Abstract:A machine-learned (ML) model is developed to enhance the accuracy of turbulence transport equations of Reynolds Averaged Navier Stokes (RANS) solver and applied for periodic hill test case, which involves complex flow regimes, such as attached boundary layer, shear-layer, and separation and reattachment. The accuracy of the model is investigated in extrapolation modes, i.e., the test case has much larger separation bubble and higher turbulence than the training cases. A parametric study is also performed to understand the effect of network hyperparameters on training and model accuracy and to quantify the uncertainty in model accuracy due to the non-deterministic nature of the neural network training. The study revealed that, for any network, less than optimal mini-batch size results in overfitting, and larger than optimal batch size reduces accuracy. Data clustering is found to be an efficient approach to prevent the machine-learned model from over-training on more prevalent flow regimes, and results in a model with similar accuracy using almost one-third of the training dataset. Feature importance analysis reveals that turbulence production is correlated with shear strain in the free-shear region, with shear strain and wall-distance and local velocity-based Reynolds number in the boundary layer regime, and with streamwise velocity gradient in the accelerating flow regime. The flow direction is found to be key in identifying flow separation and reattachment regime. Machine-learned models perform poorly in extrapolation mode, wherein the prediction shows less than 10% correlation with Direct Numerical Simulation (DNS). A priori tests reveal that model predictability improves significantly as the hill dataset is partially added during training in a partial extrapolation model, e.g., with the addition of only 5% of the hill data increases correlation with DNS to 80%.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to improve the accuracy of the Reynolds Averaged Navier - Stokes (RANS) turbulence model in predicting complex flows (such as separated - reattached flows). Specifically, the researchers developed a model based on Machine Learning (ML) to enhance the precision of the turbulence transport equations in the RANS solver, and paid special attention to the performance of the model in the extrapolation mode, that is, the test cases have larger separation bubbles and higher turbulence levels than the training cases. The key points of the paper include: 1. **Problem Background**: - Turbulence modeling and simulation are one of the largest sources of uncertainty in Computational Fluid Dynamics (CFD). - Traditional RANS models have limitations when dealing with complex flows (such as separated - reattached flows), especially when predicting Turbulent Kinetic Energy (TKE). 2. **Research Objectives**: - Develop a Machine Learning model to enhance the RANS model's ability to predict complex flows. - Pay special attention to the performance of the model in the extrapolation mode, that is, the performance of the model when dealing with flow situations that are significantly different from the training data set. 3. **Research Methods**: - Use data from Direct Numerical Simulation (DNS) and Large Eddy Simulation (LES) to train the Machine Learning model. - The training data set includes multiple flow cases with different separation bubble sizes and turbulence intensities. - A periodic hill flow case with a significantly larger separation bubble and higher turbulence level was selected as the test case. 4. **Main Findings**: - The Machine Learning model performs poorly in the extrapolation mode, and the correlation between the prediction results and DNS data is less than 10%. - Partial extrapolation mode (that is, adding a small amount of test data during the training process) significantly improves the model's prediction ability. For example, after adding 5% of the hill data, the correlation with DNS data is increased to 80%. - Feature importance analysis shows that vortex generation is closely related to the shear strain in the free - shear zone, the shear strain in the boundary - layer region, the wall distance and the local velocity - based Reynolds number, and the streamwise velocity gradient in the accelerating flow region. - Data clustering is an effective method that can prevent the Machine Learning model from over - training on more common flow patterns, and can achieve similar accuracy using almost one - third of the training data set. 5. **Conclusions and Recommendations**: - The Machine Learning model can be used as a reliable method to enhance the RANS model's ability to predict TKE generation. - Before applying the Machine Learning model for posteriori testing, a priori testing should be carried out to ensure that the model is not in a completely extrapolated mode and has reasonable accuracy. Through these studies, the paper provides new ideas and methods for improving the RANS model's ability to predict complex flows.

Assessment of Neural Network Augmented Reynolds Averaged Navier Stokes Turbulence Model in Extrapolation Modes

Analysis on Numerical Stability and Convergence of Reynolds Averaged Navier-Stokes Simulations from the Perspective of Coupling Modes

Analysis on Numerical Stability and Convergence of RANS Turbulence Models from the Perspective of Coupling Modes

Machine Learning Based Reynolds Averaged Simulation of Backward-Facing Step Flows at Different Reynolds Numbers

Deep Structured Neural Networks for Turbulence Closure Modelling

Turbulence Modeling Via Data Assimilation and Machine Learning for Separated Flows over Airfoils

A COMBINED NEURAL NETWORK AND MULTIPLE MODIFICATION STRATEGY FOR REYNOLDS-AVERAGED NAVIER-STOKES TURBULENCE MODELING

A Priori Assessment of Prediction Confidence for Data-Driven Turbulence Modeling

A Semi-Implicit Discrepancy Model of Reynolds Stress in a Higher-Order Tensor Basis Framework for Reynolds-averaged Navier–Stokes Simulations

A Divide-and-conquer Machine Learning Approach for Modeling Turbulent Flows

Data-driven RANS closures for improving mean field calculation of separated flows

Characterization of Model-Based Uncertainties in Incompressible Turbulent Flows by Machine Learning

A Divide-and-Conquer Machine Learning Approach for Modelling Turbulent Flows

Artificial neural network approach for turbulence models: A local framework

Data-driven turbulence modeling for fluid flow and heat transfer in peripheral subchannels of a rod bundle

Feature selection and processing of turbulence modeling based on an artificial neural network

Data-driven turbulence modeling in separated flows considering physical mechanism analysis

RECONSTRUCTION OF RANS MODEL AND CROSS-VALIDATION OF FLOW FIELD BASED ON TENSOR BASIS NEURAL NETWORK

Machine Learning-augmented Predictive Modeling of Turbulent Separated Flows over Airfoils

Are random forests better suited than neural networks to augment RANS turbulence models?

Physics-informed Machine Learning Approach for Augmenting Turbulence Models: A Comprehensive Framework