integrated flight/structural mode control for very ...€¦ · flexible aircraft configuration that...
TRANSCRIPT
American Institute of Aeronautics and Astronautics
1
Integrated Flight/Structural Mode Control for Very Flexible Aircraft Using L1 Adaptive Output Feedback
Controller
Jiaxing Che*, University of Connecticut, Storrs, CT 06269
Irene M. Gregory† NASA Langley Research Center, Hampton, VA 23681
Chengyu Cao ‡ University of Connecticut, Storrs, CT 06269
This paper explores application of adaptive control architecture to a light, high-aspect ratio, flexible aircraft configuration that exhibits strong rigid body/flexible mode coupling. Specifically, an 1L adaptive output feedback controller is developed for a semi-span wind
tunnel model capable of motion. The wind tunnel mount allows the semi-span model to translate vertically and pitch at the wing root, resulting in better simulation of an aircraft’s rigid body motion. The control objective is to design a pitch control with altitude hold while suppressing body freedom flutter. The controller is an output feedback nominal controller (LQG) augmented by an 1L adaptive loop. A modification to the 1L output feedback is
proposed to make it more suitable for flexible structures. The new control law relaxes the required bounds on the unmatched uncertainty and allows dependence on the state as well as time, i.e. a more general unmatched nonlinearity. The paper presents controller development and simulated performance responses. Simulation is conducted by using full state flexible wing models derived from test data at 10 different dynamic pressure conditions. An 1L adaptive output feedback controller is designed for a single test point and
is then applied to all the test cases. The simulation results show that the 1L augmented
controller can stabilize and meet the performance requirements for all 10 test conditions ranging from 30 psf to 130 psf dynamic pressure.
I. Introduction he next generation of efficient subsonic aircraft will need to be both high aspect ratio and light weight, and the
associated structural flexibility presents both challenges and opportunities. An example of advanced vehicles under consideration is illustrated in Fig. 11. Robust active aeroelastic control and Gust Load Alleviation (GLA) have a potential to substantially reduce weight by relaxing structural strength requirements on the wings. However, current tools such as traditional linear aeroelastic and aeroservoelastic analysis methods are inadequate to reliably predict aeroelastic stability and assess active aeroelastic control effectiveness in this type of vehicle. The type of vehicle represents a nonlinear stability and control problem involving complex interactions among the flexible structure, unsteady aerodynamics, flight control system, propulsion system, the environmental conditions, and vehicle flight dynamics. Furthermore, because of the inherent flexibility of the aircraft, the lower order structural mode frequencies are of the same order as the rigid-body mode frequencies. The close proximity of flexible and rigid-body * Ph.D Student, Department of Mechanical Engineering, University of Connecticut. AIAA member. [email protected] † Senior Research Engineer, Dynamic Systems and Control Branch, MS 308, NASA Langley Research Center, [email protected], AIAA Assoc. Fellow. ‡ Assistant Professor, Department of Mechanical Engineering, University of Connecticut. [email protected]
T
https://ntrs.nasa.gov/search.jsp?R=20120014594 2020-04-19T08:24:26+00:00Z
dynamics dHence, thisan importa
Ref [2-7].
verified an
recently usof a generi
control is i
This pa
to an adva
class of nothe controlpredictor wcontrol thecontroller and the actstep size oflexible veon the outpthe entire sfurther mo
control lawwell as timobservabili
flexible veproposed L
[10] is revperformancthis class o
The derevisit the of vertical loading con
does not allow s trend in aeroant challenge f
State feedbac
nd design proc
sed to enable reic transport air
in the early stag
aper presents a
anced highly f
onlinear systemller considers bwhich is designe predictor outensures unifortual system’s o
of integration. Hehicles. The conput. In output fsystem. It is fo
odification to t
w will relax theme, i.e., a mity and contro
ehicles. Contro
1L output feedb
visited. Numerce as well as q
of vehicles.
earth of publicmodel from Remotion is connsiderations to
Americ
for the more tronautics and hifor adaptive co
ck 1L adaptive
cess matured a
eal-time dynamrcraft with con
ges of applicat
Figure 1.
