(Water) in a Bulk Separation PSA Process: The ... - ACS Publications

2702

Ind. Eng. Chem. Res. 2001, 40, 2702-2713

The Effects of a Readily Adsorbed Trace Component (Water) in a Bulk Separation PSA Process: The Case of Oxygen VSA Simon J. Wilson, Chris C. K. Beh, Paul A. Webley,* and Richard S. Todd Department of Chemical Engineering, Monash University, Clayton, Victoria, 3168 Australia

This paper investigates the thermal profiles that arise in oxygen VSA, which is a prominent example of a PSA bulk gas separation process. Experimentally, it is demonstrated that the severe axial thermal profile or “cold spot” that frequently characterizes oxygen VSA can only arise if there are multilayered adsorption beds or if there are readily adsorbed trace components (such as water) that create a de facto multilayered bed. A qualitative explanation is offered to explain how this cold spot is formed. This paper also details a technique for predicting the penetration of a water-loaded zone into an oxygen VSA adsorption bed based on the method of characteristics. The results of this technique compare well with experimental and numerically simulated results. Finally, this paper demonstrates that a water-loaded zone and an inert zone of activated alumina result in very similar cyclic steady-state thermal profiles, even though the transient behaviors are markedly different. Introduction Pressure swing adsorption (PSA) has emerged as a significant unit operation for the separation of gases. One important example of this process technology is vacuum swing adsorption (VSA) for the production of industrial oxygen. Oxygen VSA is a mature technology and the preferred industrial process for the production of low tonnage oxygen [5-100 tonnes of oxygen per day (TPDc)] at purities between 90 and 95%. This paper investigates two related questions that have confronted oxygen VSA development. First, the oxygen VSA process produces a severe axial temperature profile or “cold spot” that can cause a significant alteration in process performance. Observation of this cold spot has frequently been reported in the adsorption literature without a detailed explanation.1-3 This paper reports an experimental investigation that demonstrates the process conditions for the formation of a cold spot and provides a qualitative explanation of the cold spot in oxygen VSA. Second, most molecular sieves used for oxygen VSA do not reversibly desorb water under VSA process conditions; hence, a prelayer of activated alumina or NaX is required to protect the main layer of adsorbent from water and carbon dioxide. The depth of water penetration sets the lower limit for the length of this prelayer. This paper presents experimental results for the penetration of water in a laboratory VSA unit and details an equilibrium approach to predicting the depth of water penetration under different process conditions. The central argument in this paper is that these two issues of the cold spot and water loading are very much related. The water loading creates a zone with dramatically reduced nitrogen adsorptive capacity, which, in turn, gives rise to the process conditions required for the formation of a cold spot or temperature depression at the entrance of the adsorption bed. Consequently, a readily adsorbed trace component can significantly alter the overall temperature profile and performance char* Author to whom correspondence should be addressed. E-mail: [email protected]. Fax: +61 3 99055686. Phone: +61 3 9905-1874.

acteristics of the VSA process, ostensibly a bulk gas separation. This paper also compares the behavior of an adsorption bed with a water-loaded zone to that of an adsorption bed with a completely inert prelayer. This substitution is often made in numerically modeled PSA systems to avoid having to introduce water as an additional component in the simulation. Heat generation in the water-loaded zone is associated with the adsorption and desorption of water, whereas there is no heat generation in a truly inert prelayer. The significance of this heat of adsorption is explored, in particular in regard to its effect on the overall thermal profile and performance of the VSA unit. Explanation of the Cold Spot Since the publication of Collins’ patent in 1977, the severe temperature profiles evident in oxygen PSA and VSA have been of much interest to industrial adsorption researchers and academics. The surprising nature and extent of the temperature profile has unexpected, and in many cases deleterious, effects on the performance of industrial oxygen VSA units. Adsorption textbooks are scattered with recognition of the cold spot.1-3 However, a clear and concise explanation of the effect is still lacking in the literature, although Collins and Kumar have made important and useful contributions.4,5 The cold spot in a multilayered adsorption bed and the overall axial cyclic steady-state (CSS) temperature profile can be understood as the interaction between convective heat transfer and heating and cooling caused by adsorption and desorption. If a clean two-layered bed with an inert prelayer undergoes adsorption, a step or discontinuity is formed between the inert layer and the adsorptive layer based on the heat of adsorption in the main layer, as shown in Figure 1. As the feed gas is convected past the discontinuity, the temperature of a boundary layer in the adsorptive layer is reduced to the temperature of the feed gas. This boundary layer of the bed contains adsorbed gas but has now been reduced in temperature to the feed gas temperature. When evacuation (depressurization) occurs countercurrently, this boundary layer

10.1021/ie000801a CCC: $20.00 © 2001 American Chemical Society Published on Web 05/15/2001

Ind. Eng. Chem. Res., Vol. 40, No. 12, 2001 2703

Figure 1. Axial temperature profile formation.

undergoes desorption of gas and is cooled. Simultaneously, the boundary layer is also heated by desorbed gas that is passed over the boundary layer from further into the bed. This sequence of steps is shown in Figure 1. The numbers alongside each line of Figure 1 indicate the temperature profile at the end of each step in a pressure swing cycle. Step 1 is feed and product withdrawal, step 2 is provide purge, step 3 is provide pressure equalization, step 4 is evacuation, step 5 is receive purge, step 6 is receive pressure equalization, and step 7 is feed repressurization. This is the same cycle as was used for the experiments undertaken and reported later in this paper. These profiles were generated using the in-house simulator (MINSA, Monash Integrated Numerical Simulator for Adsorption). MINSA is a generalized cycle simulator based on a quadratic upwind finite volume technique.6 In general, when the mass flow rate of gas forward through the bed is greater than the mass flow rate of gas returned on desorption, there will be a net cooling of the boundary layer. As the boundary layer is cooled, the inert layer will be cooled below the feed gas temperature will behave as a regenerative heat exchanger. The temperature of the inert layer will be reduced by the desorbed gas, and this, in turn, will precool the feed gas as it enters in the next cycle. As this process continues, the thermal boundary layer shifts further into the adsorption bed, and the temperature profile evolves. A typical pattern of evolution is shown in Figure 2. The thermal boundary layer moves through the bed as the temperature evolves. Consider the first adsorbing

