betaFeedback

WorksPhysics and Astronomy

Non-Unique Machine Learning Mapping in Data-Driven Reynolds Averaged Turbulence Models

Abstract

Recent growing interest in using machine learning for turbulence modelling has led to many proposed data-driven turbulence models in the literature. However, most of these models have not been developed with overcoming non-unique mapping (NUM) in mind, which is a significant source of training and prediction error. Only NUM caused by one-dimensional channel flow data has been well studied in the literature, despite most data-driven models having been trained on two-dimensional flow data. The present work aims to be the first detailed investigation on NUM caused by two-dimensional flows. A method for quantifying NUM is proposed and demonstrated on data from a flow over periodic hills, and an impinging jet. The former is a wall-bounded separated flow, and the latter is a shear flow containing stagnation and recirculation. This work confirms that data from two-dimensional flows can cause NUM in data-driven turbulence models with the commonly used invariant inputs. This finding was verified with both cases, which contain different flow phenomena, hence showing that NUM is not limited to specific flow physics. Furthermore, the proposed method revealed that regions containing low strain and rotation or near pure shear cause the majority of NUM in both cases: approximately 76% and 89% in the flow over periodic hills and impinging jet, respectively. These results led to viscosity ratio being selected as a supplementary input variable (SIV), demonstrating that SIVs can reduce NUM caused by data from two-dimensional flows and subsequently improve the accuracy of tensor-basis machine learning models for turbulence modelling.

Nomenclature

VariableMeaningUnit
bijb_{ij}Anisotropy tensor-
BBImpinging jet inlet widthm
CμC_{\mu}Boussinesq hypothesis parameter-
gng_nScalar coefficients of the general effective-viscosity hypothesis-
ddDistance between two arbitrary scatter points pp and qq-
HhH_hHill heightm
kkTurbulent kinetic energym²/s²
LxL_xDomain length in periodic hills casem
nCIn_{CI}Number of conflicting instances in a grid cell-
rνr_{\nu}Viscosity ratio (rν=νt/(100ν+νt)r_{\nu} = \nu_t/(100\nu+ \nu_t))-
R\mathbf{R}Non-dimensional mean rotation rate (R=krij/ε\mathbf{R} = k r_{ij}/\varepsilon)-
rijr_{ij}Mean rotation rate1/s
RetRe_tTurbulent Reynolds number (Ret=k2/νεRe_t = k^2/\nu\varepsilon)-
S\mathbf{S}Non-dimensional mean strain rate (S=ksij/ε\mathbf{S} = k s_{ij}/\varepsilon)-
sijs_{ij}Mean strain rate1/s
u+u^+Nondimensional streamwise velocity (u+=uˉ1/uτu^+ = \bar{u}_1/u_{\tau})-
uτu_{\tau}Friction velocity (τw/ρ\sqrt{\tau_w/\rho})m/s
ui′u_i'Velocity fluctuation in iith directionm/s
uˉi\bar{u}_iMean velocity in iith directionm/s
UbU_bBulk inlet velocity in periodic hills casem/s
UinU_{in}Uniform inlet velocity in impinging jet casem/s
xix_iDistance in iith directionm
xtr\mathbf{x}_{tr}Input variables of the two-dimensional GEVH (xtr≡{tr(S2),tr(R2)}\mathbf{x}_{tr} \equiv\{tr(\mathbf{S}^2), tr(\mathbf{R}^2)\})-
y+y^+Nondimensional distance from the wall (y+=x2uτ/νy^+ = x_2u_{\tau}/\nu)-
Symbol
τw\tau_wWall shear stressm²/s²
τij\tau_{ij}Reynolds stressm²/s²
δij\delta_{ij}Kronecker delta-
ε\varepsilonTurbulent kinetic energy dissipation ratem²/s³
α\alphaNon-dimensional shear velocity gradient (α=(k/ε)(∂uˉ1/∂x2)\alpha= (k/\varepsilon)(\partial\bar{u}_1/\partial x_2))-
ν\nuKinematic molecular viscositym²/s
νt\nu_tKinematic eddy viscositym²/s
ρ\rhoDensitykg/m³
β\betaSteepness factor in periodic hills case-
Subscript
1Streamwise direction
2Wall-normal direction
3Spanwise direction
ppArbitrary scatter point pp
qqArbitrary scatter point qq

Table 4.

Superscript
Reynolds-averaged
Min-max normalized quantity
Abbreviation
CFDComputational fluid dynamics
CLAComplete linkage agglomerative
CIConflicting instance
DNSDirect numerical simulation
GEVHGeneral effective-viscosity hypothesis
LESLarge eddy simulation
LSRLow strain and rotation
MLMachine learning
NNMNearest neighbours method
NUMNon-unique mapping
PBMProximity box method
RANSReynolds-averaged Navier-Stokes
SIVSupplementary input variable
SRCCSpearman’s rank correlation coefficient
SSTShear stress transport
TBMLTensor-basis machine learning
TBNNTensor-basis neural network
TKETurbulent kinetic energy

Table 1.

Introduction

Over the last few decades, advancement in computational power and algorithms has greatly improved the speed and capability of computational fluid dynamics (CFD) solvers for simulating turbulent flows. The continual drive for better accuracy and wider applicability has led to the development of a range of approaches including Reynolds-averaged Navier-Stokes (RANS) models, and scale-resolving methods such as large eddy simulation (LES) and direct numerical simulation (DNS) [50]. However, scale-resolving methods are still currently computationally infeasible for most industrial CFD practitioners [12]. Using RANS turbulence models is still therefore the most common approach for simulating turbulent flows, and it is projected that this will remain for the near future [9, 5]. Despite their popularity however, progress in improving RANS model accuracy has stalled, due to the challenge of conserving their high robustness and low computational cost in the development of new models [62]. As a result, RANS approaches still give inaccurate predictions for many complex flows, including flow separation, and impingement [9].

Improvements in computational resources and recent advances in machine learning (ML) methodologies have also led to rapid developments in ML, resulting in its prevalence in many technological applications today including natural language processing, and computer vision [16, 17]. More recently, there has been growing interest in using data-driven ML models as an alternative to existing RANS turbulence models or to augment them due to the stagnation in RANS development. Yarlanki et al. (2012) explored the capabilities of neural networks for calibrating RANS model coefficients. Similarly, Parish and Duraisamy [41] demonstrated that ML can be used to tune the magnitude of terms in the RANS equations. Although improved predictions given by both models were reported, their approaches still adopted a linear relationship between mean strain rate and Reynolds stress, which neglects rotation and higher order strain rate terms [49]. Complex flow features that RANS models fail to predict accurately such as high streamline curvature, secondary vortices and anisotropic turbulence are subsequently not fully accounted for or at all in these approaches [28]. Therefore, Reynolds stress (or its projections) should be selected as the ML model outputs.

Modelling Reynolds stress using a tensor basis has emerged as a popular data-driven approach [10]. These models – referred to as tensor-basis machine learning (TBML) models in this paper – are based on predicting the scalar coefficients of a Galilean-invariant tensor integrity basis, which gives a closed-form expression for Reynolds stress in homogeneous turbulent flows [44]. Ling et al. [29] developed the first TBML model called the tensor-basis neural network (TBNN), which reportedly predicted Reynolds stress anisotropy and mean flow fields more accurately than a linear and nonlinear RANS model when tested on canonical flow cases. As the tensor basis sensitises the Reynolds stress to mean strain rate, mean rotation rate and their higher order products, these effects are fully accounted for in the TBNN. Various enhancements to the TBNN have thereafter been proposed in the literature. Zhang et al. [66] introduced regularisation in the loss function and included y+y^{+} as an input variable. Parmar et al. [42] expanded the tensor basis to include the effect of mean pressure gradient. Jiang et al. [23] proposed an additional ML model for predicting the tensor basis and a different timescale for non-dimensionalising mean strain and rotation rate. Some studies have reported improved prediction accuracy after adopting more advanced ML techniques, including ensemble learning by Man et al. [32] and modular TBNNs by Man et al. [33]. The latter involves using multiple TBNNs – each trained and tested on specific regions of flow physics. Modifying the TBNN has also been investigated, e.g., with skipped connections by Jiang et al. [23] and with Bayesian inference by Tang et al. [54]. Different ML model types have also been explored for predicting the scalars, including symbolic regression [57, 58], random forests [25], and gradient-boosted tree models [37].

While this literature review shows that many TBML models have been proposed, most have not been developed with overcoming non-unique mapping (NUM) in mind (also known as the multi-value problem). This is a well-known source of prediction error that can occur when a deterministic ML model is trained on data containing one-to-many relations – i.e., where multiple observations in the training dataset have the same input values but different output values [3]. This ill-posed problem of conflicting observations can cause difficulty in training because their same input values lead to the loss function being minimised against different target output values, thus posing a challenge in achieving model convergence [30]. Although some investigations in the literature have explored NUM in TBML models [31, 23, 6], only NUM resulting from one-dimensional (1D) channel flow data has been well studied, and their proposed solutions, which involved modifying or including supplementary input variables were developed to eliminate NUM due to near-wall flow data exclusively. Given that most TBML models in the literature to date have been trained on data from two-dimensional (2D) flow cases containing various flow physics instead, it is crucial to investigate whether NUM in TBML models can be caused by 2D flow data. If so, an awareness in the community to develop steps that reduce or eliminate NUM resulting from data of such flows becomes necessary to improve the training convergence and subsequent predictive accuracy of these models.

Although extensive literature on modelling one-to-many relations exists in mathematics and other engineering disciplines, limited studies were found on methods for identifying them. Intuitively, one-to-many relations may be identified by creating hypersurfaces of the outputs in the input space, and detecting if the hypersurfaces overlap in the input space [47]. The data may alternatively be clustered into groups that correspond to these hypersurfaces and assessed whether the clusters share the same input space, which was attempted by Huang et al. (2013). However, both methods do not quantify the extent of one-to-many relations in the data, which would be helpful in analysing the worst affected regions in a chosen flow case. Various ML models have been proposed in the literature that can represent one-to-many relations between inputs and outputs which would address NUM, including recurrent neural networks [55], mixture density networks [3], and mixture of experts [20]. However, as these approaches would introduce significant complexity to existing TBML models, their exploration was reserved for future work. It was considered reasonable instead to attempt extending the use of a supplementary input variable (SIV) seen in 1D channel flow studies to 2D flows for this investigation.

Overall, the current work aims to investigate non-unique mapping (NUM) in data-driven turbulence models based on the tensor-basis approach for two-dimensional (2D) flows. Although true anisotropy has normally been used as the output to identify NUM resulting from 1D channel flow data, we show that this is unsuitable for 2D flows, as this can lead to false diagnoses of NUM in the tensor basis. The true scalar coefficients that tensor-basis machine learning (TBML) models aim to predict are proposed as the output for identifying NUM instead. A clustering approach for identifying and quantifying NUM throughout the input space before TBML model training is proposed. Based on unsupervised learning, this approach does not require user action, and can be used with any number of inputs. The approach was demonstrated using data from two different flow scenarios: a wall-bounded separated flow over periodic hills case, and a free shear flow in an impinging jet case. It is shown that NUM can exist in TBML models trained on these 2D flows with the standard invariant inputs. Moreover, this finding is verified with data from two very different flow cases, showing that NUM is not limited to certain flow physics in the training data. The method was extended by including a supplementary input variable (SIV) and repeated with the two cases, which demonstrated that SIVs can reduce NUM from 2D flow cases in TBML models and – with verification from a training experiment, – subsequently improve TBML model accuracy.

Background

General Effective Viscosity Hypothesis

The general effective-viscosity hypothesis (GEVH) postulated by Pope [44] states that the Reynolds stress τij\tau_{ij} under local equilibrium in Reynolds-averaged Navier Stokes (RANS) modelling of a turbulent flow can be fully determined by a finite sum of basis tensors (Zhou et al. [68]). For two-dimensional flows where the velocity and variation of mean quantities are negligible or zero in one co-ordinate direction, the GEVH contains four terms (Pope [44]):

τij=ui′uj′=2k(bij+13δij)(1)\tau_{ij} = u_i^{\prime}u_j^{\prime} = 2k\left(b_{ij} + \frac{1}{3}\delta_{ij}\right) \tag*{(1)}

where, kk is the turbulent kinetic energy (TKE) and δij\delta_{ij} is Kronecker delta. The anisotropy tensor bijb_{ij} in Eq. (1) is a normalised form of the Reynolds stress and contains three of the terms:

bij=g1S+g2(SR−RS)+g3(S2−13tr(S2)δij)(2)b_{ij} = g_1\mathbf{S} + g_2(\mathbf{S}\mathbf{R} - \mathbf{R}\mathbf{S}) + g_3\left(\mathbf{S}^2 - \frac{1}{3}tr(\mathbf{S}^2)\delta_{ij}\right) \tag*{(2)}

S(=ksij/ε)\mathbf{S} (= k s_{ij}/\varepsilon) and R(=krij/ε)\mathbf{R} (= k r_{ij}/\varepsilon) are the mean strain and mean rotation rate, respectively, non-dimensionalised by TKE and TKE dissipation rate ε\varepsilon. The dimensional mean strain rate sijs_{ij} and mean rotation rate rijr_{ij} can be calculated as:

sij=12(∂uˉi∂xj+∂uˉj∂xi),rij=12(∂uˉi∂xj−∂uˉj∂xi)(3)s_{ij} = \frac{1}{2}\left(\frac{\partial\bar{u}_i}{\partial x_j} + \frac{\partial\bar{u}_j}{\partial x_i}\right), \qquad r_{ij} = \frac{1}{2}\left(\frac{\partial\bar{u}_i}{\partial x_j} - \frac{\partial\bar{u}_j}{\partial x_i}\right) \tag*{(3)}

Lowercase ss and rr are used to denote the dimensionalised form of the mean strain and rotation rate tensors, while uppercase S\mathbf{S} and R\mathbf{R} are used to denote the non-dimensionalised form. The tensor products SR\mathbf{S}\mathbf{R}, RS\mathbf{R}\mathbf{S} and S2\mathbf{S}^2 in Eq. (2) can be rewritten in Einstein notation, e.g., SR=k2sikrkj/ε2\mathbf{S}\mathbf{R} = k^2s_{ik}r_{kj}/\varepsilon^2. The notation tr(…)tr(\ldots) denotes the trace of the product inside the parentheses. These traces are also known as invariants and can be represented in Einstein notation, e.g., tr(S2)=k2sikski/ε2tr(\mathbf{S}^2) = k^2s_{ik}s_{ki}/\varepsilon^2. Coefficients g1g_1, g2g_2 and g3g_3 are scalars that are unknown functions of the following invariants of S\mathbf{S} and R\mathbf{R}:

gn=f(tr(S2),tr(R2)),(n=1,2,3)(4)g_n = f\left(tr(\mathbf{S}^2), tr(\mathbf{R}^2)\right), \qquad(n = 1, 2, 3) \tag*{(4)}

It has been shown in the literature that Eq. (2) can be represented by a TBML model (Ling et al. [29]). Given invariants such as those in Eq. (4) as inputs, these models are trained to predict the gng_n coefficients that when combined with S\mathbf{S} and R\mathbf{R} from a RANS method on the right-hand side of Eq. (2), the summation gives a bijb_{ij} result equivalent to the true bijb_{ij} from experiments or an accurate high-fidelity simulation. These gng_n targets will be referred to as true gng_n.

