Separation Science and Technology

ISSN: 0149-6395 (Print) 1520-5754 (Online) Journal homepage: http://www.tandfonline.com/loi/lsst20

Efficiency based non-equilibrium modeling of

industrial-scale multicomponent distillation
columns

Hadi Poortalari, Javad Karimi Sabet & Farshad Varaminian

To cite this article: Hadi Poortalari, Javad Karimi Sabet & Farshad Varaminian (2017): Efficiency
based non-equilibrium modeling of industrial-scale multicomponent distillation columns, Separation
Science and Technology, DOI: 10.1080/01496395.2017.1301471

To link to this article: http://dx.doi.org/10.1080/01496395.2017.1301471

Accepted author version posted online: 20

Mar 2017.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=lsst20

Download by: [University of Newcastle, Australia] Date: 21 March 2017, At: 07:57
 Efficiency based non-equilibrium modeling of
industrial-scale multicomponent distillation columns
Hadi Poortalari1, Javad Karimi Sabet21, Farshad Varaminian1*

1-Faculty of Chemical & Petroleum and Gas Engineering, Semnan University, Semnan, Iran

t
ip
2-NFCRS, Nuclear Science and Technology Research Institute, Tehran, Iran

Abstract

cr
An efficiency-based non-equilibrium model is proposed to simulate two industrial distillation

us
columns. In this model, the behavior of trays is studied by considering the mass transfer at
an
interphase. The interphase mass transfer is simply evaluated by the simultaneous solution of the

vapor material balance equation and the definition of the Murphree efficiency. A MATLAB code
M
is developed to implement the simulation procedure. For verification, the product compositions

and plate temperatures are compared with reported experimental data. The discrepancy of the
ed

experimental and simulation results is about 0.5% and 20% for the temperature and the products

compositions for both columns, respectively.

pt

Keywords: Multicomponent distillation, Non-equilibrium modeling, Interphase mass transfer,

ce

Murphree efficiency,
Ac

1
Correspond authors: [email protected],
[email protected]

1
1- Introduction

In recent years, distillation process is highly considered in various industrial applications due to

its various advantages. Distillation is a practiced technique for the separation of the chemical

t
ip
species mixtures in the refineries and chemical industries. This process is a cost effective method

for continuous separation of different mixtures in high volumes to obtain high purity products.

cr
These advantages cause to develop the application of the distillation process in many industrial

us
fields, and therefore researchers have tried to increase the performance of this process. Due to

high energy usage and operational difficulties of distillation columns, experimental

an
investigations are significantly expensive and time consuming. Hence, various researchers have

focused on the analytical and numerical approaches to model and simulate this process. Since
M

control, management and operation of the distillation columns in the industrial scale are

accompanied with various complexities, computer programs that describe this operation should
ed

be available throughout the unit operation.

pt

In order to simulate this process, the detailed knowledge of the process theory is paramount

importance for design purposes. The mass transfer between vapor-liquid is the main phenomena
ce

that occurs during the separation inside the distillation column. Basically, two approaches are

used to model vapor-liquid mass transfer on a distillation stage: the equilibrium stage approach
Ac

and the non-equilibrium stage approach. The conventional method to model a distillation column

is the so called equilibrium or ideal stage model. The ideal stage model is easy to use because a

detailed equipment design is not required. Indeed, some limited data including only equilibrium

relationships and enthalpy data are required in this model. It is conceptually simple and

2
mathematically well-designed. Hence, it has been used to simulate and design various columns

in real world. More than a century, chemical engineers have modeled distillation columns by the

equilibrium stage approach [1]. In the equilibrium stages, the vapor and liquid phases are

assumed to mix perfectly and reach complete interaction with each other, so that the equilibrium

t
composition is obtained as streams leave the stage. The equilibrium assumption also means

ip
streams leaving a stage are at the similar temperature that is reasonable for many industrial

cr
columns.

us
Although equilibrium model has several advantages, stages rarely operate at equilibrium state in

actual operation despite attempts to approach this state by proper design and operating conditions
an
tuning. In fact, the behavior of a real tray is not ideal due to some reasons such as incomplete

contact between the phases, not enough residence time, vapor and liquid entrainment, weeping,
M

and poor vapor and liquid distribution [2]. In the design and operating of distillation columns,

deviation from equilibrium state on the stages have important effects. This is often large enough,
ed

so that it might reduce the effectiveness of the separation unit to 50-70% of ideal operation [3].
pt

Therefore, the designers have always interested in modelling of the trays with real conditions.
ce

In general, two common approaches are proposed to explain the non-equilibrium nature of

distillation stages and overcome the gap between the model and reality: the stage efficiency
Ac

method and the rate-based method. In the first approach, the concept of efficiency is used to link

the performance of practical trays to the ideal equilibrium stages. Indeed, the efficiency is

recognized as an important factor in distillation column design. For dealing with deviation from

the equilibrium condition, stage efficiencies incorporate into the equilibrium relations. The most

3
common definition for stage efficiency is the vapor phase Murphree efficiency as it is simply

jointed with the equilibrium relations. The Murphree tray efficiency is defined as the ratio of

vapor concentration change for the actual conditions to equilibrium conditions. On the second

approach, the interphase mass transfer method known as the rate-based method is used to

t
describe the deviation from equilibrium conditions. Since separation process which is actually

ip
reached in the real plates mainly depends on the rates of mass transfer from the vapor phase to

cr
the liquid phase, the actual mass transfer at plates is directly calculated in the rate-based

us
approach. The rates of mass transfer depend on the non-ideality extent of the vapor and liquid

streams with each other. This method gives more detailed temperature and composition profiles
an
at the expense of heavier computational cost and more equipment design details.