an application o
flexible aircraf
ms in the presenbounded time-
ned to predict ttput instead ofrmly bounded output is boundHowever, this ntrol law is bafeedback, the eor this reason the 1L output f
e required bouore general u
ollability, will
ol law design back approach
rical simulationquantify the ac
cly available cef. [10]. The d
nstrained to 1o single digits i
can Institute of
raditional desiggh level of unc
ontrol theory. T
control has be
gainst real wo
mic modeling oventional conf
tion to such new
New generati
of a recent exte
ft configuration
nce of unknow-varying unmathe system’s ouf the actual syoutput trackin
ded, and this peparticular exte
ased on dynamieffect on the stthat the unmat
feedback is pro
unds on the unmunmatched non
make the new
details are d, a semi-span h
n results for achievable perfo
II. Montrol-centric
details of the ph2 inches, hardin degrees. The
f Aeronautics a
2
gns based on thcertainty assoc
The fundament
een used in a
orld challenges
of the edges offiguration8. On
w challenges a
ion of efficien
ension of the L
n. The extensi
wn state-dependtched uncertaiutput with arbitystem’s outputg for the syste
erformance bouension is not thic inversion, wtate cannot be etched uncertainoposed, in part
matched uncernlinearity. Thiw 1L output f
discussed in sehigh aspect rat
a very flexibleormance bound
Model Descripnonlinear mod
hysical wind tud stop to hard se wind tunnel m
and Astronauti
he separation bciated with avatal theory of L
number of ch
s. For example
f a flight envelon the other han
as very flexible
t subsonic air
1L adaptive ou
ion of 1L outpu
dent and time-inty. The contrtrary accuracy t. It is shown em. The differund can be madhe most suitab
which will cancensured, and innty is a boundticular changin
rtainty and wilis formulationfeedback more
ection IV. To tio flexible air
e aircraft confid as a function
ption dels for very funnel model castop, and the mmodel is instru
ics
between these ailable aircraft
1L adaptive con
hallenging app
e, the 1L adapti
ope of a dynamnd, the output f
e aircraft.
rcraft.
utput feedback
ut feedback co
-varying nonlinrol algorithm c
y and a control that the adapt
rence between de arbitrarily s
ble formulationcel the effects n fact, the feed
ded function ofng the control
ll allow dependn, with the use suitable for
demonstrate rcraft wind tun
iguration illustn of the contro
flexible aircrafan be found in model angle ofumented with a
two set of dyndesign tools p
ntrol is introdu
plications, the
ive control has
mically scaled feedback 1L ad
controller in R
ontrol design i
nearities. Morconsists of an law which is utive output feethe output pre
small by reducin to deal with of uncertaintie
dback can destaf time. In this law itself. Th
dence on the ssual assumptiodealing with
the potential nnel model from
trate the algorol computer CP
ft led the authRef. [10]. The
f attack is limiaccelerometers
amics. present uced in
theory
s been
model daptive
Ref [9]
is to a
reover, output
used to edback edictor ing the highly
es only abilize paper,
he new
tate as ons on highly
of the m Ref.
rithm’s PU for
hors to e range ited by s along
American Institute of Aeronautics and Astronautics
3
the spar, strain gauges at the root and mid-spar, a rate gyro at the wing tip, and a rate gyro and accelerometers at the tunnel attachment point.
The mathematical model is linear and includes rigid body translational and rotational displacements and velocities , , ,z w q , as well as twelve flexible modes. The flexible modes are represented by generalized
displacements, i , and velocities, i . There were 10 test points for the flexible vehicle wind tunnel model, each
corresponding to a different dynamic pressure ranging from 30 psf to 130 psf shown in Table 1, and an associated linear model described above. The simulation consists of linear models with third order actuator dynamics, typical of aeroservoelastic models, included for each of the control surfaces in the simulation.
Table 1. Dynamic pressure for different test points.
Test Point
Index 1 2 3 4 5 6 7 8 9 10
Dynamic Pressure (psf)
30 35 40 45 50 60 70 80 90 130
The general structure of the flexible model is given by
0
0
0
0 0 0
e lagr rr r r r
r lage ee e e e
cmdr elag laglag lag lag lag
cmd
x xA A A A
x xA A A A
x xA A A A
x x BA
(1)
where rx represents rigid body position and rates, ex represents elastic mode deflections and rates, lagx represents
aerodynamic lag states, and x represents actuator states. For control design purposes, the model is residualized to
eliminate lag states and then is further reduced by eliminating higher frequency flexible modes. Furthermore, the actuator dynamics are neglected, and as a result, the control design model is reduced from 112 to 12 state variables consisting of 2 rigid (half-span) modes and 4 flexible modes. Thus, the model used for design has the format
e rrr r r
r eee e e
rr e
e
xx A A A
xx A A A
xy C C D
x
(2)
where 1 2 3 4[ , , , , ]T
LE TE TE TE TE and y is the output of sensors described in Ref [10]. In order to improve
tracking and add damping into the system, a proportional plus integral (PI) control structure that tracks vertical displacement is chosen for the baseline controller. In order to create a model structure compatible with 1L output
feedback several of the sensor measurements, combined in such a way as to isolate flexible wing modes, are added to the system in Eq. (2) as states. In addition, the integrator on position is augmented to the system in Eq. (2) resulting in a new system structure given by
m m
m
x A x B u
y C x
(3)
where u is a vector of control inputs (one leading edge and four trailing edges), y is a vector of sensor and
integrator outputs, and 16 16 16 5 8 16, ,m m mA R B R C R are matrices with appropriate dimensions.
