You are on page 1of 65

THE USE OF FLUX-AVERAGED

DIFFUSIVITIES TO MODEL
MULTICOMPONENT DIFFUSION

Claudio Olivera-Fuentes
TADiP Group
Thermodynamics and Transport Phenomena Dept.
Simón Bolívar University
Caracas, Venezuela

1
Contents

 Constitutive Equations
 Exact and Approximate Solutions
 Definition of Flux-Averaged Diffusivities (FADs)
 Examples
 Comments and Conclusions

2
Constitutive Equations

3
Maxwell-Stefan equations [1]

 Molecular diffusion at constant T, P


n


  
TP  i  Fik ck (vi  vk )
k 1

RT
Dik  , i, k  1,, n
cFik

 Fik friction coefficients (resistances) = Fki ?


 Dik MS diffusivities (conductances) = Dki ?

4
Maxwell-Stefan equations [2]
n   n  
xi J k  x k J i xi N k  x k N i
 
xi 
TP  i   , i  1,, n
RT cDik cDik
k 1 k 1

n n
 
 
    
N i  J i  xi N mix , N mix  Nk  Ji  0
k 1 k 1

 Ji , diffusive or relative fluxes


 Ni , net or absolute fluxes

5
Diffusive interactions

 Principle of equipresence
 Duncan & Toor (1962)
 Diffusion barrier: Potential gradient exists, yet
component does not diffuse
 Reverse diffusion: Component diffuses “uphill” from
smaller to larger potential
 Osmotic diffusion: Component diffuses although no
potential gradient exists

6
Binary diffusion

 MS for 2-component systems


   
xA  J N  xA (N A  N B )
TP  A   A   A
RT cD AB cD AB

 Fick’s “law”
   
 J N  xA (N A  N B )
x A   A   A
cD AB cD AB

 DAB binary or “ordinary” diffusivity  DBA

7
Exact v. Approximate
Solutions

8
Ideal gas mixtures

 Thermodynamic & diffusive ideality


 RT 
 TP  i  xi , Dik  D ik
xi

 Simplified MS equations
n   n  
yi J k  y k J i yi N k  y k N i
 

cyi   , i  1,, n
k 1
D ik k 1
D ik

9
Continuity equations

 Steady-state diffusion
 One-dimensional transport
 No homogeneous reaction
N iz  const . rN ir  const. r 2 N ir  const

10
Unified treatment
Diffusion region Coordinate h Net flux ni Diffusive flux ci

Rectangular z  Z1 N iz J iz
( Z 2  Z1 ) ( Z 2  Z1 )
Z1  z  Z2 Z 2  Z1 c c

Cylindrical ln( r / R1 ) rN ir R2 rJ ir R2
ln ln
R1  r  R2 ln( R2 / R1 ) c R1 c R1

Spherical 1 / R1  1 / r r 2 N ir  1 1  r 2 J ir  1 1 
     
R1  r  R2 1 / R1  1 / R2 c  R1 R2  c  R1 R2 

11
Transformed equations
 Continuity
n i  const , i  1,, n

 Constitutive
n
yin k  y kn i

dyi
 , i  1,, n , 0  h  1
dh D ik
k 1

 Determinacy: boundary values, flux ratios, etc.


L  2n  1

12
Matrix solution, exact
d y
 A y 
dh
n
nk ni
Aii  D
k 1 ik
, Aik  
D ik
, i, k  1,, n (i  k )

k i

 y    y0  @ h 0 ,  y    y1  @ h 1

 Solve as vector/matrix ODE (Amundson, 1966)


 y1   expA y0 

13
Matrix solution, linearized

 Stewart & Prober (1964), Toor (1964)


n
yi,av c k  y k ,av c i

dyi
 , i  1,, n
dh D ik
k 1

d ( y)
n mix ( y )  (n )  [ D]av
dh

 Uncouple by diagonalization, solve as Fick’s law


in pseudo-composition space

14
Equivalent diffusivity methods

 Assume pseudo-binary behavior, solve as Fick’s


law in true composition space
dyi n  yin mix
 i , i  1,, n
dh D im

 Dim “equivalent” or “effective” diffusivity


 must depend on fluxes and/or gradients, otherwise
n  n 1
 1 1  
D  
Ji
J i     0  D1m    D nm
i 1 im i 1  Dim D nm 

15
Typical assessment

 Dim’s have the advantage of simplicity, but…


 They are not system properties
 They fail to account for diffusive interaction effects
 They depend on transport fluxes which are not
always known in advance
 They vary with position or with composition in a
manner that may require iteration
 They may give large errors when used in situations
they were not designed to handle
 They are no longer justified, in view of the
availability of more rigorous analytical and
computational tools

