Model-Based Non-Linear Dynamic Control of Cable-Driven Manipulators

This study focuses on the investigation of different kinds of non-linear control strategies for cable-driven manipulators. These robots are widely employed in many contexts, especially in industrial site inspections, where their compact size, allowing them to enter narrow spaces, and lightweight, increasing their payload capabilities, and precious. Moreover, the possibility of separating actuators from the robotic arm provided by cable actuation ensures their safety from possible dangerous environments, which is another crucial benefit in many industrial contexts. Besides these advantages, cable-driven manipulators' performance risks being degraded by the elasticity of the cables, which, in the worst cases, can also lead to system instability. For this reason, the main objective of this project is to provide a set of model-based non-linear control strategies to improve the motion precision and accuracy of cable-driven manipulators by exploiting a dynamic model of cables that is more precise than the ones usually used in literature. These controllers are all based on the same structure, a non-linear full-state feedback controller, which was proposed in four different variations: (i) a fixed gain controller, (ii) a fixed gain controller with sliding mode action, (iii) a gain scheduling controller implemented using a feed-forward artificial neural network and using the reference signal as scheduling variable, and (iv) a gain scheduling controller with sliding mode action. The closed-loop stability was mathematically verified through the direct method of Lyapunov for the four proposed controllers. The second objective of this work is the validation of the employed dynamic model, which was pursued by comparing the results of simulations with the results of an experimental campaign conducted on a real robot. The third and final goal of this project was the non-linear controllers performance comparison, which was performed analysing and comparing the results of the experimental campaign. The experimental results successfully validated the cable-driven manipulator dynamic model and allowed the drawing of interesting conclusions about the controller performance comparison. In particular, it was noticed that the fixed gain with sliding mode controller provided the closed-loop system with the best performance when the reference signal was constant, a condition in which the gain scheduling performed worse. In contrast, the gain scheduling with sliding mode controller performed best when the reference signal varied with time.