Deep Learning to Droplet Spreading of the Rough Surfaces Using LST-Net

The study of droplet spreading on rough surfaces is important for several reasons across various fields, including physics, chemistry, materials science, biology, and engineering. Surface wettability plays a significant role in biological systems, including cell adhesion, protein adsorption, and biomaterial compatibility. Understanding and controlling surface wettability are essential for designing biocompatible materials, medical implants, and drug delivery systems. Analyzing surface wettability with droplet spreading, surface properties, and energies is vital in characterizing the surface or materials. However, it is computationally very expensive to simulate a model to analyze the wettability and characterize the surface. In this paper, we developed and validated a model using the two-dimensional pseudo-potential multiphase lattice Boltzmann method to generate training data for different kinds of rough surfaces. We then employed the LST-Net model, which is designed for long-term spatio-temporal datasets of droplet density, to predict the droplet spreading of different rough surfaces and identify the correlation between surface roughness and droplet spreading for constant fluid properties. The R² score, RMSE, and MAE metrics are used to evaluate the performance of the models. The proposed model can make a droplet spreading prediction for arbitrary surface and fluid properties. Our study provides various insights into the appropriateness of the LST-Net model for the prediction of droplet spreading. The comparison between numerical results and our model predictions showed a good agreement of more than 98%.