Principles of Vibration Control

Prof. Bishakh Bhattacharya

Lecture-19
Controllability and Observability of System

Welcome to the course on principles of vibration control. We have reached the fag ends of
our journey on vibration control in this course. Now in the remaining lectures I want to focus
about the active vibration control.
(Refer Slide Time: 00:33)

So towards that direction we are walking on states space control and for state space control
we have to first of all we have talked about that how to develop the state space model but
today we want to talk about certain canonical forms which will allow us to develop in the
modern control that means full state feedback control for vibration control itself.

Now the first thing we have to keep in our mind and I will actually proof you that after a few
more slides that state space representation state space representation is not unique in nature.
Remember I told you that depending on the choice of the states the state space matrix will
change. But that does not mean that the governing equation will change and the dynamics
will change.

We will prove it because one of the most important till criteria towards that is the Eigen
values of the system will not change. So no matter whatever will be your choice of state
factors we have to finally cross check that. Now since let us first of all assume that this state
space representation is not unique in nature. So we are looking for certain forms which we
are calling to the canonical forms. canonical in good old days has to has a relationship with
cannon itself anything which will be said by charge with a cannon sound would be
considered to be the last word in that direction.

But as far as our things are concerned this are certain very uniform forms that is used all over
the world and hence this canonical form term came and this we will learn corresponding to 3
cases one is for controller, another is for observer and the other one is called Jordon canonical
form. There were few more but these 3 are generally most useful. So we will first talk about
the controller canonical form.

Now if we consider the wording equation of the system in a big generalized manner, you can
see that this is the Jordon equation of the system which is nth order ordinary differential
equation and this is Y is the dependent variable and time is the independent variable.

n n−1 n n −1
d y d y d u d u
n
+α 1 n−1 + …+α n y=β 0 n + β 1 n−1 + …+ β n u
dt dt dt dt

So right hand side also i get up to in nth order or you know up to that kind of a I mean you
need not go up to n here, you can go up to something like n bar or there was something like
that were n and n bar need not be the same. So that is what is you know is our Jordon
equation of the system where left hand side talks about the states, right hand side talks about
the control effort.

Now this see this is having an nth order let say ODE of in terms of y so as a function of t. So
this equation can be actually decomposed into n first order equations. N first order equation
and that n first order equation is going to give us the state space format of it. Now in that state
space format the controller can canonical form should be search that all these coefficients are
only coming at the last row of the a matrix.
So this is what if we develop you know these nth order ODE by decomposing it fast in a nth
fast order and fast order equation and then writing it in this abcd form, then we are going to
get the controller canonical form of the system. So this form will be useful for that is why we
have to always keep this form in our mind.
(Refer Slide Time: 06:40)

There is another which is known as the observer canonical form. So this is just the dual of the
controller canonical form. And that you can see that why this is the dual part of it you can
easily see each if you look at the you know structure of the A matrix here instead of the you
know last row now it is the fast column where all these coefficients are appearing starting
from - alpha 1 to - alpha n.
And then if you can see that once again the super diagonal terms are unity here and no other
terms are coming into picture. B matrix is not 0 0 1 now, B matrix has all these based on the
coefficients drawn from this given equation. C matrix is 1 0 0 0 that is why that C matrix is
suppose to be corresponding to the output equation y = CX+Du where Y is the output and
that is why this is we put more focus on this that this is related to what you call the you know
the sensing, ok so this is the sensing.

Like B is the actuation, this is the sensing ok, influence matrix or ofcourse D is disturbance.
So here that sensing part because observer are related to sensor wherever we cannot actually
have sensors to sense the particular state we try to estimate that state and then this observer
canonical form becomes useful because we have to design, we can design observer for this
estimated state.

So that is why this part is simplified here as C is 1 and rest is 0 and D equals to beta0. So this
is known as the observer canonical form.
(Refer Slide Time: 09:05)

There is one more form that comes out this is known as Jordon canonical form and this is the
form which is for non repeated Eigen values. So once again we have the same equation front
of us and you can actually carry out laplace transform this time of the entire equation, so Y
will be Ys, U will be Us and you can write each one of them through a partial fraction. So
these are the partial fractions, so that you actually separate it out in terms of lamda n distinct
roots ok.

So that is what then becomes our characteristic equation. So now if I line up all these Eigen
values which are the distinct roots along the diagonal rest is all 0. Then that is what is by A
matrix in the Jordon canonical form. B matrix is all unity and C matrix has all the residual
corresponding to each one of these roots all the residuals and D is beta0. so this is the Jordon
canonical form.

So these 3 canonical forms are important for controller design that is why we have first
introduced theses 3 forms to you. Now this form to form this is only about it is like you know
the same person who will you are putting into different dresses. All dresses corresponding to
your convenience, but the person remains the same, how do I prove that.

