Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

Proceedings of the American Control Conference

Arlington, VA June 25-27, 2001

Control Allocation and Compensation for Over-Actuated


Systems with Non-Linear Effectors
Ram Venkataraman David B. Doman, AFRL/VACA
Veridian Engineering Control Theory and Optimization Branch
Dayton OH, 45431-1289 Wright-Patterson AFB, OH 45433-7531
Email: ram.venkataraman@wpafb.af.mil Email: david.doman@wpafb.af.mil

Abstract
Control allocation techniques for inversion-based control
laws have evolved that ensure that commands to the
control effectors do not violate rate or position limits
of the individual effectors. These allocation techniques I
typically assume that all effector dynamics are fast rel-
ative to the system being controlled and are therefore F i g u r e 2: Schematic of control effector model
neglected. Unfortunately there are practical cases where
this assumption breaks down and it becomes desirable to 2 C o n s t r a i n e d Control Allocation
compensate for lags between the commands to the effec- It is assumed that a constrained control allocation al-
tors and the effector response. In this paper, a technique gorithm [l,21 is being used that ensures that a pseudo-
for compensating for individual effector dynamics while command vector d = g(uc) is delivered by some com-
respecting actuator constraints is proposed. bination of control effector deflections uc subject to
U 5 U, 5 ii where 11 and a are the most restrictive of
-
1 Introduction the effector rate and position limits at any given instant.
The actual control effector position for each actuator U
Previously, researchers have considered controller design is physically limited and always satisfies
issues for plants with magnitude and rate-limited inputs.
In this paper, we consider controller design for the ac- Umm I U I Umaz; (1-a)
tuators themselves, and thus our "system" is magnitude
and rate-limited at the output.
-T I U 5 T. (1-b)
where U is the magnitude limie and T is the rate limit
Recent advances in constrained control allocation tech- for the desired actuator position signal uc.Without loss
niques [l,21 have enabled dynamic inversion control laws of generality, we consider the case when U m i n = 0 as in
131 to control over-actuated systems while ensuring that Figure 2.
control effector rate and position limits are not violated.
Figure 2 shows the actuator schematic [4]. One can ver-
Much of this work has been aimed at the control of
ify that the output signal in the block diagram indeed
aerospace vehicles. This technique invariably assumes
satisfies the constraints. The control allocation algo-
that the actuator dynamics are sufficiently fast relative
rithm [3] takes care of these constraints l - a - l-b) and
t o the plant dynamics such that the commanded effector
only those signals that satisfy the above constraints are
positions are attained instantaneously. Dynamic com-
passed as desired signals U = ( . ) t o the actuator. However
pensation for the linear portion of these effector dynam-
for the sake of analysis, it is important t o note that the
ics must be done carefully. Simply inserting the compen-
input signal space t o the actuator itself is much larger
sation into the pseudo-command ensures that rate and
than the set of signals that satisfy the constraints l-a -
position limits will not be violated; however, each effec-
1-b). We come to this point later.
tor will receive the same amount of compensation. Since
each effector may have a different bandwidth, the-com-
pensation must be applied individually. Simple filtering 2.1 C o m p e n s a t o r design
of the control allocation command may cause actuator Assuming that the effector position is a measurable
rate and position limits to be violated. Closed loop dy- quantity, U may be fed-back and used t o form an er-
namic compensation for slow linear dynamics of a rate ror signal e = U - U, that may be used for dynamic
and position limited effector is explored in this paper compensation. We show in this paper that a simple
(Figure 1). compensator K ( s ) = shown in Figure 1 satisfies

0-7803-6495-3/01/$10.00 0 2001 AACC 1812


~ ~~

Figure 1: Schematic of the actuator and compensator

the requirements. We show that the error between the Lk[O,CQ)}. Now the control allocation algorithm only
desired and actual effector positions can be made arbi- produces signals U, that satisfy the constraints (1-a -
trarily small by choosing k large enough. To be precise, (1-b). However, we show below that the closed-loop sys-
we claim the following: tem is capable of tracking signals that satisfy the con-
straints at almost all points (except a set of measure
1. We choose a1 = a to simplify the analysis. If
zero).
the initial states are zero and U,(.) is the worst-
case commanded input with uc(0) = 0, the error From (2), it is clear that le(t)l is maximized if IUc(t)l =
le(t)l = Iuc(t) -u(t)l 5 ,& for all t. Thus to keep T for almost all t E I . In other words, the worst case
the error le(t)l 5 E, we simply choose k 2 A. error is obtained when U,(.) is defined as follows. First
2. If the initial states are non-zero and U,(.) = 0, pick T such that T $ 5 u m a XThen. define :
then U + 0 asymptotically. T
uc(t) = r t , 0 5 t 5 -;
3. If ~ ~ ( #0 0,
) then we have le(t)l 5 & asymptot- m
2
"
ically. = U,(') T), - 1
< t 5 T (3-b)
- T(t -
1