16
Definition of Flux-
Averaged Diffusivities

17
A rigorous definition

 Average diffusive interactions for each component


1   yn
n n n
yin k  y kn i n i
 n  y k   i mix
dyi
  yi n i
dh D im 
D im  k  D im
k 1 k 1 k 1 

 Define flux-average diffusivity (FAD) implicitly as


n
yin mix  n i yin k  ykn i
Dim
 
k 1
Dik
, i  1,  , n

18
General FAD solution

 Define mean film equivalent diffusivity


1

D
1 1
D im  dh
im
0

-1 1 yi1   i ni
n mix  D im ln , i  n mix  0
yi 0   i n mix

1
n i  D im
-1
( yi 0  yi1 ) n mix  0 ( EMCD)

19
Practical application

 If correct average FAD can be identified a priori,


method is exact
 If approximate FAD must be used, this is the only
source of error in the method
 Arithmetic mean (AM) value used in the literature
is most likely in error, because exact solution
shows exponential dependence of (y) on h.

20
Example 1

Evaporation of a pure liquid into


a gas mixture in a Stefan tube

21
Slattery (1999)

 Pure liquid A evaporates into a gas mixture of A, B and C


in a Stefan tube, arranged in such a manner that the
liquid-gas interface remains fixed in space (h = 0) as the
evaporation takes place. Species B and C are insoluble in
liquid A, nB = nC = 0. At the prevailing conditions, the
gas-phase composition of A in equilibrium with the liquid
is yA0. The gas mixture blowing past the top of the tube (h
= 1) has constant composition yB1, yC1. It is desired to
find the rate of evaporation of A.

22
Problem specifications

Pure liquid Gas mixture


A
nB = nC = 0

yA0 yA1, yB1

interface top of tube


h=0 h=1

23
Exact solution: B & C (stagnant )

 From MS equations
dy B y Bn A  n 
  y B  y B 0 exp  A h 
dh D AB  D AB 
dyC yCn A  n 
  yC  yC 0 exp  A h 
dh D AC  D AC 

 Solve for nA, yB0, yC0


 n   n 
y B 0  yC 0  y B1 exp   A   yC1 exp   A   1  y A0
 D AB   D AC 

24
FAD solution: B & C (stagnant)

 From general FAD solution


 1
y k1  y k 0 exp D km nA  , k  B, C

 From general FAD definition


1 1
D km  D Ak  D km  D Ak , k  B, C

 Hence FAD solution is exact

25
Exact solution: C (mobile)
 From MS equations
dy A  y y   yB y 
 n A  B  C   y A1  y A0  n A   C 

dh  D AB D AC   D AB D AC 

 Mean film compositions


1
yi 
 y dh
0
i

 From solution for B & C


h
 y k1 
y k (h )  y k 0    y k  y k , LM , k  B, C
 yk 0 
26
The correct average
 y B, LM yC , LM 
 Exact y A1  y A0  n A   
D
 AB D AC 

1  y A1 1
 FAD ln  D Am nA
1  y A0

 i.e. mean FAD at log-mean (LM) compositions


1 y B, LM 1 yC , LM 1
D Am  
(1  y A ) LM D AB (1  y A ) LM D AC

27
What not to do?

 Do not use AM compositions to compute FAD


 and if you do, don’t blame FAD method for errors!

 Do not solve equation for C alone


 Use of FAD approach does not overrule system
determinacy
 Use A & B, A & C or B & C. All combinations give
exact solution

28
AM v LM averages [1]

 Parametric analysis
1  y A0 D AB nA y B1
Q , R , X  , Y
1  y A1 D AC D AB 1  y A1

 Exact (LM): Y exp(  X )  (1  Y ) exp(  RX )  Q

 Y (1  R) 
 Approximate (AM): X R  [1  exp(  X )]  ln( 1 / Q)
 1 Q 

29
AM v LM averages [2]
100
LM (exact)
AM, Q = 0.1
yA2 << yA1 AM, Q = 0.3
DAB << AM, Q = 0.5
DAC AM, Q = 0.7
AM, Q = 0.9

1
X

0,01
0,1 1 10
R

30
Generalization

 n – 1 stagnant components
1 1  n1 
D km  D1k  yk1  yk 0 exp   , k  2, , n
 D1k 

 1 mobile component
n
1  ( y1 )1
 (1  y )
yk , LM 1 1
D1m1   n 1  D1m1 ln
1 LM D1k 1  ( y1 ) 0
k 2

31
Example 2

Diffusion of one component in a


stagnant mixture of known
average composition

32
Treybal (1980)

 Ammonia (A) is diffusing through a stagnant gas


mixture consisting of 1/3 nitrogen (B) and 2/3
hydrogen (C) by volume at 206.8 kPa and 54 °C.
Calculate the rate of diffusion of A through a film
of gas 0.5 mm thick when the concentration
change across the film is 10 to 5 % A by volume

