0% found this document useful (0 votes)
17 views54 pages

Sigma

Uploaded by

ESLAM salah
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
17 views54 pages

Sigma

Uploaded by

ESLAM salah
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 54

System Identification and Robust Control

Lecture 7: Introduction to Multivariable Control

Eugenio Schuster

[email protected]
Mechanical Engineering and Mechanics
Lehigh University

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 1 / 54
Introduction to Multivariable Control
Transfer functions for MIMO systems [3.2]

u -G G z
+ b-
u- v G p y-
1
- G2 - +6 1
z G 
2
(a) Cascade system (b) Positive feedback system

Figure 1: Block diagrams for the cascade rule and the feedback rule

1 Cascade rule - (Figure 1(a):


G = G2 G1
2 Feedback rule - (Figure 1(b)):
v = (I − L)−1 u, L = G2 G1
3 Push-through rule:
G1 (I − G2 G1 )−1 = (I − G1 G2 )−1 G1
Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 2 / 54
Introduction to Multivariable Control
MIMO Rule: Start from the output, move backwards. If you exit
from a feedback loop then include a term (I − L)−1 where L is the
transfer function around that loop (evaluated against the signal
flow starting at the point of exit from the loop).
Example:
z = (P11 + P12 K(I − P22 K)−1 P21 )w (1)

P22 

c+- K
-P21 - ? p-P12
+

w -P11 - c?
+ -z
+
Figure 2: Block diagram corresponding to (1)

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 3 / 54
Introduction to Multivariable Control
Negative feedback control systems
d2 d+1 y
r -
+ c -
K
u- c +-
+ ?
G
+ c?
- q -
-6

Figure 3: Conventional negative feedback control system

L is the loop transfer function when breaking the loop at the plant output:

L = GK (2)

Accordingly

S = (I + L)−1
output sensitivity (3)

T = I − S = (I + L)−1 L = L(I + L)−1
output complementary sensitivity (4)

LO ≡ L, SO ≡ S and TO ≡ T .
Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 4 / 54
Introduction to Multivariable Control
LI is the loop transfer function at the input to the plant:

LI = KG (5)

Input sensitivity:

SI = (I + LI )−1
Input complementary sensitivity:

TI = I − SI = LI (I + LI )−1

Some relationships:

(I + L)−1 + (I + L)−1 L = S + T = I (6)

G(I + KG)−1 = (I + GK)−1 G (7)


GK(I + GK)−1 = G(I + KG)−1 K = (I + GK)−1 GK (8)
−1 −1 −1 −1
T = L(I + L) = (I + L ) = (I + L) L (9)
Rule to remember: “G comes first and then G and K alternate in sequence”.
Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 5 / 54
Introduction to Multivariable Control
Multivariable frequency response [3.3]

G(s) = transfer (function) matrix


G(jω) = complex matrix representing response
to sinusoidal signal of frequency ω

Note: d ∈ Rm and y ∈ Rl

d - G(s) -y

Figure 4: System G(s) with input d and output y

y(s) = G(s)d(s) (10)

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 6 / 54
Introduction to Multivariable Control
Sinusoidal input to channel j

dj (t) = dj0 sin(ωt + αj ) (11)

starting at t = −∞. Output in channel i is a sinusoid with the same frequency

yi (t) = yi0 sin(ωt + βi ) (12)

Amplification (gain):
yio
= |gij (jω)| (13)
djo

Phase shift:
βi − αj = ∠gij (jω) (14)
gij (jω) represents the sinusoidal response from input j to output i.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 7 / 54
Introduction to Multivariable Control
Example: 2 × 2 multivariable system, sinusoidal signals of the same frequency ω
to the two input channels:
   
d1 (t) d10 sin(ωt + α1 )
d(t) = = (15)
d2 (t) d20 sin(ωt + α2 )

The output signal


   
y1 (t) y10 sin(ωt + β1 )
y(t) = = (16)
y2 (t) y20 sin(ωt + β2 )

can be computed by multiplying complex matrix G(jω) by complex vector d(ω):

y(ω) = G(jω)d(ω)
y10 ejβ1 d10 ejα1
   
y(ω) = , d(ω) = (17)
y20 ejβ2 d20 ejα2

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 8 / 54
Introduction to Multivariable Control
Directions in multivariable systems [3.3.2]
SISO system (y = Gd, d and y are scalars):
The gain at a given frequency ω is

|y(ω)| |G(jω)d(ω)|
= = |G(jω)|
|d(ω)| |d(ω)|

The gain depends on ω, but is independent of |d(ω)|.