The first TBML model was the TBNN introduced by Ling et al. [29]. This was demonstrated with the three-dimensional version of Eq. (2), which contains five input invariants of S\mathbf{S} and R\mathbf{R}, and ten output gng_n coefficients (n = 1 to 10) – corresponding to ten basis tensors. Ling et al. [29] showed that the TBNN was able to learn the relationship between these inputs from RANS and true bijb_{ij} from LES and DNS for some classical flow cases during training [27]. This capability has now also been demonstrated with other TBML models, e.g., the tensor-basis random forest by Kaandorp and Dwight (2020). Although most TBML models were developed to be applicable to three-dimensional flows, they can also be used to model bijb_{ij} in one- and two-dimensional flows. In these situations, the velocity and its mean gradients would be zero in one or more co-ordinate directions, resulting in some negligible and redundant terms. More specifically for two-dimensional flows, the TBML models effectively become a representation of Eq. (2) with inputs given in Eq. (4), as the other seven tensors become zero or can be subsumed into the three tensors in Eq. (2), and the other three inputs become zero or can be represented by the inputs in Eq. (4). In particular, the quadratic term g4(R2−tr⁡(R2)δij/3)g_4(\mathbf{R}^2 - \operatorname{tr}(\mathbf{R}^2)\delta_{ij}/3) that exists in the full GEVH can be subsumed into the g3g_3 term in Eq. (2). For 1D flows where only shear velocity gradient ∂uˉ1/∂x2\partial\bar{u}_1/\partial x_2 is the only non-negligible velocity gradient component, the GEVH and subsequent TBML model simplifies to modelling bijb_{ij} as a function of ∂uˉ1/∂x2\partial\bar{u}_1/\partial x_2. The simplicity of the one-dimensional GEVH has made it an ideal starting point for demonstrating NUM in TBML models [31].

Non-Unique Mapping

Non-unique mapping (NUM) can be defined as a poor fitting of a function on a dataset, due to an occurrence or occurrences of multiple solutions satisfying a particular array of input variables in some observations of the dataset [3]. Suppose a ML model is tasked with learning a mapping from a set of input variables x≡{x1,…,xn}\mathbf{x} \equiv\{x_1, \ldots, x_n\} to a set of output variables y≡{y1,…,ym}\mathbf{y} \equiv\{y_1, \ldots, y_m\}. In practice, this is undertaken by training the ML model on a finite number of examples, i.e., a dataset {(x(i),y(i))}i=1N\{(\mathbf{x}^{(i)}, \mathbf{y}^{(i)})\}_{i=1}^{N} where i=1,…,Ni = 1, \ldots, N and NN is the total number of observations in the dataset. Let observation 1 have input variables x(1)\mathbf{x}^{(1)} and output variables y(1)\mathbf{y}^{(1)}, and observation 2 have input variables x(2)\mathbf{x}^{(2)} and output variables y(2)\mathbf{y}^{(2)}. Suppose the following one-to-many relation exists: x(1)=x(2)≡x(a)\mathbf{x}^{(1)} = \mathbf{x}^{(2)} \equiv\mathbf{x}^{(a)} and y(1)≠y(2)\mathbf{y}^{(1)} \ne\mathbf{y}^{(2)} as shown in Figure 1. The ML model would consequently have to learn a non-unique relation between x\mathbf{x} and y\mathbf{y} for x=x(a)\mathbf{x} = \mathbf{x}^{(a)}. This causes difficulty in training because the loss function would use y(1)\mathbf{y}^{(1)} and y(2)\mathbf{y}^{(2)} alternatingly as the target value to calculate error in the predicted value of y\mathbf{y} for x=x(a)\mathbf{x} = \mathbf{x}^{(a)}. Therefore, the ML model would never reach a converged state – resulting in a poor fitting, and it would always give erroneous predictions for y\mathbf{y} at x=x(a)\mathbf{x} = \mathbf{x}^{(a)} [4, 30].

Diagram showing the mapping from inputs domain $x$ to outputs codomain $y$

Figure 1. Diagram showing the mapping from inputs domain xx to outputs codomain yy. A non-unique one-to-many relation is shown between xx and yy at x=x(1)(=x(2))x = x^{(1)} (= x^{(2)}), while a unique one-to-one relation is shown between xx and yy at x=x(3)x = x^{(3)} and x=x(N)x = x^{(N)}.

Non-Unique Mapping from 1D Channel Flow Data

It has been reported in the literature that NUM can occur in TBML models when trained on cases where shear velocity gradient ∂u‾1/∂x2\partial\overline{u}_1/\partial x_2 is the only non-negligible velocity gradient, such as in fully-developed channel flow [31]. In this situation, the expressions for the bijb_{ij} components given by the GEVH reduce to:

b11=f1b22=f2b33=−(f1+f2)b12=f3(5)\begin{aligned} b_{11} &= f_1 \\ b_{22} &= f_2 \\ b_{33} &= -(f_1 + f_2) \\ b_{12} &= f_3 \tag*{(5)} \end{aligned}

where f1f_1, f2f_2, and f3f_3 are univariate functions of nondimensional shear velocity gradient, α(=(k/ε)(∂u‾1/∂x2))\alpha(= (k/\varepsilon)(\partial\overline{u}_1/\partial x_2)), due to the gng_n coefficients and basis tensors becoming univariate functions of α\alpha. Liu et al. [31] showed that for any value of α\alpha present in DNS data for 1D channel flow, there are multiple possible values of b12b_{12}. This also occurs in the normal bijb_{ij} components as shown by Cai et al. [6]. Given that α\alpha in channel flow simulated using RANS and scale-resolved methods have a similar profile, a TBML model trained on channel flows would experience NUM. This is because the model by virtue of Eq. (5) would be tasked with mapping α\alpha from RANS to multiple true values from a scale-resolved method for each bijb_{ij} component [59]. To illustrate this, Figure 2(a) shows a plot of α\alpha given by a two-equation RANS model against b12b_{12} given by well-resolved LES for a channel flow at Reτ=945Re_\tau= 945 (based on friction velocity and channel half-height). It is clear that a NUM exists between α\alpha and b12b_{12} in Figure 2(a). The annotations show that distance from the wall has a many-to-one relation with b12b_{12}. Therefore, some studies have proposed sensitising b12b_{12} to a supplementary input variable (SIV) that monotonically changes with wall distance to address this issue. One such variable is the turbulent Reynolds number Ret(=k2/νε)Re_t (= k^2/\nu\varepsilon), which was included as an SIV by Liu et al. [31] and Jiang et al. (2021). By plotting Figure 2(a) with RetRe_t on a third axis as shown in Figure 2(b), it is seen that a unique mapping exists between input variables {α,Ret}\{\alpha, Re_t\} and output variable b12b_{12}.

Figure 2. Fully-developed turbulent channel flow at Reτ=945Re_\tau= 945 (based on friction velocity and channel height) with α\alpha and RetRe_t given by a two-equation RANS approach and b12b_{12} given by well-resolved LES: (a) a non-unique mapping between α\alpha and b12b_{12} is shown, and (b) a unique mapping between input variables {α,Ret}\{\alpha, Re_t\} and output variable b12b_{12} is shown.

2D General Effective Viscosity Hypothesis Equations

While NUM due to 1D channel flow data in tensor-basis machine learning (TBML) models has been well studied, it has not yet been properly investigated with 2D flows, even though data from such cases has been mostly used for training TBML models. In a related study, Sotgiu et al. [51] briefly discussed about NUM caused by data from a 2D corrugated channel with dimensionless wall-normal Reynolds stress v2/kv^2/k as one of two inputs. However, as v2/kv^2/k is seldom used in TBML models, their discussion is not generally applicable. Therefore, the present work aims to give a detailed investigation on NUM with the standard invariant inputs, which are used in most TBML models proposed in the literature. A natural starting point is to examine the expressions and variables in the 2D GEVH. The component-wise form of the bijb_{ij} expression given by the GEVH in Eq. (2) can be simplified as:

b11=g1S11−2g2S12R12+13g3(S112+S122),b22=−g1S11+2g2S12R12+13g3(S112+S122),b33=−23g3(S112+S122),b12=g1S12+2g2S11R12(6)\begin{aligned} b_{11} &= g_1S_{11} - 2g_2S_{12}R_{12} + \frac{1}{3}g_3\left(S_{11}^{2}+S_{12}^{2}\right),\\ b_{22} &= -g_1S_{11} + 2g_2S_{12}R_{12} + \frac{1}{3}g_3\left(S_{11}^{2}+S_{12}^{2}\right),\\ b_{33} &= -\frac{2}{3}g_3\left(S_{11}^{2}+S_{12}^{2}\right),\\ b_{12} &= g_1S_{12} + 2g_2S_{11}R_{12} \tag*{(6)} \end{aligned}

and the components of nondimensional mean strain and rotation rate are calculated as:

S11=−S22=kε∂u‾1∂x1,S12=S21=k2ε(∂u‾1∂x2+∂u‾2∂x1),S_{11}=-S_{22}=\frac{k}{\varepsilon}\frac{\partial\overline{u}_1}{\partial x_1},\qquad S_{12}=S_{21}=\frac{k}{2\varepsilon}\left(\frac{\partial\overline{u}_1}{\partial x_2}+\frac{\partial\overline{u}_2}{\partial x_1}\right),
R12=−R21=k2ε(∂u‾1∂x2−∂u‾2∂x1)R_{12}=-R_{21}=\frac{k}{2\varepsilon}\left(\frac{\partial\overline{u}_1}{\partial x_2}-\frac{\partial\overline{u}_2}{\partial x_1}\right)

g1g_1, g2g_2 and g3g_3 are functions of tr⁡(S2)\operatorname{tr}(S^2) and tr⁡(R2)\operatorname{tr}(R^2) as stated in Eq. (4):

tr⁡(S2)=2(S112+S122)tr⁡(R2)=−2R122(7)\operatorname{tr}(\mathbf{S}^{2}) = 2(S_{11}^{2} + S_{12}^{2}) \\ \operatorname{tr}(\mathbf{R}^{2}) = -2R_{12}^{2} \tag*{(7)}

The combination of tr⁡(S2)\operatorname{tr}(\mathbf{S}^{2}) and tr⁡(R2)\operatorname{tr}(\mathbf{R}^{2}) as a set of input variables will be denoted as xtr\mathbf{x}_{tr}, such that xtr≡{tr⁡(S2),tr⁡(R2)}\mathbf{x}_{tr} \equiv\{\operatorname{tr}(\mathbf{S}^{2}), \operatorname{tr}(\mathbf{R}^{2})\}. As the 2D version of the GEVH is examined in the rest of this paper, gng_n will collectively represent g1g_1, g2g_2, and g3g_3 hereafter, such that n=(1,2,3)n = (1, 2, 3) unless stated otherwise. Given that the full GEVH with ten basis tensors reduces to this version when it is evaluated on 2D flows, the results and conclusions of this study are also applicable to TBML models based on the full GEVH. Eq. (6) will only be studied in the context of TBML models, so the bijb_{ij} components on the left-hand side will be considered as those of true bijb_{ij}, with S\mathbf{S} and R\mathbf{R} components from RANS on the right-hand side.

Methodology

Inputs and Outputs for Non-Unique Mapping Analysis

Appropriate input and output variables must be chosen to investigate whether NUM can exist in the 2D GEVH and subsequently, in TBML models trained on 2D flow data. For NUM analysis of 1D channel flow data, α\alpha and true bijb_{ij} have usually been chosen as the input and output variables, respectively, because the true bijb_{ij} components become univariate functions of α\alpha as shown in Eq. (5) [31, 6]. In the GEVH, the inputs are xtr\mathbf{x}_{tr} so these should be chosen as the NUM analysis inputs. For the outputs, the bijb_{ij} components are found to be functions of S11S_{11}, S12S_{12}, R12R_{12}, g1g_1, g2g_2, and g3g_3 as shown in Eq. (6). While it seems reasonable to choose true bijb_{ij} again as the outputs for NUM analysis, doing so ignores the specific operations on gng_n in the tensor-basis. This can lead to false diagnoses of NUM in a TBML model.

