Machine Learning and Complex Event Processing

Abstract

In the Industrial Internet of Things, cyber-physical systems bridge the gap between the physical and digital world by connecting advanced manufacturing systems with digital services in so-called smart factories. This interplay generates a large amount of data. By analyzing the data, manufacturers can reap many benefits and optimize their operations. Here, the value of information is at its highest with low latency to its emergence and its value decreases over time. Complex Event Processing (CEP) is a technology, which enables real-time analysis of complex events (i.e., combined data values from different sources). In this way, CEP assists in the identification and localization of anomalous process sequences in smart factories. However, CEP comes with limitations that reduce its effectiveness. Setting up CEP requires in-depth domain knowledge and is primarily declarative as well as reactive by nature. Combining CEP with machine learning (ML) is a possible extension to circumvent these technological limitations. However, there is no up-to-date overview on the integration of both paradigms in research and no review of their transferability for application in smart factories. In this article, we provide (1) a synthesis of research on the integration of CEP and ML identifying supervised learning as the predominant approach, and (2) a transfer of potentials for the use in smart factories. Here, reactive and proactive policies are used in equal frequency.