International Journal of Heat and Fluid Flow 61 (2016) 650–664

Contents lists available at ScienceDirect

International Journal of Heat and Fluid Flow

journal homepage: www.elsevier.com/locate/ijhff

Conjugate heat transfer of a rib-roughened internal turbine blade

cooling channel using large eddy simulation
Sebastian Scholl a,∗, Tom Verstraete a, Florent Duchaine b, Laurent Gicquel b
a
Von Karman Institute for Fluid Dynamics, Turbomachinery Department, Chaussee de Waterloo 44, 1640 Rhode St Genese, Belgium
b
CERFACS, 42 Avenue Gaspard Coriolis, 31057 Toulouse Cedex 01, France

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

Article history: This contribution focuses on the conjugate heat transfer computation of a rib-roughened internal tur-
Received 11 September 2015 bine blade cooling channel with a blockage ratio of 0.3 and a Reynolds number of 40,0 0 0. The work
Revised 4 July 2016
considers the coupling between two numerical tools: a Large Eddy Simulation flow solver and a solid
Accepted 16 July 2016
conduction solver. While interested in the thermal steady states of such problems, both solvers provide
Available online 29 July 2016
time dependent de-synchronized solutions. A novel weak coupling strategy, the hFFB method, is applied
Keywords: to solve the challenge of varying fluid and solid time scales. This method first computes the fluid domain
Conjugate heat transfer until it reaches a statistically steady state, from which the mean temporal solution is obtained. Then, it
Large eddy simulations solves the solid domain with the mean boundary conditions imposed from the fluid leading to an up-
Turbine blade cooling dated fluid boundary condition. This process is repeated until convergence of the coupled problem. The
Internal cooling channels coupled convective-conductive solution shows that conjugate heat transfer is important for rib-roughened
CFD
internal cooling ducts and has a large impact on the heat transfer at the solid-fluid interface. The results
Stability
were validated against experimental data provided by the von Karman Institute for Fluid Dynamics.
© 2016 Elsevier Inc. All rights reserved.

1. Introduction rib-roughened channels were done, e.g., by Rau et al. (1998) and
Casarsa and Arts (2005). Studies on the heat transfer were done
In anticipation of claims for environmental awareness and the by Cakan (20 0 0); Coletti et al. (2011) and Cukurel and Arts (2013).
rising use of energy, the turbomachinery research targets im- Cooling channels were numerically investigated with RANS by,
provements of the gas turbine efficiency, driving new gas turbine e.g., Ooi et al. (2002) and more recently, e.g., by Keshmiri (2012).
designs to higher pressure ratios and increasing turbine inlet Recent research targets and investigates further possible improve-
temperatures. Since the life of a turbine blade is reported to be ments on the RANS modeling, e.g., with non-linear eddy-viscosity
reduced by half with an increased temperature of 15–30 K (Han modelsRaisee et al. (2009). With LES, first studies were done by
et al., 20 0 0), the prediction of local heat transfer coefficients and Ciofalo and Collins (1992) and Yang and Ferziger (1993). To name
temperatures is more than ever a crucial step towards the design a few, more recent studies were done by Cui et al. (2003); Fransen
of reliable and efficient turbine blades. et al. (2012); Leonardi et al. (2004); Lohasz et al. (2006); Miyake
Internal cooling channels have largely contributed to the in- et al. (2002) and Tyacke and Tucker (2015).
crease of the turbine inlet temperature over the past decades. Despite the vast amount of studies on roughened cooling chan-
Rough surfaces have been introduced, investigated and developed nels, except for few experimental (Coletti et al. (2012) and Cukurel
to enhance the heat transfer by turbulating the flow, as reviewed and Arts (2013)), or two-dimensional, steady numerical investiga-
by Han et al. (20 0 0); Han (2010); Iacovides and Launder (1995); tions (Iaccarino et al. (2002)), the prevailing research neglects the
2007); Weigand et al. (2001) and Ligrani (2013). To date, ribs have influence of the thermal boundary condition on the heat transfer,
one of the highest potential among roughness elements to tur- even though the impact of the wall temperature distribution can
bulate the flow and increase the heat transfer, therefore, a large be significant, as indicated by the experiment of Cukurel and Arts
effort has been put to investigate the flow and the heat transfer (2013).
of these channels. Experimental investigations of the flow field in On the numerical side, current numerical methods call for new
numerical strategies. Two reasons explain the current status: first,
the complex geometries with high asperities challenge the numeri-
∗
Corresponding author. cal models. Second, the generally neglected conjugate heat transfer
E-mail address: [email protected] (S. Scholl). needs to be taken into account for reliable thermal predictions.

http://dx.doi.org/10.1016/j.ijheatfluidflow.2016.07.009
0142-727X/© 2016 Elsevier Inc. All rights reserved.
 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664 651

Table 1
Channel and flow characteristics Cukurel
Nomenclature
(2012).

