An adaptive method to update alarm deadbands for non-stationary process variables

False alarms are detrimental to the safety of industrial processes and are often removed by alarm deadbands. However, most of the existing methods design alarm deadbands based on historic data samples under the assumption of stationarity, and may not achieve the desired performance in reducing false alarms for nonstationary process variables. This paper proposes an adaptive method to update alarm deadbands for nonstationary process variables to remove false alarms. Two technical challenges are addressed in this paper. First, time instants of statistically significant changes in process variables are detected based on mean values of maximum amplitude deviations between process variables and alarm trippoints. Second, when to update an alarm deadband width after detecting a mean change is determined based on a confidence interval of the designed width developed from Bayesian estimation rule. Numerical and industrial examples are provided to illustrate the proposed method and compare with alarm deadbands with fixed widths.

Automated summary

This paper proposes an adaptive method to update alarm deadbands for nonstationary process variables in order to eliminate false alarms. The method detects statistically significant changes in process variables and determines when to update the alarm deadband width based on a confidence interval.