Design and Development of Aerospace Vehicles and Propulsion Systems

Lecture Notes in Mechanical Engineering
S. Kishore Kumar
Indira Narayanaswamy
V. Ramesh Editors
Design and
Development
of Aerospace Vehicles
and Propulsion Systems
Proceedings of SAROD 2018
Series Editors
Francisco Cavas-Martínez, Departamento de Estructuras, Universidad Politécnica
de Cartagena, Cartagena, Murcia, Spain
Fakher Chaari, National School of Engineers, University of Sfax, Sfax, Tunisia
Francesco Gherardini, Dipartimento di Ingegneria, Università di Modena e Reggio
Emilia, Modena, Italy
Mohamed Haddar, National School of Engineers of Sfax (ENIS), Sfax, Tunisia
Vitalii Ivanov, Department of Manufacturing Engineering Machine and Tools,
Sumy State University, Sumy, Ukraine
Young W. Kwon, Department of Manufacturing Engineering and Aerospace
Engineering, Graduate School of Engineering and Applied Science, Monterey,
CA, USA
Justyna Trojanowska, Poznan University of Technology, Poznan, Poland
Lecture Notes in Mechanical Engineering (LNME) publishes the latest develop-
ments in Mechanical Engineering—quickly, informally and with high quality.
Original research reported in proceedings and post-proceedings represents the core of
LNME. Volumes published in LNME embrace all aspects, subfields and new
challenges of mechanical engineering. Topics in the series include:
• Engineering Design
• Machinery and Machine Elements
• Mechanical Structures and Stress Analysis
• Automotive Engineering
• Engine Technology
• Aerospace Technology and Astronautics
• Nanotechnology and Microengineering
• Control, Robotics, Mechatronics
• MEMS
• Theoretical and Applied Mechanics
• Dynamical Systems, Control
• Fluid Mechanics
• Engineering Thermodynamics, Heat and Mass Transfer
• Manufacturing
• Precision Engineering, Instrumentation, Measurement
• Materials Engineering
• Tribology and Surface Technology
To submit a proposal or request further information, please contact the Springer
Editor of your location:
China: Dr. Mengchu Huang at [email protected]
India: Priya Vyas at [email protected]
Rest of Asia, Australia, New Zealand: Swati Meherishi
at [email protected]
All other countries: Dr. Leontina Di Cecco at [email protected]
To submit a proposal for a monograph, please check our Springer Tracts in
Mechanical Engineering at http://www.springer.com/series/11693 or contact
[email protected]
Indexed by SCOPUS. All books published in the series are submitted for
consideration in Web of Science.
More information about this series at http://www.springer.com/series/11236

S. Kishore Kumar Indira Narayanaswamy
• •
V. Ramesh
Editors
Design and Development

of Aerospace Vehicles
and Propulsion Systems
Proceedings of SAROD 2018
123
Editors
S. Kishore Kumar Indira Narayanaswamy
Gas Turbine Research Establishment M. S. Ramaiah University of Applied
Bengaluru, Karnataka, India Sciences
Bengaluru, Karnataka, India
V. Ramesh
National Aerospace Laboratories
Bengaluru, Karnataka, India
ISSN 2195-4356 ISSN 2195-4364 (electronic)

ISBN 978-981-15-9600-1 ISBN 978-981-15-9601-8 (eBook)
https://doi.org/10.1007/978-981-15-9601-8
© Springer Nature Singapore Pte Ltd. 2021
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, 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.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Contents
Mach Number Effect on Aeroacoustic Characteristics of Compressible

Jet Due to Chevron . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
S. R. Nikam and S. D. Sharma
Parametric Study of Turbulent Flow Past a Compression–
Decompression Ramp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Rhea George and Raj Kiran Grandhi
Validation of Numerical Analysis Results for Pusher Configured
Turboprop Engine Air Intake . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
C. A. Vinay and S. Bhaskar Chakravarthy
Aero-elastic Analysis of High Aspect Ratio UAV Wing—Based
on Two-Way Fluid Structure Interaction . . . . . . . . . . . . . . . . . . . . . . . . 37
Vidit Sharma and S. Keshava Kumar
Numerical Study of Effect of Adjacent Blades Oscillation
in a Compressor Cascade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Shubham, M. C. Keerthi, and Abhijit Kushari
Effect of Incoming Wakes on Losses of a Low-Pressure Turbine
of a Gas Turbine Engine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Vishal Tandon, Gopalan Jagadeesh, and S. V. Ramana Murty
Effect of Axial Location on the Performance of a Control Jet
in a Supersonic Cross Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Raj Kiran Grandhi and Arnab Roy
Effect of Chord Variation on Subsonic Aerodynamics of Grid Fins . . . . 105
Manish Tripathi, Mahesh M. Sucheendran, and Ajay Misra
Numerical Investigation on the Effect of Propeller Slipstream
on the Performance of Wing at Low Reynolds Numbers . . . . . . . . . . . . 129
K. Shruti and M. Sivapragasam
v
vi Contents
Theoretical Design and Performance Evaluation of a Two-Ramp

and a Three-Ramp Rectangular Mixed Compression Intake
in the Mach Range of 2–4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Subrat Partha Sarathi Pattnaik and N. K. S. Rajan
Sensitivity of Altitude Variation on Aerodynamics of a Typical
Launch Vehicle During Hot Separation . . . . . . . . . . . . . . . . . . . . . . . . . 167
Jiju R. Justus, Sanjoy Kumar Saha, and Pankaj Priyadarshi
Normal Shock Dynamics in Internal Supersonic Flows . . . . . . . . . . . . . 177
S. Vaisakh and T. M. Muruganandam
Hinge Moment Characterization of All Movable Control Surface . . . . . 185
Manoj Kumar, G. Kadam Sunil, V. Shanmugam, and G. Balu
3D Computational Studies of Flapping Wing in Frontal Gusty Shear
Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
M. De Manabendra, J. S. Mathur, and S. Vengadesan
Investigation of Wind Tunnel Blockage Effect on Liftoff Aerodynamics
of a Launch Vehicle Through Open-Source CFD Solver SU2 . . . . . . . . 211
Aaditya N. Chaphalkar, Amit Sachdeva, Vinod Kumar,
and Pankaj Priyadarshi
Observation of Low-Frequency Shock Oscillation Over
a Forward-Facing Step . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
Jayaprakash N. Murugan and Raghuraman N. Govardhan
RANS Computations of Hypersonic Interference Heating on Flat
Surface with Protuberances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
M. Mahendhran and C. Balaji
Multi-fidelity Aerodynamic Optimization of an Airfoil
at a Transitional Low Reynolds Number . . . . . . . . . . . . . . . . . . . . . . . . 239
R. Priyanka, M. Sivapragasam, and H. K. Narahari
Aerodynamic Optimization of Transonic Wing for Light
Jet Aircraft . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
K. Sathyandra Rao, M. Sivapragasam, H. K. Narahari,
and Aneash V. Bharadwaj
Non-adiabatic Wall Effects on Transonic Shock/Boundary Layer
Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
Sahil Bhola and Tapan K. Sengupta
An Adjoint Approach for Accurate Shape Sensitivities in 3D
Compressible Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
Srikanth Sathyanarayana, Anil Nemili, Ashish Bhole,
and Praveen Chandrashekar
Contents vii
Robust Flutter Prediction of an Airfoil Including Uncertainties . . . . . . . 305

A. Arun Kumar and Amit Kumar Onkar
Effect of Vortex Generator on Flow in a Serpentine Air Intake
Duct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
B. B. Shivakumar, H. K. Narahari, and Padmanabhan Jayasimha
Supersonic Flow Behavior in Cartridge Starter . . . . . . . . . . . . . . . . . . . 331
Ritesh Gaur, Suparna Pal, Vimala Narayanan, D. Kishore Prasad,
and N. Balamurali Krishnan
High-Speed Shadowgraph Visualization Studies of the Effectiveness
of Ventilating a V-Gutter Flame Holder to Mitigate Screech
Combustion Instability in an Aero-Gas Turbine Afterburner . . . . . . . . 343
C. Rajashekar, Shambhoo, H. S. Raghukumar, R. M. Udaya Kumar,
K. Ashirvadam, and J. J. Isaac
Passive Reduction of Aerodynamic Rolling Moment for a Launch
Vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
Pankaj Priyadarshi, Amit Sachdeva, and Leya Joseph
Design and Development of Miniature Mass Flow Control Unit
for Air-Intake Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
D. B. Singh, P. Vinay Raya, Buddhadeb Nath, N. Srinivasan,
Anju Sharma, and B. Sampath Rao
The Effect of Variable Inlet Guide Vanes on the Performance
of Military Engine Fan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
Baljeet Kaur, Reza Abbas, and Ajay Pratap
Transition Prediction for Flow Over a MAV Wing Using
the Correlation Based Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
M. B. Subrahmanya and B. N. Rajani
Rotor Flow Analysis in the Presence of Fuselage Using Unsteady Panel
Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
K. R. Srilatha, Premalatha, and Vidyadhar Y. Mudkavi
Numerical Analysis of High Reynolds Number Effects
on the Performance of GAW-1 Airfoil . . . . . . . . . . . . . . . . . . . . . . . . . . 425
D. S. Kulkarni and B. N. Rajani
Investigation of the Effect of Booster Attachment Scheme
on the Rolling Moment Characteristics of an Asymmetric
Vehicle Using CFD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
P. K. Sinha, Munish Kumar Ralh, Naveed Ali, R. Krishnamohan Rao,
and G. Balu
viii Contents
Multi-objective Optimization Approach for Low RCS Aerodynamic

Design of Aerospace Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
P. S. Shibu, Sandeep, Balamati Choudhury, and R. U. Nair
Analysis of Propeller by Panel Method for Transport Aircraft . . . . . . . 457
Premalatha, K. R. Srilatha, and Vidyadhar Y. Mudkavi
Effect of Reynolds Number on Typical Civil Transport Aircraft . . . . . . 471
Vishal S. Shirbhate, K. Siva Kumar, and K. Madhu Babu
Prediction of MultiStore Separation from a Fighter Aircraft
Using In-House Code—WISe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
S. Karthik, Jishnu Suresh, P. Karthikeya, S. Rajkumar, Mano Prakash,
Sashi Kiran, and D. Narayan
Shockwave Oscillations Over the Conical Heat Shield Region
of a Typical Launch Vehicle at Mach 0.95 . . . . . . . . . . . . . . . . . . . . . . . 491
K. N. Murugan, T. Arunkumar, M. Prasath, and V. R. Ganesan
Diverterless Supersonic Intake for a Generic Stealth Fighter
Aircraft . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
Sameer Karania, Manu Mohan, and Satya Prakash
Analysis of Missile Plume Impact Characteristics on Engine Intake
and Neighboring Stores for a Fighter Aircraft . . . . . . . . . . . . . . . . . . . . 511
Jishnu Suresh, S. Karthik, P. Karthikeya, S. Rajkumar, Mano Prakash,
Sashi Kiran, D. Maharana, and D. Narayan
Control of Tailless Aircraft . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523
T. S. Ganesh, M. C. Keerthi, Sabari Girish, S. Sreeja Kumar,
and B. Mrunalini
About the Editors
Dr. S. Kishore Kumar is a DRDO fellow in the Gas and Turbine Research
Establishment (GTRE), Associate Editor of Journal of Aerospace Sciences &
Technologies of AeSI and Technical Adviser to National Design & Research
Forum. He has previously served as Associate Director and Programme Director
(GATET), GTRE. His areas of specialization include design & CFD analysis of Gas
Turbine Engines, Optimization, mathematical modelling and dynamical systems.
He has published 7 edited volumes and more than 100 research papers & technical
reports.
Dr. Indira Narayanaswamy is a Research Professor at Ramaiah University of

Applied Sciences, Bangalore, India. She has previously served as a scientist and
Technology Director at Aeronautical Development Agency, Bangalore.
Dr. Narayanaswamy’s areas of specialization include CFD applications for
Aerodynamic Shape Optimization and Multidisciplinary Design Optimization
(MDO), MDO for MAV design and Numerical Investigations on Supersonic Retro
Propulsion (SRP). She has authored more than 150 research papers and technical
reports. She is one of the Editors of the Proceedings of the 2nd National Conference
on MDAO, published by Springer Nature Singapore Pte Ltd, 2020.
Dr. V. Ramesh is the head of the CTFD Division in NAL, an AcSIR Professor and
the Secretary of the CFD Division of the Aeronautics Society of India. His research
interests are primarily in aerospace science and technology, and includes
Computational Fluid Dynamics (CFD), mesh free methods, and computational
aero-elasticity. He has authored more than 150 research publications and reports.
ix
Mach Number Effect on Aeroacoustic
Characteristics of Compressible Jet Due
to Chevron
S. R. Nikam and S. D. Sharma
Abstract The present experimental investigation reports the effect of chevron on

flow characteristics and associated acoustic characteristics at different jet exit Mach
numbers in high subsonic compressible jet. Flow characteristics are investigated with
mean pressure measurement using miniature pitot tube in the flow, and acoustic char-
acteristics are investigated using fluctuating pressure measurement with array of four
microphones in the far field of jet. Chevron is used as a flow control device on the lip
of the nozzle. Chevron converts axisymmetric jet development into corrugated shear
layer, closer to the nozzle exit. This effect diminishes as jet grows downstream away
from the nozzle exit. Corrugation of jet shear layer closer to the nozzle exit increases
with increase in Mach number. Compressibility effect with change in Mach number is
seen from potential core length and jet growth rate for base nozzle; however, chevron
is found to reduce the compressibility effect with change in jet exit Mach number due
to enhancement in mixing. Chevron reduces far-field overall sound pressure level at
shallow polar angle (30°) by about to 2 dB at all the Mach number; however, increase
in noise level at higher frequency observed at higher polar angle is mainly due to
high-frequency noise sources produced from chevron petals. Noise level at higher
polar angles and higher frequencies increases with increase in Mach number.
Keywords Chevron · Potential core · Streamwise vortex
1 Introduction
Considering increased frequency of aircraft operation, noise pollution due to aircraft

became a major concern for the community around the airports. The regulations intro-
duced by the International Civil Aviation Organization are becoming more stringent
S. R. Nikam (B)
Mechanical Engineering Department, K. J. Somaiya College of Engineering, Mumbai, India
e-mail: [email protected]
S. D. Sharma
Aerospace Engineering Department, IIT Bombay, Mumbai, India
© Springer Nature Singapore Pte Ltd. 2021 1

S. K. Kumar et al. (eds.), Design and Development of Aerospace Vehicles
and Propulsion Systems, Lecture Notes in Mechanical Engineering,
https://doi.org/10.1007/978-981-15-9601-8_1
2 S. R. Nikam and S. D. Sharma
over a period considering the higher frequency of operation of aircraft. To meet the
increasingly tough international regulation on aircraft noise, the aircraft manufac-
turers and researchers are devising the techniques to reduce the noise level. Major
challenges in implementing these techniques are thrust and weight penalty along
with increase in cost due to these devices and hence led to a demand for quieter and
more efficient jet engine designs. Jet exhaust noise is one of the major sources of
aircraft noise especially during takeoff. Jet exhaust noise is caused by the turbulent
mixing of high-speed exhaust gases with the atmospheric air which occurs due to
velocity gradients in the shear layer of jet.
Various researchers tried different techniques like tabs [1, 2], microjet [3, 4],
vortex generators [5], notches [6, 7] and chevrons [8–11] to reduce jet engine exhaust
noise. However, chevron is found to be effective considering minimum thrust [12]
and weight penalty. Chevron is saw-tooth serrations cut in the lip of the nozzle
which produces the contra-rotating pair of streamwise vortices [8]. These stream-
wise vortices disrupt the azimuthal ring of vortices and enhance the mixing. Various
researchers [10, 13, 14] investigated the effect of chevron and its geometric param-
eters on aero-acoustic characteristics of the jet. However, effect of Mach number in
the presence of chevron on the aeroacoustics characteristics of jet flow in compress-
ible subsonic range is not discussed much in the literature. The present experimental
investigation reports the effect of Mach number (0.5–0.9) on the flow field and
acoustic characteristics in the far field with chevron.
2 Experimental Setup and Test Nozzle
Experiments were conducted in a high-pressure single flow freely expanding cold

jet facility. Jet flow facility consists of a screw compressor to supply compressed
air, pressure regulating valve and plenum chamber with the provision for attaching
a test nozzle. Pressure in the plenum chamber was maintained by operating pressure
regulating valve to get required nozzle exit Mach number. The jet exits into initially
still ambient air. To minimize sound reflection from wall boundaries, roof and floor,
a semi-anechoic enclosure was constructed using 65-mm-thick polyurethane (PU)
foam sheets surrounding to test facility. Convergent nozzle with 30-mm nozzle exit
diameter (D), 130 mm length and 1 mm lip thickness was used in the present inves-
tigation as base nozzle. The nozzle had included angle of 11° and its tapered wall
provided the outer included angle of 16°. This base nozzle was modified into a
chevron nozzle by capping a ring of chevrons with eight petals made out of 0.5-mm-
thick stainless steel sheet. Thus, chevron nozzle was constructed by fitting chevron
ring over base nozzle. Figure 1 shows details of the chevron nozzle. Figure 1a shows
a representative cut section schematic of chevron nozzle with parametric details such
as petal crest angle (β) which is 45°, petal length (L) and penetration (R-Y ). Figure 1b
shows a photograph of the base nozzle and chevron ring assembly.
Mach Number Effect on Aeroacoustic Characteristics of Compressible … 3
Fig. 1 Chevron nozzle L=14.8

a sectional details and
b photograph of chevron
nozzle 8° 5.5° R=15
Y=13.16
β
Chevron ring
(a)
(b)
3 Experimental Methods
Aerodynamic measurements are taken using a miniature pitot probe (0.7 mm diameter
stainless steel hypodermic tube) inside the jet, and acoustics measurements are taken
using array of four free-field microphones in the far field of jet. To position the
measuring probe at a desired point, a computer-controlled Dantec Dynamics traverse
is used which has a traverse range of 610 mm along each of the three orthogonal
axes with a resolution of 0.01 mm. The plenum chamber pressure is determined using
isentropic relation for required nozzle exit Mach number and maintained by operating
pressure regulating valve. Pressures were registered by using Pressure Systems make
transducers, PSI model 9116 with 16 channels.
Mean pressure was measured using pitot tube along the jet centerline for M = 0.5,
0.6, 0.7, 0.8 and across the jet in transverse plane at three streamwise locations (x/D
= 0.5, 2 and 5) in steps of 0.067D for M = 0.5 and 0.8. Chevron nozzle exit plane was
considered aligned with the trough of the notch in chevron ring. The acoustic pressure
fluctuations were measured in the far field for jet exit Mach number 0.4, 0.5, 0.6,
0.7, 0.8 and 0.9 at polar angles, θ, from 30° to 90° in steps of 10° at a fixed far-field
distance of 45D in the horizontal plane. Four B&K, 6.53 mm (1/4 ) diameter 4939
free-field condenser microphones were used along with B&K 2970 pre-amplifiers for
acoustic measurements. The microphones were powered by four-channel NEXUS
model 2690-0S4 signal conditional amplifiers. Each microphone had flat frequency
response from 4 Hz to 100 kHz and an open-circuit sensitivity of 4.5 mV/Pa. Before
commencing the measurements, the microphones were calibrated using a B&K 4226
multifunction acoustic calibrator. Data acquisition was carried out using National
Instrument PCI-4462 card-based DAQ and LabVIEW software. Hundred blocks of
narrow band data are acquired at the rate of 200 kHz with 8192 samples which gives
frequency band width of 24.41 Hz. Accuracy of acoustic measurement is within
±0.3 dB. The acquired data is post-processed to obtain sound pressure level, overall
sound pressure level (OASPL) using MATLAB.
4 Results
4.1 Aerodynamics Results
Centerline Velocity Distribution

Figure 2 shows the jet centerline velocity distribution (u) at M = 0.5, 0.6, 0.7 and
0.8. Velocity is normalized with the jet exit velocity (U exit ), and the distance is
normalized by the base nozzle exit diameter (D). At Mach 0.8, the potential core
length for the base circular nozzle is about 5D, which agrees with many of the
earlier measurements [8, 13, 15–17] by various techniques in Mach number range
from 0.78 to 0.9. Potential core length obtained from the present measurement is
Fig. 2 Axial velocity along

centerline of jet
Table 1 Comparison of
Lau et al. [16] Witze [18] Present study
potential core length with
empirically obtained values M = 0.8 4.9D 5.19D 5D
M = 0.5 4.48D 4.82D 4.5D
compared with the values obtained empirically as given by Lau et al. [16] and Witze
[18]. Comparison of potential core length given in Table 1 indicates that the present
measurements match very closely with the previous empirically established data.
Effect of Mach number is clearly seen. Reduction in potential core is observed as
Mach number is reduced from 0.8 to 0.5. Potential core length for the base nozzle at
Mach 0.5 is obtained as 4.5 diameters downstream as compared to 5 diameters at M
= 0.8. Velocity decay rate also increases with reduction in Mach number. Centerline
velocity reduces from 75% at M = 0.8 to 69% at M = 0.5 for the measuring station
located at x/D = 10. This difference in the potential core length and the centerline
decay rate with Mach number is due to compressibility effect which increases with
increase in the Mach number. Reduction in the potential core length and increase in
the centerline velocity decay are noticed at all the Mach number for chevron nozzle
as compared to base nozzle. Potential core length for chevron nozzle is about 3D at
all the Mach number. For Mach 0.8, the centerline velocity decay from about 75% for
base nozzle to 61% for chevron nozzle at the measuring station located at x/D = 10
is observed. Similar findings are reported by Bridges and Brown [13] for the chevron
nozzle with six petals. It is interesting to note that for Mach 0.5, at measuring station
x/D = 10, chevron nozzle reduces centerline velocity to 59% as compared to 69% for
the base nozzle. This reduction in centerline velocity for chevron nozzle at M = 0.5
(59%) is almost same as that obtained at M = 0.8 with chevron (61%). Thus, decay
rate is almost same for the chevron nozzle at both the Mach numbers indicating that
enhanced mixing due to chevron reduces the effect of compressibility.
Jet Flow Development
Iso-velocity contours normalized by the jet exit velocity in the transverse (y–z) plane
at x/D = 0.5 and 5 are shown in Figs. 3 and 4 for base and chevron nozzle, respectively,
at M = 0.5 and 0.8. Jet develops axis symmetrically for base nozzle; however, chevron
nozzle makes the flow corrugated in the shape of lobe. For the base nozzle as shown
in Fig. 3 at x/D = 0.5, size of the jet is same at both the Mach numbers due to
insignificant entrainment in the close proximity to the nozzle exit. However, at x/D
= 5, size of the jet is different. Due to compressibility effect, at Mach 0.8, jet size is
less as compared to Mach 0.5. For chevron nozzle (Fig. 4) at x/D = 0.5, Mach number
has a strong effect on the jet cross section. Increase in Mach number enhances the
bulging of the corrugated shear layer, thereby increasing the perimeter of the jet edge.
Jet cross section becomes circular at x/D = 5 for chevron nozzle. Compared to the
base nozzle, jets from chevron nozzle show larger cross-sectional area. It is believed
that streamwise vortices produced from chevron petals make the jet corrugated which
in turn increases the mixing. Increase in the size of lobe due to increase in Mach
number indicates that strength of streamwise vortices increases due to increase in
M = 0.8 M = 0.5
Fig. 3 Iso-velocity contours for base nozzle
Mach number. This variation in strength of vortices is due to change in the velocity
gradient in the radial and azimuthal direction. For chevron nozzle, cross section of the
jet at x/D = 5 is marginally less at lower Mach number. This observation is exactly
reverse of base nozzle. Thus, higher strength of streamwise vortices produced at
higher Mach number makes the mixing aggressive which gives slightly bigger cross
section of the jet at higher Mach number.
Mass Entrainment
The mass flow rate is estimated using the isentropic relations for the density and the
velocity distribution obtained from pressure measurement in grid at three longitudinal
position (x/D = 0.5, 2 and 5) for the base and chevron nozzle as shown in Fig. 5.
Nearly
⎧ linear

variation
⎫
of mass flow rate for the base nozzle is seen. The entrainment
⎨ d ṁ/ṁ ⎬
exit
rate for base nozzle at M = 0.8 and 0.5 is 0.15 and 0.20, respectively.
⎩ d x/D ⎭
M = 0.8 M = 0.5
Fig. 4 Iso-velocity contours for chevron nozzle
These entrainment values compare well with the 0.13 reported by Arakeri et al. [19]
at M = 0.9. Higher mass flow rate at lower Mach number is seen for base nozzle
which is due to compressibility effect. Chevron does not show linear variation in
mass flow rate. Mass flow rate shows rapid entrainment rate till x/D = 2 thereafter
entrainment decreases. Thus, most of the entrainment due to chevron takes place
within x/D = 2. Chevron shows maximum increase in mass flow rate by about 15%,
at x/D = 5 and M = 0.8, whereas, for M = 0.5, chevron increases mass flow rate by
about 5%. Increase in entrainment due to chevron is higher at higher Mach number
due to chevron. Thus, the increase in the centerline velocity decay at higher Mach
number found earlier for chevron nozzle in Fig. 2 is associated with increase in the
mass entrainment.
Fig. 5 Mass flow rate along

the jet axis
4.2 Acoustics Results
Far-field acoustic pressure fluctuations are recorded at polar angles 30° to 90° in a
step of 10° for jet exit Mach number 0.4–0.9 at fixed radial distance of 45D using
microphones. SPL and OASPL are obtained from these pressure fluctuations. Figure 6
shows sound pressure level against non-dimensional frequency Strouhal number (St)
at polar angle 30° in 1/3—octave band. Peak SPL is obtained at Strouhal frequency of
about 0.2. This peak frequency remains almost same at all the Mach number except
marginal shift toward higher value at lower Mach number. Similar observations are
reported by Bogey et al. [20]. Thus, noise at this location is mainly contributed by
low-frequency noise sources. As expected, peak level decreases by about 30 dB as
Mach number is reduced. Chevron is found to be equally effective at all the Mach
number for reducing the noise by about 3 dB at peak frequency, whereas high-
frequency noise remains almost unaffected with increase in Mach number due to
chevron. Figure 7 shows SPL at polar angle 90°. Compared to SPL at 30°, spectra at
90° is broadband in nature with reduced levels. Chevron continues to reduce the level
at low frequency at all the Mach number, but noise level increases due to chevron
at high frequency. This increase in level is highest at higher Mach number. Change
over frequency shifts toward higher value at low Mach number. Thus, chevron gives
benefit of noise reduction at wide frequency band at low Mach number.
OASPL obtained for three different ranges of frequency is shown in Fig. 8. OASPL
for entire range of frequency (200–100,000 Hz) shows that OASPL increases as polar
angle reduces. Rate of increase of OASPL reduces as Mach number reduces along
with shift in peak OASPL from 30° at higher Mach numbers to about 50° at lower
Mach number. Chevron reduces OASPL by about 2 dB at lower polar angle (30°) for
all the Mach number, whereas OASPL increases at higher angles for Mach number
Fig. 6 Far-field 30° noise spectra registered different Mach numbers for base nozzle (red color—
continuous line) and chevron nozzle (blue color—dotted line)
Fig. 7 Far-field 90° noise spectra registered different Mach numbers for base nozzle (red color—
continuous line) and chevron nozzle (blue color—dotted line)
Fig. 8 Far-field OASPL

directivity at different
frequency bandwidth and
Mach number for base
nozzle (circle) and chevron
nozzle (diamond)
0.9, 0.8 and 0.7. Such increase in noise level at higher polar angle is not seen at
lower Mach number. OASPL in lower frequency range (200–10,000 Hz) shows that
chevron reduces the noise at all the Mach number and polar positions almost by equal
amount, whereas, in high frequency range (10,000–100,000 Hz), chevron increases
the noise at all the polar angles except 30°. Noise level due to chevron increases with
increase in polar angle. This increase in level is more pronounced at higher Mach
number. Thus, chevron reduces the strength of low-frequency noise sources which
results in reduction in low-frequency noise in the far field at low polar angle. On the
contrary, chevron increases the strength of high-frequency sources which increases
the noise level at high frequency at higher polar angle in the far field. Strength of high-
frequency noise sources increases with increase in velocity due to which increase in
high-frequency noise is more at high Mach number.
5 Conclusion
Important conclusions drawn from the present experimental investigation are

• Formation of streamwise vortices from chevron petals makes the jet corrugated
close to the nozzle exit which in turn enhances the mixing and reduces the potential
core. Enhanced mixing in azimuthal and radial direction turns the corrugated jet in
to axisymmetric, within five nozzle diameters with increased mixing as compared
to base nozzle.
• Chevron reduces the compressibility effect mainly due to aggressive mixing at
higher Mach number.
• Enhanced jet mixing due to chevron reduces the low-frequency noise at shallow
angle to the jet axis in the far field along with increased high-frequency noise at
higher angles.
• For chevron, change in the Mach number does not affect much in reducing the
noise level in far field at low frequency, whereas increased strength of streamwise
vortices with increase in Mach number increases the far-field noise level at higher
polar angle. These observations are seen with one particular chevron geometry;
however, effect of chevron at different Mach numbers may vary as geometric
parameters of chevron change.
References
1. Simonich JC, Narayanan S, Barber TJ, Nishimura M (2001) Aeroacoustic characterization,

noise reduction, and dimensional scaling effects of high subsonic jets. AIAA J 39(11):2062–
2069
2. Samimy M, Zaman KBMQ, Reeder MF (1993) Effect of tabs on the flow and noise field of an
axisymmetric jet. AIAA J 31(4):609–619
3. Arakeri VH, Krothapalli A, Siddavaram V, Alkislar MB, Lourenco LM (2003a) On the use of
microjets to suppress turbulence in a Mach 0.9 axisymmetric jet. J Fluid Mech 490:75–98
4. Castelain T, Sunyach M, Juvé D, Béra J-C (2008) Jet-noise reduction by impinging microjets:
an acoustic investigation testing microjet parameters. AIAA J 46(5):1081–1087
5. Panickar P, Sharma S, Sarpotdar S, Raman G, Butler W (2011) New vortex generator design
for nozzle internal modification. Int J Aerosp Innov 3(4):249–260
6. Verma SB, Rathakrishnan E (1999) An experimental study on the noise characteristics of
notched circular-slot jets. J Sound Vib 226(2):383–396
7. Ahuja KK, Manes JP, Massey KC, Calloway AB (1990) An evaluation of various concepts of
reducing supersonic jet noise. In: 13th aeroacoustics conference, AIAA Paper-90-3982
8. Alkislar MB, Krothapalli A, Butler GW (2007) The effect of streamwise vortices on the
aeroacoustics of a Mach 0.9 jet. J Fluid Mech 578:139–169
9. Callender B, Gutmark E, Martens S (2008) Near-field investigation of chevron nozzle
mechanisms. AIAA J 46(1):36–45
10. Callender B, Gutmark EJ, Martens S (2005) Far-field acoustic investigation into chevron nozzle
mechanisms and trends. AIAA J 43(1):87–95
11. Wernet M, Brown C, Bridges J (2003) Control of jet noise through mixing enhancement.
NASA/TM-2003-212335
12. Saiyed NH, Mikkelsen KL, Bridges JE (2003) Acoustics and thrust of quiet separate-flow
high-bypass-ratio nozzles. AIAA J 41(3):372–378
13. Bridges J, Brown C (2004) Parametric testing of chevrons on single flow hot jets. In: AIAA-
Paper-2004-2824
14. Tide PS, Srinivasan K (2010) Effect of chevron count and penetration on the acoustic
characteristics of chevron nozzles. Appl Acoust 71(3):201–220
15. Yu SCM, Lim KS, Chao W, Goh XP (2008) Mixing enhancement in subsonic jet flow using
the air-tab technique. AIAA J 46(11):2966–2969
16. Lau JC, Morris PJ, Fisher MJ (1979) Measurements in subsonic and supersonic free jets using
a laser velocimeter. J Fluid Mech 93(1):1–27
17. Kolpin MA (1964) The flow in the mixing region of a jet. J Fluid Mech 18(4):529–548
18. Witze PO (1974) Centerline velocity decay of compressible free jets. AIAA J 12(4):417–418
19. Arakeri VH, Krothapalli A, Siddavaram V, Alkislar MB, Lourenco LM (2003b) On the use of
microjets to suppress turbulence in a Mach 0.9 axisymmetric jet. J Fluid Mech 490:75–98
20. Bogey C, Barré S, Fleury V, Bailly C, Juvé D (2007) Experimental study of the spectral
properties of near-field and far-field jet noise. Int J Aeroacous 6(2):73–92
Parametric Study of Turbulent Flow Past
a Compression–Decompression Ramp
Rhea George and Raj Kiran Grandhi
Abstract Shock wave-boundary layer interactions take place in many vehicle con-
figurations of practical importance such as wing–body junctures, deflected con-
trol surfaces, high-speed inlets and forward-facing steps. The associated flowfield
becomes complex when the interaction causes flow to separate. A parametric study is
carried out for supersonic flow past a compression–decompression ramp (CDR), to
determine the ramp inclination angle that minimizes the adverse effect of separated
flow. The inclination angle below which weak interaction occurs is obtained from
the numerical simulations.
Keywords Compression–decompression ramp · Flow separation · Shock

wave-boundary layer interaction
1 Introduction
Shock wave-boundary layer interaction can cause flow to separate, thereby leading to
increase in drag, loss in efficiency of control surface, increased turbulence level and
heat transfer rates. The associated flowfield becomes complex when the interaction
causes flow to separate. Shock-induced separation can cause unsteady loads that
may lead to structural damage. Shock wave-boundary layer interactions may be
utilized to improve fuel–air mixing in scramjet combustors. Hence, the study of this
phenomenon assumes great importance in supersonic flows.
A commercial computational fluid dynamics (CFD) solver ANSYS Fluent [1] is
used to simulate this interaction. The flow solver is validated and then used to perform
a parametric study of flow over a compression–decompression ramp (CDR). The
literature on compression corner flows is relatively vast, but experimental results for
compression–decompression ramp flows that can be used for CFD solver validation
R. George (B) · R. K. Grandhi

Advanced Systems Laboratory, Defence Research
and Development Organisation, Hyderabad 500058, India

16 R. George and R. K. Grandhi
are much scarcer. One of the few available papers [2] lists out the experimental
results for a CDR of 25◦ inclination at a Mach number of 2.88. The flow solver has
been validated for this configuration [3]. The present study attempts to conduct a
parametric study over a typical CDR, the ramp inclination angle being the parameter
that is varied, for constant ramp height. The governing equations are the compressible
Navier-Stokes equations.
2 Flowfield in the Vicinity of CDR
When the angle of inclination is sufficiently small, the compression waves coalesce
into a single shock and the pressure downstream is similar to the inviscid case. For
a sufficiently higher inclination angle, the boundary layer separates upstream of the
compression corner and reattaches downstream [4]. The deflection of the boundary
layer by the separated flow causes a rise in surface pressure, and the compression
waves coalesce into a separation shock. A pressure ‘plateau’ occurs over the reversed
flow region. A second compression wave system forms near the reattachment point
and coalesces into the reattachment shock wave. The two shock waves intersect to
form a λ-shock. The flow then expands about the decompression corner.
The stages of shock–boundary layer turbulent interaction in the vicinity of
compression–decompression ramps [4] have been classified as (i) unseparated flow,
(ii) intermittent separation, (iii) developing small-scale separation, (iv) large-scale
separation and (v) maximum-scale separation. As the inclination angle increases, the
flow structure changes from unseparated flow to the subsequent stages of separation
and finally maximum-scale separation. The maximum-scale separation corresponds
to detached shock for inviscid flow.
3 Solver Validation
Flow simulations are carried out for a freestream Mach number of 2.88 as per the
experimental conditions [2]. These wind tunnel conditions are shown in Table 1.
The CDR has an inclination of 25◦ as shown in Fig. 1. The origin is taken at the
compression corner.
The nondimensionalized static pressure over the walls, computed using various
turbulence models, is plotted in Fig. 2. The pressure plot shows the rise in pressure
Table 1 Experimental conditions

Mach number Temperature (K) Pressure (kPa) Velocity (m/s)
2.88 110.57 11.964 607
Parametric Study of Turbulent Flow Past a Compression–Decompression Ramp 17
Fig. 1 Compression–decompression ramp (all dimensions in mm)
5.0
std-kε
4.5 realizable-kε
sst-kω
4.0 sa
3.5 Experiment
p/p∞ →
3.0
2.5
2.0
1.5
1.0
0.5
-8 -6 -4 -2 0 2 4 6 8 10 12
x (cm) →
Fig. 2 Static pressure on walls
well ahead of the compression corner and expansion over the decompression corner.
This plot also shows that the pressure in the separation region is well predicted by
most of the models. The SST k-ω model predicts a larger reversed flow region, with
a pressure peak below the experimental value, which is in line with the findings of [5,
6]. The remaining three turbulence models predict the pressure in the separated flow
regime much more accurately. The one-equation Spalart–Allmaras (SA) turbulence
method performs just as well as the two-equation models.
The velocity vectors displaying the separation and reattachment regions near the
compression corner are presented in Fig. 3. The sonic line and the zero velocity line,
as shown in the velocity contours of Fig. 4, agree well with the measurements [4].
4 Grid Independence
Grid independence studies are carried out to ensure that the grid size, used for the
simulations, is adequate to capture the flow features. Three levels of grid sizes ranging
from coarse (504 × 160 cells) to medium (620 × 200 cells) and fine (740 × 240
cells) are utilized. The first cell height normal to the walls is unchanged for all the
grids, to maintain the wall y + less than 1 in all the cases. The results obtained from
607
562
516
470
425
379
333
287
242
196
150
Reattachment
105
59
13
-32 Separation
-78
-124
Fig. 3 Velocity vectors in the vicinity of compression corner
Fig. 4 Velocity magnitude contours in the vicinity of compression–decompression corner

4.5
coarse
4.0 medium
fine
p/p∞ → 3.5
3.0
2.5
2.0
1.5
1.0
-2 -1 0 1 2 3
x/h →
Fig. 5 Pressure variation with grids
Table 2 Flowfield conditions

Mach number Altitude (km) Temperature (K) Pressure (kPa)
2 12 225 21
these grids are compared to check for grid independence. It is observed from the
plots of pressure in Fig. 5 that the results from the medium and fine grids are almost
coincident. Hence, the medium grid is used for all the simulations.
5 Simulation Parameters
Two-dimensional, viscous simulations are performed for CDR configuration with

varying inclination angles in ANSYS Fluent, to determine the inclination angle that
minimizes or avoids flow separation. Flow simulations are carried out at the condi-
tions given in Table 2. The pressure inlet boundary is assigned over the inlet to the
domain. The domain in the normal direction is taken to be sufficiently large in order
to avoid shock reflection from the boundaries. The grid generated for this domain
consists of 0.12 million cells. The height of the first row of cells is about 1 × 10−6
m, to ensure a wall y+ less than 0.4. Spalart–Allmaras (SA) turbulence method is
used to model turbulence. The ramp inclination angles considered are 8, 10, 12, 14,
15, 16, 20 and 25◦ .
6 Simulation Results
Nondimensionalized static pressure variation over the walls, computed for various
inclination angles, is plotted in Fig. 6. It shows the rise in pressure, well ahead of the
compression corner and expansion over the decompression corner. It is observed from
2.8 ο
8ο
2.6 10
2.4 12ο
ο
14
2.2 ο
15ο
2.0 16ο
p/p∞ →
20ο
1.8 25
1.6
1.4
1.2
1.0
0.8
-5 0 5 10 15
x/h →
Fig. 6 Pressure variation with angle of inclination of compression corner
0.0025
0.0020
Skin friction coefficient →
0.0015
0.0010
0.0005 8ο
10ο
ο
0.0000 12
14ο
-0.0005 ο
15
16ο
-0.0010 20ο
25ο
-0.0015
-3 -2 -1 0 1 2
x/h →
Fig. 7 Skin friction coefficient on the walls of compression–decompression corner
this plot that as the inclination angle increases, the peak pressure rises, due to increase
in shock strength. The pressure profiles for smaller angles of inclination (8, 10 and
12◦ ) closely follow the inviscid pressure profile. This interaction is referred to as Weak
Interaction [7]. As this angle increases, the pressure profile varies considerably from
the inviscid profiles (Strong Interaction). The upstream influence of the compression
corner increases with inclination angle.
The variation of the skin friction coefficient with angle of inclination of the CDR
in the vicinity of compression corner is presented in Fig. 7. The nature of the skin
friction variation near the compression corner matches well with literature [4]. The
separation and reattachment locations correspond to the points where the skin friction
coefficient falls below zero and rises above zero, respectively.
The extent of separated flow is evident from the region where the skin friction
coefficient is negative. Separated flow region is minimal for the smaller inclination
angles (8, 10 and 12◦ ). But the flow separation is significant for higher inclination
angles. For an increase in inclination angle from 12 to 14◦ , the separated flow region
2.00
1.75
1.50
1.25
1.00
0.75
0.50
0.25
0.00
Fig. 8 Mach number contours for ramp inclination angle of 12◦
widens considerably. At inclination angles of 20 and 25◦ , a change in the nature of

the skin friction curve near the compression corner may be attributed to secondary
flow separation.
6.1 Flow Features
Mach number contours for ramp inclination angles of 12 and 25◦ are presented in
Figs. 8 and 9, respectively. The difference in the flowfield structure is evident in these
contours. The large separated flow region ahead of the 25◦ ramp is seen clearly in
Fig. 9. The velocity vectors in the vicinity of the CDR of ramp inclination angle of
12◦ are shown in Fig. 10. It is observed from this figure that reversed flow is limited to
the compression corner only. For ramp inclination angle of 25◦ , the velocity vectors
are presented in Fig. 11. This figure clearly shows the attached flow well ahead of
the CDR, separated flow just ahead of and aft of the compression corner and flow
reattachment on the ramp.
6.2 Separated Flow Extents
The separation length (L) is defined as the distance from the separation point to the
compression corner. The variation of the separation length, normalized by the ramp
height (h), with the ramp inclination angle is plotted in Fig. 12. The change in slope
2.00
1.75
1.50
1.25
1.00
0.75
0.50
0.25
0.00
Fig. 9 Mach number contours for ramp inclination angle of 25◦
602
523
445
367
289
210
132
54
-25
Fig. 10 Velocity vectors in the vicinity of ramp with an inclination angle of 12◦
602
510
418
327
235
143
52
-40
-131
Fig. 11 Velocity vectors in the vicinity of ramp with an inclination angle of 25◦
2.0
separation length
1.8
1.6
1.4
1.2
L/h →
1.0
0.8
0.6
0.4
0.2
0.0
5 10 15 20 25
Inclination angle (degrees) →
Fig. 12 Separation length variation with angle of inclination of ramp
of the plot is observed at an angle of 16◦ . This value agrees with the value given in
literature [4]. Separation initially appears at a small angle (onset of separation) and
slowly increases with inclination angle. The angle at which the slope changes defines
the onset of large-scale separation. As per literature [4], the inclination angle corre-
sponding to onset of separation ranges from 6.5 to 12◦ , while large-scale separation
onset occurs at inclination angles in the range of 15–19◦ .
7 Summary
The CFD software, ANSYS Fluent is successfully validated for compression–

decompression ramp-induced separated flows. The pressure profiles match well with
literature. It is also observed from the validation studies that the one-equation SA
turbulence model predictions are as good as the two-equation models.
It can be inferred from the present results that the changes in pressure and skin
friction coefficient are adequately captured by the Fluent simulations for supersonic
flow past compression–decompression ramps (CDR) of varying ramp inclination
angles. Ramp inclination angles less than 12◦ minimize the adverse effects of sep-
arated flow for the conditions considered here. In general, flow separation can be
avoided by reducing the ramp inclination angle.
References
1. Fluent User Manual. Fluent Inc. (2006)

2. Settles GS, Dodson LJ (1991) Hypersonic shock/boundary layer interaction database. Technical
report, NASA CR 177577
3. George R, Grandhi RK, Gupta RK (2013) Validation of Fluent for turbulent separated flow
past a compression-decompression ramp. In: ANSYS 2013 Convergence Conference
4. Babinsky H, Harvey JK (2011) Shock wave–boundary-layer interactions. Cambridge
5. Oliver AB, Lillard RP, Blaisdell GA, Lyrintzis AS (2006) Validation of high-speed turbulent
boundary layer and shock-boundary layer interaction computations with the OVERFLOW
Code. AIAA 2006-0894
6. Oliver AB, Lillard RP, Blaisdell GA, Lyrintzis AS (2007) Assessment of turbulent shock-
boundary layer interaction computations using the OVERFLOW Code. AIAA 2007-0104
7. Delery J, Marvin JG (1986) Shock-wave boundary layer interactions. Technical report, AGAR-
Dograph 280
Validation of Numerical Analysis Results
for Pusher Configured Turboprop Engine
Air Intake
C. A. Vinay and S. Bhaskar Chakravarthy
Abstract In the course of an aircraft development program of LTA, a number of

engine-related ground and flight tests have to be carried out to determine the losses
particular to the engine installation. These installation losses must be determined in
order to create the final aircraft power setting charts. During the course of configura-
tion development, design analysis is a must to ensure that the proposed design will
perform the intended operation within the estimated losses, this confidence should
be assessed either in a bench tests or in CFD analysis. The numerical study of engine
air intake performance is carried out using RANS-based k − ω SST using ANSYS
FLUENT software. The study is carried out for minimum climb condition, as it is
critical operation of any aircraft mission. The CFD results were compared with the
flight test data for validation. The instrumentation that is necessary to measure instal-
lation losses and engine performance has been installed. Computational results agree
well with the experimental results and were found satisfactory. The CFD results of
air intake show that the total pressure is extensively recovered at the engine intake
plenum (compressor inlet screen) region, and the inlet pressure loss and ram air
recovery were within the acceptable limit as recommended by engine OEM.
Keywords Nacelle air intake · Ram recovery · Total pressure loss · Oil cooler
1 Introduction
The recent interest in design and development of light transport aircraft (LTA) has
led to higher operating Mach number and higher altitudes, which certainly demands
better engine air intake performance [1]. The air intake system of LTA is designed
to provide the maximum possible total pressure at the compressor inlet screen over
C. A. Vinay (B) · S. B. Chakravarthy

CSIR-National Aerospace Laboratories, Bengaluru 560037, India
S. B. Chakravarthy

26 C. A. Vinay and S. B. Chakravarthy
a wide range of normal flight operations. The engine air intake is required to capture
and efficiently compress requisite quantity of air for engine operation. The air inlet
system delivers air to the plenum chambers, which indeed covers the inlet screen.
The air intake should be effective in providing adequate mass flow of air as demanded
by combustor. The overall vehicle performance depends greatly on the flow quality
of the incoming air. Small loss in the air intake efficiency will have a greater adverse
effect on the engine power output. Efficient ram air pressure recovery at the engine
inlet is required to obtain maximum power levels and low specific fuel consumption.
Therefore, the detailed analysis and study of flow behavior through the air intake
ducts are important aspects in the design evaluation and estimation of the engine air
intakes.
Turboprop engines are widely used in commuter category airplanes. Aircraft
design bureaus routinely conduct the flight tests to confirm the performance of the
engine intakes to ensure the rated power extraction from the engines. Air intake
designs are important because engine efficiency, specific fuel consumption, flat rating
limits and efficient matching of engine with airframe, etc. much depend on the effi-
cient intake system design. In this paper, the computational study on flow through
and around the air intake along the nacelle is carried out for the given climb condition.
The computational results were validated with the flight test data of LTA.
2 Geometry and Analysis
2.1 Geometry Details
The geometric modeling was done using CATIA V5 software. The geometry consists
of internal and external parts. The external parts include the nacelle and lip intake as
shown in Fig. 1. Internal parts consist air intake duct, stopper and compressor screen
(referred as plenum) as shown in Fig. 2. Geometric consideration when dealing with
such a complex problem is to be realistic as possible keeping analysis time and cost
into consideration. Therefore, the geometry is modeled as simple, by neglecting the
minor and inconsequential structural details. In addition, the propeller and spinner
assembly was not considered, and the propeller hub (downstream) portion is replaced
by a smooth contour near the aft end of the nacelle. A parametric approach was
adopted for the creation of flow in the domain. The completed geometry was exported
as IGES format to the ANSYS ICEM-CFD software for meshing.
2.2 Mesh Details
The commercially available ANSYS ICEM-CFD software was used to generate

unstructured mesh. The inlet and outlet boundaries are located at 10 L and 15 L
Validation of Numerical Analysis Results for Pusher … 27
Fig. 1 External view of air intake and nacelle
Fig. 2 Symmetric cross-sectional view of nacelle
upstream and downstream. For the domain, the enclosure was set at 20D, where
L is the length and D is the diameter of the nacelle. Coarse mesh was generated
around the geometry and a finer mesh along the wall. To capture flow behavior in the
air intake duct, finer mesh was generated along the air intake duct and at upstream
of nacelle intake [2, 3]. Grid independence study (GIS) was carried out with three
different grids: G1 with 14 million, G2 with 8.3 million and G3 with 4 million
elements. G2 mesh with 8.3 million elements gave satisfactory results for minimum
mesh count, and the same was selected for further computations. Figure 3 shows the
global domain, and Fig. 4 shows the surface mesh on the model.
Fig. 3 Global size of the domain
Fig. 4 Surface mesh on the nacelle
2.3 Numerical Details
In the present computations, the flow field is considered to be turbulent. While the
turbulent fluctuations are not solved, their efforts on the mean flow are modeled by
solving the Reynolds-averaged Navier–Stokes equations using commercial finite-
volume method-based code ANSYS-FLYENT software. Due to the presence of
separation and reattachment regions in the flow field, it was decided to use k − ω
shear stress transport (SST) model for the computations [2]. Spatial discretization
of governing equations was done using second-order upwind scheme. The pressure
Fig. 5 Global boundary conditions
velocity coupling was achieved using SIMPLE scheme. All the computations were
performed using double-precision arithmetic.
2.4 Boundary Conditions
Velocity inlet and pressure outlet boundary conditions are imposed at domain inlet
and outlet, respectively. Domain far-field is set to pressure outlet and nacelle, and
interior parts are set to wall boundary condition as shown in Fig. 5. At the nacelle
cover and compressor screen, interior and pressure outlet with targeted mass flow
rate is applied as shown in Fig. 6. The turbulent viscosity ratio of 5% is maintained
throughout the computation.
3 CFD Results and Validation
3.1 Computational Results
The operating conditions for given minimum climb case are shown in Table 1, and
the results (static pressure and velocity contours) obtained from the computational
study are presented in Table 2.
Fig. 6 Internal boundary conditions
Table 1 Operating condition

Altitude (ft) OAT °C, P0 Free P1 Engine Mass flow Angle of Speed m/s
(K) stream total inlet screen rate, kg/s attack (°) (M)
inlet pressure total (lb/s)
(Pa) pressure (Pa)
4500 6.1 88,578 85,908.6 4.036 6.60 71.46
(279.25) (8.90) (0.21)
Table 2 CFD results

Altitude (ft) Compressor screen Inlet (Throat)
Dynamic Static Total Dynamic Static Total
pressure (Pa) pressure pressure pressure pressure pressure
(Pa) (Pa) (Pa) (Pa) (Pa)
4500 565 1578 2160 2783 10.5 2794
Figure 7 depicts the contours of static pressure over the nacelle at minimum climb
condition. The pressure gradually increases from the throat to plenum region. The
maximum pressure is observed at the compressor inlet case. Due to high pressure at
the air intake, the stagnation region is observed. The lower pressure region (suction
pressure) is observed on the upstream of intake duct, and further downstream of the
intake duct adverse pressure (diffuser) gradient is observed. Low pressure region is
noticed on both the lower and upper portion of the nacelle due to its aerodynamic
profile. However, the pressure drop at the throat is adequately recovered near the
plenum to the significant extent.
Fig. 7 Contours of static pressure at minimum climb condition
Figure 8 depicts the contours of total pressure of the nacelle and air intake duct at
climb condition. It is observed that the total pressure is adequately recovered from
the intake lip to engine intake, which indeed meets the engine requirements which
in turn will improve the performance of engine.
Figure 9 depicts the velocity vectors for minimum climb condition. High velocity
is observed on the upper portion of the nacelle and gradually increases along the
nacelle geometry. Due to the aerodynamic profile of the nacelle, higher velocity is
observed along the half way of nacelle.
Figures 10 and 11 depict the contours of Mach number and the flow path along the
nacelle and air intake duct. As the Mach number indicates the real flight conditions,
it also represents the altitude and temperature. From Fig. 10, it is observed the Mach
contours are continuously changing along the flow field and approaches near the
nacelle surface.
Fig. 8 Contours of total pressure at minimum climb condition

Fig. 9 Velocity vectors at minimum climb condition
Fig. 10 Mach number contours on the nacelle
Fig. 11 Flow path along the nacelle and air intake duct
3.2 Instrumentation and Flight Test
The instrumentation that is necessary to measure installation losses and engine perfor-
mance has been installed at the engine intake lip and at compressor inlet case. At the
inlet plane (throat), the static and total pressure is measured. The inlet plane consists
of 5 total pressure probes and 4 static pressure probes by means of tapping as shown
in Fig. 12. The inlet pressure measurements are used to determine the inlet lip flow
behavior. Engine plenum pressure is measured from the top and bottom dead center
locations of the compressor inlet screens.
Measured Parameters During Flight Test
The measured data from the flight test during minimum climb condition is shown in
Table 3. The operating altitude and calibrated aircraft speed during minimum climb
condition were observed to be 4160 ft and 142 knots.
Data Reduction
The obtained pressure values from the actual flight test are converted to the
quantitative results by using empirical relations as shown in Eqs. 1 and 2, respectively.
P P0 − P1
Intake pressure loss = (1)
P P0
Air Intake Lip

Vane Actuator Door
Static Pressure Probes Total Pressure Probes
Fig. 12 Instrumentation of probes at air intake
Table 3 Flight test results

Air intake parameters LH RH
Air inlet static pressure (psi) 12.83 12.85
Air inlet differential pressure (psi) 0.36 0.40
Plenum chamber (compressor screen) differential pressure (psi) 12.78 12.86
Air inlet total pressure (psi) 13.19 13.25
P1 − Pamb
Ram recovery = (2)
P0 − Pamb
Mass flow rate = ρ × A × V (3)
where
P0 = Free stream total inlet pressure, psi
P1 = Engine inlet screen total pressure, psi
Pamb = Free stream static pressure, psi
A = Air intake throat area, m
V = Velocity, m/s
ρ = Density, kg/m3 .
0.17
Inlet Pressure Loss = ×100 = 1.4%
12.6
12.7 − 13.3
Ram Recovery = × 100 = 97%
13.19 − 13.3
Mass flow rate (lb/s) = 1 .11 × 0.058 × 65.3 = 4.20 kg/s

= 9.2 lb/s
Thus, from the result, it is found that it is clear that for left-hand side engine, the
intake pressure loss is 1.4% and ram recovery is 97% at 4160 ft.
3.3 Validation
The computational results are validated with flight test data of LTA. The comparison
between the flight test results and computational results is shown in Table 4.
From the flight test result, the ram recovery at the intake is 97%, and from compu-
tational study, it was found to be 93.7%. It is clear that the experimental results of
the intake pressure loss is slightly higher when compared with the CFD prediction
whereas experiment data of ram recovery and mass flow agree reasonably well with
computational results. Hence, from this baseline study, the numerical model used for
Table 4 Validation of CFD and flight test data

Air intake parameters CFD study Flight test data (LH and RH engine)
Ram recovery (%) 93.7 97 97
Intake pressure loss (%) 0.93 1.4 1.3
Mass flow (lb/s) 10.29 9.2 9.5
this study was found satisfactory. Thus, CFD results are in good agreement with the
flight test results and found 5% under prediction which is satisfactory.
4 Conclusion
In this paper, the numerical simulation for evaluating the light transport aircraft engine
air intake performance at minimum climb condition has been efficiently carried out.
From the CFD results, it is observed that the total pressure loss at the air intake
(throat) has been adequately recovered at the engine inlet plenum. The performance
parameters are compared between CFD study and the measured flight test results.
The numerical results agree well with the flight test data. Hence, the system pressure
loss and ram recovery are within the acceptable limits as recommended by the engine
OEM.
Acknowledgements The authors would like to thank all partners and associated partners for their
contribution to the program and for their permission to publish this paper.
References
1. Chakraborty D (2015) Numerical simulation of a hypersonic air intake. Def Sci J 65(3):189–195.
https://doi.org/10.14429/dsj.65.8254
2. Tu J, Yeoh GH, Liu C (2008) Computational fluid dynamics, a practical approach, 1st edn.
Elsevier, USA
3. Oliveira GL, Santos LC, Martins AL, Becker GB, Reis MVF, Spogis N, Silva RFAF (2008) A
tool for parametric geometry and grid generation for aircraft configurations. ICAS
Aero-elastic Analysis of High Aspect
Ratio UAV Wing—Based on Two-Way
Fluid Structure Interaction
Vidit Sharma and S. Keshava Kumar
Abstract Aero-elasticity of a wing is the study of airflow around an elastic wing

and their interaction. A two-way Fluid Structure Interaction (FSI) method has been
advantageous in the study of aero-elastic behavior of lifting surfaces. Preliminary
design of an UAV requires detailed understanding of aerodynamic and structural oper-
ational boundaries; to arrive at an optimal wing. High aspect ratio wings are being
employed in Medium Altitude Long Endurance (MALE) and High-Altitude Long
Endurance (HALE) UAVs. High aspect ratio wings are usually designed using com-
posite materials because of their specific stiffness and strength customization ability.
To this end, for designing a high aspect ratio wing, which is aero-elastically compliant
and performs well at all flight conditions—understanding the aero-elastic behavior
is necessary. Preliminary designs as well as detailed design incorporate reduced
ordered models, however to evaluate the efficiency of reduced ordered models—a
full scale 3D numerical model analysis or wind tunnel tests are required. Here aim
is to perform and establish a baseline study of two-way coupled static aero-elastic
analysis based on the existing experimental results of a composite flat plate wing
and a large/high-aspect ratio wing, which in turn will work as a benchmark for flut-
ter analysis and reduced ordered models to be developed. ANSYS package has been
used to demonstrate the methodology to predict the onset of divergence, a static aero-
elasticity phenomenon. The numerical simulation results from ANSYS are validated
with existing experimental results available in the literature. Numerical simulation is
carried out for a composite wing, and then the methodology is extended to high aspect
ratio wing. The results obtained using bidirectional-coupled simulation studies are
found to be in agreement with the results available in the literature.
Keywords Composite materials · FSI · CFD · FEM · Divergence
V. Sharma · S. Keshava Kumar (B)

Department of Aerospace Engineering, Defence Institute of Advanced Technology,
Pune, Maharashtra 411025, India

38 V. Sharma and S. Keshava Kumar
1 Introduction
High Altitude Long Endurance (HALE) and Medium Altitude Long Endurance
(MALE) UAVs are employed for remote sensing and aerial survey of inaccessible
regions [1]. The design of a HALE or MALE UAV demands high aspect ratio wing,
because of their aerodynamic advantages [1]. The high aspect ratio structural design
of a wing requires incorporating composite materials, for their specific strength and
specific stiffness capabilities. High strength/stiffness to weight ratio of composite
materials makes them highly eligible for overall weight reduction of unmanned aerial
vehicles, leading towards high payload capability. However, designing a tailored
composite wing design is a complex process, as there are too many parameters to
consider. Hence to design an aero-elastic compatible wing reduced ordered models
are required [2], through which quick design iterations can be performed. But there
are many reduced order models and to choose the appropriate model baseline result
are required [2]. Albeit there are many reduced ordered models, very few are feasible
for high aspect ratio wings [2]. Hence, to shortlist the appropriate model or to develop
a new reduced ordered model, which can incorporate non-linearities inherent to high
aspect ratio wings—we require benchmark results. To this end a 3D numerical sim-
ulation will serve the purpose [3]. Hence, a 3D two-way FSI simulation for a known
wing model, for which experimental results are available, is performed first. By this
the optimal simulation method of two-way FSI process shall be established, and then
adopt the similar process for analyzing high aspect ratio wing. To achieve this aim,
we proceed in a step-wise fashion. First, aero-elastic (divergence) analysis of flat
laminated composite plate is performed, for which analytical as well as wind tunnel
results are available [4–6], through this process and methodologies will be validated.
Then this work will be extended to incorporate non-linearities in the wing structure.
Among the various methods two-way Fluid Structure Interaction (FSI) provides a
high-fidelity analysis tool to capture the structural and aerodynamic non-linearities.
2 Aero-elasticity
Aero-elasticity study deals with two major areas, static aero-elasticity and dynamic
aero-elasticity. Static aero-elasticity analysis has been performed in this paper, brief
introduction of static aero-elasticity wing divergence is explained below.
Static aero-elasticity includes the wing divergence phenomena due to the elastic
nature of the wing and its interaction with fluid flow. Divergence occurs when lift
produced at the center of pressure (or assumed to be aerodynamic center) produces
a moment about the elastic axis of the wing which twists the wing to pitch up and
cause the wing to fail structurally for torsional rigid wing, or induce stall due to large
rotation of the cross-section, for torsional soft wing. For a two-dimensional airfoil
attached to torsional spring (of stiffness constant kt placed at elastic axis/shear center)
Aero-elastic Analysis of High Aspect Ratio UAV … 39
at rigid/root angle of attack, αr , subjected to a freestream air velocity, U∞ , the angle

of attack, α, can be obtained as [7]:
αr
α= , α = αr + θ (1)
1 − q/qD
where, q is the freestream dynamic pressure, qD is divergence dynamic pressure,

and θ is the wing twist. Modifying the Eq. (1) leads us to an equation of straight line,
expressed as:

1 qD 1 1
= − (2)
θ αr q qD
From Eq. (2), theoretical value of divergence dynamic pressure and divergence
speed can be obtained for failure condition; where wing twist tends to infinity or 1/θ
tends to zero. To obtain divergence pressure of a wing a minimum two data points
are required, the data points can be obtained either from the experimental study or
two-way fluid-structure interaction (CFD-FEM) [7]. The graphical representation
for the calculation of divergence dynamic pressure is given in Fig. 1, the value of
divergence dynamic pressure can be extrapolated from the fitted straight line, based
on the wings static twist data; i.e. wing tip twist, θ , at corresponding freestream
dynamic pressure, q, intersecting with 1/q-axis.
Similar method was adopted by Blair and Weisshaar [8] in their experimental
aero-elastic divergence analysis of a swept composite wing, they expressed it as
Modified Southwell plot. Southwell plot method is employed in the present study to
obtain tip divergence of the wing.
Fig. 1 Graphical
representation of divergence
dynamic pressure, qD , based
on Eq. (2)
3 Fluid-Structure Interaction
Solvers for fluid flow and structural analysis are available in various software pack-
ages, and in general they are solved separately for aerodynamic and structural prob-
lems. Both processes require discretization of geometry into small control vol-
umes/elements. Whereas, for aerodynamic study mesh around the body needs to
be more refined as compared to that of a solid structure mesh in FEM. There are
two general methods to perform FSI analysis, monolithic approach and partitioned
approach [9].
3.1 Monolithic Approach
The two participants in FSI analysis, CFD and FEM, are formulated as a single solver.
Where the governing equations of both the participants are solved numerically as
a one combined discretized set of algebraic equations. Data transfer between the
structure and the fluid domain is synchronized and provides a stable solution due to
conservation of properties at the interface of two domains. Monolithic approach is
more feasible when we have weak coupling between two participants, i.e. influence
of either of the participants on each other is negligible during simulation.
Figure 2 shows the schematic of the monolithic approach, where Rf and Rs are
the fluid and structural system equations respectively. The approach can be of steady
or transient state. Although, this approach has been proven as a more robust method
than that of the partitioned approach but it is more computationally expensive and
cannot be formulated easily as that of a software modulated partitioned approach [9].
3.2 Partitioned Approach
As opposed to the monolithic approach where conservation of interface proper-ties

between the fluid and structural participants is synchronized, in partitioned approach
conservation of properties is asynchronized because of the lag between the indi-
vidually solved participants (fluid and structural). In the coupling step integration
leading to the loss of interface properties conservation. In order to solve this problem,
interface data mapping algorithms are incorporated.
Fig. 2 Representation of monolithic approach

Fig. 3 Representation of partitioned approach
As represented in the Fig. 3, data mapping between the two participants takes
place at each coupling step and each coupling step performs internal iterations to
achieve sufficient convergence of the interface properties along with the convergence
of each participant solver. This approached has been adopted (using ANSYS 17.2) in
this paper, due to its less computational expense attributes and software modularity.
The data exchange between the two participants of FSI analysis, is known as coupling
and there are two methods to implement this approach [9].
3.2.1 One Way Coupling
To study the effect of fluid flow on a solid structure, one-way coupling is more
than enough. Where the data mapping takes place in only one direction (fluid to
solid). A representation of such process is displayed in Fig. 4. Usually it is known
as weak coupling, i.e., for aero-elasticity analysis interpolation of pressure from
CFD to FEM (using data mapping algorithms) will only provide the structural defor-
mation/stress/strain results based on the CFD output. Steady as well as transient
simulations can be performed with one-way coupling [10], providing close enough
predictions at a very low computational time, and only if the effects of solid structure
deformation on the fluid flow is negligible. It is usually employed, when fluid loads
will not induce significant deformations in the structure. As for small deformations
in the structures, will not induce substantial change in fluid domain results.
Fig. 4 Representation of
one-way coupling
Once the CFD solver performs calculation and obtains a converged solution, the
surface pressure from CFD domain gets interpolated onto the mesh of the FEM solver
and resulting forces gets transferred. Now, FEM solver performs the calculations until
it reaches convergence. Whereas, in the case of large deformations of structures like
high aspect ratio wings, the significant deformation in structure leads to substantial
changes in fluid domain, and hence two-way coupling is required.
3.2.2 Two Way Coupling
Two-way coupling as called, is important for cases where the effect of solid structure
deformation on the fluid flow is significantly large. This process is described in form
of a flow chart in Fig. 5.
For steady state case, for example static aero-elasticity (divergence study). In first
coupling step, steady state CFD solution of a rigid wing is obtained and the surface
pressure gets interpolated to the FEM solver using one of the mapping algorithms.
Now, static structural solution is obtained by FEM solver which in return interpolates
the nodal displacement to CFD solver. CFD solver deforms the fluid mesh based on
the incoming displacement, this will repeat for certain amount of internal coupling
iterations until the data transfer convergence is reached. Once the data transfer is
converged, simulation will start next coupling step. For current study of static aero-
elasticity, sufficient number of coupling steps are required so that the wing is in
equilibrium state and further coupling steps does not change the displacement and
forces of the wing, assuming the flow and solid structure to be under steady state
conditions.
For transient state case, dynamic aero-elasticity (flutter study), both solver or
participants performs transient flow analysis (CFD) and dynamic structural analysis
(FEM) while exchanging the interface properties at the interfaces and ensuring data
Fig. 5 Representation of two-way coupling

transfer convergence at every coupling time step up to desired number of time steps.
As the data exchange is initiated by both the participants, meaning in both ways,
bidirectional coupling is considered to be a strong coupling.
3.3 Data Mapping
Critical part of the FSI analysis is the coupling between the participants of the anal-
ysis, where pressure fluctuations in the CFD domain surrounding the FEM domain,
has to be imposed as loads on the structure/FEM mesh. Also, the deformation of the
structure in turn will change the flow domain, hence based on the deformations, the
fluid domain has to be remeshed. Which is basically the data transfer between CFD
and FEM solvers, where interface properties are required to be shared between the
two separate meshes of different domains and nature.
This is performed in two major processes and sub-processes, first process includes
the mapping algorithms to matching the source and target mesh. Afterwards, source
mesh sends the data to target mesh based on weight integration. There are two such
algorithms, General Grid Interface (GGI) and Smart Bucket algorithm for conserva-
tive and non-conservative quantities respectively [9].
3.3.1 General Grid Interface (GGI)
Three-dimensional ‘n’ integration points (IP) are created using the method of dividing
the elements faces of both source and target sides, which later are converted into two-
dimensional quadrilaterals, they are extrapolated into rows and columns of pixels.
These pixels are then intersected to create overlapping surface areas known as control
surfaces.
Amount of intersections among the pixels determines the mapping weight contri-
butions for each control surface, which are collected to obtain the value of mapping
weights at each node. These weights are conservative in nature, see Fig. 6 and used for
transferring the quantities like forces, mass and momentum in ANSYS Workbench
[9]. GGI is used to transfer loads from the fluid domain to structure domain.
Fig. 6 General Grid

Interface (GGI) mapping
algorithm
3.3.2 Smart Bucket Algorithm
In this process the target and source mesh are divided into a grid of buckets (group of
elements) to compute the mapping weights. For source mesh nodes associated with
the buckets of target mesh, the mapping weights are calculated. There can be two
possible cases which can exist, empty and non-empty bucket. For non-empty bucket,
where it contains elements inside, one or more elements in the bucket of target side
gets matched to the node of the source side and is executed using an iso-parametric
mapping. For an empty bucket, the closest non-empty bucket is tagged and iso-
parametric mapping is used in the same manner. Non-conservative quantities like
displacement, temperature and stress are transferred using this algorithm in ANSYS
Workbench [9], which makes this approach of profile preserving nature, see Fig. 7.
It is used to transfer data from structure domain to fluid domain.
4 Model Generation
Fluid-Structure Interaction is required to contain two geometrical domains, fluid and

solid. Geometry for the CFD analysis includes the flow domain around the wing
and FEM includes the internal structure of the wing. For validation two available
experimental studies were selected, aero-elastic analysis of laminated graphite/epoxy
flat plate wing (small aspect ratio wing, AR=9.2) by Landsberger et al. [4] and another
of a high-aspect ratio wing, AR=17.7, by Tang and Dowell [11].
4.1 Flat Composite Plate Wing Divergence Study
The cantilevered flat plate wing was designed based on the geometrical param-
eters used by [4–6], half span s = 0.305 m; chord c = 0.076 m, and thickness
t = 0.000804 m (six layers of graphite/epoxy), as shown in Fig. 8a.
The structural geometry was created using the ANSYS ACP (Pre), a compos-
ite pre-post software, consisting six layers with different composite lamina or fiber
ply orientation, θF , as shown in Fig. 9. Total of four different composite plate con-
figurations were created for this validation, [+302 /0]s , [−152 /0]s , [−302 /0]s , and
Fig. 7 Smart Bucket

algorithm
(a) Flat plate composite wing. (b) Large AR wing.
Fig. 8 Schematic of geometry
Fig. 9 Fiber orientation of composite flat plate wing for a stack of six composite ply layers
Fig. 10 CFD domain
[−452 /0]s . For validation, the zero sweep wing data has been taken form [4]. CFD
domain was created around the wing, where domain walls were kept far away from
the model to avoid boundary effects, Fig. 10a.
4.2 Large AR Wing Divergence Study
Based on the wing specifications used by Tang and Dowell [11], wing model was
created with half span s = 0.4508 m; chord c = 0.0508 m. The slender body at the
tip of the wing was created, see Fig. 8b. The wing was modeled with uniform NACA
0012 airfoil cross-section, without ribs and spar as opposed to the referred literature
[11].
The actual wing used in the literature [11] was created using steel spar, aluminum
ribs and balsa wood blocks fitted in-between the ribs. Modeling these elements in
FEM framework is complex, time consuming and computationally costly, as the non-
linearities will creep in due to contact elements, which have to be used to couple the
balsa wood blocks, aluminum ribs and steel spar. And upon careful modeling all the
elements, the properties are not guaranteed to match the actual wind tunnel model.
Hence recourse to above method is adopted, which gives the advantages of simplistic
modeling and advantage of less computational cost.
The stiffness, mass and torsional rigidity parameters are mapped using custom
an-isotropic material properties, see Tables 1 and 4. But, this method could not match
the properties one to one, as there were discrepancies in the elastic axis and center
of gravity location of the modeled wing and the wind tunnel tested model wing [11],
see Fig. 11.
Table 1 Cross-sectional properties of experimental wing setup [11] and FEM wing model
Parameter Value
Tang and Dowell [11] Present study
Span, b 0.4508 m 0.4508 m
Chord, c 0.0508 m 0.0508 m
Flap bending rigidity, EIx 0.4186 Nm2 0.4186 Nm2
Chordwise bending rigidity, EI y 0.1844 × 102 Nm2 0.1844 × 102 Nm2
Torsional rigidity, GJ 0.9539 Nm2 0.9539 Nm2
Ix x – 4.53317 × 10−10 m4
I yy – 3.02209 × 10−8 m4
Torsional constant, J – 1.80227 × 10−9 m4
Mass per unit length 0.2351 kg/m 0.2351 kg/m
Spanwise elastic axis 50 % chord 44.79 % chord
Center of gravity 49 % chord 41.81 % chord
Fig. 11 Difference between

the elastic axis and center of
gravity location in present
study and experimental study
[11] (not to the scale)
5 Mesh and Boundary Conditions
5.1 Finite Elements Analysis
Structural mesh/grid for the present study was generated using ANSYS Meshing. For
flat composite plate wing divergence study, geometry was very simple and gives us
freedom to mesh it with mapped elements. ANSYS Workbench provides automatic
FEM element assigning capabilities, thus SOLID185 elements were used to discretize
this geometry. SOLID185 element has eight nodes where each node has three degrees
of freedom, translations in nodal x, y, and z directions [12].
As for large A R wing divergence study geometry, mapped meshing was not pos-
sible due to airfoil cross-section, thus meshed with paved hexahedral elements with
sweeping along the span. SOLID186 elements were used to discretize this geometry,
as they are well suited for irregular meshes where elements may have any spatial
orientation. SOLID186 element has 20 nodes where each node has three degrees of
freedom, translations in nodal x, y, and z direction [12].
Root of the wing is fixed in all the directions for displacement; gravitational
acceleration was applied in the negative x direction for flat composite plate wing
divergence study (similar to the experimental setup) and for the next study in negative
y direction (Fig. 13).
5.2 Computational Fluid Dynamics
ANSYS ICEM CFD software was used to discretized the CFD domain into small
tetrahedron control volumes and triangular surface elements. Smooth variation of
element size is maintained from near the wing to outer boundaries with a growth rate
of 1.1, see Fig. 12.
In order to capture near wall effects, prism layers were created based on a non-
dimensional turbulence quantity known as wall y + , which is the function of first layer
height of mesh layer around the body, wall shear stress and local flow velocity. It
needs to be closer to 1 for resolving the boundary layer gradients, leading to accurate
calculation of wall forces. Prism layers were created using first layer thickness and
Fig. 12 Mesh for CFD analysis
Fig. 13 Near wall mesh on wing surface
growth ratio. The first layer thickness was chosen based on the grid independence
study, 0.00001 m, which was providing wall y + ≈ 1, excellent to resolve boundary
layer accurately using Spallart-Allmaras turbulence model.
CFD domain faces were named as inlet, outlet, symmetry and wing, as shown
in Fig. 10. Inlet was considered as velocity-inlet with velocity vector defined by
magnitude and direction specification method for root angle of attack, αr . Outlet was
considered to be at standard sea level atmospheric pressure using pressure-outlet
boundary condition. Symmetry was specified to maintain the flow symmetry for the
half wing model.
Wing was specified as no-slip wall boundary as well as system-coupling interface
using dynamic meshing. Volumetric elements were specified to be deforming using
diffusion-based smoothing and remeshing algorithms to maintain the acceptable
mesh quality while absorbing and adjusting the abrupt changes due to wing-boundary
displacement [13].
6 Solver Setup
6.1 Computational Fluid Dynamics (CFD)
Fluid flow and transport phenomena are governed by the conservation of mass,
momentum and energy. These governing equations are solved based on the method
known as Finite Volume Method (FVM), which includes dividing the CFD domain
into small control volumes, integration of governing equations over the control vol-
umes of the domain, discretization of resulting integral governing equations to obtain
a set of algebraic equations and finally solution of these algebraic equations by an iter-
ative method. For present study, three-dimensional Navier-Stokes equations with one
equation Spalart-Allmaras turbulence model available in ANSYS Fluent is selected.
A brief outline of the background procedure is given below [13]:
Decomposing the instantaneous Navier-Stokes equations into mean and fluctuat-
ing quantities (vector and scalar) and taking time average, results in the ensemble-
average momentum equations known as Reynolds-Average Navier-Stokes (RANS)
equations given as:
∂ρ ∂
+ ( ρu i ) = 0 (3)
∂t ∂ xi
∂ ∂
( ρu i ) + ( ρu i u j ) (4)
∂t ∂ xi

∂p ∂ ∂u i ∂u j 2 ∂u l ∂
=− + μ + − δi j + ( −ρu i u j )
∂ xi ∂x j ∂x j ∂ xi 3 ∂ xl ∂x j
where, the last term in Eq. (5) represents the turbulence effect and Reynolds stresses,
( −ρu i u j ) , are required to be modeled towards the first step in turbulence modeling.
Which is formulated based on Boussineq hypothesis [13]. From the various turbu-
lence models available in ANSYS Fluent, Sparlart-Allmaras model was selected
in order to solve kinematic eddy viscosity, for which the transport equation of the
turbulent kinematic viscosity, ν̃, is given by:
∂ ∂
( ρ ν̃) + ( ρ ν̃u i ) (5)
∂t ∂ xi

1 ∂ ∂ ν̃ ∂ ν̃ 2
= Gν + ( μ + ρ ν̃) + Cb2 ρ − Yν + Sν̃
σν̃ ∂ x j ∂x j ∂x j
where, G ν is the production of turbulent viscosity, Yν is the turbulent viscosity

destruction which occurs because of the viscous damping and wall blocking at the
near-wall region, ν is molecular kinematic viscosity, Sν̃ is source term, and σν̃ and Cb2
are the constants. The setup of steady state in-compressible flow simulation (for static
aero-elasticity) is established by discretized CFD domain using ANSYS Fluent see

Fig. 12. ANSYS Fluent solves the RANS equations and the transport equation based
on Finite Volume Method (FVM). Where second-order upwind spatial discretization
scheme was selected along with SIMPLE algorithm; developed by Patankar and
Spalding [14], to formulate pressure-velocity coupling. Convergence was achieved
till the residuals reached below sixth order decimal place or the forces are converged.
6.2 Finite Element Method (FEM)
Finite Element Method (FEM) is based on the finite element idealization, the struc-
tural model is discretized into elements (discrete regions) with connectivity at the
nodes to each other. Current static structural analysis was performed using ANSYS
Mechanical APDL. The equilibrium equations of a structural static analysis are:
[ K ] {u} = {F a } + {F r } (6)
N
where, {F r } is reaction load vector, [ K ] = m=1 [ K e ] is global stiffness matrix,
{u} is nodal displacement vector, [ K e ] is element stiffness matrix, and {F a } is the
total applied load vector, formulated as:
N
{F a } = {F nd } + {F ac } + {Feth } + {Fepr } (7)
m=1
N
where, {F r } = −[ M] {ac } is acceleration load vector, [ M] = m=1 [ Me ] is total
mass matrix, [ Me ] is element mass matrix, {ac } is total acceleration vector, {Fet h} is
p
element thermal load vector, and {Fe r } is element pressure load vector.
The above derived structural models linear equilibrium equations can be solved
by one of the type of solvers available in ANSYS Mechanical APDL, which are
direct and iterative solvers. For current FEM analysis direct solver has been selected,
which uses a direct elimination process known as Gaussian elimination approach to
solve the unknown vector, {u}, by employing LU decomposition [15].
6.3 System Coupling
Once the setup for ANSYS Mechanical APDL and ANSYS Fluent was complete,
ANSYS Coupling Service available in ANSYS Workbench was used to setup the
two-way coupling between the two domains (FEM-CFD) [9]. For system coupling,
10 coupling steps were selected, each with maximum 5 iterations, where at each
iteration Mechanical APDL and Fluent perform analysis, in a sequential manner, till
convergence is achieved, data transfers were established in-between the two domains
with RMS convergence criteria of each data transfer equal to 0.01 which was achieved
at each coupling step. By default, ANSYS Workbench [9] selects the data mapping
algorithms as General Grid Interface (GGI) for force transfer from Fluent to Mechan-
ical APDL, and Smart Bucket Algorithm displacement transfer from Mechanical
APDL to Fluent. Once the simulation is updated to proceed towards the solution, it
will follow the process as described in Fig. 5.
6.4 Input Parameters
Current study was performed with following fluid flow properties, air density, ρ, and
dynamic viscosity, ν, at various freestream air velocities, U∞ , and root angles of
attack, αr , for CFD solver, see Table 2.
Material properties for the flat plate composite wing are taken from the exper-
imental study performed by Landsberger et al. [4], are given in Table 3. Material
properties for simplified geometry for large A R wing divergence study were calcu-
lated based on the flap bending rigidity, chordwise bending rigidity, torsional rigidity,
and mass per unit length, by Tang and Dowell [11], see Tables 1 and 4.
Table 2 Input properties—computational fluid dynamics

Parameter Value
Air density, ρ 1.225 kg/m3
Dynamic viscosity, ν 1.7894 kg/ms
Flat composite plate wing divergence study
Freestream velocity, U∞ 1.75, 5, 6.25, 8.25, and 11.5 m/s
Root angle of attack, αr 2, 4, 6, 8, 10, 12, 14, 16, 18 and 20◦
Large AR wing divergence study
Freestream velocity, U∞ 10–35 m/s
Root angle of attack, αr 2.2◦
Table 3 Material properties of Hercules ASI/3501-6 graphite/epoxy

Parameter Value
Young’s modulus in lateral direction, E L 130 × 109 Pa
Young’s modulus in transverse direction, E T 10.5 × 109 Pa
Poission’s ratio, νLT 0.28
Shear modulus, G LT 6.0 × 109 Pa
Ply thickness 1.34 × 10−6 m
Density 1520 kg/m3
Table 4 Material properties of custom material

Parameter Value
Young’s modulus in x-direction, E x 9.2342 × 108 Pa
Young’s modulus in x-direction, E y 6.1 × 108 Pa
Young’s modulus in x-direction, E z 9.2342 × 108 Pa
Poission’s ratio, νx y 0.33
Poission’s ratio, ν yz 0.33
Poission’s ratio, νx z 0.33
Shear modulus, G x y 5.2923 × 108 Pa
Shear modulus, G yz 2.0333 × 108 Pa
Shear modulus, G x z 3.0781 × 108 Pa
Density 1108.2 kg/m3
7 Results and Discussion
7.1 Grid Independence Study
To select the grid for FSI analysis, the grid independence study was performed
on the geometry of flat composite plate wing, Fig. 8a. First, for the finite element
analysis four grids were generated with increasing number of nodes, see Table 5.
Static structural analysis was performed with uniform force applied on the top and
bottom surface of the flat plate wing and wing twist was plotted along the span of
the wing, as shown in Fig. 14a.
Based on the FEM grid independence study, it is clear that the change in the
wing twist is significantly small with average percentage of 0.004 in between FEM
Grid 3 and FEM Grid 4 as well as in between FEM Grid 2 and FEM Grid 3 with
0.02 % but in order to have excellent data mapping fine mesh is required, therefore
FEM Grid 3 was selected for the further studies. For CFD, another four grids were
generated, details are given in Table 5. Using the FEM Grid 3 and four CFD grids,
four FSI simulations were carried out and wing twist was calculated along the span
of the wing, see Fig. 14b. Average difference in between CFD Grid 3 and CFD Grid
Table 5 Grid size details

Grid ID Number of elements
For FEM For CFD
Grid 1 2,706 3,145,030
Grid 2 22,692 3,913,983
Grid 3 138,624 5,461,031
Grid 4 249,084 6,644,600
(a) Finite Element Analysis. (b) Fluid-Structure Interaction.
Fig. 14 Grid independence study
4 was found to be of 0.56 % while in between CFD Grid 2 and CFD Grid 3 was
1.43 %, thus CFD Grid 3 with approximately 5.4 million elements was selected for
present study. For other wing models, similar grid resolution was maintained in all
the three directions while maintaining wall y + ≈ 1 for CFD analysis in order to
resolve boundary layer gradient.
8 Validation
Present study is based on the static two-way fluid-structure interaction analysis using
ANSYS Workbench involving the CFD and FEM solvers. In order to validate the
current methodology, two experimental studies were selected and the static two-way
FSI analysis was performed and observations are presented below:
8.1 Flat Composite Plate Wing Divergence Study
Two-way FSI analysis was performed for the static aeroelasticity based on the com-
posite flat plate wing tested by Landsberger et al. [4]. Static deflections for composite
wing of configuration [+302 /0]s , the wing tip flap wise displacement and the wing tip
twist, are presented in Fig. 15. It was found that the wing tip flap wise displacement
obtained from current method and the experiment performed by [4] are found to be
in very good agreement for free stream velocities, U∞ = 5 and 11.5 m/s. Whereas,
the wing tip twist was found to be having similar trends as of the experimental
study, while over predicting the twist similar to the analytical method adopted by
Landsberger et al. [4].
Divergence dynamic pressure and speed was calculated by putting 1/θ = 0 in Eq.
(2) for each wing configuration using wing tip twist at different freestream velocities,
same equations for each wing configuration can be seen in Table 6, which were
(a) Flap wise tip displacement. (b) Tip twist.
Fig. 15 Static aeroelastic deflection of Landsberger et al. [4] versus present FSI analysis for com-
posite flat plate wing of configuration [+302 /s]s
Table 6 Divergence speed calculation for present FSI analysis

Plate # Equation of fitted straight line
[ −452 /0] s 1/θ = 29.091/q − 0.2127
[ −302 /0] s 1/θ = 26.750/q − 0.2089
[ −152 /0] s 1/θ = 34.928/q − 0.1895
[ +302 /0] s 1/θ = −50.27/q − 0.2143
(a) Calculation. (b) Comparison.
Fig. 16 Divergence speed; Landsberger et al. [4] versus present FSI analysis for composite flat
plate wings
obtained from Fig. 16a. The comparison of the divergence speed is presented in Table
7 and Fig. 16b, where it can be seen that the current method of two-way aero-elasticity
is in excellent agreement with experimental results for two wing configurations,
[−152 /0]s and [−302 /0]s , providing better prediction then the analytical Rayleigh-
Ritz method. Whereas, for wing configuration, [−452 /0]s , divergence dynamic speed
was over-predicted with a deviation of 10.35 %, this offset is similar to the Rayleigh-
Ritz method adopted by [4]. The reason for which was found to be the use of only
Table 7 Divergence speed, Landsberger et al. [4] versus present FSI analysis
Plate # Divergence speed, UD
Landsberger et al. [4] Present Error %, w.r.t
Exp. Ritz FSI Exp. Ritz
[ −452 /0] s 13.480 m/s 14.090 m/s 14.944 m/s 10.35 6.06
[ −302 /0] s 14.575 m/s 14.068 m/s 14.460 m/s 0.12 2.79
[ −152 /0] s 17.238 m/s 17.990 m/s 17.298 m/s 0.08 3.84
[ +302 /0] s N/A N/A N/A – –
(a) Wash-out of [−302 /0]s . (b) Wash-out of [+302 /0]s .
Fig. 17 Static aero-elastic deflection at U∞ = 8.25 m/s and αr = 4◦ illustrating the wash-out and
wash-in effect
in-plane properties to define overall material properties of the model, leading to a

different torsion-bending coupling as compared to the experimental setup.
From the fitted straight line equation for wing configuration, [+302 /0]s , it was
found that the divergence dynamic pressure for this particular case is negative, thus
the divergence speed cannot be calculated, results to the conclusion that this wing
configuration with positive ply fiber orientation will not diverge owing to its wash-in
effect as founded by Landsberger et al. [4]. The wash-in and wash-out of the wing
can been seen in Fig. 17. Overall, it is sufficiently clear that the two-way FSI method
is in good agreement with the composite flat plate wing experimental results [4].
8.2 Large AR Wing Divergence Study
Another experimental study [11], performed on the high-aspect ratio wing was repli-
cated with a simplified wing model and static two-way FSI analysis was performed
and static aero-elasticity deflections were obtained for αr = 2.2◦ with respect to
freestream velocities, see Fig. 18.
(a) Flapwise tip displacement. (b) Tip twist.
Fig. 18 Static aero-elastic deflections by Tang and Dowell [11] Versus Present Fsi Analysis at
αr = 2.2◦
(a) U∞ = 10.0 m/s. (b) U∞ = 25.0 m/s.
(c) U∞ = 28.0 m/s. (d) U∞ = 35.0 m/s.
Fig. 19 Static aero-elastic displacement contours of large A R wing
Wing tip flap wise displacement obtained from two-way FSI analysis as compared
to the experimental study of Tang and Dowell [11], was found to be following the
same trend although under predicting the finds for freestream velocities greater the
30 m/s. Whereas, wing tip twist deviates from the experimental results as two-way
FSI over-predicts the twist for freestream velocities 24 m/s and over-predicts for
freestream velocities above 24 m/s, which is the effect of elastic axis location as it
is at 44.79 % chord for present study whereas, it is at 50 % chord for experimental
study. Which creates shorter moment arm between the center of pressure and elastic
axis resulting the torsion-bending coupling to mismatch with experimental setup.
With more detailed availability of material properties in the literature developing
and performing the validation would be excellent.
Figure 19 illustrates the displacement contours on large A R wing at various air-
speeds. It is clear from the contours how lift influences the wing behavior and develops
to overcome the gravity force to balance the wing in-between U∞ = 25.0 and 28.0
m/s.
9 Conclusion
Object of the present study was to create baseline for aero-elastic analysis using
ANSYS solver package. Which was successfully executed with following observa-
tions:
1. Two-way FSI analysis with linear structural model and non-linear fluid model
was efficiently able to obtain the static aero-elastic properties of a composite wing
while capturing the effect of composite materials unidirectional properties with
maximum 10.35 % deviation in divergence speed as compared to experimental
results by Landsberger et al. [11].
2. Whereas for Large A R Wing Divergence Study, wing tip twist over wide range of
freestream velocities found to be out of trend due to the mismatch in elastic axis
and center of gravity in between experimental wing model and simplified FEM
model as well as insufficient material properties. Wing tip flap displacement was
found to be following the trend of experimental study by Tang and Dowell [11]
with acceptable agreement.
3. From Large A R Wing Divergence Study, it was clear that the exact material
properties are required in order to predict the results of a real wing, which makes
the two-way FSI analysis method susceptible to the measurement of wing material
properties and section properties.
4. As the flat plate composite wing geometry was of low-aspect ratio, the present
linear structural model was sufficient to obtain the static deflections. Although
Large/High A R Wing Divergence Study performed well but it sure will provide
more insight when geometric non-linearities will be included into the FEM model.
Overall, the two performed validations create a strong baseline for aero-elastic
analysis based on partitioned static two-way Fluid-Structure Interaction analysis
using ANSYS package, which will be used to extend to the dynamic aero-elasticity
analysis (flutter with and without active control, LCO, etc.), and validate reduced
order models against these benchmark results.
References
1. Gundlach J (2012) Designing Unmanned Aircraft Systems a comprehensive approach. AIAA

Education series
2. Afonso F, Vale J, Oliveira E, Lau F, Suleman A (2017) A review on non-linear aeroelasticity
of high aspect-ratio wings. Prog Aerosp Sci 89(2017):40–57
3. Bennett RM, Edwards JW (1998). An overview of recent developments in computational
aeroelasticity by aeroelasticity branch, structures division. AIAA J
4. Landsberger BJ, Dugundji J (1985) Experimental aeroelastic behavior of un-swept and forward-
swept cantilever graphite/epoxy wings. J Aircraft 22(8):679–686
5. Chen G, Dugundji J (1987) Experimental aeroelastic behavior of forward-swept graphite/epoxy
wings with rigid-body freedom. J Aircraft 24(7):454–462 https://doi.org/10.2514/3.45501
6. Hollowell SJ, Dungundji J (1984) Aeroelastic flutter and divergence of stiffness coupled,
graphite/epoxy cantilevered plates. J Aircraft 21(1):69–76
7. Hodges D, Pierce G (2011) Intorduction to structural dynamics and aeroelasticity, 2nd edn.
Cambridge University Press, Cambridge
8. Blair M, Weisshaar Ta (1982) Swept composite wing aeroelastic divergence experiments. J
Aircraft 19(11):1019–1024
9. ANSYS Workbench User s Guide (2017) (April)
10. Raja RS (2012) Coupled fluid structure interaction analysis on a cylinder exposed to ocean
wave loading. Chalmers University of Technology, Masters thesis 2012, p 55
11. Tang D, Dowell EH (2001) Experimental and theoretical study on aeroelastic response of
high-aspect-ratio wings 39(8)
12. ANSYS Mechanical APDL Element Reference (2017) (April)
13. ANSYS Fluent Theory Guide (2017) (April)
14. Patankar SV, Spalding DB (1972) A calculation procedure for heat, mass and momentum
transfer in three-dimensional parabolic flows. Int J Heat Mass Transfer 15:1787–1806
15. ANSYS Mechanical APDL Theory Reference (2017) (April)
Numerical Study of Effect of Adjacent
Blades Oscillation in a Compressor
Cascade
Shubham, M. C. Keerthi, and Abhijit Kushari
Abstract The present numerical study deals with the steady and unsteady aerody-
namics of an airfoil in a cascade with oscillating neighboring blades. The motivation
for the study arises from aeroelastic studies of turbomachinery blades, where the
unsteady forces acting on a blade due to different sources is examined. In order to
identify the root cause of the various phenomena affecting the aeroelastic stability,
it is necessary to observe the effect of only one source of disturbance and isolate
the others. The cascade comprises of five blades with zero stagger and low inci-
dence. The two adjacent blades to central one are oscillated with a fixed frequency
and phase difference, with the rest of the blades remaining stationary, covering a
range of frequencies and phase difference angles. The primary objective is to look
at the variation of global parameters on central blade with reduced frequency. The
moment and drag are estimated numerically and compared with experimental results
which shows good agreement. The hysteresis loops of lift and moment coefficient
with angular displacement are used to understand the effect of reduced frequency.
A laminar separation bubble is observed to be formed during part of the oscillation
cycle and its size is related to the unsteady forces on the blade. The behavior is also
a function of reduced frequency to some extent. Such an understanding of the effect
of oscillating blades in a blade row is essential for the modeling of the aerodynamic
forces in an aeroelastic problem.
Keywords Aeroelasticity · Unsteady aerodynamics · Linear cascade
1 Introduction
Due to the ever-increasing need to increase the thrust to weight ratio and efficiency
of turbomachines, their blades are designed to be highly loaded and are done so with
minimal mass. This results in blades that are thinner, which are also subjected to
large aerodynamic loading. Specifically, due to the unsteady nature of the loads, this
Shubham · M. C. Keerthi · A. Kushari (B)

