ee ae tle Perel ee ela hee LOL etd The Physics of Fluid Turbulence W. D. McCOMB OXFORD SCIENCE PUBLICATIONSOXFORD ENGINEERING SCIENCE SERIES EMERITUS EDITORS A.L. CULLEN L. C. WOODS GENERAL EDITORS J.M. BRADY C. E, BRENNEN W. R. EATOCK TAYLOR M. Y. HUSSAINI T.V. JONES J. VAN BLADELTHE OXFORD ENGIN! ING SCIENCE SERIES 19 D. R. RHODES. Symthesis of planar untenna sources L.C. WOODS: Thermodynamics of fluid systems R.N. FRANKLIN: Plasma phenomena in gas discharges G.A. BIRD: Molecular gas dynamics H.-G. UNGER: Planar optical waveguides and fibres: J. HEYMAN: Equilibrium of shell structures K. H. HUNT: Kinematic geometry of mechanisms D.S. JONES: Methods in electromagnetic wave propagation (Two volumes) W.D. MARSH: Economics of electric uility power generation P. HAGEDORN: Non-linear oscillations (Second edition) R. HILL: Mathematical theory of plasticity D.J. DAWE: Matrix and finite clement displacement alysis of stractires NW. MURRAY: introduction tothe theory of thin-walled structures R.LTANNER: Engineering rheology BM. F. RANNINEN and C. tt. POPELAK: Advanced practure mechames RH. T. BATES and M. J. MeDONNELL: Image restoration and reconstruction K. HUSEYIN: Multiple parameter stability theory and its applications R.N. BRACEWELL: The Hartley transform J. WESSON: Tokamaks P.B, WHALLEY: Boiling, condensation, and gas-tiguid flow C. SAMSON, M. LeBORGNE, and B. ESPIAU: Robot control: the task function approach HJ. RAMM: Fluid dynamies for the study of transonic flow R.R.A.SYMS; Practical volume holography W.D. MeCOMB: The physics of fluid nrbulence Z.P. BAZANT and L. CEDOLIN: Stability of sructures: principles of elasti, inelastic, and damage theories J.D. THORNTON: Seience and practice of liquid-liquid extraction (Two volumes) J. VAN BLADEL: Singular electromagnetic fields and sources M. ©. TOKHI and kK, LELICH: Active noise control LV. LINDELL: Methods for electromagnetic field analysis J. A.C. KENTFIELD: Nonsteady, one-dimensional, internal, compressible flows W.F, HOSFORD: Mechanics of erystals and polycrystalsThe Physics of Fluid Turbulence W. D. McCOMB CLARENDON PRESS - OXFORDOxford University Press, Walton Seret. Oxford OX2 6DP Oxford New York Toronto ein Hombay Calcutta Madras Karachi Kuala Lumpur Singapore Hong Kong Tokvo ‘Nairobi Dares Salaam Cape Town Melbowne Auckland Madrid and associated companies in Berlin Ibadan Oxyord isa trade mark of Oxjora Universiy Press Published in the United States by Oxford University Press Inc., New York ©W. D. McComb 1990 First published 1990 First published in paperback (with corrections) 1991 Reprinted 1992 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Oxford University Press This book ts sold subject to the condition that it shall not, by way of trade or otherwise, be lent, re-sold, hired out, or otherwise circulated without the publisher's consent in any form of binding or cover other than that in which it is published and without a similar condition including this condition being imposed on the subsequent purchaser British Library Cataloguing in Publication Data McComb, W. D. The physics of fluid turbulence. 1. Fluids, Turbulence 1. Title $37:0527 ISBN 0-19-856256-X (Pbk) Library of Congress Cataloging in Publication Data MeoComb, W. D. The physics of fluid turbulence W. D. McComb, p.em.—(Oxford engineering science series; 25) ISBN 0-19-856256-X (Pbk) 1, Turbulence, 2. Fluid mechanics. 3. Perturbation (Physics) 4. Renormalization (Physics) 1. Title. II. Series. QA913.M43 1990 532.0527 —Wie20—89-28440 Printed in Great Britain by Bookoraft (Bath) Lid. Midsomer Norton, AvenFOR MY WIFEPREFACE Over the last three decades the various topics making up the subject of engineering science have tended to become ever more scientific and sophisti- cated. In effect the emphasis has been increasingly on the ‘science’ aspect of the subject, and the study of turbulence has heen no exception Up until about the end of the 1950s, most work on turbulence involved the application of ad hoc methods to difficult practical problems. That is to say, turbulence was essentially a branch of hydraulics, and hence was treated by the methods of ‘handbook engineering’. There was only a relatively small amount of fundamental research and this was described in a rather complete way by the well-known monographs due to Batchelor, Hinze, and Townsend (for references, see Chapter 1), all of which were first published in the period 1953-1959, However, since i960 the study of turbulence has seen a great increase in activity of a more scientific kind. The invention of laser anemometry, the development of powerful computers, new methods of digital data processing and signal analysis, and the introduction of exotic (at least, in engineering terms) theoretical methods from quantum physics have all played a part. It is perhaps not altogether surprising that over this time there have been some important developments in the subject. As examples, one may note the spectacular phenomenon of drag reduction by additives, the recognition of the importance of coherent structures, the direct numerical simulation of turbulence on the computer, and the derivation of closures of the Navier Stokes hierarchy using renormalized perturbation theory. As enginccring has tended to become more scientific, there has been (pleas- ingly to one’s sense of symmetry!) a comparable trend in physics in the reverse direction. The stunning success of modern physics (ie. Einsteinian relativity and, more particularly, quantum mechanics) in solving many fundamental problems has led, by way of the search for new worlds to conquer, to a strong applied physics discipline. In turn, mutual interest has resulted in the forma- tion of many interfaces between engineering science and applied physics, especially in the areas of electronics, physical optics, and materials science. Itis at this point that the study of turbulence does prove to be an exception: the applied physics involvement is almost completely absent. In view of the extraordinary practical importance of turbulence (not to mention its intense study by other disciplines in the context of applications), thie ie at firet sight quite astonishing. Yet the reason for such apparent neglect is easily found. Quite simply the fundamental problems of turbulence are still unsolved. That is, the ‘turbulence problem’ is still a matter of pure physics. Indeed, turbulence is often referred to as the ‘unsolved problem of classical physics’vill PREFACE Considered as a problem in pure physics, turbulence has not been altoget! neglected, But, when compared with topics like condensed matter physics or high energy physics, it has been very much a minority cult. It is therefore particularly encouraging to note that over the past decade or so there have been convincing signs of an exponential growth in the study of turbulence as a branch of physics, This is, of course, a predictable trend. As fewer and fewer linear problems are left to solve, there is inevitably a growing pursuit of non-linear problems, and turbulence is the archetypal non-linear non- equilibrium problem of statistical physics, In this book our aim is to deal with certain topics which form a subset of both engineering science and physics. In doing so, we hope to assist two broad classes of reader, First there are those who are new to the subject (although not necessarily new to research), and secondly there are those who are already familiar with one or more of the traditional branches of the study of turbulence, but whose background and experience does not prepare them for the usages of quantum physics. (In passing, one notes that theoretical physicists who write papers on turbulence often appear to make little or no effort to present their work in a way which would be accessible to the more theoreti- caily inclined members of the turbulence community.) In order to cater for, in particular, the first of these groups, we have aimed at a reasonable degree of completeness. The first two chapters set out to give a concise summary of the theory and practice of turbulence up to about 1960 This material is based upon lectures, which I have given to final-year under- graduates, originally in mechanical engineering but now, in more recent years, in physics. Chapter 3 then serves two purposes. First, it tries to give the reader a broad picture of what the rest of the book is about. Thus certain topics in this category can be regarded as overviews of their more detailed treatment later in the book. The second purpose of this chapter is to cover topics which would not easily fit into the main part of the book, yet are important. Section 3.1 on anemometry and data processing is an example of such a topic. Chapters 4-10 constitute the main part of the book and deal with modern (post-1960) turbulence theory. Chapter 4 begins by presenting some back- ground material on the statistical mechanics of the classical N-body system This introduces useful concepts and terminology, and provides a context for the subsequent rigorous formulation of the turbulence problem as an example ofa non-equilibrium statistical system with strong coupling. Chapter 5 aims to ‘de-mystify’ the application of renormalized perturbation theory (RPT) to turbulence. RPT is introduced for some simpler problems by considering (a) the virial cluster expansion in dilute N-body systems and (b) the Debye- ickel screened put ample oflong range * Aninteresting and authoritative discussion of this situation was given by R. Kubo in his opening. address to the Oji Seminar on Non-linear, Non-equilibrium Statistical Mechanics (1978: Suppl Progr. Theor. Phys. 64, 1)PREFACE ix interactions. ‘Then, a general treatment of the perturbation expansion of the Navier-Stokes equation (based on a modified version of Wyld’s analysis) follows, and the chapter closes with a consideration of Kraichnan’s direct- interaction approximation as an example of a second-order truncation of the renormalized expansion. It should be noted that at this point (and at various others) we do not always keep to the strict chronological order in which the different pieces of work were carried out. Chapter 6 deals with those RPTs which do not yield the Kolmogorov spectrum as a solution and Chapter 7 deals with thse that do. Chapter 8 attempts to provide a critical assessment of RPTs. The main emphasis is on the comparison of numerical solutions of the spectral and response equations with the results of laboratory and computer experiments. The chapter closes with an appraisal of RPTs in the light of various published critiques, and their relationship with other fields of physics. In Chapter 9 we introduce the newer method of renormalizing the ansport coefficients in turbulence: the renormalization group (RG). Both this and RPT methods crop up again in Chapter 10, where we discuss the numerical simula- tion of turbulence. This is entirely appropriate, tor the general area of large- eddy simulation (which resembles other hybrid areas of physics, such as lattice gauge calculations) provides for the first time an arena where engineers and physicists can meet on equal terms and find much of mutual interest. The final part of the book, consisting of Chapters 11-14, can be seen as offering some sort of counterpoise to the exclusively theoretical (and often esoteric) nature of most of the preceding chapters. My intention here is to deal with some of the more practical aspects of the subject, while at the same time illustrating the amazing richness of physical phenomena which has emerged from experimental studies of turbulence over the last few years. What is now needed is much more interest from physicists, especially on the experimental side which has for the most past been left (o the engineers. If the material contained in Chapters 11-14 does not inspire the requisite interest, then I believe that nothing will! I took my own first steps in turbulence theory under the guidance of Sam Edwards and David Leslie, In their different ways they taught me much. It is apleasure to acknowledge my debt to them here. Also, in the course of writing this book, I have received a great deal of help from friends and colleagues. | would like to thank Francis Barnes, Phil Hutchinson, V. Shanmugasundaram, and Alex Watt for reading the manuscript (in varions drafts!) and pointing out errors and making suggestions for improvement. Finally, | would like to thank Denis Jones of Dantec Electronics for his help and advice concerning Edinburgh W.D. McC. June 1989ACKNOWLEDGEMENTS Itis a pleasure to thank the various organizations and individuals, who gave permission to reproduce figures, as follows. American Institute of Chemical Engineers Figures 3.15 and 14.2 are reprinted by permission from AIChE J. 28, 547-65. Copyright 1982. American Institute of Physics Figure 14.1 is reprinted by permission from Phys. Fluids 20, 873-9, Copyright 1977 Cambridge University Press Figure 1.7 is reprinted by permission from J. Fluid Mech. 48, 477-505. Copyright 1971, Figures 3.17 and 3.18 are reprinted by permission from J. Fluid Mech. 43, 689-710. Copyright 1970. Figures 8.6-8.14 are reprinted by permission from J. Fluid Mech, 143, 95-123 Copyright 1984. Plate Il is reprinted by permission from J. Fluid Mech. 110, 73-95. Copyright 1981 Dantec Electronics Lid Figures 3.5 and 3.6 were reproduced from negatives kindly supplied by Dantec. Institute of Petroleum Figure 3.16 is reprinted by permission from J. Inst. Petrol. 47, 329-25. Copyright 1961 Macmillan Magazines Limited Figure 14,3 is reprinted by permission from Nature (Lond.) 292, 520-2. Copy- right 1981 Macmillan Magazines Ltd. Pergamon Press Figure 13.1 is reprinted by permission from Chem. Eng. Sci. 26, 419-39. Copyright 1971. Figure 13.2 is reprinted by permission from J, Aerosol Sci. 6, 227-47. Copy- right 1975. Figure 13.3 is reprinted by permission from J. Aerosol Sci. 9, 229-313. Copy- right 1978xii ACKNOWLEDGEMENTS Springer-Verlag New York Inc Figures 8.4 and 8.5 are reprinted by permission from Statistical models and turbulence (eds M. Rosenblatt and C. Van Atta), Lecture Notes in Physics Series, Vol. 12, p. 148. Copyright 1972 (article by J. R. Herring and R. H. Kraichnan). Thanks are also due to Professor F. H. Abernathy (Figs 3.17 and 3.18), Professor Mohamed Gad-el-Hak (Fig. 11.1/Plate II), Dr Jack Herring (Figs 84and 8.5), Professor Phil Hutchinson (Fig. 13.1), and DrC. J. Lawn (Fig. 1.7). My thanks are also due to Denise McCluskey who set up the laser-Doppler anemometer shown in Plate I and to Peter Tufly who took the photograph. In addition, I thank Peter Tuffy for his technical assistance in the preparation of Plate II and Figs 3.5 and 3.6. Lastly, I wish to express my gratitude to V. Shanmugasundaram for a great deal of help in the preparation of many of the figures, and to Alex Watt for his assistance with the preparation of Appendix G. ‘The author and publishers would like to thank Dantec Electronics Ltd for their generosity in sponsoring the reproduction of the colour plates in this volume.CONTENTS NOTATION 1 THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW LL The equations of fluid motion 1.2. A brief statement of the problem 1.3. The statistical formulation 1.3.1 Equations for the mean velocity 1.3.2 The energy balance equation for fluctuations 14 Two-dimensional mean flow as a special case 141 The boundary-layer equations of motion 142 The turbulent boundary layer: length and velocity scales 143 Universal mean velocity distribution near a solid surface boundary layer 1.4.5 Empirical relationships for resistance to flow through duets 1.5. Semi-empirical theoretical methods 15.1 The eddy-viscosity hypothesis 1.5.2. The Prandi! mixing-length model for flow near a solid boundary 1.53. The mixing-length model applied to a free jet 1.6 Some experimental results for shear flows 1.6.1 Mean and root mean square velocity distributions in duct flow 1.62 The turbulent energy balance 1.63 Mean and root mean square velocity distributions in a plane jet 1.7 Further reading Note References THE FUNDAMENTAL APPROACH 2.1 The Navier-Stokes equation in solenoidal form 22. The general statistical formulation 22.1. Statistical equations and the closure problem 222 Two-point correlations 23 Reduction of the statistical equations to the form for channel flow as an example 2.4 Homogeneous isotropic turbulence 24.1 Isotropic form of the two-point correlation tensor 242. Length scales for isotropic turbulence 25 Stationary turbulence xxi 13 14 16 7 18 20 20 22 4 25 28 31 3 34 34 36 36 39 40. 42 4s a 49 st 33CONTENTS 26 Fourier analysis of the turbulent velocity field, 26.1 The solenoidal Navier-Stokes equations 2.62 Homogeneity and the velocity-field moments 2.6.3 Limit of infinite system volume 2.64 The isotropic spectrum tensor 265 The Taylor hypothesis and the one-dimensional spectrum 2.6.6 Statistical equations in wavenumber space: many-time moments 26.7 Statistical equations in wavenumber space: singie-time moments 2.7 The energy cascade in isotropic turbulence 27.1 The energy balance equation 272. Spectral picture: the Kolmogoroy hypotheses 273 Interpretation in terms of vortex stretching. 28 Closure approximations 28.1 The Heisenberg effective viseusity theory 282. The quasi-normality hypothesis. 29 Some representative experimental results for spectra 23.1 One-dimensional cncrey spectra 292 The Kolmogorov constant 2.9.3 Energy transfer in wavenumber space 2.10 Further reading References, SOME RECENT DEVELOPMENTS IN THE STUDY OF TURBULENCE, 3.1 Measurement techniques and data analysis, 31 Anemometry 3.12 The laser anemometer 3.1.3 Time series analysis and computer data processing 32 Intermittency and the turbulent bursting process 3.2.1 Fine-structure intermittency 3.2.2 Intermittency and the energy cascade 323. Intermittent generation of turbulence: the bursting process 3.3 Numerical computation of turbulent flows 3.31 Direct numerical simulation (DNS) 3.3.2 Large-eddy simulation (LES) 33.3. The use of the Reynolds-averaged Navier-Stokes equations for practical calculations 3.34 Anexample of a two-equation turbulence model 34 Turbulent drag reduction by additives 3.4.1 Historical background 342. Polymer properties 3.43. The threshold effect 344 Maximum drag reduction 3.45 Drag reduction in fibre suspensions 34 38 56 38 59 61 68 65 65 o7 B B 76 n 81 3 86 86 88 88 88 92 98 190 101 104 109 112 13 8. 125 128 130 132 134 136 139 140CONTENTS 3.5. Renormalization methods and the closure problem 3.5.1 Renormalization methods in statistical physics 3.52 Renormalized perturbation theory 3.5.3 Renormalization group (RG) methods. References STATISTICAL FORMULATION OF THE GENERAL PROBLEM 4.1 Turbulence in the context of classical statistical mechanics 4.1.1 Statistical mechanics of the classical N-particle system 4.1.2 Kinetic equations in statistical mechanics 4.1.3 The difficulties involved in generalizing statistical mechanics to include turbulence 42 Functional formalisms for the turbulence problem 42.1 The space-time functional formalism 42.2 The k-space-time functional formalism 42.3 The Hopf equation for the characteristic functional 424 General remarks on functional formalisms 4.3. Test problems in isotropic turbulence 43.1 Free decay of turbulence 43.2. Stationary isotropic turbulence 44 Further reading Note References RENORMALIZED PERTURBATION THEORY AND THE TURBULENCE CLOSURE PROBLEM 5.1 Time evolution and propagators 5.2 Perturbation methods using Feynman-type diagrams 5.3. Equilibrium system with weak interactions: an introduction to renormalized perturbation theory 5.3.1 Interactions in dilute systems and the connection with macroscopic thermodynamics 5.32. Primitive perturbation expansion for the configuration integral 5.3.3 Renormalized expansion for the free energy SA The election gas: an example with long-range forces 5.4.1 Phenomenological theory: the screened potential 5.42 Perturbation calculation of the free energy 5.5 Perturbation expansion of the Navier-Stokes equation 55.1 The zero-order isotropic propagators 55.2. The primitive perturbation expansion 553. Graphical representation of the perturbation sexies 5.54 Class A diagrams: the renormalized propagator 5.55 Class B diagrams: renormalized perturbation series Notes References 14 142 144 148. 149 154 154 156 160. 165 167 169 172 174 176 176 177 179 182 182 182 184 184 186 190 191 193 199 201 202 204 205 208 213 217 218 223 223xvi CONTENTS 6 RENORMALIZED PERTURBATION THEORIES OF THE FIRST KIND 225 6.1 The direct-interaction approximation (DIA) 25 6.11 The infinitesimal response tensor 226 6.12 Perturbation expansion of the velocity field 27 6.13 Perturbation series for the mean-response and correlation tensors 228 6.14. Second-order equations for the isotropic response and correlation functions 232 6.15 Spectral transport of energy: the inertial range 235 6.1.6 The DIA energy spectrum in the inertial range 237 6.1.7 Alternative derivation of DIA by the method of reversion of, power series 240 6.18 Concluding remarks. 