Elsevier

Chemical Engineering Journal

Volume 292, 15 May 2016, Pages 366-381
Chemical Engineering Journal

Non-isothermal numerical simulations of dual reflux pressure swing adsorption cycles for separating N2 + CH4

https://doi.org/10.1016/j.cej.2016.02.018Get rights and content

Highlights

  • A non-isothermal numerical model was developed for dual reflux pressure swing adsorption cycles.

  • Bed pressure, flow and temperature profiles from N2 + CH4 experiments were well matched.

  • Predicted CH4 mole fractions had r.m.s. deviations of 0.003 and 0.024 for the light and heavy products.

  • Simulations provide insight into the optimisation of the feed step time and mass transfer effects.

Abstract

A non-isothermal model of dual reflux pressure swing adsorption (DR-PSA) was developed using a commercially available software package to numerically solve the dynamics of the unit operation. Importantly the model includes a full energy balance, which is a feature not reported previously in the literature for simulations of DR-PSA cycles even though bed temperature swings of 10–20 K have been observed in experimental studies. The simulation allowed solution of the pressure-flow network for cycles in the PL-A configuration, where feed gas enters the middle of the low pressure column and where pressure inversion is conducted by transferring gas rich in the more adsorbed component between beds. At cyclic steady state the simulations contained material balance errors comparable in magnitude to those reported previously for isothermal DR-PSA simulations; however, these were accounted for using a robust correction scheme. Predictions of the pressure, flow and temperature profiles within the beds for a range N2 + CH4 mixtures being separated using activated carbon were in good agreement with the corresponding results of 24 DR-PSA experiments recently reported by Saleman et al. (2015). The root mean square deviation of the predicted methane mole fractions from the experimental values were 0.003 and 0.024 for the light (N2-rich) and heavy (CH4-rich) product streams, respectively. Parametric studies conducted with the model show how the cycle design can be optimised with respect to reflux flow and the feed-purge step duration, and also illustrate the need for reliable values of the sorption mass transfer coefficient.

Introduction

Pressure swing adsorption (PSA) is often an attractive process for applications requiring the separation of gas mixtures, given its relatively low energy consumption and the use of non-hazardous materials. Gas separations using PSA have been adopted in different industries, for example hydrogen purification, bulk air separations and hydrocarbon recovery [1], [2]. Conventionally, PSA aims to produce a pure light product (concentrated in the more weakly-adsorbed component) and often uses some of that light product as a reflux or purge stream to regenerate a column containing a saturated adsorbent bed. This configuration is often referred to as stripping PSA since the strongly adsorbed component is stripped from the feed, and most industrial applications fall in this category [1], [2]. An alternative to stripping PSA is enriching PSA, which aims to produce a pure heavy product (concentrated in the more strongly-adsorbed component) with some of the heavy product used as a reflux or purge stream to ‘regenerate’ a column containing an unsaturated adsorbent bed. Various models and experiments investigating enriching PSA can be found in the literature including, for example, those by Ebner and Ritter [3], Yoshida et al. [4], Reynolds et al. [5] and Zhang and Webley [6]. A disadvantage of either stripping PSA or enriching PSA is that only one of the separation’s product streams is of high-purity while the purity of the waste stream is limited [4], [7]. Consequently, it can be difficult to use one of these single reflux PSA configurations to separate a mixture into two products that both satisfy stringent composition specifications.

Dual reflux PSA (DR-PSA) involves the use of two columns operating to separate a feed mixture into two product streams simultaneously, essentially by combining a stripping and an enriching PSA cycle into a single unit. As shown in Fig. 1, the feed stream enters one of the columns at an intermediate position, product streams are drawn from one end of each column, and a reflux stream enters each column at its other end. The two columns are at different pressures, and the reflux streams entering a bed are taken from the product stream leaving the other bed and passed through a compressor or a valve to raise or lower their pressure, respectively. A typical DR-PSA cycle includes four basic steps: feed (FE), purge (PU), pressurization (PR) and blow down (BD), which occur in the pairs FE/PU and PR/BD so that every half-cycle is symmetric with each column’s state swapping during the second half. The cycle can be configured so that the feed stream enters either the high pressure (PH) column or the low pressure (PL) column. Similarly, the cycle can be configured so that the pressure inversion is carried by transferring gas between the ends of the columns that is rich in either the heavy (more adsorbed) component (A), or in the light (less adsorbed) component (B). This leads to the four DR-PSA configurations, referred to as PH-A, PH-B, PL-A and PL-B [8], [9].

There have been a number of experimental and simulation-based studies of DR-PSA. One of the first DR-PSA experiments was performed by Diagne et al. [10] and aimed to separate CO2 and air using a PH-A configuration. The experimental results showed that DR-PSA was able to simultaneously produce both light product and heavy product with high purities: 3 mol% CO2 in the light product and 80 mol% CO2 in the rich product from a feed of 20 mol% CO2. In their follow-up work, they investigated key operating parameters of the DR-PSA cycle such as pressure ratio, axial feed position, reflux ratio and throughput, and the advantages of DR-PSA over conventional stripping/enriching PSA were also demonstrated [11]. A series of DR-PSA experiments aimed at separating ethane and nitrogen using BAX-1500 activated carbon was later performed by McIntyre et al. [12]. A feed containing 0.78 mol% of ethane in nitrogen was concentrated to 40–60 mol% of ethane in the heavy product, while the ethane concentration in the light product was decreased to as low as 30 parts-per-million. Subsequently, McIntyre et al. investigated the effects of four key operational parameters on cycle performance: light reflux rate (RL), feed step time (tF), heavy product flow rate (H) and feed composition (yF) [13]. Bhatt et al. performed a total of 9 experiments separating CH4 and N2 in a PH-A configuration using Norit RB1 activated carbon. A feed containing 5 mol% CH4 was separated using a pressure ratio of 4 into a heavy product containing as much as 65 mol% CH4 and a light product with 1.5 mol% CH4 [14]. The experimental results further proved that DR-PSA cycles are able to achieve a high degree of separation with pressure ratios that are low in comparison with conventional PSA cycles. Most recently, Saleman et al. [15] reported the results of 24 PL-A experiments conducted using the activated carbon Norit RB3 to separate N2 and CH4, with the primary objective being to minimise methane content in the light product while still having a useful concentration of methane in the heavy product. In one experiment a feed containing 2.4 mol% CH4 in N2 was separated into a heavy product stream containing 35.7 mol% and a light product containing less than 0.3 mol% CH4. Saleman et al. [15] were able to improve the separation performance of the DR-PSA cycles through empirical adjustments of the cycle’s parameters; however, a reliable model of those experiments would be extremely useful for efficient cycle optimisation as well as assessing the potential for operation at industrial scale. The development and validation of such a model was the primary objective of this work.

Since PSA cycles are batch processes subject to significant dynamic transients, modelling them accurately is complex. The need to describe the internal flow loop used in the DR-PSA cycles combined with the changing flow directions between steps further increases the modelling complexity. Two different modelling approaches have generally been followed to describe DR-PSA cycles: those based on equilibrium theory and those based on numerical dynamic models. Equilibrium theory models are based on using the method of characteristics to solve the adsorption bed continuity equations as suggested by Knaebel and Hill [16] and the model can often be solved analytically. A series of assumptions is usually applied to make such an approach tractable, including linear uncoupled isotherms, binary feed gas, isothermal operation, ideal gas behaviour, zero axial pressure drop in the bed, local equilibrium between the gas and adsorbed phases, zero axial or radial diffusion and uniform axial interstitial gas velocity at any position along the bed length. The first equilibrium theory model for DR-PSA cycles was proposed by Ebner and Ritter [7]. The model assumed that perfect separation was always achieved and showed that this could be accomplished by using a very low pressure ratio. The equilibrium theory model was later advanced by Kearns and Webley [8], [9]. In their work, the four configurations (PL-A, PH-A PL-B, PH-B) were identified and the gas concentration profiles inside the column within each step were calculated. The energy consumption and the product recoveries associated with the four configurations of DR-PSA cycles were discussed and compared. Recently, the equilibrium theory model was further extended by Bhatt et al., with the objective of finding (numerically) an optimal feed position and amount of feed per cycle per kilogram of adsorbent for a specified DR-PSA configuration [17]. However, the main drawback of equilibrium theory models is the required assumption of perfect separation, so they cannot be used to predict the product purities of DR-PSA processes.

