The two day workshop was held
at Detroit Diesel. Presentations are to
be posted on the CLEERS web site in the near future, http://www.cleers.org/.
DPFs are being modeled and
tested. The deposition of soot in a
filter is controlled by the Peclet number – the ratio of inertial to diffusion
forces on a particle. Different operating
characteristics change the penetration into the filter wall, and the
density/permeability of the soot cake. The process of understanding DPFs is
very rapid, and many papers are being published. If we already understood it, there would be
few papers!
The pressure drop, filtration,
and wash coating capability are strong functions of DPF pore size and size
distribution, as well as cell density and wall thickness. These same parameters also affect the
strength. Substrates suitable for
catalyst coating are much weaker than uncatalyzed designs. A narrow pore size distribution is highly
desired. A clear understanding of material properties along with flow and
temperature distribution is needed in order to assure long life. Thermal stress is a major issue. Catalyst coatings affect physical properties,
and it is the coated filter that must survive.
System models are being
developed to include DOC, DPF, and NOx catalysts. Manufacturers would like a nice set of data
to be provided by suppliers; the data would be put into models to allow system
design. Unfortunately, it may not be
readily possible to have a simple data set to describe a complex, history
dependent device like and LNT or a DPF.
Chalmers and GM R&D have
developed a global kinetic LNT model for a model Ba/Pt/alumina catalyst. They have also done a commercial catalyst but
those results can’t be shared.
ORNL has tested different LNT
regeneration strategies on an engine and reports the speciation of
The LNT Subgroup is
developing a standardized test protocol for LNTs, intended to get enough data
to calibrate a kinetic model in a reasonable length of time.
LLNL is starting some
interesting quantum chemical computations for cases related to LNTs.
ORNL is doing DRIFTS and
other testing on model LNT catalysts to find mechanisms information. Nitrate and carbonate are the dominant
storage species. There are interaction
with PGM and supports. Water is also
important. Key deactivation mechanisms
include sintering of PGM and Adsorbates, and apparently coverage of PGM sites
by Adsorbates. Kinetic models of the
lean storage are being developed.
A reformer can generate H2
and CO from fuel onboard. These
reductants can reduce the required temperature for deSOx and deNOx, potentially
improving durability and efficiency.
There is continuing work
toward HC-SCR catalysts. Many years of
research has led to better understanding of mechanisms, but the important
chemistry is strongly dependent on what conditions (HC species, temperature,
water, etc) are being run. Adding H2 to
HC greatly improves low temperature NOx conversion and widens the conversion
window. GM has a DOE program that is
using combinatorial methods and finding improved HC-SCR catalysts.
ORNL has done careful
analysis of exhaust in a urea SCR system, looking into urea decomposition and
reaction products. At low temperature,
you might make a lot of nitrates. At
higher temperatures, you might make some deposits. There is very large storage of urea-derived
components in the SCR catalyst; a control system must model this to avoid large
NH3 break through.
Ford is continuing
development of a diesel SCR SUV.
Emission cycle results are approaching Tier 2 Bin 5 at low mileage. Work continues, and will be followed by
durability testing.
NREL is working with AVL to
include SCR catalysts in the FIRE code.
JohnsonMatthey and
Walter Putz, Senior VP
Engineering,
This is an important time to
talk about simulation for diesel aftertreatment. New standards are forcing major developments
for cars and trucks. In the truck
business, life cycle cost is critical.
We have to control aftertreatment cost and also usage costs. The engine-aftertreatment system is a complex
system and must be handled together – you can’t just put a DPF on an unmodified
engine. Simulation tools are an
important part of the design and development process.
CLEERS is Crosscut Lean
Exhaust Emissions Reduction Simulation.
“Crosscut” comes from the parent DOE Diesel Crosscut Team. CLEERS is to foster communication and
collaboration – not to write software.
Several forums have been
developed including these workshops.
Common terminology and standards are being developed.
The CLEERS team leadership is
Blint GMR, Daw ORNL, Sisken DDC, Harold Kung (NW Univ), Singh DOE.
There are subgroups for DPF,
LNT, and SCR. This emphasis came from
input by the users. The Focus Groups each have 5 or more Crosscut members, plus
sponsored participants such as catalyst companies. The Groups hold monthly net meetings. 6-7 engine mfrs, 1-2 universities, 1-2
National Labs are a typical Group makeup.
Kinetics of the catalysts is
a key issue to be studied. Controlled
lab data is needed to develop the kinetic models. ORNL is developing a set of shared data.
There is a web site that
compiles information and articles. There
are now over 500 citations on the web site.
There is a quarterly newsletter.
Mansour Masoudi,
At the last CLEERS meeting,
George Muntean reported variations in soot properties according to deposition
conditions. It usually takes 2 years for a new finding to get widespread
attention and be utilized in industrial applications.
Filter pressure drop is a
system in itself:
You can map soot mass
emission rate as a function of speed, load, EGR, temperature …but in many cases
we don’t have this data yet.
You need to understand and
quantify the roles of all three of these attributes. Most papers just give a plausible
explanation, but you really need to quantify things. This paper will deal only with the soot
structure issue.
You can write the equation
for dP; there are SAE papers that show it. To have a good understanding, you have to
know soot density and permeability.
The flow through the filter
has an enormous effect on soot density and permeability. High flow (>7-8 cm/sec normal velocity at
the filter wall) means the soot particles tend toward ballistic impact
deposit. At low flow rates, deposition
is driven more by diffusion and less by particle inertia. The latter gives a
lower density, more permeable soot deposit and thus lower dP
for the same total soot mass.
How do you quantify this?
Equations were shown to estimate mean free path (MFP) of gas molecule; Knudsen
number (relative significant of mean free path to diameter of PM),
Stoke-Cunningham factor, and Kubawara function.
These allow estimation of the soot permeability. The difference between a leaf
in the wind versus soot in gas flow is the relation between MFP and particle
size. The Stroke Cunningham factor
embodies this effect.
The Peclet number is ratio of
inertial force to diffusion force. Soot
layer porosity is a function of Peclet number.
The density of the soot layer is then related to the porosity.
Thus, two major
characteristics of the soot layer – density and permeability – can be
identified. You can now plot these parameters vs Peclet number. Around Pe=1, there is a strong increase in
density and decrease in permeability. In
other words, if deposition velocity is large you get dense layers with low
permeability. Pe is much more important
to permeability than temperature.
With these two values, you
can calculate dP knowing the filter
characteristics. Note that dP alone does not provide a reliable estimate of PM mass,
unless you know the conditions under which soot was deposited.
SAE 2003-01-0842 gave
particle layer packing density vs Pe.
This matched the theory well, both for bare and catalyzed DPF. The presence of the washcoat has little
effect on soot deposit; deposit occurs before catalysis.
A slide showed good agreement
between dP calculated using
the above analysis; it cam very close to the test data.
You could implement the
equations in a controller to predict the current soot loading in real
time.
Of course, the estimation of
soot in the DPF only works at low temperatures where soot is not
oxidizing. For a real system, you will
also have to have a regeneration model.
Does soot get “stickier” when
hotter? The effect of temperature is in the analysis – i.e. in MFP. The
model as presented does not make any assumption about SOF-related sticking etc.
If you have a loose layer,
then go to high flow does it change the soot layer that was already there? The
model assumes you don’t, but we don’t have hard data. On the other hand, the model seems to agree
well with data so maybe the assumption is not too bad.
Prof.
The Center studies fine
particles as suspensions in gas media – either as pollutants or intentional
(catalysts in industrial processes).
Their lab has 24 full time
engineers and scientists, all funded by industry contracts and grants.
To make the DPF system work,
you need a DPF simulation tool for a variety of users, and also algorithms such
as virtual sensors for aftertreatment management and control. Different users (substrate mfrs, coaters,
engine makers,…) have different uses for the models.
There are a number of DPF
configurations – Cd, SiC, sintered metal, fibrous ceramic…. This is like a Cambrian explosion – lots of
diversity that will eventually move toward a single winning version.
An interesting graph shows
number of simulation papers versus time from 1980 to now. We are about ˝ way
through sigmoid shaped curve typical of new technology development. We are right about the peaks of the publication
rate! The curve predicts that by 2012
there will be little new publication – either the technology goes away or else
you understand it by then!
There are multiple scales of
DPF systems – microns pore size, wall thickness,
entire filter, and entire exhaust system.
We have to understand all these scales.
You can’t run a single model that includes the finest scale over the
full volume in a reasonable computational time.
So, you use different models for different purposes.
At the nano/micro scale you
use
At meso scale, unit cell
filtration codes, single and multi channel DPF simulation (WALTER), and exhaust
line simulator (ELVIS).
It is important to avoid
short cuts that use heuristic models – they are not general and cause problems
later. Each level must be mathematical!
At the macro scale, couple to
CFD and FEA with user defined functions, to calculate stress, temperature
field, etc.
Diesel aggregates have a
fractal structure – soot is not spherical!