Indian Institute of Technology Kanpur, Uttar Pradesh, Kanpur 208016, India

60 Shubham et al.
has led to increased aeroelastic issues in blades. Aeroelastic instability prediction

methods are greatly needed as modern jet engines require blade designs close to the
stability boundaries of the performance map. There have been extensive undertakings
in the theoretical, experimental and computational understanding of the significant
physical components and to enhance the predictive ability of turbomachinery blade
flutter [1, 2]. However, even today, the understanding is not well enough to address
aeroelastic issues at the design phase of an engine development program. Hence,
more studies are required to investigate all aspects of the problem. One aspect of
the research is in the unsteady aerodynamics of oscillating blades in a cascade. This
is to address the requirements for modeling the unsteady aerodynamics, which is
particularly challenging in a turbomachine environment.
These issues start to some extent from an inability to model the unsteady aero-
dynamic behavior of cascades. Additionally, these models must be computationally
efficient in the event that they are to be a part of a useful design framework. Of late,
several investigators [3–6] have developed time-accurate Euler and Navier–Stokes
solvers for analyzing unsteady flows in turbomachinery. These models can provide
for small and large disturbances unstable flows in cascades. Atassi [7] developed
a linearized analytical theory for oscillating cascades in uniform incompressible
flows. Verdon and Caspar [4, 8] and Verdon and Usab [9, 10] developed a linearized
theory for solving oscillating cascade unsteady loads imposed on a steady-state solu-
tion. Ekaterinaris and Menter [11] assessed the capacity of one-and two-equation
turbulence models in predicting hysteresis effects of unsteady fully turbulent flow
over oscillating airfoils. The Baldwin-Barth [12], the Spalart-Allamaras [13], and
the SST k−ω [14] models demonstrate a noteworthy change over standard two-
condition models. The SST k−ω model gave good predictions for the attached and
the light-stall cases.
For a bladerow whose aeroelastic stability is being assessed, the aerodynamic
damping of a blade is of paramount importance, as a large negative value can lead
to large-amplitude vibrations. It is known that the aerodynamic damping of a blade
at a given inlet condition is influenced by the vibration of all blades in the bladerow
including itself. In reality, it is only the immediate neighbors that significantly affect,
with the blades further apart having a diminishing influence. For the most part, there-
fore, the aerodynamic influence of the motion of two adjacent blades and the blade
itself will determine the aeroelastic stability of the blade. The present study focuses
on the aerodynamic effects caused on a blade by the oscillating adjacent blades.
The effects considered are the global parameters such as lift, drag and moment, as
well as local effects like flow separation in the vicinity of the leading edge will be
investigated. The unsteady pressure gradient at the leading edge plays a critical role
in the formation of leading edge separation bubble [15]. An important application
of the present analysis is to use the detailed distribution of the pressure obtained
particularly close to the leading edge area for analyzing the onset of boundary layer
separation that eventually could lead to separation.
This study is performed to recreate the linear oscillating cascade experiment whose
details can be found here [16]. Experiments enable us to obtain data only at discrete
points and usually with poor spatial resolution. To have a complete flow field of
Numerical Study of Effect of Adjacent Blades … 61
oscillating cascade aerodynamics, numerical computation fills the crucial missing

gap. The cascade of five airfoils has been modeled as a two-dimensional numerical
model to reduce computation costs. The analysis has been performed for a fixed
incidence angle, four different frequency (6, 9, 12, 15 Hz) with varying phase differ-
ence angles. The results have been compared for different reduced frequency and
phase difference angles. The phase difference (PD) used in this study is an essential
parameter, as it effectively simulates the inter-blade phase angle that is present in a
real vibrating turbomachine blade row. The inter-blade phase angle is a result of the
excitation of one or more nodal diameters of the rotating bladed-disk.
2 Numerical Setup
The setup for numerical analysis is kept same as that of experiment so as compare the
behavior and variation of parameters. Figure 1 represents the computational domain
with boundary and initial condition mentioned in the figure itself.
The cascade is modeled numerically by employing a commercial solver ANSYS
CFX using a full-scale time-marching model. The model extends over five blades
with walls on both sides. The chord of the blade is 152 mm and the upstream and
downstream domain length after blade is also same as chord. The airfoil for the blade
used is EPFL standard test configuration 1. The cascade is at zero stagger, incidence
of 6° and amplitude of oscillation is 3°. The blades are oscillated for three cycles.
The simulations are conducted using a standard SST turbulence model with wall
functions.
The motion of the oscillating blade is described by a set of equations and imposed
as moving mesh boundary condition, deforming the mesh around the reference blade
in each time step. Blade oscillation is thereafter introduced first after having the flow
Fig. 1 Computational domain

62 Shubham et al.
field reaching a steady state. An inflow velocity of 29 m/s with a turbulent intensity
of 5% is specified at the upstream boundary. Airfoils are specified to be the wall
with no slip. Pressure is specified at the outlet boundary to be atmospheric pressure
(101.325 kPa). The governing equations are solved in a stationary reference frame.
Convection terms are discretized with second order upwind schemes. Convergence
criteria for absolute error of all variables used are 1 × 10−5 . Table 1 shows the values
for grid independence study. The study has been performed for three meshes namely
low, medium and high. The low density corresponds to 0.1 million nodes, medium
0.2 million nodes and high 0.4 million nodes. It can be clearly observed that medium
mesh density (Fig. 2) can be used for our further study as the percentage change in
lift and drag values are very small.
For validation, experimental data [17] obtained from a linear cascade with five
blades is used. In the experiment, the central blade has surface pressure taps at the
mid-span for unsteady pressure measurement, with the total span being 300 mm. The
flow conditions used are identical to those used for the validation case. It has to be
noted that numerical setup used above simulates the effect of side walls present in the
experimental setup, but does not account for any possible flow variations in the out-of-
plane direction due to the effect of end walls. The experimental measurements were
sampled at 500 samples per second for a duration of 10 s, resulting in about 100 cycles
of oscillation. Such a large sample size was necessary to ensure low uncertainties
in the phase-averaged results for the experiment. However, the numerical solution
Table 1 Grid independence study

Mesh density No. of elements Drag (N) Lift (N) % change drag % change lift
Low 104,290 0.00554 0.08510 – –
Medium 207,680 0.00548 0.08403 1.09 1.27
High 481,670 0.00543 0.08329 0.92 0.88
Fig. 2 A portion of the structured mesh

showed flow variables fluctuating with good periodicity and hence only about three
to four cycles are used to avoid excessive computing time.
3.1 Validation
The validation has been shown for steady-state and unsteady blade surface pressure.
Figure 3a shows steady static pressure coefficient variation with the non-dimensional
chord length. The negative x/c values represent suction surface from −1 to 0 and
positive x/c values corresponds to pressure surface location from 0 to 1. The slight
acceleration on the suction surface followed by a deceleration can be seen from
Fig. 3 Steady pressure

coefficient and pressure time
history for numerical and
experimental case
64 Shubham et al.
Fig. 4 Motion of adjacent airfoils at PD = 90°
the change in slope for negative x/c values. The presence of the accelerating region
implies adverse pressure gradient and a possibility of flow separation, although none
was observed for this case where all blades were stationary. The agreement between
numerical and experimental values is good. The numerical value for suction surface
follows closely with experimental ones while there is slight under-prediction for
pressure surface values so is the suction surface values near the leading edge.
Figure 3b shows unsteady pressure coefficient along the chord for numerical
and experimental case. The trend of variation seems to be similar in both of them.
The experimental case seems to have the presence of higher frequency components
which are partly due to flow unsteadiness and measurement uncertainty. This high-
frequency component is not captured in the present numerical solution. In the study
of aeroelasticity in turbomachines, it is usually that the fundamental frequency of
the blade will contribute to most of the aeroelastic stability, with the contribution
diminishing from the higher harmonics. As a result, the numerical scheme, which
appears to cut off the high-frequency phenomena, is considered to be adequate
for the purposes of this study. The lift, drag, moment coefficient data predicted
by the numerical simulation is comparable to the experimental values obtained by
Babu [16]. Figure 4 shows the time series of the angular displacement of blades -1
and +1 when the value of PD is 90°.
3.2 Effect of Reduced Frequency
The reduced frequency and interblade phase angle both are the most important param-
eters for aeroelastic studies. The interblade phase angle represents the structural
response of the system while reduced frequency is related to the flow field around
the blade. The reduced frequency can be defined as the ratio of time taken by a fluid
particle to travel a chord length to time period of oscillation.
ωc c/U∞
k= = 2π
U∞ T
where c = chord; U ∞ = freestream velocity; ω = Angular frequency of oscillation;

T = time period of oscillation.
In this paper, the effect of reduced frequency on the behavior of forces on the
blades is numerically investigated. The validation is already been established in the
previous section.
Figures 5 and 6 represents the variation of moment coefficient with time for four
reduced frequencies at PD = 0° and 180°. The time axis has been normalized between
0 and 1, which is within the range total time for three cycles, in order to compare
the different frequency cases. The plots have been shown for two PD values. These
two specific phase differences are chosen as they represents blades moving exactly
in phase and out of phase with each other.
The plot at PD = 0° shows little variation with reduced frequency but for PD =
180°, the variations are getting a little offset with change in reduced frequency. This
is mainly due to the presence of higher frequency components as confirmed from
their Fourier amplitude spectra. There is also a reduction in amplitude as the reduced
frequency is decreased. The lift coefficients plots are also similar to moment plots
only varying by a scale factor.
Figures 7 and 8 show similar plots for drag coefficient against normalized time.
The amplitude of drag remains the same across both phase difference and reduced
frequencies discussed. There is a presence of higher frequency component in reduced
frequencies variation at PD = 180°. The drag coefficient for reduced frequency of
Fig. 5 Moment coefficient

versus time variation for
reduced frequencies at PD =
0°
66 Shubham et al.
Fig. 6 Moment coefficient

180°
Fig. 7 Drag coefficient

0°
Fig. 8 Drag coefficient

180°
0.19 is varying slightly different than others at both PDs. The reasons seems to be
more significant presence of higher frequency components at PD = 180° than earlier
and at PD = 0°, this presence of higher frequency is introduced at this reduced
frequency itself. This low frequency forced response has introduced this unique
attribute for lower phase differences.
It can be seen from the variation of the drag forces that the amplitude is low
compared to the lift and its variation across PDs and reduced frequencies are similar.
This is suggestive of the fact that the phenomena causing the amplitude change and
non-linearity is affecting the transverse force more than the axial force. There is also
strong correlation between lift and moment is likely due to the flow phenomena on the
surface of the blade that is close to the leading and trailing edges. This also explains
the reduced effect on drag, as the pressure drag component is more influenced by the
regions upstream and downstream of the blade than the lateral surfaces.
Figure 9 shows the hysteresis behavior of blade forces in a cycle of motion.
Two important distinctions between this analysis and the hysteresis plots in typical
dynamic stall studies of isolated airfoil are to be noted: firstly, in the present study
the force and moment measured is on a static blade whereas dynamic stall studies
reference the force and moment measurements to the same blade which is oscillating;
secondly, the present oscillation amplitude (3°) is below the limits required to cause
the typical sequence of events of dynamic stall including the formation and spillage
of leading edge vortex followed by a massive suction-side separation and the inter-
action with the trailing edge vortex [18]. It is clear that even with a small amplitude
oscillations of the adjacent blades is sufficient to qualitatively change the behavior
of the static blade. The plots thus far establish that the reduced frequency is weakly
affecting the blade force responses as compared to interblade phase differences. The
hysteresis loop area is increasing with increase in reduced frequency. Though the
increase is very low it can be established that reduced frequency is affecting the
hysteresis. The mechanism for an oscillating blade to affect the pressure field of the
adjacent blade is now explored. To do this, the pressure field contours around blade 0
Fig. 9 Variation of area of

hysteresis loop
68 Shubham et al.
is compared at different instances of time period fraction (t/τ) within one oscillation.
Figure 10a, b show a grid of the pressure contours for all PD values presented as a
vertical columns with time fractions as horizontal rows. First, considering the PD =
0° case, we can see a substantial variation of the pressure field on the pressure and
suction surface sides throughout the cycle.
In particular, the leading edge stagnation point is characterized by different values
of pressure and location. The lift and moment acting on the airfoil is significantly
affected by the location of the leading edge stagnation point, as it lies in the region
of maximum pressure acting on the airfoil surface. Additionally, a dark region corre-
sponding to very low pressure can be seen on the suction surface near the leading
edge. It is interesting to note that although the low pressure region of the suction
surface varies throughout a cycle, its phase within a cycle does not vary with PD. As
can be seen from combined phase lag plots, the variation is not affected much with
change in reduced frequency.
If we compare the PD = 0° and 180°, the differences can be observed. This is due
to the fact that that the field on the pressure surface side of blade 0 where the leading
edge stagnation point region occurs is influenced by the blade closest to it, namely
blade +1, whose displacement within a cycle shown is dependent on PD.
3.3 Effect of Stagnation Point Location on Reattachment

Length
It is usually recognized that the prediction of separated flows using RANS is chal-
lenging. However, presently, the trends of variation of parameters such as reattach-
ment length are of interest. As they are consistent with the trends in variation of the
global forces such as lift, it is believed that the present choice of solver and its param-
eters is capable of supporting the proposed conclusions. In this section, separation of
flow on the suction surface of airfoil is discussed along with length of reattachment.
The reattachment length is the size of the separation bubble present on the suction
surface of the blade at a given time. Figure 11 shows the formation of separation
bubble, length of reattachment and location of stagnation point on the central and both
adjacent oscillating airfoils. The oscillating blades show a much larger reattachment
length compared to the non-oscillating central blade. The oscillating airfoils have
a separation bubble on their suction surface. This figure at PD = 0° and t/T = 6/8 is
for representation of flow features for further reference in the discussion.
Figures 12 and 13 shows reattachment length on blade 0 with time at PD = 0°
and 180°. There is a significant decrease in size of reattachment length when PD is
changed from 0° to 180°. The bubble sizes at PD = 0° is similar for all the reduced
frequency value discussed whereas for PD = 0°, a monotonic increase is seen with
an increase in frequency.
Figures 14, 15 and 16 show the maximum size of the separation bubble (or reat-
tachment length) for the three airfoils for all cases. Noting the vertical axis scales,
Fig. 10 Phase-log plot of pressure contour a PD = 0°, b PD = 180°

70 Shubham et al.
Fig. 11 Nomenclature and description of separation bubble and reattachment length
Fig. 12 Reattachment
length on blade 0 with time
at PD = 0°
it can be seen that the moving blades have a larger separation bubble compared to
the stationary one. However, despite the small size, its variation within the range
of PDs is considerable and is probably the most dominant contributor to the nature
of variation in the unsteady forces on the central blade. The reason for reduction in
moment amplitude (Figs. 5 and 6) and a slight reduction in drag amplitude (Figs. 7
Fig. 13 Reattachment
length on blade 0 with time
at PD = 180°
Fig. 14 Variation of the

maximum reattachment
length on Airfoil −1 with
phase difference
and 8) at PD = 180° can now be explained by noting the dip in separation bubble size
from Fig. 15. In particular, when the stagnation point is further downstream from
the leading edge of the airfoil’s suction surface, the maximum size of the separa-
tion bubble within the cycle is smaller and thus the load amplitude is smaller. This
is a consequence of the flow field when the adjacent blades are oscillating out of
phase. For a given PD, the change in reattachment length due to different reduced
frequency is small, although it can be seen that the lower reduced frequency reduces
the separation bubble size.
Although small, the size of the separation bubble is significant for the central
blade. This shows how the separation characteristics and the forces on a stationary
72 Shubham et al.

length on Airfoil 0 with
phase difference

length on Airfoil +1 with
phase difference
blade can be affected by the motion of neighboring blades. The minimum value of
reattachment length corresponds to values near PD = 180°, implying this is related
to the formation of lift whose amplitude also is maximum around PD = 180°. It is
possible that this influence on the central airfoil is brought about due to the changes
in the local incidence near the leading edge of blade 0. This change in incidence will
invariably have an effect on the location of the leading edge stagnation point, which
is examined subsequently.
Noting that the local flow incidence will directly affect the location of the leading
edge stagnation point, it can be said that the influence of the oscillating blades
is manifested through the pressure field to ultimately result in a local change in

incidence of blade 0.
4 Conclusions
The oscillating cascade has been studied for four different frequencies and various
interblade phase angles. After benchmarking with experimental results, efforts has
been made to understand the pressure variation and lags in the domain by varying
the parameters. The variation of moment and lift with reduced frequency showed
a small increase. The drag coefficients variation is much smaller. The change with
phase difference is more considerable than reduced frequency. Further, size of reat-
tachment bubble on central blade at phase difference 180° is increasing with increase
in frequency. For the range of reduced frequencies considered, the influence on the
unsteady pressure variation on the central blade was seen to be smaller compared to
phase difference of the motion of adjacent blades at a given frequency.
References
1. Fleeter S (1979) Aeroelasticity research for turbomachine applications. J Aircraft 16(5):320–

326
2. Cinnella P, De Palma P, Pascazio G, Napolitano M (2004) A numerical method for
turbomachinery aeroelasticity. J Turbomach 126(2):310–316
3. Hall KC, Clark WS (1993) Linearized Euler predictions of unsteady aerodynamic loads in
cascades. AIAA J 1(3):540–550
4. Verdon JM, Caspar JR (1980) Subsonic flow past an oscillating cascade with finite mean flow
deflection. AIAA J 18(5):540–548
5. He L (1990) An Euler solution for unsteady flows around oscillating blades. J Turbomach
112(4):714–722
6. Hall KC, Lorence CB (1993) Calculation of three-dimensional unsteady flows in turboma-
chinery using the linearized harmonic Euler equations. J Turbomach 115(4):800–809
7. Atassi H, Akai T (1980) Aerodynamic and aeroelastic characteristics of oscillating loaded
cascades at low mach number—part I: pressure distribution, forces, and moments. J Eng Power
102(2):344–351
8. Verdon JM, Caspar JR (1982) Development of a linear unsteady aerodynamic analysis for
finite-deflection subsonic cascades. AIAA J 20(9):1259–1267
9. Usab W, Verdon J (1991) Advances in the numerical analysis of linearized unsteady cascade
flows. J Turbomach 113(4):633–643
10. Verdon JM, Usab W Jr (1986) Application of a linearized unsteady aerodynamic analysis to
standard cascade configurations
11. Ekaterinaris JA, Menter FR (1994) Computation of oscillating airfoil flows with one- and
two-equation turbulence models. AIAA J 32(12):2359–2365
12. Baldwin B, Barth T (1990) A one-equation turbulence transport model for high reynolds number
wall-bounded flows. NASA TM 102847
13. Spalart PR, Allmaras SR (1992) A one equation turbulence model for aerodynamic flows.
AIAA J 94
74 Shubham et al.
14. Menter FR (1992) Improved two-equation k-omega turbulence models for aerodynamic flows.
NASA STI/Recon Technical Report No. 93
15. Watanabe T, Aotsuka M (2005) Unsteady aerodynamic characteristics of oscillating cascade
with separation bubble in high subsonic flow. In: ASME turbo expo 2005: power for land, sea,
and air. American Society of Mechanical Engineers
16. Babu KN (2008) Experimental investigation of the forced response of a cascade blade subjected
to transverse periodic aerodynamic loading. IIT Kanpur
17. Keerthi MC, Kushari A (2015) Effect of adjacent blade oscillation on the forces on a blade of
a compressor cascade. In: National propulsion conference, Mumbai
18. McCroskey WJ (1981) The phenomenon of dynamic stall. NASA TM-81264
Effect of Incoming Wakes on Losses
of a Low-Pressure Turbine of a Gas
Turbine Engine
Vishal Tandon, Gopalan Jagadeesh, and S. V. Ramana Murty
Abstract Modern civil aircraft engines are known for their high bypass ratio fans
that are powered by many low-pressure turbine (LPT) stages (Mahallati et al. in J
Turbomach 135/011010, [1]). LPT can contribute as much as 30% of the weight of
an aero engine (Sondergaard et al. in Toward the expansion of low-pressure turbine
airfoil design space, [2]; Curtis et al. in J Turbomach 119(3), [3]). Aerofoils of modern
LPT blades are subjected to increasingly stronger pressure gradients as designers
require higher blade loading in an effort to reduce weight and costly number of
LPT blades of an engine, which leads to better reliability and maintainability. This
decrease in number of LPT blades results in increase of thrust/weight ratio, thus
reducing the fuel consumption. But highly loaded aerofoils can reduce aerodynamic
performance, and its influences can be seen more in unsteady environment. Hence,
reduction of losses in unsteady environment, improves the turbine performance which
is a challenging task. Therefore, there are persistent efforts towards the generation
of “high-lift” blade profiles. As a result, industry and research communities are
motivated for further deep research efforts in LPT aerodynamics (Sarkar in J Turbo-
mach ASME 131, [4]). In this paper, the effect of the incoming wakes shed from
the upstream HPT blade on the downstream highly loaded transonic LPT vane are
studied to better understand the LPT flow physics. In this paper, wakes shed from
upstream HPT blade are simulated by cylinders of three different radius which are
of the order of actual HPT rotor blade trailing edge radius ~0.8 mm, so that only
the influence of wakes on vane losses can be studied. From these three cases, it is
observed that incoming wakes does not always lead to increase in vane loss coeffi-
cient. For some cases, it is observed that the loss coefficient is 12.64% lower than
the vane without any incoming wakes.
Keywords Unsteady fluid flow · Incoming wakes · Navier–stokes equations · LP

turbine vane · Pressure loss coefficient · Mach number · Strouhal number
V. Tandon (B) · S. V. Ramana Murty

Gas Turbine Research Establishment (GTRE), Bangalore, India
G. Jagadeesh
Indian Institute of Science, Bangalore, India

76 V. Tandon et al.
Nomenclature
St Strouhal number
U Flow velocity
P Total pressure
τ Shear stress
μ Dynamic viscosity
M Mach number
SST Shear stress transport
ω Specific dissipation rate
Tu Turbulence intensity
Yp Pressure loss coefficient
y Stagger angle
ξ Zweifel Loading
1 Introduction
The flow processes that occur in turbine stages are always very complicated. The
flow is three dimensional, viscous, and unsteady in nature. The flow may be incom-
pressible or compressible and may have subsonic, transonic, and supersonic regimes
that may be present concurrently in different regions. The primary flow is mainly
through the blade passages, in addition to that there are secondary flows which force
the fluid to move across the passages under the action of centrifugal and Coriolis
forces, blade loading effects results in causing incidence and deviation, flow short
circuit between the moving tips and the stationary shroud, boundary layer growth and
wakes shedding, and for transonic and supersonic blades, shock waves and shock-
boundary layer interaction in the blade passage and at the trailing edges. Another
class is unsteady effects, generated mainly by the interaction of adjacent blade rows
[5–7].
There is continued enhancement of LPT efficiency and performance. This has
resulted in intensified LPT aerodynamics research efforts by the research commu-
nities and industry. As a result, wide variety of different research approaches
were adopted. Research activities, e.g., steady and unsteady cascade aerodynamics
research and rotating turbine research, were undertaken. These research efforts have
contributed in better insight into the LPT flow physics. As a result, a modern LPT
design with a reduced number of blades was possible with marginal sacrifice of the
efficiency. This reduction in number of blades means increase in thrust/weight ratio
and results in decreasing the fuel consumption [8, 9].
In this research work, the effect of wakes shedded from the upstream blade on
the downstream low-pressure vane in a turbine stage is studied to better understand
the low-pressure turbine flow physics. In low-pressure turbines, the dominant source
of unsteadiness is produced by the wakes of the upstream blade rows. One of the
Effect of Incoming Wakes on Losses of a Low-Pressure … 77
significant consequences is the interaction of the wakes shed by upstream vane with
the downstream rotor blade suction side boundary layer. This boundary layer is often
laminar along most of its surface. The free stream turbulence also has some effect
on separation, but its role is less important to the wake effect [10, 11]. The transition
behavior of the flow can also be influenced by periodic and random disturbances [12–
16] and in some cases may reduce the risk of laminar boundary layer separation at low
Reynolds numbers [11]. In the last years, computational fluid dynamics (CFD) has
contributed to get a better understanding on the physics of flow separation, transition,
and reattachment under unsteady flow conditions [17, 18]. The effects of incoming
wakes on the transition behavior at different flux coefficients and Strouhal numbers
were investigated by a numerical URANS approach by Schwarze and Niehuis [19].
Meyer has done pioneer work in explaining the kinematics of wake convection
through a cascade passage [4, 20]. A substantial progress has been made over the past
3 decades in understanding the detailed influence of unsteady wakes on separated
boundary layers [21–28]. Initial open literature-based research on the aerodynamics
of more highly loaded low-pressure turbines appears to have been based on the family
of T104–T106 cascade aerofoils developed by Hoheisel et al. [2, 29]. Most of the
high-lift work presented in the open literature is cascade-based. The non-dimensional
Zweifel coefficient is approximately 1.04–1.07 for the T104–T106 family which
includes aerofoils with both front and aft loading. These and other related aerofoils
have been used in several cascade investigations by the research groups of both
Hodson [30] and Fottner [2, 31].
2 Description of the Problem
For the study of effect of incoming wakes on losses of a low-pressure turbine stator
vane, a highly loaded vane 2D aerofoil is selected. The vane aerofoil is designed
using Prichard 11 [32] parameter method, and the details are shown in Table 1.
Wakes shed from upstream high-pressure rotor blade are simulated by a cylinder,
so that we can study the influence of only wakes on the vane losses. Hence, upstream
of vane only cylinder is modeled not the whole high-pressure rotor blade. Also since a
unsteady simulation was planned which demands huge computational resources and
takes days sometimes even months to finish a single case, it is decided to carry out
first a 2D simulation and then later a 3D simulation can be taken up. Typical trailing
edge radius of high-pressure rotor blade is of the order of 0.8 mm. Trailing edge
Table 1 Non-dimensional
Parameter Value
parameters of LP Turbine
vane Chord (C) 63 mm
Stagger angle (y) 39.75°
Pitch/chord ratio 0.74
Zweifel loading (ξ ) 0.89
78 V. Tandon et al.
Table 2 Cylinder radius for

Case 1 Upstream cylinder radius 0.65 mm
3 simulations
needs to accommodate cooling holes for cooling of the high-pressure rotor blade,
and also, a lower trailing edge radius is not feasible from casting manufacturing point
of view. Hence, one radii below and two radii above the baseline radius were selected
for the current study as shown in Table 2.
3 Computational Methodology
The analysis is carried out by using the commercial ANSYS Fluent-14.5 [33] on
an IBM DX 360 parallel computing system. ANSYS Fluent is a three-dimensional,
multi-block and parallel fluid flow solver.
The unsteady Reynolds averaged Navier–Stokes (URANS) equations are solved.
The governing equations are discretized using finite volume method. The solution
algorithm is based on an implicit scheme coupled with multi-grid acceleration tech-
niques. k–ω SST turbulence model is selected to model the turbulence effects. It
is a two-equation turbulence model, and it is capable of correctly predicting the
wall shear stresses in flows where adverse pressure gradient is present similar to the
present study, and also, it is robust for complex flows.
4 Grid Generation
ICEMCFD 14.5 software has been used to generate the computational grid for this
study [34]. Structured grids are created using O, C, and H grids. The grids are
multi-block. For stator grid, first an O grid is placed around the airfoil. Around the
leading edge of stator and passage, a C grid is used and H grid has been used in the
trailing edge of the stator. For cylinder grid, also first an O grid is placed around
the cylinder, whereas H grid is used in the cylinder passage. The grids are joined
periodically in a many-to-one fashion in pitch direction. No overlaps are used while
attaching grids of non-moving stator and moving rotor grids. In unsteady simulation,
transient sliding interface option has been used which moves the cylinder grid to the
next relative location w.r.t the stator for each time step, which represents the true
physical position of the cylinder w.r.t stator for every time step. Flow parameters
are exchanged between the two grids by using a flux conservative algorithm. It is
a second-order interpolation algorithm. For steady state simulations, one stator and
one cylinder grids are modeled, whereas for frozen and unsteady simulations, one
stator and two cylinder grids are modeled.
Fig. 1 2D computational grid
For grid independence study, grids of 0.5 million (coarse), 2 million (medium)
and 4 million (fine) cells were considered. Same boundary conditions were used for
these three simulations. Grid independence was observed for 4 million mesh but it
is observed that for this mesh size the wakes are not properly resolved. When the
element size is reduced to 0.03 mm wakes were comfortably resolved as shown in
Fig. 3 but the cell count has become 8 million. Hence, the simulations were run with
this mesh size. Hence, the finalized stator grid size is 6 million and for cylinder grid
size is 2 million. In order to resolve boundary layer, properly enough nodes are kept
close to the airfoil surfaces. For the finalized stator and rotor grids, y+ values are kept
lower than 1, so that all flow features are captured in the boundary layer and no flow
approximations are used. Grids skew angles are kept between 25° and 155°. Aspect
ratio is maintained less than 100. Expansion ratio is kept less than 1.2. There is no
abrupt change in mesh density.
2D computational grid of the simulation is shown in Fig. 1.
5 Boundary Conditions
At the inlet boundary of the cylinder grid, uniform total pressure and total temper-
ature are used as inlet boundary condition. Whereas at exit of the LPNGV, average
80 V. Tandon et al.
static pressure has been used as exit boundary condition by matching the LP turbine
vane exit conditions. Walls of the stator and cylinders are considered as smooth and
adiabatic. Also for the walls, no-slip boundary condition is enforced. Free stream
turbulence intensity is assumed to be 5%. 10% of the inlet is taken as eddy length
scale. Flow is considered as fully turbulent. In blade-to blade plane, periodic boundary
condition is applied.
6 Computational Time and Convergence
The simulations are carried out on IBM DX 360 parallel computing system using 8
parallel processors. For the convergence of the steady stage and frozen rotor simu-
lations, the calculations were carried out until the max residuals drop below 10–4 .
Difference between inlet and outlet mass flow values is less than 0.02%. First simu-
lations were ran using coarse grids, and then the results of these simulation are
given as interpolated guess for the steady and frozen rotor simulations which took
approximately 20 h to converge.
Converged frozen rotor simulation results were given as initial guess for starting
of the unsteady simulation. In order to simulate properly the alternate wake shedding
phenomena from the upstream cylinders, the time step was calculated from Stouhal
formula
fL
= St (1)
U
where f is the wake shedding frequency, L is the characteristic dimension, U is
the flow velocity, and St is the Strouhal number. For a typical case, the frequency
was calculated as 16807 Hz. Hence, the time period is 5.65E−5. It is divided by
25 time steps so that we have enough time steps to capture the wake shedding
phenomena. Hence, the time step used in the simulation for this case was 2.26E−6.
For convergence in each time step which represent orientation of the cylinder grid w.r.t
stator grid different from the last time step, a maximum of 20 inner loop iterations are
used as converged results from frozen rotor simulation are supplied to the unsteady
simulation.
The simulation was made to run until a periodic repetition of lift on vane surface is
obtained which is taken as convergence criteria of the simulation as shown in Fig. 2.
Figure 3 shows the contours of the vorticity magnitude at a particular time step, for
the case where the upstream cylinder radius is 1 mm. For grid cell size of 0.03 mm, we
were able to resolve the alternate shedding of the wakes from the upstream cylinder,
Fig. 2 Convergence criteria for the simulations
as can be seen from the figure there are two wakes one from top end and other from
the bottom end of the cylinder being shed alternatively. For this time step, the wake
from the bottom end of the cylinder is impinging on the leading edge of the LP vane,
whereas wakes from the top end are getting convected along the suction surface.
Figure 4 shows that after some time steps are over a new wake is going to impinge
on the leading edge and the older wake has diffused. Figure 5 shows the variation in
LP vane pressure loss coefficient for different time steps. As can be seen the vane
loss coefficient is having a cyclic behavior, it increases then reaches a maximum
value and then starts decreasing and after reaching a minimum value start increasing
again. We can observe the repeating pattern for vane loss coefficient.
The vane pressure loss coefficient without any incoming wakes subjected to
same boundary conditions as these cases is found to be 0.144. When the wakes are
impinging on the leading edge the loss coefficient calculated is 0.151, an increase
of 4.9% and in subsequent time steps, the vane loss coefficient starts decreasing and
reaches the base value of 0.144 by this time the wake is fully convicted from the
leading edge towards the rear surface of the vane. After this time step again, the vane
loss coefficient start increasing as a new wake start approaching the leading edge of
the vane.
Figure 6 shows the variation in LP vane exit angle for different time steps. Here
also the same repeating trend is observed.
Figure 7 shows the position of the cylinder for which the wakes from upper
82 V. Tandon et al.
Fig. 3 Vorticity magnitude contours
cylinder are going towards pressure surface of vane and does not impinge on the
leading edge as in the previous case. Vane loss coefficient for different time steps after
the solution has converged is plotted in Fig. 8. The maximum vane loss coefficient
0.149 is observed an increase of 3.62% from base value of 0.144. The vane loss
coefficient then starts deceasing and reaches minimum value of 0.126 a decrease
of 12.64% from base value, which is even lower than the LP vane loss coefficient
without any incoming wakes. Similar trends are observed for vane exit angle.
Figure 9 shows the variation in LP vane exit angle for different time steps for the
new position of the cylinder.
Figure 10 shows the position of the cylinder for which the wakes from upper
cylinder are passing over the suction surface of vane unlike the previous case where
the wakes were going towards the pressure surface of the LP vane. The maximum
vane loss coefficient 0.156 is observed an increase of 7.92% from base value of 0.144.
The vane loss coefficient then starts deceasing and reaches minimum value of 0.149.
Similar trends are observed for vane exit angle.
Similar trends were observed for the case where the cylinder radius is 1.3 mm. The
wake size is increasing which on one side results in increased vane loss coefficient
but on other side the minimum loss coefficient is even much lower than the base vane
loss coefficient.
Fig. 4 New wake about to impinge on the leading edge of the vane
Vane Loss Coefficient PLot

0.152
0.151
Pressure loss coefficient
0.15
0.149
0.148
0.147
0.146
0.145
0.144
0.143
860 870 880 890 900 910 920
Time Steps
Fig. 5 Variation of pressure loss coefficient for different time steps
8 Conclusions
From the above three simulations, it is observed that incoming wakes does not always
lead to increase in vane loss coefficient. For the case for which wakes are passing
84 V. Tandon et al.
Vane Exit Angle Plot

Time steps
-67.41
860 870 880 890 900 910 920
-67.42
-67.43
Vane exit angle
-67.44
-67.45
-67.46
-67.47
-67.48
-67.49
Fig. 6 Variation of vane exit angle for different time steps
Fig. 7 Wakes passing towards pressure surface of the vane

Vane Loss Coefficient Plot

0.155
0.15
Pressure loss coefficient
0.145
0.14
0.135
0.13
0.125
0.12
1220 1230 1240 1250 1260 1270 1280 1290
Time steps
Fig. 8 Variation of pressure loss coefficient for different time steps
Vane Exit Angle Plot

-67.3
1210 1220 1230 1240 1250 1260 1270 1280
-67.35
-67.4
Vane exit angle
-67.45
-67.5
-67.55
-67.6
Time steps
Fig. 9 Variation of vane exit angle for different time steps
through the pressure surface of the vane, the vane loss coefficient is lower than the
vane loss coefficient without any effects of incoming wakes.
Also with increase in upstream cylinder size, the wake size is increasing which
on one side results in increased vane loss coefficient but on other side the minimum
loss coefficient is even much lower than the base vane loss coefficient. But this needs
a further detailed analysis like LES or DES to understand this behavior.
86 V. Tandon et al.
Fig. 10 Wakes passing over the suction surface of the vane
Acknowledgements The authors thank Director, GTRE for giving permission to present this work.
References
1. Mahallati A, Sjolander SA, Praisner TJ (2013) Aerodynamics of a low-pressure turbine airfoil

at low reynolds numbers—part I: steady flow measurements. J Turbomach 135/011010
2. Sondergaard R, Sjolander SA, Praisner TJ (2013) Toward the expansion of low-pressure-turbine
airfoil design space. In: Proceedings of ASME Turbo Expo 2008: power for land, sea and air,
GT2008 9–13 June 2008
3. Curtis EM, Hodson HP, Banieghbal MR, Denton JD, Howell RJ (1997) Development of blade
profiles for low pressure turbine applications. J Turbomach 119(3)
4. Sarkar S (2009) Influence of wake structure on unsteady flow in a low pressure turbine blade
passage. J Turbomach ASME 131
5. Yang X (2003) Aerodynamic loss modelling in transonic turbines (M.Sc. thesis), Concordia
University, Canada
6. Cohen R, Saravanamuttoo (2008) Gas turbine theory, chap 7, 6th edn
7. Paniagua G, Rivir RB (ed) (2006) Advances in turbomachinery aero-thermo-mechanical design
analysis. Von Karman Institute for fluid dynamics, lecture series 2007-02, Nov 2006, Rhode
Saint Genese
8. Ozturk B, Schobeiri MT (2007) Effect of turbulence intensity and periodic unsteady wake
flow condition on boundary layer development, separation, and reattachment along the suction
surface of a low-press turbine. J Fluids Eng ASME 129
9. Moustapha H, Zelesky MF, Baines NC, Japikse D Axial and radial turbines
10. Ciorciari R, Kirik I, Niehuis R (2014) Effects of unsteady wakes on the secondary flows in the
linear T106 turbine cascade. J Turbomach ASME 136
11. Volino RJ (2012) Effect of unsteady wakes on boundary layer separationon a very high lift low
pressure turbine airfoil. J Turbomach ASME 134
12. Acton P (1998) Untersuchungen des Grenzschichtumschlages an einemhochbelastetenTur-
binengitterunterinhomogenen und instationearenZustreombedingungen, Ph.D. thesis, Univer-
sitat der BundeswehrMunchen,Munchen, Germany.
13. Stadtmuller, P., (2002), “Grenzschichtentwicklung und Verlustverhalten vonhochbelasteten-
TurbinengitternunterEinflußperiodischinstationarerZustreomung,” Ph.D. thesis, Universiteat
der BundeswehrMunchen, Munchen, Germany
14. Hodson HP, Howell RJ (2005) Blade row interactions, transition, and high-lift aerfoils in
low-pressure turbines. Annu Rev Fluid Mech 37:71–98
15. Schobeiri MT, Ozturk B, Ashpis DE (2003) On the physics of the flow separation along a low
pressure turbine blade under unsteady flow conditions. J Fluid Eng ASME 127
16. Schobeiri MT, Ozturk B (2004) Experimental study of the effect of periodic unsteady wake
flow on boundary layer development, separation,and re-attachment along the surface of a low
pressure turbine blade. J Turbomach ASME 26(4)
17. Ibrahim MB, Vinci S, Kartuzova O, Volino RJ (2012) CFD simulations of unsteady wakes on
a highly loaded low pressure turbine airfoil (L1A). In: Proceedings of ASME turbo expo 2012:
power for land, sea and air, GT2012 11–15 June 2012
18. Nessler CA, Marks C, Sondergaard R, Wolff M, Sanders DD, O’Brien WF, Polanka MD (2010)
A CFD and experimental investigation of unsteady wake effects on a highly loaded low pressure
turbine blade at low reynolds number. In: Proceedings of ASME turbo expo 2010: power for
land, sea and air, GT2010 Oct 2010
19. Schwarze M, Niehuis R (2010) Numerical simulation of a highly loaded LPT cascade with
strong suction side separation under periodically unsteady inflow conditions. In: Proceedings
of ASME turbo expo 2010: power for land, sea and air, GT2010 Oct 2010
20. Meyer RX (1958) The effects of wakes on the transient pressure and velocity distributions in
turbomachines. ASME J Basic Eng 80
21. Mayle RE (1991) The role of laminar turbulent transition in gas turbine engines. J Turbomach
ASME
22. Walker GJ (1993) The role of laminar turbulent transition in gas turbine engines: a discussion.
J Turbomach ASME 115(2)
23. Halstead DE, Wisler DC, Okiishi TH, Walker GJ, Hodson HP, Shin H-W, (1997) Boundary
layer development in axial compressors and turbines—part 1: composite picture. J Turbomach
ASME 119(1)
24. Halstead DE, Wisler DC, Okiishi TH, Walker GJ, Hodson HP, Shin H-W (1997) Boundary
layer development in axial compressors and turbines—part 2: compressors. In: Proceedings of
ASME turbo expo 2010: power for land, sea and air, GT-462, vol 1
layer development in axial compressors and turbines—part 3: LP turbines. J Turbomach ASME
119(2)
layer development in axial compressors and turbines—part 4: computations and analyses. J
Turbomach ASME 119(1)
27. Wu X, Jacobs RG, Hunt JRC, Durbin PA (1999) Simulation of boundary layer transition induced
by periodically passing wakes. J Fluid Mech 398
28. Wu X, Durbin PA (2001) Evidence of longitudinal vortices evolved from distorted wakes in
turbine passage. J Fluid Mech 446
29. Hoheisel H, Kiock R, Lichtfuss HJ, Fottner L (1987) Influence of free-stream turbulence and
blade pressure gradient on boundary layer and loss behaviour of turbine cascades. J Turbomach
ASME 1 86-GT-234
88 V. Tandon et al.
30. Howell RJ, Ramesh ON, Hodson HP, Harvey NW, Schulte V (2004) High lift and aft loaded
profiles for low pressure turbines. J Turbomach ASME 3 GT-0261
31. Ladwig M, Fottner L (1993) Experimental investigations of the influence of incoming wakes
on the losses of a linear turbine cascade. J Turbomach ASME 3C GT-394
32. Pritchard LJ (1985) An eleven parameter axial turbine aerofoil geometry model. J Turbomach
ASME 1 GT-219
33. ANSYS FLUENT User Guide, Release 14.5
34. ICEMCFD User Guide, Release 14.5
Effect of Axial Location
on the Performance of a Control Jet
in a Supersonic Cross Flow
Raj Kiran Grandhi and Arnab Roy
Abstract Control jets injected into a supersonic flow cause a significant region
of separated flow over the parent vehicle in the vicinity of injection. This altered
external flow contributes to an additional force on the vehicle in addition to the
jet thrust thereby effecting its performance. In this study, numerical simulations
were carried out to estimate the effect of injection location and body attitude on the
overall performance of a Reaction Control System jet in producing a control force.
Commercial CFD software was used to solve the 3D RANS equations using the SST-
kω turbulence model. It is seen that favourable interaction results from injection at
rearward locations and positive angles of attack whereas injection from forward
locations and negative angles of attack results in an adverse interaction.
Keywords Reaction control systems · Supersonic flow · CFD
1 Introduction
Launch vehicles incorporate Reaction Control Systems (RCS) for attitude control.
These systems offer significant advantages over aerodynamic surfaces such as fins
through their ability to be used during the low dynamic pressure regimes of flight—
either at low velocities or at high altitudes as well as by not incurring a drag penalty
when not operative. Control forces are generated by RCS by expelling high pressure
gas through nozzles located at the peripheral regions of a flight vehicle. When the
vehicle is travelling at supersonic speeds, these jets act as a barrier to the external
flow over the vehicle and result in a complex flow field around the injection which
R. Kiran Grandhi (B)

Advanced Systems Laboratory, Defence Research
and Development Organisation, Hyderabad 500058, India
A. Roy
Department of Aerospace Engineering, Indian Institute of Technology,
Kharagpur 721302, India

90 R. Kiran Grandhi and A. Roy
consists of several shocks and vortical structures. The external flow is thus subjected
to a severe disruption in the presence of these control jets. This altered flow field
causes a corresponding change in the pressure distribution around the vehicle and
consequently in the aerodynamic forces acting on the vehicle.
The interaction flow field corresponding to the injection of a side jet into a super-
sonic cross flow is a subject of much interest to the aerospace community. These
interactions arise not only in RCS systems where they influence the effectiveness of
the jet in producing control forces by disrupting the flow field, but also in SCRAM-
JET engines where they affect the penetration, mixing and combustion [1] of the
injected fuel. These interactions also appear in Secondary Injection Thrust Vector
Control (SITVC) systems where again they affect the control forces produced by the
nozzle [2]. The complex flow structure resulting from this interaction [3] is shown
in Fig. 1 where the expanding jet is enclosed in a barrel shock and acts as a barrier
to the incoming external flow. This causes a bow shock to form upstream of the
injection, and this shock wraps itself around the injected jet. The boundary layer
of the cross flow separates upstream of the injection and forms a horse shoe vortex
around the jet. This is a characteristic feature of the interaction structure. The shock
structure resulting from this interaction creates a region of high pressure upstream
of the injection and a region of low pressure downstream of the injection [4].
LARGE-SCALE
STRUCTURES
M>1
BOW SHOCK
BARREL
SHOCK
MACH DISK
BOUNDARY LAYER
SEPARATED
REGION RECIRCULATION
SIDE JET ZONE
RECIRCULATION
ZONE
Fig. 1 Schematic of the jet interaction flow field [3]
Effect of Axial Location on the Performance of a Control … 91
In an earlier study, [5] have studied the effect of a number of parameters, such as
the jet pressure ratio (Po,jet /P∞ ), injection Mach number, parent body attitude with
respect to the external flow and the presence of a flare on the effectiveness of the
injected control jet. The present study focusses on the effect of injection location on
the aerodynamic force due to jet interaction.
2 Methodology
The simulations were carried out using a commercial CFD software for a typical
cone-cylinder configuration. The cylindrical section has a diameter D = 40 mm and
a length of 6.2D. The length of the conical section was 2.8D. The lateral jet was
injected from the cylindrical section, and four different models were made which
correspond to different locations of this injected jet. The three dimensional RANS
equations were solved on a hexahedral grid consisting of about 7 lakh cells using
the shear-stress transport (SST) k − ω turbulence model. A second order upwind
scheme was employed in all the simulations. The pressure distribution obtained
from the simulations was integrated to obtain the aerodynamic force acting on the
vehicle.
A summary of the simulation cases is shown in Table 1. The cross flow parameters
(P∞ = 16,400 Pa, T∞ = 100 K, M∞ = 3.0), the injection Mach number (Minj = 1)
and the jet stagnation temperature (280 K) were unchanged in all the simulations.
The effectiveness of the control jet can be defined as the ratio of the total side
force generated when the jet is operated to the jet thrust. In the presence of an angle
of attack, the total side force includes a component due to the attitude which is
accounted for in calculating the effectiveness by running a reference simulation with
the jet “off” and subtracting the corresponding forces from the simulation with the
jet “on”. The jet reaction can be computed using the rocket thrust equation:
Fjet = ṁ jet Ve + (Pe − P∞ ) ∗ Ae (1)
The additional force acting on the body due to the interaction can be computed as:
Fi = Faero − Fref (2)
Table 1 Summary of simulation cases

Parameter Variation
Injection Mach number 1.0, no injection
Injection location (from nosetip) 3.3D, 4.3D, 6.0D, 8.5D
Angle of attack −15◦ , 0◦ , +15◦
And the effectiveness of the control jet is thus computed as:
Fi
E jet = 1 + (3)
Fjet
If the additional force due to jet interaction is negligible, then Faero ≈ Fref and the
jet effectiveness is close to unity. Depending on whether the jet interaction force is
in the direction of jet reaction or opposite to it, the jet effectiveness is respectively
greater than or less than unity.
3 Validation and Grid Independence Study
The simulation methodology was validated using experimental pressure measure-

ments on a similar configuration reported by [6]. The geometry considered for val-
idation purpose and the associated computational grid are shown in Fig. 2. This
geometry is nearly identical to the model described in the preceding section. The
only difference is that in this model, there is a flare extending from 6.0D to 9.0D
with an end diameter of 1.66D. The side jet is injected from an orifice located on the
cylindrical surface at a distance of 4.3D from the nose tip. The flow structure and
surface pressure distribution as obtained from the simulation are shown in Fig. 3.
In order to establish grid independence of the obtained results simulations were
run on coarse, medium and fine meshes consisting of about 3.3 lakh, 7.7 lakh and
15.7 lakh cells respectively. A comparision of the pressure variation along the upper
centre line of the model for these different grids along with the experimental values
Injection orifice 0.1D

at X = 4.3 D
3.0D
0mm
D=4
2.8D
3.2D
Fig. 2 The hexahedral grid used for the validation studies

Pathlines coloured by velocity (m/s)
Deflected
RCS jet
Horse-shoe
vortex
Region of
separation Region of
Pressure distribution on surface (kPa) low pressure
in the wake
Fig. 3 Flow structure and surface pressure distribution
Fig. 4 Variation of pressure along the centre-line
is shown in Fig. 4. The components of the normal force coefficient due to the jet, the
interaction flow field and the overall effectiveness along with the Grid Convergence
Index (GCI) as proposed by [7] using a factor of safety of 1.25 for the present three-
grid study are shown in Table 2. The GCI was found to be less than 1%, and thus the
solution may be considered to be grid-independent.
Table 2 Variation of the force components with grid size

Size (Million) CN,total CN,jet CN,i E jet
0.33 −0.246 −0.396 0.150 0.621
0.77 −0.257 −0.396 0.138 0.650
1.57 −0.259 −0.395 0.136 0.657
GCI (%) 0.329 0.027 0.824 0.416
4 Results
The overall effectiveness of the control jet depends on several features of the complex
flow interaction structure around the injected jet as well as the body attitude with
respect to the incoming flow. Some of these features are:
• The conical shock emanating from the tip of the vehicle.
• The bow shock ahead of the injected jet.
• The barrel shock around the expanding jet.
• The horse shoe vortex forming around the jet and trailing downstream alongside
the vehicle.
• Leeward vortices generated by the separation and roll up of boundary layer due to
an angle of attack.
Depending upon the angle of attack and location of the injected jet, interaction
among the above is also possible. For example, at negative angles of attack or for
injection from the fore-regions, the conical shock can interact with the bow shock.
Similarly, the lee vortices forming due to an angle of attack can interact with the horse
shoe vortex. The surface pressure distribution over the vehicle is thus dependant on
all the above features and their interactions.
4.1 Shock Structures
The shock structures of the interaction are shown in Figs. 5, 6 and 7 wherein contours
of Mach numbers on the symmetry plane and pressure coefficient on the body surface
are depicted for injection from different locations at various angles of attack. For
a negative angle of attack, the conical shock emanating from the tip of the vehicle
impinges on the bow shock ahead of the injection as seen in Fig. 5. This impingement
exists for all the injection locations except for the aft-most. For zero angle of attack,
as seen in Fig. 6, this impingement is seen only for the fore-most injection and to a
lesser extent, injection at 4.3D. For positive angles of attack, shown in Fig. 7, there
is no impingement whatsoever as the shock is present only on the side opposite that
of the injected jet.
Cp Mach number
(a) Xinj = 3.3D (b) Xinj = 4.3D
(c) Xinj = 6.0D (d) Xinj = 8.5D
Fig. 5 Mach number and Cp contours for injection at different locations (α = −15◦ )
Cp Mach number
(a) Xinj = 3.3D (b) Xinj = 4.3D
(c) Xinj = 6.0D (d) Xinj = 8.5D
Fig. 6 Mach number and Cp contours for injection at different locations (α = 0◦ )

Cp Mach number
(a) Xinj = 3.3D (b) Xinj = 4.3D
(c) Xinj = 6.0D (d) Xinj = 8.5D
Fig. 7 Mach number and Cp contours for injection at different locations (α = 15◦ )
Cp Mach number
(a) Xinj = 3.3D (b) Xinj = 4.3D
(c) Xinj = 6.0D (d) Xinj = 8.5D
Fig. 8 Flow pathlines for injection at different locations (α = −15◦ )

Cp Mach number
(a) Xinj = 3.3D (b) Xinj = 4.3D
(c) Xinj = 6.0D (d) Xinj = 8.5D
Fig. 9 Flow pathlines for injection at different locations (α = 15◦ )
4.2 Vortex Structures
Flow pathlines around the body and from the injected jet are shown in Figs. 8 and 9.
The bow shock ahead of the injection causes the boundary layer to separate, creating
a vortex in the region of separation. This vortex curves around the barrel shock
resulting in the horse-shoe vortex that is a characteristic feature of the jet interaction.
The incoming flow also deflects the injected jet aligning it parallel to the flow. At
negative angles of attack, this deflected jet lies close towards the body and at positive
angles of attack it is deflected further away from the body. In addition, when the body
is at an of attack, the boundary layer on the lee side of the body separates and rolls up
into a pair of vortices that trail downstream along the body. These vortices interact
with the horse shoe vortex, further altering the flow structure around the vehicle.
4.3 Surface Pressure Distribution
The effectiveness of the injected jet depends on the surface pressure distribution
which is strongly influenced by the shock and vortical structures described in the
preceding sections. Figures 10, 11 and 12 highlight the key aspects of the effect of
these structures on the surface pressures. In these figures, numerical oilflow patterns
are shown over plots of the differential pressure coefficient obtained by subtracting
the pressure distribution of the reference simulation (without jet) from the surface
pressure distribution.
For the cases without an angle of attack, shown in Fig. 11, the basic features of
the interaction such as separation of the boundary layer upstream of the injection
ΔCp
(a) Xinj = 3.3D (b) Xinj = 4.3D
(c) Xinj = 6.0D (d) Xinj = 8.5D
Fig. 10 Surface flow for injection at different locations (α = −15◦ )
ΔCp
(a) Xinj = 3.3D (b) Xinj = 4.3D
(c) Xinj = 6.0D (d) Xinj = 8.5D
Fig. 11 Surface flow for injection at different locations (α = 0◦ )

ΔCp
(a) Xinj = 3.3D (b) Xinj = 4.3D
(c) Xinj = 6.0D (d) Xinj = 8.5D
Fig. 12 Surface flow for injection at different locations (α = 15◦ )
followed by the high pressure region, the region of suction in the wake of the injected
jet and reattachment further downstream of the injection location can all be clearly
seen. Also, the basic structure of the oilflow pattern and intensity of the pressure
distribution is nearly independent of the injection location. For the fore-most injection
case (X inj = 3.3D), the separation region upstream of the injection extends all the way
to the bottom of the body. This is because of the lowered pressure of the incoming flow
just after the expansion corner at the interface between the conical and cylindrical
sections. For injection from the more downstream locations, the incoming flow on
the bottom side is effected to a lesser and lesser extent.
For the cases with an angle of attack, shown in Figs. 10 and 12, the strong inter-
action of the separation vortices with the injection flow structure can clearly be seen.
For these cases, the separation of the boundary layer due to the angle of attack
occurs about about 30◦ towards the lee side. These separation lines are clearly seen
for injection from all but the fore-most location in these figures. For a negative angle
of attack, the suction in the wake of the jet that has been deflected close to the body
is significantly stronger than in the case of a positive angle of attack for which the
effect of suction is rather diminished.
4.4 Force and Moment Due to Interaction
The differential surface pressure distributions over the body were integrated to obtain
the variation of force and moment along the vehicle. These force and moment dis-
tributions are shown in Figs. 13, 14 and 15. Since these forces and moments were
based on the pressure distribution obtained after subtracting the reference pressure,
the absence of any interaction with the external flow would dictate that these are both
zero. Thus, these figures depict only the effect of interaction excluding the effect
of body attitude. The raw values of force and moment were non-dimensionalised
using Fjet and Fjet × D as the scaling factors for force and moment respectively. The
moment was computed about the injection location with pitch-up moment consid-
ered as positive. Force acting in the direction of jet thrust was considered positive.
Thus, for force, positive region of the graphs corresponds to favourable interaction
and negative region corresponds to adverse interaction.
In the absence of an angle of attack, as seen in Fig. 14, the only significant
change in the force distribution is a geometric shift that corresponds to the change
in injection location. As discussed in the preceding section, the interaction pattern is
relatively independent of the injection location for zero angle of attack and the force
distribution reflects the same. In the presence of an angle of attack, however, the
changes in the extent and intensity of the regions of suction in the jet wake coupled
with the interaction of the separation vortices with the injected jet results in more
significant changes to the corresponding force variations (Fig. 15).
(a) Force per unit length (b) Cumulative force
(c) Moment per unit length (d) Cumulative moment
Fig. 13 Non-dimensional force and moment due to interaction (α = −15◦ )

Fig. 14 Non-dimensional force and moment due to interaction (α = 0◦ )
Fig. 15 Non-dimensional force and moment due to interaction (α = 15◦ )

4.5 Jet Effectiveness
The variation of jet effectiveness in producing the desired normal force with injection
location for different angles of attack is shown in Fig. 16. For all angles of attack, it
is seen that injection from the aft-most location is most favourable as it results in an
amplification of the jet thrust. For injection from other locations, the effectiveness
is relatively independent of the injection location in the absence of an angle of
attack. Further, while a positive angle of attack results in a favourable interaction
force, a negative angle of attack results in adverse interaction and injection from
progressively upstream locations result in progressively worse effectiveness. The
non-dimensionalized moment about the injection location can be thought of as a shift
of the injection location caused by the interaction flow field, an effective moment
arm, that is given by:
Mi
X i =
Fjet
Like the jet effectiveness, X i is zero if there is no moment developed about the
injection location due to the interaction. It is negative for a pitch-down moment and
positive for a pitch-up moment. The variation of this effective moment arm is shown
in Fig. 17. It is seen that for all the cases, the overall effect of the interaction is to
cause a pitch-down moment about the injection location. Like the jet effectiveness
for force, this effective moment arm is greatest for a negative angle and for injection
from the fore-most location. For other cases, it is more modest and lies within 0.5D.
Fig. 16 Variation of jet effectiveness with location (Normal force)

Fig. 17 Variation of jet effectiveness with location (Moment)
5 Conclusion
Numerical simulations were carried out to study the effect of location and attitude on
the control characteristics of a side jet injected from an axi-symmetric parent body
consisting of a sharp cone followed by a cylinder. The methodology was validated
with published experimental data and a grid independence study was performed
to establish the accuracy. The effect of injection location was studied by creating
four models each corresponding to injection of the side jet from a distinct location
on the cylindrical region. The effect of body attitude was also studied by varying
the angle of attack. The results indicate that both injection location as well as the
angle of attack strongly influence the effectiveness of the injected jet in producing
the desired control force. Injection from aft locations as well as positive angles of
attack contribute favourably to the force amplification while injection from forward
locations and negative angles of attack contribute negatively.
References
1. Kawai S, Lele SK (2010) Large-eddy simulation of jet mixing in supersonic crossflows. AIAA
J 48(9). https://doi.org/10.2514/1.J050282
2. George R, Krishna JM, Grandhi RK, Gupta RK (2011) Flow-field analysis of SITVC in a
contoured nozzle. In: Fifth symposium on applied aerodynamics and design of aerospace
vehicles
3. Ben-Yakar A (2000) Experimental investigation of mixing and ignition of transverse jets in
supersonic crossflows. PhD thesis, Stanford University
4. Viti V, Neel R, Schetz JA (2009) Detailed flow physics of the supersonic jet interaction flow
field. Phys Fluids 21(4):046101. https://doi.org/10.1063/1.3112736
5. Grandhi RK, Roy A (2017) Effectiveness of a reaction control system jet in a supersonic
crossflow. J Spacecraft Rockets 54(4):883–891. URL https://doi.org/10.2514/1.a33770
6. Stahl B, Esch H, Glhan A (2008) Experimental investigation of side jet interaction with a
supersonic cross flow. Aerosp Sci Technol12(4):269–275. https://doi.org/10.1016/j.ast.2007.
01.009
7. Roache PJ (2003) Error bars for CFD. In: 41st aerospace sciences meeting and exhibit.
https://doi.org/10.2514.6,2003-408
Effect of Chord Variation on Subsonic
Aerodynamics of Grid Fins
Manish Tripathi , Mahesh M. Sucheendran, and Ajay Misra
Abstract Current paper deliberated the impact of gap-to-chord ratio (g/c) variation
on grid fin subsonic flow characteristics through chord modifications while retaining
the same gap and aspect ratio to explicitly decipher the role of chord on grid fin
aerodynamics through numerical analysis. Solver validation is followed by compre-
hensive examination of the pertinent aerodynamic coefficient results associated with
different grid fin chord variants. The study establishes enhanced maximum lift coef-
ficient at the expense of reduced aerodynamic efficiency in the most operable angle
of attack region for higher g/c (lower chord). Efficiency reduction is attributed to
increase in drag for higher g/c emanating due to increased pressure drag applicable
for a blunt compact geometry. Stall angle undergoes minimal deviation for different
g/c when chord is the varying parameter. However, deviations associated with grid
fin aerodynamic efficiency for varying chord were found to be appreciably signifi-
cant. The study categorically deduces the impact of chord in the g/c parameter, and
hence can be helpful for grid fin designers while selecting the optimum chord value
for enhanced aerodynamic efficiency and lift requirements. This study in conjunc-
tion with analysis carried out for gap variation can help achieve an efficient grid fin
design with respect to delayed stall angle and increased aerodynamic efficiency.
Keywords Grid fins · Cascade fins · Gap-to-chord ratio · Chord variation ·

Aerodynamic efficiency · CFD++
M. Tripathi (B) · A. Misra

Department of Aerospace Engineering, Defence Institute of Advanced Technology (DU),
Girinagar, Pune, Maharashtra, India
A. Misra
M. M. Sucheendran
Department of Mechanical and Aerospace Engineering, Indian Institute of Technology,
Hyderabad, Telangana, India

106 M. Tripathi et al.
Nomenculture
AR aspect ratio of horizontal members

b, c, th span, chord and thickness of members, m
bFP , cFP span and chord of planar fin, m
CD coefficient of drag
CFD Computational fluid dynamics
CFL Courant-Friedrichs-Lewy number
CL Coefficient of lift
C Lα Lift slope, deg −1
CL /CD Aerodynamic efficiency
CLmax Maximum lift coefficient
CP Pressure coefficient
e0 Total energy
g/c Gap-to-chord ratio
M Mach number
NS Navier-Stokes equations
p Pressure, Pa
Re Reynolds number
SA Spalart-Allmaras turbulence model
t Time, s
uj jth component of velocity vector
U∞ , P∞ , T∞ Freestream pressure (Pa), temperature (K ) and velocity (m/s)
xj jth component of position vector
x/c Non-dimensional chordwise distance
Greek letters
α Angle of attack, ◦
ρ Density of flow, kg/m3
1 Introduction
Missiles and unmanned aerial vehicles are unmanned warhead delivery vehicle sys-
tems comprising of numerous sub-systems like guidance and control, propulsion,
and warhead (payload) systems. Optimizing their guidance and control system is
imperative to achieving the desired mission objective. Control system consists of
external aerodynamic surfaces called control surfaces, which generate the required
aerodynamic forces by virtue of angular deflections to execute a desired manoeuvre.
Deliberating the aerodynamic forces generated by these surfaces, and the inherent
Effect of Chord Variation on Subsonic Aerodynamics … 107
Fig. 1 Missile with grid fins
aerodynamic efficiency plays a critical role in evaluating the most optimum geom-
etry. Grid fins also termed as lattice fins are a recent advancement in the control
surface aero-mechanical technology.
Grid fins have been globally recognized as unconventional control surfaces
ascribed to its unique structural characteristics consisting of numerous intersect-
ing planar surfaces encompassed by a thin outer frame to provide rigidity (Fig. 1).
This interlacing of multiple lifting surfaces leads to the formation of multiple cells
which act as individual lifting surfaces. Unlike conventional planar fins, these fins are
aligned transverse to the freestream flow with the air being allowed to pass through
these cells. These fins have been implemented on a multitude of aerospace applica-
tions. For instance, these were implemented as control surfaces on numerous Soviet
(now called Russia) supersonic missiles (Adder AA-12, SS-12 ‘Scaleboard’ etc.)
in the late1950s [1], as stability and air drag devices on the Soyuz TM-22 Launch
Escape Spacecraft, as maneuvering device on the USA based Massive Ordinance
Air Blast (MOAB) bomb [2], and as hypersonic control and stability surface on
reusable launch vehicle named Falcon 9, developed by SpaceX [3]. Most recently,
these were implemented on the North Korean Intercontinental ‘Musudan’ cruise
missile [4]. Owing to their unique structural and aerodynamic characteristics, these
are also being explored for under-water launched missiles and munitions as well as
interstellar space exploration spacecrafts [5, 6].
Recent upsurge in its world wide application has been attributed to its unique struc-
tural and aerodynamic characteristics leading to significant utility based advantages
compared to conventional fins. The most favorable advantage pertains to augmented
lift capability at high angles of attack and across wider Mach number (M) regimes
(subsonic and high supersonic). In-fact these have been credited with delayed stalling
even up to angle of attack (α) = 50◦ in some cases. Due to the shock waves getting
swallowed within the cells, grid fins display superior control characteristics at high
supersonic speeds [7]. Smaller chord enables significant hinge moment curtailment
leading to salient control surface actuator weight requirements [8, 9]. Due to the
reduced impact of curvature on grid fin aerodynamics, these possess superior fold-
ing capabilities [10]. Some additional advantages including improved roll control due
to attenuated body vortex interference [11], high strength-to-weight ratio, as well as

increased target area and reduced miss distance due to their elevated manoeuvrabil-
ity [12]: further increase their utility potential. In-spite of these advantages, limited
usage during their advent was linked to higher drag especially in the transonic regime
(originating due to the flow choking phenomena [7]), increased radar cross-section
area [8, 13], and complex manufacturing process.
A plethora of research has been carried out historically to contemplate the under-
lying rudimentary flow-physics linked to its unique aerodynamic characteristics in
comparison to planar fins. These studies involved experimental observations [8, 10,
11, 13–17], theoretical analysis through the application of Vortex Lattice Theory [18–
21] and Prandtl-Meyer’s shock expansion theory [22]. A theoretical study involving
missile body and grid fin integration was carried out by Ledlow et al. using pre-
existing theoretical methods and a missile aerodynamic prediction tool. The study
also carried out grid fin geometry optimization using a modified ant colony optimiza-
tion technique [12]. Additional investigations by carrying out dynamic characteriza-
tion using the small perturbation theory [23] was accomplished to analyze the control
characteristics of grid fin control surfaces. With the advancement of computational
power, several numerical examinations were carried out to decipher the grid fin’s
aerodynamic [9, 24–34], thermal [35], and aeroelastic properties [36].
Significant research oriented around comprehending the flow-physics linked to
these fins had been carried out to mitigate drag as well as improve its transonic per-
formance. These studies incorporated a number of geometrical modifications like
altered frame cross-section leading to 25% drag reduction [8], and local sweeping
[37] for drag reduction of around 38% in the supersonic regime. Drag mitigation in
transonic domain was achieved through implementation of: (1) swept-back grid fin
[38, 39], (2) combination of sharp leading edges and swept grid fins [40, 41], (3)
single cell leading edge variations with a delta configuration [42], amounting to a
drag reduction of ≈12–13, 25–30, and 40%, respectively. An investigation pertain-
ing to incorporating an optimized Busemann profile cross-section for the grid fin
cells suggested significant drag reduction in the supersonic regime due to the wave
cancellation effect [33]. Thus, drag reduction had been the main approach towards
grid fin performance enhancement. As per the authors’ knowledge, hitherto study
related to aerodynamic efficiency enhancement has been scarce.
In 2009, Misra et al. suggested a simplified grid fin variant called as cascade fins
(hereinafter referred as grid fins only). These consisted of a cascade of horizontal lift-
ing members (termed as planar members) placed parallel to each other and supported
by end plate members to provide rigidity (Fig. 2). Removal of cross-members lead to
reduced drag, and hence enhanced aerodynamic efficiency. These simplified grid fins
had been extensively utilized in subsonic flow to comprehend the impact of different
geometrical parameters like number of lifting members, gap between the members,
and cross-section of the members through experimental [43, 44], and theoretical [45,
46] analysis. These studies established enhanced high α performance for grid fins,
delayed stall angle along with lowered lifting characteristics for lower gap grid fins
as well as the fact that grid fins possess lower aerodynamic efficiency compared to
planar fins. Current research aims at carrying out numerical analysis to decipher flow-
Fig. 2 Cascade fin (simplified grid fin variant)
physics involving the impact of different grid fin geometrical parameters, and hence
suggest the most optimum grid fin geometry for enhanced aerodynamic efficiency
and delayed stall characteristics. Subsequently, validated numerical evaluations per-
taining to gap-to-chord ratio (g/c) variation through gap alterations, number of planar
member modifications, and cross-section variation were carried out by the authors
in Refs. [47–49] to underscore the rudimentary flow-physics associated with these
geometrical modifications.
Current paper ventures up on explicitly evaluating the effect of chord variation on
grid fin subsonic flow characteristics at different angles of attack through a sequence
of numerical simulations performed at different g/c values where, g is gap between
the members which remains constant and c (variable analyzed) is chord of the planar
members pertaining to different grid fin variants. Moreover, span is also varied along
with c such that aspect ratio ( A R) remained constant. The following section (Sect. 2)
describes the numerical methodology adopted to carry out the pertinent steady state
numerical simulations, followed by the results and discussions section. Conclusions
drawn out from this section deliberating the effect of chord variation on grid fin
aerodynamics have been discussed in Sect. 4.
2 Numerical Methodology
Aerodynamic analysis in the present study was accomplished using computational

fluid dynamics (CFD). Details related to the pertinent geometry being simulated, sim-
ulated flow conditions, mesh configuration, boundary conditions, and solver speci-
fications are described in the succeeding section.
2.1 Geometry
The simulations make use of grid fins consisting of three rectangular cross-section
planar members placed parallel to each other (Fig. 3a) with the following geometrical
specifications:
• Gap between the members (g): 0.05 m
• Thickness of the members (th): 2.5 × 10−3 m
• gap-to-chord ratio (g/c): 0.3 to 1.3 at increments of 0.1, and 2.0.
• Aspect ratio (AR): 2.0
Thus, planar member chord was varied to obtain the particular g/c and span (b)
was varied such that A R was maintained at 2.0 with the respective values given in
Fig. 4a. Consequently, surface area varied according to these specifications as shown
in Fig. 4b. Notably, the variant pertaining to g/c = 0.5, was used for solver validation
by comparing experimental results obtained for the same geometry as given in Ref.
[44]. The same geometry shall be taken as baseline geometry to compare the impact
of increasing or decreasing g/c on grid fin aerodynamics.
Further confirmation with respect to validity of the current numerical results was
obtained by carrying out simulations for a single planar member with the geomet-
rical specifications as given in Fig. 5. Values pertaining to the different geometrical
(a) Isometric view of grid fin (b) Top and side cross-section cut views
Fig. 3 Model view of the tested grid fin variant (*variable)
0.35 0.06
Chord variation (c)
Chord and span variation (in m)
0.3 Span variation (b) 0.05

Projected area (in m )
2
0.25
0.04
0.2
0.03
0.15
0.02
0.1
0.05 0.01
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
gap to chord ratio, g/c gap to chord ratio, g/c
(a) Span and chord for different g/c (b) Projected area variation
and constant AR
Fig. 4 Geometrical variations at different g/c

Fig. 5 Flat plate geometry for validation
Fig. 6 Computational mesh configuration
parameters were bFP = 0.02 m, cFP = 0.1 m, and same th as that of the grid fin
members mentioned earlier. The numerical results pertaining to this geometry were
compared against the experimental results obtained for the same geometry by Misra
in Ref. [43].
The study implemented cylindrical computational domain across all the simula-
tions pertaining to different grid fin variants. Its radius was adjusted such that the
frontal area of the respective geometry was less than 5% of the computational domain
frontal area (Fig. 6a). Moreover, the domain length was taken as 5 × c and 20 × c in
the upstream and downstream directions respectively (Fig. 6b). Angle of attack (α)
was simulated by rotating the domain and keeping the geometry fixed on the axis.
Additionally, it was rotated to be aligned within 5◦ of the flow direction to efficiently
capture wake region variations at different angles of attack.
2.2 Mesh and Boundary Conditions
Meshing in the current setup was carried out using MIME software (also known as
IMIME Mesher) developed by Metacomp Technologies [50]. Mesh implemented a
hybrid mesh configuration comprising of structured prism layers close to the walls
and tetrahedral cells (characterized as unstructured mesh) away from the walls. Prism
layers enable efficient flow capturing in the wall-adjacent boundary layer region
(Fig. 6c). Moreover, to optimally reduce the mesh cell size and increase efficiency,
mesh implemented a multi-block meshing environment wherein, mesh was finer
close to the walls and coarser towards the far-field boundaries (Fig. 6a, b). Meshing
was carried out for one half of the domain taking advantage of symmetry about the
midspan location as seen earlier in Fig. 3a. The total number of cells depending on
the geometry varied between 2.0 million cells for g/c = 0.3, and 0.75 million cells
for g/c = 2.0. Mesh pertaining to grid fins with g/c ranging between 0.3 and 2.0
were segregated as 56–90% tetrahedral, and 40–7% triangular prisms away from the
walls, 3.1–2.0% hexahedral, and remaining pyramids within the prismatic layers.
Height of the first layer pertaining to the prismatic layer is quantified by consid-
ering the y + value. It is a non-dimensional number signifying the first layer height
away from the walls with respect to the turbulent boundary layer. It is analogous to
Reynolds number (Re) with a lower value corresponding to first layer residing in the
viscous sub-layer and a higher value corresponding to placement in the fully turbu-
lent flow region. Hence, it delineates which part of the turbulent boundary layer the
first layer resides in. Selection of y + value also depends on the turbulent model being
used for simulations. Based on the analysis carried out in the present research, the
best results were presented by the Spalart-Allmaras model (SA), which produces the
most accurate results when the mesh is refined up to the viscous sub-layer. Thus, pris-
matic layer was refined up to y + ≈ 1 leading to a first layer height of 3.616 × 10−5
m and a growth factor of 0.8. In order to cater for the accuracy hampering high aspect
ratio cells originating at sharp corners of the grid fin geometry, mesh incorporated a
sharp turn blending feature (Fig. 6c) leading to the creation of degenerate prismatic
elements projected from geometry corners and edges. This enabled efficient meshing
within sections corresponding to high geometric curvature (Fig. 7).
The study utilized characteristic based boundary conditions for inlet, outlet and
farfield boundaries. This boundary condition is characterized by assigning flow
related M and direction values to a cell by using data supplied to an approximate Rie-
mann solver. Walls were assigned the adiabatic no-slip boundary condition, which
necessitated the presence of zero velocity along the wall surfaces. Symmetry bound-
ary condition was assigned for the symmetry plane described earlier.
Fig. 7 Degenerate mesh

cells at grid fin corners
pertaining to prism layers
2.3 Mathematical Model
Numeral analyses were carried out by solving a set of 3D Navier-Stokes equations

(NS) which are a set of time (t) dependent non-linear, coupled differential equations
consisting of one mass conservation equation (Eq. 1), three momentum conservation
equations (Eq. 2), and an energy conservation equation (Eq. 3). These equations are
solved together to obtain the values of different flow parameters namely, velocity
components (u j , jth component of velocity vector) and pressure (p) in the different
directions [49].
∂ρ ∂
+ ρu j = 0 (1)
∂t ∂x j
∂ ∂
(ρu i ) + ρu i u j + pδi j − τ ji = 0, i = 1, 2, 3 (2)
∂t ∂x j
∂ ∂
(ρe0 ) + ρu j e0 + u j p + q j − u i τi j = 0 (3)
∂t ∂x j
Here, x j , ρ, τ ji , e0 , and q j are jth component of position vector, density, viscous

shear stress, total energy, and heat flux respectively. δi j = 1 for i = j, and zero
otherwise
These coupled equations in the current numerical setup were solved using the
Reynolds-Averaged Navier-Stokes equations (RANS), which are derived from the
original NS equations by replacing the flow variables (signified by Φ in Eq. 4) with
the sum of their respective mean (Φ) and fluctuating components (Φ ) as seen in
Eq. 4. These equations are time averaged to obtain the final set of RANS equations
(Eq. 5).
Φ = Φ + Φ (4)

1
Φ≡ Φ(t)dt (5)
T T
As a consequence of this averaging process, new additional terms are introduced

called the Reynolds stresses (τi j ) which are solved using additional equations called
the turbulence model equations. Turbulence closure problem (mentioned earlier in
Sect. 2.2) in the current study was accomplished by using the one-equation Spalart-
Allmaras (SA) turbulence model which relates the eddy viscosity parameter and
length scale specifications through a one transport differential equation. This model
originally developed for external aerodynamic flows [51] has been deliberated to be
robust and computationally less expensive compared to other turbulence models. It
has been delineated to generate accurate results for external aerodynamic flow prob-
lems including low Re wall-bounded flows, flows involving adverse pressure gradient
and rotation [38, 51–53]. In the current research, a systematic comparison with results
obtained using other turbulence models also reaffirmed its higher accuracy. A brief
mathematical description of this turbulence model is elaborated in Ref. [49].
2.3.1 CFD++ Solver Specifications
Numerical simulations were performed using the commercial CFD++ solver devel-
oped by Metacomp Technologies [54] which is a three-dimensional finite volume
method based solver adopting a hybrid mesh configuration. In the current numerical
setup, it utilized a second order space discretization method along-with an implicit
time integration method for steady state evaluations. It incorporated a preconditioned
based method which is favorable for low speed flows. This method applies a precon-
ditioning operator on the time derivatives of the NS equations and mitigates the eigen
value spread alleviating the numerical diffusion (characteristic of low speed simula-
tions), and hence accelerate the rate of convergence towards steady state solutions.
A multidimensional Total Variation Diminishing (TVD) interpolation scheme was
adopted by the solver to avoid spurious numerical oscillations linked with second
or higher order numerical schemes. A local wave-model solution based approxi-
mate Riemann solver was used to evaluate inter-cell fluxes which leads to correct
signal propagation for the inviscid flow terms. Moreover, a multigrid algorithm had
been incorporated to enhance convergence acceleration and provide solution grid
independence [55]. Simulations were computed in parallel on Linux based machines
consisting of 12–24 processors (based on the mesh size) and a double precision solver
at the Parallel Computation (PARC) facility located in the Department of Aerospace
Engineering, at the Defence Institute of Advanced Technology, Pune. In order to
achieve faster convergence and adapt according to the forthcoming transients related
to a flow like sudden changes in the flow direction, the Courant-Friedrichs-Lewy
(CFL) number was ramped from 1 to 40. Normalized flow residuals were iteratively
tracked, and the solution was deemed converged when these stabilized and acquired
values less than 5 orders of magnitude.
2.4 Flow Conditions
Simulation conditions were based on the conditions as mentioned in Ref. [44] that
is, low subsonic speeds with M = 0.1176, and Re = 2.855×105 (based on chord
length of the baseline model) which amounts to a freestream velocity of 40 m/s.
Freestream was assigned standard sea level conditions that is, ambient pressure (P∞ )
= 101,325 Pa, and Temperature (T∞ ) = 288 K. Flow variation pertaining to varying
α was obtained by rotating the domain about the x-axis (Fig. 3a) and keeping the
geometry fixed. For g/c = 0.3 to 1.0, and 2.0, α was varied between 0◦ to 50◦ at
an increment of 5◦ with an additional simulation at 53◦ . Moreover, for comparative
purposes, simulations pertaining to g/c = 1.1, 1.2, and 1.3 were performed for 0◦ to
15◦ only, at intervals of 5◦ . Notably due to the varying c values, Re for the various
geometries varied between 7.1 × 104 for g/c = 2.0 to 4.76 × 105 for g/c = 0.3,
which are typically below the critical Re value for flow over a flat plate. Hence, the
results can be considered to be corresponding to qualitatively similar flow conditions.
Non-dimensionalization of the forces related to a particular geometry was carried
out by using chord length, and single planar member projected area of the respective
variant as the reference length and reference area, respectively.
3 Results and Discussions
3.1 CFD++ Validation
Validity of the numerical results for the particular flow conditions was ascertained
through comparison between the aerodynamic coefficient data obtained experimen-
tally [43] and the numerically obtained data. This comparison was carried out
between the data corresponding to:
1. grid fin of g/c = 0.5 (baseline model)
2. planar fin of A R = 2.0 (Fig. 5).
Notations used for data corresponding to the respective variants are described in
Table 1.
Deviation between the experimental and numerical results were evaluated by taking
angle-wise difference between the lift and drag coefficient values, and dividing them
by the maximum value of the experimentally obtained result for the specific force
coefficient. Further clarification related to this method can be obtained from [47].
As seen in Fig. 8a, coefficient of lift (CL ) results were within 11% for the
GF_GC0.5 case, whereas it was found to be within 15% for the PF case. The larger
value for PF can be associated with the lower maximum value of CL . Stall angle is
also found to be efficiently captured by the numerical solutions. In the case of drag
coefficient (CD ), error was within 9 and 7% for GF_GC0.5 and PF cases, respectively.
Reasons pertinent for the palpable deviation between the numerical and experimental
results can be attributed to:
Table 1 Validation plot notations

Symbol Notated value
GF_GC0.5Exp Experimental results for grid fin with g/c = 0.5
GF_GC0.5Sim Simulation results for grid fin with g/c = 0.5
PFExp Experimental results for planar fin
PFSim Simulation results for planar fin
2.5 2
GF_GC0.5 GF_GC0.5
Exp Exp
GF_GC0.5Sim GF_GC0.5Sim
2
Coefficient of drag, CD
Coefficient of lift, CL
PF 1.5 PF
Exp Exp
1.5 PFSim PFSim
1
1
0.5 0.5
0
0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Angle of attack, α (in deg) Angle of attack, α (in deg)
(a) Lift coefficient comparison with (b) Drag coefficient comparison with
experimental value experimental value
Fig. 8 Solver validation by comparison with experimental results [44]
• Differences arising due to smoothness of the simulated geometry compared to the

rough surface of the tested model
• Numerical inaccuracies related to space and time based mathematical modeling
of real continuum flows.
• Lack of transition model for laminar to turbulent flow transition
• Imposition of the symmetry boundary condition.
.
Despite these differences, the numerical results were within acceptable devia-
tions for the two geometries. Hence, the same mesh and solver configurations were
appreciated to be valid for carrying out simulations for similar geometries with the
same flow conditions, and within the particular α range. Before proceeding on to the
analysis pertaining to comparison between grid fins of different g/c, Fig. 8a, b have
been used to draw a preliminary comparison between the aerodynamic performance
of grid fins and planar fins.
Lift is found to be higher for grid fin at all angles due to its multi-planar geometry
(Fig. 8a). Moreover, stall angle is significantly higher at α = 30◦ compared to the
planar fin (α = 20◦ ). Additionally, post-stall drop in lift is smoother for grid fin
compared to the conventional fins. Thus, grid fins present a tangible improvement
in its high α lifting characteristics, establishing their ability to provide elevated
manoeuvrability to a flight vehicle they get affixed upon. Notably, the smooth stall
characteristic of planar fin in Fig. 8a can be linked to its smaller span leading to
alleviated flow separation due to the presence of increased influence of wing-tip
vortices for lower A R [56]. Drag (Fig. 8b) is significantly higher for grid fin due to
higher skin friction drag and pressure drag linked to increased number of plates and
leading edge flow separation, respectively. Thus, increased lift at higher angles and
improved stall performance for grid fins are acquired at the cost of drag advancement.
This reasserted the need to carry out further analysis to alleviate this shortcoming,
and hence improve its applicability.
Fig. 9 Aerodynamic 5
efficiency comparison of grid
fin against planar fin
4
D
Lift to drag ratio, C /C
L
3
0 GF_GC0.5
Sim
PFSim
−1
0 10 20 30 40 50 60
Angle of attack, α (in deg)
Based on these coefficients, a concerted comparison between the aerodynamic

efficiency (CL /CD ) of planar and grid fins carried out in Fig. 9 displayed an average
decrement of ≈15% through all the angles for grid fin. In-fact this reduction was
calculated to be ≈24.3% within the most operable α (α = 5◦ − 15◦ ) region. This
efficiency reduction can be linked to the higher drag values for grid fins observed
earlier. However, at higher angles, efficiency displays tangible convergence for the
two variants. Thus, aerodynamic efficiency of grid fins remain comparable to planar
fins alongside improved stall angle. Notwithstanding the better performance at high
angles of attack, reduced aerodynamic efficiency of grid fins in the most operable
α region, establishes the need to study the impact of different geometrical parameters
on grid fin efficiency, which would assist in accomplishing an optimum grid fin
geometry based on lift, stall angle, and aerodynamic efficiency requirements. As
a part of the same study, analyzing the aerodynamic impacts of chord variation
while keeping the gap constant has been carried out in this paper, discussed in the
succeeding sections. This parameter has been non-dimensionalized by using the
g/c parameter (commonly employed for the analysis of such geometry [47]). It is
noteworthy that, span (b) was also varied along-with c such that, A R retained the
same value of 2.0 for all grid fin variants.
3.2 Aerodynamic Coefficient Comparison for Different g/c
As elaborated earlier, current study carries out numerical evaluations for different g/c
variations by varying the chord and retaining the same A R (A R=2.0) and compares
them against a planar fin (PF) of A R = 2.0.
Lift coefficient and stall angle are found to be higher for all grid fin variants com-
pared to PF. CL is seen to be increasing as g/c increases (chord reduces). Difference
between the successive g/c variant reduces at lower chords. The trend linked to
g/c = 2.0 displaying a sudden dip in lift at α = 10◦ leading to lower lifting forces
compared to g/c = 0.6 − 1.0 up to 40◦ is due to the same trend. Thus, after a certain
g/c value, this increase seizes, and hence we achieve reduced lifting performance for
higher g/c. Interestingly, stall angle is similar for all the variants. This is contrary to
the results as obtained by the authors in Ref. [47], wherein, it had been established
that lower g/c attained by reducing gap and retaining the same chord leads to reduced
lift but delayed stalling. Thus, it can be established that, chord variation has minimal
impact on the stall angle of the grid fins. Moreover, post-stall behavior relating to
gradual drop in lift is also similar for all the variants. Although, minimal improvement
can be asserted for higher g/c. Thus, high α lifting characteristics remain similar for
different chord variants. However, lift availability increases as chord reduces. The
flow-physics associated with these trends shall be addressed in a later section, while
describing the pressure coefficient and the flow field visualisation results.
Comparison in Fig. 10b, presents higher CD for higher g/c. It is to be noted that,
span variation to retain the same A R leads to lower span for higher g/c. Thus, the
grid fin with the lower chord (greater g/c) and narrower span presents higher drag
force. Thus, a more compact geometry, in-spite of its smaller surface area, presents
higher drag. It is noteworthy that, drag at such low Re is attributed to the presence of
skin friction and pressure drag. Moreover, it has been established [57] that, for a blunt
geometry, drag is mainly effected by pressure drag related to flow separation. Thus,
higher g/c variant leads to higher pressure drag, and hence the higher CD values
for the compact variant. This inference shall be backed up by pressure coefficient
analysis and flow visualisation study later in the paper.
The ratio of lift and drag coefficients is seen to be lower for higher g/c. Thus,
the increase in drag is more compared to lift advancement for higher g/c leading
to reduced efficiency for higher g/c. Moreover, efficiency of the grid fins approach
that of the PF as g/c lowers. It is noteworthy that, the impact of g/c variation on
the overall aerodynamic efficiency is found to be significantly high when chord is
the varying parameter, whereas the deviation was subtle when gap was the varying
parameter in Ref. [47]. Thus, this study establishes the significant impact of chord
on the overall aerodynamic efficiency, and hence can be a very helpful tool for a
control surface designer while selecting the final c value for optimising efficiency
and maximum lift.
Diligent comparison of the lift versus α curve slope (CLα ) can provide qualitative
information about the aerodynamics of control surfaces, and hence has been used as a
tool to compare the control performance of different grid fin variants. In Fig. 11a, the
same has been evaluated for different g/c variants below α = 10◦ (assuming linear
variation). Notably, additional evaluations were carried out for g/c = 1.1, 1.2, and
1.3 for α = 0◦ to 15◦ in order to comprehensively compare their slopes within this
range. This was carried out to validate the trend associated with the sudden dip in C L
for g/c = 2.0 observed earlier. The results present increase in slope as g/c increases.
However, a sudden dip in the slope can be discerned beyond g/c = 1.0. Thus, CLα
enhancement by virtue of chord reduction is delimited up to g/c = 1.0, after which
3
2
2.5
Coefficient of drag, CD
L
Coefficient of lift, C
1.5
2
1 1.5
1
0.5
0.5
0
0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Angle of attack, α (in deg) Angle of attack, α (in deg)
(a) Variation of coefficient of lift with (b) Variation of coefficient of drag with
angle of attack for different g/c angle of attack for different g/c
4
D
Lift to drag ratio, C /C
L
−1
0 10 20 30 40 50 60
Angle of attack, α (in deg)
(c) Variation of aerodynamic efficiency

with angle of attack for different g/c
Fig. 10 Variation of aerodynamic coefficients with angle of attack for different g/c
7 4
5°
(in rad )
−1
3.5 10°
6.5
Lift to drag ratio, CL/CD
15°
3
Lα
6
Lift coefficient slope, C
2.5
5.5
2
5
1.5
4.5 1
4 0.5
0 0.5 1 1.5 2 0 0.5 1 1.5 2
gap to chord ratio, g/c gap to chord ratio, g/c
(a) Lift-α curve slope for different g/c (b) Variation of aerodynamic efficiency
with different g/c for different angles of
attack
Fig. 11 Variation of lift slope and aerodynamic efficiency for different g/c at several angles of
attack
it starts reducing. This analysis can be a useful tool for grid fin designers based on
the limiting range of chord reduction for lift enhancement.
Following inferences can be established by analyzing the aerodynamic efficiency
at angles of attack 5◦ , 10◦ , and 15◦ for different g/c presented in Fig. 11b:
• Efficiency reduces as g/c increases (chord reduces), which was observed earlier
in Fig. 10c due to drag advancement.
• Maximum efficiency is found to be present at α = 10◦ , and hence can be considered
the best operating angle.
• Efficiency is found to be lowest at α = 5◦ for g/c = 0.5 and beyond, which reaf-
firms aerodynamic performance deterioration for lower chords even at such lower
angles of attack.
3.3 Aerodynamic Performance Deviation for Different g/c in

Relation to the Baseline Model
To further assess the effect of varying chord over the performance of grid fins, a
concise comparison was carried out by comparing the gain in maximum lift coef-
ficient (Fig. 12a), and average efficiency deviation (Fig. 12b) for different g/c grid
fin variants against the baseline model (GF_GC0.5) which is indicated by the dot
dashed line in these figures. The figure tangibly illustrates gain in maximum lift and
reduction in the efficiency as chord reduces. Appropriate diligence of these results
can further comprehend the application based constraints on chord variation for grid
fins. Lift slope at lower angles which would govern the controllability of the control
surface increases for the lower chord variant at the expense of aerodynamic efficiency
reduction.
Fig. 12 Variation of maximum lift coefficient and aerodynamic efficiency with different g/c
Thus, a grid fin with wider span and chord geometry would present a more effi-
cient design with regards to the aerodynamic efficiency parameter, which can lead
to reduced fuel consumption or increased payload carrying capacity. This would
lead to increased application based advantages for a wider chord grid fin. However,
a wider chord control surface necessarily dictates elevated hinge moment require-
ments. Hence, chord selection requires further optimization based on aerodynamic
efficiency, lift availability as well as the hinge moment constraints. In order to qual-
itatively assess the underlying flow-physics related to the aforementioned trends,
further analysis using the pressure coefficient (CP ) plots and flow field visualisation
using the velocity contour distribution was carried out in the succeeding sections.
3.4 Pressure Coefficient Comparison
An approach towards elucidating the underlying flow-physics involves examina-

tion of the CP distribution. Since, the flow involved is a subsonic flow, wherein the
contribution of pressure drag is found to be significantly higher compared to the
skin-friction drag, this approach can greatly assist us in comprehending the physics
leading to drag and lift variations.
Since, M corresponding to the flow being examined resides in the low subsonic
speed regime (M < 0.2), the flow can conveniently be considered an incompressible
flow. Hence, formula for calculating CP is given by Eq. 6.
p − P∞
CP = (6)
1
2
ρU∞
2
Here, p is the local static pressure, P∞ is the freestream static pressure, ρ and U∞
are the freestream density and velocity values, respectively (described earlier).
In Fig. 13, CP has been plotted along the chord length of different planar members
(also called as simply ‘members’) at the midspan location (Fig. 3a) for α = 5◦ , 15◦ ,
and 20◦ . Distance along the chord was non-dimensionalized (x/c) by dividing the
chordwise distance (x) by the respective variants chord (c) value.
The naming scheme implemented to describe the associated results follows the
terminology described in Table 2. For instance, L1_g/c = 0.4 notates, distribution
along the lower surface of Member 1 for the grid fin of g/c = 0.4. The member
surfaces can be categorized as cascaded surfaces and non-cascaded surfaces. Cas-
caded surfaces (L1, U2, L2, U3) are the ones surrounded by neighboring members,
whereas non-cascaded surfaces (U1 and L2) are exposed to the freestream flow.
Flow corresponding to α = 5◦ , is characterised by lowered flow separation for
all the members. Although, bluntness of the leading edges lead to leading edge
flow separation for all the members (Fig. 13a–c). This is indicated by the instant
lowering of CP for all the surfaces at their leading edges. Apparently, bluntness leads
to the formation of wake region behind the leading edges leading to reverse flow,
and hence the lowered pressure value. The lowering is mitigated for lower surfaces
α Member 1 Member 2 Member 3

1 1 1
P
P
0.5 0.5 0.5
Pressure coefficient, C
0 0 0
−0.5 −0.5 −0.5

5° −1 L1_g/c=0.4 −1 L2_g/c=0.4 −1 L3_g/c=0.4
U1_g/c=0.4 U2_g/c=0.4 U3_g/c=0.4
−1.5 L1_g/c=0.7 −1.5 L2_g/c=0.7 −1.5 L3_g/c=0.7
U1_g/c=0.7 U2_g/c=0.7 U3_g/c=0.7
−2 −2 −2
0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0
Non−dimensional chordwise distance, x/c Non−dimensional chordwise distance, x/c Non−dimensional chordwise distance, x/c
(a) (b) (c)

1 1 1
P
P
0.5 0.5 0.5
0 0 0
−0.5 −0.5 −0.5

15° −1 L1_g/c=0.4 −1 L2_g/c=0.4 −1 L3_g/c=0.4
U1_g/c=0.4 U2_g/c=0.4 U3_g/c=0.4
−1.5 L1_g/c=0.7 −1.5 L2_g/c=0.7 −1.5 L3_g/c=0.7
U1_g/c=0.7 U2_g/c=0.7 U3_g/c=0.7
−2 −2 −2
0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0
(d) (e) (f )
1 1 1
P
Pressure coefficient, CP
0.5 0.5 0.5
0 0 0
−0.5 −0.5 −0.5

20° −1 L1_g/c=0.4 −1 L2_g/c=0.4 −1 L3_g/c=0.4
U1_g/c=0.4 U2_g/c=0.4 U3_g/c=0.4
−1.5 L1_g/c=0.7 −1.5 L2_g/c=0.7 −1.5 L3_g/c=0.7
U1_g/c=0.7 U2_g/c=0.7 U3_g/c=0.7
−2 −2 −2
0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0
(g) (h) (i)
Fig. 13 Pressure coefficient distribution comparison across different members of varying g/c at
several angles of attack
Table 2 Naming scheme for pressure distribution plots

