A Variable Structure Approach to Robust Control
Anthony J. CaUise Friedrich Kramer
Drexel University
Philadelphia, PA. 19104
ABSTRACT
This paper examines the application of
variable structure control theory to the design
of a flight control system for the AV-8A Bar-
rier in a hover mode. The objective in vari-
able structure design is to confine the motion
to a subspace of the total state space. The
motion in this subspace is insensitive to sys-
tem parameter variations and external distur-
bances that lie in the range space of the con-
trol. A switching type of control law results
from the design procedure.
The control system was designed to track
a vector velocity commsand defined in the body
frame. For comparison purposes,, a proportional
controller was designed using optimal linear
regulator theory. Both control designs were
first evaluated for transient response perfor-
mance using a linearized model, then a nonlin-
ear simulation study of a hovering approach to
landing was conducted. Wind turbulence was
modeled using a 1052 destroyer class air wake
model.
INTRODUCTION
This paper addresses a problem that is
present, to varying degrees, in the designL of all
feedback control systems. Host of feedback con-
trol theory is dependent on having a linear, con-
stant coefficient, state model description or
transfer function representation of the open loop
plant dynamics. Since all real plants are non-
linear to some degree, the linear representation
is arrived at by either ignoring nonlinearities,
or by expanding the equations of motion to first
order in the states and control, about a specified
operating point. This is particularly true in the
design of flight control systems. The operating
point is taken as some nominal flight condition.
The linearized perturbation model is an approxima-
tion to the aircraft dynamics at the specified
flight condition. If the controlier is to operate
over a spectrum of flight conditions within the
flight envelope, then it is customary to regard
the transitions from one flight condition to the
.next as giving rise to slow variations in the model
paraseters. This leads to the design of flight
control systems with scheduled gains, typically 'as
a function of dynamic pressure. If these transi-
tions take place slowly, then hopefully the sta-
bility and transient performance criteria used in
the linear design are preserved in actual flight.
In practice this is found to be the case to vary-
ing degrees, depending on the aircraft dy-namics
and the maneuvers performd.
i346
of VTOL Aircraft
The view point taken in this paper is that
modeling errors,, due to nonliamarities-, cm be re-
presented as umknown but bounded variations in the
parameters of the open loop,, linear state odeal
description of the plant dynamics. A nonlinear
control i-s designed that meets transient pe-rf or-
mance specifications., and at the same time is near
insensitive. (robust) with respect to same of the
uncertain.paramters that belong to a specified
class. The resulting design is referred to as a
variable structure c-ontrol, because the control
gains (and thus the dlosed l-oop system dynamics)
swiltch according to the state location with refer-
ence to a surfac-e in the state space. The control
is designed to cause the motion of the system to
reach and stay in the neighborhood of this surface.
In the case of m controls, motion i-s desired at
the intersection of m defined surfaces. When the
motion is in so called "sliding mode" along the
surface (or intersection of surfaces), the system
dynamics are insensitive with respect to a class
of paran ter variations. More important, stability
is guaranteed so long as the variations remain
within the specified bounds.
VARIABLE STRUCTURE DESIGN
The design of aircraft flight control systems
is often complicated by the presence of systemn
parameter variations, nonlinearities, cross-cou-
pling effects and external disturbances that are
ignored in the initial design phases. Tradition-
ally, these effects have been diminished through a
combination of high gain and integral feedback.
But because of the limited control power available
in STOL and VTOL type aircraft, it is likely that
such approaches lead to control saturation, even
when the demand for trim control power is modest.
The variable structure system (VSS) design approach
investigated here is near insensitive to a class
of system parameter variations and provides good
disturbance rejection properties, without the use
of high gain.
A VSS control law is a discontinuous feed-
back law whose feedback gain coefficients switc-h
