ID
× COMMENTARYCOVER STORYIN THE NEWSNEWSFRONTSCHEMENTATOR + Show More
Chemical Engineering MagazineChementator Briefs
RE-free magnets Tetrataenite (L10-FeNi) is a promising alternative to rare-earth-…
BUSINESS NEWSTECHNICAL & PRACTICALFEATURE REPORTFACTS AT YOUR FINGERTIPSTECHNOLOGY PROFILEENGINEERING PRACTICEEQUIPMENT & SERVICESFOCUSNEW PRODUCTS + Show More SHOW PREVIEWS

Comment PDF Separation Processes

Experimental Validation of Distillation Column Simulations

By Glenn Graham, Pratik Pednekar and Don Bunning |

A practical look at the need for validation, as well as conceptual considerations and a case study

This article addresses the need for experimental validation of computer-generated distillation column simulations , and discusses the main concepts involved with experimental validation. Also included is a case study of a simulation effort that initially yielded some uncertainties. These uncertainties were readily resolved with batch laboratory testing.

 distillation column simulations

Difficulties

Potential problems with physical property models. Process simulation tools, such as Aspen Plus, CHEMCAD, and ProSim Plus, are indispensable for the design and optimization of separation schemes involving distillation columns. However, the accuracy of their predictions depends highly on the physical property models used in the simulation, and it is not unusual to obtain significantly different simulation results when using different physical property models. Some level of experimental testing is often necessary to validate the simulation results.

The physical property models consist of pure component properties, thermodynamic equations and component-interaction parameters used within the thermodynamic equations. Accurate values for the pure component properties and the interaction parameters, and the correct thermodynamic equation are seldom known at the beginning of a simulation effort.

Some software packages offer built-in pure-component-property libraries containing data for various components. In these libraries, the parameters — such as those in the Antoine equation — are regressed from built-in experimental data. The user must determine if the conditions of the simulation are close enough to the conditions of the source data to allow the data to be confidently utilized. If not, literature data may be available at the conditions of interest to allow regression and estimation of new property parameters.

There are numerous thermodynamic equations that can be used to predict the vapor-liquid equilibrium (VLE) or liquid-liquid equilibrium (LLE) between various components. The differences in the various thermodynamic equations may be subtle and the selection of the most appropriate equation may not be obvious. The thermodynamic equation selection is often based on rule-of-thumb techniques. Other times, the simulator default equation can be used, or the equation may be obtained from similar simulation work in the literature, which may not have been validated. The interaction parameters between the various components that go into the thermodynamic equation are often not available in the simulation software for all of the binary pairs of chemicals. These interaction parameters often must be obtained from literature or estimated using group contribution methods like UNIFAC (universal quasichemical functional group activity coefficients) or developed from new experimental testing.

Knowledgeable and experienced engineers can generally make good decisions in the selection of the thermodynamic equation and the determination of the interaction parameters when compiling the physical property models. Nevertheless, uncertainty in the reliability of the simulation results often remains. It is critical that these difficulties and uncertainties are properly considered in the development of the physical property model to avoid distillation columns that fail to accomplish the required separation, or columns that are overdesigned and incur unnecessary costs. Lack of information about the process chemicals (especially in the case of new chemicals) and their interactions lead to uncertainty in the reliability of the predictions. Experimental distillation testing is recommended, either to verify the quality of the simulation results, or to provide data to allow the physical property models to be calibrated to the subject process conditions.

Difficulties not captured by simulations. Unexpected column chemistry may also occur, generating troublesome byproducts. These byproducts may cause product-quality problems or build up in the separation system, negatively affecting the process in other ways. If the byproduct is an intermediate-boiling compound, it may concentrate in the middle of the column, and could significantly affect the known separations within the column. An example of this “bulge” effect is shown in Figure 1. The main feed to the column in this example is methanol and heptanol, which can be easily separated. If, however, a small amount of butanol, representing an intermediate-boiling byproduct, is fed to the reboiler, the butanol accumulates in the middle of the column and its concentration builds up to very high levels within the column. This accumulation of an intermediate-boiling species would cause significant changes to the separation within the column and would dramatically affect the temperature profile and the temperature sensor(s) used for control.