segment in the main bed. Initially, the first adsorptive section of the bed is heated by the gas entering from the prelayer. As adsorption occurs in the node, the temperature of the node rises above the temperature of the gas entering from the prelayer. The gas from the prelayer now cools the segment. During the desorption steps, the node undergoes desorption (cooling) but is simultaneously heated by gas from further up in the adsorption bed. This process continues until cyclic steady state (CSS), when there is no net heating or cooling of any segment of the bed over a cycle. The final temperature profile for this case is shown in Figure 3. The initial conditions of the bed, the temperature, the pressure, and the loading all play a role in determining the evolutionary path to a CSS temperature profile, but this qualitative explanation still holds, and a singular temperature profile evolves on the basis of the process conditions and adsorption bed configuration. This qualitative model offers a simple account of the causes of the evolution of the temperature profiles and the ultimate convergence to a cyclic steady state (CSS). It follows from this explanation that any process condition affecting the heats of adsorption and the patterns of convection will shape the final CSS temperature profile. For example, changes in cycles, step times, pressure envelopes, purge-to-feed ratios, recoveries, product purities, and sieve properties will all affect the CSS temperature profile to varying extents. One critical parameter investigated experimentally in this paper is the effect of the adsorbent properties and the layering of the adsorbent bed in shaping the final CSS temperature profile. Our explanation predicts that a single-layered adsorption bed with a dry air feed

2704

Ind. Eng. Chem. Res., Vol. 40, No. 12, 2001

Figure 2. Cycle 50 and cycle 100 in the evolution to a CSS temperature profile.

Figure 3. Final cyclic steady-state (CSS) temperature profile, cycle 4000.

will not cause a temperature depression because there is no regenerative heat exchanger to trap the colder gas enthalpy on desorption. The desorbed gas will leave the bed below the feed gas temperature, but it will not precool the feed gas in the next cycle. For the case of wet feed, which is also investigated experimentally, we argue that water adsorption creates a de facto inert layer that then behaves similarly to the activated alumina described in the qualitative model. This qualitative explanation contrasts with some previous research efforts, which postulated that water adsorption might be a significant causal mechanism.7 Instead, we assert that water adsorption does not cause the thermal profile directly. A multilayered bed in which the temperature swing due to adsorption in the prelayer (or defacto prelayer) is less than the temperature swing in the main layer is a necessary condition for the

formation of a temperature depression. It also follows that, if the converse is true (i.e., the temperature swing in the prelayer is greater than that in the main layer), an inverted “temperature hump” will occur. This has, in fact, been observed experimentally and numerically. The simple explanation offered above is further complicated by secondary factors that modify the final CSS temperature profile. For example, axial conduction, heat exchange with the environment, and the temperature dependence of the isotherms are all factors that impinge on the final CSS temperature profile. These secondary factors can play a critical role in dissipating the temperature profile, as is often evident in laboratory-scale studies of oxygen VSA.8 This qualitative model also provides an explanation of the slow transient behavior that characterizes oxygen VSA. A simplified energy balance for the adsorption bed


in oxygen VSA can be written as

(

)

∂nN2 ∂nO2 ∂T ∂ 1 + qN2 ) - [u(T - To)] + qN2 ∂t ∂z Cp,s ∂t ∂t

(1)

where ) (Cp,gFjg/Cp,sFb) ≈ 0.0013. The first term on the right-hand side is the temperature change due to convection, and the second term is the temperature change due to adsorption. In this case, an average gas density is assumed, and only heats of adsorption and convection are considered. From this equation, can be identified as a perturbation parameter that is based on the ratio of the thermal capacitances of the gas and solid.9 We see that the temperature change at a point in the bed is based on a small convective term and a large adsorptive term. Thermal coupling of points in the bed occurs through the convective term. As a result, the transfer of thermal energy through the adsorption bed is very slow, and hence, closure of the energy balance is slow. This results in very slow convergence to cyclic steady state over hundreds, if not thousands, of cycles.10

Figure 4. Adsorption isotherm for water on NaX zeolite 300 K.

Equilibrium Theory of the Penetration of a Water-Loaded Zone To prevent the penetration of water and carbon dioxide into the main adsorbent layer of a lithium- or calcium-based adsorbent, a prelayer of activated alumina or NaX adsorbent is used in oxygen VSA for water and carbon dioxide removal.5 The removal of these readily adsorbed trace components could be achieved in a separate unit operation, but there are significant capital cost benefits of integrating these steps in a single adsorption bed, and this is established industrial practice. The resulting problem is to predict the penetration of water and carbon dioxide to determine the required depth of the prelayer. In this paper, we ignore the effect of carbon dioxide, as it is removed before entering the adsorption beds in our experiments. Water adsorption in the prelayer is modeled here using an equilibrium isothermal water balance based on a dual-site Langmuir model. This approach is similar to the development of LeVan.11 Oxygen and nitrogen are treated as inert species in the initial zone of the bed where water adsorption occurs. It is well-known that water is strongly adsorbed on X-type zeolites excluding nitrogen and oxygen adsorption. The mass balance for water at a point in the bed is

uP ∂y ∂Py FbRT ∂nw + + t ∂z ∂t t ∂t

(2)

Because water is a trace component, one can assume that there are no axial pressure or velocity gradients in the water-loaded zone. Water adsorbs strongly to both activated alumina and NaX, and its isotherm is highly favorable (Figure 4). A dual-site Langmuir model is used to describe the isotherm.

nw )

mw1bwPw mw2bwPw + 1 + bwPw 1 + dwPw

(3)

and dw ) do With the apwhere bw ) bo propriate isotherm parameters, the isotherm for water adsorption is as shown in Figure 4.12 eQ1/RT

eQ2/RT.

