AUTOMATIC CONTROL

Collection of Formulae

Department of Automatic Control

Lund University
June 2011
2
Matrix theory

Notations
Matrix of order m x n
a11 a12 ⋅⋅⋅ a1n 
 

 
a21 a22 ⋅⋅⋅ a2n 

 
 
A=
 
..
 

 



 . 



 
am1 am2 ⋅⋅⋅ amn

Vector of dimension n
x1 
 

 
 x2 

 

x=
 
 .. 
 




 . 


  
xn

Transpose
B = AT
bi j = a ji
( AB )T = B T AT
The matrix is symmetric if ai j = a ji .

Determinant
 
 a11 a12 ⋅ ⋅ ⋅ a1n 

 a21 a22 ⋅ ⋅ ⋅ a2n 
 
det A = p Ap =  . 
 ..


 
 an1 an2 ⋅ ⋅ ⋅ ann 
If A is of order 2 x 2, then

det A = a11 a22 − a12 a21

In general
n
X
det A = ai j (−1)i+ j det Mi j
i=1
n
X
= ai j (−1)i+ j det Mi j
j =1

where Mi j is the matrix one obtains if row i and column j are removed
from the matrix A.

Inverse
A−1 A = AA−1 = I (det A ,= 0)
If A is of order 2 x 2, then

−a12 
 
−1 1  a22
A =  
det A −a21
 
a11

1
In general,
1
A−1 = CT
det A
where the elements in C are given by

ci j = (−1)i+ j det Mi j

Eigenvalues and eigenvectors

The eigenvalues (λ i , i = 1, 2, . . . , n) and the eigenvectors ( xi , i = 1, 2, . . . , n)
are given as the solutions to the equation system

Ax = λ x

which has a solution if

det(λ I − A) = λ n + α 1 λ n−1 + α 2 λ n−2 + ⋅ ⋅ ⋅ + α n = 0

λ n + α 1 λ n−1 + α 2 λ n−2 + ⋅ ⋅ ⋅ + α n is called the characteristic polynomial.

det(λ I − A) = 0 is called the characteristic equation.

2
Dynamical systems

State-space equations
dx
= Ax + Bu
dt
y = Cx + Du
Z t
At
x(t) = e x(0) + eA(t−τ ) Bu(τ )dτ
0

Weighting function
Z t
y(t) = h(t − τ )u(τ )dτ
0
h(t) = CeAt B + Dδ (t)

Transfer function
Y (s) = G (s) U (s)
G (s) = C(sI − A)−1 B + D = L h(t)


The denominator of G is the characteristic polynomial to the matrix A.

Frequency response
u(t) = sin ω t
y(t) = a sin(ω t + ϕ )

a = p G (iω )p
ϕ = arg G (iω )

Linearization
If the nonlinear system

dx
= f ( x, u)
dt
y = ˆ( x, u)

is linearized around a stationary point ( x0 , u0 ), a change of variables

∆ x = x − x0
∆ u = u − u0
∆ y = y − y0

then gives the linear system

d∆ x €f €f
= ( x0 , u0 )∆ x + ( x0 , u0 )∆ u
dt €x €u
€ˆ €ˆ
∆y = ( x0 , u0 )∆ x + ( x0 , u0 )∆ u
€x €u

3
State-space representations
1. Diagonal form

λ1 0  β1 
   

  
 
λ2 β2 
   
dz 
 
 
 
= z+
   

.

 
.

 u
dt 

 . .





 .
.




 
 
 

   
0 λn βn
 
y =  γ 1 γ 2 . . . γ n  z + Du

2. Observable canonical form

−a1 1 0 ... 0 b1 
   

  
 
− a 0 1 0 b

 
 
 


 2 
 
 2 

dz  
 .

 
 .


= .
.

 z +

 .
.

 u
dt  
 
 


 
 
 




 −an−1 0 0 1









    
− an 0 0 0 bn
 
y =  1 0 0 . . . 0  z + Du

3. Controllable canonical form

−a1 −a2 . . . −an−1 −an 

 
1

   
1 0 0 0

 
 
 
 
0
 
dz 
 
 
 
= 0 1 0 0 z+
 
u
 
.
   
.
 
dt  
. .
   
.
  
 

.