Bi Biot number [ms−1 ] Parameter Abbr. Value

c speed of sound [ms−1 ]
Reynolds number ReDh 40,0 0 0
Dh Hydraulic diameter [m] number of ribs 6
D Thermal diffusivity [m2 s−1 ] length Lf 1597.5 mm
EF enhancement factor hydraulic diameter Dh 75 mm
rib height H 22.5 mm
F flux tensor
pitch to rib ratio P/H 10
f friction factor blockage ratio H/D h 0.3
H rib height [m]
h heat transfer coefficient [W m−2 K]
L length of channel model [m]
Ma Mach number existing challenges, only few computations of conjugate heat trans-
Nu Nusselt number fer (CHT) with LES have been done (e.g., Duchaine et al. (2009,
P pitch length in the ribbed channel [m] 2013)) and research is still needed for an efficient LES based CHT
p pressure [Nm−2 ] Strategy.
Pr Prandtl number The present contribution aims at improving the current state of
q heat flux [W m−2 ] the art thermal predictions for roughened cooling channels and in-
R gas constant [J 1 K −1 mol −1 ] vestigates the influence of the thermal boundary condition on the
Re Reynolds number heat transfer inside a rib-toughened cooling channel by compar-
T temperature [K] ing LES with iso-flux boundary conditions to coupled convective-
t time [s] conductive computations.
u, v, w velocity components [ms−1 ] This paper is laid out as follows: the next section describes the
W vector of primary variables used solvers for fluid and solid domain as well as the used cou-
w quantity at the wall pling approach for the CHT problem. The results have been com-
y+, x+, z+ non-dimensional measures of the first cell to the pared with experimental data by the Von Karman Institute for
wall Fluid Dynamics subsequently for the mean flow field, turbulence
x grid cell size [m] statistics and the heat transfer.
t time step [s]
ν kinematic viscosity [m2 s−1 ] 2. Problem definition
μ dynamic viscosity [Pa1 s1 ]
ρ density [kg1 m−3 ] The studied geometry models the experimental facility of Cakan
λ thermal conductivity [W m−1 K −1 ] (20 0 0); Cukurel (2012) and Cukurel and Arts (2013), which repre-
τw wall shear stress [W m−1 K −1 ] sented an internal rib-roughened aircraft gas turbine blade cooling
CFL Courant-Friedrich-Levy number passage with a squared cross-section, a vertical alignment of the
LES Large Eddy Simulation ribs to the flow streamwise direction, a pitch to rib ratio of 10 and
NSCBC Navier-Stokes Characteristic Boundary Condition a blockage ratio of 0.3 (Fig. 1). The cooling passage was scaled up
PIV Particle Image Velocimetry by a factor of 15 with respect to real engine conditions (Cukurel,
SGS Sub-grid Scale 2012), to warrant a high measurement resolution (Cakan, 20 0 0).
RANS Reynolds-averaged Navier-Stokes The experimental test section with its six ribs was fully repre-
∞ free stream quantity sented by the configuration for the numerical simulations ( Fig. 2).
i iteration step in coupling cycle Only a reduced length of the inlet and outlet sections were sim-
f fluid ulated with appropriate boundary conditions. The cross section
s solid dimensions were 75 × 75 mm and the test section length was
1260 mm (Table 1). Although the rib of the cooling channel re-
peats itself periodically, the total number of six ribs was kept
Conjugate heat transfer describes the thermal interaction be- similar to the experimental model to avoid possible arbitrary ef-
tween the fluid and the solid domain. If the temperature distribu- fects of periodic boundary conditions that occur as Fransen et al.
tion would be known at the interface a priori, the problem would (2012) have shown. Indeed, the flow field needs more than one
reduce to two single domain uncoupled problems. However, the rib distance to establish a fully developed pattern independent of
temperature and heat flux distributions are unknown a priori and the inflow and previous ribs. If assuming that the largest scales in-
are part of the coupled solution of the fluid and solid energy equa- side the channel are of the order of the channel diameter, which
tions. The temperature and heat flux distributions can only be corresponds to around one third of the rib to rib distance, the ac-
found by solving both domains in a coupled manner, which is the tual axial distance of the ribbed pipe is not enough to establish an
conjugate heat transfer problem as first described and named by uncorrelated flow field as it would be the case if only one pitch
(Perelmann, 1961). with periodic boundary conditions would be used for the simula-
The coupling of fluid and solid solvers is a numerically demand- tion domain. Furthermore, the use of periodic boundary conditions
ing task, as stability problems can occur. Furthermore, the time would include major complications for CHT computations. Besides
scales in the fluid and solid domain may vary by orders of mag- technical difficulties for the implementation of periodic boundary
nitude. For ribbed ducts, LES have demonstrated a high accuracy conditions for CHT computations, it would not be possible to in-
for the turbulent flow field and heat transfer predictions. clude temperature dependent material and flow properties. More-
However, LES are CPU consuming and amplify the problem of over, the temperature profiles inside the domain are not know a
varying time scales in solid and fluid domains, even though recent priori and it would be a questionable assumption that the tem-
research also targets a reduction in cost for LES using a zonal wall perature profiles at the inlet and outlet periodic boundaries are
treatment for ribbed ducts (e.g., Patil and Tafti (2013). Due to the identical, both for solid and fluid domain. This assumption would
652 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664

Fig. 1. Sketch of a modern high pressure turbine blade and the reduced model of an internal cooling channel.

Fig. 2. Sketch of the computational configuration.

include a linear increase of the temperature. At last additional sta- with the thermal conductivity
bility problems of the whole system could be provoked.
μ̄(T̄ )c p (T̄ )
λ̄ = , (3)
Pr
3. Methods where cp is the specific heat at constant pressure, Pr the Prandtl
number and μ̄ the dynamic viscosity. The solver uses tabulated
3.1. Flow solver values for the temperature dependent specific heat at constant
pressure. The fluid viscosity follows a temperature dependent
The fluid flow was computed by the CERFACS in-house LES power law. The unknowns appearing in the LES equations, i.e. the
solver AVBP (Schönfeld and Poinsot, 1999) (Mendez and Nicoud, subgrid-scale (SGS) quantities, need to be modeled. The Boussinesq
2008). Applying Favre filtering to the unsteady, compressible eddy viscosity assumption is applied for which a subgrid-scale tur-
Navier–Stokes equations (Sagaut, 20 0 0), leads to the following LES bulent viscosity ν t is modeled via the Wall Adapting Linear Eddy
equations: (WALE) model (Nicoud and Ducros, 1999). Using the SGS turbu-
lent viscosity, the SGS turbulent conductivity is modeled as λt =
∂ W̄ ρ̄νt c p /Prt , with the turbulent Prandtl number fixed at Prt = 0.7.
+ ∇ · F̄ = 0, (1)
∂t The equations are discretized by centered spatial schemes and ex-
plicit time schemes with a finite-element based two step Taylor–
with the conservative variables W̄ = (ρ̄ , ρ̄ u˜, ρ̄ v˜ , ρ̄ w
˜ , ρ̄ E˜ ) the flux
Galerkin (TTGC) scheme for the convection in a cell-vertex formu-
tensor F̄ , the density ρ̄ , the velocity components (u˜, v˜ , w ˜ ), and the
lation (Colin and Rudgyard, 20 0 0). This explicit scheme is of third
total Energy E˜ . The fluid follows the Favre filtered equation of state
order accuracy in space (on hybrid meshes) and time, providing
p̄ = ρ̄ r T˜ , with the gas constant r. The conductive heat flux is in-
low dissipation and dispersion, especially suitable for LES. The dif-
cluded in the flux tensor and given by Fourier’s law of conduction:
fusion term is discretized with a vertex-centered operator closely
following the Galerkin method and is of second order, having a
∂ T˜ 2  stencil (Colin et al., 2003) and (Donea and Huerta, 2003).
q̄i = −λ̄ , (2)
∂ xi The time step is controlled via the Courant–Friedrich–Lewy (CFL)
 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664 653

