Você está na página 1de 8

11

Nonlinear Input/Output
Control: Volterra Synthesis
Patrick M. Sain 11.1 Introduction ...................................................................................... 1131
Raytheon Company, 11.2 Problem Definition Using Total Synthesis ............................................... 1131
E1 Segundo, California, USA 11.3 Plant Representation ........................................................................... 1132
11.4 Controller Design ............................................................................... 1133
11.5 Simplified Partial Linearization Controller Design ................................... 1134
11.6 SDOF Base-Isolated Structure Example .................................................. 1135
11.7 Conclusion ....................................................................................... 1138
References ......................................................................................... 1138

11.1 Introduction a based-isolated single degree of freedom structure with non-


linear hysteretic damping.
A power series expansion, provided it exists, often provides a
useful representation of a nonlinear plant. The same can be
said for the desired closed-loop input/output map of a feed- 11.2 Problem Definition Using
back control system. Given a Volterra series representation for Total Synthesis
each of these components, then Volterra feedback synthesis
(VFS) can be used to design and realize a nonlinear controller Based on Rugh's results (1981) using Volterra series to represent
that uses a finite consecutive number of Volterra kernels of the nonlinear systems, A1-Baiyat and Sain (1986, 1989) applied the
plant and desired closed-loop input/output map. A striking use of Volterra operators to nonlinear regulator design in
feature of this method is that it permits the specification of a the context of the total synthesis problem (TSP) in 1986.
nonlinear desired closed-loop behavior. The control design Since then, Sain et al. (1990, 1991) have used the TSP frame-
takes place in the frequency domain and is realized as an work to apply Volterra operators in nonlinear servomechanism
interconnected set of linear systems. The controller possesses design, and in 1995, Doyle et al. (1995) cast the method into a
an interesting recursive structure that is readily exploited to model-predictive control design. A1-Baiyat showed that the
produce an equivalent reduced-order realization. In fact, this Volterra operators comprising the controller can be realized
simplified controller implementation can be computed to an as interconnections of linear systems, and Sain (1997) derived
arbitrarily high order in an automatic manner. an equivalent reduced-order realization. For brevity, a partial
The VFS approach is reasonably general; the specific devel- linearization regulator design (Sain et al., 1997b) is presented
opment presented herein draws upon the total synthesis prob- herein. The approach extends to servomechanisms, and the
lem (TSP) framework, and so this chapter begins with concise general controller design is given by Sain (1997).
descriptions of the TSP paradigm, Volterra plant representa- Let R, U, and Y denote real vector spaces of dimensions
tion, and controller synthesis. The length of the general for- p, m, and p, respectively representing the spaces of requests,
mulas at this point is only to be admired; the intrepid reader plant inputs, and plant outputs. Let P: U --~ Y denote an input
will note, no doubt with some relief, that the following section output description of a nonlinear plant. Define the desired
describes an equivalent simplified reduced-order implementa- closed-loop response to a command by T: R ~ Y and the
tion of the controllers represented by these formulas, complete desired plant input for a command by M: R -~ U. The oper-
with block diagrams. An example application is provided using ators P, T, M, and E are assumed to have Volterra series

Copyright (c) 2005 by Academic Press. 1131


All rights of reproduction in any form reserved.
1132 Patrick M. Sain

//
J2(xo) = ~ Cli(xo, U0)X Ii]. (11.4)
I M i=1
I E
In these equations, x = x - x 0 , h= u-Uo, and x Ill =
x ® - . . ® x, the /-fold Kronecker tensor product of x with
itself. To reduce notational complexity, ® is assumed to have
precedence over matrix and scalar multiplication. Suppressing
the explicit dependence on (x0, u0) and if flJ denotes Ufl/Ox j,
FIGURE 11.1 TSP Regulator Configuration then the following results:

n n--1
/ R\
/ \ flJ = Z A)ix[i] ÷ Z Dj(i-1)x[i] @ ~" (11.5)
/ x i=) i=j

lil A12AI:I
M / \ T
/ x
/ "x
/ \
/
p
\
-.,q
d x[21 A22 A2n ~[2]
u )v dt " z . , . .

