Constrained re-calibration of two-equation Reynolds-averaged Navier–Stokes models

Theoretical and Applied Mechanics Letters, Год журнала: 2024, Номер 14(2), С. 100503 - 100503

Опубликована: Фев. 7, 2024

Machine-learned augmentations to turbulence models can be advantageous for flows within the training dataset but often cause harm outside. This lack of generalizability arises because constants (as well as functions) in a Reynolds-averaged Navier–Stokes (RANS) model are coupled, and un-constrained re-calibration these (and disrupt calibrations baseline model, preservation which is critical model's generalizability. To safeguard behaviors beyond dataset, machine learning must constrained such that basic like law wall kept intact. letter aims identify constraints two-equation RANS so future work performed without violating constraints. We demonstrate identified not limiting. Furthermore, they help preserve model.

Язык: Английский

---

Fast flow prediction of airfoil dynamic stall based on Fourier neural operator

Physics of Fluids, Год журнала: 2023, Номер 35(11)

Опубликована: Ноя. 1, 2023

Dynamic stall on airfoil is of great importance in engineering applications. In the present work, Fourier neural operator (FNO) applied to predict flow fields during dynamic process NACA0012 airfoil. Two cases with different angles attack are simulated by Reynolds averaged numerical simulation Spalart–Allmaras (SA) model at Re=4×104. A prediction directly constructed between several previous time nodes and that future node FNO. The sequence based iterative strategy achieved for stall. results show FNO can achieve a fast accurate streamwise velocity, normal pressure, vorticity both cases. dynamics vortices around analyzed demonstrate accuracy addition, FNOs configurations tested lower error shorter training time-consuming.

Язык: Английский

---

Constrained Recalibration of Reynolds-Averaged Navier–Stokes Models

AIAA Journal, Год журнала: 2023, Номер 62(4), С. 1434 - 1446

Опубликована: Дек. 20, 2023

The constants and functions in Reynolds-averaged Navier–Stokes (RANS) turbulence models are coupled. Consequently, modifications of a RANS model often negatively impact its basic calibrations, which is why machine-learned augmentations detrimental outside the training dataset. A solution to this identify degrees freedom that do not affect calibrations only modify these identified when recalibrating baseline accommodate specific application. This approach colloquially known as “rubber-band” approach, we formally call “constrained recalibration” paper. To illustrate efficacy Spalart–Allmaras log law calibration. By subsequently interfacing data-based methods with freedom, train solve historically challenging flow scenarios, including round-jet/plane-jet anomaly, airfoil stall, secondary separation, recovery after separation. In addition good performance inside dataset, trained yield similar

Язык: Английский

---

Data-Driven Adverse Pressure Gradient Correction for Turbulence Model

AIAA Journal, Год журнала: 2025, Номер unknown, С. 1 - 17

Опубликована: Фев. 21, 2025

The Spalart–Allmaras (SA) model is widely used in engineering turbulence simulations. It has been calibrated using the logarithmic law and provides sufficient accuracy zero pressure gradient (ZPG) turbulent boundary layer (TBL) but shows poor performance with adverse (APG), especially separated flows. In this paper, distribution of important variables functions SA studied. found that, APG TBL, original exhibits significant errors encounters a multiple-value problem [Formula: see text] function destruction term. A new feature proposed based on eddy viscosity to characterize history effects overcoming problem. algebraic expression function, as shown Eq. ( 21 ), established by combining neural networks symbolic regression. results show that good generalization for nine different flows outside training set, not only maintaining behaviors ZPG, also enhancing

Язык: Английский

---

Data-Enabled Parameter Modification for the Shear-Stress-Transport Model with Progressive Neural Networks

AIAA Journal, Год журнала: 2025, Номер unknown, С. 1 - 12

Опубликована: Апрель 7, 2025

Turbulence models based on the Reynolds-averaged Navier–Stokes method are widely employed for simulation in engineering and research. Nonetheless, these have some limitations simulating flows with adverse pressure gradients due to inclusion of various assumptions, such as eddy viscosity hypothesis. As a result, many researchers focused their efforts improving turbulence models, including parameter calibration. In this paper, progressive neural network framework is utilized modify dissipation coefficients shear-stress transport (SST) model into function physical quantities. Firstly, zero gradient flat plate case obtain calibrated SST-FP. Subsequently, flow turbulent separation bubble adopted progressively acquire SST-TSB. The enhanced performs well predicting mean velocity profile, friction coefficient, other variables training cases, owing joint corrections [Formula: see text] equations. Furthermore, verification research demonstrates that SST-TSB mitigates potential damage wall-law prediction. It can also adapt forecast properly circumstances where original predicts larger or smaller zones, avoiding drawbacks Bayesian inference methods when applied

