Bringing a Machine Learning Based Novelty Detection Software Tool from Research to Production

Abstract

This paper presents the process of bringing a machine learning based novelty detection software tool from research to production. Moreover, it sums up the necessary changes that needed to be done for developing a scientific software library into a software product with an application in space operations. This process considers the needs and expectations of all stakeholders. The system for which this process is shown is the Automated Telemetry Health Monitoring System (ATHMoS) developed at the German Space Operations Center of the German Aerospace Center. In its early phase as a research software, it paved the way for the novelty detection research. After its value for the satellite engineer’s daily work became visible, it evolved to a robust and resilient software tool that can be used in a productive environment to support the engineers in their routine work. Furthermore, the integration of the system into our Visualization and Data Analysis framework is explained. This framework has a web-based front-end for the interactive exploration and analysis of satellite telemetry data.