In recent years, different researchers motivated to apply efficiency calculations along with
M

equilibrium modeling approaches to consider the non-ideal nature of distillation process.

ed

The efficiencies are conventionally obtained from experiment or empirical methods and usually

just a same fixed value (overall efficiency) is considered for the whole column. However, it has
pt

been experimentally revealed that the plate or even component efficiencies might be different
ce

with each other in multicomponent mixtures [4, 5]. Thus, the concept of overall column

efficiency cannot result in an accurate simulation and a more rigorous definition should consider
Ac

for individual efficiencies that can differ along the column. Consequently, the prediction of the

component or stage efficiencies during the simulation improves the column design by avoiding

unnecessary over-sizing, and so savings are obtained in the capital and operation costs.

4
Frequently, these calculations are based on multicomponent mass transfer models and the

estimation of binary NTUs (Number of Transfer Units). Aittamaa [7] initially applied the

multicomponent plate efficiencies in the modelling of distillation column based on equilibrium

stage approach. Ilme [8] later used this algorithm and developed it further. The methods of

t
distillation process modeling according to efficiency concept have been extensively examined by

ip
Klemola [9], Jakobsson [10], and also applied for industrial columns by Ilme et al. [11] and

cr
Jakobsson et al. [12].

us
Although various aspects of these methods are comprehensively studied, the concept of the

efficiency have not been applied in the non-equilibrium modelling approaches. In fact, the
an
researchers often focused on modifying the equilibrium algorithms using the efficiency concept

to consider the actual condition of the separation units. Furthermore, researchers frequently have
M

been applied multicomponent mass transfer models to predict individual component efficiencies

at any plate. The rigorous methods for the multicomponent efficiencies calculation are
ed

accompanied with many complexities that lead to convergence problems in calculations and
pt

consequently the difficulty of applications. Therefore, it would be very desirable to simulate a

multicomponent distillation process based on pseudo-binary efficiency calculations without

ce

significant reduction in the accuracy of results. According to this, the aim of this study is to

present a rigorously modified method for the modeling of industrial-scale distillation columns by
Ac

using pseudo-binary efficiency calculations into a simple non-equilibrium approach. In this

model, the interphase mass transfer relations are applied instead using of the phase equilibrium

relations. The interphase mass transfer coefficient is determined simply from the simultaneous

solution of the vapor phase material balance and the definition of the Murphree efficiency. The

5
main characteristics of this model is to combine the advantages of the both efficiency based

equilibrium model with interphase mass transfer non-equilibrium model to obtain the efficient

and applicable approach for the modeling of distillation process. In the proposed model, the

behavior of real plates is initially estimated in form of stage efficiencies, and then column

t
performance for separation of key components is obtained. Finally, the estimated separation is

ip
compared with actual measurements of the industrial distillation columns. According to this

cr
procedure, the column design can be analyzed, and it would be modified if needed. The

us
calculated results include the concentration and temperature profiles along the column and

finally product compositions are compared against reported industrial data.

an
2- Model description
M

2-1- Interphase mass transfer

ed

In our model, the interphase mass transfer approach is used to describe the deviation from

equilibrium stage model in two industrial scale distillation columns, a C4 splitter and a column
pt

for separating benzene, toluene, and xylene (BTX). A specific equation is required to calculate
ce

the interphase mass transfer coefficient. This relation obtained from the simultaneous solution of

the vapor phase material balance equation and the definition of the Murphree efficiency factor
Ac

( ) as follow [3]:

= (1)
1−

6
where represents the vapor mass transfer coefficient, is the effective area for mass transfer,

and is vapor flow rate, all for stage j. In derivation of this expression, it is assumed that the

bulk vapor rate does not change within the liquid and/or froth on the stage.

t
To consider the non-equilibrium behavior of real distillation trays, it is necessary to include the

ip
interface mass transfer effect in the component-material balances written for any stage in

cr
distillation column. Mass transfer on a two-phase contact stage may be described by two material

balance equations, one written for the liquid phase and one for the vapor phase. The balance

us
equation for any component in the liquid phase and the vapor phase are stated as:
an
∗
, − , − , − , + , =0 (2)
M

∗ (3)
, − , + , − , + , =0
ed

∗
where , = , , . , represents the K-value of component i at stage j.
pt

The term related to feed, the last term, should be considered in both expressions, since feed may
ce

enter either phase. These expressions for the phase material balance are derived based on the

relation of the mass transfer driving force and equilibrium state. According to this, the rate of
Ac

mass transfer is decreased as conditions proceed to equilibrium. Figure 1 schematically depicts

the flow pattern of stages according to the interface mass transfer effect.

7
In distillation modeling, the condenser and the reboiler are always considered as equilibrium

stages even though equilibrium is not obtained on the trays within the column. So, the balance

equations for condenser as first stage and reboiler as last stage are as follows [13]:

t
condenser (stage 1) : − + =−

ip
, , , ,

(4)

cr
reboiler (stage N) : , − + , =− ,
,

us
2-2- Efficiency model an
The Murphree vapor phase tray efficiency [14] is the most popular efficiency definition, and it is

the only one with practical aspects [15, 16]. It is defined as the ratio of the actual change in a
M

component vapor mole fraction to the change in mole fraction of the vapor while it is in

equilibrium with liquid at the tray output. For each tray, the Murphree vapor efficiency is defined
ed

by:
pt

–
= ∗ (5)
−
ce

∗
Ac

where represents the vapor average composition above the froth and corresponds to the

vapor composition in equilibrium with the liquid composition, , leaving the tray. Since the

compositions often vary across the interface, the point efficiency is defined at any specified

position on the tray in terms of compositions. The Murphree vapor point efficiency can be

calculated as:

8
 –
= ∗ (6)
−

where represents the point composition in the vapor phase.

t
ip
Point efficiency describes the degree of tendency to equilibrium condition at any point on the

tray. This factor is not constant on the tray because there are concentration gradients in the liquid

cr
flowing on the tray. The tray efficiency is more useful factor than the point efficiency for the

us
design purpose.
an
The Murphree point efficiency is often calculated from a binary data correlation based on type of

the tray. This correlation provides the number of liquid and vapor phase transfer units (NTUs).
M
Then, the number of overall vapor phase transfer units is calculated from the number of liquid

and vapor NTUs as below [17]:

ed

1 1
= + , = (7)
pt
ce

This is used with:

Ac

= 1 − exp(− ) (8)

where is the slope of the equilibrium line and and are the molar vapor and liquid flow

rates through the tray, respectively.