The fractal overlap seems to be around 0.174. Recently, extensive data was taken of
effective density versus mobility diameter for different engines and
speed/loads. As diameter increases,
density drops. The fractal dimension
also varies with size, running about 2.4 on average.
The normalized size
distribution is remarkably similar over a range of engines. Sigma is about 1.89. This is handy for instrumentation and for
modeling. The shape can be explained by
a model of oxidative fragmentation. That
model says the larger a particle gets, the more likely it will break. The fragmentation must be random in order to
fit the data (i.e., soot does not break in half or some other constant
fraction, but rather at a random location).
The models of DPFs include wall
scale, channel scale, and DPF scale features.
(See slide for more detail).
Current commercial models of the walls have gas temperature but not
surface temperature considered; that’s a future enhancement.
The micro scale is done with
Lattice Boltzmann. You need a
reconstruction of the wall structure; they make digital models from
pictures. This can be done with several
algorithms for granular media, with a library or grain types, packing, and
resulting structure. This can also be
done for fibers. The porous structures
we use have a lot of wasted porosity; regions the flow will not really go
through. The comparison of model to test
results is quite good.
Next, there are algorithms
for catalyst coatings – both uniform and non-uniform. Coaters sometimes think there is such a thing
as a uniform coating! A uniform coating,
however, reduces the cross section of the flow path. Non uniform coating depends on coating
method. Coating is usually by
dip-and-blow. Here, the catalyst is
removed where the flow velocity is high.
The different methods can give very different permeability versus
loading and fraction of pores filled.
Coating can also be done on a
filtration method. This gives good
results.
The dynamics of soot
filtration are being modeled also.
Initial deposition is in the wall, followed by a cake development. Layer characteristics can be predicted.
DPFs are very good at
collecting small particles; if you see small PM downstream, they have condensed
after the filter.
Soot deposits first at the
high flow locations. That area plugs
first and the flow redistributes to cleaner areas. You have to feed back soot deposit into the
flow field calculation.
Soot cake personality is
characterized by the Pe number. You get
more wall deposition at higher Pe.
Ice particle growth on a
window is very similar to soot deposition growth dynamics!
Soot cake geometry can become
non-convex at low Pe number.
A quenched diffusion flame
burner was used to provide reference soot distributions. With this, we learned that input PM size
affects the soot cake density; larger PM makes looser cake. We don’t see this so much in engines because
the soot input size is more constant than in this bench experiment.
The soot microstructure
changes as you oxidize the soot. This
has been modeled (series of slides shown).
Density is linear with oxidation rate.
The model agrees well with regeneration experiments.
At the DPF level, there are
practical issues such as the glue thermal characteristics in segmented
filters. If the glue is a good
insulator, there are larger temperature gradients in the catalyst. You can get regeneration of some segments
without other regenerating. Yu can get
partial regeneration when the temperature is non uniform.
As others have said, partial
regeneration destroys the correlation between dP and
soot mass.
They are now trying to use
hooks in CFD codes to put the DPF models in. These codes can predict soot
deposition geometry and thermal distribution.
A DOC in front helps the
temperature for regeneration on the DPF.
The ash distribution (SAE
2003-01-1963) along the length depends on the shear on the ash layer, and the
ash shear strength.
You have to know the origin
and pathways of ash. It can be by pyrolysis on the filter, or combustion upstream. Ash can re-entrain at high shear. Ash stickiness changes depending on its
history. More work is needed to characterize ash!
Future emission control
systems will be very complex, whether you like it or not. You will have multiple catalysts/DPF units,
sensors and virtual sensors, and control algorithms. Models will have to be made faster to use in
vehicle ECU. Fiat paper at FISITA
F2004V068 showed DPF control algorithms based on this work. Their paper shows the value of modeling!
The regeneration oxidation
model shown in these slides is very simple. We have more complex models but
no time today to show it. NO2 and O2
mechanisms are both important. A 2-layer
model is used: some soot in contact with catalyst, some not, so there is an effectiveness
factor. HC or wetness of soot is also
important. We don’t see an effect of wet
soot; it becomes dry by the time it is hot enough to react. You might get some added heat release from Pt
oxidation of the desorbing HC.
Greg Merkel,
You can show dP versus flow at various soot loadings; generally a
parabolic fit. Or, you can look at dP versus soot loading at various flow conditions.
dP affected by
We are particularly
interested in soot penetration into wall.
How much gets in, how deep does it penetrate, location within the pores
(uniform coating or local?), and packing density in pores.
Slide shows 6 different pore
structures developed. These have
different dP characteristics – initial dP, slope of
initial dP versus soot loading.
The simplest case is a filter
that only has a surface cake, no penetration. Useful filters, though, have an
initial steeper dP slope. Some filters have a low initial slope that
seems to be building a uniform coating on the pore walls, followed by the
steeper and then the shallower slope.
Each of the above three cases
is modeled as two overlaying slabs with different thickness and permeability.
The flow is treated with an
approximation of randomly oriented capillary pores. The Hagen-Poiseuille
equation relates dP to capillary geometry. You sum over all pores to get an equation for
the wall. Permeability can then be
estimated, and is linear in porosity and quadratic in pore diameter. K=C phi d2.
This model was used to
compare to literature data. The
agreement is pretty good to porous glass and MgO
ceramics of known characteristics.
Agreement is also pretty good for cordierite (Cd).
The soot deposit permeability
also has to be modeled. The
Konstandopoulos method was used. Soot
cake porosity is typically >90%.
These are combined then to
calculate filter dP.
Assumptions for following slides are listed; high SV, 180K.
For the case of a soot cake
forming without permeation into the wall a low slope versus loading is
predicted. For the case where soot uniformly fills surface and near surface
pores you get a steeper slope. For the case where soot deposits uniformly coat
the capillary walls you get very little pressure increase until the model
begins to close the entire pore.
One can then compare modeled dP to experimental data with Cd
DPFs.
How do you control that
initial slope? A study done with 110 Cd filters having various structures. 38-57% porosity, different
pore size and distribution. These
generated a wide range of dP slopes. The Konstandopoulos model was used to
generate permeabilities and coefficients, and these were fit to the
experimental variables.
Pore size distribution
defined by the diameter of 10, 50, and 90 percentile. Also, d90-d10 is width of
distribution. You can also normalize the
parameters: (d90-d10)/d50 etc. These parameters were part of the dP data fitting.
Permeability of a clean wall
correlates well (r2 = 0.87) with d502.
A soot filled wall
permeability is dominated by (d50-d10)/d50.
The d502 term is now minor.
High porosity and narrow pore size distribution maximize
permeability. (d50-d10)/d50
seems to be related to whether the flow path seems like large pores connected
by small necks, or more like a constant width channels. A large value is more-connected. Narrow necks increase the likelihood of dense
soot packing in the pore necks.
Summary:
Pore microstructure is
important
A random oriented capillary
model matches data
It has been extended to the two
layer wall + soot cake model
Experiments show desired pore size distribution characteristics.
Prof K: interesting
work. What “pore size”
means is a matter of definition. There
are solid reasons for the data shown.
Permeability always scales with a characteristic size squared and a
constant around 1.6. Deposition is
dominated by stagnation regions. It is a
matter of style if you prefer to model flow in pores versus flow around
objects. We think it is easier in denser
media to model flow around objects. the sharp regions of slope are related to percolation
mechanisms. I don’t think the flow
acceleration mechanism is really likely; the flow can move to another channel
instead of accelerating. The high slope
is related to the penetration depth. The
best filter is the one that keeps the filter wall separate from the soot cake –
but that is the worst if you want to get catalyst and soot into contact! Of course, the capillary pore model is
only a rough approximation.
The pore distributions are
based on mercury porosimetry, so they are volume related.
In real filters the pores
have a lot of interconnection, while your model has none. How can it match the data as well as it does?
I don’t really know! It does seem surprising that such a simple model fits
the data so well. Perhaps it is because
the porosimetry only measures the “accessible” pores.
We are also now doing
image analysis of real filters.
Tariq Shamin,
U
Most available DPF
regeneration data is at steady state. In
the real world, regeneration will have to occur under transient driving
conditions. Can we regenerate at lower
temperature by some form of modulation.
Their model is based on Bisset (like everyone else’s!). This is a fairly simple model. A
The model’s predicted
temperature matched published data pretty well for a SS
regeneration.
Using the model several cases
were evaluated. If inlet temperature is
sinusoidally varied +/-50C at 0.01 Hz, the Toutlet
varies less than the Tbed. Wall temperature is not symmetric around the
mean value. dP
also varies sinusoidally.
Exhaust flow rate modulation
does not affect the filter temperature but strongly affects dP.
For a soot
loaded filter at Tinitial 510C (below light off),
+/-50C inlet temperature modulation at 0.01 Hz you get more temperature rise
and faster regeneration. At 0.1 Hz, the
effect is damped out and it looks more like SS.
For higher temperature, 500C
– close to soot ignition temperature, you see larger effect of modulation.
Modulation of exhaust flow
rate has major dP effect, and small improvement in
regeneration.