21 6.2. The Edwards-Fokker—Planck theory 241 62.1 ‘The derivation of the Liouville equation 242 62.2. The Edwards-Fokker-Planck equation as 62.3. Evaluation of the coefficients in the expansion for the probabuity distribution of veiocities 248 624 The energy-balance equation 250 625 The response equation 252 62.6 Comparison with the DIA 253 6.2.7 The limit of infinite Reynolds number 255 63. Self-consistent field theory 257 63.4 Time-dependent SCF 262 6.32. Other self-consistent methods 264 Note 265 References 265 7 RENORMALIZED PERTURBATION THEORIES OF THE SECOND KIND 267 7.1 The low-wavenumber catastrophe 267 71.1 The infra-red divergence 267 7.12. Spurious convection effects 269 7.13. Postulate of random Galilean invariance 274 7.14 Response integrals with an arbitrary cut-off in wavenumber 276 7.2. Lagrangian-history direct-interaction theories 27 72.1 The Lagrangian-history formulation 278 7.22. The statistical formulation 281 723. DIA adapted to Lagrangian-history coordinates 282 724 Abridged LHDI theory 286 7.25 Other Lagrangian theories 288 7.3. Modified EFP theories. 290 73.1 Maximal entropy principle 290 7.32. The response function determined by a local energy balance 294 7.3.3 Local energy-transfer equations 298CONTENTS 7.4 Local energy-transfer theory of non-stationary turbulence (LET) 74.1. The velocity-field propagator 74.2 The generalized covariance equation 7.4.3 Equations for the correlation and propagator functions 744 Comparison with DIA 7.5. Near-Markovian model closures 7.5.1 Quasi-normal Markovian appronimations 75.2. The test-field model Note References AN ASSESSMENT OF RENORMALIZED PERTURBATION THEORIES 8.1 Free decay of isotropic turbulence as a test problem 8.2. Calculations of decaying turbulence at low Reynolds numbers 8.2.1 The direct-interaction approximation (DIA) 8.22 Comparison of various theories: Herring and Kraichnan 823. The LET theory 8.3. Calculations of decaying turbulence at high Keynoids numbers 84 The Kolmogorov spectrum as a test problem 84.1 Do intermittency corrections have any bearing on ‘our assessment of RPTS? 84.2 Is the Kolmogorov 5/3 law correct after all? 8.5 Application to non-isotropic turbulence 85.1 Application of DTA to inhomogeneous turbulence 85.2 The computational problems 8.53. Other applications 8.6. Appraisal of the theories 86.1 Critique of DIA: the wider justification of RPT approaches 8.6.2 Some comments on random Galilean invariance 8.7 General remarks 8.8 Postscript: some current work Notes References, RENORMALIZATION GROUP THEORIES 9.1 Background: RG applied to critical phenomena 9.1.1 Ferromagnetism and the Ising model 9.1.2 Block spins and RG 9.1.3 Space dimension and the epsilon expansion 92. Application of RG to turbulence 9.2.1 Determination of scaling laws 9.2.2. Subgrid-scale modelling 9.3. The Forster—Nelson-Stephen (FNS) theory 9.3.1 Formulation of the problem 9.3.2 The perturbation series 300 300 303, 305 306 307 307 309 310 310 aun aul 316 310 317 319) 328 329 330 331 331 333 337 337 337 340 3at 342 343 344 346 346 347 347 349 350 350 331 353 353 355CONTENTS 9.3.3 Lhe effective viscosity 93.4 Recursion relations 933 Behaviour near the fixed point 9.3.6. Some later conjectures about FINS theory 9.4 Application of RG by iterative averaging 9.4.1 General formulation 9.4.2 Partial averaging of the small scales 9.43 The statistical equations of motion 94.4 Moment hierarchy from partial averaging 9.4.5 A mean field approximation 946 The RG equations 9.4.7 Second-order calculation of the effective vise: 948 The effect of higher-order moments 9.5 Concluding remarks. Notes References NUMERICAL SIMULATION OF TURBULENCE, 10.1.1 Isotropic turbulence 10.1.2 Shear flows 102. Large-eddy simulations 10.2.1 Assessment of subgrid models 10.3 Application of renormalization methods to the subgrid modelling problem 10.3.1 Formulation of spectral LES 10.3.2 Renormalized perturbation theory 10.33 Renormalization group 10.4 Miscellaneous simulation methods 10.4.1 Methods based on the Navier-Stokes equation 10.42. Alternatives to the Navier-Stokes equation Note References COHERENT STRUCTURES ULI Coherent structures in free turbulent flows ILL Plane mixing lavers 11.1.2 Other free shear flows 11.2 Conditional sampling, intermittency, and the turbulent-non-turbulent interface 11.3 Transitional structures in boundary layers and pipes 11.3.1 Anatomy of the turbulent spot 114 Developed structures in boundary layers and pipes 114.1 Turbulent bursts 1142 Frequency of turbulent bursts 11.43 Streaky structure and streamwise vortices 11.44 Relationship between bursts and other types of, intermittency 397 359) 361 363 364 365 307 369 370 371 373 374 377 377 378 379 381 382 385 387 388 389 390 391 397 399 400 401 403 404 406 407 407 410 410 413 ais 416 417 48, 420 423,12 13 CONTENTS 115 Theoretical approaches 11.5.1 Numerical simulation of bounded turbulence 115.2 Numerical simulation of free shear layers 11.5.3 Deterministic chaos 11.54 Wave theories 11.6 Implications for other turbulence concepts 11.7 Further reading Note References TURBULENT DIFFUSION: THE LAGRANGIAN PICTURE 12.1 Diffusion by continuous movements 12.2. The problem of expressing the Lagrangian analysis in Eulerian coordinates 12.2.1 Statement of the problem 1222 Approximations based on the conjecture of Hay and Pasquill 12.2.3 Approximations based on Corrsin’s independence hypothests 12.24 Experimental measurements of Lagrangian quantities 123 Relative diffusion 12.31 Richardson’s law 123.2. Three-dimensional diffusion 124. The motion of discrete particles in a turbulent fluid 124.1 Some asymptotic results 12.42 Tehen’s analysis 12.5. Applications of Taylor's analysis to shear flows References TURBULENT DIFFUSION: THE EULERIAN PICTURE 13.1 Heat and mass transfer 13.1.1 Statistical formulation 13.1.2. Single-point equations 13.13 Two-point equations 13.14 Some experimental measurements in pipes and jets 13.2 Scalar transport in homogeneous turbulence 13.2.1 The inertial-convective range of wavenumbers, 13.2.2 Universal forms of the scalar spectrum 1323 Renormalized perturbation theory 13.3 The motion of discrete particles 13.3.1 Governing equations for particles 13.3.2. Interpretation in terms of a diffusion coefficient 13.3.3 Random walk models of particle diffusion 13.3.4 Measurements of particle motion in turbulent flows 134 Turbulent mixing Note References 425, 426 427 428 430 431 432 433 433 436 437 4a 4a 444 446 447 449 451 452 452 433 454 456 458 460 461 463, 465 467 470 47 414 475 438 479 480 482 484 489 490 491 491x 14 NO 14a 142 143 CONTENTS N- NEWTONIAN FLUID TURBULENCE Non-Newtonian fluid flow 14.1.1 Rheological aspects 14.1.2 Composite systems: Newtonian fluid with modified boundary conditions 14.1.3. Flow in pipes 14.1.4 Structural turbulence 14.1.5 Isotropic turbulence Turbulent structure in drag-reducing polymer solutions 142.1 Mean velocity distributions 14.2.2. Turbulent intensities 14.23 Spectra and correlations 14.2.4 The importance of the region near the wall 142.5 Free turbulence Turbulent structure in drag-reducing fibre suspensions 143.1 Mean velocity distributions 14.3.2 Intensities, correlations, and spectra 14.3.3 Mixed fibre-polymer suspensions 144 The effect of dray-reducing additives on turouient transport 145 146 144.1 Heat and mass transfer 144.2 Turbulent diffusion ‘Comparison of polymers and macroscopic fibres as drag-reducing additives Further reading References APPENDICES, Author Subject I Creation and dissipation of kinetic energy in a viscous fluid Probability and statistics ‘Symmetry and invariance Application of Fourier methods and Green’s functions to the Navier-Stokes equation Evaluation of the coefficients L(k,j) and L(k, k — j) Optical background to laser-Doppler anemometry Second-order term in the perturbation series as an example of the diagram calculus The Novikov functional formalism index index Plates J and II fall between pp. 92 and 93 494 494, 495 497 498 00 501 505 506 508 509 Sut 312 312 514 317 318 518 519 520 521 522 524 527 532 536 Sa sas 350 554 357 565NOTATION 1 General remarks Vectors, matrices, and tensors are represented by the standard Cartesian notation, i. A,, By, Cap,, and so on. In all cases we use Greek letters 2, B, 7, 6,....for the tensor subscripts; each subscript can take the value 1, 2, or 3. In certain cases, vectors are shown in bold face type. ‘This 1s usually the case where the vector concerned is the independent variable, i.e. U,(x,1) or even F[U(x,0)]- In the latter example, the square brackets are used to indicate that F isa functional or ‘function of a function’. We shall usually, but not always, employ the summation convention, in which repeated tensor indices are taken to be summed without the summation symbol being needed, For example, the scalar product of two vectors X and Y can be written as a X-Y=S XY, a or as XY=X,¥, All averages are denoted by Dirac brackets < ), although an overscore is sometimes used for single variables. Thus the mean velocity can be written as U =U). A local (or partial) average is introduced in Chapter 9, and this is denoted by < Yo. Position vectors are written as x, x’, ... for different positions, while times of different events are usually written as t,1', 0", Inconsistently. distinct wavevectors are normally written as k. j kkk 2 symbols a semi-width of a plane channel, radius of spherical particles Ay cross-sectional area of added particles & friction (or drag) coefficient for external flow round a solid body Cc empirical constants in “two equation’ turbulence models CoG C\-Cy constants specifying the trial spectra, as used in Chapter 8 a diameter of pipe, diameter of diffusing particle, width ofa plane jet Dit, Di. structure functions D diffusivity due to molecular motion Dy eddy diffusivity for fluid pointsxxii Re(t) Rutt) (x,0) Uy U(X 0, U,(k,1) 1(k, 0) NOTATION eddy diffusivity for added particles diffusivity associated with a macroscopic random walk total kinetic energy of fluid motion kinetic energy of fluctuating motion per unit mass of fluid energy spectrum flatness factor of a probability distribution friction factor for pipe or channel flow wavevector (or position coordinate in k-space) Kolmogorov dissipation wavenumber thermal conductivity, as used in Chapter 13 integral length scale scalar concentration field mean scalar concentration field fluctuating scalar concentration field reference level for scalar concentration field instantaneous pressure field mean value of the pressure field fluctuating pressure field spectral density stationary spectral density relative coordinate, r = (x — x’) Reynolds number for pipe flow, centroid coordinate R = (+ x2 Reynolds number for boundary layer flow microscale-based Reynolds number (or Taylor-Reynolds number) correlation coefficient Eulerian time-correlation Lagrangian time-correlation deviatoric stress tensor skewness factor of a probability distribution Eulerian integral time-scale Lagrangian integral time~ Lagrangian integral time- inertial transfer spectrum bulk mean velocity in pipe or channel flow free stream velocity in external flows centre-line velocity in pipe or channel flow instantaneous velocity field fluctuating velocity field Fourier transform of the velocity field Fourier transform of the fluctuating velocity field ale ale for diffusing particlesvo ¥(0) woe w(kst—¢’) autocorrelation of external stit oe xo XU) NOTATION xxiii Fourier components of the velocity field restricted to a band of wavenumbers 0 < k : two-point, two-time correlation of two velocities Ove, Y= Cul heal’. tu, (0". 0°)... Myp,(k) = (21) {kpDay(k) + k, Dyg(k)}: inertial transfer operator occurring in the solenoidal Navier-Stokes equation Djp(k) = 5,9 — kaky/Ik|?: projection operator Ctug(K, p(k’, t')> = Q2A/L) Ogex,oQap(kst,t defines the spectral density tensor for finite system volume Cine, p(k’, t')> = d{k + K’)Qup(k; 6,0"): defines the spectral density tensor for infinite system volume Qap(k: t,t’) = Dyg(k)Q(k; 1,1"): isotropic spectrum tensor lk; 1,1) = Q(k,t): single-time spectral density function Q(k; 0) = qik): stationary single-time spectral density E(k, 1) = 4nk?Q(k, 0): the energy spectrum E(k) — Ank2q(le: the stationary energy spectrum (fll Dfy(—Ks1')> = Daplis)w(k;,0’): defines autocorrelation of the stirring forces in isotropic turbulence w(k;t — ’) = W(08(t — 1"): defines the correlation function W(k) for stirring forces with delta function autocorrelations beeen” kg — (e/v3)"*: Kolmogorov di sipation wavenumber v = (e)"*; Kolmogorov dissipation-range velocity scale ky = (e/vD?)"**: Batchelor wavenumber ke = (e/D?)" diffusion cut-off wavenumber1 THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW Turbulence, a phenomenon which is complicated in itself, can be encountered in a wide variety of more or less complicated situations, both in the natural environment and in many industrial processes. Thus, when engineers attempt to predict the behaviour of turbulent systems, they have more to reckon with than the intrinsic difficulty of the turbulence problem. In order to carry out design calculations, they must also tackle the problems inherent in describ- ing fluid flow through complicated physical systems such as turbine rotors, chemical reaction vessels, or tube bundies in heat exchangers. Nevertheless, as computational methods improve and computers grow in power, there is continuous progress in treating the mereiy compiicated aspects of problems. Therefore, in many engineering applications nowadays, the major problem faced is the irreducible one of the turbulence itself. It hardly comes as a surprise, therefore, that a survey of the literature of turbulence reveals the classical symptoms of a subject that is of immense practical importance, yet is at the same time poorly understood. That is to say, there is much activity but relatively little consensus. Also, there is a strong tendency to form separate subject schools, each related toa particular industry or type of phenomenon, but with relatively little communication between them, Above all, there is a great variety of predictive methods, ranging from simple empirical correlations, through rigorous (but limited) deductions from the Navier-Stokes equations, to claborate statistical models which are based on dubious analogies between the chaotic behaviour of turbulence and the molecular chaos of dilute gases. In this book we are interested only in the physics of the turbulence itself, and not in engineering complications. For this reason we shall restrict our attention to flows with a rather simple geometrical configuration, such as free jets and wakes, the boundary layer next to a flat plate, and flows through pipes and plane channels. These are the classical flows upon which the general subject of fluid dynamics has very largely been based. Each of them has the virtue of allowing a general formulation of the equations of motion to be greatly simplified. They therefore form an important special case, and we shall refer to them generically as ‘two-dimensional mean flows’ (As an aside, we should perhaps emphasize that only the mean velocity is two dimensional in such flows; the turbulent fluctuating velocities are fully three dimensional.) An additional simplification can come about if the flow configuration is2. THE SEMI-EMPIRICAL PICTURE OF TU BULENT SHEAR FLOW narrow (in comparison with its length in the flow direction), and the cross- stream component of velocity is small compared with the streamwise com- ponent. This allows a further approximate reduction of the equations of motion, in what is known as the ‘thin shear layer’ or ‘boundary-layer’ ap- proximation. The first of these terms is undoubtedly the more generally correct, but the second is much more commonly employed. We shalll follow custom in this matter. Our aim in this chapter is to present a concise summary of the formulation of the turbulence problem from the point of view of engineering applications. We begin with the equations of fluid motion and take averages in order to form an equation for the mean velocity. At this stage correlations are restricted to values taken at a single point in space and timc, as opposed to the multipoint many-time correlations of the fundamental approach in Chapter 2. We immediately encounter the fundamental closure problem of turbulence, in that the mean velocity can only be calculated if the two-velocity correlation is also known, and consider some of the traditional engineering models used to overcome the problem in the context of the special case of two-dimensional mean flows. Although the statistical equations presented are not soluble, we show how they can be used to analyse and interpret experimental results. Not least, we show that simple empirical correlations can be used, not just as the only really reliable predictive methods, but also in order to indicate the presence of universal behaviour. 1.1 The equations of fluid motion Throughout this book we shall only consider fluid motion which can be regarded as incompressible. For a general discussion of the conditions under which this will be true, reference can be made to Batchelor (1967). For our present purposes, it amounts to a requirement that the fluid density p always remains constant, and the equation of continuity (which expresses conserva- tion of mass) takes the form Ug (8) oxy where U,(x,t) is the fluid velocity at position x and time t. Note that we shail almost invariably use Cartesian tensor notation, and that Greek indices such as x, f, of 7 take values of 1, 2, ot 3. We shall also employ the summation convention in which, as in eqn (1.1), repeated indices are summed For an incompressible fluid, the equation expressing conservation of momentum is dt) (12)Ll THE EQUATIONS OF FLUID MOTION 3 where s,, is the deviatoric stress tensor. For a Newtonian fluid, s,, is given by ou, aU, sua or(’ f45 *) (13) Oxy” Ox, where v is the kinematic viscosity of the fluid. As a specific example, let us consider steady shearing flow between infinite plates which lie parallel to the (x, x3) plane. The flow is taken to be solely in the direction of x,, and the velocity field reduces to U,(x, 1) = {U,(x2).0,0} Under these circumstances, the non-linear term in (1.2) becomes {Us (¢2)U yl) } and the conservation of momentum equation takes the form oP _ ass tox (1.2a) Note that the vanishing of the non-linear term is a general property of unidirectional laminar flows (Batchelor 1967), and that in this particular situation the momentum equation describes a balance between the streamwise pressure gradient and the resistive viscous stresses. For this flow, the viscous stress tensor can be obtained from eqn (1.3) as 27 o (13a) a result which is usually known as Newton’s law. This law provides both a definition for and a method of measuring the dynamic viscosity = py. Phy- ally, the effect of viscosity can be interpreted as an irreversible flux of streamwise momentum in the direction of x, With the substitution of s,, from (1.3) and the use of (1.1), the resulting equation of motion is known as the Navier Stokes equation: aU, , aU, a” Ox, : s Lap eemeoe Wile (14) pO, where V? = 07/éx,dxp These equations can be derived using the methods of continuum mechanics (eg. Landau and Lifshitz 1959; Batchelor 1967) or, more restrictively, from kinetic theory (e.g. sce Reichl 1980). They describe the motion of many common fluids, ouch ae ater, alcohol, glycerine, air, and mast gases, provided only that the density remains constant. Many other fluids (e.g. particulate suspensions, polymer solutions) require a more complicated constitutive rela- tionship between stress and rate of strain than the linear form given by eqn4 THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW (1.3). The subject of non-Newtonian fluids will be considered later in Chapter 14 which deals with turbulent drag reduction by additives 1.2 A brief statement of the problem ‘The scientific study of turbulence is generally taken as having begun with the work of Osborne Reynolds (1883). The problem which Reynolds studied was the classic one of flow through long straight pipes of constant diameter and circular cross-section. Using his ‘method of colour hands’, he was the first person to show that, for a given fluid and pipe, the flow would be orderly (laminar) for velocities below a certain critical speed. At the critical speed, the flow abruptly became turbulent at some distance from the pipe entrance. Above the critical speed, turbulence was found to be the normal state of the flow, although laminar flow could be maintained as a metastable state by carefully eliminating all disturbances or perturbations. Reynolds found that the criterion for the transition from laminar to turbu- lent flow could be expressed in general (universal) form in terms of the value taken by dimensionless group R=Udjv (15) where R is what we now call the Reynolds number. Here d and U are representative length and velocity scales, in this case the diameter and the (bulk) mean velocity. The latter quantity is obtained by measuring the vol metric flow rate and dividing by the cross-sectional area of the pipe. With this, definition, the experiments of Reynolds indicated that the minimum value of R for which turbulence could occur in a tube was about 2000. However, laminar flow in a pipe can be metastable at much larger values of the Reynolds number than this. A fuller, although still concise, discussion of transition to turbulence will be found in Goldstein (1938, pp. 69-74). If we choose pipe flow as a specific problem on which to focus our attention, then we shall find it helpful to subdivide the basic theoretical problem into two parts: (1) we would like to solve the Navier-Stokes equation for the critical value of the Reynolds number at which the transition from laminar to turbulent flow occurs; (2) we would also like to solve the Navier-Stokes equation for the mean values of quantities like the velocity and the pressure, for Reynolds numbers larger than the critical value. 