Sei sulla pagina 1di 15

Reynolds-averaged Navier–Stokes equations

Sodabunlu Ativich1, Hammarratchata Chayuttra2, Gabire Tseganeh Wabela3

Central South University, China

Abstract —Reynolds averaged Navier-Stokes physical laws on which mechanics as a science is


computations using several million grid points have based.... my attention drawn to various mechanical
become commonplace today. While many practical phenomena, for the explanation of which I
problems can be solved to acceptable accuracy with discovered that a knowledge of mathematics was
such methods at these resolutions, the drive to more essential. For the year immediately following his
complex problems and higher accuracy is requiring graduation from Cambridge he again took up a post
the solution of ever larger problems. For example,
with an engineering firm, this time as a practicing
the flow over aircraft configurations in off-design
civil engineer in the London (Croydon) sewage
configurations, such as high-lift, has been computed
with up to 25 million grid points, and cases transport system. In 1868 he was appointed
involving up to 108 grid points can be anticipated in professor of engineering at Owens College in
the near future. Manchester (now the University of Manchester),
Keywords: Reynolds-averaged, Navier-Stokes, becoming in that year one of the first professors in
Osborne Reynolds UK university history to hold the title of "Professor
of Engineering". This professorship had been newly
1. INTRODUCTION created and financed by a group of manufacturing
industrialists in the Manchester area, and they also
Before study RANS we should know about Osborne
had a leading role in selecting the 25–year–old
Reynolds who invent Reynolds-averaging and using
Reynolds to fill the position.
the Reynolds-averaged Navier–Stokes equations.
For the fluid mechanics, Reynolds most
Osborne Reynolds FRS (23 August 1842 – famously studied the conditions in which the flow of
21 February 1912) was a prominent Irish-born fluid in pipes transitioned from laminar
British innovator in the understanding of fluid flow to turbulent flow. In 1883 Reynolds
dynamics. Separately, his studies of heat transfer demonstrated the transition to turbulent flow in a
between solids and fluids brought improvements in classic experiment in which he examined the
boiler and condenser design. He spent his entire behavior of water flow under different flow rates
career at what is now the University of Manchester. using a small jet of dyed water introduced into the
He was born in Belfast and moved with his parents center of flow in a larger pipe.
soon afterward to Dedham, Essex. His father The larger pipe was glass so the behavior of the layer
worked as a school headmaster and clergyman, but of dyed flow could be observed, and at the end of
was also a very able mathematician with a keen this pipe there was a flow control valve used to vary
interest in mechanics. The father took out a number the water velocity inside the tube. When the velocity
of patents for improvements to agricultural was low, the dyed layer remained distinct through
equipment, and the son credits him with being his the entire length of the large tube. When the velocity
chief teacher as a boy. Reynolds showed an early was increased, the layer broke up at a given point
aptitude and liking for the study of mechanics. In his and diffused throughout the fluid's cross-section.
late teens, for the year before entering university, he The point at which this happened was the transition
went to work as an apprentice at the workshop of point from laminar to turbulent flow.
Edward Hayes, a well-known shipbuilder in Stony From these experiments came the
Stratford, where he obtained practical experience in dimensionless Reynolds number for dynamic
the manufacture and fitting out of coastal steamers similarity—the ratio of inertial forces
(and thus gained an early appreciation of the to viscous forces. Reynolds also proposed what is
practical value of understanding fluid dynamics). now known as Reynolds-averaging of turbulent
flows, where quantities such as velocity are
Osborne Reynolds attended Queens' College, expressed as the sum of mean and fluctuating
Cambridge and graduated in 1867 as the components. Such averaging allows for 'bulk'
seventh wrangler in mathematics.[1] He had chosen description of turbulent flow, for example using
to study mathematics at Cambridge because, in his the Reynolds-averaged Navier–Stokes equations.
own words in his 1868 application for the
professorship, "From my earliest recollection I have
had an irresistible liking for mechanics and the
Now we will come back for introduce a small converging nozzle. Such flows, whether
about RANS, The Reynolds-averaged Navier– exactly solvable or not, can often be thoroughly
Stokes equations (or RANS equations) are time- studied and understood.
averaged equations of motion for fluid flow. The 2. Turbulence => Turbulence is the time-
idea behind the equations is Reynolds dependent chaotic behavior seen in many fluid
decomposition, whereby an instantaneous quantity flows. It is generally believed that it is due to
is decomposed into its time-averaged and fluctuating the inertia of the fluid as a whole: the culmination of
time-dependent and convective acceleration; hence
quantities, an idea first proposed by Osborne
flows where inertial effects are small tend to be
Reynolds. The RANS equations are primarily used laminar (the Reynolds number quantifies how much
to describe turbulent flows. the flow is affected by inertia). It is believed, though
The Navier–Stokes equations, named not known with certainty, that the Navier–Stokes
after Claude-Louis Navier and George Gabriel equations describe turbulence properly.[19]
Stokes, describe the motion of viscous The numerical solution of the Navier–Stokes
fluid substances. Navier–Stokes equations are equations for turbulent flow is extremely difficult,
useful because they describe the physics of many and due to the significantly different mixing-length
phenomena of scientific and engineering interest. scales that are involved in turbulent flow, the stable
They may be used to model the weather, ocean solution of this requires such a fine mesh resolution
that the computational time becomes significantly
currents, water flow in a pipe and air flow around
infeasible for calculation or direct numerical
a wing. The Navier–Stokes equations, in their full simulation. Attempts to solve turbulent flow using a
