A similarity analysis method for hybrid sequences of alarm events and trend events

Effective identification of faults or abnormal conditions can help operators make corrective decisions and plan equipment maintenance. Sequence matching and cluster analysis are important methods to distinguish different faults. Most existing sequence matching methods mainly focus on alarm event sequences, which reflect the amplitude change characteristics of process data. However, due to the complexity of the equipment and the coupling between variables, alarm event sequences caused by different faults may still assemble each other in a certain extent, which makes it difficult to distinguish faults based on alarms only. To solve this problem, this paper proposes a sequence similarity analysis method combining both alarm and trend events. A qualitative trend representation method is proposed to extract trend changes as trend events. A feature event fusion method is proposed to generate a hybrid sequence to distinguish different fault sequences. The proposed method is evaluated based on data generated by the Tennessee Eastman process model.