Page 184 - Adaptive Identification and Control of Uncertain Systems with Nonsmooth Dynamics
P. 184

Adaptive Dynamic Surface Output Feedback Control of Pure-Feedback Systems  181


                               Consequently, the pure-feedback system (11.1) can be redescribed as

                                        ⎧
                                               = z i+1 ,      i = 1,...,n − 1
                                        ⎪ ˙ z i
                                        ⎨
                                                                λ n
                                           ˙ z n  = a n (¯x n ) + b n (¯x n ,u )u     (11.12)
                                            y = z 1 .
                                        ⎪
                                        ⎩
                               To facilitate the controller design, the function b n (¯x n ,u ) in (11.12)
                                                                                λ n
                            is assumed to be positive and bounded satisfying 0 < b 1 < b n (¯x n ,u )< b 2,
                                                                                     λ n
                            where b 1 and b 2 are positive constants. It should be noted that this condition
                            has been widely used in the literature [1–3] as a necessary condition to
                            guarantee the controllability of (11.1).
                               Substituting (11.3)into(11.12)yields
                                    ⎧
                                           =z i+1 , i = 1,...,n − 1
                                    ⎪ ˙ z i
                                    ⎨
                                                          λ n
                                                                         λ n
                                       ˙ z n  =a n (¯x n ) + b n (¯x n ,u )d(t)v + b n (¯x n ,u )ρ(t)  (11.13)
                                    ⎪
                                    ⎩
                                        y = z 1
                               Moreover, we can rewrite system (11.13) in the form of
                                           ⎧
                                                  = z i+1 , i = 1,...,n − 1
                                           ⎪ ˙ z i
                                           ⎪
                                           ⎪
                                                                 λ n
                                           ⎨      = a(¯x n ) + b(¯x n ,u )v
                                              ˙ z n
                                                                                      (11.14)
                                                  = F(¯x n ,u ,v) + b 0v
                                                           λ n
                                           ⎪
                                           ⎪
                                           ⎪
                                           ⎩
                                               y = z 1
                            where a(¯x n ) = a n (¯x n ) + b n (¯x n ,u )ρ(t), b(¯x n ,u ) = b n (¯x n ,u )d(t),and F(¯x n ,
                                                                   λ n
                                                      λ n
                                                                              λ n
                                                 λ n
                                                                                       λ n
                             λ n
                            u ,v) = a n (¯x n ) + (b(¯x n ,u ) −b 0 )v,and b 0 is the estimation of b(¯x n ,u ) and
                            can be obtained based on the prior modeling knowledge. Considering the
                            positive boundedness of b n (¯x n ,u ), we know that b(¯x n ,u ) is also positive
                                                                             λ n
                                                        λ n
                                                                                          ¯
                            and bounded with the lower boundary b = b 1   and upper boundary b =
                            b 2 (d l1 + d r1 ).
                               As shownin(11.14), the original system (11.14)isnowrewrittenas
                            a canonical form, which is more suitable for control design. However, in
                            the proposed coordinate transformation, the system states z i ,i = 2,··· ,n
                            are not available though the original system state x i ,i = 1,··· ,n may be
                                                                           λ n
                            measurable. Moreover, the lumped dynamics F(¯x n ,u ,v) in (11.14)are
                            unknown. Hence, they should be specifically addressed in the following
                            control design. It is noted that ESO can be used to cope with these two
                            issues simultaneously. Hence, the following subsection will introduce the
                            design of an ESO.
   179   180   181   182   183   184   185   186   187   188   189