Turbulence

1
There are no “simple” turbulent flows
Turbulent boundary layer:
• Instantaneous velocity field (snapshot)

Ref: Prof. M. Gad-el-Hak, University of Notre Dame

 DNS/LES – RANS

DNS

RANS

3
 DNS/LES – RANS

DNS

DNS (and also LES):

• 3D
• Time dependent
– Full information of turbulence scales
– Acoustics and dynamic loads
– No (limited) turbulence modelling
• Huge Reynolds number dependency

Expensive! 4
 DNS/LES – RANS
RANS:
• Reduction of dimensions –> cheap
Here: 2D and steady
• Only statistical information of turbulence scales:
Time and length scales
rms values
• Turbulence model needed

RANS

5
 Prediction of turbulent flows

standard
RANS
Approximation
level

advanced
RANS

hybrid
RANS-LES

LES
DNS
Log(CPU)
days weeks months years
 Standard RANS models

Available in commercial codes

standard •k-e
RANS
•k-w, e.g. Menter SST
Approximation

•Spalart-Allmaras
level

advanced Good (only) for:

RANS
•Attached boundary layers
hybrid •Thin shear flows
RANS-LES

LES
DNS
Log(CPU)
days weeks months years
 Advanced RANS models

•Differential Reynolds stress models, DRSM, RST

standard •Algebraic Reynolds stress models, EARSM
RANS
•Structure-based modelling (Kassinos & Reynolds)
Approximation

•Two-point correlations
level

advanced Needed for:

RANS •Rotation and swirl

hybrid •Strong 3D effects

RANS-LES •Body forces
LES
DNS
Log(CPU)
days weeks months years
 Direct Numerical Simulation – DNS

•Valid (!)
standard But:
RANS •no approximation
•Extremely expensive
Approximation

•Universal
•Only low Re
level

•Full information
advanced •Full aircraft ~2080 !?
RANS
•Research tool only
hybrid
RANS-LES

LES
DNS
Log(CPU)
days weeks months years
 Large Eddy Simulation – LES
•Valid

standard •little approximation

RANS •Universal (if correctly done)
Approximation

•Also very expensive

•Large scale dynamics:
level

•Full aircraft ~2050

•Acoustics
advanced
•Affordable in internal
RANS •Dynamic loads
flow (industrial use)
hybrid
RANS-LES

LES
DNS
Log(CPU)
days weeks months years
 Hybrid RANS – LES methods
•RANS in attached BLs
standard •LES in free turbulence,
RANS separation
Approximation

•Affordable, but still

expensive
level

Many different methods:

advanced
•DES, DDES, IDDES, VLES,
RANS
XLES, PANS, PITM
hybrid
RANS-LES

LES
DNS
Log(CPU)
days weeks months years
 Prediction of turbulent flows

standard
RANS
Approximation
level

advanced
RANS

hybrid
RANS-LES

LES
DNS
Log(CPU)
days weeks months years
Basic concepts

13
Scale separation

14
Different Reynolds numbers

15
Viscosity
• Kinematic viscosity, n
• Dynamic viscosity, µ
• Density, r

16
Boundary layers (BL)
• Thin layers
– Thickness Reynolds number dependent
• Laminar boundary layers
– Thickness related to wall skin friction
• Turbulent boundary layers
– Inner and outer scales separated
– Scale separation Reynolds number dependent

17
BL on the ONERA A-profile

18
Wall treatments for BLs
𝑦
• No slip BC (no approximation)
– BL resolved: 𝑦 ! ∼ 1
– Δ𝑦" = 1 − 100𝜇m
– 40-80 grid points

• Log-law boundary condition

– First grid point in log layer
𝑦 ! >20 AND 𝑦<0.1𝛿 0
– 10-20 grid points
– Warning for too small grid size
0 𝑈
READ SOLVER DOCUMENTATION !!!