and simplified forms, help with the design of aircraft laminar solver typically result in a time-unsteady
and cars, the study of blood flow, the design of solution, which fails to converge appropriately. To
power stations, the analysis of pollution, and many counter this, time-averaged equations such as
other things. Coupled with Maxwell's equations, the Reynolds-averaged Navier–Stokes
they can be used to model and equations (RANS), supplemented with turbulence
models, are used in practical computational fluid
study magnetohydrodynamics. dynamics (CFD) applications when modeling
The Navier–Stokes equations are also of great turbulent flows. Some models include the Spalart–
interest in a purely mathematical sense. Despite their Allmaras, k–ω, k–ε, and SST models, which add a
wide range of practical uses, it has not yet been variety of additional equations to bring closure to the
proven whether solutions always exist in three RANS equations. Large eddy simulation (LES) can
dimensions and, if they do exist, whether they also be used to solve these equations numerically.
This approach is computationally more expensive—
are smooth – i.e. they are infinitely differentiable at
in time and in computer memory—than RANS, but
all points in the domain. These are called produces better results because it explicitly resolves
the Navier–Stokes existence and the larger turbulent scales.
smoothness problems. The Clay Mathematics
2. APPLICABILITY
Institute has called this one of the seven most
important open problems in mathematics. The Navier–Stokes equations assume that the fluid
being studied is a continuum (it is infinitely divisible
The Navier–Stokes equations have 2 and not composed of particles such as atoms or
properties as 1. Nonlinearity => The Navier–Stokes molecules), and is not moving at relativistic
equations are nonlinear partial differential velocities. At very small scales or under extreme
equations in the general case and so remain in conditions, real fluids made out of discrete
almost every real situation. In some cases, such as molecules will produce results different from the
one-dimensional flow and Stokes flow (or creeping continuous fluids modeled by the Navier–Stokes
flow), the equations can be simplified to linear equations. For example, capillarity of internal layers
equations. The nonlinearity makes most problems in fluids appears for flow with high gradients. For
difficult or impossible to solve and is the main large Knudsen number of the problem,
contributor to the turbulence that the equations the Boltzmann equation may be a suitable
model. replacement. Failing that, one may have to resort
The nonlinearity is due to convective acceleration, to molecular dynamics or various hybrid methods.
which is an acceleration associated with the change Another limitation is simply the complicated nature
in velocity over position. Hence, any convective of the equations. Time-tested formulations exist for
flow, whether turbulent or not, will involve common fluid families, but the application of the
nonlinearity. An example of convective Navier–Stokes equations to less common families
but laminar (nonturbulent) flow would be the tends to result in very complicated formulations and
passage of a viscous fluid (for example, oil) through often to open research problems. For this reason,
these equations are usually rewritten for Newtonian formed the basis of many more sophisticated
fluids where the viscosity model is linear; truly closures developed later on. Investigations of
general models for the flow of other kinds of fluids decaying anisotropic turbulence, however, showed
(such as blood) do not exist. that the return-to-isotropy process cannot be
Currently, The Navier–Stokes equations are used described accurately by a linear model. Hence,
extensively in video games in order to model a wide Speziale et al. (1991) (SSG) suggested a quadratic
variety of natural phenomena. Simulations of small- pressure-strain model which has prevailed in some
scale gaseous fluids, such as fire and smoke, are form in many modern closures.
often based on the seminal paper "Real-Time Fluid
Dynamics for Games" by Jos Stam, which Even in the late 1990s and early 2000s, thirty years
elaborates one of the methods proposed in Stam's after the first full DRSMs were suggested, they were
earlier, more famous paper "Stable Fluids" from still not sufficiently used to assess their advantages
1999. Stam proposes stable fluid simulation using a over LEVMs with respect to turbomachinery
Navier–Stokes solution method from 1968, coupled applications (Bradshaw, 1996). Especially low-
with an unconditionally stable semi- Reynolds closures had not yet been widely adopted
Lagrangian advection scheme, as first proposed in (Leschziner et al., 2000). However, there are more
1992. recent developments which show encouraging
More recent implementations based upon this work results for the application of those models to
run on the game systems graphics processing industrially relevant configurations.
unit (GPU) as opposed to the central processing
unit (CPU) and achieve a much higher degree of One area of research towards the application in
performance.[36][37] Many improvements have been complex geometries is the removal of wall geometry
proposed to Stam's original work, which suffers related parameters. Craft & Launder (1996)
inherently from high numerical dissipation in both formulated a model in- dependent of the wall
velocity and mass. geometry by replacing the wall-normal vector,
3. RELATED WORK which is usually found in wall modifications, by the
gradient of a turbulence anisotropy parameter. A
More than fifty years after Reynolds laid the ground modified version of this model was applied to
for RANS turbulence models (Reynolds, 1895), compressible flows involving shock-
Chou derived the governing equations for the wave/boundary-layer interaction. The complex flow
individual Reynolds stress components (Chou, around a blunt-fin/flat-plate junction could be
1945). Rotta then analysed the pressure-strain simulated successfully and showed superior results
redistribution term in great detail and suggested a compared to the Menter SST k-ω model as well as
simple return-to-isotropy model for the slow term the DRSM of Hanjali´c and Jakirli´c (Batten et
which still forms the basis for most closures today al.,1999). The approach of Craft and Launder was
(Rotta, 1951). Also, he took up considerations by also adopted by Gerolymos and his co-workers who
