AIAA 2016-1751

AIAA SciTech
4-8 January 2016, San Diego, California, USA
AIAA Atmospheric Flight Mechanics Conference

Robust Control Design for Active Flutter Suppression

Julian Theis∗, Harald Pfifer† and Peter Seiler‡

University of Minnesota, Minneapolis, MN 55455, USA

Flutter is an unstable oscillation caused by the interaction of aerodynamics and struc-

tural dynamics. It can lead to catastrophic failure and therefore must be strictly avoided.
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on January 18, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-1751

Weight reduction and aerodynamically efficient high aspect ratio wing design reduce struc-
tural stiffness and thus reduce flutter speed. Consequently, the use of active control systems
to counter these adverse aeroservoelastic effects becomes an increasingly important aspect
for future flight control systems. The paper describes the process of designing a controller
for active flutter suppression on a small, flexible unmanned aircraft. It starts from a grey-
box model and highlights the importance of individual components such as actuators and
computation devices. A systematic design procedure for an H∞ -norm optimal controller
that increases structural damping and suppresses flutter is then developed. A second key
contribution is the development of thorough robustness tests for clearance in the absence
of a high-fidelity nonlinear model.

I. Introduction
Aeroelastic flutter involves the adverse interaction of aerodynamics with structural dynamics and pro-
duces an unstable oscillation that often results in structural failure. Conventional aircraft are designed such
that flutter does not occur within their range of operating conditions. This is usually achieved through the
use of stiffening materials and thus at the expense of additional structural mass. The use of active control sys-
tems to expand the flutter boundary could therefore lead to a decrease in structural mass and consequently
increase fuel efficiency and performance for future aircraft. The present paper contributes a systematic H∞
control design for the University of Minnesota’s mini MUTT (Multi Utility Technology Testbed) aircraft.
The mini MUTT is a small, remote-piloted aircraft that resembles Lockheed Martin’s Body Freedom Flutter
vehicle1 and NASA’s X56 MUTT aircraft.2
Early research on active flutter suppression relied to a large extend on what is known as collocated
feedback within the structural control community. Collocated feedback employs sensors and actuators in the
same location. The special property of such feedback loops is the presence of a complex pair of zeros in the
immediate vicinity of the lightly damped poles of the structural mode, see e. g. Ref. 3. A closely related
approach, termed the concept of identically located force and acceleration in Ref. 4, was successfully applied
to address the damping of structural modes on the B-1 aircraft.5, 6 A similar configuration was also used in
the first flight test beyond flutter speed, conducted in 1973 on a modified B-52 aircraft.7 The control system
on that aircraft involved two single feedback loops that fed back filtered vertical acceleration signals, acquired
on the wing, to control surfaces located nearby (outboard ailerons and flaperons). Collocated acceleration
feedback is also proposed in various other publications concerned with flutter suppression, e. g. Refs. 8–11.
Collocated controllers are, in general, easily designed using root-locus analysis and have favorable robustness
properties. For the mini MUTT aircraft, however, collocated acceleration feedback tends to destabilize the
short period dynamics. This is attributed to two facts. First, the collocated control surfaces (outboard flaps)
have a much higher pitch effectiveness on a flying-wing compared to ailerons on a conventional aircraft.
Second, the frequencies of the short period dynamics and the aeroelastic modes are very close to each other,
making a frequency separation difficult to achieve. An alternative design approach for a flutter suppression
controller is therefore required.
∗ Research Fellow, Aerospace Engineering and Mechanics Department, email: [email protected]
† Post-Doctoral Associate, Aerospace Engineering and Mechanics Department, email: [email protected]
‡ Assistant Professor, Aerospace Engineering and Mechanics Department, email: [email protected]

1 of 13

American Institute of Aeronautics and Astronautics

Copyright © 2015 by Julian Theis. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.
 A recent flight test demonstration of active flutter suppression on Lockheed Martin’s Body Freedom
Flutter vehicle is reported in Ref 12. A blend of multivariable linear quadratic Gaussian (LQG) controllers
is used, but due to the proprietary nature of the research, little detail about the design is provided. NASA
Dryden is also currently developing multivariable flutter suppression controllers based on LQG techniques
for their X-56 MUTT aircraft, see Ref. 2. An H∞ control design for the X-56 is detailed in Ref. 13 and also
extended to linear parameter-varying (LPV) control. Similar multivariable robust control designs for flutter
suppression were previously reported in References 14–16.
For the present paper, H∞ closed-loop shaping is selected as the control design method. It is believed
to be particularly suitable for the considered application for three main reasons. First, the objective of
providing damping is easy to express in terms of reducing peaks in the frequency response of closed-loop
transfer functions. Second, control activity is easy to confine to a specific frequency range. Roll-off and
wash-out filters can be directly incorporated into the controller, without the need of ad-hoc modifications.
This avoids undesired interaction of the control loop with unmodeled parts of the plant or different layers of
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on January 18, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-1751

control systems. Finally, H∞ controllers minimize a worst-case metric. They thus tend to provide a high level
of inherent robustness when all possible loop break points are included in the performance specifications.
The latter two reasons are considered advantages over other popular multivariable design techniques such as
LQG control.
A mathematical model of the mini MUTT aircraft is described in Section II, with an emphasis on
accurately capturing phase loss due to parasitic dynamics. Section III provides the necessary background
about H∞ control and details the design of an active flutter suppression controller with a mixed sensitivity
formulation. The controller is analyzed with respect to a variety of stability margins in Section IV and shown
to be very robust. The design presented in this paper parallels ongoing research on flutter suppression for the
mini MUTT vehicle based on an adaptive linear quadratic regulator formulation (Ref. 17) and on pitch rate
feedback (Ref. 18). Flight tests are scheduled for spring 2016 and will provide a comparison of the different
approaches.