MIMO system (y = Gd, d and y are vectors):


We need to “sum up” magnitudes of elements in each vector by use of some norm
sX q
kd(ω)k2 = |dj (ω)|2 = d210 + d220 + · · · (18)
j

sX q
ky(ω)k2 = |yi (ω)|2 = 2 + y2 + · · ·
y10 (19)
20
i

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 9 / 54
Introduction to Multivariable Control
The gain of the system G(s) is
p
ky(ω)k2 kG(jω)d(ω)k2 y 2 + y20
2 + ···
= = p 210 (20)
kd(ω)k2 kd(ω)k2 d10 + d220 + · · ·
The gain depends on ω, and is independent of kd(ω)k2 . However, for a MIMO
system the gain depends on the direction of the input d.
Example: Consider the five inputs (all kdk2 = 1)
         
1 0 0.707 0.707 0.6
d1 = , d2 = , d3 = , d4 = , d5 =
0 1 0.707 −0.707 −0.8
For the 2 × 2 system  
5 4
G1 = (21)
3 2

The five inputs dj lead to the following output vectors


         
5 4 6.36 0.707 −0.2
y1 = , y2 = , y3 = , y4 = , y5 =
3 2 3.54 0.707 0.2

with the 2-norms (i.e. the gains for the five inputs)
ky1 k2 = 5.83, ky2 k2 = 4.47, ky3 k2 = 7.30, ky4 k2 = 1.00, ky5 k2 = 0.28
Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 10 / 54
Introduction to Multivariable Control
Introduction to Multivariable Control
8
σ̄(G1 ) = 7.34

∥y∥2 /∥d∥2
6

0 σ(G1 ) = 0.27
−5 −4 −3 −2 −1 0 1 2 3 4 5
d20 /d10
Figure 5: Gain kG1 dk2 /kdk2 as a function of d20 /d10 for G1 in (21)
Figure
The maximum5:value
Gainof the∥G 1 d∥
gain /∥d∥
in 2(20) 2 as
as the a function
direction ofisdvaried,
of the input 20 /d10is
the maximum singular value of G,
for G1 in (7.21)
kGdk2
max = max kGdk2 = σ̄(G) (22)
d6=0 kdk2 kdk2 =1

whereas the minimum gain is the minimum singular value of G,


kGdk2
min = min kGdk2 = σ(G) (23)
d6=0 kdk2 kdk2 =1
Lecture 7 (Weeks 13-14) – p. 14

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 11 / 54
Introduction to Multivariable Control
Eigenvalues are a poor measure of gain [3.3.3]
Example:      
0 100 0 100
G= ; G = (24)
0 0 1 0
Both eigenvalues are equal to zero, but gain is equal to 100.

Problem: eigenvalues measure the gain for the special case when the inputs and
the outputs are in the same direction (in the direction of the eigenvectors).

Let d be an eigenvector of G, then y = Gd = λd. Then,

kyk/kdk = kλdk/kdk = |λ|

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 12 / 54
Introduction to Multivariable Control
For generalizations of |G| when G is a matrix, we need the concept of a matrix
norm, denoted kGk. Two important properties must be satisfied:

Triangle inequality:
kG1 + G2 k ≤ kG1 k + kG2 k (25)
Multiplicative property:
kG1 G2 k ≤ kG1 k · kG2 k (26)

As we may expect, the magnitude of the largest eigenvalue, ρ(G) = |λmax (G)|
(the spectral radius), does not satisfy the properties of a matrix norm.
Several matrix norms are possible (see Book’s Appendix)
Frobenius norm kGkF
Sum norm kGksum
Maximum column sum kGki1
Maximum row sum kGki∞
Maximum singular value kGki2 = σ̄(G)
The latter three norms are induced by a vector norm (reason for subscript i). The
induced 2-norm (maximum singular value) is of special interest because it is the
the ratio of the energy of the input/output signals.
Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 13 / 54
Introduction to Multivariable Control
Singular value decomposition [3.3.4]
Any matrix G may be decomposed into its singular value decomposition,