 
 
  
 

 
 0
0 0 1 0
 
y =  b1 b2 . . . bn  z + Du

The transfer function of the system is

b1 sn−1 + b2 sn−2 + ⋅ ⋅ ⋅ + bn
G (s) = D +
sn + a1 sn−1 + ⋅ ⋅ ⋅ + an

β 1γ 1 β 2γ 2 β nγ n
= D+ + + ⋅⋅⋅ +
s − λ1 s − λ2 s − λn

4
 The Laplace transform

Operator lexicon

Laplace transform F (s) Time function f (t)

1 α F1 (s) + β F2 (s) α f1 (t) + β f2 (t) Linearity

2 F (s + a) e−at f (t) Damping

(
f (t − a) t−a>0
3 e−as F (s) Time delay
0 t−a<0
1 s
4 F (a > 0) f (at) Scaling in t-domain
a a
1 t
5 F (as) (a > 0) f Scaling in s-domain
a a
Z t
6 F1 (s) F2 (s) f1 (t − τ ) f2 (τ ) dτ Convolution in t-domain
0
Z c+i∞
1
7 F1 (σ ) F2 (s − σ ) dσ f1 (t) f2 (t) Convolution in s-domain
2π i c−i∞

8 sF (s) − f (0) f ′ (t) Differentiation in t-domain

9 s2 F (s) − s f (0) − f ′ (0) f ′′ (t)

10 sn F (s) − sn−1 f (0) − ⋅ ⋅ ⋅ − f (n−1) (0) f (n) (t)

dn F (s)
11 (−t)n f (t) Differentiation in s-domain
dsn
Z t
1
12 F (s) f (τ ) dτ Integration in t-domain
s 0

∞
f (t)
Z
13 F (σ ) dσ Integration in s-domain
s t

14 lim sF (s) lim f (t) Final value theorem

s→0 t→∞

15 lim sF (s) lim f (t) Initial value theorem

s→∞ t→0

5
Transform lexicon

Laplace transform F (s) Time function f (t)

1 1 δ (t) Dirac function

1
2 1 Step function
s

1
3 t Ramp function
s2

1 1 2
4 2 t Acceleration
s3

1 tn
5
sn+1 n!

1
6 e−at
s+a

1
7 t ⋅ e−at
(s + a)2
s
8 (1 − at) e−at
(s + a)2

1 1
9 T e−t/T
1 + sT
a
10 sin at
s2 + a2
a
11 sinh at
s2 − a2
s
12 cos at
s2 + a2
s
13 cosh at
s2 − a2

1 1
1 − e−at


14 a
s(s + a)

1
15 1 − e−t/T
s(1 + sT )

1 e−bt − e−at
16
(s + a)(s + b) a−b

6
Transform lexicon, continued

Laplace transform F (s) Time function f (t)

s ae−at − be−bt
17
(s + a)(s + b) a−b
a
18 e−bt sin at
(s + b)2 + a2

s+b
19 e−bt cos at
(s + b)2 + a2

1
20
s2 + 2ζ ω 0 s + ω 02

1
ζ =0 ω0 sin ω 0 t

 p 
ζ <1 √1 e−ζ ω 0 t sin ω 0 1 − ζ 2 t
ω0 1−ζ 2

ζ =1 te−ω 0 t

 p 
ζ >1 √1 e−ζ ω 0 t sinh ω 0 ζ 2 − 1 t
ω0 ζ 2 −1

s
21
s2 + 2ζ ω 0 s + ω 02
 p 
0 ≤τ ≤π : ζ <1 √1 e−ζ ω 0 t sin ω 0 1 − ζ 2 t + τ
1−ζ 2

√
ω 0 1−ζ 2
τ = arctan −ζ ω 0

ζ =0 cos ω 0 t

ζ =1 (1 − ω 0 t) e−ω 0 t
a 1
sin(at − φ ) + e−bt sin φ


22
√
s2 + a2 (s + b) a2 + b2
a
φ = arctan
b

7
Laplace transform table, continued

Laplace transform F (s) Time function f (t)

s 1
cos(at − φ ) − e−bt cos φ


23
√
s2 + a2 (s + b) a2 + b2
a
φ = arctan
b

ab ae−bt − be−at
24 1+
s(s + a)(s + b) b−a

a2
25 1 − (1 + at) e−at
s(s + a)2
a
26 t − 1a (1 − e−at )
s2 (s + a)

1 (b − c) e−at + (c − a) e−bt + (a − b) e−ct

27
(s + a)(s + b)(s + c) (b − a)(c − a)(b − c)

ω 02
28
s(s2 + 2ζ ω 0 s + ω 02 )
 p 
0<ζ <1 1− √1 e−ζ ω 0 t sin ω 0 1 − ζ 2 t + φ
1−ζ 2

φ = arccos ζ

ζ =0 1 − cos ω 0 t

1 1 n −at
29 t e
(s + a)n+1 n!

