You are on page 1of 14

Automatica, Vol.30, No. 10, pp.

1527-1540,1994
ElsevierScienceLtd
Pergamon 0005-1098(93)E0057-B Printed in Great Britain

Robust Longitudinal Axis Flight Control for an


Aircraft with Thrust Vectoring*t
JAMES M. BUFFINGTON,:~ A N D R E W G. SPARKS$ and
SIVA S. BANDA$

A reduced-order H= controller is designed for full envelope flight control


of a thrust vectoring aircraft. Structured singular values and nonlinear
simulations are used to analyze the robustness of the control system.
Key Words--Robust flight control; thrust vectoring aircraft; reduced-order control; flying qualities;
H~ control.

Abstract--A full conventional envelope longitudinal axis structured and unstructured uncertainties is not
control design is presented for a fighter aircraft capable of directly addressed using classical techniques.
thrust vectoring. An inner-outer loop modular control
structure is used to provide good flying quantities in the The most natural approach for addressing
presence of highly structured uncertainty across a wide flight multiple structured uncertainties involves robust
envelope. Simple, low-order control laws are designed for a multivariable control techniques.
version of an F-18 aircraft model augmented with thrust
vectoring nozzles. A minimal-order H~ design algorithm is In most flight control problems, initial control
used to aid in the design of an inner loop equalization gains obtained from multivariable design meth-
controller. Structured singular value synthesis is used to ods must be tuned to achieve performance
design outer loop implicit model-following controllers.
Different control laws are found for high and low dynamic objectives. Control gain tuning is a laborious and
pressure conditions, and controller commands are blended extremely expensive task and is further com-
for a small region of dynamic pressure. Daisy-chaining is pounded with high-order, complex control
used to blend elevator and thrust vectoring commands.
Structured singular values are used to analyze stability systems. Typically, the aircraft design process is
robustness to structured parametric uncertainty, actuator and iterative in that the physical configuration of the
sensor unmodeled dynamics, and structured uncertainty aircraft changes numerous times throughout
corresponding to controller blending. A nonlinear simulation
is used to show that the aircraft performs well across the aircraft development. Each time the configura-
flight envelope during outer loop controller blending and tion changes, control law gains must be
thrust vectoring actuation. 'retuned' or redesigned altogether which can be
time consuming and expensive for complex
INTRODUCTION control structures. Therefore, simple low-order
HISTORICALLY, THE TREND o f t h e flight control flight control systems that can be readily
industry has been to use well-established 'retuned' or redesigned are necessary to
single-loop classical techniques for control minimize the design cost. Simple control
system design due to the excellent performance structures will also ease implementation
of these methods. The inherently simple control difficulties, such as gain-scheduling, correspond-
structure and small amount of required com- ing to the use of high-order controllers. The
putation add to the appeal of classical methods. control law design approach presented in this
However, robustness of controlled systems to paper leads to low-order controllers that require
minimal tuning and implementation effort.
One implementation difficulty inherent in
* Received 22 September 1992; revised 10 April 1993; multivariable controllers is commanding of
received in final form 1 August 1993. This paper was
not presented at any IFAC meeting. This paper was
redundant control effectors. Canards, forebody
recommended for publication in revised form by Editor blowers, and thrust vectoring nozzles are
Y. Arkun. Corresponding author J. M. Burlington. examples of redundant control effectors that
Tel. +1 513 255 8680; Fax +513 476 4000; E-mail
buff@falcon.flight.wpafb.af.mil. provide additional control power. However, the
t Contributed paper presented in the session Nonlinear addition of redundant control effectors presents
Control: Flight Applications, at the 1993 American Control the problem of allocating commands between
Conference, San Francisco, CA, 2-4 June 1993. effectors to obtain the desired response.
:~Aerospace Engineer, Control Analysis Section,
WL/FIGC Bldg 146, 2210 Eighth St. Ste 21, Wright- Generally, multivariable control methods gener-
Patterson AFB, OH 45433-7531, U.S.A. ate actuator commands without considering the
AUTO 30:I0-B 1527
1528 ,I. M. BITFFINGTONet al.

possibility of redundant effectors "lighting' each then presented followed by a summary of


