1
Full-State Feedback Control Design Examples
A. Example 1
The open-loop system is described by the following state space model.
·
x(t) = Ax(t) + Bu(t), y(t) = Cx(t) + Du(t) (1)
∙ ¸ ∙ ¸
2 −3 1 £ ¤
A = , B= , C= 1 0 , D=0 (2)
1 3 −1
Obviously, the model is second order and represents a single-input, single-output (SISO) system, that is, n = 2 and r = m = 1.
The eigenvalues of the open-loop system are λOL = {2.5 ± j1.66} . Since the open-loop system is unstable, some form of
feedback is required to produce a stable system. In this example, full-state feedback will be used. The controllability matrix
for the open-loop system is
∙ ¸
£ ¤ 1 5
P = B AB = , Rank (P ) = rP = 2 = n (3)
−1 −2
Since the rank of P is equal to n, the system is completely controllable. Therefore, the closed-loop eigenvalues can be
placed at arbitrary places in the complex plane as long as if one of the eigenvalues is complex, its complex conjugate is also
an eigenvalue. For this example, the closed-loop eigenvalues will be placed at λCL = {−1, −2} . Although this choice of
closed-loop eigenvalues may not give satisfactory performance, they are asymptotically stable and will serve for this example.
With full-state feedback, u(t) = −Kx(t) + v(t), and the closed-loop state equations are
·
x(t) = (A − BK) x(t) + Bv(t) (4)
The closed-loop eigenvalues and eigenvectors are related by
[A − BK] ψ i = λi ψ i ⇒ [λi I − A + BK] ψ i = 0 (5)
Since the r × n feedback gain matrix K is unknown, Eqn. (5) can be rewritten as shown below which will lead to a procedure
for computing the value for K.
∙ ¸ ∙ ¸
£ ¤ ψi £ ¤ ψi £ ¤
λi I − A BK =0 ⇒ λi I − A B =0 ⇒ λi I − A B ξ i = 0 (6)
ψi Kψ i
where ψ i is the n-dimensional eigenvector associated with eigenvalue λi , and ξ i is a vector of dimension n + r (3 in this
example). £ ¤
For each desired closed-loop eigenvalue λi , the matrix λi I − A B will be formed, and the Row-Reduced Echelon
(RRE) technique will be used to find the vector ξ i . Once that is done for each λi , the gain matrix K can be computed by
partitioning the ξ i into ψ i and Kψ i . The procedure is described in detail for this example.
For λ = −1, the steps in the process are:
∙ ¸ ∙ ¸
£ ¤ −3 3 1 1 −1 −1/3
−I − A B ξ 1 = 0 = ξ1 ⇒ ξ1 = 0 (7)
−1 −4 −1 1 4 1
∙ ¸ ∙ ¸ ∙ ¸
1 −1 −1/3 1 −1 −1/3 1 0 −0.0667
⇒ ξ1 = 0 ⇒ ξ1 = 0 ⇒ ξ1 = 0 (8)
0 5 4/3 0 1 0.2667 0 1 0.2667
The last matrix in (8) is in the RRE form. The first row of that matrix shows that the first element in ξ 1 equals 0.0667 times the
third element, and the second row of the matrix shows that the second element in ξ 1 equals −0.2667 times the third element. The
£ ¤T
third element of ξ 1 , which corresponds to Kψ 1 , is arbitrary. That value will be set to 1, so ξ 1 = 0.0667 −0.2667 1 .
For λ = −2, the same steps will be followed.
∙ ¸ ∙ ¸
£ ¤ −4 3 1 1 −3/4 −1/4
−2I − A B ξ 2 = 0 = ξ2 ⇒ ξ2 = 0 (9)
−1 −5 −1 1 5 1
∙ ¸ ∙ ¸ ∙ ¸
1 −3/4 −1/4 1 −3/4 −1/4 1 0 −0.0870
⇒ ξ2 = 0 ⇒ ξ2 = 0 ⇒ ξ2 = 0 (10)
0 5.75 1.25 0 1 0.2174 0 1 0.2174
2
The last matrix in (10) is in the RRE form. The first row of that matrix shows that the first element in ξ 2 equals 0.087 times the
third element, and the second row of the matrix shows that the second element in ξ 2 equals −0.2174 times the third element.
£ ¤T
The third element of ξ 2 , which corresponds to Kψ 2 , is arbitrary. That value will be set to 1, so ξ 2 = 0.087 −0.2174 1 .