FIGURE 11.2 Commutative Diagram J L 0 A,.., J L ~.[~1


Dll D12 ''" Dl(n-1) ] V ~:

representations at the point or in the region of operation. An


illustration of these operators for a regulator design is given in
+
i1 ®h (11.6)

Figure 11.1 In general, a pair (M,T) is desired such that 0 ... Dn(. 1) J
T = P o M and the diagram in Figure 11.2 commutes. In the D10
sequel, T is given, and the objective is to find and realize a
controller G: Y --~ U such that M = G o E, where E: R ~ Y; + h.
in this case, Y is the space of output errors.

11.3 P l a n t R e p r e s e n t a t i o n Equations .11.5 and 11.6 apply for 2 ~ j _ < n, or, more
compactly, x = Ax + Dx ® h+ Bh, y = CYc,~(0) = x0. In the
Within the TSP framework, consider linear analytic plants, sequel, the notation Ak denotes the matrix partition composed
that is, plants for which the state and output equations are of the leflmost k partitions of the top k rows of A; Ck denotes
given by: the leftmost k partitions of C; and Dk denotes the leftmost
k - 1 partitions in the top k rows of D, with D1 = D10.
k=f(x)+g(x)u, x(0)=Xo. (11.1) For a multiple-input, multiple-output, finite-dimensional,
causal, time-invariant plant, define the homogenous multi-
y=h(x). (11.2) linear Voherra operator as:

In equations 11.1 and 11.2, x E X; u E U; y E Y; f , g: X--+X; t "ri--1


g(x): U-+X; h: X-+Y; and f , g, and h are analytic in x. Such Pi[u(t)] =[.,. [ pi(T1 . . . . . Ti)u(t-- T1) Q " " Q u(t-Ti)dTi.., d%,
plants can be represented using a so-called bilinear approxi- 0 0
mation, obtained by denoting the approximation of the state (11.7)
and output equations )~(x, u) and ]~(x), respectively and by
taking a multivariable Taylor series expansion about the oper- where t _> "F1 ~ " ' " ~ T i Z 0 and where Pi is its ith Volterra
ating point (x0, u0), truncating all terms of order higher than kernel. A Volterra representation then has the form (A1-Baiyat
n, yielding: and Sain, 1986, 1989):
// oo
~(X0, U 0 ) = ~ _ . A l i ( x o , U0)-~ [i] y(t) = Z Pi[u(t)]. (11.8)
i=1 i=1
n--1
+ ~ Dli(xo, U0)X [i] @ h + Dlou. (11.3) The convolutional nature of this representation makes it
i=1 expedient to work in the transform domain. Define the
11 Nonlinear Input~Output Control: Volterra Synthesis 1133

multidimensional Laplace transform of the/th Volterra kernel where the following is true:
as (Bussgang et al., 1974):
t 'rl Ti 1

P,(Mj, lr(t)l ..... Mj, lr/t)l) = JI... J p,< ..... Ti)Mj,(FI)