Figure 5. Final CSS water mole fraction profiles.

The plot shows an almost rectangular isotherm. There is a substantial range of partial pressures where the loading is essentially constant (i.e., dn/dp ≈ 0) and a small region at low partial pressures where the adsorbent undergoes substantial changes in loading for small changes in partial pressure. Equation 2 can be readily manipulated to yield the following characteristic equation:

dz ) dt

u FbRT dnw t 1 + t dPw

(

)

(4)

This characteristic equation can be used to determine the velocity of the water front during the adsorption and desorption steps for either constant-pressure or pressure-varying steps. However, this equation alone does not enable determination of the position of the waterloaded zone at cyclic steady state. At cyclic steady state, the water-loaded zone penetrates to some final position in the bed and shifts periodically between adsorption and desorption. We show this schematically in Figure 5. The adsorption front propagates as a shock wave because of the highly favorable shape of the water isotherm, whereas the desorption front propagates as a simple or dispersive wave. Cyclic steady state occurs when the quantity of water entering the bed during the feed steps equals the water removed during desorption. This condition is described by eq 5

∫0z (nw,i - nw,f) w

pbπd2 dz ) 4

∫0t

f

Vf pw dt RT

(5)

where the left-hand side of the equation is the change

2706


in water loading on the sieve between adsorption and desorption (ignoring gas-phase accumulation), and the right-hand side is the total number of moles of water entering the bed during the adsorption steps, which is known experimentally. The solution technique involves calculating the total water load per cycle per bed given the experimentally determined flow rate and the relative humidity of the feed gas. This is the right-hand side of eq 5. The unknown quantity is zw on the left-hand side of eq 5. A value of zw is assumed, and the water loading at the end of adsorption (nw,i) is calculated assuming the water front propagated as a shock front during adsorption. The final pressure and feed mole fraction of water are known. To determine the water loading at the end of desorption steps (nw,f), it is necessary to solve eq 4 to determine the shape of the simple desorption wave and ultimately the water loading as a function of z at the end of the desorption step. This requires experimental data to determine the pressure as a function of time and also to determine an average velocity during the blowdown and purge steps. The approach detailed here is similar to that of previous work by LeVan.11 However, in that work, a constant-pressure purge was assumed, and a more typical trace separation, such as air-drying, was investigated. In such cases, the feed and purge steps are long and are interspersed by rapid pressurization and depressurization steps. However, in the case of oxygen VSA, the process conditions are very different. The feed and blowdown steps are long, and the purge step is short, and hence, changes in pressure must be accounted for. The water loaded at the entrance of the adsorption bed undergoes an “internal purge” with dry enriched nitrogen during the blowdown and purge steps in the oxygen VSA cycle. Purge conditions for the removal of water (trace component) are imposed by the process conditions for bulk separation of oxygen and nitrogen. This technique enables both determination of the minimum prelayer thickness and characterization of different oxygen VSA cycles with respect to their purge-to-feed ratios over the prelayer. Equally important, this model enables estimation of the velocity of the water composition wave as a function of feed humidity. For example, the position of the water-loaded zone before or after a period of humid weather could be estimated. An alternative solution strategy is to solve eq 2 numerically over the water-loaded zone for the evacuation step. The numerical technique involves estimating zw and calculating the number of moles of water desorbed from the water-loaded zone over the desorption steps. When this calculated value equals the number of moles of water fed into the bed, the correct length of the water-loaded zone has been found. This numerical approach used the in-house simulator MINSA but was simplified by using experimental pressure and velocity data, and hence, only one equation, the water component balance (eq 2), had to be solved. Unlike the analytical solution, the numerical solution incorporates gas voidage into the calculation of the quantity of desorbed water. One possible complication is that it is necessary to determine the initial conditions of the bed at the end of the countercurrent desorption step. The pressure and mole fraction at the end of the feed step are known, but two cocurrent depressurization steps occur prior to the countercurrent desorption. However, for the partial

Figure 6. Schematic diagram of laboratory oxygen VSA unit.

pressure window investigated for these experiments, depressurization occurs in the region of the isotherm where dn/dp is approximately zero. Under these conditions, it is assumed that the cocurrent depressurization steps do not result in any change in the water mole fraction. Experimental Investigations Experimental Apparatus. A schematic of the laboratory-scale VSA unit is shown in Figure 6. It consists of two adsorption beds with a 104-mm internal diameter and a 1.85-m-long packed bed. The wall of the column is composed of 3-mm PVC with external insulation. The aim of these columns is to minimize heat exchange with the environment and minimize the thermal capacitance and axial conduction along the wall, while achieving adequate mechanical properties. Previous work used more “adiabatic” adsorption columns,8 but the PVC columns are more mechanically robust and better suited to this study. Along the length of the adsorption beds are junctionexposed type-T thermocouples that enter the bed radially for temperature monitoring. Thermocouples are also located throughout the pipework to monitor process streams. The location of pressure transducers is also shown on the equipment schematic. A paramagnetic oxygen analyzer is located after the 60-L product surge tank to monitor the product oxygen concentration. Flowmeters are also located on the feed, vacuum, and product lines. The product line is a small differential pressure type flowmeter, whereas the feed and vacuum lines are measured with annubar flowmeters. The feed and vacuum valves at the bottom of the adsorption beds are pneumatically actuated 1-in. valves. All other solenoid valves are 3/4-in. pneumatically actuated valves. The control valves enable the control of the flows through all cycle steps. The feed air is supplied at a dew point of -58 °C. This is achieved through the use of a desiccant dryer that also removes carbon dioxide. The feed air is supplied after a surge tank and a regulator valve at an approximate pressure of 1.5 bar absolute. For the experiments with wet air feed, the feed gas is passed through a humidifier that increases its humidity to 35-40% relative humidity. The humidity is monitored on-line. A rotary vane vacuum pump is used to enable evacuation of the adsorption beds. The control and monitoring of the VSA unit is performed by a PLC and SCADA software. The PLC/SCADA enables unattended

