Page 167 - Neural Network Modeling and Identification of Dynamical Systems

4.3 APPLICATION OF ANN MODELS TO ADAPTIVE CONTROL PROBLEMS UNDER UNCERTAINTY CONDITIONS 157
4.3.3.3 Model Predictive Control Applied to ble to identify the requirements for the eleva-
the Angular Motion of an Aircraft tor actuator. From the results of computational
The behavior of an aircraft under the con- experiments it follows that the required rate of
trol of the MPC system is most similar to its deviation of the elevons lies in the range of
behavior under the control of the MRAC sys- ±50 deg/sec.
tem. These data can be compared with the re-
To assess the effectiveness of the adaptive sults for hypersonic research vehicle X-43 (flight
MPC algorithm, we have carried out several se- with Mach number M = 6), which are shown in
ries of computational experiments. Initially, the Fig. A.49, where the operation of the system is
MPC algorithm was tested for a stepwise ref- shown in the absence of failures and damage
erence signal concerning the angle of attack. at a more complex (random) nature of the ex-
It was required to synthesize the control law
citation signals. Fig. A.50 demonstrates the op-
for the longitudinal angular motion of the hy- erability of the system under consideration in
personic research vehicle X-43, which would
the conditions of two consecutive failures that
provide a high accuracy of stabilization of the
affect the hypersonic research vehicle dynam-
required angle of attack determined by the in-
ics. These failures caused a shift of the centering
put command signal for various combinations
by 5% back at the time t = 30 sec, and then a
of the Mach number and altitude characteris-
tic for a given aircraft. An example of the re- 30% decrease in the efficiency of the control at
t = 60 sec.
sultsfromthisseriesisshown inFig. A.40;
an extended set of these results is given in Similar computational experiments applied
Figs. A.41–A.48. From the results obtained, we to an adaptive control system with a predictive
can see that the control laws synthesized with model were performed for a maneuverable F-16
the use of adaptation mechanisms with predic- aircraft (the results are shown in Figs. A.51, A.52,
tive model provide a sufficiently high quality and A.53), as well as for the UAV (operation of
of control. Namely, for all the flight regimes the MPC control system in the nominal mode);
studied, the error of tracking a given angle of see Figs. A.54 and A.55 and, for the case of fail-
attack when it was changed dramatically by ures, see Fig. A.56.
up to 12 degrees did not exceed ±0.27 deg, The conclusions that follow from the results
and in some cases it dropped to ±0.08 deg. Af- of the computational experiments presented in
ter the transition to the new value of the an-
this section are, in general, similar to those
gle of attack was completed, the value of the that were made for MRAC systems. Namely, in
tracking error decreased almost in all cases to
most cases, the adaptive neural network con-
±(0.01 ÷ 0.02) deg.
trol system, whose structure is represented in
The above results also show the nature of the
Fig. 4.14, successfully copes with its tasks, in-
operation of the longitudinal control surface of
the hypersonic research vehicle (elevons used as cluding emergency situations.
Comparison of the MRAC and MPC schemes
elevator), required for implementing the synthe-
sized control law, by comparing the values of does not allow choosing any of them. Each of
the command signal input to the elevator ac- them has its own positive and negative features.
tuator with the deflection angle of the elevons. The final decision in favor of any one of these
In addition, by analyzing the data on the speed schemes can only be made regarding a particu-
of the deflection of the elevons required to im- lar applied problem, after a fairly extensive se-
plement the obtained control law, it is possi- ries of computational experiments.