Ind. Eng. Chem. Res.

1991,30, 1819-1824 1819

Selective Oxidation of n -Butane to Maleic Anhydride. 1. Optimization

Studies
Shyamal K. Bej and Musti S. b o *
Department of Chemical Engineering, Indian I n s t i t u t e of Technology, K a n p u r 208 016, U.P.,
India

Optimization studies have been carried out for the selective oxidation of n-butane to maleic anhydride
(MA) over vanadium-phosphorus oxide (VPO) catalysts promoted with Mo and Ce. It is known
that Mo increases the selectivity to MA while Ce increases the conversion of n-butane. Response
surface methodology has been used for finding out the optimum catalyst composition and optimum
values for the process variables, viz., inlet feed composition, reaction temperature, and contact time
for the reaction. It is observed that a catalyst having a P / V atomic ratio equal to 1.08, Mo/V weight
ratio of 0.06, and Ce/V weight ratio of 0.022 gave the maximum yield of MA (49.5%) a t a temperature
of 423 "C and an inlet n-butane concentration of 1.75% of air and n-butane mixture.

1. Introduction Table I. First-Order Response S u r f a c e Strategy (First

Move): Levels of Factors
Maleic Anhydride (MA), an industrially important lower higher
chemical, is generally produced by the selective oxidation factor code base level level (-) level (+)
of either benzene or n-butane. For the past few years, due n-butane concn in the x1 0.7 0.4 1.0
to higher prices of benzene and restrictions imposed on feed, vol % of
using benzene from a pollution point of view, C4fractions air-n-butane mixture
were found to be better feedstocks for the production of reaction temp., "C x2 375 360 390
this important chemical. Generally, vanadium-phosphorus Mo/V wt ratio %3 0.09 0.04 0.14
oxide (VPO)with or without promoter is used as a catalyst Ce/V wt ratio ~1 0.04 0.02 0.06
for this reaction. W/F,g/(mg-mql/s) ~5 12.5 10.0 15.0
P/V atomic ratio ~6 1.05 0.95 1.15
The selective oxidation of C4fractions has been reviewed
by Hucknall (1974), Varma and Saraf (1979), Hodnett
(1985), and most recently Centi et al. (1988). Factors like concentration, respectively, on the intrinsic yield of MA.
the method of preparation of VPO catalyst, phospho- Integral reactor data were obtained with a powdered
rus/vanadium (P/V) atomic ratio of the catalyst, tem- catalyst. One of the aims was to obtain a suitable catalyst
perature, and atmosphere of calcination of the catalyst composition which will be used in parts 2 and 3 of this
affect the selectivity toward MA. Several investigators series under differential reactor conditions for modeling
(Varma and Saraf, 1978; Cavani et al., 1984,1985a; Centi the intrinsic kinetics of this reaction. While extrapolating
et al., 198413) have extensively studied the influence of the results of this study to the operational design of
these variables on the selective oxidation of 1-butene to practical reactors, caution should, however, be exercised.
MA. Similarly, the influence of different factors on the Methods of obtaining the nonisothermal effectiveness
selective oxidation of n-butane to MA has been studied factors are available in literature (Smith, 1981). Also the
by a number of investigators (Cavani et al., 1985b; Centi extrapolation of these results to the region, where higher
et al., 1984a; Wenig and Schrader 1986; Hodnett and values of n-butane conversion are prevailing, should be
Delmon, 1985; Buchanan and Sundaresan 1986; Moser and carefully avoided without any evidence from experimental
Schrader 1985). support. We have selected molybdenum and cerium as
Different promoters have been used in the VPO catalysts promoters since cerium was reported to increase the con-
so as to improve the yield of MA. Most of the information version of n-butane while molybdenum is expected to en-
on promoted catalysts is in patented form. Hodnett (1985) hance the selectivity to MA.
had summarized the information on the activity and se-
lectivity (to MA) of promoted VPO catalysts.
Though there are many publications regarding the 2. Experimental Section
chemistry of VPO catalysts and possible mechanistic 2.1. Catalyst Preparation. The catalysts used in the
pathways for this reaction, one finds hardly any infor- present investigation were prepared in an organic medium.
mation regarding the optimization of all the variables to- Catalysts prepared in an organic medium have greater
gether to get the maximum yield of MA. A detailed surface areas as compared to catalysts prepared in an
one-dimensional, heterogeneous model for the selective aqueous medium. The method of catalyst preparation was
oxidation of n-butane to MA in fixed-bed reactors ac- the same as that reported by Katsumoto and Marquis
counting for interfacial and intraparticle gradients has been (1979). Vanadium pentoxide was suspended in a 1:3 (v/v)
presented by Wellauer et al. (1986). In the present in- mixture of allyl alcohol and isobutyl alcohol. Allyl alcohol
vestigation, we have tried to find out the optimum values was used as a reducing agent and isobutyl alcohol was used
of the variables to obtain the maximum yield of MA over as a solvent. The mixture was refluxed at 115 "C for 2 h
VPO catalysts promoted by molybdenum and cerium under constant stirring. Then it was cooled to 40 "C. The
added simultaneously under negligible interparticle and proper amount of orthophosphoric acid (needed to main-
intraparticle mass transfer resistances with a view to obtain tain the desired P/V atomic ratio) was mixed with isobutyl
a better insight into the effects of catalyst composition, alcohol and then added to the slurry of reduced vanadium.
reaction temperature, contact time, and feed n-butane The whole mixture was then again refluxed for 2 h at 115
"C under constant stirring. After the mixture was cooled
* To whom correspondence should be addressed. to 40 "C, molybdic acid was added to it while it was stirred
0888-5885/91/2630-1819$02.50/0 0 1991 American Chemical Society
 1820 Ind. Eng. Chem. Res., Vol. 30, No. 8, 1991
Table 11. The 2er First-Order (First Move) Design Mntrix nnd Vnluee of t h e Reeponses
factors in coded form values of responses
expt no. *I X2 x3 x4 x5 ( X I X 3 ) x6 (xIx4) X S Y
1 - - - - + + 20.2 90.4 18.3
2 +
-
- -
-
-
-
-
+
-
+
13.9 91.8 12.8
3 + 26.7 87.1 23.2
4 + + - - - - 24.2 78.4 19.0
5 - - + - - + 3.5 98.6 3.4
6 + - + - + - 19.2 84.4 16.2
7 - + + - - + 13.4 96.3 12.9
8 + + + - + - 25.8 84.8 21.9
9 - - - + + - 21.1 70.8 14.9
10 + - - + - + 16.2 81.2 13.1
11 - + - + + - 31.4 65.3 20.5
12 + + - + - + 21.9 84.6 18.5
13 - - + + - - 9.3 78.1 7.3
14 + - + + + + 20.5 79.7 16.3
15 - + + + - - 16.1 76.5 12.3
16 + + + + + + 29.4 72.1 21.2
Rewat Trials
17 + - + - + - 18.7 83.2 15.5
18 + - + - + - 19.9 86.1 17.1
19 + - + - + - 19.5 85.8 16.7

