Thermal Deactivation of Pd/CeO2-ZrO2 Three-Way Catalysts during Real Engine Aging: Analysis by a Surface plus Peripheral Site Model.

The thermal deactivation of Pd/CeO2-ZrO2 (Pd/CZ) three-way catalysts was studied via nanoscale structural characterization and catalytic kinetic analysis to obtain a fundamental modeling concept for predicting the real catalyst lifetime. The catalysts were engine-aged at 600-1100 °C and used for chassis dynamometer driving test cycles. Observations using an electron microscope and chemisorption experiments showed that the Pd particle size significantly changed in the range of 10-550 nm as a function of aging temperatures. The deactivated catalyst structure was modeled using different-sized hemispherical Pd particles that were in intimate contact with the support surface. Therefore, Pd/CZ contained two types of surface Pd sites residing on the surface of a hemisphere (Pds) and circular periphery of the Pd/CZ interface (Pdb), whereas a reference catalyst, Pd/Al2O3, contained only Pds. In all Pd particle sizes investigated herein, Pd/CZ exhibited higher reaction rates than Pd/Al2O3, which nonlinearly increased with increasing slope as the weight-based number of surface-exposed Pd atoms ([Pds] + [Pdb]) increased. This finding contrasted with that of Pd/Al2O3, where the reaction rate linearly increased with [Pds]. When the Pds sites in both catalysts were equivalent in terms of their specific activities, the activity difference between Pd/CZ and Pd/Al2O3 corresponded to the contribution from Pdb, where oxygen storage/release to/from CZ played a key role. This contribution linearly increased with [Pdb] and therefore decreased with Pd sintering. Although both Pds and Pdb sites showed nearly constant turnover frequencies despite the difference in the Pd particle size, the values for Pdb were more than 2 orders of magnitude greater than those for Pds when assuming a single-atom width one-dimensional Pdb row model. These results suggest that the thermal deterioration of the three-phase boundary site, where Pd, CZ, and the gas phase meet, determines the activity under surface-controlled conditions.

Introduction

Currently, three-way catalysts (TWCs) play a major role in the abatement of hazardous exhaust gas components, including NO, CO, and hydrocarbons, of gasoline engines. A key challenge involves reliably predicting and enhancing the long-term durability of the catalyst performance, which is impaired by thermal degradation. The most probable deactivation mode is the sintering of platinum group metals (PGMs) and CeO2-based cocatalysts, such as CeO2–ZrO2 (CZ), which reduce the active surface area and therefore the metal–CZ interface.[1,2] This interface is particularly important because the three-phase boundary (TPB) of the metal, CZ, and gas phase provides the catalytically active sites.[3−9] On these sites, CZ enables the oxygen storage/release in response to an oscillating air-to-fuel ratio (A/F),[10−18] promoting the catalytic conversion by PGMs. Oppositely, PGMs increase the rate of oxygen storage/release through spillover to/from CZ.[16,19]

Heo et al.[5] reported that the weakening of the Pd–Ce interaction decreased the oxygen storage capacity (OSC) and contributed to the deactivation of TWC. Martínez-Arias et al.[20,21] reported that the sintering-induced decrease of the interface between Pd and CZ in Pd/CZ/Al2O3 catalysts reduced the conversion efficiencies of CO and NO. Moreover, many researchers proposed that the TPB sites in PGM-supported CeO2-based oxides are active for TWC-related reactions, including CO oxidation and water–gas shift reactions.[22−27] The reaction rate linearly increases with the length of the periphery of a metal–oxide interface.[26−29] For instance, Haruta reported that the turnover frequency (TOF) of CO oxidation is related to the length of the Au–CeO2 interface and increases as the Au particle size decreases.[30] Efstathiou et al.[26] studied the water–gas shift reaction in Pt/CeO2 and revealed that as the Pt particle size increases (or Pt loading increases), the surface concentration of the active species decreases and the specific rate per length of periphery of the Pt–CeO2 interface linearly increases. Notably, in most previous studies, the length of the metal–CeO2 periphery (metal particle size) in supported catalysts is varied by changing the metal loading. Preparing a series of different-sized metal particles without changing the metal loading is probably difficult. Moreover, there are still few scientific studies of the deactivated honeycomb catalysts treated under practical conditions.

