On data-driven modelling and terminal sliding mode control of dynamic systems with applications

Publication Type:
Thesis
Issue Date:
2019
Full metadata record
This thesis addresses critical issues in system modelling and control with some applications to robotics and automation. The main content is divided into three parts, namely data-driven identification, fast terminal sliding mode control alongside underactuated crane control, and robotic pointing system for thermoelastic stress analysis (TSA). The first part is devoted to system modelling. A dynamic model can be identified from data collected (input and output data from the plant). However, the data obtained is often affected by noise. Hence, such algorithms for modelling the plant should be robust enough to accurately predict the dynamic behaviour of the system in the presence of noisy data. Taking this into account, this thesis focuses on subspace-based identification methods, and proposes an effective algorithm based on the Least-Square Support Vector Regression (LS-SVR). In the proposed algorithm, the system identification is formulated as a regression problem to be solved by applying multi-output LS-SVR. The second part of the thesis deals with the control of underactuated systems which are subjected to uncertainties including nonlinearities, parameter variations, and external disturbances. Among many control methodologies, Sliding Mode Control (SMC) is known for its strong robustness. Conventional SMC usually consists of linear sliding surfaces, which can only guarantee the asymptotic stability of the system, and hence, takes infinite time to reach the equilibrium. Requirements of finite-time stability can be fulfilled by adding the sliding function with a fractional nonlinear term to achieve the Terminal Sliding Mode, and using another attractor can lead to a faster response, called the Fast Terminal Sliding Mode (FTSM). FTSM is theoretically promising but it has limited application in real-time systems. This thesis is devoted to bridging this practical gap by developing a FTSM controller for underactuated mechanical systems. The third part of this thesis presents the applications of the proposed LS-SVR based identification algorithm and FTSM control scheme. Here, theoretical developments are implemented on a laboratorial gantry crane and an optical pointing system, respectively. Performance of both LS-SVR identification and FTSM control is verified through extensive simulation and experimental results. Notably, the work for this thesis has been applied to the RobotEye, an industrial pointing system of Ocular Robotics Pty. Ltd., which consists of a mirror integrated with other sensors such as laser sensors and vision cameras for robotic navigation or structural health monitoring with TSA.
Please use this identifier to cite or link to this item: