You are on page 1of 15

Nuclear Engineering and Design 240 (2010) 321–335

Contents lists available at ScienceDirect

Nuclear Engineering and Design


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

Simulation of a compressor cascade with stalled flow using Large Eddy


Simulation with two-layer approximate boundary conditions
N. Tauveron
Commissariat à l’Energie Atomique, Direction de l’Energie Nucléaire, 17 rue des Martyrs, 38054 Grenoble, France

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

Article history: The final objective of the study is to provide correlations and models, which are able to describe axial
Received 15 November 2007 compressor behaviour in some particular incidental or accidental conditions, such as stall. The present
Received in revised form 10 April 2008 contribution consists to establish and begin to validate numerical simulations by comparison with rep-
Accepted 2 July 2008
resentative experimental data on a planar static compressor cascade. Trio U software is used to perform
Large Eddy Simulations with a standard wall-function (SWF) or a two-layer approximate boundary condi-
tions (TBLE). The results obtained show a good behaviour of the TBLE model compared to an open literature
case of the gas turbine aircraft community. Sensitivity studies are performed on various parameters
(solidity, angle of incidence) and take part of a database.
© 2008 Elsevier B.V. All rights reserved.

1. Introduction When the incidence angle is increased, the compressor blade row
stalls (similarly to an isolated airfoil). The objective of this paper
1.1. Context of the present study is to provide knowledge on axial compressor cascade behaviour
in stall conditions, in order to propose correlations and models
As a result of a screening review of candidate technologies, adapted to this regime (mid-term objective).
CEA has selected an innovative concept of high temperature gas
cooled reactor with a fast neutron spectrum. Direct or indirect 1.2. Paper organisation
cycles are considered. Design and safety considerations require a
‘system code’ able to describe the thermal hydraulic behaviour of At first we briefly present compressor modelling for off-design
the whole plant. As both concepts – direct or indirect – use helium situations (Section 2.1). Then the paper focuses on stall regime:
compressors to maintain the flow in the core (the compressor is not we present analytical models (Section 2.2) and describe an exper-
the same for direct or indirect cycle), a special emphasis must be imental configuration, which is an open literature case of the gas
laid on compressor modelling, especially on low mass flow regimes. turbine aircraft community, with which fully stalled results were
These particular regimes, which are investigated in the present obtained (Section 3). Then we focus on the numerical model (Sec-
publication, are of great importance as they can generate ther- tion 4): Trio U (Trio U, 2007), a software for computational fluid
malhydraulic instabilities in the compressor and the circuit. These dynamics (CFD), is used to perform Large Eddy Simulations with
instabilities (surge and stall) can also impose negative mass flow a standard wall-function or a two-layer approximate boundary
in some parts of the circuit (Tauveron et al., 2007; Kikstra, 2001; conditions (TBLE). This method takes place among general hybrid
Bammert and Zehner, 1975) and generate high level loads on com- methods (RANS-LES) used to treat some unstaedy problems con-
pressor blades and thrust bearings. In this context, a description cerning the safety analysis of nuclear reactors (CSNI, 2007). Section
of the turbomachinery has been developed, based on a quasi- 5 is devoted to the validation of the numerical simulations by com-
two-dimensional approach to generate performance maps for the parison with the representative experimental data. In the last part
system code CATHARE (Tauveron et al., 2007). However, this type of (Section 6) other parameters are investigated with the simula-
model requires cascade correlations as input data (deviation angle, tions.
losses).
At the scale of the compressor, low mass flow rate regimes cor- 2. Compressor off-design performance model
respond to high incidence angle (on compressor blades) regimes.
In CATHARE simulation each turbomachine element is defined
in a 0D approach: pressure change and enthalpy change are given
E-mail address: nicolas.tauveron@cea.fr. by solving modified momentum transfer and energy equations in

0029-5493/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.nucengdes.2008.07.015
322 N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335

frame (absolute for stators, with V velocities, ˛ angles or relative for


Nomenclature rotors with W velocities, ˇ angles).
To take into account blade forces, source terms for the lift and
c blade chord drag of each row in the blade cascade are derived from correlations.
GFR gas fast reactor From a formal point of view, equations are reformulated to use the
Ma Mach number based on axial velocity correlations of total pressure losses and deviation angle.
MU Mach number based on peripheral velocity
MHTGR modular high temperature gas reactor 2.1. Off-design stable conditions
P pressure
Ps static pressure We shortly present off-design modelling in stable conditions to
Pt total pressure illustrate the similarities in treating off-design regimes (formalism,
Re Reynolds number based on the chord transposition from air to helium data) and the differences (physical
S spacing between adjacent blades phenomena, maturity of description and validation processes).
SWF standard wall function For design and off-design stable situations, blade total pressure
U peripheral velocity losses are calculated as the sum of profile, annulus, secondary and
V absolute velocity tip clearance losses, whereas the deviation angle is the sum of
W relative velocity the deviation angle at the design point, corrected for off-design
WALE wall-adapting local Eddy-viscosity regimes. The profile losses and design deviation angle are given
1 relative to inlet by Lieblein (1959, 1960) with the modifications proposed by Swan
2 relative to outlet (see Fig. 4) (1961) for the evaluation of the blade boundary layer momentum
3 relative to outlet fully mixed flow (see Fig. 4) thickness. Mach number and Reynolds number influences are taken
into account following Koch and Smith (1976) data. For secondary
Greek symbols
and annulus losses, we use expressions given by Dixon (1975). Tip
˛ absolute angle
clearance losses are evaluated through formulae from Hubert and
ˇ relative angle
Bauermeister (1963), where the constant has been adapted to com-
 ratio of specific heats
pressor case; we have chosen this model as it takes into account
c camber angle
corrections on blade deflection angles, which is a well-known effect
 stagger angle
(Lakshminarayana, 1970).
 viscosity
The correlations use physical conditions (pressure, enthalpy and
 gas density
velocity triangles) and numerous geometrical parameters (radii,
 solidity = c/S
blade camber, thickness, stagger angle, height, cord, shape, num-
˚ flow coefficient (VX /U)
ber and tip clearance for each compressor cascade). Inlet nozzle and
 pressure coefficient ( P/(1/2)U2 )
outlet diffusers are also described in our model.
ω total pressure loss coefficient
The presented case is a helium axial compressor, which was used
as the primary circulator for Fort-Saint-Vrain reactor. On this case,
air experimental data were also obtained. The data are taken from
which source terms representing overall characteristics are added.
Cavallaro and Yampolsky (1974), Brey (1987), Mac and Yampolsky
These source terms are modelled in a quasi-steady approach. The
(1987).
assumption of quasi-steady behaviour is almost justified in numer-
The cross-comparisons between experimental data and simula-
ous transient phenomena: the transit time of the working fluid
tion results show a good agreement, for design and off-design stable
through a row in the machine is far below delays in numerous
points (Fig. 2). We also conclude that the transposition between
transient phenomena if we consider variations of pressure, tem-
air and helium flows seems correctly described by our model. As
perature, speed of the different components of the system. The
a consequence we are quite confident in the pertinence of testing
following considerations are devoted to the generation of the
the model with air machines. However, more test cases are needed
source terms: pressure rise and enthalpy rise.
to verify the models.
The model – more detailed in (Tauveron et al., 2007) – is based on
velocity triangles, deflection angle ( ˇ for the rotor, ˛ for the sta-
2.2. Stall regime
tor, see Fig. 1) and total pressure losses correlations. This approach
first consists in the solution of 1D axisymmetric Navier–Stokes
For the stall regime we only investigate profile phenomena. The
equations on an axial grid (z-axis in Fig. 1) inside the machine. Mass,
previous correlations for profile losses are not adapted to the stall
axial momentum (x- or z-axis), circumferential momentum ( -axis
regime. This fact is illustrated in Fig. 3. We note that the traditional
in Fig. 1) and total-enthalpy balances are written in the appropriate
correlations are well-suited for low angles of attack.
Particular models were developed in the literature; three of
them are presented here. The main features are derived from
Cornell (1954). They are shortly described hereafter.

