Page 158 - Neural Network Modeling and Identification of Dynamical Systems

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

FIGURE 4.11 The results of a computational experiment for model reference adaptive control system without a compen-
sator (hypersonic research vehicle X-43, flight regime M = 6). Here α is angle of attack, deg; E α is the angle of attack tracking
error, deg; α ref is the angle of attack reference signal, deg; δ e, act is command signal for the elevator drive, deg; t is time, sec;
the solid line is the real plant; the dotted line is the ANN model.

The parameters of this reference model are order two). For this reason the oscillatory-
the eigenfrequency ω rm , the coefficient of rela- type unit is a basis of the reference model.
tive damping ζ rm ,and thetimeconstantofthe 2. The controlled object is an airplane with an
“prefilter” T pf . Formally the aperiodic-type unit actuator. Therefore a purely oscillatory refer-
in the reference model is not a prefilter. How- ence model is unattainable for the controller
ever, it performs some of its functions, namely since following it requires large input signals
the smoothing of the abrupt input signals for for the actuator. An aperiodic-type unit is in-
the actuator. This reference model structure is troduced into the reference model to avoid
this. This unit plays the role of a prefilter.
selected based on the following considerations:
3. Reference model parameters are selected
1. Transient processes in angular longitudinal based on its feasibility for the controlled ob-
motion are oscillatory (the system of nonlin- ject coupled with a control system, and also
ear differential equations describing the lon- by taking into account the required range of
gitudinal angular motion of the aircraft is of deflection for the controls.