Hybridization of neural and fuzzy systems by a multi agent architecture for motor gearbox control

In this paper a hybrid system for motor control on testbeds, consisting of neural networks with a self-organizing process state detection and fuzzy rulebases, is proposed. The basic mechanism used for hybridization is a multiagent system composed from loosely interconnected subsystems for the different control tasks to be accomplished. The major aims taken into account are: using standard - and approved - subsystems, realize an easily expandable system, which can handle the exchange or even failure of a component. The proposed system is implemented as a first stage using a simple motor and car simulation. First results show the system's capability to control the car simulation precisely following a given speed profile using knowledge acquisition from the fuzzy system to the neural network.