Di Erential Algebra: Roger Miranda 9 de Septiembre de 2008

Dierential algebra
Roger Miranda
1
9 de septiembre de 2008
1
R. Garrido and R. Miranda are with Departamento de Control Automatico, CINVESTAV-
IPN, 07360, Mxico DF MEXICO garrido@ctrl.cinvestav.mx
NDICE
ndice
1. Feedback linearization 1
1.1. Intuitive concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1. Feedback linearization and the canonical form . . . . . . . . . . . . 2
1.1.2. Input-state linearization . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.3. Input-Output linearization . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.4. Zero Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2. Mathematical tools 15
2.1. Lie derivatives and Lie brackets . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2. Dieomorphisms and state transformations . . . . . . . . . . . . . . . . . . 18
2.3. The Frobenius theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3. Input-State Linearization of SISO systems 22
3.1. Input-state linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2. Conditions for input-state linearization . . . . . . . . . . . . . . . . . . . . 23
3.3. How to perform input-state linearization . . . . . . . . . . . . . . . . . . . 24
3.4. Controller design based on input-state linearization . . . . . . . . . . . . . 28
4. Input-Output linearization of SISO systems 29
4.1. Generating a linear input-output relation . . . . . . . . . . . . . . . . . . . 29
4.1.1. The case of well dened relative degree . . . . . . . . . . . . . . . . 29
4.1.2. The case of undened relative degree . . . . . . . . . . . . . . . . . 30
4.2. Normal forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.3. The zero dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.4. Local asymptotic stabilization . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.5. Global asymptotic stabilization . . . . . . . . . . . . . . . . . . . . . . . . 31
4.6. Tracking control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.7. Inverse dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.8. Tracking control for non-minimum phase systems . . . . . . . . . . . . . . 31
1. Feedback linearization
The central idea of this approach is to algebraically transform a nonlinear system dy-
namics into a (fully or partially) linear one, so that the linear control techniques can be
applied. This approach diers entirely from conventional linearization in that feedback
linearization is achieved by exct state transformations and feedback, rather than by linear
approximations of the dynamics.
Feedback linearization techniques can be viewed as ways of transforming original system
models into equivalent models of a simpler form. Thus, they can be used in the develop-
ment of robust or adaptive nonlinear controllers.
1
1. FEEDBACK LINEARIZATION
Some practical problems that have been adressed with feedback linearization are: control
of helicopters, hicg performance aircraft, industrial robots, and biomedical devices. How-
ever, there are also a number of important shortcomings amd limitations associated with
feedback linearization approach.
1.1. Intuitive concepts
1.1.1. Feedback linearization and the canonical form
Feedback linearization amounts to canceling the nonlinearities in a nonlinear system so
that the closed-loop dynamics is in a linear form.
Example 1 Consider the control of the level / of uid in a tank to a specied /
d
. The
control input is the ow n in the tank, and the initial level is /
0
. The dynamic model of
the tank is
d
dt
__
h
0
(/) d/
_
= n(t) c
_
2q/ (6.1)
where (/) is the cross section of the tank and c is the cross section of the outlet pipe. If
the initial level is /
0
is quite dierent from the desired level /
d
, the control of / involves
a nonlinear regulation problem. The dynamics (6.1) can be rewritten as
(/)
_
/ = n c
_
2q/
if n is chosen as
n(t) = c
_
2q/ + (/) (6.2)
with being an equivalnt input to be specied, the resulting dynamics is linear
(/)
_
/ = c
_
2q/ + (/) c
_
2q/
_
/ =
Choosing as
= c
~
/ (6.3)
with
~
/ = /(t)d
d
being the level error, and c being strictly positive constant, the resulting
closed loop dynamics is
_
/ + c
~
/ = 0 (6.4)
which implies that
~
/(t) tends to zero as t tends to innity. Based on (6.2) and (6.3), the
actual input ow is determined by the nonlinear control law
n(t) = c
_
2q/ c
~
/(/) (6.5)
In control law (6.5), the rst part on the right hand side is used to provide the output ow
c
_
2q/, while the second part is used to raise the uid level according to the desired linear
dynamics (6.4). Similarly, if the desired level is a known time-varying function /
d
(t), the
equivalent input can be chosen as
=
_
/
d
(t) c
~
/
so as to still yield
~
/(t) 0 as t .
2
Remark 1 Feedback linearization can be simply applied to a class of nonlinear systems
described by the so called companion-form, or controllability canonical form.
A system is said to be in companion form if its dynamics is represented by
r
(n)
= , (x) + / (x) n (6.6)
where n is the scalar control input, r is the scalar output of interest, x =
_
r. _ r. .... r
(n1)
T
is the state vector, and , (x) and / (x) are nonlinear functions of the states. This form is
unique in the fact that, although derivatives of r apper in this equation, no derivative of
the input n is present. In state-space representation, (6.6) can be written as
d
dt
_
_
r
1
r
2
...
r
n
_
_
=
_
_
r
2
...
r
n
, (x) + / (x) n
_
_
and for systems that can be expressed in the controllability canonical form, using the
control input (assuming / to be non-zero)
n =
1
/
[ ,] (6.7)
we can cancel the nonlinearities and obtain the simple input-output relation (multiple
integrator form)
r
(n)
=
Thus, the control law
= /
0
r /
1
_ r ... /
n1
r
(n1)
with the /
i
chosen so that the polynomial
j
n
+ /
n1
j
n1
+ ... + /
0
has all its roots strictly in the left-half complex plane, leads to the exponentially stable
dynamics
r
(n)
+ /
n1
r
(n1)
+ ... + /
0
r = 0
which implies that r (t) 0. For tasks involving the tracking of a desired output r
d
(t),
the control law
= r
(n)
d
/
0
c /
2
_ c ... /
n1
c
(n1)
(6.8)
(where c (t) = r (t) r
d
(t) is the tracking error) leads to exponentially convergent track-
ing. Similar results would be obtained if the scalar r were replaced by a vector and the
scalar / by an invertible square matrix.
Remark 2 In (6.6) it is assumed that the dynamics is linear in n (although nonlinear in
the states). However, the method can be extended to the case when n is replaced by an
invertible function q (n) .
When the nonlinear dynamics is not in a controllability canonical form, one may have to
use algebraic transformations to rst put the dynamics into the controllability form
before using the above feedback linearization design, or to rely on partial linearization of
the original dynmics, instead of full linearization.
3
1.1.2. Input-state linearization
Consider the problem of desifning the control input n for a single-input nonlinear system
of the form
_ x = f (x. n)
The technique of int-state linearization solves this problem in two steps
1. One ndsa state transformation z = z (x) and an input transformation n = n(x. )
so that the nonlinear system dynamics is transformed into an equivalent linear-time
invariant dynamics, in the familiar form
_ z = Az + b
2. One uses standard linear techniques (such as pole placement) to design
Example 2 Consider the second-order nonlinear system
_ r
1
= 2r
1
+ cr
2
+ sin (r
1
) (6.11-a)
_ r
2
= r
2
cos (r
1
) + ncos (2r
1
) (6.11-b)
which have an equilibrium point at (0. 0). Note that the nonlinearity in the rst equation
cannot be directly canceled by the control input n. However, if we consider the new set of
state variables
.
1
= r
1
(6.12-a)
.
2
= cr
2
+ sin (r
1
) (6.12-b)
then, the new state equations are
_ .
1
= _ r
1
= 2r
1
+ cr
2
+ sin (r
1
)
= 2.
1
+ .
2
_ .
2
= c _ r
2
+ _ r
1
cos (r
1
)
= cr
2
cos (r
1
) + nc cos (2r
1
) + _ r
1
cos (r
1
)
= [.
2
sin (.
1
)] cos (.
1
) + nc cos (2.
1
) + [2.
1
+ .
2
] cos (.
1
)
= .
2
cos (.
1
) + sin (.
1
) cos (.
1
) + nc cos (2.
1
) 2.
1
cos (.
1
) + .
2
cos (.
1
)
= 2.
1
cos (.
1
) + sin (.
1
) cos (.
1
) + nc cos (2.
1
)
then
_ .
1
= 2.
1
+ .
2
(6.13-a)
_ .
2
= 2.
1
cos (.
1
) + sin (.
1
) cos (.
1
) + nc cos (2.
1
) (6.13-b)
4
Note that the new state equations also have an equilibrium at (0. 0). By using the control
law
n =
1
c cos (2.
1
)
[ + 2.
1
cos (.
1
) sin (.
1
) cos (.
1
)] (6.14)
is possible to cancel nonlinearities and is an equivalent input to be designed (equivalent
in the sense that determining amounts to determining n, and vice versa), leaing to a
linear input-state relation
_ .
1
= 2.
1
+ .
2
(6.15-a)
_ .
2
= (6.15-b)
Thus, through the state transformation (1) and (1) and input transformation (6.14), the
problem of stabilizing the original nonlinear dynamics (1)-(1) using the original control n
has been transformed into the problem of stabilizing the new dynamics (1)-(1) using the
new input . Since the new dynamics is linear and controllable,it is well known that the
linear state feedback control law
= /
1
.
1
/
2
.
2
can place the poles anywhere with proper choices of feedback gains. For example, we may
choose
= 2.
2
(6.16)
resulting in the stable closed-loop dynamics
_ .
1
= 2.
1
+ .
2
_ .
2
= 2.
2
whose poles are both placed at 2. In terms of the original state r
1
and r
2
, this control
law corresponds to the original input
n =
1
c cos (2r
1
)
[2 (cr
2
+ sin (r
1
)) + 2r
1
cos (r
1
) sin (r
1
) cos (r
1
)] (6.17)
=
1
c cos (2r
1
)
[2cr
2
2 sin (r
1
) + 2r
1
cos (r
1
) sin (r
1
) cos (r
1
)]
and the original state is given from z by
r
1
= .
1
(6.18-a)
r
2
=
.
2
sin (.
1
)
c
(6.18-b)
and since both .
1
. .
2
converges to zero, the original state x converges to zero too.
Remark 3 The closed-loop system under the above control law is represented in the block
diagram in Fig. 6.3, where we can detect two loops in this control system
5
1. Inner loop: this achieves the linearization of the input-state relation
2. Outer loop: achieves the stabilization of the closed-loop dynamics
This is consistent with (6.14), where n is seen to be composed of a nonlinearity cancellation
part and a linear compensation part.
Figure 6.3. Input-state linearization
Note, however, that we have some comments about this method
In the points
r
1
=
:
4