Ind. Eng. Chem. Res., Vol. 40, No. 12, 2001 2707 Table 1. Dual-Site Langmuir Model for Water Adsorption on UOP NaX mw1 bo Q1

9.19 6.75 × 10-7 59 497

mw2 do Q2

3.03 0.045 41 830

mol/kg bar J/mol

Table 2 bed 1/bed 2 mass AA (kg)

case and process description Case 1: dry feed air and a single layer of NaX molecular sieve Case 2: wet feed air and a single layer of NaX molecular sieve Case 3: dry feed air and a short prelayer of activated alumina (prelayer of 38 mm) and NaX (main layer) Case 4: dry feed air and a long prelayer of activated alumina (prelayer of 300 mm) and NaX (main layer) Case 5: wet feed air and a long prelayer of activated alumina (prelayer of 300 mm) and NaX (main layer)

bed 1/bed 2 mass NaX (kg) 10.4/10.3 10.4/10.3

0.3/0.3

9.8/10.1

1.9/1.9

8.5/8.6

1.9/1.9

8.5/8.6

Table 3. CSS Molar Balances

Figure 7. Eight-step VSA cycle.

operation and automatic shutdown in the event of an alarm condition. Cycle Conditions and Process Operation. For the experiments performed in this study, an eight-step VSA cycle was used. This is shown in Figure 7. On bed 1, the steps in the cycle are as follows: Steps 1 and 2 are feed repressurization followed by feed and product withdrawal; steps 3 and 4 are cocurrent depressurization steps providing purge and pressure equalization to the other bed; steps 5 and 6 are evacuation steps; and steps 7 and 8, to complete the cycle, are the receive purge and pressure equalization steps. Bed 2 is subjected to the same steps one-half cycle out-of-phase. The total cycle time is 60 s, and this represents a simplified industrial cycle. The process was run under manual control; no PID control was applied to the control valves to adjust the pressure profiles or control the purity or product flow. A typical pressure profile is shown in the next section. For details of control of oxygen VSA, see Beh et al.13 The adsorbent used for this study was 1/16-in. 13X APG pellets supplied by UOP. For the multilayered experiments where an inert prelayer was required 3/16in. F200 activated alumina (AA) from Alcoa was used. Table 1 shows the iostherm parameters for water adsorption on NaX.12 The evolution to CSS is notoriously slow in oxygen VSA. This is particularly true for the case of water penetration and evolution of the temperature profile to some cyclic steady-state position. The criteria for CSS is based on closure of the mass balance, constant product purity and performance parameters, balanced bed pressures, and constant temperature profiles over a 12-h period. Measurement of Nitrogen Loading and Water Loading. One critical part of this study was determining the length of penetration of the water-loading zone into the adsorption bed. We accomplished this by extracting small (∼2 g) samples of adsorbent from the bed at predefined distances and measuring the nitrogen

high-purity runs low-purity runs

feed flow (mmol/cycle)

vacuum flow (mmol/cycle)

product flow (mmol/cycle)

4253 ( 274 4643 ( 158

4110 ( 238 4066 ( 192

245 ( 19 664 ( 57

adsorption capacity of these samples on a Micromeritics ASAP2010M analyzer. Each sample was degassed at ambient conditions prior to the measurement of nitrogen capacity. We thus inferred the presence of water on a sample by the reduction in nitrogen loading at 810 mmHg nitrogen and 295 K. It is important not to degas the sample at elevated temperature, as this fully restores the nitrogen adsorption capacity. The use of reduced nitrogen capacity to infer the extent of water loading was considered preferable to TGA analysis, which would have determined the percent water loading on the sample. The latter does not provide information on the extent of nitrogen loading, which is the critical issue addressed in this work. It is well-known that water is strongly adsorbed onto alumina and 13X zeolite and that levels above approximately 1% w/w water prevent nitrogen and oxygen from adsorbing. Thus, measuring nitrogen adsorption in our gas adsorption analyzer allowed for a determination of the relative extent of water penetration in the beds. Experimental Results and Discussion Experimental Conditions. Ten experiments were conducted to determine the effect of the water loading on the thermal profiles and performance of an oxygen VSA unit. Table 2 shows the conditions for each experiment and indicates that the beds were equally packed with adsorbent. For each of these cases, two different product purity runs were investigated: a high-purity case (∼95% oxygen) and low-purity case (∼80% oxygen). The two different purity runs were performed to examine the effect of purity on CSS temperature profiles. Cyclic Steady-State Mole Balances and Performance Data. The mass balance and performance parameters were similar for all cases and are summarized in Tables 3 and 4, respectively. Recovery is the ratio of oxygen in the product stream to oxygen in the

2708


Table 4. CSS Performance Parameters

high-purity runs low-purity runs

O2 purity

recovery

kgPDc

BSF

95.5 ( 0.7% 80.4 ( 2.0%

24 ( 2% 48 ( 3.2%

10.5 ( 0.5 26 ( 4.0

1798 ( 256 772 ( 106

Figure 9. Axial temperature profiles at CSS for a single layer with dry air feed at low purity: (() end-of-feed repressurization, (9) end of feed, (2) end-of-provide pressure equalization, (×) end of evacuation. Figure 8. Characteristic cyclic steady-state pressure profiles for all cases: (b) high-purity runs, (9) low-purity runs.

