Hybrid Modelling in Production: Approach and Evaluation

The Fourth Industrial Revolution accompanies exponentially growing data bases and decreasing costs for computing power and data storage and is focussed on digitisation and networking of the production environment. Thus, methods of machine learning (ML) are gaining popularity. Manufacturing was shaped by research regarding the mathematical and physical description (causality based) of production technologies in the past. Data-driven approaches usually are not built on these causalities and only consider data. Combining these approaches, called HybridModelling, helps overcoming the limitations of the single ones. The idea is to integrate the available knowledge and causalities into a data-driven model to obtain a better accuracy, with lesser training. Therefore, Hybrid Modelling can provide more accurate forecasts at an acceptable cost level, and has the potential to save resources, reduce schedules, and improve manufacturing quality. This article provides guidance and recommendations for the application of HybridModelling in production processes.