Integration of CNN in a Dynamic Model-Based Controller for Control of a 2DOF Helicopter With Tail Rotor Perturbations

Abstract

This paper shows a proposal for a control scheme for the trajectory tracking problem in a Two Degree of Freedom Helicopter (2DOFH). For this purpose, a control scheme based on a feedback linearization combined with a Generalized Proportional Integral (GPI) controller is used. In order to implement linearization by feedback, it is required to know and have access to all the physical 2DOFH parameters, however, angular velocity and viscous friction are often not available. Commonly, state observers are used to know the angular velocity, however, estimating friction results out to be more complex. Therefore, we propose the use of a Convolutional Neural Network (CNN) to estimate viscous friction and angular velocity. The variables estimated by the CNN are entered into both the GPI and feedforward controllers. Thus, the system is brought to a linear representation that directly relates the GPI control to the dynamics of perturbations and non-model parameters. Finally, results of numerical simulations are shown that validate the robustness of our scheme in the presence of disturbances in the tail rotor, as well as the advantages of using a feedforward control based on a CNN. INDEX TERMSFriction estimation, Neural networks, Non-linear system, Tail rotor disturbance, Two degrees of freedom helicopter. Hence, different experimental prototypes have been developed to study helicopter dynamics [13]-[15]. One of the most popular consists of a Two Degree of Freedom Helicopter (2DOFH), which recreates the dynamic behavior of the helicopter in its pitch (θ) and yaw (ψ) rotations [15], [16].