Design of an Industrial Adsorption Process with Activated Carbon for


Design of an Industrial Adsorption Process with Activated Carbon for...

4 downloads 203 Views 138KB Size

Ind. Eng. Chem. Res. 2001, 40, 3171-3180

3171

Design of an Industrial Adsorption Process with Activated Carbon for the Removal of Hexafluoropropylene from Wet Air Marco J. G. Linders,*,† Martijn B. L. van der Weijst,† Jacques J. G. M. van Bokhoven,‡ Freek Kapteijn,† and Jacob A. Moulijn† Chemical Engineering Department, Industrial Catalysis Section, Delft University of Technology, Julianalaan 136, 2628 BL Delft, The Netherlands, and TNO Prins Maurits Laboratory, Post Office Box 45, 2280 AA Rijswijk, The Netherlands

The design of an industrial gas adsorption column for the removal of hexafluoropropylene (HFP) from wet air is reported. The column uses activated carbon and operates under the strongly fluctuating concentration conditions of an industrial plant. An important aspect of this study was to investigate the influence of water vapor in the feed on the performance of the adsorption column. During regeneration of the carbon, unforeseen reactions of HFP on carbon occurred at elevated temperatures, imposing constraints on the regeneration conditions. A two-dimensional mathematical model for the adsorber worked well. Design calculations were performed on the basis of the experimental results. The Dubinin-Radushkevich model described the adsorption equilibrium substantially better than did the Langmuir model. It is demonstrated that a correct adsorption isotherm is more crucial than particle kinetics in designing the adsorption column. Decoupling hydrodynamics from the rate processes by filling the bed with fines of an inert material worked well, enabling the breakthrough experiments to be downscaled. Introduction The removal of trace components from gaseous waste streams in chemical industry is, because of more stringent permit requirements, becoming increasingly important. The design project reported in this work was carried out for DuPont de Nemours B.V. (Dordrecht, The Netherlands).1 The main objective was to design a gas adsorption column with activated carbon for the removal of hexafluoropropylene (HFP) from wet air. This column must operate under the strongly fluctuating concentration conditions of the industrial plant. First, an appropriate carbon must be selected. The carbon must have the following properties: high adsorption capacity, fast diffusion and good desorption characteristics, hydrophobicity, and ability to be used in packed beds. Because of the presence of water vapor in the feed, an important aspect of this study was to investigate the influence of water vapor on the performance of the adsorption column. Regeneration of the adsorbed HFP, preferably with steam is desirable because, as is usually the case, steam is available on-site and the HFP/steam flow can be directly used in another part of the plant. The Emission. At the DuPont site in Dordrecht, several monomers are produced that are used for the production of fluoropolymers and fluoroelastomers or rubbers. One of these polymerization reactions is carried out batchwise and under pressure. After pressure has been relieved, the batch is dropped into a hold-up tank (see Figure 1). Subsequently, the batch is divided over coagulators in which the dispersion is coagulated. * Correspondence concerning this article should be addressed to M. J. G. Linders. Phone: +31-15-2843521. Fax: +31-15-2843963. E-mail: [email protected]. Present address of M.J.G. Linders: TNO Prins Maurits Laboratory, P.O. Box 45, 2280 AA Rijswijk, The Netherlands. † Delft University of Technology. ‡ TNO Prins Maurits Laboratory.

Figure 1. Process scheme of the emission and removal of HFP.

During this dropping and coagulating process, the solvent contains traces of unreacted HFP, which, as a consequence, are released. The HFP is removed by continuously blowing an air stream through the holdup tank and the coagulators. The coagulators are operated at an elevated temperature and in the presence of water. This results in a HFP-contaminated air exit stream that is saturated with water vapor and has a temperature of 363 K. The gas stream also contains small polymer particles, typically 0.1 µm in diameter. Figure 2 shows measured HFP emissions during test cases, which can be considered representative for the process. These emissions were measured on two subsequent days. The air flow applied was 400 m3/h. The worst-case emission can be approximated by a so-called “idealized emission profile” consisting of emissions of 90 ppm during the first 3.5 h of the 4-h test and 450 ppm during the last 0.5 h of the test. Even the low concentration level of 90 ppm HFP is undesirable: 90 ppm HFP at 400 m3/h (1 ppm at 20 °C and 1 atm ) 6.25 mg/m3) gives an emission of 0.225 kg/h at a concentration of 563 mg/m3, when a concentration of 20

10.1021/ie000932b CCC: $20.00 © 2001 American Chemical Society Published on Web 06/15/2001

3172

Ind. Eng. Chem. Res., Vol. 40, No. 14, 2001

is the macroporosity of the adsorbent particle, Dm (m2/ s) is the molecular diffusivity, τ (dimensionless) is the tortuosity, and r (m) is the radial coordinate inside a particle. The molecular diffusivity divided by the tortuosity is defined as the pore diffusivity. Both phases are assumed to be isothermal, and their coupling is assumed to be described by the equation

p

Figure 2. Measured HFP emission profiles during test cases on two subsequent days (4 and O) and the idealized emission profile (solid line) (source: DuPont field measurements). The idealized emission profile is used in the design calculations.

mg/m3 is desirable. Thus, the emission of HFP has to be reduced. The design calculations were performed using the idealized emission profile. For the design, in addition to minimizing the absolute concentration, the aim of keeping the total amount released into the environment (the emission) as low as possible is also important. Theoretical Section Adsorber Model. A two-dimensional mathematical model was used to describe the dynamic behavior of the adsorber. The transport mechanism in the gas phase is described by a dispersed plug-flow model accounting for mixing in the axial direction. The resistance to mass transfer between the bulk of the gas phase and the outer surface of the particle is located in a fictitious film around the particle. The gas-phase mass balance is then given by