II. Modeling of the Aircraft

The aircraft under consideration in this paper is the mini MUTT, built at the University of Minnesota,
Minneapolis. It is a remote-piloted flying wing aircraft with a wing span of 3 m and a total mass of about
6.7 kg. The design closely resembles Lockheed Martin’s Body Freedom Flutter vehicle1 and NASA’s X56
MUTT aircraft.2 The mini MUTT is designed such that it exhibits strong coupling of rigid body dynamics
and structural dynamics at low airspeeds. Flutter occurs above an airspeed of 30 m/s. Without active flut-
ter suppression, the inevitable result is catastrophic structural failure as shown in the picture sequence in
Figures 1a–e.

(a) (b) (c) (d) (e)

Figure 1. Open-loop flutter and catastrophic failure during a flight test slightly above 30 m/s indicated airspeed
at the University of Minnesota on August 25th 2015.

A. Grey-Box Identification and Synthesis Model

The modeling procedure is described in detail in Refs. 19,20. The model is based on a mean-axis description21
and considers only longitudinal dynamics for straight and level flight under small elastic deformations. The
state space representation contains four states associated with rigid body dynamics, namely the forward
velocity u, angle of attack α, pitch angle θ, and pitch rate q. Additionally, the first three symmetric free vi-
bration modes are included in the model. The modes are described by their generalized displacements {ηi }3i=1

2 of 13

American Institute of Aeronautics and Astronautics

 and velocities {η̇i }3i=1 . With the state vector x = [u α θ q η1 η̇1 η2 η̇2 η3 η̇3 ]T , the state equation is
   
Xu Xα −g Xq 0 0 ··· 0 0 Xδ1 Xδ2
Z /V
 u Zα /V 0 1+Zq /V Zη1 /V Zη̇1 /V ··· Zη3 /V Zη̇3 /V 

Z /V
 δ1 Zδ2 /V 

···
   
 0 0 0 1 0 0 0 0   0 0 
   
 Mu
 Mα 0 Mq Mη 1 Mη̇1 ··· Mη 3 Mη̇3 

 M
 δ1 Mδ2  
0 0 0 0 0 1 ··· 0 0 x +  0 0  δ.
   
ẋ = 
Ξ1η1−ω12
   
 0
 Ξ1 α 0 Ξ1 q Ξ1η̇1−2ω1 ζ1 ··· Ξ1η3 Ξ1η̇3 

 Ξ1δ
 1
Ξ1δ2 
 . . . . . . . .   . .. 
 . .. .. .. .. .. .. ..  .
 .  . . 
 

   
 0 0 0 0 0 1 ··· 0 0   0 0 
0 Ξ3 α 0 Ξ3 q Ξ3η1 Ξ3η̇1 ··· Ξ3η3−ω32 Ξ3η̇3−2ω3 ζ3 Ξ3δ1 Ξ3δ2
(1)
The plant input δ = [δ1 δ2 ]T consists of symmetric midboard flap deflection δ1 and symmetric outboard
flap deflection δ2 . The entries Xi , Zi , Mi and Ξk,i for i ∈ {u, α, q, η1 , η̇1 , η3 , η̇3 , δ1 , δ2 } and k = 1, 2, 3 are
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on January 18, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-1751

dimensional aerodynamic derivatives. The entries ωk and ζk are the eigenfrequencies and damping ratios
of the k th structural mode and g denotes the gravitational acceleration. The values of the aerodynamic
coefficients are initially computed by a vortex lattice method.20 Flight data, obtained in system identification
flights at 23 m/s, were used to update the coefficients Zi , Mi , Ξ1,i for i ∈ {α, q, η1 , η̇1 , δ1 , δ2 } in Ref. 19. These
coefficients are associated with the short period dynamics and the first flexible mode. An output equation
y = C x + D δ can be added to Eq. (1) using the mode shapes of the structural modes. For further details,
the reader is referred to Refs. 19, 20 and references therein.
For the flutter suppression control design, the pitch rate and the vertical acceleration at both the center
of gravity and at the wing tips are used, i. e., y = [q az,CG az,WT ]T . The controller is assigned full authority
over the outboard flaps, i. e., δ2 = u. The midboard flaps remain reserved for exclusive use by the pilot.
Keeping the flutter suppression control loop completely separate from pilot inputs reduces the risk of satu-
rating the control surfaces and facilitates a simple control design. A schematic showing the aircraft with the
sensor and actuator positions is depicted in Figure 2. In order to simplify the synthesis model, the states u
and θ are removed by truncation and the states η2 , η̇2 , η3 , η̇3 are residualized. The resulting model thus only
consists of four states, α, q, η1 , η̇1 , and can be interpreted as a short period approximation that includes the
first aeroelastic mode. Both the short period and aeroelastic mode contain contributions from all four states,
which shows that there is no clear separation between rigid body and structural dynamics. The short period
frequency is around 25 rad/s with a damping ratio 0.8 for a flight speed of 30 m/s. The aeroelastic mode at
that airspeed has a frequency of 33 rad/s and is marginally stable. This agrees well with the observed flutter
speed of slightly above 30 m/s in flight tests.
Pitch Rate Gyro
Center Accelerometer

Wing Tip Accelerometer Wing Tip Accelerometer

Midboard Flap Midboard Flap

Outboard Flap Outboard Flap

Figure 2. Schematic of the mini MUTT aircraft.

B. Time Delay and Phase Loss Modeling

The goal of this subsection is to describe and model all known parasitic dynamics. For regular flight control
systems, the sampling rate is much higher than the closed-loop bandwidth and the induced phase loss from
sensors and actuators is usually negligible. On the contrary, active suppression of the flutter instability at
high frequency requires a very high closed-loop bandwidth. Actuator and sensor dynamics are not negligible

3 of 13

American Institute of Aeronautics and Astronautics

 in this frequency regime. Time delay, introduced by digitalization effects and computation, also has a big
impact on the control loop. The question of whether the design should be carried out in discrete time rather
than continuous time naturally arises in this context. A discrete time design based on “exact discretization”
would automatically incorporate the delay due to the zero-order hold operation. This delay would, however,
only be a small part of the overall delay. The remaining part would still require a model. While the advantages
of a discrete-time design thus seem to be limited, insight into the problem would be lost to a certain extent. It
therefore appears preferable to stay in the continuous-time domain and to include the delay and digitalization
effects as part of the model. The controller can be implemented in discrete time by Tustin approximation.
Pre-warping can further be used to enhance accuracy in the critical frequency range.
Figure 3 shows all components in the feedback loop and how they are grouped into three models Gsens ,
Gdelay , and Gact . Including these dynamics in the synthesis model allows the controller to compensate for
known phase loss and hence to improve performance and robustness. The pitch rate measurement on the
mini MUTT aircraft is obtained by an inertial measurement unit (IMU) that includes a 50 Hz low-pass filter.
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on January 18, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-1751

