Heuristic design of pressure swing adsorption: a preliminary study

S. Jain a, A.S. Moharir a, P. Li b,*, G. Wozny b

b

Department of Chemical Engineering, Indian Institute of Technology, Bombay 400 076, India Institut fu r Prozess und Anlagentechnik, Technische Universita t Berlin, KWT9, 10623 Berlin, Germany

Abstract Due to its complicated nature and multiple decision parameters including plant dimensionality and operation condition, the design of pressure swing adsorption (PSA) processes is not a trivial task. Most previous studies on PSA design have been made through rigorous modeling and experimental investigation for specific separation tasks. General heuristics for a preliminary design of PSA processes are necessary but not well investigated so far. In this paper, we attempt to develop easy-to-use rules for PSA process design, based on analysis of the inherent properties of adsorbate / adsorbent systems (i.e. equilibrium isotherm, adsorption kinetics, shape of breakthrough curves, etc.) and simulation results. These rules include the selection of adsorbent, particle size, bed size, bed configuration, purge volume, pressure equalization and vacuum swing adsorption. Results of two case studies are presented to verify the rules proposed in this preliminary study.

Keywords: Pressure swing adsorption; Simulation; Design and operation; Heuristics

1. Introduction Because pressure swing adsorption (PSA) has the properties of high selectivity, high throughput and high energy efficiency, more and more PSA processes are designed and operated to carry out gas bulk separation and purification tasks in the chemical industry. In a PSA process, the adsorbent adsorbs the preferential species of a gas mixture, which is then desorbed by reduction in pressure.

Since the development of PSA, many improvements in the process have been done to make it more efficient. The most previous studies on PSA design have been made through rigorous modeling and experimental investigation for specific purification tasks. Since a PSA process is quite complicated and there are many parameters to be decided, there have been no general easy-to-use design rules so far. Because a pilot plant study of PSA is costly compared with computational study, simulation has become a viable alternative to pilot plant experiments. Based on basic kinetics and equilibrium data as well as operation parameters, simulation provides a method of predicting outlet

26

Nomenclature b C D dp G K k L N P Q q q* qs R T t v W x y z Greek symbols a o f g h r m t Subscripts 0 1, 2 act ads blow f g H I L min pres prod purg s

Langmuir constant (atm 1) concentration in gas phase (mole per m3 of fluid) intracrystalline diffusivity (cm2 s 1) diameter of particle (cm) purge/feed volume ratio Henrys Law constant overall mass transfer coefficient (s1) total height of the column (m) total number of components pressure (atm) molar flow rate adsorbed phase concentration (mole per m3 of solids) equilibrium concentration of adsorbed phase (mole per m3 of solids) Langmuir constant (mol cm 3) universal gas constant temperature (K) time (s) superficial velocity (m s1) power (kW) adsorbed phase composition gas phase composition height of the bed ( /0 at feed end, /L at product end) (m) separation factor (dimension less) bed Porosity factor defined in (Eq. (19)) ratio of specific heats in gas phase mechanical efficiency density (kg m 3) gas viscosity (Cp) residence time (s) initial components actual adsorption blowdown step feed gas phase high pressure step component low pressure step minimum value pressurization step product purge step solid phase

27

concentrations and the dynamic capacity of PSA, without recourse to experimentation. To this end, a comprehensive model and a numerical solution method are required [1,2]. Cen and Yang [3] showed that due to its simplicity, the equilibrium model can be widely used for PSA simulation. Alpay and Scott [4] used the linear driving force (LDF) model to describe the adsorption and desorption kinetics in spherical particles. Raghavan and Ruthven [5] presented numerical simulation of PSA in the recovery of trace adsorbable from an inert carrier using the linear equilibrium and linear rate expressions. They also assumed that during pressure changes (blowdown and pressurization) there was no mass transfer between the fluid and adsorbent. Raghavan and Ruthven [6] discussed numerical simulation for a simple two bed PSA process in which effects of kinetics and changes in flow rate due to adsorption are significant. They showed that when the above effects are significant, adsorption equilibrium and constant velocity assumptions are no longer valid. Chahbani and Tondeur [7] discussed the mass transfer kinetics in PSA. They showed that the choice of the pore diffusion model plays a key role for obtaining reliable simulations. They compared their results with other models like LDF and equilibrium model. Kvamsdal and Hertzberg [8] showed the effect of mass transfer during the blowdown step. According to their results, the frozen solid assumption is valid only in certain cases, and taking into account the mass transfer during the blowdown step gives a better overall model performance in the studied cases. A series modeling and simulation studies have been done aiming at finding proper PSA design and operation such as the particle size and pressure ratio [9 / 12]. These previous works mainly focused on modeling and simulation of PSA. From these studies, it is found that for design and operation of PSA, one has to go for a rigorous simulation. Modeling PSA leads to a nonlinear dynamic partial differential equation system, which is difficult to solve. The parameters in PSA models affecting the behavior of the process are highly coupled to each other. Studies on the sensitivities of these parameters

