I n d . Eng. Chem. Res. 1988,27, 208-211
208
structures on the basis of reducing loop interaction is not a valid criterion in most process control applications. The selection of the structure for a process control system should be made on its ability to effectively reject load disturbances.
Tyreus, B. D. "Optimization and Multivariable Control of Distillation Columns". Advances In Instrumentation Vol. 42, Part 1, Proceedings of ISA 87 International Conference, 1987. Waller, K. V.; Finnerman, D. H.; Sandelin, P. M.; Haggblom, K. E. "On the Difference Between Distillation Column Control Structures". Report 86-2,1986; Process Control Laboratory, Abo Akademi, Abo, Finland.
Literature Cited Buckley, P. S. Techniques of Process Control; Wiley: New York, 1964. Fisher, W. R.; Doherty, M. F.; Douglas, J. M. "Synthesis of Steadystate Control Structures For Complete Chemical Plants, Part I11 Control Structure Synthesis Strategies". Paper 82b, presented at the AIChE Meeting, San Francisco, Nov 1984. Georgakis, C . Chem. Eng. Sci. 1986,41, 1471. Luyben, W. L. Ind. Eng. Chem. Fundam. 1975,14, 321. Niederlinski, A. AIChE J . 1971, 17, 1261. Stanley, G., Marino-Galarraga, M., McAvoy, T. J. Ind. Eng. Chem. Process Des. Dew. 1985, 24, 1181.
William L.Luyben Process Modeling and Control Center Department of Chemical Engineering Lehigh University Bethlehem, Pennsylvania 18015
Received for review May 14, 1987 Revised manuscript received October 14, 1987 Accepted November 2, 1987
Kinetics of the Pyrolysis of Chlorodifluoromethane The thermal decomposition of chlorodifluoromethane has been studied a t 1.379 X lo4-4.826 X lo4 P a (2-7 psig) over the temperature range 750-950 "C and residence times of 0.1-0.25 s, which is consistent with commercial practice. Within the limits of detectability, no perfluoroisobutylene was produced in these short residence time runs. Reaction mechanisms are proposed which include all products and side products identified in the reactor effluent streams by gas chromatography-mass spectroscopy. Arrhenius preexponential factors and activation energies are estimated for each of the reactions using a numerical integration scheme coupled to a Marquardt least-squares estimation algorithm. Operation of the tubular reactor was simulated a t three different temperatures using optimal estimates for the parameter values. Changes in product concentration along the tubular reactor then define operating conditions for any desired product mix. Aliphatic fluoroolefins are made commercially by the direct pyrolysis of monochlorodifluoromethane (FC-22) or the pyrolysis of other fluoroolefins. This study concentrates on the most important commercial fluoroolefins: tetrafluoroethylene (TFE) and hexafluoropropylene (HFP). Pyrolysis of FC-22 was studied in the past (Park et al., 1947; Gozzo and Patrick, 1966; Edwards and Small, 1965), but the temperature range in these studies (500-700 "C) and uncertainties in separation and identification of the pyrolysis products convinced us of the need to further investigate the kinetics of the pyrolysis. Our study covers the operating range 750-950 "C, which is consistent with commercial practice. This study eliminates the need to extrapolate kinetic coefficients reported in the literature beyond the range of their experimental data. Pyrolysis of FC-22 is carried out commercially in tubular reactors with residence times of 0.1-0.25 s. Thus study was carried out in a tubular reactor using the same range of residence times.
alumina tower and its composition analyzed. Analysis of the gas was done on a gas chromatograph (GC) with a Puracil B column. Components of the gas were identified by using GC mass spectroscopy; thus trace amounts of fluorocarbons were identified and quantified.
Experimental Section Figure 1 is a schematic description of the experimental unit. CHClF2 (GENETRON-22, Allied Corporation) was supplied from a cylinder through a rotameter at a constant pressure to a 0.95-cm (3/s-in.) i.d. tubular reactor made of Inconel. The reactor was heated with three Lindberg single-zone tube furnaces. The first furnace of 30 cm (12 in.) was used as a preheater and the other two, with a total length of 60 cm (24 in.), defined the pyrolysis section. The gas exiting from the reactor was immediately quenched in a helical double-wall water-cooled heat exchanger. The gas was cooled below 400 "C, thus stopping any further reaction. After cooling, the pyrolysate was scrubbed with water or HCl solution to remove the HCl and dried in an
CzF4+ H C 1 5 H(CF2),Cl
Reaction Mechanisms The kinetics of the pyrolysis can be described by the following set of equations: CHClF2 7 CF2' 1
+ HCl
3 7 C2F4 2C2F4 7 5 c-C~F~
2CF2'
C2F4+ CF2' 9
I
(A)
(B) (C)
C3F6
(E)
These relations describe the major reactions taking place. A series of high boilers with a general molecular formula H(CF,),Cl are produced, but their concentration per pass is low. Normal commercial practice is to recycle the high boilers. In this study we assumed that they were part of the H(CF2),C1 (FC-124A) concentration. The key reaction of the pyrolysis is the generation of CF2' radicals, which by recombination create the desired fluoroolefins. The kinetics of each reaction was assumed to conform with an Arrhenius temperature dependence. The pyrolysis was carried out at isothermal conditions and the temperature along the reactor was monitored and 0 1988 American Chemical Society
Ind. Eng. Chem. Res., Vol. 27, No. 1, 1988 209 Table I. Kinetic Coefficients of Chlorodifluoromethane Pyrolysis
reaction
preexponential factor activation energy, cal/mol
1" 2b 3b 4" 6.92 X lOI3 9.02 X 10" 2.26 X lo7 1.02 X 70400 16 000 0 49 500
6" 1.64 X 10" 1.05 X 82000 44800
lo5
lo5
7b 8" 1.03 X lo5 1.03 X 35400 79000
gb 10" 2.21 X 10l1 0 47100
lo5
"Unit of preexponential factor is s-l. b u n i t of preexponential factor is L/(mol)(s). "Reaction 9 is irreversible. C4FB
loo-; I -
\
-5z
-'.--._
'\
$80-
I
S.C. SAMPLER
-
--- HCL _---- FC-22
............ HFP FC124A
--- --- ----_
----
0 60n -
s
TFE
------------- -----_
b-
W z
2 40V 0
ROTAMETEI
c
PYROLYSIS REACTOR (3 SECTIONS)
-
OLi-
....................................................
LII
I
0
0.02
I
I
I
I
I
I
I
I
I
I
I
I
I
0.04 0.06 0.08 0.10
I
0.12 0.14 0.16 REACTION TIME ( s e d
I
I
018
'
I
I
I
0.20 0.22
I
I
I
0.24
Figure 2. Pyrolysis of chlorodifluoromethane. Simulation of concentrations (wt %) vs residence time (s) a t 800 "C and 4 psig. FC-22 ON SCALE
D U A L COIL HEAT EXCHANGER
HCL COLLECTOR TANK
*
Figure 1. Schematic diagram of chlorodifluoromethane pyrolysis unit.
controlled. The changes in temperature were small and are the major contribution to the uncertainties in the values of the preexponential factors and the activation energies of the various reactions. A total of 46 experimental runs was conducted, covering the temperature range 750-950 "C, pressure range 1.379 X lo4-4.826 X lo4 Pa (2-7 psig), and flow rates resulting in residence times 0.1-0.25 S. The operation of the tubular reactor was simulated by integrating the kinetic equations along the reactor and comparing the results with the measurements of the gas composition. Since we did not have the analytical capability of measuring the concentration of CF2', a .quasisteady-state approximation was used for its concentration in the computations. The kinetic coefficients were estimated by using a nonlinear least-square regression algorithm (Marquardt, 1963). Initial estimates of parameter values were determined from various literature sources.
Results Table I shows ,the results of the estimation of the preexponential factors and activation energies for reactions 1-10. Using the calculated kinetic coefficients, one can simulate the changes in product concentration along the tubular reactor and define operating conditions for a desired product mix. Figures 2-4 show simulated results for three different operating temperatures. The reaction can be stopped at any desired concentration at a specific residence time by quenching the pyrolysate with water or HC1 solution. Pressure effects were not significant in the range 1.379 X lo4-5.516 X lo4Pa (2-8 psig), which is used commercially; thus the calculations were performed at
REACTION TIME (sed
Figure 3. Pyrolysis of chlorodifluoromethane. Simulation of product concentrations (wt %) vs residence time (s) a t 875 "C and 4 psig.
C4F8
--- HCL ----- FC-22 ............ HFP
FC124A TFE
40 0
c 0 13
8
20
a n
0 REACTION TIME ( s e d
Figure 4. Pyrolysis of chlorodifluoromethane. Simulation of product concentrations (wt %) vs residence time (s) at 950 "C and 4 psig.
2.758 X lo4Pa (4psig). As the temperature increases from 800 to 950 "C, conversion of FC-22 to products increases. This results in increased production of FC-124A and other high boilers, which need to be recycled. At 800'"C only
210 Ind. Eng. Chem. Res., Vol. 27, No. 1, 1988 Table 11. Error Analvsis av concn and rangeof concn in effluent component stream, mol/L FC-22 59.4 range: 31.3-71.4 TFE 31.5 range: 20.2-39.8 c-CZF, 0.88 range: 0.50-2.9 C3F6 3.80 range: 0.50-10.5 H(CF,),Cl" 3.03 range: 1.21-4.7
magnitude and ran&of av fract. error [actual predicted1 /actual 0.054 range: +0.08 t o -0.19 0.10 range: +0.13 to -0.29 0.77 range: +0.99 to -0.62 0.40 range: +0.78 to -0.95 0.32 range: +0.28 to -0.65
Includes also the concentrations of all other high-boiling components.
REACTIOL TIME lsecl
Figure 5. Pyrolysis of chlorodifluoromethane. Simulation of product concentrations (wt %) vs residence time (9) at 875 "C, 4 psig, and FC-22/TFE feed ratio = 3.5.
TFE is made in substantial quantities. At 875 "C TFE reaches its peak production of 22 wt %. At longer residence times the TFE converts to other products. At 950 OC the peak in TFE production of 25 wt % is reached at a shorter residence time. The higher temperature causes a shift in the equilibrium in eq B from TFE to CF; radicals and a concomitant increase in concentration of the higher molecular weight products. Figure 5 shows the effects of using an FC-22/TFE feed mixture with a molar ratio FC-22/TFE of 4 1 (3.5 wt ratio) at 875 "C. The introduction of TFE roughly doubles the amount of HFP, C4F8,and FC-124A that is being made, and their concentrations become quite similar.
Error Analysis Parameter estimates were based on 46 runs covering the temperature range 750-950 "C, pressure range 1.379 X lo4-4.826 X lo4 Pa (2-7 psig), and residence times of 0.1-0.25 s. The data from all runs were used to determine the optimal parameter values presented in Table I. The objective function in the Marquardt (1963) optimization algorithm was the sum of the squares of the deviations between observed and predicted values for the concentrations of all the components measured in the effluent stream. The least-squares criterion puts greater weight on fitting the components with higher concentrations in the effluent stream. The goodness of fit then is inversely proportional to the concentration of species measured. The optimal value of each parameter estimated by the least-squares criterion is also influenced by the sensitivity of the objective function to that parameter in the neighborhood of the minimum. Therefore, we can never prove that a set of estimated "optimal" parameters is unique for a coupled set of nonlinear functions. Moreover, because of the strong correlation of several of the parameters, it is not very instructive to present, for example, 95% confidence inte~alsfor the parameter estimates. An approach which gives a clearer picture of the goodness of fit is to present the average fractional error and range of predicted values for each component in the effluent stream when the optimal parameter set is used for simulation over the ranges of temperatures, pressures, and residence times studied. These values are presented in Table 11. The prediction errors appear to be randomly distributed, with positive and negative values appearing with equal frequency around a zero mean value, except for the prediction of C4Fs concentrations. For that case, the simulation consistently overestimates C4F8 concentration over all experimental conditions. Finally, the error analysis indicates that the set of optimal parameter values predict
product concentrations for all combinations of temperature, pressure, and residence times studied with approximately the same precision.
Discussion Estimated values for the preexponential factors and activation energies for reactions A and B are similar to those reported by Edwards and Small (1965). The major differences are in the values for the preexponential factor for reactions 2 and 4 and in the activation energy estimated for reaction 2. Our estimated preexponential factor for reaction 2 is approximately an order of magnitude greater than that reported by Edwards and Small (1965),and our estimated activation energy is greater than theirs by approximately a factor of 3. On the other hand, for reaction 4,their reported preexponential factor of 1016.66 is many orders of magnitude larger than our estimated value of 1.02 X lo5, while our estimate of activation energy, 70.4 kcal, is identical with their value. There are several possible reasons for the discrepancies. First, Edwards and Small only consider reactions A and B in their analysis. They recognize that side reactions C, D, and E take place but do not consider these reactions in estimating the rate constants for reactions A and B and instead use an empirical correction factor. This means that the mass balances for their reaction scheme are in error by the amount of side products produced. Since our parameter estimation procedure used all five reactions, the mass balances accounted for >99% of the products formed in each run. A second possible source of discrepancy is the temperature range studied. Edwards and Small's study covered the temperature range 533-750 "C, whereas our study covers the temperature range 750-950 "C. Lastly, all nonlinear parameter estimation techniques are sensitive to the initial guesses of the parameter values and to the effects of small changes in parameter values on the value of the objective function (in this case, the sum of the squares error) in the neighborhood of the minimum. One can never prove that a single set of parameter values is unique. The best correlation was obtained for the FC-22 conversion. Errors in the predicted values for each component using the "best-fit'' parameter estimates are inversely proportional to the concentration of the component in the pyrolysate. Errors ranged from approximately i8% for FC-22 to *35% for HFP. This is partially attributed to the scarcity of some components in the product stream and the failure to include some products in the kinetic scheme, particularly the higher boiling components. Within experimental error, the mass balances for each run checked. Reactors were run for a maximum of 6-h duration and during this time, no significant formation of carbon (fouling) was observed. We were particularly concerned about the production of perfluoroisobutylene (PFIB) in the process since PFIB is one of the most toxic substances known. It is also known that PFIB can be produced during pyrolysis of TFE and is one of the reasons that CHClF, was chosen as the
Ind. Eng. Chem. Res. 1988, 27, 211
starting material for HFP production. The other reason is that TFE is itself a commercially valuable material and, where possible, one usually starts with a cheaper raw material (CHClF2in this case). We did not experimentally observe (within the limits of detectability of the gas chromatograph-mass spectrometer system) the formation of PFIB and only trace quantities (