Vous êtes sur la page 1sur 10

Experimental and Modeling Study on Zeolite Catalysts for Diesel Engines

Pontikakis G. N., Koltsakis G. C., Stamatelos A. M.*


Noirot R., Agliany Y., Colas H., Versaevel Ph. and Bourgeois C. **

* University of Thessaly, 383 34 Volos, Greece, Tel: +30-421-74067, Fax: +30-421-74050, e-mail:
greg@antiopi.meng.auth.gr
** PSA Peugeot Citroen, 18 rue des Fauvelles, 92250 La Garenne Colombes, FRANCE

ABSTRACT
The work presented in this paper aims at detecting and understanding some critical behavior aspects
of zeolite containing diesel catalysts. An already available mathematical model for precious metal
catalysts was used as a starting point. A specially designed set of experiments provided the
information needed to improve some special modeling features. New sub-models were introduced to
account for hydrocarbon and H2O adsorption, as well as diffusion limitations in the zeolite. The effect
of flow maldistributions during real world operation is investigated experimentally and numerically.
Although a number of issues (especially regarding the de-NOx mechanisms) are not fully resolved,
significant progress was achieved as regards the understanding and prediction of diesel catalyst
operation.

1. INTRODUCTION
Diesel catalysts for gaseous emission control need very careful optimization to attain the desired
efficiency levels [1]. Although modeling can be quite helpful in this task, only a few relative efforts
have been presented in the literature so far [2, 3]. Aim of this work is to develop and experimentally
validate a mathematical model for zeolite based diesel catalysts, to be integrated in a currently
available CAE methodology for exhaust aftertreatment design.
Recently, it has been shown that relatively simple reaction schemes are sufficient to describe the light-
off and temperature window behavior of precious metal diesel catalysts [4]. The model presented in
[4] is the starting point for the modeling work described in this paper.
To develop an application oriented diesel catalyst model, we should focus on the phenomena that are
really important in real world operating conditions. After studying a series of measurements on SGB
and on a diesel catalyst equipped car driven on a driving cycle, we concluded that the following points
should be given special attention:
(a) Since the operation temperatures of the diesel catalyst are low and very near the light-off region
of CO, HC it is critical to have highly accurate oxidation reactions kinetics,
(b) For the same reason, it is very important to have a reliable prediction of the temperature levels
inside the catalyst. This implies a good prediction of heat losses, as well as prediction of the 3-D
temperature gradients in the converter.
(c) Simulation of “temperature window” phenomena for NO conversion requires good knowledge of
the reaction mechanisms and high accuracy of the respective rate expressions.
(d) The adsorption-desorption phenomena of HC, NOx complicate the reaction scheme. For these
phenomena, the published results are very few and no kinetic investigations appear in the related
literature. Hydrocarbon adsorption is important to reduce HC emissions at the first seconds of the
cycle. During this cold start phase, the catalyst temperatures favour also H 2O adsorption in the
zeolite. The latent heat released can strongly influence the thermal response of the catalyst and
thus affect the HC adsorption, which is very temperature sensitive.
(e) The usual practice in catalyst modeling is to discriminate between two HC categories (fast – and
slow burning). In the case of a zeolite containing catalyst, we should also take into account the
differences in the adsorption behavior between HC species present in exhaust. Therefore,
additional HC categories should be discriminated when simulating real driving cycles.
(f) The presence of zeolite imposes significant resistance for the diffusion of HC molecules in the
washcoat [5]. The observable effect is that for some HC species, the conversion efficiency cannot
reach 100% even at sufficiently high temperatures after light-off.
In this study, we aimed at clarifying the relative importance of the above remarks, in order to improve
our knowledge of the prevailing physical and chemical mechanisms. This, in turn, would allow us to
develop a reliable modeling approach, which could be used in the process of design optimization.

2. EXPERIMENTAL
2.1 Catalyst type
Two catalysts were tested in the frame of this work. The first catalyst (CATOX1) contained Pt (70
g/ft3) and zeolite (Alumina/zeolite = 80/20 % mass). The second catalyst (CATOX2) contained the
same amount of Pt but the ratio Alumina/zeolite was 70/30 % mass.
For the tests in the synthetic gas bench (see below), samples of 1” diameter and 3” length were used.
Ageing? (PSA).
In the driving cycle tests the catalyst (CATOX2) had a diameter of 5.6” and a length of 6’’. The
catalyst was aged for 20h at full load engine operation.

