0 Up votes0 Down votes

18 views40 pagesReview paper about TM.

Mar 29, 2016

© © All Rights Reserved

PDF, TXT or read online from Scribd

Review paper about TM.

© All Rights Reserved

18 views

Review paper about TM.

© All Rights Reserved

- 07A - Chapter 7 - Sec 7.1-7.3,App F - Black
- Adv Turb v62 01 Overview
- Torque 2010 Presentation Modelling Issues with Wind Turbine Wake and Atmospheric Turbulence
- Numerical Simulation of the Ultrasonic Cavitating Atomizer
- Hidraulica de Tuberias j p Tellius
- Lecture Seeding Particles for PIV
- 16294
- PCFD 10(5-6) Paper 13
- MFIXEquations2005-4-4
- Star 0520
- D. Kivotides et al- Quantum Signature of Superfluid Turbulence
- None
- Mesolayer in turbulent pipe and channel flows -
- 9-3-D numerical simulations of cylindrical pleated filter pa.pdf
- Basic Hyd Dyn Equations
- Distributor Design and Testing
- Presentation Invelox
- Numerical Simulation and Comparison With Experiment of Natural Convection Between Two Floors of a Building Model via a Stairwell
- Theoretics Study of The Effect of Suction and Blowing on Turbulence
- Gutheil - Turbulent Spray Combustion Modeling For Rocket Engine Applications

You are on page 1of 40

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

Review

C.D. Argyropoulos a,, N.C. Markatos b,c,

a

Department of Chemical Engineering, Imperial College London, South Kensington Campus, London SW7 2AZ, UK

Computational Fluid Dynamics Unit, School of Chemical Engineering, National Technical University of Athens, 9 Iroon Polytechniou Str., Zografou Campus,

15780 Athens, Greece

c

Metropolitan College, School of Engineering, 74 Sorou Str., Marousi, Athens 15125, Greece

b

a r t i c l e

i n f o

Article history:

Received 15 February 2013

Received in revised form 9 June 2014

Accepted 7 July 2014

Available online 14 July 2014

Keywords:

Turbulence modelling

DNS

LES

URANS

DES

Reynolds stress models

a b s t r a c t

This paper reviews the problems and successes of computing turbulent ow. Most of the

ow phenomena that are important to modern technology involve turbulence. The review

is concerned with methods for turbulent ow computer predictions and their applications,

and describes several of them. These computational methods are aimed at simulating

either as much detail of the turbulent motion as possible by current computer power or,

more commonly, its overall effect on the mean-ow behaviour. The methods are still being

developed and some of the most recent concepts involved are discussed.

Some success has been achieved with two-equation models for relatively simple hydrodynamic phenomena; indeed, routine design work has been undertaken during the last

three decades in several applications of engineering practise, for which extensive studies

have optimised these models.

Failures are still common for many applications particularly those that involve strong

curvature, intermittency, strong buoyancy inuences, low-Reynolds-number effects, rapid

compression or expansion, strong swirl, and kinetically-inuenced chemical reaction. New

conceptual developments are needed in these areas, probably along the lines of actually

calculating the principal manifestation of turbulence, e.g. intermittency. A start has been

made in this direction in the form of multi-uid models, and full simulations.

The turbulence modelling approaches presented here are, Reynolds-Averaged

NavierStokes (RANS), two-uid models, Very Large Eddy Simulation (VLES), Unsteady

Reynolds-Averaged NavierStokes (URANS), Detached Eddy Simulation (DES) and some

interesting, relatively recent, hybrid LES/RANS techniques.

A large number of relatively recent studies are considered, together with reference to the

numerical experiments existing on the subject.

The authors hope that they provide the interested reader with most of the appropriate

sources of turbulence modelling, exhibiting either as much detail as it is possible, by means

of bibliography, or illustrating some of the most recent developments on the numerical

modelling of turbulent ows. Thus, the potential user has the appropriate information,

for him to select the suitable turbulence model for his own case of interest.

2014 Elsevier Inc. All rights reserved.

Current address: Metropolitan College, School of Engineering, 74 Sorou Str., Marousi, Athens 15125, Greece (N.C. Markatos). Tel./fax: +30 210 7723126.

E-mail addresses: c.argyropoulos09@imperial.ac.uk (C.D. Argyropoulos), n.markatos@ntua.gr (N.C. Markatos).

URL: http://www.amc.edu.gr (N.C. Markatos).

http://dx.doi.org/10.1016/j.apm.2014.07.001

0307-904X/ 2014 Elsevier Inc. All rights reserved.

694

Contents

1.

2.

3.

4.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Computer modelling of turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.1.

The differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2.

Direct Numerical Simulation (DNS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.

Reynolds-Averaged Navier Stokes (RANS) models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.1.

Physical concepts of turbulence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.2.

The equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.3.

Zero-equation or algebraic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.4.

Half-equation models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.5.

One-equation models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.6.

Two-equation models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.6.1

The ke model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.6.2

Modified ke model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.6.3

The kx model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.6.4

More recent two-equation models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.6.5

Low Reynolds number modifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.7.

Non-Linear Eddy Viscosity Models (NLEVM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3.8.

Recent advances in eddy viscosity modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.4.

Differential Second-Moment (DSM) and Algebraic Stress Models (ASM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.5.

Two-fluid models of turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.6.

Large Eddy Simulation (LES) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.6.1.

Validation of the LES approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.7.

Monotone Integrated LES (MILES) and Implicit LES (ILES) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.8.

Unsteady Reynolds-Averaged NavierStokes (URANS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.9.

Very LES (VLES) and Detached-Eddy Simulation (DES). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.10.

Hybrid RANS/LES strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Applications of DNS and LES to flows in pipes and flows with a free surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.1.

DNS of turbulent pipe flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.2.

DNS of turbulent free-surface flows. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.3.

LES of turbulent pipe flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.4.

LES of turbulent free-surface flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

694

695

695

696

697

697

698

699

700

700

701

701

701

702

702

704

705

706

707

709

710

713

714

714

715

715

715

716

719

721

724

725

726

726

1. Introduction

Turbulence is the most complicated kind of uid motion, making even its precise denition difcult. Literature contains

many denitions as, for example, that included in Markatos [1]: A uid motion is described as turbulent if it is three-dimensional, rotational, intermittent, highly disordered, diffusive and dissipative.

Turbulence is a three-dimensional, time-dependent, nonlinear phenomenon. Its modelling is very attractive, as it saves

huge amounts of money, by avoiding the need to build and test prototypes, and as it transforms technologies by allowing

improved understanding of turbulence. This is particularly true in industrial ows which, apart from the complexities of

turbulence, involve also very complicated geometries and several design parameters, requiring optimisation [2]. Thus,

shape design is one of the most important drivers for the use of simulation approaches in uid-engineering industry.

Examples refer to the drag of an aircraft or ship, propulsive efciency of aeroengines or propellers, turbomachinery, chemical process engineering, among others. In comparison to experiments, Computational Modelling offers a competitive

advantage if it is able to guide the analyst to a better design.

Computer programs now exist which are capable of solving three-dimensional, time-dependent NavierStokes (NS) equations, within practical computer resources. The reason that we do not make direct computer simulations of turbulence is that

turbulence is dissipated, and momentum exchanged by small-scale uctuations [3]. The crucial difference between visualisations of laminar and turbulent ows is the appearance of eddying motions of a wide range of length scales in turbulent

ows [4,5].

A typical ow domain of 0.1 m by 0.1 m with a high Reynolds number turbulent ow might contain eddies down to 10

100 lm size. We would need computing meshes of 109 up to 1012 points to be able to describe processes at all length scales.

The fastest events take place with a frequency of the order of 10 kHz, so we would need to discretise time into steps of about

100 ls. We have estimated that the direct simulation of a turbulent channel ow at a Reynolds number of 800,000 requires a

computer which is half a million times faster than a current generation supercomputer. This estimate is analogous to the one

made by Speziale in 1991 [6], who stated that direct simulation of a turbulent pipe ow at a Reynolds number 500,000

required a computer 10 million times faster than the CRAY supercomputer of that time.

695

With present day computing power it has only recently started to become possible to track the dynamics of eddies in

relatively simple ows at transitional Reynolds number. The computing requirements for the direct solution of the timedependent NavierStokes equations for fully turbulent practical ows at high Reynolds numbers are truly phenomenal

and must await major developments in computer hardware, possibly those based on quantum computing.

Meanwhile, engineers need computational procedures which can supply adequate information about the turbulent

processes, but which avoid the need to predict the effect of each and every eddy in the ow [7]. Therefore, in quantitative

work one is obliged to use turbulence models based on using averaged NS equations and, in addition, a set of equations

that supposedly express the relations between terms appearing in the NS equations [810]. It must be realised that most

of the available models pay no respect to the actual physical modes of turbulence (eddies, velocity patterns, high-vorticity

regions, large structures that stretch and engulf...) and, therefore, obscure the physical processes they purport to represent.

Flow visualisation experiments [1116] conrm this point and demonstrate the difculty of precise denition and

modelling. It is therefore hardly surprising that the actual physics of turbulence are nowhere to be seen in the available

models; simply because nobody can see as yet how mathematics can be employed to represent them in the models. It is,

however, also true that the engineering community has fortuitously often obtained very useful results by using relatively

simple models, such as those described in Section 2.3 below, results that would have required much more man-time and

experimental cost to obtain in their absence. Therefore, cautiously exercised and interpreted the turbulence models can be

valuable tools in research and design despite their physical deciencies.

The purpose of the present effort is to provide a comprehensive review of the available turbulence modelling techniques.

The relevant material is certainly too much to be reviewed in a single paper. For this reason the authors conne attention to

what they consider the better established or more promising models. No disrespect is therefore implied for the models that

are scarcely or not at all mentioned. Extensive use has been made of the published literature on the topic and of earlier

reviews [1720,11,2140].

In addition, ERCOFTAC (European Research Community On Flow, Turbulence And Combustion) organises workshops and

special courses on best practise guidelines for CFD users. The Special Interest Group (SIG) 15 of ERCOFTAC is devoted to turbulence modelling, and provides the appropriate data (e.g. experimental, DNS, highly-resolved LES databases) for the verication and validation of turbulence models, thus promoting their use for fundamental research and for industrial

applications [41].

Turbulent heat and mass transport are not explicitly covered in this review; the interested reader is directed to the review

by Launder [9]. Multi-phase phenomena are also not explicitly covered, apart from presenting the general differential equations and some necessary, to the authors mind, discussion on considerable work done for free-surface ows.

The review concludes with a summary of the advantages and disadvantages of the various turbulence models, in an

attempt to assist the potential user in choosing the most suitable model for his particular problem.

In the remainder of this review paper: Section 2 illustrates all of the available techniques for predicting turbulent ows;

in Sections 3, a literature survey is presented for applications using the LES and DNS technique; nally in Section 4, conclusions and some recommendations for future research are outlined.

2. Computer modelling of turbulence

2.1. The differential equations

The motion of a uid in three dimensions is described by a system of partial differential equations that represent mathematical statements of the conservation laws of physics (mass, momentum, energy and concentration conservation). The

momentum conservation equations are called the NavierStokes equations. In what follows the Eulerian equations governing the dynamics and heat/mass transfer of a turbulent uid are given, in Cartesian tensor notation, using the

repeated-sufx summation convention. The equations are presented in the most general form of multi-phase ows

[4254], as the single-phase ones are easily derived by just setting the volume fraction, ri equal to unity.

A convenient assumption for deriving these equations is based on the concepts of time- and space-averaging; it is that

more than one phase can exist at the same location at the same time [46,54]. Then, any small volume of the domain of interest can be imagined as containing, at any particular time, a volume fraction ri of the ith phase. As a consequence, if there are n

phases in total,

n

X

r i 1:

i1

When ow properties are to be computed over nite time intervals, a suitable averaging over space and time must be carried

out. Following the above notion, that treats each phase as a continuum in the domain of interest, we can derive the following

balance equations:

Conservation of phase mass:

@

_ i;

q ri divqi r i ~

V i m

@t i

696

V i is the velocity vector, ri is the volume fraction of phase i; m

where qi is the density, ~

the phase i, from all sources per unit time, and div is the divergence operator (i.e. the limit of the outow divided by the

volume as the volume tends to zero).

Summation of (2) over all phases leads to the over-all mass-conservation equation:

n

X

@

i1

@t

V i 0;

qi r i divqi r i ~

Conservation of phase momentum:

@

k grad p Bik F ik lik ;

V i uik ri ~

q ri uik divqi r i ~

@t i

k

where: uik is the velocity component in the direction k of phase i; p is pressure, assumed to be shared between the phases; ~

is a unit vector in the k-direction; Bik is the k-direction body force per unit volume of phase i; Fik is the friction force exerted

on phase i by viscous action within that phase; and lik is the momentum transfer to phase i from interactions with other

phases occupying the same space.

Conservation of phase energy:

@

V i hi r i Q i H i J i ;

ri qi hi p divqi ri ~

@t

where: hi is stagnation enthalpy of phase i per unit mass (i.e. the thermodynamic enthalpy plus the kinetic energy of the

phase plus any potential energy); Qi is the heat transfer to phase i per unit volume; Hi is heat transfer within the same phase,

e.g. by thermal conduction and viscous action; and Ji is the effect of interactions with other phases.

Conservation of species-in-phase mass:

@

_ i M il ;

V i mil divri Cil grad mil r i Ril m

q ri mil divqi r i ~

@t i

where: mil is the mass fraction of chemical species l present in phase i; Ril is the rate of production of species l, by chemical

reaction, per unit volume of phase i present; Cil is the exchange coefcient of species l (diffusion); and Mil is the l-fraction of

the mass crossing the phase boundary, i.e. it represents the effect of interactions with other phases.

All of the above equations can be expressed in a single form as follows:

@

_ i Ui r i sui total source of ui ;

q ri ui divqi r i ~

V i ui divr i Cui grad ui m

@t i

where: ui is any extensive uid property; the rst term on the right-hand side expresses the whole of that part of the source

term which can be so expressed, with Cui being the exchange coefcient for ui. sui is the source/sink term for ui, per unit

_ i Ui represents the contribution to the total source of any interactions between the phases, such as any

phase volume; and m

phase change (with Ui being the value of ui in the material crossing the phase boundary, during phase change). Distribution

_ i Ui and sui is sometimes arbitrary, reecting modelling convenience. For single-phase situations, the

of effects between m

above equations are valid by setting the rs to unity.

For turbulent ow, averaging over times which are large compared with the uctuation time leads to similar equations

for time-average values of ui with uctuating-velocity effects usually represented by enlargement of Cui. More details on the

above concepts and equations may be found in [1].

2.2. Direct Numerical Simulation (DNS)

Solutions of turbulent ow problems (Eqs. (1)(4)) can be obtained by using various analytical or numerical approaches,

with different level of accuracy in each case. Among the latter approaches, the Direct Numerical Simulation (DNS) has made a

signicant contribution in turbulence research over the last decades [21], as it involves the numerical solution of the above

full three-dimensional, time-dependent NavierStokes equations without the need of any turbulence model. DNS is indeed

useful for the investigation of turbulence mechanisms, the improvement and development of turbulence models and for

assessing two-point closure theories.

Until the 1970s the DNS approach was impossible to be used due to computer systems with insufcient memory and

speed to accommodate the required resolution needed for the small-scale turbulence effects. The rst attempts for the investigation of homogeneous turbulence with DNS originated at the National Center for Atmospheric Research (NCAR) by Lilly

[55] and Orszag and Patterson [56] for 2-D and 3-D dimensional simulations, respectively. Rogallo [57] investigated the

effects of mean shear, rotational and irrotational strain on turbulence, based on the extension of Orszag and Patterson algorithm. Kim et al. [58], Moser et al. [59], Abe et al. [60], Del Alamo et al. [61] and Hoyas and Jimenez [62], among others, performed DNS for the investigation of wall turbulence for channel ows at1 Res = 180, 395, 640, 1900 and 2003.

1

Reynolds number based on the friction velocity us and the channel half width.

697

There have also been extensive investigations of DNS in turbulent boundary layers at2 Reh = 700 [63], 1410 [64], 2500 [65],

2900 [66], 4060 [67] and 6650 [68,69]. More complex problems in wall-bounded turbulent ows (e.g. square duct, homogeneous isotropic turbulence, heat transfer, turbulence control and wavy boundary) have also been studied by DNS. The interested

reader may also consider the review papers by Moin and Mahesh [21], Kasagi and Shikazono [20] and Ishihara et al. [23]. The

most recent review concerning almost all the aspects of DNS (e.g. wall-bounded turbulence, turbulence control, bluff body turbulence, turbulent ow structures and high performance computing) may be found in the work of Alfonsi [70].

The current largest scientic DNS was performed by Lee et al. [71] for turbulent channel ow at Res = 5200 and 3.5 times

more degrees of freedom than the DNS (40963 grid points) obtained by Kaneda et al. [72] and Kaneda and Ishihara [73] on

the Earth Simulator in Japan. The maximum Reynolds number obtained was approximately 1200 (Taylor microscale) which

is similar to the current capabilities obtained by laboratory experiments.

The absence of a turbulence model implies that the simulation is obtained by numerically solving over all the spatial and

temporal scales of turbulence, and its accuracy, therefore, is unrivalled by other methods. However, the DNS of high-Reynolds number ows poses overwhelming demands on present-day available computing resources (speed and storage). It

is, therefore, necessary that DNS satises the following two constraints, according to Rogallo and Moin [17] and Kasagi

and Shikazono [20]:

(1) The dimensions of the computational domain must be large enough to comprise the largest turbulence scales.

(2) Grid resolution must be ne enough to capture the dissipation length scale, which is known as the Kolmogorov microscale, g = (v3/e)1/4, where e is the average rate of dissipation of turbulence kinetic energy per unit mass, and v is the

kinematic viscosity of the uid.

As a result, the required number of grid points for a given DNS is dependent on the Kolmogorov micro-scale and Kolmogorov micro-timescale (s = (v/e)1/2) of the ow. The higher the Reynolds number, the ner the mesh should be. Hence, the

cell size in each direction of the computational domain should decrease with Re3/4 and the time step should decrease with

Re1/2 [74,3]. It is worth mentioning that the DNS time step is always smaller than the Kolmogorov micro-timescale in order

to maintain the algorithms numerical stability [75].

The required resolution for DNS in the directions parallel to the wall, according to the work of Kim et al. [58], is Dx+ = 8 and

Dz+ = 4, where Dx+ is the streamwise and Dz+ is the spanwise grid spacing, respectively. In wall-normal directions, a rule of

thumb is to place at least three grid points below y+ = 1 (non-dimensional distance from the wall to the rst grid point) and at

least 10 grid points for y+ < 10, while in the outer region such as the pipe/channel centre line a value of Dy+ = 10 must be used.

Even with modern super-computers, the applicability of DNS is limited to ows of low to moderate Reynolds numbers.

Despite this current limitation, DNS is an effective and very useful tool for turbulence research leading to satisfactory results,

and used for testing simpler turbulence models, but it is still not practical for industrial or general engineering applications.

Among other benets, DNS has contributed remarkably to testing conventional models and ideas and therefore to the development of turbulence theory, in many ways which are summarized briey below [20,74]. Furthermore, DNS data are important for the development and improvement of turbulence models, due to the ability of DNS to provide the appropriate

turbulence statistics, including pressure and all spatial derivatives.

Important dimensionless numbers such as Reynolds and Prandtl can be varied in DNS, a fact of signicant importance for

the derivation of a turbulence model with wide applicability. DNS is also suitable for studying a virtual ow which may occur

in reality now or in the near future. The latter advantage is important for the study of a dynamical turbulence phenomenon

[76] and for the evaluation of turbulence control methodologies [77].

Another important issue about DNS is the validation of the obtained results. According to Sandham [75] and Coleman and

Sandberg [78] the following are the criteria for such a validation: (a) validation of the obtained numerical data against

analytical solutions, experimental data and different numerical codes; (b) parametric studies with different grid resolutions,

domain sizes and time steps; (c) the time step (Dt) should be comparable with the Kolmogorov time-scale and the grid

spacing, Dxi, with Kolmogorov micro-scale, while the ratios of Dt/s and Dxi/s should be of order unity; (d) evaluation of

the statistical quantities budgets.

2.3. Reynolds-Averaged Navier Stokes (RANS) models

For the purpose of introducing the concepts of turbulent ow modelling we restrict attention to single-phase, incompressible ow with constant laminar viscosity. The introduction of two-phase considerations, variable density and viscosity

are nowadays relatively easy tasks in modern solution algorithms. Only a generic presentation of turbulence modelling is

attempted here, for the sake of clarity. Details on the manner in which turbulence models properly couple into multi-phase

ow solvers may be found in literature (for example [79,80]).

2.3.1. Physical concepts of turbulence

Before discussing the turbulence models a very brief description of some concepts is provided. The main characteristic of

turbulence is the transfer of energy to smaller spatial scales across a continuous wave-number spectrum, i.e. a 3D, nonlinear

2

698

process. A useful concept for discussing the main mechanisms of turbulence is that of an eddy [81,82,3]. An eddy can be

thought of as a typical turbulence pattern, covering a range of wave- lengths, large and small eddies co-existing in the same

volume of uid. The actual modes of turbulence are eddies and high-vorticity regions [4,3]. By analogy with molecular viscosity, which is a property of the uid, turbulence is often described by eddy viscosity as a local property of the uid; the

corresponding mixing length in eddy viscosity models is treated in an analogous manner to the molecular mean-free path

derived from the kinetic theory of gases. This description is based on erroneous physical concepts but has proved useful in

the quantitative prediction of simple turbulent ows [7].

The eddies can be considered as a tangle of vortex elements (or lines) that are stretched in a preferred direction by mean

ow and in a random direction by one another. This mechanism, the so-called vortex stretching, ultimately leads to the

breaking down of large eddies into smaller ones. This process takes the form of an energy cascade. Since eddies of comparable size can only exchange energy with one another [9], the kinetic energy from the mean motion is extracted from the

largest eddies [3]. This energy is then transferred to neighbouring eddies of smaller scales continuing to smaller and smaller

scales (larger and larger velocity gradients), the smallest scale being reached when the eddies lose energy by the direct action

of viscous stresses which nally convert it into internal thermal energy on the smallest-sized eddies [82]. It is important to

note that viscosity does not play any role in the stretching process nor does it determine the amount of dissipated energy; it

only determines the smallest scale at which dissipation takes place. It is the large eddies (comparable with the linear dimensions of the ow domain), characterising the large-scale motion, that determine the rate at which the mean-ow kinetic

energy is fed into turbulent motion, and can be passed on to smaller scales and be nally dissipated. The larger eddies

are thus mainly responsible for the transport of momentum and heat, and hence need to be properly simulated in a turbulence model. Because of direct interaction with the mean ow, the large-scale motion depends strongly on the boundary

conditions of the problem under consideration.

An increase in Reynolds number increases the width of the spectrum, i.e. the difference between the largest eddies (associated with low-frequency uctuations) and the smallest eddies (associated with high-frequency uctuations). This suggests

that at high Reynolds numbers the turbulent motion can be well approximated by a three-level procedure, namely, a mean

motion, a large-scale motion and a small-scale motion [83].

