Knowledge Base Decoupling in Fuzzy-Logic Control Systems

In a coupled rule base, more than one condition variable may link with more than one action variable within the same rule. This paper develops an analytical framework for decoupling such a rule base, with specific reference to fuzzy-logic control. Since control decisions are made through the use of the compositional rule of inference, the computational problem of fuzzy-logic is greatly simplified through the use of an uncoupled rule base. Conditions are derived, under which inferencing through an uncoupled rule base becomes equivalent to that through a coupled rule base. A laboratory prototype of a single-axis servo motor system which incorporates a knowkdge-based tuner is employed to demonstrate the effective use of an uncoupled rule base. Experiments are carried out by operating the system, with only a segment of the rule base active. By comparing the tuning response of the servo motor when such subgroups of rules are sequentially activated, with that when all rules are activated simultaneously, it can be established whether rule base decoupling has resulted in a performance degradation.