Similarand then uadaptation dynamics, first and sepitch mom
An ove
adaption scfilter mech
share the sclose to th
adaptive coa high levethe 1L adap
in a tradeo
r to Ref [10], tuse adaptive c
is to maintainin the presenc
econd bending ment at the pivo
erview of the c
cheme to estimhanism to get
same state predhe model used
ontroller accouel perspective, ptive controlle
ff between per
Fi
Americ
III. the general concontrol as an an the desired ce of uncertainmodes of the w
ot point.
control architec
mate the unmoda band-limited
dictor (this wilfor the LQG
unts for all the the LQG is cor is dealing wi
rformance and
igure 3. 1L ad
can Institute of
Control Objntrol approachaugmentation osystem perfor
nties or unknowwing, pictured
Figure 2
cture is illustra
deled uncertaind augmentation
ll be addressedcontroller des
model uncertaontrolling the eth the model m
stability.
daptive contro
f Aeronautics a
4
ective and Dh is to design aon the nomina
rmance definedwn variation inin Fig. 2, whil
. Wind tunne
ated in Fig. 3. T
nties. The contn control signa
d in more detasign, the outpu
inty and unmoexact plant as
mismatch and u
ol architecture
and Astronauti
Design Approa robust linear al control systd by the nomin plant dynamile executing ve
el model
The 1L adaptiv
trol law for disal. As an augm
ail in the next ut of the 1L ad
olded dynamicsthe one used f
unmodeled dyn
e for stability
ics
oach controller at o
tem. In this apinal controllerics. The object
ertical position
ve control augm
sturbance rejecmentation con
section). Whedaptive law is
s with an adaptfor the LQG panamics. The low
augmentation
one of the test pproach, the rr/vehicle closetive is to contrhold and contr
mentation uses
ction uses a lowntroller, LQG a
en the plant moalmost zero. T
tive parameter.arameter desigw-pass filter is
n.
points role of d-loop rol the rolling
s a fast
w-pass and 1L
odel is The 1L
. From gn, and s tuned
American Institute of Aeronautics and Astronautics
5
IV. 1L Augmentation Adaptive Controller Design
Consider the following system dynamics
( ) ( ) ( ) ( ) ( )
( ) ( ( ), ( ), )
( ) ( )
z w gx t Ax t z t w t Bu t
z t f x t z t t
y t Cx t
(4)
where ( ) nx t R are system states, ly R are output variables available for feedback, ( ) pu t R are control signals,
( ) mz t R are bounded input bounded output (BIBO) unmodeled system dynamics, ( )gw t is a bounded external
disturbance, , ,A B C are unknown matrices of appropriate dimensions, , ,z wA represent unknown parameters
with given conservative bounds, and f is a Lipchitz continuous unknown function.
The control objective is to design an augmentation adaptive output feedback controller, using only available measurements, ( )y t , such that ( )y t tracks a given bounded reference input ( )r t with desired transient and
asymptotic performance. Note that given a Hurwitz matrix mA such that the triple , ,m m mA B C is controllable and observable, Eq. (4) can
be rewritten as
( ) ( ) ( ) ( )
( ) ( ( ), ( ), )
( ) ( )
m m
m
x t A x t B u t t
z t f x t z t t
y t C x t
(5)
where ( )t is a vector of time varying signals that includes all the uncertainties and the disturbances, i.e
( ) ( ) ( ) ( ) ( ) ( ) ( )m z w g mt A A x t z t w t B B u t (6)
The design process is to use the parameters , ,m m mA B C , from Eq. (3),to design an LQG baseline controller to
achieve the desired tracking performance in the nominal system. Then an 1L output feedback controller is designed
to deal with the model uncertainties and the disturbances in the ( )t term.
A. LQG Baseline Controller Design:
The LQG controller is the combination of a Kalman filter with a linear-quadratic regulator (LQR). The detailed design procedure for an LQG is omitted here, but an overview using the above problem formulation is provided below. The Kalman filter is
ˆ ˆ ˆ( ) ( ) ( ) ( )( ( ))
ˆ ˆ( )m m m
m
x t A x t B u t L t y C x t
y C x t
(7)
where ˆ( )x t is the estimated state and ( )L t is the observer gain for the Kalman filter. Tuning the covariance matrix
of process noise and measurement noise in the LQG problem formulation can change the performance of the Kalman filter.
