Large-eddy simulation for an axisymmetric piston-cylinder assembly with and without swirl

K. Liu and D.C. Haworth

Department of Mechanical and Nuclear Engineering, The Pennsylvania State University, University Park, PA, 16802, USA

Introduction

Large-eddy simulation (LES) is increasingly used as a tool for studying the dynamics of turbulence in engineering ows. In LES, one explicitly captures the dynamics of the large eddies while modeling the eects of the smaller eddies on the larger ones. Because the statistics of small scale turbulence are expected to be more universal than those of the large scales, LES oers the promise of wider generality and more accurate results compared to Reynolds-averaged Navier-Stokes (RANS) where the eects of all turbulence scales are modeled. The shortcomings of RANS models in IC engine have been argued by many turbulence researchers in [1, 2, 3] and [4]. In contract to RANS, LES oers advantages which are particularly compelling for the ICengine application [1, 2, 3, 4], such as the cycle-to-cycle ow and combustion variability. So, the transition of in-cylinder CFD from RANS to LES is a natural direction. El Tahry and Haworth [1] argued in 1992 that the computational meshes typically used for RANS modeling of practical in-cylinder congurations should be sucient to capture 80-90% of the ows kinetic energy. Other arguments for pursuing LES for in-cylinder CFD have been made in [4] and [5]. Jansen and Haworth [2] reported encouraging results using LES for predicting the ensemble-averaged mean and rms velocity proles for a simplied motored engine conguration. This research addresses important outstanding issues for in-cylinder LES in the context of simplied engine-like congurations. These includes: systematic comparisons between a conventional two-equation RANS model and LES; a systematic parametric study of the eects of physical model parameters and numerical parameters; quantication of the qualify of the LES solutions through an examination of the relative magnitudes of resolved-scale and sublter-scale uctuations; and the rst LES results reported to data for an axisymmetric conguration with swirl.

Experimental congurations

The axisymmetric piston-cylinder assembly that is the subject of this study is shown schematically in gure 1. Key geometric parameters and operating conditions are summarized in gure 1 [6]. This is a pancake (at head and piston) chamber with a piston moves in simple harmonic motion. Flow enters the chamber through an annular passage through a xed, open valve. For the swirling cases, swirl vanes are added upstream of the valve around the valve stem. The working uid is air, which is treated as an ideal gas. The initial pressure and temperature are uniform at 1 atm and 300 K, respectively.
Full Professor, Department of Mechanical and Nuclear Engineering, The Pennsylvania State University, 232 Research Building East, University Park, PA 16802.

For both swirling and nonswirling cases, Laser-Doppler anemometry has been used to obtain ensemble(phase-) averaged radial proles of mean and rms velocity components at 10-mm axial increments starting from the head for crank positions of 36 , 90 , and 144 after piston top-dead-center (TDC) [6, 7]. For nonswirling cases, only the axial velocity component was measured and measurements also were reported at 270 after TDC. For the swirling case, mean and rms proles of all three velocity components (axial, radial, and tangential) were reported.

Figure 1: Axisymmetric piston-cylinder assembly. [6].

3
3.1

Physical models and numerical methods

Sublter-scale models

In LES, large scales are resolved and sublter-scale models then are introduced to account for the eects of the unresolved scales. Two SFS models are considered here: a Smagorinsky SFS model, and a one-equation SFS model. 3.1.1 Smagorinsky Sublter-scale model

The Smagorinsky model is the most commonly used SFS model [8]. It is derived from a local equilibrium assumption: the production and dissipation of SFS turbulent kinetic energy are equal. It can be written in the following form [8]: SF S = Cs 2 21/2 S , (1) here S = (Sij Sij )1/2 is the Frobenius norm of the resolved strain-rate tensor, and is taken as Vcell [9], where Vcell is the volume of a computational cell. The baseline value of the model parameter Cs is taken to be the square of the classic Smagorinsky constant (0.165) [10, 11] and the values can be adjusted following the recommendations available in the literature for shear ows and free ows [12, 13]. 3.1.2 One-equation SFS model
1 /3

Speziale [14] recommended solving a transport equation for kSF S rather than using a local equilibrium assumption for the SFS kinetic energy. A modeled transport equation for kSF S can be written as, k uj kSF S kSF S 1/2 kSF S + = SF S,ij Sij C1 SF S + [C2 kSF S ], t xj xj xj
3/2

(2)

Here the model constants C1 and C2 are derived from turbulence theory: C1 = 1 and C2 = 0.05. The turbulent kinetic energy thus obtained is then used as a velocity scale for the SFS viscosity [15, 16, 17]: SF S = C2 kSF S .
1/2

(3)

3.2

Numerical methods and computational mesh

A commercial CFD code, STAR-CD version 4.06 [18], has been used for this study. Central dierencing is used for the convective terms in the momentum equation and the PISO algorithm [16] is used for pressurevelocity coupling. Values of model parameters recommended for LES applications are provided in [19, 20]. The baseline mesh for the present study contains approximately 1.3 million cells, and corresponds to a factor of 2 mesh renement in each direction with respect to the coarse mesh. A ne mesh of approximately 2.6 million cells (additional factor of 2 mesh renement in the axial direction only) was used in some cases. To accommodate the movement of the piston, the in-cylinder cells are compressed or expanded in the axial direction without changing the mesh topology.

