9 Controller Discretization

9 Controller Discretization
In most applications, a control system is implemented in a digital fashion

on a computer. This implies that the measurements that are supplied to
the control system must be sampled. If the sampler has period T , then the
sampled value of the measurements are denoted by
yk = y(tk ), tk = kT, k = 0, 1, 2, 3, (1)
The output from the controller will take on discrete values which we denote
by uk . The actual input to the plant be controlled is usually obtained by
passing the sequence uk through a zero-order hold (ZOH). The output of the
ZOH is given by
u(t) = uk , tk t < tk+1
In this section we would like to determine mathematical models for the rela-
tionship between uk and yk and establish a technique for digital implemen-
tation of a controller design.
9.1 Discrete-Time Plant Model
Assume that the plant to be controlled is described by a state-space model

of the form
x = Ax + Bu, y = Cx (2)
Let us first of all note that the sampled value of the output y satisfies
yk = y(tk ) = Cx(tk ) = Cxk
Given the plant model in (2), the solution is

Z t
x(t) = eA(tt0 ) x(t0 ) + eA(t ) Bu( ) d
t0
Let t0 = tk , t = tk+1 , and tk+1 tk = T . Therefore,

Z t
k+1 A(t
xk+1 = e AT
xk + e k+1 ) Bu( ) d
tk
Z t
k+1 A(t
= eAT xk + e k+1 ) B d uk
tk
Letting, 0 = tk+1 , we have
Z t Z 0 Z T
k+1 A(t 0
e k+1 ) B d = eA (d 0 )B = eA d B
tk T 0
Therefore,
xk+1 = Ad xk + Bd uk (3)
yk = Cd xk
where
Cd = C
Ad = eAT
Z T
Bd = eA d B
0
This is termed the ZOH equivalent of the state-space system.
9.2 Stability of LTI Discrete-Time Systems
Consider
xk+1 = Ad xk , x0 given
It is readily verified that the solution is
xk = Akd x0
For simplicity, assume that Ad has distinct eigenvalues and permits the eigen-
decomposition
Ad = EE1
Therefore,
Akd = (EE1 )(EE1 ) (EE1 )

= Ek E1
and
xk = Ek E1 x0
Letting xk = E1 xk , it follows that
xk = k x0
where k = diag{ki }. If
|i | < 1, i = 1, , n (4)
then
ki 0 as k
k O as k
xk = Exk 0 as k
If |i | > 1 for some i, then
ki as k
||xk || as k
Theorem 1. If the eigenvalues of Ad lie within the open unit disk of the
complex plane, the discrete-time system is asymptotically stable.
Theorem 2. If any of the eigenvalues of Ad lie outside the unit disk of the
complex plane, then the discrete-time system is unstable.
Consider the ZOH equivalent in (3) and assume that A has distinct eigen-
values. If A has the eigendecomposition
A = EE1 , = diag{i }, i = i ji
then
Ad = eAT = EeT E1
Therefore
{Ad } = ei T = ei T eji T
Now, |ei T | = |ei T |. Hence if i < 0, then |ei T | < 1. Therefore, the
ZOH equivalent of an asymptotically stable continuous-time system is also
asymptotically stable.
9.3 Digital Control Design
There are two broad approaches to digital control design:

(i) direct discrete design using the ZOH equivalent as a model;
(ii) discretization of a continuous-time design based on the continuous-time
plant model.
Initially we consider (ii) using the bilinear transformation (also called Tustins
rule or the trapezoidal rule).
Assume that we have a continuous-time controller with the following model:
x = Ax + Bu, y = Cx (5)
We want to replace this with difference equations that relate the sampled
values yk and uk . Since y(tk ) = Cx(tk ) we have
yk = Cxk
Since x = Ax + Bu, it follows that

Z t Z t Z t
k+1 k+1 k+1
xk+1 xk = x dt = A x dt + B u dt
tk tk tk
Now, approximate the integrals on the right-hand side using the trapezoidal
rule:
Z t
k+1 T
x dt = (xk+1 + xk )
tk 2
Z t
k+1 T
u dt = (uk+1 + uk )
tk 2
Therefore,
T T
xk+1 xk = A [xk+1 + xk ] + B [uk+1 + uk ]
2 2
T T T
[1 A ]xk+1 = [1 + A ]xk + B [uk+1 + uk ]
2 2 2
T 1 T T T
xk+1 = [1 A ] [1 + A ] xk + [1 A ]1 B [uk+1 + uk ]
| 2 {z 2} | 2
{z 2}
Ad Bd
Consider the difference equations
zk+1 = Ad zk + Bd uk
zk+2 = Ad zk+1 + Bd uk+1
Note that setting xk = zk + zk+1 and adding the two equations gives the
desired one for xk . Hence
yk = Cxk = C(zk + zk+1 )
= Czk + C(Ad zk + Bd uk )
= C(1 + Ad ) zk + C B u
| {z d} k
| {z }
Cd Dd
Hence the discrete equivalent of the state-space model in Eq. (5) is
zk+1 = Ad zk + Bd uk (6)
yk = Cd zk + Dd uk (7)
where Ad , Bd , Cd , and Dd are defined as above.
9.4 Closed-Loop Discrete-Time Stability Analysis
Assume the plant is described by its ZOH equivalent:

