Introducing universal model-free multivariable Adaptive Neural Network Controllers for MIMO systems

Author: Arash Mehrafrooz

  • Thesis download: available for open access on 8 Jan 2023.

Mehrafrooz, Arash, 2019 Introducing universal model-free multivariable Adaptive Neural Network Controllers for MIMO systems, Flinders University, College of Science and Engineering

Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.

Abstract

In this thesis, novel universal model-free adaptive controllers based on neural networks are designed to control various types of model-free coupled Multiple-Input Multiple-Output (MIMO) systems. For each controller, a specific model-free adaptive learning algorithm, with accumulated gradients using the error back-propagation algorithm, is defined and developed to be automatically trained based on the history of inputs and outputs of the system. The system is considered as a ‘black box’ with no pre-knowledge of its internal structure. By online monitoring of the system inputs and system outputs, the controller is able to adjust itself to the new conditions such as changing the desired outputs, structural uncertainty of the system’s model, and unwanted disturbances. In this study, the design of the proposed controller is developed step-by-step as follows: (i) Firstly, by designing an Adaptive Neural Network Controller (ANNC) for controlling model-free Single-Input Single-Output (SISO) systems (in Chapter 2); (ii) Secondly, by expanding the ANNC to a Multivariable Adaptive Neural Network Controller (MANNC) to apply to square coupled MIMO systems (in Chapter 3); (iii) Thirdly, by improving the MANNC to Multivariable Adaptive Neural Network Controller with two dynamic layers (MANNC2) to deal with the non-linear characteristic of square coupled MIMO systems (in Chapter 4); (iv) and finally, by modifying the MANNC2 to a universal Multivariable Adaptive Neural Network Controller for Non-Square MIMO systems (MANNCNS), a general framework for controlling non-linear model-free non-square coupled MIMO systems is provided as the final product of this thesis (in Chapter 5); The Lyapunov stability criteria are developed for each multivariable controller and are embedded in their learning algorithms to guarantee the stability of the closed loop control system during the entire time of the adaptive control process. Several meaningful experiments by computer simulations relevant to each controller are performed in this study. The simulation results demonstrate the proper performance of the introduced controllers such as set-point tracking, accumulated error reduction, high control speed, unwanted overshoot/undershoot reduction, disturbance rejection, and dead-time compensation. These are compared to the best recent counterparts in significant examples with properties such as time-variance, non-linearity, and hybrid structure. The usefulness of the controllers in a wide range of applications, makes them potential candidatures to be implemented in industrial control software packages for use in many practical applications.

Keywords: Auto tuning, learning algorithm, Lyapunov stability, neural network control, MIMO, Multivariable control

Subject: Engineering thesis

Thesis type: Doctor of Philosophy
Completed: 2019
School: College of Science and Engineering
Supervisor: Fangpo He