Collaborative Research: CDS&E: Learning Convective Heat Transfer from Mass Transfer Visualization

Understanding convective heat transfer is crucial in designing efficient heat exchangers, thermodynamic systems, biochemical fluid transport systems, and geothermal reservoirs. However, directly measuring temperature fields in unsteady and geometrically complex convection systems is often challenging. This project will develop machine learning based heat and mass transfer analogy functions that can approximate the temperature fields in convection systems from observed concentration fields. From simple to complicated flow passages and various flow regimes, this research will investigate the impact of dataset size and quality on learned functions. The educational goal of this project is to cultivate student interest in studying machine learning and advanced flow systems. This project will develop unique hands-on activities for K-12 students and provide research opportunities for undergraduate and graduate students on how to use machine learning for intriguing applications and ways to visualize convection.

Machine learning-based heat and mass analogy functions will enable spatiotemporally-resolved thermal characterization of thermofluidic devices where direct temperature measurements are extremely challenging. Until recently, heat and mass analogy functions are expressed in mathematical forms mostly derived from simple convection systems. This research aims to overcome the limitations of traditional analogy functions by leveraging the inference ability of machine learning models. Particularly, this project will study how the machine learning models can be used to infer the temperature field from a mass concentration field for laminar mixed convections. Heat and mass transfer analogy has been observed in systems with laminar mixed convection, but the derivation of analogy functions has been challenging due to the unstable and spatially inhomogeneous nature of the flows. Furthermore, this research will elucidate the requirements of datasets, e.g., size and resolution of thermofluidic field information, as well as the capabilities and limitations of the machine learning approach when studying complex thermofluidic phenomena.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.