Feedback Linearization Continued: Consider The Control Affine SISO System

Lecture 5 Feedback linearization continued
Consider the control ane SISO system:
x = f (x) + g (x)u y = h(x)
(1)
In the previous lecture we saw how if the system has relative degree r < n around a point xo, it can represented in its normal form:
1 = 2 2 = 3 . . . r1 = r r = b(, ) + a(, )u = q (, ) y = 1 where

1 1 (, )) a(, ) = Lg Lr h ( f 1 b(, ) = Lr f h( (, ))
(2)
(3)
x = 1(, )
(4)
We also saw that it is possible to inputoutput linearise the system by using the feedback law: 1 u= (v b(, )) a(, ) (5)
So that the resulting system from the reference input v to the output y = 1 is linear. The normal form is given by: 1 = 2 2 = 3 . . . r1 = r r = v = q (, ) y = 1
(6)
Notice, however, that the non-observable internal dynamics = q (, ) are possibly nonlinear, so that the system has not been fully linearised by the feedback law. In addition the stability of the internal dynamics is important as the linearising control signal depends on .
3
Zero dynamics
The stability of the internal dynamics is closely related with the stability of the system when the output and its derivatives are forced to zero. If the output of the system is restricted to be zero, y (t) = h(x) = 1 = 0, , then it follows that: j = y (j 1) = 0, j = 1, . . . , r (7)
so that = 0. If, in addition, we restrict the rth derivative of the output to be zero, then it follows that the input is constrained to be the solution of the equation: 0 = b(0, ) + a(0, (t))u(t) (8)
Notice that by denition of relative degree a(, ) is known to be non-zero around x0. If f (x0) = 0, h(x0) = 0, and k (x0) = 0, k = r + 1, . . . , n, then the normal form is dened around (, ) = (0, 0). In this case, it is reasonable to assume that a(0, (t)) is non-zero for small values of (t). Solving Equation 8 for u gives the output-zeroing input: u(t) = b(0, (t)) a(0, (t)) (9)
where (t) denotes the solution of the dierential equation: (t) = q (0, (t)), (0) = 0 (10)
This equation represents the zero dynamics of the system.
The zero dynamics of a linear system with r < n are given by: = Q (11)
The eigenvalues of matrix Q are equal to the zeros of the transfer function of the system G(s) = N (s)/D(s).
Local asymptotic stabilization

Consider the feedback law: 1 u= (v b(, ) 01 r1r ) a(, ) (12) and assume that v = 0. When this feeedback law is applied to system (1), it results in the following normal form: = A = q (, ) y = x1 with 0 1 0 0 0 1 A= . . . 0 1 2
(13)
... 0 ... 0 ... 1 . . . r1
(14)
Proposition 1. Suppose that
1. the equilibrium = 0 of the zero dynamics of the SISO system given by Equation 1 is locally asymptotically stable
2. all eigenvalues of matrix A dened by Equation 14 have negative real parts
then the feedback law in Equation 12 with v = 0 provides an equilibrium (, ) = (0, 0) that is locally asymptotically stable. Proof. See Isidori (1995).
Bounded tracking and inversion

It is easy to modify the feedback law given by Equation 12 so that the objective is to track a desired output yR (t).
u= 1 (r ) r 1) (yR (t) b(, ) 0 e0 r1 e( ) 0 a(, ) (15) i i ei 0(t) = Lf h(x) yR (t)
where (16)
The problem of tracking a reference output trajectory is closely related to the problem of dynamic inversion.
Proposition 2. Suppose that the following conditions are satised: 1. the equilibrium = 0 of the zero dynamics of the SISO system given by Equation 1 is globally exponentially stable 2. All eigenvalues of matrix A dened by Equation 14 have negative real parts, 3. The mapping q (, ) is globally Lipschitz in . 4. The reference output yR (t) and all its derivatives are bounded. then the feedback law in Equation 15 results in bounded tracking, this is e0 and its rst r derivatives tend to zero asymptotically and the state x is bounded. Proof. See Sastry (1999).
10
Example 1. Consider the system
x 1 = x2 + u x 2 = u y = x1
(17)
where y (t) is required to track a desired output yR (t). Dierentiating of the output: y =x 1 = x2 + u (18) we note that it contains explicitly u. (Hence the system has a relative degree r = 1.) The control law: u = x2 + y R (y yR ) yields the tracking error equation: e 0 + e0 = 0 x 2 + x2 = y R e0(t) (20) (19)
where e0 = y yR , and the internal dynamics: (21)