G = U ΣV H (27)

Σ is an l × m matrix with k = min{l, m} non-negative singular


values, σi , arranged in descending order along its main diagonal;
U is an l × l unitary matrix of output singular vectors, ui ,
V is an m × m unitary matrix of input singular vectors, vi ,
q q
σi (G) = λi (GH G) = λi (GGH ) (28)

(GGH )U = U ΣΣH , (GH G)V = V ΣH Σ (29)

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 14 / 54
Introduction to Multivariable Control
Input and output directions. The column vectors of U , denoted ui , represent
the output directions of the plant. They are orthogonal and of unit length
(orthonormal), that is
p
kui k2 = |ui1 |2 + |ui2 |2 + . . . + |uil |2 = 1 (30)

uH
i ui = 1, uH
i uj = 0, i 6= j (31)

The column vectors of V , denoted vi , are orthogonal and of unit length, and
represent the input directions.

G = U ΣV H ⇒ GV = U Σ (V H V = I) ⇒ Gvi = σi ui (32)
If we consider an input in the direction vi , then the output is in the direction ui .
Since kvi k2 = 1 and kui k2 = 1 σi gives the gain of the matrix G in this direction.

kGvi k2
σi (G) = kGvi k2 = (33)
kvi k2

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 15 / 54
Introduction to Multivariable Control
Maximum and minimum singular values. The largest gain for any input
direction is
kGdk2 kGv1 k2
σ̄(G) ≡ σ1 (G) = max = (34)
d6=0 kdk2 kv1 k2
The smallest gain for any input direction is
kGdk2 kGvk k2
σ(G) ≡ σk (G) = min = (35)
d6=0 kdk2 kvk k2
where k = min{l, m}. For any vector d we have
kGdk2
σ(G) ≤ ≤ σ̄(G) (36)
kdk2
Define u1 = ū, v1 = v̄, uk = u and vk = v. Then
Gv̄ = σ̄ū, Gv = σ u (37)
v̄ corresponds to the input direction with largest amplification, and ū is the
corresponding output direction in which the inputs are most effective. The
directions involving v̄ and ū are sometimes referred to as the “strongest”,
“high-gain” or “most important” directions.
Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 16 / 54
Introduction to Multivariable Control
Example:  
5 4
G1 = (38)
3 2
The singular value decomposition of G1 is
     H
0.872 0.490 7.343 0 0.794 −0.608
G1 =
0.490 −0.872 0 0.272 0.608 0.794
| {z } | {z } | {z }
U Σ VH
 
0.794
The largest gain of 7.343 is for an input in the direction v̄ = , the smallest gain
0.608
 
−0.608
of 0.272 is for an input in the direction v = . Since in (38) both inputs affect
0.794
both outputs, we say that the system is interactive.

The system is ill-conditioned, that is, some combinations of the inputs have a strong
effect on the outputs, whereas other combinations have a weak effect on the outputs.
Quantified by the condition number; σ̄/σ = 7.343/0.272 = 27.0.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 17 / 54
Introduction to Multivariable Control
Example: Distillation process. Steady-state model of a distillation column
 
87.8 −86.4
G= (39)
108.2 −109.6

Since the elements are much larger than 1 in magnitude there should be no
problems with input constraints. However, the gain in the low-gain direction is
only just above 1.
     H
0.625 −0.781 197.2 0 0.707 −0.708
G= (40)
0.781 0.625 0 1.39 −0.708 −0.707
| {z } | {z } | {z }
U Σ VH

The distillation process is ill-conditioned, and the condition number is


197.2/1.39 = 141.7. For dynamic systems the singular values and their associated
directions vary with frequency (Figure ??).

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 18 / 54
Introduction to Multivariable Control
Singular values for performance [3.3.5]
SVD provides insight into the directionality of MIMO systems. But the maximum
singular value is very useful in terms of frequency-domain performance and
robustness. We here consider performance.