For a CDPF at 250C modulation
has a larger effect of enhancing soot regeneration. At 300C (regeneration will go by itself),
modulation increases regeneration completeness and speed.
This arises because chemical
reaction rates are nonlinear: for the same average T, modulation increases the net
reaction rate. Low frequency is
necessary since thermal mass will prevent fast modulation from having an
effect.
Did you normalize for the
total energy input? Is there an advantage in regeneration versus total energy? For
soot regeneration, you see accumulated effect.
What would you have to change
in your model to reduce the effect of frequency? Is it a material property, a soot property,
or….? I don’t know if it is mass or heat transport limited. It seems likely it is the heat capacity or
thermal mass effect.
Suresh Gulati,
Research Fellow and Consultant,
DPF requirements:
Durability:
LD needs something like 200K
miles durability. You can’t have (large
scale) cracks, although a flow through catalyst substrate can. HC needs 435K miles.
Stresses and relevant
strength characteristics:
Physical properties
Maximum stress occurs
somewhere in the center, not at the ends where the plugs are. Property
measurements focus on unplugged section.
A flex test
measures bending strength – axial, direction of flow – or face plane –
tangential.
You can do a crush test in 3
directions.
A 3D isostatic test
pressurizes in all directions. It finds
what stress can be applied during canning. A 2D isostatic tests can also be
used to predict canning.
The focus is on isostatic and
bending strengths.
Strength measurements are
usually made in a short time, while real parts have loads applied for
years. So, the service load must be
lower than the ultimate strength measured.
Web thickness and porosity
each affect strength. Wall crush strength in the axial direction drops
exponentially with porosity. Thus, 35%
porosity gives about 3000 psi, while 50% porosity has about 1500 psi strength,
about half.
Web strength is 17000 exp(-4.9/P) where P is porosity. Isostatic strength is 2.8 (web strength)
(t/L), t= wall thickness, L = cell width.
Ceramics are weaker in
tension than in compression. Canning is
more compressive, thermal stress is tensile.
The honeycombs also have a
ceramic skin; this improves canning strength significantly. The inner part is called the matrix. During
isostatic tests (or canning) the stress in the skin is much higher than in the
matrix. Even so, the matrix is where you
see failure first because it is so much weaker than the skin material. For 900/1, stress skin / stress matrix
=11. For 400/6.5, 6.6. Typically, canning loads are lower than
strength by factors of >2.5.
DPF vs
DOC
Thermal expansion: up to
about 200C, the material actually shrinks.
There is only 140 ppm length change 0-400C. This grows to 1150 ppm at
800C. axial and
circumferential growth rates are different. If the center gets too much hotter
than the outside, then the skin is put into tension and you may get a ring off
crack.
This depends on Tpeak, CTE, Tgradients, how many cycles over lifetime.
A simple model allows input
of measured temperatures. A linear T difference (hot center, cooler outside)
crates maximum stress in the middle.
When there is a T difference along the axis, you get an added
stress. You have to add these two
stresses.
The stress also depends on
the aspect ratio, L/D. Longer DPFs
experience more stress.
A table compares EX-80 to RC,
100/17 vs 200/19.
Measured temperature profiles were put into the model. The radial T
variation causes high stress. Highest
stress is in the middle (between inlet and outlet). A 6” filter would have about half the stress
of a 12” one.
To avoid cracks, the highest
stress must not exceed the statistical lowest strength. To allow less than 0.1%
failure probability, allowable stress is 0.18 So
or about 130 psi. Either you need a
stronger product, or you reduce the thermal stress by regeneration more
frequently.
For a 3 g/L soot load, stress
is about 217 psi. 7.0 č 412. If you
want to go to 1 in a million failures, you need even lower stress. (these numbers are for RC).
Regeneration stress parameter
isx defined as Emod x
(length change center vs edges). This correlates pretty well with regeneration
stress.
Summary:
Thermal cracks can be avoided
by decreasing both coated Emod and CTE by 40%.. In addition,
allowable stress can be increased about 15% by reducing variability of MOR in
the skin region.
This argues you need to
couple CFD to FEA codes in the long run.
Are we close? It is important to get exhaust flow well
distributed. Low flow at the OD
contributes to larger T differences.
Soot should also be evenly distributed.
The inlet cone needs to be a diffuser.
We could also imagine a cell geometry that helps flow distribution.
Doesn’t a DPF self-adjust the
flow top an extent? High flow channels
soot up faster, so the flow should gradually redistribute to other cells. It
might help to direct more flow to the OD – this reduces the dT. Mass and temperature distribution don’t
necessarily go together. It is also harder to regenerate the OD cells.
The T gradient can be
larger is the Tpeak is lower – since the CTE grows at
higher temperature.
Have you looked to see what
happens to the ceramic with fatigue aging? Yes; we see a slow crack growth
due to stress corrosion at high humidity – near the OD.
Controlled regeneration is
not really an issue. The only major
issue is an uncontrolled regeneration, where you start regeneration then drop
to idle.
Mark Stewart, PNNL
Model development started
this year. We are modeling flight and
deposition of individual particles.
There are multiple length
scales of interest in a DPF. We are
looking at the finest scale, discrete particles. In the long run, this work should lead to
characteristics o\at pore and channel scale – we don’t think you would
routinely use such a small scale model.
PNNL started with a 3D
digital map of filter microstructure.
Corning EX80 was cut and imaged.
Images were combined to give a 3D model.
Flow is solved with Lattice Boltzman. Time step
is 3 x 10-8.
Initially we assume that all
PM acts the same as 100 nm particles.
The PM motion is from gas velocity with Brownian motion
superimposed. Particles that hit a wall
always stick. For now there is no
re-entrainment of oxidation.
As PM deposits, the flow
field gets an added flow resistance.
Preliminary results: An example condition was run, taking 88 ours
to run on 64 Itanium2 processors at 1.5 GHz.
About 1 million particles introduced, about 14K escaped. Typically 3K in flight at
any time. The model illustrates
deep bed filtration transitioning to cake filtration. There is little change in flow deep into the
wall. On the surface, the high velocity
areas move as deposits build. The bulk
of gas flow follows a few major channels. (note:
this is EX80, a wide pore size distribution – these flow channels are a reason
for narrow pore distribution)
The model makes reasonable
predictions of porosity at two conditions.
Qualitative agreement is good between photos of real soot and the model
predictions.
A method is being developed
to run experiments for validation. A
single channel filter is placed in exhaust.
Gas is drawn out of the channel, so the outside walls are the filter
medium. You can then analyze the soot
cake without cutting up the sample.
Next steps: compare data to
projections – dP, capture efficiency, deposit location
and structure. Then, apply to other
substrate materials.
How reproducible are these
calculations? You have “random”
motion. So far, we have not changed
the random seed, so it runs the same each time.
What factors go into the
Darcy resistance when a particle hits the wall?
So far, we are using a simple linear relation between the number of
particles in a volume and the Darcy resistance. A future step will be more
rigorous.
EX-80 has a broad pore size
distribution. Have you tried anything
else? Not yet but we plan to. We would be very happy if someone gave us the
3D model! We plan to digitize other
images of the same EX-80 (different portions) to see if the results come out
the same.
Prof K noted that you would
prefer to have a model to build the filter geometry rather than the laborious
method of cutting up a piece.
Tony Triano,
Research sponsored by John
Deere and DOE.
JMI CRT is DOC + DPF,
intended to convert NO to NO2 and use NO2 to regenerate the soot.
The DOC model has routines to
calculate exhaust properties, dP, gas phase energy
conservation and species conservation.
The resulting values are input to a DPF model. The DPF model calculates exhaust properties,
velocity field, inlet temperature, regeneration chemistry, wall temperature, dP and filter effectiveness.
The DOC model is in SAE
2003-01-3176. It includes kinetics,
reaction rates, mass and energy conservation. Reaction rates are assumed first order in
both reaction species and oxygen. All
chemical reactions occur at the walls, and at wall temperature.
NO conversion to NO2 versus
temperature is well modeled against a set of experimental data.
The DPF model is like Prof K’s; DPF only, no catalyst in it. The code has been licensed to GM and U-Wisc Madison. The
NO2 +C reaction was added.
The DOC is JMI part, 10.5 x 6”, 300 cpsi. DPF: 200 cpsi, 52% porosity, 10.5 x 12”. Tested on John Deere engine.
Including the NO2 in the
model predicts much better deSoot; this agrees well with the model. NO2/C ratio needs to be high enough for CRT
to work well.
An issue: if you have a soot
cake and are oxidizing in the wall, are you still accumulating on the soot
cake?
On an engine, it is hard to
tell is you are getting oxidation at the same rate as deposition – the mass
balance is obscured by CO/CO2 in engine exhaust. You will need some lab data too.
CRT will be more of a problem
when you have lower FG NOx.
How do you know you have
repeatable data? The filter was baked in air for a long time to start clean. We have not run a lot of repeat yet.
Sponsored
by GM.
