Physics‐informed sparse causal inference for source detection of plant‐wide oscillations

Identification of the source of plantwide oscillations is a challenging problem, even with the availability of big data. Causality analysis is often used to construct causal maps and obtain a sequence of fault propagation for an in‐depth investigation. The reliability of the widely used Granger causality depends on the quality of the observed data. But since real‐world industrial data are prone to sensor errors, their accuracy is significantly compromised. Experienced engineers possess years of valuable process knowledge which when introduced into the modeling can significantly reduce the over‐dependence on data. In this article, we propose a novel approach to efficiently amalgamate expert information with the observed data to reconstruct causal maps. A new surrogate‐data‐based approach to test the significance of the causal relations obtained for oscillatory data is also proposed in this article. The efficiency of the proposed methodology is demonstrated using a simulation and an industrial case study.