other, i.e. redundant effector commands of conclusions.
opposite signs. This paper discusses the use of a
control selector within the controller to address PROBLEM AND MODEL DESCRIPTION
these issues.
The design of a fixed H~ controller for a This paper addresses robust longitudinal
fighter aircraft flying a Herbst-type maneuver stabilization for the full conventional flight
with multiple control effectors is presented by envelope of a fighter aircraft. The conventional
Chiang et al. (1990). One condition along the flight envelope is defined as the subsonic Mach
maneuver trajectory is chosen as nominal and regime between sea level and 40,000 ft. Dynamic
pressure ranges from 50 lbs/ft 2 (psf) to 1000psf.
several other conditions along the maneuver
trajectory represent uncertainty for which the These defined ranges correspond to angles of
robust controller is designed. The control attack up to approximately 25 degrees (deg).
structure consists of an outer performance loop Figure 1 shows the conventional flight envelope
wrapped around an inner equalization loop that with design and analysis model flight conditions
reduces the relative errors between the nominal indicated.
aircraft model and the off-nominal models. A high-fidelity nonlinear model of a modified
Therefore, one non-scheduled robust controller F-18 aircraft augmented with thrust vectoring
is used by Chiang et al. (1990) to perform the nozzles is used for nonlinear simulations and
entire maneuver. Performance specifications are linear model generation at trim conditions for
control design. Leading and trailing edge flaps
addressed through frequency dependent weights
on various sensitivity functions. are scheduled with angle-of-attack to provide
optimum lift-to-drag ratio. The nonlinear
Sparks and Banda (1992) considered the
Herbst-type maneuver used by Chiang et al. aircraft model has detailed aerodynamic, propul-
sion, actuator, and sensor F O R T R A N modules.
(1990) and presented longitudinal and
lateral/directional axes designs at a single flight The extensive aerodynamic and propulsion
databases were obtained from well-documented
condition along the maneuver. Structured
singular value or/x-synthesis was used to design flight and wind tunnel testing and augmented
with thrust vectoring test data. The actuator
a controller formulated in an implicit model-
models include rate and position limits. The
following framework allowing for direct in-
corporation of flying qualities specifications into sensors are modeled with infinitely fast
the design. dynamics.
This paper is an extension of the work by As alluded to above, the nonlinear model is
Sparks and Banda (1992) and Chiang et al. (1990) linearized at trim conditions to generate
in that a model-following framework is used in state-space models for design. Table 1 lists the
an inner-outer loop control structure to provide design and analysis conditions shown in Fig. 1.
a full envelope capability for the aircraft. The The linear models have ten state variables that
design goal is to obtain acceptable longitudinal represent the aircraft dynamics within a small
aircraft flying qualities, despite modeling uncer- region around the trim condition. For the flight
tainties, using conventional aerodynamic control envelope of interest, the longitudinal and
effectors and thrust vectoring for a wide flight lateral/directional dynamics are sufficiently de-
envelope. Flying qualities requirements obtained coupled to partition the dynamics into a
from Military Standard (1990) specifications are
used to assess the quality of the aircraft response 'FxlO~

to pilot inputs as sensed by the pilot. These 45! x - design condition


o - analysis condition
requirements are directly incorporated into the 4;
design procedure to guarantee satisfactory flying
qualities.
In the following sections, the problem altitude 3
addressed in this paper is defined, and a (It) 2.5

description of the aircraft model is given. The


body of the paper begins with descriptions of the
controller structure and design. Subsections are
organized in the natural order of the design
process. These subsections are organized to
present background theory followed by applica- 02 04 0.6 08 I 12

tion of the theory. Flying qualities, robustness, Mach Number

and nonlinear analyses of the control design are FIe;. 1. Flight envelope.
Robust flight control 1529

TABLE 1. DESIGN AND ANALYSIS CONDITIONS Controller

I • outer I
Condition Mach Altitude (ft) # (psf) a (deg)
%"-~+ CterrorL ~ qcmd • ~E
D1 0.3 26,000 47.5 25.2
D2, A1 0.5 40,000 68.5 16.8 I I
D3 0.6 30,000 158.4 6.8 inner ]
D4, A2 0.4 6000 189.9 6.0 I •

D5, A3
D6, A4
0.7
0.8
14,000
12,000
426.4
603.0
2.6
1.9
L
D7 0.95 20,000 614.4 1.6
D8 0.8 10,000 652.0 1.7
D9 0.8 5000 789.1 1.5
D10 0.9 10,000 825.2 1.4 FIG. 2. Control structure.
D11 0.85 5000 890.8 1.4
D12, A5 0.9 5000 998.7 1.3