The accelerometer signals are filtered by an analog first order low-pass with a bandwidth of 35 Hz. These
components are modeled by two transfer functions
2 π 35 2 π 50
Gaccel (s) = and GIMU (s) = . (2)
s + 2 π 35 s + 2 π 50
The signals provided by the sensors are processed by the mini MUTT’s flight computer that executes the
control algorithm within a 6.6 ms frame. The controller output is passed on to a microcontroller that runs
asynchronous with a 3.3 ms frame rate to generate a pulse width modulation (PWM) signal. This PWM
signal is the input to a servo controller that runs, also asynchronous, with a 3.3 ms frame. This results in
13.2 ms total computational delay. The actuator used on the mini MUTT is a Futaba S9254 servo. Its physical
inertia introduces additional low-pass characteristics. A second-order model
96710
Gact (s) = (3)
s2 + 840 s + 96710
is constructed via frequency-domain identification techniques using a chirp input signal. Validation is per-
formed in the frequency domain using a second set of data with an input chirp at a higher voltage and in
the time domain via step response data.

Pilot Input Servo Controller Midboard

Accelerometers
and Actuator Flaps
Airframe
Outboard

Wing tip az
Center az
Actuator IMU
Flaps

Gact (s) Gsens (s)

q

PWM Control
Servo Controller Microcontroller Flight Computer
Signal Signal

Gdelay (s)

Figure 3. Modeling of components involved in the feedback loop for flutter suppression on the mini MUTT.

In H∞ control, every state in the synthesis model directly results in a controller state. To keep the
controller order low, it is necessary to combine actuator dynamics, sensor dynamics, and the delay in a
low-order equivalent model. Obtaining this model requires a shift of the sensor dynamics from the plant
output to the input, which is only possible if all sensors are modeled identically. The slower dynamics
of the accelerometers are therefore also assumed for the faster IMU and both are uniformly modeled as
Gsens (s) = Gaccel (s). Further, all computational frames are added up and a factor of 1.5 is included in order
to anticipate the zero-order hold delay. To further account for actuator and sensor delays, a total delay of

4 of 13

American Institute of Aeronautics and Astronautics

 25 ms is assumed and modeled as Gdelay (s) = e−0.025 s . A second-order model is calculated from balancing
and residualization22 of Gact (s) Gdelay (s) Gsens (s), where a fifth-order Pade approximation is used for the
time delay. The resulting model is
0.966s2 − 86.33s + 5539
Gequiv (s) = . (4)
s2 + 117.6s + 5539
It captures the phase loss very accurately up to about 100 rad/s, see Figure 4a. Figure 4b further illustrates the
phase loss contributions of the known parasitic dynamics in the critical frequency range in detail. The largest
contribution comes from the time delay, followed by the actuator and sensors. The resulting simplified loop
is depicted in Figure 5. The model Gy is used for the control design and synthesis, described in Section III.
It combines the fourth-order airframe model and the second-order equivalent model for actuator dynamics,
delay, and sensor dynamics and hence has six states.
0
Magnitude (dB)

0
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on January 18, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-1751

−3

−6 47◦
−9

Phase (◦ )
−45
0
16◦
Phase (◦ )

Figure 2b
−180 9◦

−360 −90
10−1 100 101 102 10 20 30 40 50
Frequency (rad/s) Frequency (rad/s)
(a) Equivalent phase loss model for inclusion in the syn- (b) Estimated phase loss at 33 rad/s, the frequency of the aeroe-
thesis model. lastic mode.

Figure 4. Phase loss due to known parasitic dynamics: pure time delay ( ), plus actuator dynamics ( ),
plus sensor dynamics ( ), second-order approximation for synthesis model ( ).

Gy (s)

Outboard
Gequiv (s) Airframe
Wing tip accel. az,WT

Flaps
Center accel. az,CG
Pitch rate q

u
Controller

K(s)
Figure 5. Low-order equivalent modeling of parasitic components involved in the feedback loop.

III. Control Law Design

When flutter was observed at 30 m/s airspeed in a flight test, the aircraft was already running on full
throttle. An envelope expansion beyond the flutter speed thus is also limited by the propulsion system.

5 of 13

American Institute of Aeronautics and Astronautics

 Noting that the flutter dissipated a certain amount of energy, it appears possible to fly at 33 m/s once the
oscillations are controlled. The main objective for the control design is consequently to stabilize flight at
33 m/s and to provide enough safety margin to maintain stability at higher velocities that might occur due
to head wind gusts and unintended dive maneuvers. The model described in Section II is thus scaled to
an airspeed of V = 33 m/s, the desired flight point in the expanded envelope. For that flight condition, an
H∞ -norm optimal controller is designed in this section after briefly summarizing the necessary background
for H∞ control.

A. H∞ Closed-Loop Shaping
The H∞ -norm, or induced L2 -norm, of a linear time invariant (LTI) dynamic system G(s) from input d to
output e is defined as
kek2
kG(s)k = sup σ̄(G(jω)) = sup , (5)
kdk
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on January 18, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-1751

ω d∈L2 \{0} 2

where σ̄(·) denotes the largest singular value. This norm measures the maximum gain of the transfer function
G(s), i. e., the largest amplification of L2 input signals over all frequencies and input/output directions. It can
be used to specify performance for a feedback interconnection in terms of a generalized plant P . A dynamic
controller K can be synthesized by solving two Riccati equations.23, 24 The controller stabilizes the closed-loop
interconnection given by the lower fractional transformation FL (P, K) and achieves a performance index γ
that provides an upper bound on the H∞ -norm of the closed loop, i. e., kFL (P, K) k < γ. With synthesis
machinery readily available, e. g., in the Matlab Robust Control Toolbox,25 the challenging part of any H∞
design is to provide meaningful performance specifications.
The high-level objective of the flutter suppression controller is to attenuate the aeroelastic mode without
impairing handling of the aircraft by the pilot. The controller further needs to provide robustness against
a wide class of possible uncertainties in the model. The proposed generalized plant interconnection, that
translates these goals into the objective of minimizing a closed-loop H∞ -norm, is depicted in Figure 6.