2.2 Adsorption test


A series of measurements in a Synthetic Gas Bench (SGB) was conducted with two catalyst
formulations to understand the adsorption – desorption phenomena in zeolite based catalysts and
determine the parameters to be included in the model.
The test protocol is the following: Starting from a clean catalyst at a stabilized temperature of 130°C,
the flow mixture (2000 ppmC HC, 10% H2O, N2) passes through the catalyst sample (GHSV=70000
h-1) for 500 s at constant inlet temperature (130°C). During this phase, the catalyst is saturated with
HC. At time=500s, the HC injection in the gas mixture is cut-off. At the same time the temperature of
the feed gas is steadily increased at a rate of 1K/s. The adsorbed hydrocarbon starts to desorb and the
HC concentration at the exit is monitored until it becomes zero.

2.3 Light-off test


The light-off tests were also conducted in the Synthetic Gas Bench, mainly to identify the reaction
kinetics of the CO and hydrocarbon oxidation. Starting from stabilized conditions at 100°C, the feed
gas was heated at a rate of 1°C/s and the catalyst efficiency was monitored by measuring CO, HC,
NOx before and after the catalyst sample. Two types of light-off tests were conducted. The first one
(Low GHSV) corresponds to low-medium load operation: GHSV=35000 h-1, CO=500 ppm, CO2=5%,
HC = 150 ppm C1, NO = 150 ppm, O2=13%, SO2=10ppm, H2O=5%. The second (High GHSV)
corresponds to high load operation: GHSV=70000 h-1, CO=250 ppm, CO2=10%, HC = 100 ppm C1,
NO = 300 ppm, O2=10%, SO2=20ppm, H2O=10%.

2.4 MVEG driving cycle


The CATOX2 catalyst was mounted on a vehicle (PSA – vehicle – engine displacement – catalyst
positioning).
The exhaust gas analysis during the test included measurement of CO, HC, NOx, CO2, O2 before and
after the catalyst. A set of thermocouples was also installed at different positions before, after and
inside the catalytic converter.

2
3. MODELING
3.1 Background
The currently available model for PM based diesel catalysts will be employed as a starting point. The
model describes the transient heat and mass transfer phenomena, as well as the heterogeneous
reactions in the catalyst. A 2-dimensional grid is employed to simulate the cylindrical catalyst. The
governing mathematical equations have been described in detail elsewhere [4].

3.2 Flow maldistribution


In order to predict the flow maldistribution at monolith inlet, a special technique is employed, which
calculates the flow distribution by solving an equivalent hydraulic circuit, in which the diffuser is
represented by suitable flow resistances. This technique is described in more detail in a previous
paper [6].

3.3 Zeolite diffusion limitations


An additional mass diffusion resistance is considered for the hydrocarbon species. According to the
related literature, this resistance follows an exponential temperature dependence. The pre-exponential
factors and the activation energies are the independent parameters that should be fitted for each HC
species.

3.4 H2O, HC Adsorption - Desorption


A further novel feature, which is introduced in the model in the frame of the present work is the
simulation of adsorption and desorption phenomena. The long-chain and the aromatics HC species are
much keener to adsorption in the zeolite, compared to the short chain molecules. Based on the
speciation of total hydrocarbons emitted by a typical diesel engine, we can consider that the
“adsorbable” HC species can be represented by decane and toluene.
Since we deal with adsorption in microporous zeolite, we follow the principles of the Polanyi
adsorption theory, extended by Dubinin and Radushkevich [5]. According to this model, the adsorbate
in intimate contact with the solid is in liquid form. The equation of the DR isotherm gives the
adsorbed mass at equilibrium xeq as function of temperature and partial pressure:
2 2
  p   RT 
ln xeq = ln (W0 ρ ) − D ln 0  , D = A 
  p   β 
where W0 is the total volume of all micropores and A is a constant characteristic of the pore size
distribution (both depend on the zeolite only).  is the affinity coefficient, which depends on the
adsorbate. ρ is the liquid phase density, p0 is the saturation pressure and p is the partial pressure of the
adsorbate at the gas-solid interface.
To predict the adsorption-desorption rates towards equilibrium, we employ the commonly used linear
driving force:
∂x
R= = k ⋅ (xeq − x )
∂t
k is a rate constant, which is constant in case of adsorption (x<xeq), but exhibits an exponential
temperature dependence in case of desorption (x>xeq).
H2O also interferes in the adsorption process, inhibiting HC adsorption in the zeolite [7, 8]. Water
adsorption is also taken into account based on the same theory. An interesting feature of H2O
adsorption is the high amount of latent heat released during the process.

