Progress on Engine LES Using STAR-CD

Transcription

1 Progress on Engine LES Using STAR-CD A D Gosman CD-adapco Japan STAR Conference 2012, Yokohama

2 INTRODUCTION 1. Nature and motivation for LES of engines 2. LES modelling in STAR-CD 3. Collaborations 4. Validation studies 5. Applications

3 MOTIVATION FOR ENGINE LES COMPARISON LES AND RANS RANS models all scales of fluctuation around mean LES models only small scales and directly computes larger ones ACTUAL LE S RAN S ADVANTAGES OF LES OF ENGINES because only small scales modelled, is inherently more accurate. calculate individual cycles, so obtain information on cycle-to-cycle variations principal motivation small-scale modelling easier than RANS modelling in some, but not all areas. DRAWBACKS OF LES moderately more expensive/cycle, but need many cycles to get statistics more effort to set up, e.g. more detailed boundary conditions required puts more demands on numerical solver both accuracy and speed.

4 CD-adapco ACTIVITIES IN ENGINE LES IN-HOUSE DEVELOPMENT OF LES CAPABILITIES several experts in LES modelling and numerics current focus on combustion UNIVERSITY COLLABORATION AND SPONSORSHIP Penn State Univ.: Professor D Haworth Cornel Univ: Prof S Pope Darmstadt Univ: Prof J Janicka Modena Univ: Prof S Fontanesi Imperial College London: Prof W P Jones COLLABORATIVE PROJECTS WITH INDUSTRIAL PARTNERS ENGINEERING SERVICES PROJECTS

5 ENGINE LES/DES MODELLING IN STAR-CD: OVERVIEW FULL FRAMEWORK FOR MODELLING SI AND DIESEL ENGINES LES and DES modelling of flow and mixing - optional LES subgrid turbulence models - hybrid DES model combining URANS for near-wall and LES for bulk LES-adapted Lagrangian modelling of sprays (V4.20) - special subgrid modelling of gas-droplet interactions - full capabilities as in URANS version: atomisation, wall impingement, wall film General ECFM3Z-LES modelling of SI and Diesel combustion and emissions - LES version of ECFM-3Z model, with full capabilities - applicable to all combustion modes: full/partially premixed, diffusion - optional SI ignition models - under test, scheduled for release in STAR-CD V4.20

6 LES SUBGRID MODELLING OPTIONS Filtered LES equations, subgrid viscosity modelling Options 1. Smagorinsky S ij = 1 æ u i + u j 2 ç è x j x i ö ø 2. One-equation subgrid k model transport equation solved for k sgs

7 DES MODELLING I INTRODUCTION Hybrid non-zonal model: - tends to LES in resolved flow - tends to URANS in unresolved Automatic selection of length scale according to grid:turbulence length scale ratio Preferable to limit URANS to near-wall region Several variants based on different URANS models - Spalart-Almaras - k-ω SST - k-ε

8 DES MODELLING II - EXAMPLE SPALART-ALMARAS DES MODEL one-equation model: both high-re and low-re versions n = n t f n1 n t at high Re dissipation rate depends on controlling turbulent length scale d d = wall normal distance, Ψ 1 at high Re tends to Smagorinsky-type LES model when C DES Δ/d > 1

9 ECFM3Z-LES COMBUSTION MODELLING I OUTLINE Full ECFM-3Z adapted for LES coherent flame model for premixed burning eddy dissipation model for diffusion burning 3-zone modelling for sub-grid mixing equilibrium modelling for burnt gas Zeldovich NO x chemistry range of soot models range of knock models Two optional spark ignition models 1. Eulerian AKTIM 2. AKTIM adapted for LES

10 ECFM3Z-LES COMBUSTION MODELLING II LES FLAME SURFACE DENSITY EQUATION Transport equation for flame surface density T res = resolved transport T sgs = subgrid transport P = laminar propagation S res = resolved strain S sgs = subgrid strain C res = resolved curvature C sgs = subgrid curvature