Table 111. Firet-Order Responee Surface Methodology

(Firet Move): Test for Model Adequacy
Conversion
j = 19.551 + 1 . 8 2 4 ~+~4 . 0 7 1 ~- ~2 . 3 8 5 ~+~ 1 . 1 8 0 ~+~ 4.732 xg -
0.582~8 (4)
ANOVA
deg of
source sum of sauares freedom mean sauare
due to regression 794.42 6
residual 17.83 12
pure error 0.74 3 0.246
lack of fit 17.M 9 1.899
total 812.55 18
Fdcd = 7.72; Fom(e,3,
= 8.81; model is adequate
Selectivity
= 82.523 - 0.371~1- 1.862~2+ 1.301X3 - 6.474~4- 3.176~6+ valve
3.739x, (5) Figure 1. Schematic diagram of the experimental setup.
ANOVA
deg of in an atmosphere of flowing air.
source sum of squares freedom mean square 2.2. Experimental Setup. The schematic diagram of
due to regression 1202.58 6 the experimental setup is given in Figure 1. Compressed
residual 73.94 12 air was allowed to pass through a sodium hydroxide tower
pure error 5.26 3 1.753 (for removing small quantitiea of carbon dioxide) first and
lack of fit 68.68 9 7.631
then through a drying tube. The flow of air was controlled
total 1276.52 18
F,, = 4.35; F0,,e,3, = 8.81; model is adequate by a needle valve and measured by a rotameter. Similarly
the flow of high-purity n-butane was controlled by a needle
Yield valve and measured by a soapbubble flowmeter. This was
9 = 15.775 + 1.611~1+ 2.970~1- 1.77413 - 0.098~4+ 3.299~6+ then passed through a drying tube and finally mixed with
0.121~0 (6)
air in a mixing chamber. The mixture was then preheated
ANOVA and fed to the reactor. The reactor w a a stainless steel
deg of tube of 1 cm i d . and a total height of approximately 45
source sum of squares freedom mean square cm, which consisted of inlet and outlet calming sections
due to regression 415.30 6 and the reactor section. Two Chromel-Alumel thermo-
residual 21.42 12 couples were used to measure the temperature of the
pure error 1.348 3 0.449 catalyst bed, The product gases were then bubbled
lack of fit 20.073 9 2.230 through water so as to dissolve MA and then passed
total 436.72 18
through a condenser. The temperatures of the abeorber
Fdd = 4.97; Fo,mce,3,
= 8.81; model is adequate and the condenser were kept at a 10 and 6 O C , respectively,
by passing refrigerated water through their jackets. The
noncondensed gases were allowed to pass to a gas chro-
constantly. Then the slurry was filtered and ceric am- matograph for analysis.
monium nitrate was added to the solid. Both molybdic 2.3. Experimental Procedure. A known amount of
acid and ceric ammonium nitrate were added in prede- catalyst (in the form of fine powder) diluted with silica (1:l
termined quantities so as to maintain the desired Mo/V weight ratio), was taken in the reactor and sandwiched
and Ce/V weight ratios. The solid mass was then dried between silica at the exit and inlet sections. Air was peeeed
at 150 "C for 12 h, and it was calcined at 450 "C for 1 h at the rate of 100 cm3/min, and the reactor was heated in
 Ind. Eng. Chem. Res., Vol. 30,No. 8, 1991 1821
Table IV. Calculation of the Path of Steepest Ascent
factor
X1 x2 x3 x4 XS x6
base level 0.1 375 0.09 0.04 12.5 1.05
unit 0.3 15 0.05 0.02 2.5 0.1
estimated slope (b) 1.611 2.97 -1.7745 -0.0981 3.2988 0.1213
unit X b 0.4833 44.55 -0.0887 -0.00196 8.247 0.012
change in level per 6 OC change in x 2 0.065 6.0 -0.012 -0.00026 1.11 0.0016
Path of Steepest Ascent As Represented by a Series of Trial Points
exDt no. trial Doint X1 x., XQ XA XI. XR

20 1 0.165 381 0.078 0.0397 13.61 1.0516

21 2 0.830 387 0.066 0.0394 14.72 1.0532
22 3 0.895 393 0.054 0.0391 15.83 1.0548
23 4 0.960 399 0.042 0.0389 16.94 1.0564
24 5 1.025 405 0.030 0.0386 18.05 1.0580
25 6 1.090 411 0.018 0.0383 19.16 1.0596