Pi(Sl ..... si) . . . . pi('rl ..... Ti)e (s'~i+"+si~i)d%... dTi. (1 1.9)
O0 0
o o
@ "'" @ Mj~(ri)d'ri... d'rl. (11.18)
M-Baiyat and Sain (1986) showed that for a stationary time
invariant, bilinear system of the form given above, where Equating output equation expressions and suppressing the
P~ = ((Sl + . ' - + sj) I - A i ) - l D i , the transforms of the first argument [r(t) ] yields:
three kernels are as follows:
c;T; . . . . /l÷-..+j
~ o,,, . . . . . . Mjl) • (11.19)
'~itivij
Pl(s) = c , pll • (11.10) j=l i=1 jl=l ji=l
--2 --1
P2(Sl, S2) = C2P2{P1 @ Ira}. (11.11)
Equating powers of c on both sides of equation 11.19 gives:
--3 --2 --1
P3(Sl, s2, s3) = C3P3{[P2{P1 @ Im}] @ Ira}. (11.12)
T1 = P1M1. (11.20)
The general form is the following: T2 = P1M2 + P2(Mb M~). (11.21)
T3 = P1M3 + P2(Mb M2) + P2(M2, M1) + P3(M1, M1, M1). (11.22)
-- " -- "-- --2 --1
Pj(s, . . . . . sj) = CjP}{[F}-]{[-.. P2{P, @ Ira}"" ] @Im}] @ Ira}.
For Ki = ~ = 1 kj, K0 =
(11.13)
0:

~ ( i-j÷l i j-kl÷2 i-~j 2-1


11.4 Controller Design j=2 kl=l k2= ' kj 1=1
\
(11.23)
Now, the goal is to formulate an expression for the Volterra
kernels of the controller G in the frequency domain in terms of
Pj(Mk~, Mk . . . . . . Mkj ~, Mi-.j-1)~.
/
the kernels of the plant, given above, and those of the desired
closed-loop map, T. The development follows that given in A1- Applying the multidimensional Laplace transform yields:
Baiyat Sain (1986). Similarly to the Volterra operator Pi [u(t)],
define Volterra operators associated with the maps T and M rl(s) = PI(S)MI(S). (11.24)
such that:
r2(s1, s2) = PI(S 1 q- $2)M2(s1, s2)
+ P2(Sl, s2)[Ml(Sl) ® M(s2)]. (11.25)
y(t) = ~ Tj[r(t)]. (11.14) T3(s> s2, s3) = Pl(Sl + s2 + s3)M3(sl, s2, s3) (11.26)
j=l
+ P2(sl, s2 + s3)[MI(Sl) @ M2(s2, s3)]
O0
+ P2(sl + 82, s3)[M2(sl, s2) @ MI(S3)]
u(t) = Z Mk[r(t)]. (11.15)
k=l + P3(s~, s2, s3)[M~(sl) ® M~(s2) ® M~(s3)]. (11.27)

To obtain a relation between the operators Pi, Tj, and Mk, Ti(st . . . . . si) = PI(Sl + ' " + si)Mi(Sl . . . . . si)
replace the request signal r(t) by cr(t), where c is an arbitrary ~2~ / i j + l i j kl+2 i-K) 2-1
constant; because Volterra operators are multilinear, equating + ZZ
expressions for y yields: j=2 kl=l k2=l k3_1=1
Pj(Sl + " " + sK~, sK~+l+ "'" + s~2. . . . . s~i_~+l + "'" + si)
dTj[r(t)] = Z Pi dMj[r(t)] . (11.16) [Mk~(Sl..... s~,) ® M<(s~+l . . . . . sK2) ® " " ®
j=l i=1 j=l
Mi ~, ,(sKi 141..... si)] ~. (11.28)
Let ri denote r(t - Ti) and define: J

Observe that if the pair (M1 (s), T1(s)) is chosen for P1 (s), then
~_~ Pi "
dMj[r(t)] = z.. ~~J'+'"+J'Pi~I M j, [r(t)], . . . ,
... ~-"
one can proceed to the design equation involving P2(Sl, s2),
i=1 jl =1 ji=l
where the pair (M2(Sl, s2) and T2(Sl, s2)) is chosen and so on.
Mj~[r(t)]), (11.17) To complete the design, consider the synthesis of the controller
1134 Patrick M. Sain

