Suppressing Spurious Causal Chains in Alarm-Triggered Alarm Phenomena via Liang–Kleeman Information Flow–Max-Min Hill-Climbing Graphs

Due to coupling in industrial control structures, alarm variables may exhibit interconnected variations, leading to the “alarm-triggered alarm” phenomenon, which can contribute to alarm flooding. Causal inference methods can characterize dependencies among alarm variables and help identify potential propagation paths for alarm configuration optimization. However, strong autocorrelation, nonlinear dynamics, and time-delay effects in complex industrial processes may lead conventional causal inference methods to identify spurious causal relationships, thereby reducing the reliability of the resulting causal graphs. To address this issue, we propose a Liang–Kleeman information flow–Max-Min Hill-Climbing (IF-MMHC) causal graph framework that combines physics-informed information flow analysis with constrained score-based structure learning. Specifically, Liang–Kleeman information flow (LKIF) separates endogenous evolution from external driving effects and provides physically informed constraints for identifying candidate causal relationships, thereby reducing spurious causal effects associated with temporal dependence. The resulting candidate graph is then optimized using the Max-Min Hill-Climbing (MMHC) algorithm through constraint checking and score-based structural search, further removing redundant and indirect edges and improving the structural parsimony of the causal graph. Experiments on multiple fault scenarios of the Tennessee Eastman Process demonstrate the effectiveness of IF-MMHC in identifying key alarms and causal propagation relationships while reducing redundant causal connections. The results provide evidence for the applicability of the proposed framework to industrial alarm analysis and configuration optimization, while broader validation across different industrial processes remains necessary.