through rigorous modeling and simulation are quite expensive. Therefore, it is necessary to develop heuristic rules with which design and operation of PSA processes can be made without doing more rigorous simulation. Heuristic rules (or rules of thumb) simplify the design and/or operating options and shortlist a few options which could be probed further. They thus reduce the dimensionality of an otherwise combinatorial design problem. Heuristics has found wide applications in process design (usually a preliminary design). Potential designs from heuristic rules can then be fine tuned using rigorous simulation, if necessary. The number of decision parameters, including plant dimensionality and operation condition, is quite large in the design of PSA. More and more adsorbents are being developed, but adsorbent / adsorbate system characteristics themselves are not fully understood. The accumulated experiences with adsorptive separations are also insignificant compared with those in the cases of thermal distillation and chemical reaction. This work attempts to develop a set of heuristic rules for the design of PSA systems. A model of adsorptive separation is used to generate case studies. Based on the properties of adsorbate / adsorbent systems (i.e. equilibrium isotherm, adsorption kinetics, shape of break through curves etc.), systematic knowledge of the input/output relations thus gained is extracted out to evolve some simple rules for design. Product purity and recovery are the performance factors studied to derive the rules for various decision steps in design of PSA. They include decisions on size of particles, pressure levels, configuration of PSA cycle, residence time in PSA bed, design of PSA bed, choice for pressure equalization step and choice for VSA process.

2. Modeling PSA processes Fig. 1 shows a typical PSA process. The process consists of two fixed-bed adsorbers undergoing a cyclic operation of four steps: (1) adsorption, (2) blowdown, (3) purge, and (4) pressurization. By employing a sufficiently large number of beds and

28

@ Ci @ vCi (1  o ) @ qi    0; o @t @z @t

i  1 and 2

(1)

The last term in Eq. (1) is the mass transfer term between solid and gas, where qi is the concentration of component i in the solid. For pressure changing steps (pressurization and depressurization) this term is zero due to assumption (2). Summation of Eq. (1) for all the components leads to the total mass balance equation. For constant pressure steps (adsorption and purge) the total concentration of the fluid remains constant in the bed, thus the total mass balance will be C
N @ v ( 1  o ) X @ qi  0 @z o i1 @ t

(2)

Fig. 1. Basic two-bed PSA process.

while for pressure changing steps the total mass balance is @v @z  1 @P P @t 0 (3)

using more complicated procedures in changing bed pressure, PSA may be carried out as a continuous process. Additional steps such as cocurrent depressurization and pressure equalization have been added to improve the purity and recovery of products as well as to make the process more energy-efficient. A common feature of all PSA processes is that they are dynamic, i.e. they have no steady state. After a sufficiently large number of cycles, each bed in the process reaches a cyclic steady state (CSS), in which the conditions in the bed at the end of a cycle are approximately the same as those at the beginning of the next cycle. In the present work, the following assumptions are made to model a PSA process: (1) the system is isothermal with negligible pressure drop through the adsorbent beds; (2) the pressure change in the steps of pressurization and blowdown is so rapid that no significant exchange between adsorbed phase and gas phase occurs. This is also called the frozen solid assumption; (3) Langmuir isotherm is valid for the system; (4) the mass transfer rate is represented by a linear driving force expression; (5) the ideal gas law is applicable and (6) plug flow is assumed, i.e. there is no axial or radial dispersion. The component balance for species i in the bed is [1,2]:

where it is assumed that the pressure drop in the bed is negligible. Eqs. (1) /(3) are applied for the flow from z / 0 to L . If the flow is reverse, the term @ /@ z will be negative. The mass transfer kinetics is modeled using the LDF approximation, based on the simplification of Ficks second law of diffusion @ qi  ki (qi qi ) @t (4)

where q i denotes the equilibrium concentration of component i. It is calculated using either extended Langmuir isotherm, q i  qSi bi pi j n X
j 1

(5a) bj pj

1

or using Henrys law, q Ki Ci i (5b)

The following boundary conditions are considered for Eqs. (1) /(3). For the adsorption and pressurization steps, the concentration of fluid at the inlet is assumed to be equal to the feed condition, since axial and radial dispersion is

29

neglected, namely Ci jz0  Cf i for the purge step it is Ci zL  PL (Ci zL )ads PH (7) (6)

Ci (z; 0)  C0 ;

qi ( z ; 0)  q i

(13)

and for the depressurization step @ Ci @z jz0  0 (8)

The above set of equations is discretized by using finite difference and then solved by the Newton /Raphson algorithm. The performance of a PSA process is measured on the basis of product purity and product recovery. Product purity of the desired component 2 is defined as its average composition in the adsorption step
tads

The velocity boundary condition for the pressurization and adsorption step is vz0  vf (9)

g
y 2;prod 
0

y2;prod (t)dt tads (14)

where vf is the superficial velocity of feed at z /0.

while product recovery is defined as

 amount of component 2 obtained during adsorption step amount of component 2 used in purge step  Reco   amount of component 2 used during adsorption step in feed amount of component 2 fed during pressurization step  y 2;prod (PH vf tads  PL G vf tpurg ) y2;f (PH vf tads  (PH  PL =2)vf tpres ) (15)

(Velocity need not be constant with time during pressurization step unless specifically controlled. Normally, controllers are not used in PSA. Above could be seen as a simplifying assumption.) For the purge step the velocity boundary condition is vzL  Gvf (10)

Eq. (15) is valid only for a PSA process having re-pressurization step with feed.

