ETH Zürich

Department for Chemistry and Applied Biosciences

Institute for Chemical and Bioengineering

Lecture notes for the course of

Chemical Process Control
(Regelungstechnik)

Frühlingssemester 2024

Prof. Robert N. Grass

Organization of the Course

Lecture: Friday 11:45 – 13:30 in HCI H 8.1

Exercise: Wednesday 11:45 – 12:30 in HCI H 8.1
Website: https://fml.ethz.ch/lectures/RegTech.html

Lecturer:
Prof. Robert N. Grass HCI E 111 Tel. 36 344 E-Mail: [email protected]

Assistant:
Andreas Gimpel HCI E 106 Tel. 32909 E-Mail: [email protected]

Recommended Reading:

G.F. Franklin, J.D. Powell and A. Emami-Naeini

Feedback Control of Dynamic Systems
Sixth Edition, 2010, Prentice Hall, Upper Saddle River
ISBN 0-13-032393-4

D.E Seborg, T.F. Edgar and D.A. Mellichamp

Process Dynamics & Control
Third Edition, 2011, Wiley, New York
ISBN 0-471-00077-9

J. Lunze
Regelungstechnik 1
10.Auflage, 2007, Springer, Berlin
ISBN 978-3-642-53908-4

These lecture notes are intended to support students in learning for the exams in Process Control. They are
not understood as a complete and self-explanatory scriptum, and therefore are only useful in connection to
visiting the actual course.

Students willing to learn without visiting the course are advised to follow one of the books mentioned above,
or the online course by Prof. Manfred Morari published on YouTube: http://tinyurl.com/jshqba8
Chemical Process Control, Prof. Robert N. Grass |1

Table of Content

1 INTRODUCTION .............................................................................................................................. 3
ENGINEER’S VIEW OF THE WORLD .................................................................................................. 3
WHAT IS CONTROL? ......................................................................................................................... 3
BASIC ELEMENTS OF PROCESS CONTROL ........................................................................................ 4
HISTORY OF PROCESS CONTROL ...................................................................................................... 5
PROCESS CONTROL IN CHEMICAL INDUSTRY .................................................................................. 6
2 REPRESENTATION OF SYSTEMS ............................................................................................... 7
FEEDFORWARD CONTROL ................................................................................................................ 7
FEEDBACK CONTROL ....................................................................................................................... 8
3 SYSTEM MODELLING ................................................................................................................. 10
4 DIFFERENTIAL EQUATIONS ..................................................................................................... 15
NOMENCLATURE ............................................................................................................................ 15
NUMERICAL SOLUTION .................................................................................................................. 16
ANALYTICAL SOLUTION ................................................................................................................. 16
LINEARIZATION .............................................................................................................................. 17
SUPERPOSITION .............................................................................................................................. 19
CONVOLUTION................................................................................................................................ 20
5 LAPLACE TRANSFORMATION ................................................................................................. 21
LAPLACE TRANSFORMATION ......................................................................................................... 21
INVERSE LAPLACE TRANSFORMATION........................................................................................... 21
IMPORTANT LAPLACE TRANSFORMATIONS ................................................................................... 22
PROPERTIES OF LAPLACE TRANSFORMATIONS .............................................................................. 23
INITIAL CONDITIONS ...................................................................................................................... 25
6 BLOCK DIAGRAM ALGEBRA .................................................................................................... 26
SERIES OF BLOCKS ......................................................................................................................... 26
PARALLEL BLOCKS......................................................................................................................... 26
FEEDBACK CONTROL LOOP ............................................................................................................ 27
EQUIVALENT DIAGRAMS ................................................................................................................ 28
7 SYSTEM RESPONSE...................................................................................................................... 29
1ST ORDER SYSTEMS ....................................................................................................................... 29
2ND ORDER SYSTEMS....................................................................................................................... 32
SPECIFICATIONS WITHIN THE TIME DOMAIN.................................................................................. 35
8 SYSTEM STABILITY ..................................................................................................................... 36
9 SENSITIVITY .................................................................................................................................. 38
10 DYNAMIC BEHAVIOR .................................................................................................................. 40
2| Chemical Process Control, Dr. Robert N. Grass

P-CONTROLLER .......................................................................................................................... 40
STEADY STATE ERROR OF SYSTEMS WITH P-CONTROLLER ...................................................... 43
11 IMPROVED CONTROLLERS ...................................................................................................... 44
INTEGRAL-CONTROLLER............................................................................................................ 44
DIFFERENTIAL CONTROLLER ..................................................................................................... 46
COMPARISON OF CONTROLLER RESPONSES .............................................................................. 46
PID-CONTROLLER ...................................................................................................................... 47
DESIGN OF PID CONTROLLERS .................................................................................................. 48
12 FREQUENCY RESPONSE ............................................................................................................. 52
BODE DIAGRAM ......................................................................................................................... 54
NYQUIST DIAGRAM .................................................................................................................... 62
TIME DELAY ............................................................................................................................... 65
13 PROBLEMS WITH CURRENT CONTROL LOOPS ................................................................. 68
FEED FORWARD CONTROL ......................................................................................................... 68
CASCADE CONTROL ................................................................................................................... 69
MULTIVARIABLE CONTROL ....................................................................................................... 69
MODEL PREDICTIVE CONTROL .................................................................................................. 71
Chemical Process Control, Prof. Robert N. Grass |3

1 Introduction

Engineer’s View of the World

Changes of state can be described by differential equations that transform the current steady state, that
depends on certain input variables and some additional external variables into another steady state.

Figure 1-1. Different state transformations showing the important variables.

What is Control?

In short terms: Do as I say!

But how can you be sure that it is really done?
1) Do R!
2) Did you get R? No: Change something
Yes: Keep everything as it is
Telling and checking make up control as you know it from teaching kids or pets.

EXAMPLE 1.2-1: Maintaining Body Temperature Movemen Blood

t Circulation
Our body temperature is controlled by
different factors, which you can see in Figure 1-2. Sweat Clothing
37°
C
Figure 1-2. Influence factors on body
temperature.
4| Chemical Process Control, Dr. Robert N. Grass

EXAMPLE 1.2-2: Oven

The simplest idea of a temperature control system for an

oven can be seen in Figure 1-3.

Do T=170°C!

If T < 170°C → heat

If T > 170°C → do nothing Figure 1-4. Oven.

EXAMPLE 1.2-3: Production plant

Work
A very general flowchart of a chemical production plant
plant is depicted in Figure 1-4. In order that the plant can Material Energy
produce the outputs properly all inputs need to be
controlled.
Plant

Output

Figure 1-3. Plant flowdiagram.

Basic Elements of Process Control

u(t) Process y(t)

input output
manipulated variable controlled variable

Figure 1-5. Basic elements of Process Control

Process (in general) = machine/system equipment + actuators (servomotors) + measuring device (sensors)
Chemical Process Control, Prof. Robert N. Grass |5

History of Process Control

Figure 1-6. Drebbels incubator (Brutkasten)

Figure 1-7. James Watts governor for steam machines.

6| Chemical Process Control, Dr. Robert N. Grass

Process Control in Chemical Industry

Goals of the Industry: Maximize: output

Minimize: energy, work, raw materials

→ This is Chemical Engineering

The ideal case: Why not ?:

Design plant (steady state) → Turn plant Nature (external variables) not taken into
on → Walk away → DOESN’T WORK! account (temperature at
day/night/seasons, unreliable products)

Machine Nature

Chemical Engineering Chaos

Process Control

Design Dynamic

EXAMPLE 1.5-1: Coca Cola Plant

The mixing process has to run at steady state to

guarantee for the same quality of the product at any
time, but there are some problems:
• Sugar from sugar beets has different
concentrations depending on the season
etc.
• Vanilla concentration changes, too.
• Phosphoric acid from stone → natural Figure 1-8. Simplified Coca Cola mixing process.
product and therefore also unsteady.
Chemical Process Control, Prof. Robert N. Grass |7

2 Representation of Systems
Systems in process control are represented by block representation:

System

Input u(t) Mathematical function Output y(t)

g(f)

Figure 2-1. Simplified representation of a system that transfers

an input into an output.

Mathematically described by:

𝑦(𝑡) = 𝑔(𝑢(𝑡))

where g is a transfer function.

Variable with time dependence (vector of variables)

Functions transfer one variable into another
+ Mathematical operations
+

SISO Single input single output (most of this course)

MIMO Multiple input multiple output. In this case u(t) and y(t) are vectors (see last chapter).

Feedforward Control

Feedforward control is also called open loop control (deutsch: Steuerung). The controller produces a
signal u(t), which is independent of the current value of y(t).

r(t) Open loop u(t) y(t)

Process
controller

Figure 2-2. Block representation Feedforward Control.

r(t) reference point

u(t) input Just telling, no checking
y(t) output → “stupid” system because it
Controller mathematical construct/function can’t react to changes and
works only “by design”
Goal y(t) = r(t)
8| Chemical Process Control, Dr. Robert N. Grass

Feedback Control

Feedback control is also called closed loop control (deutsch: Regelung).

The controller produces u(t) based on feedback of the real value of the controlled variable y(t) and
comparison with the desired value r(t).

r(t) + e(t) u(t) y(t)

- Controller Process

ym(t)
Sensor

Figure 2-3. Block representation of feedback control.

e(t) error = r(t) – ym(t)

r(t) reference point
u(t) input → “intelligent” system, which
y(t) output can react to changes but has a
Controller mathematical construct/function disadvantage because of the
Sensor measures y(t) measurement that has to be done.
ym(t) measured output

EXAMPLE 2.2-1: An Easy Static Model

It is given that:
10 𝑚𝑝ℎ 𝑐ℎ𝑎𝑛𝑔𝑒 𝑜𝑓 𝑣𝑒𝑙𝑜𝑐𝑖𝑡𝑦
1° 𝑡ℎ𝑟𝑜𝑡𝑡𝑙𝑒 𝑎𝑛𝑔𝑙𝑒 𝑐ℎ𝑎𝑛𝑔𝑒
And:
0.5 𝑚𝑝ℎ 𝑐ℎ𝑎𝑛𝑔𝑒 𝑜𝑓 𝑣𝑒𝑙𝑜𝑐𝑖𝑡𝑦
1 % 𝑟𝑜𝑎𝑑 𝑔𝑟𝑎𝑑𝑒 𝑐ℎ𝑎𝑛𝑔𝑒

Therefore, part of the control system will be given by:

w

0.5

-
r(t) u(t) y(t)
∑ Controller ∑ 10
+ +

Figure 3-5. Control cycle of a car speed system. Green

arrows stand for feedback control.
Chemical Process Control, Prof. Robert N. Grass |9

In the case of open loop control only the black part is relevant (Controller = 1/10). For the velocity
you get:

𝑦𝑜𝑙 = 10(𝑢 − 0.5 𝑤)

𝑟
= 10 ( − 0.5 𝑤)
10
= 𝑟 − 5𝑤

This means that on a road with no grade (w = 0): 𝑦𝑜𝑙 = 55 𝑚𝑝ℎ = 𝑟. But on a road where w = 1:
𝑦𝑜𝑙 = 50 𝑚𝑝ℎ ≠ 𝑟 this are already 10 % error.

In the case of closed loop control also the green arrows have to be considered (assume controller
gain= 100). For the velocity you get:
𝑦𝑐𝑙 = 10𝑢 − 5𝑤
𝑢 = 100(𝑟 − 𝑦𝑐𝑙 )
And combined:
𝑦𝑐𝑙 = 1000𝑟 − 1000𝑦𝑐𝑙 − 5𝑤
1001 𝑦𝑐𝑙 = 1000𝑟 − 5𝑤
𝑦𝑐𝑙 = 0.999𝑟 − 0.005𝑤

For a grade where w = 1 and r = 55 mph 𝑦𝑜𝑙 = 54.94 𝑚𝑝ℎ (0.1% error). For w = 0,
𝑦𝑜𝑙 = 54.945 𝑚𝑝ℎ → there will be also an error for flat roads = remaining steady state error.

Goal of Feedback Control

• Compensation of ‘uncertainties’ (disturbances)
• Uncertainties can be compensated with bigger gains (factor that multiplies with the difference)

Problems
• Feedback gain cannot be randomly chosen (stability problems)
• Steady state error remains
10 | Chemical Process Control, Dr. Robert N. Grass

3 System Modelling
Why should the systems be modelled? → understanding of the problem/system
→ enable manipulation and control of the system

Purpose of Modelling Dynamic Systems

• Understanding through analysis of model properties and simulation
• Control unit design and testing
• Model as explicit integral part of control units (model predictive control)

Two different types of modelling methods

• Establish fundamental physical laws and constitutive relationships = fundamental model
• Adjusting of mathematical (not based on physics) descriptions based on experimental data = black
box model
→ in practice mostly a combination of both models is used.

