Aerospace Science and Technology 106 (2020) 106067

Contents lists available at ScienceDirect

Aerospace Science and Technology

www.elsevier.com/locate/aescte

Improving predictions of transitional and separated flows using RANS

modeling
Chunhua Sheng
The University of Toledo, Toledo, OH, 43606, USA

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, new concepts are presented from the aspects of viscous modeling and numerical scheme
Received 4 December 2019 in an attempt to improve the prediction of rotor hover performance and associated viscous flow physics
Received in revised form 3 July 2020 under a RANS statistic modeling framework. An improved viscous modeling is proposed to address the
Accepted 9 July 2020
overprediction of flow separations faced by RANS modeling, which modifies the linear eddy viscosity
Available online 20 July 2020
Communicated by Cummings Russell
models based on a new separation correction method. This concept is seamlessly implemented into
two popular eddy viscosity models: the Spalart-Allmaras one-equation model and Menter’s shear stress
Keywords: transport two-equation model. A new high order unstructured grid WENO scheme is adopted to improve
Rotor the prediction of mean flow resolutions as well as viscous flow structures around the hovering rotor.
Transition Numerical results show significant improvements achieved in the prediction of rotor hover performance
Separation as well as viscous flow phenomena including boundary layer transitions and flow separations.
Turbulence model © 2020 Elsevier Masson SAS. All rights reserved.
High order scheme

1. Introduction popular Menter and Langtry correlation-based γ − Reθ transition

model [5]. The central idea of the γ − Reθ model is to use a con-
cept of vorticity Reynolds number to link the local transition onset
It is apparent that turbulence modeling and simulation is sig-
Reynolds number in boundary layers. However, as pointed out by
nificant in both scientific knowledge and engineering applications.
Robinson [6] the correlation of the rotational loop and the vortic-
Boundary layer transition and separation are the two most compli-
ity can be weak, there may be other ways or quantities [7] to link
cated viscous flow phenomena, which can occur in many aerospace
the flow transition in a general sense. In the RANS modeling cate-
applications. Examples include external flows over high-lift aircraft
gory, the γ − Reθ transition model [5] is still the prevailing method
wings and helicopter rotors or internal flows in turbo machines used to predict the transition onset in engineering applications,
and ducted fans. However, as pointed out by Spalart [1], turbu- which was developed under the frameworks of the Menter shear
lence predictions using Reynolds-averaged Navier-Stokes (RANS) stress transport (SST) k − ω turbulence model [5,8] and the Spalart-
methods face three primary challenges: (a) transition of bound- Allmaras (SA) one-equation turbulence model [9,10]. Comparative
ary layer, (b) growth and separation of boundary layer, and (c) studies indicated that compatible results have been achieved in
momentum growth after separation. This is due to the fact that transition predictions using both frameworks in 2-D benchmark
the current eddy viscosity based models, which are used in most flows and 3-D realistic helicopter rotors [11].
RANS methods, generally do not contain the mechanisms to re- The prediction of separated flows, including separation incep-
spond to the full physics of turbulent flows such as anisotropy of tion (Challenge I [1]) and growth of separated flows after inception
the Reynolds shear stress or streamline curvature in vortical and (Challenge II [1]), still remains a challenge using linear eddy vis-
separated flows [2]. cosity models. Spalart, et al. [1,12] introduced the non-Boussinesq
In terms of predictions of boundary layer transition, or Chal- quadratic constitutive relationships (QCR) into the simple Spalart-
lenge Zero classified by Spalart [1], much progress has been made Allmaras one-equation model, which demonstrated the ability to
over the past few decades. Examples include the e N method by capture the secondary flow in a square channel. Hellsten [13], as
Smith and Gamberoni [3] based on the linear stability theory, the well as Wallin and Johansson [14] proposed an explicit algebraic
laminar kinetic energy method of Walters and Leylek [4], and the Reynolds stress model (EARSM), which is an intermediate formu-
lation between linear eddy viscosity and the nonlinear Reynolds
stress model (RSM) to account for the effects of anisotropy with-
out increasing the number of turbulence model equations to be
E-mail address: [email protected]. solved. A successful prediction of separation zone was reported us-

https://doi.org/10.1016/j.ast.2020.106067
1270-9638/© 2020 Elsevier Masson SAS. All rights reserved.
2 C. Sheng / Aerospace Science and Technology 106 (2020) 106067

Nomenclature

c local blade chord r rotor radial axis

Cf local skin friction coefficient ( 0.5ρτ w2 r 2 ) R rotor radius
CP rotor power coefficient ( πρ P3 R 5 ) T rotor thrust
Q γ intermittency
CQ rotor torque coefficient ( πρ  2 R5 )
T
γsep corrected intermittency
CT rotor thrust coefficient ( πρ 2 R 4 ) θ collective pitch angle

FM rotor figure of merit ( CT CT
) ν̃ transport eddy viscosity
2 CQ
k turbulent kinetic energy σ rotor solidity
M tip tip Mach number ω turbulence decay frequency
P rotor power  rotor rotational speed
Q rotor torque τw wall shear stress