A system model is needed,
including engine exhaust and thermal models, a soot emission model, and an
aftertreatment model.
We used a 1D engine model of
a 4.9L Isuzu engine. The model was
calibrated against 8 Mode
The exhaust thermal model is
1D in gas flow and 2D in heat transfer.
Flow pulsations and the resulting heat transfer effects are
included. It was created in WAVE. The model was validated at 18 SS operating
conditions. Gas stream temperatures were
within 1%, and wall temperatures within 3%.
The engine soot emission
model is a neural network and physical model. See SAE 2003-01-3227.
The DPF model is based on MTU’s adaptation of Prof K’s
model, 2003-01-0841. Input data from
WAVE through Simulink.
There will also be a model
for DPNR – DPF plus LNT models. For now
the kinetics are separate but will be integrated later.
The component models can be
combined into a system model. Practical
issues include selection of a reasonable time step, and synchronizing the
various models. Computational demands
are also an issue.
3 SS and 2 transient cases
were run to test the model. Results are
preliminary but look reasonable. Work is
continuing.
How long does it take to run
350 sec? 25 hours
on 1 Pentium 4 3 GHz PC. The bottleneck is the neural net. There
are several ideas to speed it up.
The paradigm is that
suppliers will provide a map of component behavior. An example is a turbocharger, where
performance maps are provided that manufacturers can use for performance
simulations. We would like a similar
sort of map for catalyst/DPF function. The map must include the necessary input
values over the useful range. One does
not need to know details of the catalyst if you have an adequate model.
For aftertreatment models in
system controls and strategy development we use 1D and 2D models. For real time
control in the product we use 0D models.
3D typically requires more computation time and more detailed knowledge;
we typically rely on universities or suppliers for those models.
National Labs and
universities can assist in developing more physically based maps, and detailed
micro kinetics. Suppliers can map and
provide standardized format data.
Each catalyst has its own
unique map. Further, each kind of
catalyst (DOC, DPF, LNT, SCR…) has its own map criteria.
An example of what is wanted
for a DPF:
The map must
This maybe a task for the
A map might include
The LNT team has made
progress on maps; can the DPF team do so too?
Prof. K: In systems that are
path dependents, you can’t describe by a simple map like a turbo map. You have to have an intrinsically dynamic
model, not a static map. Perhaps what is
meant is a minimum set of performance data, from which true dynamic models can
be calibrated. It might better be called
a protocol rather than a map.
We don’t have reliable
handbook values of things like permeability.
Further, since models are not standardized, the same permeability in
different models will give different results.
Prof K: you need to get more
of the basics nailed down so there is less to be calibrated from data. As you do so, you will need less and less
calibration data.
You need to develop a very
detailed test protocol. The whole
thermodynamics is involved: eat and mass transfer, chemistry,
…
George Muntean (PNNL): the
“map” is intended to be a simplification that lets you bypass detail
consideration of some variables. For
instance, a turbo could be calculated from geometry – but not easily. At this point, we haven’t even decided what
we want in a DPF model. Is it just
pressure-flow? Do you need to know
detailed soot cake characteristic? How
do you short cut the data process to make it economic and timely? Because DPFs have a history effect things get
more complex. Perhaps the map should be
rate of change rather than current value based.
The discussion is still very open.
Louissa Olsson, Chalmers
With GM R&D
Model is a storage and
shrinking core type model. There are
rates for BaCO3 to Ba(NO3)2 and vice versa. The model includes recognition that some Ba
sites never participate.
Reaction set: (see slides for
better detail)
Storage
BaCO3 + NO2
BaCO3 + NO – forms nitrites
Regeneration
BaNO3 + c3h6
NOx + CO
PGM:
NO + O2 č NO2
c3h6 + O2 č CO2 + H2O
NO + c3h6 č N2, CO2, H2O
Experiment sequence:
NO to NO2
over Pt/alumina. T ramp 25-500C, 8% O2 and 600 ppm NO
NOx storage
on Pt/Rh/Ba/alumina – 320C with long lean period. 8% O2 (0 rich), 900 ppm c3h6, varying [NO].
The NO oxidation rate
kinetics were calculated from the above data. As expected, above 350C or
so thermodynamics defines concentrations of NO and NO2. Below that, by the reaction
kinetics. The resulting model matches data well.
When Ba is present, the
ability of Pt to oxidize NO is reduced.
Similarly, the model matches
the experiment data well for the NOx storage, including the NOx spike on
regeneration. The experiment was repeated at different temperatures and [NO]. The model predicts barium nitrite and nitrate
in agreement with FTIR data.
The model was modified to
include the inactive Ba core because our experiment did not use all the Ba. Fit was
good.
The model was validated
against another data set with 4% O2 rather than 8%. Good prediction.
Summary:
A global LNT map of a model
catalyst was developed and matches experimental data. NO oxidation was investigated with a
Pt/alumina catalyst.
How important is the NO
storage in comparison to NO2? Only important initially,
when you don’t get much NO to NO2 oxidation at low temperature. It also happens at the start of the
adsorption.
The Ba nitrite should be
unstable >260C. Why do you see so
much? We should see more at low
temperature. In real catalysts, you may
oxidize the nitrite to nitrate.
There was neither water nor
CO2 in this test data. It has been done,
but is not released at this time.
There is work indicating a
diffusion limitation of barium carbonate to nitrate – there is a molar volume
changes so there is a diffusion limitation. We did not look into that.
What determines the break
trough peak? It gets larger at higher
temperature. It gets smaller with lower
O2. The model does include
exothermic heating.
A recent paper discussed the
stability of barium nitrate in the presence of CO2. You can’t get the nitrate made without being
on a surface. Shouldn’t the pure
compound limitations be relevant when you get into the bulk of Ba? I would
not think delta-G should be positive at these temperatures for nitration of Ba
in the presence of CO2. The surface
of the Ba will still form nitrate. The
nitrate is not bulk though.
Katey Lenox, ORNL
Regeneration strategies were
investigated for 3 fuels, 2 strategies, one MECA
catalyst at 300C. Extensive
instrumentation was used to characterize the operation. A paper will be presented at the SAE Fall
meeting.
Mercedes
1.7L CIDI Common Rail engine with full control. SpaciMS used with 6 capillaries. . TEOM, Celesco, LII,
and AVL smoke meter were used to look at PM.
The strategies:
1.
Delayed extended
main (DEM) injection
2.
Post injection
3.
Added EGR to get
LTC before regeneration.
A DOC and LNT are used (no
DPF). Exhaust was analyzed at engine
out, after DOC, and after LNT.
FTIR concentrations (L/R
cycle average) of HC cracking products (methane etc) drop with later delayed
main timing, but not with porst injection.
Post injection gives higher
total HC and a larger range of HC species than delayed extended main.
PM increases quite a bit with
DEM strategy but not with post injection.
BP15 and ECD-1 are likely
commercial fuels. DECSE is a research
fuel – chosen because it is different!
DEM makes more CO/H2 and lower total HC with all 3 fuels. DEM also gets better LNT catalyst conversion
(60 versus 45%). The operating
conditions were chosen to get ~50% NOx conversion so differences would be more
apparent; other timings could get much higher conversion.
The relative significance of
CO/H2 varies with temperature: toluene works better at high temperature, H2 at
low temperature.
With DEM, H2 peak at DOC out
is about the same as engine out. H2
drops very early in the LNT, and is gone by half way through the catalyst. The cycle is 57/3 L/R. CO is present still at the catalyst outlet.
A deSOx was performed. DeSOx
was done by heating to a high temperature (700C) by engine control strategy
then going rich 6-8 seconds. Temperature
rises, so you go back lean to control temperature. Total rich time was 6-8 minutes. There was probably a lot of soot but htat wasn’t measured – juts trying to get sulfur out
without destroying the catalyst.
With the sulfated catalyst,
L/R cycle average NOx conversion improves as you make the rich lambda richer,
but you reach a plateau where going richer no longer helped. This seems to be a mass transfer limit in the
deNOx regeneration.
Next,
looked at regeneration in a LTC operation mode. They started
lean, then transitioned to LTC by adjusting injection
timing and EGR. Then you put in a rich
pulse, and then go back to lean. The
regeneration is very similar to the DEM strategy; you still get soot during the
rich pulse. LTC does give low NOx.
Plans:
Detailed questions should be
directed to
300C was the LNT inlet
temperature. Didn’t the Temp change
with strategy? They did control it; details in the SAEpaper.
We think the DOC was not in
very good condition.
We are working on a better
calibration and interpretation of the UEGO rich lambda data. The values reported are “indicated” from the
UEGO.
The emission data from SpaciMS should have 5-10 Hz time resolution. Why doesn’t the O2 go to zero at the LNT
outlet? If you are really rich, there
seems to be too much O2 for the CO and HC that made it out. It might also have been O2 storage.
Stuart Daw, ORNL
The second day started with a
discussion that included several
The current proposal is a
test protocol that would give enough information on a degreened catalyst to
permit extraction of kinetic constants to be used in an LNT model. Note that sulfation and aging effects are not
included at this time; they will come later.