9
The Murphree plate efficiency is determined based on the point efficiency when the degree of

phase mixing on the plates has been determined. For trayed columns, the concentration varies

across the tray as a result of non-ideal mixing. If the liquid on the tray is perfectly mixed, the

concentration is constant across the tray and consequently the Murphree tray efficiency is equal

t
to the point efficiency. However, liquid is never perfectly mixed on a real tray and concentration

ip
gradients exist. If the concentration gradient is maximized, it leads to the extreme limit of plug

cr
flow across the tray and the difference between tray and point efficiencies will be maximized.

us
The degree of liquid mixing is characterized by the Peclet number known as the ratio of liquid

transport and longitudinal liquid mixing and is defined as follow [11]:

an
= (9)
M

where is the liquid flow path length, is the liquid phase residence time and is the eddy
ed

diffusion coefficient. A small Peclet number corresponds to a high degree of mixing and vice
pt

versa.
ce

Welch et al. [18] presented a correlation for estimation of the eddy diffusion coefficient as

follow:
Ac

= 0.088 (10)
ℎ

10
where is the volumetric liquid flow rate, is the liquid flow path width and ℎ is the clear

liquid height on the plate.

In conventional industrial scale columns, the liquid on the plate is partially mixed while the

vapor is often assumed to be perfectly mixed. For modeling the liquid mixing on the tray,

t
ip
Gautreaux and O'Connell [19] divided a tray into n perfectly mixed liquid pools and proposed

cr
the following relation between the tray and point efficiencies:

us
1+ −1
(11)
=
an
where is the number of perfectly mixed pools that can be determined from the equation given
M

by Ashley and Haselden [20]:

ed

+2
= (12)
2
pt

The advantage of the mixed pool model is that the Murphree plate efficiencies are calculated in
ce

the whole liquid mixing range. The number of perfectly mixed pools depends on the liquid flow
Ac

path on the tray.

2-3- Efficiency prediction in the multicomponent mixtures

In a binary mixture, the component efficiencies are equal, but multicomponent mixtures do not

have similar component efficiencies. However, individual component efficiencies are not often

11
 used in design practice because of the complexity of rigorous methods for the multicomponent

efficiencies calculation, the difficulty of applying for design purposes, and unknown reliabilities.

The most common and the simplest procedure used for multicomponent efficiency calculations is

pseudo binary method, and it proceeds in the following steps [21]:

t
ip
1- The selection of light and heavy key components and calculation of the pseudo mixture

cr
composition as:

us
=
+
an 3)

=
+
M

where the subscripts and refer to light and heavy key, respectively.
ed

2- The prediction of the key components binary diffusion coefficients in each phase at the
pt

mixture temperature and pressure.

ce

3- The determination of the equilibrium curve slope:

Ac

, − ,
= 4)
, − ,

4- Using the binary correlations to predict the Murphree point and plate efficiencies.

12
5- Assuming the same Murphree efficiency for all components at the any stage and apply

this value in the simulation procedure.

It is worthy to note that some judgments are required to select the pseudo keys, and two

components selected can be different for various parts of a column. Naturally the light and heavy

t
ip
key approach is more appropriate for distillation columns in which two main components are

cr
present in the feed. However, this method can be extended to allow for multi-pair efficiencies

that may be different. The choice of binary pairs depends on feed and product compositions,

us
volatility differences, components of major interest for design, and the components which are the

majority fraction of the mixture [21].

an
2-4- Efficiency prediction in the binary mixtures
M

Several methods have been proposed for the prediction of the efficiencies in a binary mixture. In
ed

the present study, two well-known methods, AIChE (1958) [22] and Chan-Fair (1984) [23,24],

are used.
pt

Based on the AIChE correlation, the vapor phase number of transfer units on both bubble caps
ce

and sieve trays is:

Ac

(0.776 + 4.57 ℎ − 0.238 + 104.8 / )

= (15)

where = is the superficial f-factor and = is the vapor-phase Schmidt

number.

13
Chan and Fair developed an alternative correlation for as follow:

10,300 − 8670
= (16)
ℎ

t
ip
where is the fractional approach to flooding defined as the ratio of the superficial velocity ( )

cr
to the correspond value under flooding conditions ( ).

us
Both AIChE and Chan-Fair methods proposed a similar correlation for the liquid-phase number

of transfer units of sieve trays only as follow:

an
= 19,700 (0.4 + 0.17) (17)
M

2
In above equations and are Fick diffusivities ( / ), ℎ is the exit weir height ( ), is
ed

3
the weir length ( ), and is the volumetric liquid flow rate ( / ). and are the vapor

phase and the liquid phase residence time, respectively that are given by:
pt
ce

(1 − )ℎ
= (18)

Ac

ℎ (19)
=

The clear liquid height, ℎ is calculated by the correlation of Bennett et al. [24]:

14
 .
ℎ = ℎ +

.
(20)
= −12.55

t
−

ip
cr
= 0.50 + 0.438 (−137.8 ℎ )

us
3- Simulation procedure an
A distillation column is considered as a set of N separation stages and includes a condenser at the

top as stage 1 and a reboiler at the bottom as stage N. A schematic view of a single separation
M

stage is shown in Figure 2 where vi,j , li,j and fi,j are component flow rates of vapor, liquid and

feed streams, respectively, and Sj and sj are dimensionless side-stream flow rates of vapor and
ed

liquid streams, respectively.

pt

The modified Newton-Raphson method [26] was used to solve the equations of present model.

