Automatically Detecting Operating Procedures from Raw Process Data using Filtering, Machine Learning, and Clustering Methods

A data analysis framework is proposed to identify unlogged recurring process transitions from raw process data. Transitions are a ubiquitous part of any continuous process occurring every time an operating procedure includes opening valves, starting pumps, or taking equipment in or out of service. However, since many of these procedural steps are accomplished manually there are no data logs that can be used with traditional data mining techniques. In this work we investigate using data filters and machine learning techniques to identify recurring transitions from process data. The information extracted can then be used to identify clusters of data and detect similar patterns representing recurring process transitions thus automatically creating operating procedure data logs.