MODERN CONTROL SYSTEMS

Lecture 6

Full order State Oserver Design

Emam Fathy

Department of Electrical and Control Engineering

email: [email protected]

http://www.aast.edu/cv.php?disp_unit=346&ser=68525
1
 Introduction
• In the pole-placement approach to the design of control systems, we
assumed that all state variables are available for feedback.

• In practice, however, not all state variables are available for feedback,
for instance, if the component is in an inaccessible location, or the
sensors are expensive.

• Then we need to estimate unavailable state variables.

 Introduction
• Estimation of unmeasurable state variables is commonly called
observation.
• A device (or a computer program) that estimates or observes
the state variables is called a state estimator, state observer, or
simply an observer.
• There are two types of state observers
– Full Order State Observer
• If the state observer observes all state variables of the system,
regardless of whether some state variables are available for direct
measurement, it is called a full-order state observer.

– Reduced Order State Observer

• If the state observer observes only those state variables which are not
available for direct measurement, it is called a reduced-order state
observer.
 Topology of State Feedback Control with Observer
Based Approach

• State feedback with state observer

Ke
 Topology of State Feedback Control with Observer
Based Approach
• State Feedback with observer
 State Observer
• A state observer estimates the state variables based on
the measurements of the output and control variables.

• Here the concept of observability plays an important

role.

• State observers can be designed if and only if the

observability condition is satisfied.
 State Observer
• Consider the plant defined by
𝒙ሶ = 𝑨𝒙 + 𝑩𝑢
𝑦 = 𝑪𝒙
• The mathematical model of the observer is basically the same
as that of the plant, except that we include an additional term
that includes the estimation error to compensate for
inaccuracies in matrices A and B and the lack of the initial error.

• The estimation error or observation error is the difference

between the measured output and the estimated output.

• The initial error is the difference between the initial state

and the initial estimated state.
 State Observer
• Thus we define the mathematical model of observer to be
෕ሶ = 𝑨෕
𝒙 𝒙 + 𝑩𝑢 + 𝑲𝒆 (𝑦 − 𝑪෕
𝒙)
• Where 𝒙 ේ is estimated output and
෕ is estimated state vector, 𝑪𝒙
𝑲𝒆 is observer gain matrix.
 Full Order State Observer
• The order of the state observer that will be discussed here is
the same as that of the plant.
• Consider the plant define by following equations
𝒙ሶ = 𝑨𝒙 + 𝑩𝑢 (1)
𝑦 = 𝑪𝒙
• Equation of state observer is given as
෕ሶ = 𝑨෕
𝒙 𝒙 + 𝑩𝑢 + 𝑲𝑒 (𝑦 − 𝑪෕
𝒙) (2)
• To obtain the observer error equation, let us subtract
Equation (2) from Equation (1):
෕ሶ = 𝑨𝒙 + 𝑩𝑢 − [𝑨෕
𝒙ሶ − 𝒙 𝒙 + 𝑩𝑢 + 𝑲𝑒 𝑦 − 𝑪෕
𝒙 ]

෕ሶ = 𝑨𝒙 + 𝑩𝑢 − 𝑨෕
𝒙ሶ − 𝒙 𝒙 − 𝑩𝑢 − 𝑲𝑒 𝑪𝒙 − 𝑪෕
𝒙
 Full Order State Observer
෕ሶ = 𝑨𝒙 + 𝑩𝑢 − 𝑨෕
𝒙ሶ − 𝒙 𝒙 − 𝑩𝑢 − 𝑲𝑒 𝑪𝒙 − 𝑪෕
𝒙
• Simplifications in above equation yields

෕ሶ = 𝑨(𝒙 − 𝒙
𝒙ሶ − 𝒙 ෕ ) − 𝑲𝑒 𝑪 𝒙 − 𝒙
෕ (3)

• Define the difference between 𝒙 and 𝒙

෕ as the error vector
e.
𝒆=𝒙−𝒙 ෕

• Equation (3) can now be written as

𝒆ሶ = 𝑨𝒆 − 𝑲𝑒 𝑪𝒆
𝒆ሶ = (𝑨 − 𝑲𝑒 𝑪)𝒆
 Full Order State Observer
𝒆ሶ = (𝑨 − 𝑲𝒆 𝑪)𝒆
• From above equation we see that the dynamic behavior of the
error vector is determined by the eigenvalues of matrix A-KeC.

• If matrix A-KeC is a stable matrix, the error vector will

converge to zero for any initial error vector e(0).

