Early fault diagnosis of rotating machinery based on wavelet packets—Empirical mode decomposition feature extraction and neural network

After analyzing the shortcomings of current feature extraction and fault diagnosis technologies, a new approach based on wavelet packet decomposition (WPD) and empirical mode decomposition (EMD) are combined to extract fault feature frequency and neural network for rotating machinery early fault diagnosis is proposed. Acquisition signals with fault frequency feature are decomposed into a series of narrow bandwidth using WPD method for de-noising, then, the intrinsic mode functions (IMFs), which usually denoted the features of corresponding frequency bandwidth can be obtained by applying EMD method. Thus, the component of IMF with signal feature can be separated from all IMFs and the energy moment of IMFs is proposed as eigenvector to effectively express the failure feature. The classical three layers BP neural network model taking the fault feature frequency as target input of neural network, the 5 spectral bandwidth energy of vibration signal spectrum as characteristic parameter, and the 10 types of representative rotor fault as output can be established to identify the fault pattern of a machine. Lastly, the fault identification model of rotating machinery with rotor lateral early crack based on BP neural network is taken as an example. The results show that the proposed method can effectively get the signal feature to diagnose the occurrence of early fault of rotating machinery.

► In this study a new fault diagnosis method based on the WPD–EMD and BP network is proposed.
► EMD energy is introduced to reflect the actual work condition and the fault pattern of rotating machinery.
► BP network served as a classifier to identify the work condition of rotating machinery.
► The method is useful for extracting the feature for weak signals and diagnosing the early fault.