Contribution to the intelligent control of robot manipulators

Abstract

This doctoral thesis presents a comprehensive study of the intelligent control of robot manipulators. It employs advanced techniques such as neural networks, optimization algorithms, and fuzzy logic to develop innovative control strategies. This thesis proposes several control techniques, compares them to existing methods, and evaluates their performances in terms of tracking accuracy of reference trajectories and disturbance rejection through simulations and experimental validations on different models of robot manipulators. The first contribution of this work is the development of a neural network model predictive controller based on Archimedes Optimization Algorithm (AOA). This proposed controller relies on a feed-forward multi-layer neural network to accurately predict the system's future outputs and employs the Archimedes optimization algorithm to compute optimal control actions. The proposed controller's trajectory tracking and disturbance rejection performances are investigated through simulation on a two degrees of freedom robot manipulator. A comparative study between the proposed control technique, the PID controller, the computed torque controller, the neural network model predictive controller based on the Teaching-Learning-Based Optimization (TLBO), and the neural network model predictive controller based on the Particle Swarm Optimization (PSO) is carried out. Additionally, the proposed control algorithm is implemented on a DSP board to control a three degrees of freedom SCARA robot manipulator and compared to the neural network model predictive control based on the TLBO algorithm and the neural network model predictive control based on the PSO algorithm. The second contribution enhances the neural network model predictive controller's robustness against external disturbances, unmodeled dynamics, and uncertainties by integrating active disturbance rejection control. The predictive controller incorporates a terminal cost constraint to ensure stability, while the active disturbance rejection controller uses an extended state observer to estimate and compensate for the total disturbances. The efficiency of the suggested control approach is demonstrated through experimental validation to control a four degrees of freedom MICO robot manipulator. In the third contribution, we combine the prescribed performance control with the neural network model predictive controller to maintain tracking errors within predefined bounds, which significantly improves the system's transient response. The performances of the proposed controller are compared against the neural network model predictive controller in simulation to control a four degrees of freedom MICO robot manipulator. The fourth contribution focuses on enhancing the computed torque controller by adding a fuzzy controller as a nonlinear element and employs the Archimedes optimization algorithm for the controller's parameters optimization. The performances of the proposed fuzzy computed torque controller are evaluated in simulation, considering the control of a six degrees of freedom PUMA 560 robot manipulator and comparing to the PID and the computed torque controllers.