Professional Documents
Culture Documents
,r._<:>...._/<C_
207117
\ :_L>"
"
i_
:_i!.<:iii
Reprinted from
Journal
ofGuidance,
Control,
and
Dynamics
Volume20, Number1, Pages90-96
I_
AIA
A_
A publication of the
American Institute of AeronauticsandAstronautics, Inc.
1801 Alexander Bell Drive, Suite500
Reston, VA 20191-4344
7/'/.
_"
3
1997
i: _
of California,
Davis,
Davis,
California
95616
Actuator rate saturation is an important factor adversely affecting the stability and performance of aircraft
flight control systems. It has been identified as a catalyst in pilot-induced oscillations, some of which have been
catastrophic. A simple design technique is described that utilizes software rate limiters to improve the performance
of control systems operating in the presence of actuator rate saturation. As described, the technique requires control
effectors to be ganged such that any effector is driven by only a single compensated error signal. Using an analysis
of the steady-state behavior of the system, req_ents
are placed upon the type of the loop transmissions and
compensators in the proposed technique. Application of the technique to the design of a multi-input/multi-output,
lateral-directional
control system for a simple model of a high-performance fighter is demonstrated as are the
stability and performance improvements that can accrue with the technique.
I.
Introduction
HE performance
requirements
of modem, high-performance,
fly-by-wire aircraft have made it imperative that the characteristics of the actuation devices be included in any control system
design procedure. For example, in the control of supermaneuverable
aircraft, the reality of control actuator limitations can determine the
overall control system design philosophyJ
Although the linear dynamics of the actuator are often modeled in such designs, the nonlinear behavior, e.g., saturation, is not often considered in explicit
fashion. "lypically, extensive a posteriori simulation is employed
to determine if and when actuator saturation is likely to occur and
then if such saturation can affect flight safety. Even time-consuming
simulation efforts are often unable to uncover situations that can affect flight safety. For example, probably no aircraft in the history of
aviation has undergone more simulator evaluation than the Shuttle
Orbiter. Yet a serious pilot-induced
oscillation
(PIO) occurred in
early free-flight testing, 2 which led to actuator rate saturation. Actuator rate saturation has also been implicated in the YF-22 crash
early in 1992 (Ref. 3) and possibly in the JAS 39 crash of 1993 (Ref.
4), both of which also experienced serious PIOs. Indeed, the pivotal
role that actuator rate saturation may play in PIOs is coming under
increased scrutiny. 5
II.
HI.
Software
Software Limiter
Rate
Limits:
SISO
Systems
Consider Fig. la, which shows a very simple SISO control system. Although the linear actuator dynamics are ignored here, it will
be assumed that actuator rate limiting does occur. The rate limited
actuator ARL is modeled as shown in Fig. lb. Figure 2 shows the
proposed software limiter SRL as part of a SISO feedback system.
SRL consists of a derivative element, followed by an amplitude limiter, followed by an integration
element. As is evident from their
structures, the input-output behavior of SRz and ARL (or an actual
rate-limiting actuator) will be different. The former device employs
no internal feedback, whereas the latter does. Consider the behav-
Background
a)
As opposed to the methods just mentioned, the research to be described employs a technique for addressing actuator rate saturation
Received Feb. 12, 1996; revision received Aug. 6, 1996; accepted for
publication Oct. 18, 1996. Copyright 1996 by the American Institute of
Aeronautics and Astronautics, Inc. All fights reserved.
*Professor, Department of Mechanical and Aeronautical Engineering. Associate Fellow AIAA.
tAssistant Professor, Department of Mechanical and Aeronautical Engineering. Member AIAA.
b)
Fig. 1 Rate limiting in a control system: a) simple SISO system and
b) the rate iimiter ARL.
90
91
"software"
Fig. 2
limiter,
SRL
ess(t) =
lim e(t)=
t--_ O0
lira sE
(3)
S-"_ 0
L t
+ L)] - D_P
_t
b)
Fig. 3 Rate limiting behavior: a) input/output behavior of software
limiter of Fig. 2 and b) input/output
behavior of rate iimiter of Figs. 1
and 2.
qlSttenn = qR __qt__
diagam
algebra
(1+
D/(1
in which
remnant
in Fig. 4
limiting
ARt can
by SRL
Fig. lb).
