OCR Text |
Show their characteristics. Even such seemingly simple features as the mean spreading angle or the mean location of a wall reattachment point have often been beyond our capabilities to predict. The fact that this is so is not for lack of trying. Many studies, experimental and computational, have been directed at these flows, as reviewed by Gupta et al. (1984). Some have met with a reasonable degree of success, while others have been less successful. Notable among the studies which have met with both are those conducted by the International Flame Research Foundation (IFRF) as part of their Near-Field Aerodynamics (NFA) research program (Weber et al., 1990). These studies show good agreement between experimental measurements and computations using the Reynolds-averaged Navier-Stokes equations as long as a second order turbulence closure (Algebraic Stress or Reynolds' Stress) is used. This paper addresses a troublesome flow which we have been investigating in connection with our studies of steady spray combustion (e.g., Edwards and Rudoff, 1990). It is the isothermal flow (air only) in a research-scale, optically accessible furnace. The furnace is nearly axisymmetric in cross section and has a sudden expansion of about 6: 1 in diameter at the inlet. The mass flowrate of air provided to the furnace is the same as that used under combusting conditions, resulting in a jet Reynolds number of 25,000 (based on the throat diameter), and a swirl number of 0.7 measured at the inlet plane. 1 To date, our attempts to compute the behavior of this flow have been based upon a timeaccurate solution of the Reynolds-averaged Navier-Stokes equations using a modified form of the KlV A-IT computer code (Amsden et al., 1990). Turbulence closure has been of frrst order, using the k-£ model. The problem which we have encountered is that the computations overpredict the rate of spread of the jet towards the wall of the furnace. Inasmuch as it has long been known that the turbulence of these flows is anisotropic (Beer and Chigier, 1972), whereas the k£ model assumes isotropic turbulence, it seems reasonable to conclude that the solution to the problem is, as Weber and coworkers suggest, use of second-order closure. The only problem with the second-order approach is that we suspect the root-cause of the discrepancy is not the turbulence closure, but instead the applicability of Reynolds' decomposition to this type of flow. Inherent in the use of Reynolds' decomposition is the assumption that the time-mean velocity field is somehow meaningful to solution of the problem of interest. In many problems in science and engineering this is certainly the case. This is particularly true when the dominant scales of the mean and fluctuating components of the velocity are widely separated (i.e., ii »u'). Meaningful results may also be obtained when this separation of scales is not as distinct, but the problem of interest is somewhat insensitive to the scales of the flow. However, if the problems of interest are local phenomena (e.g., spray dispersion and evaporation), and the scales of the fluctuations are comparable to those of the mean flow, it is expected that the time-mean flow characteristics will yield little insight into the problem. ITbe swirl number provides an assessment of the relative magnitudes of the angular momentum flux and the axial momentum flux of the flow (see, for example, Gupta, 1984). Note that in evaluation of the swirl number for this flow t possible contributions due to correlated velocity fluctuations and pressure gradients have been neglected. - 2 - |