This action might not be possible to undo. Are you sure you want to continue?
Turbulence Modeling for CFD
by
David C. Wilcox
DCW Industries, Inc. La Canada, California
Dedicated to my Wife
BARBARA my Children KI~LEY and BOB
atul my Dad
Turbulence
Modeling for CFD
Copyright
© 1993, 1994 by DCW Industries, Inc. All rights reserved.
July, 1993 November, 1994 (with corrections)
First Printing: Second Printing:
No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from DCW Industries, Inc. DCW Industries, Inc. 5354 Palm Drive, La Canada, California 91011 818/7903844 (FAX) 818/9521272 This book was prepared with UTEX, and was printed in the United States of America by Griffin Printing, Glendale, California.
Library of Congress Cataloging in Publication Data
Wilcox, David C. Turbulence Modeling for CFD / David C. Wilcox1st Includes bibliography, index and 3! inch floppy disk. 1. TurbulenceMathematical Models. 2. Fluid DynamicsMathematical Models. TA357.5.T87 W542 1993 ISBN 0963605100
ed.
93224752
About the Author
Dr. David C. Wilcox, was born in Wilmington, Delaware. He did his undergraduate studies from 1963 to 1966 at the Massachusetts Institute of Technology, graduating with a Bachelor of Science degree in Aeronautics and Astronautics. From 1966 to 1967, he was an Engineer Scientist Specialist at McDonnell Douglas Aircraft Division in Long Beach, California, working for A. M. O. Smith. His experience with McDonnell Douglas was primarily in subsonic and transonic flow calculations. From 1967 to 1970, he attended the California Institute of Technology, graduating with a Ph.D. in Aeronautics. In 1970 he joined TRW Systems, Inc. in Redondo Beach, California, as a Member of the Technical Staff. He performed studies of both high and lowspeed fluidmechanical and heattransfer problems, such as turbulent hypersonic flow and thermal radiation from a flame. From 1972 to 1973, he was a staff scientist for Applied Theory, Inc., in Los Angeles, California, responsible for scientificproject management. He participated directly in many research efforts involving numerical computation and analysis of a wide range of fluid flows such as separated turbulent flow, transitional flow and hypersonic plumebody interaction. In 1973, he founded DCW Industries, Inc., a La Canada, California firm engaged in engineering research, software development and publishing, for which he is currently the President. He has taught several fluid mechanics and applied mathematics courses at the University of Southern California and at the University of California, Los Angeles. Dr. Wilcox has published many papers and reports on turbulence modeling, computational fluid dynamics, boundarylayer separation, boundarylayer transition, thermal radiation, and rapidly rotating fluids. He is an Associate Fellow of the American Institute of Aeronautics and Astronautics (AIAA) and has served as an Associate Editor for the AIAA Journal.
III
Contents
Notation Preface
1 Introduction
Xl XVll
. ..
1
1.1 1.2 1.3 1.4 2 The 2.1 2.2 2.3 2.4
Definition of an Ideal Turbulence Model . . . . . . . . How Complex Does a Turbulence Model Have to Be? . Comments on the Physics of Turbulence A Brief History of Turbulence Modeling . . . . . . . . Closure Problem Reynolds Averaging . . . . . . Correlations . ReynoldsAveraged Equations . The ReynoldsStress Equation .
1 1 2 5 11
11
15 15
17
23
3 Algebraic Models 3.1 Molecular Transport of Momentum. 3.2 The MixingLength Hypothesis . 3.3 Application to Free Shear Flows 3.3.1 The Far Wake .. 3.3.2 The Mixing Layer . 3.3.3 The Jet . 3.4 Modern Variants of the MixingLength Model. 3.4.1 CebeciSrnith Model . 3.4.2 BaldwinLomax Model . 3.5 Application to WallBounded Flows 3.5.1 Channel and Pipe Flow 3.5.2 Boundary Layers 3.6 Separated Flows ... .. v
24 27
30
32 38 41 44
50
53 53
52 59
61
VI
CONTENTS
3.7 3.8
The 1/2Equation Model. Range of Applicability ..
65 67 73 74 77 83 84 87 90 92 95 104 105 110 122 126 126 128 131 131 132 133 138 138 146 160 163
4 Turbulence Energy Equation Models 4.1 The Turbulence Energy Equation . 4.2 OneEquation Models 4.3 TwoEquation Models 4.3.1 The kt» Model 4.3.2 The k( Model 4.3.3 Other TwoEquation Models 4.4 Closure Coefficients . 4.5 Application to Free Shear Flows .. 4.6 Perturbation Analysis of the Boundary Layer 4.6.1 The Log Layer .... 4.6.2 The Defect Layer . . . 4.6.3 The Viscous Sublayer 4.7 Surface Boundary Conditions 4.7.1 Wall Functions .... 4.7.2 Surface Roughness . . 4.7.3 Surface Mass Injection . 4.8 Application to WallBounded Flows 4.8.1 Channel and Pipe Flow 4.8.2 Boundary Layers . . . . 4.9 LowReynoldsNumber Effects. 4.9.1 Asymptotic Consistency 4.9.2 Transition . 4.10 Separated Flows . 4.11 Range of Applicability 5 Effects of Compressibility 5.1 Physical Considerations 5.2 Favre Averaging .... 5.3 FavreAveraged Equations 5.4 CompressibleFlow Closure Approximations 5.5 Dilatation Dissipation . 5.6 Compressible Law of the Wall . 5.7 Compressible Boundary Layers . . . . . . . 5.8 ShockInduced BoundaryLayer Separation
171
171 172 174
180
183 189 195 203
CONTENTS 6 Beyond the Boussinesq Approximation 6.1 BoussinesqApproximation Deficiencies. 6.2 Nonlinear Constitutive Relations 6.3 SecondOrder Closure Models ... 6.3.1 Closure Approximations .. 6.3.2 LaunderReeceRodi Model 6.3.3 Wilcox Multiscale Model 6.4 Application to Homogeneous Turbulent Flows . 6.5 Application to Free Shear Flows .. 6.6 Application to WallBounded Flows 6.6.1 Surface Boundary Conditions 6.6.2 Channel and Pipe Flow 6.6.3 Boundary Layers .... 6.7 Application to Separated Flows 6.8 Range of Applicability ..... 7 Numerical Considerations 7.1 Multiple Time Scales and Stiffness 7.2 Numerical Accuracy Near Boundaries 7.2.1 Solid Surfaces . 7.2.2 Turbulent IN ontur bulent Interfaces 7.3 Parabolic Marching Methods .... 7.4 Elementary TimeMarching Methods . . . 7.5 BlockImplicit Methods ... ....... 7.6 Solution Convergence and Grid Sensitivity . 8 New Horizons 8.1 Background Information 8.2 Direct Numerical Simulation 8.3 Large Eddy Simulation. 8.4 Chaos . A Cartesian Tensor Analysis B Rudiments of Perturbation Methods C Companion Software C.1 Overview . C.l.l Program Structure C.1.2 Program Input . C.l.3 Program Output C.2 Free Shear Flows ....
Vll
213 213 218 223 224 231 232 235 242 243 244 248 253 261 266 273 273 275 275 279 287 292 297 303 313 313 316 322 328 331 337
349
349 350 351 352 353
Vlll
CONTENTS Program WAKE: Far Wake . . . . . . . . . . Program MIXER: Mixing Layer Program JET: Plane, Round and Radial Jet. Program PLOTF: Plotting Utility Channel and Pipe Flow C.3.1 Program PIPE: Channel and Pipe Flow C.3.2 Program PLOTP: Plotting Utility .. BoundaryLayer Perturbation Analysis. . . . C.4.l Program SUBLAY: Viscous Sublayer . C.4.2 Program DEFECT: Defect Layer . . . C.4.3 Program PLOTS: Sublayer Plotting Utility C.4.4 Program PLOTD: DefectLayer Plotting Utility. Miscellaneous Routines. . . . . . . . . . . . . . . C.5.1 Function ERF: Error Function .. . . . . . . C.5.2 Subroutine NAMSYS: Fortran Portability . . C.5.3 Subroutine RKGS: RungeKutta Integration C.5.4 Subroutine RTNI: Newton's Iterations . . . . C.5.5 Subroutine TRI: Tridiagonal Matrix Inversion. Diskette Contents. . . . . . . . . . . . . . . . . . . . .
C.2.1 C.2.2 C.2.3 C.2.4
C.3
C.4
C.5
C.6
355 357 359 361 364 365 367 371 373 375 376 379 382 383 384 386 388 389 390 391 391 391 392 392 397 397 398 398 399 399 400 400 402 403 404 406 406 408 408 410 410
D Program EDDYBL D.1 Overview 0.1.1 Acknowledgments D.1.2 Required Hardware and Software D.2 Getting Started Quickly . . . . . D.3 Installing SETEBL . . . . . . . . . . D.3.l BootConsole Installation . . D.3.2 RemoteTerminal Installation D.4 Installing EDDYBL ..... D.5 Running a General Case . . . 0.5.1 Preliminary Operations D.5.2 Units Selection . . . . D.5.3 Main Parameters . . . D.5.4 Taking a Lunch Break D.5.5 Edge/Wall Conditions D.5.6 Preparing Edge/Wall Condition Data Files D.5.7 Generating Edge/Wall Conditions D.5.8 Initial Profiles. . . . . . . . . . . . . . . . D.5.9 Selecting a Turbulence Model . . . . . . . D.5.10 Logical Unit Numbers and Plotting Files D.5.11 Running the BoundaryLayer Program. D.5.12 Restart Run. . . . . . . . . . . . . . . . .
CONTENTS D.5.13 Gas Properties and Profile Printing D.5.14 Selecting Laminar, Transitional or Thrbulent Flow D.B Applicability and Limitations . . . . D.7 EDDYBL Output Parameters. ..... D.S Program PLOTEB: Plotting Utility .. D.9 Adapting to Other Compilers/Systems. D.10 Compile and Link Commands. . D.11 Additional Technical Information . . . . D.ll.1 MeanFlow Equations . . . . . . D.l1.2 kw and Multiscale Model Equations D.11.3 ke Model Equations . . D.11.4 'Transformed Equations D .12 Software Package Modules . E Plotting Program Details E.1 Font Files . . . E.2 Video Devices . . . E.3 Plotting Colors . . EA Hardcopy Devices. Bibliography Index
~ 411 412 412 413 415 419 420 423 423 424 426 429 432 433 433 433 435 435 437 456
Dn A+ o Aij bij B Cbl. English Symbols Symbol a aijkl Definition Speed of sound Rapid pressurestrain tensor Coefficients in tridiagonal matrix equation Van Driest damping constant Slow pressurestrain tensor Dimensionless Reynoldsstress anisotropy tensor Additive constant in the law of the wall Closure coefficients Skin friction based on edge velocity. CK CLl! CL2 CKleb Cp c. Tw/(tpU. Cb2 CJ CJoo Cwb Cw2.) Skin friction based on freestream velocity. pressure coefficient Closure coefficients Smagorinsky constant Specific heat at constant volume Shearlayer spreading rate An. G2 GcP GWk ! GD. a few symbols denote more than one quantity.Gn. c.GE CdiJ.. Cw3 t Gl. Cs Cv Cf! Xl .Notation This section includes the most commonly used notation in this book. In order to avoid departing too much from conventions normally used in literature on turbulence modeling and general fluid mechanics. Tw / ( pU~J Closure coefficients Closure coefficients Closure coefficients Closure coefficients Closure coefficients Kolmogorov constant Closure coefficients Specific heat at constant pressure.Bn.
III J J k = = kg kR K K()Kw Kn f fmJp fmiz K(11) L Lij M Mijkl Me . j. Cr2 NOTATION Closure coefficients Closure coefficients Closure coefficient LES crossterm stress tensor Turbulent transport tensor Drag per unit body width Production tensor. 6*/0 Heaviside step function Unit vectors in z . shape factor. C(3 Crt. axisymmetric (j 1) index Specific momentum flux (flux per unit mass) Kinetic energy of turbulent fluctuations per unit mass Geometric progression ratio Surface roughness height Distortion parameter Dimensionless selfsimilar turbulence kinetic energy Effective Karman constant for compressible flows Knudsen number Turbulence length scale.xu Cd. fv IfJ' It. smalleddy energy Total energy. channel height.e) h H(x) i. e E E(K) E(11) f. z directions Unit (identity) matrix Stress tensor invariants Twodimensional (j 0). Is F(11) FKleb(Y. characteristic eddy size Mean free path Mixing length Characteristic length scale Leonard stress tensor Mach number Rapid pressurestrain tensor Convective Mach number CfJ Gijk D Dij c.. viscous damping function Energy spectral density Dimensionless selfsimilar dissipation rate Discretization error Viscous damping functions Turbulence flux vectors Dimensionless selfsimilar streamfunction Klebanoff intermittency function Meanflow flux vectors Amplitude factor in von Neumann stability analysis LES filter Specific enthalpy Total enthalpy. G H r. G(x . 12. k I II. 6) Eh F. TimOUm/OXj + TjmOUm/OXi. C(2. y. Specific internal energy .
k1/2y/V Instantaneous strainrate tensor Sourceterm vectors Source term . PrL. Mto N(1J) NCFL Nw N( 'Ui) P Pij P Pij Pk. 'Rij (x. 'UrR/v Closure coefficients in viscous damping functions Reynolds number based on length L Turbulence Reynolds number. e Laminar.t. t Sij SUI SR SB s. Sk. ReL ReT t.production minus dissipation Mean strainrate tensor s. 0. Tim8Uj /8xm + Tjm8Ui! 8xm Net production per unit dissipation of k.s S Sij o 8ij Turbulence Mach number. Rk. Gij + ~j Dependent variable vector Cylindrical polar coordinates Pipe radius. R+ Turbulence Richardson number Nearwall turbulence Reynolds number. 12k/a Closure coefficient Dimensionless selfsimilar eddy viscosity CFL number Constant in nearwall solution for w NavierStokes oper ator Instantaneous static pressure Instantaneous momentumflux tensor Mean static pressure Production tensor.. characteristic time scale . w. perfect gas constant SGS Reynolds stress tensor Twopoint velocity correlation tensor Radius of curvature Autocorrelation tensor Sublayer scaled radius or half height. Hij R+ Rf3. x R Hij(x.NOTATION Xlll u. turbulent mean heatflux vector LES stress tensor. turbulent Prandtl number Heatflux vector Surface heat flux Laminar. t') s.r) 1(. Pw. tij T Oldroyd derivative of Sij Source terms in a similarity solution Dimensionless surface mass injection function Dimensionless surface roughness function Time Instantaneous viscous stress tensor Temperature. ReT RiT fly s. channel half height. k1/2l/v Sublayer scaled radius or half height. PrT qw qLj' Qij qTj Q r. qj r.
W Ui U U' t . z directions Fluctuating velocity in tensor notation Fluctuating velocity in vector notation Favreaveraged velocity components in X. y. distance. W" Ui U U" V" U~I I . Vrw/ Pw Velocity perturbation vector Mean velocity components in z . y. y. V. sublayerscaled. U" Urm8' U~UI. W  . W' U~ U' U. U [u: Maximum or centerline velocity Dimensionless selfsimilar streamwise velocity Mixing velocity Thermal velocity Surface injection velocity Dimensionless selfsimilar normal velocity Dimensionless selfsimilar specific dissipation rate Rectangular Cartesian coordinates Position vector in tensor notation Position vector in vector notation Dimensionless. sublayerscaled. y directions Temporal average of fluctuating velocities Friction velocity. z directions Instantaneous velocity in tensor notation Instantaneous velocity in vector notation Fluctuating velocity components in x. velocity.XIV NOTATION Freestream turbulence intensity Instantaneous velocity components in X. a* Definition Closure coefficients Closure coefficients Closure coefficients in viscous damping functions . y. Z directions Favreaveraged velocity in tensor notation Favreaveraged velocity in vector notation Favre fluctuating velocity components in x. W u. Xi X y. U U+ Urn U(TJ) Vth Vw V(TJ) W(TJ) x. ury/v y+ at first grid point above surface Inner/outer layer matching point T' U. z directions Mean velocity in tensor notation Mean velocity in vector notation Dimensionless. Z y+ yt Ym Greek Symbols Symbol 0: . V' . z directions Favre fluctuating velocity in tensor notation Favre fluctuating velocity. fluctuating molecular velocity RMS fluctuating velocity components in X. y. V. V. tJ 'Ur Vrm8 ii U.
. . (1 dy Finitedifference matrix operator Kronecker delta LES filter width Clauser thickness.X .6./c.6.6.e 11' ~ TIij P U .6. Q.y . V ur. UT' UTI. Boundary layer or shear layer thickness xv c.6.t { (d {8 (ij {ijk ( fJ () Velocity thickness. y Timestep Dissipation per unit mass Dilatation dissipation Solenoidal dissipation Dissipation tensor Permutation tensor Second viscosity coefficient Kolmogorov length scale. ILl p Kinematic eddy viscosity. thermal conductivity.( X) . Ue6* IUT Incremental change in Q. similarity variable Momentum thickness.s (1  . 6* 6* v 6x fJij Displacement thickness.6. . UL1. x. wavenumber Effective Karman constant for flows with mass injection Taylor microscale Largest eigenvalue Molecular viscosity Eddy viscosity Innerlayer eddy viscosity Outerlayer eddy viscosity Kinematic molecular viscosity. fo. (1 dy Karman constant. U* a. VT ~*. UL2 U'T2 Uk. . (fJ* ITw)dPldx Specific heat ratio.NOTATION Defectlayer similarity parameters Closure coefficients Equilibrium parameter. PT Ip Dimensionless streamwise distance Closure coefficients Coles' wakestrength parameter Pressurestrain correlation tensor Mass density Closure coefficients Closure coefficients Closure coefficients Closure coefficients f: fIe) t iJe) dy f: t ife ife) Kv A Amax IL JlT ur.
