Development of two-phase flow regime map for thermally stimulated flows using deep learning and image segmentation technique

•A deep learning technique was developed to identify a two-phase flow using multi-class segmentation.•The effects of the bubble length, velocity, and acceleration on the flow pattern are described.•A flow regime map for CLPHP was developed.•Bubbly, slug-plug, elongated plug, and annular/semi-annular flows were studied. Accurate prediction of the flow patterns is important for understanding the behavior and operation of thermo-fluidic phenomena in two-phase systems. Considering the importance of thermally induced flows in capillaries and their use in various processes, the thermo-hydrodynamics of two-phase flows in pulsating heat pipes (PHPs) have been studied. In the present research, flow regimes were identified in capillaries where the mass flow rate is difficult to deal with. A deep learning technique has been carried out to classify the flow patterns and extract their attributes in a particular capillary using multi-spectral segmentation. The modified Weber, Froude, and Bond numbers are used by considering the absolute velocities, accelerations, and bubble lengths for classifying the flow regime. Flow regime maps were plotted based on the proposed non-dimensional numbers by analyzing experimentally measured data sets and visual observation for methanol as a working fluid at six different heat levels (18,38,44,50,60, and 70 W). Four different flow regimes have been identified and classified as a bubbly flow, a slug flow, an elongated plug flow (transitional flow), and an annular/semi-annular flow, and a new flow regime map for thermally induced two-phase flows in a PHP is presented. The flow regime map gives quantifiable criteria for the calculation of bubbly, slug, and annular flow patterns. The present study will be useful for researchers analyzing a large number of two-phase flow images and to design two-phase passive heat transfer devices based on the flow regime map.