Shortcut Procedure Simulating Batch Distillation Operations: Suresh Sundaram Lawrence Evans'
Shortcut Procedure Simulating Batch Distillation Operations: Suresh Sundaram Lawrence Evans'
Shortcut Procedure Simulating Batch Distillation Operations: Suresh Sundaram Lawrence Evans'
A fast, accurate model for batch distillation simulation will be useful both in batch distillation
synthesis and in batch process design. A shortcut model using the Fenske-Underwood-Gilliland
(FUG) equations of continuous distillation design was developed. The model consists of stepping
forward in time using a first-order explicit integration scheme with a variable time step and solving
the FUG equations a t each time step. A large number of example problems were used for model
testing and validation. Agreement between the shortcut model and rigorous simulation was excellent.
The model is a powerful and computationally fast tool that can be used both in batch process design
and in synthesis of batch distillation systems.
-c.
Ne~ton,s
c
- Compute
~
Specified feed. vapor rite. number
of stages. reflux ratio of column
Solve F W quatiom:
- Guess minimumreflux ratio
- Compute minimumnumber of stages
W,i, /amin
+
Compute splits of all components
(Fcarke’r quation)
(Gillihd’s correlation)
( U d e m d ’ s equation)
(s): Reflux ratio = 1 0
a 0
0.4
0.6
0.2
0
0
,
:m0.4
50 100
*:m
(b): Reflux ratio = 25.0
0.2
0 50 100
1
Repeat
if stop criterion
i s unsatisfied
Take a small time slep
I
I
CIllcuLPte new fced mount m d compositions
I
0 ’
0 50 100 0 50 100
Figure 1. Shortcut model algorithm. %distilled % distilled
Figure 2. Comparison of shortcut model with rigorous simulation
are also specified. A preliminary estimation of the using BATCHFRAC. Binary separation: XF = 0.8 0.2; UAB 1.1;
distribution of the nonkey components is made. Column Nstg= 20 (+, component A (BATCHFRAC); X, component B
pressure and type of condenser are determined from bubble (BATCHFRAC);solid line is shortcut model).
and dew point calculations of the distillate and bottoms. (R): Reflux rntio = 1.0 (b): Reflux rntio = 25.0
The feed-phase condition can be calculated once the
column pressure is known.
‘ 7
ll++-
The next step in the procedure is the calculation of the Ij I I --===-l
minimum number of stages for the specified splits of the
two key components using Fenske’s equation:
‘hk,B \I
--
1r.n (’lk,D
A; 0.200 50 100
the specified separation of the two keys. The minimum % distilled % distilled
reflux ratio corresponds to a column of infinite stages, Figure 3. Comparison of shortcut model with rigorous simulation
and the point at which this occurs is referred to as a pinch using BATCHFRAC. Binary separation: XF = 0.1 0.9; U A B = 10.0;
point. When all components in the feed distribute to both Nstg= 3 (+, component A (BATCHFRAC); X, component B
(BATCHFRAC);solid line is shortcut model).
the distillate and bottoms, a single pinch point is obtained,
and the separation is classified as a class 1 separation
(Shiras et al., 1950). If one or more of the components a class 2 separation, the minimum reflux ratio obtained
occur in either the distillate or the bottoms but not in will be an upper bound to the true minimum. This is
both, two pinch points occur in the column, and the because the presence of the nonkey components in the
separation is classified as a class 2 separation. pinch zone increases the difficulty of the separation, thus
For the rectifying section pinch point of a continuous increasing the reflux requirement.
column, the following equation can be written from mass For a class 2 separation, x , is not simply related to the
balances and equilibrium relationships: feed composition. The two equations devised by Under-
wood in this case are
a 0.4
a 0.4
a
0.200 SO 100 0.200 50 100
Y -7
t/
w o,6K
o.8
i
0 SO
J100 - 0
0 so 100
%distilled % distilled RC distilled c distilled
Figure 4. Comparison of shortcut model with rigorous simulation Figure 6. Comparison of shortcut model with rigorous simulation
using BATCHFRAC. Binary separation: XF = 0.4 0.6;a A B = 10.0; using BATCHFRAC. Ternary separation: (XAC = 2.0; a g c = 1.5; N8e
Nsts = 3 (+, component A (BATCHFRAC); X, component B = 20; reflux ratio = 10.0 (+, component A (BATCHFRAC); X,
(BATCHFRAC); solid line is shortcut model). component B (BATCHFRAC); 0,component C (BATCHFRAC);
solid line is shortcut model).