Kolmogorov about local isotropy of turbulence at published a series of papers about DRSMs applied
high Reynolds numbers which lead to isotropic to turbomachinery-related flows. They extended
dissipation models for the Reynolds stresses. their model (Gerolymos & Vallet, 1997) in a similar
An early solution to the transport equations for the fashion, to obtain a “wall-normal free” model
Reynolds stresses and the dissipation rate of (Gerolymos et al., 2002; Gerolymos & Vallet,
turbulent kinetic energy was presented by Daly and 2002). Based on this model, they were able to obtain
Harlow. They computed the set of equations a superior predictive accuracy compared to eddy
simplified for the case of flow between two parallel viscosity models (Gerolymos & Vallet, 2007;
flat plates (Daly & Harlow, 1970). One of the first Gerolymos et al., 2010).
full differential Reynolds stress closures was
presented by Hanjali´c & Launder (1972). Their
strategy was a term-by-term modelling of the
Reynolds stress and dissipation equations. The
pressure-strain correlation was an early version of
one of the most cited pressure-strain models,
published three years later by Launder et al. (1975)
(LRR). The blocking effect of a solid wall had to be
considered explicitly in the model by modifications
of the redistribution term. The model, which is linear
in the velocity gradient and Reynolds stress tensors,
Other authors choose a different approach and focus
4. MATHEMATICAL MODEL
primarily on the prediction of mean flow quantities.
Wilcox (2006) developed a high Reynolds, The Reynolds-averaged Navier–Stokes equations
compressible DRSM based on the most recent
can be used with approximations based on
version of his 𝑘 − 𝜔 LEVM. It was calibrated for
boundary-layer flows and a range of homogeneous knowledge of the properties of flow turbulence to
turbulence flows. No wall-reflection terms are give approximate time-averaged solutions to
employed in the pressure-strain model because,
according to the author, they are not required to the Navier–Stokes equations. For a stationary,
obtain the boundary layer profile if the 𝜔-equation incompressible Newtonian fluid, these equations
is used in contrast to 𝜖-equation in original LRR-
model. The model was intensively tested and can be written in Einstein notation in Cartesian
compared to other DRSMs by Sciberras & Coleman coordinates as:
(2007). In plane turbulent channel flow, it was found
that the model has serious deficits in normal-stresses
prediction, but, due to cancellation of errors, the
mean velocity profile and skin friction coefficient
are predicted successfully. However, the authors
state that correct normal stress prediction is essential
especially for flows with separation. The left hand side of this equation
Eisfeld (2006) developed his model for the represents the change in mean momentum of fluid
application in industrially relevant external element owing to the unsteadiness in the mean
aerodynamic flows. He followed the approach of flow and the convection by the mean flow. This
Menter and combined the SSG model for free shear change is balanced by the mean body force, the
flows with the Stress-ω model for wall bounded isotropic stress owing to the mean pressure field, the
flows using the blending function developed by viscous stresses, and apparent stress owing
Menter for his SST k-ω model. Hence, close to solid to the fluctuating velocity field, generally referred to
walls, the model inherits the deficiencies of the as the Reynolds stress. This nonlinear Reynolds
Stress-ω model concerning the prediction of normal- stress term requires additional modeling to close the
stresses (Morsbach et al., 2012). Mean flow features RANS equation for solving, and has led to the
of complex 3D flows, however, are captured more creation of many different turbulence models. The
accurately than with standard LEVM approaches as time-average operator is a Reynolds operator.
shown for the ONERA M6 wing (C´ecora et al., Derivation of RANS equations
2012). Furthermore, Eisfeld (2010a) studied the
influence of the scale-determining equation on the The basic tool required for the derivation of
prediction of the shock-position for the RAE 2822 the RANS equations from the instantaneous Navier–
aerofoil. It was found that it can have a larger Stokes equations is the Reynolds decomposition.
influence than the choice of pressure-strain model. Reynolds decomposition refers to separation of the
Nguyen et al. (2011) developed near-wall extensions flow variable (like velocity ) into the mean (time-
for the model to overcome the deficiencies in averaged) component and the fluctuating
normal-stress prediction. component . Because the mean operator is
a Reynolds operator, it has a set of properties. One
The complex 3D structure of the flow through a of these properties is that the mean of the fluctuating
linear compressor cascade with tip clearance was quantity is equal to zero . Thus,
investigated using the DRSM of Hanjali´c &
Jakirli´c (1998) by Borello et al. (2007). As
expected, simulations using the DRSM resulted in
improved predictive accuracy of the mean velocity
profiles at most measured stations. The most is the position vector. Some author prefer
significant improvements could be seen in the
representation of the interacting vortices where the using in stead of for the mean term (since an
flow is 3D and the turbulence is anisotropic. Thus, overbar is sometimes used to represent a vector). In
according to the authors, DRSMs are the most this case, the fluctuating term is represented
adequate form of RANS turbulence modelling. instead by .This is possible because the two terms
do not appear simultaneously in the same equation.
To avoid confusion, the 5. RANS LIMITATION
notation will be used to represent RANS is the most widely used approach for
the instantaneous, mean, and fluctuating terms, industrial flows. RANS turbulence models are the
respectively. only modeling approach for steady state simulation
The properties of Reynolds operators are useful in of turbulent flows. The family of Reynolds-averaged
the derivation of the RANS equations. Using these Navier-Stokes (RANS) models is the largest in the
properties, the Navier–Stokes equations of motion, field of turbulence. These models attempt to close
expressed in tensor notation, are (for an the turbulence equations using viscosity terms. A
incompressible Newtonian fluid): common variable calculated in these models is k, or
the kinetic energy per unit mass of turbulent
fluctuations.

Generally, there are some limitations with RANS


models as they are based on the definition of
turbulent viscosity. These limitations are:

 Lack of physical description


 Turbulence-induced secondary flows
 Streamlined curvatures
where is a vector representing external forces.  Swirling flows or flows with rotations
Next, each instantaneous quantity can be split into  Transitional flows between turbulent and
time-averaged and fluctuating components, and the laminar
resulting equation time-averaged, [4] to yield:  Unsteady flows like internal combustion
engines
 Stagnant regions in flows

For many applications, steady state solutions are


preferable, and for many applications a good RANS
model with a good quality grid will provide all the
required accuracy. RANS is an industry standard,
Inexpensive for most engineering applications and
Well established with numerous models. However,
The momentum equation can also be written as for some flows, challenges associated with RANS
modeling can limit the level of accuracy that it is
possible to attain. It is limited to steady state and
requires a significant modelling effort to close the
system of equations. Using RANS, the
computational costs can be reduced by solving
On further manipulations this yields, the statistically averaged equation system, which
requires closure assumptions for the higher
moments. RANS models have a computing time
of only about 5% of the LES. Sophisticated RANS
models like EASM are able to capture important
flow features correctly. At low computational
costs that makes them already a useful tool in
industrial design.
where, is the mean
rate of strain tensor. To gain a deeper understanding of the limitations
Finally, since integration in time removes the time and strengths of RANS models, we will see variety
dependence of the resultant terms, the time of models which are based on RANS.
derivative must be eliminated, leaving:
RANS Single-Equation Model: Spalart-Allmaras: Marra explained that the model uses wall functions
to analytically account for the fluid velocity in the
“Spalart-Allmaras (SA) is a one-equation turbulence viscous sublayer near the wall.
model that has been developed specifically for
aerodynamic flows such as transonic flow over The technique offers good convergence and isn’t
airfoils,” said Baglietto. memory-intensive. Marra also explained that the
model is typically used for external flows with
The model is based on kinematic eddy viscosity and complex geometry. However, it is also a good
mixing length. This mixing length defines the general-purpose model.
transport of the turbulent viscosity.
Baglietto noted that the equation for epsilon is
“When we look at the benefits and drawbacks, the postulated, so it isn’t perfect. Nonetheless, the model
Spalart-Allmaras model has historically been a is used for the largest number of applications. This
strength … due to its speed and robustness,” said is partly because many of the model’s limitations are
Corson. well-known.

Figure 1 Simulation of the turbulent flow around a


NACA profile calculated using Spalart-Allmaras
within COMSOL. (Image courtesy of COMSOL.)

