A Novel Approach to Alarm Causality Analysis Using Active Dynamic Transfer Entropy

Alarm flooding is a serious safety problem in the chemical process industries. Bayesian Networks are a set of powerful tools that can be used to trace the root-cause of alarms. For highly integrated complex chemical processes, we propose a Bayesian Network based on Active Dynamic Transfer Entropy (ADTE) to establish an accurate alarm propagation network during an alarm flood. The proposed method has two primary advantages: (1) It circumvents the false causality problem caused by strong correlations and therefore can be used to mine deeper alarm propagation paths like feedback loops. (2) It provides the time of origin of an alarm as it propagates through the process network, allowing operators to respond appropriately. The proposed method involves the following elements: modular segmentation, extraction of common cause variables, calculation of alarm propagation time between variables, calculation of ADTE, identification of an underlying alarm network, and tuning of relevant parameters. The Tennessee Eastman Process (TEP) is used to demonstrate the validity and superiority of the proposed ADTE-based alarm causality method.