Variable-Lag-Aware Multivariate Granger-Based Method for Root Cause Diagnosis with Redundant Causality Removal

Due to the dynamic characteristics of chemical processes, the time lags between variables are varied, which may lead to spurious and indirect relationships in root cause diagnosis. To investigate the real-time variation of lags and reduce the misleading of the fault propagation path, a causal discovery method based on modified multivariate Granger causality is proposed. First, by analyzing the real-time changes in cross-correlation coefficients, the dynamic characteristics of variable lags are obtained. Meanwhile, the constraint of fixed lags that is assumed by the multivariate Granger causality is released to adapt to variable lags in dynamic chemical processes. On this basis, a variable lag logic is designed to enhance the discovery capability of the causal direction. Finally, the fault propagation path is illustrated through the causal-effect adjacency matrix, and the root cause of fault can be determined. The effectiveness of the proposed method is demonstrated by chemical case studies.