AIAA 2015-1506

AIAA SciTech
5-9 January 2015, Kissimmee, Florida
53rd AIAA Aerospace Sciences Meeting

A Flight Simulator for Agile Fighter Aircraft and

Nonlinear Aerodynamics

H. A. Carlson ∗ and R. Verberg †

Clear Science Corp., Harford, NY 13784, USA

M. S. Hemati ‡ and C. W. Rowley §

Princeton University, Princeton, NJ 08544, USA
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Physics-based reduced-order models have been developed that can accurately and effi-
ciently simulate key aspects of aircraft flight operations including aerodynamics, aeroelas-
ticity, and control-surface dynamics at subsonic, transonic, and supersonic flight speeds
and rapidly changing, nonlinear post-stall conditions. The modeling technology enables
flight simulations and virtual flight testing of agile (highly maneuverable) fighter aircraft.
Order reduction is effected by transforming from physical space to modal space using the
method of proper orthogonal decomposition. Modal models constructed with a relatively
small set of data from high-fidelity, computationally intensive CFD simulations (where flow
properties are computed in physical space) are capable of accurately predicting the flight
dynamics for a wide range of aggressive aircraft maneuvers in simulations that are signif-
icantly faster than real time. Model accuracy is demonstrated through comparisons with
data from CFD simulations of an open-source fighter aircraft with and without wing stores
(modeled after an F-16) and an F-16 aircraft with articulating control surfaces. Model
evaluations include both rigid and flexible versions of the aircraft.

I. Introduction
light testing of military aircraft is required for the certification of both new assets and existing aircraft
F with new configurations designed to meet expanded mission roles (e.g., new wing-store configurations).
This includes loads testing and flutter testing and constitutes a significant expense in the process of rolling out
the U. S. Air Force (USAF) fleet and maintaining its operational readiness. Modeling and simulation (M&S)
has an important role in the design and analysis of highly maneuverable, fighter aircraft, but the integration
of computational modeling, ground testing, and flight testing has not yet realized the full potential of a
synergistic engineering approach that combines data from these different sources. One limitation in CFD-
based modeling is the amount of wall-clock time required to accurately simulate the complex and often
multi-disciplinary physics associated with agile fighters.
Here, a modeling technology has been developed and demonstrated that accurately simulates the full set
of physics associated with aircraft flight operations with models that have significantly lower computational
overhead. This includes nonlinear dynamics associated with stall and separated flow. The connection
between the physics-based, reduced-order model and the governing equations in the parent CFD model
provides information that traditional models based on systems identification (SI) do not have. Conventional
empirical and semi-empirical aerodynamic models are often restricted to linear dynamics over a small range
of flight conditions, requiring bridging functions, gain scheduling, and other methods of extending the range.
The connection between the physics-based model and the parent CFD model also enables predictions of
spatially distributed surface properties and line loads—along with predictions of the integrated forces and
moments that conventional SI models are limited to providing.
∗ President, Clear Science Corp., 663 Owego Hill Road, Harford, NY 13784 ; AIAA Associate Fellow
† Research Scientist, Clear Science Corp., 663 Owego Hill Road, Harford, NY 13784; AIAA Member
‡ Post-Doctoral Research Associate, Mechanical and Aerospace Engineering, Princeton University, Princeton, NJ ; AIAA

Member
§ Professor, Mechanical and Aerospace Engineering, Princeton University, Princeton, NJ ; AIAA Senior Member

1 of 22

American Institute of Aeronautics and Astronautics

Copyright © 2015 by Clear Science Corp.. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.
 Figures of merit in the modeling technology are accuracy, computational efficiency, and extensibility to
a wide range of problems involving a commensurately wide range of physics. More accurate computational
models will engender higher fidelity analysis during the aircraft design process with higher confidence in the
results, enabling reductions in the required number of expensive and time-consuming ground and flight tests.
This translates into shorter time-to-market cycles and more affordable aircraft. Computationally efficient
models support flight simulations for the design of guidance, navigation, and control (GNC) systems, can
be utilized as pilot training software, can support flight testing to reduce risks to test pilots and aircraft,
and can be integrated into advanced control systems—utilizing the models for real-time state estimation.
Extensibility enables high-fidelity predictions for the broad range of physics that are relevant to flight op-
erations in different types of aircraft performing a range of missions. This includes six degrees-of-freedom
(6-DOF) forces and moments and flight dynamics, flight control system dynamics, aerodynamic loading and
structural dynamics, aeroservoelasticity, aerothermodynamics associated with high-speed (hypersonic) flight,
propulsion dynamics, aero-acoustics, and aero-optics.
The modeling technology that has been developed accommodates operating conditions from low subsonic
to transonic, supersonic, and hypersonic. Models of full aircraft must incorporate the interactive effects of
control-surface dynamics, stores separation, landing gear dynamics, and cargo extraction dynamics. The
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

components of the physical problem can be weakly coupled or sometimes strongly inter-connected with
zeroth-order effects produced by the coupling. Aerothermodynamic loading can alter material properties,
which in turn affect aeroelastic responses and structural integrity. Interactions between the flight control
system software (control laws) and hardware (actuators, sensors) and the flow can produce unanticipated,
nonlinear responses.
High-fidelity models typically solve a set of partial differential equations (PDEs) with time and spatial
location as the independent variables. The dependent variables are the kinematic and thermodynamic
properties of the flow in CFD models, displacements and velocities in structural models, temperatures in
thermal models, etc., computed at (often, a large number of) discrete grid points in physical space. The
framework for virtual flight testing utilizes data from these high-fidelity models to construct physics-based
reduced-order models (PB-ROMs) that replace the PDEs with a small set of ordinary differential equations
(ODEs). Time is the independent variable in the ODEs with a one-to-one correspondence between the
dependent variables in the high-dimensional, parent model and the derived ROM. Order reduction is effected
through a proper orthogonal decomposition (POD) that transforms properties from physical space to a modal
frame.
The POD was originally formulated as an unbiased technique for studying coherent structures in turbulent
flows1 and has since been applied to a number of other problems including near-wall turbulence,2, 3 aero-
optics,4, 5 and aeroelastic systems.6, 7 With properly constructed modes, the reduced system reflects the
important dynamical behavior of the original, high-order system. A direct correspondence between terms in
the governing equations of the parent model and terms in the ROM ensures full representation of the relevant
dynamics. For example, the nonlinear, convective terms in the Navier-Stokes equations are represented
by nonlinear terms in the model. Both inviscid and viscous terms in the high-dimensional equations are
represented in the model, again, reflecting both inviscid and viscous effects in the flow.

II. The POD-based Reduced-order Model

The first step in constructing the ROM utilizes the following correlation tensor of the flow variables,
denoted as q i (x, t): Z
i i
R(mn) = q(m) (x)q(n) (x)dx (m, n = 1, ..., N ) (1)
Ω
i
where indices in parentheses—e.g., “(m)” in the subscript of q(m) —denote CFD solution number. R(mn) is
(n)
symmetric, positive definite, and has an orthonormal set of N eigenvectors (φi ) corresponding to N real
positive eigenvalues. Einstein notation is used here and throughout: repeated indices in a term are summed.
The flow variables are then approximated as
q i (x, t) = a(n) (t)φi(n) (x). (2)
The set of basis functions maximizes the mean square projection of the flow variables:
Z
(n)
a(n) (t) = q i (x, t)φi (x)dx. (3)
Ω

2 of 22

American Institute of Aeronautics and Astronautics

 This means that much of the flow dynamics can be represented in a small number of POD basis functions.
a(n) (t) are referred to as the POD coefficients, and they will be dependent variables in the low-dimensional
model.
Conservation of mass, momentum, and energy are enforced by high-dimensional governing equations, the
PDEs that are discretized and solved in the CFD code. Multiplying by a variational function yields the
weak forms of these equations, and choosing the variational function to be a set of low-dimensional modes
produces the ROM. The procedure yields the following low-dimensional set of coupled ODEs to represent
compressible, viscous flow:
da(n) <0000> <0010> <0001> <0020>
 
= α̃(n) + α̃(n)(µ) η(µ) + α̃(n)(µ) η̇(µ) + α̃(n)(µν) η(µ) η(ν) + Ã<0100>
(np) + Ã <0110>
(np)(µ) η (µ) a(p)
dt  
+ Ã<0200> <0001> <0100> <0110> <0200> (4)
(npq) a(p) a(q) + β̃(n)(ρ)j ζ̇(ρ)j + β̃(n)(ρ)j + β̃(n)(ρ)(µ)j η(µ) + B̃(np)(ρ)j a(p) ζ(ρ)j
<0200>
+ β̃(n)(ρσ)jk ζ(ρ)j ζ(σ)k .

The system contains linear and bi-linear terms involving the model variables, a(n) , η(µ) , and ζ(ρ)j and is
therefore capable of modeling nonlinear dynamics in the flow. η(µ) corresponds to the the rigid and flexible
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

motion of the aircraft, and ζ(ρ)j represents flight speed or propulsion variables when the aircraft engine is
included in the model.
<0000> <0010> <0001> <0020> <0100> <0110>
α(n) , α(n)(µ) , α(n)(µ) , α(n)(µν) , A(np) , A(np)(µ) , A<0200> <0100> <0110> <0200> <0001>
(npq) , β(n)(ρ)j , β(n)(ρ)j(µ) , B(np)(ρ)j , β(n)(ρ)j ,
<0200>
and β(n)(ρσ)j are time-invariant coefficients that are computed once during the model construction phase. A
low-dimensional representation of the 6-DOF force and moment coefficients is derived in a similar manner
from the constitutive relations in the CFD model:

F i = c̃<0000>i + c̃<0010>i
(µ)
<0100>i
η(µ) + C̃(n) a(n) + d˜(ρ)j
<0100>i
ζ(ρ)j (5)

where F i is the i-component of force or moment. c̃<0000>i , c(µ) <0010>i <0100>i

, C̃(n) , and d˜<0100>i
(ρ)j are time-
invariant coefficients; again, computed once during the model construction phase.
Coupling (4) and the force approximation in (5) to a low-dimensional (modal) model of the aircraft struc-
ture produces a PB-ROM for the coupled aeroservoelastic system. By defining ξ(µ) = η̇(µ) , a solution vector,
x = (a(1) , · · · , a(N ) , ξ(1) , · · · , ξ(K) , ξ˙(1) , · · · , ξ˙(K) )T , and a generalized force vector, Q = (Q̃(1) · · · Q̃(K) )T , the
model of the coupled system becomes a set of first-order, nonlinear ODEs:

ẋ = f1 (x, Q) ; Q = f2 (x) (6)

where f1,2 are functions involving linear and bi-linear terms with respect to x. The ROM accounts for
linear and nonlinear effects of structural motion on the flow and both linear and nonlinear dynamics in the
flow that contribute to aerodynamic forcing and the structural response. The structural component of the
PB-ROM is linear but may also include nonlinear terms.