2.2.1. Analytical models


The flow is assumed to be planar, steady and incompressible.
The blades are supposed to be closely spaced, i.e.  ≥ 1, thus the
flow behaviour is assumed to be similar to that for closely spaced
flat plates. The flow separates at the leading edge producing a jet
zone and a recirculation zone. After the leading edge, an intense
mixing process in the wake occurs. The fully mixed flow is of great
interest as it is supposed to be representative of the flow entering
Fig. 1. Support of compressor modelling (from Ottavy, 2004). the next row. This is the reason why the fully mixed state is used in
N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335 323

Fig. 2. Cross-comparisons between simulation and experimental data for Fort-Saint-Vrain circulator in helium (left) and in air (right).

the correlations. Mass, and momentum conservations are solved to 2.2.3. Moses and Thomason model
give fully mixed flow. It is also assumed that the jet-wake structure Three additional hypotheses are also added (Moses and
leaves the cascade parallel to the stagger direction. Thomason, 1986):

2.2.2. Cornell models • effects of friction within the blade passage are supposed to follow
For the clarity of presentation we call these models Longley
the empirical relation:
model, as they are fully described in Longley (2007), for negative
flows. Three additional hypotheses are added:
Pt,1 − Pt,2 cos2 (˛1 )
= MT (1)
• isentropic flow is assumed in the jet structure (potential flow), (1/2) V12 cos2 (˛2 )
• in the momentum conservation equation parallel to the stagger
angle, friction is ignored, where MT is a non-dimensional loss of pressure in Moses and
• the mixing process is either: Thomason model:
1. a constant area mixing process, 
2. a constant pressure mixing process. MT = 0.15 (2)
cos (˛1 )

With this hypothesis and the free streamline theory, losses and • the sum of the pressure and shear force parallel to the chord line
deflection at the fully mixed state can be calculated. Results are pre- is zero,
sented in Fig. 3. For reasons of clarity, we only present results about • the mixing process is a constant area mixing process.
total pressure losses. In particular Taylor expansion calculations
lead to, when ˛1 → 90◦ :
With this hypothesis and the free streamline theory, losses and
• ˛3 → 90◦ deflection at the fully mixed state can be calculated. Results are
• ω3 → cos2 () presented in Fig. 3. In particular Taylor expansion calculations lead
to, when ˛1 → 90◦ :

• ˛1 → 90◦
• ω3 → 1

Unfortunately it is very difficult to validate the different models,


as the experimental data do not reach full stall regime (Fig. 3). This
is the reason why we decide to investigate a wider range of angle
of attack with CFD simulations.

3. Experimental configuration considered

To our best knowledge the experimental data by Yocum and


O’Brien (Yocum, 1988; Yocum and O’Brien, 1993) constitute one
of the most interesting data base on stall conditions in order to
establish correlations (losses, deflection) at row scale for low speed
machines. As suggested in Section 2.1, a great amount of work has
been devoted to design point, off-design and stall occurrence, fully
Fig. 3. Cross-comparisons between experimental from Yocum and O’Brien, tradi-
tional correlations and specific analytical models: total pressure loss, vs. angle of
stalled cascade performance investigation is more limited in the
attack. open literature (see Yocum, 1988 for a specific bibliography).
324 N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335

ing to the fully mixed flow, where the outlet angle is ˛3 :


S
0
VX,2 VY,2 dyS
tan (˛3 ) =  2 (5)
S
0
VX,2 dy

The deflection is ˛1 − ˛3 . The losses ω3 are


S
0
(Pt,1 − Pt,3 ) dy
ω3 = (6)
(1/2)SV12

3.2. Representativeness of the tests of Yocum and O’Brien


compared to the compressors and axial circulators of GFR

3.2.1. Parameters to consider


Turbomachine similarity studies performed have lead to under-
line the following parameters to be investigated (see for example
Tauveron et al., 2006):

1. ratio of specific heats ,


2. Mach number based on axial velocity Ma ,
3. Mach number based on peripheral velocity MU (or the flow coef-
Fig. 4. Geometry of the blade cascade.
ficient ˚ if Ma is satisfied as Ma /MU = ˚),
4. Reynolds number based on the chord Re,
5. pressure coefficient  ,
3.1. Experimental data
6. solidity .
We present only a limited view of the work. Cascades of a sin-
gle blade geometry and a solidity matching unity were studied for As cascade data do not concern parameters based on the periph-
three stagger angles and a wide range of angles of attack extending eral velocity (MU or ˚,  ), we will replace them by the stagger angle
well into the stalled flow regime. Results from velocity and pres-  and the camber angle  c .
sure measurements made in the cascade and the overall cascade
performance evaluated from these measurements are presented. 3.2.2. Reference design for the GFR
In addition, results from a numerical simulation are also available. For the direct cycle, we consider the GFR design given by (Bassi,
These measurements are essentially time averaged. As a conse- 2003) and the turbomachinery design given by (Paniagua and Van
quence other “similar” configurations (isolated airfoil, flat plate) den Braembussche, 2003); three states are considered: high pres-
results are also used in the present paper. sure (7 MPa), medium (2.5 MPa) and low (.7 MPa). For the indirect
Yocum and O’Brien define (and we will use their definitions, see cycle, we consider GFR design as published in (Garnier et al., 2006)
Fig. 4) the outlet angle ˛2 : and the turbomachinery design performed at CEA (close to MHTGR
and Fort-Saint-Vrain design, see Mac and Yampolsky, 1987); four
S
VX,2 VY,2 dy states are studied: high (7 MPa), medium (.5 and 1 MPa) and low
0
tan (˛2 ) = S 2 dy
(3) (.1 MPa) pressure.
0
VX,2
3.2.3. Comparisons
where S is the blade spacing,  is the gas density. The deflection is
Table 1 presents comparisons between Yocum and O’Brien data,
˛1 − ˛2 . The loss of total pressure (Pt ) is
low pressure compressor of the Brayton cycle of direct cycle GFR
S   and the axial circulator for primary flow in indirect cycle GFR.
S
0
(Pt,1 − Pt,2 )(VX,2 )/ 0
VX,2 dy) dy We do not detail what relates to the high pressure compressor
ω2 = (4) of the Brayton cycle of direct cycle GFR because it is very similar
(1/2)V12
with the low pressure compressor. The results show that the ranges
Mixing wake considerations (for rotating to static frames transition) investigated by the tests of Yocum and O’Brien correspond rather
conducted to define a hypothetical (3) stage (see Fig. 4), correspond- well to the situations of medium and low pressure of GFR.