Consider the following scenario where two arbitrary locations denoted L1 and L2 in a 2D flow case have the same R12R_{12} values but different S11S_{11} and S12S_{12} values that satisfy the condition: S11,L12+S12,L12=S11,L22+S12,L22S_{11,\mathrm{L1}}^{2} + S_{12,\mathrm{L1}}^{2} = S_{11,\mathrm{L2}}^{2} + S_{12,\mathrm{L2}}^{2}, where S11,L1≠S11,L2S_{11,\mathrm{L1}} \ne S_{11,\mathrm{L2}}, S12,L1≠S12,L2S_{12,\mathrm{L1}} \ne S_{12,\mathrm{L2}}, and subscripts L1 and L2 denote the location each term belongs to. This leads to both locations having the same xtr\mathbf{x}_{tr} input values. If L1 and L2 also have the same true bijb_{ij} values, then a one-to-one relation would exist between xtr\mathbf{x}_{tr} and true bijb_{ij}. However, L1 and L2 require different true gng_n values to be predicted to give the same true bijb_{ij} values, as L1 and L2 would apply different S11S_{11} and S12S_{12} values to Eq. (6). Hence, a one-to-many relation would exist between xtr\mathbf{x}_{tr} and true gng_n. This relation would cause NUM in TBML models because while the final output in TBML models is bijb_{ij}, their learnable mapping actually exists between xtr\mathbf{x}_{tr} and gng_n. Therefore, this NUM is undetectable if true bijb_{ij} is chosen as the NUM analysis output and would give a false negative result. Oppositely, it is possible that a one-to-one relation between xtr\mathbf{x}_{tr} and true gng_n may exist if L1 and L2 have different true bijb_{ij} values instead, thus returning a false positive result.

NUM in TBML models is guaranteed to be absent if one-to-many relation between xtr\mathbf{x}_{tr} and true gng_n do not exist – allowing both gng_n and bijb_{ij} to be predicted accurately. The task becomes finding the true gng_n values. McConkey et al. [37] showed that true gng_n can be approximated from a least-squares solution between both sides of Eq. (6). Mandler and Weigand [34] obtained true gng_n by performing field inversion involving successive tensor projections. However, both methods do not express true gng_n explicitly with closed-form equations, which would allow the causes of one-to-many relation between xtr\mathbf{x}_{tr} and true gng_n to be more easily understood and addressed. Jongen and Gatski [24] showed that true gng_n can be explicitly determined with the following expressions which are always valid for 2D flows:

g1=(S112(S112+S122))(b11−b22)+(S12S112+S122)b12(8)g_1 = \left(\frac{S_{11}}{2(S_{11}^{2}+S_{12}^{2})}\right)(b_{11}-b_{22})+\left(\frac{S_{12}}{S_{11}^{2}+S_{12}^{2}}\right)b_{12} \tag*{(8)}
g2=2S11b12+S12(b22−b11)4R12(S112+S122)(9)g_2 = \frac{2S_{11}b_{12}+S_{12}(b_{22}-b_{11})}{4R_{12}(S_{11}^{2}+S_{12}^{2})} \tag*{(9)}
g3=3(b11+b22)2(S112+S122)(10)g_3 = \frac{3(b_{11}+b_{22})}{2(S_{11}^{2}+S_{12}^{2})} \tag*{(10)}

Although they derived Eq. (3.1) by finding the Gram matrix for calculating true gng_n, these expressions can be obtained by treating Eq. (6) as a set of simultaneous equations and solving for g1g_1, g2g_2, and g3g_3.

As a side note: while these true coefficients can be used as the targets in TBML models, the calculation of them can also be used to tune well-established RANS models. The two-equation RANS models such as the k−εk-\varepsilon model are known to have accuracy limitations due to CμC_\mu being set as a constant in calculating Reynolds stress. Coefficient g1g_1 is directly proportional to CμC_\mu and therefore calculating true g1g_1 may allow target values of CμC_\mu to be calculated for tuning the models. Furthermore, calculating true g2g_2 and g3g_3 may allow developers working on these classical models to add accurate terms to Boussinesq hypothesis to increase the accuracy of their model.

True gng_n as the Non-Unique Mapping Analysis Output

Eq. (3.1a-c) gives the true gng_n values for any 2D flow case, and a TBML model achieves the upper performance limit if it can predict these from inputs xtr\mathbf{x}_{tr}. Hence, true gng_n should be the gng_n targets for a TBML model to predict and therefore chosen as the NUM analysis output. There is a caveat however if the a posteriori process is to be performed, in which test predictions from the TBML model are injected into the RANS equations to give improved mean flow field results. Wu et al. (2019) proposed that the linear term should be treated implicitly and give a non-negative eddy viscosity to ensure numerical stability, i.e., νt≥0\nu_t \ge0, which can only be achieved if g1≤0g_1 \le0 (Durbin & Pettersson-Reif [11]; McConkey et al. [37]). The TBML model should be trained to predict g1≤0g_1 \le0 to enforce this constraint, which requires g1g_1 targets ≤0\le0 in the training data. As true gng_n calculated with Eq. (3.1) are not constrained to obey this inequality, they should not be the training targets in this situation where true g1>0g_1 > 0. Therefore, at these locations, different gng_n targets would need to be considered that (i) obey g1g_1 targets ≤0\le0, (ii) return true bijb_{ij} when substituted into the GEVH in Eq. (6), and (iii) ideally are given by closed-form expressions so that any causes of NUM can be traced back to the flow physics.

Nonetheless, true gng_n should always be the gng_n targets where true g1≤0g_1 \le0. Therefore, if true g1≤0g_1 \le0 occurs in most of the domain of a training case, choosing true gng_n as the NUM analysis output would be applicable to most of the data, and can still give many insights on the mapping between the inputs xtrx_{tr} and true gng_n. With data from such flow cases, choosing true gng_n as the NUM analysis output would allow the main aim of this study to be achieved, which was to identify whether NUM occurs due to 2D flow data. With this reason in mind, true gng_n was chosen as the NUM analysis output, and data from flow cases that mostly contained values of true g1≤0g_1 \le0 which are introduced in the next section were used.

Flow Cases

The NUM analysis was performed on data from two different flow cases: (i) a flow over periodic hills characterised by the interaction of the fluid with repetitive hill-like structures, influencing boundary layer dynamics, flow separation, and turbulence generation, and (ii) a plane impinging jet involving various flow patterns such as stagnation, boundary layer effects, and recirculation zones. These cases were chosen to demonstrate the developed methodology on different 2D flow physics. Velocity contours of these cases with annotations are shown in Figure 3a and (b).

The 2D flow over periodic hills case is a wall-bounded flow which separates at the hilltop, creating a large recirculation zone with flow reattachment downstream of the hill. The Reynolds number is 5600 based on a bulk inlet velocity UbU_b of 0.028 m/s and a hill crest height HhH_h of 1 m. It has a steepness factor, β\beta of 1.5 which gives a domain length LxL_x of 10.9Hh10.9H_h as shown in Figure 3a. Further information regarding the geometry and boundary conditions can be found in Xiao et al. [63]. Data was extracted from McConkey et al. [36], which includes inputs xtr\mathbf{x}_{tr} modelled with RANS shear stress transport (SST), and true bijb_{ij} from a DNS performed by Xiao et al. [63]. This case was chosen for NUM analysis because periodic hill cases are commonly used to train TBML models in the literature ([25, 37, 54]).

To assess NUM on a very different canonical flow case without flow separation, data from an impingement flow was also used. As xtr\mathbf{x}_{tr} and true bijb_{ij} data of an impingement flow was not readily available, a RANS and scale-resolved simulation of one was run in this study. Jaramillo et al. [21] and Shukla and Dewan [48] simulated a particular plane impinging jet case using RANS and scale-resolved methods. This case begins with uniform velocity flow entering through a slot in the top wall of the domain, which forms a core jet as it falls towards the bottom. This leads to an impingement zone due to stagnation at the bottom wall. The flow then moves towards the sides to form wall jets, counter-rotating recirculation zones, and leaves the domain at the two side outlets. The Reynolds number is 2×1042 \times10^{4} (based on bulk inlet velocity UinU_{in} and slot width BB), and the spacing between the inlet and impingement wall is 4B4B as shown in Figure 3b. In this work, the case was modelled in OpenFOAM version 2006 with RANS SST and well-resolved LES to obtain inputs xtr\mathbf{x}_{tr}, and spanwise- and time-averaged true bijb_{ij}, respectively ([22, 38, 26]). Validation of the results was performed by comparing the recirculation zone dimensions with the two studies, and good agreement was obtained. For both the periodic hill and impinging jet cases, true bijb_{ij} was interpolated onto their respective RANS grids. The SS and RR components from RANS and interpolated true bijb_{ij} were then used to calculate true gng_n at each RANS cell centre with Eq. (3.1).

Clustering Process

Quantifying how NUM varies in the input space allows the worst affected regions to be determined, and more effective methods to reduce or eliminate NUM can be developed by targeting these regions. The challenge that 2D flow data poses is that xtr\mathbf{x}_{tr} vs. true gng_n (n∈{1,2,3}n \in\{1, 2, 3\}) exists in three dimensions. One suggestion would be to extend the method used for visualising NUM in 1D channel flow data as shown in Figure 2 by creating a surface plot with xtr\mathbf{x}_{tr} on the xx and yy axis and true gng_n (n∈{1,2,3}n \in\{1, 2, 3\}) on the zz-axis. Overlapping surfaces in regions of the xtr\mathbf{x}_{tr} space would then indicate a one-to-many relation between xtr\mathbf{x}_{tr} and the plotted true gng_n coefficient ([47]). Although the amount of input space that contains overlapping surfaces can be easily quantified when inputs xtr\mathbf{x}_{tr} are used, this becomes challenging with three or more inputs, and comparison becomes unclear, e.g., comparing overlapping surfaces to overlapping volumes. Secondly, many overlapping surfaces can lead to cluttered data visualisations and difficulties in clearly displaying regions of the input space that would be worst affected by NUM.

The proposed NUM quantification method is instead based on inputs vs. output data in the form of scatter plots with inputs xtr\mathbf{x}_{tr} plotted on the xx and yy axes, and outputs true gng_n (n∈{1,2,3}n \in\{1,2,3\}) given in colour. For each scatter plot, the method begins by grouping points with similar true gng_n (n∈{1,2,3}n \in\{1,2,3\}) values into a pre-specified number of clusters. Each data point can then be compared with other points that have very similar xtr\mathbf{x}_{tr} values to assess whether any of them belong to a different cluster. If so, this indicates that the two points have very similar xtr\mathbf{x}_{tr} but very different true gng_n (n∈{1,2,3}n \in\{1,2,3\}) values, signifying that NUM would exist at those xtr\mathbf{x}_{tr} values. This approach is based on finding regions of the input space where groups of points with similar true gng_n (n∈{1,2,3}n \in\{1,2,3\}) values overlap (Huang et al. 2013). Complete linkage agglomerative clustering in the Scikit-Learn library was used to group the points [43]. This algorithm begins by considering each point as a cluster of its own, then clusters with closest similarity are combined into larger ones until a pre-specified number of clusters remain [15]. More information regarding the clustering process can be found in Appendix § 14.

To find the ideal number of clusters, a cluster validity study was performed whereby the clustering process was run with the pre-specified number of clusters set to 2, 3, 4, …, 10. The ideal number was chosen as (i) greater than two, as it was found that two clusters still contained a large range of true gng_n (n∈{1,2,3}n \in\{1,2,3\}) values, and (ii) the value that gave the highest average silhouette score over all true gng_n coefficients [1]. The average silhouette score is a commonly used metric of cluster quality that represents how separated the clusters are from each other and how compact they are. This is achieved by considering both the intra- and inter-cluster distances between points. Readers are referred to Rousseeuw [46] for further information. Figure 4 shows that the ideal number of clusters was 3 for both cases.

Figure 4. Line plots showing number of clusters vs. silhouette score for true gng_n (n∈{1,2,3}n \in\{1,2,3\}) and their average for the (a) periodic hill case, and (b) impinging jet case.

Non-Unique Mapping Quantification Process

For any chosen point in a cluster denoted hereafter as the “focal point”, two approaches named the nearest neighbours method (NNM), and proximity box method (PBM) were undertaken to identify points that have very similar xtrx_{tr} values to it, which are hereafter referred to as “close points”. These approaches are illustrated in Figure 5(a) and (b), respectively. In NNM, the nearest nn points are taken as the close points, where nn is a pre-specified number [40]. In this study, nn was set to the number of clusters, which ensured each cluster had a chance to allocate at least one of their points as a nearest neighbour. The xtrx_{tr} distance between any two points pp and qq was determined with Euclidean distance, dd:

Figure 5. Figure 5 Close points identification methods using (a) nearest neighbours, and (b) a proximity box. The focal point has a black border and close points are highlighted in yellow.

d=(tr⁡(S2)p−tr⁡(S2)q)2+(tr⁡(R2)p−tr⁡(R2)q)2(11)d = \sqrt{\left(\operatorname{tr}(\mathbf{S}^{2})_{p} - \operatorname{tr}(\mathbf{S}^{2})_{q}\right)^{2} + \left(\operatorname{tr}(\mathbf{R}^{2})_{p} - \operatorname{tr}(\mathbf{R}^{2})_{q}\right)^{2}} \tag*{(11)}

where the subscripts indicate which point the trace is calculated from. In PBM, a box with length ll and height hh is specified [8]. The box is then centrally superimposed on the focal point, and the other points inside the box are taken as the close points. NNM and PBM are similarity query methods, which are well-established in data similarity searching [65].

If the focal point and one of its close points belong to different clusters, then a one-to-many relation is considered to exist between their xtrx_{tr} values and the true gng_n coefficient being investigated. This will be referred to as a “conflicting instance” (CI), which is proposed as a metric to quantify NUM in this work. The distribution of CIs in the input space was investigated by partitioning the xtrx_{tr} space into a grid, sequentially setting each scatter point in each grid cell as the focal point and calculating their number of CIs, then summing them to give the total number of CIs in each grid cell, which is denoted as nCIn_{CI}. This process is outlined in Algorithm 1 and was separately run for true gng_n, i.e., ∀n∈{1,2,3}\forall n \in\{1, 2, 3\}.

Algorithm 1: Find distribution of CIs as a metric of NUM in the input space

  1. Partition the input space into a grid

  2. for each grid cell c=1,2,…,Ncc = 1, 2, \ldots, N_c do

  3. Initialize new CI counter nCI=0n_{CI} = 0 for grid cell cc

  4. for each point in grid cell pc=1,2,…,Npcp_c = 1, 2, \ldots, N_p^c do

  5. Run NNM or PBM to find its close points pc′=1,2,…,Npc′p_c' = 1, 2, \ldots, N_p^{c'}

  6. for each close point pc′=1,2,…,Npc′p_c' = 1, 2, \ldots, N_p^{c'} do

  7. if focal point pcp_c and close point pc′p_c' belong to different clusters then

  8. nCI←nCI+1n_{CI} \leftarrow n_{CI} + 1

  9. end if

  10. end for

  11. end for

  12. end for

This method can be extended to include more than two input variables by performing the clustering with more dimensions and increasing the dimensions of Eq. (11) if using NNM, or applying a higher dimensional proximity space (e.g., cuboid for three inputs) if using PBM. nCIn_{CI} can thereby be easily calculated and compared to assess how NUM changes when using a greater number of, or different input variables.