III. Test Configurations

PB-ROMs are constructed and evaluated for two aircraft configurations: a tail-less open-source fighter
(OSF) with and without wing stores and an F-16 with articulating control surfaces. A CFD model of the
OSF was developed by the European Union For the Marie Curie Excellence Team Enabling Certification by
Analysis (ECERTA).8 The OSF is based on an F-16 but does not have any of the restrictions attached to
models of actual military aircraft. One of the OSF models includes wing stores (see the left panel of Figure 1),
and the CFD model is coupled to a structural model with 26 modes based on the F-16.9 The second aircraft
configuration is an actual F-16 with articulating control surfaces (see the right panel of Figure 1). The green
surfaces in the figure are sliding interface planes that permit control surface motion using overset grids.
CFD and aeroelastic CFD (CFD-AE) simulations were performed with HPCMP CREATETM –AV Kestrel,
the fixed-wing-aircraft component of the Department of Defense’s (DoD’s) Computational Research and En-
gineering for Acquisition Tools and Environments (CREATE) Program.10, 11 Kestrel utilizes unstructured
grid topologies and can simulate maneuvering fighter aircraft with six degrees of freedom, articulating control
surfaces, variable flight speed and altitude, flexible structures (utilizing deforming grids), and 6-DOF motion
(solving the equations of motion). The OSF grid has 1.77 million nodes and 4.53 million cells; the F-16 grid

3 of 22

American Institute of Aeronautics and Astronautics

 has 4.72 million nodes and 13.3 million cells. Viscous effects are included in all of the CFD/CFD-AE sim-
ulations, using a Reynolds-averaged Navier-Stokes (RANS) formulation and a Spalart-Allmaras turbulence
model with delayed detached eddy simulation (SA-DDES).
At the time of testing, Kestrel was not yet capable of simulating control surface motion and aeroelastic
responses simultaneously; therefore, these two operational features are treated separately in evaluating PB-
ROMs. In tests with control surface motion, the trailing-edge ailerons, elevators, and rudder are included,
but the leading-edge ailerons are not. Control surface deflection limits for the F-16 are ±25◦ for the elevator,
±21.5◦ for the ailerons, and ±30◦ for the rudder;12 however, the sliding interface planes in the CFD model
can only accommodate motion corresponding to half these limits. The rate limits for the F-16 elevator,
ailerons, and rudder are, 60◦ , 80◦ , and 120◦ per second, respectively, and these limits are imposed in the
CFD simulations.
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Figure 1. Tail-less open-source fighter (OSF) with wing stores (left) and F-16 with articulating control surfaces
(right).

IV. Results
Evaluations consist of comparisons between PB-ROM predictions of forces and moments and values from
the CFD/CFD-AE code at both on-design conditions (corresponding to the inputs assigned in the CFD/CFD-
AE simulations that generate data for PB-ROM construction) and off-design conditions (corresponding to
different inputs). Testing is performed in stages by incrementally increasing the number of operational
variables. Model evaluations were performed with both a rigid and flexible OSF, as described in Section A.
F-16 model testing begins with a single operational variable (flight speed), as described in Section B, followed
by evaluations with nine operational variables (6-DOF rigid aircraft motion and aileron, elevator, and rudder
motion), as described in Section C. The final test couples an aerodynamic PB-ROM of the F-16 to a flight
mechanics model and flight controller, as described in Section D).

A. Rigid and Flexible OSF with Three Degrees of Freedom

Figure 2 contains time histories of three OSF maneuvers with three degrees of freedom (roll, pitch, and yaw)
at Mach 0.8 and sea level that were simulated with Kestrel, one on-design maneuver (top row) to generate
data for PB-ROM construction and two off-design maneuvers (middle and bottom rows) to verify and validate
the PB-ROMs. The left column contains time histories of the Euler angles and the right column time histories
of angles of attack and yaw angles. From the top row, the on-design maneuver consists of a series of pseudo-
random ramps and holds designed to capture aerodynamic responses to impulsive inputs. The off-design
maneuvers were generated by an F-16 flight controller,12 and the changes in aircraft attitude are relatively
smooth. The on- and off-design maneuvers include rapid changes in vehicle orientation approaching 200◦ per

4 of 22

American Institute of Aeronautics and Astronautics

 second (the limit for pilot safety); all three extend well past the stall angle of attack for the aircraft (30◦ ).
For this reason, nonlinear PB-ROMs are constructed and evaluated, one for the rigid aircraft and one for
the aeroelastic.
The PB-ROM includes six aerodynamic (POD) modes—along with the first 10 structural modes. Fig-
ures 3 and 4 compare Kestrel and PB-ROM time histories of force and moment coefficients from the on-design
maneuver of the rigid and flexible OSFs (left and right columns, respectively). The PB-ROMs accurately
track the CFD forces and moments in both cases. Comparing the left and right columns, aeroelastic responses
increase the maxima and minima of the forces and moments by approximately 30%, and the PB-ROMs ac-
curately capture the effects. Figures 5 and 6 compare Kestrel and PB-ROM time histories of lift, drag, and
pitch coefficients (columns left to right) for the two off-design maneuvers of the rigid and aeroelastic OSFs
(left and right columns, respectively). Yaw angles are relatively small in the off-design maneuvers; therefore,
only the lift, drag, and pitch coefficients are shown. Again, aeroelastic responses increase excursions in
the forces and moments, and the PB-ROMs accurately capture the effects. With 512 processors, each 10-
simulated-second CFD-AE simulation consumes two wall-clock days. Approximately five wall-clock hours
are required to construct a PB-ROM, using data from a single CFD-AE simulation, and 288 processors.
PB-ROMs with 6–10 POD modes simulate maneuvering, rigid and aeroelastic aircraft faster than real time.
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Figure 2. Euler angles (left column) and angle of attack and yaw angle (right column) versus time from the
on-design OSF maneuver (top row) and two off-design maneuvers (middle and bottom rows).