D ~ e s i g n condition A--analysis condition


(Kper0. The controlled plant (G) is represented
by the remaining element of Fig. 2, and Mach
number and altitude dependency is indicated by
(M, h). The design and objective of each control
fifth-order model for the longitudinal axis and a system component is described in detail in the
fifth-order model for the lateral/directional axes. following subsections•
This paper describes only the longitudinal
dynamics model described by the state Control selector
variables, angle-of-attack (a), pitch rate (q), The idea behind the control selector is to
velocity (u), altitude (h), and pitch angle (0). introduce generalized controls in cases where
Since the pilot controls the trajectory states there are redundant control effectors (Chiang et
(u, h, 0), they are eliminated from the linear al., 1990). Generalized controls keep the
models thus resulting in two state short-period force-moment generation accounting decoupled
longitudinal linear models of the following form: from the remaining control system (Shaw et al.,
1988). The control selector dictates how the
o
control effectors will generate the necessary
Mq][q]+kM~E M~rv J I-6vrv J forces and moments commanded by the rest of
(1) the control system. Consider the linear model
where 6e is the elevator deflection, 6vrv is the with realization G (Ac, Be, Co, 0) and actual
thrust vectoring nozzle deflection in the pitch controls u,
axis, Z,, Z~E and Z ~ v are z-direction force 2 = A c x + Bcu, (2)
derivatives, and Ms, Mq, M~E, and M ~ v are the
and with generalized controls u*,
pitch moment derivatives. Short-period lon-
gitudinal linear models of this form are = AGX + B'U*. (3)
generated at the design conditions indicated by
To preserve the state dynamics, equations (2)
Table 1. Because the control law gains will be
and (3) must be equal or
scheduled with dynamic pressure, the design
conditions are chosen to cover a broad range of B ' u * = Bc;u. (4)
dynamic pressure.
Therefore, the mapping from the generalized
controls to the control effectors,
CONTROLLER STRUCTURE AND DESIGN
u = Tu*, (5)
The longitudinal axis manual flight control
system generates elevator and pitch thrust vector becomes the control selector
commands from angle-of-attack commands. r = B~B* (6)
These control effector commands stabilize the
short-period aircraft dynamics across a wide where B~ is the pseudo-inverse of Ba. Note that
conventional flight envelope. if Bo has full row rank, then B* is exactly
The structure of the longitudinal flight control obtainable, and the realization of the control
system is modular, where each control module is selector and open loop plant is
designed for a specific purpose. Figure 2 shows (~-(Ao, B*, Co, 0), which is used as the design
this modular control structure. There are three model plant for the inner loop equalization
control system modules: the nonlinear control controller in the next section.
selector (T), the inner equalization controller Since there are two independent control
(Keq), and the outer performance controller effectors to control state equation (1), equation
1530 J.M. B[:FFINGTONet al.
• outer
(4) becomes

l Jk0~mdJ [Ms, Ms .... 6vrvJ"


1 (7)

Continuous commanding of thrust vectoring is


not feasible because of practical issues such as
heating limitations on the thrust vectoring Inner
U = qcmd ~ . ~
nozzles. The thrust vectoring will be used only to
augment the elevator effectiveness during satu-
ration. Therefore, only one command from
equation (7) is possible. Since the elevator and Flo. 4. Inner equalization design model.
pitch thrust vectoring nozzles are primarily pitch
moment effectors, pitch accelerator commands
are most appropriate: vector command ~-.'"PTV}l~' cm
~dusing the inverse of the
thrust vector control effectiveness (Ma,),v).
0c,~d = [Mo~: Ma,.,v] 6m~v • (8)
Minimal-order H~ inner equalization controller
The actual control surface commands for the A minimal-order H~ design algorithm de-
longitudinal control system are elevator and veloped by Yeh et al. (1992) considers the linear
pitch thrust vector commands, and the general- time-invariant system in Fig. 4. The minimal-
ized control is now just pitch acceleration order H~ design algorithm generates a controller
that stabilizes the closed loop system and bounds
the H~-norm of the transfer function from the
u = 8 ll* = gl~ma B* -- . (9) disturbance (Wl) to the controlled outputs (Zl, z2)
by 3'. The controller has the Luenberger
Implementation issues, such as blending of observer-based structure shown in Fig. 5 and has
thrust vectoring commands, are also addressed dimension equal to the number of plant states
within the control selector. A daisy-chain minus the number of plant measurements. The
method is used to generate thrust vector controller state, x0, is an estimate of a linear
commands. Thrust vectoring is used only when transformation of the plant state, Tx, where T is
the aerodynamic surfaces are not able to the transformation matrix. The controller para-
generate the necessary forces and moments meters are Kf, F, T, M, and N, and they must
required for commanded maneuvers. Nonlinear satisfy the Luenberger constraints
elements, such as position and rate limits, are
required to implement the daisy-chain thus TA - F T = K f C (10)
making the control selector nonlinear. Figure 3 G- TB (11)
shows the structure of the nonlinear control
selector based upon equation (8). A limited N T + M C = Kc, (12)
elevator command (8~mJ) is generated from a for any Kc such that A - BKc is Hurwitz.
pitch acceleration command (0~md) via the According to Yeh et aL (1992), the robust
inverse of the elevator control effectiveness controller parameter Kc is given by
(M~2) and elevator rate and position limits. An
achievable pitch acceleration (O~,,~,) is com- Kc = [H~H2] IBTpI (13)
puted from the limited elevator command using with P1 -> 0 satisfying
the elevator control effectiveness (M<). The
difference of the commanded and achievable PtA + A r p j + H~HI + (1 + a ) T - zP~G~G~P~
pitch acceleration is transformed to a pitch thrust - PIB[HTH2] 1BTp~<--O, (14)

I %
FiG. 3. Nonlinear control selector. Fit;. 5. Minimal-order ff~ controller structure.
Robust flight control 1531
P
subject to the existence of a P2 -> 0 that satisfies
I q
P2F + FTp2 + NTHTH2N • outer ~ • ~E ~ ~ a I

+(I+a-')y-2e2TG1GTTTe2<_0. (15)
The design parameters a, Kf, and F represent
the freedom available to the designer and are
chosen such that a is a positive real scalar, F is
LI thl
any stable matrix, and Kf is completely arbitrary. a. C l o s e d i n n e r l o o p
The design tuning parameter a is used to tighten • ou~r
q cmd
the disturbance to controlled output H~-norm
bound as described by Buffington et al. (1992).
An initial bound y is selected that allows the b, R e f e r e n c e model

existence of a positive semi-definite solution to FIG. 6. Inner loop equalization.