G that realizes the desired mappings (M, T) shown in Figure A nice feature of these equations is their recursive nature. Once
11.1 Note that the similarity of the relation M = G o E per- Ml(s) has been found, it can be used to obtain an expression
mits expressions for the operators Mj to be obtained in a for M2(s> s2) and so forth. Under partial linearization, the
manner similar to the one for the operators T/. Also from operators are represented by Ek = 0, k > 1. Assuming El(s)
Figure 11.1, note that e = r - y, which can be written as: and Pl(s) to be invertible, then the kernels of the controller
G are given by:

Ei[r(t)] = I[r(t)] - Z ~ [ r ( t ) ] , (11.29) G1(s) = p;l (s) T1(s)E1 1(s). (11.34)


i-1 i=1
Ge(sa, s2) = -P['(sa + s2)P2(s> si)[G~(s,) ® Ga(s2)]. (11.35)
where I is the identity operator, and, thus, E1(s) = I - T~(s). G3(s1, 52, 53) = -Pll(51 -[- $2 ~- $3){P2(s1, $2 -}- 53)[G1(s1)
In addition, for i > 1, El(S, . . . . . si) = -Ti(s~ . . . . . si). A direct
® G2($2, S3)] q- P2(sl q- S2, S3)[G2(51, $2) ® Gl(S3)]
relation between the Volterra kernels of the operators T, M ,
and G can be found if the design is well-posed on the first q- P3($1, s2, $3)[GI(Sl) @ G1($2) ® G1($3) ] }. (11.36)
order or linear level, implying that El(S) has an inverse. As T is
usually open to choice, the latter requirement is reasonable. In general:

Gi(Sl . . . . . si) = -Pll(Sl + . . . + si)


11.5 Simplified Partial Linearization ~ { i j+l i-j k1+2 i /~j_2--1
Controller Design Z Z Z
j=2 kl=l k2=l kj-l=l
Consider designing a feedback system for a given nonlinear Pj(Sl + ... + s,~, snt+l + ... + &. . . . . . S/gj-l+l -~- " ' ' -~- Si)
system so that Ti = O, 1 < i < n, where Ti is the /th Volterra
[Gk, (Sl . . . . . S~;1) @ Gk2(Stvl+l..... S~2)
kernel of the closed loop. Such a design is called partial
linearization, and it has the desirable attribute of greatly
Q ... ~ Gi ,~j-,(s,%~-I . . . . . si)] ~. (11.37)
lessening the complexity of the controller design equations. J
Both the general controller design and the partial linearization
controller design, as derived using the equations above, yield Again, note the recursive nature of the controller design equa-
realizations with repeated structures. Collecting terms with like tions. Once Gl(s) has been computed, it can be used to find
coefficients yields the simplified controller design. The result is G2(sl, s2) and so on.
presented here for the partial linearization controller, with the The next step in the realization process is to substitute the
general result given by Sain (1997). expression for the plant kernels Pi in the frequency domain.
Suppose for T that the second and higher order kernels are Given G~ = Gi(sj, sj+l . . . . . sj+i-1), then:
zero. Then, if Pl(s) inverts, the following results:
G] = -(ClP~)-1C2~{ [P~G~] ® G~}. (11.38)
Ma(s) = P;~(s)T~(s). (11.30)
¢] = _(c,~e)-l{ca~{ [8¢:] ® ¢2} + <p~{ [p~¢]] ® <~}
M2(s,, s2) = -P[a(s~ + s2)P2(s~, s2)[M~(s~) ® Ml(S2)]. (11.31) + C3P~{ [/)22{[P~ G~] ® ¢~}] ® 0~}}. (11.39)
M3($1, s2, s3) = --Pll(51 -I- $2 + 53){P2($1, $2 -}- $3)[M1(s1) ~__~{ i-j+l i j-~l+2 i j-~2+3 i ~J 2 1

® M2($2, 53)]-}-P2(Sl q-$2, 53)[M2($1, $2)@ M1($3)] j=2


Z
kl=l
Z
k2=l
Z
k3=l
... ki-l=l
Z
+ P3(s~, s2, s3)[M~(Sl) ® M~(s2) ® Ma(s3)]}. (11.32) -~;l-1 G~l+l \ ]
%

