Ind. Eng. Chem. Res. 2008, 47, 4675–4682
4675
Combinatorial Assessment of the Activity-Composition Correlation for Several Alloy Nanoparticle Catalysts Xiajing Shi,†,‡ Jin Luo,† Peter N. Njoki,† Yan Lin,† Ta-Hsuan Lin,‡ Derrick Mott,† Susan Lu,*,‡ and Chuan-Jian Zhong*,† Department of Chemistry, and Systems Science and Industrial Engineering, State UniVersity of New York at Binghamton, Binghamton, New York 13902
The screening of catalysts with desired catalytic activity and selectivity for electrocatalytic fuel cell reactions is a time-consuming process. One approach to address this problem is to apply combinatorial analysis techniques. In this paper, we present the results of an investigation of the application of systematic statistical analysis techniques such as analysis of variance (ANOVA) analysis and regression modeling for the development of effective screening methods of bimetallic and trimetallic nanoparticle catalysts. Based on several sets of experimental data from the chosen catalysts, empirical models derived from statistical analysis techniques were first built to fit the experiment results for each of the electrocatalytic parameters such as catalytic peak current, peak potential, Tafel slope and mass activities. These parameters were expressed as a function of catalyst component proportion variables and process variables. The adequacy of the chosen models is verified with residual analysis. The catalyst properties were also analyzed using a response surface approach. The statistical analysis results from the available experiment data provided useful information to aid the understanding of the relationship between the catalyst activities and compositions, which may provide guidance for experimental design toward discovery of catalysts with desired activity and selectivity. Introduction The design of a catalyst with desired activity and selectivity has long been a challenging task for experimental researchers, which is especially true for the design of nanostructured catalysts.1,2 The application of statistical analysis techniques such as combinatorial analysis and optimization in catalyst data analysis has been an increasingly important approach to speeding up the process.3,4 The purpose of this article is to investigate the application of statistical analysis techniques in alloy nanoparticle catalyst design and experimental analysis. The statistical techniques are widely applied in industry to help optimize products composition, such as detergents and food products, to make them meet a number of performance criteria and at the same time minimize the designing cost. Different combinatorial analysis methods have been applied in drug design and catalyst screening.5–8 We have been working on a series of bimetallic and trimetallic nanoparticle catalysts for fuel cell reactions.9–14 Some of the bimetallic and trimetallic nanoparticle catalysts display unprecedented electrocatalytic activities for fuel cell reactions such as methanol oxidation reaction and oxygen reduction reaction. While the synthesis and processing protocols for some promising catalysts have been established, the effective screening of the catalysts with desired activity and selectivity for fuel cell reactions is a time-consuming process. To aid the screening of the multimetallic nanoparticle catalysts, the application of combinatorial analysis and statistic methods to the processing of experimental data provides an important tool for delineation and optimization in catalyst design and preparation. In the application of statistic methods in catalyst design, there are two main parts: experimental design and result analysis. This * To whom correspondence should be addressed. E-mail: cjzhong@ binghamton.edu (C.J.Z.);
[email protected] (S.L.). † Department of Chemistry, State University of New York at Binghamton. ‡ Systems Science and Industrial Engineering, State University of New York at Binghamton.
article mainly focuses on the application in result analysis, in which the properties of a catalyst are predicted with the empirical model developed by fitting the experimental data with statistical models. The model can also be used to assist our experimental design toward finding the desired catalyst. In addition, specific properties of most interest are optimized on the basis of the empirical model subjected to constraints on other less important properties. In this article, the experimental results obtained for some alloy nanoparticle catalysts are first described. This description is followed by discussion of statistic methods applied for catalyst model fitting, model validation, and optimization. Finally, the experimental and modeling data are discussed with several sets of catalyst data. Experimental Data There are four basic sets of experimental data that were analyzed in this work. These data are from the data for methanol oxidation reaction (MOR) and oxygen reduction reaction (ORR) at several different catalysts,9,10 as summarized in Tables 1–4. These data are presented in terms of their composition (Au, Pt, Fe) in the bimetallic or trimetallic nanoparticles and their responses. The responses are extracted from the electrocatalytic activities expressed by Tafel slopes (T.S., mV/dec), peak currents (Ip, mA mgMt-1 cm-2; Mt: total metal weight) and Table 1. Tafel Slope (T.S.) and Peak Current (Ip) for MOR at AuPt catalyst in 0.5 M KOH composition
response
Au
Pt
0 100 82 72 68 60 56 97
100 0 18 28 32 40 44 3
10.1021/ie800308h CCC: $40.75 2008 American Chemical Society Published on Web 06/10/2008
T.S.
