Detecting outliers and anomalies to prevent failures and accidents in Industry 4.0

Abstract

Due to the rapid development of IT and sensorial technologies in the previous decade, the fourth industrial revolution, Industry 4.0 has become a reality. Many new terminologies and approaches were introduced in this concept. Combined with the new technologies and advantages of the 5G telecommunication networks it is possible to gain much faster response times over a reliable infrastructure which paves the way towards efficient industrial control. Based on historical datasets and using the knowledge of the domain specific experts it is possible to label those records with the corresponding key performance indicators which stand out as anomalies or outliers from the normal operation behavior. This paper is concerned with developing new algorithms for fast identification of outliers, which ensure high operational reliability. Our approach is statistical driven, once the usually high dimensional data is mapped into a low dimension for tractable analysis, outliers are captured by statistical decision techniques. As the corresponding numerical analysis performed on industrial data has demonstrated, the method has a reliable performance. Thus, the new methods can prove to be appropriate tools for industrial control in Industry 4.0 environment.