You are on page 1of 5

Unit 3

Linearization

3.1 Learning Objective


The goal of this unit is

• that you get to know why linearization is necessary


• that you learn how to proceed to linearize a system description
• that you are able to determine an appropriate operation point

Please answer the questions in ILEARN after having read this unit!

3.2 Contents
3.2.1 A Nonlinear Example System
Let’s consider a three-tank system as given in figure 3.1.

Figure 3.1: Three-tank system

17
18

The tanks with footprint A are connected via tubes with foot-
print a. Velocity of the liquid in the tubes depends on the level
difference of the two connected tanks according to the law of
Torricelli:
p
vliquid = sgn(∆h) · 2g · |∆h|

with g . . . acceleration of earth.


If we define h1 to be the level in the left tank, h2 of the middle tank, and h3 of the right
tank and we regard that h1 > h2 > h3 , we obtain the following nonlinear differential
equations:
ap 1
ḣ1 = − 2g(h1 − h2 ) + q1
A A
ap ap
ḣ2 = 2g(h1 − h2 ) − 2g(h2 − h3 )
A A
ap ap 1
ḣ3 = 2g(h2 − h3 ) − 2g(h3 − H) + q2
A A A
Due to the non-linearities we cannot set-up a system description in matrix notation. In
order to get a deep insight into the system behavior it is best to have such a type of
description. That’s why we need to linearize the equations above. We proceed for that in
two steps:
1. We determine first a stationary condition where the system rests before we apply
any change in input or any desturbance, or after we did so. We call this stationary
condition operation point.

2. After that we regard only small deviations from this operation point and will such
be able to describe the system with linear equations.

3.2.2 Operation Point


The dynamic system is given with
ẋ = f (x, u, t) resp. ẋi = fi (x1 , . . . , xn , u, t) where i = 1, . . . , n
The steady state is defined such:
All xi are constant → ẋi ≡ 0 counts for all i = 1, . . . , n.
≡ 0 means the equations are valid in the complete t-interval, not only in one point.
This is only possible if all ui are constant: ui = ui,s , i = 1, . . . , p. Subscript s stands for
the steady state condition. With the state equations
0 = fi (x1 , . . . , xn , u1,s , . . . , up,s )
we get a system of n equations for the stationary values x1,s , . . . , xn,s :
 
x1,s
xs =  ... 
 
: rest position in state-space
xn,s
DIT Deggendorf, Prof. Müller Teaching Units Selected Chapters of Control 19

The operating point is a special steady state where certain system variables take their
preselected constant values. They are, as mentioned above, the values just before or long
after a change has happened.

As an example we look at the three-tank system: In the steady state it is


ap 1 p
0 = − 2g(h1s − h2s ) + q1s → q1s = a 2g(h1s − h2s )
A A
ap ap
0 = 2g(h1s − h2s ) − 2g(h2s − h3s ) → h1s − h2s = h2s − h3s
A A
ap ap 1
0 = 2g(h2s − h3s ) − 2g(h3s − H) + q2s →
A p A p A
q2s = −a 2g(h2s − h3s ) + a 2g(h3s − H)
From the second equation we see for instance that the level in the middle tank must be
the mean between the two neighbored levels:
h1s + h3s
h2s = (3.1)
2
Now, if two of the three levels are known (and the parameters like a, A, and H) we can
compute all other steady state variables from algebraic equations only. All differential
equations vanished here! But it’s important here to mark the computed variables as
steady state values. That’s why we it is mandatory to give these variables an extended
subscript. Don’t forget!

