Automated Hybrid Machine Learning System for Production

Hybrid Machine Learning (ML) models have potential for high performance, as they learn based on data and incorporate existing knowledge. In this work, an automated system is developed which creates the best possible hybrid ML model without Data Science expertise. First, the system architecture is defined including the breakdown of functionalities into a workflow along the sub-components of the system. Then, the system is validated on a use case from optics production, i.e., quality prediction during nonisothermal glass forming. A performance comparison between a baseline ML model with the automatically generated hybrid ML model is demonstrated.