Table 1
Comparisons of various parameters between Yocum and O’Brien data, low pressure compressor of the Brayton cycle of direct cycle GFR and the axial circulator of indirect
cycle GFR

Parameter Data: Yocum and O’Brien GFR

Direct cycle Indirect cycle

7 MPa 2.5 MPa .7 MPa 7 MPa 1 MPa .5 MPa .1 MPa

 1.4 1.67
 25◦ , 36.5◦ , 45◦ 16◦ (stator) 47◦ (rotor) 17◦ (stator) 43◦ (rotor)
c 23◦ 30◦ (stator) 15◦ (rotor) 31◦ (stator) 30◦ (rotor)
 1 0.75 1 (stator) 1.3 (rotor)
Re 2 × 105 2 × 106 7 × 105 2 × 105 2 × 106 3 × 105 1.5 × 105 3 × 104
Ma 0.14 0.15 0.17
N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335 325

4. CFD simulations resolved thanks to a RANS approach1 and the outer layer is a LES
region. Such methods can be appropriate in turbomachinery appli-
4.1. Presentation cations (de Saint-Victor et al., 2005; Daeninck et al., 2006) and
for nuclear safety issues (CSNI, 2007). Another particular hygrid
In gas cooled reactor applications, compressor flows are char- (URANS in the near wall region: LES away from walls) implementa-
acterized by a quite high Reynolds number (see Table 1 to have an tion is Scale-Adaptative Simulation which improves prediction of
order of magnitude for Reynolds number), and they are also often unsteady flows, compared to URANS results (CSNI, 2007).
turbulent and wall bounded flows. In particular for stall regimes In order to highlight the benefit of the hydrid methodology
the near wall unsteadiness of the flow is usually a crucial feature (URANS: LES) we can recall general results on present computa-
for determining performance of the cascade. Unsteady Reynolds tional capabilities relative to wall-bounded flows. Moreover, these
Averaged Navier–Stokes approach (URANS) is commonly used, but general results will show that the methodology is relevant to treat
is only able to take into account low frequency features of the flow. some unstaedy problems concerning the safety analysis of nuclear
Large-Eddy Simulation (LES) seems to be a relevant technique to reactors (CSNI, 2007). Chapman (1979) was probably the first to
perform flow simulations able to capture accurately the major part emphasize the Reynolds number dependency of the mesh require-
of the unsteadiness of the flow (see Section 5.1.2 to have an order ments near the wall. He evaluated that the mesh resolution should
of magnitude of frequencies). The part of the turbulent spectrum scale as Re18 in the inner layer of the boundary layer, and as Re0.4
that is explicitely simulated by LES is much more extended than the in the outer layer. Piomelli and Balaras (2002), using the same esti-
part simulated by URANS; the other part is covered by special mod- mation tools, consider that in a boundary flow case, if the Reynolds
els (see CSNI, 2007). As a result, it has been proved that RANS and number is about one million, 99% of the computational nodes
URANS simulations gave very limited results compared to experi- should be used to compute the inner layer given that this latter
mental data on flow past bluff bodies (Rodi et al., 1997; Cummings represents only 10% of the boundary layer. Moreover, Baggett et
et al., 2003), which was not the case for LES (Shah, 1998; Guénot al. (1997) estimate that the required LES mesh for a plane channel
and Aupoix, 2003). We will illustrate this fact for our particular flow scales as Re 2 .2 The friction Reynolds number can be related
application in Section 5. to the “classical” Reynolds number thanks to the Dean correlation
Wall-resolved LES requires a tremendous mesh refinement in (Benarafa, 2005). According to this criteria and Dean correlation,
the near wall region, which implies a prohibitive computational a flow with a Reynolds number of one hundred million requires
cost, especially for parametric simulations. In order to alleviate this approximately a mesh of 1012 nodes to be computed. Even with the
mesh constraint, many research works have been performed to cre- nowadays computer power, a simulation with a mesh of 109 nodes
ate wall models (see Piomelli and Balaras, 2002; Sagaut, 2003 for an remains very impressive (Benarafa, 2005). As a consequence, wall-
exhaustive review). These latters are meant to approximate the wall resolved LES applied to wall bounded flow configurations for high
boundary conditions, allowing this way, the use of coarser meshes. Reynolds numbers implies a prohibitive computational cost. This
Wall models for LES can be divided into two categories. The first explains the interest of hydrid methodology (URANS: LES) such as
one concerns the wall models based on equilibrium laws such as TBLE.
the logarithmic law for example. The other one consists on using
a different approach (most of the time a (U)RANS approach) in the 4.2. Numerical setup and modelling framework
near-wall region to provide accurately the wall shear stress to the
coarse LES mesh. 4.2.1. Governing equations
Introducing a wall model called “two-layer model” or “TBLE In this study, the flow is considered incompressible (as sug-
(Thin Boundary Layer Equations)”, Balaras (Cummings et al., 2003) gested by Yocum and O’Brien, see also Mach number value in
embedded a one-dimensional grid between the wall and the Table 1: 0.14) and turbulent so that the mass conservation and the
first LES computation node. A RANS equation is solved in this momentum conservation filtered equations can be expressed as
inner mesh (Eq. (13), see Fig. 5). As an upper boundary condi- follows (Benarafa, 2005):
tion, the outer LES mesh supply the inner mesh with the velocity
and the pressure gradient of the first LES node. In return, the ∂V j
=0 (7)
TBLE inner mesh using a non-slip condition at the wall provide a ∂xj
wall shear stress to the LES outer mesh. The key of this method
is that the pressure gradient is assumed to be constant in the ∂V̄i ∂V i V j 1 ∂P ∂
+ =− +2 (( + sgs )S ij ) (8)
fine mesh which avoids Poisson equation inversion for incom- ∂t ∂xj  ∂xi ∂xj
pressible flows, hence a huge reduction of the computational  
cost. 1 ∂V i ∂V j
with S ij = + (9)
We must also mention a second attempt of the same kind is 2 ∂xj ∂xi
the Detached Eddy Simulation (DES) used as a wall model (Spalart,
1999; Nikitin et al., 2000). In this method, the inner layer is where (·) is LES filter operator. sgs is a subgrid-scale viscosity.  is
the molecular viscosity.
As for the turbulence models, we generally used Smagorinsky
model (Sagaut, 2003). The results for isolated airfoils submitted
to high incidence angle flows seem satisfactory (Cummings et al.,
2003) with this model. However, alternative choices can be men-
tioned for the wall model (Strelets, 2001) or for the turbulence
model (Daeninck et al., 2006).