7 be q atic, At this point it w: premature to try to offer any more fundamental justification. However, it does reflect the way in which the subject is organized. The two problems are normally treated in the literature as quite separate lopics.1.2 A BRIEF STATEMENT OF THE PROBLEM 5 20 nar profile with same U as the turbulent case Turbulent 05 R=S.x10" Fig. 1.1. Comparison of laminar and turbulent mean velocity distributions in pipe flow at the same Reynolds number (R = 5 x 10°), Of the two, the second is overwhelmingly the more important in practical terms. Again, this shows up in the literature, although the subject of transition has received much more attention in recent years, mainly because of its relationship to other critical phenomena and its relevance to deterministic chaos. We make some brief comments on these developments in Chapter 11, Section 11.5.3. However, that section apart, we shall concentrate throughout on the second problem: the need to obtain a statistical description of fully developed turbulence. Let us now consider flow through a pipe when the Reynolds number is well, above the critical value. In Fig. 1.1 we show the distribution of the mean velocity with radial position in the pipe. The mean flow is taken to be in the x; direction and xz is the transverse or radial coordinate, For purposes of comparison, we have also plotted the equivalent laminar velocity profile for the same Reynolds number. This is, of course, the well-known parabolic profile which can be calculated directly from the Navier-Stokes equation. A not unreasonable goal for a turbulence theory would be to perform the analogous task for the mean turbulent velocity profile. For the present we shall take mean values to be given by time averages and, for the particular case of the mean velocity, we shall show this by an overbar. In general, the operation of taking averages (however defined) will be repre-6 THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW sented by angular brackets < ), and these brackets will be used to indicate higher-order moments. Thus, for the mean velocity, we have TyGa,0) = Usb) = 54 i} Usbo t teyde’, (1.6) oT. where 211s the averaging time which must be long enough to smooth out the turbulent fluctuations but short enough not to average out any imposed time dependence, However, for the most part, we shall only consider flows in which the mean velocity is constant in time. We can now take a preliminary look at the task of solving the Navier Stokes equation for the mean velocity profile for steady (mean) flow through a pipe. As we have already seen for the laminar case, eqn (1.2) reduces to the particularly simple form of (1.2a) for steady flow. Similar considerations apply to the equation for the mean velocity in turbulent flow; however, as the fluctuations in velocity are not unidirectional, the mean value of the non-linear term is not zero (the appropriate generalization of eqn (1.2a) to turbulent flow be found in Section i.4. nis average contribution from the turbuient fluctuations gives rise to an additional resistance to flow. It also raises the central problem of the statistical theory of turbulence, as encountered by Reynolds (1895). We can give a general (if rather simplified) explanation of this as follows. Let us rearrange eqn (1.4) to obtain which can then be written in the following highly symbolic fashion: LoU=L,UU +L,P where Lo, Ly, and L; stand for the respective differential operators, and we have temporarily suppressed the tensor indices for simplicity. Now average each term to obtain an equation for the mean velocity: Lo. At this stage we should note that P is related to U through the continuity equation: we shall discuss the implications of this in Chapter 2. For the present it implies that a solution for (U’) depends (in principle) only on the second- order moment (UU). ‘An equation for (UU) is readily obtained by multiplying each term of (1.4) by U and averaging, Lo(UU) = L,(UUU) + L, = 0. (19) Physically, this result has the simple interpretation that, on average, the fluctuations are as often positive as they are negative. Correspondingly, the average of the square of a fluctuation will not vanish and we can introduce the root mean square (t.m.s) value u;, as follows: yl, t) = uz(x, >"? (1.10) In practice wu, is often used as a convenient measure of fluctuation intensity. We should remind ourselves that the time dependence exhibited by the above mean quantities refers to slow external variations and is only retained here (and at other points) for generality. When we actually come to specific examples of real flows, we shall restrict our attention to the stationary case. 13.1 Equations for the mean velocity Let us first consider the continuity equation, as given by (I.1), If we substitute (1.7)for the mean velocity, and average according to (1.6) and (1.9), we obtain BU g(x, 0) Oxy (Lip8 THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW Subtracting this result from (1.1) yields a similar equation for u, and so the ‘mean and fluctuating velocities separately satisfy the continuity equation. This is, of course, a trivial consequence of the linearity of eqn (1-1). In order to treat the equation of motion the same way, we substitute (1.7) and (1.8) into (1.4) and average term by term. It is easily shown that the operation of time averaging commutes with the operation of differentiating with respect to time." This time we have the non-linear term to deal with, and so, noting that Tuy =, we find that aU, Uz, | é 1 OP Ae Get Gag U8 uy Male =F By + Us (1.12) Comparison with (1.4) shows that the equation for the mean velocity is just the Navier-Stokes equation written in terms of the mean variables, but with the addition of the term involving dv (1.18) where, again, we have used the relationship (Ou) = 0. Normally we shall only be interested in the energy associated with the velocity fluctuations u,(x, ). We can derive the appropriate balance equation, via an equation for the Reynolds stress tensor, as follows. Consider the equation for the fluctuating component u,(x,¢) as given by (1.14), Rewrite this as an equation for u,(x,1), Keeping all other indices and variables the same, and give it the notional equation number (1.14a). Now multiply (1.14) through by u,(x,1) and average, then multiply (1-14a) through by u,(x,) and average, and, making repeated use of the rule for differentiating a product, add the resulting equations together to obtain 5 Stuy u,) + Oe 5 kat + ae) Bee au, 2 + duty + a City ty (tH) an * a Sot ou oe +(u SEV a wlcuvia) + 97H}. (119) The viscous terms can be simplified (see the comparable procedure in Appen- dix A) and the rest of the equation rearranged a little to give 5 ets) + Oy = Oy -f co el\ \Oxp OX s / (1.20) The left-hand side gives the total rate of change with time (ie. local plus convective time derivatives) of the kinematic Reynolds stress tensor u,u,). 1, serves to ptt) > Seeves to remind us that the moment closure problem is still with us. If we set 2 = 7 in the above equation then we obtain the equation for the mean square excitation for each component of the fluctuating velocity as14 TWO-DIMENSIONAL MEAN FLOW AS A SPECIAL CASE U1 follows: A ODay aay see arg ee Gees a BD + Una ax, Cau) +74 *P>§ B c D a a hk ~ 2d tr Egad) — 2 ( *) } a2) Furthermore, if we sum over 2, then (1.21) gives us the balance equation for the turbulent kinetic energy E per unit fluid mass, which is defined by 2E = «2d. (1.22) Clearly, the only manipulation req u2> on the left-hand side (Lh.s.) by E. In words then, eqn (1.21) tells us that the total rate of change of turbulent energy with time is given by the net effect of the terms on the right-hand side (c.hs). To interpret these we note that the first and third terms (A and C) can each be written asa divergence and accordingly do not contribute to the global energy balance (see the analogous step in Appendix A on going from eqn(A.10) to(A.12)). Thus their physical effect is the diffusion of turbulent energy through space by non-linear and viscous actions respectively, The difficulty with the second term B is that apparently it cannot be written, as a divergence. However, the reader may find it instructive to form an energy-balance equation for the mean velocity, where he will find a corre- sponding term, The two terms, taken together, can be integrated and shown to conserve energy jointly. Thus the term B on the rhs. of (1.21) can be interpreted as a flow of energy from the mean field to the fluctuating velocity field. It is often referred to as the ‘production term’ (see Hinze (1975) for a fuller discussion). Obviously the last term D on the r.h.s. represents the irreversible dissipation of kinetic energy into heat, A more detailed discussion of this, and the other terms described above, will be found in Appendix A. Finally, we should note that (1.21) cannot be solved: the moment closure problem rules that out. However, the individual terms can be measured experimentally and the energy balance studied in this way. We shall return to this subject in Section 1.6.2, d iy to divide across by 2 and replace 1.4 ‘Two-dimensional mean tlow as a spectal case Our general problem is to solve the equations of mean motion for turbulent Hows. However, if we consider very simple flows, we can reduce the number12. THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW of variables involved. We can do this by means of symmetries (a rigorous method) and by ‘order-of-magnitude’ arguments (an approximate method), In this way the overall size of the problem can be reduced, although its intrinsic difficulty is unaffected. ‘As mentioned earlier, we shall consider two-dimensional mean flows, and historically the most important example of these is the boundary layer. This concept was introduced by Prandtl in 1904 (e.g. see Schlichting 1968), and takes its simplest form when a fluid flows at zero incidence over a flat plate. To be specific, we shall take x, to be the direction of flow (this will be our convention throughout this book) and the plate to lie in the (x, x3) plane. The plate is also supposed to be of infinite extent in the x3 direction and the incident stream has velocity U,,, which is constant in time and uniform in space. With these restrictions the free stream is said to be an irrotational flow. That is, its vorticity (curl U) is everywhere zero. However, when the fluid encounters the plate, the ‘no-slip’ boundary condi- tion ensures that vorticity is generated where the fluid is in contact with the surface. As the fluid moves downstream, this vorticity diffuses outwards (at right angles to the plate) with its rate of diffusion controlled by the viscosity of the fluid, In this way we obtain a physical interpretation of the idea of a boundary layer; that is, the distance travelled by vorticity away from the plate in the x, direction is a measure of the boundary-layer thickness. If we take the leading edge of the plate to lie along the x, axis, then x, is distance measured along the plate and x, is distance measured perpendicular to it, We restrict our attention to one side of the plate (x, > 0) and consider the evolution of the boundary layer for x, > 0. It is known from experiment that the boundary layer is (in the absence of disturbances) laminar for small values of x, but, as downstream distance increases, it passes through a short transition length and thereafter is fully turbulent. It is conventional to define the Reynolds number (based on distance along the plate) as Rx, = U,x1/¥. With this form of Reynolds number, transition to turbulence normally occurs when Rx, = 3.2 x 10°. (As an aside, it may be of interest to note that, if the Reynolds number for the boundary layer is based on the thickness of the layer, then the value at transition is much closer to that for pipe flow based on diameter.) The great simplification introduced by this concept is that the Navier- Stokes equations need only be solved inside the boundary layer where ap- proximations can be made, In the laminar case, this means that we only have to solve for U, and U;,as functions of x, and xz,on theinterval 0 < x3 < 6(x) where the boundary-layér thickness is arbitrarily defined by the relationship x2 = (x) when U,(x,,x3) = 0.99U,. (1.23) From symmetry considerations, it follows that U, = 0 and that variations of mean quantities with x3 can be neglected. The problem is then further14 TWO-DIMENSIONAL MEAN FLOW AS A SPECIAL CASE 13, reduced by order-of-magnitude arguments based on the smallness of 6(x,) in comparison with the length of the plate in the streamwise direction. For instance U; « U,, and variations with x, are much larger than those with x, Finally, the existence of similarity solutions of the form Uy (1,2) = Ux fl2/d), (1.24) where f is a universal function, reduces the dimensionality still further. These ideas can be extended to the turbulent part of the boundary layer by applying the same arguments to the mean velocity U,(x) — (U,, U3, 0). Of course the fluctuating field u,(x, 1) is fully three-dimensional, but even here derivatives of mean quantities with respect to x, vanish, as do off-diagonal elements of the Reynolds stress tensor ix (1.26) for conservation of mean momentum in the x, direction. The latter equation is approximate, but a further approximation can be made, Experimental results suggest that the streamwise variation of (u?) can be neglected, provided that the pressure gradient is not too large. The momentum equation for the x, direction reduces to the condition that the pressure variation between the boundary and the free stream is of the second order of small quantities. Hence the streamwise pressure gradient in the boundary layer can be found approximately from the Bernoulli equation applied to the irrotational flow outside the boundary layer: P + pU2/2 = a constant along a streamline, or sie Ue ep eae14. THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW It follows that the pressure gradient in (1.26) can be put equal to zero for the case where the free stream is parallel to the plate, U,, is constant, and there s no externally applied pressure gradient, 14.2 The turbulent boundary layer: length and velocity scales In Fig. 1.2 we show a schematic view of a turbulent boundary layer in a plane normal to the flow. The first point to be noticed is the irregularity of the outer edge. This means that we must work with the average value of the boundary- layer thickness, and so this is what we shall mean when we refer to 6(x,), or indeed to any of its various subdivisions, which we shall now consider. First of all, it is usual to divide the turbulent boundary layer into an inner layer for (approximately) the rangeO < x, < 0.25, andan outer layer bounded by 0.25 < x) < 5, This is based on the experimental observation that the total shear stress 1,> is nearly constant over the inner layer and is approximately equal to t,., its value at the surface (or wall). ‘The inner (or constant stress) iayer can be furiier subdivided according iv the relative magnitude of the viscous and turbulent parts of the total shear stress. In the present case, eqn (1.13) takes the simple form de (1.27) where the viscous part is given by Newton's law applied to the mean rate of Mean edge Instantaneous a edge I ' Outer — region | inner | region resscnocececes sone cgeeseeseeseercecreereeee Viscous sublayer Fig. 1.2. Schematic view of a turbulent boundary layer (not to scale; flow direction normal to the page).14 TWO-DIMENSIONAL MEAN FLOW AS A SPECIAL CASE 15 strain and the turbulent contribution is just the appropriate component of the Reynolds stress tensor. Near the wall, the boundary conditions on the velocities, 1, u > 0 as x) 70, imply that the product u, u; vanishes rapidly as the wall is approached. Hence the stress at the wall is entirely due to viscous shear and can be expressed as a0, = prs 12 - Oe 028 ‘We can now define the viscous sublayer as the region next to the wall where the first term on the rhs. of (1.27) is dominant. For larger values of x2, the second term on the r-h.s. of (1.27) will ultimately become dominant, and this region is usually referred to as the fully turbulent constant-stress layer. Evidently there will be an intermediate region where the two stresses will be of equal magnitude, and this is called the transition sublayer (or, often, the buffer layer). The physical extent of each of these sublayers is most conveniently ex- pressed in terms of the so-called ‘inner layer variables’, which can introduced as follows. Dimensional analysis (confirmed by experiment) velocity scale for the inner region is given by 4, = (tw/p)*7, (1.29) which then allows the relevant length scale to be written as, inner layer’ length scale = v/t, (1.30) where u, is called the ‘friction velocity’. The rather odd name can make this a puzzling concept for, at first sight, it seems to have little to do with any physical velocity. However, it is hallowed by tradition and, as we shall sce, it is of the same order of magnitude as the r.m.s. value of the velocity fluctuations. With (1.29) and (1.30), it is usual to define scaled variables for the inner region as, + Xaly % (1.31) This amounts to measuring distance from the wall in units of v/u, and, for small v (ie. large Reynolds number), it has the effect of expanding the wall region. The experimental results (Lo be discussed in more detail in Section 1.6) then inner layer 0 < x; < 0.26 outer layer 0.26 < xz < 1.06.16 THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW The inner layer is subdivided as follows: viscous sublayer 0 < xf <5 transition sublayer 5 < x} < 30 turbulent constant stress layer 30 < x}. It should be noted that the values chosen above for x, or x} to classify the various layers will vary from one source to another in the literature. This, of course, reflects the difficulty in establishing precise criteria for the boundary between any two sublayers. 1.43 Universal mean velocity distribution near a solid surface Phenomenological theories for the mean velocity have been much helped by experimental observations of scaling behaviour in the sort of simple situation. that we are considering here. For instance, in the inner region of the boundary layer, measurements of mean velocities can be reduced to the universal form ft = fed), (1.32) which is known as the ‘law of the wall’, (We are working on the assumption that the wall has a smooth surface. If the height (however defined) of any roughness elements on the wall is less than the thickness of the viscous sublayer, the surface is said to be “hydraulically smooth’, and eqn (1.32) holds. If the roughness height is greater than the sublayer thickness, then it is the roughness height which determines the inner-region length scale.) However, for the outer region, experimental results take the self-preserving form = Us = glx2/) (1.33) which is known as the ‘velocity defect law’. The functions f and g can be determined (at least for much of the boundary layer) by requiring there to be a region where the two forms (or, rather, their first derivatives) are continuous. A detailed treatment is given by Hinze (1975). The result is that f and g must be logarithms, and that eqn (1.32) becomes Uf = Aln(x) + B, (1.34) where A and Bare constants which have to be determined from a comparison of (1.34) with experimental results. periment, to such an extent that it is virtually accorded the status of a law of nature by fluid dynamicists. Yet, as we shall see, it breaks down both near the tribution is strongly supported b:14 TWO-DIMENSIONAL MEAN FLOW AS A SPECIAL CASE 17 wall and in the outer parts of the boundary layer. (It is of course evident that eqn (1.34) cannot satisfy the boundary condition U, = 0 at x = 0.) However, we can establish the limiting form of the mean velocity at the wall by considering eqn (1.27) for the total shear stress when restricted to the viscous sublayer. Recalling that the total stress is constant in this region (c12 = t,) and that the Reynolds stress tends to zero, we obtain the result av, avg forad <5 (1.35) Then, integrating with respect to x2, and using (1.29), we find or, in scaled variables, ntegration has been the boundary condition. This linear law only applies to the viscous sub- layer and, as we shall see in Section 1.6, has received ample experimental confirmation. 