Gain scheduled dynamic surface control for a class of underactuated mechanical systems using neural network disturbance observer

Abstract A gain scheduled dynamic surface control (GSDSC) based on neural network disturbance observer (NNDOB) is developed for a class of uncertain underactuated mechanical systems with multiple equilibria. Equilibrium manifold linearization (EML) transforms the nonlinear system into an equivalent model using scheduling variables. The NNDOB with a filtering adaptive law is constructed to approximate arbitrary disturbances. The dynamic surface controller is designed based on the NNDOB and EML model, which efficiently solves the mismatches and overcomes the problem of “explosion of complexity” resulting from repeated differentiation of virtual control variables. Moreover, the robust stability of the closed-loop system is proved by Lyapunov stability theorem. Finally, we apply the proposed control scheme to a power line inspection robotic system. The simulation results illustrate the approximation capability of NNDOB and the effectiveness of the presented control scheme.