Self-Learning Digital Twins for Faster Time to Market

End-to-end digital twins of the manufacturing process are getting increasingly important tools for process development and characterization. In the past companies employed Design of Experiments (DoE) to characterize each unit operation of the manufacturing process individually. However, the mutual interplay of multiple unit operations has not been explored. Here we want to present how end-to-end process models can be used as a self-learning recommender system that suggests which runs should be planned to explore the relationship between CPPs and CQAs and which runs should be planned to study the interaction between unit operations. This reduces the amount of unnecessary experiments. We will show in a simulation study that is based upon an industrial case study that this recommender system – called holistic DoE (hDoE) - can save more than 50% of the experiments for a typical process characterization study. This will open up new possibilities in experimental planning, substantially reduce experimental costs and time-to-market.