A high performance fuzzy inference processor and its evaluation system

The fuzzy inference microprocessor "FORCE" (Fuzzy Oriented Coprocessing Engine) and its evaluation system are described. In order to achieve high speed inference, antecedent processing and consequent processing are pipelined. To evaluate FORCE, the inference speed for various role sets has been measured on our system. In the case of the two stage inverted pendulum control, whose rule set is composed of thirty rules, the inference is performed in 15.2 /spl mu/s. This speed is 279 times faster than that of an 8-bit general purpose microcontroller (/spl mu/C). In FORCE, a special algorithm is employed for calculating the variance of the resultant membership function. Judging from the variance, the reliability of the inference result is examined. Also, the software tools for effectively developing the inference program are studied. A rule/membership function editor is proposed, which helps programmers to make the membership function according to the specifications of target processor.<
>