Uncertainty analysis of URANS simulations coupled with an anisotropic pressure fluctuation model

Ali Eidi, Richard P. Dwight · 2025

Read Uncertainty analysis of URANS simulations coupled with an anisotropic pressure fluctuation model by Ali Eidi, Richard P. Dwight (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-14564.txt.

Introduction

Ali Eidi and Richard Dwight study spread in averaged turbulence runs coupled to a pressure fluctuation model. Exact fluctuation forecasts matter for flow induced vibration and fatigue. The authors combine global sensitivity analysis, surrogate modeling, and Bayesian inference. Two test cases are used: turbulent channel flow and turbulent annular flow.

In channel flow, calibrated setting distributions improve agreement with source data. In annular flow, settings are harder to identify from data, yet predictions still follow high fidelity trends. The results show both promise and limits of calibration for wall bounded turbulence. The work helps engineers judge when tuned models can be trusted for fluctuation loads.

Catalog record

CREATORAli Eidi, Richard P. Dwight
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE24 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-14564

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Uncertainty analysis of URANS simulations coupled with an anisotropic pressure fluctuation model Ali Eidi, Richard P. Dwight Abstract Accurate prediction of pressure and velocity fluctuations in turbulent flows is essential for understanding flow-induced vibration and structural fatigue. This study investigates the role of turbulence model parameter uncertainty in such predictions using a combination of global sensitivity analysis, surrogate modeling, and Bayesian inference. The methodology is applied to two fluid-only flow cases: turbulent channel flow and turbulent annular flow. In the channel flow case, calibrated parameter distributions lead to improved agreement with reference data. In the annular case, limited parameter identifiability is observed, though predictions remain consistent with high-fidelity trends. The results demonstrate both the potential and limitations of model calibration strategies in wall-bounded turbulent flows. I Introduction Turbulence-induced vibrations (TIV) in nuclear reactors threaten fuel rod integrity. Axial coolant flow generates pressure fluctuations that cause vibrations, leading to wear, fatigue, and structural damage. Accurate prediction of these effects is critical to reactor safety and component longevity. Fluid–structure interaction (FSI) simulations are widely used to analyze TIV. While high-fidelity approaches such as Direct Numerical Simulation (DNS) and Large Eddy Simulation (LES) offer detailed turbulence resolution, they are often too expensive for FSI studies. As a practical alternative, Unsteady Reynolds-Averaged Navier–Stokes (URANS) simulations are commonly employed 2 , though their reliance on turbulence closure models introduces significant uncertainty in pressure fluctuation predictions. To address this, synthetic turbulence models have been proposed to enhance URANS predictions without resolving all turbulence scales 9 . One example is the pressure fluctuation model, designed to reconstruct missing pressure fluctuations 6 . The anisotropic pressure fluctuation model (AniPFM) extends this idea by incorporating anisotropic velocity fluctuations, improving pressure field predictions in TIV studies 12 . However, AniPFM depends on several tunable parameters, and URANS-AniPFM simulations often fail to match pressure statistics accurately within reasonable parameter ranges 11 . Turbulence models based on RANS formulations also introduce uncertainty due to empirical tuning and simplified assumptions 3 . Case-specific adjustments are typically needed to improve predictive performance. Recent studies have demonstrated the potential of Bayesian methods to quantify parameter variability and enhance predictive robustness in RANS modeling 5 ; 4 . Given these challenges, sensitivity analysis (SA) and Bayesian calibration have become essential tools to quantify uncertainty and refine turbulence model parameters using high-fidelity data. This study applies such a framework to URANS-AniPFM simulations using the SST k - \omega model in a fluid-only configuration. A structured methodology is employed to quantify uncertainties and calibrate model parameters, as outlined in Section 2. Section 3 presents preliminary results and key findings, followed by conclusions and an outlook on further studies in Section 4. II Methodology II.1 The k-\omega SST turbulence model The Shear Stress Transport (SST) k-\omega model 7 combines k-\omega for near-wall accuracy with k-\varepsilon for improved free-stream behavior. Its blending function enhances robustness in complex flows, particularly under adverse pressure gradients. The SST model includes multiple empirical parameters that influence its performance ( \sigma_{k} , \beta_{1} , \beta^{*} , a_{1} , and \kappa among others). These parameters are typically determined from canonical experiments and empirical tuning, but remain uncertain when applied to complex flow conditions. A detailed summary of these parameters and their uncertainty limits is provided by 10 . II.2 Anisotropic pressure fluctuation model The anisotropic pressure fluctuation model (AniPFM) enhances URANS predictions by reconstructing missing pressure fluctuations using synthetic anisotropic velocity fields. Starting from a Reynolds decomposition of the velocity and pressure fields, the pressure fluctuation p^{\prime} is obtained by solving the Poisson equation: \displaystyle\frac{\partial^{2}p^{\prime}}{\partial x_{i}\partial x_{i}} \displaystyle=-\rho\bigg[\frac{\partial}{\partial x_{i}}\left(u^{\prime}_{j}\frac{\partial\bar{u}_{i}}{\partial x_{j}}\right) (1) \displaystyle+\frac{\partial^{2}}{\partial x_{i}\partial x_{j}}(u^{\prime}_{i}u^{\prime}_{j}-\overline{u^{\prime}_{i}u^{\prime}_{j}})\bigg], where \bar{u}_{i} is the mean velocity and u^{\prime}_{i} the fluctuating velocity component. The accuracy of p^{\prime} depends on the quality of the synthetic velocity field. AniPFM generates fluctuating velocity components u^{\prime}_{i} using a spatial Fourier decomposition, where the dimensionless fluctuations w_{i}(\mathbf{x},t) are given by: w_{i}(\mathbf{x},t)=\sqrt{6}\sum_{n=1}^{N}\sqrt{q_{n}}\,\sigma_{in}\cos\left(\mathbf{k}_{n}\cdot(\mathbf{x}-\bar{\mathbf{u}}t)+\varphi_{n}\right), (2) with q_{n} the mode amplitude, \mathbf{k}_{n} the wavenumber vector, \sigma_{in} the directional cosine, and \varphi_{n} a random phase. The amplitudes q_{n} are sampled from a modified von Kármán energy spectrum: E(k)=\frac{(k/k_{t})^{4}}{\left[1+2.4(k/k_{t})^{2}\right]^{17/6}}\exp\left(-\frac{12k}{k_{\eta}}\right), (3) where k_{t} is the energy-containing peak wavenumber and k_{\eta} the dissipation-range cutoff. To enforce anisotropy, the dimensionless velocity field is transformed using the Cholesky decomposition of the local Reynolds stress tensor R_{ij} : u^{\prime}_{i}=a_{ij}w_{j},\quad\text{with }a_{ik}a_{kj}=R_{ij}, (4) ensuring that the synthesized fluctuations reproduce the anisotropic structure predicted by URANS. This model improves URANS predictions of unsteady pressure fields in wall-bounded flows, particularly in configurations relevant to flow-induced vibrations. A full formulation is available in 11 . II.3 Simulation cases Two canonical turbulent flow configurations are used to evaluate the URANS-AniPFM framework: turbulent channel flow at multiple friction Reynolds numbers and a more application-relevant turbulent annular flow representing rod-bundle geometries. II.3.1 Channel flow cases The channel flow setup is chosen for its simplicity, statistical homogeneity, and availability of high-fidelity DNS data. Simulations are conducted at four friction Reynolds numbers: Re_{\tau}=180 , 395, 640, and 1020, covering a range of turbulence intensities. Each domain consists of parallel walls with periodic boundaries in the streamwise ( x ) and spanwise ( z ) directions. A constant body force maintains the target Re_{\tau} . Structured meshes with sufficient y -direction resolution ensure grid-independent statistics. Simulations use the SST k - \omega model with second-order time integration and a CFL limit of 0.5. A two-step approach is employed: steady-state RANS using simpleFoam , followed by unsteady URANS with pimpleFoam and AniPFM. Each run covers 10–15 flow-through times. Time-averaged results are compared against DNS data from 1 and others. II.3.2 Annular flow case To test the framework in a more complex geometry, annular flow between concentric cylinders is simulated at Re_{H}=45{,}000 with a bulk velocity of U_{\text{bulk}}=3.45~\text{m/s} . The domain has length L=6D_{h} , where D_{h} is the hydraulic diameter. Axial periodicity is imposed, with no-slip walls at the inner and outer cylinders. The same simulation strategy as in the channel flow is used. The unsteady run spans about 100 flow-through times. Validation is performed against LES data from 8 . II.4 Uncertainty framework To assess the impact of turbulence model parameter uncertainty on URANS-AniPFM simulations, a four-step uncertainty quantification (UQ) framework is employed: primary sensitivity analysis (SA), surrogate modeling, Bayesian calibration, and secondary SA. The first step uses a Sobol-based global SA to rank the influence of SST model parameters on the simulation outputs. This allows focusing the calibration on the most significant parameters, reducing computational cost. A Kriging surrogate is then built to approximate the relationship between selected parameters and quantities of interest (QoIs). To account for structural discrepancies, a Gaussian process-based model inadequacy term is added: \hat{y}(\boldsymbol{\theta})=\mathcal{M}_{\text{Kriging}}(\boldsymbol{\theta})+\delta(\boldsymbol{\theta}), (5) where \hat{y}(\boldsymbol{\theta}) is the corrected prediction, \mathcal{M}_{\text{Kriging}} is the Kriging output, and \delta(\boldsymbol{\theta}) is a zero-mean Gaussian process (Kennedy and O’Hagan, 2001). This hybrid surrogate supports robust calibration and uncertainty propagation. Bayesian inference is used to calibrate model parameters with high-fidelity data. Markov Chain Monte Carlo (MCMC) sampling yields posterior parameter distributions and predictive uncertainty bounds for QoIs. A secondary SA is then performed using the surrogate model to assess the influence of calibrated parameters and verify model robustness. III Results and Discussion The uncertainty analysis considers nine independent parameters from the SST k - \omega model: \alpha_{k1} , \alpha_{k2}

Read the complete text