3
The simulation of the adsorption of H2O and two different HC species (decane, toluene) leads to a
system of 3 differential equations for the adsorbed mass of each species vs time, which is solved by an
implicit numerical procedure.

3.5 Reaction mechanisms and kinetics


The NOx conversion levels observed with the specific catalyst at the tested conditions were too low to
allow an identification of the prevailing DeNOx mechanism and the respective kinetics. Thus, we will
restrict our attention to the catalyst performance as regards only CO and hydrocarbon conversion.
This simplifies substantially our reaction scheme. H2 oxidation is typically taken into account, since it
contributes significantly in the overall reaction exothermy. The reaction scheme is thus:
1. CO + 1 O 2 
→ CO 2
2
2. H 2 + 1 O 2  → H 2O
2
3-6. C x H y + (x + 0.25 y )O 2 
→ xCO 2 + 0.5 yH 2 O for each of the 4 HC species considered

The reaction rate expressions are, as usual, of the Langmuir – Hinshelwood type with inhibition terms
[4]:
E2 Ei

E1 − −
RT
A1e RT
cCO cO2 , A2 e RT
cH 2 cO2 Ai e cC x H y cO2
R1 = R2 = , Ri = (i=3 to 6)
G1 G1 G1

with:

G1 = T (1 + K1cCO + K 2 cCxHy ) 1 + K 3cCO


2 2
(
cC2 x H y (1 + K 4 cNO
0.7
) )
The activation energies, the pre-exponential constants, as well as the inhibition constants are fitted in
order to match the experimental with the model results for a variety of SGB tests, as will be presented
below.

4. RESULTS
4.1 Adsorption – desorption
Figure 1a presents the comparison between the measured and the computed HC (decane)
concentration curves during the adsorption test with catalyst CATOX1. In the first 500 s the catalyst
is gradually saturated with decane with a characteristic asymptotic behavior, which is accurately
predicted by the model, after the fitting of the related parameters (storage capacity, adsorption rate
constant). In the desorption phase (time>500 s), the HC concentration at exit is initially low. By
increasing the temperature, the desorption becomes faster and the HC concentration at exit increases
up to a maximum value. Afterwards, due to the decreased availability of adsorbed decane in the
catalyst, the desorption is slowed down, until it is totally completed. By fitting the respective model
parameters (desorption kinetics), it is possible to obtain a very successful prediction of all the above
phenomena.

4
2 E -3
C o n c e n tra tio n [H C 1 ] Inlet
1 .6E -3
O u tlet, exp erim en tal

1.2E -3 O u tlet, com p ute d

8 E -4

4 E -4

0
0 2 00 4 00 6 00 8 00 1 000
T im e [s]

2 E -3
Inlet
C o n c e n tra tio n [H C 1 ]

1 .6E -3 O u tlet, exp erim en tal

O u tlet, com p ute d


1.2E -3

8 E -4

4 E -4

0
0 2 00 4 00 6 00 8 00 1 000
T im e [s]
Figure 1: Measured –vs- computed HC concentration at catalyst exit during decane adsorption test. (a)
CATOX1, (b) CATOX2
To verify the physical meaningfulness of the model, we checked it against the adsorption test
performed with CATOX2. The only difference between CATOX1 and CATOX2 is the 50% higher
zeolite content of the latter. Thus, we increased the HC storage capacity by 50% and kept all other
model parameters same as in CATOX1. The results are presented in Figure 1b. Obviously, the model
is able to predict with remarkable accuracy the expected response, when using larger zeolite amounts.

