Application of DTF method in disturbance propagation analysis of complex chemical process

Abstract

Oscillations usually propagate to other loops with the delivery of mass and energy, then cause plant-wide oscillation and affect the performance of whole control system in complex chemical process. DTF (Directed Transfer Function) method, which has been widely used to analyze information flow in the brain structures in biomedical area, is applied to the disturbance propagation analysis of complex chemical process in this paper. Based on MVAR (Multivariate Autoregressive) model, DTF can analyze the multivariate causality simultaneously and calculate the causality quantitatively. Based on the DTF value, one can draw the causality graph, get the disturbance propagation path and finally locate fault sources. The results of simulation on TEP (Tennessee Eastman Process) are presented to illustrate the effectiveness of the proposed approach.

From oscillatory to non-oscillatory disturbances: A comparative review of root cause analysis methods

“…Reference [42] applied transfer entropy for RCA of another type of transient disturbance with limited efficiency, since transfer entropy finds causal dependencies between all variables affected by the disturbance and is unable to deduce the root cause. [37,56], and [25] highlighted that directed transfer function combined with Granger causality and partial directed coherence provided highly accurate results for RCA of oscillations.…”

“…Reference [37] uses a 95th-percentile threshold based on values of the directed transfer function causality measure on surrogates generated with the full Causal Fourier Transform (CFT). Reference [56] uses a three-sigma threshold and generates surrogates using iterative AAFT. As for other significance tests based on surrogates, the reliability is low [25].…”

“…• Multivariate and direct causality: The autoregressive modelling is multivariate by default [56] but directed transfer function does not distinguish direct and indirect causality.…”