AN INVESTIGATIVE STUDY INTO ROBUST CONTROL DESIGN TECHNIQUES

Abstract

An important aspect of controller design is the ability to achieve a desired performance despite model uncertainties. This thesis reports the various ways of designing and implementing robust controllers. The project is divided into four parts, consisting of two case studies, a controller board design and a new robust controller algorithm design. In Part 1 and 2, the importance of the robust control technique is demonstrated by applying Quantitative Feedback Theory (QFT) control to an electro-mechanical platform rate-stabilisation loop design and H? control to a high performance aircraft pitch-rate autopilot design respectively. As advanced robust controller is usually complex, there is a need to look towards fast hardware in implementing these complex algorithms. In Part 3, a DSP-based fast universal controller board (UCDSP) is designed. The board is designed based on a novel link-up between the AT&T DSP32 digital signal processing chip and the NEC uPD78P312A microcontroller chip. The division of tasks between the two processors is such that the DSP chip handles the main bulk of the computational requirements while the microcontroller handles the task scheduling and the real time interfacing functions. The resulting system is a powerful number cruncher and yet very flexible. To achieve a truly robust controller for a wide class of systems, there is a need to appeal to the techniques of both robust controller design and adaptive controller design. A new robust H? optimal adaptive controller is proposed in Part 4. The proposed robust H? optimal adaptive scheme minimizes on-line the H? norm of the closed loop transfer operator of a feedback system. The parameters of the plant model are estimated using a set membership constrained weighted least squares algorithm. The estimated model is then used to derive on-line an approximated H? norm optimal feedback controller using a weighted least squares technique with Lawson's algorithm in the updating of the least squares weights.