Page 208 - Neural Network Modeling and Identification of Dynamical Systems

CHAPTER
6

Neural Network Semiempirical Modeling

of Aircraft Motion

6.1 THE PROBLEM OF MOTION els of nonlinear controlled dynamical systems
MODELING AND introduced in [3–6].
IDENTIFICATION OF The typical problem of system identification
AIRCRAFT AERODYNAMIC in aviation is based on the use of the motion
model for aircraft as a rigid body. Such a model
CHARACTERISTICS
is described by a system of ODEs or DAEs.
In the most common case, the motion model
In the process of forming aircraft motion
of an aircraft is described by the following ODE
models, we need to solve a problem, which is system:
very significant for practice. Namely, the initial
theoretical model of the object contains, as a rule,
˙ x = f(x,u,t), x = (x 1 ,...,x n ), u = (u 1 ,...,u m ),
elements that we cannot determine with the re-
y = h(x,t), y = (y 1 ,...,y p ).
quired accuracy without involving experimen- (6.1)
tal data on the behavior of the modeled object
due to the lack of knowledge about this object.
The right hand sides of the equations of the
For an aircraft, these are, most often, the non-
aircraft motion include, among others, the val-
linear dependencies of the aerodynamic forces
ues of the aerodynamic forces (longitudinal, lat-
and moments on the parameters characterizing eral, and normal, respectively)
its motion. The reconstruction of the form of
such dependencies from available experimental 2
ρV
data (for example, based on flight tests results X = C x ¯qS; Y = C y ¯qS; Z = C z ¯qS; ¯ q =
2
for aircraft) is a traditional system identifica-
tion task. The approach proposed in the book as and aerodynamic moments (roll, pitch, and yaw,
part of the formation process for semiempirical respectively)
ANN models provides the restoration of un-
known (or insufficiently known) dependencies L = C l ¯qSb; M = C m ¯qSb; N = C n ¯qS ¯c.
that are included in these models. We proposed
this approach to solving this problem in [1,2]. It A typical feature of the aircraft motion model
is based on the use of semiempirical ANN mod- is that it is determined up to aerodynamic forces

Neural Network Modeling and Identification of Dynamical Systems
https://doi.org/10.1016/B978-0-12-815254-6.00016-2 199 Copyright © 2019 Elsevier Inc. All rights reserved.