feed stream. kgPDc is the kilograms of oxygen produced per day, and BSF is the bed sizing factor. BSF is the mass of sieve (in kilograms) required to produce one tonne of oxygen per day and is a typical measure of sieve productivity. The mass of alumina is not included in the calculation of BSF. The mass balance error for the high-purity runs was -1.4 ( 3.0%, and the error for low-purity runs was 0.7 ( 2.3%. There is a marked difference in the high- and lowpurity runs but only a slight variation between the cases within each purity. These differences are the result of different ambient temperatures, different product purities, and different quantities of NaX. Even though only manual control was applied, there was little variation in the pressure profile for each case, as Figure 8 indicates. Error bars are included to demonstrate the variation in the pressure profiles for the different cases and the different adsorption beds. The noticeable difference between these profiles is that the high-purity runs have a slightly higher end of the feed step pressure. This is due to the fact that less product gas is being withdrawn from the product tank. Cyclic Steady-State Temperature Profiles. Case 1: Single-Layered Bed and Dry Feed. In investigating CSS temperature profiles, the simplest case is for a single-layered bed with dry air feed. The axial temperature profiles for both the high- and low-purity runs are shown in Figures 9 and 10, respectively. No temperature depression is observed for either run, although the temperature of the desorbed gas is lower than the temperature of the feed gas. These observations are consistent with the qualitative explanation of the cold spot. There is no regenerative heat exchanger (inert zone) and, hence, no cold spot. An interesting feature is the difference in the thermal profiles between the high- and low-purity runs. The bed in the low-purity runs is characterized by a thermal profile of a single gradient, whereas the thermal profile tends to flatten off and decrease in the high-purity runs.

Figure 10. Axial temperature profiles at CSS for a single layer with dry air feed at high purity.

This difference is caused by the reduced convection of gas and the reduced heats of adsorption through the end of the bed in the high-purity runs. The nitrogen front in the high-purity runs is held well back from the outlet of the bed, and only small quantities of gas are passed out of the end of the bed. Table 4 shows that production in the high-purity runs is 2.6 times lower than in the low-purity case. The very small temperature swing over the end of the bed in the high-purity cases demonstrates the absence of nitrogen adsorption. These different thermal profiles are the result of different heats of adsorption and convection at each point throughout the bed, i.e., different process conditions. Importantly, these plots indicate that cyclic steady-state temperature profiles are a useful heuristic for understanding the oxygen VSA cycle and are useful for plant tuning and troubleshooting. Case 2: Single-Layered Bed and Wet Air Feed. In case 2, a single-layered bed was fed with wet air (relative humidity ≈ 40% at 18 °C and atmospheric pressure). The resulting cyclic steady-state temperate profile displays a clear temperature depression for both the low- and high-purity runs, as is shown in Figure 11. For clarity, we show only the low-purity run.


Figure 11. Axial temperature profiles at CSS for a single layer with wet air feed at low purity.

Figure 12. Water penetration in NaX for wet feed on NaX, low purity.

A clear temperature depression forms over the first 50 mm of the bed: a temperature drop from the feed temperature to about 5 °C occurs, followed by an increase in the temperature thereafter. From the qualitative theory developed, it is postulated that the cold spot can only occur when a prelayer of adsorbent is present with a reduced adsorptive capacity compared to that of the main bed, or, more specifically, a reduced temperature swing due to adsorption in the prelayer. It is evident from Figure 11 that the temperature swing over a cycle over the first 50 mm in case 2 is less than that in case 1, which suggests reduced nitrogen adsorptive capacity. This observation was confirmed by taking samples at different axial positions and analyzing them on the ASAP2010 gas analyzer as described earlier. Figure 12 shows the nitrogen loading capacity at different axial positions along the adsorption bed after the process had reached CSS. Figure 12 indicates that the water-loaded zone is approximately 25-40 mm from the inlet of the column and that no nitrogen loading occurs in this zone. Thereafter, the nitrogen loading increases over approximately 20 mm to near full adsorption capacity. There is a slight tail in this plot that could be attributed to a mass-transfer zone that is not accounted for in the simple equilibrium model. However, there is still a sharp transition point that justifies an equilibrium model to a first approximation. The position of the “inert” zone corresponds well to the location of the temperature depression shown in Figure 11.

Calculated Length of the Water Penetration Zone. Based on the characteristic eq 4 and using the experimentally determined pressure and velocity profiles for the desorption step, the z-t plot shown in Figure 13 was generated. This plot describes lines of constant mole fraction over the duration of the countercurrent desorption steps. Bed pressure is also plotted on the graph. During the major part of the desorption step, the pressure in the bed decreases exponentially; it then increases during the purge step. On the basis of these lines of constant mole fraction and eq 5, the penetration of the water-loaded zone was calculated to be 38 mm. The relative humidity of the feed air was 40% at 18 °C and atmospheric pressure. The pressure profile and the required velocity data for calculation of the length of the water-loaded zone were taken from the low-purity run for case 2. The numerical solution using MINSA was calculated to be 36 mm. The MINSA calculation includes the change in water mole fraction within the gas phase between the desorption and adsorption steps. Both of these results compare well with the experimental result of between 25 and 40 mm. However, this large tolerance in the experimental result highlights the limitations of the experimental technique in determining the location of the water front. As a result, it is difficult to account for the shape of the experimental water mass-transfer zone with sufficient accuracy to ascertain whether the equilibrium assumption is valid. MRI (magnetic resonance imaging) is an alternative technique that has recently been applied for the determination of the position of a water front in chromographic studies.14 Potentially, MRI offers much improved accuracy, which would enable an improved comparison between experimental and modeling results. Within experimental error, it appears that a simple equilibrium model, solved by characteristics or a numerical technique, provides a useful design tool for determining the thickness of the water-loaded zone and for investigating the impact of process parameters (such as feed humidity, internal purge velocity, and pressure profile) on the CSS position of the water front. For example, the same VSA cycle, except with the feed gas having a relative humidity of 90% at 30 °C and atmospheric pressure, results in a calculated water penetration zone of 280 mm. This is reasonably consistent with current industrial practice, where a 300 mm prelayer of NaX might be used, although, in practice, an industrial unit would be subjected to higher purgeto-feed ratios. It should be noted that, if activated alumina were to be used, a shorter prelayer would be possible because of its increased adsorptive capacity for water. This analysis also highlights the need to develop VSA cycles that provide sufficient internal purge to ensure that the water front cannot penetrate beyond the design prelayer. For example, there are dangers of producing more product gas at lower purities because this reduces the ratio of internal purge to feed, and as a consequence, the water front will penetrate further into the bed, possibly breaking through the prelayer. Effect of Temperature Swing Due to Adsorption. Cases 1 and 2 and our analysis demonstrate the presence of a water-loaded zone that displaces nitrogen, but they ignore the heat generation associated with the adsorption and desorption of water in this section of the bed. As shown in Figure 11, the temperature swing due

2710


Figure 13. Lines of constant mole fraction during the desorption steps.