The LQR controller design uses the estimated state, ˆ( )x t . The baseline control law is defined as
ˆ( )u Kx t (8)
The dynamic performance and control effort is tuned by choosing different performance and control effort matrices Q and R. The overall baseline control law is defined in Eq. (8) with the estimated state, ˆ( )x t , generated by Eq. (7).
There are several limitations to the LQG design. The LQG design is optimal when using the real plant model,
, , , ,m m mA B C A B C . In reality , ,A B C are unknown. The LQR controller has limited stability when , ,A B C
American Institute of Aeronautics and Astronautics
6
have large deviations from , ,m m mA B C . The white process noise assumption on the Kalman filter is also not
satisfied in this system. Hence, the LQG design has a limited performance and stability margin due to the invalid white noise assumption.
B. 1L Control Augmentation Design:
1. State Predictor:
In addition to Eq. (7), a vector of adaptive parameters, ̂ , is introduced to the predictor.
ˆ ˆ ˆ ˆ( ) ( ) ( ) ( )( ( ))
ˆ ˆ( )m m m
m
x t A x t B u t L t y C x t
y C x t
(9)
2. Adaptive Law:
The adaptive law updates ̂ as piece-wise constant signal such that output tracking, ˆ( ) ( )y t y t is ensured.
Prior to introducing the adaptive law, certain variables need to be defined as follows. Since mA is Hurwitz, there exists 0P P that satisfies the algebraic Lyapunov equation
0m mA P PA Q
Q
(10)
From the properties of P , it follows that there exists a non-singular P such that
( )P P P (11)
Given 1( )mC P , let D be an ( 1)n n matrix that contains the null space of 1( )mC P such that
1( ( ) ) 0mD C P , and further let mC
D P
.
Let T be any positive constant, which can be associated with the sampling rate of the available CPU, and let
(1 ) RT m
m m mn
n mI O 1 be the matrix with first sub-matrix an identity matrix and all other elements zero ( 11
matrices are designed to match the dimensions in Eq. (13) ). Let ˆ( ) ( ) ( )y t y t y t be output tracking error, then the
update law for ˆ ( )t is given by
1
ˆ ˆ( ) ( ), [ , ( 1) )
ˆ ( ) ,
0,1,2,
t iT t iT i T
iT T iT
i
(12)
where ( )T and ( )iT are defined as
1 1( )
10( ) e , ( ) e ( ) , 0,1, 2,3, .m m
T A T A TT d iT y iT i 1 (13)
Letting ( )n n mumB R
be the null space of n m
mB R , so that 0T Tm umB B
and ( )=nm umBr Bank , the adaptive
parameter ˆ ( )t can be further decoupled into matched and unmatched terms
1ˆ ( )ˆ ( )
ˆ ( )m
m umum
iTB B iT
iT
(14)
American Institute of Aeronautics and Astronautics
7
In the following control law design, the matched disturbance, ˆ ( )m iT , and unmatched disturbance, ˆ ( )um iT , are
filtered with different filter bandwidths. This approach defining and updating compensation for matched and unmatched uncertainty in the state feedback formulation has been successfully flight tested on NASA GTM aircraft, e.g. see Ref [11].
3. Control law design:
In addition to the baseline LQG control signal, bu , in Eq. (8), there are two augmentation control signals, ( )mu t
and ( )umu t , for matched and unmatched uncertainties respectively. This control law is inspired by a successful
application in Ref [11]. In this paper the control law is further modified to be suitable for highly flexible vehicles. The overall control law is generated as
( ) ( ) ( ) ( )m um bu t u t u t u t (15)
1 ˆ( ) ( ) ( )m mu s C s s (16)
2 ˆ( ) ( ) ( )um umu s C s W s (17)
where 1
1( ) ( )m mH s C sI A B , 12 ( ) ( )m umH s C sI A B , and W is the DC gain matrix of the system
11 2( ) ( )H s H s . 1( )C s , 2 ( )C s are the square low pass filter matrix with unity gain for the diagonal element. 1( )C s is
the filter matrix for the matched channel and 2 ( )C s is the filter matrix for unmatched channel.
