Identification of Fuzzy Rule on Manual Control of an Unstable System

Abstract

After some training, human operators can manually control very unstable objects when some proper information is given. But they can hardly explain how they do it, because they operate them intuitively and not logically. In this paper, we study the human behavior during the control of a double inverted pendulum and identify its control rules experimentally. The motion of a double inverted pendulum is simulated by a micro-computer and some of the state variables are indicated on a CRT, observed by a subject, and controlled on a keyboard. In order to find which information is used by a subject, his visual points are examined by an eye-camera. As a result, we see that there are three phases of operation, that is, the decrease of initial deviation, the prevention of over-shoot, and the keeping of stability. Next, the motion of a pendulum is analyzed qualitatively in each phase so as to identify the control rules of a human operator. By this analysis, we see that the intuitive manipulator of the human operator is quite reasonable from the physical viewpoint, and we represent it by some linguistic rules. From these results, we suggest a hierarchical structure of fuzzy rules as a model of a human operator which is verified through experiments on fuzzy control. It is concluded that this fuzzy controller acts as a skilled operator, but its performance is far superior to humans.

Fuzzy Control Systems

“…After defining design requirements and performing system identification we need to develop a knowledge base (KB) which consists of rule base contents and structure, termset definitions, and scaling factors. The validated knowledge base is compiled to minimize memory requirements and run-time search, and it is deployed on the target microprocessor [2], The KB can be derived from knowledge engineering sessions with process operators including analysis of observed operator responses [33], from published rule bases for standard control policies such as PI and PD [41], and from linguistic models of the open and closed loop systems [3], We have mentioned in the beginning of this section that the expert control system and the fuzzy logic control system have at least one thing in common, i.e., both want to model human experiences and human decision-making behaviors. In addition, both the expert control system and the FLC contain the KB and inference engine, and most of them have been the rule-based systems until now.…”