Fault feature extraction of rotating machinery using a reweighted complete ensemble empirical mode decomposition with adaptive noise and demodulation analysis

Fault feature extraction is crucial to detect failures as earlier as possible in fault diagnosis of rotating machinery. Due to the influence of environment noise and interference, the signal to noise ratio (SNR) of fault feature is relatively low in the measured signal. Complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is an improved method based on EEMD, which has been extensively applied to signal de-noising. The key problem for CEEMDAN is to determine the fault-related degree of a decomposed intrinsic mode function (IMF), especially in the presence of both Gaussian and non-Gaussian noises or interferences. However, most of the traditional assessment criterions are developed to describe the statistical parameters of IMFs, e.g. correlation coefficient and kurtosis, which ignore the specific characteristics of the fault and are easily affected by noise components. Therefore, a new criterion is proposed to quantify the fault-related degree of a vibration signal, in which the ratio of periodic modulation components caused by fault to the generalized interferences is defined. Then, a reweighted and reconstruction strategy of the decomposed IMFs is presented to obtain the de-noised signal based on the new criterion. Furthermore, in order to detect the fault-related modulation features in multi-frequency scales, a time-frequency representation (TFR) based demodulation analysis is employed, which guarantees an accurate extraction of the fault feature at the early stage of fault. The effectiveness of the proposed fault diagnosis method comparing to traditional methods are demonstrated by both numerical simulation and experimental studies. The results show that the proposed method achieves a better performance in terms of SNR improvement and fault feature detection, it can successfully detect the fault features in the presence of Gaussian and non-Gaussian noises. (C) 2019 Published by Elsevier Ltd.