Page 162 - Neural Network Modeling and Identification of Dynamical Systems
P. 162

152             4. NEURAL NETWORK BLACK BOX MODELING OF AIRCRAFT CONTROLLED MOTION



































                         FIGURE 4.13 The results of a computational experiment for the MRAC-type control system with compensator (F-16 air-
                         craft, flight mode with a test speed V ind = 600 km/h). Adaptation to the change in the dynamics of the controlled object: the
                         shift of the centering by 10% back (t = 30 sec); 50% decrease in the effectiveness of the control (t = 50 sec). Here α is angle of
                         attack, deg; E α is tracking error for the reference angle of attack, deg; δ e is the deflection angle of the stabilizer, deg; t is time,
                         sec.


                         ter the second failure, the stability is preserved,  Another series of computational experiments
                         but the error values become quite significant  was performed to assess the significance of the
                         (E α ≈±0.5 deg).                             factors excluded from consideration in the tests
                            In Figs. A.18–A.21 we demonstrate the re-  described above. This additional series was also
                         sults of similar computational experiments for  carried out for the hypersonic research vehicle
                         the NASP hypersonic aircraft.                X-43, which cruised at Mach number M = 6.
                            In the computational experiments considered  In this series of experiments, the motion
                         above, we have used the model of the aircraft  model has the form (4.3), i.e., the interaction of
                         motion of the form (4.2) with a single control  the angle of attack and thrust is taken into ac-
                         variable δ e, act . The relation between the angle of  count. Besides, we introduce the engine thrust
                         attack α and the thrust T cr in this model was in-  control δ th in addition to control for the angle of
                         troduced through the values of the pitching mo-  attack (command signal δ e, act ) in order to coun-
                         ment coefficient C m (α,T cr ). Additional effects  teract errors in the angle of attack and tangential
                         from the thrust on the angle of attack and the  g-load.
                         angle of attack on the thrust were not taken into  Various combinations of the reference sig-
                         account; thrust control was also not introduced.  nals used for both channels were considered, as
   157   158   159   160   161   162   163   164   165   166   167