/:
2
. / = 1. 2. ...
we have a singularity in (6.14) and the controller cannot bring the system to the
equilibrium point
Input state linearization is a linearization by feedback dierent from a Jaobian
linearization
In order to implement the control law, the states (.
1
. .
2
) must be available. if they
are not physically meaningful or cannot be measured directly, the original state x
must be measured and used to compute them from (1)-(1)
If there is uncertainty in the model, e.g., uncertainty in the parameter c, this uncer-
tainty will cause error in the computation of both the new state z and the control
input n, as seen in (1)-(1) and (6.14)
Tracking control can also be considered. However, the desired motion then needs to
be expressed in terms of the full new state vector and complexcomputations may
be needed to translate the desired motion specication (in terms of physical output
variables) into specications in terms of the new states
From the above is interesting to ask:
What classes of nonlinear systems can be transformed into linear systems?
How to nd the proper transformations for those which can?
6
1.1.3. Input-Output linearization
Let us now consider the tracking problem. Consider the system
_ x = f (x. n) (6.19-a)
= /(r) (6.19-b)
and assume that the objective is to make the output (t) track a desired trajectory
d
(t)
while keeping the whole state bounded, where
d
(t) and its time derivatives up to a
sucient high order are assumed to be known and bounded.
Remark 4 An apparent diculty with this model is that the output is only indirectly
related to the input nm through the state variable x and the nonlinear state-equations
(1)-(1). Then, it is not easy to see how the input n can be designed to control the tracking
behavior of the output .
It has been seen that one might guess that the diculty of the tracking control design
can be reduced if we can nd a direct and simple relation between the system
output and the control input. This idea constitutes the intuitive basis for the so
called input-output linearization approach to nonlinear control design.
Example 3 Let the third order nonlinear system
_ r
1
= sin (r
2
) + (r
2
+ 1) r
3
(6.20-a)
_ r
2
= r
5
1
+ r
3
(6.20-b)
_ r
3
= r
2
1
+ n (6.20-c)
= r
1
(6.20-d)
and to generate the direct relationship between and n, let us dierentiate the output
_ = _ r
1
= sin (r
2
) + (r
2
+ 1) r
3
and because of theres not relation with n yet, we dierentiate again the output
= _ r
2
cos (r
2
) + (r
2
+ 1) _ r
3
+ _ r
2
r
3
=
_
r
5
1
+ r
3
cos (r
2
) + (r
2
+ 1)
_
r
2
1
+ n
+
_
r
5
1
+ r
3
r
3
=
_
r
5
1
+ r
3
_
[r
3
+ cos (r
2
)] + (r
2
+ 1) r
2
1
+ (r
2
+ 1) n
= , (r) + (r
2
+ 1) n (6.21)
where
, (r) =
_
r
5
1
+ r
3
_
[r
3
+ cos (r
2
)] + (r
2
+ 1) r
2
1
(6.22)
then, if we choose
n =
1
r
2
+ 1
[ , (r)] (6.23)
7
where is the new input to be determined, the nonlinearity in (1) is canceled, and we
obtain a simple linear double integrator relationship between the output and the new input
=
Let c = (t)
d
(t) be the tracking error and choosing the new input
=
d
/
1
c /
2
_ c (6.24)
with /
1
and /
2
being positive constants, and the tracking error of the closed loop system
is
c + /
2
_ c + /
1
c = 0 (6.25)
which represents an exponentially stable eror dynamics. Note that
1. The control law is dened everywhere, except in the singularity r
2
= 1
2. Full state measurement is neccesary in implementing the control law
The above control design strategy is referred to as the input-output linearization
approach. If we need to dierentiate the output of a system : times to generate an
explicit relationship between and n, ths system is said to have relative degree :.
Then, the above system has relative degree 2.
Remark 5 For any controllable system of order : it will take at most : dierentiations
of any output for the control input to appear, i.e.
: _ :
Remark 6 One must remenber that (6.25) only accounts for part of the closed-loop dy-
namics, because it has only order 2, while the whole dynamics has order 3. Therefore, a
part of the system dynamics has been rendered unobservable in the input-output lineariza-
tion. This part of the dinamics is called internal dynamics, because it cannot been seen
from the external input-output relationship (1).
Example 4 From the above example, the internal state can be chosen to be r
3
, because
r
2
, and _ constitutes a new set of states, and the internal dynamics is represented by
_ r
3
= r
2
1
+ n
= r
2
1
+
1
r
2
+ 1
[ , (r)]
= r
2
1
+
1
r
2
+ 1
[
d
/
1
c /
2
_ c , (r)] (1)
If the internal state is stable (in BIBO sense), the tracking control problem has indeed
been solved. Otherwise, the above tracking controller is practically maningless, because the
instability of the internal dynamics would imply undesirable phenomena such as burning-
up of fuses or the violent vibration of mechanical members.
8
Remark 7 As it can be seen, the efectiveness of the above control design, based on the
reduced-order model (1), hinges upon the stability of the internal dynamics.
Example 5 Consider the nonlinear system
_
_ r
1
_ r
2
_
=
_
r
3
2
+ n
n
_
(6.27-a)
= r
1
(6.27-b)
by supposing that the control objective is to make to track
d
, we have
_ = _ r
1
= r
3
2
+ n
then, we can use
n = r
3
2
+ _
d
c (t) (6.28)
which makes c to converge to zero
_ c + c = 0 (6.29)
The second control input is also applied to the second dynamic equation, leading to the
internal dynamics
_ r
2
+ r
3
2
= _
d
c (6.30)
which is non-autonomous and nonlinear. However, c is bounded and _
d
is assumed to be
bounded, then
| _
d
c| _ 1
where 1 is a positive constant. Thus, it can be concluded that |r
2
| _ 1
1=3
, since
_ r
2
< 0 when r
2
1
1=3
_ r
2
0 when r
2
< 1
1=3
therefore, (6.28) does represent a satisfactory tracking control law for the system (2)-(2),
given any trajectory
d
(t) whose derivative _
d
is bounded. Note, however, that if _ r
2
= n,
then the resulting internal dynamics is unstable.
Remark 8 Although the input-output linearization is motivated in the context of output
tracking, it can be also applied to stabilization problems.
Remark 9 In stabilization problems there is not reason to restrict the choice of output
= /(r) to be a physical meaningful quantity, while in tracking problems the choice of
the output is determined by the physical task.
Remark 10 Dierent choices of output function leads to dierent internal dynamics.
It is possible for one choice of output to yield a stable internal dynamics (or no inter-
nal dynamics), while another choice of output would lead to a unstable one. Therefore,
one should choose, if possible, the output function to be such that the associated internal
dynamics is stable.
9
Remark 11 When the relative degree of the system is the same as its order, i.e., when
the output has to be dierentiated : times (with : being the order of the system) to
obtain a linear input-output relation, then the variables . _ . ....
(n1)
may be used as a
new set of state variables for the system, and there is not internal dynamics associated
with this input-output linearization.
Example 6 (Internal dynamics in two linear systems) Consider the simple controllable
and observable linear system
_
_ r
1
_ r
2
_
=
_
r
2
+ n
n
_
(6.31-a)
= r
1
(6.31-b)
where (t) is requiered to track a desired output
d
(t). Then we have
_ (t) = _ r
1
= r
2
+ n
then we can use the control law
n(t) = r
2
+ _
d
(t) c (t) (6.32)
which yields to the tracking error equation
_ c + c = 0
and the internal dynamics
_ r
2
+ r
2
= _
d
c
and while
d
(and _ _
d
), r
2
remains bounded, and so does n. Therefore, (6.32)
is a satisfactory tracking controller for system (6.31-a)-(6.31-b). Let now to consider the
slightly dierent system
_
_ r
1
_ r
2
_
=
_
r
2
+ n
n
_
(6.33-a)
= r
1
(6.33-b)
The same control law yields to the same tracking error dynamics, but now leads to the
internal dynamics
_ r
2
r
2
= c _
d
which implies that r
2
, and accordingly n, both to go to innity as t . Therefore (6.32)
is not a suitable tracking controller for system (6.33-a)-(6.33-b).
For understanding the above results let us consider the transfer function for the system
(6.31-a)-(6.31-b)
G
1
(:) =
_
1 0

