Root cause diagnosis with co‐integration constrained Liang–Kleeman information flow in non‐stationary processes

Abstract To cope with the challenge of non‐stationary characteristics to causal identification, this paper proposes a Liang–Kleeman information flow framework under co‐integration residual constrained for root cause diagnosis in industrial processes. Specifically, the stationarity is first determined by combining augmented Dickey–Fuller and Kwiatkowski–Phillips–Schmidt–Shin tests, and the non‐stationary variables can be screened out. Then, the co‐integration analysis is used to identify the long‐term equilibrium relationship between variables, and the driving term is constructed with co‐integration residuals. Next, the residual matrix is embedded in the Liang–Kleeman information flow to improve the ability to characterize the changes in causal intensity in non‐stationary processes. Finally, experiments are carried out based on typical industrial process Tennessee Eastman Process to verify the effectiveness and applicability of the proposed method in root cause diagnosis.