressure P0 as described before.
There is also a pressure hole (“Pp”) situated at the schlieren centerline
position (X = 0), at 473 mm below the center of the window. This pres-
sure corresponding roughly to the model maximum thickness is measured
by means of a ±1 Bar differential DRUCK transducer with respect to the
reference static pressure.
The variable diffuser has the upper and lower walls parallel and 1200 mm
apart while the vertical walls are made out of segments hinged to one another.
The second throat is used as a critical section to control the test section Mach
number. Static pressure holes are provided on the last segment of both vertical
walls – from the Second Throat to exit.
The measured scanivalve pressures were reduced to local Mach numbers.
Examples of pressure distributions on the tunnel walls are given in
Figure 3.4–3.6. The variable diffuser Mach number variation clearly indicates
the position of the shock wave system by which the supersonic flow becomes
subsonic at approx. 3 meters downstream of the second throat.
Pressure distributions on the profile
A series of tests – from #7432 to #7444 – were devoted to measurements

of the pressure distributions on the model upper and lower surfaces at
Mach = 0.76 in buffeting conditions. The geometrical configuration of the
Fig. 3.4 Mach number distribution along the Flexible Nozzle wall
Fig. 3.5 Pressure distributions along the Variable Diffuser walls
Fig. 3.6 Pressure distributions along the Variable Diffuser walls

model is represented in Figure 3.1. Static pressure holes were provided in

seven sections, each section having 31 holes at 10 mm intervals, as shown.
The chordwise axis X with the origin at the model leading edge is positive
towards downstream while the Y axis with the origin on the model centerline
is positive towards sections 5, 6 and 7.
The pressure holes were pneumatically connected to an electronic scanning
device type ZOC 23B/32Px from Scanivalve Corporation. The device has 32
transducers of ±50 psid capacity and is capable of a scan rate of 20 kHz.
In order to address the 32 transducers of the scanning device a special
electronic interface was designed and manufactured. This interface was con-
trolled by 6 digital output channels of a PIO-96 digital input-output module
from National Instruments.
The pressure on each port (transducer) was read once before switching to
the next transducer. All the channels were read at intervals of 3 to 4 millisec-
onds. A complete scan of the 32 transducers lasted approx. 120 milliseconds.
During most runs the model angle of attack was kept at the prescribed value
and the scan was repeated 150 times which lasted approx. 18 seconds.
The pressure coefficient distributions can be averaged for the entire run
and the values can be represented in the diagrams as average values. However,
Fig. 3.7 Average, maximum and minimum Cp , section 2, 3 and 6

since the shock waves are continuously moving, one must note that the aver-
age distributions do not represent the real flow at any moment. More useful
are the maximum and minimum values of the pressure coefficient (or p/pinf,
or local Mach number) for each hole over the whole run (see the examples in
Figure 3.7).
The Cp max and Cp min represent the limits between which the flow oscillates
when the instability (buffeting) occurs. At the same time, these curves can
be compared with the theoretical (computed) pressure distributions.
The spanwise pressure distributions were measured in a similar way for
selected positions on the profile chord.
Unsteady Pressure Measurements on the Profile
In order to investigate the unsteady pressures on the profile, the biconvex

symmetric model was equipped with 5 Kulite XCS-152–0.7 BAR differential
pressure transducers placed at 40%, 50%, 60%, 70% and 80% of the profile
chord, on the top surface of the model, as shown in Figure 3.1.
All Kulite transducers were excited in parallel with the same excitation
voltage of 10 VDC from a MOXON SRC signal conditioner. The signals were
connected to the National Instrument AT-MIO-64E3 data acquisition module
via SCXI-1120 isolation amplifiers. The Kulite transducer signals were filtered
with a 10 KHz low pass filter.
The reference side of all the Kulite transducers was connected via a long
tube to the static pressure Ps measured at the reference static pressure holes
(hole no. 49) on the side wall of the tunnel, the same pressure which was used
to compute the Mach number.
The scanning device was not used at this stage and no schlieren visualiza-
tions were performed.
A new data acquisition procedure was implemented, which allows the read-
ing of all channels at 32 microseconds intervals i.e. 31.25 KHz sampling rate.
At Mach = 0.76 and Alpha = 1◦ (Run no. 7455) the buffeting appeared
clearly in the Kulite signals, as shown in Figure 3.8 below.
Fig. 3.8 Run 7455, Mach = 0.76

Figure 3.8 above shows the KULITE signals in A/D counts. Figure 3.9
below shows the local Mach number values computed using the existing
calibrations.
Fig. 3.9 Run #7455 – Local Mach number – instantaneous values
The frequency of shock wave oscillation was found to be 78 Hz. The dis-
placement speed of the shock wave between 50% and 70% of the chord is
approximately 10 m/s.
At 80% of the chord the buffeting was less visible but still present
(Kulite K1).
No buffeting was recorded at Mach numbers 0.75 or lower. The pressure
oscillations were smaller in amplitude and the main frequency was approx.
600 Hz.
Tests with the Synthetic Jet Actuators (SJA)
In an attempt to control the SWBLI a number of 5 Synthetic Jet actuators

were installed on the biconvex symmetric wing, at 65% of the chord on the
top surface of the model, as shown in Figure 3.1.
The model was equipped with piezoelectric SJ actuators having a body
diameter of 25 mm, a diaphragm diameter of 20 mm, a 1 mm hole diameter
and 1.1 mm hole height. The diaphragms could be replaced by taking apart
the upper part of the model.
Several constructions of SJ actuators were tested and measured at vari-
ous regimes using various techniques – a thermo anemometer system for air
jet velocity measurements, a laser sheet system for visualizations, schlieren
visualizations of the air jets, microphones for sound level measurements etc.
During the SJA tests the unsteady signals from the Kulite transducers were
recorded in the same way as for the basic experiments, i.e. at 32 microseconds
for approx. one second of run, in order to detect changes induced by the action
of the SJ actuators excited with different wave shapes.
Fig. 3.10 Attenuation at 78 Hz
Movies of the schlieren visualizations were recorded with an ordinary video

camera and also with a fast (1000 frames/second) Photron camera.
The effects of the flow control by SJ actuators were expected to be seen as
alterations in the amplitude and frequency of the unsteady pressure signals
recorded by the Kulite transducers.
At all the runs with Mach = 0.76 the Kulite transducers indicated the
presence of the buffeting.
However, there were changes in the amplitude and frequency of the signals
recorded by the Kulite sensors depending on the excitation of the SJA.
The effect of the SJA was felt mostly by the Kulites K1 at 70% and K2 at
80% of the chord, i.e. downstream of the actuators. Most diagrams show the
presence of the excitation frequencies on the K1 and K2 signals, but their
effect on the other Kulites upstream of the SJA is more difficult to observe.
The effects were found in the form of frequency changes and amplitude
attenuation.
Using Fourier analysis it has been observed that during some of the flow
control tests the buffeting frequency (around 78 Hz) was attenuated (see
Figure 3.10). The attenuation was recorded differently by the Kulite sen-
sors, but sensors K1, K3 and K5 recorded attenuations at all runs compared
with the reference case.
Another visible effect was a shift in buffeting frequency as shown in
Figure 3.11 below.
Fig. 3.11 Frequency shift due to SJ actuators (Blue = Run 7473, White = Ref. run
7472)
3.3 Flow Simulations

All three partners performed flow computations using their own code and
different grids as well as turbulence models (see table 3.2).
It should be mentioned that all used slightly different model geometries or
wind tunnel presentations. EADS uses a straight squared tunnel cross section
upstream of the test section as well as for the part downstream of the sword.
Also a simplified connection between sting and model was used. INCAS used
the same simplified model mounting but simulate the total surrounding wind
tunnel comprising nozzle and diffuser. IMFT applied also the total wind
tunnel but the sting was taken off. It can be assumed that the different wind
tunnel geometries may have minor influence upon the results as long as the
proper flow conditions are insured in the reference section. The differences
between the EADS/INCAS and original mounting or even the neglect ion
of the sting may have an influence upon the shock motion and is discussed
further down.
EADS
EADS has much experience with computations of unsteady flows (Rieger

2005, Haase et. al 2006, 2007, 2009) by means of hybrid RANS-LES methods.
Before starting the final computations EADS performed a detailed conver-
gence study with regard to the inner convergence of the dual time stepping
approach. Beside the comparisons with the experimental data EADS did a de-
tailed comparison between the URANS and the DDES simulation. Parts of
these comparisons are shown below.
3.3 Flow Simulations 67
IMFT
IMFT performed 2D simulations with a generic configuration of the upper

and lower wind-tunnel walls, as well as 3D simulations with the exact geom-
etry of the wind tunnel. The buffet phenomenon has been well reproduced in
all cases, at frequency values comparable with the experiment. For the case
of the complete wind tunnel geometry, IMFT performed a detailed paramet-
ric study to establish the correct flow conditions in respect of the upstream
Mach number by adapting the outlet pressure. IMFT delivered results com-
parable to the experiments in respect of the buffet amplitudes (see figures)
and frequency (78 Hz) by using the URANS/Spalart-Almaras model. The
amplitudes of the oscillations have been improved by requiring an order of
convergence error of 5 × 10−4 instead of 7 × 10−3 iterations for the inner
loop in respect of the dual time-stepping scheme. The URANS simulations
provided a stable behavior with less small fluctuations than DDES. Further-
more, IMFT performed a case of buffet control by using blowing along the
half-span, that slightly attenuated the amplitudes of the lift coefficient.
INCAS
INCAS did URANS as well as LES simulations. They used an unstructured

tetra mesh with about five million nodes with URANS and 6.4 million for
LES, with a decomposition of the domain in 8 subdomains. The simulation
was based on a 2 step approach, where the first step was a global simulation
for the whole tunnel needed for setting up the boundary conditions in the
area for the test section.
INCAS has found reasonable agreement with the experimental data for the
mean, minimum and maximum pressure as well as for the unsteady values.
They found a motion frequency of 80.1 Hz for UURANS and 76.5 Hz for LES
being quite close to experimental value of about 78 Hz. The LES simulation
seems to show somewhat higher unsteady content.
INCAS has also performed numerical simulation for the synthetic jet ac-
tuators and for the cases for complex interaction for buffeting alleviation.
EADS CFD SIMULATIONS

Grid Generation
It should be mentioned that the geometry used by EADS differs from the
experimental shape mainly at the sting profile mounting. Also the geometry
of the channel differs slightly. Instead of an inlet nozzle and an exit diffusor
straight channel geometry is applied in both places. Some grid details are
given below.
Fig. 3.12 Grid and grid details on tunnel walls
Fig. 3.13 Profile surface grid and grid cut
The unstructured hybrid grid contains about 12 million nodes. The surface
discretization on the profile surface was adapted to the assumed region of the
shock motion. In space this was also done and special grid refinement regions
cover the vicinity around the rear part of the profile where the separation
was expected to occur and in the wake. These regions of high grid density
are needed to get reliable results for the hybrid RANS/DDES mode. The
same grid was used for URANS and DDES approach and details can be seen
in the Figures 3.14 and 3.15.
The difference in the sting profile shape is shown below. It can be expected
that the blunt face of the sting head in the experiment will influence the shock
motion compared to the smooth mounting in the computations on the left
hand side.
Fig. 3.14 Geometry comparison old and new CAD definition
Fig. 3.15 Locations of points of monitored flow properties on profile along the
tunnel side wall and in the wake
Simulations
The computations were performed using the TAU release 2006.1 (Schwamborn
et. al) in the DES version of EADS. To account for turbulence in the URANS
mode the Spalart Allmaras (SA) (Spalart 1994) turbulence model was applied
while in the hybrid RANS/LES mode the SA-DES (Spalart 1997) model
was used. To avoid grid induced switching between LES and RANS regions
the Delayed Detached Eddy (DDES, Spalart 2006) version was applied. The
time integration was done by a dual time stepping scheme of Jameson type
(Jameson 1991).
The simulation was started in the URANS mode using a large time step
size until a quasi steady result was reached. Then the time step size was
diminished in several steps reaching the final value of Δt = 4.0 × 10−5 sec.
The experimental sampling rate of the Kulite equipment was comparable

3.2 × 105 seconds.
The computations were continued until a quasi periodic behavior was
reached. After that the final data extraction was started.
The procedure described above was applied to both simulations. The only
difference was that the SA-DDES simulation started with the quasi periodic
RANS result after the first cycle of motion.
The time step size was chosen according to the experience made in a num-
ber of unsteady DES simulations and seems to be small enough to resolve
the unsteadiness in the shock induced separation regions. Perhaps it is by
far suitable for the shock motion frequency which was expected to be below
hundred Hz. In the URANS as well as in the DDES computations the same
time step was used.
While the chosen time steps size results in about 330 time steps per period
of shock motion the data extraction of instantaneous values was only done at
every tenth step resulting in Δt = 4.0 × 10−4 seconds or 33 extraction points
per period. This was done to reduce the amount of output data.
The computations were continued for at least five periods of shock motion
to get reliable statistical mean and variance values. This unsteady data was
saved for each time step to get relevant data for statistical analysis and to
compare with the values of the Kulite sensors of the experiment.
In the simulation much more points in the flow were monitored compared
to the number of Kulite sensors in the experiments. Overall sixty two points
were monitored located on the profile (5 pts. in 7 sections), in the wake
(5 pts. in 3 sections) and along the tunnel wall (12 pts). The locations of
the points are given in the figure below. The monitored values are density,
velocity components u, v and w and energy.
Parameter and Convergence studies
Before the real productive simulation were started test runs were performed
in order to find and determine proper code parameters to ensure convergence
over the total period of shock motion and to fulfil good convergence within
each time step.
Within the dual time step scheme the convergence of the force coefficients
were looked at using 25, 50 and 100 inner iterations. Also the influence upon
the periodic force behaviour was controlled. According to the behaviour of
lift, drag and pitching moment it was decided to finally use hundred inner
iterations, which result in perfect inner convergence.
Comparisons between URANS and DDES simulation
The URANS and the DDES simulations were performed using the same grid
and the same parameters. The differences in the periodic behaviour of the
forces (no such experimental data is available) between both are shown below.
The SA-DDES calculation results in higher amplitudes and a slightly lower
frequency. The mean value is the same except for the drag where the DDES
seems to be higher. As one will see further down the shock induced separation
region is much larger in the case of the hybrid result.
Fig. 3.16 Differences in lift, drag and pitching moment for several periods
As one might detect, the time of each period with is not constant. The
variation is larger with the hybrid DDES result. As a consequence the fre-
quency of the motion also varies and the mean values of both results differ
between 77 for URANS and 76 Hz DDES.
The following figures show the instantaneous vorticity distribution in
a slice and in 3D. The unsteady content of the flow is obvious larger in
the SA-DDES simulation. The slice and the 3D view show for this case much
more details and structures and the size of the separated region is enlarged.
Fig. 3.17 URANS (left) and DDES (right) 3D and 2D vorticity distribution
From the extracted instantaneous data the RMS values of the velocity
components and the pressure in slices were calculated. For the u-component
the regions of increased values above and below the profile mark the region of
shock oscillation. Differences between the URANS and DDES distributions
are clearly visible. Generally speaking the RMS values are higher for DDES,
they extend farer into the space and especially in the wake the oscillations
are much higher and persist longer. In the wake particularly the vertical and
lateral velocity component represents higher unsteadiness and three dimen-
sionality of the DDES result compared to the URANS simulation.
Fig. 3.18 RMS distribution of the velocity components DDES (top) URANS (bot-
tom) y = 150 mm
The following figures 3.19 and 20 show the differences between both com-
putations in the wake region. In the wake the pressure oscillations of the
DDES result are considerable higher compared to the URANS computation
while the mean values seem to be comparable. The first point at x/C = 1.0
is situated at the trailing edge while the last one is two chords away. The
URANS curves present oscillations which reproduce more or less only the
global shock motion frequency. The damping of the oscillation which is visi-
ble at the very right is much larger in the URANS case and show a remarkable
gap compared to the DDES result until 2 chords behind the profile. The drop
at the first point of the hybrid result is due to the damping of the surface
because the point is located at the trailing edge only 0.095 mm above the
contour.
The distribution of the power spectrum density represents the higher un-
steady content of the DDES result especially for x/C = 1.25 and 1.50. At
Fig. 3.19 Pressure oscillations and RMS values in the wake at y = 150 mm (DDES
very right, URANS middle)
Fig. 3.20 Comparison of PSD of pressure in the wake with distance to the trailing
edge y = 150 mm
the trailing edge (x/C = 1.0) the oscillations are damped by the present of
the profile surface and at x/C = 2 the differences decrease. At the trailing
edge and at the next station the first amplitude peak goes along with the
basic shock motion frequency while the next are not such distinct like for the
points on the profile. They are also no exact multiples of the main frequency
as above the wing. There seems to be a relative maximum at about 1000 Hz
in the DDES curve.
The last figure in this chapter shows the behavior of the pre shock Mach
number Msh and the location of the shock in section y = 0.15 for one period of
motion. Unfortunately such data could not be extracted from the experiment.
One can observe that the motion is anti parallel between upper (blue) and
lower (red) surface. The maximum Mach number is slightly higher for the
URANS result while the shock position is a little bit more downstream.
Fig. 3.21 Msh , Xsh and Xsep behavior within one cycle of shock motion
Influence of sting geometry
The comparison of the simulations with the experiments is limited by the

small amount of wind tunnel data being available.
The following pictures presents the comparisons of the minimum, maxi-
mum and mean value of the pressure coefficient Cp compared with the exper-
iment. For section y = 0.20 generally speaking the DDES simulation shows
slightly better agreement with the experimental data.
The differences over the first half of the section are probably caused by
different shock strengths leading to differences in the separation region re-
sulting in a different effective angle of attack. Obvious are the large discrep-
ancies downstream of the shock until the trailing edge. The offset starts at
x/C = 0.60 and is visible in the min, max and mean plot as well. The reason
is to be found in the shock motion of the simulations reaching much closer
to the trailing edge (x/C = 0.8) than in experiment.
These large differences are assumed to be partly caused by the differences
in the geometry of the sting profile mounting as already mentioned before.
The blunt face of the sting head in the measurements introduces a higher
pressure level at the rear part of the profile preventing the shock from moving
further downstream. The following picture for x/C = 0.30 compared to the
figure above shows a decreasing offset towards the tip obvious supporting this
assumption.
Fig. 3.22 Comparison of Cp values in section y = 0.20 m
Fig. 3.23 Comparison of Mach number values in section y = 0.30
Code description and Timings for all simulations
As mentioned above, the TAU-code from the DLR was used for all sim-
ulations. The TAU-code is a hybrid structured/unstructured second order
finite-volume flow solver for the Euler and Navier-Stokes equations in the
integral form (Schwamborn 2006). Various numerical schemes such as the
central scheme with classic or matrix dissipation or upwind schemes such as
AUSM are implemented. Additionally to an explicit Runge-Kutta scheme,
a point implicit LUSGS scheme is implemented to advance the solutions
in time. For convergence acceleration local time stepping, implicit residual
smoothing and full multigrid are implemented. Also the Jameson type dual
time stepping scheme is available for transient simulations. This method is
not limited to the smallest time step in the flow field and all acceleration
techniques mentioned above can be applied within the inner explicit loop.
A lot of one and two equation turbulence models as well as RSM, EARSM
and several DES models are applicable.
The in-house hardware consists of a LINUX cluster with 256 Intel Xeon
CPUs. The CPUs have a tact rate of 2.66 GHz and are mounted on
dual-boards with 2GB memory. As back plane a QUADRICS-interconnect

is available.
In the standard simulations 48 CPUs were used. For the URANS compu-
tations this results in a CPU-time of 20 minutes per time step, 3.45 hours per
data extraction step (ten basic time steps) or 114h for one period of shock
motion (about 330 basic time steps). The hybrid DDES simulation needs
slightly more effort. The timing was 32 minutes per physical time step, 5.4
hours for each data extraction pass and 178 h for one period.
IMFT CFD SIMULATIONS

IMFT performed CFD simulations for this test case with the URANS –
Spalart-Allmaras [4] model, and especially devoted a special attention to
take into account the exact 3D wind tunnel geometry. The computations
have been performed using the NSMB (Navier-Stokes Multi-Block) software
[18], which is a structured multi-block code created and upgraded by means
of a European consortium since early ‘90’s including IMFT since 2002.
The code NSMB is fully parallelised in MPI and the computations have
been carried out using up to 128 processors on the national supercom-
puting centres GRID5000, CINES (Centre Inter-universitaire National de
l’Enseignement Supérieur) and IDRIS (Institut du Développement et de
Ressources en Informatique Scientifique).
The results efficiently capture the buffet phenomenon and compare quite
well with the physical experiment in respect of the predominant frequency
and of the amplitude oscillations due to the shock motion as detailed in
the following. Furthermore, IMFT provided a study for buffet control by
employing blowing jets along the half – span distance of the wing, by using
the same real geometry and URANS modeling for the present test case.
Physical parameters
Air is considered as a perfect gas following Sutherland law for viscosity.

Boundary conditions have been determined to match a flow with Mach
number, M = 0.76, and Reynolds number, Re = 6.78 × 106 in the wind
tunnel test section where the flow field is practically uniform:
Horizontal velocity U∞ = 262 m/s
temperature T∞ = 293 K
pressure P∞ = 98500 Pa
density ρ = 1.17 kg/m3
In order to reach this state, the inlet Mach number at the entrance of the
nozzle had to be set to 0.39 and the pressure at the outlet of the diffuser
had also to be imposed after performing detailed tests, to reach the physical
Mach number M = 0.76 at the far upstream region from the wing.
Grid Generation
A first test-case was performed in 2D with a generic configuration of the upper

and lower walls, by using a grid size of 82,000 cells, as well as for an initial 3D
configuration of grid size of 5.8 M points. The final test-case geometry taken
into account by IMFT is the full wind tunnel with curved nozzle and diffuser.
The difference between this computational domain and the real geometry is
the absence of the sting for the computation, because this would require
a high increase in the number of grid point in respect of structured grid
configuration. The sting geometry is more efficient to take into account by
an unstructured grid formulation, as is the case in the EADS computations.
The IMFT’s multi-block structured grid contains about 5 million nodes.
Fig. 3.24 Computational domain of the exact wind tunnel geometry and grid (5M
points), including zoom around the profile
Numerical parameters and convergence studies
The temporal discretisation is carried out by a 3rd -order Runge-Kutta scheme

and the time-marching procedure is performed by a dual-time stepping
scheme by Jameson [2]. The space discretization of the advection terms is
carried out by an implicit Roe scheme (3rd order upwind), [19] with MUSCL
scheme of van Leer [20]. The viscous fluxes are treated by central difference
scheme.
The computations were run on parallel (MPI) architectures using up to
128 processors.
The simulation time step is Δt = 10−5 s; Computations have been carried
out in sufficient time duration so that a significant number of cycles in respect
of the period of the buffet phenomenon be simulated, to provide converged
statistics. Data extraction of instantaneous values was done at every 5 time-

step Δt) which represents about 250 extractions per period of the buffet
phenomenon.
The convergence tolerance is found to influence much more the amplitude
of the oscillations than the buffet frequency. Indeed, the buffet frequency
found 78 Hz is in good agreement with the experiment and has been achieved
by using a tolerance convergence of order 7×10−3 and an order of 15 iterations
for the inner loop of the dual time-stepping. However, this case provided
a weak shock oscillations amplitude than in the experiment (Figure 3.25).
A second run has been performed by setting a convergence criterion of order
5 × 10−4 and an order of 70–100 inner-loop iterations. These simulations
provided the same buffet frequency and an improved shock amplitude in
comparison with the experiments (Figure 3.25). This improvement is also
seen in the Clift and Cdrag coefficient oscillations, that are comparable in
the second case with the EADS/DDES simulations. It is noticeable that in
the framework of NSMB code, the first order of moderate tolerance for the
residuals error is widely sufficient for a considerable number of unsteady
motion test-cases, including SWBLI. It is therefore an interesting point to
note that for several buffet phenomena, a finer tolerance threshold is needed.
Fig. 3.25 Comparison between higher (left) and lower (right) convergence rate;
pressure fluctuations
In the following, the IMFT results presented correspond to the higher

convergence rate, unless it is differently specified.
Data monitoring
The following points are monitored in the flow to produce time histories and
spectra:
• on the lateral wall in the sense of span edges, at X/C = −4.5; −3, −1.5;
0 and 1.
• on the cut y = 150 mm on upper and lower surface at X/C = 0.4; 0.5; 0.6;
0.7; 0.8
• on the cut y = 150 mm on the wake at z = z(trailingedge) + 9.5 mm at
X/C = 1; 1.25; 1.5; 2; 3.
Results
Preliminary results using a H-type mesh allowed reproducing the buffet phe-
nomenon, both in 2D and 3D computations, on a biconvex 18% thick airfoil
at angle of attack of 3 degrees.
Fig. 3.26 Clift (left) and Cdrag (right), high convergence rate – IMFT results
Fig. 3.27 Clift oscillations; IMFT lower convergence rate in comparison with
EADS simulations
Fig. 3.28 H-type 2D an 3D first meshes
Fig. 3.29 Drag and lift coefficients on the wing Re = 1×107 , M = 0.75, alpha= 3◦
This preliminary test-cases run also with the URANS/SA model, with
Reynolds number Re = 1 × 107 , M = 0.75, alpha = 3◦ . The drag and lift
coefficient oscillations show capturing of buffet phenomenon with a significant
amplitude.
The buffet phenomenon is described by means of the iso-Mach contours
shown in the above figure. In the following, the study is performed by using
the full wind tunnel geometry. For the real test-case with the full experimental
domain, adaptation of the inlet and outlet boundary conditions had to be
made, to get the appropriate pressure and Mach number in the test-section
of the wind tunnel.
The boundary conditions in IMFT’s study are the following:
Total Temperature at inlet = 326.847 K
Total Pressure at inlet = 144522.39 Pa
which correspond to an approximate Mach number equal to 0.39
At outlet, pressure is imposed to 1.1 bar
Fig. 3.30 Iso-Mach contours (left) and horizontal U velocity (right) illustrating
transonic buffet over the biconvex airfoil at Re = 1.0 × 107 , M = 0.75, alpha= 3◦
Fig. 3.31 Averaged pressure coefficient on upper (left) and lower (right) surface
of the wing
Fig. 3.32 Comparison of pressure coefficient with the experiment at span section
y = 0.1 mm
Fig. 3.33 Comparison of pressure coefficient with the experiment at span section
y = 0.2 mm (top) and y = 0.3 mm (bottom)
Buffet control
IMFT performed a first study attempting attenuation of buffet oscillations

by employing blowing jets along the half-span of the wing. Figure 3.37 shows
modification of the separation downstream of the body and a slight reduction
of the Clift oscillations.
Fig. 3.34 Pressure signal vs time on upper and lower surfaces at several locations
X/C = 0.4, 0.5, 0.6, 0.7 and 0.8
Fig. 3.35 Pressure signals versus time in the wake

Fig. 3.36 Instantaneous Mach contours corresponding practically to extreme po-

sitions of the shock motion
Fig. 3.37 Lift coefficient oscillations (left) and iso-U velocity contours in the sep-
arated region downstream, under the effect of blowing jets applied at half-span,
Ublow = 0.05U∞
INCAS SIMULATION
Grid Generation
The simulations for the buffeting case were started in a hybrid approach,
using an initial global simulation for the global tunnel flow (coarse mesh, 1.3
million thets, shock located in the second throat as requested by the wind
tunnel operation). This simulation was needed in order to assess the farfield
boundary conditions for the main test section (Figure 3.38).
In the second step we have used a fine mesh for the domain of the solid
wall test section, with special treatment on the solid walls and the model
surface. The detailed mesh was based on 5 million points (aprox.), using
domain decomposition in 8 domains (Figure 3.39).
The final mesh generated is using unstructured tetrahedras and has the
following characteristics (Figure 3.39):
Fig. 3.38 Global coarse mesh
Fig. 3.39 Domain reduction for CFD analysis
• 4.381.256 points
• 576.318 points on solid surfaces
• 218.628 points on the wing
The global domain is decomposed in 8 sub-domains using MeTHIS tool and
the global solver uses Schwarts overlapping technique. The overall number of
points in the overlapping regions is 176.136.
Fig. 3.40 Mesh details – model surface

Simulations
DxUNSp code is an in-house code development at INCAS. It is an unstruc-

tured code, using domain decomposition based on Schwartz technique. The
code is based on a choice of Roe and Osher finite volume schemes, using
second order schemes with limiter.
For URANS computations we use a four-stage Runge-Kutta scheme with
local and/or global time step selection. Turbulence model used for these com-
putations is based on a 2 equation k-eps. model that includes several non-
linear extensions.
Solution analysis and convergence is considered using detailed residual
analysis and several acceleration techniques.
For both URANS and LES cases, numerical computations were used in
step 1 to reproduce a buffeting flow configuration in the tunnel, for airfoil
at 1 incidence, Mach = 0.76 and Reynolds = 10 mil. (Figure 3.41). This
was necessary in order to impose specific boundary conditions in step 2, in
a domain localised around the model in the test section. (Figure 3.40), using
specific formulations for URANS and LES cases.
The simulation was started in both URANS and LES cases using a global
time step formulation, where the global time step was considered as the
minimum local time step competed in a previous iteration on the global
domain. The order of magnitude of the time step during the computations
was close to Δt = 1.5 × 10−5 (non-dimensionalised value).
All computations were performed until a quasi periodic motion of the
shock was reached, also identified by a cvasi-periodic variation of the global
numerical residual. The main interest in comparing numerical data with ex-
periments is related to the proper identification of the buffeting phenomena.
From numerical point of view this can be analysed through the evaluation
of the variation in time of the global integer of the pressure on the model
(Figure 3.43).
Fig. 3.41 3D simulation for the global experiment

Fig. 3.42 LES simulation (section 5) – CPmed distribution
Fig. 3.43 Global variation of lift – buffeting identification (SJ control off/on)
Boundary conditions
For URANS analysis, Riemann invariant boundary conditions are used at

both the inlet and the outlet of the channel. Such a boundary condition
should avoid reflection at the inlet and outlet regions.
The quantities imposed at these boundaries are not the ones encountered
in the experimental configuration, since no information was available so that
this type of information could be used. However, final data from numerical
analysis may be compared with local data on the solid walls measured in
several locations.
For LES case, upstream boundary conditions were considered based on
a channel flow developed using URANS results. The computations took into
account a number of 5 cvasi-periods in URANS cases and recorded the flow
generated on a length of 1.2 meter, equivalent to the cross dimension of the
Fig. 3.44 Global buffeting simulation – URANS, Iso-Mach (section 3)
tunnel. These values were stored as time dependent boundary conditions for
LES analysis.
Solid walls treatment was considered using generalized Reichard wall laws
for both URANS and LES computations.
For numerical simulations for the flow control cases, actuators are identified
by the locations of the vertices on the surface. In every region, an array of
actuators is operating in identical conditions. This makes an array of SJ to
act like an individual actuator.
Operating frequencies are considered in the range of 0 to 1500 Hz and the
maximum blowing/suction top speed is supposed to be in the range of 0 to
150 m/s. For numerical evaluations, nondimensionalizations were performed
for the reduced frequency in the range of F + = 1 to 10.
In order to use such SJ actuators for numerical simulations, global velocity
profiles have been computed and recorded in order to enable a lower compu-
tational effort and to avoid complex discretisations. Such approach has been
already investigated with successful results in previous applications.
Comparisons between URANS and LES simulation
The URANS and LES simulations have been performed using the same global
parameters for the global time step. For the basic case, URANS analysis was
used in order to define a set of upstream boundary conditions compatible for
LES simulations. The set of recorded data was corresponding to a cvasiperi-
odic motion averaged on 5 cycles.
Global pressure distribution compared for 3 gates (15, 18 and 20) are
presented in Figure 3.45.
Fig. 3.45 URANS-LES comparison – Spanwise pressure distributions
The amplitude and phase of the total lift on the model from CFD analysis
has been identified as follows:
– fURAN S = 80.1 Hz
– fLES = 76.5 Hz
– fEXP = 78 Hz (aprox.)
Code description and Timings for all simulations
DxUNSp code is an in-house code development at INCAS. It is an unstruc-

tured code, using domain decomposition based on Schwartz technique. The
code is based on a choice of Roe and Osher finite volume schemes, using sec-
ond order schemes with various limiters. This platform is based on a unified
formulation that enables a wide choice for simulations types and/or turbu-
lence models. For URANS cases, a specific implementation for the 2 eg. k-ε
model was used. For LES, a beta-gamma scheme is used and Germano dy-
namic model.
In order to have a comparable accuracy for the time integration, the time
integration scheme is deliberately taken explicit, using a 4th order optimized
Runge-Kutta formulation with global time step strategy.
DxUNSp code is implemented on a Beowulf LINUX cluster with 64
AMD X2 CPUs at 3 GHz. For simulations, domain decomposition was used
(Schwartz) in 8 sub-domains. Post-processing is based on domain reconstruc-
tion and custom tools developed for TecPLOT.
For the URANS computations, solution was obtained in a reasonable time
(aprox 25 h for a buffeting cycle, after “convergence”). In LES computations
the time for a buffeting cycle was aprox. 43 h. Solutions was extracted after
7 cycles, so global computation time for URANS was aprox. 175 h, and for
LES aprox. 200 h. In the LES case another additional 15 h were needed for
specific upstream boundary layers computations.
3.4 Comparisons with Experiments and Cross Plotting

The minimum, maximum and mean Cp and Mach number is given below. The
differences between the EADS simulations and the experiments are mainly
blamed to the different sting geometry as mentioned already above. The
IMFT simulations have been performed without the sting geometry. The
runs corresponding to the higher convergence rate shows show good agree-
ment with the experiment, especially for the fluctuation pressure minimum
amplitudes and the shock positions (Figures 3.46, 3.47).
The comparison of the periodic behavior of the lift is performed between
EADS, IMFT and INCAS simulations, since no such experimental data are
available for this quantity. The comparison (figures 3.48, 3.49 and 3.50) show
a good agreement between EADS/URANS and IMFT/URANS results (these
last ones corresponding to the higher convergence rate). The maximum am-
plitudes of the lift coefficient are of order 0.062 for both EADS and IMFT and
the minimum amplitudes are of order −0.03 for EADS/URANS and −0.04
Fig. 3.46 Comparison of minimum, maximum and mean Cp and Mach number
between experiment and simulations of EADS and IMFT
3.4 Comparisons with Experiments and Cross Plotting 91
Fig. 3.47 Comparison of minimum, maximum and mean Cp and Mach number
between experiment and simulations of INCAS
for IMFT/URANS. INCAS Clift oscillations either by URANS or LES show

a similar amplitude of order (+0.04, −0.03), close to the above mentioned
values.
Fig. 3.48 Comparison of periodic lift behavior for INCAS URANS/LES
The following sequence of plot shows the comparison of the unsteady pres-
sure oscillations of the simulations compared with the Kulite data of the
experiment for section y = 0.15. In the first figure one can see some principle
differences between the EADS URANS, DDES and experiment. The oscilla-
tions in the experiment are by far larger than in the theoretical results. Only
the DDES computation shows some comparable behavior of the data. The
DDES result also matches better the maximum level. The period time for all
is slightly different and not constant over all periods resulting in varying gap
for the time of the pressure rise between all three results.
Fig. 3.49 Comparison of EADS URANS and DDES instantaneous pressure at

x/C = 0.4, 0.6 and y = 150 mm
Furthermore, the previously mentioned comparison of IMFT pressure fluc-

tuation with the experiment shows also a good agreement (Figure 3.46), see
also Figure 3.51 below, where in addition, lower pressure peaks by IMFT
computations are in good agreement with the experiment.
As can be expected from the pressure coefficient comparison already shown
the agreement will be not as good for the more downstream points. Both
EADS simulations show for x/C greater than 0.6 a much stronger expansion
or in other words a much higher Mach number (next figure) compared to
the measurement. The experiment reaches only a slightly supersonic Mach
number at x/C = 0.70 and is completely subsonic at x/C = 0.80. Also one
can observe in this figure that not only the period time varies from period to
Fig. 3.50 Kulite K3 pressure variation – Test 7464, CFD and SJ – S3

Fig. 3.51 Comparison of EADS and IMFT instantaneous pressure for total section
y = 150 mm
period but also the magnitude of flow properties themselves. This fact can
be very nicely detect for the maximum Mach number in at x/C = 0.7 and
0.8 shown in the next figure. The IMFT result again suffers of the reduced
shock motion. The pressure amplitudes are much smaller and only at point
x/C = 0.60 really pronounced. Also the behavior is very smooth containing
no higher frequencies.
Looking at the overall values of PRMS and Pmean one can recognize again
the increasing discrepancies towards the trailing edge. For the points x/C =
0.4, 0.5 and 0.6 the agreement is good, for the DDES very good. Because
the shock motion extents till x/C = 0.8 in the simulations the theoretical
RSM is here much higher and the mean value too low because of expansion
in front of the shock. The differences we see here in the mean value perhaps
show the same tendency like we saw in the mean, minimum and maximum
Cp -comparisons.
The time series of the instantaneous flow properties allow also the calcu-
lation of the spectral distribution of the pressure Power Spectrum Density.
The distribution has distinct maxima clearly visible at the lower frequen-
cies. These peaks are located at the frequency of the shock motion and mul-
tiple of it. This is of course the dominant frequency of the total problem. The
small differences in positions are due to the slightly different period time of
each result. The unsteady content at higher frequencies decreases from the
experiment over DDES to URANS. At the lower frequency part all curves
show more or less the same amplitudes for x/C < 0.6 while for x/C = 0.7
Fig. 3.52 Comparison of EADS URANS AND DDES instantaneous Mach number
for total section y = 150 mm
Fig. 3.53
and 0.8 the experimental result is clearly below the simulations. The reason
is again the enlarged regions of the shock motion in the numerical simulation,
which extend closer to the trailing edge.
To compare the experimental schlieren pictures with the simulation EADS
tried to find within the available instantaneous slices comparable cases with
regard to the shock position at the lower surface. It was assumed that in
the experimental setup the sting mounting has lower influence on the shock
motion of this side although a coupling will be present. One should keep in
mind, that the schlieren pictures represent a view through the total wind
tunnel and show therefore the three dimensional structure of the shock while
in the simulations only one section at y = 0.15 m is shown. Compared are
Fig. 3.54 Comparison of the EADS PSD distribution of pressure y = 150 mm
the experimental pictures with gradrho, the gradient of the density of the
simulations. First one can again observe the higher unsteady content of the
flow in the DDES result compared to URANS. Also the separated regions
are larger with DDES and closer to the experiment.
The qualitatively agreement between experiment and simulations is good
as long as the shock at the upper surface is far from the trailing edge.
Fig. 3.55 Comparison between schlieren pictures and gradrho distribution of the
EADS simulations
3.5 Conclusion
Experiment
Unfortunately the experiments deliver only a limited amount of data to com-

pare with CFD. The data available comprise minimum, maximum and mean
values in several sections and the unsteady pressure readings in one section
with 5 Kulite sensors. For the selected test run a shock frequency of 78 Hz
was found.
The use of a synthetic jet to influence or reduce the shock motion showed
only a minor effect.
3.5 Conclusion 97
CFD Simulations
EADS
EADS has performed an SA-URANS and SA-DDES simulation of the current

test case for the flow conditions
M∞ = 0.76, α = 1.0◦ ; Re = 6.78 × 106
at which the experiment shows a stable and distinct shock motion. There ex-
ists a difference in the shape of the sting mounting between experiment (blunt
face) and simulations (smooth intersection; figure 3.6). This shape discrep-
ancy is assumed to influence the shock motion mainly on the upper surface
leading to a more forward location of the extreme backward shock position
in the experiment. This leads to large differences between experimental and
theoretical results for x/C greater than 0.6.
The grid generation produced a 12 Million node unstructured mesh with
refinement in all relevant parts of the flow field being influenced by the shock
motion or exhibiting flow separation. Some geometry simplifications were
applied like e.g. for the inflow nozzle and the outflow diffusers of the wind
tunnel.
Detailed convergence and parameter studies guarantee reliable results for
both URANS and DDES simulations.
In addition to comparing the agreement with the experiment selected com-
parisons between the URANS and DDES results are shown for flow regions
and flow properties not being available from the measurements. As can be
expected, the unsteady content of the DDES result is higher.
The comparison with the experimental data shows for the front part of the
profile, that the minimum, maximum and mean Cp distributions are in good
agreement, with DDES giving a small improvement compared to URANS.
The instantaneous data in the simulations present much less unsteady
content especially for the URANS result. The DDES computation fits better
in pressure level and shows some comparable higher frequency oscillations.
Both theoretical results show strong shocks even at x/C = 0.7 and 0.8 while
in the experiment at x/C = 0.7 only a weak and at 0.8 no shock is present.
The comparison of the total PRMS and Pmean distribution along the profile
shows also some improvements with DDES and good agreement for x/C 0.4,
0.5 and 0.6.
Looking at the PSD distributions again the agreement for the lower x-
stations is quite good. The amplitudes also agree quite well at the lower
frequency band while at higher values the unsteady content especially of the
URANS simulation is too low. The level of the experimental PSD for x/C 0.7
and 0.8 is obvious lower than in the computations because the shock motion
never reaches this positions with full strength.
The experimental and theoretical global shock frequencies are relatively
close together, showing 78 Hz in the experiment, 77 Hz for URANS and 76 Hz
for DDES.
IMFT
IMFT performed simulations with the URANS/SA model. First two-dimen-

sional simulations without channel show similar shock oscillations like the
experiments. A comparative study of convergence rate indicates that the con-
vergence does not play a significant role in capturing the frequency value but
the amplitudes of the shock oscillations. The final IMFT results performed
for the real 3D wind tunnel geometry without the sting accurately capture
the buffet phenomenon, especially the lower pressure fluctuation levels. They
show a shock motion quite in agreement with experiments in respect of fre-
quency (78 Hz), amplitudes oscillations and extrema of the shock motion.
The motion frequency is found to be 78 Hz, which surprisingly close to 78 Hz
of the experiment. The modification of buffet in respect of blowing actuation
along half-span indicates a slight attenuation of the oscillation amplitudes.
INCAS
INCAS performed URANS as well as LES computations. The predicted buffet

frequencies of 80.1 Hz for URANS and 76.5 Hz for LES are in good agreement
with the measurements. The Kulite readings for LES show larger amplitudes
and higher frequency oscillations compared to URANS. This is in accordance
with the differences found by EADS between URANS and DDES. Looking at
the minimum, maximum and mean Cp distributions the pressure upstream of
the shock is in good agreement with the experiment showing some improve-
ments with regard to the minimum and maximum shock position for LES.
Downstream of the shock the pressure differences increase.
Cross plots
Because only IMFT delivered data for cross plotting in a reasonable time cross
plotting was only possible between experiment, EADS and IMFT. Compar-
isons of INCAS simulations are shown in separate figures. The comparison
between EADS and IMFT shows a good agreement in respect of frequency
and amplitude oscillations. It is also shown that DDES approach carried out
by EADS has the tendency of overpredicting the minimum amplitudes of the
pressure fluctuations.
URANS compared to DDES/LES
The EADS DDES as well as the INCAS LES result show a more rich
statistical content than URANS simulations. Furthermore, IMFT URANS
simulations achieve a good description of the buffet phenomenon. Obvi-
ous improvements compared to EADS/URANS can be noticed. The overall
References 99
amplitudes as well as unsteadiness at high frequencies are improved by LES