In our previous study,[31] full-scale honeycomb TWCs with constant metal loading (ca. 2 wt % Pd/CZ) after engine aging at elevated temperatures were analyzed using in situ diffuse reflectance spectroscopy under oscillation between fuel-lean and fuel-rich conditions. In a fresh highly dispersed catalyst, Pd reoxidation rate was significantly retarded owing to oxygen storage in CZ. Conversely, in a high-temperature engine-aged catalyst in which most of the Pd particles were enlarged, Pd species that exist far from the TPB were no longer influenced by oxygen storage/release. In this case, the retardation effect was thus considerably more subdued than in the case of the fresh catalyst. This result implies that the specific activity of Pd depends on whether they are located on the Pd/CZ interface (peripheral site) or on the Pd surface far from the interface (surface site). However, the extent to which the contribution from each active site determines the catalytic performance of deactivated catalysts remains unclear.

This study aimed to extend our previous work on the spectroscopic characterization of Pd/CZ TWC and provide an insight into the modeling concept of deactivated catalyst structures to establish kinetic predictions. Real-scale monolithic honeycomb TWCs containing a Pd/CZ or Pd/Al2O3 (Pd/A) catalyst were aged in an engine dynamometer at different temperatures, and their catalytic performance was examined in the chassis dynamometer driving test cycles. The use of engine aging and chassis dynamometer tests is emphasized for real-scale performance and simulated reactions for laboratory-scale kinetic analysis. The obtained results were coupled with those of detailed structural characterizations using chemisorption, X-ray diffraction (XRD), scanning transmission electron microscope (STEM) equipped with a high-angle annular dark-field detector (HAADF) and X-ray detector, and other methods that are outlined in the subsequent content. Based on these results and our previous findings, a simple numerical description of Pd/CZ catalysts was discussed assuming the contributions from surface and peripheral Pd sites, the abundance of which depended on the Pd particle size in engine-aged catalysts.

Results and Discussion

Structural Characterization of Engine-Aged Catalysts

The catalytic performances of the fresh and engine-aged honeycomb catalysts (Pd/CZ and Pd/A) were evaluated using the chassis dynamometer test. Figure shows the plots of catalytic performances versus aging temperatures. The performance was compared in terms of the cumulative conversion efficiency of CO, NO, and total hydrocarbons (THC) in the 40–200 s period of the cold-start New European Driving Cycle (NEDC) test mode (Supporting Information, Figure S1). The conversion efficiencies of both the fresh catalysts were nearly complete but monotonically decreased with increasing aging temperature. Pd/CZ tended to preserve higher conversions than Pd/A as the aging temperature increased above 800 °C. These activity trends resemble that observed for the catalyst powder removed from each honeycomb catalyst, as measured using a simulated gas mixture at A/F = 14.6 (Supporting Information, Figure S2). The difference in the deactivation behavior was unlikely to be solely attributed to the thermal sintering of Pd, as described below. However, after aging at 1100 °C, the activity of both catalysts dropped at nearly the same point, suggesting a steep deterioration in the Pd/CZ active phase at >1000 °C. After the chassis dynamometer test, the catalyst powders were collected from the honeycomb catalysts for structural characterization using XRD, HAADF–STEM/X-ray mapping, and gas adsorption experiments. Before these experiments, the catalysts were treated in H2 at 400 °C to convert Pd oxide to Pd metal for easy detection. The XRD of the fresh and aged catalysts demonstrated the simple particle growth behavior of Pd and CZ as the aging temperature increased to up to 1000 °C (Supporting Information, Figure S3). Noticeable changes in the catalyst structure were observed at 1100 °C, where the CZ peak shift to lower 2θ with the simultaneous appearance of ZrO2 peaks suggested partial decomposition of the CZ solid solution with the cubic fluorite-type structure. In the Pd/A catalyst, the phase transformation of a transition alumina phase (mainly θ-Al2O3) to α-Al2O3 was noted. These changes resulted in a considerable loss of SBET (Table ),[32] which was a primary reason for the substantial deactivation at 1100 °C (Figure ).

Figure 1

Plots of cumulative NO, CO, and THC conversion efficiencies in the 40–200 s period from cold-start urban driving cycles in the chassis dynamometer test for Pd/CZ and Pd/A honeycomb catalysts (see Supporting Information, Figure S1).