EXAMPLE 3.0-1: Car with Cruise Control

Goal: car should drive with constant speed

Assumptions: Ignore impacts that do not contribute a lot to the problem

e.g. loss of mass due to burning fuel

Fmotor Mass Ffriction

m

Figure 3-1. Simplified model of problem.

Mathematical description:
𝑚𝑥ሷ = 𝐹𝑚𝑜𝑡𝑜𝑟 − 𝐹𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛

For air resistance it holds that the friction force is proportional to the velocities and you have two
different regimes:

If v small 𝐹𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛 = 𝑏1 𝑥ሶ
If v high: 𝐹𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛 = 𝑏2 𝑥 2ሶ
Therefore, we can write for small v:
𝑏1 𝐹𝑚𝑜𝑡𝑜𝑟
𝑥ሷ + 𝑥ሶ =
𝑚 𝑚
Chemical Process Control, Prof. Robert N. Grass | 11

How does accelerating work:

The throttle valve is what should be controlled with the controlling system.

Throttle valve

Air
Cylinder
Fuel

Figure 3-2. Acceleration system.

Road grade
Manipulated Actuator Process
variable
Controller Engine Auto body
Desired Throttle Actual
speed speed
Sensor

Speedometer
Measured
speed

Figure 3-3. Control cycle of the acceleration system in a car with cruise control.

The control cycle is represented in Figure 3-3.

EXAMPLE 3.0-2: Car Damping

If mass is distributed uniformly then:

𝑚2 = 0.25 𝑀𝑐𝑎𝑟
Physical modelling of springs:
Hook’s Law: 𝐹 = 𝑘 ∆𝑙
Damper: 𝐹=𝑑𝑣
Where d is the damping constant.

Therefore, the forces are:

𝐹𝑤ℎ𝑒𝑒𝑙 = 𝑘1 (𝑥 − 𝑟)
𝐹𝑠𝑝𝑟𝑖𝑛𝑔 = 𝑘2 (𝑦 − 𝑥)
𝐹𝑑𝑎𝑚𝑝𝑒𝑟 = 𝑑2 (𝑦ሶ − 𝑥ሶ )

Newton: Figure 3-4. Problem description.

𝑚1 𝑥ሷ = −(𝑥 − 𝑟)𝑘1 + (𝑦ሶ − 𝑥ሶ )𝑑2 + (𝑦 − 𝑥)𝑘2
𝑚2 𝑦ሷ = −(𝑦ሶ − 𝑥ሶ )𝑑2 − (𝑦 − 𝑥)𝑘2
12 | Chemical Process Control, Dr. Robert N. Grass

The process follows:

r(t) G y(t)
(transfer function)

Figure 3-5. Block description for car damping.

EXAMPLE 3.0-3: Pendulum

If an external torque 𝑇𝑐 is exerted on the pendulum the

following equations hold:
𝑇𝑐
Newton:
angular movement 𝐼 𝜃ሷ = ∑ 𝑇 = 𝑇𝑐 − 𝑚𝑔𝑙 sin 𝜃 ϴ
inertia 𝐼 = 𝑚𝑙 2 l
harmonic oscillator 𝑔
= 𝜔2
(if 𝑇𝑐 =0) 𝑙
𝑚𝑥ሷ = ∑𝐹
m

And from force equilibrium:

mg
𝑚 𝑙 2 𝜃ሷ = 𝑇𝑐 − 𝑚𝑔𝑙 sin 𝜃
𝑔 𝑇𝑐
𝜃ሷ + sin 𝜃 =
𝑙 𝑚𝑙 2 Figure 3-6. Problem description.

For 𝜃 ≪ 1 it holds that sin 𝜃 = 𝜃, therefore the solution is:

𝑔 𝑇𝑐
𝜃ሷ + 𝜃=
𝑙 𝑚𝑙 2

The corresponding block diagram is as follows:

𝑇𝑐 G ϴ
(transfer function)
(t)
Figure 3-7. Block description of a pendulum with external
torque.
Chemical Process Control, Prof. Robert N. Grass | 13

EXAMPLE 3.0-4: Water Tank

Mass balances: win

𝑑𝑀
= 𝑤𝑖𝑛 − 𝑤𝑜𝑢𝑡
𝑑𝑡
𝑑 (ℎ 𝐴𝑔 𝜌)
= 𝑤𝑖𝑛 − 𝑤𝑜𝑢𝑡
𝑑𝑡 h
Where 𝜌 and 𝐴𝑔 are independent of t, therefore: A0
wout
𝐴𝑔 𝜌 𝑑ℎ
= 𝑤𝑖𝑛 − 𝑤𝑜𝑢𝑡 Ag
𝑑𝑡
Figure 3-8. Problem description.
wout can be found using the Bernoulli equation:
1
𝑔 𝜌 ℎ + 𝜌𝑣 2 + 𝑝 = 𝑐𝑜𝑛𝑠𝑡.
2
By writing the Bernoulli equation for the top (v = 0) and bottom (h = 0) of the tank we can get the
following expression for the bottom velocity:
𝑣𝑏𝑜𝑡𝑡𝑜𝑚 = ඥ2 𝑔 ℎ
And finally:
𝑤𝑜𝑢𝑡 = 𝐴0 ඥ2 𝑔 ℎ 𝜌
By plugging this into the mass balance and rearranging for 𝑑ℎ/𝑑𝑡:
1 1
ℎሶ = (𝑤𝑖𝑛 − 𝑤𝑜𝑢𝑡 ) = (𝑤𝑖𝑛 − 𝜌 𝐴0 ඥ2𝑔ℎ)
𝜌 𝐴𝑔 𝜌 𝐴𝑔
Where 𝑤𝑖𝑛 has units: kg s-1

EXAMPLE 3.0-5: Thermocouple

Problem parameters:
Thermocouple

𝐴 Surface area thermocouple

𝐶𝑝 Heat capacity thermocouple
ℎ Heat transfer coefficient
𝛾 Thermo element constant
T
𝑦 Measured voltage 𝑦 = 𝛾𝑇 ′
𝑇 Temperature of fluid Figure 3-9. Problem description (side view).
𝑇′ Temperature of thermocouple
𝑀 Mass of thermometer
y
There will be a temperature gradient between the T‘
heat transfer
thermocouple and the surrounding water, that can
be described by Fig 3.10: T

Figure 3-10. Problem description (top view).

14 | Chemical Process Control, Dr. Robert N. Grass

Energy balance:
𝑑𝐸
= 𝐸𝑖𝑛 − 𝐸𝑜𝑢𝑡
𝑑𝑡
𝑑(𝑀𝐶𝑝 𝑇 ′ )
= ℎ𝐴(𝑇 − 𝑇 ′ )
𝑑𝑡
𝑦 𝑦 = 𝛾𝑇 ′
𝑑 (𝑀𝐶𝑝 𝛾 ) 𝑦
= ℎ𝐴 ൬𝑇 − ൰
𝑑𝑡 𝛾
𝐶𝑝 𝑀 𝑑𝑦
+ 𝑦 = 𝛾𝑇
ℎ𝐴 𝑑𝑡

𝐶𝑝 𝑀
Where is the time constant 𝜏 of the Thermocouple, i.e. high A and low M desired.
ℎ𝐴
The measured temperature is time dependent no matter how good the sensor. It will approach
the true temperature only asymptotically but with a speed proportional to 1/τ.

EXAMPLE 3.0-6: Heating

Water is heated in a radiator with hot air. We assume 𝑇𝐿 = 𝑇𝐿0 and 𝑇𝑊 = 𝑇𝑊 0
(instantaneous heat
transfer and ideal mixing) like in a CSTR. The temperatures inside the radiator are independent of
space but will change with time. The energy balance reads as follows:

𝑑𝐸 𝑑𝑇𝐿 𝑇𝐿 − 𝑇𝑊
= 𝑚𝐿 𝑐𝐿 = 𝑤𝐿 𝑐𝐿 𝑇𝐿′ − 𝑤𝐿 𝑐𝐿 𝑇𝐿 −
𝑑𝑡 𝑑𝑡 𝑅
𝑑𝐸 𝑑𝑇𝑊 ′
𝑇𝐿 − 𝑇𝑊
= 𝑚𝑊 𝑐𝑊 = 𝑤𝑊 𝑐𝑊 𝑇𝑊 − 𝑤𝑊 𝑐𝑊 𝑇𝑊 +
𝑑𝑡 𝑑𝑡 𝑅

Where 𝑐 is the heat capacity and 𝑇 the temperature of the liquid and water, 𝑤 is the mass flux
and 𝑅 the resistance to heat transfer. This is a MIMO system and already quite complicated. It
can be simplified when we look only at one in- and outlet and keep the other two constant so
we can treat it like a SISO.

Figure 3-11. Problem description Radiator.

Chemical Process Control, Prof. Robert N. Grass | 15

4 Differential Equations
General scheme of how you can deal with differential equations:

Nonlinear differential
equations

Linear differential
Solve numerically
equations

Solve analytically Understand

Nomenclature

PDE Partial differential equations 𝜕𝑢 𝜕2𝑢

More than 2 variables differentiated =𝛼 2
𝜕𝑡 𝜕𝑥

ODE Ordinary differential equations 𝑑𝑢

𝑑𝑡
=𝑢 first order
Order: highest differential 2
𝑑 𝑢 𝑑𝑢
𝑑𝑡 2
+ 4 𝑑𝑡 =2 second order

Linear: only linear terms in u 𝑛−1

(𝑛) (𝑥)
𝑢 = ∑ 𝑎𝑘 (𝑥)𝑢(𝑘) (𝑥) + 𝑔(𝑥)
𝑘=0

Example: 𝑑2 𝑢
𝑑𝑡
+ 𝜔0 𝑢 =0 linear second order
𝑑𝑢
𝑑𝑡
= 𝑢2 non-linear system
16 | Chemical Process Control, Dr. Robert N. Grass

Numerical Solution

𝑑𝑥 Δ𝑥
𝑥ሶ = 𝑓(𝑥) = ≈ ≈ 𝑓(𝑥)
𝑑𝑡 Δ𝑡

𝑥𝑖+1 − 𝑥𝑖
𝑥𝑖+1 ≈ 𝑓(𝑥)
Δ𝑡
𝑥𝑖
tangent in Forward Euler Equation:
ti 𝑥𝑖+1 = 𝑥𝑖 + Δ𝑡 𝑓(𝑥𝑖 )
slope = f(xi)
∆t
If Δ𝑡 → 𝑑𝑡 → No error

ti ti+1 t

Figure 4-1. Euler method.

Analytical Solution

General form of an ODE 𝑥ሶ = 𝐴𝑥 + 𝐵𝑢

Homogeneous 𝑥ሶ = 𝐴𝑥
𝑥(𝑡) = 𝐶𝑒 𝐴𝑡 𝐶 = 𝑐𝑜𝑛𝑠𝑡.

Inhomogeneous 𝑥ሶ = 𝐴𝑥 + 𝐵𝑢

Guess solution (variation of parameters)

𝑥(𝑡) = 𝑒 𝐴𝑡 ⋅ 𝐶(𝑡)
Differential of guess
𝑥ሶ (𝑡) = 𝐴𝐶(𝑡)𝑒 𝐴𝑡 + 𝑒 𝐴𝑡 ⋅ 𝐶ሶ (𝑡)
Compose with original equation
𝑥ሶ (𝑡) = 𝐴𝑥 + 𝐵𝑢
𝐵𝑢 = 𝑒 𝐴𝑡 ⋅ 𝐶ሶ (𝑡)
𝐶ሶ (𝑡) = 𝐵𝑢 𝑒 −𝐴𝑡
𝑡
At
x(t) = e ⋅ (𝐶(𝑡0 ) + ∫ 𝐵𝑢(𝜏)𝑒 −𝐴𝜏 𝑑𝜏) with C(t 0 ) = x0
𝑡0
𝑡
→ 𝑥(𝑡) = 𝑒 𝐴𝑡 (𝑥0 + ∫0 𝑒 −𝐴𝜏 𝐵𝑢(𝜏)𝑑𝜏)
Chemical Process Control, Prof. Robert N. Grass | 17

Linearization

Assume for small Δx: f(x) is linear, this is only valid in the near neighborhood of x.
d𝑓
𝑓(𝑥) ≈ 𝑓(𝑥̅ ) + | (𝑥 − 𝑥̅ )
d𝑥 𝑥̅
In vector format:
𝑓1 𝑥1
𝑓 𝑥2
𝑓 = ( 2) 𝑥 = (…)
…
𝑓𝑛 𝑥𝑛

𝑓(𝑥) ≈ 𝑓(𝑥̅ ) + J|𝑥̅ (𝑥 − 𝑥̅ )

d𝑓1 d𝑓1
…
d𝑓 d𝑥1 d𝑥2
J= =
d𝑥 𝑇 d𝑓2
…
(d𝑥1 )

For DGL-Systems:

𝑢 input 𝑢 𝑥ሶ = 𝑓(𝑥, 𝑢) 𝑦
𝑦 output 𝑦 = 𝑔(𝑥, 𝑢)

𝑥 state
𝑥ሶ , 𝑦 transfer Linearization
functions
∆𝑢 ∆𝑥ሶ = 𝐴 Δ𝑥 + 𝐵 Δ𝑢 ∆𝑦
Δ𝑦 = 𝐶 Δ𝑥 + 𝐷 Δ𝑢

Figure 4-2. Linearization of DGL Systems.