equation (14). The controller parameter Kc is
now computed using P1, and then the Luenber-
operating conditions. Therefore, an effective
ger constraint equations (10)-(12) are solved.
inner loop controller must be designed to reduce
Sylvester equation (10) is solved for T, and then
the relative error between P and Po as defined by
G is computed using equation (11). Parameters
equation (17).
N and M are computed using a variation of
Since P is a function of flight condition, the
equation (12):
equalization module must be robust across the
[N M ] = Kc[1C]
-1.--- (16) flight envelope or scheduled against a slow-
varying flight condition-dependent parameter,
Note that the choice of Kt and F provide the such as dynamic pressure, as suggested by
design freedom to obtain a T that insures the Shamma and Athans (1992). Because the aircraft
dynamics vary greatly across a wide envelope, it
existence of I T ] -~. If there exists a solution to is highly unlikely one robust controller can be
designed. Thus, the inner equalization controller
design equation (15), a series of y-reduction
gains are scheduled across the flight envelope as
iterations is performed on equation (14) to find functions of dynamic pressure.
the smallest bound Ymin for which positive Figure 6 shows that P is a linear model of the
semi-definite solutions P~ and P2 exist. If no control selector (T), the open loop aircraft
solution exists to equation (15), y is increased, dynamics (G), and the inner equalization
and the procedure is repeated.
controller (Keq). From the previous section, the
The relative error of a given linear model (P) realization of the combination of the control
and a nominal linear model (P0) is given by selector and open loop aircraft dynamics is
A = (P - po)p~T1, (17) (Ac, B*, Co, 0). If the performance weight has
the realization W (Aw, Bw, Cw, Dw), then the
where Po is a nonsingular reference transfer parameters of the design model from Fig. 4
function. Summarizing Safanov and Chiang's become
robustness theorem (1988), an outer loop
stabilizing controller of P0 will stabilize P
provided the maximum singular value of the 0 Aw J
relative error is less than unity across the (18)
G, = [ B*Dw]
controller bandwidth. Bw / HI=[C* 0],
The purpose of the inner loop equalization
controller is, as the name suggests, to equalize and 142 is the control signal weighting used as a
the aircraft dynamics across the flight envelope, design parameter. Since the control system will
between angle-of-attack response and com- eventually track angle-of-attack commands, the
manded pitch acceleration as shown in Fig. 6. In performance output distribution matrix is
other words, the goal of the inner loop is to
C* = [1 0]. (19)
reduce the variation of the aircraft dynamics
between operating conditions, thereby reducing With the performance weight augmented to the
the modeling uncertainty between flight condi- open loop aircraft dynamics, the number of
tion-dependent aircraft models. Successfully design states becomes three, and the number of
reducing the modeling uncertainty provides a inputs and outputs remains at one and two,
greater possibility that one robust performance respectively. Therefore, the order of the
controller will control the aircraft across all equalization controller is one (number of
1532 J . M . BtJFFIN(iTON et al.

s t a t e s - n u m b e r of measurements). Fhe perfor- 10o~ . . . . . . . . . . . . . . . . . . t


mance and control signal weights used [or the ]

inner equalization design are


A w = - lO0 Bw-I Cw = 45,918
(20) 10~_~
Dw = 510.20 //2 = 0.035.
E
Since the purpose of the inner loop controller is -~(A)I
to equalize the dynamics across the flight
t
envelope, the performance weight, Wp, given by |02 t
equation (20), is chosen as the inverse of a
first-order fit to the open loop dynamics at the
highest dynamic pressure condition (D12) in 7
Table 1. The choice of the highest dynamic
10.2 lO-t 10o lOt 102
pressure dynamics prevents positive feedback
Frequency (rad/see)
from diminishing the fast dynamics at high
dynamic pressure conditions which will in- F I G . 8. C l o s e d inner loop relative errors.

evitably occur for other choices of Wp.


Controllers are designed for each condition in
Table 1, and some controller parameters are To analyze the variation of aircraft dynamics
scheduled with polynomial fits in dynamic at different flight conditions, the relative error
pressure. The resulting controller parameters; defined by equation (17) is computed for a given
with design parameter a = 1 and design weights linear model (P) and a nominal linear model
given by equation (20), are (P0)- In order to evaluate the relative error
F=-40 Kf=[l 1] T=[0 0.0247 0] between these equalized models, the closed
inner loop model at condition D7 (Mach = 0.95,
U = N(~t) M ' = [M,(~) M2(~)]. (2l) alt = 20 kft) is arbitrarily chosen as the nominal
Only three control parameters require schedul- model (B3. Figure 8 shows the maximum
ing with dynamic pressure (~). The elements of singular values of the model errors relative to B,
N and M are fit using least-squares polynomial across the flight envelope at design conditions
curve-fits, linear in dynamic pressure: given in Table 1. Since the closed inner loop
errors relative to condition D7 are less than
N(~) = -0.312c7 + 461 unity for all frequencies and across the flight
Ml(Cl) = -0.058# + 50.5 (22) envelope, P0 is used as the open outer loop
design plant for the performance controller.
M2(q) = -0.0064 + 8.11
The controller given by equations (21) and (22) ix-Synthesis outer p e r f o r m a n c e controller
provides closely matched frequency responses of The general output feedback H~ design
the closed inner loop transfer function (P) from problem is to find a controller K ( s ) such that the
• outer
qcmd to a across the flight envelope as disturbance to controlled output transfer
shown in Fig. 7. function (~,,,) is bounded,
IIT:,. II~ < 7, (23)
for the system shown in Fig. 9. The design model