In general, the following is true: ¢t~,_2+1 -t~) 1+1~. (11.40)


JJ
M/(sl . . . . . si) = - P [ I ( s l + . . . + si)
~L~{ i j+l i-j-kl+2 i-~j 2-1 Each of equations 11.38 through 11.40 can be realized as a set
ZZ Z of interconnected linear subsystems. These realizations possess
a number of interesting features that are nicely illustrated using
j=2 kl=l k2=l kj l=l
block diagrams. In doing so, certain notational conveniences
Pj(sl + ... + sn~, s,q+l + ... + &2. . . . . s~i_~+l + . . . + Si) are employed.
[mk~(s~ . . . . . s~,) ® Vk2(<+~ . . . . . < ) First, the superscripts on the quantities P/; and G / i n equa-
tions 11.38 through 11.40 are redundant and can be neglected
® . . . ® m~ ~ ~(s~ ,+~..... ~,)] ~. (11,33) without loss (Sain, 1997). The proof lies in observing that the
ordering of terms in a controller design equation, together
11 Nonlinear Input~Output Control: Volterra Synthesis 1135

with the established use of parentheses and brackets, are suffi- The members of the common set F0, (namely c~/) and [3/)),
cient to determine what the superscripts should be. are computed and stored the first time they are encountered
Second, if e E Y represents the output error of the closed and then recalled as needed. In the sequel, the equation
loop, and 811 is its Laplace transform, then let 8~i] denote the g/) = - p [ I c j P j [ 3 i j and the following identifications are con-
/-fold tensor product of 811 with itself, then define: venient:

i
(11.41) gi = Z - p [ 1 c j P j [ 3 / ) (11.44)
k=l j=2
i
making [3il the Laplace transform of the /th order controller = Zg/). (11.45)
output. j--2
Finally, the layout of the block diagrams follows certain
conventions. If the controller design equations are fully The above algorithm for calculating the output of the simpli-
expanded and each written on one line, then their addends, fied partial linearization controller takes full advantage of the
as read from left to right, appear in the block diagrams pro- recursive nature of the design equations, reducing the number
ceeding from top to bottom. of states and floating-point computations. Furthermore, the
The fundamental observation behind the simplification pro- simplified form is amenable to the construction of general
cess is that quantities present in the realization for G) reappear software routines, capable of computing the controller output
in the realizations for G1, j > i. The motivation behind the for an arbitrary value of the controller order n.
simplification process is that in computing the output of a The simplified block diagrams for the second third-order
given controller component G), the most efficient algorithm controller components are shown in Figures 11.3 and 11.4.
is to compute these repeated quantities only once, store them A general form for the ith order component is shown in
in memory, and then reuse them as needed. The reduction Figure 11.5.
process is thus a problem of identifying a set F of elements
common to a sequence of controller components G], i = 1,
2, 3. . . . . n, and determining which ones to save for later reuse.
11.6 SDOF Base-Isolated Structure
In computing the n components &lr~(i)~ 1 < i < n, of the Example
output of an nth order controller, many quantities are com-
puted repeatedly. Therefore, a system of notation is now intro- Consider a single degree of freedom (SDOF) structure sitting
duced with the primary goal of eliminating redundant on a base isolation system consisting of hysteretically damped
calculations by identifying members of the common set F and bearings, and let the structure be subject to ground acceler-
the secondary goal of further simplifying the controller output ation 2g(t) and a control force f ( t ) supplied by a hydraulic
calculation. The identification of F presented here is not actuator. Such a system is shown in Figure 11.6 The forces
unique. Therefore, the notation F0 will be used to denote the exerted by the individual bearings are modeled collectively.
particular identification described herein and to distinguish it The equation of motion for the horizontal displacement of
from the general notion of the common set represented by F. the mass m of the structure is as follows:
Here and in the sequel, superscripts are suppressed.
The first members of F0 have already been introduced,
fll 1
n a m e l y [3ii. Given o~ij E Fo and 2 < j < i defined by oLij =
[9j-113(i 1)(j-l), then:

i-j+l FIGURE 11.3 SimplifiedG2 Realization


[3ij = Z OL(i-k+l)J @ [3kl" (11.42)
k=l
[521
The simplified partial linearization nth order controller output 0~22 )~
can now be calculated as follows. First, compute the output of 13. ~ I
the first order controller, [3n = p[l T1(1 - T1)-1811, and then
the output of the nth order controller for n _> 2 is computed
using: Pll /
i
(11.43)
i=2 j=2 FIGURE 11.4 SimplifiedG3 Realization
1136 Patrick M. Sain

13(n-I)I

13(n-2)1
?32
i331
O~(n-2)2
1321
~
O~(n 1)2
J311
13(n- 1)1 -P1-1C2/52]gn2

1?(3i+(i1)i4
O~ii
1)1
+
)(~

~(i-1)1
(i +])i

~21
(Z(n 1)/ + )(~
~11 y
~(n-I)(i- 1 ) ~ _Pl_lCi~ilgni X
[321
O~(n1)(n-1~)
1311 |
1 3 ( n - 1 ) ( n - 2 ~ _ m -lt", ~ I .,2
1311 q'n ~'loc~n 1~~ "%CJ13n(n 1)1 1 "-'n 1 n- llg(n_ l) (n_ l)'%
13(n 1)(n-1~ _Pl.lOn~ n gnn )~

FIGURE 11.5 General Form for Simplified Realization

(f , i xp >

FIGURE 11.6 A Base-Isolated SDOF Structure of Mass rnl. The structure is supported by hysteretically damped bearings bl and b2 that are
subject to ground acceleration ~g and control force f.

fit) = rn[~p(t) + ~g(t)] + cic;(t) + kxp(t) + Q[x(t), k(t)]. provided in Table 11.1 are meant to represent the behavior of
(11.46) the system with the actuator and bearings in place. The hys-
teretic restoring force Q generated by the base isolating bear-
Descriptions of the variables and parameter values are given in ings is given by:
Table 11.1 Both the bearings and the actuator affect the stiff-
ness and damping coefficients c and k, and it should be
Q[xp(t), kp(t)] -- c ~ xp(t) + (1 - a)Fyz(t), (11.47)
understood that the values identified for these parameters
11 Nonlinear Input~Output Control: Volterra Synthesis 1137

TABLE 11.1 Parameters for the SDOF System TABLE 11.3 Parameters for the Actuator

Variable Description Value Variable Description Value

c Damping coefficient 6001Tkg/s Af Cross-sectional area of actuator m 2


f Force applied by actuator (N) B Bulk modulus of hydraulic fluid N/m 2
k Stiffness coefficient 3 X 106Tr 2 kg/s2 f Force applied to SDOF structure N
m Mass of structure 3000 kg by actuator
Q Hysteretic restoring force of base-isolation N g Proportional feedback gain 2.5
bearing k¢ Flow-pressure coefficient m 4 .s/kg
xg Horizontal ground acceleration m/s 2 kq Flow gain m2/s
xp Horizontal displacement of the mass rn m xp Horizontal displacement m
fp Natural undamped frequency 5 Hz V Characteristichydraulic fluid volume m3
{p Damping ratio 1% 2BAfkqG/V Displacementcoefficient 3.6484 x 10s N/m/s
2Bkc/V Force feedback coefficient 66.67 1/s
2BA2/V Velocityfeedback coefficient 2.1891 x 107N/m
TABLE 11.2 Parameters for the Hysteretically Damped Base-Isolation
System