A subgroup has been leading
the protocol development:
The protocol includes 72 runs
including cleaning steps, and is intended to take less than 2 weeks to run (on
an automated stand). Simulated gas is used
including water, CO2, O2, NO and NO2.
“Fast” and “Slow” cycles are used to get short and long time kinetic
data. Degreening procedure is defined by
the catalyst supplier. A difficulty is
that some suppliers consider the degreening process to be confidential.
Three temperature ranges
might be used depending on the intended application
The protocol tests five
temperature spread across the applicable range.
Regeneration is done with two
levels of reductant for the fast cycle, and one level for the long cycle. The reductant can be H2/CO and/or liquid:
diesel (toluene/dodecane)/gasoline (toluene,isooctane). There are still issues about how to do
the liquid reductant tests: mixing, vaporization etc. It is intended to have fast L/R and R/L
transitions – that is also a hard problem.
20 mil thermocouples are used
at entrance, center and exit of the brick.
Exit gas composition measurement includes NO, NO2, H2, CO, O2, N2O, HCN,
NH3, total HC.
Sampling rate is 2-5 Hz for the fast cycles, 1 Hz for the slow
cycles. A blank reactor test is to be
included so you can characterize the instrument dynamic response.
There is a lot of variation
between different labs in terms of instrumentation and time resolution. It may be necessary to compromise on sampling
rates.
### offer
time resolved data to CLEERS.
Remaining Protocol questions:
To get good kinetic data at
high SV you need fast instrument response.
You can get into mass transfer issues also, that don’t show up at lower
SV.
We also need to develop how
the data will be organized and presented so it is useful and accessible.
ORNL plans to run the
method. They may use a production GDI catalyst (VW GDI from
Umicore?) since it is available without proprietary issues. There are also some model catalysts to be
tested. The first trial should be run in August.
Stuart Daw said that the data
in this protocol, combined with previous experience and modeling know how, allow you to model a catalyst. Without the prior knowledge, this would not
be enough data.
Bill Pitz, LLNL
There are a lot of basic
questions to be answered – how are nitrites and nitrates attached to the
surface? How does the carbonate affect
it? Ab
Initio molecular dynamics are used to examine
these kinds of issues. We can predict
binding energies and what the species geometry is.
Car-Parrinello
molecular dynamics is used on the LLNL MCR massively parallel computer, up to
2000 processors.
Cluster calculations (18
atoms) and surface calculations (infinite slab) are used. We find surface calculations to be faster,
but perhaps clusters are more interesting to represent things like steps on
particles.
Others have done some NO2 on BaO calculations. We
wanted to see if we get the same answer. We got the same result as Broqvist (2004). We
find ionic bonding of the NO2. N’s sit ab ove O’s.
We also did a surface
calculation with 4 layers of 16 atoms each.
Boundary conditions make it appear infinite. This calculation did not show
an electron moving from the surface – i.e., not ionic bonding. Again we see the N’s over O’s. This is very recent work, not really finished
yet.
It is not obvious which
calculation is more likely to be relevant.
The calculations take 10’s of hours on this very fast computer.
We also looked at a barium
carbonate surface. Peden and Muntean are
doing related experiments
at PNNL. The surface looks
like CO3 laying flat above the Ba plane.
The next step will be to put NOx on this surface. BaCO3 and BaO
surfaces are very different; it is amazing they can transform back and forth!
In another case, a BaO surface was loaded with NO2 and the surface taken to
300K (most literature is at 0K). The NOx
tends to bend over from the N-down initial position.
Summary: first time we know
that a BaCO3 surface is calculated, or that BaO-NO2 done at non-0 temperature.
Plans:
Will try to get the movies
available on a web site; not yet approved to release.
Have you been able to
determine the interaction of the nitrites on the surface? To some extent – we can look at electron
interactions and so on. We can compare
to FTIR data, but haven’t done so yet.
Will you be looking at
temperature effects? We hope to but not yet.
Will you later throw in some
Pt? Yes, eventually. We have the tool
and we are starting with relatively easy things. It should be fairly easy to put in a Pt atom.
It would be interesting to
look at bonding as a function of the number of NO2s. Yes; some papers show a
configuration of N-down and N-up pairs; we see that at low temperature but not
at 300K.
There are probably cluster
size effects. Yes; we have not looked in to that yet. It matters where the NO2 sits on the cluster
– step, edge, etc.
Water is so important; can
you look at how water influences the environment? All these oxides at hydroxylated,
for instance. It shouldn’t be that hard for us to do so; we should try it.
The objective is to
understand what is happening on the LNT surface. Model catalysts were studied 150-400C to
identify key reaction steps, with water and CO2 present. Regeneration effects and thermal deactivation
effects have also been studied.
The focus has been on K based
LNT: Pt/K/alumina. We have also started
looking at Ba based catalysts.
DRIFTS has been used as well as physisorption
and chemisorption, and MS to look at effluents.
DRIFTS is in-situ up to 500C with up to 10% water.
NO3- and CO2-
dominate on Pt/K/alumina. When water is
present, we do not see nitrate storage on the alumina as we did dry. Water favors nitrate formation in favor of
carbonate.
The saturated stored NOx
species is the same at all temperatures, but the amount stored changes with
temperature.
If you look at smaller time
scale where NOx is not saturated, we do not see NO
adsorption on K/alumina. Nitrites
observed when Pt is poor oxidizer of NO; these are eventually converted to
nitrates at higher saturation. NO only adsorbs
near Pt sites, but only at low temperature (at higher temperature Pt converts
NO to NO2 for adsorption.
NO2 adsorption: With NO and
O2, K/alumina does not form nitrites – everything storing is a nitrate. One route is NO č NO2 on Pt and spill over to K. Alternatively, gaseous NO2 can form nitrates
readily. There can be NO2 adsorption on
Ba sites through disproportionation but this is not shown in experimental data
(yet!).
NO2 never adsorbs as a
nitrite, only as nitrate.
Water is important to regeneration. Starting with a saturated Pt/K/alumina trap,
water inhibits the regeneration especially <200C. Water inhibits nitrate reduction. Without water, 20 seconds removes all
nitrate. With water,
only 21% removal after 20 minutes.
We need to start looking at
the species evolution to understand regeneration. That experiment is beginning.
Thermal deactivation has been
studied on K and Ba base catalysts. Key
mechanisms are sintering of Pt and Adsorbates; noble metal masking by
Adsorbate. The deSOx cycle on ORNL’s
dyno got to 780C or so.
A reactor allows flow through
or chemisorption experiments; they can measure surface areas in situ. Ran 15 hr at various
temperatures with cycling lambda (9 min lean, 1 min rich). Water and CO2 present at all times. NOx storage at 250C was checked after high
temperature exposure. H2/O2 titration
was used to find Pt surface area, and BET for total surface area.
K based: They found surface
area was affected above 850C; this is alumina phase transition. Pt size steadily
increases, with a leap at 900C.
Effective NOx capacity decreases steadily – not directly linked to
either surface area or Pt area. After
900C aging, ˝ of storage capacity is gone.
Ba based: Larger drop in
surface area. Pt size increase smaller
with only a minor decrease between 760 and 900. This suggests Ba is covering
some Pt. Pt area decreases more than Pt
size would suggest. Storage has an
initial decrease, then relatively stable until it dies at 900C.
Storage rates were measured
after the aging. The initial rate is not
dramatically lower after 900C than after lower temperature aging.
Summary:
Plans:
Ran NO2 adsorption at room
temperature; it takes a long time, more than 30 minutes. Usually I ran over night to be sure it is
fully loaded. That is when you see the
alumina effect without water; with water the process is much faster.
Have you tried separating
Pt/alumina and Ba/alumina and varying the particle size to change
“intimacy”? Chalmers tried physical
mixtures of Pt and Ba – didn’t work. You
seem to need them very close together.
LNT models can be based on
elementary kinetics – but it is difficult especially if you can’t look at the
surface composition in detail on a proprietary catalyst. Or, you can do global kinetics but there is a
risk t is less broadly applicable. We
tried a global model. We can include
mass transfer, bulk versus surface diffusion, reaction time delays, etc.
We want to predict NO and NO2
adsorption based on experimental observations and known phenomenology.
Key observations:
The global model formulation
is shown on a slide (I can’t type that fast!).
4 reversible reactions (8 rate parameters) are used for NOx-Ba
interaction. Two steps are needed to
retain the basic features: a fast step and a slower step.
The fast reaction must
involve release of NO, since experimentally we see NO immediately on NO2
exposure.
The spill over mechanism is
still quite fuzzy. We assume the sites right next to Pt are nitrated very quickly. Once those sites are filled, species have to
migrate on the surface. So you get a
fast initial adsorption followed by a slower adsorption rate. Wee assume an exponential drop in rate with
coverage. We think spill over does not
happen below 250C.
We model NO < = > NO2
over Pt. the
rate of this reaction gradually decreases with time; based on the Chalmers
(Olsson, 2004) data we think this is oxidation of the Pt over time.
