AI Techniques for Control and Observation of Nonlinear Dynamic Systems
No Thumbnail Available
Date
2026-05-19
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Souaad TAHRAOUI
Abstract
This thesis confronts the gap between theory and practice in controlling the Twin Rotor
MIMO System (TRMS). A conventional approach combining a backstepping controller with
a fixed-gain High-Gain Observer (HGO) was first evaluated, demonstrating successful performance in simulation for both setpoint regulation and trajectory tracking. When transferred
to the physical platform, the regulation performance was preserved, confirming the validity
of the design under steady-state conditions. However, the same controller failed completely
during real-time trajectory tracking, revealing a fundamental limitation of the fixed-gain observer architecture: its inability to simultaneously ensure fast state estimation and suppress
real-world sensor noise under dynamic operating conditions. To resolve this specific failure
mode, an adaptive control architecture was designed in which online-learning neural networks intelligently tune the observer gains in real time. Two distinct schemes were developed: a Feedforward Neural Network (FFNN) and a Radial Basis Function Neural Network
(RBFNN), both modulating the HGO parameters based on the live observation error. Experimental validation on a TRMS testbed confirms that both proposed methods achieve robust
and accurate tracking under the exact conditions that caused the conventional controller to
fail. This work presents a proven solution to a documented real-world control problem and
provides a direct comparative analysis of the FFNN and RBFNN approaches in this challenging application.
Description
THESIS
Presented to obtain the degree of
DOCTORATE
Major: Science and Technology
Specialty: Automation and Industrial Informatics
Keywords
High-Gain Observer, Backstepping Control, Feedforward Neural Network, Radial Basis Function Neural Network, Adaptive Gain Tuning, Twin Rotor MIMO System