Variable Description Value


structure yields the following system of equations. For con-
A Hysteresisloop-shaping parameter 1 venience, the explicit dependence on time is suppressed:
Force of equivalent hysteretic damper 40 k
tl Parameter controlling the smoothness of the 2
transition from elastic to plastic response k
m+ r'YmJxo-~ --mc xp' + l f ( t ) - ( 1 - o ~ ) F Y z - 2g.(ll.52)
m
xp Horizontal displacement m
Y Yield displacement of equivalent hysteretic damper 1mm
2BAfk@lf 2BA} . 2Bkc 2BAfkq'lf (11.53)
z Hysteretic displacement / = - V xp - ~ - X p - ~ - f ( t ) q ~ u.
Post yielding to preyielding stiffness ratio 0.2
Hysteresisloop-shaping parameter -0.25 = (Az + azn)2p. (11.54)
Y Hysteresisloop-shaping parameter 0.75
Given z w i t h n = 2, with z0 = z(t¢) and Xp0 = xp(t¢) (where t¢
represents the most recent point in time where or(t) switched
where z(t) is a dimensionless quantity representing hysteretic value yields (for o- > 0 and o- < 0) closed-form expressions for
displacement (Fan et al., 1988; Sain et al., 1997) satisfying the z are readily obtained (Sain, 1997).
first-order differential equation: An example application of a third-order VFS partial linear-
ization regulator design for the SDOF structure described
Y~(t) = ylkp(t)llz(t)ln-lz(t) - ~2p(t)lz(t)l'+Akp(t). (11.48) previously is simulated in equations 11.55 to 11.57, using the
simplified controller design derived above. Choosing the state
Table 11.2 lists descriptions of the variables and parameter vector [ x 1 x 2 x3] I = [Xp Xp f]', where z can be expressed as

values. The actuator force f applied to the SDOF structure is previously noted, depending on the sign of ¢, the resulting
governed by the equation (Dyke et al., 1995; DeSilva, 1989): state-space description has the form y = Xl, and the following
results:
j'(t) 2BAfkq~lf 2Bk~ , 2BA}
-- V [u(t) - Xp(t)] - ~ - f ( t ) - V kp(t). 21 = x2. (11.55)
(11.49)
22= ( kmq-ot .Fr_~ X l - - - xc 2 - - - - x13 - (1-or) Fy z. (11.56)
IN// m m m
The u(t) is the control signal generated by the controller.
Descriptions of the quantities in equation 11.49 are provided BAIvk 2BA} 2Bkc 2BAfkqyf
23 - 2 q'Yf xl - - V x2 - ~ x 3 + 7/ u. (11.57)
in Table 11.3.
Defining Az = A f3z = ~, and % = ~ and using the fact that
]z] = z sgn z, then: This example represents a preliminary test of the control
system to demonstrate that the actuator can be c o m m a n d e d
to move the first floor of the structure to a specific location.
~ ( t ) = [Az - (f3z + Yz sgn kp sgn z ) z ' ( t ) l k p ( t ) (11.50)
Under the hypothesis that the VFS control designs work
= [A~ + a(t)z~(t)]2p(t), (11.51) to make small errors smaller, a c o m m a n d request r = x~ =
0.004 m was specified, and a constant offset of )I0 = 0.001 m,
where o'(t) = -[[3 z + % sgn 2p(t) sgn z(t)]. Substitution of representing a biased position sensor, was added to the meas-
the expression for Q into the equation o f m o t i o n of the ured o u t p u t of the plant. Regulation about a nonzero set point
1138 Patrick M. Sain

recursive in nature and lends itself to an automated design


3.4 ~ il and simulation procedure. The procedure is applied to a regu-
lator design for a base-isolated SDOF structure with hysteretic
damping.
3.3 .............. ~ ............... ~................ i................ i................ i ............... i............... i................ i................ i ..............