Linearization of 𝑥ሶ gives:
𝑥ሶ = 𝑓(𝑥, 𝑢)
d𝑓 d𝑓
𝑥ሶ ≈ 𝑓(𝑥̅ , 𝑢̅) + | (𝑥 − 𝑥̅ ) + | (𝑢 − 𝑢̅)
d𝑥 𝑥̅ d𝑢 𝑢̅
≈ 0 + 𝐴(𝑥 − 𝑥̅ ) + 𝐵(𝑢 − 𝑢̅)
→ 𝑥ሶ = 𝐴 Δ𝑥 + 𝐵 Δ𝑢 = Δ𝑥ሶ
Same with y gives:
𝑦 = 𝑔(𝑥, 𝑢)
d𝑔 d𝑔
𝑦 ≈ 𝑔(𝑥̅ , 𝑢̅) + | (𝑥 − 𝑥̅ ) + | (𝑢 − 𝑢̅)
d𝑥 𝑥̅ d𝑢 𝑢̅
≈ 𝑦̅ + 𝐶(𝑥 − 𝑥̅ ) + 𝐷(𝑢 − 𝑢̅)
→ Δy = 𝐶 Δ𝑥 + 𝐷 Δ𝑢
18 | Chemical Process Control, Dr. Robert N. Grass

EXAMPLE 4.4-1: Water Tank

Remember from Example 3.0-4:

𝑑ℎ 1 win
= (𝑤 − 𝐴0 𝜌ඥ2𝑔ℎ)
𝑑𝑡 𝐴𝜌 𝑖𝑛
Constant A0, ρ, A, 𝑔
Variables h, win

Nonlinear in h h
𝑑ℎ A0
=𝑓
𝑑𝑡
𝜕𝑓 𝜕𝑓 Ag
𝑓 ≈ 𝑓𝑠𝑠 + | Δℎ + | Δ𝑤𝑖𝑛 Figure 4-3. Problem description.
𝜕ℎ 𝑠𝑠 𝜕𝑤𝑖𝑛 𝑠𝑠

Steady state:
𝑑ℎ 1
0= = (𝑤 𝑠𝑠 − 𝐴0 𝜌ඥ2𝑔ℎ 𝑠𝑠 )
𝑑𝑡 𝐴𝜌 𝑖𝑛

𝑠𝑠 2
1 𝑤𝑖𝑛
ℎ 𝑠𝑠 = ቆ ቇ
2𝑔 𝜌𝐴0
Linearization:
𝐴0 ඥ2𝑔ℎ 𝑠𝑠 1
𝑓 ≈0− Δℎ + Δ𝑤
𝐴 2ℎ 𝑠𝑠 𝐴𝜌 𝑖𝑛

𝑑Δℎ 𝑑(ℎ − ℎ 𝑠𝑠 ) 𝑑ℎ 𝑑ℎ 𝑠𝑠 𝑑ℎ
= = − =
𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑡

𝑑ℎ 𝑑Δℎ
𝑓= = 𝐴′Δℎ + 𝐵′Δ𝑤𝑖𝑛 =
𝑑𝑡 𝑑𝑡

Therefore, we get the following linear equation:

′
𝜌𝐴20 𝑔 1
𝐴 = − 𝑠𝑠 ; 𝐵′ =
𝐴𝑤𝑖𝑛 𝐴𝜌
Valid only close to steady state.
Chemical Process Control, Prof. Robert N. Grass | 19

Superposition

If an input signal is complex, you could split the complex input into multiple additive simpler inputs. If
the system is linear, the output of the complex input can be calculated by summing up the outputs of the
simpler inputs.

u1 y1
Linear Transfer
u2 Function y2

u3 = au1 + bu2 y3 = ay1 + by2

Figure 4.4. Superposition of multiple input signals.

EXAMPLE 4.5-1: Superposition

𝑦ሶ + 𝑘𝑦 = 𝑢

1) 𝑦1ሶ + 𝑘𝑦1 = 𝑢1

2) 𝑦2ሶ + 𝑘𝑦2 = 𝑢2

3) 𝑢3 = 𝑎𝑢1 + 𝑏𝑢2

𝑦3ሶ + 𝑘𝑦3 = 𝑢3

Demonstrate that 𝑦 = 𝑦1 + 𝑦2

Which means: 𝑦ሶ = 𝑦ሶ1 + 𝑦ሶ 2

and 𝑢 = 𝑢1 + 𝑢2

(𝑎𝑦1ሶ + 𝑏𝑦2ሶ ) + 𝑘(𝑎𝑦1 + 𝑏𝑦2 ) = 𝑎𝑢1 + 𝑏𝑢2

𝑎(𝑦1ሶ + 𝑘𝑦1 − 𝑢1 ) + 𝑏(𝑦2ሶ + 𝑘𝑦2 − 𝑢2 ) = 0

Because 𝑦ሶ + 𝑘𝑦 − 𝑢 = 0 ▪
20 | Chemical Process Control, Dr. Robert N. Grass

Convolution

Simple inputs → g(t) → outputs

Input Output
𝑢(𝑡) 𝑦(𝑡)

Pulse 𝛿(𝑡) 𝑔(𝑡)

pulse response

Pulse with time delay 𝛿 (𝑡 − 𝜏) 𝑔(𝑡 − 𝜏)

Weighted pulse 𝑢(𝜏) 𝛿 (𝑡 − 𝜏) 𝑢(𝜏) 𝑔 (𝑡 − 𝜏)

+∞ +∞
Generic Input 𝑢(𝑡)
𝑢(𝑡) = ∫ 𝑢(𝜏)𝛿(𝑡 − 𝜏)𝑑𝜏 𝑦(𝑡) = ∫ 𝑢(𝜏)𝑔(𝑡 − 𝜏)𝑑𝜏
−∞ −∞

+∞
𝑦(𝑡) = ∫ 𝑢(𝜏)𝑔(𝑡 − 𝜏)𝑑𝜏
−∞
𝑦=𝑢 ∗𝑔

If 𝑢(𝑡) = 0 for 𝑡<0

then +∞
𝑦(𝑡) = ∫ 𝑢(𝜏)𝑔(𝑡 − 𝜏)𝑑𝜏
0

And if 𝑔(𝑡) = 0 for 𝑡<0

additionally

then 𝑡
𝑦(𝑡) = ∫ 𝑢(𝜏)𝑔(𝑡 − 𝜏)𝑑𝜏
0
Chemical Process Control, Prof. Robert N. Grass | 21

5 Laplace Transformation

Laplace Transformation

The Laplace transformation of a function 𝑓(𝑡) (𝑓(𝑡) = 0, 𝑡 < 0) is defined as

∞
∫ 𝑓(𝑡)𝑒 −𝑠𝑡 𝑑𝑡 = 𝐹(𝑠) = ℒ{𝑓(𝑡)}
0

∞
(Which is equivalent to the “continuous power series” ∫0 𝑓(𝑡)𝑥 𝑡 𝑑𝑡 with 𝑠 = −ln(𝑥).
It exists for all functions, which fulfill the condition of Dirichlet:

∞
∫ |𝑓(𝑡)|𝑒 −𝜎𝑡 𝑑𝑡 < ∞
0
For a large enough 𝜎

EXAMPLE 5.1-1: for f(t) = 1

𝑡=𝐴
∞
𝑒 −𝑠𝑡 𝑒 −𝑠𝐴 1 1
𝐹(𝑠) = ∫ 1 𝑒 −𝑠𝑡 𝑑𝑡 = lim − | = lim ቆ + ቇ=
0 𝐴→∞ 𝑠 𝑡=0 𝐴→∞ 𝑠 𝑠 𝑠

Inverse Laplace Transformation

1 𝜎 +𝑗∞ 𝑓(𝑡) for 𝑡 ≥ 0

ℒ −1 {𝐹(𝑠)} = ∫𝜎 𝑗−𝑗∞ 𝑒 𝑠𝑡 𝐹(𝑠) 𝑑𝑠 = {
2𝜋𝑖 𝑗 0 for 𝑡 < 0

If 𝐹(𝑠) is a fraction, the easiest way to find 𝑓(𝑡) is the partial fraction expansion of 𝐹(𝑠) and the
use of table (5.3) of simple functions.

𝑝(𝑠) p(s), q(s) are polynomials; order of p(s) ≤ order of q(s) (true for all
ℒ −1 {𝑞(𝑠)};
physical systems).

𝑛
𝑝(𝑠) 𝐴𝑖 𝐴1 𝐴2 𝐴𝑛
= ∑ = + + ⋯+
𝑞(𝑠) 𝑠 − 𝑠𝑖 𝑠 − 𝑠1 𝑠 − 𝑠2 𝑠 − 𝑠𝑛
𝑖=1

𝑠𝑖 = roots of q(s) = 0, 𝐴1, 𝐴2, … partial fractions

22 | Chemical Process Control, Dr. Robert N. Grass

Important Laplace Transformations

The following LPT have to be remembered.

𝒇(𝒕) 𝑭(𝒔)

𝛿(𝑡) 1

𝛿(𝑡 − 𝑎) 𝑒 −𝑎𝑠

1
1 𝑠

1
𝑡
𝑠2

1
𝑒 −𝑎𝑡
𝑠+𝑎

1
𝑡𝑒 −𝑎𝑡 (𝑠 + 𝑎)2

1 −𝑎𝑡 1
𝑒 sin 𝑏𝑡
𝑏 (𝑠 + 𝑎)2 + 𝑏 2

𝑎
1 − 𝑒 𝑎𝑡
𝑠(𝑎 − 𝑠)

𝑛!
𝑡𝑛
𝑠 𝑛+1

𝑎
sin(𝑎𝑡)
𝑠2 + 𝑎2

𝑠
cos(𝑎𝑡)
𝑠2 + 𝑎2
Chemical Process Control, Prof. Robert N. Grass | 23

Properties of Laplace Transformations

1) Time shift
𝑓(𝑡 − 𝜏) ↔ 𝐹(𝑠) 𝑒 −𝜏𝑠

2) Superposition
𝑎𝑓1 (𝑡) + 𝑏𝑓2 (𝑡) ↔ 𝑎𝐹1 (𝑠) + 𝑏𝐹2 (𝑠)

3) Convolution
𝑓1 (𝑡) ∗ 𝑓2 (𝑡) ↔ 𝐹1 (𝑠) ⋅ 𝐹2 (𝑠)
𝑡
∫ 𝑓1 (𝑡) 𝑓2 (𝑡 − 𝜏)𝑑𝜏
0

4) Final value theorem

lim 𝑓(𝑡) = lim 𝑠 ⋅ 𝐹(𝑠) (not applicable if 𝑓 diverges e.g. maximum one node in the origin
𝑡→∞ 𝑠 →0
and the real part of all other nodes is negative)

5) Initial value theorem

lim 𝑓(𝑡) = lim 𝑠 𝐹(𝑠)
𝑡→0 𝑠→∞

6) Differentiation

𝑓 ′ (𝑡); 𝑓(0) ↔ 𝑠𝐹(𝑠) − 𝑓(0)

𝑓 ′′ (𝑡); 𝑓(0); 𝑓 ′ (0) ↔ 𝑠 2 𝐹(𝑠) − 𝑠 𝑓(0) − 𝑓′(0)

𝑓 𝑛 (𝑡); 𝑓(0) = 0; 𝑓 𝑛−1 (0) = 0 ↔ 𝑠 𝑛 𝐹(𝑠) All initial conditions are 0.