Magnitude (dB)
d2 e1 d1 e3 e2 20

W2 Wu (s) W1 Wz Wy 10

z
u 0
K(s) G(s)
− y 100 101 102 103
Frequency (rad/s)
(a) Generalized plant. (b) Weighting filter Wu (s).

Figure 6. Generalized plant interconnection for the flutter suppression control design.

h i
G (s)
The plant model is partitioned as G(s) = Gyz (s) , where Gy (s) includes the airframe model and the
combined actuator dynamics, sensor dynamics, and delay as described in Section II.B. The measurable
output y, used for feedback, thus consists of pitch rate (in rad/s), vertical center acceleration, and vertical
wing tip acceleration (both in m/s2 ). The generalized velocity, η̇1 , of the first structural mode is added as
an additional, non-measurable performance output z to the plant model. This results in a transfer function
Gz (s) with a band-pass characteristic and a sharp peak at the flutter frequency. The plant input is the
symmetric deflection of the outboard flaps (in rad). Disturbances are modeled both at the plant’s input
and outputs by exogenous signals d1 and d2 that are weighted by W1 and W2 . The three outputs of the
interconnection, e1 to e3 , are weighted versions of the control signal u, the measurable output y, and the
performance output z. These five signals define an input-output map e = FL (P, K) d that can be represented
in terms of six transfer functions as
    
e1 Wu −Ti Si K " #" #
 W1 d1
e2  =  Wy   Gy Si To . (6)
   

W 2 d2
e3 Wz Gz Si Gz Si K

6 of 13

American Institute of Aeronautics and Astronautics

 The transfer functions To = Gy K (I +Gy K)−1 and Ti = K Gy (I +K Gy )−1 are called output complementary
sensitivity and input complementary sensitivity. The transfer function Si = (I + K Gy )−1 is called input
sensitivity. Further, Si K and Gy Si are known as control sensitivity and disturbance sensitivity. The remaining
two transfer functions, Gz Si and Gz Si K, relate the disturbances d1 and d2 to the generalized velocity η̇1 of
the aeroelastic mode. Decreasing their gain thus corresponds to attenuating the aeroelastic mode in response
to disturbances.
The desired flutter margin is set by the weight Wz . Larger values result in higher damping augmentation,
since the weight encourages the controller to reduce the sharp peak in the frequency response, cf. Refs. 26,27.
The weight Wu is used to limit control action to a specific frequency range. The main goal is to avoid unde-
sired interaction with rigid body dynamics in the low-frequency regime and with unmodeled high-frequency
dynamics. Selecting Wu as a band-stop filter, as shown in Figure 6b, results in band-pass behavior for both
the input complementary sensitivity and the control sensitivity. Thus, both for low and high frequencies
Ti ≈ 0. The key observation is that this implies Si ≈ I and consequently Si K ≈ K. The band-stop weight
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on January 18, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-1751

Wu thus directly shapes K at low and high frequencies and imposes both a wash-out and a roll-off on
the controller. The complementary sensitivities are related to robustness against multiplicative uncertainty
at the plant output and input, respectively. When all weights are removed and the loop is closed with a
norm-bound, stable LTI dynamic uncertainty ∆1 ∈ RH∞ such that d1 = ∆1 e1 , the loop remains stable
according to the small-gain theorem28 as long as k∆1 k < 1/kTi k. The same is true when the loop is closed
with d2 = ∆2 e2 and k∆2 k < 1/kT k. Similarly, the control sensitivity can also be interpreted to guard against
an additive uncertainty d2 = ∆additive e1 with k∆additive k < 1/kSi Kk, and the disturbance sensitivity as to
account for inverse uncertainty d1 = ∆inverse e2 with k∆inverse k < 1/kGy Si k. The weights Wy , W1 , and W2
are used to adjust the relative importance of all involved transfer functions. For more details, the reader is
referred to standard robust control textbooks, e. g., Refs. 22, 28.

B. Design and Tuning

The weights for the mixed sensitivity formulation (6) are selected as W1 = 1, W2 = diag(11, 150, 200),
Wz = 0.001, and Wy = 0.0001 diag(3, 3, 6). The numbers are the result of tuning but can be seen to essentially
normalize all individual transfer functions to a maximum gain of around 0 dB. The weight for the control
effort is selected as the interconnection of a low-pass filter with DC-gain 200, crossover frequency 25 rad/s
and feedthrough gain 0.5 in series with a high-pass filter with DC-gain 0.5, 55 rad/s crossover frequency
2
and feedthrough gain 200. The resulting band-stop filter Wu = 100s2s+12700+7506 s+137500
s+1375 , shown in Figure 6b, thus
restricts activity of the flutter suppression controller to the frequency region of the aeroelastic mode. The
tuning procedure is simple and intuitive, once initial values are selected to normalize all involved transfer
functions. The desired increase in damping, and hence the flutter margin, is set by the weight Wz . To
increase robustness margins of the closed-loop (see Section IV.A), Wu (s) is increased in the frequency region
where the margin is attained. This consequently decreases the controller gain at that frequency. The relation
between input and output margins is handled by the weights W1 and W2 .
To guide the tuning procedure, Figures 7–9 are used as indicators for nominal controller performance.
Figure 7a shows the open-loop and closed-loop transfer function Gz Si used to specify damping augmentation.
The disturbance sensitivity Gy Si that relates inputs to the measurable outputs is shown in Figure 7b. The
sensitivity is in both cases lowered at the frequency of the aeroelastic mode, but as a consequence increased at
neighboring frequencies. This is an inevitable consequence of Bode’s sensitivity integrals.22, 29 One important
Singular Value (dB)

70
Magnitude (dB)

40
60
20
50
0 40
−20 30 −1
10−1 100 101 102 10 100 101 102
Frequency (rad/s) Frequency (rad/s)
(a) Sensitivity of aeroelastic mode (Gz Si ). (b) Disturbance sensitivity (Gy Si ).

