Self-Tuning and Adaptive Control

Abstract

Conventional controllers have transfer-functions with fixed parameters which are chosen from knowledge of plant dynamics or by manual tuning experiments. For cases where the dynamics are unknown a self-tuning algorithm can be used; with the advent of microprocessor technology such methods are easily implemented and are increasingly popular. This article describes the background to many current approaches to self-tuning design which are based on a Generalised Minimum Variance strategy. The plant is described as a discrete-time transfer-function with dead-time and with stochastic disturbances; a predictor model is then generated and the unknown parameters of this model are estimated from plant input/output data using a Recursive Least Squares estimator. An outline of the implementation of this self-tuner is given together with a discussion of additional features required to make it work in practice. One important limitation of the GMV approach is the need to know the plant’s dead-time; a new method called Generalised Predictive Control is shown to overcome this problem. For simple process-control applications PID regulators are entirely adequate and methods for the self-tuning of their proportional, integral and derivative terms are described — all of which are easy to embed into a microcomputer.