11 KEY NUMERICAL ALGORITHM FEATURES Implicit PISO with quasi 2 nd order time differencing Second-order centered or MARS spatial differencing for momentum MARS for species and other scalars Synthetic turbulence inlet conditions Non-reflecting inlet/outlet conditions

12 QUALITY ASSESSMENT CRITERIA FOR LES 1. A-priori ratio integral scale/mesh size = l int /Δ use RANS estimate l int = C m 0.75 k 3/ 2 /e want ratio < 0.5 accuracy depends on RANS solution 2. Fraction resolved kinetic energy k res /k tot k res 1 2 u' 2 1 u' 2 2 ' 2 u' 3 ; u i u i u i ; k tot k res k sgs want ratio > Ratio LES predicted turbulence scale/mesh size obtain length scale from energy spectrum 4. Ratio turbulent: laminar viscosity ideally close to unity, minimizes modelling error contribution 5. Other, e.g. Index of Resolution Quality

13 VALIDATION: MOTORED MODEL ENGINE (PENN STATE U): Axisymmetric, central open valve, flat piston LES simulations, without/with swirl Smagorinsky, wall functions 1.3M cell mesh, size ~ 1 mm effects of mesh (170K,2.6M), time step ( CAD), subgrid model (1 eqn) Model Engine Configuration Flow during Intake

14 MODEL ENGINE: LES AND RANS COMPARED WITH MEASUREMENT LES better than RANS for same mesh and time step

15 MODEL ENGINE: RESOLVED AND SUBGRID KINETIC ENERGY Subgrid-scale energy generally is small relative to resolved-scale energy and decreases with mesh refinement 170K cells 1.3M cells 2.6M cells

16 MODEL ENGINE: CYCLIC VARIATIONS AND POD ANALYSIS Cyclic variations investigated using Proper Orthogonal Decomposition (POD) Similar study underway for the Transparent Combustion Chamber (TCC) engine

17 VALIDATION: MOTORED 4-VALVE PENT-ROOF ENGINE (UNIV DARMSTADT) ENGINE 4-valve, pent roof, flat piston optical access measurements with laser diagnostics exhaust intake COMPUTATIONAL SET-UP Smagorinsky, MARS, Piso time-dependent inlet and outlet pressure boundary conditions fixed-temperature inlet variable time step, linked to valve motion mesh sizes 2.1M (base), 3.2M(refined) corresponding mean cell sizes1.2, 0.8mm up to 25 successive cycles exhaust intak e

18 4-VALVE PENT-ROOF ENGINE: EFFECT OF NUMBER OF CYCLES Total cycles = 25; phase averages over increasing periods Intake (270bTDC) Compression (90bTDC) rms phase averaged mean

19 y ( mm ) y ( mm ) y ( mm ) 4-VALVE PENT-ROOF ENGINE: FLOW FIELD AT 270 O BTDC (INDUCTION) base 90 btdc V [m/s] refined x ( mm ) x ( mm ) measured x ( mm

20 4-VALVE PENT-ROOF ENGINE: VALIDATION AT 270 O BTDC (INDUCTION) y = 0mm Phase averaged u velocity RMS u velocity y = -10mm

21 y ( mm ) y ( mm ) y ( mm ) 4-VALVE PENT-ROOF ENGINE: VELOCITY FIELD AT 90 O BTDC (COMPRESSION) base 90 btdc V [m/s] refined x ( mm ) x ( mm ) measured x ( mm

22 4-VALVE PENT-ROOF ENGINE: VALIDATION AT 90 O BTDC (COMPRESSION) Phase averaged u velocity RMS u velocity y = 0mm y = -10mm

23 4-VALVE PENT-ROOF ENGINE: RESOLVED KINETIC ENERGY FRACTION 90 btdc V [m/s] base refined

24 4-VALVE PENT-ROOF ENGINE: CONCLUSIONS OF STUDY 1. LES is more accurate than RANS and gives good agreement with experiment. 2. For reasonable accuracy require: Grid size not more than Δ 1.0 mm to capture 80% - 95% of the TKE (except jets) 25 to 50 engine cycles to get accurate statistical means and fluctuations. 3. Cyclic variations most likely to be caused by random nature of turbulence. 4. Accuracy of RANS is not bad, especially for mean velocity.