This approach has been widely applied to solve the equations of both the equilibrium and non-
ce

equilibrium models. Based on Naphtali and Sandholm [27] method, whole functions consist of
Ac

material balances for each phases in addition to energy balance are categorized according to

stage locations. Details of this method are presented in the appendix A.

According to Newton-Raphson method, the set of governing equations for each stage is as

follows:

15
 : 1 + , + 1+ , − , −

(21)

, − , + = 0 (1 equation)

t
ip
, ,
, : 1 + , − , + , − − , = 0 (c equations) (22)

cr
∑ , ∑ ,

us
, ,
, : 1 + , − , − , − − , = 0 (c equations) (23)
∑ , ∑ ,
an
The entire model equations, N (2c+1), are solved simultaneously based on the modified
M
Newton's method using an in-built program in MATLAB. The framework of the developed code

is represented in details in Figure 3. The simulation is carried out with non-equilibrium stage
ed

model according to interphase mass transfer method, but the reboiler and the condenser are

described as equilibrium stages. The sequence of simulation steps is depicted as a flowchart in

pt

Figure 4. The simulation model is specified by defining various specifications such as the
ce

thermodynamics of the system, the feed rate, condition and composition of the feed, the column

and plate geometry. Since the characteristics of the components are similar for both phases, the
Ac

well-known Soave Redlich Kwong (SRK) equation of state [28] is used as the VLE-method. The

simulation procedure needs to estimation of several physical properties at any calculation

iteration. The methods used for estimation of necessary properties are presented in Table 1.

16
The numerical algorithm consists of two computational loops: outer loop and inner loop. In each

iteration, a set of values for the interphase mass transfer coefficients ( ) is initially calculated

from the estimated tray efficiencies during outer loop, and then these values are used to

determine the compositions and temperature profiles based on Newton-Raphson method through

t
ip
inner loop. The point efficiencies in each computation step are calculated using the NTU

correlations of AIChE and Chan-Fair, and these values are then related to the Murphree tray

cr
efficiencies using the mixed pool model of Gautreaux and O'Connell. The calculation sequence

us
for the prediction of each tray efficiency has been shown in Figure 5.

4- Columns specifications
an
To evaluate and validate performed simulation, the results of two industrial-scale distillation
M

columns are used. The first column is a C4 splitter that was introduced by Ilme et al. [11, 12]. In

this column, i-butane and n-butane are separated from a liquid feed containing about 29.4 w% i-
ed

butane, 67.7 w% n-butane, 1.5 w% propane and 1.0 w% pentanes. In the mixture, i-butane and n-

butane are considered as light and heavy key components, respectively. The second column
pt

(BTX column) introduced by Biddulph [29] separates benzene from a liquid feed containing
ce

approximately 37.2 mol% benzene, 57.3 mol% toluene, and 5.5 mol% C8 aromatics. The C8

aromatics will be lumped together and treated as m-xylene. In this column benzene and toluene
Ac

are chosen as light and heavy key components, respectively.

A simplified flow sheet of these two columns are shown in Figure 6. As mentioned, in each

columns, samples were taken from the feed, top and bottom products when the operation is

17
steady state. At the same time, operating data including flow rates, temperature profiles and

pressure profiles was gathered from the plants instrumentation through automation systems. Due

to presence of inaccuracies in instruments measurement and analysis, the material balance are

not often satisfied in plant data. Hence, it is needed to reconcile errors in measurements and the

t
compositions should be adjusted to satisfy material balance constraints. In Figure 6, the

ip
reconciled values for compositions and flow rates are presented. The adjusted performance data

cr
for each columns are given in Table 2. The C4 column diameter is 2.9 m and its working pressure

us
is about 650 kPa. It has 74 Glitsch Ballast two-pass type V-l valve trays. The BTX column

consists of 54 plates with 2.43 m in diameter, having Koch type Kx-9-u valves. The column has
an
a sidestream taken from plate 5 which provides the pure benzene product. Details of the columns

specifications are given in Table 3.

M

5- Results and discussion

ed

To assess the quality of the separation, a pseudo-binary McCabe-Thiele diagram for considered

multi-component system at C4 splitter is presented in Figure 7. The data for equilibrium and
pt

operating curves are obtained from the results of the converged simulation. In the Figure, the
ce

axes are defined by the relative mole fractions of light key component correspond to equation 13.

As it is noticed, the staircase of visible triangles is extended to diagram corners. This fact shows
Ac

that the separation of key components has well done. In order to clearly inspect the variation, a

portion of the McCabe-Thiele diagram is enlarged to illustrate the stages extended beyond the

equilibrium curve. This means that the efficiencies are greater than 100%. However, this

problem is not unusual. In the multicomponent mixtures, the component diffusion rates are

18
affected by the concentration of all species presented in the mixture. According to Kister [21],

this leads to unbound efficiencies in the range [-∞, +∞].

