Abstract:Turbulent flows play an important role in many scientific and technological design problems. Both Sub-Grid Scale (SGS) models in Large Eddy Simulations (LES) and Reynolds Averaged Navier Stokes (RANS) based modeling will require turbulence models for computational research of turbulent flows in the future. Turbulence model-based simulations suffer from a multitude of causes of forecast uncertainty. For example, the simplifications and assumptions employed to make these turbulence models computationally tractable and economical lead to predictive uncertainty. For safety-critical engineering design applications, we need reliable estimates of this uncertainty. This article focuses on Uncertainty Quantification (UQ) for Computational Fluid Dynamics (CFD) simulations. We review recent advances in the estimate of many types of uncertainty components, including numerical, aleatoric, and epistemic. We go into further depth on the possible use of Machine Learning (ML) methods to quantify these uncertainties. Above all, we elaborate further on significant limitations in these techniques. These range from realizability constraints on the Eigenspace Perturbation Method (EPM) to the requirement for Monte Carlo (MC) approaches for mixed uncertainty. Based on this study, we pinpoint important problems that need to be addressed and offer focused solutions to move beyond these obstacles.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to quantify uncertainty in turbulence modeling. Specifically, the paper focuses on how to evaluate and reduce the prediction uncertainty caused by multiple sources in computational fluid dynamics (CFD) simulations. These uncertainties include numerical uncertainty, aleatory uncertainty, and epistemic uncertainty. The paper particularly emphasizes the model - form uncertainty in turbulence models based on the Reynolds - Averaged Navier - Stokes (RANS) equations and explores the possibility of using machine - learning methods to quantify these uncertainties. ### Main Problems 1. **Prediction Uncertainty**: Simplified assumptions and computational limitations in turbulence models lead to prediction uncertainty. These uncertainties pose challenges to the safety and reliability of engineering design applications. 2. **Model - Form Uncertainty**: Especially in RANS models, due to the limitations of the model structure and the uncertainty in parameter selection, model - form uncertainty is caused. These uncertainties cannot be completely eliminated by increasing the amount of data. 3. **Numerical Uncertainty**: Due to different methods of discretization and solving partial differential equations (PDEs), numerical uncertainty is caused. These uncertainties need to be evaluated by methods such as the Grid Convergence Index (GCI) and Richardson Extrapolation. 4. **Limitations of Machine - Learning Methods**: Although machine - learning methods show potential in quantifying uncertainty, they often have over - fitting problems and lack transparency, which makes their credibility and applicability in engineering applications questioned. ### Solutions 1. **Forward Method**: Evaluate parametric and non - parametric uncertainties by defining probability distributions and propagating them into the governing equations (such as RANS equations). 2. **Backward Method**: Estimate parametric and non - parametric uncertainties using observational data and update prior beliefs. 3. **Eigenspace Perturbation Method (EPM)**: Quantify the sensitivity of predictions to structural uncertainty by perturbing the spectral decomposition components of the expected Reynolds stress tensor. 4. **Methods in the Bayesian Framework**: Use techniques such as the Markov Chain Monte Carlo (MCMC) method and the Ensemble Kalman Filter (EnKF) to combine physical information and observational data and update the posterior distribution of parameters. ### Formula Examples - **Bayes' Theorem**: \[ p(\theta|Z)=\frac{p(Z|\theta)p(\theta)}{p(Z)} \] where \(p(\theta)\) is the prior probability, \(p(Z|\theta)\) is the likelihood function, \(p(\theta|Z)\) is the posterior probability, and \(p(Z)\) is the normalization constant. - **Covariance Kernel of Gaussian Process**: \[ K(x,x') = \sigma(x)\sigma(x')\exp\left(-\frac{|x - x'|^{2}}{L^{2}}\right) \] where \(\sigma(x)\) represents the variance and \(L\) represents the modified length. Through these methods and techniques, the paper aims to provide a comprehensive framework to evaluate and reduce the uncertainty in turbulence modeling, thereby improving the reliability and accuracy of CFD simulations.

Physics Based & Machine Learning Methods For Uncertainty Estimation In Turbulence Modeling

Uncertainty Quantification For Turbulent Flows with Machine Learning

Uncertainty Quantification in Computational Fluid Dynamics: Physics and Machine Learning Based Approaches

Motion Estimation under Location Uncertainty for Turbulent Fluid Flows

Multi-fidelity Deep Learning-based methodology for epistemic uncertainty quantification of turbulence models

Multi-Fidelity Data Assimilation For Physics Inspired Machine Learning In Uncertainty Quantification Of Fluid Turbulence Simulations

A Deep Learning Approach For Epistemic Uncertainty Quantification Of Turbulent Flow Simulations

Physics Constrained Deep Learning For Turbulence Model Uncertainty Quantification

Convolutional Neural Networks For Turbulent Model Uncertainty Quantification

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

Machine learning based uncertainty quantification of turbulence model for airfoils

Uncertainty Quantification of SSG/LRR-ω Turbulence Model Closure Coefficients

Quantifying Model Uncertainty of Neural Network-based Turbulence Closures

Quantifying and Reducing Model-Form Uncertainties in Reynolds-averaged Navier-Stokes Simulations: A Data-Driven, Physics-Informed Bayesian Approach

An Efficient Approach for Quantifying Parameter Uncertainty in the SST Turbulence Model

Extending turbulence model uncertainty quantification using machine learning

Physics-constrained Random Forests for Turbulence Model Uncertainty Estimation

Quantification of Uncertainties in Turbulence Modeling: A Comparison of Physics-Based and Random Matrix Theoretic Approaches

A Random Matrix Approach for Quantifying Model-Form Uncertainties in Turbulence Modeling

An Old-Fashioned Framework for Machine Learning in Turbulence Modeling

Physically constrained eigenspace perturbation for turbulence model uncertainty estimation