Turbine machine fault diagnosis using modified redundant second generation wavelet packet transform

Faulty features extraction is an essential problem in the field of large-scale electromechanical equipment faulty diagnosis. Classical vibration faulty features extraction is based on spectral analysis method, while the wavelet transform provides a novel tool to solve this problem. In this paper, the problem of frequency band derangement inhering in redundant second generation wavelet packet transform (RSGWPT) was explained and the causes were pointed out. Then a modified redundant second generation wavelet packet transform which can make the order of decomposed subband signals to be consistent with the linear partition order of frequency band is proposed. The modified RSGWPT discards the split and merge operations in the decomposition and reconstruction stages and directly use the constructed operators to accomplish prediction and update steps. Thus the signal length at each level is the same with the original signal, accordingly more information of the time domain features can be preserved, and at the same time the aliasing of RSGWPT can be inhibited effectively. This method was applied to analyze the simulated signals and the practical turbine machine vibration faulty signals. Testing results show that the proposed improved RSGWPT method is quite effective in extracting the faulty features from the vibration signal, so it can be effectively applied to the fault diagnosis of turbine machine.