Digital Control Systems

(EP-5511)

Lecture-10
State Space Analysis …

Belaynesh Belachew (capt.)

Defence Engineering College

1
 Discretization of Continuous Time State
Space Equations
In digital control of continuous-time plants, we need to
convert continuous-time state-space equations into
discrete-time state-space equations.
Such conversion can be done by introducing fictitious
samplers and fictitious holding devices into
continuous-time systems.
The error introduced by discretization may be made
negligible by using a sufficiently small sampling
period compared with the significant time constant
of the system.
2
 Continued
Suppose the continuous time state space system
x(t )  Ax(t )  Bu (t )
y (t )  Cx(t )  Du (t )
and apply an input that changes only at discrete
sampling intervals.
It would be nice if matrices G and H could find,
independent of t or k so that a discrete time model of
the system could obtain,

x((k  1)T )  G (T ) x(kT )  H (T )u (kT )

y (kT )  Cx(kT )  Du (kT )
The values of the matrices G(T) and H(T) will be
3 determined.
 Continued
It will turn out that while they are constant for a particular
sampling interval, they depend on the value of the
sampling interval, so for that reason they can be
written as G(T) and H(T).
Using the solution of the continuous time state space
equation to calculate the values of the state x at times
kT and (k + 1)T.
These are
( k 1)T
 e A Bu ( )d
A ( k 1)T A ( k 1)T
x((k  1)T )  e x(0)  e
0


kT
x(kT )  e AkT
x(0)  e AkT
e A Bu ( )d
0

Writing x((k+1)T) in terms of x(kT) and multiplying all

4
terms of x(kT) by eAT and solve for eA(k+1)T x(0),
 Continued


kT
e A ( k 1)T
x(0)  e x(kT )  e
AT A ( k 1)T
e  A
Bu ( )d
0

Substituting for eA(k+1)Tx(0) in x((k+1)T) we obtain:

which, by linearity of integration, is equivalent to
 ( k 1)T e  A Bu ( )d
 0
A ( k 1)T
x((k  1)T )  e x(kT )  e
AT

  e A Bu ( )d 
kT

0 

Next, notice that within the interval from kT to (k+1)T, u(t) = u(kT) is
constant, as is the matrix B, so take them out of the integral to obtain
( k 1)T
 e  A Bu ( )d
A ( k 1)T
x((k  1)T )  e x(kT )  e
AT
kT
5
 Continued

( k 1)T
 e  A d Bu (kT )
A ( k 1)T
x((k  1)T )  e x(kT )  e
AT
kT

  [kT ,  k +1 T )
( k 1)T
x((k  1)T )  e x(kT )  
AT
e A[( k 1)T  ]d Bu (kT )
kT
As τ ranges from kT to (k + 1)T the exponent of e
ranges from T to 0. Accordingly, let’s define a new
variable λ= (k +1)T−τ.
Then dλ =−dτ and λ ranges from T to 0 as τ ranges from
kT to (k + 1)T. Thus:
x((k  1)T )  e x(kT )   e d  Bu (kT ) λ ϵ [0, kT)
T
AT A
0 6
 Continued
The state update equation is:
G (T )  e ; H (T )   e A d  B
T
AT
0
As A is invertible, we can easily integrate, using the fact
that d
e AT  Ae AT  e AT A
dx
to obtain
H (T )  A  Ae d  B  A e
T
1 A 1 A T
B
0  0

 A1 (e AT  I ) B  (e AT  I ) BA1
Finally, note that while I restricted the value of τ and λ
to lie within a single sampling interval, k appears
nowhere in the expressions for G(T) and H(T). 7
 Continued
Therefore, the solution to x((k  1)T )  G(T ) x(kT )  H (T )u (kT )
is k 1
x(kT )  (G(T )) k x(0)   (G(T )) k  j 1 H (T )u ( jT ), k  1, 2,3,...
j 0

and it can be seen that at the sampling instants kT, this has
exactly the same value as is obtained using the
continuous time state space equation.
Specifically, (G(T ))k  (e AT ) k  e AkT ,
and since the input u(t) is constant on sampling intervals,
k 1

 Bu ( )d   e A( k  j 1)T A1 (e AT  I ) Bu ( jT )

kT
AkT  A
e e
0
j 0
k 1
  (G(T )) k  j 1 H (T )u ( jT ).
j 0 8
 Continued
Example 1: Obtain the discrete-time state and output
equations and the pulse transfer function of the
following continuous-time system: (T = 1)
Y ( s) 1
G( s)  
U ( s) s( s  2)
Solution:
 x1  0 1   x1  0
 x   0 2  x   1  u
 2   2  
 x1 
y  1 0  
 x2 
The desired discrete-time state equation will have the form
x((k  1)T )  G(T ) x(kT )  H (T )u(kT )
1 1 (1  e 2T ) 
9
G (T )  e AT  2
2T 
0 e 
 Continued
   T 1  0 
T 1 (1  e 2 ) 
A
H (T )   e d  B     2
2 d   1 
 0   
0 0 e
 1 (T  e 1 ) 
2 T