__
: 0
0 :
_
_
0 1
0 0
__
1
_
1
1
_
=
: + 1
:
2
10
and for (6.33-a)-(6.33-b)
G
2
(:) =
_
1 0

__
: 0
0 :
_
_
0 1
0 0
__
1
_
1
1
_
=
: 1
:
2
where we see that the two systems have the same poles but dierent zeros. Then, we can
conclude that
If the plant zeros are in the left-half plane (the plant is minimum phase) the
internal dynamics is stable, which actually is true for all linear systems. This is true
because for nonminimum-phase, perfect tracking of arbitrary trajectories requieres
innite control eort
Example 7 Consider now a thir order linear system in state space form
_
_ z = Az + bn
= c
T
z
(6.34)
and having a zero (and hece two more poles than zeros). The systems input-output lin-
earization can be facilitated if we transform it into the so-called companion form. To this
end, from linear control we note that the input output behavior of this system can be
expressed in the form
= c
T
(:I A)
1
bn =
/
1
: + /
0
:
3
+ c
2
:
2
+ c
1
: + c
0
n (6.35)
Thus, if we dene
r
1
=
1
:
3
+ c
2
:
2
+ c
1
: + c
0
n
r
2
= _ r
1
r
3
= _ r
2
this system can be equivalently represented in the companion form
d
dt
_
_
r
1
r
2
r
3
_
_
=
_
_
0 1 0
0 0 1
c
0
c
1
c
2
_
_
_
_
r
1
r
2
r
3
_
_
+
_
_
0
0
1
_
_
n (6.36-a)
=
_
/
0
/
1
0

