519.6 Surrogate model for non-isothermal flow mixing derived from 3D numerical simulation data
doi: 10.18698/2309-3684-2026-2-4465
Application of classical approaches to modeling complex fluid flow processes in technical systems is associated with significant computational costs, which limits their use in optimization problems and repeated parametric studies. Surrogate modeling is one of the effective tools for reducing computational complexity in solving such problems. This paper proposes a methodology for constructing surrogate models of complex hydrodynamic processes based on numerical simulation data. A tee-junction for mixing non-isothermal flows is considered as the object of study. A Radial Basis Function (RBF) neural network was used for approximation. The training dataset was formed based on the results of three-dimensional numerical modeling performed in the domestic software package "Logos," validated against experimental data. An analysis of the influence of RBF model hyperparameters and the size of the initial dataset on prediction accuracy was conducted. A comparison of the surrogate and numerical simulation results showed a high degree of agreement (mean absolute error less than 5% for both velocity and temperature), confirming the effectiveness of the proposed methodology. The obtained results demonstrate the potential of its application for accelerating engineering calculations and the possibility of integration into domestic system-level 1D simulation software packages, which is particularly relevant in the context of developing digital twin technologies and import-independent engineering solutions.
Минин С.В, Королева А.П., Шевченко Ю.А., Шлыкова А.О. Суррогатная модель смешения неизотермических потоков на основе результатов трехмерного численного моделирования. Математическое моделирование и численные методы, 2026, № 2, с. 44–65.