“Because we are only solving a single equation for Figure 2 Turbulent flow around a car-like model
turbulence,” Corson added, “the non-linear calculated in COMSOL using a k-epsilon model.
convergence is outstanding and the model is very (Image courtesy of COMSOL.)
forgiving of poor quality mesh, particularly in the
near wall region. The drawback is that it does have
some limitations due to the single-equation Limitations of k-epsilon include:
formulation. The turbulence length and time scales
are not as well defined as they are in other models  No-slip walls
such as SST.”  Adverse pressure gradients
 Strong curvatures
Limitations of Spalart-Allmaras include:
 Jet flows
 Shear flows  Difficulty solving for epsilon
 Under predicting separation
Despite this, the model is reliable due to its
 Decaying turbulence
predictability and numerous variants that aim to
RANS Two-Equation Model: Standard k-epsilon, improve the model for several applications.
Realizable k-epsilon, RNG k-epsilon
Perhaps the most famous variation of the model is
“In [the standard k-epsilon model] we solve for two the realizable k-epsilon model. This variation
variables, the turbulent kinetic energy, k, and the modifies the equation for epsilon and introduces the
rate of dissipation of kinetic energy, epsilon [ε],” effect of the mean flow distortion on turbulent
said Valerio Marra, marketing director at COMSOL. dissipation.

“[Realizable k-epsilon] is the default


recommendation in mainstream commercial
packages, therefore represents the most proven,
well-quantified and widely-documented of all
closures,” said Baglietto. “The model has improved
performance for planar surfaces, round jets, rotation,
recirculation and streamline curvature. It also
improves the boundary layer under strong adverse
pressure gradients or separation. But it cannot do
magic as it’s still based on [turbulent] viscosity.”

Malan clarifies that k-epsilon has also become the


“de facto” standard two-equation model because its
two-layer formulation has improved its applicability
to well-resolved boundary layers. It also has
improved results for complex separated industrial
flows.

Another popular modification is the renormalization Figure 3 Left: Simulation of a turbulent flow
group (RNG) k-epsilon model. The model was modeled with the shear stress transport (SST) k-
omega turbulence model in Altair AcuSolve. Right:
originally derived by attempting to solve for epsilon
Comparison of the convergence rate for the model
using the Navier-Stokes equation. The result was
solved using Spalart-Allmaras, SST k-omega and
very much like the original equation. However, an standard k-omega model
update of the method added a term to the epsilon
equation that accounts for the mean flow distortion “The most significant advantage of the k-omega
of turbulence dissipation. model is that it may be applied throughout the
boundary layer without further modification,” said
RANS Two-Equation Model: Standard k-omega and
Baglietto. “Furthermore, the standard k-omega
SST k-omega
model can be used in this mode without requiring the
Another popular two-equation model pairs k with computation of wall distance.”
the specific rate of dissipation of kinetic energy, or
“[k-omega] is a popular model for turbomachinery
omega (ω). Baglietto explained that the aim of the
simulations and for simulations where strong
standard k-omega model is to model near-wall
vortices are present such as those originating from
interactions more accurately than k-epsilon models.
wing tips,” said Mann. “[It] performs well for
However, he noted that k-omega can over-predict swirling flows and in the near wall region, but it
shear stresses of adverse pressure gradient boundary over-predicts separation.”
layers and that the model has issues with free stream
flows. The model is also very sensitive to inlet
boundary conditions, which is a disadvantage not Limitations of k-omega include:
seen in k-epsilon.
 Difficulty of convergence compared to k-
epsilon
 Sensitivity to initial conditions

One variant of k-omega that has gained popularity,


especially in the aeronautics area, is the shear stress
transport (SST) model. The model has gained this
popularity based on its ability to predict separation
and reattachment better when compared to k-epsilon
and the standard k-omega.

“The SST k-omega model is an enhancement of the


original k-omega model and addresses some specific
flaws of the base model, such as the sensitivity to
freestream turbulence levels,” explained Malan. “It
has the advantage that it can be applied to the
viscous-affected region without further
 