Tyy. turbulence dissipation time Reynolds stress tensor Eddy turnover time Reynolds shear stress Normal Reynolds stresses Surface shear stress Kolmogorov velocity scale.j TturnotJer Txy Txx. Tzz Tw V Other Symbol 8f/8q 8F/8Q os/oq Definition Turbulence fluxJacobian matrix Meanflow fluxJ acobian matrix SourceJacobian matrix Subscripts Symbol DNS e eq LES o v w 00 Definition Direct Numerical Simulation Boundarylayeredge value Equilibrium value Large Eddy Simulation Centerline value Viscous Wall (surface) value Freestream value Superscripts Symbol + Definition Sublayerscaled value . closure coefficient Dimensionless parameter. (vw / pu~)dP / dz Free shear layer closure coefficient Streamfunction Parabolic marching scheme coefficients Specific dissipation rate.XVI NOTATION Nonequilibrium parameter Instantaneous total stress tensor Kolmogorov time scale. vorticity vector magnitude O"(x) O"ij r Ti.
for indeed many illconceived and illfated turbulence models have appeared in engineering journals. This is not to say all turbulence modeling has been done in such a manner. However. early in his career. Even this author. turbulence modeling is regarded by many CFD researchers as "black magic. but with less detail. systematic construction of a wellfounded turbulence model is not only possible but can be an exciting and challenging research project. A secondary goal of this book is to provide a rational way for deciding how complex a model is needed for a given problem." lacking in rigor and physical foundation.Preface This book has been developed from the author's lecture notes used in presenting a postgraduate course on turbulence modeling at the University of Southern California. The engineer who feels no existing turbulence model is suitable for his or her needs and wishes to modify an existing model or to devise a new model will benefit from this feature of the text. A methodology is presented in Chapters 3 and 4 for devising and testing such equations. devised a turbulence model that violated Galilean invariance of the timeaveraged NavierStokes equations! However. This book has been written to show that turbulence modeling can be done in a systematic and physically sound manner. with judicious use of relatively simple mathematical tools. The engineer who wishes to select an existing model that is sufficient for his or her needs will benefit most from this feature of the text. As a consequence. The methodology is illustrated in great detail for twoequation turbulence models. While several computational fluid dynamics (CFD) texts include some information about turbulence modeling. the primary goal of this book is to provide a systematic approach to developing a set of constitutive equations suitable for computation of turbulent flows. Thus. Chapter 3 begins with the simplest turbulence models and subsequent chapters chart a course leading to some of the most complex models that have been applied to a nontrivial XVlJ . it is by no means limited to such models and is used again in Chapter 6 for a full Reynoldsstress model. very few texts dealing exclusively with turbulence modeling have been written.
Redekopp of USC for encouraging and supporting development of the course for which this book is intended. reviewed the entire manuscript as I . standard numerical procedures just won't work and alternative methods must be found to obtain accurate converged solutions. First. and numerical procedures. Often. Chapter 7 focuses on numerical methods and elucidates some of the commonly encountered problems such as stiffness. the text would be incomplete without companion software implementing numerical solutions to standard turbulence model equations. Dr. first or second year graduate course. which is done extensively in Chapter 2. The material presented in this book is appropriate for a onesemester. jet and far wake. L. Appendix A introduces rudiments of tensor. Bradshaw. Because turbulence modeling is a key ingredient in CFn work. A friend of many years. preferably in FORTRAN. Successful study of this material requires an understanding of viscousflow and boundarylayer theory. No book on turbulencemodel equations is complete without a discussion of numerical solution methods. Some degree of proficiency in solving partial differential equations is also needed. analysis to facilitate manipulation of the NavierStokes equation. 5 and 6 use the methods to dissect modelpredicted features of the turbulent boundary layer. largeeddy simulation (LES) and the interesting mathematical theory of chaos. Chapter 3 shows. singular perturbation methods. Anyone who has ever tried to obtain a numerical solution to a set of turbulence transport equations can attest to this. or as a reference text for a CFD course. Two things are done at each level of complexity. The methodology makes extensive use of tensor analysis. The programs all have a similar structure and can be easily modified to include new turbulence models. P. The text assumes the user has limited prior knowledge of these mathematical concepts and provides what is needed both in the main text and in the Appendices. Similarity solutions are then obtained for the turbulent mixing layer. similarity solutions. the range of applicability of the model is estimated. G. will help the reader gain maximum benefit from the companion software described in the Appendices. Appendices C and D describe several computer programs that are included on the floppy disk accompanying the book. I extend my thanks to Dr. in detail. Appendix B presents elements of singular perturbation theory. many of the applications are repeated for all of the models to illustrate how accuracy changes with complexity. The concluding chapter presents a brief overview of new horizons including direct numerical simulation (DNS). sharp turbulentnonturbulent interfaces. For example. A knowledge of computer programming. the way a similarity solution is generated. Second. and difficulties attending turbulence related time scales.XVlll PREFACE turbulent flow problem. Chapters 4.
form. Their patience was especially noteworthy. reviewed the manuscript from cover to cover and offered a great deal of physical and computational insight in the process. G. made sure I omitted the dot over every /. P. Lin. Leu. M. F. C. Finally. and taught me a lot through numerous discussions. Foley. I thank the nine students who were the first to take the course that this book was written for. D. Knight. Wallace. lowe a lifelong debt to my loving wife Barbara for tolerating the hectic pace first in college and then in the business world. Dr.S. So. this book would not have been possible. and crossed every z in Appendix B. Kale. My favorite mathematics teacher. Cohen. Menter and C. comments and suggestions that greatly improved the final draft. Thanks are also due for the support and help of several friends and colleagues. Dr. C. Wilcox . D. Taniguchi and D. Holbrook. Without her. David C. R. H. T. R. Horstman were kind enough to provide results of several of their computations in digital . T. S. Another long time friend. particularly in regard to typographical errors in the homework problems! That outstanding group of young engineers is D. Magee. C. Drs. helped me understand why I had to write this book. S.PREFACE XIX wrote it. D. N. J. most notably Drs. Roache. T. P. Speziale and R. Tadepalli.
Very precise mathematical theories have evolved for the other two key elements. This description of an ideal model serves as the main keystone of this text.. 1. By its nature .1 Definition of an Ideal Turbulence Model Turbulence modeling is one of three key elements in Computational Fluid Dynamics (CFD). Fortunately. an ideal model should introduce the minimum amount of complexity while capturing the essence of the relevant physics.2 How Complex Does a Turbulence Model Have to Be? Aside from any physical considerations. turbulence is inherently three dimensional and time dependent. to some extent. Simplicity combined with physical insight seems to have been a common denominator of the work of these great men. a throwback to the days of Prandtl.Chapter 1 Introduction 1. The field is.in creating a mathematical model that approximates the physical behavior of turbulent flows ~ far less precision has been achieved in turbulence modeling. Thus. Taylor. an enormous amount of information is required to completely describe a turbulent flow. Using their work as a gauge. grid generation and algorithm development. von Karman and all the many other clever engineers who spent a good portion of their time devising engineering approximations and models describing complicated physical flows. This is not really a surprising event since our objective has been to approximate an extremely complicated phenomenon. we usually 1 . viz.
how do we predict the physically meaningful properties of the flow? What properties of a given flow are meaningful is generally dictated by the application. Thus. we may require only the skin friction and heattransfer coefficients. refer to a basic text on the physics of turbulence such as those by Tennekes and Lumley (1983) or Landahl and MolloChristensen (1992). almost always manifest themselves as numerical difficulties! 1.2 CHAPTER 1. On the one hand. not to mention vast computer resources. Overkill is often accompanied by unexpected difficulties that. Most engineering problems fall somewhere between these two extremes. the conscientious engineer will strive to use as conceptually simple an approach as possible to achieve his ends. if all we require is skin friction for an attached flow. Rather. once the question of how much detail we need is answered. turbulence fluctuation magnitudes and scales. Given a set of initial and/or boundary conditions. Such models are well developed and can be implemented with very little specialized knowledge. if we desire a complete time history of every aspect of a turbulent flow. For the simplest applications. it is worthwhile to first discuss physical aspects of the phenomenon. in CFD applications.3 Comments on the Physics of Turbulence Before plunging into the mathematics of turbulence. Taylor and von Karman. The following discussion is not intended as a complete description of this complex topic. and in construction of a mathematical model. Such a solution requires an extremely accurate numerical solver and may require use of subtle transform techniques. Thus. qualitatively speaking. we focus upon a few features of interest in engineering applications. INTRODUCTION require something less than a complete time history over all spatial coordinates for every flow property. In the spirit of Prandtl. the level of complexity of the model follows. More esoteric applications may require detailed knowledge of energy spectra. for a given turbulentflow application. a simple mixinglength model (Chapter 3) may suffice. Taylor and von Karman [see Goldstein (1938)] proposed the following definition of turbulence: "Turbulence is an irregular motion which in general makes its appearance in fluids. only a solution to the complete NavierStokes equation will suffice. In 1937. gaseous or liquid. On the other hand. For a morecomplete introduction. we should expect the complexity of the mathematics needed for a given application to increase as the amount of required flowfield detail increases. when they flow past solid surfaces or even when neighboring streams of the same fluid flow . we must pose the following question. Certainly.
large Reynolds number corresponds to anything stronger than a small swirl or a puff of wind. Vortex stretching is absent in twodimensional flows so that turbulence must be three dimensional. In principle. let alone the fullyturbulent state. Although some degree of success can be achieved in predicting the onset of instabilities that ultimately lead to turbulence with linear theories. threedimensional N avierStokes equation contains all of the physics of a given turbulent flow. On the other hand.1." It is characterized by the presence of a large range of excited length and time scales.e. On the one hand.. which is the primary focus of this text. this aspect is not really a problem from the engineer's view. Even if we had a complete time history of a turbulent flow. The interaction is very complex because it is rotational. COMMENTS ON THE PHYSICS OF TURBULENCE 3 past or over one another. we would usually integrate the flow properties of interest over time to extract timeaverages.3. viscous) fluid. fully three dimensional and time dependent. This inherent three dimensionality means there are no satisfactory twodimensional approximations and this is one of the reasons turbulence remains the most noteworthy unsolved scientific problem of the twentieth century. the inherent nonlinearity of the NavierStokes equation precludes a complete analytical description of the actual transition process. we must use statistical methods. time averaging operations lead to statistical correlations in the equations of motion that cannot be determined a priori. The strongly rotational nature of turbulence goes handinhand with its three dimensionality. because the fluid was imagined to flow in smooth laminae. virtually all methods begin by linearizing the equations of motion. For slightly viscous fluids such as water and air. or layers. This is the classical closure problem. As noted . That this is true follows from the fact that turbulence is a continuum phenomenon. Virtually all flows of practical engineering interest are turbulent. the instabilities result from interaction between the N avierStokes equation's nonlinear inertial terms and viscous terms. Turbulent flows always occur when the Reynolds number is large. Vigorous stretching of vortex lines is required to maintain the everpresent fluctuating vorticity in a turbulent flow. so called historically. show that turbulence develops as an instability of laminar flow. The additional complexity goes beyond the introduction of an additional dimension. the timedependent. For a real (i. Thrbulence is characterized by random fluctuations thus obviating a deterministic approach to the problem. The timedependent nature of turbulence also contributes to its intractability. To analyze the stability of laminar flows. or more typically to its boundarylayer form. Careful analysis of solutions to the NavierStokes equation. The irregular nature of turbulence stands in contrast to laminar motion. Rather.
the ratio of smallest to largest scales decreases rapidly as the Reynolds number increases.000 would require a computer 10 million times faster than a Cray Y IMP. However. the smallest scales of turbulence are still extremely small. as opposed to a discrete set of scales. Apparent stresses often develop in turbulent flows that are several orders of magnitude larger than in corresponding laminar flows. Consequently. Perhaps the most important feature of turbulence from an engineering point of view is its enhanced diffusivity. In order to visualize a turbulent flow with a spectrum of scales we often refer to turbulent eddies. Thus. the smallest eddies dissipate into heat through the action of molecular viscosity. Ultimately.4 CHAPTERl. momentum and energy. While more and more progress is being made with such simulations. Speziale (1985) notes that a numerical simulation of turbulent pipe flow at a Reynolds number of 500. To make an accurate numerical simulation (i. Turbulence consists of a continuous spectrum of scales ranging from largest to smallest. as the turbulence decays. Furthermore. They are generally many orders of magnitude smaller than the largest scales of turbulence. its kinetic energy transfers from larger eddies to smaller eddies.e. large ones carrying smaller ones. The main physical process that spreads the motion over a wide range of wavelengths is vortex stretching.1). a full timedependent threedimensional solution) of a turbulent flow. "Even the smallest scales occurring in a turbulent flow are ordinarily far larger than any molecular length scale. we observe that turbulent flows are always dissipative. INTRODUCTION by Tennekes and Lumley (1983). Turbulence features a cascading process whereby. the results are very useful in developing and testing turbulence models in the limit of low Reynolds number.. The nonlinearity of the NavierStokes equation leads to interactions between fluctuations of differing wavelengths and directions. wavelengths that are not too small compared to the meanflow width interact most strongly with the mean flow. Most importantly." Nevertheless. the latter being of the same order of magnitude as the dimension of the object about which the fluid is flowing. the largerscale . Eddies overlap in space. As discussed above. The turbulence gains energy if the vortex elements are primarily oriented in a direction in which the mean velocity gradients can stretch them. all physically relevant scales must be resolved. computers of the early 1990's have insufficient memory and speed to solve any turbulent flow problem of practical interest. A turbulent eddy can be thought of as a local swirling motion whose characteristic dimension is the local turbulence scale (Figure 1. To underscore the magnitude of the problem. 'Turbulent diffusion greatly enhances the transfer of mass. the wavelengths of the motion usually extend all the way from a maximum comparable to the width of the flow to a minimum fixed by viscous dissipation of energy.
cascading energy to them.4. turbulent motion carries most of the energy and is mainly responsible for the enhanced diffusivity and attending stresses. The flow above the boundary layer has a steady velocity U. The origin of this approach dates back to the end of the nineteenth century when Reynolds (1895) published results of his research 011 turbulence. we will introduce more specific details of turbulence properties for common flows on an asneeded basis. In turn. carrying smallerscale disturbances with them. the larger eddies randomly stretch the vortex elements that comprise the smaller eddies. the eddies move at randomlyfluctuating velocities of the order of a tenth of U.e) is comparable to the boundarylayer thickness (b). Hence.1. The interface and the flow above the boundary is quite sharp (Corrsin and Kistler (1954)]. the turbulent stresses at a given position depend upon upstream history and cannot be uniquely specified in terms of the local strainrate tensor as in laminar flow. In addition to migrating across the flow. The arrival of these large eddies near the interface between the turbulent region and nonturbulent fluid distorts the interface into a highly convoluted shape (Figure 1. An especially striking feature of a turbulent shear flow is the way large bodies of fluid migrate across the flow. As we progress through the following chapters. 1. The largest eddy size (. His pioneering work proved to have such profound importance .imeaveraged NavierStokes equation. A BRIEF HISTORY OF TURBULENCE MODELING 5 u Figure 1.1: Large eddies in a turbulent boundary layer.1). they have a lifetime so long that they persist for distances as much as 30 times the width of the flow [Bradshaw (1972)].4 A Brief History of Turbulence Modeling The primary emphasis in this book is upon the t.
6 CHAPTER 1. upon flow history. Neither Reynolds nor Boussinesq attempted solution of the Reynoldsaveraged NavierStokes equation in any systematic manner. have proven to be of great value in many engineering applications. an nequation model signifies a model that requires solution of n additional differential transport equations in addition to those expressing conservation of mass. To elaborate a bit further. The Boussinesq eddyviscosity approximation is so widely known that few authors find a need to reference his original paper. and further progress awaited Prandtl's discovery of the boundary layer in 1904. we refer to a mixinglength model as an algebraic model or a zeroequation model of turbulence. While having an eddy viscosity that depends upon flow history provides a more physically realistic model. The mixinglength hypothesis. Important early contributions were made by several researchers. In modern terminology. Boussinesq has been immortalized in turbulence literature. formed the basis of virtually all turbulencemodeling research for the next twenty years. most notably by von Karman (1930). By definition. on a conceptual level. Focusing upon turbulent flows. Thus was born the concept of the socalled oneequation model of turbulence. closely related to the eddyviscosity concept. Boussinesq (1877) introduced the concept of an eddy viscosity. Such models are not without merit and. INTRODUCTION for all future developments that we refer to the standard timeaveraging process as one type of Reynolds averaging. As with Reynolds. Since the length scale can be thought of as a characteristic eddy size and since such scales are different for each flow. Prandtl (1945) postulated a model in which the eddy viscosity depends upon the kinetic energy of the turbulent fluctuations..e. takes account of the fact that the eddy viscosity is affected by where the flow has been. other than initial and boundary conditions) in advance in order to obtain a solution. That is) we must know something about the flow. i. k. In this spirit. Much of the physics of viscous flows was a mystery in the nineteenth century. The earliest attempts at developing a mathematical description of turbulent stresses sought to mimic the molecular gradientdiffusion process. He proposed a modeled differential equation approximating the exact equation for k. an incomplete model generally defines a . This improvement. To improve the ability to predict properties of turbulent flows and to develop a more realistic mathematical description of the turbulent stresses. turbulence models that do not provide a length scale are incomplete. Prandtl (1925) introduced the mixing length (an analog of the meanfree path of a gas) and a straightforward prescription for computing the eddy viscosity in terms of the mixing length. the need to specify a turbulence length scale remains. in fact. momentum and energy.
Yet another length might be needed for free shear flows. for an attached boundary layer. i. In addition to having a modeled equation for k. the displacement thickness.g. We define such a model as being complete. Such models automatically accommodate complicating effects such as streamline curvature." The reciprocal of w serves as a turbulence time scale. etc.1. a different length in this example would be needed when the boundary layer separates since 6* may be negative. For a threedimensional flow. by the early 1950's.. . and body forces.e.g. A particularly desirable type of turbulence model would be one that can be applied to a given turbulent flow by prescribing at most the appropriate boundary and/or initial conditions. Note that our definition implies nothing regarding the accuracy or universality of the model. In this model. secondorder closure models awaited adequate computer resources. one for the turbulence scale and six for the components of the Reynoldsstress tensor. The primary conceptual advantage of secondorder closure is the natural manner in which nonlocal and history effects are incorporated. Rotta (1951) laid the foundation for turbulence models that obviate use of the Boussinesq approximation. He devised a plausible model for the differential equation governing evolution of the tensor that represents the turbulent stresses. it went with virtually no applications for the next quarter century because of the unavailability of computers to solve its nonlinear differential equations.. However. incomplete models usually define quantities that may vary more simply or more slowly than the Reynolds stresses (e. rigidbody rotation.4. Thus. While this model offered great promise. a secondorder closure model introduces seven equations. viz. four main categories of turbulence models had evolved. Presumably such quantities are easier to correlate than the actual stresses. he introduced a second parameter w that he referred to as ''the rate of dissipation of energy in unit volume and time. This approach is called secondorder or secondmoment closure. This stands in contrast to eddyviscosity models that fail to properly account for these effects. The model is thus termed a twoequation model of turbulence. w satisfies a differential equation similar to the equation for k. Ideally. only that it can be used to determine a flow with no prior knowledge of any flow details. known as a kw model. the Reynoldsstress tensor. e. no advance knowledge of any property of the turbulence should be required to obtain a solution. 6*. eddy viscosity and mixing length). Kolmogorov (1942) introduced the first complete model of turbulence. As with Kolmogorov's kw model. In essence. while k1/2/w serves as the analog of the mixing length. A BRIEF HISTORY OF TURBULENCE MODELING 7 turbulence length in a prescribed manner from the mean flow.