on hypersurfaces defined in the state space. The
o
important feature of VSS is the "sliding mode"
which may occur on the intersection of the switch-
ing surfaces. 'While in sliding mode, the feedback
system is insensitive to certain paraneter varia-
tions and disturbances. The theory of VSS has
been mainly developed in the USSR (1, 21 and a
survey can be found in (3].
 To illustrate the parameter invariance and where it is necessary to e-ither estimate the dis-
disturbance rejection properties, consider a con'- turbance [6] or use a servo-compensator [7]. How-
trolable linear system in the following form: ever, in the case of scaler control, and in the
absence of external disturbances,, reaching and
existence of a sliding mode are guaranteed when
si< 0. This leads to the following control
structure
x2 A21x1+' A22x2+'Bu +g(t)() n
where x,is (n-rn) dimensional; 13g(t) and u are (8)
rn-dimensional, and B is invertabe. The matrices i-l
The gains * are switched according to
A21 A22 and B possess elements that are unknown
but lbounded. In general, they
are time varying
due to the presence of non-linearities in the ori-
ginal system. The vector g(t) represents dis- f, ca,x >
(9)
turbances and ummodelled dynamics, including cou- = $i's 1is~< 0
pling effects from other subsystems.
In variable structure control, u is chosen
so that traJectories are attained near the inter-
where aj and 8i satisfy the following inequalities:
section of a hypersurfaces. The most acceptable ai/ IBI] (10)
surfaces for design purposes are stationary hyper- sign (B) ai 4 sup[c
t
planes
5 - C1 Xl + 12 0 (2) sign (B) s inf[c ai/IBI ] (11)
where C 1is an m x (n-rn) matrixc to be t
chosen as
In (10, 11) B is now a scalar, c is a row vector
part of the design process. tben the trajectory
is in the vicinity of (2),9 it is said to be in .c = 1] [ci, (12)
sliding mode. Using (2) in the first equation in and ai is the ith colums in the composite system
(1) reduces the dynasics to
matrix
=
(A11l -AA12 CI)x1 (3)
Thus, the motion in sliding mode is invariant with A FA11 A1213
respect to system parameter variations and external
disturbances. In order to reach and maintain slid- LA21 A221
ing, a control is designed from a class of switch- Note that the spread between ai and Sincreases
ing controls: with the uncertainty in k and A.This re-
A2
sults in larger jumps in u whns changes sign.
kl1
u+ (x) ,s >O0 Since the control tends to chatter when the system
is in sliding mode, this becomes a major design
consideration. On the other hand, due to the pro-
ui(x) 0 si <
duct form in (8), the switching level decreases in
where si is the ith element in s. Several good proportion to x,and approaches zero as the sys-
numerical examples of the robustness properties of tem is driven to the origin.
variable structure control are given in [43.
The design process consists of two steps. APPLICATIOiN TO AV-8A IN HOVER
First, select C so that the system has
properties in sliding mode, then choose u 1and u
dvsirable This section summarizes a VSS design that re-
suits in a velocity coummad control system for the
to guarantee reaching and existence of thi slidiAg AV-8A Harrier dynamics in hovering flight. The
mode over the feasible part of the state space. linearized dynamic model for the Harrier was taken
The first step is an ordinary problem in control from [8] for a trim relative wind speed of 15.43
theory since it relates to the placement of the rn/s (30 knots). In addition, a linear regulator
eigenvalues of the reduced system (3). An approach design is given for the sae flight condition.
proposed in [53] is to regard 1,as a control vari- This design was done so that the closed loop eigen-
able in the first equation in l) that minimizes values closely match the dynamic-s of the VSS design
a quadratic performance. of the form in sliding mDde. The resulting linear regulator
was used as a baseline for comparing the perfor-
1w T T amace of the VSS design.
VSS Desi,gnfor Longitudinal Dynamics
where t is defined as the time when sliding motion Reference [8] gives the linearized model for
is initiated. In this case the AV-8A Harrier in hover for airspeeds between
0 and 120 knots. Selecting the values for 30 knots,
-l TP(6 6 we have the following model for the longitudinal
C R A32
where P satisfies system dynamics in the body frame:
T R-1 T p+Q_O * - Ax + Bu (14)
PAl Al -PA12 A12P+iO where
P >0 (7) (15)
Step 2 of the design process is fairly invol- T
ved for the case of vector control [1], particu- uT- [6LQ1I, TU,, RN c ] (16)
larly in the presence of external disturbances

