Simulation of Micro Flows by Taylor Series Expansion- and Least Square- based Lattice Boltzmann Method C. Shu, X. D. Niu, Y. T. Chew and Y. Peng Department of Mechanical Engineering Faculty of Engineering National University of Singapore Outline Introduction Standard Lattice Boltzmann method (LBM) TLLBM for Non-uniform Mesh and Complex geometries TLLBM for Micro Flows Numerical Examples Conclusions 1. Introduction Micro Flows in MEMS Classification of Flow Regimes according to Knudsen number 0.001 3 (10) 0.1 Continuum Flow Transition Regime (Ordinary Density Levels) (Moderately Rarefied) Kn L 100 Slip-Flow Regime Free-Molecule Flow (Slightly Rarefied) (Highly rarefied) Ma 2 Re 1 2n 2 Studies of Micro Flows – Experimental – Numerical and Theoretical Navier-Stokes Solver with Slip Condition – Only applicable to slip flow regime Particle-based Methods – MD (Molecular Dynamics) – DSMC (Direct Simulation Monte Carlo) – Applicable to slip flow and transition regimes – Huge computational effort Demanding More Efficient Solver for Micro Flow Simulation • Does LBM have Potential to be an Efficient Method? LBM is also a particle-based method Features of LBM and MD, DSMC Micromechanics of real particles by MD and DSMC; micromechanics of fictitious particles by LBM Local conservation laws are satisfied by all Tracking particles by MD and DSMC (Lagrangian solver); Dynamics of particles at a physical position and a given velocity direction by LBM (Eulerian solver) LBM More efficient than MD and DSMC 2. Standard Lattice Boltzmann Method (LBM) Particle-based Method (streaming & collision) D2Q9 Model Streaming process Collision process Isothermal Flows: f ( x ext , y eyt , t t ) f ( x, y, t ) [ feq ( x, y, t ) f ( x, y, t )] / feq 1 1 e U e U 2 U 2 2 2 4 2 2 c c 2 6 c N f 0 N U f e 0 (2 1) 2 c t P c / 2 8 2 Thermal Flows o Difficulty (stability and fixed Pr) o A number of thermal models available Multi-speed model; Hybrid model; two distributions model (passive scalar; IEDDF) o We prefer to use IEDDF for the thermal flows (He et al. 1998) f (mass) is used for velocity field g (energy) is used for temperature field Features of Standard LBM o Particle-based method o Density distribution function as dependent variable o Explicit updating; Algebraic operation; Easy implementation No solution of differential equations and resultant algebraic equations is involved o Natural for parallel computing Limitation---Difficult for complex geometry and non-uniform mesh Mesh points Positions from streaming 3. TLLBM for Non-uniform Mesh and Complex Geometry Current LBM Methods for Complex Problems o Interpolation-Supplemented LBM (ISLBM) He et al. (1996), JCP Features of ISLBM Large computational effort May not satisfy conservation Laws at mesh points Upwind interpolation is needed for stability Mesh points Positions from streaming o Differential LBM Taylor series expansion to 1st order derivatives f f f f eq ( x, y, t ) f ( x, y, t ) ex ey t x y t Features: Wave-like equation Solved by FD and FV methods Artificial viscosity is too large at high Re Lose primary advantage of standard LBM (solve PDE and resultant algebraic equations) • Development of TLLBM o Taylor series expansion f ( P , t t ) f ( P , t t ) f ( A, t t ) f ( P , t t ) x A y A x y 2 2 1 1 2 f ( P , t t ) 2 f ( P , t t ) (x A ) (y A ) 2 2 2 x 2 y 2 f ( P , t t ) x A y A O[( x A ) 3 , (y A ) 3 ] xy P-----Green (objective point) A----Red (neighboring point) Drawback: valuation of Derivatives o Idea of Runge-Kutta Method (RKM) du dt f (u, t ), u u0 , when t 0 Taylor series method: n n+1 2 3 du 1 d u 1 d u n 1 n 2 3 u u h h h , h t 2 3 dt 2 dt 6 dt Need to evaluate high order derivatives Runge-Kutta method: n Apply Taylor series expansion at Points to form an equation system n+1 Taylor series expansion is applied at 6 neighbouring points to form an algebraic equation system A matrix formulation obtained: o [ S ]{V } {g} (*) {V } { f , f / x, f / y, f / x , f / y, f / xy} 2 {g} {g i } Tg i f ( xi , yi , t ) 2 2 2 2 eq f ( xi , yi , t ) f ( xi , yi , t ) [S] is a 6x6 matrix and only depends the geometric coordinates (calculated in advance in programming) T / o Least Square Optimization Equation system (*) may be ill-conditioned or singular (e.g. Point coincide) Square sum of errors E M 2 erri i 0 g i si , jV j i 0 j 1 M 6 2 i 0,1,2,..., M ( M 5 for 2 D) M is the number of neighbouring points used Minimize error: E / Vk 0, k 1,2,...,6 Least Square Method (continue) The final matrix form: 1 {V } [ S ] [ S ] [ S ] {g} [ A]{g} T T [A] is a 6(M+1) matrix The final explicit algebraic form: M 1 f ( x0 , y0 , t t ) a1,k gk 1 k 1 (Shu et al. 2002 PRE, Vol 65) are the elements of the first row of a1,k the matrix [A] (pre-computed in program) Features of TLLBM o Keep all advantages of standard LBM o Mesh-free o Applicable to any complex geometry o Easy application to different lattice models Flow Chart of Computation Input , U , Re, e M 1 f ( x0 , y0 , t t ) a1,k gk 1 k 1 Calculating Geometric Parameter and physical parameters ( a1, k , N=0 ) N=N+1 M M 0 0 f U f e No Convergence ? Calculating eq f YES OUTPUT Boundary Treatment Bounce back from the wall Fluid Field Stream from fluid field f f Non-slip condition is exactly satisfied Square Driven Cavity (Re=10,000, Non-uniform mesh 145x145) Streamlines (right) and Vorticity contour (left) 4. TLLBM for Micro Flows LBM for macro flows is related to viscosity (m) through Chapman-Enskog expansion Bounce back rule (non-slip condition) Features of micro flows Knudsen number (Kn) dominant Velocity slip at wall Temperature jump at wall LBM for micro flows m relationship not valid. kn ? Bounce back rule not valid. New B. C. ? • Relationship of -Kn In Kinetic Theory: Relaxation time in velocity field: f m P Relaxation time in thermal field: g m Pr P (Pr m c p / ; c p R /( 1); 5 / 3) Knudsen number: Ma Kn L 2 Re • Relationship of -Kn In LBM: Mach number and Reynolds number: Ma Re U ref cs U ref H m We derived: f H Kn H Kn g Pr (cs 1 / 3 ) • Boundary Conditions For Micro Flows n Specular boundary condition Fluid Wall Diffuse-Scattering Boundary Condition (DSBC) Kinetic Theory: (ξ u w ) n f (ξ ) (ξ ' u w ) n (ξ ' ξ ) f (ξ ' ) dξ ' ( ξ ' u w )n 0 Maxwell Kernel: (ξ u w ) n (ξ u w ) 2 (ξ ξ ) exp( ) 2 RTw 2 ( RTw ) 2 1 ' w (2RTw 2 D ) 2 ((ξ u w ) n) f eq (ξ ) u u w RTw Normalized condition: ' ( )d 1 ( u w ) n 0 (( ' u w ) n 0) (incoming mass flux Equals to outgoing Mass flux) LBM Version Replace integral by summation (e u w ) n f ( e ' u w ) n 0 (e u w ) n g f (e ' e ) g (e ' e ) AN 6 (e ' u w ) n f (e ' e ) f ' BN (e ' u w ) n g (e ' ( e ' u w )n 0 AN w w ((e u w ) n) f eq ((e u w ) n) geq BN 3 ( D2Q9) e ) g ' u u w u u w , W Theoretical analysis of DSBC--a 2D constant density flow along an infinite plate with a moving velocity and a temperature u Kn 2 u U S u w U 0 Kn 2 n 2 n Kn T Kn 2T TS Tw T0 2 Pr n 2 Pr n 5. Numerical Examples High pressure Low pressure Pressure-Driven, Isothermal Micro Channel Flow U* 0.3 0.2 0.1 0.5 1 0.25 0 0.75 0 0.5 Y -0.25 0.25 -0.5 0 X Kn=0.05 Pr=2.0 Pressure-Driven, Isothermal Micro Channel Flow 0.002 0.001 V* 0 -0.001 -0.002 0.4 0 0.2 0.2 0 0.4 X 0.6 Y -0.2 0.8 -0.4 1 Kn=0.05 Pr=2.0 Non-linear Pressure Distribution along the Channel Kn=0.056, Pr=1.88 0.05 P' 0.025 Arkilic Spec. U. Ext. UCLA 0 0 0.25 0.5 X 0.75 1 Non-linear Pressure Distribution along the Channel Kn=0.155, Pr=2.05 0.05 P' 0.025 0 Arkilic Spec. U. Ext. UCLA 0 0.25 0.5 X 0.75 1 Pressure-Driven Microchannel Flows Kn=0.053 Kn=0.165 Mass Flow(Kg/s) Mass Flow(Kg/s) 5 1.2 Exe. data, Argon [23] 1 No-Slip 0.8 Present LBM Exe. data, Helium [24] 4 3.5 No-Slip Analytical [23] Present LBM 3 Analytical [23] 0.6 4.5 2.5 2 0.4 1.5 1 0.2 0.5 0 0 1 1.5 2 2.5 3 3.5 4 Pressure ratio 4.5 5 1 1.5 2 2.5 Pressure ratio 3 Comparison of slip magnitude at wall in the slip flow regime (shear-driven flow) Kn Analytic LBM 0.001 0.000998 0.0008628 0.00167 0.001667 0.0015939 0.002 0.001990 0.0019277 0.004 0.003968 0.0039492 0.005 0.004950 0.0049355 0.01 0.009800 0.0097963 0.02 0.019231 0.0192269 0.025 0.023800 0.0238064 0.03 0.028302 0.0280455 0.04 0.037037 0.0370351 0.05 0.045400 0.0454530 0.1 0.083300 0.0833333 Thermal Couette Flow High T Low T 2 1 T U 0.5 1.5 0 1 No-Slip No-Slip Kn=0.05 Kn=0.1 Kn=0.3 LBM(Kn=0.05) -0.5 LBM(Kn=0.1) 0.5 LBM(Kn=0.3) -1 -0.5 -0.25 Y 0 Velocity Profile 0.25 0.5 0 -0.5 -0.25 Y 0 0.25 Temperature Profile (Ec =10) 0.5 Slip length Temperature Jump 1.2 Ls 0.8 2.5 Present LBM Maxwel [19] MD [20] DSMC [20] Lj 2 Present LBM Maxwell [19] 1.5 1 0.4 0.5 0 0.01 0.1 Kn 1 0 0.01 0.1 Kn 1 Thermal Developing Flow (Constant Temperature on the walls Tw=10Tin) Temperature Z Y velocity Z Y X X 10 U1.5 20 1 5 15 0.5 10 x/H 0.5 5 0 y/H -0.5 0 20 15 0.5 0.25 y/H 10 0 -0.25 -0.5 0 5 x/H Thermal Developing Flow Thermal developing Channel flow---Wall coefficients along channel 55 18 Kn=0.015 Cf*Re Kn=0.03 45 Nu Kn=0.015 15 Kn=0.03 Kn=0.046 Kn=0.046 35 12 25 9 15 6 0.1 1 x/H 10 0.1 1 x/H 10 Thin Film Gas Slider Bearing Lubrication y 6mU w L / P0 H x 2 0 x=0 x=L Knudsen Number, Kn 0.8 2nd-order model 0.7 10 P 0.6 Load Carrying Capacity, W stress-density ratio model LBM 0.5 0.4 0.3 0.2 0.1 0 0 0.2 0.4 X/L 0.6 Kn=1.24, =123.2 1 10 10 0 10 =10 10-1 Boltzmann 10-2 10 0.8 0 1 Continuum stress-density ratio model second-order model LBM -3 10-1 100 Inverse Knudsen Number, D -1 Three-dimensional Thermal Developing Flow in Ducts Constant Heat flux at the wall; Initial flow is assumed static with a constant temperature; Re=0.1, Pr=0.7 Z Y H Flow W L X Three-dimensional Thermal Developing Flow in Ducts Z Y X . 0.5 Velocity distribution along duct Kn 0.06 Z Y 1.0462 1.0240 1.0110 1.0055 1.0037 1.0010 0.9998 0.9989 0.9979 X Thermal distribution along duct Velocity profile at different cross sections Z X Z Y X 1.5 Y 1.5 1 -0.5 -0.25 -0.125 z /w 0 0.125 0.25 -0.25 0.25 u/U0 u/U0 1 0.5 0 w y/ 0.5 0.5 -0.5 -0.25 -0.125 z /w 0 0.125 0.25 -0.25 0.25 0 w y/ 0.5 x=0.3442 x=0.0 Z Z X X Y Y 1.5 1.5 1 u/U0 0.5 -0.5 -0.25 -0.125 z /w 0 0.125 0.25 -0.25 0.25 0.5 x=0.7826 0 w y/ u/U0 1 0.5 -0.5 -0.25 -0.125 z /w 0 0.125 0.25 -0.25 0.25 0.5 x=1.230 0 w y/ Temperature profile at different cross sections Z X Z Y X 1.5 Y 1.5 1 1 -0.5 -0.25 -0.125 z /w 0 0.125 0.25 -0.25 0.25 T/T0 T/T0 0.5 0 w y/ 0.5 0.5 -0.5 -0.25 -0.125 z /w 0 0.125 0.25 x=0.0 -0.25 0.25 0 w y/ 0.5 x=0.3442 Z Z X X Y 1.5 1.5 1 1 -0.5 -0.25 -0.125 z /w 0 0.125 0.25 -0.25 0.25 0.5 x=0.7826 0 w y/ T/T0 T/T0 0.5 0.5 Y -0.5 -0.25 -0.125 z /w 0 0.125 0.25 -0.25 0.25 0.5 x=1.230 0 w y/ Three-dimensional Thermal Developing Flow in Ducts: Comparison Kn 0.04 Nu Us Slip model Present Slip model Present 1 0.21 0.25 2.85 3.16 0.75 0.25 0.27 2.81 3.06 0.5 0.32 0.30 2.71 2.92 Kn 0.06 1 0.28 0.31 2.69 2.87 0.75 0.33 0.33 2.62 2.79 0.5 0.41 0.36 2.48 2.66 1 0.34 0.37 2.53 2.60 0.75 0.39 0.38 2.44 2.54 0.5 0.48 0.41 2.26 2.44 Kn 0.08 6. Conclusions TLLBM is an efficient method for simulation of different micro flows New relationship of -Kn is valid Specular and diffuse-scattering boundary conditions can capture velocity slip and temperature jump Not accurate for cases with high pressure ratio (an incompressible model)