PERFORMANCE ANALYSIS OF CAUSALITY INDICATION METHODS APPLIED TO INDUSTRIAL ALARMS
Industrial Alarms Management, Causality, Cross-Correlation, Granger Causality Test.
Industrial alarms are fundamentals tools for maintenance of safety and health from complex industrials process, between monitoring of thousands of process variables, the operator’s role is reduced to work on demand of alarms occurrence. However, an industrial alarms system poorly designed becomes inefficient against industrial abnormality situation. Among the possible degrading agents of an alarm system are the causal alarms, these are characterized by the difficulty of identification and by the high impact in monitoring industrial systems. Several works in the literature study possible ways of identifying causality patterns through information from industrials process; such ways are based in statistics and/or mathematical metrics to obtain the results. In this work, the performance of two causality indication methods applied to industrial alarms was analyzed, namely: Cross-correlation and Granger causality test. As alarm data have a discrete characteristic on the time, it was necessary to perform a preprocessing on them, via a signal smoothing technique, before applying both methods. To validate the methods under study, industrial alarms data from simulation scenarios of alarms generation with very realistic characteristic were used. Lastly, industrial alarms data from Tennessee Eastman Process benchmark were used as real application results. Therefore, the results indicate that, in general aspects, the Granger causality test performed a greater efficiency than the cross-correlation in the task of indication causal relations between industrial alarms.