Kinetic Study on CO-Selective Methanation over Nickel-Based Catalysts for Deep Removal of CO from Hydrogen-Rich Reformate

: The CO-selective methanation process is considered as a promising CO removal process for compact fuel processors producing hydrogen, since the process selectively converts the trace of CO in the hydrogen-rich gas into methane without additional reactants. Two different types of efﬁcient nickel-based catalysts, showing high activity and selectivity to the CO methanation reaction, were developed in our previous works; therefore, the kinetic models of the reactions over these nickel-based catalysts have been investigated adopting the mechanistic kinetic models based on the Langmuir chemisorption theory. In the methanation process, the product species can react with the reactant and also affect the adsorption/desorption of the molecules at the active sites. Thus, the kinetic parameter study should be carried out by global optimization handling all the rate equations for the plausible reactions at once. To estimate the kinetic parameters, an effective optimization algorithm combining both heuristic and deterministic methods is used due to the large solution space and the nonlinearity of the objective function. As a result, 14 kinetic parameters for each catalyst have been determined and the parameter sets for the catalysts have been compared to understand the catalytic characteristics.


Introduction
There has been growing interest in generating electric energy from hydrogen due to serious concerns about climate change brought about by greenhouse gases. The combination of a methane-steam reformer and polymer electrolyte membrane fuel cells (PEMFCs) has been considered as an effective distributed power source due to the development of compact and low-cost fuel processors and the scalability of the fuel cell system [1][2][3][4]. In this combined system, hydrogen-rich gas fed into the PEMFCs is produced via steam reforming and water-gas shift (WGS) reactions [3,4]. The fraction of CO in the reformate gas after the WGS process is approximately 0.5-1 vol.% and it should be reduced preferably to less than 10 ppm for the platinum catalyst in the anode of low-temperature PEMFCs; otherwise, a drastic deactivation of the fuel cell anodes by CO poisoning arises. Preferential CO oxidation (PROX: CO + 0.5 O 2 → CO 2 ) has been widely used for deep CO removal in a compact fuel processor combined with low-temperature PEMFCs [2][3][4]. Although PROX has been practiced in commercial compact fuel processors, this process requires air (oxygen) supply and cooling systems to avoid undesired loss of hydrogen, which make the fuel processors complex and costly [5,6]. On the other hand, as depicted in Figure 1, the CO-selective methanation (CO-SELMET: CO + 3H 2 → CH 4 + H 2 O) process is considered as an effective alternative because the direct hydrogenation of CO occurs over proper catalysts without an additional reactant supply. Thus, the CO-SELMET process can be a practical According to the published reports on CO-SELMET catalysis over the last few decades, nickel-and ruthenium-based catalysts have been attracting a great deal of attention [5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23]. Specifically, among the various experimental results, Ru-based catalysts (e.g., Ru/TiO 2 ) have shown consistently high activity and selectivity to CO [15][16][17][18][19]. Despite the remarkable performance of the Ru-based catalyst, Ni is considered commercially appropriate as a major component due to economic aspects. Consequently, highly efficient nickel-based catalysts, showing outstanding performance under the condition similar to the actual reformate gas, can be a more practical solution to the commercialization of the CO-SELMET process [5,6,13,[21][22][23].
