Bulk gas separation by pressure swing adsorption - Industrial

---

758

Ind. Eng. Chem. Fundam. 1986, 25, 758-767

Bulk Gas Separation by Pressure Swing Adsorption Peillng Cen and Ralph T. Yang’ Department of Chemical Engineering, State University of New York at Buffalo, Amherst, New York 14260

An experimental and theoretical study has been performed on bulk gas separation by pressure swing adsorption using activated carbon. A 50/50 H,/CO mixture was separated into two products, both at well over 90% purities and recoveries and at a high throughput. The data were successfully simulated by two models, both considering an energy balance and adsorption from all components in the mixture. The purge-to-feed ratio was important to the H, product purity only at low feed pressures. The CO product purity depended crucially on the step of cocurrent depressurization in the cycle. The function of this step was illustrated by a bed-loading analysis.

Introduction Pressure swing adsorption (PSA) is a gas separation process in which the adsorbent is regenerated by rapidly reducing the partial pressure of the adsorbed component, either by lowering the total pressure or by using a purge gas. In the original PSA cycles, invented by Skarstrom (1960), two steps (adsorption and depressurization/purge) are carried out in two adsorbent beds operated in tandem, enabling the processing of a continuous feed. Since introduction of the Skarstrom cycle, many more sophisticated PSA processes have been developed and commercialized. Such processes have attracted increasing interest more recently because of their low energy requirements and low capital investment costs (Stewart and Heck, 1969). Start-of-the-art reviews of the PSA processes have been made by Keller (19831, Cassidy and Holmes (19831, Wankat (19811, and Yang (1986). In modern PSA processes, three or more beds are used to synchronize and accommodate steps additional to those in the Skarstrom cycle: cocurrent depressurization and pressure equalization. The two major applications of PSA have been air-drying and hydrogen purification. Other commercial applications are production of oxygen and nitrogen from air, octane improvement (using the Union Carbide ISOSIV process), and solvent recovery. A one-bed variation of PSA (Turnock and Kadlec, 1971; Jones and Keller, 1981) has been recently commercialized for medical oxygen generation purposes. Other PSA variations include the “vacuum swing adsorption” processes in which the pressure is varied between atmospheric and mechanical pump vacuum (Sircar, 1977-1979), “complementary PSA” in which sorbents of different (or opposite) selectivities contained in columns arranged in series or parallel are used (Knaebel, 1984; Wankat and Tondeur, 1984), and techniques for reducing the adverse temperature excursions in PSA cycles for bulk separations by using heat-exchange beds or high heat capacity inert additives in the bed (Yang and Cen, 1986). Despite the increasing industrial use of PSA, theoretical understanding of the process is still in a primitive stage. Brief reviews of the theoretical developments have been made by Cheng and Hill (1983), Wankat (1981), and Chen and Yang (1985). Linear adsorption isotherms are used in most of the published PSA models, thus limiting the models to adsorption of dilute species. Also, local equilibrium between the bulk gas phase and the intraparticle adsorbed phase is assumed, implying no mass-transfer limitations. The only exceptions are the model by Chihara and Suzuki (1983) and that by Carter and Wyszynski (1983) where the linear driving force approach for masstransfer rates is used. The analytical approach of the 0196-4313/86/1025-0758$01.50/0

