Constellation, le dépôt institutionnel de l'Université du Québec à Chicoutimi

Résumé

Superhydrophobic surfaces serving as icephobic surfaces are a passive means of limiting the icing of surfaces. Ice nucleation time depends on not only liquid properties and environmental conditions but also surface features; however, it is challenging to investigate ice nucleation time and the influencing parameters simultaneously. This manuscript presents two approaches, experimental testing and machine learning, to study ice nucleation time on exposed surfaces. Hydrophobic/superhydrophobic silicone rubber surfaces were fabricated, and these surfaces varied in their wettability and roughness parameters. Superhydrophobic surfaces characterized by a higher arithmetic average, root mean squared, ten-point height, maximum height of the profile, and a Gaussian roughness distribution—skewness near 0—had longer ice nucleation times. We then used neural networks to model icephobicity in relation to ice nucleation time. The predicted ice nucleation time of the model, trained using some of the experimental results, demonstrated a good agreement with the experimental outcomes. Furthermore, this machine learning approach determined the relative importance of roughness parameters, surface wettability, temperature, and droplet volume in determining surface icephobicity. The proposed approach provides a starting point for studying heterogeneous ice nucleation prediction through an understanding of the key parameters required to optimize the icephobic behavior of superhydrophobic surfaces.