Performance measure for SISO systems:

|e(ω)|/|r(ω)| = |S(jω)|

.
Generalization for MIMO systems ke(ω)k2 /kr(ω)k2

ke(ω)k2
σ(S(jω)) ≤ ≤ σ̄(S(jω)) (41)
kr(ω)k2

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 19 / 54
Introduction to Multivariable Control
For performance we want gain ke(ω)k2 /kr(ω)k2 small for any direction of r(ω)

σ̄(S(jω)) < 1/|wP (jω)|, ∀ω ⇔ σ̄(wP S) < 1, ∀ω


⇔ kwP Sk∞ < 1 (42)

where the H∞ norm is defined as the peak of the maximum singular value of the
frequency response

kM (s)k∞ = max σ̄(M (jω)) (43)
ω

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 20 / 54
Introduction to Multivariable Control
Typical singular values of S(jω) in Figure 7.
Typical singular values of S(jω) in Figure 6.

σ̄(S)
0
10
σ̄(S) σ(S)
Magnitude
σ(S)
−2
10
Design 1:
Design 2:
−2 0 2
10 10 10
Frequency [rad/s]

Figure Figure
7: Singular values
6: Singular values of S for aof
2 ×S forwith
2 plant a RHP-zero
2 × 2 plant wit
RHP-zero

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 21 / 54
Introduction to Multivariable Control

Bandwidth, ωB : frequency where σ̄(S) crosses √1 = 0.7 from below.


2

It is then understood that the bandwidth is at least ωB for any direction of


the input (reference or disturbance) signal.

Since S = (I + L)−1 , the singular values inequality


1
σ(A) − 1 ≤ σ̄(I+A) −1 ≤ σ(A) + 1 yields

1
σ(L) − 1 ≤ ≤ σ(L) + 1 (44)
σ̄(S)
1
low ω : σ(L)  1 ⇒ σ̄(S) ≈ σ(L)
high ω: σ̄(L)  1 ⇒ σ̄(S) ≈ 1

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 22 / 54
Introduction to Multivariable Control
Poles [4.4]
Definition: The poles pi of a system with state-space description
ẋ = Ax + Bu (45)
y = Cx + Du (46)
are the eigenvalues λi (A), i = 1, . . . , n of the matrix A. The pole or characteristic
∆ Qn
polynomial φ(s) is defined as φ(s) = det(sI − A) = i=1 (s − pi ). Thus the poles
are the roots of the characteristic equation

φ(s) = det(sI − A) = 0 (47)
Poles and stability [4.4.1]
Theorem
A linear dynamic system ẋ = Ax + Bu is stable if and only if all the poles are in
the open left-half plane (LHP), that is, <{λi (A)} < 0, ∀i. A matrix A with such a
property is said to be “stable” or Hurwitz.
Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 23 / 54
Introduction to Multivariable Control
Poles from transfer functions [4.4.3]
Theorem
The pole polynomial φ(s) corresponding to a minimal realization of a system with
transfer function G(s), is the least common denominator of all
non-identically-zero minors of all orders of G(s).

A minor of a matrix is the determinant of the matrix obtained by deleting


certain rows and/or columns of the matrix.

The notation Mcr is used to denote the minor corresponding to the deletion
of rows r and columns c in G(s).

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 24 / 54
Introduction to Multivariable Control
Example [4.7]: Consider the 2 × 2 system, with 2 inputs and 2 outputs,
 
1 s−1 s
G(s) = (48)
1.25(s + 1)(s + 2) −6 s − 2

The minors of order 1 are the four elements all have (s + 1)(s + 2) in the
denominator.
Minor of order 2
(s − 1)(s − 2) + 6s 1
det G(s) = 2 2 2
= 2
(49)
1.25 (s + 1) (s + 2) 1.25 (s + 1)(s + 2)

Least common denominator of all the minors:

φ(s) = (s + 1)(s + 2) (50)

Minimal realization has two poles: s = −1; s = −2.


Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 25 / 54
Introduction to Multivariable Control
Example [4.8]: Consider the 2 × 3 system, with 3 inputs and 2 outputs,
1
G(s) = ∗
(s + 1)(s + 2)(s − 1)

(s − 1)2
 
(s − 1)(s + 2) 0
∗ (51)
−(s + 1)(s + 2) (s − 1)(s + 1) (s − 1)(s + 1)

Minors of order 1:
1 s−1 −1 1 1
, , , , (52)
s + 1 (s + 1)(s + 2) s − 1 s + 2 s + 2

Minor of order 2 corresponding to the deletion of column 2:

(s − 1)(s + 2)(s − 1)(s + 1) + (s + 1)(s + 2)(s − 1)2


M2 = =
((s + 1)(s + 2)(s − 1))2
2
= (53)
(s + 1)(s + 2)

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 26 / 54
Introduction to Multivariable Control
The other two minors of order two are
−(s − 1) 1
M1 = , M3 = (54)
(s + 1)(s + 2)2 (s + 1)(s + 2)

Least common denominator:

φ(s) = (s + 1)(s + 2)2 (s − 1) (55)

The system therefore has four poles: s = −1, s = 1 and two at s = −2.

