Sunteți pe pagina 1din 11

1

EML 6934, Spring 2010 Instructor: Dr. W. Dixon EXAM #3

RAMIN SHAMSHIRI UFID#: 9021-3353

Ramin Shamshiri

EML 6934, Exam #3

Due April.23.2010

2
I. Given the following dynamic system 5 + cos = 1 sin 2 + + 2 Problem 1

where (), (), (), () , is a positive constant defined as 0 < < 1 and () is an unknown function that satisfies the following inequality: () < 1 where 1 is a known positive constant. Assume that () and () are measureable, design a continuous controller that shows () () (or within some ultimately bounded region of ()) where = sin . Show a complete analysis including all signal chasing arguments to prove the control can be implemented. Solution: Let () = 5 + cos , then we can write the dynamic system as follow: 1 () = sin 2 + + 2 Error system We define a position tracking error (2) and filtered tracking error (3) as follow: () = () () = () + () (2) (3) (1)

where is a positive constant. We can see from (2) and (3) that if 0, then = which means 0. Also if is bounded, then , are also bounded. Open loop error system Differentiation (3) with respect to time and multiplying by () yields the open loop error system as given in (4): = () + () = () () () () + ()() Substituting for using the dynamic in (1): 1 () = sin 2 () + + () () + () 2 Let = +
1 2

(4)

(5)

sin 2

. = + + () (6)

where (, , ) is linear regression matrix (linear in the parameters) and is the unknown constant.

Ramin Shamshiri

EML 6934, Exam #3

Due April.23.2010

3
Adaptive control design We proceed with adaptive control design to achieve trajectory tracking task (() ()) because of the presence of unknown parameters, thus, the controller should be designed based on an estimate of the system model. From the dynamic in (6), we design the control as follow: = where is just a positive control gain and represents an estimate of the system parameters. Closed loop error system Substituting (7) in (6), yields the following closed loop error system = + + (8) (7)

where = is he difference between the actual and estimated parameters. Differentiation with respect to time gives = (since = 0). In order to design , we consider a continuously differentiable, positive-definite Lyapunov function candidate given by = 1 1 2 + 1 2 2 (9)

Differentiating (9) with respect to time and substituting for with yields; ( = 0 since it is only function of ). Substituting for with (8) ; = 1 = + + Now we can design as: = The expression for can then be find by integrating , ( = ). (12) (10) (11)

Stability analysis: Substituting the designed in the expression for given by (11) yields; = 2 + Since () < 1 , we can upper-bound (13) by K
2

(13)

+ 1

(14)

Signal chasing: If = 0, the expression in (14) reduces to K 2 . Since is a positive definite function and is negative semi definite, we can conclude that . In addition, since is a function of and , then and are also . Being bounded, we have , , thus , which means . Also since , , then , thus . Therefore, is continuously differentiable since is square integrable ( 2 ) and uniformly continuous, using the corollary to Barbalats lemma, we can prove that 0 as 0, hence we can conclude that using this controller, and in the absence of , the system is asymptotically stable. The input controller () in (7) is a function of bounded terms and is therefore bounded. In the presence of a bounded function disturbance, we can conclude a globally uniformly ultimately bounded (GUUB) stability result from (14). To calculate the ultimate bound, we set K 2 + 1 = 0, which yields > 1 .

Ramin Shamshiri

EML 6934, Exam #3

Due April.23.2010

4
High gain control design In this approach, we design the control input in such a way that it crushes the uncertainties in the system through high gain. This design leads to a high control effort. Using the dynamic in (6), we design the control input ()as: = where is a positive control gain. Substituting (15) in (6) yields the closed loop error system as follow: = + + (15.1) (15)

Lyapunov stability analysis: 1 Considering a continuously differentiable, positive-definite Lyapunov function candidate given by = 2 with its time derivative given by = , substituting for from (15.1) yields: = 2 + where = + and can be upper bounded by 1 + 2 + 3 bound as follow:
2 2 2

(16) , where = . Now we can upper

(17)

we can see that 2 dominates by large , and as long as this happens, () will be decreasing. But because we have 2 , the term 2 will eventually be dominated by 2 . In the other words, at some point, will be small enough that 2 will be dominated by , and since is a positive definite, radially unbounded and decrescent function, we conclude Uniformly Ultimately bounded result. This can be shown by letting = 1 + 2 and completing the square term in (17) as follow: 1 The inequality in (18) can be written as 21 1 2 . + 2 +
2 1

2 2 + 2 42 42

(18)

(19)

where =

2 1

and =

2 4 2

. The solution to (19) is () (0)e + . Therefore, we have UUB result using high

gain control design.

