Page 936 - The Mechatronics Handbook
P. 936

0066_Frame_C30  Page 47  Thursday, January 10, 2002  4:45 PM









                       Estimator (Filter) Loop Shaping
                       If one selects

                                                    B 1 =  [ L 0 n × ]                         (30.260)
                                                               n
                                                                y
                                                   D 21 =  [ 0 n ×  mIn × ]                    (30.261)
                                                                      n
                                                            y  n  u  y  y
                                                    C 2 =  C                                   (30.262)

                                                    Θ =  mI n ×                                (30.263)
                                                            y  n y
                       then  B 1 D 21 =  0   and we have
                              T

                                                                 T1
                                                         H f =  YC 2  ---                      (30.264)
                                                                  m

                       where Y ≥ 0 is the unique (at least) positive semi-definite solution of the FARE:

                                                   T        T     T  – 1
                                                YA +  AY +  LL –  YC Θ CY =  0                 (30.265)
                       Given this, the following KFDE may be derived from the FARE:


                                                                  1
                                                                              1
                                          (
                                              [
                                                                                    (
                                                     (
                                   [ I +  G KF jw)] I + G KF jw)] H  =  I +  -------G FOL jw(  ) -------G FOL jw)  H  (30.266)
                                                                  m           m
                       where
                                                             (
                                                     G FOL =  CsI A) L                         (30.267)
                                                                   −1
                                                                –
                                                             (
                                                      G KF =  CsI A) H f                       (30.268)
                                                                   −1
                                                                –
                       Given this, the loop shaping ideas discussed earlier are applicable. A designer may use the matrix L and
                       the scalar µ > 0 to shape G FOL  in an effort to get a desirable loop G KF  . The matrix L, specifically, may be
                       used to match singular values at low frequencies, high frequencies, all frequencies, etc. Assuming that
                       (A, C) is detectable and (A, L) has no imaginary modes that are uncontrollable, then a stabilizing solution
                       is guaranteed to exist. Moreover, the resulting  G KF  loop will possess nominal sensitivity and stability
                       robustness properties—a consequence of the KFDE. The resulting filter (output injection) gain matrix
                       H f  may be used within an estimator (feedback) loop, a modified estimator (feedback) loop, or within a
                       model-based compensator.



                       30.6 Summary
                                                                                     2
                       This chapter has presented a general framework for control system design via H  optimization. While
                       the focus has been on continuous time LTI systems, the methods are very  flexible and have wide
                       application. They may be used to design constant gain state feedback control laws, constant gain state
                       estimators, dynamic output feedback controllers, and much more. Weighting functions are easily accom-
                       modated within the generalized plant framework presented. Such functions may be used to achieve closed
                       loop design objectives. All of the ideas presented may be extended with subtle (all be it very important)


                       ©2002 CRC Press LLC
   931   932   933   934   935   936   937   938   939   940   941