Table 1

Characterization Results of Pd/CZ and Pd/A after Engine Aging at Different Temperatures

catalyst	aginga	SBETb (m2 g–1)	dCOc (nm)	dEMd (nm)	dXRDe (nm)	[Pds]f (μmol g-cat–1)	[Pdb]g (nmol g-cat–1)	CZ crystallite sizeh (nm)
Pd/CZ	uncoated	67	15.3		ND	13.11	513	11
	fresh	61	24.4	11.3 ± 3.3	ND	7.61	185	12
	600 °C	53	47.4		ND	3.96	48.7	9
	700 °C	50	48.8		18	4.85	46.1	11
	800 °C	35	50.6	41 ± 12	35	3.71	42.8	17
	900 °C	27	88.2	80 ± 27	73	2.14	11.1	24
	1000 °C	17	180	119 ± 37	123	1.05	3.39	45
	1100 °C	6	481	208 ± 73	216	0.40	0.48	89
Pd/A	uncoated	150	10.7		ND	17.80
	fresh	116	14.4	8.9 ± 2.8	ND	13.20
	600 °C	105	32.9		ND	5.78
	700 °C	103	32.9		17	5.77
	800 °C	95	38.9	57 ± 21	37	4.88
	900 °C	86	73.2	122 ± 53	94	2.60
	1000 °C	70	154	198 ± 72	173	1.24
	1100 °C	44	552	489 ± 180	184	0.34

Aged in an engine dynamometer under stoichiometric-lean/rich gas cycles at the specified temperature for 40 h (Supporting Information, Figure S8). The fresh powder catalyst before coating onto a honeycomb is shown as “uncoated”. The samples were treated in 5% H2 at 400 °C.

BET surface area of Pd-loaded catalysts.

Pd particle size calculated from the amount of CO chemisorbed assuming a chemisorption stoichiometry of Pd/CO = 1:1 and hemispherical particle shape.

Surface area-weighted mean Pd particle size and standard deviation, calculated using histogram analysis of the HAADF–STEM/X-ray mapping images.

Pd crystallite size calculated by X-ray line-broadening analysis using Scherrer’s equation. ND: Pd peaks were not detected.

The number of surface-exposed Pd sites calculated from dCO.

The number of Pd sites lying along the circular Pd/CZ periphery interface calculated from dCO.

Calculated by X-ray line-broadening analysis using Scherrer’s equation.

HAADF–STEM and energy-dispersive X-ray mapping analysis were performed to determine the particle size distribution of Pd. In Figure , the Pd particles are highlighted in red in the composite images. In the fresh Pd/CZ catalyst, Pd particles of approximately 10 nm or less in size were well dispersed in the CZ support. The primary CZ particles were as small as 10–20 nm in size but weakly agglomerated into large porous structures of over 100 nm. The same structures were observed on the fresh Pd/A, although the primary particle size of Al2O3 (≤10 nm) was smaller than that of CZ. The engine-aged catalysts showed a similar dispersion state, whereas the Pd particle size increased as the aging temperature increased. Notably, aging at 1100 °C led to the obvious appearance of very large crystallites of support oxides (CZ: >100 nm and Al2O3: >500 nm) accompanied by structural deteriorations detected using XRD (Supporting Information, Figure S3). The partial decomposition of the CZ solid solution at an aging temperature of 1100 °C was also confirmed by X-ray images, which represented particles containing Ce but not Zr (Supporting Information, Figure S4). Large particles of >500 nm in Pd/A were highly crystalline α-Al2O3, which were precipitated in fine particles of θ-Al2O3.

Figure 2

HAADF–STEM and Pd X-ray mapping images (highlighted in red) of Pd/CZ and Pd/A after engine aging at elevated temperatures. The aged catalysts were treated in H2 at 400 °C.

The distributions of the Pd particle size are shown in Figure . Both fresh catalysts exhibited sharp distributions close to a Gaussian shape with maxima at approximately 10 nm. As the aging temperature increased, the Pd particle size increased and spread over a wide range, becoming asymmetric, with a greater width on the larger side of the maximum. This was clearly observed for Pd/A-1000 in which Pd particles >100 nm in size were abundant (Figure ). Conversely, in Pd/A-900, large (>100 nm) and small (<20 nm) Pd particles coexisted. Previous theoretical model studies have suggested that a size distribution skewed toward small particles can arise from Ostwald ripening, while a size distribution with a long tail toward large particles can only arise from particle migration and coalescence.[33−35] It should be remarked that such a broad distribution of the Pd particle size is a characteristic of TWCs after real engine aging, which makes it difficult to analyze and model the deactivation behavior. The broadest size distribution that formed after aging at 1100 °C was particularly significant for Pd/A than for Pd/CZ. Regardless of the difference in particle sizes, Pd particles preserved spherical morphology and intimate contact with the support surfaces.

Figure 3

Pd particle size distribution of Pd/CZ and Pd/A after engine aging at elevated temperatures. The surface area-weighted mean Pd particle size is shown as dEM.