1
No one-dimensional model and grid are used, as for TBLE model.
2
Fig. 5. Near-wall TBLE one-dimensional mesh and unstructured LES mesh (from Re␶ is the friction Reynolds number based on the friction velocity u␶ , the half
Benarafa, 2005). channel height h and the molecular viscosity .
326 N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335

Table 2 cell as it is shown in Fig. 5 instead of localizing it between the veloc-


Boundary conditions
ity nodes and the wall. On this mesh, the solved equations are the
Edge Type of boundary conditions following.
Upstream Velocity imposed (two or three components)
For the TBLE velocity field i = x or z:
Downstream Pressure imposed (zero)  
∂Vi ∂ ∂Vi ∂P
Blade Non-slip − ( + t ) =− (13)
Upper and lower Periodical ∂t ∂y ∂y ∂xi
Front and back (only for 3D) Periodical
The turbulent viscosity t is evaluated thanks to a mixing length
model with the Van Driest damping function Dvd 4 :
4.2.2. Numerical procedure  2  2
1/2
All the computations of this study were achieved with the ∂VX ∂VZ
Trio U code. This object oriented code solves Eqs. (7) and (8) in t = Dvd ( y)2 + (14)
∂y ∂y
a mixed finite volume/finite element approach for both struc-
tured and unstructured grids (Calvin et al., 2002). For the present Various boundary conditions have been tested at the interface
study, unstructured grids are considered with tetrahedral elements. between the inner TBLE one-dimensional grid and the outer LES
The unknown variables are the velocity and the pressure. Velocity mesh. Balaras et al. (1996) proposed to use the velocity of the first
unknowns are located at the center of the faces of the tetrahedral node velocity of the first LES node as boundary conditions and the
elements, whereas the pressure unknowns are located at the centre pressure gradient as a given source term. In addition to these lat-
and the vertices of the elements. The discretization method ensures ters, Diurno et al (2001) attempted to impose the continuity of the
that the fluxes through the surfaces of each cell are continuous. As a shear stress or the one of the turbulent viscosity which they con-
consequence conservation of main physical parameters is guaran- clude to be a better choice for numerical stability reasons. Wang
teed (which is not the case for finite element methods). The discrete and Moin (2002), to study a trailing edge flow, also connected the
formulation of the equations is solved using a matrix projection subgrid-scale viscosity to the TBLE turbulent viscosity at the inter-
scheme (similar to the method presented in Hirt et al (1975)). face between the LES outer mesh and the TBLE inner mesh.
Time advancement was ensured by a second order In the present TBLE implementation, the velocity of the first LES
Crank–Nicholson implicit scheme. For the momentum Eq. (8), node is interpolated at the center of the cell to compute the upper
we use a centred second order scheme for convection and diffusion boundary conditions for the TBLE mesh.
terms. An additional term due the discretization method ensures
the stability of the simulations (Heib, 2003). 5. Simulation on a configuration with a high angle of
attack (45◦ )
4.2.3. Mesh and boundary conditions
Table 2 indicates the boundary conditions for the simulations. This configuration has been chosen as full stall has occurred
and as it is extensively described in the publications by Yocum and
4.2.4. Wall modelling O’Brien.
4.2.4.1. Wall models based on equilibrium laws. Most of common
wall models (as in other CFD codes, see Benarafa, 2005) are based 5.1. 2D simulations
on equilibrium laws such as the logarithmic law (Grötzbach, 1987):
5.1.1. k−ε models
V + = y+ ; if y+ < 5 (viscous sublayer) (10)
Even if there is a strong limitation of the turbulence description
 y+
with the use of URANS simulations, k−ε computations are per-
2 d˛ formed. As suggested in CSNI (2007) results obtained with different
V+ =  ; if 5 < y+ < 30
0 1+ 1 + 4lm2+ (˛) discretization schemes are employed to provide a rough estimation
  ˛  of numerical error: k−ε simulations are performed with different
with lm+ (˛) = ˛+ 1 − exp − (buffer layer) (11) numerical schemes (first order, second order: see Table 3). All of
26 them suffer from difficulties to correctly predict the losses, which
1 is a well-known fact for (U)RANS for isolated airfoil cases (Section
V+ = log (y+ ) + A; if y+ > 30 (logarithmic layer) (12) 4.1). The use of less diffusive schemes (than the first order) improves

the quality of the results, as such models supports dissipation and
The wall Cartesian coordinates are expressed in wall units: (x+ , y+ , stall. However with such models, filtering the eddy viscosity was
z+ ) = (xv /, yv /, z v /), and the velocity is V + = V /v where v is necessary, which is a strong disadvantage. The value of filter (t,filt )
the friction velocity defined as the square root of wall shear stress can also be set in doubt.
w divided by the density .  is the kinematic viscosity of the
fluid, P is the pressure. is Von Karman constant, A is a numerical 5.1.2. “LES” models
constant of the model.3 However, debatable is the question of using LES-similar model in
2D, some authors have shown that such models can provide accept-
4.2.4.2. Velocity TBLE wall model. The TBLE wall model has been able results on complex configurations (Hassan and Ibrahim, 1997;
developed for the first time by Balaras et al. (1996). This wall model Bouris and Bergeles, 1999).
relies on the embedding of a one-dimensional mesh where RANS
Thin Boundary Layer Equations are solved to supply the LES compu- 5.1.2.1. Unsteadiness. The first characteristics of the flow is
tation with the wall shear stress. In return, the LES first computation unsteadiness. We see in Fig. 6 the shedding of two vortices, at the
node will impose the upper boundary condition. It was chosen in
this study to embed the one-dimensional mesh at the center of the
  3 
+ y
4
Dvd = 1 − exp − w
25
.
3
We have used: A = 5.32, = 0.415.
N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335 327