Naming scheme Description of the naming scheme
A P_g/c = X X A symbolizes the upper and lower surfaces of the respective members by
L and U, respectively
P is the number corresponding to the member varying as 1,2 or 3
according to the Member number shown in Fig. 2
g/c = X X represents the g/c variant with X X representing the g/c value
indicating reduced leading edge separation which is in-sync with the known physics.
Although, as flow approaches further downstream, separation region reduces, leading
to increase in pressure. However, CP increment for upper surfaces of all the members
is found to be slower for g/c = 0.7. Thus, higher g/c amounts to reduction in the
rate of leading edge flow separation deterioration. Hence, the reduced increment of
pressure for higher g/c variant would emphasize its increased drag value at lower
angle, which was established earlier in Fig. 10b. Despite these variations, due to
meagre differences with respect to flow separation at such low angles, the underlying
deviations for different g/c at such low angles is considerably small.
As α increases to 15◦ , Fig. 13d–f demonstrate absence of leading edge flow

separation over the lower surfaces. However, flow separation is higher on the upper
surfaces due to the enhanced wake region close to the leading edge corners. In-fact for
non-cascaded upper surface (U1) as seen in Fig. 13d, flow becomes fully separated,
which is indicated by the presence of constant chordwise pressure distribution. A
fully separated flow leads to the formation of a wake across the whole chord leading
to the main flow becoming independent of the freestream flow. This leads to the
formation of a near constant negative pressure distribution along the upper surface.
Notably, this angle is close to the stall angle of conventional fins. Hence, due to
the lack of cascading, the uppermost member stalls indicated by the presence of
constant CP distribution along its surface. Notably, CP distribution for non-cascaded
surfaces is comparable for both the g/c variants. However, CP distribution along the
cascaded lower surfaces (Fig. 13e, f) display increased reduction of CP along x/c for
lower g/c. This indicates the higher velocity acceleration for lower g/c (increased c
value). The trend associated with gradual CP increment for lower g/c along x/c for
the upper surface flow (as observed earlier for α = 5◦ ) was discernible at this angle
also. The elevated reduction in lower surface CP and the steeper increment in CP over
the upper surfaces of lower g/c grid fins lead to reduction in the overall pressure
differential, and hence the lowering of lift for lower g/c grid fins (as observed earlier
in Sect. 3.2). Similarly, the gradual CP recovery for the higher g/c variant indicates
higher pressure drag. This accounts for the elevated drag level for higher g/c as
established in Fig. 10b.
Similar deviation with respect to steeper decrement in CP values along x/c for
the cascaded lower surfaces (Fig. 13h) in comparison to non-cascaded lower surface
(Fig. 13i) can be delineated at α = 20◦ also . Moreover, CP advancement was again
slower for higher g/c. Thus, these plots explain the reasons leading to increased lift
and drag for higher g/c. It is also to be observed that, deviation between the two
variants for the lowermost member U3 is less prominent compared to the differences
related to CP distribution over U2. It has been established earlier by the authors
in Ref. [49], that this member stalls second after stalling of the uppermost member,
which ultimately determines the overall stall angle of the grid fin. Hence, the presence
of similar CP variation for the lowermost member of different chord, accounts for
their similar stall angle and post-stall behaviour.
3.5 Flow Field Visualization Using Velocity Contours
Qualitative understanding with respect to the aforementioned flow-physics can be

accomplished through implementation of flow-field visualisation methods. In this
section, the same was carried out by comparing the contours pertaining to the longi-
tudinal velocity component across grid fins at their respective mid-span locations in
Fig. 14 for different angles of attack and g/c. Assessment of the magnitude of this
component eventually indicates the amount of flow separation across a lifting mem-
Fig. 14 Velocity contour distribution for different g/c variants at midspan location for various
angles of attack
ber. It is noteworthy that, flow separation region using this method can be recognised
by the presence of a region consisting of negative velocity distribution.
Bluntness of the planar members lead to flow separation at the leading edges for all
the members of all g/c variants (Fig. 14a–l). Moreover, for a particular g/c variant,
the amount of flow separation increases as α increases. This separation is maximum
for the uppermost member, indicating its first-most tendency to stall. Although, for
the lower members, flow separation region is significantly smaller at higher angles
compared to the uppermost member. This results in the delayed stall characteristic
of grid fins. Differences between the various variants at low angles of attack is small
(Fig. 14a, d, g, j). Although, as α increases, disparity associated with flow between
the planar members becomes more appreciable. This is indicated by the reduction in
velocity between the members for higher g/c variants (Compare Figs. 14b, e, h, k or
14c, f, i, l). This again indicates the presence of higher drag between the members
for the grid fin with higher g/c leading to reduced velocity between their members.
This leads to increased drag values for higher g/c as mentioned earlier. It was also
noted that, size of the separation bubble remained similar for the lifting members
at same angles for various variants. This again explains the similar stall angle and
post-stall characteristic for different g/c variants (when c is the varying parameter).
This analysis further reaffirmed the presence of higher drag values for higher g/c
when chord is the varying parameter. Furthermore, stall angle similarity was also
affirmed through this analysis.
Lowering of aerodynamic efficiency alongside increase in the maximum lift coef-
ficient for increased g/c were the main observations during this study. Hence, chord
selection would require optimization based on lift, efficiency requirements, and the
associated hinge moment liabilities.
4 Conclusions
The paper discusses the impact of gap-to-chord ratio (g/c) variation on the flow-
dynamics of grid fins by virtue of chord variation while retaining the same aspect
ratio. The analysis is carried out by performing steady state numerical simulations
in the subsonic regime for high angles of attack. Some of the prominent inferences
drawn out from the study were as follows:
• The solver has been validated against experimental results available in the literature
for similar flow conditions and same angle of attack (α) range.
• Chord reduction (increasing g/c) leads to increase in the lifting performance up to
a certain bounded g/c, beyond which the lift starts reducing as g/c increases.
• Aerodynamic efficiency reduces as the g/c increases due to drag increment. Thus,
a more compact geometry with smaller chord and span would generate greater
drag forces leading to efficiency reduction.
• Variation of g/c by virtue of chord alterations, has minimal impact on the stall
angle, and a more significant impact on the maximum lifting force and aerodynamic
efficiency of the fins.
• Through elaborate comparison carried out for pressure coefficient distribution
and velocity contour plots across the different planar members of grid fins, the
underlying rudimentary flow-physics associated with these trends were explained.
• A grid fin designer can use this analysis for chord selection based on aerodynamic
efficiency requirement and lift availability. Additional consideration with respect
to hinge moment requirement would also be a criteria for chord selection.
• Further research pertaining to detailed comparison between gap and chord varia-
tion impacts would lead to a greater understanding with respect to their role on grid
fin aerodynamics. This can be collated with studies related to other geometrical
parameters like aspect ratio, etc. to determine the most optimum grid fin geometry
based on maximum lift, stall angle and aerodynamic efficiency requirements.
References
1. Zaloga S (2000) The Scud and other Russian ballistic missile vehicles. Armor at war series.
Concord Publications Co, New Territories, Hong Kong
2. Fulghum DA (2003) It is the big one (MOAB actually fits in a B-2). Aviation week & space
technology (16 Mar 2003). http://www.freerepublic.com/focus/f-news/874758/posts
3. SpaceX: Falcon 9. http://www.spacex.com/falcon9 (2015). Accessed on: 21st Oct 2017
4. Savelsberg R, Kiessling J (2016) North Korea’s Musudan missile: a performance assessment.
http://38north.org/2016/12/musudan122016 38 North Accessed on: 21st Oct 2017
5. Sreenivasulu J, Sivaramakrishnan E (2010) Aerodynamic analysis of a rocket configuration
with grid fin. In: International conference on fluid mechanics and fluid power. IIT Madras,
Chennai, India, pp 1–5
6. Palaszewski B, Field L (2012) Entry, descent, and landing with propulsive deceleration: super-
sonic retropropulsion wind tunnel testing. In: 50th AIAA aerospace sciences meeting including
the new horizons forum and aerospace exposition. Nashville, Tennessee, AIAA paper 2012-
401, pp 5808–5830
7. Fleeman EL (2006) Tactical missile design, chap. Aerodynamic considerations in tactical mis-
sile design, 1st edn. AIAA education series, AIAA, Inc., Reston, pp 40–42
8. Washington WD, Miller MS (1998) Experimental investigations of grid fin aerodynamics: a
synopsis of nine wind tunnel and three flight tests. In: Proceedings of the NATO RTO-MP-
5, missile aerodynamics, NATO Research and Technology Organization. Sorrento, Italy, pp
10–1–10–14
9. Munawar S (2010) Analysis of grid fins as efficient control surface in comparison to conven-
tional planar fins. In: 27th international congress of the aeronautical sciences, vol 1. Nice,
France, pp 1732–1737
10. Washington WD, Booth P, Miller M (1993) Curvature and leading edge sweep back effects
on grid fin aerodynamic characteristics. In: 11th AIAA applied aerodynamics conference.
Monterey. AIAA paper 93-3480-CP
11. Simpson G, Sadler A (1998) Lattice controls: a comparison with conventional, planar fins. In:
Proceedings of the NATO RTO-MP-5, missile aerodynamics, NATO Research and Technology
Organization. Sorrento, Italy, pp 9–1–9–10
12. Ledlow TI, Burkhalter JE, Hartfield RJ (2015) Integration of grid fins for the optimal design
of missile systems. In: 2015 AIAA atmospheric flight mechanics conference (AIAA Sci-Tech
Forum). Kissimmee, Florida, U.S.A., pp 620–649
13. Dupuis A, Berner C (2001) Aerodynamic aspects of a grid finned projectile at subsonic and
supersonic velocities. In: 19th international symposium on ballistics. Interlaken, Switzerland,
pp 495–502
14. Berner C, Dupuis A (2001) Wind tunnel tests of a grid finned projectile configuration. In: 39th
AIAA aerospace sciences meeting and exhibit. Reno. AIAA paper 2001–105
15. Abate G, Winchenbach G, Hathaway W (2001) Transonic aerodynamic and scaling issues for
lattice fin projectiles tested in a ballistics range. In: 19th international symposium of ballistics.
IBS 2001 symposium office thun, Switzerland, pp 413–420
16. Fournier E (2001) Wind tunnel investigation of a high L/D projectile with grid fin and con-
ventional planar control surfaces. In: 19th international symposium of ballistics. IBS 2001
symposium office thun, Switzerland, pp 511–520
17. Fumiya Hiroshima KT (2004) Grid pattern effects on aerodynamic characteristics of grid fins.
In: 24th international congress of the aeronautical sciences (ICAS). Yokohama, Japan, pp 1–11
18. Brooks RA, Burkhalter JE (1989) Experimental and analytical analysis of grid fin configura-
tions. J Aircraft 26(9):885–887. https://doi.org/10.2514/3.45856
19. Burkhalter JE, Hartfoeld RJ, Leleux TM (1995) Nonlinear aerodynamic analysis of grid fin
configurations. J Aircraft 32(3):547–554. https://doi.org/10.2514/3.46754
20. Burkhalter JE, Frank HM (1996) Grid fin aerodynamics for missile applications in subsonic
flow. J Spacecraft Rockets 33(1):38–44. https://doi.org/10.2514/3.55704
21. Pakkiri T (2007) Aerodynamic characterization of grid fins at subsonic speeds. J Aircraft
44(2):694–698. https://doi.org/10.2514/1.27653
22. Theerthamalai P, Nagarathinam M (2006) Aerodynamic analysis of grid-fin configurations at
supersonic speeds. J Spacecraft Rockets 43(4):750–756. https://doi.org/10.2514/1.16741
23. Zhaoqing X, Hailong P (2016) Dynamic characteristics analysis of cruise missile controlled by
grid fin. In: 2016 Chinese control and decision conference (CCDC). IEEE, Yinchuan, China,
pp 370–374. https://doi.org/10.1109/CCDC.2016.7531012
24. Khalid M, Sun Y (1998) A CFD investigation of grid fin missiles. In: 34th
AIAA/ASME/SAE/ASEE joint propulsion conference & exhibit. Cleveland. AIAA Paper
1998-3571
25. Chen S, Khalid M, Xu H, Lesage F (2000) A comprehensive CFD investigation of grid fins as
efficient control surface devices. In: 38th AIAA aerospace sciences meeting and exhibit. Reno.
AIAA Paper 2000-987
26. DeSpirito J, Edge H, Weinacht P, Sahu J, Dinavahi, S.P.: CFD analysis of grid fins for maneu-
vering missiles. In: 38th aerospace sciences meeting and exhibit. AIAA, Reno. AIAA paper
2000-391
27. DeSpirito J, Sahu J (2001) Viscous CFD calculations of grid fin missile aerodynamics in the
supersonic flow regime. In: 39th aerospace sciences meeting and exhibit. Reno, Nevada. AIAA
paper 2001-0257, p 257
28. DeSpirito J, Vaughn ME, Washington WD (2003) Numerical investigation of canard-controlled
missile with planar and grid fins. J Spacecraft Rockets 40(3):363–370. https://doi.org/10.2514/
2.3971
29. DeSpirito J, Vaughn Jr ME, Washington WD (2003) Subsonic flow CFD investigation of
canard-controlled missile with planar and grid fins. In: 41st AIAA aerospace sciences meeting
& exhibit. Reno. AIAA paper 2003-27
30. DeSpirito J, Washington WD, Vaughn Jr ME (2004) Numerical investigation of aerodynamics
of canard-controlled missile using planar and grid tail fins, part II: subsonic and transonic flow.
Tech. Rep. TR-3162, U.S. Army Research Lab, Aberdeen Proving Ground, MD
31. Lin H, Huang JC, Chieng CC (2003) Navier-stokes computations for body/cruciform grid fin
configuration. J Spacecraft Rockets 40(1):30–38. https://doi.org/10.2514/2.3912
32. Ma M, Deng Y, Zheng M, Zhou N (2005) Navier-Stokes computations for a grid fin missile.
In: 23rd AIAA applied aerodynamics conference, fluid dynamics and co-located conferences.
Toronto, Ontario, Canada. AIAA paper 2005-4973
33. Despeyroux A, Hickey JP, Desaulnier R, Luciano R, Piotrowski M, Hdespiritoamel N (2015)
Numerical analysis of static and dynamic performances of grid fin controlled missiles. J Space-
craft Rockets 52(4):1236–1252. https://doi.org/10.2514/1.A33189
34. Dikbas E, Baran ÖU, Sert C (2018) Simplified numerical approach for the prediction of aero-
dynamic forces on grid fins. J Spacecraft Rockets 55(4):887–898. https://doi.org/10.2514/1.
A34062
35. Li S, Jiang Z, Zhang W, Peng K (2015) Fluid-thermal-structure coupled analysis of grid fins
for hypersonic flight vehicle. In: 6th international conference on computational methods for
coupled problems in science and engineering. San Servolo Island, Italy, pp 701–712
36. Huang C, Wen L, Guowei Y (2017) Numerical studies of static aeroelastic effects on grid fin
aerodynamic performances. Chin J Aeronaut 30(4):1300–1314. https://doi.org/10.1016/j.cja.
2017.04.013
37. Schülein E, Guyot D (2006) Novel high-performance grid fins for missile control at high
speeds: preliminary numerical and experimental investigations. In: Meeting proceedings RTO-
MP-AVT-135, innovative missile systems, paper 35. RTO, Neuilly-sur-Seine, France, pp 35–
1–35–26
38. Zeng Y, Cai J, Debiasi M, Chng TL (2009) Numerical study on drag reduction for grid-fin
configurations. In: 47th AIAA aerospace sciences meeting including the new horizons forum
and aerospace exposition. Orlando. AIAA paper 2009-1105
39. Cai J (2009) Numerical study on choked flow over grid-fin configurations. J Spacecraft Rockets
46(5):949–956. https://doi.org/10.2514/1.41442
40. Debiasi M (2012) Measurements of the forces and moments generated by swept-back grid
fins. In: 30th AIAA applied aerodynamics conference. New Orleans, Louisiana. AIAA paper
2012-2909
41. Debiasi M, Zeng Y (2016) Forces and moments generated by swept-back grid fins with sharp
leading edges. J Aircraft 53(6):1964–1968. https://doi.org/10.2514/1.C033504
42. Dan W, Yong Y (2014) Numerical study on drag reduction for swept-back, swept-front, delta
grid fin with blunt and sharp leading edges. In: AIAA modeling and simulation technologies
conference. Atlanta, Georgia, U.S.A., AIAA Paper 2014-0638, pp 94–114
43. Misra A (2009) Investigation of grid and cascade fins for missile flight stabilization. Ph.D.
thesis, Indian Inst. of Technology, Kanpur, India
44. Misra A, Ghosh A, Ghosh K (2008) Cascade fins—an alternate tail stabilization unit. In: AIAA
atmospheric flight mechanics conference and exhibit. Honolulu, Hawaii. AIAA paper 2008-
6884
45. Kumar R, Misra A, Ghosh AK (2012) Nonlinear modeling of cascade fin aerodynamics
using Kirchhoff’s steady-state stall model. J Aircraft 49(1):315–319. https://doi.org/10.2514/
1.C031254
46. Kumar R, Misra A, Ghosh A (2011) Modelling of cascade fin aerodynamics near stall using
Kirchhoff’s steady-state stall model. Defence Sci J 61(2):157–164 https://doi.org/10.14429/
dsj.61.481
47. Tripathi M, Sucheendran MM, Misra A (2018) High angle of attack analysis of cascade fin in
subsonic flow. In: Singh S, Raj P, Tambe S (eds) Proceedings of the international conference on
modern research in aerospace engineering. Lecture notes in mechanical engineering. Springer
Singapore, pp. 121–131. https://doi.org/10.1007/978-981-10-5849-3_13
48. Tripathi M, Misra A, Sucheendran MM (2016) Numerical simulations of high angle of
attack subsonic flow past cascade fin for varying number of planar members. In: Interna-
tional NAFEMS conference on engineering analysis. Modeling, simulation and 3D-printing
(NAFEMS-3D). NAFEMS, Bangalore, Karnataka, India, pp 383–386
49. Tripathi M, Misra A, Sucheendran MM (2018) Effect of rectangular and airfoil planar mem-
ber cross-section on cascade fin aerodynamics. In: 2018 AIAA atmospheric flight mechanics
conference (AIAA Sci-Tech Forum). Kissimmee, Florida. AIAA paper 2018-1761
50. Metacomp Technologies: MIME user manual (2017) (Private Communication)
51. Spalart PR, Allmaras SR (1992) A one-equation turbulence model for aerodynamic flows. In:
30th aerospace sciences meeting and exhibit. AIAA, Reno. AIAA paper 1992-439
52. Salim SM, Cheah S (2009) Wall y+ strategy for dealing with wall-bounded turbulent flows. In:
Proceedings of the international multiconference of engineers and computer scientists, vol 2.
International Association of Engineers (IAENG), Hong Kong, pp 2165–2170
53. Ariff, M., Salim, S.M., Cheah, S.C.: Wall y+ approach for dealing with turbulent flow over a
surface mounted cube: Part 1–low reynolds number. In: Seventh international conference on
CFD in the minerals and process industries. Melbourne, Australia, pp 1–6
54. Technologies, M.: Accurate CFD for all regimes. http://www.metacomptech.com/index.php/
features/icfd (2017), cited on : 22nd Oct, 2017
55. Dwight RP, Lucas P, Bijl H (2010) Solution methods and acceleration techniques in computa-
tional fluid dynamics (CFD). Encyclopedia of Aerospace Engineering. Wiley, New York
56. Gabriel E, Mueller T (May 2004) Low-aspect-ratio wing aerodynamics at low Reynolds num-
bers. AIAA J 42(5):865–873. https://doi.org/10.2514/1.439
57. Anderson Jr JD (2010) Fundamentals of aerodynamics, chap. 1. Tata McGraw-Hill Education,
pp 75–80
Numerical Investigation on the Effect
of Propeller Slipstream
on the Performance of Wing at Low
Reynolds Numbers
K. Shruti and M. Sivapragasam
Abstract The flow over a flat plate airfoil with 5-to-1 elliptical leading and trailing
edge at Re = 80,000 and for different angles of attack (0°–15°) is numerically inves-
tigated by solving the Reynolds-averaged Navier–Stokes equations. The k − ω shear
stress transport equation, γ − Reθ turbulent transition model is used to address the
effect of laminar-turbulent transition. The present computed aerodynamic forces are
compared with the available experimental data for validation. Large flow separa-
tion and a single recirculation zone is found at higher angles of attack. The study is
extended to investigate the laminar separation bubble effect on a three-dimensional
Zimmerman wing planform for Re = 50,000; 100,000 and 150,000 at different
angles of attack. The present results agree well with the available experimental and
computational data. The influence of propeller on the aerodynamic performance of
a Zimmerman wing planform is investigated. The results show that the wing with
propeller configuration has lower C D values compared to wing alone case. The results
presented in this paper show the importance of modelling the propeller slipstream
effects on the aerodynamic characteristics of low aspect ratio wing.
Keywords Micro air vehicles · Low Reynolds numbers · Aerodynamic

characteristics · Propeller slipstream
K. Shruti (B) · M. Sivapragasam

Department of Automotive and Aeronautical Engineering, Faculty of Engineering and
Technology, M. S. Ramaiah University of Applied Sciences, Bangalore 560058, India
M. Sivapragasam
K. Shruti
Centre for Civil Aircraft Design and Development, CSIR-National Aerospace Laboratories,
Bangalore 560017, India

130 K. Shruti and M. Sivapragasam
1 Introduction
Micro air vehicles (MAVs) are asserting their impact in the modern aerospace
industry due to their wide range of applications. The recent interest in micro air
vehicles has motivated to study the aerodynamic performance of MAVs operate
at Reynolds numbers ranging from 104 to 105 . The key parameters of MAV are
maximum dimension less than 15 cm, operating range of 10 km and average
endurance of about 30 min [1]. At such low Reynolds numbers, the viscous effect
dominate the flow and has a tendency to separate even at mild adverse pressure
gradients leading to laminar flow separation, transition and perhaps reattachment,
leading to the formation of a laminar separation bubble. Due to this phenomenon the
aerodynamic characteristics are deteriorated.
Experiments on the aerodynamic characteristics of flat plate with 5-to-1 elliptical
leading and trailing edge and on low aspect ratio wings at the low Reynolds numbers
were done by Pelletier and Mueller [2]. Only a few experimental studies have been
conducted on the LSBs with 3D characteristics, while experiments on the 2D wing
section with the endplates have been investigated by many researches. Lain and Shyy
[3], indicated the transition can occur at Re > 104 , and therefore a transition model is
essential to capture detailed flow features. However, very little data exist on the inves-
tigation of Laminar Separation Bubble (LSB) for fixed low-aspect-ratio wings in this
Reynolds numbers range. Brandt and Selig [4], had performed many experiments
on series of propeller performances and characteristics. Propeller slipstream influ-
ence on wing aerodynamics at Re = 40,000 was studied experimentally by Makino
and Nagai [5]. Sudhakar et al. [6], had performed surface flow topology on the
surface of MAV wing at propeller-on and off condition. At propeller-off condition,
the symmetry flow patterns was observed on the wing. More recently, experimental
studies on propeller effects on wings at Re = 50,000–300,000 were presented by
Ananda et al. [7]. However, data for low aspect ratio (AR) wings relevant to MAVs
are sparse in the literature. Thus the study of aerodynamics at low Reynolds numbers
becomes pertinent and this is the main inspiration to carry out the present work.
In this paper, the aerodynamic characteristics of a flat plate airfoil with 5-to-1
elliptical leading and trailing edge with thickness-to-chord ratio 1.93% is numerically
computed at Reynolds number (Re = 80,000) for different angles of attack (α = 0°–
15°), and the study is extended to three-dimensional Zimmerman wing planform
with AR = 2. Later a propeller is attached ahead of the Zimmerman wing planform
to study the aerodynamic characteristics. The choice of the flat plate, wing planform
and the study at these Reynolds numbers, angle of attack was motivated by the
availability of the published results. The 3D computations are performed for Re =
50,000; 100,000 and 150,000 at various angles of attack (α = 0°, 4° and 8°). The
comparison between wing alone and wing with propeller configuration is studied to
understand the propeller slipstream effect on the aerodynamic characteristics, flow
behavior and span wise lift distribution. These factors alter the performance of MAVs
and these are key factors to design a MAV.
Numerical Investigation on the Effect of Propeller Slipstream … 131
Fig. 1 Cross-section of flat plate airfoil
2 Geometric Details and Numerical Model
2.1 Geometric Details
For current numerical investigations, a thin flat plate airfoil is selected as shown in
Fig. 1. Geometric details are inferred from the available experiments of Pelletier and
Mueller [2]. The flat plate was designed to have a 5-to-1 elliptical leading and trailing
edge, chord of 100 mm, thickness-to-chord ratio 1.93%. The Re based on freestream
velocity and chord length was 80,000.
2.2 Numerical Model
The Reynolds-averaged Navier Stokes equation are solved numerically using

commercial finite-volume method based ANSYS FLUENT software. An approach
to determine laminar-turbulent transition is to use transition SST model which is
available in FLUENT. Menter’s k − ω shear stress transport (SST) model is used for
the turbulence closure with the γ − Reθ transition model. The model consists two
additional transport equations, intermittency (γ ) and momentum thickness Reynolds
number (Reθt ).
γ − Reθ model typically correlates the transition momentum thickness Reynolds
number to local freestream conditions such as freestream turbulence intensity and
pressure gradient [8]. Default model constants are used for all the computations,
as they are capable of capturing the transition. Spatial discretization was done
using second order upwind scheme. The pressure–velocity coupling was achieved by
coupled scheme. All computations were performed in double-precision arithmetic.
2.3 Mesh Details
ANSYS ICEM-CFD software is used to generate a structured mesh around the flat
plate airfoil and wing planform. To minimize the far field boundary condition effects,
the domain is set at 10c upstream, 20c downstream, upper and lower boundaries are
placed at 10c, with all the distance considered from airfoil leading edge, where c
is the airfoil chord length. Total number of grid points used for computations are
15 0.04
10
0.02
5
0
y/c
y/c
0
-0.02
-5
-0.04
-10
-10 -5 0 5 10 15 20 -0.04 -0.02 0 0.02 0.04
x/c x/c
(a) (b)
Fig. 2 Mesh around flat plate airfoil and near the leading edge
42,412. Enough care is taken while generating the mesh and it is ensured for the
application of transition model, the average y+ value should be of the order 1. The
mesh around the flat-plate airfoil and mesh near leading edge is shown in Fig. 2.
Velocity inlet and pressure outlet boundary conditions are applied at the domain
upstream and downstream respectively. The wall boundary condition was imposed on
airfoil upper and lower surfaces which enforces no-slip, no-penetration of the fluid on
the airfoil surfaces. At the domain inlet, velocity was calculated using corresponding
to Re = 80,000. The turbulence intensity of 0.05% was used for this investigation.
Grid Independence Study was done for both 2D and 3D case. Enough care was taken
in generating a good quality mesh. The number of grid points for computations was
selected after careful grid independence study. For a 2D case, this study was done at
Re = 80,000 at α = 4° with three grids g1 with 19,754 (fine grid), g2 with 42,414
(medium grid) and g3 with 82,854 (coarse grid) grid points. The Grid Convergence
Index (GCI) was calculated for Cl and Cd , the results are summarized in Table 1. g2
grid had reasonably low value of discretization error and the same was used for all
the present computations.

Re = 80,000 at α = 4°
Grid No. of grid points Cl GCI for Cl Cd GCI for Cd
g1 19,754 0.4243 0.039 0.0346 4.048
g2 42,414 0.4204 0.0306
g3 82,854 0.4203 0.002 0.0298 1.007
Fig. 3 Comparison of Cl and Cd with experiment data
2.6 Aerodynamic Coefficients
A series of 2D simulations were carried out for a flat plate airfoil. The lift and drag
coefficients at Re = 80,000 at different angles of attack (α = 0°–15°) respectively.
The 2D simulations are compared with the experiment data of Pelletier and Mueller
[2]. The numerical results show good agreement with the experimental data until α
= 8°. The aerodynamic coefficients are shown in Fig. 3. At higher angles of attack,
aerodynamic coefficients are slightly low when compared to experimental data. This
is because during experiment at the 2D sectional wing, Pelletier and Muller [2] found
that endplates can lead to an increase in Cd . Menter’s transition SST model is not able
to capture the flow behavior at higher angles of attack. However, the computational
values are within the uncertainty range which was provided by Pelletier and Mueller
[2].
2.7 Flow Structure Over Flat Plate
At α = 0°, the flow remains attached on both the upper and lower surfaces of the
airfoil. In the low α range, α = 2° the flow structure is dominated by leading edge
separation, which enlarges gradually as the angle of attack increases. At α = 4°, a

very small separation region is present near the leading edge. At α = 6° the leading
edge separation bubble dominates the flow structure. At higher angles of attack α =
10°–15°, a large recirculation zone is seen on the airfoil upper surface. At α = 15°,
the single very large recirculation zone is formed on the airfoil upper surface. The
streamlines over the airfoil at α = 0°–15° is shown in Fig. 4.
The numerical results of flat plate airfoil presented in this section provided confi-
dence with the numerical simulation methodology to handle complex flow problems
α = 0° α = 2°
α = 4° α = 6°
α = 8° α = 10°
α = 12° α = 15°
Fig. 4 Streamlines over the airfoil at α = 0°–15°

involving LSB. In the next section, the numerical computations on the wing alone
and wing with propeller configuration are presented to understand the aerodynamic
characteristics and behaviour of flow.
3 3D Computations
3.1 Geometric and Mesh Details
The Zimmerman wing geometric details are inferred from the experimental studies
of Torres and Mueller [9]. The wing section profile remains same as in the 2D
study. The wing has zero camber, thickness-to-chord ratio (t/c) of 1.96%, root chord
of 0.25 m, wing span of 0.44 m and mean aerodynamic chord of 0.189 m. The
Re selected for the present computations was 100,000. For the same planform an
APC Slow-Flyer propeller was installed. Based on the experiments conducted by
Brandt and Selig [4], the APC Slow-Flyer had good performance characteristics and
a 10 × 4 [10] propeller was selected for the present computations. This propeller
was installed 0.044 m ahead of the wing leading edge, with the disk enclosing the
propeller of diameter 0.212 m. Propeller rotates in anti-clock wise direction when
viewed from front. For the present investigation, the rotational speed of propeller is
set from 540 rad/s (10,000 rpm) to 1626 rad/s (15,527 rpm).
ANSYS ICEM-CFD is used to generate structured grid. The C-H mesh topology
is used for 3D grid generation. For wing with propeller configuration a multi-block
strategy was employed. This structured mesh contains two domains, a circular domain
which encloses the propeller and a C-H outer domain containing the wing. Total
number of grid points used for computations for wing alone case was 1,630,438 (for
half wing) and for wing with propeller configuration was 3,767,830. Figure 5 shows
the mesh on the wing alone and on wing propeller configuration.
The wall and farfield boundary conditions are similar with the 2D case as mentioned
earlier. To minimize far field boundary effects, the far field domain is set at 10c
upstream, 20c downstream, upper and lower boundaries are placed at 10c, with all
the distances being from the wing leading edge. Here c is the mean aerodynamic
chord. Along the span wise direction the domain extent was set at one span length.
A symmetry boundary condition is imposed at the wing root, simulating the wing
with an AR = 2. At the symmetry plane, 3D mesh resembles the mesh which was
generated for the 2D case.
For the wing with propeller combination, boundary conditions remain same as that
discussed for wing alone case. However, symmetry boundary condition cannot be
Fig. 5 Mesh on the wing
applied due to the presence of the propeller. The full wing planform was simulated.
Interface boundary condition was imposed on the disk surrounding the propeller.
Wall boundary condition was applied on the propeller blades.
Grid independence study was performed at α = 4° for Re = 100,000. For grid

independence study three grids G1 with 820,480 points, G2 with 1,630,438 points
and G3 with 3,138,842 cells were used. The Grid Convergence Index (GCI) was
calculated for C L and C D , and the results are summarized in Table 2. G2 grid had
the reasonably low value of discretization error and the same was used for all present
computations. The wall y+ value was ensured to be less than 1 for all the computations.
Table 2 Summary of GIS

Re = 100,000 α = 4°
Grid No. of Grid Points CL GCI for C L CD GCI for C D
G1 820,480 0.1701 0.1501 0.0166 1.1643
G2 1,630,438 0.1717 0.0165
G3 3,138,842 0.1719 0.0171 0.0164 0.7520
3.4 Results and Discussion
3.4.1 Aerodynamic Coefficients
Numerical computations for Zimmerman wing alone and wing with propeller config-
uration were carried out for Re = 50,000; 100,000 and 150,000. The study is
performed at α = 0◦ , 4◦ and 8◦ . The present computational results agree well with
experimental data of Torres and Mueller [9] and an excellent agreement with the
computational data of Chen [11]. The results are shown in Fig. 6. The present compu-
tational results are within the experiment uncertainty range (10% for lift and drag)
[2].
The aerodynamic performances in terms of C L and C D at various angles of attack
is computed for three different Reynolds numbers. Typical results are presented for
Re = 50,000 and shown in Fig. 7.
Figure 7a, shows the lift coefficients of both wing alone and wing with propeller.
The lift coefficient is higher for wing-alone case when compared to wing-with
propeller at a given α. The lift coefficient for wing at α = 4° is 0.1689 and for
wing-with propeller it is 0.1060. This is because in the wing-alone case, larger LSB
occupies the upper surface of the wing which contributes to C L enhancement. In
wing-with propeller case, the LSB is washed out due to propeller slipstream effect.
Fig. 6 Comparison of C L and C D with experiment data, Re = 100,000

Fig. 7 Comparison of aerodynamic coefficients for wing alone and wing with propeller configu-
ration, Re = 50,000
Figure 7b, shows the drag coefficient. The propeller slipstream effect has a large
impact on the C D values. At α = 4°, the wing-alone has C D = 0.0186 and for
wing-with propeller is 0.0297, which is 111 drag counts higher.
3.5 Pressure Distribution
The pressure contours on the wing alone and wing with propeller is shown in Fig. 8.
The pressure distribution is different for wing alone and wing with propeller config-
urations. The flow field is symmetric about the x–y plane for the wing alone case.
However, for wing with propeller, asymmetric flow behavior is observed due to the
effect of propeller slipstream. The pressure is more on the right-side wing due to
the direction of the propeller blade rotation. The bubble is reduced with the wing
Z
Y
Cp: -1.27 -0.805 -0.34 0.125 0.59
Fig. 8 Pressure contours for wing alone and wing with propeller configuration
propeller case, due to the propeller flow and the bubble becomes asymmetric which
triggers a side force.
Figure 9 show the C p distribution at z/b = 0.2 for wing alone and wing with
propeller case at α = 4° for Re = 50,000.
The peak suction pressure is low for wing alone case and higher for wing with
propeller. Due to propeller slipstream effect the wing with propeller configuration
produces negative lift coefficients from z/b = (0 to −0.2). The separation region is
found to be more for the wing alone when compared to wing with propeller.
Fig. 9 C p distribution at z/b -4

= 0.2, α = 4° for wing alone
and wing with propeller -3 Wing alone
Wing with propeller
-2
-1
Cp
4
0 0.2 0.4 0.6 0.8 1
x/c
3.6 Flow Structure on Wing Planform
The 3D flow becomes very complicated and it needs a better understanding than 2D
flow structures. As per the topological rules by Tobak and Peake [12] we assume that
the body is simply connected and immersed in a flow that is uniform far upstream.
The topological rule is given by Eq. 1.
N − S = 2 (1)
where N = Nodal points.

S = Sadal points.
A detailed flow topology analysis is performed and the critical points on the wing
surface were identified and were found to satisfy Eq. 1.
The flow structure on the wing differs when compared between wing alone and
wing with propeller case. However, both the flow structure satisfy the topological
criteria of Tobak and Peake [12]. The flow structure on the upper surface of wing
alone and wing with propeller is shown in Fig. 10.
Fig. 10 Flow structure on the upper surface on the wing alone and wing with propeller
4 Conclusion
Numerical computation was carried out for a flat plate with 5-to-1 elliptical leading
and trailing edge at Re = 80,000, for different angles of attack. The behaviour of
aerodynamic characteristics were evaluated. The numerical results were compared
with the experimental data of Pelletier and Mueller [2]. The laminar separation bubble
(LSB) formation on the airfoil were investigated. As α increases the LSB increases
in both length and thickness on the airfoil upper surface. Different flow modes were
captured for different angles of attack and a large recirculation zone was observed
at higher angles of attack.
Zimmerman wing alone computations were performed for various Reynolds
number and for several angles of attack. The computational results are validated
with the experiment data of Torres and Mueller [9] and with the computational data
of Chen [11]. The Zimmerman wing has better aerodynamic coefficients. The flow
structure on the upper and lower surface of the wing is qualitatively represented and
the parameters of flow topology were discussed.
The Zimmerman wing with the propeller combination is computed to study the
effect of propeller slipstream on the wing. LSB is formed on the upper surface of the
wing, due to propeller slipstream effect the aerodynamic coefficients are altered. This
leads to decrease in C L and increase the C D . Propeller rotation alters the sectional
Cl distribution this leads to negative lift coefficient from z/b = (0 to −0.2).
The present study was based on the Zimmerman planform since many MAVs
employ this wing. It would be of fundamental and practical value to also study the
aerodynamic characteristics for other planforms as well. Further, in the present paper
we noticed that the propeller slipstream had some benefits in C D . However, the lift-to-
drag ratio of wing with propeller configuration was poorer. This fact would motivate
one to perform an optimization study including the propeller slipstream effects to
achieve enhanced aerodynamic characteristics for the objective functions.
References
1. Mueller TJ, DeLaurier JD (2003) Aerodynamics of small vehicles. Ann Rev Fluid Mech 35:89–
111
2. Pelletier A, Mueller TJ (2000) Low Reynolds number aerodynamic of low-aspect ratio,
thin/flat/cambered-plate wings. J Aircr 37(5):825–832. https://doi.org/10.2514/2.2676
3. Shyy W, Lain Y, Tang J, Viieru D, Liu H (2008) Aerodynamics of low reynolds number flyers.
Cambridge University Press, New York
4. Brandt JB, Seling MS (2011) Propeller performance data of low Reynolds numbers, 49th AIAA
aerospace sciences meeting, pp 1–8
5. Makino F, Nagai H (2014) Propeller slipstream interference with wing aerodynamic char-
acteristics of mars airplane at low reynolds number. In: AIAA, 52nd aerospace sciences
meeting
6. Sudhakar S, Chandankumar A, Venkatakrishnan L (2017) Influence of propeller slipstream on
vortex flow field over a typical micro air vehicle. Aeronaut J 121:95–113. https://doi.org/10.
107/aer.2016.114
7. Ananda GK, Selig MS (2018) Experiments of propeller-induced flow effects on a low-reynolds-

number wing. AIAA J
8. Menter FR, Langtry RB, Likki SR, Suzen YB, Huang PG, Volker S (2003) A correlation-based
transition model using local variables-part 1: model formulation. J Turbomach 128(3):413–422.
https://doi.org/10.1115/1.2184352
9. Torres GE, Mueller TJ (2004) Low-aspect ratio wing aerodynamics at low reynolds numbers.
J Aircr 43(5):865–873. https://doi.org/10.2514/1.439
10. Deshpande MD, Sivapragasam M, Umesh S (2014) Air suction system for chemical senor
for MAV applications. In: Compendium on advances in micro air vehicles in India, National
Design and Research Forum
11. Chen ZJ (2014) Micro air vehicle design for aerodynamic performance and flight stability,
Ph.D. Thesis, The University of Sheffield, UK
12. Tobak M, Peake DJ (1982) Topology of three-dimensional separated flows. Annu Rev Fluid
Mech 14:61–85
Theoretical Design and Performance
Evaluation of a Two-Ramp
and a Three-Ramp Rectangular Mixed
Compression Intake in the Mach Range
of 2–4
Subrat Partha Sarathi Pattnaik and N. K. S. Rajan
Abstract The present study focuses on the design and performance evaluation of
a rectangular mixed compression intake for ducted rocket ramjet applications. For
the theoretical design stage, a 1D optimization criterion has been used to fix the
ramp angles for theoretical maximum total pressure recovery (TPR), at on-design
M 2.9 (shock-on-lip). Two different ramp designs (two and three ramps) have been
considered for optimization, to find their effect on the overall performance. The throat
height has been fixed using the practical self-starting contraction ratio (CR), (i) for
the on-design Mach number of 2.9 and (ii) for the low off-design Mach number
of 2. The throat length used is about six times the throat height for the required
supercritical margin as well as to contain the shock train and the subsonic diffuser
divergence angle is about 6°. The viscous flow field has been obtained by solving
Favre averaged Navier-Stokes (FANS) equations with two-equation SST k-ω model.
The analysis shows that, for the two-ramp and three-ramp design with self-starting
contraction ratio at M 2.9, the on-design critical TPR is 0.645 and 0.67, and the
critical mass flow ratio (MFR) is 1 and 0.99, respectively. The performance at low
off-design M 2 shows, the MFR of both designs reduces to 0.51 and 0.47, respectively,
and improves to a value of 0.6 for the two-ramp configuration with self-starting CR
at M 2. This indicates that the two-ramp design has a better low off-design Mach
number performance.
Keywords Mixed compression intake · Supercritical margin · Compressible flow ·

Shock wave turbulent boundary layer interaction (SWBLI) · CFD
S. P. S. Pattnaik (B) · N. K. S. Rajan

CGPL, Indian Institute of Science, Bangalore 560012, India
N. K. S. Rajan

144 S. P. S. Pattnaik and N. K. S. Rajan
Nomenclature
α Angle of attack (AoA)

β Ramp/flow deflection angle of the jth shock
θ the shock wave angle of the jth shock
θd The lower wall divergence angle of the subsonic diffuser
Ac Maximum capture area
A∞ Freestream capture area
Mj Mach number after jth station
ṁ Captured mass flow rate
p0 Total pressure
P̄0 f Area weighted average total pressure at combustor face
P0 f max Maximum total pressure at the exit of intake
P0 f min Minimum total pressure at the exit of intake
MFR Mass flow ratio
Re Reynolds number
P̄0 f
TPR Total pressure recovery = P0∞
P0 f max −P0 f min
DI Distortion index = P̄0 f
SWBLI Shock wave boundary layer interaction
FANS Favre averaged Navier–Stokes equations
Throttling degree/ exit blockage ratio (EBR)
Subscripts
0 Stagnation/total condition
∞ Freestream condition
c Cowl
t Throat entrance
1, 2, 3 Ramp upstream station
f Intake exit/combustor entrance station
1 Introduction
In the recent years, ducted rocket ramjet engines are becoming popular for long
range air-to-air missile applications [1]. Here, the oxygen required for combustion is
taken from the atmosphere through an intake system and the missile carries only the
solid/liquid fuel. These missiles commonly use a fixed geometry rectangular or axi-
symmetric side intake. Designing such air intake systems is of key importance as it
Theoretical Design and Performance Evaluation … 145
Fig. 1 Schematic of a rectangular side intake
affects the overall performance of the engine. Their basic functions are to (i) capture
the required mass of air from the freestream at high velocity and compress it to low
velocity and high pressure before supplying it to the combustor, (ii) maximize the
static/total pressure recovery (TPR), (iii) provide a flow with good homogeneity (low
pressure distortion) and stability for each flight conditions (subcritical buzz stability
limit), (iv) operate efficiently with the pressure fluctuations due to combustion at a
range of F/A ratios and off-design conditions.
The major parts of a ramjet intake are a supersonic diffuser, throat and a subsonic
diffuser. For a rectangular intake, the supersonic diffuser can further be divided as
ramp, cowl and side plates as shown in Fig. 1. As the flow enters the intake, it goes
through a series of oblique shocks in the supersonic diffuser, a terminal normal shock
at the throat and the subsonic flow after the normal shock is compressed further in the
subsonic diffuser to the required combustor Mach number at the exit. In this process,
the flow incurs stagnation pressure loss across the shocks, viscous boundary layer
and their interactions. A better explanation of this flow field can be found from the
work of Saha et al. [2]. Hence, designing an intake involves selecting the different
geometric components and optimizing them to minimize the losses or to maximize
the performance in terms of total pressure recovery and mass capture as well as to
have minimal drag and flow field distortion at the exit.
Several studies have been reported in the literature on supersonic intake, (i) theo-
retical design approach [3–5], (ii) experimental and numerical performance evalua-
tion, focusing on the effect of various geometric concepts [6–9], (iii) intake unstart
[10, 11] as well as (iv) subcritical oscillating flow behavior (buzz) [12–14]. A detailed
review of literature shows, the supersonic compression structure, internal contraction
ratio and throat length are the major source of performance loss in an intake and their
selection is operational range dependent. Hence, the objective of the current study
is to design and evaluate the performance of a two-ramp and a three-ramp mixed
compression rectangular intake at Mach 2.9, considering the self-starting contrac-
tion ratio at the on-design Mach number of 2.9 and the lower off-design Mach number
of 2.
Consequent to this introductory note, the paper has been organized as follows:
Section 2 describes a theoretical design methodology, for fixing the ramp angles
and other geometric dimensions of intake. Section 3 describes the numerical simu-
lation procedure to find the viscous flow field. And in the final section, a detailed
comparison of performance parameters for different geometric designs is presented.
The maximum on-design TPR for the two-ramp design is close to 0.645 and that
of three-ramp design is 0.67. Here, only the results of 2D simulation are presented.
The 3D effect and complete off-design performance analysis will be considered for
further studies.
2 Modes of Operation
The operation of ramjet air intakes can be divided into three different modes, charac-
terized by the position of the terminal normal shock/shock train in the intake. These
are sub-critical, critical and supercritical modes of operation and are primarily the
function of engine demand or downstream combustor pressure condition.
At low downstream pressure, the flow at the throat expands after the cowl shock
and forms a normal shock downstream (of throat) at a higher Mach number, resulting
in higher stagnation pressure loss or low stagnation pressure recovery is known as the
supercritical mode of operation. As the downstream pressure increases, the normal
shock structure moves upstream towards the throat and gradually weakens. The
maximum recovery is achieved when the terminal shock stands at the throat, known
as critical operation. With further increase in downstream pressure, the normal shock
moves upstream of the throat resulting in large flow separation and then moves outside
the cowl, leading to unstart. This process is called as the subcritical operation. In
this case, the ramp oblique shocks and normal shock interact outside the cowl, to
form a bowed/curved shock ahead of it. And, a series of slip planes/vortex sheets
are produced at the point of interaction. Here, as the shock stands ahead of the cowl,
the MFR decreases compared to the critical operation (because of mass spillage) as
well as the total pressure recovery decreases based on the losses across the shocks
[15]. Usually, it is observed that till a certain range of subcritical operation, the
flow field remains stable known as stability margin. And below the stable subcritical
operation, flow field (Shock system) starts to oscillate, resulting in fluctuations in
pressure recovery and mass flow capture. This unstable condition is known as buzz
[14]. In order to avoid this unstable condition during the actual operation, all intakes
are conventionally designed with a supercritical margin, where the intake operates
at a supercritical mode, and as the back pressure increases, the shock system adjusts
to have a critical operation. Hence, the subcritical operation is avoided [15].
These modes of operation of the intake are conventionally represented through
an inlet characteristic diagram, as shown in Fig. 2. This shows the variation of total
pressure recovery and mass flow ratio (defined as the ratio of actual freestream mass
captured to the maximum possible mass capture) for different modes of operation.
Fig. 2 Example of inlet characteristic diagram
3 Theoretical Design
The theoretical design stage of the intake involves selection of the different internal
geometric concepts that is the selection of supersonic compression structure, throat
length and subsonic diffuser divergence angle to maximize the performance. For the
current application, the intake has been designed at an on-design Mach number of
2.9. Hence, in order to have a compromise between length, weight, external drag and
internal boundary layer effects, a mixed compression intake configuration has been
selected which is considered to be efficient above a design Mach number of 2.5 [16].
3.1 Selection of Supersonic Compression Structure
As mentioned earlier, to achieve the subsonic flow at the exit of the intake, the flow
gets compressed through a series of oblique shocks and a terminal normal shock.
Hence, for the supersonic compression geometry, higher the number of ramps for
a fixed freestream Mach number, higher is the total pressure recovery that can be
achieved. This is graphically represented in Fig. 3. But higher number of ramps leads
to increase in geometric length, weight, viscous boundary layer losses and cowl drag,
in addition the low off-design Mach number performance decreases. Hence, going
for higher and higher number of ramps or isentropic ramp is not always useful. So,
for the present study, two ramp designs, (i) a two-ramp and (ii) a three-ramp design,
have been considered to find their effect on the overall performance of the intake.
For fixing the ramp angles, a 1D optimization criterion is used. The process aims
at maximizing the TPR or minimizing the total pressure loss for a system of oblique
Fig. 3 Shock pressure

recovery for freestream
Mach number and number of
oblique shocks [4]
Fig. 4 Schematic of theoretical shock system in the designed intake
shocks, a terminating normal shock and the subsonic diffuser, as pictorially summa-
rized in Fig. 4. The optimization closely follows the method used by Benson [3] and
Ran et al. [4], for which the oblique shock relations at ramp and cowl shocks, normal
shock relations [17], Oswatitsch criterion (1944) for planar shocks and empirical
relation for subsonic diffuser pressure loss [5] have been solved combinedly for a
freestream Mach number of 2.9 using MATLAB. The optimization process for a
five-shock (four oblique shock, one normal shock) system has been described below.
Let the freestream Mach number is M∞ . The shock angle and ramp angle at
the different ramp stations be θ1 ,β1 , θ2 ,β2 , θ3 ,β3 and at cowl be θ4 ,β4 . The Mach
number at each ramp be M1 , M2 , M3 , M4 and after the terminal normal shock be
M5 . The Mach number at the end of subsonic diffuser be M6 , which is fixed by the
combustor design requirement. Here, all the angles mentioned are measured from
the flow direction, just before the particular station, considering an angle of attack of
α. The equations relating the Mach number, ramp angle, shock angle are as follows:
(1) Oblique shock relations at the ramp and cowl (Eqs. 1–8)
(γ + 1)M 2
cot β1 = tan θ1 [ 2 2 ∞ − 1] (1)
2 M0 sin θ1 − 1

M02 (γ + 1)ε1 + (γ − 1) − 2 ε12 − 1
M1 =
2
(2)
ε1 (γ − 1)ε1 + (γ + 1)
2γ M02 sin2 θ1 −(γ −1)

where ε1 = P1
P0
= (γ +1)

(γ + 1)M12
cot β2 = tan θ2 2 2 −1 (3)
2 M1 sin θ2 − 1

M12 (γ + 1)ε2 + (γ − 1) − 2 ε22 − 1
M2 =
2
(4)
ε2 (γ − 1)ε2 + (γ + 1)
2γ M12 sin2 θ2 −(γ −1)

where ε2 = P2
P1
= (γ +1)

(γ + 1)M22
cot β3 = tan θ3 2 2 −1 (5)
2 M2 sin θ3 − 1

M22 (γ + 1)ε2 + (γ − 1) − 2 ε22 − 1
M3 =
2
(6)
ε2 (γ − 1)ε2 + (γ + 1)
2γ M22 sin2 θ3 −(γ −1)

where ε3 = P3
P2
= (γ +1)

(γ + 1)M32
cot β4 = tan θ4 2 2 −1 (7)
2 M3 sin θ4 − 1

M32 (γ + 1)ε + (γ − 1) − 2 ε2 − 1
M4 =
2
(8)
ε (γ − 1)ε + (γ + 1)
2γ M32 sin2 θ4 −(γ −1)

where ε4 = P4
P3
= (γ +1)
(2) As per Oswatitsch (1944) criteria [4], to obtain maximum pressure recovery
for a system of planar oblique shocks and a normal shock, the shock strength
should be equal that is the Mach number perpendicular to the individual shocks
is equal. Hence, the oblique shock angles can be related as
M0 sin θ1 = M1 sin θ2 (9)
M1 sin θ2 = M2 sin θ3 (10)

M2 sin θ3 = M3 sin θ4 (11)
(2) From the normal shock relations, the Mach number across it can be related as
(γ − 1)M42 + 2
M52 =
2γ M42 − (γ − 1)
Hence, the number of unknown variables in the shock system is 13 and number
of equations relating them is 12. As the number of unknowns is less than that of
number of equations, to solve the system, any one of the variables that is Mach
number before the normal shock (M4 ) or the shock angle θ1 can be assumed and
solved for the remaining variables for a fixed inlet Mach number. Now, with the
known shock angles, the total pressure recovery for the system can be found as
described below.
Calculation of total pressure recovery. The following stagnation pressure recovery
(PR) relations have been used at different sections of the intake to obtain the total
pressure recovery.
(a) PR across the oblique shock
γ
Pti (γ + 1)εi + (γ − 1) γ −1 −1
PRi = = ∗ εiγ −1 (13)

Pti−1 (γ − 1)εi + (γ + 1)
where i = 1,2,3,4
(b) PR across the normal shock

(γ + 1)M42 γ γ−1 (γ + 1) 1
PR5 = ] ] γ −1 (14)
(γ − 1)M42 + 2 2γ M42 − (γ − 1)
(c) PR across the subsonic diffuser
The flow in the subsonic diffuser happens across an adverse pressure gradient.
So, if the pressure gradient is high, the flow can separate from the wall resulting
in a stagnation pressure loss. It mainly depends on the divergence angle, length
to height ratio, upstream Mach number as well as the flow condition (boundary-
layer separation) at the throat. In the literature, a number of empirical relations can
be found formulating the pressure loss across the subsonic diffuser, of which the
relation provided by Bridges [5] using the experimental data of Nicolai has been
used here. This can be defined as below:
The stagnation pressure recovery, in the subsonic diffuser is given by:

⎡ ⎤
⎢ 1 ⎥
PR6 = 1 − κ(M)⎣1 − γ γ−1 ⎦ (15)
γ −1
1+ 2
M52
where the diffuser loss co-efficient, κ(M) is:
κ(M) = 0.2867M − 0.0875
Now, the total pressure recovery in the intake is given by

6
TPR = PRi (16)
i=1
Optimized ramp angles. The above process has been repeated by assuming a range of
Mach number upstream of the normal shock (M4 ), from which the shock/ramp angles
that produce the maximum total pressure recovery are obtained by plotting the TPR
with the Mach number upstream of the normal shock. The result of 1D optimization
for the two-ramp and three-ramp system is as shown in Fig. 5. The ramp angles
corresponding to maximum TPR have been used for designing the geometry. Table 1
shows the summary of optimized ramp angles for both the designs. The theoretical
TPR and the cowl angle obtained for the two-ramp case is about 0.83 and 8.5°,
respectively, and that for the three-ramp case is about 0.86 and 16°, respectively.
3.2 Throat Design
For a mixed compression supersonic intake, the throat height and length are important
design variables as they affect the overall performance of intake. The concepts that
have been used to fix these variables are explained below.
Throat height. Usually, the cowl lip height of an intake is fixed by the overall system
or drag requirement. From the cowl lip height and the optimized ramp angles, the
inlet area can be known. Now for the fixed inlet/cowl area, the throat area or the throat
bottom surface can be found out using the “self-starting contraction ratio limit” for
the design Mach number. It is defined as the ratio of cowl area to the throat area,
above which complete incoming mass cannot be captured in the intake, resulting in
the formation of a bow shock or normal shock ahead of cowl to allow mass spillage.
And the operation of the intake is said to be unstarted.
In the literature, there are several studies aiming to find the self-starting contraction
ratio limit with upstream Mach number for different geometries, starting from the
Fig. 5 Result of 1D optimization for a two-ramp, b three-ramp intake
Table 1 Ramp angles and theoretical performance for different ramp designs
Ramp structure M0 β1 θ1 M1 P1 /P0 β2 θ2 M2 P2 /P1
Two-ramp 2.9 10.5 28.5 2.4 2.08 12.5 35.5 1.9 2.08
Three-ramp 8.05 26.4 2.51 1.773 9.1 31 2.13 1.773
β3 θ3 M3 P3 /P2 β4 θ4 M4 P4 /P3 M5 TPR
14.5 46.3 1.383 2.08 0.747 0.82
10.3 37.13 1.75 1.773 11.5 49 1.34 1.773 0.786 0.861
theoretical work of Kantrowitz [10]. Sun and Zhang [11] have recently compiled all
the data from literature and provided the practical limit of self-starting contraction
ratio with upstream Mach number. This practical limit has been used here for the
selection of the contraction ratio.
For the present study, two different contraction ratios are used for both two-ramp
and three-ramp designs. These are
(i) A ratio of 1.316, which is the self-starting contraction ratio for the internal
compression, for the 2-ramp design. And a ratio of 1.282, which is the self-
starting contraction ratio for the internal compression, for the three-ramp design
at a freestream Mach number of 2.9.
(ii) A contraction ratio of 1.111, which is the self-starting contraction ratio at off-
design freestream Mach of 2 for both two- and three-ramp design.
Throat length. At the critical mode of operation in the ramjet intake, the super-
sonic flow terminates with a normal shock at the throat. But practically, instead of
a single terminal normal shock, usually a normal shock train is formed because of
the SWBLIs, depending on the upstream flow condition [18]. The structure of shock
train and the different parameters affecting it can be found from the earlier studies in
the literature. Hence, the length of throat should be such that, it should completely
contain the shock train within to achieve the maximum performance at the intake
exit in terms of flow uniformity and low subsonic Mach number.
In the present design, as a first iteration, a fixed throat length of 80 mm has been
used which is about six times the throat height for both two-ramp and three-ramp
design.
3.3 Subsonic Diffuser Design
The subsonic diffuser lower wall divergence angle affects the axial adverse pressure
gradient across which the flow occurs, hence is an important design variable. As per
the literature, an angle of 4–10° is usually preferred for this subsonic diffuser lower
wall angle of divergence. Thus, to have minimal flow separation, for the present case,
a 6°-divergence angle has been used.
3.4 Comparison of Inlet Dimensions
Boundary layer correction. The 1D optimization procedure used here to obtain the
ramp angles does not consider the viscous effects. Thus, in order to achieve shock-
on-lip condition considering viscous boundary layer effects, the ramp angles need to
be corrected to take care of displacement thickness. As the supersonic boundary layer
thickness is small, here a simple method has been used instead of going for exact
boundary layer displacement thickness correction. To take care of this, all the ramp
angles have been reduced by 0.5° each. As will be shown from the CFD simulation,
this is good enough to keep the streamlines within the cowl lip.
Geometric dimensions. The corrected ramp angles have been used to obtain the
geometric dimensions for the intake, with a fixed cowl tip height of 40 mm. Here,
the oblique shocks generated by the three ramps are designed to focus on the cowl
leading edge. The throat and subsonic diffuser dimensions are then fixed as explained
before. The major geometric dimensions of the intake are presented in Table 2.
Table 2 Geometric dimensions

(mm) Total length Cowl angle Throat height Throat length Subsonic diffuser length
Two-ramp 400 8.5° 15.2 Fixed
Three-ramp 406.7 16° 13.5 80 250
4 Computational Methodology
To evaluate the performance of the initial design considering viscous effects and
shock wave turbulent boundary layer interactions, 2D CFD analysis has been carried
out. The numerical methodology adopted for the analysis is explained in this section.
4.1 Geometry Details and Grid Generation
The intake geometry obtained from the theoretical analysis has been used for the
numerical study. To simulate the different modes of operation, a downstream throttle
has been used to vary the back pressure by varying the throat area. The computational
domain used for the study is as shown in Fig. 6a. The top and bottom external
geometry heights are about 15 and 5 times the cowl height, respectively. This large
outer domain has been considered to avoid any influence of freestream at angle
of attack simulation. The computational domain has been meshed with complete
hexahedral cells as shown in Fig. 6b using ICEM CFD [19].
4.2 Numerical Details
To solve the flow through the intake, which is compressible and turbulent, Favre
averaged Navier–Stokes (FANS) equations with two-equation SST k-ω turbulence
model have been employed [20]. The governing equations are integrated using finite
volume technique and solved using commercial CFD software: CFX-17. Even though
the flow field in the intake is completely three-dimensional, to optimize the length
and the geometry of subsonic diffuser, only a quasi 2D simulation has been carried
out, which considers a single cell in the 3D direction. The fluid properties like specific
heat capacity (C p ), thermal conductivity (K) and viscosity are modelled as a function
of temperature, where viscosity is modelled using the model provided by Sutherland
[1893], obtained from kinetic theory for dilute gases. The specific heat variation with
temperature has been modelled using a curve fit. And the thermal conductivity of air
has been modelled using the modified Euken Model (2001) [20].
Boundary condition. The boundary conditions applied for the domain are as shown
in Fig. 6a. All the solid walls are modelled as stationary and no-slip adiabatic
Fig. 6 Schematic of the computational a domain with applied BC, b grid used
boundary. At the inlet to the supersonic intake, all the freestream conditions are
applied as a supersonic velocity inlet boundary, i.e. static pressure, static tempera-
ture and velocity are specified. The turbulence quantities at the inlet are specified
in terms of eddy viscosity ratio and turbulence intensity. The outlet boundary in the
outer regions above the cowl and below the ramp is assigned with subsonic outlet
where static pressure is specified. And the outlet after the downstream throttle has
been modelled as supersonic outlet.
Convergence metrics. The point of convergence of the numerical simulations has
been identified by considering the overall mass, momentum, energy balance in the
domain, with a criterion of residuals below 10−5 and the flow property monitors
getting stabilized.
Validation studies. To ensure the effect of turbulence model and boundary condition
uncertainties, the numerical methodology has been validated with the intake experi-
mental data of Emami et al. [21]. For this, a 2D dual-mode ramjet inlet geometry has
been considered for the simulation. The geometry consists of a supersonic diffuser,
an isolator followed by an expanding section which acts as a subsonic diffuser in the
ramjet mode of operation. The supersonic diffuser includes a 11° compression ramp,
cowl and side plates with leading-edge radius of 0.005 in. each. The isolator section
considered is a constant rectangular area of length to throat height ratio (L/Ht ) of 4.7,
where Ht is 0.4 in., followed by a rectangular 6° expanding section of length 4.25 in.
The subsonic diffuser considered has a lower wall divergence angle of 20°.
For this validation study, both 2D and 3D simulations considering the effect of side
walls have been carried out. Figure 7 shows the ramp and cowl wall static pressure
variation obtained from 2D simulations. A closer look at the points of pressure rise
clearly indicates, the separation regions due to the shock wave boundary interactions
are not captured well but the pressure peaks and the trend of shock structure are
predicted quite accurately in the 2D simulation.
The similar results of 3D simulation are shown in Fig. 8. It can be seen that, here
the effect of side plate shock and the separation regions is predicted better. Hence, the
overall favourable agreement between the predicted results with the data demonstrate
Fig. 7 Comparison of predicted non-dimensional wall static pressure from 2D simulation, along
a ramp wall, b cowl wall with experiment [21]; M ∞ − 4.0
Fig. 8 Comparison of predicted non-dimensional wall static pressure from 3D simulation along,
a ramp wall, b cowl wall midplane with experiment [21]; M∞ − 4.0
that this numerical methodology can be used for the preliminary design optimization
of the intake before final experimentation.
To have a mesh with optimal number of cells, three sets of grids have been considered
for the simulation and the wall static pressure has been compared. The grids are
systematically refined and have a resolution as given in Table 3. Figure 9 shows the
variations of ramp and cowl wall static pressure distribution for all the three grids.
Table 3 2D mesh resolution

Level of Cells First cell Maximum
resolution (millions) distance y wall y+
(mm)
Coarse 0.094 0.01 11
Intermediate 0.148 0.001 1.5
Fine 0.181 0.0008 <1
Fig. 9 Comparison of non-dimensional wall static pressure for three different grids along,
a bottom/ramp wall, b top/cowl wall of the 2-ramp intake; M∞ − 2.9, CR − 1.316
This clearly shows that the wall pressure trend is very well captured in all the grids
and the variation of pressure peaks from the intermediate mesh result to the fine mesh
is almost negligible. Hence, the intermediate mesh with 0.148 million cells has been
considered for all the simulations.
This section presents a detailed comparison of the internal flow field for the different
designs and their effect on the performance parameters.
5.1 On-Design Performance Characteristics
The various modes of operation of the intake like subcritical, critical and supercritical
modes have been simulated by controlling the downstream throttle condition/pressure
at the on-design Mach number of 2.9.
Two-ramp design. Figure 10 shows the contours of Mach number, axial density
gradient and the streamlines for two different modes of operation. And the corre-
sponding ramp wall static pressure variation for the complete range of operation is
shown in Fig. 11. It can be seen that, in the supercritical modes of operation, the
pressure increases along the two ramps through the oblique shocks and across the
incident shock from the cowl. Following the cowl shock, the flow goes through a
series of repeatedly reflected oblique shocks and expansion fans till the normal shock
formation in the subsonic diffuser. This results in a continuous change in pressure at
the throat and then decreases in the diverging section due to flow expansion. Down-
stream of the normal shock, the flow pressure increases in the subsonic diffuser till
the intake exit. As shown in Fig. 10a, in this case instead of a single normal shock,
a shock train is formed. As we go from the supercritical to critical operation, the
normal shock train moves upstream. For the operation with supercritical margin, the
complete normal shock remains within the throat. But the mixing region after the
shock train extends into the subsonic diffuser resulting a higher exit Mach number
of about 0.58. For the critical operation, the shock train starts from the throat/cowl
shoulder is the condition of maximum pressure recovery. It can be seen from the pres-
sure plot that the supercritical margin for the design is small and the complete shock
train is not contained within the throat requiring a higher length. For the subcritical
operation, the terminal moves upstream resulting in a bow shock ahead of the cowl
and the shock starts to oscillate. Therefore, as discussed in Sect. 2, further lower
subcritical operations have not been simulated here. It can be observed that the flow
separation occurs in the subsonic diffuser at all operating conditions leading to large
flow distortion at the exit.
Three-ramp design. The wall static pressure and internal flow field for the three-
ramp design are shown in Figs. 12 and 13, respectively. In this case, the flow gets
compressed through three-ramp shocks before entering the throat. And the cowl
shock impinges slightly downstream of the throat. Further downstream, a similar flow
field is observed as explained for the two-ramp design. The static pressure variation
clearly indicates that the sustainable back pressure ratio at the critical operation
increases to about 20 compared to about 19 for the 2 ramp design.
Performance comparison. The variation of on-design total pressure recovery with
different modes of operation for the two-ramp and three-ramp design is shown in
Fig. 14. Clearly, the critical pressure recovery for the two-ramp design is about 0.645
and that of three-ramp design is about 0.67. The critical mass flow ratio for two-ramp
case is 1 and that for the three-ramp is 0.99. The slight reduction in the mass flow
ratio for the three-ramp design occurs because of the fact the ramp shocks impinge
marginally outside the cowl tip.
Fig. 10 Contours of internal flow field at different modes of operation a supercritical b supercritical
margin for two-ramp design; M∞ − 2.9, CR − 1.316
Fig. 11 Non-dimensional
ramp wall static pressure
variation at different modes
of operation for the
two-ramp design; M∞ − 2.9,
CR − 1.316
Fig. 12 Non-dimensional
ramp wall static pressure
variation at different modes
of operation for the
three-ramp design; M∞ −
2.9, CR − 1.282
5.2 Off-Design Performance Characteristics
To study the performance (TPR, MFR) of the intake at off-design operation, the
intake flow field has been simulated at four off-design Mach numbers with zero
degree angle of attack (AoA). Those are Mach 2, 2.4, 3.5 and 4 with an exit throttle
condition corresponding to critical mode of operation at the on-design condition.
The performance variation is shown in Fig. 15. Here, all results for design with
self-starting CR at M 2.9 are presented.
Fig. 13 Contours of internal flow field at operation with supercritical margin for three-ramp design;
M∞ − 2.9, CR − 1.282
Fig. 14 On-design
performance characteristics
for two-ramp and three-ramp
design; M∞ − 2.9
TPR. The variation of critical total pressure recovery at the exit of the intake indicates
that the TPR falls drastically with increase in off-design Mach number. This is due
to the losses across the strong shocks produced in the intake. Further, it can be seen
that, for the three-ramp design, the off-design TPR is smaller compared to that of
two-ramp design, even though the on-design performance is higher.
Fig. 15 Variation of a TPR and b MFR with off-design Mach number operation
MFR. The variation of MFR at off-design clearly shows that the critical MFR reduces
to 0.51 and 0.47, respectively, for two-ramp and three-ramp design at M2 operation.
The drastic reduction in mass flow is due to the formation of normal/bow shock
structure ahead of the cowl to allow mass spillage, since the flow is not self-starting
at this Mach number.
The distortion index and exit Mach number are also compared, but as the throat
length used for the on-design is smaller than required, the results are not discussed
here.
5.3 Effect of Contraction Ratio
To study the effect of contraction ratio on performance, two different self-starting

contraction ratios are considered for the design for each ramp configurations. But
here, only the results of two-ramp design have been presented. The two different
contraction ratios are (i) 1.316, which is the self-starting contraction ratio for a
Fig. 16 Variation of MFR

with internal contraction
ratio for two-ramp design
freestream Mach number of 2.9. (ii) a contraction ratio of 1.111, which is the self-
starting contraction ratio at off-design freestream Mach number of 2 for the current
two-ramp design. These ratios are obtained from the experimental contraction ratio
curve provided by Sun et al. [11].
For this two-ramp design, the contraction of 1.316 leads to a throat height of
15.2 mm. And for the contraction of 1.11, the throat height is about 18.1 mm. The
effect of CR on the critical MFR at different operating Mach number is shown in
Fig. 16. It clearly shows that with reduction in internal contraction ratio, the MFR is
improved at low off-design Mach numbers. The cause for this improvement can be
seen by comparing the shock structure ahead of the cowl as shown in Fig. 17. This
clearly indicates that for the low contraction ratio case due to smaller flow separation
zone, the shock is closer to the cowl, leading to low mass spillage; hence, the mass
capture is improved. For the present two-ramp design at M2 freestream condition,
the Mach number ahead of cowl corresponds to a contraction ratio of 1.111 as self-
starting limit as per the practical CR limit curve described by Sun et al. [11]. However,
the numerical simulation still shows unstarting condition due to presence of large
boundary-layer ahead of the throat. From this analysis, it can be concluded that the
low CR design is a better choice to improve low off-design performance.
Fig. 17 Contours showing the M2 flow field for two-ramp design with different contraction ratio
a CR 1.316 b CR 1.111
6 Conclusion
The present investigation of the effect of ramp structure and internal contraction is
motivated by the need for design methods for an intake designed at M 2.9 and to
operate in the Mach range of 2–4. Therefore, theoretical design and CFD analysis
of a rectangular intake have been carried out to study these effects on the overall
performance. The theoretical design makes use of various oblique and normal shock
relations and available experimental correlations to fix the geometric dimensions
for different parts. Two different ramp designs, namely two-ramp and three-ramp
designs, have been considered to find their effect on the total pressure recovery. To
obtain the viscous flow field and the actual performance parameters, compressible
flow CFD simulation has been carried out on these geometries. The critical TPR
obtained from the 2D simulation for the two-ramp design is nearly 0.645 and that of
three-ramp design is 0.67. Even though the on-design performance of the three-ramp
intake is higher, the TPR at the off-design Mach numbers is found to be smaller than
that of two-ramp design. Similarly, at the low off-design Mach numbers, the MFR
for the three-ramp design is smaller than that of two-ramp design. The study of the
effect of contraction ratio indicates that, with reduction in contraction ratio the low
off-design MFR increases.
References
1. Kim S, Natan B (2014) Inlet geometry and equivalence ratio effects on combustion in a ducted
rocket. J Propul Power 31(2):619–631
2. Saha S, Chakraborty D (2012) Hypersonic intake starting characteristics—a CFD validation
study. Def Sci J 62(3):147–152
3. Benson TJ, Trefny CJ, Walker JF (1997) Interactive design tool for turbine based combine
cycle engines. AIAA 97:3160
4. Ran H, Mavris D (2005) Preliminary design of a 2D supersonic inlet to maximize total pressure
recovery. In: AIAA 5th aviation, technology, integration and operations conference, Virginia
5. Bridges DH (2010) Inlet stagnation pressure loss estimate for a simplified thrust model. J
Aircraft 47(5):1784
6. Woollett RR, Connors JF (1956) Zero-angle-of-attack performance of two-dimensional inlets
near Mach number 3
7. Dutton RA, Goldsmith EL (1963) Some characteristics of rectangular multi-shock and isotropic
external compression intakes at a Mach number of 2.9. Ministry of aviation report, C.P. No.
330
8. Kamman JH (1975) Effects of selected design variables on three ramp, external compression
inlet performance. Natl Aeronaut Sp Admin
9. Sakata K, Yanagi R, Murakami A, Shindo S, Honami S, Shizawa T, Omi J (1993) An experi-
mental study of supersonic air-intake with 5-shock system at Mach 3. In AIAA, SAE, ASME,
and ASEE, Joint propulsion conference and exhibit, 29 th, Monterey, CA
10. Van Wie D, Kwok F, Walsh R (1996) Starting characteristics of supersonic inlets. In: 32nd
joint propulsion conference and exhibit, p. 2914
11. Sun B, Zhang KY (2010) Empirical equation for self-starting limit of supersonic inlets. J Propul
Power 26(4):874–875
12. Ferri A, Nucci LM (1951) The origin of aerodynamic instability of supersonic inlets at
subcritical conditions
13. Dailey CL (1955) Supersonic diffuser instability. J Aeronaut Sci 22(11):733–749
14. Herrmann D, Siebe F, Gülhan A (2013) Pressure fluctuations (buzzing) and inlet performance
of an airbreathing missile. J Propul Power 29(4):839–848
15. Mahoney JJ (1990) Inlet for supersonic missiles, AIAA education series. American Institute
of Aeronautics and Astronautics, Washington, DC, p 1
16. Seddon J, Goldsmith E (1999) Intake aerodynamics. American Institute of Aeronautics and
Astronautics, Inc
17. Equations T (1953) Charts for compressible flow. NACA Report, p 1135
18. Matsuo K, Miyazato Y, Kim HD (1999) Shock train and pseudo-shock phenomena in internal
gas flows. Prog Aerosp Sci 35(1):33–100
19. ICEM (2012) CFD: ver. 14.0. ANSYS Inc., Southpointe, p 275
20. Ansys CFX (2006) Solver theory guide. Ansys CFX Release 11:1996–2006
21. Emami S, Trexler CA, Auslender AH, Weidner JP (1995) Experimental investigation of inlet-
combustor isolators for a dual-mode scramjet at a Mach number of 4, NASA T.P. 3502
Sensitivity of Altitude Variation on
Aerodynamics of a Typical Launch
Vehicle During Hot Separation
Jiju R. Justus, Sanjoy Kumar Saha, and Pankaj Priyadarshi
Abstract Staging of a launch vehicle is essential for achieving the required orbital
velocity and altitude. For the launch vehicle studied here, the boosters are detached
from the ongoing stage (core) by means of jettison motors. The jet pressure ratio is
high and the plume from the jettisoning motors expand and propagate upto the heat-
shield region. Thereby aerodynamics of the complete vehicle alters. The present study
is also focused on the effect of altitude variation on the aerodynamics of the launch
vehicle during hot separation. Two different trajectories (or altitudes) are studied
and the effect on the aerodynamic coefficients is quantified. The paper presents in
detail about the simulated conditions, the variation in aerodynamic coefficients and
the complex flow-field involved.
Keywords Launch vehicle · Jettisoning · Jet pressure ratio · Altitude
1 Introduction
The separation of the ongoing stage from the spent stage is essential for achieving the
required orbital velocity and altitude. In the present launch vehicle, two boosters are
present on each side of the central core stage (Fig. 1). The boosters are separated from
the ongoing core stage by means of firing jettisoning motors. There are six jettison
motors on each booster, three in the nose cone region and three in the base shroud
region of the booster. The jettisoning motor’s nozzle exit plane is flush with the
booster external surface and nozzles are facing towards the core. During separation,
which occur at high altitudes, the jettison motors are fired. As the jet pressure ratio
of boosters and core increases with increase in altitude, there is interaction of jets.
Due to the interaction of jets [1, 2], the base region experiences higher pressure and
thermal loads [3]. Now at the instant of separation, the jettisoning motors are also
J. R. Justus (B) · S. K. Saha · P. Priyadarshi

Aeronautics Entity, Vikram Sarabhai Space Centre, ISRO, Thiruvananthapuram, Kerala, India

168 J. R. Justus et al.
Fig. 1 Typical launch

vehicle along with Jettison
motors locations
fired into the existing complex flow-field. In the present study we investigate the
effect of jet and its interaction with free-stream on the aerodynamics of the ongoing
stage.
2 Numerical Simulations
Numerical simulations are carried out using PARAS-3D [4] to simulate the flow
over the vehicle. PARAS-3D solves the Reynolds Averaged Navier-Stokes (RANS)
equations. The simulations are carried out using k-ε turbulence model in association
with modified wall function. The k-ε turbulence model is seen to predict the follow
better for flow involving jet interactions. The time stepping is done for each cell based
on local CFL criteria and fluxes at the interface of the cell are computed by means
of an approximate Riemann solver. The explicit scheme is second order accurate in
space and TVD type, which is achieved by means of min-mod limiter. As the solution
proceeds, gradient based grid refinement is done to capture the flow features with
better resolution. The simulations are carried out for two different cases at different
altitudes, higher altitude (70.25 km) and lower altitude (63.81 km) to study the effect
of altitude (Table 1).
Sensitivity of Altitude Variation on Aerodynamics of a Typical Launch … 169
Table 1 Free-stream conditions used in numerical simulations

Case 1 Case 2
Altitude (km) 70.25 63.81
Pressure (Pa) 5.45 13.01
Mach number 6.0 6.0
The simulated jet properties are given below:

Specific heat ratio (Cp/Cv) = 1.163
Molecular Weight = 26.56 kg/ kmol
Jet exit Mach number = 3.4
Angle of Attack (α) = 0.0◦
3 Domain and Grid Independence
The numerical simulations were carried out on a domain as shown in Fig. 2. We have
taken 12.5 D in the longitudinal direction ahead of vehicle and 37.5D downstream.
Similarly, in the lateral and transverse direction ±12.5D and ±10.0D is taken respec-
tively. The free-stream condition is supersonic, hence the domain is limited to 12.5 D
upstream. In order to capture the jets from boosters and core stage, the downstream
domain is taken as 37.5 D. Initial grid distribution was close to 15 million cells and
after four refinements the cell count was close to 50 million cells. The gradient based
flow refinement is carried out to capture the flow physics accurately with higher
resolution.
The booster separation occurs at high altitudes where the jettisoning motor jets are
highly under expanded. As the jet comes out of the nozzle, it expands to match with
the free stream pressure. The top jettison motor jets expand till the heat-shield (Fig.
3) where it interacts with the free-stream to form a double shock system. Whereas,
the bottom jettison motor jets expand to such an extent that it reaches upto the
strapon-core aft-attachment region. The core jet boundary also gets compressed due
the jettisoning jets from the bottom jettisoning motors located on the boosters. Due to
the interaction of the jettisoning motor jets with the core stage, the flow-field over the
ongoing stage (core) also changes. The extent to which the jet reaches the heat shield
region is shown using the jet mass fraction palette (Fig. 4). This, in turn, alters the
pressure field and hence the aerodynamic forces and moments acting on the ongoing
stage.
Fig. 2 Domain and initial grid distribution
Fig. 3 Pressure palette in presence of Jettison jet along the core plane
Fig. 4 Jet mass fraction palette in presence of Jettison jet along the core plane
4.1 Forces and Moments Acting on the Ongoing Stage
When there is no jet impingement, the forces on the core is due to free-stream flow
and the interference between the core and the boosters. But due to the high dynamic
pressure of the jets, huge forces and moments are experienced by core stage. The
forces and moments acting on the core with and without jettison jets are given in
Table 2. It is clear that due to the jet impingement all the forces and moments alter
drastically. A significant reduction in axial force is observed. This large reduction in
axial force mainly comes from the base region of core, where jettison jet impinges
directly on the core base. Small reduction in axial force also observed at the heat-
shield boat-tail region. The augmentation is yawing moment and side forces are
smaller than the pitching moment and normal force as the jets are symmetric across
pitch plane.
Table 2 Effect of Jettison motor jet interaction on aerodynamics of core (ongoing vehicle)
Force coefficients Moment coefficients
Axial Side Normal Rolling Pitching Yawing
Jettison motors off 0.650 −0.056 0.002 0.001 0.003 0.106
Jettison motors on −0.155 0.294 3.528 0.087 4.157 −1.190
Fig. 5 Rolling moment generated at various stations
The rolling moment is induced when there is asymmetrical pressure distribution

over the vehicle. The asymmetrical protuberances on the vehicle at various stations
(3.75D, 5.0D and 8.0D) lead to the generation of a rolling moment as shown in
Fig. 5.
4.2 Effect of Variation in Altitude
If separation occurs at relatively higher altitude ( 70 km), free-stream pressure reduces

significantly (50% of earlier value, Table 1). For a given jet exit condition, jet pluming
is directly governed by the jet pressure ratio (i.e; the ratio between jet exit pressure to
the free-stream pressure). To assess the effect of this change, flow simulations were
carried out. The jettisoning jet pressure ratio being almost double, the jet pluming is
also more as compared lower altitude condition. As a result, the jettison jet engulfs
more area of the vehicle. It is observed that even some part of the heat-shield also
gets engulfed by the jet at higher altitude (Fig. 6). So, forces and moments acting on
the ongoing stage also get affected. The variation of cumulative rolling and pitching
moment is shown in Figs. 7 and 8 respectively. The rolling moment is contributed
Fig. 6 Body pressure palette in presence of Jettison jet
Fig. 7 Effect of altitude on rolling moment coefficient
mainly from the forebody portion of the vehicle. Effect of increased altitude on the
aerodynamics of the vehicle is given in Table 3. It is observed that all the aerodynamic
coefficients increases by 70–100 %, except axial force which becomes nearly six
times the original value.
Fig. 8 Effect of altitude on pitching moment coefficient
Table 3 Effect of altitude on ongoing vehicle aerodynamics

Force coefficients Moment coefficients
Axial Side Normal Rolling Pitching Yawing
Altitude 63.81 km −0.155 −0.294 −3.528 0.087 −4.157 1.190
Altitude 70 km −0.937 −0.515 −7.000 0.150 −7.300 2.135
% change 505 75 98 72 76 79
5 Conclusion
In order to study the effect of jettisoning motor jets impingement on the ongoing
stage, numerical simulations were carried out. It is observed that due to jettison
jet impingement, all the force and moment coefficient change significantly. With
the increase of altitude at the time of booster separation, the jet impingement effect
increases significantly. Hence, proper margins should be accounted for while design-
ing for structural loads during strap-on separation and thermal protection system.
Acknowledgements Authors would like to thank Dr. V. Ashok, Deputy Director, and Dr. M. M.
Patil, Group Director, ADSG, Aeronautics Entity for the valuable suggestions and the support
rendered. The authors would also like to thank the reviewers for their positive comments and
constructive criticisms.
References
1. Desikan SLN, Murugan B, Srinivasan K, Sajan S (2015) Twin jet interaction and reverse flow.
J Spacecraft Rockets 52(6)
2. Saha SK, Kessop NK, Das D (2011) Numerical investigation of twin jet interaction for a typical
launch vehicle. In: 25th national convention of aerospace engineers (NCAE)
3. Justus JR, Saha SK, Anoop P, Sundar B (2017) Estimation of aero-thermal environment at the
base region of a launch vehicle. In: Proceedings of the 24th national and 2nd international
ISHMT-ASTFE heat and mass transfer conference (IHMTC-2017)
4. Balu R, Babu TC, Ashok V, Kumar P, Das D (2002) Development of low cost parallel computing
platform for CFD applications. In: 3rd annual symposium of CFD division of Ae.S.I
Normal Shock Dynamics in Internal
Supersonic Flows
S. Vaisakh and T. M. Muruganandam
Abstract The normal shock train oscillations in a rectangular duct were studied
experimentally. The studies have been classified into three groups. The studies in
group I showed normal shock train oscillations and its structures at different duct
locations. Group II studied about the shock oscillation when it stands on a hybrid
micro vortex generator, while group III studied moving normal shock train. The
shock structures have been visualized using high-speed schlieren imaging, and its
oscillations have been quantified with time resolved image processing. It has been
showed that the tertiary shock can oscillate at larger amplitude than secondary shock.
Moreover, the injection assisted MVGs reduces shock oscillation even at its off-
design condition.
Keywords Supersonic duct flow · Unsteady normal shock · Shock-boundary layer

interaction · Micro vortex generators
1 Introduction
In supersonic intakes, the flow is decelerated through oblique and normal shock waves
[1]. These shocks interact with the intake boundary layer and can cause thickening or
separation of boundary layer in the intake duct [2]. In mixed compression intakes, the
normal shock stands inside the intake. The boundary layer approaching this normal
shock already subjected to multiple oblique shock interactions. This increases the
boundary layer shape factor (ratio of displacement thickness to momentum thick-
ness), and hence, it becomes more vulnerable to separation. Typical Mach number
ahead of the normal shock in a mixed compression intake varies in the range of
1.3–2 [2]. A normal shock-boundary layer interaction (NSBLI) of this strength can
trigger flow separation and oscillation of the normal shock in the subsonic diffuser.
S. Vaisakh (B) · T. M. Muruganandam

Department of Aerospace Engineering, Indian Institute of Technology Madras, Chennai 600036,
India

178 S. Vaisakh and T. M. Muruganandam
In NSBLI, the shock gets bifurcated and forms a "lambda foot" structure at the inter-
action region. As Mach number increases, the flow downstream of the primary shock
accelerates to supersonic region and terminated by a secondary shock. Such a shock
system with multiple shocks formed downstream of the primary shock is termed
as a normal shock train [3–5]. The shock-induced separation leads to spatial and
temporal unsteadiness in the flow. As the normal shock is standing in the beginning
of the subsonic diffuser, the separation imparted by shock grows easily. This results
in a highly distorted flow at the exit of intake, which is not desirable.
Micro vortex generators (MVGs) are passive devices used to control shock-
induced separation [6, 7]. MVGs are specially contoured profiles with a height lower
than boundary layer thickness. Micro ramps and vanes are the most commonly used
passive devices in controlling shock-induced separation. These devices sit inside the
boundary layer and send streamwise vortices toward the shock interaction region.
The streamwise vortices from the MVGs impel high momentum fluid to the near
wall region and thus energize the boundary layer. There also exist a hybrid device,
which consists of MVG and injection [8].
Most of the previous studies were concentrated on shock-boundary layer inter-
actions (SBLIs) or shock-induced separation control. At off-design conditions, the
normal shock in a mixed compression intake moves in the intake duct at different
speeds. If this normal shock movement reaches the converging part, then the shock
will be expelled and led to intake unstart. Based on the back pressure, the normal
shock can stand at different locations in the diffuser and hence, on top of the control-
ling devices also. So this paper studies about normal shock train and its nature at
off-design time. The off-design study consists of (1) its oscillation nature when its
stands over a hybrid control device and (2) the shock structure comparison between
stationary and moving shock train in a duct.
2 Experimental Details
Figure 1 shows the experimental facility and details of duct used to study the NSBLI.
One wall of the duct has a divergence of 4°. This resembles the situation in a subsonic
diffuser of a mixed compression intake. Stagnation temperature of the wind tunnel
is 302 K (±3 K). Stagnation pressure ("po ") is measured using a GE UNIK 5000
series piezoresistive-type pressure transducer with a sampling frequency of 4 kHz.
Flow in the duct chokes at the beginning of the diverging wall and the normal shock
formed stands downstream of it. The duct exits to atmosphere, and the location of
normal shock in the duct depends on the settling chamber pressure upstream to the
duct (see Fig. 1).
The studies conducted in this article can be arranged in three groups, namely I,
II, and III. Group I consists of dynamics of normal shock due to boundary layer
interaction at four locations in the duct. Group II consists of the dynamics of normal
shock, when it stands over an MVG. Here, the MVG used is a hybrid device called
micro vortex generator with injection (MVGI). The trailing edge height of the MVG is
Normal Shock Dynamics in Internal Supersonic Flows 179
Fig. 1 a Experimental facility, b Test section/duct details, and c Details of MVG with injection
(MVGI). All dimensions are in mm
1 mm, and ramp dimensions (see Fig. 1c) are similar to the one studied previously [8].
The injection through the trailing edge of MVG is driven by the pressure difference
existing between inside duct and atmosphere. When supersonic flow establishes in
the duct, the pressure at the trailing edge location of MVG is lower than outside
atmospheric pressure. This establishes a flow through the trailing edge of MVG
(MVGI) in the streamwise direction of main duct. Lastly, group C shows the dynamics
of a moving normal shock train and its interaction with the boundary layer in a duct.
The bottom block shown in Fig. 1 is changed to perform above mentioned studies.
For group I and group III studies, the bottom block used was plane, and for group
II studies, the bottom block used had MVGIs. There were four MVGIs in a row
arranged in the bottom wall center in the spanwise direction of duct. The leading
edges of these MVGIs were located 70 mm from the chocking point. The Mach
number at MVGI location is 1.35. The shock oscillations in all groups are visualized
using high-speed schlieren imaging. IDT-N4 high-speed camera is used for schlieren
imaging at 7000 fps with an exposure of 10 µs.
3.1 Normal Shock Dynamics in Duct
Figure 2 shows the schlieren images of a normal shock interacting with boundary
layer for constant settling chamber pressure. This interaction led to bifurcation of
the normal shock, flow separation, and formation of secondary and tertiary shocks
[3]. The normal shock (primary shock) is kept at four locations (A, B, C, and D)
by adjusting the settling chamber pressure. The Mach number ahead of the shock
increases from location A (M = 1.35) to D (M = 1.75) as the top wall is diverging.
It is evident from these schlieren images that flow downstream of the shock system
is more disturbed in C and D locations than A and B. This is due to the presence
of multiple shocks at locations C and D. The "lambda foot" size of primary shock
increases, as the shock location changes from A to D. This increase in "lambda foot"
size indicates larger separation due to higher adverse pressure gradient resulting from
higher Mach number. The shock system formed due to shock-induced separation
oscillates across a mean location. Figure 3 shows the standard deviation of primary,
secondary, and tertiary shock oscillations. It is observed that primary shock oscillation
increases when it is kept at a higher Mach number location. The secondary shock
oscillations at locations A and B are of same order with primary shock oscillation. At
location C, the primary, secondary, and tertiary shock oscillations are in same orders
of magnitude. At location D, the standard deviation of secondary shock oscillation is
lowered by a large value compared to its primary shock oscillation value. In addition,
the tertiary shock oscillation standard deviation increases beyond secondary shock
oscillation, but lower than its primary shock oscillation. The presence of a fourth
shock in location D is the main difference with location C. This can be the reason for
sudden variation in standard deviation values among primary, secondary, and tertiary
shocks at location D.
Fig. 2 Schlieren images of normal shock train at locations: a A, b B, c C and d D

Fig. 3 Standard deviation

plot of normal shock train
oscillations for primary
shock locations at A, B, C,
and D
3.2 Normal Shock on MVGI
MVGI is used to control shock-induced separation [8]. These devices are kept ahead
of the SBLI region so that it energizes the boundary layer (due to the generation of
streamwise vortices in the boundary layer) and hence lead to reduction in shock-
induced separation [7]. This section studies the normal shock oscillation during an
off-design location, i.e., when the shock stands over the control device (MVGI).
Figure 4 shows the schlieren image of a normal shock standing over an MVGI. The
presence of MVGs generates additional waves such as leading edge shock (LES)
and trailing edge shock (TES) from the leading and trailing edges of the MVG. The
slant edges of the MVG generate expansion waves. However, in the present case,
LES merges with leading edge of "lambda foot shock." TES is absent as the flow is
subsonic there. Three cases are studied in this section, namely no-control case, MVGI
case, and no-injection MVGI case. In no-control case, the normal shock is located
at 74 mm from the throat, and no control device is present in this case. In MVGI
Fig. 4 Schlieren image of

normal shock on MVGI
Fig. 5 Standard deviation

plot
case, the normal shock stands over the MVGI as shown in Fig. 4. In no-injection
MVGI case, there is no injection through the trailing edge of MVGI, but the normal
shock stands over the MVG. The last case is to understand the influence of pressure
fluctuation downstream of the normal shock and corresponding fluctuating injection
effect. Figure 5 shows the standard deviation values of normal shock oscillations
for above three cases. The standard deviation values of normal shock oscillation for
MVGI case and no-injection MVGI case are of same order, while a higher value
for no-control case. This shows that at off-design location (when the shock stands
over the micro vortex generator), also, the shock oscillation is reduced compared
to no-control case. Moreover, the fluctuating injection for an MVGI device due to
shock standing over the device is not increasing shock oscillation.
3.3 Moving Normal Shock Train
The normal shock is allowed to accelerate through the rectangular duct (shown in
Fig. 1b) by suddenly decreasing the settling chamber pressure. The settling chamber
pressure variation is shown in Fig. 6a. Initially, settling chamber pressure is set to
a high value so that the shock stands at a downstream location. The sudden closure
of the ball valve upstream of the settling chamber causes the shock system to travel
fast through the duct. The normal shock moves with an average velocity of about
830 mm/s at location D. Figure 6a shows the primary shock location versus time
plot at location D. It is observed that the shock train oscillates during its sudden
travel through the duct. These oscillation amplitudes were slightly lower compared
to corresponding stationary case. Figure 7 shows the schlieren images of normal
shock train at location D for stationary and moving cases. It is evident from these
Fig. 6 a Settling chamber pressure and b shock location variation for moving normal shock train
Fig. 7 Schlieren images of:

a stationary shock train and
b moving shock train
images that there is no change in shock train structure and distance between them
for stationary and moving cases.
4 Conclusions
The oscillations of normal shock in a duct at different situations are studied. These
situations are classified into three groups. The first group studied the oscillation
amplitude of each shock in a shock train at different Mach numbers/duct locations.
As the Mach number/shock location increases, the multiple shocks are formed down-
stream of the primary shock. Moreover, the “lambda foot” size and standard deviation
of primary shock increases with increase in Mach number/shock location. From the
studies performed in this paper, it can be concluded that for a normal shock train; the
standard deviation of secondary shock oscillation is equal or lower than its primary
and tertiary counter parts. The studies in the second group concentrated on normal
shock oscillations, when it stands over an MVG. The MVG used in this study was
MVGI. It can be concluded from group II studies that injection assisted MVGs
reduces shock oscillation even at its off-design condition. The third group studied
the structure of a normal shock train when it moves at a fast rate (830 mm/s) through a
duct. It is observed that there is no change in shock structure compared to a stationary
normal shock train for a given location at this rate. It is also observed that the moving
shock at this rate also oscillates during its travel in the duct.
References
1. Seddon J, Goldsmith EL (1999) Intake aerodynamics, 2nd edn. Blackwell Science, Boston
2. Babinsky H, Harvey JK (2011) Shock wave boundary-layer interaction. Cambridge University
Press, Cambridge
3. Matsuo K, Miyazato Y, Kim HD (1999) Shock train and pseudo-shock phenomena in internal
gas flows. Prog Aerosp Sci 35:33–100
4. Carroll BF, Dutton JC (1992a) Multiple normal shock wave/turbulent boundary-layer interac-
tions. J Propul Power 8(2):441–448
5. Carroll BF, Dutton JC (1992b) Turbulence phenomena in a multiple normal shock wave/turbulent
boundary-layer interaction. AIAA J 30(1):43–48
6. Lin JC (2002) Review of research on low-profile vortex generators to control boundary-layer
separation. Prog Aerosp Sci 38:389–420
7. Lu FK, Li Q, Liu C (2012) Microvortex generators in high-speed flow. Prog Aerosp Sci 53:30–45
8. Vaisakh S, Muruganandam TM (2019) Influence of multi-wall separation control on normal-
shock-induced separation in supersonic duct flows. Proc IMechE Part G: J Aerosp Eng
233(9):3184–3192
Hinge Moment Characterization of All
Movable Control Surface
Manoj Kumar, G. Kadam Sunil, V. Shanmugam, and G. Balu
Abstract A special side wall mounted test set up rig was designed and realized
for hinge moment characterization of an all movable control surface. Experiments
were conducted on a 1:2 scale down model of the control surface. In the experiment,
two different configurations (A & B) were tested. In configuration A, the control
surface is mounted on its actuator cover whereas in the configuration B, the control
surface without actuator cover is tested. In case of configuration B, a boundary
layer bypass plate was used between tunnel test section side wall and the control
surface. The test Mach numbers are 2.0, 2.5, 3.0, and 3.5. The control surface was
deflected from −10° to 25° during the test using electrical stepper motor. Integral
strain gauge force balance has been used to measure the forces and moment on
the control surface. Test data have been analyzed. Normal force co-efficient, hinge
moment co-efficient, and variation of center of pressure have been presented in
this paper. Hinge moment is significantly influenced by the presence of actuator
cover, and it is found higher compared to configuration B. Center of pressure is
ahead of the hinge line toward the leading edge of the control surface for both the
configurations. The hinge moment is unstable for both the configurations. However,
it is more unstable in case of configuration A.
Keywords Control effectiveness · Center of pressure · Hinge moment · Unstable

hinge moment
M. Kumar (B) · G. K. Sunil · V. Shanmugam · G. Balu

Defence Research and Development Organization, Hyderabad, India
V. Shanmugam
G. Balu