3. Heuristics for PSA process When an adsorption for separating a gas mixture is determined, a logical sequence of decision steps in design of a PSA process is as follows: 1) Selection of a proper adsorbent based on its equilibrium and kinetic characteristics. 2) Selection of particle size distribution and particle shape. 3) Selection of operating pressure levels for a PSA system. 4) PSA cycle configuration and duration of each individual steps.

where G is the purge-to-feed velocity ratio. For the blowdown step the velocity boundary condition is vzL  0 (11)

In the cyclic operation, the initial condition in the bed is the condition at the end of the previous step. For startup, either a clean bed or a saturated bed can be used. For a clean bed the initial conditions are Ci (z; 0)  0; qi ( z ; 0 )  0 (12)

and for a saturated bed they are

30

5) PSA bed dimensions. 6) Inclusion/exclusion of pressure equalization step. 7) Vacuum swing adsorption (VSA) as an alternative. Based on the model described in the last section, simulation studies were carried out to analyze the performances of PSA and to develop heuristics for the decision steps indicated above. The systems considered for the simulation, as given in Appendix A, cover a wide range of adsorbate /adsorbent alternatives. The resultant heuristic rules are presented in the following. 3.1. Selection of adsorbent Adsorption is achieved due to the interaction forces between the adsorbing molecules and the adsorption surface. Different substances are adsorbed with different affinities. It is this selectivity that provides the basis for adsorption separation processes. The task of adsorbent is to provide the surface area required for selective adsorption of the preferentially adsorbed species. A high selectivity is of users interest. The separation factor (a ) can be used as a measure of selectivity. The separation factor of an absorbent is defined as [1]: a12  x1 =x2 y1 =y 2 (16)

the adsorbent particle depends on the diffusivity of adsorbing molecules. For short time intervals this dependency can be approximated by [1]: p (18) qt 8 D Thus the separation factor for kinetically controlled process is calculated by s D1 a12  (19) D2 It is always useful that the separation factor is calculated by considering both equilibrium and kinetic effect. Thus the separation factor can be defined as s K1 D1 a12  (20) K2 D2 In this study, Eq. (20) is used to calculate the separation factor for all cases considered. Table 1 shows some simulation results. Adsorbents with different selectivities are used to separate a twocomponent fluid mixture (such as say air with 20% oxygen and 80% nitrogen) with a mole-fraction of 0.2 for the highly adsorbed species (species 1). While doing simulation the other parameters are kept constant. Fig. 2 shows a graphical representation of these results. From Table 1 and Fig. 2, it can be seen that the product purity will be increased, if a higher separation factor is selected. Thus a high separation factor is a key for a quick screening of various adsorbents.
Table 1 Comparison of adsorbents based upon separation factor p K1/K2 / D1 =D2 a Product purity of component B 1 1 1 1 0.11 2.5 10 2.14 7 10 0.68 2.14 6.76 21.4 6.76 6.76 6.76 1 1 1 0.68 2.14 6.76 21.4 0.75 17.1 67.6 2.14 7 10 0.78 0.821 0.952 0.999 0.787 0.993 0.999 0.812 0.942 0.954

where x1 and y1 are, respectively, the mole fraction of component 1 in adsorbed phase and fluid phase. The separation factor depends on the adsorption property, either adsorption kinetics or adsorption equilibrium, or both. In an equilibrium controlled adsorption process, it is simply the ratio of the equilibrium constants. For an extended Langmuir isotherm and a linear isotherm, this separation factor is the ratio of Henrys constants a12  K1 K2 (17)

In a kinetically controlled adsorption, the selectivity depends on the difference of kinetic parameters. The time-dependent concentration within

The conditions for the simulation are given in the caption of Fig. 2.

31

@P @z

180mv (1  o )2
2 dp

o3

; z  [ 0; L ]

(21)

From Eq. (21), as the porosity increases the bed pressure drop will decrease. Fig. 4 shows the pressure drop versus the bed porosity. For a given system, Eq. (21) can be rewritten as 
Fig. 2. Effect of separation factor on product purity (PH /3 atm, PL /1 atm, tads /tpurg /40 s, tpres /tblow /5 s, o /0.4, vf /0.007 m s  1, purge/feed /2, L /0.65 m).

@P (1  o )2 f @z o3

(22)

where f is given as f 180mv

2 dp

(23)

3.2. Size distribution and shape of particles In adsorption processes, the particle size distribution and particle shape decide the bed porosity. Porosity affects bed performance in two ways. If other parameters remain constant, a lower porosity leads to a higher bed pressure drop. On the other hand, a lower porosity means a higher adsorbent content of the bed and hence a higher adsorption capacity. Moreover, a lower porosity means less loss of the adsorbable component during the blowdown step and less requirement of product gas for an effective extraction during the purge step. Both of these properties increase the product recovery. Bed porosity can be varied using different shapes and sizes of adsorbent particles and to some extent packing techniques. Simulation studies are made using various values of bed porosity. The results are shown in Fig. 3. The effect of bed porosity on product purity is significant for both bulk separation and purification processes. The effect on product recovery is significant only in bulk separation cases. Fig. 3a /d show that as porosity increases both the product purity and recovery will decrease. From Fig. 3e, it can be seen that as the porosity increases, product recovery remains almost constant for purification processes (where the highly adsorbing species is in very low concentration, e.g. air containing moisture). On the other hand, the porosity of a bed also affects the bed pressure drop. In a packed bed the pressure drop can be calculated from Blake / Kozeny equation

