Machine learning-based predictive modeling of contact heat transfer

Heat transfer phenomena at the interface between two contacting solids are highly complex involving multiple influencing factors. Over the years, a large amount of experiments were carried out to determine the contact heat transfer coefficients between two dissimilar joint materials. However, there are still no existing theoretical or physics-based models that satisfactorily predict the contact heat transfer coefficients. By taking advantage of the existing data, in contrast, machine learning promises a powerful method, capable of predicting the contact heat transfer coefficients for different material pairs and contact conditions. This research introduces a robust machine learning-based model that succeeds in precisely estimating the heat transfer across the interfaces between glass and steel, a material pair widely used in hot forming of glass. The data used for training and validating the machine learning models were determined experimentally by means of infrared thermography. The datasets consisted of contact heat transfer coefficients with dependence on three factors - interfacial temperature, contact pressure, and surface finishes. Aim of this study is to analyze the prediction accuracy and interpretability of various supervised learning algorithms in order to realize the machine learning models that are able to capture the underlying physics governing the heat transfer phenomena at the glass-mold interface. Finally, the results were compared with those estimated by a theoretical model and a numerical simulation model. The comparison demonstrates enhancements in prediction accuracy enabled by the data-driven method. This study indicates accurate and efficient strategies for solving thermal problems in hot glass forming processes.