Viscosity does not usually affect the larger-scale eddies which are primarily responsible for turbulent mixing, with the

exception of the viscous sublayer very close to a solid surface. Furthermore, the effects of density uctuations on turbulence

are small if, as in the majority of practical situations, the density uctuations are small compared to the mean density, the

exception being the effect of temporal uctuations and spatial gradients of density in a gravitational eld. Therefore, one can

usually neglect the direct effect of viscosity and compressibility on turbulence. It is also important to note that it is the uctuating velocity eld that drives the uctuating scalar eld, the effect of the latter on the former usually being negligible.

2.3.2. The equations

Eqs. (8)(11) below constitute the mathematical representation of uid ows, under the assumptions that the turbulent

uid is a continuum, Newtonian in nature and that the ow can be described by the NavierStokes equations. For turbulent

ows, the latter represent the instantaneous values of the ow properties [1,84,85].

The equations for turbulence uctuations are obtained by Reynolds de-composition which describes the turbulent motion

as a random variation about a mean value [1]:

/0 ;

//

its time- mean value and /0 the uctuating part. The time-average of the ucwhere / is the instantaneous scalar quantity, /

0

is dened as:

tuating value is zero / 0, and the mean value /

/x limt!1

1

Dt

t 1 Dt

/x; tdt

t 1 Dt t 2 ;

t1

where t1 is the time scale of the rapid uctuations and t2 the time scale of the slow motion (for time-dependent mean value,

i.e. for non-stationary turbulence). By substituting Eq. (8) into the form of Eqs. (1)(3) for single-phase, incompressible ows

and then taking the time-mean of the resulting equations, one derives the following continuity and NS equations:

i

@u

0;

@x

10

i

i

@u

@

1 @p

@2u

@

i u

j

u

m

u0 u0 ;

@t @xj

q @xi

@xj @xj @xj i j

11

i is the mean velocity, u0i the uctuating velocity, q the uid density and v the kinematic viscosity. Eq. (11) is known

where u

as the Reynolds-Averaged NavierStokes (RANS) equation, while the term u0i u0j is the Reynolds-stress tensor:

12

which is a symmetric tensor with six independent components. It is observed from Eqs. (10) and (11) that the number of

unknown quantities (pressure, three velocity components and six stresses) is larger than the number of the available

699

equations (continuity and NavierStokes). As a result, the system of equations is not yet closed and the problem of closure

is reduced to the modelling of the Reynolds-stresses, in terms of mean-ow quantities.

The most popular approach to resolving this problem of closure is the use of the Boussinesq eddy-viscosity approximation [86]. The latter is based on an analogy between molecular and turbulent motions, in order to correlate Reynolds stresses

to the rate of strain of the mean motion. The turbulence eddies are thought of as parcels of uid, which like molecules, collide

and exchange momentum, obeying the kinetic theory of gases. Thus, in analogy with the molecular viscous stress, the Reynolds (turbulence) stresses are modelled as follows [1]:

i @ u

j

@u

;

@xj @xi

1

1 02

02

k u0i u0i

u1 u02

2 u3 ;

2

2

2

3

13

14

where k is the turbulence kinetic energy and mt lt =q is the turbulence or eddy (kinematic) viscosity which, in contrast to

the molecular (kinematic) viscosity is not constant; and may vary signicantly from ow to ow and from point to point [1];

and dij is the Kronecker delta. Substituting Eq. (13) into Eq. (11) leads to:

i

i

@u

@

1 @p

@

@u

i u

j

:

m m t

@t @xj

q @xi @xj

@xj

15

p

2k

The isotropic part of the Reynolds-stress tensor is absorbed normally into the pressure term as p

.

3

Dimensional analysis dictates that the unknown vt must be proportional to the product of a characteristic velocity Vt and

a characteristic length scale Lt. The difference between zero-equation, one-equation and two-equation models, discussed

below, lies in the way they choose to calculate them [1]. Thus, zero-equation models prescribe both characteristic velocity

and length-scale as algebraic expressions. One-equation models consider as characteristic velocity the square root of the turbulence kinetic energy and prescribe algebraically the length scale, therefore:

p

vt C v 1 kL;

16

where C v 1 is a dimensionless constant. Two-equation models, such as ke and kx [1], described below in this subsection,

use differential equations to compute both the characteristic velocity and length scale and then estimate the value of vt by

the following equations:

vt

C l f l ke

ke models

a xk

kx models

17

where fl is a damping function, Cl and a are constants, e is the turbulence energy dissipation rate and x the dissipation per

unit turbulence kinetic energy.

Recent developments have led to the construction of non-linear eddy viscosity models, aiming at including non-linear

terms of the strain-rate [40]. More details for these models are presented in Section 2.3.7.

The traditional linear-eddy-viscosity RANS models may be divided into the following four main categories [1,87]: (a) algebraic (zero-equation) models, (b) half equation models (c) one-equation models and (d) two-equation models.

In the remainder of this section, a number of the better-established and most promising, according to the present authors

experience, linear and non-linear eddy viscosity models, along with some more recently advanced ones, will be presented and

discussed.

2.3.3. Zero-equation or algebraic models

Zero-equation or algebraic models use partial differential equations only for computing the mean elds, while only algebraic expressions for the turbulence quantities [1]. This class of models is the oldest one, it is characterised by simplicity to

implement and has given good results for some applications of engineering relevance. For example, the best known of this

class, Prandtls mixing length model [88], is suitable for the prediction of thin-shear-layer ows such as boundary layers, jets,

mixing layers, and wakes. According to Prandtl, [88], in a boundary layer ow the eddy viscosity is given by:

@ u

vt 2mix ;

@y

18

where mix is the mixing length, that depends upon the type of ow, and is specied algebraically, while y is the direction

normal to the wall. This model is not suitable for predicting ows with recirculation and separation.

More modern variants of this category, following the contribution of Van Driest, Clauser and Klebanoff modications [86],

are the CebeciSmith [89] and BaldwinLomax models [90]. These models are characterised by two-layer mixing-length

eddy viscosities, one as an inner and one as an outer layer viscosity. The second model is distinguished from the rst because

of the different outer-layer length viscosity equation. Both are suitable for predicting turbulent ows in aerodynamics (e.g.

around airfoils) with similar accuracy, but are unreliable for separated ows. The mathematical formulations of these models

may be found in the textbook by Wilcox [86]. Nowadays, zero-equation models are used rarely and only for getting an initial

prediction of the ow eld [91].

700

In 1985 Johnson and King [92] developed a two layer model for the investigation of pressure driven separated ows. The

JK model has been improved by Johnson [93] and Johnson and Coakley [94] in order to become applicable to compressible

ows as well. In addition, the JK model has been extended for 3-D dimensional ows by Savill et al. [95]. The mathematical

equations along with comparisons against computational and experimental data may be found in Wilcox [86]. Even though

the JK model has improved the classical algebraic models for predicting turbulent, transonic separate ows, it still suffers

from the same drawbacks as the CebeciSmith and BaldwinLomax models.

2.3.5. One-equation models

One-equation models are characterised by formulating one additional transport equation for the computation of a turbulence quantity, usually the turbulence kinetic energy (k). For all of them there is still a need of prescribing a length-scale

distribution (L), which is dened algebraically and is usually based on available experimental data. For elliptic ows, like

recirculating and separated ones, experimental data is generally not available, making it difcult to prescribe algebraically

such a length scale. Therefore, most researchers decided to adopt two- or even more-equation models [1]. One-equation

models were used mainly in the nuclear and aeronautics industries (e.g. aircraft wings, fuselage, nuclear reactors) and the

most well known for aerospace applications are Baldwin and Barth [96] and Spalart and Allmaras [97] models. The SpalartAllmaras was designed and optimised for ows past wings and airfoils and produced very good results. It is also easy

to implement for any type of grid (e.g. structured or unstructured, single-block or multi-block) [40]. However, both models

create enormous diffusion, in particular for regions of 3-D vortical ow [40]. Improvements of the aforementioned models

are presented in the works of Spalart and Shur [98], Dacles-Mariani et al. [99] and Rahman et al. [100], regarding the effects

of curvature, rotation, decrease of diffusion and for near-wall effects. Recent studies with the SpalartAllmaras model have

been presented by Karabelas and Markatos [101] and Karabelas [102] for ow over an airfoil and past a apping multi-element airfoil, respectively. Karabelas [102] performed simulations past a plunging multi-element airfoil at Re = 6 105

(Fig. 1).

Accurate resolution of such ows still constitutes a great mathematical challenge for RANS modelling. It is well known

that the latter is often inaccurate even in terms of integrated quantities, such as lift and drag coefcients. This is due to

Fig. 1. Turbulence simulations of the ow past a plunging multi-element airfoil at Re = 6 105: Path-lines and pressure distribution at three xed

geometric angles of attack, soaring ight regime (left), mid-time of the up-stroke (middle) and mid-time of the down-stroke phase (right), reproduced with

the authors permission [102]. Reprinted by permission of the American Institute of Aeronautics and Astronautics, Inc.

701

the fact that in ow regimes past multi-element airfoils, multiple transition processes from laminar to turbulent states and

vice versa could occur. In this study, we mention the one-equation model of Spalart Allmaras because its performance was

found to be superior [103] even over other more complete two-equation models. Further tuning of the latter models to

include transition effects did not increase considerably the accuracy of the simulations. It is worth mentioning, that in

the workshop held by NASA [104], for validating turbulence modelling of the ow past the multi-element airfoil MDA

30P-30N, mainly one-equation models were used in the governing equations.

Recently, Fares and Schroder [105] developed a complete and general one-equation model based on the two-equation k

x model for predicting general turbulent ows such as wakes, jets, boundary layers and vortex ows. The new model proved

more accurate compared to the SpalartAllmaras model, especially for jets and vortex ows. More information for zero-,

half- and one-equation models is provided in detail in the review papers by Markatos [1], Alfonsi [106] and in the classic

text by Wilcox [86].

2.3.6. Two-equation models

Two-equation models use, in addition to the mean-ow NavierStokes equations, two transport equations for two turbulence properties. The rst one is usually that for the turbulence kinetic energy (k) and the second any other from a variety

that includes: the dissipation rate of turbulence kinetic energy (e), the specic dissipation rate (x), the length scale (l), the

product of k l, the time scale s, the product of k and s, among others [1]. This class of models is the most preferred by industry it looks like remaining so for the foreseeable future [85]. Two-equation eddy viscosity models are still the rst choice for

general CFD calculations, with the standard ke model [107] and kx [108] being the most widely used. There is no particular reason for this preference, but at least those models have been applied so widely, that we know their behaviour

beforehand.

In this section only the ke and kx models are presented, as being representative of the two-equation models, along

with their improvements and some interesting low-Re versions.

2.3.6.1. The ke model

The ke model is by far the most widely used and tested two-equation model, with many improvements incorporated

over the years. The standard ke model of Launder and Sharma [107] is specied as follows:

Kinematic eddy viscosity (vt) equation:

mt C l

19

i

@k

@k

@ m vt @k

@u

j

e sij

u

:

@t

@xj @xj

rk @xj

@xj

20

@e

@e

@ m vt @ e

e @ ui

e2

j

C e1 sij

u

C e2 ;

@t

@xj @xj

re @xj

k @xj

k

21

where rk = 1.0 and re = 1.3 are the Prandtl numbers for k and e, respectively. The remaining model constants are: Cl = 0.09,

Ce1 = 1.44, Ce2 = 1.92. The standard k-e model behaves very in predicting turbulent shear ows, in many applications of engineering interest. However, this model is unable to predict accurately ows with adverse pressure gradients and extra strains

(e.g. streamline curvature, skewing, rotation [91]). As a result it yields poor results for separated ows, whilst it is rather

difcult to be integrated through the viscous sublayer [86]. Despite the above shortcomings, the ke model is recommended

for an at least gross estimation of the ow eld and for cases such as combustion, multiphase ows and ows with chemical

reactions [91].

2.3.6.2. Modied ke model

Improvements and modications of the standard ke model are many (for example, for ows with strong buoyancy, [1])

with probably the most important being the realisable ke model [109] and the Renormalization Group (RNG) ke model

[110].

The rst model is based on the satisfaction of the realizability constraints on the normal Reynolds stresses and the

Schwartz inequality for turbulent shear stresses. Beside this, the Cl constant of standard ke model is not anymore a

constant but it is computed in this improved model by an eddy-viscosity equation. Performance is substantially improved

for jets and mixing layers, channels, boundary layers and separated ows compared to the standard k-e model [109]. The

constants of the realisable k-e model are: Ce1 = 1.44, Ce2 = 1.9, rk = 1.0 and re = 1.2.

The RNG ke model [110] is a modication of the classical ke model with better predictions of the recirculation length in

separating ows. The model is represented by the same equations (19)(21) of standard ke model but with a modied coefcient, Ce2, which is computed by the following equation:

C e2 C e2

C l g3 1 g=g0

;

1 b1 g3

22

702

Sk

q

2Sij Sij ;

i @ u

j

1 @u

;

2 @xj @xi

23

where S denotes the mean strain-rate of the ow and Sij the deformation tensor. The model constants are: Cl = 0.085,

Ce1 = 1.42, Ce2 = 1.68, rk = re = 0.72, b = 0.012 and g0 = 4.38. This model gives better results than the standard ke model

for separating ows, but fails to predict ows with acceleration [91].

Implementation of standard ke model and RNG ke model for pollutant dispersion from large tank-res [111114] and

street canyon ows [115], respectively, have been recently undertaken by the authors. In addition, high Reynolds number

turbulent ow past a rotating cylinder has also been examined by the authors and their colleagues. In Fig. 2, supercritical

streamline patterns are illustrated compared with laminar ones at Re = 200 and for some common examined cases (same

rotational rate) for ow past a rotating cylinder. More details for the study can be found in the work of Karabelas et al. [116].

2.3.6.3. The kx model

Another successful model and also widely used is the kx model. The initial form of the model was proposed by

Kolmogorov in 1942 [117]. An improved version of the model was developed by the Imperial College group under Prof. B.

Spalding [118]. Further development and application of kx model was performed by many scientists and engineers, but

the most important development was by Wilcox [108]. In the present paper, the most recent version of the model (Wilcox

(2006) kx model) is presented below [86,108]:

Kinematic eddy viscosity (vt) equation:

mt

s)

2Sij Sij

~ max x; C lim

x

;

b

(

k

;

~

x

C lim

7

:

8

24

@k

@k

@

j

u

@t

@xj @xj

i

k @k

@u

b kx sij

v r

:

x @xj

@xj

25

@x

@x

@

j

u

@t

@xj @xj

vr

@x

rd @k @ x

x @ ui

bx2

a sij

:

x @xj

x @xj @xj

k

@xj

k

26

The auxiliary relations and closure coefcients of the model are specied as follows:

a 0:52;

rd

b b0 f b ;

8

>

< 0;

> rd0;

:

b0 0:0708;

@k @ x

@xj @xj

60

@k @ x

@xj @xj

> 0;

b 0:09;

1 85vx

fb

;

1 100vx

vx

X X S

ij jk ki

3 ;

b x

Xij

i @ u

j

1 @u

;

2 @xj @xi

27

28

where Clim is the stress-limiter strength, fb the vortex-stretching function, vx the dimensionless vortex-stretching parameter

and Xij the mean-rotation tensor. The kx model is superior to the standard ke model for several reasons. For instance, it

achieves higher accuracy for boundary layers with adverse pressure gradient and can be easily integrated into the viscous

sub-layer without any additional damping functions [86]. In addition, the recent version of Wilcox (2006) kx model is

much more accurate for free shear ows and separated ows. The model still suffers from weaknesses when applied to ows

with free-stream boundaries (e.g. jets), according to the review paper by Menter [119].

2.3.6.4. More recent two-equation models

A more advanced turbulence model is the Shear Stress Transport (SST) model by Menter [120]. This model combines the

advantages of ke and kx models in predicting aerodynamic ows, and in particular in predicting boundary layers under

strong adverse pressure gradients. The model has been validated against many other applications with good results such as

turbomachinery blades, wind turbines, free shear layers, zero pressure gradient and adverse pressure gradient boundary

layers. Recent improvements of the model are an enhanced version for rotation and streamline curvature [121] and the

replacement of the vorticity in the eddy viscosity with the strain rate [119]. The mathematical formulation of the model

is not repeated due to space limitations, but it may be found in the above mentioned references.

Another class of two-equation models is the two-time scale models, with signicantly improved results compared to the

ke model. Hanjalic et al. [122] proposed a multi-scale model in which separate transport equations are solved for the

turbulence energy transfer rate across the spectrum. The mathematical formulation of the proposed turbulence model is

as follows [123].

kkP

;

vt C l

eP

29

703

Fig. 2. Streamline patterns at Re = 200, 5 105, 106, 5 106 and rotational rates a = 2, 3, 4, 5, 6, 7 and 8. L1 and L2 stagnation points are apparent for low

rotational rates (laminar ow), while A, B, C and Z are addressed for the super-critical Reynolds numbers [116].

704

DkP

@

@kP

Pk eP ;

m vt

@xj

Dt

@xj

30

DkT

@

@kT

v vt

eP eT ;

@xj

Dt

@xj

31

l @ u

i

DeP

@

@ eP

eP

e2

@u

C P1 Pk C P2 P C 0P1 kP

v vt

elmk eijk ;

@xj

Dt

@xj

kP

kP

@xm @xj

32

DeT

@

@ eT

eP eT

eT

C T1

v vt

C T2 2 :

@xj

Dt

@xj

kT

kT

33

The form of Eq. (29) has been obtained from simplifying the mean-Reynolds-stress (MRS) equation by considering normal

Reynolds stresses proportional to k and by taking the time scale for pressure strain to be kePP . The above mentioned Eqs.

(29)(33) use the following set of coefcients and functions:

kP

k

1

kT

; C T1 1:08

eP

; C 1:15;

eT T2

34

where P k ui uj @@xui . Here kp and kI are, respectively, the turbulence kinetic energy in the production and dissipation ranges,

j

Pk is the rate at which turbulence energy is produced (or extracted) from the mean motion, ep is the rate at which energy is

transferred out of the production range, eT is the rate at which energy is transferred into the dissipation range from the inertia range and e is the rate at which turbulence energy is dissipated (i.e. converted into internal energy).

It is worth mentioning that the proposed version of ke model performs better than the standard (single-scale) ke model

due to the fact that ep (rate at which energy is transferred out of the production range) replaces Pk (turbulence production) in

the dissipation rate (e) equation, simply because in ows where Pk is suddenly switched off, e is not expected to decrease

immediately. The present model gives better predictions than the single-scale ke model in plane and round jets [1].

The main advantage of the two-scale ke model is the combination of modelling the cascade process of turbulence kinetic

energy and of solving complex ows such as separating and reattaching ows. Improvements of the model may be achieved

by accounting for the proper empirical coefcients which affect the spectrum shape. Applications of the model for breaking

waves [124], plane synthetic and swirling jet [125], wake-boundary-layer interaction and compressible ow [126], may be

found in the literature. Finally, new models for non-equilibrium ows have also been developed by Klein et al. [127] with

satisfactory results.

2.3.6.5. Low Reynolds number modications

Most of the above models are applicable for turbulent ows at high-Re numbers, but are inaccurate for the prediction of

the ow in the vicinity of the wall, where viscous forces dominate. In order to treat this shortcoming, many scientists and

engineers have proposed a number of near-wall modications. These models with near-wall modications are referred to as

Low-Reynolds Number (LRN) models. A full list of these models is presented in the text by Wilcox [86] and in the review

paper by Patel et al. [128]. In the present paper two popular LRN models will be presented, the LamBremhorst ke model

[129] and Bredberg et al. kx model [130]. The mathematical formulation of the rst model can be written in the following

boundary layer form:

u

@k

@k

@

t

@x

@y @y

2

vt @k

@u

vt

v

e;

@y

rk @y

35

u

@ ~e

@ ~e

@

t

@x

@y @y

2

~e

~e2

vt @ ~e

@u

C e1 f 1 v t

v

C e2 f 2 E;

@y

rk @y

k

k

36

e e0 ~e

37

2

k

vt C l f l :

~e

38

The damping functions f 1 ; f 2 ; f l ; e0 and E and closure coefcients for the LamBremhorst ke model are presented below:

!3

2

0:5

20:5

0:05

k

k y

; f1 1

f l 1 exp0:0165Ry 1

; f 2 1 expRe2t ; Ret ; Ry

;

~ev

Ret

fl

v

2

39

705

The development of LRN ke models improved the original ke model by making it more compatible with the law of the

wall. However, LRN modications did not improve the problem with strong adverse pressure gradient. More details for the

difculty of LRN ke models to predict turbulent ows with pressure gradients may be found in Wilcox [86]. Finally, Patel

et al. [128] claim that any improvement in predicting ows with adverse pressure gradient would require modications to

the original ke model itself.

Another class of LRN models is the kx models and one of the most popular is the standard kx model by Wilcox [108].

Extensions and improvements of the model have been proposed by Wilcox [131], Peng et al. [132], Bredberg et al. [130],

among others.

The mathematical equations of Bredberg et al. kx model are as follows:

@k

@

@

j k

u

@t @xj

@xj

vt @k

Pk C k kx;

v

rk @xj

@x

@

@

j x

u

@xj

@t @xj

vt

rx

40

v v @k @ x

@x

x

t

Cx

C x 1 P k C x 2 x2 :

k k @xj @xj

@xj

k

41

vt C l f l

42

f l 0:09 0:91

!"

Re3t

( )#

2:75

Ret

1 exp

:

25

43

rk 1; rx 1:8:

44

The model of Bredberg et al. [130] presents improved results against the original Wilcox kx model, compared to DNS

and experimental data for three different cases (channel ow, backward facing step ow and rib-roughened channel ow).

Recently, an extension of the model to viscoelastic uids was proposed by Resende et al. [133].

2.3.7. Non-Linear Eddy Viscosity Models (NLEVM)

As mentioned earlier in Section 2.3.2, the Non-Linear Eddy Viscosity Models (NLEVM), may be dened as non-linear

extensions of the eddy-viscosity models in which Eqs. (13) and (15) can be rewritten in a more general form, in order to

include non-linear terms of the strain-rate [40]:

2

3

N

X

n

an T ij ;

45

n1

i

i

@u

@

1 @p

@

@u

i u

j

N:S:T:;

m m t

@t @xj

q @xi @xj

@xj

46

where N.S.T are non-linear source terms deriving from Eq. (45).

This class of models has been developed to overcoming the deciencies of eddy-viscosity models, in particular for twoequation models. There is a large number of NLEVM in the literature and they may be categorised as quadratic and cubic

models. Popular quadratic models have been proposed by Gatski and Speziale [134] and Shin et al. [135], among others.

The rst model is a high-Re ke model which supports separation in adverse pressure gradient ows. The model of Shin

et al. [135] has shown improved results for backward facing step compared to classical linear eddy viscosity models, but

it also suffers with rotational effects especially for channel ow [136].