1.44 A simple way of culculuting the friction drag due to a boundary layer ‘The simplest possible practical calculation method is to follow the approach taken with laminar boundary layers and use the equation of motion to derive an integral relation based on a similarity solution of the type G,(e., x2) = Uz hte/) (1.38) where h is chosen to satisfy the boundary conditions. Setting the gradients of P and (u?) to zero, and integrating each term of eqn (1.26) with respect to x2, then yields (e.g. see Goldstein 1938) the von Karman momentum-integral equation 1 = Ua {{. (t= ihaxs (1.39) In the analogous laminar case, we would fit h by a low-order polynomial, solve (1.39) simultaneously with (1.28) for the viscous shear stress to obtain 5, and hence obtain the wall shear stress as a function of x,. This is not possible in the turbulent case, mainiy because the shape of the mean velocity profile is not easily represented by a polynomial. For this reason, we are forced to take two rather arbitrary steps.18 THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW First, we introduce an empirical formula for the shear stress at the wall. This is due to Blasius (Goldstein 1938, p. 340) and takes the form Ty = 0.023pU2 (ca) : (1.40) The r.h.s. of this expression can be equated to the r-h.s. of (1.39) in order to solve for the boundary-layer thickness. The second step is to use the empirical mean velocity profile given by h = (x,/5)"". (141) The value of n depends on the Reynolds number and, for example, for mod- erate Reynolds numbers (Rx; © 107) we can take n = 7. If it is assumed, for simplicity, that the boundary layer is fully turbulent from x, = 0, the above procedure yields the result u,\"5 dts,) = 038(%*) Fd (1.42) and the drag force can be obtained by integrating the r-hs. of either (1.39) or (1.40) over the surface of the plate. Alternatively, the result of the calculation is often expressed in terms of the friction (ot drag) coefficient, as defined by the relationship pUz Thus even the simplest practical calculation of a turbulent flow depends on some prior input from experiments. It has to be said that, at the time of writing, this is true of all practical methods used in engineering calculations. G (1.43) 1.4.5 Empirical relationships for resistance to flow through ducts In the preceding sections we have been concerned with developing flow, where mean quantities vary (however weakly) with the streamwise coordinate x,. We shall devote this section to a brief consideration of well-developed flows, in which mean quantities are independent of x,. In particular, we shalll discuss the two-dimensional (mean) channel flow obtained by considering a second plate in the (x, x3) plane which is placed at x, — 2a, above our original plate at x, = 0, Then, sufficiently far downstream (ie. for large values of x,), where the upper and lower boundary layers have merged, the turbulent flow will be weii developed. The approximations used in deriving eqn (1.26) remain valid, but we can now simplify this equation further by putting the gradients of U, and (x2 =a) ay: (1.46) However, at the wall (e.g. for x, = 2a) we have the important result dP =a La7] waa (L47) which offers a simple method of determining the shear stress at the wall in terms of two quantities which are readily measured. Also, we can adapt the definition of the friction coefficient (see eqn (1.43)) to duct flows if we replace the free-stream velocity by the bulk mean velocity U. In this case the friction coefficient is renamed the friction factor f and eqn (1.43) becomes Sf = 24/pU? (1.48) In fact, we should mention that many of the deductions in preceding sections about the boundary layer at zero incidence—especially the mean velocity distributions—apply equally to flow in ducts. We can extend these results to pipe and channel flows by replacing the outer-region variables U,, and 6 by the bulk mean velocity U and by the radius (circular pipes) or the semi-width (rectangular channels). Lastly, for completeness, we should note that, for flow through ducts, there is an alternative to the Blasius formula as given by eqn (1.40), This is the logarithmic (or Prandtl~Karman) law for the resistance to flow through ducts. IL lakes the form (Goidsicin 1938, p. 338} Lf? = 40 log,o( Rf) — 0.40 (1.49) where f is the friction factor as given by eqn (1.48)20 THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW 1.5. Semi-empirical theoretical methods Since the early days of research into turbulence, there have been many at- tempts to reconcile the basic ideas that underlie the kinetic theory of gases with the continuum concepts (especially vorticity and vortex motion in gen- eral) encountered in macroscopic fluid motion. The result has been many theories of turbulence based on analogies between the chaotic motion of eddies and the random motion of molecules in dilute gases. The mixing-length model of Prandtl (e.g. see Schlichting 1968; Hinze 1975) is perhaps the best known, and provides us with a representative example to discuss here. We begin by considering the associated concept of an effective turbulent viscosity. 15.1 The eddy-viscosity hypothesis The notion that the collective interaction of eddies can be represented by an increased coefficient of viscosity is an attractive one. Traditionally it is intro- duced by analogy with the result from kinetic theory (e.g. as in eqn (1.3a)): : aU, mean viscous shear stress = pys-+, dx, It is then tempting to guess that the turbulent shear stress can be written in the analogous form a0, os (1.30) PKU a> = pvr(%2) where vp(1,) is the kinematic eddy viscosity. Despite the fact that the apparent unsoundness of such analogies was clear even to the earliest workers in turbulence, this hypothesis still receives much critical attention. At a later stage we shall be considering the support given. to the idea of eddy viscosity by recent developments in the theory of renor- malization, albeit with some restriction. At this stage, we merely note the pragmatic criticism that, in flows where dU, /dx, and (u, u, > do not have their zeros at the same point, the eddy viscosity (as defined by (1.50)) may be either zero or infinite at certain points in the flow (e.g. Bradshaw 1972). Evidently —if ‘we were to pursue an analogy with the subject of continuum mechanics, rather than kinetic theory—the ‘constitutive relationship’ for turbulence must in general be more complicated than the purely ‘Newtonian’ one implied by eqn (1.50). 1.5.2. The Prandtl mixing-length model for flow near a solid boundary The mixing-length model is a more ambitious attempt to build on analogies with kinetic theory. We start from the recognition that the Reynolds shear1.5. SEMI-EMPIRICAL THEORETICAL METHODS a stress p(u,u,) represents a flux of xy-momentum in the direction of x, Prandtl assumed that this momentum was transported by discrete ‘lumps’ of fluid, which moved in the x, direction over a distance ! without interaction (ie. momentum is assumed to be conserved over distance !) and then mixed with existing fluid at the new location. Clearly |, which is called the mixing length, is supposed to play the part of a mean free path in this process. The essentials of the analysis are then as follows. A fluid element dV is carried from x, to x; +1 by a velocity fluctuation uz. It carries net x,- momentum to its new location, owing to the difference between U,(x;) and U(x, +1). The result is a fluctuation in x;-momentum and hence in x,- velocity u, This can be expressed as pu, AV = p[O,(x2) — U(x, + N1dV (151) to first order in i, and ence u,=—1 (1.52) We note that U, is an increasing function of x,, and so a fluid ‘lump’ moving in the direction of positive x, (ie. one which corresponds to a positive fluctuation u,) causes a negative fluctuation in w,. Thus the Reynolds stress will be negative, and the correlation can be written in terms of the rms. velocity components as (uit) = —Ryat = -Ch> (1.53) where R,2 is the correlation coefficient. The second step follows from the experimental observation that 13 is of the same order of magnitude as u, in the constant-stress layer. Also, Ry, has been absorbed into the constant of proportionality C. Hence, from (1.52), ‘av, \? —pP(— 1.54) a (=) (1.54) where now the constant C has been absorbed into I At this stage we need three further assumptions: pKa a> (a) in the constant stress laver we can put t15 = Tai (b) for xf > 5, we can neglect the viscous term in the shear stress; (©) = kx, where k is known as the von Karman constant. Then, using (1.27) and (1.29), eqn (1.53) becomes22. THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW ! Gy, (155) and hence Ww, (1.56) Finally, taking the square root of both sides and integrating with respect to xz, We obtain the logarithmic profile in the form a, = Fine, + D (1.57) where D is a constant of integration. This result can be matched to the linear profile (see eqn (1.37)) by making an appropriate choice of the constant D. The result is that the logarithmic profile given by eqn (1.57) can be seen to satisly the ‘law of the wail’ form, as shown in eqn (1.34). In Section 1.6.1 we shall see that there is substantial experimental support for the logarithmic mean velocity distribution. 1.53 The mixing-length model applied to a free jet Jets, wakes, and mixing layers are other flows in which the criteria for a thin shear layer are often satisfied. Thus we can employ the boundary-layer ap- proximation, despite the fact that the absence of rigid boundaries may make the name seem something of a misnomer. Also, we shall use the mixing-length model again, as it will be instructive to apply this theory to a free shear flow. The boundary-layer equations for the mean velocity, as given by (1.25) and (1.26), can again be applied: aU, , a0, Get geno (125) while (1.26) reduces to 0,24, gt. 2? (1.58) &, where we have followed the usual practice (established on experimental grounds) of neglecting the streamwise variation of the pressure and the Reynolds stress. For the same reasons, we have also neglected the viscous term iu Comparison with the term involving the Reyuolds sizes. If we wished to study wake flows, then it would be convenient to express (1.58) in terms of the velocity defect, that is, the amount by which the velocity in the wake is reduced below that of the free stream. However, here we shall dx; oxy15 SEMI-EMPIRICAL THEORETICAL METHODS 2B take the case of the plane turbulent jet as an example, and (1.58) can be used. as it stands. Let us consider a jet emerging from a thin slot into a still fluid. As always, we take the mean velocity to be in the x, direction and the jet spreads out in the x, direction. We assume similarity in different sections sufficiently far downstream (an assumption borne out in practice) and look for a general solution of the form o-#% o,--# 59) oxy oxy and an gf X2. n Y= Xt om) = iS, (1.60) where the introduction of the stream function y ensures that the continuity equation is automatically satisfied (see Batchelor (1967) for a general discus- sion of the use of stream functions) and the similarity variable 7 is given by = x2/xT. (1.61) We need a ‘closure’ for eqn (1.58), and we shall take the mixing-length form as given by (1.54). Then (1.58) becomes g ee ee B{POU /0x2)"} (1.62) dx, 7 ax, oxy The next step is to fix values for the exponents n and m in (1.60). The first condition for this is that eqn (1.62) should be self-consistent when we substitute (1.59) and (1.60) for U, and Uz. This requirement gives m = 1. Our second condition is obtained by integrating (1.62) with respect to x, from —v0 to +00, which yields the condition of constant momentum flux at different streamwise positions along the jet: {fen Orda} (1.63) Satisfying this condition then gives us n = 1/2. In the previous section, it was assumed that the mixing length was linearly proportional to the distance from the wall. In free-shear flows it is usual to assume that | is proportional to the width of the layer. In the present case, with m = Land n = 1/2, we have 1A, ax, = mfr and eqn (1.62) can be shown to reduce to f+ pf" =4af’s” (1.65)24 THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW where the primes denote differentiation with respect to 7 and (1.65) has to be solved subject to the conditions of symmetry and of zero mean velocity at the edge of the jet. A first integral of (1.65) is easily obtained, but thereafter it has to be solved numerically. Tolmien (1926; see Goldstein 1938, p. 593) found that conditions for the edge of the jet must be satisfied at = X,/x, = 3.040, (1.66) and comparison with experiment gave a value for the mixing-length constant of a = 0,0165. In Fig. 1.8 (Section 1.6.3) we make a comparison of Tolmien’s calculation with some recent results for the mean axial velocity distribution in a plane jet, and it can be seen that the agreement is quite good. The same analysis can also be carried out for the round free jet, in which case n = I and m=1 Full discussions of the various mixing-length types of theory will be found in Goldstein (1938), along with good discussions of their application to various flows. From onr present point of view. an interesting feature of the application discussed above is that the mixing-length model is used quite explicitly as a closure of the statistical equations of motion. This was less apparent in the previous section, where we relied quite heavily on the idea of the constant- stress layer. 1.6 Some experimental results for shear flows There is an enormous amount of data on turbulent flows, much of it amassed a long time ago. Indeed, the interested reader can obtain quite a good his- torical impression of the subject by consulting the book Modern developments in fluid dynamics (Goldstein 1938, two volumes). In the hope of presenting an uncluttered picture, we shall discuss only a very few investigations. We shail also mainly restrict our attention to duct flows, although we shall also give some results for the free jet. In order to give a representative picture of flow through ducts, we have drawn on the investigations of Nikuradse (1932; see Goldstein 1938), Laufer (1954), and Lawn (1971), all of which deal with flow through a pipe of circular cross-section, Results for the other main duct configuration —plane channel flow—are not very different from those for pipe flow. But, for completeness, we should mention the investigations into channel flow by Laufer (1951), Hussain and Reynolds (1975), and Kreplin and Eckelmann (1979). There is also no shortage of results for free jets, but we would mention in particular the papers by Wygnanski and Fiedier (i969) on the axisymmetric jet, and by Bradbury (1965) and Ramaprian and Chandrasekhara (1985) on the plane jet. The latter reference is especially interesting, as it involves the use of a frequency-shifted laser-Doppler anemometer. The importance of this16 SOME EXPERIMENTAL RESULTS FOR SHEAR FLOWS — 25 measuring device is that it determines the direction, as well as the magnitude, of the fluid velocity and hence (unlike the traditional pitot tube and hot-wire devices) can take proper account of the effect of the flow reversals which occur at the edge ofa jet where it entrains ambient fluid. A discussion of anemometry will be found in Chapter 3. Lastly, hefore turning to a discussion of the actual experimental results, we should define our coordinate systems for the different geometries of flow. For duct flows, x, is the axial (streamwise) coordinate, x, is the distance from the wall in the radial direction, and x, is the circumferential (spanwise) coordinate. In jets, x, is again the flow direction, but now x, is the distance from the centre line in the radial or transverse direction. 1.6.1 Mean and root mean square velocity distributions in duct flow In Fig. 1.3 we show mean velocity distributions in pipe flow for three widely separated Reynolds numbers, The results have been selected from those hed by Nikuradse (1922; ace Goldstein 1938). hed by Nikuradse (1922; see Goldstein 1928). pub! of turbulence, with a steep rise near the wall and a flatter profile in the core region. Clearly, this behaviour becomes more marked as the Reynolds number increases. 08 02 aya Fig. 1.3. Mean velocity distributions in pipe flow at various Reynolds numbers: @, R= 4000; m, R = 110000; a, R = 3200000 (Nikuradse 1932; see Goldstein 1938),26 THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW . | = 20 SUP 95.75 log x} 45.5 Fig. 1.4, Logarithmic mean velocity distributions in pipe flow: universal form in law of the wall coordinates (symbols for experimental data are the same as those in Fig. 1.3). These results of Nikuradse can be replotted (along with those of other workers (Goldstein 1938) to verify the ‘velocity defect law’ as given by eqn (1.33), but we shall not pursue that here. ‘The reduction to the universal ‘law of the wail’ form is demonstrated in Fig, 1.4, using the same three sets of results. As the abscissa is the log (ie. logarithm to base 10) of xf, the straight line indicates a satisfactory logarithmic depen- dence for most of the data, ‘This result has been confirmed by many investigators (e.g. see Goldstein 1938; Hinze 1975), which means that the velocity distribution given by eqn (1.34) is in good agreement with experimental data, except close to the wall. However, there is less than complete agreement in the literature on the values of the constants A and B. In Fig. 1.4, the straight line represents the equation U = 5.75 logio(x$) + 5.5, and, converting to natural logarithms, this implies that A = 2.50 and hence the von Karman constant k is 0.40, But. even with the restricted data presented here, it is clear that experimental scatter would permit some variation in the values of A and B. ‘A specific indication of the extent of dicagreement on thie paint can he obtained from the discussion in Hinze (1975), where it is stated that taking A= 2.44 and B = 49 represents ‘a good average’ of various investigations. Yet Ilinze (1975, p. 626) also suggests that many workers in the field are now16 SOME EXPERIMENTAL RESULTS FOR SHEAR FLOWS | 27 inclined to prefer the values A = 2.5 and 5.2 < B < 5.5, with a slight bias in favour of B = 5.5. In other words, these values are very much in line with the above equation, which represents the original conclusion based on Niku- radse’s results. Given the central position of the ‘log law’ in the phenomenology of turbu- lence, itis really rather surprising that there should be so little unanimity on the experimental value of the universal constant A. Indeed, as recently as the carly 1970s it was suggested that, for low Reynolds numbers, A actually depended on the Reynolds number, although this seems to have heen some- thing of a false alarm (Huffman and Bradshaw 1972; Purtell, Klebanoff, and Buckley 1981). In contrast, the linear part of the mean velocity profile in the viscous sublayer is a rigorous result. In addition, although we do not show any experimental points in this region of Fig. 1.4, the linear law of the wall has also been verified experimentally (for a fairly recent reference, see Bakewell and Lumley (1967)). For turbulence quantities the classic measurements were those of Laufer (1954), who used a hot-wire anemometer to obtain values of the three com- ponents of fluctuating velocity in air low through a pipe. In Fig. 1.5, we show a sketch of the rms. velocities u;, uj, and u4, each divided by the friction velocity u,, plotted against x,/a. Evidently each rms. component is roughly of the same order as the friction velocity. 2p 2ap Fig. 1.5. Sketch showing the radial variation of the three components of the rms. velocity in turbulent flow through a round pipe of diameter 2a (note that x, = 0 corresponds to the position of the wall): ——, uj us; —-—, w/t; =~ —, 05 /t).3B ‘ THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW 10 os! 06 oa 02 0 02 aa 06 08 10 Fig. 1.6, Sketch showing the radial variation of the Reynolds shear stress and its associated correlation coefficient in turbulent flow through a pipe: ——, 0, production due to this term is negative and the turbulence decreases. In turn, this can lead to relaminarization.16 SOME EXPERIMENTAL RESULTS FOR SHEAR FLOWS 31 Striking examples of this effect have been demonstrated, with the help of flow visualization, in boundary-layer flows with strong pressure gradients (e.g, Van Dyke 1982, Plate 164). 1.6.3 Mean and root mean square velocity distributions in a plane jet We have already referred to Fig. 1.8, where we compare the mixing-length calculation of Tolmien with the experimental results of Ramaprian and Chan- drasekhara (1985), These authors found that their mean axial velocity profiles exhibited self-similarity for downstream distances (as measured from the jet exit) greater than 10D, where D is the width of the jet at its exit. These self-similar profiles took a Gaussian form: Un) U,0) where 7 is the similarity variable, as given by eqn (1.64), and the experimental value of the constant is C = In(0.5). The measured velocities differed signi- ficantly from the Gaussian profile at the jet edges, where the authors noted that the mean velocities were slightly negative. As we pointed out earlier, traditional measuring techniques would be insensitive to the sign of the velocity. As the jet spreads out, there must also be a lateral component of the mean velocity, and we sketch a typical result for the profile of this quantity in Fig. exp{—Cn?