Operational zone-specific univariate alarm design for incipient faults

Abstract

Alarm systems are essential components of industrial process monitoring, supporting both safety and operational efficiency by detecting deviations from normal conditions. Traditional alarm design methods often assume stationary, which limits their ability to reflect the evolving nature of incipient faults. These faults develop gradually and, if not properly addressed, can lead to critical failures. Timely and accurate detection is therefore vital to minimize false alarms, reduce missed detections, and improve response effectiveness. This study proposes a time-variant statistical modeling framework to characterize the behavior of process variables affected by incipient faults. A new alarm system design methodology is introduced, guided by three key performance indices: Missed Alarm Rate (MAR), False Alarm Rate (FAR), and Average Alarm Delay (AAD). The methodology uses the Narrowest Over Threshold change-point detection technique to segment the process into distinct operational zones, including the Normal Operating Zone (NOZ), Rising Zone (RZ), Fault Zone (FZ), and Return to Normal (RTN). By employing a piecewise time-variant model, the alarm system’s performance is assessed in a manner that captures local trends and transitions. The resulting indices are dynamic, offering a more detailed projection of the process variable’s behavior over time. In particular, the AAD metric reflects realistic delay patterns and avoids the misleading interpretations often associated with stationary models. The proposed method is validated through Monte Carlo simulations and demonstrated using the Tennessee Eastman Process benchmark. Results show that the time-variant model provides a more accurate and interpretable representation of process dynamics and alarm behavior than traditional approaches.