5 of 22

American Institute of Aeronautics and Astronautics

 2.5 2.5

CFD CFD
DRAG COEFFICIENT (cD)

DRAG COEFFICIENT (cD)

2.0 ROM 2.0 ROM

1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0
0 2 4 6 8 10 0 2 4 6 8 10
TIME TIME
0.50 0.50
CFD CFD
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

SIDE FORCE COEFFICIENT (cS)

SIDE FORCE COEFFICIENT (cS)

ROM ROM
0.25 0.25

0.00 0.00

-0.25 -0.25

-0.50 -0.50
0 2 4 6 8 10 0 2 4 6 8 10
TIME TIME
3 3
LIFT COEFFICIENT (cL)

LIFT COEFFICIENT (cL)

2 2

1 1

0 0

-1
CFD -1
CFD
ROM ROM
-2 -2
0 2 4 6 8 10 0 2 4 6 8 10
TIME TIME

Figure 3. Time histories of aerodynamic force coefficients from the PB-ROM (blue curves) and the CFD
model (black curves) from the on-design maneuver of the rigid and flexible OSFs (left and right columns,
respectively).

6 of 22

American Institute of Aeronautics and Astronautics

 0.9 0.9
CFD CFD
ROM ROM
ROLL COEFFICIENT (cR)

ROLL COEFFICIENT (cR)

0.6 0.6

0.3 0.3

0.0 0.0

-0.3 -0.3
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

-0.6 -0.6
0 2 4 6 8 10 0 2 4 6 8 10
TIME TIME
1.5 1.5
CFD CFD
1.0 ROM 1.0 ROM
YAW COEFFICIENT (cY)

YAW COEFFICIENT (cY)

0.5 0.5

0.0 0.0

-0.5 -0.5

-1.0 -1.0

-1.5 -1.5
0 2 4 6 8 10 0 2 4 6 8 10
TIME TIME
9 9
PITCH COEFFICIENT (cP)

PITCH COEFFICIENT (cP)

6 6

3 3

0 0

CFD CFD
-3 ROM -3 ROM

-6 -6
0 2 4 6 8 10 0 2 4 6 8 10
TIME TIME

Figure 4. Time histories of aerodynamic moment coefficients from the PB-ROM (blue curves) and the CFD
model (black curves) from the on-design maneuver of the rigid and flexible OSFs (left and right columns,
respectively).

7 of 22

American Institute of Aeronautics and Astronautics

 2.8 2.8
DRAG COEFFICIENT (cD)

DRAG COEFFICIENT (cD)

2.1
CFD 2.1
CFD
ROM ROM

1.4 1.4

0.7 0.7
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

0.0 0.0
0 2 4 6 8 10 0 2 4 6 8 10
TIME TIME
2.8 2.8

CFD CFD
ROM ROM
LIFT COEFFICIENT (cL)

2.1 LIFT COEFFICIENT (cL) 2.1

1.4 1.4

0.7 0.7

0.0 0.0
0 2 4 6 8 10 0 2 4 6 8 10
TIME TIME
10 10

CFD CFD
PITCH COEFFICIENT (cP)

PITCH COEFFICIENT (cP)

8 8

ROM ROM
6 6

4 4

2 2

0 0
0 2 4 6 8 10 0 2 4 6 8 10
TIME TIME

Figure 5. Drag, lift, and pitch coefficients from the first off-design maneuver of the rigid and flexible OSFs
(left and right columns, respectively).

8 of 22

American Institute of Aeronautics and Astronautics

 2.0 2.0
DRAG COEFFICIENT (cD)

DRAG COEFFICIENT (cD)

1.5
CFD 1.5
CFD
ROM POD ROM

1.0 1.0

0.5 0.5
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

0.0 0.0
0 2 4 6 8 10 0 2 4 6 8 10
TIME TIME
2.5 2.5

CFD CFD
ROM ROM
LIFT COEFFICIENT (cL)

LIFT COEFFICIENT (cL)

2.0 2.0

1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0
0 2 4 6 8 10 0 2 4 6 8 10
TIME TIME
9 9

CFD CFD
PITCH COEFFICIENT (cP)

PITCH COEFFICIENT (cP)

6
ROM 6
ROM

3 3

0 0
0 2 4 6 8 10 0 2 4 6 8 10
TIME TIME

Figure 6. Drag, lift, and pitch coefficients from the second off-design maneuver of the rigid and flexible OSFs
(left and right columns, respectively).