} Fig. 18. Distribution of axial mean velocity in a plane jet. Comparison of the mixing- length theory with some representative results: ———, Mixing-lengtn theory, as calculated by Tollmien (1926; see Goldstein 1938, p. 593); X, representative experimen- tal value of Ramaprian and Chandrasekhara (1985), Note that the mean velocity is divided by its own value at the centre line and plotted against transverse distance divided by ¥. Here ¥ is the value of x» where U,(x3) = 0.50; (0)32. THE SEMI-EMPIRICAL PICTURE OF TURBULENT SHEAR FLOW 0.006 4 r T 0.04 4 0:10, (0) 0.02 0.04 l 0,06 L 1 L L “7 OC wy Fig. 1.9. Sketch showing the distribution of the transverse mean velocity in a free jet. 0.06 0.05+ oot 0.03 0.02 o.01F 0 0.01 0.02 0.03 peeeepeeeey LL a es nr | WY Fig. 1.10, Sketch of the variation of the Reynolds stresses across a free jet: ——, normal stress (u?)/U}(0); ~~-, shear stress G(x,x') = 5(x — x’) 26) subject to the condition 2G(%x) _ an S06) 9 rons. 7} én Discussions of Green’s function methods can be found in the book by Roach (1970), The formal solution of (2.3) can be written down immediately38 THE FUNDAMENTAL APPROACH UpexOU, Px =-p \ Px’G(x, x) LAUD) iv OXp0Xy + of x'Glx, x’) A 28) f on The first term on the r.h.s can be modified in two ways which will later prove helpful. First, we note that the index a is now repeated and hence is to be summed. Accordingly we rename this dummy index 7 in order to avoid confusion with the labelling index « in eqn (2.2). Second, we carry out a partial integration (twice), and use the boundary conditions and the symmetry of Gix,x’) under interchange of x and x’ to take the differentials outside the integration, Our final expression for the pressure then takes the form P= —p [aoe xt NUN + Fe xox, 20st pie!) 29) tov | dxGtnx'my Js Clearly, our next step will be to use (2.9) to express the pressure gradient in (2.2) in terms of the velocity field. This will take the form of a non-linear tensor expression involving tensor indices 2, f, and >. We can partly.anticipate this process by rewriting (2.2) as au a 1aP + 92, 14, - (2.2 ae Ve Poa OH Ca (22a) where the Kronecker delta d,, acts as a substitutional symbol. Then, with (2.9) for the pressure, (2.2a) becomes ou, ot Dy(V)LUp(X, U(X, 01 — Lap V)LU,(%,0)1, (2.10) Oxy where the operators D,,(¥) and L,,(¥) can be defined in terms of their effect on an arbitrary function f(x): DAVE F09] = O5;.f(8) — x, Ox. BPX'G(X,X)J(X) (211) and é f(x’ Ly W)LP00)] = f BPx'G(x,x’ng(x’) a PFO) (2.12) Two final steps are now needed. First, we can put the non-linear term on the r.h.s. of (2.10) in a more symmetric form. Noting that the interchange of the dummy indices ff and 7 must leave this term unchanged, we introduce the22. THE GENERAL STATISTICAL FORMULATION 9 symmetric operator M,»,(V) such that Myp(V) = (2.13) Second, we can extend this formulation to include flows driven by a constant externally imposed pressure gradient. In this case the external pressure P.. would be such that ve. a, = constant and hence would satisfy Laplace's equation VPP. = 0. (2.14) Thus P,,, can be added on to P(x,1), as given by (2.9), without affecting the solution of Poisson's equation. With these points in mind, we can now write (2.10) as Mag (V)[UAX, )U,(8,1)] ~ Lawrtuyouon 22, as which is the Navier-Stokes equation in divergenceless form. This may seem to be a rather cumbersome procedure, and, although we have eliminated the pressure from the problem, we do seem to have paid the price of additional complication in the equation of motion. Fortunately, this disadvantage is more apparent than real. In most of the fundamental work to be considered in this book, we shall be able to neglect the surface integral in (2.15). Also, as we shall see, for most purposes the non-linear term in (2.15) is really no more complicated than the original non-linearity in eqn (2.2). 2.2. The general statistical formulation In eqn (1.6) we have previously introduced the time average as the basis of a statistical approach. This is the method of averging normally used by experimenters, but from now on we shall base all our theoretical work on the ensemble average, It is usual to assume that the two methods of taking averages are equivalent: this property is known as ergodicity. We shall defer a discussion of ergodic theory, along with a consideration of what constitutes a turbulent ensemble, until Chapter 4. For the moment let us consider how we would vbiain au cuscuble a © for some experiment. We can imagine carrying out N simultaneous independent ident- ical versions of our experiment and then wishing to know the mean value of some quantity f(x.) for particular values of x and t. If we associate a40 THE FUNDAMENTAL APPROACH superscript n with each version (or realization) of our experiment, then we can write the ensemble average as f(x.) = lim Nove N Yo (2.16) 7 for any given values of x and t. Evidently this definition satisfies a necessary condition for an average: KM =P. (2.17) The essential feature of (2.16) is its linearity, and various important results follow from this. For example, it is now obvious that the operations of differentiation and taking averages commute; that is, eae a a or \ax / Ox and of course this result can readily be extended to any independent variable. Alea, if 4 is 2 conctant, it ie ancy ta chow that (2.20) where g(x, t) is another random variable in the domain. Finally, the linearity of (2.16) implies that + (2.21) even when f and g are not statistically independent. Of course, in our study of turbulence we shall normally be concerned with variables which are not statistically independent or which may only become independent of each other under certain restricted circumstances. In order to quantify the degree of statistical dependence of two or more variables, we make use of the correlation, of which ¢ fg) is an example. This concept will be developed further in the next section. Also, a brief general introduction to some relevant aspects of probability and statistics will be found in Appendix B. 2.2.1 Statistical equations and the closure problem We now follow the Reynolds averaging procedure again. But this time we use ‘eqn (1.7) to set up the formal many-point many-time moment hierarchy based on the solenoidal form of the Navier-Stokes equal Recalling that the velocity can be decomposed into mean and fluctuating, parts (eqn (1.7)) U,(%,1) = Olt) + Ul)22 THE GENERAL STATISTICAL FORMULATION 41 where Tx, 1) = (U,Lx,0)) and = 0, (2.23) we can set up the infinite sequence of velocity ficld moments as follows: QaplX,X'5 1,0) = CUgl(X, f)ug(x', t°)) (2.24) Qap (XXX GEESE) = Ual% uplX EWE) (2.2: and, to any order, Opp eee xe et te) = ugly t) ug (Xt) ("1") ug(X"07)... ‘Two quite general points can be made about each of these moments. First, their value cannot be affected by the order in which we take the measuring ities each satisfy the cantinnity ond, the cans condition so that the moments must also satisfy eqn (1.1). Taking the second- order moment as an example, we can express these two conditions in turn as Dag Xs X00) = Opals X50'50) (2.26) é Gal XStt) = 7 Oaglr'stt') =O Q2n We now wish to derive a more general form of the Reynolds equations which we obtained in Chapter 1. Essentially we follow the same procedure and substitute (1.7) for the velocity field into the Navier-Stokes equation, although this time we use the solenoidal form as given by (2.15). Averaging then yields the equation for the mean velocity: 8 wy2) tu.) = 1 Fe 4 a4. g ILTpla. T0021 7 Lh) = Gee + Maga WILT. 00, + A(X, Xi.) — Lap WO p(x, 0). (2.28) The equation for the fluctuating velocity u,(x,1) is obtained (as in the preceding chapter) by subtracting (2.28) for the mean velocity from (2.15) for the total velocity. It is easy to show that this takes the form é G - w) u,(X, 0) = 2Myp,(V)(Op(x, t)u,(x, 0] + + Map,(V) [uglx, u,(x, 0) — = Qpy(%¥5t,2)] — LaglV)[up(x)] (2.29) where we have used the property Myy, = Maypa THE FUNDAMENTAL APPROACH Equation (2.29) can be used to obtain an cyuation for a moment of any order. We shall illustrate this process by forming the governing equation for the vitally important second-order moment. Multiply each term in (2.29) by u,(x’,t’) and average throughout. The result is (5 1?) Ona) = 2Mag (VLU p> 0Q,(%%'s60)] | + Myo AV) [OprolX, %x56,60')] — — Lag V)[Qpol%s x60). (2.30) Of course, in order to solve this for the second-order moment we first have to solve a similar equation for the third-order moment which in turn requires the fourth-order moment, and so on, We have previously referred to this as the closure problem, and this is its most general form. 2.2.2 Two-point correlations The second-order two-point correlations (or moments) play a leading part in turbulence theory. Indeed, much of the rest of this book will be concerned in one way or another with these correlations or (more usually) their Fourier transforms. Accordingly we shall spend a little time (in this section and in Section 2.3) in considering the relationship between the two-point (fundamen- tal) and the single-point (engineering) approaches to the turbulence problem. In particular, as the multipoint theory is the more general, we should be interested in knowing under what circumstances it can be expected to reduce to the single-point form, Let us begin by considering how we would actually measure the correlation of velocities at two points in, say, a pipe flow. We can assume that we have an anemometer which will measure ail three scalar components of the fluctuating velocity at a single point x. (The subject of anemometry will be discussed in the next chapter.) If we have a second such anemometer at the point x’, then in principle all we have to do is multiply each pair of signals together and average the resultant product in order to obtain the nine scalar components of Q,y. Each component of the correlation tensor is itself a function of eight scalar variables. Therefore before we start generating a mountain of data, we should carefully consider how we can fashion In general terms, the velocity correlations can be expected to depend on two things. First, as we move the measuring points apart, we would expect22 THE GENERAL STATISTICAL FORMULATION 43 Fig. 2.1. Centroid (R) and relative (r) coordinate system for two-point correlations. the correlation to die away. Thus correlations will depend on the distance between the measuring points. Second, the magnitude ot the correlations must (at fixed separation) depend on the absolute values of the constituent velocities. If we refer to the experimental results given in the previous chapter (cg. Fig. 1.5), then clearly the value of the correlation at a given separation can be expected to depend on absolute radial position in the pipe. In practice, quite a good way of tackling this task would be to mount the two anemometers rigidly on a traversing system which would allow the distance between them to be varied in a controlled way. Then this system could itself be attached to a second traversing mechanism which would move it bodily to various different locations in the pipe, with the effect of variable separation being investigated independently at each such location Formally this procedure corresponds to a change of variables, as illustrated schematically in Fig, 2.1. That is, we express the two-point correlations in terms of R(x +x'/2 231) r=(x—x’), (2.32) where R is the centroid (or absolute) coordinate and relative) coordinate A similar transformation can be carried out for the basic time variables and ¢’, and in general we have is the difference (or Dap 85585875 — Qaplt BiG T a where Ta(e+ey24 THE FUNDAMENTAL APPROACH and tetce. (235) Clearly if we put x = x’ in (2.33) we recover the type of single-point correla- tion which was of interest to us in the previous chapter, where we met it as the Reynolds stress. In this case its dependence on absolute position was the important feature. However, if we wish to achieve a fundamental understanding of turbulence, then we should conventrate on how the correlations depend on the relative coordinate, Let us try to see why this is so. For example, consider two variables A and B, each a random function of time and each with zero mean. In simple terms, the mean of A vanishes because Ais as often negative as it is positive and so its integral over all time is zero. The same is true of B. However, the mean squares of A and B do not vanish because the integrands are always positive. What then is the implication for = Qaplt, Ro) (2.44) Now we use the fact that the mean properties of the flow must be unaffected by reflections of the velocity vectors in any one of the three coordinate planes. ig shown that Q13 = Qs: = constant = 0 Q23 = —Qaz = constant = 0 ea24 HOMOGENEOUS ISOTROPIC TURBULENCE a but Qy2 = —Qz; = an antisymmetric function of x, (2.46) Thus, with (2.44), (2.45), (2.46), «= .and x = x’, the rhs. of (2.28) reduces to L dP dQr2 has of (2.28) = this of (2.28) pax, dx, (2.46) Tt is readily seen that (2.43) and (2.46) taken together are equivalent to eqn (1.52) with Qi2l¥2) = (uy u2 ). ‘We conclude this section by noting that the above results and arguments can be used to reduce eqn (2.30)—the general equation of motion for the two-point correlation tensor—to the single-point energy balance equation for channel flow. However, it should also be noted that this form will not be identical with that previously given in eqn (1.68), as we have now eliminated the pressure as a dependent variable. 24 Homogeneous isotropic turbulence In the previous section we saw how the statistical equations obtained by averaging the Navier-Stokes equation could be very much simplified by specializing them to the case of channel flow. In order to study the physics of turbulence we need to concentrate on the simplest non-trivial problems, and it has long been widely agreed that homogeneous isotropic turbulence presents the very simplest such problem. We shall begin by bricfly discussing the two concepts in turn. It should be noted that we shall not need the time variables in this section, and accordingly we can make the algebra more compact by not showing them explicitly. Homogeneity is really a contraction of ‘spatial homogeneity’ and indicates that mean properties do not vary with absolute position in a particular direction, Thus the channel flow considered previously was assumed to be homogeneous in the x; and x, directions. However, the unqualified term ‘homogeneous’ is normally applied to fields which are translationally invari- ant in all three mutually perpendicular coordinate directions, and this is the case which will concern us here. The most important implication of this restriction is that (2.24) for the correlation can be written as a function of felative position only: = (uh) = , (2.51) where where E is the kinetic energy of turbulent fluctuations per unit mass of fluid and 6, is the Kronecker delta symbol. Thus the single-point correlation tensor has been expressed in terms of only one scalar constant. As we shall see shortly, the two-point correlation tensor can also be reduced to a single scalar, which is in this case not a constant but a function of the distance between the points x and x’, Now let us consider the question of how realizable is isotropic turbulence? Referring back to the results for pipe flow as a typical example, we see from Fig. 5 that the three rms. velocity components are quite different from each other, and hence (2.51) is not satisfied. Likewise, Fig, 1.6 shows that , (2.56) where f and g are even differentiable functions of r and satisfy f(0) = g(0) = 1 as well as (0) — g'(0) = 0. We should also recall that xr We can relate f and g to the coefficients A and B in (2.54) by considering the special case where we take r to lie along the x, axis, That is, we put x = 0 and r = (r;,0,0). Then, putting x = f = 1 in (2.54), we obtain O17) = Ar? + B= (wf (2.57) where the last step follows from (2.55). Similarly, we can puta — fi — 2in (2.54) to obtain, Q22(t) = B= (F = gE + 0? adap 259) Our last step is to eliminate g in terms of f. We do that by differentiating (2.59) with respect to r, and invoking the continuity relation in the form of eqn (2.27). With some straightforward algebra and a little rearrangement we readily find that the two-point isotropic correlation can be expressed in terms of a single scalar function f as 1 Qu) = PDL + 5 DHF (Buy — 4) 2.60) where the prime denotes differentiation with respect to r, For the particular case r = 0, we note that LF unl0) = Cu?» Ott dyy = 34u?> =2E, (2.61) where tr stands for ‘trace’, that is, the sum of the diagonal elements of a matrix. 2.4.2 Length scales for isotropic turbulence Two important length scales can be formed from the longitudinal correlation coefficient f(r), Tie fusi uf these is the Taylor microscale 2. length scale and is obtained in the following way. Suppose that we expand f(r) about r = 0, Then, recalling that f must be a symmetric function of r2 THE FUNDAMENTAL APPROACH 40) — Fig 23. Sketch of the longitudinal correlation coefficient (note that the integral scale is based on the erroneous assumption that the correlation coefficient can be approxi- ‘mated by an exponential function, and is intended to be illustrative only). and hence f’(0) = 0, we can write 2 fiy=t oo + Or) (2.62) where gine) (2.63) That is, we define the microscale by fitting a parabola to the correlation function for small values of r. This is shown schematically in Fig. 2.3. The physical importance of the Taylor microscale is indicated by the relationship connecting it to the rate of dissipation of energy in isotropic turbulence, w! (2.64) shall discuss this further in Section 2.8. The longitudinal integral scale L is just another particular case of the general integral length scale as defined by eqn (2.40). It is given by (2.65) We can idiusirate the pirysical ougit nui aitogeiiier rigor ously) by fitting an exponential form to the correlation function. (In practice this can be quite a good approximation, although clearly it is invalid at r where we require f’ = 0, from symmetry.)25 STATIONARY TURBULENCE 33 Purely for this purpose, therefore, let us take f = exp(—br), where 0 is a parameter with dimensions of inverse length. Substituting this form into (2.65), we readily find that L = 1/6, Or, in other words, if the correlation were exponential in form then the integral length scale would be the distance over which the correlation coefficient declined from f(0) = 1 to f = 1/e. Again, we show this schematically in Fig. 23. 2.5 Stationary turbulence Turbulence is intrinsically a time-dependent phenomenon. Nevertheless we can distinguish between those situations where mean velocities are dependent on time and those where they are not. For instance, an everyday example is when we use a hose pipe connected to a water tap. Now suppose that we turn the tap up or down, While we are doing this, the mean velocity of the water through the pipe will vary with time. But if we allow external factors (such as tap setting, nozzle adjustment, or ambient conditions) to remain constant, then normally the mean velocity of the water will also remain constant, a condition we referred to as ‘steady mean flow’ in Chapter 1. The extension of this idea from single-time means to many-time moments brings us to the concept of stationarity. Formally, we say that u,(x,0) is a stationary random variable if the associated many-time probability distribu- tion depends only on the differences between measuring times and not on their absolute values. As an example let us consider the two-time correlation. Temporarily sup- pressing space variables and tensor indices as irrelevant to the work of this section, we have from eqns (2.24) and (2.33) = Q(T) (2.66) where t and T are the relative (or dillerence) and absolute times respectively. Then if u(t) is a stationary random function, (2.66) becomes Ku(eju(e’)> — Q(z) — Q(t’) (2.67) such that Ott) = QW — 0. (2.68) Thus stationarity is, in effect, homogeneity in time. We have already defined an integal time-scale in eqn (2.39). For stationary fields we can also define a microscale in the same way as the Taylor microscale was introduced for homogeneous turbulence in the previous section. If R(t) is any Eulerian correlation function (e.g. f(r, ) in isotropic turbulence, R,, in channel flow, etc), then for small values of t = t — t’, we can write 2 R,(t) =1 x” + O(c) (2.69)sa THE FUNDAMENTAL APPROACH where the microscale ty is given by I = —Ri0) (2.70) te and the primes denote differentiation with respect to t. In order to carry out time averages, we require the turbulence to be stationary, or at least approximately so, on time-scales which are long com- pared with, say, the integral time-scale. It is often argued that small eddies may evolve so rapidly, compared with large eddies, that they can be regarded as being in a quasi-stationary state, rather like the idea of local isotropy which we discussed previously. In this context reference is often made to an ‘equilibrium range’ of eddy sizes. However, it seems to the present writer that this usage invites confusion with the idea of thermal equilibrium. Therefore we shall only use the word equilibrium in the latter sense and not asa synonym for ‘steady state’ ier analysis of @ velocity fiekd The introduction of Fourier analysis leads to three main benefits. It converts differential operators into multiplicrs, it gives us a relatively simple picture of the physics of turbulence, and it allows us to define the degrees of freedom of the turbulent system. This last aspect will turn out to be particularly important, both for discussion of turbulence from the point of view of statistical mechanics and for our consideration of the problems of the numerical solution of the equations of motion. We begin by considering the turbulent fluid to be occupying a cubic box of side L. The velocity field (or any of the other dynamical variables) can be expanded in a Fourier series as follows: u(x, 0) = Yuk, Dexptik x} (271) r where the wavevector k is given by 2n T tremens} (2.72) k and n,,1z, and n3 are integers, each of which is summed over the range from x to +26 It should be noted that we use the same notation for fields, irrespective of whether we are referring to the actual velocity or its Fourier component. This, aithough contrary to the normai usage to be found in mathematicai textbooks, is quite usual in this sort of work. In practice there should be no confusion, and indeed we shall find it helpful to think of u,(k,t) as the velocity field in wavenumber space.26 FOURIER ANALYSIS OF THE TURBULENT VELOCITY FIELD 55 For completeness, we present a brief summary of Fourier methods in Appendix D, with emphasis being given to those results which are most relevant to the theory of turbulence. However, we shall not consider the more profound questions concerning the validity of the Fourier representation of stationary random functions, and the interested reader who wishes to pursue such questions should refer to the books by Batchelor (1971) or Monin and Yaglom (1975). Nevertheless, one particular point should perhaps be mentioned here. In order to have rigorous isotropy, we must allow the box size to go to infinity. This means that Fourier sums become Fourier integrals in the usual way. However, the random nature of the velocity field causes difficulties for the implementation of the limiting process inherent in the transition from a summation to Riemannian integration, In our view, the simplest way of dealing with this is to defer taking the infinite system limit unul after we have taken averages. Then the problem simply does not arise. This also seems quite an appropriate procedure for the additional reason that homogeneity (a necessary precondition for isotropy) is purely a statistical property of the velocity field 2.6.1 The solenoidal Navier-Stokes equations We begin by transforming the continuity equation, as this provides us with a relatively easy start. As we are restricting our attention ta isotropic fields with zero mean, eqn (1.7) now becomes: U, (x,t) = u(% 0). Then substituting (2.71) into (1.1) and carrying out the differentiation, we obtain ¥.