Neuro-observer based control of double gimbal control moment gyro systems

In this paper, a new third-order nonlinear dynamics solving the mismatched problem is proposed for double gimbal control moment gyros (DGCMGs) affected by friction and coupling torques, unmodeled dynamics, or parameter uncertainties; the new nonlinear dynamics with the control inputs and the disturbances into the same equations should not be linearized and any control method handling multivariate systems with cross-coupling between channels can be then used to design a controller. Secondly, a feed-forward neural network based observer with disturbance rejection ability is proposed to estimate both the exogenous/endogenous disturbances and the states of the system, the number of necessary sensors being decreased. Thirdly, in order to control the rotations of the inner/outer gimbals, compensate the disturbances, enhance the robustness of the control system, and handle the channel interferences, we design, evaluate, and compare two novel control architectures using the Lyapunov theory, the backstepping method or the dynamic inversion control technique. Each of the two control architectures uses two interconnected controllers (an inner gimbal controller and an outer gimbal controller), a neuro-observer, and two reference models to generate the necessary fictive control signals. The simulation results show that the proposed control approaches provide very good angular rate precision of the DGCMG system and successfully handle a wide range of disturbances, parameter uncertainties, and channel interferences. The smaller fluctuations of the rotation angles, angular rates, and angular accelerations in the gimbal channels indicate that the backstepping control provides better ability to reject disturbances, smaller overshoot and convergence time than the dynamic inversion control technique. (C) 2020 Elsevier Masson SAS. All rights reserved.

Automated summary

A new third-order nonlinear dynamics solving method is proposed in this paper for DGCMGs, along with a feed-forward neural network observer and two novel control architectures using Lyapunov theory and backstepping method. These approaches enhance the robustness of the control system and reject disturbances effectively.