We assume disproportionation
and spill over are independent. Pt
passivation is by NO2 + Pt č PtO + NO. O2 gas
concentration is not calculated – there is lots of O2 around. The model is constrained to measured
isotherm. Equilibrium includes material
heterogeneity and complexity.
The sorbent shifts NO2/NO
equilibrium toward NO. The NOx storage
isotherm is measured.
The model is compared to some
data sets with good matching.
Evidence of the Pt oxidation
is that [NO] has a peak in time when the input NOx is mainly NO2. You quickly get some Pt passivation.
The passivation behavior is
important. The Pt at the front of the
LNT may get close to saturation even in 30-60 seconds. Experimentally, you have to saturate the
whole catalyst to measure the kinetic effect.
NO to NO2 conversion cannot
be modeled by a simple first order reaction rate. A slide shows relative activity versus time
for lean mix with either NO or NO2 input.
Summary:
Plans:
There is obviously something
that slows down the reaction – shrinking core, bulk diffusion, Pt passivation, …? The
model has a deactivation factor – any of these mechanisms might be what causes
the need for this term. The passivation
is a function of time, not degree of saturation.
Another possible reason for
the slow down is that there can be many different sorts of Ba sites – perhaps
the easier sites fill first and then it gats harder to store.
We find the Pt deactivation
faster is there is Ba than if you have only Pt/alumina. O2 oxidation of Pt is much slower than with
NO2, we think.
We think the CLEERS
protocol will generate enough data to calibrate this model.
Navon Khadiya, ArvinMeritor
H2 and CO are very good
reductants for LNTs, both deNOx and deSOx.
The plasma fuel reformer is a generator of H2 and CO. Preliminary results of deSOx will be shown.
The device uses electrical
power to generate a plasma to do the main reforming; a
catalyst is used to maximize [H2]. A
fuel rich mixture is treated to get ideally 24% H2 and 26% CO. The actual hardware (without catalyst)
processes diesel fuel into 6-8% H2 with very little soot. With catalyst, [H2] increases to 23%, still
very low soot.
The plasma unit can start
very fast. There is about an 8 second
rise time in measured [H2] (catalyst plus plasma).
You can rapidly cycle on and
off. Data shown for
3000 cycles; the same as fresh [H2] response.
Engine tests were run to test
LNT deSOx benefits of reformate. “Low
temperature” LNT used in dual leg configuration. The catalyst can reportedly be deSOxed at
550C with normal means. This system
delivered reformate while the LNT was of line.
This was compared to the same system with diesel fuel injected in the
“dead” leg for regeneration.
The catalysts were sulfated
at 260C with 35 ppm SO2. Performance
measured on ESC Mode 7 before and after sulfation, and again after deSOx.
100% recovery of activity was
obtained after deSOx at 510C with diesel in 6.3 minutes. At 450C, diesel gave only 8% recovery. With reformate, 400C gave full recovery in
6.3 minutes. At 350c,
100% in 7.3 minutes.
### add a reformer to the
LNT/SCR cost picture
Clearly, reformate is able to
deSOx at temperatures more than 100C lower than diesel fuel.
There is significant work on
S effect over TWCs, including H2S formation. By
comparison, there is little modeling of LNT sulfur effects.
A simple global model was
proposed here. When H2 and CO are
present, there are added deSOx pathways: H2S, COS. Pt catalyzes the H2 + SO2 č H2S reaction.
Pt also thought to influence the decomposition of sulfated adsorbents.
H2/CO may permit deNOx as low
as 100-120C on Pt and Pd.
Summary:
The performance of the
catalyst in the reformer deSOx tests was poorer than in the diesel deSOx. The
LNT had been damaged by diesel deSOx before the reformer tests.
The deSOx temperatures
listed are catalyst inlet. Bed
temperature was not measured (they are doing so now). Outlet temperatures were 100-140c higher than
inlet in all cases. There was some
leakage (1-3%) past the shut off valve. For diesel fuel, regeneration lambda
was ~0.7; the reformer was somewhat richer. If the bed temperatures aren’t
lower, you will still get damage! The
outlet temperature was not so high as to indicate catalyst damaging bed
temperatures.
There is some good sulfation
data on LNTs in the SAE literature. That
is probably more relevant than the TWC kinetics.
Do you see PtS being made? No;
we see it from literature but not in our data.
Many millions of dollars have
been spent on research…. and we don’t really know much yet!
There are a number of
mechanistic studies. See slides! Organo-nitrogen species are involved on some
catalysts. On others, inorganic nitrogen
species are important. With enough patience, you can find almost any species
you want – but are they relevant? What
are the simplest possible explanations that cover the data?
Factors:
The conclusions you reach
depend on the experiment you do! For
instance, propene adsorbs strongly on Pt.
Pt is in a reduced state all the time.
Propane does not adsorb on Pt; Pt is oxidized, and NO goes to NO2 rather
then alumina. The NO2 can move to
another site and get reduced there. The
mechanism is quite different even on the same catalyst. Octane acts a lot like propene, juts at a
higher temperature. N2O formation is an
issue.
You can get this under lean
conditions. A temporal analysis of
products (TAP) reactor was used. It has
good transient time resolution while analyzing the surface. A pulse of NO can decompose, but subsequent
pulses are less decomposed; Pt gets oxidized.
You need a reductant to keep the catalyst functioning.
Propene is effective in
reducing the Pt. On the other hand,
oxygen completely kills the CO/NO reaction.
A Japanese group looked at
single crystals, and reduce CO/NO or CO/O2 at 400C. In either case, CO2 comes from COads + Oads. This is evidence for NO dissociation.
Evidence
for organo-nitrate species at lower temperature.
Good FTIR evidence for cyanates etc.
However, this species is changing over very long times (100s of
minutes).
On another catalyst, -CN seen
on the order of minutes. But, water was
not present and there was no CO2 in the inlet gas. Isocyanates react very rapidly with water. With water, you see HNCO and NH3 in the gas
phase.
It is hard to resolve which
species are really the important reaction intermediates.
Real world effects
Exhaust temperatures in
driving vary. Low temperature is
important. Here, H2 is a good
reductant. But, you get mostly N2O.
NO decomp
was seen over Pt at 100C in the TAP.
But, you need a reductant to keep the Pt from oxidizing. You never see N2O, unlike the steady state
experiments. If you add H2, you see N2
formation and not N2O initially, but N2O appears after several pulses as the
oxygen coverage increases.
Conclusions: you can
decompose NO on a reduced Pt surface.
You do not form N2O on a reduced Pt surface.
SSITKA experiments were run
to look into mechanisms. Here, an
isotopic tracer gas is used, and the transient is isotopic rather than
concentration.
These show that N2 is formed
from adsorbed + gas phase NO. On the
other hand, N2O is almost all from gas phase NO. This suggests different mechanisms. Looking at how the concentrations change in
time can suggest the number of intermediates that exist.
This method suggests an (NO)2 dimer intermediate.
Density function theory of
the formation of N2O on Pt
Can the modeling help explain
the data?
The most favored N2O
mechanism still has a barrier of 1.78 eV; defects etc may lower to <1. On the other hand the dimer mechanism at a
step has a barrier of only 23 kCal/mole. The dimer mechanism becomes dominant on
partially oxidized surface.
Adding H2 to a propane-NO mix improves both peak NOx conversion and lower
temperature activity. H2 is also good in
combination with octane. H2 also
broadens the temperature window where HC is partially oxidized. The NOx conversion light off is about the
same for H2 with a wide range of HC reductants.
H2 also enhances formation of
NO2 – even going above equilibrium at 800C.
This is surprising since NO2 poisons the Ag NO activity. One theory is that H2 encourages formation of
Ag clusters. There is some controversy:
is it Ag-aluminate, Ag+ ions, clusters??? UV evidence for clusters is not conclusive.
Replacing H2 with CO causes
NO activity to slowly decline at low temperature. Activity recovers immediately when H2 comes
back on. We also ran H2/CO mix. It seems like CO does nothing at low
temperature (not inhibiting, just not working).
Another issue is that H2 may
decrease nitrate inhibition.
It is pretty clear you don’t
want metallic Ag; you want to prepare ionic Ag on the surface.
Ag catalyst is interesting:
low temperature conversion, no N2O.
Catalyst scientists are
falling into the surface science trap: the right experiment is very hard, so
they do the experiment they can do to get it published. We need to go after a real problem.
Is H2 also that important on
other catalysts? We find Ag unique in that.
We have tried Cu, zeolites, etc.
Ag is not easy to oxidize; the oxide decomposes above 150-200C or
so. You have to form something like a
nitrate to activate the HC. Once you
activate it, it can stick to the surface where it will eventually react with
NOx.
Is there an optimal H2/HC
ratio? There seems to be some storage effect.
There won’t be any
publications soon, and we won’t show any new materials… because we can’t show
it! We believe that there are better
HC-SCR catalysts than people have found so far.