The mathematical formulation of the cubic model by Craft et al. [137] is selected for presentation here, as a general form

of the category. The anisotropic tensor and turbulence kinetic energy are dened as:

aij

ui uj 2

dij

3

k

and k

1

uk uk :

2

47

vt

vt

1

vt

vt

1

vt k

Sik Skj Skl Skl dij c2 Xik Skj Xjk Skl c3

Xik Xjk Xlk Xlk dij c4 2 Ski Xlj

Sij c1

~e

~e

~e

~e

3

3

k

vt k

2

vt k

vt k

Skj Xli Skl c5 2 Xil Xlm Smj Sil Xlm Xmj Slm Xmn Xnl dij c6 2 Sij Skl Skl c7 2 Sij Xkl Xkl ;

~e

~e

~e

3

aij

where Sij is the mean strain-rate tensor and Xij the mean vorticity tensor:

48

706

Table 1

The model coefcients [136].

c1

c2

c3

c4

c5

c6

c7

f

0:05 f q

l

f

0:11 f q

l

f ~

S

0:21 f ~Sq X~ =2

l

0.8fc

0.5fc

0.5fc

Cl

fl

11:8n

1:1

rn

fq

fc

rn

0:5

10:0086g2

r 2n

10:45n2:5

10:6A2 0:2A3:5

2

r

n

o

~

Rt

1 1 exp2A2 3 1 4 exp 20

Sij

i @ u

j

@u

@xj @xi

p~e

~

e10:8 expRt =30

Xij

i @ u

j

@u

:

@xj @xi

49

~ the dimenThe empirical coefcients of the model are presented in Table 1, where ~

S is the dimensionless strain parameter, X

sionless vorticity parameter, ~e the homogenous dissipation rate, which are dened as:

~S k

~e

q

Sij Sij =2;

~

X

k q

Xij Xij =2;

~e

vt c l

k

:

~e

50

The model appears better compared to ordinary linear eddy viscosity models (e.g. for impinging jet ows). The computational time is approximately 20% more compared to a low-Re ke model. One drawback of the model is its performance for

convex surfaces, according to Craft et al. [137].

Another popular cubic low-Re ke model was developed by Apsley and Leschziner [138], with its free parameters calibrated with data from DNS data for channel ow. The model leads to better results for airfoil and diffuser ows compared

to other linear and nonlinear EVM. For more details for the NLEVM, the interest reader is directed to the review paper by

Hellsten and Wallin [136].

2.3.8. Recent advances in eddy viscosity modelling

It is worth mentioning some recent eddy-viscosity models, such as Durbins t2-f model (also known as v2f) [139] and ff

model [140]. The t2f model is based on the elliptic relaxation concept and employs two additional equations, apart from the

k and e ones. One for the velocity scale t2 and one for the elliptic relaxation function, f. The main motivation for the development of this model was the improved modelling in the vicinity of the wall (near-wall turbulence). More applications (e.g.

rotating cylinder, rotating channel ow, axially rotating pipe and square duct) and validation of the model with experimental

and DNS data, may be found in the work of Durbin and Petterson [141].

The ff model is based on the similar concept of elliptic relaxation, but instead of solving the t2 equation it solves for the

velocity scale ratio f = t2/k [140]. The full equations of the ff model, which are similar to those of the t2f model, are [140]:

vt C l fks;

51

Dk

@

vt @k

P e;

v

Dt @xj

rt @xj

52

De

@

Dt @xj

vt @ e

C e1 P C e2 e

v

;

re @xj

s

L2 r2 f f

Df

@

Dt @xk

P

2

f

;

c1 C 02

3

s

e

vt @f

f

f P:

v

k

rf @xk

53

54

55

Completeness of the model is achieved by Durbins [142] realizability constraints, combined with the lower bounds (Kolmogorov time- and length- scale):

"

!

#

v0:5

a

p

;C

;

s max min ;

e 6C l jSjf s e

k

"

L C L max min

1:5

!

3 0:25 #

0:5

k

v

; p

; Cg

;

e

6C l jSjf

56

57

where a 6 1 (recommended a = 0.6 [140]). The coefcients of this model are: Cl = 0.22, Ce1 = 1.4(1 + 0.012/f), Ce2 = 1.9,

c1 = 0.4, C 02 0:65, rk = 1, re = 1.3, rf = 1.2, Cs = 6.0, CL = 0.36, Cg = 85.

707

Fig. 3. Mean velocity streamlines over the swept wing surface, highlighting regions of interest [145].

It should be noted that the ff model is more stable [141] compared to the t2f model. Both models are better for computing wall-bounded ows compared to the classical low-Re two equation models (e.g. kx and ke), but they are still weak

against DSM (introduced next in Section 2.4) and advanced NLEVMs. The ff model has been extensively validated with

experimental and DNS data for plane channel, backward-facing step and multiple-impinging jets ows, presenting satisfactory agreement.

Recently, a new robust version of the t2f model was proposed by Billiard and Laurence [143] with improved numerical

stability, known as the BL-t2 =k. The model is based on the elliptic blending method of Manceau and Hanjalic [144] and was

validated for pressure induced separated ows, as well as buoyancy impairing turbulent ows, with satisfactory results.

More detailed evaluation of the model against other turbulence models and test cases (e.g. 3-D diffuser and swept wing)

is presented in the work of Billiard et al. [145]. In Fig. 3, the mean velocity streamlines for swept wing are compared to data

from Implicit Large Eddy Simulation (LES). Both models capture the leading edge vortex but the secondary vortex region

(red-dashed line) is reproduced by the EBRSM model and secondly from the BL-t2 =k model. A detailed review of t2f model

evolution may be found in the work of Billiard and Laurence [143].

2.4. Differential Second-Moment (DSM) and Algebraic Stress Models (ASM)

A type of turbulence closure models with great expectations to replace the widely used ke model is the Differential

Second-Moment (DSM) or Reynolds Stress (RS) or Mean-Reynolds Stress (MRS) model. DSM presents natural superiority

compared to the two equations turbulence models, as it is physically the more complete model (history, transport and

anisotropy of turbulent stresses are all accounted for). More specically, DSM closure models explicitly employ transport

equations for the individual Reynolds Stresses, u0i u0j (as well as for u0j T 0 ), each of them representing a separate velocity scale.

The transport equation of u0i u0j for an incompressible uid, excluding effects of rotation and body force, may be written in

general symbolic tensor form as follows:

58

where Lij is the local change in time, Cij the convective transport, Pij the production by mean-ow deformation, /ij the stress

redistribution tensor due to pressure strain, Dij the diffusive transport and eij the viscous dissipation tensor. The Lij, Cij and Pij

terms do not require any modelling and are given by the following equations:

Lij C ij

@

@

m

u0 u0 ;

u

@t

@xm i j

j

i

@u

@u

:

Pij u0i u0m

u0j u0m

@xm

@xm

59

60

The remaining terms, /ij , Dij and eij need to be modelled. The simplest way to model the viscous dissipation tensor is by

assuming local isotropy:

2

3

eij edij ; e v

@um @um

;

@uk @uk

61

where e is the turbulence dissipation and dij the Kronecker unit tensor. The diffusion of turbulence (Dij) is usually treated by

the popular DalyHarlow model [146]:

708

0 0

Dij

@ui uj

@

k

C s u0l u0m

@xk

e

@xl

!

with C s 0:25:

62

Instead of the popular DalyHarlow model, there are more advanced models that can be used and have been developed

over the years, such as the models by Magnaudet [147] and Nagano and Tagawa [148]. More details and validation of the

above mentioned models can be found in the review paper by Hanjalic [149].

The radically new feature of the DSM-equation is the pressure-strain redistribution term (/ij ), which does not appear in

the exact solution of ke equation. This suggests that the pressure-strain term only serves to redistribute the turbulence

energy among its components and to reduce the shear stresses, thus tending to make the turbulence more isotropic. The

unknown correlations appearing in the DSM-equation are either determined by a transport equation or else they are

expressed in terms of second-order correlations u0i u0j themselves; the latter procedure, closing the DSM-equation at its

own level, is often referred to as second-order closure. The redistribution term, /ij , is usually modelled by the Isotropization

of Production (IP) model [150]:

2

e

dij

/ij C 1 u0i u0j dij k

C 2 Pij P kk

with C 1 1:8 and C 2 0:6:

3

k

3

63

Improvements of the IP pressure-strain model are the LRR-QI model by Launder et al. [151], the SSG model by Speziale

et al. [152], the CL model by Craft and Launder [153], and LT model by Launder and Tselepidakis [154]. For detailed evaluation of the models, the interest reader is referred to the work of Hanjalic and Jakirlic [155]. DSM closure models predict

more accurate physical phenomena which involve streamline curvature, strong pressure gradients, swirling and system rotation effects [155].

It is important to mention that the initial versions of DSM models could not perform very well in handling the return to

isotropy [1]. They may, however, work well in ows dominated by other effects. In addition, the DSM models do not always

perform better than the two-equation models. For instance, recent numerical simulations in street canyon ows performed

by Koutsourakis et al. [115] indicated that DSM performance was not good enough according to the theoretical expectations,

while the RNG ke model exhibited better results. The evaluation of the models (DSM, RNG ke and standard ke) was done

comparing to different experimental and numerical data (LES). Figs. 4 and 5 present numerical results for the velocity proles compared with experimental data. It is concluded that DSM, at least as it has been applied in that study, is not more

useful than simpler models for practical use. However, both modelling and experimental uncertainties are high, so that extra

attention is required in order to draw any denitive conclusions about the quality of the models.

Main disadvantages of DSM are the difculty in the modelling of more terms in the turbulence equations and the

increased demand on computer resources. The new generation of DSM closure models have solved most of the above-mentioned difculties but the computational demands are roughly twice as large as those for the two-equation models, for high Re number ows using wall functions [155].

It is worth noting the Elliptic Blending DSM by Manceau and Hanjalic [144], which belongs to the category of advanced

DSMs. The model is based on the DSM of Durbin [159] but, instead of resolving equations for the stress components, it adopts

a single elliptic equation [160]. Implementation of EBDSM for predicting impinging jet ows can be found in the work of

Thielen et al. [161].

Another interesting approach is the hybridization of DSM with an eddy viscosity model, recently developed by Basara and

Jakirlic [162]. The model combines the advantages of DSM along with the robustness of ke model and it is known as Hybrid

Turbulence Model (HTM). It may be used as an initialization model for DSMs in order to stabilize it and reduce the

Fig. 4. Experiment 4: Non-dimensional horizontal velocity proles at leeward and windward side of a street canyon with aspect ratio H/D = 1 and very

rough walls. Uref is the free stream velocity, equal to 8 m/s. Experimental data are extracted from Kovar-Panskus et al. [156]. CFD results with the k-e model

from the original paper are also included [115]. N. Koutsourakis, J. Bartzis, N. Markatos, Evaluation of Reynolds stress, k-e and RNG k-e turbulence models in

street canyon ows using various experimental datasets, Environmental Fluid Mechanics 12 (2012) 379403. This is Fig. 6 in the publication in which the

material was originally published. With kind permission from Springer Science and Business Media.

709

Fig. 5. Experiment 6: Performance of the three turbulence models against experimental data extracted from Li et al. [157] in a street canyon with aspect

ratio H/D = 2. LES results of Li et al. [158] are also included [115]. N. Koutsourakis, J. Bartzis, N. Markatos, Evaluation of Reynolds stress, k-e and RNG k-e

turbulence models in street canyon ows using various experimental datasets, Environmental Fluid Mechanics 12 (2012) 379403. This is Fig. 8 in the

publication in which the material was originally published. With kind permission from Springer Science and Business Media.

computational time. The model has been tested for many ows such as ow around a car, axially rotating pipe, 180 turned U

bend, backward-facing step and round jet impinging on a at plane. The results were better than EVMs and closer to those

obtained by full DSM.

Simplied versions of DSMs are the Algebraic Stress Models (ASM) or Explicit Algebraic Reynolds Stress Models (EARSM)

[136], obtained by eliminating the transport terms, using instead those of the kinetic energy equation. EARSM have the reputation of being simple and easy to implement for boundary layer ows, while for elliptic, recirculating ows they are very

unstable. In addition, their performance is dependent on the DSM from which they were derived. This category constitutes

an intermediate-level between DSM and eddy viscosity models. EARSM are characterised by less computational demands

and higher accuracy compared to LEVM.

First attempt for the derivation of EARSM was done by Pope [163] for two-dimensional ows and it was later extended

and rened by Gatski and Speziale [134] and by Jongen and Gatski [164] for 3-D ows. The EARSM are popular for predicting

aeronautical ows, in particular with the model by Wallin and Johansson [165].

Due to space limitations, the interested reader is directed to the recent review papers by Hellsten and Wallin [136] and

Alfonsi [106], for the mathematical equations of EARSM.

2.5. Two-uid models of turbulence

Undoubtedly, the main idea behind a large number of turbulence models derives from the notion of Boussinesq, who

introduced the idea of an effective viscosity, and of Prandtl who conceived the notion of turbulence mixing phenomena being

very similar to those treated by the dynamical theory of gases. The mathematical background of convectional turbulence

theory reects only the unstructured diffusion of the molecular-collision process; large structure formation and growth,

and ne-structure creation and stretching, are nowhere to be found.

The above-mentioned facts led Spalding [166], among others, to the development of the two-uid theory, briey presented below. The origins of two-uid model ideas are to be found back in the 1940s and 1950s, particularly in the eld

of turbulent combustion.

The following two-uid model results in the prediction of the intermittency and of the conditional ow variables within

the turbulent and non-turbulent zones of the ow. The model was proposed by Spalding [166] and then developed by Malin

[167] for the investigation of intermittency in free turbulent shear ows.

The Spalding model is designed on an analogy between intermittent ows and two-phase ows, rather than on any specic and rigorous closure of conditional-averaged transport equations. It is supposed that two-uids share occupancy of the

same space, although not necessarily at the same time, their share of space being measured by the volume fractions. There

are many ways in which the two-uids can be distinguished. In case of turbulence intermittency computations, it is convenient to dene the uids as turbulent and non-turbulent. The equations governing the motion of the turbulent and nonturbulent uids are given in detail by Markatos [1,79].

Improvement and expansion of the model for turbulent combustion was presented by Markatos and Kotsifaki [168]. Furthermore Shen et al. [169] used a two-uid model to simulate turbulent stratied ows, while Yu et al. [170] established a

modied two-uid model to simulate the ow and heat transfer characteristics of air curtains in an open vertical display

cabinet. For RayleighTaylor mixing so far only two models have been developed, namely Youngs model [171] and twostructure two-fluid two-turbulent (2SFK) model [172]. Liu et al. [173] applied two-uid model for predicting the ow in

UV disinfection reactor. Finally, Cao et al. [174] used the model for the design of air curtains for open vertical refrigerated

display cases. The numerical results compared with experimental data present good agreement.

710

Fig. 6. Turbulence energy spectrum. The gure has been redrawn based on [175].

Another modelling approach, promising to be more accurate and of wider applicability than RANS and less computationally demanding than DNS, is the Large Eddy Simulation (LES) approach. In LES of turbulence, the important large scales are

fully resolved whilst the small sub-grid scales are modelled. The main advantage of LES compared to RANS models is that in

the former only the small, isotropic turbulent scales are modelled and not the entire spectrum (Fig. 6) as it is the case in the

latter. The LES approach is extremely useful for the investigation of turbulence at high Reynolds numbers, for the development and assessment of new turbulence models, and for the prediction of complex ows where other turbulence models

may prove inadequate [176,177].

The rst attempts of LES are found in the pioneering works of Smagorinsky [178] and Lilly [179] in meteorology and Deardorff [180] in engineering. Since then, LES has seen a tremendous popularity for the study of turbulent ows. The development and testing of LES has concentrated at rst on isotropic turbulence by Kraichnan [181] and Chasnov [182] and on

turbulent channel ow by Deardorff [180], Schumann [183], Moin and Kim [184] and Piomelli [185]. The basic steps of

LES method, according to Pope [24] and Berselli et al. [25] are: (a) a ltering operation of the NS equations; (b) a closure

model for SGS stress tensor; (c) imposition of the appropriate boundary and initial conditions with special care for near wall

modelling; (d) selection of the suitable numerical method for spatial and temporal discretization of NS equations and (e)

performance of the numerical simulation.

Instead of time-averaging, a spatial ltering approach is adopted in order to separate the resolved (large-eddy) eld from

the small-eddy (sub-grid) eld. A lter operation is dened by the convolution [186]:

/x

dx0 ;

/x0 Gx; x0 ; D

64

is the lter width. The lter function is responwhere D is the domain of integration, G is the specied lter function, and D

sible for determining the structure and size of the small scales that are captured [32]. More details about spatial ltering

techniques are presented in Aldama [187], Pope [24] and Sagaut [26]. An illustrative example of LES for single-phase, incompressible ow is given below.

Assuming that the ltering operation presented above commutes with temporal and spatial derivatives, the ltered governing equations can be expressed as follows:

Filtered continuity equation:

i

@u

0:

@xi

65

i @ u

i u

j

@ sij

i

@u

1 @p

@2u

v

;

@t

@xj

q @xi @xj

@xj @xj

66

where sij denotes the Sub-Grid Scale (SGS) stress tensor, dened as:

67

The SGS stress tensor is symmetric, invariant to Galilean transformation and plays important role for the dynamic coupling between the small and large scales of turbulence.

as:

Denoting the uctuation /0 of / with respect to /

/ /0 ;

/

68

711

sij ui uj ui uj ui uj ui uj ui u0j u0i uj u0i u0j ;

|{z}

Lij

|{z}

C ij

|{z}

69

Rij

where Lij is the Leonard term, C ij is the Cross term, and Rij the Reynolds term according to the work of Clark et al. [188].

Each of these terms describes physical interactions between the scales that arise from the LES approach. More specically,

the Leonard term, Lij , expresses the interactions between resolved scales and it is an explicit term that can be computed from

the ltered eld without a modelling approach. Subsequently the use of a spectral or sharp cut-off-lter transforms the

Leonard terms in aliasing error [32]. Moreover, the Cross term, C ij , represents the interactions between sub-grid and large

scales. The Leonard and Cross terms in RANS modelling are equal to zero. On the other hand, in LES the Cross and Leonard

terms are approximately equal. Finally, the Reynolds term, Rij , represents the interactions between sub-grid scales.

The decomposition of the SGS stress tensor, as presented above in Eq. (69) is known as Leonard or triple decomposition

[186]. According to Sagaut [27], another way to decompose Eq. (69) is by double decomposition but with a sharp cut off lter.

Hence Eq. (69) can be written as:

sdij ui uj ui uj ui u0j u0i uj u0i u0j :

|{z}

C ij

|{z}

70

Rij

It is necessary to nd a satisfactory model for the sum of Cross and Reynolds terms which constitute a signicant part of

the turbulence energy spectrum. The accuracy of LES is dependent on the model for the SGS stresses, which should ensure

the accurate transfer of energy between unresolved and resolved turbulent scales [32]. The suitability of an ideal sub-grid

scale model depends on several properties given by Boris [189] and not repeated here.

In the literature, there are many available SGS models for treating the Reynolds and Cross terms, but none of them can

provide a fully satisfactory solution for the SGS modelling. The most common selection for SGS modelling is based on the

eddy-viscosity hypothesis. In this category of models, the length and velocity scales are specied and combined with a Boussinesq relationship. The most well known and oldest model of this category is the Smagorinsky model [178], which suffers

from many weaknesses, such as: the failure to predict the inverse energy transfer (backscatter), the tuning of Smagorinsky

constant (Cs), the failure to eliminate the eddy viscosity near to the walls and the requirement of a damping function.

An implementation of the Smagorinsky model along with a wall-sensitive length scale for predicting turbulent ow

study past a rotating cylinder has recently been investigated by Karabelas [190]. In Fig. 7, the selected O-type grid for the

simulations is presented together with the system of reference for the numerical set up.

In that Karabelas study, the physics and the load performance of a rotating cylinder subjected to uniform ow are

explored. The ow is resolved by LES (Fig. 8) and the Re number based on the diameter is equal to 150,000. Based on this

Re number, the ow is characterised as separated turbulent ow implying that it becomes turbulent close to the point

of separation. Further downstream the wake is fully turbulent. One of the major outcomes from that study is that the rotation of the cylinder alters the place and nature of transition. At high rotational rates, turbulence is generated without being

necessarily accompanied by immediate separation of the ow. The total resolved kinetic energy is plotted in Fig. 9. The high

values of kinetic energy in the wake are not only due to turbulence but also due to the considerable vortex shedding. Phase

averaging is needed to distinguish the high frequency from the low frequency modes of the uid motion.

Fig. 7. Computational domain (side section) and the system of reference for the numerical simulations. The azimuth angle h is measured in a clockwise way

while the cylinder rotates in the counter-clockwise direction. A panoramic and a close-up view of the structured grid adopted are also presented.

Reproduced with the authors permission [190].

712

Fig. 8. Time-averaged dimensionless stream-wise velocity along the centerline y = 0. Comparison of the numerical results against the Breuers and

experimental data of Cantwell and Coles, reproduced with the authors permission [190].

Fig. 9. Illustration of the total kinetic energy of the uctuations kf plotted in the near wake for the examined spin ratios, reproduced with the authors

permission [190].

In order to cope with the limitations of the classical Smagorinsky model an improved dynamic version has been developed by Germano et al. [191]. The main idea behind the dynamic version is that the coefcients of the model are specied as

the calculation progresses, such that a suitable local value for Cs is determined. The aforementioned idea is accomplished by

a second lter operation (test lter) to the already ltered equations of LES. The main advantage of this procedure is the good

behaviour near to the walls (eddy viscosity tends asymptotically to zero). However, this approach leads to a system of equations which, in general, is very difcult to satisfy with a single constant Cs. The selected treatment of the aforementioned

deciency distinguishes the newer versions of the model. The Lilly model [192] adopts a least-square approach, in order

to minimize the error for constant Cs and is the most widely used approach. However, both versions of the model may suffer

from large uctuations in Cs, which can create instabilities and as a result a local averaging in homogeneous directions is

713

necessary. In order to avoid it, the use of the localised dynamic model by Ghosal et al. [193] or the Lagrangian dynamic model

by Meneveau et al. [194] is recommended.

Another important class of SGS models accounting for the backscatter phenomenon is the similarity models

[188,195,196]. This category of models assume that the estimation of SGS tensor can be achieved by using a second ltering

technique with lter size equal to or larger than the initial lter for the resolved scales. The most well known model of this

class is the Bardina model [195] using polynomial inversion [197]. The Bardina model and, in general, all the similarity models are characterised by the presence of the backscatter phenomenon. However, the similarity models under-predict the SGS

dissipation and demand high computational cost because of the several explicit ltering processes involved.

An alternative class of models that combine the good characteristics of eddy-viscosity and similarity models, is known as

mixed models [198200]. Thus, a mixed model is obtained by introducing an additional expression for eddy-viscosity in a

similarity model. Mixed models exhibit better results compared to the classical Smagorinsky model and also present the

backscatter phenomenon. Mixed models have been proposed by Zang et al. [198] and Vreman et al. [199], among others.

Another type of SGS models which is based on the high-order approximation of the inverse lter is known as deconvolution models [201,202]. The Approximate Deconvolution Model (ADM) by Stoltz and Adams [201], in which a high-pass lter is adopted for modelling the interaction between the unresolved and resolved scales, has been tested for channel ows

[201], incompressible wall-bounded ows [203] and compressible ows [202]. The numerical results are compared to DNS

data and exhibit good agreement. More applications of the model may also be found in the work of Adams et al. [204] and in

the textbook by Geurts [205].

Three more classes of SGS models are the Structure function models [206,207], the SGS velocity models [208,209] and the

regularisation models [210215]. One of the advantages of the latter class of model is that the regularisation principle [213]

allows a transparent modelling between the model equations of the system and the NavierStokes equations. As a result,

fundamental properties of NS equations can be shared between them. In addition, there is no need for any model coefcient

or width of test lter. In this category, the most widely used model is the Leray-a [211,216] which presents a general regularisation form of NS equations. It is also worth mentioning that Leray-a model has been used as a closure model for large

Reynolds number channel and pipe ows [212214]. The numerical results obtained appear in excellent agreement with

empirical data from large Reynolds number ows [217].