Launder's kc model.f. SecondOrder Closure Models With the coming of the age of computers since the 1960's. Even the model's demonstrable inadequacy for flows with adverse pressure gradient [c. the BradshawFerrissAtwell model most faithfully reproduced measured flow properties. OneEquation Models. TwoEquation Models 4. Of the four types of turbulence models described above. Ferriss and Atwell (1967). Algebraic Models. With no prior knowledge of Kolmogorov's work. The following overview lists a few of the most important modern developments for each of the four classes. By far. the oneequation model has enjoyed the least popularity and success. is as well known as the mixinglength model and is the most widely used twoequation model. Goldberg (1991) and Spalart and Allmaras (1992)]' motivated primarily by the ease with which such model equations can be solved numerically. Baldwin and Barth (1990). There has been some renewed interest in oneequation models [c. INTRODUCTION Models 2. In the 1968 Stanford Conference on Computation of Turbulent Boundary Layers [Coles and Hirst (1969)] the best turbulence models of the day were tested against the best experimental data of the day. Saffman (1970) formulated a kw . it remained in obscurity until the coming of the computer. further development of all four classes of turbulence models has occurred.f. TwoEquation Models. Baldwin and Lomax (1978) have proposed an alternative algebraic model that enjoys widespread use. To remove some of the difficulties in defining the turbulence length scale from the shearlayer thickness. Van Driest (1956) devised a viscous damping correction for the mixinglength model that is included in virtually all algebraic models in use today. Perhaps the most successful model of this type was formulated by Bradshaw. relative to twoequation models and secondorder closure models. While Kolmogorov's kw model was the first of this type.8 1. In this author's opinion. Rodi and Scheuerer (1986) and Wilcox (1988a)] has done little to discourage its widespread use. Cebeci and Smith (1974) refined the eddyviscosity /mixingIength model to a point that it can be used with great confidence for most attached boundary layers. Algebraic (ZeroEquation) CHAPTER 1. of all the models used. where £ is proportional to the product of k and w. the most extensive work on twoequation models has been done by Launder and Spalding (1972) and a continuing succession of students and colleagues. OneEquation Models 3.
The primary emphasis is upon examining the underlying physical foundation and upon developing the mathematical tools for analyzing and testing the models. Wilcox and Rubesin (1980). The text is not intended to be a catalog of all turbulence models. Wilcox and Alber (1972). As pointed out by Lakshminarayana (1986).4. . Reece and Rodi (1975). Speziale (1985. The most noteworthy efforts were those of Donaldson [Donaldson and Rosenbaum (1968)]. This book investigates all four classes of turbulence models. The latter has become the baseline secondorder closure model: more recent contributions by Lumley (1978). Daly and Harlow (1970) and Launder. they have thus far found their way into a relatively small number of applications compared to algebraic and twoequation models. 1987a) and Reynolds (1987) have added mathematical rigor to the closure process. we approach each class of models in a generic sense. additionally. because of the large number of equations and complexity involved in secondorder closure models. Rather.1. A BRIEF HISTORY OF TURBULENCE MODELING 9 model that enjoys advantages over the ke model. for example. However. especially for integrating through the viscous sublayer and for predicting effects of adverse pressure gradient. and Wilcox (1988a). Saffman and Wilcox (1974). references are given for most models. SecondOrder Closure Models. By the 1970's. have pursued further development and application of kw models. ki» models are the second most widely used type of twoequation turbulence model. sufficient computer resources became available to permit serious development of this class of model. Wilcox and Traci (1976). Detailed information is provided for models that have stood the test of time.
i. As we will see in this chapter.e. we express an instantaneous flow variable as f(x. i. the spatial average and the enfor stationary turbulence. establishing a sufficient number of equations for all of the unknowns. we use a statistical approach. We then form the time average of the continuity and NavierStokes equations. a turbulent flow that. Our purposes are best served by using the procedure introduced by Reynolds (1895) in which all quantities are expressed as the sum of mean and fluctuating parts. Reynolds averaging assumes a variety of forms involving either an integral or a summation. is defined by semble average. We then derive equations for these stresses and the resulting equations include additional unknown quantities. In general. These momentum fluxes are unknown a priori.e.1 Reynolds Averaging We begin with the averaging concepts introduced by Reynolds (1895).. t) dt (2. on the average. FT(x). This illustrates the issue of closure.Chapter 2 The Closure Problem Because turbulence consists of random fluctuations of the various flow properties. Its time average. The three forms most pertinent in turbulencemodel research are the time average. the nonlinearity of the NavierStokes equation leads to the appearance of momentum fluxes that act as apparent stresses throughout the flow. 2. For such a flow. Time averaging is appropriate FT(x) = T+oo lim T 11t+T t f(x. t).. does not vary with time.1) 11 .
12 CHAPTER 2.. THE CLOSURE PROBLEM Spatial averaging is appropriate for homogeneous turbulence. the quantity velocity defined by Ui(X) is the timeaveraged. or mean. t). using Equation (2. in terms of measurements from N identical experiments where lex.4) As in Equation (2. The timeaveraging process is most clearly explained for stationary turbulence. t) . uHx. the average is FE.2) As an idealized example.5) The time average of the mean velocity is again the same timeaveraged value.Ui(x)] dt = Ui(x) . FE(X.e. and a fluctuating part.3) For turbulence that is both stationary and homogeneous. Ui(X). The time average of the fluctuating part of the velocity is zero. For such a flow. we have Fv (t) = lim V+CXl _!_ V Iff lex. we may assume that these three averages are all equal. we express the instantaneous velocity.7) While Equation (2. t) in the nth experiment. That is. Ui(X.5) is mathematically well defined. we can never truly realize infinite T in any physical flow. i. where Ensemble averaging is the most general type of averaging. u~ = lim T T+CXl 11t+T t [Ui(X. (2. which is a turbulent flow that. as the sum of a mean. We average over all spatial coordinates by doing a volume integral. t) n=l L (2. This is not a serious problem in . t) dV (2. is uniform in all directions. time averaging is the most appropriate form of Reynolds averaging. t) = !n(x. This assumption is known as the ergodic hypothesis. so that (2. t) = N+CXl lim 1N N '"'" !n(x.6). t). on the average. (2.6) where an overbar is shorthand for time average. Calling the average Fv . Because virtually all engineering problems involve inhomogeneous turbulence.Ui(X)  =0 (2.1).
1.8) and (2. In other words. Clearly. we might impose a slowly varying periodic pressure gradient in a duct or we might wish to compute flow over a helicopter blade. an integration time of 20 seconds would probably be adequate. There are some flows for which the mean flow contains very slow variations with time that are not turbulent in nature. TI.5) must be modified to accommodate such applications. We are implicitly assuming that time scales T'.4) and (2. For instance. rather than formally taking the limit T ~ 00. method is to replace Equations (2.2. we just select a time T that is very long relative to the maximum period of the velocity fluctuations. and T2 exist that differ by several .1: Time averaging for nonstationary turbulence. t Figure 2.5) with T » TI. In forming our time average. for flow at 10 m/sec in a 5 em diameter pipe. REYNOLDS AVERAGING 13 practice however.1 illustrates these concepts. In this time the flow moves 4. A word of caution is in order regarding Equation (2. Figure 2. As an example.9). The simplest. we do the indicated integration in Equation (2.000 pipe diameters.5) with (2. Equations (2.4) and (2.9) where T2 is the time scale characteristic of the slow variations in the flow that we do not wish to regard as belonging to the turbulence. but a bit more arbitrary.
9) for such flows because there is no distinct boundary between our imposed unsteadiness and turbulent fluctuations. The notion is simply that we have a slow time scale and a fast time scale.HT !!_(U. For nonstationary turbulence.~+ u. involving integrals over time. otherwise. By contrast. For such flows. an alternative method such as Large Eddy Simulation (Chapter 8) will be required. Thus if c! and C2 are constants while a and b denote any two flow properties. t + T) . i. i. we must look a little closer. commutes with spatial differentiation.10) Because we are dealing with definite integrals..e.13) Although it may seem a bit unusual to be taking the limit T + 00 and T + 0 in the same equation.12) vanishes because T effectively approaches 00 on the time scale of the turbulent fluctuations.12) The second term on the righthand side of Equation (2. In a perturbation analysis of such a problem. T is very small relative to the time scale of the mean flow. THE CLOSURE PROBLEM orders of magnitude. Very few unsteady flows of engineering interest are guaranteed to satisfy this condition. . and (2. the process can be fully justified using the twotiming method from perturbation theory [see Kevorkian and Cole (1981)]. the limit tiT! + 00 corresponds to the limit t IT2 + O. the mean and fluctuating components are correlated. for example. m: Iat. t) T + u~(x. Thus.14 CHAPTER 2. for any scalar p and vector Ui.Ui(X.8) and (2. ensemble averaging is necessary. dependent variables become functions of two independent time variables (essentially tiT! and tIT2)' In the normal spirit of perturbation theory.u~(x. Phase Averaging (see Problems) can be used.11) The time average of an unsteady term like au. We know that _!_ T J. Clearly our time averaging process../ at is obviously zero for stationary turbulence. For a rigorous approach. then (2. If the flow is periodic. Hence. similar to the case of small damping on a linear oscillator. time averaging is a linear operation. this is known as the spectral gap problem. t + T) . t) T (2.~) dt _ at t Ui(X.. so that the first term is the value corresponding to the limit T + 0.e. We cannot use Equations (2. the time average of their product is nonvanishing. (2. In meteorology.
' are said to be correlated if ¢'1j. There is no a priori reason for the time average of the product of two fluctuating quantities to vanish. The equations for conservation of mass and momentum are (2.2 Correlations Thus far we have considered time averages of linear quantities.16) (2. ¢' or have zero mean. (2.18) where Il is molecular viscosity and Sij is the strainrate tensor.14) tells us the mean value of a product. Thus. CORRELATIONS 15 2.15) Again. When we time average the product of two properties. ~". there is no a priori reason for the cubic term.14) where we take advantage of the fact that the product of a mean quantity and a fluctuating quantity has zero mean.17) The vectors u. and Xi are velocity and position.2. for a triple product. p is density and tij is the viscous stress tensor defined by (2. to vanish..3 ReynoldsAveraged Equations For simplicity we confine our attention to incompressible flow. we have the following: ¢¢ = (<I + ¢/)(W + . terms linear in ¢/. t is time.pI) = 4>w + ~¢' + w¢' + ¢'. The quantities ¢' and '1/. e 2.19) .p' = ~w + ¢'¢' (2. Effects of compressibility will be addressed in Chapter 5. Similarly. p is pressure. Equation (2. differs from the product of the mean values. They are un correlated if ¢'¢' = O.. ¢1j. say ¢ and ¢. we find (2.' i. ¢'¢'e'. As with terms quadratic in fluctuating quantities.2.O..
we need a prescription for computing u~uj..e.21) yields the Reynolds averaged equations of motion in conservation form.24) Equation (2.16 CHAPTER 2. the only difference between the timeaveraged and instantaneous momentum equations is the appearance of the correlation u~uj . = . we have gained no Tij Tji By inspection.17) through (2..22) (2. Unfortunately. In order to compute all meanflow properties of the turbulent flow under consideration.21) Time averaging Equations (2. Subtracting Equation (2. is identical to the instantaneous Equation (2. viz. we have produced six unknown quantities as a result of Reynolds averaging.25) so that this is a symmetric tensor. also has zero divergence.22). u~.23) can be written in its most recognizable form by using Equation (2.24) is usually referred to as the Reynoldsaveraged NavierStokes equation.20) where we take advantage of Equation (2. Herein lies the fundamental problem of turbulence for the engineer. (2.16) shows that the fluctuating velocity.16) and (2. i. we rewrite the convective term in conservation form. Combining Equations (2.16) with the mean velocity replacing the instantaneous velocity. There follows OUi 8U p+pU·=~+i at ) aXj aP OXi aXj a (s ·iPU·U· .20) in reverse. pat 8Ui + p(UjUz) OXj a = _ OXi op + (2pSjd a OXj (2. and thus has six independent components. Equation (2.23) The timeaveraged conservation of mass.22) from Equation (2. Thus.) ' 21' ) ) t (2. (2.20) yields the NavierStokes equation in conservation form. The quantity _pu~ujknown as the Reynoldsstress is tensor and we denote it by Tij. Hence. Aside from replacement of instantaneous variables by mean values. Equation (2. THE CLOSURE PROBLEM To simplify the timeaveraging process. (2.16) in order to drop udJuj/olJj.
27) In order to derive an equation for the Reynolds stress tensor. for economy of space. }. we introduce some special notation. + uj). we form the following time average.2. 2 f . To close the system. That is.U .4 The ReynoldsStress Equation In quest of additional equations.. pressure and the three velocity components.. This means our system is not yet closed. (pu~uj _ OTij ). Let N( Ui) denote the NavierStokes operator.kk.u· t pUHUj 2.22)] and the three components of Equation (2. Also.. t + pu J.29) at . To illustrate the process. viz... the NavierStokes equation can be written symbolically as (2. we must find enough equations to solve for our unknowns. viz.28) Note that. Along with the six Reynoldsstress components.. }. the resulting equation is also symmetric in i and j. we thus have ten unknowns. First... we can take moments of the NavierStokes equation. t + pu'... Using this procedure. we can derive a differential equation for the Reynoldsstress tensor.Pu'..t . For the sake of clarity. we have four unknown meanflow properties.24) for a grand total of four. Thus. Our equations are mass conservation [Equation (2. U.k = Ui..26) The viscous term has been simplified by noting from mass conservation (for incompressible flow) that Ski. (2. Z ).UJ' t + pu~u t + pu'. we multiply the NavierStokes equation by a fluctuating property and time average the product. 2. } 2. Pu .4.t + puj(Ui + UD. So. we proceed term by term.. Z.. THE REYNOLDSSTRESS EQUATION 17 additional equations. we consider the unsteady term. for general threedimensional flows.t (2.. u~ t " . we use tensor notation for derivatives throughout the time averaging process. (2. consistent with the symmetry of the Reynolds stress tensor.
(2.k + pU~U~(Uj + Uj). . The pressure gradient term is straightforward.tU·U· J ' kkt 2" '.33) ...k . the viscous term yields I ') ( J.j + uj(P + P').18 CHAPTER 2." k') In order to arrive at the final line of Equation (2.32) Collecting terms.p J' i UHP t + pl).u· (2.31) Finally.k ).k + pU~U~Uj. Bp' J J ax' t (2.k + + + + UU(Ui + UD.J J u·+u·t !:Ix.::.J' t + u'. u'.30) OXk o (PUiUjU + :. op' . U . + (2.iii k J. we have PUHUk + u~)(Uj + uj }.k + PUk UiUj uk OT.j OUj UkOXk OXk UXk ou.Tjk '(") pUj"u i.30).Tik.p . THE CLOSURE PROBLEM Turning to the convective term. we arrive at the equation for the Reynolds stress tensor.k Uk (puiuj + pUjU~(Ui + UD. we use the fact that Buj)8xk = O.k .t .i U~p' .k PU~UkUj.p'. J. + u'.tu".k puj(Uk puj Ukui.
THE REYNOLDSSTRESS EQUATION 19 We have gained six new equations. we dose the system.37) This exercise illustrates the closure problem of turbulence.35) (2.OXk {)Xk . and introduce no additional physical principles. one for each independent component of the Reynoldsstress tensor. At no point will this procedure balance our unknowns/equations ledger. . + P UiUjk + P . In making such approximations. this is not a particularly surprising situation.c I Ic UjUik (2 . However. as we take higher and higher moments. viz. Because of the nonlinearity of the N avierStokes equation. pu~uju~ OU~ OU] 21'1. we generate additional unknowns at each level. we have the following. The function of turbulence modeling is to devise approximations for the unknown correlations in terms of flow properties that are known so that a sufficient number of equations exists. After all.Opl uia+Uja Xj I ~ ~ 10 unknowns 6 unknowns 6 unknowns Op' Xi ~ With a little rearrangement of terms.4. we have also generated 22 new unknowns! Specifically. where IIij = p' ( OU~2 __ aXj aU") + __3 OXi (2.36) C ijk  I I PUiUjuk .2. On physical grounds. we can cast the Reynoldsstress equation in its most recognizable form.. such operations are strictly mathematical in nature. accounting for all symmetries..
i) n=O where r is the period of the imposed excitation. t) > = U(x) + u(x. U(x) is defined as in Equation (2. 2.20 CHAPTER 2. by definition. < u(x. t). t) = U(x) + u(x. t) > . (c) Repeat Parts (a) and (b) for du/dt. a standard way of decomposing flow properties is to write u(x. t) >= u(x. i) is the turbulent fluctuation.1 Suppose we have a velocity field that consists of: (i) a slowly varying component U(t) = Uoetlr where Uo and r are constants and (ii) a rapidly varying component u' = aUo cos (27rt/t2r) where a and t are constants with e «: 1. t) where U(x) is the meanvalue. We want to show that by choosing T = cr. (a)<U>=U (b) it = 0 (c) u' 0 = (d) « o' » = 0 (e)itv'=O (f) <uv> = u<v> < fw' > = 0 (h) < U » » = U < v > (g) . t + nr) < u(x. t) > = U(x). (b) Replace T by er in the slowly varying part of the time average of u and let if = (2r in the fluctuating part of u to show that U + u' = U(t) + O(E) where O( E) denotes a quantity that goes to zero linearly with E as ( + o. and u'(x. < u(x. Then. THE CLOSURE PROBLEM Problems 2.lim 2_ Noo N L u(x.2 For an imposed periodic mean flow. t) + u'(x. the limiting process in Equation (2. u(x.9) makes sense. We also use the Phase Average defined by Nl < u(x.5). (a) Compute the exact time average of u = U + u'. Verify the following. t) is the organized response component due to the imposed organized unsteadiness.
Pl d«. i. we have an imposed freestream velocity given by u(x.34) follows from Equation (2...33). Integrating over one period.ax) + Uoax sin 271" It where a is a constant of dimension l/length. t) = Uo(l . compute the average pressure gradient. ~'I13T. = .34). (b) Contract Equation (2.PROBLEMS 21 2. p+pu= au at au ax ax 8p 2. and I is frequency.e. set i = j and perform the indicated summation.4 Compute the difference between the Reynolds average of a quadruple product ¢1jJev and the product of the means.e. for f = 0 and I 1= 0 in the freestream where the inviscid Euler equation holds.34). to derive a differential equation for the kinetic energy of the turbulence per unit mass defined by k tu~u~. d.3 For an incompressible flow. (a) Show how Equation (2. i. Uo is a constant reference velocity. 2.5 Consider the Reynolds stress equation as stated in Equation (2.