_
_
r
1
r
2
r
3
_
_
(6.36-b)
Now we perform input-output linearization based on this form and we obtain
_ = /
0
_ r
1
+ /
1
_ r
2
= /
0
r
2
+ /
1
r
3
11
= /
0
_ r
2
+ /
1
_ r
3
= /
0
r
3
+ /
1
[c
0
r
1
c
1
_ r
1
c
2
r
1
+ n]
= /
0
r
3
+ /
1
[c
0
r
1
c
1
r
2
c
2
r
3
+ n]
= /
1
n /
1
[c
0
r
1
+ c
1
r
2
+ c
2
r
3
] + /
0
r
3
then, the relative degree is two. Thus, the control law
n =
_
c
0
r
1
+ c
1
r
2
+ c
2
r
3
/
0
/
1
r
3
_
+
1
/
1
(
d
/
2
_ c /
1
c) (6.38)
yields an exponentially stable error
c + /
2
_ c + /
1
c = 0
and the internal dynamics can be described for only one state equation. We can use r
1
to complete the state vector, since r
1
. . _ are related to r
1
. r
2
. r
3
through a one-to-one
transformation. Then the internal dynamics is
_ r
1
= r
2
=
1
/
1
( /
0
r
1
)
that is
_ r
1
+
/
0
/
1
r
1
=
1
/
1
(6.39)
where (6.36-a)-(6.36-b) were used, and since is bounded ( = c +
d
), stability of the
internal dynamics depends of the location of the zero /
0
,/
1
of the transfer function in
(6.35). If the system is minimum phase, then the zero is in the left-half plane, which
implies that the internal dynamics (6.39) is stable, independently of the initial conditions
and of the magnitudes of the desired .
(1)
. ....
(r)
(where : is the relative degree).
1.1.4. Zero Dynamics
We have seen that for linear systems the stability of the internal dynamics is simply
determined by the locations of the zeros, but it is interesting to see whether this relation
can be extended to nonlinear systems. However, to do this is neccessary to extend the
notion of zeros to nonlinear systems.
The above is nontrivial, because linear system zeros are based on its transfer function,
which cannot be dened for nonlinear systems. Futhermore, zeros are intrinsic properties
of a linear plant, while for nonlinear systems the stability of the internal dynamics may
depend on the specic control input. A way to approach these diculties is to dene the
so called zero-dynamics for a nonlinear system.
Denition 1 (Zero-dynamics) The zero-dynamics is the internal dynamics of the sys-
tem when the system output is kept at zero by the input.
For example, for the system (2)-(2) we have that for keeping = r
1
= 0 we need n = r
3
2
,
which lead us to the zero-dynamics
_ r
2
+ r
3
2
= 0 (6.45)
12
We see that the internal dynamics is an intrinsic property of a nonlinear system. The
zero-dynamics (6.45) is tested with the Lyapunov function \ =
1
2
r
2
2
, then
_
\ = r
2
_ r
2
= r
2
_
r
3
2
_
= r
4
2
< 0
then (6.45) is asymptotically stable. Similarly, for (6.34) the zero-dynamics is (from (6.39))
_ r
1
+
/
0
/
1
r
1
= 0
Then, for this linear system the poles of the zeros dynamics are the zeros of the
system. This result is general for linear systems, and therefore, in linear systems having
all zeros in the left-half complex plane guarantees the global asymptotic stability of the
zero-dynamics.
Remark 12 The reason for studying the zero-dynamics is that we want to nd a simpler
way of dening the stability of the internal dynamics.
Remark 13 For linear systems, the stability of the zero-dynamics implies the global sta-
bility of the internal dynamics: the left-hand side of (6.39) completely determines its sta-
bility characteristics, given that the right-hand side tends to zero or is bounded.
Remark 14 In nonlinear systems, however, the relation is not so clear. For stabilization
problems, it can be shown that the local asymptotic stability of the zero dynamics is enough
to guarantee the local asymptotic stability of the internal dynamics. Extensions can be
drawn to the tracking problem. However, unlike the linear case, no results on the global
stability or even larga range stability can be drawn for internal dynamics of nonlinear
systems, i.e., only local stability is guaranteed for the internal dynamics even if the zero
dynamics is globally exponentially stable.
Remark 15 As in the linear case, a nonlinear system whose zero dynamics is asymptot-
ically stable is called an asymptotically minimum phase system.
Remark 16 The zero dynamics is an intrinsic feature of a nonlinear system, which does
not depend on the choice of the control law or the desired trajectories.
Remark 17 Examining the stability of zero dynamics is much easier than examining the
stability of internal dynamics, because the zero-dynamics only involves the internal states,
while the internal dynamics is coupled to the external dynamics and desired trajectories.
The control design based on input-output linearization can be made in three steps
1. Dierentiate the output until the input n appears
2. Choose n to calcel the nonlinearities and guarantee tracking convergence
13
3. Study the stability of the internal dynamics
Remark 18 If the relative degree associated with the input-output linearization is the
same as the order of the system, the nonlinear system is fully linearizedn and this procedure
indeed leads to a satisfactory controller (assuming that the model is accurate)
Remark 19 If the relative degree is smaller than the system order, then the nonlinear
system is only partially linearized, and whether the controller can indeed be applied depends
on the stability of the internal dynamics.
Remark 20 The study of the internal dynamics can be simplied locally by studying that
of the zero-dynamics instead.
Remark 21 If the zer-dynamics is unstable, dierent control strategies should be sought
14
2. MATHEMATICAL TOOLS
2. Mathematical tools
We shall call a vector function , : R
n
R
n
a vector eld in R
n
, because to every vector
function , corresponds a eld of vectors in an :dimensional space. Inthe following, we
shall be interested only in smooth vector elds, where by smoothness of a vector eld,
we mean that the function f (r) has continuous partial derivatives of any required order.
Given any smooth scalar function /(r) of te state x, the gradient of / is denoted by \/,
where
\/ =
J/
Jx
and it is represented by a row-vector of elements
(\/)
i
=
J/
Jr
i
Similarly, given a vector eld f (r), the Jacobian of f is denoted by \f
\f =
Jf
Jx
which is represented by a : : matrix of elements
(\f )
ij
=
Jf
i
Jr
j
2.1. Lie derivatives and Lie brackets
Given a scalar function /(r) and a vector eld f (r), we dene a new scalar function 1
f
/,
called the Lie derivative of / (or simply the derivative) with respect to f .
Denition 2 (Lie derivative) Let / : R
n
R be a smooth scalar function, and f :
R
n
R
n
be a smooth vector eld on R
n
, then the Lie derivative of / with respect to f is
a scalar function dened by
1
f
/ = \/f
Thus, the Lie derivative of / is simply the directional derivative of / along the direction
of the vector f . Repeated Lie derivatives can be dened recursively
1
0
f
/ = /
1
i
f
/ = 1
f
_
1
i1
f
/
_
= \
_
1
i1
f
/
_
f . i = 1. 2. ...