Based on distributions of Pd particle size, the surface area-weighted mean size (dEM) was determined for comparison with that obtained from the amount of CO chemisorption (dCO) assuming a spherical particle shape and Pd/CO stoichiometry of 1:1. The in situ FTIR results suggested that CO adsorption bands attributed to linear (∼2080 cm–1) and bridged (1980–1990 cm–1) carbonyl species were observed for the fresh catalysts, whereas the bridged species were dominant for the aged catalysts (Supporting Information, Figure S5). The stoichiometry for CO chemisorption on supported Pd catalysts is a controversial topic. Each linear and bridged CO occupies one- and two-surface Pd atoms, respectively, at the saturation coverage.[36,37] Conversely, a 1:1 ratio of bridged CO to Pd has been reported by several researchers.[38,39] Owing to the very low proportions of linear to bridged carbonyl species obtained, except the fresh catalysts in this study, a constant Pd/CO stoichiometry of 1:1 was selected to calculate the Pd particle size, which showed good agreement with electron microscopy data.

Table compares dEM and dCO with the mean crystallite size calculated from X-ray line-broadening analysis using dXRD. Aging Pd/CZ in a moderate temperature range (≤900 °C) agreed well with the Pd particle size trend between dEM and dCO. Conversely, the agreement did not hold true for catalysts aged at higher temperatures (≥1000 °C), where dCO revealed significantly larger values than dEM. A similar discrepancy was also observed for Pd/A catalysts (except Pd/A-1000) and explained in terms of large errors while determining small amounts of CO chemisorbed onto large Pd particles grown with limited specific surface areas. However, the discrepancies at such large Pd particles sizes do not matter when considering the number of surface Pd sites per catalyst weight. Although dXRD yielded a volume-weighted mean crystallite size,[40] which was generally larger than the surface-area-weighted average, several exceptions were found in Pd/CZ-700 and Pd/A-700, which indicated that Pd particles comprising few crystallites were present. In the following experiments, geometrical parameters of the Pd particle determined via CO chemisorption were used to construct the model structures of aged catalysts.

Structural Models for Deactivated Catalysts

The structural model of Pd/CZ and Pd/A catalysts was proposed based on the characterization results, as schematically illustrated in Figure a. For simplicity, Pd particles were considered hemispherical in shape and in contact with the support surfaces. This assumption was reasonable because the Pd particles observed in Figure were nearly spherical and well dispersed on the support matrix. Each CZ crystallite in engine-aged Pd/CZ was smaller in size than Pd particles but weakly agglomerated into larger porous structures. A similar structure was observed for Pd/A; the crystallite size of Al2O3 was smaller than that of Pd particles, but the size of weakly agglomerated Al2O3 was considerably larger. Therefore, the structure of these supported catalysts could be approximated to monodispersed Pd hemispheres on a two-dimensional support surface.

Figure 4

(a) Schematic model structures of Pd/CZ and Pd/A catalysts to show the contributions of surface Pd sites (Pds) and peripheral Pd sites (Pdb) and (b) their numbers and (c) fraction of Pdb in Pd/CZ as a function of Pd particle size (dCO).

In the Pd–CZ model, two types of surface-exposed Pd atoms, residing on the surface of a hemisphere (Pds) and circular periphery of the Pd/CZ interface (Pdb), are responsible for TWC. This hypothesis is based on the fact that the interface or the TPB of the metal, support, and gas phase of many catalyst systems comprising metals and CeO2-based supports provides active sites for the catalytic reaction owing to the oxygen storage/release functions to/from the support.[3−9,20] Other reported deep interactions such as decoration,[41] alloying,[41] Pd encapsulation,[42] and surface complex (PdCeO2) formation[43,44] are not considered here. The decoration of Pd by Ce oxide and Pd–Ce alloying were reported to occur only under a high-temperature reducing atmosphere.[41] The encapsulation of Pd by a CZ support and the surface complex formation were reported when the support surface area was fallen by a factor of 1 or 2 orders of magnitude.[43] These situations can be ruled out under the stoichiometric gas air-rich gas-cycled aging condition of the present study, where the largest loss of SBET was only up to 72% (Pd/CZ-1000, Table ). Although one may point out that the periphery of Pd on Al2O3 also affects the catalytic behavior, we assume that its effect should be negligible due to lack of redox ability of Al2O3 and large Pd particle sizes (≥10 nm).

The total number of surface-exposed Pd atoms (Pds and Pdb) was assumed to be equal to that of CO chemisorption, which was close to that estimated from the surface area-weighted mean Pd size determined via histogram analysis of electron microscopic images, as shown in Table . However, the Pd/A catalyst exhibited active sites on the surface of a hemisphere (Pds), the specific activity of which was assumed to equal that of Pd/CZ. Therefore, the activity difference between Pd/CZ and Pd/A stemmed from the presence of Pdb when their Pd particle sizes were the same. Based on the model shown in Figure a, the total number of Pd atoms on the surface, [Pds,tot], was expressed using a Pd particle diameter determined by CO chemisorption (dCO), Pd loading (wPd/wcat = 0.0182), density Pd metal (ρPd = 12.023 g cm–3), number of Pd atoms per unit surface (sPd = 1.260 × 1019 atom m–2), and Avogadro constant (NA).

A factor 3 in eq represents an arbitrary way of considering the fraction of the bottom surface of the hemisphere that is in contact with the support surface, which does not contribute to the catalytic process. The number of peripheral sites, [Pdb], can be estimated from the length of the perimeter of the Pd/CZ interface (m g-cat–1), considering a circular geometry for the interface. Similar calculations have been reported elsewhere in Pt/CeO2 for water–gas shift reactions, Pd/CeO2 for CH4 combustion, and Pt/LaFeO3 for NO–H2 reactions.[26,45,46]

Assuming that a single-atom Pdb row aligned along the periphery of the Pd/CZ interface, the as-calculated length was divided by the Pd atomic distance in the Pd metal crystal (0.275 nm) to convert into [Pdb] (mol g-cat–1). The number of Pds could therefore be derived using the following equation:

The as-calculated values of [Pds] and [Pdb] are plotted in Figure b and listed in Table . The number of [Pdb] was approximately 4% of [Pds,tot] for the fresh (uncoated) Pd/CZ, and the ratio decreased to 0.1% as the Pd particle size of the aged catalysts increased (Figure c). Notwithstanding the small number of Pdb sites on the Pd surface, their impact on the observed catalytic performance was substantial, as discussed in more detail later.

Kinetic Analysis of Deactivated Catalysts

Considering that the number of Pds and Pdb sites in the present catalysts is solely dependent on dCO, their contributions to specific kinetic rates provide detailed insights into the mechanism of thermal deactivation. A model TWC reaction was performed using Pd/CZ and Pd/A catalysts under steady-state conditions to study the relationships between the kinetically controlled reaction rates of NO, CO, and C3H6, and [Pds,tot]. Figure shows the plots of these relationships at 280 °C for Pd/A and Pd/CZ as functions of [Pds,tot]. Clearly, the reaction rate over Pd/A was nearly proportional to [Pds,tot] that was equal to [Pds] for Pd/A, demonstrating that the rate per each Pds (TOF) was almost independent of the dispersity of Pd. This finding indicates that each Pd atom exposed on the surface equally contributed to the overall rate. This so-called structure-insensitive behavior is known to occur in CO–O2 and CH4–O2 reactions in Pd catalysts.[47−49] However, Pd/CZ exhibited higher reaction rates than Pd/A, which nonlinearly increase with increasing slope as [Pds,tot] increases. Similar reaction rate dependences were also confirmed for CO/C3H6 oxidation rates. Furthermore, the shape of the plot in Figure tends to remain unchanged if [Pds,tot] is calculated from dEM and/or dXRD instead of dCO used in eq (Supporting Information, Figure S6).

Figure 5

Dependences of kinetically controlled reaction rates on the total number of Pd atoms on the surface (Pds,tot).

Based on the model shown in Figure a, the Pds sites in Pd/CZ and Pd/A are equivalent in their intrinsic activities, resulting in the straight line, as shown in Figure . However, the activity difference between the two lines could be attributed to the presence of Pdb in Pd/CZ or the absence of Pdb in Pd/A. In relation to the hypothesis that the difference between the two lines corresponds to the contribution of periphery Pdb sites along the interface with CZ, this contribution steeply increases as the [Pds,tot] increases and therefore as the Pd particle size decreases. This situation is schematically illustrated in Figure a. By approximating a linear function to the actual rate data for Pd/A, the straight line could be empirically expressed as rs = TOFs[Pds]. This was subtracted from the observed reaction rates over the Pd/CZ catalysts with different [Pds,tot] to estimate the rate contribution of the Pdb site. Figure b shows the plots of the obtained partial rates for NO, CO, and C3H6 at 280 °C for the aged Pd/CZ as functions of [Pdb]. Clearly, linear correlations were obtained for all three gas reactants, implying a nearly constant TOF for the periphery Pdb site; rb = TOFb[Pdb]. Notably, the apparent values for TOFb are more than 2 orders of magnitude higher than those for the surface Pds site (TOFs), as detailed in Table . This suggests that active sites along the Pd/CZ interface (Pdb) possess a considerably higher intrinsic reactivity than those on the surface (Pds). This result is in accordance with the well-known idea that most active sites in TWC involve contacts between Pd and the Ce-based oxide supports.[20,21]

