Empirical modeling and hybrid machine learning framework for nucleate pool boiling on microchannel structured surfaces

Abstract

Micro-structured surfaces influence nucleation characteristics, capillary liquid wicking, and bubble dynamics besides increasing the heat transfer surface area, thus enabling efficient nucleate boiling heat transfer. Modeling the pool boiling heat transfer characteristics of these surfaces under varied conditions is essential in diverse applications. A new empirical correlation for nucleate boiling on microchannel structured surfaces for heat flux input condition has been proposed with the data collected from various experimental studies in the literature, as the accuracy and applicability of the existing correlations are limited. This study further examines various Machine Learning (ML) algorithms and Deep Neural Networks (DNN) for predicting the nucleate pool boiling Heat Transfer Coefficient (HTC) on the microchannel structured surfaces for heat flux and wall superheat input conditions. With an aim to integrate both the ML and domain knowledge, the study proposes a Physics-Informed Machine Learning-Aided Framework (PIMLAF), which employs an empirical correlation as the prior physics-based model and a DNN to model the residuals of the prior model. This hybrid framework achieved the best performance in comparison to the other ML models and DNNs for wall superheat as an input. This framework is able to generalize well for different datasets because the empirical correlation provides the baseline knowledge of the boiling behavior. The verification of the models involves two tests — one against the blind data from the ML pool and the other against the blind data from other than the ML pool. Also, SHAP interpretation analysis identifies the critical parameters impacting the model predictions and their effect on HTC. This analysis further makes the modeling more robust and reliable.