and DDES.
This leads to the conclusion that DDES and LES are able to perform
reasonable flow analysis for shock induced buffeting and turn out to have
some advantages compared to URANS simulations.
References
[1] Schwamborn, D., Kroll, N., Heinrich, R.: The DLR TAU-code: recent Appli-
cations in Research and Industry. In: European Conference on Computational
Fluid Dynamics, Egmond ann Zee, ECCOMAS CFD 2006 (2006)
[2] Jameson, A.: Time Dependent Calculations Using Multigrid with Application
to Unsteady Flows past Airfoils and Wings. AIAA Paper 91–1596 (1991)
[3] Rieger, H.: First Experiments with Detached-Eddy Simulations in the Aero-
nautics Industry. In: Symposium on Hybrid RANS-LES methods, Stockholm,
Schweden, pp. 14–15 (July 2005)
[4] Spalart, P.R., Allmaras, S.R.: A One-Equation Turbulence Model for Aerody-
namic Flows. La Recherche Aerospatiale 1, 1–21 (1994)
[5] Spalart, P.R., Jou, W.H., Strelets, M., Allmaras, S.R.: Components on the fea-
sible of LES for wings and on hybrid RANS-LES approach. In: First AAFOSR
International Conference on DES/LES, Ruston, Louisiana, August 4-8 (1997)
[6] Spalart, P.R., Deck, S., Shur, M.L., et al.: A new Version of Detached-Eddy
Simulation, Resistant to Ambiguous Grid Densities. Theoretical and Compu-
tational Fluid Dynamics 20(3) (2006)
[7] MacDevitt, J.B.: Supercritical Flow about a Thick Circular-Arc Airfoil, NASA
TM 78549
[8] MacDevitt, J.B., Levy Jr., L.L., Deiwert, G.S.: Transonic Flow about a Thick
Circular-Arc Airfoil. AIAA Journal 14(5), 606
[9] Levy Jr., L.L.: Experimental and Computational Steady and Unsteady Tran-
sonic Flows about a Thick Airfoil. AIAA 16(564)
[10] Haase, W., Aupoix, B., Bunge, U., Schwamborn, D. (eds.): FLOMANIA –
A European initiative on flow physics modelling. Notes on Numerical Fluid
Mechenics and Multidisciplinary Design, vol. 94. Springer, Heidelberg (2006)
[11] Haase, W., Peng, S.-H. (eds.): Advances in Hybrid RANS-LES Modelling.
Notes on Numerical Fluid Mechenics and Multidisciplinary Design, vol. 94.
Springer, Heidelberg (2008)
[12] Haase, W., Braza, M., Revell, A. (eds.): DESider – A European Effort on
Hybrid RANS-LES Modelling. Notes on Numerical Fluid Mechenics and Mul-
tidisciplinary Design, vol. 103. Springer, Heidelberg (2009)
[13] McDevitt, J.B., Levy, L.L., Deiwert, G.S.: Transonic flow about a thick
circular-arc airfoil. AIAA Journal 14(5), 606–613 (1976)
[14] Tijdeman, H.: Investigation of the transonic flow around oscillation airfoils.
NLR TR 77090 U, National Aerospace Laboratory, The Netherlands (1977)
[15] Nae, C.: Numerical Simulation of the Synthetic Jet Actuator. In: ICA 0.266,
ICAS 2000, Harrogate, UK (2000)
[16] Mabey, D.G.: Oscillatory flows from shock induced separations on biconvex
airfoils of varying thickness in ventilated wind tunnels. In: AGARD CP-296,
France, September 14-19, pp. 11.1–14 (1980)
100 References
[17] Nae, C.: Efficient LES using Beta-Gamma Scheme and Wall Laws. In: ICFD
2001, Oxford, U.K. (2001)
[18] Vos, J., Chaput, E., Arlinger, B., Rizzi, A., Corjon, A.: Recent advances in
aerodynamics inside the NSMB (Navier-Stokes Multi-Block) consortium. In:
36th Aerospace Sciences Meeting and Exhibit, AIAA Paper 1998–0802, Reno,
USA (1998)
[19] Roe, P.L.: Approximate Riemann Solvers, Parameter Vectors, and Difference
Schemes. J. Comp. Phys. 43, 357–372 (1981)
[20] Van Leer, B.: Toward the ultimate conservation difference scheme: A second-
order sequel to Godunov’s method. J. Comp. Phy. 32, 101–136 (1979)
Chapter 4
NACA0012 with Aileron
(Marianna Braza)
4.1 The IoA Experiment
Introduction
This study integrates the Institute of Aviation (IoA) experiments and the nu-
merical simulations and turbulence modelling of the University of Liverpool
(ULIV)and by the Institut de Mécanique des Fluides de Toulouse (IMFT).
The IoA experiment focused on unsteady flow characteristics and buffet phe-
nomena arising as the result of the shock wave / boundary layer interaction
in transonic flow over a NACA 0012 airfoil equipped with an aileron. The
transonic buffet is a natural and self-sustaining oscillation of the shock wave
and the separated flow region, caused by pressure fluctuation.
Experiments are performed in the N-3 wind tunnel of the IoA. The perfo-
rated lower and upper walls of the wind tunnel test section were used in the
early stage of the project. However, to achieve explicit boundary conditions
for numerical calculations in further wind tunnel tests perforated walls were
replaced by solid ones as agreeds the UFAST partners.
The first objective is the investigation of the effect of a flap deflection on
the transonic flow without and with shock wave induced separation includ-
ing the buffet onset boundary. The second objective is the determination of
unsteady flow field characteristics at flow conditions above the buffet onset
boundary. Then periodic flap oscillations have been used aimy to attenuate
of the transonic buffet.
Contributors: G. Barbut and M. Braza, M. Miller, W. Kania, Y. Hoarau,
G. Barakos, A. Sévrain.
Experimental set-up
The N-3 wind tunnel of the Institute of Aviation (IoA) is a trisonic blown-
down wind tunnel with partial flow recirculation and Mach number range
springerlink.com
102 4 NACA0012 with Aileron (Marianna Braza)
from 0.3 to 2.3. The Mach number range of subsonic and transonic flow is
0.3 to 1.15. The tested NACA 0012 model is supported by rotatable mecha-
nism which provides the continuous sweep of the angle of attack during run.
The transonic test section has a square cross section of 0.6 m × 0.6 m and
a length of 1.58 m. The details of the test section with the NACA 0012 model
installed are shown in Figure 4.1.
The NACA0012 airfoil model equipped with 22.6% of airfoil chord flap and
modified gap is shown in Figure 4.4. The model’s chord is equal to 180 mm
and flap length is 40.69 mm. The axis of rotation of the main part of the
model is at 35% of the chord. There is a smaller, 0.3 mm gap between airfoil
main part and flap leading edge. The flap leading edge radius is 4.69 mm.
The axis of the aileron / flap rotation is at 80% of the airfoil model chord.
The span of the model is 596 mm.
The stagnation pressure and temperature were measured in the settling
chamber downstream of the honeycomb and nets set.
Fig. 4.1 Sketch of the test section with installed NACA 0012 flapped model in the
N-3 wind tunnel (all dimensions in mm)
Fig. 4.2 The geometry of the NACA 0012 model equipped with the aileron (all
dimensions in mm)
4.1 The IoA Experiment 103
The determination of the subsonic or transonic Mach number at each angle

of attack of the model was based on:
• Measured plenum chamber pressure which was checked during test section
calibration for the cases with perforated lower and upper walls,
• Mean value of measured side wall pressures at two reference pressure taps
on both side walls for solid lower and upper walls case.
There are independently activating flap oscillation:
• The amplitude of the flap deflection angle up to Δδf = ±10◦ ,
• Oscillation frequency of the flap, ff = 0 to 10 Hz.
From the wind tunnel tests carried out during the UFAST project it was
determined that the buffet frequency exceeds an acceptable limit of the os-
cillation frequency of the flap. This resulted in modifications of the control
mechanism for the aileron oscillations.
The slider-crank mechanism have been changed to allow for the increased
frequency of flap oscillations. The maximum calculated frequency of flap os-
cillations with the new mechanism was ff = 100 Hz and the calculated am-
plitude of the flap deflection was Δδf = ±2.5◦ . The protection system of the
electric motor driving the flap limits both the frequency and the amplitude
of flap oscillations because of the higher loads expected during wind tun-
nel tests. The modified control mechanism is presented in Figure 4.4. After
modification the aileron drive mechanisms delivers:
• Amplitude of flap deflection up to Δδf = ±2◦ ,
• Flap oscillation frequency ff = 0 to 35 Hz.
Fig. 4.3 Schematic of the control mechanism with the NACA 0012 model
Fig. 4.4 Aileron control mechanism

Comparison of the NACA 0012 airfoil wind tunnel tests

The results of the tests with solid section walls and for undeflected aileron
were compared with tests for a single-element NACA 0012 airfoil model per-
formed in the NASA Ames High Reynolds Number Facility [3]. The buffet
onset angles of attack αbo determined in both wind tunnels at the range of
Mach numbers M = 0.70 to 0.80 were compared in Figure 4.5. Very good
agreement was achieved at Mach numbers up to M = 0.78 (about 0.1◦ α
deviation). Somewhat greater difference occured at Mach number M = 0.80.
Fig. 4.5 Comparison of buffet onset boundaries for the NACA 0012 airfoil mea-
sured in IoA N-3 wind tunnel (solid walls) and Ames HRN Facility [3] (model
without flap)
Results of steady measurements over the NACA0012

with aileron / flap
The static pressure distribution on the airfoil was measured using three 16-
channel electronic scanners (EPS-16 HD Pressure System) installed outside
the model closed to the test section, and 48 pressure taps of 0.4 mm diameter
that were located on the upper and lower model surfaces.
Pressure distribution on the NACA 0012 model with undeflected flap at

M = 0.7
The comparison of the pressure coefficient distribution on the NACA 0012

model with modified gap in the test section with solid walls for angles of
attack α = 6◦ , 7◦ and constant flap deflection δf = 0◦ at Mach number
M = 0.7 is presented in Figure 4.6 (a-b).
Fig. 4.6 NACA 0012 model with modified gap and solid walls (Run 84/07)
Effect of the flap deflection on pressure distribution of NACA 0012 airfoil
Mach number M = 0.75. The influence of flap deflection angles δf = 0◦ to

6◦ on the pressure coefficient distribution on the NACA 0012 model with
modified gap in the test section with solid walls for selected angles of attack
α = 0◦ and5◦ for Mach number M = 0.75 is presented in Figure 4.7.
Mach number M = 0.7 and flap deflection δf = 0◦ and 6◦ . Some supplemen-

tary experiments were made for the purpose of numerical comparison. In this
series of experiments a small correction for the inclination angle of NACA
0012 model was input to remove differences of the pressure coefficient Cp on
the upper and lower sides of the model at angle of attack α = 0◦ . Pressure
coefficients Cp on the model at M = 0.70 and Re = 2.63 × 106 and the flap
deflection respectively, δf = 0◦ and 6◦ for angles of attack α = 0◦ to 7◦ are
presented in Figure 4.8.
Effect of flap deflection on the buffet onset of the NACA 0012 airfoil model
The buffet onset angle of attack was determined based on the divergence of
trailing edge pressure coefficient.
Figure 4.9 shows the effects of flap deflection on the buffet onset angles of
attack αbo of NACA 0012 airfoil model at Mach numbers M = 0.70, 0.75,
0.78 and 0.80.
Fig. 4.7 NACA 0012 model with gap and solid walls (upper surface)
Fig. 4.8 Pressure coefficients Cp on NACA 0012 at Mach = 0.70 and Re = 2.63 ×
106 and the flap deflection δf = 0◦ for various angles of attack α = 0◦ to 7◦
Fig. 4.9 Effect of flap deflection on buffet onset angle of attack of NACA 0012
model
In Figure 4.10 the effects of flap deflection on the buffet onset lift coeffi-
cients CL bo of NACA 0012 airfoil model at Mach numbers M = 0.70, 0.75
and 0.80 are shown.
Fig. 4.10 Effect of flap deflection on the buffet onset lift coefficient
Shadowgraph flow visualisation
The flow visualisation was performed at Mach numbers of 0.7, 0.75, 0.78 and
0.8 and for a range of angles of attack α up to seven degrees. A single-pass
shadowgraph system and a digital video camera for recording were used.
Photographs extracted from the shadowgraph movies at M = 0.70 and
angle of attack below (α = 2◦ and 4◦ ) and on the buffet boundary (α = 5◦
and 6◦ ) are shown in Figure 4.11. At α = 7◦ the airflow is fully stalled.
Fig. 4.11 Shock wave/boundary layer interaction on the NACA 0012 airfoil with
undeflected flap at Mach number M = 0.70 and Reynolds number Re = 2.69 × 106
Results of unsteady pressure measurements

The unsteady pressure measurements [4] were performed using two miniature
Kulite LQ-125 pressure transducers mounted on the upper surface of the
NACA 0012 airfoil model at 50% and 75% of the chord (approximate location
of the shock wave at buffet onset). An ESAM TRAVELLER data acquisition
system by ESA Messtechnik GmbH was used for acquiring the date and
analysing the results.
Buffet onset determination based on unsteady pressure measurements
The RMS pressure fluctuation at 75% of an airfoil chord point was used to
determine buffet onset of the NACA 0012 model with aileron / flap. The
RMS of pressure fluctuation versus the angle of attack α at Mach numbers
M = 0.70, 0.75 and 0.80 for the NACA 0012 model with undeflected flap are
shown in Figure 4.12.
Fig. 4.12 The pressure fluctuation RMS at the point of 75% of chord versus angle
of attack at Mach number M = 0.70, 0.75 and 0.80 for NACA 0012 model with
undeflected flap
The comparison of buffet onset angle of attack αbo determined based on

unsteady pressure measurements and static trailing edge pressure divergence
for the NACA 0012 model with undeflected flap are shown in Table 4.1.
Table 4.1 The comparison of buffet onset angle of attack
M 0.70 0.75 0.80

αbo unsteady 4.5◦ 2.8◦ 0.5◦
αstatic pTE 4.8◦ 3◦ 0.9◦
Effect of flap deflection on unsteady determined buffet onset angle of attack
The angles of attack of the buffet onset as determined by the RMS [5] of the
pressure fluctuations at 75% of the chord for Mach number M = 0.7 and for
flap deflections of δf = 0◦ and 6◦ are shown in Figure 4.13.
Results and analysis of unsteady pressure measurement at Mach number of

M = 0.7
The unsteady pressure measurements at 50% and 75% of the chord on the
upper surface of the NACA 0012 model were performed at Mach number
Fig. 4.13 Buffet onset determination based on the RMS of the pressure fluctuation
at Mach number M = 0.7 and for flap deflections of δf = 0◦ and 6◦
Fig. 4.14 Experimental signal (up); autoregressive modeling (AR) (middle), FFT
spectrum (down), M = 0.7, a = 6.8, Re = 2.63 × 106
M = 0.7 and a range of angle of attack from α = 0◦ to 6◦ . For unsteady pres-

sure data acquisition and results analysis the same apparatus as mentioned
above was used. The bandwidth of 1000 Hz was analysed.
The power spectral densities of the measured pressure fluctuation ats point
of 50% of the chord over angles of attack between α = 0◦ and 6◦ are shown
in Figure 4.14. The buffet onset frequency is about 74 Hz at angle of attack
α = 5◦ . Further increase of the angle of attack α in the range of the buffet
causes an increase of the buffet frequency up to about 110 Hz.
The power spectral densities of the measured pressure fluctuations at 75%
of chord over angles of attack α = 0◦ to 6◦ are shown in Figure 4.15. The
buffet onset frequency of about 82 Hz at the angle of attack α = 4.5◦ was
determined. Further increase of α increases the buffet frequency as for the
previous case.
Fig. 4.15 AR signal processing for forced flaps frequency f = 35 Hz and flap
deflection of 2◦
Advanced signal processing

The experimental signals display a strong chaotic character, interspersed by
peaks of more organised events during the buffet phenomenon (figure 4.14).
This complex behaviour was enhanced by the fact that the upper and lower
walls ware not porous but solid. IMFT has applied advanced signal process-
ing techniques, as autoregressive modelling (AR), Burges algorithm [6] and
Morlet’s wavelets analysis [7], to extract the organised part and to assess the
contribution of the organised and random parts in the physical process.
The AR model indicates organisation of the flow according to a bi-modal
pattern with a predominant frequency of order 95 to 100 Hz, corresponding
to the buffet phenomenon and a first subharmonic of order 50 Hz. The third
sub-figure in Figure 4.14 shows the FFT results that provide a multitude of
frequencies for the present case.
Fig. 4.16 Morlet’s wavelets signal processing the forced flaps at frequency
f = 35 Hz amd deflection of 2◦
It is noticeable that these frequency values, evaluated towards the end of

the study, did not influence the computations that have been carried out
independently and suggested values of order 95 Hz, in a good agreement with
the experiment as presented in the following section.
Figure 4.15 shows a similar behaviour obtained from the Morlet’s wavelet
analysis, according to the previously mentioned bi-modal pattern. For the
case of forced flaps oscillation at a frequency f = 35 Hz, the organisation
of the signal shown by AR and by Mormet’s wavelets indicates a shift of

the organised motion around the forcing frequency (35 Hz), as well as the
appearance of a first harmonic of this frequency, of order 70 Hz.
4.2 Numerical Simulation of the IoA Test-Case
Summary
The transonic flow case of the IoA was studied for clean and flapped wing
configurations. From the very start, the influence of the wind tunnel walls
was identified as a key problem and a special mesh was put together to allow
for modelling of the walls and easy comparison with free-stream conditions.
At the time this chapter was compiled few experimental data were available
for comparisons and the exact flow case conditions were not specified. For
this purpose, the University of Liverpool (ULIV) and IMFT focussed on
a set of cases and comparisons with test data were made where possible.
The numerical grids are those of the ULIV, modified also by IMFT to get
more refinement in the upper and lower walls in respect of the y + value (less
than 1). Therefore, the IMFT final grids were of the order of 3.3 Million and
3.75 Million cells.
Special attention was attributed to correctly mesh the gap region and to
provide CFD on a computational domain with wind-tunnel walls.
Overall, the employed turbulence models, including the URANS Spalar
Almaras (SA), k-ε and k-ω models, the Detached Eddy Simulation DES-k-ω
and the DDES-SA (delayed DES with the SA model for the URANS part)
were able to predict the buffet frequency and produce the low-frequency
unsteadiness of the buffet. The effect of the flap actuation was also quanti-
fied with actuation frequencies lower, and up to the buffet frequency tested.
The comparisons with the experiment have been carried out by means of
advanced signal processing methods (IMFT) based on Autoregressive mod-
elling (Burges algorithm) and Morlet’s wavelet analysis, to extract properly
the buffet frequency and to quantify the organised and chaotic turbulence
processes in the experimental signal. Thanks to these techniques, a bi-modal
organisation has been depicted, where the complex flow process is governed
by a fundamental frequency of order 100 Hz and of a lower frequency close
to the first subharmonic, of order 50 Hz.
Results and discussion – CFD of the IoA flow case

A view of the employed mesh is shown in Figure 4.17. The mesh was generated
using the ICEM-CFD Hexa grid generator. ULIV used the in house PMB code
and IMFT the NSMB (Navier-Stokes MultiBlock code [20] in which IMFT
has implemented higher-order turbulence models [21].
4.2 Numerical Simulation of the IoA Test-Case 115
The resolution of the wing surface was higher than the side walls which
were put at exactly the same location as the experiment. Care was taken to
allow adequate grid resolution near the front 20% of the NACA-0012 wing
section. This would allow for shocks to be resolved and capture any separation
due to shock/boundary-layer interaction. The work focussed at relatively high
incidence angles (up to 6.8 degrees) for which buffet clearly appears and for
this reason, the upper surface of the wing was also meshed at a finer level. In
total 2.65 million cells were used and the baseline grid included 18 blocks at
each span-wise station. Multiples of this set of 18 blocks were then used for
parallel efficiency. Figure 4.17 presents the dimensions and topology of the
baseline mesh along with a close view of the mesh near the aerofoil.
Fig. 4.17 View of the baseline mesh constructed in ICEM-CFD Hexa. Top: overall
domain configuration and block boundaries; bottom: mesh around the flap gap.
Investigation of Flow Conditions
For this test case, several conditions were computed to establish the validity of
the employed method for flap actuation and the potential of the trailing edge
flap as a flow control device. Preliminary computations using the URANS
version of the URANS k-ω model specified in [1] suggested that the flow was
buffeting at incidence angles around 5 degrees and Mach number of 0.7. This
was further ascertained using computations with and without the wind tunnel
walls. This was a very encouraging result since buffet predictions are known
to be sensitive to the employed turbulence model and mesh. In addition, it
was reported by the IoA that buffet was evident at a Mach number of 0.7
and angle of attack of approximately 5 degrees.
Figure 4.18 presents an overview of the flow configuration obtained for the
wing with the tunnel walls present. The surface colour represents pressure
and the one can see the strong shock present near the leading edge of the
wing at a Mach number of 0.7 and 5 degrees of angle of attack. The strength
of the shock is enough to cause separation of the boundary layer and the flow
near the side-walls of the tunnel is dominated by flow re-circulation regions.
An interesting remark is that the near the mid-span of the wing, there is an
extended region where the flow is shows little span-wise variation at least for
this set of conditions. This suggests that one could perhaps impose a zero
gradient condition in this direction and model a slice of the wing near the
centre. This approach could result in lower computational cost. However, the
top and lower walls have a stronger effect and as was investigated, their effect
must be accounted for.
The effect of the upper and lower walls can be seen in Figure 4.19 where
the surface pressure coefficient on the mid-span is plotted. The Mach number
was kept at 0.75 and the angle of attack was zero degrees for this test case.
At the time of the computation, the exact Reynolds number was not known
but an estimate was made at 2.81 million, based on the chord of the wing. As
can be seen, computations with the upper tunnel wall and with free-stream
conditions result in different surface pressure coefficients. The free-stream
computation shows lower values near the leading edge while the wind tunnel
walls result in higher values. This significant influence suggested that the
wind tunnel walls must be modelled for this test case. Overall, the presence
of the walls resulted in higher levels of surface pressure coefficient and it
appears that without the wall present, the onset of the buffet phenomenon is
shifted to higher Mach numbers and angle-of-attack. It is therefore vital to
include the upper and lower walls, if any comparison is to be made against
experimental data.
To further ascertain this, computations of several incidence angles were
made and the results were compared with some of the experimental data made
available by the IoA. For this set of computations, span-wise symmetry was
imposed, with and without the upper and lower walls and, in addition, full 3D
results were also obtained. Figure 4.20 presents some of these comparisons
Fig. 4.18 Overview of flow configuration (flow recirculation and corner vortices
near the side walls of the tunnel) for Mach number of 0.7 and angle-of-attack of 5◦ .
Streamlines (left) above the leading edge and iso contours of pressure coefficicient
(left). Iso-U contours (right and down).
Fig. 4.19 Comparison of surface pressure coefficient distributions obtained with

and without upper tunnel walls. M = 0.75, a = 0◦ . Re ≈ 2.81 million based on the
aerofoil chord. ULIV results use the new URANS model.
Fig. 4.20 Comparison between CFD and experimental data for the effect of the
tunnel walls on the obtained surface pressure coefficient distribution. The full 3D
configuration appears to be the most suitable for computations and results in fair
agreement with the experiments. The slotted mesh with a gap between the main
and flap elements was used.
Fig. 4.21 Blended flap configuration (a) un-deformed mesh, (b) mesh deformed
by −2 degrees and (c) mesh deformed by +2 degrees
and one can see that the full 3D configuration compares better with the
experimental data. An interesting observation is that the flow near the flap
may be influenced by the exact geometry of the gap between the main element
and the flap of the wing.
Starting from the above observation, further computations were made to
quantify this effect. Figure 4.21 shows a blended flap configuration which can
be used along with trans-finite mesh interpolation to model the actuation of
the flap.
This configuration results in a C0 surface continuity and a simpler multi-
block topology. It also approximates better the improved design of the IoA
where the gap between the flap and the main element was blocked. The qual-
ity of the mesh is preserved during the actuation and the employed technique
was found to be adequate for flap deflections approaching 10 degrees before
some deterioration of the mesh quality was observed.
Figures 4.22 and 4.23 present results obtained using the slotted and
blended flap configurations. The conditions were kept at a Mach number
of 0.75 and Reynolds number of 2.81 million. The angle of attach of the wing
Fig. 4.22 Comparison between blended and slotted flap configurations for the
surface pressure coefficient distribution. Slotted flap configuration has a small effect
near the trailing edge. Re ≈ 2.81 × 106 .
Fig. 4.23 Wall-pressure coefficients for angles of attack from 0◦ to 4◦ , URANS/k-ε

with Chien’s damping functions; no buffet obtained – steady state flow
was changed between 0 and 5 degrees (where the onset of the buffet was
close). At the same time the flap was deflected by three degrees. A smaller
set of computations was performed using a slightly higher Mach number of
0.78. For all cases, blended and slotted flaps were used.
For the same test case, some flow visualisation is shown in Figure 4.24 for
the flow region near the slot. The conditions were selected so that the pressure
difference between the lower and upper wing surfaces drives the flow through
the gap. As a result a small flow distortion is shown. Interestingly enough,
this flow distortion is also evident in the surface pressure coefficient plots
shown in Figure 4.22.
IMFT has carried out a first set of computations with the URANS/k-ε
model with Chien’s damping functions, for a first set of low angles of attack.
The results, concerning the airfoil with gap are shown in figure 4.23. There
is no buffet for angles of attack up to 4◦ .
Fig. 4.24 Flow visualisation near the trailing edge of the blended (a) and slotted
(b) flap configurations. Flap deflected by 3◦ .
Computations beyond the buffet onset
The onset of buffet was first investigated using a mesh without the upper and
lower walls. At the time, the exact conditions of the test were not known and
a zero flap deflection was selected. The Mach and Reynolds number were kept
at 0.7 and 2.63 million, respectively and the angle-of-attack of the wing was
slowly increased. Figure 4.25 shows results for the lift coefficient (multiplied
by span) from these computations. As can be seen, at 4.8 degrees of angle of
Fig. 4.25 Results for buffeting case with and without the upper and lower wind
tunnel walls. Buffet is present at about 6.8 degrees of angle of attack. For compu-
tations including the wind tunnel walls, buffet was present at lower angles of about
5 degrees. Lift coefficient multiplied by span is shown.
attack, the flow becomes steady and there is no buffet. At one degree higher
(5.8 degrees) the flow buffets initially but the buffet amplitude is reducing
with the simulation time. It was only at about 6.8 degrees where buffet was
present and the amplitude of the lift oscillations remained constant with time.
Including the walls, the buffet was present at lower angles of attack. As
can be seen in Figure 4.25b, the flow was steady at 4 degrees while results
for this case suggest a well-established buffet at about 5 degrees. This result
is encouraging and agrees with previous observations made about the effect
of the walls on the obtained surface pressure distribution.
The walls confine the flow and result in slightly higher suction values of the
surface pressure coefficient. A stronger shock is present near the leading edge
of the wing and separation due to shock/boundary-layer interaction appears
at lower incidence angle. Due to the complexity of the flow and the lengthy,
unsteady flow computations required, no attempts were made to compare the
buffet onset at the same wing loading instead of the same angle-of-attack.
It is evident thought from the results of Figure 4.25 that the lift coefficient
oscillation amplitude for the case without walls at 6.8 degrees is about the
same as the oscillation amplitude at 5 degrees, if walls are included. This
observation led to the question about the buffet mechanism. From a first
flow visualisation of the results, it appears that the presence of the tunnel
walls is not changing the buffet mechanism itself. The conditions where buffet
occurs may be different though the fundamentals of the flow are similar.
A further observation is that the fundamental frequency shown in the results
of Figure 4.25 is about 90 Hz though this depends on how the results are
scaled. At the time the report was compiled, there was no information from
the IoA about the tunnel temperature or the flow velocity in the tunnel. It is
therefore quite difficult to scale the results to a specific frequency. Assuming
no tunnel pre-heating or cooling a value of about 90 Hz was obtained by ULIV.
Figure 4.26 presents some flow visualisation for the buffeting cases. Two
sets of results are shown for the flow in the mid-plane of the tunnel. As can be
Fig. 4.26 Flow visualisation using iso-Mach lines during buffet. (a) 5 degrees of
angle-of-attack and (b) 6 degrees of angle-of-attack. The results were obtained using
the k-ω model, M = 0.7, Re ≈ 2.63 × 106 .
seen, regardless of flow conditions, a strong shock is present near the leading
edge of the wing. The shock separated the boundary layer and the separation
region grows moving the shock slightly upstream. The shock is then strength-
ened and a larger separation occurs extending towards the trailing edge. This
re-distributes the momentum in the boundary layer resulting in some shock
alleviation and consequently some displacement of the shock downstream.
The computations for this test case were also repeated using the full DES
model. However, as the project advanced, a slightly modified set of conditions
was put forward. Computations were therefore undertaken for an incidence
angle of 6.8 degrees at the same Mach number of 0.7 and 2.63 million of
Reynolds number.
Figures 4.27 to 4.29 show the time history of the wing loads obtained from
this refined set of computations. The case was computed long enough for
several buffet cycles to be obtained (at least 6 are visible in Figure 4.27).
The results for this case were also visualised in Figures 4.30 and 4.31 where
the averaged data obtained from all cycles are used. The fundamental buffet
frequency of 90Hz is still present and interestingly enough several more fre-
quencies are also present up until 450 Hz. It has to be noted that the employed
time step allowed for frequencies up to 1000 Hz to be resolved.
The flow reversal near the mid-plane of the wing and walls is clearly visible
in Figure 4.31.
The compressibility effects and the unsteady SWBLI (ShockWave Bound-
ary layer Interaction) upstream of the gap are shown by means of iso-U ve-
locity component on Figure 4.28, together with the wall pressure distribution
coefficient.
A zoom of the flow within the gap region is shown in Figure 4.29, to-
gether with the unsteady pressure coefficients at tdifferent pressure side
and suction side positions, upstream and downstream of the gap. As the
Fig. 4.27 Instantaneous iso-U contours and wall pressure distribution, M = 0.73,
a = 5◦ , Re ≈ 2.65 × 106
Fig. 4.28 Instantaneous iso-U contours in the gap region (up); unsteady pressure
coefficients, wing with gap, M = 0.73, a = 5◦ , Re ≈ 2.65 × 106
Fig. 4.29 Lift and drag coefficients, URANS-SA and DDES-SA, flow parameters
as in previous figure
phase-averaged Navier-Stokes equations have been used in the URANS ap-

proach, the above time-dependent evolutions are phase-averaged and have
attenuated high-frequency noise appearing in the experimental signals. The
buffet frequency obtained by the above IMFT results is of order 95 Hz, in
good agreement with ULIV results and with the physical experiment.
Fig. 4.30 RMS of spanwise pressure coefficient illustrating the compressibility

effects in the shock region (left) and iso-U contours (right), flow parameters as in
Figure 4.28
Fig. 4.31 Iso-streamlines along the span illustrating the 3D flow structure (left)
and on the median plane (right)
Fig. 4.32 Mean wall pressure coefficient at 50% and 75% spanwise positions (left
and middle), experimental wall pressure coefficient at mid-span (right), URANS –
SA model
The mean pressure coefficients at two different spanwise positions (Fig-

ure 4.32) as well as the rms spanwise variation of the pressure (Figure 4.30)
illustrate the compressibility effect. The pressure coefficients are in good
agreement with the experiment.
Flap actuation for flow control
Due to delays with the release of the conditions for the flap actuated cases,
Liverpool decided to impose a flap actuation based on the obtained buffet
frequency. Efforts focussed in two main areas where the flap was oscillating
at an amplitude of 2 degrees at the buffet frequency and at lower frequency
taken as 1/3 of the buffet.
Figure 4.27 (right), shows the influence of the phase in application of the
flap motion at the natural buffet frequency. it is found that the buffet mech-
anism finally follows in phase the flap motion oscillations.
Furthermore, the blended flap was used and results are shown in
Figures 4.33 and 4.34. The flap actuation of the present work did not signifi-
cantly affect the flow buffet. Figure 4.33 shows the results for the case where
the flap was actuated at the buffet frequency. After an initial time where the
flow adapts from a partially-converged steady-state to buffet, the loads show
an almost periodic response. The flap motion is also shown in Figure 4.33
where as shown, buffet is still present.
The results of Figure 4.34 are more encouraging, in the sense that the flap
appears to introduce extra frequencies in the flow and induce loads variation.
Figure 4.35 compares the obtained results with respect to the initial un-
controlled signal. Some reduction of the loads is present for the low frequency
case.
A very good comparison of the mean wall pressure coefficient with the
DES-SA is obtained (Figure 4.36, ULIV). Comparison of the flap effect (fixed
and actuation with forcing frequencies of 30 Hz) is shown in Figure 4.37
(results by ULIV) by means of the aerodynamic coefficients. It has been
shown that a slight attenuation of the Clift amplitudes is achieved by the flap
motion.
Figure 4.38 shows the influence of forcing at 100 Hz (very close to the nat-
ural buffet frequency), actuated at twi different phases, first at a minimum
of the Clift coefficient and secondly at a maximum. Both cases lead to in-
crease of amplifications of the lift coefficient. Figure 4.38 (right) shows the
3D structure of the shock region along the span in the first forced case.
Figure 4.39 shows the wall pressure fluctuations within the gap and at 75%
of the chord, at mid section of the span in the case of forcing by 100Hz, (first
phase case). It is shown that the selected flap motion leads to an increase of
the amplitudes and of the absolute mean values of the pressure fluctuations.
Fig. 4.33 Flap oscillation of 2 degrees at the buffet frequency (a) Clift , (b): Cdrag .
M = 0.7, Re = of 2.63 × 106
Fig. 4.34 Flap oscillation of 2 degrees at a third of the buffet frequency (same
parameters as in Figure 4.33)
Fig. 4.35 Comparison of controlled and uncontrolled cases (flow parameters as in

previous figure)
Fig. 4.36 Comparison of mean wall pressure of the DES-SA computations with
the experiment
Fig. 4.37 Comparison of the flap oscillations at 30 Hz of forcing frequency and

deflection of 2◦ with the fixed flap case
Fig. 4.38 Flap actuation at 100 Hz of forcing frequency and deflection of ±2◦
at two different phases (max/min) of the Clift natural buffet oscillations (left),
isosurface of negative velocity along the span (right), M = 0.73, a = 5◦ , Re =
2.63 × 106
Fig. 4.39 Wall pressure at 100 Hz of forcing frequency and flap deflection of ±2◦
in the gap and at 75% of chord at mid-span distance, as well as comparison with
static flap case, M = 0.73, a = 5◦ , R = 2.63 × 106
References 131
4.3 Conclusions
A detailed physical experiment has been carried out by the Insitute of Avia-
tion (IoA) concerning the transonic interaction around a NACA0012 profile
with trailing edge aileron (flap). An experimental data base has been pro-
vided that allowed evaluation of the mean and fluctuating pressure values,
these last ones are responsible for the buffet phenomenon. The unsteady wall
pressure measurements allowed advanced signal processing, especially using
Wavelet analysis and Autoregressive modelling, that extracted from the com-
plex chaotic behaviour due to the turbulence, the organised frequencies of the
buffet phenomenon. A special attention has been attributed to examine the
influence of a low amplitude (2◦ ) flap motion, forced at a low frequency
(35 Hz), much lower than the buffet frequency (order of 95 Hz), to attenuate
the amplitudes of buffet.
Detailed numerical simulations have been carried out by respecting the
exact wind tunnel geometry, by the University of Liverpool, ULIV and the
Institut de Mécanique des Fluides de Toulouse, IMFT. The approaches that
have been used in the context of URANS and DES, show ability of captur-
ing the buffet frequency. Comparisons were first carried out between the two
sets of simulations without knowing the experimental values and afterwards,
detailed comparisons with the experiment, showing a good agreement. Fur-
thermore, the flap motion was forced at a frequency close to the natural
buffet, as well as at a frequency of 30 Hz. A slight attenuation of the aerody-
namic coefficients has been achieved. This first study opens a promising way
towards a further significant attenuation or control of the buffet, by using
this kind of simulation approaches within an optimisation kernel, to estimate
appropriate forcing in respect of frequency, amplitude and phase, concerning
the presently simulated buffet phenomenon.
References
[1] Batten, P., Goldberg, U., Chakravarty, S.: Sub-grid turbulence modeling for
unsteady flow with acoustic resonance. AIAA Paper No 00–0473 (2000)
[2] Miller, M., Kania, V.: Parameters for aerofoil and aileron oscillations, UFAST
Deliverable D3.1.3 (March 2009)
[3] Barakos, G., Drikakis, D.: Numerical simulation of transonic buffet flows using
various turbulence closures. International Journal of Heat and Fluid Flow 21(5),
620–626 (2000)
[4] Bourguet, R., Braza, M., Hoarau, Y., Harran, G., El Akoury, R.: AIAA J.
(2007); J. Fluids & Struct. (2009)
Part II
Nozzle Flows
Chapter 5
Nozzle Forced Shock Oscillations with
Wall Bump (Reynald Bur)
5.1 Introduction
The purpose of the present experimental study is to test the effects of control
devices – mechanical vortex generators (VGs) – both on a steady shock wave
and on a forced shock oscillation configurations interacting with a separated
boundary layer. Oscillation of the shock wave is forced thanks to a periodic
variation of the downstream throat section of the channel given by a rotating
elliptical shaft. The unsteady case under study is a forced shock oscillation
at 30 Hz which produces a shock oscillation amplitude of around 30 mm. The
quasi-steady shock wave configuration is obtained for downstream conditions
leading to a large separation of the boundary layer. Numerical simulations
of the forced shock oscillation configuration are performed both by URANS
and DES approaches, respectively for the reference case and a co-rotating
VG control device.
5.2 Presentation of the Experiments

Experimental set-up
Experiments are performed in the S8Ch wind tunnel of the ONERA Meudon
Center (see Figure 5.1a). This facility is a continuous wind tunnel supplied
with desiccated atmospheric air. The transonic channel is 100 mm high and
has a span of 120 mm at the entrance of the test section. The test set-up
consists of a rectilinear upper wall and a lower wall equipped with a contour
profile (bump). The shape of the bump has been specially designed to induce
a strong interaction between the boundary layer and the shock when it takes
place at the level of the rear part of the bump. Such an interaction induced
an extended separated zone. The stagnation conditions were near ambient
pressure and temperature: pst = 0.96 × 105 ± 300 Pa and Tst = 300 ± 10 K.
The unit Reynolds number is around 14 × 106 . The nominal Mach number,
at the rear part of the bump, is equal to 1.45.
springerlink.com
136 5 Nozzle Forced Shock Oscillations with Wall Bump (Reynald Bur)
Nearly-sinusoidal pressure perturbations were introduced at the down-

stream end of the channel by periodic variation of the second throat section
thanks to a rotating elliptical shaft located near this throat, in the middle
of the channel (see Figure 5.1a). This system caused forced shock wave os-
cillations at a known adjustable frequency. In these experiments, the shock
oscillation frequency is equal to 30 Hz, which produces a shock oscillation
amplitude of around 30 mm.
Fig. 5.1 Test set-up in the S8Ch wind tunnel

5.2 Presentation of the Experiments 137
The origin of the co-ordinate system is at the beginning of the lower wall
(see Figure 5.1b). The X-axis is along the lower wall in the streamwise direc-
tion, Y is normal to the lower wall and the bump chord and Z is along the
spanwise direction (Z = 0 in the median plane). The rectilinear upper wall is
inclined from −0.55◦ to the X-direction. The elliptical (2×3.6 mm axis) shaft
is located at X = 575 mm near the second throat section.
Control devices
Mechanical vortex generators (VGs) are tested to control both a steady shock
wave and a forced shock oscillation interacting with the separated boundary
layer. The VGs are located in the spanwise direction of the test set-up along
a line situated 10 mm downstream of the bump crest, at XV G = 261.37 mm.
So they are implemented upstream of the shock foot region and the boundary
layer separation. The VGs are triangular elements whose angular position is
fixed at 18◦ regarding to the main flow direction. The effects of co-rotating
VGs are compared to those of counter-rotating VGs (see Figure 5.2); several
spacing between elements have been retained, mainly referred to literature
based on Pearcey [1] and a review of Lin [2]. An important parameter taken
into account in the study is the VGs height, noted h. Conventional VGs (with
h/δ = 1) and sub-VGs (with h/δ = 0.5) are tested, δ being the incoming
boundary layer physical thickness (equal to 4 mm). Finally, six configurations
of mechanical vortex generators are under study in these experiments: two of
co-rotating VGs and four of counter-rotating VGs. They are represented in
Figure 5.3 and all their geometrical parameters are given in Table 5.1.
Fig. 5.2 Definition of the mechanical vortex generators

Fig. 5.3 Six configurations of VGs under study
Table 5.1 Geometrical parameters of the mechanical vortex generators
Co-rotating VGs Counter-rotating VGs

CoC1 CoS1 C1 C2 S1 S2
h/δ 1 0.5 1 1 0.5 0.5
l/h 2.5 2.5 2.5 1.25 2.5 1.25
L/h – – 3 1.5 3 1.5
λ/h 6 6 10 5 10 5
number 5 9 3 5 5 11
Flow investigation techniques

Flow field visualizations
A schlieren apparatus is used to visualize the flow field and control the shock
wave positions in the test section. Schlieren visualizations are performed by
means of a high speed camera Phantom V4.1 coupled with a spark light (spark
duration: 20ns). The high-speed camera characteristics are: a resolution of
512×512 pixels and an acquisition speed of 1000 frames per second.
Coloured oil flow visualizations are performed only for the steady shock
wave configuration, for the reference case and several VG devices, allowing
flow topology comparison.
Pressure taps and sensors
The lower wall of the test set-up is equipped with 39 continuous pressure
taps located on a line at 10 mm from the median plane of the test section
and 12 unsteady pressure sensors (named P and G) located on the median
plane and in the spanwise direction of the test section, in the shock oscillation
region (see Figure 5.4). For the flow control experiments, the pressure tap at
X = 265 mm does not exist due to the presence of the VG devices and the
G1 and G2 sensors are not be used.
The pressure taps have a diameter of 0.4 mm and are connected to
StathamT M transducers via rubber tubes. The unsteady pressure transducers
Fig. 5.4 Sketch of the pressure taps and Kulite sensors positions along the lower
wall of the channel
are absolute or differential 15psi KuliteT M XCS093 sensors with a diameter

of 0.8 mm.
LDV system and phase-averaged technique
Probings of the forced shock oscillation configuration for the reference case
without control have been performed by using a two-component laser Doppler
velocimetry system (LDV) synchronized with a signal emanating from the
rotating shaft [3]. The longitudinal extension of the acquisition domain is:
285 ≤ X (mm) ≤ 445, which includes the shock oscillation zone. Velocity
measurements were obtained with the velocimeter operating in forward scat-
tering mode, a Bragg cell unit being used to produce frequency shift of 15MHz
in order to determine the velocity direction. The radius of the probe volume
was estimated to be around 0.2 mm, which leads to reliable measurements
at 0.3 mm of the wall. The flow was seeded by submicronic DEHS (Di Ethyl
Hexyl Sebacate) particles introduced in the settling chamber of the wind
tunnel in order to assure an homogeneous seeding.
The phase-averaged technique is applied to LDV measurements to char-
acterize the velocity field. It can distinguish between the “coherent” motion
related to the periodic excitation and a random fluctuating part. The velocity
U (X, t) is broken down into three terms:
U (X, t) = U (X) + Ũ (X, t) + U (X, t)
where U (X) is the ensemble average component, Ũ(X, t) the cyclic compo-
nent and U (X, t) the fluctuating component. The phase-averaged component
is defined as:
U (X, t) = U (X) + Ũ (X, t)
The remaining fluctuating component should be interpreted as residue
characterizing events, which are not in phase with the main signal. In the
present case, the phase-averaged velocity is evaluated using a time informa-
tion. One oscillation period is divided into 60 intervals of 6◦ each. The begin-
ning of the period is given by an electronic circuit, which locates a marker
placed on the rotating shaft. The use of the TSI RMR (Rotating Machinery
Resolver) then allows for each laser sample to be arranged in its arrival time
in the period.
Tested configurations and boundary conditions

