Page 790 - The Mechatronics Handbook
P. 790

0066_Frame_C24  Page 38  Friday, January 18, 2002  5:38 PM





                       Lemma 24.4 If rank {ΓΓ ΓΓ [A,  C]}  =  k  <  n,  there exists a similarity transformation  T  such that with
                                          o
                             1
                            –
                                      1
                                     –
                       x =  T x, A =  T AT, C =  CT  , then C  and A  take the form
                                                A =  A o  0  ,    C =  [ C o  0]               (24.225)
                                                     A 21 A no
                              has dimension k and the pair (C o, A o)  is completely observable.
                       where A o
                         This result has a relevance similar to that of the controllability property and the associated decompo-
                       sition. To appreciate this, we apply the dual of Lemma 24.1 to express the (transformed) state and output
                       equations in partitioned form as


                                              x o t()  A o   0  x o t()  B o
                                              ˙
                                                    =                 +     u t()              (24.226)
                                              ˙ x no t()  A 21 A no  x no t()  B no
                                                               x o t()
                                                 y t() =             +  Du t()                 (24.227)
                                                        C o 0
                                                               x no t()

                         The above description reveals why one can be in trouble when trying to control a system using only
                                                                          .
                       the system output. The output has no information on the state  x no
                         The  observable subspace of a model is the space composed of all states generated through every
                       possible linear combination of the states in  . The stability of this subspace is determined by the location
                                                       x o
                                         .
                       of the eigenvalues of A o
                         The unobservable subspace of a model is the space composed of all states generated through every
                                                         . The stability of this subspace is determined by the location
                       possible linear combination of the states in x no
                                          .
                       of the eigenvalues of  A no
                         If the unobservable subspace is stable we say that the system is detectable.
                         A key feature of the descriptions (24.226) and (24.227) arises from the fact that the transfer function
                       is given by
                                                                   1
                                                                   –
                                                   H s() =  C 0 sIA o) B o +  D                (24.228)
                                                            (
                                                              –
                         Equation (24.228) says that the eigenvalues of the unobservable subspace do not belong to the set of
                       poles of the system transfer function. This implies that there is a cancellation of all poles corresponding
                                     –
                       to the roots of (sIA no  ).
                       Observability Canonical Form
                       There are also duals of the canonical forms given in Lemmas 24.2 and 24.3. For example, the dual of
                       Lemma 24.3 is:
                       Lemma 24.5 Consider a completely observable SISO system. Then there exists a similarity transformation
                       that converts the model to the observer canonical form:


                                                    – α n−1 1           b n−1
                                             x ˙ t() =       O    xt() +     ut()              (24.229)
                                                                1
                                                     – α 0  0   0        b 0

                                               yt() =  [ 10  …  0]xt() + Dut()                 (24.230)


                       ©2002 CRC Press LLC
   785   786   787   788   789   790   791   792   793   794   795