Figure 7. Open-loop ( ) and closed-loop ( ) transfer functions.

7 of 13

American Institute of Aeronautics and Astronautics

 aspect of the control design is to confine this sensitivity degradation to a specific frequency region. Figure 7
shows that this is indeed achieved and that neither the low frequency phugoid nor the high frequency elastic
modes are affected by the flutter suppression controller.
In order to further assess the interaction with pilot commands, a comparison of open-loop and closed-loop
step responses to midboard flap deflection is shown in Figure 8 for two different airspeeds. These flaps are used
by the pilot to control the longitudinal motion of the aircraft, see Figure 3. The pilot essentially closes a pitch
angle feedback loop, since his main visual indicator for control is the vehicles attitude. Maintaining a pitch
response as close as possible to the open-loop aircraft is thus considered desirable. The pitch response for the
low airspeed of 24 m/s is barely altered by the presence of the flutter suppression controller, see Figure 8. At
the naturally unstable airspeed of 33 m/s, the highly oscillatory and divergent pitch rate response is effectively
damped out and stabilized. This is achieved without affecting the initial transients up to about 0.15 s. The
aircraft’s immediate response to pilot inputs is thus identical with and without flutter suppression, both for
low and high airspeeds. The flutter suppression controller introduces no additional delay or phase lag, that
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on January 18, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-1751

could impair handling. The effectiveness of the controller is further visible in the acceleration responses in
Figure 8b.
6

Center az (m/s2 )
Pitch angle (◦ )

0
4
−1
2
−2
0
−3
10 Wing tip az (m/s2 ) 6
Pitch rate (◦ /s)

0 4

2
−10
0
−20
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
Time (s) Time (s)
(a) Pitch response. (b) Acceleration response.

Figure 8. Open-loop responses at 24 m/s ( ) and 33 m/s ( ), and closed-loop responses at 24 m/s ( )
and 33 m/s ( ) to step input at midboard flaps.

The effect of the flutter suppression controller on the pole locations is shown in Figure 9. The open-loop
model exhibits flutter at airspeeds above 30 m/s, indicated by the poles of the aeroelastic mode crossing
into the right half plane at about 33 rad/s in Figure 9a. Figure 9b shows that the locus of the aeroelastic
mode is altered by the controller to stay within the left half plane with a drastic improvement in damp-
ing. Extrapolation of the model to higher airspeeds further shows that flutter now occurs at 40 m/s. This
corresponds to an envelope expansion of 10 m/s (33 %) and is deemed a more than sufficient safety margin
for the desired flight point at 33 m/s. A noticeable side effect of the flutter suppression controller is related
to the short period poles. While their damping ratio is only marginally affected, their frequency is lowered
for airspeeds below the design point and increased for airspeeds above the design point. Judging from the
time-domain responses in Figure 8, the effect is not expected to cause handling quality degradation. With
no significant effect on short period damping and the pilot controlling the aircraft almost entirely through
its phugoid mode, the effect is of no concern.
The resulting controller K(s) is shown in Figure 10. The desired band-pass behavior is apparent. The
controller has eight states. Its fastest pole is at 106 rad/s and thus well within the permissible region for
digital implementation on the flight computer. The peak gain for both center acceleration and wing tip
acceleration signals is attained at the same frequency around 40 rad/s, but their phase differs considerably.
The wing tip acceleration lags the center acceleration by up to 40◦ . This shows that the proposed controller
would be impossible to obtain by a simple combination of the acceleration signals in a single loop.

8 of 13

American Institute of Aeronautics and Astronautics

 40 40

Aeroelastic Mode Aeroelastic Mode

30 30
Flutter above 30 m/s airspeed Envelope expansion
Im(s) (rad/s)

Im(s) (rad/s)
to 40 m/s airspeed
20 20
Short Period Short Period

10 10
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on January 18, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-1751

0 0
−40 −30 −20 −10 0 −40 −30 −20 −10 0
Re(s) (rad/s) Re(s) (rad/s)
(a) Open-loop pole locations at 24 m/s ( ), 27 m/s ( ), (b) Closed-loop pole locations at 24 m/s ( ), 27 m/s ( ),
30 m/s ( ), and 33 m/s ( ). 30 m/s ( ), and 33 m/s ( ).

Figure 9. Effect of the flutter suppression controller on pole locations.

Pitch rate (rad/s) Center acceleration (m/s2 ) Wing tip acceleration (m/s2 )
Magnitude (dB)

−40

−60

−80

90
0
Phase (◦ )

−90
−180
−270
−360
100 101 102 100 101 102 100 101 102
Frequency (rad/s) Frequency (rad/s) Frequency (rad/s)

Figure 10. Bode plot of the flutter suppression controller.

IV. Controller Robustness Evaluation

Given the catastrophic consequences of flutter, it is paramount that the controller is highly robust.
Without a high-fidelity nonlinear model for evaluation and with limited possibilities for testing outside
of the critical flight regime, a thorough linear analysis is required. The robustness tests described in this
section aim at maximizing the likelihood of a successful flight with the developed controller. Disk margins,
both for single and multivariable loops, are considered. These margins measure robustness with respect to
simultaneous phase and gain variations and hence avoid the pitfalls of classical gain and phase margins.
Further, structured singular values are used to evaluate robustness with respect to parametric uncertainties
in the aircraft model. Specifically, uncertainty in the structural model, the aerodynamics model, and the
actuator model is considered. All robustness calculations are performed on a model that includes sensor
dynamics, actuator dynamics, delay, the first three structural modes and complete rigid body dynamics. The
analysis results of this section are thus to be understood “on top” of all known parasitic dynamics.