(a): o ~ c = 2 . 0osc=l.5 (b): o*c=6.0 o,gc=l.l
( a ) : O ~ c = 2 . 0Oec=1.5 (b): %c=6.0 0sc=1.2
0.8 1 0.8
0.8 1
'I
0.6
a o.6
0.4
a 0.4
0.6
a
0.4
0.2 0.2
0.2
0 0
0 so 100 0 50 I00 n
"0 50
""f
1
0.6 0.6
R 0.4
0.6
R
0.200
::M 50
%distilled
100 0.2
0
0
%distilled
100
using BATCHFRAC. Ternary separation: XF = 0.33 0.33 0.34;N 8 ~ Figure 7. Comparison of shortcut model with rigorous simulation
= 3; reflux ratio = 10.0 (+, component A (BATCHFRAC); X, using BATCHFRAC. Ternary separation: ZF = 0.4 0.5 0.1;N.Q =
component B (BATCHFRAC); 0,component C (BATCHFRAC); 3;reflux ratio = 10.0 (+,component A (BATCHRRAC);X, compOnent
solid line is shortcut model). B (BATCHFRAC); 0 , component C (BATCHFRAC); solid line 1
shortcut model).
Equation 3 is first solved for m roots of 8 where m is one
lees than the number of distributing Components. Qua- equation form of Gilliland's correlation is (Eduljee, 1975)
tion 4 is then written for each value of 8, and the m
equations are solved simultaneously to yield R m i n and the
splits of the distributing components.
The feed stage location can be located using the
At this point in the design, the minimum reflux ratio, Kirkbridge equation. The distribution of the nonkey
the minimum number of stages, and the splits of all the components at actual reflux is approximated to be close
componentsat totaland minimum reflux are known. The to that estimated by Fenske's equation at total reflux.
actual reflux ratio is generally established at some multiple
of the minimum reflux. The value of R I R m i n usually lies The Shortcut Model Algorithm
between 1.1 and 1.5. The number of theoretical stages for
the specified separation is then determinedfrom Gilliland's The shortcut model is made up of two nested loops, as
empirical correlation relating Rmin, N m i n , R,and N. One shown in Figure 1. For the outer loop, the differential
614 Ind. Eng. Chem. Res., Vol. 32, No. 3, 1993
balances for total mass in the pot and for each component
can be written as follows
dM -v 0 50 100
dt R + l % distilled
where Vis the vapor rate of the column and R is the reflux Figure 8. Comparison of shortcut model with rigorous simulation
ratio, and using BATCHFRAC. Quaternary separation: XF = 0.25 0.25 0.25;
N,, = 6; reflux ratio = 10.0 (+, component A (BATCHFRAC); 0 ,
component B (BATCHFRAC); X, component C (BATCHFRAC);1,
(7) component D (BATCHFRAC); solid line is shortcut model).
the actual reflux ratio. The final equation is that the sum
of the overhead compositions should be equal to unity.
