Introduction to Monolith

Reactor Modelling:
Mass, Energy and Momentum Transport

R.E. Hayes
Department of Chemical and Materials
Engineering, University of Alberta,
Edmonton, Alberta, Canada

MODEGAT 14-15 September 2009

1
Objectives of this presentation

• To give an introductory overview to the

area of monolith reactor modelling

• To high light some of the factors that

must be taken into consideration when
choosing a model.

2
 Why use computer modelling?
• Experiments are expensive and take time.
• Analysis of reactors can be difficult, especially for
fast highly exothermic reactions. Non-intrusive
measurement methods may not be available.
• Modelling can eliminate scenarios that do not
offer promise.
• Faster design and optimisation.
Models are not intended to eliminate experiments completely.
3
Typical catalytic converter
•The catalytic converter uses a metal or ceramic monolith
support structure.
•Typical automotive converters use 400 CPSI cell density

Hayes and Kolaczkowski, Chem.

Eng. Sci., 49, 3587, 1994.

Heck and Farrauto, Catalytic Air

Pollution Control, 2002
4
Observations
• The problem involves the transport of mass, energy
and momentum. It is a multi-physics problem.

• Transport occurs at different scales in space.

• The transient response of the gas phase and solid
phase occurs at different time scales.

• Modelling of monolith reactors is thus an exercise in

multi-scale modelling.

The monolith reactor presents a challenging modelling problem.

5
 Heterogeneous catalytic reactions
Heterogeneous gas phase catalytic reactions typically comprise the same steps.

Fluid flow over the solid results in a

boundary layer through which heat and
mass diffuse to the gas/solid interface.

Reactants diffuse into the porous

catalyst.

External Bulk gas Reactants adsorb on the active sites.

catalyst
surface The reaction occurs on the active sites,
located on the walls of the tortuous pores.

catalyst Products desorb and diffuse back to the

pore surface, and then through the boundary
layer to the bulk fluid.
6
Model Scales
Micro scale: At the atomic/molecular level, the basic chemistry
involves a set of adsorption/desorption and surface reaction
steps. The genuine mechanism may be very complicated.

Meso scale: Single channel of a monolith. Includes gas

Gas Phase
phase flow, transport to the surface of the washcoat by
Catalyst diffusion, diffusion and reaction in the washcoat. The
basic process in a heterogeneous catalytic reaction are
present at the meso scale.

Macro scale: Entire catalytic converter, which

includes transport and interaction among channels.

7
Model complexity: Points to note
• How much complexity is wanted and or needed?
– At what level do you wish to model?
– How much information do you want to include?
– Do you need to include all of the scales?
• The model should be as simple as possible, but not
simpler (Albert Einstein).

• Models do not have to be perfect to be useful.

Gas Phase

Catalyst

8
Mathematical formulation
• The fundamental model is represented by a set of
equations, which include key assumptions
– Algebraic equations, linear or non-linear
– Ordinary differential equations (IVP or BVP)
– Partial differential equations

• Practical monolith modelling problems do not have

analytical solutions; models are solved numerically.

• The model selected and the way it is represented will

determine the numerical method used for its solution.

9
Mathematical formulation
• Reactors are fundamentally distributed parameter
systems ; Concentration, temperature and velocity
vary with spatial coordinate.

• The dimensionality (and hence complexity) can be

reduced by using average parameter values over
specified model dimensions. The result is referred
to as a lumped parameter model.

• Lumped parameter models are approximations.

10
Basic models in common use

• Single channel of monolith structure

– 1D, 2D or 3D
– Diffusion in the washcoat may be considered

• Whole reactor with variable channel conditions

and channel interactions
– 2D or 3D , several model options exist
– The upstream and downstream duct work might
be included

11
Single channel model (SCM)
• Single channel of monolith structure
• For a radially uniform monolith, the SCM is
representative of the whole reactor

• There are many approaches to writing a single

channel model:
– PDE models (all 2D and 3D models, some 1D)
– ODE models (some 1D models)
– Algebraic equations only (some 1D models)

12
The 3D SCM
• The real monolith does not have
axial symmetry.
• The washcoat can have a variety
of possible shapes.
• 3D is required for exact channel
representation

13
The 3D SCM
• A minimum of a one-eighth
section must be considered.
• For complex shapes such as the
“sinusoidal” channel, the entire
cross section must be
considered.
• There is no analytical solution
for the velocity.
• Must solve the transport
equations in three dimensions.