Figure 1.  Relatively small amounts of intermediate-boiling compounds can build up to large concentrations within a distillation column, interfering with an otherwise easy separation. Feed to stage 10 is 1.0 kg/h of 50/50 methanol-heptanol. Feed to reboiler (stage 20) is 1.0 g/h butanol. Pressure is 760 mm Hg and the reflux ratio is 2

Figure 1. Relatively small amounts of intermediate-boiling compounds can build up to large concentrations within a distillation column, interfering with an otherwise easy separation. Feed to stage 10 is 1.0 kg/h of 50/50 methanol-heptanol. Feed to reboiler (stage 20) is 1.0 g/h butanol. Pressure is 760 mm Hg and the reflux ratio is 2

Other problems that cannot be predicted by simulation include foaming and fouling. Foam generation can reduce the capacity of a distillation column. In severe cases, foaming will make the column completely non-operational. Fouling of contactor surfaces may reduce the column’s capacity and could affect the efficiency of the contactor.

The case for experimental validation. These uncertainties associated with the physical properties, as well as the potential for the other issues that simulations do not address, dictate that empirical testing of distillation simulations be considered (a case study demonstrating the need for experimental validation of simulation work is given later in the article). Experimental testing may only be required for part of the separation scheme. A considerable amount of judgement is necessary to decide when experimental testing is appropriate and what level of testing is needed. Often, especially for early work, adequate experiments can be conducted in a simple batch column. In other situations, a continuous column is needed. Later-stage work often requires equipment representing the entire process. A secondary decision is required regarding the scale of the equipment. The “scale” is generally thought of as laboratory-scale equipment or a pilot-scale facility, but “lab-scale” and “pilot-scale” can cover a wide range of sizes, which might overlap.

 

Concepts for experimental validation

In any attempt to compare experimental distillation results to simulation predictions, there are three main concepts that interact to influence the comparison. These are as follows:

  • VLE of the components (and possibly LLE)
  • Relative liquid-to-vapor flowrates (L/V ratios within the column)
  • Vapor-liquid contactor efficiency (tray efficiency or packing height equivalent to the theoretical plate, or HETP)

When a disagreement exists between experimental distillation data and the corresponding simulation results, it will likely be a difficult challenge to determine which one, or more, of these effects are responsible for the discrepancy. Some additional detail is justified to explain their interaction.

For the sake of simplicity, the following discussion will refer to a binary system of components and a conventional continuous column with an overhead distillate, a bottoms product, and a feed that is introduced somewhere within the column, as is shown in Figure 2. Although real systems are seldom binary, the relevant concepts are more readily explained using binary components. These same concepts will apply to multicomponent systems in a similar, but more complex form. Likewise, ideas presented using a conventional continuous column can be extrapolated to more complex columns with multiple feeds, sidestreams and so on.

Figure 2.  Shown here is a schematic diagram of a conventional distillation column with a feed, and distillation and bottoms products

Figure 2. Shown here is a schematic diagram of a conventional distillation column with a feed, and distillation and bottoms products

Vapor-liquid equilibrium. The relative volatility of an ideal binary chemical system is equal to the ratio of the pure component vapor pressures. If the components’ interactions are not far from ideal and the two components have sufficient relative volatility (>2), then small errors in the VLE predictions may not significantly affect the separation. However, if the relative volatility is small (<1.5), any deviations between predicted and actual VLE will lead to large differences between the actual and predicted number of stages required to accomplish the desired separation.

Non-ideal systems will often have a pinch at the upper or lower end of the binary VLE curve. The term “pinch” means that the vapor and liquid compositions of the light-boiling component are nearly equal (this concept also applies to the heavy-boiling compound). The pinch is usually at the upper (right-hand) end of the VLE curve, which is the case for binary pairs that repel each other. In this situation, the actual vapor pressure of the heavy-boiling component is nearly equal to that of the light-boiling component, due to the increased liquid-phase activity coefficient of the heavy-boiling component at infinite dilution. With sufficient non-ideality, the activity coefficient of the heavy-boiling component will become even greater, causing the heavy-boiling component’s actual vapor pressure to exceed that of the light-boiling component at infinite dilution of the heavy-boiling component. This is the case for minimum-boiling azeotropes, and at the azeotrope composition, the actual mixture vapor pressures (pure component vapor pressure times activity coefficient at the component’s liquid concentration) of the two components will be equal.

