High-fidelity noise-reconstructed empirical mode decomposition for mechanical multiple and weak fault extractions

The premise to ensure the safe operation of rotating machinery is to accurately and timely capture all kinds of damages and fault signatures, whose challenging issues are the multiple and weak faults. Ensemble noise-reconstructed empirical mode decomposition (ENEMD) is a smart method of adaptively decomposing and denoising for mechanical fault diagnosis. However, the original ENEMD and its derivative methods suffer from the drawbacks of critical noise estimation on the high accuracy, leading to the powerless capability for accurate multiple and weak fault signature extraction. Thus, high-fidelity noise-reconstructed empirical mode decomposition (HNEMD) is proposed to overcome the drawbacks, whose compositions include: (1) The cepstral neighboring coefficient editing is proposed to obtain the pre-whitening signal with the actual white noise, from which the major noise can be well purified by the minimax thresholding. (2) The high order singular value decomposition (HOSVD) local reconstruction is developed to estimate the minor noise from the pre-characterizing signal, where the tensor with the reasonable singular order is constructed by the sliding window and Hankel matrix. (3) The high-fidelity noise recovered by the major noise and the minor noise is reconstructed and combined with the basic ENEMD algorithm for accurate signature extraction, especially for the multiple or/and weak fault ones. Three repeatable simulations are analyzed to illustrate the signature extraction capability and noise estimation process of this method. Moreover, the effective detections of weak and multiple faults from a hot strip finishing mill gearbox and an aerospace bearing further validate the feasibility of the proposed method.