Table V. Results of Experiments along the Path of and then cooled in a condenser (kept at 6 "C). The
Steepest Ascent products which were dissolved in water were analyzed by
expt no. trial point exptl yield predicted yield gas chromatography using a 2-m-long Porapak-QS column
20 1 20.3 19.1 and a flame ionization detector. The product was found
21 2 23.9 22.6 to be entirely MA. The noncondensed gases were analyzed
22 3 29.9 26.1 by use of two different columns, viz., Porapak-Q and
23 4 37.1 29.5 molecular sieve 5A. The hydrocarbons and carbon dioxide
24 5 38.5 32.9 were separated in a 2-m-long Porapak-Q column and an-
25 6 38.7 36.4 alyzed by use of a thermal conductivity detector. Oxygen,
nitrogen, and carbon monoxide were separated in a 2-m-
Table VI. First-Order Response Surface Strategy (Second
Move): Levels of Factors long molecular sieve 5A column and analyzed with the use
of a thermal conductivity detector. After all the analyses
lower higher
factor code base level level (-) level (+) were completed, checks on carbon balance were done. In
all cases, MA, carbon dioxide, and carbon monoxide were
n-butane concn in the rl 1.3 0.9 1.1 found to be the only products.
feed, vol W of
air-n-butane mixture
reaction temp, "C XZ 405 390 420 3. Results and Discussion
Mo/V wt ratio x3 0.04 0.01 0.07
Ce/V wt ratio x4 0.02 0.01 0.03 All experiments were carried out in the presence of
W/F,g/(mg-mol/s) ~6 18.5 15.0 22.0
negligible intra- and interparticle resistances. The absence
P/V atomic ratio xg 1.125 1.o 1.25
of intraparticle resistance was verified by comparing the
rates of MA formation and n-butane consumption using
a controlled manner. When the desired temperature was the catalyst in the form of pellets and fine powder. The
reached, air flow was controlled according to the require- interparticle transport effect was checked by changing the
ment and n-butane was allowed to pass through. The feed rate at a constant contact time of the reactants. The
reaction was allowed to proceed for about 2 h when steady difference between the inlet and outlet temperatures of
state was reached. At that stage the products were with- the catalyst bed was found to be not more than 1 "C.
drawn and analyzed. This study has been carried out with a view to determine
2.4. Analytical Methods. The reactor effluent was the effects of different variables on the yield of MA and
allowed to pass through a water bubbler (kept at 10 "C) their optimum levels for obtaining the maximum yield of
Table VII. The p*First-Order (Second Move) Design Matrix and Values of the Responses
factors in coded form values of responses
expt no. xl X2 x3 x4 %6 (Xlx3) x6 (xlx4) X S Y
26 - - - - + 51.3 + 74.5 38.2
21 + - - - - - 59.5 68.2 40.6
28 - + - - + +- 66.4 63.2 42.0
29 + + - - - 75.1 51.5 43.5
30 - - + - - 39.5 + 92.3 36.5
31 + - + - + - 62.8 66.5 41.8
32 - + + - - 51.2 + 82.1 42.0
33
-
+ -
+ + - +
- 74.6 54.8 40.9
34
-
- + -
+ - 73.8 53.2 39.3
35 -+ -
- + + 57.5 70.5 40.5
36 + + + - 89.8 41.8 37.5
37
-
+ -
+ - + - + 71.3 57.5 41.0
38 + + - - 56.5 72.3 40.8
39 +- - + + +- +- 60.4 69.6 42.0
40 + + + 72.6 60.3 43.8
41 + + + + + + 74.5 56.8 42.3
Repeat Trials
42 + + -
-
-
-
-
-
-
-
76.9 56.7 43.6
43 + + 76.5 57.9 44.3
44 + + - - - - 74.8 58.5 43.1
1822 Ind. Eng. Chem. Res., Vol. 30, No. 8, 1991