As an introductory demonstration, the method was applied to 1D channel flow containing input variable α\alpha and the results are detailed in Appendix § 15. While the present approach may be extended to 3D flows, it is not possible for the true output coefficients to be calculated analytically as was undertaken in this study with Eq. (3.1). Instead, they must be calculated numerically – e.g., by least squares or successive tensor projections [37, 34]. Furthermore, a minimum of five invariant inputs are required. This method is also not limited to TBML models; it can also be applied to other data-driven turbulence models where the inputs and target outputs of the training cases are well-defined. With the two typical TBML model inputs in xtr\boldsymbol{x}_{tr}, this method enables NUM caused by 2D flow data to be clearly visualised and quantified as demonstrated in the following results.

Periodic Hills Case Results

Non-Unique Mapping Quantification

The NUM quantification results for the periodic hills case are shown in figure 6. It is observed in figure 6(a)-(c) that the data points are grouped according to their true gng_n (n∈{1,2,3}n \in\{1, 2, 3\}) values, and the groups overlap in some parts of the xtr\boldsymbol{x}_{tr} space. This can be easily visualised in the low strain and rotation (LSR) subset of xtr\boldsymbol{x}_{tr}, which is shown in the zoomed-in views. For example in figure 6(a), true g1g_1 at tr(S2)≈2tr(\mathbf{S}^2) \approx2 and tr(R2)≈−1tr(\mathbf{R}^2) \approx-1 is approximately both greater than 0.02 and less than −0.1-0.1. This is also observed with true g2g_2 and true g3g_3 for the same input values, thereby qualitatively showing that NUM can exist between xtr\boldsymbol{x}_{tr} and true gng_n in 2D flow data. Another part of the xtr\boldsymbol{x}_{tr} space where there are overlapping groups (and thus NUM can exist) is near and along the tr(S2)=−tr(R2)tr(\mathbf{S}^2) = -tr(\mathbf{R}^2) diagonal. These points were taken from locations that have approximately pure or homogeneous shear flow ([53, 39]).

A comparison between figure 6(a)-(c) and figure 6(d)-(f) shows that the scatter points have been reasonably clustered according to their true gng_n values. These clusters enabled the nCIn_{CI} heat maps calculated with NNM shown in figure 6(g)-(i) to be made, which support the scatter plot observations. In particular, the heat maps show that there is high nCIn_{CI} located in the LSR subset where tr(S2)<2tr(\mathbf{S}^2) < 2 and tr(R2)>−2tr(\mathbf{R}^2) > -2 and near the pure shear line, amounting to 77%, 76% and 74% of total nCIn_{CI} in the true g1g_1, g2g_2, and g3g_3 heat maps, respectively, found in the heat map grid cells that the pure shear line passes through. While it is difficult to identify NUM in the rest of the xtr\mathbf{x}_{tr} space using the scatter plots in figure 6(a)-(c), the heat maps show that CIs exist there but in significantly lower quantities. Figure 17(a)-(Figure 17c) in Appendix § 16 shows that the distribution trends of nCIn_{CI} given by Algorithm 1 with PBM are very similar to those of figure 6(g)-(i), and therefore the results of both approaches support each other.

Heat maps of $n_{CI}$ for true $g_1$, $g_2$, and $g_3$ for the periodic hills and impinging jet cases

Figure 17. Heat maps of nCIn_{CI} between inputs xtr\mathbf{x}_{tr} and outputs true gng_n given by PBM. The heat maps for the periodic hill case are shown in subfigures (a), (b), and (c), while the heat maps for the impinging jet case are shown in subfigures (d), (e), and (f).

Panels showing true $g_1$, true $g_2$, and true $g_3$ values, and true $g_1$, true $g_2$, and true $g_3$ clusters

Figure 6.

Figure 6 (PDF p. 18). NUM results for the flow over periodic hills data. Subplots (a)-(c) show scatter plots of inputs xtr\mathbf{x}_{tr} vs. outputs true gng_n, (d)-(f) show clusters represented by colours: red, blue, and green, and (g)-(i) show heat maps of nCIn_{CI} given by NNM (n=3n = 3), with the blue line indicating pure shear flow. The red boxes in subplots (a), (d), and (g) mark the low strain and rotation subset, and the grey boxes mark the near pure shear subset described in the following two sections.

Low Strain and Rotation Regions

To relate the findings from figure 6 to the physical domain, contours of the NUM analysis inputs and outputs are shown in figure 7(a)-(b) and figure 7(c)-(e), respectively. The true gng_n contour distributions in figure 7(c)-(e) are found to compare well with optimal gng_n computed using field inversion in Mandler and Weigand [35], thereby validating the approach of using Eq. (3.1) to calculate true gng_n.

The domain regions in figure 7 with red hatching represent where scatter points located in the LSR subset of tr(S2)<2tr(\mathbf{S}^2) < 2 and tr(R2)>−2tr(\mathbf{R}^2) > -2 in figure 6 were taken from. Two regions that satisfy these constraints can be identified: the larger region in the freestream and the smaller region by the lower wall – both spanning the whole length of the domain. The constraints can be rewritten in terms of velocity gradients by substituting Eq. (7) into them:

(∂uˉ1∂x1)2+14(∂uˉ1∂x2+∂uˉ2∂x1)2<ε2k214(∂uˉ1∂x2−∂uˉ2∂x1)2>ε2k2(12)\begin{aligned} \left(\frac{\partial\bar{u}_1}{\partial x_1}\right)^2+\frac{1}{4}\left(\frac{\partial\bar{u}_1}{\partial x_2}+\frac{\partial\bar{u}_2}{\partial x_1}\right)^2 &< \frac{\varepsilon^2}{k^2} \\ \frac{1}{4}\left(\frac{\partial\bar{u}_1}{\partial x_2}-\frac{\partial\bar{u}_2}{\partial x_1}\right)^2 &> \frac{\varepsilon^2}{k^2} \tag*{(12)} \end{aligned}

The presence of ∂uˉ1/∂x2\partial\bar{u}_1/\partial x_2 in both expressions of Eq. (12) and its large range compared to other velocity gradients suggests that it is important in determining whether a data point belongs to red hatching or not in this case. It is found that the red hatched regions in figure 7 contain inflection points of ∂uˉ1/∂x2\partial\bar{u}_1/\partial x_2 (Xiao et al. [63]). Therefore, the LSR flow is found to have only been extracted from regions with very low values of ∂uˉ1/∂x2\partial\bar{u}_1/\partial x_2 – approximately one order of magnitude lower than the rest of the domain.

Figure 7(c) shows discontinuities exist in true g1g_1 as marked by label (1). These are caused by terms in the true g1g_1 expression in Eq. (3.1a) tending towards infinity or negative infinity as S11S_{11} and S12S_{12} approach zero. Consequently, low S112+S122S_{11}^{2} + S_{12}^{2} values (on the order of 10−310^{-3} to 10−110^{-1} in this case) can cause discontinuities, such that true g1g_1 may change from high magnitude negative values to high positive values across them. Discontinuities in true g2g_2 and g3g_3 can also be observed as marked by label (2) in figure 7(d) and figure 7(e), respectively. These are also caused by terms tending towards infinity and negative infinity as S11S_{11} and S12S_{12} approach zero. Proof of these limits are given in Appendix § 17.

Negligible change in tr(S2)tr(\mathbf{S}^{2}) and tr(R2)tr(\mathbf{R}^{2}) can be observed in figure 7(a)-(b) where the discontinuities occur. This shows that one-to-many relations exist between xtrx_{tr} and true gng_n at these xtrx_{tr} values. The clustering was able to capture the large differences in true gng_n across these discontinuities, resulting in high nCIn_{CI} in the LSR subset at tr(S2)<2tr(\mathbf{S}^{2}) < 2 and tr(R2)>−2tr(\mathbf{R}^{2}) > -2 as shown in figure 6(g)-(i).

However, it is found that NUM from the discontinuities has a negligible effect on the accuracy of bijb_{ij} given by TBML models. For example, consider a TBML model that is able to perfectly predict the true gng_n fields shown in figure 7(c)-(e). The model would then typically multiply its gng_n predictions (which are true gng_n in this scenario) with S\mathbf{S} and R\mathbf{R} tensor products in the GEVH to give bijb_{ij} predictions as shown in Eq. (6). It is found that any discontinuities in gng_n vanish upon substitution in Eq. (6), and negligible differences in bijb_{ij} are obtained whether gng_n have high negative or positive magnitude predictions in regions containing low S11S_{11} and S12S_{12}. Further explanations are given in Appendix § 18. This indicates that as long as the loss function in the TBML model is calculated using bijb_{ij} error, NUM caused by true gng_n discontinuities in these regions can be ignored.

Near Pure Shear Regions

The white hatched regions in figure 7 represent where scatter points in figure 6 that were not included in the red hatching and satisfy the following constraints were taken from:

−1<tr(S2)+tr(R2)<1tr(S2)<10tr(R2)>−10(13)\begin{aligned} -1 &< tr(\mathbf{S}^{2}) + tr(\mathbf{R}^{2}) < 1 \\ tr(\mathbf{S}^{2}) &< 10 \\ tr(\mathbf{R}^{2}) &> -10 \tag*{(13)} \end{aligned}

The first constraint captures points that are in close proximity to the pure shear line, while the second and third constraint excludes points in a region of the xtrx_{tr} space that has nil or very low nCIn_{CI}, as shown in figure 6(g)-(i). Substituting Eq. (7) into the first constraint shows that these points satisfy:

−ε22k2<(∂uˉ1∂x1)2+∂uˉ1∂x2∂uˉ2∂x1<ε22k2(14)-\frac{\varepsilon^{2}}{2k^{2}} < \left(\frac{\partial\bar{u}_{1}}{\partial x_{1}}\right)^{2} + \frac{\partial\bar{u}_{1}}{\partial x_{2}} \frac{\partial\bar{u}_{2}}{\partial x_{1}} < \frac{\varepsilon^{2}}{2k^{2}} \tag*{(14)}

and in the same way, the criterion for pure shear, tr(S2)=−tr(R2)tr(\mathbf{S}^{2}) = -tr(\mathbf{R}^{2}) can be found as:

(∂u‾1∂x1)2=−∂u‾1∂x2∂u‾2∂x1(15)\left(\frac{\partial\overline{u}_1}{\partial x_1}\right)^2=-\frac{\partial\overline{u}_1}{\partial x_2}\frac{\partial\overline{u}_2}{\partial x_1} \tag*{(15)}

There are two main regions of white hatching in figure 7: (i) the region at the centre of the domain, which includes freestream, recirculation, and reattachment flow, and (ii) the near-wall region by the upper wall. This shows that different flow physics can satisfy Eq. (4.3) to be considered flow that is almost pure shear. These different flow physics are found to cause NUM between xtr\boldsymbol{x}_{tr} and true gng_n in various ways. For example, some regions have almost constant values of tr(S2)tr(\boldsymbol{S}^2) and tr(R2)tr(\boldsymbol{R}^2) but contain a large range of true gng_n, such as near the separation point labelled (3) in figure 7. A special case of this occurs where there is pure shear flow, such that tr(S2)=−tr(R2)tr(\boldsymbol{S}^2)=-tr(\boldsymbol{R}^2). As one input is equal to the negative of the other, the number of inputs effectively reduces to one at these locations. Figure 7 shows that some pure shear flow regions have almost constant values of tr(S2)tr(\boldsymbol{S}^2) (=−tr(R2)=-tr(\boldsymbol{R}^2)) but contain a large range of true gng_n values, such as the freestream region labelled (4). This shows that the standard two inputs can be insufficient in mapping to true gng_n accurately. NUM can also be found where the inputs and outputs vary in different directions. For example, some regions contain tr(S2)tr(\boldsymbol{S}^2) and tr(R2)tr(\boldsymbol{R}^2) varying in the x2x_2 direction, while true gng_n varies in the x1x_1 direction, such as those labelled (5). These causes of NUM justify why ηCI\eta_{CI} along the pure shear diagonal is higher compared to the rest of the xtr\boldsymbol{x}_{tr} space in figure 6(g)-(i). As the majority of the white hatching contains true g1≤0g_1 \le0, its gng_n targets would mostly be unaffected by the a posteriori caveat, so the NUM there must be addressed.

Figure 7.

Figure 7 (PDF p. 21). Contour plots of (a)-(b) xtr\mathbf{x}_{tr} inputs, (c)-(e) true gng_n, and (f) viscosity ratio rνr_\nu for the periodic hills case. Red hatching shows regions of low strain and rotation flow where tr⁡(S2)<2\operatorname{tr}(\mathbf{S}^2)<2 and tr⁡(R2)>−2\operatorname{tr}(\mathbf{R}^2)>-2. White hatching shows regions of near pure shear flow that were not included in the red hatching and satisfy −1<tr⁡(S2)+tr⁡(R2)<1-1<\operatorname{tr}(\mathbf{S}^2)+\operatorname{tr}(\mathbf{R}^2)<1, tr⁡(S2)<10\operatorname{tr}(\mathbf{S}^2)<10 and tr⁡(R2)>−10\operatorname{tr}(\mathbf{R}^2)>-10.