Although a great deal of previous research studies on catalytic preparations and morphological analyses have been reported, kinetic studies on CO-SELMET catalysis have barely been carried out. In fact, a mechanistic kinetic model for the CO methanation reaction over ordinary nickel catalysts has intensively been investigated since the 1970s [24][25][26][27][28]. Then, recently, kinetic studies on the CO 2 methanation process, utilizing captured CO 2 or CO 2 -rich biogas, were also reported [29][30][31][32]. However, most of the previous research studies deal with independent CO and CO 2 methanation reactions under the condition where no steam exists in the feed gas. They are not applicable for developing global kinetic models of all the plausible reactions in the CO-SELMET process for deep CO removal from hydrogen-rich reformate gas because there must be competitive CO and CO 2 reactions, e.g., methanation and WGS reactions, due to the existence of CO, CO 2 and residual steam. Thus, the objective of this work is to develop global kinetic models regarding all plausible reactions in the CO-SELMET process. For an exemplary practice, we have investigated the kinetics of the reactions in the CO-SELMET process over the two different types of nickel-based catalysts suggested in our previous research studies for comparison: (1) the iridium-doped nickel catalyst [5] and (2) the nickel-zirconium composite catalyst [21]. In order to determine appropriate mechanistic kinetic models for the catalytic reactions, the Langmuir-Hinshelwood models have been adopted according to the previous research studies on reaction mechanisms and all the parameters of the kinetic models have been globally determined by the efficient optimization algorithm, combining both heuristic and deterministic methods.

Mechanistic Reaction Model and Kinetic Parameter Estimation Method
The CO-SELMET process for deep CO removal from the hydrogen-rich gas produced by fuel processors deals with three competitive reactions: WGS or reverse WGS (RWGS), CO and CO 2 methanation reactions, as follows: CO  It is generally accepted that CO methanation proceeds via adsorption of CO and H 2 on the active metal sites and dissociation followed by the hydrogenation of the adsorbed carbonaceous species [25][26][27][28]. However, the CO 2 methanation mechanism is explained in a different way: CO 2 is firstly adsorbed on the catalyst support surface, and then, the hydrogenation reaction follows possibly at the active metal-support interface so CO 2 adsorption does not compete with CO according to this explanation [33,34]. In addition to the two methanation reactions, WGS and RWGS reactions also occur in the CO-SELMET process due to the coexistence of CO, CO 2 and steam. Based on the chemisorption mechanisms and the reaction equilibrium theory, well-accepted mechanistic kinetic studies were already proposed in the 1980s [35,36].
To develop the kinetic models of all these possible reactions in the CO-SELMET process, we have adopted the Langmuir chemisorption theory so the rate Equations (1)-(3) are derived as reported in a number of published works since the 1980s [24][25][26][27][28][29][30][31]35,36]. Using the kinetic models and stoichiometry, the generation or consumption rate of each species can be expressed as the ordinary differential forms (Equations (4)-(9)). The balance Equations (4)- (9) are valid under the assumption that the only three reactions occur in the catalytic reaction system. This assumption is confirmed by calculating the carbon balance in the reaction experiments, which will be briefly discussed in the next section. In addition, it should be noted that the optimal kinetic parameters have to be globally determined because all the plausible reactions simultaneously occur. In other words, the reactions cannot separately occur: for example, if CO and H 2 are fed into the reactor for the methanation reaction, CO can also react with the steam (WGS) formed via the methanation reaction.