°I l (G) must satisfy certain assumptions given by


Doyle et al. (1989) to insure the existence of a
K ( s ) that bounds IIL~. II~.
The structured singular value 0 x) of a matrix
M • C "×" is defined as the inverse of the size of
the smallest perturbation, A e A=
diag(A1 . . . . . A ) , Ai e C ...... , that makes I -
AM singular, or

Ix(M) %* 1 (24)
min {~(A) :det (I - AM) = 0}'
104
10-2 104 100 101 102
Frequency (rad/sec) where M and A are shown in Fig. 10. Since ix
FIG. 7. Full envelope equalization. cannot generally be computed exactly, tight
Robust flight control 1533

~ w
r- 0t 0~idnl Z

FIG. 9. H~ output feedback.

FIcJ. 11. Outer performance design model.

FIG. 10. M - A format.


,

I , outer I
bounds of the following form are computed: I qemd r" - "3 Cterr°rl
t_ _ _ .1Kperfr4- - - -1
m a x p ( Q M ) < - t z ( M ) <- inf d,(DMD-1), (25) t-_J
QEQ D~_D

where FIG. 12. Aircraft M - A format.

Q = { Q e_A:Q*Q=In×,,}
- (26)

D = {diag (dllm,× . . . . . . , d,I,,,.×,~°):dj E ~+}. Using this ideal model-following approach, the
outer performance controller design model is
The details of structured singular values and developed in Fig. 11. P0 and IM are the design
their bounds are given by Doyle (1982) and plant and ideal model described previously, and
Doyle et al. (1982). Wp and W~ are design weights.
Combination of general output feedback H~ With the augmentation of an uncertainty block
design theory and structured singular value corresponding to performance specifications
analysis theory form a design technique known (Ap), the overall uncertainty structure is given by
as /x-synthesis, and the details are given by
Doyle (1985) and Balas et aL (1991). By sealing [6d 0I (29)
A= 0 Ap'
the system with D shown in equation (25), the
bound (3') on IITzwll~ from equation (23) is and the performance specification represented
reduced. Using this idea, the D - K iteration by Wp is guaranteed in the presence of the
synthesis technique is defined by
u n c e r t a i n t y A d if

Stabimiinl~gK(s){Oilseed l[-D(s)FIl(P, K ) _ O ( s ) - I H~},


sup {/ZA(M)} --< 1 (30)
eo

(27) according to Doyle (1982). Figure 12 shows the


where D(s) is a real-rational, stable, minimum formulation of the aircraft longitudinal axis outer
phase transfer function that approximates the D loop design model of Fig. 11 into the standard
over frequency since optimum D-scaling is M-A format of Fig. 10 based upon the
performed at each frequency. E(P, K) is the uncertainty structure of equation (29). The
lower linear fractional transformation (Doyle et robust performance controller (Kperf) for P0 is
al., 1991) of the design plant P and stabilizing designed using /z-synthesis with the perfor-
controller K. mance and input uncertainty weights set as
The purpose of the outer performance follows
controller is to generate a pitch acceleration 0.25s + 50 10s + 1000
command from an angle-of-attack command Wp - Wa = - - - - (31)
s+5 s + 10000
that produces a desired robust angle-of-attack
response from the aircraft. This problem is The process of choosing these design weights is
formulated as an implicit model-following somewhat empirical. The tracking of the ideal
problem, where the ideal model to be followed is angle-of-attack response is important at fre-
chosen to be a second-order filter based on quencies most sensitive to the pilot (low
desired flying qualities frequencies), whereas robustness to input uncer-
2
tainties is important at different frequencies
Ct'ideal
=
O-) i d e a l
2 • (28)
(high frequencies). Therefore, the performance
O~cmd S 2 Jr- 2~idealC..OidealS q- ¢.OideaI weight is chosen as a low-pass filter, and the
1534 J . M . B t l v v l x c ; r o N et al.

