ORCID

0000-0003-2034-5404

Date of Award

2024

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical Engineering

First Advisor

Luis Duffaut Espinosa

Abstract

Non-linear systems are difficult to model due to a lack of comprehensive understandingand inaccuracies in the model parameters. Model-Free Control (MFC) is a non-linear control technique that aims to solve this problem by replacing the model of the system with an Ultra-Local Model (ULM). The ULM parameters are estimated using sensor data and the input that goes into the system. Conventional MFC techniques estimate a single parameter in the ULM while the tuning parameter is selected by an expert based on the application. This dissertation focuses on eliminating the dependence on expert knowledge for selecting the tuning gain by introducing a Kalman Filter (KF) based methodology that can estimate both parameters of the ULM simultaneously. The sensor data used to predict the ULM parameters is prone to outliers due to the failure of the sensor or because of the noise in it. The presence of outliers will have a detrimental effect on the estimation of the ULM which consequently has a severe effect on the system being controlled. To overcome this robustification techniques can be applied to the sensor data coming from the system, once the sensor data has been robustified it can then be used by the ULM. One such technique that can be used to robustify the sensor data is called the Robust Generalized Maximum Likelihood Estimator (RGMKF). This dissertation provides a novel methodology for estimating the ULM parameters by using an extended KF approach where the overall dynamics of the system are estimated. The extended KF approach is further modified to an optimized constrained KF approach to estimate both parameters of the ULM. Next, this dissertation introduces a methodology to integrate RGMKF into MFC to introduce robustification to sensor outliers. Additionally, modifications to RGMKF are also introduced that can be used for the detection of outliers in an alternative method. Examples and simulations have been provided for each of the proposed methodologies, and this is followed by the conclusion and future work.

Language

en

Number of Pages

134 p.

Available for download on Thursday, November 27, 2025

Share

COinS