Note MIMO-poles are essentially the poles of the elements. A procedure is needed
to determine multiplicity.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 27 / 54
Introduction to Multivariable Control
Zeros [4.5]
SISO system: Zeros zi are the solutions to G(zi ) = 0.
Example:
s+2
Y = 2 U
s + 7s + 12
Compute the response when

u(t) = e−2t , y(0) = 0, ẏ(0) = −1


1
L{u(t)} =
s+2
s2 Y − sy(0) − ẏ(0) + 7sY − 7y(0) + 12Y = 1
2
s Y + 7sY + 12Y + 1 = 1
⇒ Y (s) = 0

Assumption: g(s) has a zero z, g(z) = 0. Then for input u(t) = u0 ezt the
output is y(t) ≡ 0, t > 0. (with appropriate initial conditions)
In general, zeros are values of s at which G(s) loses rank.
Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 28 / 54
Introduction to Multivariable Control
Zeros from state-space realizations [4.5.1]
The state-space equations of a system can be written as:
     
x 0 sI − A −B
P (s) = , P (s) = (56)
u y C D
The zeros are then the values s = z for which the polynomial system matrix P (s)
loses rank, resulting in zero output for some non-zero input
  
zI − A −B xz
=0
C D uz
Non-trivial solution: The zeros are the solutions of
 
zI − A −B
det =0
C D

MATLAB:
zero = tzero(A,B,C,D)
Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 29 / 54
Introduction to Multivariable Control
Zeros from transfer functions [4.5.2]
Definition: zi is a zero of G(s) if the rank of G(ziQ
) is less than the normal rank
nz
of G(s). The zero polynomial is defined as z(s) = i=1 (s − zi ) where nz is the
number of finite zeros of G(s).

Theorem
The zero polynomial z(s), corresponding to a minimal realization of the system, is
the greatest common divisor of all the numerators of all order-r minors of G(s),
where r is the normal rank of G(s), provided that these minors have been
adjusted in such a way as to have the pole polynomial φ(s) as their denominators.

Remark. This gives the same result as the Smith McMillan form introduced by
Rosenbrock (See Section 4.6).

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 30 / 54
Introduction to Multivariable Control
Example [4.9]:
 
1 s−1 4
G(s) = (57)
s+2 4.5 2(s − 1)
The normal rank of G(s) is 2.
2
−18
Minor of order 2: det G(s) = 2(s−1)
(s+2)2
= 2 s−4
s+2
.
Pole polynomial: φ(s) = s + 2.
Zero polynomial: z(s) = s − 4.

Note: Multivariable zeros have in general no relationship with the zeros of the
transfer function elements.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 31 / 54
Introduction to Multivariable Control
Example [4.7 Continued]:
 
1 s−1 s
G(s) = (58)
1.25(s + 1)(s + 2) −6 s−2

The normal rank of G(s) is 2.


(s−1)(s−2)+6s 1
Minor of order 2: det G(s) = 1.252 (s+1)2 (s+2)2 = 1.252 (s+1)(s+2)
2
Pole polynomial: φ(s) = 1.25 (s + 1)(s + 2)
Zero polynomial: z(s) = 1 ⇒ no multivariable zeros.

Example [4.10]:  
s−1 s−2
G(s) = (59)
s+1 s+2
The normal rank of G(s) is 1.
No value of s for which both elements of G(s) (minors of order 1) become zero
⇒ G(s) has no zeros.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 32 / 54
Introduction to Multivariable Control
Directions of poles and zeros
Let G(s) = C(sI − A)−1 B + D.