Ip
138 169.5 156 165 154 106
8092 349 7545 2491 6254 4482 729 1869
4676 Ind. Eng. Chem. Res., Vol. 47, No. 14, 2008 Table 2. Tafel Slope (T.S.) and Kinetic Current (Ik) for ORR at AuPt catalyst in 0.5 M H2SO4 composition Au 56 65 97 72 0
response Pt
T.S.
Ik
44 35 3 28 100
96.3 112.3 161.2 108.3 86.1
76 25 5 14 292
Table 3. Ip and Ep for MOR at AuPtFe catalyst in 0.5 M KOH composition
response
Au
Pt
Fe
Ip
Ep
1 0.9 0.8 0.68 0.6 0.41 0 0.5 0 0.01 0.38 0
0 0.1 0.2 0.32 0.4 0.59 1 0.25 0.4 0.08 0 0
0 0 0 0 0 0 0 0.25 0.6 0.91 0.62 1
1744 1982 1357 1079 2582 4699 1696 1300 762 1746 1429 802
-145 -177 -197 -178 -200 -245 -179 -175 -255 -205 -168 -384
Table 4. Mass Activity (M.A.) for MOR at PtNiFe catalyst Treated at Different Temperatures in 0.5 M H2SO4 composition Pt
Ni
Fe
response M.A.
process variable temperature
32.67 31 29.29 26.73 26.73 30 26 29 28.3 25 31 28 29.7 29.13 28.57
27.72 34 36.36 37.63 37.63 29 20 20 18.3 36 34 40 28.7 28.16 27.62
39.61 35 34.35 35.64 35.64 41 54 51 53.4 39 35 32 41.6 42.71 43.81
2.8 4.7 3.15 3.3 1.8 3.6 3.1 4.4 4.2 4 3.7 4.4 3.5 3.9 4
400 400 400 400 400 400 500 500 500 500 500 500 500 500 500
peak potential (Ep, mV vs Ag/AgCl 8 sat’d KCl) for MOR, and Tafel slopes, kinetic currents (Ik, mA mgMt-1cm-2), or mass activity (M.A., mA mgMt-1cm-2) for ORR. Considering the case of AuPt catalysts as an example, whereas the presence of Au in Pt increases the lattice distance of Pt, the higher electronegativity of Au than that of Pt could cause an increase in the amount of change being transferred from Pt to Au, and consequently the increase in the d-orbital vacancy in the PtAu. On the basis of the dependence of MOR M.A. on the bimetallic composition,1 the composition was found to significantly modify the electrocatalytic properties of both Au and Pt. One of the most significant findings is that the M.A. in the alkaline electrolyte exhibits a maximum around 65- 85%Au, in contrast to the small and gradual increase from no activity of Au to high activity of Pt in the acidic electrolyte. The variation of T.S. appears to be relative small (∼120 mV/dec). The display of the maximum M.A. for AuPt/C catalysts comparable or higher than that for Pt/C catalyst in the alkaline electrolyte is remarkable in view of the fact that only 15-35% Pt was present in the catalyst. The Au atoms surrounding Pt atoms in the AuPt alloy are believed to have played an important role in either removing the intermediate CO-like species and/ or providing oxygenated species in the methanol oxidation
process. The bifunctional activity in the alkaline electrolyte involves the participation of CO and OH adsorption on Au sites in the catalytic reaction of Pt in the alloy via a combination of reaction steps, including the adsorption of MeOH on Pt followed by dehydrogenation, the formation of intermediate COad/Pt, the transfer of COad/Pt to neighboring Au atop sites forming COad/ Au, which is supported by the favorable adsorption of CO on Au nanoparticles known from both experimental measurements and theoretical calculations, the formation of OHad/Au or surface oxides on gold. For ORR, the experimental data allow us to compare their mass activities for catalysts with different bimetallic composition in both alkaline and acidic electrolytes. The mass activities with units of mA cm-2 mg-1 are given in this report. The comparison of the M.A. for these catalysts with different bimetallic compositions is based on the Ip extracted from the RDE data. We note that the changes of the T.S. were found to be relatively small compared with the changes in the mass activities. The slopes determined for Au/C and AuPt/C catalysts were in fact quite close to the value obtained for Pt/C and thin-film Pt electrodes9 in the alkaline electrolyte (52-60 mV/dec). In