(iky)up(k, exp fik- x} = 0 (2.73) ¥ which must hold for arbitrary exp{ik-x}. Thus the continuity relation be- comes Kgup(k, t) = 0. (2.74) Or, u(k) and k are mutually orthogonal The general solenoidal equation of motion is given by (2.15). We wish to apply this to a system where the boundary is at infinity and there are no mean velocities or externally applied pressure gradients. If we then invoke the boundary condition that u,(x,) goes to zero as x goes to infinity, eqn (2.15) reduces to (; a wv) ont = May (V)[up(x, )u,(x, 0]. (2.75) Bo56 THE FUNDAMENTAL APPROACH The technicalities of Fourier transforming this equation are discussed in Appendix D. However we can do it straightaway if we bear in mind just two points (a) as we saw in the process of deriving (2.74), when we substitute (2.71) into (2.75) the net effect will be to replace each differential operator by its analogous wavenumber expression, (b) the non-linear term involves the product of two velocities in x-space and hence, by the Convolution Theorem, will give rise to the convolution of their respective Fourier transformer in k-space. Thus, by inspection, we can write down the result obtained by substituting (2.71) into (2.75) and carrying out the appropriate differentiations: Ee we) u,(k,t) = Mapy(k) Y gli, thu,(k — 5,0), (2.76) } j where (2.77) and Kaky a1 Dag iat (2.78) Clearly, M,»,(k) and D,g(k) are just the Fourier transforms of M,g,(V¥) and D,9(¥), as defined by eqns (2.13) and (2.11) respectively Itis easy to verify that solutions of (2.76) also satisfy the continuity condi as given by (2.74), If we multiply each side by k,, the vanishing of the 1.h.s. follows trivially from (2.74). The r-hs. of (2.76) also vanishes, and this follows from an important property of D,9(k), namely that k,Dag(k) = 0. (2.79) 2.6.2 Homogeneity and the velocity-field moments In order to develop a formalism based on eqn (2.76), we need to know something about the general properties of the moment hierarchy in wave- number space. We begin by considering the implications of homogeneity, meaning that moments are translationally invariant in configuration (x) space. Temporarily dropping the time arguments, we can write the expression for the Fourier coefficients in (2.71) as uth) (z) | Px us(xexp{—ik-x}, (2.80) from which it follows that the (wo-point correlation in k-space can be related26 FOURIER ANALYSIS OF THE TURBULENT VELOCITY FIELD 57 to the corresponding form in x-space by Cug(hug(k)> = (Gy flex dx! Cu,(Xug(X)> x x exp{ ik: x — ik’-x7} = (Z) f [ara reatoute =n) x x exp{ —i(k + k’)-xfexp{ik’-r} 281) We now invoke the property of invariance under translation. That is, Cualx)u p(X = 1)> = Cua(O)tsg(t), and (2.81) becomes eee dughkuy(k)) = (7) | f Fx dru (Ougle)> x x exp{ —i(k + k’):x}expfik’-1}. (2.82) [At this stage we can perform the integration with respect to x, noting (e.g. see Appendix D) that (i) fervent +h)X} = deieyos so that the correlation in wavenumber space takes the form = 0, unless k +j +1=0, (2.84) and, in general, the homogeneity property for a moment of any order can be written as ug (kup 4, (0)... WelP)> That is, the n-order homogeneous moment, involving the set of n distinct wavevectors {k,jl.-.,p}, is zero unless all n wavevectors sum to zero. It should be noted that all these results apply to denumerable k, jl...» P. In the next section, we shall consider the transition to continuous wavevector variables 0, unlessk +j-+i+-+p=0. (2.85)8 THE FUNDAMENTAL APPROACH 2.6.3. Limit of infinite system volume The Fourier decomposition of the velocity field into harmonic components is given by eqn (2.71), with (2.80) as the inverse relationship which specifies the Fourier coefficients. It should be noted that the integral in the latter equation is over the volume L3. What we now wish to obtain is the corresponding pair of relationships connecting the two-point correlations, as a preliminary to taking the limit as L + oo. In the previous section we found that the fact that the two-point correlation (1u,(X)uy(x’)> depended on only one space variable was enough to ensure that = (f) feProgineso( —ik-r). 287) It should be noted that (2.87) is analogous to eqn (2.80) for the Fourier coefficients in the expansion of the velocity field. Formally, we now make the transition from (2.86) and (2.87) to the Fourier integral representation in the usual way (e.g. see Appendix D) by defining the (continuous) spectral density tensor Q,9(k) as follows: oust = (4) centr Then, taking the limit as L + 20, we can introduce the Fourier transform pair Q,p(t) and Q,,(k) through the relationships b>. (2.88) Q,ght) = | dk O,p(kexp {ik -r} (2.89) and Ler Qap(k) = () | dr Q,,(exp{—ik th. (2.90) This procedure can be extended to all higher-order moments. Choosing the constituent wavevectors to be such that they satisly the general homogeneity property (2.85), we can write down the higher-order analogues of (2.88) as Qap p(k} LY\s () Cuslke)ug(jJu,(—k — j)> 291) and26 FOURIER ANALYSIS OF THE TURBULENT VELOCITY FIELD 59 Quprolk.5.)) (se) =0 unless j +k —j=0, and hence the rh.s. of the equation for {u,(k,t> is zero as Myp,(0) = 0. This result is, of course, consistent with our earlier claim that we could only have homogeneous isotropic turbulence if the mean velocity was zero or, at Worst, constant over space. The equation for the two-time correlation can be formed by multiplying each term of (2.76) by u,(—k,1’) and averaging: é (5 + «) Halk, Dita —k, 0°) a = Myyy(k) Y. G = May (A) Edythe = fil bat)? + + Map (WY. Cupli.uj(—k = jugtky > 2.114) or, taking the infinite system limit ( : a ae) Quel, t) \ee J = Myth) [Op 9+ fhe kare} 15)27 THE ENERGY CASCADE IN ISOTROPIC TURBULENCE 65 where the second term on the right is generated from the first by interchanging k with —k, and a with o. Similarly we can obtain the equation for the triple moment by first genera- ting three equations from (2.76) for u,(k,t), u,(1,0), and u,(p,t). Then each of these equations is multiplied through by the other pair of velocities. When all three equations are added together and averaged, the result is easily found to be pte P+ *) utg(k)u4p (Du, (P)> Mayy(k) © + i +{kokae p} + {kopaco}, 2.116) where the time arguments have been dropped in the interests of conciseness Further equations in this sequence will be found in the paper by Orszag and Kruskal (1968). 2.7 The energy cascade in isotropic turbulence : In this section we shall again (as in Section 1.3.2) be concerned with the energy equation for turbulent fluctuations. However, it is probably as well to remind ourselyes that we are now dealing with a situation where the turbulent energy is constant throughout space. Thus when we consider the transport of turbulent energy, this will be in wavenumber, rather than configuration, space. Alternatively, we can envisage a transfer from one range of eddy sizes to another; this process is known as the energy cascade. 2.7.1 The energy balance equation Our starting point is cqn (2.115) for the single-time correlation. We specialize the Ls. to the isotropic case by using (2.97b) to reduce the spectral tensor to its isotropic form a 2 (j + 2vk’ ) bth 9 Map(k) BI QpilbK — 5-0) — - Map(h | BQ pia —k — LKSD, (2.117) where we have also used the property M,p, —k) = —M,),(W0. It should also66 THE FUNDAMENTAL APPROACH be noted that we have added the third vector argument to the triple correla- tion, Normally we only give the two arguments, as the third one can be deduced from the homogeneity requirement that all the vector arguments add up to zero, However, in this section it will be helpful to be reminded which vector corresponds to which tensor suffix. Now we can form the equation for the energy spectrum E(k. tl. as defined by (2.101), but now generalized to the time-dependent case. If we set ¢ = sum over « and (remembering that tr D,9(k) = 2) multiply each term of (2.117) by 2nk?, we obtain d (¢ + aw) E(k,t) = T(k,) (2.118) where the non-linear term T(k,1) is given by Tibs1) = 2ak? My) | PH{Qp alk — 5 =k: 0) — (2.119) When we discussed the turbulent energy balance in Section 1.3.2, we saw that the non-linear terms conserved energy: a global result proved in Appendix A. The implication of this is that T(k,t) can only redistribute energy in wavenumber space, and so if we integrate each term in (2.119) over k, it follows from (2.100) that we must have = +f 2k E(k, dk (2.120) dpe The rate of decay of the total kinetic energy of fluctuations (per unit mass of fluid) is just the dissipation rate (see Appendix A). Hence (2.120) provides us with a simple result for the dissipation rate ¢ for isotropic turbulence: dE (a a -| vk? tk, dk. (2.121) dt ° Itis quite easily checked that (2.121) can also be obtained by Fourier trans- forming the result given at the end of Appendix A. Itis of some interest to verify that the integral of T(k, t) does indeed vanish when integrated over all wavenumbers. We can do this in the following way. From the definition of Dyp(k), a given by (2.78), it can be shown that Day(kug(k) = u,(h), (2.122) and thus Dak) Qpyalib —K) = Opyplb —k), (2.123) where we have temporarily introduced the new dummy variable I = k — J27 THE ENERGY CASCADE IN ISOTROPIC TURBULENCE 67 Then, from (2.119), (2.123), and (2.77), we can put the integral of T(k,¢) in the form 2T (kt) dk = Jo ak fora {k,Qpplbb —k) — ky Qpyp(ist — 2k, k)}, (2.124) where dummy variables have been interchanged and dummy tensor indices renamed as appropriate. At this point we again invoke the continuity condi- tion in the form Lu, () = 0, and hence 1,Qp,a(i1, —k) = 0. Clearly this allows us to replace k, in the first term on the right of (2.124) hy k, =I, = jy, and so find {fac fr fo : | (iv @nrath — — k, Qnyglinl — 2K,k)} (2.125) ‘As each triple moment is symmetric under interchange of k and j, it follows that the above integrand is antisymmetric under interchange of k and j and therefore vanishes when integrated over all space with respeel (o these variables, from which we conclude that (2.126) It will be seen later that this an important result, and alsa the form of the proof will be found to be very helpful in later sections of the book 2.7.2. Spectral picture: the Kolmogorov hypotheses The usual interpretation of eqn (2.118) is that the energy in the system at small k (large scales) is transferred by the non-linear term T(k, ) to large k (small scales), where it is dissipated (ie. turned into heat) by the viscous term. Evidently the non-linear term, which represents the collective action of all the other modes on one particular mode k, is intrinsically complicated. However, despite this, its overall effect is the energy cascade and this makes an appealingly simple picture. In this section we shall find it helpful to rewrite eqn (2.118) as eee su k — jl) dj — 2vk2 BU, 2) (2.127) a dy68 THE FUNDAMENTAL APPROACH where the definition of $ van be inferred from (2.119). It follows from (2.125) and (2.126) that S$ has the property {a Jose ik — . (2.128) for ky 2n/L, and tends asymptotically to the form given by eqn (2.137) in the inertial subrange of wavenumbers. There is no agreed theoretical form for F, but, as we shall see later, experi- mentai results suggest that some from of exponential law might be appropriate. Discussions of the Kolmogorov hypotheses often refer to the universal range of wavenumbers (in which (2.135) is supposed to be valid) as the ‘equilibrium range of wavenumbers’. This terminology is based on the argu- ment that the small eddies will evolve much more rapidly than the large eddies which contain most of the energy. Thus eddies in the universal range can adjust so quickly to changes in external conditions that they can be assumed to be always in a state of ‘local equilibrium’. In fact this seems to be a rather confusing way of referring to a process which is, in the thermodynamic sense, very far from equilibrium. Possibly a better term would be ‘quasi-stationary’. Of course in practice many real flows are stationary, and this is therefore the concept of isotropic turbulence by means of an artifice. Let us introduce an input term W(k) to the energy equation. This is an arbitrary step which we shall discuss in more detail in Chapter 4, For the moment we merely specifyn THE FUNDAMENTAL APPROACH the input by f W(k) dk (2.142) Clearly, once a stationary state has been reached under the influence of our arbitrary stirring forces, the rate of doing work will be equal to the rate of dissipation Under these circumstances, eqn (2.127) can be written as | dj S(k, j,k — jl) + W(k) — 2vk7 E(k) = 0. (2.143) ° Now, in order to have a well-posed problem with well-separated input and dissipation ranges of wavenumbers, we choose a wavenumber k’ such that re r | W(k)dk ~e~ -| Jo wk? E(k) dk. (2.144) ie This means that we require the input term to be peaked near k = 0, and that the Reynolds number should not be too low. With all this in mind, we can obtain two energy-balance equations by first integrating each term of (2.143) fram zero up to k’, and then integrating each term from infinity down to k’. The result can be written as Kp re f af) dk S(k,j, 1k — jl) + W(kj}dk =0 (2.145) moar Jo oP pe j anf dj S(ejlk — jl) — | 2vktE() dk = 0, (2.146) ew Jo J where we have used (2.128) to eliminate contributions from the double integral for 0 < k,j < k’ in the first case, and for k’ < k, j < 00 in the second, The first of these equations tells us that energy supplied directly by the production term to modes k < A’ is transferred by the non-linearity to modes > k’. Thus, in this range of wavenumbers, T(k) behaves like a dissipation and absorbs energy. Similarly, from eqn (2.146), we see that the non-linearity transfers energy from modes j k’. Therefore in this range of wavenumbers T(k) behaves like a source of energy, and this input is dissipated by the viscous term A consideration of these energy-balance equations, taken in conjunction with (2.144), allows us to put another interpretation on eas the rate at which that the turbulence is stat Lastly, we should note that the various arguments and hypotheses pre- sented in his section are not limited to the mathematical idealization of from low to high27 THE ENERGY CASCADE IN ISOTROPIC TURBULENCE 73 isotropic turbulence, Indeed it is usual to argue that, as the cascade proceeds to ever smaller eddy sizes, the conversion of velocity fluctuations to scalar pressure fluctuations (and vice versa) will lead to the ironing out of directional preferences, Thus, provided that the eddies concerned are small compared with any spatial inhomogeneity of the flow, we can think of a state of ‘local isotropy’. Under these circumstances (which, again, amount to the require- ‘ment that the Reynolds number be large enough), the Kolmogorov spectrum can be expected to apply to shear flows as well. 2.7.3 Interpretation in terms of vortex stretching High levels of fluctuating vorticity are characteristic of all turbulent flows. This fact (which is obvious from a glance at any of the well-known photo- graphs showing turbulent flow visualization) taken in conjunction with the historical development of fluid dynamics as a subject very much concerned (through classical hydrodynamics) with vortex motions, made it natural to interpret turbulence in terms of vortex stretching (e.g. Taylor 1938b), Nowadays the vortex-stretching interpretation of turbulence may seem a rather underdeveloped branch of the subject compared with, for instance, the statistical theory based on the velocity field. Yet it is a topic which is rapidly growing in importance, particularly in the study of the so-called coherent structures, We shail be returning to this, and other applications of vortex models, in later chapters. Here we will consider only some of the basic ideas. Let us return to configuration space. Reminding ourselves that the vorticity is just the curl of the velocity field, we take the curl of each term of (2.2) to obtain (Batchelor 1967) Do, Guy, t= ay + Wg, (2.147) ay where D/Dt stands for the total time derivative and contains the convective derivative u,2/2xp which acts on o,. This is the equation of motion for the vorticity, It tells us that the rate of change of vorticity is controlled by the interaction of the vorticity with the velocity gradients and by direct viscous dissipation. It will help us to understand the effect of the interaction term on the rhs, of (2.147) if we consider a vortex tube in the direction of x,. We can envisage this structure as @ long thin cylinder of fluid, rotating about its own axis, which lies parallel to x,. This means that we take f/ = 1 in (2.147) and examine what happens physically when a = 1, 2, or 3. In what follows, viscous effects will be neglected on the grounds that inertial processes will be much faster, provided that the scales involved are appreciably larger than those of the dissipation range. This allows us to invoke Kelvin’s theorem (Batchelor 1967), which states that the circulation (in effect, them4 THE FUNDAMENTAL APPROACH vorticity) moves with the (inviscid) fluid, Thus we can argue that the action of the velocity gradients on the vorticity is (to a good approximation) much the same as that on the fluid. Now consider the cases « =2 or x= 3. In either case the gradient is shearing, and will tend to rotate a fluid element. Clearly, the effect on our hypothetical vortex tube will be to tilt its axis towards the x, or x3 axes respectively. Thus the effect of this kind of interaction is to exchange vorticity between the three scalar components of «». The remaining case corresponds to « — 1, and here the gradient is exten sional, If we assume that Ou, /2x, is positive then it will cause the vortex tube to be streched out in the x, direction, with a consequent decrease in its cross-sectional area. Conservation of angular momentum (per unit fluid mass) can then be expressed in the form wir = constant where r is the radius of the vortex tube. Therefore, as r decreases under the cries of the extensional gradient, the angular velocity (proportional to @)} will increase, and accordingly the energy («,r)* associated with scale r will also increase. Hence energy is transferred to the small scales. Of course, in this picture of turbulence the velocity gradients in the above discussion should be regarded as belonging to some other vortex tube. Thus we can interpret the interaction term in (2.147) as really being a vorticity— vorticity interaction. Bradshaw (1971) has presented an ingenious ‘family tree” to illustrate how a series of such interactions on decreasing scales can lead to an energy cascade. ‘An important question hanging over the above discussion is: what is the overall effect of negative values of éu, /2x,? Clearly, any individual interaction involving negative strain rate would have the opposite effect to those discussed above, with the vortex tube being compressed rather than stretched, Discussions of this aspect often resort to the argument that it is a general property of random walks that (ou avetage) the distance between any two marked points will increase with time. Or, alternatively, we could rely on the proof that infinitesimal line elements are on average stretched in isotropic turbulence (Batchelor 1952; Cocke 1969; Orszag 1970a; Corrsin 1972; Dhar 1976), although the proof cannot be rigorously extended to vorticity or to finite line lengths. However, as only a statistical answer has any meaning for us, it seems more logical to pose the question in a more formal way by considering the equation for the mean square vorticity. We can do this by multiplying ean (2.147) through by 2«, and averaging to obtain Dio?) Di (ee se). (2.148) ~2(osenge) —20( Se28 CLOSURE APPROXIMATIONS 15 where the first term on the r.h.s. represents the mean rate of production of vorticity from the fluctuating velocity field. It was shown by Taylor (1938b) that this production term was always positive, thus leading to an increase in mean square vorticity limited only by the viscous effects represented by the second term on the rhs. of (2.148), Tt was also shown by Taylor (1938b) that the above vorticity budget would hold for the more general case of turbulent shear flows, provided that the Reynolds number was large enough. That is, terms involving the mean velocity can be neglected and the vorticity is predominantly generated by the fluctuating velocity field. As the only possible mechanism for the energy cascade is some form of vortex stretching, this result reinforces the concept of universal behaviour in the small scales for all flows. Although the assumption that the turbulent cascade consists of a tangle of vortex tubes, interacting through induced velocity fields, seems very natural, other geometric forms have also been proposed. For example, Townsend (1951) has used both vortex sheets and vortex lines to model energy transfer processes in the far dissipation region. The basic hypothesis is that ofa balance between straining processes tending to concentrate vorticity and the tendency of viscous effects to diffuse it. The result is a prediction for the energy spectrum at very large wavenumbers, and certainly a plot of k?