Towards reproducible wetting studies: Automated contact angle determination by molecular simulations

Abstract

Accurate determination of contact angles in molecular modeling traditionally requires manual intervention and parameter tuning, which limits reproducibility and efficiency. In this paper, we present PANDA-NN, a fully automated pipeline that determines liquid–liquid–solid contact angles. Our method uses a PointNet++ neural network to classify the shapes of interfacial boundaries between immiscible liquids in a slit pore, allowing us to automatically select the appropriate analytical equation to describe the density profile. Interfacial shape classification achieves robust performance across all classes: the confidence of the model in surface type classification of MD systems was greater than 99%. Then we use gradient optimization procedure to calculate the angle by minimizing the difference between simulated and theoretical density profiles. Molecular dynamics simulations demonstrate the high precision of the presented method (mean absolute percentage error$<$2°) without manual pre-processing. PANDA-NN improves the reproducibility of nanoscale wetting studies by eliminating operator bias, and facilitates investigations of interfacial phenomena. This provides researchers with a robust tool to quantify molecular wetting with a high level of automation, potentially accelerating materials discovery for applications ranging from enhanced oil recovery to microfluidic device development.