Download as ppt, pdf, or txt
Download as ppt, pdf, or txt
You are on page 1of 73

MODERN CONTROL SYSTEMS

STATE-SPACE CONCEPTS
In the modeling of systems in the classical
sense, we assume a cause and effect
relationship only
 DRAWBACKS OF CLASSICAL CONCEPT OF
TRANSFER FUNCTION
 Defined under zero initial conditions

 Applicable to LTI systems

 Restricted to SISO systems

 Provides no info on internal state

 Trial-and-error procedures
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 IMPLICATIONS OF DRAWBACKS
 Complex for MIMO systems

 May not yield an optimum control


system wrt to performance
 Lack of info on feedback and internal
states may affect stability and
performance
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
In order to overcome these limitations,
especially when dealing with Time-
Varying Systems:
“SYSTEMS IN WHICH ONE OR MORE
OF THE PARAMETERS VARY WITH
TIME”
The idea of the state-variable was born.
NOTE: This was very critical in the
attempt to go to space
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 STATE: “A set of numbers x1 , ..., xn  such
that the knowledge of these numbers and
the input functions u1 , ...,
will,un with the
equations describing the dynamics,
provide the future state and output
 y1 ,system”
of the ..., yn 
u1 y1
STATE
un SYSTEM yn

x1 xn
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 In deriving a mathematical model for a physical
system one usually begins with a system of
differential equations.
 We will call this system of differential
equations a state space representation.
 The solution to this system is a vector that
depends on time and which contains enough
information to completely determine the
trajectory of the dynamical system.
 This vector is referred to as the state of the
system, and the components of this vector are
called the state variables
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Advantages of using the state space
representation compared with the classical method
  
 Ability to handle MIMO systems;
 State system model includes the internal state
variables as well as the output variable;
 State model provides a time-domain solution;
 The form of the solution is the same as for a
single 1st-order differential equation;
 Effect of initial conditions can be easily
incorporated in the solution;
 The matrix/vector modeling is very efficient from
a computational standpoint for computer
implementation
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 State space model: a representation of
the dynamics of an nth order system as a
first order differential equation in an n-
vector, which is called the state.


x  Ax  Bu
State Equation


y  Cx  Du
Output Equation

 A=State Matrix (n x n)
 B= Input Matrix (n x m)
 C= Output Matrix (p x n)
 D= Direct Transition Matrix (p x m)
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Find the state model of the ff D.E:

5 4 3 2
d y d y d y d y dy
4 5  6 4  3 3  2 2   3 y  4u
dt dt dt dt dt
 NOTE: The number of state variables is
equal to the highest order of the
differential
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 DERIVATION OF TRANSFER FUNCTION
FROM STATE MODEL
 Recall:
x  Ax  Bu
y  Cx  Du
 Assuming zero initial conditions and laplace
transforming:

sX ( s )  AX ( s )  BU ( s ) (1)

 Y ( s )  CX ( s )  DU ( s ) (2)
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 From (1):
X s sI  A  BU s 
 X s   sI  A BU s 
1
 (3)
 Putting (3) into (2)
Y s   C sI  A BU s   DU s 
1

 Y s   C sI  A B  D U s 
1

Y s 
 T s    C sI  A B  D
1

U s 
In matrix a lg ebra form
adjsI  AB
T s   C D
detsI  A
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The adjoint of a matrix is found by replacing
each element by its co-factor and transposing
(interchanging rows and columns)
1 T
 a11 a12 a13   c11  c12 c13 
a a a  
  c21 c22  c23 
 21 22 23   
 a31 a32 a33   c31  c32 c33 
 c11  c21 c31 
  c12 c22  c32 
 
 c13  c23 c33 
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 ss2tf

 Syntax: [num,den] = ss2tf(A,B,C,D,iu)


 Description: ss2tf converts a state-space
model of a given system to an equivalent
transfer function representation. It returns
the transfer function
Y s 
T s    C sI  A B  D
1

U s 
 from the iuth input.
 NOTE: iu must be specified for systems with
more than one input.
:•.

