Modeling of frost formation under cryogenic conditions based on the hybrid physics-informed data-driven method

Physical properties of a frost layer under cryogenic conditions were predicted by utilizing the hybrid physicsinformed data-driven method. Frosting experiments were performed on a cryogenic cooling surface under forced convection conditions. Subsequently, a model was developed by leveraging both experimental data and a partial differential equation (PDE) representing physical laws. The suggested model was validated using the determination coefficient (R2) and RMSE values. The R2 scores of frost thickness and density were 0.988 and 0.981, with corresponding RMSE values of 0.173 and 2.448, respectively. The model's ability to extrapolate and predict cryogenic frost growth behavior beyond its training range was also evaluated. Despite being trained on only half of the total three-hour experimental duration, the model could predict frost growth behavior in the subsequent periods. Accordingly, it effectively addressed the data imbalance problem between frost thickness and density by concurrently incorporating both data and governing equations.