Machine Learning Algorithms for Flow Pattern Classification in Pulsating Heat Pipes

Owing to their simple construction, cost effectiveness, and high thermal efficiency, pulsating heat pipes (PHPs) are growing in popularity as cooling devices for electronic equipment. While PHPs can be very resilient passive systems, operation relies on the establishment persistence of slug/plug flow dominant regime. It is, therefore, paramount predict regime accurately a function various operating parameters design geometry. Flow pattern maps that capture regimes nondimensional numbers (e.g., Froude, Weber, Bond numbers) have been proposed literature. However, prediction patterns based deterministic models is challenging task ability explaining complex underlying phenomena or measure parameters, such bubble acceleration, which difficult know beforehand. In contrast, machine learning algorithms require limited priori knowledge system offer an alternative approach classifying regimes. this work, experimental data collected two working fluids (ethanol FC-72) PHP at different gravity power input levels, were used train three classification (namely K-nearest neighbors, random forest, multilayer perceptron). The previously labeled via visual using results. A comparison resulting accuracy was carried out confusion matrices calculation scores. algorithm presenting highest performance selected development map, indicated transition boundaries between annular flows. Results indicate that, once available, could help reducing uncertainty improve predictions