Weak time-frequent feature enhancement method using improved ensemble noise-reconstructed empirical mode decomposition and its application

Based on the noise utilization mechanism, ensemble noise-reconstructed empirical mode decomposition (ENEMD) uses the noise component inherent in the input data to ameliorate the mode mixing problem and to cancel each other out by a collection in the mean IMFs given enough empirical mode decomposition (EMD) trials, yielding the signal denoising. However, the analogous hard thresholding is adopted in the pivotal noise estimation technique, ignoring the relativity among the coefficients. Thus, the noise estimation using the neighboring coefficient principle is investigated to improve the precision of noise estimation. On the basis, improved ENEMD is introduced to Hilbert-Huang transform (HHT) and weak time-frequent feature enhancement method using improved ENEMD is proposed. In the method, the instantaneous frequency by IMFs without the mode mixing could accurately characterize the weak fault signals. Meanwhile, by the denoised IMFs, the signal-to-noise ratio of HHT is effectively improved and the noise of HHT is restrained, which heavily enhance the resolution and weak faults of time-frequency features, highlighted the local fault symptoms. The proposed method provides an effective tool for mechanical early and weak fault identification. The engineering applications showed that the method could effectively reveal the impact and friction fault symptom from the compressor air separation, and successfully extract the early weak impact fault feature from the heavy oil catalytic cracking unit.