Wall resolving LES is a method that near to the wall, where the energy spectrum is made up of anisotropic currents, performs a direct numerical simulation; away from the wall isotropic currents make up a large fraction of the energy spectrum

and are modelled by the use of SGS model. Therefore, the mesh pitch away from the walls may be greater and computational

time is shorter than required by direct numerical simulation. Hence, LES is only meaningful on a rened grid; on a regular

grid ne enough to resolve wall currents the subgrid scale model is unnecessary.

The required resolution of a wall-bounded resolved LES should be sufcient enough in order to resolve the wall layer

according to Piomelli [218]. More specically, Dx+

50150, Dz+

1540 and the rst point in the wall normal direction

is at y+ < 1. However, with the help of the appropriate wall model the required resolution can be adjusted to Dx+

100

160, Dz+

100300 and the rst point in the wall-normal direction should be at y+ = 30150 [218]. More details for wall

modelling may be found in the review paper by Piomelli and Balaras [219].

2.6.1. Validation of the LES approach

One of the most important issues of LES approach is the validation of the model, mainly for accuracy, sensitivity and efciency. A common approach for validation is based on the capability of a given LES to present satisfactory results on a class of

ows [27].

Two related issues are the efciency and sensitivity of LES approach. Sensitivity expresses the robustness of the method,

which is the variability of the results according to model parameters such as the mesh size, explicit constants, etc. Efciency

estimates the cost that should be available in order to achieve a given level of accuracy for the various ows.

Accordingly, the above-mentioned issues denote that a good LES depends strongly on the denition of the error. Therefore, a LES approach can be very good for one given error measure and bad for another. For instance, a LES method can produce satisfactory results for engineering applications and very bad results for high-level statistics. As a result, the validation

methodology should be explicitly associated with a clear purpose in terms of future use. The validation process for LES

approach is characterised by two alternatives [26]:

A priori validation: In this approach data from a DNS database are used to give exact solutions, in order to compare with

various hypotheses or models in a purely static way. Hence, this method presents good results relating to the nature of nonlinear interactions, but is characterised by the disability to predict the time properties of sub-grid closures.

The second alternative is the a posteriori validation. This approach is characterised by performing numerical simulations

with LES method and comparing its results with a reference solution. This dynamic approach takes all the numerical factors

into consideration.

Geurts and Leonard [220], for reasons described in their paper, proposed the following three criteria for a sound and predictable LES: (a) adjusting the lter width while adjusting the grid resolution; (b) avoid numerical methods with dissipative

characteristics; (c) implementation of the dynamic modelling approach.

The full validation problem for a given ow is equivalent to nding the space of the solutions spanned by a given LES

method and calculating the solution(s) with the minimum error [27]. The control and estimation of the error in LES which

is generated by the SGS modeling and the adopted numerical method (numerical errors) are difcult to deal with. A recent

714

attempt to quantifying the error by the sub-grid modelling and discretization method was presented by Geurts [221,222]. In

addition, Celik et al. [223] proposed the LES_IQ method which represents an index of LES quality. Recently, several attempts

for the quantication of the error in LES have been undertaken by Geurts and Frohlich [224], Meyers et al. [225] and Freitag

and Klein [226], who described inroads into attributing errors to numerical methods versus SGS models [227].

2.7. Monotone Integrated LES (MILES) and Implicit LES (ILES)

The implementation of ltering operation can be invoked explicitly and implicitly. In explicit LES an extra forcing term

(SGS model) is added to the NavierStokes equations in order to eliminate the resolution error. On the other hand, in Implicit

LES (ILES) no extra term is added to the NavierStokes equations, but the numerical method is selected in such a way that the

numerical and resolution errors will cancel out [26].

Instead of explicit SGS model a class of Non-oscillatory Finite Volume (NFV) methods is adopted for incorporating the

effects of unresolved scales. These methods preserve the monotonicity to the integral form of NS equations and, as a result,

the correct rate of energy dissipation on the resolved scales is obtained. In this category of NFV schemes belong the FluxCorrected Transport (FCT) by Boris and Book [228], the Monotonic Upstream-Centered Scheme for Conservation Laws (MUSCL) by Bram Van Leer [229], the Piecewise Parabolic Method (PPM) by Colella and Woodward [230], the Total Variation

Diminishing (TVD) by Harten [231] and the second-order Godunov method by Colella [232]. The above mentioned schemes,

initially used for predicting compressible ow with high accuracy, have been extended to deal with subsonic compressible

and reactive ows. It is also worth noting the implementation and testing of ILES for incompressible ows [233] (e.g. turbulent channel ow, ow over a cylinder and sudden expansion), supersonic jet ows [35] and large-scale urban ows [234].

Boris [235] introduced the use of the term Monotone Integrated Large Eddy Simulation (MILES) [236]. The basic idea

behind the MILES approach is the use of monotone advection schemes which mimic the action of what turbulence modelling is supposed to do. Monotone advection schemes which satisfy physical properties (e.g. positivity and causality) preserve

stability regardless of turbulence and can have a minimal LES lter and a built-in sub-grid turbulence model coupled continuously to the grid-scale errors [236]. According to Boris [236] the aforementioned physical properties are adequate to provide efcient transfer of the sub-grid motions, by minimizing the effect of the numerical lter of the well-resolved scales in

the resolved grid.

The MILES approach has been adopted for predicting turbulent ows such as beam-channel interactions, open-air ammunition-destruction, shock waves, geophysical and atmospheric ows [236]. It is concluded, after many years of research on

MILES, that the monotone (positivity-preserving) algorithms appear to lead to good agreement with theory and data, being

free of any eddy viscosity or explicit sub-grid model. MILES is very useful for complex problems which include compressible

and multi-physics ows with complex geometries because it is easy to implement.

2.8. Unsteady Reynolds-Averaged NavierStokes (URANS)

Unsteady Reynolds-Averaged NavierStokes (URANS) or Transient Unsteady Reynolds-Averaged NavierStokes (TRANS)

is an alternative approach to LES. URANS is characterised as the most complex version of RANS. In URANS, a time solution of

the conventional RANS is performed for 3D unsteady problems, with or without special treatment for the ow instability

[85]. The equations of URANS are similar to the RANS models but include the additional unsteady term. The most important

aspect for an URANS model is the correct selection of the suitable turbulence model and the discretization scheme (which

must be not too dissipative). URANS performed well for vortex shedding behind bluff bodies [237] and surface mounted cube

[238] and around cars [234], among other applications. Application of TRANS for predicting environmental turbulent ows

and pollutant dispersion have been investigated by Hanjalic and Kenjeres [160] with good results. For more details, the interest reader is directed to the reviews by Speziale [240], Spalart [241], and the recent one by Kenjeres and Hanjalic [242].

The second generation URANS (2G-URANS) models, according to the terminology of Frohlich and von Terzi [36], is also

worth mentioning. They are the Scale-Adaptive Simulation (SAS) model [243] and the Partially Filtered NavierStokes

(PANS) model [244]. The former is based on revisiting the kkl model of Rotta [245] by Menter and Egorov [246], and the

latter is based on the transformation of a RANS model into a URANS by just changing the dependant coefcients.

The PANS model is now well proven with many practical and complex turbulent ows applications such as channel ow,

external car aerodynamics, and simplied train geometry with a crosswind [247]. Recently, a low-Re number version of the

model has been developed by Ma et al. [248] and another version with an embedded LES model has been proposed by Davidson and Peng [249].

The SAS model was modied by Menter and Egorov [246] with the use of the kx SST model and as a result a new version

of the model was developed, known as SST-SAS. The model has been tested for different type of ows such as axisymmetric

hill, channel and asymmetric diffuser [250]. It is concluded that the SST-SAS is an equivalent VLES (introduced next in Section 2.9), because it represents the spectrum of resolved scales between LES and URANS [175].

A similar model to PANS is the Partially Integrated Transport Model (PITM) [251,252] with both models using RANS and

LES framework. In the LES framework both models avoid to adopting a length scale as the width of a lter function. These

models resolve a large part of the turbulence energy spectrum and as a result they are named as second generation URANS

[36]. The difference of PANS against PITM is that in the rst the coefcients which are responsible for the turbulence diffusion in k and e are modied. The PITM model has been validated against numerical and experimental data for the decay of

715

homogeneous turbulence, fully turbulent channel ow and unsteady turbulent ow with periodic forcing, with promising

and encouraging results [251,252].

2.9. Very LES (VLES) and Detached-Eddy Simulation (DES)

The Very Large Eddy Simulation (VLES) was proposed by Speziale [240], and was rened and reformulated by Johansen

[253]. VLES combines RANS and LES in such a way that use is made of the advantages of those approaches. The combination

of these strategies involves numerical resolution in time and space only for the very large eddies, while a signicant part of

the turbulence spectrum requires modelling [85]. VLES constitutes a turbulence approach with coarser grid requirements

compared to LES and the idea of VLES originates from the atmospheric-science community. The main advantage of this

method is that it can preserve efciency and accuracy within reasonable computer-time demands.

A similar approach with the same characteristics of VLES is the Partially Resolved Numerical Simulation (PRNS) technique

by Liu and Shih [254]. PRNS is based on the assumption that the small scales of turbulence have small timescales and as a

result a lter with a xed width is adopted in order to distinguish the large scales of turbulence. Hence, according to the

value of the temporal lter width the PRNS approach operates between RANS and LES modes. The PRNS approach was tested

for turbulent pipe ow [175,254], non-reacting ow in a single injector ame tube [254], internal reacting and external static

stall ows [255].

There are more approaches that follow the VLES methodology but it is not possible to describe all of them, due to space

limitations. However, the interested reader may nd more information in the Limited Numerical Scales (LNS) by Batten et al.

[256], the approach by Ruprecht et al. [257], the self-adapting turbulence model by Perot and Gadebusch [258,259] , the

method by Hsieh et al. [260], and the recent turbulence modelling approaches by Labois and Lakehal [261] and Han and Krajnovic [262].

The Detached-Eddy Simulation (DES) originated in 1997 [263] and was rst used in 1999 [264], with the purpose of coping with massively separated ows at high-Reynolds number. DES combines LES and RANS approaches, based on the turbulence length scale and the grid spacing. Hence, LES is used for regions of massive separations and RANS within the boundary

layer. The ofcial denition of DES, in accordance with Travin et al. [265] is A three-dimensional unsteady numerical solution using a single turbulence model, which functions as a sub-grid scale model in regions where the grid density is ne

enough for a large-eddy simulation and as a Reynolds-averaged model in regions where it is not. More details about the

equations, advantages, limitations and implementation of DES can be found in the review paper of Spalart [264].

Improvements of the DES model such as Delayed DES (DDES), Improved DDES (IDDES) and SST-DDES have been presented

for different case studies. DDES is characterised as a modied version of DES approach in order to tackle the nonphysical

behaviour in the vicinity of the boundary layers. Therefore, a more generic formulation of the shielding function was proposed which is dependent only on the wall distance and the eddy viscosity. The term IDDES was originated by Shur et al.

[266], who indicated that the original DES could be transformed with wall modelled LES capabilities. Finally, the SST-DDES

constitutes the combination of DDES by Spalart [267] with the SST-DES approach of Strelets [268]. More details, applications

(e.g. sharp edged delta wings, bluff bodies, ground vehicles, active ow control by suction/blowing, vibrating cylinders with

strakes, cavitation inkjets, building, air inlets, aircraft in spin, high-lift devices [264]) and modications for the above mentioned versions of DES can be found in the works of Spalart et al. [267], Spalart [264] and Gritskevich et al. [269].

2.10. Hybrid RANS/LES strategies

All the aforementioned approaches in the last two Sections 2.8 and 2.9 are expressed by the combination of LES with

RANS models. According to Leschziner et al. [270] all those approaches, excluding very few cases, can be divided into four

basic types; wall laws or wall functions, zonal schemes, seamless schemes and hybrid schemes.

The hybrid schemes were introduced in order to overcome one of the main disadvantages of LES for wall-bounded turbulent ows. For a wall-bounded turbulent ow the demand of ne grid resolution is great, in particular for high Reynolds

number ows, in the vicinity of the walls. Therefore, the hybrid RANS/LES strategies decrease the demand for ne grid resolution at the wall region. This can be achieved by using a low-Re RANS model at the wall region, while for the outer region

an LES approach is adopted. Thus the near wall turbulence is modelled instead of being fully resolved and thus a coarser grid

may be used.

Much work has been dedicated to Hybrid RANS/LES approaches [271276]. Due to space limitation the interested reader

is directed to the review paper by Frohlich and von Terzi [36],

3. Applications of DNS and LES to ows in pipes and ows with a free surface

This section is devoted to listing interesting applications of the DNS and LES approaches to two well-researched ow

types. This discussion is valuable (a) in order to present interesting practical applications of these two important turbulence

modelling methods, (b) in order to highlight some important turbulent-ow physics, and (c) in order to present several more

references on turbulence modelling applications, thus completing the present review. LES and particularly DNS applications

716

Table 2

Previous studies in DNS of turbulent single-phase pipe ows.

Reference

Mesh

Reb

Geometry

Method

Nikitin [279]

Eggels et al. [280]

Zhang et al. [281]

Loulou [282]

Orlandi and Fatica [284]

Orlandi and Ebstein [285]

Schmidt et al. [286]

43 64 43

96 128 256

75 128 128

72 160 192

128 96 257

129 96 193

Cyl: 150 elements

Car: 64 elements

70 240 486

96 128 256

152 256 394

200 160 256

32 256 64

48 256 128

300 1024 2048

430 512 1024

256 1024 2048

18.67 106, 121.4 106, 4374 106, 2.184 109

20004000

5300

2500, 4000

5600

4900

4900

4910

Straight pipe

Straight pipe

Straight pipe

Straight pipe

Rotating pipe

Rotating pipe

Straight pipe

FDM + SM

FVM

SM

B-spline/spectra

FDM

FDM

SEM

10,300

5300

10,300

6000

4000

6000

44,000

24,500, 61,000

24,580

5300, 11,700, 19,000, 37,700

Straight pipe

Straight pipe

Straight pipe

Elliptical pipe

Elliptical pipe

Elliptical pipe

Straight pipe

Straight pipe

Straight pipe

Straight pipe

FVM

FDM

SM

FDM

FDM

FDM

FDM

SM + FDM

FDM

SEM

Fukagata and Kasagi [289]

Veenman [290]

Nikitin and Yakhot [291]

Voronova and Nikitin [293]

Voronova and Nikitin [294]

Wu and Moin [295]

Boersma [296]

Wu et al. [297]

Khoury et al. [298]a

FVM: Finite Volume Method.

SM: Spectral Method.

SEM: Spectral Element Method.

Car: for Cartesian coordinates.

Cyl: for Cylindrical coordinates.

Reb: Reynolds number based on bulk-mean velocity and pipe diameter of the pipe.

a

Number of grid points for each considered Reb, respectively.

lead to much more detailed results than those obtained by RANS, as they allow the computation of high order statistics and

they also reveal secondary ows that are very difcult for RANS model to reveal.

3.1. DNS of turbulent pipe ows

Turbulent pipe-ow numerical studies (Table 2) historically have not been as popular as channel ows. This is due to the

strong restrictions imposed on the time step choice and the very ne grids required to simulate pipe ows, in order to elaborate on the singularity at the centerline. Some available turbulence pipe experimental studies can be found in the review

papers by Marusic et al. [277] and Mullin [278].

An early attempt to perform DNS computations of three-dimensional turbulent ows in pipes was performed by Nikitin

[279]. In that work, a mixed nite differences and spectral method was used for the numerical simulation of the ow within

a range of Reynolds number from 2000 to 10,000. The numerical results, which describe the evolution of motion and the

characteristics of the ow, presented fair agreement with the available experimental data.

The same year, Eggels et al. [280] presented DNS of fully turbulent pipe ow at3 Reb = 5300 in a computational domain

with length of 10R. Statistical results on DNS were investigated in order to examine whether the axisymmetric geometry affects

the velocity uctuations of the ow or not. The numerical results were compared with the DNS data of channel ow by Kim

et al. [58] at the same Re number, with the aim to observe any possible differences between axisymmetric pipe and the plane

channel geometries. The DNS results were also validated with experimental data obtained by PIV and LDA measurements. It was

noticed that the statistics of velocity uctuations exhibit less effects for the pipe compared to plane channel geometry. However,

it was observed that the turbulence-intensity differences appear to be negligible between the pipe and channel ow. The skewness factor differs signicantly between the two geometries, caused probably by the impingement mechanism of the wall due to

the transverse curvature effect. The high order statistics data and the energy budget computations between the channel and

pipe geometries exhibit fair and excellent agreement with available experimental data, respectively.

DNS for fully turbulent pipe ows has also been investigated by Zhang et al. [281] and by Loulou [282]. The former used a

full spectra method and presented features of turbulent pipe ow at Re = 2500 and 4000, while the latter used a B-spline/

Fourier embedded method with the divergence-free Galerkin method of Leonard and Wray [283] at Reb = 5600.

Orlandi and Fatica [284] performed DNS using nite difference methods for investigating the effects of ow in a rotating

circular pipe. The rotation number, N, was in a different range compared to the work of Eggels [280], but was not high

enough to investigate the near-wall vortex structures. The numerical results in the non-rotating case were validated by

the data of Eggels [280]. They concluded that a strong interaction exists between numerical and sub-grid dissipation. However, a degree of drag reduction demonstrated by numerical and experimental data and the differences of turbulent statistics

were explained by the tilting of the near-wall -vortex -structures in the direction of rotation.

3

717

Fig. 10. Visualization of the turbulent pipe ow over the surface of 1 r = 0.1 using contours of instantaneous uz. Red represents higher values of uz.

(a) Reb = 5300. (b) Reb = 44 000 [295]. X. Wu, P. Moin, A direct numerical simulation study on the mean velocity characteristics in turbulent pipe ow,

Journal of Fluid Mechanics 608 (2008) 81112, reproduced with permission. (For interpretation to colours in this gure, the reader is referred to the web

version of this paper.)

Extension of the previous study was the work of Orlandi and Ebstein [285] with the increase of the rotation number, N, up

to 10 and the derivation of Reynolds stress budgets. No signicant changes compared to previous results were observed

[284].

Schmidt et al. [286] investigated a DNS of turbulent pipe ow at Reb = 4910, by employing a spectral element method, in

both cylindrical and Cartesian coordinates. The numerical results were validated by available experimental and numerical

data of Den Toonder and Nieuwstadt [287] and Eggels et al. [280], respectively. The mean velocity prole presented excellent

agreement against the experimental data by both methods, while urms values are almost identical for both cases, excluding

some small differences at the outer layer. Finally, the shear stress distribution did not exhibit almost any differences among

all data.

Wagner et al. [288] conducted DNS for fully turbulent pipe ow at Reb = 10,300, by using a nite volume technique, similar to Eggels et al. [280]. Turbulence quantities were investigated and showed signicant low Re effects, particularly for the

turbulence kinetic energy budget, the Reynolds stress tensor, vorticity and pressure uctuations and the two point correlations for velocity.

The same year, Fukagata and Kasagi [289] presented DNS results at Res = 180 by using a second-order nite difference

method. The length of computational domain was 10R and the adopted mesh was 96 128 256 with Dr+ = 0.46 (wall)

to Dr+ = 2.99 (center). Their numerical results compared with DNS data by Eggels et al. [280] and DNS data of channel ow

by Moser et al. [59], presenting good agreement.

Some years later, Veenman [290] investigated the DNS of turbulent pipe ow at Reb = 10,300 by using a pseudo-spectral

approach. The numerical results obtained focused on statistical parameters which are important for the development of

Lagrangian stochastic turbulence models. The proposed stochastic turbulence model was used to describe the dispersion

ow of a passive scalar from a point source in a pipe. The obtained dispersion statistics showed satisfactory agreement with

available DNS and experimental data.

Nikitin and Yakhot [291] presented DNS of turbulent ow in elliptical ducts. They implemented Immersed Boundary

Method (IBM) according to the research of Kim et al. [292] and estimated that the mean ow characteristics, the Reynolds

718

Fig. 11. Pseudo-colour visualisation of the instantaneous axial velocity uz normalised by the bulk velocity Ub. (a) Reb = 5,300; (b) Reb = 11,700;

(c) Reb = 19,000; (d) Reb = 37,700. Here, the colours vary from 0 (black) to 1.3 (white) [298]. G.K. Khoury, P. Schlatter, A. Noorani, P. Fisher, G. Brethouwer, A.

Johansson, Direct numerical simulation of turbulent pipe ow at moderate high Reynolds numbers, Flow, Turbulence and Combustion 91 (2013) 475495.

This is Fig. 4 in the publication in which the material was originally published. With kind permission from Springer Science and Business Media.

(For interpretation to colours in this gure, the reader is referred to the web version of this paper.)

stresses and turbulence intensities present similarities compared to data from channel ow. In addition, a reduction of turbulence effect near to the wall region of the major axis was noticed due to the transverse curvature effect. Moreover, the

presence of instantaneous velocity structures near the wall region was identied as streaks. However, these structures were

not present near to the major axis endpoints.

Extension of the previous study was the work of Voronova and Nikitin [293,294]. Both studies performed DNS in elliptical

ducts but for higher Reynolds number, 4000 and 6000, respectively.

Wu and Moin [295] performed DNS (Fig. 10) for fully turbulent pipe ow at Reb = 5300 and 44,000 based on bulk velocity,

by using a nite difference technique of second order on 630 106 grid nodes. A computational domain with length equal to

15R was used at both Re numbers, while the mesh at Reb = 5300 and 44,000 was 256 512 512 and 300 1024 2048

along the r, h and z directions, respectively. Their numerical results for mean ow statistics appear in good agreement with

available experimental data from Princeton Superpipe at Reb = 41,727 and 74,000, while for the second-order statistics good

agreement is observed at Reb = 74,000.

Recently, Boersma [296] presented DNS results of turbulent pipe ow at Reb = 24,500 and 61,000, by using a combination

of a pseudo-spectra method and a 6th order nite difference method. The pseudo-spectra method was used for the axial and

azimuthal directions while a 6th order nite-difference method was adopted for the radial direction. Turbulence statistics

were computed and autocorrelation functions and one-dimensional energy spectra were also presented. The numerical

results for the case of Reb = 24,500 appear in excellent agreement with experimental data by Den Toonder and Nieuwstadt

[287].

Continuation of the previous study of Wu and Moin [295] was the work of Wu et al. [297] by using the same numerical

method. They presented numerical results for DNS of pipe ow with a computational domain equal to 30R at Reb = 24,580. A

grid with resolution of 256 1024 2048 was used along the r, h and z directions, respectively. The numerical results highlighted the importance of large and very large scale motions. The very large scales indicate streamwise acceleration of the

ow in the vicinity of the wall, based on force spectra. The small scales seem to decelerate the mean streamwise prole

according to observations in force spectra of previous experimental studies. It was the rst time that net force spectra computations were performed in the buffer layer.

Khoury et al. [298] performed DNS for investigating incompressible ow in a smooth pipe of radius R and length 25 R, by

using a high-order spectral element method. Numerical results were obtained at Reb = 5300, 11,700, 19,000 and 37,700. In

719

Fig. 12. (a) Pseudo-colours of the instantaneous axial vorticity xz for Reb = 19,000 together with the spectral element boundaries. (b) Zoomed view of an

upper right part of (a) [298]. G.K. Khoury, P. Schlatter, A. Noorani, P. Fisher, G. Brethouwer, A. Johansson, Direct numerical simulation of turbulent pipe ow

at moderate high Reynolds numbers, Flow, Turbulence and Combustion 91 (2013) 475495. This is Fig. 5 in the publication in which the material was

originally published. With kind permission from Springer Science and Business Media. (For interpretation to colours in this gure, the reader is referred to

the web version of this paper.)