Non-Correlation Between True gng_n and True bijb_{ij}

One may speculate whether the mapping between xtr\mathbf{x}_{tr} and true gng_n is the same as, or similar to between xtr\mathbf{x}_{tr} and true bijb_{ij}. If so, using true bijb_{ij} instead of true gng_n as the NUM analysis outputs may be simpler and acceptable if most of the data points do not cause false positive and false negative diagnoses of NUM detailed in § 4.1. Figure 8 shows scatter plots of xtr\mathbf{x}_{tr} plotted against true bijb_{ij} components from the periodic hills data. By qualitatively comparing Figure 8 with Figure 6(a)-(c), it can be seen that the mappings are not similar. For quantitative validation, Spearman’s rank correlation coefficients (SRCCs) between true gng_n and true bijb_{ij} were calculated [7]. The SRCC between two variables is a nonparametric measure of the statistical dependence between their ranked values, thereby indicating their monotonicity with a scale from -1 to 1, in which a SRCC of 1 or -1 indicates that the relationship is perfectly monotonic [18]. The SRCC values in table 1 demonstrate that the relationships between the true gng_n coefficients and true bijb_{ij} components are not monotonic, hence the relationships in xtr\mathbf{x}_{tr} vs. true gng_n must be different to those in xtr\mathbf{x}_{tr} vs true bijb_{ij}. Therefore, true gng_n must not be substituted by true bijb_{ij} in NUM analysis of 2D flow data. Figure 8 also shows that xtr\mathbf{x}_{tr} has one-to-many relations with true bijb_{ij}, as the data points in each subplot may be grouped according to their true bijb_{ij} values, and the groups may overlap in some parts of the xtr\mathbf{x}_{tr} space.

Scatter plots of $\mathbf{x}_{tr}$ against true $b_{ij}$ components

Figure 8.

Image not extracted

Figure 8 (PDF p. 22). Scatter plots of inputs xtrx_{tr} vs. (a) true b11b_{11}, (b) true b22b_{22}, (c) true b33b_{33}, and (d) true b12b_{12} components for the flow over periodic hills data.

true b11b_{11}true b22b_{22}true b33b_{33}true b12b_{12}
true g1g_10.08-0.540.420.52
true g2g_2-0.240.66-0.34-0.46
true g3g_30.300.36-0.83-0.17

Table 1 (PDF p. 22). Spearman’s rank correlation coefficients between true gng_n coefficients and true bijb_{ij} components

Impinging Jet Case Results

Non-Unique Mapping Quantification

The NUM quantification results for the impinging jet case are shown in Figure 9. The scatter plots in Figure 9(a)-(Figure 9c) show that most of the points are located in the LSR subset where tr(S2)<2tr(S^2) < 2 and tr(R2)>−2tr(R^2) > -2, as well as near the pure shear line. Similar to the periodic hills data, overlapping groups can be observed in these parts of the xtrx_{tr} space for all true gng_n coefficients. For example, Figure 9(a) shows that true g1≈−0.08g_1 \approx-0.08 or −0.02-0.02 for any value of xtrx_{tr} along the pure shear line in the zoomed-in view, thus demonstrating one-to-many relations between xtrx_{tr} and true g1g_1. Similar observations along the pure shear line can be made with true g2g_2 and g3g_3 in Figure 9(b) and Figure 9(c), respectively.

Image not extracted

Figure 9.

20 000 data points were randomly selected for the clustering process due to memory constraints. Although this amounts to only 18% of all data points in Figure 9(a), a comparison with Figure 9(d) shows that the selected points were still able to capture the overall distribution in the xtrx_{tr} space, and thereby represent it. Figure 9(d)-(Figure 9f) shows that the points have been reasonably grouped according to their true gng_n (n∈{1,2,3}n \in\{1, 2, 3\}) values by the clustering algorithm. For example, a comparison between Figure 9(b) and Figure 9(e) shows that the points along the pure shear diagonal that give true g2≈−0.02g_2 \approx-0.02 have been clustered in one group, true g2≈−0.06g_2 \approx-0.06 in another group and true g2<−0.1g_2 < -0.1 have been allocated to a third group. These clusters enabled the nCIn_{CI} heat maps calculated with NNM shown in Figure 9(g)-(Figure 9i) to be made. Like the periodic hills results, the heat maps support qualitative NUM observations made around Figure 9(a)-(Figure 9c), as nCIn_{CI} is mostly concentrated in LSR and near the pure shear line in the xtrx_{tr} space: 89%, 87% and 90% of the total nCIn_{CI} in the true g1g_1, g2g_2, and g3g_3 heat maps, respectively, are found in the heat map grid cells along the pure shear line. Furthermore, the heat map results are supported by Figure 17(d)-(Figure 17f), which show that running Algorithm 1 with PBM instead gives similar nCIn_{CI} distribution trends.

(g) heat map of nCIn_{CI} for true g1g_1 (h) heat map of nCIn_{CI} for true g2g_2 (i) heat map of nCIn_{CI} for true g3g_3

Figure 9 NUM results for the impinging jet flow data. Subplots (a)-(c) show scatter plots of inputs xtr\mathbf{x}_{tr} vs. outputs true gng_n, (d)-(f) show clusters represented by colours: red, blue, and green for 20 000 random points, and (g)-(i) show heat maps of nCIn_{CI} given by NNM (n=3n = 3), with the blue line indicating pure shear flow.

Low Strain and Rotation Regions

To identify regions in the domain that contain LSR and near pure shear flow, contours of the NUM analysis inputs and outputs are shown in Figure 10a-(Figure 10b) and Figure 10c-(Figure 10b (PDF p. 26)), respectively. The same hatching constraints used for the periodic hills data were applied. Two regions were found to contain LSR where tr(S2)<2tr(S^2) < 2 and tr(R2)>−2tr(R^2) > -2 as indicated by red hatching: one near the top wall and another near the bottom wall. Similar to the periodic hills case, these regions contain ∂uˉ1/∂x2\partial\bar{u}_1/\partial x_2 inflection points, and satisfy Eq. (12) due to low values of ∂uˉ1/∂x2\partial\bar{u}_1/\partial x_2 compared to the rest of the domain.

The mean streamwise velocity uˉ1\bar{u}_1 profiles in the x2x_2 direction at the recirculation zones are flatter near the top wall compared to near the bottom wall. Therefore, low values of ∂uˉ1/∂x2\partial\bar{u}_1/\partial x_2 are sustained for a greater x2x_2 distance near the top wall, which lead to greater x2x_2 thickness in the red hatched region there ([21]).

Figure 10c-(Figure 10b (PDF p. 26)) show both red hatched regions contain discontinuities in true gng_n with examples labelled (1). The cause of these discontinuities is the same as those in the periodic hills case – namely, very low S11S_{11} and S12S_{12} magnitudes. Although the discontinuities would result in NUM between xtr\mathbf{x}_{tr} and true gng_n, it has been explained in § 6.2 that the discontinuities are eliminated upon substitution in Eq. (6), and can therefore be ignored.

Near Pure Shear Regions

Figure 10 shows that near pure shear flow satisfying Eq. (13) marked by white hatching dominates the domain. Further indication of this is given by comparing Figure 10a with (b), which demonstrates that the distributions of tr(S2)tr(S^2) and tr(R2)tr(R^2) are almost perfectly negatively correlated with each other. It is found that the white hatched regions contain absolute values of S11S_{11} that are less than 0.5 and one order of magnitude smaller than S12S_{12} and R12R_{12}. Therefore, S12S_{12} and R12R_{12} must have similar absolute values to satisfy the condition in Eq. (13), which is only possible if ∣∂uˉ1/∂x2∣≫∣∂uˉ2/∂x1∣\left|\partial\bar{u}_1/\partial x_2\right| \gg\left|\partial\bar{u}_2/\partial x_1\right| as shown in Eq. (6). It was observed that ∣∂uˉ1/∂x2∣\left|\partial\bar{u}_1/\partial x_2\right| is two orders of magnitude greater than ∣∂uˉ2/∂x1∣\left|\partial\bar{u}_2/\partial x_1\right| in these regions. This shows that ∂uˉ1/∂x2\partial\bar{u}_1/\partial x_2 is the only non-negligible velocity gradient in these regions of this case and therefore, inputs tr(S2)tr(S^2) and tr(R2)tr(R^2) can be approximated as:

tr(S2)≈κ22ε2(∂uˉ1∂x2)2tr(R2)≈−κ22ε2(∂uˉ1∂x2)2(16)\begin{aligned} tr(S^2) &\approx\frac{\kappa^2}{2\varepsilon^2}\left(\frac{\partial\bar{u}_1}{\partial x_2}\right)^2 \\ tr(R^2) &\approx-\frac{\kappa^2}{2\varepsilon^2}\left(\frac{\partial\bar{u}_1}{\partial x_2}\right)^2 \tag*{(16)} \end{aligned}

Eq. (16) supports the almost perfect negative correlation found between tr(S2)tr(\mathbf{S}^{2}) and tr(R2)tr(\mathbf{R}^{2}) in the white hatched regions. Moreover, its derivation gives insight into how near pure shear flow arises in the impinging jet case. Figure 3(b) shows various flow physics including recirculation, wall jets and their interface are contained in the white hatched regions and satisfy Eq. (16).

Flow over periodic hills velocity contour

(a) Flow over periodic hills at Re=5600Re = 5600, based on bulk inlet velocity of 0.028 m/s and hill crest height of 1 m.

Plane impinging jet velocity contour

(b) Plane impinging jet flow at Re=2×104Re = 2 \times10^{4} based on bulk inlet velocity of 3 m/s and slot width of 0.1 m.

Figure 3. Schematic velocity contour plots of (a) flow over the periodic hills case, and (b) the impinging jet case.

It was discussed in the periodic hills results that NUM exists where there are almost constant values of xtr\mathbf{x}_{tr} and a large range of true gng_n. Figure 10 shows that this occurs in the white hatched regions with examples labelled (2). Hence, the standard inputs xtr\mathbf{x}_{tr} are insufficient in mapping to true gng_n accurately in this case as well. Furthermore, figure 10(a)-(b) show that tr(S2)tr(\mathbf{S}^{2}) and tr(R2)tr(\mathbf{R}^{2}) vary in both x1x_1 and x2x_2 directions in the white hatched regions, resulting in the magnitude of the inputs increasing towards the free jet. Contrastingly, figure 10(c)-(e) show that true gng_n vary primarily in the x2x_2 direction. For example, the magnitude of true g1g_1 increases towards the horizontal centreline where x2/B=2x_2/B = 2, and the magnitude of true g2g_2 and g3g_3 approach zero. Therefore, in the outer parts of the white hatched regions such as where 2<tr(S2)<42 < tr(\mathbf{S}^{2}) < 4 and −2>tr(R2)>−4-2 > tr(\mathbf{R}^{2}) > -4 labelled (3), one-to-many relations (and therefore NUM) exist between xtr\mathbf{x}_{tr} and true gng_n. This justifies why nCIn_{CI} increases as the magnitude of the inputs decrease along the pure shear line as shown in figure 9(g)-(i). Similar to the periodic hills case, the NUM in the white hatched regions must be addressed as it only contains true g1≤0g_1 \leq0.

Image not extracted

(a) (a) tr(S2)tr(\mathbf{S}^{2})

Image not extracted

(b) (b) tr(R2)tr(\mathbf{R}^{2})

Image not extracted

(c) (c) true g1g_1

Figure 10.

(a)

Contour plot of true $g_3$

(b)

Figure 10 (PDF p. 26). Contour plots of the (a)-(b) xtr\boldsymbol{x}_{tr} inputs and (c)-(e) true gng_n for the impinging jet case. Red hatching shows regions of low strain and rotation flow where tr(S2)<2tr(\boldsymbol{S}^2) < 2 and tr(R2)>−2tr(\boldsymbol{R}^2) > -2. White hatching shows regions of near pure shear flow that were not included in the red hatching and satisfy −1<tr(S2)+tr(R2)<1-1 < tr(\boldsymbol{S}^2) + tr(\boldsymbol{R}^2) < 1, tr(S2)<10tr(\boldsymbol{S}^2) < 10 and tr(R2)>−10tr(\boldsymbol{R}^2) > -10.

Non-Unique Mapping Reduction

Viscosity Ratio rνr_\nu as a Supplementary Input Variable

Having demonstrated that data from 2D flows can cause NUM in the GEVH, developing methods to reduce or, better still, eliminate NUM becomes a necessary step to improve TBML model training convergence and predictive accuracy. Previous studies showed that including a supplementary input variable (SIV) enabled NUM from 1D channel flow data to be eliminated, and SIVs have already been included in some TBML models in the literature [25, 31, 13]. Therefore, the hypothesis that SIVs reduce or eliminate NUM caused by 2D flow data was tested. The tasks became choosing an SIV that could address NUM and demonstrating the improvement.

While one-to-many relations were found to be mostly concentrated in the LSR and near pure shear flow regions in both cases, § 6.2 and § 7.2 explained that those in LSR regions can be ignored. Therefore, the SIV was chosen to primarily address NUM in near pure shear flow, which can be attributed to causing the most nCIn_{CI} anyway – approximately 45% of total nCIn_{CI} in both flow cases. In this study, the SIV was chosen to be tailored for the periodic hills case. It was explained in § 6.3 that the standard xtr\boldsymbol{x}_{tr} inputs are almost constant but the values of true gng_n vary greatly near its separation point as shown in figure 7. As the separated flow is the defining flow feature, an SIV that can address NUM there was sought after, and it was believed that this could be achieved if the distribution of the SIV is similar to the true gng_n coefficients there. While different SIVs proposed in the literature were investigated as shown in Appendix § 19, only two qualified with their distributions: the ratio of convection to production of TKE, and the turbulence Reynolds number RetRe_t that was used in the 1D channel flow studies. As the latter is simpler and its effectiveness in addressing 1D channel flow NUM has already been recognised, a variant of it which is more often found in data-driven turbulence models including TBML models called the viscosity ratio rν=νt/(100ν+νt)r_\nu= \nu_t/(100\nu+\nu_t) was chosen as the SIV, where νt\nu_t and ν\nu are the eddy and kinematic viscosity, respectively [56, 13] (Wu et al. 2022). Eddy viscosity was calculated as νt=Cμk2/ε\nu_t = C_\mu k^2/\varepsilon, where CμC_\mu is a constant equal to 0.09. Figure 7(f) shows rνr_\nu in the periodic hills case, which highly varies near the separation point in a similar manner to true gng_n as labelled (6). While rνr_\nu was chosen as the SIV, the authors do not rule out well-established or new SIVs in the literature outperforming rνr_\nu in reducing NUM. An investigation on comparing the performance of different SIVs would provide insightful results and may be subject to future work.

