摘要
•A innovative state space equation enables multiscale causal interaction analysis among fault variables.•A multi-timescale Granger causality framework is developed for root cause diagnosis.•Case studies on TEP and motor bearing systems validate the proposed approach.
As modern industrial processes become increasingly complex and integrated, accurate root cause diagnosis of faults has become increasingly challenging. The fault propagation process in industrial systems involves complex multi-time scale dynamics, which are essential for a thorough analysis of propagation mechanisms but are seldom considered in existing Granger causality-based root cause diagnosis algorithms.
To fill this gap, this paper proposes a multi-time scale Granger-based fault root cause diagnosis algorithm, which can accurately infer the root cause and provide valuable information on fault propagation. Specifically, the algorithm begins by redefining the state process to transform the vector autoregressive model into an innovative state space equation after filtering. This is followed by down-sampling, and the state error variance matrix is obtained by solving the discrete algebraic Riccati equation, updating the form of the innovative state equation. Subsequently, the covariance matrices of the residuals from both full and restricted models are used to calculate the Granger causality strength, and statistics are constructed for the significance test. Finally, the maximum Granger causality strength at different time scales is employed to construct a causal diagram, infer the fault root cause, and analyze propagation mechanisms.
The proposed algorithm is validated using both the Tennessee Eastman process benchmark and a real-world motor bearing system. These two representative case studies demonstrate the algorithm's effectiveness across different industrial scenarios. The results illustrate its strong applicability in accurately inferring the root causes of faults in complex industrial processes.
[Display omitted]