Machine learning for heat transfer correlations

This paper explores machine learning approach as a heat transfer correlation. Machine learning significantly reduces the effort to develop multi-variable heat transfer correlations, and is capable of readily expanding the parameter domain. Random forests algorithm is used to predict the convection heat transfer coefficients for a high-order nonlinear heat transfer problem, i.e., convection in a cooling channel integrated with variable rib roughness. For 243 different rib array geometries, numerical simulations are performed to train and test ML model based on six input features. Machine learning model predicts closely to numerical simulation data with high determination of coefficient (R²), e.g., R² > 0.966 for the testing dataset. The capability and limitation of random forests algorithm are discussed with validation dataset.