ing EARSM at the wing-fuselage junction of a wing-body-nacelle- 2. Methodologies

pylon (WBNP) configuration [15]. However, as stated by Spalart [1],
it is unlikely for a RANS model to provide the accuracy needed in In this section, two main concepts or strategies are presented to
a variety of vortical and separated flows. These complex viscous improve the prediction of complex viscous flows involving bound-
flows may require large eddy simulation (LES) or direct numerical ary layer transition and flow separation. The first strategy is to
simulation (DNS) [16], which is currently infeasible for industrial introduce a separation correction method or a stall delay model
applications at a full-scale Reynolds number due to prohibitively (SDM) [17] into the current linear eddy viscosity models to boost
the turbulence production in separated flows. This strategy is com-
large amount of computing resources required [1,2].
bined with the Langtry and Menter transition model [5] to provide
It is envisioned that the RANS modeling will remain a pre-
modeling capabilities for both boundary layer transition and flow
vailing method in practical engineering applications in a foreseen
separation under two popular linear eddy viscosity models: the
future, such as rotorcraft aerodynamic design and analysis. How- Spalart-Allmaras one-equation turbulence model [9] and Menter’s
ever, one particular difficulty encountered in rotor RANS modeling shear stress transport (SST) two-equation κ − ω turbulence model
is the overprediction of flow separations on blade surfaces [17]. [8]. The second strategy is based on a new high order WENO
It is widely believed that the underestimation of turbulent shear scheme recently developed by Zhong and Sheng [27], which has
stresses in separated flow region is attributed to this issue in the achieved a high order of accuracy on benchmark flows imple-
current RANS methods. For example, Aupoix, et al. stated in their mented on general structured and unstructured grids. The intent of
article [18] that “The prediction of the separation point is now fairly the high order scheme is to obtain increased resolutions on mean
well understood. However, the model behavior in the separation region, flows as well as viscous flow structures.
especially close to the separation and reattachment points, is still an is- All strategies are developed based on a three-dimensional un-
sue as models usually underestimate turbulence in this region.” Rumsey steady RANS CFD framework using unstructured mixed element
[19] also made a similar statement that “All models – including full grid topology [29]. Three-dimensional unsteady compressible gov-
erning equations are solved using a vertex-centered finite-volume
second-moment Reynolds shear stress models (RSM) – underpredicted
scheme. In order to solve flows from essentially incompressible
the level of turbulent kinetic energy and turbulent shear stress in the
to supersonic Mach number, a preconditioning technique [29] is
separated shear layer.” These underlying issues in RANS modeling
introduced into the time derivative term of the compressible gov-
have caused inaccurate predictions of rotor aerodynamic perfor- erning equations. Convective fluxes are computed using the Roe’s
mance (such as figure of merit), which have been reported in the flux approximate scheme [30]. Viscous fluxes are evaluated using a
recent series of workshops on rotor hover simulations [20–26]. directional directive method described in [31]. The nonlinear sys-
This article presents new concepts and strategies to improve the tem of equations is solved using the Newton’s implicit method,
prediction of rotor viscous flows involving boundary layer transi- with the Newton sub-iterations as the primary method to march
tion and flow separation based on RANS statistic modeling frame- the unsteady solution at each time step, and symmetric Gauss-
works. There are two main aspects involved in these strategies. Seidel relaxations as the secondary method to solve the linear
The first one is to introduce a separation correction method [17] system of equations at each Newton’ iteration. In addition, there
to linear eddy viscosity models in order to address the issue as- is an unstructured mesh deformation technology using radial ba-
sociated with the overprediction of separated flows. This method sis functions [32] developed in the solver for rotor collective pitch
is seamlessly combined with the transition modeling [5] and pro- sweep studies. Details of the numerical methods and solution pro-
cedures can be found in [29,32].
vides predictive capabilities for both transition and separation in
rotor aerodynamic analysis. The second aspect in the current strat-
2.1. Separation correction method
egy is to adopt a new high order unstructured grid scheme [27] to
improve the mean flow resolution as well as turbulence structure,
In the context of RANS statistical modeling, there is a well-
based on a weighted essentially non-oscillatory scheme (WENO-
known issue of underestimated turbulence in predicting viscous
ZQ) [28]. In the following, methodologies regarding these strategies flows close to separation point [16]. This is even true with the full
are described in Section 2. Assessments of the improved viscous second moment Reynolds shear stress model [19]. The underesti-
models are presented in Section 3, and assessments of the new mated turbulence, reflected as reduced turbulent kinetic energy or
high order unstructured grid scheme are presented in Section 4. low eddy activity, may cause premature flow separation or overes-
Concluding remarks and future works are summarized in the final timation of separation zone. This has been reported in both fixed-
Section 5. wing and rotary wing applications [19–26,33]. For example, data
 C. Sheng / Aerospace Science and Technology 106 (2020) 106067 3

Fig. 1. Streamwise velocity over NACA 4412 airfoil, shows trailing edge separation. (a) F reattach only shows effect on attached boundary layers. (b) Old γsep [5] only corrects
the laminar separation zone near leading edge. (c) F θ t shows effect on both laminar and turbulent separation zones. (d) New γsep [17] corrects both laminar and turbulent
separation zones. (For interpretation of the colors in the figure(s), the reader is referred to the web version of this article.)

scattering among the CFD participants for a common transport inar and turbulent flow separations. It is important to note that
aircraft in the Drag Prediction Workshops [19] was attributed to this modification does not affect the original turbulence model or
incorrect predictions of separated flows at the wing-body juncture transition model in fully attached laminar and turbulent flows, and
area. An overestimation of the side-of-body flow separation was only takes effect within the separation zone. In fact, this modifica-
linked to the original simple linear constitutive model that caused tion has a similar effect as the non-Boussinesq quadratic consti-
thicker boundary layer at the corner due to low momentum flows. tutive relationships (QCR) proposed by Spalart [1,12], which alters
Yamamoto, et al. [33] applied the quadratic constitutive relation the shear stresses directly by introducing the non-linear eddy vis-
(QCR) proposed by Spalart [1] and reduced the corner flow sepa- cosity terms in separated flows.
ration considerably. Rumsey [19] found that QCR [12] and EARSM The new separation correction method [17] was obtained by
[13,14] had a similar effect on the reduction of corner flow sep- removing the blending function F reattach from the original method
arations. In rotary-wing applications, Gardarein and Le Pape [23] [5], which can be written as
reported an overpredicted separation zone for two Sikorsky S-76       
Re v
rotor tip shapes (swapped and rectangular tips) using the k − ω full γsep = min S 2 · max 1, min − 1 , 1 Fθt , 2
 θ t tran- 3.235 · Reθ c
turbulence model [8] and the Menter and Langtry γ − Re
sition model [5], when the rotor collective pitch angle approaches (1)

to 10◦ or above. Their computational results showed a quickly re- γeff = max γ , γsep (2)
duced figure of merit (FM) at high rotor collective pitch angles.
where S 2 is a separation correction factor that is carefully cal-
Jung et al. [25] predicted a reversed flow near the S-76 rotor tip at
ibrated for the turbulence models of interest, with a suggested
a 10◦ collective pitch using a mixed structured and unstructured
value of 2.0–2.5 for the SA model [9] and 1.25–1.5 for the SST
mesh methodology. Sheng, et al. [20–22] also reported overesti- k − ω model [8]. γ is the original intermittency and γsep is the
mation of flow separations for both conventional rotors (S-76) and modified intermittency. The elevated effective intermittency γeff is
highly twisted proprotors (XV-15) at high thrust levels using the used to multiply the production term in the SA and SST turbulence
Spalart-Allmaras one-equation model [9] and the Menter SST k − ω models, which boosts the turbulent kinetic energy or eddy activ-
turbulence model [8], which led to quickly reduced FM predictions ity within the separation zone. F θ t in Eq. (1) is a blending function
compared to the experimental data. to turn off the source term outside the boundary layer so that the
To improve the prediction of rotor hover performance, Sheng separation correction is only applied in the boundary layer region.
[17] proposed a separation correction method or stall delay model Fig. 1 illustrates the difference between the original and new
(SDM) to address the overestimation of flow separations experi- separation correction methods in the prediction of trailing edge
enced in rotor hover predictions. The concept was similar to the flow separation over a NACA 4412 airfoil [34]. Using the old
separation correction used in the Langtry and Menter transition method, the separation region is only identified within the tran-
model [5], but a new criterion was introduced to identify the area sition zone close to the leading edge, due to a blending function
of separations for model correction. In the Langtry and Menter F reattach [5] that excludes most separated flows over the airfoil.
method [5], the laminar separation bubbles were suppressed by Therefore, the modified intermittency γsep is unable to affect the
increasing the intermittency factor to above one, which was con- prediction of separated flows over the trailing edge of the airfoil
fined within the transition region only. However, the new criterion (Fig. 1 (a) and (b)). On the other hand, the new separation correc-
proposed in [17] boosts the production of turbulence in both lam- tion method has accurately identified the area of separated flows,
4 C. Sheng / Aerospace Science and Technology 106 (2020) 106067