• That is, 𝒙
෕(𝒕) will converge to 𝒙(𝒕) regardless of the values of
x(0).

• And if the eigenvalues of matrix A-KeC are chosen in such a

way that the dynamic behavior of the error vector is
asymptotically stable and is adequately fast, then any error
vector will tend to zero (the origin) with an adequate speed.
 Full Order State Observer
𝒆ሶ = (𝑨 − 𝑲𝒆 𝑪)𝒆
• If the plant is completely observable, then it can be proved
that it is possible to choose matrix Ke such that A-KeC has
arbitrarily desired eigenvalues.

• That is, the observer gain matrix Ke can be determined to yield

the desired matrix A-KeC.
 Duality Property
• The design of the full-order observer becomes that of
determining an appropriate Ke such that A-KeC has
desired eigenvalues.

• Thus, the problem here becomes the same as the pole-

placement problem.

• In fact, the two problems are mathematically same.

• This property is called duality.

 Duality Property
• Consider the system defined by
𝒙ሶ = 𝑨𝒙 + 𝑩𝑢
𝑦 = 𝑪𝒙
• In designing the full-order state observer, we may solve
the dual problem, that is, solve the pole-placement
problem for the dual system.
𝒛ሶ = 𝑨∗ 𝒛 + 𝑪∗ 𝑣
𝑛 = 𝑩∗ 𝒛
• Assuming the control signal 𝑣 to be

𝑣 = −𝑲𝒛
 Duality Property
• If the dual system is completely state controllable, then the
state feedback gain matrix K can be determined such that
matrix A*-C*K will yield a set of the desired eigenvalues.

• If 𝜇1 , 𝜇2 ,…, 𝜇𝑛 , are the desired eigenvalues of the state

observer matrix, then by taking the same 𝜇𝑖′ 𝑠 as the
desired eigenvalues of the state-feedback gain matrix of
the dual system, we obtain
𝑠𝐼 − (𝑨∗ − 𝑪∗ 𝑲 ) = (𝑠 − 𝜇1 )( 𝑠 − 𝜇2 ) ⋯ 𝑠 − 𝜇𝑛
• Noting that the eigenvalues of A*-C*K and those of A-K*C
are the same, we have
𝑠𝐼 − (𝑨∗ − 𝑪∗ 𝑲 ) = 𝑠𝐼 − (𝑨 − 𝑲∗ 𝑪 )
 Duality Property
𝑠𝐼 − (𝑨∗ − 𝑪∗ 𝑲 ) = 𝑠𝐼 − (𝑨 − 𝑲∗ 𝑪 )
• Comparing the characteristic polynomial 𝑠𝐼 − (𝑨 − 𝑲∗ 𝑪 ) and
the characteristic polynomial for the observer system
𝑠𝐼 − (𝑨 − 𝑲𝑒 𝑪 ) , we find that Ke and K* are related by

𝑠𝐼 − (𝑨 − 𝑲∗ 𝑪 ) = 𝑠𝐼 − (𝑨 − 𝑲𝑒 𝑪 )

𝑲∗ = 𝑲𝑒

• Thus, using the matrix K determined by the pole-

placement approach in the dual system, the observer
gain matrix Ke for the original system can be determined
by using the relationship Ke=K*.
 Observer Gain Matrix
• Direct Substitution Method
• Ackermann’s Formula
• Using Transformation Matrix Q
 Observer Gain Matrix
• Direct Substitution Method

𝑘1
𝑘2
𝑲𝒆 = ⋮
𝑘3
𝑘𝑛
 Observer Gain Matrix
• The feedback signal through the observer gain matrix Ke serves
as a correction signal to the plant model to account for the
unknowns in the plant.

• If significant unknowns are involved, the feedback signal

through the matrix Ke should be relatively large.

• However, if the output signal is contaminated significantly by

disturbances and measurement noises, then the output y is not
reliable and the feedback signal through the matrix Ke should
be relatively small.
 Observer Gain Matrix
• The observer gain matrix Ke depends on the desired
characteristic equation
(𝑠 − 𝛽1 )( 𝑠 − 𝛽2 ) ⋯ 𝑠 − 𝛽𝑛 = 0
• The observer poles must be two to five times faster than the
controller poles to make sure the observation error (estimation
error) converges to zero quickly.

• This means that the observer estimation error decays two to

five times faster than does the state vector x.