3.2.3 Linearization Around a Steady State


Starting point is
ẋi = fi (x1 , . . . , xn , u1 , . . . , up ) where i = 1, . . . , n
Now, we regard only small deviations of the steady state
∆ẋi = ẋi − 0, ∆xi = xi − xi,s , ∆uj = uj − uj,s

∆ẋi = fi (x1,s + ∆x1 , . . . , xn,s + ∆xn , u1,s + ∆u1 , . . . , up,s + ∆up )

We then apply Taylor’s law

n p
X ∂fi X ∂fi
∆ẋi = fi (x1,s , . . . , xn,s , u1,s , . . . , up,s ) + ·∆xk + ·∆uj
| {z } ∂x k s
k=1 | {z }
∂u j
j=1 | {z }s
= 0 steady state aik bij
+ρi (∆x1 , . . . , ∆xn , ∆u1 , . . . , ∆up )

ρ contains only terms of second order which can be neglected in a sufficient small envi-
ronment of the steady state. This leads to a linearized system description:
n p
X X
∆ẋi = aik ∆xk + bij ∆uj , i = 1, . . . , n
k=1 j=1
20

This can be written in vector notation as


∆ẋ = A ∆x + B ∆u where
 ∂f1 ∂f1
  ∂f1 ∂f1 
∂x1
· · · ∂x   ∂u1
··· ∂up  
 .. . . .
n
∂f ∂f
B =  ... ... ..  =
 
A= . . ..  = . 
∂fn ∂fn
∂x s ∂fn ∂fn
∂u s
∂x1
· · · ∂x n ∂u1
··· ∂up

We must assume hereby that the fi are continuously differentiable twice around (xs , us ),
especially there must be no step or bend of a characteristic.

Three-tank system example:


∂ ḣ1 ∂ ḣ1 ∂ ḣ1
ḣ1 = f (h1 , h2 , q1 ) → ∆ḣ1 = ∆h1 + ∆h2 + ∆q1
∂h1 ∂h2 ∂q1
s s s
√ √ r
∂ ḣ1 a 2g ∂ a 2g 1 a g
= − · (h1 − h2 )1/2 =− · √ =− ·
∂h1 A ∂h1 A 2 h1s − h2s A 2(h1s − h2s )
s
r
∂ ḣ1 a g
= ... = + ·
∂h2 A 2(h1s − h2s )
s

∂ ḣ1 1
=
∂q1 A
s

In the same way we find


∂ ḣ2 ∂ ḣ2 ∂ ḣ2
ḣ2 = f (h1 , h2 , h3 ) → ∆ḣ2 = ∆h1 + ∆h2 + ∆h3
∂h1 ∂h2 ∂h3
s s s
r
∂ ḣ2 a g
= + ·
∂h1 A 2(h1s − h2s )
s
r r r
∂ ḣ2 a g a g with (3.1) a g
= − · − · = −2 ·
∂h2 A 2(h1s − h2s ) A 2(h2s − h3s ) A 2(h1s − h2s )
s
r
∂ ḣ2 a g
= + ·
∂h3 A 2(h1s − h2s )
s

And finally
∂h3 ∂ ḣ3 ∂ ḣ3
ḣ3 = f (h2 , h3 , q2 ) → ∆ḣ1 = ∆h2 + ∆h3 + ∆q2
∂h2 s ∂h3 ∂q2
s s
r
∂ ḣ3 a g
= =+ ·
∂h2 A 2(h1s − h2s )
s
r r
∂ ḣ3 a g a g
= − · − ·
∂h3 A 2(h1s − h2s ) A 2(h3s − H)
s
DIT Deggendorf, Prof. Müller Teaching Units Selected Chapters of Control 21

∂ ḣ3 1
=
∂q2 A
s

If we introduce the abbreviations


r r
a g a g
f1 := · and f2 := ·
A 2(h1s − h2s ) A 2(h3s − H)
we can write the linearized system description in matrix notation as
   1 
−f1 f1 0 A
0
∆ẋ =  f1 −2f1 f1  ∆x +  0 0  ∆u.
0 f1 −f1 − f2 0 A1
If we further neglect the deltas in the equation above we obtain our well-known linear
system description

ẋ = A x + B u.

3.3 Exercises
Exercise 6.

You might also like