This paper considers five time domain features of pump vibration: P,P p ,R rms ,V var and C crest as minimum combination of diagnosic parameters(MCDP), and uses BP neural networks to fuse and synthesize these features. A on line NN based condition monitoring and fault diagnosic system(NNCMFDS) for hydraulic pump is presented. The paper also discusses two modes of data representations which are single node data mapping(SNDM) and spread encoding(SE).The simulation and bench test results demonstra...