CO47
 [z:35 - .02 -9.8 0 .002] (29)
-.105
0
.014
0
0
0
0
1
-. 13
-.

0
309
(17) weeai > Max(d1/.079} , 8i > Mund /.0791 (30)
I..0056 0 0 0 -4.86
.00161
Allowing for possible variations of 100% in the
parameters in (17) with the exception of A(l, 3),
0 -9.8 0
the following selecgtiosswere made:
.06 3 .28 0 .2321 [0
(13)
79 0
a- 13.9 8- 11.41 (31)
[LO079 4.86J [ 19.5 116.0
The engine response time constant (1/4.86 s) was L.041 J 1 L.-005J
taken from Table 14 of [91 for an RN - 86.5%. The nozzle angle control in (29) was used to con,
The VSS design for surge velocity control was trol rate in reaching the sliding surface, and
based on the controller structure shown in Fig. 1. does not affect the dynamics in sliding uode4 In
Since the VS control in this structure is used for general it was found that, while the VS control in
attitude stabilization, the sliding surface is (28) is sufficient to guarantee reaching of the
defined in terms of 0 and q: sliding surface, the time duration of the reaching
sac 166+q ,s60- e-e (19) phase can be large in the absence of a term pro-
portional to s. In the lateral dynamics this term
ec k1dU ou u uc (20) was added directly to 6LAT, since direct side
force control is not available in the Harrier.
and u is regarded as a constant or slowly varying
input. In sliding mode (s - o) we have from (17) VSS Design for Lateral Dynam_ics
and (19). Fromn the data in [8] at 30 knots, the lateral
dynamics are defined by:
~c 1 =/C 1 (1
T0 =-0 + ,t
1 (1 x Ax +Bu (32)
which is stable for any C1 > 0. The transient re- where T *,,(3
sponse is dictated by C01 and is invariant with x v,[
, r, p] 3
respect to remaining state variables. The design T
u =[SLAT, 6RUIDJ (34)
of k1and C is based on Fig. 2. The closed loop
poles were c'hosen to give a natural frequency of 0O 0 0 I 017
1.0 rad/s and a damping ratio of .70. The result- 10 0 0 0 11
ing values for C1 and k1 are : A = 0 9.8 -.042 0 (35)
10 0 -.007 - .06 -.0751
C1=l1.4 sI k - .0729 s/rn (22) [O 0 -.039 .11 -.2601
The heaving motion is controlied using a -conven-
tional proportional control law that results in a 1~~~~~
natural frequency of 3.5 rad/s and a damping'of .70.
0
0!
B-m 0 -.27j (36)
RN = 8.04 Sw
c. (23) .0055 .0851

dv = w - w (24) L..177 -.033J

c
This is accounted for by modifying A5 2 in (1 7) In this case, it was decided to first deter-
from 0.0 to 3.9.1 and deleting the last colurmn in min dRUD) as a conventional control to maintain a
(18),. desired heading for,the stability axis frane. Ig-
To guarantee reaching and existence of the noring A4 3 and A4 5 in (35), and for a natural
sliding mode, it is sufficient that frequency'and dampinig of 1.5 rad/s and .8 respec-
sAi < 0 (25) tive-ly, we obtain
Differentiating (19) and. assuinng u - 0, we obtain ORUD = -26.5 6* - 28.2 r (37)
Si ss8[ E
5~~~~~~
+ .079 6LGI
where
di- xi 64, - 4, - 4, (38)
i-l Substituting (35 into (32) results in the fol-low-
+ 9.8 Cl k1 TaflN] (26) ing single input system
* - A x + b SLAT (39)
where the nominal values f or d are : where
T
d1
.0056 + C d - 1(.0351 c1- .-13 x - [d6,*, d,6v, r, p1 (40)
d2 - .014 +0C1 k1 (.02), d5. - .0016 - C1k1(.002) 0 0 0 I ol
0 0 0 0 II
d3 - C1 k1 (9.8) (27) A-n 7.21 9.8 -.042 7.67 01 (41)
To satisfy (25), the following control structure is -2.25 0 -.007 -2.46 .075
used .875 0 -. 309 1.04 .:26J
{(xi paxi
*01
S 0 >
01
dLON --Z *ixi ; *, - (28) 01 (42)
i-I. .0055
$i I sxi < 0
L.177 j

*348
 dv - v-v (43) we have the following variable structure gai-ns
c
Note that dRIUD has a significant affect on lateral
velocity. This can be seen in comparing (41) with a [:] 8 .o
(35).
In accordance with the design procedure out-
lined in Section 2,, the coefficients of the slid- 17:88] L4 .585j
ing surface were obtained by regarding p as a The gain kt in (51) was adjusted to yield a rea-
control variable for the reduced system: sonable tisle to reach the sliding surface.
-
J,
Al x,+ b, p ~~~(44) Linear RegulatorDsi
where Aland b Iare the upper left and right blocks In order to have a baseline design for pur-
of A in (41) A sliding surface poses of drawing comparisons between the VSS de-
sign based on proportion state feedback, an opti-
ps= C 1x1 -R71b 1 (45) mal linear regulator design was carried out for
both the longitudinal and lateral control. As in
'was determined that minimizes a quadratic perfor- the VSS design,, RNc and dRUDl were first replaced
mance index of the form in (5). It was fouad in terms of state variables in accordance with
after several iterations that the folloving choice (23) and (37). Then the Q and R weighing matrices
of Q and R resulted in a suitable set of closed were chosen such that the closed loop system dy-
loovp eigenvalues: namics approximated the VSS design in sliding mode.
Q diag [1, 1, 0.1, 1]
- (46) The resulting longitudinal and lateral proportion-
R - 10 al feedback controls have the form
A natural frequency of 1.0 rad/s with adequate u =-Kx
damping appears to be a suitable design criterion where
for both longitudinal and lateral velocity control -~Nm LO I -.168 .149 14.7 18.8 01 (4
for attitude only systems, wfhich is the case here
in sliding mode. Wdhile this has not been validated LKHN -.056 0.0 .475 .171 oJ
completely, the results in [10] indicate a pilot
preference for sluggish response in deference to KLAT - [2.33
14.2 .361 .5.40 12.5] (55)
large attitude excursions. The eigenvalues and eigenvectors for these designs
The above design resulted in a sliding surface are given in Tables 2 and 3. Note that the natu-
s - Cx (46) ral frequency and damping of the forward and lat-
C [.1041, 1.4, .091, .312, 1.01
- (47) eral velocity modes are well matched to the corre-
Once again, to guarantee reaching and exis-
sponding sliding modes in the VSS design.
Linear Resp2onses
tance of the sliding mode, we must insure that The longitudinal responses for a 3 mIs com-
s <.0. Differentiating (47) and assulming that mand in velocity with an initial attitude error of
-? -C 0, we have
c c~~~~~~~~C -0.15 rad are given in Figures 3-4, and 5-6 for
the VSS and optimal linear regulator designs,
ss- s[ E di x + d6 6LATJ (48) respectively. For the VSS design, the sliding sur-
face was reached in 0.4 seconds, and the remaining
where the nominal values for the d are: response is characteristic of an attitude only
system with a natural frequency on 1.0 rad/s.
d1 7.21 C 3 - 2.25 C4 + .875
Note the similarities between Figures 3 and S for
-2m9.8 C3 the surge and pitch-responses and Figures 4 and 6
C3- for the control perturbations. Also note from
d3 -.042 C4 - .D039
.007 (49)
Figures 3 and 4 that despite the presence of a
d4 1 m
+75C3 -2.4 C4 +.93 switching controlier, the resulting velocity and
d5-'C2 -.-075C4 - .26 attitude excursions are quite smooth.
Nonlinear Responses
d6- .0055 C 4 + .177 This section compares the lateral performance
Hence, the variable structure control th,tat guaran- of the VSS and the optimal linear regulator designs
tees sA c 0 has the form in the presence of aerodynamic nonlinearities, en-
5 gine and actuator dynamics, and wind disturbances.
dLAI--E *i It (50) First, a 10 second transient response was generated
i-I using a 3 mIs step conmmad in lateral velocity.
The aircraft was initially aligned with a constant
{ a,sx > 0 30 knot wind, to match the flight condition used to
k -4.0 (51)
<0 x,sx obtain the linear design model. The resulting
lateral velocity errors are shown in Figure 7.
In the selection of aL and 8, we allowed for a 100% Note that the response for the VSS is as predicted
variation in the coefficients of A sand B in (35) on the basis of the linear model; whereas, the
and (36), with the exception of A322'
and B51. Then on the basis that
B31,2,' B4,02 response for the linear control is quite sluggish.
This illustrates the robustness of the VSS design.
The differences in the lateral performance of the
ai: max {dii /min {6 (52)
controller can be attributed to the mismatch be-
tween the linear design model and the dynamic
%i sin {di) Imax {d6} model.

1049
 Next, a velocity canmmad generator was used 4. Drazenovic, B., "The Invariance Conditions
s0 that the aircraft was comumaded to capture and inL Variable Structure Systems", Automatica,
track an Approach Trajectory (AT). The AT was Vol. 5, pp. 287-295, 1969.
aligned with the freestream Wind Over Deck (WOD),
at 50' to port, passing over the touchdown center. 5. Utkin, V.I., Young, K-K. D., "Me-thods for
In this case, a ship airwake turbulence model was Const-ructionL for DisconLtinuity Planes in
used. The wind disturbances are modeled from ex~- Multidimensional Variable Structure Systems,"
perimental data taken on a 1052 class fast frigate Auto. Remote Contr., pp. 1466-1470, 1978.
model (9], and corresponds to a 45 knot freestream
WOD. Initial conditions were chosen such that the 6. Young, K-K. D., Kwatny, H.G., "Formulation
aircraft was 50 m from the landing point, offset and Dynamic Behavior of a Variable Structure
laterally from the AT by 11 metems and heading in- Servomechanism," Joint Automatic Control Con-
to the wind. ference, June 17-19, 1981.
Time history responses of the lateral channel
are shown in Figures 8 and 9 for the VSS and opti- 7. Calise, A.J., Ramuan, K.V., "A Servo Compensa-
mal linear control designs, respectively. All tor Design Approach for Variable Structure
responses are for a 20 second duration.. In com- Systems," Ninteenth Allerton Conference,
paring the VSS and optimal linear control, the key University of Illinois, Sept. 30 - Oct. 2,
feature is the lateral velocity error and the AT 1981.
error at about 10 seconds into the traJectory(Figs.
8 and 9, b-c). This corresponds to the region- 8. Lebacqz, J.V., Aiken, E.W., "A Study to De-
behind the ship where turbulence is the. greatest. termine the Feasibility of Simulating the
Note that DV and RG are significantly larger for AV-8A- Harrier with the X-22A Variable Sta-
the optimal linear control by roughly a factor of bility Aircraft," Calspan Report No. AK-5876-
2.5. The lateral stick shows a higher level of F-l, July, 1976.
activity (Figs. 8 and 9,d); whereas, the commnands
are quite smooth (Figs. 8 and 9,a). This illus- 9. Nave,, R.L., "A Computerized VSTOL/Small Plat-
trates the benefit of a full authority flight con- form Landing Dynamic Investigation Model,"
trol system from a piloting viewpoint. The dis- Report No. NADC~-77O24-3O,, September,, 1977.
tance from the lateral switching surface is-shown
in Figure 8, e. 10. Carpenter, C., Equivalent System Analysis of
Translation Rate Control Systems for Hover
CONCLUSIONS and Low Speed Flight," AIAA 8th Atmospheric
It has been shown that variable structure Flight Mechanics Conference, August 19-21,
control theory can be used in the design of flight 1981.
control systems. The resulting design is insensi-
tive to a class of unknown but bounded parameter u, w, RN
variations in the linearized model for the plant
dynamics. Although the controller is nonlinear and
contains switching elements, the resulting response
(in the absence of uncertainties) is-nearly identi-
cal to that obtained using linear control theory.
Thus, current design specifications in terms of
response setting time, overshoot and other piloting
related criteria can be used to design the variable
structure control. Variable structure control in
combination with a hilgh gain proportional component
can also.be used to reduce the effects of external
disturbances, such as those due to wind turbulence.
A,CKNOWLEDGHENTS Figure 1. Controller Structure for Surge Dynamics
This work was supported by the Naval Air Sys-
tems Conmmad under Contract No. N00019-81-C-0178,
and by NASA Ames Research Center under Grant No.
NAG 2-8.
REFERENCES
1. Itkis, U.,, Control Systems of.Variable Struc-
ture, New York: Wiley, 1976.
2. Utkin,' V.1., Sliding Modesand Their 4pplica-
tions to Variable Structure Systems, Moscow:
Nauka, 1974 (in Russian), Mim, 1977 (in En-
glish).
3. litkin, V.I., "fVariable Structure Systems with Figure 2. Approximate Model for Surge Dynamtics in
Sliding Mode: A Survey,," IEE Trans. Automat. Sliding Mode.
Contr., Vol. AC-22, pp. 212-222, April, 1977.

%J5O
 t-)
r)
TABLE - 1
EIGENVALUES AND EIGENVETORS FOR
cl 2.
LATERAL DYNAMICS IN SLIDING MODE ir

,7j
C')
S1 ATt KLGEANVALUES
i
LJ
VAAIA5LES -.727 4 i .563 -1.1. + 1 .sos
.io
f..U, 11
A.;,
f la2 -0Y+i11 12.4 1 11.4

I.
+110'U -4.4 + 16.74 -l.15 +1 2.60 &'
IV 1~~~~.0 1.0 VV/-'
X10 2
1 -.M4 +1 .75 -2.50 + 1 2.23
I ;',.o ot).

TABLE - 2 Figure 4. Longitudinal stick (1) and nozzle

EIGENVALUES AND EIGENVECTOR.S FOR CLOSED LOOP angle (2) with VS control for u - 3
LONGITUDINAL DYNAMICS - LINEAj.RZGIJATOR DESIGN rn/sec and e(0) = -0.15 rad.C

STATE: SICDIVALUES
%AIIA&LLS -.690 + 1 .652 -2.4$ ± i 2.52 -.634 (I

6. 1.0 -.001 1.0

Sw -.0i5 .061 + .065 .006 C)

laIto0 3.41 :; i 10 .2 -.00 + 1 .020 3.76

cl.
C)
q 102 4.56 + 1 9.4 -.028 i.7 31
C')
RH -.316 + 1.050 1.0 .061 9
,i,;
I

1. C)
m9
73.14-
TABLE - 3
EIGENVALUES AND EIGENVECTORS FOR CLOSED LOOP
LATERAL DYNAMICS - LINEAR REGULATOR DESIGN C)
9g..)

STrAlL LWWUVAUIKS C-)

Li
VARIABLE.S -.695 41i .712 -1.25 ± ± .924 - 1.11 IJ,
InI to1) Ililt j, iOtLj
4.0u S. U C.U
s# x 10.532 ± i .772 22. 3 * 7.02 -2.50
1 102
Figure 5. Surge (I) and pitch (2) perturbation
--.37 + 1 6.5. .12.0 +i 8.42 -11.2 with proportional control for u- 3
6. . 1.0 1.0 rn/sec and 0'(0) - -0.15 rad.
r 10 2 - .921?
+i .160 -21.4 ± 29.4 2 .72
p 02 -.413 sO
i19. 22.7 +.51S 12.4

r,11
I i"I I
,C-1
C.'i
C-)
cI
C-)
2.
6
C,
C-)
9- -C)
.11
I
a:,' c-SC
F ci
ef CV C,

0
9(F) I.

c-,
1.
.0
3.j'0 4.0
). 0 6. 0
C.0 1.0 t
I It
J. 0
(biL)
4.0 5.0 6l.0
Figure 3. Surge (1) and pitch (2) perturbation Figure 6. Longituldinal stic-k (1) and nozzle
with VS control for u - 3 u/s sand angle (2) perturbation with propor-
0(0) - -0.15 rad.c tional control for u. 3 u/see.
 5 .00 5.00

DV-14/ S I vc0H-H/S

-5 .00 i -5 .00
5 .00
(a) VfSS. 5 .00
la)Laeral veocity command.

DV-i{/ S i i Ft V-
0
i--
I Dy-M/S

.13.00 -5 .00
(bN Otinmt linear regultor. 20.00

Fi gure 7. Lateral velocity error tine histories

(10 sec) for a 3m/s lateral velocity
commaid. RCG-M

-20 .00
10.00
Ic) Lateral offset from AT.

5 .00

IRLATSTK-GCM

VCOM-H/S
-10 .00

-5.00 0 .50

5 .00 (aLtaktea velocity comrnund.

S-RAD/S

Dy-H/S
-0 .50

(e) Distanc from laera swiching surftac.

-5.-00
20.00 Fi gure 8. VSS time histories (20 sec).

RG-M

-20.00
5 .00 (c) Lateral off set f rom AT.

RLATSTK-04M

-5 .00

Figure 9.
(20 sec).

V52