Data mining algorithm for correlation analysis of industrial alarms

Preprocessing of Alarm Data for Data Mining

“…The latter approach is more common in alarm research. 9,10,22 In this paper, we consider both cases and discuss the differences whenever necessary. Let ON 1 , ON 2 , ON 3 , ... be the sequence of alarm activation times, and OFF 1 , OFF 2 , OFF 3 , ... the sequence of RTN times for a unique alarm.…”

“…Many research results confirm that there are some correlations between alarm data. 9,10,22 Here, we use this correlation, captured by temporal patterns, to impute missing alarm messages. Kong et al proposed a method to discover multitemporal patterns in large databases.…”

“…3 This can generally bring the alarm system within the limits defined by alarm standards. Many resources and methods have been suggested over the past decades to reach this goal, 4 including alarm processing techniques (e.g., filters, 5 deadbands and delay timers, 6 and time-deadbands 7 ), graphical tools, 8,9 correlation methods, 10 cause-and-effect analysis, 11 signed digraphs, 12 and root-cause analysis. 13 Data mining, which is a general process of discovering patterns in data, 14 has also been used for the analysis of alarm systems as early as 2019.…”

“…9 Other works include a general weight-based multistate sequential algorithm to find temporal patterns between alarms, which is particularly useful in detecting successive alarms. 10 From a different perspective, weighted fuzzy association rules were developed by Wang et al to identify cause-and-effect relationships of frequent alarms. 18 Although the proposed method reduces the search region, it requires process data for implementation, which is a limitation.…”

“…Or, Hu et al utilized an itemset mining method to detect frequent alarm patterns during alarm floods . Other works include a general weight-based multistate sequential algorithm to find temporal patterns between alarms, which is particularly useful in detecting successive alarms …”