condition for compressible flows: CF L = (U+cx)t  0.7, with the For instance, modeling the solid domain as a semi-infinite one-
grid cell size x and the time step t, U and c being the local dimensional solid, with the lowest occurring frequency of temper-
flow and sound speeds. The Navier-Stokes Characteristic Boundary ature fluctuations of 10 Hz, assuming the penetration depth Xt at
Conditions (NSCBC) of Poinsot and Lele (1992), are applied at the which
√ 1% of the temperature disturbance is felt, as Xt = 3.643 ·
inlet and the outlet. Dt , with the thermal diffusivity D, would lead to a penetration
depth of 0.2 mm.
3.2. Solid solver Therefore an exchange of mean fluxes and temperatures is suffi-
cient and is conducted in terms of temporally integrated LES over a
The conduction solver AVTP uses a Galerkin diffusion scheme, duration of around 10 flow-throughs. The conduction solver, tech-
second order accurate in space with an explicit time advancement nically being an instationary solver is computed such that a high
of first order accuracy to solve the energy conservation: amount of iteration grants a full convergence after each coupling
iteration. This weakly coupled treatment of the problem is known
∂ T ( x, t ) ∂ qi to be prone for stability problems which are considered in the fol-
ρs c s =− , (4)
∂t ∂ xi lowing section.
with the temperature in the solid T, the solid density ρ s , the 3.3.2. Stability consideration
heat capacity cs and the heat flux q, given by Fourier’s law of Weak CHT coupling strategies are known to become unstable
conduction: under certain conditions. Verstraete (2008) and Errera and Baqu
∂T (2013) describe the influence of the kind of exchanged bound-
qi = −λs . (5)
∂ xi ary condition, which is important for convergence and stability.
Verstraete (2008) distinguishes four different coupling procedures,
3.3. Coupling method for each of which the stability depends on the Biot number:
hLs,d
Bi = . (6)
Conjugate heat transfer methods for numerical simulations can λs
be categorized into two main approaches which are the monolithic
The physical meaning of this dimensionless number can be inter-
approach (often referred to as the conjugate method or strong cou-
preted as the ratio of conductive resistance in the solid Ls, d /λs ,
pling) using one solver for both domains and the partitioned ap-
with the solid conduction coefficient λs and the convective resis-
proach (often referred to as the coupled method or weak coupling)
tance in the fluid boundary layer 1/h with the heat transfer co-
using different solvers for each domain.
efficient h. The heat transfer coefficient is a frequently used en-
Monolithic coupling strategies are inflexible and have the dis-
gineering parameter during the thermal design. It is derived from
advantage that for a disparity of time scales as in the present case
Newton’s law of convection:
the convergence is slow, leading to unjustifiable high computa-
tional cost, especially when using LES. q = h(Tw − T∞ ), (7)
with the wall temperature Tw , the free-stream temperature T∞ , the
3.3.1. Time scale disparity heat flux q and the heat transfer coefficient h. Four different types
Considering the current case of the model channel, operated at of weak coupling exist and will be explained in the following.
typical conditions with a velocity of 80m/s and a channel dimen-
The flux forward temperature back method (FFTB)
sion with a length of around 0.1 m, a temperature disturbance at
The FFTB algorithm is frequently used method for weakly cou-
the inlet will arrive at the outlet in around 0.00125 s. In com-
pled CHT computations (Verdicchio et al., 2001) (Illingworth et al.,
parison, the response time of the metal surface temperature typ-
2005). The FFTB method is named from the perspective of the fluid
ically lies in the order of seconds, hence three orders of magni-
domain. It imposes a wall temperature distribution to the fluid
tude higher than the fluid. Expressed in terms of characteristic
solver. After the fluid calculation the resulting heat flux distribu-
time scales and assuming the same characteristic length of the
tion is imposed to the solid conduction solver, which returns an
solid domain, the characteristic diffusion time scale in the solid is
ρ f C p L2solid updated temperature in turn imposed to the fluid domain. The ex-
τd = kf
with a specific heat for steel of 460 J/kgK, a density change of boundary conditions is repeated until continuous tem-
of 80 0 0 kg/m3 and a thermal conductivity of 15 W/m2 K, leading peratures and heat fluxes between both domains are reached. The
to a ratio of characteristic convective fluid to characteristic diffu- FFTB procedure is stable for Bi < 1 (Verstraete, 2008). With a re-
sive solid of ττs  1940, more than three orders of magnitude dif- laxation factor β , simulations for Bi > 1 can be stabilized with the
f
ference. drawback of slower convergence. The relaxation is used for the ex-
The large discrepancy of time scales becomes more severe in change of the boundary conditions as follows:
f luid,i−1
case of unsteady simulations. As pointed out by L.He and Oldfield f luid,i
Twall = βT · Twall + (1 − βT ) · Twall
solid,i−1
, (8)
(2011), for a statistically stationary oscillatory flow, a resolved har-
and:
monic needs at least 20 time steps. In the current case of the f luid,i−1
LES more than three orders of harmonics are resolved, leading to qsolid,i
wall
= βq · qsolid,i
wall
−1
+ (1 − βq ) · qwall , (9)
20,0 0 0 time steps, if one large harmonic is to be adequately cap- with the current coupling iteration step i, fluxes q and tempera-
β
tured, this would lead to 40,0 0 0,0 0 0 time steps if the time scales tures T. The stability criterion is then: Bi < 1+
1−β
.
in solid and fluid domain were kept consistent.
Hence, for the current case and most practical applications, a The temperature forward flux back method (TFFB)
monolithic coupling approach disqualifies for an LES based cou- The TFFB algorithm imposes a heat flux distribution as bound-
pling strategy and a partitioned, and more efficient weak coupling ary condition to the fluid solver (He et al., 1995) (Heidmann et al.,
strategy needs to be applied. 20 0 0). The resulting temperature is imposed to the solid solver,
Albeit dealing with unsteady flow simulations, the main tar- which returns a new heat flux distribution to the fluid domain. The
gets are only mean statistically stationary quantities, in particular TFFB method is stable if Bi > 1. With a relaxation factor β simu-
the temperature fields. The large discrepancy in time scales, thus, lations for Bi < 1 are possible; again, with the drawback of slower
−β
also leads to a low penetration depth of temperature fluctuations. convergence. The stability criterion is then: Bi > 11+ β.
654 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664

cient is used: hˆ > 2h , with the imposed heat transfer coefficient hˆ

and the actual heat transfer coefficient h.

The heat transfer coefficient forward flux back method (hFFB)

Verstraete (2008) proposed a fourth method, which imposes a
heat flux distribution to the fluid domain returning mixed type
boundary conditions to the solid solver (Fig. 4). Again, for the
mixed type boundary condition, Newton’s law of convective heat
transfer is used:

qwall = hˆ · (Twall − T f luid ). (11)

This method is stable for Bi < 1, but also for Bi >= 1, if an ap-
propriate virtual heat transfer coefficient for the method is used:
hˆ < 2h, with the imposed virtual heat transfer coefficient hˆ and the
actual heat transfer coefficient h.
Even tough the heat flux qwall and respectively the temperature
Fig. 3. Sketch of stable regions for the standard coupling methods for weakly cou- Twall are never directly exchanged for the hFTB and respectively
pled conjugate heat transfer computations. the hFFB method, continuity of the heat flux qwall and respectively
the temperature Twall is assured at the interface if convergence is
reached (Amano, 1994).
The heat transfer coefficient forward temperature back method (hFTB)
The hFTB method (Amano, 1994) (Heselhaus, 1998), starts with 3.3.3. Choice of coupling algorithm
an initial temperature distribution as boundary condition for the The Biot number inside the channels varies from values be-
fluid computation. It uses Newton’s law of convective heat trans- low unity to values larger than unity (Fig. 3). Along the pitch this
fer to return a Robin (mixed type) boundary condition to the solid value can be expected to be lower, while at the rib corners higher
domain: Biot number peaks occur, which make an introduction of a relax-
qwall = hˆ · (Twall − T f luid ), (10) ation coefficient needed for both the FFTB and the TFFB coupling
methods, in turn slowing down the convergence. Moreover, using
with the virtual heat transfer coefficient hˆ , the heat flux qwall , the the FFTB method would lead to an ill-posed problem for the cou-
temperature at the wall Twall and the ambient fluid temperature pled computation, since only flux boundary conditions could be
Tfluid . Twall and qwall are passed to the solid solver as Robin condi- imposed on both sides of the solid domain.
tion. Note that h and Tfluid have to be computed for this method. Therefore, a Robin type boundary condition is advantageous.
This is done with an algorithm that sets a fixed virtual heat trans- Being aware of varying Biot numbers inside the channel and its
fer coefficient. The advantage of this method is that it is stable for dependence on the choice of reference length, revisiting the stabil-
Bi > 1 and also for Bi <= 1, if an appropriate heat transfer coeffi- ity criteria presented in the previous section, the Robin boundary

Fig. 4. Flow chart of applied hFFB coupling algorithm.