Then, the dry-basis compositions of the product gas can be calculated by Equation (10) and be compared with the experimentally measured data. Thus, the kinetic parameters should be optimally determined to minimize the differences between the predicted and the measured values. Thus, as mentioned above, the kinetic parameters are globally determined unlike a number of previous works focusing on the separate kinetic model. CO methanation rate [mol/h/g cat ]: WGS reaction rate [mol/h/g cat ]: r CO,wgs = − k wgs /P H 2 P CO P H 2 O − P H 2 P CO 2 /K eq,wgs CO 2 methanation rate [mol/h/g cat ]: dF CO dw = r CO = r CO,m + r CO,wgs Dry-basis composition at the outlet of the catalyst bed, calculated in the model [%]: For the kinetic experiments, a fixed-bed quartz reactor was used, and the pressure was maintained at the atmospheric level. Multiple experiments where the temperature and gas compositions were varied in the wide ranges (see Table 1) were carried out and the measured dry-basis molar composition of each species was used for the kinetic parameter estimation. Cases 1 and 2, CO and CO 2 methanation reactions, are separated and Case 3 is similar to the gas compositions of the actual reformate gas from an ordinary natural-gas-based fuel processor. Detailed information on the experiments, such as catalyst preparation, experimental apparatus, etc., is provided in Section 3. To define the kinetic parameters, an optimization problem is defined. The objective function for the kinetic parameter estimation is shown as Equation (11), which is an ordinary root means square error equation, and this complicated optimization problem has been solved using the genetic algorithm (GA), which is a widely used heuristic optimization method for parameter estimation because of the non-linearity and the extremely large solution space of the objective function [38,39]. However, the heuristic optimization methods cannot guarantee the optimality of the solutions and may require a number of ineffective iterations near the optimal solutions. Therefore, combination with a deterministic optimization method can provide synergic effects, reducing the aforementioned drawbacks of the GA when solving such high-dimensional optimization problems [38][39][40][41]. Thus, we have used the GA combined with the fmincon function in the Global Optimization Toolbox 8.0 of MATLAB R2017b [39] and the ordinary differential equations (ODE) were solved using the ode23t solver, which is the common MATLAB ODE solver for stiff functions. Thus, the 14 kinetic parameters for each catalyst were estimated with the experimental data. For the parameter estimation trials, approximately 400-1000 iterations were carried out by the GA to find the best solution. The detailed options for the kinetic parameter estimation are summarized in Table 2. Objective function: The experimental data and the kinetic model predictions for both Ir/Ni-MgO/γ-alumina and Ni 0.5 Zr 0.5 O 2 catalysts are illustrated in Figures 2 and 3, comparing the compositions of the carbonaceous species. It should be noted that the absence of side reactions other than the three main reactions mentioned above was confirmed by the carbon balance calculations made at all data points of these experiments.
To discuss the reaction phenomena and kinetic properties of the Ir/Ni-MgO/γ-alumina catalyst, in Case 1, the CO 2 methanation is considered negligible and the tiny amount of CO is consumed by both the hydrogenation and the WGS reactions. The experimental data show that the CO methanation reaction is readily accelerated from 190 • C (see Figure 2a). On the other hand, CO 2 is considered to be generated by only the WGS (CO + H 2 O → CO 2 + H 2 ) reaction in Case 1. Figure 2a shows that the CO 2 concentration starts to increase at or above 190 • C, but remains under 0.04 mol%. In Case 2, CO 2 methanation is expected to be dominant in the hydrogen-rich gas without CO feeding. For the sake of an efficient deep CO removal process where the residual CO centration has to be maintained below a few dozen ppm, the catalyst should be designed to suppress the CO 2 reaction, avoiding unnecessary consumption of hydrogen in the reformate gas. When using the Ir/Ni-MgO/γ-alumina catalyst, the increase in CH 4 concentration by the hydrogenation of CO 2 in the product gas is observed from 230 • C, despite the abundant amount of CO 2 and hydrogen, as shown in Figure 2b. This means that the CO 2 hydrogenation is efficiently suppressed under 230 • C. This is the most important attribute of the CO-SELMET catalyst for the deep removal of CO in the reformate gas [5][6][7][8][9][10]21,33,34]. Finally, in Case 3, the reactions proceed under conditions similar to the actual CO-SELMET process applied to the ordinary fuel processor having a methane-steam reformer followed by a WGS reactor. The catalyst selectively and efficiently converts the CO into CH 4 , minimizing the undesired hydrogen loss by CO 2 methanation up to 230 • C (see Figure 2c).