the acidic electrolyte, the slope for Au/C (144 mV/dec) was larger than that for Pt/C (86 mV/dec), whereas the slope for AuPt/C (96-161 mV/dec) fell approximately in between those from Au/C and Pt/C. The mass activities were found to be strongly dependent not only on the bimetallic composition (AumPt100-m) but also on the nature of the electrolyte. The strong dependence of the mass activities on the bimetallic composition in the alkaline electrolyte is evident by the exhibition of a maximum in the composition region of 60-80% Au, which is higher than those for Pt/C and Au/C by a factor of 2-3. This finding is in contrast to the gradual increase of M.A. in the acidic electrolyte displaying a relatively smooth transition from the low activity of Au to the high activity of Pt. In acidic solution, Au is not capable of proving adsorption sites for -OH and the electrocatalytic activity is thus rather low. Although the M.A. in the acidic electrolyte could reflect a collective effect of the activities from both Au and Pt, the concurrence of a maximized activity in the 60-80% Au region in the alkaline electrolyte suggests the operation of a remarkable synergistic effect. Because this composition range represents an alloy structure in which Pt atoms are surrounded by Au atoms, the possibility of an optimal fraction of Au atoms surrounding Pt could have played an important role in the observed activity maximum. Statistical Analysis Methodologies The goal of applying statistical analysis methods in catalyst components proportioning in this study is to find empirical models for the response/properties Y of AuPt alloy nanoparticle catalyst in terms of the catalyst component proportion. The model is built by fitting the available experimental data using regression models. Thus the properties of catalyst for other catalyst composition can be predicted with the empirical model. In addition, the model can be used to locate the optimum composition which can yield desired catalyst properties. In dealing with multiple properties, a specific property of the most interested could be optimized subject to constraints on other properties with the model. The general steps of applied statistical analysis here can be summarized as the following two major steps: First, a set of trial experiments covering a chosen range of proportions for each component based on expert experience are carried out. Second, the results are analyzed using statistical analysis techniques. The empirical models are chosen to fit the data for
Ind. Eng. Chem. Res., Vol. 47, No. 14, 2008 4677
each property criterion using analysis of variance (ANOVA) techniques, in which each response/property such as Ip, Ep, or T.S. is expressed as a function of catalyst component proportion variables and process variables. The adequacy of the chosen model is then verified with residual analysis. Finally, the target catalyst properties are optimized using response surface plot (contour plot) approach. Model Fitting. Scheffe’s first three order polynomial functions15 are used as the empirical models for data fitting in this study because of their simple form, well-known and understood properties, and moderate flexibility of shapes, along with being independent of underlying metrics and computationally easy to use.16 The linear polynomial model is given as15,16 n
Y)
n
∑ a X + e, where∑ X ) 1 i i
(1)
i
i)1
i)1
Here, Y and Xi denote the response/properties of the catalyst and catalyst component proportion variables, respectively. The error term e is assumed to be normal distribution (N (0, σ2)) and independent of variables Xi. It is the same here, with the quadratic polynomial model and cubic polynomial model represented as below15,16 n
Y)
n
∑ a X + ∑ ∑ b X X + e, where∑ X ) 1 i i
ij i j
i)1
i
j