Data-Driven RANS Closures Using a Relative Importance Term Analysis Based Classifier for 2D and 3D Separated Flows

Abstract

This study presents a novel approach for enhancing Reynolds-averaged Navier-Stokes (RANS) turbulence modeling through the application of a Relative Importance Term Analysis (RITA) methodology to develop a new zonally-augmented $k-\omega$ SST model. Traditional Linear Eddy Viscosity Models often struggle with separated flows. Our approach introduces a physics-based binary classifier that systematically identifies separated shear layers requiring correction by analyzing the relative magnitudes of terms in the turbulence kinetic energy equation. Using symbolic regression, we develop compact correction terms for Reynolds stress anisotropy and turbulent kinetic energy production. Trained on two-dimensional configurations, our model demonstrates significant improvements in predicting separation dynamics while maintaining baseline performance and fully attached flows. Generalization tests on Ahmed body and Faith Hill three-dimensional configurations confirm robust transferability, establishing an effective methodology for targeted enhancement of RANS predictions in separated flows.

References in corpus (10)

• Turbulence Modeling in the Age of Data
• A Physics Informed Machine Learning Approach for Reconstructing Reynolds Stress Modeling Discrepancies Based on DNS Data
• Perspectives on Machine Learning-augmented Reynolds-averaged and Large Eddy Simulation Models of Turbulence
• Discovery of Algebraic Reynolds-Stress Models Using Sparse Symbolic Regression
• Searching for turbulence models by artificial neural network
• RANS Equations with Explicit Data-Driven Reynolds Stress Closure Can Be Ill-Conditioned
• A neural network approach for the blind deconvolution of turbulent flows
• Learning dominant physical processes with data-driven balance models
• Customized data-driven RANS closures for bi-fidelity LES-RANS optimization
• Comparison of neural closure models for discretised PDEs