s a(b − c) e−at + b(c − a) e−bt + c(a − b) e−ct

30
(s + a)(s + b)(s + c) (b − a)(b − c)(a − c)

as t
31
2 sin at
s2 + a2 2

1 1
32 √ √
s πt
√

Z ∞
1 1 2 /4t
33 √ F s √ e−σ f (σ ) dσ
s πt 0

8
 G G
G(s)=(1+sT), 1
1+sT
20 T=1

Bode plot. Standard curves

10
90

.
01 2 5 1 2 5 10
1 0
Radians/s

0.5

0.2

90
0.1

0.05

0.02
180
0.01
9
10
G G
G(s)=e-sL
20 L=1

10
90

.
01 2 5 1 2 5 10
1 0
Bode plot. Standard curves, continued

Radians/s

0.5

0.2

90
0.1

0.05

0.02
180
0.01
 G G
ω2
G(s)= 0
s2 + 2ζω0s + ω02
20 ω0=1

Bode plot. Standard curves, continued

10
ζ
90
ζ
5 ζ

2 ζ

.
01 2 5 1 2 5 10
1 0
Radians/s

0.5 ζ

ζ
0.2

90
0.1
ζ
0.05 ζ
ζ
ζ
ζ
0.02 ζ
180
0.01
11
Stability

Stability conditions for low-order polynomials

s + a1 a1 > 0
s2 + a1 s + a2 a1 > 0, a2 > 0
3 2
s + a1 s + a2 s + a3 a1 > 0, a2 > 0, a3 > 0, a1 a2 > a3

Routh’s algorithm
Consider the polynomial

F (s) = a0 sn + b0 sn−1 + a1 sn−2 + b1 sn−3 + ⋅ ⋅ ⋅

Assume that the coefficients ai , bi are real and that a0 is positive. Form
the table
a0 a1 a2 . . .
b0 b1 b2 . . .
c0 c1 c2 . . .
d0 d1 d2 . . .
..
.
where
c0 = a1 − a0 b1 /b0
c1 = a2 − a0 b2 /b0
..
.
d0 = b1 − b0 c1 /c0
d1 = b2 − b0 c2 /c0
..
.
The number of sign changes in the sequence a0 , b0 , c0 , d0 ⋅ ⋅ ⋅ equal the num-
ber of roots for the polynomial F (s) in the right half plane Re s > 0. All
the roots of the polynomial F (s) lie in the left half plane if all numbers a0 ,
b0 , c0 , d0 , . . . are positive.

12
Stability margins

0.8

0.6 1/ Am

0.4

0.2

G0 (ω o )
0
ϕm
−0.2

−0.4

−0.6

−0.8
G0 (ω c )

−1

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1

1
10

ωc
0
10
Gain

1
Am
−1
10

−2
10
−1 0
10 10

−50

−100
Phase

−150
ϕm

−200 ωo

−250
−1 0
10 10
Frekvens [rad/s]

Gain margin:
Am = 1/ p G0 (iω 0 )p
Phase margin:
ϕ m = π + arg G0 (iω c )
Delay margin:
L m = ϕ m /ω c

13
State feedback and Kalman filtering

State feedback
If the system
dx
= Ax + Bu
dt
y = Cx
has the control law
u = − Lx + {r r
then the closed-loop system is given by

dx
= ( A − B L) x + B {r r
dt
y = Cx

Criterion for controllability. The controllable states belong to the li-

near subspace which is spanned by the columns of the matrix
 
n−1
Ws =  B AB ⋅⋅⋅ A B

A system is controllable if and only if the matrix Ws has rank n.

Kalman filtering
Assume that only the output signal y can be directly measured. Introduce
the model
d x̂
= A x̂ + Bu + K ( y − C x̂)
dt
The reconstruction error x̃ = x − x̂ satisfies

d x̃
= ( A − K C) x̃
dt

Criterion for observability. The subspace of unobservable states is the

null space of the matrix
C 
 

 
CA 

 
 
Wo = 
 
 .
.


.

 


 

n−1
 
CA
A system is observable if and only if the matrix W0 has rank n.

14
Lead-lag compensation

Lag compensator

s+a 1 + s/a
G K (s) = =M M >1
s + a/ M 1 + sM /a
The rule of thumb
a = 0.1ω c
guarantees that the phase margin is reduced by less than 6○ .

Lead compensator

s+b 1 + s/b
G K (s) = K K N = KK N>1
s + bN 1 + s/(bN )
The maximum phase advance is given by the figure below:

60

50

40
∆ϕ m

30

20

10

0
1 2 3 4 5 6 7 8 9 10

The peak of the phase curve is located at the frequency

√
ω =b N

The gain of the compensator at this frequency is

√
KK N

15
The Ziegler–Nichols methods

The Ziegler–Nichols step response method

Consider the step response for the open-loop system. The tangent is drawn
from the point on the step response with the maximal slope. From the
intersection of the tangent and the coordinate axes the gain a and time b
are found. The PID-parameters are calculated from the table below.

a t

Controller K Ti Td
P 1/a
PI 0.9/a 3b
PID 1.2/a 2b 0.5b

The Ziegler–Nichols frequency method

This method is based on observations of the closed-loop system. Outline of
the procedure:
1. Disconnect the integral and the derivative part of the PID-controller.
2. Adjust K until the system oscillates with constant amplitude. Denote
this value of K as K 0 .
3. Measure the period T0 for the oscillation. The different settings for
the controller parameters are given in the table below.

Controller K Ti Td
P 0.5K 0
PI 0.45K 0 T0 /1.2
PID 0.6K 0 T0 /2 T0 /8

16