Simple neuron-fuzzy tool for small control devices

Abstract

Small control computers, running a kind of Fuzzy controller, are more and more used in many systems from household machines to large industrial systems. The purpose of this paper is firstly to describe a tool that is easy to use for implementing self learning Fuzzy systems, that can be executed in a large group of micro computers and secondly to illustrate the use of the tool by an example. The tool is called FuNNy. FuNNy generates Fuzzy systems and consists of a compiler and a C learning library. The compiler translates a Fuzzy system (written in a dedicated language, called FuNNy language) to C. The C learning library contains the learning algorithm. The generated C code is simple standard C and therefore it can be applied to all computers which can be programmed in C. The learning algorithm is a gradient descend method based on a numerical calculation of the gradient. The input fuzzyfication can be described by four different kinds of membership functions. The output fuzzyfication is based on singletons. The rule base can be written in a natural language. The result of the learning is a new version of the Fuzzy system described in the FuNNy language. A simple shower control example is shown. This example shows that FuNNy is able to control the shower and that the learning is able to optimize the Fuzzy system.