 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664 655

Similarly, Tyacke and Tucker (2015) and Sewall (2005) observed on

a ribbed channel with a blockage ratio of 0.1 that with applied
turbulence fluctuation at the inlet, the turbulence intensity was
higher than experimental data while it was lower if fluctuations
were omitted. These observations were restrained only to regions
before the second rib and no influence could be observed for the
flow after the second rib.
Furthermore, the turbulence intensity of the experimental
source for the current study by Cukurel and Arts (2013) was 5.3%
in case of the channel area averaged value and 1.6% towards the
center of the inlet plane and can be considered to have a negligi-
ble influence.
To simulate the exact flow conditions with the small Mach
number of the experimental campaign of Cukurel (2012) at Ma =
0.02, the computational time (using a compressible solver with ex-
Fig. 5. Sketch of stable regions for the standard coupling methods for weakly cou- plicit time integration) would have been unjustifiably long. Hence,
pled conjugate heat transfer computations.
the Mach number was increased to Ma = 0.2, keeping the same
Reynolds number, based on the bulk velocity and the hydraulic di-
condition type coupling procedures show their dependency on the ameter, at 40,0 0 0. This scaling granted a decrease of the compu-
chosen virtual heat transfer coefficient hˆ . The hFTB method is only tational cost and a re-scaling to more realistic engine conditions,
stable for a sufficient large hˆ , exceeding the largest value of h oc- while maintaining the similarity with the experimental data and
curring in the domain. As the value of h approaches infinity for keeping the simulated Mach number small enough to avoid com-
sufficiently refined meshes near leading edges, the hFTB method pressibility effects.
is inoperable for the current application in view of the very re-
fined grid resolution near sharp corners leading to high values of 4.2. Mesh description
h (Fig. 5). In turn, the hFFB method is stable if hˆ is smaller than
the largest occurring h, which is practical to follow. Accordingly, Due to the ubiquitous presence of walls in the internal cool-
the choice for the present study was the hFFB coupling algorithm ing passage and the need for resolving the walls instead of us-
as depicted by Fig. 4. ing wall-functions to obtain a realistic heat transfer coefficient,
The starting point of the CHT computations for the applied al- the mesh requirements are high, leading to demanding and costly
gorithm was a convective LES imposing a constant heat flux at wall-resolved LES.
the interface wall qwall . With the resulting wall temperature Twall , The unstructured LES solver AVBP was in particular designed to
the heat flux qwall and Newton’s law of cooling a ambient virtual deal with hybrid meshes, therefore a hybrid mesh (consisting of
fluid temperature was computed that was used to impose a Robin tetrahedra within the channel and prisms at the walls) has been
boundary condition with the chosen virtual heat transfer coeffi- used for the current application. Fig. 6 shows the tetrahedra of a
cient to the solid domain. The solid computations lead to updated used mesh around two ribs of the channel in the symmetry plane.
heat flux distribution at the wall that were again imposed to the To obtain the correct temperature and heat transfer, the closest
fluid domain to restart the iteration loop. The solid solver was con- grid point needs to be within the viscous sublayer, leading to small
sidered as steady state, i.e., it was run until full convergence. The grid cells at the wall. Fig. 6(b) shows a zoom at the corner of a
fluid solver was considered to be statistically converged using 10 rib visualizing the prism layer. Fransen et al. (2012) and Fransen
flow throughs for each iteration and for the exchanged quantities (2013) observed that using one prism layer results in a good mesh
of interest including the transient evolution of the thermal bound- quality for a hybrid mesh in such a ribbed geometry, on which was
ary condition, which is beneficial for the stability of the procedure. build upon in this work.
Increasing the amount of prism layers penalized the quality in
4. Computational setup the rib to bottom wall corners.

4.1. Flow and boundary conditions 5. Results

No-slip conditions were applied at all walls. The inlet and out- The results of the current numerical work were compared with
let boundaries are positioned far from the actual part of inter- aerodynamic data of Casarsa and Arts (2005) and heat transfer
est avoiding major influences on the flow around the ribs. Adi- measurements of Cukurel and Arts (2013), which were both done
abatic conditions were applied on the lateral walls and the top at the same Reynolds number of 40,0 0 0. Fig. 7 shows coordinate
wall, while iso-flux boundary conditions were applied to the rib- system and the local coordinate around one rib xc used in the
surrounding surface. For the first coupling iteration, a uniform figures of this section. The next section presents an estimate for
static temperature was set at the inlet. The static pressure was set the quality of the present LES including a mesh convergence study
at the outlet. A turbulent mean profile for the mass-flow was ap- showing four levels of refinement. Subsequently, the flow struc-
plied as inlet condition. The inlet boundary condition was set with- ture, the aerodynamic mean flow field and turbulent statistics are
out imposed fluctuations. The omission of inlet fluctuations was presented. Finally, the coupled CHT results are shown and com-
justified by previous studies. Indeed, Keylock et al. (2012) demon- pared with the experimental data.
strated on a single rib with a lower blockage ratio (0.15) than in
the present study that the rib has such a strong effect on the flow 5.1. Resolution
that the precise nature of the inlet condition, with regards to the
turbulent fluctuations and length scales, is rapidly forgotten. In- To estimate the resolution and to asses the quality of the
deed, only small differences for the flow separation and reattach- present LES, the following steps have been carried out: the dis-
ment may appear. The reattachment lengths and many flow quan- tance of the nearest grid point to the wall has been estimated
tities are preserved reasonably well, regardless of the inlet profile. following common guidelines for boundary layers in viscous units
656 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664

Fig. 6. Mesh around two ribs and rib corner.

Table 2
Ribbed slab characteristics.

Parameter abbr. Value

length Lsolid 1250 mm

height, depth Hsolid , dsolid 75 mm
thickness Ls,d 47.5 mm
material AISI 304 Steel 1260

Table 3
Used meshes. y+ is defined using the wall-normal distance y, the
friction velocity uτ the wall shear stress τ w as: y+ = uντ y . x+ and
Fig. 7. Sketch of the coordinates around rib 5. z+ are defined with the streamwise and spanwise mesh resolu-
tion, determining the aspect ratio of the cell.

