Publication Date

Spring 4-13-2023


School of Engineering and Computational Sciences


Engineering: Industrial and Systems


data streams, statistical process control, quality control, aggregation, control charts


Operations Research, Systems Engineering and Industrial Engineering


This work outlines an approach for managing sensor data streams of continuous numerical data in product manufacturing settings, emphasizing statistical process control, low computational and memory overhead, and saving information necessary to reduce the impact of nonconformance to quality specifications. While there is extensive literature, knowledge, and documentation about standard data sources and databases, the high volume and velocity of sensor data streams often makes traditional analysis unfeasible. To that end, an overview of data stream fundamentals is essential. An analysis of commonly used stream preprocessing and load shedding methods follows, succeeded by a discussion of aggregation procedures. Stream storage and querying systems are the next topics. Further, existing machine learning techniques for data streams are presented, with a focus on regression. Finally, the work describes a novel methodology for managing sensor data streams in which data stream management systems save and record aggregate data from small time intervals, and individual measurements from the stream that are nonconforming. The aggregates shall be continually entered into control charts and regressed on. To conserve memory, old data shall be periodically reaggregated at higher levels to reduce memory consumption.