Figure 6

(a) Estimated contributions of peripheral Pd site (Pdb) and surface Pd site (Pds) in Pd/CZ on the reaction rate. (b) Plots of rb versus the number of Pdb in Pd/CZ.

Table 2

Estimated TOF of Pds and Pdb Sites in Pd/CZ (280 °C)

TOF	NO	CO	C3H6
TOFsa (min–1)	1.67	37.0	2.82
TOFbb (min–1)	4.14 × 102	105.7 × 102	6.02 × 102

Estimated from the kinetic data of Pd/A and [Pds] assuming that the Pds sites in Pd/A and Pd/CZ are equivalent.

Estimated from the differential kinetic data between Pd/A and Pd/CZ divided by [Pdb] assuming the single-atom width 1D Pdb row model, as shown in Figure a.

One may point out that the present calculation based on a single-atom width one-dimensional (1D) Pdb row model overestimates the TOFb (Figure a). Considering that a possible contribution of several-atom width Pdb rows cannot be ruled out, the TOF depends on the number of Pd atom rows from the periphery that play a role as the Pdb, which is simulated in Figure . Here, the width of Pdb rows in the TPB region is defined as an arc length, L, where the O transfer to/from CZ affects the catalytic activity of Pd. The fraction of Pdb increases with increasing L but is always greater in Pd/CZ-fresh (dCO = 24.4 nm) than in Pd/CZ-1000 (dCO = 180 nm), as shown in Figure a. Based on these fractions, TOFs and TOFb for both catalysts for NO conversion at 280 °C were recalculated as a function of L. Clearly, Pd/CZ-fresh shows TOFb decreasing but TOFs increasing with L, as shown in Figure b. However, even if a 10 atom width Pdb row was assumed (L = 2.75 nm), the calculated TOFb was still over 20-fold higher than TOFs, which represented a remarkable superiority of Pdb compared with Pds. Even after thermal sintering (Pd/CZ-1000), the superiority of Pdb compared to Pds remains unchanged in a wide range of L (Figure c). A significant relationship was also observed between [Pdb] and the OSC of aged Pd/CZ (Supporting Information, Figure S7). The only exception was Pd/CZ-1100, which showed the deterioration due to CZ phase separation, as described above.

Figure 7

(a) Estimated fraction of Pdb in Pd/CZ and TOF of NO conversion (280 °C) for (b) Pd/CZ-fresh and (c) Pd/CZ-1000 as a function of TPB width (L).

Another important feature of the surface plus peripheral site model shown in Figure a is that it was useful for predicting the catalyst deactivation behavior. As evident in Figure a, the ratio of activity contributions from Pds and Pdb significantly depends on [Pds,tot], i.e., the mean Pd particle size, representing different degrees of thermal sintering. In the case of the NO conversion rate for the Pd/CZ-fresh catalyst, rb bears 87% of the relative activity, whereas rs bears 13%. According to the approximate lines in Figure a, this activity ratio of rb/rs monotonically decreases from 87:13 to 29:71 as the aging temperature increases to 1000 °C. This clearly indicates that the superiority of the activity of Pd/CZ to that of Pd/A remains unchanged even after significant sintering where the Pd particles become very large (>100 nm). However, the contribution of the peripheral Pdb site decreases as the Pd sintering progresses because the Pd/CZ line tends to asymptote to the Pd/A line with decreasing [Pds]. Thus, the reference catalyst, Pd/A, corresponds to an extreme case of deactivated Pd/CZ in which the periphery Pdb site and thus TPB are nearly lost. Information relating to the activity ratio is useful for kinetic modeling of the TWC catalyst deactivation and estimation of the catalyst lifetime. One exceptional case occurred after thermal aging at 1100 °C in which the partial phase separation of CZ into ZrO2 and CeO2 deteriorated the oxygen storage/release functions of the catalyst; this caused the most serious deactivation and OSC loss, as shown in Figure and Figure S7, respectively.

