Flow Characterization in Inclined Intermittent Flow Using Improved U-Net and Particle Image Velocimetry

Gas-liquid intermittent flow is of great engineering significance for the design and optimization of pipeline systems in complex terrains, while current research on the flow characteristics influenced by pipe inclination angle remains insufficient. In this study, the flow characteristics of intermittent flow in horizontal, 5°, and 10° inclined pipelines are systematically analyzed by combining the improved deep learning model with particle image velocimetry (PIV) technology. First, the improved U-Net model integrating the Convolutional Block Attention Module (CBAM) is employed to achieve high-precision segmentation of the gas-liquid phase. Experimental results show that the improved model achieves 98.83% Pixel Accuracy (PA) and 97.01% Mean Intersection over Union (MIoU) in phase segmentation tasks, which surpasses benchmark models including DeepLabV3 and HRNet. By comparing the three inclined configurations, the pipe inclination angle is found to significantly alter the flow structure by increasing the axial gravitational component, which manifests in reduced length of elongated bubbles, decreased thickness of liquid film, and enhanced asymmetry in flow velocity distribution. Furthermore, the increase of the inclination angle will trigger the flow regime transition to unstable slug flow. The flow pattern transition boundary is established based on the mixed Froude number (Fr), revealing that the critical Fr value for the transition from slug flow to plug flow in the 10° inclined pipe decreased by 22% compared to the horizontal pipe. Research results provide essential parameter references for flow stability prediction and numerical simulation in pipeline design across complex terrains.