Fault diagnostic system on knowledge for steam turbogenerator

In this paper a new neural network model is proposed for fault diagnosis of steam turbogenerator, which is a self-organization network model that forms a continuous topological feature mapping from vibration frequency spectrum to a kind of fault. The self-organization network model is trained without supervision. The knowledge base in the corresponding expert system is built up and maintained automatically. We discuss the setting and running of the network and review the implementation of fault diagnosis when a turbogenerator is started or stopped. The network performs the diagnosis automatically. The topological feature mapping is observed by using graphics. It can quickly gives a direct answer as to whether a fault exists and what type of fault it is, without requiring the interference by a human expert