7) Integral:
𝑡 𝐹(𝑠)
∫ 𝑓(𝜏)𝑑𝜏 ↔ 𝑠
0

EXAMPLE 5.4-1: Laplace Transformation

d𝑦
+ 𝑎𝑦 = 0
d𝑡
𝑠⏟𝑌(𝑠) − 1 + 𝑎𝑌(𝑠) = 0
𝑠(𝑌(0)−𝑓(0))
1
𝑌(𝑠) = → 𝑦(𝑡) = 𝑒 −𝑎𝑡
𝑠+𝑎
24 | Chemical Process Control, Dr. Robert N. Grass

EXAMPLE 5.4-2: Linear Dynamic System

d 𝑦(𝑡)
𝜏 + 𝑦(𝑡) = 𝛾 𝑇(𝑡)
d𝑡
𝜏(𝑠 𝑌(𝑠) − 𝑦(0)) + 𝑌(𝑠) = 𝛾𝑡(𝑠)
𝛾 𝜏(𝑦(0))
𝑌(𝑠) = 𝑇(𝑠) +
𝜏𝑠+1 𝜏𝑠 + 1

EXAMPLE 5.4-3: Final Value Theorem

→ REMEMBER: the final value theorem can only be used if the finite final value exists.

Easy example:
3(𝑠 + 2)
𝑌(𝑠) =
𝑠(𝑠 2
+ 2𝑠 + 10)
3(𝑠 + 2) 3∙2
𝑦(𝑡 → ∞) = lim 𝑠 2
= = 0.6
𝑠→0 𝑠(𝑠 + 2𝑠 + 10) 10

But if there is no finite final value, the final value theorem does not work, even though you might get a
solution:

3
𝑌(𝑠) =
𝑠(𝑠 − 2)
→ with the final value theorem you would get:
3
𝑦(𝑡 → ∞) = −
2
But actually:
3
𝑦(𝑡) = (−1 + 𝑒 2𝑡 ) → lim 𝑦(𝑡) = ∞
2 𝑡→∞
In this case you cannot use the final value theorem because the function diverges.

In General 𝑈(𝑠) 𝐼𝑛−1

𝑌(𝑠) = +
𝑎0 𝑠𝑛 + 𝑎1 𝑠 𝑛−1 + ⋯ + 𝑎𝑛 𝑎0 𝑠 + 𝑎1 𝑠 𝑛−1 + ⋯ 𝑎𝑛
𝑛
Output = Input + Initial condition

Characteristic polynomial 𝑎0 𝑠 𝑛 + 𝑎1 𝑠 𝑛−1 + ⋯ + 𝑎𝑛

!
Characteristic equation 𝑎0 𝑠 𝑛 + 𝑎1 𝑠 𝑛−1 + ⋯ + 𝑎𝑛 =
⏞0

→ Solutions of characteristic equations → roots of system 𝑠𝑖 , 𝑠𝑗

Chemical Process Control, Prof. Robert N. Grass | 25

Initial Conditions

In process control we are often looking at the system behavior relative to a reference state. We start in the
steady state (𝑥𝑠𝑠 . 𝑦𝑠𝑠 , 𝑢𝑠𝑠 ) and look at the deviation variables (𝛥𝑥, 𝛥𝑦, 𝛥𝑢), which represent the deviation
from it. This is useful as we obtain a simple transfer function that expresses changes from the steady state.

Looking at the example of the thermocouple (EXAMPLE 3.0-5), where we had

dy
𝜏 + 𝑦 = 𝛾𝑇
dt

In the steady state dy𝑠𝑠

𝑇 = 𝑇𝑠𝑠 =0
dt

Deviation variables ΔT = T − 𝑇𝑠𝑠 Δy = y − 𝑦𝑠𝑠

dΔy Δ𝑦(0) = 0
𝜏 + Δ𝑦 = 𝛾Δ𝑇
dt Δ 𝑇(0) = 0

Laplace transformation ̅̅̅̅̅̅ + 𝑌(𝑠)

𝜏 𝑠 𝑌(𝑠) ̅̅̅̅̅̅ = 𝛾 𝑇(𝑠)
̅̅̅̅̅̅

𝛾
̅̅̅̅̅̅
𝑌(𝑠) = ̅̅̅̅̅̅
𝑇(𝑠)
𝜏𝑠 + 1

In words: „output = transfer function × input“

26 | Chemical Process Control, Dr. Robert N. Grass

6 Block Diagram Algebra

Block diagram algebra can be used for the modeling of systems. When the initial conditions are zero (see
5.5) the transfer function 𝐺(𝑠) relates the inlet 𝑈(𝑠) to the outlet 𝑌(𝑠).

𝑌(𝑠) = 𝐺(𝑠) ∙ 𝑈(𝑠)

Series of Blocks

Two blocks in series behave multiplicative:

Figure 6-1. Blocks in series.

The transfer function is given by:

𝑌1 = 𝐺1 ∙ 𝑈1
𝑌2 = 𝐺1 ⋅ 𝐺2 ⋅ 𝑈1
𝑌2 (𝑠)
= 𝐺1 ⋅ 𝐺2
𝑈1 (𝑠)

Parallel Blocks

Parallel blocks behave additive as stated by the superposition principle:

Figure 6-2. Parallel blocks.

The transfer function is given by:

𝑌 = 𝑌1 + 𝑌2
𝑌 = (𝐺1 + 𝐺2 ) ⋅ 𝑈
𝑌(𝑠)
= 𝐺1 + 𝐺2
𝑈(𝑠)
Chemical Process Control, Prof. Robert N. Grass | 27

Feedback Control Loop

Y1(s)
Figure 6-3. Control cycle of Feedback control.

Transfer function is given by:

𝑌1 = 𝐺1 ⋅ 𝑈1
𝑈1 = 𝑅 − 𝑌2
𝑌2 = 𝐺2 ⋅ 𝑌1
𝑌1 = 𝐺1 (𝑅 − 𝐺2 𝑌)
𝐺1
𝑌= ⋅𝑅
1 + 𝐺1 ⋅ 𝐺2

Or written in a different way:

Y(s) 𝑻𝑭input to output

= 𝑻𝑭closed loop =
𝑅(𝑠) 𝟏 − 𝑻𝑭loop
In the example above:
𝑻𝑭input to output = 𝐺1
𝑻𝑭loop = −G1 ∙ G2
28 | Chemical Process Control, Dr. Robert N. Grass

Equivalent Diagrams

For the representation of more complex systems, rearranging and the use of the formulas above can be used
to simplify and obtain the transfer function. Additional rules are listed below.

• Nodes can be moved by introducing blocks with the inverse of the block:

Figure 6-4. Moving nodes by introducing the inverse of the block.

• Blocks can be moved in front of a sum by introducing this block to all contributing inputs:

Figure 6-5. Moving a block in front of a sum by introducing it to all inputs.

• The feedback loop block diagram can be rearranged in the following way (more suitable for certain
applications):

Figure 6-6. Rearranging of the feedback control loop.

Chemical Process Control, Prof. Robert N. Grass | 29

7 System Response

The following system is considered

U(s) Y(s)
G(s)

Figure 7-1. First order system.

1st Order Systems

Systems described by a first order ODE have a first order response, e.g. a CSTR without reaction.
The transfer function 𝐺(𝑠) reads:
𝑘
𝐺(𝑠) =
𝜏𝑠 + 1

where 𝑘 is the gain (Verstärkungsfaktor) and 𝜏 is the time constant (Zeitkonstante).

As was introduced in the previous chapter, the transfer function can be used to express the dependence of
the output signal 𝑌(𝑠) on the input signal U(s):

𝑘
𝑌(𝑠) = ∙ 𝑈(𝑠)
𝜏𝑠 + 1

Pulse Input

In this case the input signal U(s) is:

𝑈(𝑠) = 1

Therefore, the Laplace transform of the output signal Y(s) becomes:

𝑘
𝑌(𝑠) = ⋅1
𝜏𝑠 + 1

The back transformation (using the table on p.20) of Y(s) gives the output signal as a function of time y(t):
𝑘 𝑡
𝑦(𝑡) = ∙ 𝑒 −𝜏
𝜏
And the slope at t = 0:
𝑑𝑦 𝑘
| = − 2
𝑑𝑡 𝑡=0 𝜏
𝑘 𝑘
Initial value theorem 𝑦(𝑡 = 0) = lim 𝑠 ⋅ 𝑌(𝑠) = lim 𝑠 ⋅ =
𝑠→∞ 𝑠→∞ 𝜏𝑠 + 1 𝜏
𝑘
Final value theorem 𝑦(𝑡 → ∞) = lim 𝑠 ⋅ 𝑌(𝑠) = lim 𝑠 ⋅ =0
𝑠→0 𝑠→0 𝜏𝑠 + 1
30 | Chemical Process Control, Dr. Robert N. Grass

u(t) y(t)
k/τ

t/τ t/τ

Figure 7-2. Input (left) and output (right) signal of a pulse input.

Step Input

The input signal U(s) of a step function can be written as:

1
𝑈(𝑠) =
𝑠
Therefore, the output signal Y(s) becomes:
𝑘 1
𝑌(𝑠) = 𝐺(𝑠) ⋅ 𝑈(𝑠) = ⋅
𝜏𝑠 + 1 𝑠

Back transformation gives the output as a function of time y(t):

𝑡
𝑦(𝑡) = 𝑘(1 − 𝑒 −𝜏 )

𝑘 1
Initial value theorem 𝑦(𝑡 = 0) = lim 𝑠 ⋅ ⋅ =0
𝑠→∞ 𝜏𝑠 + 1 𝑠
𝑘 1
Final Value Theorem 𝑦(𝑡 → ∞) = lim 𝑠 ⋅ ⋅ =𝑘
𝑠→0 𝜏𝑠 + 1 𝑠
𝑘 1 𝑘
Initial slope 𝑦ሶ (𝑡 = 0) = lim 𝑠 ⋅ 𝑠 ⋅ 𝑌(𝑠) = lim 𝑠 ⋅ 𝑠 ⋅ =
𝑠→∞ 𝑠→∞ 𝜏𝑠 + 1 𝑠 𝜏

Remark: This approach also works for any other derivative.

u(t) y(t)
k

t/τ t/τ
Figure 7-3. Input and output signal of a step input.
Chemical Process Control, Prof. Robert N. Grass | 31

The process signal does not respond immediately to the change in input. The response is delayed and
approaches the new steady state value only asymptotically. After a time equal to the time constant 𝜏 it has
reached 63.2% of the input value.

How does the signal change by varying the parameters k and τ? Note that the axes are no more normalized.

y(t) y(t)

t t
Figure 7-4. Left: Change of output signals with varying parameter k (gain).
Right: Change of output signals with varying parameter τ (time constant).

If k increases: If τ increases:
𝑦(𝑡 ∞) multiplied  y(t) increases slower to asymptote

By changing the gain, the new steady state which the system approaches can be altered. It increases with
increasing gain. When we vary the time constant, the speed with which the system approaches the new
steady state changes. The higher the time constant, the slower the system evolves.

Ramp Input

The Laplace transformed ramp input signal 𝑈(𝑠) is:

1
𝑈(𝑠) =
𝑠2
Therefore, the output signal 𝑌(𝑠) is given by:
𝑘 1
𝑌(𝑠) = ⋅
𝜏𝑠 + 1 𝑠 2
Back transformation gives:
𝑡
𝑦(𝑡) = 𝑘(𝑡 − 𝜏 + 𝜏𝑒 −𝜏 )
32 | Chemical Process Control, Dr. Robert N. Grass

Initial value theorem 𝑦(𝑡 = 0) = 0

Final slope 𝑦ሶ (𝑡 → ∞) = 𝑘
Initial slope 𝑦ሶ (𝑡 = 0) = 0

y(t)
kτ

t/τ
Figure 7-5. Input and output signal in case of a ramp input.

Note: If you plot the input and output signal in non-normalized axes, then the difference between the two
signals would be the time constant τ.

2nd Order Systems

Most mechanical systems or two coupled systems of 1st order are 2nd order systems. All systems with 2nd
order derivatives have a second order response, e.g. pendulum and many other mechanical systems (𝐹 =
𝑚𝑥ሷ ).

The transfer function G(s) reads:

𝑘 𝑘𝜔𝑛2
𝐺(𝑠) = =
𝜏 2 𝑠 2 + 2𝜏𝜉𝑠 + 1 𝑠 2 + 2𝜉𝜔𝑛 𝑠 + 𝜔𝑛2

where 𝑘 is the gain, 𝜏 the time constant, 𝜔𝑛 the natural frequency and 𝜉 the damping factor.

