Research On Vibration Fault Diagnosis System Of Rare-Earth Permanent Magnet Machines

Rare-earth permanent magnet machines are used in the production of industry and agriculture no more than 30 years. It played a great role in national production, People pay more and more attention on it, but there is little research to make on fault diagnosis of rare-earth permanent magnet machines, so it is more meaningful to do a work on vibration fault diagnosis system of rare-earth permanent magnet machines.Permanent magnet machines use permanent magnet for exciting instead of electric excitation, so there is no such fault as broken rotor bar, winding short circuit, casting defects produced by large current or overload. For this there more reliable in permanent magnet machines, but there still has fault in stator such as inter-turn short circuit, layer short circuit, phase to phase short circuit and stator loosening. In this paper has been analyzed electromagnetic force wave of permanent magnet synchronous motor when it is working, and obtained the vibration characteristics of unbalance stator, paper also concluded main machinery vibration faults' characteristic.On fault signal processing, multi-resolution analysis (MRA) based on wavelet analysis theory is put in forward in this paper to decompose fault signal from high to low frequency. The fast wavelet algorithm-Mallat pyramidal algorithm is realized by VB program and applied it to do multilevel wavelet decomposition, then compare signal processing results with FFT analysis, it proved that wavelet analysis is better than Fourier analysis in processing multilevel frequency and transient signal.Paper use improved BP neural network algorithm which theory has been fully developed and more practical to identify faults. Applied momentum BP algorithm to resolve norm BP neural network algorithm has long training time and easily fall into minimization by sample data normalization and increasing momentum factor. Combined with fault character frequency of permanent magnet synchronous motor defined input sample and output goal data. Paper identified faults by 3-layer network and optimized the number of middle layer and learning rate. Training results prove that optimized network has improved in convergence speed. Testing training results by input new fault sample show that it can identify fault sample pattern.Paper use VB program to create a diagnosis system based on wavelet analysis and BP neural network algorithm, basis on ISO2372 standards classified equipment. Applied system to analyzing 55kW rare-earth permanent magnet machines used in pumping unit, proved that wavelet analysis is suit to analyzing fault signal and the results of BP diagnosis is consistent with fact.