( )

1 - b ∂c ∂2c vs ∂c kfa′(c - c*) ) Dax 2 ∂t b ∂z b ∂z

(1)

where c (mol/mg3) is the gas-phase concentration, t (s) is the time, Dax (m2/s) is the axial dispersion coefficient, z (m) is the distance from the bed inlet, vs (m/s) is the superficial velocity, b (mg3/mr3) is the bed porosity, kf [mg3/(mp2 s)] is the mass transfer coefficient, a′ (mp-1) is the external surface area per unit particle volume, and c* (mol/mg3) is the concentration at the gas-particle interface. The fluid velocity is considered to be constant across the bed length. This assumption is valid when the adsorptive species is present in low concentrations, which is the case for the measured emission level. Intraparticle gas transport is described by a diffusive contribution assuming Ficks law and an effective molecular diffusivity as the transport parameter. From a differential mass balance on a spherical shell element assuming rapid adsorption in the micropores, i.e., equilibrium between the intraparticle gas phase and the adsorbed phase, the continuity equation over the solid phase becomes

(

)

(

)

∂cp Dm ∂2cp 2 ∂cp ∂q + p ) p + ∂cp ∂t τ ∂r2 r ∂r

(2)

where q (mol/mp3) is the amount adsorbed, cp (mol/mg3) is the intraparticle gas-phase concentration, p (mg3/mp3)

( )

Dm ∂cp τ ∂r

r)deqv/2

) kf(c - c*)

(3)

where deqv is the equivalent spherical diameter. The continuity equation over the solid phase is obtained for spherical particles, whereas in this work, cylindrical extrudates were used. These particles are modeled as spheres by calculating a so-called equivalent spherical diameter.1 The molecular diffusivity is estimated from the empirical correlation of Fuller et al.2-4 The axial dispersion coefficient is calculated from the particle Peclet number, which is taken to equal 2.5 The mass transfer coefficient6-8 is determined on the basis of the ChiltonColburn factor, which is calculated using the empirical correlation from Dwivedi et al.9 Equilibrium between the gas- and solid-phase concentrations is described with either the DubininRadushkevich adsorption isotherm

( [ ( )] )

q ) qsat exp -B2 ln

ps p

2

with B )

RT βE0

(4)

or the Langmuir adsorption isotherm

bp q ) qsat 1 + bp

(5)

where q (mol/g) is the amount adsorbed, p (Pa) is the pressure, qsat is the amount adsorbed at the saturation pressure ps, R (J/mol/K) is the gas constant, T (K) is the temperature, β (dimensionless) is the affinity coefficient of the adsorptive species, E0 (J/mol) is the characteristic energy of the adsorbent, and b (Pa-1) is the Langmuir equilibrium constant. Calculation Procedure. The set of partial differential equations was implemented in a Fortran program. The equations were solved with the numerical method of lines using Fortran routines of the DSS/2 package.10,11 The essence of the method of lines is to transform a partial differential equation into a set of ordinary differential equations (ODEs) in time by using a discretization scheme for the spatial coordinates z and r in the axial and radial directions, respectively. The equations have been discretized using a one-dimensional five-point centered approximation for first-order derivatives for z, subroutine DSS004, and a two-dimensional, three-point centered approximation for first-order derivatives for r, subroutine DSS022. The time dependency of the resulting set of ODEs was solved using the numerical integrator LSODES,10,11 which is especially suited for sparse Jacobian matrices, as in this discretization method. Experimental Section Materials. After contact with the vendor, it was decided to consider two activated carbons, viz., steamactivated R3 carbons Norit RB3 and Sorbonorit B3, which are cylindrical extrudates with diameters of about

Ind. Eng. Chem. Res., Vol. 40, No. 14, 2001 3173 Table 1. Characterization Results Derived from Mercury Porosimetry and Nitrogen Adsorption

carbon Norit RB3 Sorbonorit B3

BET pore bulk macropore surface micropore volume volume density volume area (cm3/g)c (cm3/g)d Fb (g/cm3) (cm3/g)a (m2/g)b 0.686 0.605

0.511 0.645

1038 1453

0.419 0.602

0.455 0.755

aObtained by mercury porosimetry. b BET surface area determined from data with relative pressures such that 0.005 < pr < 0.15. c Micropore volume calculated from the amount adsorbed at pr ) 0.16 using the method of Horvath-Kawazoe;12 this volume corresponds to pore diameters d < 2 nm. d Pore volume calculated from the amount adsorbed at pr ) 0.90 using the BJH method;13 this volume corresponds to pore diameters d < 20 nm.

