Optimization Approaches To Multi-Variable Alarm Thresholds With Priorities In Process Productions

Alarm systems are eligible to effectively monitor operational conditions in industrial processes so as to ensure safe and stable operations. However, there are always flooding alarms in practical process productions, which are identified to be mainly caused by unreasonable specifications of alarm thresholds. It is conceivable that a most effective solution is to optimize the alarm thresholds. Considering strong interactions among the process variables, a novel approach to optimization of multi-variable alarm thresholds is explicitly introduced, which can deal with alarm floods more reasonably and guide operations more effectively.This thesis initially conducts a review on process alarm systems and alarm threshold optimization methods, introducing performance assessment indices of alarm systems. Subsequently, a novel method for multi-variable alarm threshold optimization which combines a correlation analysis of multi-variable alarms is proposed. The details are presented as follows.Alarm events are extracted from process historical data before pseudo continuous time series are generated from the original binary data by Gaussian kernel methods. To deal with the resultant pseudo-data, correlation coefficients among alarms are calculated with Pearson’s correlation coefficient analysis. Alarms are clustered based on the correlations and the alarm priorities are created by weight assignment approaches based on the difference driven decision principles.In the view of clustering and ranking of the alarms, aiming at handling alarms of the highest priority, a mathematical model with respect to probabilities of missed and false alarms is established to optimize the alarm thresholds along with gradient descent solvers.The TE process is employed for the application research. Comparing the proposed approach with univariate alarm threshold optimization methods, the numbers of both false and missed alarms can be considerably reduced, which demonstrates the potential benefits of the contribution.