9 of 13

American Institute of Aeronautics and Astronautics

 A. Robustness Margins
The most common metric to quantify robustness for a control system is given by the classical gain and phase
margins. The former specifies how much gain variation a single loop-transfer function can tolerate before
instability occurs. The second measures the amount of phase loss that this loop can tolerate. Both margins
are independent of each other and common practice in aerospace usually requires at least 6 dB gain margin
and 45◦ phase margin. These numbers are, however, derived from experience with rigid aircraft and certain
highly structured control architectures. They are thus not necessarily sufficient for the problem at hand.
The flutter suppression controller in Ref. 7, for instance, was designed with 60◦ phase margins, while the
design in Ref. 8 only required 30◦ . Considering only classical margins can also easily overlook destabilizing
combinations of gain and phase that independently are considered safe. It is therefore important to take into
account simultaneous gain and phase variations. The corresponding metric is known as disk margin and can
be calculated from kSo − To k and kSi − Ti k.30 Disk margins provide a higher level of robustness compared
to classical margins. They are also easily extended to the multivariable case, allowing for simultaneous
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on January 18, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-1751

perturbation of several loops. A single-loop input-output disk margin is obtained by breaking the loop at
both the input and at one output at the same time. This margin considers simultaneous perturbations at
the input and output of a single feedback loop, with all other loops closed. It is regarded as useful for the
present design because independent sensor uncertainties for every channel appear overly conservative, given
the same sensor type and data acquisition system for the accelerometers. A simultaneous input and output
uncertainty, on the other hand, is inevitably present.
The design requirements are selected as minimum single-loop disk margins of at least 8 dB (45◦ ) and
minimum single-loop input/output disk margins of at least 6 dB (37◦ ). These requirements are indicated
in Figure 11 as horizontal lines. Further, a minimum single-loop delay margin of 19.8 ms is required, cor-
responding to one dropped frame from every computational unit and the induced zero-order hold delay.
The robustness margins are depicted in Figure 11 as a function of airspeed. All margins uniformly increase
with lower airspeed to a similar extent. This indicates a smooth variation without any particular robust-
ness bottlenecks. The input disk margin is above 8 dB (45◦ ) and single-loop output disk margins are all well
above 11 dB (60◦ ). The single-loop input/output disk margins also satisfy the requirement of 6 dB (37◦ ). The
multi-loop output disk margin, corresponding to simultaneous perturbation of all outputs, is also calculated
and remains above 6 dB (37◦ ). If independent perturbations of all outputs and the input are considered, the
margin is known as multi-loop input/output disk margin. It remains above 3.5 dB (23◦ ), which is considered
an acceptable level of degradation with respect to the single-loop margins. The delay margins at the outputs
are infinite for airspeeds below 33 m/s and the lowest margin is 47 ms at 33 m/s. The lowest delay margin at
the input is 22 ms and also attained for 33 m/s.
90

18 75
Gain Margin (dB)

Phase Margin (◦ )

60

12
45
9
30
6

15
3

0
24 27 30 33 24 27 30 33
Airspeed (m/s) Airspeed (m/s)

Figure 11. Minimum robustness margins as a function of airspeed: single-loop disk margin at input ( )
and output ( ), single-loop input-output disk margin ( ), multi-loop output disk margin ( ), and
multi-loop input-output disk margin ( ).

10 of 13

American Institute of Aeronautics and Astronautics

 B. Structured Singular Value Analysis
The margin analysis of Section IV.A aimed at capturing generic model uncertainty. In this subsection, the
analysis is narrowed to specific sources of uncertainty within the model structure. The models for both struc-
tural dynamics and aerodynamics are best described as uncertain with respect to real parameters. Structured
singular value analysis provides an efficient way to calculate stability margins for such structured uncertain-
ties, see Refs. 31, 32. Three different sets of uncertainties are considered in this subsection. Structural model
uncertainty in the following refers to a real parametric uncertainty in the eigenfrequency ω1 of the first struc-
tural mode. The parameter ω1 in Eq. (1) is hence replaced by (1 + ∆) ω1 , where ∆ ∈ R with nominal value
zero spans the range of possible variation, e. g., |∆| < 0.1 for 10 % uncertainty. Likewise, aerodynamic uncer-
tainty refers to real perturbations in the aerodynamic coefficients for pitch moment (Mα , Mq ), lift (Zα , Zq ),
and influence of the first structural mode (Ξ1α , Ξ1q , Ξ1δ2 ) in Eq. (1). All real parametric uncertainties
are “complexified” with a 5 % dynamic uncertainty to regularize the resulting computational problem, see
Refs. 25, 33 for details. Actuator uncertainty refers to a norm-bound complex multiplicative uncertainty in
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on January 18, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-1751

the actuator model, i. e., Gact is replaced by (1 + ∆) Gact , where ∆ ∈ RH∞ is a norm-bound, stable LTI
dynamic uncertainty with nominal value zero that represents the range of variation, e. g., k∆k < 0.1 for 10 %
uncertainty. Figure 12a shows the stability boundaries for parameter variations along with a robust perfor-
mance analysis. The performance index is calculated as the ratio kGz Si k/kGz k of the worst-case H∞ -norm.
It thus measures the amount of damping augmentation that is provided by the flutter suppression controller.

4 4
unstable

unstable

unstable
kGz Si k / kGz k

unstable
kGz Si k / kGz k
3 3
2 2
1 1

0 10 20 30 40 50 60 0.5% struct 1.5% struct 2.5% struct

5% aero 15% aero 25% aero
2.5% act 7.5% act 12.5% act

Uncertainty (%) Uncertainty

(a) Robust performance analysis for three sets of structured (b) Robust performance analysis with a large uncer-
uncertainties. tainty set.

Figure 12. Robust stability and performance analysis for parametric uncertainties in the structural
model ( ), aerodynamic model ( ), actuator model ( ), and a combination of these ( ).

