Znd. Eng. Chem. Res. 1988,27, 81-85 Smith, 0. J. M. Chem. Eng. Prog. 1957,53(5), 217-219. Vogel, E. F.; Edgar, T. F. Proceedings of Joint Automatic Control Conference, San Francisco, CA, 1980; paper TP5-E. Wellstead, P. E.; Zanker, P. Opt. Control Appl. Methods 1982, 3, 305-322. Widrow, B.; McCool, J. M.; Larimore, M. G.; Johnson, C. R. Proc. IEEE 1976, 64, 1151-1162. Widrow, B.; McCool, J. M.; Madoff, B. P. Conf. Rec. of 12th Asilomar Conference on Circuits, Systems and Computers, Asilomar, NOV1978; pp 90-94.
81
Widrow, B.; Shur, D.; Shaffer, S. Conf. Rec. of 15th Asilomar Conference on Circuits, Systems and Computers, Nov 1981; pp 185-189. Widrow, B.; Walach, E. Preprints of IFAC Workshop on Adaptive Systems in Control and Signal Processing, San Francisco, CA, 1983. Received for review April 17, 1986 Revised manuscript received August 24, 1987 Accepted September 24, 1987
Oxygen Enrichment by Pressure Swing Adsorption Anthony S. T. Chiang,* Mau-Yeuh Hwong, Ting-Yueh Lee, and Tsao-Wen Cheng Department of Chemical Engineering, National Central University, Chung-Li, Taiwan, R.O.C. 32054
An automated experimental apparatus was build to study the PSA oxygen enrichment process. Mass flux and concentration of all streams connected to the adsorption bed were continuously monitored so that the transient behavior of the process could be examined. In particular, a four-step process operated under various production, pressurization, and blowdown rates was studied in detail. The experimental data were then compared with the simulation results of a simple, isothermal model. The adsorption isotherm and t h e mass-transfer rate constants used in the model were taken directly from literature data. T h e simulated and the experimental transient results were in reasonable agreement when the purge step was long, but for cases with a short purge step, a slower than predicted desorption rate was observed. T h e thermal effects not considered in the model were believed t o be responsible for this difference. The principle of oxygen enrichment from air by a PSA process is relatively simple. Typical adsorbate includes zeolites A and X and mordenite. Compressed air enters the adsorption bed a t a high pressure. Since nitrogen is adsorbed appreciably more than oxygen at high pressure, oxygen-enriched air can be easily obtained at the outlet. The bed is then blown down to a lower pressure to desorb the nitrogen in zeolite. A purge step may sometimes be used to sweep the nitrogen out of the bed before a new cycle begins (Skarstrom, 1972). There has been a variety of mathematical models to describe a PSA process. Early models were mostly based on a local equilibrium assumption. For example, Shendalman and Mitchell (1972) developed a model assuming a linearly adsorbed trace component at local equilibrium conditions. Weaver and Hamrin (1974) assumed a constant separation factor and local equilibrium conditions to study the separation of hydrogen isotopes. These models were however only good for a purification process. An important difference between purification and bulk separation is that the gas velocity in bulk separation will vary along the length of the column due to the adsorption. This change of gas velocity was first considered by Fernandez and Kenney (1983). Recently, Cheng and Hill (1985) studied the separation of He and methane. A finite mass-transfer rate was considered in addition to the change of gas velocity due to adsorption. Raghavan et al. (1985) further improved the model by taking the pressure dependency of the mass transfer rate into consideration. Cen et al. (1985) and Yang et al. (1985) have developed models that considered surface diffusion or pore diffusion as the mass-transfer controlling step. Most of the above-mentioned studies compared the respective model prediction with their experimental results. However, except for the work of Yang et al. (1985), the comparison was always made based on the overall performance of the process. An overall index such as an average product concentration or a steady-state production
.