The predicted values for the number of transfer units of each tray for both phases are given in

Figure 8. Figure 8a shows the results of two NTU calculation methods (AIChE and Chan-Fair)

t
ip
for C4 splitter. It can be seen that the number of liquid transfer units (NTUL) in the stripping

cr
section is almost constant and equals to 3.5 while values of NTUL in the enriching section

oscillate within value of 4. Also, it can be observed that the number of vapor transfer units

us
(NTUV) across the column obtained by Chan-Fair method is nearly more uniform rather than

AIChE method. The average value for NTUV according to Chan-Fair method is 2.8 while AIChE
an
method predicts a mean value of 2.6 and 3 for the enriching section and the stripping section of

column, respectively. Figure 8b displays the variation of NTUs for each phase based on AIChE
M

method regarding to BTX column. Alike C4 splitter, the values of NTUV value in enriching

section are lower than stripping section while NTUL values in enriching section is higher than
ed

stripping section of column. It is found that NTUV value is approximately 1.6 and 2.7 in
pt

enriching and stripping sections, respectively. However, the value of NTUL is around 3.4 and 2.9

in enriching and stripping sections, respectively. So it can be concluded that in both columns the
ce

liquid mass transfer in top section is more than bottom section while this trend for vapor mass

transfer is vice versa.

Ac

Figure 9 presents the estimated vapor phase Murphree point and tray efficiencies. The tray

efficiencies were achieved from point efficiencies using the partial liquid mixing method. As

shown in Figure 9a, the point efficiencies do not vary significantly from plate to plate along the

19
C4 splitter column, so that the mean value for point efficiencies is equal to 80%. Alike the point

efficiencies, the calculated tray efficiencies also do not change considerably along the C4 splitter

column, and the average value is equal to 110%. On the other side, Figure 9b shows that in BTX

column both efficiencies are almost unchanged in stripping section, with 66% for point

t
efficiency and 101% for tray efficiency, while efficiency values oscillate in the enriching section.

ip
It is clearly observed that the Murphree tray efficiencies considerably differ from point

cr
efficiencies in both columns. Since the liquid mixing on trays is incomplete, the tray efficiencies

us
appreciably deviate from the point efficiencies. This will enhance the tray efficiency compared to

point efficiency and consequently improve the separation.

an
Figure 10 displays the liquid composition profiles in C4 splitter, resulted in AIChE method. The

profiles of each eight components presented separately in Figure 10a where the mole fractions
M

are shown on a logarithmic axis so that all composition profiles can simply be seen. It is found

that each component exhibits a different composition profile curve. It is obvious that the
ed

composition of other components in comparison with key components composition is negligible,

pt

so it can be concluded that this process is essentially a binary separation and the separation solely

occurs between i-butane (LK) and n-butane (HK). Figure 10a also shows that i-butane (LK) mole
ce

fraction decrease smoothly and continuously from the top to the bottom of the column except at

the feed zone and top end of the column while the trend is contrary for n-butane (HK). i-Butane
Ac

is more concentrated below the feed stage while the main separation of n-butane occurs above

the feed stage. This trend has more precisely plotted in Figure 9b, where the liquid mole fractions

are presented in the normal scale. The results for both mass transfer models are also plotted in

20
this Figure. It can be seen that both AIChE and Chan-Fair methods predict approximately the

same profiles along the C4 splitter column.

The liquid composition profiles along the BTX column is shown in Figure 11. It is clear that the

column has been able to completely separate benzene at stage 7. As it is mentioned in the

t
ip
reference work, the function of further stages above this stage is the removal of some small

cr
quantities of non-aromatics components. In contrast, toluene is concentrated at bottom section of

column. Since the main goal is not the complete separation of toluene, the concentration of this

us
component reaches to about 90% and the rest is xylene in bottom product.
an
A comparison between simulated results obtained by both mass transfer methods and the

reconciled experimental data related to C4 splitter column is provided in Figure 12. This Figure
M
contains the results of distillate and bottom product compositions. The agreement between the

predicted column separation based on the efficiency calculation method and the reconciled
ed

compositions is fairly good.

The quantitative comparison of products composition between simulation results and correspond
pt

experimental data for both C4 splitter and BTX column are presented in Table 4 and Table 5,
ce

respectively. Since the results of two efficiency calculation methods are almost similar,

compositions resulted from AIChE method are only reported. The comparison of simulated and
Ac

reconciled experimental compositions of the products shows that the presented model fairly

predicts the compositions of all components. The comparisons also illustrates that in C4 splitter

the maximum deviation of the top product is occurred in the nButane while the iButene

composition presents the highest discrepancy in the bottom product. On the other hand, in BTX

21
column all compositions of the products except composition of toluene at top product are

predicted accurately by our model. In the both columns, deviations in the components

compositions of products are so small that it can be confidently said that the accuracy of

predicted values is acceptable for design purposes.

t
ip
The predicted temperature profiles along the columns is compared with measured values at some

cr
trays in Figure 13. As it is observed, the maximum deviations between predicted and measured

temperatures occurs on the plate 9 (0.3%) for C4 splitter and the plate 16 (0.5%) for BTX

us
column. As expected, the temperature decreases from the bottom to the top of the column. A

dramatic decrease in temperature occurs from bottom (reboiler) to top (condenser) in both
an
columns. This is caused by the subcooled condenser of columns, so that the reflux stream is

returned to the column with temperature lower than its bubble point.
M

The experimental reboiler heat duty for C4 splitter column is 10.24MW, while the simulation
ed

gives the value of 10.27MW. The difference of duties is about 0.3% which sounds reasonable,

because our simulation model does not consider heat losses of the real column.
pt

Finally it worthy to notice that in this work we tried to simulate distillation columns based on
ce

estimated stage efficiency profiles with a proposed non-equilibrium approach. These information

about stage efficiencies can be applied in efficiency-based modified equilibrium approach.

