Evolution and identification method of gas-liquid flow pattern in M-type rigid crossover pipe based on deep learning

Abstract Multiphase flow pattern identification is one of the common fluid phenomena in engineering, which involves many fields such as energy, environment and chemical industry. The study optimizes an efficient network model to obtain better performance with limited computational resources through a composite model scaling approach. The simulation part of the Metal-type rigid spanning receiver has been carried out using ANSYS FLUENT software. A fluid volume fraction model has been utilized in the numerical simulation to monitor the interfaces of various fluids, guaranteeing that the total of the phase-by-phase volume fractions is always 1. To increase the accuracy of the turbulent motion simulation, the study has also used a technique based on the Reynolds-averaged Navier–Stokes system of equations. In the realm of multiphase flow pattern identification and analysis, the study’s suggested method has significant theoretical value as well as real-world applications. Additionally, the findings also have potential applications in air conditioning and heating systems, where efficient multiphase flow control can improve energy efficiency and overall system performance. The experimental results revealed that the gas–liquid two-phase flow showed a dispersed bubble distribution in the initial stage, which coalesced over time to form a stratified flow. The Efficient Network-B5 plus Adam optimizer network model performed best after testing on 3271 flow pictures with an accuracy of about 97%. Its stability and generalization ability were verified by the relationship between training period and loss function and learning accuracy. This approach can be extended to optimize airflow patterns and heat exchange in air conditioning and heating systems, leading to more efficient thermal management. The study uses an efficient network model to identify multiphase flow patterns in M-type rigid crossover pipes, providing important data support for the optimization design of subsea production systems.