8 The Open Aerospace Engineering Journal, 2008, 1, 8-27 1874-1460/08 2008 Bentham Science Publishers Ltd. Assessment of the Performances of RANS Models for Simulating Swirling Flows in a Can-Combustor K. Khademi Shamami and M. Birouk * Department of Mechanical and Manufacturing Engineering, the University of Manitoba, Winnipeg, Manitoba, R3T 5V6, Canada Abstract: The paper presents an assessment of the performances of RANS turbulence models for simulating turbulent swirling can-combustor flows with different inlet swirl intensities (i.e. S=0.4 and S=0.81). The predictions compared against published experimental data reveal that the eddy-viscosity models can not show the central recirculation zone in the case of a weakly swirling flow. However, although they reveal the existence of this region in a strongly swirling flow, they are incapable of predicting its correct size. On the other hand, the Reynolds stress models are able to predict the cor- ner and the central recirculation zones in both flow cases. The predictions of turbulence intensities by using the realizable k- and the SST k- are comparable to those of the Reynolds stress closures. The shear stresses are not well predicted by all the tested models. Both the eddy-viscosity and the Reynolds stress closures show relatively less approximation errors in the weakly swirling flow. 1. INTRODUCTION Swirling flows are used in a wide variety of engineering applications, such as furnaces and gas turbine combustors. The use of swirl in these power systems has several benefits. It is recognized that a swirling flow produces an adverse pressure gradient that can cause flow reversal or vortex breakdown. The swirling flow’s central recirculation zone may result in decreasing pollutants emission by bringing hot species back to the combustion zone as well as lowering the possibility of flame blow-off. Moreover, swirl causes further mixing between the fuel and the oxidant. To improve the performance of a combustor, an accurate insight into the flow structure is needed. Due to the complex turbulent nature of a swirling flow in a combustor, accurate numerical calculations of the flow parameters require a care- ful choice of turbulence models. These models are needed to calculate the turbulent stress terms in the mathematical equa- tions that describe the flow dynamics. A review of the litera- ture reveals that numerous studies are reported on the mathematical calculations of swirling flows in a combustor. It is shown that the standard k- model [1-2] and its different versions (e.g. References [3-5]) which can perform reasona- bly well for simulating simple turbulent flows, appear inade- quate for simulating swirling flows [6-28]. Using different versions of the k- turbulence model, Hogg et al. [6], Jones et al. [7], Sharif et al. [8], Chen et al. [9], Yaras et al. [10], and Yang et al. [11], carried out numerical simulation of a highly swirling flow (S=2.25) in a cylindrical combustor measured by So et al. [29]. It is reported that the k- model exhibits an excessive level of turbulent diffusion and its pre- dictions for the mean flowfield of the studied case [29] are not satisfactory. The deficiency of the k- model in *Address correspondence to this author at the Department of Mechanical and Manufacturing Engineering, The University of Manitoba, Winnipeg, Manitoba R3T 5V6, Canada; E-mail: biroukm@cc.umanitoba.ca predicting the turbulent diffusion is recognized in the simula- tion of other swirling flows in different combustor geo- metries and in a wide range of swirl numbers [12-28,30]. For example, Tsao et al. [28] simulate a can-type gas turbine combustor for two swirl numbers (S=0.74, and 0.85) and show that the k- model predicts a relatively higher level of deceleration of the axial velocity in the centerline region of the combustor which is a sign of excessive diffusion and hence higher level of swirl entrainment. However, later ver- sions of the k- model show improvement over the standard k- model in predicting the characteristics of swirling flows but still less accurate as compared to experimental data [16,19,26,31-34]. The persistent deficiency of these models is believed to be a result of their use of isotropic eddy- viscosity concept, while the structure of turbulent swirling flows is mostly anisotropic [35]. In addition, the eddy- viscosity models have difficulties in accounting properly for turbulence-swirl interactions. For instance, the RNG k- model [36] is employed to simulate several configurations of confined swirling flows [16,19,37]. Recall that the RNG k- and the standard k- differ mainly in the expression of the dissipation ( ) equation. In the RNG k- model a new term is introduced into the dissipation ( ) equation which results in an apparent success of this version of k- models in predict- ing the length of recirculation zones of several separating flows [37-39]. However, in some cases predictions of the RNG k- and the k- are not much different. For example, Xia et al. [19] examine both the standard k- and the RNG k- models for predicting a strongly swirling flow (S=1.68) in a water model combustion chamber, and find that both of the models give fairly accurate results near the inlet region but fail to reproduce accurately the downstream flow characteris- tics, although the RNG k- model is found to make a slightly improved predictions near the flow inlet. A major weakness of the standard k- model or other traditional k- models, such as RNG k- model, lies in their way of modeling the dissipation ( ) equation. The realizable k- model [40] is intended to address the deficiencies of these k- models by