3 mm. The two carbons are quite similar, the only difference being the sizes of their micropore volumes. Sorbonorit B3 has a larger micropore volume because of its longer steam activation time. Some characteristics of the two carbons are given in Table 1. Adsorption Equilibrium. Single-component isotherms and binary adsorption equilibria of HFP and water were measured on the two carbons at 303 K using the headspace analysis method.14,15 A known amount of carbon was put into a vial of known volume, which was sealed gastight with a rubber cap. Subsequently, the adsorptive species was injected into the vial. The exact amount injected was determined by weighing. After equilibration in a thermostated bath, the gasphase concentration of the adsorptive species was analyzed by gas chromatography. The amount adsorbed on the activated carbon was calculated from the mass balance. In the case of a liquid, e.g., water, a glass tube was placed inside the vial. The liquid was injected into this tube to prevent direct contact between the liquid and the carbon. Kinetics in Single Particles. Kinetic experiments were performed with HFP on the two carbons in a thermogravimetry-differential scanning calorimetry apparatus that was coupled to a mass spectrometer (TG-DSC-MS). So-called uptake experiments were performed at 303 K to determine pore diffusivities. The pore diffusivities were determined by estimating the molecular diffusivity and fitting the uptake curves by adjusting the tortuosity. Temperature-programmed desorption (TPD) experiments were performed to measure desorption as a function of temperature and time. The carbon was equilibrated with HFP at 303 K, after which the temperature was increased with heating rates of 1.0 and 5.0 K/min. Breakthrough Measurements. All breakthrough experiments were performed at a temperature of 295 K and ambient pressure. The Norit RB3 and Sorbonorit B3 carbons were tested in an adsorption column with an internal diameter of 13.3 mm. Both single-component and binary experiments were performed. The HFP concentration was 44.3 g/m3 (pr ) 1.0 × 10-3), and the superficial gas velocity was 0.013 m/s. Approximately 1.6 g of carbon was used, resulting in a bed heights of about 28 mm for Norit RB3 and about 38 mm for Sorbonorit B3. Binary experiments with preconditioned carbon were performed at a relative humidity of 30%. The carbon was preconditioned with a humid helium flow for one night. In addition, binary experiments were performed at a relative humidity of 80% starting with dry carbon, i.e., without conditioning the carbon overnight. In addition to these adsorption experiments, some

regeneration experiments were performed. First, the carbon bed was saturated with HFP at a temperature of 295 K, as in a regular adsorption experiment, and then it was regenerated with water vapor at a temperature of 353 K. The regeneration was carried out by saturating the helium flow with water vapor, which is not the same as using steam. In this way, a steam regeneration experiment was mimicked. The gas effluent concentration of the adsorption column was analyzed with a gas chromatograph. Because the ratio of the column diameter to the particle diameter is rather small, the bed-filling technique, known from catalysis, was applied. As a general rule, the radial concentration profile can be neglected when the ratio of the column and particle diameters is larger than 20.16,17 However, according to Sie,17 filling the empty volume between the particles with fines of an inert material allows packed-bed experiments to be performed in situations where the ratio of the column and particle diameters is even less than 5, i.e., when the criterion for a flat velocity profile is not satisfied. This should result in an ideal breakthrough curve, where wall and packing effects can be neglected. Thus, in addition to the “regular” experiments, some breakthrough experiments were performed with beds of both carbons, which were filled with silicon carbide (SiC) fines with an average diameter of 363 µm. First, the carbon was “rained” into the column, and then the SiC fines were added. Because the latter particles are much smaller, they fall or flow along and between the carbon particles in such a way that they fill the voids between them. The carbon bed length was not affected by this process. Thus, the actual particle size and packing of the Norit RB3 and Sorbonorit B3 particles were not changed. This is called the decoupling of the hydrodynamics and kinetic processes. On a somewhat larger scale, breakthrough experiments were performed in a column with an internal diameter of 0.1 m. Approximately 320 g of carbon was used, which resulted in bed heights of about 0.08 m for Norit RB3 and about 0.09 m for Sorbonorit B3. The carbon was rained into the column and then vibrated for 30 min at 170 Hz. This is a standard procedure for obtaining good packing of a carbon bed. The HFP concentration was 8.2 g/m3 (pr ) 1.9 × 10-4) at a total flow of about 1.0 m3/h. Air was used as the carrier gas. The concentration was determined with a Miran analyzer at a wavelength of 9.6 µm. Only part of the breakthrough curve was obtained, because the range of concentrations that can be measured with the Miran instrument is limited. The HFP concentration that can be measured ranges between 0.001 and 3 g/m3. Because a relatively large amount of carbon was consumed by these experiments, the carbon was regenerated. The standard procedure for regenerating carbon for reuse is placing it in a vacuum oven at a temperature of 393 K for at least one night. Results The adsorption column for the removal of HFP can be designed on the basis of the experimental results. The most important results from the experimental study are summarized here. For a more detailed discussion, the reader is referred to Linders.1 Adsorption Equilibrium. The HFP adsorption isotherm on Norit RB3 is shown in Figure 3. The experimental data are described with the Dubinin-Radush-

3174

Ind. Eng. Chem. Res., Vol. 40, No. 14, 2001

Figure 3. Adsorption isotherm of hexafluoropropylene on Norit RB3 at a temperature of 303 K. (b) Experimental data. Comparison between fits with the Langmuir (dashed) and the DubininRadushkevich (solid) model.

Figure 5. Sorption of HFP on Norit RB3 and Sorbonorit B3 at a temperature of 303 K, a concentration of 396 g/m3, and a gas flow rate of 200 mL/min.

Table 2. Adsorption Isotherm Parameters of Hexafluoropropylene on Norit RB3 and Sorbonorit B3 for the Langmuir and Dubinin-Radushkevich Models at a Temperature of 303 K model

parameter

Langmuir

qsat b σ2 a

(mmol/g) (10-3 Pa-1) (10-3 (mmol/g)2)

D-R

qsat B σ2 a

(mmol/g) (-) (10-3 (mmol/g)2)

Norit RB3

Sorbonorit B3

3.14 0.957 76.6

3.43 0.639 62.7

4.14 0.1523 8.9

4.51 0.1603 10.5

a Variance σ2 ) sum of squared residuals divided by d.f.; d.f. ) degrees of freedom ) number of experimental data points minus number of parameters in the isotherm equation.