equilibrium models is similar to earlier models for the process of parametric pumping (Sweed and Wilhelm, 1969; Jenczewski and Myers, 1968; Pigford et al., 1969; Aris, 1969), including the complexities of the blowdown and pressurization steps. The method of characteristics is used in these models to derive simple algebraic equations in order to estimate the steady-state performance for simplified PSA cycles. The equilibrium models include those of Shendalman and Mitchell (1972), Weaver and Hamrin (1974), Fernandez and Kenney (1983), and Hill et al. (1982), for one adsorbate, and Chan et al. (1981), Nataraj and Wankat (1982), and Knaebel and Hill (1985), for two or more adsorbates. The other approach involves numerical solution of mass and energy balance equations plus the mass-transfer rate equations and provides better approximations a t the expense of increased computation time (Chihara and Suzuki, 1983a; Carter and Wyszynski, 1983). All of the above-mentioned models pertain to dilute systems, and thus linear isotherms are used. In addition, the mass-transfer limitation in PSA is approximated by the linear driving force approach. In bulk separation, Le., for gases containing high concentrations of adsorbates (more than 10 wt % according to Keller (1983)),linear isotherms can no longer be used. The performance of PSA for bulk separation of a 50/50 Hz/CH4mixture has been successfully simulated by a pore diffusion model (Yang and Doong, 1985), which embodies adsorption isotherms for mixtures, pore diffusion in each particle, an energy balance, and modeling of all basic steps in the PSA cycle. This model is an extension of a pore-diffusion model for temperature swing separation (Tsai et al., 1983), with additional simplifying assumptions and, hence, less numerical computation. An interesting noncyclic countercurrent flow model has been proposed by Suzuki (1984), which may simulate the steady-state PSA behavior when the steps involving pressure changes are relatively short. The separation of Hz/CO mixtures is an important step in fuel, synthetic, chemical, and other industries. Beside the cryogenic process, there are two new commercial processes for Hz/CO separation. The first is the COSORB process, based on absorption of CO in toluene solutions of cuprous tetrachloroaluminate, and the second is stripping by heating and depressurization (Walker, 1975). High-purity CO (over 99%) can be produced by this process. However, the energy requirement is high and the copper solvent is easily deactivated by trace amounts of polar gases such as H20, SOz, and H,S. They must be removed to less than 1-ppm concentration for the process. The more recent one is the PRISM separator based on the faster permeation of H2 through a bundle of hollow mernbranes (Narayan et al., 1981). The separation ratios are 1986 American Chemical Society

Ind. Eng. Chem. Fundam., Vol. 25, No. 4, 1986

low, especially for CO, and the product recoveries are consequently low. The required differential pressure is high, typically 500 psi. This paper describes experimental and theoretical results of separation of H 2 / C 0 mixtures by PSA with activated carbon as the sorbent. Both components are recovered as products. An equilibrium model and a linear driving force model are presented and compared. In both models adsorption from mixture and energy balance are accounted for.

Process Description In a typical PSA process, each bed goes through some or all of the following five basic steps forming a cycle: (I) pressurization, (11)adsorption, (111)cocurrent depressurization, (IV) countercurrent blowdown, and (V) purge. Two of more interconnected beds are operated in-phase so continuous feed and product flows are possible. All five steps are used in separation of HS/CO. In step I, the bed is pressurized to the feed pressure by H2, which is obtained from another bed. In step 11, the high-pressure feed flows through the bed while a portion of the effluent H2 product is used to purge another bed at a reduced pressure. Step 111, cocurrent depressurization or blowdown, is used to recover the H, remaining in the voids and to allow time for CO to desorb in the bed. The end pressure of step I11 depends on bed utilization, i.e., the fraction of the bed traversed by the CO wave front. Steps IV and V, both operated countercurrent to the feed direction, produce CO product and provide a cleaned bed for the next cycle. H2 is used in step V. The performance of a PSA process is determined by three interrelated sets of results: product purity, product recovery, and productivity. The productivity is judged as the rate of feed processed per unit amount of sorbent. For bulk separation, both H2 and CO must be considered for product recovery and purity. Nonisothermal Models The temperature excursions in a PSA cycle for bulk separation are not negligible as in the air-drying process. For example, certain points in the bed experienced a 40 "C swing during a cycle in our experiments, whereas a 5 "C swing was encountered in air-drying (Chihara and Suzuki, 198313). Energy balance equations must, therefore, be included in the models. The simplifying assumptions and approximations made in the models are the following: 1. The ideal gas law applies. (The compressibility factor for the gas mixture was calculated to be 0.99 under our experimental conditions of 34 atm and 25 "C.) 2. The axial pressure gradient across the bed is neglected. 3. Thermal equilibrium is assumed between the fluid and the particles. 4. No variation exists in the radial direction for both concentration and temperature. 5. Due to the narrow temperature range of the process, ka in the linear driving force model is assumed to be independent of temperature. The value of ka is further assumed to be the same for both H2 and CO. The total material balance in the column can be written as