Instability occurs first for uncertainty in the structural mode frequency. This frequency is obtained from
ground vibration tests for the present model in Ref. 34 and expected to be known very accurately, up to
about 2 %. Thus, the stability margin of over 10 % is more than sufficient. The highest uncertainty is expected
in the aerodynamics model. The analysis shows that the controller is highly robust with respect to this
uncertainty, tolerating up to 40 % perturbations. The permissible actuator uncertainty is even higher and is
above 60 %. The performance degradation for all three cases is qualitatively similar and can be characterized
as graceful. Small variations result in small performance degradation, that only start to increase significantly
close to the stability boundary. For individual uncertainties below 7 % in the structural model, 25 % in the
aerodynamics model, and 48 % in the actuator model, the ratio of closed-loop and open-loop gain is less
than one. In these cases, the controller provides additional damping to the aeroelastic mode and hence
achieves robust performance. A fourth analysis is shown in Figure 12b for an uncertainty set that combines
all aforementioned uncertainties. Even in this case, performance degradation is smooth and graceful. The
stability margin is considerably lower than for the individual uncertainties but still encouraging. Stability
is certified up to simultaneous 2.5 % structural mode uncertainty, 25 % aerodynamic uncertainty and 12.5 %
actuator uncertainty. Robust performance is achieved up to simultaneous 1.5 % structural mode uncertainty,
15 % aerodynamic uncertainty and 7.5 % actuator uncertainty.

C. Rate Limits and Saturation

Since the system is open-loop unstable, saturation and rate limits must be strictly avoided. This is the main
reason for keeping the flutter suppression loop and the control surfaces it uses completely separate from pilot
inputs. Doing so prevents the pilot from saturating (and hence disabling) the flutter suppression system with

11 of 13

American Institute of Aeronautics and Astronautics

 his or her control inputs. Still, the response of the control system alone must be verified to remain inside
the allowed boundaries. Usually, nonlinear simulation is used to assess the likelihood of hitting either rate or
deflection limits. Since no such model is available for the present aircraft, a singular value analysis of the linear
closed-loop model is performed instead. This analysis does not establish any time-domain guarantees, but
provides an estimate of the controller response to worst-case inputs. The maximum deflection of the outboard
flaps, used by the controller, is δmax = 35◦ . Hardware tests further led to an estimate of 400–1200◦ /s for the
rate limits of the actuators. For the following analysis, the more restrictive bound δ̇max = 400 ◦ /s is used.
The model is shown in Figure 13. The pilot has direct control over the midboard surfaces, with a maximum
deflection of 35◦ . Gusts are assumed to directly alter the angle of attack by up to 5◦ . Both pilot and gusts
are modeled as first order low-pass filters with 5 rad/s bandwidth and a steady-state gain corresponding to
their maximum input magnitude. Modeling gusts in accordance with the Dryden spectrum would result in a
similar filter with around half the bandwidth and one third of the gain. The employed model thus provides
an additional safety margin. The singular values for both the control signal and its rate are depicted in
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on January 18, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-1751

Figure 13b, normalized by their respective limits. The singular values measure the amplification of a worst-
case combination of gust and pilot inputs. A singular value of 0 dB represents the maximum permissible
deflection and rate, respectively. The deflection can be seen to max out around −15 dB, indicating that
less than about 20 % of the available deflection will actually be used by the controller. The rate increases
significantly around the frequency of the flutter frequency, but never exceeds −5 dB. This analysis is very
conservative because it largely overestimates the influence of gusts and at the same time considers the worst
possible combination with pilot inputs. Still, both the deflection and rate limits are satisfied. This indicates
that neither should be of concern for the flutter suppression controller.

Singular Values (dB)

Midboard 0
Pilot Model
Flaps
Angle of −20
Gust Model Attack Aircraft Model
−40
−60
Outboard
Flaps −80
100 101 102
K
Frequency (rad/s)
(a) Deflection and rate limit verification model. (b) Normalized singular values for deflection
(K Si /δmax ) and rate (s K Si /δ̇max ).

Figure 13. Saturation analysis by means of singular values.

V. Conclusions
The present paper developed a systematic multivariable robust control design for a small, unmanned
flexible aircraft. The controller requires a large bandwidth in order to stabilize the aircraft. Consequently,
all known parasitic dynamics are included in the synthesis model in order to anticipate phase loss. Since
the model is highly uncertain, special emphasize is put on a design that is robust with respect to a wide
variety of uncertainties. Linear analyses are performed to demonstrate both the high level of robustness and
the absence of adverse interaction with low-frequency rigid body dynamics and high-frequency structural
dynamics beyond the targeted aeroelastic mode. Validation of the flutter suppression controller in flight tests
is planned for the spring of 2016. These flight tests will also provide a comparison to other control approaches
and help to establish a benchmark.

Acknowledgement
The authors thank Dale Enns for his insights and helpful discussion. This work was supported by NASA
NRA No. NNX14AL36A entitled Lightweight Adaptive Aeroelastic Wing for Enhanced Performance Across
the Flight Envelope. Mr. John Bosworth is the technical monitor and Mr. Dan Moerder is the acting Technical
Monitor.

12 of 13

American Institute of Aeronautics and Astronautics

 References
1 Beranek, J., Nicolai, L., Buonanno, M., Burnett, E., Atkinson, C., Holm-Hansen, B., and Flick, P., “Conceptual Design

of a Multi-Utility Aeroelastic Demonstrator,” 13th AIAA/ISSMO Multidisciplinary Analysis Optimization Conference, 2010,
pp. 2194–2208.
2 Ryan, J. J., Bosworth, J. T., Burken, J. J., and Suh, P. M., “Current and Future Research in Active Control of Lightweight,

Flexible Structures Using the X-56 Aircraft,” AIAA SciTech, 2014, doi:10.2514/6.2014-0597.
3 Preumont, A., Vibration Control Of Active Structures, Kluwer Academic Publishers, New York, 2nd ed., 2002.
4 Wykes, J. H., “Structural Dynamic Stability Augmentation and Gust Alleviation of Flexible Aircraft,” AIAA 5th Annual

Meeting and Technical Display, 1968.

5 Wykes, J. H., Borland, C. J., Klepl, M. J., and MacMiller, C. J., “Design and Development of a Strutural Mode Control

System,” Contractor Report 143846, NASA, 1977.

6 Wykes, J. H., Byar, T. R., MaeMiller, C. J., and Greek, D. C., “Analyses and Tests of the B-1 Aircraft Structural Mode

Control System,” Contractor Report 144887, NASA, 1980.