14
The 2D approximate SCM
• The 2D model approximates the
channel as a right circular cylinder
with a uniform annular washcoat
layer, thus giving axi-symmetry.
• The flow area, washcoat area and
substrate area should be the same
for each system.
• The 2D model is still the most
complex model that is used
routinely.
• The fully developed flow field can
be approximated, thus reducing the
number of equations to solve.

15
 2D and 3D SCM
• 2D and 3D SCM are both distributed parameter
models.
• Temperature, concentrations and velocities vary with
both axial (flow direction) and radial (transverse to
flow) directions.
• The most general model requires the solution of a set
of coupled non-linear partial differential equations.
• The equations are the same in each case – only the
number of space dimensions changes.

16
The 2D SCM
• All of the physical/chemical processes present
in 2D and 3D can be illustrated using a two
dimensional model.

Mass Transfer Heat Transfer

Convection Radiation
Direction of flow

washcoat Diffusion and reaction Conduction

substrate Conduction

Centreline of wall 17
Basic solution procedure
• Each physical space (fluid, washcoat and substrate)
is a computational domain governed by the
conservation equations.

• Write the governing differential equations for each

domain (fluid, washcoat and substrate)

• Often only the solid state energy balance equations

retain the transient terms, owing to the relative
magnitude of the time scales for fluid and solid.

18
Modelling the fluid domain
Velocity and pressure are obtained from the Navier-Stokes equation
and the equation of continuity:
⎡
( ) ⎛ 2η
ρ ( v i∇ ) v = ∇i ⎢− p I + η ∇ v + ( ∇ v ) − ⎜
⎣
T

⎝ 3
⎞ ⎤
− κ ⎟ ( ∇ iv ) I ⎥
⎠ ⎦
∇ i( ρ v ) = 0 The density depends on temperature.

The mass balance (assuming Fick’s law) is:

∇ ⋅ ( DA ρ ∇wA ) − v ρ ⋅ ∇ ( wA ) = 0
Species conservation should be
The energy balance is: expressed in terms of the mass
fraction because mass is conserved.

( ) (
∇ i k f ∇T − ρ C p ) f vi∇T = 0
19
Observations
• The fluid equations are very classical and can be
readily solved.

• The mass and energy balances are coupled to the

washcoat at the gas solid interface.

• A zero slip condition is usually imposed at the gas

solid interface and a zero velocity boundary
condition imposed.

• One mass balance equation needed for each

species.
20
Modelling the washcoat domain
• The washcoat is a porous matrix containing solid
and fluid fractions.
• The diffusion/reaction occurs in the fluid fraction,
which is comprised of a network of tortuous pores.
• It is too complicated to model the actual diffusion
in the porous network (and the structure is
unknown in any case)
• Diffusion is thus assumed to occur across the entire
matrix and is governed by an effective diffusion
coefficient, with a Fickian type of approach.

21
Washcoat balance equations:
The mass balance equations are solved for the washcoat:

∇ ⋅ ( ( Deff )A ρ ∇wA ) − (− R A ) = 0 One mass balance for each species

The energy balance is:

∂T
(
∇ i( keff ∇T ) + ( −ΔH R ) (− R A ) = ρ C p )S ∂t

Normally the momentum balance equation is not solved in the washcoat.