The effect of mechanical vortex generators was tested for a steady shock and
for a forced shock wave oscillation configurations interacting with a separated
boundary layer.
The inlet flow conditions could be chosen as the stagnation conditions for
a computational domain starting at the beginning of the converging part of
the wind tunnel, where the flow velocity is very low. However, LDV probing
has been performed at X = 135 mm, in the front part of the bump where the
flow is still subsonic. At this station, the boundary layer on the bump has
the following characteristics: δ = 3.9 mm; δ ∗ = 0.46 mm; θ = 0.25 mm and
Hi = 1.63.
The outlet flow conditions are those localized in the second throat section.
The quasi-steady shock wave configuration was obtained for downstream con-
ditions leading to a large separation of the boundary layer. The quasi-normal
shock location in the channel is 12.5 mm downstream of the end of the bump
on the lower wall; the downstream conditions to obtain this shock location
are a height of the second throat equal to 92.7 mm with the elliptical shaft
fixed in its horizontal position.
The unsteady case was a forced shock oscillation at 30 Hz which produces
a shock oscillation amplitude of around 30 mm. For the reference case, the
height of the throat is Y = 93.4 mm, the rotating shaft not being taken
into account in this value. In order to approximate the periodic pressure
signal delivered by the elliptical shaft, the forced downstream pressure could
be modelled by a planar pressure wave moving with a sinusoidal motion,
at a frequency f = 30 Hz, around an averaged pressure pav given by the
experiment. Then, the time evolution of the downstream pressure is given by:
p = pav + Δp, with: Δp = δpav sin(2πf t)
Main experimental results

A synthesis of the experimental results is carried out in [4], and this section
points out the main results obtained for the steady shock and the forced
shock wave oscillation configurations in presence of control devices.
Steady shock wave configuration
Figure 5.5 compares the schlieren visualizations between the reference and
the controled cases with counter-rotating VG-S2. In the reference case (see
Figure 5.5a), the interaction region is characterized by a large λ-shock struc-
ture in the core flow. The boundary layer is destabilized with massive sepa-
ration and large vorticed structures are developed in the shear stress layer.
For the control case (see Figure 5.5b), the size of the λ-shock structure is re-
duced compared to the reference case. The trails of the vortices generated by
the VGs are observed near the curved wall. The extension of the separation
region is reduced under the effects of VGs and the mixing layer is grow-
ing. The presence of VGs generates expansion waves followed by moderate
compression waves in the channel flowfield.
The lower wall pressure distributions are plotted in figure 5.6 for the ref-
erence case without control and the six VGs devices. The reference case
Fig. 5.5 Schlieren visualizations for steady shock wave configuration w/o con-
trol (a) and with VG-S2 (b)
Fig. 5.6 Wall pressure distributions for steady shock wave configuration with and
withouot VGs
distribution, plotted in yellow dots, reveals the existence of a large separa-

tion region corresponding to a quasi-plateau pressure level after the intense
recompression of the flow. The separation point is located near X = 325 mm
and the reattachment point around X = 390 mm. When control by VGs is
applied, the plateau pressure seems to disappear and the efficiency of VGs
to delay the separation region increases with the number of vortices created
by VGs. Then, the VG-S2 device (11 pairs of counter-rotating sub-vortex
generators) seems to suppress the separation in the near median plane of
the test section. Even in this configuration with a large number of VGs in
the spanwise direction, the pairs of vortices do not merge in the boundary
layer upstream of the interaction region. Moreover, the flow perturbations
– expansion and compression waves observed on schlieren visualization (see
Figure 5.5b) – generated by the VGs have a slight intensity and only have
a local effect on the pressure distributions. In terms of drag penalties, the
way to reduce the effects of these perturbations is to impose the height of the
VGs at a value less than the sonic line of the incoming boundary layer.
Coloured oil flow visualizations have been performed for the reference case
without control and with several VG devices. A comparison between the ref-
erence case and the control case with VG-C1 is made in Figure 5.7a. One
Fig. 5.7 Oil flow visualizations for steady shock wave configuration w/o control
and with several VG devices
notices an important modification of the flow separation pattern: the sepa-

ration line located in the rear part of the bump is replaced by a corrugated
separated line due to the presence of intense vortices. The flow seems to be
partially reattached under the action of VGs vortices. The merging of two
vortices created by two close VG devices to form a large vortex structure,
is well evidenced by its footprint on the wall. The effect of VG height is
discussed in Figure 5.7b by a comparison between the conventional VG-C2
and the sub-VG-S1. The VG height seems to have a slight influence on the
flow topology, especially on flow separation extension. One notes a signifi-
cant interaction with the corner flow, which is not controlled in this study.
Then, a comparison between co-rotating VG-CoS1 and counter-rotating VG-
S1 is carried out in Figure 5.7c. A non symetric flow topology is obtained
for the co-rotating VGs with smaller footprints of vortices at the wall. This
co-rotating VGs configuration is more adapted to control a full 3-D flow, for
instance on a swept wing.
The number of VG devices distributed along the spanwise of the chan-
nel appears to be an important parameter to control the separation flow
region. When this number is increased, the distance between each VG device
is decreased and the merging vortices are more efficient in reducinge the sep-
aration. This result is cross-checked with wall pressure measurements carried
out in the interaction region (see Figure 5.6): the plateau pressure seems to
disappear when the VG-S2 configuration (11 pairs of counter-rotating vortex
generators) is applied.
Forced shock oscillation configuration
Figure 5.8 shows the extreme positions of the shock wave in the channel for
the shock oscillation frequency of 30 Hz, considering both the reference case
without control and the control case with VG-S2 device. For the reference
case (see Figure 5.8a), the corresponding shock oscillation amplitude is equal
to 30 mm. The behavior of the boundary layer is different between these two
shock positions: in the upstream position, the boundary layer is separated
with important vortices structures in the mixing layer; in the downstream po-
sition, the boundary layer is attached. For the control case (see Figure 5.8b),
the size of the λ-shock structure is reduced compared to the reference case.
The trails of the vortices generated by the VGs are observed near the curved
wall and the mixing layer is growing for both shock locations under the in-
fluence of the VGs. Compared to the reference case, the leading shock of
the λ-shock system is not moving far in the upstream direction due to the
effect of the VGs (see left visualizations of Figure 5.8), which leads to a slight
reduction of the forced shock oscillation amplitude.
Figure 5.9 shows the longitudinal evolution of the pressure fluctuations
spectra obtained from signals measured by sensors located in the test sec-
tion median plane, at the 30 Hz shock oscillation, for the reference case
Fig. 5.8 Schlieren visualizations for 30 Hz-forced shock oscillation configuration

w/o control (a) and with VG-S2 (b) – Upstream position (left) and downstream
position (right) of the shock wave
Fig. 5.9 Longitudinal evolution of the pressure fluctuations spectra for 30 Hz-
forced shock oscillation configuration w/o control
without control. These spectra are represented in Sound Pressure Level (SPL)
expressed in dB, by:

Sp p
SPL(dB) = 20 × log10
pref
where Sp p is the spectrum modulus, in Pa2 , and pref is equal to 2 × 10−5 Pa.
The sample frequency is 6000 Hz for each sensor; the spectrum average is
obtained from 50 blocks and 8192 samples.
The P1 sensor spectrum (at X = 281.4 mm) exhibits no particular fre-
quencies because it is located in the supersonic zone of the flow and is not
sensible to downstream perturbations. The G9 sensor (at X = 316.4 mm)
gives a peak at the shock oscillation fundamental frequency f = 30 Hz, and
some harmonics. These peaks are weak because the sensor is in border of the
shock wave oscillation. The G6 sensor (at X = 336.4 mm) has several harmon-
ics and a very energetic fundamental peak at f = 30 Hz. This sensor picks up
very well the shock wave oscillation. The G3 sensor (at X = 356.4 mm, just
downstream of the bump) gives the fundamental peak (at f = 30 Hz), and
only one harmonic is observed. The fundamental peak is less energetic than
that of the G6 sensor (160 dB instead of 170 dB). This can be explained by
the location of the sensor. Indeed, according to the schlieren visualizations
(see Figure 5.8a), this sensor is always located downstream of the leading
shock, so all the harmonics – except one – have disappeared. The P2 sen-
sor (at X = 421.4 mm) and P3 sensor (at X = 575 mm, under the shaft)
have almost the same spectrum. These two sensors, which are located in the
subsonic zone, are mainly sensible to the downstream pressure perturbations.
Figure 5.10 gives comparisons between the reference case and control
cases with VG devices, for the 30 Hz forced shock oscillation configuration,
on the evolution of the pressure fluctuations by their RMS levels. Concerning
the G9 sensor, control by VGs allows to remain the RMS pressure level at
Fig. 5.10 RMS pressure levels for 30 Hz-forced shock oscillation configuration w/o
control and with VGs
5.3 Numerical Simulations 147
the level measured in the upstream supersonic region (P1 sensor), whereas
this level is strongly increased without control. So, the leading shock of the
λ-shock structure is not travelling so far upon this sensor and consequently,
the shock oscillation amplitude is reduced due to the VGs effect. The very
high RMS pressure level obtained with the G6 sensor is related to the pres-
ence of a recirculation bubble at this location; i.e., the rear part of the bump.
For the G3 sensor, discrepancies on the RMS pressure level are observed be-
tween the VGs devices, this location corresponding to the reattachment flow
region. The moderate RMS level remaining for the P2 sensor (compared to
the upstream residual P1 level) means that the reattached boundary layer is
not yet relaxed to a new equilibrium.
5.3 Numerical Simulations
University of Liverpool solver
The CFD code of the University of Liverpool solves the 3-D URANS equa-
tions with a variety of turbulence models including LES/DES approaches.
This is a Parallel Multi-Block (PMB) solver, which uses shared and dis-
tributed memory and multi-block structured grids. The numerical procedure
follows the implicit time marching approach and for space, a central-difference
approximation is used for viscous fluxes and LES. The convective fluxes are
evaluated by means of Roe’s and Osher’s schemes and MUSCL for formally
3rd order accuracy.
The unsteady case – forced shock oscillation – was computed using URANS
based on Menter’s SST model [5]. Due to the high Reynolds number (up-
stream unit Reynolds number is of the order of 14×106 ) and the need to
resolve the walls of the wind tunnel, no Large-Eddy Simulation (LES) was
attempted for this case. Instead, an approach based on Detached Eddy Sim-
ulation (DES) was adopted, using the well-known DES model by Spalart et
al. [6] in a Delayed DES formulation. Although this approach involved more
turbulence modelling than LES, it was the only pragmatic way of simulating
some of the turbulence of this flow.
Grid generation
For the analysis of the full domain, a model was generated in ICEMCFD
where the contraction of the tunnel, the bump, the rotating shaft and the
second throat of the tunnel were meshed (see Figure 5.11). Regardless of the
obvious advantages of this configuration regarding the clarity of the applied
boundary conditions, the extent of the domain was large and this would
certainly cause problems with the required mesh size and the efficiency of
the computations.
(a)
(b) (c)
(d) (e)
Fig. 5.11 Domain and multi-block topology for the reference case: (a) full domain,
(b) overview of the multi-block topology, (c) topology around the sliding interface,
(d) topology around the rotating cylinder and (e) mesh around the sliding interface
5.3 Numerical Simulations 149
To help with the selection of the best modelling approach, meshes with
a reduced domain were also created. For this case, the geometry was speci-
fied along with the boundary conditions presented in section 2.4. The inlet
flow condition was an input velocity profile measured by LDV and the out-
let was a downstream exit pressure condition which alleviated the need to
compute the whole domain up to the rotating shaft and the second throat of
the tunnel. The reduced domain is shown in Figure 5.12 where the shape of
the bump can also be seen. Cells with aspect ratio of almost 1 were used
for the XY (streamwise, normal) plane near the shock location. This mesh
minimizes numerical effects and helped with the resolution of the separated
flow region. The cell aspect ratio on the transverse plane varied from 1 to 35.
Exponential stretching allowed for 20 points inside the boundary layers of
all four walls indicating that finer grids may be needed. The overall mesh in-
cluded approximately 3.5 million cells and could be run on up to 64 processors
without substantial loss of parallel efficiency.
Figure 5.12c presents a comparison between experiments and CFD for
the velocity profile at a station X = 135 mm upstream of the interaction
region. One can see that the CFD results (obtained using the URANS method
and the SST model of Menter) suggest a slight underprediction of the peak
velocity. This behaviour is not unexpected since a reduced domain is used
and the predictions of the CFD near the wall of the tunnels influence the flow
and depend heavily on the employed turbulence model. For this reason, the
experimental profile has been imposed at the inlet of a reduced domain for
Fig. 5.12 Domain for the reference case: (a) co-ordinates of the bump, (b) re-
duced domain and (c) comparison between URANS results and experiments for
the velocity profile at X = 135 mm indicated in (a)
Fig. 5.13 Domain and multi-block topology for the control case: (a) full domain,
(b) surface view of the vortex generators on the bump and (c) multi-block topology
around the CoC1-VGs device
5.4 Comparison between Experimental and Numerical Results 151
computations, coupled with an imposed white noise on velocity fluctuations

at the level of 0.1%.
For the control test case, it was decided that the coarser configuration of
the vortex generators was to be used. This means that the conventional co-
rotating CoC1-VGs device was retained for computations with only 5 VGs in
the spanwise direction, see Figure 5.3. Figure 5.13 shows the vortex generators
arranged on the bump and the details of the employed multi-block topology
around the VGs. Three blocks were used for each side of the vortex generator
and the thickness of each vortex generator was also modelled, which added
12500 points near each VG. This approach avoids modelling problems with
representation of VGs as momentum sources in the equations.
5.4 Comparison between Experimental and Numerical

Results
Reference case
Figure 5.14 presents numerical shadowgraphs generated from the URANS
results for the extreme positions of the shock wave in the channel for the shock
oscillation frequency of 30 Hz. Compared to the experiments (see schlieren
Fig. 5.14 Numerical shadowgraphs for 30 Hz-forced shock oscillation configuration

w/o control (a) upstream position and (b) downstream position of the shock wave
visualizations on Figure 5.8a), it is interesting to notice that the lower λ-foot

of the shock changes structure and size as the shock moves. On the upper
wall, somehow smaller changes are obtained.
Figure 5.15 shows a comparison of the Mach number fields obtained by
LDV measurements and URANS computations for the extreme positions of
the shock wave in the channel for the shock oscillation frequency of 30 Hz.
The computed iso-Mach number distributions were generated using the same
levels as the LDV data and the same scaling was used for the dimensions
of the domain. The large λ-shock structures of the interaction region are
well-captured by the computations. With the shock in the upstream position
(see Figure 5.15a), the Mach number in front of the shock was around 1.3
and URANS shows a smaller re-circulation area compared to experiments.
Moreover, the LDV measurements allow for the separated flow to be quanti-
fied: the longitudinal extension of the separated bubble was equal to 62 mm
(from X = 328 mm to 390 mm). The maximum negative velocity is equal to
−77 m/s. When the shock is in the downstream position (see Figure 5.15b),
the Mach number is equal to 1.45 and there is no measured re-circulation
region, whereas an incipient flow separation is predicted by computations.
Figure 5.16 presents the comparisons between experimental and computa-
tional histories of the shock position and the size of the separated flow region
for two full forced oscillations of the shock. Three oscillations were computed
and results from the last two appeared to be similar. This suggested that the
URANS solution reached a periodic state, and was then used for comparisons
with the experiments. The shock positions, as a function of time, were based
on comparisons of the wall surface pressure. As can be seen in Figure 5.16a,
the URANS results are in rather good agreement with the experiments al-
though the change of the shock position is not as rapid as the experiments and
some discrepancy exists for the maxima and minima which correspond to the
upstream and downstream locations. However, the size of the separated flow
region is under-predicted by almost 45% (see Figure 5.16b). Interestingly, the
results show that URANS predicts the attached phase of the oscillation quite
well and problems are concentrated for the case where the flow has large sep-
aration extending downstream of the shock. Another observation is that the
experiment suggests a more rapid change of the separated flow region with
a sharp rise of the length of the re-circulation while the URANS results show
a slower development of the separation.
The obtained URANS results suggested that a refined level of modelling
was necessary to resolve the flow physics of such complex configuration, es-
pecially the dynamics of the separated region. Subsequently, results were
obtained using a Delayed Detached Eddy Simulation formulation with the
model of Spalart et al. [6] (SA-DES). No flow forcing was used at the inflow
of the domain and 5 cycles of the shock oscillation were computed before
the final two were kept for comparisons. Due to the volume of the data, only
a small number of flow modes from Proper Orthogonal Decomposition (POD)
were kept. The employed time step was adequate to resolve frequencies up
Fig. 5.15 Mach number distributions obtained by LDV measurements (left) and
URANS computations (right) for 30 Hz-forced shock oscillation configuration w/o
control (a) upstream position and (b) downstream position of the shock wave
Fig. 5.16 Comparison between experimental and URANS computational histories

of shock position (a) and size of the separated flow region (b) for 30 Hz-forced shock
oscillation configuration w/o control
to 100 Hz which covers most of the range of frequencies obtained during the
experiments.
Figure 5.17 presents the comparisons between experimental and compu-
tational histories of the size of the separated flow region for two full forced
oscillations of the shock. DES results are predicting a larger separated flow
region then the URANS ones: the size of the re-circulation bubble is under-
predicted by almost 17% when the finer mesh (3.5 million points) is used.
As was the case for the URANS though, the change of the separation size
is not as rapid for the DES as for the experiments. The observed differences
between DES and URANS are mainly due to the near-wall formulation of
the models.
Fig. 5.17 Comparison between experimental and DES computational histories of

size of the separated flow region for 30 Hz-forced shock oscillation configuration w/o
control
Figure 5.18 attempts to show in 3D the two extreme configurations with

the shock upstream and downstream positions, by plotting the iso-surfaces of
negative streamwise velocity at the level of 1% of the pre-shock streamwise
component. A substantial separated region exists when the shock is in up-
stream position. For the downstream shock position, an incipient separation
is observed on the lower wall and a small amount of re-circulation is visible
on the top wall as well. Although these results correspond to instantaneous
data, one can still appreciate the complexity of this interaction.
The averaged DES wall pressure data appear to agree well with the exper-
iment, as can be seen on Figure 5.19. The average pressures on both upper
and lower walls suggest that the interaction is well-resolved with some un-
derestimation of the separated flow region.
Figure 5.20 shows a comparison of experimental and DES computational
unsteady wall pressure spectra obtained by 4 transducers located on the
Fig. 5.18 Evolution of the size of the separation region obtained by DES com-
putations for 30 Hz-forced shock oscillation configuration w/o control (a) upstream
position and (b) downstream position of the shock wave
Fig. 5.19 Comparison between experimental and DES computational averaged

pressure distributions on the upper wall (a) and the lower wall (b) for 30 Hz-forced
shock oscillation configuration w/o control
(a) P1 (Upstream Flow) (b) G9 (Leading Shock)
(c) P2 (Boundary Layer Reattachment) (d) P3 (Under the Shaft)
Fig. 5.20 Comparison between experimental and DES computational unsteady

wall pressure spectra for 30 Hz-forced shock oscillation configuration w/o control
median axis: P1 upstream of the interaction, G9 around the leading shock

location of the λ-foot, P2 near the reattachment of the boundary layer and P3
under the shaft. The fundamental peak is well-predicted in terms of frequency
(a slight underestimation of 2 Hz) but its magnitude is under-predicted by
approximately 7 dB. The first harmonic is predicted by computations, which
is not the case for the second harmonic at 90 Hz. The DES results could not
resolve frequencies above 100 Hz (due to the employed time step) and the
low-frequency parts of the spectra (below 10 Hz) show lower amplitude in
comparison to the test data ones.
Control case by CoC1-VG device
Results were also obtained using the SA-DES for the 30 Hz-forced shock oscil-
lation configuration with the CoC1-VG device applied. Figure 5.21 presents
visualisations of the vortices behind the VGs and surface streaks deduced
from averaged DES results. One notices that the surface flow topology due
to the presence of the 5 co-rotating VGs is similar between DES and experi-
ments (see oil flow visualization on Figure 5.7c).
Figure 5.22 shows the three components of the velocity and the Mach
number fields mid-way between CoC1-VGs and the end of the bump ob-
tained using the time-averaged DES results. At this streamwise location, the
5 vortices created by the co-rotating VGs have a strong intensity and amplify
the effet of the corner flow on the interaction.
Fig. 5.21 Vortices behind VGs and surface streaks obtained using the time-
averaged DES results for 30 Hz-forced shock oscillation configuration with the
CoC1-VG device
Fig. 5.22 Velocity and Mach number fields mid-way between VGs and the end
of the bump obtained using the time-averaged DES results for 30 Hz-forced shock
oscillation configuration with the CoC1-VG device
Figure 5.23 presents a flow snap-shot from the DES solution by plotting
streamwise velocity contours on vorticity iso-surfaces. One notices low fre-
quency content regions upstream of the interaction, whereas an increase of
small structures is observed past the shock. The shock appears to follow the
back pressure in phase and is stabilised under the effect of VGs, although the
corner vortices appear to have a dominant role. DES is mainly active near
the centre of the domain behind the interaction region.
Fig. 5.23 Flow snap-shot from the DES solution for 30 Hz-forced shock oscillation
configuration with the CoC1-VG device
Figure 5.24 presents the Mach number distributions for three spanwise
stations obtained by DES computations. The results suggest that the shock
is much more stable with the VGs but the flow is mainly influenced by the
corner vortices. Moreover, a very small separated flow region is evident near
the mid-span of the channel.
Fig. 5.24 Mach number distributions for three spanwise stations obtained using
DES solution for 30 Hz-forced shock oscillation configuration with the CoC1-VG
device
Interestingly, the forces on the vortex generators are not equal. Figure 5.25
presents the history of the streamwise forces for VG 1, located in the vicinity
of the sidewall of the channel, and for VG 2, located far away from the
sidewall. Clearly, the loading is not the same which also suggests that the
main flow variation is now in the spanwise direction.
Fig. 5.25 Comparison of streamwise forces for the CoC1-VGs 1 and 2
The averaged DES lower wall pressure data appear to agree well with the
experiment, as can be seen on Figure 5.26. Separation is still under-predicted
in comparison to the experiments. Weak perturbations induced by the VGs
are visible upstream of the shock.
Fig. 5.26 Comparison between experimental and DES computational averaged

lower pressure distributions for 30 Hz-forced shock oscillation configuration with
the CoC1-VG device
Figure 5.27 shows a comparison of experimental and DES computational

unsteady wall pressure spectra for the G3 transducer located on the median
axis, in the rear part of the bump where the flow is separated. The frequency
of the fundamental peak is well-predicted by DES but its magnitude is under-
predicted by approximately 12 dB. Regarding the first harmonic, experiment
shows that its magnitude is increased under the VGs effect, which reduced the
Fig. 5.27 Comparison between experimental and DES computational unsteady

wall pressure G3-spectra for 30 Hz-forced shock oscillation configuration w/o control
and with the CoC1-VG device
extension of the separated region, in particular around the G3 location. On

the contrary, DES reduced strongly the intensity of the first harmonic, which
is not coherent with the under-prediction of separation (see Figure 5.26).
Concerning the low-frequency parts of the spectra (below 10 Hz), DES results
show a lower level (of 10 dB) in comparison to the test data ones.
5.5 Conclusion
The aim of the study was to control by mechanical vortex generator devices
the strong interaction between a shock wave and a separated turbulent bound-
ary layer in a transonic channel. Control devices – co-rotating and counter-
rotating vortex generators – were implemented upstream of the shock foot
region and tested both on a steady shock wave and on a forced shock oscil-
lation configurations. Wall pressure measurements and optical non intrusive
means of investigation allowed to quantify the effects of vortex generators on
the flow separation region and the shock oscillation amplitude.
The number of vortex generator devices distributed along the spanwise
of the channel appeared to be an important parameter to control the flow
separation region. When this number is increased, the distance between each
device is decreased and the vortices merging is more efficient to reduce the
separation. Moreover, their placement upstream of the shock wave is determi-
nant to ensure that vortices have mixed momentum all spanwise long before
they reach the separation line, so as to avoid separation cells. The expansion
and compression waves generated by the vortex generators have a slight in-
tensity. But in term of drag penalties, the way to reduce the effects of these
perturbations is to impose the height of the vortex generators at a value less
than the sonic line of the incoming boundary layer.
The experiments presented a very challenging flow due to the well
known origins of both the flow unsteadiness and the boundary conditions.
The work undertaken using URANS and the SST model resulted in some
References 161
under-predictions of the separated flow region though all computations sug-

gest that the essential physics was captured. The DES approach had to be
used for these experiments due to the high Reynolds number and the need
to resolve all channel walls. Some improvement was offered by DES on the
re-circulation prediction, though the results were still not in perfect agree-
ment with experiments. Regarding the wall pressure evolutions, the results
were in fair agreement with the tests. The advantage of the DES was that
it provided some unsteady spectra which captured at least two of the tones
suggested by the experiments. The overall level of peaks was, however, lower
than the experiments suggesting that further work is needed to capture all
flow features. The flow control results were obtained with DES only, and
the geometric details of the co-rotating vortex generators had to be resolved.
Each vortex generator produced a well-defined vortex which travelled under
the sonic surface to the region of the interaction. The vortices allowed the
separation to be incipient and the shock to stabilise near the end of the bump,
the flow being somehow pacified. Nevertheless, the efficiency of vortex gen-
erators to reduce the amplitude of the forced shock wave oscillation was not
confirmed by experiments, only a slight delay on the upstream displacement
of the leading shock being observed.
References
[1] Pearcey, H.H.: Shock induced separation and its prevention by design and
boundary layer control. In: Lachmann, G.V. (ed.) Boundary Layer and Flow
Control, vol. 2, pp. 1312–1314. Pergamon Press, Oxford (1961)
[2] Lin, J.C.: Review of research on low-profile vortex generators to control bound-
ary layer separation. Progress in Aero. Sci. 38, 389–420 (2002)
[3] Galli, A., Corbel, B., Bur, R.: Control of forced shock-wave oscillations and
separated boundary layer interaction. Aero. Sci. and Tech. 9, 653–660 (2005)
[4] Bur, R.: Deliverable 3.2.11 of the UFAST Project – Final Report of the ONERA-
DAFE activities (May 2009)
[5] Menter, F.R.: Two-equation eddy-viscosity turbulence models for engineering
applications. AIAA J. 32(8), 1598–1605 (1994)
[6] Spalart, P., Jou, W.H., Strelets, M., Allmaras, S.: Comments on the feasibil-
ity of LES for wings, and on a hybrid RANS/LES approach. In: Advances in
DNS/LES, 1st AFOSR International Conference on DNS/LES, Columbus, OH
(August 1997)
Chapter 6
Nozzle Forced Shock Oscillations
(Holger Babinsky)
6.1 Introduction
In the present study, the response to periodic forcing of a normal shock

wave in a parallel walled channel has been investigated. Attention has been
focused on the interaction between the normal shock wave and the turbulent
boundary layer developing on the tunnel floor. The main aim of the study
is to characterise the unsteady behaviour of the interaction and to identify
the key factors that influence its nature. Tests with freestream Mach numbers
ahead of the shock wave of 1.3, 1.4 and 1.5 have been conducted to investigate
interactions with varying degrees of shock induced boundary layer separation.
Experiments without forcing have also been conducted to provide a basis for
comparison. The effects of flow control devices – in the form of mechanical
micro-vane vortex generators – on the steady and oscillating interaction have
also been studied.
CFD was performed in the form of unsteady RANS calculations for the
steady and unsteady flow without control at Mach numbers 1.3 and 1.4.
A number of turbulence models were applied. In the case of M = 1.3, one
further simulation was performed using DES. During the course of the inves-
tigation it was found that a converged solution could not be achieved for the
strong shock case (M = 1.5). It is suggested that this is due to corner effects.
6.2 Experimental Conditions

Figure 6.1 shows a sketch of the general working section arrangement for
unsteady tests (left). Detailed descriptions of the experimental set-up, flow
conditions and the measurement technology can be found elsewhere. Tests
were also conducted with a modified wind-tunnel set-up to achieve a perfectly
steady shock wave. This is also shown in Figure 6.1.
springerlink.com
164 6 Nozzle Forced Shock Oscillations (Holger Babinsky)
Fig. 6.1 Sketch of the wind tunnel working section (left: unsteady tests, right:
steady tests)
6.3 Numerical Method

The Euranus flow solver from Numeca International has been used in order
to perform these simulations. It is a structured multi-block Navier-Stokes
code using finite volume approach. Central-space discretization is employed
together with Jameson type artificial dissipation. A four-stage Runge-Kutta
scheme is selected for the temporal discretization. Multi-grid, local time step-
ping and implicit residual smoothing are also used in order to speed-up the
convergence. Three grid levels are used for all the simulations performed,
a typical grid is shown in Figure 6.2. Time-accurate computations are done
with a dual-time stepping method. The temporal derivative is discretised
using a second order backward Euler difference. Further details of the nu-
merical method and grid can be found in the individual reports produced by
the partners.
Fig. 6.2 Typical computational grid (3,300,000 cells)
6.4 Flow Case Results: Steady Tests

Figures 6.3–6.8 show for each Mach number: a schlieren image of the steady
interaction, a photograph of the tunnel floor surface flow topography, the
centreline pressure profile; and plots of the mean and fluctuating streamwise
velocity contours.
6.4 Flow Case Results: Steady Tests 165
Fig. 6.3 Steady M = 1.3 interaction
Fig. 6.4 Mean and fluctuating velocity contours, steady M = 1.3 interaction
Fig. 6.6 Mean and fluctuating velocity contours for steady M = 1.4 interaction
6.4 Flow Case Results: Steady Tests 167
Fig. 6.8 Mean and fluctuating velocity contours for steady M = 1.5 interaction
6.5 Flow Case Results Part II: Unsteady Tests

Mechanism for varying the downstream pressure at M = 1.4 and
1.5
Rotation of the elliptical shaft downstream of the interaction causes an al-

most sinusoidal variation of tunnel back pressure as seen in Figure 6.9 (top).
Each line plotted in the figure is calculated by averaging conditionally sam-
pled results from a number of individual periods of unsteady forcing. The
amplitude of pressure perturbation is seen to be very weakly dependent on
forcing frequency; decreasing slightly at higher frequencies. The shape of the
pressure profile is close to sinusoidal in all cases, though some skewing occurs
at high frequencies (most clearly seen for M = 1.4). Experimental limita-
tions prevented unsteady experiments at M = 1.3. Also seen in Figure 6.9
are the measured shock velocities across a cycle. It can be seen that there is
a similar magnitude of peak velocities for all frequencies, suggesting that the
shock motion is primarily determined by the magnitude of the back-pressure
perturbation.
Fig. 6.9 Variation of downstream pressure (shock velocity is also plotted for com-
parison) at M = 1.4 and 1.5
Schlieren images of the unsteady interaction at M = 1.4 at three dif-

ferent forcing frequencies are shown in Figure 6.10. Mach number contours
from LDA velocity measurements are shown in Figure 6.11, superimposed on
schlieren images. The measured static pressure profile through the interac-
tion at four points in the shock’s cycle of motion is plotted in Figures 6.12 &
6.13. Equivalent results for M = 1.5 are shown in Figures 6.14–6.16.
6.5 Flow Case Results Part II: Unsteady Tests 169
Fig. 6.10 Schlieren images at different frequencies, M = 1.4
Fig. 6.11 LDA measurements of the unsteady M = 1.4 interaction, f = 43 Hz

Fig. 6.12 Comparison of the pressure rise during upstream and downstream shock
motion. M = 1.4, f = 43 Hz
Fig. 6.13 Comparison of the pressure rise at extreme shock positions. M = 1.4,
f = 43 Hz
6.5 Flow Case Results Part II: Unsteady Tests 171
Fig. 6.14 Schlieren images at different frequencies, M = 1.5
Fig. 6.15 Mach number contours for the unsteady M = 1.5 interaction, f = 43 Hz
Fig. 6.16 Comparison of the pressure rise during upstream and downstream shock
motion. M = 1.5, f = 43 Hz
6.6 Comparison with CFD 173
6.6 Comparison with CFD

For the steady shock cases at M = 1.3 there was relatively good agreement
between CFD and experiment. Figures 6.17–6.20 show comparisons of flow
structure, surface pressures and boundary layer profiles before and after the
interaction. It particularly noteworthy that the inflow boundary layer profile
(Figure 6.19) is very well predicted by CFD, despite the fact that the exact
conditions at nozzle entry are not known. This justifies the additional expense
of computing the complete flow through the nozzle (see Figure 6.18) in order
to correctly establish the boundary-layer development.
Fig. 6.17 Comparison of experimental and numerical schlieren images. M = 1.3,

steady.
However, significant differences were also observed, notably the boundary

layer growth through the interaction was overpredicted. Among the turbu-
lence models tested, the SST model appeared to give the best agreement of
the boundary layer development, particularly the inflow was well captured.
For the stronger Mach number (1.4) the differences observed between
experiment and computation downstream of the shock became more pro-
nounced (see Figures 6.21–6.24), although the inflow continues to be well
captured. In general, the effect of the interaction on the centre-line bound-
ary layer was more severe in the CFD results. More significantly, several of
Fig. 6.18 Comparison of experimental and numerical surface pressure. M = 1.3,

steady.
Fig. 6.19 Comparison of inflow velocity profiles (x = −150). M = 1.3, steady.

Fig. 6.20 Comparison of post-shock velocity profiles (x = 90). M = 1.3, steady.
Fig. 6.21 Comparison of experimental and numerical schlieren images. M = 1.4,

steady
the numerical simulations predicted asymmetric flowfields (depending on the

turbulence model used), which were not observed in the experiments (see
Figure 6.25).
On closer investigation it was concluded that one likely cause of this asym-
metry was due to exaggerated corner effects. All asymmetric solutions fea-
tured one corner which was much more severely separated than the others,
and often the diagonally opposite corner displayed the smallest amount of
separation. It is thought that once the corner separation grows above a cer-
tain limit, it begins to affect the flow at all other corners, leading to asym-
metric solutions. There was also evidence that the corner effects observed
in the numerical solutions were more severe than seen in the experiments
Fig. 6.22 Comparison of experimental and numerical surface pressure. M = 1.4,

steady.
Fig. 6.23 Comparison of inflow velocity profiles (x = −150). M = 1.4, steady.
Fig. 6.24 Comparison of post-shock velocity profiles (x = 90). M = 1.4, steady.

Fig. 6.25 Demonstration of asymmetric results: CFD surface streamlines for var-
ious turbulence models. M = 1.4, steady.
Fig. 6.26 Comparison of surface streamlines along tunnel floor from experiment
and CFD. M = 1.4, steady.
as demonstrated by Figure 6.26, which might explain why such asymmet-

ric solutions were not observed experimentally. However, there is evidence
from experiments elsewhere, that asymmetric solutions do occur in similar
circumstances1 . However, while asymmetry reported elsewhere is generally
quasi-two-dimensional, the asymmetries seen here are generated at a corner.
The exact reason for these discrepancies remains an open question.
The hypothesis that asymmetry is triggered by excessively large (compared
to the channel size) corner separations was to some extent supported by
further simulations run with enforced symmetry: here the size of the corner
separation was considerably reduced compared to the fully 3-d cases.
Although many open questions remain, it is clear that corner effects are
significantly less well captured by the CFD methods employed here than the
centreline flow.
A review of the unsteady CFD results for M = 1.4 shows that the global
effects are well captured. For example, Figure 6.27 compares the shock os-
cillation amplitudes for two frequencies with the experimental data and
Figure 6.28 shows the shock velocities observed during an unsteady cycle.
Here it can be seen that the numerical results follow the experimental data
very well indeed. Figure 6.29 compares the streamwise pressure distribution
and the general flowfield for M = 1.4 and, once again, the main trends agree
well. However, the changes in shock structure through an unsteady cycle
are less well captured. For example the changes to the size and shape of the
Fig. 6.27 Shock oscillation amplitudes for various frequencies (CFD and Expt).
M = 1.4.
1
Papamoschou D. and Zill A., 2004, “Fundamental investigation of supersonic noz-
zle flow separation”. AIAA 2004–1111. Bourgoing A. and Reijasse Ph., 2001, “Ex-
perimental investigations of an unsteady and asymmetrical supersonic seprated
flow”, CASI – 8th Aerodynamics Section Symposium, Toronto, Canada, 2001.
lambda shock foot appear to be different from the experimental observations.

Figure 6.30 compares the triple point height through a shock oscillation cycle
and it can be seen that, the behaviour is quite different. The magnitude of
changes to the height of the lambda foot is much greater for the numerical
results than observed in the experiments. Also, the phase relationship (not
seen here) is quite different with the maximum lambda foot height occurring
at different times in the cycle. This suggests that the inviscid effects of the
enforced unsteadiness are well captured by CFD, whereas the more subtle
viscous effects are not.
Fig. 6.28 Shock velocities through one cycle (CFD and Expt). M = 1.4.
Fig. 6.29 Comparison of various unsteady CFD and expt. results. M = 1.4.
Fig. 6.30 Comparison of triple point height through an unsteady cycle. M = 1.4,
f = 40 Hz.
6.7 Other Observations
Prediction of shock oscillation amplitude
The amplitude of unsteady shock wave motion is observed to decrease with

increasing excitation frequency. Pressure measurements in the region of un-
steady shock wave motion suggest that the pressure rise across an unsteady
SBLI depends primarily on the relative flow Mach number in the shock wave
reference frame. This implies that the velocity of shock wave motion can be
determined analytically for a given imposed (time-varying) pressure ratio.
Furthermore, with knowledge of the pressure variation driving shock wave
motion, the shock wave trajectory can easily be calculated by integrating
the predicted shock wave velocities, thus yielding the amplitude of shock
wave motion. Figure 6.28 also includes the results predicted by this analyti-
cal method for M = 1.4. The agreement between analytical and experimental
results is good, supporting the concept that shock wave motion is simply the
result of the shock wave adjusting its strength to satisfy an imposed vary-
ing pressure ratio. As an extension to this project, the theoretical approach
was expanded to include the effects of non-constant duct cross-sections and
this was also found to be capable of predicting shock oscillation amplitude
observed experimentally.
Measurements of the evolving pressure signal a short distance downstream
of the shock wave support the above theory at low frequencies, where the
temporal variation of velocity is seen to closely match the temporal varia-
tion of downstream pressure. However, at higher frequencies, peak upstream
velocities are higher than expected for part of the cycle. The causes of this
discrepancy are thought to be a combination of viscous effects in the bound-
ary layer and a phase lag between the upstream transmission of pressure
information in the freestream and boundary layer.
6.8 Conclusions and Further Work 181
The velocity discrepancy is more significant for the M = 1.4 case than
for the M = 1.5 case. This may be because the flow at M = 1.4 is only
intermittently separated and so small changes in relative Mach number may
greatly affect shock wave motion through variations in the extent of boundary
layer separation. At M = 1.5, the flow is strongly separated and so changes
in relative Mach number would have less of an effect.
6.8 Conclusions and Further Work
In simple bullet form, the following are the main conclusions reached from
this flow case:
• Shock oscillations caused by downstream pressure fluctuations are domi-
nated by simple inviscid (one-dimensional) effects
• It is possible to predict shock oscillation amplitudes in ducts (of constant
and varying cross-section) with simple analytical tools
• Unsteadiness does lead to changes in the SBLI structure but the role of
viscous effects is not well understood
• URANS methods are capable of predicting the global dynamics of un-
steady pressure driven flows well
• However, subtle viscous effects are not well captured by CFD, especially
unsteady effects
• Corner effects can dominate the complete flow-field in a confined channel
• Flow control (micro-vortex generators) are capable of reducing shock-
induced separation away from corners
• No flow control tested has been able to reduce or eliminate corner separa-
tion
• Flow control has negligible effect on shock oscillations (presumably because
they are dominated by inviscid effects)
• Corner flows may be the biggest challenge (for CFD and experiment)
As a result of this study, a number of interesting topics for future study
emerge:
• Understanding separated supersonic corner flows and their control
• Improving CFD methods for corner flows
• Understanding the interplay between inviscid and viscous effects in shock
unsteadiness.
Chapter 7
Natural Shock Unsteadiness in Nozzle
and Curved Channel (Piotr Doerffer)
7.1 Introduction
This flow case concerned experiments carried out at the Institute of Fluid
Flow Machinery of the Polish Academy of Sciences in Gdansk. Here un-
steadiness in shock wave – boundary layer interaction could only be induced
by natural factors, such as: upstream boundary layer excitation, separation
unsteadiness, vortex shading from the separation zone or reattachment un-
steadiness.
Basic flow cases were followed by flow control in which mainly air jet
vortex generators were used. In addition, suction through the test wall was
implemented, but only in the curved channel. This was done for comparison
with earlier EUROSHOCK European project results.
The work share concerning numerical simulations is presented in the table
below:
Table 7.1
Test case RANS/URANS LES DES
Straight nozzle
Basic exp, straight nozzle, M = 1.23
Basic exp, straight nozzle, M = 1.33 IMP
Basic exp, straight nozzle, M = 1.45 IMP, LIV LIV IMP/
NUMECA
AJVG, straight nozzle, M = 1.33
AJVG, straight nozzle, M = 1.45 IMP, LIV periodic LIV periodic
Curved nozzle
Basic exp., curved nozzle, M = 1.33
Basic exp., curved nozzle, M = 1.43 IMP, LMFA LIV
AJVG, curved nozzle, M = 1.33
AJVG, curved nozzle, M = 1.43 IMP periodic LIV periodic
Suction, curved nozzle, M = 1.33
Suction, curved nozzle, M = 1.43 LMFA
springerlink.com
184 7 Natural Shock Unsteadiness in Nozzle (Piotr Doerffer)
Comparison of CFD results with experiments concerned:

• Middle plane cut – iso-contours of Mach number instantaneous picture
• Static pressure distribution along the wall in the test section centre Ps /P0
• Streamlines at the lower wall for comparison with oil flow pictures
• velocity profiles in the marked traverses up to 20 mm from the wall
• CTA measurements of velocity fluctuations in chosen boundary layer or
separated area locations
• fluctuations of the shock location
The experimental work at the IMP PAN focused on natural shock wave
oscillations in nozzles of two different design concepts. These concepts had
an important effect on the unsteady behaviour of interaction.
7.2 Basic Flow Cases

The first type of test section concerned half a straight de’Laval nozzle,
designed for a uniform outlet stream (Figure 7.1). A typical triangular area of
uniform velocity was formed at the outlet extending a long distance along the
rectilinear wall. The extent of the constant velocity made the shock location
easy to destabilise. Nevertheless, the shock wave spanned the whole height of
the nozzle and it was stabilised along the upper wall by the velocity gradient.
Thus a configuration was produced where a shock could react to the natural
flow excitations.
However, this type of test section is inconvenient in that it can only be
applied for a single Mach number value at the outlet. Therefore each flow case
required a new test section, i.e. a new upper wall contour, see Figure 7.1.
A typical shock structure is presented in Figure 7.2 (M = 1.45 upstream of
the shock wave). Here the oblique disturbance originating from the rectilinear
wall upstream of the shock wave (left bottom corner of Figure 7.2) merely
resulted from a joint between two wall sections.
Figure 7.3 presents the typical behaviour of static pressure along the rec-
tilinear wall, characteristic for the centreline of a typical Laval nozzle. Static
pressure value was reduced along the wall due to the acceleration, reaching
M = 1.45 and maintaining this value until the shock wave. The length of the
Fig. 7.1 Straight nozzle