rate might be adequate for design purposes, but it could very unlikely be used to differentiate the detail mechanism of a dpamical model. Therefore, if one wanted to extend the range of applicability of a particular model, transient data must be obtained. The prediction of a model could then be examined in detail. For example, the studies by Kotsis and Argyelan (1981, 1982, 1983) and Argyelan and Kotsis (1982) found that for short cycles, the process might be limited by the macropore diffusion rate in the zeolite pellets. The adsoprtion rate of oxygen was found to be twice that of nitrogen for the 5A zeolite pellets they used. The limited experimentalresults of Fernandez and Kenney (1983) indicated that when the half cycle time was longer than 40 s, a isothermal local equilibrium model could give a satisfactory prediction. It should be important to define the range of operation conditions under which a simple model is adequate. It would also be useful to find the particular assumption that should be removed to extend the model ability. In this study, an apparatus was designed in which the flow rates and the concentrations of all streams connected to the adsorption bed were continuously monitored. With this apparatus, the transient behavior of a PSA oxygen enrichment process could be followed closely. These transient results were then compared with the prediction of a simple model and used to demonstrate the correctness or the inadequacy of some assumptions made to derive the model. One important application of oxygen-enriched air is in the field of combustion. Boiler feed with such air can reach a higher flame temperature and a more complete burning of the fuel. Oxygen-enriched air has also been used in the steel industry to increase the efficiency of blast furnaces. However, a feed air too high in oxygen may overincrease the furnace temperature and damage the firebrick. Therefore, only a small percentage of oxygen enrichment should be used. It was anticipated that the operation conditions of a PSA process to produce such a low en-
0888-5885/88/2627-0081$01.50/0 0 1988 American Chemical Society
82 Ind. Eng. Chem. Res., Vol. 27, No. 1, 1988
pd
RV
PR
-- vent
sv c v
RV
pressurization, production, blowdown, and purge in repeating cycles. The direction of blowdown and purge steps were countercurrent to that of the pressurization and production steps. Feed was a t 500 kPa and predried. Purge was done by the same compressed air regulated to about 120 kPa. The duration of each step was preset by an appropriate control program on the programmable logic controller. Prior to the cyclic operations, tests were done to estimate the time needed to fill and to vent the bed under different needle valve settings. Therefore, one can change the duration of each step and still obtain the same pressure ratio. To use the feed air instead of product as purge gas was not a common practice. However, since we were interested in producing only low-purity oxygen, this arrangement could give higher productivity. In addition, the experimental results would be easier to analysis if the purge gas concentration was kept constant.
Mathematical Model The material and energy balances of a binary gas in an adsorption column could be written as
Figure 1. Schematic diagram of the experimental system.
richment of oxygen should be different from that of a common process producing more than 90% purity oxygen. For this reason, we focused in this study on cases that the product gas contains less than 40% oxygen. A single-bed process was chosen in this study. T o further simplify the system, low-pressure air was used to purge the adsorption bed instead of the product gas, as is usually done in air separation processes. Even in such a simple operation, there were many process variables involved (pressure ratio, purge ratio, step time, product quantity, etc.). In this study, the pressure ratio and the purge quantity was fixed. Only the effects of the step duration as well as the production rate were examined.
Experimental Studies The 5A zeolite pellets of 0.159-cm size were purchased from Union Carbide Corp., New York. The compressed air used was the dry instrumental air supply at China Steel Corp. The adsorption column was 10.16 cm in diameter and 150 cm long, with flow distributors on both ends. The flow distributor was a solid polypropylene cylinder with many small holes. The total pre- and postbed dead volumes were less than one-hundredth of the packed volume. All pipings were 0.935-cm i.d. in size and were made as short as possible. A schematic diagram of the system is shown in Figure 1. The top and bottom of the column were connected to an oxygen analyzer (Sensor Medics Co., Model OM-11, California), the response time of which was less than 0.1 s. There were four mass flow sensors (Linde Co., Model FM-4575, New York) used to monitor the flow rate of feed, blowdown, purge, and product stream, respectively. All the measurements of flow rate and oxygen concentration were digitized by an A/D converter and send to a personal computer. The A/D converter was a HP3478 digital multimeter which had an automatic range switching capability and a resolution of 10 mV compared to the 5-V output range of the sensors. The solenoid valves in the apparatus were controlled by a programmable logic controller. The 5A zeolite was activated at 350 "C for about 24 h before they were packed into the adsorption bed. The packed bed was purged with dried cylinder air for a sufficiently long period of time before each experiment. The process was operated a t ambient temperature of roughly 25-30 "C. The operation was a four-step process-
