Inlet Conditions for Large Eddy Simulation of Gas-Turbine Swirl Injectors M. H. Baba-Ahmadi * and G. R. Tabor † University of Exeter, Exeter, England EX4 4QF United Kingdom DOI: 10.2514/1.35259 In this paper, we present a novel technique for generating swirl inlets for large eddy simulation. The velocity a short distance downstream of the inlet to the main domain is sampled and the ﬂow velocity data are reintroduced back into the domain inlet, creating an inlet section integrated into the main domain in which turbulence can develop. Additionally, variable artiﬁcial body forces and velocity corrections are imposed in this inlet section, with feedback control to force the ﬂow toward desired swirl, mean, and turbulent proﬁles. The method was applied to ﬂow in an axisymmetric sudden expansion, with and without swirl at the inlet, and compared against experimental and literature large eddy simulation data and against similar results in the literature. The method generates excellent results for this case and is elegant and straightforward to implement. I. Introduction S WIRL injectors have been widely adopted in combustion systems such as gas-turbine engine combustors to stabilize the ﬂame for efﬁcient and clean combustion. Breakdown of the incoming swirl vortex in the central toroidal recirculation zone creates high shear rates and strong turbulence intensities that act as a ﬂame stabilization mechanism. In addition, the swirl also produces high rates of entrainment and fast mixing. Investigation of these mechanisms is obviously of great interest. Traditionally, designers have relied heavily on empirical correlations for determining overall geometries, dimensions, etc. This approach is now supplemented with theoretical and computational modeling techniques, which have the ability to predict physical phenomena over a wide range of conditions, in addition to providing a better insight into the ﬂuid dynamics. Modeling of these processes, however, is extremely complicated. In particular, swirling ﬂows are difﬁcult to model with Reynolds-averaged Navier–Stokes (RANS) methods due to the effect of the mean ﬂow streamline curvature [1], and so this is one example in which large eddy simulation (LES) methods have come to the fore. However, there are still numerous technical issues to be overcome in implementing LES as a technique [2]. In particular, the provision of adequate boundary conditions (for the case of swirl injection, this particularly means inlet conditions) is one very signiﬁcant hurdle to be overcome, and this is the subject of the current paper. Implementing inlets for LES is signiﬁcantly more challenging than is the case for RANS models; the inlet ﬂow has to include the grid-scale (GS) turbulence, and so has to include a stochastically ﬂuctuating component that satisﬁes a range of conditions (such as the correct temporal and spatial correlation). Thus, the topic of this paper is of great importance for the adoption of LES in this area. Two approaches to creating inlet conditions for swirling ﬂow have been applied in the literature. The simplest approach is to create a mean ﬂow proﬁle by determining the axial and tangential mean ﬂow components, either from previous computational work (using RANS), from experiment, or from theory, and to impose a speciﬁed level of ﬂuctuation on top of this, usually as Gaussian white noise. Examples of this approach include [3,4]. However, such approaches suffer problems related to the nonphysical nature of the turbulence introduced at the inlet, leading to incorrect prediction of turbulent kinetic energy and energy spectra downstream of the inlet [5]. Creating an appropriate inlet condition for LES is considerably more challenging than is the case for RANS; because there is no implicit scale separation in LES between simulated and turbulent ﬂow, the grid-scale (explicitly simulated) ﬂow contains a transient component due to turbulent velocity ﬂuctuations, a component that has to be included at the inlet. Moreover, this transient component has to possess most, if not all, of the characteristics of the turbulence that it is representing, including correct spatial and temporal correlation properties, something that white noise fails to satisfy. More sophisticated synthesis techniques have been developed using approaches such as digital ﬁltering and the Fourier series to introduce appropriate correlations [6–8], but these have not been applied to swirling ﬂows to date. The alternative approach to generate a turbulent inlet for LES is via a turbulence-library database. Typically, this involves running a precursor simulation on a simpler geometry (e.g., a cyclic channel), to create fully developed turbulence; successive time steps of this simulation are then saved and replayed into the inlet of the main simulation. Various variants of the technique have been tried, for example, running the precursor simulation in parallel with the main simulation (thus obviating the need to store a limited database of information [9]) and scaling the data using the Reynolds stress (to adjust an existing database to another Reynolds number [10]). In the context of swirling ﬂows, most versions of this technique make use of a method developed by Pierce and Moin [11] for generating swirl within a cyclic channel by imposing a constant tangential body force on the ﬂow. Having computed a library of turbulent swirling ﬂow in this way, either as a saved precursor database or “on the ﬂy” in parallel with the main calculation, the ﬂow conditions from the secondary calculation can be fed into the main computation [12,13]. As an example of this, Wang and Bai [5] used Pierce and Moin’s [11] method to create a 10,000 time-step library for lookup, which was then cycled through as appropriate. The library does not, however, meet the speciﬁcations for the required ﬂow, and so the data are rescaled to meet the desired statistical properties (speciﬁed mean and variance of velocity). However, this rescaling does cause problems; the level of turbulent kinetic energy is seen to decrease downstream of the inlet, which the authors attribute to the unphysical turbulence at the inlet adapting to become true turbulent ﬂow further downstream. Schlüter et al. [14] also implemented and compared various inlet conditions for swirl: speciﬁcally, a laminar inﬂow (no ﬂuctuations), inﬂow with random ﬂuctuations, and various precomputation methods. As Received 22 October 2007; revision received 21 February 2008; accepted for publication 26 February 2008. Copyright © 2008 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved. Copies of this paper may be made for personal or internal use, on condition that the copier pay the $10.00 per-copy fee to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923; include the code 0001-1452/08 $10.00 in correspondence with the CCC. * Senior Lecturer, School of Engineering, Computer Science and Mathematics. † Ph.D. Student, School of Engineering, Computer Science and Mathematics. AIAA JOURNAL Vol. 46, No. 7, July 2008 1782