33
Problem specifications

Gas mixture nB = nC = 0 Gas mixture

Average A-free composition y´av

yA0 yA1

h=0 h=1

34
What average? [1]

 This work
1


yk
y k  dh , k  B, C
1  y1
0

1 yB yC
D Am  
D AB D AC

ln( 1 / Q)
X
R  (1  R) yB

35
What average? [2]

 Hsu and Bird (1960)


1

 y dh
k
yk
y k0  1
0
 n
, k  2,, n

 (1  y )dh  y
0
1
k 2
k

1 1  Q exp( X )  1  exp( RX )  1 
1   
y B0 R  exp( X )  1  Q exp( RX )  1 

1
 Does not give D Am

36
Not all averages are equal!
1,0

0,8
mean filmA free composition-

0,6

Eq (55)
0,4
Eq (51), Q = 0.1
Eq (51), Q = 0.3
0,2 Eq (51), Q = 0.5
Eq (51), Q = 0.7
Eq (51), Q = 0.9
0,0
0 2 4 6 8 10
R

37
Example 3

Condensation of mixed vapors in


presence of a noncondensable gas

38
Sherwood et al. (1975)

 A condenser operates with a mixture of vapors of


ammonia (A), water (B) and hydrogen (C) at 3.36 atm
and 200 °F. Mole fractions are yA0 = 0.30, yB0 = 0.40 in
the gas bulk (h = 0), and yA1 = 0.455, yB1 = 0.195 at the
interface (h = 1), where equilibrium with the local liquid
is assumed. Binary diffusivities are DAB = 0.294, DAC =
1.14, DBC = 1.30 cm2.s-1. It is desired to estimate the ratio
of condensation fluxes of water and ammonia.

39
Problem specifications

Gas mixture Condensate

nC = 0

yA0, yB0 yA1, yB1

bulk gas interface


h=0 h=1

40
Exact solution

 Gilliland (in Sherwood, 1937)


nA nB yC1
  ln
D AC D BC yC 0

 y A1  y 
  1  P B1  1 1 1
 A   B  D AC  D AB
n A  n B  D AB ln , P  1 1
 y A0  y  D BC  D AB
  1  P B 0  1
 A   B 

 cf. Toor (1957), Sherwood et al. (1975), Taylor (1981)

41
Exact, yet false [1]

 The special value …


 y A1  y 
  1  P B1  1 1 1
  B D AC  D AB
n A  n B  D AB ln  A   , P  1 1
 y A0  y  D BC  D AB
  1  P B 0  1
 A   B 
 … reduces both equations to
yC1
n mix  D AB ln
yC 0

 Discarded on physical grounds

42
Exact, yet false [2]

 System is always satisfied by …


nA nB yC1
  ln
D AC D BC yC 0

n A n B  0

 … but then this problem becomes overspecified


(nmix = 0 in addition to original data!)

43
FAD solution

 From general FAD solution


-1 1 y A1   A 1 y  B 1 yC1
n mix  D Am ln  D Bm
-1
ln B1  D Cm
-1
ln
y A0   A y B0   B yC 0

 For any stagnant component (C in this case) …


n
k
D -1 1 yi1
-1
D im  , n mix  D im ln , yi  yi, LM
ik yi 0
k 1

 … but no obvious suggestion for the others

44
FAD-1

 Literature: Use AM compositions …

yi, AM  k  yk , AM  i
n
Dim- 1 yi, AM  i    , i  1,  , n
k 1
Dik

 … but AM is inconsistent with exact solution

45
FAD-2

 Use LM compositions for all components …

 n  n
yi , LM  k  yk , LM  i
Dim- 1


 yi , LM   i
k 1
yk , LM  


k 1
Dik
, i  1,  , n

 … but LM is exact only for stagnant components

46
FAD-3
Not yi, but xi  yi – i!

 Recall general FAD solution …


-1 1 yi1   i
n mix  D im ln
yi 0   i

 … use flux-averaged compositions


 n  n
x i , LM  k  x k , LM  i
Dim- 1 x i , LM   i

k 1
x k , LM  


k 1
Dik
, i  1,  , n

47
Exact v FAD solutions
1 1 1
Solution nmix n B/ n A -1
D Am -1
D Bm -1
D Cm
Gilliland 0.1955 4.603 0.2378 0.4933 1.268
FAD-1 (AM)
A&B 0.1800 5.667 0.2537 0.4785 -
A&C 0.1956 4.742 0.2437 - 1.269
B&C 0.1953 4.412 - 0.4868 1.267
FAD-2 (LM)
A&B 0.1820 5.504 0.2518 0.4815 -
A&C 0.1956 4.728 0.2431 - 1.269
B&C 0.1954 4.435 - 0.4876 1.267
FAD-3 (x-LM)
A&B 0.1889 5.010 0.2453 0.4880 -
A&C 0.1956 4.670 0.2406 - 1.268
B&C 0.1954 4.552 0.4916 - 1.268

