Lecture 2

System models and characteristic properties of linear

time-invariant systems in the time-domain

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Differential equation as model of LTI systems

A LTI-System is a linear time-invariant dynamic system and can be modeled in SISO case using a
linear ordinary differential equation with constant coefficients, which defines the relationship
between the system input u(t) and system output y(t)
dn y d n −1 y dy d pu d p −1u du
n
+ a n −1 n −1
+ ... + a1 + =
a 0 y b p p
+ b p −1 p −1
+ ... + b1 + b0u (2.1)
dt dt dt dt dt dt
It is always n ≥ p (only those systems are physically realizable!) and the highest number n of
differentiations of the system output y is called the order of the system.
Eq. (2.1) represents the input-output relation of the dynamic system and is one of the standard
mathematical models used for system modeling.
The model equation (2.1) can describe a physical process (plant), a controller or a closed-loop
control as well.
Which system properties can we recognize from this standard form of system model? To answer the
question, we first consider its simplest variant, the first-order dynamic LTI systems.

Notice:
An exact definition of linearity and time-invariance of dynamic systems will be given later.

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Design problem 1: cruise control (1)

Process input: gas pedal (throttle) u

Process output: cruising velocity v
Desired output (reference signal): desired velocity vr (constant)
Disturbance: slope θ

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Design problem 1: cruise control (2)

Simplification assumptions:
• Essential dynamics relates velocity and force
• The force responds very fast to a change in the throttle
• All relations are assumed to be linear (small signal behavior)

Process model:
A simplified mathematical description (first-order differential equation):
dv
m + Dv = F − mg sin θ ≈ F − mgθ
dt
With reasonable parameters
dv
u − 10θ
+ 0.02v =
dt
dn y d n −1 y dy d pu d p −1u du
n
+ a n −1 n −1
+ ... + a1 + =
a 0 y b p p
+ b p −1 p −1
+ ... + b1 + b0u
dt dt dt dt dt dt

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 First-order dynamic LTI systems

We know from the theory of differential equations that the first-order homogeneous equation
dy (t )
+ ay (t ) =
0 (2.2)
dt
has the solution

y (t ) = Ce − at (2.3)

where C = y(0) is an initial condition depending constant. The solution of the inhomogeneous equation
dy (t )
+ ay (t ) =
bu(t ) (2.4)
dt
with u as system input is known as
t
=
y (t ) Ce − at
+ b ∫ e − a ( t −τ )u(τ )dτ (2.5)
0
and consists of the free response (the first Term, depending on initial condition only) and the forced
response (the second Term, depending on system input only).
Note:
-a is the solution to the equation s+a=0.

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Design problem 1: cruise control (3)

System behaviour:
According to the solution of first order differential equation the velocity dynamic is described by:
t
v (t ) = v0e + ∫ e −0.02 ( t −τ ) (u − 10θ )dτ
−0.02 t
  0
  
free response
forced response

where v0 [m/s]: initial speed.

Simulation using MATLAB/SIMULINK:

θ
10
slope [rad/s] u
u
0.02*v0 1
xo s 3.6
actuation signal for
90 km/h Integrator v [km/h]
v0
initial speed 0.02
90 km/h

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Design problem 1: cruise control (4)

Free Response Free- and Forced Response

Without (feedback) control the velocity cannot be kept constant in case of disturbance!

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Design problem 1: cruise control (5)

Construction of a closed-loop control: θ

10 v0
Disturbance forces
vr u
vr u F
Fdis
v
kp ∫ v
Car
Contr Eng
Body
throttle 0.02

The closed-loop system behavior:

Again by solving the first order differential equation we
P (Proportional) controller: have:
= u k P ( vr − v ) ( vr = 100) − (0.02 + k P )t
t

 ∫
−( 0.02 + k P )( t −τ )
v (t ) = v0e + e ( k P vr − 10θ )dτ
The closed-loop system is described by 0
 
free response
dv forced response
+ (0.02 + k P )v = k P vr − 10θ In steady-state with t → ∞ we have:
dt
k P vr 10θ
=v∞ −
0.02 + k P 0.02 + k P
Prof. Dr.-Ing. Steven Liu
Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Active excise

Stop and think!

How is the behavior of the equations

dv dv
u − 10θ
+ 0.02v = and + (0.02 + k P )v = k P vr − 10θ
dt dt
and how is the closed-loop behavior influenced
by the controller parameter kP?
How should we choose kP to achieve a "good" control behavior?

