Autonomous Systems:Tutorial 4

Mechatronics Engineering

Dr. Islam S. M. Khalil

German University in Cairo

Faculty of Engineering and Material Science

March 25, 2018

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 1 / 17

Input State Linearization
In the previous tutorial, we discussed the issue of controllability
for a non linear system,and were able to study the sytem’s
controllability easily, if we were able to represent the system as:
ẋ = f (x) + g(x)u
However, An appropriate control input (u) to achieve a certain
objective (stability or trajectory tracking) is to be selected.
For Linear Systems: the concept was established, that the
same system can be presented in infinitely many
representations, where
ẋ = Ax + Bu
can also be presented as
ż = Az
e + Bu
e

while maintaining the same behaviour of the system.

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 2 / 17
To be able to go from one representation to another, this is
achieved by using state transformation, like:
Z = Tx
For example,     
z1 2 0 x1
= (1)
z2 0 5 x2
For Non-Linear Systems, can be presented in a different
representations if we were able to find a transformation of the
form
z = T (x)
Note that: T (x) should be diffeomorphism i.e its inverse exists.
Thus, the system represented as:
ẋ = f (x) + g(x)u
Maybe able to be presented as:
ż = Az + B w(z)[u
e − φ(z)]
e
Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 3 / 17
If we were able to present the system in this form:
ż = Az + B w(z)[u
e − φ(z)]
e

Then, if we select the input to be of the form

u = φ(z)
e e −1 ν
+ w(z)
will make the system be reduced to become:
ż = Az + Bν
which is similar to the linear system representation.
Now, the non linear system is represented in a linearized way.
Therefore, all the tools from linear control theory can be used.
One of these tools is the state feedback controller: ν = −Kz.

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 4 / 17

Converting the system from:
ẋ = f (x) + g(x)u

Using the state transformation

z = T (x)

to the linearized form:

ż = Az + Bν

Which can be done only if the following two conditions are

satisfied:
(1) The system is controllable.
(2) and involutive.

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 5 / 17

The proof of the desired transformation:

z = T (x) (2)

d
ż = T (x) (3)
dt
∂T (x) ∂x
ż = (4)
∂x ∂t
where, ∂x
∂t is the dynamics of the system i.e. ẋ.
Therefore,
∂T (x)
ż = [f (x) + g(x)u] (5)
∂x
∂T (x) ∂T (x)
ż = f (x) + g(x)u (6)
∂x ∂x

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 6 / 17

From another side, we want to reach the system representation:

ż = Az + B w(z)[u
e − φ(z)]
e (7)

To be able to represent the previous equation as function in x

not z, we will take the fact that:

z = T (x), x = T −1 (z) (8)

where, w(z)
e = w(x), φ(z)
e = φ(x).

ż = AT (x) + Bw(T −1 (z))[u − φ(T −1 (z))] (9)

ż = AT (x) + Bw(x)[u − φ(x)] (10)

ż = AT (x) − Bw(x)φ(x) + Bw(x)u (11)

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 7 / 17

For a controllable system, we are able to represent these
matrices as following:
   
0 1 0 0 .. 0
 0 0 1 0 ..   0 
   
A → Ac =   0 0 0 1 ..  , B → Bc =  .. 
   (12)
 .. .. .. .. ..   .. 
0 0 0 0 0 1

and,  
T1 (x)

 T2 (x) 

T (x) = 
 .. 
 (13)
 .. 
Tn (x)

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 8 / 17

Now, equating both equations (6) and (11), we may reach:
∂T (x)
f (x) = Ac T (x) − Bc w(x)φ(x) (14)
∂x
∂T (x)
g(x) = Bc w(x) (15)
∂x
By substituting with the previous matrices Ac , Bc and T (x) in
equations (14) and (15).
    
0 1 0 0 .. T1 (x) 0
 0 0 1 0 ..   T2 (x)   0 
∂T (x)     
  ..  −  ..  w(x)φ(x)
f (x) =  0 0 0 1 ..
∂x  
 .. .. .. .. ..   ..   .. 
  

0 0 0 0 0 Tn (x) 1
    (16)
T2 (x) 0
 T3 (x)   0 
∂T (x)    
f (x) =  ..  − 
   ..  (17)
∂x  ..  