Zero directions [4.5.3]. Let G(s) have a zero at s = z. Then G(s) loses rank at
s = z, and there exist non-zero vectors uz and yz such that

G(z)uz = 0, yzH G(z) = 0 (60)

uz = input zero direction (uH


z uz = 1)
yz = output zero direction (yzH yz = 1)
yz gives information about which output (or combination of outputs) may be
difficult to control.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 33 / 54
Introduction to Multivariable Control
Pole directions [4.4.4]. Let G(s) have a pole at s = p. Then G(p) is infinite,
and we may write
G(p)up = ∞, ypH G(p) = ∞ (61)
up = input pole direction (uH
p up = 1)
yp = output pole direction (ypH yp = 1)

Singular Value Decomposition. The zero/pole directions may in principle be


obtained from an SVD of
G(z/p) = U ΣV H
uz = last column in V , yz = last column of U
(corresponding to the zero singular value of G(z))
up = first column in V , yp = first column of U
(corresponding to the infinite singular value of G(p))

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 34 / 54
Introduction to Multivariable Control
Example: Plant in (57) has a RHP-zero at z = 4 and a LHP-pole at p = −2.
 
1 3 4
G(z) = G(4) =
6 4.5 6
   H
1 0.55 −0.83 9.01 0 0.6 −0.8
= (62)
6 0.83 0.55 0 0 0.8 0.6

The zero input and output directions are associated with the zero singular value of
G(z), i.e.    
uz = −0.800.60
y z = −0.83
0.55
(63)

For pole directions consider


 
1 −3 +  4
G(p + ) = G(−2 + ) = 2 (64)
 4.5 2(−3 + )

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 35 / 54
Introduction to Multivariable Control
The SVD as  → 0 yields
   H
1 −0.55 −0.83 9.01 0 0.6 −0.8
G(−2 + ) = 2
 0.83 −0.55 0 0 −0.8 −0.6
   
0.60 −0.55
up = −0.80
yp = 0.83
(65)

Note 1: Locations of poles and zeros are independent of input and output
scalings, their directions are not. Thus, the inputs and outputs need to be scaled
properly before making any interpretations based on pole and zero directions.

Note 2: This method is not numerically reliable.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 36 / 54
Introduction to Multivariable Control
Remarks on poles and zeros [4.6]
1. For square systems the poles and zeros of G(s) are “essentially” the poles
and zeros of det G(s).

This fails when zero and pole in different parts of the system cancel when
forming det G(s).
 
(s + 2)/(s + 1) 0
G(s) = (66)
0 (s + 1)/(s + 2)

det G(s) = 1, although the system obviously has poles at −1 and −2 and
(multivariable) zeros at −1 and −2.

2. System (66) has poles and zeros at the same locations (at −1 and −2).
Their directions are different. They do not cancel or otherwise interact.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 37 / 54
Introduction to Multivariable Control

3. There are no zeros if the outputs contain direct information about all the
states; that is, if from y we can directly obtain x (e.g. C = I and D = 0).

4. Zeros usually appear when there are fewer inputs or outputs than states

5. Moving poles. (a) Feedback control (G(I + KG)−1 ) moves the poles. (b)
Series compensation (GK, feedforward control) can cancel poles in G by
placing zeros in K (but not move them). (c) Parallel compensation (G + K)
cannot affect the poles in G.

6. Moving zeros. (a) With feedback, the zeros of G(I + KG)−1 are the zeros
of G plus the poles of K. , i.e. the zeros are unaffected by feedback. (b)
Series compensation can counter the effect of zeros in G by placing poles in
K to cancel them, but cancellations are not possible for RHP-zeros due to
internal stability (see Section 40). (c) The only way to move zeros is by
parallel compensation, y = (G + K)u, which, if y is a physical output, can
only be accomplished by adding an extra input (actuator).

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 38 / 54
Introduction to Multivariable Control
Stability [4.3]
Definition
A system is (internally) stable if none of its components contains hidden
unstable modes and the injection of bounded external signals at any place in the
system results in bounded output signals measured anywhere in the system. The
word “internal” implies that all the states must be stable not only inputs/outputs.