There are two opposite effects of porosity: high product purity requires low porosity and a low pressure drop is achieved with high porosity. From Fig. 4, it can be seen that for bed porosity lower than 0.3, the pressure drop increases rapidly. Thus this should be the lower limit of porosity (which can be achieved for most practical particle shapes and sizes). For bed porosity higher than 0.5, the change in pressure drop becomes moderate, thus this can be considered as the upper limit. Similarly, the effect of the porosity on the product purity, as shown in Fig. 3a /e, indicates that when bed porosity is higher than 0.5, the purity will decrease significantly, thus confirming the use of 0.5 as the upper limit of the bed porosity. 3.3. Selection of adsorption pressure Selection of the adsorption pressure is based on the equilibrium relationship of the system. An isotherm describes the equilibrium loading of a species, which is dependent on the partial pressure of the species in an adsorption process. As the adsorption pressure increases, the amount of fluid adsorbed on the adsorbent will increase. To determine the pressure level for adsorption, one should keep in mind that the larger the difference between the capacities of the competing adsorbates, the purer the raffinate will be. The selectivity of an extended Langmuir isotherm and a linear isotherm is constant [1,2]. These isotherms are, therefore, called constant selectivity isotherm. For constant selectivity systems, if the pressure is

32

Fig. 3. Effect of porosity on bed performance (a): Bed porosity vs. product purity. (b): Bed porosity vs. product recovery. (c): Product recovery for system 3. (d): Product purity for systems 4 and 5 (purication process). (e): Product recovery for systems 4 and 5 (purication process).

increased, which causes more adsorption of the highly adsorbed species, the product purity will be increased. The less favorable species is also ad-

Fig. 4. Effect of bed porosity on bed pressure drop.

sorbed more, but in comparison to the more favorable species it is always less, as long as the more adsorbing species is present in significant quantities in the bulk phase. Therefore, for such systems a higher pressure always leads to a purer product. Besides this advantage of high pressure for constant selectivity systems, there is one disadvantage too, that is energy loss. A higher pressure level leads to higher compression costs and a higher loss of energy in the blowdown step. If an adiabatic compression is assumed, the power requirement can be approximated as [1]:

33

where PH and PL are high and low pressure levels, respectively, Q is the volumetric flow rate, and T is the operating temperature. Simulation results of extended Langmuir and linear isotherms are given in Fig. 5a /c for various adsorption pressure levels. It can be seen that as the pressure increases, the product purity will be increased. At the same time the power required for compression also rises. If a high purity product is desired, the adsorption pressure should be as high as possible. Fig. 6 shows that the power requirement increases with the increase of the adsorption and the desorption pressure ratio. Considering the trade-off between product quality and power requirement, the adsorption pressure should be taken as the value at which the change of the adsorbed phase concentration with pressure becomes moderate. If the selectivity varies with the operating pressure, the adsorption pressure should be that at which the selectivity is maximum. To study the effect of pressure for varying selectivity systems, two hypothetical systems (with imputed isotherm constants) are taken and simulated. The simulation results are shown in Fig. 7a and b. It is

Fig. 5. Effect of adsorption pressure on product purity with yf /(0.99, 0.01). (a) System 1 and 2. (b): System 3. (c): Systems 4 and 5 (purication process).

 W

g g1

QRT h



PH PL

(g1)=g

1

(24)

Fig. 6. Effect of pressure ratio on power requirement.

Fig. 7. Effect of pressure on bed performance for varying selectivity. System A: Having same properties as system 2 of Fig. 6 except equilibrium selectivity. System B: Having same properties as system 2 of Fig. 6 except equilibrium selectivityand kinetic selectivity. The kinetic data are: k1 /0.002 s  1, k2 /0.00018 s  1. (a): Selectivity. (b): Product purity.

34

illustrated that the product purity is maximum at the pressure where the selectivity is also maximum.

3.4. Selection of adsorption time In a PSA process, the duration of the adsorption step is determined by studying the breakthrough curve. The term breakthrough curve refers to the response of the initially clean bed to an influent with a constant composition. It can be seen by monitoring the concentration of the effluent. Breakthrough occurs when the effluent concentration reaches a specific value. The adsorbate concentration in the flow at any given point in a bed is a function of time, resulting from the movement of concentration front in the bed. The breakthrough curve for a gas containing a single adsorbate can be obtained by the solution of the mass balance equations for both the bed and adsorbent particles, along with the equilibrium isotherm. The duration of the adsorption step is the time period needed for breakthrough to occur. After this time the product purity will decline, and before this time the full bed capacity will not be employed. Thus the adsorption time should be near the breakthrough time. This time depends upon isotherm, diffusivity and residence time of the feed in the bed. From simulation studies, it can be observed that beyond a certain value of the adsorption time, the change in product purity becomes insignificant. The product purity decreases as the adsorption of less adsorbable species increases. Fig. 8 shows how the change in the adsorption time affects the product purity. It can be seen that the time of

Fig. 9. Effect of distribution of total adsorption time on bed performance. (a): Product purity. (b): Product recovery.

Fig. 8. Effect of adsorption time on product purity. (System 2: PH /4 atm, PL /1 atm, purge/feed /2).

Fig. 10. Effect of distribution of total adsorption time on bed performance for purication process (a): Product purity. (b): Product recovery.