where rA =
kA(WA* - WA)
rB = kB(WB* -
WB)
(4) (5)
In here, rA and rB were the local adsorption rates of oxygen and nitrogen, respectively. To write these equations, it was assumed that the pressure is uniform in the column, the gas is ideal, and the mass-transfer rates can be represented by a linear driving force model as suggested by Chihara and Suzuki (1983). Air was also assumed to be a mixture of only oxygen and nitrogen. The equilibrium sorption amounts WA*and WB* were functions of the pressure and the gas-phase composition yA and could be calculated from one of the adsorption isotherms given by Verelst and Baron (1985), Sorial et al. (1983),or Miller et al. (1987). A comparison of these works indicated that the results of all these studies were close to within lo%, even though the zeolites they used were from different manufacturers. However, the study of Sorial et al. was done at the highest pressure, and therefore their parameters for the statistical thermodynamic model (Ruthven, 1976) were employed in this study. Thus, WA = [ K A ~+AC C ( K A P A ) ' ( K B P B-) ~~ (P~A / vi
J
- l)!j!]/[l + KAPA + KBPB + CC(KAPA)'(KBPB)'(l - iP,/v - j @ B / v ) ' + l / i ! j ! ]
j(lg/V)'+J/(i 1 1
where KA = 0.031 exp (1741.48/T), KB = 0.005 exp(1645.75/T) mol/(g Pa), @ A / v = 1.282 exp(-959.06/T), and &/v = 0.202 exp(-175.63/T). The lumped mass-transfer coefficients k A and kBwere taken to be 1 and 0.5 s-l for oxygen and nitrogen, respectively, as suggested by Kotsis and Argyelan (1983) for
Ind. Eng. Chem. Res., Vol. 27, No. 1, 1988 83 Table I. System Characteristics and Operation Conditions" void fraction 0.367 packing density, g/cm3 0.764 300 temp, K low pressure, kPa 170 500 high pressure, kPa TI, 45 T2, s 30 7'3, 40 T4, 65 purge flow rate, NBV/min 1.85
k(O,),
1.0
k(Nz), s - ~
0.5
&
-z
05 OL
2 03
i
L
02
C
W
01
" T I ,T2,T3,and T 4and purge flow rate were as given here unless specified otherwise.
00
0
the 0.159-cm 5A pellets and corrected by the relation given by Raghavan et al. (1986). This mass-transfer coefficient is a lumped factor of the film-transfer coefficient as well as the pore diffusivities. Data on the adsorption heat were more difficult to find. They might also depend on the loading of the zeolite (Barrer, 1978). In addition, the equations would be more difficult to solve when the temperature effect is considered. Therefore, the model was simplified by neglecting the thermal effect. The results of this isothermal model might not be correct in certain cases, but the effect of this assumption on the transient result will be examined. Putting the gas quantities in units of equivalent STP bed volume, and assuming the system was isothermal, the above equations became
1
2
3
4
5
6
Productlor rate (NBV/Cycle )
Figure 2. Effects of pressurizationrate on enrichment in each cycle. Experimental points at T I = 45, 30, and 20 s. Solid lines are the simulation results of T I = 45 and 20 s. ~
-
-
_
_
0 6 /
-&
05 Ob
Z
E03 E
r
5
02
C
W
01
00 0
where XAis the amount of oxygen adsorbed per unit bed volume. Corresponding to the four steps of operation, there were four different sets of boundary conditions: (i) pressurization, at n = 0, y = 0.21, a t n = 1, Q = 0, d@/dt = (aH - @,,)/TI;(ii) production, a t n = 0, y = 0.21, at n = 1, Q = &,a = constant, CP = aH;(iii) blowdown, a t n = 1, Q = 0, at n = 1, ayA/an= 0, d a / d t = (aL- a H ) / T 3(iv) ; purge, at n = 1, Q = Q p g = constant, at n = 1, y = 0.21, = aL. Here, the product and purge flow rates were assumed to be constant. Although it is unlikely that these flow rates were constant in reality, an average value might be used as an approximation. These equations were then solved by the cell model approximation (Ikeda, 19791, which resulted into a set of ordinary differential equations. Starting with column equilibrated with air, these equations were integrated following the pressurization, production, blowdown, and purge sequences until a cyclic steady state was established. These equations were however very difficult to integrate by using convension procedures such as the Runge-Kutta method. The final integrating procedure used was the STIFF-3 program described by Villadsen and Michelsen (1978). It was known that the choice of cell number corresponds to the states of mixing in the bed. It is equivalent to the effect of the axial dispersion coefficient in a second-order model of packed beds (Ikeda, 1979). However, since the integration was slow even with the STIFF-3 routine, and it took about six cycles to reach a steady state, we had to scale down the problem by using less mixing cells. A total of eight cells was actually used in the simulation. It was a relatively small number for a long column and would tend to smooth out the calculated transient concentration.