Material balance for component i yields a(uYj) € A ~ ( P Y J T ) + -AP - - a(4xi) - 0 az + -R at at

v,

(2)

759

In the equilibrium model, the amount adsorbed is equal to the equilibrium amount

aq - = -aq* (3) at at whereas in the linear driving force model (approximating intraparticle diffusion) aq - = Ka(q* - q ) (4) at The equilibrium adsorption from the CO/H2 mixture on PCB activated carbon has been measured and expressed as (Ritter, 1985) (5)

These equations are the loading ratio correlation used previously for zeolites (Yon and Turnock, 1971). The use of these noniterative equations substantially simplifies the computation. The constants are qmi = ai + bi/T and Bi = exp(ci + di/T) (6) The energy balance in the column is given by

The molar heat capacity of the adsorbed phase is assumed to be equal to that in the gas phase, thus

c,, = c,

= CCPiyi

(8)

and C,i = Ai

+ BiT + C i P + D i p

(9)

The initial conditions for the PSA process are, a t t = 0, which is the time after step I (H2pressurization) after start-up of the process Y A = X A = 1.0 YB X B = 0.0

P = Po; T = To; = 4 0 (10) The boundary conditions for the ensuing cycles are as follows. step 11: at z = L (the feed end), YA = ym; = ysf; P = Pf; T = T f ; u = uf a t z = 0, P = Pf step 111: at z = L , u = 0 a t z = 0, u = u ( t ) or P = P(t) step IV: at z = L, u = u ( t ) or P = P ( t ) atz=O, u=O step V: at z = L, u = u ( t ) or P = P(t) a t z = 0, u = upurge step I: (hydrogen pressurization from the end opposite to the feed) a t z = L , u=O at z = 0, u = u ( t ) or P = P ( t ) The final state of each cycle (step I1 to step I) is a t t = At, P = Pf The initial conditions for step I1 in the next cycle are the same as the final conditions of step I, thus continuing the cyclic process.

760

Ind. Eng. Chem. Fundam., Vol. 25, No. 4, 1986 Table I. Characteristics of AdsorDtion column height, L column inside radius, r activated carbon, U J particle size, dp bulk density, p interparticle void fraction, cy intraparticle void fraction, E total void fraction heat capacity of carbon, C,,

F i g u r e 1. Schematic diagram of apparatus for pressure swing adsorption for gas separation: SP, sampling port; PG, pressure gauge; CV, check valve; PT, pressure transducer; SV, solenoid valve; TC, thermocouple; LPR, line pressure regulator.

Method of Solution. Equations 1-3 (for the equilibrium model) or eq 4 (for the linear driving force model), 5, and 7 are solved for u , t , q , q * , qI*,y I , and I,. These equations have been solved by the finite-difference backward method. An iteration scheme has been used. In a typical computation 30 space steps and 1860 time steps were used in a whole PSA cycle. In all calculations, P = P ( t ) was used for the boundary conditions because the entire pressure history was recorded by a pressure transducer / recorder. Experimental Section The experimental apparatus was designed for simulating all five basic steps in the PSA process, capable of widerange conditions (high pressure, temperature, flow rate, adiabatic, isothermal, etc.) for multicomponent, bulk separation. The apparatus was automated and controlled, the only manual operation being sample collection and analysis and the recording of flow rates. Apparatus for PSA. A schematic of the apparatus is shown in Figure 1. The adsorption column was a stainless steel (Type 304) pipe 60 cm long and 4.1-cm i.d. The bottom plate of the column was a stainless steel sintered plate that was glazed to the column. The column was packed with 20-60-mesh activated carbon (designated PCB, manufactured by the Calgon Corp.). Glass wool was compressed on top of the bed to prevent the carry-over of particles. An additional in-line filter (60-pm holes) was installed to keep fine particles from entering the gas lines. All lines were 1/4-in.stainless steel. Three solenoid valves located at the feed, cocurrent end, and countercurrent end points were used alternatively to direct the flow into and out of the column. These solenoid valves were activated by electronic timers, which were preset to operate at desired time cycles. A check valve in the purge line prevented any backflow. A pressure transducer outside the bottom of the column was connected to a recorder which provided the pressure history of the process. In addition to the transducer, a pressure gauge connected at the top of the column provided easy reading of pressure. A layer of insulation covering the column helped both to make the temperature gradient in the radial direction as small as possible and to simulate adiabatic operation. These are approximately the conditions existing in commercial PSA processes where large-diameter beds are used. The axial temperature distribution was measured and recorded a t three locations in the bed (nominally top, middle, and bottom). This was done by using three fine thermocouples sheathed in a l/s-in. thin stainless steel protection tube which was inserted in the center of the packed column. Gas samples were taken from two sampling ports, which were fitted with septa, by syringes. The syringes were equipped with locks so samples could be collected in short