and previous empirical studies with the present LES solver. There- mesh M1 mesh M2 mesh M3 mesh M4

fore, y+ values were kept reasonably small to obtain a statistically cells 3.53M 8M 20.3M 50.6M
steady grid independent solution of the heat transfer around the nodes 0.69M 1.5M 3.74M 9.4M
ribs. The heat transfer as the quantity of interest was evaluated in prisms no yes yes yes
terms of the enhancement factor as introduced in the subsequent mean y+ 18.7 8 3.4 1.9
x+, z+ ∼1 ∼5 ∼8 ∼8
section. Furthermore, the friction factor as another integral quan-
tity was compared with the experimental data. Moreover, energy
spectra, two-point correlations and the ratio of the SGS viscosity
to the molecular viscosity were considered. temperature Tbulk . Due to uniform heating and periodical flow con-
A comparison of different mesh resolutions resulting in a con- ditions, the bulk temperature is assumed to linearly increase inside
vergence study, and its effect on the enhancement factor (EF) the channel. Considering the values around ribs, the linear increase
has been done. The enhancement factor quantifies the increase of of the bulk temperature is further simplified with a sequence of
heat transfer in comparison with a correlation valid for flow in a step functions (which results in a constant bulk temperature along
smooth pipe at the same Reynolds number. It is defined as: the pitch). This assumption allowed a fair comparison with previ-
ous results, which used the same data reduction approach (Cakan,
Nu
EF = . (12) 20 0 0).
Nu0 Table 3 summarizes the mesh properties. Mesh M1 lacked qual-
The Nusselt number is defined as: ity to resolve major flow structures and the needed resolution for
h · Dh the heat transfer. From mesh M2 on, using a higher resolution and
Nu = , (13) a prism layer in the boundary layer, the aerodynamics were well
λ (T )
represented. To further improve the heat transfer predictions, mesh
with the heat transfer coefficient h, the hydraulic diameter Dh and M3 used larger cell aspect ratios for the wall prisms, while still
the fluid conductivity λ. The Dittus–Boelter correlation [19] gives remaining lower than 8 (z ࣃ x ࣃ AR · y). The histogram in
the denominator for the enhancement factor (valid for 0.7 < Pr < Fig. 8 presents the y+ distribution around the ribs for meshes M3
120) and is defined as: Nu0 = 0.023Re4D/5 P r 0.4 . The heat transfer co- and M4. The smaller wall distance of the first grid point of meshes
h
efficient h is defined as in Eq. (7), while T∞ is defined as the bulk M3 and M4 keeps the majority of first node distances within the
 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664 657

Fig. 8. EF of used meshes along the unfolded rib and y+ value distribution around the ribs of meshes M3 and M4. y+ is defined using the wall-normal distance y, the friction
velocity uτ the wall shear stress τ w as: y+ = uντ y .

viscous sub-layer and y+ < 5. For accurate heat transfer predic- Table 4
Normalized friction factor.
tions inside rib-roughened cooling channels with the current used
solver and numerical schemes, Fransen et al. (2012) suggests keep- f /f 0
ing y+ < 10. Collado et al. (2012) have recently shown with the
Experiment (Casarsa, 2003) 12.3 ± 0.31
same numerical scheme on another application that a y+ value LES, mesh M1 22.0
of 4 leads to good predictions for the heat transfer. The mean y+ LES, mesh M2 18.5
value for mesh M3 and M4 in the current study are even smaller. LES, mesh M3 12.9
Since the heat transfer changed very little (Fig. 8(a)) from mesh LES, mesh M4 12.6
M3 to mesh M4, the result for mesh M4 can be considered grid-
converged. Hence, mesh M3 is assumed to deliver reliable results
and will be used for the analyses in the following sections.
The average pressure drop over one pitch interval is also an im- mesh M3 matches the experimental friction factors exactly and
portant integral parameter to demonstrate the simulation quality Mesh M4 only deviates by less than 3% which further validates the
(Arts et al., 2007). The pressure drop is expressed using the fric- simulations.
tion factor, Incopera and Witt (2007), Fig. 9 shows the ratio of the turbulent viscosity over the molec-
D h P ular viscosity, as often used as means for an evaluation of the LES
f = , (14) quality, e.g., recently by Schneider et al. (2015). The ratio of the
2L p ρ U 2
turbulent viscosity over the molecular viscosity for the current
with Lp being the pitch length. The rib effect can be demon- case is reasonable and lies in a typical range for LES in the whole
strated by normalizing the friction facgtor with the friction fac- domain as Fig. 9 demonstrates. The ratio is small where the mesh
tor for a smooth pipe f0 = 0.046Re−0.2 . Table 4 shows the ratio refinement is large, but also depends on the turbulence intensity
of these friction factors for the simulation compared with the ex- and is therefore highest in shear layer regions with the largest tur-
periment (Casarsa and Arts, 2005). P is the static pressure drop bulent production, where it can reach higher values of around 40
along the pitch, which was measured from the slope of the wall also for very fine grid resolutions Davidson (2009).
static pressure using pressure tappings on the smooth top wall and The energy spectra at distinct points (Fig. 10) show that a large
on the opposite ribbed wall. The difference between the experi- range of frequencies are resolved with the current LES, exhibiting a
ment and the simulations is shown by Table 4. The friction fac- −5/3 slope, as expected for an inertial region of a fully developed
tors for meshes M1 and M2 deviate from the experiment, whereas turbulent spectrum.
658 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664

Table 5
Sizes of recirculation zones in the symmetry plane in streamwise direction (see
Fig. 4); ∗ experiment by Casarsa and Arts (2005).

V1 V2 V3 V4

∗
Experiment 3.76H-3.84H 0.225H-0.28H 0.6H-0.9H 1.04H-1.5H
Simulation 3.7H 0.31H 0.86H 1.46H

recirculation zones and on the heat transfer and, therefore, need

to be accurately predicted.
The next section quantifies the described flow features and val-
idates the conducted LES with the experimental data.

5.3. Mean flow field

This section presents results for time-averaged statistics. The

Fig. 9. Profiles of sub-grid scale turbulent viscosity over molecular viscosity ν t,SGS /ν averaging time of the simulations was equivalent to 22 flow
at different locations.
throughs, which is the time a fluid particle needs to travel through
the whole channel, based on the bulk velocity. Note that the chan-
After assessing the quality of the used mesh, the next section nel contains six ribs, which would translate to 132 flow throughs
describes the flow inside the channel. in a periodic case, ensuring sufficient statistical convergence.
Fig. 11 shows the streamlines of the mean flow field in the
5.2. Flow structure symmetry plane around rib n0 4, displaying the four main re-
circulation zones in streamwise direction. Table 5 compares their
The high blockage ratio of the channel leads to a high level of lengths with the experimental study of Casarsa (2003) as well as
turbulence production with the largest levels on top of the ribs. Casarsa and Arts (2005). The values of the recirculation lengths
Before reaching the rib, the flow accelerates while a fairly steady are evaluated by applying the same criteria as defined by Casarsa
recirculation zone in front of the rib persists. On top of the rib (2003) to assure a direct comparison. The reattachment lengths
another recirculation bubble is created, fluttering such that a vor- are therefore measured at a small distance above the wall, done
tex shedding behind the rib affects the large recirculation bubble, in the experiment due to the resolution of the PIV measurement
which, in turn, determines the reattachment point. Behind the rib, close to the wall, with a negative streamwise velocity for V1 at
the flow creates a strong shear layer after a clear separation point y/H = 0.05 and for V3 at y/H = 0.01, a positive streamwise veloc-
at the rib top back corner. The recirculation zone size variation be- ity for V2 at y/H = 0.05 and a vertical velocity, exceeding a thresh-
hind the rib has a length of about one third of the pitch length. old of 0.05m/s (scaled for LES due to a change in Mach number) at
The large curvature of the streamwise flow path imposed by the y/H = 0.05.
blockage of the ribs leads to secondary flow structures, which have Evaluating the LES case for the recirculation zones, the sepa-
a large impact on the pressure losses, caused by the mean flow ration downstream of the rib (V1) is in good agreement with the

Fig. 10. Energy spectra.