The design objective of ( )mu t and ( )umu t is to compensate for the effect of model mismatch. This
decomposition enables filter design of different bandwidths for the matched channel and unmatched channel separately. Compensating the unmatched uncertainties usually leads to stability challenges. By tuning these two filter bandwidths of 1( )C s , 2 ( )C s separately, the stability margin of the controller can be configured to match the
actual range of the system parameter. A detailed analysis is summarized in Ref [12]. An important modification in this paper is that instead of dynamic inversion of 1
1 2( ) ( )H s H s , the DC gain
matrix, W , of the system 11 2( ) ( )H s H s is inverted instead. This modification is necessary because the flexible
vehicle model structure has a high system dimension and high frequency lightly damped modes. Under nominal conditions, dynamic inversion will improve the tracking performance, but the control signal will easily excite the high frequency mode shapes, especially when dealing with high system dimensions with large parameter deviations due to wide range of encompassed flight conditions. The frequencies of the flexible modes are uncertain and the first and second flexible modes have frequencies close to rigid body dynamics and are close to each other making cancellation by inversion completely impractical. In addition, the rank of the controllability matrix is low compared to the real system dimensions, which means many of the high frequency mode shapes are nearly uncontrollable. To avoid exciting the uncertain flexible mode frequencies, the dynamic inversion is changed to a static inversion, thus, only compensating the static error which makes more sense for the flexible structure. This results in a different stability condition for the 1L output feedback controller.
Revisiting the baseline control signal ( )bu s in terms of the system formulation in Eq. (9) is given by
ˆ( ) ( ) ( ( ) ( ) ( ))b g p cu s Kx s K K C s r s y s (18)
where K is the feedback gain from the LQR controller design, ˆ( )x t is the estimated state from Eq. (6), and
1 1K ( )mg CA B . This choice for K g ensures that the diagonal elements of the transfer matrix1[ ( ) K ]m gmC sI A B
have DC gain equal to one, while the off-diagonal elements have zero DC gain. pK is a
proportional design factor used to add damping to the adaptation loop inside the controller and to tune frequency response and robustness margins when there is model parameter deviation. ( )cC s
is a command filter with unity DC
gain. Together with the pK design, the baseline control law, ( )bu t , makes the system behavior close to the
command filter ( )cC s . ( )r s is the reference input signal.
American Institute of Aeronautics and Astronautics
8
V. Simulation Results The controller was evaluated in simulation on the original full state linear wind tunnel models derived at 10
different test dynamic pressures. The simulation also includes third order actuator dynamics and time-delay in the sensor feedback loop. The simulation time step and, hence, the sampling time was set at 0.005T . For the purposes of this publication, the magnitude of control deflections and vehicle attitude has been normalized with respect to the trim values.
Let ( , , ), 1,....10ir ir irA B C ir , where 112 112 112 5 8 112, ,A R B R C R represent the full state linear model
derived from the wind tunnel test at a given test dynamic pressure with ir associated with a particular dynamic pressure from Table 1. Let ( , , ), 1, 2,...10mi mi miA B C i represent the reduced order model used for controller design
at a particular test dynamic pressure with i associated with a particular dynamic pressure from Table 1. An LQG controller is designed as the baseline controller for each test point. The particular controller is then
evaluated at different test conditions in addition to the nominal test condition. In other words, a particular baseline control signal ( )bu t , given in Eq. (18), is designed using ( , , ), 10mi mi miA B C i is tested for the plant
( , , ), 1,....10ir ir irA B C ir . The stability results for each LQG baseline controller are given in Table 2 where shaded
cells indicate matched design and evaluation conditions, e.g. LQG controller designed using reduced order model at 130 psf is evaluated on full order linear model at 130 psf (ir=10). As seen from Table 2, in the simulation the baseline controller is limited in its ability to stabilize plant models different from the model used to design the controller. For example, the LQG controller designed for model 10 (130 psf test condition) can stabilize full linear models at 90 and 80 psf (ir=9, 8) as indicated by checkmarks in Table 2, but cannot stabilize models for dynamic pressure below 80 psf (ir=7-1, 70 to 30 psf test conditions). The LQG controller does somewhat better in the middle ranges of tested dynamic pressure by stabilizing the system up to 3 test points below its design condition covering the range of 90 to 60 psf. As dynamic pressure drops towards minimum necessary for flight, the range over which LQG can provide stability also decreases. It is interesting to note that in all case, the baseline LQG stabilizes the vehicle for dynamic pressure at or below its design condition. There are two reasons for the unstable response. One is the large parameter variation with dynamic pressure especially in the control effectiveness. The other challenge is the unmodeled dynamics not captured in the reduced order model used for control law design. In summary, the baseline controller designed with reduced order model i appears to stabilize the system for
, 1, 2,ir i ir i ir i and in some cases for 3ir i , which means when the real plant changes, the baseline controller can work for at most 4 test conditions.