0 ~ i f i i i i i i
/~ i i i i i i
References
¢- 3.2
............. •I!~"~ ............... i ................ ::............... i ............... ! ............... i ............... i ................ i .............. M-Baiyat, S.A., and Sain, M.K. (1989). A Volterra method for non-
.0 i I~.I linear control design. Preprints IFAC Symposium on Nonlinear Con-
trol System Design, 76-81.
A1-Baiyat, S.A., and Sain, M.K. (1986). Control design with transfer
.ii .............................. ......... ............... ::............... i ............... ';................ ...............................
functions associated to higher order Volterra kernels. Proceedings of
the 25th IEEE Conference on Decision and Control, 1306-1311.
Bussgang, J.J., Ehrman, L., and Graham, J. (1974). Analysis of non-
linear systems with multiple inputs. Proceedings of the IEEE 62,
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 1088-1119.
Time [sec] DeSilva, C.W. (1989). Control sensors and actuators. Englewood Cliffs,
NJ: Prentice Hall.
FIGURE 11.7 Position of First Floor Following Initial Transient for Doyle, III, EJ., Ogunnaike, B.A., and Pearson, R.K. (1995). Nonlinear
Linear Controller Design (Dashed) and Third-Order Controller model-based control using second-order Volterra models. Automa-
Design (Solid) tica 31,697-714.
Dyke, S.J., Spencer, Jr., B.E, Quast, E, and Sain, M.K. (1995). Role of
control-structure interaction in protective system design. ASCE
was chosen because the higher order Taylor series coefficients Journal of Engineering Mechanisms 121,322-338.
of the control model are zero if ~1 = 0. Primarily for computa- Fan, EG., Ahmadi, G., and Tadjbakhsh, I.G. (1988). Base isolation of
tional convenience, the desired closed-loop map Tl(s) was a multi-story building under a harmonic ground motion A com-
given four poles at s = -1000 in this example. parison of performances of various systems. Technical Report.
A result is shown in Figure 11.7 The initial transient re- NCEER-88-0010. National Center for Earthquake Engineering
sponse was omitted so as not to obscure the effect of the higher Research, State University of New York at Buffalo.
Rugh, W.J. (1981). Nonlinear system theory: The Volterra/Wiener ap-
order controller. The response of the closed-loop system,
proach. Baltimore, MD: Johns Hopkins University Press.
meaning the position of the first floor, xl, is shown for the
Sain, EM. (1997). Volterra control synthesis, hysteresis models, and
linear controller design and the third-order controller design. magnetorheological structure protection. Ph.D. dissertation. Depart-
Note that the magnitude of the oscillations in the response as ment of Electrical Engineering., University of Notre Dame, Notre
the output tends toward the commanded position are much Dame, Indiana.
smaller for the third-order design when compared to those Sain, EM., Sain, M.K., and Spencer, Jr., B.E (1997). Models for
produced by the linear controller design. hysteresis and application to structural control. Proceedings of the
American Control Conference, 16-20.
Sain, EM., Sain, M.K., and Spencer, Jr., B.E (1997). Volterra feed-
11.7 Conclusion back synthesis: A systematic algorithm for simplified Volterra con-
troller design and realization. Proceedings of Thirty-Fifth Annual
Allerton Conference on Communication, Control, and Computing,
Volterra feedback synthesis is shown to provide a capability for
1053-1062.
specifying nonlinear input output behavior for a closed-loop Sain, EM., Sain, M.K., and Michel, A.N. (1991). Nonlinear model-
feedback system. The procedure is developed for the class of matching design of servomechanisms. Proceedings of the First IFAC
linear analytic plants that admit Volterra series representation, Symposium on Design Methods of Control Systems, 594-599.
and a simplified means of systematically realizing the Volterra Sain, EM., Sain, M.K., and Michel, A.N. (1990). On coordinated
kernels and computing the output of the resulting controller is feed-forward excitation of nonlinear servomechanisms. Proceedings
provided, up to an arbitrary order. The simplified form is of the American Control Conference, 1695-1700.

Você também pode gostar