Time-Frequency Analysis of Plant-Wide Oscillations Using Multivariate Intrinsic Time-Scale Decomposition

Abstract

Oscillations that propagate throughout a plant can have severe impacts on an industrial process. It is important and challenging to automatically detect and diagnose such oscillations in order to maintain efficient operation. A novel algorithm, multivariate intrinsic time-scale decomposition (MITD), is proposed to characterize the oscillatory behavior. It extends univariate intrinsic time-scale decomposition (ITD) by solving an overdetermined system of linear equations. To optimize the decomposition performance, several approaches are proposed which include (i) uniformly projection method, (ii) redefinition of proper rotation component for ITD and (iii) end effect restraining. The advantage of the developed method is that the investigated multi-dimensional signal can be analyzed under both time and frequency scales. Unlike other methods, both the regularity of the plant-wide oscillations (in frequency domain) and evolution of the local characteristics (in time scale) can be well captured via MITD. It is a...