Real-time anomaly detection in distributed manufacturing: a high-resolution statistical SCADA framework

Abstract

Real-time anomaly detection in distributed manufacturing environments often relies on rigid timeout thresholds that may fail to identify missing workpieces or abnormal mechanical behavior until a critical failure occurs. This study presents a high-resolution, statistics-based anomaly detection framework for a Fischertechnik 24 V platform coordinated through a centralized Python supervisory control and data acquisition (SCADA) layer with multiple ESP32 microcontrollers. A statistically normal behavior was established from 10 identical production runs at 10 ms resolution using the mean and standard deviation of movement durations and event timing. The real-time SCADA Watchdog was evaluated through repeated intentional physical fault injection across a complex manufacturing process. The results showed that the framework successfully detected missing/stuck expected events and duration-based temporal deviations in real time, provided precise fault localization, stopped the process for critical missing/stuck anomalies, and logged non-terminal duration anomalies under the implemented monitoring configuration. These findings demonstrate that statistical monitoring can provide effective real-time anomaly detection for distributed manufacturing systems.