Abstracting Process Mining Event Logs From Process-State Data To Monitor Control-Flow Of Industrial Manufacturing Processes

In industrial manufacturing, a production process usually consists of multiple manufacturing steps during the transformation of raw material into a complete product. Depending on varying product specifications, the production process may differ within the performed manufacturing steps and within the definition of process "normality", which refers to the multi-dimensional sensor data captured during the manufacturing process. Process Mining offers excellent process discovery and monitoring capabilities; however, it relies on well-defined event logs. In Industry 4.0 manufacturing, this requirement poses a significant challenge as data commonly generates by a range of low-level sensor measurements, i.e. process-state data such as temperature or humidity and product-state data such as quality measurements or material compositions. In this paper, we investigate the applicability of Process Mining methodologies in combination with concepts of data-driven Digital Twins to discover and monitor process control-flow. We introduce a concept called self-discovering and event-log-aware Digital control-flow Twins to bridge the gap of Process Mining, Digital Twins and continuous process data. We overcome the abstraction gap by generating discrete event logs backed by descriptive statistics of reoccurring process-state characteristics using unsupervised clustering. Our approach does not rely on preexisting information of manufacturing processes in the form of discrete event logs, making it suitable for various information systems that incorporate cyclic process behaviour. (C) 2022 The Authors. Published by Elsevier B.V.