Figure 4. Comparison between the Dubinin-Radushkevich fits of the HFP adsorption isotherms at a temperature of 303 K for the carbons Norit RB3 and Sorbonorit B3.

kevich and Langmuir equations. The results for the fitted adsorption parameters of both carbons are given in Table 2. The Langmuir model describes the experimental data less well than the Dubinin-Radushkevich model does. Figure 4 compares the Dubinin-Radushkevich fits of the HFP adsorption isotherms for the two carbons. At low relative pressures (pr < 2.0 × 10-3), Norit RB3 adsorbs slightly more HFP than Sorbonorit B3, when the capacity is expressed per gram of carbon. At higher relative pressures (pr > 2.0 × 10-3), this trend is reversed, as Sorbonorit B3 adsorbs more HFP. This is due to the larger pore volume of Sorbonorit B3. However, in terms of column volume, the capacity of

Figure 6. Temperature-programmed desorption of HFP on Norit RB3 and Sorbonorit B3. The carbon was equilibrated with HFP at 303 K, and then the temperature was increased at a rate of 1.0 or 5.0 K/min while HFP was not present in the feed.

Norit RB3 is larger; that is, because the bulk density of Norit RB3 is about 13% higher than the density of Sorbonorit B3, a certain column volume can contain considerably more Norit RB3 carbon, resulting in a higher column capacity. Previous binary equilibrium measurements revealed that the HFP capacity is reduced significantly for both carbons under humid conditions.1 At relative humidities below about 50%, the single-component capacity is only slightly reduced for Sorbonorit B3, whereas for Norit RB3, a larger reduction is observed. At higher relative humidities, the effect is stronger, although the capacity on Sorbonorit B3 is somewhat less affected than the Norit RB3 capacity. Kinetics in Single Particles. Kinetic experiments, i.e., uptake and temperature-programmed desorption, revealed that both adsorption and desorption are faster on Sorbonorit B3 (see Figures 5 and 6. This difference can be explained by the fact that Sorbonorit B3 has a larger macropore volume, which leads to better accessibility to the micropores. However, the diffusion inside the particles is fast for both carbons and is, therefore, not a limiting factor in the adsorption process. The kinetic experiments also showed that Sorbonorit B3 has a lower heat of adsorption than Norit RB3, which results in smaller temperature effects. Breakthrough Measurements. Figure 7 shows breakthrough results for both carbons. HFP breakthrough under dry conditions occurs at around 4500 s for Norit RB3 and at 4100 s for Sorbonorit B3. The breakthrough time is decreased when the carbon is conditioned at a relative humidity of 30% and also when a binary feed is applied on a dry carbon bed. Clearly,

Ind. Eng. Chem. Res., Vol. 40, No. 14, 2001 3175

Figure 8. Normalized effluent concentrations of HFP as functions of time. The breakthrough experiments were performed in a 13.3mm-diameter column with Norit RB3 and Sorbonorit B3. First, the carbon bed was saturated with HFP at a temperature of 295 K, and then it was regenerated with water vapor at a temperature of 353 K. Adsorption and desorption are both normalized on the feed concentration.

Figure 7. Normalized effluent concentrations of HFP as functions of time. The breakthrough experiments were performed in a 13.3mm-diameter column: (a) Norit RB3, (b) Sorbonorit B3. (O) Singlecomponent feed; (×) single-component feed, bed filled with SiC; (0) binary feed at RH ) 30%, preconditioned carbon; (4) binary feed at RH ) 80%, dry carbon bed.

the influence of water vapor is more pronounced for Norit RB3 than for Sorbonorit B3, which means that Sorbonorit B3 is the better carbon to use under humid conditions. Because the ratio of the column and particle diameters is rather small for the experiments with the Norit RB3 and Sorbonorit B3 carbons, the HFP breakthrough experiments were also performed with beds of both carbons that were filled with silicon carbide fines. The figures show that filling the bed with SiC results in steeper breakthrough curves than those observed in the case no SiC is used. Thus, less axial dispersion takes place. Through the use of SiC, the hydrodynamics are decoupled from the sorbent particle size, as is done in microreactor studies for trickle-bed operation.17 Figure 8 shows the result of regeneration experiments. For both carbons, total breakthrough (c/cfeed ) 1) is reached after 5500 s. It is clear that desorption at 353 K is faster than adsorption at 295 K. For both carbons, regeneration is complete already after 2500 s. The rate of desorption does not follow a single exponential function; rather, in the beginning, it is relatively high, and then the rate decreases after about 500 s. Experimentally, it was observed that some condensation of water vapor occurred in the carbon bed despite the heating of the tubes. This condensation might have resulted in blocking of the tubes and/or the carbon particles, which slowed the desorption. Graphically, the areas above the breakthrough curve and below the desorption curve have to be equal, when the scaling of the y axis is taken into account (“a graphical mass balance”). For both carbons, however, the area below the desorption curve is about 15% lower than the area above the breakthrough curve. The discrepancy can be explained by the difference in the operation temperatures of adsorption, T ) 295 K, and desorption, T ) 353

Figure 9. Normalized effluent concentrations of HFP as functions of time. The breakthrough experiments were performed in a 0.1m-diameter column with Norit RB3 (triangles) and Sorbonorit B3 (squares).