Y = D + sGR
- (sGPY/s)
L)Y
= D + sGR
- LY
+ L) _=Dn = Y -- [sGR/(1
(la)
(Ib)
= D +sGR
+ L)I = Y - YL
(lc)
i_
E = [R/(1
i_ _ ,
+ L)] -- (DnP/s)
(4)
(5a)
(5b)
(6)
qL _> qR
(7a)
> qe + 1
or
qC > 1
(7b)
G =
2
0.05s + 1
(8)
(2)
A fundamental
assumption in this analysis is that the closed-loop
system of Fig. 2 is stable, even in saturation. This will allow the final
value theorem to be applied to Eq, (2). Note that open-loop stability
is not required, because if P(s) has right half-plane poles, they will
be canceled by identical poles of D_(s), a condition guaranteed by
(9)
92
HESSAND SNELL
SR
:,
/r
r
t
I
I
,_
1
I
I
IL):')!_
:
C
{ i/:'J
Fig. 4
Quasilinear
representation of Fig. 2.
2
>,
1.2:.
0.4-
""/
f /
I
_(0 in Fig.12
:
1.2 ...__SY....-d
......... wi_houtAimltlng
.........
............
il.............
!!
..............
P_--L- c(0 in Fig. !a
il\ !
wi,outlimifi_g
!
b
[
c(Oin Fig. 2
I
0.4 [---/ ..... :__..._.__._!........ with_limitiog............
c(0
_._
-o.4-
::
:!.......
-_
'
I with lim}ting
I
'7
c(0
r
_...........
,_- -0.4
-1.2- ._
-2
10
'
15
20
-1.2
25
I0
(7) require
15
20 -
25
time, s
time, s
that
(IO)
qL>qP+l=2
(11)
1.2,
I F'-u(t)i
in big. 2
::
uft)
-0.4.
"
::(!
r(t) = sin(3t)
(12)
p--oco
i.Fi=.
I
_2
This modification
of the original compensator
G will cause some
change in the performance
of the system of Fig. 2 as compared with
that of the system of Fig. la when no saturation occurs; i.e., there
will be a modest reduction in stability margins. Finally, the s appearing before the limiter in Fig. 2 will be included in the compensator
in actual implementation.
Although the resulting compensator sG'
is now only proper, as opposed to being strictly proper, the integrator following the limiter will restore the latter property. This can
be important from the standpoint of sensor noise propagation to the
plant actuator.
Figure 5 shows the response of the system of Fig. la to a pulsive
input of unity magnitude and 5-s duration with and without rate saturation. For the case with saturation, the rate limit was 0.2Is. Figure 6
shows the response of the system of Fig. 2 [with compensator G'(s)
and the software limiter] to the same input, with and without the
rate limit of 0.2Is. The performance
improvement
with saturation is
obvious. An examination
of the actuator rate time histories in this
I0
time, s
Fig. 7 Comparison of signals u(t) in systems of Figs. la and 2 with
sinusoidal input and rate saturation.
was applied. Here the rate limit was set to 2.0Is. The characteristics
of u (t) indicate that both systems were in nearly constant rate saturation of-4-2.0/s throughout.
Because of this significant limiting, the
tracking behavior of both the systems of Figs. la and 2 was poor.
What is important,
however, is the fact that the response
u(t) of
Fig. la exhibits an effective incremental
time delay of 0.55 s compared with the same system without rate limits. In comparison,
the
response u(t) of Fig. 2 exhibits an effective incremental
delay of
only 0.12 s compared with the same system without rate limits. The
effective incremental
delays were calculated
between corresponding instants when ti(t) changes sign. If the systems of Figs. la and 2
were representing the stability and command augmentation
system
of an aircraft, with r (t) being the pilot's input, the much larger delay
of the system of Fig. la would probably lead to this system being
more prone to PIOs than the system of Fig. 2 (Ref. 5).
IV.
Software Limiters
Any application of the technique introduced in See. III to a realistic aircraft flight control problem must allow consideration
of
,/
//
i'3
_',,
93
Pe = P" K
By a procedure
show
(16a)
(16b)
Dn2 = D2/(1+L2)
analogous
s R1
sE1
- P,,2Pqt
(16c)
(1 + L1)(1
Dn2Pen(1
GI
+ L2)
(1 + L2)(1
Dn_Pe2,(1
Dn2L2
o_
+ P_, G 0
+ Lt)
(1+ L_)
(15)
P, =P'K
SPz,
I
"l
r-s7,er--Tx
,vc-7
AIRCRAFT I
'1
'1
, I
e_
..111
'.
I---
a)
AMP.
WiTH
I
&
RAVEI
LIMI_t)
I
ACTUATORS
.-.c_
-J
i(
RI _
Dt
G t _-_
S _
L___I
; '
'
.............
r- .............
-_
-
t-_
I
-_
,'7
I _
I /_
I
_
AIRCRAFT
I " I
WITH
[ IAMP.&RATE
I _ [ LIMITED
_ cl
+_i
_ i:_
L
._,
b)
:"
Fig. 8
(17b)
(l+ P,,,O,)
= Y2-Y2L
Each YiL in Eqs. (15) can be defined as the signal Yi that would exist
if no rate limiting was occurring in the ith software limiter, alone.
As Fig. 8a indicates, amplitude, as well as rate limiting, can occur
in the control system actuators. However, the Dni in the quasilinear
systems of Figs. 4 and 8b have been assumed to arise from software
(17a)
(1 + P,_G2)
SPe21G1R1
1 + L2
Eqs. (1),
+ Pe22G2)
(1 + LO
sR2
sE2 =
1 + L1
D., Ll
(14)
the
(13)
= YI--YIL
opening
where
Dnl = D1/(I+LI)
can-
LI = Gl P_H + G2A
1 + Pe22G2
A = P,,, P_
K = RL V
limiting
A 2 x 2 MIMO system: a) with software rate limiting and b) a quasilinear representation of panel a.
94
HESSAND SNELL
Equation (17a):
linear
model
(s) = $2+2_On$+O2n
(18)
q3rd=rm = qD_ I + qLl _ q01
q4thterm = qO_ 2 -I- ql'e12 -- qL_ + qZ_ -- q_Z2 -- qO2
Equation
(17b):
qlStterm = qR2 _ q_
q_dtcm
_ 1
LIMINT
"1
q4thtc'rra= qDn I ..1_qPe2! __ qL2 ..]_qLl _ qPel! __ qOl
!
!
I
I
In obtaining Eqs. (18) and (19) it has been assumed that qL_ > 0
and qP_u o, _>0, criteria invariably met in the satisfactory design of
the feedback system of Fig. 8a. The design task is to select qL, and
q_(i = 1, 2) so that
q/th,,_
< -1
(j = 1, 2 ....
II
8t
(20)
4"
"*[--L..!
!I
t___
in Eqs. (18) and (19). As in Eqs. (6) and (9), allowing
qD_ = _qL,
q& < 1
Fig. 9
" dz_''
(21)
thrust vector input 8Yrv were used to control sideslip/_, and the differential elevator input 8Dr, aileron input 8A, and differential pitch
thrust vector input d_rv were used to control roll rate p,. Obviously,
other choices could have been made. In terms of Eqs. (13),
Eqs. (20) and (21) lead to Eqs. (18) and (19) requiring
qL1 > 1
dSn. a
qOZ > 1
(22)
and
Rz =
qL2 >_ 1
_ qt%, + 1
qO2 > 1
deg/s
0
(23)
6ol
(25)
and inequality
=0
i = 1,2
(24)
0
The qL_ are governed by the q or as evident in Eqs. (16a) and (16b).
Of course, the methodology
just outlined can be extended to 3 x 3,
4 x 4 systems, etc. As in the case of the SISO design, open-loop
stability is not required.
Ganged Controls
The explanation
for imposing ganged controls can now be given.
First, grouping each software limiter with a compensator ensures
that the type of each of the n_ loop transmissions
qL_ (nr = number
of controlled outputs) can be determined by a single compensator;,
e.g., in Eq. (16), qL_ can be governed by G1, etc. Second, ganging
the effectors and defining the nonzero elements of the K matrix as
in Eq. (13) allows nr software limiters to prevent rate saturation of
n e actuators, with n e > n r.