Similarly, if g is another vector eld, then the scalar function 1
g
1
f
/(x) is
1
g
1
f
/(x) = \(1
f
/(x)) g
15
Example 8 Let the system
_ x = f (r)
= /(x)
then, the derivatives of the output are
_ =
J/
Jx
_ x = 1
f
/
=
J [1
f
/]
Jx
_ x = 1
2
f
/
and so on. Similarly, if \ is a Lyapunov function candidate for the system, its time
derivative
_
\ can be written as 1
f
\ .
Denition 3 (Lie brackets) Let f and g be two smooth vector elds on R
n
. The Lie
bracket of f and g is a third vector eld dened by
[f . g] = \g f \f g
and it can be written as
[f . g] = cd
f
g
Repeated Lie brackets can be dened recursively as
cd
0
f
g = g
cd
i
f
g =
_
f . cd
i1
f
g
. i = 1. 2. ...
_ r
1
= 2r
1
+ cr
2
+ sin (r
1
)
_ r
2
= r
2
cos (r
1
) + ncos (2r
1
)
which can be rewritten as
_ x =
_
2r
1
+ cr
2
+ sin (r
1
)
r
2
cos (r
1
)
_
+
_
0
cos (2r
1
)
_
n
= f (r) + g (r) n
16
then
[f . g] = \g f rf g
= \
_
0
cos (2r
1
)
_
_
2r
1
+ cr
2
+ sin (r
1
)
r
2
cos (r
1
)
_
r
_
2r
1
+ cr
2
+ sin (r
1
)
r
2
cos (r
1
)
_
_
0
cos (2r
1
)
_
=
_
0 0
2 sin (2r
1
) 0
_ _
2r
1
+ cr
2
+ sin (r
1
)
r
2
cos (r
1
)
_
_
2 + cos (r
1
) c
r
2
sin (r
1
) cos (r
1
)
_ _
0
cos (2r
1
)
_
=
_
0
2 [2r
1
+ cr
2
+ sin (r
1
)] sin (2r
1
)
_
_
c cos (2r
1
)
cos (r
1
) cos (2r
1
)
_
=
_
c cos (2r
1
)
4r
1
sin (2r
1
) 2cr
2
sin (2r
1
) 2 sin (r
1
) sin (2r
1
) + cos (r
1
) cos (2r
1
)
_
Lemma 1 Lie brackets have the following properties
1. Bilinearity
[c
1
f
1
+ c
2
f
2
. g] = c
1
[f
1
. g] + c
2
[f
2
. g]
[f . c
1
g
1
+ c
2
g
2
] = c
1
[f . g
1
] + c
2
[f . g
2
]
where f . f
1
. f
2
. g. g
1
. g
2
are smooth vector elds and c
1
. c
2
are onstant scalars.
2. Skew-commutativity
[f . g] = [g. f ]
3. Jacobi-identity
1
ad
f
g
/ = 1
f
1
g
/ 1
g
1
f
/
where /(r) is a smooth scalar function of x.
The Jacobi-identity can be used resursively to obtain useful technical identities. Using
twice yields
1
ad
2
f
g
/ = 1
ad
f
(ad
f
g)
/
= 1
f
1
ad
f
g
/ 1
ad
f
g
1
f
/
= 1
f
[1
f
1
g
/ 1
g
1
f
/] [1
f
1
g
1
g
1
f
] 1
f
/
= 1
2
f
1
g
/ 21
f
1
g
1
f
/ + 1
g
1
2
f
/ (6.47)
ans imilar identities can be obtained for higher-order Lie brackets.
17
2.2. Dieomorphisms and state transformations
A dieomorphism can be seen as a generalization of the familiar concept of coordinate
transformation and is dened as follows.
Denition 4 (Dieomosphism) A function c : R
n
R
n
, dened in a region , is
called a dieomosphism if it is smooth, and if its inverse c
1
exists and is smooth.
If is the whole space R
n
, then c(r) is called a global dieomorphism. Global dieo-
morphisms are rare, and therefore one often looks for local dieomorphisms, i.e., for
transformations dened only in a nite neighborhood of a given point.
Lemma 2 Let c(r) be a smooth function dened in a region R
n
. if the Jacobian
matrix \c is non-singular at a point x = x
0
of , then c(r) denes a local dieorphism
in a subregion of .
Remark 22 A dieomorphism can be used to transform a nonlinear system into another
nonlinear system in terms of a new set of state variables.
_ x = f (r) + g (r) n
= /(x)
and let the new set of states be dened by
z (x) = c(x)
then
_ z =
Jc
Jx
_ x =
Jc
Jx
[f (r) + g (r) n]
and then
x = c
1
[z]
and
_ z =
Jc
J
_
c
1
[z]
_f
_
c
1
[z]
_
+
Jc
J
_
c
1
[z]
_g
_
c
1
[z]
_
n
= f
(z) + g
(z) n
= /(x) = /
_
c
1
[z]
_
= /
(z)
where
f
(z) =
Jc
J
_
c
1
[z]
_f
_
c
1
[z]
_
g
(z) =
Jc
J
_
c
1
[z]
_g
_
c
1
[z]
_
/
(z) = /
_
c
1
[z]
_
18
Example 11 The nonlinear function
_
.
1
.
2
_
= c(r) =
_
2r
1
+ 5r
1
r
2
2
2 sin r
2
_
(6.48)
is well dened for all r
1
. r
2
and its Jacobian matrix is given by
\c(r) =
_
2 + 5r
2
2
10r
1
r
2
0 2 cos r
2
_
then
det [\c(r)] = 4 cos r
2
+ 10r
2
2
cos r
2
and
det [\c(r)]
(0;0)
= 4
so it has rank 2 at (0. 0). Therefore, Lemma 2 indicates that (6.48) denes a local dieo-
morphism around the origin. In fact the dieomorphism is valid in the region
=
_
(r
1
. r
2
) : [r
2
[ <
:
2
_
2.3. The Frobenius theorem
Frobenius theorem provides a neccessary and sicient condition for the solvability of a
special class of partial dierential equations. Consider the set of rst-order partial dier-
ential equations
J/
Jr
1
,
1
+
J/
Jr
2
,
2
+
J/
Jr
3
,
3
= 0 (6.49-a)
J/
Jr
1
q
1
+
J/
Jr
2
q
2
+
J/
Jr
3
q
3
= 0 (6.49-b)
where ,
i
(r
1
. r
2
. r
3
) . q
i
(r
1
. r
2
. r
3
) . i = 1. 2. 3 are known scalar functions of r
1
. r
2
. r
3
and
/(r
1
. r
2
. r
3
) is an unknown function. This set of partial dierential equations is uniquely
dened by the two vectors f = [,
1
. ,
2
. ,
3
]
T
. g = [q
1
. q
2
. q
3
]
T
. If a solution /(r
1
. r
2
. r
3
) ex-
ists for the above partial dierential equations, we shall say the set of vector elds f . g
is completely integrable. But the question is: when these equations are solvable. Froe-
nius theorem says that (2)-(2) has a solution /(r
1
. r
2
. r
3
) i there exist scalar functions
c
1
(r
1
. r
2
. r
3
) . c
2
(r
1
. r
2
. r
3
) such that
[f . g] = c
1
f + c
2
g
i.e., if the Lie bracket [f . g] can be expressed as a linear combination of f and g. This
condition is called involutivity condition on the vector elds f . g. Geometrically it
means that the vector [f . g] is in the planed formed by the two vectors f and g. Then,
Frobenius theorem states that the set of vectors f . g is completely integrable i it is
involutive.
19
Denition 5 (Set completely integrable) A linearly independent set of vector elds
f
1
. f
2
. .... f
m
on R
n
is said to be completely integrable if, and only if, there exist :
: scalar functions /
1
(x) . /
2
(x) . .... /
nm
(x) satisfying the system of partial dierential
equations
\/
i
f
j
= 0 (6.50)
where 1 _ i _ : :, 1 _ , _ :, and the gradients \/
i
are linearly independent.
Denition 6 (Involutivity) A linearly independent set of vector elds f
1
. f
2
. .... f
m
is
said to be involutive if and only if there are scalar functions c
ijk
: R
n
R such that
[f
i
. f
j
] (x) =
m
k=1
c
ijk
(x) f
k
(x) . \i. , (6.51)
Remark 23 Involutivity means that if one forms the Lie bracket of any pairs of vector
elds from the set f
1
. f
2
. .... f
m
, then the resulting vector eld can be expressed as a linear
combination of the original set of vector elds.
Note that:
Constant vector elds are always involutive. Indeed, Lie bracket of two constant vec-
tors is simply the zero vector, which can be trivially expressed as linear combination
of the vector elds
A set composed of a single vector f is involutive. Indeed
[f . f ] = 0
Checking whether a set of vector elds f
1
. f
2
. .... f
m
is involutive amounts to check-
ing whether
:c:/ (f
1
(x) . f
2
(x) . .... f
m
(x)) = :c:/ (f
1
(x) . f
2
(x) . .... f
m
(x) . [f
i
. f
j
] (x))
for all x and all i. ,.
Teorema 1 (Frobenius) Let f
1
. f
2
. .... f
m
be a set of linearly independent vector elds.
The set is completely integrable if, and only if, it is involutive.
Example 12 Consider the set of partial dierential equations
4r
3
J/
Jr
1