K. The gas volume changes between these temperatures by about 15%. The measured effluent concentration is, therefore, 15% lower during desorption. Thus, the areas are the same when the density difference is taken into account, and no HFP remains on the column. Figure 9 shows the breakthrough results with the wider column. Sorbonorit B3 exhibits a shorter breakthrough time than Norit RB3. This is in agreement with the adsorption isotherm. At the applied relative pressure, the capacity is lower for Sorbonorit B3. Because the same weight of carbon is used in the experiments, a lower total capacity is available on Sorbonorit B3, and thus, breakthrough is faster. The two experiments performed with Sorbonorit B3 agree well with each other, in contrast to the two Norit RB3 experiments, which show a large difference in the breakthrough time. It appeared that the Norit RB3 carbon had not been completely regenerated when it was used in the second experiment, explaining the relatively poor performance. The Norit RB3 carbon was again regenerated by placing it in a vacuum oven at 393 K. After 3 days, the oven was opened, and the following phenomena were observed: (i) the glass window of the oven and the glass cup containing the carbon had been etched; (ii) an irritating, acidic smell was released from the oven; and (iii) less than 10% of the original amount of adsorbed HFP had been desorbed. Thus, it was clear that a

3176

Ind. Eng. Chem. Res., Vol. 40, No. 14, 2001

Figure 10. Mass spectra obtained with the TG-DSC-MS instrument: (a) pure HFP, (b) components released from contaminated Sorbonorit B3 carbon, unknown peaks at masses 85 and 97 are marked with arrows.

Figure 11. Possible reaction routes of HFP on activated carbon. Hydrolysis (left) and oxidation (right).

reaction had occurred on the carbon, which will be referred to as the contaminated carbon. The same observations were made when Sorbonorit B3 was placed in the vacuum oven for regeneration. It should be noted that the vacuum in this oven is not maintained overnight. It is, therefore, possible that air from outside enters the oven. Some of the contaminated particles were taken and placed in the TG-DSC-MS apparatus. The temperature was instantaneously raised to 473 K. Figure 10 shows the mass spectra of pure HFP and of the component(s) released from the contaminated carbon. Some HFP is still present on the contaminated carbon (molecular peak at mass 150), but in addition, two new peaks have arisen at masses 85 and 97 (marked with arrows). These peaks are not fragmentation peaks of HFP, so at least one extra component has to be present on the carbon. Discussion Reaction of HFP. Very little was found in the literature about reactions of HFP on activated carbon. However, a number of reactions were found for perfluoroisobutene (PFIB, C4F8), a compound related to HFP that is highly toxic.18,19 In analogy to the reaction routes of PFIB, two kinds of reactions can occur with HFP: hydrolysis and oxidation. Furthermore, it is not unlikely that surface groups on the high-surface-area active carbon act as a catalyst. The hydrolysis reaction is well-known for PFIB18 and is derived for HFP in Figure 11. Hydrogen fluoride (HF) is formed, which explains the etched glass. However,

the unknown peaks at masses 85 and 97 do not correspond with the molecules formed in this reaction route. Furthermore, the formed hydrofluoric acids are volatile. Thus, the chance that these acids remain adsorbed on the carbon under vacuum at 393 K is rather small. Therefore, this reaction route is capable of explaining only one of the three observed phenomena. The oxidation route appears to be more likely19 (see Figure 11). Again, acids are formed. The product HFPO can further react and produce HF (not shown in the figure). It is expected that some of the molecules formed, i.e., HFPO, HFPA, and HFPA‚H2O, are difficult to desorb. It is known that this is the case for HFPA‚H2O. The formation of HFPO can lead to oligomerization, which yields high-molecular-weight products that will be even more difficult to desorb. The unknown peak at mass 97 might be due to a fragment of HFPO, namely, C2F3O (see Figure 11). To obtain more insight into the reactions that had taken place, fresh Norit RB3 particles were placed in the TG-DSC-MS apparatus at a temperature of 473 K. A feed of HFP in air was passed by the instrument. Peaks at mass 85 and 97 appeared, as shown in Figure 12a. The peak at mass 85 became even larger when water vapor was added to the HFP/air feed (see Figure 12b). Experiments that were performed with helium/ HFP or helium/HFP with water vapor did not produce the peaks at masses 85 and 97. However, the absence of these peaks does not necessarily mean that the reactions do not occur. It could be possible that the concentrations of reaction products are below the detection limit as a result of a slow reaction rate. Neverthe-

Ind. Eng. Chem. Res., Vol. 40, No. 14, 2001 3177

Figure 12. Mass spectra obtained with the TG-DSC-MS instrument when (a) HFP/air and (b) HFP/air saturated with water are fed to fresh Norit RB3 carbon at a temperature of 473 K.

