Page 103 - Adaptive Identification and Control of Uncertain Systems with Nonsmooth Dynamics
P. 103

96   Adaptive Identification and Control of Uncertain Systems with Non-smooth Dynamics


                        tion f (x) can be represented by

                                                    N

                                         f (x) = p 0 +  p r σ r (0,x − α r ,β r − α r )  (6.2)
                                                   r=1
                        where α r and β r are the lower and upper boundaries of x confined in the
                        r-th subinterval, respectively. σ r (0,x−α r ,β r −α r ) is the basic function which
                        is given by

                                        σ r (a,b,c) = max(a,min(b,c)) ∀a,b,c ∈ R     (6.3)
                        and p r (r = 0,...,N) is the unknown coefficient that can be determined
                        by the least squares. It has been proven in [13] that any piecewise linear
                        function can be represented by (6.2).
                           The use of (6.2) to approximate any piecewise linear function is on the
                        precondition of partitioning the domain of x into finite closed subinter-
                        vals whose interiors do not overlap (see Lemma 1 in [13]). Moreover, the
                        boundaries α r , β r satisfy: 1) α r <β r ;2) β r = α r + 1. Then it is noted that
                        the basic function σ r (0,x − α r ,β r − α r ) is actually a special piecewise linear
                        function which can be decomposed as the difference of two simpler piece-
                        wise linear function, i.e., α r (0,x − α r ,β r − α r ) = max(0,x − α r ) − max(0,x −
                        max(α r ,β r )). In fact, the coefficient p r is the slope of the local linear func-
                        tion given by α r = x − α r when α r ≤ x <β r . Furthermore, it can be seen
                        that α r = 0when x <α r or α r = β r − α r which is a constant for x ≥ β r .


                        6.3.2 DPPR Modeling of Friction
                        The expression (6.2) provides a potential way for model-free parameteri-
                        zation of frictions since the friction dynamics appear sectionally and nearly
                        linear in the high speed operation region [6]. However, in the low veloc-
                        ity operation region, the Stribeck effect plus the Coulomb friction force
                        mainly contributes to the friction, which makes the friction highly non-
                        linear and non-smoothing in the low speed especially operation near zero
                        crossings. In addition, the static friction has a jumping behavior at zero
                        velocity where the direction of motion may change.
                           In order to solve the approximation of Stribeck effect and jumping be-
                        havior at zero velocity, two additional terms are introduced into (6.2)to
                        describe the friction dynamics:
                        1) A jump term h 1 (v) related to the maximum static force is introduced
                            into (6.2) to represent the reversal behavior of friction force when the
                            motion direction is changed;
   98   99   100   101   102   103   104   105   106   107   108