Achieving consistency and predictability in the liquid filling process of vials is crucial, with the goal of minimizing fill weight variations during sterile drug product development and manufacturing. This works presents a hybrid Computational Fluid Dynamics (CFD) and Machine Learning (ML) approach to predict droplet formation at the end of the fill cycle. We investigate the effects of contact angle, surface tension, and liquid viscosity on droplet behavior. A baseline volume of fluids (VOF) CFD simulation is established to provide accurate and efficient results. To explore the design space effectively, a Design of Experiments (DOE) approach is employed, utilizing a Latin Hypercube sampling method to systematically vary the three input parameters. The number of design points is adaptive, allowing for continuous refinement to enhance the accuracy of droplet size and shape predictions. Ultimately, the integration of ML techniques enables the development of a surrogate digital twin model that serves as a predictive tool for droplet characteristics, thereby stabilizing the filling process, minimizing fill weight variation, as well as reducing the risk of needle clogging and product loss.
