Integrated CFD and machine learning investigation of pressure and temperature distributions on a cone at supersonic and hypersonic Mach numbers

Résumé fourni par la source

A comprehensive Computational Fluid Dynamics (CFD) and Unified Machine Learning (ML) framework is developed to study high-speed flow regimes over a cone. The study evaluates the distributions of total pressure and total temperature along the slant length of the cone for freestream Mach numbers between 4.2 and 6 and semi-cone angles between 2° and 20°. Flow properties are extracted at non-dimensional positions (x/L) across the range 0 to 1 in 0.1 increments. The CFD results show that the cone tip experiences a significant total pressure loss due to intense shock interactions, leading to downstream pressure stabilization. The total temperature shows a gradual change pattern, resulting from both viscous dissipation and boundary-layer growth, which becomes more pronounced with increasing Mach number and larger cone angles. The CFD results are validated against the Taylor-Maccoll theory. The machine learning models use Support Vector Machines (SVMs), K-Nearest Neighbors (KNN), and Ordinal Logistic Regression (OLR) to train on CFD data using k-fold cross-validation method. Later, the models' generalizability was tested on the test dataset. The analysis shows that SVM achieves the best total pressure prediction accuracy, while OLR outperforms other methods in total temperature prediction accuracy. KNN demonstrates moderate accuracy, but it struggles to apply its knowledge to new situations. The combined CFD–ML solution enables researchers to achieve their results with less computational work while delivering accurate predictions. It would serve as a powerful resource for high-speed aerodynamic research and design optimization.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.