186 M. Kumar et al.
Abbreviations
CHM Hinge moment co-efficient

CN Normal force co-efficient
Cr Root chord
H.L. Hinge Line
M Mach number
Xcp Location of center of pressure
δ Control surface deflection in degree
1 Introduction
Aerodynamic control is one of the means of controlling the vehicle during its flight.
In this type of control, movable control surfaces are used to generate the required
control forces and moments to trim the vehicle or to provide the required maneu-
verability. Control surfaces are rotated about its hinge lines using control actuators
to generate the required control forces and moments. In all movable control surface,
the entire control surface rotates about its hinge line. Many aerospace vehicles use
all movable control surfaces for trim and maneuver because higher control effective-
ness is obtained by utilizing all the potentially available planform area of the control
surfaces [1].
The selection of control actuator mainly depends on the amount of control force
(normal force) and hinge moment (moment of normal force about hinge line), which
are experienced by the control surface during the course of its flight. The control
surface, which provides the adequate control force but smaller hinge moment, is a
better choice from the actuator selection point of view. The hinge moment depends
on the control force and the location of center of pressure with respect to the hinge
line.
In the present experimental study, two types of configurations (A & B) have been
tested. In configuration A, the control surface is mounted on its actuator cover. In this
configuration, the control actuator is kept outside the vehicle body and is covered with
an aerodynamically shaped cover called actuator cover. This is because of limited
space available inside the vehicle body. The generic vehicle configuration is shown
in Fig. 1. Due to limitation in sharing the details of the configuration, the concept is
provided as a generic and is shown in Fig. 1. In the other configuration B, isolated
control surface without actuator cover has been tested. In this case, a boundary layer
bypass plate [2] is introduced between the control surface and the tunnel test section
side wall.
Normal force co-efficient, hinge moment co-efficient, and center of pressure vari-
ation have been presented in this paper. Test results have been compared for both the
test configurations A and B.
Hinge Moment Characterization of All Movable Control Surface 187
Fig. 1 Generic vehicle configuration
2 Experimental Details
2.1 Model Details
The airfoil shape of the control surface is modified double wedge. The details of
the isolated control surface model (scale, 1:2) are shown in Fig. 2. The control
surface is common for both the test configurations A and B. The model assembly
of the control surface with the actuator cover is shown in Fig. 3 for configuration
A. The actuator cover is designed and optimized by design team to suit the overall
mission requirement of the vehicle configuration. In this paper, design details of the
Fig. 2 Isolated control

surface
188 M. Kumar et al.
Fig. 3 Control surface with actuator cover, configuration A
actuator cover are not presented. However, the test results like the effect of actuator
cover on the normal force co-efficient, hinge moment co-efficient, and location of
center of pressure are presented. Figure 4 shows the assembly of control surface with
boundary layer bypass plate for configuration B. The gap between the control surface
and boundary layer bypass plate is same as that in between the control surface and
actuator cover.
2.2 Experimental Setup Details
A special side wall mounted test setup rig has been designed and realized for hinge
moment characterization of an all movable control surface. Tests were conducted in
0.3 m × 0.3 m supersonic wind tunnel at Aero Test Facilities, DRDL Hyderabad.
The side wall mounted test setup rig is shown in Fig. 5 for both the test configurations
(A & B). The test setup rig mainly consists of control surface, strain gauge balance,
electrical stepper motor, encoder, actuator cover (in case of configuration A), and
boundary layer bypass plate (in case of configuration B). The purpose of putting
boundary layer bypass plate is to isolate the control surface from the influence of
tunnel side wall boundary layer. The gap between the tunnel test section side wall
and boundary layer bypass plate is decided based on the boundary layer thickness on
the tunnel side wall near the location of trailing edge of the boundary layer bypass
plate. This gap is kept sufficiently higher than the boundary layer thickness so that
Fig. 4 Control surface with

boundary layer bypass plate,
configuration B
tunnel side wall boundary layer does not influence the control surface. Boundary
layer bypass plate size (diameter) is decided primarily based on model size (root
chord), and its thickness is decided based on starting load on the plate and ease
of putting the screw fasteners to fix the plate with the tunnel test section side wall
flange. The wedge angle of the boundary layer bypass plate is kept sufficiently low
so that wedge shock is always attached for all the test Mach numbers. The electrical
stepper motor of 30 N-m rated capacity is used to deflect the control surface during
the blow down. The control surface was deflected from −10° to 25° at the rate of
2°/second. Encoder is used to take angular positional feedback of the control surface.
The measurement of forces and moment on the control surface was carried out using
five-component integral type strain gauge force balance. The balance was calibrated
and has measurement accuracy of 0.3% of full scale.
Figure 6 shows the isometric view of the tunnel test section along with the contour
of convergent-divergent nozzle. Also, the model mounted in the tunnel test section
can be seen for both configurations (A & B).
Signal conditioning was carried out using NI-based SCXI platform. The test was
carried out using 3 PC’s separately for data acquisition, electrical stepper motor
control, and tunnel blowdown pressure control. PC-based single-axis controller was
used for the operation of the stepper motor. The data acquisition system comprises
of signal conditioners of NI make having configuration as PCI-6036E, SCXI-1001,
SCXI-1125, SCXI-1124, and SCXI-1520. The data acquisition and control software
190 M. Kumar et al.
Side wall mounted test set up rig, Configuration A
Side wall mounted test set up rig, Configuration B
Fig. 5 Side wall mounted test setup rigs

Fig. 6 Isometric view of tunnel test section with model
were developed in Laboratory VIEW 7.1. The data acquisition was carried out at
1000 samples/second for each channel.
3 Test Facility Details
The tests were conducted in 0.3 m × 0.3 m intermittent blow down type supersonic
wind tunnel at Aero Test Facilities division of DRDL, Hyderabad. The achievable
Mach numbers are 1.4, 1.7, 2.0, 2.5, 3.0, 3.5, and 4.0. Fixed geometry replaceable
nozzle blocks made in two pieces (top and bottom) are being used for different test
Mach numbers. The Reynolds number ranges from 2.7 × 107 to 7.6 × 107 /m for the
entire test envelope.
4 Test Results and Discussion
Test data has been analyzed and presented in this section. Figure 7 shows the normal
force co-efficient with respect to control surface deflection for configuration A at
different Mach numbers. Normal force co-efficient is higher for lower Mach number.
Figure 8 shows the variation of center of pressure (Xcp) with respect to control
surface deflection for configuration A. It is presented as percentage of root chord
(Xcp/Cr) from leading edge (L.E.) at root chord. It can be seen from the figure that
the center of pressure is much ahead of the hinge line toward the leading edge of the
control surface. Therefore, hinge moment is unstable, and this type hinge moment
experienced by the control surface is generally not preferred [3, 4] because a small
incremental deflection of control surface will add an incremental control force on the
control surface. This incremental control force will create a diverging (non-restoring)
hinge moment which will try to rotate (deflect) the control surface in direction of
incremental deflection. Variation of hinge moment co-efficient with control surface
deflection for different Mach numbers is shown in Fig. 9.
192 M. Kumar et al.
Fig. 7 Normal force

co-efficient with respect to
control surface deflection
Fig. 8 Center of pressure

variation with respect to
Fig. 9 Hinge moment

co-efficient with respect to
Comparison of normal force co-efficient for the configuration A and configuration

B is shown in Fig. 10 at Mach 2.5. Configuration A, which is with actuator cover,
is showing lower normal force co-efficient. Due to the presence of actuator cover,
control effectiveness is reducing. Figure 11 shows the variation of center of pressure
for configuration A and configuration B at Mach 2.5. It can be seen from the figure that
center of pressure is closer to the hinge line location for configuration B compared
to configuration A. Due to the presence of the actuator cover in configuration A,
the center of pressure is moving upstream (toward the leading edge) away from the
hinge line. Therefore, hinge moment is more unstable in case of configuration A.
Hinge moment co-efficient for the two configurations are shown in Fig. 12 at Mach
2.5. It can be seen from the figure that the hinge moment co-efficient is significantly
influenced by the actuator cover and is higher for configuration A. Normal force
co-efficient and hinge moment co-efficient both are influenced by the presence of
Fig. 10 Comparison of
normal force co-efficient for
configurations A and B
center of pressure variation
for configurations A and B
194 M. Kumar et al.
hinge moment co-efficient
for configuration A and B
actuator cover. However, the impact of the presence of actuator cover is very high
on the hinge moment co-efficient compared to normal force co-efficient.
Before conducting the actual experiment, offline (in the absence of air flow) work
back load check has been carried out to assess the accuracy of measurement chain.
Known values of load and moment have been applied on a panel (calibration plate)
shown in Fig. 13 and are compared with the output. From this exercise, the accuracy
of normal force channel and hinge moment channel is obtained which are 3 and
4.2%, respectively. Normal force co-efficient and hinge moment co-efficient from
two repeat runs at Mach 2.5 for configuration B are shown, respectively, in Figs. 14
and 15.
Fig. 13 Setup for offline

work back load check
Fig. 14 Normal force

co-efficients from repeat runs
Fig. 15 Hinge moment

co-efficient from repeat runs
5 Conclusion
Experiments have been conducted on an all movable control surface model for hinge
moment characterization. Two configurations (A & B) have been tested. In config-
uration A, the control surface is mounted on its actuator cover whereas in the other
configuration B, isolated control surface without actuator cover is tested. Normal
force co-efficient, hinge moment co-efficient, and center of pressure variation have
been compared for both the configurations as function of Mach number and control
surface deflection. Hinge moment co-efficient is higher for configuration A. For
both the configurations, hinge moment is unstable. However, for configuration A,
hinge moment is more unstable. Normal force co-efficient and hinge moment co-
efficient, both are influenced by the presence of actuator cover. However, the impact
196 M. Kumar et al.
of the presence of actuator cover is very high on the hinge moment co-efficient
compared to normal force co-efficient.
Acknowledgements The authors express thanks to Aero Test Facilities team, DRDL Hyderabad
for providing the necessary support during the progress of the test program. The authors would
also like to thank Directorate of Engineering, DRDL Hyderabad for providing model fabrication
support and other hardware which were required for conducting the experiments.
References
1. Nielsen JN, Goodwin FK (1982) Preliminary method for estimating of hinge moments of all
movable controls. Nielsen Engineering and Research, Inc
2. Czarnecki KR, Lord DR (1953) Hinge-moment characteristics for several tip controls on a 60°
sweptback delta wing at Mach number 1.61. NACA Washington, January 22, 1953
3. Fleeman EL, Schetz JA (2012) missile design and system engineering. AIAA-2012
4. Thompson RF, Moseley Jr. WC (1957) Effect of hinge line position on the oscillating hinge
moment and flutter characteristics of a flap-type control at transonic speed. NACA Washington,
June 10, 1957
3D Computational Studies of Flapping
Wing in Frontal Gusty Shear Flow
M. De Manabendra , J. S. Mathur, and S. Vengadesan
Abstract The present paper reports findings of the 3D computational studies of the
effect of frontal gusty shear flow on the force patterns of a flapping wing. A rigid
wing with semi-elliptical wing planform with asymmetric 1 DoF flapping kinematics
was exposed to a gusty shear flow. The shear gradient of the flow was varied from
−10 to +10 in steps of 5. Computation studies were carried out for Re = 150
which lies in the typical flight Reynolds number range of natural flyers like a fruit
fly and anthropogenic flyers like a Pico Aerial Vehicle. 3D, unsteady, laminar, and
incompressible Navier-Stokes equations were solved using finite volume formulation
based commercial code ANSYS Fluent. Wing kinematics and gusty inflow conditions
were modelled into the solver by User Defined Functions (UDFs). Wing motion was
simulated using the dynamic meshing technique. The effect due to variation in the
frontal inflow condition was studied quantitatively and qualitatively. Comparisons of
the instantaneous and gust cycle averaged forces and moment coefficients about the
wing root mid chord point and 3D phase space projections of the forces and moment
coefficients was carried out. Qualitative studies were carried out by comparing the
static pressure over both the surfaces of the wing and the vortex patterns near the
flapping wing using λ2 –criteria. It was observed from these studies that negative shear
gradient resulted in a rise in the vertical force and moment and a minor reduction
in the horizontal force. Positive shear gradient resulted in a minor rise in horizontal
force and a significant reduction in vertical force and moment.
Keywords 1 DoF asymmetric flapping · Gusty shear flow · Pico aerial vehicle ·
λ2 –criteria
M. De Manabendra (B) · J. S. Mathur

CSIR-National Aerospace Laboratories, Bengaluru, India
J. S. Mathur
S. Vengadesan
Indian Institute of Technology Madras, Chennai, India

198 M. De Manabendra et al.
Nomenclature
Aw Planform area of the wing, m2

c Wing chord length, mm
fg Gust frequency, Hz
fw Wing-flapping frequency, Hz
r Distance between wing’s flapping pivot point and the centroid of the wing,
m
t Instantaneous time, s
u Flow velocity, m/s
−
→ug Velocity of the moving mesh, m/s
C Hori Coefficient of horizontal force
C Mom Coefficient of moment
C Vert Coefficient of vertical force
C GA_Hori Gust cycle averaged coefficient of horizontal force
C GA_Mom Gust cycle averaged coefficient of moment
C GA_Vert Gust cycle averaged coefficient of vertical force
F Hori Horizontal force, N
F Vert Vertical force, N
M Piv Moment about wing’s flapping pivot point, N-m
Re Reynolds number
Sφ Source term
S Symmetric part of gradient velocity tensor
T Period of flapping, s
Tg Period of gust, s
Ug Gust amplitude, m/s
Uw Root mean square average flapping velocity at the tip of the wing, m/s
UG Gust velocity, m/s
U∞ Mean free-stream velocity, m/s
U Velocity field
V Gradient velocity tensor
-V Arbitrary control volume
Greek symbols
ø Any scalar quantity

ρ Fluid density, kg/m3
Diffusion coefficient
Asymmetric part of gradient velocity tensor
3D Computational Studies of Flapping Wing … 199
1 Introduction
Research on flapping wing Pico Aerial Vehicles (PAVs) has attracted the academia
and industry. Aerodynamics of various flight modes like ascending, forward, hovering
and descending flights, wing kinematics, different wing configuration like single and
tandem wings with different relative orientations, and different wing flexibility are
few niche areas that have been studied. Flapping wing with ground effects has also
been studied. Flapping wing PAV has good mobility and surreptitious characteristics.
But, as they have less inertia (max. weight of 0.5 g), low flight speeds (0–10 m/s)
and small size (max. chord of 50 mm), they are highly sensitive to gust. Hence, it is
important to understand their performance when they are subjected to gust.
Lian and Shyy [1] numerically investigated the impact of the gusty atmosphere
on static and flapping airfoil in 2D reference frame. Wan and Huang [2] employed
commercial code ANSYS Fluent to study the influence of gust on flapping wing in 2D
reference frame. Lian [3] employed an innovative algebraic moving grid approach for
computationally studying flapping airfoil under different gusty conditions. Viswanath
and Tafti [4] developed and used an in-house LES code to study the impact of
frontal gusts on a rigid wing in forward flight in 3D reference frame. Prater and
Lian [5] computationally studied single and tandem wing configurations subjected
to oscillatory inflow profile in 2D reference frame. Sarkar et al. [6] reported 2D
studies of wings undergoing pure and sinusoidal hovering and subjected to gust. Zhu
et al. [7] performed 2D computation of flexible foil subjected to gust. Jones and
Yamaleev [8] computationally studied the performance of flapping wings under the
influence of gusts from different directions.
It was observed that these efforts primarily aimed to understand the lift and thrust
individually. Symmetric flapping kinematics was predominantly considered for these
studies. Gust of temporally varying nature was only considered. In view of these
observations from the above cited researches, the present study focused on studying
flapping wings with Re = 150 in asymmetric 1 DoF flapping subjected to a frontal
spatiotemporal gust in 3D reference frame. A semi-elliptical wing planform was
considered for the present study.
2 Numerical Formulation, Wing Kinematics and Domain
3D, unsteady Navier-Stokes equations were solved using finite volume approach.
Flow was considered to be incompressible and laminar. Methodology reported by
Ferziger and Peric [9] was employed for solving the mass and momentum equations.
Generalized equation of conservation for a scalar quantity, ø in a generic control
volume, V with moving bodies is as shown in (1).

∫ ρ∅dV + ∫ ρ∅ u − −
→
d
u g · d A = ∫ Γ ∅ · d A + ∫ Sφ dV (1)
dt V ∂V ∂V V
is flow velocity vector, −

Here, ρ is fluid density in kg/m3 , u →
u g is mesh velocity,
represents the diffusion coefficient, and Sφ represents a source. Second-order upwind
spatial discretization and second-order implicit time discretization were employed.
Pressure-velocity coupling was carried out by the PISO algorithm reported by Issa
[10]. Convergence criteria for mass and momentum were set to O(10−6 ) in magnitude.
ANSYS Fluent was used for carrying out the simulations.
It needs to be mentioned here that the iterative time-advancement scheme for
carrying out the time accurate unsteady calculation is computationally expensive due
to the large numbers of outer iteration at each time step. ANSYS Fluent provides two
Non-Iterative Time-Advancement (NITA) schemes which significantly speed ups the
transient simulation. However, in doing so, the transient features of the flow are not
captured. Since the present study aimed at carrying out a time accurate unsteady
calculation to capture the transients during each time step, the NITA accelerating
device, offered by ANSYS Fluent, was turned off while carrying out the time accurate
unsteady simulations.
The asymmetric 1 DoF wing kinematics is shown in Fig. 1. Wing flapped about
X-axis. The flapping angle was 60°. The nature of flapping was asymmetric about
Z-axis. This asymmetry resulted in non-zero vertical force during a flapping cycle.
Similar kinematics was previously studied by Hong and Altman [11].
Domain architecture and boundary types are shown in Fig. 2. The wing was
enclosed in a sphere with non-deforming quadrilateral grids. This spherical domain
was externally surrounded by a cuboid with deforming quadrilateral grids. When the
wing flapped as per the prescribed kinematics, the grids inside the cubical domain
changed their shape and size and adjusted amongst each other. Face ABCD was
attributed with velocity inlet condition. Different gusty shear inflow conditions were
defined at this face. Face EFGH was attributed with pressure outlet condition. Faces
Fig. 1 Wing kinematics

(viewing from upsteam)
Fig. 2 Domain details
ABFE, BCGF, CGHD and HEAD were attributed with symmetric conditions so as
to reduce their effects on the solution.
A semi-elliptical shape was chosen as wing’s planform shape. Wing’s aspect ratio
was considered as four. The half-span length was considered as 0.155 m. Wing
thickness of 3e-4 m was considered. Flapping was modelled about X-axis. The flap-
ping frequency was considered as 250 Hz (typical for a fruit fly). Mean free-stream
velocity, air density and viscosity were considered in such a manner that Re was
limited to 150.
3 Methodology
Verifying and validating the present computational model were carried out by
comparing the predictions with the wind tunnel data reported by Fejtek and Nehera
[12]. A User Defined Function (UDF) was developed to virtually move the wing in
the prescribed 3D kinematics. Dynamic meshing was used to simulate the flapping
kinematics. In order to arrive at a smallest domain, grid resolution and time step for
the simulation, domain, grid and time-step independence studies were conducted.
Subsequently, the present model was validated against the results reported by Fejtek
and Nehera [12]. Details are included in the next section. Numerical predictions
were observed to be similar to the experimental data. Subsequently, asymmetric 1
DoF flapping kinematics was simulated. UDF subroutine was coded for modelling
the 1 DoF flapping of wing in 3D reference. A spatiotemporal gust was modelled.
Mathematical expression is shown below:

UG U∞ Ug fg Vgrad
= + sin 2π t ± y (2)
Uw Uw Uw fw Uw
UDF for specifying this gusty shear inflow was coded. Here, U G represented
inflow gust velocity, U ∞ represented mean free-stream velocity, U g represented gust
amplitude, V grad represented slope of velocity gradient along Y-axis, y represented
a location along Y-axis, f g represented gust frequency, and t represented time. For
present simulation, U g was considered as 10% of U w , and f g was considered as
10% of f w . Here, U w was the root mean square wingtip velocity, and f w was the
wing-beat frequency. For the present study, five values of V grad viz. ±10, ±5 and 0
were considered.
Findings were studied quantitatively and qualitatively. Quantitative comparisons
were made based on plots of instantaneous and gust cycle averaged forces and
moment coefficients about the mid-point of the root chord and 3D phase space
projections of the force and moment coefficients. Qualitative comparative studies
were based on the static pressure plots on the upper and lower surfaces of the wing
and flow visualization of the vortex patterns near the flapping wing using λ2 –criteria
at t/T g = 0.25 and 0.75. Here, T g is the period of gust.
Findings of the validation studies and various other studies are presented in this
section.
4.1 Validation Studies
Details of the simulation are reported in Table 1. Before commencing the numerical
studies, the present model was validated against results reported by Fejtek and Nehera
[12]. Findings are reported in Table 2.
Table 1 Details of
Parameter Particulars
simulation
Wing’s cross section NACA 8318
Wing size Chord—0.076 m; span—0.305 m
Flapping frequency 3.3 Hz
Flapping amplitude 45 deg
Inlet velocity 21.4 m/s
C Vert C Hori
non-dimensional coefficients
Fejtek and Nehera [12] 0.412 −0.0286
Present approach 0.419 −0.0271
The predicted values of flapping cycle averaged non-dimensional force coeffi-

cients, C Vert and C Hori , were found to be within 5% of the reference benchmark case.
Hence, this validation study established the robustness of the present approach to
simulate unsteady flow past a flapping wing.
4.2 Instantaneous Forces and Moment Coefficients
Time series of the instantaneous forces and moment coefficient during a gust cycle
is shown in Fig. 3(i)–(iii). The coefficients were calculated using the following
equations:
FVert FHori
CVert = , CHori = &
1
2
× ρ × A w × U∞
2 1
2
× ρ × A w × U∞
2
MPiv
CMom = (3)
1
2
× ρ × A w × U∞
2 ×r
Here, ρ represented the fluid density in kg/m3 ; Aw represented planform area in

2
m ; U ∞ represented free-stream velocity in m/s. F Vert and F Hori represented instanta-
neous forces generated along Y- and X-axis by the flapping wing in Newton, respec-
tively, and M Piv represented moment about wing’s flapping pivot point in N-m. r
represented the distance between the pivot point and the centroid of the wing plan-
form in m. The instantaneous forces and moment coefficients were plotted against
non-dimensionalized time. The reference time scale for the simulation was the period
of one gust cycle, T g .
It was seen from these comparative graphs that the shear gradient influenced the
force patterns during the entire gust cycle. The influence was seen to be more distinct
during pronation and supination for the vertical force and moment coefficients. The
effect of shear gradient was of similar nature on the vertical force and moment. The
shear gradient induced a change in the horizontal force pattern during the whole gust
cycle.
(i) Vertical force pattern (ii) Horizontal force pattern (iii) Moment pattern
Fig. 3 Variation of instantaneous coefficients

(i) Vertical force pattern (ii) Horizontal force pattern (iii) Moment pattern
Fig. 4 Variation of gust cycle averaged coefficients with respect to V grad
4.3 Gust Cycle Averaged Forces and Moment Coefficients
Comparative plots of the variations in gust cycle averaged forces and moment patterns
due to the change in the shear gradient are shown in Fig. 4(i)–(iii).
It was seen from the collective plots that there was a reduction in the magnitude
of the vertical force pattern and moment pattern when the velocity shear gradient
changed from −10 to +10. These reductions were quite significant in magnitude.
As opposite to this observation, the gust cycle averaged horizontal force coefficient
increased for the same change in the velocity shear gradient. However, this increase
was marginal in nature. The overall observation was that the changes in the forces
and moment patterns were non-linear in nature. As only two shear gradients on either
side of the zero shear inflow were considered in the present studies, further detailed
studies are desirable to understand these non-linear natured variations.
From the practical application point of view, these observations are important. For
example, during a phase of flight, when the anthropogenic flyer needs more lift or
vertical force, the inclination of stroke plane of flapping should be adjusted in such a
manner that a relative negative shear gradient is realized, and the anthropogenic flyer
can take advantage of the increased vertical force. During a phase of flight, when
more thrust is necessary, the inclination of stroke plane of the flapping should be
adjusted in such a manner that a relative positive shear gradient is realized, and the
flyer gets the benefit of the additional horizontal force. This would help in not only
fine-tuning the existing control algorithms of flapping wing flyers but also enhance
their endurance and in-flight stability. These techniques would help to improve the
vertical take-off capabilities of the anthropogenic flyers by appropriately modulating
the inclination of the stroke plane of flapping within a stable limit.
4.4 Phase Space Projection
3D phase space projections of force coefficients and moment coefficient were plotted.
Plots for V grad = −10, −5, 0, +5 and +10 are shown in Fig. 5a–e.
Fig. 5 3D phase space projections
Phase space diagrams primarily indicate the co-relationship between the forces
and moment coefficients. It was observed from the plots that the pattern of behaviour
of the system was similar in nature. The behaviour of the system was bounded and
periodic. Due to the difference in the shear gradient, the patterns exhibited a slight
difference. However, the overall system behaviour did not differ due to change in
the shear gradient component of the gusty inflow condition within the range of V grad
considered for the present study.
4.5 Surface Static Pressure Contours
Surface static pressure contours were plotted. Surface integration of the pressure over
the wing’s suction and pressure surface is carried out to estimate the forces along
the horizontal (X-axis) and vertical (Y-axis) directions. Hence, comparison of the
static pressure patterns over the flapping wing with varying V grad helped us in the
physical interpretation of the variations in the forces and moment patterns as reported
in Sect. 4.4. Static pressure patterns over both the surfaces of the wings at temporal
instances of t/T g = 0.25 and 0.75 are shown in Fig. 6.
It was observed from the surface pressure contour plots that the shear gradient
affected the surface pressure distribution over both the surfaces of the wing. Low-
pressure zones over the peripheral boundaries of the wing were observed. These zones
were because of the attached leading edge vortex and trailing edge vortex. As shear
Fig. 6 Static pressure pattern on suction and pressure surfaces of the wing
gradient increased, static pressure over both the surfaces of the wing was observed to
increase. However, the overall difference between the pressure over both the surfaces
of the wing was observed to have reduced. This endorsed the observations made in
the previous section wherein the vertical force decreased when the shear gradient
varied from −10 to +10. During both the instances of the gust cycle, i.e. t/T g = 0.25
and 0.75, the observations were similar in nature. As both the instances were during
the downstroke movement of the wing, the pressure contour patterns on the top and
bottom surfaces were identical in nature.
4.6 Vortex Structures Identified by λ2 –Criteria [13]
Numerical flow visualization of the vortex core in the proximity of flapping wing was
carried out by employing the λ2 –criteria approach. The λ2 -criterion is methodology to
identify a vortex core in a fluid field. This approach is a Galilean invariant, implying
that even if the reference frame of the flow field undergoes a transformation, the
output of the analysis remains same. At first, a gradient velocity tensor, V, is defined
as below:
⎡ ⎤
∂ x u x ∂ y u x ∂z u x
V = ∇ U = ⎣ ∂x u y ∂ y u y ∂z u y ⎦ (4)
∂ x u z ∂ y u z ∂z u z
Here, U is the velocity field. The symmetric and asymmetric components of the
gradient velocity tensor are then splitted as shown below:
V + VT V − VT
S= &= (5)
2 2
Here, T represents the transpose operator. Subsequently, three eigenvalues viz. λ1 ,
λ2 and λ3 of S 2 + Ω 2 are then computed. These three eigenvalues are arranged in such
a fashion that λ1 ≥ λ2 ≥ λ3 . A point in the velocity field, where two out of the three
eigenvalues are negative, i.e. λ2 < 0 is marked as a portion of vortex core. Hence,
this approach is known as λ2 —criteria or λ2 —method. The vortex structures at t/T g
= 0.25 and 0.75 were plotted and rendered with velocity magnitude colour coding
to differentiate them from each other and assess the spanwise transport of fluid flux.
The vortex structures are shown in Fig. 7.
It was observed from the plots that as the shear gradient of the gusty inflow
varied from −10 to +10, the vortex structures over the wing changed. While there
were substantial secondary shedded vortices observed over the wing for V grad = −
10, there were few secondary vortices observed for V grad = +10. These secondary
vortices affected the circulation over the wing. Also, these vortices interacted with
the primary vortex structures attached to the wing and affected their strength. It
was also observed that the magnitude of fluid flux in these vortices, represented by
the spanwise velocity magnitude colour code, changed as the shear gradient varied
from -10 to +10. The differences in the force coefficients and moment coefficients
reported in the previous section are attributed to these observations.
Fig. 7 Vortex structures identified by λ2 —criteria
5 Conclusion
Findings of a 3D computational study of flapping wing exposed to frontal gust with

shear inflow condition have been reported in the present paper. A canonical asym-
metric 1 DoF flapping kinematics was considered. Flapping wing was exposed to
a spectrum of spatiotemporal gusty inflow condition. Simulation Reynolds number
was limited to 150, which falls in a typical operating range of fruit fly and flap-
ping wing Pico Aerial Vehicles (PAVs). Effect of positive and negative shear gusty
inflow on forces and moment patterns was assessed qualitatively and quantitatively.
It was observed that the negative shear gradient element of the gusty frontal inflow,
V grad , induced a considerable increase in the gust cycle averaged vertical force and
moment and a marginal decrease in gust cycle averaged horizontal force. Posi-
tive shear gradient element of the gusty frontal inflow, V grad , induced a marginal
increase in gust cycle averaged horizontal force but a substantial decrease in gust
cycle averaged vertical force and moment.
From the application point of view, this know-how would help to improve the
existing control algorithms for kinematics of flapping wings and hence stabilize the
flight. By mere modulation of the stroke plane angle of flapping, the anthropogenic
flyers can take advantage of the additional lift (vertical) force and thrust (horizontal)
forces. This would, in turn, have bearing on enhancing the flight endurance, manoeu-
vrability and vertical take-off capability of the anthropogenic flyer. Subsequently, it
would lead to the realization of anthropogenic flyers like PAVs with better flight
stability and hence better performance.
References
1. Lian Y, Shyy (2007) Aerodynamics of low Reynolds number plunging airfoil under gusty
environment. In: 45th AIAA aerospace sciences meeting and exhibit, AIAA Paper 2007-70.
https://doi.org/10.2514/6.2008-523
2. Wan T, Huang C (2008) Numerical simulation of flapping wing aerodynamic performance
under gust wind conditions. In: 26th international congress of the aeronautical sciences, ICAS
2006-3.5 (ST 1)
3. Lian Y (2009) Numerical study of a flapping airfoil in gusty environments. In: 27th AIAA
applied aerodynamics conference, AIAA Paper 2009-3952. https://doi.org/10.2514/6.2009-
3952
4. Viswanath K, Tafti DK (2010) Effect of frontal gusts on forward flapping flight. AIAA J
48(9):2049–2062. https://doi.org/10.2514/1.J050263
5. Prater R, Lian Y (2012) Aerodynamic response of stationary and flapping wings in oscillatory
low Reynolds number flows. In: 50th AIAA aerospace science meeting including the new
horizons forum and aerospace exposition, AIAA Paper 2012-0418. https://doi.org/10.2514/6.
2012-418
6. Sarkar S, Chajjed S, Krishnan A (2013) Study of asymmetric hovering in flapping flight. Eur
J Mech B Fluids 37:72–89. https://doi.org/10.1016/j.euromechflu.2012.07.002
7. Zhu J, Jiang L, Zhao H, Tao B, Lei B (2015) Numerical study of a variable camber plunge
airfoil under wind gust condition. J Mech Sci Technol 29(11):4681–4690. https://doi.org/10.
1007/s12206-015-1015-z
8. Jones M, Yamaleev NK (2016) Effect of lateral, downward and frontal gusts on flapping-wing
performance. Comput Fluids 140:175–190. https://doi.org/10.1016/j.compfluid.2016.08.016
9. Ferziger JH, Peric M (2002) Computational methods for fluid dynamics, 3rd edn. Springer,
Berlin
10. Issa RI (1985) Solution of the implicitly discretized fluid flow equations by operator-splitting.
J Comput Phys 65:40–65. https://doi.org/10.1016/0021-9991(86)90099-9
11. Hong YS, Altman A (2008) Lift from spanwise flow in simple flapping wings. J Aircr 45:1206–
1216. https://doi.org/10.2514/1.34100
12. Fejtek I, Nehera J (1980) Experimental study of flapping wing lift and propulsion. Aeronaut J
84:28–33. https://doi.org/10.1017/S0001924000030463
13. Jeong J, Hussain F (1995) On the identification of a vortex. J Fluid Mech 285:69–94. https://
doi.org/10.1017/S0022112095000462
Investigation of Wind Tunnel Blockage
Effect on Liftoff Aerodynamics
of a Launch Vehicle Through
Open-Source CFD Solver SU2
Aaditya N. Chaphalkar, Amit Sachdeva, Vinod Kumar,

and Pankaj Priyadarshi
Abstract Aerodynamic coefficients for a launch vehicle are estimated using wind
tunnel tests as well as CFD simulations. Estimation of coefficients at liftoff is often
carried out using low speed wind tunnel tests. Presence of tunnel walls affect aero-
dynamic coefficients due to the blockage introduced by the model to the flow. The
effect of blockage is investigated in the current study through CFD simulations with
and without wind tunnel walls on a launch vehicle configuration in the presence of
launch pad and umbilical tower (UT). It is seen that the aerodynamic characteristics
are significantly influenced by the presence of wind tunnel walls. Even though the
model blockage is 4%, variation of 8.5% is seen in total force coefficients.
Keywords Blockage effect · Liftoff aerodynamics · CFD · SU2
Nomenclature
CS Coefficient of side force
CN Coefficient of normal force
CNt Coefficient of total force
φ Roll plane angle
UT Umbilical Tower
A. N. Chaphalkar (B) · A. Sachdeva · V. Kumar · P. Priyadarshi

Vikram Sarabhai Space Centre, Thiruvananthapuram, India
A. Sachdeva
V. Kumar
P. Priyadarshi

212 A. N. Chaphalkar et al.
1 Introduction
Aerodynamic coefficients are required for all phases of launch vehicle from liftoff till
the end of sensible atmosphere. Coefficients for the liftoff phase are often obtained
using low speed wind tunnel tests carried out with launch vehicle being stationed at
various heights with respect to the ground. But these coefficients need to be corrected
for blockage caused due to confinement of the flow within the wind tunnel and model
as well as its wake blocking part of the flow area [1]. In general, this estimation is
carried out by testing a smaller scale model thus reducing the blockage. But this
also does not eliminate the blockage fully. An attempt has been made to estimate
blockage through CFD by carrying out simulations with and without the wind tunnel
walls. In addition, CFD analysis can provide insights into the influence of walls on
the flow field around the launch vehicle. In this study, the blockage correction is
estimated for a particular configuration of launch vehicle standing on the launch
pad with Umbilical Tower (UT). This geometry is shown in Fig. 1 along with sign
convention. It is essential to estimate coefficients in this configuration accurately as
they are used to design base structures of the launch vehicle. Inaccuracies in these
coefficients can lead to non-optimal design of the structures.
Fig. 1 Launch vehicle on launch pad and associated sign convention

Investigation of Wind Tunnel Blockage Effect on Liftoff … 213
2 Methodology
2.1 Configuration Details
Wind tunnel tests are carried out on a scaled model. Hence simulations in this study
are carried out on a scaled model with and without wind tunnel walls. Blockage area
introduced by the vehicle along with the UT is 4% of the tunnel cross-sectional area.
2.2 Grid Generation
Grid required for these studies was generated using Pointwise software. Body-
conforming hybrid grid with prism layers was generated. First cell height of 6µm
was imposed to get y+ < 1. A feature of ‘baffle faces’ in Pointwise was used to
refine the volume grid locally. This local refinement was employed in front of the
vehicle to capture upstream influence of the body and also downstream of the vehicle
to capture large wake produced. All the protrusions on the body were captured in
this grid. Total cells in the grid are 78.2 million. A view of the surface grid is shown
in Fig. 2 and a cross-sectional view of the volume grid is shown in Fig. 3.
2.3 CFD Studies
Simulations were carried out in open-source software suite SU2 [2]. Typical wind
speeds on launch pad result in very low subsonic Mach numbers. In order to simulate
flow at these low Mach number, AUSM+-UP scheme implemented by Sachdeva et
al. in the SU2 was used [3]. All simulations were carried out using 2nd order scheme
with Venkatakrishnan limiter. SST turbulence model was used for RANS simulations.
All the coefficients converged within 2%. A typical convergence plot of CS is shown
in Fig. 4. Convergence was achieved by first simulating a first order accurate solution
followed by changing the order of accuracy to second order. CFL ramping was also
used to aid the convergence.
Simulations are carried out at various wind plane angles (φ) to simulate various wind
directions. In a wind tunnel test, the model is mounted on a turntable and rotated about
the vertical axis to simulate various wind plane angles. In the present CFD study, a
similar methodology is followed. The model is rotated about the vertical axis and
grid is generated for each wind plane angle. Streamlines at a typical horizontal cross-
Fig. 2 Surface grid generated using Pointwise
Fig. 3 Volume grid generated using Pointwise

Fig. 4 Convergence of CS
Fig. 5 Streamlines overlaid on Mach palette for wind plane angle of 135◦ at a typical horizontal
cross-section with zoomed view
section for wind plane angle of 135◦ are shown in Fig. 5. Streamlines for the case
without wind tunnel walls are shown in Fig. 6. Simulations without the wind tunnel
walls were carried out on a single grid by rotating the free-stream flow vector.
Large wake is formed behind the UT. Flow also passes through the gap between
the vehicle and the UT. These flow features cause the streamlines to bulge. Both
these phenomena are not altered by the presence of the walls. Streamline pattern
also does not show significant change visually. But flattening of the streamlines due
to the presence of the walls is seen as bulging of the streamlines is constrained by
the walls in this low subsonic flow. An estimate of the wake blockage area can be
made by considering the wake as part of the model being tested. This consideration
shows the blockage area as nearly 8%. Coefficients obtained with and without walls
at various roll angles are shown in Figs. 7 and 8.
Fig. 6 Streamlines overlaid on Mach palette for wind plane angle of 135◦ at a typical horizontal
cross-section (without walls)
Fig. 7 Effect of blockage on CN and CS for various wind plane angles
CNt is the coefficient of total force acting on the vehicle. It is seen that trend of the
coefficients does not change with and without walls as flow features do not change
by presence of the walls but, the value of the coefficient changes due to presence of
the walls. Change is more in the CS (25%) than CN (5%). This is due to flattening
of the streamlines in the presence of walls. Blockage effect is found to be similar at
all the wind plane angles as is seen from variation of CNt (8.5%) . As this blockage
effect is due to flattening of streamlines, most of this effect can be attributed to the
wake blockage. It is seen that the change in coefficient is lesser for φ = 90◦ . This is
attributed to lower wake blockage in this case as seen from Fig. 9.
Fig. 8 Effect of blockage on CNt for various wind plane angles
Fig. 9 Streamlines overlaid on pressure coefficient for wind plane angle of 90◦ at a typical horizontal
cross-section
It can be inferred that when the open area between the launch vehicle and the
UT is aligned with the flow, lateral extent of the wake is smaller resulting in smaller
blockage. When it is at 45◦ with the flow high velocity flow through the open area
causes the wake to increase laterally in turn increasing the blockage.
4 Conclusions
This study shows that the blockage does not affect trend of the aerodynamic coeffi-
cients but it affects their magnitudes significantly. This is mainly due to confinement
of streamlines in the presence of walls in the presence of large wake. Even though
blockage offered by the model is 4% of the flow area, wake blockage area is as high
as 8%. Coefficient in the direction of the flow is more affected. Change in the CS is
found to be as high as 25% whereas variation in CN is within 5%. Variation of CNt is
8.5%. As the coefficient in direction normal to the flow is also affected, it suggests
turning of the streamlines near of the vehicle. Hence blockage in this configuration
can be attributed mainly to the wake blockage. This blockage effect is found to be
weakly dependent on the roll plane angle as the lateral extent of wake structure varies
slightly with variation in the roll plane angle. Minimum effect is found to be at 90◦
roll plane angle.
Acknowledgements The authors would like to thank all engineers of Aerodynamic Design and
Synthesis Group along with Group Director, ADSG/AERO and Deputy Director (AERO), VSSC
for their support throughout this work.
References
1. Garner HC, Rogers EWE, Acum WEA, Maskell EC (1966) Subsonic wind Tunnell wall correc-
tions. AGARDograph 109
2. Palacios F, Colonno M, Aranake A, Campos A, Copeland S, Economon T, Lonkar A, Lukaczyk
T, Taylor T, Alonso J (2013) Stanford University unstructured (SU2): an open-source integrated
computational environment for multi-physics simulation and design. In: 51st AIAA aerospace
sciences meeting
3. Sachdeva A, Kumar V, Priyadarshi P (2017) CFD simulations over launch vehicle in the presence
of umbilical tower at liftoff. In: 19th annual CFD symposium, Bengaluru
4. Sahini D (2004) Wind tunnel blockage corrections: a computational study. Master of Science
thesis
Observation of Low-Frequency Shock
Oscillation Over a Forward-Facing Step
Jayaprakash N. Murugan and Raghuraman N. Govardhan
Abstract An experimental study has been conducted in a Mach 2.5 flow to under-
stand the shock-wave boundary layer interaction (SWBLI) over a forward-facing
step (FFS) of step height equals twice the oncoming boundary layer thickness. Parti-
cle Image Velocimetry (PIV) is the primary methodology used to study the unsteady
shock oscillation phenomena. Prior studies show shock oscillates one or two orders
smaller than the characteristics frequency of the incoming boundary layer. Particle
Image Velocimetry has done both in cross-stream plane and spanwise plane. Instan-
taneous PIV images in the cross-stream plane shows shock foot oscillates over a dis-
tance of approximately one step height. From PIV study in the cross-stream plane, it
is observed that the incoming boundary layer velocity fluctuations (upstream param-
eter) is corelated to the shock foot oscillation and the separation bubble (downstream
parameter) is correlated to the shock motion for the forward-facing step of step height
equals twice the incoming boundary thickness.
Keywords Shock wave-boundary layer interaction · Separation flow · Particle

image velocimetry · Forward-facing step
1 Introduction
Shock wave-boundary layer interactions (SWBLIs) have been studied for more than
past fifty years [1]. The interactions due to SWBLI, especially the ones leading to
shock induced separation (due to adverse pressure gradient), are typically unsteady
in nature. This can lead to large fluctuating pressure and thermal loads on the vehicle
structure. The resulting shock oscillations are generally composed of large-scale low-
J. N. Murugan (B)
School of Mechanical Engineering, Vellore Institute of Technology, Vellore 632014, India
R. N. Govardhan
Department of Mechanical Engineering, Indian Institute of Science, Bangalore 560012, India

220 J. N. Murugan and R. N. Govardhan
Fig. 1 Schematic of high speed flow over a two-dimensional forward-facing step (FFS). In the
figure, M∞ is the free-stream Mach number, δ is the oncoming boundary layer thickness, h is the
forward-facing step height and x, y and z (not shown in figure) represents streamwise, wall-normal
and spanwise directions, respectively
frequency oscillations and small-scale high-frequency oscillations. In particular, the

source of the low-frequency large-scale low-frequency shock oscillations remains
a subject of debate [2]. We study in the present work, the SWBLI at a forward-
facing step (FFS), as shown schematically in Fig. 1 to understand the reason for
shock oscillation. This configuration can be considered as an extreme case of the
compression ramp geometry which has been well studied. Prior studies on supersonic
flow over a FFS have reported the measured steady and unsteady wall pressure field
[3]. However, to the best of authors’ knowledge, there have been no studies that have
studied the flow field using particle image velocimetry (PIV) for FFS configuration.
This experimental study on supersonic flow over a FFS will help in the debate on
understanding the source of the unsteadiness in SWBLIs.
The goal of the current work is to experimentally study and understand shock
oscillations at a forward-facing step using velocity field. The velocity field is obtained
using particle image velocimetry (PIV) in the cross-stream (x-y) plane.
2 Experimental Methodology
2.1 Supersonic Facility
The present experimental studies were conducted in a Mach 2.5 supersonic blow-
down tunnel at the Department of Mechanical Engineering, Indian Institute of Sci-
ence, Bangalore. The test section had a length 35 cm and cross-section dimensions
100 mm width, 80 mm height. Two ELGI compressors were used to compress 1 bar
atmospheric air to 30 bar and stored at 30 bar in 6 m3 storage tanks. The boundary
layer developed on the nozzle surface and undergoes natural transition to turbu-
lent boundary layer under approximately adiabatic wall temperature conditions. The
free-stream velocity within the test section 585 m/s and the free-stream temperature
calculated using the Crocco-Busemann relation is approximately 129 K. The free-
Observation of Low-Frequency Shock Oscillation Over … 221
stream turbulence level was found to be less than 1.5% of the free-stream velocity.
The boundary layer thickness, δ has been measured using pitot tube and was found
to be around 6 mm. The unit Reynolds number (Re/m) and Reynolds number based
on momentum thickness (Reθ ) are 2.8×107 and 11,000, respectively. The stagna-
tion pressure and stagnation temperature in the settling chamber during bare tunnel
experiments were approximately 3.5 bar and 300 K, respectively. A useful stable run
time of about 40 s was available for this case. In the present experimental study of
SWBLI, a forward-facing step of height, h = 12 mm in a boundary layer of thickness,
δ = 6 mm is used. The length and width of the FFS 17.5 cm and 6 cm, respectively.
The step is placed symmetrically on the base of the test section leaving 2 cm (> 3δ)
gap on both sides of the FFS to avoid the effect of the side wall boundary layer on
SWBLI [4].
2.2 Particle Image Velocimetry
Particle image velocimetry (PIV) is a flow visualization technique primarily used in

the present study. In this work, we used two-dimensional two-component (2D2C) par-
ticle image velocimetry (PIV) system from Dantec Dynamics, Denmark. It comprises
of a New Wave 120 mJ energy, 532 nm dual cavity Nd-YAG laser, 1.92 Megapixel
(1600×1200 pixel resolution) CCD sensor camera, timer card for synchronization,
grabber card for image acquisition, beam to sheet converter (convex and concave lens
assembly) and PIV software DynamicStudio. Laser beam was made in to a sheet using
a spherical convex and cylindrical concave lens assembly, with adjustments possible
for the laser sheet thickness. For the present experiments laser sheet thickness was
kept approximately at 1.5 mm. The camera used had a minimum inter frame time of
0.3 µs when operated in the double exposure mode. In the present experiments, an
inter frame time of 0.5–1.0 µs was used. Nikon Micro-N60 mm f/2.8 D and 105 mm
f/2.8 G lens were used with the camera. During PIV experiments, the surface of
the model was painted with black color which helped in reducing the laser reflec-
tion from the surface. The flow was seeded with olive oil particles for PIV (Melling
1997). Olive oil particles were produced using in-house built Laskin nozzle based
seeder. Oil particles from this were injected into the settling chamber perpendicu-
lar to the flow direction 1 m upstream of the test section. Oil particles after mixing
with the incoming flow passed through a coarse mesh, honey comb and a series of
fine meshes. Two vertical side plates 2 mm thickness were welded inside the settling
chamber upstream of the honeycomb to minimize smearing of oil particles on the
Perspex window of the test section. This helped in getting clear PIV images and also
reduced the need for cleaning of the test section windows between runs.
The PIV flow visualization in the cross-stream plane (x-y) were done separately both
upstream and downstream of the forward-facing step of height, h/δ = 2.0 (δ is the
oncoming boundary layer thickness). The PIV velocity field results obtained from
these measurements, both upstream and downstream were processed separately and
then time-averaged. The time-averaged fields from both upstream and downstream
cases were then stitched into a single velocity vector field, shown in Fig. 2. It is
observed from the mean velocity vector field figure, the incoming boundary layer
deflects approximately at a distance 4h upstream of the step. A continuous straight
line is drawn to represent the sharp change in the flow deflection which is called as
separation shock. The flow separating at x/ h = −4 forms a large separation bubble
(recirculation region) upstream of the step. The flow after the separation bubble turns
around the top corner of the step, through a series of expansion waves. The series
of expansion waves at the top step corner is marked as dotted straight line. The zero
mean streamwise velocity contour line (u = 0) is drawn in the Fig. 2 as a dotted
curved line.
The mean separation shock location (xs ) is found by locating velocity field points
far away from the wall (outside the boundary layer), where the mean flow is deflected,
as shown schematically in Fig. 2. Mean separation shock location (xs ) at a particular
wall-normal location (y) is found by locating the streamwise (x) position where the
flow is deflected by greater than 6.5◦ ; this threshold deflection criteria being one-half
of the mean flow deflection angle. This method is followed at different wall-normal
locations from y = 1.5h to 2.1 h and flow deflection streamwise locations are marked
as solid circles (•) in Fig. 2. A continuous straight line is fitted through these solid
Fig. 2 Mean velocity vector field obtained after averaging approximately 800 instantaneous vector
fields over forward-facing step of h/δ = 2. Downstream of the separation shock and upstream of
the step face, a separation bubble (reverse flow region represented as curved dotted line, u = 0)
exists. For visibility, every fourth and third velocity vector is shown in streamwise and wall-normal
directions, respectively
circles (•) using least square method (shown as solid line through the solid circles)
and it is extended up to the bottom wall, shown by the dotted line, to locate the mean
separation shock location (xs , ). It should be noted that this mean separation shock
location (xs , ) is about 4h upstream of the step, and represents the mean location
of the separation shock as measured in the bulk flow outside the boundary layer.
This may be compared with the separation location on the surface determined from
surface oil flow, both being approximately close to 4h for the h = 2δ case [2].
Figure 3 shows two sample instantaneous velocity fields upstream of the forward-
facing step. Figure shows the sample instantaneous velocity fields with the separation
shock (dotted straight line) and the zero streamwise velocity contour (dotted curved
line) representing the reverse flow region shown in both cases. Figure 3a shows a
small reverse flow region with the separation shock closer to the step face, while in
Fig. 3b, the reverse flow region is much larger with separation shock being pushed
further upstream. Clearly, the flow field is highly unsteady with large-scale shock
oscillations.
From every instantaneous velocity vector field in the cross-stream plane, we
extract incoming boundary layer velocity field as upstream parameter and sepa-
ration region or reverse flow region as downstream parameter. The relation of these
Fig. 3 Instantaneous velocity vector plot showing a small and b big reverse flow region (u = 0).
Dotted straight line represents separation shock and dotted curved line represents zero streamwise
velocity contour (u = 0). For visibility, every third velocity vector is shown in the streamwise and
wall-normal directions
parameters with the corresponding instantaneous shock location obtained from the
velocity field is studied. The upstream parameter is obtained as the line-averaged
streamwise velocity at a wall-normal location of y = 0.5δ (=0.25h) over a distance
of 2δ, which is equivalent to the step height (h). Similarly, downstream parameter is
obtained as the area of the reverse flow region (hereafter called as separation region).
The angle formed by the mean separation shock wave (solid line) with the wall is the
shock wave angle represented as βs . Similarly instantaneous separation shock angle
(βs ) is also defined based on flow well outside the boundary layer (as shown in [3])
for all the instantaneous velocity fields. The variation of these parameters along with
the shock location is shown in Figs. 4, 5, 6 and 7. The instantaneous separation shock
location and separation shock angle is found out by following the same procedure
followed to locate the mean separation shock location and shock angle.
Figure 4 shows the variation of the line-averaged streamwise velocity fluctuations
measured between streamwise location of x/h = −6 to −5 upstream of the step
for each one of the 800 measured instantaneous velocity fields. The line-averaged
streamwise velocity fluctuation varies from −50 to 50 m/s. The positive line-averaged
streamwise velocity fluctuation implies that the incoming boundary layer is thinner
Fig. 4 Estimated values of 50

the line-averaged streamwise
velocity fluctuations
measured between x/ h = −6 25
to −5 upstream of the step
for each one of the 800
u′ (m/s)
measured instantaneous 0
velocity fields. The velocity
fluctuations varies from −50
to 50 m/s -25
-50
0 200 400 600 800
PIV Image

the reverse flow or separation
region for each one of the
800 measured instantaneous 300
velocity fields. It varies from
A (mm2)
no separation to separated
region of bubble size 200
approximately 300 mm2
100
0
0 200 400 600 800
PIV Image
Fig. 6 Estimated values of -3

the instantaneous shock
location normalized using
step height for each one of
the 800 measured -4
instantaneous velocity fields.
xS/h
It varies from −3.5 to −5h
-5
-6
0 200 400 600 800
PIV Image

the separation shock angle
for each one of the 800
measured instantaneous 40
velocity fields. It varies from
28 to 40◦
βS
35
30
25
0 200 400 600 800
PIV Image
and negative line-averaged streamwise velocity fluctuation implies that the incoming
boundary layer is thicker.
Figure 5 shows the reverse flow or separation region for each one of the 800
measured instantaneous velocity fields. The separation bubble varies from almost
small size to big size of approximately 300 mm2 . When the bubble size is big, we
can observe from Fig. 5, that separation shock is far from the step. Similarly, when
the bubble size is smaller, the separation shock is closer to the step.
Figures 6 and 7 shows separation shock location and shock angle for each one of
the 800 measured instantaneous velocity fields, respectively. The separation shock
location varies approximately from −3.5 to −5h upstream of the step. Separation
shock oscillates approximately 1.5 h. The separation shock angle varies from 28 to
40◦ .
It is observed from the cross correlation between the separation shock location and
incoming streamwise boundary layer velocity fluctuation that the separation shock
is weakly correlated to the incoming streamwise boundary layer velocity fluctuation.
At the same time, there is strong correlation between separation shock and separa-
tion bubble or reverse flow region. Based on this, we conclude that the separation
shock foot motion is correlated to incoming boundary layer velocity fluctuations and
separation shock oscillation is correlated to reverse flow or separation region.
4 Conclusions
Shock wave boundary layer interaction was studied over a forward-facing step of
step height equals two times the incoming boundary layer. Particle image velocime-
try is the primary methodology used to study the interaction between shock wave
and boundary layer. The purpose of using step height greater than the boundary layer
thickness is to obtain bigger separation bubble downstream of the separation shock
wave. Instantaneous PIV measurements shows separation shock oscillates over a
distance of step height about the mean separation shock position. Similarly, instan-
taneous PIV vector fields show for large separation bubble separation shock is far
from the step face and for smaller separation bubble, separation shock is near to the
step face. In general, correlation of separation shock with incoming boundary layer
velocity fluctuations and reverse flow regions shows that separation shock foot is
more related to reverse flow region.
References
1. Dolling, David S (2001) Fifty years of shock-wave/boundary-layer interaction research: what

next? AIAA J 39(8):1517–1531
2. Clemens NT, Narayanaswamy V (2014) Low frequency unsteadiness of shock wave/turbulent
boundary layer interactions. Annu Rev Fluid Mech 46:469–492
3. Zukoski EE (1967) Turbulent boundary-layer separation in front OFA forward-facing step.
AIAAJ 39(8):1746–1753
4. Murugan J, Govardhan R (2016) Shock-wave/boundary layer interaction in supersonic flow
over a forward-facing step. J Fluid Mech 807:258–302
RANS Computations of Hypersonic
Interference Heating on Flat Surface
with Protuberances
M. Mahendhran and C. Balaji
Abstract RANS computations are performed for M ∞ = 8.2, Re∞ /m = 9.35 × 106
flow past a flat surface with protuberance of size of the same order as that of the
boundary layer. The local intensification of heat transfer rates in the vicinity of the
protuberance is considered as the main focus of the study. The effect of separated and
unseparated boundary layer is considered by varying the protuberance angle. The
results are compared with available experimental data. The maximum heat transfer
rate, a major design parameter, is found to be fairly predicted by computations.
However, accurate prediction of parameters such as upstream separation length and
heat flux distribution requires further research and calibration of available turbulence
models.
Keywords Hypersonic flow · Flat surface with protuberance · CFD++
1 Introduction
Protuberances in the form of rivet heads, screws, etc., present on the surfaces of
missiles and other aero-vehicles traveling at hypersonic speeds interfere with the
otherwise undisturbed flow past these surfaces and result in the appearance of shock
waves around them, altering the evolution of the boundary layer and intensifying
local heat transfer rates to the vehicle’s surface [1].
Estruch et al. [2] have performed experiments involving hypersonic flow past
protuberances with finite span and height of the order of the boundary layer thickness
in order to characterize the augmentation in heat transfer to the vehicle’s surface
and, depending on the configuration of the protuberance, the location where the
maximum heat transfer rates can be expected. The complex interaction between
M. Mahendhran (B) · C. Balaji

Metacomp Technologies Pvt. Ltd, #14, 3rd Cross St., Karpagam Gardens, Adayar, Chennai
600020, India
C. Balaji

228 M. Mahendhran and C. Balaji
the shock wave and boundary layer at hypersonic speeds makes this localized heat
transfer dependent on many parameters, including the protuberance deflection angle,
the state of boundary layer, flow Reynolds and Mach numbers.
Computing such flow fields involving shock wave/boundary layer interactions
at hypersonic speeds and predicting the augmentation in heat transfer rates and its
location in the vicinity of protuberances is a challenging task. In this paper, the
hypersonic flow past a protuberance considered by Estruch et al. [2] is computed
using the CFD software suite, CFD++ [3]. Computations are performed for both
unseparated and separated boundary layer ahead of the protuberance by varying the
protuberance angle (α), and the results are compared with experimental data.
2 Geometry and Mesh
The protuberance is a wedge of height 5 mm and width 13.5 mm. It protrudes from
a flat plate of length 255 mm, width 155 mm and thickness 6 mm and is located
at 175 mm from the leading edge of the plate (x le ), and it extends up to 80 mm
downstream (see Fig. 1).
The surface mesh is generated using triangles and quadrilaterals, and the near
wall and core regions are constructed with prism layers and hexahedral elements,
respectively. Prism layers are constructed such that the wall y+ for the first layer is
less than one in most regions of the domain and reaches a maximum value of 2.5 at
the side edge corner of the protuberance. In the volume regions that are marked with
constant mesh size, hexahedral cells resembling a structured mesh are employed (see
Fig. 2). Pyramids and tetrahedral cells are used as required to bridge the near-wall
prismatic cells and the surface mesh with the core hexahedral cells. The total mesh
size of this 3D hybrid unstructured mesh is ≈20 million cells. A cut section through
the mid-span of a representative mesh employed for this study is shown in Fig. 2.
175 80
6
5
30° wedge angle
All dimensions are in mm
Fig. 1 Plate with protuberance in side view showing important dimensions

RANS Computations of Hypersonic Interference Heating … 229
Fig. 2 Unstructured computational mesh in the vicinity of the protuberance
3 Computational Details
The fluid flow is modeled using the 3D compressible Navier–Stokes equations with
due consideration of viscous heating in the energy balance. The working fluid, air,
is treated as real gas, with the consideration of specific heat variation with respect
to temperature. Its material properties—viscosity, thermal conductivity and their
dependence on temperature—are modeled using Sutherland’s law.
Considering the complexity of the flow field and the uncertainty of the best suited
turbulence model to simulate turbulence in such flow fields, three different turbulence
models are employed in this study, viz., the one-equation Goldberg Rt model [4], the
two-equation SST model and the two-equation nonlinear cubic k-ε model [5] with
variable Prandtl number treatment [6].
A second-order spatial discretization is employed along with a continuous-type
slope limiter in order to avoid spurious numerical oscillations across sharp variations
in solution variables [7]. Partial blending from the first-order discretization (up to
20%) is required to support stability and improve convergence while employing cubic
k-ε model for computations involving protuberance with boundary layer separation
(α ≥ 45°).
Pressure switches [7] are used to control additional diffusion only at locations of
sharp variation in pressure, ensuring stability with minimal penalty on the compu-
tation accuracy. An implicit iterative advancement of the solution coupled with an
implicit treatment of boundary conditions and an algebraic agglomeration multi-grid
helps in achieving rapid convergence toward the steady-state flow field.
Vortex generators (VGs) are employed by Estruch et al. [2] to trip the boundary
layer. The VGs are located 20 mm from the leading edge, and vortex breakdown is
reported to occur at about 100 mm from the leading edge. In an attempt to transition
the boundary layer to turbulent, numerical trip wires are placed 100 mm from the
leading edge [7].
4 Results
Unless otherwise stated, computations are performed for M ∞ = 8.2, Re∞ /m = 9.35
× 106 . The walls are kept at 295 K, as in experiment. The near-wall shear stresses
and heat transfer rates are directly computed using a solve-to-wall approach without
resorting to wall functions.
Tables 1 and 2 summarize the boundary layer height and heat transfer rates,
respectively, obtained for an undisturbed boundary layer flow past the plate (without
protuberance). The results include the measurement of boundary layer height and
heat transfer to the plate at x le = 175 mm, both in the presence and absence of
boundary layer trip.
Table 1 Boundary layer height at x le = 175 mm for an undisturbed boundary layer

BL trip BL height (mm)
Exp. CFD++
Rt SST Cubic k-ε
No 2.25–2.75 2.58 2.54 2.61
Yes 4.5–5.5 5.51 2.53 2.64
Table 2 Heat transfer rate at x le = 175 mm for an undisturbed boundary layer

BL trip Heat transfer rate (W/cm2 )
Exp. CFD++
Rt SST Cubic k-ε
No 1.62–1.98 1.9 1.74 1.35
Yes 5.31–6.49 5.85 2.39 1.35
In the absence of a boundary layer trip, the boundary layer height predicted by
all the models lies within the experimental range. The corresponding heat transfer
rates predicted by Rt and SST models lie within the experimental range, whereas it is
underpredicted by 15% from the mean value when the cubic k-ε model is employed.
Only the Rt model is observed to respond to the placement of numerical trip
wires, turning the boundary layer into a turbulent state with a corresponding increase
in boundary layer height and with heat transfer rates that lie within or very close to
experimental range. The presence of turbulent dissipation rates in the SST and cubic
k-ε model makes the flow relaminarize rapidly after the numerical trip. The boundary
layer state remains laminar for these models and is found to have a profound impact
on the evolution of flow field (see Fig. 3) and the resulting heat transfer that occurs
in the vicinity of the protuberance (see Figs. 4, 5, 6, 7 and 8).
The presence of the protuberance increases the local heat transfer rate in its vicinity
(see Fig. 4) and induces a separation shock ahead of it, whose intense adverse pressure
gradient causes the boundary layer to separate, resulting in a recirculation zone ahead
of the protuberance (see Fig. 5).
The maximum heat transfer rate in the vicinity of the protuberance—a signifi-
cant parameter for engineering design requirements—is predicted to a fair extent by
Fig. 3 Mach number contours in the protuberance region for α = 90° case
50
45 exp flow α
40 Rt
SST
35
Cubic k-eps
Stmax x 10-3
30
25
20
15
10
5
0
0 15 30 45 60 75 90 105 120 135 150
α (°)
Fig. 4 Comparison of maximum heat flux variation with respect to α
70
separation
60 shock
flow α
50 exp L
Rt
40
RL x 104
SST
30 Cubic k-eps
20
10
0
0 15 30 45 60 75 90 105 120 135 150
α (°)
Fig. 5 Non-dimensional separation length ahead of protuberance for different α
computations, up to α = 90°, see Fig. 4. For α = 135°, heat transfer rates in the
vicinity of protuberance converged to a constant value for the Rt model, whereas
oscillations are noted for SST and cubic k-ε, with notably large amplitudes for the
cubic k-ε model. This could possibly indicate the presence of a large-scale structures
14
exp, α=30°
num, α=30° flow α 12
exp, α=45° xk
10
num, α=45°
exp, α=60° 8
St x 10-3
num, α=60°
exp, α=90° 6
num, α=90°
4
0
-30 -25 -20 -15 -10 -5 0
Rex x 104
Fig. 6 Comparison of heat flux distribution, obtained using Rt turbulence model ahead of the
protuberance along the centerline for different α
14
exp, α=30°
flow α 12
num, α=30°
exp, α=45° xk
10
num, α=45°
exp, α=60° 8
St x 10-3
num, α=60°
exp, α=90° 6
num, α=90°
4
0
-30 -25 -20 -15 -10 -5 0
Rex x 104
Fig. 7 Comparison of heat flux distribution, obtained using SST turbulence model, ahead of the
protuberance along the centerline for different α
14
exp, α=30°
flow α 12
num, α=30°
exp, α=45° xk
10
num, α=45°
exp, α=60° 8
St x 10-3
num, α=60°
exp, α=90° 6
num, α=90°
4
0
-30 -25 -20 -15 -10 -5 0
Rex x 104
Fig. 8 Comparison of heat flux distribution, obtained using Cubic k-ε turbulence model, ahead of
the protuberance along the centerline for different α
in the vicinity of protuberance possibly caused by a laminar boundary layer ahead

of the protuberance at these angles.
Figure 5 shows the variation of separation length ahead of the protuberance from
the computations. It is underpredicted by Rt model and is overpredicted by SST and
cubic k-ε. The primary reason for overprediction by SST and cubic k-ε is the failure
of the numerical trips to transition the boundary layer to a turbulent state, leaving the
laminar boundary layer to be influenced to a greater extent by the adverse pressure
gradient introduced by the separation shock. This also affects the heat transfer rates
ahead of the protuberance as seen in Figs. 6, 7 and 8.
As expected, the Rt model with a turbulent oncoming boundary layer (Fig. 6) fares
better than the SST (Fig. 7) and cubic k-ε (Fig. 8) models, with oncoming laminar
boundary layers, wherein a general underprediction in heat transfer rates is noted.
Considering the fair representation of the flow field by the Rt model, further anal-
yses are performed using this model, for α = 90° unless otherwise stated explicitly.
This includes the dependence of results on the type of mesh employed (unstructured
or structured), mesh resolution and the location of numerical trip.
The structured mesh with similar mesh resolution as in the unstructured mesh was
prepared. The heat transfer parameters such as the maximum heat transfer rate in the
vicinity of the protuberance (a major design parameter), the heat transfer distribution
ahead of the protuberance and along the centerline obtained from the unstructured
and structured meshes are compared in Table 3, Figs. 9 and 10, respectively. Table 4
Case Maximum heat flux ahead of protuberance
maximum heat transfer ahead
(Stmax × 10–3 )
of protuberance between
unstructured and structured Experiment 11.022
meshes for α = 90° case Unstructured 10.593
employing Rt turbulence
Structured 10.579
model
Fig. 9 Heat flux distribution around the protuberance for α = 90° base case
12
unstructured mesh
10
structured mesh
8
St x 10-3
0
-30 -25 -20 -15 -10 -5 0
Rex x 104
Fig. 10 Comparison of heat flux ahead of protuberance along centerline between unstructured and
structured meshes for α = 90° case employing Rt turbulence model
Mesh type Separation distance (ReL × 104 )
separation distance ahead of
protuberance between Experiment 21.638
unstructured and structured Unstructured 12.479
meshes for α = 90° case
Structured 13.522
employing Rt turbulence
model
compares the non-dimensional separation length ahead of the protuberance obtained

from unstructured and structured meshes.
Only a marginal difference in results is noted because of the change in the type
of mesh employed in the computations. The unstructured mesh with the hex-core
feature prepared using IMIME [8] produces results that compare very closely with
its structured mesh counterpart. Further analyses are performed only using hex-core
unstructured mesh unless otherwise stated explicitly.
Computations are repeated by refining the mesh in the shock/boundary layer
interaction region ahead of the protuberance and by locating the numerical trip to an
upstream location, 20 mm from the leading edge.
The mesh count for the refined mesh is almost doubled from the base case. The
mesh is refined in the region ahead of the protuberance to predict the separation
shock’s location and strength more accurately. The placement of the numerical trip
at an upstream location is expected to change the momentum of the boundary layer.
Figure 11 compares the heat flux distribution ahead of the protuberance for these
cases. Mesh refinement has increased the upstream heating marginally and has
14
exp
base case 12
fine mesh 10
upstream trip
8
St x 10-3
0
-30 -25 -20 -15 -10 -5 0
Rex x 104
Fig. 11 Heat flux ahead of protuberance along centerline for α = 90° case
Table 5 Separation distance

Case Separation distance (ReL × 104 )
ahead of protuberance for α
= 90° case Experiment 21.638
Base case 12.479
Fine mesh 13.540
Upstream trip 13.885
Table 6 Maximum heat flux

Case Maximum heat flux ahead of protuberance
ahead of protuberance for α
(Stmax × 10–3 )
= 90° case
Experiment 11.022
Base case 10.593
Fine mesh 10.518
Upstream trip 7.700
resulted in corresponding increase in separation distance, see Table 5. The upstream

trip placement has resulted in similar increase in separation distance; however, it has
resulted in much lower heat transfer to the plate. Maximum heat flux obtained for
these cases is summarized in Table 6. The base and refined mesh results fall within
the experimental error of 10%; however, the upstream trip case underpredicts the
maximum heat flux significantly.
5 Summary and Conclusion
Hypersonic flow past wedge-shaped protuberances whose dimensions are of the

same order as that of the boundary layer height are computed using three turbulence
models, viz. Rt, SST and cubic k-ε models, and salient results are compared with
experimental data.
Initially, an undisturbed boundary layer flow over the flat plate is computed. All
the models fare well in the absence of boundary layer transition and the heat flux and
boundary layer height showed a very good match with the experimental results.
Numerical trip wires are employed in the computations to transition the boundary
layer to turbulent state as in experiments. Only Rt model is observed to respond to
numerical trip placement and shows very good comparison of boundary layer height
and heat transfer rates. Under the flow conditions considered, the numerical trips are
essentially ineffective in the case of the SST and cubic k-ε models. This is likely due
to the turbulence dissipation rate determined from the second transport equation in
both these models.
Subsequently, the flow past wedge-shaped protuberances are computed for both
attached and separated boundary layers, obtained by varying the wedge angle.
Computations predict the maximum heat flux with reasonable accuracy, and this
is found not to depend much on the state of boundary layer. However, the state of
the boundary layer has a profound impact on the separation distance and the spatial
extent of upstream heat transfer. The Rt model with a turbulent boundary layer is
found to underpredict the separation distance, whereas the SST and cubic k-ε models
with laminar boundary layer overpredict the separation distance. This suggests that
the physical boundary layer, post-trip, has insufficient time to become fully devel-
oped and remains transitional. Heat transfer distribution ahead of the protuberance is
represented reasonably well by the Rt model, but with severe underprediction noted
for SST and cubic k-ε models.
In concurrence with the experimental findings [2], the computations indicate that
the state of the oncoming boundary layer plays an important role in the flow field
evolution in the vicinity of the protuberance. Further research is required to sensitize
models with length scale determining transport equations (SST and cubic k-ε) to
numerical trip placement, to conduct studies involving boundary layer transition,
and subsequently to evaluate their performance with models that lack a free-stream
turbulence decay mechanism (e.g., Rt).
References
1. Guoliang M, Guiqing J (2004) Comprehensive analysis and estimation system on thermal

environment, heat protection and thermal structure of spacecraft. Acta Astron 54(5):347–356
2. Estruch D, MacManus DG, Stollery JL, Lawson NJ, Garry KP (2010) Hypersonic interference
heating in the vicinity of surface protuberances. Exp Fluids 49:683–699
3. Metacomp Technologies Inc. Homepage. https://www.metacomptech.com. Last accessed
2018/09/26
4. Goldberg U (2003) Turbulence closure with a topography-parameter-free single equation model.
IJCFD 17(1):27–38
5. Palaniswamy S, Goldberg U, Peroomian O, Chakravarthy S (2001) Predictions of axial and
transverse injection into supersonic flow. Turbul Combust 66:37–55
6. Goldberg UC, Palaniswamy S, Batten P, Gupta V (2010) Variable turbulent Schmidt and Prandtl
number modeling. Eng Appl Comput Fluid Mech 4(4):511–520
7. ICFD++ theory manual, Metacomp Technologies Inc.
8. IMIME theory manual, Metacomp Technologies Inc.
Multi-fidelity Aerodynamic Optimization
of an Airfoil at a Transitional Low
Reynolds Number
R. Priyanka, M. Sivapragasam , and H. K. Narahari
Abstract Aerodynamic shape optimization is carried out for (a) maximizing lift
coefficient, C l and (b) maximizing endurance factor (C l 3/2 /C d ) of an airfoil at a
transitional low Reynolds number (Re = 5 × 104 ) using surrogate-based optimiza-
tion technique. Bezier curves are used to parameterize the airfoil. Latin hypercube
sampling (LHS) technique is used to sample the initial set of samples. The aero-
dynamic response is estimated using low- and high-fidelity datasets. A co-kriging-
based multi-fidelity surrogate is built using these datasets, and the response surface is
used in the optimization procedure for the search of optima. Expected improvement
strategy is used to update the surrogate at every iteration, and optimization is termi-
nated once the convergence criterion is met. The objective function improvement for
maximizing C l and maximizing (C l 3/2 /C d ) is 18.08% and 38.70%, respectively.
Keywords Low Reynolds number · Laminar separation bubble · Co-kriging ·

Aerodynamic optimization
1 Introduction
Micro air vehicles (MAV) find widespread use in many civilian and military applica-
tions. These small-sized vehicles typically cruise at a Reynolds number (Re) of the
order of 105 based on the flight speed and chord length. In the low Reynolds number
regime, viscous forces have a significant effect on the aerodynamic characteristics.
R. Priyanka (B) · M. Sivapragasam · H. K. Narahari

Technology, M.S. Ramaiah University of Applied Sciences, Bangalore, India
M. Sivapragasam
H. K. Narahari

240 R. Priyanka et al.
The flow properties in this regime have significant effects on the aerodynamic perfor-
mance of MAVs. Hence, the understanding of aerodynamics at low Reynolds number
regime becomes important. In low Reynolds number regime, the flow is very sensi-
tive to small changes in flow parameters such as airfoil geometry, Reynolds number,
freestream turbulence and angle of attack. In this regime, the airfoil often encoun-
ters laminar boundary layer separation, transition to turbulence and reattachment on
the suction surface which forms a laminar separation bubble (LSB) on the airfoil.
The LSB on the suction surface deteriorates the aerodynamic performance of the
MAV [1]. Optimization techniques are useful in improving the aerodynamic perfor-
mance. The application of such formal optimization methods is sparse in the literature
involving transitional low Re.
In the present paper, a surrogate-based optimization methodology is formu-
lated to achieve enhanced aerodynamic performance characteristics. Optimization
is performed for two different objective functions: (a) maximizing lift coefficient,
C l and (b) maximizing endurance factor (C l 3/2 /C d ), at Re = 5 × 104 . For each of
these objective functions, multi-fidelity surrogate models are constructed, and genetic
algorithm is employed for optimization.
2 Airfoil Parametrization
The choice of a suitable parametrization technique is important in optimization. Two

important points have to be kept in mind while parametrizing a given geometry:
Firstly, the parametrization technique must be capable of representing the given
geometry very accurately, and secondly, it must do this by utilizing a minimum
number of control variables for ease in optimization. In the present study, NACA
0012 airfoil is chosen as the baseline and is parametrized using Bezier curves.
The airfoil is represented by 12 control points with six each on upper and lower
surfaces of the airfoil. Among the 12 control points, six control points are fixed to
maintain the leading edge continuity and trailing edge closure. The remaining six
control points were allowed to participate in the design process. The control points
are allowed to move in y-direction. The bounds on the design variables were set to
50% of their initial values on either side of the control point [2]. Figure 1 shows the
Bezier approximation of NACA 0012 airfoil along with the control points and their
bounds.
3 Co-kriging-Based Surrogate Modeling
In the present study, co-kriging-based surrogate modeling was employed. A large set
of cheap data was coupled with a comparatively smaller set of expensive data build the
surrogate. Co-kriging-based surrogate model takes advantage over standard kriging
model by constructing an accurate model with a fewer high-fidelity evaluations. In
Multi-fidelity Aerodynamic Optimization … 241
Fig. 1 Bezier approximation

of NACA 0012 airfoil also
showing the control points
and their bounds
the present study, high-fidelity evaluations were done in ANSYS FLUENT, and low-
fidelity evaluations were done in XFLR5. High-fidelity solutions are more accurate
and expensive, but low-fidelity solutions are less accurate and cheap. Co-kriging
model was built by combining the set of data from high-fidelity and low-fidelity
evaluations as follows:
⎡ ⎤ ⎡ ⎤
x (1) yc x (1)
c
c
⎢ . ⎥ ⎢ .. ⎥
⎢ .. ⎥ ⎢ . ⎥
⎢ ⎥
⎢ ⎢ (n c ) ⎥
⎥ ⎢ (n c) ⎥
Xc ⎢ xc ⎥ Y c (xc ) ⎢ yc x c ⎥
x= = ⎢ (1) ⎥ and y = =⎢
⎢
⎥
⎥ (1)
Xe ⎢ xe ⎥ Y e (xe ) ⎢ ye x (1) ⎥
⎢ ⎥ ⎢
e
⎥
⎢ .. ⎥ ⎢ .. ⎥
⎣ . ⎦ ⎣ . ⎦
(n c ) (n e)
xe ye x e
where X denotes a vector of design variables, X c and X e denotes the sampling plan
of the cheap and expensive dataset, respectively. Y c and Y e represents the corre-
sponding functional values at these training points. The expensive code using the
auto-regression model of Kennedy and O’Hagan [3] is expressed as
Z e (x) = γ Z c (x) + Z d (x) (2)
where Z c (x) and Z e (x) denotes the local features of cheap and expensive codes,
respectively, while Z d (x) denotes a Gaussian process which represents the difference
between the γ Z c (x) and Z e (x). γ denotes the scaling parameter. The covariance
matrix with both high- and low-fidelity dataset is given as

γc2 ψc (X c , X c ) ργc2 ψc (X c , X e )
C= (3)
ργc ψc (X e , X c ) ρ γc ψc (X e , X e ) + γd2 ψd (X e , X e )
2 2 2
ψc (X c , X e ) represents the correlation matrix between the data X c and X e . There are
two correlation and four hyper-parameters to be determined because of the usage of
two datasets. For the given data and the values of estimated hyper-parameters, the
co-kriging predictor is given as. Forrester et al. [4] can be referred for more detailed
information.
y e (x) = μ̂ + cT C −1 y − 1μ̂

(4)
4 Computational Procedure and Grid
The high-fidelity computations are performed using ANSYS FLUENT. The two
dimensional, steady and incompressible Navier–Stokes equations in Eq. 5 are solved
numerically.
∇ · u = 0
1
u · ∇ u = − ∇ p + ν∇ 2 u − u i u j (5)
ρ
Transition SST k-ω model is used here for computations. The transition γ −Reθ
model is coupled with SST k-ω model where is turbulence kinetic energy and
is specific dissipation rate, is intermittency, Reθ is Re based on momentum thick-
ness. Spatial discretization was done by using a formally second-order accurate
discretization scheme, and double-precision arithmetic solver is used for calcula-
tions. Boundary conditions applied are velocity inlet and pressure outlet at the domain
upstream and domain downstream, respectively. Velocity inlet boundary condition,
calculated based on the required Re, is imposed on the upper and lower boundaries.
The no-slip boundary condition is imposed on the airfoil. The convergence of the
residuals was 10–6 .
The sample airfoil geometries are constructed using CATIA. Structured mesh of
C-grid topology is generated around airfoil using ANSYS ICEM-CFD. The domain
is extended 10c upstream, above and below the airfoil and 20c downstream from the
leading edge, where c is airfoil chord length. Mesh generation is automated using
script file in ANSYS ICEM-CFD. Figure 2a shows the mesh in the computational
domain and (b) shows the grid around the airfoil.
The computational domain had 24,500 grid points which was chosen after a careful
grid independence study. Grid independence study is performed to ensure that the
results are independent of the number of grid points. Three different meshes are
Fig. 2 a Mesh in the computational domain. b Grid around airfoil
created with different number of grid points (fine mesh, medium mesh, coarse mesh).
To perform grid independence study, CFD simulations are carried out at Re = 5 ×
104 , Tu = 0.3% and α = 6°. Table 1 shows the values of C l and C d for different
number of grid points. The grid convergence index (GCI) between G2 and G3 is less
when compared between G1 and G2 . From the grid independence study, grid G2 is
selected for the further computations. Y + is maintained 0.6 for all the computations.
The present computational results were validated with the experimental results
of Kim et al. [5] at Re = 4.8 × 104 , Tu = 0.3% and α = 6˚. Figure 3 shows the
comparison of C p distribution with experimental results. A good agreement was seen
validating the present computational procedure.

α = 6°
Grid No. of grid points Cl GCI for Cl Cd GCI for Cd
G1 11,712 0.5629 1.41180 0.0332 2.29432
G2 24,500 0.5814 0.0312
0.35175 0.47963
G3 49,200 0.5933 0.0308
Fig. 3 Comparison of C p -2
distribution with
Present CFD
experimental result Expt. (Kim et al., 2011)
-1.5
-1
Cp
-0.5
0.5
1
0 0.2 0.4 0.6 0.8 1
x/c
5 Optimization Methodology
In the present work, surrogate-based multi-fidelity optimization was performed. Co-

kriging-based multi-fidelity surrogate was built using low- and high-fidelity datasets.
Latin hypercube sampling (LHS) technique [4] is used to sample the design space.
The sample points given by LHS are converted into the actual coordinates of the
design variables based the upper and the lower limits in the y-direction. A sample
of 150 airfoils is generated using LHS technique, and the objective functions were
evaluated using XFLR5. Thirty airfoil samples were randomly chosen from the set
of 150 airfoils, and the objective function is evaluated using FLUENT. The sample
geometries were constructed using CATIA. Automated mesh generation was done
using script files in ICEM-CFD. Co-kriging surrogate model was constructed using
these 150 cheap and 30 expensive datasets. The surrogate model was subjected to
genetic algorithm toolbox in MATLAB. Expected improvement infill technique [4]
was used to update the surrogate. Expensive evaluations were carried out for the
optimal geometry identified by the GA, and it was used to update and reconstruct the
surrogate. The procedure was repeated until the convergence criterion is met. The
Fig. 4 Quality of 0.6

multi-fidelity surrogate
0.55
MFS prediction
0.5
r2 = 0.757
0.45
0.4
0.4 0.45 0.5 0.55 0.6
Independent data set

convergence criteria are f k∗ − f k−1
∗
≤ 10–3 where f k∗ is current objective function
∗
value and f k−1 pervious objective function value. Figure 5 shows the flowchart of
the expected improvement-based optimization process.
A co-kriging-based multi-fidelity surrogate model was built using 150 low-fidelity
and 30 high-fidelity datasets. The quality of this surrogate model was estimated by
obtaining the model predictions and comparing with an independent high-fidelity
dataset as shown in Fig. 4. This independent high-fidelity dataset was obtained from
a separate study [2]. The correlation coefficient, r 2 , between the two sets of data, was
0.757. It is important to note that this value of r 2 is only for the initial build of the
surrogate. In the iterative process described in the previous paragraph and illustrated
in Fig. 5, high-fidelity infill point, evaluated from the expected improvement method,
is used to update the surrogate model at every iteration, and thereby, the quality of
the model improves subsequently.
6 Results and discussion
The objective procedure was carried out as detailed in Sect. 5. The convergence of
the objective function with iterations for both the objective functions is shown in
Fig. 6. It can be seen that the objective function converges in about eight iterations.
The optimal airfoil geometries for both the objective functions are shown in Fig. 7.
The optimal airfoils had camber and a reduced thickness. The maximum t/c for
optimal geometry is 12.6% at 30% chord, and the maximum camber of the optimal
airfoil is 0.4% occurring at 19% chord for maximizing C l . The maximum t/c for
Objective function
Design variables
Design
Constraints
Design of experiments
Cheap evaluation Expensive evaluation
Multi-fidelity surrogate
Genetic algorithm
CFD
No
Con-
vergence
Yes
Optimal geometry
Fig. 5 Flowchart of multi-fidelity optimization procedure

Fig. 6 Convergence of 0.6

objective function for
a maximizing C l and
b maximizing (C l 3/2 /C d ) 0.5
0.4
0.3
Cl 0.2
0.1
0
0 2 4 6 8
Itrerations
(a)
20
18
Cl /Cd
16
3/2
14
0 2 4 6 8
Itrerations
(b)
optimal geometry is 9.2% at 40% chord, and the maximum camber of the optimal
airfoil is 1.6% occurring at 17% chord for maximizing (C l 3/2 /Cd ).
The streamlines over the optimal airfoils are shown in Fig. 8. A prominent LSB
was observed on the upper surfaces of the optimal airfoils. The separation, tran-
sition and reattachment locations and the length of the LSB were evaluated from
C f distribution on the upper surface of the airfoils shown in Fig. 9. They are tabu-
lated in Table 2. There was early separation and early reattachment resulting in a
reduced bubble length for the optimal airfoil for maximizing C l case. However, for
Fig. 7 Comparison of Optimalgeometry

Baseline
optimal and baseline
geometry for a maximizing 0.05
C l and b maximizing
(C l 3/2 /C d )
y/c
0
-0.05
0 0.2 0.4 0.6 0.8 1

x/c
(a)
Optimalgeometry
Baseline
0.05
y/c
-0.05
0 0.2 0.4 0.6 0.8 1

x/c
(b)
Fig. 8 Streamlines over the

optimal geometry for
a maximizing C l and
b maximizing (C l 3/2 /C d )
Fig. 9 C f distribution on the 0.03

upper surface of the optimal
geometry for a maximizing
C l and b maximizing
C l 3/2 /C d 0.02
0.01
Cf
0
0 0.2 0.4 0.6 0.8 1
x/c
-0.01
(a)
0.03
0.02
0.01
Cf
0
0 0.2 0.4 0.6 0.8 1
x/c
-0.01 (b)
Table 2 Comparison of LSB of optimal and baseline airfoil for a maximizing C l . b maximizing
C l 3/2 /C d
(a)
(x/c)s (x/c)t (x/c)r LLSB
Baseline 0.23 0.66 0.79 0.56
Optima 0.21 0.61 0.69 0.48
(b)
Baseline 0.23 0.66 0.79 0.56
Optima 0.19 0.67 0.78 0.59
the optimal airfoil for maximizing (C l 3/2 /C d ) case, there was early separation, but
the reattachment location was only slightly early. This resulted in the little longer
bubble length compared to the baseline airfoil.
Figure 10a shows the pressure distribution over the optimal and baseline airfoils
for maximizing C l case. From C p plot, we can see that the change in pres-
sure distribution on the upper and lower surfaces of the airfoil contributes to C l
improvement.
However, the pressure distribution on the upper surface of optimal airfoil for
maximizing (C l 3/2 /C d ) case was nearly the same as the baseline airfoil. This is due
to the fact that the upper surface of both baseline and optimal airfoils had similar
Fig. 10 Pressure -2
distributions over optimal
and baseline airfoil for -1.5
a Maximizing C l , Optimal geometry
Baseline
b Maximizing (C l 3/2 /C d )
-1
Cp
-0.5
0.5
1
0 0.2 0.4 0.6 0.8 1
x/c
(a)
-2
-1.5
Optimal geometry
Baseline
-1
Cp
-0.5
0.5
1
0 0.2 0.4 0.6 0.8 1
x/c
(b)
Table 3 Aerodynamic
(a)
coefficients for the baseline
and optimal geometries for Cl
a maximizing, C l Baseline 0.459
b maximizing (C l 3/2 /C d )
Optimal 0.542
(b)
Cl Cd (C l 3/2 /C d )
Baseline 0.459 0.0258 12.04
Optimal 0.552 0.0245 16.70
geometry; see Fig. 7b. Much of the C l improvement was brought about by the lower
surface and the C d improvement is due to the reduced thickness.
The aerodynamic coefficient of the baseline and optimal airfoils for both the
objective functions is listed in Table 3. The optimal geometry for maximizing C l
has 18.08% improvement compared to the baseline. The optimal geometry for
maximizing endurance factor (C l 3/2 /Cd ) has 38.70% improvement compared to the
baseline.
It is interesting to note that C l for maximizing (C l 3/2 /C d ) objective function is
slightly more than that achieved by the maximizing C l objective function. To see why
this happens, we plotted the camber distribution of the optimal airfoils in Fig. 11. It
is observed that the optimal airfoil for (C l 3/2 /C d ) objective function had picked up
larger camber than the other optimal airfoil. The maximum camber was about 1.6%
occurring at 17% chord. This higher camber leads to higher C l .
Fig. 11 Camber distribution 0.035

for the optimal airfoils
0.030
Camber for max. Cl
0.025 Camber for max. (C3/2
l / Cd )
0.020
0.015
y/c
0.010
0.005
0.000
0 0.2 0.4 0.6 0.8 1
-0.005
x/c
-0.010
7 Conclusion
Aerodynamic shape optimization of an airfoil was carried out to obtain optimal airfoil
shapes at a transitional low Reynolds number. Parameterization of the airfoil was
done using Bezier curve. The Bezier control points were the design variables. Latin
hypercube sampling technique was used to sample the design space. Simulations
were carried out at Re = 5 × 104 , α = 4° and Tu = 0.3%, using a laminar–turbulent
transition model. Optimization was performed for two objective functions: (a) maxi-
mizing C l and (b) maximizing (C l 3/2 /Cd ). Co-kriging-based surrogate models were
built using multi-fidelity datasets. The surrogate models were subjected to genetic
algorithm to obtain the optima. The optimal airfoil for both the objective functions
had camber. The objective function improvement for maximizing C l and maximizing
(C l 3/2 /C d ) was 18.08 and 38.70%, respectively, compared with the baseline NACA
0012 airfoil.
References
1. Shyy W, Lian Y, Tang J, Viieru D, Liu H (2011) Aerodynamics of low Reynolds number flyers.
Cambridge University Press, Cambridge
2. Pranesh C, Sivapragasam M, Narahari H (2019) Multi-fidelity aerodynamic shape optimization
of an airfoil at transitional low Reynolds number. In: Ncmdao.org. https://www.ncmdao.org/
docs/NCMDAO2018_Schedule.pdf. Accessed 30 Aug 2019
3. Kennedy M, O’Hagan A (2000) Predicting the output from a complex computer code when fast
approximations are available. Biometrika 87:1–13. https://doi.org/10.1093/biomet/87.1.1
4. Forrester A, Sobester, A, Keane A (2019) Engineering design via surrogate modelling: a prac-
tical guide (eBook, 2008) [WorldCat.org]. In: Worldcat.org. https://www.worldcat.org/title/eng
ineering-design-via-surrogate-modelling-a-practical-guide/oclc/264714649. Accessed 30 Aug
2019
5. Kim D, Chang J, Chung J (2011) Low-Reynolds-number effect on aerodynamic characteristics
of a NACA 0012 airfoil. J Aircraft 48:1212–1215. https://doi.org/10.2514/1.c031223
Aerodynamic Optimization of Transonic
Wing for Light Jet Aircraft
K. Sathyandra Rao, M. Sivapragasam , H. K. Narahari,

and Aneash V. Bharadwaj
Abstract Light jet aircraft capable of carrying 6–10 passengers over short and
medium ranges find great demand in the present and future civil aviation market.
Light jets often cruise at high subsonic Mach numbers enabling greater range. Conse-
quently, the drawback associated with high Mach number flight is that it leads to
the deterioration of the aircraft’s aerodynamic characteristics. Aerodynamic opti-
mization of a typical transonic wing was performed to reduce the wing cruise drag
coefficient at a fixed Mach number and lift coefficient. A surrogate-based 2D opti-
mization of wing sections was performed. Objective function values were obtained
by solving Reynolds-averaged Navier–Stokes equations. The optimal baseline wing
was constructed using the optimal airfoil sections obtained from the 2D optimiza-
tion. Subsequently, parametric studies were conducted using vortex lattice method to
arrive at an optimal wing by varying the wing twist distribution. The optimal airfoils
showed 2% and 4% reduction in drag coefficient compared to the corresponding
baseline airfoils. The optimal wing showed about 4% reduction in drag coefficient
compared to the baseline wing.
Keywords Transonic regime · Light business jets · Supercritical airfoil ·

Aerodynamic optimization
K. S. Rao (B) · M. Sivapragasam · H. K. Narahari

Technology, M.S. Ramaiah University of Applied Sciences, Bangalore, India
M. Sivapragasam
H. K. Narahari
A. V. Bharadwaj
Genser Aerospace & IT Pvt. Ltd, Bangalore, India

254 K. S. Rao et al.
Abbreviations
c Chord length (m)

Cd Airfoil drag coefficient
CD Wing drag coefficient
Cp Coefficient of pressure
C L /C D Lift to drag ratio
Cl Airfoil lift coefficient
CL Wing lift coefficient
M∞ Freestream Mach number
t/c Thickness to chord ratio
α Angle of attack (°)
1 Introduction
Transport aircraft capable of flying at high subsonic and transonic speeds have domi-
nated the civil aviation market since the 1950s [1]. Light jet aircraft, capable of
carrying 6–10 passengers operate over various ranges. There has been an increasing
demand for such class of aircraft. These aircraft cruise in the transonic regime where
the wing’s aerodynamic performance is limited by the presence of shock waves
leading to wave drag penalty.
This led to the development of supercritical airfoils which had higher drag diver-
gence Mach number compared to other conventional airfoil designs [2]. The aft
lower section of the supercritical airfoil has a concave shape, which contributes to
the overall C l , thereby reducing the loading on the upper surface of the airfoil. This
leads to a weaker terminal shock wave on the upper surface and hence, a lower wave
drag.
There is a possibility of further drag reduction of a supercritical airfoil to improve
its performance in transonic flight regime through optimization methods. In the
present study, a transonic wing was optimized which cruises at a Mach number of
0.79, at a C L of 0.30. The baseline wing consisted of 12% and 10% t/c supercritical
airfoils at the root and tip sections, respectively.
A family of NASA supercritical airfoils were represented using a set of geomet-
rically orthogonal basis functions. The weighting coefficients of the basis functions
were the design variables. A high-fidelity surrogate model was constructed in the
design space. Genetic algorithm was applied on the surrogate model to obtain optimal
airfoil shapes for minimizing C d . The optimal airfoils were used to construct a swept,
cranked-wing whose twist distribution was tweaked to arrive at an optimal wing. The
aerodynamic characteristics were evaluated.
Aerodynamic Optimization of Transonic Wing for Light Jet Aircraft 255
2 Methods
2.1 Computational Procedure
NASA SC(2)-0714 airfoil was chosen for validation studies. O-grid topology was
used to generate the grid around the airfoil in ANSYS ICEM-CFD. The domain size
was chosen after carrying out domain independence study. Boundary influence from
the far-field domain was found to be prominent even when the pressure far-field
boundary condition was imposed. The domain boundaries were fixed at a distance of
10c from the leading edge of the airfoil in the upstream, upward and downward direc-
tions of the airfoil. The downstream boundary was fixed at a distance of 20c from the
airfoil leading edge. This type of domain configuration will be represented as 10/20c
henceforth. Here c stands for the chord length of the airfoil. The drag coefficient (C d )
was obtained from RANS simulation. Similarly, computations were performed with
domains of 20/40c, 30/60c, 40/80c, and 50/100c dimensions. When this study was
performed, the grid around the airfoil in 10/20c domain was unchanged while moving
towards 20/40c. Similarly, this was applied during generating grids for subsequent
larger domains. This is a systematic way of performing a domain independence study,
without the grid affecting the variation in flow simulation results. From this study,
50/100c domain was chosen to be suitable for the present study as the change in C d
between 50/100c and 40/80c domains was only 1.72%. Also, the computational cost
estimated for the optimization process influenced the selection of 50/100c domain.
A grid-independence study was performed using the 50/100c domain, to choose a
computational grid containing about 2.5 × 105 cells for the present study. The steady-
state, compressible Reynolds-averaged Navier–Stokes equations were solved using
the finite-volume flow solver ANSYS Fluent. An implicit density-based solver using
the Roe-FDS was employed. Spatial discretization was done using the Green-Gauss
node based second order upwind scheme. Turbulence closure was achieved by using
Menter’s SST k-ω model. Pressure far-field boundary condition was imposed on the
domain extents and no-slip boundary condition on the airfoil. Computations were
performed for the conditions corresponding to the experiments of [3]. The pressure
distributions over the airfoil at C n = 0.8715 obtained from experiment and present
computation are compared in Fig. 1. Figure 2 shows a comparison between the
computational and experimental drag polar. A good agreement is seen validating the
computational procedure.
2.2 Airfoil Optimization
The airfoils to be used in construction of the transonic wing were optimized for
minimizing C d, at a fixed C l . To realize the effects of the optimized airfoil over the
wing, pressure distribution transformation is required. Hence, the cosine rule was
Fig. 1 Comparison of C p distribution
Fig. 2 Drag polar comparison

employed based on the half chord sweep location of the wing [4]. In this transfor-
mation, the C l , airfoil thickness to chord ratio (t/c) and effective Mach number from
the wing case were transformed for optimizing the 2D airfoil case. The transformed
values on the airfoil corresponding to the wing requirements were:
• C l = 0.43
• M ∞ = 0.72
• t/c at root = 13.22%
• t/c at tip = 11.019%
Based on these considerations the optimization problem can be formally stated
as follows:
Minimize: Cd
Subject to:
Cl = 0.43
M∞ = 0.72
In the present study a set of orthogonal basis functions were used to represent the
airfoil geometry.
A series of select supercritical airfoils were represented by a linear combination
of basis functions as:
j

n
j
Zi = bk Z ik (1)
k=1
j j
that best fitted the selected airfoils [5, 6]. Here, Z i are basis function coordinates, bk
the weighting coefficients and n the number of basis functions. In the present study n
was chosen to be 2 and the weighting coefficients b1 and b2 represent maximum t/c
and camber, respectively. The two basis functions are represented in Fig. 3 derived
by fitting the mean of the source airfoils with the NASA SC(2)-0412 airfoil.
The limits on the t/c of the airfoils were:
0.75 ≤ b1 ≤ 1.2
−1.5 ≤ b2 ≤ 1.5
corresponding to maximum t/c bounds of 9% ≤ t/c ≤ 14%.