Table 3
Cascade performance with k−ε model

˛2 (◦ ) ˛1 − ˛2 (◦ ) ω2 ˛3 (◦ ) ˛1 − ˛3 (◦ ) ω3

Experimental data 31 39 0.46 50 20 0.77


First order 21.5 48.5 0.30 51.6 18.4 0.41
Second order (centered), t < t,filt /2 35.2 34.8 0.43 51.6 18.4 0.59
Second order (centered), t < t,filt 34.2 35.8 0.43 50.7 19.3 0.60
Second order (Muscl), t < t,filt 25.7 44.3 0.39 46.6 23.4 0.59

Table 4
Cascade performance, simulations performed with various options

˛2 (◦ ) ˛1 − ˛2 (◦ ) ω2 ˛3 (◦ ) ˛1 − ˛3 (◦ ) ω3

Experimental data 31 39 0.46 50 20 0.77


Standard case 28.6 41.4 0.80 42.0 28.0 1.11
Slow time marching 28.7 41.3 0.79 42.4 27.6 1.10
Second order convection scheme 27.5 42.5 0.89 37.5 32.5 1.11
WALE model 25.6 44.4 0.93 36.7 33.3 1.11
TBLE model 28.5 41.5 0.84 41.6 28.4 1.16

leading edge and at the trailing edge of the blade profile. They have 5.2. 3D simulations
an opposite sign and the same frequency; their birth is in opposition
of phase. From a quantitative point of view their intensity is approx- An important characteristic of the flow is its tridimensionality.
imately identical and their period is five times the period between
two figures, which is 0.0008 s. These points are coherent with Sarp-
kaya views (Sarpkaya, 1975). The size of each vortex approximately 5.2.1. Tridimensionality
respects the order of one tenth of the length scale, here the blade We first give results of the simulations performed with a small
spacing (Rodi, 1980). spansize (7% of the cord) in Table 5. They do not improve the quality
The literature on the subject indicates that Strouhal number St of results.
characterizing this vortex shedding on a purely flat plate at high These results are coherent with simulations performed with a
incidence is approximately equal to 0.15–0.17 (Fage and Johansen, larger spansize (Figs. 9 and 10): we find small 3D structures and a
very developed 2D structure, close to that described in (Guénot and
1927; Abernathy, 1962). We have adopted the  same formalism as Aupoix, 2003) for an isolated airfoil. In Fig. 10, we note structural
Sarpkaya, showing spectral evolution of ∂/∂t V dl. We have seen
in the simulations the frequency of 195 Hz. Expressed in Strouhal elements similar to those described by (Naijar and Vanka, 1995)
number, we have and (Hussain and Hayakawa, 1987) for a flat plate and a cylinder:
“rib”-like structures between two adjacent spanwise rollers. These
fc sin (˛) structures occur in pair of opposite vorticity. They show the impor-
St = (15)
V tance of the tridimensionality of the flow and the necessity of 3D
where f is the frequency, c is the cord, ˛ is the incidence angle on simulations.
the plat, V is the upstream velocity. We find 0.18 for St, showing
the same tendency as those experimentally obtained with an iso-
lated flat plate (Yocum and O’Brien, 1993 did not measure these 5.2.2. Time average fields
Fig. 11 shows the time averaged velocity field. It clearly illus-
frequencies).
trates the “immediate” full stall at the leading edge, for the
suction side and the attached flow for the pressure side. The
5.1.2.2. Time average fields. Fig. 7 shows the time averaged velocity
size of the detached zone is important (more than half of the
field. It clearly illustrate the “immediate” full stall at the leading
“outlet” section), but less important than in 2D simulations
edge, for the suction side and the attached flow for the pressure
(Fig. 7).
side. The size of the detached zone is important (more than the
Fig. 12 shows the time averaged (Pt,1 − Pt )/(1/2)V12 field. On the
half of the “outlet” section). Such zone is a zone of recirculation
suction side we see a zone with large losses (detached zone). The
(backflow due to adverse pressure gradient) and of large losses.
attached (potential) zone shows a lower value of losses than in 2D
Upstream, we see a mixing zone between detached and attached
(Fig. 7), but not completely equal to zero.
flows.
Fig. 7 shows the time averaged Pt,1 − Pt /(1/2)V12 field. The zero
value is the upstream zone. On the suction side we see a zone with 5.2.3. Results
large losses (detached zone). The attached (potential) zone shows The angular results in Table 6 are of good quality, as for this
a very low value of losses, but not equal to zero. The mixing zone is stalled regime the difference between simulations and measure-
also well illustrated in Fig. 8. ments is only 2.6◦ or 0.4◦ . For the stage (3) the difference is limited
to 3.4◦ or 2.1◦ , which is a significant improvement if we compare
5.1.2.3. Cascade performance. LES simulations are performed with to 2D results. This can be explained by the definitions of ˛2 and
different numerical and physical features (time marching man- ˛3 (Eqs. (3) and (5) and Table 7). All the simulations in LES (2D,
agement, convection scheme, turbulence model, wall function, see 3D standard wall function, 3D TBLE) underestimate the size of the
Table 4). All of them overestimate the losses and the deflection, S S 2 dy are
attached zone, as a consequence VX,2 VY,2 dy and VX,2
which is a coherent with observations on an isolated airfoil (Guénot S 0 0
and Aupoix, 2003). The overestimation of losses is coherent with underestimated. As 0 VX,2 dy is correct (mass conservation), ˛3 is
some remarks about Fig. 7, where the attached zone has a signifi- less correctly estimated than ˛2 . When the attached zone is well
cant loss of head. The use of various options does not significantly estimated (3D TBLE > 3D standard wall function > 2D), the angles
improve the quality of the results. are better estimated.
328 N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335

Fig. 6. Instantaneous vorticity with velocity (arrows).

As far as losses are concerned, simulations results overestimate


ω2 and ω3 . The classification of models is the same as for angles:
the better estimation is performed with 3D TBLE. It seems that Table 5
Cascade performance in 2D and Pseudo-3D
the overestimation of the detached zone size is directly respon-
sible of the overestimation of losses. We go further in detail with ˛2 (◦ ) ˛1 − ˛2 (◦ ) ω2 ˛3 (◦ ) ˛1 − ˛3 (◦ ) ω3
Figs. 13 and 14. On the total pressure loss ((Pt,1 − Pt,2 )(y)/(1/2)V12 ) Experimental data 31 39 0.46 50 20 0.77
we see that the detached zone (i.e. zone with high pressure losses) 2D 28.6 41.4 0.80 42.0 28.0 1.11
is more extended in the simulations, which is coherent with the Pseudo-3D 28.3 41.7 0.88 41.3 28.7 1.16
N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335 329

Fig. 7. Time averaged velocity (2D).