Ramin Shamshiri

EML 6934, Exam #3

Due April.23.2010

5
RISE control design According to class notes, since we have the bounded term () in our system, using high gain controller, we got UUB result. In general we will be limited to this result if we use a continuous controller. Another alternative is discontinuous controller such as SMC which is the only way to take care of linear disturbance. So the challenge if we can use a continuous controller to get Asymptotic result instead of UUB, that what we can get with RISE controller. Error system If we define the following error system 1 () = () () 2 () = 1 () + 1 1 () () = 2 () + 2 2 () (20) (21) (22)

where , 1 are just positive constants. We can see from (20),(21) and (22) that if 0, then 2 = 2 2 which means 2 , 2 0. Also if 2 0 then 1 , 2 0. If is bounded, then , are also bounded. Open loop error system Multiplying (22) with : = 2 + 2 2 = (1 + 1 1 ) + 2 2 = () () + ()(1 1 + 2 2 ) The motivation is that since the only way to get rid of the linear bound is to use the discontinuous function , but what is we could design the time derivative of the control as function. So we will be implementing the integral of the of the error which is continuous. Substituting for using the dynamic in (1), we can write as: 1 = sin 2 + + 5 cos + ()( + ) 2 Let = sin 2 cos , and then adding and subtracting in (24)
2 1

(23)

(24)

1 = sin 2 + + 5 cos + + + 2 1 1 = sin 2 + + 5 cos sin 2 + cos + 2 2 + +


1 1

(25)

Letting = sin 2 sin 2 + cos cos + + , then equation (25) can be 2 2 written as (26) = + + 5 + Control design We design = , where is our estimate for and is the RISE term, a control signal to be designed. Substituting this control in (26) yields = + 5 + = + 5 + where = . Differentiation of () with respect to time is + (), = 5 + + (28) (27)

Ramin Shamshiri

EML 6934, Exam #3

Due April.23.2010

6
We put all the state dependent terms in and we collect all the terms bounded by constant and put inside , so 1 = 5 + 2 2 = + Designing the RISE term (according to class notes):

(29) (30)

= + 1 2 + 1 2 0 +

( + 1) 2 3 + 2

(31)

The term + 1 2 is like a PD controller and 0 ( + 1) 2 3 + 2 is like Integral controller ( is some control gain), so we have the PID controller plus the integral of the 2 and all the magic happen because of the integral of sgn. Lyapunuv analysis 1 1 = + 2 2 2 2 1 = + + 2 ( 2 2 ) 2 Differentiating (31) with respect to time: = + 1 2 + + 1 2 2 + 2 = + 1 + 2 Substituting (29), (30) and (32) into (28) yields 1 = + + 2 + 1 2 2 We can upper-bound
1 2 2

(32)

(33)
2

and = 5 + =>

+ 5, where = 1

Design = 1 , where = cos + sin sin cos which is bounded. We consider = 2 0 2 0 0 0 , so, = 2 Let = , now we can put in out Lyapunuv function:

1 1 1 1 = 1 1 + 2 2 + + + 1 2 2 2 2 and 1 = 1 2 1 1 + 2 2 2 + 2 1 + + + + 2 2 2

(34)

(35)

Ramin Shamshiri

EML 6934, Exam #3

Due April.23.2010

7
= 1 2 1 1 1 2 2 2 + + 1 + 2

+ 2

(36)

We can upper-bound as follow:


2

. .

(37)

Using completing squares: ( 2 ) 4


2

(38)

Therefore, if we have Signal chasing:

2 4

> 0, then 0 and we get semi-global stability result.

We have > 0 and < 0 which means is bounded, so , , therefore which means 1 , 2 , .Also from the square integrable lemma, 2 , therefore 2 which means 1 , 2 , 2 . also implies that . Since and are both functions of measurable parameters, they are also measurable. All the terms in the control are measurable and bounded, therefore the controller is implementable

Ramin Shamshiri

EML 6934, Exam #3

Due April.23.2010

8
Problem 2 Given the same dynamics as in Problem #1, design a continuous controller that shows () () (or within some ultimately bounded region of ()) assuming that only () is measurable. Show a complete analysis including all signal chasing arguments to prove the control can be implemented. Solution: Output feedback design Since () is not measurable, we cannot define = + , because will not be measurable and cannot be used in our controller. Define + + (39)

where is called filter error and is defined through P-filter. Since is not measurable, its purpose is to help us with our stability analysis. P-filter + + 1 + 2 + 1 (40) (41)

where is the solution of . It can be seen that is a signal that we can use in out feedback, because it has which is measurable, also is measurable, so and are both implementable in the controller, however, is problematic. Taking derivative of (39) with respect to time and multiplying both sides by () yields: = () + () + () Substituting for from (2) and for () from the dynamic in (1), we have: = 1 sin 2 + + () + () + () 2 (42) (41)

