ANN based MRAC-PID controller implementation for a furuta pendulum system stabilization Academic Article in Scopus uri icon

abstract

  • © 2020 ASTES Publishers. All rights reserved.Nowadays, process automation and smart systems have gained increasing importance in a wide variety of sectors, and robotics have a fundamental role in it. Therefore, it has attracted greater research interests; among them, Underactuated Mechanical Systems (UMS) have been the subject of many studies, due to their application capabilities in different disciplines. Nevertheless, control of UMS is remarkably more difficult compared to other mechanical systems, owing to their non-linearities caused by the presence of fewer independent control actuators with respect to the degrees of freedom of the mechanism (which characterizes the UMS). Among them, the Furuta Pendulum has been frequently listed as an ideal showcase for different controller models, controlled often through non-lineal controllers like Sliding-Mode and Model Reference Adaptive controllers (SMC and MRAC respectively). In the case of SMC the chattering is the price to be paid, meanwhile issues regarding the coupling between control and the adaptation loops are the main drawbacks for MRAC approaches; coupled with the obvious complexity of implementation of both controllers. Hence, recovering the best features of the MRAC, an Artificial Neural Network (ANN) is implemented in this work, in order to take advantage of their classification capabilities for non-linear systems, their low computational cost and therefore, their suitability for simple implementations. The proposal in this work, shows an improved behavior for the stabilization of the system in the upright position, compared to a typical MRAC-PID structure, managing to keep the pendulum in the desired position with reduced oscillations. This work, is oriented to the real implementation of the embedded controller system for the Furuta pendulum, through a Microcontroller Unit (MCU). Results in this work, shows an average 58:39% improvement regarding the error through time and the effort from the controller.

publication date

  • June 1, 2020