9 of 22

American Institute of Aeronautics and Astronautics

 B. F-16 at Varying Flight Speeds from Subsonic to Supersonic
The next test varies the flight speed of the rigid F-16. Control surface angles are fixed at zero degrees, and
the aircraft pitch and yaw angles are fixed at 3◦ and 0◦ . Figure 7 contains on-design flight speed variations
and the corresponding force and moment time histories in the top row, and off-design flight speed variations
and the corresponding force and moment time histories in the bottom row. The variations in speed cover the
subsonic, transonic, and supersonic regimes, and the PB-ROM is nonlinear with 12 modes. High-frequency
oscillations occur in force and moment coefficients from the CFD simulations at supersonic speeds and are
not picked up by the PB-ROM. More modes in the model would capture more of the high-frequency content,
but 6-DOF CFD simulations indicate it has little effect on aircraft motion because of inertia.
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Figure 7. On-design F-16 flight speed variations and corresponding time histories of force and moment coeffi-
cients in the top row; off-design flight speed variations and corresponding time histories of force and moment
coefficients in the bottom row.

C. Maneuvering F-16 with Articulating Control Surfaces

In the next test, nine operational variables are included: 6-DOF rigid motion, control surface deflections,
and flight speed. Aircraft angles of attack and yaw angles are prescribed—rather than computed from
the equations of motion—in the on- and off-design CFD simulations. Figure 8 contains time histories
of vehicle attitude, Mach number, and control surface angles from the on-design maneuver and two off-
design maneuvers. Again, the on-deign maneuver consists of pseudo-random ramps and holds in all of the
operational variables and covers 18 seconds of simulated time. Mach number variations cover all three flight
regimes—subsonic, transonic, and supersonic. The off-design maneuvers are constructed from an F-16 flight
controller, using an empirical aerodynamic model of the F-16.12 The initial flight speed is supersonic in the
first off-design maneuver, and it gradually decreases as the aircraft pitches up and drag increases. In the
second off-design maneuver, the Mach number increases initially from 0.8 to 1.2 and then gradually decreases
as the angle of attack and drag increase. Commands from the controller to the control surfaces are frequent
in the off-design maneuvers, reaching both the deflection limits and rate limits on numerous occasions.
The PB-ROM contains eight POD modes. Figure 9 contains time histories of the 6-DOF force and
moment coefficients from the on-design maneuver. The dominant force and moment coefficients are drag,
lift, and side force and pitching moment; the roll and yaw moment coefficients are five to ten times smaller
in magnitude. Comparing the black and blue curves in the figure, the PB-ROM accurately tracks all six
components—even the roll and yaw moments. Figure 10 contains time histories of the 6-DOF force and
moment coefficients from the first off-design maneuver. Again, the dominant forces and moments are drag,
lift, and side force and pitching moment, and the PB-ROM tracks these well. The largest discrepancies occur
in the roll and yaw moments. The roll moment coefficient from the PB-ROM (the blue curve in the bottom
left pane) is shifted lower than the CFD (black) curve; although, roll moment is an order of magnitude lower
than the other force and moment coefficients. Figure 11 contains force and moment coefficients from the
second off-design maneuver. Again, the largest discrepancies between the PB-ROM and CFD model occur

10 of 22

American Institute of Aeronautics and Astronautics

 in the (smaller) roll and yaw moment coefficients.
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Figure 8. Angle of attack and yaw angle, Mach number, and control surface deflection time histories from
the on-design F-16 maneuver (top row) and two off-design maneuvers (middle and bottom rows) with nine
operational variables.

Figure 12 contains snapshots of surface pressure coefficients from the on-design CFD simulation, along
with predictions from the PB-ROM. The maneuver produces significant pressure variations—both temporal
and spatial—and the PB-ROM captures the spatially distributed properties. Figure 13 contains snapshots
of surface pressure coefficients from the first off-design CFD simulation, along with predictions from the PB-
ROM. Not surprisingly, differences between the CFD and PB-ROM pressures are greater in the off-design
maneuver; however, the spatial and temporal variations are complex, and the PB-ROM captures most of
both. Errors in the PB-ROM predictions of the surface pressure coefficients are defined as
v " #
u1 Z T Z Z TZ
u

2
2
E = 100% × t cCFD
p − cROM
p dx dt ÷ cCFD
p dx dt (7)
T 0 Ω 0 Ω

where Ω denotes the aircraft surface and T is the total simulated time of the maneuver. The error in the
on-design simulation is 4.98%, and the errors in the two off-design simulations are 11.6% and 12.1%.

11 of 22

American Institute of Aeronautics and Astronautics

Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Figure 9. Force (left column) and moment (right column) coefficients from the on-design F-16 maneuver with
nine operational variables.

12 of 22

American Institute of Aeronautics and Astronautics

Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Figure 10. Force (left column) and moment (right column) coefficients from the first off-design F-16 maneuver
with nine operational variables.

13 of 22

American Institute of Aeronautics and Astronautics

Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Figure 11. Force (left column) and moment (right column) coefficients from the second off-design F-16 ma-
neuver with nine operational variables.

14 of 22

American Institute of Aeronautics and Astronautics

Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Figure 12. Pressure coefficients on the aircraft surface at intervals of 2.5 seconds from the on-design maneuver.

15 of 22

American Institute of Aeronautics and Astronautics

Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Figure 13. Pressure coefficients on the aircraft surface at intervals of approximately 1.4 seconds from the first
off-design maneuver.