Also, substituting for with ( ) and substituting for with ( + ) and the adding and subtracting yields: = sin 2 + + + + + + -
2 1 1 1

(43)

The motivation to add and subs tract is by looking at and sin 2 . We first let = sin , then 2 2 2 we have: , , = + . It should be noted that this adding and subtracting is called Desired Compensation Adaptation Law (DECAL). In this case, (44) can be written as: = () 2 + + + + + + - (44)

Now we need to figure out a way to deal with () 2 . Using the definition of and , we can expand and write it as: 2 = +
2

(45)

Ramin Shamshiri

EML 6934, Exam #3

Due April.23.2010

9
Substituting (45) into (46): = + + + + + + + -
2

(46)

Expanding the expression in (46) and substituting for with ( ) and for with 2 + 1 + 1 , yields: = + + + + + () + = + + + + + + 1 + 2 + 1 + Now we substituting for in (47) with + and then expand: = + + + + + + 1 ( + ) + 2 + 1 + Simplifying (48) yields: = + + + + () + 2 + 2 + + = + + + + + + + + = + + + + + + = +
2 2 2 2 2 2 2

(47)

(48)

(49)

(50)

(51)

+ + + (1 ) 2 +

(52)

The term (1 ) is a good term as long as we pick > 1 , because it will eventually give us 2 , where = ( 1) , but 2 is a bad term and we should upper-bound it. In fact, we through all the bad terms in (52) into a bucket called and then upper bound by 1 + 2 2 where = 3 . (The term 2 2 in the 2 upper-bound comes from 2 ). Expanding + : +
2 2 2 = ( 2 2 2 + 2 + 2 + 2 2 + 2 + )

(53)

Ramin Shamshiri

EML 6934, Exam #3

Due April.23.2010

10
It should also be noted that can be upper-bounded by can write as: = + + + 1 + + Designing the controller as: = + ? At this point we need to work with Lyapunuv analysis to get more intuition and to know what to put in ? . We know that we ultimately need something like 2 in the Lyapunuv. = + + 1 + + + ? Considering a continuously differentiable, positive-definite Lyapunov function candidate as follow: 1 1 1 1 2 = 2 + 1 + 2 + 2 2 2 2 (56) (55) < and can be upper bounded by 1. So we
2 1

(54)

The reason we put and in the Lyapunuv is that we have = + + which does not mean that if , then and are also . So since we dont know that the elements of are bounded, we put them in the stability analysis. Taking time derivative of (56), = 1 + + + 1 + + 2 + 1 = + + 1 + + + ? 1 + + + 1 + + 2 + 1
Design = to cancel . Then can be found using integration by part: ( + ) + t 0

(57)

=
0

Integration by part for


t

t , 0 t

T = Yd e

d ()

For all these terms, , , are all known and we can use numerical integration, so we can implement this adaptive law without velocity only because this desired trajectory doesnt depend on . Expanding (57): = 2 + 1 2 + + + ? 2 + + + + + 2 + + + ( 2 + ) (58)

It can be seen that 2 is a good term. So we need to put + into ? in order to cancel with and with . In this case, the control input will become: = +

Ramin Shamshiri

EML 6934, Exam #3

Due April.23.2010

11
And hence becomes:
2 = 2 + 1 2 + + 2

(59)

At this point, we only need to take care of the term. In fact, is like out trash can, it includes all the terms that we couldnt handle. We can upper-bound as follow: = 1 2
2

+ 6 1

+ 1 + 1 + 2

(60)

Once again, it should be noted that we need to pick the control gain > 1. If we let = 1, then we can write (60) as: = 1 2
2

+ 1 + 1 + 2

(61)

Since is just a control gain, we can let = 1 + 1 1 + 2 2 . This is to make stability analysis clean. 1 and 2 are conservative upper-bond on the disturbance terms that are in . Also 1 and 2 are just some number that we pick. If we proceed with completing the squares we will eventually se that is always decreasing or zero, so is always bounded by its initial condition. So if a function is positive and its time derivative is either decreasing or zero, it will never grow beyond the initial condition. Therefore will be bounded for some inequality condition which yields a semi-global stability result. It means if we know the initial condition then we can pick our gain big enough.

Ramin Shamshiri

EML 6934, Exam #3

Due April.23.2010

S-ar putea să vă placă și