Design of an expandable fuzzy inference coprocessor with application to FLC hardware implementation

Abstract

Fuzzy logic controller(FLC) is widely used for applications ranging from industrial automation systems to consumer electronics products. Previous research revealed that FLCs have, in many cases, a superiority over the conventional control algorithms such as PID controller. This property is a nothing surprise because the fuzzy logic on which the FLC is based is more closer to the human decision making and the FLC is inherently a nonlinear controller. In order to realize the FLC, there are, in general, 3 ways to choose. The controller can be implemented by using a look-up table in which controller output values are calculated and stored beforehand corresponding to possible input values. This so-called fuzzy memory controller is simple but requires huge memory space when the number of input/output is increased or the resolution of input variables is increased to have better performance. Secondly, one can utilize a software program executed on a general purpose computer. The method is suitable for low speed applications because general purpose computers are generally Von-Neumann type whereas fuzzy inference procedure requires multiple data computations. In this thesis is investigated the methodology of utilizing specialized hardware processors in order to construct FLCs, which is better than the above two methods. It uses a straightforward manner and dedicated hardware processors can be exploited by fast processing and its simplicity. There has been many researches about the hardware, but those processors are, in general, inflexible to expand. Therefore, a novel architecture of fuzzy inference coprocessor consisting of IF modules and THEN modules is proposed in this thesis. Using the architecture, one can expand the capacity of an FLC by simply adding another module. Furthermore, pipelined parallelism is adopted in the architecture. By the way, it is well known that there is no systematic way to design FLCs. Although many researchers have tackled the problem but th...