We are looking for them.
Partners are Engelhard, GM,
and Accelerys (software).
Combinatorial methods are
used to look for new materials. Parallel
synthesis and multiple sample reactors are used along with informatics. Hundreds of samples per month are evaluated. Engelhard makes the supported catalysts. GM does the lab reactor testing. Data is put
into a database to allow searching and analysis. Engine and vehicle testing are in the
future. HC-SCR is the target, not
decomposition or urea.
If a product is developed,
Engelhard will be able to sell to anyone.
Over 3000 materials
evaluated; something around 10% have been interesting
enough to look at further.
Future combustion regimes
such as HCCI may permit us to be happy with 40-50% NOx conversion, and this
seems possible with HC-SCR. Low
temperature activity improvement is needed; perhaps both low and high temperature
materials would be used.
Engelhard makes a range of
materials and runs the combinatorial testing.
Promising materials are evaluated in more detail at GM, followed by
another down select. The information also feeds back into the discovery phase of
the program.
GM’s reactor evaluations use
diesel fuel.
Criteria for selection of
interesting candidates is peak conversion >70-80% (have moved to at least
150%) of the benchmark standard catalyst, a wider temperature window, or
unexpected positive performance.
Disqualifiers are N2O, HCN, NH3, excessive adsorption or poisoning.
All compositions are binary,
tertiary or higher.
All samples are tested
degreened and lightly aged. On the GM reactor a temperature sweep is run. At the peak conversion temperature, the
NO/NO2 ratio is swept. Several different
reductants are tested. Another
temperature sweep is then done at optimal NOx ratio and reductant conditions. Sulfur effects are next if still interesting.
We have correlated the
Engelhard bench to the GM bench and to the engine.
Informatics is important.
Instead of sparse data, you have a ton of data – composition, synthesis, and
analysis info. You have to extract the
right info to decide about scale up and further testing. The database is set up with secure
connections so all participants have secure access. You have to have tools to get your instrument
data into the data base.
Trend analysis is a part of
the learning. This is not mechanistic learning but rather statistical
analysis. What preparation and
composition steps gave good performance?
Can we predict a better one, then make it and try.
Summary:
Do you think DOE action is
what lead to this?
Would it have happened without them? Yes. The start up cost of informatics and so on
would have kept people from getting it started.
You probably could
buy the software now; Accelerys owns
the software and is allowed to market it.
National Labs get a certain number of paid licenses.
Have you tried genetic
techniques for the trend analysis? Yes. Initial test data sparsely populates a large
data space. With that data you can do
genetic or neural routines to search for interesting regions.
The original LEP CRADA for
this work started back in 1992. This
work is nearly the last of that project, although there is still some fuel
funding to continue part of this work.
SNL and LANL were also
involved in this CRADA. The goal was to
develop and characterize improved urea SCR catalysts. ORNL tested full scale
catalysts on the engine.
Urea decomposition begins at
80C and melting at 135C; there is no gas phase urea. A chemical isomer is ammonium
isocyanates. For aftertreatment we use
32.5% in water because that is the minimum freezing point. Decomposition is endothermic.
What might happen to urea in
hot exhaust? A slide shows several
possible paths. You might get biuret, cyanic acid, ammelide, ammeline, cyanuric acid,
melamine, cyanamide, dicyanimide,
or ammonium nitrate. Cyanuric acid and
melamine make an H bonded solid that we have observed in exhaust at higher
temperatures. We have seen all of these
compounds at some level at some time.
Analysis is by FTIR, photoacoustic analyzer for NH3 and N2O, and filter
analyses. A two stage filter gets PM on
the first, while a second stage can gather some gas phase components. Impingers were also used.
For light duty works, around
250C, we don’t see the above compounds.
The exhaust system has a DOC plus urea SCR. Urea is injected after the DOC. The urea injection is by air atomizing
nozzle. The atomizing air was humidified
before use to improve metering and measuring accuracy.
The engine was operated
without EGR. Temperature was varied by
changing load, so NOx input rises with temperature.
We wanted to measure urea
decomposition before the catalyst. The
initial work used a 6” SCR catalyst brick; this was later changed to 3” to
reduce storage time delay effects.
It is very hard to measure
ammonia upstream of the SCR. We wanted
to see if the urea had decomposed. The
trouble is that if you collect any urea on the sample filter, then you see
ammonia even at very low filter temperature.
If you run the filter cooler, you get condensation. It would be better to use an in situ ammonia
measurement but we don’t have one that works well.
You expect to see urea
decompose to ammonia and isocyanic acid, and the acid hydrolyzes to another
ammonia and a CO2.
You would also like the NOx
going in to the SCR to be 50% NO2. The
DOC tended to convert NO more to NO2 at higher temperature, and less than 50%
at low temperature.
We got 65% NOx conversion at
262C, dropping to ~20% at 155C – although at the lowest temperature it may be
storage or conversion to ammonium nitrate rather than real conversion.
Nitrate is as much as 20% of
the total NOx at 210C. It is all
particle phase, probably ammonium nitrate or nitric acid. This is not made over the DOC. HONO was very low after the SCR.
There was significant
reductant storage. You can see this in
that turning off the urea and increasing the load (raises NOx and temperature)
causes large ammonia desorption. NO
conversion continues for a while. This
could be a problem in a vehicle system.
Another slide showed 165C
accumulation followed by a burn off at 190C.
There is a complex result of NO, NO2, and NH3 concentration. We think isocyanic acid is stored in the
catalyst. As temperature increases, you
get SCR on the surface and ammonium nitrate decomposing into NH3 and HNO3. At higher temperature NH3 SCR stops as you
exhaust the storage, and NO2 increases faster than NO.
Modeling needs to include all
these reactions and storage effects. It
seems hard to incorporate into a real time model. At low temperatures, you can assume no gas
phase urea decomposition. (current models assume you
are getting NH3 in the inlet!) We need
to understand the nitrate formation, storage, release, reactions. How does rapid warmup (sudden acceleration)
affect this?
Plans:
Questions:
Have you thought of using the
SpaciMS? NH3 is right next to water so it is hard
to measure in MS. Also, NH3 tends to
plug the capillaries.
At high temperature, the
SCR catalyst works well but you make the various side products (the downstream
clean up cat gets rid of them). At low
temperature, we don’t see the same products but we do see the nitrates.
Urea distribution is
probably important. In our low
temperature case, think of it as spray drying – the urea probably arrives as a
fine dried powder
At the low temperature,
even the 3” brick took ~2 hours to stabilize.
Bob Weber, TIAX
Or, microkinetrics of lean NOx catalysis.
You have to be careful not to
over simplify the models. Global models
are appropriate for steady state chemical plants, but vehicles are virtually
never steady state. Global models will
capture lab data but may not be so useful for real systems.
How complex a model do you
need to capture the behavior of HC-SCR catalysts? How do you construct the model in a way that
is extensible for future enhancements?
How do you get enough data to calibrate and validate the model?
To build a simple model with
as few parameters as possible, we start by ignoring S and aging effects. The
model includes oxidation of
A software package has been
built that allows you to build the models. You select reactions and define the
reactor. We can use more than one
experiment for parameter fitting.
A syngas SCR model over Ag
catalyst was presented as a simple case.
We find that the surface is largely covered by oxygen; other sites are
at the ppm levels, making them hard to see in IR studies.
The butane model an be added. The
previous parameters were not changed when butane was added. Experience shows that several of the
reactions listed are in fact insignificant.
You could next add a decane
model. This uses a joining with the CH4
network.
Without having calibrated the
model to any data set, you get a predicted NOx conversion with peak temperature
and characteristics that are reasonable.
There are certainly enough parameters you could use to fit the data, but
mostly you don’t need to since standard values give reasonable answers. In
fact, you only need to calibrate a few large-effect parameters. Calorimetry and
other supporting experiments can be used for calibration too. Quantum Chemistry may help.
DFT was combined with this
modeling approach to ask what would happen is you have a few atoms of S under
the surface of Pt(111) you decrease the heat of adsorption of oxygen near the
sulfur, moving TPD to lower temperatures.
Models of this sort are an important
part of the discovery-testing-using cycle.
Model reduction is necessary if you are going to put the model into your
powertrain controller in real time.
You want to choose models
that have “optimal sloppiness”!
What you are doing sounds a
lot like combustion codes. It turned out
there were a lot of rates we don’t know.
How do you fill your database with meaningful rates? Easy; hire us! Parameters come from chemical intuition of
bond strength, transition state theory, and thermodynamic consistency. The
hardest part is a plausible reaction network, but the literature is full of
good suggestions.
We would generally suggest
using a reduced model to couple to a CFD code rather than both detailed models
at once.
If your
end result is a neural net model of a model, why not just use the neural net in
the first place? You could do that. But, you are limited to what you
measured. With a more physical model an
extrapolation is more rational. Thus, we
can tune the neural net beyond measured data.
Also, you can use literature data for some inputs.