  21 2T 
2
 2 (1  e ) 
Thus
 x1 ((k  1)T )   1 1 (1  e 2T )  
 x1 (kT )   2
1 (T  e2T 1 ) 

 x2 ((k  1)T )  0
2
2T     1 (1  e 2T ) 
2 u (kT )
 2  2
x ( kT )
 e 
 x1 (kT ) 
y (kT )  1 0  
 2
x ( kT ) 
Therefore:
 x1 (k  1)   1 0.4323  x1 (k )   0.2838 u (k )
 x2 (k  1)  0 0.1353  x2 (k )  0.4323
 x1 (k ) 
10
y (k )  1 0  
 2 
x ( k )
 Lyapunov Stability Analysis
Lyapunov stability anaylsis plays an important role in
the stability analysis of control systems described by
state-space equations.
There are two methods:
First method and the second method
Lyapunov first or indirect method: it a linearization
technique. Start with a nonlinear system
x(k + 1) = f(x(k))
Expanding in Taylor series around equilibrium (xe) and
neglecting higher order terms.
Δx(k + 1) = GΔx(k) (linearized)
where f
G
11
x xe
 Continued
Then the nonlinear system is asymptotically stable
around xe if and only if the linear system is; i.e., if all
eigenvalues of G are inside the unit circle.
The above method is very popular because it is easy to
apply and it works well for most systems, all we
need to do is to be able to evaluate partial
derivatives.
One disadvantage of the method is that if some
eigenvalues of G are on the unit circle and the rest
are inside the unit circle, then we cannot draw any
conclusions, the equilibrium can be either stable or
unstable.
12
 Continued
Lyapunov second or direct method: this is a
generalization of Lagrange’s concept of stability of
minimum potential energy.
Consider the nonlinear system above. Without loss of
generality, assumes the origin as the equilibrium
point of the system.
Suppose that there exists a function, called ‘Lyapunov
function’, V(x) with the following properties:
1. V(0) = 0
2. V(x) > 0, for x ≠ 0
3. V(x) < 0 along trajectories of the nonlinear system.
Then, origin is asymptotically stable.
13
 Continued
The method hinges on the existence of a Lyapunov
function, which is an energy-like function, zero at
equilibrium, positive definite everywhere else, and
continuously decreasing as we approach the
equilibrium.
The method is very powerful and it has several
advantages:
• answers questions of stability of nonlinear systems
• can easily handle time varying systems
• determines asymptotic stability as well as normal
stability
• determines the region of asymptotic stability or the
14 domain of attraction of an equilibrium
 Continued
Lyapunov Matrix Equation: It is also possible to find a
Lyapunov function for a linear system.
For a linear system of the form x(k + 1) = Gx(k) we
choose as Lyapunov function the quadratic form
V(x(k)) = xT(k)Px(k)
where P is a symmetric positive definite matrix. Thus
V(x(k)) = V(x(k+1))−V(x(k))=x(k+1)TPx(k+1)−xT(k)Px(k)
Simplifying the above equation and omitting k
V(x) = (Gx)TPGx − xTPx
= xTGTPGx − xTPx= xT(GTPG − P)x
= −xTQx
Where GTPG − P= −Q
If the matrix Q is positive definite, then the system is
15
asymptotically stable.
 Continued
Therefore, putting Q = I, the identity matrix and solve
GTPG − P= −I
for P and see if P is positive definite.
The equation GTPG−P= −Q is called Lyapunov matrix
equation for discrete time systems and can be solved
through MATLAB by using the command dlyap.
Example 2: Determine the stability of the following
system by solving Lyapunov matrix equation.
 1 1 
x(k  1)    x(k )
 1 1
Solution: Let us take
1 0   P11 P12 
Q and P  
16

0 1   P21 P22 
 Continued
Putting these into Lyapunov matrix equation, GT PG  P   I
 1 1   P11 P12   1 1   P11
T
P12  1 0 
 1 1  P        
   21 P22   1 1  P21 P22   0 1 
 1 1   P11  P12 P11  P12   P11 P12  1 0 
 1 1   P  P P  P    P P    0 1 
   12 22 12 22   21 22   
 2 P12  P22  P11  P22  P12  1 0 
 P  P  P    
 11 22 12 P11  2 P12   0 1 
Thus 2p12 + p22 = −1; −p11 + p22 − p12 = 0; p11 − 2p22 = −1
Solving: p11 = −1; p12 = p21 = 0; p22 = −1
which shows P is a negative definite matrix. Hence the system
is unstable. To verify the result, if you compute the eigenvalues
17
of G you would find out that they are outside the unit circle.
 Continued
Asymptotically stable: All the eigenvalues of the system
have in the unit circle.
1) If the system is asymptotically stable for all the
initial states
2) The equilibrium point is said to be asymptotically or
exponentially stable in the large.
3) It is also called globally asymptotically stable.
3
2
1 Curve 1: asymptotically stable
0
x0 Curve 2: marginally stable
Curve 3: unstable
18
19