input uncertainty weight is chosen as a high-pass with respect to the uncertainty structure oi
filter. The selection of the poles, zeros, and gains equation (29) and Fig. 11, of the closed loop
of the filters is iterative. Initial filter values are system with full- and reduced-order controllers.
chosen, a controller is designed, and time It is seen that reduction of the full 13th-order
responses are examined. The filter values are controller to a fourth-order controller does not
tuned until satisfactory tracking occurs with destroy the original robustness. The final
minimal control effort. fourth-order low and high dynamic pressure
Unlike the responses in the lateral/directional controllers are
axes, the desired pitch response is not uniform 1.30 × 102s3 + 7.89 × 104s 2
across the flight envelope. The pilot would like
to feel a faster response at higher speeds. To + 6.06 × 105s + 4. 19 × 10~
account for the different performance require- Kperf(S) S4 + 1.25 × 103s 3 + 6.99 × 103s2
ments, two robust performance controllers are + 4 . 1 4 × 104s + 1.50X 103
designed using two ideal models based on (32)
equation (28): (1) a low dynamic pressure
controller using a slow ideal model (g~d~= 1.20 × 102s3 + 2.91 × 105s2
8, Wide,,l=3rad/s): and (2) a high dynamic hi
+ 3.15 × 106s + 2.38 × 107
pressure controller using a fast ideal model gPerf(S) - s 4 + 1.64 × 102s :~ + 9.76 × 103S 2 '
(~'id~,,, = 0.8, a)ia~, = 5 rad/s). Implementation of + 1.40 × 105s + 1.01 × 104
the two outer loop performance controllers is
discussed later in this section. Figure 14 schematically shows the combination
Structured singular value synthesis typically of controller commands generated from the high
generates controllers of an order much higher and low dynamic pressure controllers of
than the original plant because of the design equation (32). Implementation of the controllers
weights and the frequency-dependent scalings. is achieved through a simple linear blending
The additional state variables in the controllers parameter (6b) that generates a combination of
are a result of the design method. Therefore, controller commands for a region of dynamic
good performance is possible with a reduced- pressure. Figure 14 also shows the blending
order controller. Truncation of a minimal, parameter as a function of dynamic pressure.
balanced realization of the full-order controller The dynamic pressure blending region between
is a common method of controller reduction 170psf and 200psf and the piece-wise linear
(Glover, 1984). A balanced realization of the low function were chosen to maximize the use of the
and high dynamic pressure controllers reveals "fast" high dynamic pressure controller. Once
that the original 13th-order controllers are the high dynamic pressure controller caused an
feasibly reduced to fourth-order controllers. elevator rate saturation for a 5 deg angle-of-
Figure 13 shows the structured singular values, attack step input (-200psf), the low dynamic
1 I
. . . . . .m . . . . . . . . . . . . . . . / ~., . . . . . . . .
0[ Low ~Controller / \ i o9 High q Controller
0~ 0B ' '",
! {)7

06] 06

ij. 05'

°'i
03
\ 01

02 0]
full-order upper b o u n d ~ i
{)1 -- 4th-order upper bound " ~ o, - ~ - 4th:order upp;; bou
10l I{)i 10~ 101 1{)1 I~2 l0 I I{)o 10~ I0~
Frequency (rad/sec) Frequency (rad/sec)
Fl(,. 13. Closed loop system/z-bounds.

5b

% r r o r ~ r
1

q (psi'3
170 200
F'l~i. 14. Controller blending.
Robust flight control 1535

pressure controller command was blended with where ~sp is equivalent short-period damping,
the high dynamic pressure controller command. O~sp is equivalent short-period frequency, To,
The piece-wise linear function was chosen for corre.sponds to the pitch rate zero, and re is
simplicity, and a blending region of about 30 psf equivalent pitch time delay. Details of LOES
seemed to minimize the transient blending and acceptable values of the LOES parameters
effects. are given in the Military Standard (1990). Since
the flying qualities parameters are based upon
ANALYSIS OF RESULTS the LOES given by equation (33), the mismatch
This section describes the results and analysis between the high-order systems and the LOES is
of the results from implementation of the control analyzed for the frequency range most notice-
system described in the previous sections. Flying able to the pilot (0.1-10rad/s). The mismatch
quality results based on low-order equivalent is quantified in the form of maximum unnotice-
short-period approximations are first presented. able added dynamics and mismatch values
Robustness to unmodeled dynamics and para- (Military Standards, 1990) from a transfer
metric uncertainties is then assessed, and finally function matching program. The complete
nonlinear time responses are presented. scheduled flight control system, full-order ac-
tuator models, and second-order aircraft short-
Flying qualities analysis period models are used to generate high-order
Several flying qualities measures are used to closed loop aircraft linear models. Low-order
analyze the handling characteristics of the equivalent systems are generated from the
aircraft and the designed control system. The high-order models using an equivalent system
measures for short-term pitch response are given transfer function matching program.
in terms of low-order equivalent systems The control anticipation parameter (CAP) is
(LOES) that represent pitch rate (q) responses defined as
to pilot inputs (O/emd). 2
O)sp

q
Kq(s+lle-'~o
To,/
O/cm d 2 ,
S 2 + 2~spWspS + COsp (33)
and Fig. 15 shows that all flight conditions meet

. ' • "

co (rad)
sp

I0 ~
lOO 101 ]02

n-"~za(m~d)

tx~l

O~spTO2
I,x,,d2
ioo

~e 0l
L~ll
(see)

%, ' ...... i0. ' ' ' '""~o, 1oo ~ 30o ~ ~o ~oo 7oo IDo 900 I~O
~,p (psf)
FIG. 15. Flying qualities parameters.
1536 .I.M. BI'FFIN(;TON et al.

