Data-Driven Turbulence Models in CFD

Introduction

• Reynolds-averaged Navier–Stokes (RANS) equations are a widely preferred approach for industrial CFD applications because of their robustness and computational speed.
• RANS simulations rely on a Reynolds stress tensor (RST) model to predict turbulence physics.
• The accuracy of RANS simulations depends on the performance of the RST models used.
• Data-driven methods are emerging as a way to improve the performance of RANS turbulence models.

Data-Driven Turbulence Models

• Data-driven models use available high-fidelity data of RST values during training, improving predictions obtained from empirical models.
• The majority of studies focus on correcting the RST prediction or modifying existing empirical models.

Progressive Augmentation of Data-Driven Models

• The study focuses on progressively improving the performance of data-driven turbulence models using simulation-driven Bayesian optimisation and Kriging surrogates.
• The goal is to improve the prediction of secondary flow in the 𝑘 − 𝜔 SST model while preserving its performance for canonical cases like channel flow.
• The new models are designed to predict secondary flows that the standard model fails to predict and are tested on channel flow cases to confirm their performance.

Generalisability of Augmented Models

• The generalisability of the new models is assessed by testing them on unseen test cases.
• The results show a significant improvement in the prediction of secondary flows and streamwise velocity, highlighting the potential of this progressive approach to enhance the performance of data-driven turbulence models.
• The study demonstrates the potential of data-driven methods to improve the performance of turbulence models while maintaining the robustness and stability of the solver.

Explore the emerging field of data-driven turbulence models within Reynolds-averaged Navier–Stokes (RANS) equations for computational fluid dynamics (CFD). This quiz dives into the significance of Reynolds stress tensor models and how data-driven approaches enhance their performance. Gain insights into the integration of simulation-driven Bayesian optimization and Kriging surrogates in this cutting-edge research area.