Christine Lambert, Ford
The program started 3 years
ago, and has one year left. It is part
of the DOE Ultra Clean Fuels program. We
target 2007 Tier 2 Bin 5 standards, 0.07 NOx and .1 gm/mi PM. ExxonMobil and FEV are subcontractors.
An F250 diesel truck is the
test bed. Durability testing is
scheduled to go 120K equivalent miles, starting July 2004.
To hit Bin 5 we need >90%
NOx conversion. We are using CDPF +
urea-SCR.
Models have been used to
predict system performance, NOx, FE, and direction for vehicle system design.
Models are also used to help diagnose problems.
An example is the demonstration that you need a fast warmup strategy to
meet FTP NOx emissions.
The vehicle exhaust has a
small upstream DOC, another DOC where it fits, then the SCR, then CDPF. Urea is added after the DOCs. The SCR is 2X engine displacement volume.
the vehicle took about 150 seconds to get to 150C with a
base calibration. With a fast warmup
strategy we would expect to meet Bin 8 NOx level.
Urea mixing is
important. We use a spray deflector to
get acceptable mixing in the length available.
you have about 12” to mix in a 4” diameter
pipe.
NO2/NO ratio is
important. You have to pick the right
DOC to get about the right ratio over the temperature range.
SCR catalysts are also good
HC adsorbers. When you adsorb HC you
lose NOx activity. Heating to 500C for
10 minutes regains NOx conversion.
ExxonMobil is working on a
demonstration of a urea co-fueling system.
A vehicle and fuel dispenser have been modified
and are being tested, including cold conditions. Durability and reliability testing are
underway. Some antifreezes
have been tried, but so far all are HC based and poison the SCR. Instead, electric heating is required for
very cold weather.
Plans:
What aging cycle will you
use? What is your control strategy? We have control strategies for both urea
injection and DPF regeneration during the driving cycles. We stay lean at all times. Sulfur and HC can be removed from the SCR
with an active heating, lean strategy.
Is Ford developing the
sensors? NGK NOx sensors; we are looking at several potential NH3 sensors.
Anything
special for cold starts? Yes,
rapid warmup strategy. We reduced thermal mass, increased exhaust temperature
by injection and EGR calibration. An
ensemble of many items is needed to get enough heat.
Have you experienced issue
like Storey mentioned? If we get the SCR warm enough fast enough we should
not have big issues. Do you get
repeatable data on FTPs? The FTP data is better if
you have some NH3 stored in it. We do
track the NH3 storage in the strategy.
Do you have a cost
target? Yes, $4 per kW – a
challenging target! We see a path to get
clost to that target but we are not there yet.
Matt Thornton, NREL
FIRE is AVL’s
CFD code. It is used for flow
distribution, can handle spray into exhaust, catalyst kinetics, A/F mixture
distribution. A fuel cell model is being developed.
Modeling issues:
Urea spray is modeled. Urea
decomposition was modeled with CHEMKIN in the pipe upstream of the SCR. The catalyst is modeled with global lumped
chemistry. The catalyst is modeled with
an Eley Riddel mechanism.
For the catalyst model, see
SAE 2003-01-0845.
Exhaust geometry is defined,
and input boundary conditions like exhaust mass flow and inlet
temperature. Catalyst substrate, cell
density etc are defined. Urea injector
design is specified along with urea injected quantity. Parameters specify washcoat activity.
This model is to be used to
support several DOE fuels research programs.
Three of the APBF-DEC programs are looking at LNT and one at SCR. The latter is an SWRI HD SCR/DPF project.
The SCR project is using a
Caterpillar C12 12L engine. early results have shown very low emissions. Zero hours hit 0.24 gm/hp-hr NOx, about equal
to the 2010 levels. Access to the engine
controls is limited. The system is being
aged, now at 200 of 6000 planned test hours.
We want to do a system validations and compare engine data using the CFD
code. This will set parameters. Then, we will use the model to look at system
optimization.
Summary:
David Bergeal,
JMI Royston
We do mainly global reaction
modeling. We also do CFD but generally
not together with chemistry.
Lab reactor
experiments and adsorption studies. We use this data to calibrate the model and
validate against engine data.
The model is 1D heat and mass
transport, surface reactions, convection and heat conduction.
The model was described in
SAE 2003-01-????.
The model
shown as a vanadium catalyst. The model uses NH3, not urea.
3 NH3 reactions are in the
model. In addition, 3 NH3 oxidation
reactions (2 going to NO) are included. NO oxidation is not observed on these catalysts.
The coverage term is given by
a Temkin isotherm.
The desorption energy changes with the NH3
coverage: it is less easy to adsorb more if you have high coverage.
NH3 adsorption was measured
and modeled. First, adsorb at about
200C, then ramp the temperature to about 350C and watch desorption. This gives
information on adsorption and desorption energy.
NH3 oxidation was measured
with NH3-O2 mixtures.
Next,
measure NO conversion; NO+NO2 conversion; NO2 conversion versus temperature. 100-200
experiments are needed for each new formulation.
Once you build up this data,
we calibrate the model and then compare to engine data.
Slides show effect of catalyst
volume on the HD 13 mode test. Also, model versus data for cold start.
The DOC can over-oxidize NO,
creating too large an NO2/NO ratio and causing a dip in NOx conversion versus
temperature. System optimization
suggests a smaller DOC to avoid this problem.
But, you have to trade low temperature performance and HC/CO control.
The model can also
investigate different urea injection strategies.
The model projections match
engine data pretty well. The model is
useful for predictions of catalyst volume, cell density, injection rate, test
cycles, DOC catalyst and its properties.
You model NH3 coverage. The surface coverage lags behind
considerably. Did you account for
this? Yes – but only algebraic, not
differential equation. It doesn’t
seem like that should work – you ought to need a differential equation. Somebody said they thought it did …. some confusion here!
The model is a single
channel catalyst model – we assume good flow distribution etc. You have to use the CFD to get flow distribution.
Why did you choose vanadium
catalyst? It is popular in
Chris Depcik,
U of
Work supported by Ford
(Baker, Lambert, Laing, Johnson), TACOM, and EPA.
Why a 1D model? 3D is much harder and runs much slower. 1D models are
faster, use calibrated global kinetics, and have easier input boundary
conditions.
A slide gave 1D catalyst
modeling history. There was some
discrepancy between the 1982 and 1992 equations, so he re-derived the equations
from the conservative Euler equations.
Constant density is assumed, allowing independent solution of the energy
equation. It also eliminates the sped of
sound from the equations. Constant
velocity is also assumed; eliminates the need for a momentum conservation
equation. The resulting formulation is
the same a Harmed in 1972. Some terms
are commonly eliminated.
For bulk gas temperature,
equations were also re-derived. People usually neglect the time derivative.
Surface species are gas
species at the surface, and also surface coverage of intermediates. Also re-derived.
These equations allow for all possible catalyst formulations.
Surface temperature is like
the literature with the addition of a volume heating terms and a pseudo-2D heat
transfer network.
A slide listed the model
terms and assumptions. The model runs at approximately real time on a PC.
This general model is applied
to a particular catalyst: different
chemical species, diffusion properties etc can be in a table. The different catalysts have different
reaction mechanisms. That reaction
scheme is a separate subroutine. It has
a graphical user interface.
It was applied to urea
SCR. A 3-reaction mechanism was
postulated. Arrenhius constants were fit
to some vehicle data.
The model and derivations are
available in his dissertation and on the web. Preprints of the published paper
can be made available.
1D is strictly useful only for
uniform flow distribution at front of monolith. You could divide the catalyst
into a few segments and solve the 1D model a few times.
There have been other
important modeling contributions between 1982 and 1994!
You seem to be driven more by
computational convenience than by valid computations! When temperature varies, density must change
too.
Dr. Christopher Depcik’s comments
John,
I was looking at your notes on
the web for my part in the CLEERS conference and I noticed that you included
the discussion after my presentation.
However, you did not include any rebuttal that I had to those questions...
For instance,
I mentioned at the beginning of
the slide that it was a brief synopsis of the history of catalyst
modeling. Of course there have been
important contributions between 1982 and 1994, but in terms of the general
model for catalyst that everyone tends to use, there was no significant change
to the traditional model equations.
Yes, I was driven by
computational convenience, however I mentioned in the conference that I have
done the full compressible flow simulations and it is not practical to do for
optimization over a full transient FTP cycle. Also, in order to get the model
equations that everyone uses for catalyst modeling (see Oh and Cavendish, 1982
for instance) from the governing equations of fluid flow (Euler), you have to
make similar assumptions that I did.
When looking at the comments
as-is, it appears that I did not do my homework in regards to catalyst
modeling, however that is anything but the case. I am in the process of publishing a complete
description of my work that should enlighten those who had comments on my
work. I don't know if you were able to
catch my rebuttal comments at the time, so I thought I'd expand on them here.
Sincerely,
Chris Depcik
Staff Technical Specialist
Ford Motor Company Research and Advanced
Phone (313) 594-1316
FAX (313) 594-2963
Email jhoard@ford.com