Fig. 11, the instantaneous axial velocity is presented, according to the increase of Re number and therefore the increase in the

range of the scales. Fig. 12 depicts the instantaneous axial vorticity at Reb = 19,000 with the numerical grid. Strong vortex

motion in the vicinity of the walls is observed as well as the conguration of small rotating vortices. The numerical results

were extensively compared with numerical data from pipe, channel and boundary layer turbulent ows. Turbulence statistics such as turbulence kinetic energy budgets and velocity uctuations were also presented and evaluated. It was concluded

that the variation of pressure and velocity uctuations depend on the Reynolds number for the examined ows. High degree

of similarity was exhibited for all the considered types of ows and important differences were observed in the wake region.

Turbulence kinetic energy budgets present independence, in the inner region, from the type of ow up to y+

100, along

with great differences for the wake region.

Some important DNS work involving turbulent ows with free surfaces, with and without shear is listed here.

Lam and Banerjee [299] conducted DNS with the use of a FourierChebyshev pseudo-spectral method for the examination

of the inuence of shear, and of boundary conditions, on the turbulent structures near to the wall and the free slip surface.

The aim of their research was focused on the determination of the critical parameter which was responsible for the conguration of turbulent streaky structures. However, they did not examine the relationship between turbulence structure and

scalar transport mechanism.

Komori et al. [300] performed DNS calculation for the clarication of turbulence structure at the interface of an open

channel ow. In contrast to the study of Lam and Banerjee [299] the proposed methodology describes the mechanism of scalar transfer for a gasliquid interface with zero-shear. The numerical simulations were conducted with the use of nite difference method and a Boundary Fit Coordinate (BFC) system without any approximations. The numerical results for the

bursting frequency and mass transfer coefcient were compared with turbulence statistics by Laser Doppler Velocimetry

(LDV) and authors previous studies, presenting good agreement. They also noticed that in the vicinity of the wall there were

large eddies. These large eddies are lifted up towards the interface position and renew it. Furthermore, they promote the

mass transfer across the gasliquid interface.

Lombardi et al. [301] investigated via DNS the coupling between gas and liquid phase, with keeping at the interfacial

region (Fig. 13) of the ow. The numerical results focused on the turbulence statistics and the clarication of the mechanics

of coupling ows between the phases. It was concluded that the turbulence structure on the gas side has great similarities

with that at the wall. They also observed that in some applications of two-phase ows the gas side might perceive the liquid

phase as a solid surface.

De Angelis [302] presented a continuation of the research of Lombardi et al. [301] with the aim at extending his work to a

non-at interfacial region, while the stratied ow was accounted for with a freely deformable interface.

De Angelis et al. [303] also examined stratied ows with the use of wind stresses on the gasliquid interfacial region,

while the wavy interface was transformed to a simple geometry in order to facilitate a simple computation. They concluded

that the scalar exchange rates depended in a similar way on the activities on the gas side, with high exchange rates occurring

with sweeps, whereas low exchange rates with ejections [304].

Fulgosi et al. [305] performed DNS to investigate the turbulence near a freely deformable interface (Fig. 14). The research

was focused on the interfacial sub-layer and particularly on the gas side, with the aim at identifying how the wavy-induced

mechanisms affect the ow characteristics. Numerical results on the gas side were compared with channel ow data at the

720

Fig. 13. Sketch of the physical problem [301]. Reprinted with permission from P. Lombardi, V.D. Angelis, S. Banerjee, Direct numerical simulation of nearinterface turbulence in coupled gas-liquid ow, Physics of Fluids 8 (1996) 16431665. Copyright [1996], AIP Publishing LLC.

Fig. 14. Sketch of the simulated problem. The elevation of the waves has been amplied by a factor 5 [305]. M. Fulgosi, D. Lakehal, S. Banerjee, V.D. Angelis,

Direct numerical simulation of turbulence in a sheared air-water ow with a deformable interface, Journal of Fluid Mechanics 482 (2003) 319345,

reproduced with permission.

same shear Reynolds number. However, the numerical method was not the most appropriate for ows with large deformable

interface, due to the possible entrainment of the liquid phase and the formation of large waves. It was observed that the

numerical results do not exhibit great differences compared to the previous study of Lombardi et al. [301].

Banerjee et al. [306] presented DNS for the clarication of the transfer mechanisms at deformable interfaces. They

increased the friction velocities with the purpose to present surface deformations of high waveslops, up to the point which

they do not lead to wave breaking. In the cases of high shear rate because of gas, the turbulence is presented in the vicinity of

the interface, as being similar to that at solid boundaries. Fig. 15 illustrates contours of instantaneous interfacial heat transfer

coefcient for wavy and at interface. The top cases are for the gas phase and the other two at the bottom for the liquid

phase. It was concluded that the surface divergence model is suitable for the prediction of gas transfer for non-sheared interface, but also for wind-shear cases.

Lin et al. [307] developed an airwater coupled method in order to simulate the interaction of two fully developed

turbulent layers (air and water), above and below the interface. In that study the wind-wave generation processes were

investigated by performing DNS and coupling of two turbulent ows (air and water) across a deformable interface. Limitation of the model is the linearization of the boundary condition at the interface, so that the model is suitable only for

small-amplitude waves. It was observed that in the initial stage of the simulated waves there were great similarities of their

characteristics with eld and experimental data.

721

Fig. 15. Instantaneous patterns of the interfacial heat transfer coefcient at the at and wavy interface. (upper panel) gas side, (lower panel) liquid side

[306].

Numerical simulations of turbulent condensing vapourliquid ow were conducted by Lakehal et al. [308]. A DNS

approach was selected for the investigation of turbulence characteristics of the two-phase ow, accounting for condensation

effects. The numerical results showed that the effects of condensation are important on the turbulence characteristics of the

ow for both phases (liquid and vapour). More specically, the turbulence statistics and Reynolds stresses increase at the

vapour side, while they decrease at the liquid side. Fig. 16 depicts the coherent structures with and without condensation

for both phases of the ow. The results for the obtained velocity proles were compared with DNS and experimental data.

It was also observed that condensation affects the conguration of interfacial waves by weakening them.

Trontin et al. [309] performed DNS of two-phase interfacial ows. The main goals of their study were to investigate the

interaction between small turbulence scales and the deformable interface. The pioneering part of the research was the presentation for the rst time of DNS results for the interface/turbulence interactions, in case of a widely deformed interface. A

energy budget and turbulence statistics were also exhibited.

LES of three-dimensional turbulent pipe ow studies (Table 3) are not as numerous as for channel ows due to the formers centreline singularity.

An early attempt to perform LES of turbulent pipe ow was undertaken by Unger and Friedrich [310] for complex geometries at Reb = 50,000, by using second-order nite volume method. The classical Smagorinsky model was used on a mesh

size of 96 128 256. Numerical results for instantaneous turbulence quantities were presented.

Eggels and Nieuwstadt [311] implemented a Smagorinsky SGS model for a LES approach and showed that their numerical

results were in good agreement with experimental data.

Boersma and Nieuwstadt [312] investigated the effects of turbulent ow in a curved pipe, by means of LES technique and

the classical Smagorinsky SGS model at Reb = 20,000. Numerical results for turbulence statistics, mean velocity prole and

secondary motion of the ow were presented and discussed; they were also compared with available experimental data,

presenting satisfactory agreement.

722

Fig. 16. Three-dimensional distribution of vortical structures in the presence of interfacial mass exchange, made visible by the use of isosurfaces of

k2 = 0.03. [(a) and (b)] Case R2. [(c) and (d)] Case C21. The amplitude of the waves has been magnied ten times in the vertical direction [308]. Reprinted

with permission from D. Lakehal, M. Fulgosi, S. Banerjee, G. Yadigaroglu, Turbulence and heat exchange in condensing vapor-liquid ow, Physics of Fluid 20

(2008) 06510118. Copyright [2008], AIP Publishing LLC.

Table 3

Previous studies in LES of turbulent single-phase pipe ows.

Reference

Mesh

Reb

Geometry

SGS model

Method

Eggels and Nieuwstadt [177]

Boersma and Nieuwstadt [190]

Yang [191]

96 128 256

96 128 256

40 114 200

192 64 128

50,000

59,500

20,000

20,000

Straight pipe

Rotating pipe

Curved pipe

Rotating pipe

FVM

FVM

FVM

FVM

Schmidt et al. [154]

36,700

16,000

Straight pipe

Straight pipe

Cyl: 80 elements

Car: 105 elements

65 39 65

Smagorinsky

Smagorinsky

Smagorinsky

Smagorinsky

Dynamic

Smagorinsky

Smagorinsky

4900, 7400

Rotating pipe

FDM

Jordan [194]

Vijiapuraru and Cui [195]

64 141 401

64 96 64

8000

5000, 30,000

Straight pipe

Jung and Chung [197]

96 128 128

128 256 256

100,000

700036,000

Straight pipe

Smagorinsky

Dynamic

Dynamic

Smagorinsky

Dynamic

Dynamic

Dynamic

FVM: Finite Volume Method.

SEM: Spectral Element Method.

Car: for Cartesian coordinates.

Cyl: for Cylindrical coordinates.

Reb: Reynolds number based on bulk-mean velocity and pipe diameter of the pipe.

p/k: ratio of rib periodicity where p is the distance between two successive ribs and k is the rib height.

SEM

SEM

FDM

FDM

FVM

FDM

723

Table 4

List of the described numerical approaches for simulating turbulent ows in the present review paper, excluding the cases for pipe ows (DNS and LES) which

are presented in Tables 2 and 3, respectively.

Approaches-models

DNS

Lilly [55], Orszag and Patterson [56], Rogallo [57], Kim et al. [58], Moser

et al. [59], Spalart [64], Schlatter et al. [65], Kaneda et al. [72], Kaneda

and Ishihara [73], Abe et al. [60], Del Alamo et al. [61], Hoyas and

Jimenez [62], Lee et al. [71], Schlatter and Orlu [67], Sillero et al. [68]

RANS

Zero-equation models

Prandtls mixing length [88], Cebeci-Smith [89], Baldwin-Lomax [90]

Half-equation models

Johnson and King [9295]

One-equation models

Baldwin and Barth [96,99,100], Spalart and Allmaras [97,98,101,102],

Fares and Schroder [105]

Two-equation models (including LRN)**

Standard ke [107], realisable ke [109], RNG ke [110], LRN kx

[108,131], Wilcox (2006) kx [86], SST [119121], Lam-Bremhorst

ke [129], Bredberg et al. kx [130], Pang et al. kx [132]

NLEVM

Quadratic models [134,135], Cubic models [137,138]

Advanced EVM

Durbins u2f model [139,142], ff model [140], BL- u2/k [143,145]

DSM

Models for the diffusive term (Dij): [146148]

The redistribution term (Uij) is usually modelled by IP model [150].

Improvements are: LRR-QI model [150], SSG model [152], CL model

[153] and LT model [154]. EBSDM model [144], Durbins DSM [159]

and HTM model [162]

Type of ow

Two-dimensional turbulence [55], homogeneous isotropic turbulence

[56,57,72,73], channel [5862,71], boundary layer [6369]

Pressure driven separated ows [92], transonic separated ows [93],

compressible [94]

Airfoil [99102], wake, jet, boundary layer [105]

dispersion [111114], street canyon [115], rotating cylinder [116],

turbomachinery blades, wind turbines, free shear layers [119121]

For separated ows in adverse pressure gradient, turbine blade, ow

plate, jets [134138]

Rotating cylinder, rotating channel ows, axial rotating pipe and square

duct [141]

Street canyon ows [115], impingent jet ows [144,162]

Flow around a car, axially rotating pipe, 180 turned U bend, backwardfacing step [162]

ASM

Pope [163], Gatski and Speziale [134], Jongen and Gatski [164] and Wallin

and Johansson [165]

2-D ows [163], 3-D ows [134,164], channel ows, boundary layer,

wings and impinging shock [165]

Two-uid models

Spaldings model [166], Malin [167], Markatos and Kotsifakis [168], Shen

et al. [169], Yu et al. [170], Youngs model [171],

2SFK model [172], Lin et al. [173] and Cao et al. [174]

Free shear ows [167], combustion [168], stratied ows [169], ow and

heat transfer of air curtains [170,174], RayleighTaylor mixing [171,172]

and UV disinfection reactor ow [173]

LES

Smagorinsky [178], Lilly [179,192], Deardorff [180], Schumann [183],

Kraichnan [181], Chasnov [182], Moin and Kim [184], Piomelli [185],

Bardina [195], Zang et al. [198], Vreman et al. [199]. Germano [191],

Ghosal et al. [193], Germano [191], Meneveau et al. [194], Stolz and

Adams [201], Stolz et al. [202], Leray-a [211,217,216], Clark-a [210],

Metais and Lesieur [206], Ducros et al. [207], Domaradzki and Saika

[208], Domaradzki and Loh [209]

MILES-ILES

Boris and Book [228], Boris [235,236], Fureby [233], Patnaik et al. [234],

Colella and Woodward [230],

URANS (including 2G-URANS)

Johansson et al. [237], Person and Davidson [238,239], Speziale [240],

Kenjeres and Hanjalic [242], Hanjalic and Kenjeres [160], SAS model

[243,246], PANS model [244,247250], PITM [251,252]

VLES

Speziale [240], Johansen [253], PRNS [175,254,255], LNS [256], Ruprecht

et al. [257], Parot and Gadebusch [258,259], Hsieh et al. [260], Labois

and Lakehal [261], Han and Krajnovic [262]

185,201,204,205,212214,217], wall-bounded ows [203], compressible

ows [202], rotating cylinder [190]

Channel ow, ow over a cylinder [233], supersonic jet ows [35], largescale urban ow [234]

Vortex shedding behind bluff bodies [237], surface mounted cube [238],

ow around car [239], environmental turbulent ows and pollutant

dispersion [160]

Homogeneous turbulence, channel ow and unsteady ow with periodic

forcing [251,252]

Turbulent pipe ow [175,254], non-reacting ow in a single injector

ame tube [254], internal reacting and external static stall ows [255]

724

Table 4 (continued)

Approaches-models

DES

Spalart et al. [267], Spalart [264], IDDES [266], DDES [266], SST-DES [288],

Gritskevich [269]

Hybrid LES/RANS

Leschziner et al. [270], Davidson [271], Hambla [2003], Temmerman et al.

[2005], Tucker and Davidson [274], Davidson and Dahlstrom [275],

Xiao et al. [276]

Type of ow

Sharp-edged delta wing, bluff bodies, ground vehicles, active ow control

by suction/blowing, vibrating cylinders with strakes, cavitation in jets,

building, air inlets, aircraft in spin, high-lift devices [264]

Channel ow [271275], separated ow [273], plane asymmetric diffuser

ow [275]

Yang [313] performed LES for the investigation of swirl effects driven by a rotational wall at Reb = 20,000. A nite volume

method was adopted with a computational domain of 4D for the numerical simulations. The performance of two sub-grid

scale models, namely dynamic and classical Smagorinsky, has been investigated. The numerical results appeared to be in fair

agreement with measurements and both sub-grid scale models performed equally well.

Rudman and Blackburn [314] and Schmidt et al. [286] presented results from an LES of turbulent pipe ow with bulk ow

Reb = 36,700, with the classical Smagorinsky model and van Driest type wall damping. The numerical results were compared

with experimental measurements, exhibiting fair agreement.

Feiz et al. [315] performed LES of a rotating pipe at moderate Reb numbers 4900 and 7400. The numerical results were compared with simulations obtained by DNS data [284]. Two different SGS models (Smagorinsky and dynamic) were examined for

the prediction of turbulent ow with and without rotation. The best performance was exhibited by the dynamic model.

The same year, Jordan et al. [316] conducted LES for the investigation of turbulent ow inside a cylindrical ribbed duct

with ratio of rib periodicity equal to p/h = 5 and Reb = 8000. They employed a nite difference technique for the discretization

of NavierStokes equations and adopted a dynamic model for the description of SGS stress eld. Turbulence statistics such as

streamwise intensity, radial intensity and Reynolds stress were computed and compared with DNS data of Eggels et al. [280]

and at-sea measurements (Reb = 4 106). The numerical results and at-sea measurements presented essentially equivalent pressure core loss at the center of the circular duct, which proves that the turbulence physics for both Reynolds numbers are scale-similar.

Vijiapuraru and Cui [317] investigated the fully developed turbulent ow by means of LES for two different Reynolds

numbers, Reb = 5000 and 30,000. A nite difference method was employed for the discretization of NavierStokes equations

in cylindrical coordinates, while the classical Smagorinsky and dynamic model were selected for the subgrid scale modelling.

Mean velocity calculations and computations of high order turbulence statistics were presented and validated with available

numerical and experimental data. It was concluded that the proposed methodology is suitable for handling turbulent pipe

ows at low and moderate Re numbers.

Vijiapuraru and Cui [318] examined also turbulent ows in circular ribbed pipes by using LES, and the nite volume

method. The numerical results obtained for Reb = 100,000 were compared with experimental data and presented fair agreement for the considered p/k = 2, 5 and 10. It was concluded that LES is suitable for predicting the turbulence statistics but

that it needs at least one order more CPU time compared to the classical RANS models.

Jung and Chung [319] undertook numerical simulations for the investigation of accelerated turbulent ow in cylindrical

pipe by means of LES. A second order central difference scheme was employed for the discretization of incompressible

NavierStokes equations and the numerical simulations were performed for linearly increasing value of Reb from 7000 to

36,000 due to the acceleration. Statistics for the skin friction coefcient, mean velocity, velocity uctuations and quadrant

analysis were presented and discussed. That LES study was the rst attempt to treat temporal acceleration in a fully-developed turbulent pipe ow and it was concluded that the turbulence anisotropy increased during the temporal acceleration.

3.4. LES of turbulent free-surface ows

To the authors best knowledge two-phase pipe LES studies are not available except for the recent work of Lakehal [320].

Lakehal [320] performed Large Eddy Interface Simulation (LEIS) for modelling the slug formation of a two-phase pipe ow. A

BFC grid was adopted for the computational domain and a level set method for the interfacial region. The numerical results

were compared with analytical and experimental data, predicting the slug speed (tail and centre) with reasonable accuracy.

For turbulent ows with free surfaces some important work involving LES with and without shear has been performed, as

follows.

Reboux et al. [321] used an LES approach for a turbulent interfacial two-phase ow. The effects of a modied Smagorinsky

model with and without shear interface treatment and the Variational Multiscale approach were examined. The numerical

results conrmed that both proposed methods were suitable for handling the anisotropy of turbulence in the liquid side

underneath the interfacial region. They also predict the formation of boundary layer stratication at the gas side. In addition,

turbulence statistics were calculated by both mentioned approaches, with better results for the Variational Multiscale

approach compared to the modied Smagorinsky model.

725

Christensen [322] performed LES for the study of wave breaking and the interaction of the turbulence interface. A VOF

method was adopted for the simulation of the interface. Periodic spilling and plunging breakers were simulated by means

of LES along with the classical Smagorinsky model. The undertow, set-up and turbulence levels, obtained were compared

with experimental data. It was concluded that although the grid was relatively coarse, satisfactory results were obtained

for the wave height decay and undertow, while the turbulence levels were overestimated. The performance of the numerical

model was considered satisfactory.

The same year, Lubin et al. [323] conducted two-phase LES for the numerical simulation of plunging breaking waves in

airwater congurations, by using a dynamic (mixed scale) model by Sagaut [26] for the sub-grid scale modelling. In addition, they investigated the air entrainment mechanism that occurs during the wave breaking process. The numerical model

was validated with available experimental data and analytical solutions. In general, the proposed numerical model appeared

to be a reliable tool for the description of two-phase ows and gave very satisfactory results.

Lakehal and Liovic [324] undertook numerical simulations for the investigation of turbulence scales and their interaction

with wave breaking by means of Large-Eddy Interface Simulation (LEIS). LEIS is a combination of LES and an interface tracking method (e.g. VOF). The ltered single-uid equations along with an ILES framework were employed to the MFVOF-3D

ow software. SGS modelling was treated by using the proposed SGS model of Liovic and Lakehal [325]. The numerical model

was validated with experimental (phase-averaged) data by Ting and Kirby [326] which are suitable for the validation of

breaking wave simulations. It was concluded that the numerical simulations were adequate in providing detailed information for the airliquid turbulent coherent structures, and their connexion with the local incidence transient mechanisms. For

the rst time a conditional zonal analysis was undertaken for the investigation of transient mechanisms for turbulence statistics (turbulence kinetic energy, diffusion, decay and transport) and their effect and dependencies on the wave breaking

process.

4. Conclusions

This paper reviews the problem and successes of computing turbulent ows. The review is primarily concerned with the

most recent methods for such computer predictions. The successes and problems are demonstrated by listing and briey

discussing several applications of DNS and LES to ows in pipes and free-surface ows.

In Table 4, a full list with the appropriate references of all the above mentioned turbulence strategies and models are

given, so that the potential user may easily nd the appropriate information, for him to select the suitable turbulence model

for his own case of interest. The potential user is also recommended to examine the following review papers and guidelines

concerning verication and validation in CFD work, by Roache [327], Oberkampf and Trucano [328], Roy [329], Stern et al.

[330], ASME [331], NEA [332], ERCOFTAC [333], AIAA [334], in order to be able to assess the accuracy and reliability of his

numerical result.

The LES approach appears, from the given references that describe its applications, to have reached maturity. However,

further work is required to improve the characteristics of the method for more types of turbulent ow, in particular for complex industrial ows. There are still challenges facing LES of turbulence such as the development of advanced sub-grid scale

models, high-order discretization techniques for eliminating the numerical errors, implementation on unstructured grids,

control of the numerical errors, interaction with other physical mechanisms and a simple wall stress model for wallbounded complex ows. In this context, the subject of discretization techniques for convection is very important. Thus,

highly dissipative schemes must be avoided, as they smooth out several frequencies and therefore one may not attribute

errors correctly, either to the turbulence model or to the numerical scheme. On the other hand monotonic increasing

schemes appear benecial (see MILES). This topic is also vast and, in several aspects, misunderstood, and is not covered

in the present work. The interested reader may nd useful information in [335339].

ILES techniques were developed recently and have gained a lot of application space due to the drawbacks of LES. Many

researchers replace the classical LES formulations, by introducing implicit SGS modelling with the use of nonlinear algorithms (e.g. MILES). However, some challenges for MILES and in general for ILES, are the appropriate physical and mathematical framework for their analysis and development, and the interaction of the numerical schemes among the implicit SGS

models.

Hybrid LES/RANS methods are a good alternative to LES, because they combine the accuracy of LES and the speed of RANS.

These methods are suitable for ows dominated by large coherent structures and strong unsteady proles (e.g. bluff body

ows, IC engines, among others). Hybrid LES/RANS methods along with the aforementioned PANS, PITM provide higher accuracy compared to the DES, SAS and URANS approaches, but they are also more demanding in CPU time.

The continuing progress of computer-hardware development is promising for DNS, in the near future, with major

improvements expected in statistical samples and in considering a variety of several physical parameters, for better understanding of the turbulence nature. Some nal comments are:

Some success has been achieved with two-equation models for relatively simple hydrodynamic phenomena; indeed, routine design work can now be undertaken in several applications of engineering practise, for which extensive studies have

optimised these models.