Column 64 cm 2.05 cm 412 g 0.055 cm 0.498 g/cm3 0.43 0.61 0.78 0.25 cal/(g "C)

Table 11. Loading Ratio Eq 5 a n d 6 P a r a m e t e r s a n d Heat of Adsorption n

a

b

C

d

H. kcal/mol

0.97 1.02

87.68 65.11

42392 17992

-12.336 -9.442

1219.3 1286.3

2.6 2.6

~

H 2

CO

time intervals (ca. 20-50 s) and stored for later GC analysis. Experimental Procedure. The procedure described below was for bulk separation of a Hz/CO (50/50 volume) mixture. The bed was cleaned before each run by degassing with a mechanical pump. Step I, pressurization, was initiated by opening the solenoid valve connected to Hz. The desired column pressure was controlled by the pressure regulator connected to the Hz cylinder. Step 11, high-pressure adsorption, started when the bottom solenoid valve was opened by the timer. The flow rate in step I1 was controlled by adjusting a needle valve in the line. Step 111,cocurrent depressurization, was effected by closing the feed valve. Step IV, countercurrent blowdown, was achieved by simultaneously closing the bottom valve and opening the valve in the top exhaust line. When the column pressure was lowered below 3 psig, the check valve opened, which started the hydrogen purge step, step V. Experimental PSA Cycle. In all but one experiment, the total cycle time was 15.5 rnin with the following distribution: step I (pressurization): 0.5 rnin step I1 (feed): 4 min step I11 (cocurrent depressurization): 6 min step IV (countercurrent blowdown): 1.5 min step V (purge): 3.5 min The experiments were designed to obtain an understanding of the effects of various operating parameters on separation: feed pressure, purge/feed ratio, feed rate, cocurrent depressurization, etc. The feed pressure was 200, 300, or 500 psig (or 14.6, 21.4, and 35.0 atm). The productivity was in the range of 244-416 L (STP) of feed/(h kg of carbon), which was in the range for commercial PSA processes. Because of the relatively long cycle time, steady state was reached rather rapidly, both by experiment and by model simulation. Generally, five cycles were required to reach steady state after start-up of the process with a clean bed.

Model Input Parameters The characteristics of the adsorber column are given in Table I. The parameters for heat capacities of H, and CO are given in Table 11. Additional input data necessary for the models are equilibrium amount of adsorption of mixture per unit mass of carbon, heats of adsorption, and the heat- and mass-transfer coefficients. Equilibrium Adsorption Data. Single-gas isotherms for the specific system under this investigation (H, and CO on PCB activated carbon) have been measured by Ritter in this laboratory (19851, covering the temperature

Ind. Eng. Chem. Fundam., Vol. 25, No. 4, 1986

Table 111. Heat Cauacity Ea 9 Parameters A B c

H2 CO

6.483 7.373

2.215 X -3.07 X loF3

-3.289 X lo4 6.662 X lo+

D 1.826 X -3.037 X

range 295-375 K and pressures up to 1000 psig. The data were best fitted by a hybrid Langmuir-Freundlich isotherm:

For adsorption of mixtures, it has been shown that the loading ratio correlation (LRC) equations, eq 5, can be derived from the single-gas isotherms, eq 11 (Chen and Yang, 1985; Yon and Turnock, 1971). The LRC equations fit the data for adsorption of mixtures as closely as the ideal adsorbed solution theory and the Polanyi potential theory. For adsorption of the H 2 / C 0 mixture on activated carbon in the T and P ranges used in this study, the LRC equations fitted the experimental data within f15% (Ritter, 1985). However, they do not require iteration as the two thermodynamic theories do, and thus they reduce the amount of numerical computation. The parameters for the LRC equations for the H,/CO/activated carbon system are listed in Table 111. The isosteric heats of adsorption were also measured. These values are rather constant over surface coverage and are also listed in Table 111. Heat-Transfer Coefficient. A layer of insulation was fitted around the adsorber column in an attempt to simulate adiabatic conditions. However, it was found in early model computations that the temperature history in the bed and, consequently, the separation results were very sensitive to the heat-transfer term in the energy balance equation, eq 7. It was found necessary not to neglect this term, i.e., r d h ( T - To). The temperature sensitivity is inherent to the cyclic process, not due to the models. Neglecting this term, i.e., for an adiabatic system, the temperature excursion will grow cumulatively to reach nearly 100 "C in a steady-state cycle (Yang and Cen, 1986). A detailed calculation has been made for the value of the overall heat-transfer coefficient, h, for the heat-transfer rate between the bed (at T ) and the ambient air (at To). Four consecutive resistances were bed-to-steel wall, conduction in steel wall, conduction in the insulation layer, and free convection, the resistance in the insulation layer being dominant. Since h varied only slightly with bed temperature, a constant value was used in the model. The constant h value was taken as 9.58 X cal/(cm2 K s), which corresponded to T = 40 "C and To = 20 "C. Other input parameters included pressure history, P ( t ) , feed flow rate, uo, and purge flow rate, upurge.P ( t ) was measured and recorded throughout the experiments. The value of uo was calculated by mass balance based on the effluent flow rate and concentration, both measured. The purge flow rate was also obtained from measured data. For both equilibrium and linear driving force models, we obtained a computation time of approximately 1.5 min/cycle using a VAX-780 computer; approximately five cycles were required for the process to reach a steady state, which agreed with experimental observation.

Results and Discussion In all commercial PSA processes, only the raffinate, i.e., the weakly adsorbed component which escapes the bed in adsorption step, is the desired product. Either the strongly adsorbed component is unwanted, e.g., in air-drying and air separation, or its purity is unimportant as in hydrogen

761

purification. In this study both components are wanted products. The major problem is to recover a high-purity product composed of the strongly adsorbed component; it is relatively easy to obtain a high-purity raffinate. Experiments have been performed to separate a 50/50 (by volume) mixture of H 2 / C 0 under various PSA conditions. The goal of the experiments was to obtain highpurity products of both Hz and CO. The cocurrent depressurization (step 111) was found necessary to recover CO; a high-purity CO product could be obtained by lowering the end pressure of step 111, at the expense of Hz product purity and the recoveries of both products. Presentation of Data of a Typical Run. For all experimental data and model simulations, the main results will be presented in terms of the following three important quantities: concentrations of (both) products, recoveries of (both) products, and sorbent productivity (as the throughput of feed), all at steady state. The effluents from steps I1 and I11 are collected as the H2-rich product; those from steps IV and V are the CO product. The product purity is calculated as the volume-averaged concentration over the entire product, expressed in percent by volume. The feed rate or throughput is presented as the total feed volume (at STP) per cycle, which can be converted into sorbent productivity by dividing by the amount of carbon sorbent. The feed rate (in step 11) was not measured. Instead, it was calculated by mass balance based on the measured CO effluents in steps 11-V, as the concentration and flow rate of the effluent were measured in these steps. The product recoveries are defined as Hz from I1 and I11 - H2 used in I and V Hz recovery = Hz in feed (in 11) (12) CO from IV and V CO recovery = (13) CO in feed (in 11) The detailed experimental data are presented in Table IV for a typical run, run 1. These were for steady-state operation, which was reached after approximately the 5th cycle from start-up with a clean bed. The pressure swing in each cycle for run 1 was 35 atm for step 11, 8.5 atm at the end of step 111, and 1.4 atm for pure (step V). (The pressure history is also depicted in Figure 2.) The experimental results of run 1showed that product purity and recovery for both components of well over 90% could be achieved. The adsorbent productivity for run 1 was 400 L (STP)/(h kg), a moderately high value as compared with that from commercial PSA processes. The values predicted by the equilibrium model and by the linear driving force model are also given in Table IV. The model predictions were reasonably good. Note that the lineardriving-force model could predict the behaviors of all steps in PSA, whereas in the previous mass-transfer (nonequilibrium) model, the three important steps involving pressure changes (I, 111, and IV) were assumed "frozen"; i.e., no adsorption or desorption was allowed to occur (Chihara and Suzuki, 1983a). As shown in Table IV, all process characteristics for all five steps could be predicted by the models. (Step I is not shown as there was no effluent from this step.) As shown in Table IV, as well as in results to be given later, the linear-driving-force (LDF) model was slightly better than the equilibrium (EQ) model. Dynamic Behavior in Bed. It is instructive to visualize the temperature and concentration distributions in the bed and how they vary during a steady-state cycle.