Table VIII. First-Order Response Surface Strategy MA. We have adopted response surface methodology
(Second Move): Test for Model Adeauacv ~~
(RSM) for this purpose. RSM is discussed extensively by
Conversion Kittrell and Erjavec (1968), Box et al., (1978), Davies
4 = 64.837 + 2.210~1+ 7.175X2 - 3.325~3+ 4.712~4+ 4.362~5- (1978), and Khuri and Cornel1 (1987).
5.825~6 (7) RSM coltsists of several steps. The first step is to design
ANOVA a set of experiments and conduct them to get reliable
deg of
values of the response. Factorial and fractional factorial
source sum of squares freedom mean square designs are used for this purpose. The details of these
due to regression 2590.28 6
designs are discussed by Box et al. (1978) and Rao and
residual 32.76 12 Iyengar (1984). The second step is to propose a suitable
pure error 2.587 3 0.862 mathematical model to fit the experimental data and test
lack of fit 30.173 9 3.352 for model adequacy through lack-of-fit F testa (Draper and
total 26 33.0 4 18 Smith, 1981). The third and last step is to find out the
F&d = 3.89; F0.05(9,3) = 8.81; model is adequate optimum conditions of the independent variables which
will produce the maximum (or minimum) value of the
Selectivity
3 = 65.025 - 2.287~1- 6 . 0 6 2 ~+~4.425X3 - 5.025~4- 4.612r5 + response. A t a point that is remote from the optimum,
5.386~8 (8) there is little curvature in the true response surface and
first-order models will be satisfactory. In the vicinity of
ANOVA the optimum, higher order models will be required due to
deg of the presence of curvature in the response surface.
source sum of sauares freedom mean souare RSM is sequential in nature. Some experiments are
due to regression 2367.67 6 carried out, valuable information is gathered, and the next
residual 8.76 12 stage is designed for getting better values of the response.
pure error 1.71 3 0.570 Generally, the method of steepest ascent (or descent) is
lack of fit 7.05 9 0.783
total 2376.43 18 applied for moving sequentially along the direction of
FCdd= 1.37; FO,o6(9,3) = 8.81; model is adequate
maximum (or minimum) increase (or decrease) in response.
3.1. First-Order Design (First Move). 3.1.1. Iden-
Yield tification of Variables and Their Levels. Important
4 = 40.794 + 0 . 7 8 1 ~ 1+ 0.831~2+ 0.46%~ + 0.106~4- 0.294~5- variables that influence the selective oxidation of n-butane
0.231X6 (9) to MA are n-butane concentration in the feed stream (q),
ANOVA reaction temperature (q), Mo/V weight ratio (xJ, Ce/V
deg of weight ratio (xJ, contact time (XJ, and P/V atomic ratio
source sum of squares freedom mean square of the catalyst ( x 6 ) . The levels of these variables in the
due to regression 19.94 6 first stage of design are given in Table I. The responses
residual 42.46 12 are conversion of n-butane, selectivity to MA, and yield
pure error 0.387 3 0.129 of MA. These are defined as
lack of fit 42.07 9 4.675
total 62.40 18
X = YO conversion =
Fcalcd= 36.24; FO,O6(9,3) = 8.81; model is not adequate no. of moles of n-butane consumed
Yield
x 100
no. of moles of n-butane fed
Ji +
= 40.794 + 0.781~1+ 0.831~2 0.469~3+ 0.106~4- 0.294~5-
0.231~6- 0 . 4 8 1 ~ 1 ~-20 . 5 8 1 ~ 2 ~-40 . 6 5 6 ~ 2 ~+5 0.431XzX6 + S = 7'0 selectivity =
0 . 8 5 6 ~ 3 ~ 4(10)
no. of moles of MA formed
ANOVA x 100
no. of moles of n-butane consumed
deg of
source sum of squares freedom mean square Y = 5% yield of MA =
due to regression 56.65 6 no. of moles of MA formed
residual 5.75 12 no. of moles of n-butane fed x 100
pure error 0.387 3 0.129
lack of fit 5.363 9 0.596
total 63.40 18 xs
=-
F&d = 4.62; Fo.O5(9,3) = 8.81; model is adequate 100

Table IX. Calculation of the Path of Steepest Ascent

factor
XI XP x3 x4 X5 X6
base level 1.3 405 0.04
0.02 18.5 1.125
unit 0.4 15 0.03
0.01 3.5 0.125
estimated slope ( b ) 0.781 0.831 0.469
0.106 -0.294 -0.231
unit X b 0.3124 12.46 0.0141
0.0011 -1.029 -0.0289
change in level per 6 "C change in x 2 0.1504 6 0.00679
0.00053 4.495 -0.0139
Path of Steepest Ascent as Represented by a Series of Trial Points
expt no. trial point XI X2 X3 X4 X6 X6
45 1 1.4504 411 0.046 79 0.020 53 18.005 1.1110
46 2 1.6008 417 0.053 58 0.021 06 17.510 1.0971
47 3 1.7510 423 0.060 37 0.021 59 17.051 1.0832
48 4 1.9016 429 0.067 16 0.022 12 16.556 1.0693
49 5 2.0520 435 0.073 95 0.022 65 16.061 1.0554

Table X. Results of Experiments along the Path of
Steepest Ascent
expt no. trial point exptl yield predicted yield
~ 3.2.2. First-Order 2"2 Fractional Factorial Design
45 1 42.1 41.5
46 2 46.4 42.2
47 3 49.5 42.8
48 4 45.7 43.2
49 5 42.6 43.1
3.1.2. First-Order 26-2Fractional Factorial Design
(First Move). Since there are six variables, a 26-2 frac-
tional factorial design around the base levels was employed.
The base levels of the factors were chosen on the basis of
an a priori knowledge. The design matrix is shown in
Table 11. The basis for choosing x 5 as ~ 1 x and
3 x 6 as ~ 1 x 4
is that these combinations are expected to have minimum
interactions. Experiments were carried out in a random-
ized sequence to avoid bias. Values of all the responses
(conversion, selectivity, and yield) for the first move of
experiments as planned in the design matrix are given in
Table 11. Experiment 6 was repeated four times so as to
obtain an estimate of the pure error.
3.1.3. Model Fitting. The following first-order model
has been fitted to the experimental data.
+
9 = 60 + 61x1 + 62x2 + 63x3 + 64x4 65x5 + 66x6 (1)
where h0, hl, b2, b3, h4, b5, and 6 6 are the parameters, xi is
the ith variable in its coded form, and j , is the predicted
value of the response (conversion, selectivity, or yield). The
coefficients can be estimated by the least-squares tech-
nique as
6 = (XTX)-lXTy (2)
wher? X is the design matrix, Y is the vector of responses,
and b is the vector of coefficients. The details of the
method are given by Draper and Smith (1981). This is
then tested for its adequacy by a lack-of-fit F test. With
the experimental results as given in Table 11, the fitted
first-order models for conversion, selectivity, and yield are
given in Table 111. Models for conversion and selectivity
are not required for optimizing yield. However, these are
given to have information on the effect of different vari-
ables on conversion and selectivity.
3.1.4. Calculation of the Path of Steepest Ascent
and Conduct of Experiments along This Path. In-
formation obtained from the models was used to locate the
path of maximum increase in yield. The calculation of the
path of steepest ascent in which maximum value of the
yield would be expected is given in Table IV.
Additional experiments were then performed according
to trial points 1-6. The experimental results and values
predicted by the first-order model are given in Table V.
It is observed that, after point 3, experimental yields were
nearly constant. Here, though the yield of MA was be-
coming steady, the first-order model was fitting the data,
indicating that the point might not be the true optimum.
A possible reason for this type of ambiguous result may
be due to the wrong choice of levels of variables in the first
set of design. Therefore, a second set of Ze2 fractional
factorial design (first order, second move) was done around
the new levels of variables and conducted to study the
effect of the variables at their new levels.
3.2. First-Order Strategy (Second Move). 3.2.1.
Variables and Their Levels. The base levels of the
factors for the second move were chosen on the basis of
the optimum conditions obtained in the first move, the
magnitude of the effects of factors as obtained from the
coefficients of the models, and the constraint of not using
high reaction temperature so as to avoid explosion. The
Ind. Eng. Chem. Res., Vol. 30, No. 8, 1991 1823
new levels of different variables for the second move using
the first-order design are given in Table VI.
(Second Move). The design matrix is exactly same as in
the first move. The values of the responses are given in
Table VII.
3.2.3. Model Fitting. The models that fitted the ex-
perimental values of the responses are given in Table VIII.
The first-order model did not fit the yield data. Therefore,
a higher order model including some important cross
product terms was fitted. The cross products that are
confounded with the main effect cannot be determined
from here. The cross products that are confounded with
the main effects can be identified as follows. For this type
of design, generators are I = 135 = 146 = 3456. Neglecting
third-order and higher order interactions, we get the
confounding pattern as follows:
1=35=46; 3=15; 4=16; 5=13; 6 = 1 4
Besides the cross product terms that are confounded with
the main effects, the other cross product terms that have
a strong influence on yield are
612 = 4.481; 624 4.581; 6 2 5 = -0.656;
~
b26 = +0.431; b34 = +0.856
Therefore, a better model is proposed incorporating these
terms with the first-order model. The model then takes
the following form
+
9 = 60 + 61x1 62x2 + 63%3 + 64x4 6 5 x 5 + +
612xlx2 + 624xZx4 + b25X2X5 + b 2 6 X 2 X S + b34X3X4 (3)
The inadequacy of the first-order model indicates the
proximity to the optimum.
3.2.4. Calculation of the Path of Steepest Ascent
and Conduct of Experiments Along This Path. The
path of the steepest ascent to obtain maximum yield is
calculated in Table 1X. Results of additional experiments
performed at those points that would give better yields are
given in Table X. The results show that at trial point 3
a maximum yield of MA (49.5%) was obtained.
4. Conclusions
The promoted VPO catalyst having a P/V atomic ratio
of 1.08, Mo/V weight ratio of 0.06, and Ce/V weight ratio
of 0.022 gave the maximum yield of MA at an inlet n-bu-
tane concentration of 1.75% and a W / F ratio of 17.051
g/ (mg-mol/s). The responses (conversion, selectivity, and
yield) can be described in terms of fitted models.
An increase in the value of reaction temperature, WIF,
n-butane concentration in the inlet feed, or Ce/V weight
ratio increases the conversion of n-butane at the cost of
selectivity towards MA. On the other hand, an increase
in the value of P/V atomic ratio of Mo/V weight-ratio
increases the selectivity at the cost of conversion. This
optimization technique selects those values of the variables
which keep the conversion and selectivity at such levels
which maximize the yield of MA.
Nomenclature
bo, bl, ..., b6 = parameters in eq 1
F = total feed flow rate, mg-mol/s;also used for the ratio of
two independent estimates of the same variance
S = selectivity toward MA
T = transpose of matrix
W = weight of catalyst, g
x i = ith factor in coded form
X = conversion of n-butane
X = design matrix
1824 Ind. Eng. Chem. Res. 1991,30, 1824-1828
y = response
Y = yield of MA
Y = vector of responses
= least-squares estimate
Registry No. PO, 1314-56-3; MA, 108-31-6; Mo, 7439-98-7;
Ce, 7440-45-1; H(CH2),H, 106-97-8; V, 7440-62-2.
Literature Cited
Box, G. E. P.; Hunter, W. C.;Hunter, J. S. Statistics for Experi-
menters; Wiley: New York, 1978; Chapters 10 and 12.
Buchanan, J. S.; Sundaresan, S. Kinetics and Redox Properties of
Vanadium Phosphate Catalysts for Butane Oxidation. Appl.
Catal. 1986, 26, 211-226.
Cavani, F.; Centi, G.; Trifiro, F. The Chemistry of Catalysts Based
on Vanadium-Phosphorus Oxides Note IV Catalytic Behaviour
of Catalysts Prepared in Organic Medium in the Oxidation of C4
Fraction. Appl. Catal. 1984,9, 191-202.
Cavani, F.; Centi, G.; Manenti, I.; Trifiro, F. Catalytic Conversion
of C, Hydrocarbons on Vanadium-Phosphorus Oxide: Factors
Influencing the Selectivity of l-Butene Oxidation. Ind. Eng.
Chem. Prod. Res. Deu. 1985a, 24, 221-226.
Cavani, F.; Centi, G.; Trifiro, F. Study of n-Butane Oxidation to
Maleic Anhydride in a Tubular Flow Stacked-Pellet Reactor.
Influence of Phosphorus on the Selectivity. Appl. Catal. 1985b,
15, 151-160.
Centi, G.; Fornasari, G.; Trifiro, F. On the Mechanism of n-Butane
Oxidation to Maleic Anhydride: Oxidation in Oxygen-Stiochiom-
etry-Controlled Condtions. J . Catal. 1984a, 89, 44-51.
Centi, G.; Manenti, I.; Riva, A.; Trifiro, F. The Chemistry of Cata-
lysts Based on Vanadium-Phosphorus Oxides Note I11 Catalytic
Behaviour of Different Phases in 1-Butene Oxidation to Maleic
Anhydride. Appl. Catal. 1984b,9, 177-190.
Centi, G.; Trifiro, F.; Ebner, J. R.; Franchetti, V. M. Mechanistic
Aspects of Maleic Anhydride Synthesis from C4 Hydrocarbons
over Phosphorus Vanadium Oxide. Chem. Reu. 1988,88,55-80.
Davies, 0. L. The Design and Analysis of Industrial Experiments;
Longman Group: New York, 1978; Chapter 11.
Draper, N. R.; Smith, H. Applied Regression Analysis, 2nd ed.;
Wiley: New York, 1981; pp 1-125.
Hodnett, B. K. Vanadium-Phosphorus Oxide Catalysts for the Se-
lective Oxidation of C4 Hydrocarbons to Maleic Anhydride. Catal.
Rev. Sci. Eng. 1985, 27 (3), 373-424.
Hodnett, B. K.; Delmon, B. Factors Influencing the Selectivity of
Vanadium-Phosphorus Oxide Catalysts for n-Butane Oxidation
to Maleic Anhydride. Appl. Catal. 1985, 15, 141-150.
Hucknall, D. J. Selective Oxidation of Hydrocarbons; Academic
Press: London, 1974; Chapter 4.
Katsumoto, K.; Marquis, D. M. US. Patent 4 132 670, 1979.
Khuri, A. I.; Cornell, J. A. Response Surfaces Designs and Analyses;
Marcel Dekker: New York, 1987; Chapters 1, 2, 3, and 5.
Kittrell, J. R.; Erjavec, J. Response Surface Methods in Heteroge-
neous Kinetic Modelling. Ind. Eng. Chem. Process Des. Deu.
1968, 7 (3),321-327.
Moser, T. P.; Schrader, G. L. Selective Oxidation of n-Butane to
Maleic Anhydride by Model V-P-0 Catalysts. J . Catal. 1985,92,
216-231.
Rao, M. S.; Iyengar, S. S. In Computer Modelling of Complex Bio-
logical Systems; Iyengar, S. s.,Ed.; CRC Press: Boca Raton, FL,
1984; pp 29-53.
Smith, J. M. Chemical Engineering Kinetics, 3rd ed.; McGraw-Hill:
New York, 1981; Chapter 11.
Varma, R. L.; Saraf, D. N. Oxidation of Butene to Maleic Anydride
I. Kinetics and Mechanism. J . Catal. 1978,55, 361-372.
Varma, R. L.; Saraf, D. N. Selective Oxidation of C, Hydrocarbons
to Maleic Anhydride. Ind. Eng. Chem. Prod. Res. Deu. 1979, 18
(11, 7-13.
Wellauer, T. P.; Cresswell, D. L.; Newson, E. J. Optimal Policies in
Maleic Anhydride Production through Detailed Reactor Modell-
ing. Chem. Eng. Sci. 1986, 41 ( 4 ) , 765-772.
Wenig, R. W.; Scharder, G. L. Vanadium-Phosphorus-Oxygen In-
dustrial Catalysts for n-Butane Oxidation: Characterization and
Kinetic Measurements. Ind. Eng. Chem. Fundam. 1986, 25,
612-620.
Receiued for review July 2, 1990
Reuised manuscript received February 6, 1991
Accepted February 12,1991

Selective Oxidation of 11 -Butane to Maleic Anhydride. 2. Identification

of Rate Expression for the Reaction
Shyamal K. Bej and Musti S. Rao*
Department of Chemical Engineering, Indian Institute of Technology, K a n p u r 208 016, U.P.,
India

The modeling of the selective oxidation of n-butane to MA has been done in a systematic manner.
Both Langmuir-Hinshelwood and redox models have been proposed and tested with differential
rate data. Four redox models have been fitted to the experimental data. Discrimination among
these models has been achieved with the use of a sequential experimental design. The redox model
having first-order dependency on n-butane partial pressure and zero-order dependency on oxygen
partial pressure was found to be the most appropriate.

1. Introduction Buchanan and Sundaresan (1986) have proposed different

The selective oxidation of n-butane to maleic anhydride
(MA) is a complex reaction from the mechanistic and
modeling point of view. Besides the main reaction of
selective formation of MA from n-butane, two other side
reactions, viz., the complete oxidation of n-butane and of
MA to oxides of carbon, take place simultaneously.
Different types of models have been proposed by dif-
ferent investigators (e.g., Wohlfahrt and Hofmann (1980),
Centi et al. (19851, Buchanan and Sundaresan (1986),
Lerou and Weiher (1986), and Schneider et al. (1987)).
Wohlfahrt and Hofmann (1980), Centi et al. (1985),and
* T o whom correspondence should be addressed.
0888-5885/91/2630-1824$02.50/0
forms of redox models for this reaction. An adsorption
model has been proposed by Lerou and Weiher (1986).
Though a number of studies are available on the modeling
of this reaction, there is some doubt regarding the order
of reaction with respect to oxygen and also the activation
energy. In this present work we have tried to clarify some
of these doubts by systematic modeling studies.
The reaction has been studied over a Mo- and Ce-pro-
moted VPO catalyst (P/V atomic ratio = 1.0s; weight ratio
of Mo/V = 0.06; weight ratio of Ce/V = 0.0221, the op-
timum composition of which was established in part 1 of
this series. The catalyst had a surface area of 20 m2/g.
The details of catalyst preparation were also given in part
1 of this series.
0 1991 American Chemical Society