ACTIVE LEARNING BASED PROCESS MONITORING - AN APPLICATION REPORT

摘要

Automated production systems require reliable monitoring systems. Unfortunately, the effort of adapting classical monitoring systems to a specific production process is very high and hence only pays off for processes with large batch sizes. The Fraunhofer Institute for Manufacturing Engineering and Automation developed a new approach for monitoring systems that explicitly fits the needs of small batch production processes. This new approach uses active learning methods to model the underlying production process. As a result, this monitoring system requires only a small set of training data for an initial adaption to the process. Subsequently, active learning algorithms refine the monitoring system if required. This refinement happens while the monitored production process is running. By using this new approach, the effort of initially adapting the monitoring system can be reduced to a minimum. Therefore, it is worthwhile even for small batch processes. This paper briefly presents the basic ideas of active learning and shows how this concept can be applied in process monitoring systems. Subsequently, it describes the application of such a monitoring system in a real-world injection molding process.