11
We see from these equations that since the error equation is stable, then y (t) tends to yR (t), so does y tends to y R . Hence, the internal dynamics are stable and x2 remains bounded and so does the control input u. The zero dynamics, this is the internal dynamics of the system with the restriction that the output and its rst r time derivatives are zero, are given by: x 2 + x2 = 0 which is stable. Notice that for a slightly dierent system: x 1 = x2 + u x 2 = u y = x1 the zero dynamics are unstable (prove it).
12
(22)
(23)
In the Laplace domain this system can be expressed as follows: 1 (x2(s) + u(s)) s 1 u(s) + u(s) = s s 1 1 = + 1 u(s) s s s1 = u(s) 2 s So that the transfer function: y (s) = x1(s) = y (s) s1 = u(s) s2
(24)
(25)
has a zero on the right hand side of the complex plane. What happens when we try to invert a system with a right hand side zero?
13
Example 2. Consider the dynamic model of a single link manipulator: x 1 = x2 x 2 = a sin x1 bx2 + cu y = x1 This system has relative degree r = 2 (hence no zero dynamics) and is already in normal form. We want to design a linearising feedback to track a reference signal yR (t), with bounded derivatives y R and y R . Taking: e0 = x1 yR (t) e 0 = x 1 y R (t) = x2 y R (t) (27) (26)
The feedback law is given by Equation 15: u= 1 ( yR (t) + (a sin x1 + bx2) 0e0 1e 0 ) c (28)
14
Figure 2: Inverse tracking linearising control of the single link manipulator
15
Figure 2: Internal diagram of the inverse tracking linearising controller.
16
1.5
0.5 y(t), y (t)
0.5
1.5
3 t
Figure 2: Simulation results for a = c = 10, b = 1, 0 = 400, 1 = 20, x(0) = [0, 0]T . Solid line: yR (t) and y (t) for the case of perfect model and initial conditions (in this case the tracking is perfect). Discontinuous line: y (t) for the case of oset initial conditions x(0) = [0.5, 1]T and perfect model (in this case the tracking is asymptotic). Dotted line: y (t) for the case of model mismatch: bm = 0.5, cm = 5, which corresponds to doubling the mass.
17
Exact input-to-state linearisation

In the case when r = n (the system has full relative degree) then there are no internal or zero dynamics, and the feedback laws given by Equations 5 and 12 provide a full inputto-state linearisation. In particular, using the feedback law given by Equation 12, the normal form of the system is given by: = A + v y = 1 where A is given by (14). (29)
18
Example 3 (Isidori, 1995). Consider a one link robot arm whose rotary motion about one end is controlled by means of an elastically coupled actuator.
Figure 1: Single link robot arm with joint elasticity.
Let q1 be the angular position of the actuator shaft and q2 be the angular position of the link. The actuator equation can be written as follows: K q J1 q 1 + F1q 1 + q2 1 = T (30) N N
19
where J1 is the inertia, F1 is the friction constant, K is the elasticity constant, N is the transmission gear ratio and T is the torque produced at the actuator axis. The link equation is as follows: q J2 q 2 + F2q 2 + K q2 1 + mgd cos q2 = 0 N (31) where m and d represent the mass and position of the centre of gravity of the link. Dene the state vector as follows: q1 q x = 2 q 1 q 2

(32)
20
Then the system can be represented as a control ane system with:

f (x) =
K x + K x F1 x J1 N 2 J1 3 J1 N 2 1 K x K x mgd cos x F2 x 2 J2 N 1 J2 2 J2 J2 4
x3 x4
0 g (x) = 1 J1
(33)
(34)
The input is u = T and the output is the angular position of the link: y = h(x) = x2 (35)
21
Calculate the following Lie derivatives: Lf h(x) = f2(x) = x4 L2 f h(x) = f4 (x) f4 f4 f4 3 Lf h(x) = x3 + x4 + f4(x) x1 x2 x4 K K mgd = x3 + + sin x2 x4 J2 N J2 J2 F K K mgd F 2 x1 x2 cos x2 2 x4 J2 J2 N J2 J2 J2 (36) Since f4(x) does not depend on x3, we have: Lg h(x) = Lg Lf h(x) = Lg L2 f h(x) = 0 (37) 1 K 3 3 Lg Lf h(x) = Lf h(x) = J1 x3 J1J2N Therefore the system has a full relative degree r = 4 = n at each point xo of the state space.
22
Hence we can fully linearise this system using, for example, the following control law: 1 4h(x) u(t) = v ( t ) L f Lg L3 h ( x ) f J J N v (t) L4 = 1 2 f h(x) K
(38)
23

Feedback Linearization Continued: Consider The Control Affine SISO System

Uploaded by

Copyright:

Available Formats

Feedback Linearization Continued: Consider The Control Affine SISO System

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Feedback Linearization Continued: Consider The Control Affine SISO System

Uploaded by

Copyright:

Available Formats

Lecture 5 Feedback linearization continued

Consider the control ane SISO system:

x = f (x) + g (x)u y = h(x)

1 = 2 2 = 3 . . . r1 = r r = b(, ) + a(, )u = q (, ) y = 1 where

This equation represents the zero dynamics of the system.

Local asymptotic stabilization

... 0 ... 0 ... 1 . . . r1

Proposition 1. Suppose that

2. all eigenvalues of matrix A dened by Equation 14 have negative real parts

Bounded tracking and inversion

Example 1. Consider the system

where e0 = y yR , and the internal dynamics: (21)

Figure 2: Inverse tracking linearising control of the single link manipulator

Figure 2: Internal diagram of the inverse tracking linearising controller.

0.5 y(t), y (t)

Exact input-to-state linearisation

Figure 1: Single link robot arm with joint elasticity.

Then the system can be represented as a control ane system with:

You might also like