Академический Документы
Профессиональный Документы
Культура Документы
Mathematical Preliminaries
This chapter presents a few denitions and notions that are used throughout
the monograph.
q = f (qq , q , u )
s(qq , q , q) = 0
s(qq , q , q) = 0
h(qq ) = 0
then the constraint satises the complete integrability property and is called
a holonomic constraint and the system is called a holonomic system . If the
constraint cannot be expressed as in (2.3), then the system is called a nonholonomic system . We present a few examples.
R.N. Banavar and V. Sankaranarayanan: Switch. Fin. Time Contrl., LNCIS 333, pp. 720, 2006.
Springer-Verlag Berlin Heidelberg 2006
2 Mathematical Preliminaries
No lateral motion
x cos + y sin = r
Pure rolling
sin() cos() 0 0 y
=0
cos sin 0 r
This matrix representation of the constraint conveys the fact that permissible
motions of the coin are such that allowable velocity directions at any point
on the manifold (xp , yp , p , p )- described by the vector
x
y
(xp ,yp ,p ,p )
sin() cos() 0 0
cos sin 0 r
dt
If yes, then
1 (qq ) = c1
2 (qq ) = c2
and this implies that the system evolves on a manifold of reduced dimension
(in this case reduced by 2) and the constraints are integrable (or holonomic).
However, in this particular case, it is not possible to nd such functions i s and
the constraints are hence non-integrable. We arrive at this conclusive negative
answer by employing a result from dierential geometry called Frobenius
theorem [33]. Before we state this, we recast the problem in dierential geometric parlance.
The constraints can be viewed in terms of annihilator codistributions as
x
sin() cos() 0 0
y
=0
cos sin 0 r
Permissible motions of the coin are such that the vector eld is annihilated
by the codistribution
=
sin() cos() 0 0
cos sin 0 r
d1
d2
If yes, then the constraints are integrable (or holonomic). Frobenius result
states
10
2 Mathematical Preliminaries
11
where the rst equation (with the right hand side being zero) denotes the lack
of actuation at the rst joint and is a non-integrable acceleration level constraint. 2 is the torque applied at the second joint. The acrobot can assume
any conguration but cannot assume arbitrary accelerations.
m1, m
2
l1 , l2
I1 , I
2
= link lengths
link 2
= link masses
l c2
2
Actuator
l c1
link 1
q
1
qa + f2 (q, q,
q) = F
12
2 Mathematical Preliminaries
fuel tank
where q = (qu , qa ), qu corresponds to the conguration variable of the pendulum, qa corresponds to the conguration variable of the outer rigid body and
F is the external force.
While the acceleration level constraint in the acrobot arises due to purpose
of design or loss of actuation, in the case of the launch vehicle it is the inability
to directly actuate the uid dynamics.
2.2 Stability
Consider an autonomous system described by
(2.6)
x)
x = f (x
where x IRn , f : D IRn is a local map from an open set D IRn into
xe ) = 0. Without loss of
IRn . Let x e D be an equilibrium point, that is, f (x
generality, let us assume x e = 0.
Denition 2.3. (Lyapunov stability) The equilibrium point x e = 0 is said
to be
2.2 Stability
13
t0
x) 0 x D
V (x
asymptotically stable if
x) < 0 x D \ {0}
V (x
The above denitions and theorems are valid in a neighbourhood D around an
equilibrium point. The following theorem concerns global asymptotic stability.
Theorem 2.6. Let V : IRn IR be a continuously dierentiable function
x) > 0, x = 0
V (0) = 0 and V (x
x) is radially unbounded (||x
x|| = V (x
x) ) and
If V (x
x) < 0, x = 0
V (x
then the equilibrium point x e = 0 is globally asymptotically stable.
Denition 2.7. (Invariant set) A set S in the phase space is said to be
invariant with respect to (2.6) if
x(0) S x(t) S t IR
14
2 Mathematical Preliminaries
x 1 = 2x2
x 2 = x2 + x1
x 1 = 2x1
x 2 = 2x2
The trajectories that originate in O converge to (0, 0). Now consider a Lyapunov candidate function.
x ) = x1 2 + x2 2
V (x
Its time derivative on the set O is
x) = 4x21 4x22
V (x
15
5
4
3
2
x2
1
0
1
2
3
4
5
5
0
x1
x) < x \ {0}
V (x
Hence the set O is positively invariant. Even though the origin is not stable
in the classical sense but it is stable in a hyperplane. This motivates us to
search for some weaker notion of stability which can be used to describe the
behaviour of the saddle point.