Definition
State stabilizable, state detectable and hidden unstable modes. A system is
state stabilizable if all unstable modes are state controllable. A system is state
detectable if all unstable modes are state observable. A system with unstabilizable
or undetectable modes is said to contain hidden unstable modes.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 39 / 54
Introduction to Multivariable Control
Internal stability of feedback systems [4.7]
Note: Checking the pole of S or T is not sufficient to determine internal stability
Example [4.14]: (Figure 7). In forming L = GK we cancel the term (s − 1) (a RHP
pole-zero cancellation) to obtain

k s
L = GK = , and S = (I + L)−1 = (67)
s s+k
S(s) is stable, i.e. transfer function from dy to y is stable. However, the transfer
function from dy to u is unstable:

k(s + 1)
u = −K(I + GK)−1 dy = − dy (68)
(s − 1)(s + k)
Consequently, although the system appears to be stable when considering the output
signal y, it is unstable when considering the “internal” signal u, so the system is
(internally) unstable.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 40 / 54
Introduction to Multivariable Control

du dy
K G
r + + y
+d
- - k(s+1) -?d u- s−1 -?d q-
s(s−1) + s+1 +
-
6

Figure 7: Internally unstable system

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 41 / 54
Introduction to Multivariable Control

y du
6
+
q - −K -?d
+

d
+
G  q
+6
dy u
?

Figure 8: Block diagram used to check internal stability of feedback system

For internal stability consider


u = (I + KG)−1 du − K(I + GK)−1 dy (69)
−1 −1
y = G(I + KG) du + (I + GK) dy (70)

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 42 / 54
Introduction to Multivariable Control

Theorem
The feedback system in Figure 8 is internally stable if and only if all four
closed-loop transfer matrices in (69) and (70) are stable.

Theorem
Assume there are no RHP pole-zero cancellations between G(s) and K(s), i.e. all
RHP-poles in G(s) and K(s) are contained in the minimal realizations of GK and
KG. Then, the feedback system in Figure 8 is internally stable if and only if one
of the four closed-loop transfer function matrices in (69) and (70) is stable.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 43 / 54
Introduction to Multivariable Control
Implications of the internal stability requirement

1 If G(s) has a RHP-zero at z, then L = GK, T = GK(I + GK)−1 ,


SG = (I + GK)−1 G, LI = KG and TI = KG(I + KG)−1 will each have a
RHP-zero at z.

2 If G(s) has a RHP-pole at p, then L = GK and LI = KG also have a


RHP-pole at p, while S = (I + GK)−1 , KS = K(I + GK)−1 and
SI = (I + KG)−1 have a RHP-zero at p.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 44 / 54
Introduction to Multivariable Control
Introduction to MIMO robustness [3.7]
Motivating robustness example 1: Spinning Satellite [3.7.1]
Angular velocity control of a satellite spinning about one of its principal axes:
s − a2
 
1 a(s + 1)
G(s) = 2 ; a = 10 (71)
s + a2 −a(s + 1) s − a2
A minimal, state-space realization, G = C(sI − A)−1 B + D, is
 
  0 a 1 0
A B  −a 0 0 1 
=  (72)
C D  1 a 0 0 
−a 1 0 0
Poles at s = ±ja For stabilization:
K=I
 
1 1 a
T (s) = GK(I + GK)−1 = (73)
s+1 −a 1
Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 45 / 54
Introduction
Introduction to Multivariable
to Multivariable Control Control
The stability
Nominal distillation process
(NS). is ill-conditioned,
Two closed −1 and
loop poles at s =and the condition
number is 197.2/1.39 = 141.7

0 a
. For dynamic
 
1 a
 systems

−1 0
the
singular
A cl values and their associated directions vary with
= A − BKC =
−a 0

−a 1
=
0 −1
frequency (Figure 6).
Nominal performance (NP). Figure 9(a)
2
10
2
σ̄(G)

Magnitude
10
Magnitude

1
10

σ̄(G) 0 σ(G)
0
10
10

σ(G)
−1 −2
10 10 −4 −2 0
−2 0 2
10
Frequency
10
[rad/s]
10
10 10 10
Frequency [rad/s]
(a) Spinning satellite in (7.71) (b) Distillation process in (7.76)

Figure 6:9:Typical
Figure plotsofofsingular
Typical plots singular values
values

Lecture 7 (Weeks 13-14) – p. 25/67


Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 46 / 54
Introduction to Multivariable Control

σ(L) ≤ 1 ∀ω poor performance in low gain direction


g12 , g21 large ⇒ strong interaction

Robust stability (RS).

