A Data-Driven Framework for Anomaly Detection in Industrial Systems Using Log Data

Abstract

Reliability engineering plays a crucial role in modern industrial systems, aiming to minimize costly downtime and prevent safety hazards. The feasibility of automating this process largely depends on the available data. While sensor-level data analysis can reveal crucial insights into system health and operational states, often only event-driven log data is available due to technical or cost constraints. Although extensive research on log-based failure diagnosis has been conducted, particularly in the information technology (IT) sector, the application of these methods remains challenging in real-world industrial systems. Hence, we propose a modular and interpretable framework for log-based anomaly detection in industrial systems to address the interpretability and reliability shortcomings of previous approaches. The results obtained from a real-world production system validate the framework’s ability to support timely root cause analysis and facilitate predictive maintenance.

Authors 5

1. Merle Hewing

   Aixtron (Germany)

   Affiliation as printed

   AIXTRON SE,Herzogenrath,Germany

2. Yuanchen Zhao Aachen Chair of Information and Automation Systems for Process and Material Technology

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University,Chair of Information and Automation Systems for Process and Material Technology,Aachen,Germany

3. Manuel Vossel

   Aixtron (Germany)

   Affiliation as printed

   AIXTRON SE,Herzogenrath,Germany

4. Paul Nieschler

   Aixtron (Germany)

   Affiliation as printed

   AIXTRON SE,Herzogenrath,Germany

5. Tobias Kleinert Aachen Chair of Information and Automation Systems for Process and Material Technology

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University,Chair of Information and Automation Systems for Process and Material Technology,Aachen,Germany