The implementation of a generalised robust adaptive controller

Abstract

An adaptive controller is developed, comprising a robust parameter estimator and an explicit pole assignment controller design. The controller is reformulated to have a standard PID structure. A practical implementation is facilitated on a digital microcomputer, connected to a physical process. Test results are presented for this real process subject to variable dead-time and an external disturbance. Simulation results are also presented, for a nominally non minimum-phase process subject to variable dead-time and large open-loop gain changes. Robust performance is demonstrated under all of these circumstances. Recommendations are given for the choices and considerations required in a robust practical implementation. Much research has been done in the field of adaptive control over the past few decades. However, a let needs to be learned about the robustness of adaptive control algorithms. This research investigates the implementation of a practical adaptive control algorithm, with numerous features incorporated to improve the robust performance of such a controller. Parameter estimation is performed using Recursive Least Squares (RLS), with various signal conditioning filters to reduce estimator sensitivity to noise and modelling errors. The control design is based on closed-loop pole assignment, with adaptive feed forward compensation included. Further, provision is made in both the estimation model and the feedback control structure to eliminate deterministic immeasurable disturbances, and to track deterministic set point variations. This is based on the Internal Model Principle. Measured random disturbance signals are included in the estimation model, for which "transfer function" polynomial coefficients are estimated and then used in the feed forward control d e sign. A new shift- operator, namely the 6-operator, is used in all controller and estimator formulations. This has been shown to have better numerical properties and to correspond more closely to continuous-time control, than the traditional q operator of z-domain discrete control. A practical implementation on a digital computer is investigated, applied to a real plant typical of an industrial application. Simulation results are also obtained for plant with non minimum-phase zeros and variable dead-time.