 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664 659

Fig. 11. Recirculating zones at rib n0 4 visualized with streamlines in the center plane.

Fig. 12. Velocity profile comparison.

measurement, being close to the range of the experimental uncer- 5.4. Verification for convective heat transfer
tainty. The downstream corner region (V2) is only slightly over-
estimated. The recirculation bubble sizes agree well for the top The most challenging part in modeling complex flow fields in
(V3) and front (V4) regions. Fig. 12(a) and (b) show the streamwise such configurations is the prediction of the heat transfer. First, the
and vertical velocity component mean profiles. The LES agrees very aerodynamic field needs to be accurately predicted to insure a cor-
well anew with the PIV data. Moreover, the velocity fluctuations rect heat transfer prediction, which has been shown in the previ-
are in good agreement (Fig. 13), too. Finally, the secondary flow ous section. Fig. 15 shows the comparison of the enhancement fac-
structures are all very well represented quantitatively and qualita- tor around one rib between the experiment by Cukurel (2012) and
tively as evidenced by Fig. 14. In addition to the curvature induced a convective LES case (Scholl et al., 2015).
main secondary flow structures on top of the rib, the simulation The overall EF and the EF on the pitch was in good agreement,
also reveals small secondary flow structures in the corners of the while the differences occurring especially around the rib corners
channel, which are typical for turbulent flow in internal rectan- could be explained through the difficulty of imposing the same
gular channels (Melling and Whitelaw, 1976), but which were not boundary conditions in the experiments and in the simulations. A
measured with the PIV. sharp peak in the Nusselt number for the simulation can be seen
660 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664

Fig. 13. Streamwise velocity fluctuations as urms .

Fig. 14. Streamline comparison on top of the rib at xc /h = 0 and on the pitch at xc = 5.16.

more significantly, a physical reason can be attributed analogous

to the stagnation point of an airfoil or the leading edge of a flat
plate, where the thermal boundary layer is absent and the Nusselt
number approaches infinity. Therefore, the Nusselt number is ex-
pected to be high at this point. The experiment on the other hand,
approaches very high measurement uncertainties toward the rib
corner with limited spatial resolution, incapable of measuring the
high peak in the direct vicinity of the corner.
The convective LES case was used as starting condition, or it-
eration 0 for the weakly coupled CHT approach, which the next
chapter presents.

5.5. Conjugate heat transfer

Fig. 15. Convective heat transfer comparison. Experiment by Cukurel and Arts
(2012). In the case of coupled convective conductive computations, a
correct heat transfer prediction is only a first requirement that
at x = –0.5. The reason for this peak is twofold: first, the numer- needs to be fulfilled. Furthermore, the coupled computations need
ical model assumes an infinitely sharp corner that would in real- to be stable and need to converge. Applying the hFFB coupling
ity always be rounded, which would lower the peak. Second, and algorithm (Fig. 4), the required heat transfer coefficient to insure
 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664 661

Fig. 16. Evolution of flux responses for CHT using LES. Fig. 19. Comparison of flux with experiments by Cukurel and Arts (2013) and
Agostini and Arts (2005).

Fig. 17. Evolution of normalized difference in flux responses.

Fig. 20. Comparison of EF with experiment.

particular on top of the rib the convergence trend is rather straight,

while on the back rib corner and directly after the rib, where the
heat transfer coefficient is the smallest, the convergence oscillates.
This is due to a difference in local heat transfer.
Fig. 19 compares the fluxes of two experiments with the LES.
While on the pitch all results have a high level of consistency, in
the rib vicinity, for |xc /H| < 2.5, there is a larger spread of heat
flux. The difference between the two experiments can be explained
by the applied measurement techniques and modeling differences.
Agostini and Arts (2005) measured the temperature with an in-
frared camera. The solid-fluid interface was made of a polyethy-
lene film, which transmitted a fraction of the infrared radiation. On
Fig. 18. Evolution of normalized difference in flux responses. the other hand Cukurel and Arts (2013) used steady wide-band liq-
uid chrystal thermometry leading to differences in measurements.
To obtain the heat flux from the measurements, the lack of mod-
stability is hˆ < 2 · h in the whole domain. With an approximate eling of the applied paint layer by Agostini and Arts (2005) and
mean heat transfer coefficient of around 800 W/m2 K for the iso- the application of adiabatic boundary conditions lead to superfi-
flux computation, the imposed heat transfer coefficient was set to cially large temperature gradients in the case of the experiments
a value of 500 W/m2 K. by Agostini and Arts (2005) in turn also leading to a large spread
Fig. 16 shows the heat flux response obtained from the con- in fluxes around the ribs. Nonetheless, the comparison reveals a
duction computation for different iterations over one unfolded rib peak in front of the first rib that is not present in the experiment
at the symmetry plane. The heat flux is normalized with the ref- of Cukurel and Arts (2013), but is present for the LES results.
erence value uniformly imposed under the slab. The flux redistri- The good comparison of the LES with one of the two experi-
bution due to the conduction is apparent. In comparison with the ments for the fluxes also leads to a good agreement for the en-
reference value, the heat flux is increased up to 130% on top of the hancement factor as Fig. 20 shows. The distribution of the en-
rib and is decreased to around 40% in the rib downstream region. hancement factor on the pitch agrees well, while differences occur
Fig. 17 shows the difference between two coupling iteration steps at the rib corners, where, as mentioned in the previous section, it
of the symmetry plane heat flux distribution. It demonstrates that is impossible to measure a temperature accurately, while the LES
the coupling procedure converged well. Fig. 18 shows the conver- provides results. The major trend is common both in the experi-
gence at several distinct points. It is visible that on the pitch and in ment and for the LES: the enhancement factor peaks on the rib
662 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664

Fig. 21. Comparison of temperature distributions between conjugate and convective case.

Table 6
Enhancement factor comparison; ∗ experiment by Cukurel and Arts (2013).

Rib

Front Top Back Mean Inter Rib Average

∗
experiment 3.2 2.86 1.89 2.65 2.2 2.31
experiment cht 2.34 2.99 0.86 2.07 2.01 2.02
LES iso-flux 2.65 3.72 1.7 2.7 2.04 2.2
LES coupled 2.28 3.59 1.18 2.34 2.01 2.05

Fig. 23. Fluid streamlines and solid iso-thermals.