Ac

According to this, distillation columns simulated once more and calculations repeated. The

obtained results in this condition did not show noticeable difference, so that both efficiency

modeling approaches led to almost same results. This means that the accurate estimation of stage

efficiencies has central importance in distillation modeling procedure.

22
6- Conclusion

In this study, a rigorous efficiency based non-equilibrium model is presented to investigate and

model two industrial-scale multicomponent fractionators. This approach conducts a more

t
ip
realistic distillation simulation without imposing a large computation cost. This model is based

on the interphase mass transfer evaluation at each separation stage. In present model, the mass

cr
transfer at liquid-vapor interphase was considered simply by the simultaneous solution of the

us
vapor phase material balance equation and the definition of the Murphree efficiency. In order to

estimate the stage efficiencies, two popular methods (AIChE and Chan-Fair methods) were used.
an
Estimated efficiencies were applied to predict the mass transfer coefficient in each tray. A

MATLAB home-code was used to implement the sequence of efficiency based non-equilibrium
M

model for obtaining the profile of compositions and temperature along the columns. Our results

show that the average point efficiency for trays is about 80% for C4 splitter and about 66% for
ed

BTX column, while the mean value of the plate efficiency is almost 110% for C4 splitter and

almost 94% for BTX column. The results for the product composition and temperature profiles
pt

along the columns were also compared with the reported experimental data. The simulation
ce

results gave an excellent agreement (deviation less than 1%) with the measured values for

temperature profile. Furthermore, comparisons display a fairly consistence (average deviation

Ac

about 20%) between results for products composition. According to these results, it can be

concluded that the accuracy of predicted values form the proposed efficiency based non-

equilibrium model is acceptable for the accurate modeling of industrial-scale distillation

columns.

23
Appendix A. Newton-Raphson method

Equations (24)-(26) are solved simultaneously by the Newton-Raphson iterative method in

which successive sets of the output variables are produced until the values of the ML, MV, and H

t
ip
functions are driven to within some tolerance of zero. During the iterations, nonzero values of

the functions are called discrepancies or errors. The functions and output variables be grouped by

cr
stage in order from top to bottom as:

us
= , ,…, ,…, (A.1)
an
and
M

T
= , ,…, ,…, (A.2)
ed

where X, is the vector of output variables for stage j arranged in the order:
pt

= , ,…, ,…, , , , ,…, ,…, (A.3)

ce

, , , , , , , ,
Ac

and Fj is the vector of functions for stage j arranged in the order:

= , , , , ,…, , ,…, , , , , , ,…, , ,…, , (A.4)

24
The Newton-Raphson iteration is performed by solving for the corrections ΔX to the output

variables, which in matrix form becomes:

( )
( )
=− ( ) (A.5)

t
ip
cr
These corrections are used to compute the next approximation to the set of output variables from:

us
( ) ( ) ( )
= +
an (A.6)

where t is a nonnegative scalar step factor. At each iteration, a single value of t is applied to all
M
output variables. For each iteration, an optimal value of t is sought to minimize the sum of the

∂F
squares. The quantity is the Jacobian matrix that is a (N×N) matrix of blocks of partial
∂X
ed

derivatives of all functions with respect to all output variables as:

pt

0 0 .. 0
0 .. 0
..
ce

0 0
∂F
..
∂X
= . . . .

(A.7)
..
.. ..
0 .. 0
Ac

0 0
0 ..

Jacobian has a block tridiagonal form because functions for stage j are only dependent on output

variables for stages j-1, j, and j+1. Each A </ >, B </ >, or C </ > block represents a

25
(2C+1) by (2C+1) submatrix of partial derivatives, where the arrangements of output variables

and functions are given by (A.3) and (A.4), respectively. Blocks Aj </ >, Bj </ >, and

Cj </ > correspond to submatrices of partial derivatives of the functions on stage j with respect

to the output variables on stages j-1, j, and j+1, respectively. Thus, using (24), (25), and (26) and

t
ip
denoting only the nonzero partial derivatives by +, we find that the blocks have the following

form:

cr
us
< > Aj
+ + +
1
=
an
(A.8)
⋮ ⋱
= 1
M

⋮
ed

=−
pt

=− − , = 1:
, ,
ce

,
=− , = 1:
Ac

26
 < > Bj
+ +
= = + +
(A.9)
⋮ ⋱ ⋱
+ +

t
+ +

ip
⋮ ⋮ ⋮
+ +

cr
= 1+ + , = 1:

us
, ,

= 1+
an
+ 1+
M
= 1+ + , = 1:
, ,
ed

, , , 1 ,
= − + , = 1:
, ,
pt

, , ,
=
ce

, , , , , ,
= 1+ + + − , = 1:
Ac

, ,

, , , 1 ,
= 1+ − − + , = 1:
, ,

27
 , , ,
=

, , , , , ,
=− + − , = 1:
, ,

t
ip
cr
us
an
M
ed
pt
ce
Ac

28
 t
ip
< > Cj

cr
+ + +
= 1

us
(A.10)
⋮ ⋱
= 1 an
⋮
M

=− − , = 1:
, ,
ed

=−
pt

,
=− , = 1:
ce

,
Ac

29
Nomenclature

vapor flow rate, mol/s

t
ip
, component vapor flow rate, mol/s

cr
liquid flow rate, mol/s

, component liquid flow rate, mol/s

us
an
feed flow rate, mol/s
M

component feed flow rate, mol/s

ed

vapor side stream flow rate, mol/s

pt
ce

vapor dimensionless side-stream flow rate

Ac

liquid side stream flow rate, mol/s

liquid dimensionless side-stream flow rate

30
, component vapor composition

, component liquid composition

t
ip
, component feed composition

cr

enthalpy, J/mol

us
heat load, J/mol
an
, equilibrium constant
M

mass transfer coefficient

ed

effective mass transfer area, m2 /m

pt
ce

Murphree vapor phase tray efficiency

Ac

point efficiency for a tray

number of vapor phase transfer units

31
 number of liquid phase transfer units

number of overall vapor phase transfer units

t
ip
slope of equilibrium curve

cr
stripping factor

us
dimensionless Peclet number
an
dimensionless Schmidt number
M

eddy diffusivity, m2 /s
ed

liquid flow rate, m3 /s

pt

residence time on the tray, s

ce

liquid flow pass length, m

Ac

number of perfectly mixed pools

32
 outlet weir length, m

ℎ exit weir height, m

t
ip
ℎ clear liquid height, m

cr
superficial f-factor

us
superficial velocity, m/s
an
superficial velocity under flooding conditions, m/s
M