The closer the composition is to a pinch zone or azeotrope, the greater will be the effect of any errors in the prediction of the VLE on the separation. Figure 3 shows how a positive error in the VLE diagram can result in a significant underestimation of the number of stages required. Also, near a pinch or azeotrope, small changes in the reflux ratio can cause significant changes in the separation. A small error in the VLE prediction by the simulator could have a similar effect to a small change in the reflux ratio. It may be difficult to determine which is causing any discrepancies between a simulation prediction and the experimental results. If the components are so non-ideal as to cause phase separation within the column, then accurate LLE predictions will be necessary to produce accurate VLE predictions.

Figure 3.  A positive error in the predicted VLE curve of a pinched binary system can cause the required number of stages to be significantly underestimated (The graph at the right is an expanded view of graph at the left)

Figure 3. A positive error in the predicted VLE curve of a pinched binary system can cause the required number of stages to be significantly underestimated (The graph at the right is an expanded view of graph at the left)

Liquid/vapor ratios. The relative vapor-to-liquid flowrates within the column are not only determined by the feedrate, reflux ratio, boilup ratio, and distillate-to-bottoms ratio, but also by heat effects. Excessive heat loss from a high-temperature column with many stages can significantly increase the internal reflux. This heat loss reduces both the liquid and vapor flowrates as one progresses up the column. The L/V ratio will be the lowest at the top of the column, where it can be experimentally determined from the reflux and distillate flowrates. However, the L/V ratio will progressively increase as one moves down the column, due to condensation of the vapor. This effect can be thought of as an increase in the effective reflux ratio going down the column. This increase in L/V ratios due to heat loss will cause the column to have a better separation than predicted by a computer simulation, but at the expense of higher energy costs and lower capacity than expected (see Figure 4). Without understanding that the actual L/V ratios are different than expected, the VLE curve would appear to be more open (higher relative volatility or less pinched) than is actually the case. Preheated feeds and reflux from the condenser that have heat qualities different than expected can have similar effects on the column internal flows. Sub-cooled feed or reflux will cause condensation, which increases the L/V ratio below its point of entry into the column. Superheated or flashing feed may produce vapor flowrates different than anticipated, affecting the L/V ratio above the feed point.

Figure 4.  The left figure represents a column with no heat loss, whereas in the right figure, 50% of heat from reboiler is lost through the column walls. Laboratory columns with many trays at high temperature could easily have 50% heat loss or greater. The increased internal reflux improves separation, but also increases reboiler duty and reduces column capacity

Figure 4. The left figure represents a column with no heat loss, whereas in the right figure, 50% of heat from reboiler is lost through the column walls. Laboratory columns with many trays at high temperature could easily have 50% heat loss or greater. The increased internal reflux improves separation, but also increases reboiler duty and reduces column capacity

Efficiencies. Process simulators generally use theoretical stages to represent the trays in a distillation column. However, in real columns, the vapor leaving a tray is seldom in equilibrium with the liquid from that tray. The difference between the actual and equilibrium vapor compositions can be expressed as an efficiency. A common method of defining efficiencies for actual distillation trays is the Murphree vapor efficiency (see Figure 5). Values for Murphree vapor efficiency tend to be in the range of 50 to 75% for laboratory- and pilot-scale columns, but can lie outside of this range. The Murphree vapor efficiency is an indication of how closely the vapor composition in an actual distillation tray approaches equilibrium with the liquid leaving that tray, as shown in the left-hand diagram of Figure 5. For a binary chemical mixture, if the vapor composition for every actual tray were plotted against the liquid composition on that tray, it would appear as a pseudo-equilibrium curve, as shown in the right diagram of Figure 5. The McCabe-Thiele concept could be applied to this pseudo-equilibrium curve, in which actual trays would be drawn, instead of theoretical stages, as for a true equilibrium curve.