Scatter plots of true gng_n with xtr\mathbf{x}_{tr} on the xx- and yy-axes, and rνr_\nu on the zz-axis for the periodic hills data are shown in Figure 11a-(Figure 11c). By comparing these to figure 6(a)-(c), it can be seen that including rνr_\nu enables more distinct true gng_n distributions, such that the points are more separated but closer to others with similar true gng_n values, which improves the smoothness of the mapping between inputs and true gng_n. This can be clearly observed along the pure shear line of tr(S2)=−tr(R2)tr(\mathbf{S}^2)=-tr(\mathbf{R}^2), thus qualitatively demonstrating that including rνr_\nu as an SIV should reduce NUM in near pure shear flow. To facilitate quantitative investigation of the NUM reduction, the clustering process detailed in § 4.4 was extended to three dimensions. The modifications for this are detailed in Appendix § 14. For consistency with the results obtained from inputs xtr\mathbf{x}_{tr} in figure 6, all true gng_n coefficients were grouped into three clusters, which are shown in figure 11(d)-(f). A comparison with Figure 11a-(Figure 11c) shows that the points have been reasonably clustered in accordance with their true gng_n, (n∈{1,2,3}n \in\{1,2,3\}) values, and the clusters collapsed in the direction of the rνr_\nu axis resemble those in figure 6(d)-(f).

true $g_1$ values

(a) true g1g_1 values

true $g_2$ values

(b) true g2g_2 values

true $g_3$ values

(c) true g3g_3 values

Figure 11.

NUM results for the flow over periodic hills data: true $g_1$, $g_2$, and $g_3$ clusters and heat maps of $n_{CI}$

Figure 11 (PDF p. 28). NUM results with inputs {xtr,rν}\{x_{tr}, r_{\nu}\} for the flow over periodic hills data. Subplots (a)-(c) show scatter plots of inputs xtrx_{tr} and viscosity ratio rνr_{\nu} vs. outputs true gng_n, (d)-(f) show clusters represented by colours: red, blue, and green, and (g)-(i) show summed heat maps of nCIn_{CI} given by NNM (n=3n = 3), with the blue line indicating pure shear flow.

Non-Unique Mapping Reduction Quantification

To quantify the NUM, the {xtr,rν}\{x_{tr}, r_{\nu}\} input space was partitioned into a 3D grid, and the NUM quantification method detailed in § 5 was run with NNM extended to three dimensions, whereby Eq. (11) was modified as:

d=(tr⁡(S2)p−tr⁡(S2)q)2+(tr⁡(R2)p−tr⁡(R2)q)2+(rνp−rνq)2(17)d = \sqrt{\left(\operatorname{tr}(S^2)_p-\operatorname{tr}(S^2)_q\right)^2+\left(\operatorname{tr}(R^2)_p-\operatorname{tr}(R^2)_q\right)^2+\left(r_{\nu p}-r_{\nu q}\right)^2} \tag*{(17)}

and 3D heat maps of nCIn_{CI} were generated. To compare these results with the 2D heat maps given by just inputs xtrx_{tr} in figure 6(g)-(i) denoted HM1, the 3D heat maps were summed in the direction of the rνr_{\nu} axis to give 2D heat maps of nCIn_{CI} in the xtrx_{tr} space as shown in figure 11(g)-(i). These were subtracted from HM1 to give difference arrays, which show the difference in nCIn_{CI} throughout the xtrx_{tr} space after including rνr_{\nu} as an SIV. The process of generating the difference arrays is depicted in Figure 12.

Figure 12. Process of obtaining the difference arrays: (a) simplified schematic of a 3D heat map containing nCIn_{CI} for inputs xtrx_{tr} and rνr_{\nu}, (b) summation of the 3D heat map in the rνr_{\nu} axis to give the flattened heat map, (c) simplified schematic of the 2D heat map containing nCIn_{CI} for inputs xtrx_{tr}, (d) subtraction of the flattened heat map from the 2D heat map to give the difference array.

Figure 13(a)-(Figure 13c) show the difference arrays for the periodic hills data. These results demonstrate that NUM can reduce in the vast majority of the domain when an SIV is included. Choosing rνr_{\nu} as the SIV led to most improvement being located along the pure shear line as initially targeted. Table 2 shows including rνr_{\nu} reduced nCIn_{CI} by 45.7%, 60.3%, and 57.9% for true g1g_1, g2g_2, and g3g_3, respectively, in total. Difference arrays were also generated for the impinging jet data as shown in Figure 13(d)-(Figure 13f). It can be observed that much of the improvement is again located along the pure shear line, thus rνr_{\nu} as an SIV was able to target this subset of the xtrx_{tr} input space effectively in this case too. However, large increases in nCIn_{CI} can also be observed – especially in the true g3g_3 difference array, which led to lower total nCIn_{CI} reductions of 44.5%, 34.2%, and 5.8% for true g1g_1, g2g_2, and g3g_3, respectively, in the xtrx_{tr} space.

Difference arrays showing the change in number of conflicting instances

Figure 13. Difference arrays showing the change in number of conflicting instances ΔnCI\Delta n_{CI} throughout the xtr\mathbf{x}_{tr} space after including viscosity ratio rνr_{\nu} as a supplementary input variable. For the periodic hill case, ΔnCI\Delta n_{CI} for true g1g_1 is shown in (a), for true g2g_2 in (b), and for true g3g_3 in (c), while for the impinging jet case, ΔnCI\Delta n_{CI} for true g1g_1 is shown in (d), for true g2g_2 in (e), and for true g3g_3 in (f).

Casetrue g1g_1true g2g_2true g3g_3
Periodic hills (%)45.760.357.9
Impinging jet (%)44.534.25.8

Table 2. Total percentage reduction in conflicting instances across entire xtr\mathbf{x}_{tr} space after including viscosity ratio as a supplementary input variable.

It is suspected that when the input space increases in dimension due to the inclusion of another input variable, the close points of a focal point belonging to the same cluster may possibly move further away from the focal point in the input space. This can cause other points in close proximity to be promoted as the focal point’s close points. If these new close points do not belong to the same cluster as the focal point, then an increase in the number of conflicting instances occurs. Where the points move to in the new dimension is due to the choice of SIV. This indicates that the choice of SIV can lead to different amounts of NUM reduction in (i) different true gng_n coefficients and (ii) different flow case data. The NUM reduction in the periodic hills data may be significantly greater than in the impinging jet data due to rνr_{\nu} being specifically chosen to reduce NUM in the periodic hills data near the separation point. Therefore, this shows that causes of NUM in 2D flow data used for TBML model training should be identified, and the choice of SIVs should be tailored towards the flow data. Nonetheless, reductions in NUM for both cases and all true gng_n coefficients show that including SIVs should be a generally viable approach to reduce NUM.

Training Experiment

To demonstrate the effect of including rνr_{\nu} as an SIV in TBML model training, two TBNNs were trained. The TBNNs were identical, except the inputs of the first model were the two invariants shown in Eq. (4), while the inputs of the second model were the two invariants and rνr_{\nu}. These models will be referred to as tbnn-1 and tbnn-2. The periodic hill case that has been analysed thus far in the paper with steepness factor β=1.5\beta= 1.5 was used for training, whereas flow over a periodic hill with β=1\beta= 1 and 1.21.2 were used for validation and testing, respectively. As the cases are 2D, the two-dimensional version of TBNN with three tensor basis was used [44]. The models had three hidden layers with 10 hidden nodes per layer and rectified linear unit activation. Training was undertaken with a learning rate of 0.001, batch size of 32 and the Adam optimizer. Mean squared error loss between predicted bijb_{ij} and DNS bijb_{ij} was used as the true gng_n coefficient values can tend towards infinity in some flow regions. The models were evaluated on the validation set after every two epochs. If the average validation error over the last three evaluations was higher than the average over the three before, then model training was stopped early.

As shear anisotropy governs the shear flow downstream of the hill crest and subsequent reattachment, contours of predicted b12b_{12} are shown in Figure 14 (Xiao et al. [63]). The colorbar has been set to the lower and upper realizability bounds of -0.5 and 0.5. A comparison between the RANS prediction in Figure 14(a) with the DNS prediction in Figure 14(d) shows that while the distribution of the RANS prediction matches closely with the DNS, differences in magnitudes can be observed close to the wall, and above the hill crest. The tbnn-1 prediction does not match the DNS magnitudes closely either, due to discontinuities in the predictions which are believed to be caused by NUM. The tbnn-2 prediction contains magnitudes closer to the DNS compared to RANS, especially near the upper wall, above and downstream of the hill crest. Furthermore, it contains significantly less discontinuities compared to tbnn-1, hence suggesting a reduction in NUM can reduce discontinuous TBML model predictions.

Figure 14. Shear anisotropy b12b_{12} predicted by (a) RANS SST by McConkey et al. [36], (b) TBNN with the two standard invariant inputs: tr⁡(S2)\operatorname{tr}(\mathbf{S}^{2}) and tr⁡(R2)\operatorname{tr}(\mathbf{R}^{2}), (c) TBNN with the two standard invariant inputs and rνr_\nu as a SIV, and (d) DNS by Xiao et al. [63] for the β=1.2\beta= 1.2 periodic hill test case.

The mean squared error of the RANS, tbnn-1 and tbnn-2 predictions compared to DNS for the non-zero anisotropy components are given in Table 3. These quantitative results show that for all components, tbnn-1 gave more accurate predictions than RANS, and the inclusion of input rνr_\nu in tbnn-2 improved the predictions further. For example, a 24% improvement in the accuracy of b12b_{12} was observed between tbnn-1 and tbnn-2.

Methodb11b_{11}b12b_{12}b22b_{22}b33b_{33}
RANS0.03050.00410.02770.0095
TBNN without rνr_\nu0.00650.00240.00670.0066
TBNN with rνr_\nu0.00440.00140.00540.0044

Table 3. Mean squared error of predicted anisotropy components compared to true values from DNS for the β=1.2\beta= 1.2 periodic hill test case

Conclusion

A method for quantifying non-unique mapping (NUM) in data-driven Reynolds averaged turbulence models was proposed and deployed on the general effective viscosity hypothesis, which forms the basis of many such models in the literature, such as tensor-basis neural networks. Using data from a flow over periodic hills case and an impinging jet case, this study confirms for the first time that NUM can exist in these models with the standard invariant inputs when 2D cases are used to train them. The magnitude of this finding is underscored by the fact that such flow cases have been used to train most of these models in the literature. Furthermore, this discovery was demonstrated on two very different flow cases, which show that NUM is not limited to certain flow physics in the training data.

To identify types of flows that cause NUM, a pre-training metric for quantifying NUM throughout the input space called conflicting instances nCIn_{CI} was proposed in the method. For both flow cases, nCIn_{CI} was found to be highly concentrated in regions containing (i) low strain and rotation (LSR), and (ii) near pure shear flow. The former is caused by discontinuities in the target outputs, and the latter is due to various causes of one-to-many relations between the invariant inputs and the target outputs. These regions combined were found to cause approximately 76% and 89% of total nCIn_{CI} in the periodic hills and impinging jet case data, respectively. While it was found that NUM caused by LSR can be ignored due to vanishing discontinuities in the models, near pure shear flow gave the highest nCIn_{CI} regardless – approximately 45% of total nCIn_{CI} in both cases. These consistent findings should urge the community to examine their training data from cases containing near pure shear flow to potentially reduce NUM significantly in their data-driven turbulence models. As LSR and/or near pure shear flow (including homogeneous shear flow) exists in many flow cases, these findings are widely applicable and can help develop more accurate models for cross-case training, which is a critical challenge in the field today.

The method was repeated with viscosity ratio as a supplementary input variable (SIV) which was chosen for reducing NUM from pure shear flow in the periodic hills case. This led to total nCIn_{CI} being reduced for all outputs in both different flow cases. Therefore, it is demonstrated for the first time that including SIV(s) is a viable approach for reducing NUM caused by 2D flows in these data-driven turbulence models. This result also demonstrates the merit in choosing viscosity ratio and the proposed solution for reducing NUM. However, the overall nCIn_{CI} reduction varied with different output coefficients and flow cases: 46-60% for the periodic hills data and 6-45% for the impinging jet data. Smaller reductions for the impinging jet case were partly due to increases in nCIn_{CI} for some outputs in certain regions of the input space, which are believed to be caused by the choice of SIV. Given that viscosity ratio was chosen specifically to reduce nCIn_{CI} in the periodic hills data from near the separation region, this shows that the input and target outputs of each training case should always be analysed, and the choice of SIV(s) should be tailored towards the flow physics of the cases in order to effectively reduce NUM. Including the SIV in training a tensor basis neural network was found to improve prediction accuracy for all anisotropy components. As the types of testing flow cases are unknown in model deployment, SIVs that effectively reduce NUM over different types of flows should be investigated in the interest of cross-case training.

Acknowledgements

The authors would like to acknowledge the assistance given by Research IT and the use of the Computational Shared Facility at The University of Manchester.

Funding

This work was supported by the UK Engineering and Physical Sciences Research Council (grant numbers EP/W033542/1 and EP/T012242/2).

Author Declarations

The authors have no conflicts to disclose.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Clustering Process

Min-Max Normalization

While the clustering algorithm was required to group points of similar true gng_n (n∈{1,2,3}n \in\{1, 2, 3\}) values together, it was desirable to a lesser extent for it to also group points with similar xtr\mathbf{x}_{tr} (and rνr_\nu) values, as this led to points with similar inputs vs true gng_n (n∈{1,2,3}n \in\{1, 2, 3\}) relation to belong to the same cluster.

