As reported in
Section 2 and
Figure 3, both the Spalart–Allmaras and Shear Stress Transport turbulence models fail to predict vortex interactions at moderate to high flight angles. The members of the AVT-316 group consistently made the same observation, apart from one outlier. We have proposed a few methods to improve the prediction performance. Then, we combined these to achieve better results. The first method is the application of well-known rotational corrections. Applying these corrections improved the
-
ω-based results at the mentioned flow range, but the improvement did not resolve the problem, as will be shown in
Section 6.4.
We have experimented with the idea of using a more vortex-capable turbulence model. One notable candidate is Rotta’s
-based model [
25]. Menter and Egorov implemented a new turbulence model based on Rotta’s original idea [
9] and claimed that the new model has LES-like properties [
10,
26]. This feature is promising for the vortical flows, since the separated flows are often simulated better with LES-based methods. The
k-kL method is developed and reported by Abdol-Hamid, and compressible extensions are provided [
11,
12]. Therefore, we have implemented and tested Abdol-Hamid’s method on a compressible separated flow around a missile fin [
14].
4.1. Verification of the Implementation
The
k-kL-MEAH2015 turbulence model formulation, development procedure, and extensive reference data for the validation cases were provided by Abdol-Hamid et al. [
11,
12,
27]. The model formulation is available on the NASA Turbulence Modeling Resource website, which forms a valuable collection and classification of standard turbulence models [
28]. The authors of the article followed the available documentation, while implementing the model into their solver. The formulation of the implemented model is given in (
2) and (
3). The turbulence quantities influence the flow variables via the turbulent viscosity parameter, computed as
.
The production term of the turbulence transport equations is calculated by (
4). The major component of turbulence production is the term containing the local shear strain appearing in the standard version. According to this formulation, the major component of the turbulence source is
, given that
.
We have followed the verification steps in the original studies of Abdol-Hamid. The implementation was verified with the standard test case simulations for which the computational data are available [
28]. Verification with the 2D flat plate and 2D bump cases was reported previously by the authors [
14]. Verification data is composed of pressure and skin friction distributions throughout the wall surface and the velocity and turbulent parameter (
k,
) profiles within the boundary layer extracted at a couple of sections. The published verification cases were run using the CFL3D and FUN3D codes, maintained by NASA, and the TAU code, maintained by the German Aerospace Center (DLR). All flow field parameters were very close to those calculated by the other three codes. An extensive discussion regarding the verification of our implementation of the
k-
-MEAH2015 model is given in the study of Dikbaş and Baran [
14].
The results of the
k-
model for standard turbulence test cases are in-line with other standard models, showing that the
k-
model is a reliable turbulence model. In order to reveal the scale-adaptive characteristics of the model claimed by Menter et al. and also to evaluate the model’s performance in vortical flows, the implemented model was further validated with a simplified vortical flow case. The reference experimental study conducted by Beresh et al. [
15,
16,
17] investigates the dynamics of a single vortex that is separated from an isolated fin exposed to an external flow. The authors of the current paper showed that the
k-
turbulence model predicts more realistic turbulent characteristics within the vortex core, compared to other equation models. Therefore, better agreement with the experimentally determined flow field could be obtained. These validation results can also be seen in Ref. [
14]. The results show that the vortex core originating from the fin is preserved much better with the
k-
method than with the standard models. The study claims that the
k-
model prevents excessive turbulence production at the vortex core, unlike the SA and SST models.
4.2. Overview of Improvements on the k- Model
As the second step in our search for a better vortex-capable turbulence model, we have implemented the
k-
turbulence model, a relatively new method tested for the current type of problem for the first time. As it will be shown in
Section 6, the results were surprisingly good, yet there is room for improvement. To achieve more accurate predictions for the transonic flow problems involving strong vortex interactions, we have put effort into the optimization of the
k-
turbulence model.
Several sources of error in the flowfield calculations involving vortical compressible flows are as follows:
The prediction of vortex flows is extremely sensitive to domain resolution and numerical settings [
5,
29,
30].
The turbulence models estimate excessive turbulent viscosity and, therefore, tend to dissipate the vortex rapidly. Therefore, the dissipation characteristics of the numerical model are also important [
5,
6,
31].
A shear layer is formed between the vortex and the freestream flow, and the interaction effects with shock waves and surfaces should be considered carefully [
32].
We are addressing these three problems in our research program. The third problem is determined as the main focus of this paper; therefore, improvements of the
k-
turbulence model are sought with the aim of enhanced RANS simulations for the described test problem. With the application of different turbulence production formulations and compressibility and free jet corrections, more realistic predictions could be made, as presented in
Section 6.
For this purpose, we have implemented the
k-
model to allow modifications to the turbulence production term. The new implementation permits choosing alternative production formulations. These production formulation alternatives may make use of either local strain magnitude (
S), local vorticity magnitude (
), or both
S and
. The original implementation of Abdol-Hamid [
11] is based on total strain tensor magnitude. We have adopted different production formulations based on the reference works done for other turbulence models for the
turbulence model. The designations associated with each option are recommended considering the principles endorsed in NASA Turbulence Modeling Resource [
28]. In this study, we have replaced the major component of the production term with Kato’s [
33] formulation. Therefore, the major total strain term,
, in the original model is replaced by
. Likewise, another alternative that uses only vorticity magnitude can be generated, i.e.,
. We expect that the employment of the vorticity magnitude in the calculation of turbulence production helps avoid the unphysical accumulation of computed turbulent quantities caused by shear stress.
The
k-kL model implemented in our solver also supports the vorticity corrections [
34,
35] widely used for a broad range of rotational flows. Among these, we have adopted the rotation/curvature (RC) correction, which is the one customized by Menter et al. [
36] for the two-equation SST model. We have also refined the RC correction for our
k-kL turbulence model implementation. It has already been shown that the vorticity corrections are able to provide better vortex predictions via compensation of the over-dissipative character of Boussinesq hypothesis-based turbulence models [
31,
37]. The present test case, however, will give the opportunity for the performance evaluation of the RC correction in the context of the prediction of the formation of a leading edge vortex.
Table 2 summarizes the alternative turbulence production and vorticity correction formulations.
Another improvement to the
k-
model in our implementation is the support for different types of compressibility corrections. In our solver, the compressibility correction is accompanied by free shear correction, as suggested by Abdol-Hamid [
12] in their jet correction definition for the
k-
-MEAH2015+J turbulence model. The formulation of the compressibility correction in the current implementation is generalized with modifiable coefficients. As given in Equation (
5), the coefficients of destruction terms in
k and
equations are increased by factors depending on the compressibility correction parameter,
. This parameter is given in Equation (
6) as a function of the turbulent Mach number,
. In order to eliminate low subsonic speeds from the compressibility correction, a cut-off value for
, denoted as
may be used as suggested by Wilcox [
38]. The original formulation of Sarkar et al. [
39] does not contain such a threshold. The free shear correction is active unless the compressibility correction is turned off.
Table 3 summarizes some alternatives that can be applied with the improved
k-
model. The modified Wilcox correction represents the alternative, described by Abdol-Hamid et al. [
12], as
k-
-MEAH2015+J.
It should be noted that the alternative production formulations actively change the turbulence model only at high-curvature or high-vorticity flows. These formulations do not alter the results for standard turbulence configurations, as several earlier studies have suggested (see, e.g., [
40,
41,
42]). We have repeated our validation test campaign as discussed in
Section 4.1. The standard compressibility correction is already verified in our previous study. New compressibility corrections simply change the parameters, and they are only effective in higher Mach numbers and jet flows. Hence, they do not alter the results in the standard test case, either.