7.2 Basic Flow Cases 185
Fig. 7.2 Schlieren visualization of the shock at nozzle for M = 1.45
Fig. 7.3 Static pressure distribution along the test wall
constant velocity area allowed us to have the same Mach number value at
a flow control device and the shock wave boundary layer interaction location.
It should be emphasized here that RANS and URANS as well as DES
turbulence modelling all under-predict separation. This is primarily apparent
in the static pressure distribution along the wall centreline downstream of the
shock wave, as shown in Figure 7.4.
Fig. 7.4 Wall static pressure distribution comparison with RANS, URANS, DES
Initially, upstream of the shock wave, the static pressure at the wall is
very well predicted. As is the pressure rise location and gradient of the shock.
Further downstream, however, the discrepancy between measurements and
numerical simulations becomes very significant. This is due to differences in
separation size prediction. Over-prediction of pressure jumps is usually the
effect of a very short separation region.
An oil flow visualisation indicated a very large separation zone at M =
1.45, see Figure 7.5. Here a relatively wide zone of oil accumulation appears
just upstream of the separation. This may be related to shock wave unsteadi-
ness. The separation zone is rather long, almost as much as the wind tunnel
span.
The URANS numerical results are shown in Figure 7.6. These confirm
that the separation length is significantly shorter in the CFD than in the
experiment. In the numerical simulations the separation area is concentrated
in the middle of the test wall, tapering off towards the side walls, adjacent
to a corner vortex.
Nevertheless, it could be pointed out that shear stress distribution close to
the wall could lead to new observations. In Figure 7.7 one can see that very
low shear (blue and green) covers a much larger area than the reverse flow.
This low shear zone size is very similar to the separated area in Figure 7.5.
Fig. 7.5 M = 1.45 oil flow visualisation of the separation zone
Fig. 7.6 URANS of M = 1.45 case by IMP PAN
It is worth mentioning that the new PSP (Pressure Sensitive Paint) mea-
surement technique allowed us to resolve an interesting issue that until now
had been difficult to explain. PSP provides pressure measurement on the
entire wall surface. Therefore it can provide information on pressure varia-
tion also across the channel – in a spanwise direction. The obtained pressure
contours are presented in Figure 7.8. For the first time it became evident
that the pressure iso-contours under the shock wave are bow-shaped, bulging
upstream in the wall centre.
This very interesting and new observation is related to the curvature of
the oil front at the upstream edge of the separation area. A thick gray line in
oil visualisation shows the shape corresponding to the iso-pressure contour
in the PSP picture. The upstream edge of oil accumulation is straighter and
there is virtually no effect close to the side walls.
Fig. 7.7 Contours of shear stress
Fig. 7.8 Contours of static pressure at the test wall measured by PSP and oil
visualisation
PSP allows for a comparison to be made between the experiment and CFD
concerning pressure distribution on the test wall. There is very good coinci-
dence in the pressure contour at the end of compression. The dashed line in
both plots is of identical shape. It fits both PSP and CFD results exception-
ally well. There is, however, discrepancy in the upstream compression close
to the side wall. As indicated by the dotted ellipse, the CFD shows a con-
siderable upstream effect in the corner, while the PSP does not confirm this
behaviour. However, at the other side wall such a tendency is also present in
Fig. 7.9 PSP result and RANS numerical simulation
Fig. 7.10 Comparison of stagnation pressure profiles between experiment and

RANS and URANS in three traverses upstream and downstream of the shock wave,
M = 1.45
the PSP. It should also be mentioned that the CFD in the upstream part of
the compression does not show any bow-type pressure contours in the wall
centre.
Weaker separation in the CFD than in the experiment (Figure 7.6 and 7.5
respectively) may also be seen in the velocity traverses at the locations
marked in Figure 7.5.
The CFD results from IMP PAN concerning RANS and URANS coin-
cide very well, indicting that for such analyses RANS simulation results are
sufficient. Upstream the interaction (X = 212) boundary layer profile pre-

dicted by the CFD coincides very well with the experiment. Only the stagna-
tion pressure value at the boundary layer outer edge is slightly over-predicted.
The other two traverses, X = 267 and X = 292, are located in the separation
area (see Figure 7.6). In the experiment, the stagnation pressure within the
separated zone maintains virtually constant value. At X = 267, just after the
separation, the height of the separation is smaller than at X = 292, which
is closer to the re-attachment. The CFD shows significant over-prediction of
the stagnation pressure in the main part of the boundary layer. This shows
that CFD predicts much less loss, hence the effect of the shock wave in the
CFD is much weaker. Above the boundary layer the situation is different.
Lower stagnation pressure shows that the losses above the boundary layer
are higher, so the shock is stronger.
Unsteady effects
Interaction unsteadiness was measured in two ways. Firstly, as the shock

wave unsteadiness, and secondly, as the fluctuations in chosen locations in
the separated boundary layer measured by the CTA.
A typical result of the shock wave oscillation at M = 1.45 is presented in
Figure 7.11 (left). The recorded signal contains high frequency oscillations as
well as low frequency. Results obtained from URANS (as shown in Figure 7.11
– right) have an amplitude and period comparable to the high frequency
oscillations of the shock movement in the experiment. The low frequency
component in the CFD was not observed probably due to the short time
frame of the simulations (0.4 sec). Under-predicted separation length may
also be responsible for the lack of low frequency oscillations.
Unsteady behaviour of the shock wave depends on the Mach number of the
interaction. The RMS of the shock position is presented in the table below,
with the average shock oscillation given in millimetres.
The results identified interesting features of unsteadiness. One is that Mach
number reduction enhances the shock unsteadiness. The largest shock oscil-
lations have been observed for the lowest Mach number, at which the flow
Fig. 7.11 Shock wave oscillation M = 1.45, experiment (left), URANS (right)
Table 7.2
Mach number 1.23 1.33 1.45

RMS 1.44 0.986 1.066
is only weakly destabilised and fully attached. In the case of M = 1.33, at

which an incipient separation appears, the shock wave oscillations are the
smallest. At the highest Mach number, with developed separation, the shock
oscillations become stronger. All these factors indicate that the intensity of
natural shock unsteadiness reaches a minimum which is associated with in-
cipient separation.
In the cases of separation a λ-foot is formed, which is also subjected to
unsteadiness. It has been observed that λ-foot unsteadiness appears as a pul-
sation of the size of the entire λ-foot structure.
Issue of asymmetry in CFD of straight nozzle
The results presented in Figure 7.6 and 7.7 were obtained using SPARC [2]
code with a two-equation k-tau (Speziale-Abid-Anderson) turbulence model.
These results are very important because the shock and separation are sym-
metric in the nozzle. In Figure 7.12 the constant pressure surface is shown,
which confirms the flow field symmetry.
Fig. 7.12 Iso-pressure surface downstream of the shock

Before the above symmetric structure was found, the separation obtained
in numerical simulations looked as the one at the bottom wall in Figure 7.13.
Fig. 7.13 Typical asymmetry of separation at the rectilinear wall
Fig. 7.14 Iso-pressure surface in case of asymmetric flow
The asymmetry is caused by upstream penetration of separation through

one of the streamwise corners. It is well visualised by the shape of the iso-
pressure surface in Figure 7.14.
In order to find a reason for the appearance of this asymmetry, the mesh
was checked for symmetry and orientation of blocks. All available codes were
used: SPARC, FINE/Turbo of NUMECA and FLUENT with the same mesh
and the same boundary conditions. Nevertheless, in all cases the solution
was asymmetric. In the case of FINE/Turbo, the central difference scheme
and second order upwind were applied. In both cases asymmetry appeared.
Asymmetric flow structure was also obtained from Fluent 6.3 with a MUSCL
numerical scheme.
It is important to note that on a coarse mesh the flow structure was sym-
metric. However, the existence, size and structure of separation varied with
each numerical scheme.
After using three codes, a number of numerical schemes and many tur-
bulence models, the next step was to find out whether further mesh refine-
ment could influence the flow structure. The basic mesh was doubled in each
direction and simulations were performed using SPARC. The refined mesh
consisted of ≈ 22 × 106 cells. The convergence history was very good and
on the finest mesh level, residues were decreased by 7 orders of magnitude.
Nevertheless, the flow structure remained asymmetric and the solution was
steady.
To complete the analysis, a new turbulence model in Fluent k-ω SST was
applied without success. Using DES (based on Spalart-Allmaras) did not
help either. This led us to suspect that eddy viscosity models, having in
general difficulty in predicting streamwise corner secondary vortices, were
the reason for these disappointing results. There was a great deal of hope in
using Reynolds Stress Model, but, unfortunately, it also produced asymmetric
results.
As it was shown in Figure 7.14, the asymmetry was always in the form of
significantly enlarged separation in one of the channel’s streamwise corners,
extending far upstream from the shock location. It is therefore clear that
numerical simulations have some problem in flow modeling along a corner.
If one wants to avoid corner effects in experiments, the best method is to
chamfer the corners. This was done in our nozzle with 5 mm chamfers, which
corresponded to the boundary layer thickness. First of all two chamfers were
introduced adjacent to the rectilinear test wall. The effect was significant,
causing the symmetry of flow structure on the rectilinear wall. It turned out,
however, that the strong asymmetry simply shifted to the upper contoured
wall. Next we introduced chamfers to the remaining two corners adjacent to
the upper wall. With all four chamfers in place, the flow structure became
fully symmetric.
With this interesting result a suspicion emerged that the problem might
simply concern the implementation of numerical schemes and turbulence
models in numerical codes. Therefore two additional attempts were made.
In first the chamfering in the corners included only three volumes in the
corner. In the second the chamfer size was 1 mm. Unfortunately, these two
attempts again produced an asymmetric flow structure.
Thus this problem in the UFAST project remains unsolved. It may be char-
acterised as an unphysical eruption of separation along one of the streamwise
corners caused by the shock wave. The condition for this behaviour in nu-
merical simulations is the very slow gradient of flow parameters along the
channel. Exactly the same problem was observed by NUMECA in the case
of a nozzle from Cambridge in Flow Case 2.2.
The second type of a test section used in IMP experiments was
a curved passage, as shown in Figure 7.15. Such a test section was previously
used in EUROSHOCK projects [1]. There were significant differences in the
flow structure in respect to the first test section. Firstly there was a contin-
uous acceleration along the convex test wall. For this reason the shock wave
Fig. 7.15 Curved duct test section
was much more stable than in the first test section. Another difference was
that a local supersonic area appeared on the convex wall of the curved duct,
hence the shock was strong only at the convex wall and vanished towards the
concave wall. Therefore the shock induced separation also only took place
at the convex wall. This is an advantage in comparison to a rectilinear test
section, where separation takes place at all four walls.
Variation of the Mach number upstream of the shock was obtained by ad-
justing the outlet section size to vary the mass flow rate. This affected the
local supersonic area size. It followed that changes in the pre-shock Mach
number also changed the shock location. Although this was inconvenient for
measurements, this allowed us to use one test section for many flow con-
ditions. Increases in the interaction Mach number shifted the shock wave
downstream. In the accelerated flow along the convex wall the boundary layer
gathered more momentum and was less sensitive to the disturbances than in
the case of a rectilinear wall. Therefore in the second test section incipient
separation started at a higher pre-shock Mach number. Here, only two cases
of pre-shock Mach number were studied: M = 1.33 as an ’un-separated’ flow
case, and 1.43 as a flow case with strong separation.
The schlieren image of the interaction flow field structure at M = 1.43
is presented in Fig 16. The shock wave forms a λ-foot at the convex wall
and becomes weaker in the middle of the channel to finally disappear in the
direction of the concave wall. The λ-foot terminates at the triple point, from
where the shear layer originates, visible in the picture as a gray shaded border.
The vertical white lines show the location of the measurement traverses of
the boundary layer.
The curved channel flow case was simulated using a number of methods.
Figure 7.17 shows the static pressure distribution along the convex wall.
Although generally good, the CFD results reveal a difficulty in reproducing
pressure distribution along the entire channel length. The LMFA results are
good at the shock and at the channel outlet, but show the strongest deviation
between the two. The IMP results are good at the shock and downstream of
Fig. 7.16 Visualisation of flow structure at M = 1.43
Fig. 7.17 Static pressure distribution at a convex wall, M = 1.43

it, but show deviation at the channel outlet. Here DES does not show any
superiority.
The general over-shoot of results downstream of the shock wave is asso-
ciated with separation under-prediction by all CFD methods. This also con-
cerns the results which predict shock downstream pressure distribution well.
The boundary layer profile results for the three traverses shown in Figure 7.16
are presented in Figure 7.18.
It should be pointed out that Figure 7.18 shows the profiles of stagnation
pressure, as this was directly measured in the experiment. One could show
velocity profiles, but these would contain much more processing of measure-
ment data, so the uncertainty would increase. The undisturbed boundary
layer profile was measured at traverse X = 15. Looking at CFD results it
becomes clear that the boundary layer thickness is generally over-predicted.
Only the Liverpool DES results come close to the measured boundary layer
profile. Liverpool’s new turbulence model implementation provides an even
thinner layer.
The next traverse, at X = 67, is located downstream of the shock wave
in the separated region, as the measured shape of the stagnation pressure
profile indicates. In general, the CFD significantly under-predicts stagnation
pressure disturbance. One could expect a thicker boundary layer in the CFD
upstream of the shock to result in stronger separation. Yet the results disprove
this. The Liverpool RANS with standard k-ω model results come closest to
the measurements near the wall. The most filled profile is delivered by DES.
Another challenge for all CFD methods is the reproduction of the shear
layer, generated by the triple point, which is visible in the schlieren pictures
(Figure 7.16). Its presence is also indicated by the stagnation pressure change,
marked by the circle for the X = 67 traverse in Figure 7.18. Above the shear
layer only normal shock is present and here all CFD results coincide very well.
However, below the shear layer, downstream of the λ-foot no CFD is able to
predict a characteristic reduction of loses in comparison to normal shock.
This results from an insufficient mesh resolution in the λ-foot front shock
formation area within the boundary layer. In this respect, the standard RANS
approach of Liverpool delivers the best coincidence with the experiment. The
λ-foot is a flow detail which requires exceptionally high mesh resolution. Such
a resolution is never used in typical flow analyses, even when using DES/LES
methods. Unfortunately, details concerning λ-foot structure are necessary to
correctly predict the flow above the separation zone.
The last traverse (X = 145) is located just downstream of the reattach-
ment. The pressure distribution results from all the CFD methods differ from
the experiment. A slight λ-foot triple point effect, so clearly visible in the ex-
periment, is only found in the Liverpool DES/LES and a RANS with a new
k-ω turbulence model results.
The size of the separation zone is presented in Figure 7.19 by means of
the oil flow visualisation. The first compression, from the dashed line to the
reversed flow edge, is much thinner (in the middle of the wall) than in the
Fig. 7.18 Comparison of measured and simulated boundary layer stagnation

pressure
Fig. 7.19 Oil visualisation of separation at M = 1.43

case of the straight nozzle in Figure 7.5. This is probably due to the fact that
in the curved section of the shock is much more stable than in the rectilinear
nozzle.
The URANS simulation results from Liverpool and IMP PAN reveal
a much shorter separation length, as shown in Figure 7.20. In the IMP plot,
moreover, the contours of wall shear stress are shown. It is visible that the low
stress zone is much larger than the reversed flow area alone. This behaviour
is similar to that in the rectilinear nozzle in Figure 7.7.
Fig. 7.20 CFD surface streamline visualisation
As expected, the natural oscillations of the shock wave in the curved duct
test section were much smaller than in the straight nozzle without a pressure
gradient. Indeed, the shock can be considered stable. The RMS for both Mach
numbers are shown in below. The shock wave movement with separation at
a higher Mach number is twice as large as in the case without separation.
However, these RMS values are one order of magnitude smaller than in the
straight nozzle.
Table 7.3
Mach number 1.33 1.43
RMS 0.104 0.240
Conclusions from reference flow cases
The project presented here allowed us to investigate the unsteady aspects of

shock wave – boundary layer interaction in typical flow configurations. The
natural unsteadiness investigated here was generally neglected in the past.
This inherent unsteadiness becomes particularly important when different
flow control techniques are applied.
Natural unsteadiness is especially enhanced by low Mach numbers of in-
teraction M < 1.2 and also by the appearance of separation at M > 1.4.
7.3 Flow Control 199
This implies that natural unsteadiness of the shock wave is the weakest at
incipient separation (1.3 < M < 1.35).
Shock unsteadiness is also sensitive to streamwise flow development. When
the flow parameter gradients along the nozzle are very weak, shock unsteadi-
ness is enhanced.
There are many reasons for shock wave unsteadiness. It is often excited by
upstream disturbances in the incoming boundary layer. Likewise, interplay
between shock induced separation and the downstream throat may lead to the
flow unsteadiness inherent to the test section, especially in curved channels,
although this was not observed in our experiments. Generally, low frequency
unsteadiness did appear in our experiments, but the reason for this is not
clear.
7.3 Flow Control

Rectilinear test section
Up to now, in most tests, the shock interaction parameters were provided,

but information on streamwise vortex generation was either insufficient or
absent. In the rectilinear test section (as shown in Figure 7.21) there is an area
along the lower wall with a constant Mach number. Located here is not only
shock wave boundary layer interaction, but also vortex generators. Thanks
to our test section concept, the shock interaction and the flow control devices
have the same Mach number value, thus enabling us to obtain a unique flow
configuration definition.
Fig. 7.21 Rectilinear test section
AJVG effectiveness depends on two jet inclination angles, α and θ (see Fig-
ure 7.22), and jet diameter ϕ. The stagnation parameters of AJVG are equal
to the stagnation parameters of the main stream. This assumption eliminates
the necessity of jet energizing equipment, hence our AJVG is considered to
be a passive method. In our experiments three configurations of AJVG were
tested: a standard configuration (α = 90◦ , θ = 45◦ ) with a ϕ = 0.5 mm di-
ameter and a ϕ = 0.8 mm diameter and an optimised one (α = 75◦ , θ = 30◦ )
with a ϕ = 0.5 mm diameter only [3,4,5].
Fig. 7.22 AJVG configuration
The investigations revealed that co-rotating AJVG induce a spanwise cell-

like structure, including the separation line, as seen in the oil flow visualisa-
tion on the wall (Figure 7.23).
Fig. 7.23 M = 1.45 interaction case without and with AJVG

As seen in Figure 7.23, AJVGs can even eliminate the separation but
leave a separation line structure which spreads the shock wave and thus re-
duces shock losses. One can also see that streamwise vortices are also present
downstream of the interaction. In the case of weaker streamwise vortices
(Figure 7.24, with 0.5 mm jet diameter) the separation zone is only reduced
and the vortices terminate at the reattachment line. Co-rotation of vortices
induces differences in flow structure close to the side walls.
Fig. 7.24 Separation structure for ϕ = 0.5 mm
The arrows in Figure 7.23 indicate where boundary layer measurements

were taken. The obtained results show that the boundary layer thickness and
displacement thickness with streamwise vortices are slightly lower already
upstream of the shock wave. But the strong effects are especially visible at
the last traverse, where AJVGs cause flow reattachment.
Figure 7.25 shows differences in the thickness of the boundary layer and
displacement in the three traverses along the wall with and without AJVGs,
thus measuring the effect of AJVGs.
One can see in Figure 7.25 that the effects of AJVG in boundary layer in-
tegral parameters can be observed already in the separation area but strong
differences are present in the reattachment zone. In Figure 7.26 velocity pro-
files are compared between a reference one and three AJVG configurations.
In the reference case there is a fully separated zone. Application of any of
the AJVGs reduces the thickness of the boundary layer. In the outer area of
the boundary layer one can notice that the optimised AJVG (ϕ = 0.5 mm) is
more effective than the standard one, providing a fuller velocity profile. Larger
differences between AJVGs occur very close to the wall. For the ϕ = 0.5 mm
jets a substantial improvement is obtained using optimized AJVG geometry.
The significant effect of the jet diameter is also proved in Figure 7.26.
Fig. 7.25 Effect of AJVG on the δ and δ1 along the rectilinear wall at M = 1.45
Fig. 7.26 Stagnation pressure profiles for reference case and all AJVG
Numerical simulations of the flow with AJVGs are very challenging because
it is impossible to include all AJVGs in the simulation, as the size of the grid
would be much too large. The only possibility is to simulate a slice of the
channel including only one AJVG with periodic boundary conditions on the
sides of the simulation domain. This is equivalent to the case of an infinitely
deep test section. Unfortunately, these differences in approach have significant
consequences, as shown in Figure 7.27 (without AJVGs). The differences
already appear when using the RANS method.
Fig. 7.27 Differences between full geometry simulation and a slice-periodic ap-
proach, M = 1.45
The first difference is in the wall static pressure distribution (Figure 7.27a).
Predicted pressure downstream of the shock is much lower than in the full
geometry simulations. It is even lower than in the experiment. This implies
that the shock induced separation in the slice approach simulations is even
larger than in the experiment. Formation of larger separation is not only
caused by the absence of a side wall boundary layer and corner separation
but also by a difference in the incoming boundary layer thickness, as seen in
Figure 7.27b.
These differences are more visible in the downstream traverse as shown
in Figure 7.28. In this plot one can see that the slice-periodic assumption
provides a velocity profile which is much closer to the experiment than the
full geometry simulations. This is especially true at the wall.
The structure of a simulated flow field in the slice case is presented in
Figure 7.29. Indicated here are the CFD separation and reattachment points,
as well as the size and location of separation in the experiment (box). In
CFD reattachment is located exactly in the same point as in the experiment.
Separation, however, is noticeably further upstream in the CFD. The reason
for this is the larger size of the λ-foot in the CFD, which means that the same
main shock location leads to a more upstream start of compression than in
the experiment.
Unfortunately, numerous slice-approach simulations using AJVG did not
lead to a converged solution, in spite of long simulation times.
Fig. 7.28 Velocity traverses at X = 292, M = 1.45
Fig. 7.29 Separation length in a slice approach, without – AJVG
AJVGs introduce a disturbance which is small but acts constantly and

thus has a stabilising effect on shock wave oscillations. Streamwise vorticity
destroys typical spanwise coherent structures induced by the spanwise sepa-
ration line. The table below gives the RMS of shock oscillations (mm). The
amplitude of shock oscillation is reduced by about 50%. The scope of shock
unsteadiness is proportional to the size of separation. Therefore, AJVGs with
0.8 mm diameter holes, which eliminate separation, have the greatest effect
in reducing shock unsteadiness.
Table 7.4
Main shock RMS 0.5 mm stand. VG 0.8 mm stand. VG 0.5 mm optim. VG

reference
1.390 0.862 0.621 0.781
Connected with the stabilisation of the shock wave is the considerable

stabilisation of fluctuations (hot film measurements) in the boundary layer
downstream of the interaction (Figure 7.30). This stabilisation is obtained by
eliminating the low frequency part of the spectrum, which implies that large
scale spanwise eddies and shock movements are strongly reduced.
Fig. 7.30 CTA measurements in the separationzone
Curved channel test section
In the curved channel test section two different control methods were inves-
tigated, see Figure 7.31. One concerned an optimised AJVG with a 0.5 mm
jet diameter. The other involved suction through a perforated wall in the
interaction area.
Fig. 7.31 Curved nozzle test section
The suction method is closely connected with experiments in the EU-

ROSHOCK II project. A hybrid method was investigated there, which used
passive control and followed by additional suction. These methods have shown
the ability to bring the boundary layer to the same state as in the case with-
out passive control.
After the EUROSHOCK II project, the application of suction to the pas-

sive cavity was proposed. This showed that suction upstream of the shock
eliminates the λ-foot and reduces the boundary layer, which remains thin
even downstream of the interaction. The goal of the UFAST project mea-
surements was to investigate the effect of suction on shock unsteadiness and
for the first time to carry out numerical simulations of this flow field.
The effect of both methods of flow control on static pressure distribution
along the test wall (M = 1.43) is presented in Figure 7.32. The reference case
is strongly separated. Higher pressure downstream of the interaction implies
weaker separation. Figure 7.32 shows that the suction method in particular
significantly counteracts separation, to a much greater extent than in the case
of AJVGs.
Fig. 7.32 Static pressure distribution along the test wall
The effect of both methods is most visible when comparing velocity pro-
files in two traverses downstream of the shock wave. These are shown in
Figure 7.33. As far as profile fullness is concerned, one may say that AJVGs
have a positive effect, but suction provides even better results.
As already shown in Figure 7.18, there is a local zone of reduced losses
downstream of the λ-foot, which means higher stagnation pressure (Fig-
ure 7.18) or velocity, as in Figure 7.33. The location of this local velocity
maximum in Figure 7.33 shows that suction considerably reduces the size of
the λ-foot.
The effect of the AJVGs and suction can be judged by the values of
the boundary layer displacement thickness along the convex wall. These are
shown in Figure 7.34 (M = 1.43). AJVGs considerably reduce the growth of
Fig. 7.33 Velocity profiles for both flow control methods at X = 67 and 145
Fig. 7.34 Effect of both flow control methods on the displacement thickness
M = 1.43
Fig. 7.35 Oil visualisation M = 1.43, reference case (left), with application of
AJVG
the δ1 . Application of suction further reduces δ1 growth even downstream of

the suction cavity.
The effect of AJVGs on shock induced separation is presented in
Figure 7.35. A comparison of both pictures shows that this flow control
method is able to three times reduce the separation size. It was not pos-
sible to make an oil flow visualisation on the perforated wall. But numerical
simulations show that separation is fully eliminated by suction.
Numerical simulation of AJVGs is very difficult because they require very
fine mesh resolution around the jet and downstream, where the vortex is
created. Therefore, as in the case of the rectilinear nozzle, a slice of the
channel with periodic boundary conditions and one AJVG was used. In effect
this was the simulation of an infinitely deep channel.
Figure 7.36 shows that the “slice” approach in the case of the curved
channel and the AJVG does not have a noticeable effect on static pressure
distribution along the wall. This is an important difference compared with
the rectilinear test section results.
The coincidence of pressure distribution in Figure 7.36 for the ”slice” ap-
proach with and without AJVG implies a similar separation behavior in both
cases. In Figure 7.37 the length of separation with and without AJVG is
marked by two vertical lines. In the case of AJVG application the flow is
strongly 3-D but the distance between separation and reattachment is nearly
the same as in the reference case.
The coincidence of separation lengths in the numerical results, demonstrat-
ing that AJVGs have no effect, contradicts the experiment results. Figure 7.37
shows also the length of separation and its location in the experiment (box).
The experiment shows a significant effect of AJVGs on the separation length
(as can be seen in the oil visualization in Figure 7.35). This implies that the
slice method flow approximation deviates too much from physical reality.
In case of suction, it is difficult to visualize flow structure on the wall.
One can not use oil flow visualization because the oil would stop the holes
and block the transpiration stream. Boundary layer profile measurements
Fig. 7.36 Simulations with full geometry and periodic with and without AJVG
Fig. 7.37 Flow structure in slice assumption without and with AJVG and
experiment
(Figure 7.33) indicate that there is no separation downstream of the shock

wave when suction is applied.
The only way to examine the flow structure in the transpiration area is
through numerical simulation. These were carried out by LMFA using the
URANS method. Suction through the perforated plate was distributed ac-
cording to the pressure variations on the test wall surface. Figure 7.38 shows
the spanwise distribution of normal velocity on the wall surface. Suction close
to the side walls is stronger than in the middle of the channel. This must be
due to corner flow effects.
Along the centre line of the wall, suction intensity reflects the pressure
distribution (Figure 7.39). Due to the shock wave, a rapid increase of pres-
sure occurs somewhere in the middle of the perforated plate. Pressure in
the entire cavity is uniform. In the upstream part of the perforated plate
the tangential stream is supersonic, with low pressure reducing the pressure
drop and suction intensity over the plate. Downstream of the shock wave the
pressure rapidly increases. Thus the pressure drop and suction over the plate
increases, too.
The distributions in Figures 7.38 and 7.39 show that the structure of tran-
spiration flow is rather complicated. The suction distribution over the per-
forated plate is presented in Figure 7.40. The contours are lines of constant
normal velocity, drawn on the first layer of cells adjacent to the wall. There-
fore, the contours are limited to the perforated plate only. It may be noticed
that the greatest suction takes place in the downstream corners of the plate,
close to the side walls. Figure 7.41 shows that a short distance above the wall,
above the strongest suction, there is a considerable velocity in the opposite
direction.
This complicated flow behavior can be explained by the CFD flow visuali-
sation presented in Figure 7.42. This is a view from the upstream side of the
channel.
Fig. 7.38 Normal velocity at perforated plate – spanwise

Fig. 7.39 Normal velocity at perforated plate – streamwise, along centre line
Fig. 7.40 Normal velocity at wall surface
Fig. 7.41 Normal velocity above the wall

Fig. 7.42 Secondary flow structure
One can see the perforated plate (dark green). It is important to notice the
strips marked inside the side wall boundary layer. Just downstream of the
shock wave this secondary flow moves towards the convex wall. This is typical
behaviour in a curved tunnel. No separation is induced by the shock wave
at side walls. Instead, there is a strong down wash inside the boundary layer
towards the convex wall. This secondary flow, when reaching the convex wall,
is sucked into the downstream corner of the perforated plate. This flow area
is part of the separated zone. The main stream moves above the separation
and therefore, as shown in Figure 7.41, a positive normal velocity component
appears.
The natural shock oscillations in the curved nozzle are very weak, and in
this respect not much can be obtained using flow control methods. Neverthe-
less, the table below shows the scope of shock wave oscillation.
Table 7.5
Main shock M = 1.43 0.5 optim. VG suction

reference
0.240 0.267 0.151
This table shows that AJVGs cannot reduce shock wave oscillation in this
configuration. However, suction is able to reduce these oscillations nearly by
half.
7.4 Conclusions 213
7.4 Conclusions
The UFAST project included a wide range of basic and flow control exper-
iments whose main objective was to investigate the unsteady effects of the
shock wave – boundary layer interaction. In the flow case presented here the
main objective was to examine natural unsteadiness. Numerical simulations
were performed for cases selected from our experimental database. Compar-
ison of the numerical results with experimental data required a very close
correspondence in terms of geometry and boundary conditions. This was es-
pecially important when investigating natural unsteadiness arising from the
shock wave – boundary layer interaction.
The numerical investigations were carried out for a straight nozzle and
a curved nozzle, as the basic case and the flow control case using air jet
vortex generators (AJVGs).
The numerical simulation of a straight nozzle with a high Mach No. (M =
1.45) proved to be very challenging. The main feature of the nozzle was
the constant Mach number at the outlet. In order to keep numerical model
conditions as similar as possible to those in the test section, the location of
the shock was adjusted by regulating the second throat, not by varying the
outlet pressure. This required performing more simulations in order to find
the proper shock location. Such an approach was necessary because unsteady
effects are important in numerical simulations.
The results obtained using the Spalart-Allmaras turbulence model revealed
an asymmetric flow structure downstream of the shock wave. This was con-
firmed using SPARC, FINE/Turbo Numeca and Fluent. The simulations with
these codes were conducted with the identical boundary conditions and an
identical mesh. In all cases the flow structure was asymmetric, although the
location of separation varied. Yet the existence of asymmetry is not confirmed
in experiments. To find a reason for the asymmetry, the influence of mesh
resolution was examined and other turbulence models (SST and RSM) were
tried. Only one turbulence model, the two-equations k-tau (Speziale-Abid-
Anderson) implemented in SPARC, produced a symmetric flow structure.
It was finally concluded that the reason for the asymmetry in flow struc-
ture was the over-prediction of corner vortices. After numerous simulations
it was found that symmetry may be obtained by chamfering the corners of the
test section. One can conclude that corner flow modeling is crucial in some
cases concerning the flow structure in nozzles. The lack of proper corner flow
(corner vortex) prediction causes considerable differences between the flow
topologies of numerical results and experiments. Our experience in numerical
simulations has shown that such strange behavior is obtained in the case of
very small flow parameter gradients along the nozzle length.
The numerical results obtained for the straight nozzle and the curved noz-
zle without flow control devices (basic flow cases) showed much less separation
than in the case of experiments.
214 References
In the case of both nozzles the numerically predicted shock wave lambda
foot is smaller than the one in experiments.
Under-predicted separation bubble lengths in CFD explains the lower os-
cillations of the shock wave than those measured in experiments. The fact
that low frequency oscillations were not displayed in CFD may have been
partly due to the simulation time frame being too short. In the case of the
curved nozzle no oscillations were detected. On account of the low amplitude
and high frequency oscillations, the RANS results were very close to URANS
results.
The air jet vortex generator numerical simulations used simplified geome-
tries, as they required very high mesh resolution. It was decided to make
simulations for one AJVG pitch, i.e. one slice of the whole domain was taken
into consideration. In such an approach, periodic boundary conditions on
bounding planes were applied. The lack of the sidewall boundary layers influ-
enced the boundary layer upstream of the shock. In this case, the boundary
layer was thicker and influenced the size of the lambda foot. The lambda foot
was much larger than in the case the full domain simulations as well as that
of the experiments. Downstream of the shock, the boundary layer profiles
also differed greatly from the experimental ones. To conclude, slice method
of AJVG simulations are not an adequate means of modeling wind tunnel
flows. Although requiring enormous computer resources, it seems that only
full test section simulations may deliver satisfactory results.
References
[1] Doerffer, P., Bohning, R.: Shock wave – boundary layer interaction control by
wall ventilation. Aerospace Science and Technology 7, 171–179 (2003)
[2] Magagnato, F.: SPARC – Structured Parallel Research Code. TASK Quar-
terly 2(2), 215–270 (1998)
[3] Flaszyński, P., Szwaba, R.: Optimisation of streamwise vortex generator. De-
velopments in Mechanical Engineering 2 (2008)
[4] Flaszyński, P., Szwaba, R.: Experimental and numerical analysis of streamwise
vortex generator for subsonic flow. Chemical and Process Engineering 27, 985–
998 (2006)
[5] Szwaba, R., Flaszyński, P., Szumski, J., Telega, J.: Shock Wave – Bound-
ary Layer Interaction Control by Air-Jet Streamwise Vortices. In: Proceed-
ings of the 8th International Symposium on Experimental and computational
Aerothermodynamics of Internal Flows, vol. 2, pp. 541–547 (2007)
[6] Ryszard, S., Piotr, D., Krystyna, N., Oskar, S.: Flow structure in the region of
three shock wave interaction. Aerospace Science and Technology 8(6), 499–508
(2004)
[7] Doerffer, P., Bohning, R.: Aerodynamic performance modeling of porous
plates. Aerospace – Science & Technology Journal 4(8) (2000)
[8] Doerffer, P., Szwaba, R.: Shock wave – boundary layer interaction control by
streamwise vortices. In: XXI ICTAM, Warsaw, Poland, August 15-21 (2004)
References 215
[9] Doerffer, P., Boelcs, A., Hubrich, K.: Streamwise vortices generation by air jets
for a shock wave – boundary layer interaction control. In: ASME Confrence,
Vienna, Austria, June 14-17 (2004)
[10] Doerffer, P., Zierep, J., Bohning, R.: Perforated plate aerodynamics for passive
shock control. In: Symposium Transonicum IV, Goettingen, September 2-6
(2002)
[11] Szwaba, R., Flaszyński, P., Szumski, J., Telega, J.: Shock Wale – Boundary
layer interaction control by Air-Jet Streamwise vortices. In: Proceedings of 8th
ISAIF Conf., Lyon, France, pp. 541–547 (July 2007)
[12] Szwaba, R.: Shock Wave Induced Separation Control by Air-Jet Vortex Gen-
erator in the Curved Nozzle, American Istitute of Aeronatics and Astronatics
(AIAA). In: Proceedings XIX International Symposium on Air Breathing En-
gine (ISABE), Montreal, Kanada (2009)
Part III
Shock Reflection
Chapter 8
Oblique Shock Reflection at M = 1.7
(Sergio Pirozzoli)
8.1 Presentation of Flow Case
The present report summarizes the activities that have been performed for
flow case 3.1 (TUD impinging-shock case) within the context of the Euro-
pean Commission 6th Framework Programme ‘UFAST – Unsteady Effects
in Shock Wave Induced Separation’, contract number 012226 (AST4-CT-
2005-012226). The experimental measurements for the case study have been
performed at the High Speed Laboratory of the Delft University of Technol-
ogy department of Aerospace Engineering (partner 7: TUD) The flow case
consists in the reflection of an oblique shock from a planar surface in the low
supersonic regime. Data have been acquired for a free-stream of Mach 1.7 with
an incident shock wave corresponding to a flow deflection of 6◦ . The peculiar
feature of the experiment is the very large Reynolds number (Reθ ≈ 50000), in
view of the substantial boundary layer thickness (δ99 = 17 mm) and high stag-
nation pressure (230 kPa) upstream of the interaction zone. Particle Image
Velocimetry (PIV) has been applied as major diagnostic tool for the investi-
gation of the interaction, both in the standard form (planar two-components
and stereo) and in more advanced configurations (notably, Dual-plane and
Tomographic PIV).
A campaign of numerical investigation has been conducted by means of
both RANS (steady and unsteady) and LES. In particular, the group of
the University of Rome “La Sapienza” (partner 10: URMLS) has conducted
both RANS and LES simulations; the group of the University of Southamp-
ton (partner 9: SOTON) has carried out LES simulations; and the group
of the Ecole Centrale de Lyon / Laboratoire de Mecaniques des Fluides et
Acoustique (partner 17: LMFA) has performed RANS calculations.
The report is organized as follows: in Section 8.2 the experimental method-
ology is described; in Sections 8.3 and 8.4 the setup of the numerical
simulations is illustrated; in Section 8.5 the main results of comparison of
experiments and numerics are presented; the main achievements (and limi-
tations) of the study are listed in Section 8.6.
springerlink.com
220 8 Oblique Shock Reflection at M = 1.7 (Sergio Pirozzoli)
8.2 Experimental Methodology (Source TUD

Deliverable 2.3.10)
Measurements of SBLI under conditions of incipient flow separation have
been performed by the group of the Technical University of Delft (TUD).
The main results are reported in the TUD final report, and in a series of
publications [16, 15, 17].
The experiments were carried out in the blow-down transonic-supersonic
wind tunnel (TST-27), shown in Figure 8.1. The facility is capable to gen-
erate flows in the Mach number range 0.5 to 4.2 in the test section. The
Mach number is set by means of a continuous variation of the throat sec-
tion and flexible nozzle walls. The air is supplied from a storage vessel of
300 m3 charged to 42 bar and stored at ambient temperature (280 to 290 K
typical), which allows a blow-down operating use of the tunnel of approxi-
mately 300 seconds maximum (depending on operating conditions), before
recharging is required. Typical run times applied in practice are 30 sec up to
2 min. In the present experiment, tunnel operation is in the low-supersonic
regime (below Mach 2). The wind tunnel operation conditions are typically
requiring a stagnation pressure of 200 to 300 kPa, yielding a unit Reynolds
number in the test section of 25 to 40 × 106 m−1 . The settling chamber of
the wind tunnel has a square cross section of 800 mm × 800 mm. In the first
section of the converging nozzle the flow channel contracts in lateral direc-
tion only, to the final (constant) channel width of 280 mm. Subsequently, the
Fig. 8.1 TUD wind tunnel geometry

8.2 Experimental Methodology (Source TUD Deliverable 2.3.10) 221
nozzle upper and lower walls provide the (continuously adjustable) contoured
converging-diverging shape, symmetrical with respect to the tunnel centre
line, to produce the required Mach number in the test section. The distance
from the throat to the centre of the test section measures approximately 2
meters. The major components and dimensions of the nozzle and test section
are indicated in Figure 8.1. For more details and technical drawings of the
nozzle geometry, see TUD UFAST deliverable 2.3.3.
For the present experiments, the boundary layer along the test section
top wall is considered. For representation purposes, however, all geometrical
representations will be inverted to show the test wall at the bottom. The
shock generator is placed in the free-stream flow to generate the incident
planar shock wave of prescribed strength (flow deflection angle). The shock-
generator configuration is a side-wall mounted semi-wedge; with a span of
270 mm it spans the width of the test section, apart from a small clearance
(about 1 cm) on the unsupported side.
Flow conditions
The flow conditions for the interaction described in this chapter are summa-
rized below, corresponding to a Mach 1.7 free stream and 6 degree incident
shock deflection. The final choice of Mach number and flow deflection angle
was dictated by the minimum Mach number permitting a proper operation
of the tunnel in view of choking with the given shock generator installed. The
total pressure was set to a nominal value of 230 kPa. The total pressure was
verified not to have a significant effect on the tunnel operating conditions
and flow quality. The flow conditions are as follows:
• Free-stream Mach number: M = 1.69
• Flow deflection angle: θ = 6.0◦
• Total pressure: P0 = 230 kPa
• Total temperature T0 = 273 K
• Free-stream velocity: U∞ = 448 m/s
• Unit Reynolds number: 35.9 × 106 (1/m)
Boundary layer properties (upstream of interaction):
• Boundary layer thickness: δ99 = 17.3 mm
• Momentum thickness: θ = 1.40 mm
• Momentum Reynolds number: Reθ = 50.0 × 103
Measurement procedure
Particle Image Velocimetry (PIV) was used as main diagnostic technique.