• Slip wall boundary condition

– BL neglected

19
Approximation of BLs
• Slip wall boundary condition
– Boundary layer completely neglected
– Euler (non-viscous) computations possible
– Slip BC can also be applied to viscous & turbulent CFD
• No slip boundary condition
– Boundary layer completely resolved (y+=1)
– Extreme resolution needed (Dy=1-100µm)
– 40-80 grid points within the boundary layer
• Log-law boundary condition (turbulence)
– First grid point within log layer (y+>20 AND y<0.1d)
– 10-20 grid points within the boundary layer
– Warning: standard log-law BCs inconsistent with too small
grid size. READ SOLVER DOCUMENTATION !!!

20
What’s in Fluent?
• Standard and Non-Equilibrium Wall
Functions:
– “Wall adjacent cells should have
y+values between 30 and 300–500” –
(remember y < 0.1d ! )
– “The mesh expansion ratio should be
small (no larger than around 1.2)”
– “Non-equilibrium wall function method
attempts to improve the results for
flows with higher pressure gradients,
separations, reattachment and
stagnation”
• Scalable Wall Functions:
– Consistent for all y+values

21
What’s in Fluent? …
• Enhanced Wall Treatment Option
– Combines a blended law-of-the wall
and a two-layer zonal model.
– Suitable for low-Re flows or flows with
complex near-wall phenomena.
– Generally requires a fine near-wall
mesh capable of resolving the viscous
sublayer
– y+< 5, and a minimum of 10–15 cells
across the “inner layer” for best results
– Valid for all y+
– Available for all k-e and k-w models
– Not yet for Spalart-Allmaras (y+<3 OR
y+>15)

22
Recommendations for Fluent
• For K-e models
– use Enhanced Wall Treatment: EWT-e
• If wall functions are favored with K-e models
– use scalable wall functions
• For K-w models
– use the default: EWT-w

23
Wall bounded turbulence

24
Empirical relations for BLs
• Friction coefficient

– Turbulent

– Laminar

• Boundary layer thickness

– Turbulent

– Laminar

25
 Empirical relations plotted

Laminar Turbulent

Laminar Turbulent

26
Empirical relations plotted

Turbulent

Laminar

27
Near-wall mesh size
• Compute 𝑅𝑒! ≡ 𝑈"𝑥/𝜈
• Derive 𝐶# and 𝛿/𝑥 from empirical relations
– Gives B.L. thickness 𝛿
" &
– and wall viscous scale 𝑙∗ = , 𝑢$ = 𝑈% 𝐶
#! ' (

• First cell thickness Δ𝑦*

– Wall functions: Δ𝑦& ≲ 0.1𝛿 and 𝑦 ) ≳ 20 → Δ𝑦& ≳ 20𝑙∗
Read documentation carefully and avoid using
standard wall functions. ”Extended”, ”scalable”,
”generalized” wall treatment can handle 𝑦 ) < 20.
– Wall resolved: 𝑦 ) ≈ 1 (or 𝑦 ) < 5)
→ Δ𝑦& ≈ 𝑙∗ (or Δ𝑦& < 5𝑙∗ )
Also here read documentation.

28
Turbulence modelling
• Reynolds decomposition
𝑢! 𝒙, 𝑡 = 𝑈! 𝒙 + u#" 𝒙, 𝑡
where 𝑈! 𝒙 = 𝑢! 𝒙, 𝑡 and u#" 𝒙, 𝑡 = 0
• The ”mean” is time average, ensemble average or
averaging in homogeneous directions. 𝑈! 𝒙 may
actually vary slowly in time.
• The mean of the Navier-Stokes eq.s -> RANS eq.s
𝜕𝑈!
=0
𝜕𝑥!
𝜕𝑈! 𝜕𝑈! 1 𝜕𝑃 𝜕 𝜕𝑈!
+ 𝑈$ =− + 𝜈 − 𝑢!# 𝑢$#
𝜕𝑡 𝜕𝑥$ 𝜌 𝜕𝑥! 𝜕𝑥$ 𝜕𝑥$