Fig. 8. Non-dimensional total pressure loss (2D).

previous observations. We also see that the difference is a kind of up to “zero axial flow” regimes. Secondly, we investigate solidity
“constant low shift” between simulations and experimental data influence which is usually considered as an important parameter
(Fig. 13), especially in the attached zone. This could be explained for losses. At the present time we have not investigated stagger
by numerical effects of the stabilization technique (Section 4.2.2). angle and Reynolds number influence as experimental data partly
As far as the static pressure difference ((Pt,1 − Ps,2 )(y)/(1/2)V12 ) treated these aspects. Camber angle influence will be treated in the
is concerned, we see a good agreement between simulations and future.
measurements (Fig. 13). Some simple analytical developments lead
to
6.1. Investigation of a larger range of angle of attack
S S 2 dy 2 2
(Pt,1 − Ps,2 ) dy 0
VX,2 VX,3 VY,3
0
ω3 = −2 + − (16) When the angle of attack is increased the recirculation zone
(1/2)SV12 SV12 V12 V12 becomes more extended: this is clear if we compare Figs. 15 and 11.
We see in this equation that the overestimation of losses can again If we consider the results obtained on cascade performance
S S 2 dy (Table 8), we first notice that the fact that ˛2 ∼  (and as a con-
be explained by underestimation of VX,2 VY,2 dy and VX,2
0 0 sequence ˛1 − ˛2 ∼ ˛) seems approximately valid. This is certainly
(Table 7).
due to the facts that near the pressure surface of the blades, the
6. Use of CFD simulations velocity remains parallel to the chord line (Fig. 15), i.e. following
 and that the recirculation zone has a very small contribution
S S 2 dy, due to the small velocity.
As LES model with TBLE wall modelling gives satisfying results in 0 VX,2 VY,2 dy and in 0 VX,2
compared to experimental data, we can explore a wider range of The behaviour of ˛1 − ˛3 seems qualitatively logical (an increase
parameters variations than those investigated by the experimental followed by a decrease): it is a well-known fact that for an iso-
data. First we investigate the angle of attack up to very high values, lated airfoil the circulation (and lift) decreases when stall occurs
330 N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335

Fig. 9. Surfaces of instantaneous spanwise vorticity.

Table 6 and develops. As far as losses are concerned, the increase of


Cascade performance with LES (two wall modelling)
both ω2 and ω3 can be explained by the size of the recirculation
˛2 (◦ ) ˛1 − ˛2 (◦ ) ω2 ˛3 (◦ ) ˛1 − ˛3 (◦ ) ω3 zone. On this last point also, the fact that a value (not equal to
Experimental data 31 39 0.46 50 20 0.77
unity) seems reached at higher angles of attack is more surpris-
LES SWF 28.4 41.6 0.74 46.6 23.4 0.98 ing, but can be explained by Cornell models (described in Section
LES TBLE 31.4 38.6 0.65 47.9 22.1 0.95 2.2).

Fig. 10. Surfaces of instantaneous streamwise vorticity.


N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335 331

Fig. 11. Time averaged velocity.

Fig. 12. Non-dimensional total pressure loss with TBLE model.

Table 7
Cascade results with LES (two wall modelling)
S  S 
S S S S (Pt,1 − Pt,2 )(VX,2 )/ VX,2 dy
VX,2 VY,2 dy VX,2
2
dy (Pt,1 − Ps,2 ) dy (Pt,1 − Pt,2 ) dy 0 0
S
0
= VY,3 (m/s) 0S (m/s) 0
(1/2)SV12
0
(1/2)SV12 (1/2)SV12
dy
VX,2 dy VX,2 dy
0 0

Experimental data 19.5 32.5 1.37 0.9 0.46


LES SWF 17.3 32.1 1.45 1.12 0.74
LES TBLE 17.85 28.63 1.37 1.08 0.63

6.1.1. Conclusion 6.2. Preliminary study of the influence of solidity


The points we have added with this study are shown in Fig. 16.
We can notice that Longley model with constant area mixing One of the most important parameter to consider for turbo-
(called “Longley1” in Fig. 16) gives a decreasing curve for the machinery performance is the solidity (see Section 2.1). We recall
total pressure loss for high angles of attack, which is the same that solidity is the ratio between the the chord size and the spac-
trend as the curve given by CFD simulations. Moreover, the final ing between adjacent blades. However, the solidity dependency
value of cos2 () seems to be a reasonable limit value for CFD on losses is not easy to forecast: on the one hand a high solidity
results. We must also mention that the CFD simulations results implies a large number of blades and as a consequence larger fric-
must be carefully considered, as the typical error range is indi- tion losses, on the other hand a low solidity implies a wide space
cated by the difference between CFD and experimental data for between blades and as a consequence a tendency for the flow to
˛ = 45◦ . separate at the suction side. We performed simulations with three
332 N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335

Table 8
Cascade performance with four different angles of attack

˛2 (◦ ) ˛1 − ˛2 (◦ ) ω2 ˛3 (◦ ) ˛1 − ˛3 (◦ ) ω3

˛ = 30 25.5 29.5 0.355 30 25 0.5
˛ = 50◦ 30.9 44.1 0.66 57.4 17.6 0.98
˛ = 55◦ 29 51 0.64 68.5 11.5 0.88
˛ = 60◦ 27 58 0.82 73.5 11.5 0.89

Fig. 13. Non-dimensional downstream total pressure loss.

Fig. 16. Cross-comparisons between experimental data from Yocum and O’Brien,
traditional correlations, specific analytical models and CFD simulations: total pres-
sure loss, vs. angle of attack.

size would have also affected Reynolds number, blade geometry


and blade mesh. As these different variations would be difficult
to analyse, we decided to change spacing. However, the doubling
of spacing leads to a doubling of the grid size if we do not want
to change mesh refinement. As LES are large CPU consuming, we
decided, at a first step, to consider k−ε simulations. Due to the con-
siderations about the analytical models (Section 2.2) we present
separately the cases  ≥ 1 and  ≤ 1.
Fig. 14. Non-dimensional downstream static pressure difference.

different spacings, the second is the half of the first, and the third 6.2.1.  ≥ 1
is twice the first, which is a quite large range of variation for such a If we compare Figs. 17 to 11, we see that the main stall phe-
parameter. For solidity variation we had the choice between: chang- nomena are quite similar, with the development of a half zone of
ing the chord size and changing the spacing. Changing the chord recirculation. The flow is better conducted with a larger number of

Fig. 15. Time averaged velocity with ˛ = 60◦ .


