Nonlinear Control Systems - Lecture 24

  Output Tracking   Model Matching   Disturbance Decoupling
Perfect Output Tracking: Consider the problem of reproducing the reference output ¡£¢£¤¦¥¨§ . This roblem tries to find pairs of initial states, ©
, and input functions

) ¢  . So from the system’s normal form.  ¡ ¢ and we see that the input satisfies ¢  ¡ ¢  ¤¦¥¨§213¤ ) ¢£¤¦¥¨§54¨6¤¦¥¨§§87(9:¤ ) ¢. ¦  ¤¦¥¨§¡ ¢  ¤¦¥¨§ for all ¥ and !#"%$&"(' . + + ¡ ¢  ¨ . 0 ¢ .¤¦¥¨§54¨6¤¦¥¨§§ . / *+ ¡ ¢ . . . ¤ ¥¨§ defined for all ¥ in a neighborhood of ¥ such that the output ¡¤¦¥¨§¡¢¤¦¥¨§ . Let -.

 ON < M ONP < S R RTR  <   ¢  O O¢P <  6  ¡  We choose OQ R RSR T O¢ 1¤ ) ¨ 4 6U§V7(9:¤ ) 46>§ . ¤¦¥¨§ where Again consider the < system in normal form.

=>¤ ) ¨ 4 6U§   ¢ ]^ ¦ ¤A  B_¡ ¢  .

 9 ¤ ) 4¨6U§XW ! ¢  BY1¤ ) 4¨6U§V7Z¡ ¢  B [  ¨ §a` where 4 O4 RTRSR 4 ¢  are real numbers. ^ ^ ^ Define an error b ¤¦¥¨§ < 6 ¤ ¥¨§=>¤ ) ¢£¤ ¥¨§?4¨6¤ ¥¨§¨§   ¨ ¤¦¥¨§ for !e"f$g"h' (by construction). we Since Od@¡  can rewrite our control as ¢  ¨ ) 4¨6U§V7Z¡ ¢  ¢  B [ !   b   Y B 3 1 ¤ ` .

Note also that we can express .¢ at the output. 9:¤ ) 4¨6U§XW i\j^ Note that if b ¤¦¥¨§2 for all ¥ then the above control reduces to the same control to force perfect tracking ¡.

in the original coordinates as ¢ ¦   ¨ k ¢m n 7D¡ ¢  ¢  B [ !  B  po k m  q ` n B_¡ ¢  .

6 ¤ . klOk m  ¢ ¨ n ¤ § W ^ ]  8 \  © < b ¤¦¥¨§¡¤¦¥¨§CBc¡£¢¤ ¥¨§ ¤ ¥¨§ So if the output ¡¤¦¥¨§ is to ) track ¡¢ then the ) ¢£¤ . then the input to enforce perfect tracking is . Once we’ve selected 6 @6¤A§ .§ exactly.  # §  initial system state must be where ¤ § can be chosen arbitrarily.

¤ ¥¨§ ¡ ¢ ¢  ¤¦¥¨§CBD13¤ ) ¢ ¤¦¥¨§54¨6¤¦¥¨§§ 9:¤ ) ¢. and output . Note that the above two equations represent a system ) with input ¢ .¤¦¥¨§54¨6¤¦¥¨§§ < the solutution to where 6¤¦¥¨§ denotes 6¤¦¥¨§E =>¤ ) ¢ ¤¦¥¨§54¨6¤¦¥¨§§ 6¤ §F 6  ¢ yields  Using < this control in our equation for ¨ ¢  ¢ ¨   b  O¢p¡Jr¡ ¢  B ¢  B RTRSR B  b  B b ^ ^ ^ which implies that ¢ ¢ ¨ s b  7 ¢  b  7 ^ RSRTR 7 ^  b  ¨ 7 The preceding dynamics correspond to the dynamics describing the “internal” behavior of the system when initial state and input have been chosen to constraint ¡:¤ ¥¨§ to follow ¡ ¢ ¤¦¥¨§ . state 6 .

¤¦¥¨§ . whereas our original system in normal form is a system with input .

¤¦¥¨§ . so we now consider tracking to ¡ ¢ when we aren’t free to specify the internal state. In this case we seek . ) ¢ and output . setting 6 to a specified value is impossible. Asymptotic Output Tracking: In practice. So the preceding system is like an inverse to our original system. state 6 .

