Event-Triggered Bayesian Control Chart

Control charts are practical tools for fault detection and recovery. However, traditional control charts rely on random samples collected from a production process at fixed time intervals, causing late detection if sampling intervals are too long or excessive sampling if the intervals are too short. In “Event-Triggered Bayesian Control Chart,” Abbou and Makis develop a novel control chart leveraging real-time data from smart sensors to jointly decide when to collect samples and when to stop the production process, leading to quick fault detection and recovery using few samples. Applying optimal stopping theory and dynamic programming analysis, the authors establish the average-cost optimality of their control chart and propose an efficient procedure for computing the optimal sampling and stopping thresholds. Through an empirical study, the control chart is shown to achieve substantial cost savings compared to benchmarks. Furthermore, thanks to its event-triggering mechanism, the proposed control chart requires little data communication from sensors, which is crucial from an energy-efficiency perspective.