Measurements were carried out with different resolutions and considering
different regions of the flow. The fields of view (FOV) applied in the inves-
tigation are shown in Figure 8.2, where the grey areas represent the tunnel
Fig. 8.2 Overview of the interaction and relative placement of the fields of view
(FOVs) of the different PIV experiments
wall and the tunnel window. A wide field of view (‘Panoramic’) was obtained
by the simultaneous use of two cameras in a side-by-side arrangement. In
addition, two FOVs with increased magnification were used, one to cover the
incoming boundary layer (‘Boundary Later Zoom’) and another one for the
central part of the interaction (‘Interaction Zoom’). Besides standard PIV,
also dual PIV has been applied by operating two identical PIV systems (op-
tically separated by polarization) at variable time delay, in order to obtain
time-correlated information of the interaction.
All particle image recordings were made with 1376 × 1040 pixel camera(s)
at a recording rate of 5 Hz. The particle images were cross-correlated with
a window-deformation iterative multi-grid scheme, applying a final window
size of 31 × 31 pixels with 75% overlap. As a general characteristic of the PIV
measurements, the last four data points closest to the wall have to be consid-
ered unreliable due to surface reflection and wall overlap of the correlation
windows.
The entire dual-PIV data set for all different time delays was employed for
an overall statistical analysis of the flow, resulting in a large ensemble size of
about 4000 (which is the total sum of all image pairs) that proved beneficial
for convergence the statistical flow properties.
Characterization of incoming boundary layer
The incoming boundary layer has been assessed from the PIV data with the
different FOVs (excluding the Interaction Zoom, where it is not visible). As
the BL zoom does not cover the entire boundary layer height, this data was
predominantly used to extrapolate the velocity profile in the near wall region
for the log-law fit, whereas the integral parameters were determined from
the velocity profiles extracted from the other two FOVs. Unless otherwise
stated, the results from the dual-PIV data set will be taken as reference,
the panoramic data set providing nearly identical results. The BL velocity
statistics are computed using only validated data (no substituted vectors were
used), while additional averaging in the streamwise direction was applied to

improve convergence and to reduce the influence of measurement noise.
The boundary layer mean velocity profiles from the different experiments
are depicted in Figure 8.3, where the velocity is nondimensionalized by the
free-stream value, and the y coordinate by the boundary layer thickness δ99 .
As previously remarked, note that for each data set the last four data points
closest to the wall are not reliable.
Fig. 8.3 Mean velocity profile in incoming boundary layer
The boundary layer integral parameters are listed in Table 8.1. The dis-
placement thickness, δ ∗ , momentum thickness, θ, and shape factor, H, were
obtained by integration of the velocity profile, where the temperature and
density profiles of the boundary layer were estimated from the modified
Crocco-Busemann relation, assuming adiabatic wall conditions and a recov-
ery factor r = 0.89. The accuracy of the integration was increased by means of
a power-law extrapolation of the velocity profiles towards the wall (the value
of the power-law exponent n is included in the table). Furthermore, the val-
ues of the ‘incompressible’ displacement thickness, δi∗ , momentum thickness,
θi , and shape factor, Hi , are also provided.
Using the quantities determined above, the following Reynolds numbers
have been obtained:
• Re = ρ∞ u∞ /μ∞ = 3.59 × 107 (1/m)
• Reδ = ρ∞ u∞ δ99 /μ∞ = 6.17 × 105
Table 8.1 Integral properties of boundary layer upstream of interaction
Data set n δ ∗ (mm) θ (mm) H δi∗ (mm) θi (mm) Hi

Dual-PIV 7.90 3.31 1.39 2.38 2.00 1.60 1.25
• Reδ∗ = ρ∞ u∞ δ ∗ /μ∞ = 1.19 × 105

• Reθ = ρ∞ u∞ θ/μ∞ = 5.00 × 104
The skin friction coefficient and friction velocity have been estimated from
the boundary layer velocity profile, following two approaches: using semi-
empirical relations based on the Reynolds number and, secondly, a log-law
fit. The main results of the fitting procedure are illustrated in Table 8.2.
In summary, the skin friction coefficient is estimated at Cf = 1.49 · 10−3
(uτ = 15.0 m/s) with an uncertainty of 5%.
Table 8.2 Skin friction estimation from log-law fit
Data set Cf uτ [m/s] C method

BL Zoom 1.57 · 10−3 15.46 6.22 Full-fit (uτ & C)
Dual-PIV 1.46 · 10−3 14.89 7.31 uτ prescribed
BL Zoom 1.48 · 10−3 14.99 7.00 C prescribed
Dual-PIV 1.49 · 10−3 15.07 7.00 C prescribed
Panoramic 1.49 · 10−3 15.06 7.00 C prescribed
The log law representation of the velocity profiles extracted from the three
data sets is depicted in Figure 8.4. The solid vertical line indicates the bound-
ary layer thickness in inner scaling (δ + ≈ 9300), while the solid horizontal
line indicates the free-stream velocity (u+∞ ≈ 32). It may be noted that the
data points closest to the wall diverge from the log-law as a consequence of
the spatial resolution limitations (wall overlap of the interrogation window,
in particular). Based on the log-law fit it can be concluded that the mean
velocity measurements are reliable down to y + = 700 (y/δ = 0.066) for the
Dual-PIV case and the Panoramic case. In the Boundary layer zoom, the
mean profile extends reliably down to y + = 200 (y/δ = 0.019).
In Figure 8.5 the distributions of the velocity fluctuations (turbulent nor-
mal stresses) and the Reynolds shear stress are reported. Good agreement
is observed between the two available data sets (Panoramic and Dual). For
further comparison, the same data are displayed using Morkovin’s scaling
in Figure 8.6. Klebanoff’s reference data for incompressible flow is included
for comparison. Good overall correspondence with Klebanoff’s data is ob-
served for both the velocity fluctuations and the Reynolds shear stress. The
erroneous overestimation of the near-wall peak in the streamwise velocity
Fig. 8.4 Mean velocity profile upstream of interaction in semi-log scale
Fig. 8.5 Streamwise and wall-normal velocity fluctuations (left) and Reynolds
shear stress (−u v ), right
fluctuation close to the wall is due to the previously mentioned measure-

ment errors close to the wall, but apart from this effect, this property it
is considered to be resolved correctly down to y/δ = 0.1 (corresponding to
the interrogation window size). The wall-normal velocity component and the
Reynolds stress display a premature fall-off when approaching the wall, and
Fig. 8.6 Turbulence statistics in Morkovin scaling
appear to be resolved (with maximum error of approximately 15%) down

to y/δ = 0.3 for these two data sets considered (Panoramic and Dual-PIV).
The underestimation of the Reynolds stress, in particular in the lower part
of the boundary layer, is thought to be mostly due to an underestimation of
the wall-normal fluctuation, both due to the small magnitude of the quantity
under consideration and the small magnitude of the v-component itself.
Characterization of interaction zone
The mean velocity field is represented in nondimensionalized form, with the

velocity components scaled with the free-stream velocity and the x- and y-
coordinates by the boundary layer thickness. The origin is taken at the ex-
trapolated wall impact point of the incident shock wave. Figure 8.7 shows
the overall view for both in-plane velocity components as obtained with the
panoramic view, while Figure 8.8 shows a close-up of the interaction region
as provided by the interaction zoom FOV (for the u component only). The
thin vertical line in the plot for the vertical component at approximately
x/δ = −2.1 that can be observed in Figure8.7 is an artefact of the stitching
procedure to combine the results of the two cameras.
Figure 8.9 provides an overview of the velocity fluctuation statistics (tur-
bulence components). The same way of representation in non-dimensional
form has been used as for the mean flow data in the previous sections. The
results from the dual-PIV data set have been taken as reference values, as this
data set provides a good spatial resolution, while the large data ensemble size
results in good statistical convergence of the statistical quantities considered.
Unsteady flow characteristics
The flow topology of a sequence of 100 instantaneous realisations has been

evaluated to determine the extent of the excursions of the reflected shock.
Fig. 8.7 Panoramic dataset mean streamwise and wall-normal velocity compo-
nents in the interaction zone
Fig. 8.8 Mean streamwise velocity component with increased resolution (interac-
tion zoom FOV)
Fig. 8.9 Streamwise and wall-normal velocity fluctuations and Reynolds shear
stress in the interaction zone
Furthermore, the approximate interaction length has been determined based

on the extrapolated reflected shock foot and the extrapolated point of impact
of the incident shock. It can be concluded accordingly that the interaction
length is approximately 2δ, and the total extent of the excursions of the
reflected shock is δ/2.
Although no mean flow reversal was observed, substantial regions of re-
versed flow occurred occasionally in the instantaneous flow realisations.
A map of the separation (flow reversal) probability distribution is shown in
Figure 8.10. This quantity shows the probability of the occurrence of reverse
flow (negative velocity) for the complete flow field, based on all realisations
within the ensemble. It indicates the maximum extent of the separation bub-
ble and the maximum rate of the occurrence of a separation bubble within
the ensemble size as a function of the size of the separation bubble. The
maximum probability of reverse flow occurrence is found to be of the order
of 40%.
Fig. 8.10 Flow reversal probability (for reference purposes, the grey lines indicate
V-component contour levels, the black line indicates sonic line)
8.3 Overview of RANS Simulations

The results of (U)RANS computations are reported in the present Section.
Contributions have been provided by URMLS (see deliverable 4.3.7) and
LMFA (deliverable 4.3.14). RANS simulations have the advantage of provid-
ing reasonably accurate predictions at affordable computational price, and
in particular, RANS makes feasible the reproduction of the full-scale experi-
ment (i.e. at the true Reynolds number), and allows for reproduction of the
full geometrical complexity of the flow configuration. Several computations
8.3 Overview of RANS Simulations 229
have been reported, covering different levels of geometric complexity, dif-

ferent Reynolds numbers, different turbulence models, in either 2D and 3D
environments. In the following we provide a short account of the numerical
methodology used by the investigators, and an overview of the main prelim-
inary findings, while the reader is referred to the original deliverables for a
full account of the results.
RANS model equations
Upon suitable simplifications, the filtered- and the Reynolds averaged-

Navier-Stokes equations have the same form (although the meaning of the
variables is different), which allows to use similar strategies for numerical
discretization [5]. In Cartesian coordinates (x1 = x corresponding to the
streamwise direction, x2 = y to the wall-normal direction, and x3 = z to the
spanwise direction), the equations are cast in conservation form as follows
∂ρ ∂(ρ u j )
+ = 0, (8.1)
∂t ∂xj
∂(ρ ui ) ∂(ρ u i u
j )
+ = ∂p
∂xi + ∂
∂xj σij − τij ) ,
( i = 1, 2, 3, (8.2)
∂t ∂xj

∂(ρ E) ∂(ρE + p)uj
+ = ∂
∂xj σij − τij ) ũi ] −
[( ∂
∂xj (q̃j + Qj ) , (8.3)
∂t ∂xj
where ρ, ui , p, E, σij and qi are the density, the velocity vector, the pressure,
the total energy, the viscous stress tensor and the heat flux vector, with
∗
σij = 2 μ Sij , qi = −κ ∂x
∂T
, (8.4)
i
∗ ∂uj
∂xi − 3 ∂xk δij .
∂ui 1 ∂uk
Sij = 12 ∂x j
+ (8.5)
The molecular viscosity μ is evaluated using Sutherland’s law,

3/2
μ̃ T̃ 1+C
= ,
μ∞ T∞ T̃ /T∞ + C
and the thermal conductivity κ is related to μ through κ = cp μ/P r (P r =

0.72). The overbar denotes the spatial filtering operator for LES, and the
Reynolds ensemble averaging operator in the case of RANS. The tilde is used
to denote density-weighted (Favre) averages, f = ρf /ρ; fluctuations with
respect to Reynolds and Favre averages are denoted with a single or double
prime, respectively.
The unresolved terms in the momentum and energy equations are to be
interpreted in RANS as the effect of turbulent fluctuations on the mean flow,
with τij = ρ u u , Q = T
i j j
u . A simple linear eddy-viscosity assumption
j
(Boussinesq approximation) is made to model τij and Qj in the RANS models

that have been considered for the UFAST project, i.e.
1 ∂ T
τij − δij τkk = −2 ρ νt Sij
∗
, Qj = −κt , (8.6)
3 ∂xi
where νt denotes the eddy viscosity, and the turbulent thermal conductivity
κt is determined as κt = cp ρ νt /P rt (P rt = 1). Both two-equations and one-
equation models have been considered, which are briefly described below.
Spalart-Allmaras model
The Spalart-Allmaras model [18] is based on an empirically-derived differ-

ential transport equation for the turbulent viscosity including a destruction
term that depends explicitly on the distance to the wall. For the near-wall
treatment, the Spalart-Allmaras model uses the working variable ν̌, related
to the turbulent viscosity through a damping function fv1 . The standard for-
mulation of the transport equations for ν̌ in compressible flows is as follows
2
Dν̌ 1 ∂ ∂ ν̌ ∂ ∂ ν̌ ∂ ν̌
ρ = cb1 Sˇν ρ ν̌ + ρν + ρ ν̌ + cb2 ρ
Dt σ ∂xj ∂xj ∂xj ∂xj ∂xj
2
ν̌
−cw1 fw ρ , (8.7)
d
with

ν̌ ν̌
νt = ν̌fv1 , Šν = S̃ν + fv2 , S̃ν = Ω̃ = 2Ω̃ij Ω̃ij , χ≡ ,(8.8)
κ2 d2 ν
16
χ3 χ 1 + c6w3
fv1 = 3 , fv2 =1− , fw = g , (8.9)
χ + c3v1 1 + χfv1 g 6 + cw36
ν̌
g = r + cw2 (r6 − r), r≡ . (8.10)
Šν κ2 d2
The model constants are here set to
κ = 0.41, σ = 2/3,
cb1 = 0.1355, cb2 = 0.622, cv1 = 7.1, (8.11)
2
cw1 = cb1 /κ + (1 + cb2 )/σ, cw2 = 0.3, cw3 = 2.
k-ω model
The two-equations turbulence model developed by Wilcox [22, 23] is based on

the solution of a model transport equation for the turbulence kinetic energy
(k) and an equation for ω = /k, which an be regarded as a characteristic
frequency of turbulence,

Dk ∂ ũi ∗ ∂ ∂k
ρ = τij − ρβ kω + (ρ ν + σk ρ νt ) , (8.12)
Dt ∂xj ∂xj ∂xj

Dω ω ∂ ũi ∂ ∂ω
ρ = α τij − ρβω 2 + (ρ ν + σω ρ νt ) , (8.13)
Dt k ∂xj ∂xj ∂xj
with νt = k/ω, and
σk = 0.5, σω = 0.5, α = 5/9, β ∗ = 9/100, β = 3/40.
The k-ω model does not involve damping functions, thus allowing simple
Dirichlet boundary conditions, and it is known to provide good prediction
(better than standard k − ) in the logarithmic region of boundary layers,
even under adverse pressure gradient conditions.
Fares-Schröder model
Starting from Wilcox’ k-ω model, Fares and Schröder [3] proposed a one-
equation model which has the same behavior as the Spalart-Allmaras model
for wall-bounded flows, and that promises slightly better predictioms for jet
and separated vortical flows. Starting from Equation 8.13, Fares and Schröder
[3] considered the transport equation for νt

Dνt 1 Dk Dω
= − νt , (8.14)
Dt ω Dt Dt
which, upon suitable simplifications, and exploiting Bradshaw’s hypothesis,

eventually leads to
Dν̌ ν̌ ∂ ũi
ρ = 2(1 − α)ρ Sij − (β ∗ − β)ρ ν̌ω (8.15)
Dt ω ∂xj

∂ ∂ ν̌ (ρ ν + σρ ν̌) ∂ ν̌ ∂ω
+ (ρ ν + σρ ν̌) +2 ,
∂xj ∂xj ω ∂xj ∂xj
with the following coefficients
σ ∗ = 0.5, σ = 0.5, α = 0.52, βc∗ = 0.09, βc = 0.072, (8.16)
and where the closure functions are defined as follows


1 + 680ψk2 1 ∂k ∂ω
β ∗ = βc∗ fβ∗ , fβ∗ = , ψk = max 0, 3 , (8.17)
1 + 400ψk2 ω ∂xj ∂xj

Ω̃ Ω̃ S
1 + 70ψω ij jk ki
β = βc fβ , fβ = , ψω = , (8.18)
1 + 80ψω (βc∗ ω)3
Note that the functions ψk , ψω do not play a significant role close to solid
walls, since they both attain zero value.
Computational setup for RANS
The flow geometry for the TUD flow case is sketched in Figure 8.11. Differ-
ent levels of geometrical complexity have been considered by the numerical
investigators. LMFA has analyzed the flow in the full geometry reported
in Figure 8.11, including the shock generator, and including the region of
boundary layer growth in the wind tunnel. The data delivered by LMFA also
includes the full effect of three-dimensionality, since the side-wall are included
in the calculation. This of course requires the use of significant computational
resources, and the use of a code capable to handle non-trivial geometries.
A different choice was made by URMLS, that considered a limited portion of
the flow domain, highlighted with a dashed line in Figure 8.11. This type of
approach has the advantage of allowing the use of Cartesian grids, and very
efficient flow solvers. On the other hand, the specification of accurate inlet
conditions at the left boundary of the domain is made necessary, and the in-
coming shock wave must be artificially enforced at the top boundary through
careful application of the Rankine-Hugoniot jump conditions. For a complete
description of the inlet boundary conditions and of the enforcement of the
incoming shock, see URMLS deliverable 4.3.1.
With regard to the numerical discretization of the governing equations,
RANS does not pose a significant challenge, since all turbulence scales are
modeled, rather than being resolved, and numerical schemes with low dissipa-
tion and dispersion are not necessarily required. For instance, the algorithm
used by LMFA solves for the RANS equations with Spalart-Allmaras tur-
bulence model, and it is based on a structured multi-block, finite-volume
solver, using Roe fluxes with minmod limiting function, and implicit time
stepping. The structured multi-bloc mesh used for the LMFA simulation, in-
cluding 4.4 × 106 grid points split into 16 blocks, is shown in Figure 8.12.
The URMLS RANS computations were performed using a Cartesian mesh.
The convective fluxes in the governing equations are discretized by means of
a seventh-order WENO scheme, whereas the additional transport equations
for ν̃, k, ω are discretized by means of a standard second-order TVD scheme
with Van-Leer limiter, to ensure positivity of the transported properties. Vis-
cous fluxes are approximated by standard second-order central differences,
and time advancement is performed by means of the classical four-stage,
fourth-order explicit Runge-Kutta algorithm.
Fig. 8.11 Sketch of computational domain for RANS of TUD impinging shock
flow case
Fig. 8.12 Computational domain for LMFA RANS simulation
Overview of RANS results
The main features of the TUD flow case can be appreciated from inspection
of the 3D RANS results of LMFA, reported in Figure 8.13, showing the
distribution of the Mach number in several planes. The x − y section of
the flow field clearly highlights the formation of the impinging shock past the
wedge, and its interaction with both the bottom wall (which is the subject
of the experimental investigation) as well as with the upper wall. The figure
also shows the diffraction of the reflected shock upon interaction with the
expansion fan that originates at the trailing edge of the shock generator, and
the occurrence of a complex flow pattern in its wake. A section of the flow field
Fig. 8.13 Mach number fields from LMFA-RANS simulation. (a) x − y section;
(b) x − z section; (c) z − y section.
Fig. 8.14 Comparison of RANS-LR with URANS-LR: distribution of mean skin

friction (left) and wall pressure (right)
parallel to the wall (Figure 8.13b) clearly shows the occurrence of secondary
motions near the side-walls, with substantial flow separation. However, as
also seen in the z − y flow sections, the flow is very nearly two-dimensional
in the vicinity of the mid-plane, where the experimental data are collected.
Severals simulations have been performed by URMLS, with the objective
to clarify the sensitivity of the results upon: i) Reynolds number; ii) grid size;
iii) turbulence model; iv) three-dimensional effects; and v) (un)steadiness of
the interaction pattern.
With regard to the sensitivity to the Reynolds number, two sets of numer-
ical simulations have been performed: one at a Reynolds number that closely
matches the experimental one (Reθ ≈ 45000, also see the later Table 8.4),
and labeled as ‘high-Reynolds’ (HR), and one at a reduced Reynolds number
(Reθ ≈ 2800), also attainable from LES, and labeled as ‘low-Reynolds’ (LR).
The RANS-LR test case was preliminarily used to assess the effect of
the mesh spacing upon the interaction parameters. The study (reported in
Deliverable 4.3.1) showed that a mesh spacing of the order of 100 wall units
in the wall-parallel direction was sufficient to guarantee grid-independent
results, whereas the first point off the wall must be placed at a distance of
1÷2 wall units. The LR case was also used to address the possible onset of self-
sustained unsteadiness in the interaction zone. For that purpose, a URANS
calculation was performed by running the same code used for RANS in three-
dimensional, unsteady mode, and stimulating it by means of synthetic inlet
disturbances [13]. The study has shown that, while some unsteadiness can be
generated in the interaction zone, with a breathing motion of the separation
bubble and associated excursion of the reflected shock, it critically depends
upon the type and intensity of the forcing. In any case, the time-average fields
were found to be very similar to the ones obtained from steady RANS. For
illustrative purposes, in Figure 8.14 the distributions of the mean skin friction
coefficient and wall pressure from RANS-LR and URANS-LR are compared.
The high-Reynolds number test case (RANS-HR) was used to assess the
effect of the turbulence model upon the main features of the interaction. The
comparison of the mean streamwise and wall-normal velocity fields obtained
with the three RANS models previously illustrated is shown in Figure 8.15. No
macroscopic difference in the size and in the overall pattern of the interaction
is observed when the RANS model is changed. Some differences (reported in
Deliverable 4.3.1) are observed in the value of the skin friction in the recovery
zone. In particular, the Spalart-Allmaras model exhibits a slower return to
equilibrium past the interaction, which better fits the experimental data.
A three-dimensional computation of the RANS-HR case was also per-
formed including sidewalls in the simulation, and using the Spalart-Allmaras
turbulence model. A three-dimensional view of the flow field is presented
in Figure (8.16), in terms of pressure iso-surfaces, and iso-contours of the
streamwise and spanwise velocity. The presence of side-walls is the cause of
secondary, recirculating motions, detectable from strong shock curvature and
from the spanwise velocity peak in the wall-parallel plane at y + = 18. In the
present case, the interaction with the sidewalls is rather weak, and confined to
a distance of ≈ 2δ0 from the walls. No significant acceleration of the flow ve-
locity in the center of the channel is detected, and the interaction is mainly
two-dimensional over a large span of the wind tunnel (≈ 10δ0 ), consistent
with the measurements performed by TUD.
The primary effect of three-dimensionality on the interaction in this case
is to shift downstream the whole interaction pattern, since the ‘true’ shock
impingement point is different from the nominal one, as predicted from the as-
sumption of straight inviscid shock. The distributions of skin friction and wall
pressure for the 2D and 3D (in mid-plane) RANS-HR, shown in Figure (8.17),
further confirm the idea that the main effect of three-dimensionality is a shift
in the origin of the interaction. Indeed, distribution obtained with 2D and
3D RANS very well collapse when reported in terms of the distance from
the ‘true’ point of shock impingement. The figure also show a minor effect of
three-dimensionality right past the shock impingement point, where the 3D
simulation exhibits a larger value of pressure, and a slightly different speed
of return to equilibrium.
Fig. 8.15 Comparison of turbulence models for RANS-HR test case: mean stream-
wise velocity u/u∞ (left) and mean wall-normal velocity v/u∞ (right)
Fig. 8.16 Three-dimensional RANS-HR computation including sidewalls with S-

A model. Iso-surfaces of pressure p/p0 = 1.2 (blue) and p/p0 = 1.65 (green) with:
(a) iso-plane at y + = 18 of streamwise velocity u (left); (b) iso-plane at y + = 18 of
spanwise velocity w (right).
Fig. 8.17 Comparison of 2D and 3D simulations of RANS-HR test case: distribu-

tion of skin friction (left) and wall pressure (right). Three-dimensional results are
also shown with a shift in the x coordinate of 0.65 δ0 to compensate the effect of
shock distortion.
8.4 Overview of LES Simulations

The results of LES computations are reported in the present Section. Contri-
butions have been provided by URMLS (see deliverable 5.3.3) and SOTON
(deliverable 5.3.13). LES has the power (by its own nature) to get reliable
insight into the unsteady, three-dimensional dynamics of shock/turbulence
interactions, and as such, it constitutes an ideal candidate for the purposes
of the UFAST project. Unfortunately, given the limitation of today’s com-
putational resources, it requires substantial computational efforts even for
8.4 Overview of LES Simulations 239
relatively simple flow geometries, such as the one considered in the present
flow case. Furthermore, the need to fully resolve the near-wall layer (where
the majority of turbulence production occurs) prevents LES from reproduc-
ing/predicting high-Reynolds number flows. As regards the present flow case,
the decision was taken to carry out LES simulations for the TUD test case at
a (substantially) reduced Reynolds number. Several preliminary studies were
presented both by SOTON and URMLS; at the end of this exploratory stage,
a decision was taken for a final case to be computed by the two partners using
the same computational grid and the same sub-grid-scale (SGS) model, to
isolate the effect of the numerical discretization and/or boundary conditions
on the quality of the results. In the following we briefly describe the numerical
methodology used by the investigators, and refer to the original deliverables
for a full account of the results.
LES model
The unresolved terms in the momentum and energy equations (see Equa-
tion 8.3) are to be interpreted in LES as the effect of the sub-grid scales of
motion onto the resolved ones, and
τij = ρ (u
i uj − u
i uj ), Qj = T
uj − Tuj .
A simple linear eddy-viscosity assumption is made to model τij and Qj in

the LES models that have been considered for the UFAST project
1 ∂ T
τij − δij τkk = −2 ρ νt Sij
∗
, Qj = −κt , (8.19)
3 ∂xi
where νt denotes the subgrid-scale viscosity, and the turbulent thermal con-
ductivity κt is determined as κt = cp ρ νt /P rt (P rt = 0.60).
After a series of preliminary studies by SOTON and URMLS, it was de-
cided to converge on the Mixed-Time-Scale sub-grid-scale model of Inagaki
et al. [6], which guarantees the correct asymptotic behavior to the eddy
viscosity at solid walls without using ad-hoc wall damping functions. The
subgrid-scale viscosity is determined from
CMTS
νt = −1
Δ kes , (8.20)
1 + (R CT )
√
where R = kes /(Δ S∗ ), and kes is the subgrid-scale turbulent kinetic energy,
u−u
kes = (
)2 . The hat symbol indicates the test filter, derived from the
trapezoidal rule [11], and Δ is the filter width, defined as Δ = (Δx Δz)1/2 .
As suggested by Touber and Sandham [20], the model constants are set to
CMTS = 0.03, CT = 10.
Numerical discretization and boundary conditions
The governing equations 8.3 are solved using with slightly different strategies
from the SOTON and URLMS groups, which are explained in the following.
SOTON uses a fourth-order central spatial differencing scheme for the
spatial derivatives and third-order explicit Runge-Kutta time stepping. The
SOTON code makes use of the entropy splitting of the Euler terms and
Laplacian formulation of the viscous terms to enhance the stability of the
non-dissipative central scheme [12]. In addition, a variant of the standard
total variation diminishing scheme is used for shock capturing [24], coupled
with the Ducros sensor [2].
The URMLS code exploits a finite-difference formulation that was devel-
oped in previous works both for isotropic decaying compressible turbulence
and for wall bounded turbulent supersonic flows [9, 10], whereby the con-
vective fluxes are discretized by means of a hybrid seventh-order WENO /
central scheme, with a switch based on the Ducros sensor. Viscous fluxes
(cast in Laplacian form) are approximated with second-order central differ-
ences, and time integration is performed by means of the classical four-stage,
fourth-order explicit Runge-Kutta algorithm. Both codes were made parallel
in all three coordinate directions using MPI message-passing libraries.
The computational domain used for the LES simulations, schematically
shown in Figure 8.18, was identical for SOTON and URMLS, and has an
overall size of Lx × Ly × Lz = 36δ0 × 7.4δ0 × 4.1δ0 where x, y and z stand,
Fig. 8.18 Sketch of computational domain for LES of TUD impinging shock flow
case
8.4 Overview of LES Simulations 241
respectively, for the streamwise, wall normal, and spanwise directions, and
δ0 is the boundary layer thickness at the inlet of the computational domain.
The domain is discretized with a grid consisting of 451 × 151 × 141 points,
with a grid resolution in wall units Δ+ x × Δyw × Δz ≈ 40 × 1.6 × 14, and 90
+ +
grid points are placed inside the boundary layer.

The no-slip condition is enforced at the bottom wall, which is set to be
isothermal. The top (freestream) and outflow boundaries make use of an
integrated characteristic scheme in order to minimize unwanted reflections
from the computational boundaries [19]. The oblique shock is introduced at
the top boundary using the Rankine-Hugoniot jump relations.
Large eddy simulation of turbulent flows requires the prescription of three-
dimensional, unsteady inflow boundary conditions to get fast transition to
a fully turbulent state. Such achievement critically depends upon the ‘correct’
enforcement of forcing disturbances at the inlet of the computational domain.
In the UFAST project different approaches have been pursued by SOTON and
URMLS, which are both based on the idea of superposing quasi-deterministic
perturbations onto a prescribed mean velocity profile (the same mean profile
was used for the two simulations).
SOTON’s approach is based on the so-called digital-filtering approach,
originally developed by Klein et al. [7], in the modified version of Touber and
Sandham [20]. All the details for the prescription of the inlet disturbances is
provided in the SOTON deliverable 5.3.13.
Table 8.3 Parameters for synthetic inlet forcing of URMLS-LES (as from Equa-
tion 8.22). δ0 and δv = νw /uτ are, respectively, the boundary layer thickness and
the viscous length scale at the inlet station.
j yj aj bj ωj ucj λz j φj
0 12.0δv 1.20 −0.25 0.12uτ /δv 10uτ 120δv 0.00
1 0.25δ0 0.32 −0.06 1.2u∞ /δ0 0.9u∞ Lz /3 5.01
2 0.35δ0 0.20 −0.05 0.6u∞ /δ0 0.9u∞ Lz /4 4.00
3 0.5δ0 0.08 −0.04 0.4u∞ /δ0 0.9u∞ Lz /5 3.70
4 0.6δ0 0.04 −0.03 0.2u∞ /δ0 0.9u∞ Lz /6 0.99
The synthetic inlet conditions approach proposed by Sandham et al. [13]

was used by URMLS. Time-dependent perturbations are introduced at the
inlet mimicking coherent boundary layer structures, and specified as follows

u (x, y, z, t) = ρ(y)
ρw
u∞ 4j=0 aj A1 j (y) Fj (x, t) Gj (z), (8.21)

v (x, y, z, t) = ρ(y)
ρw
u∞ 4j=0 bj A2 j (y) Fj (x, t) Gj (z), (8.22)
with
A1j (y) = (y/yj ) e−y/yj , (8.23)

2
A2j (y) = (y/yj )2 e−(y/yj ) , (8.24)

Fj (x, t) = sin ωj (x/uc j − t) , (8.25)

Gj (z) = cos 2πz/λz j + φj . (8.26)
The mode j = 0, is associated with inner layer streaks and streamwise vortices
with spanwise spacing λ+ +
z = 120, and length λx = 520, whereas the modes
j = 1, . . . , 4, correspond to large eddies scaling in outer units. The amplitudes
aj , bj , and the length- and time- scales of the various modes, selected so as
to provide fast transition to fully turbulent state with correct statistics, are
provided in Table 8.3. In order to suppress symmetries, divergence-free ran-
dom velocity fluctuations with maximum amplitude u /u∞ = 4% have been
also added within the boundary layer [8]. The spanwise velocity component
is finally determined assuming that the inlet velocity field is solenoidal.
Both approaches were found to provide transition to developed turbulence
within a distance of approximately 15δ0 from the inlet station. However, some
remnants of the inlet conditions are found even at large downstream locations
for the synthetic inlet approach, associated with undulations of the Reynolds
stress distributions in the outer part of the boundary layer, as will be shown
in the next Section.
An extensive description of the results of the final LES simulations of
SOTON and URMLS is provided in the next Section. For a full account of
the intermediate results and an exhaustive description of the computational
strategies, one is referred to Deliverables 5.3.3 and 5.3.13.
8.5 Comparison between Experiment and CFD

The results of numerical simulations of the TUD flow case are critically re-
view in the present Section, and compared to the TUD experimental data.
On the basis of the discussion developed in the previous Sections, however,
only selected data are reported. As far as LES is concerned, only the re-
sults from the last set of numerical simulations are presented, which were
performed on the same grid, and with the same subgrid-scale model. The
two LES simulations will be referred to as SOTON-LES and URMLS-LES,
respectively, and they mainly differ by the inlet conditions and the numerical
discretization. With regard to the (U)RANS simulations, the results pre-
sented in Section 8.3 clearly showed minor effects of unsteadiness and three-
dimensionality, as well as of the turbulence model. Therefore, only the results
from steady, two-dimensional RANS are shown for URMLS, one performed
at the same Reynolds number as the LES (labeled as URMLS-RANS-LR)
and one at a Reynolds number similar to the experimental conditions (la-
beled as URMLS-RANS-HR); whereas for LMFA the results of steady three-
dimensional RANS are shown (labeled here as LMFA-RANS); all data were
obtained using the Spalart-Allmaras modes. The main parameters for the set
of data presented in the following is provided in Table 8.4.
8.5 Comparison between Experiment and CFD 243
Table 8.4 Characteristic parameters for shock/boundary layer interaction experi-

ments and simulations (the subscript r denotes properties measured at the reference
station upstream of the shock)
TUD- SOTON- URMLS- LMFA- URMLS- URMLS-

-DUAL -LES -LES -RANS -RANS-LR -RANS-HR
Reθ r 50000 2678 2678 7229 2833 44954
Cf r 0.00149 0.00244 0.00255 0.00594 0.00259 0.00171
δi∗r /δr 0.116 0.145 0.162 0.107 0.148 0.108
Hir 1.25 1.41 1.39 1.29 1.40 1.26
L/δr 1.92 2.91 2.85 3.56 2.51 1.67
L/δi∗ r 16.6 20.1 17.6 33.4 17.0 15.4
Overall flow organization
A qualitative understanding of the flow organization can be gained from the

analysis of instantaneous slices of the flow field, as reported in Figure 8.19.
The instantaneous temperature fields in the x − y plane extracted from the
two LES simulations reveal the existence of complex organized motion in the
outer part of the boundary layer, with the occurrence of turbulent bulges
inclined at an acute angle with respect to the wall. As observed experimen-
tally both in subsonic and supersonic turbulent boundary layers [14] these
structures are separated from the surrounding essentially irrotational fluid
by sharp interfaces that have a three-dimensional character. For weakly com-
pressible flows, temperature can be interpreted as a passive scalar; hence,
Figure 8.19 confirms the enhancement of mixing generally observed in shock
wave / turbulent boundary layer interactions [14], and shows that the in-
terfaces separating the rotational and irrotational zones become sharper past
the interaction. The temperature distributions found in the SOTON-LES and
URMLS-LES, even though taken at different times, exhibit close similarities.
The flow pattern observed in RANS is much simpler (but qualitatively simi-
lar), as all turbulent fluctuations are filtered out, resulting in a steady flow.
Figure 8.20 depicts the instantaneous streamwise velocity distribution
taken from URMLS-LES in wall-parallel planes at various distances from
the wall (zones of instantaneous flow reversal are marked with solid lines).
The figure clearly shows the occurrence of elongated streaky patterns of al-
ternating high- and low-speed in the very near-wall region upstream of the
interaction zone, whose characteristic spacing in the spanwise direction is of
the order of 100 wall units, and whose length is of the order of 1000 wall
units. Inside the interaction zone, scattered spots of flow reversal are ob-
served, which are prevalently found upstream of the nominal impingement
point. Moving away from the wall, the structures become less elongated,
and more nearly isotropic, and the probability of instantaneous flow rever-
sal decreases. The analysis of the flow animations has further shown that
Fig. 8.19 Iso-contours of temperature (T /T∞ ) in x − y plane: (a) SOTON-LES;

(b) URMLS-LES; (c) URMLS-RANS-LR
excursions of the instantaneous separation line from a straight shape are

frequently associated with zones of high/low u upstream of the shock [4].
In Table 8.4 the characteristic parameters associated with the shock/
boundary layer interaction are shown for the TUD experiments and for the
numerical simulations. For the sake of comparison, not that all lengths are
made nondimensional with respect to the 99% boundary layer thickness right
upstream of the interaction, say δr , and the nominal shock impingement point
is assumed as the origin for the streamwise coordinate x. In all cases, δr was
conventionally measured at the streamwise location (x − x0 )/δr = −4 but
for the LMFA-RANS simulation, were the reference station was taken at
(x − x0 )/δr = −6. Several integral parameters are reported to characterize
Fig. 8.20 Iso-contours of streamwise velocity (u/u∞ ) for URMLS-LES in x − z

planes at: a) y + = 2; b) y + = 20; c) y + = 100 (black lines mark zone of negative u)
the state of the boundary layer upstream of the interaction, including the
skin friction coefficient Cf , the ‘incompressible’ displacement thickness δi∗ ,
and the ‘incompressible’ shape factor, Hi . The size of the interaction zone
is characterized in terms of the interaction lengthscale L, defined as the dis-
tance of the nominal shock impingement point from the apparent origin of
the reflected shock (see Figure 8.21).
The table reveals some interesting trends. First, the simulations performed
at a reduced Reynolds number (SOTON-LES, URMLS-LES, URMLS-RANS-
LR) exhibit comparable features, but, as expected, they predict boundary
layer properties at the reference station which are sensibly different from
those measured in the experiment. In particular, the shape factor is found
to be larger than in the TUD experiment, indicating a ‘less full’ velocity
profile in the inner part of the boundary layer; consistently, the skin friction
is substantially underestimated. On the other hand, simulations performed
at the ‘true’ Reynolds number (URMLS-RANS-HR, LMFA-RANS) seem to
Fig. 8.21 Definition of characteristic length-scale for shock/boundary layer

interaction.
predict the boundary layer structure with reasonable accuracy. However, dif-
ferences in the skin friction coefficient of the order of 15% are found for the
URMLS-RANS-HR simulation (which can be explained in the light of the
experimental uncertainties in the estimation of Cf ), whereas much larger
differences are found in the LMFA-RANS simulation. With regard to the
size of the interaction zone, all the low-Reynolds number simulations pre-
dict a substantially large extent, whereas the high-Reynolds number RANS
(URMLS-RANS-HR) yields a serious underprediction. A much larger extent
of the interaction zone is predicted by LMFA-RANS. Note that a much bet-
ter collapse of the interaction lengthscale is obtained by taking as a reference
the ‘incompressible’ displacement thickness upstream of the interaction, as
suggested by Delery and Marvin [1].
Statistical properties
The mean defect velocity distributions at the reference station are shown in
Figure 8.22, upon suitable density scaling. Very good agreement of the ve-
locity distributions are observed for SOTON-LES in the outer region, as well
as for the URMLS-RANS simulations. Some discrepancies are observed for
URMLS-LES, which predicts a stronger wake contribution in the outermost
part of the boundary layer. Very significant deviations are observed in the
LMFA-RANS simulation, presumably due to the observed wrong estimation
of the skin friction coefficient. The velocity distributions are also reported in
semi-log scale in Figure 8.23, where
u 1/2
ρ
uVD = du, (8.27)
0 ρw
is the effective (van Driest) velocity. The LES simulations exhibit a very
similar distribution in the inner part of the boundary layer, and a logarithmic
region with log-layer constant of about 6, which is similar to what previously
Fig. 8.22 Mean defect velocity across boundary layer upstream of the interaction.
Fig. 8.23 Mean van Driest-transformed velocity distribution upstream of the in-
teraction (in inner units).
found in LES of canonical boundary layers and channels. As also observed in