7 Roger, K. L., Hodges, G. E., and Felt, L., “Active Flutter Suppression—A Flight Test Demonstration,” Journal of

Aircraft, Vol. 12, No. 6, 1975, pp. 551–556.

Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on January 18, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2016-1751

8 Adams, W. M., Christhilf, D. M., Waszak, M. R., Mukhopadhyay, V., and Srinathkumar, S., “Design, Test, and Evaluation

of Three Active Flutter Suppression Controllers,” Technical Memorandum 4338, NASA, 1992.
9 Waszak, M. R. and Srinathkumar, S., “Flutter Suppression for the Active Flexible Wing: A Classical Design,” Journal

of Aircraft, Vol. 32, No. 1, 1995, pp. 61–67.

10 Mukhopadhyay, V., “Flutter Suppression Control Law Design And Testing For The Active Flexible Wing,” Journal of

Aircraft, Vol. 32, No. 1, 1995, pp. 45–51.

11 Mukhopadhyay, V., “Transonic Flutter Suppression Control Law Design and Wind-Tunnel Test Results,” Journal of

Guidance, Control, and Dynamics, Vol. 23, No. 5, 2000, pp. 930–937.
12 Holm-Hansen, B., Atkinson, C., Benarek, J., Burnett, E., Nicolai, L., and Youssef, H., “Envelope Expansion of a Flexible

Flying Wing by Active Flutter Suppression,” Proceedings of the Association for Unmanned Vehicle Systems International,
2010.
13 Hjartarson, A., Seiler, P. J., and Balas, G. J., “LPV Aeroservoelastic Control using the LPVTools Toolbox,” AIAA

Atmospheric Flight Mechanics Conference, 2013, doi:10.2514/6.2013-4742.

14 Barker, J. M., Balas, G. J., and Blue, P. A., “Gain-scheduled Linear Fractional Control for Active Flutter Suppression,”

Journal of Guidance, Control, and Dynamics, Vol. 22, No. 4, 1999, pp. 507–512.
15 Barker, J. M. and Balas, G. J., “Comparing Linear Parameter-Varying Gain-Scheduled Control Techniques For Active

Flutter Suppression,” Journal of Guidance, Control, and Dynamics, Vol. 23, No. 5, 2000, pp. 948–955.
16 Waszak, M. R., “Robust Multivariable Flutter Suppression for Benchmark Active Control Technology Wind-Tunnel

Model,” Journal of Guidance, Control, and Dynamics, Vol. 24, No. 1, 2001.
17 Danowsky, B. P., Thompson, P. M., Lee, D., and Brenner, M., “Modal Isolation and Damping for Adaptive Aeroservoe-

lastic Suppression,” AIAA Atmospheric Flight Mechanics Conference, 2013, doi:10.2514/6.2013-4743.

18 Schmidt, D. K., “Stability Augmentation And Active Flutter Suppression Of A Flexible Flying-Wing Drone,” submitted

to Journal of Guidance, Control, and Dynamics, 2015.

19 Pfifer, H. and Danowsky, B., “System Identification of a Small Flexible Aircraft,” AIAA SciTech, 2016.
20 Schmidt, D. K., Zhao, W., and Kapania, R. K., “Flight-Dynamics and Flutter Modeling and Analysis of a Flexible

Flying-Wing Drone,” AIAA SciTech, 2016.

21 Waszak, M. R. and Schmidt, D. K., “Flight dynamics of aeroelastic vehicles,” Journal of Aircraft, Vol. 25, No. 6, 1988,

pp. 563–571.
22 Skogestad, S. and Postlethwaite, I., Multivariable Feedback Control, Prentice Hall, Upper Saddle River, NJ, 2nd ed.,

2005.
23 Glover, K. and Doyle, J. C., “State-space Formulae for All Stabilizing Controllers that Satisfy an H -norm Bound and
∞
Relations to Risk Sensitivity,” Systems & Control Letters, Vol. 11, No. 3, 1988, pp. 167–172.
24 Doyle, J. C., Glover, K., Khargonekar, P. P., and Francis, B. A., “State-space Solutions to Standard H and H
2 ∞ Control
Problems,” IEEE Transactions on Automatic Control, Vol. 34,8, 1989, pp. 831–847.
25 Balas, G., Chiang, R., Packard, A., and Safonov, M., Robust Control Toolbox User’s Guide R2014b, MathWorks, 2014.
26 Hanel, M., Robust Integrated Flight and Aeroelastic Control System Design for a Large Transport Aircraft, Ph.D. thesis,

University Stuttgart, Stuttgart, Germany, 2001.

27 Theis, J., Pfifer, H., Balas, G., and Werner, H., “Integrated Flight Control Design for a Large Flexible Aircraft,” American

Control Conference, 2015, pp. 3830–3835, doi:10.1109/ACC.2015.7171927.

28 Zhou, K., Doyle, J. C., and Glover, K., Robust and Optimal Control, Prentice Hall, Upper Saddle River, NJ, 1995.
29 Stein, G., “Respect the Unstable,” IEEE Control Systems Magazine, August 2003, pp. 12–25.
30 Blight, J. D., Lane Dailey, R., and Gangsaas, D., “Practical control law design for aircraft using multivariable techniques,”

International Journal of Control, Vol. 59, No. 1, 1994, pp. 93–137.

31 Doyle, J., “Analysis of Feedback Systems with Structured Uncertainties,” IEE Proceedings D (Control Theory and

Applications), Vol. 129, IET, 1982, pp. 242–250.

32 Packard, A. and Doyle, J., “The Complex Structured Singular Value,” Automatica, Vol. 29, No. 1, 1993, pp. 71–109.
33 Packard, A. and Pandey, P., “Continuity Properties of the Real/Complex Structured Singular Value,” IEEE Transactions

on Automatic Control, Vol. 38, No. 3, 1993, pp. 415–428.

34 Gupta, A. and Seiler, P., “Ground Vibration Test on Flexible Flying Wing Aircraft,” AIAA SciTech, 2016.

13 of 13

American Institute of Aeronautics and Astronautics