16 of 22

American Institute of Aeronautics and Astronautics

 D. Maneuvering F-16 with Articulating Control Surfaces and Flight Control
Demonstrations of the modeling technology end with a series of F-16 flight simulations, using the aerodynamic
ROM in conjunction with a flight mechanics model and a flight controller. The first two maneuvers start at
a trimmed flight condition of Mach 0.8, sea level altitude, three degrees angle of attack, and zero yaw angle.
Initial elevator, rudder, and aileron deflections at the trim condition are 0.6856, 0.2102, and 6.645 × 10−3
degrees, respectively. Model evaluations are performed by 1) generating a maneuver with the ROM, 2)
prescribing the same maneuver in a CFD simulation, and 3) comparing forces and moments from the ROM
with values from the CFD model. The maneuvers combine open-loop control (prescribed control surface
motion) and closed-loop control with the controller described below.
The flight mechanics model treats the aircraft as a rigid body of constant mass. Forces and moments are
attributable to the aerodynamics, propulsion, and gravity; forces and moments from other sources (engine
component motion, sloshing fuel, etc.) are not modeled. Engine thrust acts through the aircraft center
of mass along the longitudinal body axis, and the engine is modeled as an instantaneous thrust producer.
The Earth is flat and fixed in inertial space, and both the gravitational field and atmospheric properties are
assumed to be constant (i.e., no variations due to altitude changes). Table 1 identifies the aerodynamic states
and control inputs in the flight simulator. The x, y, z-component forces are Cx,y,z , and the roll, pitch, and
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

yaw moment coefficients are Cl,m,n . Additional variables are air density (ρ), freestream dynamic pressure
2
(q̄ = ρU∞ /2), reference chord length (c̄), reference wing area (S), and aircraft mass (m).

Table 1. Thirteen aerodynamic states and four control inputs in the flight simulator.

Symbol Description
U∞ flight speed
u velocity component in the x direction
v velocity component in the y direction
w velocity component in the z direction
xNorth North displacement (in x direction)
yEast East displacement (in y direction)
h altitude
φ Euler (roll) angle
θ Euler (pitch) angle
ψ Euler (yaw) angle
p roll rate
q pitch rate
r yaw rate
δe elevator deflection angle
δa aileron deflection angle
δr rudder deflection angle
T engine thrust

The vector of aerodynamic states is

h iT
x = U∞ α θ q xNorth h β φ ψ p r yEast , (8)

and the control input vector is

h iT
u = δe δa δr T . (9)

17 of 22

American Institute of Aeronautics and Astronautics

 The six nonlinear equations of motion for rigid-body translation and rotation of the aircraft are
q̄SCx + T
u̇ = rv − qw − g sin θ + (10)
m
q̄SCy
v̇ = pw − ru − g cos θ sin φ +
m
q̄SCz
ẇ = qu − pv − g cos θ cos φ +
m
ṗ = (c1 r + c2 p) q + q̄Sb (c3 Cl + c4 Cn )
c5 pr − c6 p2 − r2 + c7 q̄Sc̄Cm


q̇ =
ṙ = (c8 p − c2 r) q + q̄Sb (c4 Cl + c9 Cn )

where
2
(Iy − Iz ) Iz − Ixz (Ix −Iy +Iz )Ixz Iz
c1 = 2
; c2 = 2
Ix Iz −Ixz ; c3 = 2
(11)
Ix Iz − Ixz Ix Iz − Ixz
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Ixz Iz −Ix Ixz

c4 = 2
; c5 = Iy ; c6 =
Ix Iz − Ixz Iy
1 2
(Ix −Iy )Ix −Ixz Ix
c7 = ; c8 = 2
Ix Iz −Ixz ; c9 = 2
.
Iy Ix Iz − Ixz

The trimmed condition of the aircraft is denoted as

ẋo = f (xo , uo ) = 0. (12)

In general, the trim condition is not unique. First, the state vector (x = x0 ) is specified for a realistic flight
condition, and then the corresponding control inputs (control surface deflections, u = u0 ) are determined by
either computing the roots of (12), using a simplex algorithm, or iteratively reducing the 6-DOF forces and
moments to zero. At trim, the Mach number is 0.8, angle of attack is 0.0068◦ , yaw angle is −0.0004◦ , engine
thrust is 6,211 lbs., and the elevator, rudder, and aileron deflection angles (δe,r,a ) are 0.6856◦ , 0.2102◦ , and
0.0066◦ . These are the initial and final flight conditions in all the maneuvers.
The flight controller is based on a linearized approximation of the fully nonlinear, coupled flight mechanics
model and aerodynamic ROM. Linearizing about the trim condition, the dynamical estimator is ẋ = Ax+Bu.
The time-invariant matrices are computed from approximations of the gradients of (12) with respect to the
aerodynamic states and control inputs:

∂f (x, u)
A ≈ ∈ Rn×n (13)
∂x xo ,uo
∂f (x, u)
B ≈ ∈ Rn×m (14)
∂u xo ,uo

where n is the number states, m is the number of inputs, and ∆xj and ∆uj are perturbations of the j-th
element of the state and control vectors. The approximations are computed iteratively by reducing the
perturbation size until ∂f i (x, u)/∂xj and ∂f i (x, u)/∂uj have converged.
The control law of the linear quadratic regulator (LQR) is u = −Klqr x, and the quadratic cost function
to be minimized is
1 ∞ T
Z
x Qx + uT Ru dt


J= (15)
2 0
where Q ∈ Rn×n and R ∈ Rn×m are positive semi-definite matrices representing differences between the
state and the target state and the amount of required control input, respectively. The optimal gain is
Klqr = R−1 B T P where P = P T ∈ Rn×n is the solution to the algebraic Ricatti equation (ARE): AT P +
P A + Q − P BR−1 B T P = 0. The associated minimum cost is Jmin = xT (t = 0) P x(t = 0)/2. The LQR
uses the following control law to track a prescribed reference state, xref (t): u = −Klqr (x − xref ) + uo where
(xo , uo ) is the trim state about which the dynamical system has been linearized.