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Design problem 1: cruise control (6)

100
θ

Vehicle velocity [km/h]

10 v0 kP = 2
95
vr u
kp ∫ v
90
kP = 0.5
0.02 85

80
0 5 10 15 20
Time [s]

How to choose the controller parameter?

• The larger the controller gain kP, the better the dynamic behavior
• The larger the controller gain kP, the smaller the steady-state velocity error
• Question: Can we choose kP → ∞?
• No, since throttle variation range is limited → saturation!
• Steady-state velocity error will remain!

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Active excise

Stop and think!

t
=
y (t ) Ce − at
+ b ∫ e − a ( t −τ )u(τ )dτ
0

Which system parameter characterizes the system behavior?

…

Parameter a!

The "system stability" (whether the system output goes to infinity)

depends only on a!
Prof. Dr.-Ing. Steven Liu
Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Second-order dynamic LTI systems (1)

The solution of the second-order homogeneous equation

d2 y dy
2
+ a1 + a0 y =
0 (2.6)
dt dt
can be given as

y (t ) C1eα1t + C2eα2t
= (2.7)

where α1, α2 are roots of the so-called characteristic equation

s 2 + a1s + a0 = ( s − α1 )( s − α 2 ) = 0 (2.8)
and C1, C2 are constants to be determined using initial conditions. The solution of the inhomogeneous
LTI equation, consisting again of the free and the forced response, is for the case α1 ≠ α2:
t
∫
α1t α 2t
y (t )= C1e + C2 e + (α 2 − α1 ) −1
( eα2 ( t −τ ) − eα1 ( t −τ ) )u(τ )dτ (2.9)
0

free response forced response

The system behavior for different system inputs u can be conveniently studied by computer simulations.

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Second-order dynamic LTI systems (2)

Basically, the system response, i.e., the solutions of the differential equation, depends on the type
(real, imaginary, or complex) of the roots α1, α2 of the characteristic equation as well as on their
signs. See also appropriate mathematical text books for more details.

a0
a0

ydot y
1/s 1/s

Integrator1 Integrator2 Scope

a1

a1

MATLAB/SIMULINK model for simulation studies

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Design problem 1: cruise control (6)

Choose a different controller: θ

PI (Proportional-Integral) controller: 10 v0
t kP
u k P ( vr − v ) + k I ∫ ( vr − v )dτ
= vr u
∫
0
The closed-loop system is now described by v
d 2v dv dθ
dt 2
+ (0.02 + k P )
dt
+ k I v = k I v r − 10
dt
∫ kI
0.02
In steady-state with θ = const. we have
v∞ = vr (independent of vr and θ!)
How can we determine the performance of the closed-loop control?
The control error e = vr − v can be described by the 2nd order differential equation:
10θ
e + (0.02 + k P )e + k I e =

Comparison with the normalized resonance oscillator (mass-spring-damper or LCR-oscillator) equation
x + 2d ω0 x + ω02 x =
 0 d: damping factor ,ω0 : undamped natural frequency
Give 2d ω0 − 0.02
kP = and ω02
kI =

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Design problem 1: cruise control (7)

Simulation using MATLAB/SIMULINK:

θ
10
open-loop slope [rad/s] u
system 0.02*v0
u
1
xo s 3.6
actuation signal for
90 km/h Integrator v [km/h]
v0
initial speed 0.02
90 km/h

θ
10
slope change kp
u
closed-loop with vr
vr u
1
PI controller Reference speed 1 xo s 3.6
s ki v [km/h]
Integrator
I v0
initial speed 0.02
90 km/h

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Design problem 1: cruise control (8)
3 4.5

Normalized throttle [m/s2]

Normalized throttle [m/s2]

105 105
Vehicle velocity [km/h]

Vehicle velocity [km/h]

100 2 100 3

95 1 95 1.5
d = 0.7 d = 1.0
ω0 = 0.8 ω0 = 0.8
90 0 90 0
0 5 10 15 20 0 5 10 15 20
3 6

Normalized throttle [m/s2]

Normalized throttle [m/s2]

105 105
Vehicle velocity [km/h]

Vehicle velocity [km/h]

100 2 100 4

95 1 95 2
d = 0.7 d = 0.7
ω0 = 0.4 ω0 = 1.6
90 0 90 0
0 5 10 15 20 0 5 10 15 20
Time [s] Time [s]