The roots of the characteristic polynomial

𝜏 2 𝑠 2 + 2𝜏𝜉𝑠 + 1

i.e. the solutions of the characteristic equation

𝜏 2 𝑠 2 + 2𝜏𝜉𝑠 + 1 = 0
Chemical Process Control, Prof. Robert N. Grass | 33

1
𝑠1,2 = − (𝜉 ± ඥ𝜉 2 − 1)
𝜏

We can describe the shape of the solution of these systems by looking at the damping factor 𝜉:

0<𝜉<1 roots are imaginary → underdamped system, oscillation

1<𝜉 roots are real → overdamped system, no oscillation
1=𝜉 only 1 discrete root → critically damped

Step input

The output signal Y(s) of a step function of a second order system looks as follows:

𝑘 1
𝑌(𝑠) = ⋅
𝜏 2 𝑠 2 + 2𝜏𝜉𝑠 + 1 𝑠

For a graphical representation of the different cases of damping for a second order system see Figure 7-6.

y(t)

t/τ
Figure 7-6. Step response of a second order system with different damping values and 𝒌 = 𝟏.

Higher Order Systems

Step inputs for higher order systems can be formulated analogously (see Figure 7-7). This also models the
transition of a CSTR to a PFR as a cascade of infinite CSTRs.

𝑛
𝑘 1
𝑌(𝑠) = ൬ ൰ ⋅
𝜏𝑠 + 1 𝑠
34 | Chemical Process Control, Dr. Robert N. Grass
𝑛
1 ℒ −1
lim ( ) = 𝑒 −𝜏𝑠 → time shift 𝑓(𝑡 − 𝜏)
𝑛→∞ 𝜏 𝑠 + 1
𝑛

Remember that exponential functions in the Laplace space are just time shifts in the time space.
In this case the time constant of the total system would be equal to the sum of the time constants of the
individual systems. For 𝑛 = ∞ we would again have a plot in the form of the initial pulse but with a delay,
like for a PFR.

y(t)

t/ n τ
Figure 7-7. Step response of nth order systems.
Chemical Process Control, Prof. Robert N. Grass | 35

Specifications within the Time Domain

For a damped second order system without roots the following holds:

1.8
Rise time: 𝑡𝑟 ≅
𝜔𝑛
5% 𝜉 = 0.7
𝜋𝜉
Overshoot: 𝑀𝑝 = exp( ) ≅ { 16 % 𝜉 = 0.5
ඥ1−𝜉 2
35 % 𝜉 = 0.3

4.6
Settling time: 𝑡𝑠 ≅ 𝜔
𝑛𝜉

𝜋
Peak time: 𝑡𝑝 =
𝜔𝑛 ඥ1−𝜉 2

• We can find the peak time and the overshoot if we set the derivative of the function zero and look
for the maximum
• An additional root in the left half plane (LHP) enforces the overshoot if the root is less then factor
4 distant from the real part of the poles.
• An additional root in the right half plane (RHP) inhibits overshooting and can even lead to
undershooting.
• An additional pole extends the rise time, if the additional pole is less then factor 4 distant from the
complex part of the complex poles.

Figure 7-8. Specifications within the time domain. Figure from [FrPE06]
36 | Chemical Process Control, Dr. Robert N. Grass

8 System Stability
A system is asymptotically stable, if there are no ‘internal’ state variables that go towards infinity but all
go to zero when t → ∞. When perturbed slightly the system will go back into the steady state.

Consider the following transfer function of a system:

𝑝(𝑠) 𝑝(𝑠)
𝐺(𝑠) = = 𝑛
𝑎𝑛 𝑠𝑛 + 𝑎𝑛−1 𝑠 𝑛−1 + ⋯ + 𝑎0 ∏𝑖=1(𝑠 − 𝑝𝑖 )

Assume that all roots 𝑝1 , 𝑝2 , … 𝑝𝑛 of the characteristic equation

𝑎𝑛 𝑠 𝑛 + 𝑎𝑛−1 𝑠 𝑛−1 + ⋯ + 𝑎0 = 0

are distinct (einfach). A differential equation with this characteristic equation has a solution of the form:

𝑦(𝑡) = ∑ 𝐾𝑖 𝑒 𝑝𝑖 𝑡
𝑖=1

This system is stable, if all terms in the series go to zero for t → ∞.

In other words:

If all 𝑝𝑖 < 0 for 𝑡 → ∞ 𝑦(𝑡) = 0 converging poles

stable system
If one 𝑝𝑖 > 0 for 𝑡 → ∞ 𝑦(𝑡) → ∞ diverging poles
unstable system
If one 𝑝𝑖 = 0 for 𝑡 → ∞ neither con- nor diverging poles
asymptotically stable system

If all poles are zero, these simple rules can’t be applied to predict system behavior. This is due to the fact
that the linearization introduces an error into the calculation. It is therefore highly unlikely that all roots
are truly zero and a different approach would be needed or more terms included into the linearization.

If there are complex conjugate poles 𝑝𝑖 , the solution would be given by an oscillatory response:

𝑦(𝑡) = 𝐴1 𝑒 𝑝𝑖𝑡 = 𝐴(cos(𝑡) + 𝑖 sin(𝑡))

.
Chemical Process Control, Prof. Robert N. Grass | 37

By looking at the location of the poles one can get a feeling of the system behavior and analyze the stability.
A pole to the right of the imaginary axis, with positive real part is called RHP (right-half plane pole). Such
a pole will lead to an unstable system as it results in a system response which grows without bound. The
term of a LHP (left-half plane pole) in contrast will go to zero and the system will approach the stable steady
state.

Roots on the right hand side of the figure lead to unstable systems which are not desirable for process control
applications. The more on the left, the faster the system evolves towards its steady state. When we compare
the evolution speed of two roots (with roots with purely negative real parts), we must look at the rightmost
root (the one closest to zero) as this one governs the speed. The system with the more negative right root
will approach its steady state faster.

Figure 8-1. Root Locus Impulse responses as a function of the position of the roots.
38 | Chemical Process Control, Dr. Robert N. Grass

9 Sensitivity
With a sensitivity analysis we can determine the effect of changes in 𝐴 or due to uncertainties (if 𝐴 is not
known precisely).

Assumption: Engine gain goes from 𝐴 to 𝐴 + 𝛿𝐴.

Open loop control:

U(s) Y(s)
K A

Figure 9-1. Open loop control system

𝐾𝐴 = 𝑇𝑜𝑙
𝐾(𝐴 + 𝛿𝐴) = 𝑇𝑜𝑙 + 𝛿𝑇𝑜𝑙

Where 𝑇𝑜𝑙 is the open loop transfer function.

Rearrange
𝑇𝑜𝑙 𝛿𝐴 𝛿𝑇𝑜𝑙
𝐾(𝐴 + 𝛿𝐴) = (𝐴 + 𝛿𝐴) = 𝑇𝑜𝑙 ൬1 + ൰ = 𝑇𝑜𝑙 + 𝛿𝑇𝑜𝑙 = 𝑇𝑜𝑙 ൬1 + ൰
𝐴 𝐴 𝑇𝑜𝑙
𝛿𝐴 𝛿𝑇𝑜𝑙
1+ =1+
𝐴 𝑇𝑜𝑙
Relative change:
𝛿𝑇𝑜𝑙 𝛿𝐴
=
𝑇𝑜𝑙 𝐴

Hendrik W. Bode named this ratio the Sensitivity of the transfer behavior on changes of A.

𝛿𝑇/𝑇 𝐴 𝑑𝑇
𝑆𝐴𝑇 = ≈
𝛿𝐴/𝐴 𝑇 𝑑𝐴

For open loops

𝑇
𝑆𝐴 𝑜𝑙 = 1

When a relative change in 𝐴 occurs, for an open loop, it will translate into a change in 𝑇 of equal size. Thus,
if the engine gains 5%, the velocity will also rise by 5%.

Closed loop control:

When we add a feedback loop to the system above

U(s) Y(s)
K A

Figure 9-2. Closed loop control system

Chemical Process Control, Prof. Robert N. Grass | 39

The transfer function changes and

𝐴𝐾
= 𝑇𝑐𝑙
1 + 𝐴𝐾

(𝐴 + 𝛿𝐴)𝐾
= 𝑇𝑐𝑙 + 𝛿𝑇𝑐𝑙
1 + (𝐴 + 𝛿𝐴)𝐾

Using the above formula for the sensitivity 𝑆𝐴𝑇 :

𝑇 𝐴 (1 + 𝐴𝑘)𝑘 − 𝑘(𝐴𝑘) 1
𝑆𝐴 𝑐𝑙 = =
𝐴𝑘 (1 + 𝐴𝑘)2 1 + 𝐴𝑘
1 + 𝐴𝑘
Therefore, for closed loops:
𝑇 1
𝑆𝐴 𝑐𝑙 =
1 + 𝐴𝑘

This illustrates the advantage of feedback control. Sensitivity to changes is usually smaller than 1 and closed
loop control reduces the effect of changes in the gain factor 𝐴 by a factor of 1 + 𝐴𝑘 when compared with
open loop control. By choosing 𝑘 large, sensitivity and the effect of disturbances on the system can be
reduced.

EXAMPLE 9-1: Advantage of Feedback

Consider a closed loop with the gain 𝑘 is such that

1 + 𝐴𝑘 = 100
we can calculate the sensitivity
𝑇 1
𝑆𝐴 𝑐𝑙 = = 0.01
1 + 𝐴𝑘
a 10% change in A
𝛿𝐴
= 0.1
𝐴

would then lead to a change of

𝛿𝐴 𝑇𝑐𝑙
𝛿𝑇 = ∙ 𝑆 = 0.001
𝐴 𝐴

The change in steady state gain in a closed loop would thus be 0.1%. When compared with the change
for an open loop under the same conditions, which would be 10% as the sensitivity is one, the closed
loop is much less sensitive.

One of the main reasons for the use of feedback control is not that it can respond easier to changes but the
fact that the system is not precisely known. In order to compensate for such uncertainties a low sensitivity
is desirable as it will reduce the effects of changes in the transfer function.
40 | Chemical Process Control, Dr. Robert N. Grass

10 Dynamic Behavior

P-Controller

• A static k does not change the dynamics of an open loop control system
• A static k changes the dynamics of closed loop control systems

Consider the following closed loop control system with a P controller:

R(s) + E(s) U(s) A Y(s)

- k
(τ1s +1)(τ2s +1)

Figure 3-3. Control cycle with a second order system.

This system for example could represent two stirred tanks in series with residence times 𝜏1 and 𝜏2 .

The transfer function of this system (only with a P-controller) is:

𝑌(𝑠) 𝑘𝐴
= with 𝐴>0
𝑅(𝑠) (𝜏1 𝑠 + 1)(𝜏2 𝑠 + 1) + 𝑘𝐴

The dynamics of the system can be understood by looking at the roots of the transfer function.

→ the roots of the characteristic equation as a function of the gain 𝑘 are:

(𝜏1 𝑠 + 1)(𝜏2 𝑠 + 1) + 𝐴𝑘 = 0

𝜏1 𝜏2 𝑠 2 + (𝜏1 + 𝜏2 )𝑠 + 1 + 𝐴𝑘 = 0

−(𝜏1 + 𝜏2 ) ± ඥ(𝜏1 + 𝜏2 )2 − 4𝜏1 𝜏2 (1 + 𝐴𝑘)

𝑠1,2 =
2𝜏1 𝜏2
−(𝜏1 + 𝜏2 ) ± ඥ(𝜏1 − 𝜏2 )2 − 4𝐴𝑘𝜏1 𝜏2
=
2𝜏1 𝜏2
−1
For 𝑘 = 0 (no P-controller) we get 𝑠1,2 = 𝜏
1,2

For 2nd order systems:

• 𝑘 affects the roots => affects the dynamics of the response
• first, increasing 𝑘 will shift the roots towards the left half plane and thus will result in faster
response (desired)
Chemical Process Control, Prof. Robert N. Grass | 41

• at some point the discriminant becomes zero and there will be a multiple root
• further increasing k results in a negative discriminant and complex conjugate poles and thus in an
oscillating response
• if 𝑘 is big there is a “trade-off” between low stationary error (low sensitivity, see previous
chapter) and bad dynamic behavior.
→ complex controllers with more degrees of freedom are required (see next chapter)

Plotting the behavior of the roots as a function of the parameter k gives the so called root locus:

Figure 10-2. Root locus

With increasing k the roots approach the multiple root from both
sides (on the horizontal) and then split when they reach it and
continue on the vertical line drawn.

(𝜏1 −𝜏2 )2
In the range of 0 < 𝑘 < the roots are both real and negative, the system is thus stable and non-
4𝐴𝜏1 𝜏2
oscillating.

