Page 168 - Neural Network Modeling and Identification of Dynamical Systems

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

4.3.4 Adaptive Control of Angular (t = 20 sec) is manifested as a shift of the center
Aircraft Motion Under by +10%, the second (t = 50 sec) as a decrease in
Uncertainty Conditions longitudinal control efficiency by 50%.

4.3.4.1 Influence of Atmospheric 4.3.4.2 Adaptation to Uncertainties in
Turbulence on the Efficiency of an Source Data
Adaptive Control System for the The presence of uncertainties in the source
Aircraft Longitudinal Motion data leads to the fact that the neural network
One of the issues traditionally difficult for model and the neurocontroller will be tuned in-
adaptive control systems is their ability to with- accurately and this inaccuracy has a negative ef-
stand disturbing external effects of a random na- fect on the quality of control.
ture. In the computational experiments conducted,
In this section, an attempt is made to evaluate, inaccurate knowledge of the aircraft dynamics
for the MRAC system, how well a synthesized was simulated by the fact that at the beginning
system can cope with disturbances, including of the simulation, an ANN model or neurocon-
cases when emergencies arise. troller tuned to a different flight mode was spec-
As a model of atmospheric turbulence, the ified. Thus, the synthesis of the control law was
well-known Dryden model was used, as de- applied to one flight mode (Mach number and
scribed in the MIL-F-8785C standard and imple- altitude of flight), and then the control law be-
mented in the Simulink modules of the gan to work quite in other conditions. The cor-
Aerospace Toolbox for the Matlab package. The respondence of the flight regime, for which the
effect of turbulence is manifested through ad- synthesis of the control law and the regime in
ditional components for the vertical velocity V z which this control law was tested in the compu-
and the pitch angular velocity q. tational experiments, is given in Table 4.1.
All results were obtained for the maneuver- Two options were considered.
able F-16 aircraft, for the flight regime H = According to the first variant, the adapta-
100 mand V = 600 km/h. They are presented tion mechanism was activated (its main features
in Figs. A.57–A.60. were considered in Chapter 1), which allowed
Two cases are considered: for the correction of the control law in relation to
the operating conditions in which it appeared.
1. A disturbing effect of
V z, turb in the range
The second approach was to assess the impor-
±10 m/sec,
q turb in the range ±0.2 deg/sec
tance of the introduced adaptation mechanisms
(Figs. A.57 and A.58).
by revealing their contribution to the overall
2. A disturbing effect of
V z, turb in the range
task of ensuring the necessary control quality
±20 m/sec,
q turb in the range ±2 deg/sec
under changing operating conditions. The adap-
(Figs. A.59 and A.60).
tation mechanisms were disconnected to imple-
To evaluate the effect of atmospheric turbu- ment this approach, that is, the control law was
lence on the characteristics of the controlled sys- not adjusted, and the problem of ensuring the
tem, for each of the cases considered, two ver- control quality for the system was entirely car-
sions were calculated, i.e., with and without tur- ried out by the robustness mechanisms, includ-
bulence. ing the compensating circuit introduced into the
In all variants, the behavior of the system was systems in Section 4.3.2.3.
considered with successive emergence of two The results of the computational experiments
failure situations introduced in the same way within the first of the variants listed above are
as wasdoneinSection 4.3.2.5. The first of them presented in the next two paragraphs. The sec-