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Active excise

We consider again the closed-loop dynamic of our cruise control design example with a PI
controller

 10θ
e + (0.02 + k P )e + k I e = resp. e + 2d ω0e + ω02e =
 0
The solution is greatly influenced by the parameters damping factor d and undamped natural
frequency ω0

Stop and think!

d < −1
−1 ≤ d < 0
What happens for d =0 ?
0 < d ≤1
d >1
How is the solution influenced by ω0?
Prof. Dr.-Ing. Steven Liu
Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Step response and impulse response

Unit step function (Heaviside Function): Step response bzw. Sprungantwort:

σ(t)
σ(t) (n) ( n −1)
h(t)
0 ∀t < 0 y + an −1 y +  + a1 y + a0 y
σ (t ) =
1
1 ∀t ≥ 0
( p)

u0 σ(t) = bp u + ... + b1u + b0u u0h(t)

t
0

Unit impulse function (δ impulse, Dirac impuls): Impulse response bzw. Impulsantwort:
δ(t)
δ (t ) = 0 ∀ t ≠ 0
1
δ(t) (n) ( n −1)
y + an −1 y +  + a1 y + a0 y g(t)
∞
∫ δ (t )dt = 1 ( p)

-∞ u0δ(t) = bp u + ... + b1u + b0u u0g(t)

t
0
∞
Sifting property of unit impulse: ∫
−∞
δ (t − τ ) ⋅ f (τ )dτ =f (t )
d d
Relation between σ and δ: σ (t ) = δ (t ) and h (t ) = g (t )
dt dt
Prof. Dr.-Ing. Steven Liu
Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Second-order dynamic LTI systems (3)

y (t ) + 2d ω0 y (t ) + ω02 y (t ) =
 ω02u(t )
The corresponding characteristic equation ( s + d ω0 + jω0 1 − d 2 )( s + d ω0 − jω0 1 − d 2 ) =
0

Impulse response Step response

1 2
d=0 d=0
d = 0.2
d = 0.2
d = 0.4
d = 0.6 d = 0.4
0.5 1.5
d = 0.8 d = 0.6
d = 1.0 d = 0.8

y(t)
y(t)

0 1
d = 2.0

-0.5 0.5
d = 1.0

d = 2.0
-1 0
0 5 10 15 20 0 5 10 15 20
ω0t ω0t

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Design problem 1: cruise control (9)

What is different between P and PI controlled system (in this special case)?
P control:
• Remaining steady-state control error
• System response is real exponential function, no overshoot (1st order dynamics)
• One common parameter for both the transient and stationary behavior of the closed-loop
PI control:
• No steady-state control error (due to integral action)
• System response is complex exponential function, overshoot possible (2nd order dynamics)
• Stationary behavior independent of controller parameters, closed-loop transient behavior can
be separately (oscillation tendency and response dynamic) determined by proper choice of
controller parameters

The dynamic order of the closed-loop system is essential for the transient behavior of
the controlled system. It is the sum of the dynamic orders of the plant (process) and
the controller.

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 High-order dynamic LTI systems (1)

The nth-order homogeneous equation (no input, free response) is given by

dn y d n −1 y dy
n
+ a n −1 n −1
+  + a1 + a0 y =
0 (2.10)
dt dt dt
The corresponding characteristic equation with the characteristic polynomial ∆(s) is then

∆( s ) = s n + an −1s n −1 +  + a1s + a0 = 0 (2.11)

If ∆(α) = 0, then y(t) = eαt is a solution of (2.10)! For the case that the roots of the characteristic
equation αk, k = 1,…, n, are completely distinct, the general solution of the homogeneous equation
(2.10) can be stated as a linear combination of the single solutions
n
y ( t ) = ∑ Ck e − α k t (2.12)
k =1
Notes: ( n −1)
1) The coefficients Ck can be given as linear combination of the initial conditions y(0), y (0), … , y (0) .
2) αk can be real or complex. The complex roots are always conjugate complex pairs. By combining
the corresponding conjugate complex solutions one obtains finally exponential oscillation terms
eσtsinωt and eσtsinωt with σ = Re{αk}.

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 High-order dynamic LTI systems (2)

The dynamic of a high-order system is the sum of first-order and second-order dynamics. It is
essentially determined by the roots of the characteristic equation!

Special case: multiple roots of the characteristic equation

If α is a solution of multiplicity k to the characteristic equation, then a (k-1)-degree polynomial C(t)
exists so that y(t) = C(t)eαt is a solution to the homogeneous differential equation. Thus, the
general solution to equation (2.10) can be given as
nk
y (t ) = ∑ Ck −1 (t )e − ak t (2.13)
k =1
where ak, k = 1,…nk are roots of the characteristic equation.

Note:
Generally there are n solutions to the characteristic equation with some of them possibly in
multiplicity. The corresponding polynomial Ck-1(t) to a root of multiplicity k is of degree k-1. In case
of distinct roots the polynomials Ck-1(t) reduce to constant coefficients.!

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 High-order dynamic LTI systems (3)

Before starting to discuss the solution for the general system model (2.1) we consider first the
following special case

dn y d n −1 y dy
n
+ a n −1 n −1
+  + a1 + a0 y =
u (2.14)
dt dt dt
The solution to above equation is known as
n

∑C
t
=
y (t ) k −1 (t )e ak t
+ ∫ g h (t − τ )u(τ )dτ (2.15)
0
k =1
where gh(t) is the solution to the homogeneous equation
d n gh d n −1 g h dg h
+ an −1 n −1 +  + a1 + a0 g h =
0 (2.16)
dt n dt dt
with following special initial conditions
( n − 2) ( n −1)
=
g h (0) 0,=
g h (0) 0,  , =
g h (0) 0,=
g h (0) 1 (2.17)
Based on this solution the general problem can be now tackled.

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 High-order dynamic LTI systems (4)

The solution to the general system model of n-th order

dn y d n −1 y dy d pu d p −1u du
n
+ a n −1 n −1
+  + a1 + =
a 0 y b p p
+ b p −1 p −1
+ ... + b1 + b0u (2.18)
dt dt dt dt dt dt
comprises again two terms
n

∑C
t
=
y (t ) k −1 (t )e − ak t
+ ∫ g (t − τ )u(τ )dτ (2.19)
0
k =1

free response forced response

where g(t) is referred to as the impulse response

( p) ( p −1)
=
g (t ) bp g h (t ) + bp −1 g h (t ) + ... + b1 g h (t ) + b0 g h (t ) (2.20)

with gh(t) being the solution to (2.16).

From now on we assume generally zero initial conditions (no free response), unless mentioned
otherwise, to drop our attention on the system's input-output behavior.

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Active excise

Stop and think!

For the purpose of controller design it is usually too

complicated to solve the differential equation for
the consideration of system behavior
Is there an alternative (and easier) way?
Can we just "read off" the most important properties
from a more concise system description?

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 System description in the Laplace-domain

By transforming the total system model on zero initial conditions into the frequency-domain (also
known as Laplace-domain) using Laplace transform, one obtains
( s n + an −1s n −1 +  + a1s + a0 )Y ( s=
) (bp s p + bp −1s p −1 + ... + b1s + b0 )U ( s ) (2.21)
where s is the Laplace variable. By defining the denominator and numerator polynomials
N ( s ) = s n + an −1s n −1 +  + a1s + a0
(2.22)
p −1
Z ( s=
) bp s + bp −1s
p
+ ... + b1s + b0
the solution to the system equation can be easily given in the Laplace-domain
Z (S ) bp s p + bp −1s p −1 + ... + b1s + b0
=
Y ( s ) G=
( s )U ( s ) = U ( s) U (S )
N ( s) s n + an −1s n −1 +  + a1s + a0
(2.23)
( s − β1 )( s − β 2 )  ( s − β p )
=K U (S )
( s − α1 )( s − α 2 )  ( s − α n )
The system function G(s) is called the transfer function. The zeros αi of the denominator polynomial
N(s) are the poles of the dynamic system (or the transfer function) and those of the numerator
polynomial Z(s), βi , are known as the system zeros (or zeros of the transfer function).

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Transfer function and impulse response

Comparing the definitions for the convolution integral and the transfer function and taking into
account the convolution properties of the Laplace transform (s. appendix at the end of this
chapter), it is easy to see that the transfer function (Laplace domain) is simply the Laplace
transform of the impulse response (time-domain).
Thus, the general behavior of a dynamic system on zero initial conditions can be characterized
equivalently either in the Laplace-domain by the transfer function G(s) or in the time-domain by
the impulse response g(t). Because of this fundamental property, the impulse response is also
called natural response of a system.

