Use of Fast Multivariate Empirical Mode Decomposition for Oscillation Monitoring in Noisy Process Plant

Abstract

Industrial signal measurement and processing are increasingly being deployed in control performance assessment, particularly in oscillation monitoring applications. In this paper, we present a novel oscillation detector, which mainly leverages the recently developed fast multivariate empirical mode decomposition (FMEMD). Our use of FMEMD is motivated by the following facts: (i) considerably fewer techniques are now available for monitoring both single-loop and plant-wide oscillations; (ii) the presence of noise and signal intermittency can severely deteriorate the detecting performance of most decomposition-based time–frequency methods; and (iii) the currently popular multivariate empirical mode decomposition (MEMD)-based methods are limited by high computational load and overdecomposition. The proposed approach is tested against the related techniques, including empirical mode decomposition (EMD), direct multivariate intrinsic time-scale decomposition (DMITD), and the noise-assisted MEMD (NA-MEMD) by extensive simulations, and is shown to offer considerable progress, especially in noise robustness, mode-mixing reduction, and intermittent and nonlinearity-induced oscillation extraction. The validity of this work is finally demonstrated by industrial cases.