Prediction of shockwave location in supersonic nozzle separation using self-organizing map classification and artificial neural network modeling

One of the novel technologies for natural gas dehydration and natural gas dew-point conditioning is supersonic separation, which has remarkable features, including compact and maintenance-free design. Due to its complex design and the difficulty of experimental analysis, researchers tend to conduct numerical modeling for behavior investigation of the nozzle focusing on shockwave which is the main phenomena inside the nozzle. The present NN-model outperforms a selection of data and proposes an efficient NN-based algorithm for shockwave position estimation as the key nozzle geometry parameter. Data for the shockwave location was collected from a wide range of results from the literature and then a neural network based self-organizing map was adapted to the dataset. This created a classified dataset and the use of unreal weight and repeated experimental results from different research were avoided. A neural network was employed for modeling the shockwave location through the nozzle using a better quality dataset. Additionally, the one-dimensional inviscid theory was utilized in the recursive approach for comparison to the main proposed model. Simulation results presented in this research reveal the effectiveness of the proposed neural network technique for supersonic nozzle modeling and make it possible to determine the shockwave location from the nozzle pressure boundary conditions. The results showed that the supersonic nozzle separation have capability to be used in both low-pressure applications and high pressure ones. The dimensionless length for shockwave location is predicted in the range of 0.82–0.92 for the former and 0.72 to 0.95 for the later, depending on pressure recovery ratio.