A two-level servo structure with knowledge-based tuning

Abstract

This paper presents a two-level structure for servo controllers. The lower level is occupied by a conventional servo loop of the proportional-integral-derivative type. The upper level consists of a knowledge-based tuner for the servo parameters. Expert knowledge expressed in the form of linguistic fuzzy rules represents the knowledge base for servo tuning. These rules, along with membership functions for condition and action variables, are processed through fuzzy logic to obtain a multi-dimensional membership function for the knowledge. Tuning inferences are made from the response of the servo loop by applying the compositional rule of inference to the knowledge base. This paper presents a general theory for the development of a knowledge-based tuner. Specifically, the concepts of fuzzy resolution and rule dissociation are introduced. A parallelism between conventional parameter-adaptive control and fuzzy tuning control is pointed out. Through rule dissociation to an elementary level, the fuzzy resolution is arbitrarily improved. By introducing a monotonicity assumption, an analytical framework is established. It is shown that in this framework, fuzzy tuning may be expressed as an algebraic scheme, and consequently the dynamic characteristics such as stability of the overall tuning system may be established purely from the dynamics of the original system. A numerical example is presented to illustrate the basic scheme of fuzzy tuning.