Non-Convex Operating Zone Establishment and Alarm Limit Design for Multivariate Alarm Systems Based on Grid Clustering

Alarm systems are core components in large complex industrial facilities to ensure safe and efficient operation. Traditional univariate alarm system design methods ignore the correlation of process variables and thus may generate massive false alarms and missed alarms. An effective way of designing a multivariate alarm system is to establish the normal operating zones (NOZs) based on historical data under normal conditions. These NOZs serve as the basis for setting alarm limits in multivariate alarm systems. Given that the actual correlations between related variables are frequently unknown, a robust modeling approach is necessary to effectively handle both convex and non-convex NOZs. Thus, this paper presents a systematic method for nonconvex operating zone establishment and alarm limit design for multivariate alarm systems. A multivariate non-convex NOZ is established based on grid clustering. Then, the multidimensional grid boundary is extracted and used to determine the alarm limits for real-time samples. A numerical example is used to demonstrate the effectiveness of the proposed method.