Based On Wavelet Analysis And Data Mining Rotating Machinery Fault Diagnosis

Research on rotating machinery fault diagnosis is of great significance for reducing economic losses. With the the development of science and technology, experts home and abroad advanced a lot of new methods and new thought for fault diagnosis, which were appied in practice, while, in practice, it is not worth of destroying the machine to gain the fault database of rotating machinery. This dissertation conducts profound research and discussion of rotating machinery fault diagnosis.1) Based on studying large number of documents, the four fault conditions of rotating machinery, that is imbalance, misalignment, rubbing and oil whirl, were simulated on the experimental instruments, known as Bently RK4, and vibrational displacement signal of the four faults has been gotten, which is as the input data for wavelet analysis.2) According to research, this dissertation summarizes basic theory of wavelet analysis, containing the development of wavelet analysis, continue wavelet transform, discrete wavelet transform, multi-resolution analysis theory, Mallat algorithm and wavelet-packet decomposition algorithm, which are applied to analyse the vibrational displacement signal of the four faults.3) This dissertation researches the application of connection between scale and energy modulus to represent characteristic vector of signal, selects appropriate scale to ascertain the dimension of vector better, and gets characteristic vector based on scale-energy modulus.4) This dissertation summarizes basic theory and implementation procedure of data mining, and researches three algorithms of data mining, that is neural network algorithm, support vector machine algorithm and fuzzy cluster algorithm, which are used to recognise the above-mentioned characteristic vector.5) The solving process of optimal hyperplane of SVM has been deduced in detail.6) Based on the fault diagnosis of rotating machinery by adopting wavelet analysis and data mining, merits and faults of the three algorithms have been compared.