1
2
3
4
5
6
Productlon Rate (NBV/cycle)
Figure 3. Effects of purge step duration on enrichment. Experimental points at T4 = 65, 40, and 20 s. Solid lines are the simulation results.
Results and Discussion The characteristics of the column and the operation conditions as well as the time sequences of each step are listed in Table 1. These were also the parameters used in the simulation. It is worth mentioning that all parameters used in the simulation were either experimentally measured or from literature data. In order to compare the performances of different operating conditions and to analyze the product quality on a common basis, an overall quantity was defined as [enrichment] = [O,in product] - [N, in product/0.79]0.21 (8) This quantity was particularly useful if the oxygen enrichment produced was somewhat higher than required and could be diluted with normal air. In other words, the oxygen-enriched air was considered a mixture of pure oxygen (the enrichment) and the flow through air (the second bracket in right-hand side of 8). As long as the enrichment remained the same, operations that produced pure oxygen or oxygen-enriched air were of the same effect as far as the final usage is concerned. For example, any process that produced 1 STP L enrichment could always be diluted to gie 79 STP L of 22% oxygen air, as long as the original product purity was higher than 22 % . This quantity was calculated from experimental data and plotted against the production rate in Figures 2 and 3. In general, the enrichment increased as the production rate was increased. In addition, when the duration of the
84 Ind. Eng. Chem. Res., Vol. 27, No. 1, 1988
0.4 N
t
I
I
0.2
0
I
0.2 I
I
TIME
Figure 5. Exit concentration history at the bottom of the column during the blowdown and purge steps. The conditions were the same as the corresponding curves in Figure 4. TIME
Figure 4. Exit concentration history at the top of the column during production step. Dotted lines are the experimental measurements. Solid lines are from simulation results. From top down the production rate was 0.61, 4.75, 0.43, and 3.84 NBV/cycle. The purge step duration was 65 s for the first two curves and 20 s for the last two curves.
compression step changed from 45 to 20 s, there was a small decrease in enrichment. If the local equilibrium model is, correct, the concentration profile after pressurization should be independent of the filling rate (Fernandez and Kenney, 1983). The observed decrease indicated that a finite adsorption rate should be considered. However, since a shorter cycle time means a higher throughput, the 20-s compression step was in fact more favorable from the production point of view. In Figure 3, three sets of experimental data were presented. They were operated at different purge step durations with the same purge flow pate. The enrichment decreased with decreasing purge step duration. A 20-s purge was obviously not enough to restore the full capacity of the column. The solid lines in Figures 2 and 3 were the simulation results of the model. Taking into account that the adsorption isotherm and rate constants were taken directly from literature data, the agreement between experiment and theory could be considered as reasonable. Moreover, the decrease of enrichment at a faster compression step as well as a shorter purge step was also correctly predicted by the model. However, it would be difficult to judge from these figures as to which assumption in the model had lead to the deviations. In the production stage, the exit concentration a t the top of the column was monitored. Shown in Figure 4 as the dotted lines are the transient product concentrations from the first to the fourth cycles. The solid lines were the corresponding simulation results. One immediately notices that, when the production rate was high, it took hbout four cycles to reach a steady concentration wave form. At the same time, the oxygen peaks were smaller and steeper than the other cases. On the other hand, more cycles were needed for the low production rate cases to reach a steady wave form. Since the first cycle was started with the same initial conditions of the column, it could be compared independently of the purge step duration. Thus, one observes that the point of maximum oxygen concentration moves from the end to the beginning of the step as the production rate increases. The final steady wave forms of the concentration were quit different for different operations. Only in the case of a long purge and a low production rate did the oxygen concentration remain high
