Abstract: From the viewpoint of information source, the system of monitoring and diagnosis for machinery is one of multi-source information processing systems. Various features can be reduced to three types: numeric, linguistic and graphics. Through translating the non-numeric symptom into numeric one, information of various types can be denoted by multi-dimension vector. So, the idea of features fusion of various types is proposed through information compression, and the method of how self-organizing feature mapping (SOFM) network deals with it is studied. With the trace of active nodes on output layer, the underlying features varying of state represented by multi-source information can be observed correctly and visually, so occurrence and varying trend of faults can be identified early. The high performance of this method proposed is exemplified by handling fusion in experiments and field work.

Key words: Fault diagnosis, Information fusion, Self-organization feature mapping