Язык: Английский

---

Gray area mitigation in grid-adaptive simulation for wall-bounded turbulent flows

International Journal of Mechanical Sciences, Год журнала: 2025, Номер unknown, С. 110305 - 110305

Опубликована: Апрель 1, 2025

Язык: Английский

---

Enhancing generalizability of machine-learning turbulence models

AIAA SCITECH 2022 Forum, Год журнала: 2024, Номер unknown

Опубликована: Янв. 4, 2024

This work aims to incorporate basic calibrations of Reynolds-averaged Navier-Stokes (RANS) models as part machine learning (ML) frameworks. The ML frameworks considered are tensor-basis neural network (TBNN), physics-informed (PIML), and field inversion & (FIML) in J. Fluid Mech., 2016, 807, 155-166, Phys. Rev. Fluids, 2017, 2(3), 034603 Comp. Phys., 305, 758-774, the baseline RANS one-equation Spalart-Allmaras model, two-equation ��-�� seven-equation Reynolds stress transport models. trained against plane channel flow shear-layer data. We compare study whether machine-learned augmentations detrimental outside training set. findings summarized follows. due TBNN detrimental. PIML leads that beneficial inside dataset but it. These results not affected by model. FIML’s two eddy viscosity models, where an inner-layer treatment already exists, largely neutral. Its augmentation does exist, improves mean prediction a channel. Furthermore, these FIML mostly non-detrimental dataset. In addition reporting results, paper offers physical explanations results. Last, we note conclusions drawn here confined flows this study. More detailed comparative studies validation verification needed account for developments recent years.

Язык: Английский

---

A priori screening of data-enabled turbulence models

Physical Review Fluids, Год журнала: 2023, Номер 8(12)

Опубликована: Дек. 21, 2023

$A$ $p\phantom{\rule{0}{0ex}}o\phantom{\rule{0}{0ex}}s\phantom{\rule{0}{0ex}}t\phantom{\rule{0}{0ex}}e\phantom{\rule{0}{0ex}}r\phantom{\rule{0}{0ex}}i\phantom{\rule{0}{0ex}}o\phantom{\rule{0}{0ex}}r\phantom{\rule{0}{0ex}}i$ validation and verification of black box machine learned turbulence models is time consuming not always fruitful. We discuss a theoretical framework that allows $a$ $p\phantom{\rule{0}{0ex}}r\phantom{\rule{0}{0ex}}i\phantom{\rule{0}{0ex}}o\phantom{\rule{0}{0ex}}r\phantom{\rule{0}{0ex}}i$ screening machine-learned are based on feed-forward neural networks. It requires no knowledge the weights bias only activation function. The method tells one whether model preserves basic calibrations like law wall.

Язык: Английский

---

Robust experimental data assimilation for the Spalart-Allmaras turbulence model

Physical Review Fluids, Год журнала: 2024, Номер 9(8)

Опубликована: Авг. 22, 2024

The presented methodology fuses computational models with experimental data to enhance the Spalart-Allmaras (SA) turbulence model for Reynolds-averaged Navier-Stokes equations. By leveraging Ensemble Kalman filtering approach (EnKF), this study refines SA model's coefficients, ensuring improved performance on separated flows without any accuracy trade-off already well captured by SA. Validated different flow conditions, including a backward-facing step and NASA wall-mounted hump, recalibrated demonstrates significant improvements in key metrics.

Язык: Английский

---

Adapting Reynolds-averaged Navier Stokes Models while Preserving the Basic Calibrations

AIAA SCITECH 2022 Forum, Год журнала: 2024, Номер unknown

Опубликована: Янв. 4, 2024

The constants and functions in Reynolds-averaged Navier Stokes (RANS) turbulence models are coupled. Consequently, modifications of a RANS model often negatively impact its basic calibrations, which is why machine-learned augmentations detrimental outside the training dataset. A solution to this identify degrees freedom that do not affect calibrations only modify these identified when {\color{black}re-calibrating} baseline accommodate specific application. This approach colloquially known as ``rubber-band'' approach, we formally call ``constrained re-calibration'' article. To illustrate efficacy Spalart-Allmaras (SA) log law calibration. By subsequently interfacing data-based methods with freedom, train solve historically challenging flow scenarios, including round-jet/plane-jet anomaly, airfoil stall, secondary separation, recovery after separation. In addition good performance inside dataset, trained yield similar

Язык: Английский

---