(𝜏1 −𝜏2 )2
For 𝑘 > 4𝐴𝜏1 𝜏2
the roots are a complex conjugate pair and the system is oscillating.

From this analysis, k is chosen as large as possible without creating imaginary roots:
→ the more left the rightmost root, the faster the response (see Chapter 8)
42 | Chemical Process Control, Dr. Robert N. Grass

EXAMPLE 10.0-1: Pendulum

Remember the variables from last time:

𝑇𝑐 external force
𝑚, 𝑔, 𝑙, 𝐼 constants
𝜃, 𝑇𝑐 variables

And the describing function:

𝑓 = 𝐼𝜃ሷ = 𝑇𝑐 − 𝑚𝑔𝑙 sin 𝜃

In steady state:
0 = 𝑇𝑐 − 𝑚𝑔𝑙 sin 𝜃
If there is no external force 𝑇𝑐 = 0
0 = −𝑚𝑔𝑙 sin 𝜃
𝑠𝑠
𝜃 = 0 and 𝜃 𝑠𝑠 = 𝜋
→ two steady states exist

Linearization
𝜕𝑓 𝜕𝑓
𝑓 ≈ 𝑓 𝑠𝑠 + | Δ𝑇𝑐 + | Δ𝜃
𝜕𝑇𝑐 𝑠𝑠 𝜕𝜃 𝑠𝑠
𝐼Δ𝜃ሷ = Δ𝑇𝑐 − 𝑚𝑔𝑙 cos 𝜃 𝑠𝑠 Δ𝜃

Laplace Transformation
̅ = 𝑇̅𝑐 − 𝑚𝑔𝑙 cos 𝜃 𝑠𝑠 Θ
𝐼𝑠 2 Θ ̅
̅
Θ 1
= 2
𝑇̅𝑐 𝐼𝑠 + 𝑚𝑔𝑙 ⋅ cos 𝜃 𝑠𝑠

Stability of the roots

𝐼𝑠 2 + 𝑚𝑔𝑙 ⋅ cos(𝜃 𝑠𝑠 ) = 0
𝑚𝑔𝑙
𝑠2 = − ⋅ cos 𝜃 𝑠𝑠
𝐼

For 𝜃 𝑠𝑠 = 0 (pendulum)
𝑚𝑔𝑙 𝑚𝑔𝑙
𝑠2 = − cos 0 = −
𝐼 𝐼
𝑚𝑔𝑙
𝑠1/2 = ±𝑖√
𝐼
Imaginary roots, oscillating response (constant, stable), 𝑅𝑒(𝑠) = 0.

For 𝜃 𝑠𝑠 = 𝜋 (inverse pendulum)

𝑚𝑔𝑙 𝑚𝑔𝑙
𝑠2 = − cos 𝜋 = +
𝐼 𝐼
𝑚𝑔𝑙
𝑠1/2 = ±√
𝐼
One positive real part, unstable
Chemical Process Control, Prof. Robert N. Grass | 43

𝑚𝑔𝑙
𝑠1/2 = ±√
𝐼
One positive real part, unstable

EXAMPLE 10.0-2: Inverse Pendulum + Controller

R(s) + E U(s) = 1 Y(s)= ϴ

- k Tccccc 10s2-10

Figure 10-3. Control cycle.

𝐼 = 20 𝑚𝑔𝑙 = 10

New overall transfer function

1
𝑌 𝑘 2 − 10 𝑘
= 20𝑠 =
𝑅 1+𝑘 1 2
20𝑠 − 10 + 𝑘
20𝑠 2 − 10

New roots
20𝑠 2 − 10 + 𝑘 = 0
ξ50 − 5𝑘
𝑠1/2 = ±
10
for 𝑘 > 10 purely imaginary roots, oscillating
for 𝑘 < 10 1 positive, 1 negative root, unstable

The controller can change the dynamics of the inverse pendulum (originally unstable) to a stable (but
oscillating) response.
44 | Chemical Process Control, Dr. Robert N. Grass

11 Improved Controllers

Goal P-Controller I-Controller D-Controller

Good final value, i.e. small For large k no steady state error fast (reacts on slope
steady state error (response increases with of error, i.e. before
time when error error occurred)
persistent)
Influence dynamics slow (least negative root sensitive to sensor
a) Faster a) For large k larger than for P- noise
b) Less oscillating b) for small k control)
influence stability for large k to for large 𝑇𝐼 unstable poor influence on
some extent stability

Integral-Controller

The integral controller is also called I-controller

r(t) e(t) u(t)

+ I-controller
-
C(s)

Figure 11-1. Part of a control cycle with an I-controller

The “transfer function” of this part is given by:

𝑘 𝑡
𝑢(𝑡) = ∫ 𝑒(𝜏)𝑑𝜏
𝑇𝐼 0
Laplace transformation then gives:
𝑘
𝑈(𝑠) = 𝐸(𝑠)
𝑇𝐼 𝑠
with:
𝑘
𝐶(𝑠) =
𝑇𝐼 𝑠
1
Where 𝑇𝐼 is the integral or reset time (Nachstellzeit) and 𝑇 is the reset rate.
𝐼
𝑢(𝑡) changes until ∫ 𝑒 = 0 and then 𝑢(𝑡) is constant and has reached its steady state. Therefore, for a
controller with I portion the controller error will always go to zero. The steady state is reached when the
error is zero.
Chemical Process Control, Prof. Robert N. Grass | 45

Consider the following 2nd order system with an I-controller:

R(s) + E k U(s) = A Y(s)= ϴ

- TI s Tccccc (τ1s + 1)( τ2s + 1)

Figure 11-2. Control cycle of a second order system with integral control.

The transfer function of such a system is:

𝑘
𝑌 𝑇𝐼 𝐴
=
𝑅 𝜏 𝜏 𝑠 3 + (𝜏 + 𝜏 )𝑠 2 + 𝑠 + 𝐴 𝑘
1 2 1 2 𝑇 𝐼

In comparison to the transfer function with the P-controller the characteristic equation has changed. The
order has increased by 1 and also the roots and the root locus have changed.

Figure 11-3. Root locus of a second order system with integral control.

There is an additional root for the controller at zero for k= 0. As 𝑘 increases the roots move left in the plane
(except for the middle one which will move right) and eventually a complex conjugate pair forms (at some
point even with positive real parts). Thus, for increasing 𝑘 the damping will be noticeably weaker until the
system will eventually become unstable.

The advantage of the I-controller is that it has no steady state error but its response is slow. As we have seen
it also tends to oscillate and get unstable with increasing 𝑘.

As the error for this kind of controller will always go to zero, we only have to look at the dynamics. The I-
controller is not usually used on its own but coupled with a P-controller.
46 | Chemical Process Control, Dr. Robert N. Grass

Differential Controller

r(t) e(t) u(t)

+ D-controller
-
C(s) = k TD s

Figure 11-4. Part of a control cycle with differential control.

𝑑(𝑡) = 𝑘𝑇𝐷 𝑒ሶ (𝑡)

𝐷(𝑠) = 𝑘𝑇𝐷 𝑠𝐸(𝑠)
𝐶(𝑠) = 𝑘𝑇𝐷 𝑠
Where 𝑇𝐷 : Derivative time (Vorhaltezeit)

Differential controllers respond to the difference in slope. For steep slopes the response will be much
stronger than for lower slopes, therefore they have a fast response.
As the response is very fast, the controller is sensitive to noise and is usually implemented in the feedback
loop.

Figure 11-5. Comparison of different controller responses with e(t) being the assumed error.

Comparison of Controller Responses

→ P-controllers multiply the error by a constant (gain k) and result in a steady state error.
→ I-controllers are slow but adjust the steady state error.
→ D-controllers are fast but sensitive to background noise and tend to oscillate.

As all the controllers have different advantages and disadvantages is common to combine them into a PID-
controller with three elements. The Proportional term closes the feedback look, the Integral term assures
that there is no error remaining and the Derivative term improves the dynamic response and stability.
Chemical Process Control, Prof. Robert N. Grass | 47

PID-Controller

Usually, a combination of P-, I- and D-control is used (industrial standard).

1 𝑡
𝑢(𝑡) = 𝑘 (𝑒(𝑡) + ∫ 𝑒(𝜏)𝑑𝜏 + 𝑇𝐷 𝑒ሶ (𝑡))
𝑇𝐼 0
1
𝑈(𝑠) = 𝑘 ∙ 𝐸(𝑠) ൬1 + + 𝑇𝐷 𝑠൰
𝑇𝐼 𝑠

𝑇𝐼 , 𝑇𝐷 , 𝑘 are selected to obtain an optimal response.

For a second order system all roots can be independently chosen with those three parameters. The three
parameters have to be determined to get a good response. This is the aim of process control.
Sometimes PD or PI-controllers are used when the desired effect can be achieved with setting only two
parameters.

Figure 11-6. Transient behavior of different controllers after (a) a unit disturbance and (b) a reference step. Figure from [FrPE06]

In the left plot we can immediately recognize the controllers with an integral part as their steady state error
goes to zero. When only the P-controller is used, y does not go to zero.
Comparing the PID and PI-controller we can say that the better response (not oscillating) must belong to
the controller with more degrees of freedom when it is set efficiently.
On the right we can again spot the (small) steady state error of the P-controller and the smaller overshoot
for the PID-controller when comparing it to the PI version.
48 | Chemical Process Control, Dr. Robert N. Grass

Design of PID Controllers

+
- PID System

Figure 11-7. Control cycle of a system with PID control.

The main aim of control unit design is the stability. PID controllers are often used because their stability is
well understood.

Experimental Methods

The methods below are used when models for the system and more advanced mathematical tools are not
available. They were both developed by Ziegler and Nichols, two chemical engineers who were looking
primarily at chemical systems.

Ziegler + Nichols (1942)

Most real systems have a “similar“ response for a step function which is characterized by reaction rate 𝑅
and lag 𝐿. The method aims at designing controllers for a system based on its step response and is also called
step method.

Figure 11-8. Process response. Figure from [FrPE06].

The response will have a shape as above and can be approximated by a step response with the transfer
function
𝑌(𝑠) 𝐴 exp(−𝑠𝑡𝑑 )
=
𝑈(𝑠) 𝑠
A step function is introduced into a process (open loop) without a controller
 R and L can be determined from the response
 the goal is to tune the control unit in such a way that the amplitude in each oscillation should decrease
by 3/4 compared to the previous oscillation (see Figure 11-9)  decay ratio of 0.25
Chemical Process Control, Prof. Robert N. Grass | 49

Figure 11-9. Aim of the Ziegler and Nichols

method. Figure from [FrPE06].

 Ziegler and Nichols developed rules to tune P, PI and PID controllers in order to achieve the desired
response (see table below) and the parameters 𝑘, 𝑇𝑖 and 𝑇𝐷 can calculated from 𝑅 and 𝐿.

 The method is fast and reliable but only applicable when measurements are possible and when the step
response has a shape as in Figure 11-8. It is suitable for slow processes when we aim mostly at correcting
disturbances.

𝑘 𝑇𝑖 𝑇𝐷

1
P / /
𝑅𝐿

0.9 𝐿
PI /
𝑅𝐿 0.3

1.2
PID 2𝐿 0.5 𝐿
𝑅𝐿
50 | Chemical Process Control, Dr. Robert N. Grass

Ultimate Gain method

Ziegler and Nichols have developed a second method based on the experience that 𝑘 shall not exceed a
certain value 𝑘𝑢 to prevent instability.
In the experiment a closed loop system with only P control is used and 𝑘 is increased until the system
becomes marginally stable. The oscillations are stable and constant.
 this 𝑘 is the ultimate gain 𝑘𝑢
 the period at 𝑘𝑢 is the ultimate period 𝑃𝑢 of the oscillation at 𝑘𝑢
 the period and gain are measured at the point of marginal stability
 based on these values the parameters for the controllers can be determined from the table below

Figure 11-9. Ultimate gain method where Pu is the ultimate period.

Figure from [FrPE06].

𝑘 𝑇𝑖 𝑇𝐷

P 0.5 𝑘𝑢 / /

1
PI 0.45 𝑘𝑢 𝑃 /
1.2 𝑢

1 1
PID 0.6 𝑘𝑢 𝑃 𝑃
2 𝑢 8 𝑢
Chemical Process Control, Prof. Robert N. Grass | 51

Mathematical Method

Root-Locus
Disadvantage: necessary to compute roots of characteristic equation; has multiple solution and is difficult
to solve for higher order systems

2nd order binomial equation

3rd order Cardano equation
nth order no algebraic solution (n>4)

Alternative: Routh stability criterion (before computers were invented)

Characteristic equation
𝑎𝑛 𝑠 𝑛 + 𝑎𝑛−1 𝑠 𝑛−1 + 𝑎𝑛−2 𝑠 𝑛−2 + ⋯ + 𝑎0 = 0
Necessary condition for stability: 𝑎𝑖 > 0

Routh table
52 | Chemical Process Control, Dr. Robert N. Grass

12 Frequency Response
This method is very useful to get a first impression of the system behavior and is often used in practice. The
frequency response will allow us not only to predict the response for any sinusoidal input but will give
qualitative information which is useful in determining the stability and robustness of a system.