As the flow cases had different ranges of tr(S2)tr(\mathbf{S}^2), tr(R2)tr(\mathbf{R}^2), rνr_\nu, and true gng_n values, min-max normalization was performed to ensure that the clustering would not be affected by their scales of magnitude ([14, 67]):

x~=a(x−min⁡(x))max⁡(x)−min⁡(x)\widetilde{x} = \frac{a(x-\min(x))}{\max(x)-\min(x)}
a={0.1,x∈{tr(S2),tr(R2),rν}1,x∈{true gn}(18)a = \begin{cases} 0.1, & x \in\{tr(\mathbf{S}^2), tr(\mathbf{R}^2), r_\nu\} \\ 1, & x \in\{\text{true }g_n\} \end{cases} \tag*{(18)}

where xx in this appendix is an arbitrary variable that may represent tr(S2)tr(\mathbf{S}^2), tr(R2)tr(\mathbf{R}^2), rνr_\nu, or true gng_n (n∈{1,2,3}n \in\{1, 2, 3\}), and x~\widetilde{x} is their min-max normalized value. The factor aa ensured the range of normalized true gng_n values was 10 times greater than the range of normalized input values, enabling the clustering algorithm to be more weighted towards grouping points with similar true gng_n values than similar tr(S2)tr(\mathbf{S}^2), tr(R2)tr(\mathbf{R}^2) or rνr_\nu values ([45]).

Complete Linkage Agglomerative Clustering

The normalized data points were clustered using complete linkage agglomerative (CLA) clustering. Agglomerative clustering algorithms begin by treating each point as a cluster of its own. The two clusters that are separated by the closest distance are combined, and this step is repeated until a pre-specified number of clusters remain ([14]). In CLA, the distance between two clusters is given by the distance between their furthest points. The Euclidean distance was used with inputs xtr\mathbf{x}_{tr}:

d(p,q)=(tr(S2)~p−tr(S2)~q)2+(tr(R2)~p−tr(R2)~q)2+(gnt~p−gnt~q)2(19)d(p,q) = \sqrt{\left(\widetilde{tr(\mathbf{S}^2)}_p-\widetilde{tr(\mathbf{S}^2)}_q\right)^2+\left(\widetilde{tr(\mathbf{R}^2)}_p-\widetilde{tr(\mathbf{R}^2)}_q\right)^2+\left(\widetilde{g_n^t}_p-\widetilde{g_n^t}_q\right)^2} \tag*{(19)}

and with inputs xtr\mathbf{x}_{tr} and rνr_\nu:

d(p,q)=(tr(S2)~p−tr(S2)~q)2+(tr(R2)~p−tr(R2)~q)2+(rν~p−rν~q)2+(gnt~p−gnt~q)2(20)d(p,q) = \sqrt{\left(\widetilde{tr(\mathbf{S}^2)}_p-\widetilde{tr(\mathbf{S}^2)}_q\right)^2+\left(\widetilde{tr(\mathbf{R}^2)}_p-\widetilde{tr(\mathbf{R}^2)}_q\right)^2+\left(\widetilde{r_\nu}_p-\widetilde{r_\nu}_q\right)^2+\left(\widetilde{g_n^t}_p-\widetilde{g_n^t}_q\right)^2} \tag*{(20)}

where dd is the Euclidean distance between two arbitrary data points pp and qq, and true gng_n is represented by gntg_n^t. The tilde overbar x~\widetilde{\phantom{x}} denotes a min-max normalized quantity from Eq. (18), and subscripts pp and qq indicate which data point the quantity is from.

Channel Flow Analysis

Figure 15(a) shows the nondimensional velocity u+u^{+} and shear velocity gradient du+/dy+\mathrm{d}u^{+}/\mathrm{d}y^{+} profiles for the channel flow case described in Figure 2. It is clear that the velocity follows the well-established law of the wall profile given by equations u+=y+u^{+}=y^{+} and u+=ln⁡(y+)/κ+C+u^{+}=\ln(y^{+})/\kappa+C^{+}, where κ=0.41\kappa=0.41 and C+=5.5C^{+}=5.5. Figure 15(b) shows profiles of the inputs and true outputs of a TBML model for 1D channel flow without SIV. Non-dimensional shear velocity gradient α\alpha is the only input as explained in § 3.3, and true gng_n (n∈{1,2,3}n\in\{1,2,3\}) are the outputs. Drawing horizontal lines across Figure 15b shows that α\alpha has a one-to-many relation with the true outputs.

Velocity and shear velocity gradient profiles, and input and true output profiles
Velocity $u^{+}$ and shear velocity gradient $\mathrm{d}u^{+}/\mathrm{d}y^{+}$ profiles

(a) Velocity u+u^{+} and shear velocity gradient du+/dy+\mathrm{d}u^{+}/\mathrm{d}y^{+} profiles of the channel flow case given by RANS SST. The velocity and distance from the wall are nondimensionalised as u+=uˉ1/uτu^{+}=\bar{u}_1/u_{\tau} and y+=x2uτ/νy^{+}=x_2u_{\tau}/\nu, where friction velocity uτ=τw/ρu_{\tau}=\sqrt{\tau_w/\rho} and τw\tau_w is the wall shear stress.

Non-dimensional shear velocity gradient and true output profiles

(b) Non-dimensional shear velocity gradient α\alpha (=(k/ε)(∂uˉ1/∂x2)(k/\varepsilon)(\partial\bar{u}_1/\partial x_2)) given by RANS SST, true g1g_1, true g2g_2, and true g3g_3 profiles against nondimensional distance from the wall y+y^{+}.

Figure 15. Profiles of (a) velocity and velocity gradient, and (b) input α\alpha and true gng_n (n∈{1,2,3}n\in\{1,2,3\}) for the 1D channel flow case.

Figure 16 shows the NUM quantification results for the channel flow case. As α\alpha is the only input, the input space simplifies to one dimension which can be represented on a number line. This allowed cell-centre values of α\alpha to be plotted against true gng_n coefficients in Figure 16(a)-(Figure 16c). Drawing vertical lines on these plots shows there are one-to-many relations between input α\alpha and outputs true gng_n coefficients. The clustering approach described in § 4.4 was used to cluster these points, and the allocated cluster of each point are indicated by their color – blue or red – in Figure 16(a)-(Figure 16c). This representation allows clear visualisation and assessment on whether the clustering approach can identify the two different branches shown in Figure 2(a). Figure 16(a)-(Figure 16c) show that the approach can approximately cluster the points based on the two branches. While the clustering approach did not identify the branches perfectly (as some clusters contain points from both branches), most points that share similar values of α\alpha but have significantly different true gng_n (n∈{1,2,3}n\in\{1,2,3\}) values were successfully allocated into different clusters. Therefore, we believe the current clustering approach gives acceptable performance.

Figure 16. NUM results for the channel flow data. Subplots (a)-(c) show scatter points of the input α\alpha vs. outputs true gng_n (n∈{1,2,3}n \in\{1, 2, 3\}) data colored in blue or red according to their cluster. Subplots (d)-(f) show heat maps of nCIn_{CI} given by NNM (n=2n = 2).

With these clusters, Euclidean distance dd was calculated as d=(αp−αq)2d = \sqrt{(\alpha_p - \alpha_q)^2} instead of Eq. (11) to generate heat maps of nCIn_{CI} with the NNM described in § 5. Figure 16(d)-(Figure 16f) show that these heat maps effectively capture regions of the input space containing conflicting instances as indicators of NUM – especially where 3.25<α<3.53.25 < \alpha< 3.5 in the true g2g_2 and true g3g_3 results which contains the log-law region. In future work, we believe the clustering algorithm may be improved by enabling neighbouring points in the physical flow domain to be easily clustered together, which would allow the two branches to be more easily identified and more accurate results of nCIn_{CI} can be calculated.

Proximity Box Method Results

Figure 17 shows heat maps of nCIn_{CI} obtained by running Algorithm 1 with PBM. The box length ll and height hh depicted in Figure 5(b) were set to 1/8 times the grid spacing for both cases, which allowed the box dimensions to scale with the grid spacing. Comparing figure 6(g)-(i) with Figure 17(a)-(Figure 17c), and Figure 9(g)-(Figure 9i) with Figure 17(d)-(Figure 17f) shows that the heat maps obtained with NNM and PBM have very similar nCIn_{CI} distribution trends. Both methods show that nCIn_{CI} is highly concentrated in the LSR subset of xtr\mathbf{x}_{tr}, and near the pure shear line where tr(S2)=−tr(R2)tr(\mathbf{S}^{2})=-tr(\mathbf{R}^{2}). Therefore, the results given by NNM have been validated with those from PBM. The NNM and PBM results only differ in the magnitude of nCIn_{CI}, as the use of a proximity box in PBM naturally exacerbates nCIn_{CI} values in regions of the input space that contain densely packed points belonging to different clusters and reduces nCIn_{CI} where points are sparse.

Limits of True gng_n Terms

The conventional (ε,δ)(\varepsilon,\delta) definition of a limit for a multivariable infinite limit can be written as follows:

Let ff be a function of two variables (x,y)(x,y) whose domain DD includes points arbitrarily close to (a,b)(a,b).

We say that ff tends to ∞\infty as (x,y)(x,y) approaches (a,b)(a,b), and we write:

lim⁡(x,y)→(a,b)f(x,y)=∞(21)\lim_{(x,y)\to(a,b)} f(x,y)=\infty \tag*{(21)}

if for every number M>0M>0 there is a corresponding number δ=δ(M)>0\delta=\delta(M)>0, such that for all (x,y)∈D(x,y)\in D with Euclidean distance 0<(x−a)2+(y−b)2<δ0<\sqrt{(x-a)^2+(y-b)^2}<\delta, we have f(x,y)>Mf(x,y)>M ([2, 52]).

The expanded expressions for true g1g_1, and true g2g_2 shown in Eq. (3.1a), and Eq. (3.1b) respectively, contain terms of the form x/(x2+y2)x/(x^2+y^2), where xx and yy may represent S11S_{11} or S12S_{12}. The expression for true g2g_2 given in Eq. (3.1b) can be considered without R12R_{12} in the denominator to give terms of this form, as true g2g_2 is always multiplied by R12R_{12} in Eq. (6) when calculating bijb_{ij}. With the multivariable infinite limit definition in Eq. (21), it can be found that

lim⁡(x,y)→(0+,0)xx2+y2=∞,lim⁡(x,y)→(0−,0)xx2+y2=−∞.\lim_{(x,y)\to(0^+,0)}\frac{x}{x^2+y^2}=\infty, \qquad \lim_{(x,y)\to(0^-,0)}\frac{x}{x^2+y^2}=-\infty.

Therefore, as the (x,y)(x,y) variables approach (0,0)(0,0), the true g1g_1, and true g2g_2 expression terms tend towards negative infinity and infinity when the numerator is negative and positive, respectively. Hence, a discontinuity occurs where the numerator xx changes sign. The terms in the true g3g_3 expression in Eq. (3.1c) can be proven to tend towards infinity or negative infinity as (x,y)→(0,0)(x,y)\to(0,0) with the same method.

Smoothness of bijb_{ij} Terms

The expanded expressions for true g1g_1, and true g2g_2 are given in Eq. (3.1a), and Eq. (3.1b), respectively. Substituting these into the 2D GEVH in Eq. (6) gives terms of the form x2/(x2+y2)x^2/(x^2+y^2) and xy/(x2+y2)xy/(x^2+y^2), where xx and yy may represent S11S_{11} or S12S_{12}. To assess whether the limits that cause discontinuities in true gng_n terms also affect bijb_{ij} terms, the same limits as those in Appendix D were taken on these functions. However, it was revealed that the limits do not exist at (x,y)=(0,0)(x,y)=(0,0). Nonetheless, by plotting f(x,y)=x2/(x2+y2)f(x,y)=x^2/(x^2+y^2) and g(x,y)=xy/(x2+y2)g(x,y)=xy/(x^2+y^2), it is understood that both functions are continuous and differentiable except where (x,y)=(0,0)(x,y)=(0,0), thereby demonstrating that discontinuities do not exist in these functions ([52]). Hence, any discontinuities found in true g1g_1 and true g2g_2 would be eliminated upon substitution in Eq. (6). In addition, f(x,y)f(x,y) and g(x,y)g(x,y) are bounded between 0 and 1, and between −0.5-0.5 and 0.5, respectively. This implies that any changes in f(x,y)f(x,y) or g(x,y)g(x,y) between two points (which would be most significant near (x,y)=(0,0)(x,y)=(0,0)) would not exceed these bounds. Substituting the true g3g_3 expression given in Eq. (3.1c) into Eq. (6) allows (S112+S122)(S_{11}^2+S_{12}^2) to be cancelled out, thereby removing S11S_{11} and S12S_{12} from these bijb_{ij} terms.

Other Supplementary Input Variables

Figure 18 shows contours of other supplementary input variables (SIV) for the periodic hills case that are commonly used for data-driven turbulence modelling.

(a) Turbulence Reynolds number k2/νεk^2/\sqrt{\nu\varepsilon} (=νt/Cμν=\nu_t/C_\mu\nu)

Wall distance-based Reynolds number contours

(b) Wall distance-based Reynolds number min⁡(kd/50ν,2)\min(\sqrt{kd}/50\nu, 2)

(c) Turbulence intensity k/(0.5UiUi+k)k/(0.5U_iU_i+k)

Ratio of turbulent time scale to mean strain rate time scale contours

(d) Ratio of turbulent time scale to mean strain rate time scale ∥S∥k/(∥S∥k+ε)\lVert S\rVert k/(\lVert S\rVert k+\varepsilon)

(e) Non-dimensional Q-criterion (∥R∥2−∥S∥2)/(∥R∥2+∥S∥2)(\lVert R\rVert^2-\lVert S\rVert^2)/(\lVert R\rVert^2+\lVert S\rVert^2)

Ratio of total Reynolds stresses to normal Reynolds stresses contours

(f) Ratio of total Reynolds stresses to normal Reynolds stresses ∥ui′uj′∥/(k+∥ui′uj′∥)\lVert u_i^{\prime}u_j^{\prime}\rVert/(k+\lVert u_i^{\prime}u_j^{\prime}\rVert)

Figure 18.