An analysis of the degrees of freedom of the inner loop
system of equations was conducted. The total number of
equations are
Fenske's equation 1
Equation 9 can be integrated numercially after substituting Underwood's equations N-1
for dM from eq 6. For example, if first-order integration Gilliland's correlation 1
sum of X~,D'S 1
with at as the time step is used, dXi,B can be approximated
as (xy- x $ ) and dM can be approximated as (Mnew -
Mold). Equation 9 would then become The unknowns in this system of equations are
overhead compositions x ~ , ~ N
minimum reflux ratio 1
minimum number of stages 1
with
The number of equations and number of unknowns both
total N + 2. The system is thus a well-determined system
which can be solved. The equations can be simplified to
At the beginning of the distillation, the feed composition a single nonlinear equation in one unknown as shown in
(xfi = Zi,F) and the total mass of the feed (Mold)are the Appendix. It is solved using Newton's method for
known. Assuming a small positive at, the new pot which the required derivative is derived in the Appendix.
compositions and the new maas in the pot can be calculated
from eqs 10and 11,assuming the vapor rate of the column, Model Testing and Validation
the reflux ratio, and the overhead compositions corre-
sponding to the initial feed composition are known. The model was extensively tested on a number of
After each iteration of the outer loop, the pot compo- examples. The examples presented here fall into three
sitions and amount, the vapor rate of the column, the classes: (1)binary feed separations where feed compo-
number of stages, and the reflux ratio are known. The sition, relative volatility, number of stages, and reflux ratio
inner loop uses the FUG equations to compute the were systematically varied; (2) ternary feed separations
overhead vapor compositions from these given values. with the same variations as the bindary examples; (3)
Fenske's equation (eq 1) is used to determine the specific multicomponent examples with varying reflux
minimum number of stages for the system. For batch ratio policies. In each case, the separation was simulated
distillation, the pinch has to occur in the rectifying section. using BATCHFRAC, and the results from the shortcut
It will be assumed that all components distribute, making simulation are compared to those from BATCHFRAC.
it a class 1separation. Hence the applicable Underwood's Binary Examples. These were the first set of com-
equations are eq 2 with xi,.. replaced by x~,B. Gilliland's parisons to be made. Twenty-four separations were
correlation is used to relate the minimum number of stages, simulated using the shortcut model with three values of
minimum reflux ratio, the actual number of stages, and the feed composition and two values each of the relative
Ind. Eng. Chem. Res,, Vol. 32, No. 3, 1993 515
Table 11. Test Problems Used for Comparison of Shortcut Model with Rigorous Simulation
Binary Comparisons
feed composition, XA QAB N8-e reflux ratio remarks
0.1 1.1 3 1.0
0.1 1.1 3 25.0
0.1 1.1 20 1.0
0.1 1.1 20 25.0
0.1 10.0 3 1.0 Figure 4
0.1 10.0 3 25.0 Figure 4
0.1 10.0 20 1.0
0.1 10.0 20 25.0
0.4 1.1 3 1.0
0.4 1.1 3 25.0
0.4 1.1 20 1.0
0.4 1.1 20 25.0
0.4 10.0 3 1.0 Figure 5
0.4 10.0 3 25.0 Figure 5
0.4 10.0 20 1.0
0.4 10.0 20 25.0
0.8 1.1 3 1.0
0.8 1.1 3 25.0
0.8 1.1 20 1.0 Figure 3
0.8 1.1 20 25.0 Figure 3
0.8 10.0 3 1.0
0.8 10.0 3 25.0
0.8 10.0 20 1.0
0.8 10.0 20 25.0
Ternary Comparisons
feed composition Q
feed composition ff
volatility, the number of stages, and the reflux ratio. For A list of all the comparisons made is shown in Table 11.
each separation, distillation was simulated until 99 % of Comparisons with rigorous simulations for all these test
the initial charge had been distilled. Pot liquid compo- problems are provided in Sundaram (1991).
sition and instantaneous overhead vapor composition Separation of Multicomponent Organic Feed. This
profiles were compared to those obtained using BATCH- example is identical to the one used by Boston et al. (1981)
FRAC to simulate the same separation. When using in their paper on BATCHFRAC. The problem description
BATCHFRAC, constant molal overflow and constant is given in Table 111. The feed consists of four components,
relative volatility were assumed. propane, butane, pentane, and hexane. The column
Some comparisons are shown in Figures 2-4. The results consists of eight internal stages. There are five operating
indicate that the shortcut model does a good job of steps, with stopping criteria given for each operating step.
predicting batch distillation composition profiles, at least The objectives are to produce butane at 99.0% purity and
for binary separations. hexane at 99.98% purity. The purpose of the first two
steps is to remove the most volatile component, propane.