Figure 5.  The Murphree vapor efficiency represents how close the actual vapor composition comes to reaching equilibrium with the liquid on an actual distillation tray.  The concept of the non-attainment of equilibrium can be represented as a pseudo-equilibrium line on a McCabe-Thiele diagram, in which actual trays are drawn instead of theoretical trays [5]

Figure 5. The Murphree vapor efficiency represents how close the actual vapor composition comes to reaching equilibrium with the liquid on an actual distillation tray. The concept of the non-attainment of equilibrium can be represented as a pseudo-equilibrium line on a McCabe-Thiele diagram, in which actual trays are drawn instead of theoretical trays [5]

If the Murphree vapor efficiencies for the trays of a distillation column are 100%, the number of actual trays required to accomplish a separation would be equal to the number of theoretical trays required. When the Murphree vapor efficiencies are less than 100%, the required number of actual trays will be greater than this theoretical number. Therefore, if the actual tray efficiencies are less than predicted, more actual trays will be required to accomplish a desired separation than would be anticipated based on the required number of theoretical trays and the predicted efficiency. Without properly accounting for the reduced efficiency, the separation will appear to be more difficult than predicted.

The efficiency of packed columns is routinely expressed as HETP (height equivalent to a theoretical plate). If the actual HETP is greater than predicted (less efficient), the column will contain less theoretical stages than predicted and this will produce similar results to a column with trays that have reduced efficiencies.

The tray efficiency is dependent on several factors, including the configuration of the tray (hole size, number of holes, hole area) or packing (surface area, shape, void fraction), the properties of the chemical mixture, the liquid and vapor flowrates, the L/V ratios, and the pressure. Several authors provide methods for estimating tray efficiencies [1–4].

Effect of pressure on the separation. The previous discussion shows how the VLE, L/V ratio, and contactor efficiency are intertwined, and the importance of knowing the values of each of these for the experimental validation of distillation columns. Another variable that needs to be considered is pressure. Pressure not only affects the temperature, but can also change the VLE curve, as shown in Figure 6. Lower pressure often makes the VLE curve more open and the separation becomes easier at lower pressures.

Figure 6.  For the methanol-ethanol binary system, lower pressure opens up the VLE curve, which would make a distillation separation less difficult

Figure 6. For the methanol-ethanol binary system, lower pressure opens up the VLE curve, which would make a distillation separation less difficult

Trayed plant columns usually have significantly higher pressure drops than laboratory or pilot columns. Therefore, if a laboratory or pilot column is used to validate simulation results, and it is operated at the same head pressure as that intended for the plant column, the laboratory or pilot column will have a lower base pressure than the corresponding plant column. For columns operating in the range from moderate vacuum to above atmospheric pressure, the effect on temperature and separation may not be significant. However, for high-vacuum columns with many trays, a laboratory or pilot column may have a much lower base pressure than the commercial column and a correspondingly lower base temperature. Generally, this lower pressure in the experimental column allows easier separation properties than the corresponding plant column.

In addition to overestimating the separation capabilities of the plant column, the labboratory or pilot column may underestimate the amount of degradation that will occur in the base of the plant column, since the plant column’s base pressure and temperature will be higher than for the laboratory or pilot column. To overcome these problems, it may be necessary to operate the small-scale column at the same base pressure as intended for the plant column, or to run separate experiments, matching the head pressure in one test and the base pressure in another.

 

A case study

An example of how simulation discrepancies can be resolved using batch distillation experiments is presented here.

A client was interested in developing a method to continuously separate a mixture of ethylene glycol (EG), propylene glycol (PG) and 1,2 butanediol (1,2 BDO) in water. Distillation was one of the options being considered for the continuous commercial separation. The components had similar boiling points and the desired purity could not be achieved under modest positive pressures. From past experimental knowledge and experience, a subject-matter expert (SME) was aware of how this separation would take place under vacuum, and the simulation work progressed to consider vacuum separations. Initial simulation studies evaluated the separation of a known composition of the mixture using a combination of columns. Different physical property models, including NRTL (non-random two-liquid), UNIQUAC and Wilson, were tried based on the expected interaction of the components. The interaction parameters were estimated for the expected range of conditions for each model. However, the simulation results did not always demonstrate the expected separation. The results did not appear correct and often were very different from one physical property model to another.