22
Diffusion in the washcoat
• Diffusion in the pore is governed by a combination
of bulk diffusion (molecular collisions) and Knudsen
diffusion (collisions of molecules with the wall).
• The relative importance of the two depends on the
gas pressure and the pore size.
• The effective diffusion coefficient is then computed
by adjusting for catalyst porosity and tortuosity.
• For a real catalyst with a given pore size
distribution, the calculation of effective diffusion
coefficients is not obvious.
23
Diffusion in the washcoat
• For binary eqi-molar counter diffusion of molecules
A and B we write, for the diffusion of A:
0.5
⎛ 1 1 ⎞
0.01013 ⎜ + ⎟ T
DAB = ⎝ MA MB ⎠ T 1.75 ( DK )A = 48.5 d p
(( ∑ν i )A )
MA
+ ( ∑ν i )
0.333 0.333 P
B
Knudsen diffusion
Bulk diffusion: Fuller correlation

ε
1 1 1 ( Deff ) A = ( D pore ) A
= + τ
( D pore )A DAB ( DK )A
Effective diffusion coefficient
Diffusion coefficient in the pore in the washcoat
24
Modelling the substrate domain

The energy balance is:

∂T
(
∇ i( k eff ∇T ) = ρ C p )S ∂t

The substrate is a porous medium and is modelled using effective properties.

25
Remarks: 2D and 3D models
• The basic equations are the same for 2D and 3D
models.

• Solution can be done using commercial or custom

software 3D models will take much longer to solve.

• Normally, a finite element or finite volume method

will be used.

• For steady state models the substrate equations

can be eliminated.

26
Comparing 2D and 3D models
• The channel reconfiguration can cause model
differences.

Increasing boundary
temperature

Reactant concentration at two temperatures in a washcoat fillet.

27
Comparing 2D and 3D models

Reactant concentration in a complex shape

28
Comparing 2D and 3D models
• The following light-off simulations show some
comparisons between 2D and 3D models.

• 2D model uses a circular washcoat and a

circular substrate (circle/circle)

• 3D model 1 uses a circular washcoat in a

square substrate (circle/square)

• 3D model 2 uses a square washcoat in a

square substrate (square/square).

More et al., Topics in Catalysis, 37 155 (2006)

29
 Comparing 2D and 3D models
100 100

Circle/circle Circle/circle
Circle/square Circle/square
80 80
Square/square Square/square

Percentage Conversion
Percentage Conversion

60 60

40 40

20 20

0 0
475 500 525 550 475 500 525 550 575

Inlet Gas Temperature /K Inlet Gas Temperature /K

Inlet temperature ramp rate of 30 K/min at a GHSV of 20 000 h-1 and 80 000 h-1.

More et al., Topics in Catalysis, 37 155 (2006)

30
Comparing 2D and 3D models
100
Circle/circle
Circle/square
Square/square
80
Percentage Conversion

60

40

20

0
500 520 540 560 580 600 620 640
Inlet temperature / K

Inlet temperature ramp rate of 300 K/min at a GHSV of 80 000 h-1.

More et al., Topics in Catalysis, 37 155 (2006)

31
The 1D SCM approximation
• 2D and 3D models can be time consuming to run
and may not be necessary.
• The radial dimension is eliminated using radial
average properties. The model is transformed into a
lumped parameter model in the radial direction.
• Parameter lumping introduces a discontinuity at the
gas solid interface which is handled using mass and
heat transfer coefficients.
• The washcoat and substrate are collapsed into a
single one dimensional equation.
32
Review of boundary layer flow
Flow over a flat plate with exothermic
U∞
U∞ reaction at the surface:
δ
Boundary layers develop for velocity,
temperature and concentration.

T∞ The boundary layer thickness increases with

length
T∞
δT
TS
The boundary layers have different thickness:

C∞
C∞
δ δ δT Scn
= Sc n = Pr n =
δC δC δT δC Pr n
CS

33
Thermal boundary layer

T∞
T∞
δT
TS

∂T
The heat transfer at the surface is : q′′ = − k
∂y y =0

The heat transfer can be defined as: q ′′ = h ( TS − T∞ )

The heat transfer coefficient h assumes a value to make the equality true.
If the temperature gradient at the surface is known, then h can be calculated.

