Design and implementation of a real-time adaptive learning algorithm controller for a 3-DOF parallel manipulator / Mustafa Jabbar Hayawi

Hayawi, Mustafa Jabbar (2015) Design and implementation of a real-time adaptive learning algorithm controller for a 3-DOF parallel manipulator / Mustafa Jabbar Hayawi. PhD thesis, Universiti Teknologi MARA.

[img]
Preview
Text
TP_MUSTAFA JABBAR HAYAWI EM 15_5.pdf

Download (6MB) | Preview

Abstract

A parallel manipulator is a closed loop mechanism which consists of a moving platform that is connected to a fixed base by at least two kinematic chains in parallel. Parallel manipulators can provide several advantages, such as high stiffness, high accuracy, and low inertia but also have some disadvantages, such as small workspace, complicated structures, a high cost, and also pose a major challenge to their analysis and control. To overcome the above shortcomings, progress on the development of parallel manipulators with less than 6-DOF has been accelerated. In this thesis, a new parallel manipulator with three degrees of freedom DOF is designed. Kinematic of the manipulator including inverse kinematic, Jacobian matrix and velocity equation are analyzed. Performance indices such as workspace, dexterity and stiffness, of the parallel manipulator are studied. The parallel manipulator is optimized based on the performance indices to obtain on the optimal design parameters for achieved maximum performance of the parallel manipulator. A prototype was fabricated to demonstrate the manipulator. An electronic board, transistor relay driver circuit, is designed for the purpose of establishing communication interface between the computer, adaptive learning algorithm and the actuator mechanism. Design and development an adaptive learning algorithm controller ALAC of position the actuators is presented in real time parallel manipulator based on artificial neural network ANN. The control of a parallel manipulator based on joint space control is implemented by determine the lengths of the electrohydraulic actuators EHA individually. EHA are known to have nonlinear parameters and dynamic factors such as frictions, load variations and leakage. These effects, if not controlled, would lead to large scale oscillations that would damage system components. System identification is a prerequisite to analysis of a dynamic system. It is performed using neural network Auto Regressive with exogenous (NNARX) model based on the input/ output experimental data. There some criteria such as loss function, Aikeke's Information Criterion, goodness of fit and correlation analysis of the residual are analyzed to determine the adequate model for representing the EHA. The real time ALAC is coded in MATLAB/ SIMULINK and consists of two inverse ANN, feed forward neural network ANN1 and feedback network ANN2, and proportional feedback controller (P). The ANN2 is used to update the weights and biases online while ANN1 is used to implement the controller. Pcontroller is designed to improve the controller system and ensure the stability. The updated weights and biases of the ANN1 are same set obtained ANN2. The results of the proposed controller were compared with direct inverse neural controller DINC. It was found from experimentation that the ALAC was able to improve the position control of the parallel manipulator and adjusted the weights according to changing condition and control the system without any changes in the controlling algorithm. It is still effective with few overshoot compared with DINC.

Item Type: Thesis (PhD)
Creators:
CreatorsEmail
Hayawi, Mustafa JabbarUNSPECIFIED
Subjects: T Technology > TJ Mechanical engineering and machinery > Robotics. Robots. Manipulators (Mechanism)
Divisions: Faculty of Mechanical Engineering
Item ID: 21610
Uncontrolled Keywords: Implementation; Real-time adaptive learning; Algorithm controller; 3-DOF parallel manipulator
Last Modified: 27 Sep 2018 02:22
Depositing User: Staf Pendigitalan 2
URI: http://ir.uitm.edu.my/id/eprint/21610

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year