Figure 18 (PDF p. 44). (g) Ratio of convection to production of TKE1 (h) Pressure gradient along streamline2 (i) Trace of squared asymmetric tensor associated with TKE gradient3 (j) Trace of squared asymmetric tensor associated with pressure gradient4. Contour plots of commonly used supplementary input variables in data-driven turbulence modelling demonstrated with the periodic hills case. (a) turbulence Reynolds number, (b) wall distanced-based Reynolds number, (c) turbulence intensity, (d) ratio of turbulent time scale to mean strain rate time scale, (e) non-dimensional Q-criterion, (f) ratio of total Reynolds stresses to normal Reynolds stresses, (g) ratio of convection to production of TKE, (h) pressure gradient along streamline, (i) trace of squared asymmetric tensor associated with TKE gradient, and (j) trace of squared asymmetric tensor associated with pressure gradient.

References

  1. [1]Aggarwal, C.C. & Reddy, C.K. 2014 Data Clustering: Algorithms and Applications. 1st ed. CRC Press, Boca Raton, Florida.
  2. [2]Bartle, R.G. & Sherbert, D.R. 2000 Introduction to Real Analysis. 3rd ed. John Wiley & Sons. Inc., New York.
  3. [3]Bishop, C.M. 2006 Pattern Recognition and Machine Learning. 1st ed. Springer Science+Business Media, LLC, Singapore.
  4. [4]Bishop, C.M. 1994 Mixture Density Networks. Technical report NCRG/94/004, Aston University, Birmingham.
  5. [5]Bush, R.H., Chyczewski, T., Duraisamy, K., Eisfeld, B., Rumsey, C.L. & Smith, B.R. 2019 Recommendations for Future Efforts in RANS Modeling and Simulation. Proceedings of the AIAA Scitech 2019 Forum, San Diego.DOI
  6. [6]Cai, J., Angeli, P-E., Martinez, J-M., Damblin, G. & Lucor, D. 2022 Reynolds Stress Anisotropy Tensor Predictions for Turbulent Channel Flow using Neural Networks. arXiv 2208.14301.
  7. [7]Corder, G.W. & Foreman, D.I. 2014 Nonparametric Statistics: A Step-by-Step Approach. 2nd ed. Wiley, New Jersey, USA.
  8. [8]de Berg, M., van Kreveld, M., Overmars, M. & Schwarzkopf, O. 2000 Computational Geometry Algorithms and Applications. 2nd ed. Springer, Berlin, Germany.
  9. [9]Duraisamy, K., Spalart, P.R. & Rumsey, C.L. 2017 Status, Emerging Ideas and Future Directions of Turbulence Modeling Research in Aeronautics. NASA Technical Memorandum 219682.
  10. [10]Duraisamy, K., Iaccarino, G. & Xiao, H. 2019 Turbulence Modeling in the Age of Data. Annual Review of Fluid Mechanics 51, 357–377.DOI
  11. [11]Durbin, P.A. & Pettersson Reif, B.A. 2011 Statistical Theory and Modeling for Turbulent Flows. 2nd ed. John Wiley & Sons Ltd, Chichester, United Kingdom.DOI
  12. [12]Durbin, P.A. 2018 Some Recent Developments in Turbulence Closure Modeling. Annual Review of Fluid Mechanics 50, 77–103.DOI
  13. [13]Ellis, C.D. & Xia, H. 2023 Data-driven turbulence anisotropy in film and effusion cooling flows. Physics of Fluids 35, 105114.DOI
  14. [14]Everitt, B.S., Landau, S., Leese, M. & Stahl, D. 2011 Cluster Analysis, 5th ed. John Wiley & Sons Ltd, Chichester, United Kingdom.
  15. [15]Gan, G., Ma, C. & Wu, J. 2007 Data Clustering: Theory, Algorithms, and Applications, 1st ed. Society for Industrial and Applied Mathematics, Philadelphia.DOI
  16. [16]Goodfellow, I., Bengio, Y. & Courville, A. 2016 Deep Learning, 1st ed. MIT Press, Cambridge, Massachusetts.
  17. [17]Gupta, A., Watson, S. & Yin, H. 2021 Deep learning-based aerial image segmentation with open data for disaster impact assessment. Neurocomputing 439, 22–33.
  18. [18]Hollander, M., Wolfe, D.A. & Chicken, E. 2014 Nonparametric statistical methods, 3rd ed. John Wiley & Sons, Inc., New Jersey.
  19. [20]Jacobs, R.A., Jordan, M.I., Nowlan, S.J. & Hinton, G.E. 1992 Adaptive mixtures of local experts. Neural Computation 3, 79–87.DOI
  20. [21]Jaramillo, J.E., Trias, F.X., Gorobets, A., Pérez-Segarra, C.D. & Oliva, A. 2012 DNS and RANS modelling of a turbulent plane impinging jet. International Journal of Heat and Mass Transfer 55, 789–801.DOI
  21. [22]Jasak, H., Jemcov, A. & Tukovic, Z. 2007 OpenFOAM: A C++ library for complex physics simulations, in International Workshop on Coupled Methods in Numerical Dynamics, Dubrovnik, Croatia.
  22. [23]Jiang, C., Vinuesa, R., Chen, R., Mi, J., Laima, S. & Li, H. 2021 An interpretable framework of data- driven turbulence modeling using deep neural networks. Physics of Fluids 33, 055133.DOI
  23. [24]Jongen, T. & Gatski, T.B. 1998 General explicit algebraic stress relations and best approximation for three-dimensional flows. International Journal of Engineering Science 36, 739–763.DOI
  24. [25]Kaandorp, M.L.A. & Dwight, R.P. 2020 Data-driven modelling of the Reynolds stress tensor using random forests with invariance. Computers and Fluids 202, 104497.DOI
  25. [26]Kim, W-W. & Menon, S. 1995 A new dynamic one-equation subgrid-scale model for large eddy simulations, in 33rd Aerospace Sciences Meeting and Exhibit, Reno, USA.
  26. [27]Kutz, J.N. 2017 Deep learning in fluid dynamics. Journal of Fluid Mechanics 814, 1–4.DOI
  27. [28]Lien, F.S., Chen, W.L. & Leschziner, M.A. 1996 Low-Reynolds-Number Eddy-Viscosity Modelling Based on Non-Linear Stress-Strain/Vorticity Relations. Engineering Turbulence Modelling and Experiments 3, 91–100.DOI
  28. [29]Ling, J., Kurzawski, A. & Templeton, J. 2016 Reynolds averaged turbulence modelling using deep neural networks with embedded invariance. Journal of Fluid Mechanics 807, 155–166.DOI
  29. [30]Liu, D., Tan, Y., Khoram, E. & Yu, Z. 2018 Training Deep Neural Networks for the Inverse Design of Nanophotonic Structures. ACS Photonics 5, 1365–1369.DOI
  30. [31]Liu, W., Fang, J., Rolfo, S., Moulinеc, C. & Emerson, D.R. 2021 An iterative machine-learning framework for RANS turbulence modelling. International Journal of Heat and Fluid Flow 90, 108822.DOI
  31. [32]Man, A., Jadidi, M., Keshmiri, A., Yin, H. & Mahmoudi, Y. 2022 Optimising a Machine Learning Model for Reynolds Averaged Turbulence Modelling of Internal Flows. Proceedings of the 16th International Conference on Heat Transfer, Fluid Mechanics and Thermodynamics and Editorial Board of Applied Thermal Engineering, online.
  32. [33]Man, A., Jadidi, M., Keshmiri, A., Yin, H. & Mahmoudi, Y. 2023 A divide-and-conquer machine learning approach for modeling turbulent flows. Physics of Fluids 35, 055110.DOI
  33. [34]Mandler, H. & Weigand, B. 2022 A realizable and scale-consistent data-driven non-linear eddy viscosity modeling framework for arbitrary regression algorithms. International Journal of Heat and Fluid Flow 97, 109018.DOI
  34. [35]Mandler, H. & Weigand, B. 2023 Feature importance in neural networks as a means of interpretation for data-driven turbulence models. Computers and Fluids 265, 105993.DOI
  35. [36]McConkey, R., Yee, E. & Lien, F-S. 2021 A curated dataset for data-driven turbulence modelling. Scientific Data 8:255.DOI
  36. [37]McConkey, R., Yee, E. & Lien, F-S. 2022 On the Generalizability of Machine-Learning-Assisted Anisotropy Mappings for Predictive Turbulence Modelling. International Journal of Computational Fluid Dynamics 36:7, 555–577.DOI
  37. [38]Menter, F.R. 1993 Zonal two equation k-ω turbulence models for aerodynamic flows, in Proceedings of the 23rd Fluid Dynamics, Plasmadynamics, and Lasers Conference, Orlando, USA.DOI
  38. [39]Mishra, A.A. & Girimaji, S.S. 2013 Intercomponent energy transfer in incompressible homogeneous turbulence: multi-point physics and amenability to one-point closures. Journal of Fluid Mechanics 731, 639–681.DOI
  39. [40]Papadopoulos, A.N. & Manolopoulos, Y. 2005 Nearest Neighbor Search: A Database Perspective, 1st ed. Springer, New York, USA.DOI
  40. [41]Parish, E.J. & Duraisamy, K. 2016 A paradigm for data-driven predictive modeling using field inversion and machine learning. Journal of Computational Physics 305, 758–774.DOI
  41. [42]Parmar, B., Peters, E., Jansen, K.E., Doostan, A. & Evans, J.A. 2020 Generalized Non-Linear Eddy Viscosity Models for Data-Assisted Reynolds Stress Closure. Proceedings of the AIAA Scitech 2020 Forum, Orlando.
  42. [43]Pedregosa, F. et al. 2011 Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research 12, 2825–2830.arxiv.org/abs/1201.0490
  43. [44]Pope, S.B. 1975 A more general effective-viscosity hypothesis. Journal of Fluid Mechanics 72, 331–340.DOI
  44. [45]Rebala, G., Ravi, A. & Churiwala, S. 2019 An Introduction to Machine Learning, 1st ed. Springer, Cham, Switzerland.
  45. [46]Rousseeuw, P.J. 1987 Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics 20, 53–65.DOI
  46. [47]Shizawa, M. 1994 Regularization Networks for Approximating Multi-Valued Functions: Learning Ambiguous Input-Output Mappings from Examples. Proceedings of 1994 IEEE International Conference on Neural Networks, Orlando.DOI
  47. [48]Shukla, A.K. & Dewan, A. 2019 OpenFOAM based LES of slot jet impingement heat transfer at low nozzle to plate spacing using four SGS models. Heat and Mass Transfer 55, 911–931.DOI
  48. [49]Singh, A.P., Medida, S. & Duraisamy, K. 2017 Machine-Learning-Augmented Predictive Modeling of Turbulent Separated Flows over Airfoils. AIAA Journal 55, No. 7, 2215–2227.arxiv.org/abs/1608.03990
  49. [50]Slotnick, J., Khodadoust, A., Alonso, J., Darmofal, D., Gropp, W., Lurie, E. & Mavriplis, D. 2014 CFD Vision 2030 Study: A Path to Revolutionary Computational Aerosciences. NASA Contractor Report 218178.
  50. [51]Sotgiu, C., Weigand, B., Semmler, K. & Wellinger, P. 2019 Towards a general data-driven explicit algebraic Reynolds stress prediction framework. International Journal of Heat and Fluid Flow 79, 108454.DOI
  51. [52]Stewart, J. 2008 Calculus: Early Transcendentals, 6th ed. Thomson Learning, Inc., Belmont, California.
  52. [53]Taghizadeh, S., Witherden, F.D., Hassan, Y.A. & Girimaji, S.S. 2021 Turbulence closure modelling with data-driven techniques: Investigation of generalizable deep neural networks. Physics of Fluids 33, 115132.DOI
  53. [54]Tang, H., Wang, Y., Wang, T., Tian, L. & Qian, Y. 2023 Data-driven Reynolds-averaged turbulence modeling with generalizable non-linear correction and uncertainty quantification using Bayesian deep learning. Physics of Fluids 35, 055119.DOI
  54. [55]Uno, Y., Fukumura, N., Suzuki, R. & Kawato, M. 1995 A computational model for recognizing objects and planning hand shapes in grasping movements. Neural Networks 8, 839–852.DOI
  55. [56]Volpiani, P.S., Meyer, M., Franceschini, L., Dandois, J., Renac, F., Martin, E., Marquet, O. & Sipp, D. 2021 Machine learning-augmented turbulence modeling for RANS simulations of massively separated flows. Physical Review Fluids 6, 064607.DOI
  56. [57]Weatheritt, J. & Sandberg, R. 2016 A novel evolutionary algorithm applied to algebraic modifications of the RANS stress–strain relationship. Journal of Computational Physics 325, 22–37.DOI
  57. [58]Weatheritt, J. & Sandberg, R. 2017 The development of algebraic stress models using a novel evolutionary algorithm. International Journal of Heat and Fluid Flow 68, 298–318.DOI
  58. [59]Wilcox, D.C. 2006 Turbulence Modeling for CFD, 3rd ed. DCW Industries, La Canada, California.
  59. [62]Xiao, H. & Cinnella, P. 2019 Quantification of model uncertainty in RANS simulations: A review. Progress in Aerospace Sciences 108, 1–31.arxiv.org/abs/1806.10434
  60. [63]Xiao, H., Wu, J-L., Laizet, S. & Duan, L. 2020 Flows over periodic hills of parameterized geometries: A dataset for data-driven turbulence modeling from direct simulations. Computers and Fluids 200, 104431.arxiv.org/abs/1910.01264
  61. [65]Zezula, P., Amato, G., Dohnal, V. & Batko, M. 2006 Similarity Search: The Metric Space Approach, 1st ed. Springer Science+Business Media, Inc., New York.
  62. [66]Zhang, Z., Song, X-D., Ye, S-R., Wang, Y-W., Huang, C-G., An Y-R. & Chen, Y-S. 2019 Application of deep learning method to Reynolds stress models of channel flow based on reduced-order modeling of DNS data. Journal of Hydrodynamics 31(1), 58–65.DOI
  63. [67]Zheng, A. & Casari, A. 2018 Feature Engineering for Machine Learning, 1st ed. O’Reilly Media, Inc., Sebastopol, California.
  64. [68]Zhou, X-H., Han, J. & Xiao, H. 2021 Learning nonlocal constitutive models with neural networks. Computer Methods in Applied Mechanics and Engineering 384, 113927.DOI

Paper details

vlnc:W_x4vkz7dz
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents