Cross-process alarm flood similarity analysis based on abstracted alarm descriptors

A typical industrial facility is usually large-scale, consisting of many processes or units; some of them could be similar in functionalities or identical in architecture. Failures of the same types of equipment may lead to analogous consequential alarms, which usually have different tag names, but the same types. Therefore, if such similar alarm floods across different processes are captured, the results could help discover common root causes of alarm floods in similar processes or units, and may also give general solutions to address these alarm floods. Motivated by such a practical problem, this paper proposes a new method to identify similar alarm floods across different processes or units. This method has three main steps: 1) distill key words from detailed alarm descriptions in the alarm and event (A&E) log through word preprocessing; 2) reconstruct abstracted alarm descriptors based on key words to generalize alarm representations; 3) conduct cross-process alarm flood similarity analysis through sequence alignment. The effectiveness of the proposed method is demonstrated by an industrial case study involving real alarm data from a large-scale industrial facility.