Y ( s ) = G ( s )U ( s )
U(s) G(s) Y(s)
y (t ) = g (t ) ∗ u (t ) g(t)
u(t) y(t)
{G ( s )U ( s )}
t
∫ τ τ τ −1
y (t ) = g ( t − ) u ( )d =
L
0

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Poles and system stability

The system poles determine the components of the time function building up the solution of the
differential equation (i.e., the system response to initial excitation or system input).

A dynamic LTI system is said to be bibo-stable (bounded-input bounded-output), if the system

output is always bounded for a bounded input signal.

Therefore, a dynamic LTI system is bibo-stable, if and only if all the system poles have negative
real part, i.e., they are located in the left half of the complex s-plane (LHP = left half-plane).
In order to design a stable control loop, the controller parameters are to be chosen in the way
that the closed-loop transfer function only has stable poles.

Stop and think!

What happens, if one or several poles are located in the right
half-plane (RHP) or on the imaginary axis?
Prof. Dr.-Ing. Steven Liu
Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Pole location (σ+jω) and free response

Real pole with σ < 0 Real pole with σ > 0

Complex poles with σ < 0

Damping time constant
t t
-1/σ

Exponential envelope

Complex poles with σ > 0

Real pole with σ = 0
t

Complex poles with σ = 0

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Pole location and dynamic behavior

The pole location, marked as "x" in the zero-pole-diagram, determines the fundamental behavior
of the system response.
jω

LHP RHP

σ
stable unstable

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Influence of an additional pole
h(t) Step Response
1.2
1
α=∞
0.8 α=5
α = 0.5
0.6 α=2
α = 0.2
0.4 α = 1
0.2
0
0 1 2 3 4 5 6 7 8 9 10 ω0t

αω03
G( s) = 2
( s + 2d ω0 s + ω02 )( s + αω0 )
MATLAB command script:
a=0.2;
sys=tf(a,conv([1 2*0.6*1 1],[1 a]));
step(sys,10)

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Interpretation of zeros

Stop and think!

Influence of poles is well-understood. But what is about zeros? ...

If s = β is a zero of the transfer function

Z ( β=
) bp β p + bp −1β p −1 + ... + b1β + b=
0 0 (2.24)

and u(t) = C(t)eβt, it follows that for the right-hand side of equation (2.18)

d pu d p −1u du
bp p + bp −1 p −1 + ... + b1 = )Ce β t 0
+ b0u Z ( β= (2.25)
dt dt dt
A zero s = β blocks the transmission of the signal (i.e. eliminates the effect of the input excitation)
u(t) = C(t)eβt.

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Influence of an additional zero

We assume d = 0.6, ω0 = 1 and consider now Step Response

2
s
ω02 ( + 1) β = 0,5
β (2.26) β=2
T ( s) =
s 2 + 2d ω0 s + ω02
1

Amplitude
for β = 0.5, β = 2.0, β = ∞, β = -2.0 and β = -0.5.
h0(t) (β = ∞)
1 s 1 
h (t ) =
L -1  ⋅ 2 + 2  β = -2
 β s + 1.2 s + 1 s + 1.2 s + 1  0
1  β = -0,5
= h0 (t ) + h0 (t ) (2.27)
β
• Zeros increase the peak value. The closer to origin -1
they are located, the stronger is their influence! 0 2 4 6 8 10
Time (sec)
• Zeros in the RHP cause a deflection in the opposite
direction at the beginning of the response (non-minimum phase behavior)!

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 System static gain

We consider again the general input-output system model

dn y d n −1 y dy d pu d p −1u du
n
+ a n −1 n −1
+ ... + a1 + =
a 0 y b p p
+ b p −1 p −1
+ ... + b1 + b0u
dt dt dt dt dt dt
For constant input and output, i.e., u(t) = u0 and y(t) = y0, it obviously yields

a0 y0 = b0u0
The ratio
b0
V= (2.28)
a0
is called static gain of the system (different from K from (2.23)!).

Active excise:
Calculate the static gain of an integrator!

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Theory review

What have we learned from the theory in this chapter?