Such high activity and selectivity to CO molecules are also confirmed by the estimated kinetic parameters. A comparison between k m and k m2 values (reaction frequency factors at the active site of the catalyst for CO and CO 2 hydrogenation reactions, respectively) is shown in Figure 4a,b. They help understand the aforementioned mutually competitive hydrogenation reactions: the k m value of the Ir/Ni-MgO/γ-alumina catalyst has a far larger order of magnitude than its k m2 value in the whole temperature range, which means the CO hydrogenation rate is approximately 10 9 times faster than the CO 2 hydrogenation rate. In order to understand such a remarkable difference in the k m and k m2 values, the estimated kinetic parameters should be discussed as follows: in Table 3, it is found that the activation energy of the CO methanation reaction (Ea m = 185.647 kJ/mol) of the Ir/Ni-MgO/γalumina catalyst is much lower than the one of CO 2 methanation (Ea m2 = 272.479 kJ/mol). However, its adsorption parameters related to CO and CO 2 reveal the relatively stronger preference for CO 2 over CO (see Table 4), which may not be helpful to maintain the high selectivity of the catalyst to CO molecules. In other words, although the adsorption of CO 2 molecules is favorable, the reaction rate of CO hydrogenation at the active sites of the Ir/Ni-MgO/γ-alumina catalyst is excessively faster than the CO 2 hydrogenation. This is the way that the Ir/Ni-MgO/γ-alumina catalyst efficiently and selectively removes CO, avoiding the CO 2 reaction.     Following the same manner for the Ir/Ni-MgO/γ-alumina catalyst, the kinetic parameters for the Ni 0.5 Zr 0.5 O 2 catalyst are also determined, and the model prediction results and the experimental data are compared and shown in Figure 3. First of all, in Case 1, the Ni 0.5 Zr 0.5 O 2 catalyst readily removes CO by accelerating the hydrogenation reaction from 150 • C, faster than the Ir/Ni-MgO/γ-alumina catalyst, and the CO 2 concentration is nearly zero in the whole range of the temperature. This means that the RWGS reaction may not have occurred yet, or even if the RWGS reaction generates a small amount of CO 2 , it is rapidly converted into CH 4 by the hydrogenation reaction. The experimental data of Case 2 show the stronger activity to CO 2 compared with the Ir/Ni-MgO/γ-alumina catalyst (compare the CH 4 compositions in Figures 2b and 3b). The undesired hydrogen consumption by the CO 2 hydrogenation is the drawback of the catalyst. In spite of the strong activity to the CO 2 hydrogenation reaction, the catalyst removes CO under the condition similar to the actual reformate gas (Case 3), as shown in Figure 3c. As the temperature increases, the RWGS reaction (CO 2 + H 2 → CO + H 2 O) can produce CO and the CO molecules are probably converted into methane, resulting in a high concentration of methane in the product gas.
The aforementioned differences in kinetic parameters are also considered for the Ni 0.5 Zr 0.5 O 2 catalyst. The k m values of the Ni 0.5 Zr 0.5 O 2 catalyst are less than the ones of the Ir/Ni-MgO/γ-alumina catalyst (see Figure 4a), which means that the CO hydrogenation rate at the active sites of the Ir/Ni-MgO/γ-alumina catalyst is faster. On the other hand, the k m2 values of the Ni 0.5 Zr 0.5 O 2 catalyst are approximately 10 30 times higher than the ones of the Ir/Ni-MgO/γ-alumina catalyst, as shown in Figure 4b, which demonstrates the extremely stronger activity to CO 2 and the simultaneous decrease in the CO selectivity. The lower activation energy for the CO 2 hydrogenation reaction (E am2 ), shown in Table 3, corresponds to a strong activity to CO 2 .
In addition to the comparison of the surface reaction rates (k m and k m2 ), adsorption characteristics also contribute to the overall reaction rate. The strong activity of the Ni 0.5 Zr 0.5 O 2 catalyst to CO can be explained by its adsorption characteristics. The stronger adsorption energy of CO (∆H CO = −66.144 kJ/mol for the Ni 0.5 Zr 0.5 O 2 catalyst and −12.057 kJ/mol for the Ir/Ni-MgO/γ-alumina catalyst in Table 3) is able to increase the surface concentration of the activated CO molecules, resulting in high K CO values. The K CO can contribute to the acceleration of the CO hydrogenation reaction (see the reaction rate equation of r CO,m ). This is the key difference between the two catalysts in controlling the CO-removal reaction rate: the Ir/Ni-MgO/γ-alumina catalyst has the fast and selective surface reaction and the Ni 0.5 Zr 0.5 O 2 catalyst has a stronger affinity to the CO molecules.