xk+1 = Ad xk + Bd uk
yk = Cd xk
Assume that the controller is described by a discrete equivalent of the form
zk+1 = Ad zk + Bd yk
uk = Cd zk + Dd yk
Closing the loop gives:
xk+1 = Ad xk Bd Cd zk Bd Dd yk
= (Ad Bd Dd Cd )xk Bd Cd zk
Also,
zk+1 = Ad zk + Bd Cd xk
Combining these two gives

xk+1 Ad Bd Dd Cd Bd Cd xk
=
zk+1 Bd Cd Ad zk
| {z }
Acomp
For stability {Acomp } must lie within the unit disc.
9.5 Discrete-Time Optimal Control
Assume that the system to be controlled is described by difference equations

of the form
xk+1 = Axk + Buk , x0 = c (8)
Consider a performance index of the form
N 1
1 T 1 X T
J = 2 xN SxN + 2 xk Qxk + uTk Ruk (9)
k=0
where
R = RT > O, Q = QT O, S = ST O,
We would like to determine uk , k = 0, 1, 2, , N 1 to minimize J .
The state equation is treated as a constraint which we rewrite as
Axk + Buk xk+1 = 0 (10)
Let us adjoin (10) to (9) using Lagrange multipliers k , k = 1, , N :
1
NX
L=J + Tk+1 (Axk + Buk xk+1 ) (11)
k=0
Necessary conditions for optimality are

L
= 0, k = 1, , N
xk
L
= 0, k = 1, , N 1
uk
L
= 0, k = 1, , N
k
We have
1
NX
L = 21 xTN SxN + [ 12 xTk Qxk + 12 uTk Ruk + Tk+1 (Axk + Buk xk+1 )]
k=0
Hence the optimality conditions become

L
= SxN N = 0
xN
L
= Qxk + AT k+1 k = 0, k = 1, , N 1
xk
L
= Ruk + BT k+1 = 0, k = 1, , N 1
uk
L
= Axk1 + Buk1 xk = 0, k = 1, , N
k
We can rewrite these as
xk+1 = Axk + Buk , x0 = c (12)
k = AT k+1 + Qxk , N = SxN (13)
uk = R1 BT k+1 (14)
Substituting (14) into (12) leads to
xk+1 = Axk BR1 BT k+1 , x0 = c
k = AT k+1 + Qxk , N = SxN (15)
This is a discrete two-point boundary value problem. Since N = SxN , let
us assume
k = Pk xk , k = 1, , N (16)
with PN = S. Substituting (16) into (14) gives:
uk = R1 BT Pk+1 xk+1
= R1 BT Pk+1 (Axk + Buk )
(R + BT Pk+1 B)uk = BT Pk+1 Axk
uk = R1 T
k+1 B Pk+1 Axk
= Fk xk , Fk = R1 T
k+1 B Pk+1 A
where
Rk+1 = R + BT Pk+1 B
Hence, the optimal control is state feedback.
In order to determine Pk , let us substitute (16) into (15):
Pk xk = AT Pk+1 xk+1 + Qxk
= AT Pk+1 (Axk + Buk ) + Qxk
= AT Pk+1 Axk AT Pk+1 BR1 T
k+1 B Pk+1 Axk + Qxk
Since this must hold for all xk , the coefficient matrix on each side must match.
Therefore,
Pk = AT (Pk+1 Pk+1 BR1 T
k+1 B Pk+1 )A + Q (17)
This is called the discrete-time Riccati equation. It can be solved backwards
given the terminal condition PN = S. When Pk , k = N, N 1, , 0 is
known, the optimal feedback gains Fk can be determined.

9 Controller Discretization

Uploaded by

Copyright:

Available Formats

9 Controller Discretization

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

9 Controller Discretization

Uploaded by

Copyright:

Available Formats

9 Controller Discretization

In most applications, a control system is implemented in a digital fashion

yk = y(tk ), tk = kT, k = 0, 1, 2, 3, (1)

9.1 Discrete-Time Plant Model

Assume that the plant to be controlled is described by a state-space model

yk = y(tk ) = Cx(tk ) = Cxk

Given the plant model in (2), the solution is

Let t0 = tk , t = tk+1 , and tk+1 tk = T . Therefore,

This is termed the ZOH equivalent of the state-space system.

9.2 Stability of LTI Discrete-Time Systems

Akd = (EE1 )(EE1 ) (EE1 )

If |i | > 1 for some i, then

9.3 Digital Control Design

There are two broad approaches to digital control design:

Since x = Ax + Bu, it follows that

Consider the difference equations

9.4 Closed-Loop Discrete-Time Stability Analysis

Assume the plant is described by its ZOH equivalent:

Assume that the system to be controlled is described by difference equations

Necessary conditions for optimality are

Hence the optimality conditions become

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.