Page 999 - The Mechatronics Handbook
P. 999

0066_frame_Ch33.fm  Page 23  Wednesday, January 9, 2002  8:00 PM









                       3rd set: gw = 0.5; ga = 0.07; gb = 0.03; nepoc = 200; nesant = 21; niter = 200; threshold error = 0.001;
                       Vmax = 0.047515.



















                       FIGURE 33.32  Membership functions after learning for the variable x 2 .




















                       FIGURE 33.33  The surface obtained after learning.

                       33.8     Software Considerations

                       The MM of electrohydraulic axis studied in this thesis is supported by a physical installation existing in
                       the mechatronics laboratory of UAS-Konstanz (see Fig. 33.1b). Two variants of nonlinear MM are set
                       forth in Section 33.3 and add in static and dynamic nonlinearities that arise in the function of electro-
                       hydraulic axis [23, 58].
                         The MM of hydraulic drive presented in the structure of ROBI_3 was implemented in SIMULINK in
                       order to study the dynamic behavior of the axis [26, 27]. The extended variant of MM hydraulic axis
                       was done taking into account the relative motion of the constituent parts of this servodrive.
                         The neural and neuro-fuzzy controller (Takagi-Sugeno) was developed in Borland C++ and imple-
                       mented in SIMULINK for controlling the electrohydraulic axis. SIMULINK offers the user a FUZZY
                       LOGIC library that allows the designing and modeling of SUGENO or MAMDANI fuzzy inference
                       systems. The lack of a dedicated software to design neuro-fuzzy controllers persuaded the implementation
                       of such a controller in C++ and afterwards the use of it in SIMULINK [26,27,28].
                         The support for simulation, SIMULINK 2.1 and MATLAB 5.2 (under Windows), offers solutions to
                       implement our controllers as modules and corresponding icon in a specialized toolbox. In our experi-
                       ments, we used the facility offered by S-functions and C MEX in conjunction with Borland C++ 5.0 to
                       compile them. We have chosen the C S-function because of the speed necessary to process the information
                       in our block that implement the controller. The block that implements the controller has two inputs
                       (even three inputs are available, though the adaptation process is more complicated) and one output.


                      ©2002 CRC Press LLC
   994   995   996   997   998   999   1000   1001   1002   1003   1004