726

Failures are still common for many applications, particularly those that involve strong curvature, intermittency, strong

buoyancy inuences, low-Reynolds-number effects, rapid compression or expansion, strong swirl, and kinetically-inuenced chemical reaction. New conceptual developments are needed in these areas, probably along the lines of actually

calculating the principal manifestation of turbulence, e.g. intermittency. A start has been made in this direction in the

form of multi-uid models, and full simulations.

Although some of the latest concepts hold promise of describing some of the most important physical consequences of

turbulence, they have not yet reached a denite stage of development.

From this point of view, the older and simpler methods can still be recommended as the starting point (and sometimes

the nishing point) for engineering simulations.

LES is currently the most accurate method available for practical computations and its use is expected to rise fast over the

next few years.

DNS is obviously the method that provides the most precise and detailed description of turbulence but it is still out of

reach of the available everyday computer power, i.e. it cannot be used for everyday engineering design. It is, however,

even today very useful as it serves as a test of the other model predictions and of any new ideas on turbulence calculations. Until it becomes also a practical tool the authors recommend the use of two-uid models that appear very

promising (but need further renement), and LES along with its derivatives and the hybrid methods that have already

reached maturity.

Acknowledgements

The authors wish to thank Dr Djamel Lakehal, for kindly providing us with constructive remarks and recommendations on

an initial draft of this paper. Furthermore, special thanks are due to Dr. Stavros Karabelas and PhD Student Nektarios Koutsourakis for offering gures and useful material from their studies. The authors are indebted to the anonymous referees, for

extensively reviewing this paper and their useful comments. Finally, the author (C.A) wishes to express his sincere gratitude

to Prof. Omar Matar for the extension of his bursary and his encouragement during this study.

This work has been undertaken within the Joint Project on Transient Multiphase Flows and Flow Assurance. The Author

(C.A) wishes to acknowledge the contributions made to this project by the UK Engineering and Physical Sciences Research

Council (EPSRC) and the following: ASCOMP, GL Noble Denton, BP Exploration, CD-adapco, Chevron, ConocoPhillips, ENI,

ExxonMobil, FEESA, FMC Technologies, IFP Energies nouvelles, Granherne, Institutt for Energiteknikk, Kongsberg Oil & Gas

Technologies, MSi Kenny, PDVSA (INTEVEP), Petrobras, PETRONAS, SPT Group, Shell, SINTEF, Statoil and TOTAL.

References

[1] N.C. Markatos, The mathematical modelling of turbulent ows, Appl. Math. Modell. 10 (1986) 190220.

[2] M. Hinze, T. Rung, Hydrodynamic shape-optimisation for turbulent industrial ows, Universitat Hamburg, Project C2, Poster presented at the

workshop Recent Trends and Future Developments in Computational Science and Engineering, Plon, 2014.

[3] H. Tennekes, J.L. Lumley, A First Course in Turbulence, The MIT Press, Cambridge, Massachusetts, 1972.

[4] J.C.R. Hunt, J.C. Vassilicos, Turbulence Structure and Vortex Dynamics, Cambridge University Press, 2000.

[5] C. Vassilicos, Intermittency in Turbulent Flows, Cambridge University Press, 2001.

[6] C.G. Speziale, Analytical methods for the development of Reynolds-stress closures in turbulence, Ann. Rev. Fluid Mech. 23 (1991) 107157.

[7] H.K. Versteeg, W. Malalasekera, An Introduction to Computational Fluid Dynamics. The Finite Volume Method, second ed., Pearson Education Limited,

2007.

[8] W.C. Reynolds, T. Cebeci, Calculation of turbulent ows, in: P. Bradshaw (Ed.), Ch. 5 in Turbulence. Topics in Applied Physics, second ed., vol. 12,

Springer Verlag, Berlin, 1978, pp. 193229.

[9] B.E. Launder, Heat and mass transport, in: P. Bradshaw (Ed.), Ch. 6 in Turbulence. Topics in Applied Physics, second ed., vol. 12, Springer Verlag, Berlin,

1978, pp. 231287.

[10] P. Bradshaw, Introduction to turbulence, in: P. Bradshaw (Ed.), Ch. 1 in Turbulence. Topics in Applied Physics, second ed., vol. 12, Springer Verlag,

Berlin, 1978, pp. 144.

[11] K.R. Sreenivasan, R.A. Antonia, The phenomenology of small-scale turbulence, Annu. Rev. Fluid Mech. 29 (1997) 435472.

[12] P.S. Klebanoff, NACA Report 1247, National Bureau of Standards, Washington D.C., 1995.

[13] F.M. White, Viscous Fluid Flow, second ed., McGraw Hill, New York, 1991.

[14] J. Paret, P. Tabeling, Intermittency in the two-dimensional inverse cascade of energy: experimental observations, Phys. Fluids 10 (1998) 31263136.

[15] M. Kholmyansky, A. Tsinober, On the origins of intermittency in real turbulent ows, in: C. Vassilicos (Ed.), Intermittency in Turbulent Flows,

Cambridge University Press, 2001, pp. 183192.

[16] S. Tavoularis, U. Karnik, Further experiments on the evaluation of turbulent stresses and scales in uniformly sheared turbulence, J. Fluid Mech. 204

(1989) 457478.

[17] R.S. Rogallo, P. Moin, Numerical simulation of turbulent ows, Annu. Rev. Fluid Mech. 16 (1984) 99137.

[18] Z.-S. She, E. Jackson, S.A. Orszag, Structure and dynamics of homogeneous turbulence. models and simulations, Proc. R. Soc. London Ser. A Math. Phys.

Sci. 434 (1991) 101124.

[19] L. Kleiser, T.A. Zang, Numerical simulation of transition in wall-bounded shear ows, Annu. Rev. Fluid Mech. 23 (1991) 495537.

[20] N. Kasagi, N. Shikazono, Contribution of direct numerical simulation to understanding and modelling turbulent transport, Proc. R. Soc. London Ser. A

Math. Phys. Sci. 451 (1995) 257292.

[21] P. Moin, K. Mahesh, Direct numerical simulation: a tool in turbulence research, Annu. Rev. Fluid Mech. 30 (1998) 539578.

[22] R. Friedrich, T.J. Huttl, M. Manhart, C. Wagner, Direct numerical simulation of incompressible turbulent ows, Comput. Fluids 30 (2001) 555579.

[23] T. Ishihara, T. Gotoh, Y. Kaneda, Study of high-Reynolds number isotropic turbulence by direct numerical simulation, Annu. Rev. Fluid Mech. 41 (2009)

165180.

[24] S.B. Pope, Turbulent Flows, Cambridge University Press, 2000.

[25] L.C. Berselli, T. Iliescu, W.J. Layton, Mathematics of Large Eddy Simulation of Turbulent Flows, Springer, Berlin, New York, 2006.

727

[26] P. Sagaut, Large Eddy Simulation for Incompressible Flows: An Introduction, third ed., Springer, Berlin (New York), 2006.

[27] F.F. Grinstein, L.G. Margolin, W.J. Rider, Implicit Large Eddy Simulation: Computing Turbulent Fluid Dynamics, Cambridge University Press,

Cambridge, UK, 2007.

[28] J. Meyers, B.J. Geurts, P. Sagaut, Quality and Reliability of Large-Eddy Simulations. ERCOFTAC Series, Springer, Berlin, New York, 2008.

[29] E. Garnier, N. Adams, P. Sagaut, Large Eddy Simulation for Compressible Flows, Springer, New York, 2009.

[30] X. Jiang, C.-H. Lai, Numerical Techniques for Direct and Large Eddy Simulations, Chapman & Hall/CRC, Boca Raton, 2009.

[31] M. Lesieur, O. Metais, New trends in large-eddy simulations of turbulence, Annu. Rev. Fluid Mech. 28 (1996) 4582.

[32] U. Piomelli, Large-eddy simulation: achievements and challenges, Prog. Aeros. Sc. 35 (1999) 335362.

[33] C. Meneveau, J. Katz, Scale-invariance and turbulence models for large-eddy simulation, Annu. Rev. Fluid Mech. 32 (2000) 132.

[34] S.B. Pope, Ten questions concerning the large-eddy simulation of turbulent ows, New J. Phys. 6 (2004) 35.

[35] C. Fureby, Towards the use of large eddy simulation in engineering, Prog. Aerosp. Sci. 44 (2008) 381396.

[36] J. Frohlich, D. von Terzi, Hybrid LES/RANS methods for the simulation of turbulent ows, Prog. Aerosp. Sci. 44 (2008) 349377.

[37] R. Bouffanais, Advances and challenges of applied large-eddy simulation, Comput. Fluids 39 (2010) 735738.

[38] N.J. Georgiadis, D.P. Rizzetta, C. Fureby, Large-eddy simulation: current capabilities, recommended practices, and future research, AIAA J. 48 (2010)

17721784.

[39] P.G. Tucker, S. Lardeau, Applied large eddy simulation, Philos. Trans. R. Soc. A Math. Phys. Eng. Sci. 367 (2009) 28092818.

[40] T.B. Gatski, C.L. Rumsey, Linear and nonlinear eddy viscosity models, in: B.E. Launder, N.D. Sandham (Eds.), Closure Strategies for Turbulent and

Transitional Flows, Cambridge University Press, Cambridge, UK, 2002, pp. 946.

[41] ERCOFTAC, <www.ercoftac.org/special_interest_groups/15_turbulence_modelling/>.

[42] G.F. Hewitt, J.M. Delhaye, N. Zuber, Multiphase Science and Technology, vol. 5, CRC Press, 1989.

[43] M. Ishii, H. Takashi, Thermo-Fluid Dynamics of Two-Phase Flow, second ed., Springer-Verlag, Berlin, 2010.

[44] D.A. Drew, S.L. Passman, Theory of multicomponent FLUIDS, Applied Mathematical Sciences, vol. 135, Springer-Verlag, New York, 1998.

[45] P. Jenny, S.H. Lee, H.A. Tchelepi, Adaptive fully implicit multi-scale nite-volume method for multi-phase ow and transport in heterogeneous porous

media, J. Comput. Phys. 217 (2006) 627641.

[46] D.B. Spalding, A general purpose computer program for multi-dimensional one- and two-phase ow, Math. Comput. Simul. 23 (1981) 267276.

[47] S. Whitaker, The transport equations for multi-phase systems, Chem. Eng. Sci. 28 (1973) 139147.

[48] B. Lecordier, D. Deware, L.M.J. Vervisch, Estimation of the accuracy of PIV treatments for turbulent ow studies by direct numerical simulations of

multi-phase ow, Meas. Sci. Technol. 12 (2001) 13821391.

[49] T. Berning, M. Odgaard, S.K. Kaer, A study of multi-phase ow through the cathode side of an interdigitated ow eld using a multi-uid model, J.

Power Sources 195 (2010) 48424852.

[50] D.A. Drew, Mathematical modeling of two-phase ow, Ann. Rev. Fluid Mech. 15 (1983) 261291.

[51] A.E. Bergles, J.G. Collier, J.M. Delhaye, G.F. Hewitt, F. Mayinger, Two-phase Flow and Heat Transfer in the Power and Process Industries, Hemisphere

Publishing Corp., New York, 1981.

[52] D. Jamet, O. Lebaigue, N. Coutris, J.M. Delhaye, The second gradient method for the direct numerical simulation of liquidvapor ows with phase

change, J. Comput. Phys. 169 (2001) 624651.

[53] A. Prosperetti, Gretar Tryggvason, Computational Methods for Multiphase Flow, Cambridge University Press, 2007.

[54] N.C. Markatos, Modelling of two-phase transient ow and combustion of granular propellants, Int. J. Multiphase Flow 12 (1986) 913933.

[55] D.K. Lilly, Numerical simulation of developing and decaying two-dimensional turbulence, J. Fluid Mech. 45 (1971) 395415.

[56] S.A. Orszag, G.S. Patterson, Numerical simulation of three-dimensional homogeneous isotropic turbulence, Phys. Rev. Lett. 28 (1972) 7679.

[57] R.S. Rogallo, Numerical experiments in homogeneous turbulence, Technical report, NASA TM 81315, 1981.

[58] J. Kim, P. Moin, R. Moser, Turbulence statistics in fully developed channel ow at low Reynolds number, J. Fluid Mech. 177 (1987) 133166.

[59] R.D. Moser, J. Kim, N.N. Mansour, Direct numerical simulation of turbulent channel ow up to Res = 590, Phys. Fluids 11 (1999) 943945.

[60] H. Abe, H. Kawamura, Y. Matsuo, Direct numerical simulation of a fully developed turbulent channel ow with respect to the Reynolds umber

dependence, J. Fluids Eng. 123 (2001) 382393.

[61] J.C. Del Alamo, J. Jimenez, P. Zandonade, R.D. Moser, Scaling of the energy spectra of turbulent channels, J. Fluid Mech. 500 (2004) 135144.

[62] S. Hoyas, J. Jimenez, Scaling of the velocity uctuations in turbulent channels up to Res = 2003, Phys. Fluids 18 (2006) 011702.

[63] J. Komminaho, M. Skote, Reynolds stress budgets in couette and boundary layer ows, Flow Turbul. Combust. 68 (2002) 167192.

[64] P.R. Spalart, Direct simulation of a turbulent boundary layer up to Reh = 1410, J. Fluid Mech. 187 (1988) 6198.

[65] P. Schlatter, R. Orlu, Q. Li, G. Brethouwer, J.H.M. Fransson, A.V. Johansson, P.H. Alfredsson, D.S. Henningson, Turbulent boundary layers up to

Reh = 2500 studied through simulation and experiment, Phys. Fluids 21 (2009). 0517024.

[66] A. Ferrante, S. Elghobashi, Reynolds number effect on drag reduction in a microbubble-laden spatially developing turbulent boundary layer, J. Fluid

Mech. 543 (2005) 93106.

[67] P. Schlatter, R. Orlu, Assessment of direct numerical simulation data of turbulent boundary layers, J. Fluid Mech. 659 (2010) 116126.

[68] J. Sillero, J. Jimenez, R.D. Moser, N.P. Malaya, Direct simulation of a zero-pressure-gradient turbulent boundary layer up to Reh = 6650, J. Phys: Conf.

Ser. 318 (2011) 022023.

[69] J.A. Sillero, J. Jimenez, R. Moser, One-point statistics for turbulent wall-bounded ows at Reynolds number up to + 2000, Phys. Fluids 25 (2013)

105102.

[70] G. Alfonsi, On direct numerical simulation of turbulent ows, Appl. Mech. Rev 64 (2011) 133.

[71] M. Lee, N. Malaya, R.D. Moser, Petascale direct numerical simulation of turbulent channel ow on up to 786K cores, in: Proceedings of the

International Conference on High Performance Computing, Networking, Storage and Analysis, Denver, Colorado, 2013.

[72] Y. Kaneda, T. Ishihara, M. Yokokawa, K. Itakura, A. Uno, Energy dissipation rate and energy spectrum in high resolution direct numerical simulations of

turbulence in a periodic box, Phys. Fluids 15 (2003) L21L24.

[73] Y. Kaneda, T. Ishihara, High-resolution direct numerical simulation of turbulence, J. Turbul. 7 (2006) 117.

[74] W.C. Reynolds, The potential and limitations of direct and large eddy simulations, in: John L. Lumley (Ed.), Whither Turbulence? Turbulence at the

Crossroads, Lecture Notes in Physics, 357, Springer-Verlag, NY, 1989, pp. 313342.

[75] N.D. Sandham, Introduction to direct numerical simulation, in: B.E. Launder, N.D. Sandham (Eds.), Closure Strategies for Turbulent and Transitional

Flows, Cambridge University Press, Cambridge, UK, 2002.

[76] J. Jimenez, P. Moin, The minimal ow unit in near-wall turbulence, J. Fluid Mech. 225 (1991) 213240.

[77] H. Choi, P. Moin, J. Kim, Active turbulence control for drag reduction in wall-bounded ows, J. Fluid Mech. 262 (1994) 75110.

[78] G.N. Coleman, R.D. Sandberg, A primer on direct numerical simulation of turbulence-methods, procedures and guidelines. Tech. Rep AFM-09/01a.

Aerodynamics & Flight Mechanics Research Group, School of Engineering Sciences University of Southampton, Southampton, UK, 2010.

[79] N.C. Markatos, Computer simulation techniques for turbulent ows, Encycl. Fluid Mech. 6 (1) (1985) 12591275.

[80] N.C. Markatos, Dynamic computer modelling of environmental systems for decision making, risk assessment and design, Asia-Pac. J. Chem. Eng. 7

(2012) 182205.

[81] P. Bradshaw, An Introduction to Turbulence and its Measurements, Pergamon, Oxford, 1971.

[82] W. Frost, J. Bitte, Statistical Concepts in Turbulence, Handbook of Turbulence, vol. 1, Plenum Press, New York, 1977.

[83] L.S.G. Kovasznay, Turbulent shear ows, Convegno Sulla Teoria Della Turbulenza, Rome, 1970.

[84] K. Hanjalic, One-point closure models for buoyancy-driven turbulent ows, Annu. Rev. Fluid Mech. 34 (2002) 321347.

[85] K. Hanjalic, Will RANS survive LES? a view of perspectives, J. Fluids Eng. 127 (2005) 831839.

728

[86]

[87]

[88]

[89]

[90]

[91]

[92]

[93]

[94]

[95]

[96]

[97]

[98]

[99]

[100]

[101]

[102]

[103]

[104]

[105]

[106]

[107]

[108]

[109]

[110]

[111]

[112]

[113]

[114]

[115]

[116]

[117]

[118]

[119]

[120]

[121]

[122]

[123]

[124]

[125]

[126]

[127]

[128]

[129]

[130]

[131]

[132]

[133]

[134]

[135]

[136]

[137]

[138]

[139]

D. Wilcox, Turbulence Modelling for CFD, third ed., DCW Industries, Inc, 2006.

W. Rodi, Turbulence models and their application in hydraulics, A state-of-the-art review, in: IAHR Monographs, third ed., CRC Press, 1993.

L. Prandtl, Report on investigation of developed turbulence, ZAMM 5 (1925) 136139.

T. Cebeci, A.M.O. Smith, Analysis of Turbulent Boundary Layers, Academic Press, New York, 1974.

B.S.Baldwin, H. Lomax, Thin layer approximation and algebraic model for separated turbulent ows, AIAA 16 Aerospace Sciences Meeting, 1978.

K. Hanjalic, Closure models for incompressible turbulent ows (lecture series), in: J.P.A.J. Beeck, C. Benocci (Eds.), Introduction to Turbulence

Modelling, Von Kaman Institute for Fluid Dynamics, Belgium, 2004, pp. 175.

D.A. Johnson, L.S. King, A mathematically simple turbulence closure model for attached and separated turbulent boundary layers, AIAA J. 23 (1985)

16841692.

D.A. Johnson, Transonic separated ow predictions with an eddy-viscosity/-Reynolds-stress closure model, AIAA J. 25 (1987) 252259.

D.A. Johnson, T.J. Coakley, Improvements to a non-equilibrium algebraic turbulence model, AIAA J. 28 (1990) 20002003.

A.M. Savill, T.B. Gatski, P.A. Lindberg, A pseudo-3D extension to the Johnson-King model and its application to the EuropeExpt S-duct, in: O.

Pironneau, W. Rodi, I.L. Rhyming, A.M. Savill, T.V. Truong (Eds.) Numerical Simulation of Unsteady Flows and Transitions to Turbulence, Cambridge,

1992, pp 158165.

B.S. Baldwin, T.J. Barth, A one-equation turbulence transport model for high-Reynolds number wall-bounded ows. AIAA, Aerospace Sciences

Meeting, 1991.

P.R. Spalart, S.R. Allmaras, A one-equation turbulence model for aerodynamics ows, La Recherche Aerospatiale 1 (1994) 521.

P.R. Spalart, M. Shur, On the sensitization of turbulence models to rotation and curvature, Aerosp. Sci. Technol. 5 (1997) 297302.

J. Dacles-Mariani, G.G. Zilliac, J.S. Chow, P. Bradshaw, Numerical/experimental study of a wingtip vortex in the near eld, AIAA J. 33 (1995) 1561

1568.

M.M. Rahman, T. Siikonen, R.K. Agarwal, Improved low-Reynolds-number one-equation turbulence model, AIAA J. 49 (2011) 735747.

S.J. Karabelas, N.C. Markatos, Water vapor condensation in forced convection ow over an airfoil, Aerosp. Sci. Tech. 12 (2008) 150158.

S.J. Karabelas, High Reynolds number ow past a apping multi-element airfoil, in: 43 AIAA Fluid Dynamics Conference, San Diego, USA, 2013.

S.M. Klausmeyer, J.C. Lin, Comparative results from a CFD challenge over a 2D three-element high-lift airfoilm, NASA TM, 1997.

J.C. Lin, C.J. Dominik, Parametric investigation of a high-lift airfoil at high Reynolds numbers, J. Aircr. 34 (1997) 485491.

E. Fares, W. Schroder, A general one-equation turbulence model for free shear and wall-bounded ows, Flow Turbul. Combust. 73 (2004) 187215.

G. Alfonsi, Reynolds-averaged NavierStokes equations for turbulence modelling, Appl. Mech. Rev. 62 (2009) 120.

B.E. Launder, B.I. Sharma, Application of the energy dissipation model of turbulence to the calculation of ow near a spinning disk, Lett. Heat Mass

Transfer 1 (1974) 131138.

D.C. Wilcox, Reassessment of the scale-determining equation for advanced turbulence models, AIAA J. 26 (1988) 12991310.

T.-H. Shih, W.W. Liou, A. Shabbir, Z. Yang, J. Zhu, A new ke eddy viscosity model for high Reynolds number turbulent ows, Comput. Fluids 24 (1995)

(1995) 227238.

V. Yakhot, S.A. Orszag, S. Thangam, T.B. Gatski, C.G. Speziale, Development of turbulence models for shear ows by a double expansion technique,

Phys. Fluids A Fluid Dyn. 4 (1992) 15101520.

C.D. Argyropoulos, M.N. Christolis, Z. Nivolianitou, N.C. Markatos, Assessment of acute effects for re-ghters during a fuel-tank re, in: International

Conference Working On Safety Net, Prevention of Occupational Accident in a Changing Work Environment, Crete, Greece, 2008.

N.C. Markatos, C. Christolis, C. Argyropoulos, Mathematical modelling of toxic pollutants dispersion from large tank res and assessment of acute

effects for re ghters, Int. J. Heat Mass Transfer 52 (2009) 40214030.

C.D. Argyropoulos, G.M. Sideris, M.N. Christolis, Z. Nivolianitou, N.C. Markatos, Modelling pollutants dispersion and plume rise from large

hydrocarbon tank res in neutrally stratied atmosphere, Atmos. Environ. 44 (2010) 803813.

C.D. Argyropoulos, M.N. Christolis, Z. Nivolianitou, N.C. Markatos, A hazards assessment methodology for large liquid hydrocarbon fuel tanks, J. Loss

Prev. Process. Ind. 25 (2012) 329335.

N. Koutsourakis, J. Bartzis, N. Markatos, Evaluation of Reynolds stress, ke and RNG ke turbulence models in street canyon ows using various

experimental datasets, Environ. Fluid Mech. 12 (2012) 379403.

S.J. Karabelas, B.C. Koumroglou, C.D. Argyropoulos, N.C. Markatos, High Reynolds number turbulent ow past a rotating cylinder, Appl. Math. Modell.

36 (2012) 379398.

