This study presents a Machine-Learning (ML) framework built on high-fidelity Computational Fluid Dynamics (CFD) data to obtain a robust numerical solution for three-dimensional (3D) turbulent flow in stenotic arteries. The reference solutions are generated using a finite volume discretization of the Reynolds-Averaged Navier–Stokes (RANS) equations with the Shear–Stress–Transport (SST) k– \(\omega \) turbulence model. We propose an approach to accelerate Finite Volume Method (FVM)-based turbulent flow simulations by employing Artificial Neural Networks (ANNs). In particular, Multilayer Perceptrons (MLPs) are trained to predict the velocity magnitude and pressure fields directly from the CFD dataset that was generated from the 3D turbulent blood flow in stenotic arteries. Our findings demonstrate that the trained ANN serves as a fast and accurate surrogate for traditional numerical solvers, achieving substantial reductions in computational cost while preserving predictive accuracy. Overall, the results highlight a promising pathway for leveraging machine learning to model complex hemodynamic behaviors in both symmetric and non-axisymmetric stenotic arteries, and to advance next-generation CFD-based diagnostic and design tools.