Next we introduce Prandtl's mixinglength hypothesis and discuss its physical implications and limitations. known as the HalfEquation Model. algebraic models are. the eddy viscosity (and hence the mixing length) depends upon the flow. improves agreement between computed and measured flow properties. The latter applications illustrate the limit to the algebraic model's range of applicability. the eddy viscosity.Chapter 3 Algebraic Models The simplest of all turbulence models are known as algebraic models. in turn. by definition. An interesting separatedflow replacement for algebraic models. is often computed in terms of a mixing length that is analogous to the mean free path in a gas. We begin this chapter by first discussing molecular transport of momentum. 23 . We discuss two modern algebraic turbulence models that are based on the mixinglength hypothesis. These models use the Boussinesq eddyviscosity approximation to compute the Reynolds stress tensor as the product of an eddy viscosity and the mean strainrate tensor. in contrast to the molecular viscosity that is an intrinsic property of the fluid. including applications to attached and separated wallbounded flows. Because the eddy viscosity and mixing length depend upon the particular flow under consideration they must be specified in advance. incomplete models of turbulence. For computational simplicity. The mixinglength model is then applied to free shear flows for which selfsimilar solutions hold. Thus. The chapter concludes with a discussion of the range of applicability of algebraic models. We will find that.
1: Shear flow schematic. that the analogy between turbulent and molecular mixing is questionable. Molecules migrating across y = 0 are typical of where they come from. to say the least! It is nevertheless fruitful to pursue the analogy to illustrate how important it is to check the premises underlying turbulence closure approximations. They are so different that we will ultimately find. At the molecular level. ALGEBRAIC MODELS 3. At first glance. we decompose the velocity according to u = U + u" . it is instructive to discuss momentum transport at the molecular level.2) Figure 3. However.1 depicts such a flow. molecules moving up bring a momentum deficit and vice versa. molecules and turbulent eddies are fundamentally different.2. y U(y) ~Q \ \ \ \ \ x 2fmjp I 1 (3.1) where i is a unit vector in the x direction. That is. the model just cannot stand up under close scrutiny. This gives rise to a shear stress txy. We begin by considering a shear flow in which the velocity is given by u = U(y) i (3. As we will see.1 Molecular Transport of Momentum To understand the motivation for the Boussinesq approximation.24 CHAPTER 3. in Section 3. noting that molecular motion is random in both magnitude and direction. mimicking the molecular mixing process appears to be a careful exercise in physics. We consider the flux of momentum across the plane y = 0. as a word of caution. Figure 3.
consider the average number of molecules moving across unit area in the positive y direction.. the average vertical speed is vth/2. Also. MOLECULAR TRANSPORT OF MOMENTUM 25 where U is the average velocity defined in Equation (3. molecular velocities follow the Maxwellian distribution so that all directions are equally probable. For a perfect gas. Referring again to Figure 3. u" and o".5) Equation (3. The average molecular velocity is the thermal velocity.g. As pointed out by Tennekes and Lumley (1983). Jeans (1962)] to determine txy in terms of U(y) and the fluid viscosity.1) and u" represents the random molecular motion. The instantaneous flux of any property across y = 0 is proportional to the velocity normal to the plane which. This is not a coincidence.1. is simply »".3. appear in place of the random molecular fluctuations. First. Vth. Thus. where tij is the viscous stress tensor. where n is the number of molecules per unit volume. On average.tij. It is customary in fluid mechanics to set (J"ij = pbij . we can appeal to arguments from the kinetic theory of gases [e. half of the molecules move in the positive y direction while the other half move in the negative y direction. the turbulent fluctuations.4) By definition. dpxy. Integrating over a hemispherical shell. This similarity is the basis of the Boussinesq eddyviscosity approximation.5) bears a strong resemblance to the Reynoldsstress tensor.1. Thus. Thus. the stress acting on y = 0 is given by (J" xy = dPxy / dS. The only real difference is that. the mean free path being the average distance a molecule travels between collisions with other molecules. txy = _pUll v" (3. Now consider the transfer of momentum that occurs when molecules starting from point P cross the y = 0 plane.. p. the instantaneous flux of zdirected momentum. for this flow. the average vertical component of the velocity is Vth cos ¢ where ¢ is the angle from the vertical.3) Performing an ensemble average over all molecules. we find dPxy = pu" u" dS (3. is one mean free path away. across a differential surface area dS is dpxy := p(U + u")v" dS (3. at the macroscopic level. a stress that is generated as a momentum flux can always be written in this form. u' and v'. the average number of molecules moving across unit area in the positive y direction is nVth/4. We assume molecules are typical of where they come from which. on the molecular scale. which is approximately 4/3 times the speed of sound in air. Each molecule starting from .
molecules moving from a point Q above y = 0 bring a momentum surplus of m[U(fmJp) .8) Hence. and the momentum flux from above is Consequently. we have truncated the Taylor series appearing in Equations (3.9) where Jt is the molecular viscosity defined by (3.P+.7) at the linear terms.10) are approximate and only roughly represent the true statistical nature of molecular motion. wherefore txy = D. dy (3. The length scale L defined by (3. First.P_ and D. ALGEBRAIC MODELS a point P below y = 0 brings a momentum deficit of m[U(O) .6) and (3. Similarly.26 CHAPTER 3. the net shearing stress is the sum of D.9) and (3.P_ + D. we have made two implicit assumptions in our analysis that require justification.P+ 1 R:: '2PVthfmJp dU dy (3. we conclude that the shear stress resulting from molecular transport of momentum in a perfect gas is given by t xy = dU jJ. Hence.6) and used the fact that p = mn. Jeans (1962) indicates that a precise analysis yields Jt = OA99Vthfmjp. Interestingly.10) The arguments leading to Equations (3. the momentum flux from below is We have replaced U( fm/p) by the first two terms of its Taylorseries expansion in Equation (3. wherefore our approximate analysis is remarkably accurate! However.11) .U(O)]. where m is the molecular mass and fmJp is the mean free path. we must have fmjpld2U Idy21 ~ IdU Idyl. For this approximation to be valid.U( fmJp)].
we can say that similar to Equation (3. This will be true if molecules experience many collisions on the time scale of the mean flow.3. Second.e.14) Vmix. Thus. we also require that Vth (3.2. He visualized a simplified model for turbulent fluid motion in which fluid particles coalesce into lumps that cling together and move as a unit. In analogy to the molecular momentum transport process with Prandtl's lump of fluid replacing the molecule and fmix replacing fmJp.13) mfp f «: ldU Idyl Since Vth is of the same order of magnitude as the speed of sound..12) For most practical flow conditions. 3. the righthand side of Equation (3. fmix. the average time between collisions is fmJplVth.15) .13) defines yet another meanflow length scale. 'Txy = '2 PVmix£mix 1 dU dy (3. that he called the mixing length. Kn.8). has not been specified. we can always absorb the constant in Equation (3. As above. i. (3.1.9) is valid provided the Knudsen number. the mean free path is several orders of magnitude smaller than any characteristic length scale of the mean flow. the linear relation between stress and strainrate implied by Equation (3. the lumps retain their xdirected momentum for a distance in the y direction. is very small. The formulation is not yet complete because the mixing velocity. Equation (3. the mean free path is several orders smaller than this length scale for virtually all flows of engineering interest. The characteristic time scale for the mean flow is IdU IdylI.2 The MixingLength Hypothesis Prandtl (1925) put forth the mixinglength hypothesis. Prandtl further postulated that Vmix = constant· fmix dU dy (3. Thus. we assumed that nil remained Maxwellian even in the presence of shear. Because fmix is not a physical property of the fluid. Thus. He further visualized that in a shear flow such as that depicted in Figure 3. Now.15) which makes sense on dimensional grounds.12) is satisfied for virtually all engineering problems. in computing the rate at which molecules cross y = 0. THE MIXINGLENGTH HYPOTHESIS 27 is a length scale characteristic of the mean flow.
9) and (3. This turns out to be a reasonably good approximation over a limited portion of a turbulent boundary layer. experimental measurements indicate lu'v'l ~ O.4urmsvrms (3.3.17). Prandtl's mixinglength hypothesis leads to 'Txy = J.18) shows that the mixinglength model implies Vrms and Urms are of the same order of magnitude. Townsend (1976) states that in all turbulent shear flows. must be proportional to an appropriate average of v' such as the RMS value defined by Vrms = (v'2)1/2. comparison of Equations (3. Vmix. However. This is generally true although Urms is usually 25% to 75% larger than vrms. if Vrms Vmix.16) where /IT is the eddy viscosity given by JlT = pRmix 2 dU dy (3.17) can be deduced directly from dimensional analysis.19) Consequently. fmix. Thus. p. (mix. wakes and mixing layers.10). the mixing length is proportional to the width of the layer. The point is.u'v' = fmix dy (3. Assuming molecular transport of momentum is unimportant relative to turbulent transport. and the velocity gradient. in analogy to Equations (3. our assumed mixing length.) A straightforward dimensional analysis yields Equation (3. ALGEBRAIC MODELS and the factor 1/2 in Equation (3.17) Our formulation still remains incomplete since we have replaced Boussinesq's empirical eddy viscosity. Another interesting observation follows from replacing 'Txy by its definition so that dU 2 . we expect molecular viscosity has no significance in a dimensional analysis. /IT. Also. for example) and must be known in advance to obtain a solution. f'"V . (The eddy viscosity cannot depend upon U since that would violate Galilean invariance. Prandtl postulated further that for flows near solid boundaries the mixing length is proportional to distance from the surface.28 CHAPTER 3. The only other dimensional parameters available in a shear flow are the fluid density. each of these flows requires a different coefficient of proportionality between fmix and 8.14) in the mixing length. for free shear flows such as jets. As we will see in Section 3.15) and (3. 8.18) The mixing velocity. dU Idy. Note that Equation (3. the mixing length is different for each flow (its ratio to the flow width.lT dy dU (3. with Prandtl's empirical mixing length.
"the general conclusion must be that turbulence in a shear flow cannot possibly be in a state of equilibrium which is independent of the flow field involved.15) shows that this time is proportional to \dU Idy\I. i. Kn = emix/ L ~ 0. THE MIXINGLENGTH HYPOTHESIS 29 At this point. measurements indicate that emix ~ 0.41 (3. . the usefulness of and justification for any of its approximations ultimately lies in how well the model performs in applications. for example. y. Specifically. The turbulence is continually trying to adjust to its environment. Consequently.e. the mixing length is approximately linear with distance from the surface. Tennekes and Lumley (1983) describe the situation by saying.3. its applicability depends upon the Knudsen number being small. we will see that despite its theoretical shortcomings. the velocity follows the wellknown law of the wall [see Schlichting (1979)]. Reference to Equation (3." Thus. the linear stress/strainrate relation of Equation (3. and we defer to the applications of the following sections as its justification. the Knudsen number is of order one. as the entire formulation is empirical in its essence.21 ) Equation (3. As a pleasant surprise.16) is suspect. the assumed lifetime of Prandtl's lumps of fluid is emz"x /Vmix. we have assumed that the Boussinesq approximation holds and that the turbulence is unaltered by the mean shear. this is a forewarning that a turbulence model built on this foundation is unlikely to possess a very wide range of applicability. Unfortunately.13) is 'mix ~ \dUI dy\ Vmix (3. This is indeed consistent with the observed nature of turbulent shear flows. the theoretical foundation of the mixinglength hypothesis is a bit flimsy to say the least.21) tells us that the lumps of fluid will undergo changes as they travel from points P and Q toward y = O. It can be easily calibrated for a specific class offlows. In the same vicinity. Close to a solid boundary. without ever succeeding. Concerning the effect of the mean shear on the turbulence. the length L defined in Equation (3. Hence. On the other hand. neither condition is rigorously satisfied in practice! Concerning the Boussinesq approximation. Because we have made a direct analogy to the molecular transport process. Specifically.41y. the analog to Equation (3. Thus. we have implicitly made the same two basic assumptions we made for molecular transport.11) is equal to y. we need to examine the appropriateness of the mixinglength hypothesis in representing turbulent transport of momentum.20) Hence. and the velocity gradient varies inversely with y. the mixinglength model does an excellent job of reproducing experimental measurements..2. On the one hand.
y) = uo(x)F(yjfJ(x» (3. for the case shown in the figure. In this spirit. All three of these flows approach what is known as self similarity far enough downstream that details of the geometry and flow conditions near x = 0 become unimportant. Flows with this property are also referred to as self pre. there are no solid boundaries so that we avoid the complications boundaries add to the complexity of a turbulent flow. without ever succeeding. serving. A wake forms downstream of any object placed in a stream of fluid. What they actually mean is a relatively simple flow with slowly varying properties.. Figure 3.22) This amounts to saying that two velocity profiles located at different x stations have the same shape when plotted in the scaled form U(x. turbulence is continually attempting to adjust to its environment. First. Second. for example. Nevertheless. A flow is termed free if it is not bounded by solid surfaces.30 CHAPTER 3. ALGEBRAIC MODELS and the model's predictions are consistent with measurements provided we don't depart too far from the established data base used to calibrate the mixing length. where the Reynoldsaveraged equations of motion can be . can be expressed in the form U(x. streams moving at different speeds. we will consider only the twodimensional wake. y). and the jet. the lower stream is at rest. most turbulence researchers describe certain flows as equilibrium turbulent flows. the term equilibrium is nonsensical in the context of turbulent shear flows since. the far wake.3 Application to Free Shear Flows Our first applications will be to incompressible free shear flows.2 illustrates three different types of free shear flows. Most flows of this type can be accurately described by a mixinglength computation. viz. Strictly speaking. a fitting definition of equilibrium turbulent flow might be a flow that can be accurately described using a mixinglength model! 3. The velocity component U(x. y)juo(x) versus yjfJ(x). Eddy viscosity models based on the mixing length have been fine tuned for many flows since 1925. most notably by Cebeci and Smith (1974). We will assume the jet issues into a quiescent fluid. A mixing layer occurs between two parallel. as noted above. they are mathematically easy to calculate because similarity solutions exist. and we will analyze both the (twodimensional) plane jet and the (axisymmetric) round jet. the mixing layer. A jet occurs when fluid is ejected from a nozzle or orifice. Free shear flows are interesting buildingblock cases to test a turbulence model on for several reasons.
3. Since all three flows have constant pressure.23) (3. . Prandtl also proposed a simpler eddy viscosity model specifically for free shear flows. This greatly simplifies the task of obtaining a solution. boundary conditions are different. In this model. (b) mixing layer. in addition to the mixinglength model. and. APPLICATION TO FREE SHEAR FLOWS 31 Uoo 11 ""'_. Third. laminar flow solutions can be generalized for turbulent flow with. and X is a dimensionless empirical parameter. and excellent results can be obtained if X is assumed to be constant across the layer. The appropriate boundary conditions will be stated when we discuss each flow.2: Free shear flows: (a) far wake.+pV.25) where Umax and Umin are the maximum and minimum values of mean velocity in the layer.xy) a Of course. reduced to ordinary differential equations.24) au au 1 pU.= x y yJ a 'aaY (y3T.:::.3. (3. The standard boundarylayer approximations hold for all three of the shear flows considered in this Section. there is a wealth of experimental data available to test model predictions against. This model is very convenient for free shear flows because it is a function only of x by construction. the equations of motion are (with j = 0 for twodimensional flow and j = 1 for axisymmetric flow): au ax + yi ay (y3 v) = 0 1 a . As a historical note. fJ is the half width of the layer. (c) jet. while the equations are the same for all three flows. Additionally. (3.: (a) (b) (c) Figure 3. molecular transport of momentum is negligible compared to turbulent transport. Consequently.
27) The classical approach to this problem is to linearize the momentum equation. We begin by analyzing the far wake in Subsection 3. an approximation that is strictly valid only in the far wake [Schlichting (1979)]. we study the plane jet and the round jet in Subsection 3.3. Thus.32 CHAPTER 3.17).1.U (3.3. While an analytical solution is possible for the mixing layer. The relevant boundary conditions follow from symmetry on the axis and the requirement that the velocity approach its freestream value far from the body. Complete details of the similarity solution method are given for the benefit of the reader who has not had much experience with the method.3. Hence. we solve for 0 < y < 00. we say that U(x.y)+O ail = 0 as y + 00 (3. ALGEBRAIC MODELS at most. We leave application of this model to the Problems section. Thus.26) au ay =0 at y =0 (3. minor notation changes. we proceed to the mixing layer in Subsection 3. The far wake is especially attractive as our first application because a simple closedform solution can be obtained using the mixinglength model.30) (3.3.3. Then.2. All of the applications in this Section will be done using Equations (3.29) il(x. Finally.1 The Far Wake Clearly the flow in the wake of the body indicated in Figure 3. the boundary conditions are as y + 00 (3. y) = Uoo i .28) where 1ft] <t:: u00' The linearized momentum equation and boundary conditions become (3. numerical integration of the equations proves to be far simpler. 3.2( a) is symmetric about the x axis.31 ) ay at y=O .16) and (3.
we assume that the velocity can be written as tt(x. so required for existence of the similarity 5. we proceed in a series of interrelated steps. 6(x).34) by comparison with experimental data. Determine lution.3.34) fmix = a6(x) where a is a constant that we refer to.34). APPLICATION TO FREE SHEAR FLOWS 33 There is also an integral constraint that must be satisfied by the solution. We begin by assuming the similarity solution can be written in terms of an as yet unknown velocity scale function. 6(x) [see Figure 3. Transform 4. to close our set of equations. as a closure coefficient. Unfortunately.3.32) where D is the drag of the body per unit width. The sequence is as follows. If we consider a control volume surrounding the body and extending to infinity. 1. Our fondest hope would be that the same value of a works for all free shear flows. Thus. Assume the form of the solution. Transform 3. 1 o 00 pU(Uoo  U) dy = . y) = uo(x)F(17) (3. the equations the boundary the conditions of motion. this is not the case so that the mixinglength model must be recalibrated for each type of shear flow. Solve the resulting ordinary differential equation formed boundary conditions. Thus. 2. so = 2 P£mix oil ou oy oy (3. We use the mixinglength model to specify the Reynolds that Txy stress Txy. To obtain the similarity solution to Equations (3.D 2 1 (3.33) Finally. conservation of momentum leads to the following requirement [see Schlichting (1979)). and the wake half width.29) through (3. subject to the trans In addition to these 5 steps. uo(x). we say that (3. we assume the mixing length is proportional to the halfwidth of the wake.2(a)]. Step 1. we will also determine the value of the closure coefficient a in Equation (3.35) . conditions and the integral constraint.
34 where the similarity variable. 1] is defined by (3. (3. Note that a subscript means that differentiation is done holding the subscripted variable constant. "') space which means that derivatives must be transformed according to the chain rule of calculus. For example. In order to transform Equation (3. (3.41) . We now proceed to transform Equation (3.29).37)."luo6' dF 0 6 d1] U (3.29) and using the mixinglength prescription for the Reynolds stress leads to the transformed momentum equation. ALGEBRAIC MODELS 1].38) A prime denotes ordinary differentiation so that 6'( x) = d6 / dx in Equation (3. Thus.29). derivatives transform according to the following rules. the derivatives of u are  au ax = U 'F() "I .36) = y/6(x) Step 2. We are transforming from (x.37) (3.40) au o dF ay = r.39) (3. CHAPTER 3. we have to take account of the fact that we are making a formal change of dependent variables. d1] Proceeding in this manner for all terms in Equation (3.y) space to (x.