convective LES, confirming that the major difference occurs on the

rib back.
For the practical case of a real turbine blade, assuming a coolant
temperature of 900 K and a metal temperature at the material
limit of 1350 K, a temperature difference in the blade of around
7.5 K is obtained if the overall change in EF is applied. Note that
this rough evaluation may induce larger local differences because
of the larger local enhancement factor and temperature differences.
Since the life of a turbine blade is reported to be half for an in-
creased temperature of 15 K in case of creep, this temperature dif-
ference demonstrates the severe impact of the applied boundary
condition on the design application.
Fig. 22. EF difference for LES convective case against LES coupled case.
The convective-conductive interaction is well revealed in Fig. 23
that shows a cut through the symmetry plane for the solid and
fluid part. The fluid domain is shown in terms of time-averaged
top, while it drops significantly on the rib back. Table 6 shows the streamlines, while the solid domain shows iso-thermals. The rib
EF for different parts of the interface. Both experiment and sim- acts as a heat sink, where the largest amount of heat flux is re-
ulation show a decrease on the front part of the rib and a larger leased on top of the rib, while smaller amounts are released on
decrease on the back, while the EF on the pitch remains in the the rib front and the rib back. The large flow recirculation zone,
same range as in the convective case. located downstream on the rib back, hinders the heat to enter the
Fig. 21 shows a comparison between the temperatures of the fluid part as visible through the iso-thermals that circumnavigate
coupled and the uncoupled case, illustrating the effect of the rib on around the large recirculation structure leading to an asymmetric
the conjugate heat transfer. The temperature has been normalized course of the iso-thermals. This large recirculation zone also leads
with the inlet temperature Tmin and the maximal occurring tem- to a smaller heat release through the rib back than through the rib
perature Tmax of the convective case as = (T(max
T −T min )
−T min )
. While on front, where the mean flow impinges.
the largest part of the pitch, for x/H < 3 and x/H > 3.75, the tem-
perature has slightly increased, on the rib back and downstream 6. Conclusions
part of the pitch between x/H > 1.5 and x/H < 3.75, the area with
the largest influence of the recirculation zone V1, the temperature This contribution presents a numerical analysis of the turbulent
has magnificently decreased, which is further depicted by Fig. 22 flow and the heat transfer inside a rib-roughened cooling channel
showing the enhancement factor difference between conjugate and using an LES flow solver coupled with a conduction solver for the
 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664 663

solid domain. The aerodynamic results were compared with PIV Donea, J., Huerta, A., 2003. Finite Element Methods for Flow Problems. John Wiley
measurements and with heat transfer and temperature measure- and Sons, Ltd., Chichester, UK.
Duchaine, F., Corpon, A., Pons, L., Moureau, V., Nicoud, F., Poinsot, T., 2009. Devel-
ments finding a good overall agreement between the simulations opment and assessment of a coupled strategy for conjugate heat transfer with
and the experiments. The mean flow structures are well repre- large eddy simulation: Application to a cooled turbine blade. Int. J. Heat Fluid
sented by the simulations. The enhancement factor along the chan- Flow 30, 1129–1141.
Duchaine, F., Maheu, N., Moureau, V., Balarac, G., Moreau, S., 2013. Large-eddy sim-
nel is also in good agreement with experimental studies. ulation and conjugate heat transfer around a low-mach turbine blade. J. Turbo-
A weak coupling strategy was applied for conjugate heat trans- mach.
fer leading to a good statistical convergence. The results show an Errera, M., Baqu, B., 2013. A quasidynamic procedure for coupled thermal simula-
tions. Int. J. Numer. Methods Fluids.
important influence of the thermal boundary condition on the heat
Fransen, R., 2013. - Simulation aux Grandes Echelles pour la modélisation aérother-
transfer inside the channel. Comparing the convective with the mique des aubages de turbines refroidies. Université de Toulouse - MeGeP -
conjugate heat transfer LES, important overall differences in heat Dynamique des Fluides Ph.D. thesis.
Fransen, R., Gourdain, N., Gicquel, L., 2012. Steady and unsteady modeling for heat
transfer occur. Locally, the change of the heat transfer from uncou-
transfer predictions of high pressure turbine blade internal cooling. In: Proceed-
pled to coupled simulations is as large as 80% , leading to local ings of ASME Turbo Expo.
changes of the enhancement factor of more than 50% , occurring Han, J., Dutta, S., Ekkad, S., 20 0 0. Heat Transfer Cooling Technology. Taylor and Fran-
in particular on the back of the rib, where the large recirculation cis, Reading, MA.
Han, J.-C., 2010. Recent studies in turbine blade internal cooling. Heat Transfer Re-
zone amplifies the conjugate heat transfer effect. For temperature search.
predictions with design application the imposition of constant heat He, M., Kassab, A.J., Bishop, P.J., Minardi, A., 1995. An iterative fdm/bem method
flux leads to significant differences in comparison with the true for the conjugate heat transfer problem – parallel plate channel with constant
outside temperature. Eng. Anal. Bound. Elements 15 (1), 43–50.
physics. To accurately and reliably predict the enhancement factor, Heidmann, J., Rigby, D., Ameri, A., 20 0 0. A three-dimensional coupled exter-
conjugate heat transfer computations are inevitable. nal/internal simulation of a film-cooled turbine vane. ASME J. Turbomach 124,
348–359.
Acknowledgments Heselhaus, A., 1998. A hybrid coupling scheme and stability analysis for coupled
solid/fluid turbine blade temperature calculations. Asme Turbo Expo. Paper No.
98-GT-88.
The research leading to these results has received funding from Iaccarino, G., Ooi, A., Durbin, P., Behnia, M., 2002. Conjugate heat transfer predic-
the European Community’s Seventh Framework Programme (FP7, tions in two-dimensional ribbed passages. Int. J. Heat Fluid Flow.
Iacovides, H., Launder, B., 1995. Computational fluid dynamics applied to internal
2007–2013), PEOPLE programme, under the grant agreement No gas-turbine blade cooling: a review. Int. J. Heat Fluid Flow.
FP7-290042(COPAGT project). Iacovides, H., Launder, B.E., 2007. Internal blade cooling: The cinderella of compu-
tational and experimental fluid dynamics research in gas turbines. Proc. Inst.
References Mech. Eng. Part A 221 (3), 265–290.
Illingworth, J.B., Hills, N.J., Barnes, C.J., 2005. 3d fluid-solid heat transfer coupling of
Agostini, F., Arts, T., 2005. Conjugate heat transfer investigation of rib-roughened an aero engine pre-swirl system. Asme Turbo Expo. Reno-Tahoe, Nevada. Paper
cooling channels. In: Proceedings of GT2005 ASME Turbo Expo 2005: Power for No. GT2005-68939
Land, Sea and Air June 6–9, 2005, Reno-Tahoe, Nevada, USA. Incopera, F., Witt, D.D., 2007. Fundamentals of Heat and Mass Transfer. John Wiley
Amano, R.S., 1994. A study of rotor cavities and heat transfer in a cooling process and Sons.
in a gas turbine. ASME J. Turbomach 116, 333–338. Keshmiri, A., 2012. Numerical sensitivity anaylsis of 3- and 2- dimensional
Arts, T., Benocci, C., Rambaud, P., 2007. Experimental and numerical investigation rib-roughened channels. Heat Mass Transfer.
of flow and heat transfer in a ribbed duct. 3rd International Symposium on In- Keylock, C.J., Tokyay, T.E., Constantinescu, G., 2012. A method for characterizing the
tegrating CFD and Experiments in Aerodynamics, U.S. Air Force Academy, CO, sensitivity of turbulent flow fields to the structure of inlet turbulence. J. Turbu-
USA. lence.
Cakan, M., 20 0 0. Aero-Thermal Investigation of Fixed Rib-Roughened Cooling Pas- Leonardi, S., Orlandi, P., Djenidi, L., Antonia, R., 2004. Structure of turbulent channel
sages. Universite Catholique de Louvain and Von Karman Instititue for Fluid Dy- flow with square bars on one wall. Int. J. Heat Fluid Flow 25, 384–392.
namics Ph.D. thesis. He, L., Oldfield, M., 2011. Unsteady conjugate heat transfer modeling. ASME J. Tur-
Casarsa, L., 2003. Aerodynamic Performance Investigation of a Fixed Rib-Roughened bomach 133, 031022–1–12.
Internal Cooling Passage. Universita Degli Studi Udine and Von Karman Insti- Ligrani, P., 2013. Heat transfer augmentation technologies for internal cooling of tur-
titue for Fluid Dynamics Ph.D. thesis. bine components of gas turbine engines. Int. J. Rotat. Mach.
Casarsa, L., Arts, T., 2005. Experimental investigation of the aerothermal perfor- Lohasz, M.M., Rambaud, P. Benocci, C., 2006. Flow features in a fully developed
mance of a high blockage rib-roughened cooling channel. ASME J. Turbomach ribbed duct flow as a result of miles flow. Turbul. Combust. 77, 59–76.
127, 580–588. Melling, A., Whitelaw, J., 1976. Turbulent flow in a square duct. J. Fluid Mech.
Ciofalo, M., Collins, M.W., 1992. Large-eddy simulation of turbulent flow and heat Mendez, S., Nicoud, F., 2008. Large eddy simulation of a bi-periodic turbulent flow
transfer in plane and rib-roughened channels. Int. J. Numer. Methods Fluids 15, with effusion. J. Fluid Mech 598, 27–65.
453–489. Miyake, Y., Tsujimoto, K., Nagai, N., 2002. Numerical simulation of channel flow with
Coletti, F., Armellini, A., Arts, T., Scholtes, C., 2011. Aerothermal investigation of a a rib-roughened wall. J. Turbul. 3, 035.
rib-roughened trailing edge channel with crossing jets - part 2: Heat transfer Nicoud, F., Ducros, F., 1999. Subgrid-scale stress modelling based on the square of
analysis. ASME J. Turbomach 133, 031024–1–031024–8. the velocity gradient. Flow Turb. Combust. 62, 183–200.
Coletti, F., Scialanga, M., Arts, T., 2012. Experimental investigation of conjugate heat Ooi, A., Iaccarino, G., Durbin, P., Behnia, M., 2002. Reynolds averaged simulation of
transfer in a rib-roughened trailing edge channel with crossing jets. ASME J. flow and heat transfer in ribbed ducts. Int. J. Heat Fluid Flow.
Turbomach 134. Patil, S., Tafti, D., 2013. Large-eddy simulation with zonal near wall treatment of
Colin, O., Benkenida, A., Angelberger, C., 2003. 3d modeling of mixing, ignition and flow and heat transfer in a ribbed duct for the internal cooling of turbine
combustion phenomena in highly stratified gasoline engine. Oil Gas Sci. Tech 58 blades. J. Turbomach 135(3).
(1), 47–62. 111. Perelmann, L., 1961. On conjugate problems of heat transfer. Int. J. Heat Mass Transf.
Colin, O., Rudgyard, M., 20 0 0. Development of high-order taylor-galerkin schemes 3, 293–303.
for unsteady calculations. J. Comput. Phys 162 (2), 338–371. Poinsot, T., Lele, S., 1992. Boundary conditions for direct simulations of compressible
Collado, E., Gourdain, N., Duchaine, F., Gicquel, L., 2012. Effects of free-stream turbu- viscous flows. J. Comput. Phys. 101 (1), 104–129.
lence on high pressure turbine blade heat transfer predicted by structured and Raisee, M., Naeimi, H., Alizadeh, M., Iacovides, H., 2009. Prediction of flow and heat
unstructured les. J. Heat Mass Transf. transfer through stationary and rotating ribbed ducts using non-linear k-epsilon
Cui, J., Patel, V.C., Lin, C.L., 2003. Large-eddy simulation of turbulent flow in a chan- model. Flow, Turbul. Combust. 82, 121–153.
nel with rib roughness. Int. J. Heat Fluid Flow 24, 372–388. Rau, G., Cakan, M., Moeller, D., Arts, T., 1998. The effect of periodic ribs on the local
Cukurel, B., 2012. Conjugate Heat Transfer Investigation of a Fixed Rib Roughened aerodynamic and heat transfer performance of a straight cooling channel. ASME
Cooling Passage. Purdue University and Von Karman Institute for Fluid Dynam- J. Turbomach. 120, 368–375.
ics Ph.D. thesis. Sagaut, P., 20 0 0. Large Eddy Simulations for Incompressible Flows. Scientific com-
Cukurel, B., Arts, T., 2012. Film cooling extraction effects on the aero-thermal char- putation series. Springer-Verlag.
acteristics of rib-roughened cooling channel flow. ASME GT2012-68680. Schneider, H., Terzi, D.V., Bauer, H., Rodi, W., 2015. Coherent structures in trailing
Cukurel, B., Arts, T., 2013. Local heat transfer dependency on thermal bound- edge cooling and challenge for turbulent heat transfer modelling. Int. J. Heat
ary condition in ribbed cooling channel geometries. ASME J. Heat Transf. 135, Fluid Flow.
101001–1–101001–11. Scholl, S., Verstraete, T., Torres-Garcia, J., Duchaine, F., Gicquel, L., 2015. Influence
Davidson, L., 2009. Large eddy simulations: how to evaluate resolution. Int. J. Heat of the thermal boundary conditions on the heat transfer of a rib-roughened
Fluid Flow. cooling channel using les. Proc. Inst. Mech. Eng. Part A.
664 S. Scholl et al. / International Journal of Heat and Fluid Flow 61 (2016) 650–664

