Molecular dynamics data-driven study of leidenfrost phenomena in context to liquid thin film phase transformation

In many micro and nanoscale applications of thin film phase transition, identifying the circumstances that allows stable liquid contact with a heated surface is critical. Using molecular dynamics (MD) simulations, our current data driven machine learning-based study attempts to examine the properties of liquid interaction with a solid surface at nanoscale. A few nanometer-thick liquid argon layer over a platinum surface has been used to simulate a liquid-solid contact system. The wall temperature is raised linearly after necessary initial equilibration of the entire system, with different boundary heating rates for various surface wetting conditions, namely hydrophobic, hydrophilic, and superhydrophilic. Our current investigation shows that the heating condition, liquid film thickness, and surface wetting condition all have a significant impact on the type of liquid wall contact that persists during the phase transition phenomena of thin liquid argon film (i.e., normal evaporation or explosive boiling). In the event of normal evaporation, a stable liquid contact with the solid surface continues, however in the case of explosive boiling, the liquid film is splashed away from the solid surface resembling the macroscopic Leidenfrost effect. For various system configurations in regard to liquid initial film thickness, liquid heating rate as well as solid-liquid interaction, a wide variation of the onset time as well as the wall temperature of boiling explosion have been found in the present study. An accurate mapping of the Leidenfrost conditions in context to nanoscale thin film liquid-vapor phase process has been generated using a predictive model based on deep neural networks that has been designed, trained and cross validated against molecular dynamics data of the present study.