Ternary Examples. The shortcut model was used to In actual operation, the second propane removal step is
simulate three-component separations for various values terminated when the mole fraction of butane reaches 0.985
of feed compositions, relative volatilities between com- in the distillate. Then the main accumulator is dumped,
ponents, number of stages, and reflux ratios. Some of the and material containing 40% butane and 60% hexane is
results are shown in Figures 5-7. Again, the shortcut model added to the pot, before beginning the third step. This
accurately predicts composition profiles for three-com- step is terminated when the mole fraction of butane in the
ponent separations. accumulator reaches 0.9900. The accumulator is dumped
Quaternary Feed Separation. The problem descrip- again before the first of two pentane removal steps is
tion is given in Table I. The feed consists of four started. At the end of the last operation step, the mole
components and distillation is conducted in a column with fraction of hexane in the step is 0.9998, and this may be
six stages until 99% of the initial charge is distilled. The drained off as the final hexane product.
column is operated at a reflux ratio of 10.0. Instantaneous The problem was simulated using BATCHFRAC
vapor and pot liquid compositions from the shortcut model assuming zero holdups and constant relative volatilities.
are compared to those obtained from rigorous simulation Material balances for each operating step were obtained
in Figure 8. from the simulation, and the percent distilled at the end
516 Ind. Eng. Chem. Res., Vol. 32, No. 3, 1993
Table 111. Problem Description: Multicomponent Organic Potential Shortcomings of the Model
Feed Sesaration
column specification The shortcut model did an excellent job of matching
number of internal stages 8 simulation results for all of the test cases studied.
pressure, psia 14.7 However, the model is an approximate one and cannot be
initial charge, lb-mol 100 expected to perform well in all possible situations. To
feed composition (mole fraction) determine possible causes of model deficiency, the effects
propane 0.1 of all the major assumptions made during the development
n-butane 0.3
n-pentane 0.1 of the model equations were examined.
n-hexane 0.5 The first major assumption is that, during each time
step, the batch column can be represented using the
Operation Step Specifications
Fenske-Underwood-Gilliland equations. The implicit
OPSTEP purpose reflux ratio distillate rate assumption is that, at every time step, the vapor and liquid
1 propane removal 5 2 at each stage in the column are in equilibrium and a quasi-
2 propane removal 20 2 steady-state profile is established equivalent to the profile
3 butane production 25 2 for a continuous column. This assumption effectively
4 pentane removal 15 2 neglects the effect of holdup on the dynamic behavior of
5 pentane removal 25 2
the column.
Physical Property Modele: Zero Holdup Case A second set of assumptions is related to the FUG
component re1 volatility method itself, and ita implementation. In the model, it
propane 4.58
was assumed that all components present in the column
n-butane 1.00 distribute between the bottoms and the distillate. The
n-pentane 0.258 lightest component was assumed to be the light key
n-hexane 0.067 component, and the heaviest component was assumed to
be the heavy key component. This means that at minimum
reflux, the zone of constant composition (or the pinch)
of each operating step was computed. This was used as lies adjacent to the feed stage (Shiras et al., 1950). This
the stopping criteria for the shortcut model. The com- allowed the use of a simplified version of Underwood's
parison of results from the shortcut model and from equation, thereby considerably simplifying the solution
BATCHFRAC are shown in Table IV. From Table IV, it procedure.
can be seen that the material balances agree well with There are two exceptionsto the assumed behavior. The
those obtained from rigorous simulation. Since constant first is when all components distribute but the zone of
molal overflow and constant distillate rate were assumed constant composition is not adjacent to the feed stage.
for both the shortcut model and BATCHFRAC, the This happens when t t z system is sufficiently nonideal so
distillation times were identical for both simulations. as to give the equivalent of a tangent pinch point. Such