34
Boundary layers in a tube
• At the entrance, boundary
layers grow from the surface.

• After the entry length, the

boundary layers meet.

• The fully developed region can

be considered a boundary
layer flow.

• For laminar flow in a tube, the

velocity, temperature and
concentration fields can be
obtained from the solution of
the governing PDE.
With reaction at the wall, the concentration
boundary layer development is similar.

35
Heat transfer in a tube
• In a tube, the flow becomes fully developed after
an entry length.

• There is no well defined reference temperature

external to the boundary layer, and therefore the
heat transfer coefficient is defined in terms of the
mixing cup temperatures (average radial
temperature)

• The mass transfer coefficient is defined in a

similar way.

36
Heat and mass transfer
• Heat and mass transfer coefficients can be
expressed in terms of dimensionless groups, the
Nusselt and Sherwood numbers.

• For a circular tube:

hD km D
= Nu = Sh
k DA

• The heat and mass transfer coefficients can be

calculated by solving the 2D or 3D flow problems.

37
1D SCM: Model equations
• The form of the equation depends on the channel
geometry and the basis for the reaction rate:

• Example is shown for a circular tube with the rate

expressed in terms of washcoat volume:

38
1D SCM mass balances
The fluid phase mass balance expressed in terms of the mixing
cup concentration and the mass transfer coefficient is:

∂ ⎛ ∂ wA,f ⎞ ∂ wA,f ⎛ 4 ⎞
⎜ DA ρ f
∂z ⎝ ∂z ⎠
⎟ − ( f )i
ρ v
∂z
( )
− km ρ f wA,f − wA,S ⎜ ⎟ = 0
⎝ DH ⎠

The axial diffusion coefficient is now a dispersion coefficient.

The velocity is an average velocity.

39
1D SCM mass balances

The solid phase mass balance is:

4D
(
k m ρ f wA, f − wA,S ) = η ( − RA )S
( 2
DWC −D 2
)

Washcoat diffusion is included via the effectiveness factor.

The rate here is evaluated at the surface temperature and

concentration.

The rate is expressed in terms of washcoat volume.

40
 1D SCM energy balances
The fluid phase energy balance is:

d ⎛ dT f ⎞
( )
d Tf
⎜ I
dz ⎝
k ⎟
dz ⎠
− vm ( ρ C p ) f dz
+
4
D
h TS − T f = 0

The fluid thermal conductivity is an effective value that accounts for laminar flow.

The solid phase energy balance is:

∂ ⎛ ∂ TS ⎞ ⎛ DWC
2
− D2 ⎞ ∂ TS
⎜ S
∂z⎝
k ⎟ −
4D
∂ z ⎠ ( DS − D )
2 2
(
h TS − T f ) − η ( ΔH R ) ⎜ 2
⎜ D − D 2 ⎟⎟ ( − R )
A S = ( ρ C )
P S
∂t
⎝ S ⎠

The solid thermal conductivity is an average value for washcoat and substrate.

41
Nusselt and Sherwood numbers
• The correct values of Nu and Sh are not obvious.
• In the fully developed region, the heat and mass
transfer coefficients are constant. The value
depends on the wall boundary condition.
• The classical boundary conditions are constant wall
temperature and constant wall flux.
• With a reaction on the surface, neither boundary
condition is true.

42
Nusselt and Sherwood numbers
• For flow in a circular tube with constant physical
properties, the classical result is:

43
Nusselt and Sherwood numbers
• Equations for the local Nu for Pr=0.7 with combined
entry length have been proposed. They can be
expressed generically as:

⎡ ⎛ 50 ⎞⎤ The constants depend on channel

Nu = C1 × ⎢1 + C2 Gz exp ⎜ − ⎟⎥
shape and wall boundary condition
⎣ ⎝ Gz ⎠⎦

The value with reaction is found by interpolation between the values for
constant wall temperature and constant wall flux.:

2
Nu H ⎛ Nu H ⎞
2 Nu = Nu H − Da + ⎜ Nu H − Da ⎟ + 4 Da Nu H
NuT ⎝ NuT ⎠

44
Nusselt and Sherwood numbers
• However, when the properties are not constant, the
entry length can be more complicated:
7
Local Nusselt number
T = 500 K obtained from entry length
solutions for laminar flow in a
6 T = 450 K
T = 550 K

channel with different wall

5
temperature. No reaction.
Nu

More, MSc Thesis, University

of Alberta, 2007
3
0.001 0.01 0.1 1

1/Gz

45
Nusselt and Sherwood numbers
• So, what should we choose for the correct value?
• On the bright side, for circular channels, it might not be
that important:
100

Nu = 4
80 Nu = 3
Nu = 5
Axial conversion for a
typical monolith channel for
Percentage Conversion

Nu = 30

60 CO oxidation.

40
More, MSc Thesis,
20
University of Alberta, 2007

0
0.00 0.02 0.04 0.06 0.08 0.10

Axial Distance, m
46
 Nusselt and Sherwood numbers
• However, for non-circular channels,
or non-uniform washcoat, the Nu and (a)

Sh can vary around the washcoat

perimeter.
6
Sherwood number along the
5 washcoat/channel interface
for the case of a circular
Sherwood number

4
550K channel in a square
600K
650K substrate. Interface length
3 700K
750K shown corresponds to 1/8th
800K
850K of the total interface. Fully
developed flow.
2 900K

1
0 50 100 150 200 250 300 350 400 Hayes et al. Chem. Eng. Sci., 59
Distance along the interface, microns 3169 (2004). 47
Comparing 1D and 2D models

• 1D models can give very close agreement to

2D models in many situations.

• Generally the most difficult region to map

exactly will be the developing flow region, but
that length is often not significant.

Several studies in the literature comparing 1D and 2D models,

e.g. Groppi and Tronconi, Aiche Journal 1996.

48
Simplified 1D models
• Dispersion in the gas phase is sometimes ignored,
and plug flow assumed.

• If axial conduction in the solid phase is ignored, the

model can be reduced to an initial value problem.

• With these two assumptions, the channel can be

modelled as a finite number of stirred tanks in
series.

49
The effectiveness factor
Because the washcoat thickness is not explicitly
modelled, the washcoat diffusion is incorporated
using an effectiveness factor, defined as:

Average reaction rate in washcoat

η=
Rate evaluated at surface conditions
The effectiveness factor can be estimated from a sub-
model that usually involves solving a 1D diffusion-reaction
problem. Washcoat geometry effects may be included.
Hayes, et al Chemical Engineering Science, 62 2209-2215, 2007.
Hayes, et al Chemical Engineering Science, 59 3169-3181, 2004.
50
Models for the converter
• In reality the channels in the converter may not
all have the same operating conditions.
• The substrate constrains the velocity and
concentrations to each channel, but heat transfer
can occur among the channels.
• A model of the entire converter may therefore be
required. There are two general methods for
modelling a converter:
– Discrete models
– Continuum models

51
Discrete converter models
• In a discrete model both the fluid and solid
domains are solved:

End view of a quarter section of a circular

monolith with imposed temperature
gradient.

Heat transfer occurs through a combination

of gas and solid with very different
properties.

Hayes et al., Catalysis Today, 2009

52
 Discrete converter models

• A passenger car converter has

about 7 – 10 thousand channels

• Laboratory tests often use 1-2”

diameter cores with 300 – 1200
channels

Discretizing the entire physical space result in a problem too large to solve
by conventional means. Until new methods are proven typically use ….
53
Continuum converter models
• Classical porous media models treat the domain as
a continuum containing fluid and solid phases.
• The common term for developing the resulting
equations is called volume averaging – the balances
are made on a representative elemental volume
that contains both fluid and solid
• Each node in the computational domain represents
both a fluid and a solid – the spatial identity is lost.
• The conservation equations are then fairly classical:

Zygourakis, K. (1989). Chemical Engineering Science, 44, 2075-2086.