J/
Jr
2
= 0
r
1
J/
Jr
1
+
_
r
2
3
3r
2
_
J/
Jr
2
+ 2r
3
J/
Jr
3
= 0
The associated vector elds are f
1
. f
2
, with
f
1
=
_
_
4r
3
1
0
_
_
. f
2
=
_
_
r
1
r
2
3
3r
2
2r
3
_
_
20
in order to determine if the set of partial dierential equations is solvable (or whether
f
1
. f
2
is completely integrable), let us check the involutivity of the set vector elds f
1
. f
2
.
Then
[f
1
. f
2
] =
_
_
1 0 0
0 3 2r
3
0 0 2
_
_
_
_
4r
3
1
0
_
_
_
_
0 0 4
0 0 0
0 0 0
_
_
_
_
r
1
r
2
3
3r
2
2r
3
_
_
=
_
_
4r
3
3
0
_
_
_
_
8r
3
0
0
_
_
=
_
_
12r
3
3
0
_
_
= 3f
1
+ 0f
2
then f
1
. f
2
is involutive and the two partial dierential equations are solvable.
21
3. INPUT-STATE LINEARIZATION OF SISO SYSTEMS
3. Input-State Linearization of SISO systems
Let the nonlinear system
_ x = f (r) +g (r) n (6.52)
with f and g being smooth vector elds. We shall study
1. When such nonlinear systems can be linearized by state and input transformations
2. How to nd such transformations
3. How to design controllers based on such feedback transformations
Systems in the form (6.52) are said to be linear in the control or ane. If a nonlinear
system has the form
_ x = f (r) +g (r) n[n + c(x)]
with n an invertible scalar function and c an arbitrary dunctional, then by using =
n[n + c(x)] we can obtain
_ x = f (r) +g (r)
which is in the form (6.52) and one can design a control law for and then compute n by
inverting n, i.e.,
n = n
1
c(x)
3.1. Input-state linearization
Denition 7 (Input-state linearization) A single-input nonlinear system in the form
(6.52), with f (r) and g (r) being smooth vector elds on R
n
, is said to be input-state
linearizable if there exists a region in R
n
, a dieomorphism c : R
n
, and a nonlinear
feedback control law
n = c(x) + , (x) (6.53)
such that the new state variables z = c(x) and the new input satisfy a linear time
invariant relation
_ z = Az + b (6.54)
where
A =
_
_
0 1 0 0
0 0 1 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 0 1
0 0 0 0
_
_
. b =
_
_
0
0
.
.
.
0
1
_
_
The new state z is called the linearizing state and the control law (6.53) is called the
linearizing control law. We will often use z to denote not only the transformed state,
but the dieomorphism c itself, i.e.,
z = z (x)
22
Remark 24 Note that A and b are in companion form, then, if (6.53) can be transformed
into a linear system, it can be transformed into the form prescribed by (6.54) by using an
extra transformation.
Remark 25 Is easily to see from the canonical form (6.54) that feedback linearization is
a special case of input-output linearization, qhere the output function leads to a relative
degree :. This means that if a system is input-output linearizable with relative degree :,
then it must be input-state linearizable.
Remark 26 If a system is input-state linearizable, with the rst new state .
1
representing
the output, the system is input-output linearizable with relative degree :, as stated next.
Lemma 3 An :
th
o:dc: nonlinear system is input-state linearizable if, and only if,
there exists a scalar function .
1
(x) such that the systemss input-output linearization with
.
1
(x) with .
1
(x) as output function has relative degree :.
3.2. Conditions for input-state linearization
Teorema 2 The nonlinear system
_ x = f (r) +g (r) n
with f (x) . g (x) being smooth vector elds, is input-state linearizable if, and only if, there
exists a region such that the following conditions hold
1. The vector elds
_
g. cd
f
g. ....cd
n1
f
g
_
are linearly independent in
2. The set
_
g. cd
f
g. ....cd
n2
f
g
_
is involutive in
Remark 27 The rst condition can be seen as a controllability condition for the nonlinear
system (6.52), because for linear systems we have
_
g. cd
f
g. ....cd
n1
f
g
_
=
_
b. Ab. .... A
n1
b
_
and their independence is equivalent to the invertibility of the controllability matrix.
Remark 28 The involutivity condition is less intuitive, which is trivially satised for
linear systems, but not generically satised in the nonlinear case.
Lemma 4 Let . (x) be a smooth function in a region . Then, in , the set of equations
1
g
. = 1
g
1
f
. = ... = 1
g
1
k
f
. = 0 (6.55-a)
is equivalent to
1
g
. = 1
ad
f
g
. = ... = 1
ad
k
f
g
. = 0 (6.55-b)
for any positive integer /.
23
3.3. How to perform input-state linearization
Input state linearization can be performed through the following steps
1. Construct the vector elds
g. cd
f
g. .... cd
n1
f
g
for the given system
2. Check whether the controllability and involutivity cinditions are satised
3. If both are satised, nd the rst state .
1
(the output function leading to input-state
linearization of relative degree :) from equations
1
g
.
1
= 1
ad
f
g
.
1
= ... = 1
ad
n2
f
g
.
1
= 0
i.e.,
\.
1
cd
i
f
g = 0. i = 0. .... : 2 (6.62-a)
\.
1
cd
n1
f
g ,= 0 (6.62-b)
4. Compute the state transformation
z (x) =
_
.
1
1
f
.
1
... 1
n1
f
.
1
T
and the input transformation (6.53), with
c(x) =
1
n
f
.
1
1
g
1
n1
f
.
1
(6.63-a)
, (x) =
1
1
g
1
n1
f
.
1
(6.63-b)
Example 13 Consider the control of a exible-joint mechanism whose equations of mo-
tion can be easily derived as
1
1
+ `q1sin
1
+ / (
1
2
) = 0 (6.64-a)
J
2
/ (
1
2
) = n (6.64-b)
Note that nonlinearities appear in the rst equation, while the control input appears only
in the second equation, then, there is not obvious way of designing a large range controller.
Let us consider now when input-state linearization is possible. Let the state system to be
x = [
1
. _
1
.
2
. _
2
]
T
then
_ x =
_
_
r
2
MgL
I
sin (r
1
)
k
I
(r
1
r
3
)
r
4
k
J
(r
1
r
3
)
_
_
+
_
_
0
0
0
1
J
_
_
n
24
then we have
g =
_
_
0
0
0
1
J
_
_
cd
f
g = 0
41
_
0 1 0 0
MgL
I
cos (r
1
)
k
I
0
k
I
0
0 0 0 1
k
J
0
k
J
0
_
_
_
_
0
0
0
1
J
_
_
=
_
_
0
0
1
J
0
_
_
cd
2
f
g = cd
f
(cd
f
g) =
_
_
0 1 0 0
MgL
I
cos (r
1
)
k
I
0
k
I
0
0 0 0 1
k
J
0
k
J
0
_
_
_
_
0
0
1
J
0
_
_
=
_
_
0
k
JI
0
k
J
2
_
_
cd
3
f
g =
_
_
0 1 0 0
MgL
I
cos (r
1
)
k
I
0
k
I
0
0 0 0 1
k
J
0
k
J
0
_
_
_
_
0
k
JI
0
k
J
2
_
_
=
_
k
JI
0
k
J
2
0
_
_
and we check the controllability condition as
det
_
q. cd
f
q. cd
2
f
g. cd
3
f
g
= det
_
_
0 0 0
k
JI
0 0
k
JI
0
0
1
J
0
k
J
2
1
J
0
k
J
2
0
_
_
=
/
2
J
4
1
2
which has rank 4 for / 0, 1J < . Futhermore, since [q. cd
f
q. cd
2
f
g. cd
3
f
g] is constant,
they form an involutive set, then the system (2)-(2) is input-state linearizable. Now, let us
nd the state transformation z = z (x) and the input transformation n = c(x) + , (x)
so that input-state linearization is achieved. From (6.62-a)-(6.62-b), the rst component
.
1
of the new state vector z should satisfy
_
@z
1
@x
1
@z
1
@x
2
@z
1
@x
3
@z
1
@x
4
_
0
0
0
1
J
_
_
= 0.
_
@z
1
@x
1
@z
1
@x
2
@z
1
@x
3
@z
1
@x
4
_
0
0
1
J
0
_
_
= 0
_
@z
1
@x
1
@z
1
@x
2
@z
1
@x
3
@z
1
@x
4
_
0
k
JI
0
k
J
2
_
_
= 0.
_
@z
1
@x
1
@z
1
@x
2
@z
1
@x
3
@z
1
@x
4
k
JI
0
k
J
2
0
_
_
,= 0
from where we can conclude that .
1
must satisfy
J.
1
Jr
4
= 0.
J.
1
Jr
3
= 0.
J.
1
Jr
2
= 0.
J.
1
Jr
1
,= 0
25
Then, .
1
must be a function of r
1
only. The simplest solution to the above equations is
.
1
= r
1
(6.65-a)
and the other states can be computed from .
1
.
2
= 1
f
.
1
= \.
1
f =
_
1 0 0 0

