The Research Of Intelligent Methods On Fault Diagnosis For Centrifugal Compressor

In recent years machinery fault diagnosis technology has been greatly developed all over theworld and the technology is very important in national production. It is rather difficult tomonitor the state of centrifugal compressor owing to its complex construction and so manystimulation sources. Although the researches have been made and some achievements have beengained, the total diagnosis level isn't very high and does not match with practical application.The research on centrifugal compressor fault diagnosis is of great significance owing to theparticularity of its application. The research work which is combined with practice in this thesisis based on predecessors' achievements, intelligent diagnosis methods are adopted and faultdiagnosis is effectively solved for centrifugal compressor valves.In this thesis, the denoising method and the feature extraction method are researchedaccording to the vibration characteristic of centrifugal compressor rotor system that periodicsignals, impact signals and random signals etc are intermixed, signals feature is diffcultlyextracted and so on. Acorrding to the character that wavelet transform technique can localize thetime-domain and frequency-domain, it can focalize any detail of signals and can recognize theabrupt signal and nostationary signal. The characteristics of typical fault for the centrifugalcompressor rotor are extracted by wavelet transform, and the characteristics vector is foundedby using energy as element. The method of fault diagnosis for centrifugal compressor based onBP neural network is researched in the thesis too. The neural network has the ability to processnonlinear, autolearn, compute parallel, diagnose online, approach function and recognize mode.It is very popular in the field of nonstationary time serial, but the shortage that easily get intomin. and convergence rate is slow limts the application of BP neural network. Therefore, thethesis use RBF neural network to diagnose the fault of the centrifugal compressor rotor systemwith the view of mode recognization, and train the characteristics vector which is founded byusing wavelet decomposition. The result is well tested through gathering signals , which testifiesthis system has usability.