.. 
0 w(x)φ(x)
Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 9 / 17
  
T2 (x)
 T3 (x) 
∂T (x)  
f (x) =  ..  (18)
∂x  
 .. 
−w(x)φ(x)

Similarly,
   
0 0
 0   0 
∂T (x) 
 w(x), ∂T (x) g(x) = 
  
g(x) =  .. ..  (19)
∂x   ∂x  
 ..   .. 
1 w(x)

Now using equations (18) and (19) we will be able to find the
required transformation T (x) to go from one representation to
another.

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 10 / 17

Example 1

e x2 − 1
     
ẋ1 0
= + u (20)
ẋ2 ax12 1
This system representation is ẋ = f (x) + g(x)u.
How can we transform it to another representation:
ż = Az + B w(z)[u
e − φ(z)]
e

Solution: we are searching for T(x) where,

 
T1 (x)
T (x) =
T2 (x)
as we have two states only. Using equation (19):
 
∂T (x) 0
g(x) = (21)
∂x w(x)

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 11 / 17

Therefore,
h i 0  ∂T1 (x)
∂T1 (x) ∂T1 (x)
∂x1 ∂x2
=0→ =0 (22)
1 ∂x2

From equation (22) we can conclude that T1 (x) is a function

only in x1 and doesn’t depend on x2 at all.
h i 0  ∂T2 (x)
∂T2 (x) ∂T2 (x)
∂x1 ∂x2
= w(x) → = w(x) (23)
1 ∂x2

Using equation (18):

 
∂T (x) T2 (x)
f (x) = (24)
∂x −w(x)φ(x)

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 12 / 17

Therefore,
h i  ex2 − 1  ∂T1 (x) x2
∂T1 (x) ∂T1 (x)
∂x1 ∂x2 2 = T2 (x) → (e −1) = T2 (x)
ax1 ∂x1
(25)
∂T1 (x)
x2 = 0 from equation (22). and,
h i  ex2 − 1 
∂T2 (x) ∂T2 (x)
= −w(x)φ(x) (26)
∂x1 ∂x2 ax12

∂T2 (x) x2 ∂T2 (x)

(e − 1) + (ax12 ) = −w(x)φ(x) (27)
∂x1 ∂x2
Now we have equations (22),(23),(25) and (27) to work with to
be able to continue we can start from the information that T1 (x)
is not a function in (x2 ) and assume T1 (x) to be any function in
(x1 ). Most simple case is to assume that:
T1 (x) = x1

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 13 / 17

Using equation (25):
∂T1 (x) x2
T2 (x) = ∂x1 (e − 1) = 1(ex2 − 1) → T2 (x) = (ex2 − 1)
Thus we reach that:
 
x1
T (x) = x
e −1
2

From equation (23):

∂T2 (x) ∂ x2
w(x) = ∂x2 = ∂x2 (e − 1) = ex2 → w(x) = ex2
From equation (27):

∂T2 (x) x2 ∂T2 (x)

(e − 1) + (ax12 ) = −w(x)φ(x) (28)
∂x1 ∂x2

0 + ex2 (ax12 ) = −ex2 φ(x) (29)

φ(x) = −ax12 (30)

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 14 / 17

Now substituting in the form:
ż =  − φ(z)]
 Az + B w(z)[u
e e
 
0 1 x1 0
ż = + ex2 [u + ax12 ]
0 0 ex2 − 1 1
But we want to represent the system as a function in the states
(z), Thus, we can use the info that
   
x1 z1
z = T (x) → =
ex2 − 1 z2
x
e 2 = z2 + 1
ax12 = az12
Therefore, we can reach this representation:
      
ż1 0 1 z1 0
= + (z2 + 1)[u + az12 ]
ż2 0 0 z2 1
which is in the form of :
ż = Az + B w(z)[u
e − φ(z)]
e
Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 15 / 17
where, w(z)
e = z2 + 1, φ(z)
e = −az12 Thus, if we used the input
(u) in this form:
u = φ(z)
e e −1 (z)ν
+w
we will be able to linearize the system u = −az12 + (z2 + 1)−1 ν
 
ż1
=
ż2
    
0 1 z1 0
+ (z2 + 1)[−az12 + (z2 + 1)−1 ν + az12 ]
0 0 z2 1
which will be reduced to:
      
ż1 0 1 z1 0
= + ν
ż2 0 0 z2 1
which is a linearized representation of the non linear system we
initially started with.
ż = Az + Bν

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 16 / 17

 Any Questions ?
Thank you

Dr. Islam S. M. Khalil Autonomous Systems:Tutorial 4 17 / 17