High-Fidelity Simulation and Data-Driven Modeling of Drop Aerobreakup at Moderate Weber Numbers

The aerodynamic breakup of drops is essential to many industry applications such as fuel injection. Due to the complexity of the physics and the wide range of spatial scales involved, both experimental and numerical methods face significant challenges. This study investigates the deformation of water droplets in a uniform air stream through direct numerical simulations, focusing on the low to moderate Weber number regime. At low Weber numbers, drop oscillation is dominant, while moderate Weber numbers result in bag breakup. We model the drop interface using a superposition of spherical harmonic modes and perform a 2D axisymmetric, interface-resolved simulation using the mass-momentum consistent Volume of Fluid (VOF) method with the multiphase-flow solver Basilisk. Mesh refinement is achieved through Adaptive Mesh Refinement (AMR). The study examines the temporal evolution of drag and spherical harmonic mode coefficients, identifying correlations between them. Furthermore, we propose point-particle model that can be implemented in Lagrangian spray simulations involving large droplet population. Given the complexity of the interactions between droplet deformation and drag forces, analytical models are difficult to formulate. As a result, we employ a data-driven approach, utilizing high-fidelity DNS results to train a Non-linear Auto-Regressive with Exogenous input Neural Network (NARXNN). This model predicts the evolution of spherical harmonic mode coefficients, which are subsequently used as inputs to another NARXNN model to predict the drag coefficient.