Constructing workflow models of alarm responses via trace labeling and dependency analysis

Human factors are critical to industrial alarm monitoring. Learning how operators cope with alarms and manage abnormal situations would be helpful to prevent the repetition of human errors and provide decision supports for realtime alarm responses. The events of operator actions and alarm notifications are recorded in Alarm & Event (A&E) logs, making it available to capture the experience of skilled operators from historical data. This paper presents a data driven method to construct workflow models of operator actions in response to alarms, either univariate or multivariate, from historical A&E logs. The discovery of operator responses is formulated as a causal net model construction problem and solved via a process discovery method comprised by two steps, namely, the trace labeling and the dependency analysis. The effectiveness of the proposed method is demonstrated based on A&E data from a real industrial facility.