• A dynamic LTI system can be described by a linear ordinary differential equation (ODE) with
constant coefficients.
• The solution of the ODE can be analyzed to determine the system behavior.
• The solution of first and second order ODE is easy to get, but difficult for high-order systems
(convolution integral for the forced response!). The free response, however, is just the additive
sum of first and second order free responses.
• Basic element of the solution is the exponential function, real or complex (exponential function
with sinusoidal oscillation). The real part of the roots of characteristic equation is completely
decisive on the stability of the solution!
• Making use of the Laplace transform (and transfer function) a more concise system analysis is
possible. The poles determine essentially the dynamic performance (stability, damping, natural
frequency). What is the influence of the system zeros on dynamic performance? We have some
idea, but still need more insight.
• Step and impulse response are often used as "benchmarking" signals for controller design. How
can they be determined by experiments?

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Conclusions for practical control design

How can this theoretical knowledge help in practical control design?

• The roots of the characteristic equation (= poles of the dynamic system) of the closed-loop
control must be placed in the LHP to ensure the stability of the control (indication for the proper
choice of controller structure and/or parameters!)
• The closed-loop static gain should be chosen to unity in order to eliminate steady-state control
error (indication for the proper choice of controller structure and/or parameters!)
• The dynamic performance (= transient behavior) of the control is essentially influenced by the
closed-loop pole location. A real pole results in an exponential convergence and a conjugate
complex pair of poles causes exponential oscillation, which, in case of small damping, leads to
overshoot in transient situations (indication for the proper choice of controller structure and/or
parameters!)
• Simulation of step and impulse response is useful to "tune" the control performance
What else?:
• Transfer function is an efficient and very useful form of system description. Important for
controller design: how can plant (process), controller and control loop be linked together using
transfer functions? How can a systematic design make use of such linked structures?
• How can the controller be realized, if designed in form of a transfer function?
Prof. Dr.-Ing. Steven Liu
Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Design example: once again cruise control (1)

dv
Simplified process model: + 0.02v = u − 10θ
dt
t
PI controller: u k P ( vr − v ) + k I ∫ ( vr − v )dτ
=
0

The error equation: e= vr − v

Taking Laplace transform yields ( s + 0.02)V ( s ) = U ( s ) − 10θ ( s )

k
=
U ( s) kP E ( s) + I E ( s)
s
E= ( s ) Vr ( s ) − V ( s )
Manipulating the equations one obtains (cp. also page 2-7)
s( s + 0.02) 10s
E ( s ) = GV ( s )Vr ( s ) + Gθ ( s )θ ( s ) = V ( s ) + θ ( s)
s + (0.02 + k P ) s + k I s + (0.02 + k P ) s + k I
2 r 2

Behavior on Disturbance (road slope)

reference change behavior

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Design example: once again cruise control (2)

Both GV(s) and Gθ(s) have a zero at s = 0. Therefore, a step change on reference speed or road slope
will, independently of controller parameters, cause no steady-state control error. In other words, the
control error will be eliminated by the controller (effect of controller integral action)!
The denominator polynomial is for both transfer functions the same! For determining the transient
behavior we choose e.g. (cp. page 2-9) d = 0.8 (almost no overshoot). The corresponding curves of
step and impulse response show that dynamic changeover is almost completed after ca. ω0t = 5. if
we for instance specify this transition with a real time of 2.5 [s] , then it is ω0 = 5/2.5 = 2 [1/s]. From
the values for d and ω0 we can finally calculate the parameters of the controller as
= =
k P 3.18 kI 4
To study the disturbance behavior in more detail we assume that Vr = const.. It is then
10s
E ( s) = θ ( s)
s + (0.02 + k P ) s + k I
2

The solution to a step change of θ in the Laplace-domain is (θ(s) = ∆θ/s)

10s ∆θ
=E ( s) ⋅
s 2 + 3.2 s + 4 s

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Second-order dynamic LTI systems (3)

y (t ) + 2d ω0 y (t ) + ω02 y (t ) =
 ω02u(t )
The corresponding characteristic equation ( s + d ω0 + jω0 1 − d 2 )( s + d ω0 − jω0 1 − d 2 ) =
0

Impulse response Step response

1 2
d=0 d=0
d = 0.2
d = 0.2
d = 0.4
d = 0.6 d = 0.4
0.5 1.5
d = 0.8 d = 0.6
d = 1.0 d = 0.8

y(t)
y(t)