This is giving me the same characteristic equation, so what it means is that the transformation
matrix has only changes the form, it has not change the Eigen values of the system. So the
Eigen values are actually in variant in this case. That is the very important conclusion that
help us that from fall to fall very smoothly we can jump provided this transformation matrix
exists. That is the only precondition that we have to keep with us.
(Refer Slide Time: 15:52)

No with that we need to see a few things. For example how do we check the stability of a
system. Well in state space form the stability of a system will always depend on the Eigen
values of A which may be obtained from the characteristic equation. If you remember that
when we are talking about these transfer function based approach then the roots I was always
telling that the real parts of the roots should not go, should not become positive.

It should not go to the right hand side, that means if I plot the S plane if you remember a real
part of S verses the imaginary part of S. I said that no poles can be placed in the right hand
side, no poles because then it will be unstable. Now in this state space form this pole is to be
replaced by the Eigen values, otherwise it remains the same. That means the real parts of the
roots are Eigen values are strictly negative.

Then the system will be considered to be asymptotically stable. Let us consider you know a
kind of system like say this A, this is a system if you look at it that this system looks like as it
in the control canonical form, but we cannot guarantee because we have not seen the B yet.
Only when we see the B also then only we can say yes it is in control canonical form.

And if I do find out the determinant I am going to get this polynomial, which is a cubic
polynomial of lamda. Now by using MATLAB or by using other formulations you can
actually or Newton-Raphson options you can find out what is lamda and you will see that this
are the 3 roots of lamda and in that means that this the stable system because the lamdas are
all in the left half plane. So-1, -2 and -3. So that means this is going to be a stable system.
So that is the way we can first of all check the stability of a system. Next what we can check
is actually what is this are called you know the control canonical form of the system.
(Refer Slide Time: 18:50)
(Refer Slide Time: 18:55)

Next we want to talk about the controllability and observability of the system. Because
particularly we will talk about the controllability and also in observability will check as a side
by side with respect to the observer canonical form. Now let us look in to this forms. So how
do you define the controllability of a system. The controllability the term itself is telling us
that if we have a system were I have some actuator through which I am trying to control the
vibration.

The vibration should be this actuator should be placed in such a manner that it can control the
vibration. For example that we give something like a physical significance of what we are
talking about, suppose this is a cantilever beam and this beam is vibrating. Now if you want
to you know control this vibration of the cantilever beam you have to place actuators. So if
you place an actuator here you will be able to control.

If you place it somewhere here the effort will be increasing what still you will be able to
control. If you place it very close to the structure you may be need much more force and
finally if you were here on the nodal point here itself it will be impossible for you to control.
So what it means is that based on my position of actuator the controllability of the vibration
mode will be actually you know determine that whether it is controllable or not.

Now for you know a particular state x1 of a system will call it controllable when if all initial
conditions X0 at any previous time t, t can be 0 or any previous time can be transfer to a
future x1 in a finite time by some control function u(t,x0). And this also has to be a finite
control effort. So that means suppose we are in a 3 dimensional thing let us say so I have you
know a position here and this is what let say my x0 position with respect to 3 steps suppose
ok.

X1 and x2 and x3. Now from here I am going to some location here in this states space which
is some x(t). If I can go from these state to this state within a finite time t and with a finite
control effort u then only we will say that the systems controllable and this is simply not
possible in this case that I have discuss that within a finite time if you place your actuator
here you will not be able to control the vibration of the system.

So if all these states like that are controllable then the system is completely controllable. If
controllability is restricted to depend up on t0 then it is called controllable at time t0 that is
within this time it is controllable and if it is less it is not if it is more, it will be. If a particular
output can be obtain from any arbitrary x0 at t0 then the system is called output controllable.
So these are some of the subdivisions of controllability of a system. So that is the definition
of controllability.
(Refer Slide Time: 22:33)
So by checking these particular condition if this is satisfied we can say yes it is controllable.
For designing before designing any controller in state space control is controllability must be
check always.

So that is the first condition. If you get it check then the transformation the inverse of the
transformation will actually exist.
(Refer Slide Time: 24:04)

[ ]
x́= 3 0 x +[ 1 ]u
0 −2 0
(Refer Slide Time: 25:01)

Now I told you that observer canonical form gives us that is the dual problem of it that is
from the same thing prospective. That a particular state x1(t) at some given time is
observable, if the knowledge of u(t) and output y(t) over a finite segment of time can
completely determines the states x1t. So this needs this for controllability we needed the A, B
for controllability if you remember.

That is just an example to actually keep with controllability and observability. Thank you. In
the next lecture we will talk about how to apply all these concepts together for full state
feedback control.