Conclusions

The thermal deactivation mechanism of Pd/CZ TWC in comparison with a reference Pd/Al2O3 catalyst was studied using full-scale honeycombs after engine aging under stoichiometric-lean/rich gas cycles at elevated temperatures (600–1100 °C) and chassis dynamometer driving tests. The nanoscale structure of the deactivated Pd/CZ catalysts was modeled as different-sized hemispherical Pd particles that were in intimate contact with the CZ support surface based on detailed structural characterization using HAADF–STEM/X-ray mapping analysis and the CO chemisorption technique. The model contained two types of surface-exposed Pd sites residing on the surface of a hemisphere (Pds) and circular periphery of the Pd/CZ interface (Pdb), while the reference catalyst, Pd/Al2O3, contained only Pds. The presence of the Pdb site in which oxygen storage/release to/from CZ occurred, was a key reason for the higher catalytic performance of Pd/CZ than Pd/Al2O3 for the same Pd particle size. The catalytic kinetic analysis under surface-controlled conditions revealed that the reaction rate for Pd/CZ nonlinearly increased with increasing slope as the weight-based number of surface-exposed Pd atoms increased, which was in contrast with the linear behavior of Pd/Al2O3. The Pdb site along the Pd/CZ periphery should possess different intrinsic site activities depending on the Pd particle size. Assuming that the Pds sites in both catalysts are equivalent in their activities, the activity difference between Pd/CZ and Pd/Al2O3 should reflect the contribution from the Pdb site. Based on the single-atom width 1D Pdb row model, the calculated TOF for Pdb was more than 2 orders of magnitude greater than that for Pds. The superiority of Pdb remained unchanged when a 10 atom width Pdb row was assumed. Therefore, it is concluded that the residual amounts of the TPB, where Pd, CZ, and the gas phase meet, determines the specific activity of engine-aged Pd/CZ catalysts.

Experimental Section

Preparation of Real-Scale Honeycomb Catalysts

Two types of supported Pd catalysts, 2 wt % Pd/CeO2–ZrO2 (Pd/CZ) and Pd/Al2O3 (Pd/A), were prepared via wet-incipient impregnation using Pd(NO3)2 (Tanaka Precious Metals, Japan), CeO2–ZrO2 (46.5 mol % Ce, 46.5 mol % Zr, 1.7 mol % La, and 5.3 mol % Nd, Brunauer–Emmett–Teller surface area (SBET) = 67 m2 g–1, Daiichi Kigenso Kagaku Kogyo, Japan), and γ-Al2O3 (3 wt % La2O3 added, SBET = 150 m2 g–1, Sasol Chemicals, U.S.A.). After impregnation, drying, and calcination at 600 °C for 3 h in air, the supported catalysts were obtained. Mixtures of catalyst powders, Al2O3 sol (Nissan Chemical Industries, Japan), and distilled ion-exchanged water were ball-milled into a homogeneous slurry. The as-prepared slurry was coated onto a cordierite honeycomb (diameter × length = 105.7 mm × 114.0 mm, volume = 1.0 L, cell density = 600 cells in–2, 4.3 mil, NGK Insulators, Japan) via a simple dip-coating technique. After drying at 90 °C, the coated honeycombs were air-calcined at 600 °C. The total amount of washcoat was 110 g L–1 (1.82 wt % Pd), containing Al2O3 sol (10 g L–1) and Pd (2 g L–1).

Engine Aging and Chassis Dynamometer Tests

As-prepared honeycomb catalysts were tested before and after gasoline engine aging, as described in further detail in our previous report.[31] The aging was conducted at almost constant temperatures of 600–1000 °C for 40 h. Meanwhile, three gas feeds, i.e., a stoichiometric gas (A/F = 14.6, 25 s), air (2.5 s), and rich gas (A/F = 12.0, 2.5 s), were sequentially cycled at a flow rate of 178,000 L h–1 (Supporting Information, Figure S8). Hereafter, these catalyst samples are referred to as X-Y, where X and Y denote the catalyst type and aging temperature, respectively. Therefore, Pd/CZ-1000 specifies the Pd/CZ catalyst after engine aging at 1000 °C. The chassis dynamometer tests of fresh and aged honeycomb catalysts were performed in an FEB-DNR apparatus (Meidensha, Japan), using a Mitsubishi Mirage-CVT gasoline vehicle with a 1.2 L engine and close-coupled TWC. Urban driving test cycles accorded with the NEDC were performed with a cold engine at the beginning of the cycle. The concentrations of CO, NO, and THC in the effluent gas were analyzed online using a motor exhaust gas analysis system (MEXA 9100, Horiba, Japan) equipped with nondispersive infrared detectors and a flame ionization detector. The OSC of honeycomb catalysts was evaluated in an engine dynamometer at 450 °C. The transient signals from linear air/fuel sensors, which were installed near the inlet and outlet of the catalytic converter, were recorded while switching the gas feed between A/F = 13.5 and 15.5 supplied at a space velocity of 30,000 h–1. The OSC was obtained by integrating the recorded signal differences.

