CONSTRAINED-OPTIMIZATION-BASED ADAPTIVE ROBUST CONTROL: THEORY AND APPLICATION

<p dir="ltr">In this dissertation, a two-layer constrained-optimization-based adaptive robust control (CARC) design is proposed and analyzed for a class of general nonlinear single-input single-output (SISO) systems. The design achieves multiple critical control objectives simultaneously, including accurate reference tracking, robust system performance and strict constraint compliance. Both simulation and experimental results are presented to validate the proposed design. This dissertation is organized as follows:</p><p dir="ltr">In Chapter 1, the historical development of control theory is first reviewed along several main lines, with emphasis placed on the recent demand for controller structures capable of achieving multiple objectives. The proposed CARC design is then presented and explained to address the limitations of traditional approaches. The design adopts a two-layer structure. In the outer layer, a high-level constrained-optimization-based trajectory planner generates a new (sub)optimal reference trajectory, addressing constraints and transient performance, when the system significantly deviates from the desired trajectory. In the inner layer, a low-level strong model-compensation-based adaptive robust feedback controller (ARC) maintains good robust tracking performance to the online replaned trajectory. The interactions between the two layers are well considered to achieve coordinated and consistent control performance.</p><p dir="ltr">In Chapter 2, the formal problem is first formulated for a class of general nonlinear SISO systems in semi-strict feedback form, subject to unmatched disturbances, parameter uncertainties, input variations, and hard state and input constraints. The CARC design is then formally proposed, with the transient and steady-state behavior analyzed in details. After that, comparative simulation results will be provided for a second-order unmatched system to verify the proposed design.</p><p dir="ltr">In Chapter 3, neural networks are trained to approximate the solver for the constrained optimization problem in the outer layer, with the aim of reducing online computational complexity and mitigating issues associated with the slow planning rate. Detailed procedures are shown, simulation results are provided, and the pros and cons of this approach are discussed.</p><p dir="ltr">In Chapter 4, the proposed CARC design is tested on a home-made flexible rotary joint system. The chapter starts with the modeling and system identification of the flexible joint system. After the model has been validated, a CARC controller is designed and implemented. After that, the neural networks are trained on a simplified case for demonstration purposes, due to the limited computation resources available. It is clearly shown in the results that even with significant nonlinearities and high-frequency dynamics, the proposed CARC design still achieves promising performance with all hard constraints being satisfied.</p><p dir="ltr">In Chapter 5, future work is suggested to further improve the design.</p>