Contents I. Introduction
There are a lot of fault diagnosis systems based on neural network, but if the network has deficiency in knowledge acquisition and representation, reasoning may be vulnerable. Generally, fault diagnosis research based on neural network acquires numerical value knowledge from cases, and represents knowledge with the connection weights of trained network. Limitation of this kind of knowledge base in fault diagnosis systems is that they cannot express uncertain or imprecise information and the knowledge for reasoning is not complete and perfect. Besides, classification of fault cause is absolute, i.e. either this cause or another one. So the fault diagnosis systems are of little use when multiple faults exist. It is hoped that neural network, which is trained by each single fault sample, has the ability to expand its function to deal with cases of multiple faults. However, the network has limited ability of association and generalization performance, and the accuracy of fault diagnosis is not high enough. Fuzzy neural network (FNN) is a combination of fuzzy system and neural network. It not only has advantage of neural network's numerical computation, but also has the ability of fuzzy system handling expert knowledge and diagnosing multiple faults simultaneously. So it has been attached much importance by researchers.