3.3

Initial and boundary conditions and the estimation of mean quantities

The initial pressure and temperature for the IC engine are uniform at 1 atm and 300 K, respectively. The no-slip boundary condition is applied at solid walls, and all walls are adiabatic. There is a large plenum upstream of the valve so that no inow-outow boundary conditions are needed. The volume of the plenum is approximately 50 times larger than that of the in-cylinder region, so that the global pressure and temperature vary less than 2% through each engine cycle. The computations begin at bottom-dead-center (BDC). For nonswirling cases, the simulations are run through seven engine cycles, and the rst two cycles are discarded to avoid contamination by initial conditions. Mean quantities then are estimated by azimuthal-averaging and ensemble-average the resolved-scale computed quantities over ve engine cycles. For the swirling case, the simulations were run for ten cycles and the rst ve cycles were discarded. A rotating body force was added in the momentum equation in the region upstream of intake valve to mimic the eect of swirl vanes. The magnitude of the body force was adjusted to give the correct global level of in-cylinder swirl.

4
4.1

Results
Nonswirling cases
t = 2 CA Cs = 0.00 Cs = 0.30 Ck = 0.00 Ck = 0.20 Coarse mesh 170,000 cells t = 1 CA Cs = 0.01 Cs = 0.50 Ck = 0.01 Ck = 0.30 Baseline mesh 1,300,000 cells t = 0.5 CA Cs = 0.02 Cs = 0.80 Ck = 0.02 Ck = 0.40 Finest mesh 2,600,000 cells t = 0.25 CA Cs = 0.10 Cs = 1.00 Ck = 0.05 Ck = 0.50 t = 0.1 CA Cs = 0.20 Cs = 1.50 Ck = 0.10 Ck = 0.80

Time Sensitivity Smagorinsky SFS model One-equation SFS model Mesh size

Table 1: LES run matrix for the nonswirling case. Values in bold font correspond to the baseline case.

Table 1 shows the run matrix for the nonswirling case. The parameters in bold are the baseline parameters. The computed and measured radial proles of axial rms velocity at 36 after TDC with baseline parameters are shown in gure 2. In these gures, only the resolved-scale contribution to the rms values is 3

shown. The results at other crankangles are also studied. In general, computed mean velocity proles are in good agreement with the experimental proles at 36 , 144 and 270 . The ow undergoes a transition in structure at 90 , and it is dicult to capture the phasing of the transition precisely. Better agreement between model measurement can be achieved at 90 with dierent model parameters, but at the expense of poorer agreement at 36 and 144 . The computed rms proles (resolved-scale uctuations only) at z = 20 mm at 36 underpredicts the experimental proles. If should be noted that at this measurement section the local rms velocity is of the same magnitude as the local mean velocity. While the agreement between model and measurement is not perfect, both the mean and rms computed proles show better agreement with measurements than has been reported using any RANS-based model

(a)

(b)

Figure 2: Computed (lines) and measured (symbols) radial proles at 36 after TDC for the baseline case. (a) Mean axial velocity proles. (b) Axial rms velocity proles.

The contribution of sublter-scale velocity uctuations is investigated. Here w SF S has been estimated 2 as the square root of 3 kSF S . At most locations, the sublter-scale (model) contribution is small compared to the resolved-scale contribution. The sensitivity of LES to key numerical and physical model parameters has been investigated. Based on these results, it is recommended that the computational time step should correspond to a maximum material Courant number of 0.04 or smaller. Results are especially sensitive to mesh and to the SFS turbulence models. Satisfactory results can be obtained using simple viscosity-based SFS turbulence models, although there is room for improvement. No single model coecient gives uniformly best agreement between model and measurements.

4.2

Swirling case

A systematic parametric study also has been performed for the swirling case (Table 2). The general trends are similar to those found in the parametric study for the nonswirling case. Hence the comparison in this section will be limited to mesh sensitivity and comparison between LES and RANS with a standard k turbulence model. The one-equation SFS model with Ck = 0.05 gives the best results overall and that is the value used for all LES results that are shown here. Time Sensitivity Smagorinsky model One-equation model Mesh size t = 1 CA Cs = 0.02 Ck = 0.02 Coarse mesh 170,000 nodes t = 0.1 CA Cs = 0.20 Ck = 0.05 Baseline mesh 1,300,000 nodes Cs = 0.30 Ck = 0.10

Table 2: LES run matrix for the nonswirling case

Figure 3 shows the completed and measured mean and rms of azimuthal velocity components proles at 4

36 after TDC. In general, the LES results are in at least as good agreement with experimental as for the nonswirling cases, and arguably are even better here. The LES results show a linear improvement over the RANS results with k turbulence model.