35

the adsorption step should be where the product purity is maximum. This is the time when breakthrough occurs. 3.5. Effect of distribution of adsorption time In a two-bed, four-step PSA system, adsorption takes place in both the pressurization and adsorption steps. The apportion of the total adsorption time to these two steps affects the performance of PSA. Figs. 9 and 10 show the simulation results corresponding to different time allocations to these two steps. In Fig. 10, where the results shown are those of purification processes, it can be seen that as the ratio of the pressurization time and the adsorption time increases, the product recovery will decrease, but its purity will increase. With a higher time ratio, less product will be obtained. At the same time, higher time ratio improves product quality. Moreover, it can be seen from the simulation results that the rate of the decrease of product recovery is much greater than the rate of the increase of product purity. Thus, the change in recovery is a dominating factor to determine the adsorption time apportion. Having these two opposite effects and taking the fact that the recovery effect is more dominant, the ratio of the pressurization time to the adsorption time should be low. The upper limit should be 0.2, according to the simulation results. 3.6. Effect of purge-to-feed ratio The purge step in PSA is a desorption step that regenerates adsorbents by desorbing the adsorbed species. In a PSA process, saturated adsorbents are regenerated by lower pressure, thus a low pressure purge step is required. Generally, from the product vessel with the raffinate at a high pressure, a fraction of the product stream is withdrawn to purge the bed and expended to a low pressure. The volume required in the purge step affects the product quality as well as its recovery. As the purge volume increases, purging becomes more effective, providing a regenerated bed with adsorbents of less loading and leading to an increased product purity. In principal, the bed should be fully regenerated with adsorbents completely un-

Fig. 11. Effect of purge/feed ratio on bed performance (a): Product purity. (b): Product recovery.

saturated, thus more purging is necessary. At the same time, since purging is done by utilizing the product, increase in the purge volume decreases the product recovery. Generally, the purge volume specification for PSA is given by the purge-to-feed volume ratio. Simulation was made with different values of the purge-to-feed volume ratio. The results are presented in Fig. 11. The effect of this ratio on product purity and recovery is shown for two cases. It can be seen that as this ratio increases, the product purity increases as well, but the recovery decreases. The rate of the increase in purity is much slower than the rate of the decrease in recovery. From these results, one may infer that the purge-to-feed ratio should be neither too low nor too high. A guideline regarding this ratio from the simulation results is that it should be between 1.0 and 2.0, if purging is done by product. These are volumetric ratios. It should be noted that, although the volume of purge used is more than

36

the volume of feed, the mass of purge is less than the mass of feed or product, since the pressure of purging is much lower than the feed pressure.

3.7. Residence time determination The residence time of species in a bed is the ratio of the bed holdup to the volumetric feed rate. Sufficient residence time should be provided, so that the desired product purity can be achieved. For species with lower diffusivities, greater residence time is required. The choice of residence time is critical in adsorption, since if the residence time is too short, there will be no significant adsorption. Increase in residence time can be made by reducing the feed rate or by increasing the bed volume. Since the feed rate is decided by the desired capacity of the unit, required residence can be achieved by changing the bed volume. However, in an existing unit, residence time can be altered only by adjusting feed rate or feed pressure or both.

Simulations were made to study the effect of residence time on bed performances. Fig. 12 shows the effect of residence time and feed composition on the product purity for system 2 (refer Appendix A). Similar profiles can be obtained for other systems too. For different systems, the shape of profiles remains the same. The main features of these plots are that, at low residence time, no high product purity can be obtained, and the relationship between feed composition and product composition is linear. When keeping the feed composition constant, product purity can be increased by increasing residence time. If the feed composition of the desired species is high, product purity will increase linearly with residence time. Fig. 13 shows the effect of the decrease in diffusivity with a factor of 0.1 for the highly adsorbing species. Comparing Fig. 13 with Fig. 12, it can be observed that for the same residence time, a decrease in diffusivity leads to a decrease in product purity. This means that for the same product purity, more residence time is needed for systems with lower diffusivity. To calculate the residence time for a given system and a given product purity, the concept of the minimum residence time is introduced in the following. 3.7.1. The minimum residence time A PSA process reaches its cyclic steady state after a certain number of cycles of operation. A steady state PSA model can be developed using the same assumptions as stated before. The governing equations may be obtained as follows. First, by combining Eqs. (1) and (4) and letting the timedifferential terms be zero, the following equation can be gained

Fig. 12. Effect of residence time and feed composition on product purity. System 2: PH /3 atm, PL /1 atm, tads / tpurg /40 s, tpres /tblow /5 s, L /0.65 m. (a): Effect of residence time. (b): Effect of feed composition.

Fig. 13. Effect of residence time and feed composition on product purity for system 2 by decreasing diffusivity of the highly adsorbing species with a factor of 0.1.

37

o dCi dz  (1  o ) ki (qi qi ) v

(25)

If the maximum driving force is applied, i.e. by taking qi / 0 in the above equation, one has  o dCi dz  (1  o ) ki q v i (26)
Fig. 15. Effect of kinetic separation factor on the ratio of actual-to-minimum residence time for system 2. PH /4 atm, PL /1 atm, tads /tpurg /40 s, tpres /tblow /5 s, vf /0.007 m s  1, L /0.65 m.