Fractional approach to flooding

ed

liquid composition of a light key component in a pseudo-binary mixture

pt

vapor composition of a light key component in a pseudo-binary mixture

ce

Greek Letters
Ac

Density, kg/m3

33
 Viscosity, Pa.s

Subscripts

t
ip
liquid phase

cr
vapor phase

us
light key component
an
heavy key component
M
ed

References
pt

R. Krishna, R. Kooijman, H.A. Taylor (2003). Modeling of distillation. Chem. Eng. Prog., 99,

28.
ce

2. F.M. Khoury (2005). Multistage separation processes. New York: CRC Press.
Ac

3. R.C. Waggoner, G.D. Loud (1977). Algorithms for the solution of material balance

equations for non-conventional multi-stage operations. Computers and Chemical Engineering, 1,

49-56.

34
4. K. Klemola, J. Ilme (1996). Distillation efficiencies of an industrial-scale i-Butane/n-

Butane fractionator. Ind. Eng. Chem. Rs., 35, 4579-4586.

5. K. Klemola, J. Aittamaa, L. Nystrom, J. Ilme (2001). Calculating distillation efficiencies

of multicomponent i-Butane/n-Butane column. Chem. Eng. Comm., 184, 1-21.

t
ip
6. M.W. Ashton, N.F. Biddulph (1977). Deducing multicomponent distillation efficiencies

cr
from industrial data. Chem. Eng., 14(7), 7-15.

us
7. J. Aittamaa (1981). Estimating multicomponent plate efficiencies in distillation. Kemia-

Kemi, 8, 295.
an
8. J. Ilme (1997). Estimating plate efficiencies in simulation of industrial scale distillation
M
columns, Ph.D. Thesis. Finland: Lappeenranta University of Technology.

9. K. Klemola (1998). Efficiencies in distillation and reactive distillation. Acta Polytechnica

ed

Scandinavica, Chemical Technology Series, 257.

pt

10. K. Jakobsson, J. Aittamaa (2001). Comparison of plate efficiency estimation models to

experimental results of pilot scale: a case study. Separation Technology (pp. 98-103). AIChE
ce

Annual meeting.
Ac

11. J. Ilme, K. Keskinen, V. Markanen, J. Aittamaa (2002). Simulation of industrial

multicomponent MTBE distillation column using an equilibrium stage model with efficiency

calculation. Distillation & Absorption. Inst. Chem. Engrs, Symp. Series.

35
12. K. Jakobsson, J. Aittamaa, K.I. Keskinen, J. Ilme (2002). Plate efficiencies of industrial

scale dehexanizer. Baden-Baden, Germany: Distillation & Absorption conference.

13. E.J. Henley, D.K. Roper, J.D. Seader (2011). Separation process principles, chemical

and biochemical operations. New York: John Wiley.

t
ip
14. E. Murphree (1925). Rectifying column calculations with particular reference to n-

cr
component mixtures. Ind. Eng. Chem., 17, 747-750.

us
15. M.R. Rahimi, H. Karimi (2011). Modeling distillation mass transfer efficiency.

International Journal of Chemical and Environmental Engineering, 2(3).

an
16. K. Jakobsson (2005). Modeling and simulation of process configurations combining
M
distillation and reaction, Chemical engineering report series. Espoo: Kemian laitetekniikan

raporttisarja, No. 47.

ed

17. R. Taylor, R. Krishna (1993). Multicomponent mass transfer (1st ed.). New York: John

Wiley.
pt

18. L. Durbin, C. Holland, N. Welch (1964). Mixing on valve trays and in downcomers of a
ce

distillation column. AIChE, 10, 373.

Ac

19. M. O'Connel, H. Gautreaux (1955). Effect of length of liquid path on plate efficiency.

Chem. Eng. Progr., 51, 232-237.

36
20. M.J. Haselden, G.G. Ashley (1970). The calculation of plate efficiency under conditions

of finite mixing in both phases in multiplate columns, and the potential advantage of parallel

flow. Chem. Eng. Sci., 25, 1665.

21. H. Kister (1992). Distillation design. New York: McGraw-Hill.

t
ip
22. AIChE. (1958). Bubble tray design manual; Prediction of fractionation efficiency. New

cr
York: AIChE.

us
23. H. Fair, J. R. Chan (1984a). Prediction of point efficiencies on sieve trays: 1. Binary

systems. Ind. Eng. Chem. Proc. Des. Deu., 23, 814-819.

an
24. H. Fair, J. R. Chan (1984b). Prediction of point efficiencies on sieve trays: 2.
M
Multicomponent systems. Ind. Eng. Chem. Proc. Des. Deu., 23, 820-827.