_
_
r
2
MgL
I
sin (r
1
)
k
I
(r
1
r
3
)
r
4
k
J
(r
1
r
3
)
_
_
= r
2
(6.65-b)
.
3
= 1
2
f
.
1
= 1
f
(1
f
.
1
) =
_
0 1 0 0

_
_
r
2
MgL
I
sin (r
1
)
k
I
(r
1
r
3
)
r
4
k
J
(r
1
r
3
)
_
_
(6.65-c)
=
`q1
1
sin (r
1
)
/
1
(r
1
r
3
)
.
4
= 1
3
f
.
1
=
_

MgL
I
cos (r
1
)
k
I
0
k
I
0

_
_
r
2
MgL
I
sin (r
1
)
k
I
(r
1
r
3
)
r
4
k
J
(r
1
r
3
)
_
_
(6.65-d)
=
_
`q1
1
cos (r
1
) +
/
1
_
r
2
+
/
1
r
4
=
`q1
1
r
2
cos (r
1
)
/
1
(r
2
r
4
)
and the input transformation is
n =
1
n
f
.
1
+
1
g
1
n1
f
.
1
where
`q1
1
r
2
cos (r
1
)
/
1
(r
2
r
4
)
1
4
f
.
1
=
_
MgL
I
r
2
sin (r
1
)
_
k
I
+
MgL
I
cos (r
1
)
_
0
k
I
_
r
2
MgL
I
sin (r
1
)
k
I
(r
1
r
3
)
r
4
k
J
(r
1
r
3
)
_
_
=
`q1/
1
2
sin r
1
+
`q1/
1
2
r
1
cos r
1
`q1/
1
2
r
3
cos r
1
+
/
2
1
2
r
1
/
2
1
2
r
3
+
`q1
1
r
2
2
sin r
1
+
/
2
J1
r
1
/
2
J1
r
3
+
(`q1)
2
1
2
cos r
1
sin r
1
26
1
g
1
3
f
.
1
= 1
g
_
`q1
1
r
2
cos (r
1
)
/
1
(r
2
r
4
)
_
=
_
MgL
I
r
2
sin (r
1
)
k
I