(a)

(b)

Figure 3: Computed (lines) and measured (symbols) radial proles at 36 after TDC for the baseline case. (a) Mean azimuthal velocity proles. (b) Azimuthal rms velocity proles.

Conclusion

LES has bee performed for an axisymmetric piston-cylinder assembly with and without swirl. For both swirling and nonswirling cases, the mean and rms velocity proles show better level of agreement with experimental data than a RANS-based k calculation. For the meshes have been used in the study, more than 80 % of the TKE is resolved, and the amount that is resolved increases as the mesh is rened. The sensitivity of LES to key numerical and physical model parameters has been investigated. Based on these results, it is recommended that the computational time step should correspond to a maximum material Courant number of 0.04 or smaller. Results are especially sensitive to mesh and to the SFS turbulence models. Satisfactory results can be obtained using simple viscosity-based SFS turbulence models, although there is room for improvement. No single model coecient gives uniformly best agreement between model and measurements. Although good results has been obtained, we are still not able to nd the perfect coecients in Smagorinsky and Kinetic sublter-scale model that give the best results at all crank angles and all locations. More advanced sublter-scale model such as dynamic SFS model and Rutland nondissipative one-equation structure function model will be studied in the future. Also, we are very interested in apply the LES on more realistic engine such as the Transparent Combustion Chamber (TCC) engine, which is a simplied real engine with a pancake chamber and inlet, outlet valves movement. We will made quantitative comparison between LES and PIV measurements.

Acknowledgments
The authors thanks Mr. Navtej Singh of CD-adapco Application Support for many helpful discussions related to the use of STAR-CD. The authors also acknowledge the support from the GM R&D Center, and from CD-adapco.

References
[1] S. E. Tahry and D. Haworth, Directions in turbulence modeling for in-cylinder ows in reciprocating engines, AIAA Journal of Propulsion and Power, vol. 8, pp. 10401048, 1992. 5

[2] D. Haworth and K. Jansen, Large-eddy simulation on unstructured deforming meshes: towards reciprocating ic engines, Computers and Fluids, vol. 29, pp. 493524, 2000. [3] D. Haworth, Large-eddy simulation of in-cylinder ows, in Oil and Gas Science and Technology, vol. 54, pp. 175185, 1999. [4] D. Haworth, A review of turbulent combustion modeling for multidimensional in-cylinder cfd, SAE paper, no. 2005-01-0993, 2005. [5] M. Drake and D. Haworth, Advanced gasoline engine development using optical diagnostic and numerical modeling, proceedings of the combustion Institute, vol. 31, pp. 99124, 2007. [6] J. W. AP. Morse and M. Yianneskis, Turbulent ow measurement by laser doppler anemometry in a motored reciprocating engine, Imperial College Department of Mechanical Enginnering Report FS/78/24, 1978. [7] J. W. AP. Morse and M. Yianneskis, The inuence of of swirl on the ow characteristics of a reciprocating piston-cylinder assembly, Imperial College Department of Mechanical Enginnering Report FS/78/41, 1978. [8] J. Smagorinsky, General circulation experiments with the primitive equations, Monthly Weather Review, vol. 91(3), pp. 99164, 1963. [9] R. Peyret, Handbook of Computational Fluid Dynamics. Academic Press, London, 2000. [10] D. Lilly, The representation of small-scale turbulence in numerical simulation experiments, in IBM Science and Computing Symposium on Environmental Science, (Yorktown Heights, New York, USA), 1967. [11] F. D. Mare, Large eddy simulation of reacting and non-reacting turbulent ows in complex geometries. PhD thesis, University of London, 2002. [12] U. Piomelli and J. Chasnov, Large-eddy simulations: theory and applications, vol. 7. Kluwer Academic, Dordrecht, 1996. [13] J. Deardor, On the magnitude of the sub grid scale eddy coecient, Journal of Computational Physics, vol. 7, pp. 120133, 1970. [14] C. Speziale, Analytical methods for the development of reynolds-stress closures in turbulence, Annual Review of Fluid Mechanics, vol. 23, pp. 107157, 1991. [15] A. Yoshizawa, Statistical theory for compressible turbulent shear ows, with the application to subgrid scale modeling, Physics of Fluids, vol. 29, pp. 21522164, 1986. [16] R. Issa, Solution of the implicitly discretised uid ow equations by operator-splitting, Journal of Computational Physics, vol. 62, pp. 4065, 1986. [17] A. G. R.I. Issa and A. Watkins, The computation of compressible and incompressible recirculating ows by a non-iterative implicit scheme, Journal of Computational Physics, vol. 62, pp. 6682, 1986. [18] http://www.cd-adapco.com/products/star-cd/index.html, [19] User guide for STAR-CD VERSION 4.06. CD-adapco, 2008. [20] Methodology for STAR-CD VERSION 4.06. CD-adapco, 2008.