The columns of the modal matrix M of eigenvectors are the first n = 2 rows of the vectors ξ 1 and ξ 2 . The third elements
of those two vectors form a matrix Q (not to be confused with the observability matrix). For this example, M and Q are
∙ ¸
0.0667 0.0870 £ ¤
M= , Q= 1 1 (11)
−0.2667 −0.2174
£ ¤ £ ¤
Since Q = Kψ 1 Kψ 2 and M = ψ 1 ψ 2 , it should be clear that Q = KM, and the gain matrix can be solved
from
∙ ¸−1
−1
£ ¤ 0.0667 0.0870 £ ¤
K = QM = 1 1 = 5.6667 −2.3333 (12)
−0.2667 −0.2174
The closed-loop matrix ACL = A − BK and its eigenvalues are given by
∙ ¸ ∙ ¸ ∙ ¸
2 −3 1 £ ¤ −3.6667 −0.6667
ACL = − 5.6667 −2.3333 = , λCL = {−1, −2} (13)
1 3 −1 6.6667 0.6667
The final closed-loop system is
∙ ¸ ∙ ¸
· −3.6667 −0.6667 1 £ ¤
x(t) = x(t) + v(t), y(t) = 1 0 x(t) (14)
6.6667 0.6667 −1
B. Example 2
For this example, the open-loop system model is the same is in the first example. The only change is in the desired closed-
loop eigenvalues. Now the desired eigenvalues are λCL = {−2, −2} . Since the eigenvalues are repeated, there is a possibility
that there will be one eigenvector and one generalized eigenvector for this system. If that is the case, the RRE procedure will
need to slightly£ modified. ¤
The matrix λI − A B is given by
∙ ¸
£ ¤ λ−2 3 1
λI − A B = (15)
−1 λ − 3 −1
£ ¤
The rank of λI − A B equals 2 regardless of the value of λ. Since the matrix has 3 columns and the eigenvalue’s
multiplicity is 2, the degeneracy is q = 3 − 2 = 1. Therefore, for the repeated eigenvalue λ = −2, there is one eigenvector
and one generalized eigenvector (the simple degeneracy case). The eigenvector can be found using the same procedure as in
Example 1. Since the eigenvalue
£ in this example is the¤Tsame as the second eigenvalue in the first example, the same eigenvector
will be obtained. Thus, ξ 1 = 0.087 −0.2174 1 .
The generalized eigenvector will be found using the chain rule discussed in Chapter 7 of the text1 . The eigenvalue/eigenvector
expression in Eqn. (5) with the repeated eigenvalue becomes
[λ2 I − A + BK] ψ 2 = −ψ 1 (16)
The RRE method can be applied to this equation as before, but it must also be applied to the vector −ψ 1 . An easy way to
accomplish this is to include that vector with the [λ2 I − A + BK] matrix and perform RRE on the following matrix.
h i
.. .. (17)
λ2 I − A . B . −ψ 1
That matrix and the result of the RRE process are shown below.
h i ∙ −4 3 1 −0.087
¸ ∙
1 0 −0.087 −0.0095
¸
.. ..
−2I − A . B . −ψ 1 = −1 −5 −1 0.2174 ⇒
0 1 0.2174 −0.0416
(18)
As before, the third element in the ξ 2 vector will be arbitrary. Letting that value be α2 for the moment, the first row
of the final RRE matrix shows that the first element of ξ 2 equals 0.087α2 + 0.0095, and the second element of ξ 2 equals
−0.2174α2 + 0.0416. The following vectors show the resulting ξ 2 for several different values of α2 .
1 Modern Control Theory, 3rd Edition, William Brogan, Prentice Hall, 1991
3
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
0.0964 0.2703 0.4442 −0.1645 1.1399 0.0095
⎣ −0.1758 ⎦ , ⎣ −0.6106 ⎦ , ⎣ −1.0454 ⎦ , ⎣ 0.4764 ⎦ , ⎣ −2.7845 ⎦ , ⎣ 0.0416 ⎦ (19)
1 3 5 −2 13 0
Any one of the vectors shown in (19) used along with the first vector ξ 1 for this example will produce the same K matrix.
For example, if α2 = −2 is used, the gain matrix is
∙ ¸−1
£ ¤ 0.087 −0.1645 £ ¤
K = QM −1 = 1 −2 = 7.3333 −1.6667 (20)
−0.2174 0.4764
and if α2 = 13 is used, the same gain matrix is obtained.
∙ ¸−1
£ ¤ 0.087 1.1399 £ ¤
K = QM −1 = 1 13 = 7.3333 −1.6667 (21)
−0.2174 −2.7845
The closed-loop matrix ACL = A − BK and its eigenvalues are given by
∙ ¸ ∙ ¸ ∙ ¸
2 −3 1 £ ¤ −5.3333 −1.3333
ACL = − 7.3333 −1.6667 = , λCL = {−2, −2} (22)
1 3 −1 8.3333 1.3333
and the final closed-loop system is
∙ ¸ ∙ ¸
· −5.3333 −1.3333 1 £ ¤
x(t) = x(t) + v(t), y(t) = 1 0 x(t) (23)
8.3333 1.3333 −1