4.2 CO, Hydrocarbon light-off in SGB


The CO conversion as function of temperature in the SGB tests follows the characteristic light-off
curve. By adjusting the rate parameters and the self inhibition term (K1) it was possible to obtain a
very good agreement between measurements and predictions for both light-off tests (low and high
GHSV).
The light-off tests have been carried out with different HC species in the feed gas, as well as different
HC combinations. It was noted that the CO conversion curves are not sensitive on the type of HC
present (less than 2°C difference in temperature for 50% CO efficiency).

5
10 0 10 0

80 80
E fficie n cy [% ]

E fficie n cy [% ]
60 60

Lo w G H S V H ig h G H S V
40 40
C om p u te d C om p u te d
20 20 E xp erim e ntal
E xp erim e ntal

0 0
10 0 1 50 20 0 25 0 30 0 35 0 10 0 1 50 20 0 25 0 30 0 35 0
T e m pe ra tu re [°C ] T e m pe ra tu re [°C ]

Figure 2: Measured –vs- computed CO conversion during light-off tests (Low – High GHSV)
Figure 3a presents the measured –vs- computed conversion curves for decane in the light-off tests.
The computed results have been obtained after adjusting the reaction rate parameters for decane, as
well as the diffusion resistance in the zeolite, which limits the efficiency in the high temperature
region. The coupling between adsorption – desorption phenomena and decane oxidation poses a
further challenge in this simulation. Nevertheless, the discrepancy between measurements and
predictions is considered acceptable.
100 100 100

80 80 80
E fficiency [% ]

60 60 60

40 40 40

L o w G H S V , e xp.

Low G HSV , cm p.
20 20 20
L o w G H S V , e xp.

H ig h G H S V , cm p .

0 0 0
100 150 200 250 300 350 100 150 200 250 300 350 100 150 200 250 300 350
T e m peratu re [°C ] T e m perature [°C ] T e m peratu re [°C ]
Figure 3: Measured –vs- computed HC conversion during light-off tests (Low – High GHSV) (a)
Decane, (b) Propene+Propane, (c) 4 HC
Figure 3b presents the results obtained with a mixture of propene/propane (molar ratio 3/1). For this
mixture, there was no test available with high GHSV. In order to be more realistic, we carried out a
light-off test with a mixture of 4 hydrocarbons (decane 50%, toluene 20%, propene 22.5%, propane
7.5%). This composition has been selected in order to represent as close as possible the actual HC
speciation in real exhaust. The kinetics of each one of the HC species have been obtained by
conducting tests with one species every time (not shown here). In Figure 3c we can see that the model
performs quite efficiently in the very challenging test with 4 HC species, which have different
adsorption, oxidation and diffusion characteristics.

6
4.3 MVEG test cycle
The ultimate goal of a realistic catalyst model is to predict the catalyst performance under real world
conditions. The measurements carried out on a car driven on a legislated driving cycle (MVEG) are
employed to assess the modeling capabilities and any possible deficiencies.
A main difficulty in the simulation of catalyst operation in real conditions is the prediction of flow
distribution at catalyst inlet face during the cycle. Since, flow measurements are not available, we
tried to fit the parameters of our flow distribution sub-model indirectly, taking into account the
measured temperatures after the catalyst at different radial positions. These temperature
measurements are strongly dependent on the local flow rate. After some trial and errors, we managed
to get an acceptable fit between measured and computed temperatures both at the centerline and the
periphery of the catalyst, as shown in Figure 4. Obviously, the temperature gradients in the radial
direction of the catalyst are very important which justifies the need of a 2-D simulation.
In the same figure, we also include for comparison the temperature prediction obtained with a
simulation, in which we ignored the effect of H2O adsorption. Clearly, the heat generated by the
adsorption and phase change of H2O in the zeolite is crucial for the catalyst thermal response in the
first seconds of the cycle.

