Mass Transfer in Turbulent Flow
ChEn 6603
References:• S. B. Pope. Turbulent Flows. Cambridge University Press, New York, 2000.
• D. C. Wilcox. Turbulence Modeling for CFD. DCW Industries, La Caada CA, 2000.
• H. Tennekes and J. L. Lumley. A First Course in Turbulence. MIT Press, Cambridge, MA, 1972.
• R. O. Fox. Computational Models for Turbulent Reacting Flows. Cambridge University Press, 2003.
1Friday, April 15, 2011
William H. Cabot and Andrew W. CookNature Physics 2, 562 - 568 (2006)
Rayleigh-Taylor instability (DNS calculation) Photograph of Jupiter from Voyager
~15,000 Miles(2 Earth diameters)
Mixing in reacting flow (DNS)
~ 6 cm
Methane pool fire
2Friday, April 15, 2011
Energy balance perspective.• Consider steady, isothermal, fully developed turbulent flow in a horizontal
pipe‣ Increasing pressure drop does not increase flow rate proportionally. Why? Where is
the energy going? How?
‣Work done by pressure forces balanced by work done by viscous forces
• Energy provided at “large” scales, dissipated at “small” scales.‣ Length scales reduce to meet demand of energy balance.
‣ Smaller length scales ⇒ steeper gradients ⇒ more dissipation.
Origins of Turbulence
pressure work viscous dissipation
∂ρk
∂t+∇ · (ρkv) = −v ·∇ · τ − v ·∇p + ρ
n�
i=1
ωiv · fi
0 = v ·∇p + v ·∇ · τ
∆pKinetic energy
equation:What assumptions?
3Friday, April 15, 2011
Velocity Length ScalesL - largest length scale (m)η - smallest length scale (m)U - velocity at L-scale (m/s)
ν - kinematic viscosity (m2/s) = µ/ρε - kinetic energy dissipation rate (m2/s2· s-1)
Most kinetic energy is contained in “large” length scales (L). It is dissipated primarily at “smallest” (Kolmogorov) length
scales (η) by molecular viscosity (ν).
Can we form a length scale from ε and ν?
η ∝�
ν3
�
�1/4
� ∝ UU
L/U
kinetic energy
“integral” or “large” time scale
Note: ε doesn’t depend on ν. ν just determines the smallest
length scale in the flow.
η ∝�
ν3
�
�1/4
∝ L1/4� ν
U
�3/4
L
η∝
�LU
ν
�3/4
= Re3/4 Key result! Tells us how length-scales separate!
4Friday, April 15, 2011
Scalar Length ScalesℓB - smallest scalar length scale (Batchelor scale)
L
�B=
L
η
η
�B∝ Re3/4Sc3/4L
�B=
L
η
η
�B∝ Re3/4Sc1/2
Sc ≡ ν
DSc > 1“mixing paint” - ℓB <η - scalar only feels straining
from smallest velocity scales.(mass diffuses slower than momentum)
Sc < 1ℓB>η - at ℓB, there are still velocity fluctuations,
but the scalar field is uniform (mass diffuses faster than momentum)
�2B ∝ Dt
�B ∝ D1/2�ν
�
�1/4
�B
η∝ D1/2
�ν
�
�1/4�
ν3
�
�−1/4
∝�
D
ν
�1/2
= Sc−1/2
Form a time scale from the “Kolmogorov” time scale
(i.e. from ν and ε).
Relevant parameters are D, ε. (ν only dominant near η).�B ∝
�D3
�
�1/4
�B
η∝
�D
�
�1/4 �ν3
�
�−1/4
∝�
D
ν
�3/4
= Sc−3/4
Gases: Sc ≈ 1, Liquids: Sc ~ 103.5Friday, April 15, 2011
Sc→∞
http://sensitivelight.com/smoke2
6Friday, April 15, 2011
Solution OptionsDirect Numerical Simulation (DNS)• Resolve all time/length scales by solving the governing equations directly.• Restricted to small problems.• Cost scales as Re3 for turbulence alone! (Species with Sc>1, and/or
complex chemistry could further increase cost)‣(L/η∼Re3/4, 3D, time)
Large Eddy Simulation (LES)• Resolve “large” spatial & temporal scales• Model “small” (unresolved) time/space scales
Reynolds-Averaged Navier Stokes (RANS)• Time-averaged.• Describes only mean features of the flow.