Let us now consider an another example. Consider the following nonsmooth system
x 1 = x1
x2
x 2 =
x1
x : x1 = 0, x2 = 0}. Please note that this set does not
Let us dene a set O = {x
contain the point (0, 0). If x (0) O , then the closed-loop system becomes
x 1 = x1
x 2 = 0
The trajectories originating in the set O converge to the point (0, 0) as time
gets large. This is another example that motivates us to dene a weaker notion
of stability. Furthermore, this example mimics the systems considered in this
thesis. Now we dene a weaker notion of stability called relative stability.
16
2 Mathematical Preliminaries
O
B
B O
respec-
17
x(t)|| k||x
x(0)||et t 0
x (0) B O = ||x
Remark 2.13. Please note that the equilibrium point should be a closure point
of the set O.
(2.9)
where = ab where a and b are odd and a < b. Note that the function x is
not Lipschitz. The solution of (2.9) can be written as
1
(2.10)
x(t) =
[(1 )t + C] 1 t T
0
tT
where
C
1
Note that x reaches zero in a nite-time T and the solution is well dened
in forward time. The detailed analysis of n-th order nite-time dierential
equations and nite-time stability has been presented in [11]. Now we state
nite-time stability of a dynamical system which admits a unique solution in
forward time.
C = [x(0)](1) and
T =
18
2 Mathematical Preliminaries
x)ui ,
gi (x
x = f (x
x) +
x(0) = x0
i=1
(2.11)
y = h(x
x)
19
x, u ) M IRm : f (x
x) +
(x
x)ui = 0 .
gi (x
i=1
(2.12)
where
z = x xe
v = u ue
and
m
A=
x)
x)
gi (x
f (x
+
ui
x
x
x
x
i=1
e
xe ,u
ue )
(x
x ) . . . gm (x
x )].
B = [g1 (x
xe , u e ) i
The system (2.11) is said to be linearly controllable at (x
rank = n
where
= B AB . . . An1 B .
The system (2.11) is said to be linearly controllable if the linearized system
xe , u e ) is controllable. This
(2.12) at any of the system equilibrium points (x
property however, does not hold for many underactuated mechanical systems.
x0 , T ) deThis motivates us to dene other notions of controllability. Let R V (x
note the set of reachable points in M from x 0 at time T > 0, using admissible
controls u (t) and such that the trajectories remain in the neighborhood V of
x0 for all t T . Furthermore, dene
(2.13)
x0 , T ) =
RV (x
0tT
x0 , t) , RV (x
x0 ) =
RV (x
t>0
x0 , t)
RV (x
20
2 Mathematical Preliminaries
Denition 2.18. Let x 0 M and int(U ) refer to the interior of the set U .
x0 )) = for any neighbourhood
1. is locally accessible from x 0 if int (RV (x
V . If this holds for any x 0 M then the system is called locally accessible.
x0 , T )) = for each
2. is locally strongly accessible from x 0 if int (RV (x
T > 0 and for any neighbourhood V .
x0 )) for any neighbour3. is locally controllable from x 0 if x 0 int (RV (x
hood V .
4. is small time locally controllable (STLC) from x0 if there exists T > 0 so
x0 , t)) for each t (0, T ] and for any neighbourhood
that x 0 int (RV (x
V.
x0 ) = M .
5. is globally controllable from x 0 if RV (x
Sucient conditions to check dierent notion of controllability is presented in
Appendix A.
2.6 Stabilizability
Stabilizability is a property which is related to the existence of a feedback
control that makes the closed-loop system stable. Here we present Brocketts
theorem [15] which gives a necessary condition for the existence of a C 1 feedback control law for asymptotic stability.
Theorem 2.19. Let
x, u )
x = f (x
xe , 0) = 0 and f (., .) is continuously dierand x IRn and u IRm . Let f (x
xe , 0). A necessary condition for the
entiable function in a neighborhood of (x
1
xe , 0) asymptotically stable is that
existence of a C control law which makes (x
1. the linearized system should have no uncontrollable modes associated with
eigenvalues whose real part is positive
xe , 0) such that for each N there
2. there exists a neighborhood N of (x
exists a control u (.) dened on [0, ) such that this control steers the
x, u ) from x = at t = 0 to x = xe at t = .
solution of x = f (x
3. the mapping
: IRn IRm IRn
x, u ) f (x
x, u ) should be onto an open set containing 0.
dened by : (x