Input = y(t) = ?
System
𝐴 sin(𝜔𝑡)

Figure 12-1. System with frequency input.

What happens at the output for 𝑡 → ∞ (in the stationary case)?

 For a linear system we will have oscillation with the same frequency 𝜔 but a different amplitude 𝐴 and
phase 𝜙

Input
𝑢(𝑡) = 𝐴 sin(𝜔𝑡)
Laplace transformation
𝐴𝜔
𝑈(𝑠) =
𝑠2 + 𝜔2

𝐴𝜔
𝑌(𝑠) = 𝑈(𝑠)𝐺(𝑠) = 𝐺(𝑠)
𝑠2
+ 𝜔2

1
𝐺(𝑠) =
(𝑠 − 𝑠1 )(𝑠 − 𝑠2 ) …

Partial fraction expansion

𝛼1 𝛼2 𝛽 𝑖𝛽
𝑌(𝑠) = + + ⋯+ +
𝑠 − 𝑠1 𝑠 − 𝑠2 𝑠 + 𝑖 𝜔0 𝑠 − 𝑖 𝜔0
Inverse LT
𝑦(𝑡) = 𝛼1 𝑒 + 𝑠1 𝑡 + 𝛼2 𝑒 + 𝑠2 𝑡 + ⋯ + 2|𝛽| cos(𝜔𝑡 + 𝜙)

For 𝑡 → ∞ and 𝑝𝑖 < 0

𝑦(𝑡 = ∞) = 2 |𝛽| cos(𝜔𝑡 + 𝜙)

 System oscillates with the same frequency, new amplitude 𝑀 = 2 |𝛽| and phase shift 𝜙.

To find the values of 𝑀 and 𝜙 we find 𝑦(𝑡) for 𝑢(𝑡) = 𝐴 sin 𝜔𝑡 by means of convolution

𝑦(𝑡) = 𝑢(𝑡) ∗ 𝑔(𝑡)

1
And we apply the relation for exponential and trigonometric functions sin 𝑥 = 2𝑖 (𝑒 𝑖𝑥 − 𝑒 −𝑖𝑥 )
Chemical Process Control, Prof. Robert N. Grass | 53

∞
𝐴
𝑦(𝑡) = ∫ 𝑔(𝜏) (𝑒 𝑖𝜔(𝑡−𝜏) − 𝑒 −𝑖𝜔(𝑡−𝜏) ) 𝑑𝜏
−∞ 2𝑖

This integral is not trivial to solve. We must find simplification for the convolution of 𝑣(𝑡) = 𝑒 𝑠𝑡

∞ ∞
𝑣(𝑡) ∗ 𝑔(𝑡) = ∫ 𝑔(𝜏) 𝑒 𝑠(𝑡−𝜏) 𝑑𝜏 = ∫ 𝑔(𝜏) 𝑒 −𝑠𝜏 𝑒 𝑠𝑡 𝑑𝜏
−∞ −∞

∞
= 𝑒 𝑠𝑡 ∫ 𝑔(𝜏) 𝑒 −𝑠𝜏 𝑑𝜏 = 𝐺(𝑠) 𝑒 𝑠𝑡
−∞

Definition of the Laplace transform →𝐺(𝑠)

We could show that 𝑣(𝑡) = 𝑒 𝑠𝑡 is the response in the time domain for the Laplace transform of the function.
This can be applied to 𝑠 = 𝑖𝜔 which leads to

𝐴
𝑦(𝑡) = (𝐺(𝑖𝜔) 𝑒 𝑖𝜔𝑡 − 𝐺(−𝑖𝜔) 𝑒 −𝑖𝜔𝑡 )
2𝑖

Transform to radial coordinates 𝐺(𝑖𝜔) = 𝑀𝑒 𝑖𝜑

𝐴
𝑦(𝑡) = 𝑀(𝑒 𝑖(𝜔𝑡+𝜑) − 𝑒 −𝑖(𝜔𝑡+𝜑) )
2𝑖

And apply again the definition of the sine and compare with the expression found for 𝑦(𝑡 = ∞)

𝑦(𝑡) = 𝐴𝑀 sin(𝜔𝑡 + 𝜑)
𝐴𝑀 = 2|𝛽| 𝜑=𝜙

Now we can find 𝑀 and 𝜙.

Amplitude ratio 𝐴𝑅
𝐴𝑀
𝐴𝑅 = = 𝑀 = |𝐺(𝑖𝜔)| = |𝐺(𝑠)|𝑖𝜔 |
𝐴

Phase shift 𝜙
𝐼𝑚(𝐺(𝑖𝜔))
𝜙 = tan−1 ቆ ቇ
𝑅𝑒(𝐺(𝑖𝜔))

With the amplitude ratio and phase shift we can find the outlet for any sine or cosine by plugging in 𝑖𝜔
instead of 𝑠.
54 | Chemical Process Control, Dr. Robert N. Grass

EXAMPLE 12-1: Frequency response

For a system with the transfer function

1
𝐺(𝑠) =
𝑠+𝑘
We plug in 𝑖𝜔
1
𝐺(𝑖𝜔) =
𝑖𝜔 + 𝑘
And we can calculate the amplitude ratio
1
𝐴𝑅 =
ξ𝜔 2 + 𝑘 2
As well as the phase shift at the outlet
𝜔
𝜙 = tan−1 (− )
𝑘

Bode Diagram

For a 1st order system the transfer function is given by:

𝑘
𝐺 = 𝑇𝐹 =
𝜏𝑠 + 1

The input of the system is again a wave function as in Fig 12-1. The oscillation of the output will have the
same frequency as we have seen, however amplitude and phase change. When the input frequency is low,
the corresponds to slow changes, and we are likely to still have the same amplitude at the outlet (only
amplified by 𝑘). As the frequency increases, the changes occur faster and we approach the time constant of
the system. The behavior changes to a lower response amplitude and phase shift (see figure 12-2).

Figure 12-2. Bode diagram for a 1st order system.

Bode plots depict the amplitude ratio and phase shift as a function of the input frequency. Both 𝐴𝑅 and 𝜔
are on a log scale (𝐴𝑅 most often in dB), 𝜙 is drawn in degrees.

The plots can be determined experimentally or constructed by hand (see example 12.1-1) in a quick and yet
sufficiently accurate way. Bode plots of more complex transfer functions can be constructed in an additive
way, where the transfer function is split up into simpler components. The plots for these parts can be drawn
easily and are added up to give the Bode plot of the more complex system.
Chemical Process Control, Prof. Robert N. Grass | 55

Simple Graphical Bode Plot Design

Computing of |𝐺(𝑖𝜔)| is not necessary as certain classes of functions behave the same. Below the plotting
of these classes will be discussed.

Standard classes of TF:

a) K𝑠 𝑛 = K(iω)n
→ log(𝐾|𝑖𝜔|)𝑛 = log(𝐾) + 𝑛 log(|𝑖𝜔|)
→ 𝐴𝑅: line with slope 𝑛 through 𝐾 at 𝜔 = 1
→ 𝜙 = 𝑛 ⋅ 90°

0.1 1 10 0.1 1 10
Figure 12-3. Simple Bode diagram general cases of 𝐊𝒔𝒏 .

b) (𝜏𝑠 + 1)𝑛 = (𝜏(𝑖𝜔) + 1)𝑛

𝐴𝑅: constant until 𝜔 = 1/𝜏 (break point), then slope 𝑛
𝜙 = 0 until 𝜔 = 1/𝜏, then 𝜙 = 𝑛 ⋅ 90°

Figure 12-4. Simple Bode diagram general cases of (𝝉𝒔 + 𝟏)𝒏 .

The plots above are only asymptotically correct and would have rounded edges and smooth transitions.
56 | Chemical Process Control, Dr. Robert N. Grass

EXAMPLE 12.1-1: Drawing Bode Diagrams

The following transfer function is given:

2000 (𝑠 + 0.5)
𝑇𝐹 =
𝑠 (𝑠 + 10)(𝑠 + 50)

Rearranging to have 𝜏𝑠 + 1 in the denominator:

2(2 𝑠 + 1)
𝑇𝐹 =
1 1
𝑠 (10 𝑠 + 1) ( 𝑠 + 1)
50

We can identify four components, one of standard class a) (of form 𝐾𝑠 𝑛 ) and three of standard class b)
(of form (𝜏𝑠 + 1)𝑛 ) with time constants:
𝜏1 = 2
1
𝜏2 =
10
1
𝜏3 =
50
Let us first look at the magnitude plot. The term of class a) is first order and in the denominator will lead
to a slope of 𝑛𝑎 = −1 throughout the entire 𝜔 range. The terms of class b) will identify the behavior
only after their break point 𝜔𝑖 which we can find by the inverse of the time constant. The effect on the
slope will be positive or negative depending on whether the term is in the nominator or denominator.

𝜔1 = 0.5 𝑛1 = +1
𝜔2 = 10 𝑛2 = −1
𝜔3 = 50 𝑛3 = −1

As the slopes behave additive, we can construct the magnitude plot piecewise. Initially the slope will be
−1 until after 𝜔 = 0.5 it will be horizontal as the 𝑛1 = +1 effect of the 2𝑠 + 1 in the nominator cancels
1
the −1 slope. When 𝜔 reaches a value of 10 the effect of the 10 𝑠 + 1 term in the denominator will come
in and decrease the slope to −1 and the last b) type term further to −2 after 𝜔 = 50.

To draw the plot we should start with the first asymptote before the effect of the class b) terms. This
asymptote has a slope of -1 (or 20 dB per decade) and we can locate it as it would pass through the value
2 (the value of 𝑘 in this case) at 𝜔 = 1. The asymptotes can then be constructed by changing the slope
at the break points and the amplitude plot is then obtained by connection of the lines.
Chemical Process Control, Prof. Robert N. Grass | 57

For the phase plot the same break points play a role for type b) terms and a) terms will again have an
influence over the entire 𝜔 range. The effect can also be read from the value of 𝑛𝑖 and will be 𝑛𝑖 ∙ 90°.
Phase shifts also show additive behavior. We therefore start at -90°, go to 0°, to -90° and finally -180°.
The obtained staircase like function will be only approximate with the actual shape being much
smoother.

This procedure yields the final Bode diagram in figure 12-5 and 12-6.

Figure 12-5. Magnitude plot for example 12.1-1. Figure from [FrPE06].

Figure 12-6. Phase plot for example 12.1-1. Figure from [FrPE06].
58 | Chemical Process Control, Dr. Robert N. Grass

Bode Diagram of PID Controllers

Figure 12-7. Bode diagrams for P, PD, PI and PID controllers with different C(s).
Chemical Process Control, Prof. Robert N. Grass | 59

Bode Plots of 2nd Order Systems

𝑘
𝐺(𝑠) =
𝜏2𝑠2 + 2 𝜉 𝜏𝑠 + 1
With damping factor 0 < 𝜉 < 1
→ oscillating response to step input
→ maximum amplitude (if it exists) at the natural frequency 𝜔𝑛 is the greater, the smaller the damping
factor 𝜉
→ medium phase of -90° at the natural frequency 𝜔𝑛 and the steeper the smaller the damping factor 𝜉

Figure 12-8. Bode diagrams for oscillating responses for different values of ξ. Figure from [FrPE06].

Bode Stability Criterion

The Bode stability criterion is a simple method to analyze the stability of feedback control systems. Instead
of the mathematically more complicated closed loop, the stability of the open loop is analyzed.

R(s) + G(s) Y(s)

- C(s)

Figure 12-9. Control cycle with an open loop (green) and a closed loop (green and black).

If the open loop is stable, the closed loop will also be stable if at the critical frequency 𝜔𝑐 the condition
𝐴𝑅(𝜔𝑐 ) < 1 holds, where 𝜔𝑐 is the frequency when 𝜙 = −180°.

The criterion can be applied only to overall decreasing Bode plots and when the open loop is stable.
60 | Chemical Process Control, Dr. Robert N. Grass

Figure 12-10. Bode diagram with phase shifts up to -180°

Why is a phase shift of -180° critical?