so that modified intermittency γsep can be applied to boost the 2.3. High order scheme
production of turbulence within the separation zone (Fig. 1 (c) and
(d)). However, the effective intermittency γeff remains the same as The importance of high order schemes is well received in the
the original one outside the trailing edge separation zone, because CFD community, such as in the prediction of vortex dominated
the blending function F θ t confines the area of correction to sepa- flows using RANS modeling and simulation of turbulence struc-
rated flows only. tures or acoustic waves using large eddy simulation (LES) and/or
direct numerical simulation (DNS) [16]. However, a majority of
2.2. Integration into turbulence models practical RANS CFD solvers, especially in the unstructured grid
category, still use traditional second-order schemes that have pro-
The above separation correction method is combined with the duced too much numerical dissipation and dispersion errors. In
Langtry and Menter γ − Re  θ t transition model [5] and is seamlessly
the current numerical method, the governing equations are dis-
integrated into the Spalart-Allmaras one-equation model [9] and
cretized using a vertex-centered finite-volume scheme where nu-
Menter’s SST k − ω two-equation model [8] for predicting compli-
merical fluxes are constructed and integrated over the face of each
cated viscous flows involving both transitions and separations. The
control volume. The Roe’s flux approximate scheme [30] is em-
central idea is to control the turbulence production by multiplying
ployed to calculate the numerical flux at the interface of a control
the effective intermittency parameter (γeff ) into the source term of
the respective turbulence models. The value of intermittency (γeff ) volume as
varies between zero (laminar) and one (turbulent) in the attached 1 1

+
f̂ i +1/2 = ( f (u − ) + f (u + )) −
A
(u − u − ) (8)
boundary layer, but can go above one in separated flows to boost 2 2
the turbulent kinetic energy in order to suppress the premature where u ± are the left and right values on the interface of a con-
separation inception or control the size of separated flows using trol volume. In the traditional second order scheme [29], u ± are
the current RANS modeling methods. evaluated based on a liner distribution of the solution within each
In the Spalart-Allmaras (SA) turbulence model [9], a transport control volume. A high order representation of u ± can be achieved
eddy viscosity referred to as ν̃ is solved by the following equation, based on a high degree polynomial of the solution within each
which relates to the kinematic eddy viscosity (νt ) and the molec- control volume.
ular kinematic viscosity (ν ) as follows: Recently, Zhu and Qiu [28] proposed a new fifth order WENO
    scheme (WENO-ZQ) based on an adaptive stencil formula, which
∂ ν̃ ∂ ν̃ 1 ∂  ∂ ν̃ ∂ ν̃ ∂ ν̃
+uj = P ν − Dν + ν + ν̃ + cb2 is different from the three equidistant stencils used in the classic
∂t ∂xj σ ∂xj ∂xj ∂ xi ∂ xi WENO-JS scheme [35]. It is composed of a large stencil and two
(3) small stencils, where the large stencil is used to calculate smooth
regions and the two small stencils are used to capture discontinu-
Since the SA turbulence model does not compute the turbu-
ities. The adaptive formula proposed by Zhu and Qiu [28] is written
lent kinetic energy, the free-stream turbulence intensity is used to
 θ t transition as:
substitute the local turbulence intensity in the γ − Re  
1 γ1 γ2
model. Another way to estimate the local turbulence intensity is u ± (x) = ω3 p 3 (x) − p 1 (x) − p 2 (x)
based on a decayed turbulence intensity from the far-field using a γ3 γ3 γ3
method proposed in [11]. + ω1 p 1 (x) + ω2 p 2 (x) (9)
Similarly, the Menter shear stress transport (SST) k − ω model
where γ1 , γ2 , γ3 are the positive linear weights under the condi-
[8] solves two transport equations for the turbulent kinetic energy
tion of γ1 + γ2 + γ3 = 1 (γ3 = 0) . ω1 , ω2 , ω3 are the nonlinear
(k) and turbulent dissipation frequency (ω ). The transport equa-
weights based on the linear weights and the smoothness indicators
tions for the turbulent kinetic energy and turbulent dissipation
as described in [28]. p 1 (x), p 2 (x) are the two small stencils and
frequency can be expressed as: p 3 (x) is the large stencil. u ± (x) are the final reconstructed vari-
  ables that are the high-order approximation to the solution u (x, t)
∂(ρ k) ∂(ρ U j k) ∂ ∂k
+ = P k − Dk + (μ + σk μt ) (4) in each control volume.
∂t ∂xj ∂xj ∂xj The WENO-ZQ scheme [28] can guarantee positive linear
  
∂ (ρω) ∂ ρ U j ω γ 2 ∂ ∂ω weights in both uniform and nonuniform grids, which is a unique
+ = P k − β ρω + (μ + σω μt ) advantage over the classic WENO-JS scheme [35]. However, the
∂t ∂xj υt ∂xj ∂xj
WENO-ZQ scheme was originally developed under a structured
ρσω2 ∂ k ∂ ω grid topology where adaptive stencils can be constructed straight-
+ 2(1 − F 1 ) (5)
ω ∂xj ∂xj forward based on one-dimensional high degree polynomials [28].
For the current unstructured mesh, the WENO-ZQ scheme requires
In the above two linear eddy viscosity models, the original pro-
an excessive large number of stencils to be constructed. For ex-
duction term P and destruction term D ( P ν and D ν in the SA
ample, a third order scheme needs 16 stencils to be constructed
model, and P k and D k in the SST k − ω model) are multiplied by based on a cell-centered finite-volume scheme on tetrahedron un-
the effective intermittency (γeff ) calculated from expressions (1), structured meshes [36]. This makes the original WENO-ZQ scheme
(2), which are written as: difficult and costly to be adopted in general unstructured meshes,
especially for vertex-centered finite-volume schemes.
P eff = γeff · P (6) In order to develop a compact high order WENO-ZQ scheme
 suitable for general structured and unstructured grid topologies,
D eff = min(max γeff , 0.1 , 1.0) · D (7)
Zhong and Sheng [27] proposed a new strategy to construct the
The new production and destruction terms in the SA and SST WENO-ZQ adaptive stencils. The central idea is to use a least-
k − ω models contain a mechanism to predict boundary layer tran- squares approximation to calculate the phantom points on one-
sition and also to correct separated flows, which provides a unified dimensional polynomials. A series of high order WENO schemes
method to solve complex viscous flows in aerodynamic applica- (WENO3, WENO4, WENO5, and WENO6) were proposed and vali-
tions. dated in [27], which demonstrated up to a fourth order of accuracy
 C. Sheng / Aerospace Science and Technology 106 (2020) 106067 5

Fig. 2. Dissipation and dispersion properties of various WENO schemes [27].

achieved in the validation of benchmark problems. Fig. 2 illus- 3.1. S-76 scaled rotor
trates the dissipation and dispersion properties of various WENO
schemes performed using a nonlinear approximate periodic rela- The S-76 rotor is a Sikorsky conventional helicopter rotor that
tion (ADR) analysis [37], which indicates that WENO4 and WENO6 is 1/4.71 scale to the full size [38]. This rotor has been extensively
have superior dissipation and dispersion properties among all investigated by researchers around the world in a series of work-
schemes. shops on rotor hover simulations [20–26] hosted by the American
In the present study, the new high order unstructured grid Institute of Aeronautics and Astronautics (AIAA). This four-bladed
scheme developed in [27] is adopted for rotor hover predictions, rotor possesses a −10◦ linear twist and a solidity of 0.0704. It
along with the aforementioned viscous modeling strategy. Since has a radius of 56.04 inches, a tip chord of 3.1 inches, and uses
most viscous modeling methods were calibrated and validated the SC1095 airfoil outboard and SC1094R8 airfoil inboard, tran-
sition to the SC1394R8 at the root cutout. The rotor hover tests
based on traditional second order schemes in the past [20–26], the
were performed by Balch, et al. [38] in the 80’s at Sikorsky Model
present study will assess the effect of the high order discretization
Hover Test Facility. The rotor geometry presented in this study has
on the prediction of both mean flows and turbulence structures
a blade tip with a 35◦ swept tapered shape, as shown in Fig. 3.
under the current viscous modeling framework. This may provide
Computational meshes are generated with a single blade in the
insight whether recalibrations for the existing eddy viscosity mod-
computational domain with a total of 14 million mesh points, and
els are needed when integrated with the new high order numerical
axisymmetric boundary conditions are applied for the hovering ro-
method. tor. The y+ value is set to one for the first layer of mesh off the
blade surface, and the maximum surface mesh spacing is around
3. Assessment on modeling strategy 2% of the blade tip chord. Fig. 3 also shows the surface and volume
mesh resolutions for the S-76 rotor.
Conventional rotors usually have a relatively long and flat blade
In this section, new viscous modeling strategy including tran- geometry, in which large laminar or transitional flows can develop.
sition and separation correction method is first assessed for ro- Therefore, different viscous models listed above are investigated
tor aerodynamic predictions. Two rotor cases are selected here to to assess their impact on the S-76 rotor hover prediction. The
show the effect of viscous modeling strategy: a scaled Sikorsky rotor collective pitch angle varies from 4◦ to 12◦ to cover the en-
conventional rotor S-76 [38] and a full-scale Bell XV-15 propro- tire thrust range. A time-accurate computation is conducted at a
tor [39]. Two linear eddy viscosity models including the standard given blade collective pitch angle in a rotating frame using a min-
Spalart-Allmaras turbulence model (SA) and the Menter SST k − ω imum time step equivalent to one azimuthal deg per time step.
turbulence model (SST) are employed as the baseline models here A converged solution is usually obtained in about 10 to 15 rotor
for comparative studies. The Langtry-Menter transition model [5] revolutions to reach a periodic force pattern. Solutions at different
is integrated with the SA and SST k − ω models (denoted as SA-TM rotor collective pitch angles are obtained by restarting the con-
verged solution at the previous collective pitch using the mesh
and SST-TM) to predict the transitional phenomena. The separa-
deformation method [32]. Fig. 4 (a) shows the predicted figure of
tion correction method or stall delay model (SDM) [17], which is
merit (FM) versus the blade thrust loading coefficient (CT /σ ) using
implemented into both SA and SST-based models (denoted as SA-
various viscous models. Predictions using the standard SA and SST
SDM and SST-SDM), is used to address the overprediction of rotor
models show a consistent trend of underpredicted FM over the en-
separations at high thrusts.
tire thrust range. This is because these standard turbulence models
All rotor computations are performed using the traditional sec-
cannot capture the transitional phenomena on the rotor blade sur-
ond order scheme in order to provide direct comparisons among face, resulting in larger rotor drag or power consumption in the
various models in predictions of rotor hover performance as well hover prediction. However, introducing the transition model (SA-
as viscous phenomena such as transitions and separations. The grid TM and SST-TM) into computations significantly improves the FM
convergence studies have been performed and reported for these prediction at most collective pitch angles, except for at high col-
rotors in the past [20–22], and therefore are not repeated here. lective pitch of 10◦ or above as shown in Fig. 4 (a). This indicates
The current CFD meshes are selected based on the best practices that transition modeling is helpful for improving the S-76 rotor
and guidelines developed by the author and other researchers [24], prediction at most thrust levels owning to correctly captured ro-
with a basic requirement of maximum surface mesh point spacing tor physics (laminar and transitional flows). But there is still an
lesser than 2% of the blade chord and y+ value lesser than or equal issue remaining in the rotor hover prediction at high thrusts or
to one for the first layer of the boundary layer grid points off the high collective pitch angles, as computed FM values are suddenly
viscous wall. reduced when CT /σ is approaching to 0.1. In fact, similar issues
6 C. Sheng / Aerospace Science and Technology 106 (2020) 106067

Fig. 3. S-76 rotor blade profile, surface, and volume mesh resolutions.

Fig. 4. S-76 rotor predictions using different turbulence models and CFD codes.

were also reported by other researchers in the S-76 rotor hover ration. As shown in Fig. 5 (c) and (f), the size of flow separations
simulations [20–26], whose results showed quickly reduced FM in the blade tip region is effectively reduced using the separation
predictions at high thrust, see Fig. 4 (b). These computational re- correction model (SA-SDM and SST-SDM), and the predicted FM
sults, which were obtained using the standard turbulence model or values are restored closer to the experimental data as shown in
the transition model under different structured and unstructured Fig. 4 (a).
grid RANS frameworks, have shown the same trend of underpre- The relative effects of different viscous models on the rotor
dicted FM at high thrusts or high collective pitch angles. force prediction are further analyzed in Fig. 6, which shows the
As mentioned in the Introduction, the issue of suddenly re- non-dimensional thrust (CT ) and torque (CQ ) predicted for the S-
duced FM predicted at high thrusts is due to overestimation of 76 rotor at a collective pitch angle of 12◦ . These thrust and torque
flow separations in the blade tip region using the standard viscous results are collected from the final converged solution over one ro-
models. To illustrate this issue, Fig. 5 shows the viscous stream- tor revolution (including the mean values shown in the plots). It is
line patterns predicted on the S-76 blade upper surface at a 12◦ seen that predicted thrust and torque show an oscillatory behav-
collective pitch, where all SA and SST-based viscous models are ior due to unsteady nature of the viscous flow around the rotor,
employed. The standard full turbulence model (SA and SST) can- such as trailing tip vortices and flow separations on blade sur-
not capture any transition phenomena, but the onset of transitions faces. However, the magnitude of oscillations in force and torque,
is captured near the leading edge of the blade using the transition which is associated with the size of flow separations on the rotor,
model (SA-TM and SST-TM). However, the results using either the is significantly smaller when computed using the separation cor-
standard turbulence model (SA and SST) or the transition model rection model comparing to using other models. In terms of the
(SA-TM and SST-TM) show a large separation zone in the blade mean values of CT and CQ , they show similar results obtained us-
tip region, a main reason for suddenly reduced FM predicted at ing the standard turbulence model (SA and SST) and the transition
this collective pitch angle. As discussed earlier, the fundamental model (SA-TM and SST-TM). However, the mean values of CT and
reason for the overprediction of flow separations is because these CQ obtained using the separation correction method (SA-SDM and
linear eddy viscosity models cannot correctly predict the level of SST-SDM) are 3.3 ∼ 3.8% larger in CT and 0.3 ∼ 0.8% smaller in
turbulence or shear stress in separated flows. On the other hand, CQ than the results obtained using other models. The increased CT
the separation correction method presented earlier is able to cor- and decreased CQ values predicted by SA-SDM and SST-SDM are
rect the model’s behavior by boosting the turbulence production attributed to improved FM perditions shown in Fig. 4 (a). These
in separated flows and thus reduces or eliminates the flow sepa- computations show a significant impact of the separation correc-
 C. Sheng / Aerospace Science and Technology 106 (2020) 106067 7

Fig. 5. Predicted transition and separation patterns on the S-76 rotor upper surface at θ = 12◦ , Mtip = 0.65.

Fig. 6. S-76 rotor thrust and torque convergence histories using various models at θ = 12◦ , Mtip = 0.65.

tion model on the rotor hover prediction, which is essential for taining a single blade. The mesh size is 23.6 million points in a
capturing viscous flow phenomena and obtaining accurate predic- single blade computational domain. The maximum surface point
tions of the rotor FM especially at high rotor collective pitch or spacing is set to 1% of the blade tip chord and a y+ value is set to
thrust level. one for the first layer of nodes off the blade surface. All previous
viscous models are evaluated here for the XV-15 proprotor, which
3.2. XV-15 proprotor are shown in Fig. 8 (a) and (b) at a design tip Mach number of
0.69 and a reduced tip Mach number of 0.56, respectively. Compu-
The effect of viscous modeling strategy is further investigated tational results show that predicted rotor FM values are similar for
for another type of rotor called proprotor, characterized by a rela- all viscous models when the blade collective pitch angles are less
tively short blade radius and high twist angle. Bell XV-15 proprotor than 10◦ . However, the results start to show differences among dif-
[39] is a full-scale three-bladed rotor, which has a twist angle of
ferent viscous models when the blade collective pitch angles are
−37.35◦ from the root cutoff to the tip and a blade radius of 150
greater than 10◦ (or CT /σ > 0.10). In general, the FM values pre-
inches. A geometric sketch of this rotor blade is shown in Fig. 7,
dicted by the standard turbulence model (SA and SST) are smaller
including surface and volume mesh resolutions near the tip re-
than that predicted by the transition model (SA-TM and SST-TM)
gion and around the blade. Comparing to conventional rotors such
especially for the SST-based models. This is consistent with the re-
as S-76 [38], the XV-15 proprotor presents different aerodynamic
sults reported by Wissink, et al. [40] for the same XV-15 proprotor.
characteristics and viscous phenomena. For example, Wadcock and
Yamauchi [39] performed detailed wind tunnel measurements for However, these models experience quickly reduced FM predicted at
the XV-15 rotor flow field. They observed not only transition phe- high rotor collective angles (CT /σ > 0.14), especially at the reduced
nomena on the blade surface but also revised flows in the middle tip Mach number of 0.56 shown in Fig. 8 (b). This is caused by the
and inboard section of the blade and leading-edge separation bob- overprediction of flow separations on the XV-15 rotor, the same is-
bles in the tip region [39]. Therefore, the ability to accurately cap- sue as encountered by the previous S-76 conventional rotor. Again,
ture both flow transition and separation phenomena is necessary the separation correction method (SA-SDM and SST-SDM) effec-
to investigate the aerodynamic performance of the XV-15 propro- tively reduces the size of flow separations in computations and
tor. thus restores the predicted FM values closer to the experimental
Computations for the full-scale XV-15 proprotor are performed data at both tip Mach numbers, see Fig. 8.
in a similar manner as the previous conventional S-76 rotor, where Fig. 9 shows the complex viscous flow features and streamline
a computational mesh is generated in a 120◦ sector domain con- patterns predicted on the upper blade surface at a high collective
8 C. Sheng / Aerospace Science and Technology 106 (2020) 106067

Fig. 7. XV-15 proprotor blade profile, surface, and volume mesh resolutions.

Fig. 8. XV-15 proprotor predictions at two tip Mach numbers.

pitch angle of 16.6◦ and at the reduced tip Mach number of 0.56. It the results are shown in Figs. 10–13. These plots show the skin
clearly illustrates the effect of different viscous models on the pre- friction coefficients predicted at four collective pitch angles (3◦ and
dictions of complex viscous flow phenomena at this high collective 10◦ at the tip Mach number of 0.69, and 14◦ and 16.6◦ at the
pitch angle. In particular, flow separations are predicted not only in tip Mach number of 0.56). Computational results are also validated
the blade tip region but also on the blade surface around the mid- against the wind tunnel test data [39] measured at an inboard lo-
chord, which extend from inboard radius to across a half radius. cation (r/R = 0.28) and an outboard blade location (r/R = 0.83),
Therefore, it is essential to accurately predict these flow separa- respectively.
tions and associated characteristics in order to obtain an accurate Fig. 10 shows comparisons of the skin friction coefficients pre-
assessment of the XV-15 rotor hover performance. However, both dicted and measured at the low rotor collective pitch angle of
the standard turbulence model (SA and SST) and the transition 3◦ . The leading-edge flow transition is a dominant viscous fea-
model (SA-TM and SST-TM) produce the results that show enlarged ture on the blade surface at this low collective pitch, which is
flow separations in the blade tip region as well as along the mid- well captured at the blade inboard location (r/R = 0.28) but is
chord on the blade surface, which is a major hurdle for accurate captured earlier at the outboard location (r/R = 0.83). This may
predictions of the XV-15 FM at high thrusts. On the other hand, be due to a lack of crossflow effect in the transition model. As
the separation correction model (SA-SDM and SST-SDM) shows the the rotor collective pitch angle increases to 10◦ , a reversed flow is
ability to reduce the size of flow separations in the tip region formed at the blade inboard location r/R = 0.28 from x/c = 0.7-
and on the mid-chord of the blade surface, and thus improves 1.0, where negative skin friction values are predicted as shown in
the FM prediction at this high thrust level as shown in Fig. 8. Fig. 11. Computations using the transition model (SA-TM and SST-
These XV-15 proprotor computations, along with the previous S- TM) show enlarged flow separations at the blade inboard location,
76 conventional rotor case, have all demonstrated the efficacy of as reflected by much smaller negative values predicted in compar-
the current viscous modeling strategy to address the overpredic- ison with the wind tunnel measurements [39]. However, improved
tion of flow separations in rotor aerodynamic predictions. skin friction predictions are obtained using the separation correc-
As described in Section 2, the current separation correction tion model (SA-SDM and SST-SDM), whose results are closer to the
model (SA-SDM and SST-SDM) only modifies the turbulence mod- experimental data as shown in the plot. This indicates that the sep-
el’s behavior within the separation zone, but does not affect aration correction model is able to control the flow separation and
the prediction in fully attached laminar, transitional or turbulent improve the rotor hover prediction. Furthermore, both the transi-
boundary layers. To demonstrate this feature, comparative studies tion model (SA-TM or SST-TM) and the separation correction model
are performed between the transition model (SA-TM and SST-TM) (SA-SDM or SST-SDM) produce nearly identical skin friction coef-
and the separation correction model (SA-SDM and SST-SDM), and ficients predicted outside the flow separation zone. This confirms
 C. Sheng / Aerospace Science and Technology 106 (2020) 106067 9

Fig. 9. Predicted transition and separation patterns on the XV-15 rotor upper surface, θ = 16.6◦ , M = 0.56.

Fig. 10. Comparison of XV-15 skin friction distributions at θ = 3◦ , Mtip = 0.69.

Fig. 11. Comparison of XV-15 skin friction distributions at θ = 10◦ , Mtip = 0.69.

that the proposed separation correction method does not affect the rotor power consumption leading to reduced figure of merit (FM)
original transition model in the prediction of fully attached flows. predicted in Fig. 8. The skin friction coefficients predicted by SA-
However, earlier transition onsets are also predicted at the blade SDM and SST-SDM are significantly improved and matched with
outboard (r/R = 0.83) compared with the wind tunnel measure- the measurements within the separation zone. This indicates that
ments. This indicates that improvements for the current transition the separation correction factors used in the present computations
model are necessary such as to include a crossflow effect [41] or
are appropriate (S 2 = 2.0 for SA-SDM and S 2 = 1.25 − 1.5 for SST-
to use a new criterion [7] to identify flow transitions.
SDM). At the blade outboard location (r/R = 0.83), the transition
Similar behaviors are also obtained in the XV-15 rotor predic-
onset occurs near the leading edge of the blade. No significant
tions at two high collective pitch angles of 14◦ and 16.6◦ , which
are shown in Figs. 12 and 13, respectively. A consistent trend differences are found using the transition model and the separa-
shows that enlarged flow separations are predicted by the transi- tion correction model at the blade outboard location. This again
tion model (SA-TM and SST-TM) at the blade inboard radius (r/R = confirms that the current separation correction method does not
0.28), with much smaller negative skin friction values in compari- affect the prediction of attached laminar and turbulent flows that
son with the experimental data. This is also the reason for enlarged are outside the separation zone.
10 C. Sheng / Aerospace Science and Technology 106 (2020) 106067

Fig. 12. Comparison of XV-15 skin friction distributions at θ = 14◦ , Mtip = 0.56.

Fig. 13. Comparison of XV-15 skin friction distributions at θ = 16.6◦ , Mtip = 0.56.

4. Assessment on numerical strategy of the current high order unstructured WENO scheme in the PSP
rotor prediction.
The viscous modeling strategy including the transition and sep-
aration correction models was evaluated in the previous section, 4.1. NASA PSP rotor
which has significantly improved predictions of rotor viscous flows
and aerodynamic performance. In this section, a numerical strategy The NASA PSP rotor [42] is a four-bladed scaled conventional
based on a high order unstructured WENO scheme [27] is as- rotor, which has a similar blade geometry and tip shape as the S-
sessed for rotor hover predictions. Viscous flows around rotors are 76 conventional rotor [38]. Fig. 14 shows the rotor blade geometric
complicated, which requires that numerical schemes are able to profile as well as the surface and volume mesh resolutions in the
resolve viscous flow structures in multiple length and time scales, blade tip and the surrounding region. The maximum surface mesh
such as trailing tip vortices, boundary layer transitions, inboard point spacing is selected at 1.5% of the blade tip chord, and y+
and tip flow separations, etc. Therefore, low dissipative and high value is set to one for the first boundary layer nodes off the blade
order discretization schemes are always desired in rotor predic- surface. The overall CFD mesh has 19.3 million nodes in a single-
tions in order to preserve the strength of trailing tip vortices and blade computational domain, where periodic boundary conditions
capture turbulent energy spectrum. Since the efficacy of the sepa- are applied.
ration correction model (SA-SDM and SST-SDM) has been validated Computations are performed in a fixed rotating frame where
in the previous section, it will be used as the standard option in the CFD mesh is stationary, initially using a local time step to help
the assessment of the numerical strategy in this section. The fo- settle down the flow field and finally switching to a minimum time
cus here is to assess the impact of the new fourth order WENO step to obtain time-accurate unsteady solutions. Three Newton’s it-
unstructured grid scheme [27] in comparison with the traditional erations and six Gauss-Seidel relaxations are typically used at each
second order scheme [29] to predict the rotor hover performance time step. Numerical tests indicate that the convergence behavior
and capture viscous flow physics. The rotor case considered here of the new fourth order unstructured WENO scheme is similar to
is a NASA pressure sensitive paint (PSP) scaled conventional rotor the traditional second order scheme, where a converged solution is
[42], which is similar to the previous S-76 rotor [38]. Preliminary obtained in about 4500 time steps or 12 rotor revolutions. Fig. 15
experimental data for the installed PSP rotor have been obtained at shows the convergence histories of the nondimensional thrust co-
NASA Langley Research Center, including the hover FM measured efficient (CT ) and torque coefficient (CQ ) for the PSP rotor at five
at free transition and forced transition conditions [42]. These mea- rotor collective pitch angles obtained with the current fourth or-
sured transition onset locations will be used to assess the accuracy der unstructured scheme.
 C. Sheng / Aerospace Science and Technology 106 (2020) 106067 11

Fig. 14. PSP rotor blade profile, surface, and volume mesh resolutions.

Fig. 15. NASA PSP rotor thrust and torque convergence histories, fourth order scheme.

Fig. 16. Comparison of PSP rotor hover predictions using second and fourth order schemes.