£y,(k,) against wave- number agrees rather well with experimental results, with the model based on vortex sheets performing better than the one based on vortex lines, In fact we have no firm reason for choosing one geometrical form rather than another, However, it is interesting to note that Kuo and Corrsin (1972) analysed experimental data from grid-generated turbulence in terms of various hypotheses about vortex shape (i. blobs’, ‘rods’, and ‘slabs’). They concluded that fine-structure regions in the decaying turbulence were more ‘rod like’ than like either of the other two shapes. 28 Closure approximations There have been many attempts to derive a closed form of the equation for the energy spectrum. The situation is in fact very much like that described in Chapter 1, in connection with single-point equations, with a similar variety of ad hoc methods being employed. As in that case, the concept of a turbulent or effective viscosity has proved helpful. We shall discuss the effective viscosity method of Heisenberg (1948a, b) as a representative example. Inthis section we shall also consider the quasi-normality hypothesis. Unlike the Heisenberg model, this was not even a particularly successful theory. But 318 of historical importance as the first truiy generai anaiyticai theory of turbulence. In other words, it operates by an appeal to a general principle rather than by making very specific assumptions. The distinction that we are making should become clearer as we discuss the two theories. Also, an6 THE FUNDAMENTAL APPROACH important—if pragmatic—reason for treating the quasi-normality hypothesis in detail at the present stage is that in the process we derive the main algebraic results which are then needed for nearly all the turbulence theories to be discussed in the main part of the book. 2.8.1 The Heisenberg effective viscosity theory We begin by writing eqn (2.118) for the energy spectrum in the form dE(k,t) — 2wh2 | Ge Tt) — we) = iE Stk,jk — jl) dj — Wh B(K, 0, 0 where the second line is the same as the rhs, of eqn (2.127). The detailed structure of S(K,j,|k — jl) can be inferred from (2.119), although at this point we shaii oniy need to know that ii is aniisyus i and j The simplest way of closing the above equation is just to assume that T is proportional to E. We can find some physical justification for such a step as follows. Integrate cach term of the spectrum equation up to some arbitrary wavenumber k’: “(a6 Ke f (Ph ac- [ ax [ syst iik —i0~ { 2vk E(k, dk (2.149) jo Uae Je 0 ° where the integral of $ over the range 0 , 2.156) [As before, we suppress time arguments in the interests of conciseness. Then, making the helpful abbreviation 2n\6 ye ths. of (2.156) = {7 Hap kik — 59, (2.157) we can formally write down the solution of (2.156) as Qup,(— IPG -9} x — ht) “I dsexp{—vik? +? + (k 0 Hyy,(-k jk — 59). (2.158) At this stage we introduce the quasi-normality hypothesis (Proudman and Reid 1954; Tatsumi 1957), which is essentially to the effect that all even-order moments are assumed to be related as if for a normal distribution. It should be noted that this is a much weaker step than assuming that the distribution of turbulent velocities is normal. That would be unphysical, if only because it would be inconsistent with the existence of the triple correlation, which, as we have seen, is responsible for turbulent energy transfer. In constrast, the limited assumption of quasi-normality is quite well supported by experimental measurements (Frenkiel and Klebanoff 1967; Van Atta and Chen 1969; Van Atta and Yeh 1970, Frenkiel and Klebanoff 1973), with cven order correla: tions apparently being related in this way, except perhaps at small separations of the measuring points. The quasi-normality hypothesis is used to close the moment hierarchy by expressing the quadruple moments on the r-hs. of (2.156) in terms of products of second moments. If we denote the quadruple correlation symbolically by <1, 2, 3,4), then for a normal distribution we have (e.g. see Birnbaum 1964) 1,2, 3,4) = (1, 2) €3, 4) + C1, 3> <2, 4) + C1, 4) 2, 3D. Then application of this relationship to each of the three fourth-order moments on the rhs. of (2.156) generates nine such products of second-28 CLOSURE APPROXIMATIONS ” order moments—a degree of algebraic complexity which may seem rather formidable. However, in practice this will turn out not to be as bad as it seems at first sight. We shall see that three of these terms are identically zero, and, of the six remaining, two terms can be combined into one, as can (ultimately) the other four, leaving anly two separate and distinct terms at the end of the calculation. We can begin by taking the first term on the r.h.s. of 2.156) as an example Mapal p> { + + Cugtm)up(j) > + + Cu ,lam)u,(k — j)> }- We now invoke the homogeneity condition for second-order moments (sce eqn (2.83)) in order to deal with each of the above products as follows: produci, homogeneity in M,»,(k) = O for k = 0, this term is zero. Second product: homogeneity implies m + j Third product: homogeneity implies m + k — Hence, summing over m (ie. replacing m by —jin the second product and by —k + jin the third), using (2.88) to introduce the pair-correlation tensors, and exploiting the symmetry Maal —k) = Maop(—k) under interchange of p and 5, we reduce the above expression for the first term on the r.h.s. of (2.156) to ag \6 2( *) Mays —K)QypD Ook —D. By applying the same methods to the other two quadruple moments, we obtain for Hyp,(—k,j,k — jst) as defined by (2.157) 1) = 2Mypa—K)Qpp(IQj(k — I) + + Mgpal DQpalWIQs,(k — j) + + My pslk — PQpa(K)Qsp(i)- (2.159) Hag)(—k, jk — Nw we can combine (2.159), (2.158), and (2.155) to give the inertial transfer term as80 THE FUNDAMENTAL APPROACH T(kt) = 4k? Myp,(k) x x fos f asexoi va? +? +k iP 9} x X [2M, aK) Q,p(b,8)Qz,(k — js) + + 2MpyslI)Qpatk, )Qs,(k + 2M,,s(k — DOpalK.S)Qspli.9)]. (2.160) js) + At this stage we take the following two steps. First, we restrict the formulation to the isotropic case. That is, we use eqn (2.97b) to express the spectral tensor Qplk,t) in terms of the projection operator D,g(k) and the scalar spectral density function Q(k,1), Second, we note that inside the integration with respect to j, the variable k — j can he treated on an equal footing with j as a dummy variable, Thus. in the last term of (2.160) we can interchange k — j with j The algcbraic detail (2.160) reduces to the comparatively simple form T(k,t) = 4nk? | d »f dsexp{—v(k? + j? + Ik —j/?)(t — )} x x 2Ltk.JO(Ik — jl.s\[OU,s) — Otk. 3). (2.161) where (KI) = — Magy (ke) Mapa —K)Dyp( I) Day(k — i) = 2Myp,(k) Mpa i)Ds(k — 5) _ Walk + 7?) — kil + 2?) = Be Ki (2.162) B+ P= kin and is the cosine of the angle between the vectors k and j. Formally (2.161) gives us a closed equation for the energy spectrum when we substitute for T(k, 1) in (2.118). However, in order to be consistent, we have the choice of either using (2.101) to convert all the spectral density functions in (2.161) to spectra or of expressing the L.h.s. of (2.118) in terms of the spectral density function O(k, 1) The latter course seems easier, and also fits in better with the sort of a to introduce the spectral density function on the Lhs., obtain the closure approximation based on the hypothesis of quasi-normality:29 SOME EXPERIMENTAL RESULTS FOR SPECTRA 81 qd 2 te + Qk 00, ny = 2 fay f dsexp{—w(k? +}? + Ik — Jo Py(r— 8} x x L&DQIK jls)LQU,8) — tks). (2.163) If we prescribe Q(k,1) at t = 0, then (2.163) can be integrated forward in time for the case of freely decaying turbulence. Unfortunately, despite the apparently reasonable nature of the basic assumption, when (2.163) was solved numerically (O’Brien and Francis 1962; Ogura 1963; for a more modern critical account, see Orszag 1970b) it was found that Q(k, t) became negative for certain values of the wavenumber k. This unphysical result had a profound effect on theorists, which persists to the present day as a slightly excessive concern with the ‘physical realizability’ of theories. As Orszag points out, there is no reason why an approximate theory should not violate realizability to some small extent, The problem with quasi-normality is that the negative spectra do not constitute a smaii effect. Our treatment of quasi-normality is justified by more than its historical interest. As we shall see later, there is much to be learned about the physics of turbulence from a consideration of why quasi-normality failed (and indeed from the methods that can be used to patch it up; currently this is a very active field of research and is discussed further in Section 7.5). But, not least, in deriving eqn (2.163) we have already covered the algebra which many people find quite intimidating when encountered in conjunction with modern closure approximations. Thus the results of this section are worth mastering, as they will be extremely useful later on. 2.9 Some representative experimental results for spectra At various later points in the book, we shall look at experimental measure- ments of correlations and spectra in detail. In this section we shall confine our attention to some very general results about spectra. In particular, we shall be interested in the Kolmogorov hypotheses and the question of how universal they are, We shall also take a preliminary look at the experimental picture of the energy transfer processes in wavenumber, and, lastly, a neat method of using the Kolmogorov spectrum to obtain a measurement of the local dissipation rate at a point in pipe flows. 29.1 One-dimensional energy spectra In Section 2.6.5 we introduced the one-dimensional spectrum E, ,(k,), which is defined on the interval — 0 < k, < «0 and which is related to the mean82 THE FUNDAMENTAL APPROACH w a r wf Nati © None wh we Re Prova stn 2b hw 2 Pa's (1965) coos vicar Hom S0FTehation Fig, 2.4. Measured one-dimensional spectra for a wide range of Reynolds numbers and physical situations (note that both spectra and wavenumbers have been scaled by the appropriate Kolmogorov variables): ¥. Stewart and Townsend (1951); m, Uberoi and Freymuth (1969); a, Comte-Bellot and Corrsin (1971); 4, Champagne, Harris, and Corrsin 1970; » Laufer (1954): «, Uberoi and Freymuth (1969); @, Coantic and Favre (1974); ¢, Grant et al, 1962) square velocity through (2.107). In practice experimentalists often use ¢(k,), which is defined on the interval 0 < k; < co and satisfies fc Hk, dk, 0 Clearly ¢(k,) is just twice our F, ,(k,), but for simplicity we shall follow the experimentalists’ practice when presenting results. In Fig. 2.4 we present experimental data for 4(k,) obtained from many diverse experimental situations, ranging from measurements in laboratory wind-tunnels to the classic sea-borne investigation of Grant, Stewart, and Moilliet (1962) in a tidal channel, As the physical sources of the data are so diverse (and we are really only interested in small-scale structures), we follow the usual practice and characterize each data set by a Reynolds number R, based on the Taylor microscale, that is, R, = July (2.165) uz. (2.164) where wis the rms, velocity, v is the kinematic viscosity of the fluid, and the Taylor microscale 4 is given by (2.63).29 SOME EXPERIMENTAL RESULTS FOR SPECTRA 8 It can be seen that the spectra have been made dimensionless in terms of the Kolmogorov variables, as given by (2.132) and (2.133), and plotted against the dimensionless ratio k/ky. Clearly, the spectra at the higher wavenumbers collapse to a universal form, thus supporting Kolmogorov’s first similarity hypothesis as summarized in eqn (2.135). As the Reynolds number increases, it is also clear that the spectra show longer ranges (in wavenumber) of universal behaviour, with a tendency to de- part asymptotically from the k~*? law (as predicted by the second Kolmogorov similarity hypothesis, see (2.137)) at low wavenumbers Incidentally, it should be noted that the constant asymptote of each spec- trum at low wavenumbers is purely an artefact, resulting from the fact that the one-dimensional spectrum is merely a ‘slice’ through the three-dimensional spectrum, In physical terms this means that the low wavenumber part of #(k,) is strongly affected by ‘aliasing’ from higher wavenumbers moving at an angle to the x, axis (e.g. see Tennekes and Lumley 1972, p. 249). Referring to eqn (2.108), it is easy to show that E, ,(k,) is finite at k = 0 and depends on the ‘mean square level of the turbulence and on the integral length scale. 2.9.2 The Kolmogorov constant ‘The constant of proportionality « in the Kolmogorov spectrum has long been a target for theoretical predictions and accordingly its experimental value is amatter of some importance, To begin with, if we substitute (2.137) into (2.109) we find Eyulky) = ae, (2.166) and (recalling that #(k,) is twice E, ,(k,)) it follows that an experimental measurement of Uk) = 229 (2.167) implies that the spectral constant in (2.137) is given by y 55 a= (2.168) The results of Grant et al, (1962), which would be regarded by many as among the most reliable, suggest that 2, — 0.47 + 0.02 and hence, from (2.168), « = 1.44 + 0.06. Other investigations have given slightly different results, but there appears to be wide agreement among workers in the field that the Kolmogorov constant is about i.5. However, this agreement is not entirely unanimous. Kraichnan (1966) has commented that the estimate made of the spectral constant will depend to some extent on where the boundary between the inertial and dissipation84 THE FUNDAMENTAL APPROACH ranges is chosen. Normally this is taken to be about k = 0.1k, but, as reference to Fig. 24 will show, it must be difficult to be at all precise about this. Evidently we could do with an analytical form which could be fitted to both ranges of wavenumbers Various models and correlations have been proposed for this purpose. Probably the best known is that due to Pao (1965). Essentially, Pao’s argu- ment was that the rate at which energy is transferred through wavenumber space has the same dependence on viscosity as the energy spectrum. Thus the ratio of these two quantities is independent of viscosity: this is true in the inertial range (on the Kolmogorov picture) and hence the —5/3 spectrum is recovered there. If this hypothesis is extended to the dissipation range as well, the result is that (2.141) becomes Hy = a0 exo -(2)(K) ' (2.169) UNIT Me which, as Fig, 2.4 shows, agrees quite well with experiment The basic a it is not presented here as a spectral theory (although attempts have been made to refine it (Pao 1968; Tennekes 1968; Lin 1972) or extend it to lower wavenumbers (Driscoll and Kennedy 1983), Essentially it seems to be a good way of analysing experimental results. From our point of view, perhaps the most noteworthy feature of this analysis is the way in which it draws attention to the fact that values of «as large as 2.2 (or possibly larger) are not incompat- ible with the data 29.3 Energy transfer in wavenumber space The turbulent energy balance for wavenumber k during free decay is given by egn (2.118). Hence the energy transfer spectrum T(k), which determines the rate at which energy is transferred from large to small eddies. can be obtained experimentally for grid turbulence by measuring the rate of decay dE/dt and the dissipation spectrum D(k, t) = 2vkE(k, 0), and taking the difference of the two using (2.118). This was first done by Uberoi (1963). Later Van Atta and Chen (1969) pioneered the use of fast Fourier trans- forms to calculate the transfer spectrum directly from measured third-order correlativn functions. This method has the virtue that it can also be applied to steady shear flows, and this has permitted a comparison of spectral energy budgets between non-isotropic turbulence (in a free jet) and (nearly) isotropic grid turbulence (Helland, Van Atta, and Stegen 19//) In Fig. 2.5 we show the three terms which make up the spectral energy balance as given by eqn (2.118). Each term has been multiplied by a factor (k/k,o)? in order to make it dimensionless and also to spread it out in k-space. These particular results were obtained by computing a version of the local29 SOME EXPERIMENTAL RESULTS FOR SPECTRA 85 06 gee wi we wt 10 lk Fig. 2. Sketch of the three-dimensional energy, dissipation, and transfer spectra in otrop They —ARUk ayer: —— —2vk? E(k. tk each multiplied by a factor (k/kgo?), where ky and v are the Kolmogorov dissipation wavenumber and velocity scale, respectively) valence ( energy transfer (LET) theory (see Chapter 7). At this stage we are only interested in a qualitative picture, and so it is instructive to draw an analogy with Fig. 1.7 which shows the turbulent energy balance in x-space for pipe flow. In that case, we saw that the role of the non-linear term was to transfer cnergy from the neighbourhood of the pipe wall (where production is greater than viscous dissipation) to the core region where the reverse is the case. In other words, the inertial transfer is from source to sink. Similarly we can see from Fig. 2.5 that T(k, 1) absorbs energy at small wavenumbers (where dE/dt or the production spectrum can be assumed to be large) and re-emits it at higher wavenumbers where it is ‘absorbed’ by the dissipation spectrum. However, in drawing this analogy we should bear in mind that the energy cascade is actually from one eddy size to another, and in (for example) pipe flow such a cascade would be going on at every point in the flow. For our last topic in this chapter we again hark back to our discussion in Section 1.6.2 of the turbulent energy balance in pipe flow. There we discussed the experiments of Lawn (1971) with pipe flows, and noted that he used two methods of checking his dissfpation curve, both based on assumptions of isotropy. 1 first of these was io measure the Taylor micruscaies aud use a geucial- ization of eqn (2.64) in order to relate the dissipation rate to the mean square turbulence level and the microscales; for further details the interested reader should refer to Lawn’s paper. The corresponding results are denoted by crosses in Fig, 1.7.86 THE FUNDAMENTAL APPROACH The second method relied on the fact that measured spectra were found to have the Kolmogorov —5/3 dependence for a limited range of wavenumbers. There is, of course, no reason why the Kolmogorov hypotheses should not apply to pipe flows. We merely require local isotropy, and Bradshaw (1967) has shown that the conditions for that to hold in shear flows are less restrictive than had originally been supposed. He also proposed that this fact could he made the basis of a measurement of the dissipation rate. Thus, by fitting the region of his spectrum which had a —5/3 slope to the theoretical one-dimensional spectrum (see eqn (2.167) in the present work), and taking a specific value for the spectral constant (2, = 0.55), Lawn was able to estimate the dissipation rate. Full circles denote these particular results in Fig. 1.7, and clearly all three methods are in reasonable agreement. 