A simple batch distillation experiment was set up to perform a validation experiment to confirm that the separation could be accomplished. The apparatus used was a 1-in. glass Oldershaw column with an evacuated and silvered jacket. Fiberglass insulation was added around the column and around the top of the glass kettle. The starting mixture of the chemicals was heated in the base and the column was operated in total reflux until it reached steady state. After that, the reflux ratio was changed and a total of ten distillate samples were collected.

A software package called Aspen Batch Modeler was used to simulate the unsteady-state batch distillation. The software simulates the batch distillation process dynamically, allowing the user to specify changing experimental conditions, including reflux ratios, distillate collection and so on. This software accesses the same Aspen properties library as Aspen Plus, which was being used to simulate the commercial distillation columns. Thus, once the optimum physical property model and parameters were found that allowed the batch simulation to match the batch experimental data, this model and the corresponding parameters could be used directly in Aspen Plus to simulate the commercial continuous case.

The mass of the components in the overhead distillate samples in the batch experiment were compared to the simulation. Because of the expected component interactions, activity coefficient models were used. Specifically, the NRTL, UNIQUAC and Wilson property models were individually evaluated. For each of these models, different parameter estimation techniques were tried. Binary parameters were calculated by regressing data from the literature. The Antoine’s parameters were calculated from literature data at the conditions of interest. Also, UNIFAC was used to estimate the binary parameters at the operating conditions. The best match to the experimental data was obtained using the Wilson property model. Figure 7 shows the close fit of the simulation results using the Wilson model and the contrasting poor fit when using the NRTL model.

Figure 7.  The Wilson property model provides a much better fit than the NRTL model. The above figures show the results for the mass of ethylene glycol (EG) and propylene glycol (PG) collected in the overhead sample in the experiment as compared to the simulation. The figures on the left compare the experimental data with the simulation when using the NRTL property model, while those on the right show a comparison when using the Wilson property model. The binary parameters were estimated using UNIFAC for both cases

Figure 7. The Wilson property model provides a much better fit than the NRTL model. The above figures show the results for the mass of ethylene glycol (EG) and propylene glycol (PG) collected in the overhead sample in the experiment as compared to the simulation. The figures on the left compare the experimental data with the simulation when using the NRTL property model, while those on the right show a comparison when using the Wilson property model. The binary parameters were estimated using UNIFAC for both cases

The results for the NRTL case showed only EG and water (not shown in the Figure 7) being recovered in the distillate. No PG recovery was seen in the simulation in contrast to the experimental data. The results of the Wilson case, however, compared much better with the experimental results. Using the Wilson property model and optimized parameters, the results for the mass fraction of EG, PG and 1,2 BDO are compared with the experimental data in Figure 8. The data show a very good fit. The quality of this fit demonstrates that the Wilson property model corresponding parameters are appropriate for the subject conditions, and that they can be used with confidence in the simulation of the continuous commercial columns.

Figure 8.  Mass fractions of the components in the distillate overhead samples are compared with the optimized batch column simulation results when the using the Wilson property model. A very good match was obtained using the Wilson property model

Figure 8. Mass fractions of the components in the distillate overhead samples are compared with the optimized batch column simulation results when the using the Wilson property model. A very good match was obtained using the Wilson property model

The commercial distillation column simulations in Aspen Plus were then re-developed utilizing the Wilson model and the data from the experimental batch distillation. The revised simulations showed distillation to be a feasible separation technology, including from a capital and operational cost perspective. Later, a client SME validated the distillation column simulation results based on experience in an existing plant with similar separations.

In this case study, the batch distillation experiment was useful not only to validate the feasibility of the separations, but it also provided data to calibrate the simulation model and to thereby add confidence to the commercial predictions.