2O 200[
is- "
150

io.

', I00

Gam~ ................ i Phase


-,<

(dB) o,____ ...............i - ~ (deg) sot


sa. ..........
i
-10~

.15~-

-20, -I~ ~
l0 t 100 l0 b 10 100 101
frequency (rad/see) frequency (rad/sec)
Fl(i. 16. Unnoticeable added dynamics.

Category A, Level 1 flying qualities require- The angle-of-attack and pitch rate sensor
ments for C A P and Wsp except at condition D4 dynamics are captured entirely as unstructured
which is slightly below Level 1. Figure 15 also uncertainty. The sensor dynamics are estimated
shows that Level 1 requirements are met for from flight test data of high performance aircraft
°)spTo2, ~'sp, and r~. Obtaining Level 1 flying and fit to real rational transfer functions
qualities for parameters shown in Fig. 15, representing measurement uncertainty weighting
corresponds to the most acceptable level of functions
handling qualities specifications.
21.9s 2 + l120s + 91,100
The mismatch values from the equivalent
system transfer function matching program range W,~ = s2 + 574s + 1,140,000
from 0.009 to 4.376 which suggest a close LOES (36)
approximation (Military Standards, 1990). Figure 0.745s 3 + 152s 2 + 95.9s + 1.38
16 shows that the command to pitch rate added Wq - s 3 + 626s 2 + 173,000s + 235,000
high-order dynamics are well within the maxi-
mum allowable unnoticeable added dynamics An extensive aerodynamic uncertainty data-
represented by the dashed lines. The mismatch base, developed from wind tunnel and flight test
values and unnoticeable dyfiamics show that the data, is used to generate structured uncertainty
LOES estimates the full-order system well across models of stability and control derivatives. These
the flight envelope for frequencies of interest. models are then translated into state-space
element uncertainty models. The state-space
Robustness analysis" element uncertainties depend on flight condition.
Structured singular value analysis techniques However, for simplicity, the uncertainty model
are used to analyze stability robustness of the for each state-space element,
control system design to uncertainties corres-
ponding to unmodeled actuator and sensor AA¢; = l ±0.02 rA,;(1.1)l 0 l
dynamics, parameters in the plant model, and [+0.04 rA(;(2. 1)i i 2 rA,;(2, 2)1
(37)
blending of high and low dynamic pressure outer = [+0.22 ]&;(l, l)l
performance controllers. ABe; L±0.04 p&;(2, 1) '
A high-order detailed elevator actuator model
is reduced to a simple fourth-order model to is held constant at the worst flight condition case
avoid very small time-step integration during thus making the parameter" uncertainty models in
nonlinear simulation. The reduced-order act- a sense conservative.
uator model is used in the analysis model, and Recall from previous sections that for better
the difference between the high-order and flying qualities, high and low dynamic pressure
low-order models represents the unmodeled performance controllers are designed and imple-
actuator dynamics. The actual error dynamics mented by blending both controller commands
between these models are fit to a real rational with the blending parameter 6 8 . The blending
transfer function that represents the weighting mechanism is normalized such that the blending
function for the actuator uncertainty parameter for analysis, ~b, varies between 1
and 1. Therefore, analyzing robustness to 6b is
0.63s 2 + 3.03s + 0.078
WE = ~ (35) equivalent to analyzing robustness for all outer
s" + 68.4s + 1900 loop controller command combinations.
Robust flight control 1537

IIe

FIG. 17. R o b u s t n e s s analysis model.

combinations of outer loop controller commands


are used. Figure 19 shows the nonlinear
0.~
responses of the aircraft model to identical
la - upper bounds
03
angle-of-attack unit step inputs at the three
different conditions. The elevator and angle-of-
0.¢
attack responses in Fig. 19 are actually the
0..'
difference between the actual responses and the
0.1 corresponding trim values for comparison at
0.3 different flight conditions. The high and low
A4 j
0.2
dynamic pressure conditions use only the high
and low dynamic pressure controllers, respec-
0.1
tively. The difference in response speed between
0
10.2 10a l0 0 101 102 the high and low dynamic pressure flight
frequency (rad/sec)
conditions is due to the different ideal models
FIG. 18. R o b u s t stability. being followed. However, the middle dynamic
pressure condition uses a combination of both
controllers resulting in responses between the
Figure 17 shows the robustness analysis model high and low dynamic pressure conditions. Note
including the structure of the uncertainty as well that for this benign maneuver, thrust vectoring is
as the normalization weighting elements. Figure not required since the elevator does not rate nor
18 shows the robust stability plots of the position saturate.
structure shown in Fig. 17 for all analysis Figure 20 shows the nonlinear time responses
conditions (A1-A5) in Table 1. It is interesting for a more drastic 10 deg angle-of-attack step
to note the absence of a peak near the short input that requires pitch thrust vectoring at the
period frequency suggesting insensitivity of the low dynamic pressure condition. The angle-of-
short period dynamics to the given uncertainty attack response is a nicely damped second-order
structure. The peak at approximately 10rad/s response due to the ideal model-following
corresponds to sensitivity to elevator actuation. framework of the outer loop controller design.
Recall that actuator models were not included This maneuver is severe enough to rate saturate
in the design of the control systems, thus the the elevator and thus engage the thrust vectoring
sensitivity to actuation may be reduced by without going unstable and still maintaining
including such models. good performance. Recall that the daisy-chaining
logic within the control selector commands
Nonlinear analysis thrust vectoring only when the elevator is
Longitudinal stick step inputs are used to saturated.
demonstrate performance of the control system
during nonlinear simulations at high, intermedi-
ate, and low dynamic pressure conditions. Recall CONCLUSIONS
that two outer loop controller commands are A simple inner-outer loop control structure is
blended over a range of dynamic pressure presented for robust flight control design.
according to Fig. 14. Three different flight Daisy-chaining is used as a practical mechanism
conditions are chosen such that three different for blending thrust vector commands with
Robust flight control 1539

