Dynamic Analysis of Evolving Industrial Alarm Floods Using an Adaptive Causal Directed Graph

The increasing complexity of highly automated industrial plants necessitates advanced tools for managing alarm floods and enhancing situational awareness for human operators. While Causal Directed Graphs have proven effective in these roles due to their low computational and informational requirements, their application has largely been confined to static scenarios. This is a limitation given the dynamic nature of many abnormal situations in industrial settings. To address this gap, this contribution introduces a novel extension to Causal Directed Graphs that incorporates real-time actuator states. Our approach is engineered to adapt to changing causal relationships, thereby enabling the identification of not only the root cause but also other causally important alarms in dynamic alarm flood scenarios. The effectiveness of our method is validated on the well-known "Tennessee-Eastman" process using multiple test scenarios from an openly accessible alarm management dataset.