Clearly. Taking account of Equations (3. Note that we could have introduced a third constant in the integral constraint.a corresponds to 17+ O.42) F ( 17)1. the boundary conditions in Equations (3.00 dF d'IJ =0 at fJ =0 (3. we require the following three conditions. Finally. Thus. of course. If this is true. (3. The solution to these three simultaneous equations is simply (3. and then fall monotonically to zero approaching the freestream. the righthand side of Equation (3. be constant. absorb the third constant in 6).31) transform to (3.30) and (3. we expect that F( 17)will have its maximum value on the axis.44) Step 4. y 1.00 and y 1. APPLICATION TO FREE SHEAR FLOWS 35 Step 3. Also. Thus.00 corresponds to 171.0 as fJ 1.41) have coefficients that in general vary with x.48). we are attempting to make a separation of variables. then F'(1J) will be negative for all values of 1J and we can replace its absolute value with F'(fJ).49) .44) is a function of x.46) (3.45) through (3.43) and the integral constraint becomes (3. In seeking a similarity solution.45) The quantities al and a2 must. in effect. but it is unnecessary (we.48) Step 5.47) (3. The condition for existence of the similarity solution is that these three coefficients be independent of x. The two terms on the lefthand side of Equation (3.3.3. the momentum equation now simplifies to (3.
50) Integrating once and imposing the symmetry condition at TJ tion (3. for values of 71 in excess of 71e. This . imposing the integral constraint.52) where C is a constant of integration and 71e is given by (3.56) (3.42)] is to use Equation (3. The only way we can satisfy the far field boundary condition [Equation (3.52) for 0 < 71 < TJe and to use the trivial solution. yields an equation for the constant C. we can set TIe = 1. To understand this.}F dTJ = 0 [Equa(3. we find that the solution for F(TJ) is (3.43)] yields dF a= ..57) and If the closure coefficient a were known. Integrating once more. our solution would be completely determined at this point with Equation (3.55) Therefore. C = V20/3 = 1.. Equation (3.54) Finally.36 CHAPTER 3.44).53) This solution has a peak value at TJ = a and decreases monotonically to zero as 71 ~ TJe.57) specifying a2. F( 71) = 0. we have (3. ALGEBRAIC MODELS The second term is a perfect differential so that Equation (3. 3aC = Jri2 (3.491 (3.. note that TJ/TJe = Y/[71eo(x)]. It then increases without limit for TJ > TJe. Therefore. by setting TJe = 1 we simply rescale the TJ coordinate so that 6(x) is the wake half width as originally planned.49) can be rewrit ten as (3..51) where we observe that F'(TJ) is everywhere less than zero. Hence. With no loss of generality. Performing the integration..
2 0. QI (3.4 0. i.648 = 0.0 Figure 3.58) Comparison of Equations (3.~ 0. APPLICATION TO FREE SHEAR FLOWS 37 y/{jf 1.34)] is unknown a priori for this flow.60) Collecting all of this.46) and (3.58) shows that the value of a2 is a2 = 0.0 o.0 0.61) .3: Comparison of computed and measured velocity profiles for the far wake.8 1. according to the mixinglength model is (3.57).3.e. 0 Fage and Falkner. Mixing length. which show that the wake half width grows according to (3.f.4 0. is the nature of an incomplete turbulence model.6 0. we appeal to experimental data [c.6 0.8 0.0 0.59) The value of the coefficient QI then follows from Equation (3. The coefficient QI is unknown because the mixing length [Equation (3. To complete the solution.3.. the final solution for the far wake.18 (3. Schlichting (1979)].
This manifests itself in the nonanalytic behavior of the solution at y/6 = 1.58). 3. once calibrated.. we have found a sharp turbulent. Time averaging would thus smooth out the sharpness of the physical interface. the stream with velocity U1 lies above y = 0 and U1 > U2.1 U2 as as Y t +00 00 (3. the mixinglength model.65) The boundary conditions on 1/J are  81/J By 81jJ {jy t U1 Uz as as Y y t +00 (3. y) t U.3.63) t y t The most convenient way to solve this problem is to introduce the strearnfunction. As shown. the mixinglength model is predicting a nonphysical feature.23) is automatically satisfied and the momentum equation becomes 2 2 81/J 8 1jJ _ 81jJ 8 1/J _ !!_ 8y 8x{)y 8x 8y2 . Figure 3.66) (3. does an excellent job of reproducing measured values. Hence. y) U(x. U = 81jJ 8y and v = _ 81jJ 8x (3.67)  t t 00 . we consider two parallel streams with velocities U1 and U2. The boundary conditions are thus U(x. Specifically.38 CHAPTER 3. Measurements confirm existence of such interfaces in all turbulent flows.8y [f? mu. 2 8 1jJ 821jJ] 8y2 {}y2 (3.2 The Mixing Layer For the mixing layer. being subjected to a nearGaussian jitter. the timeaveraged interface is continuous to high order.64) Equation (3. i. all derivatives of U above (PU /8y2 are discontinuous at y/6 = 1. 1jJ. ALGEBRAIC MODELS where 6(x) is given by Equation (3./nonturbulent interface. As a final comment.e. this solution has an interesting feature that we will see in many of our applications.3 compares the theoretical profile with experimental data of Fage and Falkner (1932). By convention. Consistent with this smoothing. The velocity components are given in terms of 'IjJ as follows. we should actually expect analytical behavior approaching the freestream.62) (3. However.
34) to determine the mixing length.65) by noting that we expect a solution with aU jay = a2t/Jjay2 > O.o(X) = AU1X (3. (3.3. note that our assumed form for "p [Equation (3.70) (3. is defined by 1] = yj6(x) (3.68)] is consistent with letting F( '!]) vanish at '!] = O. Equation (3. a similarity solution exists provided we choose 1/. APPLICATION TO FREE SHEAR FLOWS 39 Because the velocity is obtained from the streamfunction by differentiation.69) into Equation (3. Thus. This is known as the dividing streamline.72) Note that we have removed the absolute value sign in Equation (3. The choice will become obvious when we set up the similarity solution.3. 1].72). As an immediate consequence. Using Equation (3. F('!]).65).75) .74) dF U2jU1 1]00 (3. "p involves a constant of integration. expanding the first term leads to the following linear equation for the transformed streamfunction.71) (l(x) = A~ where A is a constant to be determined.68) and (3. For the sake of uniqueness. we assume (3.69) As can be verified by substituting Equations (3.68) where the similarity variable. we can specify an additional boundary condition on "p. our boundary conditions are dF d1] d1] + 1 as as 1] +00 + (3. Specifically. As with the far wake. we can simplify Equation (3. although at this point it is unclear where we should impose the extra boundary condition.73) To determine the constant of integration in the streamfunction.65) transforms to (3.
0 0. for the mixing . the solut. we consider the limiting case U2 = O. 0 Liepmann and Laufer. The traditional definition of spreading rate. Furthermore.08 o .071 (Mixing Layer) (3. as wit. This problem can be solved in closed form using elementary methods. ~Mixing length. F(O) =0 (3. Unfortunately.4 0.40 .2 0.24 0.6 0.ion is a bit complicated.h the farwake solution.77) This value of Q' is nearly identical to the value (0. ALGEBRAIC MODELS y/x .247 and 0: = 0.08 .4 compares computed and measured velocity profiles.16 . t. The easier way t.4: Comparison of computed and measured velocity profiles for a mixing layer.0 uto.he mixinglength model predicts a sharp turbulent/nonturbulent interface and it becomes a rather difficult chore to determine a straightforward relationship between the closure coefficient.8 1.16 CHAPTER 3. Figure 3. optimum agreement between computed and measured [Liepmann and Laufer (1947)] velocity profiles occurs if we choose A = 0. C6. Q' and the constant A.o proceed is to solve the equation numerically for various values of Q'2 / A and compare with experimental measurements to infer the values of 0: and A. Proceeding in this manner (see Program MIXER in Appendix C).76) For simplicity.070) quoted by Launder and Spalding (1972). Figure 3.
we assume the jet issues into a stagnant fluid. or round jet.81) where J is the momentum flux per unit mass.82) The momentum equation thus becomes y J' 81jJ 82 'IjJ  8y 8x8y :y (yj ~~)] (3. which can be generalized account for the axisymmetry of the round jet..e. To solve. The boundary conditions for both the plane and the round jet are = = U(x.23) and (3. our solution must satisfy the following integral constraint: 7(J '100 " o 1 u'2y1 dy = J 2 (3. y) 8y + 0 as at y + 00 (3. we take advantage of the symmetry about the x axis and solve for a < y < 00. As with the far wake.U2)2 /(U1 .3 The Jet We now analyze the twodimensional.3.83) .. the overall agreement between theory and experiment is remarkably good. or plane jet. i.24) govern the motion with j a corresponding to the plane jet and j 1 corresponding to the round jet.3. 3.3.2(c). and the axisymmetric. Referring to Figure 3. The values of A and a have been selected to match the experimentally measured spreading rate. yi U = 8'IjJ 8y and yiv = _ o'IjJ ox (3.80) To insure that the momentum in the jet is conserved. (3. viz.we introduce the streamfunction. The jet entrains fluid from the surrounding fluid and grows in width downstream of the origin.79) 8U =0 y=O (3.78) While the computed velocity goes to zero more rapidly than measured on the low speed side of the mixing layer. APPLICATION TO FREE SHEAR FLOWS 41 layer is the difference between the values of y/x where (U .U2)2 is 9/10 and 1/10. specific momentum to flux. Equations (3. or.
91 ) (3.92) = 0. F(7J).88) F(O) :::::0 as [1 dF] d7J rP d7J d (3.075) quoted by Launder and Spalding (1972). By contrast.87) y+oo (3. and comparing with experiment yields A = 0.83) with a minus sign. the appropriate forms for 1Po(x) and 6(x) are (3. we expect to have {)U lay < O. (3.098 (Plane Jet) (Round Jet) (3. the following ordinary differential equation for the transformed streamfunction. ALGEBRAIC MODELS (3.85) where A is a constant that will be determined by comparison with experimental data.86) This equation must be solved subject to the following conditions.87) through (3.080 The values for Q are about 8% larger than corresponding values (0.90). .91) and (3.233 a = 0. Bradbury (1965)] velocity profiles for the plane and round jets. (3.92) have been obtained using an accurate solver (see Program JET in Appendix C).090 and 0.5 and 3. Somewhat larger discrepancies between theory and experiment are present for the plane jet than for the round jet.6 compare computed and measured [Wygnanski and Fiedler (1968).~dTJ 'T}J (F')2 =1 (3. the values quoted in Equations (3.246 A and and Q = 0.69). Figures 3.90) Performing a numerical solution of Equation (3. The LaunderSpalding results were obtained using numerical procedures of the 1960's and are unlikely to be free of numerical error. For the jet.89) 1 a 00 ~.68) Assuming a similarity solution of the form given in Equations and (3.42 CHAPTER 3.86) subject to Equations (3. Using this fact to replace the absolute value in Equation (3.84) 6(x) :::: Ax (3. results.
2 0.4 0.05 o 0.15 .0 0.O) Figure 3. 0) Figure 3.10 . Mixing length.0 U/U(x.0 0. Mixing length.3.05 o 0.8 1.20 .15 .6: Comparison of computed and measured velocity profiles for the round jet.6 0. . APPLICATION .5: Comparison of computed and measured velocity profiles for the plane jet.0 U/U(x.8 1.2 0. 0 Bradbury.3.25 y/x TO FREE SHEAR FLOWS 43 .10 .20 .4 0. 0 Wygnanski and Fiedler . .6 0.25 y/x .
A few final comments will help put this model into proper perspective. The mixinglength computational results shown in Figures 3. In fact.93) This concludes OUf application of the mixinglength model to free shear flows.071 = 0.100 and 0. and we have found that it must be changed for each flow.080 a While fairly close agreement has been obtained between computed and measured velocity profiles. ALGEBRAIC MODELS The traditional definition of spreading rate.4 Modern Variants of the MixingLength Model For free shear flows.{D.34) that the mixing length is proportional to the width of the shear layer.095 for the round jet. 3. C6. Prandtl originally postulated that for flows near solid boundaries the mixing length is proportional to the distance from the surface.086 and 0. We postulated in Equation (3.180 a = 0.6 correspond to C . Experimental data indicate Cf. .086 (Round Jet) (3. we established the value of our closure coefficient by forcing agreement with the measured spreading rate. not too surprisingly. turbulence behaves differently and. this postulate is consistent with the wellknown law of the wall. As we will demonstrate shortly. If we are only interested in farwake applications or round jets we might use this model with the appropriate closure coefficient for a parametric study in which some flow property might be varied. a.5 and 3. which has been observed for a wide range of wallbounded flows.098 a = 0. However. we have seen that the mixing length is constant across the layer and proportional to the width of the layer. 100 (Plane Jet) {J 0. OUf theory thus has a single closure coefficient. is between 0. The following values are optimum for the four cases considered.44 CHAPTER 3. we must use a different prescription for the mixing length. Far Wake Mixing Layer Plane Jet Round Jet a = 0.110 for the plane jet and between 0. for the jet is the value of y / x where the velocity is half its peak value. For flow near a solid boundary. we have not predicted the all important spreading rate. we must proceed with some degree of trepidation knowing that our formulation lacks in universality.
The viscous sublayer is the region between the surface and the log layer. or viscous.. The velocity asymptotes to the law of the wall as y/o + 0. Figure 3..7 shows a typical velocity profile for a turbulent boundary layer. MODERN VARIANTS OF THE MIXINGLENGTH MODEL 45 u+ 60 . The quantity y+. the law of the wall holds in the log layer. where the upper boundary is dependent upon Reynolds number.4.. sometimes referred to as the "fully turbulent wall layer . Of particular interest to the present discussion. stress is negligible compared to the Reynolds stress.10.Defect ~ Layer ____j 20 u+=y+ _ _o 1 10 u+ = ~£ny+ +B y+ Figure 3. the viscous sublayer. viz. the velocity varies approximately linearly with y+.101)]' is dimensionless distance from the surface. By definition. and gradually asymptotes to the law of the wall for large values of y+. which will be defined below [Equation (3. The defect layer lies between the log layer and the edge of the boundary layer.." is the portion of the boundary layer sufficiently close to the surface that inertial terms can be neglected yet sufficiently distant that the molecular.7: Typical velocity profile for a turbulent boundary layer. and makes a noticeable departure from the law of the 1 . This region typically lies between y+ = 30 and y = 0.3. the log layer. _ Log Layer "t. Close to the surface.. t. the log layer and the defect layer.40 Sublayer _ . From an experimenter's point of view three distinct regions are discernible.
94) pu au ax + pv au = ay [JJ au + Txy 1 ay a ay (3. we choose to call the overlap region the log layer. the Reynolds stress is much larger than the viscous stress in the log layer.96) where subscript w denotes value at the wall and Ur is known as the friction velocity.. while the viscous sublayer is the region where the inner expansion holds.46 CHAPTER 3. Equation (3. we will find the log layer to be useful because of the simplicity of the equations of motion in the layer . viz. the viscous sublayer and the defect layer. there are actually only two layers.97) can be integrated immediately to U ~ in y Ii Ur + constant (3. In the present context. (3. the defect layer is the region in which the outer expansion is valid.95) Because the convective terms are negligible in the log layer. Chapter 4 discusses these three layers in great detail. we envision the existence of an overlap region. In the parlance of singular perturbation theory (Appendix B). in which both the viscous sublayer and defectlayer solutions are valid. Strictly speaking. we can say au fla + y Txy R: JJ (au) 8 y = Tw = PUr2 w (3. Nevertheless. but rather the asymptotic limit of the inner and outer layers. The flow is governed by the standard boundarylayer equations.98) =~ where yield Ii is a constant.99) . Consequently. From a mathematician '8 point of view. Consider a constantpressure boundary layer. Hence. In performing the classical matching procedure. ALGEBRAIC MODELS wall approaching the freestream.97) If we say that the mixing length is given by (3. As noted above. according to the mixinglength model. (3. the log layer is not a distinct layer. the sum of the viscous and Reynolds shear stress must be constant.
100) (3. MODERN VARIANTS OF THE MIXINGLENGTII MODEL 47 This equation assumes a more familiar form when we introduce the dimensionless velocity and normal distance defined by U+ = U]«.0 (3. y+ (3. +B (3. The first key modification was devised by Van Driest (1956) who proposed that the mixing length should be multiplied by a damping function.100) and (3.98) have evolved.101) = ury/v 1 Introducing Equations (3. U+ ~ in y+ K.3. Specifically. Rather.105) = = where the constant At is At:::: 26 (3. Using Equation (3.41 5.101) into Equation (3. Since the mixing length was first postulated.104) B~ Note that this is not intended as a derivation of the law of the wall. considerable effort has been made aimed at finding a suitable prescription for boundarylayer computations.Y throughout the boundary layer.98) with the law of the wall. Of course. Coles and Hirst (1969) found from correlation of experimental data for a large number of attached. with some theoretical support but mainly as a good fit to data. See Schlichting (1979) or Hinze (1975) for a morecomplete history of the mixinglength model's evolution. viz. not even Prandtl expected that imix = K. and B is a dimensionless constant.98) all the way from y 0 to Y b. three of which deserve our immediate attention. it simply illustrates consistency of Equation (3. Several key modifications to Equation (3.103) (3. that the mixing length should behave according to (3..99) yields the classical law of the wall.102) The coefficient K.4. the Van Driest modification improves our description of the Reynolds stress in the . the mixinglength model fails to provide close agreement with measured skin friction for boundary layers. incompressible boundary layers with and without pressure gradient that K. Van Driest proposed. ~ 0. is known as the Karman constant.106) Aside from the primary need to improve predictive accuracy.
34)]. the coefficient of the y3 term in a Taylor series expansion for Txy must be very small as measurements are as close to Txy . Results of DNS studies (Chapter 8) indicate that indeed Txy '" y3 as y t O. and a is a closure coefficient. Since the fluctuating velocity satisfies the continuity equation. Nevertheless. However. is the velocity at the edge of the layer.. Hence. Measurements indeed indicate that the turbulent boundary layer exhibits wakelike characteristics in the defect layer. turbulence structure differs a lot between a boundary layer and a wake. In the limit of small y the Van Driest mixing length implies Txy goes to zero as y4 approaching the surface.. ALGEBRAIC MODELS O.. the mathematical approximations that yield accurate predictions for a wake and for the outer portion of a turbulent boundary layer are remarkably similar. the noslip boundary condition tells us that u' 0 at y O. as noted by Hinze (1975).8 illustrates Coles' notion that the defect layer resembles a wake flow while the wall constraint is felt primarily in the sublayer and log layer. Since there is no a priori reason for au' / 8y to vanish at the surface.. Strictly speaking. "a typical boundary layer flow can be viewed as a wakelike structure which is constrained by a wall. With imix given by Equationi3. the terminology "wake component" is conceivably a bit misleading from a conceptual point of view.. although the value 0. As pointed out by Coles and Hirst (1969). y the Reynolds shear stress must go to zero as y3.108) where h is boundarylayer thickness...09 is half the value we found for the far wake.. the Reynolds shear stress Txy '" y2 as y t O." Figure 3. 6* is the displacement thickness. Using an eddy viscosity appropriate to wake flow in the outer portion of the boundary layer also improves our physical description of the turbulent boundary layer. we conclude that u' '" y as y t O.. Escudier (1966) found that predictive accuracy IS improved by limiting the peak value of the mixing length according to (3. we also conclude that v' . Clauser specifies that = = (3.107) where PTo is the eddy viscosity in the outer part of the layer.98).25).y4 as they are to Txy '" y3 when y t O. In analogy to Prandtl's special form of the eddy viscosity for wake flows given in Equation (3. U.48 limit y t CHAPTER 3. . The second key modification was made by Clauser (1956) who addressed the proper form of the eddy viscosity in the defect layer. 2. Escudier's modification is the same approximation we used in analyzing free shear flows [Equation (3. In a similar vein. However. Hence..
