Visualization experiment and machine learning modeling for falling-film systems

Abstract

The liquid film coverage over the tube wall is important in flow hydrodynamics, and flow hydrodynamics are important in heat transfer performance in falling-film systems as well. An experimental setup is used with the aid of a high-speed camera to visualize the flow behavior, which is challenging to see with the naked eye. The set of experiments are carried out by varying the important working parameters such as Reynolds number (250−458) and inter tube distance (20/30/40 mm). The Sobel image analysis method is developed to quantify flow parameters such as jet diameter and axial film thickness. The liquid jet diameter increases with increasing tube spacing for the same Reynolds number and decreases with increasing inter tube distance for the same Reynolds number. The formation of connected droplets causes disturbances near the liquid jet head. The connected droplet phenomenon causes the liquid jet diameter to increase and decrease in this region. Furthermore, there is a distinct pattern in the axial film coverage beneath and above the tube wall. Three data sets are generated for each parameter using an image analysis approach, which can then be used to train and develop machine learning models. The findings indicated that the Decision Tree, Random Forest, and Gradient Boosting models consistently better performed the other models on all features. Although a few models have lower error percentages below the datum line, consistency in each feature is still required. The results of the flow visualization experiments, as well as the machine learning models developed, can be used to quantify flow parameters and improve flow characteristics for better design.