300
C ore , e xp erim en ta l
C ore , co m pu ted

C ore , co m pu ted , n o H 2 O ad s orp tion

P e riph e ry, e xpe rim e n ta l


P e riph e ry, c om p ute d
200
T em perature [°C ]

P e riph e ry, c om p ute d , n o H 2 O a d so rp tio n

100

0
0 100 200 300 400 500
T im e [s]
Figure 4: Comparison between measured and computed temperatures at catalyst core region and
periphery.
Figure 5 illustrates the agreement between measured and computed instantaneous HC emissions
before and after the catalyst. As input to the model, we assumed the same proportion between the four
HC species as in the light-off tests in the SGB (see above). The model is able to predict the adsorption
phenomena taking place during the first seconds, as well as the HC oxidation after the warm-up phase.
For clarity reasons, the graph does not include the rest part of the cycle, where the model predictions
are equally successful.

7
1 5 00

In le t
1250 O u tle t, e xp erim e n ta l

O u tle t, co m p u te d
H C concen tration [ppm ]

1000

750

500

250

0
0 100 200 300 400 500
T im e [s]
Figure 5: Comparison between measured and computed instantaneous HC emissions in the MVEG
driving cycle (first 400 s).
Figure 6 presents a comparison between measured and computed CO, HC cumulative emissions in the
MVEG driving cycle. The accuracy of the model predictions is impressive for both pollutants, taking
into account the uncertainties regarding the flow distribution, the adsorption – desorption kinetics, the
reaction kinetics and the variability of HC species in the real exhaust. The small deviations in the
extra-urban part of the cycle can be attributed to the neglection of the de-NOx reactions, which would
consume further HC.
14 2 .0

1 .8
12
1 .6

10 1 .4

1 .2
8
C O [g]

H C [g]

1 .0
6
0 .8
C O e xp e rim e nta l

4 C O co m p u te d 0 .6

H C exp e rim en ta l 0 .4
2 H C co m p u te d
0 .2

0 0 .0
0 200 400 600 800 1 0 00 1 2 00
Tim e [s]
Figure 6: Comparison between measured and computed CO, HC cumulative emissions in the MVEG
driving cycle.

8
CO (efficiency [%]) HC (efficiency [%])
I II III total I II III total
Measurement 11.4 32.1 99.1 42.1 45.3 55.1 91.8 62.5
Model results 2.5 32.5 99.6 40.3 46.6 56.7 86.9 62.3
Assumptions
Uniform flow distribution 2.7 38.2 100 43.4 46.9 64.6 93.0 68.1
HC adsorption neglected 2.4 31.5 99.5 39.7 3.8 34.4 87.1 41.3
2 HC species (propene/propane) 2.4 32.2 99.6 40.1 7.3 50.6 83.7 49.6
H2O adsorption neglected 2.9 35.5 99.5 41.9 46.8 59.0 87.1 63.6
HC diffusion resistance neglected 2.5 32.9 99.6 40.5 52.1 61.9 96.4 68.7

Table 1: Summary of the model sensitivity to various assumptions. The CO and HC mass emissions
are given for each of the 3 MVEG test phases (Phase I: 0-196s, Phase II: 197-784s, Phase
III: 785-1180s) and as totals.
In Table 1, the measured vs computed results in the driving cycle are summarized. To better clarify
the different effects, the results are presented in the form of CO and HC efficiency (mass average) for
each of the 3 MVEG test phases (see legend) and as totals. From the comparisons between
measurements and model results in the first two lines of Table 1 it seems that the only significant
deviation regards the CO emissions in Phase I, where the model underestimates the catalyst
efficiency. This can be attributed to inaccuracies of the kinetics in the very low temperature regime.
It is interesting to check the relative importance of some model features, in order to determine which
phenomena should be paid major attention in future investigations. To do this, we performed a series
of MVEG driving simulations, introducing each time a different assumption. The results are
summarized in the bottom part of Table 1.
It is interesting to see that the assumption of uniform flow distribution at catalyst face would lead to a
significant overestimation of catalyst efficiency as regards both CO (+3.1%) and HC (+5.8%). This is
also an indication of the importance of designing flow-optimized cones at catalyst inlet.
The neglection of HC adsorption phenomena does not impact the CO prediction, as expected.
However, the model predicts that HC adsorption is responsible for about one third of the total HC
efficiency. Interestingly, HC adsorption does not only affect the cold start phase (Phase I), but also to
a lesser extent the other parts of the cycle. This result illustrates the importance of using HC adsorbent
materials in diesel catalysts.
The importance of employing a mixture of 4 hydrocarbons in the simulations, instead of the
commonly used discrimination between fast and slow HC is obvious. With the propene/propane
assumption, it is not possible to simulate the different adsorption and oxidation properties of the
various HC species present in the real exhaust.
Although the effect of H2O adsorption on the catalyst thermal response in the first 100 s of the cycle
is very strong (Figure 4), there is hardly any influence on the CO and HC emissions. Indeed, the
model results show that the temperature increase accompanying the H2O adsorption and phase change
is not enough to inhibit the adsorption of hydrocarbons during cold start.
Ignoring the diffusion limitations in the zeolite, the model predicts a substantial increase (+6.4%) in
total HC efficiency. As expected, this increase is much more evident in the Phase III (extra-urban part
of the cycle), where the diffusion limitations are more important due to the higher flow rates. On the
other hand, this result implies that the presence of the zeolite in the catalyst may have a slightly