Schönfeld, T., Poinsot, T., 1999. Influence of boundary conditions in les of premixed Verstraete, T., 2008. Multidisciplinary Turbomachinery Component Optimization
combustion instabilities. Ann. Res. Briefs. Center Turbul. Res. NASA Ames and Consifering Performance, Stress, and Internal Heat Transfer. Universiteit Gent
Stanford University, pp. 73–84. Ph.D. thesis.
Sewall, E., 2005. Large eddy simulation of flow and heat transfer in the developing Weigand, B., Semmler, K., Wolfersdorf, J.V., 2001. Heat transfer technology for in-
and 180 bend regions of ribbed gas turbine blade internal cooling ducts with ternal passages of air-cooled blades for heavy-duty gas turbines. In: Annals of
riotation - effect of coriolis and centrifugal buoyancy forces. Virgina Polytechnic the New York Academy of Sciences. In: Heat Transfer in GAS Turbine Systems,
Institute and State University Ph.D. thesis. volume 934, pp. 179–193.
Tyacke, J., Tucker, P., 2015. Large eddy simulation of turbine internal cooling ducts. Yang, K., Ferziger, J.H., 1993. Large-eddy simulation of turbulent obstacle flow using
Comput. Fluids. dynamic subgrid-scale model. AIAA J. 32 (8), 1406–1413.
Verdicchio, J.A., Chew, J.W., Hills, N.J., 2001. Coupled Fluid/Solid Heat Transfer Com-
putation for Turbine Discs. ASME TURBO EXPO. New Orleans, Luisiana, Paper
No. 2001-GT-0205.