FLEXIBLE NEURO-FUZZY SIMULATION-BASED ON ABSTRACT MODEL AND INTERFACEDEFINITIONS

This paper first presents an overview of neuro-fuzzy systems. Then a n umber of evaluation criteria for neuro-fuzzy simulation tools are esta blished. An analysis of tools points out the difficulties in simultane ous fulfillment of this criteria, and leads to the development of desi gn principles to overcome these difficulties. The basic idea is to imp lement simulators by two abstract descriptions-a model and an interfac e definition. The MiND system, which has been built according to this idea, is then presented. On the one hand, MiND offers a comfortable gr aphical user interface, and on the other hand, it is flexible and exte ndible, and supports the development of custom applications. As a hard ware component, the MiND system includes a Peripheral Compound Interco nnect (PCI) neuroboard for the acceleration of neural networks. The pa per closes with a discussion of the MiND system with respect to the ev aluation criteria established at the beginning.