762

Ind. Eng. Chem. Fundam., Vol. 25, No. 4, 1986

Table IV. Steady-State Results of Run 1 of a 50/50 Mixture of HJCO" experiment EQ model step time, s yco, % flow rate, L (STP)/min flow rate, L (STP)/min I1 30 7.22 0.00 8.70 7.22 90 8.70 0.00 7.21 150 0.02 8.64 210 7.22 0.00 8.56 I11 270 4.11 0.04 2.81 2.95 330 4.10 0.07 2.75 3.05 390 1.63 3.53 "91 2.85 450 2.08 2.99 510 60.93 570 75.45 1.95 3.21

LDF model

70

0.01 0.02 0.03 0.05

flow rate, L (STP)/min 7.24 7.74 7.71 7.63

yco, Yo 0.01 0.02 0.04 0.08

0.08 0.19 1.14 5.88 63.75 80.39

3.04 3.11 3.18 2.98 2.65 2.80

0.16 0.28 1.47 6.94 58.38 76.51

yco,

IV

615 645 67.5

9.30 9.02 9.02

89.89 97.34 97.29

8.06 6.11 4.92

88.03 92.61 96.48

8.66 6.82 5.77

87.88 93.70 98.06

V

705 735 765 795 825 855 885

5.62 2.52 1.42 1.12 1.00 0.90 0.83

97.22 97.04 96.91 97.31 93.19 89.97 84.43

3.78 3.50 2.28 1.49 1.45 1.41 1.38

96.63 94.64 92.20 89.47 86.83 84.49 82.34

3.92 3.55 2.41 1.52 1.46 1.41 1.38

96.66 94.04 92.08 88.41 87.67 85.76 84.72

The pressure history is shown in Figure 8. 100

I

Y2.0

I

1

I

I

#

80

Q

i 60

Q

z

$- 4 0

; 4 !I

20

0

20

40

60

DISTANCE FROM FEED END, C M

Figure 3. Gas-phase CO concentration profiles in bed in a steadystate PSA cycle for run 1 as predicted by equilibrium model. Time distribution: step I, 0.5; 11, 4; 111, 6; IV, 1.5; V, 3.5 min.

I

-n

n.

&

0

3

I

1

20

E 100

5

10

15

TIME, min Figure 2. Steady-state PSA temperature histograms (for run 1) at 12.7 (A), 33.0 (B), and 53.3 cm (C) from the top of the bed (60-cm height): solid line, experiment; dashed line, equilibrium model.

The experimental data on temperatures at three points (upper, middle, lower) in the bed during a steady-state cycle in run 1 are shown in Figure 2. The temperatures predicted by the EQ model are also shown in Figure 2; they compared reasonably well with the data. It was observed that a temperature excursion as large as 40 O C was reached a t the middle point during each cycle. The temperature histories are helpful in understanding the dynamics of the process. A practical utility of these data is the indication of the location of the mass-transfer zone (MTZ) or the wave front of the strongly adsorbed