18 of 22

American Institute of Aeronautics and Astronautics

 To design the controller, the linearized, dynamical system is decomposed into a component representing
the longitudinal dynamics of the aircraft and another representing the lateral dynamics:
       
U̇∞ U∞ " # β̇ β " #
 α̇   α  δe  ṗ  p δa
       
  = Along   + Blong and   = Alat   + Blat (16)
 q̇   q  T  ṙ  r δr
θ̇ θ φ̇ φ

where Along ∈ R4×4 and Alat ∈ R4×4 and Blat ∈ R4×2 are derived from the corresponding elements of A
and B, respectively. The decomposition assumes that the longitudinal and lateral dynamics are de-coupled.
Errors in the aerodynamic ROM introduce some coupling; however, the eigenvalues of Along ∈ R4×4 and
Alat ∈ R4×4 contain two longitudinal modes (short period and phugoid) and three lateral modes (spiral, roll,
and Dutch roll), which are all consistent with conventional aircraft stability theory.13
The controller consists of two LQRs, one for the longitudinal dynamics and another for the lateral. This
involves separate gains (Klqr,long and Klqr,lat ) and separate weighting matrices, Qlong and Rlong for the
longitudinal controller, and Qlat and Rlat for the lateral controller. All are designated as diagonal. The
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

corresponding AREs and optimal gains are computed in Matlab with the function lqr. To improve tracking
performance, an integral of the error (difference between the state and target) is added by augmenting the
state vector with e = (x − xref ) ∈ Rn : " # " #
ẋ Ax + Bu
= . (17)
ė x − xref
Optimal gains are computed by the Matlab function lqi, and the resulting control law is

u = −Klqr (x − xref ) − Ki e + uo . (18)

The first maneuver consists of an open-loop elevator doublet, followed by a return to the trim condition
with closed-loop control. From the left panel of Figure 14, the elevator is initially deflected to +3.686◦ , held
briefly in that position, then moved to an angle of −9.314◦ and held for 1.04 seconds, at which point the
closed-loop controller adjusts the elevator deflection to return the aircraft to the trimmed state. The other
control inputs are adjusted by the controller from the onset of the maneuver to maintain the other states
at trim. From the right panel of Figure 14, the angle of attack reaches a maximum value of 30◦ , which is
the stall condition of the aircraft. However, the thrust required to reach this attitude exceeds the maximum
engine thrust of the F-16 (approximately 30,000 lbs) by a considerable amount (red curve in the middle
panel of Figure 14). The control surface deflections are limited to one-half of the actual limits in the F-16
because of the sizes of the sliding interface boundaries in the CFD grid and therefore do not have enough
control authority at lower speeds and the corresponding lower dynamic pressures.

Figure 14. F-16 open-loop elevator doublet maneuver, followed by a return to the trim condition with closed-
loop control.

To evaluate the accuracy of the flight simulator and associated aerodynamic ROM, a CFD simulation
was performed, prescribing both the control inputs and the aerodynamic states from the flight simulator.
Figure 15 compares the lift and drag coefficients from the CFD simulation and the ROM-based flight simula-
tion. These are the dominant forces acting on the aircraft during the maneuver; the other four aerodynamic

19 of 22

American Institute of Aeronautics and Astronautics

 force and moment coefficients are an order of magnitude lower. The ROM tracks the CFD values of lift and
drag quite well with the exception of under-predictions of the maxima that occur when the angle of attack
reaches its maximum.
Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Figure 15. Lift and drag coefficients from the elevator doublet maneuver.

The second maneuver consists of an open-loop rudder doublet, followed by a return to the trim condition
with closed-loop control. From the left panel of Figure 16, the rudder is initially deflected to +14.21◦ , then
moved to an angle of −13.79◦ and held for 0.26 seconds, at which point the closed-loop controller adjusts
the rudder deflection to return the aircraft to the trimmed state. The other control inputs are adjusted by
the controller from the onset of the maneuver to maintain the other states at trim. From the right panel of
Figure 16, the yaw angle reaches maximum and minimum values of 14◦ and −8◦ . Again, the thrust required
to reach these vehicle orientations exceeds the maximum engine thrust of the F-16 by a considerable amount
(red curve in the middle panel of Figure 16), because the control surface deflections are limited to one-half
of the actual limits.

Figure 16. F-16 open-loop rudder doublet maneuver, followed by a return to the trim condition with closed-
loop control.

Figure 17 compares the dominant force and moment coefficients (side force and yaw moment) from the
CFD simulation (using prescribed inputs and states from the flight simulator) and the ROM-based flight
simulation. The ROM tracks CFD values of the side force most closely but also tracks the yaw coefficient
reasonably well.
The last maneuver is designed to push the aircraft into a deep stall (α > 30◦ ) without exceeding the
maximum thrust level of the F-16. The previous maneuvers were limited to 10 seconds of simulated time in
order to avoid long compute times. By relaxing the limit and allowing flight speed to drop below Mach 0.5,
the elevator doublet in the left panel of Figure 18 takes the aircraft to a 40◦ angle of attack (right panel of
the figure). Because the Mach number does drop below 0.5 (middle panel), a ROM is constructed from a
new on-design maneuver with a Mach number range from 0.1 to 1.2. The initial elevator deflection is −5◦ .