The boundary condition for this equation is the same as given in Eq. (6). It is easy to solve this equation, and the solution for the highly adsorbed component will give the residence time for a system under the conditions of a constant velocity and the maximum driving force. Therefore, the solution of Eq. (26) gives the minimum residence time for a given feed and desired product composition. Initially, one may expect the existence of certain relationship between the ratio of the actual and the minimum residence time, with the kinetic parameters and separation factor. These relations are shown in Figs. 14 and 15. Fig. 14 gives the relationship between the separation factor and the time ratio. As the separation factor increases, the required ratio of actual-to-minimum residence time also increases. The results of actual-to-minimum residence time versus the mass transfer coefficient ratio are shown in Fig. 15. But this relation is less clear. As an explanation, one may argue that the ratio of actual-to-minimum residence time depends on the total amount of gas adsorption, rather than the kinetic selectivity (the ratio of diffusivities). Basis for this argument is that in the calculation of the minimum residence time, the maximum driving force is considered (taking qi /0), but this is not the case of the actual

Fig. 16. Effect of summation of mass transfer coefcients on the ratio of actual-to-minimum residence time. PH /4 atm, PL /1 atm, tads /tpurg /40 s, tpres /tblow /5 s, vf /0.007 m s  1, L /0.65 m, feed composition (0.21, 0.79).

residence time. The difference arises due to the difference in driving force, which is the adsorbed phase concentration in a practical cycle. The concentration in the adsorbed phase depends on the mass transfer of each component, which relies on the individual mass transfer coefficient. In this way, the total effect on the adsorbed phase concentration depends on the sum of the mass transfer coefficient, rather than the kinetic selectivity. To test this argument, in Fig. 16, the time ratio is plotted against the sum of the mass

Fig. 14. Effect of adsorption separation factor on the ratio of actual-to-minimum residence time for k1/k2 /0.0055 in system 2. PH /4 atm, PL /1 atm, tads /tpurg /40 s, tpres / tblow /5 s, vf /0.007 m s  1, L /0.65 m.

Fig. 17. Relation between (k1/k2)(K1/K2) and the ratio of actual-to-minimum residence time.

38

transfer coefficients. A more obvious relationship between these two quantities can be seen. In Fig. 17 an attempt is made to combine the relations of Figs. 14 and 16, from which the relation between (K1/K2)(k1/k2) and the ratio of actual-to-minimum residence time can be observed. It should be noted that this ratio also depends on the feed composition, but this relation is not so clear. This is because in calculation of the minimum residence time, the assumption of a constant velocity is taken, which is invalid if the concentration of the highly adsorbing species is high in the feed. If the concentration of the highly adsorbing species in the feed increases, the product flow rate will be low as compared with the feed flow rate, and the minimum residence time will be much less than the actual residence time. Thus the ratio of actual-to-minimum residence time will also depend on feed composition. This relation is illustrated in Fig. 18, in which various cases (including systems 1, 2 and 3) are tested. It is seen that all these data fit a straight line, which can be approximated with the following relation   tact K  1  100(k1  k2 ) 1 yf 1 (27) tmin K2 Most of the systems have (k1/k2)(K1/K2)yf1 B/ 0.015, where the value of the time ratio is in between 1 and 2. From this result, a quick estimation of the bed size can be made. The bed volume is determined based on the required residence time. To specify the bed size, i.e. the bed diameter and bed height, some criteria should be kept in mind. The choice of the bed diameter depends on the fluidizing velocity, which

is the minimum velocity required to fluidize a bed. The maximum velocity in the bed should not exceed 70% of the minimum fluidizing velocity [9]. For velocities greater than this value, entrainment of adsorbents in effluent stream may occur and also the pressure drop in the bed would be very high. After determining the fluidizing velocity, the bed diameter can be calculated. Another important criterion for bed specifications is the crushing strength of the solids. The height should be such that no crushing occurs in the bed.

3.8. Pressure equalization The first improvement over Skarstroms cycle is the introduction of a pressure equalization step, as shown in Fig. 19. After the first bed has been purged and the second bed has completed its highpressure adsorption step, instead of blowing down the second bed directly, the two beds may be connected to each other through their product ends in order to equalize their pressures. The first bed is thus partially pressurized with gas from the outlet region of the second bed. After the pressure equalization, the two beds are disconnected and the first bed is pressurized with feed gas while the second bed is vented to complete the blowdown. The pressure equalization step conserves energy, because the compressed gas from the high-pressure bed is used to partially pressurize the low-pressure bed. Since this gas is partially depleted of the strongly adsorbed species, the degree of separation is conserved and the blowdown losses are reduced. Based on these considerations, a pressure equal-

Fig. 18. Relation between (k1/k2)(K1/K2)yf1 and the ratio of actual-to-minimum residence time.

Fig. 19. Pressure equalization step in a PSA process.

39

Table 2 Comparison of PSA process with and without pressure equalization step for air separation on CMS Product purity 0.914 0.945 0.96 0.972 Recovery without pressure equalization 0.418 0.374 0.348 0.325 Recovery with pressure equalization 0.532 0.51 0.488 0.469

ization step is often incorporated in the PSA process. Simulation results are presented in Table 2 indicating the PSA performance with and without a pressure equalization step. It can be seen that a pressure equalization step favors product recovery. But if the pressure swing is sufficiently low, the inclusion of a pressure equalization step may be impractical. Otherwise, the pressure equalization step should always be incorporated in the PSA process.

