Hexarotor Longitudinal Flight Control with Deep Neural Network, PID Algorithm and Morphing

Abstract

Unmanned Aerial Vehicles (UAV) have become an integral part of life, from military operations to entertainment. With this popularity, the interest of researchers on these UAVs has gradually increased. The use of UAVs in many areas and the researches carried out by researchers have revealed the limitations of these devices. At the beginning of these restrictions is the passage through the narrow space. In this study, longitudinal flight control of a hexarotor UAV is discussed with morphing. The limitation of this study is that it is difficult to estimate the moment of inertia and proportional integral derivative (PID) coefficient values according to the arm length, since there is a change in the hexarotor arm lengths with the morphing and the rigid body model changes accordingly. In this study, it is aimed to obtain these parameters with the Deep Neural Network(Artificial Neural Network(ANN) with two or more hidden layers) to overcome this problem. 15 drawings of hexarotor morphing states were drawn in Solidworks program. A training set was created by obtaining the PID coefficients for the longitudinal flight of these drawings from the Matlab/Simulink program. The training set was taught to the Deep Neural Network and moments of inertia and PID coefficients were obtained according to the arbitrarily estimated arm extension or shortening rates. In addition, the hexarotor dynamic model was derived according to the Newton-Euler approach and modeled using the state space model. The longitudinal flight of the hexarotor is simulated with the state space model. The moment of inertia and PID coefficients were estimated by Deep Neural Network according to the values determined by the program randomly together with the initial state. Simulations were made with these parameters and the results were given in graphics.