MODERN CONTROL SYSTEMS


STATE-SPACE CONCEPTS
 Vector den contains the coefficients of the
denominator in descending powers of s.
 The numerator coefficients are returned in array
num with as many rows as there are outputs y.
 Ex: Given the state model of a system as follows

 :

X   0 1  1 1
  x   u
 25  4 0 1
 Determine the transfer function of the system
1 0 
y x
0 1 

 Determine
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 tf2ss: Transfer function to state-space
conversion.
 [A,B,C,D] = tf2ss(num,den)

calculates the state-space representation:


x  Ax  Bu
y  Cx  Du
of the system:
nums 
 T s  
dens 
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 TESTS FOR CONTROLLABILITY AND
OBSERVABILITY
 The concept of controllability and observability
in the design of controllers for dynamic
systems mainly involves verifying from the
outset whether a solution exist to the control
problem.
 If a solution does exist, it gives the designer
the privilege of placing the poles anywhere
along the s-plane that will give the desired
performance.
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Two fundamental questions need to be answered
in determining whether a control solution exists
or not:
 “Can we transfer the system from any initial
state (x(to)) to any other desired state (x(t)) in
finite time by application of a suitable control
force?”
 “Knowing the output vector (y(t)) for a finite

length of time, can we determine the initial


state of the system (x(to))?”
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The answers to these questions were provided by
Kalman:
 “A system is completely state controllable if it is
possible to transfer the system from any initial
state (x(to)) to any other desired state (x(t)) in
finite time (t-to › 0) by application of a suitable
control force u(t)”
 A system is completely state observable if the
initial state (x(to)) can be completely identified
by measurements of the output y(t) over a finite
time interval”
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Allphysical systems are practically controllable
and observable whereas mathematical models
may not be.
 Any such mathematical model in the long run
may not be an accurate representation of the
system and as such may not yield an optimal
control solution.
 Any system that is controllable is also
stabilizable
 Any system that is observable is also
detectable
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 We will consider two approaches to determining
controllability and observability: KALMAN and
GILBERT tests
 The Gilbert’s test provides a better physical
insight into the problem
 The Gilbert’s test requires that the system be
transformed into a canonical form while the
Kalman’s test can be applied to any system
 The Gilbert’s test considers an n-th order SI
LTI while an n-th order MI or MIMO LTI is
considered by the Kalman’s test
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 KALMAN MATRIX TEST
 A system is completely state controllable if:

Mc  B AB ... An 1 B 
 is of FULL RANK i.e. it has n-linearly
independent columns
 A system is completely state observable if:

  C 
 CA 
Mo   
 ... 
CA n 1 
 
 is of FULL RANK i.e. it has n-linearly
independent rows
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 clc;
 j=0.0106;
 fo=0.00286;
 km=1.168;
 kb=1.168;
 Ra=13.3;
 La=3;
 A=[-fo/j km/j;-kb/La -Ra/La];
 B=[0;1/La];
 C=[1 0];
 D=0;
 [num,den]=ss2tf(A,B,C,D)
 Mc=[B A*B]
 ans=det(Mc)
 if (det(Mc)~=0)
 'The System is Completely Controllable'
 else
 'The System is NOT Completely Controllable'
 end
 Mo=[C; C*A]
 ans=det(Mo)
 if (det(Mo)~=0)
 'The System is Completely Observable'
 else
 'The System is NOT Completely Observable'
 end
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The Gilbert’s Test gives a better physical
insight into SI LTI systems and involves
transforming the state model into a canonical
form.
 It involves the following steps:

 Given the state model, determine and solve the


characteristic equation
A  I   0
 Determine the eigenvalues (i.e. roots) of the
characteristic equation
1 , 2 , 3
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Formthe Modal Matrix M using the Vander
Monde type (for example for n=3)
1 1 1 
M   1 2 3 
 2 2 2

1 2 3 
 Determine the vector Matrix
~ 1
BM B
 If
~
no element of B is zero, the system is said to
be completely state controllable
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 If there is no path from any input to one of the
decoupled differential equations, the
corresponding mode is termed UNCONTROLLABLE
 It means that portion of the system is not
affected by any input
 If it has no uncontrollable modes, it is said to be
‘completely’ CONTROLLABLE
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Determine the output vector matrix
y  cMz
 If no element of y is zero, the system is said
to be completely state observable
 Determine the state model in canonical form

~
z  z  B u
y  cMz
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 If one of the state variables of the decoupled
differential equations is not added to any of the
system’s output, the corresponding mode is
termed UNOBSERVABLE
 The response of that system is not visible from
any input
 A system with no unobservable modes is said to
be ‘completely’ OBSERVABLE
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Determine whether the given system is
completely state controllable and/or state
observable using the Gilbert’s Test. Hence
determine the state equation in canonical form
 1 0 1  1 
X   0 1 1  x  0u
   
 0 0  2 1
Y   1 1 2x
 We can also use the CLI of Matlab to solve this
problem
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 All control systems exhibit transient behaviour
(causing the output to deviate from the input
instantaneously) to certain degree before a
steady state is reached.
 This is because inertia, mass, inductances which
are all energy storage devices are inherent in
physical systems.
 Transient response is defined as that part of the
system response that goes to zero as time goes
to infinity while the steady state response
indicates where the system output ends up as
time approaches infinity.
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The response of a second order system
describing the transient and steady state
parameters: Peak Overshot, Settling Time,
Steady State Error, etc is as shown
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 RESPONSE OF THE DC MOTOR TO A STEP SIGNAL
A step signal is one of the standard test signal
used to carry out mathematical and graphical
analysis of a control system.
 Used to investigate the behaviour of a dynamic
system when the control problem under
investigation is a function of sudden
disturbance.
 The values obtained from the response of the
system to a step signal are then compared with
the design/performance specifications
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The following Matlab program is used:
 sys = ss(A, B, C, D)
 t = 0:0.01:3
 step(sys, t), title('Open loop Response')
 NOTE: The stepinfo command can also be used to obtain Open loop Response

the shown values 1.4


System: sys
Peak amplitude: 1.09
1.2 Overshoot (%): 30.4
At time (sec): 0.51

1 System: sys
Settling Time (sec): 1.65

0.8
Am plitude

System: sys
System: sys Final Value: 0.833

0.6 Rise Time (sec): 0.21

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3
Time (sec)
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 STATE TRANSITION MATRIX
 (Solution of the State Equation)

 Consider the homogeneous linear (unforced)


  
system given by: x  Ax t ; x 0  x
0

 Itcan be shown that the solution is of the


form:
 1 22 1 i i
xt    I  At  A t  ...  A t  x0
 2! i! 
 Each term in the bracket is an n x n matrix
 The entity in the bracket is a matrix
exponential defined as eAt
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The initial state x0 at t=0 is driven to a state

x(t) at a time t by the application of the matrix


exponential eAt
 The state transition matrix e   t 
At

 There are several methods that can be used to


solve the state equations
 Classical Method

 Laplace Transformation Method

 Canonical Transformation Method

 Cayley-Hamilton Theorem

 The Classical Method will not be considered as it


is limited in application to only simple cases
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 LAPLACE TRANSFORMATION METHOD
 Consider: x  Ax
 Laplace transforming:
sX s   x0   AX s 
sI  AX s   x0
X s   sI  A x0
1

 In order to find x(t), we take the inverse


Laplace transform:
xt   L sI  A x0
1 1

 x(t) is the unforced response of the system


MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
t   e  L sI  A  L1s 
At 1 1

s   sI  A is the Resolvant matrix


1
 Where

 FOR A FORCED SYSTEM: x   Ax  Bu


 Laplace transforming:
sX s   x0  AX s   BU s 
sI  AX s   x0  BU s 
X s   sI  A x0  sI  A BU s 
1 1

 By inverse Laplace transform, we find x(t)


xt   L sI  A x0  L
1 1 1
sI  A
1

BU s 
xt   t x0  L1 s BU s 
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
A linear time-invariant system is given by the
homogeneous state equation: 1 0
x    x
1 1 
 Compute the solution of the state equation,
assuming: x  1
0 0 
 
 Consider now that the system has a forcing
function and is represented by: 1 0 0 
x    x   u
1 1 1 
 Where u is a unit impulse function

 Compute the solution of the state equation


assuming same initial conditions
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 To solve the problem in Matlab, we make use of
the following commands in the CLI:
 syms

 ilaplace

 inv

 exp
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 CANONICAL TRANSFORMATION METHOD
 It involves the following steps:

 Given the state model, determine and solve the


characteristic equation
A  I   0
 Determine the eigenvalues (i.e. roots) of the
characteristic equation
1 , 2 , 3
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Form the Modal Matrix M using the Vander
Monde type (for example for n=3)
1 1 1 
M   1 2 3 
 2 2 2

1 2 3 
1
 Determine the diagonal Matrix
  M AM
 Determine the diagonalized transition matrix
t
e 1
0 0 
t  t 
e  0 e 2
0 
t
 0 0 e 
3
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Determine the state transition matrix
At t 1
e  Me M
 Determine the state solution
xt   e x0 
At

 Ex: Consider a system with the state model


0 1 0 
x    x   u
 2  3  2
0 
x0    
1 
 Compute the state solution of the system
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 To solve the problem in Matlab, we make use of
the following commands in the CLI:
 syms

 eig

 inv

 exp
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 CAYLEY-HAMILTON THEOREM

 The theorem is more computationally convenient


for large systems compared to the other 2
methods
 “Every Square Matrix A satisfies its own
Characteristic Equation”
 If q    I  A  n  a n 1  ...  a
1 n 1  an  0
 is the characteristic equation of A, then

q A  An  a1 An1  ...  an1 A  an I  0


 The theorem provides a simple procedure for
evaluating the function of a matrix
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Given a matrix polynomial in terms of the λ as f(λ)
and the characteristic polynomial as q(λ)
 If f(λ) is divided by q(λ) then:

f   R 
 Q  
q  q 
f    Q q   R 
 Where R(λ) is the remainder polynomial of the
form: R    0  1   2   ...   n 1
2 n 1

 If we evaluate f(λ) at the eigenvalues (λi); then

q(λ)=0 and we have f    R ; i  1, 2, ..., n


i i
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The coefficients  0 , 1 , ...,  n 1 can be obtained by
successively substituting 0 , 1 , ..., n 1into equation
for q(λ)
 Substituting A for λ: f  A  Q Aq  A  R  A

 Since q(A) is identically zero, it follows that:


f  A  R  A
  0 I  1 A  ...   n1 An1
 which is the desired result
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The formal procedure of evaluation of the
matrix polynomial f(A) is given as:
 Find the eigenvalues of matrix A

 If all the eigenvalues are distinct, solve n


simultaneous equations given by f(λi) for the
coefficients  0 ,1 , ..., n 1
 If A possesses an eigenvalue λk of order m, then

only one independent equation can be obtained


by substituting λk into equation for f(λi). The
remaining (m-1) linear equations, which must be
obtained in order to solve for αi can be found
by differentiating both sides of:
f    Q q   R 
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Since
d q 
j

j     0; j  0,1, ..., m  1
d k

 It follows that:
d f  
j
d R 
j

j    j    ; j  0,1, ..., m  1
d k

d k

 The coefficients α1 obtained and equation


(f(A)=R(A)) yield the required result
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The Cayley-Hamilton theorem allows us to attack the
problems of computation of eAt , where A is a constant
n x n matrix.
 The power series for the scalar eλt :

t 1 22 1 n n
e  1  t   t  ...   t
converges for all finite2λ! and t. n!
 It follows from this that the matrix power series

converges for 1 all2 2finite t.1


At all A and for  n n
e   I  At  A t  ...  A t 
 2! n! 


MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Thematrix polynomial f(A)=eAt can be expressed
as a polynomial in A of degree (n-1) using the
technique presented

0 3
 Ex: Find f  A  A for A  
5

  2  5
0 1 
f  A  e for A  
At

  1  2 
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 STATE OBSERVER

 In the design of Control Systems, it is assumed


that all state variables are accessible for
feedback
 However, this is not the case in practice and as
such there must be mechanisms to allow us to
estimate for the unavailable state variables e.g
introducing a state observer
 Its essence is to ensure that the state variables
are accessible for measurement and control
purposes
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 There are 3 steps to ensuring accessibility of
the state variables:
 Assume that all state variables are measureable
and utilize them in a full-state feedback control
law
 Construct an observer to estimate the states
that are not directly sensed and available as
outputs
 Appropriately connect the observer to the full-
state feedback control law
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 AnOBSERVER is “a device that can be driven by
the available inputs and outputs to approximate
the state vector and whose output can then be
used to implement the feedback control law”
u  kx  r
A control system with an observer is as shown
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 If the observer ‘observes’ all the state variables
even if some are not available for direct
measurement, it is called a FULL-ORDER
OBSERVER.
 If the observer assumes that certain state
variables are already available as system outputs
and do not need to be estimated, it is called a
REDUCED-ORDER OBSERVER.
 The design criterion should be to minimize the
difference between y=Cx (system output) and the
output y  Cxˆ as constructed by the observed
vectorx̂
x  xˆ
 This is equivalent to minimizing
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 However, since x may not be accessible always,
we attempt instead to minimize y  y ˆ
 The design criterion then is: “Design Ke which is

an n x 1 vector so that y  yˆ is minimized”


xˆ  Axˆ  K e Cx  Cxˆ   Bu
xˆ   A  K eC xˆ  K eCx  Bu
 x  xˆ   A  K eC x  x 
Let ~
x  x  xˆ
~
x   A  K eC ~
x
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Key to the determination of the state observer
is the determination of an appropriate observer
gain matrix Ke
 Feedback signal through Ke serves as a
correction signal to the plant.
 If the unknowns to be accounted for are
significant, the feedback signal must be large
 If the output contains noise or is unreliable,
the feedback must be small
 Several Kes must be tested based on several

different characteristic equations and the best


one chosen
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The vector x̂
can be made to decay if the
eigenvalues of (A-KeC) are properly selected and
if the system is observable, then they can be
chosen arbitrarily.
 We will consider two approaches to determining
Ke:
1. We solve the characteristic equation:
I   A  K eC   0
and equate it to the characteristic equation
formed by the given eigenvalues in order to
determine the coefficients of Ke.
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Consider a linear system

1 2 0 1 
X   2  1 1  x   2u
   
0
 1 0 1 
 
Y  0 0 2x

 Assume that the states x1, x2 and x3 are not


accessible for feedback. An observer system is
to be designed to reconstruct x so that the
eigenvalues are at (-3, -4, -2).
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 To solve the problem in Matlab, we make use
of the following commands in the CLI:
 syms

 eye

 poly
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
2. We can also use the Ackermann’s formula
1
 C  0 
 CA  0 
   
 .  .
K e    A  .
.
   
n2
CA  0 
 CA n 1  1 
   
And
 s   s  1 s   2 ...s   n 
where 1 ,...,  n are the desired eigenvalues
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The full-order state observer is given by:

~
x   A  K eC ~
x  Bu  K e y

 Ex:

 Consider the system: x  Ax  Bu


y  Cx
where:A  0 10; B  0; C  0 1
   
2 0 1 
Assume the desired eigenvalues of the observer
matrix µ1=µ2=-10, determine the full-order state
observer
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 To solve the problem in Matlab, we make
use of the following commands in the CLI:
 syms

 eye

 transpose (‘)

 acker
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 LINEAR QUADRATIC OPTIMAL REGULATOR

 The LQR method provides a systematic way of


computing the state feedback gain matrix,Ke.
 Once we can obtain an optimal solution, the
system can be STABLE irrespective of the
stability of the open-loop system.
 NOTE: We seem more interested in stability in
the classical domain and ignore optimality of
performance
 Given: x  Ax  Bu
 We need to determine matrix Ke of the optimal

system u t    K e xt 
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 This is in order to minimize the performance

J   x Qx  u Ru dt

index: * *

0
 Where Q and R are positive definite matrices
(pdm) and account for the relative importance
of the errors
 (u*Ru) accounts for expenditure of the energy
of the system.
 q11 q12 
 NOTE: A given matrix Q 
q q 
is a pdm iff:  21 22 
2
q12
q11  0; q12  q21; q22 
q11
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 It can be shown that: 1
Ke  T* 1
T  *
BP
K e  R 1 B* P
 The Reduced-Riccati Matrix (RRM) equation is
given by: A* P  PA  PBR 1 B* P  Q  0

 The LQR methodology is as follows:


 Solve the RRM to obtain P, which must be a
pdm. This implies system stability
 Substitute P into the equation for Ke. The

resulting matrix is the optimal gain matrix


MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Consider the system shown. Assuming the
control signal to be: u t    kxt , determine the
optimal feedback gain matrix k such that the
following performance index is minimized
J   xT Qx  u 2 dt

where: Q  1 0 0 1 
; R  1; A  
0 
;B   
  
0 2  0 0  1
 u x2 x1
∫ ∫

-k
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 To solve the problem in Matlab, we make
use of the following commands in the CLI:
 lqr
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 LYAPUNOV STABILITY xt  
 Consider a system: r
x  f x 
s 
 Three concepts of
sr 
stability are consi-
dered with respect
s  xt0  
to non-linear systems

 The system is STABLE at the origin if, for


every initial state x(t0) which is close to the
origin, x(t) remains near the origin for all t
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The system is ASYMPTOTICALLY STABLE if
x(t) approaches the origin as t → ∞
 (Every motion starting in S(r) converges to the
origin as t → ∞)
 The system is ASYMPTOTICALLY STSBLE IN
THE LARGE (Global Stability) if it is
asymptotically stable for every initial state
regardless of how near or far it is from the
origin (Every motion will approach the origin)
 The primary method of testing stability of non-
linear systems is the Lyapunov’s Direct Method,
which is concerned with determining the Lyapunov
Function
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Given x  f x , f 0   0
 If there exists a scalar function V(x) that
satisfies the following properties
V x   0;x  0
V x  is positive definite scalar function
V 0   0 
V x  has continuous partial derivative wrt x
dv  dv
 0 is negative semi  definite scalar function
dt  dt
 Then the system is STABLE at the origin
 NOTE: Inability to find the required V(x) does
not imply instability but that the attempt to
establish stability has failed
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 A scalar function f(x) is Positive Definite if:
f x   0 for all x
f 0   0 iff x  0
 Then g(x) = -f(x) is said to be a Negative
Definite Function
 A matrix Q is semi-definite if any of the
det[Q]=0
 A matrix Q is negative definite (or semi-
definite) if –Q is positive definite (or semi-
definite)
 If Q is positive definite so also will be Q2 and
Q-1
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
A scalar function V(x) that satisfies the
following conditions is said to be a Lyapunov
Function
V x  is a pdf 
 defines stability for linear systems
V x  is a ndf 
V x  must be radially unboundedV x    as t   
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The Lyapunov direct method stability criterion:

 t   Axt   Bu t 
 For a linear system x

y t   Cxt   Du t 
Suppose that u=0 and there exists two pdms P>0
and Q>0, it is stated thus:
 “A linear system is asymptotically stable at the
origin iff given any symmetric, pdm Q, there
exists a symmetric pdm P which is the unique
solution of: AT P  PA  Q “
 The choice of Q can be made arbitrarily (as far
as it is a pdm) but it is most common to set
Q=I, the identity matrix.
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 The test of P’s positive definite status is
carried out using Sylvester’s theorem:
 “The necessary and sufficient conditions for a
matrix P to be positive are that all successive
principal minors of P be positive”
 P11 P12 P13 
P   P21 P22 P23 
 
 P31 P32 P33 
P11  0
 P11 P12 
P  0
 21 P22 
detP   0
MODERN CONTROL SYSTEMS
STATE-SPACE CONCEPTS
 Using the Liapunov Direct Method, determine the
stability-in-the-large of a system described by
the following: X  Ax with 1 0
A
1 1

 NOTE: The lyap command is used to solve this


problem in the Matlab CLI

You might also like