Figure 8.22, the URMLS-LES seems to under-predict the streamwise velocity
in the outer part of the boundary layer. The experimental data (which are
only reliable starting from y + ≈ 300) are well reproduced by URMLS-RANS-
HR, except for a shift in the log-law constant, which is found to be C ≈ 6.3
in the experiment.
Fig. 8.24 Distribution of Reynolds components across boundary layer upstream

of the interaction (in inner scale)
The components of the density-scaled Reynolds stresses at the reference

station are reported in Figure 8.24 in wall units. Note that for the LES
simulations the resolved stresses are reported, whereas for the RANS sim-
ulations, turbulent stresses are estimated from Boussinesq approximation.
The Reynolds stress distributions found from the two LES are very similar,
especially in the inner layer, and both agree quite well with the experimen-
tal data provided by TUD. Some discrepancies are observed in the outer
layer, where the URMLS-LES simulation exhibits an unphysical undulation
in the streamwise component, probably associated with the effect of the syn-
thetic inlet conditions, whereas the SOTON-LES simulation correctly shows
monotonic decrease. With regard to the RANS simulations, as expected the
normal stresses are not correctly predicted, whereas the distribution of the
tangential stress is qualitatively reproduced by both URMLS-RANS-LR and
URMLS-RANS-HR, with an overprediction of about 50% in the peak value.
The distribution of the skin friction coefficient and of the mean wall pres-
sure are shown in Figure 8.25. Note that the TUD dataset does not include
direct measurements of wall properties. Therefore, the skin friction coefficient
was extrapolated on the basis of the available velocity fields. An estimate of
the wall pressure from the PIV data was also made based on the approach to
integrate the momentum equations introduced by van Oudheusden [21]. As
expected, the friction coefficient upstream of the interaction is only correctly
predicted by URMLS-RANS-HR, which also predicts the correct trend past
the interaction zone. However, the upstream influence is much shorter than
in the TUD experiment. Note that both the SOTON- and the URMLS-LES
predict a small extent of separation, which is not observed in the experi-
ment, and exhibit similar values of Cf both upstream and downstream of
the interaction, even though the interaction extent is significantly smaller
in the URMLS simulation. As far as the issue of mean flow separation is
concerned, we point out that in the LES the height of the mean separation
bubble amounts to very few wall units, which would imply that mean flow
separation (if any) would not be detected in the TUD experiment due to lack
of sufficient resolution near the wall. The wall pressure distributions shows
significant similarities between experiment and the two LES, especially in the
recovery zone, whereas the upstream influence mechanisms is greatly over-
estimated. The occurrence of an inflection point in the experimental pressure
distribution is likely to be due to problems in integrating the pressure gradi-
ent equation across the impinging shock foot.
The distributions of the boundary layer thicknesses and shape factor across
the interaction zone are show in Figure 8.26. The LES simulations reproduce
the same trend found in the experiments, but the size of the interaction zone
is very different, as well as the shape of the boundary layer. On the other
hand, URMLS-RANS-HR very well reproduces the growth and the change of
shape of the boundary layer across the interaction region.
The distributions of the mean velocity components, scaled by the ref-
erence friction velocity and by the reference density profile, are shown in
Fig. 8.25 Distribution of average skin friction coefficient (a) and mean wall
pressure (b)
Figures 8.27–8.28. The figures indicate very good qualitative correspondence

of all low-Reynolds-number simulations (both LES and RANS) with the
experimental data, and in particular their capability to reasonably repro-
duce the maximum vertical velocity inside the interaction zone (the largest
values found in the experiments may be due to problems with the PIV tech-
nique in the vicinity of the shock foot). However, note that the low-Reynolds
number simulations predict a very gradual growth of the boundary layer even
upstream of the reflected shock foot, and the latter is found to diffract and
spread significantly above the boundary layer. In the TUD experimental data
the boundary layer is seen to be undisturbed up to (x − x0 )/δr ≈ −2, and the
Fig. 8.26 Distribution of boundary layer thicknesses: δ99 (a); δi∗ (b); and Hi (c)
Fig. 8.27 Distribution of mean defect velocity (ρr /ρw (u − u∞ )/uτ r ) in the in-
teraction zone. (a) TUD; (b) SOTON; (c) URMLS-LES; (d) URMLS-RANS-LR;
(e) URMLS-RANS-HR; (f) LMFA-RANS.
reflected shock is found to be relatively straight. Such interaction pattern is

very similar to the one found in URMLS-RANS-HR, even though the overall
scale of the interaction there is much smaller.
The distributions of the Reynolds stress components in reference units
are shown in Figures 8.29–8.31. The longitudinal Reynolds stress component
shows the occurrence of a maximum near the foot of the reflected shock, and
a ridge of intense turbulent activity extending well downstream of the nom-
inal impingement point. Such pattern is consistent with the observations of
Pirozzoli and Grasso [10], who pointed out that the generation of turbulent
stress is strictly related to the unsteady shedding of coherent vortical struc-
tures associated with the occurrence of inflection points in the instantaneous
velocity profiles. Both LES simulations well capture the phenomenon quali-
tatively; however, the Reynolds stress peak lies somewhat farther away from
the wall, and its strength is underestimated by approximately 25%. Simi-
lar observations can be made with regard to the other two Reynolds stress
Fig. 8.28 Distribution of mean wall-normal velocity (ρr /ρw v/uτ r ) in the inter-
action zone. (a) TUD; (b) SOTON; (c) URMLS-LES; (d) URMLS-RANS-LR;
(e) URMLS-RANS-HR; (f) LMFA-RANS.
components. However, the wall-normal stress peaks right past the end of the
interaction zone, while the shear stress has two distinct peaks (the first one
being absent in the experimental data). It is interesting to observe that the
correct pattern of the Reynolds shear stress is recovered in the RANS simu-
lations, which also predict a turbulence level past the interaction similar to
the experiment.
Unsteady features
The analysis of the unsteady properties of the flow field is useful in that it
can provide useful insight for the prediction of the fluctuating loads occurring
in the interaction zone. LES is a natural candidate to get such information.
In first instance we observe that, although very little mean flow reversal is
observed, the adverse pressure gradient is sufficiently strong to locally cause
scattered spots of reversed flow in the interaction zone. The distribution of
1/2
Fig. 8.29 Distribution of longitudinal Reynolds stress ( ρr /ρw u u /uτ r ) in
the interaction zone. (a) TUD; (b) SOTON; (c) URMLS-LES; (d) URMLS-RANS-
LR; (e) URMLS-RANS-HR.
the intermittency of local flow reversal, defined as the statistical frequency

of points where u < 0, is shown in Figure 8.32. The figure indicates that
zones of flow reversal are found with non-zero probability in the range −3 ≤
(x−x0 )/δr ≤ 0, with a maximum of approximately 60% inside the separation
bubble. The shape of the iso-probability confirms that both LES overestimate
the extent of the interaction zone. However, the SOTON-LES seems to be
more effective in predicting the occurrence of points of flow reversal even at
significant distance from the wall up to y/δr ≈ 0.15.
The unsteadiness in the interaction zone can be usefully interpreted revert-
ing to the analysis of the pressure time histories at the wall. The distributions
of the weighted power spectral densities (PSD) at selected point at the bot-
tom wall are shown in Figure 8.33 (note that only the LES data are available).
The spectra upstream of the interaction show the typical behavior found in
canonical boundary layers, with energy concentrated around St ≈ 1, which is
1/2
Fig. 8.30 Distribution of wall-normal Reynolds stress ( ρr /ρw v v /uτ r ) in the
interaction zone. (a) TUD; (b) SOTON; (c) URMLS-LES; (d) URMLS-RANS-LR;
(e) URMLS-RANS-HR.
the typical frequency for large eddies of size of the order of the boundary layer
thickness. The energy spectra across the interaction zone exhibit significant
strengthening of the lowest-frequency modes, which can be an indication of
the occurrence of low-frequency motions associated with the appearance of
global unstable modes [20]. In addition, the spectral peak is found to move to
lower frequencies, which may be due either to the thickening of the boundary
layer, or to the onset of frequencies associated with the shedding of vorti-
cal structures in the mixing layer. Good agreement is observed between the
SOTON-LES and the URMLS-LES simulations, which is indicative of similar
dynamics. However, note that the SOTON simulation has been advanced for
a much longer time, thus giving a more reliable insight into low-frequency
processes.
Fig. 8.31 Distribution of Reynolds shear stress ((ρr /ρw ) u v /uτ 2r ) in the inter-
action zone. (a) TUD; (b) SOTON; (c) URMLS-LES; (d) URMLS-RANS-LR;
(e) URMLS-RANS-HR.
Fig. 8.32 Probability of flow reversal in interaction zone (percent scale is used).
(a) TUD-ZOOM; (b) SOTON-LES; (c) URMLS-LES.
Fig. 8.33 Pre-multiplied pressure spectra across the interaction zone at (x −

x0 )/δr = −3.9 (a); −2.4 (b); 0.16 (c); 3.96 (d). Solid lines: SOTON-LES; sym-
bols: URMLS-LES.
8.6 Conclusions 259
8.6 Conclusions
The TUD impinging shock flow case has received a considerable number of
contributions. The experimental data from TUD include a complete set of
PIV measurements of the velocity field across the interaction zone, covering
the mean field and the Reynolds stress components. Genuinely unsteady data
include the distribution of the probability of flow reversal, and the temporal
self-correlation of the streamwise velocity field. Unfortunately, no detailed
measurements of the wall properties, nor of the pressure spectra, are available.
The TUD flow case was one of the most challenging among impinging
shock/boundary layer interactions considered in the UFAST project. Even
though no substantial mean separation is observed, the Reynolds number is
large enough to be unattainable from LES, which unfortunately precludes
definite quantitative statements with regard to the predictive capabilities of
the latter.
The results of (U)RANS simulations have clarified some important issues:
i) the effect of three-dimensionality for the present flow case is minor, con-
firming the experimental findings; ii) unsteadiness in RANS computations
could only be obtained upon strong stimulation of the upstream boundary
layer, thus leading to conclude that no spontaneous large scale unsteadi-
ness develops in the absence of massive flow separation; iii) the influence of
the turbulence model (at least for the models considered in the study) on
the gross features of the interaction (e.g. the size of the interaction zone) is
minor.
The LES simulations were carried out by two partners, which has allowed
a cross-check of the results. The results obtained from those separate studies
are encouragingly quite similar. As common to all LES of wall-bounded flows,
the velocity field in the incoming boundary layer was found to exhibit a log-
layer, with a constant which is however somewhat larger that the commonly
accepted value. The structure of the upstream turbulence field (in terms of
the Reynolds stress components) conforms remarkably well with canonical
boundary layers, as well as with the TUD experimental data.
The results related to the extent of the interaction zone are rather puzzling.
Indeed, whereas LES and RANS performed at a reduced Reynolds number
consistently predict an interaction size that is comparable (but larger) than
found in the experiment, RANS performed at the full-scale Reynolds number
predicts a smaller size of the interaction zone, even though the boundary
layer properties upstream and downstream of the impinging shock are rather
well predicted. Assuming that the LES can be used as a validation tool for
RANS, and assuming that the RANS predictive capabilities are maintained
even at large Reynolds numbers, the reason for the disparity in the size of
the interaction zone with respect to the experiments is not clear, and also
the predictive capabilities of LES cannot be straightforwardly extrapolated
to large Reynolds numbers, thus remaining an open issue.
260 References
With regard to the structure of the interaction zone, both experimental

and numerical data consistently indicate an amplification of the Reynolds
stress components occurring near the apparent origin of the reflected shock.
Such amplification is associated with the formation of a mixing layer that
develops nearly parallel to the wall past the interaction zone, in which large
vortical structures are embedded. Such features are clearly observed in both
LES, and are also predicted with good accuracy by RANS.
A comparison of the unsteady features of the interaction in terms of the
probability of local flow reversal shows that, although no substantial flow
separation is observed, spots of instantaneously reversed flow are frequently
found throughout the interaction zone. The comparison of the LES maps with
the experimental data shows reasonable qualitative agreement, especially for
the SOTON-LES simulation. The wall pressure spectra show a good deal
of agreement between the two LES simulations, which seem to indicate the
occurrence of low-frequency dynamics inside the interaction zone.
The overall conclusion of the study is that, while LES has the power to
get insight into the physics of the interaction, and it certainly produces plau-
sible dynamics for this flow case, its application to non-trivial high-Reynolds
number flows is far from today’s computational capabilities, and on the basis
of the present data no conclusive statements can be made regarding its supe-
riority with respect to RANS. The application of wall models, which was not
explored in this study, would make full-scale computation of the flow case
feasible, at the price of introducing additional modeling uncertainties. This
might be the subject for further investigation.
With regard to RANS, the good news come from the good agreement with
the LES calculations at reduced Reynolds number. RANS (even a very simple
model such as Spalart-Allmaras) seems to predict the gross features of the
interaction quite precisely, and it is capable to capture the formation of the
mixing layer. On the other hand, RANS performed at the full-scale Reynolds
number shows substantial difference with respect to the TUD experiment
as far as the interaction size is concerned, whereas it reproduces quite well
the boundary layer structure both upstream and downstream of the interac-
tion zone. The reasons for the observed discrepancies would probably require
additional investigation, since they do not seem to be related to either the
turbulence model, to the mesh resolution, or to three-dimensional effects.
References
[1] Delery, J., Marvin, J.G.: Shock-wave boundary layer interactions. AGARDo-
graph 280 (1986)
[2] Ducros, F., Ferrand, V., Nicoud, F., Darracq, D., Gacherieu, C., Poinsot,
T.: Large-eddy simulation of the shock/turbulence interaction. J. Comput.
Phys. 152, 517–549 (1999)
[3] Fares, E., Schröder, W.: A general one-equation turbulence model for free shear
and wall-bounded flows. Flow, Turbulence and Combustion 73, 187–215 (2004)
References 261
[4] Ganapathisubramani, B., Clemens, N.T., Dolling, D.S.: Effects of upstream

boundary layer on the unsteadiness of shock-induced separation. J. Fluid
Mech. 585, 369–394 (2007)
[5] Hanjalic, K.: Will RANS survive LES? a view of perspectives. J. Fluids
Eng. 127, 831–839 (2005)
[6] Inagaki, M., Kondoh, T., Nagano, Y.: A mixed-time-scale sgs model with fixed
model-parameters for practical LES. J. Fluids Eng. 127(1), 1–13 (2005)
[7] Klein, M., Sadiki, A., Janicka, J.: A digital filter based generation of inflow
data for spatially developing direct numerical or large eddy simulations. J.
Comput. Phys. 186, 652–665 (2003)
[8] Li, Q., Coleman, G.N.: DNS of an oblique shock wave impinging upon a turbu-
lent boundary layer. In: Geurts, Friedrich, Metais (eds.) Direct and Large-Eddy
Simulations. Ercoftac Series, pp. 387–396. Kluwer, Dordrecht (2003)
[9] Pirozzoli, S., Grasso, F.: Direct Numerical Simulation of isotropic compressible
turbulence: Influence of compressibility on dynamics and structures. Phys.
Fluids 16(12), 4386–4407 (2004)
[10] Pirozzoli, S., Grasso, F.: Direct Numerical Simulation of impinging shock-
wave/turbulent boundary layer interaction at M = 2.25. Phys. Fluids 18(6),
1–17 (2006)
[11] Sagaut, P.: Large-Eddy Simulation for Incompressible Flows: An Introduction
to Large-Eddy Simulations. Springer, Heidelberg (2001)
[12] Sandham, N.D., Li, Q., Yee, H.C.: Entropy slitting for high-order numerical
simulation of compressible turbulence. J. Comput. Phys. 178, 307–322 (2002)
[13] Sandham, N.D., Yao, Y.F., Lawal, A.A.: Large-Eddy Simulation of transonic
turbulent flow over a bump. Int. J. Heat and Fluid Flow 24(4), 584–595 (2003)
[14] Smits, A.J., Dussauge, J.P.: Turbulent Shear Layers in Supersonic Flow. Amer-
ican Institute of Physics, New York (2006)
[15] Souverein, L., Dupont, P., Debiéve, J., Dussauge, J.P., van Oudheusden, B.W.,
Scarano, F.: Unsteadiness characterization in a shock wave turbulent boundary
layer interaction through dual-PIV. AIAA Paper 2008-4169 (2008a)
[16] Souverein, L.J., van Oudheusden, B.W., Scarano, F., Dupont, P.: Unsteadiness
characterization in a shock wave turbulent boundary layer interaction through
dual-PIV. AIAA Paper 2008-4169 (2008b)
[17] Souverein, L.J., van Oudheusden, B.W., Scarano, F., Dupont, P.: Applica-
tion of a dual-plane particle image velocimetry (dual-PIV) technique for the
unsteadiness characterization of a shock wave turbulent boundary layer inter-
action. Meas. Sci. Technol. 20, 074003.1–074003.16 (2009)
[18] Spalart, P., Allmaras, S.: A one-equation turbulence model for aerodynamic
flows. La Recherche Aérospatiale 1, 5–21 (1994)
[19] Thompson, K.W.: Time dependent boundary conditions for hyperbolic sys-
tems. J. Comput. Phys. 68, 1–24 (1987)
[20] Touber, E., Sandham, N.D.: Large-Eddy Simulation of low-frequency unsteadi-
ness in a turbulent shock-induced separation bubble. Theoretical and Compu-
tational Fluid Dynamics (2009) (accepted for publication)
[21] van Oudheusden, B.W.: Principles and application of velocimetry-based planar
pressure imaging in compressible flows with shocks. Exp. Fluids 45, 657–674
(2008)
262 References
[22] Wilcox, D.: Reassessment of the scale-determining equation for advanced tur-
bulence models. AIAA J. 11, 1299–1310 (1988)
[23] Wilcox, D.: Turbulence modeling for CFD. DCW Industries (1998)
[24] Yee, H.C., Sandham, N.D., Djomehri, M.J.: Low-dissipative high-order shock-
capturing methods using characteristic-based filters. J. Comput. Phys. 150,
199–238 (1999)
Chapter 9
(Neil Sandham)
9.1 Acronyms
CFD Computational Fluid Dynamics
CTA Constant Temperature Anemometry
DES Detached-Eddy Simulation
DNS Direct Numerical Simulation
HWA Hot Wire Anemometry
ITAM Institute of Theoretical & Applied Mechanics
IUSTI Institut Universitaire des Systèmes
Thermiques Industriels
LES Large-Eddy Simulation
LMFA Laboratoire de Mécanique des Fluides
et d’Acoustique
PDF Probability Density Function
PSD Power Spectral Density
RANS Reynolds-Averaged Navier–Stokes
RMS Root Mean Square
SBLI Shock-Wave/Boundary-Layer Interaction
SOTON University of SOuthampTON
UAN A. N. Podgorny Institute for Mechanical
Engineering
URANS Unsteady Reynolds-Averaged Navier–Stokes
springerlink.com
264 9 Oblique Shock Reflection at M = 2.0 (Neil Sandham)
9.2 Nomenclature
Roman letters
c local speed of sound

f dimensional frequency
kx stream-wise wavenumber
L interaction length, X0 − Ximp
Lsep separation length, Xat − Xsep
M Mach number
p pressure
Re Reynolds number
T total temperature
Tu turbulence intensity used at the RANS inlet
U time-averaged stream-wise velocity
uc convection velocity

uτ friction velocity, (μw /ρw )[∂u/∂y]w
x, y, z stream-wise, wall-normal, span-wise direction
Xat mean boundary-layer-reattachement location
Ximp incident-shock impinging point
Xsep mean boundary-layer-separation location
X0 location of the extension of the reflected shock to
the wall
Greek letters
δ99 99% boundary-layer thickness

sw
δ99 wind-tunnel side-wall 99% boundary-layer
thickness
∞
δ1 displacement thickness, 0 1 − ρρu t
∞ U∞
dy
imp
δ99 boundary-layer 99% thickness at inviscid-
impingement location in the absence of the shock
μ dynamic viscosity
ρ fluid density
σ standard deviation
τ time lag in the correlation functions

∞
θ momentum thickness, 0 ρρu t
∞ U∞
1 − ut
U∞ dy
9.4 Description of the Flow Case 265
Operators, Subscripts and Superscripts
aα α-averaged value of a a

avd , avd van-Driest transformed field, 0 ρt /ρw t da
w quantity at the wall

0 stagnation quantity
∞ quantity in the upstream potential flow
+
denotes that the variable is expressed in
wall units, y + = yuτ /νw , u+ = ut /uτ
9.3 Introduction
This report focuses on the ITAM shock-reflection flow case of the UFAST
project. The collaboration involved the experimental investigations of ITAM
and the numerical simulations of UAN (using 2D/3D (U)RANS), LMFA (us-
ing 3D (U)RANS) and SOTON (using LES). Each of the four partners have
completed their individual final reports and the reader is encouraged to read
them for further details on particular aspects not covered in the following
pages. While the partner reports are more focused on each individual work
and technique, this document is devoted to a more systematic evaluation of
the results.
In the first and second sections, the flow configuration and methods will be
reviewed including a brief discussion about the changes made from the origi-
nal to the final configuration. The third section is devoted to steady aspects
while the fourth section will cover some aspects of the flow unsteadiness.
The flow-control part of the project is restricted to the experiments and is
described in the ITAM final report. Finally, the main results from this flow
case are summarised in the conclusion section.
9.4 Description of the Flow Case

The ITAM flow case corresponds to the interaction between an oblique shock
wave and a turbulent boundary layer (SBLI). The measurements are carried
out in the ITAM T-325 wind tunnel (see figures 9.1 and 9.2). The imping-
ing shock is generated by a 7- or 8-degree wedge (see figures 9.3 and 9.4).
The incoming turbulent boundary layer is at Mach 2 with a Reynolds num-
ber in the order of 104 based on the upstream freestream velocity U∞ , the
boundary-layer edge dynamic viscosity μ∞ and the boundary-layer displace-
ment thickness upstream of interaction δ1 . The stagnation temperature is
about 288 K.
Fig. 9.1 Picture of the ITAM T-325 wind tunnel
Fig. 9.2 Picture of the T-325 test section
One interesting aspect of the ITAM experimental setup is the use of a plate
model to generate the boundary layer. This is of interest for numerical
simulations since, in principle, a low-Reynolds-number experiment is feasi-
ble, making a DNS or resolved LES approach conceivable. This was in fact
the original motivation for SOTON. In practice, the model-plate boundary-
layer required the use of an artificial transition to achieve turbulent flow,
which caused several experimental difficulties. Furthermore, the use of a rel-
atively thin boundary layer raised a number of technical problems due to the
Fig. 9.3 CAD view of the experimental setup
Fig. 9.4 Picture of the experimental setup
resulting need to make the measurements at higher frequencies than usually

encountered in such setups. Therefore, it was decided at the end of the 18-
month period to extend the plate model to increase the Reynolds number by
about 30%. Its final value is in the order of 104 (based on the displacement
thickness), which is about half of the value achieved in the similar IUSTI flow
case.
Early investigations performed on the extended-plate and 8◦ configuration

showed large discrepancies between the CFD and experimental interaction
lengths (see SOTON’s deliverable 5.3.2). This issue was believed to be re-
lated to strong 3D mean effects not captured by the simulations (see UAN
deliverable). It was therefore decided to reduce the wedge angle to 7◦ , hoping
to resolve this issue. Note that this final configuration was defined only at
the end of the 30-month period. Because the extended-model and 7◦ -wedge
angle configuration is the only one to be completed by all CFD partners,
it is the one considered in this report. A summary of the main settings is
provided in table 9.1 and a draft of the experimental geometry is reproduced
in figure 9.5. As shown in figure 9.5, the origin of the axis system is taken to
be the leading edge of the plate (and the wind-tunnel median plane).
Table 9.1 Summary of the 7◦ configuration
M wedge Reδ1 (a) P0 (bar) T0 (K) uτ (m s−1 )(b)

2.0 7◦ 1.14 × 104 0.8 288 23.8
(a)
with δ1 = 1.08 mm at x = 260 mm (δ99 = 4.30 mm and θ = 0.33 mm)
(b)
at x = 260 mm
Fig. 9.5 Draft of the 7◦ experimental setup
Figure 9.6 shows a schlieren picture of the interaction region. The main
shock system is easily seen. The multiple shock lines are due to the spanwise-
integration effect of the schlieren approach, providing clear evidence of the 3D
nature of the flow. The multiple shock lines can originate from the interaction
with the wind-tunnel side walls, possible spanwise corrugations and unsteady
motion of the shock. Figure 9.7 is an oil-flow visualisation on the flat-plate
model, where a quasi-2D separation region with a straight separation line
and curved reattachment line seems to appear. One could argue that the two-
dimensional section of the separation spans about 70% of the model width.
Fig. 9.6 Schlieren image of the 7◦ case
Fig. 9.7 Oil flow visualisation of the 7◦ case
However, the oil-flow visualisation suggests that the flow is highly 3D near
the corner formed by the plate and the wind-tunnel side walls. Nevertheless,
it can be argued that the interaction is statistically two-dimensional near the
wind-tunnel mid-plane.
9.5 Brief Summary of Methods Used

Below is a short summary of the methods used by each partner.
ITAM
The experimental investigations were based primarily on a in-house

constant-temperature-anemometry technique. Details about these hot-wire
measurements can be found in ITAM’s final report and references therein. In
addition, mean pressure measurements and oil-flow visualisations on the plate
were performed. Finally, high-speed schlieren visualisations were completed.
UAN
The 2D/3D (U)RANS simulations of UAN were performed using the group’s
in-house solver FlowER. The 3D compressible Reynolds-averaged Navier–
Stokes equations are solved using an implicit time-integration scheme com-
bined with a second-order accurate finite volume approach for the spatial
discretisation. The code uses an exact Riemann-problem solver. The turbu-
lence model chosen for this study is the k–ω SST model. More details on the
ENO approach and limiters used in the code can be found in UAN’s final
report.
LMFA
The 3D (U)RANS computations from LMFA were obtained using the

elsA code (developed at ONERA). The 3D compressible Reynolds-averaged
Navier–Stokes equations are solved using an implicit first-order time integra-
tion scheme combined with a second-order accurate finite volume approach
for the spatial discretisation. The code uses the Roe approximate Riemann
solver. The turbulence model chosen for this study is the Spalart–Almaras
model. More details on the settings of elsA can be found in LMFA’s final
report and from ONERA’s website.
SOTON
The contribution of SOTON consists in running a LES of this test case.

The filtered 3D compressible Navier–Stokes equations are solved using a 4th -
order central spatial differencing scheme for the spatial derivatives and the
3rd -order explicit Runge–Kutta scheme to integrate in time. The boundary
treatment is also of 4th order. The code makes use of the entropy splitting of
9.6 Steady Aspects 271
the Euler terms and the laplacian formulation of the viscous terms to enhance
the stability of the non-dissipative central scheme. In addition, a variant of
the standard total variation diminishing scheme is used for shock capturing,
coupled with the Ducros sensor. The sub-grid scale model used is a modified
Mixed-Time-Scales model. More details can be found in SOTON’s final and
progress reports (including the modified digital-filter approach to specify the
inflow conditions).
9.6 Steady Aspects

This section presents the main comparisons between the experimental and
numerical time-averaged results. Figures 9.8 and 9.9 compare the incoming
boundary-layer properties. Despite a clear improvement over the first inflow-
velocity profile comparisons (see SOTON’s 18-month progress report), there
remain substantial differences between the partners reference profiles, partic-
ularly near the wall. One reason could be the relatively low Reynolds number,
at which the RANS models may not perform well. Additionally, the relatively
thin boundary layer makes the measurements challenging because of the in-
herent high frequencies that must be resolved. Finally, the LES suffers from
the need to prescribe artificial time-varying and three-dimensional inflow con-
ditions. The digital-filter technique used here leads to a long inflow transient
(extending up to 20δ99 in the streamwise direction), making the prescription
of both the correct skin friction and displacement thickness at x = 260 mm
nearly impossible. Nevertheless, it is accepted that all the incoming boundary
layers presented in figure 9.8 are turbulent and share similar properties.
Figure 9.9 compares the streamwise root-mean-squared (RMS) velocity
fluctuations from the hot-wire measurements (using the technique described
in ITAM’s final report) and the LES. The RMS levels are scaled according
to Morkovin’s scaling and compared with the incompressible DNS data of
Spalart at Reθ = 1410 [8]. The disagreement between the LES and HWA
data has been discussed at length in the progress reports. In summary, the
lack of fluctuations in the HWA data is due to poor performances of the
constant-temperature-anemometry (CTA) system at high frequencies, there-
fore cutting off the near-wall high-frequency turbulent fluctuations. ITAM
has corrected the high-frequency response of the CTA measurements using
a transfer function (see ITAM’s final report for details). The data presented
in figure 9.9 include the correction. It can be seen that the fluctuations in the
inner region are still underestimated. However, the outer-layer region is close
to the DNS data and the correction seems to apply there. The LES shows
the opposite trend. The inner layer is relatively near the DNS data but the
outer region is overestimated. The higher LES levels can be explained by the
position of the reference plane, which is too close from the inflow plane. The
boundary layer at the reference station is still recovering from the inlet condi-
tions. It is shown in [9] that the outer-layer region recovers more slowly than
the inner layer. In fact, if the reference plane had been further downstream
Fig. 9.8 Upstream velocity profiles (at x = 260 mm) in semilogarithmic (left hand)
and linear (right hand) scales
Fig. 9.9 Upstream velocity fluctuations (at x = 260 mm)

from the inlet plane, the outer profile would have a shape closer to the DNS.
Finally, the
LES inflow plane produces a weak shock and is fairly noisy, with
values of pw pw t /τw ≈ 5, artificially increasing the RMS levels.
Figure 9.10 compares the wall-pressure evolution along the interaction be-
tween the experiment, the LES, the 2D and 3D RANS calculations of UAN
and the 3D RANS calculations of LMFA. Despite the use of a reduced wedge
angle, the interaction length in the experiment is larger than in the simula-
tions. There are several interesting observations to make here.
Fig. 9.10 Wall-pressure distribution
First, the interaction length is seen to correlate with the back-pressure

level. In this respect, the experiment overshoots the theoretical back-pressure
value which could force the separation to move upstream. The reason why the
back pressure overshoots the theoretical value is not fully understood, but the
3D RANS investigations of UAN offer some useful insight. Figure 9.11 gives
the mid-section wall-pressure sensitivity to different side-wall-boundary-layer
thicknesses, as found in the 3D RANS investigations of UAN. It is seen that
as the side-wall boundary layer is made thicker, the pressure overshoot is en-
hanced and the start of the pressure rise moves upstream (i.e. the interaction
length increases).
The aforementioned influence of the side-wall boundary layer on the size of
the interaction is believed to be primarily due to the corner-flow recirculation
formed at the wall/plate junction under the influence of the incident shock.
This was mentioned in figure 9.7 and further described by UAN in their
final report. Figure 9.12 shows the streamlines near the plate and near the
Fig. 9.11 Effect of the side-wall boundary-layer thickness on the mid-section pres-
sure evolution from UAN’s RANS computations. Line 1 corresponds to T u = 0.1%,
sw sw
δ99 = 10 mm, line 2 to T u = 0.1%, δ99 = 1 mm, line 3 to T u = 0.1%, 2D RANS
sw
and line 4 to T u = 5%, δ99 = 10 mm. The dots correspond to the wall-pressure
measurements of ITAM.
sidewalls. It is readily seen that the corner flow has a greater influence as the
ratio of the sidewall to the plate boundary-layer-thickness is increased.
In addition, it should be noted that the choice of incoming turbulence
intensity in the RANS calculation has a significant impact on the predicted
interaction region (see figures 9.11 and 9.12). Although UAN and LMFA did
not use the same turbulence models (two-equation k–ω SST and Spalart–
Almaras models, respectively), the smaller interaction found in the 3D RANS
of LMFA, compared to the 3D RANS results of UAN could be due to the
large difference in the incoming turbulence intensity, where LMFA used 1%
and UAN 0.1%. From figures 9.11 and 9.12 it can be seen that when UAN
used 5%, the interaction length was considerably smaller.
One other aspect of the wall-pressure distribution is the apparent pressure
decrease near the end of the graph in figure 9.10, a feature not captured by
the LES. This is believed to be due to the expansion fan coming from the
wedge trailing edge, which is included in the RANS but not in the LES (see
figures 9.13 and 9.14). It should be noted that the incident-shock impinge-
ment point found in the experiment does not exactly match the theoretical
values. As discussed with the ITAM group, the experimental incident-shock
impingement point is about 2.2 mm earlier than the expected location. The
difference comes from the fact that the wedge leading edge is not perfectly
sharp. The plots presented in this report already account for this shift.
The separation and interaction lengths are summarised in table 9.2. The
strong influence of both the inflow turbulence intensity and the side-wall
boundary layers on the SBLI region can be quantified. The most sensitive
parameter appears to be the choice of inflow turbulence intensity: the greater
the RANS inflow turbulence intensity, the shorter the interaction. Similarly,
Fig. 9.12 Limiting streamlines at the sidewall (left) and at the plate (right) for
different side-wall to plate boundary-layer-thickness ratios and different turbulence
intensities
Fig. 9.13 LES pressure field vs. schlieren picture
Fig. 9.14 Pressure contours at the mid section of the UAN 3D RANS, with
sw
T u = 0.1% and δ99 = 10 mm
Table 9.2 Summary of the interaction and separation lengths for the 7◦
configuration
Partner ITAM UAN LMFA SOTON

Method Experim. 2D RANS 3D RANS 3D RANS LES
sw
δ99 (mm) O(10) – 1 5 10 10 10 10 3 –
T u (%) O(0.1) 0.1 0.1 0.1 0.1 1.0 2.0 3.0 1.0 <1
Lsep(a) (mm) ?? 16.0 17.9 19.8 20.7 19.8 16.0 8.5 4.1 9.3
L(b) (mm) 23 ?? ?? ?? ?? ?? ?? ?? ?? 13.5
(a)
separation length, distance from the mean separation point to the mean
reattachment point
(b)
interaction length, distance from the extrapolation of the reflected shock to the
wall to the inviscid-shock impingement point
9.7 Unsteady Aspects 277
the thinner the side-wall boundary layer, the shorter the interaction. However,
it should be recalled that RANS models are not expected to perform well
in such a complex flow configuration and the results reported here must
be considered with care. Nevertheless, they offer an indication about the
sensitivity of the RANS model (two-equation k–ω SST for that matter) to
the inflow conditions.
9.7 Unsteady Aspects
This section is devoted to some results concerning the flow unsteadiness. Since
the URANS computations did not show any sign of unsteadiness, the com-
parison is limited to the HWA measurements and the LES results. First of all,
it must be emphasised that the low-frequency study is still very challenging
for LES. This is due to the fact that the most energetic low frequencies are
found to be two orders of magnitude smaller than the characteristic frequency
of the incoming boundary layer (U∞ /δ99 ). In addition, the low frequencies
are broadband in nature and it is necessary to cover several cycles (ideally
more than 50) to achieve a proper convergence of the spectral analysis. Con-
sequently, in order to resolve both the turbulence and the broadband low-
frequency motions, the LES must span times of the order of 104 δ99 /U∞ with
a time resolution of about 10−3 δ99 /U∞ , leading to an impressive frequency
range of 7 decades.
Fig. 9.15 Correlation function of the momentum fluctuations at a station crossed

by the reflected shock
Fig. 9.16 PSD of the momentum fluctuations at a station crossed by the reflected
shock
Fig. 9.17 Probability of reversed flow (in percent). The white line indicates the
mean zero-streamwise-velocity contour and the dash-dotted line corresponds to the
interpolation to the wall of the mean reflected-shock location.
In practise, the present LES signal covers about 20 ms (at a rate of 47 MHz)
while the HWA measurements span about 350 ms (at a rate of 0.75 MHz).
Thus, the LES signal is about 17 times shorter than its experimental coun-
terpart and can only cover about 10 periods at 500 Hz. Therefore, the low-
frequency spectral analysis of the LES is not fully converged and comparisons
with the HWA results should be undertaken with care. Nevertheless, the data
is sufficient to provide reasonable trends.
Figure 9.15 gives the auto-correlation functions computed from the mo-
mentum time series obtained by a hot wire probe and the numerical
Fig. 9.18 Wall-pressure auto-correlation function as a function of streamwise lo-

cation. All spanwise computational nodes were included so that the correlation
function is also averaged in the spanwise direction. The signals were low-pass fil-
tered prior to computing the auto-correlation function using a 6th -order low-pass
non-causal Butterworth filter with cutoff frequency 0.5U∞ /L (using Matlab idfilt
function). The grey-scale is linearly distributed between correlation levels ranging
from −0.1 (black) to +0.1 (white) using 256 intensity increments. The dashed lines
delimit a region exhibiting significant low-frequency motions. Those limits will be
used in figure 9.19.
equivalent at a fixed point crossed by the reflected shock wave (Rρu =

[ρu] (t0 ) [ρu] (t0 +τ )/[ρu] (t0 ) [ρu] (t0 )). As mentioned in the previous para-
graph, the hot-wire signal is about 17 times longer than the LES one. To see
the short-signal effect on the interpretation of the LES data, the experimen-
tal signal was cut in segments the same length as in the LES. One particular
segment is provided in figure 9.15 for illustration purposes. It shows that for
the frequencies of interest here, the LES signal, and to some extent the exper-
imental signal, are too short to consider the statistics to be fully converged.
Nevertheless, it is possible to infer that the experimental and LES signals
both exhibit similar low-frequency motions.
Figure 9.16 is the Fourier transform of the auto-correlation functions shown
in figure 9.15. By definition, this corresponds to the power-density distribu-
tion (PSD). It is clear that both the LES and the experiment are experiencing
significant low-frequency “tones”. As the time spanned by the signals is in-
creased, the spiky aspect of the spectra will be reduced and the spectra will
become more and more broadband. In figure 9.16, significant oscillations are
found around 0.3 kHz, consistent between experiment and LES. Finally, the
Fig. 9.19 Difference in the weighted wall-pressure PSD between the first half of the
interaction region with the region immediately upstream of interaction. ΔE(f ) =
GI (f )−GU (f ) where GI (f ) is the premultiplied PSD (f ×P SD(f )) integrated over
the region embedded between the two dashed lines in figure 9.18 and normalised
by the total resolved power in the same region. GU (f ) is computed in the same
manner as GI (f ) with the difference that the integration is performed other the
region spanning from (x − Ximp )/L = −2 to the first dashed line in figure 9.18.
The resulting quantity highlights the contribution of each frequency band to the
wall-pressure variance in the initial part of interaction relatively to the incoming
boundary layer. It is aimed at quantifying the build-up of significant low-frequency
oscillations as perceived in figure 9.18.
PSD levels below 0.1 kHz suggest there might be significant very low-frequency
oscillations but given the relatively poor convergence on that end of the spec-
trum, it is premature to draw any conclusion.
Figure 9.17 illustrates the intermittent nature of the reversed flow. It shows
that although the mean separation bubble is very shallow, it can sometimes
extend much further away from the wall. In the LES data, the peak span-
and time-averaged reversed flow is only 0.7% of the upstream freestream
velocity but the strongest reversed velocity recorded was up to 43% that
of the upstream freestream velocity. Such levels of intermittency are surely
challenging for the RANS models (and for a single-wire hot-wire probe).
Moreover, the mean separation bubble is found to have an aspect ratio of
about 150 with a height not exceeding 10 upstream-wall units. Such scales
suggest that URANS and even DES approaches have little chances of properly
capturing the unsteady motions.
Fig. 9.20 Wall-pressure frequency/wavenumber diagrams, derived from the 2D

Fourier transform of the 2-point correlation function, where the reference point was
taken at (x − Xsep )/Lsep ≈ 0.3. The contours show the premultiplied PSD on a log-
arithmic scale. Unlike for the previous figures, the turbulence content was retained
(no low-pass filter was applied). However, the contours of the premultiplied PSD
were made smoother by filtering out the high-frequency fluctuations in the contour
lines. Similarly to figure 9.18, the 2-point correlation function is also averaged in
the spanwise direction. Hence, only the streamwise wavenumbers are considered.
The yellow lines indicate the dispersion relation for various acoustic waves. The
solid lines correspond to an acoustic wave travelling in the upstream potential flow
(U1 ± c1 ) for the positive-wavenumber quadrant. The (−cw ) upstream-propagating
acoustic wave at the wall is also shown in the negative-wavenumber quadrant. The
equivalent relations for regions 2 and 3 are shown with dotted and dashed lines,
respectively, where region 2 is the potential flow located between the oblique im-
pinging and reflected shockwaves while region 3 is the potential flow located after
the shock system. Finally, the thick solid blue and white lines indicate an upstream
propagation at 5% of the upstream freestream velocity.
The question of the origin of the low frequencies is still debated and the
reader is referred to [6, 1, 5, 3, 7, 2, 9, 4]. More work is ongoing and this
report does not discuss this issue. Nevertheless, the following remarks can be
made.
Figure 9.18 gives the wall-pressure auto-correlation function as a function
of the streamwise location. One can see a clear band with large-scale black
and white structures, corresponding to high levels of correlation over large
time lags. These larger-scale structures are seen to emerge slightly upstream
Fig. 9.21 Shock-foot probability density function
of the mean shock-foot position and they stop in the middle of the interaction.
This implies that the low-frequency motions are more energetic in the first
part of the interaction. This is consistent with the experimental findings (see
ITAM’s final report).
Moreover, no large-scale structures are found upstream of interaction in
figure 9.18. This suggests that the low-frequency motions emerge inside
the interaction. To further quantify the perceived low-frequency build-up in
figure 9.18, one can compute the contribution of each frequency band to the
wall-pressure variance relative to the upstream fluctuations, as described in
figure 9.19. It is found that significant fluctuations below St = f L/U∞ = 0.03
emerge in the initial part of the interaction. This is comparable with the ex-
perimental findings reported in ITAM’s final report.
The above analysis can be refined further by computing the streamwise-
wavenumber/frequency diagram, shown in figure 9.20. It reveals the pres-
ence of upstream-propagating pressure waves. A fast one corresponding to
the upstream-propagating acoustic wave at the wall and slow-upstream-
propagating/large-wavelengths wave (where uc /U∞ ≈ −0.05). Note that re-
placing U∞ by uc for the velocity-scale in the Strouhal-number definition
would make the Strouhal number be close to unity. Given the choice of ref-
erence point at (x − Xsep )/Lsep ≈ 0.3 to compute the 2-point correlation
function, it can be inferred that the slow upstream-propagating waves orig-
inate further downstream. Combined with the findings in figure 9.18, it is
probable that the source point is somewhere between a third and half the
9.8 Conclusions 283
way down the separation bubble. As for the upstream acoustic wave, its main
possible source is at vortex/shock or vortex/wall interaction. The possibility
of having a self-sustained acoustic feeback-loop mechanism as proposed in [5]
is reasonable, although a simple Rossiter-type scaling would expect the first
mode to be an intermediate frequency range (St ∼ O(0.1)).
Finally, figure 9.21 compares the experimental and LES shock-foot mo-
tions. The experimental shock position was extracted from high-speed
schlieren visualisations (see ITAM’s final report), while the shock motions
in the LES were detected using an algorithm based on the divergence of
the velocity field. From there, the shock-foot time series were obtained using
a best-line fit on the instantaneous shock positions. The probability-density
functions of both the experimental and numerical shock-foot motions are
found to agree remarkably well (despite the poor agreement in the time-
averaged fields). Interestingly, they both match the normal distribution.
9.8 Conclusions
The case of an oblique shock generated by a 7◦ -wedge angle and impinging
on a Mach 2 turbulent boundary layer at Reynolds number of about 104
based on the displacement thickness has been investigated both experimen-
tally and numerically. A challenging low-Reynolds number target was set for
the experiment. The high frequencies emerging from the use of a relatively
thin turbulent boundary layer proved to be a difficult problem for the mea-
surement campaigns, not to mention the issues related to the need for an
artificial by-pass transition to turbulent flow.
The CFD and experimental results exhibit relatively large differences be-
tween them. In particular, the simulated interaction region is consistently
smaller in the simulations than in the experiment.
The RANS investigations have evidenced a great sensitivity of the results
to the choice of inflow turbulence intensity, with a separation bubble that
could more than double in length by changing the inflow turbulence intensity
from 3% to 0.1%. The bubble expansion was seen to occur by an upstream
shift of the separation point when the inflow turbulence intensity was in-
creased. Such sensitivity to the inflow turbulence intensity could be observed
in both the 2D and 3D RANS approaches.
The 3D RANS investigations were found to produce significant corner
flows near the junction between the plate and the wind-tunnel side-walls.
The corner flows were seen to affect the predicted interaction length at the
wind-tunnel mid section by increasing its size compared to the 2D RANS.
This could explain the smaller LES value.
The aforementioned corner flows were found to be sensitive to the ra-
tio between the side-wall boundary-layer thickness and the plate boundary-
layer thickness, in addition to the choice of incoming turbulence intensity.
284 References
Increasing the side-wall boundary-layer thickness was found to enhance the

separation-bubble size. The corner-flow issue, which was originally intended
to remain an experimental issue, turned into a key factor for a successful com-
parison with the experimental data. It should be the focus of more detailed
studies in the future.
With respect to the unsteady investigations, the URANS attempts did
not exhibit significant unsteadiness. However, both the experimental and
LES results have shown the existence of low-frequency oscillations (two or-
ders of magnitude smaller than the frequency associated with the incom-
ing boundary-layer properties). These low-frequency oscillations are more
energetically-significant in the first half of the interaction. The LES wall-
pressure data analysis tend to argue in favour of an intrinsic source for the
low frequencies.
In addition, the LES data revealed that the mean separation bubble is
shallow, with an aspect ratio well above 102 with a maximum height not
exceeding the buffer region of the incoming boundary layer. However, the
reversed-flow region can be intense at times with speeds that can be up to
half the freestream one. Such a level of intermittency of the reserved-flow
region is not expected to be properly modelled by the (U)RANS turbulence
models or captured by simple DES approaches. More modelling in this area
may be necessary.
Interestingly, the shock motions as found in the experiment and in the LES
agreed remarkably well and appear to follow a normal distribution. However,
the overall agreement regarding the time-averaged fields is poor in contrast
with the IUSTI case. This raises questions about this particular choice of
configuration. The study of the corner-flow physics is potentially interesting
in an attempt to improve the reliability of RANS approaches.
References
[1] Dupont, P., Haddad, C., Debiève, J.F.: Space and Time Organization in a
Shock-Induced Separated Boundary Layer. J. Fluid Mech. 559, 255–277 (2006)
[2] Dussauge, J.-P., Piponniau, S.: Shock/boundary-layer interactions: Possible
sources of unsteadiness. Journal of Fluids and Structures 24, 1166–1175 (2008)
[3] Ganapathisubramani, B., Clemens, N.T., Dolling, D.S.: Effects of Upstream
Boundary Layer on the Unsteadiness of Shock-Induced Separation. J. Fluid
Mech. 585, 369–394 (2007)
[4] Piponniau, S., Dussauge, J.P., Debiève, J.F., Dupont, P.: A simple model for low
frequency unsteadiness in shock induced separation. Accepted for publication
in Journal of Fluid Mechanics (2009)
[5] Pirozzoli, S., Grasso, F.: Direct Numerical Simulation of Impinging Shock
Wave/Turbulent Boundary Layer Interaction at M = 2.25. Physics of Flu-
ids 18(6) (2006)
References 285
[6] Plotkin, K.J.: Shock Wave Oscillation Driven by Turbulent Boundary-Layer

Fluctuations. AIAA Journal 13, 1036–1040 (1975)
[7] Robinet, J.-C.: Bifurcations in Shock-Wave/Laminar-Boundary-Layer Interac-
tion: Global Instability Approach. J. Fluid Mech. 579, 85–112 (2007)
[8] Spalart, P.R.: Direct Simulation of a Turbulent Boundary Layer up to Reθ =
1410. J. Fluid Mech. 187 (1988)
[9] Touber, E., Sandham, N.D.: Large-eddy simulation of low-frequency unsteadi-
ness in a turbulent shock-induced separation bubble. Accepted for publication
in Theoretical and Computational Fluid Dynamics (2009)
Chapter 10
(Eric Garnier)
10.1 Introduction
This chapter summarizes the activities conducted within the UFAST project
on the IUSTI shock reflection test case experiment [1,2]. This type of inter-
action is in particular representative of issues encountered in air intakes of
supersonic aircrafts. The interactions resulting from two different shock de-
flector angles were investigated using a large diversity of numerical methods
from RANS to LES. Furthermore, IUSTI has studied the effect of a fluidic
control of the interaction by continuous AJVG (Air Jet Vortex Generator).
This flow was computed using both RANS and Stimulated DES. Six UFAST
partners were involved in this test case: IMFT, IUSTI, NUMECA, ONERA,
SOTON, and UAN.
10.2 Flow Case Description