3.4. MODERN VARIANTS
OF THE MIXINGLENGTH
MODEL
49
Wall Component
Wake Component
Composite Profile boundary layer. [From Coles
Figure 3.8: Coles' description of the turbulent and Hirst (1969)  Used with permission.]
The third key modification is due to Corrsin and Kistler (1954) and Klebanoff (1956) as a corollary result of their experimental studies of intermittency. They found that approaching the freestream from within the boundary layer, the flow is not always turbulent. Rather, it is sometimes laminar and sometimes turbulent, i.e., it is intermittent. Their measurements indicate that for smooth walls, the eddy viscosity should be multiplied by (3.109) where {j is the boundarylayer thickness. This provides a measure of the effect of intermittency on the flow. All of these modifications have evolved as a result of the great increase in power and accuracy of computing equipment and experimental measurement techniques since the 1940's. The next two subsections introduce the two most noteworthy models in use today that are based on the mixinglength concept. Both include variants of the Van Driest, Clauser, and Klebanoff modifications. Although it is not used in these two models, the Escudier modification has also enjoyed great popularity. As a final comment, we have introduced two new closure coefficients, At and o , and an empirical function, FKleb. As we continue in our journey through this book, we will find that the number of such coefficients increases as we attempt to describe more and more features of the turbulence.
50
CHAPTER
3. ALGEBRAIC
MODELS
3.4.1
CebeciSmith Model
The CebeciSmith model [Smith and Cebeci (1967)] is a twolayer model with PT given by separate expressions in each layer. The eddy viscosity is
/l'T = {
IlTi'
/l,To'
Y::; Yrn Y > Ym
(3.110)
where Ym is the smallest value of Y for which IlTi = IlTo' The values of J.LT in the inner layer, Ilr." and the outer layer, PTa' are computed as follows. Inner Layer: (3.111) (3.112) Outer layer: (3.113) Closure Coefficients:
I\,
=
OAO,
a=
0.0168,
A
+
= 26 [ 1+ Y
dP/dx
1/2 ]
2
(3.114)
PUT
The function FKleb is the Klebanoff intermittency function given by Equation (3.109), U; is boundarylayer edge velocity, and 8~ is the velocity thickness defined by (3.115) Note that velocity thickness is identical to displacement thickness for incompressible flow. The coefficient A + differs from Van Driest's value to improve predictive accuracy for boundary layers with nonzero pressure gradient. The prescription for /LTi above is appropriate only for twodimensional flows; for threedimensional flows, it should be proportional to a quantity such as the magnitude of the vorticity vector. There are many other subtle modifications to this model for specialized applications including surface mass transfer, streamline curvature, surface roughness, low Reynolds number, etc. Cebeci and Smith (1974) give complete details of their model with all its variations.
3.4. MODERN VARIANTS OF THE MIXINGLENGTH MODEL
51
The CebeciSmith model is especially elegant and easy to implement. Most of the computational effort, relative to a laminar case, goes into computing the velocity thickness. This quantity is readily available in boundarylayer computations so that a laminar flow program can usually be converted to a turbulent flow program with just a few extra lines of instructions. Figure 3.9 illustrates a typical eddy viscosity profile constructed by using JlT; between y 0 and Y Ym, and JlTQ for the rest of the layer. At Reynolds numbers typical of fullydeveloped turbulence, matching between the inner and outer layers will occur well into the log layer.
=
=
Y
Ym
T
JlT
Figure 3.9: Eddy viscosity for the CebeciSmith model. We can estimate the value of y~ as follows. Since we expect the matching point to lie well within the log layer, the exponential term in the Van Driest damping function will be negligible. Also, the law of the wall [Equation (3.99)] tells us aU jay ~ Ur j(KY). Thus, tir, ~ pK2y2
U
r
KY
~
PKUTY
= KJlY+
(3.116)
Since the mat.ching point also lies close enough to the surface that. we can say yjo <t: 1, the Klebanoff intermittency function will be close to one so that (with 6; = 6"'): (3.117) PTv ~ apUeo'" = apRefl· Hence, equating JlT. and JlTo' we find that
Y! ~ Refl·
K
a
~ 0.04Reb",
(3.118)
Assuming a typical turbulent matching point will lie at y~
t'V
boundary layer for which Reb. "'" 104, the 400.
52
CHAPTER
3. ALGEBRAIC
MODELS
3.4.2
BaldwinLomax
Model
The BaldwinLomax model [Baldwin and Lomax (1978)J was formulated for use in computations where boundarylayer properties such as fJ, fJ~ and U e are difficult to determine. This situation often arises in numerical simulation of separated flows, especially for flows with shock waves. Like the CebeciSmith model, this is a twolayer model. The eddy viscosity is given by Equation (3.110), and the inner and outer layer viscosities are as follows: Inner Layer: (3.119) (3.120) Outer Layer: (3.121) (3.122)
(3.123) where
Ymax
is the value of
Y
at which
fmix
Iw I achieves
its maximum
value.
Closure Coefficients:
/'i,
=
Ccp
0.40, = 1.6,
a = 0.0168, C Kleb = 0.3,
At
Cwk
= 26 } =1
(3.124)
The function FKleb is Klebanoff''s intermittency function [Equation (3.109)] with fJ replaced by YmaxlCKleb, and w is the magnitude of the vorticity vector, i.e.,
w
=
[
(av _ au)2 + (aw _ av)2 + (au _ aw)2 ax ay ay az az ax
1/2 ]
(3.125)
for fully threedimensional flows. This simplifies to w = lavlax in a twodimensional flow. If the boundary layer approximations in a twodimensional flow, then w = lau layl.
 au layl
are used
3.5. APPLICATION
TO WALLBOUNDED
FLOWS
53
U dif is the maximum value of U for boundary layers. For free shear layers, Udij is the difference between the maximum velocity in the layer and the value of U at Y = Ymax' The primary difference between the BaldwinLomax and CebeciSmith models is in the outer layer, where the product CcpFwake replaces Ueo;. To avoid the need to locate the boundarylayer edge, the BaldwinLomax model establishes the outerlayer length scale in terms of the vorticity in the layer. On the one hand, using Fwake = YmaxFmax, we in effect replace 6~ by Y~axw/Ue. On the other hand, using Fwake = CwkYmaxUlij/Fmax effectively replaces the shear layer width, 0, in Prandtl's eddyviscosity model [Equation (3.25)] by U dij /lw I. As a final comment, while Equation (3.124) implies this model has six closure coefficients, there are actually only five. The coefficient Ccp appears only in Equation (3.121) where it is multiplied by a, so aCcp can be treated as a single constant.
3.5
Application to WallBounded Flows
We turn our attention now to application of the CebeciSmith and BaldwinLomax models to wallbounded flows, i.e., to flows with a solid boundary. The noslip boundary condition must be enforced for wallbounded flows, and we expect to find a viscous layer similar to that depicted in Figure 3.7. This Section first examines two internal flows, viz., channel flow and pipe flow. Then, we consider external flows, i.e., boundary layers growing in a semiinfinite medium.
3.5.1
Channel and Pipe Flow
Like the free shear flow applications of Section 3.3, constantsection channel and pipe flow are excellent buildingblock cases for testing a turbulence model. Although we have the added complication of a solid boundary, the motion can be described with ordinary differential equations and is therefore easy to analyze mathematically. Also, experimental data are abundant for these flows. The classical problems of flow in a channel, or duct, and a pipe are the idealized case of an infinitely long channel or pipe (Figure 3.10). This approximation is appropriate provided we are not too close to the inlet of the channel/pipe so that the flow has become fullydeveloped. For turbulent flow in a pipe, flow becomes fully developed approximately 50 pipe diameters downstream of the inlet. Because, by definition, properties no longer vary with distance along the channel/pipe, we conclude immediately that
54
CHAPTER 3. ALGEBRAIC MODELS
Boundary Layer Edge
:c
Figure 3.10: Fullydeveloped
flow in a channel or pipe.
==0
aU
ax
(3.126)
Denoting distance from the center of the channel or pipe by r . conservat ion of mass is
aU + ~~
ax
r}
Br
[riV]
=0
(3.127)
where j == 0 for channel flow and j 1 for pipe flow. In light of Equation (3.126), we see that V does not vary across the channel/pipe. Since V must vanish at the channel/pipe walls, we conclude that V == 0 throughout the fullydeveloped region. Hence, for both channel and pipe flow 1 the inertial terms are exactly zero, so that the momentum equation simplifies to
=
dP 1 0=+. d [rl. (dU P+Txr ) dx r} dr dr
1
(3.128)
In fullydeveloped flow pressure gradient must be independent of x and if V ::::0 it is also exactly independent of y. Hence, we can integrate once to obtain (3.129) Now, the establishes a stress at the radius of the Reynolds stress vanishes at the channel/pipe direct relationship between the pressure gradient walls. If we let R denote the halfheight of the pipe, applying Equation (3.129) at r = R tells walls and this and the shear channel or the us that (3.130)
Tw
=
dP j + 1 dx
R
3.5. APPLICATION
TO WALLBOUNDED
FLOWS
55
Hence, introducing the friction velocity, Un the momentum equation for channel/pipe flow simplifies to the following firstorder, ordinary differential equation. 11dr
dU
+ Txr
= PUr
2
R
r
(3.131)
Noting that both channel and pipe flow are symmetric about the centerline, we can obtain the complete solution by solving Equation (3.131) with r varying between 0 and R. It is more convenient however to define Y as the distance from the wall so that
y=Rr
(3.132)
Hence, representing the Reynolds stress in terms of the eddy viscosity, IlT, we arrive at the following equation for the velocity. (3.133) Finally, we introduce sublayer coordinates, U+ and y+ from Equations (3.100) and (3.101), as well as = I'T /1'. This results in the dimensionless form of the momentum equation for channel flow and pipe flow,
I'f
VIZ.,
so+ (1 + 1'+)T dy+
where
= ( 1
y+) R+
(3.134)
(3.135) Equation (3.134) must be solved subject to the noslip boundary condition at the channel/pipe wall. Thus, we require
U+(O) = 0
(3.136)
At first glance, this appears to be a standard initial value problem that can, in principle, be solved using an integration scheme such as the RungeKutta method. However, the problem is a bit more difficult, and we find that for both the CebeciSmith and BaldwinLomax models, the problem must be solved iteratively. That is, for the CebeciSmith model, we don't know U; and 6~ a priori. Similarly, with the BaldwinLomax model we don't know the values of UdiJ and Ymax until we have determined the entire velocity profile. This is not a serious complication however, and the solution can be obtained after just a few iterations. The equations for channel and pipe flow can be conveniently solved using a standard overrelaxation iterative procedure. Appendix C describes a
56
CHAPTER 3. ALGEBRAIC MODELS
program called PIPE that yields a numerical solution for several turbulence models, including the CebeciSmith and Baldwin Lomax models. Figure 3.11 compares computed twodimensional channelflow profiles with direct numerical simulation (DNS) results of Mansour, Kim and Moin (1988) for Reynolds number based on channel height and average velocity of 13,750. As shown, the CebeciSmith and BaldwinLomax velocity profiles are within 8% and 5%, respectively, of the DNS profiles. Computed Reynolds shear stress profiles for both models differ from the DNS profiles by no more than 2%. Computed skin friction for both models differs by less than 2% from Halleen and Johnston's (1967) correlation of experimental data, viz.,
Cj 1/4 = O .0706R eH
(3.137)
where the skin friction and Reynolds number are based on the average velocity across the channel and the channel height H, i.e., Cj = Twl( ~pU;Vg) and ReH UavgH [t». Figure 3.12 compares model predicted pipeflow properties with the experimental data of Laufer (1952) for a Reynolds number based on pipe diameter and average velocity of 40,000. BaldwinLomax velocity and Reynolds shear stress differ from measured values by no more than 3%. As with channel flow, the CebeciSmith velocity shows greater differences (8%) from the data. Computed skin friction is within 7% and 1% for the CebeciSmith and BaldwinLomax models, respectively, of Praudtl's universallaw of friction for smooth pipes [see Schlichting (1979)] given by
=

1
VCi
= 410910 (2ReDyIcj)
 1.6
(3.138)
where cJ and ReD are based on average velocity across the pipe and pipe diameter, D. These computations illustrate that subtle differences in the Reynolds shear stress can lead to much larger differences in velocity for pipe and channel flow. This means we must determine the Reynolds shear stress very accurately in order to obtain accurate velocity profiles. To some extent this seems odd. The Reynolds stress is a higherorder correlation while velocity is a simple time average. Our natural expectation is for the mean velocity to be determined with great precision while higherorder quantities such as Reynolds stress are determined with a bit less precision. The dilemma appears to stem from the fact that we need the same precision in Txy as in aulay. As we advance to more complicated turbulence models, we will see this accuracy dilemma repeated, although generally with less severity. As applications go, channel and pipe flow are not very forgiving.
3.5. APPLICATION
TO WALLBOUNDED FLOWS
57
0
......
N I
0
~
,.... N I
\J
\ \
If)
0
·
4
U
If)
J)
I
· 0 .0
0
0
0
0
n
O.S U/Um
1.0
10
0 N
a
10
4
10
5
ReH
......
,.... N I
\.J
\
If)
0
·
\
:J
~
0
\
J) 0 0
:J
......
*'
· 0.0
u
O.S
' v'
1.0
7'
2
0
10
a
10
U7'~/l/
1
10
2
/U
Figure 3.11: Comparison of computed and measured channelflow properties, ReH 13,750. BaldwinLomax;    CebeciSmith; 0 Mansour et al. (DNS); 0 HalleenJohnston correlation.
=
. o o~===~~~ 0.CebeciSmith: 0 Laufer..5 1....I 10 0 N 3 10 4 10 5 10 6 ReD 0 . N I 0 ~ "" 0 v N \ \ J) If) 0 . .0 0..0 0..12: Comparison of computed and measured pipeflow properties....0 u?v'/U 7' 2 o 10 0 10 2 Figure 3. o Prandtl correlation... U 4 m I 0 0 . .5 U/Um 1.0 0 .58 CHAPTER 3. ALGEBRAIC MODELS 0 . 0... ReD = 40. N lII"! 0 o \ J) \ ::J ::J \ 0 . BaldwinLomax.000.
Consequently. y3 (Jl + jtT) (3.139) 8U] 8y 8U + pV 8U oy = dP dx + yj 8y 18[.13. The CebeciSmith model has been applied to a wide range of boundarylayer flows and has enjoyed a great deal of success. for the higher Reynolds number pipe flow. and shape factor. the twodimensional (j = 0) and axisymmetric (j = 1) boundarylayer equations are as follows. 0) U(x. The model remains reasonably accurate for favorable pressure gradient and for mild adverse pressure gradient. 3. we must account for pressure gradient. Results are expressed as functions of Reynolds number based on momentum thickness. integral parameters such as momentum thickness and shape factor often show 10% differences from measured values. The opposite is true for the lower Reynolds number channelflow case. model predictions virtually duplicate correlated values. presumably. However.5. Cebeci and Smith (1974) have devised lowReynoldsnumber corrections for their model which. y) as y>b(x) } where 6(x) is the boundarylayer thickness. would reduce the differences from the DNS channelflow results.14 compares computed and measured boundary layer properties for two of the flows considered in the 1968 AFOSRIFPStanford . Reo.139) and (3. Figure 3.O) (3. Ignoring effects of normal Reynolds stresses and introducing the eddy viscosity to determine the Reynolds shear stress. for a typical boundary layer. As shown.141) V(x. differences between computed and measured velocity profiles generally are small. compares computed skin friction. we must solve Equations (3. more acceleration is predicted with the CebeciSmith model than with the BaldwinLomax model. au + ~i_ (yiV) ox yJ 8y pU ox =0 (3.2 Boundary Layers In general.3. ej.5. APPLICATION TO WALLBOUNDED FLOWS 59 Interestingly. H. for a constantpressure (flatplate) boundary layer with Coles' [Coles and Hirst (1969)] correlation of experimental data. as we approach the boundarylayer edge. Figure 3. for example.140) subject to V(x.140) The appropriate boundary conditions follow from the no slip condition at the surface and from insisting that U + U. Because the model has been fine tuned for boundarylayer flows.
60 Cf CHAPTER 3.003 .2 2. The flow considered is an incompressible boundary layer in an increasingly adverse pressure gradient which has been studied experimentally by Samuel and Joubert [see Kline et al. shape factors differ by less than 5%. The BaldwinLomax model also closely reproduces correlated values of flatplate boundarylayer properties. The close agreement between theory and experiment for this flow is actually remarkable. 0 Coles.103 H 1. For both cases. .] Conference on the Computation of Turbulent Boundary Layers (this conference is often referred to colloquially as Stanford Olympics J). ALGEBRAIC MODELS .13: Comparison of computed and correlated shape factor and skin friction for flatplate boundary layer flow. ~~ CebeciSmith model.4 1. Despite the close agreement in velocity profiles overall. Flow 3100 is two dimensional with a mild favorable pressure gradient. It performs reasonably well even for adverse pressure gradient as evidenced in Figure 3. [From Kline et al. (1981)]. Flow 3600 is axisymmetric with an adverse pressure gradient.5 1. differences in shape factor are between 8% and 10%. For this flow. This boundary layer was presumed to be a "simple" flow for the 198081 AFOSRHTTMStanford Conference on Complex Turbulent Flows (known colloquially as Stanford Olympics II). (1969) ~ Used with permission.002 .3 1. computed and measured velocity profiles are nearly identical.004 .001 2.15.103 Figure 3.
532 1.292 8 .125 4. (1969) . as we will discuss further in Chapter 4.1 .578 1..40e 1.0nl 3. 3. H 1.6 .1 .441 1.] However. It was generally regarded as one of the most difficult to predict.535  ee MUENT METHOO EXPERIMENTAL .4 a X  0 PRESENT METHOD EXPERIMENTAL EXPERIMENTAL. Figure 3.16 compares computed and measured skin friction for Flow 3300 of the 1968 AFOSRIFPStanford Conference on the Cornputation of Turbulent Boundary Layers. both models predict skin friction significantly higher than measured.6 Flow 3100 Flow 3600 Figure 3.282 1.6.7 .3.. As shown.8 1.8 .e 2.461 1. Separation occurs .61~ I. SEPARATED FLOWS 61 u: U e IIFEET 1.408 1.446 1.424 1. by 36%.0 4.S41 1.) is 22% higher than the measured value. also known as Bradshaw Flow C.460 o· 0 .14: Comparison of computed and measured boundary layer velocity profiles and shape factor for flows with nonzero pressure gradient.403 1. CebeciSmith model. The CebeciSmith value for Cj at the final station (x = 7 ft. This flow.3 .2 . has a strongly adverse pressure gradient that is gradually relaxed and corresponds to an experiment performed by Bradshaw (1969).5 .5 . The BaldwinLomax value exceeds the measured value at x = 7 ft.4 .6 Separated Flows All of the applications in the preceding section are for attached boundary layers.4 1INCHES .2 FEET 0 0 0 3.476 1.23 yINCliES . We turn now to flows having an adverse pressure gradient of sufficient strength to cause the boundary layer to separate.412 X FEET 1. [From Kline et al.Used with permission. it proved to be the Achilles heel of the best turbulence models of the day.4 4.261 CALCULATED Ii 1.6 .440 1.600 1.0 . of all flows considered in the Conference.44e EXPERIMENTAL H 1.541 CALCULATED Ii 1.
0 2. o o 2 o o  1 a 2 3 4 5 6 x(ft) 7 Figure 3....5 Figure 3.. .5 3..15: Computed and measured skin friction for SamuelJoubert's adverse pressure gradient flow.5 2. BaldwinLomax model.62 CHAPTER 3.0 1. ALGEBRAIC MODELS 2 1 o o 1.0x(m)3.. CebeciSmith.. 0 Bradshaw... 103Cj 3 . . 0 SamuelJoubert.16: Comparison of computed and measured skin friction for Bradshaw Flow C....BaldwinLomax.
143) Figure 3. + U· . . along with all the "equilibrium" approximations implicit in an algebraic model. We must solve the full Reynoldsaveraged NavierStokes equation [Equation (2. The experiment was conducted by Driver (1991). SEPARATED FLOWS 63 in many practical applications including stalled airfoils.6. Attempts have been made to remedy the problem of poor separated flow predictions with the CebeciSmith model.142) where Sij is the mean strainrate tensor defined by 8.3. etc. algebraic models are quite unreliable for separated flows.24)]' which includes all components of the Reynoldsstress tensor. the mean strainrate tensor undergoes rapid changes in a separated flow associated with the curved streamlines over and within the separation bubble. concludes from analysis of experimental data that when a turbulent boundary layer is perturbed from its equilibrium state. flow near the stern of a ship. On the other hand. Rotta (1962). Menter (1992b) applied the BaldwinLomax model to an axisymmetric flow with a strong adverse pressure gradient.) . Shang and Hankey (1975) introduced the notion of a relaxation length.[U·t. the turbulence adjusts to changes in the flow on a time scale unrelated to the mean rate of strain. a new equilibrium state is not attained for at least 10 boundarylayer thicknesses downstream of the perturbation. separated flows are very much out of "equilibrium. The corresponding rise in pressure over the separation region is 15% to 20% higher than measured. we set (3. In analogy to Stokes hypothesis for laminar flow. In other words. the streamlines are no longer nearly parallel to the surface as they are for attached boundary layers.' 1 (3. L." The Boussinesq approximation. When a boundary layer separates. for example. can hardly be expected to provide an accurate description for separated flows.17 is typical of separated flow results for an algebraic model. Inspection of the skin friction shows that the BaldwinLomax model yields a separation bubble nearly twice as long as the experimentally observed bubble. skin friction and heat transfer. Unfortunately. to account for upstream . flow through a diffuser.J t) 2 ). As pointed out by Menter. Engineering design would be greatly enhanced if our turbulence model were a reliable analytical tool for predicting separation and its effect on surface pressure. the CebeciSmith model yields similar results. On the one hand. It is not surprising that a turbulence model devoid of any information about flow history will perform poorly for separated flows.
113).110) through (3. x = Xl. is the value of the eddy viscosity at a reference point.8~~~~~ 2 e[D 4 Figure 3.Lomax. viz. They introduced what they called a relaxation eddy viscosity model and determined the eddy viscosity as follows..144) The quantity JlTeq denotes the equilibrium eddy viscosity given by Equations (3. Baldwin. Typically. In a similar vein. However. ALGEBRAIC MODELS Cp 4~rrr~ 2 . while tir.144). The principal effect of Equation (3. (3. 0 Driver. Hung (1976) proposed a differential form of Equation (3. He was able to force close agreement between computed and measured locations of the separation point and the surface pressure distribution.145) Hung (1976) exercised these relaxation models in several supersonic shockseparated flows.17: Computed and measured flow properties for Driver's separated flow. turbulence history effects. This mimics the experimental observation that the Reynolds stress remains nearly frozen at its initial value while it is being convected along streamlines in the separation region. the relaxation length is about 561. (3. . he found that these improvements came at the expense of increased discrepancies between computed and measured skin friction.144) is to reduce the Reynolds stress from the "equilibrium" value predicted by the CebeciSrnith model. heat transfer and reattachmentpoint location. where (h is the boundarylayer thickness at x = Xl. upstream of the separation region.64 103Cjoc CHAPTER 3. and approaches a new equilibrium state exponentially.
Ur + .. Additionally Ur is the conventional friction velocity and Pw is the density at the surface. THE 1/2EQUATION MODEL 65 3. and two primary velocity scales. = L .... = where subscript m denotes the value at the point.. In its original form.147)." u(x).153) = . Ur t. Their starting point is a socalled "equilibrium" algebraic model in which the eddy viscosity is (3.3. Urn G1b.{I\. JLTi' is similar to the form used in the CebeciSmith and BaldwinLomax models. y. < G1/1\.154) .7.152) (3.. this model used only the velocity scale Um in Equation (3. assumes its maximum value denoted by Tm = (Txy )max. so that: (3.exp Us (  UDY/V 2 )] A+ I\. Txy.146) Inner Layer: The inner layer viscosity.150) (3...UsY (3.151) (3.. mUm > G1/1\. Y = O. Later.148) (3.. Y = Ym.p.'1'2) + v'P:JP Um'1'2 '1'2 = tanh(y/ Lc) t. Outer Layer: The "nonequilibrium" feature of the model comes in through the appearance of a "nonequilibriurn parameter. Ur and Um. at which the Reynolds shear stress. This scale proved to provide better predictions of velocity profile shape for separated flows than the velocitygradient prescription of Prandtl [Equation (3.15)].p. as follows: JLTi = P [ 1. y'P.Ym.rp Ur (1 .7 The 1/2Equation Model Johnson and King (1985) [see also Johnson (1987) and Johnson and Coakley (1990)] have devised a "nonequilibrium" version of the algebraic model. However... the secondary velocity scales Us and UD were added to improve predictions for reattaching flows and for flows with nontrivial effects of compressibility. the dependence on velocity gradient has been replaced by explicit dependence on distance from the surface..149) (3. Ym/b Ym/b JTm/Pm ...147) (3..
66 CHAPTER 3. = 0.70 } (3. It is unclear whether this means it has half the number of dimensions (but then. = 0. (1981)] and . the coefficient u(x) is determined so that the maximum Reynolds shear stress is given by (3. /lTo' is equal to the prescription used in the CebeciSmith model multiplied by u(x). Menter (1992b) has applied the JohnsonKing model to the attached boundarylayer flow of Samuel and Joubert [see Kline et al.09. wherefore the value from a previous iteration or an extrapolated value must be used in solving Equation (3. G1 = 0. The second term is an estimate of the effect of turbulent diffusion on the Reynolds shear stress. Closure Coefficients: K. Equation (3.157) The general idea of this model is that the Reynolds shear stress adjusts to departures from "equilibrium" at a rate different from that predicted by the algebraic model.113) shows that the outer layer viscosity. The first term on the righthand side of Equation (3. The JohnsonKing model solves the following ordinary differential equation for the maximum value of the Reynolds shear stress: Um. the /IT distribution is adjusted to agree with Tm. it would have to be a 1/3Equation Model for threedimensional applications) or if partial differential equations are twice as hard to solve as ordinary differential equations. the turbulence community has chosen the curious terminology 1/2Equation Model to describe this model.Ym] u1/2(x)j (3.155) where (um)eq is the value of Urn according to the "equilibrium" algebraic model [u(x) = 1]. 0: A+ = 17 C2 o otherwise = 0. Because this equation is an ordinary.40.155) is reminiscent of Hung's relaxation model [Equation (3. The ordinary differential equation for Tm is used to account for the difference in rates.156) That is.0168.!!_ (Tm) dx pm = al [(um)eq . As the solution proceeds.145)]. differential equation.155) for Tm. al Gdil = 0.155) is solved along with the Reynoldsaveraged equations to determine Tm.50 for O"(x) > 1. ALGEBRAIC MODELS Comparison of this equation with Equation (3.um) (Tm) Lm Pm _ CdiJ (Tm/Pm)3/211_ [C26 . as opposed to a partial. In using this model. = 0.25. computations must be done iteratively since O"(x) is unknown a priori.
The user must always be aware of the issue of incompleteness. results for the BaldwinLomax model are also included.B. both the CebeciSmith and BaldwinLomax models faithfully reproduce skin friction and velocity profiles for incompressible turbulent boundary layers provided the pressure gradient is not too strong.0 1. There is very little hope of extrapolating beyond the established data base for which an algebraic model is calibrated.5 Figure 3.BaldwinLomax.18: Computed and measured skin friction for SamuelJoubert's adverse pressure gradient flow. to Driver's (1991) separated flow.3. Figures 3. For the separated case.5 3. JohnsonKing predictions are much closer to measurements.8 Range of Applicability Algebraic models are the simplest and easiest to implement of all turbulence models. . they should be replaced only where demonstrably superior alternatives are available. Because algebraic models are so easy to use. )3..19 compare computed and measured values. . We need only recall that for the four free shear flows considered in Section 3. the JohnsonKing model predictions are somewhat closer to measured values for the SamuelJoubert flow.18 and 3.5 2.. As shown.0 2. JohnsonKing.bounded flows! On balance. most notably in the size of the separation region. They are conceptually very simple and rarely cause unexpected numerical difficulties. o SamuelJoubert. 3. These models will work well only for the flows for which they have been fine tuned.0 ( x m. RANGE OF APPLICABILITY 67 2 1 o o 1. four different values for the mixing length are neededand none of these lengths is appropriate for wall.3.
e. The number of ad hoc closure coefficients has increased from three to seven..19: Computed and measured flow properties for Driver's separated flow. i.. . the model is highly geometry dependent. Despite this wellknown limitation. Its other differences from the CebeciSmith model are probably accidental. Consequently. On the negative side. The J ohnson. within its verified range of applicability. the JohnsonKing model has been applied to many transonic flows that tend to be particularly difficult to predict with modern turbulence models. On balance. it shares many of the shortcomings of the underlying algebraic model.BaldwinLomax. However. ALGEBRAIC MODELS 4~rrr~ 2 c. the JohnsonKing model provides no information about the turbulence length scale and is thus incomplete.68 l03cJoo CHAPTER 3. many incautious researchers have applied these models to extraordinarily complex flows where their only virtue is that they don't cause the computations to blow up. Neither model is clearly superior to the other: the accuracy level is about the same for both models. . However. The chief virtue of the BaldwinLomax model over the CebeciSmith model is its independence from properties such as c5~ that can be difficult to compute accurately in complex flows.King model offers a promising modification that removes much of the inadequacy of algebraic models for separated flows. 0 Driver.8r~~~~ 2 o 2 xjD 4 Figure 3. this model appears to be a useful engineering design tool.. The model's track record has been quite good with such flows. and the model inherently requires an iterative solution procedure. neither model is reliable for separated flows. . On the positive side. the improved agreement between theory and experiment has been gained at the expense of the elegance and simplicity of the CebeciSmith model. The model is also formulated specifically for wallbounded flows and is thus restricted to such flows. JohnsonKing. like algebraic models.
begin with Equation (3.68) . introduce Equations (3.25) to represent the eddy viscosity. and determine the value of X by forcing agreement with the corresponding uo( x) and 8( x) derived in this chapter.83) and derive Equation (3.. 3.(3. determine the constant.2 For the mixing layer. beginning with Equation (3.3 For the jet.71) and derive Equation (3.1 For the far wake.6 Show that using Equation (3.65).L+).73).86).45) is given by Equations (3.7 Using a standard numerical integration scheme such as the RungeKutta method. as y+ + 0 3. generate a simi larity solution for the plane jet. generate a sim ilarity solution for the far wake. solve the following equation for U+..48). U+ ~y + y 3 2 ( +)3 + .(3. The following integrals will be useful in deriving the solution.PROBLEMS 69 Problems 3.5 Using Equation (3. The following integral will be useful when you apply the integral constraint.46) . That is.98) for the mixing length in the viscous sublayer yields a velocity that behaves according to: K. in the law of the wall implied by the mixinglength model. Obtain the exact closedform solution.4 Using Equation (3.25) to represent the eddy viscosity. 3.) c + constant 3. Jo (Xi ee de = ft 2 3. verify that the solution to Equations (3. Obtain the exact closedform solution. J 2 dx 2 cx = ~ tanh1 c (:. 3. B. dU+ (1 + J.= 1 T dy+ . and determine the value of X by forcing agreement with the corresponding uo(x) and 8(x) derived in this chapter.
at y+:::: 200..0+ sm~ '" (7rY) 2ti Also. noting that cJ :::: u.0 45R eh 1/4 and 1 s * ~ti8 where Reb:::: Ueti/v is Reynolds number based on ti..9 Assume the velocity in a boundary layer for y+ » 1 is given by o+ ~ 1 fny+ '" 1.8 For a constantpressure 0.300. Compute the quantities YmaxFmax and CwkYmaxUJif / Fmax for this boundary layer.70 CHAPTER 3.m~x + .fny+ I\. . by definition.400 and 500 Do the computation with the mixing length given by: (a) Equation (3. ALGEBRAIC MODELS Integrate from y+ ::::0 to y+ ::::500 and calculate the limiting value of B as y+ + 00 from examination of 1 B ::::U+ . make a graph of Ym / ti and Y..!. the skin friction and displacement thickness are approximately Cf ~ 3.105) NOTE: To avoid truncation error. dy+ (+)2 m~x (+)4 +2£. Then.118).' +5. You should first rewrite the equations for {tTl and ""To in terms of y/6 and Reo./U.. au+ ~1£. versus Reo for the CebeciSmith mode1. turbulent boundary layer. determine the largest value of cJ for which 2 Fwake ::::YmaxF"'max. Then./U. verify the following limiting form of the equation for dU+ /dy+. Cj :::: 2u. Use this asymptotic form very close to y+ ::::O. solve the resulting equation for Ym/ti with an iterative procedure such as Newton's method. Compare your numerical results with Equation (3. Assuming the matching point always occurs in the log layer so that au / ay :::: Ur / ("'Y) .. Let Reb vary between 104 and 106. Note also that. assume that Ymax :» 26v/ur for the BaldwinLomax model. 3.98) (b) Equation (3.
Introduce the friction velocity.06(1  y/ R)4 where R is pipe radius.!. Compare computed skin friction with Equation (3. or modify Program PIPE (Appendix C).137) and (3.138). what is the approximate form of the equation derived in Part (a) if we use the CebeciSmith model? (c) Verify that the solution to the simplified equation of Part (b) is 2 U J1 Vw T + Vw U/u.£ny + constant K 3. i. 3.10 For a turbulent boundary layer with surface mass transfer. Compare computed skin friction with Equations (3.12 Generate a solution for pipe flow using a mixinglength model with the mixing length given by Nikuradse's formula.. the momentum equation in the sublayer and log layer simplifies to: where Vw is the (constant) vertical velocity at the surface. (b) Focusing now upon the log layer where VT » u. = .138).. £mix/ R = 0. NOTE: To assist in presenting your results. See NOTE below.. Use a numerical integration scheme such as the RungeKutta method.11 Generate a solution for channel and pipe flow using a mixinglength model with the mixing length in the inner and outer layers given by .0.0. Use a numerical integration scheme such as the RungeKutta method. UT. or modify Program PIPE (Appendix C). Inner Layer Outer Layer where R is channel halfheight or pipe radius.e. in stating your integrated equation. verify that Cj = 2/(UJvg)2 and ReD = 2Ud.08(1  y/ R)2 . (a) Integrate once using the appropriate surface boundary conditions.14 .. See NOTE below.gR+ where R+ = uTR/v and Uavg is the average velocity .PROBLEMS 71 3.
Also.72 CHAPTER 3. .iJl t O. Use this asymptotic form very close to y+ = O. ALGEBRAIC MODELS 1 across the channel/pipe. to avoid truncation error verify the following limiting form of the equation for so+ /dy+ in the limit l!.
with most of the emphasis on the latter. viz.. Then. that we use to analyze modelpredicted features of the turbulent boundary layer. singular perturbation theory. This chapter discusses two types of turbulence energy equation models. but differ in one important respect. Our first applications are to the same free shear flows considered in Chapter 3. our attention focuses upon a very powerful tool. twoequation models provide an equation for the turbulence length scale or its equivalent and are thus complete. OneEquation Models and TwoEquatjon Models. Our final applications are to separated flows. we introduce twoequation models with specific details of the two most commonly used models. We proceed to a general discussion of oneequation models including examples of how such models fare for several flows. Next. Oneequation models are incomplete as they relate the turbulence length scale to some typical flow dimension. These models both retain the Boussinesq eddyviscosity approximation. and the ability of twoequation models to predict transition from laminar to turbulent flow. We discuss the issue of asymptotic consistency approaching a solid boundary. By contrast. We apply the twoequation model to attached wallbounded flows and compare to corresponding algebraicmodel predictions. The concluding section discusses the range of applicability of oneand twoequation models.Chapter 4 Turbulence Energy Equation Models As computers have increased in power since the 1960's. turbulence models based upon the equation for the turbulence kinetic energy have become the cornerstone of modern turbulence modeling research. 73 . The chapter begins with a derivation and discussion of the turbulence energy equation.
i. The quantity k should strictly be referred to as specific turbulence kinetic energy ("specific" meaning "per unit mass"). k. and to avoid confusion with the mixing length used in algebraic models. i. The sum . l. k = u~u~= _(U'2 + 2JZ 2 1 1  V'2 + w12)  (4. Prandtl (1945) postulated computing a characteristic velocity scale for the turbulence..e. p ok U ak _ at + p i ax'J  Tij oo.74 CHAPTER 4. the trace of the Reynolds stress tensor is proportional to the kinetic energy per unit volume of the turbulent fluctuations..J pe + a [ak ax'J jJ ax'J  "2PUiUjUj 1 1" . Tij.15)].3) tu puiui 2pk = = Thus. He chose the kinetic energy (per unit mass) of the turbulent fluctuations.4) The quantity e is the dissipation per unit mass and is defined by the following correlation. and k.34) leads to the following transport equation for the turbulence kinetic energy. The question now arises as to how we determine k.e. Equation (2. as the basis of his velocity scale.4) represent physical processes occurring as the turbulence moves about in a given flow. which yields the following. c = au~ {)u~ V_I _I (4. We can derive a corresponding equation for k by taking the trace of the Reynolds stress equation. (4. contracting Equation (2.1 The Turbulence Energy Equation Thrbulence energy equation models have been developed to incorporate nonlocal and flow history effects in the eddy viscosity. TURBULENCE ENERGY EQUATION MODELS 4. p. dimensional arguments dictate that the eddy viscosity is given by jJT = constant pk1/2l (4. but is often just called turbulence kinetic energy. The answer is provided by taking the trace of the Reynolds stress tensor.P'Uj '] (4.2) Note that we drop subscript "mix" in this chapter for convenience.5) aXil: aXk The various terms appearing in Equation (4. in terms of the density. Equation (3.1) Thus. ax.f. Vmix. Noting that the trace of the tensor ITij vanishes for incompressible flow.34). a turbulence length scale. In Chapter 2 we derived a differential equation describing the behavior of the Reynolds stress tensor. thus obviating the need for assuming that Vmix ~ lmix lau / ayl [c.
THE TURBULENCE ENERGY EQUATION 75 of the two terms on the lefthand side. the unsteady term and the convection. this term is seen to be the rate at which work is done by the mean strain rate against the turbulent stresses. = .6) is needed to obtain the proper trace of Tij.6) yields Tii = 2pk in accord with Equation (4. Note that the second term on the righthand side of Equation (4.1. convection and molecular diffusion are exact while production. another form of turbulent transport resulting from correlation of pressure and velocity fluctuations. Dissipation is the rate at which turbulence kinetic energy is converted into thermal internal energy. The first term on the righthand side is known as Production. equal to the mean rate at which work is done by the fluctuating part of the strain rate against the fluctuating viscous stresses. Our hope is that we can find universally valid closure approximations that make accurate solutions possible. we say that the Reynolds stress tensor is given by (4. since Sii 0 for incompressible flow. We will discuss this point in greater detail when we introduce twoequation models.3).6) where Sij is the mean strainrate tensor. and represents the diffusion of turbulence energy caused by the fluid '8 natural molecular transport process.. turbulent transport and pressure diffusion involve unknown correlations. This process is by no means rigorous.e. and regard it as the rate at which turbulence energy is transported through the fluid by turbulent fluctuations. contracting Equation (4. turbulent transport and pressure diffusion. That is. dissipation. is the familiar Eulerian derivative of k that gives the rate of change of k following a fluid particle. ReynoldsStress Tensor: For the class of turbulence models considered in this chapter. The conventional approach to closure of the k equation was initiated by Prandtl(1945) who established arguments for each term in the equation.4. We refer to the triple velocity correlation term as Turbulent Transport. dissipation. and represents the rate at which kinetic energy is transferred from the mean flow to the turbulence. This termbyterm modeling approach amounts to performing drastic surgery on the exact equation. replacing unknown correlations with closure approximations. The last term on the righthand side of the equation is called Pressure Diffusion. To close this equation. Thus. i. we assume the Boussinesq approximation is valid. The unsteady term. Rewritten as Tij Sij. The term involving j1ok/{)xj is called Molecular Diffusion. The closure approximations are no better than the turbulence data upon which they are based. we must specify Tij.
7). (4. Dissipation: The manner in which we determine the dissipation is not unique amongst turbulence energy equation models.g. .. In the following sections. this statement applies closure coefficients. we hope the model is realistic enough that Uk be constant. and the sum assumed to behave as a gradienttransport process. the pressure diffusion term has generally been grouped with the turbulent transport.. we will explore the various methods that have been devised to determine the length scale.lT {)k = . L If both properties are assumed to be strictly functions of the turbulence independent of natural fluid properties such as molecular viscosity. and the dissipation. Thus. we can write the modeled version of the turbulence kinetic energy equation that is used in virtually all turbulence energy equation models. Note that Equation (4. f. we say that uj¢i .4) and (4. Fortunately. to all turbulence entered although. Combining Equations (4. The equation assumes the following form.OXj J Uk (4. Kim and Moin (1988)] indicate that the term is quite small for simple flows. Mansour. recent DNS results [e. we assume that pu~u~u'. + t t 1 2  J J..6)...8) Hence. J. Unfortunately. At this point. there is no corresponding straightforward analog for the pressure diffusion term. TURBULENCE ENERGY EQUATION MODELS Turbulent Transport and Pressure Diffusion: The standard approximation made to represent turbulent transport of scalar quantities in a turbulent flow is that of gradientdiffusion. can be chosen to where Uk is a closure coefficient..7) As stressed by Bradshaw (1992). no approximation has of course.9) where Tij is given by Equation (4. It suffices at this point to note that we still have two unknown parameters. we still need a prescription for the length scale of the turbulence in order to close our system of equations.76 CHAPTER 4.. In analogy to molecular transport processes. purely dimensional arguments [Taylor (1935)] show that (4. which are the turbulence length scale. For want of definitive experimental data. p'u'.lT{)CP/{)Xj.7) simply defines Uk.
Prandtl postulated that the dissipation assumes the form quoted in Equation (4. is given by Equation (4.12) Note that at this point we make an implicit assumption regarding the "constant" in Equation (4. we obtain (u'v'f/2jfmix Cnk3/2jf so that f ex: fmix if +u'v' jk constant. note that in a thin shear layer. Wolfshtein (1967) found that by introducing damping factors in the dissipation and eddy viscosity similar to the Van Driest . the dissipation is (4. JlT will not.2. the length scale. no reason why "constant" should really be constant. i.07 and 0.6) and the eddy viscosity is (4. In reality.g. Their length scale distributions were similar to those used for the mixinglength model.. Prandtl had sufficient confidence that he could generalize established prescriptions for the turbulence length scale C. Only in equilibrium flows for which production and dissipation balance are meanflow and turbulence scales proportional ~ and then either can be used for PT.)1/2jfmix' Hence.4. Consequently.09. pu'v') to a mean flow quantity (e. an unknown mix of scales is needed.. Of course.8).g. Before the model can be used in applications. JlT is the ratio of a turbulence quantity (e. production balancing dissipation means we have aUjfJy (u'v. i. there is no a priori reason why JlT should depend only upon k and E. To see this. Introducing a closure coefficient that we will call CD. au jay + aVjax).2).e. Given twenty years of experience with the mixinglength model.11) where ii. That is. and the closure coefficients (Tk and CD must be specified. in general. ONEEQUATION MODELS 77 4. f ex fmix only if the ratio of production to dissipation is constant. precisely follow meanflow scales such as Ue and {)* or turbulence scales such as k and E. Thus.10) and the turbulence length scale remains the only unspecified part of the model. Emmons (1954) and Glushko (1965) applied this model to several flows with some degree of success using (1} = 1 and CD ranging between 0. which has been set equal to one. Otherwise.. the first OneEquation Model appears as follows: = = = (4. Emmons (1954) proposed essentially the same model in an independent research effort.2 OneEquation Models To complete closure of the turbulence kinetic energy equation.
lT / p. vr J. the PrandtlEmmonsGlushko oneequation model is certainly straightforward and elegant. postulated a phenomenological transport equation for the kinematic eddy viscosity.number corrections.16]. This is a direct consequence of introducing Equation (4. the BradshawFerrissAtwell model's skin friction for boundary layers in adverse pressure gradient was closest of the various models tested in the 1968 Conference to measured values.3pk (4. is constant. Oneequation models have been formulated that are based on something other than the turbulence energy equation. TiCy. to the turbulence kinetic energy. Measurements [Townsend (1976)] indicate that for boundary layers.1 compares computed and measured skin friction for Flow 3300 of the 1968 AFOSRIFPStanford Conference on the Computation of Turbulent Boundary Layers.King models.105)]' more satisfactory results can be obtained with this model for lowReynoldsnumber flows. Although more complex than an algebraic model. two damping functions. As originally postulated it involves two closure coefficients and one closure function (the length scale). k. TURBULENCE ENERGY EQUATION MODELS factor [Equation (3. The model has four closure coefficients = . Overall. Even with Wolfshtein's lowReynolds. The equation involves terms similar to those appearing in Equation (4. the number of closure coefficients increases by only two so that the model actually has fewer closure coefficients than the BaldwinLomax and J ohnson. Rather than introduce the Boussinesq approximation. wakes and mixing layers the ratio is very nearly the same and given by TiCy ~ 0. Ferriss and Atwell formulated a oneequation model based on the turbulence kinetic energy. Nee and Kovasznay (1968).13) in modeling the k equation's turbulent transport term rather than a gradientdiffusion approximation. they argue that for a wide range of flows. the differences between theory and experiment are much less than those obtained using the CebeciSmith and BaldwinLomax models [see Figure 3. Ferriss and Atwell (1967) formulated a oneequation model that avoids introducing a gradientdiffusion approximation. As shown. The resulting equations are thus solved using the method of characteristics. Bradshaw. for example. Figure 4.13) Building upon this presumably universal result. Bradshaw. A novel feature of their formulation is that the equations are hyperbolic for boundary layers rather than parabolic. For attached flows. and a closure function for the length scale. Goldberg (1991) has refined the model even further.78 CHAPTER 4. Goldberg's number of closure coefficients and empirical functions more than doubles for separated flows. the Goldberg model has five closure coefficients.11). More recently. the ratio of the Reynolds shear stress.
_BradshawFerrissAtwell. K./K. The BaldwinBarth model is as follows. two empirical damping functions and a function describing the turbulence length scale. c. Figure 4. + o. At 2 = 26... and requires prescription of the turbulence length scale.14) Turbulence Reynolds Number :t (vRT) = 1.17) Auxiliary Relations P = VT [(OUi {}Xj + OUj) aUi _ ~ OUk aUk] OXi aXj 3 aXk {}Xk (4.' +( v + VT/ Closure Coefficients Cd (J' c ) 82(vRT) 8XkOXk   1 8VT 8(vRT)  (J'cOXk OXk (4..18) . More recently.15) Gt:2= 1_( _ (J'f_ 2.2.4.. includes seven closure coefficients. Baldwin and Barth (1990) and Spalart and Allmaras (1992) have devised even more elaborate model equations for the eddy viscosity. The BaldwinBarth model.=0.__. 0 Bradshaw.. = 0.. Kinematic Eddy Viscosity lIT = GJ. for example. 8~' (vRT) = (G(2h ~ Cd) J J vRTP _..16) 1fT Cf2CfdyCp.lvRTD1Dz (4.41 (4. ONEEQUATION MODELS 79 2 3 4 5 6 x(lt) 7 .09.0.2.1: Comparison of computed and measured skin friction for Bradshaw Flow C. At = 10 (4.
= 0.aUj/aXi) is the rotation tensor and d is distance from the closest surface.21) (4.80 CHAPTER 4. Although not listed here.3. 0.23) (4. The model includes eight closure coefficients and three damping functions.26) (4. Cw3 2.25) v X =.3 ' 1X . /'i. = 0. Cw2 Cvl = 7.19) The SpalartAllmaras model is also written in terms of the eddy viscosity. f vI ' !w = 9 [ 96 1 + c6 + cw3 ~3 ]1/6 ( 4.1355. v  g=r+Cw2(r V 6 r). /'i. The tensor nij = !(aUi/aXj .24) Cwl = 2 + (1 + Cb2) .622. (J Cb2 = 0.41 Auxiliary Relations X3 !vl = XCvI 3 + . = (J = 2/3 = (4.27) S=S+2"2!v2.2d2 (4. the model even includes a transition correction that introduces four additional closure coefficients and two more empirical functions.1.22) Closure Coefficients Cbl Cbl /'i. r = :::_ v d S/'i. Kinematic Eddy Viscosity vr Eddy Viscosity Equation = V!vl (4. . Its defining equations are as follows. TURBULENCE and ENERGY EQUATION MODELS (4.f v2 1 + X.
the SpalartAllmaras model yields cf for the SamuelJoubert case that lies about as far above the measurements as the BaldwinBarth C f lies below [see Spalart and Allmaras (1992)].. . ~'t. \ "Oil . we clearly have not yet arrived at anything .2 illustrates how well the BaldwinBarth model reproduces correlations of measured skin friction [see Hopkins and Inouye (1971)] for constantpressure boundary layers._ ' . the BaldwinBarth model skin friction deviates from measured values even more than the BaldwinLomax model (see Figures 3.l) .. BaldwinBarth and SpalartAllmaras models have achieved closer agreement with measurements for a limited number of separated flows than is possible with algebraic models. only a modest advantage is gained in using a oneequation model rather than a mixinglength model. For both flows.~:. Baldwin and Barth and Spalart and Allmaras show improved predictive capability (relative to early oneequation models) for some flows. the Goldberg.15 and 3. "'. ONEEQUATION 5 103Cf 4 MODELS 81 : 3 .l) r.0) 1 •\ 2 1_ . their track record remains spotty.4 show how the BaldwinBarth model fares for the two key flows considered by Menter (1992b). · . oo ~~~20000 10000 30000 Reo Figure 4. .1 and Mach 2.' ••••••••••••••••• : . BaldwinBarth model.. On the other hand.. · ~ · · · . [From Baldwin and Barth (1990).. In summary.'.2._. On the one hand.3 and 4.0) . Theory (M=2.~~~~~~~RH~~~~_J 1 ~. While these newer models appear promising for separated flows. Figures 4. Given all of these facts.\01.. the BaldwinLomax model appears to be superior to the BaldwinBarth model for the relatively simple SamuelJoubert flow and for Driver's separated flow._ Calculation (M=O.] Figure 4.4..2: Comparison of computed and measured flatplate boundary layer skin friction at Mach 0. While the recent developments by Goldberg. more research and testing is needed.. Although not shown.17). : '_: Oneequation model _.0. Theory (M=O.Calculation (M=2. ·~·~:.
.0 1. 0 Driver.82 CHAPTER 4. _BaldwinBarth. BaldwinBarth.0 2.0 x(m) 3.8~~~P~ 2 ~ Figure 4. TURBULENCE ENERGY EQUATION MODELS 2 1 o 1. 0 SamuelJoubert.3: Computed and measured skin friction for SamuelJoubert's adverse pressure gradient flow.5 3.5 2. 1~~= 4~~~r.4: Computed and measured flow properties for Driver's separated flow.5 Figure 4.
To reach a morenearly universal model. the eddy viscosity. This quantity has dimensions of (time)l. Equation (4. (4. or equivalently. For example. oneequation models share a few of the failures as well as most of the successes of the mixinglength model. They are.3. pointed out that a second transport equation is needed to compute the socalled specific dissipation rate. can be used to predict properties of a given turbulent flow with no prior knowledge of the turbulence structure.. abrupt changes from wallbounded to free shear flows (e.1. especially for separated flows. twoequation models are complete.9). flow at a trailing edge of an airfoil) cannot be easily accommodated.. t.e. rv rv rv Chou (1945) proposed modeling the exact equation for L In terms of this formulation.28) J. f k1/2/W. the turbulence length scale. On dimensional grounds. almost all of the computations done for the 198081 AFOSRHTTMStanford Conference on Complex Turbulent Flows used twoequation turbulence models. The starting point for virtually all twoequation models is the Boussinesq approximation. 4.29) Rotta (1951) first suggested a transport equation for the turbulence length scale and later (1968) proposed an equation for the product of k and t.lT pk/w.30) . f. i. the simplest complete model of turbulence. Kolmogorov (1942). for example. TWOEQUATION MODELS 83 resembling a universal turbulence model. in fact.g. the eddy viscosity and turbulence length scale are (4. there is a nonuniqueness in the way we determine the dissipation. Consequently.6). In general. we must seek a model in which transport effects on the turbulence length scale are also accounted for. turbulence length scale and dissipation can be determined from ! wk (4. As pointed out at the end of Section 4.4. w.3 TwoEquation Models TwoEquation Models of turbulence have served as the foundation for much of the turbulence model research during the past two decades. While there is a smaller need for adjustment from flow to flow than with the mixinglength model. In either case. and the turbulence kinetic energy equation in the form of Equation (4. but also for the turbulence length scale or equivalent. The rest of this chapter is devoted to investigating models that indeed include transport effects on the turbulence length scale. These models provide not only for computation of k.
Specifically. it unveils nothing about the physics underlying its implied scaling relationships.31) Regardless of the choice of the second variable in our twoequation model.1 The kw Model As noted above. and can be expected to be inaccurate for many nonequilibrium turbulent flows. twoequation turbulence models are no more likely than oneequation models to apply universally to turbulent flows. TURBULENCE ENERGY EQUATION MODELS More recently. Saffman (1970) formulated a ki» model that would prove superior to the Kolmogorov model. The physics is in the choice of variables.3. Thus.. we should always be aware that while dimensional analysis is extremely useful.84 CHAPTER 4. modeled the differential equation governing its behavior. most notably with the ki» model that help clear up some. dimensional analysis has been one of the most powerful tools available for deducing and correlating properties of turbulent flows. As with oneequation models. of the uncertainty. His second parameter was the dissipation per unit turbulence kinetic energy. f. we see a recurring theme. Abid and Anderson (1990) have postulated an equation for T. interesting developments have occurred. it is worthwhile to pause and note the following. Speziale. Historically. there is no fundamental reason that JtT should depend only upon turbulence parameters such as k. In his kw model. the ratio of individual Reynolds stresses to mean strain rate components depends upon both meanflow and turbulence scales. In general. However... For these models. One of the key conclusions of the 198081 AFOSRHTTMStanford Conference on Complex Turbulent Flows was that the greatest amount of uncertainty about twoequation models lies in the transport equation complementing the equation for k. w. Further. the dissipation. Zeierrnan and Wolfshtein (1986) introduced a transport equation for the product of k and a turbulence dissipation time. pkr. 4. JtT . As part of the Imperial . In the decade following the Conference. Before proceeding to details of twoequation models... Kolmogorov chose the kinetic energy of the turbulence as one of his turbulence parameters and. T. e or w. Kolmogorov (1942) proposed the first twoequation model of turbulence. Also. With no prior knowledge of Kolmogorov's work. but not all. wsatisfies a differential equation similar to the equation for k. like Prandtl (1945). eddy viscosity and length scale are all related on the basis of dimensional arguments. which is essentially the reciprocal of Kolmogorov's w. (4. it was even unclear about what the most appropriate choice of the second dependent variable is.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.