Fault Diagnosis System For Rotating Machinery Based On Harmoinc Wavelet Packet And Neural Network

Now, detecting vibration signal is the mian approach of fault diagnosis system forlarge ratating machinery. It will arouse complicated dynamic non-stationary signalgenerally when ratating machinery goes wrong, so how to extract the characteristic vectorsof this kind of signal exectly is the prerequisite for diagnosis. Harmonic wavelet theory isvery suitable to extract characteristic of non-stationary and singularity signal because of itsstrict box-like spectrum. But it has not yet be used widely in the field of rotatingmachinery diagnosis due to the complexity of rotor fault signal. This paper adoptedharmonic wavelet packet and proposed a practical method to extract energy features ofrotor fault signal,based on other scholars' study, this method could avoid the effect ofrotor speed and sampling frequency.Firstly, this paper simply introduced the importance of ratating machinery faultdiagnosis and its present development situation, choosed some classic traditional meansfor signal decomposition to compare their merit and demerit; Then studyed the harmonicwavelet theory systematically, analysed its superiority in extrecting signal's features forweak signal, partical abrupt signal, nearly frequency signal; In the next place, this paperintroduced the structure and algorithm of Elman neural network, and compared it with BPnetwork to show its advantage in learning stability, convergence speed and diagnosticaccuracy; Finally, proposed a thinking of combining harmonic wavelet packet and Elmanneural network, and designed the basic structure of fault diagnosis system for rotatingmachinery based on it.Combining LabVIEW with MATLAB, this paper completed fault diagnosis systemperfectly, and simulated four classic faults of rotor on rotor test bed, acquired fault signalto test the fault diagnosis system, the result showed that this system's property was good.