Data-driven root cause diagnosis of process disturbances by exploring causality change among variables

Abstract

Granger causality (GC) analysis is a widely used method in root cause diagnosis; however, the current GC-based method has deficiencies that need to be improved. Causality exists in a large number of chemical processes; therefore, causal relationships obtained by utilizing only the data after disturbance in existing analysis methods are usually too complex to provide reliable disturbance-related causality. To address this problem, this study proposes a novel root cause diagnosis framework based on GC analysis, which comprehensively utilizes the information of both normal data (the data in normal operation) and disturbance data (the data after a disturbance occurs) to filter the disturbance-unrelated causality and obtain compact and useful disturbance-related information. The Kolmogorov–Smirnov test (KS-test) is used to detect the significance of the causal intensity change before and after the disturbance occurs. The effectiveness of the proposed method is verified by a case study in the Tennessee Eastman process dataset, which not only makes the results of root cause diagnosis compact but also brings convenience to the analysis of the disturbance propagation path.