The comparison of the fourth order scheme with the traditional tal data at low thrust levels, but are slightly lower at high thrust
second order scheme is shown in Fig. 16 for the prediction of an levels such as CT /σ ≥ 0.08. This may be due to a lack of fuse-
isolated PSP rotor in hover. Computed rotor figure of merit (FM) lage for the isolated rotor in the computational model, in contrast
and power loading coefficient (CP /σ ) verse thrust loading coeffi- to an installed rotor in the experimental test [42]. Comparative
cient (CT /σ ) are obtained at five rotor collective pitch angles of studies by Jain [43] showed that the effect of the fuselage may
4◦ , 6◦ , 8◦ , 10◦ , and 12◦ . All computations are performed using the increase the FM prediction by 2 counts (2.5-4%) compared with an
separation correction model (SA-SDM and SST-SDM) as described isolated PSP rotor prediction. In terms of the effect of the high or-
earlier. In general, predicted FM values using either the second der scheme, the fourth order solution in general produces 2.35 ∼
order or the fourth order scheme match well with the experimen- 3.69% higher FM values compared with the second order solution
12 C. Sheng / Aerospace Science and Technology 106 (2020) 106067

Fig. 17. Comparison of vorticity contours predicted by second and fourth order schemes, SST-SDM, θ = 12◦ .

when the SST-based model (SST-SDM) is used. However, no signif- high collective pitch angles of 10◦ and 12◦ , which is due to a lack
icant differences are found between the low and high order solu- of the fuselage effect in the current isolated rotor computation.
tions when the SA-based model (SA-SDM) is used. This indicates
that the current high order scheme has a different effect on the 5. Conclusions
SA and SST-based turbulence models, and the SST-based viscous
model seems to be more sensitive to the change of mean flow res-
In this paper, strategies are presented to improve the predic-
olutions obtained using the high order scheme. This can be further
tion of complex viscous flows from both viscous modeling and
visualized in Fig. 17 for the rotor tip vorticity contours computed
numerical scheme aspects under a RANS modeling framework. A
using the SST-SDM model at a 12◦ high collective pitch, where
new separation correction model is proposed to address the over-
significantly refined mean flow structures are obtained using the
prediction of flow separations faced by RANS modeling, which
high order scheme compared with the low order scheme. Not only
has been the main hurdle for obtaining an accurate prediction of
large-scale tip vortices are captured with a greater strength and a
aerodynamic performance for hovering rotors. The modified vis-
longer wake age, but also small-scale wake structures are captured cous modeling has shown a significant improvement on the rotor
around the rotor with great details using the high order scheme. hover predictions including improved rotor FM and skin friction
The improved mean flow resolutions also affect the prediction of profiles on blade surfaces at high thrusts. In addition, a new high
viscous flow structures as shown next. order unstructured WENO scheme is adopted to improve the mean
Like the conventional S-76 rotor, the boundary layer transition flow resolution such as trailing tip vortices as well as the viscous
is also a predominant viscous flow feature on the PSP rotor in flow structure such as transition patterns on the blade surface.
hover. To investigate the effect of the high order scheme on the Future works should be focused on enhancements of the transi-
prediction of this important viscous phenomenon, computed skin tion model to include a crossflow effect or to introduce a new
friction coefficients on the upper and lower blade surfaces using criterion for transition identification. It is also worth further inves-
the second and fourth order schemes are shown in Figs. 18 (a) and tigating the mutual influence between the mean flow resolution
(b), respectively. These results are obtained at five rotor collective and the viscous turbulent structure under the framework of high
pitch angles using the SST-SDM model, which illustrate different order schemes. It is believed that a combination of the strategies
impacts of the low and high order schemes on the prediction of presented here should provide a viable solution for improved pre-
boundary layer transition features on the PSP rotor. The color maps dictions of rotor hover performance, with enhanced understanding
in the figures show the magnitude of the skin friction coefficients of complex viscous flow physics in realistic rotary-wing applica-
predicted on the blade surface, where the dark blue color denotes tions.
laminar flow regions and the light green color denotes the turbu-
lent flow regions. The red dots in these figures show the transition
Declaration of competing interest
onset locations measured in the wind tunnel experiment [42]. An
interesting phenomenon in these figures is the irregular transition
patterns formed on the lower blade surface at collective pitch an- The author certifies that he has NO affiliations with or involve-
gles from 4◦ to 8◦ , which are captured in the computations and ment in any organization or entity with any financial interest (such
as honoraria; educational grants; participation in speakers’ bu-
were also measured in the wind tunnel experimental tests [42].
reaus; membership, employment, consultancies, stock ownership,
These irregular transition patterns are triggered by the leading-
or other equity interest; and expert testimony or patent-licensing
edge separation bubbles, or called separation-induced transition
arrangements), or non-financial interest (such as personal or pro-
[17], which is typical on conventional rotors such as S-76. The
fessional relationships, affiliations, knowledge or beliefs) in the
transition onset locations predicted using the second order scheme
subject matter or materials discussed in this manuscript.
show some discrepancies compared with the experimental data at
4◦ and 6◦ collective pitch angles, but improved predictions are ob-
tained with the fourth order scheme especially at the 6◦ collective Acknowledgements
pitch. This shows the benefit of improved mean flow resolutions
on the prediction of viscous flow structures using the high order Graphic images and plots shown in this work were created us-
scheme. A slight discrepancy is found between the computed and ing FieldView, which is provided by Intelligent Light under its Uni-
the measured transition onsets on the lower blade surface at two versity Partners Program. This support is gratefully acknowledged.
 C. Sheng / Aerospace Science and Technology 106 (2020) 106067 13

Fig. 18. (a) Comparisons of computed and measured skin friction distributions, SST-SDM, second order. (For interpretation of the colors in the figure(s), the reader is referred
to the web version of this article.)

Fig. 18. (b) Comparisons of computed and measured skin friction distributions, SST-SDM, fourth order. (For interpretation of the colors in the figure(s), the reader is referred
to the web version of this article.)

References [11] J. Wang, C. Sheng, A comparison of a local correlation-based transition model

coupled with SA and SST turbulence model. AIAA-2015-0587, in: 53rd AIAA
[1] P.R. Spalart, Strategies for turbulence modeling and simulation, Int. J. Heat Fluid Aerospace Sciences Meeting, AIAA SciTech 2015 Forum, 5–9 January 2015,
Flow 21 (2000) 252–263. Kissimmee, Florida.
[2] F.R. Menter, Turbulence modeling for engineering flows, Technical Paper, ANSYS [12] M. Mani, D.A. Babcock, C.M. Winkler, P.R. Spalart, Predictions of a supersonic
Inc., 2011. turbulent flow in a square duct, AIAA Paper 2013-0860, January 2013.
[3] A.M.O. Smith, N. Gamberoni, Transition, pressure gradient and stability theory, [13] A. Hellsten, New advanced k-omega turbulence model for high-lift aerodynam-
Douglas Aircraft Co. Rept. ES26388, El Segundo, California, 1956. ics, AIAA J. 43 (9) (2005) 1857–1869.
[4] D.K. Walters, J.H. Leylek, A new model for boundary-layer transition using a [14] S. Wallin, A.V. Johansson, An explicit algebraic Reynolds stress model for in-
single-point RANS approach, J. Turbomach. 126 (1) (2004) 193–202, https:// compressible and compressible turbulent flows, J. Fluid Mech. 403 (2000)
doi.org/10.1115/1.1622709. 89–132.
[5] R.B. Langtry, F.R. Menter, Correlation-based transition modeling for unstruc- [15] F.R. Menter, A.V. Garbaruk, Y. Egorov, Explicit algebraic Reynolds stress models
tured parallelized computational fluid dynamics codes, AIAA J. 47 (12) (Dec. for anisotropic wall-bounded flows, in: Invited Lecture: EUCASS – 3rd European
2010) 2894–2906. Conference for Aero-Space Sciences, July 6–9th 2009, Versailles.
[6] S.K. Robinson, Coherent motion in the turbulent boundary layer, Annu. Rev. [16] P. Sagaut, S. Deck, M. Terracol, Multiscale and Multiresolution Approaches in
Fluid Mech. 1991 (23) (1991) 601–639. Turbulence, 2nd edition, Imperial College Press, 2013.
[7] C. Liu, Y. Gao, X. Dong, et al., Third generation of vortex identification methods: [17] C. Sheng, Role of transition modeling in rotor hover predictions, J. Aircr. 55 (1)
omega and liutex/rortex based systems, J. Hydrodyn. 32 (2) (2019), https:// (2018) 23–37.
doi.org/10.1007/s42241-019-0022-4. [18] B. Aupoix, D. Arnal, H. Bezard, et al., Transition and turbulence modeling, J.
[8] F.R. Menter, Two-equation eddy-viscosity turbulence models for engineering Aerosp. Lab. (2) (March 2011).
applications, AIAA J. 32 (8) (August 1994) 1598–1605. [19] C.L. Rumsey, Turbulence modeling verification and validation, AIAA-2014-0201,
[9] P.R. Spalart, S.R. Allmaras, A one-equation turbulence model for aerodynamic in: AIAA 2014 Aerospace Sciences Meeting, January 2014.
flows, Rech. Aérosp. 1 (1994) 5–21. [20] C. Sheng, J. Wang, Q. Zhao, Improved rotor hover predictions using advanced
[10] S. Medida, J.D. Baeder, Application of the correlation-based transition model turbulence modeling, J. Aircr. 53 (5) (2016) 1549–1560.
to the Spalart-Allmaras turbulence model, in: 20th AIAA Computational Fluid [21] C. Sheng, Predictions of JVX rotor performance in hover and airplane mode us-
Dynamics Conference, Honolulu, HI, June 2011. ing high-fidelity unstructured grid CFD solver, in: Proceedings of the American
14 C. Sheng / Aerospace Science and Technology 106 (2020) 106067

