This study aims to develop, implement and validate computationally efficient algorithms based on deep neural networks (DNNs) and finite volume simulations for studying coaxial droplet generators.
The approach integrates high-fidelity volume of fluid (VOF) simulations with computational intelligence techniques. VOF simulations generated reference data sets for training DNNs, enabling the prediction of droplet formation under unexplored conditions while significantly reducing computational costs.
The combined VOF–DNN methodology allows identification of characteristic flow patterns and establishes robust correlations between the system’s operational parameters and resulting flow regimes and droplet sizes. Cross-validation against VOF simulations and experimental bibliographic data demonstrates that neural-network-based predictions reliably predict coaxial droplet generation behavior. The computational cost of the proposed method is negligible compared to standard numerical simulations.
This study presents a novel hybrid framework that leverages computational intelligence to complement traditional high-fidelity simulations, enabling rapid, accurate and cost-effective exploration of coaxial droplet generators. The methodology enhances predictive capabilities for droplet generator design, bridging the gap between detailed numerical modeling and practical R&D requirements.