• Such faster decay of the observer error compared with the

desired dynamics makes the controller poles dominate the
system response.
 Observer Gain Matrix
• It is important to note that if sensor noise is considerable,
we may choose the observer poles to be slower than two
times the controller poles, so that the bandwidth of the
system will become lower and smooth the noise.

• In this case the system response will be strongly influenced

by the observer poles.

• If the observer poles are located to the right of the

controller poles in the left-half s plane, the system
response will be dominated by the observer poles rather
than by the control poles.
 Observer Gain Matrix
• In the design of the state observer, it is desirable to determine
several observer gain matrices Ke based on several different
desired characteristic equations.

• For each of the several different matrices Ke , simulation tests

must be run to evaluate the resulting system performance.

• Then we select the best Ke from the viewpoint of overall system

performance.

• In many practical cases, the selection of the best matrix Ke boils

down to a compromise between speedy response and
sensitivity to disturbances and noises.
Example 1
 Example
• Consider the system
𝑥ሶ1 20.6 𝑥1
0 0
= 𝑥 + 𝑢(𝑡)
𝑥ሶ 2 10 2 1
𝑥1
𝑦= 0 1 𝑥
2

• We use observer based approach to design state feedback control

such that
𝑢 = −𝑲෕ 𝒙

• Design a full-order state observer assume that the desired eigenvalues of

the observer matrix are 𝛽1 = −10, 𝛽2 = −10.
 Example
𝑥ሶ1 0 20.6 𝑥1 0
= 𝑥 + 𝑢(𝑡)
𝑥ሶ 2 1 0 2 1
𝑥1
𝑦= 0 1 𝑥
2

• Let us examine the observability matrix first

𝐶 0 1
𝑂= =
𝐶𝐴 1 0

• Since rank(O)=2 the given system is completely state observable and the
determination of the desired observer gain matrix is possible.
Direct Substitution Method
 𝑥ሶ1 020.6 𝑥1 0
= 𝑥2 + 1 𝑢(𝑡)
𝑥ሶ 2 1 0
𝑥1
𝑦= 0 1 𝑥
2

• The characteristic equation of observer error matric is

𝑠𝑰 − 𝑨 + 𝑲𝑒 𝑪 = 0

• Assuming 𝑘𝑒1
𝑲𝑒 =
𝑘𝑒2

𝑠 0 0 20.6 𝑘𝑒1
𝑠𝑰 − 𝑨 + 𝑲𝑒 𝑪 = − + 0 1
0 𝑠 1 0 𝑘𝑒2

= 𝑠 2 +𝑘𝑒2 𝑠 − 20.6 + 𝑘𝑒1

• The desired characteristic polynomial is

(𝑠 − 𝛽1 )( 𝑠 − 𝛽2 ) = 𝑠 2 + 20𝑠 + 100
• Comparing coefficients of different powers of s

𝑠 2 + 20𝑠 + 100 = 𝑠 2 +𝑘𝑒2 𝑠 − 20.6 + 𝑘𝑒1

120.6
𝑲𝑒 =
20
• We get the same Ke regardless of the method employed.

• The equation for the full-order state observer is given by

෕ሶ = 𝑨෕
𝒙 𝒙 + 𝑩𝑢 + 𝑲𝒆 (𝑦 − 𝑪෕
𝒙)

෕ሶ 𝟏
𝒙 0 ෕1
20.6 𝒙 0 120.6 ෕1
𝒙
= + 𝑢(𝑡) + (𝑦 − 0 1 )
෕ሶ 𝟐
𝒙 1 0 ෕
𝒙2 1 20 ෕
𝒙2

෕ሶ 𝟏
𝒙 0 ෕1
20.6 𝒙 0 120.6 120.6 ෕1
𝒙
= + 𝑢(𝑡) + 𝑦− 0 1
෕ሶ 𝟐
𝒙 1 0 ෕2
𝒙 1 20 20 ෕2
𝒙

෕ሶ 𝟏
𝒙 ෕1
0 −100 𝒙 0 120.6
= + 𝑢(𝑡) + 𝑦
ሶ𝒙
෕𝟐 1 −20 ෕
𝒙 2 1 20
End of Lec

35
Using Transformation Matrix Q
 𝑥ሶ1 0 20.6 𝑥1 0
= 𝑥 + 𝑢(𝑡)
𝑥ሶ 2 1 0 2 1
𝑥1
𝑦= 0 1 𝑥
2

• The given system is already in the observable canonical form. Hence,

the transformation matrix Q is I.
 𝑥ሶ1 0 20.6 𝑥1 0
= 𝑥 + 𝑢(𝑡)
𝑥ሶ 2 1 0 2 1
𝑥1
𝑦= 0 1 𝑥
2

