Sensor Data Stream Selection and Aggregation for the Ex Post Discovery of Impact Factors on Process Outcomes

Abstract

One target of process analysis, monitoring, and prediction is the process outcome, e.g., the quality of a produced part. The process outcome is affected by process execution data, including (external) sensor data streams, e.g., indicating an overheating machine. Challenges are to select the “right” sensors –possibly a multitude of sensors is available– and to specify how the sensor data streams are aggregated and used to calculate the impact on the outcome. This paper introduces process task annotations to specify the selected sensors, their aggregation, and initial impact functions. The initial impact functions are then refined, e.g., threshold values and the impact of sensor data streams are determined. The approach is prototypically implemented. Its applicability is demonstrated based on a real-world manufacturing scenario.