20 of 22

American Institute of Aeronautics and Astronautics

Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Figure 17. Side force and yaw coefficients from the rudder doublet maneuver.

Using the closed-loop controller, the aircraft is returned to its initial state in 14 seconds. The maximum
thrust during the maneuver does not exceed the F-16’s maximum thrust. Figure 19 compares the dominant
force coefficients (lift and drag) from the CFD simulation and the ROM-based flight simulation.

Figure 18. Second F-16 open-loop elevator doublet maneuver, taking the angle of attack to 40◦ , followed by a
return to the initial condition with closed-loop control.

V. Conclusions
The PB-ROM that includes control surface deflections, flight speed, and aircraft attitude as operational
variables is capable of simulating the agile F-16 over a wide range of flight conditions: subsonic, transonic,
and supersonic speeds, post-stall, and rapid changes in all of the variables. The model was constructed with
data from a single high-fidelity CFD simulation—consuming 58 wall-clock hours with 512 processors in the
CFD simulation and an additional five wall-clock hours with 384 processors for model construction. With
the PB-ROM, the ratio of simulated time to wall-clock computing time is 71.2. For example, eight hours of
wall-clock time on a single processor will simulate 569 hours of flight time. The low computational overhead
enables rapid analyses of the full operational space. The POD modes of the model facilitate predictions of
not only the integrated forces and moments acting on the aircraft but also the spatially distributed loading.
This feature enables virtual flutter testing, loads testing, and even predictions of long-term fatigue loading.

Acknowledgments
This material is based upon work supported by the U.S. Air Force Office of Scientific Research (AFOSR)
under Contract FA9550-13-C-0012. Any opinions, findings, conclusions, or recommendations expressed in

21 of 22

American Institute of Aeronautics and Astronautics

Downloaded by ROKETSAN MISSLES INC. on January 8, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-1506

Figure 19. Lift and drag coefficients from the second elevator doublet maneuver.

the material are those of the authors and do not necessarily reflect the views of the U. S. Air Force. CFD data
were generated with the Kestrel CFD code, a product of the CREATETM –AV element of the Computational
Research and Engineering for Acquisition Tools and Environments (CREATE) Program sponsored by the
U. S. Department of Defense HPC Modernization Program Office.

References
1 J. L. Lumley, “The structure of inhomogeneous turbulent flows,” Atmospheric Turbulence and Radio Wave Propagation,

ed. A. M. Yaglom and V. I. Tatarski, Moscow: Nauka, pp. 166–178, 1967.

2 N. Aubry, P. Holmes, J. L. Lumley, and E. Stone, “The dynamics of coherent structures in the wall region of a turbulent

boundary layer,” J. Fluid. Mech., 192, pp. 115–173 (1988).

3 G. Berkooz, H. Carlson, P. Holmes, and J. L. Lumley, “Progress in understanding the dynamics of coherent structures

in the wall layer,” in Near-Wall Turbulent Flows edited by So et al., Elsevier, Amsterdam, 1993, pp. 1–21.
4 V. Thirunavukkarasu, H. A. Carlson, R. D. Wallace, P. R. Shea, and M. N. Glauser, “Model-Based Feedback Flow

Control Development and Simulation for a Pitching Turret,” AIAA J., 50, No. 9, pp. 1834–1842, 2012.
5 R. D. Wallace, P. R. Shea, M. N. Glauser, T. Vaithianathan, and H. A. Carlson, “Simulation-Guided, Model-Based

Feedback Flow Control for a Pitching Turret,” AIAA J., 50, No. 8, pp. 1685–1696, 2012.
6 H. A. Carlson, R. Verberg, J. A. Lechniak, and K. K. Bhamidpati, “Flutter Predictions with a Reduced-order Aeroelastic

Model,” AIAA Paper 2012-1232, 50th Aerospace Sciences Meeting, Nashville, TN, January 2012.
7 H. A. Carlson, R. Verberg, J. A. Lechniak, and D. A. Reasor, “A Reduced-Order Model of an Aeroelastic Wing Exhibiting

a Sub-Critical Hopf Bifurcation,” AIAA Paper 2013-1111, 51st AIAA Aerospace Sciences Meeting, Grapevine, TX, January
7–10, 2013.
8 European Union For the Marie Curie Excellence Team Enabling Certification by Analysis (ECERTA):

www.cfd4aircraft.com/4ecerta intro.php/.
9 Charles M. Denegri, Jr., James A. Dubben, and Daniel L. Maxwell, “In-Flight Wing Deformation Characteristics During

Limit-Cycle Oscillations,” J. Aircraft, 42, No. 2, pp. 500–508, 2005.

10 High Performance Computing Modernization Program Computational Research and Engineering Acquisition Tools and

Environments (CREATE), https://portal-create.hpc.mil.

11 S. A. Morton, D. R. McDaniel, D. R. Sears, B. Tillman, and T. R. Tuckey, “Kestrel—A Fixed Wing Virtual Aircraft

Product of the CREATE Program, AIAA Paper 2009-0338, 2009.

12 B. L. Stevens and F. L. Lewis, Aircraft Control and Simulation: Appendix A, Wiley, Hoboken, NJ, Second Edition,

October 2003.
13 R. F. Stengel, Flight Dynamics, Princeton University Press, Princeton, NJ, 2004.

22 of 22

American Institute of Aeronautics and Astronautics