N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335 333

Fig. 17. Time averaged velocity with  = 2.

Fig. 18. Time averaged velocity with  = 1/2.

blades. In the case of smaller spacing, the extent of the recirculation model is quite weak. Phenomena are not governed by friction, but
zone seems more limited. The downstream zone seems also more by the separation process.
homogeneous, which is normal as we have four blades instead of
two. 6.2.2.  ≤ 1
We see in Table 9 that results are quite similar, with a lower If we compare Figs. 18 to 11, we see that the main stall phenom-
extend of the stall zone for the lower spacing, as suggested before. ena are qualitatively similar, with the development of a wide zone
In a first correlation building, it can be supposed that the solidity of recirculation. The extent of the recirculation zone seems more
dependency is weak. important in the case  = 1/2. Fig. 18 seems also comparable with
We can conclude that Longley models give total pressure loss isolated airfoil simulations results (Hoarau and Braza, 2005), if we
independent from solidity which is the same trend as the curve consider the recirculation zone size.
given by CFD simulations, whereas Moses and Thomason gives a If we recall that (see Moses and Thomason, 1986) the adapted
contrary trend. In fact solidity dependence in Moses and Thomason normal coefficient Cn , characterising the lift Fn , is expressed by
Fn 2 V2
Cn = − = cos (˛1 )(sin (˛1 − ) − sin (˛2 − ))
V12 c/2  V1
Table 9
Cascade performance with two different spacings P2 − P1
+ sin() (17)
Reference configuration ˛2 (◦ ) ˛1 − ˛2 (◦ ) ω2 ˛3 (◦ ) ˛1 − ˛3 (◦ ) ω3 V12 /2
 =1 35.2 34.8 0.43 51.6 18.4 0.59 We find that Cn ( = 1) ∼ 0.45, Cn ( = 0.5) ∼ 0.85. In the case of an
 =2 37.2 32.8 0.39 51.35 18.65 0.49
isolated airfoil, Cn ∼ 1.3 (from Guénot and Aupoix, 2003). The
334 N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335

