Optical to infrared mapping of vapor-to-liquid phase change dynamics using generative machine learning

Abstract

Infrared thermography is a powerful tool for studying liquid-to-vapor phase change processes. However, its application has been limited in the study of vapor-to-liquid phase transitions due to the presence of complex liquid dynamics, multiple phases within the same field of view, and experimental difficulty. Here, we develop a calibration framework which is capable to studying one of the most complex two-phase heat transfer processes: dropwise condensation. The framework accounts for non-uniformities arising from dynamic two-phase interactions such as droplet nucleation, growth, coalescence, and departure, as well as substrate effects particularly observed on micro- and nanoengineered surfaces. This approach enables high-resolution temperature measurements with both spatial (12 $μ$m) and temporal (5 ms) precision, leading to the discovery of local temperature phenomena unobservable using conventional approaches. These observed temperature variations are linked to droplet statistics, showing how different regions contribute to local condensation heat transfer. We extend the developed method to quantify local thermal parameters by fusing it with a generative machine learning model to map visual images into temperature fields. The model is informed of the physical parameter by incorporating vapor pressure embedding as the conditional parameter. This work represents a significant step toward simplifying local temperature measurements for vapor-to-liquid phase change phenomena by developing a methodology as well as a machine learning approach to map local thermal phenomena using only optical images as the input.