less, the reactions of the analogous PFIB19 and the experimental results both indicate that oxidation is the most likely reaction route of HFP on carbon. It is concluded that oxygen is present through the leakage of air into the vacuum oven. These results lead to the recommendation that, apart from other constraints, the adsorption process should be carried out at low temperatures and the regeneration at elevated temperatures with exclusion of oxygen. Design Considerations. It is clear that Sorbonorit has some advantages over Norit RB3. If the relative humidity cannot be reduced below 50% because of excessive costs, the best carbon to use is Sorbonorit B3. However, if the relative humidity can be reduced well below 50%, Norit RB3 is a good alternative because of its higher capacity per unit volume. The gas stream contaminated with HFP exits the coagulators at 363 K and is saturated with water vapor. Under these conditions, littleadsorption of HFP on the carbon takes place. Thus, the gas stream should be cooled anyway before it is sent to the adsorption column. This cooling results in condensation of a part of the water vapor present. Subsequent heating of the stream will further decrease the relative humidity. For example, when the gas stream is first cooled to 283 K and then heated to 303 K, the relative humidity is reduced to 30%.20 In that case, the influence of the water vapor will be rather small for both carbons. It is clear that a wide variety of temperature and relative humidity conditions can be set. An important design criterion is that the carbon can be regenerated, preferably with steam. Steam is available on-site, and the HFP/steam flow can be used directly in another part of the plant. It was demonstrated that, at a regeneration temperature of 353 K, desorption is already faster than the adsorption process at 295 K (see Figure 8). However, scaling up from these experiments to an industrial column is not trivial. First, the column has to be heated to the regeneration temperature, and second, the column has to be cooled again to enable extensive adsorption. Each of these steps can last up to 2 h.21 A higher temperature would, of course, increase the desorption rate. Industrial columns usually apply regeneration temperatures of 423-473 K. However, at these temperatures, reaction of HFP with oxygen and/or water vapor can occur. Thus, an optimal regeneration temperature has to be chosen, one high enough to ensure fast desorption but not so high that it causes the occurrence of the reaction. Another option is to install a third bed in case the desorption time is

still too long. As (i) desorption rates are sufficiently high already at 353 K and (ii) it is quite possible to carry out regeneration under the exclusion of air, it seems possible to regenerate the column under practical conditions without creating the conditions for the oxidation reaction to occur. In the next section, design calculations are presented on the basis of a standard container filter. The design calculations were performed with the two-dimensional mathematical model given in the theoretical section. It was shown by Linders1 that this model describes experimental breakthrough profiles fairly well using independently determined parameters such as the adsorption isotherm, kinetic/transport data, activated carbon properties, etc. Design Calculations Unless mentioned otherwise, the following process conditions were applied in the calculations. The dimensions of the standard container filter, which were obtained from Norit N.V., are a diameter of 1.3 m and a length of 1.5 m. After most of the water vapor has been removed, it is assumed that the adsorption process is carried out at a temperature of 303 K. Of course, some water vapor will still be present, resulting in a filter with a somewhat shorter cycle time than is calculated. The pressure is kept at atmospheric, equal to the pressure in the coagulators. As long as the superficial gas velocity is below 0.12 m/s, the pressure drop across the packed bed remains less than 1.0 kPa/m.22 This is rather low, so the pressure drop is not of any concern. The feed concentration follows from the field measurements performed by DuPont and is approximated by an idealized profile (see also Figure 2). The field measurements were performed at a gas flow rate of 400 m3/h. This results in a superficial gas velocity of 0.084 m/s. All relevant parameters are given in Tables 3 and 4. The calculations were performed with an isothermal model, although a heat effect is shown in Figure 5. However, this heat effect would result in a temperature increase of only a few degrees Celsius. Furthermore, the concentration at which this heat effect is measured is a few orders of magnitude (2-3) higher than the concentration that is present in the industrial adsorption process, implying even smaller heat effects. Therefore, heat effects can be neglected in the design calculations, and an isothermal model is sufficient. In Figure 13, the simulation results are given for Norit RB3 and Sorbonorit B3. Breakthrough occurs after

3178

Ind. Eng. Chem. Res., Vol. 40, No. 14, 2001

Table 3. Parameter Values Used in the Column Design Calculations of HFP on Norit RB3 and Sorbonorit B3 parameter

Norit RB3

column length, L (m) column diameter, dcol (m) mass of bed, m (kg) bed porosity, b (-) equivalent particle diameter, deqv (mm) particle porosity, p (-) tortuosity, τ (-) capacity,a qsat (kg/mb3) D-R parameter,a B ) RT/βE0 (-) saturated vapor concentration,a Cs (kg/m3) molecular diffusivity,b Dm (m2/s) gas-phase density,c Fg (kg/m3) gas-phase viscosity,c η (kg/(m s))

1.5 1.3 960 0.31 3.33 0.314 4.0 424.4 0.1523 52.8 7.3 × 10-6 1.18 18.3 × 10-6

Sorbonorit B3 1.5 1.3 847 0.31 3.42 0.331 3.0 409.4 0.1603 7.3 × 10-6 1.18 18.3 × 10-6

a Dubinin-Radushkevich adsorption isotherm parameters at 303 K. b Diffusivity of HFP in air at 303 K and 101 325 Pa. c Gasphase properties of air at 303 K and 101 325 Pa.

Table 4. Flow Rate, Superficial Velocity, and Concentration for the Column Design Calculations of HFP on Norit RB3 and Sorbonorit B3 concentration,a cfeed (g/m3)

flow rate (m3/h)

superficial gas velocity, vs (m/s)

0 < t < 3.5 h

3.5 < t < 4 h

400 100 80

0.0837 0.0209 0.0167

0.5625 2.250 2.8125

2.8125 11.250 14.0625

a

Figure 14. Predicted effluent concentrations of HFP as functions of time for Norit RB3 at a temperature of 303 K. The effects of the tortuosity and use of the Langmuir isotherm model on the breakthrough behavior are shown. The idealized emission profile was taken as the feed concentration.

Based on the idealized concentration profile, see Figure 2.

Figure 15. Predicted effluent concentrations of HFP as functions of time for Norit RB3. Two cases are shown: (a) the minimum allowable operation flow of 80 m3/h compared with 100 m3/h at a temperature of 303 K, (b) an operation temperature of 283 K and a flow of 100 m3/h. The idealized emission profile was taken as the feed concentration.

Figure 13. Predicted effluent concentrations of HFP as functions of time for Norit RB3 and Sorbonorit B3. The calculations were performed for flows of 400 and 100 m3/h at a temperature of 303 K. The idealized emission profile was taken as feed concentration.

about 3 days for Sorbonorit B3 and 4 days for Norit RB3. This was expected because the capacity of Norit RB3 is larger in terms of column volume. It is possible to diminish the flow through the coagulators, which increases the concentration. A higher concentration results in a larger adsorption capacity and, thus, a longer cycle time. The column will be designed for 100 m3/h, although the absolute minimum is 80 m3/h (determined by the process). The feed concentration then becomes 2.25 g/m3 during the first 3.5 h and 11.25 g/m3 during the last 0.5 h of the cycle (1 ppm ) 6.25 mg/m3). From Figure 13, it is seen that the cycle time of the column is doubled, 6 days for Sorbonorit B3 and 8 days for Norit RB3. In addition, calculations were performed to show the influence of the model parameters on the breakthrough curve. These calculations were performed for Norit RB3, a flow of 100 m3/h, and a temperature of 303 K (unless mentioned otherwise). The tortuosity of Norit RB3 was set equal to 2 instead of 4, thus increasing the effective

diffusion inside the particles. Figure 14 shows that the effect of changing the effective diffusivity is rather small. The choice of Langmuir instead of DubininRadushkevich as the isotherm model has a dramatic effect on the breakthrough time, which is decreased from 8 to only 2 days. This is in agreement with the fits of the adsorption isotherm in Figure 3, where it is shown that the Langmuir isotherm underestimates the equilibrium amount adsorbed at relative pressures below 1.0 × 10-3 (the feed concentrations of 2.25 and 11.25 g/m3 correspond to relative pressures of 0.4 × 10-4 and 2.0 × 10-4, respectively). It is obvious that the correct isotherm is more crucial than the particle kinetics in designing an adsorption column. For the presented situation in particular and for column design purposes in general, shortcut calculations are recommended. If accurate data are available on adsorption equilibrium, the column lifetime can be well predicted on the basis of the adsorption capacity only. In that case, the calculation time is drastically reduced. A further decrease of the flow to the minimum allowable value of 80 m3/h results in a small improvement in the column performance (see Figure 15). A considerable improvement is obtained when the column is operated at lower temperatures. The operation time of the column is almost 2 weeks at a temperature of 283 K. In that case, of course, the extra cooling costs

Ind. Eng. Chem. Res., Vol. 40, No. 14, 2001 3179

have to be compared to the advantage of lower costs for the carbon inventory. In the case of a standard container filter, such as considered here, the carbon has to be disposed of after use. In practice, however, this loaded carbon cannot be handled because of the excessive fluor concentration present. This option is therefore not viable. A better alternative is to scale down the adsorber and add a small steam regeneration system for which the capital costs will be reduced. The predicted cycle time will become proportionally shorter as a result of this downscaling. An optimal regeneration temperature has to be chosen, one high enough to ensure fast desorption but not so high that it causes the occurrence of reaction of HFP. It was mentioned in the Introduction that the contaminated air stream also contains dust particles. This dust has not been considered in this study. It is clear that a filter to prevent clogging of the adsorption column is favorable. As an alternative solution, a monolithic structure could be used. It is expected that the dust particles will pass without any problem through the open channels of the monolith. Of course, the adsorption and diffusion parameters for the carbon will have to be determined in that case. Conclusions Filling the carbon bed of a laboratory-scale adsorption column with inert silicon carbide fines yielded better defined flow conditions. The breakthrough curves were steeper, and the reproducibility of the experiments increased. The use of SiC confirms the catalytic application of decoupling hydrodynamics from the rate processes. It enables the testing of large and real-sized (commercial) particles in small-diameter columns for adsorption purposes. Design calculations were performed on the basis of a standard container filter. A correct isotherm is more crucial than the particle kinetics, i.e., internal diffusion behavior, in designing an adsorption column. The Dubinin-Radushkevich model describes adsorption equilibrium substantially better than the Langmuir model. The choice of Langmuir instead of Dubinin-Radushkevich as the isotherm model has a dramatic effect on the predicted column performance. In the case when the stream contains water vapor, the cycle time is decreased. The effect of water vapor is expected to be negligibly small at relative humidities below 30%. Steam regeneration proved possible, but only in a small window of temperature; at temperatures higher than ∼393 K, reactions take place, leading to the evolution of, among other byproducts, HF and deposits on the activated carbon that lead to a strong reduction of the capacity. It is tentatively concluded that the reactions are oxidation reactions. The hydrolysis of HFP appears to be less likely. Generally, in process development, it is important to consider both the regeneration step and the adsorption step. Regeneration of activated carbon/halogen-containing hydrocarbon systems at excessive temperatures and in the presence of oxygen can lead to reactions. The environmental implications of fluorohydrocarbons, such as the green house effect, toxicity, and HF and other acid formation, are important motives for prohibiting the release of these compounds into the environment.

Acknowledgment DuPont de Nemours B.V. (The Netherlands) is gratefully acknowledged for its financial support and approval for publication. Notation a′ ) external surface area per unit particle volume, 1/mp b ) Langmuir equilibrium constant, Pa-1 B ) parameter in Dubinin-Radushkevich equation c ) gas-phase concentration, mol/mg3 c* ) concentration at the gas-particle interface, mol/mg3 cp ) intraparticle gas-phase concentration, mol/mg3 deqv ) equivalent spherical diameter, m Dax ) axial dispersion coefficient, m2/s Dm ) molecular diffusivity, m2/s E0 ) characteristic energy of adsorbent, J/mol kf ) mass transfer coefficient, mg3/mp2/s p ) pressure, Pa ps ) saturated vapor pressure, Pa pr ) relative pressure q ) amount adsorbed, mol/kg, mol/mp3 qsat ) amount adsorbed at saturation, mol/kg, mol/mp3 r ) radial coordinate inside a particle, m R ) gas constant, J/(mol K) t ) time, s T ) temperature, K vs ) superficial velocity, m/s z ) distance from the bed inlet, m Greek Letters β ) affinity coefficient of the adsorptive species b ) bed porosity, mg3/mr3 p ) macroporosity of the adsorbent particles, mg3/mp3 τ ) tortuosity

Literature Cited (1) Linders, M. J. G. Prediction of breakthrough curves of activated carbon based sorption systems: From elementary steps to process design. Ph.D. Dissertation, Delft University of Technology, Delft, The Netherlands, 1999. (2) Fuller, E. N.; Ensley, K.; Giddings, J. C. Diffusion of halogenated hydrocarbons in helium. J. Phys. Chem. 1969, 73, 3683. (3) Fuller, E. N.; Schettler, P. D.; Giddings, J. C. A new method for prediction of binary gas-phase diffusion coefficients. Ind. Eng. Chem. 1966, 58, 19. (4) Reid, R. C.; Prausnitz, J. M.; Sherwood, T. K. The Properties of Gases and Liquids; McGraw-Hill Book Company: New York, 1977. (5) Gunn, D. J. Axial and radial dispersion in fixed beds. Chem. Eng. Sci. 1987, 42, 363. (6) Ruthven, D. M. Principles of Adsorption and Adsorption Processes; John Wiley & Sons: New York, 1984. (7) Perry, R. H.; Green, D. W. Perry’s Chemical Engineers’ Handbook; McGraw-Hill Book Company: Singapore, 1984. (8) Wakao, N.; Funazkri, T. Effect of fluid dispersion coefficients on particle-to-fluid mass transfer coefficients in packed beds. Chem. Eng. Sci. 1978, 33, 1375. (9) Dwivedi, P. N.; Upadhyay, S. N. Particle-fluid mass transfer in fixed and fluidized beds. Ind. Eng. Chem. Process Des. Dev. 1977, 16, 157. (10) Schiesser, W. E. The Numerical Method of Liness Integration of Partial Differential Equations; Academic Press: London, 1991. (11) Silebi, C. A.; Schiesser, W. E. Dynamic Modeling of Transport Process Systems; Academic Press: London, 1992.

3180

Ind. Eng. Chem. Res., Vol. 40, No. 14, 2001

(12) Horvath, G.; Kawazoe, K. Method for the calculation of effective pore size distribution in molecular sieve carbon. J. Chem. Eng. Jpn. 1983, 16, 470. (13) Barrett, E. P.; Joyner, L. S.; Halenda, P. P. The determination of pore volume and area distributions in porous substances. I. Computations from nitrogen isotherms. J. Am. Chem. Soc. 1951, 73, 373. (14) Schoene, K.; Bo¨hmer, W.; Steinhanses, J. Determination of vapour pressures down to 0.01 Pa by headspace gas chromatography. Fresenius’ J. Anal. Chem. 1984, 319, 903. (15) Linders, M. J. G.; van den Broeke, L. J. P.; van Bokhoven, J. J. G. M.; Duisterwinkel, A. E.; Kapteijn, F.; Moulijn, J. A. Effect of the adsorption isotherm on one- and two-component diffusion in activated carbon. Carbon 1997, 35, 1415. (16) Wolff, E. H. P. Regenerative sulfur capture in fluidized bed combustion of coal: A fixed bed sorption study. Ph.D. Dissertation, Delft University of Technology, Delft, The Netherlands, 1991. (17) Sie, S. T. Miniaturization of hydroprocessing catalyst testing systems: Theory and practice. AIChE J. 1996, 42, 3498.

(18) Hall, C. R.; Lawston, I. W.; Tinsley, A. M. Protection provided by activated carbon vapour filters against perfluoroisobutene. Chem. Ind. 1989, 145. (19) Martini, T.; Schumann, C. Uber die umsetzung von hexafluorpropen und perfluor-2-methyl-2-penten mit wasser. J. Fluorine Chem. 1976, 8, 535. (20) Smith, J. M.; Stammers, E.; Janssen, L. P. B. M. Fysische transportverschijselen I; Delftse Uitgevers Maatschappij b.v.: Delft, The Netherlands, 1981. (21) Cheremisinoff, N. P.; Cheremisinoff, P. N. Carbon Adsorption for Pollution Control; Prentice Hall: New York, 1993. (22) Datasheets from Norit N.V., Amersfoort, The Netherlands, 1995.

Received for review October 30, 2000 Accepted May 3, 2001 IE000932B