modification, which is one reason it has become a denoted by y1  10 and y1  30 in the NPARC
popular choice in aerospace applications where the
study. The above values are representative of the
flow is deemed too complex for Spalart-Allmaras.” 
maximum y1 value at the grid points nearest to the
The SST model accounts for cross-diffusion which plate as computed by the NPARC WIND code. The
better marries the k-epsilon and k-omega models. computational grids are simple Cartesian meshes
Using a blended function based on wall distance, and contain 110  80 quadrilateral elements with
engineers can include cross-diffusion when away 96 elements lying on the plate and 14 elements ahead
from the wall but not near it. In other words, using of the plate leading edge. Grid lines are clustered
the wall distance as a switch, SST works like k- both around the plate leading edge and near the wall
epsilon in the far field and k-omega near the target surface. The distance of the first grid line parallel to
geometry. the plate,
divided by the plate length, is equal to
“Purists may object strongly that the blending
function crossover location is arbitrary and could
y1  6.437 107 , y2  4.288 106 , y3  2.543 105
obscure some critical feature of the turbulence,” or the three considered grids. The grids are here
noted Baglietto. Clearly the model isn’t perfect; it labelled as G1, G2 and G3 and we remark that the
also requires limiters to improve the prediction of y1 values computed by our code over almost the
stagnant regions of the flow. Additionally, it has entire plate are much lower. In fact we have found
issues predicting turbulence levels and complex y1  0.25, y1  1.67 and y1  10 for the G1,
internal flows and it doesn’t take buoyancy into
G2 and G3 grids, respectively. All the solutions have
account.
been computed through a sequence of P0, P1 and P2
Malan added, “Some people claim that the model approximations, starting from an uniform flow field
has superior performance to the k-epsilon model in at freestream conditions and using the backward
simulating boundary layers with adverse pressure Euler scheme for time integration. In all cases the P0
solution has been run for 500-time steps and after
gradients. Ultimately, though, the performance of
that the P1 and P2 approximations took just few tens
SST k-omega is not very different from the
of time steps to reach fully converged high-order
realizable k-epsilon two-layer model. The choice solutions.
between the two will typically be made based on The skin friction distributions of the P1 and P2
user preference.” solutions computed on the G1, G2 and G3 grids are
compared with the experimental data in Figs. 1 and
6.APPLICATION
2. These figures clearly show that even the P1
Numerical results– The method has been applied to solution on the near-wall coarsest G3 grid is very
compute the steady turbulent flow over a flat plate accurate as far as the skin friction coefficient is
and the unsteady turbulent vortex shedding concerned. The near-wall behavior of the P1 and P2
phenomenon past the trailing edge of a turbine solutions is presented in terms of velocity and
blade. turbulence properties profiles. The quantities
u   u / u , k   k / u2 and     w / u2 are
6.1. Turbulent boundary layer over a flat plate plotted
The flat plate flow here considered is that reported
as functions of y  yu /  w at x / L  0.923

by Wieghardt and included in the 1968
AFOSR- IFP Stanford Conference. This test case is (the last measuring station of the Wieghardt data).
one of the validation cases used by the Notice that the lines of the computed results drawn
NPARC Alliance to evaluate the accuracy of its in the plots depict the actual high-order
CFD codes, in particular the turbulence models polynomial approximation of the solution inside the
implementations and their sensitivity to near- wall elements.
grid spacing [ 18] . For the purpose of convergence
acceleration, the Mach number ahead of the plate,
equal to 0.096 in the experiment, has been increased
up to 0. 2 in the simulations. The Reynolds number
based on the plate length is equal to 11.110 and
6

the inlet turbulence quantities are determined from


the prescribed turbulence intensity and turbulent
4
Reynolds number, both equal to 10
For our computations we have used the same
grids of the NPARC Alliance study and the results
here presented are those obtained on the grids
that, as w depends on the grid spacing, there are
three curves representing the law-of-the-wall in
the viscous sub-layer.
Figs. 3 and 4 show that all the computed
velocity profiles agree pretty well with each other,
with the exception of the P1, G3 and, to a lesser
extent, of the P2, G3 solutions in the viscous sub-
layer. The turbulent kinetic energy profiles, reported
in Figs. 5 and 6, display plateau and peak values of
k  which are in fairly good agreement with the
‘‘average’’ experimental data reported in.
Notice however that on the G3 grid even the P2
solution is not accurate enough to eliminate the
inaccuracy of the P1 solution in the region of

maximum k .Finally, Figs. 7 and 8 put in evidence
the influence of the near- wall grid resolution and of
the degree of polynomial approximation on
the accuracy of the   solution in the viscous sub-
layer.

6.2 Unsteady vortex shedding behind a turbine blade

In the second test case we consider the


subsonic flow in a cascade of turbine blades with
rounded and thick trailing edges ( TE) . The
experimental investigation of this cascade has been
the