3.8.1. Selection of the intermediate pressure The intermediate pressure is the pressure after the pressure equalization step. As the intermediate pressure increases, the degree of saturation of the highly adsorbed species increases in the bed, which is unfavorable. If it is higher than a certain value, this saturation causes decrease in product purity and recovery. Simulation results for different intermediate pressures are shown in Fig. 20. It can be seen that once the intermediate pressure is increased beyond a threshold value, both the product recovery and purity are adversely affected. This means that the intermediate pressure should be below this threshold value. If the intermediate pressure is denoted as PI, then the following relation holds approximately PI PL  (0:5 to 0:8) PH PL (28)

3.9. Vacuum swing adsorption Vacuum swing adsorption (VSA) is also a Skarstrom cycle in which the low-pressure purge step is replaced by a vacuum desorption. The product end of the bed is kept closed and the vacuum is applied through the feed end, as shown in Fig. 21. In a VSA process, using the same high operation pressure as a Skarstrom cycle and for the same product purity, the loss of the less favorably adsorbed species in the evacuation step is normally less than the corresponding loss in the purge. The gain in raffinate recovery is achieved at the expense of the additional mechanical energy required for the evacuation step. A significant

Fig. 20. Effect of intermediate pressure (after purge) on product purity and product recovery (a):System 1: Feed composition: (0.4, 0.6), PH /5 atm, vf /0.007 m/s, L /0.6 m, purge/feed /1.5. (b): System 2. Feed composition: (0.21, 0.79), PH /5 atm, vf /0.007 m/s, L /0.65 m, purge/feed /1.5.

Fig. 21. The sequence of VSA cycle (only one bed is shown).

40

amount of energy can be saved, if the adsorption takes place slightly above the atmospheric pressure and the desorption is done at a very low pressure. A VSA cycle will, therefore, be advantageous over a normal Skarstrom cycle, if a low-pressure product is acceptable. In kinetically controlled separation, a major disadvantage using a normal Skarstrom cycle is that the slowly diffusing raffinate product would be continuously adsorbed during the purge step. This problem can be avoided by using VSA. In kinetically controlled processes there is a little difference in isotherms of feed components (e.g. nitrogen separation from air using zeolite 4A) but a large difference in diffusivity. In such a system, purging with the product to remove the highly diffusing species from the bed is undesirable. This is so because apart from wasting product (a certain fraction of nitrogen), the raffinate gas will be adsorbed during this step, thereby reducing the capacity for oxygen during the next adsorption step. For such type of systems a VSA process is worth considering. Simulation is performed for the nitrogen separation from air using carbon molecular sieve (CMS). The results are shown in Table 3, which illustrate that the recovery of nitrogen is greater in a VSA process than in an ordinary Skarstrom cycle. From the results and the above explanation, it can be concluded that for kinetically controlled processes, VSA is a better choice over a normal Skarstrom cycle.

According to the proposed heuristics, recovery increases if pressure equalization is considered. Air separation using CMS was considered by Nilchan [15]. The author discussed optimization approach for PSA processes. The objective function used in the case study is minimizing the power requirement. The data are given in Table A1 in the Appendix A with system 2. Table 5 compares the results in [15] using optimization study with the results obtained using the heuristic rules proposed in this study. Both the case studies show that the results obtained using the heuristics are close to the results obtained using optimization. It demonstrates that heuristics based synthesis is useful for preliminary design and screening of PSA processes. In addition, in both case studies the flow to be separated is air, but the adsorbent is different (in case 1 Zeolite 5A, and in case 2 CMS), due to different product requirement. In the first case oxygen is the desired component, and in the second case nitrogen is the desired component. The separation factor for nitrogen in the second case is 41, while in the first case it is 0.16. If nitrogen is the desired component from air, CMS adsorbent should be used rather than Zeolite 5A. The heuristic rule proposed for this case also suggests using the adsorbent with a higher separation factor.

4. Two case studies Heuristics developed in the previous section have been tested using two case studies. Rajasree and Moharir [16] discussed simulation based synthesis, design and optimization of PSA processes. The system used was air separation using Zeolite 5A. The data are given in Table A1 in the Appendix A with system 6. Table 4 compares the results in [16] with the results obtained using our heuristic rules. In [16] the pressure equalization step was used for some cases. They showed that product (oxygen) recovery increases from 28.5 to 32.8% when pressure equalization step is included.
Table 3 Performance comparison of VSA cycle with ordinary Skarstrom cycle Product purity % Recovery of nitro- % Recovery of nifor nitrogen in % gen in Skarstrom cycle trogen in VSA cycle 89.5 92.5 94.2 95.1 98.2 56.4 53.7 49.7 42.1 21.6 88.3 83.3 77.6 72.5 60.1

The blowdown pressure used is 1 atm and the vacuum pressure used is 0.25 atm.

41

Table 4 Comparison of results by the heuristic rules and optimization (Case Study 1) Purity of oxygen 87.8% and recovery is 36.8% o PH (atm) P /F tads tdes (s) tpress/tads tmin (s) tactual (s) From Rajasree and Moharir [16] 0.376 6 6.125 65 65 0.077 104.00 270.0 Using heuristic rules proposed

0.3 /0.5 Heuristic suggests to use high pressure to obtain higher purity 1.0 /2.0 Both are equal 0.0 /0.2 104.00 316.0

5. Conclusions In PSA design and operation, heuristics developed in this study may be summarized as follows: Rule 1 : Adsorbent, which gives the largest separation factor, should be used. Rule 2 : Bed porosity should be in the range 0.3 /0.5. Particle size distribution and shape should be such that bed porosity is within these limits. Rule 3 (a ): For systems whose isotherms are given either by the Henrys Law or by the extended Langmuir isotherm expression, the adsorption pressure should be as high as possible, subjected to the power requirement constraint.