In terms of the accuracy of the kinetic model for the catalyst, the predicted values based on the optimally estimated parameters show good agreement with the experimental data, as shown in Figures 2 and 3 and Figure S1. Thus, the estimated kinetic parameters are considered to provide acceptable analyses for the reaction characteristics. As discussed so far, the estimated kinetic parameters will be compared with other published works dealing with various nickel-based methanation catalysts in the next section, so the validity and physicochemical properties of the estimated kinetic parameters can be discussed.

Comparisons of the Kinetic Parameters with Other Ni-Based Catalysts
The kinetic parameters of the two developed nickel-based catalysts showing the outstanding CO-removal performance are compared with published works dealing with nickel-based CO and/or CO 2 methanation catalysts [24,28,30]. Though the catalysts differ in nickel content, support and preparation method, they use the same active metal. Thus, some kinetic parameters can be compared with the results of this study, despite the fact that somewhat different kinetic models and related assumptions are employed.
Firstly, in terms of the CO methanation reaction, pre-exponential factors cannot be directly compared because the published papers adopt different kinetic models in detail, e.g., definition of the denominator in the kinetic models based on the Langmuir chemisorption theory, the power of the partial pressure of CO, etc. Instead, the activation energies, which is one of the most important catalytic properties related to the temperature dependence, can be essentially compared. As shown in Table 3, in the published works, the activation energy values of the CO methanation reaction (Ea m ) are in the range of 100-230 kJ/mol. The Ea m values of the Ir/Ni-MgO/γ-alumina and the Ni 0.5 Zr 0.5 O 2 catalysts are approximately 190 kJ/mol, so the estimated parameters are considered feasible. Secondly, in this way, the activation energies (Ea m2 ) of the CO 2 hydrogenation reaction can also be compared: they increase in the order of Ni 0.5 Zr 0.5 O 2 < Ni/γ-alumina < Ir/Ni-MgO/γ-alumina. The relatively strong activity of the Ni 0.5 Zr 0.5 O 2 catalyst to CO 2 is confirmed in the experiments as discussed in the previous section; the Ir/Ni-MgO/γ-alumina catalyst, on the other hand, efficiently suppresses the CO 2 reaction, resulting in higher selectivity to CO. Although all catalysts use nickel as an active species, it has been confirmed that the activation energy values (Ea m and Ea m2 ) are varying in the wide range. Such difference probably depends on the content of nickel, dispersion degree, support, etc., and as a result, these have direct effects on the activity and selectivity to both CO and CO 2 molecules. Lastly, the kinetic parameters for the WGS reaction have also been compared. It is obvious that all the Ea WGS values of the different Ni-based catalysts remain similar in Table 3.
Regarding adsorption characteristics, only the adsorption energies of the CO molecule with respect to the different nickel-based catalysts can be compared due to the lack of information on other adsorption/desorption parameters. All the ∆H CO values are in a similar order of magnitude so the parameters determined in this study seem to be in an acceptable range. Comparing ∆H H2 and ∆H H2O with Xu and Forment's work [35], the difference between the two catalysts is relatively notable. Xu and Froment specifically estimated the kinetic parameters in the steam-methane reforming process where the reaction temperature is much higher (generally around 700-800 • C) and the partial pressures of species (CH 4 , CO, CO 2 and steam) are far different from the selective CO methanation process. In addition, Xu and Froment used the nickel supported on Mg spinel catalyst. Except the active metal, the support, promoter and preparation methods are different from the two catalysts that we developed. These aspects may affect the competitive adsorption characteristics, resulting in different adsorption parameters.