25 COMBUSTION IN 4-VALVE HIGH-SPEED GDI TURBOCHARGED ENGINE (U.MODENA) ENGINE 4-valve GDI pent-roof, shaped piston turbocharged, rpm test-bed measurements COMPUTATIONAL SET-UP STAR-CD development version time-varying inlet/outlet pressures from tuned 1D analysis Smagorinsky, wall functions, ECFM-LES, simple ignition model 1.3M mesh, refined around spark plug initialisation from RANS solution up to 25 cycles research in progress

26 4-VALVE HIGH-SPEED GDI: CYLINDER PRESSURE PREDICTIONS AND COMPARISON

27 4-VALVE HIGH-SPEED GDI: CYCLIC VARIATIONS IN FLAME PROPAGATION 27

28 4-VALVE HIGH-SPEED GDI: CYCLIC VARIATIONS IN FLOW AND EQ RATIO Interaction between velocity field & flame development Equivalence Ratio 28

29 4-VALVE HIGH-SPEED GDI: KNOCK ONSET PREDICTION FOR HIGH PRESSURE CYCLE Experimental Spark Highly Increased Spark Advance Advance Spark Advance - Moderate Increase Yellow Isosurface - RVB=0.5 Red Isosurface - Autoigniting Condition

30 4-VALVE HIGH-SPEED GDI: KNOCK ONSET PREDICTION FOR LOW PRESSURE CYCLE Experimental Spark Advance Spark Advance High Increase Spark Advance - Moderate Increase Yellow Isosurface - RVB=0.5 Red Isosurface - Autoigniting Condition

31 SUMMARY 1. Adaptation of STAR-CD for LES and DES of engines is well advanced, with involvement of internal and external experts and collaborations with university and industrial partners. 2. Extensive validations have been performed for motored operation, with satisfactory results. 3. Current development and validation activity for engine combustion is also producing promising results. 4. The results confirm the clear benefits of LES in improved accuracy and capability to predict cyclic variability. 5. The experience gained is also helping to identify best practices for LES/DES.

32 ECFM3Z-LES COMBUSTION MODELLING III BASIC IGNITION MODEL Initialise reactedness profile evolution of time mean surface area wrinkled surface area wrinkle factor evolution convert Sm into flame surface density???

33 ECFM-3Z LES Ignition (two options in STAR-CD). Basic Model [1,2] AKTIM [4] with extensions for LES Basic Model: Initialise profile for burnt gases volume fraction c as: with constraint with rk the given initial kernel radius from the RANS model Model evolution in time of the mean flame surface area Sm(t) with:

34 ECFM-3Z LES Basic Model (continued): The mean surface Sm will be wrinkled by the turbulence: with Ξ the wrinkling factor given by the equation: with S_l, a_t the laminar flame speed and the sgs strain. Sm is converted into the Flame Surface Density FSD by:

35 Extensive parametric studies have been performed Formulation Computational mesh Pressure-correction algorithm LES subfilter-scale turbulence model RAS turbulence model Computational timestep Piston motion LES or RAS 170K, 1.3M, or 2.6M cells SIMPLE or PISO Constant-coefficient Smagorinsky (C S = ) or 1-eqn. subfilter-scale k model (C k = ), with wall damping Standard high-re k-e w/wall functions Dq = CAD Fixed number of cells w/deformation or mesh addition/deletion Baseline: LES, 1.3M cells, PISO, 1-eqn. w/c k =0.3, Dq=0.1 CAD, fixed # of cells In all cases: