VOL. 26, NO. 11, NOVEMBER 1988 AIAA JOURNAL 1299) Reassessment of the Scale-Determining Equatio’ for Advanced Turbulence Models David C. Wileox* DCW Industries, Inc., La Canada, California {A comprehensive and critical review of closure approximations fo two-equation turbulence models has been for adverse pressure gradi ion in an attempt to find the optimum perturbation tends to pendent variables exists that much . Based on the analysis, a two-equation turbulence model postulated that is shown to be quite acurate for atached boundary layers in adverse pressure gradient, ‘compressible boundary layers, and fre shear flows. With no and without the ‘lows with surface mass adaton. I. Introduction URING the past 20 years, a great deal of research has focused on the task of devising closure approximations for the long-time-averaged Navier-Stokes equations suitable for predicting properties of turbulent flows, Prior to 1968, ‘virtually all turbulence closure schemes were “incomplete,” ice, their implementation required some advance knowledge about the flowfield under consideration in order to obtain a soliton, The best-known incomplete turbulence model is the mnixing-length model.! This model is incomplete because the appropriate form of the mixing length must be determined empirically for each new application; in general, it cannot be specified a priori ‘In 1968, the first Stanford conference? on turbulent flows ‘was held to test existing turbulence models against the best ‘experimental data available. The data base was confined to incompressible two-dimensional boundary layers. The compe- tition was won, more or less, by the “incomplete” model of Bradshaw et al ‘The trend in turbulence modeling since the first Stanford conference has been toward development of complete models. For clarity, note that the terminology ‘‘complete model of, ‘urbulenee,”” as used in this paper, means a set of equations that can be used to predict a given turbulent flow with no ‘advance information other than boundary conditions requited {in order to achieve a solution. The terminology isnot intended to imply anything with regard to the range of applicability of the theory, ‘Over the past 15 years, the most vigorous modeling efforts hhave been conducted by Donaldson et al.,* Launder eta. and Wileox etal." Recognizing the substantial progress the various researchers seemed to be making, the second Stanford ‘conference on turbulent flows was held in 1980 and 1981." ‘This time, however, the scope of the experimental data was ded tremendously to include complicating effects of samline curvature, surface mass transfer, paration, secondary motions, etc. that is, Hibs aie AIA 2nd Ads Sens ie reed Jan 3, ean PGi © Anca tio Axo i I. ih teed ow Ain ‘ous damping ofthe mode's closure coefficients ‘of wall functions, the model equations can be integrated through the viscous sublaye Surface boundary condition are presented that permit accurate predictions virtually every complicating effect known to man was inched iF experimental data of reliable quality existed From this researcher's viewpoint, results of the second of the two Stanford conferences were af once very encouraging and disappointing. On the one hand, the state-of-the-art has teen shown to have advanced dramatically since the fist Stanford conference. It was hard to imagine in 1968 that separated flowfels could be routinely predicted with any degre of accuracy just 13 short years later. (OF course, turbulence modelers should receive only part of the credit, ‘magnificent advances in numerical methods, such as those of MacCormack have played a very important role, 0 sa) the least) On the other hand, although such predictions can be routinely made, obtaining results consistent with. measur- ments isnot nery as routine. Far worse i wasnt even clear from the results presented a the second Stanford conference that effects of an adverse pressure gradient onthe turbulent boundary layer could be predicted any more accurately than in 1968, Clearly, progress in turbulence modeling hasbeen it In light ofthis situation, this study was inated by taking 4 modest step backward to review and assess the original closure approximations forthe cass of turbulence modes nov as two-equstion modes thats, closure being accom plished sing she long time-averaged Navier-Stokes equations ind te addons dient eatos, The rata or starting at what would seem to be a very elementary level stems fron a key observation made at the second Stanford conference: the greatest amount of uncertainty and contro- ‘ersy over two-equation and hisher-order models lies in the seale-determmininig equation. It is even unclear