Characterization of Catalysts

After the chassis dynamometer tests of engine-aged catalysts, a small cylindrical catalyst fragment (diameter × length = 20 mm × 50 mm) was excised from the central portion of the honeycombs (Supporting Information, Figure S9). Hereafter, the catalyst powder was removed from the washcoated surface layer. The as-prepared noncoated Pd/CZ and Pd/A catalyst powders were used as references. Prior to the following characterization, all the catalyst powders were pretreated in a stream of 5% H2/N2 at 400 °C. XRD was conducted under monochromated Cu Kα radiation (40 mA, 45 kV, Cu tube, X’pert MPD, Spectris PLC, UK). X-ray fluorescence (EDXL300, Rigaku, Japan) was used to analyze the Pd metal loading. The measurement value agreed with the expected value (1.82 wt %). N2 adsorption isotherms for the SBET calculation were obtained at ∼196 °C (Belsorp-mini, Microtrac-BEL, Japan). The crystallite sizes of Pd and CZ were calculated from the XRD data using the Scherrer formula. The microstructure was observed using HAADF–STEM and energy-dispersive X-ray mapping analyses, performed on a Tecnai Osiris apparatus (FEI, U.S.A.) with an accelerating voltage of 200 kV. To determine the Pd particle size distributions, X-ray images were processed on 100–500 particles using the ImageJ software (NIH, U.S.A.). This software enables thresholding images and quantitative measurements of the areas of highlighted particles. The number of pixels within the areas was automatically counted. Based on the scale bars applied on the images, the size of the area was able to be quantified and converted to the diameter of spherical particles. The surface area-weighted mean particle size (dEM) was calculated using the following equation:[40]where ni represents the number of particles with diameter di.

The amount of CO chemisorbed onto Pd was determined using the pulsed injection method at 50 °C (Quadrasorb SI, Quantachrome, Austria). Because CO also adsorbs onto CZ as the carbonate species, this method tends to overestimate the CO chemisorption. To prevent such an overestimation, CO chemisorption measurements were conducted using the CO2 preinjection technique.[50] When determining the Pd particle size (dCO), a hemispherical shape and chemisorption stoichiometry (Pd/CO = 1:1) were assumed. The in situ Fourier transform infrared spectroscopy (FTIR) spectra of the chemisorbed substrate on each catalyst were acquired in a transmission mode using a Nicolet iS50 FTIR spectrometer (Thermo Fisher Scientific, U.S.A.) with a temperature-controllable transmission cell and HgCdTe detector. A catalyst pellet with a diameter of 10 mm was placed in the cell at 50 °C under a supply of a gas stream containing 1% CO balanced with He for 5 min. The cell was subsequently purged with pure He for 10 min to remove gaseous CO before the spectrum was recorded.

Catalytic Reactions

The catalyst powder removed from the washcoated surface layer was used for the catalytic kinetic analysis (no H2 reduction applied). The as-prepared noncoated Pd/CZ and Pd/A catalyst powders were used as references. The steady-state catalytic activity for simulated TWC reactions was performed in a flow reactor using a stoichiometric gas stream, which contained NO (0.050%), CO (0.49%), C3H6 (0.039%), O2 (0.40%), H2O (5%), and He (balance). The catalyst granules (20 mesh) were fixed in a quartz tube (4 mm inner diameter) packed with quartz wool at both ends of the catalyst bed. The temperature range and catalyst weight were varied from 250 to 350 °C and 1.0 to 50 mg, respectively, to determine the value of the fractional conversion and reaction rate under differential reactor conditions (fractional conversion was lower than 20%). The total flow rate F was maintained constant at 100 cm3 min–1. A steady state was confirmed through constant conversions versus time on stream, which was maintained below 20%. The compositions of the inlet and effluent gas streams were analyzed online using Horiba VA-3000 and FIA-510 nondispersive infrared detectors (CO, NO, N2O, and NH3), flame ionization detectors (C3H6), chemiluminescence detectors (NO), and O2 sensor (JKO-O2LD3, Jikco, Japan).