Table 2. Stability of baseline LQG controller designed for test conditions.
Index number of model used for the full state plant (used for evaluation)
1 2 3 4 5 6 7 8 9 10
Inde
x nu
mbe
r of
mod
el u
sed
for
cont
roll
er d
esig
n
1 √ 2 √ √ 3 √ √ √ 4 √ √ √ √ 5 √ √ √ √ 6 √ √ √ √ 7 √ √ √ √ 8 √ √ √ √ 9 √ √ √ √
10 √ √ √
In Eq. (18) r R is the command for vertical position in the wind tunnel. Figs. 4 to 10 illustrate the response of the baseline controller designed with ( , , ), 10 (130 )mi mi miA B C i psf when the real plant varies from 130 psf to 50
psf, corresponding to 10ir to 5ir . The discussion is similar to the example describing the results in Table 2. When 10ir i , which means the full state model ( , , )ir ir irA B C is at the same test condition as the design and is
close to the model ( , , )mi mi miA B C used for controller design, the system response in Fig. 4 tracks with no error.
When the real plant is changed to 9,8ir , there are some higher frequency oscillations that become apparent in the
pitch rate baseline cocondition i
Figu
and in controontroller can ni , the closed lo
(a) Ver
(c) Model
ure 4. Measure
Americ
l signal, especno longer staboop system goe
rtical tunnel po
attitude θ (deg
ed output resp
can Institute of
cially the leadbilize the systes unstable mu
osition (in)
g) and q (deg/s)
ponses for the
f Aeronautics a
9
ding edge (LE)em. Fig. 7 illu
uch more rapidl
)
baseline contr
and Astronauti
) as shown inlustrates that aly.
(b) Control
(d) Ve
roller (130psf)
ics
n Figs. 5 and as ir gets furth
deflections (de
ertical velocity
f) evaluated at
6. When 7ir her from the
eg)
Vz (in/s)
t 130 psf (ir =
7 , the design
10).
Fig
Fig
(a) V
(c) Model
ure 5. Measur
(a) V
(c) Model
ure 6. Measur
Americ
ertical tunnel p
attitude θ (deg
red output res
ertical tunnel p
attitude θ (deg
red output res
can Institute of
position (in)
g) and q (deg/s)
sponses for the
position (in)
g) and q (deg/s)
sponses for the
f Aeronautics a
10
)
e baseline con
)
e baseline con
and Astronauti
(b) Contro
(d) Ve
ntroller (130ps
(b) Contro
(d) Ve
ntroller (130ps
ics
ol deflections (
ertical velocity
sf) evaluated a
ol deflections (
ertical velocity
sf) evaluated a
(deg)
Vz (in/s)
at 90 psf (ir =
(deg)
Vz (in/s)
at 80 psf (ir =
8).
9).
(a
Fig
With th
can stabilizto 1. Figs.
Even for t1ir , in F
system and17 because
to stabilize
1L adaptiv
controller cThe sim
very flexib
a) Vertical tunn
gure 7. Measur
he 1L adaptive
ze the system w8 to 17 show th
the largest chaFig. 17 the cond maintain rease the 1L adaptiv
e the system anve output feedb
can maintain stmulation resultble vehicle mod
Americ
nel position at
(c)
red output res
e augmentation
when the real fhe system perf
ange in dynamntrol signal has sonable performve estimation c
nd compensateback controlle
tability and pers are encouragdels.
can Institute of
70 psf (ir = 7)
Vertical tunne
sponses for th
n, the augmen
full order plantformance of the
mic pressure frosmall magnitu
mance. There ican estimate th
e the effect of er are listed in
rformance bouging and the me
f Aeronautics a
11
(b
el position (in)
e baseline con
nted controller
t ranges from e 1L augmente
om nominal 3ude oscillationsis an increase ihe model mism
the model misn Table 3. Sim
unds for all 10 tethodology is r
and Astronauti
(b) Vertical tun
at 50 psf (ir =
ntroller (130ps
with paramet
130 to 30 psf ted controller de
0 psf, which is. The augmenin the control e
match and gener
smatch in the omulation resul
test points ready to be app
ics
nnel position at
6)
sf) evaluated a
ters ( , ,mi miA B C
test conditionsealing with the
is barely enounted controller effort as evidenrate an augmen
output. The stalts show that a
plied to more c
t 60 psf (ir = 6)
at 70, 60, 50 p
), 10 (130miC i , ir changes fr
e unknown real
ugh to sustain can still stabili
nt from Fig. 8 tntation control
ability results fa single 1L ad
challenging non
)
sf.
