Application of machine learning-based object detection and tracking techniques for understanding bubble formation behavior on falling film evaporator tube surface

The evaporator is a key component in absorption heat pump systems, and its optimization significantly enhances industrial process efficiency. In this study, the heat transfer performance of a falling-film evaporator with small-diameter, smooth tubes is investigated under various operating conditions. A high-temperature heat source is supplied inside the evaporator tube, with inlet temperature ranging from 85 to 94 °C while the saturation pressure varies between 27 and 56 kPa and the film Reynolds number varies from 65 to 550. Transition Reynolds numbers are observed at 107 and 165 under saturation pressures of 32 and 45 kPa, respectively. Increasing the inlet temperature or decreasing the saturation pressure improves the wall heat flux, thereby improving heat transfer performance. Bubble formation on the tube surface is observed at elevated heat source temperatures. A machine learning-based model is developed and validated for the detection and tracking of boiling bubbles. The detection model, optimized through hyperparameter tuning, achieves remarkable accuracy, with mAP50 and mAP50–95 values of 99.2 and 88.0, respectively, after 678 training epochs. The tracking model shows excellent performance, maintaining frequent correct tracking with MOTA values of 98.36 and 65.61 for untrained videos under IoU_thres of 0.5 and 0.9, respectively. Bubble parameters extracted using the optimized model are applied to analyze bubble formation and to clarify the relationship between bubble dynamics and operating conditions. Overall, the proposed model offers a highly accurate and reliable approach for visualizing bubble behavior in falling-film evaporators, providing valuable insights for the advancement of industrial heat pump processes.