2 2 -
Due to space constraints, we will indicate the main = uc(t - nT), for nT 5 t 5 ( n + 1)T(3-c)
points of the proofs without giving all the details. As
indicated in the list above, there are several cases to where n E N. As e(0) = 0 , for t E J = I n [0, g],
study. First of all we point out that given a commanded T
deflection U,(.) that satisfies the magnitude and rate e(t)= (1 - exp(-kat)) (4)
constraints, there are two important cases to consider.
The first is when uc(0) = 0 (item 1 above), and the Next we compute the signals U(.), and w(.) in J (Figures
second is when uc # 0 (item 3). Other cases pertain to 1 and 2).
non-zero initial states (item 2) and choice of the com-
pensator zero, which is not done here due to space con-
straints.
Proofs : Observing Equation (5-b), we can say that Iv(t)( <
It is assumed that u(t) = .li(t)for all t - this models the for all t E [0, $1 if u ( t ) never achieves its upper bound
physical phenomenon. However, w(t) may not be equal U,,,. we can conclude from the definition of T ,Equa-
to w(t) for all t. tions (3-a) and (5-a), that for all t E [O, $1, U ( . ) < U m a x
1. As u c ( 0 )= 0 and the initial states are zero, we have
e(0) = uc(0)- u(0) = 0. Thus at t = 0, v(0) = 0. Thus V t E [0, $1, both U(.) and &(.) take values in
Thus by continuity, there exists an open interval of time (:, f ) and [O,umaz)respectively. Observe that v ( t ) =
I = ( 0 , i ) such that W ( T ) = v ( ~ )for all T E I . This kv(t).
means that for t E I we can ignore the non-linearities
and consider the linear transfer functions alone. After Next, we analyze the signals in [$,TI. By direct com-
some calculations we see that for t E I : e = U, - kae putations, we see that V t E [:,TI; Iv(t)l < f . We also
and so, observe that v ( t ) = k e ( t ) for t E [:,TI.
e ( t )=
I' exp (-ka(t - T)).',(T)&
We note from Equation 2 that the closed-loop compen-
(2) By similarly analyzing the signals in the time intervals
[nT , ( n+ 1) TI for n E IN,we arrive at the same conclu-
sion that the signals w and 6 do not violate their limits
sator plus effector system is well-defined for inputs U,(.) (except for h(0) = O!). This conclusion does not depend
that belong in the set F = {fIf
E L&[O,co)&f E on the parameter k (as long as k > O!).

1813
The magnitude of the error is determined by k. If we
desire the error to be bounded by E then we need to pick
k such that

k?: a € (6)
2. If the initial states are non-zero and U , = 0, then we
have two cases. First is the case when the rate and mag-
nitude limits are not reached. In this case, as we have a
stable linear system, the states decay to zero asymptot-
ically. This implies that the output U + 0 asymptoti-
cally. In the second case, when the rate and magnitude
limits are reached, then we can show with some calcu-
ations that the states decay so that the signals v and
zi eventually lie in ( - f , f ) and (0, umax)respectively. (a) No compensation
Then we have the same situation as the first case. Thus
again the output U + 0 asymptotically. 094 , , , ~, , , , , ,
3. Here we consider the case when uc(0) # 0. For
simplicity, we examine the case when U,(.) = c is a
0 *a
I
constant function of time. Here subtleties concerning
the anti-windup circuit in Figure 2 enter, and so we
consider the generic case of c < umaXonly. Consider
the case when U is greater than f during an interval
of time ( t l ,tz). The case when v becomes less than -f
is handled similarly. Note that it is possible (if c is
small enough) that the v and zi always lie in (- f , f ) and
(0, umax)respectively. In ( t l ,t 2 ) we can write down the
system equations and then after some minor calculations
we observe that there must exist a time t , E ( t l ,t 2 ) such (b) With compensation
that U(-) decreases at least linearly in (t*,t 2 ) . Thus t 2 Figure 3: Desired vs actual control effector positions.
2
cannot be 00 and thus U ( - ) 5 for t 2 t 2 . Then we have
the same situation as in item 1. 0
3 Simulation results and Dynamics, vol. 17, pp. 330-336, Mar-Apr 1994.
In this section, we present simulation results that sup- [2] J. M. Buffington, “ Modular Control Law De-
port the conclusions made earlier. For one of the con- sign for the Innovative Control Effectors (ICE) Tail-
trol effectors, the parameters where : T = 2.1; a = less Fighter Aircraft Configuration 101-3,” Tech. Rep.
7; umaZ = 1.14. Figure 3(a) shows the desired versus AFRL-VA-WP-TR-1999, Air Force Research Labora-
actual control effector positions without any compen- tory, 1999.
sator. Figure 3(b) shows the same signals with the com- [3] D. B. Doman, A. D. Ngo, D. B. Leggett, M. A.
pensator discussed above with Ic = 7.5. It was further Saliers, and M. Pachter, “Development of a Hybrid Di-
verified that the relationship between k and E holds. rect/Indirect Adaptive Control System for the X-33 ,”
4 Conclusion in Proceedings of the 2000 Guidance, Navigation and
In this paper, we study a compensator design prob- Control Conference, AIAA 00-4156, August 2000.
lem for a system with rate and magnitude limits on [4] M. R. Anderson and A. B. Page, “Unified Pilot-
the output. The desired signal also satisfies the same Induced Oscillation Theory Volume 111: P I 0 Analysis
constraints. The problem is t o design a compensator Using Multivariable Methods,” Tech. Rep. WL-TR 96-
that accounts for the dynamics of the system. We have 3030, Wright Laboratory, 1996.
studied a simple compensator design and derived a re-
lationship between the gain of the compensator and the -
desired error between the input and output signals. Sim-
ulations show the effectiveness of the compensator.

References
[l] W. C. Durham, “Constrained control allocation:
Three moment problem,” Journal of Guidance, Control

1814

You might also like