Description of the baseline configuration
The experiment that we intend to reproduce numerically is carried out in

the supersonic wind tunnel of IUSTI. The rectangular shaped nozzle (height
120 mm, width 170 mm) is designed for a nominal Mach number of 2.3. A de-
flector which can be seen as a flat plane with a sharp leading edge is held
by the wind tunnel upper wall (Figure 10.1). It generates an oblique shock
which reflects on the lower wall. Depending on the deviation angle imposed
to the flow, the interaction between the shock and the lower wall boundary
layer is more or less strong and, for a deflector angle larger than or equal to
8 degrees, the boundary layer separates. Low frequency unsteady movements
of the reflected shock are then registered. In the particular case of a wedge
inclined at 9.5 degrees on which this study is grounded, three dimensional
vortical structures have been observed in the separated zone.
Before the interaction (at x = 240 mm), the boundary layer is fully
turbulent and its thickness is about 11 mm. The stagnation pressure and
springerlink.com
288 10 Oblique Shock Reflection at M = 2.25 (Eric Garnier)
Fig. 10.1 Description of the interaction (schlieren picture from IUSTI)
temperature being respectively equal to 0.5 bar and 300 K, the Reynolds num-
ber based on the momentum thickness is about 5000.
Description of the flow controlled case
The effect of AJVG has only been investigated on the 9.5 degrees case. Ten
AJVG have been positioned at x = 212 mm (about 6 boundary layer thick-
nesses before the interaction). The step between each jet is equal to 10 mm.
The flow is then controlled over 10/17 of the whole span. Their skew angle
is equal to 90 degrees and their inclination with respect to the plate (pitch
angle) has been fixed to 45 degrees. The diameter of the holes was chosen
equal to 0.8 mm. These continuous jets are all linked by a settling chamber
(Figure 10.2). The stagnation pressure in the chamber can be adjusted in
the range P0 = [0.05–0.5] bar. 0.05 bar corresponds to ’jets off’ situation for
which the pressure is in equilibrium with the local static pressure in the test
section. Since the jets are not equipped with nozzles, the flow is sonic for
a sufficient stagnation pressure (0.08 bar). The mass flux of AJVG represents
about 3% of the boundary layer flow deficit.
Available measurements
Significant efforts have been performed to document the inflow boundary

layer characteristics [2]. The mechanical vortex generators used to trigger
transition in the converging part of the nozzle have been removed since their
wakes remained present in the inflow boundary layer.
The main mean of investigation used in this study is the PIV (two and three
components). 4 planes in the wall parallel direction (3C) are available for
10.2 Flow Case Description 289
Fig. 10.2 Description of the AJVG device. Top: global view of the plate. Bottom:
view of the pressure chamber.
both controlled and baseline configurations. In the latter configuration, the

symmetry plane in the longitudinal direction (2C) has been provided whereas
in the controlled flow the spanwise variation of the flow was investigated
using 3 longitudinal planes (the symmetry plane z = 0, z = −2.5 mm and
z = −5 mm). These PIV results have been cross-validated with previous hot
wire and LDA data. They are well converged since about 3000 fields were
used to compute the first and second moments of the velocity field. Two
datasets termed PIV2006 and PIV2007 have been provided by IUSTI for the
baseline configuration. They differ by the height of the inflow boundary layer
(a momentum thickness of 0.96 mm for PIV2006 and 0.87 mm for PIV2007).
This change in the boundary layer height is tentatively attributed to the
modification of the PIV seeding device. The flow controlled case results are
only provided for the 2007 set-up. The four first moments of the pressure
fluctuations have also been supplied for the 2006 database. Additionally, hot
wire measurements have been used by IUSTI in particular to justify the

construction of a model of flow unsteadiness [8].
10.3 Important Flow Features

Baseline configuration
Figure 10.3 presents the longitudinal mean velocity and the longitudinal ve-
locity fluctuations for the 8 and 9.5 degrees cases. It is observed that the
interaction is much stronger in the latter case than in the former one. In
both cases, the flow is separated but the separation bubble length is twice
longer in the 9.5 degrees case than in the 8 degrees case. In the former case,
its height reaches about one third of the inflow boundary layer thickness.
The amplification of the fluctuations is strong in the reflected shock area.
This amplification is both due to the movement of the reflected shock and to
the turbulence enhancement. The production of turbulence is intense in the
shear layer located above the separation bubble. The return to a canonical
situation where the maximum of fluctuations is located in the wall vicinity
does not occur before the end of the measurement domain.
Fig. 10.3 Longitudinal mean velocity (top) and fluctuations (bottom) for the 9.5
(left) and the 8 (right) degrees cases (2007 dataset)
For the 8 degrees case, it is worthwhile to notice that the thicker boundary
layer of the 2006 database increases the interaction length by about 10%.
More impressively, the bubble height is twice as large in the 2006 database
as in the 2007 one. The 9.5 degrees case is much less sensitive to the state of
the inflow boundary layer than the 8 degrees case and the change in term on
interaction length remains lower than 5%.
10.3 Important Flow Features 291
The longitudinal velocity in a plane parallel to the wall is shown in

Figure 10.4. It is observed that at this altitude the length of the separa-
tion bubble is two times larger in the 9.5 degrees case than in the 8 degrees
case. More importantly, the flow topology is different and in the 9.5 degrees
case the reflected shock is slightly curved and vorticies are identified on the
external edges of the separated area which appears as very three-dimensional.
Comparatively, the separated zone of the 8 degrees case is much more two-
dimensional. This raises the issue of the origin of this three-dimensionality.
It can be solely due to the lateral wall presence or to an intrinsic instability
in such interaction or to an amplification of the latter by the lateral wall.
Unfortunately, the zones close to the lateral walls are not accessible to the
PIV technique; one can expect that the simulations could contribute to the
interpretation of experimental results.
Fig. 10.4 Longitudinal mean velocity in a plane parallel to the wall at z = 0.6 mm.
Left: 9.5 degrees case, right: 8 degrees case.
Another fundamental feature of such flows is the low frequency unsteadi-

ness of the reflected shock. Defining X∗ as X∗ = (X − X0 )/L where X0 is
the maximum of the pressure fluctuations and L the interaction length, it is
observed in Figure 10.5 that energetic frequencies of few hundreds of hertz
are found near the reflected shock mean position. Such low frequencies are
not observed in the inflow boundary layer. This suggests that the flow un-
steadiness is intrinsic to the interaction and related to the separated bubble
dynamics.
In Figure 10.6 which shows both the conditional average of the PIV lon-
gitudinal velocity field with respect to the shock position and the vertical
velocity fluctuations conditioned to the bubble height for different altitude,
it is clearly observed that the contractions of the bubble are related to
Fig. 10.5 Spectrum of the wall pressure fluctuations in the symmetry plane
Fig. 10.6 Conditional average of the PIV longitudinal velocity field with respect
to the shock position (left). Vertical velocity fluctuations conditioned to the bubble
height for different altitude.
downstream movements of the reflected shock whereas dilatations are related

to upstream movements of the reflected shock.
Significant progress in the understanding of such unsteadiness has been
performed by the IUSTI team [8] which has proposed a model which per-
mits an estimation of the frequency unsteadiness from a limited number of
parameters characterizing the interaction:
fl l
Sl = = Φ(Mc )g(r, s) .
u1 h
10.3 Important Flow Features 293
The Strouhal number of the fluctuations then depends on the bubble aspect
ratio (l/h), a function (φ) of the convective Mach number which character-
izes the spreading rate of the shear layer above the separation bubble and
a function g of the velocity and density ratios in the shear layer.
Flow controlled case
ONERA and UAN have chosen to simulate the case for which P0 = 0.4 bar.
For this flow condition, the jets static temperature is 240 K and their static
pressure is equal to 21000 Pa (five times the static pressure in the wind tun-
nel). The jets are then underexpended. The presented results concern this
particular case for which the AJVG effects are the strongest.
The corrugation of the shock due to the wake of the AJVG is clearly
observable in Figure 10.7 which presents the longitudinal velocity in the plane
z = 1 mm. These corrugations are stronger on the separation length than on
the shock. The reattachment position is only barely affected by the wake of
the AJVG at least at this altitude. In fine, the spanwise fluctuations of the
bubble length reach about 20% of its mean value. The global effect of the
AJVG is then limited. Furthermore, it is worthwhile to notice that a backward
displacement of the recirculation area is associated to a local maximum of
longitudinal fluctuations.
Fig. 10.7 Longitudinal velocity in a plane parallel to the wall z = 1 mm (PIV

results from IUSTI). Left: baseline. Right: controlled case (P j = 0.4 bars).
The penetration length of the jet of about 5 mm can be estimated from

Figure 10.8 (right). The backflow intensity appears weaker in the controlled
case than in the baseline one but as already mentioned the effect of the
AJVG on the separation length is only of limited intensity. Nevertheless,
IUSTI has evidenced that the slight reduction of the separation length leads
to a measurable increase of the frequency of reflected shock movements.
Even if the effect of AJVG was not sufficiently marked for application
purpose, it allows to investigating the behaviour of the separated flow when
submitted to a change in the inflow condition (a fuller velocity profile).
Fig. 10.8 Longitudinal velocity obtained by 2C PIV. Left: baseline configuration

in the plane y = 0. Right: flow controlled configuration in the plane y = −2.5 mm.
10.4 Numerical Investigations

Uncontrolled flow
Summary of existing computations
Table 10.1 summarizes the computations that have been performed on this
flow test case. Most of the simulations were dedicated to the 9.5 degrees case
and they have taken into account the whole wind tunnel geometry. The fact
that a wide range of turbulence modelling (from RANS to LES) has been
used will permit to draw some conclusions concerning this issue.
Table 10.1 Summary of the computations carried on the IUSTI test case
Modelling Partner Case Remark

RANS NUMECA 9.5 SA k-ε
IMFT 9.5 SA
UAN 8 k-ω SST
UAN 9.5 k-ω SST
ONERA 9.5 SA
URANS NUMECA 9.5 SA
IMFT 9.5 SA
UAN 9.5 k-ω SST
DES MUMECA 9.5 SA based
IMFT 9.5 SA based
DDES NUMECA 9.5 SA based
IMFT 9.5 SA based
ONERA 9.5 SA based
SDES ONERA 9.5 SA based
LES SOTON 8 Different grids,
models, Inflow
conditions
ONERA 9.5 Span = 10 cm
A brief description of each code has also been provided in Table 10.2. It
may help for the results interpretation.
10.4 Numerical Investigations 295
Table 10.2 Brief description of the codes used in this study
Partner Temporal scheme Spatial scheme Modelling Remark

NUMECA Runge Kutta Jameson SA, k-ε, SST, Multigrid
+IRS v2-f, DES-SA
IMFT DTS Roe (Van SA, DES, OES
Leer limiter)
UAN Implicit O(2) ENO Godunov k-ω SST Realisability cond.
O(2) in turb. model
SOTON RK3 Centered O(4) MTS, Dyn. Digital based filter
+TVD filter for inf. Cond.
ONERA Implicit O(2) Roe, O(2) MSM, SA, SEM for inflow
+TVD filter xDES SA based cond.
The 9.5 degrees case
RANS results
We first focus on the 9.5 degrees case for which most of the data is available.
Before comparing the results obtained by the UFAST partners on this test
case, we provide some first comments about the three-dimensionality of the
flow.
The Figure 10.9 highlights the flow topology extracted from a RANS com-
putation. The isocontour u = 0 is represented in brown and the incident
shock position is identified by an isovalue of the pressure. It is observed that
the main separated zone covers only about 60% of the whole wind tunnel
Fig. 10.9 Flow topology. Brown: isovalue u = 0. Light blue: isovalue P = 5500 Pa.
Black: shock generator.
span. The interaction also leads to separations in corners. Between the main
separated zone and corner structures the flow is accelerated and does not
separate.
It results from these observations that a 2D RANS computation has no
chance to provide accurate results. This statement has been checked and
it has been measured that the separation length is equal to 35 mm with
a 2D RANS (35 mm) whereas a 3D computation gives a separation length
of 60 mm. The separation is not directly measured in the experiment but
its value should be close to the interaction length which is equal to 71 mm.
The first conclusion of this study is that 3D computations are mandatory to
treat the 9.5 degrees case. This explains that the LES of a periodic slice of
10 cm performed by ONERA was not successful in reproducing the interaction
topology.
Quantitative comparisons of the longitudinal velocity and pseudo-stream-
lines in a plane parallel to the wall are presented in Figure 10.10. The solution
appears to be very sensitive to the turbulence model, the SA model giving the
better results. It is speculated that the modelling of corner flow which is still
a topic of research even for incompressible flows is the key parameter for the
simulation of this configuration. Indeed, it is observed that bigger corner flows
are associated to a stronger curvature of the reflected shock which in turn
might enhance the interaction and increase the length of the main separation
bubble. Following this idea, it is proposed that the vortices located in the
separation bubble are simply induced by the vertical vorticity of the corner
vortices. NUMECA and ONERA results with the SA model suggest that the
coarser grid gives results closer to the experiment. But one should also keep
in mind that the spatial scheme and the model implementation are not the
same in the two codes. Moreover, the gap between shock generator and lateral
walls is only taken into account in the ONERA computation.
The correlation between larger corner vortices and a thicker separation
bubble is clearly evidenced comparing Figure 10.10 and Figure 10.11 for the
NUMECA SA and k-ε results. Since IMFT and NUMECA have employed
both the same grid and the same turbulence model obtaining significantly
different results, it is worth noticing that spatial scheme and model imple-
mentation issues can not be neglected. In general, the better balance between
every numerical ingredient (grid, model, numerical dissipation) is found on
the NUMECA SA computation.
Results from unsteady computations
It has been pointed out in the literature that the lack of accuracy of RANS
methods in such configurations is intrinsic to the steadiness of these methods.
Even if most of the error on the flow general topology seems to be due in this
configuration to the treatment of corner flows, it is clear that the smoothing
of the pressure gradient by the shock movement can not be reproduced nu-
merically by RANS. Moreover, from an application point of view, the flow
Fig. 10.10 Longitudinal velocity and pseudo streamlines in a plane parallel to the
wall
unsteadiness leads to large scale fluctuations which can have for example
deleterious consequences in the compressor of a jet engine. The access to
such information can only be given by unsteady computations. Consequently
the capacities of URANS, DES, DDES and SDES have been assessed in the
9.5 degrees case.
Concerning URANS, NUMECA and IMFT with the SA model and UAN
with the k-ω SST model agree on the fact that their computation remains
steady. The instability of such flow is here not sufficient to trigger the un-
steadiness.
The original DES97 technique has been assessed by NUMECA and IMFT
on the 4.6 × 106 points grid used for RANS. The grid is locally sufficiently
fine to promote a switch of the DES in LES mode but since the modelled
Reynolds stresses are not replaced by resolved stresses the simulation tends
to laminarize and gives unphysical solutions. This phenomenon is known as
Fig. 10.11 Longitudinal velocity and pseudo streamlines in the symmetry plane
MSD (Modelled Stress Depletion). The DDES [9] has latter been developed
to avoid MSD. NUMECA, IMFT and ONERA have then carried out DDES.
The ONERA grid being finer than the one used by NUMECA and IMFT
(23 × 106 and 4.6 × 106 points respectively), the simulations have exhibited
different behaviours. On the coarser grid DDES remains steady and converges
toward the RANS solution. On the finer grid, the simulation switched in LES
mode only in corner flows, the main separated zone remaining steady. Another
attempt was performed at ONERA forcing corner flows in RANS mode. It
led to a switching of the DDES in LES mode in the main separation, but the
simulation still suffered from the incorrect treatment of corner flows by RANS
technique. An extension of DDES [14] was also tested but after a promising
beginning the computation eventually converged toward a RANS solution.
Coming to the conclusions that the mesh needs to be specifically refined in
DDES with respect to RANS and that instability of the flow is not strong
enough to generate LES content both in the main separation and in corner
flows, it has been proposed to introduce turbulent fluctuations at the inflow
of the computational domain as in LES [10,12]. This approach is denoted
Stimulated DES (SDES). The chosen technique of inflow turbulence gener-

ation is the Synthetic Eddy Method (SEM) [5,6]. Figure 10.12 illustrates
clearly the fact that LES content (resolved eddies) is introduced at the en-
trance of the computational domain over the boundary layer height. Never-
theless, lateral boundary layers are treated in RANS mode.
Fig. 10.12 Visualization of the flow (one isovalue of the Q criterion coloured by
the longitudinal velocity). In purple, one isovalue of the pressure highlighting the
incident shock.
Fig. 10.13 Longitudinal velocity and pseudo streamlines in the plane located at
1.2 mm from the wall. Left PIV (IUSTI), right: SDES computation (ONERA).
It is observed in Figure 10.13 that even if some improvement is registered

with respect to the RANS computations performed on the same grid (see
Figure 10.10), it seems that in this case the extent of the corner flows is too
small. This is tentatively attributed to the fact that lateral wall are treated
in RANS at the inflow.
The agreement between SDES and PIV is generally better in the symmetry
plane (see Figure 10.14) even if bubble aspect ratio is larger in the SDES
computation than in the experiment. According to the model proposed in
[8], this should lead to an increase of the frequency of the reflected shock
movement.
Fig. 10.14 Longitudinal velocity in the symmetry plane. Left: PIV; right: SDES.
Even if the results of these SDES simulation are far from being perfect, it is
the only one coming from unsteady computation that exists on the 9.5 degrees
case on the full geometry. Some work has then been performed to analyze
the unsteady data hoping that it could complement the experimental results
in a useful way.
Figure 10.15 presents isocontours of wall pressure fluctuations and stream-
lines which allow a visualization of the flow topology. Upstream of the
interaction, pressure fluctuations are weak and they are due to turbulent
fluctuations present in the boundary layer. A local pressure maximum is
observable near the separation at x = 0.25. This quantity is used in the ex-
periments to identify the beginning of the interaction zone. Nevertheless, one
can observe that the maxima of pressure fluctuations can be found in corner
flows and downstream from the interaction. In the latter case, these fluctu-
ations are associated to Kelvin-Helmholtz type vortices which are generated
in the shear layer above the separation bubble. More generally, these results
indicate that the unsteady movements of highest intensity are localized in
corner flows and a possible statistical link between these corner flows and the
main separation area must be investigated.
The longitudinal evolution of the spectral energy density of wall pressure
fluctuations premultiplied by the frequency and normalized by the signal vari-
ance is presented in Figure 10.16. A frequency resolution FR of 200 Hz has
been chosen to limit the statistical error due to short signal duration (only
80 ms). Very high frequency fluctuations are present in the inflow bound-
ary layer (x∗ < −0.05). In the range −0.05 < x∗ < 0.15, the energy is
Fig. 10.15 Wall pressure fluctuations and streamlines
Fig. 10.16 Power spectral density of wall pressure fluctuations premultiplied by

the frequency (f G(f )/σ in the plane (X∗, f ))
concentrated in a very low frequency band with an energy peak in the range
300 to 500 Hz. In the experiment, a bump centred around 200 Hz is evidenced
in the pressure spectra. The fact that higher frequencies are found in the com-
putation is surprising if one scales the Strouhal number of the low frequency
movements on the interaction length. Nevertheless, in the IUSTI model [8],
the frequency depends on the bubble aspect ratio, a larger aspect ratio (as
in this computation) giving a higher frequency.
In the range 0.4 < x∗ < 0.8, the spectrum is really broadband and, for
x∗ > 0.8, we can observe that energy concentrates in the range 3 to 10 kHz
as in the experiment. Such frequencies correspond to large Kelvin-Helmholtz
type structures formed in the shear layer above the separation bubble.
The computation allows getting information about the frequency content

of the pressure fluctuations in the whole flow. In particular, the relative con-
tribution of frequencies lower than 1000 Hz with respect to total fluctuations
is plotted in Figure 10.17 for the semi-plane y < 0.085. It can be observed
that low frequencies contribute for more than 70% of the total energy under
the footprint of the reflected shock. In the separated zone, 20% of pressure
fluctuations are attributable to low frequencies. This figure is in agreement
with the experimental observations of [7]. A slightly larger level (30%) of low
frequency content is found in the corner flows shear layers.
Fig. 10.17 Relative part of pressure fluctuations in the range 80 to 1000 Hz with
respect to total fluctuation in the semi-plane y < 0.085
Next, it was tried to find a statistical link between sensors whose locations
are defined in Figure 10.15. No significant correlation was found between C1-
C2, C8-C9, C5-C8 and C1-C3. This shows that both corner flows and counter
rotating vortices are not statistically linked to the reflected shock movement
for frequencies larger than 200 Hz. Unfortunately, due to the short signal
duration, lowering the frequency range of interest increases the error on the
coherence evaluation. The coherence at lower frequency will then remain to
be investigated. As expected, a strong link is found between C3 and 10 with
values of the coherence reaching 0.8 for the lowest frequencies. The phase
difference between the two signals is quasi null. This suggests the existence
of an ensemble motion of the reflected shock.
A significant link (the correlation reaches 0.6) is evidenced between the
beginning of the separation region (C3) and the reattachment point (C12).
Moreover, the phase between these two signals being constant and equal to
π on the low frequency range, the anti-correlation is clearly demonstrated.
This has already been observed experimentally by Dupont et al. [7] who have
concluded to the existence of global movement excluding the predominance

of propagation phenomena.
The 8 degrees case
RANS results
As already mentioned in the comment of Figure 10.4, with the lower shock
intensity, corner flows are of more limited extent than in the 9.5 case degrees.
This is illustrated in Figure 10.18 which shows that the flow is separated from
one lateral wall to one another, the isosurface u = 0 (in brown) connects
corner flows and the central separated zone.
Fig. 10.18 Flow topology. Brown: isovalue u = 0; Light blue: isovalue P = 5500
Pa; Black: shock generator (8 degrees case, RANS computation with k-ω SST model
from UAN)
Quantitatively, it is observed in Figure 10.19 that corner flows are over-

estimated by the k-ω SST computation. One may regret that other partners
did not compute this case using RANS. It might have permitted to judge of
the RANS results quality on a flow which is supposed to be less sensitive to
the treatment of corners. It is worthwhile to notice that despite of perfectible
flow topology, the UAN computation is good in reproducing the wall pressure
distribution in the symmetry plane.
LES results
Most of the work on the 8 degrees case was performed by SOTON which
has performed many LES computations assuming that the flow is sufficiently
two-dimensional to permit the use of periodic condition in the spanwise di-
rection. They have discussed methodological issues such as the choice of span
Fig. 10.19 Longitudinal velocity and streamlines in a plane parallel to the wall.
Left: PIV (IUSTI); right: k-ω SST computation (UAN).
of the computational domain, subgrid scale model, grid resolution and inflow
condition type. The results of this parametric study are published in [11]. As
a summary, the chosen resolution is 40×1.6×13.5 wall units. Digital filters
[13] have been preferred to synthetic turbulence and Mixed Time Scale Model
has been selected in place of the dynamic model. A short span computation
(0.7δ0 ) permits to collect statistics over long computational time. A large
span computation (7δ0 ) allows for evaluating the consequences of a limited
span on the results. The grid used for the latter computation encompasses
about 20 × 106 points.
Velocity fluctuations in a plane parallel to the wall evidence the presence
of low and high velocity streaks that populates canonical boundary layers.
After the separation (identified by the first dashed line), the size of turbulent
structures in the spanwise direction significantly increases and further down-
stream the turbulence slowly relaxes toward its canonical state. This figure
illustrates the fact that the simulation is capable of capturing most of the
finest turbulent structures present in a supersonic boundary layer.
Quantitative comparisons in the symmetry plane are proposed in
Figure 10.21. The agreement between experiment and simulation is very good
in the symmetry plane for the longitudinal velocity except in the separation
bubble region. Nevertheless, it is important to mention that this region is
very sensitive to the nature of inflow perturbations since a large variability
of the results in this area has already been observed in the experiment, the
2006 data differing from the 2007 one specifically in this region. The agree-
ment with the experiment is also generally satisfactory on the Reynolds shear
stress.
Longitudinal evolution of turbulence spectra in the spanwise direction are
presented in Figure 10.22 for both large and narrow span simulations. In the
separation region, it appears that a large part of the energy is contained in the
Fig. 10.20 Velocity fluctuations in a plane parallel to the wall
Fig. 10.21 Comparison of PIV2007 and LES results. Left: longitudinal velocity.
Right: Reynolds shear stress.
Fig. 10.22 Longitudinal evolution of turbulence spectra in the spanwise direction

for two altitudes
small wave numbers in the large span computation. The cutoff wave number
imposed by the finite span is too large in the narrow span simulation. This
forces the energy to concentrate at smaller scales and affects the results.
Low frequency movements of the reflected shock are clearly observed in
Figure 10.23. As in the experiment the frequency of the power spectral density
maximum is located at St = 0.03. The agreement on the energy distribution
between the narrow span computation and the experiment is very good. To
obtain this result the computation was averaged over 90 cycles of the most
energetic low-frequency oscillation. This represents a significant progress with
respect to previous studies of this kind.
Fig. 10.23 Left: longitudinal evolution of pressure spectra in the streamwise di-
rection. Right: pressure spectra at x = 0.
Intercorrelations between one reference point taken at x∗ = −0.2 and every

other point in the separation region (on the wall) allows to computing the
phase angle of propagating information. One can notice in Figure 10.24 that
for two different (low) frequencies (St = 0.036 and St = 0.054), this phase
changes abruptly by a π increment near x∗ = 0.3. This location corresponds
to a node in global modes extracted from a stability analysis (not shown).
The reader is encouraged to consult [11] to get more information about the
comprehensive physical analysis of this flow performed by SOTON.
10.5 Controlled Flow 307
Fig. 10.24 Phase evolution at three different frequencies with respect to a refer-
ence probe at x∗ = −0.2
10.5 Controlled Flow
This test case was only computed by UAN and ONERA. In the experiment
the hole shape is elliptic and its section is equal to 1.4 × 1 mm2 . In both
simulations, the jet is imposed by a wall boundary condition. In the UAN
computation, the grid has been refined specifically for this computation but
at the AJVG location one grid cell represents a surface of 1.7 × 1.2 mm2
on the finest grid (6.2 × 106 points). No particular grid adaptation has been
performed on the ONERA computation (18×106 points) for which the surface
of one cell is equal to 0.7 × 0.7 mm2 . One can then anticipate that in both
computations the mechanisms of vorticity generation will be affected by the
marginal resolution in the jet vicinity. ONERA has employed the same SDES
technique as for the uncontrolled flow and UAN the same RANS k-ω SST
model.
The effect of AJVG in the computations is consistent with that is expected
from the literature of jets in crossflow. On the SDES mean field presented
in Figure 10.25 (left), one can notice that every AJVG generates a main
vortex which rotates anticlockwise. Secondary vortices of smaller size which
quickly dissipate are also clearly identified. This observation is fully consistent
with the PIV data presented in Figure 10.25 (right) where the secondary
vortices are evidenced in the vorticity field. Furthermore, horseshoe vortices
Fig. 10.25 Left: Visualization of the mean flow of the SDES computation (SDES).
One isovalue of the Q criterion is coloured by the longitudinal vorticity. The sym-
metry plane is coloured by the longitudinal velocity. In purple: an isovalue of the
pressure marks the shock location. In yellow, the isovalue u = 0 m/s. Right: recon-
struction of the vorticity field with PIV planes (IUSTI).
are also shown upstream of each AJVG. They appear to be connected each
other but this effect may be amplified by the marginal resolution. In the
separation zone, additional longitudinal vortices are generated between some
AJVG main vortices. They follow the upper bound of the separation bubble.
AJVG leads to the existence a longitudinal velocity deficit which is clearly
identified in Figure 10.26. As previously mentioned, AJVG wakes corrugate
the upstream side of the separation bubble. Before the shock, it is observed
that velocity deficit is stronger in the SDES case than in the experiment.
The contrary is noticed for the UAN RANS case for which the AJVG effect
is underestimated. The lack of resolution or/and an over-dissipation of the
k-ω SST model can be incriminated to justify this observation. Consequently,
no clear AJVG effect is observed on the separation on the UAN case whereas
their effect tends to be overestimated on the SDES case despite of a lack of
spanwise resolution.
Comparisons in the symmetry plane show that the penetration length of
the AJVG is correctly estimated by the SDES computation. Nonetheless,
Fig. 10.26 Longitudinal velocity in a plane parallel to the wall. Left: PIV – IUSTI;
center: SDES – ONERA; right: RANS – UAN.
10.6 Conclusions 309
Fig. 10.27 Longitudinal velocity in the symmetry plane. Left: PIV IUSTI; Right:
SDES-ONERA.
this computation overestimates the velocity deficit downstream of the fluid

injection.
10.6 Conclusions
This test case is certainly one of the most documented experiments of shock
boundary layer interaction. It has been the subject of a strong interest from
UFAST partners involved in. As a result, a large amount of experiment and
numerical data has been collected. The comparison between experiment and
computations is made difficult by the fact that this flow is very sensitive
to inflow condition (at least for small separation in the 8 degrees case) and
influenced by corner flows for large separation.
Concerning numerical methods, it can be said that RANS gives results of
variable quality. Plausible solutions are found by every partner but only one
simulation which results from a good balance between every numerical ingre-
dient (grid resolution, model, numerical dissipation) matches satisfactorily
with the experiment. A comprehensive sensitivity study should be necessary
to identify the most relevant parameters. This test case is doubly challenging
for RANS since the modelling must account both for the shock unsteadiness
and the presence of significant corner flow. It is worthwhile to notice that
only simple closures have been considered in this study and more advanced
models (of RSM type for example) might provide some improvement.
An important conclusion of this study come from the fact that URANS,
DES and DDES have failed to reproduce the unsteadiness of the flow for
different reasons (weak natural instability of the flow, lack of resolution, ...).
It has then been necessary to use the Stimulated DES which is no more
than a LES ersatz to obtain a reasonable agreement with the experiment and
perform some analysis of the unsteady data which have shown that even if
corner flows are subjected to low frequency movements, these movements do
not seem to be linked with the ones of the reflected shock. Provided that
the computational domain is large enough in the spanwise direction, the
most reliable approach is the LES which in this case can only be applied on
the “2D” 8 degrees case due to current computer limitations. As shown by
310 References
SOTON, this approach permits to investigate deeply the physics of shock

boundary layer interaction taking information for a number of sensors that
might never be available in an experiment. In the IUSTI case, where the
Reynolds number is reasonably low, LES computations of the whole wind
tunnel should be available in a near future. Nevertheless, the most significant
advance in terms of physical understanding of such flow performed within
this project is the model of shock unsteadiness proposed by IUSTI.
The IUSTI study constitutes one of the rare (maybe the unique) existing
experiment of shock boundary layer interaction with AJVG where PIV has
been carried out. Even if the AJVG effect on the separation bubble remains
limited from an application point of view, it permits to study the response of
the separation bubble to different inflow conditions (more or less full inflow
profile). AJVG impose challenging griding issue to simulations that should
have been treated more carefully within UFAST. In the future, a possible
solution to improve the local resolution while keeping a reasonable amount
of points is to use Chimera method. Nevertheless, the under-resolved SDES
computation has permitted to represent some of the important features of the
interaction of AJVG with both the inflow boundary layer and the separated
bubble.
References
[1] Souverein, L.J., Debiève, J.F., Dupont, P., Dussauge, J.P.: Control of an inci-
dent shock wave/turbulent boundary layer interaction at M = 2.3 by means
of Air Jet Vortex Generator, UFAST deliverable 3.3.3 (2008)
[2] Dussauge, J.P., Debiève, J.F., Dupont, P., Piponniau, S.: Report on the mea-
surements of shock reflection at M = 2.25, UFAST project D. 2.3.4 (2007)
[3] Spalart, P.R., Allmaras, S.R.: A one equation turbulence model for aerody-
namics flows, AIAA Paper 92–0439 (January 1992)
[4] Deck, S., Weiss, P.E., Pamies, M., Garnier, E.: On the use of Stimulated De-
tached Eddy Simulation for spatially developing boundary layers. In: Peng,
S.-H., Haase, W. (eds.) Advances in Hybrid RANS-LES Modelling. Notes on
Numerical Fluid Mechanics and Multidisciplinary Design, vol. 97, Springer,
Heidelberg (2008)
[5] Jarrin, N., Benhamadouche, S., Laurence, D., Prosser, R.: A synthetic eddy
method for generating inflow conditions for Large Eddy Simulations. Interna-
tional Journal of Heat and Fluid Flow 27(4), 421–430 (2006)
[6] Pamies, M., Weiss, P.E., Garnier, E., Deck, S., Sagaut, P.: A Generation of
synthetic turbulent inflow data for large-eddy simulation. Physics of Fluids 21,
045103 (2009)
[7] Dupont, P., Haddad, C., Debiève, J.F.: Space and time organisation in a shock-
induced separated boundary layer. J. Fluid Mech. 559, 255–277 (2006)
[8] Piponniau, S., Dussauge, J.P., Debiève, J.F., Dupont, P.: A simple model for
low-frequency unsteadiness in shock-induced separation. J. Fluid Mech. 629,
87–108 (2009)
References 311
[9] Spalart, P.R., Deck, S., Shur, M.L., Squires, K.D., Strelets, M., Travin, A.: A
new version of detached-eddy simulation, resistant to ambiguous grid densities.
Theor. Comput. Fluid Dyn. 20, 181–195 (2006)
[10] Garnier, E.: Stimulated Detached Eddy Simulation of three-dimensional
shock/boundary layer interaction. To be published in International Journal
of Shock Waves
[11] Touber, E., Sandham, N.: Large-eddy simulation of low-frequency unsteadiness
in a turbulent boundary shock-induced separation bubble. Theor. Comput.
Fluid Dyn. 23, 79–107 (2009)
[12] Garnier, E.: UFAST deliverable 5.3.1: LES/DES of the IUSTI shock/reflection
case, RT 1/10261 DAFE/DAAP (2008)
[13] Klein, M., Sadiki, A., Janicka, J.: A digital filter based generation of inflow
data for spatially developing direct numerical or large eddy simulations. J.
Comput. Phys. 186, 652–665 (2003)
[14] Riou, J., Garnier, E., Deck, S., Basdevant, C.: An improvement of Delayed
Detached Eddy Simulation applied to a separated flow over a missile fin. AIAA
J. 47(2), 345–360 (2009)
Part IV
Summary by Work Packages

Chapter 11
WP-2 Basic Experiments
(Jean-Paul Dussauge)
11.1 Introduction
WP2 had three main objectives. A first objective was to constitute a well
documented data base elaborated from experiments run in interaction with
numericians, in order to provide measurements of quality with the pertinent
information for computational purposes. The second objective was to define
baseline flows, before applying any control, so that the effect of control could
be deduced without ambiguity. Lastly, an obvious objective was to bring
some new insight and understanding in the physics of unsteadiness of shock
boundary layer interactions. In the following the main achievements in WP2
are briefly listed, and some conclusions, lessons learned and consequences
under the form of possible prospects are given.
11.2 An Overview of the Main Results of WP2

The experimental work of WP2 has been examined in great details in the
flow case reports. We will list here briefly the main results obtained in the
basic experiments.
For the transonic interactions on profiles INCAS, Bucharest and IoA, War-
saw have provided well controlled data on airfoils. The experiment at INCAS
was made on a biconvex airfoil of large size, with an attempt to minimize the
effect of the walls. In particular, the detailed measurements have included
the effect of the sting holding the model, and have proved their efficiency
in the comparisons with numerical simulations. The experiment at IoA was
made and on a NACA12 profile with aileron. The yaw of the profile and of
the aileron can be changed independently, and in particular the movement
of the aileron can follow periodic laws; this makes it possible to control the
unsteadiness. This model and set up have contributed to experiments of high
flexibility. The experiments at QUB, Belfast consisted in the study of the flow
around a bump formed by a circular arc. The incoming conditions are tur-
bulent and transonic. The results have underlined the importance of the flow
springerlink.com
316 11 WP-2 Basic Experiments (Jean-Paul Dussauge)
generating conditions, in particular the fluid properties, moist, etc, on the

characteristics of the interactions. Moreover, having explored these difficul-
ties, the experiments have led to the conclusion that in the chosen conditions,
the unsteadiness is controlled by the separated zone.
Two of the normal shock experiments in a channel considered forced shock
oscillations one on a bump placed at the wall (ONERA/DAFE, Meudon) and
another one in a plane channel in non-separated and separated conditions
(UCAM, Cambridge); two other experiments considered natural unsteadiness
in plane channel flows in different states of separation and in a curved chan-
nel (IMP-PAN, Gdansk). The forced shock oscillation experiments showed
that in first approximation the dominant importance of the plane shock can
be described from quasi steady considerations. However, the details of the
interactions, such as the length of separation, do not follow the quasi-steady
pattern; moreover the experiments have shown the possibly overwhelming im-
portance of the interactions in the corner flows, which can be strong enough
to change the flow on the centreline of the wind tunnel. The natural oscilla-
tions in a plane channel (IMP-PAN) showed non-monotonic variations with
Mach number, the minimum occurring for incipient separation. The case of
the curved channel provided on the convex wall a boundary with fuller main
profiles producing consequently more limited separation. Correlatively, the
unsteadiness of the shock wave was less important. Note that the spectra of
the shock motion have been measured, and that they provide the material for
further work on the correlations on natural shock unsteadiness in separated
flows.
For the shock reflection cases, three interactions were considered, at Mach
numbers 1.7 (TU Delft), 2.0 (ITAM, Novosibirsk) and 2.25 (IUSTI Marseille).
The Reynolds numbers ranged from Reθ = 50 000 at TU Delft to 5000 in
IUSTI and about 3000 in ITAM. The flow studied in Delft is not separated
(6◦ deviation), while the other cases show evidence of separation (6◦ and 7◦
flow deviation for ITAM and 8◦ and 9.5◦ deviation for IUSTI). In the Delft
cases, instantaneous pockets of separation are present, although the averaged
flow is not separated. Above these instantaneous pockets, large scale vorti-
cal structures are formed, suggesting the development of Kelvin-Helmholtz
like structures. No correlation is found between shock and separation fluctua-
tions. This implies that in this non separated case, the shock motion depends
predominantly on upstream flow fluctuations.
In the ITAM experiment, extensive hot wire and wall hot film measure-
ments have been performed. The dominant frequency in this low Reynolds
number interaction is found in agreement with the results at larger Reynolds
numbers. It was found moreover that the fluctuations at the foot of the re-
flected shock and the fluctuations in the separated flow are out of phase.
A result of this sort had been already found on wall pressure measurements;
this is verified now in the whole flow. The detection of the mixing layer
formed at the edge of the recirculating zone is also derived from the hot wire
measurements. Moreover, measurements have been performed between the
11.3 Lessons Learned and Open Issues 317
wall upstream of the shock (by a hot tube placed at the wall) and the shock
itself and the separated zone. Although the authors conclude to dependence
between upstream turbulence and shock motion, the level of coherence re-
mains rather modest (less than 0.1). Moreover, some spanwise aspects have
been explored, showing the spanwise extent on the shock of the turbulent dis-
turbances. The side wall turbulent boundary layers have also been explored
to define their main characteristics. One of the striking results obtained by
ITAM is that the dominant fluctuations of the shock motion correspond to
a Strouhal number based on the length of interaction and on the external
velocity of 0.03, as in many other interactions. This point will be discussed
in the next section.
In the IUSTI experiment, some data already existed for wall pressure fluc-
tuations for the 8◦ and 9.5◦ deviation case. These results have been com-
plemented by extensive PIV measurements in a vertical plane on the axis of
symmetry of the wind tunnel. The unsteady aspects were documented from
wall pressure spectra and from hot wire measurements in the external flow.
Mean velocity and Reynolds stresses were measured in the interaction. It was
found that the Strouhal number of the dominant frequency is close to 0.03
as previously mentioned. A compilation of this Strouhal number has been
proposed showing the variations of this Strouhal number with Mach number;
this gave evidence that, roughly speaking, this Strouhal number is almost
constant for large enough Mach number, for M > 2, say. Different charac-
teristics of the separation have been derived, for example the wave lengths
involved by the vortex shedding. Moreover, it was proposed to interpret this
low frequency of the shock motion as a consequence of the flapping of the
mixing layer of the separated bubble. This will be rediscussed in the next
section.
11.3 Lessons Learned and Open Issues

A first series of conclusions has been drawn from the study of interactions
in channel flows, which has underlined the importance of the corner flow
interactions, which should be taken in account to predict accurately enough
transonic interaction, even in the middle of the channel. Moreover, it turns
out that the details of the forced interaction and the dynamics of separation
is just poorly predicted by simple quasi-steady considerations, and finally
poorly understood.
For supersonic interactions produced by reflection of an oblique shock or
in compression ramp flows, or a blunt body, a compilation of the dominant
frequency of the shock motion has been proposed. Its analysis has been made
in terms of flapping of the recirculated zone: it is supposed that this flapping
produces shock motion at low frequency and with a large amplitude. Mass
conservation has been applied to the recirculated zone to derive dimensional
dependence. This indicated that the dominant frequency varies like the nor-
malized growth rate of the supersonic mixing layer. This is supported by the
existing experiments, in particular by the experiments of UFAST. Recent re-

sults obtained at Queen’s University, Belfast and at ITAM, Novosibirsk are in
agreement with these values. This suggests more particularly that, for turbu-
lent upstream boundary layers this frequency does not depend on Reynolds
number, but only on the state of the interaction, separated or non-separated.
Of course, transitional interactions in which eddies of particular shape and
dynamics are formed can have different behaviours.
With these elements, it is possible to conjecture about the possible origin
of the shock unsteadiness. This is summarized in the following table.
Table 11.1
Flow Phenomenon Frequency Normalized Order of

frequency magnitude
Transonic Acoustic
interaction coupling (a2 − u2 )/ δ

Φ(M∞ − 1) < 10−2
Separated, Mass
Supersonic conservation
U∞
h
F (Mc ) g(r, s) d
h
F (Mc ) g(r, s) < 10−1
Non separated Eddy convection U∞ /d 1 1
Three possible sources of shock motion are listed. The first one is the in-
coming turbulence. This is the dominant phenomenon when the flow remains
attached. The dominant frequency is the frequency of passage of the large
eddies of the incoming boundary layer; it is of the order of U∞ /d (U∞ is
the external velocity, δ the incoming boundary layer thickness). The second
phenomenon under consideration is the flapping of the separated bubble.
This corresponds to the previous analysis, and results from mass conserva-
tion principle applied to the entrainment of air by the mixing layer of the
separated zone. The resulting frequency is Uh∞ F (Mc ) g(r, s), where h is the
height of the recirculating zone, Φ(Mc ) is the normalized spreading rate of
the compressible mixing layer, and g(r, s) is a weak function of velocity and
density ratios across this mixing layer. The last case examined here is the
acoustic coupling, as it can exist in buffeting or buzz phenomena. The dom-
inant frequency is produced by an acoustic feedback loop with a speed of
propagation equal to a − u, where a is the speed of sound and u the velocity.
If λ is the characteristic length of the problem, the resulting frequency is of
the order of (a − u) /.
These frequency scales are compared to the incoming turbulence and there-
fore are divided by U∞ /d for normalization. There are some approximations
to arrive at the expression for acoustic coupling. The orders of magnitude are
given in the last column. In most separated cases δ/h is order of 1. Φ(Mc ) < 1,
and for flows around M = 2, it is about 0.2; g(r, s) is about 0.2. The result is
that this ratio is less than 0.1. For the acoustic coupling, the ratio δ/λ is of
the order of the rate of spatial growth of the boundary layer, typically 10−2 .
The Mach number under consideration is here generally slightly larger than
11.3 Lessons Learned and Open Issues 319
1, so that the order of magnitude of the normalized frequency is less than

10−2 . The shock wave can respond to all the possible solicitations. However,
an interesting feature in this classification is that it shows clearly that there
is not a unique source of unsteadiness in shock boundary layer interactions,
and that in all cases, the dominant frequency of the shock motion remains
lower than the characteristic frequency of incoming turbulence or equal to it,
and in some cases smaller by two orders of magnitude.
The results of UFAST and the resulting analysis have brought insight into
the physics of the interactions and of the shock unsteadiness. They have
contributed to propose correlations about the predominant frequency of the
shock beating. Further processing of the large amount of data obtained during
the UFAST program can still be performed and represent an amount of data
for developing practical correlations. Nevertheless, some open issues have still
to be explored in the future.
It has been mentioned that in interactions in channels, the corner inter-
actions may have an overwhelming influence, and this aspect is not totally
documented. Moreover, in the case of forced oscillations, it is clear that for
the details of the shock properties, in particular for the motion of the foot of
the lambda shock, elements of understanding are still needed. The previously
quoted analyses refer to two-dimensional situations. Therefore, it would be
interesting to explore three-dimensional interactions to extend our knowledge.
Another point is that if the unsteadiness depends on mass conservation,
any modification in the mass budget could bring some changes: mass bleeding,
suction or modifications of density.
Another point seems to be the influence of the Reynolds number. Some
results have confirmed, in the case of the shock reflection, that the unsteadi-
ness frequency does not depend on Reynolds number, but on the state of the
interaction, separated or not. However, the separation itself depends on the
upstream boundary layer and in particular on its state of turbulence. Perhaps,
a more detailed investigation of incoming conditions would be necessary. This
can be related to another issue. In many cases, it has been shown beneficial
to compute the whole span of the interaction, including side walls, even in
nominally 2-d interactions. In some cases, computations started upstream of
the sonic neck. If interactions are sensitive to the state of turbulence of the
incoming flow, it would be interesting to compute the flow from the settling
chamber of the test sections. It is known that obstacles placed at the wall in
a settling chamber can modify wall friction even several meters downstream,
in the supersonic part of the nozzles where experiments are installed. Prob-
ably, the injection systems used for LDA or PIV measurements should be
considered as intrusive as far as they bring a perturbation to the wall which
is felt far downstream. This may contribute to understand the differences
on the onset of separation found in different tunnels. Therefore, it would
be interesting and useful in the future to design experiments in which some
measurements could be performed in the settling chamber to characterize the
boundary layers in this part of the facilities.
Finally, as far as Reynolds number issues are concerned, an important one

is found when the boundary layer negotiating the interaction is transitional,
producing eddies with particular dynamics. Detailed description and analysis
in this case seems badly needed, and could be the opportunity to investigate
basic problems with important applied consequences.
Chapter 12
WP-3 Flow Control Application
(Holger Babinsky)
12.1 Introduction
As part of the UFAST project, eight partners investigated the effect of flow
control on unsteady shock-induced separation in a wide variety of flow sit-
uations. Before commenting on their results it is useful to first consider the
types of unsteadiness investigated within the framework of UFAST. As shown
in Figure 12.1 we propose to define three categories of shock unsteadiness.
Fig. 12.1 Types of SBLI unsteadiness investigated within UFAST
There was also a wide variety of flow control methods, and these are listed
in Table 12.1.
Most control methods investigated within UFAST fall into the category
of boundary-layer control. In this class of flow control techniques the aim
is to improve boundary-layer health upstream of an adverse pressure gradi-
ent in order to delay separation. Most methods achieve this by introducing
streamwise vorticity or enhancing the turbulence in the near-wall flow so
that a typical controlled boundary layer exhibits increased mixing and fuller
velocity profiles with a reduced shape factor. These control methods are ex-
pected to influence the disturbances in an incoming boundary layer directly,
as well as affecting the dynamics of the shock wave boundary layer interaction
springerlink.com
322 12 WP-3 Flow Control Application (Holger Babinsky)
Table 12.1 Flow control methods employed within UFAST
Type of control Institution

Local/boundary- Vortex generators (VGs) UCAM, ONERA (DAFE)
layer controls Air jet VGs IUSTI, IMP-PAN
Synthetic/pulsed jets INCAS, QUB
Suction IMP-PAN
Electric Discharge ITAM
Global controls Oscillating flap IOA
through a reduction and/or elimination of shock induced separation. While

this reduction in separation can in turn lead to a different global flow, and
thus affect downstream pressure perturbation, the primary effect of boundary
layer flow control on shock wave unsteadiness is through local and viscous
disturbances (in the context of Figure 12.1).
One research group (IoA) also investigated a more ‘global’ flow control
device, namely an oscillating trailing edge flap on a transonic aerofoil. This
method directly affects pressure disturbances originating downstream of the
shock wave. Here, no active feedback was employed and it is therefore thought
that the main effect of this control method was to introduce defined distur-
bances rather than damping naturally existing pressure fluctuations through
some active process.
12.2 Main Observations

The lessons learned within UFAST can be broken down into two groups, as
follows:
Effect of flow control on shock-induced separation
All research partners investigating boundary layer controls (with the excep-
tion of the electric discharge studied by ITAM) reported that separation was
reduced by the application of control. Here, bleed/suction was found to be the
most effective control method, being able to completely eliminate separation
in some circumstances (IMP-PAN). Apart from bleed, all other techniques
were of a three-dimensional nature (flow control was applied at discrete span-
wise locations) and this was reflected in the observed effects on separation.
Here, most partners reported that separated regions were reduced in size and
their shape was changed. In several cases, separated regions were broken into
cells of separated flow surrounded by fully attached areas. As an example,
Figure 12.2 shows a typical result from IMP, where it can be seen that air-
jet VGs almost completely eliminated shock induced separation, except for
a number of regular, but small, ‘cells’ of separated flow.
12.2 Main Observations 323
Similar results were reported by other groups. No ideal compromise be-

tween control size or strength and control effectiveness was found. While
larger and stronger devices were found to be more effective at reducing sepa-
ration, it was also observed that these devices would incur greater penalties
in the form of additional device drag or activation power.
One intriguing observation emerged in a number of experiments where it
could be seen that whenever flow control was able to reduce separation in
the centre of the channel, corner effects appeared to become more prominent
(see Fig.2). Furthermore, no partner reported a reduction of corner separation
through flow control. However, it should be noted that no control methods
were employed to specifically deal with such corner interactions. The problem
of 3-dimensionality and corner effects was not a focus of the UFAST inves-
tigations and therefore it is still too early to make general statements but
nevertheless it is thought that the interaction of corner and centreline flow
separation and its control would make an important topic for future research.
Fig. 12.2 Normal SBLI controlled by air-jet VGs, uncontrolled flow at top. It can
be seen that separation is almost completely eliminated but small, regular regions
of attached flow remain (IMP-PAN)
Effect of flow control on unsteadiness
The introduction of controlled perturbations upstream of an impinging

oblique shock wave interaction through electric discharge actuators by ITAM
demonstrated that shock wave unsteadiness responded to such upstream dis-
turbances. In particular a clear inverse relationship between shock oscillation
amplitude and disturbance frequency was observed as seen in Figure 12.3.
Fig. 12.3 Amplitude vs frequency for shock oscillations induced by controlled

disturbances introduced upstream of an impinging shock interaction (ITAM)
It is noteworthy that this oscillation amplitude response to disturbances

introduced upstream, inside the boundary layer, mirrors the behaviour ob-
served in driven transonic shock experiments where the shock oscillation is
caused by controlled pressure perturbations introduced downstream of the
shock in the inviscid flowfield (UCAM, ONERA-DAFE). This would suggest
that inviscid effects may well be the prime cause for the often observed inverse
amplitude-frequency relationship in shock wave oscillations.
In flow cases where global or combined unsteadiness (according to
Figure 12.1) was observed, flow control was often found to have little or no
direct impact (UCAM, IMP-PAN). In some cases boundary layer control was
found to reduce oscillation amplitudes by a small amount (ONERA-DAFE,

INCAS), but often this is better understood by the effect that flow control
has on flow separation. Reduced separation changes the global flowfield and
thus the shock wave position and this in turn affects shock oscillations. There
is little evidence that boundary layer control can directly affect the unsteady
response of a shock wave to disturbances.
Somewhat disappointingly, an investigation utilising a control method
specifically designed to affect the global flowfield, namely a trailing edge
flap on a transonic airfoil (IoA) has had little effect on shock oscillations seen
under buffet conditions. It was expected that such a control method, which
can change the pressure field in the inviscid flow would have a strong effect
on the dynamic behaviour of the shock wave. However, closer examination
revealed that the control frequencies were a factor of 3 away from the buffet
frequencies and it is thought that this frequency mismatch was the cause for
the lack of control effectiveness. Also, it might be necessary to employ active
feedback to achieve a significant effect. Future work will aim to increase the
achievable oscillation frequencies of the trailing edge flap to match the buffet
range.
12.3 Conclusions
As discussed in more detail in the closing remarks for WP2, SBLI unsteadi-
ness has a variety of causes. Generally, shock waves are seen to respond to dis-
turbances which range from small-scale/high-frequency (such as found in the
incoming turbulent boundary-layer) to more large-scale and low-frequency
(when unsteady flow separation causes larger changes in the flowfield or when
external inviscid pressure fluctuations are imposed). Shock wave oscillations
are seen to be receptive to such disturbances and oscillation amplitudes typ-
ically exhibit an inverse relationship with frequency.
Flow control applied to the boundary layer ahead of an unsteady SBLI
(modifying the properties of the incoming boundary layer) has little direct
effect on the response of shock waves to incoming disturbances. However,
where disturbances are generated or amplified as a result of shock-induced
separation, boundary layer control can reduce shock oscillation via an indirect
mechanism: flow control reduces flow separation and this in turn modifies the
magnitude and frequency of unsteady disturbances.
In a comparison of all types of flow control considered here, it is noted that
the more powerful methods also have the greatest effect. Particularly the well-
known traditional techniques of vortex generators and wall suction (‘bleed’)
perform well, with boundary layer suction being particularly effective. No
attempt has been made here to assess the adverse effects of flow control
devices (‘drag penalty’) but it is thought that there is likely to be an optimum
control size/strength which balances effectiveness with installation drag.
Many of the experiments performed in UFAST exhibited considerable 3-

dimensionality. Often it was observed that this increased when flow control
was employed and several groups expressed a view that their control method
was not as effective as expected because the 3-d effects were not specifically
controlled. In channel flows in particular it would be desirable to investigate
methods to favourably influence the corner separations and this is recom-
mended as an avenue for future research.
Chapter 13
WP-4 RANS/URANS Simulations
(Charles Hirsch)
13.1 Introduction
The UFAST project is centred on three main configurations with shock in-
duced separations: transonic flow around a profile; normal shock on a flat
plate or on a curved surface; and an oblique shock reflection, as summarized
in figure 13.1. For each case, the effect of control devices was also investigated.
Experimental data were compared to (U)RANS, DES, and LES simulations,
for all these configurations. This section is focused on the issues related to
(U)RANS simulations.
Fig. 13.1 The three basic configurations investigated within UFAST
The (U)RANS simulations did generate some expected results concerning

the turbulence models, but also some unexpected results related to corner
separation and non-symmetrical solutions, as well as a number of still unan-
swered questions.
We will cover successively the three basic UFAST configurations of
figure 13.1.
13.2 Transonic Flow Around a Profile

This covers two test cases, the Wall mounted circular arc bump (QUB ex-
periments) and the biconvex airfoil, tested at INCAS.
springerlink.com
328 13 WP-4 RANS/URANS Simulations (Charles Hirsch)
Wall mounted circular arc bump (QUB experiments)
This experiment was performed at the Queens University of Belfast (QUB) in

a transonic wind tunnel with a maximum Mach number of 1.4. Some issues re-
mained with the experimental data, as described in the synthesis of WP2, but
this was not critical to the URANS assessment. Figure 13.2 illustrates typical
results. Globally, the (U)RANS solvers are able to reproduce the observed
pressure distribution along the bottom wall. They produce a 3D structure of
the flow downstream of the shock that is similar to the one observed with
China Clay visualization.
However, the length of the recirculation region is overestimated when com-
pared to experimental data. Also, steady flow is predicted by many (U)RANS
models; the Spalart-Allmaras (SA) turbulence model has not been applied
on this test case, but similar results are anticipated.
Fig. 13.2 Some results for the QUB test case. Left: experimental schlieren view;
centre: China Clay visualization (QUB); right: k-ω URANS result of the separated
region from LIV.
The Biconvex airfoil tested at INCAS
This flow configuration, with an upstream Mach number of 0.76, generated

a spontaneous buffeting of the shock boundary layer interaction, as illustrated
in Figure 13.3, showing experimental schlieren pictures and the corresponding
URANS results, as obtained by EADS.
The URANS solvers are able to reproduce the observed buffeting phe-
nomenon with reasonable agreement for the amplitude, and produce a fre-
quency that compares well to experimental data. The experimental frequency
of shock motion was measured at 78 Hz; the URANS simulation from INCAS
predicts 80.1 Hz; the EADS prediction is 77 Hz; while IMFT predicts a fre-
quency of 79 Hz; all calculations having been performed with the SA model.
13.3 The Normal Shock Experiments 329
Fig. 13.3 Top: experimental schlieren pictures from INCAS; below URANS results
from EADS
13.3 The Normal Shock Experiments

Three configurations were tested: at University of Cambridge (UCED); at
IMP and at ONERA, and they are fully described elsewhere in this book.
They comprise forced as well as steady shock data, at several Mach numbers
in the range 1.3 to 1.5.
Nozzle forced shock oscillation (ONERA and CUED)
The oscillations are forced by the presence of a downstream cam at a fixed

frequency and separate URANS simulations of the cam geometry have shown
that they generate indeed a periodic pressure field. The Mach number range
is between 1.3 and 1.5 (depending on the shock position).
Typical results, as obtained by University of Liverpool (LIV) are shown
on figure 13.4.
The URANS predictions appear to globally predict the right shock motion,
for this largely pressure dominated flow field, although the amplitude of the
shock motion is slightly underpredicted, The URANS simulations show also
some deficiencies in predicting the size of the re-circulation region.
For the CUED experiments, similar conclusions appear: a comparison of
unsteady results for M = 1.3 and M = 1.4 shows that the global, inviscid
Fig. 13.4 URANS results from University of Liverpool (LIV), with a k-ω turbu-
lence model, showing the intensify of the shock motion
dominated, effects are well captured. For example Figure 13.5 shows a com-
parison of shock oscillation amplitudes for two frequencies with the experi-
mental data and with an analytical model, developed by Bruce and Babinsky
(2008), at M = 1.4. The amplitude of unsteady shock wave motion is observed
to decrease with increasing excitation frequency. Pressure measurements sug-
gest that the pressure rise across an unsteady SBLI region depends primarily
on the flow Mach number relative to the shock wave. This implies that the
velocity of shock wave motion can be determined analytically for a given
(varying) pressure ratio. Furthermore, given the pressure variation driving
shock wave motion, the shock wave trajectory can easily be calculated by
integrating the predicted shock wave velocities, thus yielding the amplitude
of shock wave motion. The agreement between analytical and experimental
results is good, supporting the concept that the shock wave motion is the
result of the shock wave adjusting its strength to satisfy an imposed varying
pressure ratio.
Figure 13.6 compares instantaneous surface pressure and flow fields at two
instants during an unsteady cycle, with the experimental data. Although the
numerical results follow the major trend quite well, particularly the shock
wave dynamics; the changes in size and shape of the lambda shock foot ap-
pear to deviate (particularly in terms of phase) from the experimental obser-
vations. This suggests that the inviscid effects of the enforced unsteadiness
are well captured by CFD, whereas the more detailed viscous effects are more
sensitive to the current deficiencies of turbulence models.
Fig. 13.5 Snapshot of shock motion, with k-ε URANS model (NUMECA), and
predicted shock amplitude, compared to the CUED experiments and the (inviscid)
analytical model of Bruce and Babinsky (2008)
Fig. 13.6 Comparison of unsteady CFD (NUMECA) and experimental results

(top right) from CUED at M = 1.4
The steady normal shock experiments
For the steady shock cases at M = 1.3 relatively good agreement between
CFD and experiments was observed, although the boundary layer growth
through the interaction was overpredicted. Among the turbulence models
tested, the SST model appeared to give the best agreement of the boundary
layer development; particularly the inflow velocity profile was well captured.
However, figure 13.7 shows that even in this ‘simple’ flow case, the turbu-
lence models, comparing SA, SST and k-e models, show global non-negligible
differences.
Fig. 13.7 Comparison of experimental (CUED) and numerical schlieren pictures

at M = 1.3, for three turbulence models (NUMECA)
Non-symmetrical solutions at M = 1.4 and 1.5
A large number of simulations were performed on the steady normal shock

experiments of CUED, but the main findings were related to the appearance
of non-symmetrical solutions, at M = 1.4 and 1.5, where the experiments
were globally symmetrical.
Figure 13.9 shows computed surface streamlines for six different turbulence
models, ranging from the algebraic Baldwin-Lomax (BL) model to the v2-f
four equation model.
It was finally demonstrated, by numerical investigations and confirmed by
experiments, that the main source of asymmetry was the presence of large
corner separated vortices. At M = 1.4, where the experiments are largely
symmetrical, the turbulence models tend to significantly overpredict the size
Fig. 13.8 Comparison of experimental (CUED) and numerical surface pressure at

M = 1.3, for three turbulence models (NUMECA)
Fig. 13.9 Demonstration of asymmetric results: CFD limiting streamlines along

the end-walls, for various turbulence models for the CUED experiment at M = 1.4,
from NUMECA
of these corner vortices, leading to an asymmetry. This is particularly the

case with all turbulence models, with the exception of k-ε.
Similar effects wee predicted with the IMP test cases. Tests have been
performed with different flow solvers (FINE/Turbo, FLUENT, SPARC) and
different turbulence models (SA, SST, RSM and k-τ ), all producing asym-
metrical solutions, with the exception of the k-τ turbulence. These tests from
IMP, ar summarized in figure 13.10.
Fig. 13.10 Asymmetrical solutions on the IMP experiment, with different codes
and TU-models, from IMP
Among the numerical experiments, tests have been performed by adding

corner chamfers. It appears that the non-symmetrical solution is suppressed
with a thick enough chamfer, confirming that the asymmetrical solution is
due to the corner flow separation.
The main findings can be summarized as follows; see also Hirsch and Tart-
inville (2009):
• an asymmetrical flow solution can be produced for Mach numbers 1.4
and 1.5.
• The asymmetry is not as important in the experiment but can be observed
at M = 1.5.
• The asymmetry depends on the turbulence model used and appears when
the mesh is refined.
• The structure of the asymmetry depends on the initial solution used. The
solver will tend to converge towards a symmetrical solution and then a bi-
furcation appears, leading to the non-symmetrical solution.
13.4 Oblique Shock Reflection 335
• The asymmetry has been reproduced with different flow solvers and with
different numerical schemes (2nd order upwind or central scheme).
• This was actually already observed by other authors.
In addition, none of the URANS simulations did reproduce the
local shock unsteadiness observed in the experiments.
13.4 Oblique Shock Reflection
Experiments were provided by TU Delft (M = 1.7), ITAM (M = 2) and

IUSTI (M = 2.25).
The ITAM experiments
Two turbulence models were applied to this test case by UAN with the SST
turbulence model whereas LMFA applied a SA turbulence model.
The upstream boundary layer was predicted with a moderate level of agree-
ment and the experimental data show an earlier separation and higher over-
shoot in pressure compared to the predictions.
Figure 13.11 shows a typical comparison with the experimental data and
also with the LES results from Univ of Southampton (SOTON).
The influence of side walls and free-stream turbulence intensity appears to
be important, and all the URANS models produce a steady solution.
Fig. 13.11 Axial velocity distribution upstream of the shock and pressure distri-
bution along the bottom wall. The dots are the experimental data compared to
RANS results from IAN and LMFA and the LES results from SOTON.
The IUSTI experiments
A large number of simulations were performed on this test case with different
turbulence models and different grid sizes, up to 23 million points.
Figure 13.12 summarizes some of the predictions obtained with grid sizes of
the order of 2 to 5 million points, showing the axial velocity distributions on
a vertical cut at mid-span. The variation and differences between the different
Tu-models can clearly be seen, in comparison to the PIV experimental data
from IUSTI.
Fig. 13.12 axial velocity distributions on a vertical cut at mid-span for the IUSTI
oblique shock experiment at M = 2.25
The RANS models are able do all reproduce the main physics, including
the size and shape of the recirculation zone. The turbulence models appear
however to have an important effect on the prediction of the size of the
recirculation zone.
The corner flow does also have an impact on the central recirculation
region, as seen from figure 13.13. Figure 13.13 shows indeed, a top view of the
recirculation region downstream of the shock on the wind tunnel floor. The
results are obtained with two different codes, on two very different grids, 4.6
and 23 million points respectively, but with the same SA turbulence models.
The main difference seems to be connected to the difference in size of the
corner vortices on both corners.
This interesting result indicates that more effort should have been done
on assessing grid effects.
As in the preceding test cases, all the URANS models produce a steady
solution and do not capture the high frequency oscillations at the foot of the
shock.
Fig. 13.13 Axial velocity distribution at 1 mm away from the bottom wall, for the
IUSTI experiment
13.5 Conclusions
All flow cases are characterized by strong three-dimensional and viscous ef-
fects, making them probably more challenging than initially anticipated.
The discovery of non-symmetrical solutions, connected to large corner sep-
arations is a major outcome of the UFAST project.
The RANS methods are able to capture most of the flow features, but
many shortcomings have been identified:
• URANS underestimate or are unable to produce natural shock motion
(from QUB, IUSTI, TUD, ITAM and IMP test cases).
• URANS are able to capture forced oscillations (from ONERA and CUED
test cases). Though the upstream velocity profile is well captured, the
prediction of the recirculation region and the downstream profile needs to
be improved.
• Asymmetrical flow solutions are produced on symmetrical configurations
(from CUED and IMP test cases), due to overprediction of the corner
separation, which appears to be sensitive to the turbulence model used
(from CUED, IMP and IUSTI test cases) and to the grid resolution. Due
to the three-dimensionality of the experiments, corner flows have an im-
pact on the mid-span recirculation, making the separation zone strongly
dependent on the turbulence model.
The major outcome of the (U)RANS simulations within UFAST, is that
the turbulence models need to be improved in order to better model 3D
separations and corner vortices, as well as for accurate capturing of shock
wave boundary layer interactions and related separation.
Also, current Tu-models appear as too dissipative, at least on the grids
tested, to capture the spontaneous unsteadiness of the shock boundary layer
interactions.
338 References
References
[1] Bruce, P.J.K., Babinksky, H.: Unsteady shock wave dynamics. J. Fluid
Mech. 603, 463–473 (2008)
[2] Bruce, P.J.K., Babinsky, H., Tartinville, B., Hirsch, C.: An experimental and
numerical study of an oscillating transonic shock wave in a duct. In: Proc. 48th
Aerospace Sciences Meeting, Orlando, AIAA Paper-2010-xxxx (2010)
[3] Hirsch, C., Tartinville, B.: RANS modeling for industrial applications and some
challenging issues. Int. J. CFD 23(4), 295–303 (2009)
Chapter 14
WP-5 LES and Hybrid RANS/LES
(George Barakos)
14.1 Motivation, Objectives and Work Share
This paragraph summarises the progress achieved during the UFAST project
on the 5th work-package that was devoted to the analysis of unsteady shock
boundary layer interaction using Large-Eddy Simulation (LES) as well as hy-
brid techniques of LES and Unsteady Reynolds-Averaged (URANS) models.
All three configurations of UFAST namely the transonic interaction, the nor-
mal shock, and the reflected shock were analysed along with a good range of
the flow control methods used for the various experiments. The work-package
was motivated by the need to gain deeper insight into the physics govern-
ing the unsteadiness of the shock, the shock/boundary layer interaction, the
development of buffeting, together with a study on efficient methods for con-
trolling these phenomena. In addition, LES methods were applied to resolve
the large coherent structures that govern Shock Wave – Boundary Layer In-
teraction (SWBLI) and consequently investigate the range of applicability
between RANS/URANS and LES.
The above broad motivation was distilled to the following set of objectives:
(i) the assessment of LES for simulating shock-induced separation for generic
flow geometries, (ii) the exploitation of LES and hybrid RANS/LES methods
to analyse complex SWBLI and suggest future model improvements, (iii) to
properly simulate physical phenomena, specifically time averaged flow fea-
tures, and unsteadiness, (iv) to explore the applicability of LES at increased
Reynolds numbers, (v) to put forward efficient hybrid RANS/LES methods
suitable for industrial use with lower CPU requirements and ability to re-
solve the SWBLI physics and (vi) investigate the potential of LES solutions
for calculating the production, dissipation, diffusion and other model terms
in the root mean square (RMS) quantities of the turbulence models used by
the partners of the URANS work-package.
A good range of partners contributed to this work-package including:
the Romanian Institute for Aeronautics (INCAS), the Institute of Fluid
Flow Machinery of Gdansk (IMP-PAN), the Deutschland GmbH Military
springerlink.com
340 14 WP-5 LES and Hybrid RANS/LES (George Barakos)
Aircraft (EADS-M), the Foundation for Research and Technology of Greece

(FORTH), the NUMECA CFD company of Belgium, the Office National
d’Etudes et Recherches Aérospatiales of France (ONERA), the University of
Southampton in the UK (SOTON), the University of Rome “La Sapienza”
(URLMS) of Italy, the University of Liverpool (ULIV) UK, and the Institut
de Mécanique des Fluides de Toulouse (IMFT), France.
The partners contributed a wide range of codes and turbulence simulation
techniques including several versions of LES, Organized Eddy Simulation
(OES), as well as variants of Detached Eddy Simulation (DES) and wall-
modelled flavours of LES. The different CFD codes were used with individual
numerical schemes. The work was well integrated with the work packages 2
and 3 that provided the basis for the evaluation of the work carried out during
the project although a direct contribution to the experiments was not possible
from the start of the project. A strong link with the RANS work straight from
the start of the project was evident since most partners contributed to both
work-packages.
Within WP5 all three shock/boundary layer interactions were studied:
transonic, normal shock and reflected shock. For most cases, flow control
devices were also included and simulations highlighted the potential of LES
for flow control studies. In the next paragraphs a consolidated view of the
work and lessons learnt during WP5 are presented.
14.2 Summary of Observations – Transonic Flow Cases

For the transonic flow over a bump that was studied experimentally at the
Queen’s University of Belfast (QUB), the main challenge was the near-wall
resolution. Since the Reynolds number for all cases considered was high, the
wall-modelled LES approach was used by ULIV. This hybrid method pro-
duced results showing some unsteadiness in the SWBLI, in fair agreement
with the experiments. These findings were further corroborated by results
using the OES method of IMFT and a plausible flow mechanism was pro-
posed. In contrast to URANS, the employed LES and OES methods resulted
in unsteady flow and resolved at least the fundamental shock motion fre-
quency. The averaged surface pressure distribution along the bump was also
well predicted. The CFD was also used to establish the range of applica-
bility of the Synthetic Jet control method in terms of jet frequency, even if
the actual implementation of the synthetic jet could not deliver the required
actuation.
A similar challenge was present for the two aerofoil test cases studied
in the tunnels of the IoA (flapped NACA0012) and INCAS (circular arc
aerofoil). Both of these cases were related to flow buffeting and the obtained
results depended upon the resolution of the flow around the aerofoil as well
as the influence of the wind tunnel walls. In addition, the onset of buffet was
found to be sensitive to the conditions of the calculation and the incidence
angle and Mach number had to be tuned to match the exact settings of the
14.2 Summary of Observations – Transonic Flow Cases 341
experiments. For these test cases, the Reynolds number based on the aerofoil
chord was of the order of 10 millions making the application of a wall-resolved
LES extremely costly in terms of CPU time. A different modelling approach
was taken for this case adopting Delayed DES (DDES). This approach was
justified by the good agreement that was obtained with experiments for the
buffet frequency and flight conditions. Figure 14.2 presents a snap-shot of the
Delayed DES simulations contributed by EADS-M for the INCAS test case.
The overall shock configuration and motion frequency were well-predicted by
Fig. 14.1 QUB test case: experiment, OES results of IMFT and averaged wall-
function LES results of ULIV for the isentropic Mach number over the bump
Fig. 14.2 Comparison between experiment (INCAS) and CFD (EADS-M) for the
flow configuration during buffet on the circular arc aerofoil
DDES. Looking further in the shock details, the DDES resulted in higher
instantaneous surface pressure peaks and more oscillations than URANS.
However, even for DDES the shock configuration was found to deviate from
test further away from surface.
The same challenges were faced for the IoA case where again the CFD
conditions had to match the experiment for the buffet onset. The presence
of walls, the high Reynolds number and the use of an actuated trailing edge
flap made this case far too challenging for LES. Again, the DES option was
exploited by IMFT and ULIV that contributed results for this test case. The
use of DES was also costly since full 3D computations including all walls of
the tunnel had to be undertaken. The fundamental buffet phenomenon was
well-resolved for this case (Figure 14.3). Given the high Re of this flow, the
Detached Eddy simulation method was the only viable option. The obtained
results suggested that DES was a good hybrid method for aerofoil cases since
DES performed well for this test case and buffet was demonstrated by both
partners contributing to this case (ULIV and IMF).
Fig. 14.3 Spectral analysis of the aerofoil loads at 50% of the chord for the IoA
case
The DES results obtained by IMFT show clearly the dominant buffet fre-
quency near 90 Hz.
A common theme for all cases simulated for the transonic interaction is
that the instrumentation of the experiments provided some measure of the
frequencies of the flow structures present near the interaction region. These
were mainly the outcome of the use of pressure transducers which although
enough for comparisons against URANS provided small spatial resolution for
comparing instantaneous data against the detailed flow-fields produced by
the CFD.
14.3 Summary of Observations – Normal Shock Cases

The second, and perhaps the most challenging, interaction case concerned
normal shocks that were studied in the wind tunnels of ONERA, IMP and
14.3 Summary of Observations – Normal Shock Cases 343
UCAM. ONERA forced a shock oscillation using a rotating shaft placed

downstream the tunnel test section. Due to the size of the shock, the ex-
tend of separation, the relatively high Reynolds number, and the need to
resolve all walls of the tunnel the DES approach was adopted by ULIV for
this test case. The forcing frequency was relatively low and was resolved us-
ing URANS. Further there was a clear separation between the frequencies
associated with large flow scales and smaller ones. This separation of scales
was the main reason URANS resolved some of the flow unsteadiness. The
DES results of Liverpool resulted in improved prediction of the separation
flow region, better agreement of the surface pressure spectra with experi-
ments and good overall prediction of the shock motion. Nevertheless, DES
was challenged by this case since some of the unsteady flow spectra suggested
that DES was under-resolving the high flow frequencies and perhaps more
work was needed to improve the near-wall predictions. Figure 14.4 shows
some of the results obtained by ULIV. The work with the DES continued for
the flow control cases where vortex generators were placed inside the tunnel
to control the flow. The DES was again capable of providing a good set of
results capturing most of the effect of the stream-wise vortices on the shock
and the separated flow region.
Fig. 14.4 The ONERA test case, (a) experiment, (b) surface pressure with vortex
generators present and (c) DDES results for the surface flow behind the vortex
generators
The case of IMP included two configurations: shock in straight and shock
in curved channel. For these cases the flow unsteadiness was natural while
a shock configuration with forced unsteadiness was investigated by UCAM.
In contrast to the case studied by ONERA, and the curved IMP channel,
the straight channel configurations were designed to have a constant up-
stream Mach number. Remarkably, the straight channel configurations were
the hardest cases to analyse. This was due to the effort needed to obtain
the correct boundary conditions for CFD as well as due to the presence
of flow asymmetries in the CFD solutions. The requirement to resolve four
walls around the interaction and the close proximity of the interaction region
with corner vortices present near the vertical walls of the tunnels made these
cases extremely difficult for LES. The required CPU resources were in excess
of what was available to any of the partners and attempts to use span-wise
symmetry conditions did not result in the correct shock configuration. For
this reason, fewer studies were undertaken using Large-Eddy Simulation. The
attempt of ULIV used a zonal LES approach with wall-functions used for the
side walls of the tunnel to avoid excessive grid densities required for the near-
wall region. A similar hybrid method based on DES was attempted by IMP
using the CFD solver of NUMECA. Some shock unsteadiness was predicted
by the CFD solutions though the flow was relatively under-resolved and the
comparison against the experiments was not favourable. The lack of symme-
try in the solution required the use of long simulation times so that some
average to be reached (Figure 14.5). Although a relatively larger separated
flow region was achieved in comparison to URANS cases, the lack of flow
symmetry prevented the partners from performing detailed investigations of
this test case. Clearly, further work is needed to clarify the origin of the
asymmetries in the CFD solutions as well as to compare the shock motion
with the test data. No further insight in this interaction was obtained from
using LES and hybrids for this test case.
Fig. 14.5 Experiment and CFD results for the separated flow region for the flat
channel of IMP case (M145 case). The numerical solution was obtained by IMP
using the DES model of the NUMECA CFD solver. Flow direction is left to right.
Computing the straight channel cases was a humbling experience since no

2D approximation or span-wise symmetry could be applied. The influence
of the wall was not possible to ignore and overall the LES computations
were outside today’s capability. The hybrid methods again provided some
insight but these were limited by the lack of ability to resolve the corner flows
and the flow control devices. These deceptively simple cases were by far the
hardest to compute even with DES. The case by Cambridge was computed by
NUMECA without hard evidence of shock unsteadiness. The results showed
flow asymmetries and did not correlate well with the experiments. The control
was not attempted from any of the CFD partners working on this case.
14.4 Summary of Observations – Reflected Shock Cases 345
14.4 Summary of Observations – Reflected Shock

Cases
The reflected shock cases were certainly the highlight of the UFAST project
and the WP5. In total three test cases were contributed by TUD, ITAM and
USTI. This set of cases was computed by most WP5 partners and covered
a wide range of Reynolds and Mach numbers. The case studied by the TUD
experiments was perhaps the most challenging due to the high Re and the
intermittently separated flow that occurred near the interaction region. The
CFD work highlighted the complex physics of the phenomenon and the strong
3D character of the flow. For this test case, most partners attempted an
LES using a periodic domain in the span-wise direction and at lower Re
than the experiment. The results were encouraging, suggesting that with
more computer resources and better near-wall resolution the flow could be
computed with LES. The results also suggest that LES definitely gives the
correct physics of the turbulence amplification near the interaction and the
qualitative structure of the interaction was well-predicted in terms of mean
and statistical properties. The presence of shedding of coherent vortices in
the mixing layer was confirmed.
Fig. 14.6 Experimental and LES resuls for the TUB shock interaction case. Maps
of streamwise velocity component are shown.
For the experiments of ITAM, LES showed low-frequency motions even

with careful choice of upstream conditions that suggests an internal bubble
mechanism in the interaction region and shows some evidence for an unstable
global mode. Wave number-frequency analysis showed acoustic and convec-
tion processes near the separation bubble along with weak upstream acoustic
propagation and convection, combined with downstream acoustic radiation.
Excellent comparison of the shock-foot location probability was obtained by
the LES results of SOTON as shown in Figure 14.7.
The IUSTI case was by far the most popular case in WP5 and was com-
puted by many partners. This generated a wealth of simulation data and
allowed for comparison of codes and techniques. In total, DES, DDES, SDES
and LES were used for this test case. The lower Re of the experiment al-
lowed for moderate CPU time requirements. Interestingly for this case most
partners found that the classic (SA) DES and DDES failed to predict the
correct flow physics and resulted in steady-state flows for coarse grids. For
this case, the effect of the tunnel walls was less important and for this reason
Fig. 14.7 ITAM test case. LES results of SOTON compared with the experiments
for the shock foot location.
several cases were computed with span-wise expansion boundary conditions.

The LES was the approach to select for the quasi-2D flow where the shock
generator angle was kept at 8 degrees. The Stimulated DES (SDES) tech-
nique of ONERA was also a good, and cheaper, alternative resulting in fair
predictions of the overall flow configuration.
These two techniques were able to complement the experiment for the
analysis of the unsteady data since they provided time resolved information
of thousands of sensors simultaneously. Results from this effort for the stream-
wise velocity component are shown in Figure 14.8.
Fig. 14.8 Comparison between SDDES of ONERA and experiments of IUSTI for
the 8-degree case. Maps of stream-wise velocity component are shown
On the flow control front, several ideas were put forward by the experi-
mentalists including the use of air jet vortex generators, synthetic jets, con-
ventional vortex generators, flaps and porous walls. Due to the complexity of
the computations using LES for these cases relatively less work was carried
out with some devices left out of simulations altogether. For certain cases,
the WP5 partners raised to the challenge of simulating these complex de-
vices and resolving the flow physics. For all computed cases, the results were
encouraging giving deeper insight in the control aspects of SWBLI. Several
issues were identified and new challenges were set for local resolution near
the devices, modelling of devices and strategies for assessment of flow control
14.5 Summary of Conclusions and Suggestions for Further Work 347
Fig. 14.9 Flow visualization for the ONERA test case. Computations of ULIV
fort the flow control case of co-rotating conventional vortex generators. Stream-
wise velocity contours on vorticity iso-surfaces. Flow direction is right to left.
effects. Figure 14.9 presents results from the ULIV efforts to resolve the flow
around vortex generators with DES for the experiments of ONERA.
14.5 Summary of Conclusions and Suggestions for

Further Work
The cases computed as part of work-package 5 were certainly demanding and
some of these were at the edge of what is currently possible with modern com-
puters. For the cases where the Reynolds number was within the capacity of
current computers a full, wall-resolved LES solution was possible even if some
of the span-wise extend of the computational domain had to be simplified.
Such computations were mainly performed for the reflected shock cases. As
the Reynolds number increased, the cases had to be computed using either
some wall-modelled LES, this was the case for some of the transonic flow
cases or even DES and hybrids. For the most complex cases where buffeting
and full aerofoils and flapped sections had to be studied, DES was the only
realistic option available to the UFAST partners.
Overall, WP5 met its objectives by performing some very challenging com-
putations. As part of this work package some remarkable examples of SWBLI
were computed that offered improved predictions with respect to URANS for
the vast majority of cases. The high Re of several cases resulted in the use of
hybrid methods based on several approaches like the Detached Eddy Simu-
lation, the Organised Eddy Simulation as well as several wall-modelled LES
methods. The variation of the methods provided insight in the problems at
hand and the employed methods. All the attempts with hybrid models show
clearly that there is a demand for more simulation and less modelling in
CFD and this can today be achieved via the hybrid methods that represent
attempts to take LES further towards conditions of practical applications.
In addition to the work carried out in WP5 several flow challenges were
identified. The first one is associated with the resolution of corner flows.
Since most cases were computed in tunnels where walls influence the flow,
further insight in the corner flow physics is needed. For the UFAST cases,
the corner flows made the SWBLI more 3D and did not allow the use of
span-wise symmetry conditions and expansions to allow for economies in
CPU time and mesh points. Clearly, detailed experiments looking at corner
flows are needed and these should be combined with LES studies with wall-
resolved flow physics. The study of corner flow should be made with special
experiments and LES, designed to complement each other. A second challenge
is associated with the need to provide wall-resolved LES for higher and higher
Re numbers. This objective can be achieved with the use of larger massively
parallel computers, Graphics Processing Units and faster algorithms as well
as Grid Computing. The current methods appear capable, but their efficiency
must be improved. Since we cannot just wait for computer power to become
available, the use of databases of computations could be an option so that
maximum understanding can be generated by sharing existing data. The use
of hybrid and zonal methods proved popular within the WP5 of UFAST
and some of the employed methods were quite generic rather than tuned
for a particular case with the main question associated with their range of
applicability that is not known a-priori for most cases. As always, clarity
about the details of the methods and their capabilities and limitations is
necessary to allow for correct interpretation of the results. Unfortunately,
it appears that there was no way of an a priori estimate of the suitability
of hybrid methods for each test case and the use of experiments and wall-
resolved LES as well as method-to-method comparisons are still needed in
this areas.
The extraction of understanding from LES and hybrid methods was also
an area of active research during the WP5 of UFAST. In reality, one has to
generate insights from the obtained data and this brings forward the issues
of data post-processing of large data sets. Proper Orthogonal Decomposition
was a popular compression method in UFAST though it appears that no
universal solution to the problem exists. The generation of insight was also
assisted via partner-to-partner comparisons as well as complementary stud-
ies carried by two or more partners independently. The LES work was not
available early enough in the project to provide any input to the RANS and
URANS models employed in WP4. For this reason it was not possible to look
at extracting turbulent flow statistics from LES to be used for comparisons
with URANS. Clearly this should be addressed as part of a future project. In
addition, areas that need attention include investigations to challenge the as-
sumption of local equilibrium in URANS, the development of new anisotropic
stress tensor ideas, and of course more and better LES models.
14.5 Summary of Conclusions and Suggestions for Further Work 349
The work of WP5 is well-presented in the database of UFAST, has been

disseminated to several conferences and workshops and has generated ideas
for follow-on research and projects. Due to the complexity of the available
flow cases and the cost of the LES and hybrid method calculations, more is to
be done in the flow control area and further CFD study and experimentation
is needed. The modelling and resolution of flow control devices puts forward
new challenges for turbulence simulation and hybrid RANS/LES methods.

2011 Book UnsteadyEffectsOfShockWaveIndu

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

2011 Book UnsteadyEffectsOfShockWaveIndu

Uploaded by

Copyright:

Available Formats

114 Notes on Numerical Fluid Mechanics and Multidisciplinary Design (NNFM)

Piotr Doerffer, Charles Hirsch,

ISBN 978-3-642-03003-1 e-ISBN 978-3-642-03004-8

Notes on Numerical Fluid Mechanics

Library of Congress Control Number: 2010937341

Printed on acid-free paper

Prof. Dr. Wolfgang Schröder Prof. Dr. Sergio Pirozzoli

Prof. Dr. Michael A. Leschziner

Volume 79: B. Schulte-Werning, R. Gregoire, A. Malfatti, G. Matschke (eds.): TRANSAERO - A Euro-

1 Introduction – UFAST Project Overview . . . . . . . . . . . . . . . . 1

Part I Transonic Interactions

2 Bump at a Wall (George Barakos) . . . . . . . . . . . . . . . . . . . . . . . 13

3 Biconvex Aerofoil (Stefan Leicher) . . . . . . . . . . . . . . . . . . . . . . . 55

4 NACA0012 with Aileron (Marianna Braza) . . . . . . . . . . . . . . 101

Part II Nozzle Flows

5 Nozzle Forced Shock Oscillations with Wall Bump

6 Nozzle Forced Shock Oscillations (Holger Babinsky) . . . . . 163

7 Natural Shock Unsteadiness in Nozzle

Part III Shock Reﬂection

8 Oblique Shock Reﬂection at M = 1.7

8.4 Overview of LES Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . 238

9 Oblique Shock Reﬂection at M = 2.0 (Neil Sandham) . . . 263

10 Oblique Shock Reﬂection at M = 2.25 (Eric Garnier) . . . . 287

Part IV Summary by Work Packages

11 WP-2 Basic Experiments (Jean-Paul Dussauge) . . . . . . . . . 315

12 WP-3 Flow Control Application (Holger Babinsky) . . . . . . 321

13 WP-4 RANS/URANS Simulations (Charles Hirsch) . . . . . 327

14 WP-5 LES and Hybrid RANS/LES (George Barakos) . . . 339

Partners Responsible for the Experiments

1.1 Project Summary

1.2 Project Objectives

turbulent scales in respect of the unsteadiness and compressibility eﬀects

The ﬁrst objective of the UFAST project was to provide a comprehensive

shock movement and turbulence ampliﬁcation/decay at the shock wave. This

1.3 List of Participants

Table List of partner organisations in UFAST project

Part. no. Participant name Participant short name Country

1.4 Project Structure

Transonic interaction Nozzle ﬂow Oblique shock reﬂection

Fig. 1.1 UFAST conﬁgurations of ﬂow with shock waves

The selection of three conﬁgurations implied a high number of ﬂow cases.

Fig. 1.2 Graphical presentation of work program

1.5 Project Organisation

In order to secure the exchange of information obtained from experiments

Flow cases structure:

Shock/boundary layer interaction at transonic ﬂow conditions had been the

2.2 Introduction – The QUB Tunnel Setup

2.3 Contoured Upper Wall

and result in a good transonic interaction. It has to be clariﬁed, however,

M∞ Mmax % shock Max Mach Height of supersonic

2.4 CFD Tools and Mesh Generation

Table 2.2 Summary of grids for the QUB test case

Control Domain Size QUB Porous QUB Contoured

2.5 Boundary Conditions

2.6 Experimental Findings

Fig. 2.15 Isentropic Mach number measured at ‘dry’ conditions

2.7 URANS Results

near-wall plane showing substantial ﬂow distortion. As discussed in section 4,

More URANS results were obtained by IMFT using several turbulence

2.8 Flow Control Attempts

Fig. 2.28 Example of synthetic jet actuator employed for preliminary

Fig. 2.29 Vortices generated by the synthetic jet

2.9 LES and OES Results