q
(deg/see)
(deg)

i 2 ) a s

time (see) t i m e (see)

i,

i m
n q
Z ~.~
~),.: (psf)

L 2 ) 4 s I 2 3 4

t i m e (see) t i m e (see)

5 S
E PTV
(deg) (deg)

t 2 ) 4 $ i 2 2 4

time (see) time (see)


FIG. 20. Nonlinear 10 deg angle-of-attack step response at Mach = 0.3, h = 10 kft.

conventional elevator commands. H~ and Tekawy (1990). A fixed H~ controller for a super-
maneuverable fighter aircraft performing the Herbst
structured singular value theories are used to maneuver. Presented at the 29th IEEE Conference on
design and analyze a control system that Decision and Control, Honolulu, HI.
accounts for desired flying qualities using an Doyle, J. C. (1982). Analysis of feedback systems with
structured uncertainties. IEEE Proceedings D, 129(6),
implicit angle-of-attack ideal model. The full 242-250.
conventional envelope controller is robust to Doyle, J. C., J. Wall and G. Stein (1982). Performance and
parametric uncertainty and unmodeled dynam- robustness analysis for structured uncertainty. In
ics and performs well in the nonlinear simula- Proceedings of the 21st IEEE Conference on Design and
Control, Orlando, FL, pp. 629-636.
tion. The flight control design presented is a Doyle, J. C. (1985). Structured uncertainty in control system
good first step towards high angle-of-attack design. In Proceedings of the 24th 1EEE Conference on
supermaneuvering control design. Decision and Control, Ft. Lauderdale, FL, pp. 260-265.
Doyle, J., K. Glover, P. Khargonekar and B. Francis (1989).
State-space solutions to standard /-/2 and Ha control
REFERENCES problems. IEEE Trans. Aurora. Control, 34(8), 831-847.
Doyle, J., A. Packard and K. Zhou (1991). Review of LFTs,
Balas, G. J., A. K. Packard, J. C. Doyle, K. Glover and R. LMIs, and/~. In Proceedings of the 30th IEEU Conference
Smith (1991). Development of advanced control design on Decision and Control, Brighton U.K., pp. 1227-1232.
software for researchers and engineers. In Proceedings of Glover, K. (1984). All optimal Hankel-norm approximations
the 1991 American Control Conference, Boston MA, pp. of linear multivariable systems and their L error bounds.
996-1001. Int. J. Control, 39, 1115-1193.
Burlington, J. M., H. H. Yeh and S. S. Banda (1992). Robust Military Standard (1990). Flying qualities of piloted
control design for an aircraft gust attenuation problem. In airplanes. MIL-STD-1797A.
Proceedings of the 31st Conference on Decision and Safanov, M. G. and R. Y. Chiang (1988). Model reduction
Control, Tucson, AZ, pp. 560-561. for robust control: A Schur relative error method. Int. J.
Chiang, R. Y., M. G. Safanov, K. P. Madden and J. A. Adaptive Control Signal Processing, 2, 259-272.
1540 J.M. BUVFtNGTON et al.

Shamma, J. S. and M. Athans (1992). Gain scheduling: Sparks, A. and S. Banda (1992). AppLication of structured
potential hazards and possible remedies. 1EEE Control singular value synthesis to a fighter aircraft. In Proc. 1992
Systems Magazine, 12(3), 101-107. American Control Conference, Chicago, IL, pp. 1301-1305.
Shaw, P. D., K. R. Haiges, S. R. Rock, J. H. Vincent, A. Yeh, H. H., J. L. Rawson and S. S. Banda (1992). Robust
Emami-Naeini, R. Anex, W. S. Fisk and D. F. Berg (1988). control design with real-parameter uncertainties. In Proc.
Design Methods for Integrated Control Systems. 1992 American Control Conference, Chicago. IL, pp.
Technical Report, AFWAL-TR-88-2061, Wright-Patterson 3249-3256.
AFB, OH.

You might also like