0 psf )
rom 10 l plant.
flight, ize the to Fig. signal
for the daptive
nlinear
Figure
(a) V
(c) Model
e 8. Measured
(a) V
Americ
ertical tunnel p
attitude θ (deg
d output respo
ertical tunnel p
can Institute of
position (in)
g) and q (deg/s)
nses for the L
position (in)
f Aeronautics a
12
)
1 adaptive con
and Astronauti
(b) Contro
(d) Ve
ntroller (130ps
(b) Contro
ics
ol deflections (
ertical velocity
sf) evaluated a
ol deflections (
(deg)
Vz (in/s)
at 130 psf (ir =
(deg)
= 10).
Figur
Figur
(c) Model
re 9. Measure
(a) V
(c) Model
re 10. Measure
Americ
attitude θ (deg
d output resp
ertical tunnel p
attitude θ (deg
ed output resp
can Institute of
g) and q (deg/s)
onses for the L
position (in)
g) and q (deg/s)
ponses for the
f Aeronautics a
13
)
L1 adaptive co
)
L1 adaptive c
and Astronauti
(d) Ve
ontroller (130p
(b) Contro
(d) Ve
ontroller (130
ics
ertical velocity
psf) evaluated
ol deflections (
ertical velocity
0psf) evaluated
Vz (in/s)
d at 90 psf (ir =
(deg)
Vz (in/s)
d at 80 psf (ir
= 9).
= 8).
Figur
(a) V
(c) Model
re 11. Measure
(a) V
Americ
ertical tunnel p
attitude θ (deg
ed output resp
ertical tunnel p
can Institute of
position (in)
g) and q (deg/s)
ponses for the
position (in)
f Aeronautics a
14
)
L1 adaptive c
and Astronauti
(b) Contro
(d) Ve
ontroller (130
(b) Contro
ics
ol deflections (
ertical velocity
0psf) evaluated
ol deflections (
(deg)
Vz (in/s)
d at 70 psf (ir
(deg)
= 7).
Figur
Figur
(c) Model
re 12. Measure
(a) V
(c) Model
re 13. Measure
Americ
attitude θ (deg
ed output resp
ertical tunnel p
attitude θ (deg
ed output resp
can Institute of
g) and q (deg/s)
ponses for the
position (in)
g) and q (deg/s)
ponses for the
f Aeronautics a
15
)
L1 adaptive c
)
L1 adaptive c
and Astronauti
(d) Ve
ontroller (130
(b) Contro
(d) Ve
ontroller (130
ics
ertical velocity
0psf) evaluated
ol deflections (
ertical velocity
0psf) evaluated
Vz (in/s)
d at 60 psf (ir
(deg)
Vz (in/s)
d at 50 psf (ir
= 6).
= 5).
Figur
(a) V
(c) Model
re 14. Measure
(a) V
Americ
ertical tunnel p
attitude θ (deg
ed output resp
ertical tunnel p
can Institute of
position (in)
g) and q (deg/s)
ponses for the
position (in)
f Aeronautics a
16
)
L1 adaptive c
and Astronauti
(b) Contro
(d) Ve
ontroller (130
(b) Contro
ics
ol deflections (
ertical velocity
0psf) evaluated
ol deflections (
(deg)
Vz (in/s)
d at 45 psf (ir
(deg)
= 4).
Figur
Figur
(c) Model
re 15. Measure
(a) V
(c) Model
re 16. Measure
Americ
attitude θ (deg
ed output resp
ertical tunnel p
attitude θ (deg
ed output resp
can Institute of
g) and q (deg/s)
ponses for the
position (in)
g) and q (deg/s)
ponses for the
f Aeronautics a
17
)
L1 adaptive c
)
L1 adaptive c
and Astronauti
(d) Ve
ontroller (130
(b) Contro
(d) Ve
ontroller (130
ics
ertical velocity
0psf) evaluated
ol deflections (
ertical velocity
0psf) evaluated
Vz (in/s)
d at 40 psf (ir
(deg)
Vz (in/s)
d at 35 psf (ir
= 3).
= 2).
Figur
Stable /
Am=1
This pa
mode contmotion. Th
1L adaptiv
simulation
adaptive cdesigned aconditions
(a) V
(c) Model
re 17. Measure
Table 3.
/not 1
10 √
aper presents a
trol of a high his model exhibve controller is
of the baseline
controller stabiat a single test .