Figure 14. Case 3 low-purity run, dry air feed.

to adsorption is less in the water-loaded zone than in the main bed, but it is still not zero. The temperature swing associated with water in the prelayer can be calculated by solving an energy balance including convection and the heat of adsorption associated with water over the water-loaded zone. An approximate solution to the energy balance gives a temperature swing of 2.3 °C, which is less than the swing in the nitrogen layer (approximately 4.3 °C in the singlelayered bed with a dry air feed). This analysis suggests that one possible simplified model is to treat the waterloaded zone as an inert prelayer and ignore the heat effects associated with water adsorption/desorption. This idea is explored in cases 3-5. Cases 3-5: Temperature Profiles with an Inert Prelayer. In case 3, dry air feed was used, and a short prelayer of activated alumina was placed at the entrance of the adsorption bed to simulate an inert zone as measured earlier. The thickness of the prelayer was ∼40 mm, which was based on the experimentally determined thickness of the water prelayer in case 2. The resulting CSS temperature profile, shown in Figure 14, is similar to the profiles for case 2 (Figure 9), suggesting that wet loaded NaX and the dry alumina cause similar thermal profiles because both prelayers do not adsorb nitrogen. Evidently, the effect of heat generation due to water adsorption/desorption plays a minor role in the formation of the overall thermal profile.

Figure 15. Case 4 low-purity run, dry feed, and long prelayer: (() end-of-feed repressurization, (9) end of feed, (2) end-of-provide pressure equalization, (×) end of purge.

Unfortunately, the feed temperature was not the same for cases 2 and 3. However, it is easy to show that different feed temperatures lead to uniform translation of the entire profile.4,8 Therefore, it is more important to look at the temperature drop when comparing results. In case 2, a temperature drop of 15 °C was observed, whereas a 10 °C drop was observed in the latter case. Part of this difference is due to uncertainty in the thickness of the water-loaded zone, as shown below. One limitation with this analysis is that the exact thickness of the water-loaded zone could not be determined, and the effect of any water mass-transfer zone in case 2 has not been taken into account. To overcome this deficiency, additional experiments were performed with a long prelayer of activated alumina with dry (case 4) and wet (case 5) feed air. The objective of these experiments was to consider only the effect of water adsorption in the prelayer and to ensure that the prelayer is sufficiently thick that water cannot penetrate through to the main nitrogen adsorbing layer. Any thermal effects due to water adsorption in the overall thermal profiles are isolated by these experiments. The resulting CSS temperature profiles shown in Figures 15 and 16 are compelling. These plots are very similar and confirm that water loading on activated alumina has only a marginal effect on the overall temperature


Figure 16. Case 5 low-purity run with wet air feed.

profile. This confirms that formation of a cold spot is determined by differences in adsorptive capacity of the two layers and that the heat of adsorption of water plays only a minor role. An interesting difference in the temperature profiles is that the temperature swing at the entrance of the bed for the wet feed (1.4 °C) is greater than that for the dry air feed (0.4 °C) and that there is a slight increase in the temperature of the bed above the feed temperature in the wet air feed case. Also, the temperature minimum for the dry air feed case is slightly lower than that for the wet air feed. These results are consistent with the heat of adsorption due to the water loading in the wet air feed case, but together, these differences have only a very small effect on the overall temperature profile. Figures 15 and 16 (cases 4 and 5, respectively) also display more pronounced cold spots than the previous cases. This can be explained using the qualitative explanation already developed, because with a thicker prelayer, the role of the prelayer as a regenerative heat exchanger is enhanced. This also suggests that there is an optimal ratio of prelayer to main layer of adsorbents that gives the maximum temperature depression. Evolution of Thermal Profiles. The evolution of thermal profiles provides further insight into the role of water adsorption in oxygen VSA. Two interesting evolutionary plots are shown below. Figure 17 displays a very sharp temperature increase at the front of the bed as wet feed gas initially contacts a dry prelayer of 300 mm of alumina. The VSA unit was at CSS with dry air feed (case 4) when it was switched to wet air feed (cycle zero). Immediately, there was a sharp increase in the temperature profiles resulting from water adsorption on clean alumina. Over 1000 cycles, this sharp temperature rise was convected through the front section of the bed (as shown by thermocouples downstream) and largely disappeared. This transient behavior is not evident with dry feed air cases. The overall effect is an axial temperature profile at cycle zero that is very similar to the profile at cycle 1000. This confirms, as previously observed, that the CSS profiles are very similar for the dry and wet feed cases. However, the CSS profiles neglect this important difference in the transient behavior in the wet and dry feed cases. The implications of this transient profile are

Figure 17. Transient effect of introduction of wet air feed. Axial positions of thermocouples: ([) 25 mm, (b) 100 mm, (() 200 mm, (0) 300 mm.

