| Summary: | <p>The current steady-rising interest in using the unmanned multi-rotor aerial vehicles (UMAV) designed to solve a wide range of tasks is, mainly, due to their simple design and high weight-carrying capacity as compared to classical helicopter options. <strong></strong></p><p>Unfortunately, to solve a problem of multi-copter control is complicated because of essential nonlinearity and environmental perturbations. The most widely spread PID controllers and linear-quadratic regulators do not quite well cope with this task. The need arises for the prompt adjustment of PID controller coefficients in the course of operation or their complete re-tuning in cases of changing parameters of the control object.</p><p>One of the control methods under changing conditions is the use of the sliding mode. This technology enables us to reach the stabilization and proper operation of the controlled system even under accidental external exposures and when there is a lack of the reasonably accurate mathematical model of the control object. The sliding principle is to ensure the system motion in the immediate vicinity of the sliding surface in the phase space. On the other hand, the sliding mode has some essential disadvantages. The most significant one is the high-frequency jitter of the system near the sliding surface. The sliding mode also implies the complete knowledge of the system dynamics. Various methods have been proposed to eliminate these drawbacks. For example, A.G. Aissaoui’s, H. Abid’s and M. Abid’s paper describes the application of fuzzy logic to control a drive and in Lon-Chen Hung’s and Hung-Yuan Chung’s paper an artificial neural network is used for the manipulator control.</p><p>This paper presents a method of the quad-copter control with the aid of a neural network controller. This method enables us to control the system without a priori information on parameters of the dynamic model of the controlled object. The main neural network is a MIMO (“Multiple Input Multiple Output”) system approximating the control signal for the system motion in the immediate vicinity of the sliding surface. The auxiliary neural network approximates the corrective control signal required to smooth out the high-frequency jitter effect near the sliding surface.</p><p>In the course of the study a quad-copter model was designed in the MATLAB environment according to the dynamic equations as well as a controller for three angles (roll, pitch and yaw). The controller consists of a neural network for approximating the main control signals and three neural networks for approximating corrective control signals (one per the axis). Environmental perturbations are involved in model.</p><p>Based on the system behavior simulation the effectiveness of the proposed control method is shown. Each of the orientation angles (roll, pitch and yaw) follows the desired trajectory with high accuracy. The stability of the system motion in the sliding surface vicinity is proved by Lyapunov method. The simulation results of the neural network controller and a quad-copter dynamic model in the MATLAB environment allow us to draw conclusion that the proposed control method ensures the stable motion along a given trajectory even despite environmental perturbations.</p>
|
|---|
