Denoising of Industrial Oscillation Data Based on EEMD-DFA-CCA

The oscillation data collected from process industry is often corrupted with random noise and external disturbances, which hamper the monitoring of oscillation and muddle the subsequent root cause diagnosis. Therefore, the removal of noisy artifacts from the original oscillation signal is of great significance. This paper proposes an integrated denoising framework to cater for the above requirement, which is featured by following steps: (1) The ensemble empirical mode decomposition (EEMD) is used to decompose the single-loop oscillation data into series of intrinsic mode functions (IMFs). (2) By leveraging the detrended fluctuation analysis (DFA), the original IMF set is separated into noise-dominated IMFs and oscillation-dominated IMFs. (3) The canonical correlation analysis (CCA) algorithm is finally served for further extracting the effective oscillation information from the noise-dominated IMFs, then the output of CCA and the oscillation-dominated IMFs are reconstructed to obtain the denoised oscillation data. The practicality of the raised denoising strategy is evidenced by both numerical and real-word cases. The results indicate that our method offers considerable enhancement in improving the quality of industrial oscillation data.