25. R. Agrawal, P.J. Cook, D.L. Bennet (1983). New pressure drop corellation for sieve tray
ed

distillation columns. AIChE, 29(3), 434-442.

pt

26. K. Fujiwara, T. Nakata, N. Okamoto (1993). Method foe determining relaxation factor

for modified Newton-Raphson method. IEEE Transaction on Magnetics, 29(2), 1962-1965.

ce

27. L.M. Naphtali, D.P. Sandholm (1971). Multicomponent separation calculations by

Ac

linearization. AIChE, 17(1), 148-153.

28. A. Danesh (1998). PVT bihaviour of petroleum reservoir fluids. Heriot Watt University,

Scotland: Elsevier.

37
29. R. C. Reid (1987). The properties of gases and liquids (forth ed.). New York: McGraw-

Hill, Inc.

t
ip
cr
us
an
M
ed
pt
ce
Ac

38
Figure 1. Interphase mass transfer in a multistage operation

t
ip
cr
us
an
M
ed
pt
ce
Ac

39
Figure 2. Schematic of a single separation stage

t
ip
cr
us
an
M
ed
pt
ce
Ac

40
Figure 3. The framework of developed MATLAB code

t
ip
cr
us
an
M
ed
pt
ce
Ac

41
Figure 4. The sequence of simulation steps

t
ip
cr
us
an
M
ed
pt
ce
Ac

42
Figure 5. The sequence of tray efficiency prediction

t
ip
cr
us
an
M
ed
pt
ce
Ac

43
Figure 6. Simplified flow sheets in addition to reconciled experimental data

t
ip
cr
us
an
M
ed
pt
ce
Ac

44
Figure 7. Pseudo binary McCabe-Thiele diagram for C4 splitter column with an expanded view

t
ip
cr
us
an
M
ed
pt
ce
Ac

45
Figure 8. Predicted number of transfer units

t
ip
cr
us
an
M
ed
pt
ce
Ac

46
Figure 9. The estimated vapor phase efficiencies

t
ip
cr
us
an
M
ed
pt
ce
Ac

47
Figure 10. Liquid phase mole fraction profiles for C4 splitter column

t
ip
cr
us
an
M
ed
pt
ce
Ac

48
Figure 11. Liquid phase mole fraction profiles for BTX column

t
ip
cr
us
an
M
ed
pt
ce
Ac

49
Figure 12. Comparison of reconciled and simulated compositions of products for C4 splitter
column

t
ip
cr
us
an
M
ed
pt
ce
Ac

50
Figure 13. The predicted temperature profiles accompanied by some experimental data

t
ip
cr
us
an
M
ed
pt
ce
Ac

51
Table 1. Estimation of necessary physical properties [29]

Property Method
Vapor pressure Antoine equation
Vapor binary diffusion coefficient Wilke and Lee method
Liquid binary diffusion
at infinite dilution Reddy and Doraisway method
at finite dilution Vignes method
Pure vapor viscosity Method of Chung

t
Mixture vapor viscosity Method of Wilke

ip
Pure liquid viscosity Correlation of experimental liquid viscosity data
Fugasity SRK equation of state

cr
us
an
M
ed
pt
ce
Ac

52
Table 2. The adjusted columns performance data

C4 splitter BTX column

feed stage 37 28
feed flow rate 26,122 kg/h 76.6893 mol/s
bottoms product flow 17,999 kg/h 48.1769 mol/s
Top product flow rate 8123 kg/h 28.5124 mol/s
reflux flow rate 92838 kg/h 184 mol/s
reflux temperature 18.5 oC 123 oF

t
column top pressure 650 kPa 9.9 psig

ip
feed pressure 892.67 kPa 13.3 psig

cr
us
an
M
ed
pt
ce
Ac

53
Table 3. Columns specifications

C4 splitter column BTX column

column diameter 2.9 m 2.43 m
number of trays 74 54
type of trays two-pass Ballast V-1 valve single crossflow Koch type Kx-9-u valves
weir length (side) 1.859 m
2.30029 m
weir length (center) 2.885 m
outlet weir height 51 mm 54.08 mm

t
liquid flow path length 967 mm per pass 748.429 mm

ip
active area 4.9 m2 2.57507 m2
downcomer area 0.86 m2 0.283258 m2
tray spacing 600 mm 610 mm

cr
hole diameter 39 mm 50.8 mm
total hole area 0.922 m2 0.630425 m2

us
an
M
ed
pt
ce
Ac

54
Table 4. The quantitative comparison between simulation and experimental product
compositions for C4 splitter column

Top product Bottom product

Component Relative error Relative error
Simulation Experimental Simulation Experimental
(%) (%)
propane 0.064222832 0.064108685 0.18 3.91E-17 0

t
iButane 0.925763524 0.927395158 0.18 0.00345924 0.003008596 14.98

ip
nButane 0.003102686 0.001968992 57.58 0.98349419 0.983810835 0.03
iButene 0.003175131 0.0023457 35.36 0.00049914 0.000831119 39.94
1-Butene 0.003735827 0.004181465 10.66 0.00129627 0.001038899 24.77

cr
neoPentane 8.87E-12 0 0.00128919 0.001373424 6.13
iPentane 1.84E-20 0 0.00902436 0.009048441 0.27
nPentane 8.64E-23 0 0.0009376 0.000888686 5.50

us
an
M
ed
pt
ce
Ac

55
Table 5. The quantitative comparison between simulation and experimental product
compositions for BTX column

Top product Bottom product

Component Relative error Relative error
Simulation Experimental Simulation Experimental
(%) (%)
Benzene 0.999150691 0.999958 0.081 1.59E-08 3.3E-07

t
Toluene 8.4931E-04 4.19E-05 0.9115992 0.9116191 0.002

ip
Xylene 1.0381E-10 7.E-08 0.0884008 0.0883806 0.023

cr
us
an
M
ed
pt
ce
Ac

56