Model all effects of the flow field• Useful only for some classes of problems (usually interfaces like walls)• Commonly done in heat transfer & mass transfer (also for some problems
involving aerodynamics )
Incr
ease
d M
odel
ing
7Friday, April 15, 2011
Time-Averaging (RANS)Constant density, viscosity:
� ∞
0∇ · vdt = 0
∇ ·�� ∞
0vdt
�= 0
∇ · v = 0
Continuity:
Momentum:
Definition of time-average:
index (Einstein) notation: − ∂
∂xj
� ∞
0
vivjdt = − ∂
∂xjvivj .
∇ · v = 0∂v∂t
= −∇ · (v ⊗ v)− 1ρ∇p + ν∇2v
� ∞
0
∂v
∂tdt = −∇ ·
� ∞
0vv dt− 1
ρ∇p+ ν∇2v
0 = −∇ ·� ∞
0vv dt− 1
ρ∇p+ ν∇2v
∂
∂xjvivj +
1
ρ
∂p
∂xi+ ν
∂
∂xj
∂vi∂xj
φ ≡ limT→∞
1
T
� t0+T
t0
φ(t) dt
8Friday, April 15, 2011
The Closure Problem
vivj = (vi + v�i)(vj + v�
j)
= vivj + viv�j + v�
ivj + v�iv
�j
vivj = vivj + v�iv
�j
For large Re, µt≫µ (molecular viscosity is negligible).
“Fluctuating” component
φ� = 0 ¯φ = φφϕ� = 0
φ� ≡ φ− φ
Model this term using a “gradient diffusion” model.
∂
∂xj
�v�iv
�j
�≈ −µt
ρ
∂
∂xi
∂
∂xjvj
∂
∂xj(vivj) +
1
ρ
∂p
∂xi+ ν
∂
∂xj
∂vi∂xj
+∂
∂xj
�v�iv
�j
�= 0
∂
∂xj(vivj) +
1
ρ
∂p
∂xi+ 1
ρ (µ+ µt)∂
∂xj
∂vj∂xj
= 0
9Friday, April 15, 2011
Time-Averaged Species Equations
@ Large Re, ji,turb ≫ji
(molecular diffusion is negligible)
A very difficult problem...∇ · ωiv +1ρ∇ · ji = si/ρ
∇ · (ωiv) +1ρ∇ · ji +∇ ·
�ω�
iv��
= si/ρ
∇ · (ωiv) +1ρ∇ ·
�ji + ji,turb
�= si/ρ
• D°turb - turbulent diffusivity (for mass flux relative to mass avg. velocity)
• µturb - eddy viscosity• Typically, Scturb is specified.
MODEL for turbulent species diffusive flux:
∇ · (ωiv) +1ρ∇ · ji,turb = si/ρ
Multicomponent effects are irrelevant at sufficiently high Re.
constant properties & density...
ji,turb = −ρD◦turb∇ωi
Scturb =νturbD◦
turb
=µturb
ρD◦turb
10Friday, April 15, 2011
Spatial Averaging (LES)
G(x) - filter kernel functionremoves “high wavenumber” components of ϕ.
• Filter governing equations. (similar procedure as for RANS, but a little more complicated).
• Write models for unclosed terms.• Solve filtered equations (for filtered variables).• Provides time-varying solutions at a “coarse”
level.
Courtesy R.J. McDermott
Courtesy R. J. McDermott
φ ≡� ∞
−∞φ(x)G(x)dx
¯φ �= φ
φ� �= = 0
φϕ� �= 0
11Friday, April 15, 2011
Variable Density∂ρωi
∂t= −∇ · ρωiv −∇ · ji + si.
Favre-averaging (RANS)Favre-filtering (LES) φ ≡ ρφ
ρ−→ ρφ = ρφ
Leads to many additional complications, most of which are typically ignored...
example: ji =n�
k=1
ρDoik∇ωk
?≈ ρn�
k=1
Doik∇ωk
LES: ifΔ = “filter width”
∆� �B , ∆� η then ji,turb � ji
∂ρωi
∂t= −∇ · ρ�ωiv −∇ · ji + si
= −∇ · ρωiv −∇ ·�ji + ji,turb
�+ si
12Friday, April 15, 2011