Latin Hypercube Sampling technique [7] was used for sampling the design space.
The number of initial samples generally required to represent the design space well
is about 10 times the number of design variables. Since two design variables were
considered in the present study, the number of samples generated was 20. The aero-
dynamic properties of the sampled airfoils were evaluated using ANSYS Fluent at
a fixed C l = 0.43, M ∞ = 0.72 and Re = 10 × 106 . The constraint of fixed C l made
the evaluation of the objective functions expensive, as a minimum of three CFD
simulations were necessary to arrive at the required C l . A Kriging-based surrogate
Fig. 3 Basis functions
model [7] was constructed using these responses. The objective function landscape
of the surrogate model is shown in Fig. 4. The coordinate values of design variables
b1 and b2 are non-dimensionalized based on the maximum t/c bounds.
The quality of this surrogate model was estimated using the leave-one-out cross-
validation method [7]. The root-mean-square error was calculated to be 4.4 × 10–3 %.
Fig. 4 Objective function landscape

Fig. 5 Comparison of optimal and baseline airfoils. a Wing root, b wing tip
The surrogate model was subjected to genetic algorithm which resulted in optimal
airfoils at required t/c values. The optimal airfoil obtained for the root section is shown
in Fig. 5a along with the 14% t/c baseline airfoil. The optimal airfoil for the root
section had a C d value which was about 2% lower than the baseline airfoil. The
optimal airfoil for the tip section had about 4% lower C d than the 12% t/c baseline
airfoil. The optimal airfoil obtained for the tip section is shown in Fig. 5b along with
the 12% t/c baseline airfoil. The suction peak in the C p distribution was found to be
lower in the optimal airfoils, subsequently, a reduced shock strength.
Initially, the baseline wing was constructed using the 12 and 10% t/c baseline airfoils
at root and tip of the wing, respectively. A structured grid with 3.0 million cells was
generated for the computations. Computations were performed for the wing made of
baseline airfoils at C L = 0.3, M ∞ = 0.72 and Re = 10 × 106 . Later, the optimal airfoils
obtained from the 2D optimization study were transformed to the corresponding t/c
required on the wing. Computations were performed for the wing made of optimal
airfoils for the same flow conditions.
A parametric study was performed over the optimal baseline wing (wing made of
optimal airfoils). The parameter studied here was the wing twist distribution. This
was done for wing with optimal airfoil profiles in order to enhance its aerodynamic
performance further. The twist distribution was decided based on the spanwise lift
distribution on the wing. At the location of the trailing edge kink there was a sudden
dip in the load distribution. Hence, it was necessary to obtain higher lift at the kink
location, so that the span efficiency factor ‘e’ can be improved by driving towards
the ideal elliptic loading. This parametric study was performed in the low-fidelity
flow solver OpenVSP [8]. In the present study, vortex lattice method was used to
simulate the flow over the wing. Before performing the parametric study, the results
from the OpenVSP were compared with that of high-fidelity CFD data obtained from
ANSYS Fluent. The comparison of drag polar is shown in Fig. 6. The data from the
low-fidelity flow solver, OpenVSP was able to match the high-fidelity data.
An optimal wing configuration was achieved through the study, which had a wash-
in of 3.5° from the root to the kink location and a wash-out of 3.5° from the kink
to the tip of the wing. The comparison of the three wing configurations is shown in
Table 1. The shock strength listed in Table 1 is a ratio of increase in static pressure
across a shock wave to the freestream static pressure.
A comparison of the C P contours over the upper surface of the baseline and
optimal wings is shown in Fig. 7. The C P contour lines are more closely spaced near
Fig. 6 Comparison of drag polar from Fluent and OpenVSP
Table 1 Comparison of the three wings

Wing α˚ CL CD % CL /C D % Increase Shock %
Reduction in C L /C D strength Reduction
in C D in shock
strength
Baseline 1.10 0.3 0.01378 – 21.78 – 1.51 –
Optimal −0.19 0.3 0.01327 3.7 22.60 3.8 1.19 21.19
baseline
Optimal −2.19 0.3 0.01327 3.7 22.60 3.8 1.24 17.88
(a) (b)
Fig. 7 C p contour on upper surface. a Baseline wing, b optimal wing
the leading edge of the wing on the outboard sections of the baseline wing indicating
shock occurrence. This is not incurred in the optimal wing.
The Mach number contours over the upper surfaces of the baseline and optimal
wing are compared in Fig. 8. The maximum Mach number has reduced from 1.20 to
1.10 for the case of optimal wing. Consequently, the shock strength has reduced to
1.24 for the optimal wing when compared to 1.51 for the baseline wing.
The wing was divided into 20 equidistant stations along its span. The C P distribu-
tions at various spanwise stations are plotted in Fig. 9. The critical C P is also plotted
in these figures to indicate the occurrence and strength of the shock wave generated
over the wing sections. The C P plot comparisons at various spanwise stations show
a reduced peak Mach number for the optimal baseline and optimal wings. As seen
from Fig. 9e a favorable upper surface platform pressure gradient is observed in the
optimal baseline and optimal wings. This helps in attaining required C L at lower
drag due to a weaker shock. The shock strength is also reduced for the optimal and
optimal baseline wings as seen from the maximum C P above the C P critical line.
Computations were extended for a range of angles of attack at Mach number
of 0.79 for the three wings. The aerodynamic performances of the three wings are
compared in Fig. 10. There is an upward vertical shift observed in the C L versus
α curve, for the case of optimal wing compared to both the baseline and optimal
baseline wings. This indicates that a lower angle of attack is required by the optimal
wing to achieve the same C L . This aids in reducing the wing setting angle which
(a) (b)
Fig. 8 Mach number contour on upper surface. a Baseline wing, b optimal wing
has an advantage of lower form drag for the aircraft. The maximum C L attained
by the optimal baseline wing is higher than the baseline wing. The optimal wing is
found to have lower values of C D at more negative angles of attack compared to the
other two wings. This is mainly because the C L obtained at a given angle of attack
is higher for the optimal wing case. Hence, the associated induced drag component
is higher, which shows up in the total value of C D . The drag polar of the optimal and
the optimal baseline wings have a leftward shift at higher values of C L as seen in
Fig. 10c. This indicates that the drag acting on the wings for a given value of C L is
lower compared to the baseline wing. The C L /C D curve shifts more towards lower
values of α for the optimal wing case. The optimal wing performs better than the
baseline wing by achieving the same C L /C D at a much lower angle of attack. Also,
the maximum C L /C D achieved by the optimal wing is higher than the baseline wing
by 3.8%.
The spanwise lift distribution of the optimal wing is plotted in Fig. 11 along with
the equivalent elliptic loading. The optimal wing has a lift distribution much closer
to the equivalent elliptic loading at a C L of 0.3. This helps to minimize the induced
drag component of the wing. Hence, the optimal wing performs better even at higher
C L ranges.
(a) (b)
(c) (d)
(e) (f)
Fig. 9 C P distributions at Z(/b/2) = a 0.10 b 0.30 c 0.35 d 0.40 e 0.50 f 0.80

(a) (b)
(c) (d)
Fig. 10 Comparison of: a C L versus α, b C D versus α, c drag polar, d C L /C D versus α
4 Conclusion
A surrogate-based optimization procedure was employed to obtain optimal airfoils

for minimizing C d . The optimal airfoils had about 2% and 4% lower C d compared
to corresponding baseline airfoils at the root and tip sections, respectively. Optimal
baseline wing was constructed using the optimal airfoils. The optimal baseline wing
was tweaked to obtain an improved spanwise lift distribution, by varying the wing
twist angle. The spanwise lift distribution was found to be close to the equivalent
elliptic lift distribution in the optimal wing as against the baseline wing. The optimal
wing had about 4% lesser C D than its baseline counterpart. An increase in the Oswald
span efficiency factor of around 0.4% was obtained for the optimal wing as against the
baseline. The aerodynamic advantages gained by the optimal wing over the baseline
wing makes it a good contender for medium/long range transonic flights.
Fig. 11 Spanwise C L distribution of optimal wing
References
1. Vos R, Farokhi S (2015) Introduction to transonic aerodynamics. Springer, London

2. Whitcomb RT, Clark LR (1965) An airfoil shape for efficient flight at supercritical Mach
numbers. NASA-TM-X-1109, L-4517. NASA, Washington
3. Jenkins R, Hill A, Ray E (1988) Aerodynamic performance and pressure distributions for a
NASA SC(2)-0714 airfoil tested in the Langley 0.3-meter transonic cryogenic tunnel. NASA
Technical Memorandum 4044
4. Zhao T, Zhang Y, Chen H et al (2016) Supercritical wing design based on airfoil optimization
and 2.75D transformation. Aerosp Sci Technol 56:168–182. https://doi.org/10.1016/j.ast.2016.
07.010
5. Robinson GM, Keane AJ (2001) Concise orthogonal representation of supercritical airfoils. J
Aircraft 38:580–583. https://doi.org/10.2514/2.2803
6. Keane AJ, Nair PB (2005) Computational approaches for aerospace design: the pursuit of
excellence. Wiley, Chichester, pp 429–479
7. Forrester AIJ, Keane AJ, Sobester A (2008) Engineering design via surrogate modelling: a
practical guide. Wiley, Chichester
8. OpenVSP (2018) https://openvsp.org/wiki/doku.php?id=start
Non-adiabatic Wall Effects on Transonic
Shock/Boundary Layer Interaction
Sahil Bhola and Tapan K. Sengupta
Abstract Direct simulations are carried out to investigate the influence of unsteady
heat flux transfer on transonic shock-boundary layer interaction; for flow past SHM-1
airfoil at a free-stream Mach number M∞ = 0.72 and angle of attack α = 0.38◦ . Flux
is added in a periodic manner through a region (8−18% of the chord ) located on the
suction side of the airfoil, with multiple values of exciter time period (Te = 2, 4) con-
sidered for our simulation. We show that addition of unsteady heat flux delayed shock
formation, along with significant modifications in it’s structure. The time-averaged
Cp distributions revealed a shift in the shock towards the aft, by approximately 5% of
the chord; along with an increased lift near the trailing edge, suggesting a nose-down
stabilizing influence. Primarily, it is noted that the additional heat flux resulted in an
overall increase of the aerodynamic efficiency (lift to drag ratio) by approximately
10%.
Keywords DNS · SBLI · Transonic flow
1 Introduction
The phenomenon of shock-wave boundary layer interaction (SBLI) in transonic flow

regime has significant influence on the aerodynamic and the thermodynamic design
through substantial modifications of the flow field. It involves several repercussions,
such as alteration of the wall pressure, modification of the highly viscous boundary
layer, flow unsteadiness, increased drag etc. An adequate understanding of this highly
time dependent phenomenon, by solving the compressible Navier-Stokes equations
S. Bhola (B)
Thapar Institute of Engineering and Technology, Patiala, India
T. K. Sengupta
High Performance Computing Laboratory, Department of Aerospace Engineering,
IIT Kanpur, Kanpur, India

268 S. Bhola and T. K. Sengupta
(NSE) is required to predict critical design factors such as shock location, aerody-
namic forces, along with efficient techniques for it’s control.
Several studies[1–4] have been carried out to understand the interaction of shock
wave (SW) with boundary layer (BL), however all these works implemented the adi-
abatic wall boundary conditions. Direct numerical simulations (DNS) to investigate
the influence of wall temperature on the shock wave and turbulent boundary layer
interaction for a wedge reflected shock, has been carried by Bernardini et al. [5] at a
free stream Mach number M∞ = 2.28 and wall-recovery temperature values (Tw /Tr )
of 0.5, 0.75, 1.0, 1.4, 1.9. It was reported that in case of cooling (Tw /Tr <1), there
is an increase of skin friction along with considerable reduction in the interaction
scales, whereas the opposite could be observed in case of wall heating (Tw /Tr >1).
Numerical simulations to investigate the effect of steady heat transfer on transonic
flows over NACA0012 airfoil using the thin-layer NSE with Baldwin-Lomax turbu-
lence model were reported by Raghunathan and Mitchell [6] , where surface cooling
resulted in decrease of the drag coefficients along with a fuller velocity profile, as
compared to the adiabatic wall case. Qualitative understanding of the control of the
periodic shock motion through heat transfer on a 14% bi-convex airfoil at a transonic
Mach number M∞ = 0.83, Re = 9 × 106 , α = 0◦ were reported by Raghunathan et
al. [7] , where transfer of energy from the mainstream to boundary layer (cold wall)
made the boundary layer more resistant to separation and reduced the amplitude of
shock oscillation. Prior scientific work pertaining to SBLI and it’s control through
heat transfer addresses the case of steady heat flux transfer across the wall and to
our knowledge, no high-fidelity simulations have been carried out to investigate the
effect of unsteady heat flux transfer across the wall on SBLI. The authors report
DNS results to investigate the influence of unsteady heat flux transfer on SBLI,
across a natural-laminar-flow (NLF) airfoil, the SHM-1 [8] profile. Several numeri-
cal schemes [9–14] have been proposed to accurately capture the dynamics of shock
wave in a compressible flow, and for the present work to attain near spectral accuracy
using relatively compact stencils, we use the dispersion relation preserving (DRP),
optimized upwind compact schemes [15–17]. It was shown by Sengupta et al. [18]
that these compact schemes which were originally developed for direct simulation
of incompressible flows, could be used to clearly capture the SBLI in case of com-
pressible transonic flows. Computational validation for flow past SHM-1 without
the unsteady flux transfer across the airfoil surface is done through the experimental
data reported by Fujino et al. [8] for the two dimensional transonic wind-tunnel test-
ing. Although, problems related to two-dimensional testing in transonic wind tunnels
[19, 20] such as wall interference, three-dimensional effects in two-dimensional tests
might affect the reliability of the data.
The manuscript is organized as follows. Followed by the introduction, the numer-
ical strategy and the flow conditions of the simulations are described in Sect. 2 and
the results of the DNS are presented in Sect. 3. Concluding remarks drawn from the
obtained results are finally provided in Sect. 4.
Non-adiabatic Wall Effects on Transonic Shock/Boundary Layer Interaction 269
2 Computational Setup
2.1 Flow Solver
To investigate the highly time dependent phenomenon of SBLI, the two-dimensional

(2D) NSE is solved in strong conservation form for a perfect compressible gas, with
the application of Prandtl analogy for thermal conductivity and Sutherland’s law to
compute the dynamic viscosity. Molecular heat conductivity in the flow has been
evaluated using the Fourier heat law, and Newtonian viscous stress components have
been used to compute the shear stress in the flow. The governing equations have been
discretized on a transformed curvilinear coordinate system (ξ, η) by introducing a
transformation [ξ = ξ(x, y), η = η(x, y)], using an in-house developed solver based
on finite-difference schemes, which has been previously validated by Sengupta et
al. [18] to compute 2D transonic flows over NACA0012 and SHM-1 airfoils. Con-
vective flux derivatives were evaluated through high accuracy optimized, implicit
upwind compact scheme [21] OUCS3 and its symmetrized version S-OUCS in the η
and ξ directions, respectively. These implicit schemes are highly robust, have supe-
rior stability, DRP [15, 16, 22] properties and provide near-spectral accuracy by
optimizing the numerical dispersion and dissipation in the spectral plane [17, 23].
To provide closure at the boundary and the near boundary points, explicit schemes
have been used, as proposed by Sengupta et al. [15]. Derivatives in the isotropic
viscous flux terms are evaluated using the second order central differencing (CD2)
scheme in self-adjoint form, preserving isotropic nature. The schemes are used in
parallel computing framework using Message Passing Interface (MPI) libraries, by
domain decomposition technique [24, 25] , which require overlap of the sub-domains
to attenuate non-physical growth near the sub-domain boundaries. Time integration
has been performed using a third-order, low-storage, explicit Runge-Kutta scheme
[26]. The governing equation in non dimensional, strong conservation form in the
physical (x, y)-plane is given by
∂ Q̂ ∂ Ê ∂ F̂ ∂ Eˆv ∂ Fˆv
+ + = + (1)
∂t ∂x ∂y ∂x ∂y
where the conservative variables are
Q̂ = [ρ, ρu, ρv, ρet ]T
The convective flux vectors, Ê and F̂ are
Ê = [ρu, ρu2 + p, ρuv, (ρet + p)u]T

F̂ = [ρv, ρuv, ρv 2 + p, (ρet + p)v]T
and the viscous flux vectors Eˆv and Fˆv are given by
Eˆv = [0, τxx , τxy , (uτxx + vτxy − qx )]T

Fˆv = [0, τyx , τyy , (uτyx + vτyy − qy )]T
In Eq. (1), the variables ρ, u, v, et , p represent the non-dimensional density, Carte-

sian velocity components, total specific energy, and pressure. All the terms were
non-dimensionalized√with scales such as the free stream density (ρ∞ ), free stream
velocity (U∞ = M∞ γ RT∞ ), free stream temperature (T∞ ), free stream dynamic
viscosity (μ∞ ), Length scale (L = c) and (c/U∞ ) for time; where γ denotes the heat
capacity ratio (1.4 for air). Stokes hypothesis (3λ + 2μ = 0), where λ and μ are the
Lame’s coefficient has been assumed, thereby neglecting the effect of molecular bulk
viscosity while computing the viscous flux components. The heat conduction terms
involved are given by
μ ∂T
qx = −
PrRe∞ (γ − 1)M∞
2 ∂x
μ ∂T
qy = −
PrRe∞ (γ − 1)M∞ ∂y
2
μ ReM 2 γ RT
and are non-dimensionalized using ( ∞ L∞∞ ∞ ) scale. Governing equations in
the transformed curvilinear coordinate system, written in strong conservation form,
preferred for suitable capturing of shocks [16] is given by
∂Q ∂E ∂F ∂Ev ∂Fv
+ + = + (2)
∂t ∂ξ ∂η ∂ξ ∂η
where the corresponding state variables and flux vectors are given by
Q = Q̂/J
E = (ξx Ê + ξy F̂)/J
F = (ηx Ê + ηy F̂)/J
Ev = (ξx Eˆv + ξy Fˆv )/J
Fv = (ηx Eˆv + ηy Fˆv )/J
where J is defined as the Jacobian of the transformation and is given by
1
J =
xξ yη − xη yξ
The grid metrics terms ξx , ξy , ηx , ηy are computed as
ξx = Jyη ; ξy = −Jxη ; ηx = −Jyξ ; ηy = Jxξ

Heat conduction terms in the transformed plane are given by

μ
qx = − (ξx Tξ + ηx Tη )
2
μ
qy = − (ξy Tξ + ηy Tη )
2
While performing the reported computations, we have not used any model for
transition and/or turbulence. However, to eliminate the spurious high wavenumber
(k) components, which may lead to numerical instability, we have used explicit two-
dimensional filters [16, 27] in the whole domain. The two-dimensional filter stencil
is given by,

m
an
ûi,j + α2 (ûi−1,j + ûi+1,j + ûi,j−1 + ûi,j+1 ) = (ui±n,j + ui,j±n )
n=0
2
where as per convention, 2 m represents the ‘order’ of filter and α2 is the two-
dimensional filter coefficient, which ranges from −0.25 to 0.25; where α2 = 0.25
indicates no filtering of the solution. To attenuate the numerical errors and the possi-
ble Gibbs’ phenomenon while computing high wavenumber components, a transfer
function [16] (Ti,j ) (defined in the k-plane) was chosen. This is given by

a0 + m i=1 ai [cos(mki hi ) + cos(mkj hj )]
T(ki hi ,kj hj ) =
1 + 2α2 [cos(ki hi ) + cos(kj hj )]
In the present computations, we have used a ’Sixth’ order filter to represent the
filtered quantities (û) in terms of unfiltered variables (u) in an explicit manner. For
moderate filtering of the solution, filter coefficient (α2 ) has been taken as 0.2. Data
filtering has been performed at the final stage of each ORK3 time-integration step
and to obtain the filtered physical quantities, the iterative technique of BI-CGSTAB
[28] have been used till the residue converges to 1 × 10−6 .
2.2 Flow Conditions and Computational Arrangement
Flow past the SHM-1 airfoil is computed using an orthogonal, body-conforming

O-grid topology which is generated using hyperbolic grid generation technique [17,
29] , with 959 points in azimuthal (ξ ) direction and 500 points in the wall-normal
(η) direction. This required the introduction of a mathematical cut-(ξ -direction) in
the computational domain, along which periodicity of flow variables is used. The
mapping from the physical plane to the transformed plane is one-to-one, with uniform
grid spacing in the transformed plane ( ξ = η = 1).
Fig. 1 Computational
domain and boundary
segments
A schematic view of the flow configuration investigated is shown in Fig. 1, which

shows a non-adiabatic region along the top surface of the airfoil, placed between
0.08c (ξi = 100) to 0.18c (ξf = 150). Location of this non-adiabatic region is chosen
so that it lies a head of the supersonic pocket developed in case of the completely
adiabatic wall; thereby, locally accelerating the flow, according to the ideal Rayleigh
flow. Exchange of heat flux in the non-adiabatic region takes place along the wall-
normal direction, and to avoid it’s discontinuity along the airfoil surface, a Gaussian
function is prescribed for the wall-normal heat flux (qη ) in the transformed plane as
∂q ξ −ξo 2
= qη = A(t)[exp( 10 ) ] (3)
∂η
where, ξo corresponds to the location of maximum flux transfer in the transformed

plane and A(t) is the time-varying periodic amplitude given by

−6 −7 2π t π
A(t) = 1.01 × 10 + 4.95 × 10 sin − (4)
Te 2
This spatio-temporal dependence of wall normal heat flux function ensure the flux
expenditure remained in the physical limits, with a mean transfer rate of approxi-
mately 422 W m−2 (Te = 2) and 217 W m−2 (Te = 4). This is a small perturbation.
The heat conduction terms in the non-adiabatic region are given by
qx = qη ηx
qy = qη ηy
and for the adiabatic wall region, there is no transfer of heat flux
qx = 0
qy = 0
On the airfoil surface, no-slip boundary condition is imposed in terms of velocity

components, given by
u=v=0
Assuming one dimensionality of the flow (in η-direction) at the far field, non-
reflective characteristic-like boundary conditions [30, 31] , based on the one-
dimensional Euler equations are enforced. Depending upon the sign of eigenvalues,
of the linearized 1-D Euler equations, flow at the inflow or at the outflow could either
be subsonic or supersonic. Characteristic variables are used to enforce the bound-
ary conditions at the free stream, where any information entering the computational
domain is set to it’s free stream values and the information leaving the computa-
tion domain, is extrapolated using the interior flow variables. The outer boundary
of the computational domain is located at approximately 22c from the airfoil sur-
face, for accurate implementation of the far-field boundary conditions and to avoid
contamination of the flow-field from the acoustic wave reflection at the boundaries.
Numerical simulations have been carried out at values of Te = 2, 4 and for a
completely adiabatic wall (without exciter); which and are labeled as SBLI-T2,
SBLI-T4, SBLI-U, respectively. For all the cases free-stream initial conditions are
implemented on all the flow variables, with free-stream Mach number M∞ = 0.72 ,
Re = 16.2 × 106 , α = 0.38◦ , T∞ = 222.36 K and ρ∞ = 1.101552 kg m−3 .
This section presents the results of the simulations with adiabatic wall conditions,
essentially validated with the flight test data reported by Fujino et al. [8] and also
computations carried out with the added unsteady heat flux. The instantaneous flow
field analysis along with time-averaged data analysis have been done to underline
the salient features of the transonic SBLI.
3.1 Validation of Methodologies for Compressible Flow

Calculation and Shock Capturing
For transonic flow past SHM-1 airfoil, without any unsteady flux transfer across
the wall (SBLI-U), numerically obtained pressure coefficients are compared with
computed and experimental
Cp -distribution for flow
around SHM-1 airfoil,
without exciter(SBLI-U).
Averaging is done over
t = 50–80
reference experimental values, and are shown in Fig. 2. To obtain a clear view of
the transonic compressible flow, we have time-averaged the computed solution over
meaningful intervals, removing the initial transient state resulting from introducing
the airfoil into the computational domain. A good agreement is observed between
the experimental and computed time-averaged pressure coefficients, with a slight
deviation in shock location, which could be accounted through an error of 5% [4] in
experimentally determined inflow Mach number. As reported by Sengupta et al. [18]
, these flows show significant time-dependent behavior prior to the formation of the
shock and such unsteadiness are augmented with shock formation, due to the physical
dispersion of high wavenumber components. Discontinuities of physical variables
across the shock excites wide spectrum of physical variables, consequentially one
could notice from Fig. 3, the downstream convection of pressure pulses which orig-
inated near the shock. It is to be further noted, that we did not observe any spurious
upstream propagating q-waves [16], which is in agreement that such unsteadiness
for high wave numbers is due to physical dispersion and not numerical dispassion
[18]. Instantaneous numerical Schlieren pictures for the selected flow conditions are
shown in Fig. 4. In this case, Robert’s cross edge detector model [32] is used as a
methodology to numerically visualize the strong density gradients developing within
the flow field.
From the instantaneous Schlieren images, we note a self-sustained oscillatory motion
of the shock wave with varying amplitude, on the suction side of the airfoil. Lee et
al. [33] identified this 2D transonic buffeting as a closed loop mechanism based on
the coupling between the shock and the trailing edge, through pressure waves. Due
to the physical dispersion at the shock discontinuity, pressure pulses are generated
at regular intervals, which propagate downstream within the boundary layer. At the
trailing edge, these pressure waves diffract and cause wake deflection, consequen-
tially generating upstream propagating pressure waves, termed as Kutta waves [34].
Fig. 3 Comparison of computed instantaneous Cp -distribution with experimental values for flow
around SHM-1 airfoil, without exciter (SBLI-U)
Fig. 4 Visualization of Kutta-wave generation at the trailing edge by Robert’s cross edge detector
model for flow around SHM-1 airfoil, without exciter (SBLI-U). The red line denotes the sonic line
These Kutta waves propagate outside the boundary layer, and while moving upstream
exhibit non-linear interaction with one another, adding on to their strength. Further
upstream, these waves merge into the shock-wave, resulting in an unsteady shock
motion and strength modulation [4, 33]. For a thermodynamic systems where the
total enthalpy is conserved, generation of vorticity (−
→
ω ) is an indication of the pres-
ence of entropy gradients across any streamline, as given by the Crocco’s theorem
[35],
−
→ T dS RT dp0
ω = =−
u dη up0 dη
where S is the entropy and p0 is the stagnation pressure. Such physical gradients
of entropy are inherently created in cases of shock formation, consequentially we
did notice the creation of rotational effects within the shear layer in the proximity of
shock foot as shown in Fig. 4. Generally, it is not possible to numerically capture these
phenomena as these excite a wide spectrum of spatial and temporal scales, however
the use of high fidelity numerical methods with near-spectral accuracy and superior
dispersion properties enable us to capture these with good resolution. For the SHM-1
airfoil used, a spatial resolution of sη = 2.702 × 10−5 has been used to provide near
spectral accuracy for the employed compact scheme [15, 16]. Computations have
been carried out at a temporal resolution of t = 2.5 × 10−6 , which has been used
to accurately capture the wide spectrum of spatial and temporal scales, by providing
better DRP properties and numerical stability [16] for the employed space-time
discretization schemes.
3.2 Influence of Unsteady Heat Flux
To provide an overview of the flow organization and a qualitative perception of the

influence of unsteady heat flux on SBLI, we report in Fig. 5, contours of mean Mach
number and density. It is to be noted that irrespective of addition of heat flux we could
observe the typical characteristics of transonic SBLI, identified by it’s mixed flow
nature, i.e presence of locally supersonic region ahead of the shock and a subsonic
region downstream of it. Furthermore, it could be observed that addition of heat
flux resulted in a marked dilation of the supersonic region along with a shift in the
shock location by approximately 5% of the chord further downstream, as also noted
from the time averaged Cp distributions reported in Fig. 6. It is evident from these
visualizations that the shock structure has been significantly modified, with a much
smoother pressure recovery observed, as compared with the SBLI-U case.
The Gaussian function imposed on the wall-normal heat flux is revealed in Fig. 6,
where the flow experiences a stronger favorable pressure gradient, and is locally
accelerated as a consequence of increase in kinetic energy, which is derived from
the surge in internal energy due to flux addition. Use of a smooth Gaussian function
for flux transfer could be clearly elucidated with the fact that we do not observe any
sudden discontinuity in the physical variables at the boundary of the non-adiabatic
region, which could potentially be sites for Gibbs’ phenomenon. At the aft region of
the airfoil, we could note that there is an increase in suction, whereas the pressure
distribution on the bottom surface showed minimal variations. These pressure distri-
butions suggested that under the same flow parameters, addition of unsteady heat flux
resulted in an increased lift near the trailing edge of the airfoil in conjunction with a
stabilizing pitch-down moment. By performing time-accurate simulation we are able
to capture the unsteady nature of flow separation and the formation of the laminar
separation bubble at the shock foot due to the fluctuations in longitudinal pressure
Fig. 5 a Computed mean Mach contours, b computed mean density contours for M∞ = 0.72,
Re = 16.2 × 106 , α = 0.38◦ . Time averaging is done over t = 50–80
gradients. These pressure variations over the airfoil surface resulted in the formation
of multiple separation points, which are broadly identified in the near vicinity of the
shock foot and the trailing edge, as depicted through instantaneous Cf variations in
Fig. 7.
The flux amplitude function A(t), as given by Eq. (4) has been chosen, such that
there is continuous heat flux addition from across the airfoil surface to the flow. The
periodic nature of the amplitude function could be reflected through the distribution
of instantaneous wall temperature (Tw ) on the suction side of the airfoil as reported
in Fig. 8. Physically, the maximum flux transfer rate peaked to approximately 1772
W m−2 (SBLI-T2) and 886 W m−2 (SBLI-T4), as a consequence of which we could
note strong amplifications in the wall temperature distribution. However, it was fur-
ther noted that due to the strong convection of heat energy ( ξ dominated ), these
fluctuations were not of periodic nature themselves and could reach a maximum
value of 2223.6 K approximately at around 13% of chord. As previously discussed,
there exists a self sustaining closed feedback mechanism for shock oscillation on
the surface of the airfoil due to which the pressure and shear forces become highly
computed and experimental
Cp -distribution for flow
around SHM-1 airfoil.
Averaging is done over
t = 50–80
time-dependent. Impulsively introducing the airfoil into the computational domain

with a uniform initial flow condition excites a wide band of spectrum in the spectral
plane and produces results that are physically not relevant. In most cases, it took
about t 20 for the flow to develop into a meaningful viscous state. Hence, in order
to evaluate the time evolution of the loads acting on the airfoil we show the computa-
tional results from t = 20 to t = 80 which physically correspond to a time duration
of 0.278 s. The coefficient of lift (Cl ) and drag (Cd ) are computed by integrating
the static pressure and shear stress over the airfoil surface. Addition of unsteady
heat flux introduces a time scale into the flow which alters the pressure and shear
force distribution over the airfoil, as one could observe through Figs. 9a, 10a and
11a, which report the instantaneous Cl , Cd and Cf excursions, respectively. To bring
out the contrast between the load distribution signature, corresponding fast Fourier
transformation (FFT) are shown in Fig. 9b, 10b, 11b. These FFT reveal additional
heat flux modulated the natural frequency of the signal, as marked in the figures.
Slight variations could be noted in case of Cl and Cd spectrum, however in the case
of Cf , we observe a surge in amplitude of the dominant frequency. These variations
suggest that the major effect of heat flux on drag is through skin friction variation,
which is a function of dynamic viscosity (temperature dependent). Table 1 shows the
consolidated mean load values along with the lift to drag ratios, to comprehend the
influence of heat flux on pressure and shear forces. We do notice that there is a sig-
nificant increment in the mean lift coefficient (C̄l ) by approximately 26% (SBLI-T2)
and 26.8% (SBLI-T4), as we could also observe through the Cp distributions reported
in Fig. 6. Overall, an increment in the lift to drag ratio was noted by approximately
8.9% (SBLI-T2) and 10% (SBLI-T4), as compared to the completely adiabatic wall
case, indicating an increase in the aerodynamic efficiency of the control surface.
Fig. 7 Instantaneous skin friction coefficient distribution on the suction side
4 Summary and Conclusion
The present paper has presented first-time results from a simulations of unsteady heat
flux transfer from an airfoil at M∞ = 0.72, α = 0.38◦ , Re = 16.2 × 106 . The objec-
tive is to augment the knowledge on transonic Shock-wave boundary layer interaction
(SBLI) by performing high-resolution, high-accuracy simulations to capture the phe-
nomenon involved in a transonic flow over the SHM-1 airfoil. First, computations
are performed with the adiabatic wall boundary condition, which have shown good
agreement of Cp distribution with flight test data [8] , with a slight underestimation
in the shock location. The instantaneous flow field analysis reveals a self sustaining
closed feedback mechanism of shock oscillation, as a consequence of physical dis-
persion and wake deflection. Typically, these highly time-dependent phenomenon
are difficult to resolve as they excite large spectrum of spatial and temporal scales,
however with the usage of compact schemes with near-spectral accuracy and good
DRP properties, these are captured nicely. Moreover, presented computations do
not require any model for transition and turbulence, except the two-dimensional
Fig. 8 Instantaneous wall temperature Tw distribution for flow around SHM-1 airfoil
Fig. 9 a Instantaneous Cl distribution and b FFT for flow around SHM-1 airfoil. t = 20–80
Fig. 10 a Instantaneous Cd distribution and b FFT for flow around SHM-1 airfoil. t = 20–80
Fig. 11 a Instantaneous Cf distribution and b FFT for flow around SHM-1 airfoil. t = 20–80
Table 1 Computed mean loads for viscous flow past SHM-1 airfoil. Averaging is done over
t = 20–80
Case C̄l C̄d C̄f Lift-drag
ratio (C̄l /C̄d )
SBLI-U 0.2092 6.671 ×10−3 8.724 ×10−4 31.35
SBLI-T2 0.2634 7.710 ×10−3 9.134 ×10−4 34.16
SBLI-T4 0.2654 7.707 ×10−3 9.148 ×10−4 34.43
Padé filters to remove the spurious high wavenumber components that might lead to
numerical instability.
Computations performed with the unsteady heat flux transfer (SBLI-T2 and SBLI-
T4) reveal significant dilation of the supersonic region along with a shift in the shock
location by approximately 5% of the chord, downstream. The time averaged Cp
distribution further suggested a higher lift near the trailing edge, which provides a
nose-down stabilizing influence on the airfoil surface. By computing time-accurate
flow, we are able to capture the longitudinal pressure variations, which are responsible
for the creation of multiple sites of unsteady separation near the shock foot and at
the trailing edge of the airfoil. It is noted that primarily unsteady heat flux influences
the airfoil drag, through variation in the skin friction, which further depends on the
dynamic viscosity of the fluid medium. Overall, we are able to increase the lift to drag
ratio of the control surface by approximately 8.9% (SBLI-T2) and 10% (SBLI-T4),
reflecting a significant increment in the aerodynamic efficiency.
Hence, these computations lay down the groundwork to study the influence of
rapid unsteady heat rejection across a control surface on transonic SBLI. Typically,
they could be used to study the load fluctuations and the delay in shock formation
across a control surface in a stabilized altitude vehicle. Primarily, modifications in the
shock structure is studied by placing a non-adiabatic region on the airfoil surface; in
the subsonic flow regime ahead of the supersonic pocket. Since energy conservation
remains the primary objective of the aerospace industry, it is necessary to study the
flow control using minimal energy input. Further investigations are required to study
transonic SBLI using different duty cycles for the heat flux, unlike the sinusoidal
cycles used here which resulted in continuous addition of heat flux into the flow.
Additionally, multiple magnitudes of flux transfer rates and exciter location should
be considered to augment our knowledge on the outcomes of unsteady heat flux
transfer on SBLI.
References
1. Toure PSR, Schuelein E (2018) Numerical and experimental study of nominal 2-D shock-
wave/turbulent boundary layer interactions. In: AIAA 2018-3395, fluid dynamics conference,
3395
2. Gross A, Lee S (2018) Numerical analysis of laminar and turbulent shock-wave boundary layer
interactions. In: AIAA 2018-4033, fluid dynamics conference , 4033
3. Quadros R, Bernardini M (2018) Numerical investigation of transitional shock-wave/boundary-
layer interaction in supersonic regime. AIAA J 56:2712–2724
4. Hermes V, Klioutchnikov I, Olivier H (2013) Numerical investigation of unsteady wave phe-
nomena for transonic airfoil flow. Aerosp Sci Technol 25(1):224–233
5. Bernardini M, Asproulias I, Larsson J, Pirozzoli S, Grasso F (2016) Heat transfer and wall
temperature effects in shock wave turbulent boundary layer interactions. Phys Rev Fluids
1(8):084403
6. Raghunathan S, Mitchell D (1995) Computed effects of heat transfer on the transonic flow over
an aerofoil. AIAA J 33(11):2120–2127
7. Raghunathan S, Early JM, Tulita C, Benard E (2008) Periodic transonic flow and control.
Aeronaut J 112(1127):1–16
8. Fujino M, Yoshizak Y, Kawamura Y (2003) Natural-laminar-flow airfoil development for a
lightweight business jet. J Aircr 40(4):609–615
9. Chakravarthy S, Harten A, Osher S (1986) Essentially non-oscillatory shock-capturing schemes
of arbitrarily-high accuracy. In: AIAA 1986-339, 24th aerospace sciences meeting, 339
10. Ha Y, Kim CH, Lee YJ, Yoon J (2013) An improved weighted essentially non-oscillatory
scheme with a new smoothness indicator. J Comput Phys 232(1):68–86
11. Allaneau Y, Jameson A (2009) Direct numerical simulations of a two-dimensional viscous flow
in a shocktube using a kinetic energy preserving scheme. In: 19th AIAA computational fluid
dynamics, 3797
12. Chiu EKY, Wang Q, Jameson A (2011) A conservative meshless scheme: general order for-
mulation and application to Euler equations. In: 49th AIAA aerospace sciences meeting, 651
13. Ohwada T, Shibata Y, Kato T, Nakamura T (2018) A simple, robust and efficient high-order
accurate shock-capturing scheme for compressible flows: towards minimalism. J Comput Phys
362:131–162
14. Borges R, Carmona M, Costa B, Don WS (2008) An improved weighted essentially non-
oscillatory scheme for hyperbolic conservation laws. J Comput Phys 227(6):3191–3211
15. Sengupta TK, Ganeriwal G, De S (2003) Analysis of central and upwind compact schemes. J
Comput Phys 192(2):677–694
16. Sengupta TK (2013) High accuracy computing methods: fluid flows and wave phenomena.
Cambridge University Press
17. Sengupta TK (2004) Fundamentals of computational fluid dynamics. Universities Press, Hyder-
abad, India
18. Sengupta TK, Bhole A, Sreejith NA (2013) Direct numerical simulation of 2D transonic flows
around airfoils. Comput Fluids 88:19–37
19. Garbaruk A, Shur M, Strelets M (2003) Numerical study of wind-tunnel walls effects on
transonic airfoil flow. AIAA J 41(6):1046–1054
20. Binnion TW (1979) Limitations of available data. AGARD-AR 138
21. Dipankar A, Sengupta TK (2006) Symmetrized compact scheme for receptivity study of 2D
transitional channel flow. J Comput Phys 215(1):245–273
22. Sengupta TK, Ganerwal G, Dipankar A (2004) High accuracy compact schemes and Gibbs’
phenomenon. J Sci Comput 21(3):253–268
23. Sengupta TK, Dipankar A, Sagaut P (2007) Error dynamics: beyond von Neumann analysis. J
Comput Phys 226(2):1211–1218
24. Sengupta TK, Dipankar A, Rao AK (2007) A new compact scheme for parallel computing
using domain decomposition. J Comput Phys 220(2):654–677
25. Dolean V, Lanteri S, Nataf F (2002) Optimized interface conditions for domain decomposition
methods in fluid dynamics. Int J Numer Methods Fluids 40(12):1539–1550
26. Rajpoot MK, Sengupta TK, Dutt PK (2010) Optimal time advancing dispersion relation pre-
serving schemes. J Comput Phys 229(10):3623–3651
27. Sengupta TK, Bhumkar YG (2010) New explicit two-dimensional higher order filters. Comput
Fluids 39(10):1848–1863
28. Van Der Vorst HA (1992) Bi-CGSTAB: a fast and smoothly converging variant of Bi-CG for
the solution of nonsymmetric linear systems. SIAM J Sci Stat Comput 13(2):631–644
29. Bagade PM, Bhumkar YG, Sengupta TK (2014) An improved orthogonal grid generation
method for solving flows past highly cambered aerofoils with and without roughness elements.
Comput Fluids 103:275–289
30. Hoffmann KA, Chiang ST (1993) Computational fluid dynamics for engineers, vol 2. Engi-
neering Education System Wichita, KS
31. Pulliam TH (1986) Solution methods in computational fluid dynamics. In: Notes for the von
Kármán institute for fluid dynamics lecture series
32. Samtaney R, Morris RD, Cheeseman P, Sunelyansky V, Maluf D, Wolf D (2000) Visualiza-
tion, extraction and quantification of discontinuities in compressible flows. In: International
conference on computer vision and pattern recognition (2000)
33. Lee BHK, Murty H, Jiang H (1994) Role of Kutta waves on oscillatory shock motion on an
airfoil. AIAA J 32(4):789–796
34. Tijdeman H (1977) Investigations of the transonic flow around oscillating airfoils. Nationaal
Lucht-en Ruimtevaartlaboratorium
35. Liepmann HW, Roshko A (1957) Elements of gasdynamics. Courier Corporation
An Adjoint Approach for Accurate Shape
Sensitivities in 3D Compressible Flows
Srikanth Sathyanarayana, Anil Nemili, Ashish Bhole,

and Praveen Chandrashekar
Abstract This paper presents the development of a robust discrete adjoint approach
for accurate computation of shape sensitivities in three-dimensional inviscid com-
pressible flows. The adjoint Euler solver is generated by applying algorithmic dif-
ferentiation techniques to the underlying primal solver. The novelty of the proposed
framework is that the geometry subroutine that computes cell volumes, surface areas
and normals is integrated to the subroutine that performs the primal fixed point
scheme so that the adjoint code directly yields the desired shape sensitivities. The
applicability of the developed adjoint approach is demonstrated on ONERA-M6
wing test case. The consistency and accuracy of the adjoint solver are assessed by
comparing the adjoint shape sensitivities with the values from finite differences and
tangent linear code. Numerical results show that the adjoint residual inherits the
asymptotic rate of convergence of the primal residual.
Keywords Aerodynamic optimisation · Shape sensitivities · Discrete adjoints ·

Algorithmic differentiation · Compressible flows
S. Sathyanarayana (B) · A. Nemili

Department of Mathematics, BITS-Pilani, Hyderabad Campus, Hyderabad 500078, India
A. Nemili
A. Bhole
Université de Nice Sophia-Antipolis, C.N.R.S. U.M.R. & Inria Sophia-Antipolis, Biot, France
P. Chandrashekar
TIFR Centre for Applicable Mathematics, Bengaluru 560065, India

290 S. Sathyanarayana et al.
1 Introduction
Aerodynamic shape optimisation has been a subject of intense research for almost
three decades. Typically, these problems are characterised by a large number of shape
variables, also known as design or control variables. An efficient way of finding the
optimal set of control variables is by employing the gradient-based optimisation
algorithms. Central to the success of gradient algorithms is the accurate and efficient
computation of the sensitivity gradients of an objective function with respect to shape
variables. This information is then used in finding a descent direction that changes
the initial shape such that there will be an improvement in the design. This process is
then repeated until we achieve the desired optimal solution. One way of computing
the sensitivities is by using the finite differences [10] or the Complex Taylor Series
Expansion (CTSE) method [19]. These methods provide a simple and flexible means
of evaluating the desired sensitivities. However, both these methods are not practical
as the computational costs increase linearly with the number of shape variables.
Alternatively, control theory based techniques [13] can be employed to compute the
sensitivities by solving the continuous [11] or the discrete adjoint equations [4]. An
advantage of the adjoint approach is that the computational cost of solving the adjoint
equations and thus evaluating the sensitivities is bounded and also independent of
the number of shape variables.
In continuous adjoint methods, the adjoint versions of the governing (primal)
PDEs and the boundary conditions are first derived. The adjoint PDEs are then dis-
cretised and solved numerically. However, deriving adjoint equations for high-fidelity
simulation models can be mathematically very challenging. Another drawback is that
the sensitivities based on continuous adjoints may not be consistent with the values
obtained from primal numerical simulations [2]. On the other hand, in discrete adjoint
methods, the primal PDEs are first discretised. The adjoint equations and then the
adjoint solver are constructed by linearising the non-linear discretised residuals. The
linearisation can be performed either manually or by employing Algorithmic Dif-
ferentiation (AD) techniques [6]. An advantage of AD is that it performs the exact
linearisation of the residuals. This results in an adjoint solver that is always consistent
to the corresponding primal solver and therefore yields accurate shape sensitivities.
However, the computational cost of AD based discrete adjoint solver is more than
the continuous adjoint solver. Furthermore, these costs vary depending on the way
AD is employed.
In earlier works [5, 14], AD is used selectively to minimise the CPU and memory
requirements as black-box application of AD can lead to very poor performance of the
adjoint code. In this approach, AD is used to linearise the discrete residuals and to con-
struct the Jacobian matrix of the adjoint linear system. A lot of hand coding was still
required in the development of discrete adjoint codes. This is a laborious process and
is prone to errors. In fact, developing adjoint versions of industry standard CFD codes
by hand may take several years and their maintenance can become a tedious task.
With significant progress in the development of AD tools and their performance,
another line of research has gained considerable acceptance [18]. In this approach,
An Adjoint Approach for Accurate Shape Sensitivities in 3D Compressible Flows 291
as a first step, the adjoint codes are constructed by applying AD directly to the
underlying primal codes. This results in an adjoint solver that is always consistent
to the primal and therefore yields accurate sensitivities. Since the differentiation is
performed in a black-box fashion, the adjoint code demands substantial amount of
memory and run-time. In the second stage, the computational efficiency is enhanced
by employing various advanced techniques of AD like reverse accumulation and
checkpointing [6]. Computationally, these adjoint codes are still found to be more
expensive than the hand-discrete codes. Numerical experiments have shown that the
run-time of fully AD based adjoint codes is around 4–10 times more than their primal
codes. However, the lack of computational efficiency of AD based adjoint solvers is
well compensated by robustness, easy maintenance of adjoint codes and availability
of affordable computational resources. In fact, this approach has matured to the
level of applying to industry standard CFD codes to generate accurate and efficient
adjoint codes. [1, 9, 15]. It should be noted that any further enhancement in the
computational efficiency may require better understanding of the primal solver and
also the underlying AD tools. At times, a re-organisation of the primal code structure
may result in more efficient adjoint codes.
In this paper, we pursue the development of a robust and accurate discrete adjoint
solver for accurate computation of shape sensitivities in three-dimensional inviscid
compressible flows. The primal Euler code is organised in such a way that AD results
in an efficient discrete adjoint Euler code that directly yields shape sensitivities.
This paper is organised as follows. Section 2 presents the governing inviscid fluid
flow equations. Section 3 presents the details pertaining to the primal numerical
solver. Section 4 presents the development of an adjoint approach for accurate shape
sensitivities. Numerical results to validate the performance of the adjoint solver are
shown in Sect. 5. Finally, conclusions are drawn in Sect. 6.
2 Governing Equations
The equations that govern the 3D inviscid compressible flows are given by
∂U ∂Fx ∂Fy ∂Fz

+ + + =0 (1)
∂t ∂x ∂y ∂z
Here, U is the vector of conserved variables, Fx , Fy and Fz are the flux vectors
along the coordinate directions x, y and z directions respectively. These quantities
are defined by
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
ρ ρu1 ρu2 ρu3
⎢ρu1 ⎥ ⎢ p + ρu2 ⎥ ⎢ ρu1 u2 ⎥ ⎢ ρu1 u3 ⎥
⎢ ⎥ ⎢ 1 ⎥ ⎢ ⎥ ⎢ ⎥
⎢
U = ⎢ρu2 ⎥ , Fx = ⎢
⎥
⎢ ρu1 u2
⎥ , Fy = ⎢ p + ρu2 ⎥ , Fz = ⎢ ρu2 u3 ⎥
⎥ ⎢ 2 ⎥ ⎢ ⎥
⎣ρu3 ⎦ ⎣ ρu1 u3 ⎦ ⎣ ρu2 u3 ⎦ ⎣ p + ρu32 ⎦
ρe (p + ρe) u1 (p + ρe) u2 (p + ρe) u3
(2)
Here, ρ is density, p is pressure, u1 , u2 , u3 are the components of fluid velocity along

the coordinate directions x, y, z respectively. e is the total energy per unit mass,
given by
p 1 2
e= + u + u22 + u32 (3)
ρ (γ − 1) 2 1
3 Primal Numerical Solver
The UG3 code is based on a finite volume method for the compressible Euler and
Navier-Stokes equations on hybrid, unstructured grids. It is capable of operating on
a mesh comprised of tetrahedra, pyramids, prisms and hexahedra. The discretization
is based on vertex-centered scheme which means that all solution variables are stored
at the vertices of the mesh, also called the primary mesh. A dual mesh is constructed
from the primary mesh on which the conservation law is satisfied. The main task is to
estimate the inviscid and viscous fluxes across the faces of the dual mesh. For inviscid
fluxes, we make use of Riemann solver type fluxes and also those based on central
schemes. The viscous fluxes are computed in a central manner with a compact stencil
consisting of only the first neighbours of any vertex, similar to a P1 Galerkin method.
To achieve high order accuracy, a linear reconstruction of some primitive variables
is performed in combination with limiters in case of discontinuous solutions. Edge-
based MUSCL-type limiters and multi-dimensional limiters of Barth-Jesperson and
Venkatakrishnan are available in the code. The code is essentially 3D but can also
solve 2D problems by extruding a 2D mesh in the third direction and using periodic
boundary conditions in the third direction.
3.1 Grid and Cell Topology
The grid is composed of non-overlapping polyhedral cells like tetrahedra, pyramid,

triangular prism and hexahedra, which will be referred to as the primary cells. The
unknowns are stored at the vertices of the mesh around which a dual cell is con-
structed. The dual cell is obtained by joining the mid-point of edges with the average
points of faces and cells. Within each primary cell, the geometric entities comprising
the cell like vertices, edges and faces have a local numbering which is based on the
convention in Gmsh and helps us to compute various geometric quantities required
in the finite volume method. The dual cell is obtained by joining the arithmetic center
of each cell, face and edge as shown in Fig. 1. Note that the faces of dual cells are
comprised of triangular surfaces. Let
Fig. 1 Definition of faces of dual cells
N (i) = {all vertices j connected to vertex i by an edge}

C(i) = {all primary cells containing vertex i}
F(i, c) = {all interior dual faces of vertex i inside primary cell c}
B(i) = {all boundary dual faces containing vertex i}
Each element of F(i, c) and B(i) is a triangular face as shown in Fig. 1. For a
tetrahedral grid, we also define
G(i, c) = {face of cell c opposite to vertex i}
3.2 Numerical Scheme
The finite volume scheme is based on satisfying the conservation laws on each control
volume, which in our case are the dual cells constructed around each vertex. The
semi-discrete finite volume scheme can be written as
dU i
Vi + F̂ij Sij + F̃f Sf = Ĝ f Sf + G̃ f Sf (4)
dt j∈N (i) f ∈B(i) c∈C(i) f ∈F(i,c) f∈B(i)
The second term on the left is the inviscid flux across interior faces while the third term
contains boundary fluxes. The numerical approximation of differential operators on
unstructured grids is described in [7], see also [12]. The inviscid flux F̂ij is computed
using a numerical flux function
F̂ij = F̂(Qij , Qji , nij )
where Qij , Qji are primitive variables at the middle of the edge ij obtained by some
reconstruction scheme. The primitive variables are taken to be
Q = [u1 , u2 , u3 , ρ, p] (5)

In the code we have a variety of MUSCL type reconstruction schemes along with
limiters for computing flows with shocks and other discontinuities. The reconstruc-
tion process required the knowledge of the gradient of primitive variables at the
vertices which is obtained by applying the divergence theorem over the dual cell in
a very careful manner to account for hybrid grids leading to a scheme which is exact
for linear functions even on hybrid grids. The above equations resulting from the
semi-discrete finite volume method are a system of non-linear ordinary differential
equations. For a time accurate solution, we integrate these equations using a Runge-
Kutta schemes which are strong stability preserving while for steady state solutions,
we use local time stepping to accelerate convergence to steady state.
4 A Discrete Adjoint Approach for Shape Sensitivities
Consider the problem of finding an optimal shape that maximises or minimises an

objective function of particular interest in aerodynamic applications. This amounts
to a PDE-constrained optimisation problem. A general formulation of this problem
can be given as
max/min J (U, α)
(6)
subject to C (U, α) = 0
where J is the objective function, U is the conserved vector, α is the vector of

control variables comprising the shape coordinates (x, y, z). In the present work,
C (U, α) = 0 represents the inviscid Euler equations in Eq. (1) along with boundary
conditions. In the semi-discrete form, the constraints can be written as
dU
+ Rs (U, α) = 0 (7)
dt
Here, Rs (U, α) is the discrete residual vector obtained after the finite volume dis-
cretisation of the spatial derivatives in the governing equations. Approximating the
temporal derivative using the first order forward difference formula, the state-update
formula can be written as
U n+1 = U n − tRs (U, α) (8)
Since we are interested in the steady state solution, at each pseudo-time iteration n,
the residual equations are solved for the solution of the state vector U using a fixed
point formula
U n+1 = G U n , α (9)
Here, G represents an iteration of the finite volume scheme employed for the numer-
ical solution of the 3D Euler equations. The above fixed point iteration converges to
the steady state solution U, given by
U = G (U, α) (10)
In the discrete form, the optimisation problem defined in Eq. (6) can then be written
as
max/min J (U, α)
(11)
subject to U = G (U, α)
The Lagrangian functional associated with the above constrained optimisation prob-
lem is given by
L = J (U, α) − T {U − G (U, α)} (12)
Here is the adjoint state vector or the Lagrangian multiplier. The first order nec-
essary conditions (KKT conditions) for optimality of the Lagrangian function are
given by
∂L
=0 (State equations) (13a)
∂
∂L
=0 (Adjoint equations) (13b)
∂U
∂L
=0 (Control equation) (13c)
∂α
From Eq. (13b), the discrete adjoint equations can be derived in the fixed point form
as
∂G ∂J
n+1 = T n + T (14)
∂U ∂U
From this equation, it is clear that the adjoint iterative scheme requires the converged
flow solution U to evaluate G and J . Therefore, to solve the adjoint equations the
primal solution should be available a priori. A general notation for the adjoint fixed
point iterative scheme can be written as

n+1 = G n , U, α (15)
where G represents a pseudo-time iteration of the discrete adjoint solver for 3D Euler
equations. The above fixed point iteration converges to the adjoint solution , given
by
= G (, U, α) (16)
The solution of the adjoint equations is then substituted in Eq. (13c) to evaluate the
sensitivities of the objective functional with respect to the shape variables as
dL ∂J ∂G
= + T (17)
dα ∂α ∂α
From Eqs. (14) and (17), it is clear that accurate computation of the shape sensitivities
require the exact differentiation of J and G. Note that the primal fixed point iterator
G consists of discrete residuals due to higher order convective scheme and limiters.
The exact differentiation of these terms by hand is laborious and prone to errors. Any
approximation made by neglecting the differentiation of these terms will result in
inaccurate computation of sensitivities [17]. One way to circumvent this difficulty is
by employing Algorithmic Differentiation (AD) techniques. An advantage of AD is
that it performs the exact differentiation of these terms with much ease.
In the present work, at the first step, AD is applied in a black-box fashion to
differentiate the computer code that represents the primal 3D Euler solver with respect
to the vector of shape variables α. Since all terms in the primal solver are differentiated
exactly, the adjoint solver is always consistent to the primal solver and therefore gives
accurate sensitivities at any residual level achieved by the primal solver.
A major drawback of the black-box AD approach is that the discrete adjoint code
demands expensive memory and computational time. Therefore, adjoint calculations
on practical configurations involving millions of grid points can become an infeasible
task. Algorithms 1 and 2 show the general structure of the primal Euler solver and the
corresponding black-box AD based discrete adjoint Euler solver. From Algorithm 2,
it is clear that the excessive demand for computational resources is due to the black-
box application of AD to the primal fixed point iterative solver. In this approach, the
adjoint code stores the primal solutions U n for all N iterations that yield a desired
accuracy and convergence. The stored solutions are then used to compute the adjoints
in the reverse sweep. Note that for large values of N , the storage costs can become
significant.
One way to circumvent the storage of primal solutions is by employing the reverse
accumulation technique [3]. This approach makes use of the iterative structure of the
adjoint fixed point scheme in Eq. (15). From this equation, it is clear that the adjoint
solver requires only the converged primal solution U. The structure of an efficient
adjoint solver based on this technique is shown in Algorithm 3. Here, M represents
the number of pseudo-time iterations required for the convergence of adjoint solution.
In [16], this approach has been successfully employed in accurate computation of
sensitivities in flows governed by incompressible URANS equations.
In the present work, the discrete adjoint Euler code UG3A is developed by algo-
rithmically differentiating the primal UG3 code using the AD tool Tapenade [8]. The
resulting adjoint solver precisely performs the adjoint fixed point scheme shown in
Algorithm 3. The novelty of the present approach is that the geometry subroutine that
computes cell volumes, surface areas and normals is integrated to the subroutine that
performs the primal fixed point scheme so that AD generated adjoint code directly
yields the desired shape sensitivities. To enhance the computational efficiency, the
primal code is organised in such a way that the computational requirements of the
adjoint code are as minimum as possible. Numerical investigations have shown that
the run time of the adjoint code is around a factor of 5 compared to the primal code.
The performance of the discrete adjoint Euler code in accurate computation of shape
sensitivities is demonstrated in numerical results.
Algorithm 1: Primal Euler Solver

Initialize U 0
for n ← 1 to n ≤ N do
perform the primal iteration U n+1 = G(U n , α)
end
return Objective function J n (U n , α)
Algorithm 2: Black-box AD based adjoint Euler solver

Initialize U 0
store (U n )
end
compute the objective function J n (U n , α)
Initialize N
for n ← N to n ≥ 1 do
perform the adjoint iteration n−1 = G( n , U n , α)
retrieve (U n )
end
∂J
return Shape sensitivities ddLα = ∂α + T ∂G
∂α
Algorithm 3: Adjoint Euler solver based on reverse accumulation

Initialize U 0
end
compute the objective function J n (U n , α)
Initialize N
for n ← 1 to n ≤ M do
perform the adjoint iteration n+1 = G( n , U, α)
end
∂J
return Shape sensitivities ddLα = ∂α + T ∂G
∂α
5 Numerical Results
In this section, we present the numerical results to demonstrate the performance of

the discrete adjoint solver in accurate computation of shape sensitivities. The test case
under investigation is the transonic flow past ONERA-M6 wing with Mach number
M = 0.84 and angle of attack AoA = 3.06◦ . The computational domain consists
of an unstructured grid with 72,791 points. The wall boundary is resolved with
30,635 grid points. For the discrete optimisation problem, the objective functions
are defined as the minimisation of inviscid aerodynamic drag coefficient and total
Table 1 Sensitivities of the drag coefficient with respect to the shape variable x
Index Finite differences Tangent linear code Discrete adjoint code
160 7.92357672529942E–05 7.92366912337142E–05 7.92366912337503E–05
170 –8.07455664830758E–05 –8.07444484751142E–05 –8.07444484751254E–05
1524 7.33930052149167E–04 7.33931859299297E–04 7.33931859299497E–04
25,930 4.82636153265048E–05 4.82641982087865E–05 4.82641982087881E–05
The bold values signify the accuracy of the values from discrete adjoint solver (third column) with
respect to the first and second column. It shows the extent till which the third column values match
with the first and second columns
Table 2 Sensitivities of the drag coefficient with respect to the shape variable y
160 7.00035862610803E–05 7.00033454518258E–05 7.00033454517270E–05
170 8.08812392061320E–05 8.08812570771250E–05 8.08812570771694E–05
1524 –5.81716619318939E–04 –5.81716715574034E–04 –5.81716715573846E–04
25,930 1.06139865993504E–03 1.06136000617902E–03 1.06136000617888E–03
Table 3 Sensitivities of the drag coefficient with respect to the shape variable z
160 7.72512453312002E–04 7.72514007628485E–04 7.72514007628787E–04
170 8.47840010004263E–04 8.47838916150582E–04 8.47838916150576E–04
1524 –4.33925430309845E–04 –4.33924914359014E–04 –4.33924914358905E–04
25,930 –6.35399718951212E–06 –6.35426243142565E–06 –6.35426243142373E–06
entropy produced in the computational domain. The control variables are the shape
coordinates that define the surface of the wing. This implies that at each wall point,
we have sensitivities due to x, y and z coordinates. In total, this results in 91,905
control variables.
Tables 1, 2 and 3 show a comparison of drag sensitivities obtained with finite
differences and AD based tangent linear and discrete adjoint solvers at 4 randomly
selected wall points. The point 160 is near the leading edge of the wing tip and the
point 170 is near the trailing edge of the wing tip. The point 1524 is located near the
leading edge of the midsection and 25,930 is near the trailing edge of the midsection.
Note that the tangent linear code gives accurate sensitivity information with respect
to only one control variable. Obviously, for large number of control variables this
approach is practically not feasible as the computational costs grow linearly with the
number of control variables. However, the tangent code is very useful in building
and validating the adjoint code.
2 2
0 0
-2 -2
Entropy sensitivty Cd sensitivty
Residue
Residue
-4 -4
-6 -6
-8 -8
-10 -10
-12 -12
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 0 1000 2000 3000 4000 5000 6000 7000 8000 9000
Iterations Iterations
Fig. 2 Transonic flow past ONERA-M6 wing. Comparison of the convergence histories of primal
and discrete adjoint residuals
From these tables it can be clearly seen that the shape sensitivities based on the
discrete adjoint code are in excellent agreement with the values obtained from the
second order finite differences with a step size of δ = 10−7 . Furthermore, the adjoint
sensitivities match even better with the values obtained from the tangent linear code.
It has been observed that the percentage error in sensitivities due to finite differences
and discrete adjoint approach is found to be varying between 10−3 and 10−4 . On the
other hand, the error in sensitivities based on the adjoint and tangent codes is less
than 10−11 , which is an outcome of the exact differentiation of the primal fixed point
iterator G. Fig. 2 shows a comparison of primal and adjoint residual convergence
history. It can be observed that the residues continue to decrease to machine zero.
Furthermore, the adjoint residual inherits the asymptotic rate of convergence of the
primal residual. Figure 3 shows the convergence of the objective function Cd and
its sensitivity with respect to the shape coordinate y at the point 25,930. Figure 4
shows the corresponding plots for the total entropy. From these plots we can argue
that the convergence of sensitivity gradients may require more number of iterations
compared to the objective function.
The pressure contours in Fig. 5 clearly show the λ-shock structure on the upper
surface of the wing. The contours of the sensitivities of the drag coefficient with
respect to the y coordinates of the shape are shown in Fig. 6. From the sensitivities
plot, it is clear that the maximum positive sensitivities occur at the trailing edge close
to the root of the wing. We can also observe mild negative sensitivities in the λ-shock
region. Figure 7 shows the contours of entropy produced on the wing surface. These
contours show that the maximum entropy production on this grid resolution occurs
near the leading edge close to the root of the wing. Figure 8 shows the contours of
the magnitude of the total entropy sensitivities. From this plot, it can be observed
that the dominant sensitivities occur near the leading edge region from midspan to
the tip of the wing. We can also observe mild sensitivities in the λ-shock region.
0.2 0.0015
0.15 0.001
Cd sensitivty w.r.t y
Cd
0.1 0.0005
0.05 0
0 -0.0005
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 0 1000 2000 3000 4000 5000 6000 7000 8000 9000
(a) (b)
Fig. 3 Transonic flow past ONERA-M6 wing. a Convergence history of the objective function Cd .
b Convergence history of the sensitivity of Cd with respect to the shape coordinate y at wall point
25,930
1000 150
Total entropy sensitivty w.r.t y
800 100
Total Entropy
600 50
400 0
200 -50
0 -100
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 0 1000 2000 3000 4000 5000 6000 7000 8000 9000
Fig. 4 Transonic flow past ONERA-M6 wing. a Convergence history of the objective function
total entropy. b Convergence history of the sensitivity of total entropy with respect to the shape
coordinate y at wall point 25,930
Fig. 5 Transonic flow past

ONERA-M6 wing. Pressure
contours on the upper
surface of the wing

ONERA-M6 wing.
Sensitivities of the drag
coefficient with respect to
the shape coordinate y on the
upper surface of the wing

ONERA-M6 wing. Entropy
contours on the upper
surface of the wing

ONERA-M6 wing.
Magnitude of the total
entropy sensitivities on the
upper surface of the wing
6 Conclusions
In this paper, we presented the development of a discrete adjoint approach for accu-
rate computation of shape sensitivities in inviscid compressible flows. The three-
dimensional discrete adjoint code was developed by algorithmically differentiating
an unstructured grid-based finite volume Euler code. The performance of the adjoint
solver in accurate computation of sensitivities of the drag coefficient and total entropy
with respect to the shape variables was assessed. Numerical results for the ONERA-
M6 wing test case have shown that the sensitivities based on the discrete adjoint code
were in excellent agreement with the values obtained from the tangent linear code
and second order finite differences. The run-time of the adjoint code is found to be
around a factor of 5 compared to the primal code.
In future, we want to extend the discrete adjoint solver to turbulent flows governed
by Reynolds-averaged Navier Stokes (RANS) equations. Research will also focus
on further enhancement in the computational efficiency of the adjoint solver.
Acknowledgements The authors gratefully acknowledge the financial support from the Science
and Engineering Research Board, Department of Science and Technology, Government of India,
under the project number EMR/2016/003182.
References
1. Albring T, Sagebaum M, Gauger N (2016) A consistent and robust discrete adjoint solver for
the Stanford University Unstructured (SU2) framework—Validation and application. Notes
Numer Fluid Mech Multidiscip Des 132:77–86
2. Carnarius A, Thiele F, Özkaya E, Nemili A, Gauger N (2013) Optimal control of unsteady
flows using a discrete and a continuous adjoint approach. IFIP Adv Inf Commun Technol
391:318–327
3. Christianson B (1994) Reverse accumulation of attractive fixed points. Optim Methods Softw
3:311–326
4. Elliott J, Peraire J (1997) Practical three-dimensional aerodynamic design and optimization
using unstructured meshes. AIAA J 35(9):1479–1485
5. Giles M, Ghate D, Duta M (2008) Using automatic differentiation for adjoint CFD code devel-
opment. CFD J 16(4):434–443
6. Griewank A, Walther A (2008) Evaluating derivatives: principles and techniques of algorithmic
differentiation. SIAM
7. Ham F, Mattsson K, Iaccarino G (2006) Accurate and stable finite volume operators for unstruc-
tured flow solvers. Annual Research Brief, Centre for Turbulence Research, Stanford University
8. Hascoët L, Pascual V (2013) The Tapenade automatic differentiation tool: principles, model,
and specification. ACM Trans Math Softw 39(3). https://doi.org/10.1145/2450153.2450158
9. Hay JA, Özkaya E, Gauger N, Schönwald N (2016) Application of an adjoint CAA solver for
design optimization of acoustic liners. In: AIAA paper 2016–2775
10. Hicks RM, Henne PA (1978) Wing design by numerical optimization. J Aircr 15(7):407–412
11. Jameson A (1988) Aerodynamic design via control theory. J Sci Comput 3:233–260
12. Khalighi Y, Ham F, Nichols J, Lele S, Moin P (2015). Unstructured large eddy simulation
for prediction of noise issued from turbulent jets in various configurations. Am Inst Aeronaut
Astronaut. https://doi.org/10.2514/6.2011-2886
13. Lions J (1971) Optimal control of systems governed by partial differential equations.
Grundlehren der mathematischen Wissenschaften. Springer. http://books.google.nl/books?
id=1TYgWwbmFGMC
14. Mohammadi B (1997) A new optimal shape design procedure for inviscid and viscous turbulent
flows. Int J Numer Meth Fluids 25(2):183–203
15. Nemili A, Özkaya E, Gauger N, Cramer F, Höll T, Thiele F (2014) Optimal design of active
flow control for a complex high-lift configuration. In: AIAA paper 2014–2515
16. Nemili A, Özkaya E, Gauger NR, Kramer F, Thiele F (2017) Accurate discrete adjoint approach
for optimal active separation control. AIAA J 55(9):3016–3026
17. Nielsen EJ, Anderson WK (1999) Aerodynamic design optimization on unstructured meshes
using the Navier-Stokes equations. AIAA J 37(11):185–191
18. Özkaya E, Gauger N (2010) Automatic transition from simulation to one-shot shape optimiza-
tion with Navier-Stokes equations. GAMM-Mitteilungen 33(2):133–147
19. Squire S, Trapp G (1998) Using complex variables to estimate derivatives of real functions.
SIAM Rev 40(1):110–112
Robust Flutter Prediction of an Airfoil
Including Uncertainties
A. Arun Kumar and Amit Kumar Onkar
Abstract This work presents the robust stability analysis of 2DoF airfoil by
including various uncertainties. These uncertainties arise due to several factors such
as modeling and manufacturing errors as well as disturbances in the flight condi-
tions. The approach adopted to study the uncertain aeroelastic system is based on
the structured singular value (μ-method). In this approach, the aeroelastic system is
formulated in a robust stability framework by parameterizing around dynamic pres-
sure and introducing uncertainties in the system parameters to account for errors and
disturbances. This results in the perturbed aeroelastic system which is then repre-
sented using Linear Fractional Transformation (LFT). Then, the nominal and robust
stability analysis of the perturbed aeroelastic system is carried out using μ method.
In this work, first the validation of μ method is done for 2DoF airfoil with quasi-
steady aerodynamics having uncertainties in the structural and aerodynamic proper-
ties. Further, the robust flutter boundary of 2DoF airfoil with Theodorsen’s unsteady
aerodynamics is studied using μ method in the presence of stiffness, damping, and
aerodynamic uncertainties.
Keywords 2DoF airfoil · Uncertainty · LFT · μ-method · Robust flutter
1 Introduction
Aeroelastic stability is one of the important criteria for the safety of flight vehicles.
Since the aeroelastic phenomena such as flutter and LCO can result in catastrophic
structural failure of the aircraft, these instabilities should be avoided in the whole
operational envelope. In order to predict these instabilities accurately, it is neces-
sary to consider uncertainties that arise due to the differences between the model
A. A. Kumar · A. K. Onkar (B)

CSIR-National Aerospace Laboratories, Bangalore 560 017, India
A. A. Kumar

306 A. A. Kumar and A. K. Onkar
and the actual flight vehicle, in the conventional aeroelastic analysis. In general, this
is handled by carrying out various parametric studies by enumerating all possible
combinations of uncertainties. This approach is relatively simple but can take enor-
mous computational hours and also possibility of omitting the worst-case situation
[1]. Recently, the concept of robustness from the control theory is introduced into the
aeroelastic analysis to study the flutter margins in the presence of uncertainties. In
this technique, the flutter analysis considering various uncertainties can be treated as
the robust stability problem under the perturbation of uncertainty set. This stability
problem can be solved by the structured singular value μ-method and proves as an
effective tool for robust aeroelastic studies [1].
2 Objectives
The objective of this work is to find the robust flutter boundary of 2DoF airfoil
with various structural and aerodynamic uncertainties. For this purpose, a state-
space model of 2DoF airfoil system is built along with perturbation in the system
parameters considered for uncertainties. Then, this is represented in LFT form as a
feedback interconnection between the perturbation in uncertainties and the nominal
state-space model. First, the present μ method is validated for 2DoF airfoil system
with quasi-steady aerodynamics having structural and aerodynamic uncertainties.
Further, the robust flutter boundary of 2DoF airfoil system is studied using μ method
with Theodorsen’s unsteady aerodynamics in the presence of various uncertainties.
3 Problem Formulation
In this study, a 2DoF airfoil system considered for the robust stability analysis is
shown in Fig. 1. Here, b represents the half-chord length, ba is the dimensional
distance between the centerline and the elastic axis, bx α is the dimensional distance
between the elastic axis and the center of gravity of the airfoil. The governing equa-
tions of motion of 2DoF airfoil system having heave (h) and pitch (α) degrees of
freedom can be expressed as [2]:

[M] Ẍ + [C] Ẋ + [K ]{X } = {Q} (1)
where [M], [C], [K] are the mass, stiffness and damping matrices respectively,
and {X} is the displacement vector whose components are given as:

m mxα b Ch 0 Kh 0 h
[M] = , [C] = , [K ] = , {X } =
mxα b Iα 0 Cα 0 Kα α
Robust Flutter Prediction of an Airfoil Including Uncertainties 307
b b
bxcg
ba bxα
L c.g.
V∞ M
α
Kα
Z h Cα
Kh Ch
X
Fig. 1 A typical 2DoF airfoil system
Here m, I α , K h , K α , C h and C α are mass, mass moment of inertia about elastic

axis, heave stiffness, pitch stiffness, heave damping and pitch damping respectively.
The expression for unsteady aerodynamic load vector in matrix form is given as
[3]:

{Q} = q [A1 ] Ẍ + [A2 ] Ẋ + C(k)[R] [S1 ] Ẋ + [S2 ]{X } (2)
where C(k) is the Theodorsen’s function, k = ωb/U is the reduced frequency and q
is the dynamic pressure. Here, ω and U are the frequency and freestream velocity,
respectively. Further, the matrices [A1 ], [A2 ], [S1 ], [S2 ] and [R] are defined as:

bS −π π ba S 0 −π b
[A1 ] = , [A 2 ] =
U 2 π ba −π b2 0.125 + a 2 U 0 −π b2 (0.5 − a)

−clα 0 S 1 b(0.5 − a) 01
[R] = , [S1 ] = , [S2 ] = S
0 cmα U b b2 (0.5 − a) 0b
where clα = 2π and cmα = 2π(0.5 + a). In the present study, the following
approximation is used for Theodorsen function as given in [4]:
0.165 0.335
C(k) = 1 − − for k < 0.5 (3)
1− k i
0.045
1 − 0.3
k
i
The above expression of Theodorsen’s function which is in reduced frequency

domain is converted to time domain by using Laplace transform. Here, aerodynamic
state variables are introduced resulting in additional aerodynamic degrees of freedom.
The final form of unsteady aerodynamic load vector is:

{Q} = q [A1 ] Ẍ + [A2 ] Ẋ + [R] [A3 ] Ẋ + [A4 ]{X } + [A S ]{X }a (4)
where

S 1 b(0.5 − a) S 01 S Ub c2 c1
[A3 ] = , [A 4 ] = , [A S ] = .
2U b b2 (0.5 − a) 2 0b b U c2 c1 b
The governing equation of aerodynamic state variables resulted from the Laplace
transformation of the Theodorsen’s function is:

Ẋ a = [B1 ]{X }a + [B2 ] Ẋ + [B3 ]{X } (5)
where

0 1 0 0 0 0
[B1 ] = U 2 U , [B2 ] = , [B3 ] = .
− b c4 − b c3 1 b(0.5 − a) 0U
The values of c1 , c2 , c3 and c4 are 0.10805, 0.006825, 0.3455 and 0.01365, respec-
tively. Substituting Eq. (4) in Eq. (1), the aeroelastic equation of motion is written
as:

[M] Ẍ + [C] Ẋ + [K ]{X }

= q [A1 ] Ẍ + [A2 ] Ẋ + [R] [A3 ] Ẋ + [A4 ]{X } + [A S ]{X }a (6)
4 Flutter Analysis Using μ Method
In this work, the nominal and robust stability of 2DoF system is solved by μ method
[2]. Here, the stability of nominal/robust aeroelastic system (P) having a set of unity
norm bounded uncertainties ={ : ∞ ≤ 1} is measured by structured singular
value μ(P) defined as [2]:
μ(P) = 1 min{σ () : det(I − P) = 0} (7)

∈
If there is no ∈ that causes det(I − P) = 0, then μ = 0. If μ(P) <

1, then the system is robustly stable under all unity norm bounded uncertainties.
The importance of μ method is that it results in less conservative stability margin
for systems having structured uncertainty descriptions when compared to analyzing
stability using small gain theorem [5].
4.1 Nominal Flutter Analysis
Initially, the nominal flutter analysis of 2DoF airfoil system is carried out using
the nominal (or mean) values of the structural and aerodynamic properties. Many
methods such as k-method, p-k method, p-method and μ-method are available for
nominal flutter analysis. In the current work, μ-method is chosen since it can be
directly extended to carry out robust stability analysis including various uncertainties.
In the μ method for nominal flutter analysis, the state-space model of the aeroe-
lastic governing equation is derived by introducing perturbation in the dynamic pres-
sure [2]. Then, this is represented in LFT form to perform nominal stability analysis
using μ method. Here, the μ analysis treats dynamic pressure as uncertain parameter
and searches for the minimum perturbation in this uncertainty that causes instability.
This minimum perturbation corresponds to the required minimum dynamic pressure
change that may result in aeroelastic instability.
Following the procedure described in [2], the total dynamic pressure q is defined
by considering an additive perturbation, δ q , on the nominal dynamic pressure, q0 as:
q = q0 + δq (8)
Equation (8) is then substituted in Eq. (6) and after appropriate simplifications
and grouping results in:
−1 −1 −1 −1
Ẍ = M C Ẋ + M K {X } + M q0 [R][A S ]{X }a + M {w}q
(9)
where

M = [M] − q0 [A1 ], C = q0 ([A2 ] + [R][A3 ]) − [C], K = q0 [R][A4 ] − [K ]
The feedback input {w}q to the dynamic pressure perturbation block is given by:
{w}q = δq {z}q (10)
where
−1
−1 [A1 ] M C + [A2 ]
{z}q = [A1 ] M K + [R][A4 ] {X } + Ẋ
+[R][A3 ]
−1 −1
+ [A1 ] M q0 [R][A S ] + [A S ] {X }a + [A1 ] M {w}q (11)
The LFT representation of the aeroelastic system state-space model [P] along
with the dynamic pressure perturbation is shown in Fig. 2.
Fig. 2 LFT representation

of the nominal aeroelastic
system
4.2 Robust Flutter Analysis
The robust flutter analysis of 2DoF airfoil system is carried out by considering
uncertainties in the damping and stiffness of the structural system as well as in the
aerodynamic coefficient matrix [R]. Following the similar derivation in the previous
section for nominal flutter analysis, the governing equations for robust flutter analysis
are presented. Consider the uncertainties in damping, stiffness and aerodynamic
coefficients matrices as [2]:
[C] = [C0 ] + [WC ][C ], [K ] = [K 0 ] + [W K ][ K ], [R] = [R0 ] + [W R ][ R ]

(12)
where subscript 0 represents nominal values of the respective structural and aero-
dynamic matrices; ([C ], [ K ], [ R ]) ∈ R2×2 represents the unity norm bounded
uncertainty operators and ([WC ], [W K ], [W R ]) ∈ R2×2 represents the weighting
matrices. For a given uncertain parameter, the weighting matrix along with its asso-
ciated uncertainty operator [] with ∞ ≤ 1 defines a range of values that should
be considered in the stability analysis [2]. Substituting these uncertainties (Eq. 12) in
Eq. (9) and grouping the feedback terms appropriately, after simplifications results
in:
−1 −1 −1
Ẍ = M C Ẋ + M K {X } + M q0 [R0 ][A S ]{X }a
−1 −1 −1 −1
+ M {w}q + M q0 {w} R − M {w}C − M {w} K (13)
The feedback inputs {w}q , {w} R , {w}C , and {w} K to the uncertainty block are
given respectively as:

{w}q = δq {z}q , {w} R = [ R ]{z} R
{w}C = [C ]{z}C , {w} K = [ K ]{z} K (14)
where
−1
−1 [A1 ] M C + [A2 ]
{z}q = [A1 ] M K + [R0 ][A4 ] {X } + Ẋ
+[R0 ][A3 ]
Fig. 3 LFT representation

of the aeroelastic system
including uncertainties
−1 −1
+ [A1 ] M q0 [R0 ][A S ] + [R0 ][A S ] {X }a + [A1 ] M {w}q
−1 −1 −1
+ [A1 ] M q0 + [I ] {w} R − [A1 ] M {w}C − [A1 ] M {w} K

{z} R = [W R ] [A3 ] Ẋ + [A4 ]{X } + [As ]{X }a

{z}C = [WC ] Ẋ , {z} K = [W K ]{X } (15)
The LFT representation of the aeroelastic system state-space model [P] including
perturbation in the dynamic pressure, structural, and aerodynamic uncertainties is
shown in Fig. 3.
First, the nominal and robust flutter velocities obtained from the present approach
are validated with those given in [6] for 2DoF airfoil considering quasi-steady aero-
dynamics. The input data considered for the analysis are [2]: m = 12.387 kg; I α =
0.065 kg m2 , b = 0.135 m, bx α = 0.033291 m, K h = 2844.4 N/m, K α = 3.525 Nm/rad,
C h = 27.43 kg/ s, C α = 0.036 kg m2 /s and ba = 0.0675 m. The uncertainties consid-
ered in the airfoil system for robust flutter analysis are: plunge damping (C h ) =
5 kg/ s, pitch stiffness (K α ) = 0.35 Nm/rad, lift coefficient (clα ) = 0.1 and moment
coefficient (cmα ) = 0.1. The nominal and robust flutter analysis are conducted at U
= 6 m/ s and ρ = 1.225 kg/ m3 .
Table 1 gives the comparison of nominal and robust flutter dynamic pressures
as well as frequencies for 2DoF airfoil obtained from the present approach with
the literature [6]. It is observed that there exists a good agreement between the two
Table 1 Comparison of flutter dynamic pressure and frequency for 2DoF airfoil system using
quasi-steady aerodynamics
Present results Reference [6]
q (Pa) ω (Hz) q(Pa) ω (Hz)
Nominal flutter 55.256 1.524 55.3 1.52
Robust flutter 41.693 1.415 41.8 1.41
q = 62.05 Pa
1.5 q = 41.69 Pa
q = 32.05 Pa
0.5
0
0 2 4 6 8 10 12 14 16 18
(rad/s)
Fig. 4 Variation of structured singular values (μ) with frequency (ω) at different dynamic pressures
for 2DoF airfoil system under quasi-steady aerodynamics with various uncertainties
results. The variation of structured singular value (μ) with frequency (ω) at various
dynamic pressures is also shown in Fig. 4. It can be observed that μ ≈ 1 is found at q
= 41.693 Pa with ω = 8.891 rad/s indicating the minimum destabilizing perturbation.
Next, the robust flutter boundary of 2DoF airfoil system due to uncertainties in the
stiffness, damping, and aerodynamic coefficients is studied based on Theodorsen’s
unsteady aerodynamic theory. Here, the nominal input data considered for the anal-
ysis are same as defined earlier except the position of elastic axis which is assumed
to be ba = −0.081 m and U = 12 m/s. The uncertainties in the stiffness (K α ) and
damping (C h ) are assumed to be 10% and 20% respectively [2]. The uncertainties
in the lift and moment coefficients (clα and cmα ) are assumed to be 5% and 10%,
respectively [7].
Table 2 shows the converged nominal and robust dynamic pressure of 2DoF airfoil
system based on unsteady aerodynamics. It can be observed that the worst flutter
dynamic pressure due to uncertainty in the aeroelastic system is about 17% lower than
the nominal flutter. The variation of μ with ω for different dynamic pressures is also
shown in Fig. 5. At q = 89.502 Pa, the value of μ is approximately 1 corresponding
to the frequency of 11 rad/s. Thus, the smallest perturbation in q to make the 2DoF
airfoil system unstable under the given set of parametric uncertainties is qr ob =
89.502 Pa.
Table 2 Flutter dynamic

Present approach
pressure and frequency of
2DoF airfoil system using q(Pa) ω (Hz)
unsteady aerodynamics Nominal flutter 107.518 1.959
Robust flutter 89.502 1.752
2.5
q = 107.518 Pa
q = 89.502 Pa
2
q = 74.212 Pa
1.5
0.5
0
0 4 8 12 16 20 24 28 32 36
(rad/s)
Fig. 5 Variation of structured singular value (μ) with frequency (ω) at different dynamic pressures
for 2DoF airfoil system under unsteady aerodynamics including various uncertainties
6 Conclusions
In this paper, the robust stability analysis of an airfoil is carried out considering uncer-
tainties in the aeroelastic system using structured singular value (μ) method. First, the
validation of μ method is carried out for 2DoF airfoil by considering uncertainties in
stiffness, damping and aerodynamic coefficients under quasi-steady aerodynamics.
The results obtained from the present approach agree well with those given in the
literature. Further, the method is extended by incorporating Theodorsen’s unsteady
aerodynamics to study the robust flutter characteristics of 2DoF airfoil system with
parametric uncertainties in stiffness, damping and aerodynamic coefficients. It is
observed that there is a reduction of 17% in the flutter dynamic pressure due to
uncertainties considered in the aeroelastic system.
References
1. Zhigang W, Chao Y (2008) A new approach for aeroelastic robust stability analysis. Chin J
Aeronaut 21:417–422
2. Lind R, Brenner M (1999) Robust aeroservoelastic stability analysis—flight test applications,
1st edn. Springer, London
3. Bisplinghoff RL, Ashley H, Halfman RL (1955) Aeroelasticity. Addison-Wesley, Cambridge
4. Fung YC (1955) An Introduction to the theory of aeroelasticity. Wiley, NewYork
5. Lind R, Brenner M (1998) Incorporating flight data into a robust aeroelastic model. J Aircraft
35(3):470–477
6. Akmese A, Comert, MD, Platin BE (2004) Aeroservoelastic analysis of missile control surfaces
via robust control methods. In: IFAC automatic control in aerospace. Elsevier IFAC publications,
Saint-Petersburg, Russia, pp 719–723
7. Sadraey M (2009) Robust nonlinear controller design for a complete UAV mission. VDM
Publishing
Effect of Vortex Generator on Flow
in a Serpentine Air Intake Duct
B. B. Shivakumar, H. K. Narahari, and Padmanabhan Jayasimha
Abstract Low Radar Cross Section (RCS) is one of the important requirements of
aircraft design for stealth. Rotating engine face components are one of the major
sources of radar reflection and hence need special attention. To this end, present day
aircraft air intake ducts are designed to hide the rotating components from the radar
signal by incorporating multiple bends. This type of intake ducts ensures that there is
no direct line of sight from the entrance of the duct to engine face components and are
generally called as Serpentine Ducts. Design of serpentine duct, therefore involves
sharp bends that are likely to cause flow separation and consequent instability and
distortion in the flow. In order to realize optimal performance from the engine, it is
necessary to reduce the losses and minimize distortion at the Aerodynamic Interface
Plane (AIP) that is near the outlet of the duct. This requires use of flow control
techniques to alleviate the effects of flow separation. A Serpentine duct described in
(Hamstra et al. in Active inlet flow control technology demonstration. ICAS, 2000
[1]) was selected for the present study, as the geometry details and experimental
results were reported in the paper. Based on this, a duct was created and CFD flow
simulations were performed. Results so obtained were compared with the test results
in order to establish a baseline duct. Subsequently, passive vortex generators of
trapezoidal shape were introduced and CFD simulations were carried out. The results,
in comparison with the base design, indicate an enhancement in the flow uniformity
at the AIP although with a 3% reduction in the pressure recovery. Further course of
study is also indicated.
Keywords Serpentine intake duct · Trapezoidal vortex generator · Passive flow

control
B. B. Shivakumar (B)
M.S. Ramaiah University of Applied Sciences, Bangalore 560 054, India
H. K. Narahari
Department of Automobile and Aeronautical Engineering, Faculty of Engineering and
Technology, M.S. Ramaiah University of Applied Sciences, Bangalore 560 054, India
P. Jayasimha
HAL-ARDC, Bangalore, India