0 1
d = 2.0

-0.5 0.5
d = 1.0

d = 2.0
-1 0
0 5 10 15 20 0 5 10 15 20
ω0t ω0t

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Design example: once again cruise control (3)

Converting this result to a time function we find the expression for the control error:
25
=
e( t ) ∆θ ⋅ e −1.6t sin(1.2t )σ (t )
3
3

2.5
Control error in [km/h]

1.5

0.5

-0.5
0 0.5 1 1.5 2 2.5 3 2.5 4 4.5 5
Time in [s]

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Appendix 3A: Laplace transform

The unilateral (or one-sided) Laplace transform (for causal systems):

∞
=
F ( s) L=
{ f (t )} ∫ f (t )e− stdt with Laplace variable s = σ + jω
-0

f (t )  F ( s)
or as linear mapping:
time-domain Laplace-domain
(real) (complex)

σ + j∞
1
Inverse Laplace transform: =f (t ) L= ∫
-1 st
{F ( s)} F ( s )e ds
2πj σ − j∞

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Properties of Laplace transforms (1)

f (t ) −• F ( s ); g (t ) −• G (s)

Linearity:

⇒ a 1 f ( t ) + a 2 g ( t )  − • a 1 F ( s ) + a 2G ( s ) a1, a2 arbitrary complex constants

Time and frequency shift:

f (t − τ) −• e − τs F ( s ) τ arbitrary real number

F ( s − α) •− e αt f (t ) α arbitrary complex number

Differentiation:
d
f (t ) −• sF ( s ) − f ( −0)
dt
f ( n) (t ) −• s n F ( s ) − s n −1 f ( −0) −  − f ( n −1) ( −0)
Prof. Dr.-Ing. Steven Liu
Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Properties of Laplace transforms (2)

Integration:
t
1
∫
−0
f (τ )dτ −•
s
F ( s)

Convolution:

f (t ) ∗ g (t ) −• F (s) ⋅ G (s) Convolution in time-domain

F (s) ∗ G (s) −• f (t ) ⋅ g (t ) Convolution in frequency-domain

Initial and final value theorem:

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

t →+0 s →∞
lim f (t ) = lim(sF ( s ) ) Final value theorem
t →∞ s →0

The final value theorem is only valid, if such a value exists!

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Correspondence table of Laplace transform

No. f (t ) F (s ) No. f (t ) F (s )
2
1 δ (t ) 1 7 t ⋅e −a t
(s + a )3
2

1 a
σ (t ) 1 − e−a t s ⋅ (s + a )
2 8
s
ω
sin (ω t )
1
3 t 9
s2 s2 + ω 2
t n −1 s
cos(ω t )
4 1
10
(n − 1)! sn s2 + ω 2
ω
sin (ω t )
1 −a t
5
e−a t s + a
11 e
(s + a )2 + ω2
1 s+a
6 t ⋅e−a t (s + a )2 12 e − a t cos(ω t ) (s + a )2 + ω 2
Prof. Dr.-Ing. Steven Liu
Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern
 Solving differential equations using Laplace transform

Exp. 3A1: solve the second-order linear differential equation

y + 5 y + 6 y =u + u

if the initial conditions are y(0) = 2, y (0) = 1, and the input u(t) = e-4tσ(t).
Taking the Laplace transform of the differential equation including the initial values we have
s 2Y ( s ) − sy (0) − y (0) + 5[ sY ( s ) − y (0)] + 6Y ( s ) = sU ( s ) − u(0) + U ( s )
Collecting all the terms of Y(s) we easily can solve the algebraic equation in the Laplace-domain
s +1 2 s + 10
=Y ( s) U ( s ) +
s 2 + 5s + 6 s 2 + 5s + 6
By substituting U(s) = 1/(s+4) and expanding the right-hand side into partial fractions we have
 −1/ 2 2 3/ 2   6 4 
Y ( s )=  + − +
  − 
 s + 2 s + 3 s + 4   s + 2 s + 3
The time-domain solution can be obtained via inverse Laplace transform
( − 12 e −2 t + 2e −3t − 23 e −4 t )σ (t ) + (6e −2 t − 4e −3t )σ (t ) =
y (t ) = ( 112 e −2 t − 2e −3t − 23 e −4 t )σ (t )

forced response free response

Prof. Dr.-Ing. Steven Liu

Lehrstuhl für Regelungssysteme
Technische Universität Kaiserslaut ern