48
Exact v FAD-3
1,0
NA, exact
NA, FAD-3 A&B
NA, FAD-3 A&C
NA, FAD-3 B&C
0,5 NB, exact
NB, FAD-3 A&B
NB, FAD-3 A&C
B

NB, FAD-3 B&C


A,

0,0

-0,5
0 0,1 0,2 0,3 0,4 0,5 0,6
y A0

49
Example 4

Diffusion with rapid


heterogeneous catalytic reaction

50
Treybal (1980)

 Ammonia (A) is being cracked on a solid catalyst


according to the reaction 2NH3  N2 + 3H2 at 1 atm and
200 °C. The bulk gas (h = 0) contains 33,33% NH3,
16,67% N2 and 50,00% H2 by volume. NH3 diffuses from
the bulk gas to the catalyst surface (h = 1), and the
reaction products N2 (B) and H2 (C) diffuse back, through
a gas film 1 mm thick. Estimate the rate of cracking of
NH3 if the reaction is so rapid that the concentration of
NH3 at the catalyst surface drops to zero

51
Problem specifications

Gas mixture Catalyst

B , C from stoichiometry

yA0, yB0 yA1

bulk gas surface


h=0 h=1

52
Exact solution

 Matrix only, no explicit algebraic form


 y1   expBn mix  y0 

n
k
Bii  Dk 1 ik
, i  1,  , n

k i

i
Bik   , i, k  1,  , n (i  k )
Dik

53
FAD solution

-1 1 y A1   A 1 y  B 1 yC1   C
n mix  D Am ln  D Bm
-1
ln B1  DCm
-1
ln
y A0   A yB0   B yC 0   C

 Use FAD-1, 2, 3
 Use A & B, A & C, B & C

54
Exact v FAD
-0,29

exact

-0,30
mix

-0,31 FAD-1 (AM)

FAD-2 (LM)

FAD-3 (xi-LM)

-0,32
0,33 0,34 0,35
y B1

55
Comments and
Conclusions

56
Equivalent diffusivities …

 … depend on transport fluxes which are not


always known in advance
 True. So do the matrices [A] and [B] in the exact
solution. They have to depend on the fluxes,
otherwise they would not represent the diffusive
interactions correctly.

57
Equivalent diffusivities …

 … vary with position or with composition in a


manner that may require iteration
 True. Iteration is an inescapable part of the solution
of multicomponent problems. Again, the
computational difficulties and convergence
problems of the matrix solution have been the
subject of a great many publications. The same
hardware used to compute a matrix solution will
handle a FAD method just as easily and speedily.

58
Equivalent diffusivities …

 … may give large errors when used in situations


they were not designed to handle
 Read the manual. There exists at least one FAD
definition, used in this work, that is 100% rigorous,
designed to handle all situations, and gives exact
results if the correct mean film values can be
identified. It is not fair on the method to knowingly
use clearly incorrect averages and then complain of
poor results…

59
Equivalent diffusivities …

 … are not system properties


 True. If anything, the concept resembles that of
mass transfer coefficients, entities defined to
provide a simple mathematical and physical picture
of a more complex reality. Should this detract from
their usefulness?

60
Equivalent diffusivities …

 … fail to account for diffusive interaction effects


 False. In the course of solving typical mass transfer
problems, cases are frequently found where a FAD
becomes zero (diffusion barrier), infinite (osmotic
diffusion) or negative (reverse diffusion). Such
cases illustrate in vivid fashion that diffusion in
multicomponent systems can be much more
complex than in binary mixtures, and that “strange”
behavior may be the norm rather than the
exception.

61
Equivalent diffusivities …

 … are no longer justified, in view of the


availability of more rigorous analytical and
computational tools
 Your choice. In the author’s experience, FAD
methods constitute an excellent form of introducing
students to multicomponent mass transfer, at least
at the undergrad level where mathematical
elements required for a more advanced matrix
approach may not be available. Why else teach
binary diffusion? There are no binary systems in
industry!

62
In conclusion

 The FAD method can be a useful practical tool


that, despite its mathematical simplicity, provides
a true representation of multicomponent diffusive
interactions, usually with only a small reduction
in the accuracy of the computed fluxes. FAD-3 in
particular appears promising, but must undergo
extensive testing before being recommended as a
general technique.

63
Thanks…

 To the Organizing Committee, for their kind


invitation to give this conference
 To my students in the TF3261 (Transport
Phenomena) course at USB, for their (not
completely forced) work as beta-testers

64
Questions?

(and hopefully answers …)

65

You might also like