A Study on Machine Learning Models’ Capability as an Alternative for CFD in Modeling Heat Transfer

This research assesses the viability of utilizing machine learning models as alternatives to computational fluid dynamics for heat transfer modeling. Driven by a heightened interest in evaluating the benefits of machine learning for precise predictions in forced convection heat transfer, the study investigates the potential of artificial neural networks, super-gradient boosting, and random forests as alternatives to traditional methods. Employing artificial intelligence algorithms and implemented through Python software, the methodology conducts a meticulous analysis of a dataset comprising 210 data points. The dataset includes critical heat transfer parameters such as nanoparticle characteristics, size, Reynolds number, Nusselt number, and volume fraction. The selected machine learning algorithms are systematically applied to predict forced convection heat transfer outcomes, and their accuracy is rigorously assessed through comparisons using machine learning R-Squared, Mean Absolute Error, and Root Mean Squared Error values. The results demonstrate promising predictive capabilities, with super-gradient boosting, random forest, and artificial neural network models achieving accuracies of 91%, 90%, and 86%, respectively. The corresponding mean squared error values of 1.07, 1.65, and 16.1 underscore the high accuracy and predictive prowess of these machine learning models in simulating forced convection heat transfer processes.