A.N. Kolmogorov, Equations of turbulent motion of an incompressible uid, Izv Acad. Sci. USSR Phys. 6 (1942) 5658.

B.E. Launder, D.B. Spalding, Mathematical Models of Turbulence, Academic Press, London, 1972.

F.R. Menter, Review of the shear-stress transport turbulence model experience from an industrial perspective, Int. J. Comput. Fluid Dyn. 23 (2009)

305316.

F.R. Menter, Two-equation eddy-viscosity turbulence models for engineering applications, AIAA J. 32 (1994) 15981605.

P. Smirnov, F.R. Menter, Sensitizing of the SST turbulence model to rotation and curvature by applying the Spalart-Shur correction term, in:

Proceedings of ASME Turbo Expo, Berlin, 2008.

K. Hanjalic, B.E. Launder, R. Schiestel, Multiple-time-scale concepts in turbulent transport modelling, in: L.J.S. Bradbury et al. (Eds.), Turbulent Shear

Flows, 2, Springer-Verlag, Berlin, 1980, pp. 3649.

O. Heynes, M. Cotton, T. Craft, Eddy-viscosity and stress-transport turbulence models in application to a plane synthetic jet, Flow Turbul. Combust. 91

(2013) 931947.

Q. Zhao, S. Armeld, K. Tanimoto, Numerical simulation of breaking waves by a multi-scale turbulence model, Coast. Eng. 51 (2004) 5380.

O. Gregoire, D. Soufand, S. Gauthier, R. Schiestel, A two-time-scale turbulence model for compressible ows: turbulence dominated by mean

deformation interaction, Phys. Fluids 11 (1999) 37933807.

K.R. Kim, M.A. Cotton, T.J. Craft, O.R. Heynes, On the dynamics and frequency response of fully-pulsed turbulent round jets: computations using twotime-scale/strain sensitized eddy viscosity models, Int. J. Heat Fluid Flow 29 (2008) 16501669.

T.S. Klein, T.J. Craft, H. Iacovides, Development in two-time-scale turbulence models applied to non-equilibrium ows, TSFP-7, 2011.

V.C. Patel, W. Rodi, G. Scheuerer, Turbulence models for near-wall and low Reynolds number ows a review, AIAA J. 23 (1985) (1985) 13081319.

C.K.G. Lam, K.A. Bremhorst, Modied form of ke model for predicting wall turbulence, ASME J. Fluids Eng. 103 (1981) 456460.

J. Bredberg, S.-H. Peng, L. Davidson, An improved kx turbulence model applied to recirculating ows, Int. J. Heat Fluid Flow 23 (2002) 731743.

D.C. Wilcox, Simulation of transition with a two-equation turbulence model, AIAA J. 32 (1994) 247255.

S.H. Peng, L. Davidson, S. Holmberg, A modied low-Reynolds-number kx model for recirculating ows, J. Fluid Eng. 119 (1997) 867875.

P.R. Resende, F.T. Pinho, B.A. Younis, K. Kim, R. Sureshkumar, Development of a low-Reynolds-number kx model for FENE-P uids, Flow Turbul.

Combust. 90 (2013) 6994.

T.B. Gatski, C.G. Speziale, On explicit algebraic stress models for complex turbulent ows, J. Fluid Mech. 254 (1993) 5978.

T. Shih, J. Zhu, J. Lumley, A new Reynolds stress algebraic equation model, Comput. Meth. Appl. Mech. Eng. 125 (1995) 287302.

A. Hellsten, S. Wallin, Explicit algebraic Reynolds stress and non-linear eddy-viscosity models, Int. J. Comput. Fluid Dyn. 23 (2009) 349361.

T.J. Craft, B.E. Launder, K. Suga, Prediction of turbulent transitional phenomena with a nonlinear eddy-viscosity model, Int. J. Heat Fluid Flow 18

(1997) 1528.

D.D. Apsley, M.A. Leschziner, A new low-Reynolds-number nonlinear two-equation turbulence model for complex ows, Int. J. Heat Fluid Flow 19

(1988) 209222.

P.A. Durbin, Near-wall turbulence closure modelling without damping functions, Theor. Comput. Fluid Dyn. 3 (1991) 113.

729

[140] K. Hanjalic, M. Popovac, M. Hadziabdic, A robust near-wall elliptic relaxation eddy-viscosity turbulence model for CFD, Int. J. Heat Fluid Flow 25

(2004) 10471051.

[141] P.A. Durbin, B.A. Pettersson, The elliptic relaxation method, in: B.E. Launder, N.D. Sandham (Eds.), Closure Strategies for Turbulent and Transitional

Flows, Cambridge University Press, Cambridge, UK, 2002, pp. 127152.

[142] P.A. Durbin, On the ke stagnation point anomaly, Int. J. Heat Fluid Flow 17 (1996) 8990.

[143] F. Billard, D. Laurence, A robust ket2/k elliptic blending turbulence model applied to near-wall, separated and buoyant ows, Int. J. Heat Fluid Flow

33 (2012) 4558.

[144] R. Manceau, K. Hanjalic, Elliptic blending model: a new near-wall Reynolds-stress turbulence closure, Phys. Fluids 14 (2002) 744754.

[145] F. Billard, A. Revell, T. Craft, Application of recently developed elliptic blending based models to separated ows, Int. J. Heat Fluid Flow 35 (2012) 141

151.

[146] B.J. Daly, F.H. Harlow, Transport equations in turbulence, Phys. Fluids 13 (1970) 26342649.

[147] J. Magnaudet, The modelling of inhomogeneous turbulence in the absence of mean velocity gradients, Fourth Eur. Turbulence Conf., Delft, The

Netherlands, 1992.

[148] Y. Nagano, M. Tagawa, Turbulence models for triple velocity and scalar correlations, Turbulent Shear Flows, vol. 7, Springer, Berlin, 1991, pp. 6378.

[149] K. Hanjalic, Advanced turbulence closure models: a view of current status and future prospects, Int. J. Heat Fluid Flow 15 (1994) 178203.

[150] D. Naot, A. Shavit, M. Wolfshtein, Two-point correlation model and the redistribution of Reynolds stresses, Phys. Fluids 16 (1973) 738743.

[151] B.E. Launder, G.J. Reece, W. Rodi, Progress in the development of a Reynolds-stress turbulence closure, J. Fluid Mech. 68 (1975) 537566.

[152] C.G. Speziale, S. Sarkar, T.B. Gatski, Modelling the pressure-strain correlation of turbulence: an invariant system dynamic approach, J. Fluid Mech. 227

(1991) 245272.

[153] T.J. Craft, B.E. Launder, A Reynolds stress closure designed for complex geometries, Int. J. Heat Fluid Flow 17 (1996) 245254.

[154] B.E. Launder, D.P. Tselepidakis, Application of a new second-moment closure to turbulent channel ow rotating in orthogonal mode, Int. J. Heat Fluid

Flow 15 (1994) 210.

[155] K. Hanjalic, S. Jakirlic, Second-moment turbulence closure modelling, in: B.E. Launder, N.D. Sandham (Eds.), Closure Strategies for Turbulent and

Transitional Flows, Cambridge University Press, Cambridge, UK, 2002, pp. 47101.

[156] A. Kovar-Panskus, P. Louka, J.F. Sini, E. Savory, M. Czech, A. Abdelqari, P.G. Mestayer, N. Toy, Inuence of geometry on the mean ow within urban

street canyonsa comparison of wind tunnel experiments and numerical simulations, Water Air Soil Pollut. 2 (2002) 365380.

[157] X.X. Li, D.Y.C. Leung, C.H. Liu, K.M. Lam, Physical modeling of ow eld inside urban street canyons, J. Appl. Meteorol. Clim. 47 (2008) 20582067.

[158] X.X. Li, C.H. Liu, D.Y.C. Leung, Large-eddy simulation of ow and pollutant dispersion in high-aspect-ratio urban street canyons with wall model,

Bound-Layer Meteorol. 129 (2008) 249268.

[159] P.A. Durbin, Reynolds stress model for near-wall turbulence, J. Fluid Mech. 249 (1993) 465498.

[160] K. Hanjalic, S. Kenjeres, Some developments in turbulence modelling for wind and environmental engineering, J. Wind Eng. Ind. Aerodyn. 96 (2008)

15371570.

[161] L. Thielen, H.J.J. Jonker, K. Hanjalic, Symmetry breaking of ow and heat transfer in multiple impinging jets, Int. J. Heat Fluid Flow 24 (2003) 444453.

[162] B. Basara, S. Jakirlic, A new hybrid turbulence modelling strategy for industrial CFD, Int. J. Numer. Methods Fluids 42 (2003) 89116.

[163] S.B. Pope, A more general effective-viscosity hypothesis, J. Fluid Mech. 72 (1975) 331340.

[164] T. Jongen, T.B. Gatski, General explicit algebraic stress relations and best approximation for three-dimensional ows, Int. J. Eng. Sci. 36 (1998) 739

763.

[165] S. Wallin, A.V. Johansson, An explicit algebraic Reynolds stress model for incompressible and compressible turbulent ows, J Fluid Mech. 403 (2000)

89132.

[166] D.B. Spalding, Chemical reaction in turbulent uids, J. Physico-chemical Hydrodyn. 4 (1983) 323336.

[167] M.R. Malin, D.B. Spalding, A two-uid model of turbulence and its application to heated plane jets and wakes, J. Physico-chemical Hydrodyn. 5 (1984)

339362.

[168] N.C. Markatos, C.A. Kotsifaki, One-dimensional, two-uid modelling of turbulent premixed ames, Appl. Math. Modell. 18 (1994) 646657.

[169] Y.M. Shen, C.-O. Ng, A.T. Chwang, A two-uid model of turbulent two-phase ow for simulating turbulent stratied ows, Ocean Eng. 30 (2003) 153

161.

[170] K.Z. Yu, G.-L. Ding, T.-J. Chen, Modied two-uid model for air curtains in open vertical display cabinets, Int. J. Refrig. 31 (2008) 472482.

[171] D.L. Youngs, Representation of the molecular mixing process in a two-phase ow turbulent mixing model, in: Proc. Fifth Int. Workshop on the Physics

of Compressible Turbulent Mixing, World Scientic, Singapore, 1995, pp. 8388.

[172] A. Llor, P. Bailly, A new turbulent two-eld concept for modelling RayleighTaylor, RichtmyerMeshkov, and KelvinHelmholtz mixing layers, Lasers

Beams 21 (2003) 311315.

[173] D. Liu, C. Wu, K. Linden, J. Ducoste, Numerical simulation of UV disinfection reactors: evaluation of alternative turbulence models, Appl. Math. Modell.

31 (2007) 17531769.

[174] Z. Cao, H. Han, B. Gu, A novel optimization strategy for the design of air curtains for open vertical refrigerated display cases, Appl. Therm. Eng. 31

(2011) 30983105.

[175] H. Tinoco, H. Lindqvist, W. Frid, Numerical simulation of industrial ows, numerical simulations examples and applications in computational

uid dynamics, Prof. Lutz Angermann (Ed.), ISBN: 978-953-307-153-4, InTech, DOI: http://dx.doi.org/10.5772/13216, 2010. Available from:

<http://www.intechopen.com/books/numerical-simulations-examples-and-applications-in-computational-uid-dynamics/numerical-simulation-ofindustrial-ows>.

[176] M. Meinke, Th. Rister, F. Rutten, A. Schvorack, Simulation of internal and free turbulent ows, in: H.J. Bungartz, F. Durst, C. Zenger (Eds.), High

Performance Scientic and Engineering Computing, Springer-Verlag, 1998, pp. 6179.

[177] J. Frohlich, W. Rodi, Introduction to large eddy simulation of turbulent ows, in: B.E. Launder, N.D. Sandham (Eds.), Closure Strategies for Turbulent

and Transitional Flows, Cambridge University Press, Cambridge, UK, 2002, pp. 267298.

[178] J. Smagorinsky, General circulation experiments with the primitive equations, Mon. Weather Rev. 91 (1963) 99164.

[179] D.K. Lilly, The representation of small-scale turbulence in numerical simulation experiments, in: Proceedings IBM Scientic Computing Symposium

on Environmental Sciences, 1967, pp. 195210.

[180] J.W. Deardorff, A numerical study of three-dimensional turbulent channel ow at large Reynolds numbers, J. Fluid Mech. 41 (1970) 453480.

[181] R.H. Kraichnan, Eddy viscosity in two and three dimensions, J. Atmos. Sci. 33 (1976) 15211536.

[182] J.R. Chasnov, Simulation of the Kolmogorov inertial subrange using an improved subgrid model, Phys. Fluids A 3 (1991) 188200.

[183] U. Schumann, Subgrid scale model for nite difference simulations of turbulent ows in plane channels and annuli, J. Comput. Phys. 18 (1975) 376

404.

[184] P. Moin, J. Kim, Numerical investigation of turbulent channel ow, J. Fluid Mech. 118 (1982) 341377.

[185] U. Piomelli, High Reynolds number calculations using the dynamic subgrid-scale stress model, Phys. Fluids A 5 (1993) 14841490.

[186] A. Leonard, Energy cascade in large-eddy simulations of turbulent uid ows, Proceedings of Turbulent Diffusion in Environmental Pollution,

Charlottesville, vol. 18, Elsevier, 1974, pp. 237248.

[187] A. Aldama, Filtering techniques for turbulent ow simulations, Lecture Notes in Engineering, vol. 56, Springer, Berlin, 1990.

[188] R.A. Clark, J.H. Ferziger, W.C. Reynolds, Evaluation of subgrid-scale models using an accurately simulated turbulent ow, J. Fluid Mech. 91 (1979) 1

16.

[189] J.P. Boris, F.F. Grinstein, E.S. Oran, R.L. Kolbe, New insights into large eddy simulation, Fluid Dyn. Res. 10 (1992) 199228.

[190] S.J. Karabelas, Large eddy simulation of high-Reynolds number ow past a rotating cylinder, Int. J. Heat Fluid Flow 31 (2010) 518527.

730

[191] M. Germano, U. Piomelli, P. Moin, W.H. Cabot, A dynamic subgrid scale eddy viscosity model, Phys. Fluids A 3 (1991) 17601765.

[192] D.K. Lilly, A proposed modication of the Germano subgrid-scale closure method, Phys. Fluids A 4 (1992) 633635.

[193] S. Ghosal, T.S. Lund, P. Moin, K. Akselvoll, A dynamic localization model for large-eddy simulation of turbulent ows, J. Fluid Mech. 286 (1995) 229

255.

[194] C. Meneveau, T.S. Lund, W.H. Cabot, A lagrangian dynamic subgrid-scale model of turbulence, J. Fluid Mech. 319 (1996) 353385.

[195] J. Bardina, J.H. Ferziger, W.C. Reynolds, Improved subgrid scale models for large eddy simulation, in: AIAA 13th Fluid and Plasma Dynamics

Conference, Snowmass, Colorado, 1980.

[196] A. Leonard, Large-eddy simulation of chaotic convection and beyond, Tech. Rep. 970204 AIAA, 1997.

[197] B. Geurts, Inverse modelling for large eddy simulation, Phys. Fluids 9 (1997) 35853587.

[198] Y. Zang, R.L. Street, J.R. Koseff, A dynamic mixed subgrid scale model and its application to turbulent recirculating ows, Phys. Fluid 5 (1993) 3186

3196.

[199] B. Vreman, B. Geurts, H. Kuerten, On the formulation of the dynamic mixed subgrid-scale model, Phys. Fluid 6 (1994) 40574059.

[200] A. Leonard, G.S. Winckelmans, A tensor-diffusivity subgrid model for large-eddy simulation, in: P.R. Voke, N.D. Sandham, L. Kleiser (Eds.), Direct and

Large-Eddy Simulation III, 1999, pp. 147162.

[201] S. Stolz, N.A. Adams, An approximate deconvolution procedure for large-eddy simulation, Phys. Fluids 11 (1999) 16991701.

[202] S. Stolz, N.A. Adams, L. Kleiser, The approximate deconvolution model for LES of compressible ows and its application to shock-turbulent-boundarylayer interaction, Phys. Fluids 13 (2001) 29853001.

[203] S. Stolz, N.A. Adams, L. Kleiser, An approximate deconvolution model for large eddy simulations with application to incompressible wall-bounded

ows, Phys. Fluids 13 (2001) 9971015.

[204] N.A. Adams, S. Hickel, T. Kempe, J.A. Domaradzki, On the relation between subgrid-scale modelling and numerical discretization in large-eddy

simulation, in: Complex Effects in Large Eddy Simulations, Springer, Berlin Heidelberg, 2007, pp. 1527.

[205] B.J. Geurts, Elements of direct and large-eddy simulation, R.T. Edwards Inc, Flourtown, 2003.

[206] O. Metais, M. Lesieur, Spectral large-eddy simulation of isotropic and stably stratied turbulence, J. Fluid Mech. 239 (1992) 157194.

[207] F. Ducros, P. Comte, M. Lesieur, Large-eddy simulation of transition to turbulence in a boundary layer developing spatially over a at plate, J. Fluid

Mech. 326 (1996) 136.

[208] J.A. Domaradzki, E.M. Saika, A subgrid-scale model based on the estimation of unresolved scales of turbulence, Phys. Fluids 9 (1997) 21482164.

[209] J.A. Domaradzki, K.C. Loh, The subgrid-scale estimation model in the physical space representation, Phys. Fluids 11 (1999) 23302342.

[210] C. Cao, D. Holm, E.S. Titi, On the clark-a model of turbulence: global regularity and long-time dynamics, J. Turbul. 6 (2005) 111.

[211] A. Cheskidov, D.D. Holm, E. Olson, E.S. Titi, On a Leray-a model of turbulence, Proc. R. Soc. A Math. Phys. Eng. Sci. 461 (2005) 629649.

[212] S. Chen, C. Foias, D.D. Holm, E. Olson, E.S. Titi, S. Wynne, The CamassaHolm equations and turbulence, Phys. D 133 (1999) 4965.

[213] S. Chen, C. Foias, D.D. Holm, E. Olson, E.S. Titi, S. Wynne, A connection between the CamassaHolm equations and turbulent ows in channels and

pipes, Phys. Fluids 11 (1999) 23432353.

[214] S. Chen, C. Foias, D.D. Holm, E. Olson, E.S. Titi, S. Wynne, CamassaHolm equations as a closure model for turbulent channel and pipe ow, Phys. Rev.

Lett. 81 (1998) 53385341.

[215] B.J. Geurts, D.D. Holm, Regularization modelling for large-eddy simulation, Phys. Fluids 15 (2003) 1316.

[216] J. Leray, Essai sur le mouvement dun uide visqueux emplissant lspace, Acta Math. 63 (1934) 193248.

[217] A.A. Llyin, E.M. Lunasin, E.S. Titi, A modied-Leray-a subgrid scale model of turbulence, Nonlinearity 19 (2006) 879897.

[218] U. Piomelli, Large-eddy and direct simulation of turbulent ows, in: CFD2001 9e confrence annuelle de la socit Canadienne de CFD, Kitchener,

Ontario, Canada, 2001.

[219] U. Piomelli, E. Balaras, Wall-layer models for large-eddy simulations, Annu. Rev. Fluid Mech. 34 (2002) 349374.

[220] B.J. Geurts, A. Leonard, Is LES ready for complex ows?, in: B.E. Launder, N.D. Sandham (Eds.), Closure Strategies for Turbulent and Transitional Flows,

Cambridge University Press, Cambridge, UK, 2002, pp. 720739.

[221] B.J. Geurts, Interacting errors in large-eddy simulation: a review of recent developments, J. Turbul. 7 (2006) 116.

[222] B.J. Geurts, Analysis of errors occurring in large eddy simulation, Proc. R. Soc. A Math. Phys. Eng. Sci. 367 (2009) 28732883.

[223] I. Celik, M. Klein, M. Freitag, J. Janicka, Assessment measures for URANS/DES/LES: an overview with applications, J. Turbul. 7 (2006) 127.

[224] B.J. Geurts, J. Frohlich, A framework for predicting accuracy limitations in large-eddy simulations, Phys. Fluids 14 (2002) 4144.

[225] J. Meyers, B.J. Geurts, M. Baelmans, Database analysis of errors in large eddy simulation, Phys. Fluids 15 (2003) 27402755.

[226] M. Freitag, M. Klein, An improved method to assess the quality of large eddy simulations in the context of implicit ltering, J. Turbul. 7 (2006) 111.

[227] D. Drikakis, W. Rider, High Resolution Methods for Incompressible and Low-speed Flows, Springer-Verlag, Berlin Heidelberg, 2005.

[228] J.P. Boris, D.L. Book, Flux-corrected transport. I. SHASTA, a uid transport algorithm that works, J. Comput. Phys. 11 (1973) 3869.

[229] B. van Leer, Towards the ultimate conservative difference scheme. I. The quest of monotonicity, in: H. Cabannes, R. Temam (Eds.), Lecture Notes in

Physics, vol. 18, Springer-Verlag, Berlin, 1973, pp. 163168.

[230] P. Colella, P.R. Woodward, The piecewise parabolic method (PPM) for gas-dynamical simulations, J. Comput. Phys. 54 (1984) 174201.

[231] A. Harten, High resolution schemes for hyperbolic conservation laws, J. Comput. Phys. 49 (1983) 357393.

[232] P. Colella, A direct Eulerian MUSCL scheme for gas dynamics, SIAM. J. Sci. Statist. Comput. 6 (1985) 104117.

[233] C. Fureby, M. Liefvendahl, U. Svennberg, L. Persson, T. Persson, Incompressible wall-bounded ows, in: F.F. Grinstein, L. Margolin, B. Rider (Eds.),

Implicit Large Eddy Simulation: Computing Turbulent Fluid Dynamics, Cambridge University Press, 2007, pp. 301328.

[234] G. Patnaik, F.F. Grinstein, J.P. Boris, T.R. Young, O. Parmhed, Large-scale urban simulations, in: F.F. Grinstein, L. Margolin, B. Rider (Eds.), Implicit Large

Eddy Simulation: Computing Turbulent Fluid Dynamics, Cambridge University Press, 2007, pp. 502530.

[235] J.P. Boris, On large eddy simulation using subgrid turbulence models, comment 1, in wither turbulence? Turbulence at the crossroads, Lect. Notes

Phys. 357 (1990) 344353.

[236] J.P. Boris, More for LES: a brief historical perspective of MILES, in: F.F. Grinstein, L. Margolin, B. Rider (Eds.), Implicit Large Eddy Simulation:

Computing Turbulent Fluid Dynamics, Cambridge University Press, 2007, pp. 938.

[237] S. Johansson, L. Davidson, E. Olsson, Numerical simulation of vortex shedding past triangular cylinders at high Reynolds numbers using a ke

turbulence model, Int. J. Numer. Methods Fluids 16 (1993) 859878.

[238] S. Perzon, L. Davidson, On CFD and transient ow in vehicle aerodynamics, in: SAE Technical Paper 2000-01-0873, Detroit, 2000.

[239] S. Perzon, L. Davidson, On transient modelling of the ow around vehicles using the Reynolds equation, in: J.-H. Wu, Z.-J. Zhu, F.-P. Jia, X.-B. Wen, W.

Hu (Eds.), ACFD 2000, Bejing, 2000.

[240] C. Speziale, Turbulence modelling for time-dependent RANS and VLES: a review, AIAA J. 36 (1998) 173184.

[241] P.R. Spalart, Strategies for turbulence modelling and simulations, Int. J. Heat Fluid Flow 21 (2000) 252263.