Table 10 Chapman, D.R., 1979. Computational aerodynamics development and outlook. AIAA
Cascade performance with two different spacings Journal 17, 1293–1313.
CSNI (Committee on the Safety of Nuclear Installations), 2007. Best practice guide-
Reference configuration ˛2 (◦ ) ˛1 − ˛2 (◦ ) ω2 ˛3 (◦ ) ˛1 − ˛3 (◦ ) ω3 lines for the use of CFD in nuclear reactor safety applications. Technical Report.
 =1 35.2 34.8 0.43 51.6 18.4 0.59 Nuclear Energy Agency.
Cornell, W.G., 1954. The stall performance of cascades. In: Proceedings of the ASME
 = .5 37.4 32.6 0.68 47.3 22.7 0.8
Applied Mechanics Conference, p. 21.
Cummings, R.M., Forsythe, J.R., Morton, S.A., Squires, K.D., 2003. Computational chal-
lenges in high angle of attack flow prediction. Progress in Aerospace Sciences
39, 369–384.
behaviour of a cascade with a wide spacing looks like the behaviour Daeninck, G., Medic, G., Templeton, J.A., Kalitzin, G., 2006. On near-wall dynamic
of an isolated airfoil, which is quite logical. coupling of LES with rans turbulence model. In: Proceedings of the ASME
Turbo’Expo.
We see in Table 10 that results are different, as suggested before. de Saint-Victor, X., Houdeville, R., Aupoix, B., 2005. Simulations URANS et DES pour
In a first correlation building, it can be supposed that the solidity des applications à des turbomachines. In: Procedings of the 40ème colloque
dependency is more important when  ≤ 1. d’aérodynamique appliquée.
Diurno, G.V., Balaras, E., Piomelli, U., 2001. In: Geurts, B. (Ed.), Mathematical Princi-
ples of Classical Fluid Mechanics. RT Edwards, PA, pp. 207–222.
7. Conclusion Dixon, S.L., 1975. Fluid mechanics Thermodynamics of Turbomachinery, 2nd ed.
Pergamon Press.
Fage, A., Johansen, F.C., 1927. On the flow of air behind an inclined flat plate of infinite
The final objective of the study is to provide models or correla- span. Proceedings of the Royal Society 116, 170–180.
tions, adapted to describe compressor behaviour in very off-design Garnier, J.-C., Bassi, C., Blanc, M., Bosq, J.-C., Chauvin, N., 2006. Contribution to GFR
design option selection. In: Proceedings of the 2006 International Conference
conditions (stall conditions) which can be used in our compressor on Advances in Nuclear Power Plants.
performance generation model. In this paper, we have established Grötzbach, G., 1987. Direct and large Eddy simulation of turbulent channels flows.
and begun to validate the numerical simulations with Trio U soft- In: Cheremisinoff, N.P. (Ed.), Encyclopedia of Fluids Mechanics. Gulf Publ, New
Jersy, pp. 1337–1391.
ware by comparison with representative experimental data on stall
Guénot, D., Aupoix, B., 2003. Etude de l’approche DES pour prévoir les écoulements
regime (for profile phenomena). Large Eddy Simulations with a instationnaires à grande échelle. In: Proceedings of the 16ème Congrès Français
two-layer approximate boundary conditions (TBLE) has given the de Mécanique.
Hassan, Y.A., Ibrahim, W.A., 1997. Turbulence prediction in 2D bundle flows using
most precise results compared to the open literature case (in air).
les. Nuclear Technology 119, 11–19.
Simulations with a larger span size domain will be performed to Heib, S., 2003. Nouvelles discrétisations non structurées pour des écoulements
improve the quality of simulations. Numerical aspects will also be de fluides à incompressibilité renforcée. PhD Thesis. Université Paris VI,
investigated. p. 22.
Hirt, C.V., Nichols, B.D., Romero, N.C., 1975. SOLA—a numerical solution algorithm
However, we have used the present numerical model to for transient flow. Technical Report Report LA-5852. Los Alamos National Labo-
perform sensitivity studies on various parameters of inter- ratory.
est (solidity, angle of incidence) to improve the experimental Hoarau, Y., Braza, M., 2005. Simulation d’écoulements turbulents autour d’une aile:
comparaison des méthodes OES et DES. In: Proceedings of the 40ème colloque
database. Enrichment of database shows analytical model with d’aérodynamique appliquée.
constant section mixing trends are valuable. The limited impact Hubert, G., Bauermeister, K.J., 1963. Probleme der sekundärströmung in axialen
of solidity when the solidity is greater than unity is also estab- turbomaschinen. VDIForschungsheft 29(B), 1–31.
Hussain, F., Hayakawa, H., 1987. Eduction of large scale organized structures in a
lished. turbulent plane wake. Journal of Fluid Mechanics 180, 161–181.
In the future, other parameters influence will be investigated to Kikstra, J.F., 2001. Modelling, design and control of a cogenerating nuclear gas turbine
enrich this data base (wider ranges of incidence, stagger and cam- plant. PhD Thesis. Delft University of Technology.
Koch, C.C., Smith, L.H., 1976. Loss sources and magnitudes in axial-flow compressors.
ber angles, Reynolds number). Negative velocities are also of great
ASME Journal of Engineering for Power, 411–424.
interest for simulating stall and surge transients. A large amount Lakshminarayana, B., 1970. Methods of predicting the tip clearance effects on
of work is still required to evaluate the importance of additional axial flow turbomachinery. ASME Journal of Basic Engineering 92, 467–
482.
effects (3D, role of tip clearance, etc.) and finally simulate stall in a
Lieblein, S., 1959. Loss stall analysis of compressor cascades. Transactions of ASME,
compressor. 387–400.
Lieblein, S., 1960. Incidence and deviation-angle correlations for compressor cas-
cades. Transactions of ASME, 575–587.
References Longley, J.P., 2007. Calculating stall and surge transients. In: Proceedings of the
International Gas Turbine and Aeroengine Congress and Exposition. Paper
Abernathy, F.H., 1962. Flow over an inclined plate. ASME Journal of Basic Engineering GT2007-27378.
84, 380–388. Mac Donald, C.F., 1987. Helium compressor aerodynamic design considerations for
Baggett, J.S., Jimenez, J., Kravchenko, A.G., 1997. Resolution requirements in large- MHTGR circulators. In: IAEA Specialists’ Meeting on Gas Cooled Reactor Coolant
Eddy simulations of shear flows. Annual Research Briefs: Center of Turbulence Circulator and Blower Technology.
Research. Moses, H.L., Thomason, S.B., 1986. An approximation for fully stalled cascades. Jour-
Balaras, E., Benocci, C., Piomelli, U., 1996. Two-layer approximate boundary condi- nal of Propulsion, 188–189.
tions for large-eddy simulations. AIAA Journal 34, 1111–1119. Naijar, F.M., Vanka, S.P., 1995. Effects of intrinsic 3D on the drag characteristics of a
Bammert, K., Zehner, P., 1975. Back flow in turbines of nuclear closed-cycle gas tur- normal flat plate. Physics of Fluids 7, 2516–2518.
bine plants in case of circuit pipe rupture. ASME Journal of Engineering for Power Nikitin, N.V., Nicoud, F., Wasistho, B., Squires, K.D., Spalart, P.R., 2000. An approach
97, 189–194. to wall modeling in large-Eddy simulations. Physics of Fluids 12, 349–
Bassi, C., 2003. Transient behaviour of a gas cooled fast reactor. In: Proceedings of 374.
the 2003 International Conference on Advances in Nuclear Power Plants. Ottavy, X, 2004. Cours de turbomachines à fluides compressibles. Technical Report.
Benarafa, Y., 2005. Application du couplage RANS/LES aux écoulements turbulents INPG.
à haut nombre de Reynolds de l’industrie nucléaire. PhD Thesis. Université Paris Paniagua, G., Van den Braembussche, R.A., 2003. Aerodynamic study of the helium
VI. turbine and compressor of a scaled model of the GT-MHR 600 MWpower plant.
Bouris, D., Bergeles, G., 1999. 2D time dependent simulation of the subcritical flow Technical Report. VKI-Contract report 2003-19.
in a staggered tube bundle using a subgrid scale model. International Journal of Piomelli, U., Balaras, E., 2002. Wall-layer models for large-eddy simulations. Annual
Heat Fluid Flow 20, 105–114. Review of Fluid Mechanics 34, 349–374.
Brey, H.L., 1987. Fort St. Vrain circulator operating experience. In: IAEA Specialists’ Rodi, W., 1980. Turbulence models and their application in hydraulics. IAHR Delft.
Meeting on Gas-cooled Reactor Coolant Circulator and Blower Technology. Rodi, W., Ferziger, J.H., Breuer, M., Pourquie, M., 1997. Status of les: result of a work-
Calvin, C., Cueto, O., Emonot, P., 2002. An object-oriented approach to the design of shop. ASME Journal of Fluids Engineering 119, 248–262.
fluid mechanics software. Mathematical Modelling and Numerical Analysis 36, Sagaut, P., 2003. Large Eddy Simulation for Incompressible Flows, 2nd ed. Springer-
907–921. Verlag.
Cavallaro, L., Yampolsky, J., 1974. Design and development of a steam turbine driven Sarpkaya, T., 1975. An inviscid model of two-dimensional vortex shedding for tran-
circulator for high temperature gas-cooled reactors. Nuclear Engineering and sient and asymtotically steady separated flow over an inclined plate. Journal of
Design 26, 135–157. Fluid Mechanics 68, 109–128.
N. Tauveron / Nuclear Engineering and Design 240 (2010) 321–335 335

Shah, K.B., 1998. Large Eddy simulation of the flow past a cubic obstacle. PhD Tauveron, N., Saez, M., Ferrand, P., Leboeuf, F., 2007. Axial turbomachine modelling
Thesis. with a 1D axisymmetric approach—application to gas cooled nuclear reactor.
Spalart, P.R., 1999. Strategies for turbulence modeling and simulation. In: Pro- Nuclear Engineering and Design 237, 1679–1692.
ceedings of the Fourth International Symposium on Engineering Turbulence Trio U, 2007. http://www-trio-u.cea.fr.
Modeling and Measurements, pp. 3–17. Wang, M., Moin, P., 2002. Dynamic wall modeling for large-Eddy simulation of com-
Strelets, M., 2001. Detached eddy simulation of massively separated flow. AIAA, plex turbulent flows. Physics of Fluids 14, 2043–2051.
2001-0879. Yocum, A.M., 1988. An experimental and numerical investigation of the performance
Swan, W.C., 1961. A practical method of predicting transonic-compressor perfor- of compressor cascades with stalled flow. PhD Thesis. Virginia Polytechnic Insti-
mance. ASME Journal of Engineering for Power, 322–330. tute and State University.
Tauveron, N., Saez, M., Ruby, J.-P., Geffraye, G., Tenchine, D., Pialla, D., Germain, T., Yocum, A.M., O’Brien, W.F., 1993. Separated flow in a low-speed two-
Hervieu, E., 2006. Preliminary design of a small air loop for system analysis. In: dimensional cascade: Parts I and II. ASME Journal of Turbomachinery 115,
Proceedings of the Third International Topical Meeting on High Temperature 409–434.
Reactor Technology (HTR-2006).

You might also like