To estimate the product purities and other key dynamic features of DR-PSA cycles, numerical models have been developed. The first was presented by Diagne et al. for CO2/N2 separations and the predicted product purities showed a relatively close match with their experimental results [18]. The model used a simplified method to predict the product purities and assumed the system to be isothermal. Another key simplification of the model was the assumption that none of the light component (N2) was adsorbed. An enhanced numerical method was proposed by Sivakumar and Rao, and the simulation results were presented for CO2/N2 [19] and O2/N2 [20] separations (no experiments with which the results could be compared were reported). In the latter case, competitive adsorption between the two components was considered, and they investigated a modified cycle design intended to increase the achievable product purities by mitigating the effects of transient variations in product composition during the feed-purge step. By eliminating energy balance considerations and implementing the model directly into FORTRAN, their simulations were able to run relatively quickly (1 h).

Bhatt et al. [14] developed an isothermal numerical model to describe their experimental methane and nitrogen separations using Norit RB1 activated carbon with a PH-A configuration; their model was implemented in the commercially available software package Aspen Adsorption [21]. The simulation results predicted a CH4 mole fraction in the heavy product stream that differed by between −0.02 and 0.08 from their reported experimental results, for which the average CH4 mole fraction was about 0.6. To reduce the computational complexity, rather than numerically calculating the full dynamics for both columns across the entire cycle, the model of Bhatt et al. [14] utilised so-called interaction columns to map the properties of product streams produced from a bed at one stage of a cycle to the properties of a future feed stream to that same bed. Consequently, the model could only demonstrate half of the DR-PSA cycle at any given time. Other simplifications included the use of a valve to act as a compressor with a positive pressure drop but a constant, specified reverse flow to achieve the pressure inversion. The assumption of isothermal operation was based on the experimental measurement of a 2 K temperature swing on the column’s external surface; no experimental information about the temperature variation inside the adsorbent bed was available. Bhatt et al. subsequently reported another set of numerical simulations of the ethane–nitrogen separation experiments carried out by McIntyre et al. in 2010 [13], [22]. A total of 19 scenarios were simulated to demonstrate the effect of different parameters, including axial feed position (zF), tF, RL, H and yF. However, the same isothermal approach was used, which meant that the significant experimental temperature variation of 25 K reported by McIntyre et al. was ignored by the model [13]. This could help explain the over-prediction of the ethane content in the heavy product, yH, which amounted to more than 5 mol% (out of about 60 mol%) in many scenarios.

Incorporating the ability to predict bed temperatures into a model of DR-PSA cycles is clearly important, given that laboratory-scale experiments with mixtures of methane + nitrogen [15] and ethane + nitrogen [13] on activated carbon produced temperature swings of 10 K and 25 K, respectively. The sizes of adsorbent beds used for industrial scale separations usually have larger volume-to-surface area ratios and hence tend to be more adiabatic than laboratory scale apparatus; thus it is even more important for simulations of larger scale columns to include energy balance to capture the effects of temperature variations in the bed. Furthermore, bed temperature data are usually a key output of laboratory-scale DR-PSA experiments, and play a central role in understanding the cycle performance. For example, Saleman et al. [15] used bed temperature measurements to infer the location of the adsorption wave front during cycles and study how the variation of cycle parameters such as product flow rates, reflux flow rates and feed step time affected separation performance.