𝑈(𝑠) = + 𝑌(𝑠) =
- G(s) C(s)
sin(𝜔𝑡) sin(𝜔𝑡 − 180)

sin(𝜔𝑡 − 180)
= − sin(𝜔𝑡)

Figure 12-11. Control cycle with phase shift of – 180°

→ For a phase shift of -180° and the same amplitude the output will be exactly the opposite of the input
→ As it is subtracted this is the same as adding again the input signal
→ So after the back coupling there will be constructive interference! (If the amplitude ratio is greater than
unity is will amplify, otherwise it disappears)

The Bode stability criterion can also be used for controller design. When we introduce another controller
into the system or change the parameters of the implemented controllers, the Bode plot will change. The
new controller/parameters should change it in such a way, that the stability criterion is fulfilled and we can
find it by construction. This procedure is illustrated in the following example.
Chemical Process Control, Prof. Robert N. Grass | 61

EXAMPLE 12.3-1: Controller Design using Bode Plots

The following transfer function is examined in this exercise

1
𝐺(𝑠) =
(0.2𝑠 + 1)(2𝑠 + 1)(𝑠 + 1)2

The Bode diagrams for the system with a P and a PI-controller are drawn below. When only a P-
controller with 𝑘 = 10 is introduced, the function becomes 𝐺(𝑠) ∙ 𝑘 and according to the Bode stability
criterion will be slightly unstable.

With a PI-controller with 𝜏𝐼 = 1 we can achieve stability when the criterion is considered.

Figure 12-12. Bode diagram of the system with a P- and a PI-controller. The
critical frequencies for both set ups are indicated in red.
62 | Chemical Process Control, Dr. Robert N. Grass

Stability Margin
The Bode criterion can be extended to analyse the stability also in a quantitative way and to determine not
only whether the system is stable or not but how stable. This described by the gain and phase margin.

As a chemical process is rarely unchanged as the process conditions may vary or it might be disturbed, this
is very useful. The margins indicate how likely such changes will result in instability.

Gain margin
at 𝜙 = −180°, how far away is 𝐴𝑅 of 1
Phase margin
at 𝐴𝑅 = 1, how far away is 𝜑 of 180°

Bode stability is only valid for a system with stable open loops
 Nyquist stability criterion for closed loops with unstable open loops and complex systems

Nyquist Diagram

Nyquist Stability Criterion

The imaginary and real for all ω between -∞ and ∞ are plotted against each other. The amplitude ratio and
phase shift can be read from the plot but not the corresponding ω value of the frequency.

𝐴𝑅 = |𝐺(𝑖𝜔)| = distance from origin

Im(G(s))

𝐼𝑚(𝐺(𝑖𝜔))
𝜑 = tan−1 ൬𝑅𝑒(𝐺(𝑖𝜔)) ൰ = angle
AR

ɸ
Re(G(s))
Nyquist point: (−1,0)

Figure 12-12. Nyquist diagram of G(s). The red dot is called

Nyquist point.

Criterion: If G(s) in the open loop is unstable, the closed loop is stable if the Nyquist plot encircles (−1,0)
in counter-clockwise direction once per unstable root.

Remember: Positive angles are left-turning.

Chemical Process Control, Prof. Robert N. Grass | 63

EXAMPLE 12.2-1: Inverse Pendulum

Figure 12.10. Problem description.

𝑔 𝑇𝑐
𝜃ሷ + 𝛼𝜃ሶ + sin 𝜃 = 2
𝑙 𝑚𝑙
with 𝛼 as a damping constant

Steady states for no external force 𝑇𝑐 = 0

𝑔
0+0+ sin 𝜃 = 0
𝑙
steady states for sin 𝜃 = 0
𝜃=0
𝜃=𝜋

Linearized system
𝑇𝑐 𝑔
𝑓 = 𝜃ሷ + 𝛼𝜃ሶ =2
− sin 𝜃
𝑚𝑙 𝑙
d𝑓 d𝑓
𝑓≈ | Δ𝑇 + | Δ𝜃
d𝑇𝑐 𝑠𝑠 𝑐 d𝜃 𝑠𝑠

1 𝑔
𝑓 ≈ 2
Δ𝑇𝑐 + (− cos 𝜃 𝑠𝑠 ) Δ𝜃
𝑚𝑙 𝑙

with cos 𝜋 = −1
Δ𝑇𝑐 𝑔
𝑓 = Δ𝜃ሷ + 𝛼Δ𝜃ሶ = + Δ𝜃
𝑚𝑙 2 𝑙
Laplace transform
𝑇̅𝑐 𝑔
̅ + 𝛼𝑠Θ
𝑠2Θ ̅= + Θ̅
𝑚𝑙 2 𝑙
̅
Θ 1
= 2 2 = 𝐺(𝑠)
𝑇̅𝑐 𝑚𝑙 𝑠 + 𝑚𝑙 2 𝛼𝑠 − 𝑚𝑙𝑔
𝑚
With 𝑚 = 1 𝑘𝑔, 𝑙 = 1 𝑚, 𝛼 = 0.01 and 𝑔 = 10 2
𝑠

Roots
𝑠 2 + 0.01𝑠 − 10 = 0
𝑠1/2 = ±3.15
64 | Chemical Process Control, Dr. Robert N. Grass

Root locus

It is not possible to tell from the open loop if the closed loop is stable or not.

1
Nyquist plot of 𝑠2 +0.01𝑠−10
1
𝐺(𝑖𝜔) =
−𝜔 2 + 0.01𝑖𝜔 − 10

for 𝜔 ∈ ℝ

Plotting 𝑅𝑒(𝐺(𝑖𝜔)) versus 𝐼𝑚(𝐺(𝑖𝜔)) gives the Nyquist plot:

Stability criterion: Encircle (−1, 0) counterclockwise once per unstable root of the open loop to
obtain stability for closed loop.

This criterion can be fulfilled by adding a controller to move/scale the behavior of the system.

Nyquist plot for P-control

For 𝐷(𝑠) = 𝑘 and 𝑘 > 10  stability

Chemical Process Control, Prof. Robert N. Grass | 65

1
Bode plot of 𝑠2 +0.01𝑠−10

Time Delay
In many chemical systems the problem of time delay occurs and affects the controller.

𝑘
𝑇𝐹 =
𝜏𝑠 + 1

𝑇𝐹 with time delay 𝜆

𝑘
𝑇𝐹𝑇𝐷 = 𝑒 (−𝑠𝜆)
Figure 12-13. Problem description. 𝜏𝑠 + 1

To illustrate the effect of time delay, we consider the following system:

R + TFTD
Y
- D

Figure 12-12. Control Cycle of a system with time delay.

66 | Chemical Process Control, Dr. Robert N. Grass

With the transfer function

𝑘
𝑌 𝐷 𝜏𝑠 + 1 𝑒 (−𝑠𝜆) 𝐷𝑘𝑒 (−𝑠𝜆)
= =
𝑅 1 + 𝐷 𝑘 𝑒 (−𝑠𝜆) 𝜏𝑠 + 1 + 𝐷𝑘𝑒 (−𝑠𝜆)
𝜏𝑠 + 1

Time delay makes it hard to use root locus method to evaluate stability. We can still apply the criterion for
Bode plots.

Bode Plot with Time Delay

Without time delay With time delay

𝑘 𝑘
𝑇𝐹 = 𝑇𝐹𝑇𝐷 = 𝑒 (−𝑠𝜆)
𝜏𝑠 + 1 𝜏𝑠 + 1

Figure 12-13. Bode diagrams of systems with time delay.

The Bode plot can still be used to assess stability of a system with time delay (stability criterion: 𝐴𝑅 <
1 for 𝜑 = −180°).

When time delay occurs the AR will be unaffected but the phase shift will be unbounded and decrease by
−𝜔𝜆 and the point of 𝜑 = −180° is approached sooner. This is detrimental to the stability of closed loop
systems.
Chemical Process Control, Prof. Robert N. Grass | 67

EXAMPLE 12.3-1: Time Delay and Instability

For a system with the transfer function

10
𝐺(𝑠) = 𝑒 −0.5𝑠
(0.2𝑠 + 1)(2𝑠 + 1)(𝑠 + 1)

With a time constants in minutes and time delay of 0.5 minutes (in the exponential term).

The Bode plots of the system with (left) and without time delay (on the right) are drawn below.

The amplitude ratios of the two systems are the same. However, the system with time delay approaches
the critical phase shift of -180° much sooner. The critical frequency is indicated with a red line.

One can see that the system without time delay is stable, as the amplitude ratio is below 1 at the critical
frequency. The phase shift of the delayed system decreases much faster and reaches the critical phase
shift earlier, when the amplitude ratio is still above 1, indicating that the system will be unstable.

Figure 12-14. Left: Bode Plot of system with time delay of 0.5 min
Right: Bode Plot of system without time delay
68 | Chemical Process Control, Dr. Robert N. Grass

13 Problems with Current Control Loops

So far:
R + Y
- D(s) G(s)

Figure 13-1. Control cycle.

find 𝐷(𝑠), i.e. PID

 experimental
 root locus
 Bode/Nyquist stability

Problems:
 only for single input, single output
 slow: first error, then reaction

Feed Forward Control

Df(s)
Gw(s)

+ + G(s) +
- D(s)
+ +

Figure 13-2. Control cycle of a system with feed forward control.

𝐷𝐺 𝐷𝑓 𝐺 + 𝐺𝑤
𝑌= 𝑅+ 𝑊
1 + 𝐷𝐺 1 + 𝐷𝐺

Idea: 𝐷𝑓 predicts the disturbance and reacts to it before it has altered the system response (significantly)
Goal: 𝑌 independent of disturbance 𝑊
Remove 2nd term
𝐺𝑤
Choose 𝐷𝑓 = − 𝐺
such that 𝑊 has no effect on system
Prerequisites
- 𝑊 measurable
- 𝐺 and 𝐺𝑤 have to be stable
- 𝐷𝑓 has to be stable
Chemical Process Control, Prof. Robert N. Grass | 69

Cascade Control

Figure 13-3. Control cycle of a system with cascade control.

In cascade control an inner closed loop 𝐺2 is added to the system which compensates for disturbances
within the inner system. This system has to be faster than the outer system 𝐺1 in order to be effective.

Cascade control is used for example in heating to correct for disturbances in the valve before their effect
becomes measurable by a change in room temperature.

EXAMPLE 13.1-1: pH Control

Figure 13-4. Problem description.

Inner loop: flow control Outer loop: pH control

Multivariable Control

Figure 13-5. Multivariable system (MIMO)

When there are multiple input variables which influence more than one output variable in a system, we call
this system multivariable system or MIMO (multiple input, multiple output).
In some cases 𝑢1 might have no or negligible effect on 𝑦2 and 𝑢2 might not influence 𝑦1 significantly. Then
the system can be treated as two single variable systems as there is no coupling.
70 | Chemical Process Control, Dr. Robert N. Grass

When the coupling is stronger, decoupling compensators are used to make up for the effects. The transfer
functions in front of the process are chosen in such a way that they cancel out the coupling and again the
system can be treated as two single variable systems. This is achieved with a control system as drawn in the
figure below.

Figure 13-6. Decoupling compensator.

The strength of the coupling is quantified through the relative gain measured along the loop represented in
𝑦
green in the setup below. The effect of 𝑔𝑐1 on 𝑢2 is determined and the relative gain 𝑚 is defined such that
2

𝑦2
|𝑔 = 0
𝑢2 𝑐1
𝑚=𝑦
2 |𝑔
𝑢2 𝑐1 ≠ 0

and will have values between 0 (no coupling) and ∞ (very strong coupling). It is often normalized with the
maximum gain so that 0 < 𝑚 < 1.

Figure 13-7. Relative gain array.

Chemical Process Control, Prof. Robert N. Grass | 71

Model Predictive Control

Control systems today are mostly based on model predictive control.

𝑢1 𝑦1
𝑢2 System 𝑦2
𝑢3 𝑦3

𝑢1′ 𝑦1′
𝑢2′ Model 𝑦2′
𝑢3′ 𝑦3′

Figure 13-8. Model Predictive Control.

The system is limited by the precision of the model that predicts the system behavior. It is based on
differential equations that must be solved fast enough.

Procedure:
 Measure current values
 Solve differential equations of the model to get path which reaches target
 Implement only the first step
 Repeat measurements and calculations

In order to compensate for the fact that the model is not able to represent the system precisely, the whole
path to reach the target is simulated but only the first step is implemented. This compensates for uncertainties
but a reasonable good model is still necessary as well as large computing power.

Figure 13-9. Model Predictive Control over time, here the target is the reference trajectory.
Figure from http://new.abb.com/control-systems/features/model-predictive-control-mpc, 8.6.16