9
adverse effect on HC efficiency. However, this effect is too small to counterbalance the significant
advantages of using HC adsorbing materials.

5. CONCLUSIONS
With the perspective to model as reliably as possible a zeolite containing diesel catalyst, the present
study led to a number of observations, useful for understanding and simulating the catalyst behavior.
Specially designed adsorption tests carried out in the synthetic gas bench enabled us to understand
and quantify the HC adsorption phenomena on the zeolite material, which are important especially
during the cold start. A series of light-off tests in the same bench provided valuable information on
the oxidation reaction kinetics of CO and four different HC species. At this stage, we did not have
enough information to model the de-NOx reaction mechanisms and estimate the respective kinetics.
The predictions of CO, HC emissions as well as catalyst temperatures were impressively successful in
the simulation of the MVEG driving cycle, which is a quite demanding modeling case. The relative
importance of different model assumptions was investigated. It was verified that a reliable modeling
approach for zeolite containing diesel catalysts should take into account: (a) adsorption phenomena
for hydrocarbons and water, (b) flow maldistributions, (c) accurate estimation of the oxidation
kinetics, (d) accurate representation of the actual exhaust hydrocarbons by four HC species, (e)
diffusion limitations in the zeolite.

REFERENCES

1 Stamatelos A.M., Koltsakis G.C., Kandylas I.P., Pontikakis G.N., “Computer Aided Engineering in
the Design of Diesel Exhaust Aftertreatment Systems”, SAE paper 1999-01-0458 (1999).
2 Ansell G., Bennet P., Cox J., Frost J., Gray P., Jones A., Rajaram R., Walker A., Litorell M. and G.
Smedler, "The development of a model capable of predicting diesel lean NOx catalyst
performance under transient conditions", Applied Catalysis B: Environmental, 10, pp. 183-
201, 1996.
3 Wilber R. A., Boehman A. L., “Numerical Modeling of the Reduction of Nitric Oxide by Ethylene
over Cu-ZSM-5 under lean conditions”, Catalysis Today 50 (1999) 125-132.
4 Kandylas I.P., Koltsakis G.C., Stamatelos A.M, “Mathematical Modelling of Precious Metals
Catalytic Converters for Diesel NOx Reduction”, Proceedings of the Insitution of Mechanical
Engineers, Journal of Automobile Engineering, Vol. 213 (1999), pp. 279-292.
5 Thomas J. M. and Thomas W. J., “Principles and Practice of Heterogeneous Catalysis” (VCH
Verlagsgesellschaft mbH, Weinheim, Germany, 1997)
6 Koltsakis G. C., Stamatelos A. M., “Thermal Response of Automotive Hydrocarbon Adsorber
Systems”, Transactions of the ASME: Journal of Engineering for Gas Turbines and Power,
January 2000.
7 Otto K., Montreuil C. N., Todor O., McCabe R. W., Gandhi H. S., “Adsorption of Hydrocarbons
and Other Exhaust Components on Silicalite”, Ind. Eng. Chem. Res. (1991), 30, 2333-2340.
8 Lafyatis D. S., Ansell G. P., Bennett S. C., Frost J. C., Millington P. J., Rajaram R. R., Walker A.P.,
Ballinger T. H., “Ambient temperature light-off for automobile emission control”, Applied
Catalysis B: Environmental 18 (1998) 123-135.

10