316 B. B. Shivakumar et al.
Abbreviations
AIP Aerodynamic interface plane

CFD Computational fluid dynamics
DC Distortion coefficient
VG Vortex generator
Pto Free stream total pressure
PtAIP Total pressure at AIP
qAIP Dynamic pressure at AIP
Ps Static pressure
Pt Total pressure
1 Introduction
Research and development is an ongoing activity in the aerospace industry as

new requirements are continuously evolving. Contemporary advanced capabilities
are expected based on changing technological scenario. Extreme performance and
lowered detectability are of primary importance to Fifth Generation of combat
aircraft. These aircraft are characterized by high thrust-to-weight ratio and improved
stealth characteristics without compromising the performance qualities. Major reduc-
tion in the radar signal reflection is achieved by hiding the rotating parts of the engine,
since they are a main source of returning radar signal. These rotating components
are hidden from the radar signal by a multiple curvature duct so that there is no direct
line of contact from the entrance of the duct. Further, the radar signals that enter
the duct get reflected on to the duct walls and are absorbed by Radar Absorbing
Structure (RAS) specially designed for this purpose. This prevents reflection of the
radar signal from the intake.
Air intake duct is an important component of the aircraft propulsion system.
Principle objective of the intake duct is to ensure uniform airflow to the engine at
all operating conditions with maximum pressure recovery. However, the geometry
of the duct plays a major role in the quality of flow that ensues. Flow in serpentine
duct is complex in nature due to the effect of sharp bends between the inlet of duct
and the aerodynamic interface plane (AIP), which may be construed as the outlet
of the duct. As the flow moves towards a bend, there is a velocity difference in the
transverse direction due to the centrifugal forces acting on the flow, causing the outer
surface flow to accelerate and the inner surface flow to decelerate. After the bend, the
decelerated flow may not be able to overcome the adverse pressure gradient, leading
to flow separation. However, the separation bubble may close downstream. A similar
situation would exist at the second bend. If the AIP is close to be bend, the separation
bubble may not close leading to regions of low total pressure and consequently high
distortion index and low pressure recovery.
Effect of Vortex Generator on Flow … 317
A significant body of work already exists where a host of flow control devices and
techniques have been explored. Researchers have studied both active flow control
impinging micro-jets, and passive devices like guide vanes and vortex generator
(VG) attached to the duct wall. Encouraging results in the form of improved flow
quality at the AIP have been reported. Present study is to explore trapezoidal VG as a
passive flow control method mounted upstream of the bends to reduce the formation
of secondary flow. A few positions and orientations of the VG have been studied and
presented in this paper. It is proposed to continue the study further to optimize the
shape, size, location and orientation of VG.
2 Literature Review
Serpentine duct and vortex generators have been studied extensively, both in wind
tunnel and CFD. The results have been presented in [1] by Hamstra et al. Since the
base duct geometry and experimental results were available in this reference, it has
been selected as a basis for comparison in the present research work reported in
this paper. Effect of active flow controls such as micro-vane effectors and micro-jet
effectors has been reported. Logdberg [2] has clearly explained the effect of vortex
generator on the turbulent boundary layer and concluded that they were effective in
controlling the separation, albeit with increase in the drag. Experimental studies of
flow in S-Ducts have been reported in [3]. A parametric study of the vortex generator
on a flat plate, airfoil, and diffuser has been reported in [4]. It has been inferred that
flow mixing is accentuated by the formation of vortices due to VG. Chima et al.
[5] have analyzed through CFD, the Versatile Integrated Inlet Propulsion Aerody-
namic Rig (VIIPAR) located at NASA Glenn Research Centre. The rig consisted of
serpentine inlet, rake assembly, fan stage, exit rakes or probes and exhaust nozzle.
Experimental investigations have been done and pressure recovery measurements
compared with results from CFD-codes. The conflicts and compromises during the
design of fighter aircraft intakes have been explained in [6].
3 Geometry of the Serpentine Duct
A serpentine duct was designed by Lockheed Martin Aeronautics Company and

tested in NASA Glenn Research Centre for unmanned vehicle applications [1]. It
featured 2 bends of approximately 45° each, biconvex inlet shape with a ratio 4:1
(major axis of 490 mm and minor axis of 124 mm), a circular outlet of 254 mm
diameter and a total length of 632 mm. Based on that, a similar serpentine duct
was designed for the present research using CATIA, a commercial CAD software.
Figure 1 shows the baseline duct used for the present study. Since all the dimensions
were not available from [1], appropriate assumptions had to be made and therefore
the two ducts may not be identical.
Fig. 1 Baseline duct
4 Grid Independence Study
The initial step, for CFD simulation, was to discretize the domain using hexahedral
or tetrahedral elements. Accuracy of results obtained from finite volume method of
CFD depended to a large extent on the number of grids in the domain. Therefore,
it was necessary to carry out the grid sensitivity study to confirm that the grid size
selected for the given model was adequate to capture the flow physics with minimum
error. In grid sensitivity study, a series of grids were created starting with coarse
elements followed by systematic enrichment in x, y and z-directions. For each of
these grids, CFD simulation was performed and results tabulated. When results from
two successive grids did not vary by greater than 5%, the coarser of the two grids
was chosen for further computations. Present study started with a hexahedral grid
topology consisting of 90 elements along the circumference () of the outlet, 63
elements along the diameter (D) and 158 elements along the axis (L) of the flow.
Nature of the flow inside the duct being complex, with zones of separated and
secondary flows due to the bends, it was important to capture the flow near the
wall region. The often quoted standard parameter to assess proper resolution of
boundary layer flows is Y + . Mach number for the present research was 0.65 and
using corresponding density, frictional factor and shear stress values, Y + computed
turned out to be 5.
Hexahedral grids were generated using ANSYS-ICEM-CFD for the baseline
serpentine duct without Vortex Generators (VG), as shown in the Fig. 2. Table 1 lists
different grids used for simulation, giving details of grid size along each coordinate,
first layer thickness and total grids used for grid dependence study.
Fig. 2 Baseline duct without Vortex generator
Table 1 Grid sensitivity

Case Grid size First layer thickness Total number of
criteria
grids
G1 , D, L 4.00E−06 507,232
G2 , 2D, L 4.00E−06 1,173,216
G3 2, D, L 4.00E−06 1,277,760
G4 2, 2D, L 4.00E−06 3,066,624
G5 2, 2D, 2L 4.00E−06 5,068,800
Compressible flow simulation was carried out for the inlet total pressure
101,325 Pa and temperature 288.15 K. The outlet static pressure was varied till
the obtained inlet Mach number was 0.65. For the grid sensitivity study, K-ε stan-
dard turbulence model was used. Since the objective of the study was to improve the
flow at AIP, pressure recovery and distortion coefficient (DC) were calculated for all
5 grids and compared.
Pressure recovery was defined as the ratio of total pressure at AIP to the free
stream total pressure. Distortion coefficient was defined as the ratio of average total
pressure at AIP minus 60° sector minimum average of the total pressure at AIP to the
average dynamic pressure at AIP. Sector could be selected as 45°, 60° and 90° for
the DC calculation and based on the selection of sector angle, the index of distortion
coefficient is defined as DC-45, DC-60 and DC-90. For the present work, sector of 60°
was selected for DC-60 calculation. For discretization, 36 circumferential locations,
at 10° interval, were defined. For each circumferential location, 19 equidistance radial
points were created as stated in [1]. Total 684 points were created to measure the
distortion coefficient. Figure 3 shows the points created for the calculation of DC-60.
Fig. 3 684 points at AIP for

DC-60 calculation
Table 2 Grid independent study results

Results-Grid Independent Study
Coefficient G1 G2 G3 G4 G5
DC-60 (%) 21.97 21.04 24.69 17.96 17.78
Pressure recovery 0.97156 0.97115 0.97081 0.97160 0.97157
Equations used for the calculation of pressure recovery, DC-60 and static pressure
ratio were as follows:
Pressure recovery: PtAIP /Pt0

Distortion Index: DC-60: ((PtAIP − Pt. min 60 avg)/qAIP ) ∗ 100
Static pressure ratio: Ps /Pt0
Pressure recovery and DC-60 values were calculated for different grids, and
compared in Table 2.
From the grid sensitivity study, G4 (180 grid points along tangential direction, 126
grid points along the radial direction and 158 grid points along the flow direction)was
selected as adequate for the study.
5 Turbulence Model Study
Among the various turbulence models available in the commercial code ANSYS-
FLUENT, four models were selected for this study. Models chosen were Spalart-
Allamaras (SA) one equation model, Standard K-ω two equation model, Shear Stress
Transport(SST) K-ω two equation model and Realizable K-ε two equation model.
Figures 4 and 5 show the static pressure ratio on the center plane of top and bottom
surface of the duct for all 4 turbulence models compared with the experimental
Fig. 4 Static pressure ratio on the center plane of top surface
Fig. 5 Static pressure ratio on the center plane of the bottom surface
Table 3 CFD simulation results and comparison with [1] for different turbulence model
Coefficient Spalart-Allamaras K-Omega-standard SST-K-Omega Realizable Experimental
K-Epsilon results [1]
DC-60 24.35 29.67 24.93 58.84 38.00
[%]
Pressure 0.967 0.956 0.963 0.952 0.910
recovery
results. Experimental results of [1] were compared with CFD results for different
turbulence models in Table 3.
When the simulation results were compared with the experimental results, it was
observed that there were significant deviations in the simulation results. Similar
deviations were observed in [1] between CFD and experiment. Commenting on the
delayed prediction of flow separation on both top and bottom surfaces, the authors
have concluded that it is mainly attributed to the inadequacy of turbulence models.
Therefore, it was decided that the selection of the turbulence model would be based
on the closeness to the experimental results. Accordingly, it was decided to choose
the K-Omega Standard turbulence model for further study. It may also be noted
that the top contour could be different from the experimental contour as adequate
geometric information was not available for construction. Sensitivity to this factor
would be studied later.
6 Vortex Generator Study
From the baseline duct simulation, flow separation was noticed on the bottom surface
after the first bend and the top surface after the second bend. Initial studies indicated
a boundary layer thickness around 2.5 mm near the first bend. Hence VG dimen-
sions were chosen such that they are not submerged inside the boundary layer. The
trapezoidal VG of leading edge height 4 mm, trailing edge height 20 mm, length of
40 mm, thickness of 1 mm were the dimensions considered for the design of VGs for
the study. Figure 6 shows the VG dimension and geometry considered for the study.
Once the VG design was finalized, it was necessary to decide on the position of
the VG inside the duct. Since flow separation is noticed after the turning at the bend,
it was decided to place the VGs exactly at the beginning of the first bend. Six possible
locations and orientations of VGs have been studied, as indicated in Figs. 7, 8, 9,
10, 11 and 12. Figure 7 shows D1 design with a pair of 2 VGs located on top of the
serpentine duct equally positioned at a distance of 80 mm from the center axis.
Figure 8 shows the D2 design, which is similar to D1, but with VG placed at the
bottom surface. Figure 9 shows the D3 design with 3 pairs of VGs at the bottom
surface of the serpentine duct at throat. One pair of VG was placed exactly at the
center and remaining 2 pairs were equally positioned at a distance of 120 mm from
the center axis. Figure 10 shows the D4 design with 6 VGs and their orientation at
Fig. 6 Vortex generator geometry
Fig. 7 D1-serpentine duct with 2 pair of VGs at the top surface of the throat
the bottom surface of the serpentine duct at throat. Figure 11 shows the D5 design
which was similar to D4 but with VGs on the top surface of the duct. Figure 12 shows
the D6 design with 6 VGs and their orientation at the top and bottom surface of the
serpentine duct at throat.
For all 6 design cases, G4 type hexahedral grids were generated using ANSYS-
ICEM-CFD commercial code. Compressible flow simulation was carried out for an
entry Mach number of 0.65.
A comparison of pressure recovery and DC-60 values between baseline duct and all
the 6 design case is shown in Table 4. Placing two pairs of contra rotating VG on
the top surface at the entry plane (D1) indicated marginally higher distortion and a
little lower pressure recovery compared to the baseline duct without VG. Introducing
Fig. 8 D2-serpentine duct with 2 pair of VGs at the bottom surface of the throat
Fig. 9 D3-serpentine duct with 3 pair of VGs at the bottom surface of the throat
Fig. 10 D4-serpentine duct with 6 VGs at the bottom surface of the throat
Fig. 11 D5-serpentine duct with 6 VGs at the top surface of the throat
the same pair on the bottom surface (D2) made the distortion higher and pressure
recovery lower compared to the top position (D1). Adding one more pair, at the center
plane on the bottom surface (D3), again lowered level of distortion, but also lowered
pressure recovery in comparison with the baseline duct. In the next configuration,
three VG on either side (D4), with the leading edge towards the center line, showed
further lower levels of distortion, about 9% lower than baseline, and the pressure
recovery better than D3, but lower than the baseline by 2.7%. Next configuration
Fig. 12 D6-serpentine duct with 6 VGs at the top and bottom surface of the throat
Table 4 CFD results and comparison with experiment result

Parameters S-duct, without D1-top surface, D2-bottom D3-bottom
VG VG-2 surface, VG-2 surface, VG-3
DC-60 [%] 29.67 30.94 33.12 23.55
Pressure recovery 0.956 0.950 0.918 0.917
Parameters D4-Bottom surface, 6 D5-Top surface, 6 D6-Bottom and Top
VGs-oneside oriented VGs-Oneside oriented surface, 6 VGs-Oneside
oriented
DC-60 [%] 20.71 31.28 44.67
Pressure recovery 0.929 0.949 0.895
D5 tried was with six VG, similar to D4, but on the top side. The results were not
encouraging. So D6 configuration with 12 VG, six on top and six on the bottom side,
was tried. This led to substantially higher distortion and substantially lower pressure
recovery.
Obviously configuration D4 appears the best among the six configurations tried.
Further study would be around this configuration, with the effect of VG geometric
parameters varied.
The static pressure ratio along the centerline is presented in Figs. 13 and 14. From
the graphs it can be seen that, VGs placed on the top surface of the duct (D1, D5) had
no impact on the flow inside the duct. Surprisingly, when the VG were placed at the
bottom side of the inlet, the static pressure distributions seemed to match better with
the experimental data. This requires to be studied in greater detail. For example, in
the case of D4, the point of separation on the bottom side very closely matched with
the experimental data. However, the differences observed on the top side could be
due to geometric differences between the experimental model and the CFD model.
Small changes to be geometry would be tried to verify this aspect at a later stage.
Figures 15 and 16 show the stream line plots inside the duct for baseline duct and
Fig. 13 Static pressure ratio on top of the mid surface of the duct
Fig. 14 Static pressure ratio on bottom of the mid surface of the duct
Fig. 15 Streamline plot for baseline duct
Fig. 16 Streamline plot for duct for D4 VG design
duct with D4 VG design. Figures 17 and 18 show the static pressure plot at the center
plane for both baseline duct and duct with D4 VG. Figure 19 shows the velocity at the
AIP for baseline duct and duct with D4 VG. Further study is required to understand
the reason for the reduction in DC60.
Fig. 17 Static pressure plot for baseline duct

Fig. 18 Static pressure plot for the D4 VG design
Baseline duct D4 VG design
Fig. 19 Velocity plot at AIP for base design duct and D4 VG design
8 Conclusion
A CFD study of Serpentine duct was conducted to evaluate the effect of Vortex Gener-
ators to improve the flow characteristics. A duct, as described in a reference paper,
was modeled using CATIA software. ANSYS-ICEM-CFD was used to generate the
grid and RANS option of ANSYS-FLUENT was used to solve the flow problem.
Grid independence study was conducted and it was established that for the given
problem, G4 grids (180 grid points along tangential direction, 126 grid points along
the radial direction and 158 grid points along the flow direction) was adequate.
Various turbulence model options in the software were tried out and it was found
that Standard K-Omega turbulence model was the best suitable for the present study.
Flow characteristics through CFD of a baseline duct were compared with experi-
mental data. As the trends from CFD were similar to those presented in the reference
paper, with the deviations from experiments following same pattern, it was decided
to accept the baseline duct as the best match possible with the experiment for this
study.
Six different configurations of VG were tried out and it was found that D4 config-
uration (six of them on the bottom surface—three co-rotating on either side of center-
line at the entry section) gave the best results. There was a substantial reduction of
Distortion Index DC60 from nearly 30% in baseline to 21%, even though the pressure
recovery had reduced by nearly 2.7%. The static pressure ratios on the centerline for
the D4 configuration matched better with the experimental trend.
The difference in pressure recovery between experiment and CFD could also be
due to the type of bell mouth used in the experiment. There could be a pre-entry loss
that is missing in CFD, as the flow was assumed to enter the duct with no loss in
Total pressure at the inlet.
Further work is planned to reconcile the major differences noticed between CFD
and the experiment. After that, effect of placing VG on the top surface at the second
bend and the effect of changing the geometry of VG would be studied.
References
1. Hamstra JW, Miller DN, Truax PP (2000) Active inlet flow control technology demonstration.
ICAS
2. Logdberg O (2006) Vortex generators and turbulent boundary layer separation control. Technical
Report, Royal Institute of Technology, Stockholm, Sweden, Oct 2006
3. Wellborn SR, Reichert BA, Okiishi TH (1992) An experimental investigation of the flow in a
diffusing S-duct. NASA Technical Memorandum, AIAA-92-3622
4. Computational studies of Passive Vortex generators for Flow control. Technical Report, Royal
Institute of Technology, Stockholm, Sweden, Dec 2009
5. Chima RV, Arend DJ, Castner RS, Slater JW (2010) CFD models of a serpentine inlet, fan and
nozzle. In: AIAA 2010, 33 Jan 2010
6. Sobester A (2007) Tradeoffs in jet inlet design, a historical perspective. J Aircraft 44(3), June
2007
Supersonic Flow Behavior in Cartridge
Starter
Ritesh Gaur, Suparna Pal, Vimala Narayanan, D. Kishore Prasad,
and N. Balamurali Krishnan
Abstract Hot flow analysis was undertaken in a cartridge starter using NUMECA
software. Cartridge starter provides high temperature gases with kinetic energy which
impinges on to the rotor blades of the turbine at the initial stage of starting a gas
turbine engine which sets gas turbine to rotate. Other starting mechanisms which
can be used are air starter, electric motor or a small gas turbine engine. The cartridge
starter assembly consists of gas generator section with perforated central core, vertical
tube/pipe which bifurcates into two pipe lines, called short arm and long arm, which
provide high velocity hot gases to the high pressure turbine rotor blades. These
two pipelines form convergent–divergent nozzles. Computational fluid dynamics
(CFD) analyses were carried out for nine operating points for the inlet pressure
ranging from 90 to 157 bar. Analyses considered the cartridge starter gas flow domain
only and burning model is not considered. Hexahedral cells are generated with cut
cell approach using fine Hexpress software. The problem is considered as three-
dimensional, steady, compressible with turbulent flow for the flow analysis. The aim
of the project is to determine the mass flow rate, its distribution in two arms and the
total pressure loss. Spalart–Allmaras turbulence model with extended wall function
was chosen for this analysis. First analysis showed that the flow was unsteady at the
exit plane of nozzle. Therefore, analysis was carried out by considering unsteadiness
in the flow and averaged quantities are derived. Flow is found to diffuse in the
divergent nozzle after a normal shock. Mass flow rate has reduced due to the formation
of shock. It is also found that the total pressure loss has reduced near the perforated
central core of cartridge starter.
Keywords Cartridge starter · HPT rotor blade · Starting mechanism
R. Gaur (B) · S. Pal · V. Narayanan · D. Kishore Prasad · N. Balamurali Krishnan

Gas Turbine Research Establishment, CV Raman Nagar, Bangalore 93, India

332 R. Gaur et al.
1 Introduction
The use of small gas turbine engines as propulsion sources for cruise missiles and
UAVs is based on the ability to realize the engine including its peripheral systems
at low cost and offer high reliability. The starting system being a critical peripheral
system is required to provide the initial cranking of the gas turbine engine followed
by initiation of combustion in the combustor till the propulsion system takes over.
Hence, two common steps in the engine starting sequence include cranking and
ignition. The selection of the cranking device, also called as the starter, depends
on many factors like the engine configuration, operating comfort, engine stat time
requirements, safety and reliability. Among the various alternatives available to start a
small turbo fan engine, cartridge type of starter is found suitable for initial cranking of
small expendable engines owing to its simple, compact design which is independent
of any other source, high torque to weight ratio, high reliability and ability to meet
lower starting time requirement.
The purpose of a starter is to transmit the required starting torque to the engine
rotating assembly such that it is able to accelerate the engine to enable light up and
taking over by turbine such that the starter assistance can be removed. A typical
starting cycle is shown in Fig. 1 indicating different speed requirements with time
for starting of a typical gas turbine engine.
Cartridge starter discussed in the current paper is sized to supply torque to over-
come the drag torque of the engine such that it is used to accelerate the engine to the
required speed.
The cartridge starter discussed in the current work comprises of a solid propellant
based gas generator. The cartridge starter is initiated through supply of electrical
voltage to an pyro cartridge. Upon activation, the starter generates high temperature
and high pressure gases that impinge on the HP turbine rotor assembly and provide
the initial torque. Starting nozzles are used at the exit of the cartridge starter to
maximize the exit velocities and hence the momentum for the given mass flow rate
of gases. The high acceleration rate of the turbine helps to ensure lower exposure
time of the turbine blades to the high temperature gases.
Anil and Balamuralikrishan [2] have described the approach for sizing and config-
uring pyro-based starting system. They explained the challenges involved in attaining
light up speed for establishing sustained combustion. It deals with high acceleration
of spool and the assessment of all inertia loads. They have detailed the starting process
and developed a torque model which can predict the achievable spool speed using
pyro gas.
In the work presented, hot flow analysis was undertaken for a cartridge starter
using NUMECA software. Cartridge starter provides high temperature gases with
kinetic energy which impinges on to the rotor blades of the turbine at the initial
stage of starting a gas turbine engine which sets gas turbine to rotate. Other starting
mechanisms which can be used are air starter, electric motor or a small gas turbine
engine. The cartridge starter assembly consists of gas generator with perforated
central core, vertical tube/pipe which bifurcates into two pipe lines which provide
Supersonic Flow Behavior in Cartridge Starter 333
Fig. 1 Starting cycle [1]
high velocity hot gases to the high pressure turbine rotor blades. These two pipelines
form convergent–divergent nozzles.
2 Modeling and Grid Generation
Grid generation is a process of dividing the domain of observation into a set of

small control volumes, which are associated to several flow variables (e.g., velocity,
pressure, temperature, etc.). The generated grid has an impact on the accuracy of
the results, which shows the importance of this process. The grid was generated
using the software named Hexpress [3] which is an unstructured hexahedral grid
generation software designed to automatically generate meshes for the discretization
of complex 2D and 3D geometries using cut cell approach. This gives the advantage
of structural mesh in the accuracy of solver and flexibility of an unstructured grid
for capturing the details of the domain. It starts its mesh generation process from
334 R. Gaur et al.
an initial mesh which encompasses the whole computational domain. An initial

mesh is automatically proposed which corresponds to an isotropic subdivision of the
computational domain bounding box. The next step consists of two major actions,
refinement and trimming.
Grid refinement is a process in which cells are successively subdivided such that
specific geometrical criteria (to make grid adapt the shape of the geometry) are
satisfied. Number of refinements can be decided using following formula:

D
ln di
n= (1)
ln(2)
where ‘n’ is number of refinements, ‘D’ size of the initial cell in xyz direction and
di minimum cell size required to capture the smallest detail of geometry. To capture
the geometry, it is required to define the refinement level for different surface. This
refinement level ensures that the smallest detail of the geometry is captured.
Trimming involves two operations viz. snapping and optimization. The trimming
step removes all the cells intersecting or located outside of the geometry. Software
automatically finds the cells located inside the computational domain as both domain
coordinates and cell coordinates are known. At the end of the process, a staircase
mesh is obtained including all the interior cells. Finally, the mesh is smoothed by
moving outer grid points on the surfaces and in the volume in order to obtain a mesh
with a good but not guaranteed quality. The mesh obtained after the snapping action
may involve poor quality cells usually located close to corners and curves. Some of
these cells can be concave, twisted or may even present a negative volume. Software
uses specific algorithms to convert concave cells to convex ones by slightly displacing
their vertices. A novel approach is also implemented to improve the orthogonality
of convex cells and hence the quality of overall grid based on the preset criteria. The
convex cells are important to ensure the stability and the robustness of a flow solver.
The grid generated at the end of this step can be used for Euler solver.
Software uses a very specific approach for the insertion of layers of large aspect
ratio cells to accurately resolve boundary layers. The technique is based on successive
subdivisions of the cells connected to the walls; this contrasts with other techniques
which insert layers by extrusion. This refinement technique has the advantage of
robustness and speed. Besides, the inflation algorithm improves the mesh quality in
viscous layers. Once optimized grid is available, the software asks for the boundary
layer refinements for different surfaces. This is where first cell width can be specified
and can also be calculated for a required Y+ using the following equation:
Vref −7 L ref 1 +
ywall = ( )8( ) 8 y1 (2)
v 2
Y+ distribution over the entire domain for the plain passing through the center of
long leg is shown in Fig. 2. Depending upon the flow reference velocity, reference
length and kinematic viscosity, the grid with close to desired Y+ is generated. Since
Fig. 2 Y+ details of the

generated grid
Y+ is close to 40, turbulence model with extended wall function is used. The focus
was to generate a good quality grid for the domain particularly for the long and short
nozzles. The grid size for the entire domain is 1.23 million without any negative,
concave or twisted cells. Grid quality is not relaxed for any domain. Other grid quality
details are presented in Table 1. Figure 3 is showing the arrangement of cartridge
starter with details of CFD domain’s inlet and outlet.
3 Governing Equations
The well-known Navier–Stokes equations of motion describe the mechanism of

flow and associated energy transfer for any fluid. Although the fluid considered is
incompressible, the generalized equations are shown which are applicable for both
incompressible and compressible flows.
The general Navier–Stokes equations written in a Cartesian frame can be
expressed as:

∂ S − −
→
U d + F.d G.d S = ST d (3)
∂t
x s
where is the control volume, S is the control surface, and U is the set of conservative
variables:
336 R. Gaur et al.
Fig. 3 Arrangement of
cartridge starter with details
of CFD domain’s inlet and
outlet
⎡ ⎤
ρ
⎢ ρu ⎥
⎢ ⎥
⎢ ⎥
U = ⎢ ρv ⎥ (4)
⎢ ⎥
⎣ ρw ⎦
ρE
−
→ −
→
and F and G are the advective and diffusive part of the fluxes, respectively, also
can be denoted as inviscid and viscous fluxes. When U is taken as ρ (density) and
the diffusive part of the flux is zero (as mass cannot diffuse) along with source
term (as mass neither can be created nor can be destroyed), it represents the mass
conservation equation, popularly known as continuity equation. Next three values
of U (ρu, ρv,ρw) represent the momentum equation and are popularly known as
Navier–Stokes equations. When U is taken as ρ E (internal energy), it represents the
energy equation.
⎛⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎞
ρu ρv ρw
⎜⎢ ρu 2 + p ⎥ ⎢ ρuv ⎥ ⎢ ρuw ⎥⎟
→ ⎜
− ⎢
⎜⎢
⎥ ⎢
⎥ ⎢
⎥ ⎢
⎥ ⎢
⎥⎟
⎥⎟
F = ⎜⎢ ρuv ⎥, ⎢ ρv 2 + p ⎥, ⎢ ρvw ⎥⎟ (5)
⎜⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎟
⎝⎣ ρuw ⎦ ⎣ ρvw ⎦ ⎣ ρw 2 + p ⎦⎠
ρu H ρv H ρw H
where H is the total enthalpy per unit mass:
H = E + p/ρ (6)
⎛⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎞
0 0 0
⎜⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎟
⎜⎢ τx x ⎥ ⎢ τ yx ⎥ ⎢ τzx ⎥⎟
→ ⎜
− ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎟
G =⎜⎢
⎜⎢ τx y ⎥, ⎢
⎥ ⎢ τ yy ⎥, ⎢
⎥ ⎢ τzy ⎥⎟
⎥⎟ (7)
⎜⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎟
⎝⎣ τx z ⎦ ⎣ τ yz ⎦ ⎣ τzz ⎦⎠
uτx x + vτx y + wτx z − q x uτ yx + vτ yy + wτ yz − q y uτzx + vτzy + wτzz − qz
As only Newtonian fluids are considered (which is the case for most gases and
common liquids), the stress tensor is a linear function of the shear rate as shown in
the equation.

−→ → −→ →T 2 −
→− −
→→
τ =μ ∇ ⊗−
v + ∇ ⊗−
v − ∇ .→
v I + ζ ∇ .−
v (8)
3
where μ is the dynamic molecular viscosity, − →v is velocity vector and I represents

Identity (unit) tensor.
The heat flux vector is defined by Fourier’s law:
−
→
−
→
q = −k ∇ T (9)
where T is the static temperature and k is the molecular thermal conductivity.

⎞ ⎛
0
⎜ρf ⎟
⎜ ex ⎟
⎜ ⎟
ST contains the source terms : ST = ⎜ ρ f ey ⎟ (10)
⎜ ⎟
⎝ ρ f ez ⎠
wf
−
→
with the vector f e expressing the effects of external forces and W f , the work
performed by those external forces is accounted as
−
→ →
W f = ρ f e .−
v (11)
338 R. Gaur et al.
The Spalart–Allmaras turbulence model is one equation turbulence model which

estimates the kinematic turbulent viscosity based upon the value specified in the
boundary conditions. Details regarding all the terms and coefficients of the effective
turbulence eddy viscosity can be found in manual of FINE-Open [4].
4 Computational Details
CFD analyses were carried out for nine operating points for the inlet pressure ranging
from 90 to 157 bar. Analyses considered the cartridge starter gas flow domain only and
burning model is not considered. Air is considered as perfect gas with static pressure
and temperature values at the inlet as reference values for property definition. The
problem is considered as three-dimensional unsteady, compressible with turbulent
flow for the flow analysis. The aim of the analysis is to determine the mass flow rate,
its distribution in two arms and the total pressure loss. Spalart–Allmaras turbulence
model with extended wall function was chosen for this analysis. Since the flow was
unsteady at the exit plane of nozzle, unsteady computation was carried out. Second-
order central difference scheme is employed in space for discretizing the variables
in the governing equations (Navier–Stokes equations) using commercial software
package FINE-Open [4]. The minimum convergence criterion of 10–6 is defined for
the governing equations.
5 Results
The computational results are extracted and relevant performance indicators are
derived to assess the performance of the cartridge started under various operating
conditions. Computations are carried out for nine different operating conditions from
80 to 160 bar as inlet total pressure value. Flow is found to diffuse in the divergent
nozzle after a normal shock. Mass flow rate has reduced due to the formation of
shock. It is also found that the total pressure loss has reduced near the perforated
central core of cartridge starter and maximum loss is associated with the shocks in
the CD nozzles.
The variation of pressure loss in two arms of the cartridge starter is presented in
Fig. 4. Pressure loss in the short arm remained almost constant. Pressure loss in the
long arm is slightly more for most of the cases. For 80 bar case, Mach number was
found to be comparatively lesser as shown in Fig. 5, which has resulted in lesser loss.
For case 120 bar, mass flow from the CD nozzle was comparatively increased for
long arm, as shown in Fig. 6, and recirculation zone has reduced which resulted in
lesser loss in this arm.
Mach number variation at the exit of the long and short arms is shown in Fig. 5
which shows that Mach number remains almost constant from 80 to 120 bar for both
Variation of Pressure Loss

25
20
Pressure loss (%)
15
10 Long_Arm
Short_Arm
5
0
8000000 10000000 12000000 14000000 16000000
Total Pressure (Pa)
Fig. 4 Variation of pressure loss with total pressure for different inlet total pressure values
VariaƟon of Mach Number with Total Pressure

Long Arm Short Arm
2.60
2.55
2.50
Mach Number
2.45
2.40
2.35
2.30
2.25
Total Pressure (Pa)
Fig. 5 Mach number distribution with total pressure for different inlet total pressure values
long and short arm and then reduce and settle for lesser value after 140 bar case
which indicates the higher mass flow and lower losses close to design condition.
340 R. Gaur et al.
Total Pressure Vs Mass Flow DistribuƟon

0.70
Total Mass_Flow
0.60
Short Arm
Mass Flow (kg/s)
0.50 Long Arm
0.40
0.30
0.20
0.10
0.00
Total Pressure
Fig. 6 Mass flow distribution with total pressure for different inlet total pressure values
6 Conclusion
Unsteady CFD analyses are carried out for nine different operating conditions from
80 to 160 bar as inlet total pressure value. Flow is found to diffuse in the divergent
nozzle after a normal shock. Mass flow rate has reduced due to the formation of
shock. CFD is successfully applied for studying the supersonic flow behavior in
cartridge starter. The velocity of the high pressure high temperature gas coming out
from the exit of converging diverging nozzles predicted using CFD study is found to
be in good agreement with the design value.
Table 1 Grid quality details

Quality criteria Value
Minimum orthogonality 13.4
Average equiangular skewness 80.7
Maximum expansion ratio 7.9
Maximum aspect ratio 6.7
References
1. https://megiovandi.wordpress.com/tag/starting-system/
2. Anil Kumar K, Balamuralikrishnan N (2017) Pyro assisted starting of small gas turbine engine
for unmanned application. In: Proceedings of the 2017 gas turbine india conference, 7–8 Dec,
Bangalore
3. User manual Hexpress version 7.2 (2018)
4. Theoretical manual FINE-Open version 7.2 (2018)
High-Speed Shadowgraph Visualization
Studies of the Effectiveness of Ventilating
a V-Gutter Flame Holder to Mitigate
Screech Combustion Instability
in an Aero-Gas Turbine Afterburner
C. Rajashekar, Shambhoo , H. S. Raghukumar, R. M. Udaya Kumar,

K. Ashirvadam, and J. J. Isaac
Abstract Screech combustion instabilities are high frequency (>1000 Hz) trans-
verse periodic oscillations driven by combustion and which are then manifested
as large amplitude oscillations in the afterburner duct pressure, accompanied by
the characteristic high-pitched audible tones. These screech instabilities which are
detrimental to the engine are conventionally suppressed by embedding Helmholtz
resonator arrays in the afterburner liner. This method has been found inadequate when
mixed mode combustion instability oscillations occur and also when the frequencies
of oscillation were lower. The design of practical Helmholtz resonator arrays is clas-
sified and so is not available in the open domain. Hence, it was necessary to evolve a
robust design solution to mitigate screech combustion instabilities in an afterburner.
In an afterburner, V-gutters are used as flame stabilizers. The high Reynolds number
flow past a V-gutter array is dominated by the presence of vortices characterized by
the Kelvin–Helmholtz instability, which is a convective flow instability related to
the shear layers separating from the V-gutter lips and the Benard–von Karman insta-
bility which is related to the asymmetric vortex shedding of the flow in the flame
holder wake. The shedding of von Karman vortices at non-reacting and near the
blowout conditions, and the transition from a Kelvin–Helmholtz instability to that of
a Bernard–von Karman instability during near flame blowout create conditions for the
frequency to get locked-on to the duct transverse acoustic mode frequency; screech
is triggered. Hence, a smart flame stabilization method which has the intrinsic prop-
erty of preventing the lock-on between the frequency of the vortex shedding from
the V-gutter and the duct transverse acoustic frequency was developed. The test rig
with optically accessible critical zones around the V-gutter flame stabilizer had the
capability to operate the afterburner model under simulated inlet conditions of pres-
sure and temperatures. A FastCam SA-4 Photron high-speed camera was used in this
experimental investigation and high-speed shadowgraph flow visualization studies
C. Rajashekar (B) · Shambhoo · H. S. Raghukumar · R. M. Udaya Kumar · J. J. Isaac

Propulsion Division, CSIR-National Aerospace Laboratories, Bengaluru, Karnataka, India
K. Ashirvadam
ABES, Gas Turbine Research Establishment, DRDO, Bengaluru, Karnataka, India

344 C. Rajashekar et al.
were carried out to develop a comprehensive method of introducing an aerodynamic

splitter plate concept through a ventilated V-gutter; mitigation of screech combustion
instability has been demonstrated.
Keywords Combustion instability · Ventilated V-gutter · Screech · Afterburner ·

Screech mitigation
List of Abbreviations
CAF Compressed Air Facility

CSIR-NAL Council of Scientific and Industrial Research, National Aerospace
Laboratories
PXI Peripheral Extended Instrumentation
DAS Data Acquisition System
FFT Fast Fourier Transform
KH Kelvin Helmholtz
B-VK Benard–von Karman
1 Introduction
Afterburner combustion instability is one of the serious problems that is encountered

by aero-gas turbine engineers in the development phase of advanced military engines
[1, 2]. The serious detrimental effects of screech combustion instability phenomena
need to be solved through a robust design solution. The occurrence of combustion
instabilities is found to happen when there is coupling between the unsteady pressure
oscillations and the combustion heat release, thus satisfying the Rayleigh criterion.
Screech refers to an acoustic instability of resonant combustion, with the frequency
and phase of the wave oscillations corresponding to those for the transverse acoustic
resonance of the afterburner duct.
It is essential to understand the cause of the problem when addressing it through
robust and innovative solutions. Schlieren and Shadowgraphy are simple, but very
effective, non-intrusive techniques used to study complex flow phenomena such as
unsteady combustion in propulsion systems like the afterburner. Flow visualization
studies help in understanding the flow phenomena, to evolve and verify flow models,
help make theoretical predictions and also validate the computed solutions [3].
High-Speed Shadowgraph Visualization Studies … 345
2 Experimental Test Facility
A comprehensive experimental investigation has been carried out in a specially

set up versatile single V-gutter flame holder afterburner test rig (Fig. 1). The test
facility consisted of an air feed line of 152.4 mm diameter with a maximum air
delivery pressure of 1 MPa. This test rig was connected to the main reservoirs of the
compressed air facility (CAF) available at the wind tunnel center, CSIR-NAL. The
test rig consisted of a kerosene fueled slave combustor to heat up the air to simu-
late the afterburner inlet conditions, a settling chamber to provide stabilized uniform
entry conditions to the test section and a transition duct connecting the circular outlet
of the settling chamber to the rectangular section afterburner duct where the V-gutter
flame holder/fuel injector was fitted. The required air mass flow, metered by an
orifice plate, was allowed into the afterburner test rig model and was controlled by
a motorized valve. Metered kerosene fuel was fed to the slave combustor and the
afterburner using a nitrogen gas top loading method. The hot section of the test rig
had a proper cooling arrangement to avoid over heating of the critical afterburner
section of the test rig. The dimensions of the rectangular section afterburner were
200 mm × 70 mm, which housed the single V-gutter to generate screech combustion
instability at selected frequencies. The length of the afterburner duct was around 1 m.
The afterburner fuel injector was located upstream of the V-gutter and the fuel was
injected transversely to the air flow through 14 nos 0.5 mm holes.
Fig. 1 Schematic and photograph of the experimental test set up

The critical section of the afterburner test facility was appropriately modified with
optical access to enable the high-speed flow visualization studies. The test section
in the region of the V-gutter flame holder could be fitted with a metal cover plate or
quartz glass depending on the type of experiments. Two collimating concave mirrors
of diameter 300 mm and focal length 2.8 m were used for flow visualization studies.
A fiber optic illuminator having a 150 W halogen bulb was used as the light source. A
Photron FastCam SA 4 camera was used to capture the unsteady flow structures in the
V-gutter region; camera speeds up to 13,500 frames per sec at a resolution of 512 ×
464 pixels was used during the study. Figure 1 shows a schematic and photograph
of the complete test set up and the conventional Z-type mirror arrangement for the
flow visualization studies.
2.1 V-Gutter Configurations Considered

for the Experimental Studies
Figure 2a shows a CAD drawing of the flame stabilizer that was used for the exper-
imental studies. The material used was Nimonic 263 alloy. The flame stabilizer had
a rectangular slit at the center which could be varied from zero ventilation to various
widths of bleed using an insert and also to use a special metal insert which could
isolate the top and bottom halves of the V-gutter while completely covering the slit
as shown in Fig. 2b.
The V-gutter was designed in such a way that a passage was provided to channel
ventilated air directly to the flame holder base to create the best vortex suppression
effect. Figure 2c shows the assembly of the V-gutter inside the test rig and also the
mechanical splitter plate assembled V-gutter in position.
2.2 Instrumentation Details
The details of the test rig instrumentation scheme are shown in Fig. 3. The DAS
included pressure (steady and unsteady), temperature and mass flow rate measure-
ments. The pressure of air and fuel was measured using YOKOGOWA pressure
transmitters. All pressure signals were calibrated and connected to the NI PXI DAS
system. The temperatures at the exit of the slave combustor and at the inlet to after-
burner section were measured using K-type thermocouples. The mass flow rates of
air and fuel were measured using orifice and Coriolis mass flow meters, respectively.
The dynamic pressure measurements were carried out using Kulite unsteady pres-
sure sensors mounted in the afterburner section of the test rig. The other critical
parameters that were monitored during the experiments were the afterburner inlet
temperature, inlet pressure and overall afterburner equivalence ratio.
Fig. 2 a Dimensions of the V-gutter flame stabilser. b V-Gutter hardware. c Assembly of ventilated
V-gutter flame stabilizer in afterburner duct, and V-gutter with mechanical splitter plate
Fig. 2 (continued)
Fig. 3 Instrumentation scheme of the test rig
2.3 Experimental Procedure
Experiments were conducted to generate a specified controlled and sustained screech

frequency of 2000 Hz by a special technique at afterburner inlet temperatures of
around 600 K. The duct pressures were maintained around 150 kPa bar, abs. The air
flow to the afterburner test rig was controlled from a 152.4 mm diameter motorized
valve and the required air flow and pressure was regulated in the test rig for smooth
ignition of the slave combustor. After ignition of the slave combustor, the fuel flow
and the line air pressure were controlled to get the required pre-set afterburner inlet
pressure and temperature. The afterburner fuel flow was carefully regulated, and after
the ignition of the afterburner and stable operating conditions attained, the afterburner
fuel flow was carefully further increased until screech was observed. The occurrence
of screech was noted from the distinct high frequency audible tone. Simultaneously,
the FFT of the fundamental mode and the overtones were observed on the control
room DAS monitor. The raw data and the derived data during the entire test run were
acquired in the DAS. Simultaneously, the shadowgraph images were captured by the
monochrome high-speed video camera and color flame pictures by a Nikon D 800
still camera.
The following sets of experiments were conducted during the investigation:
1. V-gutter with zero ventilation
2. V-gutter with a mechanical splitter plate
3. V-gutter with an aerodynamic splitter plate by inserting three selected metal
inserts with ventilation of width 42 mm and gap height of 10, 14 and 16 mm as
shown in Fig. 2a.
Figure 4 shows a high-speed shadowgraph of the flame holder wake at different

operating conditions. Figure 4a shows a high-speed shadowgraph of the flame holder
wake when the afterburner (A/B) was off, while the main combustor was kept on and
the AB inlet temperature was maintained at around 600 K. Corresponding pressure
signals measured with a dynamic pressure transducer are shown in Fig. 5a. At A/B
off conditions, fluctuations in pressure signals were very small and were of random
nature.
The Kelvin–Helmholtz (K-H) instability was observed in the separating shear
layers originating from the V-gutter tips, when afterburner was lit at low equivalence
ratios Fig. 4b. Pressure signal fluctuations had increased as compared to A/B off
condition but still small and random in nature. At higher equivalence ratios this
K-H instability changed into a Benard–von Karman (B-VK) instability, which was
characterized by the presence of periodic asymmetric vortex shedding behind the
flame holder as shown in Fig. 4c. Pressure signals, at this condition (afterburner on,
high equivalence ratio), exhibited large fluctuations with definite periodicity in the
pressure variation in the duct.
From the trace of the vortex evolution, extracted from high-speed video shad-
owgraphy, the vortex shedding frequency was found to be 1930 Hz, which was very
close to the frequency obtained from the FFT of measured dynamic pressure as shown
in Fig. 6, first mode:2000 Hz). Interestingly, this frequency (~2000 Hz) is not the
expected Strouhal frequency. Here, the vortex shedding frequency had ‘locked-on’
to duct resonant acoustic transverse mode frequency (first mode).
Asymmetric vortex shedding, large pressure oscillation, ‘lock-on’ of vortex shed-
ding frequency to the duct acoustic frequency indicated that the root cause of the
screech combustion instability was vortex shedding behind the flame holder. Suppres-
sion of this vortex shedding could be the key in eliminating the screech combustion
instability [4–6].
(a) Main combustor-on and (b) Afterburner on; no ventilation,

afterburner-off with inlet temperature overall equivalence ratio ~ 0.35
of ~ 600 K (Kelvin-Helmholtz instability),
No screech
(c) Afterburner on; no ventilation, equivalence

ratio ~0.45 (Benard-Von Karman instability), Screech
Fig. 4 High-speed shadowgraph images in the V-gutter region
Vortex shedding can be suppressed by introducing a mechanical splitter plate in

the wake behind the flame holder, which suppressed formation of any large-scale
shedding vortex structure in the flame holder wake. Experiments were conducted
using the V-gutter with a mechanical splitter plate as shown in Fig. 2b under similar
afterburner inlet conditions that were maintained for the V-gutter with zero ventila-
tion. Figure 7a, b shows high-speed shadowgraphs of flame holder wakes showing
the effect of mechanical splitter at afterburner inlet temperature of 600 K. Mechan-
ical splitter plate was successful in suppressing the formation of large-scale shedding
vortex structure at low equivalence ratio (up to ~Ø = 0.7) which is better than an
earlier result (without splitter plate case where screech was observed at Ø = 0.45),
but at higher equivalence ratio (Ø > 0.7), screech was observed. Figure 7c shows
Fig. 5 Unsteady pressure signal measured behind the flame holder for different operating condition
a Afterburner off; main combustor on, afterburner inlet temperature ~ 600 K. b Afterburner on;
no ventilation, overall equivalence ratios ~0.35, No screech (c) Afterburner on; no ventilation,
equivalence ratios ~0.45, screech
Fig. 6 Typical test result showing the screech frequency of 2000 Hz for an inlet pressure of 150 kPa
abs and inlet temperature ~600 K and equivalence ratio ~0.45
Fig. 7 a, bHigh-speed shadowgraphs of flame holder wakes showing the effect of the mechanical
splitter plate. c Screech frequency of 2000 Hz for an inlet pressure of 1.5 bar abs and inlet temperature
of ~600 K and equivalence ratio of ~0.85
the FFT of the unsteady pressure signal. The pressure excursion over the duct mean
pressure was found to be around 14%.
At low equivalence ratio, feedback energy from heat release fluctuation was small
to the acoustic field, which was easily dampened by the presence of a mechanical
plate. As the equivalence ratio increased, feedback energy also increased, and due to
fluid–structure interaction, the plate started to oscillate. Finally, the vortex shedding
got locked-on to the duct acoustic resonance frequency. Relative strength of the plate
and strength of feedback energy play an important role in determining the operating
condition at which screech might be triggered. A very heavy and strong plate might
be required to suppress the screech over a wide range of operating condition. This
will add a weight penalty to the system, notwithstanding the plate cooling difficulties,
which is not desirable in aero-systems. Hence, an innovative design was adopted to
create the same effect, as a mechanical splitter plate, over a wider range of operating
condition, but without weight or cooling penalty.
A passage was provided to channel ventilated air directly to the flame holder base
to create the best vortex suppression effect. If the channel was not provided, due to
a Coanda effect, the ventilated air would move along the inside V-gutter face and
would also experience bi-stability and wake asymmetry. The ventilation coefficient
‘v’ is defined as the ratio of the ventilation slit gap width (h) to the V-gutter base
height (H).
Figure 8a shows the effect of ventilation at an afterburner inlet temperature ~600 K.
The center jet from the V-gutter (for v = 10.6%) successfully eliminated any large-
scale shedding vortex structure, and no screech was observed at all operating condi-
tions for this ventilation coefficient value. While in other cases (for v = 6.67%), low
strength of the center jet allowed shedding vortex structure and screech combustion
instability was observed at such a low ventilation coefficient (low bleed). Figure 8b
shows the FFT of dynamic pressure measured for different ventilation coefficient
(0%, 6.6%, 9.3%, 10.6%). As the ventilation coefficient increased, the strength of the
central jet increased which suppressed the formation of large-scale shedding vortex
structure and hence the amplitude of screech combustion instability decreased as
shown in Fig. 9. At v = 10.6%, screech combustion instability had been completely
suppressed. High-speed shadowgraphs as shown in Fig. 7 and FFT of dynamic pres-
sure signal as shown in Fig. 8 both confirmed the suppression of screech at all
operating conditions.
Figure 10 shows the afterburner flame photographs at similar operating test condi-
tions. With no bleed condition, the screech was found to occur at an equivalence ratio
~0.45, whereas with 16 mm (v = 10.6%) bleed, it was possible to go to higher equiv-
alence ratios of the order 0.7. The intensity of the flame also clearly indicated that
it was possible to test at higher equivalence ratios without encountering any screech
combustion instability problems. It clearly indicated that the concept of aerodynamic
bleed was successful in breaking the communication between the top and bottom
halves of the V-gutter, suppressing the vortex shedding getting locked-on to the duct
acoustic frequency and thus mitigating screech.
Fig. 8 a High-speed shadowgraphs of flame holder wakes showing the effect of ventilation. After-
burner inlet temperature ~600 K. b FFT showing the effect of ventilation on the intensity of screech
combustion instability, afterburner inlet temperature ~600 K, equivalence ratios ~0.5
Fig. 9 Variation of screech amplitude with ventilation coefficient. Afterburner inlet temperature
~600 K, equivalence ratio ~0.8 (Screech frequency—2000 Hz)
4 Conclusions
A method of enabling controlled screech has been evolved. Using this, a technique
of incorporating an aerodynamic splitter plate using ventilation in the V-gutter flame
holder base has been successfully developed to mitigate screech in a model aero-
gas turbine afterburner. High-speed shadowgraphy has been shown to be a powerful
technique of studying V-gutter flame holder wake structures when screech combus-
tion was present. The transition from a Kelvin–Helmholtz to a Benard–von Karman
instability characterized the onset of screech combustion instability. For the chosen
operating conditions of pressure of 1.5 bar abs and temperature of ~600 K, ventilation
coefficient of atleast 10.6% was found to give good screech mitigation.
Fig. 10 Comparison of afterburner flame structure with aerodynamic splitter plate and with bleed
ventilation of 16 mm (v = 10.6%)
Acknowledgements The authors thank the Director, Gas Turbine Research Establishment, DRDO
for sponsoring this project under the Gas Turbine Enabling Technology (GATET) scheme. The
authors thank the Director, NAL and the Head, Propulsion Division for granting permission to take
up this project.
They would also like to thank Mr Fakruddin Goususab Agadi, Technician-1 Propulsion Division
for his technical support.
References
1. Sotheran A (1987) High performance turbofan afterburner systems. In: AIAA-87–1830

AIAA/SAE/ASEE 23rd joint propulsion conference, 29 June–2 July 1987
2. Houshang E (2006) Overview of gas turbine augmentor design, operation, and combustion
oscillation. In: AIAA-2006–4916 AIAA/SAE/ASEE, 42nd joint propulsion conference, 9–12
July 2, 2006
3. Culick FEC (1988) Combustion instabilities in liquid-fueled propulsion systems—an overview.
AGARD 72B PEP meeting, Bath, England
4. Briones A, Sekar MB (2012) Effect of von Karman vortex shedding on regular and open slit
V-gutter stabilized turbulent premixed flames. Spring Technical meeting of the Central States
Section of the Combustion Institute
5. Yiqing Du, Qian R, Peng S (2016) Coherent structure in flow over a slitted bluff body. Commun
Nonlinear Sci Numer Simul 11(3):391–412
6. Jing T, Chang W, Go L (1994) Flame stabilization in the wake flow behind a slit V-gutter.
Combust Flame 99(2):288–294
Passive Reduction of Aerodynamic
Rolling Moment for a Launch Vehicle
Pankaj Priyadarshi, Amit Sachdeva, and Leya Joseph
Abstract Aerodynamic rolling moment on a core-alone launch vehicle due to the

presence of a wire tunnel and the associated roll dynamics has been studied. Com-
putational Fluid Dynamics (CFD) simulations across the roll angles from φ = 0◦ to
180◦ and across the Mach number range indicated the wire tunnel to be the major
cause of the rolling moment. A passive means of roll moment reduction has been pro-
posed by adding dummy wire tunnels symmetrically around the vehicle. It was found
that adding one dummy wire tunnel diagonally opposite to the existing wire tunnel
did not reduce the peak rolling moment as the leeward wire tunnel is ineffective.
However, adding two dummy wire tunnels reduced the rolling moment substantially.
Addition of the third dummy wire tunnel was also helpful in reducing the rolling
moment further, though marginally. In addition to the CFD studies, the maximum
roll rates and roll errors of the different configurations have been compared through
roll dynamic simulations. A novel linear superposition methodology has been pro-
posed and validated to obtain the rolling moment coefficient for multiple wire tunnel
configurations.
Keywords Aerodynamic rolling moment · Roll dynamics · Wire tunnels ·

Core-alone launch vehicle · CFD · SU2
P. Priyadarshi (B) · A. Sachdeva · L. Joseph

Vikram Sarabhai Space Centre, Thiruvananthapuram, India
A. Sachdeva
L. Joseph

360 P. Priyadarshi et al.
Abbreviations
M Mach number
φ Roll angle
CRM Rolling moment coefficient
CN Normal force coefficient
α Angle of Attack
t Time
φ̇ Roll rate
Izz Second moment of inertial about the roll axis
C Moment generated by the control system
R Aerodynamic rolling moment
q Dynamic pressure
S Reference area
d Reference length
Cmax Amplitude of moment generated by control system
K φ Gain associated with roll error
K φ̇ Gain associated with roll rate
φ Roll error
S Sigmoid function
1 Introduction
Attitude angles of a launch vehicle need to be controlled to ensure that a near-

nominal trajectory is achieved with acceptable peak angular rates for the survival of
the vehicle’s instrumentation and payload. At high angular rates, the gyro sensors
can lose their reference and become inaccurate. Launch vehicles, especially core-
alone vehicles, have low inertia about the roll axis and are susceptible to large roll
rates and roll angle errors due to aerodynamic rolling moments. Hence, either the
inherent rolling moment has to be reduced or appropriate roll control system to
counter the disturbing rolling moments has to be incorporated. Typically, launch
vehicles employ aerodynamic control surfaces such as fins and propulsive systems
like Reaction Control System (RCS) to control the roll rates and roll angle errors.
Pure hydrazine based roll control thrusters are used in European launch vehicles like
Ariane 5 and VEGA [1].
In this work, a passive technique for rolling moment reduction for a core-alone
launch vehicle configuration is presented. Asymmetry introduced by the wire tunnel,
in the one-wire tunnel configuration, is the major source of rolling moment when
the vehicle is at an angle of attack. RCS sizing studies on a single wire tunnel
configuration indicated large roll-control demand. The current proposal is based on
a detailed study on how the rolling moment, roll rate and roll error can be reduced in
a passive manner. This is investigated by progressively adding dummy wire tunnels
Passive Reduction of Aerodynamic Rolling Moment for a Launch Vehicle 361
to induce symmetry about the roll axis such that the rolling moment generated by
the overall system is significantly reduced. Such a technique would also significantly
bring down the roll control demand. The rolling moment demand due to CG offset
and thrust offset are easily corrected by the pitch and yaw thrust vector control of
the vehicle and do not cause any rolling moment on the vehicle.
The methodology adopted in the aerodynamic and roll dynamic studies is pre-
sented in Sect. 2. Results of CFD simulations are presented in Sect. 3.1. The appli-
cability of various techniques presented in Sect. 2 has been validated in the same
section. Sect. 3.2 presents the results of the roll dynamics studies. Conclusions are
presented in Sect. 4.
2 Methodology
This section presents the methodology followed for the study. Section 2.1 presents the
approach used to characterize the aerodynamic rolling moment coefficients across
various Mach numbers and roll angles. This includes the proposed linear superpo-
sition based method to obtain rolling moment coefficients for multiple wire tunnel
configurations by employing data from single wire tunnel simulations. Section 2.2
presents the methodology adopted in the roll dynamics studies to assess the maximum
roll rates and roll angle errors for various configurations.
2.1 Characterization of Aerodynamic Rolling Moment
First the methodology for the Computational Fluid Dynamics (CFD) simulations is
presented. Thereafter, the proposed approach employing linear superposition to com-
pute rolling moment for two, three and four wire tunnel configurations, by employing
results from single wire tunnel configuration, is presented. It is followed by descrip-
tion of the scaling approach to obtain variation of rolling moment with roll angle
across Mach numbers.
2.1.1 CFD Studies
An initial database of rolling moment has been derived using CFD simulations on a
one wire tunnel configuration of the core-alone launch vehicle at Mach number 1.6
and roll angles ranging between 0◦ and 180◦ , at an angle of attack of 4◦ . The simula-
tions have been carried out using the in-house CFD software, PARallel Aerodynamic
Simulator-3 Dimensional (PARAS-3D) [2]. The solver is based on rectangular adap-
tive Cartesian mesh and can solve both compressible Euler equation and Reynolds
Averaged Navier-Stokes (RANS) equations. The solver utilizes finite volume tech-
nique with explicit time marching. Primitive Variable Riemann Solver (PVRS) is
(a) Initial grid (b) Grid after final refinement
Fig. 1 Grid used for CFD simulations
used to evaluate inviscid fluxes, whereas central difference scheme is used for vis-
cous fluxes. The solver also has utility to refine the grid at any stage based on flow
gradients. In this study, RANS equation and k- turbulence model have been used
for simulations.
The domain considered for CFD simulations extends across −50 to +50 D in
the longitudinal direction, and from −40 to +40 D in the lateral directions, where
D is the diameter of the configuration. D and its associated circular area are the
reference length and reference area used in computation of aerodynamic coefficients
respectively. Adaptive grid refinement has been carried out at sufficient iteration
intervals. Figure 1 shows the initial Cartesian grid and the final grid after a sequence
of flow refinements till convergence in rolling moment coefficient has been achieved.
Figure 2 shows typical convergence plots for vehicle normal force coefficient (CN )
and rolling moment coefficient (CRM ). The grid is refined based on gradients in the
flow field at user specified iterations. Thus, after each refinement, the number of
cells in the grid increases as seen in Fig. 2. The convergence is measured across
the solutions before and after the last refinement, which are essentially the solutions
from two different grids. The CFD results had convergence in CRM within 0.8%.
Limited CFD simulations have been also carried out for other Mach numbers
and configurations with one and four wire tunnels to validate the PARAS results
and linear superposition explained in subsequent subsections. These simulations
have been carried out in the open-source CFD code, SU2 [3] using the body fitted
grid generated in Pointwise [4]. SU2 is capable of high-fidelity analysis solving
compressible and incompressible Euler, Navier-Stokes, and RANS solvers, along
with optimal shape design and adaptive grid refinement [5].
0.45 3 0.004 3
Rolling Moment Coefficient

0.4 0.0035
Normal force coefficient
No of Cells(in million)
No of Cells(in million)
0.35 0.003
0.3 2.5 0.0025 2.5
0.25 0.002
0.2 0.0015
0.15 2 0.001 2
0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160
Iteration/1000 Iteration/1000
(a) Variation of normal force coefficient (b) Variation of CRM with iterations
with iterations
Fig. 2 Convergence plots for normal force and rolling moment coefficients
2.1.2 The Proposed Linear Superposition Approach to Obtain Rolling

Moment Coefficients for Non-simulated Wire Tunnel
Configurations
As discussed earlier, most of CFD simulations have been carried out on a single
wire tunnel configuration. The rolling moment coefficients for configurations with
multiple wire tunnels were derived from the CFD results of single wire tunnel config-
uration using the principle of linear superposition. The simulated core-alone launch
vehicle configuration has very low CRM , which is primarily contributed by the wire
tunnel. The rolling moment contribution of the wire tunnel and the protrusions was
extracted from the single wire tunnel CFD simulations. Then, these result were used
to construct the rolling moment for the multiple wire tunnel configurations using
the principle of linear superposition. The linear superposition of CRM is justifiable
when the separation between the wire tunnels is significantly high so as to avoid
interference effects between the tunnels, and when the wire tunnel diameter is very
small in comparison to the launch vehicle diameter. The validation studies for this
approach have been presented in Sect. 3.1.
Consider a modified configuration of the core-alone vehicle with an additional
wire tunnel placed diametrically opposite to the already existing wire tunnel. The
modified configuration has an additional wire tunnel at a phase difference of 180◦
from the wire tunnel in the single wire tunnel configuration. The contribution from
the two wire tunnels can be captured as a linear superposition of two single wire
tunnel cases at a phase difference of 180◦ . The linear superposition is schematically
shown in Fig. 3a and can be mathematically expressed for a given roll angle φ and
Mach number M as follows:
(2) (1) (1)
CRM (φ, M) = CRM (φ, M) + CRM (φ + 180◦ , M) (1)
(1) (2)
where CRM and CRM represent rolling moment coefficients of one and two wire
tunnel configurations in the absence of protrusions respectively.
Similarly, the CRM for three and four wire tunnel configurations are obtained
through linear superposition of individual wire tunnel contributions as follows:
(3) (1) (1) (1)
CRM (φ, M) = CRM (φ, M) + CRM (φ + 120◦ , M) + CRM (φ + 240◦ , M) (2)
(4) (1) (1) (1)

CRM (φ, M) = CRM (φ, M) + CRM (φ + 90◦ , M) + CRM (φ + 180◦ , M)
(1)
+ CRM (φ + 270◦ , M) (3)
(3) (4)
where CRM and CRM represent rolling moment coefficients of three and four wire
tunnel configurations in the absence of protrusions respectively. The schematic rep-
resentations for three and four wire tunnel configurations are shown in Fig. 3b, c
respectively. The contribution of protrusions to CRM is also linearly superposed as
follows:
(n+prot) (n) (prot)
CRM (φ, M) = CRM (φ, M) + CRM (φ, M) , n = 1 . . . 4 (4)
(n+prot) (prot)
where CRM and CRM represent the CRM of the configuration with n wire tunnels
in the presence of protrusions and the CRM due to protrusions respectively.
2.1.3 Rolling Moment Coefficients at Non-simulated Mach Numbers
Rolling moment variation with roll angle (φ) has been characterized extensively for
the single wire tunnel configuration at Mach 1.6. However, roll dynamics studies
require aerodynamic moments at time-varying Mach number across the vehicle’s
trajectory. Thus, to derive the CRM variation with roll angles at a range of Mach
numbers (M = 0.8, 0.95, 1.05, 1.2, 2 and 4), available CFD simulation results at φ
= 45◦ [6] at these Mach numbers were made use of. The variation of CRM with φ
captured from Mach 1.6 simulations are scaled to other Mach numbers using the
CRM data at φ = 45◦ at these Mach numbers.
(1+ pr ot)
(1+prot) CRM (45◦ , M) (1+prot)
CRM (φ, M) = (1+prot)
× CRM (φ, 1.6) (5)
CRM (45◦ , 1.6)
2.2 Roll Dynamic Studies
The roll dynamics of the launch vehicle in presence of aerodynamic disturbances was
studied through 1-DOF simulations governed by the following equations of motion.
Fig. 3 Schematic representation of linear superposition of CRM for various configurations
d φ̇ C(φ, φ̇) + R (t)

= (6)
dt Izz (t)
dφ
= φ̇ (7)
dt
where φ, φ̇, C and R represent roll angle, roll rate, moment generated by control
system and aerodynamic rolling moment respectively. Aerodynamic rolling moment
is obtained as,
R (t) = S d q(t) CRM (M, φ) (8)
where q, S and d represents dynamic pressure, reference area and reference length
respectively.
A 1-DOF code was written in Octave [7] and was used to carry out analysis of
the vehicle’s roll dynamics in the presence of aerodynamic moments alone, i.e., with
C(φ, φ̇) = 0. Also simulations with a PI controller with rate feedback and with the
rolling moment modeled as a sigmoid function, were carried out. Moment generated
by control system is modeled using sigmoid function as follows:

C (t) = (−2Cmax ) × S φ̇, φ (9)

where the sigmoid function S φ̇, φ is given by,
⎛ ⎞
1
S φ̇, Δφ = ⎝ − 0.5⎠ (10)
1 + e−K φ̇ φ̇(t)−K φ φ(t)
Cmax and φ represent maximum moment that can be produced by the control
system and the deviation from initial roll angle respectively. K φ and K φ̇ represent the

PI controller gains associated with roll error and roll rate respectively. As S φ̇, φ
varies from 0 to 1, C (t) varies from −Cmax to Cmax .
The system of differential equations Eqs. (6) and (7) can be solved with the
appropriate initial conditions on roll angle and roll rate as follows:
φ (0) = φ0 (11)
φ̇ (0) = φ̇0 (12)
where φ0 and φ̇0 represent the initial values of roll angle and roll rate respectively.
The assumptions involved in the roll dynamics simulations are as follows:
– All simulations have been carried out with α = 4◦ throughout the flight for conser-
vative estimates. The variation of CRM corresponding to φ and M is considered.
– Off-nominal aerodynamic rolling moment data have been considered with 100%
dispersion for conservative estimates.
– Upper bound trajectory has been used for variation in Mach number and dynamic
pressure with time.
– Aerodynamic damping has been neglected.
– Sources of rolling moment other than aerodynamics, have not been considered.
– Nominal variation of moment of inertia about the roll axis has been accounted as
a function of time.
– Ideal control without any delay is considered.
– Control has been modeled using a sigmoid function.
– Wind plane azimuth is considered to remain invariant throughout the flight.
– Pitch-yaw coupling has been neglected.

– CRM variation with φ at Mach 1.6 has been scaled to other Mach numbers using
φ = 45◦ data at the respective Mach numbers.
– Linear superposition has been used to obtain CRM for two, three and four wire
tunnel configurations.
3 Results
Section 3.1 presents the rolling moment characterization for the core-alone vehicle
at different M and φ. The validation studies for the linear superposition method has
been discussed in the same section. Section 3.2 presents the results of the 1-DOF
simulations and compares the one, two, three and four wire tunnel configurations
based on the maximum roll rates and roll angle errors.
3.1 Characterization of Aerodynamic Rolling Moment
CFD simulations in PARAS-3D were carried out on the one wire tunnel configuration
of the vehicle at Mach 1.6, when the vehicle encounters peak dynamic pressure.
Figure 4 shows the variation of CRM with roll angle at this Mach number. As can be
observed from the plot, the peak aerodynamic rolling moment is generated when the
vehicle with one wire tunnel is at a roll angle around 67.5◦ and 292.5◦ . It can also be
observed that CRM is very low for φ in the roll angle range [120◦ , 240◦ ], when the wire
tunnel is on the leeward side. Figure 4 also compares the CRM values obtained from
PARAS-3D and SU2 . The values are fairly closely matching, thus giving confidence
on the CFD results especially when the wire tunnel is on the windward side.
Fig. 4 CRM variation with φ 0.003

PARAS-3D
for one wire tunnel launch SU2
vehicle configuration at 0.002
M = 1.6
0.001
CRM
-0.001
-0.002
-0.003
0 45 90 135 180 225 270 315 360
φ (deg)
Rolling moment coefficients for two, three and four wire tunnel configurations as
obtained from the linear superposition of results from one wire tunnel configuration
are shown in Fig. 5a. The maximum value of CRM remains similar for both one and
two wire tunnel configurations. This is due to the low rolling moment generation by
the wire tunnel on the leeward side, due to which the moment generated by the wire
tunnel on the windward side is not significantly reduced. Three and four wire tunnel
configurations have been simulated and both of them are found to be attractive. The
configurations with three and four wire tunnels leads to substantially lower CRM as
compared to one and two wire tunnel configurations at M = 1.6 (Fig. 5). The phase
difference between the wire tunnels is such that the rolling moment contributions
add up to provide very low net CRM . This is clearly seen from Fig. 5b where the
contribution from each wire tunnel is plotted along with the sum of contribution from
all wire tunnels for the three wire tunnel case. The contribution from the protrusion
is also plotted and is linearly superposed to obtain the final results.
Figure 6 shows the CRM variation with roll angle for one wire tunnel configuration
at various Mach numbers obtained by scaling the results at M = 1.6. A similar
approach has been applied on the linearly superimposed CRM of 2, 3 and 4 wire
tunnel cases.
3.1.1 Validation Studies
CFD simulation of the four wire tunnel configuration was carried out in SU2 and
was compared with the results from linear superposition. The maximum CRM for
the single wire tunnel case is 0.00267 which gets reduced by 80.5% to 0.0005 when
computed using linear superposition. Computational Fluid Dynamics (CFD) simu-
lations using SU2 show that the value reduces by 85% to 0.0004, which is very close
considering the small magnitude of CRM .
3.2 Roll Dynamics Studies
Time evolution of roll angle and roll rate obtained from 1-DOF simulations in the
absence of control, with initial roll angles varying from 0◦ to 180◦ are shown in Fig. 7.
As mentioned earlier, simulations were carried out using 100% dispersion in CRM
and considering an angle of attack of 4◦ throughout the trajectory. The initial roll rate
is considered to be zero for all simulations. As can be observed from results for single
wire tunnel configuration in Fig. 7a and Table 1, the system tries to achieve the nearest
roll trim point with a maximum roll rate of 15.2◦ /s. However, for most of the cases,
the system has non-zero roll rates at the end of simulation along with significant roll
errors. The reduction in roll rate with the addition of dummy wire tunnels are evident
from Fig. 7b–d. However, it is also observed from these figures that in the absence
of control, the roll angle errors are lesser in the configurations with two and three
additional wire tunnels as compared to the one wire tunnel configuration. Three and

N = 1 (CFD) N = 3 (LS)
for one, two, three and four
N = 2 (LS) N = 4 (LS)
wire tunnel configurations at 0.003
M = 1.6
0.002
0.001
CRM
0
-0.001
-0.002
-0.003
0 50 100 150 200 250 300 350 400
Roll angle (deg)
(a) CRM variation with φ for 1,2,3, and 4 wire tunnel configurations
0.004
3-WT Config. Contribution: WT 3
Contribution: WT 1 Contr.: Protrusions
0.003 Contribution: WT 2
0.002
0.001
CRM
-0.001
-0.002
-0.003
0 50 100 150 200 250 300 350
Roll Angle (deg)
(b) CRM variation with φ for three wire tunnel configuration. The
contribution from each wire tunnel have been superimposed

M = 0.8
and M for one wire tunnel M = .95
0.003
configuration M = 1.05
M = 1.2
0.002 M = 1.6
M = 2.0
0.001 M = 4.0
CRM
-0.001
-0.002
-0.003
-0.004
0 50 100 150 200 250 300 350 400
Roll angle (deg)
(a) Results for single wire tunnel configuration
(b) Results for two wire tunnel configuration
(c) Results for three wire tunnel configuration
(d) Results for four wire tunnel configuration
Fig. 7 Roll error and roll rate evolution with time obtained from simulations without control
Table 1 Maximum roll rate and roll angle errors from simulations in the absence and presence of
control
No. of wire Without control With 400 Nm control
tunnels
φmax φ̇max φmax φ̇max
1 444.7 15.2 267.8 13.0
2 790.7 13.2 137.3 11.5
3 387.2 6.0 40.4 2.1
4 390.0 5.2 16.6 0.7
four wire tunnel configurations have fairly low maximum roll rate and roll error as
compared to one and two wire tunnel configurations (Table 1).
1-DOF simulations in the presence of roll control with varying maximum roll
control capability have been carried out. Results of studies with a maximum control
capability of 400 Nm are shown in Table 1. In these studies, initial roll angle is varied
from 0◦ to 180◦ . Similar to the observation in the absence of control, the maximum
roll rates of 3 and 4 wire tunnel configurations are significantly low as compared to
the 1 and 2 wire tunnel configurations, and are ≈2.1◦ /s and ≈0.7◦ /s respectively. The
evolution of roll rate and roll angle for various configurations with time in the presence
of 400 Nm control are shown in Fig. 8. As observed from Fig. 8c, d, the maximum
roll rates are significantly low for three and four wire tunnel configurations. All the
configurations tend to return to the initial roll angles in the presence of control. As
mentioned earlier, the three and four wire tunnel configurations experience very low
deviation from the initial roll orientation and this is clearly evident from Fig. 8c, d.
As can be observed from Fig. 9a, b, one wire tunnel configuration experiences
the highest roll rate and roll error. Even with a control of 400 Nm capability, the
maximum roll angle error and roll rate are significant; the values being ≈268◦ and
≈13◦ /s respectively in comparison to ≈445◦ /s and ≈15.2◦ /s in the absence of control.
The addition of wire tunnels imparts symmetry to the system and hence is a passive
technique for rolling moment reduction. The two wire tunnel configuration has a
marginal decrease in maximum CRM (2.8%), which translates to a marginal decrease
in maximum roll rate. Even with a control of 400 Nm capability, the maximum roll
angle error and roll rate for the configuration are ≈137◦ and ≈11.5◦ /s respectively.
Inclusion of two additional wire tunnels reduces the CRM by 77%. The three wire
tunnel configuration thus enables significant reduction in roll rate and roll angle error.
Further, a reduction in maximum roll angle error from ≈387.2◦ in the absence of
control to ≈40.4◦ in the presence of 400 Nm control is possible. The roll rate also
reduces from ≈6◦ /s to ≈2.1◦ .
The four wire tunnel configuration has the least aerodynamic rolling moment
compared to the other three configurations. CRM reduces by ≈80% with the inclusion
of three dummy wire tunnels. The configuration has a maximum roll angle error and
roll rate of ≈390◦ and ≈5.2◦ /s in the absence of control. It is to be noted that the
maximum roll rate is significantly lower than that achievable with 400 Nm control
(a) Results for single wire tunnel configuration
(b) Results for two wire tunnel configuration
(c) Results for three wire tunnel configuration
(d) Results for four wire tunnel configuration
Fig. 8 Results of One Degree of Freedom (1-DOF) Simulations with 400 Nm control
Fig. 9 Comparison of 350

100 Nm
configuration with different 200 Nm
number of wire tunnels and 300 300 Nm
400 Nm
different maximum control 500 Nm
250
moments
Max. Δ φ(deg)
200
150
100
50
0
1 2 3 4
Number of wire tunnels
(a) Maximum roll angle error
16
100 Nm
200 Nm
14 300 Nm
400 Nm
12 500 Nm
Max. φ (deg/s)
10
8
.
0
1 2 3 4
Number of wire tunnels
(b) Maximum roll rates
in case of a single wire tunnel configuration. With a 400 Nm control, the maximum
roll angle error and roll rate further reduces to ≈16.6◦ and ≈0.7◦ /s respectively.
As observed from Fig. 9a, b, the maximum roll rate and roll angle error reduces
with the increase in control moment for all the configurations.
4 Conclusions
CFD studies carried out on the single wire tunnel configuration of the core-alone
launch vehicle indicated that the wire tunnel is the major source of asymmetry,
and hence the major source of aerodynamic rolling moment. The results also showed
significant rolling moment when the wire tunnel was positioned in the windward side.
A passive mode of CRM reduction has been attempted by adding dummy wire tunnels
to impart symmetry to the configuration. Addition of a wire tunnel diametrically

opposite to the existing wire tunnel turned out to be of minor consequence, as the
wire tunnel positioned in the leeward side couldn’t oppose the rolling moment of
the windward wire tunnel efficiently. However, the addition of two and three dummy
wire tunnels reduced the CRM of the system by ≈77% and ≈80% respectively.
A new, linear superposition based method has been proposed to estimate the CRM
for two, three and four wire tunnel configurations. First we extract the rolling moment
contribution of the wire tunnel and the protrusions from the CFD simulations on a
single wire tunnel configuration at various roll angles. These are linearly superposed
to predict CRM for two, three and four wire tunnel configurations. The method has
been validated by carrying out CFD simulations on the four wire tunnel configuration.
Roll dynamics simulations have been carried out to compare the maximum roll
rates and roll errors for the 4 configurations in presence of only aerodynamic rolling
moments as the source of perturbation. Conservative studies assuming upper bound
trajectory and α = 4◦ throughout the trajectory have been carried out. The results
indicate that the maximum roll rate in the absence of control is around 15.2◦ /s for
single wire tunnel and 13.2◦ /s for two wire tunnel configuration. The same reduces
to 6◦ /s with three wire tunnels and to a mere 5.2◦ /s with three wire tunnels. It is
however not recommended to rely on aerodynamics alone (without the use of any
roll control system). This is because the roll errors are found to persist throughout
the region of prominent atmosphere and the system doesn’t settle to the roll trim
points within this regime for many of the initial conditions used for simulations.
However, with the application of control, it is found that the maximum roll errors
can be reduced. The maximum roll rates are also found to decrease tremendously
with the application of control. With a 400 Nm control, the maximum roll rates
for one, two, three and four wire tunnel configurations are 13.0, 11.5, 2.1 and 0.7
respectively. The dummy wire tunnels add to the efficiency and reliability to the
launch vehicle due to their simplicity, lower mass and passive nature.
Therefore, provision of a minimum of two dummy wire tunnels provides an elegant
and passive aerodynamic solution to the problem of rolling moments generated by
the single wire tunnel configuration. The proposed design solution is likely to add to
the reliability and efficiency of the launch vehicle due to its passive nature, simplicity,
ease and cost of implementation, and likely to have a lower mass penalty.
References
1. Ulrich G, Stephan K, Dietmar W, Daniel F, Pierre M, Christian D, Santiago C, Bastian G, Rainer

K (2015) Development and test of a 3D printed hydrogen peroxide flight control thruster. In:
51st AIAA, SAE, ASEE joint propulsion conference, 27–29 July 2015, Orlando. Florida, USA
2. Kumaravel G, Gopalsamy M (2011) PARAS 3D user’s manual, version 4.1.0,
VSSC/ARD/GN/01/2011
3. Palacios F, Colonno MR, Aranake AC, Campos A, Copeland SR, Economon TD, Lonkar AK,
Lukaczyk TW, Taylor TWR, Alonso JJ (2013) Stanford University Unstructured (SU2): an
open-source integrated computational environment for multi-physics simulation and design. In:
AIAA paper 2013–0287, 51st AIAA aerospace sciences meeting and exhibit, 7th–10th Jan 2013.
Grapevine, Texas, USA
4. Pointwise user manual V18
5. Economon TD, Palacios F, Copeland SR, Lukaczyk TW, Alonso JJ (2016) AIAA J 54(3):828–
846
6. Sreenivasulu J, Saha SK (2018) Analysis of rolling moment coefficients for SSLV configuration.
VSSC/ACD/TM/006/2018
7. Eaton JW, Bateman D, Hauberg S, Wehbring R (2014) GNU octave version 3.8.1 manual: a high-
level interactive language for numerical computations. CreateSpace Independent Publishing
Platform. ISBN 1441413006, http://www.gnu.org/software/octave/doc/interpreter
Design and Development of Miniature
Mass Flow Control Unit for Air-Intake
Characterization
D. B. Singh, P. Vinay Raya, Buddhadeb Nath, N. Srinivasan, Anju Sharma,

and B. Sampath Rao
Abstract Wind tunnel tests on high-speed air-intake configuration needs an accurate

simulation of mass flow through the intake ducts. This leads to the requirement of
miniature mass flow control device which provides linear variation in the throat area.
Here, an attempt has been made to design and develop new miniature mass flow
control device to characterize the air-intake model in the 1.2 m wind tunnel. This
paper describes the design and development aspect of miniature mass flow control
unit, drive electronics and its utility in wind tunnel testing of air-intake models.
Emphasis is placed on experimental results obtained from an electrically actuated
plug which controls the critical flow area at the downstream end of the intake model.
Keywords Mass flow control unit · Air intake · Plug controls · Mass flow
calibration
1 Introduction
Simulating the required mass flow in wind tunnel experiments is a challenging prob-
lem. The air flow through the model has to be accurately simulated during the wind
tunnel tests to assess the intake performance parameters like total pressure recovery,
engine face total pressure distortion and the intake instability/buzz phenomenon [1,
2]. Mass Flow Control Unit (MFCU) is used for controlling the mass flow through
the air-intake duct. It is required to calibrate the airflow passing through the MFCU
for various axial positions of the plug by simulating the pressure ratios across the
MFCU as observed during the wind tunnel tests. The effective throat area for various
MFCU plug positions is established through detailed calibration tests. Figure 1 shows
the schematic view of the typical air-intake model. It was proposed to have intake
studies on missile model. The dimensions of the wind tunnel model were finalised
according to the standard procedure to suite 1.2 m Trisonic wind tunnel of NTAF,
D. B. Singh (B) · P. Vinay Raya · B. Nath · N. Srinivasan · A. Sharma · B. Sampath Rao

National Aerospace Laboratories, CSIR, Bangalore, India

378 D. B. Singh et al.
Fig. 1 Schematic view of the typical air intake model with Y duct
CSIR-NAL [3]. According to the approved model, duct size available is of diameter
60 mm. The existing mass flow controller Unit (MFCU) is bigger and cannot be used
for simulating required mass flow. Hence it was decided to make a new miniature
mass flow control unit with LabVIEW based control. The developed unit is miniature
and it can be used for duct size of 50 to 150 mm with traverse length of 10–60 mm.
The unit is robust in design and having a backlash-free driving mechanism. Higher
motor torque availability with planetary gear mechanism of the gear ratio of 1:62,
this makes it more accurate and allows to cater for higher torque required during wind
tunnel testing. High-resolution encoder makes the system more readable regarding
the position of the conical plug. System electronics developed based on digital logic
and complete control and data acquisition system is realised on National Instru-
ment cDAQ based hardware and LabVIEW based software. Most of the components
(Motor, gear, encoder and clock buffer circuitry) are housed inside the protective
casing preventing uncertainties during the wind tunnel tests.
2 Experimental Setup
Present tests are conducted using two independent systems, one system used for
controlling the MFCU unit and another system used for data acquisition. Both the
systems are tightly synchronised. Tests were conducted in continuous/step mode, here
at every step with fixed free stream condition rake data and plug position was acquired
and processed using LabVIEW based programs. Pressure probes and thermocouple
placed near orifice plate to calculate the primary mass flow. 40 probe total pressure
rake followed by MFCU unit were placed downstream to the model. The pressurised
Design and Development of Miniature Mass Flow . . . 379
chamber uses to simulate the tunnel pressure was made to check the amount of
air entering into the duct with and without ejector in place for different upstream
pressures ranging from 16 to 45 psig. Pressure regulating valve place in upstream of
the experimental setup for precise control of the pressure in the chamber.
3 Mass Flow Control Unit (MFCU)
The Schematic view of MFCU plug unit is shown in Fig. 2 The unit is fixed at the
downstream end of the intake model. It consists of a plug whose shape is contoured
having a radius of curvature of 288 mm. The movement of the plug axial and is
controlled through a drive motor. A geared DC motor, drives the plug and the plug
position is obtained using incremental optical encoder mounted on the motor shaft
at the rear extension. The encoder provides 500 pulses per rotation. An interface unit
has been built to provide an integrated pulse counting with 7 segment LED (Light
Emitting Diode) display and buffered TTL (Transistor-Transistor logic) equivalent
data to the associated data acquisition system for the position of the plug. The total
traverse length of the MFCU is 60 mm which corresponds to 80,000 counts in the
controller. At different axial locations, different throat areas are formed between the
plug and the insert ring. This indicates the area that controls the mass flow through
the intake model for a given upstream condition. The flow-through (throat) area
varies from a maximum to a minimum value (non-zero) at the two extreme traverse
limits. Figure 3 shows the block diagram of the MFCU unit. The developed unit
provides local mode operation/controls for forward/reverse drive of the plug and for
resetting the counter to zero at the fully open condition. The unit offers an interface
for remote operations from the PC for a programmed control operations. PC based
control is developed using NI cDAQ 9188 chassis with NI 9401 module to acquire
Fig. 2 Schematic view of a designed MFCU plug unit

Fig. 3 Block diagram of developed MFCU control unit
Fig. 4 Photograph of MFCU setup
20 bit digital information (5 digit plug position), NI 9401 for directly reading clock
and other handshake signals, NI 9205 to acquire analog input from pressure sensor
which is plotted against plug counts/area to detect the occurrence of buzz. Provision
has been made to reverse the plug manually in case of buzz occurrence. Figure 4
shows the photograph of the complete system.
3.1 Salient Features of the Newly Developed MFCU System
Guide Shaft: The guide shaft is designed in such a way that it caters the simple
supported structure for a respective ball-screw rod this makes evenly load distribution
on the driving pins during operation. Pitch of the MFCU can be adjusted with different
ball screw to get different traverse speed. Photograph of the developed system is
shown in Fig. 5
Ball Screw and Nut assembly: Provision has been made to use two ball screws with
a pitch of 2 and 4 mm, this will operate the plug to and fro 60 mm (2 mm pitch) and
50 mm (4 mm pitch) length in 26–12 s respectively. With the ball-screw arrangement,
the system has increased positional accuracy and ware and tare that may occur due
to the long run. As all the parts lead screw, nut and balls of the ball screw assembly
are hardened; there will be very negligible ware in the long run. As Ball-screw is
placed with the simple support system, it experiences only axial load acting on the
plug, keeping frictional loads minimum and frictional loss negligible.
Web: Web is the structure designed to keep the MFCU in the centre of the duct/
MFCU casing. The web is designed to take all the loads acting on MFCU, it has three
arms place at 120◦ apart evenly. The web has two bearings on which coupling are
placed; this will act as an intermediate part between the drive and driven mechanism
of the MFCU. It is necessary to keep the web diameter lesser than the maximum plug
diameter to have a proper choking effect only at throat region. It is recommended
to have an area greater than 120% of area available between casing and web region
concerning that of throat area in fully opened condition.
Drive and Feedback mechanism: Drive mechanism has two major components; DC
motor and planetary gearbox, as mentioned above. The appropriate motor has been
selected to provide necessary torque in smaller form factor. To achieve the required
Fig. 5 Schematic view of a designed MFCU plug unit

Fig. 6 Figure shows the logic implementation of MFCU control and motor drive logic
Fig. 7 MFCU continuous and step mode operation response
torque appropriate gearbox with 1:62 gear ratio is selected for the current design.
The unit contains the feedback device (encoder) having 500 pulses per rotation.
Drive circuit designed with latest digital logic circuitry as shown in Fig. 6. Motor
drive circuit implemented using H-bridge concept. Safety for ball-screw and motor
is taken care during design. Figure 7 shows the response of MFCU in continuous
and in step mode of operation, the system maintains linearity. In step mode system
response has been checked for the step size varying from 2000 to 10,000. Counts
have checked repeatability in full span of the traverse and by physical measurement
of distance traveled and found good repeatability in the traverse of the plug.
4 Plug Control for Air Intake Experiments
LabVIEW based data acquisition and control programs are developed and used for
controlling and acquiring position data of mass flow control plug. Here plug position
feedback is acquired through attached encoder giving 500 pulses per rotation. As
it required transmitting clock signal from the model to control situated 30 m away,
the clock signal was buffered onboard. Clock output is used for electronic counting
of up/down motion of plug. The local display is provided using five-digit seven
segment display. Further, position information is converted to BCD (5 digits, 20 bit)
and made available for three PCs used for acquiring model steady pressure data and
for model unsteady pressure data. With the developed program, it was ensured that all
the system, i.e. Steady pressure data acquisition, unsteady pressure data acquisition
and plug control PC are tightly synchronised. In plug control PC provision has been
made to display any of models Kulite pressure sensor output against the plug counts
Fig. 8 Screenshot of LabVIEW based plug control GUI
or plug area, this makes real-time monitoring of buzz onset and control of the plug
motion. Figure 8 shows a screenshot of LabVIEW based GUI used for plug motion
control and coordinating the data acquisition events on other systems. MFCU control
algorithm represented through flow-chart shown in Fig. 9.
5 Discussion of Test Results
Improved mass flow control unit with listed specification in Table 1 has been designed
and developed at CSIR-NAL Bangalore; the system is tested rigorously at bench
level and at high-speed flow facility. System performance matched as per design
specification. The developed system was calibrated at high-speed test rig; typical
result presented in Fig. 10. Many repeat tests were conducted and found an excellent
match in calibration data.
6 Conclusion
Improved MFCU has been developed and tested at high-speed flow facility. The
system is extensively checked for its load capability, accuracy, repeatability and
positional resolution. The system developed based on the latest electronic circuitry
and tested with portable NI data acquisition and control. Complete system perfor-
mance in continuous and step mode was established during mass flow calibration.
Many repeat tests were conducted and found an excellent match at similar conditions.
Fig. 9 MFCU operational

Start MFCU Unit
flow-chart
Enter parameters and press Enter

(For remote operation)
Enable Reverse motion: Local/System
no
Reverse limit reached ?
yes
Enable Reset: Local/System
Enable Forward motion: Lo-

cal/System (till reach initial position)
Enable forward motion with Po estd.
Forward limit switch no

reached or Keyboard hit ?
yes
Enable reverse motion: Local/System
no
Is Count = 5000 ?
yes
Stop
MFCU Unit
Table 1 Specification of the developed MFCU unit

Precision ball screw Pitch 4 mm and shaft diameter 10 mm
Precision ball screw Pitch 2 mm and shaft diameter 10 mm
Length of screw rod 95 mm
Traverse length 59 mm
Motor outer diameter 32 mm
Motor type DC motor 24 V
Torque required 2.5 Nm
Motor length 60 mm
Planetary gear box 1:62
Encoder details 500 PPR
Limit switches details Micro switch
Bearing type Ball bearing
Outer diameter 47 mm
Inner diameter 25 mm
Model no. and make SKF 6005 Z (2 no.s)
Fig. 10 Plot shows a typical calibration result of MFCU unit

Acknowledgements The authors are thankful to Shri. Rajeev G, Head, NTAF for his support and
guidance. The author would like to appreciate the support of the vendor Yottec, Bangalore for
realising the hardware of MFCU. Help of Design office and 1.2m wind tunnel staff acknowledge
with thanks.
References
1. Chidananda MS et al. Development of an ejector for ADA intake model tests. Project Document,
NAL PD PR 8805
2. Jaganatha Rao B et al. Static tests and mass flow calibration of 1:7 scale PV2GE air intake
model. Project Document, NAL PD PR0501
3. Narayan G et al. Instrumentation, control and data acquisition system for the LCA Kaveri air
intake model tests in NAL 1.2 m trisonic wind tunnel. Project Document, NAL PD NT 0007
The Effect of Variable Inlet Guide Vanes
on the Performance of Military Engine
Fan
Baljeet Kaur, Reza Abbas, and Ajay Pratap
Abstract This paper discusses the effect of variable inlet guide vanes (VIGV) on the
performance of military engine fan. The variable inlet guide vanes (VIGV) are neces-
sary in order to safely start up multi-stage axial compressors. It not only improves
surge margin at off-design condition but also gives nearly flat efficiency for wider
speed range. The present fan design is a transonic three-stage axial flow compressor
without inlet guide vanes. The aerodynamic performance of fan was limited by issues
of lower stability range at part speeds, and blade flutter was observed in engine
testing. So, it was proposed to redesign the fan with VIGV for addressing the above-
mentioned design issues. Fixed-flapping type variable inlet guide vanes are designed
to improve the performance of present fan. The design concept of VIGV itself will
be a new design and development in our country. As a result of introducing VIGV,
aerodynamic performance of redesigned fan is improved in terms of better surge
margin at speeds lower than 95% due to reduced inlet incidence on rotor—stage1.
The baseline fan was having stall flutter at part speed 80–85%; it is expected to get
eliminated due to reduced incidence and improved stall margin.
Keywords Fan · Variable inlet guide vane design · Inlet flow angle · Aerodynamic
performance · Surge margin
B. Kaur (B) · R. Abbas · A. Pratap

Gas Turbine Research Establishment, Bangalore, Karnataka, India
R. Abbas
A. Pratap
Defence Research & Development Organisation, Ministry of Defence, New Delhi, India

388 B. Kaur et al.
1 Introduction
Axial flow compressors are one of the most critical components in a gas turbine engine
pertaining to its operation in an adverse pressure gradient. To satisfy the demands of
high thrust to weight ratio, reduced fuel consumption and performance in a modern
day gas turbine, compressors are being pushed towards operating in the transonic
regime. The current research in aero-compressor field is devoted towards designing
compact and high performance compressors with characteristics such as high blade
loading, greater tip speeds, fewer stages and lesser axial gap between stages [1]. The
challenges associated with off-design operation of transonic compressors lies in a
mismatch of the incidence angle with that of the design conditions. This increases the
risk of blade stall leading to compressor surge, in addition to low efficiency and surge
margin of the compressor [2]. One method of countering the above problem is multi-
spooling, wherein the compressor stages are divided into two or more spools termed
as low-pressure and high-pressure spools that are coupled to respective turbines
through separate shafts. This enables us to rotate the low-pressure and high-pressure
compressor at different speeds, thus matching the design point incidence angle. In
many cases, variable inlet guide vanes (VIGV) are necessary in order to safely start
up multi-stage axial compressors. It not only improves surge margin at off-design
condition but also gives nearly flat efficiency for wider speed range. Alternative
method is to implement bleed control system but that causes loss to the engine
performance. Inlet guide vanes work as a regulating valve by adjusting the stagger
angle of each airfoil. It will lead to turn of the blades around a radial axis, so that inflow
to the following rotor is influenced in a positive way. As a result, the high incidences
on following rotor blades and correspondingly higher aerodynamic loadings can be
avoided [3]. Depending on the interface restriction and other engine requirements,
both the above methods are used at the discretion of the designer. For example, GE
F110 has a fan pressure ratio of 3.2:1 is equipped with VIGV, while Eurojet EJ200
employs the multiple spool configuration with three stages of low-pressure and five
stages of high-pressure compressor.
There have been many works dedicated towards establishing the performance
characteristics of centrifugal compressors equipped with variable guide vanes. Simon
et al. [4] reported the improvement in performance of centrifugal compressor by
simultaneously adjusting the inlet guide vanes and diffuser vanes. Experimental
determination of closure schedule for simultaneous adjustment of guide vane and
diffuser vanes was carried out in accordance with the required performance charac-
teristics. Rodgers [5] carried out compressor rig testing in a single-stage centrifugal
compressor, with moderately high specific speed and high inducer Mach number.
Testing showed that the surge margin increased with IGV regulation. The maximum
static pressure recovery and highest work done factor of the impeller occurred near
the stall conditions.
Similar research has been carried out that investigates the causes and effects of
off-design operation in axial flow compressors. Stephenson [6] highlighted that the
surging in axial flow compressor happens due to operation in condition far from
The Effect of Variable Inlet Guide Vanes … 389
design point. Limitation in the pressure ratio of the compressor and mass flow rate
is analysed. Handel et al. [7] investigated the transition and separation phenomenon
on inlet guide vane with symmetric profile used in an axial flow compressor. Exper-
imental investigation at different stagger angle and Reynolds number was carried
out. Results indicated that for acute stagger angles the flow separation occurred on
the pressure side trailing edge. Shaw et al. [8] experimentally investigated and vali-
dated using CFD, the effects of inlet guide vanes in performance characteristics of
transonic compressor. The experimental tests were conducted for both clean inlet
flow and distorted flow with and without IGV. The results indicated the accuracy of
CFD methods in predicting the flow characteristics in a transonic compressor with
inlet distortion. The maximum pressure and mass flow rate at stall were found to be
within 2% and 3% of the measured values, respectively. The VIGV showed a positive
effect on the performance of the fan when the swirl angle upstream of the rotor was
found to be more uniform. Broichhausen et al. [9] conducted experiments using static
wall pressure taps and semi-conductor pressure transducers, pneumatic hole probes
with incorporated pressure transducers to establish the off-design performance of
supersonic compressors with fixed and variable geometry. Semi-empirical models
were integrated with the through-flow algorithm to estimate in detail the losses and
deviation of the compressor performance in off-design condition. Williams et al. [10]
used an open-loop flow control algorithm to induce swirl in the flow downstream to
IGV using smart inlet guide vanes. Unlike mechanically actuated IGV, smart IGV
uses Coanda effect negating the need of articulating mechanism. The experiments
conducted demonstrated the ability of the guide vane to regulate the flow by 14° at
85% corrected RPM.
From the above discussion, it can be concluded that there have been significant
research carried out regarding the usage on inlet guide vanes and variable diffuser
vane in centrifugal compressors. In the case of axial flow compressors, there have
been many experimental works carried out to establish the effects of off-design
operation on the performance of transonic compressors. The effects of inlet distortion
and IGV articulation have been verified experimentally, which require sophisticated
test rigs and measuring probes to determine the compressor performance. However,
there is very less work that focuses on utilising numerical tools to verify the effect
that inlet guide vane actuation with varying compressor speed would have effect on
the surge margin and compressor stability.
The present fan design is a three stage transonic axial compressor without inlet
guide vanes. The aerodynamic performance of fan was limited by lower stall margin
at part speed, and blade flutter was observed at part speed in engine testing [11]. So,
it was proposed to redesign the fan with application of variable inlet guide vanes
(VIGV) to address the above-mentioned performance issues. Fixed-flapping type
variable inlet guide vanes are designed to improve the performance of present fan.
The design concept of VIGV itself will be a new design and development in our
country.
390 B. Kaur et al.
Table 1 Fan aerodynamic

