Linear Control system

MIMO State Feedback Design

Lecture 9
Dr. Mostafa Abdel-Geliel
 Topics
• Pole placement in SISO model using state feedback
– Direct substitution
– Transformation using Controllable form
– Ackerman Formula
• Pole placement in case of uncontrollable state
• Pole Placement for MIMO model
– Difference between SISO and MIMO design
– Limitation
– Pole placement using Eign structure technique
• Pole placement using output-feedback
 State Feedback
Open loop system
x& (t ) = A x(t ) + Bu(t )
y (t ) = Cx(t ) + Du(t )
After state feed back u(t)=r(t)-Kx(t)

r(t)

x& (t ) = Af x(t ) + Br (t ) Af = A − BK
Closed loop system
y (t ) = C f x(t ) + Dr (t ) C f = (C − DK )
 The problem here is to Find the feedback vector K so that
the closed loop eign values (poles) will be place at
λ1 , λ 2 , L , λ n
The selection of the closed loop eign values is depend on
either
1- the system stability i.e. to stabilize the unstable system
2- transient performance; the dominant closed-loop poles
3- the quadratic optimal control approach
t
Br (τ )dτ
A f ( t −τ )
x(t ) = e x(0) + ∫ e
Af t
The Solution
0

Improving the steady state error is improved using either

1- forward gain u(t)=kd r(t)-kf x(t)
2- integral part (will discussed later)
 Determination of Feedback Matrix K

1- Using Direct Substitution Method

2- Using Transformation Matrix T.

3- Using Ackermann’s Formula

٥
 Determination of Matrix K Using Direct
Substitution Method

If the system is of low order (n 3), direct substitution of matrix K into the
desired characteristic polynomial may be simpler.

For example, if n=3, then write the state feedback gain matrix K as

Substitute this K matrix into the desired characteristic polynomial

And equate it to

٦
Since both sides of this characteristic equation are polynomials in s, by
equating the coefficients of the like powers of s on both sides, it is possible to
determine the values of k1, k2 , and k3 .

This approach is convenient if n=2 or 3. (For n=4, 5, 6, p , this approach may

become very tedious.)

Note that if the system is not completely controllable, matrix K cannot be

determined. (No solution exists.)

٧
 Consider the regulator system shown in Figure

The system uses the state feedback control u=–Kx. Let us choose the desired
closed-loop poles at

Determine the state feedback gain matrix K.

٨
First, we need to check the controllability matrix of the system. Since the
controllability matrix M is given by

we find that |M|=–1, and therefore, rank M=3. Thus, the system is completely
state controllable and arbitrary pole placement is possible.

By defining the desired state feedback gain matrix K as

٩
and equating with the desired characteristic equation, we obtain

١٠
١١
Using Transformation Matrix T.
If the SS model of a system

in the form of controllable form

Discuss the relation between new and old transfer function

x = Tx c
Ac = T −1 AT ; Bc = T −1 B
Cc = CT ; K = K cT −1
the transformation between any form and controllable cannonical
form can be obtained as follow
let U = [b Ab L An −1b] is the controlability matrix of given
system and
U = [b A b L A n −1b ]is the controlability matrix of given
system in controllable form. then
the charactristic equation in the form
s n + an −1s n −1 + L + a0
t1 = B
t2 = At1 + an −1t1 = AB + an −1 B
t3 = At2 + an −2t1 = A2 B + an −1 AB + an − 2 B
M
tn = Atn −1 + a1t1 = An −1 B + an −1 An − 2 B + L + a2 AB + a1 B
T = [tn tn −1 L t1 ]
K = [α 0 − a0 α1 − a1 L α n−1 − an −1 ]
K = K T −1
where
ai are the cofficient of the given charactristic equation
s n + an −1s n−1 + L + a0
α i are the cofficient of the required charactristic equation
s n + α n −1s n−1 + L + α 0
example : design a state feedback for the given system where
the new eign values will be placed at
- 1, - 2 and - 3 which means the new char. eq
∆ c ( s ) = s 3 + 6 s 2 + 11s + 6
1 6 − 3 1
x& =  − 1 − 1 1  x + 1u
  
− 2 2 0  1
y = [0 0 1]x
solution
K = [α 0 − a0 α1 − a1 L α n−1 − an−1 ]
∆( s ) = sI − A = s − 3s + 2;
3
= [4 14 6]
λA = {1,1,−2}
 1   4 
t1 = B = 1  ; t 2 = AB + a 2 B =  − 1  ;
   
1   0 
−5
t 3 = A 2 B + a 2 AB + a 1 B =  − 6 
 
 − 13 
−5 4 1
T = −6 − 1 1
 
 − 13 0 1 
K = K T −1 1 4 − 2 
U = [b Ab A2b ] = 1 − 1 − 3 
−1  
−5 4 1 1 0 − 10 

= [4 14 6 ] − 6 −1 1 0 0 1 
  U = 0 1 0  ;
 
 − 13 0 1  1 0 3
 Using Ackermann’s Formula

α (A)
−1
K = [0 0 K 1] Q c

where
Qc = controlability matrix
α (.) = new charactristic equation
regarding to the previous example
K = [0 0 1]Qc−1 [A 3 + 6 A 2 + 11 A + 6 I ]
−1
1 4 −2
Qc−1 = 1 − 1 − 3 
 
1 0 − 10 
K = [5 6 − 5]
 Uncontrollable System
The assignment of all eignvalues by using state feedback is possible if and only if
the system is controllable.
If the system has some uncontrollable state, it is possible to find a similarity
transformation to decompose the system into controllable and noncontrollable
parts. That means A and b matrix will be transformed into

 A11 A12  b1 

where A =  0  , b =  , c = [c 11 c 12 ]
 A22  0
A = Q AQ
−1

b = Q b; c = cQ
−1

A11 , b1 is the controllable

then the eigns of A11 can be orbitrary assigned
A22is the uncontrollable part
 hence
∆( s) = c( sI − A) −1 b = C11 ( sI − A11 ) −1 b1

Q = [q1 L q p v1 L vn − p ]
is a nonsingular matrix where q1 to q p represent the
controlable state they are obtained from controlability matrix
U = [b Ab A 2 B L An−1b]

Select the linear independent columns (qi) in U with number equal rank(U)
And choose the vi where it is independent and Q matrix has inversion
Example
− 5 − 10 10  1 4
1 0 0
A= 2 − 1 − 2; B = 1 0 ; C = 
     0 1 0
 0 − 4 1  1 2
1 4 − 5 0 5 − 20
U = 1 0 − 1 4 − 3 8 ; rank (U ) = 2
 
1 2 − 3 2 1 − 14 
Q = [u1 u2 v1 ]; let v1 = [1 0 0]T
1 4 1 0 1 0
Q = 1 0 0, and Q −1 = 0 − 5.5 0.5
   
1 2 0 1 1 − 2
1 4 2
At = Q _ 1 AQ = − 1 − 1 − 1,
 
 0 0 − 3
1 0
1 4 1
Bt = Q B = 0 1, Ct = CQ = 
_1
  1 0 0
0 0
Design a state feedback so that closed loop pole can be place at -4, -5 and -6
− 5 − 10 10   4
A= 2 − 1 − 2 ; B = 0; C = [1 0 0]

eign( A) = {− 1 ± j 2,−3} stable
   
 0 − 4 1  2
but it is uncontrollable
4 0 − 20
U = 0 4 − 8 ; rank (U ) = 2 it is seen that the 3rd state is uncontrollable
 
2 2 − 14  so it can not move
Q = [u1 u2 v1 ]; let v1 = [1 1 0]T eign( Af 11 ) = {− 4,−5}
4 0 1
Q = 0 4 1, and
using Akerman formula for A11 , then
 
2 2 0 K t1 = [7 1]
0 − 5 − 3 K t = [ 7 1 0]
At = Q _1 AQ = 1 − 2 1 ,
  K = K t Q −1 = [0.75 − 0.75 2]
0 0 − 3
1 
the new eign value
Bt = Q B = 0, Ct = CQ = [4 0 0]
_1
 
eign( Af ) = eign( A − BK ) = {− 4,−5,−3}
0
0 − 5  1 
A11 =   , B = 0, C1 = [4 0]
1 − 2
1
 
MIMO case
In case of MIMO the feedback matrix K has unknown
parameters more than the system poles.
Each input contains n feedback gain so
# of gains=n*m; m=number of inputs

It this case the system may be

1- completely controllable
2- incompletely controllable: use controllability decomposition
 1- Completely controllable case
There are two ways to design the feedback
1- arbitrary using the same technique of SISO
2- eignstructure Assignment

Eignstructure Assignment
It depends on the eign value and ( A − BK ) v = λ v i i i

eign vector concept it can be rearanged

The system after feedback should [λ I − A B] v  = 0
i
i

satisfy  Kv  i

 v  v 
pi =  i  =  i 
 Kv i  q i 
then p i is in the NULL space of [λi I − A B ]
then
KV = Q
K = QV −1
Example
0 0 1 1 0
x& = − 5 1 − 3 x + 0 0u; y = [0 0 1]x
   
 0 1 0  0 1 
system eign value λA = {− 1,1 ± j 2}
the required eign values are
λA = {− 2,−3,−4}
f

Then

Ti = [λi I − A B ] has three matrix for each eign value

T1 = [− 2 I − A B ] ; p1 = α1,1n1,1 + α1, 2 n1, 2
T2 = [− 3I − A B ] ; p 2 = α 2 ,1n2 ,1 + α 2 , 2 n2 , 2
T3 = [− 4 I − A B ] ; p 3 = α 3,1n3,1 + α 3, 2 n3, 2
V 
P= =
Q 
5.5 − 1.5 2.5  This can be done using command
K = QV =  −1
 Place in matlab
 2. 5 − 1 . 5 4. 5 
 Application 1
The Magnetic-tape-Drive system is a MIMO system shown in Fig. There is an
independently controllable drive motor on each end of the tape; therefore, it is
possible to control the tape position over the read head, x3, as well as the tension in
the tape. The tape is modeled to be a linear spring with small amount of viscous
damping. The goal of the control system is to enable commanding the tape to specific
position over the read head while maintaining a specified tension in the tape at all
times. The desired specifications are that the tape position must be adjusted if the
tape head is moved 1mm with 1% settling time of 2.50 sec and overshoot less than
20%. The tape tension, Te, should be controlled to 2 N with constraint that 0 < Te < 4.
The current is limited to 1A at each drive motor.
b
x3 x2
x1
It is required to design a state feedback
controller to achieve the required
performance
K

i1 i2

Figure C1: Schematic diagram of magnetic tape drive

 The system model is given by
 0 0 - 10 0  0 0 
Jθ&&1 = −Te r + K m i1 ,  0 0 0 10  0 0 
x& =  x+  u
Jθ&& = −T r + K i ,
2 e m 2
3.315 - 3.315 - 0.5882 - 0.5882
 
8.533 0 
 
Te = k ( x 2 − x1 ) + b( x& 1 − x& 2 ),
3.315 - 3.315 - 0.5882 - 0.5882 0 8.533

x 3 = ( x 2 + x1 ) / 2, 0.5 0.5 0 0 
y= x
- 2.113 2.113 0.372 0.375

where i1 and i2 are the current into drive motors 1, 2, respectively,

Te tension in tape (N),
θ1 , θ2 angular position of motor, r redias
x1and x2 position of tape over read head (mm),
J=0.006375 kg.m2, motor and capstan inertia,
r =0.1 m,
Km=0.544 N.m/A, motor torque constant,
k=2.113 N/m, tape spring constant
b=0.375 N sec/m, tape damping constant.
the state vector x=[x1 x2 ω1 ω2]T, input vector u=[i1 i2]T and
the output vector y=[x3 Te]T is
 Application 2

Tank cross section area (At) 0.0154 m2

Pipe cross section area (Ap) 5*10-5 m2
Outflow coefficient (µ mn) µ 13=µ 32=0.5 , µ 20=0.675
Maximum flow rate (Qmax) 1.2*10-4 m3/s
Maximum level (Lmax) 0.62 m Deduce the linearized state space model
Q1=0.35*10-4 m3/s
Operating point Q2=0.375*10-4 m3/s Then design state feedback that adjust
L1=0.4 m , L2=0.2 m , L3=0.3 m
The input flow q1 and q in order to make
2

= q1 − cv13 δg (L1 − L3 )
dL1 L1=0.4 m
A1
dt L2=0.2 m
dL
A3 3 = cv13 δg (L1 − L3 ) − cv 32 δg (L3 − L2 ) L3=0.3 m
dt
dL q1o= 0.35*10-4 m3/s
A2 2 = q2 + cv 32 δg (L3 − L2 ) − cv 20 δgL2
dt q20=0.375*10-4 m3/s
where L1 ≥ L3 ≥ L2
1- Output feedback design
2- Observer design