Helicopter Society 70th Annual Forum, May 20–22, 2014, Montreal, Quebec, [33] K. Yamamoto, K. Tanaka, M. Murayama, Effect of a nonlinear constitutive re-
Canada. lation for turbulence modeling on predicting flow separation at wing-body
[22] C. Sheng, Q. Zhao, M. Hill, Computational investigation of a full-scale proprotor juncture of transport commercial aircraft, AIAA 2012-2895, in: 30th AIAA Ap-
hover performance and flow transition, J. Aircr. 55 (1) (2018) 122–132. plied Aerodynamics Conference, 25–28 June 2012, New Orleans, Louisiana.
[23] P. Gardarein, A. Le Pape, Numerical simulation of hovering S-76 helicopter rotor [34] Turbulence Modeling Resource, NASA Langley Research Center, https://
including far-field analysis, AIAA-2016-0034, in: 54th AIAA Aerospace Sciences turbmodels.larc.nasa.gov.
Meeting, AIAA SciTech 2016 Forum, 4–8 January 2016, San Diego, California. [35] G. Jiang, C.-W. Shu, Efficient implementation of weighted ENO schemes, J. Com-
[24] P.R. Narducci, Overflow simulation of rotor in hover: the Boeing company, put. Phys. 126 (1996) 202–228.
AIAA-2014-0208, in: 52nd AIAA Aerospace Sciences Meeting, AIAA SciTech [36] J. Zhu, J. Qiu, A new third order finite volume weighted essentially non-
2014 Forum, 13–17 January 2014, National Harbor, Maryland. oscillatory scheme on tetrahedral meshes, J. Comput. Phys. 349 (2017)
[25] M.K. Jung, J.Y. Hwang, O.J. Kwon, Assessment of rotor aerodynamic perfor- 220–232.
mances in hover using an unstructured mixed mesh method, AIAA-2014-0042, [37] S. Pirozzoli, On the spectral properties of shock capturing schemes, J. Comput.
in: 52nd AIAA Aerospace Sciences Meeting, AIAA SciTech 2014 Forum, 13–17 Phys. 219 (2) (2006) 489–497.
January 2014, National Harbor, Maryland. [38] D.T. Balch, J. Lombardi, Experimental Study of Main Rotor Tip Geometry and
[26] B.Y. Min, B. Wake, Parametric validation study for a hovering rotor using UT- Tail Rotor Interactions in Hover. Vol. I – Text and Figures, NASA CR 177336,
GENCAS, AIAA-2016-0301, in: 54th AIAA Aerospace Sciences Meeting, AIAA Vol. 1, February 1985.
SciTech, 4–8 January 2016, San Diego, California. [39] A.J. Wadcock, G.K. Yamauchi, D.M. Driver, Skin friction measurements on a hov-
[27] D. Zhong, C. Sheng, A new method towards high-order WENO schemes on ering full-scale tilt rotor, J. Am. Helicopter Soc. 44 (4) (1999) 312–319.
structured and unstructured grids, Comput. Fluids 200 (2020) 104453. [40] A. Wissink, W. Staruk, S. Tran, et al., Overview of new capabilities in Helios
[28] J. Zhu, J. Qiu, A new fifth order finite difference WENO scheme for solving version 9.0, AIAA-2019-0839, in: AIAA SciTech 2019 Forum, 7–11 January 2019,
hyperbolic conservation laws, J. Comput. Phys. 318 (2016) 110–121. San Diego, CA.
[29] C. Sheng, A preconditioned method for rotating flows at arbitrary Mach num- [41] S. Medida, J. Baeder, A new crossflow transition onset criterion for RANS tur-
ber, Model. Simul. Eng. (ISSN 1687-5591) 2011 (January 2011) 537464, https:// bulence models, AIAA-2013-3081, in: 21st, AIAA Computational Fluid Dynamics
doi.org/10.1155/2011/537464. Conference, June 24–27, 2013, San Diego, CA.
[30] P.L. Roe, Approximate Riemann solvers: parameter vectors and difference [42] A.D. Overmeyer, P.B. Martin, Measured boundary layer transition and ro-
schemes, J. Comput. Phys. 43 (1981) 357–372. tor hover performance at model scale, AIAA Paper 2017-1872, AIAA SciTech
[31] D. Hyams, K. Sreenivas, C. Sheng, W. Briley, D. Marcum, D. Whitfield, An in- 2017 Forum, in: 55th AIAA Aerospace Sciences Meeting, 9–13 January 2017,
vestigation of parallel implicit solution algorithms for incompressible flows on Grapevine, TX.
unstructured topologies, AIAA-2000-0271, in: The 38th AIAA Aerospace Sci- [43] R. Jian, CFD performance and turbulence transition predictions on an installed
ences Meeting, 10–13 January 2000, Reno, Nevada. model-scale rotor in hover, AIAA-2017-1871, in: AIAA SciTech 2017 Forum,
[32] C. Sheng, C. Allen, Efficient mesh deformation using radial basis functions on 9–13 January 2017, Grapevine, TX.
unstructured meshes, AIAA J. 51 (3) (2013) 707–720, https://doi.org/10.2514/1.
J052126.