MgL
I
cos (r
1
) 0
k
I
_
0
0
0
1
J
_
_
=
/
J1
then
c(x) =
1
4
f
.
1
1
g
1
3
f
.
1
=
J1
/
_
MgLk
I
2
sin r
1
+
MgLk
I
2
r
1
cos r
1
MgLk
I
2
r
3
cos r
1
+
k
2
I
2
r
1
k
2
I
2
r
3
+
MgL
I
r
2
2
sin r
1
+ ...
... +
k
2
JI
r
1
k
2
JI
r
3
+
(MgL)
2
I
2
cos r
1
sin r
1
_
, (x) =
1
1
g
1
3
f
.
1
=
J1
/
and
n =
J1
/
[ c(x)] (6.66)
and nally
_ .
1
= .
2
_ .
2
= .
3
_ .
3
= .
4
_ .
4
=
Remark 29 The above input-state linearization is actually global, because the dieomor-
phism z (x) and the input transformation are well dened everywhere. Specically, the
inverse of the state transformation (6.65-a)-(2) is
r
1
= .
1
r
2
= .
2
r
3
= .
1
+
1
/
_
.
3
+
`q1
1
sin .
1
_
r
4
= .
2
+
1
/
_
.
4
+
`q1
1
.
2
cos .
1
_
which is well dened and dierentiable everywhere. The input transformation (6.66) is
also well dened everywhere, of course.
Remark 30 In this example, the transformed variables have physical meanings, because
.
1
is the link position, .
2
the link velocity, .
3
the link acceleration and .
4
the link jerk
27
Remark 31 In hindsight, of course, we also see that the same result could have been de-
rived simply by dierentiating equation (2) twice, i.e., from the input-output linearization
perspective of Lemma 3.
Note that (6.62-b) can be replaced by the normalization equation
\.
1
cd
n1
f
g = 1
without aectiing the input-state linearization. This equation and (6.62-a) constitute a
total of : linear equations
_
cd
0
f
g cd
1
f
g ... cd
n2
f
g cd
n1
f
g

_
_
@z
1
@x
1
@z
1
@x
2
.
.
.
@z
1
@x
n1
@z
1
@x
n
_
_
=
_
_
0
0
.
.
.
0
1
_
_
and given the independence condition on the vector elds, the partial derivarives
@z
1
@x
1
. ....
@z
1
@x
n
can be computed uniquely from the above equations. The state variable .
1
can be found,
in principle, by sequentially integrating these partial derivatives.
Remark 32 Note that analytically solving this set of partial dierential equations for .
1
may be a nontrivial step, although numerical solutions may be relatively easy due to the
recursive nature of the equations.
3.4. Controller design based on input-state linearization
With the state equation transformed into a linear form, one can easily design controllers
for either stabilization or tracking purposes. For the exible link example we have
.
(4)
1
=
and if we desire to have the link position .
1
track a prespecied trajectory .
d
1
(t), then
the control law
= .
(4)
d
1
c
3
c
(3)
1
c
2
c
(2)
1
c
1
_ c
1
c
0
c
1
c
1
= .
1
.
d
1
leads to the tracking error dynamics
c
(4)
1
+ c
3
c
(3)
1
+ c
2
c
(2)
1
+ c
1
_ c
1
+ c
0
c = 0
which is exponentially stable if c
i
0 are chosen properly. To nd the physical input n
one simply uses (6.66).
28
4. INPUT-OUTPUT LINEARIZATION OF SISO SYSTEMS
4. Input-Output linearization of SISO systems
Here is discussed the input-output linearization of single-output nonlinear systems de-
scribed by the state-space representation
_ x = f (x) +g (x) n (6.67-a)
= /(x) (6.67-b)
where is the output. By input-output linearization we mean the generation of a linear
dierential relation between the output and a new input and will be discussed
How to generate a linear input-output relation for a nonlinear system
What are the internal dynamics and zero dynamics associated with the input-output
linearization
How to design stble controllers based on input-output linearizations
4.1. Generating a linear input-output relation
Tha basic approach is to dierentiate the output repeatedly until the input n appears,
and then design n to calcel the nonlinearity. However, in some cases the second part of
the approach cannot be carried out, because the systems relative degree is not dened.
4.1.1. The case of well dened relative degree
Let us place in a region (an open connected set)
x
in the state space. We have that
_ = \/(f + gn) = 1
f
/(x) + 1
g
/(x) n
If 1
g
/(x) ,= 0 for some x = x
0
in
x
, then, by continuity, that relation is veried in a
nite neighborhood of x
0
. In the input transformation
n =
1
1
g
/
(1
f
/ + )
results in a linear relation between and , namely _ = . If 1
g
/(x) = 0 for all x
x
we can dierentiate _ to obtain
= 1
2
f
/(x) + 1
g
1
f
/(x) n
If 1
g
1
f
/(x) is again zero for all x
x
, we shall dierentiate again and again
(i)
= 1
i
f
/(x) + 1
g
1
i1
f
/(x) n
until for some integer :
1
g
1
r1
f
/(x) ,= 0
29
for some x = x
0

x
. Then, by continuity, the above relation is veried in a nite
neighborhood of x
0
. In the control law
n =
1
1
g
1
r1
f
/
(1
r
f
/ + ) (6.68)
applied to
(r)
= 1
r
f
/(x) + 1
g
1
r1
f
/(x) n (6.69)
yields the simple linear relation
(r)
= (6.70)
The number : of dierentiations is called the relative degree of the system, and one
necessarily has : _ :, qhere : is the system order. If : = :, then input-output linearization
actually yields input-state linearization in , as stated in Lemma 3. Based on the above
procedure, we can give the following formal denition.
Denition 8 (Relative degree) The SISO system is said to have relative degree : in a
region if for all x
1
g
1
i
f
/(x) = 0. 0 _ i _ : 1 (6.71-a)
1
g
1
r1
f
/(x) ,= 0 (6.71-b)
4.1.2. The case of undened relative degree
Often we are interested in the system properties about a specic operating point x
0
. Then
the denition of relative degree requires particular care. If
1
g
1
r1
f
/(x
0
) ,= 0
this also implies, by continuity, that (2) is veried over a nite neighborhood of x
0
. We
shall say that the system has relative degree : at the point x
0
.
However, it could be possible that 1
g
1
r1
f
/(x) ,= 0, but 1
g
1
r1
f
/(x
0
) = 0. In this case
the relative degree is undened at x
0
.
Example 14 Consider the system
r = j (r. _ r) + n
qhere j is a smooth nonlinear function of the state x = [r. _ r]
T
. If
= r
is dened as the output of interest, then the system is in companion form with relative
degree 2. However, if one takes
= r
2
30
as output, then
_ = 2r _ r
= 2r r + 2 _ r
2
= 2rj (r. _ r) + 2 _ r
2
+ 2rn
then
1
g
1
f
/(x) = 2r
and at x = 0 the system has neither relative degree 1 nor relative degree 2.
4.2. Normal forms
a
4.3. The zero dynamics
a
4.4. Local asymptotic stabilization
a
4.5. Global asymptotic stabilization
a
4.6. Tracking control
a
4.7. Inverse dynamics
a
4.8. Tracking control for non-minimum phase systems
a
31

Di Erential Algebra: Roger Miranda 9 de Septiembre de 2008

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Di Erential Algebra: Roger Miranda 9 de Septiembre de 2008

Hochgeladen von

Copyright:

Verfügbare Formate

Dierential algebra

Das könnte Ihnen auch gefallen