Materials and Methods
In this work, the two aforementioned outstanding catalysts, Ir/Ni-MgO/γ-alumina and Ni 0.5 Zr 0.5 O 2 catalysts, were prepared according to the published works [5,21]. To prepare the Ir/Ni-MgO/γ-alumina catalyst, first of all, Ni-MgO/γ-Al 2 O 3 catalysts were prepared by the deposition-precipitation method using Ni(NO 3 ) 2 (H 2 O) 6 (Aldrich), Mg(NO 3 ) 2 (H 2 O) 6 (Aldrich) and γ-Al 2 O 3 (PURALOX SCFa-140, SASOL) supports. The precipitation pH of the solution was adjusted at 11 with 10 wt.% KOH solution as a precipitation agent. The contents of the nickel and the magnesium were 30 wt.% and 5 wt.% in their metallic state, respectively. The solution was stirred at 80 • C for 24 h and then, it was filtered and washed several times with deionized water. The filtered powder was dried at 80 • C overnight and then, it was calcined at 400 • C for 1 h, where the rate of the temperature increase was 1 • C/min. After that, IrCl 3 ·xH 2 O (Aldrich) was used as Ir precursor so 1 wt.% was impregnated onto the nickel catalyst by the incipient wetness method. Then, it was calcined again in the same manner mentioned above [5]. The Ni 0.5 Zr 0.5 O 2 catalysts was prepared using ZrO(NO 3 ) 2 (Aldrich) as zirconium precursor and Ni(NO 3 ) 3 ·9H 2 O (Junsei) as Ni precursor, where the molar ratio of Ni and Zr is 1:1. They were taken and dissolved separately in a minimum amount of distilled water. Urea (1 mol per metal ion) was also dissolved separately in a minimum amount of distilled water. Then, all these three solutions were mixed thoroughly by ultrasonicating for 2 min. The mixed solution was then transferred to a crystallizing dish, and it was heated on a laboratory hot plate at 200 • C. The resulting solid products were further heated at 400 • C for 1 h to obtain the nanostructured Zr 0.5 Ni 0.5 O 2 catalyst powders [21].
For the experiments, a fixed-bed quartz reactor (outer diameter: 0.5 inch) was used and the pressure was maintained at the atmospheric level. The prepared catalyst powder was pressed into a tablet and then crushed and sieved. The size of the sieved particles was less than 75 µm and the volume of the catalyst bed was 0.8 mL. For accurate analysis of the intrinsic activity, the particle size was studied with the pre-experiments so it is confirmed that approximately 75 µm powder catalysts had no (or minimized) diffusion resistance. The reaction temperature was monitored by a thermocouple placed at the center of the catalyst bed and the bed temperature is thoroughly controlled by the temperature-programmed PID controller (KP-1000, Chino, Tokyo, Japan). Before the reaction, the catalyst was reduced under the optimized reduction conditions [5,21]. The product gas passed through a water trap to condensate and remove the residual steam, and then, the gas compositions were analyzed with an on-line gas chromatograph (Agilent 6890 GC, Agilent Technologies, Inc., Santa Clara, CA, USA) equipped with both a thermal conductivity detector (TCD, for N 2 ) and a flame ionization detector (FID, for CO, CO 2 and CH 4 ) with a catalytic methanizer. To make the data consistent, the reaction temperature was held at each point for one hour so that the gas compositions measured by the GC remained stable. The concept of the experimental apparatus used in this work is depicted in Figure 5.

Conclusions
The kinetic models of the two different nickel-based catalysts for the CO-SELMET process have been developed based on the Langmuir chemisorption theory using an efficient optimization method combining the GA and the deterministic algorithm. The globally determined parameters showed acceptable predictions of the overall reaction rates in wide ranges of reaction temperature. In addition, comprehensive explanations on the different catalytic characteristics for the catalysts have been provided, considering the kinetic parameters related to each plausible reaction. Thus, the differences in mechanistic kinetics, such as the activity and selectivity to CO and CO 2 molecules, for the reactions in a mutually competing relationship, are clearly explained. Accordingly, the methodology of this kinetic study can be useful to (1) understand catalytic properties, (2) optimize or modify the efficient catalysts (3) determine optimal operating windows with respect to the catalysts and (4) optimize the deep CO removal reactor comprising efficient CO-SELMET catalysts.