29
Reynolds stresses
• Not “small”
• Significant effects on the flow
• Needs to be modelled in terms of mean flow quantities
• Reduces the problem to steady (or slowly varying)
• 2D assumptions possible

• Equation can be derived from Navier-Stokes equations

• Need modelling

30
Eddy-viscosity models (EVM)
• Assume: Reynolds stresses related to an ”eddy viscosity”, 𝜈*
1
𝑢+, 𝑢-, − 𝑢., 𝑢., 𝛿+- = −2𝜈* 𝑆+-
3
& /0" /0#
where 𝑆+- ≡ +
' /1# /1"
/0
• Based on Boussinesq (1877) 𝑢, 𝑣′ = −𝜈*
/2
• Eddy viscosity ~ turbulence velocity and length scales
𝜈* ∼ 𝑉𝐿

31
One-equation models
• One transport equation for K (turbulent kinetic energy) or nT.
• Additional information from global conditions (typically wall
distance)
• Works well for attached boundary layers
• Not very general, but more than algebraic models

• Example: Spalart-Allmaras (1992)

– reasonable and robust model for external aerodynamics
– Boeing’s ”standard model”

32
 Two-equation models
• Two transport equations for the turbulence scales (K–e or K–w)
• Completely determined in terms of local quantities (except near-
wall corrections which may be dependent on wall distance)
• Works well for attached boundary layers
• Somewhat more general than zero-, one-equation models
• Model transport equations loosely connected to the exact equations.

• Examples:
– Standard K–e model (Launder & Spalding 1974)
– Wilcox K–w (1988, ...) models
– Menter (1994) SST K–w model (performing reasonable well also in
separated flows)
Airbus’ ”standard model”

33
Eddy-viscosity models …

34
LES and LES/RANS hybrids
• Simulation of only the large scale turbulence (compare with
DNS, simulation of all scales)
– Always time dependent and 3D -> expensive
• Wall free turbulence simulations almost Re independent
• Wall bounded turbulence largely Re dependent
– fully resolved near-wall region very expensive (almost as DNS)
– wall-function or near-wall RANS coupling saves computational
cost
– hybrid RANS-LES (RANS in attached BLs and LES in wall-free
separated regions) a very active research field, eg DES

35
LES and LES/RANS hybrids …
• LES in academic research for:
– low Re generic flows
– complement to DNS for higher Re
– gives detailed knowledge about turbulence
• LES in industrial use in:
– internal flow with complex geometries
– flows around blunt bodies (with large separated regions)
– atmospheric boundary layers (e.g. weather forecasts)
– combustion simulation
– other complex flow physics at moderate Re
• Warning: LES is extremely expensive in high attached and slightly
separated wall-bounded flows, if properly resolved.

36
How expensive is DNS?
• DNS of flat plate turbulent boundary layer
– Schlatter, et al., KTH, Dept. of Mechanics
– http://www.youtube.com/watch?v=4KeaAhVoPIw

– 7.5×109 nodes, Rex = 1.4×106

– CPU time: 3 months @ 4000 CPU cores = 1 unit

• DNS of model airplane, same Reynolds number (Rex = 1.4×106)

– Only a narrow stripe – wing requires about 1 000 stripes
– Nnodes = 1013
– CPU = 103 units

• See also DNS of aerofoil below (KTH)

– 3.2×109 nodes, Rex = 0.4×106
– https://www.youtube.com/watch?v=hz7UjN_vYuw
DNS – full scale airplane
• Re scaling – wall bounded flow

– Nodes:

– Time steps:

– CPU time:

• DNS of Airplane (Rex = 70×106) (factor of 50)

– Nnodes = 1017
– CPU = 109 units
Supercomputer development

109 units
DNS of full
airplane
year 2055?

present
45 year?

doubling of CPU
every 18 months

wikipedia