component (CO). (Direct measurements of concentration profiles in the bed are far more difficult.) The heads of adsorption of CO and H2 are nearly the same. However, much more CO than H2was adsorbed; hence, the CO wave front was accompanied by a peak or rise in temperature. The data in Figure 2 indicated that the CO wave front had passed the midpoint of the bed at the end of step 11, but only slightly. The large temperature peak in curve C (for the lower point in bed) indicated the readsorption of a large amount of CO desorbed from the upper portion of the bed during the cocurrent depressurization step. The desorption of CO (endothermic) in the upper portion of the bed during step I11 was also indicated by the decline in temperature. Further desorption in steps IV and V was also indicated by the continuous decline in temperature. A small temperature rise occurred in step 5 (H, purge). This was not seen in run 1 but was predicted by the models and was seen in some other runs. It was due to heating by the purge gas, which was at 20 "C. The concentration profiles in the bed and their variations during a steady-state cycle are shown in Figures 3 and 4. These data were simulated by the EQ model. From these figures, the location of the wave front and the sharpness of the wave front are both seen. These are two crucial factors determining respectively the CO product

Ind. Eng. Chem. Fundam., Vol. 25, No. 4, 1986 763

\ 0

20

40

\

BO

\

D I S T A N C F n D M FEED E N D , C M

Figure 4. Concentration profiles of CO in adsorbed phase in bed in a steady-state PSA cycle for run 1 as predicted by the equilibrium model. The time distribution is given in Figure 3.

purity and the Hz product purity, as will be discussed in the following section. A further discussion on the simulated concentration data will be given later. Two Major Factors Determining Purities of Products. For a high purity of raffinate (H, in this case), it has been well-known that a sharp wave front or concentration profile is necessary in the MTZ. A sharp wave front can be achieved by using a high purge/feed ratio, with the weakly adsorbed component in step I, and by operating under conditions which result in rapid pore diffusion and low axial dispersion in the bed. For a high purity of the strongly adsorbed component (CO in this case), it has been found in this study that the fraction of bed covered by the wave front, Le., the bed utilization in step 11, is a crucial factor and that step I11 is necessary to recolier CO. This effect can be understood by the analysis shown below. Bed-Loading Analysis. The bed contains two phases of the mixture: the gas in the voids and the adsorbed phase on the adsorption surface. The void fraction is usually very high, 0.78 in our experiments which corresponded to 1.61 cm3 void space/g of carbon. The voids behind the concentration wave front (or covered by the MTZ) hold a significant amount of H,, which limits the CO product purity and H2 recovery. In the zone covered by the concentration wave front, the ratio of the amounts of CO to H2 held in the bed is a strong function of total pressure and gas-phase concentration, as shown in Figure 5. The amount of each component is calculated as the sum of the amounts in the gas phase (from the ideal gas law) and in the adsorbed phase (from the loading ratio correlation). The role played by step I11 can be understood from Figure 5. The ratio of CO/H2 in the zone covered by the wave front a t the end of step I1 lies on curve I1 in Figure 5. Without step 111, going directly to steps IV and V, curve I1 sets the limit of the CO product purity. If we use run 1 as an example, point A in Figure 5 indicates the ratio of total amounts of CO/H2, or 3.2. Without step 111, the highest purity of CO that steps IV and V can accomplish is 76% (3.2/4.2). With step 111, the total pressure is decreased and the gas-phase concentration of CO in the zone covered by the concentration wave front is increased, e.g., to over 80% due to desorption. (See Figure 3, a t t = 10.5 min.) The ratio of CO/H2 in the bed covered by the wave front is shifted to a much higher value, e.g., point B on curve IV. The difference between curves I1 and IV is due to the amount of H, eluted out of the wave front and subsequently out of the bed during step 111. This is pos-

I

I

I

10

20 TOTAL

30

0

PF)EBBUEIEr DTF

Figure 5. Total loading analysis. Ratio of total (adsorbed plus gas phases) amounts in activated carbon with 0.78 void at 300 K, at gas phase CO/H2 = 1 (11) and 4 (IV). A and B indicate conditions at respectively the end of step I1 and the beginning of step IV. Line I11 indicates the shift performed by step 111.

sible because the bed not utilized in step I1 is used to adsorb CO and a small amount of H2 during step 111. Therefore, it is seen that the role of step I11 is to shift the loading ratio from curve I1 to curve IV. From the loading aiialysis, the effects of the pressure in step I1 and the end pressure of step I11 can also be understood and predicted. A lower pressure in step I1 will favor CO product purity but will decrease the throughput or productivity. A lower end pressure in step I11 will also increase CO product purity. However, this will allow breakthrough of CO in step I11 and will lead to lower H, product purity and to lower the recoveries for both products. Thus, optimal conditions must be found depending on the specific requirement for separation. Effect of Purge-to-FeedRatio. Purge of the bed by the weakly adsorbed component in commercial processes is used primarily to clean the bed in order to obtain high-purity raffinate. The effect of purge on the raffinate purity has been relatively well understood as it has been studied in most of the previous PSA literature. Since the lengths of time for purge and feed differed in this study, the purge-to-feed (P/F) ratio was defined as amt (L (STP)) of H 2 used in step V P/F = (14) amt (L (STP)) of H, feed in step I1 The effects of P/F ratio on bulk separation were studied in four experiments (runs 2-5) a t a feed pressure of 21.4 atm (300 psig), two experiments (runs 6 and 7) at 14.6 atm (200 psig), and two (runs 1 and 9) a t 35 atm (500 psig). Other conditions were maintained as nearly the same as possible. The results of runs 2-5 are shown in Figure 6 and in Table V. The results of runs 6 and 7 are shown in Table VI. Results obtained from simulation by the EQ and LDF models are also shown and compared. Although the LDF model was slightly better than the EQ model, both models were capable of predicting the effects. As in the commercial PSA processes, the major result of purge is to increase the product purity of H,. Results in Tables V and VI showed that the H2 product purity

Ind. Eng. Chern. Fundarn., Vol. 25, No. 4 , 1986

764

Table V. Steady-State Results of PSA Separation for 50/50 H,O/CO Mixture with Different Purge/Feed Ratio" 4 3 31.9 34..3 I .46 2.3 0.092 0.1:14

____

__

LDF input H,, L, output, I. ?'Ha, 55 output. I. >'H1, '2 output. 1, ,IC".

13.57 19.32 99.76 12.66 71.70 9.2(1 51.1 .:I2 ?.E5

'i;

output. 1, .\C().

'15.8:;

%

H, reco\er\ % C O reco\er\ q r

9609 7 5 71

9753

7061

9796 82 86

9588

'35.72

-1 50

-r

d..~.80

expt 12.93 19.17 99.99 12.36 84.,51 9.91 95 11 4.85 95.00

EQ 15.06 21.54 99.92 13.91 70.94 9.37 92.87 3.60 88.58

95.15 88.00

93.23 74.57

LDF

~

LDF

13.45 21.09 99.90 11.75 72.93 10.08 92.65 1.89 88.71

expt 11.46 18.63 99.99 13.48 86.83 8.84 95.49 7.02 96.42

EQ 15.01 24.18 99.86 13.17 69.12 10.21 93.34 4.05 86.11

12.74 22.93 99.82 11.64 69.69 10.38 93.33 1.08 86.69

92.63 80.19

96.21 89.30

92.94 76.09

94.87 80.87

"Adsorption pressure, 21 4 atm, LLcle time, 1 5 imin

Table VI. Steady-State Results of PSA Separation for 50/50 H,O/CO Mixture with Different Purge/Feed Ratio" run 6 27.5 29.4 teed, L purge H,, L 0.8 2.06 ourneifeed ratio 0.058 0.15

-_step

I IT

-~ input H,, I. output, I, >H2*

111

rv v

YC

output, I, \H1.

%

output, I, ?cos 7 c output. I, .\cos 7c

2 0

-

c

a a

expt

EQ

LDF

expt

9.3 17.27 99.90 8.23 71.52 7.02 94.36 4.88 97.61

9.75 18.25 99.71 9.02 (59.82 6.62 93.91 4.16 92.44

8.62 17.40 99.68 8.30 61.21 6.80 94.7:1 4.42 91.26

95.36 82.60

94.88 73.21

94.%58 95.02 76.18 85.15

_

_

9.2 17.26 99.99 9.97 79.94 6.98 93.87 6.45 95.29

EQ

_

b .

~

2

9.96 18.98 99.80 10.6