In this work, we present a comprehensive numerical model of a DR-PSA cycle in which many of the assumptions utilised in the past have been eliminated. In particular, a dynamic numerical model consisting of material, momentum and energy balances as well as a complete pressure-flow network for DR-PSA cycles was implemented in the software Aspen Adsorption [21]. The model was benchmarked against the experimental results reported by Saleman et al. [15]; all the parameters used in the model were determined from independent measurements and none were adjusted to minimise the difference between the experimental DR-PSA results and the model’s predictions. In this context, the agreement achieved between the measured and predicted dynamic bed pressures, product purities and flows, and bed temperature profiles is excellent. Finally, the model is used to investigate the sensitivity of the DR-PSA cycle and its performance to reflux flow rate, the adsorbent’s mass transfer rate, and the degree of discretisation used in the finite difference approximation of the bed.

Section snippets

Equations and numerical solution

The full numerical model of the DR-PSA cycle was developed in the commercial simulation software package, Aspen Adsorption [21], which is capable of modelling dynamic batched adsorption processes with rigorous pressure flow relationships as well as mass and energy balances. The basic assumptions and governing equations used in the simulation of the adsorbent bed are listed in Table 1 and may be summarised as follows [21]:

  • Momentum balance and pressure drop described by the Ergun Equation.

Dynamic pressure, flow and temperature profiles at CSS

Comparisons between the CSS bed pressures, reflux flow rates and bed temperature at three axial locations calculated for a simulation of a specific experiment (Run 13) and the corresponding measured values are shown in Fig. 3, Fig. 4, Fig. 5, respectively. Summaries of the results for all 24 simulations of the experiments conducted by Saleman et al. [15] are presented in Table 5. For the purpose of the comparison shown in Fig. 3, two simulations of Run 13 were conducted with the normal

Bed temperatures and reflux flows

Fig. 5 shows that the local bed temperature swing varied with position: at the light reflux end of the bed (T1) it was about 4 K whereas at the heavy reflux end (T11) it was about 8 K. This reflects the nature of the PL-A cycle configuration in which the pressure inversion occurs via the transfer of gas that is rich in CH4 through the heavy reflux end of the bed, making it the main region in which adsorption/desorption occurred. Furthermore, the simulation also accurately predicted a temperature

Conclusions

A numerical model of a DR-PSA cycle was constructed using the commercial software package Aspen Adsorption [21], which included a full energy balance and solved the pressure-flow network without the use of interaction units. The model was used to simulate the 24 DR-PSA experiments conducted by Saleman et al. [15] who separated N2 + CH4 mixtures using activated carbon and a PL-A cycle. All of the parameters in the model were set to either values specified in the experiment or determined from

Acknowledgements

This research was supported by the Department of Environment Regulation of Western Australia through its Low Emissions Energy Development (LEED) Fund and by the Australian Research Council through GL’s Discovery Early Career Researcher Award (DE140101824). YZ thanks the Department of Chemical and Materials Engineering at the University of Auckland and the Chevron Chair Endowment at the University of Western Australia for providing a scholarship.

References (34)

  • S.P. Reynolds et al.

    Heavy reflux PSA cycles for CO2 recovery from flue gas: Part I. Performance evaluation

    Adsorption

    (2008)
  • J. Zhang et al.

    Cycle development and design for CO2 capture from flue gas by vacuum swing adsorption

    Environ. Sci. Technol.

    (2008)
  • A.D. Ebner et al.

    Equilibrium theory analysis of dual reflux PSA for separation of a binary mixture

    AlChE J.

    (2004)
  • D. Diagne et al.

    New PSA process with intermediate feed inlet position operated with dual refluxes: application to carbon dioxide removal and enrichment

    Chem. Eng. Jpn.

    (1994)
  • D. Diagne et al.

    Parametric studies on CO2 separation and recovery by a dual reflux PSA process consisting of both rectifying and stripping section

    Ind. Eng. Chem. Res.

    (1995)
  • J.A. McIntyre et al.

    High enrichment and recovery of dilute hydrocarbons by dual-reflux pressure-swing adsorption

    Ind. Eng. Chem. Res.

    (2002)
  • J.A. McIntyre et al.

    Experimental study of a dual reflux enriching pressure swing adsorption process for concentrating dilute feed streams

    Ind. Eng. Chem. Res.

    (2010)
  • Cited by (38)

    View all citing articles on Scopus
    View full text