Page 234 - Neural Network Modeling and Identification of Dynamical Systems

REFERENCES 225
gitudinal motion of the maneuverable aircraft REFERENCES
are RMSE V = 0.0026 m/sec, RMSE α = 0.183 deg,
RMSE q = 0.0071 deg/sec. [1] Egorchev MV, Kozlov DS, Tiumentsev YV. Neural net-
The identification of the accuracy of the aero- work adaptive semi-empirical models for aircraft con-
trolled motion. In: 29th Congress of the International
dynamical coefficients C x , C z ,and C m is demon- Council of the Aeronautical Sciences; 2014.
strated by the data shown in Figs. 6.13, 6.14,and [2] Egorchev MV, Tiumentsev YV. Learning of semi-
6.15. empirical neural network model of aircraft three-
In each of these figures, the upper part of the axis rotational motion. Opt Memory Neural Netw
(Inf Opt) 2015;24(3):210–7. https://doi.org/10.3103/
figure shows the actual values (according to the S1060992X15030042.
data from [20,21]) of the required coefficients de- [3] Egorchev MV, Kozlov DS, Tiumentsev YV, Cherny-
pending on the angle of attack α and the de- shev AV. Neural network based semi-empirical mod-
flection angle of the controlled stabilizer δ e .The els for controlled dynamical systems. Her Comput Inf
Technol 2013;(9):3–10 (in Russian).
lower part shows the approximation errors for
[4] Egorchev MV, Kozlov DS, Tiumentsev YV. Aircraft aero-
the corresponding ANN modules. We can see dynamic model identification: A semi-empirical neu-
that the accuracy achieved is very high. ral network based approach. Her Mosc Aviat Inst
2014;21(4):13–24 (in Russian).
The results presented above allow us to draw
[5] Egorchev MV, Tiumentsev YV. Neural network semi-
the following conclusions.
empirical modeling of the longitudinal motion for ma-
Just as in the case described in the previous neuverable aircraft and identification of its aerody-
section and in [1,2] for the identification problem namic characteristics. Studies in computational intelli-
of the aerodynamic coefficients C y , C z , C l , C n , gence, vol. 736. Springer Nature; 2018. p. 65–71.
[6] Egorchev MV, Tiumentsev YV. Semi-empirical neural
C m , the semiempirical ANN modeling methods network based approach to modelling and simula-
successfully provide the solution to the identifi- tion of controlled dynamical systems. Proc Comput Sci
cation problem of the axial force coefficient C x , 2017;123:134–9.
given the known engine model. If this model is [7] Klein V, Morelli EA. Aircraft system identification: The-
ory and practice. Reston, VA: AIAA; 2006.
unavailable, then the solution to the identifica-
[8] Tischler M, Remple RK. Aircraft and rotorcraft system
tion problem would provide the approximation identification: Engineering methods with flight-test ex-
for the total axial force coefficient, dependent on amples. Reston, VA: AIAA; 2006.
the control variable δ th as well. As follows from [9] Jategaonkar RV. Flight vehicle system identification: A
time domain methodology. Reston, VA: AIAA; 2006.
(6.14)and (6.15), this is quite enough to simulate
[10] Berestov LM, Poplavsky BK, Miroshnichenko LY.
themotionoftheaircraft. Frequency domain aircraft identification. Moscow:
The second important conclusion, which fol- Mashinostroyeniye; 1985 (in Russian).
lows from the obtained results, is that the “rep- [11] Vasilchenko KK, Kochetkov YA, Leonov VA,
Poplavsky BK. Structural identification of mathematical
resentation power” of the semiempirical model
model of aircraft motion. Moscow: Mashinostroyeniye;
is quite sufficient to approximate complicated 1993 (in Russian).
nonlinear functional dependencies defined on a [12] Ovcharenko VN. Identification of aircraft aerodynamic
wide range of values of their arguments, pro- characteristics from flight data. Moscow: The MAI
Press; 2017 (in Russian).
vided that there is a training set with the re-
[13] Etkin B, Reid LD. Dynamics of flight: Stability and con-
quired level of informativity. trol.3rd ed.New York,NY: John Wiley&Sons, Inc.;
The results of computational experiments 2003.
show high accuracy of both the obtained ANN [14] Boiffier JL. The dynamics of flight: The equations.
Chichester, England: John Wiley & Sons; 1998.
model of the longitudinal motion of the aircraft
[15] Roskam J. Airplane flight dynamics and automatic
and the corresponding aerodynamic character- flight control. Part I. Lawrence, KS: DAR Corporation;
istics. 1995.