Turbulent channel flow: DNS relaminarizes and LES shows wavy flow at finer grid

[nipinl]

I’m attempting DNS of turbulent channel flow by modifying the channel3D LES example: same geometry, increased resolution, wall functions replaced with bounce-back, and dynamics switched to BGK D3Q19 / MRT D3Q27. I get sustained turbulent flow for coarser(y+ = 4.5) mesh but gets laminar solution for fine mesh(y+ = 2.25), both when started from an LES warm-up and when started directly from initialization.

Interestingly, LES case (the example comes with version 1.7) works well for coarser mesh but show wavy flow at finer mesh. Setup notes: OpenLB 1.7; Mach capped < 0.1; forcing feedback retained; initialization with TI = 0.05. Velocity contours are attached.

What might I be overlooking in my DNS setup? Are there known pitfalls when using Smagorinsky on fine grids?

I can provide logs, converter prints, and the relevant code snippets if helpful. Thanks!

[nipinl]

Image

• This reply was modified 1 month ago by nipinl. Reason: Added Image

[nipinl]

Velocity contours: Velocity Contours

LOG: Y+ 2.25 DNS :

`[UnitConverter] —————– UnitConverter information —————–
[UnitConverter] — Parameters:
[UnitConverter] Resolution: N= 200
[UnitConverter] Lattice velocity: latticeU= 0.0254542
[UnitConverter] Lattice relaxation frequency: omega= 1.99611
[UnitConverter] Lattice relaxation time: tau= 0.500973
[UnitConverter] Characteristical length(m): charL= 0.00199
[UnitConverter] Characteristical speed(m/s): charU= 7.88506
[UnitConverter] Phys. kinematic viscosity(m^2/s): charNu= 1e-06
[UnitConverter] Phys. density(kg/m^d): charRho= 1000
[UnitConverter] Characteristical pressure(N/m^2): charPressure= 0
[UnitConverter] Mach number: machNumber= 0.0440879
[UnitConverter] Reynolds number: reynoldsNumber= 15691.3
[UnitConverter] Knudsen number: knudsenNumber= 2.80971e-06
[UnitConverter]
[UnitConverter] — Conversion factors:
[UnitConverter] Voxel length(m): physDeltaX= 9.95e-06
[UnitConverter] Time step(s): physDeltaT= 3.21201e-08
[UnitConverter] Velocity factor(m/s): physVelocity= 309.775
[UnitConverter] Density factor(kg/m^3): physDensity= 1000
[UnitConverter] Mass factor(kg): physMass= 9.85075e-13
[UnitConverter] Viscosity factor(m^2/s): physViscosity= 0.00308226
[UnitConverter] Force factor(N): physForce= 0.00950033
[UnitConverter] Pressure factor(N/m^2): physPressure= 9.59605e+07
[UnitConverter] ————————————————————-
[main] ———————————————————————-
[main] Converge time(s): 0.266661
[main] Lattice converge time: 8301991
[main] Max. Phys. simulation time(s): 0.355548
[main] Max. Lattice simulation time: 11069321
[main] Frequency Statistics Save(Hz): 25
[main] Statistics save period(s): 0.04
[main] Lattice statistics save period: 1245326
[main] ———————————————————————-
[main] Channel height(m): 0.00199
[main] y+ value: 2.25004
[main] y+ value spacing: 4.4776

[FBukreev]

Hello,

if you want to make DNS, then BGK and bounce back should be enough. Sometimes by too small time steps the flow becomes strange. You can try to take higher lattice Velocity in the converter if your maximal CFL number during actual simulation is too low.

[nipinl]

Hi Bukreev,
Thank you for your reply and confirming the dynamics and BC. I did not use a higher lattice velocity because I wanted to be conservative on the Mach Number side. Ma correspond to the characteristic velocity is 0.044 and maximum velocity goes 1.3 times the characteristic velocity. That said, a slightly higher lattice velocity gives similar result.

[FBukreev]

Hi,
BGK at higher resolution should give the turbulent result, because BGK has very low numerical diffusion itself.

[nipinl]

Hi Fedor,
As you suggested on the other thread, using DNS with ThirdOrderRLBdynamics solved the problem. Thanks again !!

[Author]

Posts