100
K=RL.r=
deg/s
oooj
60
and
=
[_DT,
_A,
_R,
_RTV,
6rrv] r
A classical loop-shaping
procedure was used to determine the
compensators
Ga and Gv, shown in Table 1. The crossover frequencies of each loop were selected as
to,_ = 5.0 rad/s
e% = 4.0 rad/s
(26)
i '' )i
MIMO Example
The aircraft
example
to be examined
involves
the lateraldirectional
control of a model of a high-performance
fighter aircraft. The simplified aircraft and actuator dynamics were taken from
Ref. 13 and are repeated here in Table 1. Rather than employing the
simple rate limiter shown in Fig. lb, a more accurate representation
of a rate and amplitude limited second-order
actuator was employed.
This actuator model is shown in Fig. 9 (ReL 14). The controlled outputs are sideslip/_ and roll rate Ps (t) about the x stability axis. Note
/J
In this application,
require that
qZ# > 1,
qLp > 1,
q_
> 1,
qO2 > 1
(27)
(28)
ThUS,
G#, = G/_ (S + 0.5)
s
Ge' = G, (s + 0.4)
s
(29)
Table I
M = 0.5,
- -0.1354
A =
0.09036
9.93
-0.01091
0.2757
8rrv] r
-0.9949
-1A69
0.5126
2.181
-0.01482
-0.1277
12.12
-0.005695
-0.2797
0.9416
-0.7419
_ith
P' .:
sonware ,_ .
limiter _
Pswithodt
_"_'software ]imiter
xl
,
_\
!
-10.37
0.01555
B =
I
300J
/
X=[_,p,r]
/ ,
500
Flight condition
Altitude = 20,000 ft,
Vehicle dynamics
_=Ax
+Bu
i.
95
__
-
0
0.3977
-0.001175
!
_
"I
...........
//.....
'
lj'
. oo]
...........
,...........
"U.........
i...........
:...........
0.00489 1
0.02817
-0.3861]
g i\ !//:V,:-
P"
/
0
where
State variables
::
10
time, s
[0.707, 30]
L
ampl. limit = -4-17.5deg
Fig. 10 Roll rate response ofsystem ofFig. 8a tu input of E,qs. (30), with
and without software rate llmiter (rate and amplitude limited actuators
always present).
,,_
50
8 withoutJsoftware !
limiter ]
i
30.................
L.I..:
......................
:............
;
qaiAi
....
/-i-i--ig-----!;
....
deg
11 I
'
l i I \, :1/ _ i ',i
IHl "f!i
Vi
8a:
752
[0.59, 75]
'
8R:
-50_0
722
[0.69,
72]
! .l'_..wi_ softwar,
;
limher
i
Compensators b
42613(I) 2
G# = (0)(100)
2
,
,_ (0.5)
G# =t,_-._-
pc(t)
= 0 deg/s
(30)
where
Gp=
22.8(2)
(0)(I00)
,
_ (0.4)
GpfOp-_
Prefilters
!i /
22
102
Fp = [0.707, 2]
Fp = [0.707, 10]
Increasing the type qL# ofqLo from 1 to 2 will improve the performance of the software limiters, provided, of course, that stability
margins are not appreciably reduced by this increase, and they were
not in this application. Note that the zeros in Eq. (29) were chosen
to be a decade below the crossover frequencies
of the associated
(
= 0deg
pc(t)
96
V.
i_ii'':
_Yi:_,i'
VL
, ,{i
:{
),
i/_
/_ i !
Discussion
Acknowledgments
This research was supported by NASA Grant NAG1-1744 from
NASA Langley Research Center. The Technical Manager was Barton Bacon.
References
ISnell,
S. A., Enns,
Inversion
of Guidance,
Control,
and Dynamics,
3Domheim,
M. A., "Report Pinpoints Factors Leading to YF-22 Crash,"
Aviation Week and Space Technology,
Nov. 1992, pp. 53, 54.
4Anon., "Why the Gripen Crashed" Aerospace America,
Vol. 32, No. 2,
1994, p. 11.
5MeKay, K., "Summary
of an AGARD
Workshop
on Pilot Induced Oscillation"
AIAA Paper 94-3668, Aug. 1994.
6Bryson, A. E., Jr., and Ho, Y. C., Applied Optimal Control, Hemisphere,
New York, 1989.
7Kothare, M. V., Campo, P. J., Morari, M., and Nett, C. N., "A Unified
Framework
for the Study of Anti-Wmdup
Designs"
Automatica,
No. 12, 1994, pp. 1869-1883.
SKapesouris,
P., Athans, M., and Stein, G., "Design of Feedback
Vol. 30,
Control