separated flow

Flow separation in overexpanded single expansion ramp nozzles (SERN) involves complex phenomena, such as shock waves, expansion waves, turbulent boundary layers, and shear layers. Computational fluid dynamics plays a crucial role in studying unsteady flow behaviour in supersonic nozzles, allowing for an investigation into the dynamic flow field characteristics. However, the application of OpenFOAM as a numerical tool for studying SERN in the field of compressible flows, particularly in the overexpansion state where the flow field characteristics are more complex, has received relatively less attention. In this study, the flow field characteristics of an overexpanded SERN under different turbulence models are investigated through a combination of experiments and numerical calculations. The qualitative and quantitative predictive performance of two compressible flow solvers in OpenFOAM, namely, rhoCentralFOAM and sonicFOAM, are compared in terms of flow separation pattern and separation pattern transitions within the overexpanded SERN. The ability of rhoCentralFOAM and sonicFOAM to accurately predict complex flow states is evaluated. Results indicate that the numerical simulations conducted using rhoCentralFOAM and sonicFOAM successfully capture flow separation, separated shock waves, separated bubbles and shear layers for two types of restricted shock separation patterns at the same nozzle pressure ratio (NPR), demonstrating agreement with experimental results. However, sonicFOAM initiates the transition in the separation pattern 0.0773 NPR earlier than rhoCentralFOAM during the whole separation pattern transition process of the SERN. The transition process in sonicFOAM lasts longer and exhibits a greater variation in NPR. SonicFOAM fails to accurately predict certain aspects, such as the pressure rise after the separation bubble, the reattachment shock wave, and tends to overestimat the length of the separation shock length. Consequently, sonicFOAM cannot be recommended as a suitable solver for accurately capturing the separation pattern of an overexpanded nozzle.

In this study, pressure drop for oil–water flow in horizontal pipes is represented by using artificial neural network (ANN). Results were compared with Al-Wahaibi correlation and Two-fluid model. This research has used a multilayer feed forward network with Levenberg Marquardt back propagation training for prediction of pressure drop. Original data were divided into two parts where 80% of data was used as training data and remaining 20% of data was used for testing. In this method inputs are oil superficial velocity, water superficial velocity, ratio of density, ratio of viscosity, diameter of pipe and roughness of the pipe wall. The number of neurons is set on four. The feasibility of ANN, Al-Wahaibi correlation and Two-fluid model has been tested against 11 pressure drop data sources. The average absolute percent error of Al-Wahaibi correlation and two-fluid model are 12.73 and 15.84 while this average for the same systems using neural network is only 6.36.so the ANN is in good agreement with experimental data.