important for understanding the effect of varying humidity on temperature profiles in the bed. The observed temperature “spike” results from the significant heat of adsorption of water initially in a very thin section of clean bed giving a substantial temperature rise. The heat of adsorption will be highest on a clean bed because of the large isosteric heat of adsorption at zero loading. This heat is convected through the bed, and the temperature spike is slowly dissipated. The thermal wave travels slowly because of the slow speed of the convection of thermal energy through the adsorption bed. From eq 1, the velocity of a simple thermal wave (nonadsorptive) is the product of the gas velocity and the quantity . As discussed earlier, is the ratio of the gas thermal capacitance and the solid thermal capacitance ( ≈ 0.013). It should also be noted that this temperature wave resulting from water adsorption is decoupled from the water composition wave, which is held inside the initial section of the bed. As the waterloaded zone increases in length, the absolute thermal capacitance of the water-loaded zone increases. In effect, the heat generated by water adsorption occurs over a larger mass of sieve, leading to a much smaller temperature increase at cyclic steady state. A significant feature of oxygen VSA is the slow evolution of CSS temperature profiles. The qualitative explanation developed in this paper argued that this was due to the low ratio of the thermal capacitance of the gas compared with the thermal capacitance of the solid and, hence, the slow propagation of heat through the adsorption bed by convection. Figure 18 presents a typical plot of the evolution from one CSS temperature profile to another. In this case, there is a gradual evolution from an initial CSS condition to a final CSS condition over 800 cycles. This slow transient behavior makes numerical simulation of oxygen VSA complex and underscores the need for acceleration schemes to reduce the computational time to CSS. Also, it suggests that at least 800 cycles are needed to fully assess the impact of a process change on performance of a PSA bulk gas separation process. Evolution of the Water-Loaded Zone. In addition to determining the evolution of the temperature profile, it is also possible to estimate the time required to

2712


Figure 18. Asymptotic convergence to CSS of several axial temperatures. Axial positions of thermocouples: (() 500 mm, (b) 100 mm, (() 300 mm, (0) 200 mm.

not possible to isolate performance differences between the deeper temperature depression and the reduced mass of adsorbent. Third, automatic control was not used in this study, and as a result, there were minor variations in purity, flows, feed temperature, and pressure profiles among the cases investigated. These differences all impinge on performance, and again, the effects of the cold spot cannot be readily isolated. These limitations suggest that the best method for investigating performance changes due to temperature is to use a single-layered bed and adjust (both heat and cool) the feed gas temperature to simulate the effect of a cold spot of varying magnitude. Interestingly, there is evidence that NaX performance is enhanced at lower temperatures and that the cold spot would benefit NaX performance.16 This is apparently inconsistent with previous industry concerns about the cold spot. Recent patents recognize that the performance of different sieves differs markedly with decreasing temperature and that the cold spot, under some circumstances, could enhance overall VSA performance. For example, matching different sieves with different axial temperature zones would give rise to an optimal multilayered bed.17 Conclusions

Figure 19. Evolution of the length of the water-loaded zone.

achieve CSS of the water-loaded zone. This involves solving eq 2 from cycle zero to CSS and is shown in Figure 19. The water-loaded zone evolves to 98% of its final value with 970 cycles (0.7 days). This dynamic response is of a similar order of magnitude as the evolution of the temperature profile, as is shown in Figure 18. Implications of this Study. The principal motivation of industrial gas companies to understand the cold spot has been its unexpected, and in many cases deleterious, effect on oxygen VSA performance.15 This study has not demonstrated any marked performance differences between different cases. There are several reasons for this. First, the temperature depressions observed are substantially smaller in magnitude than those typically found in industrial oxygen VSA units. These lower cold spots are due, in part, to the substantially adiabatic nature of industrial oxygen VSA units. Also, current industrial processes use a main layer of adsorbent, such as LiX, that has far higher loading capacity and larger heat of adsorption than the NaX used in this study. This leads to a larger temperature swing in the main layer between adsorption and desorption and a deeper temperature depression at cyclic steady state. Second, in cases 4 and 5, where the temperature depression was more marked, there was substantially less adsorbent in the beds. Under these conditions, it is

A main conclusion of this work is that a temperature depression will not occur in a single-layered bed. Both an inert prelayer (activated alumina with a dry air feed) and a water-loaded zone give rise to the process conditions amenable to the formation of a temperature depression or cold spot. Moreover, both a water-loaded zone and an inert zone of the same thickness result in similar cyclic steady-state temperature profiles, as was observed when a single-layered bed with wet air feed (case 2) was compared with a dry air feed and short alumina prelayer (case 1). We have also demonstrated that wet feed air loading on to activated alumina and dry air feed with activated alumina display similar overall temperature profiles. There is only a marginal difference in the temperature profiles due to the heat of adsorption of water. However, water adsorption has a marked effect on the transient temperature profiles, and interestingly, this effect is dissipated with time. Water is strongly adsorbed onto both NaX and activated alumina, and this creates a short water-loaded penetration zone. This zone evolves slowly to a cyclic steady state where there is no net accumulation of water on the adsorption bed. The length of this water penetration zone can be estimated using a simple isothermal, equilibrium model that can be readily solved by the method of characteristics or by numerical methods. This calculated result matches reasonably well with experimental results and provides a simple technique for determining process conditions that impinge on the thickness of the water-loaded zone. Finally, this work has provided a simple qualitative account of the cold spot that has been the subject of much interest in the broad adsorption community. The effect is the result of the interaction between the heats of adsorption and convection heat fluxes imposed by process conditions in a bulk gas separation. A more detailed multiple-scale analysis of this problem has been previously presented.10 There are several implications of this study. To simulate wet air feed conditions in the laboratory,