Normalised parameter Design intent Achieved
performance at design point
(ISA-SLS) Mass flow 1.0 1.0154
Pressure ratio 1.0 0.994
Isentropic efficiency 85 86
(%)
Flutter Flutter free Flutter in rotor 1 at
NLR of 83% [13]
2 Fan–Baseline
Fan module consists of fan inlet casing and three stage transonic fan. Inlet casing
has eleven fixed struts inlet casing which has met requirement of uniform flow to the
inlet of fan and minimum loss. The gradients of velocity and static pressure in radial
direction are less than 1%, and total pressure loss is ~0.5% [12]. The aerodynamic
features of fan are given as below Table 1.
3 Variable Inlet Guide Vane (VIGV) Design
The VIGV system is capable of improving the compressor performance with the
turning of the vanes leading to an increase in vane incidence angle. This can cause
separations in the VIGV which reduces the desired flow deflection and produce addi-
tional losses. The classical configuration of IGVs consists of un-cambered or slightly
cambered profiles that are staggered. This configuration produces disadvantageous
flow physics, which cause comparatively large flow losses for small stagger angles.
A better configuration for IGVs consists of profiles with mechanical flaps as shown
in Fig. 1. VIGVs with mechanical flaps are applied in some military aircraft engines
for many years.
Inflow Outflow
Fig. 1 VIGV configuration consisting of profiles with flaps

3.1 Design Configuration of VIGV
The objective of incorporating VIGV was to assess relative improvement in flutter

and operability margins of fan which is presently integrated with gas turbine engine.
The fan axial length and tip diameter for are constraints from the baseline design
fan module. Keeping the inlet casing length same as baseline engine, available axial
length for VIGV is 160 mm. Considering tip solidity as one, 15 number of vanes was
decided for VIGV.
Design Trial 1
Initial design for VIGV was configured with chord-wise fixed to flapping part ratio
1:1. VIGV aerofoil considered is of symmetrical design as there is no preswirl at
design condition. The strength requirement of the inlet casing is same as the base-
line design. So, NACA0008 aerofoil is used for hub and tip section of IGV with
keeping the wetted area same. The above-mentioned configuration was analysed
using ANSYS CFX. It was found that maximum normalised mass flow of 0.977 was
passing through the inlet casing. Choking was occurred in the hub region, where
solidity was as high as 2.5. Distribution of swirl angle was not uniform also at the
hub locations, flow turning was very high. After this initial trial, it was decided that
hub chord to be reduced.
Design Trial 2
To address design problem from trial 1, the hub chord of VIGV reduced from 160
to 130 mm, which resulted in reduced hub solidity of 2. This design changes nearly
restored mass flow rate with pressure loss of 1%. But non-uniformity in the swirl
angle all most remains the same as baseline configuration of VIGV.
Design Trial 3
The objective of this trial was to have uniform swirl angle variation from hub to
tip. This can be done by increasing the chord-wise fixed to flapping part ratio at tip
location and reducing the same at hub location of VIGV. Numerous designs iterations
and analyses were carried out to finalise the split ratio. The final design of the VIGV
is described as given in Table 2.
Final configuration of the inlet casing with VIGV was analysed. This design has
the capacity to handle intended mass flow rate with a pressure loss of less than 1%
Table 2 Design parameters

Parameters Hub section aerofoil Tip section aerofoil
of VIGV
Aerofoil series NACA0008 NACA0006
Chord (mm) 130 160
Fixed part length 78 64
(mm)
Flapping part length 52 96
(mm)
392 B. Kaur et al.
Table 3 Aero design

Flap angle (º) Flow angle (º) IGV pressure loss (%)
parameters of VIGV
0 0 0.6
19 17.8 0.7
30 25.3 1.7
Fig. 2 Swirl angle

distribution from hub to tip
with meeting the flow angle requirements of the fan at part speeds. It has swirl angle
variation of ±2° from hub to tip. The detailed result of the VIGV is given in Table
3. Swirl angle distribution from hub to tip and flow filed around VIGV is shown in
Figs. 2 and 3, respectively.
4 Numerical Modelling and CFD
Reynolds averaged Navier-Stokes equation modified for turbomachinery calcula-

tions along with shear stress transport turbulence model [14] are used to numerically
simulate the axial flow compressor. CFX (ANSYS 17.0) is a pressure-based finite
volume solver with multi-block grid facility. In turbomachinery applications, conser-
vation of mass is an important parameter for judging the accuracy of the numerical
model [15], and for the same reason, finite volume model was used.
Fig. 3 Flow field around

IGV
4.1 Geometry and Grid Generation
The various inlet guide vane configurations with different vane angles were generated
by importing the co-ordinate files of the vanes, hub and the shroud. ANSYS turbo
grid 17 was used for the grid generation. The machine type is chosen as an axial
compressor with the principal axis of rotation as the X-axis. There are 15 vanes in
this system. For the hub, shroud and the vane, the B-spline curve type is used as it
generates a smooth curve obtained by interpolating the points provided in the co-
ordinate files. Initially, there are two layers: at the hub (span = 0) and at the shroud
(span = 1). In addition to that, three more layers are generated at spans of 0.25, 0.5
and 0.75. These layers project the topology onto a given span and improve the quality
of the mesh by creating a curve for the mesh to follow from the hub to the shroud.
Following this step, the 3D mesh is generated, and refinements are made to improve
the mesh quality. After the refinement, the mesh quality parameters are ascertained,
and necessary changes are made. The above steps were followed for generating grid
of all the rotor and stator configurations.
The topology of ’O’ grid close to surface and ’H’ grid in the blade to blade passage
with rotors mesh size of varying from 90,000 to 70,000 elements, and for stators,
varying from of 70,000 to 50,000 elements were used in decreasing order from first
stage to third stage, respectively; along with IGV, mesh size of 70,000 elements was
used. Rotor tip clearance of 0.5 mm was considered during meshing. A 3D model of
fan is shown in Fig. 4.
4.2 Pre-processing and Boundary Conditions
In the preprocessor, a turbomachinery simulation file is set-up. An axial compressor

is chosen as the machine type with X-axis as the rotational axis, and the analysis type
is chosen as steady state. Rotor–stator interface is defined by a stage model which
394 B. Kaur et al.
Fig. 4 3D view of the fan showing rotating and stationary stage
uses span-wise variation of pitch averages of all properties and redistributes mass,
momentum and energy based on pitch differences between the rows. The meshes of
the next stage rotors and stators are imported in a sequential order and integrated with
the meshes generated for the different vane angles. The rpms are varied from 100%
(design condition) for the fully open configuration of the vane to 70% for the fully
closed configuration. There are three boundary conditions available in the ANSYS
CFX-Pre 17 module, namely:
• P-Total Inlet P-Static Outlet
Here, the total pressure and the total temperature values at the inlet along with the
static pressure at the outlet need to be specified for evaluation of the problem.
• P-Total Inlet Mass Flow Outlet
In this condition, the total pressure and the total temperature at the inlet along
with the mass flow rate at the outlet are specified.
• Mass Flow Inlet and P-Static Outlet
In this condition, the mass flow rate and the total temperature at the inlet along
with the static pressure at the outlet are specified.
The boundary condition used for the analysis under consideration is the P-total
inlet P-static outlet. The inlet total pressure and the total temperature values are found
at ISA-SLS conditions which are 101.325 kPa and 288 K, respectively. The flow was
initialised with a velocity of 220 m/s. A high-resolution advection scheme is used,
and the number of iterations for each back pressure is set to 400. For convergence
control, the RMS value limit is set to 1E-6. Two user-defined expressions are created
to monitor the mass flow rate at the inlet and outlet conditions throughout the entire
solving phase. After completing the entire set-up, the preprocessor file is converted
into a ’.def’ file and used as the solver input.
4.3 Solver
The solver runs for either the total number of iterations as specified by the user or for
the convergence criteria set by the user, whichever is satisfied first [14]. Following
the completion of the run for starting back pressure, the static pressure at the outlet
is increased by 10 kPa and evaluated again using the previous run’s output as the
initial input for the next file. After the completion of each run, total pressure ratio
and the isentropic efficiency were evaluated, and subsequently, the next run file was
set-up. The above steps are repeated till the total pressure ratio starts to drop for a
particular back pressure.
4.4 Post-processor
The pressure and temperature variables at the inlet and outlet for the entire fan
components were extracted. Along with this, we extract the mass flow rate at the
inlet and outlet, velocities and Mach numbers at various spans of the blades. These
variables were to assess the performance of each component as well as plotting the
compressor maps.
4.5 CFD Model Validation
The present CFD model was validated by comparing the model with other similar
analysis published by the AGARD researchers [15]. In turbomachinery applications,
the commercial CFD codes under predict the efficiency of the rotor due to over
prediction of the tip leakage losses. Also there is an error in the pressure ratio predic-
tion due to the negligence of corner stall in the leading edge of the rotor blades [12].
Simoes et al. [16] used the similar CFD model and verified the suitability of the
three available turbulence models viz. k-ε, k-ω and shear stress transport model in
turbomachinery applications. Shear stress transport is the most suitable turbulence
model as it is confirmed with the fan analysis.
5 Performance Analysis
5.1 Characteristic Maps
The compressor maps are collective plots of the pressure ratio and efficiency vs. the
corrected mass flow rate at constant speeds. The pressure values and the mass flow
rates were taken from CFD post for each case of analysis. The isentropic efficiency
396 B. Kaur et al.
is calculated from the following equation:

γ −1

P2
P1
γ
−1
η= T2
(1)
T1
P and T denote total pressure and total temperature, respectively. Subscript 1

refers to inlet, and Subscript 2 refers to outlet. γ denotes ratio of specific heats for
air equal to 1.4.
The surge margin is given by the following equation [17]:

(Pr/m)surge
surge margin(%) = − 1 ∗ 100
(Pr/m)opearting
Pr and m denote total pressure ratio and total mass flow rate, respectively.
Subscript surge refers to stall point, and operating refers to design point.
5.2 Redesigned Fan with VIGV Performance Results
The 3D analysis was carried to estimate aerodynamic performance at all speeds. The
variability of inlet guide vane (IGV) has been introduced at off-design speeds to
lower the incidence on rotor 1 to enhance performance. This involved carrying out
the analysis at each of the lower speed with several iterations of different rotational
setting of the inlet guide vane to optimise the final choice of rotational setting for the
inlet guide vane to have satisfactory aerodynamic performance for that speed. Inlet
guide vane (IGV) is having a symmetrical aerofoil throughout with increasing chord
from hub to tip. For the variability of IGV, 40% of rear hub section was rotated through
desired angle at a particular speed; while at the tip, 60% rear section was rotated. This
configuration for rotational movement for IGV was chosen through various iterations
from the initial guess of configuration based on earlier experience and the literature
available regarding the same kind of fans and on the basis of configuration suitable
for mechanical design point of view. The optimised closure schedule followed for
variability of VIGV and stator at various speeds arrived through analysis is mentioned
below and shown in Fig. 5 (Table 4).
The performance analysis with optimised closure schedule for inlet guide vane
at design speed (10,312 rpm) and at low speeds, i.e. (100, 90, 85, 80 and 75%) was
carried out. The characteristics plots (overall pressure ratio vs. corrected mass flow
rate and overall efficiency vs. corrected mass flow) are shown in Fig. 6.
Fig. 5 VIGV closure schedule
Table 4 Variable inlet guide

NLR % VIGV flow an2008014gle (°)
vane (VIGV) closure schedule
100 and 95 0
90 5
80 15
85 25
75 35 (constant for speeds below 75% NLR)
5.3 Surge Margin Improvement
The surge margin improvement is listed below from the baseline fan to redesigned
fan with VIGV [18]. The change is surge margin was defined with respect to fan
operating line arrived after engine component matching vis compressor, combuster,
turbine, etc. (Table 5).
Characteristics at 75% Querypart speed were not obtained due to software solution
divergence as flow separation was very high in first rotor itself.
6 Results and Conclusion
The performance comparison of baseline fan and redesigned fan with VIGV is shown
in Figs. 7 and 8. The effect on aerodynamic performance of fan with introduction of
VIGV is summed up as given below.
398 B. Kaur et al.
Fig. 6 Characteristics plots—fan with VIGV
Table 5 Surge margin comparison

Change in surge margin with mod 2 Ncorrected (%) 100 95 90 85 80 75
+ VIGV Estimate −0.6 1.5 2.6 0.4 1.7 –
Fig. 7 Comparison characteristics plot (overall pressure ratio vs. corrected mass flow rate)
Fig. 8 Comparison characteristics plot (overall efficiency vs. corrected mass flow rate)
• The number of vanes are increased to 15 in count for VIGV as compared to that
earlier inlet casing was having 11 number of struts. As a result of more blocked
inlet area, the corrected mass flow at design point is reduced by 0.3 kg/s and
subsequently affected the operating point at lower speeds.
• It was observed that considerable reduction in the inlet Mach numbers, showing
that there is a corresponding reduction in shock losses by employing VIGV.
400 B. Kaur et al.
• At speeds lower than 95%, characteristics are shifted to lower mass flow rate due
to VIGV closure but stalling pressure has not changed significantly.
• The surge margin is improved at speeds lower than 90% due to reduced inlet
incidence on rotor—stage 1.
• The efficiency of redesigned Fan was improved at off design conditions as rotor
incidences were brought down at operating line and mass flow rate from choke to
stall got extended.
• The baseline Fan was having stall flutter at part speed 80–85%. With the incor-
poration of VIGV, it is expected to get eliminated due to reduced incidence and
improved stall margin.
• Part speed stall witnessed at high altitude low Mach No. during engine restart
accelerating is expected to get alleviated due to reduced fan mass flow and Fan
rotor 1 blade incidences.
References
1. Sanders AJ, Fleeter S (2000) Experimental investigation of rotor-inlet guide vane interactions
in transonic axial-flow compressor. J Propul Power 16:421–430
2. Kerrobrock JL (1981) Flow in transonic compressors, AIAA 80–0124R. AIAA J 19:4–19
3. Farokhi S (2014) Aircraft propulsion, 2nd edn. Wiley, West Sassex, United Kingdom
4. Simon H, Wallmann T, Monk T (1987) Improvements in performance characteristics of single-
stage and multistage centrifugal compressors by simultaneous adjustments if inlet guide vanes
and diffuser vanes. J Turbomach 109:41–47
5. Rodgers C (1991) Centrifugal compressor inlet guide vanes for increased surge margin. J
Turbomach 113:696–702
6. Stephenson JM (1981) A solution of stall problem in axial-flow compressors. Readers Forum
67–69
7. Handel D, Rockstroh U, Niehuis R (2014) Experimental investigation of transition and sepera-
tion phenomenon on an inlet guide vane with symmetric profile at different stagger angles and
Reynolds Number. In: 15th international symposium on transport phenomenon and dynamics
of rotating machinery, pp 1–9
8. Shaw MJ, Heilel P, Tucker PG (2014) The effect of inlet guide vanes on inlet flow distortion
transfer and transonic fan stability. J Turbomach 136
9. Broichhausen KD, Gallus HE, Monig R (1988) Off-design performance of supersonic
compressors with fixed and variable geometry. J Turbomach 110:312–321
10. Williams DR, Cornelius D, Acharya M, James D, Marshall T (2006) Smart inlet guide vanes
for active flow vectoring in an axial compressor. In: 3rd AIAA flow control conference, 1–7.
11. Compressor Group, Gas dynamic stability of fan and compressor module. Report No. 0002 00
03 00 06 004 056, Gas Turbine Research Establishment, Bangalore
12. Compressor Group, Design of K9+ LP compressor. Report No. 0001 00 02 09 00 002 019, Gas
Turbine Research Establishment, Bangalore
13. Compressor Group, Aeromechanical testing of MOD II fan with casing treatment. Report No.
DRDO-GTRE-CPR-FSR-020–2012, Gas Turbine Research Establishment, Bangalore
14. User & Theory Documentation, CFX-ANSYS 17.0, ANSYS Inc.
15. Dumhan J (1988) CFD validation for propulsion system components. AGARD Advisory Report
355, AGARD, France
16. Simoes MR, Montojos BG, Moura NR, Sujian JS (2009) Validation of turbulance models for
simulation of axial flow compressor. In: 20th international congress of mechanical engineering,
1–9
17. Cohen H, Rogers GFC, Saravanamutoo HIH (1996) Gas Turbine Theory, 4th edn. Longman
Group Limited, Essex, England
18. Compressor Group, Aerodynamic design of new fan with high inlet pressure distortion toler-
ance. DRDO-GTRE-CPR-FSR-141–2015, Gas Turbine Research Establishment, Bangalore
Transition Prediction for Flow Over
a MAV Wing Using the Correlation
Based Model
M. B. Subrahmanya and B. N. Rajani
Abstract In this work, a low aspect ratio MAV fixed wing at a relatively low
Reynolds number wherein the flow undergoes transition is analysed. The effective-
ness of the correlation based transition model γ -Reθ SST proposed by Menter and
Langtry (Correlation based transition modeling for unstuctured parallelized compu-
tational fluid dynamics codes. AIAA J 47:2894–2906 [7]) is brought out by making
vis-a-vis comparison with the pure turbulence model SST (Turbulence, heat and mass
transfer vol 4. Begell House Inc., pp 625–626 [6]). The transition model is able to
handle separated flow transition and gives more insight to flow than the turbulence
model. Some of the results depicting the transitional flows are presented and the
superiority of the transitional model over the pure turbulence model is demonstrated.
Keywords γ -Reθ SST Transition model · Low Re · Laminar separation bubble ·

Transition onset · Pressure based finite volume method
1 Introduction
The aircraft industry in the last couple of decades has seen a rapid growth in the
miniaturised and unmanned aerial vehicles like the MAVs which find application
for both military and civil purposes. MAVs are designed based on specific missions
they are meant to accomplish. These are low aspect ratio vehicles with high lift thin
wings either fixed or flexible or flapping and flying at low speeds and are usually
made of light weight composites to enhance their range and/or duration of flight. The
flow is very complex and transition plays a significant role because of low Reynolds
number and low aspect ratio thin wing sections. Owing to their dimensions and flight
Supported by CSIR-NAL
M. B. Subrahmanya (B) · B. N. Rajani

CSIR-NAL, Bengaluru 560017, India
B. N. Rajani

404 M. B. Subrahmanya and B. N. Rajani
speeds they pose lot of challenges to the designers and especially to the computational
fluid dynamics (CFD) research community. Therefore to design an efficient MAV, a
thorough research in multiple disciplines is needed. This has led to lot of innovations
and research in small and more efficient motors and batteries, light weight airframe
structures and high lift generating surfaces, highly efficient propellers, light-weight
and efficient autopilots and control systems, light weight sensors and cameras and so
on. The stability of the vehicle is another important factor due to its dimensions and
weight. The fixed wing MAVs are known to be more stable compared to the flexible
and flapping wings.
Now looking at the fluid dynamics point of view, the flow regime of MAVs is
104 < Re < 105 where transition, separation and reattachment are very predomi-
nant features. Further in this regime, the flow tends to become unsteady even at lower
angles of attack. Even today the transitional flows are not fully understood and lot of
research is still on going in the area of CFD to explore this grey area. The behaviour
and the underlying aerodynamics of MAVs is quite different and more complex from
conventional aircrafts which are their larger counterparts. The flow over MAV wings
are predominantly transitional leading to the formation of the laminar separation
bubble (LSB) and we lack a proper understanding of these complex flows coupled to
their geometrical complexities. There are few approaches to handle transitional flows
numerically. One can use either a simple empirical method or a more sophisticated
approach like the direct numerical simulation (DNS). But from the designers point of
view, a viable tool like RANS is most desirable since this approach greatly reduces
the computational cost and time compared to the much complicated DNS while an
acceptable accurate mean quantities are still possible to compute. Several researchers
have developed RANS based approaches to handle transitional flows. Some of them
are, Abhu-Ghannam and Shaw correlation based model coupled to turbulence mod-
els, low Reynolds number versions of turbulence model, point transition approach
and transport equation based transition models viz. kT -kL -ω model and γ -Reθ SST
transition model. The transport equations based transition model is one of the recent
approach to model transition. Here additional transport equations are solved to han-
dle the effects of transition. A correlation based transition model γ -Reθ SST was
proposed by Menter and Langtry [7] using local variables based on his SST turbu-
lence model. Along with the original SST model transport equations, this model uses
additional two transport equations. One of the equation is for the intermittency (γ )
and the other for the momentum thickness based Reynolds number (Reθ ) inorder
to calculate the location of transition onset. The current work is on a 300 mm class
fixed wing MAV where flow is transitional because of the flight speed and its physical
dimensions. The γ -Reθ SST transition model has been used to model the transitional
flow. The effects of transition modelling is brought out by comparing results with the
results of base turbulence model SST. The flow solution code used is 3D-PURLES
(3D Pressure based Unsteady RANS LES solver) which is an in-house developed
URANS and LES code equipped with various turbulence and transition models [10].
This code is extensively used for both external and internal aerodynamics problems.
The γ -Reθ transition model recently implemented in 3D-PURLES is validated for
the ERCOFTAC flat plate test cases and SD7003 airfoil [5].
Transition Prediction for Flow Over a MAV Wing Using the Correlation Based Model 405
2 Mathematical Formulation
Three-dimensional incompressible flow simulation is carried out by numerical solv-

ing of the Navier Stokes (NS) equations. The numerical simulation of the NS equation
needs an appropriate mathematical model to handle the geometrical complexities and
the physical complexities.
2.1 Governing Equation
The Reynolds averaged Navier Stokes equations for unsteady incompressible flow
in the coordinate-free form:
Mass conservation:
.ρU = 0 (1)
Momentum conservation:
Dt ρU = − P + .((μ + μt )(U + t U )) (2)
where μ and ρ are fluid viscosity and density, p and U are the pressure and velocity
vector, respectively. The eddy viscosity μt is evaluated through turbulence/transition
models. The γ -Reθ SST model implemented in 3D-PURLES uses the constants,
damping function, etc. as given by Menter [7] except for the two correlation functions
viz. ReθC and Flength which are adopted from Malan et al. [4]. However, the ReθC
correlation has been suitably modified as given below based on our validation study
carried out for the flat plate T3A test case [5].

ReθC = min 0.665 R̃eθt + 66.5, R̃eθt (3)
2.2 Flow Solution Procedure
The simulations are carried out using the in-house multi-block finite volume flow
solution code 3D-PURLES [3, 10]. This code uses the SIMPLE algorithm to solve
the three-dimensional unsteady incompressible Navier Stokes equation in non-
orthogonal curvilinear coordinates using the collocated variable arrangement. The
SIMPLE algorithm is suitably modified for the collocated variable arrangement [2]
to avoid the checkerboard oscillations of the flow variables. The system of linear
equations derived from the finite volume procedure is solved in a decoupled manner
for the velocity components, pressure correction and turbulence scalars using the
strongly implicit procedure of Stone [11]. The effect of turbulence and transition
is modelled either using different eddy viscosity based Reynolds averaged Navier
Stokes (RANS) models or Large Eddy Simulation (LES).
The transition prediction capability of the γ -Reθ SST model for different angles of
attack is analysed for the 300 mm MAV wing at Re = 1.7 × 105 based on the mean
chord (C) of the root section.
3.1 Computational Details
The MAV considered for this study is a modified cropped delta wing with modified
Eppler-61 as its cross-section. The wing semispan is 0.625C having a planform area
of 0.5140C 2 . H-H grid topology consisting of 12 blocks is generated for the semi
span MAV wing using the in-house grid generation code. The computational domain,
boundary conditions and grid used for the present simulation are shown in Fig. 1.
At the farfield boundary, depending on the sign of the convective flux either an inlet
boundary condition is applied for the velocity components and turbulent scalars or an
outflow boundary is applied where the normal gradients of the velocity components
and turbulent scalars are made zero. At the wing surface, no slip condition (U = 0,
V = 0 and W = 0) is specified. At the block boundary, one overlap control volume
is provided on the either side of the block interface boundary for appropriate transfer
of the solution from the neighbouring block. The near wall boundary conditions for
the turbulence scalars (k, ω γ and Reθ ) have been appropriately used depending on
the model used. The turbulence intensity (Tu) at the inlet is assumed to be 1% of the
freestream velocity.
3.2 Effect of Grid Refinement
The MAV wing analysis using the in-house flow solution code 3D-PURLES is car-
ried up to the stall angle using a coarser grid (193 × 126 × 97) having a near wall
y + ranging between 11 to 15. The effect of grid refinement on the aerodynamic
characteristics of the wing is studied using the SST turbulence model. For the grid
refinement study, a finer grid of size 295 × 166 × 129 with near wall y + < 1 is
generated and simulations are carried at three typical angles of attack (low, middle
and high). The aerodynamic coefficients obtained using the coarse and finer grid
are compared with the NAL wind tunnel measurement data [1] and is shown in
Fig. 2. It is quite evident from this figure that refining the grid has not brought in a
Fig. 1 Boundary conditions

and H-H grid used in the
present simulation
(a) Boundary conditions

Y
X
(b) Grid on the symmetry plane and the wing surface
significant improvement in aerodynamic coefficients indicating that the coarser grid

will suffice. It is further observed that, a reasonable good match with the measurement
data is obtained for the drag coefficient (Cd ) and moment coefficient (Cm ). However
the present computation has underpredicted the lift coefficient with the measurement
predicting a very early stall (α = 16◦ ). This early stall behaviour is not observed in
the measurements [8, 12] for similar class of low aspect ratio (AR) wings (AR < 3)
where the stall in general is obtained around α = 22◦ .
1.4 0.6
1.2
1
0.4
0.8
0.6
0.2
0.4
0.2
0 0
0 4 8 12 16 20 24 28 0 4 8 12 16 20 24 28
(a) Coefficient of lift (b) Coefficient of drag
-0.1
-0.2
-0.3
-0.4
-0.5
0 4 8 12 16 20 24 28
(c) Coefficient of moment
Fig. 2 Effect of grid size on aerodynamic coefficients using SST turbulence model, Re = 1.7 × 105
3.3 Effect of Transition Model
The coarser grid is used to simulate the effect of modelling the transition in the
RANS framework. The γ -Reθ transition model implemented in the in-house code
3D-PULRES is used for this simulation and its performance is asserted by comparing
with fully turbulent SST results. The flow characteristics obtained by γ -Reθ transition
model is compared with SST in detail for α = 12◦ along with the aerodynamic
characteristics. The surface pressure coefficient (−Cp ) and skin friction coefficient
(Cf ) distribution at three spanwise locations (root, mid and near tip) on the wing
surface obtained by the transition and turbulence model for α = 12◦ are shown
in Fig. 3. The Cp distributions predicted by the transition model as well as the
2 0.05
1.5 0.04
Near Tip
1 0.03
Mid span
-Cp
Root
Cf
Mid span
Root
Near tip
0.5 0.02
0 0.01
-0.5 0
-1 -0.01
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/c x/c
(a) Surface pressure coefficient (b) Surface skin friction coefficient

Fig. 3 Variation of the wing surface coefficients at α = 12◦ , Re = 1.7 × 105 (solid line: γ -Reθ
model, dashed line: SST model)
turbulence model at the three locations are very similar. However, a notable difference
is observed in the Cf distribution with the difference increasing as we move from the
root to tip which may be due to the difference in the turbulence level obtained by the
two models. The eddy viscosity ratio (μt /μ) contours on the cross planes (Fig. 4)
obtained by γ -Reθ model for α = 12◦ indicate that the flow is laminar almost upto
the mid wing. The γ -Reθ model has predicted a small negative Cf on the tip section
close to the leading edge (0.264 ≤ x/c ≤ 0.28) of the wing which indicates that there
is a flow separation which is also evident in the surface streamlines (Fig. 4a). The
separation is laminar since the flow becomes turbulent at this span location beyond
x/c = 0.35 as indicated in the Reynolds shear stress contours (Fig. 5a). The Reynolds
shear stress (τx z ) contours for α = 12◦ at different planes along the wingspan are
shown in Fig. 5. The dashed line shown in Fig. 5a is the indicative transition onset
line drawn based on the shear stress criteria [9]. In this criteria, the onset location is
determined wherever the Reynolds shear stress (τx z ) exceeds more than 0.001 of the
mean kinetic energy and a clear visible rise is observed. The τx z contours obtained
by the SST turbulence model (Fig. 5b) clearly indicate that the flow is turbulent right
from the leading edge of the wing.
The shift of transition onset location with angle of attack is captured by the γ -Reθ
model as depicted by eddy viscosity ratio contours at different cross planes (Fig. 6)
and τx z contours predicted at different spanwise planes (Fig. 7). The trend of the
transition onset shifting upstream with increase in angle is captured by the present
computation. The level of turbulence is observed to be greatly enhanced at the higher
angle leading to an early transition when compared to the lower angle of attack.
(a) γ-Reθ SST model (b) SST Turbulence model
Fig. 4 Surface streamlines and eddy viscosity ratio contours on the different cross planes α = 12◦ ,
Re = 1.7 × 105
xz:
(a) γ-Reθ model (b) SST Turbulence model
Fig. 5 Reynolds shear stress contours at different planes across the wingspan at α = 12◦ , Re =
1.7 × 105 (dashed line: indicative transition onset)
The transition onset line shown in Fig. 7 indicates that at α = 4◦ the onset at
root of the wing occurs at x/c = 0.3 and moves slightly downstream at wing tip
(x/c = 0.42) whereas for α = 28◦ the onset occurs close to the leading edge of the
wing (x/c = 0.25) and remains almost constant along the wingspan. The surface
streamlines indicate that at α = 4◦ (Fig. 6a) and α = 12◦ (Fig. 4a) the flow is smooth
with the formation of the separation line at the tip of the wing which eventually forms
the tip vortex. Whereas, at α = 28◦ (Fig. 6b) the flow is complex with separation
occurring at the root itself. The volume streamlines at three different angles of attack
are shown in Fig. 8 which clearly indicate the formation of the tip vortex and its
strength increases with angle of attack.
(a) α = 4◦ (b) α = 24◦
Fig. 6 Surface streamlines and eddy viscosity ratio contours on the different cross planes planes at
xz:
(a) α = 4◦ (b) α = 24◦

Fig. 7 Reynolds shear stress contours at different planes across the wingspan for Re = 1.7 × 105
using γ -Reθ model (dashed line: indicative transition onset)
The aerodynamic coefficient predicted by γ -Reθ SST model using the two grids
are compared with the present SST results and NAL measurement and is shown
in Fig. 9. Inspite of the difference observed in the flow features predicted by the
transition model and SST turbulence model the aerodynamic coefficients are almost
identical in the linear regime. However the transition model has predicted the stall
early by 2 degree. Simulations using the finner grid are also carried out at few angles
of attack. Similar to the SST turbulence model, no significant improvement in the
aerodynamic coefficients was obtained for the γ -Reθ transition model by using the
finer grid.
(a) α = 4◦ (b) α = 12◦
(c) α = 24◦
Fig. 8 Volume streamlines obtained at different angles of attack at Re = 1.7 × 105
4 Concluding Remarks
The transitional flow analysis past a 300 mm class fixed wing of a MAV is carried
out. The simulation done within the RANS framework has shown the significance
of modelling the transition while comparing the results of base turbulence model.
The γ -Reθ SST transition model is able to capture the separated flow transition.
Thus an engineering approach to handle the transitional flows past an aerodynamic
configuration like MAV is demonstrated.
1.4 0.6
x
1.2
x
1 x
0.4
x
0.8 x
x
x
0.6 x
x
0.2
0.4 x
x
x x
0.2
x
x x x
0 0
0 4 8 12 16 20 24 28 0 4 8 12 16 20 24 28
0
x
x
-0.1
x
-0.2 x
x
-0.3
x
x
-0.4 x
-0.5
0 4 8 12 16 20 24 28
(c) Coefficient of moment

Fig. 9 Effect of transition on the aerodynamic coefficients
References
1. Arivoli D, Ravi D, Roshan A, Suraj CS, Ramesh G, Sajeer A (2011) Experimental studies
on a propelled micro air vehicle. In: 29th AIAA applied aerodynamics conference, Honolulu,
Hawaii, 2011-3656, pp 1–10
2. Majumdar S (1988) Role of underrelaxation in momentum interpolation for calculation of flow
with non-staggered grids. Numer Heat Transf 13:125–132
3. Majumdar S, Rajani BN, Kulkarni DS, Mohan S (2003) RANS computation of low speed
turbulent flow in complex configuration. In: Symposium on state of the art and future trends
of CFD. NAL SP0301. NAL, Bangalore, pp 31–48
4. Malan P, Suluksna K, Juntasaro E (2009) Calibrating the γ -Reθ transition model for commercial
CFD. In: 47th AIAA aerospace sciences meeting, Orlando, Florida, 2009-1142
5. Manu YC, Rajesh A, Subrahmanya MB, Kulkarni DS, Rajani BN (2015) Simulations using
transition models within the framework of RANS. In: Sengupta TK et al (eds) Advances
in computaion. Modeling and control of transitional and turbulent flows. World Scientific,
Singapore
6. Menter FR, Kuntz M, Langtry R (2003) Ten years of industrial experience with the SST
turbulence model. In: Hanjaliac K, Tummers M (eds) Turbulence, heat and mass transfer vol
4. Begell House Inc., pp 625–626
7. Menter FR, Langtry R (2009) Correlation based transition modeling for unstuctured parallelized
computational fluid dynamics codes. AIAA J 47:2894–2906
8. Mizoguchi M, Itoh H (2013) Effect of aspect ratio on aerodynamic characteristics at low
Reynolds number. AIAA J 51(7):1631–1639
9. Radespiel R, Windte J, Scholz U (2007) Numerical & experimental flow analysis of moving
airfoils with laminar separation bubbles. AIAA J 45:1346–1356
10. Shetty P, Subrahmanya MB, Kulkarni DS, Rajani BN (2013) CFD simulation of flow past MAV
wings. Int J Aerosp Innov 5(1):19–27
11. Stone HL (1968) Iterative solution of implicit approximations of multidimensional partial
differential equations. SIAM J Numer Anal 5:530–530
12. Torres, Gabriel E, Thomas JM (2004) Low-aspect-ratio wing aerodynamics at low Reynolds
numbers. AIAA J 42:865–873
Rotor Flow Analysis in the Presence of
Fuselage Using Unsteady Panel Method
K. R. Srilatha, Premalatha, and Vidyadhar Y. Mudkavi
Abstract Panel methods are known to be simple yet effective during initial design
stages. With the advent of advanced CFD methods and high speed computers, there
is a feeling that these methods are no longer needed. On the contrary, experience
with many practical problems shows that their utility is significant. There are also
some problems where panel methods may be even superior to CFD approaches. One
such problem is that of flow field analysis of rotors in the presence of stationary
fuselage. In this paper we make use of an unsteady panel method which is a simple
extension of a steady panel method to assess the effect of presence of fuselage on the
rotor wake. Another significant advantage an unsteady panel method offers is that it
is particularly easy to ‘fly’ the rotor in the presence of fuselage and assess the wake.
Typical flight scenarios include hover, forward flight, ascent and descent flights.
Some specific applications include assessment of wake flow for weapon separation
from helicopters and estimates of noise due to unsteady rotor loading.
Keywords Unsteady panel method · Rotor · Fuselage
1 Introduction
Panel methods [1, 2] are lowest in the hierarchy of flow analysis tools and are gener-
ally not classified as mainstream CFD. Notwithstanding this, they played significant
role in aircraft design and development. Almost every design house made use of panel
methods during initial design stages. More recently, there is a perception that these
K. R. Srilatha (B) · Premalatha

CTFD Division, CSIR NAL, Bangalore, India
Premalatha
V. Y. Mudkavi
CSIR Fourth Paradigm Institute, Bangalore, India

416 K. R. Srilatha et al.
methods can be dispensed with owing to advances in CFD and availability of signif-
icant computing power. However, this perception is misplaced as they still continue
to play major role in several areas. In fact, when flow regimes are genuinely linear,
these methods are as good as any advanced CFD method, provided there are no flow
separations and compressibility effects are small. It should, however, be noted that
in aerospace applications, flow separations still remain challenging to CFD methods
and when separations do occur, the flow is generally nonlinear.
Some of the applications where panel methods are still effective include rotor
flows [3], both unsteady and steady, rotor noise (as the noise in this case is due to
fluctuating loads) [4], fluid-structure interaction, simultaneous analysis of stationary
and moving parts such as rotor and fuselage interaction studies.
At CSIR National Aerospace Laboratories (CSIR-NAL), panel methods have been
used for a variety of problems. Wind turbine analysis and marine propeller [5, 6]
analyses were carried out with success assuming steady flow. Propeller performance
degradation for light transport aircraft in pusher configuration was estimated com-
bining panel method under quasi-steady assumption with Euler solution for a live
project [7, 8]. Recently the capability of the panel method was extended to include
unsteady effects and applied to pitching wing and interaction of helicopter main and
tail rotor wakes. In this paper we describe an attempt to study the interaction of
helicopter main rotor and fuselage flow which has applications in stores separation
from helicopter platforms.
2 Methodology
Panel method solves the linearised potential flow equation, also known as the Prandtl-
Glauert equation. The basic assumptions are that the flow is linear (small perturba-
tions), irrotational and inviscid. The Prandtl-Glauert equation is transformed into
the Laplace equation which, in turn, is converted into an integral equation over the
bounding surfaces. This reduces the three-dimensional problem by one order and
makes it possible to obtain flow solutions about completely arbitrary configurations
with little effort. The integral equation is solved by discretization of the configuration
surface into plane panels and distributing singularities such as sources and doublets
whose strengths are determined subject to boundary conditions.
The mathematical formulation is described briefly here [1, 2]. With respect to the
body fixed frame of reference, the body surface is denoted by
Sb (r ) = 0. (1)
where r is the radial coordinate and the subscript b denotes the body. The governing
equation for the velocity potential is the Laplace equation
∇ 2 (r, t) = 0. (2)
Rotor Flow Analysis in the Presence of Fuselage Using Unsteady Panel Method 417
In panel method analysis, this potential is split into potential at infinity ∞ which
represents uniform undisturbed flow U∞ and a disturbance potential ϕ. Then
(r, t) = ∞ + ϕ(r, t), ∇∞ = U∞ (3)
At infinity, ϕ vanishes and on the boundary of S we have

∇ϕ(r, t) − U∞ (r, t) · n(r, t) = 0. (4)
where n is the normal.

The problem is solved by distributing sources σ and doublets μ. The sources
account for non-lifting part of the solution and the doublets represent the lifting part
of the solution. On using Green’s theorem, the problem is recast as

σ ∂ 1 ∂ 1
ϕ(r, t) = + μ dS + μ dS. (5)
r ∂n r ∂n r
b w
where the second integral is carried out over the wake w.

As noted before, the body being analysed is discretised into plane polygonal (nor-
mally quadrilateral) panels and closed form expressions for ϕ and induced velocities
for a planar polygonal element with constant distribution of source [1, 2] are used
to cast this problem into a matrix problem.
The numerical procedure consists of dividing the wing and mean surface into a
number of small polygonal (quadrilateral, in practice) planar surfaces. Only constant
sources, whose strengths are unknown at the moment, are distributed on each of
the surface panels. Vortex panels which are equivalent to the doublet distribution
are placed on mean camber surface of lifting components much like a collection of
horse-shoe elements forming a number of columns. The vortex elements leave the
trailing edges where one Kutta panel is placed for each horse-shoe column. Assuming
that the strength of doublet elements are proportional to the local thickness of the
lifting element, one needs to only determine the constant of proportionality per Kutta
panel by applying Kutta condition. The application of boundary condition leads to
a system of linear equations which is easily solved for singularity strengths from
which one can calculate other flow parameters.
This part of the procedure is common to both steady and unsteady panel methods.
When the flow is unsteady [3], the procedure is as follows. At time t = 0, the body
is impulsively started from rest at a given initial position. At this instant there is no
wake. The singularity (sources and doublets) strengths are determined by solving a
linear system as described above. Using these singularities, the induced velocities
at the downstream corner point of the Kutta panel are evaluated. A straight vortex
element of known strength is released from the downstream points of each Kutta
panel. The ends of this vortex element is moved with the induced velocities and
other movement of the body, if any. Now this forms the next configuration at time
t + t where t is the time step. Once a vortex element (or equivalently a doublet
element) enters the wake, its strength is kept constant. For the next time step, the new
system of equations with wake elements is solved. With each time step the vortex
panels in the wake grows and are never removed from the computations.
At each time step the potential and induced velocities at all the control points
is evaluated from the known singularity distributions. The pressure coefficient at a
control point is evaluated by unsteady Bernoulli’s equation
2
|U | 2 ∂ϕ
cp = 1 − + 2 ∂t
. (6)
U∞ U∞
The time derivative of the potential in Eq. 2 is evaluated by the following approx-
imate expression
∂ϕ ϕ(t + t) − ϕ(t)
= . (7)
∂t t
2.1 Non-dimensionalization
For the purpose of non-dimensionalization, we prescribe U∞ in terms of its compo-

nents Vx , Vy , Vz in m/s and use it as reference velocity. However, when U∞ is zero,
as in the case of hover, the reference velocity is taken as the rotor tip speed.
We present the results for a representative rotor with 4 blades fixed atop ROBIN
fuselage [9] (without the pylon). Four flight cases are considered, viz., vertical climb,
forward climb, hover and forward motion. Results with and without the presence of
fuselage are compared to assess the effect of the presence of the fuselage on the flow.
The fuselage is represented using 400 panels while each blade is represented
using 301 panels. The time steps are coarse for these studies mainly because one
is interested in a quick assessment. The paneling is shown in Fig. 1. We use body
fixed coordinate system where x is along the fuselage positive from tail to nose, z is
positive vertically up and y is positive towards portside of the fuselage.
Y
Z
X
Y
Z X
Y X
Fig. 1 Three views of Robin fuselage and 4 bladed rotor after panelling. Pylon over the fuselage
is not considered for the present case
3.1 Vertical and Forward Climb
For the case of vertical climb, vertical velocity of Vz = 1 m/s is prescribed and the
analysis is carried out for two rotor revolutions. Time step corresponds to increment of
azimuthal angle by 15◦ . Thus, 48 time steps are required to complete two revolutions.
For the case of forward climb, an additional forward velocity of Vx = 1 m/s is
prescribed over and above vertical climb of 1 m/s. The time step is the same as for
vertical climb. Figures 2 and 3 shows the results. Here, CA is the coefficient of force
along z, CW is the force coefficient along x. Thus, CA is akin to lift while CW is akin
to drag. CM represents the pitching moment about y.
For the rotor alone case, the force and moment coefficients do not exhibit much
variations. There are mild oscillations whose period coincides with rotor motion.
However, the addition of fuselage shows that the force and moment coefficients are
significantly modified. The flow is characteristically different.
In the case of combined forward and climb motion, there is a significant variation
in CA and CW . This is expected because the rotor will undergo significant change in
angle of attack due to rotation and forward motion. The periodicity is same as that of
rotor motion. While CA and CW are nearly symmetric over one rotation, CM exhibits
asymmetry which is also as expected. The most interesting feature in contrast to
climb alone case is that the addition of fuselage has very little effect on CA and CW
while there is reasonable modification to CM .
0.06
CA
0.04
Rotor + Fuselage Rotor
-0.16
CW
-0.2
-0.5
CM
-0.55
-0.6
0 2 4 6 8 10 12 14
Ω
Fig. 2 Comparison of force and moment coefficients with and without fuselage for vertical climb.
Vz = 1.0, Vx = Vy = 0. Azimuthal step = 15◦
3.2 Hover and Forward Flight
Next we consider hover and forward flight. The panelling is the same as the previous
case. We also retain the same time step and perform simulations for two revolutions
of the rotor. The results are shown in Figs. 4 and 5.
0.04
0.02
0
CA
-0.02
-0.04
-0.1
-0.15
CW
-0.2
-0.25
-0.3
CM
-0.35
0 2 4 6 8 10 12 14
Ω
Fig. 3 Comparison of force and moment coefficients with and without fuselage for forward climb.
Vx = Vz = 1.0, Vy = 0. Azimuthal step = 15◦
For the case of hover, we observe very little difference between case with and
without fuselage. Essentially, hover is very close to steady state and the fuselage
shows insignificant changes. The periodicity is maintained in both the cases.
For the forward flight, we specify forward velocity of Vx = 1 m/s. We note that
there is a characteristic difference in CW and CM . The effect of CA is much less.
-0.005
-0.006
CA
-0.007
0.00017
0.00012
7x10-5
CW
-5
2x10
-5
-3x10
-5
-8x10
0.012
0.008
CM
0.004
0
0 2 4 6 8 10 12 14
Ω
Fig. 4 Comparison of force and moment coefficients with and without rotor for hover flight.
Vx = Vy = Vz = 0.0. Azimuthal step = 15◦
-0.002
CA
-0.006

CW
-0.0005
-0.001
0.006
CM
0.002
0 2 4 6 8 10 12 14
Ω
Fig. 5 Comparison of force and moment coefficients with and without rotor for forward flight.
Vx = 1.0, Vy = Vz = 0.0. Azimuthal step = 15◦
Applications of an unsteady panel method for complex problem of rotor flow field
analysis in the presence of fuselage are carried out. Though panel methods suffer
from several theoretical limitations, these applications indicate that essential features
of flow phenomenon can be extracted. The results show the effect of fuselage is
enhanced in some flight conditions like vertical climb and forward flight while in
other cases effects are minimal. The method is useful in understanding qualitative
flow features while it may be as yet difficult to ascertain accuracy of results for
such complex flows unlike when the flow conditions are genuinely linear. It may
be possible to enhance their utility by performing computations for a much better
resolved geometry and associated smaller time steps. The flow features for the present
calculations, however, exhibit sufficient smoothness.
References
1. Hess JL, Smith AMO (1962) Calculation of non-lifting potential flow about arbitrary three-
dimensional bodies. Technical report E. S. 40622. Douglas Aircraft Company, USA
2. Hess JL (1972) Calculation of potential flow about arbitrary three-dimensional lifting bodies.
Technical report MDC J5679-01. Douglas Aircraft Company, USA
3. Ahmed SR, Vidjaja VT (1994) Numerical simulation of subsonic unsteady flow around wings
and rotors. In: AIAA applied aerodynamics conference. No. AIAA-94-1943-CP, pp 938–950
4. Yin JP, Ahmed SR (1999) Aerodynamics and aeroacoustics of helicopter main rotor/tail rotor
interaction. In: AIAA applied aerodynamics conference. No AIAA99-1929, A99-27283. pp
834–844
5. Narayana CL (1998) Panel method calculations for the DTMB P4119 propeller. In: ITTC propul-
sion committee propeller RANS/Panel method workshop, Grenoble, France
6. Narayana CL (2002) Analysis of propeller by panel method. In: NRB seminar on marine hydro-
dynamics. NSTL, Visakhapatnam, India
7. Narayana CL, Srilatha KR (2004) Aerodynamic analysis of Saras propeller by a panel method.
CSIR-NAL Project Document CF 0403. CSIR National Aerospace Laboratories, Bangalore,
India
8. Srilatha KR, Narayana CL, Premalatha, Mudkavi VY (2007) Performance analysis of Aft
mounted propeller for a light transport aircraft. In: 7th Asian computational fluid dynamics
(ACFD7) conference, Bangalore, India. pp 107–115
9. Schweitzer S (1999)Computational simulation of flow around helicopter fuselages. Masters
thesis, The Pennsylvania State University, USA
Numerical Analysis of High Reynolds
Number Effects on the Performance
of GAW-1 Airfoil
Abstract The aerodynamic behaviour of the GAW-1 airfoil at high Reynolds num-
ber is analysed numerically using the Spalart-Allmaras (SA) turbulence model.
Initially an inter-code comparison is carried out at Re = 6 × 106 and the aerody-
namics characteristics obtained using the in-house flow solution code 3D-PURLES
and the open source CFD tools SU2 and OpenFOAM are validated with the available
measurement data. Based on this validation exercise, 3D-PURLES is used to study
the effect of increasing Reynolds number (1, 3, 6, 9, 12 millions) on the aerodynamic
characteristics.
Keywords Spalart Allmaras turbulence model · Aerodynamic characteristics ·

Drag count · SU2 · OpenFOAM · 3D-PURLES
1 Introduction
The GAW-1 is a 17% thick airfoil which has been designed for general aviation
applications. Detailed measurements for this airfoil was conducted by McGhee and
Beasley [1] for a range of Mach number (0.1 ≤ M ≤ 0.28) and Reynolds num-
ber (2 × 106 ≤ Re ≤ 2 × 107 ) at different angles of attack (−10◦ ≤ α ≤ 24◦ ). The
performance of the wing is greatly influenced by the airfoil characteristics. Therefore
it becomes important to understand the effect of Reynolds number on airfoil char-
acteristics. The Clmax and Cdmin are two important parameters [2] to characterize
the airfoil especially at high Reynolds number. The high Reynolds number flows are
usually encountered in wind turbines [3, 4] and transport aircrafts [5]. In the present
study, numerical simulation of GAW-1 airfoil is initially carried out using three flow
solution codes viz. in-house 3D-PURLES (3D Pressure based Unsteady RANS LES
D. S. Kulkarni · B. N. Rajani (B)

CSIR-NAL, Bengaluru 560017, India
D. S. Kulkarni

426 D. S. Kulkarni and B. N. Rajani
solver), OpenFOAM and SU2 which is validated against the measurement data [1]
at Mach number 0.2 and Reynolds number 6 × 106 . On the basis of this validation
exercise, the in-house flow code 3D-PURLES is used to analyze the effect of high
Reynolds number on the aerodynamic performance of GAW-1 airfoil.
2 Methodology
The in-house grid generation code [6] is used to generate the 2D grid around GAW-
1 airfoil. This 2D grid generation code is based on simple differential-algebraic
hybrid procedure to generate boundary-orthogonal body fitted grids. All the flow
computations are carried out using the Spalart-Allmaras (SA) turbulence model [7]
on the C-grid having a grid size of 527 × 101 with the farfield placed at 15 times the
airfoil chord ( C ). The near wall grid spacing is adjusted by stretching the grids along
the normal direction in order to obtain the desired near wall y + < 1. The boundary
conditions and grid used for present simulation are shown in Fig. 1. At the wall
boundary, no slip condition is applied for the velocity and the turbulent scalar ν̃ is set
to zero. The turbulence intensity at the far field is prescribed as 1% of the freestream
velocity for all the computations. The convergence criteria has been fixed to be 10−5
for all the three solvers. The details of the solvers are described briefly below.
2.1 3D-PURLES—In-House Incompressible Flow Solver
3D-PURLES [8] is an in-house flow solution code to solve the unsteady turbu-
lent incompressible flows. The code is based on SIMPLE algorithm modified for
Wall
Farfield Farfield
(a) C-grid (527 × 101) for GAW-1 airfoil (b) Boundary condition
Fig. 1 Grid and boundary condition used for airfoil simulations

Numerical Analysis of High Reynolds Number … 427
collocated variable arrangement [9]. The system of linear equations derived from the
finite volume procedure is solved in a decoupled manner for the velocity components,
pressure correction and turbulence scalars using the strongly implicit procedure of
Stone [10]. For the present study, 3D-PURLES computations are carried out using
QUICK scheme for spatial discretization of convective flux of the momentum equa-
tions. The upwind spatial discretization scheme is used to the compute the convective
flux of the turbulent scalar equation. Whereas, the diffusion (viscous) flux is com-
puted using the central difference scheme. The farfield boundary is treated either as
an inflow (flow is prescribed) or as an outflow (zero gradient) depending on the sign
of the convective flux on the relevant face. For the inflow condition, uniform velocity,
the turbulent viscosity (μt ) equal to the laminar viscosity (μ) and the turbulent scalar
ν̃ = μρ are prescribed.
2.2 OpenFOAM—Open Source Incompressible Flow Solver
OpenFOAM is a free open source CFD software developed in C++. This software
package is capable of simulating a wide variety of fluid flow processes. The package
has over 80 solver modules, to handle incompressible turbulent flows, compressible
turbulent flows, chemically reacting flows, porous medium, multiphase physics, heat
transfer, acoustics and fluid structure interactions. In the present study, OpenFOAM
computations are carried out using simpleFoam coupled to the second order linear
upwind scheme for the convective flux of the momentum equation and the central
difference scheme for the viscous flux. The convective flux of the turbulent equation is
discretized using the upwind scheme. At the farfield, freestream boundary condition
is applied which switches between fixed (inlet) value and zero gradient based on
the sign of the flux. For the inlet condition, uniform velocity, μt = μ and ν̃ = μρ are
prescribed.
2.3 SU2—Open Source Compressible Flow Solver
The Stanford University Unstructured (SU2) open source software suite was devel-
oped for solving partial differential equations (PDE) and PDE constrained optimiza-
tion problems. The core suite is a RANS solver capable of simulating the com-
pressible, turbulent engineering problems. The SU2 computations are carried out for
M = 0.2 using the second order Roe scheme for spatial discretization of convec-
tive flux with Venkatakrishnan’s limiter for the primitive variables. The convective
flux of the turbulent scalar and the viscous flux is calculated using upwind scheme
and corrected average gradient method respectively. Implicit local time-stepping has
been used to obtain the steady-state solution, and the linear system is solved using
the GMRES method. The characteristic based farfield boundary condition is applied
to the outer domain and adiabatic wall condition is enforced on the airfoil surface.
At the inlet, uniform velocity is prescribed. μt = 10μ and ν̃ = 3 μρ are fixed in the
code.
3.1 CFD Code Validation at Re = 6 × 106
The aerodynamic coefficients obtained for GAW-1 airfoil at Re = 6 × 106 using

the three flow solution codes are shown in Fig. 2 and is observed to closely follow the
measurement data [1]. The variation of drag polar obtained by OpenFOAM and 3D-
PURLES are almost identically and closer to the measurement when compared to the
SU2. However all the three codes have consistently predicted a higher drag coefficient
(Cd ) when compared to the measurement. On the other hand, all the three codes have
underpredicted the pitching moment coefficient (Cm ) with 3D-PURLES and SU2
being closer to the measurement. Table 1 compares the aerodynamic performance
parameters obtained by the three flow codes. Stall angle (αstall ) predicted by the
3D-PURLES and the OpenFOAM are identical and closer to the measurement data,
whereas SU2 has predicted an early stall. The Clmax obtained by 3D-PURLES is
observed to be closest to the measurement data. The angle of attack at which the lift
coefficient (Cl ) becomes zero obtained by the three codes is observed to be slightly
overpredicted when compared to the measurement with OpenFOAM having the
maximum difference of 0.4◦ . The Cd at this angle of attack (Cd0 ) is also overpredicted
with maximum error of 24% obtained by 3D-PURLES and SU2 whereas OpenFOAM
(a) Drag polar (b) Coefficient of pitching moment
Fig. 2 Variation of aerodynamic coefficients for GAW-1 airfoil at Re = 6 × 106

Table 1 Aerodynamic performance parameters for GAW-1 airfoil at Re = 6 × 106

Data set Clmax αstall Cd0 Cdmin
3D-PURLES 1.92 19◦ 107.1 × 10−4 , 104.7 × 10−4 ,
α = −4.25◦ α = −2◦
OpenFOAM 1.870 19◦ 101.4 × 10−4 , 99.7 × 10−4 ,
α = −4.40◦ α = −2◦
SU2 (Ma = 0.2) 1.872 16◦ 106.3 × 10−4 , 105.8 × 10−4 ,
α = −4.25◦ α = −4◦
Measurement [1] 1.943 18◦ 86.1 × 10−4 , 73.3 × 10−4 ,
α = −4◦ α = −2◦
is having the minimum error of 18%. Similar to to the Cd0 , the minimum drag
coefficient (Cdmin ) is also overpredicted. The overprediction of the drag coefficient
may be due to the errors and ad-hoc assumptions associated with the turbulence
modelling.
3.2 Effect of Reynolds Number
The in-house code 3D-PURLES is used to analyze the effect of Reynolds number
on the aerodynamic characteristics of GAW-1 airfoil. The variation of the aerody-
namic coefficients obtained for five Reynolds numbers (1, 3, 6, 9 and 12 million) are
shown in Fig 3 and has captured the expected trend [4, 11], The maximum lift and
stall angle are observed to increase with Re whereas the drag coefficient reduces with
Re. This effect is also reflected in the L/D plot with maximum L/D occurring at the
same angle of attack (α = 8◦ ) for all the Re. The momentum coefficient (Cm ) is also
found to increase with Re. The maximum difference in the aerodynamic coefficient
is found between 1 million Reynolds number to 3 million Reynolds number and this
difference reduces at higher Re. The difference is almost negligible between 9 and
12 million Reynolds number indicating the insensitivity of the aerodynamic char-
acteristics beyond Re = 9 × 106 similar to the experimental results [4]. Figure 4
shows the trend of the minimum drag coefficient (Cdmin ) expressed in drag count
(D.C = Cd × 104 ), maximum lift coefficient (Clmax ), maximum ((L/D)max ) and lift
slope (between −4◦ and 4◦ ) with Reynolds number. A decreasing trend of the min-
imum drag coefficient and increasing trend of both maximum lift coefficient and
maximum L/D is observed as Reynolds number increases. It is further noticed that
the above trend tends to become almost constant beyond 9 million Reynolds number.
On the other hand, the slope of the lift coefficient curve in the linear region obtained
at different Reynolds is found to be nearly same and hence may be treated as constant
over this Reynolds number range.
430
(c) Lift to drag ratio (d) Coefficient of pitching moment

Fig. 3 Variation of aerodynamic coefficients with the angle of attack for different Reynolds numbers
(a) Minimum drag coefficient (b) Maximum lift coefficient
(c) Maximum L/D (d) Lift slope
Fig. 4 Variation of aerodynamic parameters with Reynolds number
The flow simulations with SA turbulence model were carried out for GAW-1 air-
foil at Re = 6 × 106 using the two incompressible flow solvers viz., 3D-PURLES,
OpenFOAM and the compressible flow solver SU2. The results indicate that the stall
angle and Clmax obtained by 3D-PURLES were closer to the measurement data but
has overpredicted the drag coefficient when compared to OpenFOAM. Based on this
code validation study, the in-house code 3D-PURLES was successfully used to study
the effect of high Reynolds numbers (1, 3, 6, 9 and 12 million) on the aerodynamic
performance of GAW-1 airfoil. This study has captured the expected trend with lift
coefficient increasing and drag coefficient decreasing with Reynolds number. The
Cdmin was found to decrease with Reynolds number where as an increasing trend
was obtained for Clmax , stall angle and (L/D)max . The maximum difference in the
aerodynamic coefficient was obtained between 1 and 3 million Reynolds numbers
and this difference was found to reduce at higher Reynolds numbers. Further, this
difference was observed to be quite negligible for Reynolds number between 9 and
12 million indicating the insensitivity of the aerodynamic coefficient beyond 9 mil-
lion Reynolds number. The slope of the lift curve in the linear regime was found to
be almost same for all the five Reynolds numbers.
Acknowledgements The authors acknowledge the Director, CSIR-NAL, Bangalore and Head,
CTFD Division, CSIR-NAL, Bangalore for giving us the opportunity to carry out this work and
permitting us to publish. We sincerely thank Dr. V. Ramesh Jt. Head CTFD Division, CSIR-
NAL, Bangalore for his support and fruitful discussion. We wish to express our special thanks to
Mr. K. Madhu Babu Scientist CTFD Division, CSIR-NAL, Bangalore for providing the airfoil
geometry data.
References
1. McGhee RJ, Beasley WD (1973) NASA TN D-7428

2. Yamauchi GK (1983) Trends of Reynolds number effects on Two-dimensional airfoil charac-
teristics for helicopter rotor analyses. Technical report TM 84363, NASA
3. Kang S, Li J, Fan Z (2009) In: Inaugural US-EU-China thermophysics conference-renewable
energy 2009 (UECTC 2009 proceedings
4. Pires O, Munduate X, Ceyhan O, Jacobs M, Snel H (2016) J Phys : Conf Ser 753:1
5. Clark PW, Pelkman RA (2001) In: 39th AIAA aerospace meeting and exhibit. Reno, NV
6. Fathima A, Baldawa NS, Pal S, Majumdar S (1994) Grid generation for arbitrary 2D config-
urations using a differential algebraic hybrid method. Technical report PD CF 9461. National
Aerospace Laboratories, Bangalore
7. Spalart PR, Allamaras SR (1992) In: AIAA paper 92–0439
8. Shetty P, Subrahmanya MB, Kulkarni DS, Rajani BN (2013) Int J Aerosp Innov 5(1):19
9. Majumdar S (1988) Numer Heat Transfer 13:125
10. Stone HL (1968) SIAM J Numer Anal 5:530
11. Abbott HI, Doenhoff AEV (1949) Theory of wing sections. Dover Publications Inc., New York
Investigation of the Effect of Booster
Attachment Scheme on the Rolling
Moment Characteristics
of an Asymmetric Vehicle Using CFD
P. K. Sinha, Munish Kumar Ralh, Naveed Ali, R. Krishnamohan Rao,

and G. Balu
Abstract The rolling moment characteristics of a launch vehicle (LV) arising out
of joining an air-breathing cruise vehicle (CV) and a booster is investigated through
CFD to explain experimental observed behaviour. The launch vehicle has a stabilising
fin at the rear and is placed on bearing to have free roll with respect to body. The basic
and control rolling moment of LV (without stabilising fin, to mimic freely rolling
fin), obtained experimentally, is compared with the corresponding experimental data
for CV and found that the values differ, contrary to that observed in literature. CFD is
used to investigate the reason for this difference. Investigation in two representative
Mach numbers 0.8 (subsonic) and 2.0 (supersonic), has revealed that the interstage
flare in LV, has significant effect in modifying the rolling moment contribution of
the CV fin. Apart from that, the booster attachment arm, launch shoes are also
having some impact in modifying the rolling moment. It is evident that for a canard
controlled vehicle, even if the stabilising fin is freely rolling, the after-body geometry
has significant effect in dictating the aerodynamic characteristics, especially rolling
moment.
Keywords Asymmetric vehicle · Rolling moment · Booster attachment ·

Experiment · CFD
Abbreviations
LV Launch Vehicle
CV Cruise Vehicle
RF Rolling Fin
CRM Coefficient of Rolling Moment
CN Coefficient of Normal Force
Cm Coefficient of pitching moment
P. K. Sinha (B) · M. K. Ralh · N. Ali · R. Krishnamohan Rao · G. Balu

Defence Research and Development Laboratory, Hyderabad 500 058, India

434 P. K. Sinha et al.
ϕ = Phi Roll orientation

α = A = Alpha Angle of attack
δ=D Fin deflection
M Mach number
L Length of vehicle
1 Introduction
Aerodynamics of an asymmetric vehicles, like the one shown in Fig. 1, is very

challenging. The vehicle, called CV, is basically an air-breathing vehicle having
two side mounted intakes. The configuration, in consideration, has intakes in closed
condition. It has four control fins in cruciform position at the rear end, for vehicle
control. The length behind the fins are very small ~0.38 d where d is the smallest
diameter. It can be seen from Fig. 1 that the configuration is symmetric about pitch
plane but it is asymmetric about yaw plane. Due to these geometric features, the
longitudinal characteristics show a non-linear variation with angle of attack [1] due to
cross-coupling and also the characteristics vary with roll orientation. The longitudinal
characteristics of the vehicle and comparison of CFD with experimental data is
already discussed in Ref. [1]. The lateral characteristics for this vehicle are also
asymmetric with roll orientation [2], in contrast to the behaviour observed in case of
symmetric configurations. In fact it is seen that, in intake closed condition, beyond α
~7° the rolling moment keeps on increasing very fast because of the cross coupling.
Hence, it is desirable that the vehicle roll orientation is maintained around the view
shown in Fig. 1, defined as ϕ = 0 orientation where the intakes are in the windward
side and also the angle of attack is kept low. Any deviation from ϕ = 0 orientation or
any asymmetric disturbance also makes the body vulnerable to roll. Hence, a tight
control on the vehicle roll is desired from vehicle performance point of view. Proper
estimates of lateral characteristics especially rolling moment, both basic and control,
is very important for these kind of vehicles as they play an important role in vehicle
performance and control. It is also required to have a good understanding of the
Actuator cover
Launch shoe
1.5°Flare
Wire Tunnel
Control Fin =00 roll posion

Intake
Fig. 1 Cruise vehicle (CV) configuration

Investigation of the Effect of Booster Attachment Scheme … 435
Inter stage flare Rolling fins

Aachment arm
Control fins
d
Intake
1.62d
CV Booster
Fig. 2 Launch vehicle (LV) configuration
factors which affect their values. Though experiment is the only reliable means of
estimating the characteristics, it may not be cost-effective to understand the factors
responsible for the values through experiments. In such cases, a validated CFD code
can throw sufficient light on those aspects. Though the vehicle is designed for air
launch, to prove certain technology, ground launch is preferred.
For ground launch, booster with a specified diameter, (almost twice that of CV)
is attached at the rear end of the CV. Ground booster is used to boost the vehicle to a
desired Mach number and altitude from where the CV can be brought to required air
launch condition. CV and LV are connected through a flare at the interstage due to
diameter difference between the two. Also, the interstage length is restricted in order
to restrict the overall length of LV which has increased the flare angle. The interstage
length is kept at 1.62d, where, d is the smallest diameter of LV. Integrated vehicle
termed as Launch Vehicle or LV, shown in Fig. 2, is designed to meet the vehicle
stability and the control requirement. Since, pitch/yaw control is not effective in boost
phase, it was decided to make the vehicle sufficiently stable so that pitch/yaw control
can be avoided. To maintain the roll orientation and control vehicle roll rate, only roll
control using CV fin is used in this phase. Vehicle stability is achieved by placing
stabilising fin on booster. During the initial studies using CFD, it was observed
that a huge roll rate is developed if the stabilising fin is fixed and an acceptable fin
misalignment is considered. The roll control effectiveness of the control fin, which
is part of the CV, is unable to keep the roll rate within allowable limit due to induced
rolling moment on the stabilising fin. In order to decouple the induced rolling moment
on the stabilising fins, the fins are placed on a bearing as suggested by Auman and
Kreegar [3]. This makes the fin free to rotate and the rolling moment contribution
from the booster fin do not get transferred to the body. As a result, it is expected that
the rolling moment characteristics will be governed mostly by that of CV. Auman
and Kreegar [3] has shown experimentally, that for the body with single diameter,
the rolling characteristics of the vehicle are nearly same with rolling tail fin and
without tail fin. Hence, it is expected that when only cylindrical booster (without any
protrusions) is attached to the CV, the rolling characteristics of LV will be same as that
of CV. But in LV configuration, CV is attached to booster through some attachment
scheme using a flare configuration, as mentioned earlier. Though the stabilising fin is
on bearing, the presence of other portrusions viz. launch shoes, booster attachment
arms may have influence on the rolling moment characteristics of LV. Further, though
rolling moment due to stabilising fin is decoupled, the yawing moment and pitching
moment contribution of these fins will be present during roll control and has to
be considered for LV characterization. This is because it becomes equivalent to

canard controlled vehicle having lifting surface and the vehicle characteristics are
governed by CV fin downwash effect on the portrusions and the stabilising fin.
In order to characterize LV aerodynamics, experiments are performed for the LV
configuration [4]. To validate the findings of Auman and Kreegar [3], the rolling
moment characteristics of LV, viz basic and control, are compared with that of CV
obtained from separate test [2]. A difference in rolling moment values is observed
which will be discussed in detail in the subsequent sections. In order to understand
the reason for the difference, CFD simulations are performed. In this paper, the
experimental observations in respect of rolling moment will be discussed and the
CFD based explanations to justify the difference will be dealt with in detail.
2 Experimental Observation
The effect of booster attachment scheme on rolling moment characteristics of LV, is

estimated through wind tunnel tests. The tests are performed in 0.6 m × 0.6 m tunnel
of NTAF, NAL for configuration without stabilising fin [4] which will be equivalent
to that with fin sitting on a free bearing as the rolling moment contribution from
booster fin is removed in both the cases. The CV characteristics are also obtained
from wind tunnel tests, carried out in 1.2 m × 1.2 m tunnel of NTAF, NAL [2]. To
confirm the findings of Auman and Kreegar [3], on rolling moment characteristics, the
corresponding data (both basic and control) of LV and CV were compared. Typical
comparison of basic rolling moment for CV and LV (without booster fin) are made in
Fig. 3a, b respectively for M = 0.8 and M = 2.0 at roll orientation, ϕ = 22.5 (non-zero
roll orientation is selected for better comparison). It is observed that for α > 5°, the
basic rolling moment of LV is lower than that of CV (by maximum 15%) at M = 0.8
whereas at M = 2.0, the difference is much less. Similarly, the comparison of rolling
moment with control deflection 10° (in roll sense) for CV and LV is presented in
Fig. 4a, b respectively for M = 0.8 and 2.0. It is found that control rolling moment is
lower for LV in comparison to CV. The difference is found to be more at lower angle
of attack and less at higher angle of attack for both Mach = 0.8 and 2.0. The difference
is found to be more (maximum ~20%) in M = 0.8 (transonic Mach numbers) and
CRM Vs. Alpha,Phi=22.5 deg.,M=0.8 CRM Vs. Alpha,Phi=22.5 deg.,M=2.0

1 1
CV,Expt. CV,Expt.
0.8 0.8
LV,Expt
LV,Expt.
0.6 0.6 LV,CFD
LV,CFD
CRM
CV,CFD
CRM
0.4 0.4
CV,CFD
0.2 0.2
(b)
0
(a) 0
0 5 10 15 0 5 10 15
-0.2 -0.2
Alpha Alpha
Fig. 3 Comparison of computed CRM for LV and CV with Experiment a M = 0.8, b M = 2.0
CRM Vs. Alpha,M=0.8,Droll=10 Deg. CRM Vs. Alpha,M=2.0,Droll=10 Deg.

-1
-1
-10 -5 0 5 10 15
-10 -5 0 5 10 15 -1.1
-1.2 -1.2
CV,Expt CV,Expt
-1.3
LV,Expt. LV,Expt
-1.4
-1.4
CV,CFD CV,CFD
CRM
-1.5
CRM
-1.6 LV,CFD
LV,CFD
-1.6
-1.8 -1.7
-1.8
-2
-1.9 (b)
(a) -2
-2.2
Alpha
Alpha
Fig. 4 Comparison of Rolling moment for CV and LV from experiment at Droll = 10° a M = 0.8,
b M = 2.0
comparatively less (~10%) at M = 2.0 (supersonic Mach numbers). It is expected that

at supersonic Mach numbers, as the disturbance from the flare will be insignificant,
the difference in rolling moment should be negligible. But experiment showed a
difference, especially for δ(Droll) = 10°. Understanding the cause of this difference
in rolling moment characteristics through experiment is not cost effective. Though
it is known that predicting rolling moment values through CFD is very challenging,
it is planned to be investigated numerically with a view that it will be possible to
obtain the relative difference in the values and the reason for it.
3 CFD Simulations
To understand the experimentally observed rolling moment characteristics, CFD

simulations are carried out for both LV (without booster fin) and CV to obtain basic
rolling moment and control rolling moment at conditions identical to experiment i.e.
M = 0.8 and 2.0, δ = 0° and 10° (roll sense).
3.1 Details of Solver and Grid Generator
Commercial CFD software, HEXPRESS [5] is used for grid generation and Fluent-
15 [6] is used for flow simulation. The grid generator generates a hybrid grid and
the flow solver solves 3-D Reynolds Averaged Navier Stokes (RANS) equation on
unstructured grid based on finite volume approach. It also solves one of the following
turbulence models viz. k-ε, k-ω or SST turbulence model, etc. along with the RANS
equations.
The computational domain for the subsonic flow simulation of CV is selected
with boundaries located sufficiently far from the body as shown in Fig. 5, where, L
being the length of the vehicle, whereas for supersonic flow simulation boundaries
Fig. 5 Typical
computational domain for
CV
5L
1.5L 4L
Subsonic Domain
5L
Fig. 6 Typical surface grid

for CV
are placed at a maximum distance of 2L from the body. Typical surface grid for CV
is shown in Fig. 6. It can be seen that grids are sufficiently refined in the region of
intake entry, wire tunnel, launch shoes and fins. Typical grid size for CV is about 17
Million. At inflow boundary, uniform conditions of Mach number, static pressure and
static temperature are imposed. At outlet boundary, the supersonic outflow boundary
condition is imposed. No-slip and adiabatic wall boundary conditions are prescribed
at the solid wall. Second-order scheme for mass, momentum and energy and first-
order scheme for turbulence quantities are used for solution. A log normalized rms
residue of about 1e–04 has been set as the convergence criteria. Same philosophy
is followed in domain selection and grid generation for LV. Typical surface grid for
LV (zoomed around interstage) is shown in Fig. 7. The final grid size selected for
simulation is about 18 Million.
3.2 Validation of Longitudinal Characteristics for CV and LV
The CFD computation for CV and LV is carried out according to the solution method-
ology described above. The longitudinal characteristics especially the normal force
Fig. 7 Typical surface grid

for LV (zoomed around
interstage region)
CN(Closed Intake),M=0.8 Cm(Closed Intake),M=0.8

10 50
Expt.
Expt.
CFD
5 CFD 0
-10 -5 0 5 10
Cm
CN
0 -50
-10 -5 0 5 10
-5 -100
Alpha Alpha
CN(Closed Intake),M=2.0 Cm(Closed Intake),M=2.0

10 50
Expt. Expt.
CFD CFD
5 0
-10 -5 0 5 10
Cm
CN
0 -50
-10 -5 0 5 10
-5 -100
Alpha Alpha
Fig. 8 Validation of longitudinal characteristics for CV
coefficient CN and pitching moment coefficient Cm are compared for CV in Fig. 8

and for LV in Fig. 9. Good match of the values are observed. The rolling moment
coefficient obtained from the same simulations are compared in subsequent sections.
3.3 Comparison of Rolling Moment
The basic rolling moment coefficient (CRM) data for LV and CV, obtained from CFD
is first validated with experiment at ϕ = 22.5°. Typical comparison of computed basic
rolling moment for LV (without booster fin) and CV at ϕ = 22.5° with experiment,
is presented in Fig. 3a, b for M = 0.8 and 2.0, respectively. It is observed from the
RF Conf.: CmNose, φ = 0o, M=0.8
500 Expt.
Spline fit and EI
400
CFD
300
200
100
Cm(nose)
0
-100
-200
-300
-400
-500
-20 -15 -10 -5 0 5 10 15 20

Alpha(deg)
Fig. 9 Validation of longitudinal characteristics for LV
comparison that in case of CFD also, at α = 5° the CRM of both LV and CV is nearly
same as in experiment but at α = 10°, the basic rolling moment for LV is lower than
CV. The trend is same at both the Mach numbers i.e. M = 0.8 and M = 2.0. It may
be observed from Fig. 3a, b that at α = 10°, the relative difference in CRM values is
more at M = 0.8 (~20%) than at M = 2.0 where it is ~10% as observed in experiment.
It may be further observed that though CFD is unable to capture absolute values of
CRM, it is able to capture well the trend of CRM variation. At supersonic Mach
numbers, the variation of relative difference in CRM between CV and LV with angle
of attack could not be captured through CFD.
Similarly, the comparison of control rolling moment between experiment and
CFD is shown in Fig. 4a, b for the same Mach number and angle of attack conditions
as above. In this case also, the CFD is unable to capture the absolute values of CRM,
but once again, it has predicted lower CRM for LV in comparison to CV as in the
case of experiment. Though CFD data is able to capture the angle of attack effect
on the relative difference of CRM between CV and LV for M = 0.8, it has failed
to capture the effect in M = 2.0. The variation of computed CRM of LV (without
booster fin) for 10° deflection of fins (in roll sense), with Mach number is plotted
against that from experiment for LV in Fig. 10a, b for α = 5° and 10° respectively. It
CFD Vs. Expt. CFD Vs. Expt.

2.0 2.5
(a)
(b)
2.0
1.5
A=5,D=0,CFD 1.5
1.0 A=10,D=0,CFD
A=5,D=10(roll),CFD
CRM
CRM
1.0 A=10,D=10(roll),CFD
A=5,D=0,Expt
0.5 A=10,D=0,Expt.
A=5,D=10(roll),Expt. 0.5
A=10,D=10(roll),Expt.
0.0
0 0.5 1 1.5 2 2.5 3 0.0
0 0.5 1 1.5 2 2.5 3
-0.5 -0.5
Mach No.
Mach No.
Fig. 10 Comparison of CRM LV (without B2 Fin), a A = 5° b, A = 10°

is observed that CFD is able to predict the experimental trend of CRM variation with
Mach number well for α = 5°, but for α = 10° the difference between experiment
and computation is high at supersonic Mach number. As both experiment and CFD
are showing lower rolling moment for LV, in comparison to CV, with and without roll
control deflections, it is necessary to understand the reason for it. Since, CFD is not
able to capture the experimental trend of CRM variation for M = 2.0, the analysis is
restricted to M = 0.8.
3.4 Discussions and Analysis
In this case, CFD data has been used for analysis, where component wise contribution
to CRM can be obtained. Component wise contribution of major roll producing
components of LV are plotted in comparison to CV for α = 5° and 10° for M =
0.8 in Fig. 11. It is found that at M = 0.8, there is significant reduction in CV
fin contribution for LV. In addition, the four arms, that attaches CV to the booster,
generate an opposing rolling moment. Because of the presence of the interstage flare
and attachment components, the pressure distribution on fin, intakes and actuators
are getting altered in LV case in comparison to CV case. The pressure contours for
the CV fin with and without booster attachment are compared in Fig. 12. It clearly
brings out the effect of booster attachment in altering the fin pressure distribution
Component contribuƟon to CRM,

Component contribuƟon to CRM,
M=0.8,A=10 Deg.,D=10 Deg.
M=0.8,A=5 Deg.,D=10deg.
LV
AƩachment
LV (b)
AƩachment
CV
CV
(a) Actuator
Actuator
CV fin
CV fin
Intake Intake
-0.5 0.0 0.5 1.0 1.5 2.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
Fig. 11 Component wise contribution to CRM for M = 0.8 a A = 5 and b A = 10°
(a) (b)
Fig. 12 Pressure distribution at Fin TE a CV b LV (with booster) for M = 0.8, α = 10° and δ roll
= 10°
around trailing edges (TE) which is responsible in reducing the fin contribution to
rolling moment. As a result, their contribution is reducing. But, in case of M = 2.0,
reasons are not very clear which is causing the lower value of rolling moment in
case of LV. In summary, it can be stated that the reason for the difference in rolling
moment between LV and CV is mainly due to reduction in fin contribution apart from
contribution from intake fairing, actuator and attachment arms in subsonic/transonic
Mach numbers. It brings that after body has significant effect on the rolling moment
characteristics with canard controlled fin geometry.
4 Conclusions
The effect of booster attachment scheme on the rolling moment characteristics are
investigated both experimentally and numerically. It is found from the experimental
data, obtained from wind tunnel tests, that the rolling moment (both basic and control)
of Launch vehicle is lower than that of Cruise Vehicle. The difference in values
decreases with angle of attack. Two Mach numbers viz. M = 0.8 and 2.0 are selected
for analysis which represents subsonic and supersonic Mach number regime. It is
found from experiment that the relative difference in values are more at subsonic
Mach number. To understand the reason for this difference, CFD computations are
performed. CFD code is first validated for the longitudinal characteristics of the
configurations and the detail analysis has been made in respect of rolling moment. It
is found that computed rolling moment values differ considerably from experiment.
However, it is able to capture the experimental trend, in respect of LV values lower
than that of CV. It also captured the trend of the effect of angle of attack on rolling
moment difference between CV and LV for M = 0.8, but it failed to capture the
trend for M = 2.0. The analysis showed that the CV fin has the maximum effect
in reducing the rolling moment for LV. Actually, booster flare has significant effect
in altering the pressure in fin trailing edge region which is responsible in bringing
down fin contribution to rolling moment. Apart from control fin, booster attachment
components are also having marginal contribution in reducing the rolling moment
of launch vehicle. It may be noticed that in the literature, the free-spinning tail was
on the same diameter as the canard fin and was effective in eliminating roll coupling
and reducing other out of plane forces and moments. But in this case, though the
tail fin is free to roll, the rolling moment is affected as the tail fin and canard fin
are placed on two different diameters. It may be concluded that even if the tail fin is
freely rolling, the control fin after-body has significant effect in altering the rolling
moment characteristics of a canard controlled vehicle.
Acknowledgements The authors would like to acknowledge the support and encouragement of
Director, DRDL and Project Director for their constant support and encouragement in their activity
of characterizing one of most complex aerodynamic configuration. The support of Ms. Ankita Jain,
Contract engineer, in mesh generation and CFD simulation is highly acknowledged.
References
1. Ralh MK, Sinha PK, Theerthamalai P (2017) Aerodynamic characteristics of an Aerospace

vehicle with side mounted air-intake in closed condition at subsonic and supersonic flow. In:
Proceedings NCWT-05, MIT, Chennai, Mar 2017
2. Nath B, Sreenivasan N, Sampath Rao B Force measurement on a 1:3.56 scale model of SFDR
configuration. Doc. No. PD-NTF/2016/1011
3. Auman LM, Kreegar RE Aerodynamic characteristics of a canard controlled missile with a free
spinning tail. AIAA-1998–0410
4. Sreenivasan N, Nath B, Joshi K, Sampath Rao B Force measurement on 1:10.2 scale model of
SFDR with ground booster configuration. Doc. No. PD-NTF/2017/1014
5. User’s Manual for NUMECA-FINE/Hexa V2.10–4”, Numeca International, Brussels, Belgium
6. Ansys Fluent-14.5 theory and users guide-Ansys inc, India (2013)
Multi-objective Optimization Approach
for Low RCS Aerodynamic Design
of Aerospace Structures
P. S. Shibu, Sandeep, Balamati Choudhury, and R. U. Nair
Abstract This paper presents a multidisciplinary, multi-objective optimization

approach towards low Radar Cross section (RCS) aerodynamic design of aerospace
structures. As a typical example, design and analysis of aircraft intake duct for stealth
behavior by integrating computational fluid dynamics (CFD) and computational
Electromagnetics (CEM) has been demonstrated. The design constraints (inlet area,
throat area, exit area, and diameter) are calculated based on the RAE M2129 diffuser
and subsonic flow condition with Mach number 0.8 is considered at the Indian stan-
dard atmospheric conditions (ISA_SL + 15). Inlet shaping parameters for intake are
modified based on super ellipse equation by retaining the area as constant. Shaping
parameter samples have been generated by limiting major axis to maximum length
and minor axis to minimum length using MATLAB code. Based on each curva-
ture parameter, three geometries were opted and CAD models are generated from
sample space. For these geometries, CFD and CEM analysis has been performed and
corresponding pressure recovery and RCS at 10 GHz (X-band) has been estimated.
The CFD-CEM performance analysis has been presented for the optimized intake
duct geometry. Particle swarm optimization (PSO) in-conjunction with CFD solver
and CEM developed indigenous RCS solver has been used for optimization of the
designed duct towards stealth characteristics.
Keywords CFD-CEM optimization · Radar stealth · Intake duct · RCS reduction
1 Introduction
Stealth technology is one of the most important requirements for the next-generation
combat aircraft. Most of the stealth aircraft design focuses on low radar signature to
P. S. Shibu · Sandeep · B. Choudhury (B) · R. U. Nair

Centre for Electromagnetics, CSIR—NAL, Bangalore 560017, India
P. S. Shibu

446 P. S. Shibu et al.
reduce the detection possibility. Stealth technology allows the aircraft to enter into
the enemy territory without being detected by enemy radars. Across the world, only
few successful aircrafts are flying which are stealthier in nature. The stealth aircraft
design with low radar signature depends on the nature of mission. The fighter aircraft
which take part in air to air battle is designed for reduced front and back radar cross
section, whereas in bombers design, even side and bottom RCS reductions are also
considered. Cavities such as intake, exhaust, landing gear gates, air breath sensors,
etc., are the main RCS hotspots. The best way to reduce RCS is by removing or
hiding the RCS contributing portions from the enemy radar vicinity. Some of the
components which cannot be avoided have to be redesigned for low radar signature
characteristic.
Design of stealth aircraft involves the disciplines like computational fluid
dynamics, structural and computational electromagnetics. An integrated approach
has to be considered for the stealth design process. The shape feature of the aircraft
is most important in the improvement of the stealth characteristics to avoid detection
from radars. There are two kinds of radar systems, one is ground based radar detec-
tion system and other is air to air radar detection systems. Both radar system works on
same principle, i.e., transmitting the signal and receiving the reflected waves. Based
on the incoming wave characteristics, the RCS will be calculated for the object. RCS
contributions from the cavities like intake, exhaust, landing gear gates, air breath
sensors, etc., contribute more due to accumulation of electromagnetic energy. In
order to avoid this high contribution, these cavities need to be designed to achieve
the requirement of low radar signature. Air intake duct is considered for the anal-
ysis of the aerodynamic and electromagnetics characteristics. Intakes are designed for
suction of air from the atmosphere to fulfill the requirement of engine. Reduced pres-
sure loss is desirable across the duct while retaining the distortion levels to adequate
for the best performance of the engine and to avoid instabilities. As aerospace plat-
forms are always multidisciplinary in nature, the design optimization also should
follow an integrated approach to achieve the goal. The optimization problem of an
inlet shape of duct for minimum pressure loss and low radar signature with maximum
mass flow rate has been discussed in this paper.
2 Objective
The design of advanced fighter aircraft intake ducts leans to compact nature due to
its stealth factor. This leads to the unsteady in aerodynamic performance. As a result,
there should be a need for a compromise between aerodynamic attributes and radar
signature. Understanding of flow physics and electromagnetic scattering mechanism
of duct to predict the unsteady aerodynamics and assess of radar signature has a great
significance for stealth design. The present work attempts to use Computational Fluid
Dynamics—Computational Electromagnetics (CFD-CEM) integrated framework to
optimize the inlet shape of duct to ensure the maximum mass flow rate with minimum
pressure loss and radar signature. This would have a positive influence on the ability
Multi-objective Optimization Approach for Low RCS Aerodynamic … 447
to improve association between low observability and aerodynamic effect on the

designed structures. These two domains have been analyzed integrally to design a
stealthier intake duct.
3 Approach and Literature
To design optimal intake duct with stealth characteristics, needs an integrated research
of an aerodynamic and electromagnetic characteristic of aircraft intake duct. Initially
to understand the flow physics RAE M2129 S-shaped diffuser model of Royal
Aircraft Establishment has been identified for the computational fluid dynamic simu-
lation and compared with experimental results that are available for assessment [2].
The geometry of M2129 S-shaped diffuser is shown in Fig. 1.
The dimension of the duct are: Dthroat (0.1288 m) is the throat diameter, Dc
(0.1440 m) is cowl diameter, DEF (0.1524 m) is engine face diameter and L (0.4572 m)
is the duct length.
Intake duct scattering characteristic depends on the shape of the duct longitudi-
nally as well as inlet opening and it should meet the aerodynamic performance. Longi-
tudinal shape variation of air intake duct has strong impact on the radar cross-section
and aerodynamic performance. A straight cylindrical-shaped duct (Fig. 2a) has been
designed based on the design criteria values that mention in Fig. 1. Straight cylin-
drical duct design has mean RCS contribution of −8.7629 dBsm with the 97.6189%
pressure recovery. Inside RCS contribution without cowl was estimated based on in-
house MATLAB code. Aerodynamically it meets good performance but with high
radar signature.
Intake duct shape has altered to single curved nature through longitudinally as
shown in Fig. 2b which leads to mean RCS contribution of −25.0767 dBsm and
96.79% of pressure recovery. Radar signature has lowered while pressure recovery
has slightly reduced. The duct shape has further changed to double curve nature
through longitudinally as shown in Fig. 2c. The mean RCS of −42.1962 dBsm
Fig. 1 RAE M2129 diffuser

geometry
Fig. 2 Intake ducts a Straight. b Single curved. c Double curved. d Co-centric single curved
and 92.4044% pressure recovery is achieved. Radar signature is reduced consid-

erably but the aerodynamic performance has drastically degraded which leads to
aerodynamically instable structure.
Further, the design criteria have been shifted by retaining the single curve nature of
the duct longitudinally and altering the inlet shape in which the open end cavity space
should not visible to the engine face. The design is shown in the following Fig. 2d.
Mean RCS of −30.0253 dBsm and 96.2873% of pressure recovery was observed.
Co-centric single curved duct radar signature was further lowered when compared
to the single curved duct and the pressure recovery is approximately similar. It can
be observed that the inlet shape change has an impact on the radar signature. Hence,
the problem of optimization of an inlet shape of duct for minimum pressure loss and
radar cross-section with maximum mass flow rate is of great importance.
The super ellipse equation is considered to design inlet shape of air intake duct.
Super ellipse is a closed curve analogous to the ellipse [1]. The generalized two-
dimensional cartesian form of super ellipse is represented as
x n y m

+ =1 (1)
a b
where ‘a’ and ‘b’ are the semi diameters of the curve, while ‘n’ and ‘m’ are curvature
factor. Different combination of ‘n’ and ‘m’ is capable of generating number of
diverse shapes having range of symmetries. Variation of such shapes are shown in
Fig. 3.
Curvature factor m = n is assigned to retain the symmetry nature. To design the
inlet shape of air intake duct the following equation is considered. Variation of such
shapes are shown in Fig. 4.
x n y n

+ =1 (2)
a b
Fig. 3 Super ellipse shape variations with a = 2, b = 1, n = 1–5 with step size of 1, m = 0.5–2
wit step size of 0.5
Fig. 4 Super ellipse shape variations with a = 2, b = 1, n = 1–5 with step size of 0.5
To design optimal intake duct with stealth characteristics needs an integrated

research of aerodynamic and electromagnetic characteristic of aircraft intake duct.
The conceptual design and analysis approach of cavity structure such as intake duct
of aircraft is shown in Fig. 5.
Fig. 5 Intake duct conceptual design using CFD-CEM integrated approach
4 Intake Duct Design
4.1 Design (Super Ellipse)
Based on Eq. (2), three variable parameters alter the shape in two dimension. Curva-
ture factor ‘n’ influences the shape of the curve, while semi major axis ‘a’ and semi
minor axis ‘b’ has an impact on the dimension of the shape. An algorithm was devel-
oped in MATLAB to determine the ‘a’ and ‘b’ values based on the constrained area of
13029.31575 mm2 . The variables ‘a’, ‘b’ and ‘n’ has certain dimension limitations.
The range of ‘a’ confined between 50 and 300 mm, while ‘b’ bounded to 30–300 mm.
The range of ‘n’ varies from 1 to 5 with step size of 0.5.
Based on these limits, 154 samples are generated. For each ‘n’ we opted three
combinations based on the limiting ‘a’ and ‘b’ ranges into three sections. The limita-
tion of each section of ‘a’ and ‘b’ depends on the ‘n’ value. Each section is termed as
set 1, set 2 and set 3. For a particular ‘n’ three combinations of ‘a’ and ‘b’ values are
opted. Twenty-seven combinations are chosen to study integrated analysis of aerody-
namic and electromagnetic characteristics of the inlet shape of duct. Designed ducts
with respective ‘n’, ‘a’ and ‘b’ are shown below (Table 1).
Table 1 Inlet shape configurations

Case Set 1 Set 2 Set 3
1
n = 1, a = 81.13 mm n = 1, a = 130.94 mm n = 1, a = 199.83 mm

b = 80.299 mm b = 49.753 mm b = 32.601 mm
2
n = 1.5, a = 71.13 mm n = 1.5, a = 116.14 mm n = 1.5, a = 142.26 mm

b = 66.905 mm b = 40.976 mm b = 33.4525 mm
3
n = 2, a = 70.00 mm n = 2, a = 105.80 mm n = 2, a = 133.65 mm

b = 59.248 mm b = 39.20 mm b = 31.0315 mm
4
n = 2.5, a = 66.18 mm n = 2.5, a = 91.33 mm n = 2.5, a = 120.56 mm

b = 58.2315 mm b = 42.196 mm b = 31.9655 mm
5
n = 3, a = 62.05 mm n = 3, a = 90.61 mm n = 3, a = 117.99 mm

b = 59.4295 mm b = 40.6975 mm b = 31.2535 mm
6
(continued)
Table 1 (continued)
Case Set 1 Set 2 Set 3
n = 3.5, a = 69.30 mm n = 3.5, a = 70.83 mm n = 3.5, a = 112.43 mm
b = 51.7105 mm b = 50.5935 mm b = 31.8735 mm
7
n = 4, a = 63.81 mm n = 4, a = 92.36 mm n = 4, a = 110.13 mm

b = 55.065 mm b = 38.0435 mm b = 31.905 mm
8
n = 4.5, a = 73.72 mm n = 4.5, a = 88.65 mm n = 4.5, a = 106.34 mm b =

b = 46.994 mm b = 39.0795 mm 32.5785 mm
9
n = 5, a = 61.29 mm n = 5, a = 89.35 mm n = 5, a = 95.81 mm

b = 55.9345 mm b = 38.3685 mm b = 35.7815 mm
4.2 Computational Fluid Dynamics
The aerodynamic design constraints have been calculated based on the Indian Refer-
ence Atmospheric (IRA = ISA_SL + 15) conditions. The duct has been designed
for subsonic flow condition with 0.8 Mach number. Accordingly, the inlet velocity
is set to 272 m/s with mass flow rate of 3 kg/s. The throat dimension is calculated
to allow maximum mass flow rate through duct. Accordingly, 27 designs have been
modeled using available CAD software. The aerodynamic flow analysis has been
performed using steady-state flow equations solved in 3D field. In solver the flow
field parameters have been computed based on Reynolds Averaged Navier Stokes
(RANS) equations, Ideal gas flow equation along with SST K-omega turbulence
model has been used. The results such as mass flow rate, pressure recovery, and total
pressure loss coefficient have been estimated at the engine face.
4.3 Computational Electromagnetics
Monostatic RCS analysis is carried out at 10 GHz (X-band). High-frequency ray

tracing Geometric Optic (GO) method is used to estimate RCS in FEKO software.
The scattering energy is observed at azimuth = 0 with an elevation variation of −10
to 10.
5 Results
The variation of curvature factor with pressure recovery values of the inlet designs
are shown in Fig. 6. It depicts that pressure recovery for all the design are greater
than 97% which makes the structure aerodynamically feasible. Similarly variation
of semi-major axis (a) and semi-minor axis (b) with pressure recovery values of the
inlet design are represented in (Figs. 7 and 8).
From Fig.6 it portrays that set 1 contribution has good pressure recovery when
compared to set 2 and set 3. Figure 9 shows the mean RCS contribution of the ducts
over the curvature factor. It depicts that the set 3 has low radar signature contribu-
tion when compared to the set 1 and set 2. Similarly, the mean RCS distribution
over semi-major and semi-minor axis is shown in Figs. 10 and 11. To design a low
radar signature duct, shaping parameter samples has to be opted by selecting the
major axis to maximum length and minor axis to minimum length. A nature based
optimization technique namely particle swarm optimization technique has been used
in conjunction with CFD and CEM solvers. Particle swarm optimization is based on
Fig. 6 Pressure recovery variations over curvature factor (n) of all duct inlet models
Fig. 7 Pressure recovery variation over semi-major axis (a) of all duct inlet models
Fig. 8 Pressure recovery variation over semi-minor axis (b) of all duct inlet models
swarm intelligence for collection of high density food in the field and has been imple-
mented here as a multiobjective optimization technique to select the best design with
maximum mass flow and minimum RCS [6]. The optimized design (a: 95.81 mm,
b: 32.57 mm and n: 4.5) that has been obtained using PSO provides a mean RCS of
-3.1431 dBsm and a pressure recovery of 98%.
Fig. 9 Mean RCS (elevation = −10 to 10, Azimuth = 0) variation over curvature factor (n) of all
duct inlet models at 10 GHz Horizontal polarization
Fig. 10 Mean RCS (elevation = −10 to 10, Azimuth = 0) variation over semi-major axis (a) of
all duct inlet models at 10 GHz Horizontal polarization
6 Conclusion
This paper provides an insight to CFD-CEM integration approach in conjunction

with nature base optimization techniques towards design of a low RCS air intake
duct for aerospace applications. Inlet shape optimization of an air intake duct for
Fig. 11 Mean RCS (elevation = −10 to 10, Azimuth = 0) variation over semi-minor axis (b) of
all duct inlet models at 10 GHz Horizontal polarization
maximum mass flow rate with minimum pressure recovery loss and radar cross
section has been carried out with simulation of adequate models based on CFD-CEM
integration approach in conjunction with multi-objective algorithm. Twenty-seven
intake ducts have been evaluated for flow analysis and electromagnetic scattering
mechanism. Based on these performance values multi-objective design optimization
has been employed to design an optimized inlet of an intake duct which provides a
mean RCS of −3.14 dBsm with 98% pressure recovery. The results suggest that the
multi-objective design optimization can help for the design of the intake duct inlet
in CFD-CEM integrated approach with best compromise performance.
References
1. Gridgeman NT (1970) Lamé Ovals. Math Gaz 54:31–37

2. Berens TM, Delot A, Chevalier M, Muijden JV, Waaijer RA, Tattersall P (2012) GARTEUR
AD/AG-43 application of CFD high offset intake diffusers, garteur final report, ARTEUR TP-
173, pp 18–24, Oct 2012
3. Kumar I (2015) 6th generation stealth aircraft. Int J Res Sci Eng 1(2):7–12
4. Richardson D (2001) Stealth warplanes: deception. Evasion and Concealment in the Air. MBI
Publishing Company, New York
5. Raymer DP (1989) Aircraft design: A conceptual approach, 5th edn. AIAA Publishers,
Washington
6. Balamati C, Jha RM (2015) Soft computing in electromagnetics: methods and applications.
Cambridge University Press, Cambridge, CB2 8BS UK, USA, (ISBN: 9781107122482), p 215
Analysis of Propeller by Panel Method
for Transport Aircraft
Premalatha, K. R. Srilatha, and Vidyadhar Y. Mudkavi
Abstract The new aviation policy in place has given much impetus to local con-
nectivity that has been re-emphasized with the launch of UDAN program. Certainly,
this program will propel introduction of a significant number of propeller-driven
aircraft suitable for short hauls. It is also well known that propeller-driven aircraft
can be far more fuel efficient. In this context, CSIR-NAL initiated development of
14-seat Saras, a twin engine propeller-driven aircraft in pusher configuration. Being
essentially an ab-initio design, one needs to understand complex aerodynamics that
results from a pusher configuration. There are a number of technical issues such as
propeller-fuselage interaction, propeller induced noise, power-on drag that need to
be understood for design inputs. It is also known that full blown CFD methods for
such analysis are still not very mature. Even if they are, they consume enormous
computing power and clock-time to provide meaningful design inputs. In this paper,
we present application of NAL’s unsteady Panel Code (Unsteady panel method anal-
ysis. PD-CTFD/2016/1009, CSIR National Aerospace Laboratories, Bangalore [1])
for the analysis of Hartzel propeller for a combination blade setting and advance
ratios. This is an initial step towards more complex analysis wherein fuselage and
other components can also be added in a much simpler fashion. Results indicate that
there is a good confidence in this approach and as such one can generate significant
design data at initial design stage.
Keywords Panel method · Propeller · Transport aircraft
Premalatha (B) · K. R. Srilatha

CTFD Division, CSIR NAL, Bangalore, India
K. R. Srilatha
V. Y. Mudkavi
CSIR Fourth Paradigm Institute, Bangalore, India

458 Premalatha et al.
1 Introduction
With UDAN program set to boost regional transport, a need will soon be felt to
introduce turbo-prop aircraft as they are far more efficient when compared to jet
equivalents. Towards this, CSIR National Aerospace Laboratories (CSIR-NAL) had
initiated development of turbo-prop regional transport aircraft. Much earlier, NAL
also began development of 14-seat Saras in pusher prop configuration keeping North-
East segment in mind. Such developments require quick and robust propeller flow
analysis methods. It is well known that full blown CFD methods require high end
computing systems and long turn around time to produce reasonable solutions needed
for initial design studies. It is interesting to note that panel methods continue to play
a useful role in analysis of propeller flow fields, in spite of their complexity, requiring
very little compute power and extremely short turn around times to generate design
data.
In this paper, we present validation of in-house developed unsteady panel code for
propeller flows. We consider five-bladed propeller used for powering NAL developed
Saras. We analyze the flow for free propeller at various flow conditions defined by
blade setting angles (β). Flow quantities such as thrust and torque coefficients are
compared with experimental results.
For inviscid, incompressible and irrotational flow past an arbitrary three-
dimensional body, we have
∇ 2 (r, t) = 0 (1)
where is the velocity potential, r is the position vector, and t is the time. We can
decompose the velocity potential as
(r, t) = ∞ + φ(r, t) (2)
where ∞ is the velocity potential in the absence of the body and φ(r, t) is the
perturbation potential due to the presence of the body. Usually, this perturbation is
regarded as small. Assuming that the undisturbed flow U∞ = ∇∞ is uniform and
steady, Eq. (1) reduces to
∇ 2 φ(r, t) = 0 (3)
Since the flow disturbances are assumed to decay at infinity and that body is
impermiable, the boundary conditions become
v = ∇φ → 0 as r → ∞ (4)
(U∞ + v) · n = 0 (5)
where n is the unit normal on the boundary surface.