54
Momentum balance equation
Velocity and pressure are obtained from the Volume Averaged Navier-
Stokes equation and the equation of continuity:

⎡
( )
⎛ 2η
ρ ( v i∇ ) v = ∇i ⎢ − p I + η ∇v + ( ∇v ) − ⎜
⎣
T
⎝ 3
⎞ ⎤
− κ ⎟ ( ∇ iv ) I ⎥ + Si
⎠ ⎦
∇ i( ρ v ) = 0
The velocity is the superficial velocity. The Brinkman
equation is a classical
The source term S accounts for the porous inertia. version of the VANS

⎛μ 1 ⎞ K is the permeability.
Si = − ⎜ vi + C2 ρ f v vi ⎟
⎝K 2 ⎠ For laminar flow, the second term is zero.

55
Energy and mass balances
• There are two main classes of porous media models.
• Heterogeneous model:
– The fluid and solid temperatures and concentrations
are taken to be different. The two phases are coupled
using heat and mass transfer coefficients.
• Pseudo-homogeneous model
– The fluid and solid temperatures and concentrations
are taken to be the same. This situation corresponds
to infinitely large heat and mass transfer coefficients.

For transient simulations the heterogeneous model is recommended.

56
Energy balances
The fluid phase energy balance is:

∂Tf
φρ f CP , f
∂t
( )
+ ρ f C P, f v i∇T f = ∇i keff,f ∇ T f − h av (T f − TS )

The solid phase energy balance is:

∂ TS m
(1 − φ ) ρS CP ,S
∂t
( )
= ∇ i keff, S ∇TS + h av (T f − TS ) + ∑ (ΔH R )i η ( Ri )
i =1

The effective thermal conductivity for each phase depends on the

porosity and material properties.

see Hayes et al., Catalysis Today 2009

57
Mass balances

The fluid phase mass balance is:

∂
φ
∂t
( ) ( ) (
ρ f wi, f + ρ f v i∇wi, f = ∇i φρ Di ,m∇wi , f + k m av ρ f wi , f − wi, S )
Mass diffusion occurs only in the flow direction

The solid phase mass balance is:

(
0 = ηi Ri + km av ρ f wi , f − wi ,S )

Washcoat diffusion is included via the effectiveness factor

58
Continuum model
• In effect, the continuum model solves a set of 1D
single channel models with channel coupling via
solid phase heat transfer.

• However, every channel is not modelled.

• Heat and mass transfer coefficients are the same
as used in the 1D SCM.

• The continuum model consists of a set of coupled

PDE, which are solved using the FEM or the FVM.

59
Example:
3D reverse flow CondorTM converter
3D heterogeneous model. 10 litre converter
for methane IC engine with reverse flow.
Fluent was used, dual zone model.

Liu et al. Computers and Chemical Engineering, 31 292 (2007)

60
Model validation
• All models must be validated for:
– (1) Coding (errors in model implementation)
– (2) Numerical errors
– (3) Simplifying assumptions

• Models can be validated by

– Comparison to experiments
– Analytical solutions

• No model can be completely validated.

61
Model validation - Example
• Oxidation of methane in air
• Reactor wall temperature measured along reactor
length

• Step changes in methane concentration

• Kinetics determined in separate experiments
• 2D single channel model with washcoat diffusion

62
Model validation - Example

Heating curve with step increase Cooling curve when

in methane concentration methane switched off

Hayes et al, I&EC Research, 35 406 (1996)

63
Concluding remarks
• An appropriately chosen model can provide a
useful design aid.
• Models can be used as a tool to investigate the
phenomena occurring in the reactor.
• Recognize the limitations imposed by the
simplifying assumptions.
• Good values for the physical properties are
essential.

64
 Advertising supplement

R.E. Hayes and S.T. Kolaczkowski

Introduction to Catalytic
Combustion

Taylor and Francis, 1997

65