what the ‘optimum choice of it variables is for a two-equation todel, As tel of tis study, we fel we have found the optimum choice and, based on this choice, we have postulated new two-equation turbulence model ‘Sesion I summarizes the new model, including arguments that set values of all but two of the closure coefficients appeatng inthe postulated equations. Section II presents ‘esl ofa perturbation analysis of the icompresble defect layer, including effects of pressure gradient. Predictions of he sew ode are compared with those ofthe Jone-Launder and the Wileox-Rubesin'® models. Section IV uses perturba- tion methods to analyze the viscous sublyer including effects of surface roughness and surface mass injection, Section V1300 D.C. WILCOX includes results of attached boundary-layer computations for boundary layers subjected to adverse pressure gradient, susface mass injection and compressibility, and free shear flows. II, Equations of Motion ‘This section states the postulated equations of motion, including established values of all closure coefficients. Physi- ‘al interpretations of turbulence field properties are given and, ‘additionally, arguments are presented that have been used in setting values of several of the closure coefficients. ‘A. Postulated Equations For general compressible turbulent fluid flows, the turbu- lence model equations are written in terms of Favre!? mass- averaged quantities as follows. Mass conservation: ae ae ® Momentum conserv te Jou +e mom a @ Mean eneray conservation: 38 geyed neg u OF) oul=5e [net ‘ar) 3 «| @ ‘Turbulent mixing energy aay a 2, ge au Ob + ay, Gouh = 5, —B* pak 2 [rotary 2 Et [ure ‘ard x) @ Specific dissipation rate: Zoom (6m) (e/krry = teed [orm where is time, x, position vector, u velocity vector, p density, ‘pressure, y molecular viscosity, 4, the sum of the molecular land Reynolds stress tensors, and g, the sum of the molecular ‘and turbulent heat flux vectors. In Eq. (3), the quantities Ewe tk +uui/2 and H=h+ k + ugi/2 are total eneray and enthalpy, respectively, with h = € + p/p; e and h denote internal energy and enthalpy. Additionally, rj s the Reynolds ‘ress tensor. The turbulent mixing energy k and the specific pon ae ar needed to dette ety aly ry is given by AIA JOURNAL, ‘where, by definition, the mean strain-rate tensor S, is 5S ae] ® We invoke the Bouin appoxinaton ht he Resa Stes tenor ib proportional te men anu tesr ook ete 3m raters, 14], 8) Finally, the heat flux vector g is approximated as (wr) ah a= ee (0) where Pr, and Pry are the laminar and turbulent Prandtl ‘numbers, respectively. ‘Several closure coefficients, namely, 8, 6°, 7,7". 0, and o* appear in Eqs. (1-10). A key objective of this study has been to review typical arguments used in establishing values of such coefficients in a model of ths type. In the next subsection and in later sections, the arguments are presented. The values are summarized in the following equations: 8=3/40, B= 9/100, =5/9, y*=1, nn 20° ay Before proceeding. to further discussion of the closure coefficients, it is worthwhile to pause and diseuss the form of the model equations snd the physical meanings of the quantities k and «As in other two-equation models of turbulence, the quantity k represents a measure ofthe kinetic energy of the turbulence. Whether k is specifically identified as being the exact kinetic energy of the turbulence or, alternatively, the kinetic energy of the fluctuations in the direction of shear isnot eitically important. All we require fon physical grounds is that k be proportional othe square of the velocity at which local turbulent mixing occurs, The second quantity introduced inthe model, w is referred to as, the specifi dissipation rate. Its dimensions are inversely proportional to time, and is, in fact, the same variable used by this author in all prior turbulence modeling. studies. Perhaps the simplest physical interpretation of is that it is the ratio of the turbulent disipation rate «to the turbulent mixing energy. Alternatively, is the rate of disipation of turbulence per unit energy ‘Asis obvious from inspection of Eq. (4), the equation for k is modeled directly after the exact, long-ime-averaged equa- forthe turbulent kinetic energy. On tis pont, the model is consistent with virtually ll other two-equaion Odes. The second of the two model equations is similar in fotm to the equation fork. Although it adds no rigor to the exuation for u can be regarded asthe form of the civation that would result from 1) writing for turbulent kinetic eneray and dissipation the formal change of dependent variables ‘The primary life study and the Bp for our choice we analyze