Asymptotic Model Matching: Sometimes the reference outut is the output of a reference model which is deriven by some input t . ^ b So the error function satisfies a linear differential equation of order ' whose coefficents can be arbitrarily chosen to force asymptotic output tracking. < a linea model. As an eaxmple consider v  w v ( 7 x t v ¡£¢P y . So we can pose the problem of finding a feedback control that causes ¡cGu¡¢ for any t and any initial condition. We call this asymptotic model matching. ¤ ¥¨§ that forces ¡¤¦¥¨§HGI¡¢£¤¦¥¨§ as ¥JGLK . This is the problem of asymptotic tracking.

To solve the asymptotic model matching problem we might try the same input as before. !  o B k ¢m n 7D¡ ¢  ¢   .

Then for our linear system ¢  N y#xf2y#wzxf RTRSR |y#w xf which implies that ]  ¡ ¢  y w v ¦ # ¤ ¥¨§ ¤  } " $"}' B ¢    ¢ ¢ ¡ ~  y w v # ¤ ¥¨§87%y#w x t If we rewerite the state euations choosing the same coordinates used for the normal form. we see that Ž   Ž   Ž © n ©   Ž ¥ Ž ¥  Ž ¥ k  m © n 7 kl n  © 7 kŒ n  . k l k ¢mL n ¢  ¨ ¦   zo k m B [ q ` n Bc¡ ¢  iV^ where we replace ¡¢ with outputs of our model so that ]    ¡ ¢  ¤¦¥¨§F y#w v ¤ ¥¨§87%y#w x t 7 RSRTR   N{  ¨ 7#y#wYx t  ¤¦¥¨§87(y#x t  ¤¦¥¨§ But this requires differentiating the input t which in practice makes the system very sensitive to noise on the input signal t . But what if we suppose our model has a relative degree larger or equal to ' (relatice degree of the original system).

.

k m n  N  A similar situation happens for the oter normal states N through O¢  . So we get< ONP  Q  < RSRTR  RSRSR O¢   ¢£¤¦¥¨§ < For  ¢ ’s state equation we get ¢  ¢ k  ¢  m n 7 k l k mLn .

 k Œ k ¢mL n  . ¢ ¢ ¢  !   B k m n 7%y#w v ¤¦¥¨§j7(y#w x t . The remaining equabecause we required 6 tions for will have t in them. So our system in < normal form is <    N ONP OQ ! § Replacing this into our control yields.

Disturbance Decoupling Problem: Consider a system of the form <  ƒ ¤ © §V7D„:¤ © § .¢ for any t driving our special reference model. klOk ¢mL nD ¢ ¦  ¦  v   o k m  B [ q ` n B€y#w ] ^ By this construction we can guarantee ¡_G‚¡.

7†…8¤ © § t C © ¡  n ¤© § where t is a disturbance. We want to examine conditions under which there exists a feedback law. < RSRTR   ¢  < M  ¢<  6  ¡  RTRSR  ¢ 1¤ ) ¨ 4 6U§V7(9:¤ )  4 6>§ .

) ) =>¤ ¨ 4 6U§V7%‘¤  4 6U§ t   Suppose we use the following state feedback law. Š 13¤ ) 46U§ .

<    N  N   Q < S R RTR   ¢  < M  ¢<  6  Š RTRSR  ¢ =>¤ ) 4¨6U§V7%‘¤ ) 46U§ t . ’B 9:¤ ) 46>§ 7 9:¤ )  4 6U§ < Then these equations become.

‡ˆ¤ © §87Z‰¤ © §aŠ that results in a closed loop system whose output ¡ is independent (i..e. decoupled) from the disturbance t . Let the system have relative degree ' at ©‹ and suppose the vector field …j¤ © § is such that kŒ3k m n ¤ § © for all  "($"%' B ! and all © near ©‹ . . This problem is called the disturbance decoupling problem. k Œ k m n  So the condition is sufficient for disturbance decoupling. in which we easily see that t is decoupled from the output ¡“ . It is also suppose to be possible to show that this condition is necessary too. We’ll look ath the normal form of the system.