Neural Network Based Non-Linear Control Methods with Observer Design for Robotic Manipulators

Abstract

Robotic manipulators are often used in applications requiring high precision. It is inevitable to use a controller for the satisfactory operation of such manipulators. In general, an open-loop system subjected to torque disturbances and parameter uncertainties causes instability. Therefore, to ensure the global asymptotic stability, gravity compensation derived from either conventional or non-linear control methodologies with independent joint control is essential, resulting in a closed loop. As a result, several controllers have emerged during the last decades for improving the system stability with better disturbance rejection and small tracking error. Since then, many derivatives and refinements to the classical controllers have been proposed. However, a fusion/hybridization of hard control (proportional integral derivative controller) and soft control (computational intelligence technique based) is an alternative choice for better performance. Therefore, an effort towards the designing of such fusion-based controllers is worth investigating. With this motivation, several hybrid controllers as applied to robotic manipulators are proposed. First, the control strategy for robotic manipulator based on the coupling of artificial neuro-fuzzy inference system (ANFIS) with sliding mode control (SMC) is proposed. As a part, boundary sliding mode control (SMCB), boundary sliding mode control with PID sliding surface (PIDSMCB) and backstepping sliding mode control (BSMC) are developed for the best optimal criterion by using the genetic algorithm (GA) and particle swarm optimization (PSO). Further, they are applied for the control of 2-Degree of freedom (DOF) robot manipulator. The proposed neuro-fuzzy-based adaptive controller offers several advantages such as the consistent estimation and considerable robustness to parameter variation and external disturbance. Second, control strategies for 3-DOF rigid robot manipulator based on the coupling of neural network (NN)-based adaptive observer with SMC are proposed. A radial basis function neural network (RBFNN)-based observer is used to estimate the trackingposition and velocity vectors of overhead transmission line de-icing robot manipulator (OTDIRM). To overcome local minima problem, the weights of both NN observer and NN approximator are adjusted off-line using PSO. All the developed controllers are simulated extensively in MATLAB/SIMULINK. Numerical simulations using the dynamic model of a single-link rigid robot manipulator with two and three DOF in the presence of input torque disturbances are performed. Finally, the obtained simulation results considering various torque disturbances and uncertainties in terms of path tracking and disturbance rejection are validated through a set of experiments for a 2-DOF manipulator.