Americ
ertical tunnel p
attitude θ (deg
ed output resp
Stability of 1L
2 √
an output feedb
aspect ratio, hbits a high leveused as an au
e and the 1L ad
ilizes the modcondition can
can Institute of
position (in)
g) and q (deg/s
ponses for the
1 augmented c
3 4√ √
VI.
back 1L adapti
highly flexibleel of rigid bodyugmentation on
daptive control
del when the bn maintain stab
f Aeronautics a
18
)
L1 adaptive c
controller for
Air from the f
4 5 √ √
Conclusio
ve control fram
e, semi-span wy flexible moden a baseline co
l conglomeratio
baseline alonebility for a wid
and Astronauti
(b) Contro
(d) Ve
ontroller (130
models at diff
full state plant6 √
on
mework applie
wind tunnel me interactions aontroller design
on show prom
e cannot do thde range of dy
ics
ol deflections (
ertical velocity
0psf) evaluated
ferent test con
7 8 √ √
ed to an integr
model capable and full body fned at a single
mising results. T
he job. The auynamic pressur
(deg)
y Vz (in/s)
d at 30 psf (ir
nditions.
9 √
rated flight/stru
of pitch and pflutter dynamice test condition
The addition of
augmented conres, all 10 of th
= 1).
10 √
uctural
plunge cs. The n. The
f an 1L
ntroller he test
American Institute of Aeronautics and Astronautics
19
While the results are preliminary since the evaluation models are linear with no significant time delay in the system, they are nevertheless encouraging and warrant further application to more sophisticated fully nonlinear very flexible aircraft models.
VII. Acknowledgments The authors would like to thank John Cooper, Jie Luo in AIM lab at University of Connecticut for their
comments and suggestions.
VIII. References: [1] Del Rosario, R, G. Follen, R. Wahls, “NASA Subsonic Fixed Wing Project Annual Review,” Fundamental Aeronautics
Program Annual Review, 18-19 November 2010. [2] Kharisov, E., N. Hovakimyan, and K. Astrom, Comparison of Several Adaptive Controllers According to Their
Robustness Metrics, In Proceedings of AIAA Guidance, Navigation and Control Conference, Toronto, Canada, AIAA-2010-8047, 2010.
[3] Cao, C. and N. Hovakimyan, “Design and Analysis of a Novel L1 Adaptive Control Architecture with Guaranteed Transient Performance,” IEEE Transactions on Automatic Control, vol.53, No.2, pp. 586-591, 2008.
[4] Cao, C. and N. Hovakimyan, “L1 Adaptive Controller for Systems with Unknown Time-varying Parameters and Disturbances in the Presence of Non-zero Trajectory Initialization Error”, International Journal of Control, Vol. 81, No. 7, 1147–1161, July 2008.
[5] Cao, C. and N. Hovakimyan, “Stability Margins of L1 Adaptive Controller,” IEEE Transactions on Automatic Control, vol. 55, No. 2, pp. 480-487, 2010.
[6] Cao, C. and N. Hovakimyan, “L1 Adaptive Controller for a Class of Systems with Unknown Nonlinearities: Part I,” American Control Conference, Seattle, WA, pp. 4093-4098, 2008.
[7] Cao, C. and N. Hovakimyan, “L1 Adaptive Controller for Nonlinear Systems in the Presence of Unmodeled Dynamics: Part II,” American Control Conference, Seattle, WA, pp. 4099-4104, 2008.
[8] Gregory, I. M. “From Dynamically Scaled High Risk Flight Testing to Learn-to-Fly: Control Enabled Capabilities for Autonomous Flight,”, Aerospace Decision and Control Workshop, Georgia Institute of Technology, 11-12 June 2012.
[9] Cao, C., Naira Hovakimyan,“L1 Adaptive Output-Feedback Controllers for Non-Strictly-Positive-Real Reference Systems: Missile Longitudinal Autopilot Design.” Journal of Guidance, Control, and Dynamics, vol. 32, No. 3, pp.717-726, May-June, 2009.
[10] Gregory, I.M., C. Cao, V. Patel and N. Hovakimyan, “Adaptive Control Laws for Flexible Semi-Span Wind Tunnel Model of High-Aspect Ratio Flying Wing,” in Proceedings of AIAA Guidance, Navigation and Control Conference, AIAA-2007-6525.
[11] Gregory, I. and Xargay, E. and Cao, C. and Hovakimyan, N. “Flight Test of L1 Adaptive Controller on the NASA AirSTAR Flight Test Vehicle,” in Proceedings AIAA Guidance, Navigation and Control Conference 2010.
[12] Hovakimyan, N. and Cao, C. L1 Adaptive Control Theory: Guaranteed Robustness with Fast Adaptation. Siam 2010.