Predicting Energy Budgets in Droplet Dynamics: A Recurrent Neural Network Approach

ABSTRACT The application of neural network‐based modeling presents an efficient approach for exploring complex fluid dynamics, including droplet flow. In this study, we employ Long Short‐Term Memory (LSTM) neural networks to predict energy budgets in droplet dynamics under surface tension effects. Two scenarios are explored: Droplets of various initial shapes impacting on a solid surface and collision of droplets. Using dimensionless numbers and droplet diameter time series data from numerical simulations, LSTM accurately predicts kinetic, dissipative, and surface energy trends at various Reynolds and Weber numbers. Numerical simulations are conducted through an in‐house front‐tracking code integrated with a finite‐difference framework, enhanced by a particle extraction technique for interface acquisition from experimental images. Moreover, a two‐stage sequential neural network is introduced to predict energy metrics and subsequently estimate static parameters such as Reynolds and Weber numbers. Although validated primarily on simulation data, the methodology demonstrates the potential for extension to experimental datasets. This approach offers valuable insights for applications such as inkjet printing, combustion engines, and other systems where energy budgets and dissipation rates are important. The study also highlights the importance of machine learning strategies for advancing the analysis of droplet dynamics in combination with numerical and/or experimental data.