• The characteristic equation of the given system is

𝑠𝐼 − 𝐴 = 𝑠 2 − 20.6 = 0

• We have
𝑎1 = 0, 𝑎2 = −20.6
 𝑥ሶ1 0 20.6 𝑥1 0
= 𝑥 + 𝑢(𝑡)
𝑥ሶ 2 1 0 2 1
𝑥1
𝑦= 0 1 𝑥
2

• The desired characteristic equation of the system is

(𝑠 − 𝛽1 )( 𝑠 − 𝛽2 ) = (𝑠 + 10) 𝑠 + 10

(𝑠 − 𝛽1 )( 𝑠 − 𝛽2 ) = 𝑠 2 + 20𝑠 + 100
• We have
𝛼1 = 20, 𝛼2 = 100
• Observer gain matrix Ke can be calculated using following formula
𝛼2 − 𝑎 2
𝑲𝑒 = 𝛼 − 𝑎
1 1
• Where

𝑎1 = 0, 𝑎2 = −20.6 𝛼1 = 20, 𝛼2 = 100

100 − (−20.6)
𝑲𝑒 =
20 − 0

120.6
𝑲𝑒 =
20
Ackermann’s formula
 𝑥ሶ1 0 20.6 𝑥1 0
= 𝑥 + 𝑢(𝑡)
𝑥ሶ 2 1 0 2 1
𝑥1
𝑦= 0 1 𝑥
2

• Using Ackermann’s formula

−1
𝑪 0
𝑲𝑒 = ∅(𝑨)
𝑪𝑨 1
• Where
∅ 𝑨 = 𝐴2 + 𝛼1 𝐴 + 𝛼2 𝐼

𝛼1 = 20, 𝛼2 = 100

∅ 𝑨 = 𝐴2 + 20𝐴 + 100𝐼
 ∅ 𝑨 = 𝑨2 + 20𝑨 + 100𝑰
2
0 20.6 0 20.6 1 0
∅ 𝑨 = + 20 + 100
1 0 1 0 0 1

120.6 412
∅ 𝑨 =
20 120.6
• Using Ackermann’s formula
−1
𝑪 0
𝑲𝑒 = ∅(𝑨)
𝑪𝑨 1
−1
120.6 412 0 1 0
𝑲𝑒 =
20 120.6 1 0 1

120.6
𝑲𝑒 =
20
• We get the same Ke regardless of the method employed.

• The equation for the full-order state observer is given by

෕ሶ = 𝑨෕
𝒙 𝒙 + 𝑩𝑢 + 𝑲𝒆 (𝑦 − 𝑪෕
𝒙)

෕ሶ 𝟏
𝒙 0 ෕1
20.6 𝒙 0 120.6 ෕1
𝒙
= + 𝑢(𝑡) + (𝑦 − 0 1 )
෕ሶ 𝟐
𝒙 1 0 ෕
𝒙2 1 20 ෕
𝒙2

෕ሶ 𝟏
𝒙 0 ෕1
20.6 𝒙 0 120.6 120.6 ෕1
𝒙
= + 𝑢(𝑡) + 𝑦− 0 1
෕ሶ 𝟐
𝒙 1 0 ෕2
𝒙 1 20 20 ෕2
𝒙

෕ሶ 𝟏
𝒙 ෕1
0 −100 𝒙 0 120.6
= + 𝑢(𝑡) + 𝑦
ሶ𝒙
෕𝟐 1 −20 ෕
𝒙 2 1 20
Example 2
 Example-2
• Design a regulator system for the following plant:
𝑥ሶ1 0 20.6 𝑥1 0
= 𝑥 + 𝑢(𝑡)
𝑥ሶ 2 1 0 2 1
𝑥1
𝑦= 1 0 𝑥
2
• The desired closed-loop poles for this system are at 𝜇1 = −1.8
+ 𝑗2.4, 𝜇2 = −1.8 − 𝑗2.4. Compute the state feedback gain matrix K
to place the poles of the system at desired location.

• Suppose that we use the observed-state feedback control instead of

the actual-state feedback. The desired eigenvalues of the observer
matrix are 𝛽1 = −8, 𝛽2 = −8.

• Obtain the observer gain matrix Ke and draw a block diagram for the
observed-state feedback control system.
End of Lec

48