during the entire production step. For all other cases, the oxygen purity quickly dropped to a lower value. In the bottom set of concentration waves, two peaks were observed. It seems to suggest that two different adsorption zones were developed in the bed after the purge step. More on this will be discussed late. The simulation results were in better agreement with the experiment data in the case of longer purge and higher production rate. It should be pointed out that the product as well as the purge flow rate used in the simulation was the average of the respective measurements. The product and purge flow rates were in fact varied with time. If the flow rates were averaged, some inaccuracy would be obviously introduced into the calculation. As mentioned earlier, the calculation was done with only eight cells to approximate the column. Thus, a smoother than real transient predicted by the model was expected. The concentration history of the exit stream during the depressurization and purge steps is given as dotted lines in Figure 5. In all the experimental curves, there is a short, oxygen-enriched period that exists in front of the waves. This might be in part because the oxygen analyzer measures the product gas before switching to the vent stream. This enriched period however was higher and longer for low production rate cases. It indicated that the oxgenenriched gas in the void volume was first blown out of the column before any desorption of nitrogen from the zeolite had occurred. It was also observed that only for the case of long purge duration and small production rate could the oxygen content of the blowdown gas rise back to 21 % . This indicated that the nitrogen adsorbed previously has been completely desorbed after 65 s of purge. Because of better desorption, the product oxygen concentration was maintained at a high level as seen in Figure 4. On the other hand, the downward trend of oxygen purity was terminated abruptly at the end of the purge step in the last case in the figure. This suggests that near the venting end of the column, the gas phase would still be nitrogen enriched after the purge step. This nitrogenenriched portion in the bed would then split the oxygen front developed in the next pressurization step and give the double transient peaks observed in Figure 4. However, this phenomenon was not adquately described by the model, since we were using too few mixing cells to simulate the process. The simulation results in this figure seemed to drop earlier and deeper than the experimental results. Even with the consideration of flow rate averaging, this observation still suggested an overestimation of the desorption rate in the model. It has been pointed out by Raghavan
Ind. Eng. Chem. Res., Vol. 27, No. 1, 1988 85 et al. (1986) that the rate of adsorption and desorption may be different in the high- and low-pressure cycles. Part of the reason might be the pressure dependence of the diffusion coefficient, but the local thermal effect definitely played a major role in changing the effective adsorption rate. The endothermic desorption process as well as the adiabatic expansion of the gases in the bed could lead to a lower temperature in the desorption step and thus decrease the desorption rate. In our simulation, the same rate constant was used for both the desorption and the adsorption steps. During the adsorption cycle, simulation results were closer to the experimental observation than the desorption cycle. It could be concluded that different rates were indeed experienced in the two half-cycles. Since the system was better described by the model in the pressurization step than the blowdown and purge steps, the overall process performance was more in accordance with the model when the purge time was long and the desorption was more complete. Efforts were made to improve this model by changing the pressure history in the model during the blowdown step, without success. Therefore, a nonisothermal model with a pressure- and temperature-dependent rate constant should be able to improve the model prediction.
Conclusions An automatic experimental PSA system was set up to study the enrichment of oxygen from air by 5A zeolite. For the limited operation conditions studied, the maximum oxygen purity produced was about 40%. The process reaches a steady-state condition in about 4-10 cycles. The overall results indicate that about 0.5 normal bed volume of enrichment could be produced in each cycle under the present operation conditions. The emphasis of this study was to measure the transient behavior of the effluent gas streams so that the comparison between model prediction and experimental results could give more insight to the detail process mechanism. This has not been done in most of the previous studies. A simple isothermal model was used, with literature data on equilibrium and rates, to simulate the experimental results. The prediction of overall enrichment by the model was close but not exactly the same as the observation. However, on the basis of these comparisons, one could not differentiate which part of the model was in error. The transient results were more informative than the overall enrichment results. It indicated that the model, as well as the parameters used, was quite reasonable in cases where the purge step was long and production rate was small. However, large deviations between the model and the experiments were found under other conditions. A slower than predicted desorption rate was evidently observed from the experimental data. The thermal effects was believed to be the main reason for a slow desorption rate. For a better prediction under such conditions, a nonisothermalmodel with temperature-dependent rate was thus suggested as more appropriate. Acknowledgment We acknowledge the financial support given by the China Steel Corp. to complete this study.
