Structural-material-operational performance relationship for pool boiling on enhanced surfaces using deep neural network model

Abstract

An accurate predictive model of the enhanced pool boiling heat transfer on various surface modifications is essential to operate the pool boiling and design the optimal surface designs. However, the existing predictive models generally predict the enhanced pool boiling heat transfer on various surfaces with very large errors as high as ±50%, mainly due to the complex nature of the pool boiling processes. In this study, we unlock the complex relations among four geometrical, nine thermophysical properties, and two operational conditions to accurately predict the Heat Transfer Coefficient (HTC) on the enhanced surfaces using an optimized Deep Neural Network (DNN) model. The six dimensionless numbers are identified based on geometries, operation conditions, and thermophysical properties, which are used as input parameters for the DNN model for the first time. This results in the Mean Absolute Percentage Error (MAPE) below 5%, compared to the existing empirical correlations having 5.04–45.37% MAPE on the selected 1256 data points. Also, the developed DNN model outperforms the prediction accuracy of the existing correlation for the data in much different experimental conditions, showing the 20% MAPE for the pre-trained DNN model (without additional training) and 38% MAPE for the existing correlation. Moreover, the sensitivity analysis was performed to identify the key dimensionless parameters for the HTC on the enhanced surfaces. The developed DNN model with the dimensionless parameters shed light on understanding the complex pool boiling process on the enhanced surfaces.