 

Concluding remarks

The use of process simulation tools for modeling distillation columns is invaluable for designing plant-scale columns. However, it can be difficult to know if the simulator is generating accurate predictions. Additionally, there are potential problems associated with distillation columns that process simulators do not address. Hence, experimental distillation studies are quite important to either verify the simulation results or to provide a path forward to improve the simulations. In this article, a case study was presented of a simulation effort that yielded some uncertainties, which were then readily resolved with laboratory batch-column testing.

When contemplating the need to experimentally validate distillation simulations, key points include the following:

  • For process simulations, developing the proper physical property model by selecting the correct pure component properties, thermodynamic equations and component interaction parameters can be a challenging task, especially for moderate to severe non-ideal mixtures.
  • Without experimental validation, it is difficult to know if the simulations are accurate
  • Byproduct generation, foaming and fouling are difficult or impossible to predict by process simulators.
  • Experiments should be conducted to validate distillation simulations and to determine if other non-simulated problems will occur.
  • The main concepts that affect the separation in a distillation column are the VLE, L/V ratios, and contactor efficiencies (pressure also plays a role).
  • When a disagreement exists between experimental distillation data and the corresponding simulation results, it will likely be a difficult challenge to determine which one, or more, of these effects are responsible for the discrepancy. n

References

1. Stichlmair, J. G., and Fair, J. R., “Distillation: Principles and Practice,” John Wiley & Sons, Inc., Hoboken, N.J., 1998.

2. Kister, H. Z., “Distillation Design,” McGraw-Hill, Inc., New York, N.Y., 1992.

3. Treybal, R. E., “Mass-Transfer Operations,” McGraw-Hill, Inc., New York, N.Y., 1987.

4. Van Winkle, M., “Distillation,” McGraw-Hill, Inc., New York, N.Y., 1967.

5. Kister, H. Z., “Distillation Design,” McGraw-Hill, New York, N.Y.,1992.

Authors

Glenn Graham photoGlenn Graham is a distillation SME and senior chemical engineer at MATRIC (Mid-Atlantic Technology, Research, & Innovation Center, P.O. Box 8396, South Charleston, WV 25303; Phone: 304-552-6554; Email: glenn.graham@matricresearch.com; Website: www.matricinnovates.com). Prior to his position at MATRIC, Graham worked for Union Carbide Corp. and The Dow Chemical Co. as a distillation specialist in their R&D separations groups. Graham holds a M.S.Ch.E. from Montana State University.

Pratik Pednekar photoPratik Pednekar is a project manager and research chemical engineer at MATRIC (Email: pratik.pednekar@matricinnovates.com). He works in the areas of process conceptualization and development, laboratory experimentation, pilot plant testing and technology evaluation. Prior to working at MATRIC, Pednekar was awarded his Ph.D. from West Virginia University in the field of process development, involving simulation, optimization and reactor modeling.

Don Bunning photoDon Bunning is currently employed by MATRIC (Phone: 304-720-1049, Email: don.bunning@matricresearch.com). He has extensive experience in process development, reaction engineering, new catalyst development, demonstration of new technologies in pilot units, commercialization, on-site startup support, and licensing of the technologies. His more than 45 years of R&D experience in the chemical industry includes technical and management positions at Union Carbide and Dow Chemical. He hold a M.S.Ch.E. from the University of West Virginia and is a registered P.E. in West Virginia.

Related Content
A fluorine-free membrane for fuel cells
Polymer-electrolyte fuel cells (PEFCs) are promising devices for clean power generation in automotive, stationary and portable applications. Up to now,…
Multilayer membrane for climate control
A triple-layer, porous membrane system developed at the Functional Materials Laboratory at Eidgenössische Technische Hochschule Zürich (www.fml.ethz.ch) can be fashioned…

Chemical Engineering publishes FREE eletters that bring our original content to our readers in an easily accessible email format about once a week.
Subscribe Now
Trinseo Digitizes Control System Migration Projects to Achieve Fast ROI
Purdue University Saves $400,000 Annually with Local Vacuum Networks
Bag filter Housings/Vessels
Innovative Backwashable Media Filter
Automated Vertical Tower Filter Press

View More

Live chat by BoldChat