An adaptive learning algorithm for controlling a two-degree-of-freedom serial ball-and-socket actuator

Abstract

This article presents the development and implementation of an artificial neural network (ANN) for controlling a new two-degree-of-freedom (2DOF) serial ball-and-socket actuator. The ANN is a well-known algorithm for simulating the ability of the human brain to learn and predict sets of information. In this approach, ANN will learn the control parameters to obtain the angular displacement, angular velocity, and angular acceleration of the end-effector without any prior knowledge of the actuator. The ball-and-socket actuator has been proposed as an alternative actuator to the conventional one-degree-of-freedom (1DOF) revolute actuator. The actuator was fabricated from a ball-and-socket joint powered by two electrohydraulic cylinders. Experimental control data had been collected manually and provided for ANN to learn in off-line mode. The training process was carried out to build control knowledge. Thus, the adaptive learning algorithm adopts any modification in the actuator mechanism and hydraulic power system through updating the control knowledge. The results of implementing the build control knowledge for on-line operation of the ball-and-socket actuator shows a fully compliant actuator end-effector to the desired dynamic behaviour within the workspace.