Flow-pattern recognition and dynamic characteristic analysis based on multi-scale marginal spectrum entropy

Rod bundles are widely used in energy chemical equipment. The prediction and dynamic characteristics of flow patterns have been expansively studied in the last few decades because of their significant influence on heat transfer. In most studies, flow patterns were usually predicted by analyzing the temperature distribution on the tube and by visualization using high-speed cameras. Further, the studies used a square channel composed of 6 × 6 rod bundles as the experimental section. However, owing to the limitations of various temperature sensors, the measurement accuracy is low and the experimental setup is unreliable. In the present study, fine bubble flow, fine churn-bubble flow, churn flow, and annular flow were experimentally observed, and the pressure-difference signals for fast reaction flow patterns were collected. An innovative method based on multi-scale marginal spectrum entropy was developed to identify flow patterns precisely, and the results revealed the dynamic characteristics. Multi-scale marginal spectrum entropy is a characteristic quantity, and the input support vector machine can recognize the flow pattern. The results indicate that the new method is feasible for the identification of flow patterns and analysis of the dynamic characteristics of different gas-liquid two-phase flows, which provide a reliable guide for flow-pattern control design and the safe operation of equipment.