Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd
Upload
54 views

State Observers (p751)

The document discusses state observers, which are used to estimate states that cannot be directly measured. A full-order state observer generates estimates of all states based on the measurable outputs and control inputs. The observer dynamics are defined by a matrix (A-KeC) where Ke is chosen so that the estimation error approaches zero over time. Once designed, the observer and controller can be designed independently according to the separation principle. An example demonstrates designing an observer and state feedback controller to speed up a system's response.

Uploaded by

Abdullah Abu Lehyah
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
54 views

State Observers (p751)

The document discusses state observers, which are used to estimate states that cannot be directly measured. A full-order state observer generates estimates of all states based on the measurable outputs and control inputs. The observer dynamics are defined by a matrix (A-KeC) where Ke is chosen so that the estimation error approaches zero over time. Once designed, the observer and controller can be designed independently according to the separation principle. An example demonstrates designing an observer and state feedback controller to speed up a system's response.

Uploaded by

Abdullah Abu Lehyah
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 6

State Observers ( p751)

A disadvantage of pole assignment is that it requires all states of a system


to be available for direct measurement in order to be fedback. Such
availability may not be the case as some states are inaccessible,
inconvenient to measure , sensor unavailability , cost of measurement , to
the extent that some states may be non-physical and therefore cannot be
measured.

In such cases another system, named an observer is constructed to give


estimates of the states in the steady state. The observer can be a physical
hardware or a software . It requires the outputs, y ( which are necessarily
available and measurable ) and the control signal u to generate signals
approaching those of the actual states in the steady state .

Full-Order State Observer .

All states are observed even if some of them are available for direct
measurement , hence, the name full-order , see Figure 10-11 p 753.
The states x are assumed unmeasurable (therefore cannot be fedback )
whereas the estimated states ~x are necessarily available for direct
measurement , hence, used for feedback as clarified later on.

The dynamics describing the observer is given by

ẋ−~
˙ Ax+ Bu− A ~
x= x−Bu−K e ( Cx−C ~
x)

ẋ−~
˙ ( A−K e C ) ( x−~
x= x)

Defining e=x−~x as the estimation error yields

ė=( A−K e C ) e

For the error signal to approach zero as time approaches infinity the
eigenvalues of A−K e C should have negative real parts. Hence, a need for
eigenvalue assignment to ensure stability and proper dynamics of the error
signal. To change all the eigenvalues arbitrarily, the system has to be
completely observable.

Any eigenvalue assignment method you know of can be used after


T
adaptation to observer design, i.e. K e is calculated as K e =( K B ) where
K B does the assignment to a system described by AT ∧BT .

Note that K B is a row matrix , while K e is a column matrix .

The Ackermann's method has been adapted ( p756) to calculate K e as


1
 C  0
 CA  0
K e   ( A)    
    
 n 1   
CA  1 

As a rule of thumb, the observer poles are to be 2 to 5 times faster than the
controller poles to ensure a quick error signal convergence to zero. Usually
compromises are taken in the presence of measurement noise.
Having designed an observer , a state feedback controller is sought of the
form u=r−K ~x to satisfy system requirement .

It may be suspected that a conflict may arise in the selection of K e and K


necessitating a need for compromises in their selection. Fortunately , this
is not the case as the separation principle shows ( end of p760) that the
determination of K e and K are independent of each other, i.e. each one can
be designed separately in isolation of the other.

Example : consider the following system


 0 1 0 
x(0)   0.4 0.8
T
x  x   u ;
 2 3  2
y   1 1 x

The observer is to have


-5,and -5 as eigenvalues . Using MATLAB

≫Ke=(acker(A,B,[ -5 -5 ]))’
5
Gives Ke   
 2 
Simulating the system and the observer using Simulink as shown below
confirms that the observer faithfully estimates the two states.

The system may be judged as slow ( settling time is about 4 units), in which
case a state feedback controller is required to speed it up. The original
eigenvalues of the system are -2, and -3. Let us speed up the system by
choosing -4, and -4 as system closed loop eigenvalues . Hence,

≫K=acker(A,B,[-4 -4]) giving K= [ 7 2.5 ].

Note that the states fedback are the estimated states not the original ones
as they are assumed unmeasurable. Note also from where the controller u
is taken and how it is applied to the observer .

Apply feedback to the system through K where r is taken as a unit step.


Running the system , the results are as shown below . Note now the system
is now faster as the settling time is about 2.5 units, however, the steady state
has been altered.
Exercise : Calculate alternative Ke∧K . Observe the difference.

Exercise : Re-run the example worked out in the book.

Exercise : Design and implement an observer for an unstable system.

You might also like