A Benchmark for Real-Time Anomaly Detection Algorithms Applied in Industry 4.0

Abstract

Industry 4.0 describes flexibly combinable production machines enabling efficient fulfillment of individual requirements. Timely and automated anomaly recognition by means of machine self-diagnosis might support efficiency. Various algorithms have been developed in recent years to detect anomalies in data streams. Due to their diverse functionality, the application of different real-time anomaly detection algorithms to the same data stream may lead to different results. Existing algorithms as well as mechanisms for their evaluation and selection are context-independent and not suited to industry 4.0 settings. In this research paper, an industry 4.0 specific benchmark for real-time anomaly detection algorithms is developed on the basis of six design principles in the categories timeliness, threshold setting and qualitative classification. Given context-specific input parameters, the benchmark ranks algorithms according to their suitability for real-time anomaly detection in production datasets. The application of the benchmark is demonstrated and evaluated on the basis of two case studies. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.