A robust method for coupling detection among process variables

Abstract

Investigating complex interaction patterns among multiple process variables (PVs) is an important task. This paper demonstrates a robust method to estimate direction and strength of interactions among process variables. The method captures rapid changes in process variables through wavelet analysis. It uses single degree of freedom (SDOF) modelling to approximate a non-linear system in terms of linear damped forced oscillators. Phase interaction theory then extracts coupling direction and strength among process variables. The robustness of the proposed technique is verified on simulated Van der Pol oscillators with known directionality and coupling strength with varying signal to noise ratio (SNR). The effectiveness and feasibility of the proposed method have also been demonstrated on simulated data emanating from Canada Deuterium Uranium (CANDU) nuclear power plant steam generator level control mechanism. The extracted patterns of interaction structure among PVs aid to uncover the polishing mechanisms and provide more insights to investigate fault propagation scenarios.