Nomenclature A = cross-sectional area of column, cm2 C, = average heat capacity of the mixed gas, J / ( K mol)
C,i = heat capacity of the component gases or solid D = column diameter, cm h = heat-transfer coefficient of column external surface 12 = effective mass-transfer coefficient, min-' L = column length, cm n = z/L P = pressure, kg/cm2 q = molar flow rate, mol/min 8 = qRT/ (ALP& r = sorption rate, mol/(g s) R = gas constant t = time, min T = temperature, K T I = duration of pressurization step, min T , = duration of production step, min T, = duration of blowdown step, min T4 = duration of purge step, min W = amount of gas adsorbed, mol/g of solid X = =WRT/P, yA = mole fraction of oxygen z = distance measured from column inlet, cm Greek Symbols = porosity of the packed column p = bulk density of the adsorbent = PIP, t
Subscripts and Superscripts A = oxygen B = nitrogen H = high operation pressure L = low operation pressure pd = production step pg = purge step s = solid phase 0 = ambient condition * = equilibrium state Registry No. 02,7782-44-7.
Literature Cited Argyelan, J.; Kotsis, L. Hung. J . Znd. Chem. 1982, 10, 155. Barrer, R. M. Zeolites and clay Minerals as sorbents and molecular sieues; Academic: London, 1978; Chapter 4. Cen, P. L.; Chen, W. N.; Yang, R. T. Ind. Eng. Chem. Process Des. Deu. 1985, 24, 1201. Cheng, H. C.; Hill, F. B. AZChE J. 1985, 31,95. Chihara, K.; Suzuki, M. J. Chem. Eng. Jpn. 1983, 16, 53. Fernandez, G. E.; Kenney, C. N. Chem. Eng. Sci. 1983, 38, 827. Ikeda, K. Chem. Eng. Sci. 1979,34,941. Kotsis, L.; Argyelan, J. Hung. J. Znd. Chem. 1981, 9, 73. Kotsis, L.; Argyelan, J. Hung. J. Znd. Chem. 1982, 10, 143. Kotsis, L.; Argyelan, J. Hung. J. Ind. Chem. 1983, 21, 417. Miller, G. W.; Knaebel, K. S.; Ikels, K. G. AIChE J . 1987, 32, 194. Raghavan, N. S.; Hassan, M. M.; Ruthven, D. M. AZChE J . 1985,31, 385. Raghavan, N. S.; Hassan, M. M.; Ruthven, D. M. Chem. Eng. Sci. 1986,41, 2787. Ruthven, D. M. AIChE J . 1976,22, 753. Shendalman, L. H.; Mitchell, J. E. Chem. Eng. Sci. 1972, 27, 1449. Skarstrom, C. W. In Recent Developement in Separation Science; Li, N. N., Ed.; CRC: New York, 1972; Vol. 2, p 95. Sorial, G. A.; Granville, W. H.; Daly, W. 0. Chem. Eng. Sci. 1983, 38, 1517. Verelst, H.; Baron, G. V. J . Chem. Eng. Data 1985, 30, 66. Villadsen, J.; Michelsen, M. I. Solution of differential equation models by polynomial approximation; Prentice-Hall: Englewood Cliffs, NJ, 1978; Chapter 8. Weaver, K.; Hamrin, C . E. Chem. Eng. Sci. 1974,29, 1873. Yang, R. T.; Doong, S. J.; Cen, P. L. AZChE Symp. Ser. 1985,242, 84. Received for review June 9, 1986 Revised manuscript received August 25, 1987 Accepted September 25, 1987