Statistical Process Control for Real-Time Industrial Data Streams

Abstract

Effective quality monitoring in modern industries requires robust statistical process control (SPC) methods capable of detecting shifts in real-time data streams. This study compares the performance of Shewhart, Cumulative Sum (CUSUM), and Exponentially Weighted Moving Average (EWMA) control charts using a simulated dataset of 300 observations, with a deliberate mean shift introduced at observation 150. Descriptive results showed that the process mean increased from 49.91 to 53.56 (a shift of +3.65 units) while variability remained stable. The Shewhart chart detected the shift at observation 158 with only 11 signals, demonstrating reliability for large shifts but slower detection. The EWMA chart (λ = 0.2) detected the change at observation 156, generating 142 alarms, offering greater sensitivity to gradual changes. The CUSUM chart signaled the shift earliest, at observation 17, but produced 209 alarms, reflecting excessive sensitivity and false positives. Overall, the findings show that CUSUM is most effective for rapid detection, EWMA balances responsiveness with stability, and Shewhart provides robustness for larger shifts. These results highlight the need to select SPC methods according to process characteristics and tolerance for false alarms.