Analysis of Propeller by Panel Method for Transport Aircraft 459
From Green’s theorem [2], it follows that any solution of Laplace equation can
be expressed as an integral given by

σ (S, t) ∂ 1
φ P (r, t) = + μ(S, t) dS
r (S, P) ∂n r (S, P)
body
(6)
∂ 1
+ μ(S, t) dS
∂n r (P, q)
wake
where σ and μ are the strengths of the source and doublet singularities, respectively.
r (S, P) is the distance from the integration point to the field point P where the
potential is being evaluated. n is the normal along the surface S. The source singularity
is used to take into account the thickness effects, and the doublet singularity is
used to account for the lifting effects. As the velocity induced by constant doublet
distribution has the same magnitude as the vortex line of same strength situated
around the perimeter of the panel, the doublet distribution on quadrilateral panels is
replaced by vortex rings around the same panel. This enables one to use Biot–Savart
law for the calculation of velocities at any point due to quadrilateral vortex ring. This
method is well-established [3].
1.1 Numerical Procedure
The numerical procedure consists of dividing the lifting body such as wing and its
mean surface into a number of quadrilateral planar surfaces. Only constant sources
whose strengths are unknown are distributed on each of the surface panels. The
camber surface is also panelled similar to surface panelling. The vortex panels are
placed on camber surface from leading to trailing edge. The chordwise panels on
each column are grouped together, and only constant doublets proportional to section
thickness are distributed. The boundary condition is set on Kutta panel downstream
of trailing edge. Imposing the condition Eq. (6) in Eq. (5) results in a system of
algebraic equation of the form,
Ax = b (7)
where A is known as the influence coefficient matrix, x is the vector of unknowns

consisting of source and doublet strengths, and b is the known vector that accounts
for boundary conditions. The solution of this system gives the strengths of σ and μ
at each instant of time.
If the flow is steady, then A is constant and the solution is simple [4, 5]. However,
if the flow is unsteady, as in the present case, then we proceed as follows.
At time t = 0, there is no wake when the body is started impulsively from rest.
With the solution of Eq. (7), the strengths of the singularities are determined. Next the
induced velocity at the downstream corner point of Kutta panels for lifting surfaces
is evaluated, and a straight vortex element is released whose length is the distance
traversed by local velocity in that time interval. This row of vortex filaments forms
the first increment of wake. The vortex strength of this row remains same for all
subsequent calculations. For the next time step, Eq. (7) is solved taking into account
all the body bound singularities and the induction due to first row of wake panels.
Repeating the procedure, this results in the second wake row. The process continues
for subsequent time steps yielding unsteady wake as it develops.
Figure 1 shows the model of the blade surface and the development of the wake
on it.
In order to calculate other parameters such as thrust and torque, we need pressure
coefficient Cp which is calculated using the following unsteady Bernoulli’s equation
2
|V | 2 ∂φ
Cp = 1.0 − − 2 ∂t
(8)
VRef VRef
where V is the total velocity and VRef is any reference velocity.

The time derivative of potential in Eq. (8) is evaluated using
∂φ ϕ(t + t) − ϕ(t)

= (9)
∂t t
Fig. 1 Panelling strategy adopted for lifting surfaces. Doublets are distributed on the mean surface,
accounting for lift, while sources are distributed on the actual surface, accounting for thickness.
Panels are generated by taking sections span-wise and chordwise. A Kutta panel is inserted which
is an extension of the mean surface. Wake panels are introduced beyond the Kutta panel
Once the pressure coefficient acting at all panels is calculated, the forces and
moments acting on individual panels and the configuration as a whole are easily
obtained at every time instant [1].
In what follows, we present an application of this technique to the five-bladed
Saras propeller for various flow conditions.
Figure 2 shows the propeller configuration and their panel discretization. The pro-
peller was first analysed for five blade angle settings of β = 20◦ , 22.5◦ , 25◦ , 27.5◦
and 30◦ . These were compared with the values provided by the design team. Sub-
sequently, results were generated for four more blades angles β = 19◦ , 23◦ , 27◦
and 31◦ .
We primarily are interested in the thrust coefficient CT defined by
T T
CT = = (10)
1
2
ρU∞
2 S ρn 2 D 4
where T is the magnitude of the thrust calculated as the axial component of the
aerodynamic force, U∞ is the reference velocity which is the undisturbed flow at
infinity, S is the propeller swept area, n is the revolutions per second (rps) of the
propeller, ρ is the density, and D is the diameter of the propeller.
The torque coefficient CQ is defined by
Q
CQ = (11)
ρn 2 D 5
where Q is the magnitude of the torque.

Figure 3 shows the comparison of thrust coefficient as a function of advance ratio
with NAL experiments for different blade setting angles. For all cases, it is clear
that the trend is extremely well-captured except for a near constant shift. This may
be attributed to the fact that the blade setting angle measured on actual propeller
blade, and the simulations may have some differences. It is also not clear if the blade
setting angle is taken at 70% of the propeller centerline or from the edge of the hub.
However, for design purposes, these results can be used with confidence if the blade
setting angle correction is added. A simple exercise shows that the difference in the
results, i.e., a constant shift, can be accounted for by adding 5◦ to the blade setting
angle. This is shown as a composite plot in Fig. 4. Here Rtip refers to propeller blade
radius.
Even though the method is simple in nature, the wake can get fairly complex as
time evolves. Since the wake is significant and has major effect on moments and
forces, it is important to assess the convergence of these quantities. Figures 5 and
6 show evolution of thrust coefficient with time for different blade setting angles
X
Y
Fig. 2 Five-bladed Saras propeller showing the panelling. The blade geometry was obtained by
physically scanning the propeller blade. For the purpose of these calculations, data was cleaned
before panelling
for axial velocity of 70 m/s and 100 m/s, respectively. In about 1 time unit (non-
dimensional), the propeller would have executed about half a revolution. It is clear
that thrust coefficient converges smoothly rapidly in about one propeller revolution
that corresponds to about 2 time units. The convergence is a little slower for smaller
blade setting angles. The blade setting angles examined were provided by the design
team.
Similarly, Fig. 7 shows the convergence history of thrust coefficient for blade
setting angle of β = 31◦ for various axial velocity values. It is again clear that con-
vergence is smooth and rapid.
Next we examine behaviour of torque coefficient. Figure 8 shows the variation
with respect to advance ratio for various blade setting angles. Figure 9 shows the
variation with axial velocity. The trends are right. Also, it is notable that variations
are smooth. However, it should be noted that the results may not be reliable for lower
advance ratio as the method does not account for any separation which is likely to
occur for lower values.
As mentioned earlier, wake is developed as part of the solution that may get fairly
complex. For this purpose, we examine the physical structure of the wake for a
qualitative assessment. Figures 10 and 11 show the wake structure for axial velocity
of 40 m/s and 90 m/s, respectively. The structure for axial velocity of 40 m/s is
Blade Angle = 20o Blade Angle = 22.5

o
0.4 0.4
Panel Panel
Expt. Expt.
0.3 0.3
CT
CT
0.2 0.2
0.1 0.1
0 0
0.4 0.6 0.8 1 1.2 1.4 1.6 0.4 0.6 0.8 1 1.2 1.4 1.6
Advance Ratio Advance Ratio
o o
Blade Angle = 25 Blade Angle = 27.5
0.4 0.4
Panel Panel
Expt. Expt.
0.3 0.3
T
CT
0.2 0.2
C
0.1 0.1
0 0
0.4 0.6 0.8 1 1.2 1.4 1.6 0.4 0.6 0.8 1 1.2 1.4 1.6
Advance Ratio Advance Ratio
Blade Angle = 30 o
0.4 Experiment Data from Report:
Panel "Power, Fuel Flow and Thrust data
Expt. Report for Saras",
0.3 SARAS/ AERO/213, May 2013
CT
0.2
Panel Computation:
RPM = 1700, Rtip = 1.32 m
0.1
0
0.4 0.6 0.8 1 1.2 1.4 1.6
Advance Ratio
Fig. 3 Comparison of thrust coefficient as a function of advance ratio with experiments for various
blade setting angles
0.4
RPM = 1700, Rtip = 1.32 m
o
Experiment 20
Panel 25o
Experiment 22.5 o
0.3
Panel 27.5o
Experiment 25 o
Panel 30o
CT
0.2
0.1
0
0.4 0.6 0.8 1 1.2 1.4 1.6
AdvanceRatio
Fig. 4 Comparison of thrust coefficient as a function of advanced ratio with NAL experiments after
adjusting for blade setting angle by 5◦
Axial Velocity = 70 m/s, RPM = 1700, Rtip = 1.32 m

0.3
0.2
0.1
CT
-0.1 Blade Angle = 19

o
Blade Angle = 23o

Blade Angle = 27o
Blade Angle = 31o
-0.2
-0.3
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75
Time
Fig. 5 Comparison of thrust coefficient as a function of time for various blade setting angles and
axial velocity = 70 m/s

0.3
o
Blade Angle = 19
o
0.2 Blade Angle = 23
o
Blade Angle = 27
o
Blade Angle = 31
0.1
CT
-0.1
-0.2
-0.3
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75
Time
Fig. 6 Comparison of thrust coefficient as a function of time for various blade angles and axial
velocity = 100 m/s
Blade Angle = 31o, RPM = 1700, Rtip = 1.32 m

0.3
0.2
0.1
CT
0
Vaxial = 40 m/s
Vaxial = 50 m/s
Vaxial = 60 m/s
-0.1 Vaxial = 70 m/s
Vaxial = 80 m/s
Vaxial = 90 m/s
Vaxial = 100 m/s
-0.2
Vaxial = 110 m/s
-0.3
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75
Time
Fig. 7 Comparison of thrust coefficient as a function of time for blade setting angle of 31◦ for
various axial velocities
0.04
Blade Angle
o
19
23 o
0.02 27 o
31 o
0
CQ
-0.02
-0.04
RPM = 1700, Rtip = 1.32 m

-0.06
0.4 0.6 0.8 1 1.2 1.4
Advance Ratio
Fig. 8 Comparison of Torque coefficient as a function of advanced ratio for various blade setting
angles
0.04
Blade Angle
19o
23 o
0.02 27 o
31 o
0
CQ
-0.02
-0.04
RPM = 1700, Rtip = 1.32 m

-0.06
40 50 60 70 80 90 100 110
Axial Velocity
Fig. 9 Comparison of Torque coefficient as a function of axial velocity for various blade setting
angles
o
Blade Angle = 19 , Axial Velocity = 40m/s, RPM = 1700, Rtip = 1.32
Y
X
Fig. 10 Wake plot for blade angle 19◦ , 1700 rpm and axial velocity = 40 m/s
o
Blade Angle = 19 , Axial Velocity = 90 m/s, RPM = 1700, Rtip = 1.32
Z
Y
X
Fig. 11 Wake plot for blade angle 19◦ , 1700 rpm, axial velocity = 90 m/s
o o
Blade Angle = 19 Blade Angle = 23
468
Z Z
Y Y
X X
Blade Angle = 27 o Blade Angle = 31 o

Z Z
Y Y
X X

Premalatha et al.
Fig. 12 Wake plot depicting comparison for various blade angles, 1700 rpm and
axial velocity = 40 m/s
shown for about two revolutions while that for 90 m/s is for about one revolution.
The helical structure is well-maintained and is expected when the wake is nearby
to the propeller. Wake distortions are expected to happen far downstream. It is also
clear that wake expands after about one revolution. In Fig. 12, we examine wake
structure for different blade setting angles. While the structure is in general smooth
and well-organized, a close examination shows that for higher blade angle the wake
distortion is initiated. However, this is not of any consequence since relevant design
quantities are already converged.
3 Conclusions
A modified panel method to account for unsteady flows is applied for Saras five-
bladed propeller. Convergence of thrust and torque coefficients, relevant for initial
design, is examined. Results converge smoothly and rapidly. Wake structure is also
examined, and it shows that the picture is qualitatively meaningful. Wake distortions
are expected to happen much later. However, this is not of consequence as the results
are already converged within about one propeller revolution. The results are compared
with NAL experiments and show that but for a near constant shift, the match is
excellent. The shift in computation and experimental result may be attributed to
the fact that the blade setting angle measured on actual propeller blade and the
simulations may be different. However on adjusting the blade setting angle by about
5◦ , the method yields results that can be used for all practical design calculations.
References
1. Srilatha KR, Premalatha, Ahmed SR, Narayana CL, Mudkavi VY (2016) Unsteady panel method
analysis. PD-CTFD/2016/1009, CSIR National Aerospace Laboratories, Bangalore
2. Lamb H (1932) Hydrodynamics. Dover Publications, New York
3. Ahmed SR, Vidjaja VT (1994) Numerical simulation of subsonic unsteady flow around wings
and rotors. AIAA-1994-1943, A94-30939 10-02 938-951
4. Narayana CL, Srilatha KR (2004) Aerodynamic analysis of SARAS propeller by a panel method,
NAL PD CF 0403. CSIR National Aerospace Laboratories, Bangalore
5. Srilatha KR, Narayana CL, Premalatha, Mudkavi V (2007) Performance analysis of AFT
mounted propeller for a light transport aircraft. In: Proceedings, 7th Asian CFD conference,
pp 107-115
Effect of Reynolds Number on Typical
Civil Transport Aircraft
Vishal S. Shirbhate, K. Siva Kumar, and K. Madhu Babu
Abstract Reynolds Averaged Navier–Stokes (RANS) simulations have been

performed over a civil transport aircraft at Mach number 0.17. The present work aims
to understand the effect of various Reynolds number on the aerodynamic perfor-
mance of aircraft since the Reynolds number is considered as a most important
parameter in fluid dynamics. The current study also helps to identify the various
hot spots for higher drag so that the design team can focus on these specific areas
to minimise the total drag. The present CFD solver uses a Roe scheme for convec-
tive flux discretisation and Spallart-Allmaras turbulence model for eddy viscosity
computations. Calculations showed that with an increase in Reynolds number, the
maximum lift coefficient increases and minimum drag coefficient decreases. The
paper also highlights the sectional Cp plots and pressure contours along with the
velocity streamlines at different span-wise stations of an aircraft wing. The strength
of vortex over a midboard flap gradually decreases with the increase in Reynolds
number. The drag of various aircraft components is also shown in the form of Pie
chart so that the major drag contributing parts can be identified.
Keywords RANS · Wind-tunnel · Reynolds number · Civil transport aircraft ·

Vortex
V. S. Shirbhate (B) · K. Siva Kumar · K. Madhu Babu

Computational and Theoretical Fluid Dynamics Division, Council of Scientific and Industrial
Research—National Aerospace Laboratories, Bangalore, Karnataka 560017, India
K. Siva Kumar
K. Madhu Babu

472 V. S. Shirbhate et al.
1 Introduction
Today, the regional civil transport aircraft market has a huge potential in India. The
government is also very keen on the regional connectivity program for tier-II and
tier-III cities since India’s civil aviation industry is on a high-growth trajectory [1].
India aims to become the third largest aviation market by 2020 and largest by 2030
as per the report posted by India Brand Foundation Equity website. According to the
latest market analysis, turbo-propeller aircraft is going to share the 24% of market
share in an Asia–Pacific region. Towards this, smaller passenger aircraft segment of
around 14–40 seater aircraft will play a vital role to fulfil the market demand.
In the view of the above context, the current paper focuses on the computational
aerodynamic studies performed on a typical civil transport aircraft. The Reynolds
number plays a vital role in determining the aerodynamic coefficients and type of
fluid flow (laminar or turbulent). The effect of various Reynolds number over aerofoil
has been studied experimentally [2]. The authors observed from the experiment that
the minimum drag decreases and the maximum lift increase on the aerofoil with
the increase in Reynolds numbers. The extensive work has also been performed
to determine the effect of Reynolds number for applications like smaller MAVs and
UAVs [3–6]. However, such investigations are not much helpful at civil aircraft level.
Therefore, an attempt has been made to perform the RANS simulations over a typical
civil transport aircraft for different Reynolds numbers 1.0 million, 5.0 million and
10.0 million at a freestream Mach number of 0.17. The stall behaviour for all the
three Reynolds number has been studied and the flow characteristics at corresponding
region are thoroughly assessed and reported.
2 Methodology
2.1 Grid Generation
The unstructured grid is recommended for complex geometries like full aircraft
configuration integrated with engine and nacelle. An unstructured tetrahedral grid
has been generated using Pointwise software for present aircraft configuration as
shown in Fig. 1. The present grid consists of 2 million elements on the surface and
68 million volume elements. A cylindrical domain is used for a far-field boundary
which is placed at 10 body-lengths from the nose and tip of the wing and the 20 body
lengths behind the aircraft. The prismatic layer is created using a special technique
called T-Rex (anisotropic tetrahedral extrusion). The first grid cell spacing of the
mesh is 1.5e–05 and the y + considered for the present computations is less than 1.0
on the lifting surfaces. The drag prediction workshop guidelines have been followed
for building a mesh.
Effect of Reynolds Number on Typical Civil Transport Aircraft 473
Fig. 1 Unstructured grid generated over aircraft
2.2 Solver Details
Three-dimensional steady RANS computations [7, 8, 10] have been performed using
a commercial CFD package ANSYS Fluent. This solver is the most powerful and
flexible general-purpose tool used to model flow, turbulence, heat transfer and reac-
tions for many aerospace applications. All present CFD simulation uses Roe [9]
scheme for convective flux discretisation and Spallart-Allmaras turbulence model
for eddy viscosity computations. A special initialisation technique called Full Multi-
Grid (FMG) initialisation has been used for better initial guess values and accel-
erating the convergence. FMG solves the flow problem on a sequence of coarser
meshes using Euler equations before transferring the solution onto the actual mesh
[11]. FMG initialisation is useful for complex flow problems involving large pressure
gradients on finer meshes. The computations have been performed on ICE cluster
super-computing facility of CSIR-4PI.
The lift coefficient comparison between CFD computations for three different
Reynolds numbers; 1.0 million, 5.0 million and 10.0 million and wind tunnel experi-
ment are shown in Fig. 2. The wind tunnel data shows a slow and gradual stall pattern
compared to sudden stall pattern predicted by CFD results. The computations have
Fig. 2 CL versus AoA comparison between CFD and WT
predicted a similar stall behaviour for all the three Reynolds number whereas exper-
iment data slightly varies in the same region. It is observed from the CFD results that
the maximum lift coefficient (CL) increases with increasing Reynolds number. The
lift coefficient at 0° angle of attack remains nearly equal for all the Reynolds number.
The best comparison can be observed for Re = 1.0 million which also corresponds to
the wind tunnel data. The drag coefficient versus angle of attack comparison between
computations and wind tunnel data is plotted in Fig. 3. The numerical results have
predicted a lower drag than the wind tunnel data. It is also seen that the drag coeffi-
cient decreases with the increase in Reynolds number. Figure 4 shows the lift versus
Fig. 3 CD versus AoA comparison between CFD and WT

Fig. 4 CL versus CD comparison between CFD and WT
drag plot and it is depicted that the CFD computations have predicted a lower drag as
compared to the experimental data till the stall region. The zero lift drag coefficient
(CD0 ) reduces as the Reynolds number increases. The drag polar trend for 5.0 and
10.0 million Reynolds numbers is identical in nature until the stall region. It has
also been observed that for a Re = 1.0 million, the drag prediction is slightly higher
than the other two Reynolds number cases and lower than wind tunnel data. The
lift versus pitching moment coefficient comparison is depicted in Fig. 5. It is seen
that the pitching moment slope is negative for a range of alpha sweep and the slope
remains almost constant till it reaches to stall angle. The computations under-predict
Fig. 5 CL versus Cm comparison between CFD and WT

the pitching moment at zero lift compared to experiments for all the three Reynolds
numbers.
At the stall angle, the sectional Cp plots and Cp contours along with streamlines
on different spanwise sections are shown in the Figs. 6, 7 and 8. It is understood
from the figures that a vortex forms over a trailing edge of the midboard flap and
also in the gap between main wing and inboard aileron. It can also be noticed in the
corresponding sectional Cp plots that the pressure distribution is slightly disturbed
near that region. The same vortex becomes stronger as it moves towards the midboard
aileron region. This phenomenon finally results in a wing stall.
The strength of vortex over a midboard flap gradually decreases with the increase
in Reynolds number as depicted in figures. The above flow characteristics are clearly
visible in the zoomed view of all pressure contour pictures. The Cp plots illustrate
that the peak of negative pressure coefficient is maximum at the highest Reynolds
Fig. 6 Sectional Cp plot (left), Cp contours along with streamlines (full view: middle, zoomed
view: right) at Re = 1.0 million: mid-board flap, inboard aileron, and midboard aileron
number. Since experimental data is not available for sectional Cp plots, results are
presented only for computations.
The zero lift drag coefficient (CD0 ) for various aircraft components at different
Reynolds numbers are shown in Fig. 9. It is presented in the form of Pie charts so
that the major drag generating areas can be easily identified. The Pie charts clear
that the wing drag contribution is maximum to the total drag of aircraft, followed
by fuselage and engine nacelle. The percentage of zero lift drag increases on a wing
and decreases slightly on the fuselage with the increase in Reynolds number.
(HT-Horizontal Tail, VT-Vertical Tail, N&SW-Nacelle & Stub-Wing, VF-Ventral
Fin).
4 Conclusion
The present CFD study is carried out over a civil transport aircraft for various
Reynolds number and it has been observed that the sudden stall pattern remains
similar for all the Reynolds number. The maximum lift coefficient increases and
the minimum drag coefficient decreases with the increase in Reynolds number. It
is also noted that a vortex is observed between the main wing and aileron region
which causes a massive flow separation. It is also seen that the strength of vortex
over a midboard flap gradually decreases with the increase in Reynolds number. It
can be concluded that the stall pattern and corresponding flow characteristics are
well captured by current CFD methodology and can be helpful in future design
modifications.
Fig. 9 Contributions of various aircraft components (shown in %) to drag coefficient at α = 0o
Acknowledgements The authors would like to thank Dr. J.S. Mathur, Head and Dr. V. Ramesh
Jt. Head, CTFD Division and Director, CSIR-NAL for their kind support. The authors also wish
to thank Dr. V.Y. Mudkavi, Head, CSIR-4PI for arranging reservation of nodes on CSIR-4PI ICE
HPC facility.
References
1. Flight Global Flight Fleet, Forecast Executive Summary (2017)

2. Pires O, Munduate X, Ceyhan O, Jacobs M, Snel H (2016) Analysis of high Reynolds numbers
effects on a wind turbine airfoil using 2D wind tunnel test data. J Phy. https://doi.org/10.1088/
1742-6596/753/2/022047
3. Mueller TJ, Batill SM (1982) Experimental studies of separation on a two-dimensional the
airfoil at low Reynolds numbers. AIAA J 20(4):457–463
4. Selig MS, Donovan JF, Fraser DB (1989) Airfoils at low speeds. Soartech 8, Soartech
Publications, Virginia Beach, Va, USA
5. Brown L, Filippone A (2003) Aerofoil at low speeds with Gurney flaps. Aeronautical J
107(1075):539–546
6. Traub LW, Agarwal G (2008) Aerodynamic characteristics of a gurney/jet flap at low Reynolds
numbers. J Aircraft 45(2):424–429
7. CTFD Division, CSIR-NAL (2014) Bangalore, IIT Kanpur: Aerodynamic analysis of dornier
228 aircraft with and without external antennae
8. Shirbhate VS (2012) Reynolds averaged navier-stokes analysis for civil transport air aircraft
using structured and unstructured grids. In: Proceedings, 14th annual CFD Symposium, Indian
Institute of Science, Bangalore, CP-054
9. Shirbhate VS, Sonil S (2012) Viscous computations for 4 and 5 abreast NCA configuration
using HiFUN code: results and discussions. NCAD/AE-02/0001/2012
10. Mathur JS, Dhanalakshmi K, Ramesh V, Chakrabarty SK (2008) Aerodynamic design and
analysis of SARAS aircraft. Comput Fluid Dyn J 16(3):320–334
11. ANSYS Fluent https://www.ansys.com/Products/Fluids/ANSYS-Fluent
Prediction of MultiStore Separation
from a Fighter Aircraft Using In-House
Code—WISe
S. Karthik, Jishnu Suresh, P. Karthikeya, S. Rajkumar, Mano Prakash,

Sashi Kiran, and D. Narayan
Abstract An efficient and economical in-house code WISe (Weapon Integration

and Separation) has been developed to predict the trajectory of stores (1000 lb bomb
and 250 kg bomb) released from an Indian fighter aircraft. First the stores are released
in isolated mode and validated with a commercial code CFD++ and flight test data.
After successful validation, the in-house code is extended to predict the trajectory of
1000 lb bombs when they are mounted on the mid-board and inboard stations on either
sides of the Fighter Aircraft. The bombs are released in the following sequence—
mid-board (left) followed by mid-board (right) followed by inboard (left) followed by
inboard (right). Time interval considered between subsequent releases is 100 ms. A
heavy-duty ejector release unit (ERU) has been used for inboard bombs whereas for
mid-board bombs a light duty ejector has been used. The histories of store positions,
orientation and miss distance are presented.
1 Introduction
During operational firing from a fighter aircraft, number of bombs are required to
be dropped simultaneously or with minimum time interval between two subsequent
releases. Accurately simulating the trajectory of the bombs dropped from the parent
aircraft is a challenging task, especially when they are jettisoned simultaneously or
within very short interval of time. The problem becomes more complex when their
ejection force is not same for all the stations. In the event of multi-store jettison
the store separation engineers are to define operational release envelope of the
stores by working out safe release sequence and minimum time interval between
two subsequent releases. Hence the adopted store separation techniques should be
robust in terms of complexity of configuration and range of flight conditions. Also
S. Karthik (B) · J. Suresh · P. Karthikeya · S. Rajkumar · M. Prakash · S. Kiran · D. Narayan

Aeronautical Development Agency, Bangalore, India
J. Suresh

482 S. Karthik et al.
Fig. 1 Schematic view of

bombs
1000 lb. Bomb
250 Kg Bomb
the technique should be capable of providing accurate results with low turnaround
time.
In the present study, to predict the trajectory of stores, an in-house code WISe
(Weapon Integration and Separation) has been developed. The code works based on
decay factorization sheme [2]. Trajectories of two different type of bombs—1000 lb
and 250 kg (shown in Fig. 1) released from an Indian fighter aircraft have been
predicted using WISe code and validated with a commercial code CFD++ [1] and
flight test data. The bombs are released from mid-board station in presence of a fuel
tank at inboard station in isolated mode.
After successful validation, the in-house methodology is extended to predict the
trajectory of 1000 lb bomb when they are mounted on the mid-board and inboard
stations on either sides of the Fighter Aircraft. The bombs are released in following
sequence—mid-board (left) followed by mid-board (right) followed by inboard (left)
followed by inboard (right). Time interval considered between subsequent releases
is 100 ms.
2 Methodology
Any store trajectory prediction program consists mainly of two distinct modules: (i)
an aerodynamic prediction module that computes the forces and moments acting on
the store and (ii) a 6-DOF time integration module that generates the trajectory of
a store using the forces and moments acting on the store and its inertial properties,
such as mass and moments of inertia. During the trajectory prediction process the
aerodynamic forces and moments may be generated from wind tunnel tests or through
CFD approaches.
Prediction of MultiStore Separation from a Fighter Aircraft … 483
2.1 In-House Code—WISe
The WISe code computes store aerodynamic forces and moments using two basic
sets of aerodynamic data—(i) carriage loads when the stores are attached to the
aircraft (ii) free-stream data base as a function of attitudes. The variation of store
aerodynamic load from installed to the freestream is described by a decay-function
which is assumed to be dependent on store’s vertical displacement. A linear decay
function [2] is used to compute the store aerodynamic coefficients from carriage
flight to its free-stream value when it is away from the carriage location. The decay
function is defined as follows:

Z (t)
d_fact = max 0, 1 − (abs (1)
D ∗ R1
where
D = diameter of store.
R1 = constant.
Z = vertical distance of store w.r.t. aircraft which is function of time (t).
This decay factorization system has to be tuned by matching with flight tests
results. The tuned and approved mathematical model then can be used to define the
safe separation envelope. In the present study, the required data base is generated
using a CFD approach. A Cartesian grid based Euler code—PARAS [3] is used to
generate the required installed load and free-stream data base. The viscous crossflow
terms are calculated based on empirical relation given by Jorgensen [4] and added
to the inviscid aerodynamic coefficients. The code can also compute store dynamic
derivatives from store sectional loads2 and added to the above aerodynamic coeffi-
cients. This store aerodynamic prediction module is integrated with 6-DOF module.
The store force and moment data apart from other parameters are fed into the 6-DOF
module to predict the store’s new location and orientation.
2.2 Commercial Code—CFD++
In CFD++ the flow field over the parent aircraft and stores are simulated using
unsteady RANS calculation. Flow is assumed to be turbulent, the effects of which
are computed in terms of eddy viscosity using Spalart-Allmaras model. The compu-
tational domains are discretized with unstructured hybrid grids using Ansys ICEM
CFD [5]. The grids comprise of prism cells in near wall zones and tetrahedrons in
the remaining flow domains. The first layer height is fixed to be 0.0012 mm in grid
over the bomb as well as in the aircraft grid. The bomb grid was constructed with
a smaller domain and is merged with the background aircraft grid to form a single
set after concatenation of the overlapping grids using the overset mesh technique in
CFD++. The near wall resolution in grids resulted in average first layer y+ <1. The
mesh generated for isolated bomb consists of nearly 3.1 million cells and aircraft
mesh consists of nearly 32.3 million cells. The combination of aircraft and bomb
consists of 35.5 million cells in the concatenated mesh. Figure 2 displays surface
mesh over the aircraft and 1000 lb bomb at time t = 0.1 s after release from the
aircraft. Figure 3 displays the cut-plane view of volume mesh over the 1000 lb bomb
at mid-board location at time t = 0.1 s after release from the aircraft.
2.3 Photogrammetric Method
The trajectory from the flight test is captured using an Airborne Separation Video
System (ASVS). ASVS captures the separation sequences through multiple high-
0speed digital cameras with wide angle lenses to cover the entire store body within
the near field of the aircraft. A specially designed pod was fitted on the outboard
station to carry the cameras. The cameras are controlled by Multi System Controller
(MSC). This controller upon receiving the trigger signal from the pilot for the bomb
release, initiates the high speed cameras to start recording the sequences for specified
duration. Once event is completed the captured images are downloaded automatically
to the hard disk in the MSC. The system has adequate safety features to preserve
the valuable separation data. The images captured by the high speed cameras are
analyzed in Track Eye Software [6] which provides the separation characteristic of
the stores under test.
2.4 Miss Distance
The most important factor for safe store separation is the ability to accurately predict
the store miss-distance, which is the smallest distance (function of time) between
any part of the store and any part of the aircraft (store, pylon, fuselage, etc.) during
the early stages of the trajectory. A computer code has been developed to compute
the miss distance using store time history data and aircraft and store geometries.
The parent aircraft and the store are represented by discrete points and the store is
placed at a position based on the corresponding displacements and orientations at
each time step and the miss distance is computed. The program also identifies the
corresponding nearest store and aircraft surfaces.
In the present paper, the trajectories for two different types of bombs—1000 lb
and 250 released from an Indian fighter aircraft have been predicted using WISe
code and validated with a commercial code CFD++ and flight test data when they
are released in isolated mode.
(a) Surface Mesh over Aircraft and 1000lb Bomb
(b) Zoomed View of 1000lb Bomb mesh
(c) Bottom view of mesh
Fig. 2 Surface Mesh over Aircraft and 1000 lb Bomb at t = 0.1 s after release from aircraft
(a) Cut Plane view of Volume Mesh over 1000lb bomb – Side View
(b) Cut Plane view of Volume Mesh over 1000lb bomb – Bottom View
Fig. 3 Volume Mesh over Aircraft and 1000 lb Bomb at t = 0.1 s after release from aircraft
3 Results
3.1 Single Store Release
Figure 4a shows the comparison of trajectory traces between predictions (CFD++

& WISe) and flight test of 1000 lb bomb at M = 0.9, α = 3.342 and altitude =
7 km. Figure 5 shows the comparison of translational and angular time histories of
the same results. The comparison is found to be in good agreement and separation
of the bomb is safe as indicated by a time history of miss distance plot as shown in
Fig. 4b.
(a) Front view (b) Miss distance plot
Fig. 4 Trajectory of 1000 lb bomb—single release from mid-board station at different time
instances
(a) Store X,Y,Z Locations (b) Store Pitch and Yaw Angles
Fig. 5 Trajectory validation of 1000 lb bomb—single release

(a) Front view (b) Miss distance plot
Fig. 6 Trajectory of 250 Kg bomb—single release at different time instances
(a) Store X,Y,Z Locations (b) Store Pitch and Yaw Angles
Fig. 7 Trajectory validation of 250 Kg bomb—single release
Comparison of trajectory traces between predictions and flight of 250 kg bomb

for M = 0.7, α = 3 and altitude = 2 km are shown in Fig. 6a. Figure 7 shows
the comparison of translational and angular time histories of the same results. The
comparison is again found to be encouraging and separation of the bomb is safe.
Comparison of time history of miss distance is shown in Fig. 6b.
3.2 Multi-store Release
The in-house methodology is further extended to predict the trajectory of 1000 lb

bombs when they are mounted on the mid-board and inboard stations on either sides
of the fighter aircraft. As mentioned above the bombs are released in the following
sequence—mid-board (left) followed by mid-board (right) followed by inboard (left)
Fig. 8 Trajectory of 1000 lb bomb released in Salvo mode at different time instances
followed by inboard (right) with time interval between subsequent releases being
100 ms. Inboard bombs are released with a heavy duty ejector release unit (ERU),
whereas for mid-board bombs a light duty ejector has been used. The bombs are
released at M = 0.9, α = 2 deg and altitude = 2 km. The trajectory traces for all
the bombs at three different time instances (t = 0 s, 0.6 and 1 s) is shown in Fig. 8.
Figure 9a shows the miss distance between the aircraft and bombs. Figure 9b shows
the miss distance between bomb to bomb.
4 Conclusion
An in-house code WISe has been developed to predict the trajectory of stores released
from the fighter aircraft. The code works based on Decay Factorization approach.
The predicted trajectory has been successfully validated with the trajectory computed
using a commercial code CFD++ and flight test results. The trajectory from the flight
test is captured using an Airborne Separation Video System (ASVS). The images
captured by the high speed cameras were analyzed in Track Eye software which
provides the separation characteristics of the stores. The code WISe is further used
to predict the trajectory of bombs released in salvo mode with time interval between
subsequent releases is 100 ms. The miss distance history shows the separation of
stores are safe.
(a) Aircraft-to-store miss distance (b) Store-to-store miss distance
Fig. 9 Miss distance plots of 1000 lb bombs released in Salvo mode
References
1. CFD++ Ver 15.1, Metacomp Technology Inc., USA

2. Deslandes R Strategies for modeling aerodynamic interference during store separation,
DASA/LME12/S/R/1713/1995
3. Ashok V, Kumar P, Babu TC, Devasia KJ (2000) User’s manual for PARAS 3D, Report no.
VSSC/ARD/TR-033/2000
4. Jorgensen LH Prediction of static aerodynamic characteristics for space shuttle-like and other
bodies at angles of attack from 00 to 1800 . NASA TN D-6996
5. ANSYS ICEM CFD User’s Manual ver 17.0, ANSYS, Inc., USA
6. Narayan D, Maharana D (2006) Validation of store trajectory predicted by CFD and
CTS approaches with flight test data, Report No. ADA/LCA/010601/136, April 2009. No.
ADA/TD/WT/070, Nov 2006
Shockwave Oscillations Over the Conical
Heat Shield Region of a Typical Launch
Vehicle at Mach 0.95
K. N. Murugan, T. Arunkumar, M. Prasath, and V. R. Ganesan
Abstract The occurrence of shock wave over the typical launch vehicle model
with conical heat shield at a free-stream Mach number of 0.95 was studied using
flow visualization techniques. Both oil flow pattern, as well as shadowgraph image
techniques, have been used to trace the flow features over the model. The study was
conducted at a various angle of attack from 0◦ to 4◦ in steps of 1◦ . Instantaneous
shadowgraph images of shock wave pattern over the surface of the conical heat
shield region were captured using a highspeed camera. The image shows a series
of compression weak waves which forms a standing normal shock wave on the
heat shield region at angles of attack of 0◦ , 1◦ , and 2◦ . Further increase in angle of
attack, the normal shock wave splits into lambda-shaped shock pattern. At an angle
of attack 3◦ and 4◦ , shock-induced separation and reattachment process result in
shock oscillations along the surface of constant heat shield cylindrical region. The
frequency of shock wave oscillations was found to increase with the angle of attack
from 3◦ to 4◦ . The oil flow pattern clearly evident that the average distance covered
during the shock oscillations.
Keywords Normal shock wave · Lambda shock wave · Separated flow ·

Shockwave oscillations
1 Introduction
Transonic flow over the conical heat shield region with typical launch vehicle expe-
riences more unsteady pressure fluctuations and buffet loading due to shock waves
and associated boundary-layer interaction [1]. Chevalier and Robertson [2] were first
to report the occurrence of alternating flow phenomena and observed the unsteady
K. N. Murugan (B) · T. Arunkumar

NTAF Division, CSIR-NAL, Bangalore 560037, India
M. Prasath · V. R. Ganesan
VSSC, Thiruvananthapuram 695022, India

492 K. N. Murugan et al.
pressure fluctuations on the heat shield region of launch vehicle. There have been
experimental [3] as well as computational studies [4] to understand the transonic flow
over such kind of configuration. Shockwave location and associated boundary layer
separation point on the heat shield cylindrical (HSCY) region were highly dependent
on the nose geometry, angle of attack, boat tail angle for the transonic flow region
[1, 2, 5, 6]. The available literature is limited to study the shock wave dynamics on
the heat shield region of typical launch vehicle. Therefore, the present study aimed
to understand the dynamics of shock wave oscillation over the conical heat shield
region of a typical launch vehicle by means of the shadowgraph technique. The con-
tinuous change of shock wave pattern over the configuration as the angle of attack
increased from 0◦ to 4◦ in steps of 1◦ . Time series data of shock wave oscillation also
traced using image processing.
2 Experimental Methods
2.1 Test Facility
Flow visualization study was carried out in 0.6 m wind tunnel at CSIR-NAL. This
facility is an intermittent blowdown type tri-sonic wind tunnel and has a test section
of 0.6 m×0.6 m with a test section Mach number ranges from 0.2 to 4.0. The transonic
test section 2 m long and is of slotted wall configuration with top and bottom walls
having an open area ratio of 6% and side walls with an open area ratio of 4%. The
opaque window size of 9.5 times model diameter (D) of either side of tunnel test
section have been used for optical flow visualisations. The models are mounted on
the sting from the existing model support system of the tun-nel. The model angle of
incidence may be varied from −12◦ to 24◦ continuously or in step modes during a
run and model can be rolled and locked from 0◦ to 360◦ prior to run. The static and
total pressures of the tunnel were measured with a 0.1034 MPa differential pressure
transducers with an accuracy of 0.1% of full scale. The model angle of incidence
was fixed for each run and free stream Mach number was maintained nearly constant
with an allowable variation of 0.005.
2.2 Configuration Details
Wind tunnel experiments have been carried out on a typical launch vehicle model
having bulbous Pay Load Fairing (PLF). The schematic as well as photograph view of
scaled model mounted in the wind tunnel is shown in Fig. 1. The complete configura-
tion features a central cylindrical body with a diameter of D and heat shield diameter
of 1.25D attached with conical nose (20◦ ) cone and two strap-on boosters.The strap-
ons are a cylindrical body with a diameter of 0.8D. The distance between nose cone
Shockwave Oscillations Over the Conical Heat Shield Region … 493
Fig. 1 Schematic sketch of tunnel view of model mounted 0.6 m tunnel
Fig. 2 Photograph of typical launch vehicle model mounted 0.6 m tunnel
to strap-on nose cone for the configurations about 4.4D. The juncture of the con-
ical nose to the HSCY region as a starting location, X and length of HSCY, were
considered in this study (Fig. 2).
2.3 Test Condition
The model was tested in 0.6 m tunnel at free-stream Mach number of 0.95 with a
total pressure of 1.72369 bar and static pressure of 0.969 bar. Model orientation was
kept in yaw plane as shown in Fig. 1 and for all the angle of incidences (θ = 0◦ , 1◦ ,
2◦ , 3◦ , and 4◦ ).
2.4 Flow Visualisations and Image Processing
Figure 3 shows the Z-type schileren arrangement used for the optical flow visual-
isation. Present flow visualisation study have been carried out without knife-edge
which makes shadowgrpah. The shadowgraph videos were captured using PHON-
TAM v341 high-speed camera. The videos were acquired 1000 frames per second at
each angle of incidence (θ = 0◦ , 1◦ , 2◦ , 3◦ , and 4◦ ) with the resolutions of 640×480
and an exposure time of 4s for the total time of 3 s. Further, the shadowgraphs were
acquired at the higher frame rate (4096 fps) with the resolutions of 512×384 in the
HSCY region at an angle of incidence (θ = 3◦ , and 4◦ ) to capture the time traces of
shock wave oscillations on the HSCY region.
Shadowgraph videos were converted into images with the time interval of 1/fps
using the Phantom software. These images are cropped in the region of interest using
the MATLAB image processing toolbox. The complete length of the cylindrical
region was taken to trace the shock wave oscillation. The position was traced at few
fixed transverse locations and later extended to the surface of HSCY region. The
position (X), was normalized by the total length of HSCY (L) as the function of
time. Power spectral density of the time traces has been carried out to capture the
characteristics frequency of shock oscillations.
An oil flow pattern was captured to complement the average shock wave foot
location and associated flow feathers in the HSCY region. The mixture of Olic acid,
oil and titanium di-oxide in the ratio of 1:2:3 has been used to sprinkle on the model
and allowed to exposure in short duration of required test flow Mach number. Images
of oil flow pattern captured for the angles of θ = 3◦ , and 4◦ at Mach number of 0.95.
Fig. 3 Z-type schliren arrangement 0.6 m tunnel: PM-plane mirror, CM: collimeter mirror
Instantaneous shadowgraph images of the transonic flow over conical heat shield
region of a typical launch vehicle model for the various angles of incidence (θ =
0◦ 1◦ , 2◦ , 3◦ ,and 4◦ ) are shown in Figs. 4 and 5. From the shadowgraph images,
presence of normal shock wave and followed by boundary layer separation at the end
of the conical heat shield region was evident for θ = 0◦ , 1◦ , and 2◦ . Interestingly, the
shock-induced separation point S moves forward as the angle of incidence increases
from θ = 0◦ to 2◦ . This qualitative information supports the results of mean and
fluctuating pressure measured on the typical launch vehicle models [6].
Further, increase in angle of incidence to θ = 3◦ and 4◦ , boundary layer separation
point S induces the normal shock wave to detect and splits into the two shocks (C1:
oblique shock wave and C2: separation shock wave) with the triple point (I) to form
the lambda like shock pattern shown in Fig. 5. The situation is similar to the case of
Fig. 4 Typical shadowgraph images of transonic flow over the bulbous heat shield region at angle
of incidence of θ = 0◦ to 2◦ : S-separation point
Fig. 5 Typical shadowgraph images of transonic flow over the bulbous heat sheild region at various
angle of incidence θ = 3◦ and 4◦ : S-separation point, C 1 -incident shock, C 2 -separation shock, C 3 -
transmitted shock, I-intersection point
Type VI shock-shock interference as reported by Delery and Bur [7]. An upstream

disturbance caused by the separated shear layer which induces the shock wave to
move forward and back-ward along the entire length of the HSCY region. Violent
shock wave movements on the heat shield region were observed. The influence of
shock oscillations on the heat shield region of the typical launch vehicle model
induces more unsteady pressure fluctuations level and it has been already reported
in the earlier work on a similar model [6].
A typical instantaneous images of different time interval have been shown in
Figs. 6 and 7 for both the angle of incidence (θ = 3◦ and 4◦ ). These images are clearly
shown the presence of a shock wave and its associated boundary layer separation point
moves with the time. The images shown here are chosen randomly to show the typical
shock location movement on the heat shield cylindrical region at a different instance
of time. The time traces of shock wave movement were traced using MATLAB image
processing tool. Power spectral density was taken to this time series to represent in
the frequency domain.
Figure 8 shows the time series data of shock wave displacement over the heat
shield length (L) at an angle of incidence, θ = 0◦ , 1◦ , and 2◦ . As expected, the shock
wave displacements on the HSCY region found to be stationary at θ = 0◦ . As the
angle increases (θ = 1◦ , shock wave is found to move downstream location on the
starting point of boat tail region.
The physical location of the normal shock wave on the heat shield region is 0.8 L
0.88 L for the θ = 0◦ and 1◦ , respectively. These locations are close with the boat-tail
starting point and followed by the flow separation. However, θ = 2◦ , the data, as well
as image, show the negligible amount of shock wave movement over the time period.
Fig. 6 Instantaneous shadowgrap images of shock wave displacement associated with the flow
separation on heat shield at θ = 3◦
Fig. 7 Instanteneous shadowgrap images of shock wave displacement associated with the flow
separation on heat shield at θ = 4◦
Fig. 8 Time series data of shock wave displacement on the bulbous heat shield region at various
angle of incidence
Fig. 9 Time series data of shock wave displacement on the bulbous heat shield region at various
angle of incidence
This may be true since this angle is just below the angle where violent unsteady shock
movement starts.
As the angle of incidence increases further to the θ = 3◦ and 4◦ , the time series
data extracted from high speed images shows violent shock wave movements on the
heat shield region as seen in Fig. 9. The time trace of shock wave displacement for
both angle of incidence (θ = 3◦ and 4◦ ) found to be similar. Interestingly, the periodic
pattern were able to infer from the time series data.
Power spectral density of time series data was carried to obtain the characteristics
frequency of shock wave oscillation. Figure 10 shows the comparison of normalized
power spectral density of shock wave oscillations over the range of frequency for the
angle of incidence θ = 3◦ and 4◦ on the leeward side of HSCY region. The shock wave
oscillates more dominantly at angles of incidence of 4◦ as compared to 3◦ angles.
Interestingly, the fundamental frequency of shock wave oscillations at these angles
is 150 and 290 Hz. However, a close comparison of the spectral data reveals that the
Fig. 10 Normalized power spectral density of shock wave oscillation for an angle of incidence, θ
= 3◦ and 4◦
Fig. 11 Oil flow pattern at an angle of attack a θ=3◦ , and b θ = 4◦ for Mach 0.95: S-separation
point, R-re-attachment point
model at an angle of attack of 4◦ found to more prominent of the amplitude as well

as frequency.
The imprint was visualized in oil flow pattern and shown in Fig. 11. The oil flow
pattern shows the shock foot imprint on the leeward side of the HSCY region for both
θ = 3◦ and 4◦ followed by the separation and flow re-attachment. Coincidentally, the
shape of shock foot curvature resembles the spectral density of time traced data. Broad
shock foot curvature was found to θ = 3◦ , whereas narrow shock foot curvature in θ
= 4◦ . These flow pattern over the model will be very useful towards the aerodynamic
designing of HSCY region for launch vehicles. In addition to that understanding and
identifying the shock-induced flow separations in the HSCY region mainly helps to
the designer for considering acoustic loads.
4 Conclusions
Flow visualization study has been conducted to capture the shock wave oscillation
over the heat shield of a typical launch vehicle at free-stream Mach 0.95 and angle of
attack of θ = 0◦ to 4◦ in steps of 1◦ . A standing normal shock wave associated with
local separation point was observed in the region of 0.8–0.88 L in the heat shield at
the angle of incidence 0◦ , 1◦ , and 2◦ . As the angle of attack increases to 3◦ and 4◦ , the
normal shock wave splits into lambda-shaped shock wave pattern and followed by
flow separations. In addition to that, it was found to oscillates over the entire HSCY
region. Time series data of the shock wave oscillation were obtained using image
processing. The frequency and amplitude of shock wave oscillations over HSCY
region were found be increases with the angle of attack. The oil flow pattern was
supporting the observation from the shadowgraph images. The shock foot location,
flow separation, and re-attachment were clearly imprinted in the oil flow pattern.
These findings from flow visualizations will definitely help towards aerodynamic
design which will be compact for aeroacoustics loads.
Acknowledgements This work was funded by Vikram Sarabhai Space Centre (VSSC), Thiru-
vananthapuram. Authors thank Director, NAL and Head, NTAF for their approvals. The author also
thanks Dr. G K Suryanarayana and Mr. Nitin Khamesra for his valuable suggestion and discussion
during the course of experiments. The authors would like to thank Mr. Baburajan and Mr. Roopesh
for their continuous support and dedication towards capturing quality images using a high-speed
camera. Also, the authors greatly acknowledge and thanks for the cooperation of staff members
from 0.6 m tunnel, model shop and design office. Finally, Dr. Jayaprakash, N Murugan, Mr. Rajan
Kurade, Mr. E Narendran NAL for their support towards image processing and analyzing the results.
References
1. Coe Charles F, Steady and fluctuating pressures at transonic speeds on two space-vehicle payload
shapes. NASA TM X-503
2. Chevalier HL, Robertson JE, Pressures fluctuations resulting from an alternating flow separation
and attachment at transonic speed. AEDC-TDR-63-204
3. Ramasamy MA, Rajendra G (1978) Experimental investigation of transonic flow past a blunt
cone-cylinder. J Spacecraft Rockets 15:120123
4. Purohit SC (1986) Navier-stokes solution for a conical payload shroud. J Spacecraft Rockets
23:590–596
5. Kumar R, Viswanath PR, Prabhu A, Mean and fluctuating pressure field in boat-tail separated
flows at transonic speeds. In: 39th AIAA aerospace sciences meeting and exhibit, AIAA 2001-
0582
6. Nath B, Srinivasan N, Murugan KN, Arun Kumar T, Suryanarayana GK, Prasath M, Ganesan
VR (2015) Experimental investigation of fluctuating pressures on a scaled model of typical
launch vehicle at transonic Mach numbers. SAROD, page 161
7. Delery JM, Bur RS (2000) Mean and fluctuating pressure field in boat-tail separated flows at
transonic speeds. In: European congress on computational methods in applied sciences and
engineering, Barcelona, 11– 14 Jan 2000
Diverterless Supersonic Intake
for a Generic Stealth Fighter Aircraft
Sameer Karania , Manu Mohan, and Satya Prakash
Abstract The design of air intake systems for modern fifth generation fighter aircraft
is of great significance as it directly affects the stealth characteristics of the aircraft.
The major sources of Radar Cross Section (RCS) in an aircraft are its cavities such as
air intake, diverter passage, weapon bays, etc. A diverterless air intake provides two
fold advantages in supersonic military aircraft. In comparison with traditional intake
with boundary layer diverter, a diverterless intake offers better aerodynamic perfor-
mance at high speeds. It also gets rid of the diverter passage and the bump ahead of the
entry to the intake provides additional shielding to the incident radar waves, thereby
reduces RCS of the aircraft. It also eliminates the need to have moving surfaces used
for efficient compression at high supersonic speeds. This paper presents the design
approach of a Diverterless Supersonic Intake (DSI) on a generic stealth fighter aircraft
and performance comparison with a conventional intake with boundary layer diverter.
Keywords Air intake · Diverterless supersonic intake · Stealth aircraft
1 Introduction
Fifth generation fighter aircraft have requirements to cruise at supersonic speeds

without getting detected by enemy radar systems. This can be achieved by designing
the aircraft for low radar cross section and infrared signature. In order to keep infrared
signature low, the net thrust available must be high enough to overcome drag without
using afterburners. One of the important factors responsible for reduction in available
thrust at supersonic speeds is the reduction in intake performance due to loss in total
S. Karania (B) · M. Mohan · S. Prakash

M. Mohan
S. Prakash

502 S. Karania et al.
Fig. 1 Forebody and intake

with diverter
pressure across shocks and flow separation due to shock boundary layer interactions.
To prevent the ingestion of boundary layer, traditional aircraft offset the intake from
the fuselage thereby creating a cavity which is called diverter (Fig. 1). To avoid
shock losses, various mechanisms like a translating cone or variable ramp systems
are employed. All these systems increase weight, complexity and RCS. In addition,
the aircraft generally employ boundary layer bleed system and spill doors at high
supersonic speeds to avoid instability in intake operation due to shock boundary layer
interaction which may cause intake buzz [1]. All these systems not only increase
weight of the aircraft (thereby adversely affecting aerodynamic performance) but
also lead to increase in mechanical complexity, cost of production, and maintenance.
The concept of Diverterless Supersonic Intake (DSI) was first introduced in 1956
by Simon Brown and Huff of NACA [2]. A diverterless intake does not require
a boundary layer diverter, moving components for supersonic compression or a
boundary layer bleed system. Therefore, it reduces weight and complexity asso-
ciated with conventional air intakes for supersonic fighter aircraft. These aspects of
DSI lead to a better fighter aircraft design with reduction in empty weight, production
cost and maintenance requirements [3–6]. At the present times, DSI can be seen on
many military aircraft namely, the American F-16 (block 30 aircraft, experimental)
[7], X-35, F-35, and the Chinese JF17 [8, 9], J10, J20, and J31.
This paper presents the design and performance assessment of a generic stealth
fighter aircraft with DSI. The total pressure recovery characteristics of DSI are
compared with that of a conventional intake with diverter on a fifth generation fighter
like configuration. This study concentrates on supersonic flight regime up to M =
1.8 at zero angle-of-attack (α) and angle-of-sideslip (β). Subsonic performance is
also presented for the sake of completeness.
Diverterless Supersonic Intake for a Generic Stealth Fighter … 503
2 Design Procedure
A diverterless supersonic intake consists of a three-dimensional bump and a forward

swept cowl lip. The bump serves the dual purpose of external compression as well
as diversion of low energy boundary layer developed on the fuselage. The boundary
layer is diverted by the combined action of bump and forward swept cowl.
The bump design is derived from supersonic flow over a cone (Fig. 2). Streamlines
released in a plane in a CFD derived flow field over a cone is converted into a three-
dimensional surface (Fig. 3) with some modifications to accommodate the influence
of forward fuselage. This compression surface is connected to an aft body which will
then continue as a duct.
Fig. 2 Supersonic flow over

a cone
Fig. 3 Streamlines in conical flow field

Fig. 4 Cowl and Bump

integration
The combined compression surface and the aft body is then integrated with a
forward swept cowl lip. The relative positions of cowl lip and bump is based on the
design Mach number (Fig. 4).
Since the engine fan is a major source of radar cross section, stealth aircrafts must
hide their engine face by a serpentine duct which merges smoothly with the bump
and cowl surfaces.
3 Modeling and Flow Simulation
The flow features and intake characteristics are studied numerically by solving
governing equations of field. All the flow solutions are computed on unstructured
grids. The procedure is discussed in the sections that follow.
3.1 Geometry Modeling
A half body configuration with a symmetry plane is modeled for both conventional
and Diverterless intake configurations (Figs. 5 and 6).
An outflow boundary is placed 2.7 m behind the intake forward most point. A
spherical far field boundary of radius 125 times the engine diameter is modeled in
Fig. 5 Conventional intake

Fig. 6 Diverterless intake
Fig. 7 Computational model
computational process. The outlet and symmetry plane will cut the far field as shown
in Fig. 7.
3.2 Flow Simulation Procedure
Unstructured mixed element mesh is generated using ANSYS ICEMCFD [10]. The
near wall regions are resolved using prism layers with appropriate first layer height
to achieve a y + of 1.The grid is refined near the intake region in order to capture
the steep gradients due to the presence of shock waves. The current grid is arrived at
after a thorough grid independence study.
The flow solutions are computed using steady Reynolds-averaged Navier–Stokes
(RANS) equations using CFD+ + flow solver [11]. The one-equation Spalart–
Allmaras turbulence model is used considering its applicability in aerospace domain
[12]. The flow calculations are made at the Mach numbers varying from subsonic to
Table 1 Flight conditions

Mach 0.2 0.4 0.9 1.1 1.3 1.5 1.8
supersonic for various mass flow rates at zero angle-of-attack and angle-of-sideslip
(Table 1).
4 Calculation of Pressure Recovery
The total pressure recovery of air intake is defined as
PtAIP
PressureRecovery =
Pt∞
where
Pt AI P —is the area averaged total pressure at the Aerodynamic Interface Plane
(AIP) just ahead of the engine entry.
Pt∞ —is the free stream total pressure at the given condition.
The performance of diverterless intake is evaluated in comparison to conventional air

intake in terms of the pressure recovery characteristics at different mass flow rates.
The pressure recovery of the two intakes is compared at all the Mach numbers
given in Table 1 in Figs. 8, 9, 10, 11, 12 and 13. As seen from Figs. 8, 9, 10 and
11 the subsonic and transonic intake performance of a DSI is similar to that of
conventional intake. At higher end of operating engine mass flow rates at supersonic
speed of M = 1.3, the diverterless intake shows an improvement of about 0.5% in
total pressure recovery. At low mass flow rates, the conventional intake exhibits drop
is recoveries with a positive slope, which is indicative of unstable intake behavior.
On the other hand, the diverterless intake shows stable behavior with a negative
slope of the recovery curve throughout the operating mass flow rate at a given Mach
number (Fig. 12). The decisive advantage of DSI over conventional intake can be
seen in Figs. 13 and 14 for Mach numbers of 1.5 and 1.8. The DSI total pressure
recovery characteristics are stable throughout the mass flow range at M = 1.5 and 1.8.
The simulations for conventional intake does not indicate acceptable behavior. The
flow calculations with conventional intake exhibit convergence difficulties, indicating
strong flow unsteadiness at higher Mach numbers, thus no data could be obtained in
operating engine mass flow rate at this flight Mach number.
Fig. 8 Pressure recovery M = 0.2
Fig. 9 Pressure recovery M

= 0.4

= 0.9

= 1.1
The diverterless intake significantly outperforms the conventional intake, espe-

cially at supersonic flight Mach numbers. This is evident in M = 1.8 performance
where the intake behavior is still stable even at lower mass flow rates.

= 1.3

= 1.5
6 Conclusion
The DSI configuration provides an alternative to the conventional intake design

together with additional benefits such as reduced drag, reduced weight, improved
supersonic performance, improved stealth, and reduced complexity. It outperforms
the conventional intake, especially at supersonic flight Mach numbers and shows

= 1.8
stable pressure recovery characteristics at lower mass flow rates. These traits helps
aircraft perform better at supersonic speeds.
References
1. Shedon J, Goldsmith EL (1999) Intake aerodynamics 2nd edn AIAA Education series
2. Prior BJ, Hall CN (1957) Subsonic wind-tunnel tests of various forms of air intake installed in
a fighter-type aircraft, Aeronautical Research Council Reports and Memoranda no. 3134, Sept
1957
3. Hamstra et al (1998) Transition shoulder aystem and method for diverting boundary layer Air,
U.S. Patent Documents no. US5749542, 12 May 1998
4. Hamstra et al (1998) System and method for diverting boundary layer air, U.S. Patent
Documents no. US5779189, 14 July 1998
5. Sander et al (2004) Supersonic external-compression diffuser and method for designing same,
U.S. Patent Documents no. US6793175, 21 Sept 2004.
6. Henne et al (2008) Isentropic compression inlet for supersonic aircraft, U.S. Patent Documents
no. US20080271787, 06 Nov 2008
7. codeonemagazine.com web article
8. Masud J, Akram M (2010) Flow field and performance analysis of an integrated diverterless
supersonic inlet, 48th AIAA aerospace sciences meeting including the New Horizons Forum
and Aerospace Exposition, 4–7 Jan 2010, Orlando, Florida
9. Luo W, Wang Q (2015) Aerodynamic design of a forebody-integrated bump inlet with a S-Shape
Diffuser, IFEESM
10. ICEMCFD (2016) User’s manual. ANSYS INC, USA
11. CFD++ v16.1 User’s manual, Metacomp Technologies Ltd., USA.
12. Spalart PR, Allamaras SR (1994) A One-Equation Turbulence Model for Aerodynamic Flows.
Recherche Aerospatiale, no. 1:5–21
Analysis of Missile Plume Impact
Characteristics on Engine Intake
and Neighboring Stores for a Fighter
Aircraft
Jishnu Suresh, S. Karthik, P. Karthikeya, S. Rajkumar, Mano Prakash,

Sashi Kiran, D. Maharana, and D. Narayan
Abstract This paper deals with validation of predicted trajectory of a BVR missile
launched from an Indian fighter aircraft with flight test results. The trajectory of the
missile is predicted using an in-house code WISe (Weapon Integration and Separa-
tion). The code works based on decay factorization scheme, where the assumption
is that the missile aerodynamic interactions with the parent aircraft during sepa-
ration process is dependent on missile’s horizontal displacement. In the absence
of quantitative flight test trajectory data, the images of trajectories of the missile
captured at various instants from the chase aircraft during test are compared with
the predicted trajectories (pictures) of missile at corresponding time steps. The paper
also contains generation of static plume using a commercial code Fluent. The missile
plume path along with its trajectory is modeled from the static plume using two
different approaches (Gleissl and Deslandes in simulation of missile plumes for
aircraft store compatibility assessments, [1])—(i) Flashlight method and (ii) Particle
ejection method. An assessment of plume ingestion into the air intake has been
carried out and compared with flight test data obtained from temperature sensor. After
successful validation, the rigid plume obtained from flashlight method and flexible
plume from particle ejection method is also used to predict the static temperature at
defined section of a Laser Guided Bomb (LGB) seeker.
1 Introduction
Analyses of safe separation of missile launched from a fighter aircraft not only include
study of its trajectory at various flight conditions but also the thermal interference
between the missile plume and launching aircraft. The hot exhaust gas emanating
J. Suresh · S. Karthik (B) · P. Karthikeya · S. Rajkumar · M. Prakash · S. Kiran · D. Maharana ·

D. Narayan
J. Suresh

512 J. Suresh et al.
from missile motor can impact the neighboring stores and aircraft intake systems
such as engines, air-conditioning, and coolers. The exhaust plume impact on neigh-
boring weapons could severely degrade the weapon systems. The aircraft engine
could blowout if plume impact temperature exceeds the engine relight temperature
envelope.
In the present study, trajectory of a BVR missile launched from an Indian fighter
aircraft is predicted using an in-house code Weapon Integration and Separation
(WISe) [2]. Because of unavailability of quantitative flight test trajectory data, the
images of trajectories of the missile captured at various instants from the chase
aircraft during test are compared with the predicted trajectories (pictures) of missile
at corresponding time steps. Generation of missile plume path along with its trajec-
tory is carried out using a static plume. The static plume has been generated using
a commercial code Fluent [3]. Two different approaches [1]—(i) Flashlight method
and (ii) Particle ejection method have been employed to model the plume path. The
later approach is found to be more realistic and accurate.
Further, the plume ingestion into the air intake has been monitored by computing
the static temperature at each time step of the trajectory at defined section of air
intake and compared with flight test data obtained from temperature sensor. The
plume profile is also used to predict the static temperature at defined section of a
Laser Guided Bomb (LGB) seeker.
2 Methodology
2.1 In-House Code—WISe
The code is developed based on the assumptions that missile aerodynamic interac-
tions with the parent aircraft during separation process are dependent on missile’s
horizontal displacement. Thus, the code requires two basic sets of aerodynamic
data—(i) carriage loads when the missile is attached to the aircraft and (ii) missile
free-stream data base as a function of attitudes. A linear decay function [4] is used to
compute the missile aerodynamic coefficients from carriage flight to its free-stream
value when it is away from the carriage location. A commercial code CFD ++ [5] is
used to generate the required installed load and free-stream data base. The viscous
cross-flow terms are calculated based on empirical relation given by Jorgensen [6] and
added to the inviscid aerodynamic coefficients. The code can also compute missile
dynamic derivatives from the sectional loads and added to the above. The aerody-
namic coefficients variation of missile thrust with time has been modeled based on
the data provided by the missile manufacturer. This missile aerodynamic prediction
module is integrated with 6-DOF module. The missile force and moment data apart
from other parameters are fed into the 6-DOF module to predict the missile’s new
location and orientation.
Analysis of Missile Plume Impact Characteristics … 513
Far-field
Out-flow
D=10
Inlet
L=30
Nozzle
Outlet
Nozzle
Inlet
Fig. 1 CFD domain and nozzle mesh
2.2 Computation of Missile Static Plume
The static missile plume has been generated using a commercial code Fluent by
solving Navier-Stokes fluid dynamics equations of the ideal gas with piecewise
polynomial of Cp function. A standard k-ω turbulent model suited for calculating
the exhaust jet coming through a round aperture is employed for the computation. A
computational domain of 30 m × 10 m is used in order to prevent the flow field from
a boundary effect. A mesh size of 39 thousand has been used in the present study.
The mesh generated using ICEM CFD [7] and the associated boundary conditions
are shown in Fig. 1.
2.3 Modeling of Missile Plume Path
(a) Flashlight method

In this approach, 6-DOF trajectory of the launched missile without plume is
obtained from any trajectory simulation approach. In the present study, the missile
trajectory with respect to the aircraft has been predicted using WISe code. Plume char-
acteristics with respect to longitudinal distance versus radial extension as isotherms,
isomach, and isobar plotlines are generated either from realistic experimental plume
data or from approved validated static plume models. In the present study, as
mentioned above the plume characteristics, shape, and size are generated using
Fluent. The outer radius is considered to have closer to ambient temperature. Plume
radius, temperature, and velocity data at different sections along plume length are
taken from the static plume model obtained from Fluent, and at intermediate sections,
data is interpolated. The temperature and velocity data are extracted from the plume
centreline.
After defining the plume geometry from above-mentioned data, the entire plume
is assumed as rigid body and attached to the missile base with the instantaneous
orientation of missile. This is done by rotating isolated plume with instant Euler
angle of missile with respect to the aircraft, followed by translating the plume to
the missile base. Such plume representations may lead to sometime inaccurate and
unrealistic results.
(b) Particle ejection method
The trajectory of missile and its static plume characteristics is computed with the
approach as mentioned above. It is assumed that a plume simply consists of finite
clouds of gas, exhausting the back end of the missile in a specific way, depending
on the missile motion and attitude. These clouds are time-dependently ejected at the
missile nozzle as very hot and high-speed finite volume gas element which slows
down and cools down to ambient conditions of the surrounding air after passing
specific phase of expansion and contraction. The plume path is obtained by three sets
of known parameters—the motion of the aircraft, the relative position, and motion
parameters of the missile and by the specific plume data of the missile to the actual
missile in motion. The plume path can now integrated in finite time steps of some
milliseconds using all the information above. The path will consist a finite crowd of
clouds aligned to each other according to the order of their age and along the trail of
the missile exhaust area. The path of plume obtained by this method will be flexible,
more realistic, and accurate.
3 Results
The static plume computation has been carried out for M = 0.77, Altitude = 7.6 km
using Fluent as mentioned above. Figure 2 shows the missile plume for the above
flight condition. It can be observed that flow phenomena like shock waves, expan-
sion waves, Mach disk, and shock diamonds have been captured in the simula-
tion. Figures 3 and 4 show the Mach number contour, and the radius of the 253 K
iso-temperature line along the plume axial distance, respectively.
Figures 5 and 6 show the representation of top and front view of rigid and flexible
plume when a missile fired from the fighter aircraft using a hypothetical trajectory
for distinctive differentiation between them. Static plume has been generated at M =
0.77 and altitude = 7.6 km. Rigid plume forms a straight path while flexible plume
follows the missile path which is evident from figures.
Figure 7 shows the predicted translational and angular states of the missile trajec-
tory launched at M = 0.77, altitude = 7.6 km, α = 5°, and β = −2°. As mentioned
Fig. 2 Simulation of missile plume for M = 0.77 and altitude = 7.6 km
Fig. 3 Mach contours of plume for M = 0.77, H = 7.6 km
above, in the absence of quantitative flight test trajectory data, the images of trajecto-
ries of the missile launched at above said flight conditions are captured for time t =
0.2, 0.3, 0.4, 0.5, 0.6, and 0.7 s from the chase aircraft during test are compared with
the predicted trajectories (pictures) of missile at corresponding time steps (Fig. 8).
The figure shows only the core of missile plume attached to the missile base and not
the entire plume. The qualitative comparison is found to be very encouraging.
Further, to assess the plume ingestion into the air intake, flexible and rigid plume
profiles at the flight test condition are generated as explained above. At each time
step of trajectory, static temperatures are monitored at defined sections of air intake
as shown in Fig. 9. The computed static temperature for each time step at the left
intake face is compared with the flight test data obtained from temperature sensor.
Figure 10 shows the comparison of measured and predicted static temperature for
the above-mentioned condition at left-hand side air intake section using rigid plume
Fig. 4 Isoline for 253 K along plume axial distance at M = 0.77, H = 7.6 km
a b
Fig. 5 Rigid plume
approach. It can be observed that the peak temperature has been captured but with
a time shift of 0.076 s. No plume ingestion is observed on the right-hand side air
intake section (Fig. 11).
Figure 12 shows the comparison of measured and predicted static temperature at
left-hand side air intake section using flexible plume approach for the same flight
condition. It can be observed that the peak temperature has been captured exactly at
0.876 s. As observed from figure, the flexible plume approach gives a better match
with flight test data compared to the rigid plume approach. Again, no plume ingestion
is observed on the right-hand side air intake section (Fig. 13).
After successful validation, the rigid and flexible plume profile are also used to
predict the static temperature at defined section of a Laser Guided Bomb (LGB)
a b
Fig. 6 Flexible plume
Fig. 7 Predicted store translational and angular state
seeker as shown in Fig. 14. The LGB is mounted at port side of in-board station of
the aircraft (Fig. 9).
Figure 15 shows the predicted static temperature at each time step on the LGB
seeker section using rigid and flexible plume approaches. LGB seeker experiences
the temperature changes at different time periods because of continuously changing
missile position, orientation, and the flexibility of the plume.
4 Conclusion
In the absence of quantitative trajectory data from flight test, the images of trajectories
of the missile captured at various instants from the chase aircraft during test are
Prediction (T=0.2 s) Flight Test (T=0.2 s)
Fig. 8 Comparison of predicted trajectory and flight test trajectory of missile at different time
instances
Fig. 9 Air intake section

considered for monitoring
static temperature
measured and predicted
static temperature at
left-hand side air intake
section with rigid plume
compared with the predicted trajectories (pictures) of missile at corresponding time

steps. The qualitative comparison is found to be very encouraging.
Missile plume path has been modeled using two different approaches—(i) Flash-
light method (Rigid plume) and (ii) Particle ejection method (Flexible plume).
Thermal interference between the missile plume and engine air intake has been
studied using both rigid plume and flexible plume approaches. The results have been
validated at various time steps with the data available from air intake temperature
sensors. The analysis with flexible plume led to a better comparison with the flight
test data in comparison with rigid plume analysis.
Fig. 11 Rigid plume only entering port side air intake
measured and predicted
static temperature at
left-hand side air intake
section with flexible plume
Fig. 13 Flexible plume only entering port side air intake
Fig. 14 LGB seeker section

considered for monitoring
static temperature
Fig. 15 Predicted static

temperature at LGB seeker
section with Rigid and
Flexible plume approaches
Plume obtained from the above approaches has also been successfully integrated
with the missile trajectory to carry out plume ingestion studies on LGB seeker.
References
1. Gleissl R, Deslandes RM (2005) Simulation of missile plumes for aircraft store compatibility
assessments. 43rd AIAA aerospace sciences meeting and exhibit, 10–13 Jan 2005, Reno, Nevada
2. Jishnu S, Karthik S et al (2018) Validation of In-house code—WISe with EADS-MAST and
Flight test results. ADA/ARD&P/TR/273, Aug 2018
3. FLUENT Ver 13, ANSYS Inc.
4. Deslandes R (1995) Strategies for modeling aerodynamic interference during store separation.
DASA/LME12/S/R/1713/1995
5. CFD++ Ver 15.1, Metacomp Technology Inc., USA
6. Jorgensen LH, Prediction of static aerodynamic characteristics for space shuttle-like and other
bodies at angles of attack from 0° to 180°”, NASA TN D-6996.
7. ICEM CFD Ver 13, ANSYS Inc.
Control of Tailless Aircraft
T. S. Ganesh, M. C. Keerthi, Sabari Girish, S. Sreeja Kumar,
and B. Mrunalini
Abstract This paper presents the details about lateral-directional and longitudinal
control of a tailless aircraft for landing condition. Tailless aircraft configurations are
specifically designed for stealth and the absence of a vertical fin aids in low visibility
to RADAR. For such configurations, yaw control becomes a major challenge. Usage
of control surfaces, known as spoilers, specifically to obtain yaw-controllability
is one of the approaches. For the considered tailless aircraft configuration, it was
observed that due to deflection of spoilers, the effectiveness of elevons was signifi-
cantly hampered. This decrease in effectiveness demands higher deflection of elevons
which is not preferred due to actuator requirements. Another approach to tackle the
yaw-controllability would be to use the existing elevons present in aircraft. This
approach is discussed and analyzed in this paper. Critical landing condition have
been chosen as the design point to cater for, namely: (a) to generate sufficient yawing,
rolling and pitching moments to restore the aircraft (b) by using minimum deflection
angles to control (c) to attain required CL at landing condition. Few results of this
study are presented here.
Keywords Flying wing · Tailless aircraft · Lateral-directional control
T. S. Ganesh · M. C. Keerthi · S. Girish (B) · S. Sreeja Kumar · B. Mrunalini

T. S. Ganesh
M. C. Keerthi
S. Sreeja Kumar
B. Mrunalini

524 T. S. Ganesh et al.
Nomenclature
CL Coefficient of lift
Mx Rolling moment coefficient
My Pitching moment coefficient
Mz Yawing moment coefficient
M Mach number
α Angle of attack
β/AoSS Angle of sideslip
δ Elevon deflection angle
IB Inboard
OB Outboard
1 Introduction
A flying wing is a tailless wing aircraft where the payload and equipment are housed
inside the wing structure. There are a few operating aircraft like Northrop Grumman
B-2 Spirit, Boeing X-45, Dassault neuROn etc., which are tailless. The concept of
flying wing has come into existence because of its excellent payload and range capa-
bilities. These configurations produce less drag compared to conventional aircraft,
as these do not include distinct fuselage and empennage that often are the major
contributors to drag. By reducing drag, performance is improved which in turn leads
to lower fuel consumption. Design of flying wing configuration depends on chal-
lenging aerodynamic parameters like appropriate wing sweep angle, wing twist,
airfoil cross-section, control surface placement, sizing etc. The planform of the
considered tailless aircraft is as shown in Fig. 1 which has two spoilers and four
elevons for control around all three stability axes, namely pitch, roll, and yaw. In
flying wing configurations, spoilers are used for yaw control and elevons are used
for both pitch control/trim and for producing the required CL during various flight
phases. The spoilers in this configuration are positioned ahead of the elevons such
that sufficient control arm is available which is required for generating restoring
yawing moment for directional controllability of the aircraft. The purpose of the
present study is to investigate the aerodynamics of a generic flying wing aircraft
with particular focus on lateral, directional and longitudinal control at high angle of
attack crosswind landing condition.
Use of spoilers to control lateral/directional moments reduces the elevon effec-
tiveness as elevons are positioned behind the spoiler. This is due to the formation of
a strong wake zone created just behind the spoiler when it is deflected. The wake
generated by the spoiler propagates to elevons also. Low pressure zone is created in
the rear of spoiler due to separation; this highly disturbed flow affects the efficiency
of elevons drastically as can be observed from the surface Cp contours shown in
Fig. 2a, b. Due to the deflection of spoiler, the flow over the entire outboard elevon
Control of Tailless Aircraft 525
Fig. 1 Planform view of

aircraft
b.
Fig. 2 Surface Cp contours
and major portion of the inboard elevon is disturbed which reduces their effective-
ness. Amount of pitching moment created by one degree deflection of elevons is
defined as the elevon power i.e., dCPM/dE. Similarly, the change in coefficient of
lift due to unit deflection of elevon is defined as elevon effectiveness i.e., dCL/dE.
Due to the decrement in elevon effectiveness, there is a demand of 8 times the
elevon deflection angle to trim the aircraft when the spoilers are deflected which is not
preferable for various reasons; namely: higher actuator requirements and increased
radar signature. Decrement in elevon effectiveness due to spoiler is a function of
spoiler deflections, elevon deflections and angle of attack. Elevon power decrement
is around 50% when the elevons are deflected upwards and downwards by δ° and is
shown for five angles of attack in increasing order in Fig. 3a, b. Hence, an alternate
a: Elevon power decrement for upward deflection of elevons
b : Elevon power decrement for downward deflection of elevons
Fig. 3 Elevon power decrement
solution for spoiler has been considered using the available elevons to control the
aircraft in all the three directions. Stenfelt et al. have considered the usage of elevons
for lateral-directional control of a similar experimental tailless aircraft configuration
using low speed wind tunnel experiments [1, 2]. Their experiments indicate the
feasibility of usage of elevons for controlling the aircraft about both roll and yaw
axes.
2 Procedure
A systematic approach to account for the contribution of each elevon on aerodynamic

moments has been developed using aerodynamic coefficients obtained using CFD.
Since cross-wind landing is one of the critical operational requirements of a tailless
configuration, operating conditions such as altitude, Mach number, angle of attack
and angle of side slip corresponding to landing condition are considered in this
study. The contribution of each control surface to the roll, pitch and yawing moment
coefficients for the aircraft has been estimated for the considered landing conditions
using CFD simulations. For each elevon, the effect of both upward and downward
deflections, as shown in Fig. 4a–d, have been estimated.
For estimating the aerodynamic coefficients, RANS based solver with Spalart-
Alamaras turbulence model in SU2, an open source Finite volume based solver
from Stanford University has been used. A finite volume, unstructured, hybrid mesh
(tri-tetra-prism) comprising of 28 million finite volume cells was made in Ansys
ICEM CFD after performing grid independence studies for the considered aircraft
configuration at the operating conditions. Surface, volume and prism mesh of the
aircraft are shown in Fig. 5a–c.
To control the three aerodynamic moments, three equations have been considered,
as given in Eqs. (1)–(3), to calculate deflection angles of three control surfaces by
keeping the deflection angle of fourth control surface fixed. Left hand side of the
equations are set to zero for trimming the aircraft. Database for the four configurations
shown in Fig. 4a–d and configuration with no deflections have been generated using
CFD for landing condition. The derivatives shown in the equations indicate the slope
of the aerodynamic coefficient with and without the deflection. In these equations,
the effect of cross-coupling, i.e., the effect of the neighboring elevon’s deflection on
itself has been neglected. The effect of aero-elasticity has also not been considered
as landing condition is subsonic and the aero-elasticity effects are negligible at this
dynamic pressure.
CRM = (CRM)β=10◦ ,δ=0◦ + (∂CRM/∂δ1) ∗ δ1ib-port-down + (∂CRM/∂δ2) ∗ δ2ob-port-up

+ (∂CRM/∂δ3) ∗ δ3ib-star-down + (∂CRM/∂δ4) ∗ δ4ob- star-down (1)
a: IB-star elevon deflected up b: IB-star elevon deflected down
c: OB-star elevon deflected up d: OB-star elevon deflected down
Fig. 4 Elevon combinations studied

a: Surface mesh from top view
b: Volume mesh at a constant X-station
c: Prism layers at a constant X-station
Fig. 5 Mesh details
CPM = (CPM)β=10◦ ,δ=0◦ + (∂CPM/∂δ1) ∗ δ1ib - port - down + (∂CPM/∂δ2) ∗ δ2ob - port - up
+ (∂CPM/∂δ3) ∗ δ3ib - star - down + (∂CPM/∂δ4) ∗ δ4ob - star - down (2)
CYM = (CYM)β=10◦ ,δ=0◦ + (∂CYM/∂δ1) ∗ δ1ib - port - down + (∂CYM/∂δ2) ∗ δ2ob - port - up
+ (∂CYM/∂δ3) ∗ δ3ib - star - down + (∂CYM/∂δ4) ∗ δ4ob - star - down (3)
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
δ2 ∂CRM/∂δ2 ∂CRM/∂δ3 ∂CRM/∂δ4 (CRM)base + ∂CRM/∂δ1∗δ1
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣ δ3 ⎦ = inv⎣ ∂CPM/∂δ2 ∂CPM/∂δ3 ∂CPM/∂δ4 ⎦ ∗ −⎣ (CPM)base + ∂CPM/∂δ1∗δ1 ⎦ (4)
δ4 ∂CYM/∂δ2 ∂CYM/∂δ3 ∂CYM/∂δ4 (CYM)base + ∂CYM/∂δ1∗δ1
Around 1200 combinations of deflection angles for all the four control surfaces
have been explored for which the aircraft can be trimmed about roll, pitch and yaw
axes. The solutions are deemed feasible if all the four control surface deflection
angles are within 20° upwards or downwards. Among the feasible solutions, the
solution with minimum deflection angles has been considered. An algorithm has
been developed for the prediction of the deflection angles required and the check of
feasibility of the solution and has been implemented using MATLAB. The algorithm
involves an iterative procedure in which for a fixed deflection of one control surface,
the deflections of the other three control surfaces is iterated by employing the control
surface contribution slopes in Eqs. 1–4 for trimming the aircraft. The program devel-
oped is modular and is capable of estimating the deflections required for roll, pitch
and yaw control separately as well as for controlling about either two of the axes or
all the three axes combined.
The obtained solutions from the MATLAB program are subsequently evaluated by
performing CFD simulations for configurations with the control surfaces deflected at
the angles obtained from the program. Moment coefficients with respect to a nominal
moment reference point for three probable landing condition angles of attack are as
shown in Table 1. Aircraft landing conditions at a constant AoSS which, correspond to
crosswind conditions, have been simulated using CFD and the aircraft aerodynamic
coefficients are tabulated in Tables 1, 2, 3 and 4. Controllability of aircraft is estimated
about a single axis, two axes and about the three axis. The corresponding results are
as mentioned in Tables 2, 3 and 4.
Table 1 Aerodynamic coefficients for base configuration, i.e., all control surfaces are undeflected
M α β CRM CPM CYM
0.3 Landing α − 2° −10° 0.0167 0.0040 0.0010
0.3 Landing α −10° 0.0187 0.0041 0.0016
0.3 Landing α + 2° −10° 0.0194 0.0037 0.0030
Table 2 Result for yaw control

M α β Mx My Mz
0.3 Landing α −10° 0.0179 0.0026 0.0005
Table 3 Result for Roll/Yaw control

M α β CRM CPM CYM
0.3 Landing α 10° −0.0021 −0.0178 −0.0008
Table 4 Result for Roll/Pitch/Yaw control

M α β CRM CPM CYM
0.3 Landing α −10° 0.0012 0.0015 0.0000
Fig. 6 Yaw control

combination
3.1 Yaw Control
For the yaw controllability of aircraft, trimming about yaw axis has been considered
(Eq. 3). Deflection angle of λ° upward deflection for outboard elevons and downward
deflection of inboard elevon on starboard side wing for sideslip in the same direction
of deflections as shown in Fig. 6 is considered. The data was calculated analytically
and validated with the CFD result as shown on Table 2. It is evident from comparing
Tables 1 and 2 that the yawing moment coefficient (CYM) has been reduced by
~69%.
3.2 Roll-Yaw Control
For the yaw and roll controllability of aircraft, 2 equations have been solved for
CRM and CYM. Elevon deflection angle of λ + 5° downward for both inboard and
outboard on port side wing for sideslip in the opposite direction of deflections as
shown on Fig. 7 is considered. Results indicated a decrement is CRM and CYM as
shown in Table 3.
Fig. 7 Yaw/Roll control

combination
Fig. 8 Roll/Pitch/Yaw
control combination
3.3 Pitch-Roll-Yaw Control
Trimming of aircraft in all the directions has been studied in detail in this section.
One of the feasible solutions, which has least control surface deflection of λ° from the
obtained designs, is as shown in Fig. 8. Both analytically and CFD result indicated
decrement in moments in all the directions. Roll and yaw moment were decreased
greater than 95% and CPM decrement is seen up to 65%. Results are as shown in
Table 4.
4 Future Work
To acquire the required CL at landing condition is challenging task, hence solving of

four equations will be studied in detail at critical landing conditions. The effect
of cross coupling, i.e., the effect of the neighboring elevon’s deflection on its
contribution has to be considered on Eqs. 1–3.
Acknowledgements We wish to thank PGD (CA) & Director-ADA for the continuous technical
support and computing facilities to carry out this work. We acknowledge the invaluable support
and guidance given by Dr. Santhosh P. Koruthu, Mr. Biju Uthup, Mrs. Archana Mhaske, Ms. Renu
Bhatt and Mr. Abhishek Trivedi of ADA, Bangalore, in the execution of this work.
References
1. Stenfelt G, Ringertz U (2009) Lateral stability and control of a tailless aircraft configuration. J
Aircraft
2. Stenfelt G, Ringertz U (2010) Yaw control of a tailless aircraft configuration. J Aircraft

Design and Development of Aerospace Vehicles and Propulsion Systems

Uploaded by

Copyright:

Available Formats

Design and Development of Aerospace Vehicles and Propulsion Systems

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

Design and Development of Aerospace Vehicles and Propulsion Systems

Uploaded by

Copyright:

Available Formats

Lecture Notes in Mechanical Engineering

More information about this series at http://www.springer.com/series/11236

Design and Development

ISSN 2195-4356 ISSN 2195-4364 (electronic)

Mach Number Effect on Aeroacoustic Characteristics of Compressible

Theoretical Design and Performance Evaluation of a Two-Ramp

Robust Flutter Prediction of an Airfoil Including Uncertainties . . . . . . . 305

Multi-objective Optimization Approach for Low RCS Aerodynamic

Dr. Indira Narayanaswamy is a Research Professor at Ramaiah University of

S. R. Nikam and S. D. Sharma

Abstract The present experimental investigation reports the effect of chevron on

Keywords Chevron · Potential core · Streamwise vortex

Considering increased frequency of aircraft operation, noise pollution due to aircraft

© Springer Nature Singapore Pte Ltd. 2021 1

2 Experimental Setup and Test Nozzle

Experiments were conducted in a high-pressure single flow freely expanding cold

Fig. 1 Chevron nozzle L=14.8

4.1 Aerodynamics Results

Centerline Velocity Distribution

Fig. 2 Axial velocity along

Fig. 3 Iso-velocity contours for base nozzle

Fig. 4 Iso-velocity contours for chevron nozzle

Fig. 5 Mass flow rate along

4.2 Acoustics Results

Fig. 8 Far-field OASPL

Important conclusions drawn from the present experimental investigation are

1. Simonich JC, Narayanan S, Barber TJ, Nishimura M (2001) Aeroacoustic characterization,

Rhea George and Raj Kiran Grandhi

Keywords Compression–decompression ramp · Flow separation · Shock

R. George (B) · R. K. Grandhi

© Springer Nature Singapore Pte Ltd. 2021 15

2 Flowfield in the Vicinity of CDR

Table 1 Experimental conditions

Fig. 1 Compression–decompression ramp (all dimensions in mm)

Fig. 2 Static pressure on walls

Fig. 3 Velocity vectors in the vicinity of compression corner

Fig. 4 Velocity magnitude contours in the vicinity of compression–decompression corner

Fig. 5 Pressure variation with grids

Table 2 Flowfield conditions

Two-dimensional, viscous simulations are performed for CDR configuration with

Fig. 6 Pressure variation with angle of inclination of compression corner

Fig. 7 Skin friction coefficient on the walls of compression–decompression corner

Fig. 8 Mach number contours for ramp inclination angle of 12◦

widens considerably. At inclination angles of 20 and 25◦ , a change in the nature of

6.1 Flow Features

6.2 Separated Flow Extents

Fig. 9 Mach number contours for ramp inclination angle of 25◦

Fig. 12 Separation length variation with angle of inclination of ramp

The CFD software, ANSYS Fluent is successfully validated for compression–

1. Fluent User Manual. Fluent Inc. (2006)

C. A. Vinay and S. Bhaskar Chakravarthy

Abstract In the course of an aircraft development program of LTA, a number of

C. A. Vinay (B) · S. B. Chakravarthy

© Springer Nature Singapore Pte Ltd. 2021 25

2 Geometry and Analysis

2.1 Geometry Details

2.2 Mesh Details

The commercially available ANSYS ICEM-CFD software was used to generate

Fig. 1 External view of air intake and nacelle