Check stability: one loop at a time.


z1 w1
6
-+c - p-
- G
-+c 6 - p -
-
6


K


Figure 10: Checking stability margins “one-loop-at-a-time”

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 47 / 54
Introduction to Multivariable Control

z1 ∆ 1
= L1 (s) = ⇒ GM = ∞, P M = 90◦ (74)
w1 s
1 L1 (s) 1
Note that T11 (s) = s+1 = 1+L1 (s) ⇔ L1 (s) = s

Good Robustness? This result suggests YES ... but NO


Consider perturbation in each feedback channel

u01 = (1 + 1 )u1 , u02 = (1 + 2 )u2 (75)


 
0 1 + 1 0
B =
0 1 + 2
Closed-loop state matrix:
    
0 a 1 + 1 0 1 a
A0cl = A − B 0 KC = −
−a 0 0 1 + 2 −a 1

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 48 / 54
Introduction to Multivariable Control

Characteristic polynomial:

det(sI − A0cl ) = s2 + (2 + 1 + 2 ) s +
| {z }
a1

+ 1 + 1 + 2 + (a2 + 1)1 2
| {z }
a0

The perturbed system is stable if and only if both the coefficients a1 and a0
are positive. The system is always stable if we consider uncertainty in only
one channel at a time: (−1 < 1 < ∞, 2 = 0) and (1 = 0, −1 < 2 < ∞).
This confirms GM=∞.

But only small simultaneous changes in the two channels: for example, let
1 = −2 , then the system is unstable (a0 < 0) for
1
|1 | > √ ≈ 0.1
a2 +1
Summary. Checking single-loop margins is inadequate for MIMO problems.
Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 49 / 54
Introduction to Multivariable Control
Motivating robustness example 2: Distillation Process [3.7.2]
Idealized dynamic model of a distillation column,
 
1 87.8 −86.4
G(s) = (76)
75s + 1 108.2 −109.6
(time is in minutes).
2.5 Nominal plant:
Perturbed plant:
2
1.5
y1
1
0.5
y2
0
0 10 20 30 40 50 60
Time

Figure 11: Response with decoupling controller to filtered reference input


r1 = 1/(5s + 1). The perturbed plant has 20% gain uncertainty as given by (79).
Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 50 / 54
Introduction to Multivariable Control
Inverse-based controller or equivalently steady-state decoupler with a PI controller
(k1 = 0.7)
 
k1 k1 (1 + 75s) 0.3994 −0.3149
Kinv (s) = G−1 (s) = (77)
s s 0.3943 −0.3200

Nominal performance (NP).


0.7
GKinv = Kinv G = I
s
First order response with time constant 1/0.7 = 1.43 (Fig. 11).
Nominal performance (NP) achieved with decoupling controller.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 51 / 54
Introduction to Multivariable Control
Robust stability (RS).
s 1
S = SI = I; T = TI = I (78)
s + 0.7 1.43s + 1
In each channel: GM=∞, PM=90◦ .

Input gain uncertainty (75) with 1 = 0.2 and 2 = −0.2:

u01 = 1.2u1 , u02 = 0.8u2 (79)

 
1 + 1 0
L0I (s) 0
= Kinv G = Kinv G =
0 1 + 2
 
0.7 1 + 1 0
(80)
s 0 1 + 2

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 52 / 54
Introduction to Multivariable Control
Perturbed closed-loop poles are

s1 = −0.7(1 + 1 ), s2 = −0.7(1 + 2 ) (81)

Closed-loop stability as long as the input gains 1 + 1 and 1 + 2 remain positive


⇒ Robust stability (RS) achieved with respect to input gain errors for the
decoupling controller.

Robust performance (RP).


Performance with model error poor (Fig. 11)
SISO: NP+RS ⇒ RP
MIMO: NP+RS 6⇒ RP
RP is not achieved by the decoupling controller.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 53 / 54
Introduction to Multivariable Control
Robustness conclusions [3.7.3]
Multivariable plants can display a sensitivity to uncertainty (in this case input
uncertainty) which is fundamentally different from what is possible in SISO
systems.

Prof. Eugenio Schuster ME 450 - System Identification and Robust Control Spring 2020 54 / 54

You might also like