Prediction of Gear Defect Based on WPT & CEEMDAN & SVD

After the defected gears are determined, a novel method, combined with wavelet packet decomposition, complementary ensemble empirical mode decomposition with adaptive noise and singular value decomposition, is put forward. It is utilized to exclude disturbance of irrelevant signals that generated by the defect gears. Firstly, wavelet packet decomposition is used to extract the defect signals and retain original features. The processed signal is called S1 and the irrelevant frequency bands could be filtered out. Secondly, complementary ensemble empirical mode decomposition with adaptive noise decomposes S1 into a series of intrinsic modal functions. The correlations between S1 and intrinsic modal functions are analyzed. The intrinsic modal functions that are highly correlated with S1 are screened out and reconstructed into a new signal, called S2. The disturbance of irrelevant signals could be further filtered out, but some of them still disturb the judgement. Thirdly, singular value decomposition decomposes S2 into several singular values. The large singular values are choose to represent S2 and the disturbance of irrelevant signals could be reduced to the minimum. After applying above processes, the disturbance of irrelevant signals is nearly eliminated. The defect feature of gears can be easily and accurately judged. Finally, the judgement is proved to be true. This method provides a novel idea for gear defect analysis.