2.10 Further reading The monograph by Batchelor (1971) is widely regarded as a classic. For the newcomer it provides a lucid introduction to the subject of homogeneous isotropic turbulence, although the Fourier-Stieltjes notation failed to catch on and now looks rather dated. This work would be admirably complemented by the encyclopaedic book by Monin and Yaglom (1975), which also employs the same notation for the Fourier integral Other more general books on turbulent shear flows (e.g. Hinze 1975; Townsend 1976) have sections on isotropic turbulence which can be helpful, if only to give another viewpoint, The first of these references is particularly good at giving rather full mathematical details of the various derivations. Lastly, for those who find the generally abstract treatment of the subject rather daunting, the book by Tennekes and Lumley (1972) may be helpful ‘These authors make a point of working out many specific examples in detail, especially in order to illustrate new or difficult concepts. References Barcuttor, G.K. (1952). Proc. R. Soc. A 243, 349. (1967). An introduction to fluid dynamics. Cambridge University Press, Cam- bridge. (1971). The theory of homogeneous turbulence (2nd edn). Cambridge University Press, Cambridge. BIRNBAUM, Z. W, (1964). Introduction to probability and mathematical statisties. Harper, New York, Brapstaw, P. (1967). NPL Aero Report 1220. 4) Am introduction wo wréuience und ity measurement, Pergamon Press, Ox ford. BROWN, LW. B. ANTONIA, R.A., and RASAGOPALAN, S. (1983). Phys. Fluids 26, 1222. Cuampacye, FH, Hannis, V.G, and Corrsin, 8. (1970) J. fluid Mech. 41, 81 ‘Cou, P. Y. (1945) Q. appl. Math. 3, 38 Coanmic, M. and Favre, A. (1974). Adv. Geophys. 18A, 391.REFERENCES 87 Cocke, W. J. (1969). Phys. Fluids 12, 2488. Comre-BELLor, G. and Corrstw, S.(1971). J. fluid Mech. 48, 273. Corrsin, S. (1972). Phys. Fluids 13, 1370, Duar, D. (1976). Phys. Fluids 19, 1059. Driscott, R.J. and KeNNeDy, L.A. (1983). Phys. Fluids 26, 1228. FReNKIEL, F.N. and KLEBANOFF, P.S, (1967). Phys. Fluids 10, 1737. and” (1973). Phys. Fluids 16, 725. Goxpstemn, S. (1938). Modern developments in fluid dynamics, Clarendon Press, Oxford. (Reprinted by Dover Publications, 1965.) Grant, H.L,, Stewart, R, W., and MOmLLr, A. (1962). J. fluid Mech. 12, 241 Hersexnerc, W. (1948a). Z. Phys. 124, 628. — (1948b). Proc. R. Soc. A 195, 402. ELLAND, K.N., VAN Arta, C.W., and STeGEN, G.R. (1977) J. fluid Mech, 79, 337, Hinze, J, 0. (1975), Turbulence (2nd edn). McGraw-Hill, New York. KoLmoGorov, A.N. (1941a). C. R. Acad. Sci. URSS 30, 301. — (19416). C.R. Acad. Sci. URSS 32, 16. Krarcunan, RH (1966). Phys Fluids 9, 172 Kuo, A. ¥. and Corrsty, S. (1972). J. fluid Mech. 6, 447. Laurin, J. (1954), NACA Tech. Rep. 1178. Lawn, C.J.(1971). J. fluid Mech. 48, 477. Lesttz, D.C. (1973). Developments in the theory of turbulence. Clarendon Press, Oxiord Lax, ICT. (1972). Phys. Fluids 18, 205. Monty, A.S. and YAGLoM, A. M. (1975). Statistical fluid mechanies, Vol..2, Mechanics of turbulence, MIT Press, Cambridge, MA. O'BRIEN, E. E. and FRANCIS, G.C. (1962). J. fluid Mech. 13, 369, Ocura, ¥. (1963). J. fluid Mech. 16, 33. ORSzAG. S.A. (1969), Stud. appl. Maths. 48, 275, ——(1970a). Phys. Fluids 13, 2203. (1970b). J. fluid Mech. 41, 363, —— and KRUSKAL, M.D. (1968). Phys. Fluids 11, 43. Pao, Y.-H. (1965) Phys. Fluids 8, 1063. — (1968) Phys. Fluids 11, 1371. ProupMan, I and Reip, W.H. (1954). Phil. Trans. R. Soc. Lond. A 247, 163. Roactt, G.F. (1970). Green's functions: introductory theory with applications. Van Nostrand Reinhold, London Rone sox, H. P. (1940). Proc. Camb, Phil, Soc. 36, 209. Stewart, R.W. and TOWNSEND, A. A. (1951). Phil. Trans. R, Soc. Lond. A 243, 359. Tarsumt, T. (1957). Proc. R. Soc. Lond. A 239, 16. Tavtor, G.I. (1935). Proc. R. Soc. Lond. A 151, 421. Tavtor, G.T.(1938a). Proc. R. Soc. Lond. A 164, 476. TAYLOK, G.I. (1938). Proc. R. Soc. Lond. A 164, 15. TeNNekES, H, (1968). Phys. Fluids 11, 246. Tennres, H., and Lumwey, J.L. (1972). A first course in turbulence. MIT Press, ‘Cambridge, MA. Townsenn, A.A, (1951). Proc, R. Soc. Lond. A 208, 534. —— (1976). The structure of turbulent shear flow (2nd edn), Cambridge University Press. Cambridge UneRo1, M.S. (1963), Phys. Fluids 6, 1048. ‘and FREYMUTH, P. (1969). Phys. Fluids 12, 1359. VAN ATTA, C.W. and CHEN, W. Y. (1969). J. fluid Mech, 38, 743. and YEH, T.. (1970) J. fluid Mech, 41, 169. Zaman, K.B.Q. and HUSSAIN, A.K.M.F. (1981). J. fluid Mech, 112. 379,3 SOME RECENT DEVELOPMENTS IN THE STUDY OF TURBULENCE This chapter is somewhat miscellaneous in character and is intended to serve two purposes. The first of these is to cover certain topics which would not easily fit into the main part of the book, yet which are important. The sections on anemometry, data processing, and engineering models are the main topics in this category. The second purpose is to try to give the reader a broad picture of what the rest of the book is about. That is, certain sections of this chapter should act, as both an introduction to and an overview of more detailed and advanced treatments later in the book. For example, the section on renormalized per- turbation methods is mainly descriptive, using only rather simple mathematics to explain the general method 3.1 Measurement techniques and data analysis The subject of flow measurement is vast, but here we are only concerned with the quantitative measurement of the fluid velocity at a point, Our main objective will be to discuss the new developments of laser anemometry and digital data processing, which together are in the process of revolutionizing the experimental study of turbulence. We begin by considering the older methods of making local measurements of fluid velocity in order to provide a background for the newer material. 3.1.1 Anemometry Historically, methods of continuously monitoring fluid velocities have de- pended on restricting the flow in some way, and using the Bernoulli equation to relate the flow through the restriction to the pressure difference across it. Discussions of such methods can be found in almost any elementary text on fluid mechanics. From our present point of view, the only one of these methods to qualify as measuring local velocities is the pitot tube, Essentially this is just a thin tube placed with its open end facing the oncoming flow and its other end connected to a manometer in order to measure the total stagnation pressure, The other side of the manometer is then connected to a pressure tapping which is positioned such that it registers the free-stream pressure. The accuracy of the pitot tube can be established by comparing its value for the bulk mean velocity in (say) pipe flow with the result obtained by31 MEASUREMENT TECHNIQUES AND DATA ANALYSIS 89 measuring the volumetric flow rate and dividing by the cross-section area, Surprisingly, even for turbulent flows, the pitot tube turns out to be quite an accurate measuring device. The only proviso is that we must restrict its use to mean velocities, as the inertia of the system is such that it cannot respond rapidly to turbulent fluctuations, However, for anything other than the simplest of practical applications, it is precisely the fluctuations that we are interested in. Thus virtually the whole quantitative treatment of turbulence during most of this century has depended on the hot-wire anemometer, which has been developed to the point where it can handle the fastest turbulent fluctuations with ease. In its simplest form, this consists of a short fine metal wire which is heated by the passage of an electric current through it. If the wire is cooled by a flowing fluid, its electrical resistance is reduced and (all other parameters being constant) this reduction in resistance can be related to the fluid velocity. Typically wires for this purpose are made of platinum or tungsten and have a diameter of a few microns. The operating principle is readily understood as follows. We assume that the wire 1s mitially at the same temperature 1; as the luid and has electrical resistance R;. If we then heat up the wire to some temperature T,, the resulting resistance R,, will be given by the well-known formula Ry = Rel + o(Ty — T)} Bl) where o is the temperature coefficient of resistance. Ifa current J flows in the wire, then heat is generated at a rate given by Joule heating rate = [?R,. This heat is then transferred from the wire to the fluid at a rate given by rate of cooling = AS(T,, — T) where his the heat transfer coefficient and S$ is the surface area of the wire For thermal equilibrium, these two rates can be equated: PR, = hS(T, — T,) = WSR =e) (3.2) where the last step follows upon substitution from (3.1) for the temperature difference. Evaluation of the heat transfer coefficient h is a rather specialized matter and we shail not go into the details here. However, provided that free convec- tion can be neglected and the cooling process is entirely due to the fluid velocity Uy perpendicular to the wire, a reasonably general result is PR, Ry — Re = A+B 6.3)90 RECENT DEVELOPMENTS IN THE STUDY OF TURBULENCE where 4 and B are assumed (o be independent of the fluid velocity. This formula is often referred to as *King’s law’, although strictly that term should be applied to the case of a cylindrical wire, where (3.3) can be obtained from first principles (King 1914) In practice it is always best to treat eqn (3.3) only as a guide and to calibrate the hot-wire anemometer against a pitot tube in laminar flow. Then the resulting /? can be plotted against ./Uy in order to obtain values for 4 and B. This calibration can be extended to low intensity turbulance, but it is best to usc an electronic linearizer such that the output of the anemometer becomes proportional to Us The usual method of measuring the change of resistance in the hot wire is to incorporate the device in one arm of a Wheatstone bridge. Then the usc of a power supply with a high internal resistance ensures that changes in bridge resistance do not affect the current and the voltage differences across the bridge can be detected. This is known as the constant-current mode of opera- tion, It is limited to low turbulent intensities by its poor response at high frequencies. ‘Nowadays the intrinsic sluggishness of thermal devices can be overcome by using electronic feedback to keep the temperature (and hence the resistance) of the wire constant. In this way the upper frequency limit can be increased from about 500 Hz to about 1.2 MHz, which is adequate for almost any turbulent flow. ‘The operating circuit of the constant temperature anemometer is shown schematically in Fig, 3.1. The mode of operation can be summarized as follows: (a) The bridge is initially in balance: no error voltage. (b) Flow changes, probe resistance changes: hence an error voltage is generated, (6) The error voltage is fed to the servo amplifier. Bridge top Servo ampli Error voltage oI Probe, in. Bridge 7D) Y t vote aa Fig. 2.1. Circuit diagram for a constant-temperature hot-wire anemometer.31 MEASUREMENT TECHNIQUES AND DATA ANALYSIS 91 (d) The servo amplifier increases ur decreases (he vperating voltage at the bridge top in order to restore the balance. (6) The velocity change appears as a change in the bridge voltage. It shared be noted that the higher the gain of the amplifier, and the faster its response, the lower is the error voltage needed to compensate a temperature change in the sensor. This is the basis of the fast frequency response, despite the slow thermal response of the wire. Hot wires (and hot films) can be used in all sorts of configurations, some of them very complicated, depending on the precise purpose of the anemometer. ‘The interested reader will find much further information in the relevant chapter in Hinze (1975), in the review article by Comte-Bellot (1976), and in the specialist texts by Perry (1982) and Lomas (1986). We shall conclude this section with a discussion of the simplest, and most important, application. Let us consider a single hot wire in a turbulent flow with mean velocity U, Although the mean flow is unidirectional, we know that the fluctuating velocity has three scalar components, which we denote by (u,,42,us). Our problem now is to distinguish between these three components, but this turns out to be a lot simpler than it might appear. We align the wire with the x, axis, The physical situation is illustrated schematically in Fig. 3.2. This means that the u, component does not (at least toa good approximation) contribute to the cooling of the wire. That is, we are assuming that the rate of cooling is entirely determined by the velocity normal to the major axis of the wire, which is given by the vector sum of U, + u, and uy, OF (0, + uy)? + ug}? (3.4) Manse ‘ ae Os Side view Fig. 32. Hot-wire configuration to measure the streamwise turbulent fluctuation in a unidirectional mean flow92 RECENT DEVELOPMENTS IN THE STUDY OF TURBULENCE ‘Then, provided that the turbulent intensities (ic. u,/U, or u2/0,) are smail, we can usefully approximate this result by a Seas (tt Oy =U, 41+ + 0( (3.5) | G, (&)t | Thus, to the first order of small quantities, Uy 2 OU, +, (3.6) or in other words, the effective cooling velocity is approximately equal to the instantaneous streamwise velocity, provided that the velocity fluctuations are small compared with the mean, 3.1.2 The laser anemometer The invention of the laser anemometer is usually attributed to Yeh and Cummins (1964), who showed that steady fluid velocities could be measured by observing the Doppler shift in the frequency of laser light scattered by small particles moving with the fluid, Over the following decade the method was extended, refined, analysed, and generally improved to the point where it is now routinely used in research in fluid dynamics and (in particular) in turbulence. (Plate 1) Many of the pioneering workers were content to interpret the basic operat- ing principle in terms of the Doppler shift of scattered light, and this fact has left us with the legacy that the method is normally referred to as laser-Doppler anemometry’ (LDA) or (sometimes) as ‘laser-Doppler velocimetry’ (LDV). However it is now widely accepted that the operation of the laser anemometer can be more easily understood in terms of conventional optical concepts like interference and diffraction. Thus, although we are stuck with the above acronyms, in this chapter we shall introduce laser anemometry without relying on the Doppler effect. However, in the fuller discussion of the background to the subject given in Appendix F we shall find it helpful to make use of the general analysis in terms of the Doppler shift of the scattered light. The laser anemometer can be set up in many different optical configura- tions. In Fig. 3.3 we show an optical arrangement which is probably the most widely used and is certainly the easiest to understand, The incident beam of laser light is split into two separate beams by a suitable beam splitter. There are, of course, many ways of doing this, and the precise form of device is not critical. However, it is generally advantageous if the resulting beams are of about equal intensity and symmetrically disposed about the incident beam. they form a pattern of interference fringes. This is the measuring point and, for con- venience, we take it as the origin of coordinates. The angle of intersection is 26 and this is simply related to the beam separation and the focal length ofPlate I. A laser-Doppler anemometer being used tc This device measures the fluid velocity at the point where the light beams i (Courtesy of Dantec Electronics Lid)3 MEASUREMENT TECHNIQUES AND DATA ANALYSIS 93 He-Ne laser Beam Lens . Lens Photo multiplier splitter 20 Side view Fig. 3.3. The laser anemometer: a typical optical arrangement, faa ee -_ LILLY TM, Cf. er, Fig. 3.4. Schematic view of a particle with velocity vector w crossi fringes 1g the interference the lens. The set-up is completed by a second lens which focuses scattered light onto a photodetector, which in turn outputs a voltage proportional to the incident light intensity. In order to have a definite example, we show the anemometer set up to measure the axial component of velocity u, in a free jet from a nozzle, The incident beams lie in the x,x, plane, In Fig. 3.4 we show a cross-section through the fringe pattern in the x,x3 plane (of course, the fringe pattern to.as the probe volume and itis only on the overall scale of the flow that we can regard it as a point). When a particle in the jet moves through the fringe pattern with velocity u, it scatters light from each bright fringe. This light is94 RECENT DEVELOPMENTS IN THE STUDY OF TURBULENCE Fig. 35. Oscilloscope trace showing the photomultiplier output voltage correspond- ing to the light scattered by one particle crossing the fringe pattern collected and focused onto the photodetector, which produces an output current proportional to the scattered light inten: Figure 3.5 shows a typical laser anemometer signal in the form of an oscillograph. This is for one particle crossing the fringes and reflects the sinusoidal intensity variation (modified by the lascr beam’s own Gaussian profile) in the fringe pattern. We should also note that the alternating signal isona pedestal or d.c. level. Details of these optical technicalities can be found in Appendix F; here we shall concentrate on the relationship between the static fringe pattern and the time-dependent signal on the oscilloscope. As the particle crosses the fringes, clearly only the u, component of its velocity can contribute to the oscilloscope signal; the u, and u, components do not cause fringe crossings. If the distance between the fringes is d, then the time between fringes at a speed u, is d/u,. Hence it follows at once that the frequency observed on the oscilloscope is fo =u /d (3.7) where the subscript D stands for Doppler. It is shown in Appendix F that the fringe spacing is given by where / is the wavelength of the laser light (typical value 633 nm for a He-Ne31 MEASUREMENT TECHNIQUES AND DATA ANALYSIS 95 fayer), Thus, eqn (3.7) eam be written as fod 1 35nd (3.8) This is a special case of the general result of an analysis in terms of light scattering and the Doppler effect. In that context, this particular optical arrangement is known as the ‘differential Doppler mode’ Equation (3.8) is the operating relationship for the laser anemometer, and two points stand out at once. First, unlike the relationship for a hot-wire anemometer, it isa linear expression, Second, the constant of proportionality in the relationship between the measured Doppler frequency and the particle velocity depends only on the wavelength of the laser light and the angle of intersection of the beams, both of which can be established in advance. Hence the instrument does not require any calibration and can be regarded as a ptimary standard. These points, taken in conjunction with the obvious fact that the laser anemometer does not perturb the flow in any way, make this a formidable mstrument lor research m turbulence. However, on the debit side, the fluid must be transparent, there may be some uncertainty about how well the scattering particle is representative of the actual fluid motion, and the signal- to-noise ratio is not usually as good as for the best hot-wire anemometers (e.g. see Melinand and Charnay 1978; Lau, Whiffen, Fisher, and Smith 1981). Normally when a laser anemometer is heing used, there will he several particles in the probe volume at any one time, and the effect on the output signal can be seen from Fig. 3.6. It should be noted that the trace in Fig. 3.6(a) is just that of Fig. 3.5 after high-pass filtering to remove the d.c. pedestal level. In Fig. 3.6(b), the effect of scattering from several particles is seen to reduce the signal-to-noise ratio, but, by way of compensation, the signal takes a more or Jess continuous form which can be helpful when it comes to signai processing. Some ways in which the continuously varying Doppler frequency (due to the continuously varying fluid velocity) output can be turned into a voltage are discussed in Appendix F. We conclude this section by considering the important technique of fre- quency shifting, It is helpful to introduce this concept through a consideration of the signal spectrum, If we denote the output current of the photodetector by i), then we can obtain the corresponding spectrum i(f) in the usual way by Fourier trans- formation, Let us begin by considering the single particle signal as exemplified 3.5. If this signal were just a cosine wave and a d.c. pedestal level, then rier transform would consisi of a spike ai zero frequency and anuther at the Doppler frequency. In fact there is also a finite-size effect, which in this case shows up as the Gaussian modulation due to the intensity profile of the laser beams, As a Gaussian is self-reciprocal under Fourier transformation,