[242] S. Kenjeres, K. Hanjalic, Tackling complex turbulent ows with transient RANS, Fluid Dyn. Res. 41 (2009) 132.

[243] F.R. Menter, M. Kuntz, R. Bender, A scale-adaptive simulation model for turbulent ow predictions, AIAA Paper, 2003.

[244] S.S. Girimaji, Partially-averaged NavierStokes model for turbulence: a reynolds-averaged NavierStokes to direct numerical simulation bridging

method, J. Appl. Mech. 73 (2006) 413421.

[245] J. Rotta, Recent attempts to develop a generally applicable calculation method for turbulent shear ow layers, in: Proceedings AGARD Conference on

Turbulent Shear Flows, London, 1971.

[246] F.R. Menter, Y. Egorov. A scale-adaptive simulation model using two equation models. AIAA paper 2005-1095, Reno, NV, 2005.

[247] B. Basara, S. Krajnovic, Z. Pavlovic, P. Riugqvist, Performance analysis of partially-averaged NavierStokes methods for complex turbulent ows, in:

6th AIAA theoretical uid mechanics conference, Honolulu, Hawaii, 2011.

731

[248] J. Ma, S.H. Peng, L. Davidson, F. Wang, A low Reynolds number variant of partially-averaged NavierStokes model for turbulence, Int. J. Heat Fluid

Flow 32 (2011) 652669.

[249] L. Davidson, S.H. Peng, Embedded large-eddy simulation using the partially averaged Navier Stokes model, AIAA J. 51 (2013) 10661079.

[250] L. Davidson, Evaluation of the SST-SAS model: Channel ow, asymmetric diffuser and axi-symmetric hill, in: ECCOMAS CFD 2006, The Netherlands,

2006.

[251] B. Chaouat, R. Schiestel, A new partially integrated transport model for subgrid-scale stresses and dissipation rate for turbulent developing ows,

Phys. Fluids 17 (2005) 065106.

[252] R. Schiestel, A. Dejoan, Towards a new partially integrated transport model for coarse grid and unsteady turbulent ow simulations, Theor. Comput.

Fluid Dyn. 18 (2005) 443468.

[253] S.T. Johansen, J. Wu, W. Shyy, Filter-based unsteady RANS computations, Int. J. Heat Fluid Flow 25 (2004) 1021.

[254] N.S. Liu, T.H. Shih, Turbulence modelling for very large-eddy simulation, AIAA J. 44 (2006) 687697.

[255] T.H. Shih, N.S. Liu, Modelling of internal reacting ows and external static stall ows using RANS and PRNS, Flow Turbul. Combust. 81 (2008) 279

299.

[256] P. Batten, U. Goldberg, S. Chakravarthy, Interfacing statistical turbulence closures with large-eddy simulation, AIAA J. 42 (2004) 485492.

[257] A. Ruprecht, T. Helmrich, I. Buntic, Very large eddy simulation for the prediction of unsteady vortex motion, in: Modelling Fluid Flow, Springer-Berlin

Heidelberg, 2004, pp. 229246.

[258] J.B. Perot, J. Gadebusch, A self-adapting turbulence model for ow simulation at any mesh resolution, Phys. Fluids 19 (2007) 115105.

[259] J.B. Perot, J. Gadebusch, A stress transport equation model for simulating turbulence at any mesh resolution, Theor. Comput. Fluid Dyn. 23 (2009)

271286.

[260] K.J. Hsieh, F.S. Lien, E. Yee, Towards a unied turbulence simulation approach for wall-bounded ows, Flow Turbul. Combust. 84 (2010) 193218.

[261] D. Lakehal, M. Labois, A new modelling strategy for phase-change heat transfer in turbulent interfacial two-phase ow, Int. J. Multiphase Flow 37

(2011) 627639.

[262] X. Han, S. Krajnovic, An efcient very large eddy simulation model for simulation of turbulent ow, Int. J. Numer. Methods Fluids 71 (2013) 1341

1360.

[263] P.R. Spalart, W-H Jou, M. Strelets, S.R. Allmaras, Comments on the feasibility of LES for wings, and on a hybrid RANS/LES approach, in: C. Lin, Z. Lin

(Eds.), Advances in DNS/LES, Creyden Press, Columbus, OH, 1997, pp. 137147.

[264] P.R. Spalart, Detached-eddy simulation, Annu. Rev. Fluid Mech. 41 (2009) 181202.

[265] A. Travin, M. Shur, M. Strelets, P. Spalart, Detached-eddy simulations past a circular cylinder, Flow Turbul. Combust. 63 (2000) 293313.

[266] M.L. Shur, P.R. Spalart, M.K. Strelets, A.K. Travin, A hybrid RANS-LES approach with delayed-DES and wall-modelled LES capabilities, Int. J. Heat Fluid

Flow 29 (2008) 16381649.

[267] P. Spalart, S. Deck, M. Shur, K. Squires, M. Strelets, A. Travin, A new version of detached-eddy simulation, resistant to ambiguous grid densities, Theor.

Comput. Fluid Dyn. 20 (2006) 181195.

[268] M. Strelets, Detached eddy simulation of massively separated ows, AIAA J. (2001) 118.

[269] M. Gritskevich, A. Garbaruk, J. Schiitze, F. Menter, Development of DDES and IDDES formulations for the k-x shear stress transport model, Flow Turb.

Comb. 88 (2012) 431449.

[270] M. Leschziner, N. Li, F. Tessicini, Simulating ow separation from continuous surfaces: routes to overcoming the Reynolds number barrier, Philos.

Trans. Roy. Soc. A Math. Phys. Eng. Sci. 367 (2009) 28852903.

[271] L. Davidson, Hybrid LES-RANS: back scatter from a scale-similarity model used as forcing, Philos. Trans. R. Soc. A Math. Phys. Eng. Sci. 367 (2009)

29052915.

[272] F. Hamba, A hybrid RANS/LES simulation of turbulent channel ow, Theor. Comput. Fluid Dyn. 16 (2003) 387403.

[273] L. Temmerman, M. Hadziabdic, M.A. Leschziner, K. Hanjalic, A hybrid two-layer URANS-LES approach for large eddy simulation at high Reynolds

numbers, Int. J. Heat Fluid Flow 26 (2005) 173190.

[274] P. Tucker, L. Davidson, Zonal kl based large eddy simulation, Comput. Fluids 33 (2004) 267287.

[275] L. Davidson, S. Dahlstrom, Hybrid LES-RANS: an approach to make LES applicable at high Reynolds number, Int. J. Comput. Fluid Dyn. 19 (2005) 415

427.

[276] X. Xiao, J.R. Edwards, H.A. Hassan, Blending functions in hybrid large eddy/Reynolds-averaged NavierStokes simulations, AIAA J. 42 (2004) 2508

2515.

[277] I. Marusic, B.J. McKeon, P.A. Monkewitz, H.M. Nagib, A.J. Smits, K.R. Sreenivasan, Wall-bounded turbulent ows at high Reynolds numbers: recent

advances and key issues, Phys. Fluids 22 (2010) 065103.

[278] T. Mullin, Experimental studies of transition to turbulence in a pipe, Annu. Rev. Fluid Mech. 43 (2011) 124.

[279] N.V. Nikitin, Direct numerical modelling of three-dimensional turbulent ows in pipes of circular cross section, Fluid Dyn. 29 (1994) 749758.

[280] J.G.M. Eggels, F. Unger, M.H. Weiss, J. Westerweel, R.J. Adrian, R. Friedrich, F.T.M. Nieuwstadt, Fully developed turbulent pipe ow: a comparison

between direct numerical simulation and experiment, J. Fluid Mech. 268 (1994) 175210.

[281] Y. Zhang, A. Chandi, A.G. Tomboulides, S.A Orszag, Simulation of pipe ow, in: AGARD Conference Proceedings on Applications of Direct and Large

Eddy Simulations to Transition and Turbulence, 1994.

[282] P. Loulou, R.D. Moser, N.N. Mansour, B.J. Cantwell, Direct numerical simulation of incompressible pipe ow using a B-spline spectral method, NASA

report, 1997, pp. 1138.

[283] A. Leonard, A. Wray, A new numerical method for simulation of three-dimensional ow in a pipe, Lect. Notes Phys. 170 (1982) 335342.

[284] P. Orlandi, M. Fatica, Direct simulations of turbulent ow in a pipe rotating about its axis, J. Fluid Mech. 343 (1997) 4372.

[285] P. Orlandi, D. Ebstein, Turbulent budgets in rotating pipes by DNS, Int. J. Heat Fluid Flow 21 (2000) 499505.

[286] S. Schmidt, D.M. McIver, H.M. Blackburn, M. Rudman, G.J. Nathan, Spectral element based simulation of turbulent pipe ow, in: 14th Australasian

uid mechanics conference, Adelaide University, Adelaide, Australia, 2001.

[287] J.M.J. Den Toonder, F.T.M. Nieuwstadt, Reynolds number effects in a turbulent pipe ow for low to moderate Re, Phys. Fluids 9 (1997) 33983409.

[288] C. Wagner, T.J. Huttl, R. Friedrich, Low-Reynolds-number effects derived from direct numerical simulations of turbulent pipe ow, Comput. Fluids 30

(2001) 581590.

[289] K. Fukagata, N. Kasagi, Highly energy-conservative nite difference method for the cylindrical coordinate system, J. Comput. Phys. 181 (2002) 478

498.

[290] M.P.B. Veenman, Statistical analysis of turbulent pipe ow: numerical approach, Faculty of Mechanical Engineering, Eindhoven University of

Technology, Eindhoven, 2004.

[291] N. Nikitin, A. Yakhot, Direct numerical simulation of turbulent ow in elliptical ducts, J. Fluid Mech. 532 (2005) 141164.

[292] J. Kim, D. Kim, H. Choi, An immersed-boundary nite-volume method for simulations of ow in complex geometries, J. Comput. Phys. 171 (2001)

132150.

[293] T.V. Voronova, N.V. Nikitin, Direct numerical simulation of the turbulent ow in an elliptical pipe, Comp. Math. Math. Phys. 46 (2006) 13781386.

[294] T.V. Voronova, N.V. Nikitin, Results of direct numerical simulation of turbulent ow in a pipe of elliptical cross-section, Fluid Dyn. 42 (2007) 201211.

[295] X. Wu, P. Moin, A direct numerical simulation study on the mean velocity characteristics in turbulent pipe ow, J. Fluid Mech. 608 (2008) 81112.

[296] B.J. Boersma, Direct numerical simulation of turbulent pipe ow up to a Reynolds number of 61000, J. Phys. Conf. Ser. 318 (2011) 042045.

[297] X. Wu, J.R. Baltzer, R.J. Adrian, Direct numerical simulation of a 30 R long turbulent pipe ow at R+ = 685: large- and very large-scale motions, J. Fluid

Mech. 698 (2012) 235281.

732

[298] G.K. Khoury, P. Schlatter, A. Noorani, P. Fischer, G. Brethouwer, A. Johansson, Direct numerical simulation of turbulent pipe ow at moderately high

Reynolds numbers, Flow Turbul. Combust. 91 (2013) 475495.

[299] K. Lam, S. Banerjee, On the condition of streak formation in a bounded turbulent ow, Phys. Fluids A 4 (1992) 306320.

[300] S. Komori, R. Nagaosa, Y. Murakami, S. Chiba, K. Ishii, K. Kuwahara, Direct numerical simulation of three-dimensional open-channel ow with zeroshear gas-liquid interface, Phys. Fluids A Fluid Dyn. 5 (1993) 115125.

[301] P. Lombardi, V.D. Angelis, S. Banerjee, Direct numerical simulation of near-interface turbulence in coupled gasliquid ow, Phys. Fluids 8 (1996)

16431665.

[302] V.D. Angelis, Numerical investigation and modelling of mass transfer processes at shared gasliquid interface (Ph.D. thesis), USCB, 1998.

[303] V.D. Angelis, P. Lombardi, P. Andreussi, S. Banerjee, Microphysics of scalar transfer at air-water interfaces, in: Proceedings of the IMA conference:

wind-over-wave coupling, perspectives and prospects, 1999, pp. 641651.

[304] G.H. Yeoh, J. Tu, Computational Techniques for Multiphase Flows, A Butterworth-Heinemann, 2009.

[305] M. Fulgosi, D. Lakehal, S. Banerjee, V. De Angelis, Direct numerical simulation of turbulence in a sheared airwater ow with a deformable interface, J.

Fluid Mech. 482 (2003) 319345.

[306] S. Banerjee, D. Lakehal, M. Fulgosi, Surface divergence models for scalar exchange between turbulent streams, Int. J. Multiphase Flow 30 (2004) 963

977.

[307] M.Y. Lin, C.H. Moeng, W.G. Tsai, P.P. Sullivan, S.E. Belcher, Direct numerical simulation of wind-wave generation processes, J. Fluid Mech. 616 (2008)

130.

[308] D. Lakehal, M. Fulgosi, S. Banerjee, G. Yadigaroglu, Turbulence and heat exchange in condensing vaporliquid ow, Phys. Fluids 20 (2008). 06510118.

[309] P. Trontin, S. Vincent, J.L. Estivalezes, J.P. Caltagirone, Direct numerical simulation of a freely decaying turbulent interfacial ow, Int. J. Multiphase

Flow 36 (2010) 891907.

[310] F. Unger, R. Friedrich, Large eddy simulation of fully-developed turbulent pipe ow, in: Proceedings 8th symposium on turbulent shear ows, 1991,

pp. 19-13-1119-13-16.

[311] J.G.M. Eggels, F.T.M. Nieuwstadt, Large-eddy simulations of turbulent ow in an axially rotating pipe, in: Proceedings of the 9th symposium on

turbulent shear ows, 1993, pp. 310313.

[312] B.J. Boersma, F.T.M. Nieuwstadt, Large-eddy simulation of turbulent ow in a curved pipe, J. Fluids Eng. 118 (1996) 248254.

[313] Z. Yang, Large eddy simulation of fully developed turbulent ow in a rotating pipe, Int. J. Numer. Methods Fluids 33 (2000) 681694.

[314] M. Rudman, H.M. Blackburn, Large eddy simulation of turbulent pipe ow, in: Second International Conference on CFD in the Minerals and Process

Industries, Melbourne, Australia, 1999, pp. 503508.

[315] A.A. Feiz, M. Ould-Rouis, G. Lauriat, Large eddy simulation of turbulent ow in a rotating pipe, Int. J. Heat Fluid Flow 24 (2003) 412420.

[316] S.A. Jordan, The turbulent character and pressure loss produced by periodic symmetric ribs in a circular duct, Int. J. Heat Fluid Flow 24 (2003) 795

806.

[317] S. Vijiapuraru, J. Cui, Large eddy simulation of fully developed turbulent pipe ow, ASME Conf. Proc. (2004) 675679.

[318] S. Vijiapuraru, J. Cui, Performance of turbulence models for ows through rough pipes, App. Math. Modell. 34 (2010) 14581466.

[319] S.Y. Jung, Y.M. Chung, Large-eddy simulation of accelerated turbulent ow in a circular pipe, Int. J. Heat Fluid Flow 33 (2012) 18.

[320] D. Lakehal, LEIS for the prediction of turbulent multiuid ows applied to thermal-hydraulics applications, Nucl. Eng. Des. 240 (2010) 20962106.

[321] S. Reboux, P. Sagaut, D. Lakehal, Large-eddy simulation of sheared interfacial ow, Phys. Fluids 18 (2006) 105105.

[322] E.D. Christensen, Large eddy simulation of spilling and plunging breakers, Coast. Eng. 53 (2006) 463485.

[323] P. Lubin, S. Vincent, S. Abadie, J.-P. Caltagirone, Three-dimensional large eddy simulation of air entrainment under plunging breaking waves, Coast.

Eng. 53 (2006) 631655.

[324] D. Lakehal, P. Liovic, Turbulence structure and interaction with steep breaking waves, J. Fluid Mech. 674 (2011) 522577.

[325] P. Liovic, D. Lakehal, Multi-physics treatment in the vicinity of arbitrarily deformable gasliquid interfaces, J. Comput. Phys. 222 (2007) 504535.

[326] F.C.K. Ting, J.T. Kirby, Dynamics of surf-zone turbulence in a spilling breaker, Coast. Eng. 27 (1996) 131160.

[327] P.J. Roache, Quantication of uncertainty in computational uid dynamics, Annu. Rev. Fluid Mech. 29 (1997) 123160.

[328] W.L. Oberkampf, T.G. Trucano, Verication and validation in computational uid dynamics, Prog. Aerosp. Sci. 38 (2002) 209272.

[329] C.J. Roy, Review of code and solution verication procedures for computational simulation, J. Comput. Phys. 205 (2005) 131156.

[330] F. Stern, R. Wilson, J. Shao, Quantitative V&V of CFD simulations and certication of CFD codes, Int. J. Numer. Methods Fluids 50 (2006) 13351355.

[331] ASME, Standards for Verication and Validation in Computational Fluid Dynamics and Heat Transfer, The American Society of Mechanical Engineers

(ASME), 2009.

[332] NEA, Best practice guidelines for the use of CFD in nuclear reactor safety applications, Nuclear Energy Agency (NEA), Committee on the safety of

nuclear installations, NEA/CSNI/R(2007) 5, 2007.

[333] ERCOFTAC, Best Practice Guidelines, European Research Community on Flow, Turbulence and Combustion (ERCOFTAC), Special Interest Group on

Quality and Trust in Industrial CFD, Ver. 1, 2000.

[334] AIAA, Guide for the Verication and Validation of Computational Fluid Dynamics Simulations, AIAA Guide G-077-1988, 1998.

[335] M.K. Patel, M. Cross, N.C. Markatos, An assessment of ow-oriented schemes for reducing false diffusion, Int. J. Numer. Methods Eng. 26 (1988) 2279

2304.

[336] M.K. Patel, N.C. Markatos, M. Cross, Methods of reducing false-diffusion errors in convection-diffusion problems, Appl. Math. Modell. 9 (1985) 302

306.

[337] M.K. Patel, N.C. Markatos, An evaluation of eight discretization schemes for two-dimensional convection-diffusion equations, Int. J. Numer. Methods

Fluids 6 (1986) 103112.

[338] M.K. Patel, M. Cross, A.C.H. Mace, N.C. Markatos, An evaluation of eleven discretization schemes for predicting elliptic ow and heat transfer in

supersonic jets, Int. J. Heat Mass Transfer 30 (1987) 19071925.

[339] D.P. Karadimou, N.C. Markatos, A novel ow-oriented discretization scheme for reducing false-diffusion in three-dimensional ows: an application in

the indoor environment, Atmos. Environ. 61 (2012) 327339.

- 07A - Chapter 7 - Sec 7.1-7.3,App F - BlackUploaded byStevina Langrimari
- Adv Turb v62 01 OverviewUploaded byStefano Capra
- Torque 2010 Presentation Modelling Issues with Wind Turbine Wake and Atmospheric TurbulenceUploaded bypierre_elouan
- Numerical Simulation of the Ultrasonic Cavitating AtomizerUploaded byvedanth_s
- Hidraulica de Tuberias j p TelliusUploaded byDanny Garcia
- Lecture Seeding Particles for PIVUploaded byKaffel
- 16294Uploaded bySakthivel Arumugam
- PCFD 10(5-6) Paper 13Uploaded bydatam38
- MFIXEquations2005-4-4Uploaded bySrinivas Gowrishetty
- Star 0520Uploaded byShrigovind Tiwari
- D. Kivotides et al- Quantum Signature of Superfluid TurbulenceUploaded byJuaxmaw
- NoneUploaded byandhika_pw
- Mesolayer in turbulent pipe and channel flows -Uploaded bynoorafzal786
- 9-3-D numerical simulations of cylindrical pleated filter pa.pdfUploaded byUmanath R Poojary
- Basic Hyd Dyn EquationsUploaded bysivajirao70
- Distributor Design and TestingUploaded byjokish
- Presentation InveloxUploaded byPramodPradhan
- Numerical Simulation and Comparison With Experiment of Natural Convection Between Two Floors of a Building Model via a StairwellUploaded byVicky Rajasekharan
- Theoretics Study of The Effect of Suction and Blowing on TurbulenceUploaded byInnovative Research Publications
- Gutheil - Turbulent Spray Combustion Modeling For Rocket Engine ApplicationsUploaded byRUHAN PONCE
- Fm Lesson Plan ImprovedUploaded byPragya Shahdeo
- Experimental study of rotating dry slag granulation unit: Operating regimes, particle size analysis and scale upUploaded byKali Prasad
- Technical Challenges With the Small Scale LNG TerminalsUploaded byJosé Nunes
- NomenclatureUploaded byRochdi Demagh
- 154854_ORUploaded byDennis Angelo Pablico
- Methods for VortexUploaded byQuike Flores Suárez
- thesis.pdfUploaded bymuhammad numan iqbal
- 193028939-MATERI-7-Kehilangan-Energi.pptUploaded bybaizubiraji
- Modeling the dispersion of viable and total Escherichia coli cells in the artiﬁcial semi-enclosed bathing area of Santa Marinella (Latium, Italy)Uploaded byPriscila Vargas Babilonia
- Heat Transfer Enhancement in Solid Cylindrical & Cylindrical With Perforated Fins in Staggered & in Inline ArrangementUploaded byMANTECH Publications

- Hemophilia VWDUploaded byJhvhjgj Jhhgty
- ElectroplatingUploaded byvibhorsoni123
- bias in the workforceUploaded byapi-298895368
- Acute Exacerbation of Copd by Dr Irappa MadabhaviUploaded byIrappa Madabhavi
- Dunne v. Massachusetts Board of Bar Examiners et al - Document No. 5Uploaded byJustia.com
- Judd 1962Uploaded byRaquel Gamez
- Troliu Rugged Ridge Heavy Duty 8500 LBS 12V 3856 KG Cu Cablu Sintetic SK 75 Si Set de Accesorii Fisa TehnicaUploaded byNanescu Liliana
- 240782866-21st-Century-Oboe.pdfUploaded byFernando
- PONZI SCHEMES.pptxUploaded byshah_gen89
- AIPMT Solved Paper Retest July 2015Uploaded byNickOoPandey
- Support for Hierarchical Scheduling in FreeRTOSUploaded byAviral Mittal
- research paper(2).docxUploaded byMahesh Shende
- Building EssaysUploaded byTeuku Muktasim
- DepressionUploaded bySpoodie
- Baseline Document OutlineUploaded byOlsjon Baxhija
- UML for DesignersUploaded bylaxmankurnool
- Applied Calculus for Business StudentsUploaded byDr Srinivasan Nenmeli -K
- cogniciónUploaded byIsabel Henriquez Psicog
- A Psychotherapy of LoveUploaded bysvetlari
- 734-1175-1-PBUploaded byManoza Hevina
- Charge ListUploaded byKristine Browning
- Eat Chapter 2Uploaded byPG Chong
- DocumentUploaded byFehr Rivera
- Morgan Stanley ReportUploaded byricheshs
- Customer Sample Report Bsr for s4hanaUploaded byNitin
- Breaking the SilenceUploaded byRAM KRISHNA SINGH
- Unit3(2)Uploaded byDion Philip
- 152255182-Kenton-Knepper-Friends-Great-Minds-Think-Alike.pdfUploaded byKim Acosta
- Wind Induced Conductor Motion - EPRIUploaded byAnonymous 4FOiDbnWv
- gdbUploaded byViswanath Vellaiappan

## Much more than documents.

Discover everything Scribd has to offer, including books and audiobooks from major publishers.

Cancel anytime.