approximately 70 mm of alumina can be used with a dry air feed. This simulates the effects of water loading (∼50% relative humidity at 25 °C) without the need for a wet air feed stream. A water-loaded zone with a length of 70 mm would have substantially higher humidity than that used in this study. However, 70 mm does not include any safety margin that would be required in an industrial plant. Second, in industrial practice, it might be more appropriate to use alumina than NaX for water removal. Alumina is cheaper than NaX, and after NaX is initially loaded with water, it has virtually no nitrogen loading capacity. Another advantage of using alumina as a prelayer is that the resulting cyclic steady-state temperature profile and process performance are essentially independent of humidity. One option is to use two layers: alumina in the first section for removal of water at typical design humidities and temperatures and then NaX as a safety zone for infrequent but extreme conditions. Future work on this project involves optimizing the oxygen VSA unit performance given the presence of thermal profiles. There are numerous optimization strategies to be explored, including layering of adsorbent recognizing that gas loading and separation performance are functions of temperature. Alternatively, changing bed flow paths can be used to adjust the CSS temperature profiles. Nomenclature Cp,g ) heat capacity of gas [1000 J/(kg K)] Cp,s ) heat capacity of solid [1000 J/(kg K)] d ) diameter of adsorption bed (0.104 m) mw1, mw2, bo, do, bw, dw, Q1, Q2 ) dual-site Langmuir parameters for water adsorption on NaX nN2, n02, nw ) loadings of nitrogen, oxygen, and water, respectively [mol/(kg of bed)] nw,i, nw,f ) loadings of water at the beginning and end, respectively, of the desorption step [mol/(kg of bed)] P ) total pressure (bar abs) Pw ) partial pressure of water (bar abs) Pf ) total pressure at the end of the feed step (bar abs) qN2, qO2 ) heats of adsorption for nitrogen and oxygen, respectively [J/(mol K)] R ) universal gas constant 8.3143 × 10-5 [bar m3/mol K] T ) temperature (K) To ) reference temperature (K) t ) time (s) tf ) time duration of feed step (s) u ) superficial gas velocity (m/s) Vf ) volumetric flow rate of feed gas (m3/s) y ) mole fraction of water in the gas phase yf ) mole fraction of water in the feed gas z ) axial bed length (m) zw ) length of penetration of the water-loaded zone (m) ) dimensionless ratio of the thermal capacitances of the gas and solid t ) total bed voidage (0.7) Fg ) gas density (kg/m3) Fb ) bulk adsorbent density (700 kg/m3)

important work in this area undertaken by Bob Thorogood and Roger Whitley. Thanks are particularly due to Dr. Doug Dee (APCI) who provided valuable input to the ideas contained in this paper. Thanks are also due to the technical staff at Monash University who provided mechanical, electrical, and computer assistance in the installation and operation of the oxygen VSA unit. Literature Cited (1) Ruthven, D.; Farooq, S.; Knaebel, K. Pressure Swing Adsorption; VCH Publishers: New York, 1994. (2) Yang, R. Gas Separation by Adsorption Processes; Butterworths: Markham, ON, Canada, 1987. (3) Wankat, P. Rate Controlled Separations; Elsevier Science: New York, 1990. (4) Collins, J. Air Separation by Adsorption. U.S. Patent 4,026,680, May 1977. (5) Kumar, R. Vacuum Swing Adsorption Process for Oxygen ProductionsA Historical Perspective, Sep. Sci. Technol. 1996, 31 (7), 877-893. (6) Webley, P.; He, J. Fast Solution-Adaptive Finite Volume Method for PSA/VSA Cycle Simulation 1. Single Step Simulation. Comput. Chem. Eng. 2000, 23, 1701-1712. (7) Stegmair, M. Temperatureffekte beim Zusammenspeil zwischen Lufttrocknung und Lufttrennung bei technischen Druckwechseladsorptionsprozessen. Institut fur Chemische Verfahrenstechnik, Universitat Stuttgart, Stuttgart, Germany, July 1996. Minor thesis, undergraduate. (8) Wilson, S.; Webley, P.; He, J. Thermal Effects in O2 VSA. 1998 CHEMECA Conference, Port Douglas, Queensland, Australia, Sept 28-30, 1998. (9) Harriott, G.; Tsirukis, A. Convective Approximation of Adsorption Processes. Proceedings of the Fifth International Conference of Fundamentals of Adsorption; LeVan, M. D.; Kluwer Academic Publishers: Boston, MA, 1995; pp 353-360. (10) Wilson, S.; Webley, P. A Technique for Accelerated Convergence of Cyclic Steady State in Oxygen VSA Simulations; 4th Topical Conference on Separations Science & Technology, AIChE Annual Meeting, Dallas, TX, Nov 1999. (11) LeVan, D. Pressure Swing Adsorption: Equilibrium Theory for Purification and Enrichment. Ind. Eng. Chem. Res. 1995, 34, 2655-2660. (12) ADSIM User Manual, version 6.1; Aspentech: Cambridge, MA, 1997. (13) Beh, C.; Wilson, S.; Webley, P.; He, J. The Control of the VSA Process for Air Separation. Proceedings of the Second Pacific Basin Conference on Adsorption Science and Technology; Do, D., Ed.; World Scientific: Singapore, 2000. (14) Karsten-Bar, N.; Balcom, B.; Ruthven, D. Direct Measurement of Transient Concentration Profiles in Molecular Sieve Particles and Columns by MRI. Proceedings of the Second Pacific Basin Conference on Adsorption Science and Technology; Do, D., Ed.; World Scientific: Singapore, 2000. (15) Watson, C.; Whitley, R.; Meyer, M. Multiple Zeolite Adsorbent Layers in Oxygen Separation. U.S. Patent 5,529,610, 1996. (16) Leavitt, F. Low-Temperature Pressure Swing Adsorption with Refrigeration. U.S. Patent 5,169,413, 1992. (17) Notaro, F.; Mullhaupt, T.; Leavitt, F.; Ackley, M. Adsorptive Process and System Using Multilayer Adsorbent Beds. U.S. Patent 5,674,311, October 1997.

Acknowledgment We acknowledge the contributions of Air Products and Chemicals (APCI) and the Australian Research Council to the funding of this project. We also acknowledge the

Received for review September 12, 2000 Accepted March 15, 2001 IE000801A

(Water) in a Bulk Separation PSA Process: The ... - ACS Publications

Recommend Documents