Rule 3 (b ): For systems with pressure dependent selectivity, adsorption pressure should be the pressure, which gives maximum selectivity. Rule 4 (a ): The adsorption time should be near the adsorption breakthrough time. Rule 4 (b ): For a two bed PSA process, the adsorption and desorption time should be equal. Rule 5 : The maximum limit for the ratio of the pressurization time to the adsorption time should be 0.2. Rule 6 : The ratio of purge-to-feed volume should be in the range 1.0 /2. Rule 7 (a ): The bed diameter should be such that the velocity within the bed does not exceed 70% of the minimum fluidizing velocity. Rule 7 (b ): Bed height should not cross the crushing strength of adsorbent particles.

Table 5 Comparison of results by the heuristic rules and optimization (Case Study 2) Purity of nitrogen 87.1% and recovery is 67.6% o PH (atm) P /F tpress/tads tmin (s) tactual (s) From Nilchan [15] Using heuristic rules proposed

0.4 2.36 Purging was not considered 0.086 18.2 21.2

0.3 /0.5 Power requirement decides the adsorption pressure Zero purge results in high impurity in the product. As in the case study, without purge consideration high impurity (12.89%) was achieved 0.0 /0.2 18.2 (solution of Eq. (26)) 19.3

42

Rule 7 (c ): The Ratio of actual-to-minimum residence time should be in the range 1 /2. Eq. (27) may be used to calculate the ratio. Rule 8 (a ): For processes with high swing in pressure, a pressure equalization step should be included. Rule 8 (b ): The ratio of intermediate pressure to the low pressure should be in the range 0.5 /0.8 of the ratio of the high pressure to the low pressure. Rule 9 : For a kinetically controlled process, VSA should be considered. The work for developing heuristics for PSA has been initiated in this paper. Some rules could appear obvious or trivial, considering the accumulated knowledge on PSA at this stage. More detailed models and more extensive simulation studies in future would help modify the rules proposed here and develop more elaborate heuristics. Moreover, the study of interactions between individual decision parameters is also part of the further work.

Table A2: Purification systems Number System Component 1 Component 2 K1 k1 (s1) 4 CO2 /He /Silica Gel [13] CO2 1% He 99% 9084 2.583 e-4 5 H2O /Air /Alumina [13] H2O 1% Air 99% 52.7 4.67 e-2

References
[1] D.M. Ruthven, S. Farooq, K.S. Knaebel, Pressure Swing Adsorption, VCH Publishers, New York, 1994. [2] R.T. Yang, Gas Separation by Adsorption Processes, Butterworth Publishers, Boston, 1987. [3] P.L. Cen, R.T. Yang, AIChE Symposium Series 80 (1985) 68. [4] E. Alpay, D.M. Scott, Chemical Engineering Science 47 (1992) 499. [5] N.S. Raghavan, D.M. Ruthven, The American Institute of Chemical Engineers Journal 31 (1985) 385. [6] N.S. Raghavan, D.M. Ruthven, The American Institute of Chemical Engineers Journal 31 (1985) 2017.

Appendix A: Systems considered in the simulation studies

Table A1: Bulk separation systems Number System 1 2 3 6

CH4 /N2 /CMS Air separation /CMS Hypothetical Air separation /Zeolite [14] [15] system 5A [16] Composition of compo- N2 40% O2 21% 40% N2 79% nent 1 Composition of compo- CH4 60% N2 79% 60% O2 21% nent 2 qs1 (mol cm3) 0.00182 0.00264 0.0033 0.0258 qs2 (mol cm3) 0.00255 0.00264 0.00065 0.00344 b1 (atm 1) 0.26 0.14 0.161 0.0155 b2 (atm 1) 0.62 0.154 0.164 0.057 k1 (s 1) 9.99 e-4 2.7 e-3 0.002 0.0098 k2 (s 1) 4.82 e-6 5.9 e-5 0.0018 0.0032

43 [7] M.H. Chahbani, D. Tondeur, Separation and Purication Technology 20 (2000) 185. [8] H.M. Kvamsdal, T. Hertzberg, Chemical Engineering Science 50 (1995) 1203. [9] Z.P. Lu, J.M. Loureiro, M.D. LeVan, A.E. Rodrigues, The American Institute of Chemical Engineers Journal 38 (1992) 857. [10] Z.P. Lu, J.M. Loureiro, M.D. LeVan, A.E. Rodrigues, Gas Separation and Purication 6 (1992) 15. [11] Z.P. Lu, J.M. Loureiro, M.D. LeVan, A.E. Rodrigues, Industrial Engineering and Chemical Research 32 (1993) 2740. [12] Z.P. Lu, J.M. Loureiro, A.E. Rodrigues, M.D. LeVan, Chemical Engineering Science 48 (1993) 1699. [13] W.L. McCabe, J.C. Smith, Unit Operations of Chemical Engineering, fth ed., McGraw-Hill, New York, 1993. [14] A.I. Fatehi, K.F. Loughlin, M.M. Hassan, Gas Separation and Purication 9 (1995) 199. [15] S. Nilchan, The optimization of periodic adsorption processes, Ph.D. Thesis, Department of Chem. Tech., Imperial College of Science, Technology and Medicine, London, 1997. [16] R. Rajasree, A.S. Moharir, Computers and Chemical Engineering 24 (2000) 2493.