In the plots of the near- wall u and  profiles we


 

have also drawn the corresponding law-of-the-


wall profiles according to the equations

Where   0.41 is the von Karman constant,


B  5 and  * and  are model constants.
Observe
topic of a European research project, aimed at According to the experiments, this test case has been
studying the unsteady wakes behind the computed for a downstream
trailing edge of turbine blades. The experimental
isentropic Mach number M 2is  0.4 a Reynolds
results have shown that the dominant vortex
shedding frequency does not appear as a single peak number based on the downstream isentropic
in the frequency spectrum but rather as a conditions and on the blade chord
bandwidth. The frequency range is strongly ReC 2is  2 106 and an inlet turbulence intensity
influenced by the thickness and by the type (laminar,
turbulent or transitional) of the boundary layers that Tuin  1.15 102 .The computational grid, shown
develop on the blade pressure (PS) and in Figs. 9 and 10, contains 5411 triangular
suction (SS) sides. elements and consists of a layer of anisotropic
elements all around the blade surface connected
with the set of isotropic elements that fill the rest of
the computational domain.
The solution has been computed by using P0, P1 and periods. For comparison purposes also the P1
P2 elements. Starting from an initial flow computation has been continued for the same
field at rest, the P0 solution is advanced in time for number of vortex shedding cycles of the P2 solution.
700 time steps by means of the backward Euler Figs. 11 and 12 show a snapshot of the
scheme. The time step magnitude t in this phase Mach number and of the turbulence intensity fields
corresponds to about 1/20 of the vortex of
shedding period T . Starting from the computed P0 the P2 solution. In these figures the alternate vortex
flow field the integration in time continues for shedding structure and the marked widening of the
800 time steps using the P1 approximation, the turbulent wake are clearly evident. Fig. 13 presents
second-order Runge–Kutta scheme and a time step the evolution of the density field during

t T / 80 .Then the solution switches to P2


approximation and continues, using the same
integration scheme and the same time step of the P1
solution, for about 100 vortex shedding
say that, whereas the computed average pressure
compares favorably with the experimental
data, the RMS of the computed pressure fluctuations
is largely overestimated, especially
for the P2 approximation, with respect to the
experiments. This discrepancy has also
been observed by other authors, who have used
different numerical methods and

a vortex shedding cycle. The vortex shed by the


pressure side appears to be stronger than that shed
by the suction side. This is due to the different
boundary layer development on the two sides of the
blade.
Next we focus on the pressure field around
the TE, where the origin of the curvilinear
coordinates, pointing towards the PS, is fixed on the
center of the TE. The base pressure distribution

