Hostname: page-component-7b7567c8c-7pwfn Total loading time: 0 Render date: 2026-11-16T22:00:27.290Z Has data issue: false hasContentIssue false

Toward generalisable turbulence modelling via radial-basis-function-based corrections using data assimilation with fused multi-fidelity observations

Published online by Cambridge University Press:  30 March 2026

Lifeng Gou
Affiliation:
National Key Laboratory of Science and Technology on Aero-Engine Aero-thermodynamics, Beihang University , Beijing 100191, PR ChinaResearch Institute of Aero-engine, Beihang University, Beijing 102206, PR China
Jian Ye
Affiliation:
School of Energy and Power Engineering, Chongqing University, Chongqing 400044, PR China
Zhengping Zou*
Affiliation:
National Key Laboratory of Science and Technology on Aero-Engine Aero-thermodynamics, Beihang University , Beijing 100191, PR ChinaResearch Institute of Aero-engine, Beihang University, Beijing 102206, PR China
*
Corresponding author: Zhengping Zou, zouzhengping@buaa.edu.cn

Abstract

This paper employs the ensemble-based data assimilation method to develop a closed-form correction term for the Spalart–Allmaras (S–A) turbulence model to enhance predictive accuracy in separated flows through model-form uncertainty reduction. A compact radial-basis-function expression is proposed as correction model to supersede conventional modification procedures in classic field inversion and machine learning frameworks, achieving computational economy through spatially bounded correction regions. The correction model is derived via the Ensemble Kalman method with effective utilisation of synthesised observations based on the multi-fidelity data aggregation. The modified compact expression trained on a single case is systematically evaluated against unseen separation scenarios and the results show that the developed model can improve the prediction accuracy of flow separation in different validation cases, and the effectiveness of the method is verified. Compared with other black-box models, the correction based on the radial-basis-function form offers reduced complexity and high suitability for direct integration into numerical solvers. This approach facilitates cost-effective data assimilation and enables dynamic adaptation of the correction, thereby enhancing the generalisation capability for similar flow separation conditions.

Information

Type
JFM Papers
Copyright
© The Author(s), 2026. Published by Cambridge University Press
Figure 0

Figure 1.Schematic of the FIML-Classic framework with the one-time-correction approach, including the standalone ML model construction and the four stages involving RANS solver usage in both training and testing. The predicted correction is frozen and embedded into the RANS solver for the final augmented computation. Here, $q_i$ denotes the $i$-th input feature.

Figure 1

Figure 2.Schematic of the FIML-Direct framework, implementing the correction through parameter inference of a predefined functional form and eliminating the need for a separate ML model construction and input features collection. The RBF-based correction structure is embedded into the RANS solver prior to data assimilation, then the assimilation process estimates the optimal parameters by matching high- and low-fidelity (Lo-Fi) data. The identified functional correction is then directly embedded into the RANS solver for augmented simulations. Here, $\eta$ denotes the input feature, and $a,\,b,\,\text{and}\,c$ denote the coefficients of the predefined functional form.

Figure 2

Figure 3.Comparison of pressure distribution predictions by the convolutional neural-network-based multi-fidelity model (CNN-MFM) and linear interpolation using different amounts of Hi-Fi data. (a) Dense Hi-Fi samples. (b) Sparse Hi-Fi samples.

Figure 3

Table 1.Comparison of prediction errors $\varepsilon$ at Hi-Fi locations for the linear interpolation and CNN-MFM approaches under different Hi-Fi data densities.

Figure 4

Table 2.Input features selection aimed at separated-flow modelling.

Figure 5

Figure 4.Schematic of a model-consistent turbulence modelling framework with RBF-based corrections and ensemble-based assimilation, consisting of the following three phases: (a) initialisation of the ensemble by randomly sampling the parameters of the RBF; (b) embedding of the correction into the RANS solver, including both structural incorporation of the RBF formulation and numerical embedding through iterative updating of the correction values during the RANS simulation; (c) the RBF parameters are updated based on the observation data, and the updated parameters are re-embedded in phase B, forming a closed-loop FIML process.

Figure 6

Table 3.Overview of the training and testing cases including the Pitch to chord ratio (t/c), Mach numbers (Ma) and Reynolds numbers (Re).

Figure 7

Figure 5.Sketch of the T106A blade geometry and domain boundaries.

Figure 8

Figure 6.Identification of the correction region using GMM-based clustering for T106A. The viscosity ratio and pressure gradient magnitude from the baseline S–A model are used as two-dimensional clustering features. The region with higher viscosity ratio and lower pressure gradient magnitude is where the correction expression is activated.

Figure 9

Table 4.Comparison of different synthetic observations generation methods during the data assimilation process.

Figure 10

Figure 7.Eigenvalue spectrum $(\lambda_i/\lambda_0)^2$ of the observability Gramian versus the mode index $i$, indicating strong separation between the eigenvalues with the RBF-based correction.

Figure 11

Figure 8.Distribution of ensemble members during data assimilation with linearly interpolated synthetic observations. (a) Iteration 1. (b) Iteration 2.

Figure 12

Figure 9.Distribution of ensemble members during data assimilation with CNN-MFM synthetic observations for iteration 1.

Figure 13

Table 5.Convergence assessment of the EnKF data assimilation for T106A.

Figure 14

Figure 10.Convergence process of the EnKF data assimilation for T106A. (a) Iteration 0. (b) Iteration 1.

Figure 15

Figure 11.Distribution of pressure coefficient for T106A.

Figure 16

Figure 12.Distribution of wall shear stress for T106A.

Figure 17

Figure 13.Mach number contours for T106A: (a) original model and (b) RBF-augmented model.

Figure 18

Figure 14.Comparison of model predictions for test cases.:(a) T106C; (b) T106D.

Figure 19

Table 6.Comparison of prediction performance between the baseline RANS model and the RBF-based corrected model across different cases. ($\varDelta$: pressure distribution error, $\circ$: flow separation behaviour, AE: predicted separation location absolute error.).

Figure 20

Figure 15.Distribution of wall shear stress for T106D. Here, MOD–SA and ORI–SA correspond to the Modified S–A and Original S–A models, respectively.

Figure 21

Figure 16.Mach number contours for the T106C: (a) original model and (b) RBF-augmented model.

Figure 22

Figure 17.Contours of the eddy-viscosity difference $\Delta \nu _t=\nu _{t,{aug}}-\nu _{t,{base}}$ showing the local effect of the RBF-based correction on the separated shear-layer regions: (a) T106C; (b) T106D.

Figure 23

Table 7.Comparison of the current framework and FIML-Classic in terms of computational complexity.