摘要
In industrial processes, fault diagnosis is a core technology for ensuring the stable operation of industrial systems. Accurate identification of causal relationships between variables is crucial for root cause localization. Conventional Granger causality analysis can only capture linear causal interactions, making them unsuitable for the complex nonlinear causal relationships in industrial processes. This paper proposes a nonlinear Granger fault root cause diagnosis method based on the Deep Autoregressive Model (DeepAR). This method combines the time series probabilistic prediction model with the Granger analysis, leveraging DeepAR to model uncertainty and to learn similar time-series trends. Experiments using numerical simulation and the Tennessee Eastman process demonstrate that the proposed method can accurately identify the fault root cause variable and detect causal relationships consistent with the mechanistic analysis. © 2026 IEEE.