turbulence models. However it is worth mentioning


that subsequent experimental investigations
on the same blade at an higher Mach number
resulted in much greater pressure fluctuations.
The pressure fluctuations at the PS and SS
separation points are displayed in Figs. 16 and 17 for
is illustrated by Figs. 14 and 15, which show the time both P1 and P2 solutions. Figs. 18 and 19 show the
averaged pressure distribution, made results of the FFT analysis of the pressure
non- dimensional by the inlet total pressure, and the
RMS of the pressure fluctuations, made
non- dimensional by the downstream dynamic
pressure. The averaging time interval corresponds
to the last 28 vortex shedding periods out of the more
than 100 that have been computed.
The experimental data, are not indicated in the
figures. However, we can
fluctuations shown in Figs. 16 and 17. Maximum
peaks occur in the range of Strouhal numbers
S  fDTE / v2is  between 0. 26 and 0. 29. This
result is in excellent agreement with the
experimental where the authors claim that flow periodicity is
data reported. obtained after a number of shedding
Despite the large number of computed cycles lower than the one of the present work. This
vortex shedding cycles, Figs. 16 and 17 indicate that discrepancy could be explained by considering
the flow regime is not periodic. This is in turbulence effects are accounted for by means of
disagreement with the computational results algebraic models, whereas
reported in the present work we use the 𝑘–𝜔 model which
can simulate the transition in the boundary layer. We
have noticed, in fact, that the pressure fluctuations in
the trailing edge region propagate upstream and
influence the location of the boundary layer
transition.
However, the most striking feature of these
figures is the fact that pressure fluctuations appear
to be modulated in amplitude. This phenomenon
surely deserves a deeper investigation in order to
clarify both physical and numerical aspects that
could have determined it.
Finally, the residuals time histories shown
in Figs. 20 and 21 confirm that a perfectly periodic
flow regime has not been reached yet.
[ 5] Bassi F, Rebay S. GMRES discontinuous Galerkin
solution of the compressible Navier–Stokes equations. In:
Discontinuous Galerkin methods, theory, computation and
applications. Lecture notes in computational science
and engineering, vol. 11. Springer- Verlag; 2000. First
International Symposium on Discontinuous Galerkin
Methods, May 24–26, 1999, Newport, RI, USA.
[ 6] Ilinca F, Pelletier D. Positivity preservation and
adaptive solution for the k– _ model of turbulence. AIAA
J 1996;36(1):44–50.
[ 7] Gerolymos GA, Vallet I. Implicit computation of the
three-dimensional compressible Navier–Stokes equations.
AIAA J 1996;34(7):1321–30. F. Bassi et al. / Computers
& Fluids 34 (2005) 507–540 539
[ 8] Menter FR. Two-equation eddy-viscosity turbulence
models for engineering applications. AIAA J 1994;32(8):
1598–605.
[9] Hellsten A. On the solid-wall boundary condition of x
in the k– x- type turbulence models. Technical Report B–
50, Helsinky University of Technology, Laboratory of
Aerodynamics, 1998.
[10] H€anel D, Schwane R, Seider G. On the accuracy of
upwind schemes for the solution of the Navier–Stokes
equations. AIAA Paper 87- 1105 CP, AIAA, Proceedings
of the AIAA 8th Computational Fluid Dynamics
Conference, July 1987.
[11] Brezzi F, Marini D, Pietra P, Russo A. Discontinuous
finite elements for diffusion problems. In: Francesco
Brioschi ( 1824- 1897) Convegno di Studi Matematici.
Incontro di Studio No. 16, Milano, 22–23 October, 1997,
1st Lomb Acc Sc Lett, 1999. p. 197–217.
[ 12] Bassi F, Rebay S. Ahigh- order accurate
discontinuous finite element method for the numerical
solution of the compressible Navier– Stokes equations. J
Comput Phys 1997;131:267–79.
[ 13] Bassi F, Rebay S. Numerical evaluation of two
discontinuous Galerkin solvers for the compressible
Navier– Stokes equations. Int J Numer Methods Fluids
2002;40:197–207.
[ 14] Iannelli GS, Baker AJ. A stiffly- stable implicit
Runge– Kutta algorithm for CFD applications. AIAA
Paper 88-0416, AIAA, 1988.
[ 15] Bassi F, Rebay S. Ahigh order discontinuous
Galerkin method for compressible turbulent flows. In:
Discontinuous Galerkin methods, theory, computation and
REFERENCES applications. Lecture notes in computational science and
engineering, vol. 11. Springer- Verlag; 2000. First
[ 1] Bassi F, Rebay S, Mariotti G, Pedinotti S, Savini M. International Symposium on Discontinuous Galerkin
Ahigh-order accurate discontinuous finite element method Methods, May 24–26, 1999, Newport, RI, USA.
for inviscid and viscous turbomachinery flows. In: [ 16] Wieghardt K, Tillman W. On the turbulent friction
Decuypere R, Dibelius G, editors. 2nd European layer for rising pressure. Technical Memorandum 1314,
Conference on Turbomachinery Fluid Dynamics and NACA, 1951.
Thermodynamics, Technologisch Instituut, Antwerpen, [ 17] Coles DE, Hirst EA. Computation of turbulent
Belgium, 1997. p. 99–108. boundary layers–1968 AFOSR-IFP-Stanford Conference.
[ 2] Brezzi F, Manzini M, Marini D, Pietra P, Russo A. Stanford University, 1969.
Discontinuous Galerkin approximations for elliptic [ 18] Yoder DA, Georgiadis NJ. Implementation and
problems. Numer Methods Part Different Equat validation of the chien k– _ turbulence model in the wind
2000;16:365–78. Navier–Stokes code. AIAA Paper 99-0745, AIAA, 1999.
[ 3] Arnold D, Brezzi F, Cockburn B, Marini D. [19] Patel VC, Rodi W, Scheuerer G. Turbulence models
Discontinuous Galerkin methods for elliptic problems. In: for near- wall and low Reynolds number flows: a review.
Discontinuous Galerkin methods, theory, computation and AIAAJ 1985;23(9):1308–18.
applications. Lecture notes in computational science [ 20] Cicatelli G, Sieverding CH. The effect of vortex
and engineering, vol. 11. Springer-Verlag; 2000. p. 89– shedding on the unsteady pressure distribution around the
101. First International Symposium on Discontinuous trailing edge of a turbine blade. ASME Paper 96- GT- 39,
Galerkin Methods, May 24–26, 1999, Newport, RI, USA. ASME, 1996.
[ 4] Wilcox DC. Turbulence modelling for CFD, DCW [ 21] Zunino P, Ubaldi M, Campora U, Ghiglione A. An
industries Inc., La Ca~nada, CA91011, USA, 1993. experimental investigation of the flow in the trailing edge
region of a turbine cascade. In: Decuypere R, Dibelius G,
editors. 2nd European Conference on Turbomachinery
Fluid Dynamics and Thermodynamics. Technologisch
Instituut, Antwerpen, Belgium, 1997. p. 247–54.
[ 22] Arnone A, Pacciani R. Numerical prediction of
trailing edge wake shedding. ASME Paper 97- GT- 89,
ASME, 1997.
[ 23] Manna M, Mulas M, Cicatelli G. Vortex shedding
behind a blunt trailing edge turbine blade. Int J Turbo Jet
Engines 1997;14(3):145–57.
[24]CD-Adapco, 2009. CD-Adapco Training Material.
Course Notes on Turbulence Modelling. CD-Adapco
https://www.engineering.com : Choosing the Right
Turbulence Model for your CFD Simulation.

Potrebbero piacerti anche