Sie sind auf Seite 1von 3

June 2011, Volume 2, No.

3
International Journal of Chemical and Environmental Engineering


Modeling distillation mass transfer efficiency

Mahmood Reza Rahimi and Hajir Karimi
Chemical engineering department, Faculty of engineering, Yasouj University, Yasouj 75918-74831, Iran
*Corresponding author: mrrahimi@mail.yu.ac.ir

Abstract
Mass transfer efficiency was modeled by developing a computational fluid dynamic based artificial neural network model. A CFD
model was developed and validated for hydrodynamics and mass transfer characteristics of distillation column. Generated data by
CFD include a wide range of operating conditions, tray geometries and material properties were used for training and testing network.
The model results were close to experimental data and shown that CFD-ANN model can be used as a powerful tool in tray design and
analysis.
Keywords: Distillation, computational fluid dynamics, artificial neural networks, mass transfer efficiency, sieve trays.
1. Introduction

Distillation is a separation process of major
importance in chemical and petroleum industries. Increase
of separation efficiency as well as its prediction has been
a major task in design and operation of distillation
columns.
Determination of theoretical stages required for a
desired separation is the first stage of distillation column
design, thereafter, by using column efficiency, the actual
number of trays is determined. This column efficiency is
calculated from the Murphree tray efficiency, and this
efficiency is calculated from the point efficiency.
Therefore, the knowledge of the point efficiency is
essential for design of distillation columns. The
predication of industrial distillation tray efficiencies is
done usually by the following procedures [1]:1-
Comparison with the tray efficiency of similar operating
columns, 2- Scaling up from laboratory columns,3-
Empirical correlation, 4- Theoretical to semi theoretical
mass-transfer methods. The best and surest method of tray
efficiency calculation, to day is to use the value of a
similar column as a reference [1]. Unfortunately, such
data are seldom available, they should be used as basis for
efficiency of separation [1]. On the other hand, the
prediction of point efficiency remains uncertain and little
real progress has been achieved since the AIChE Bubble
Tray design Manual [2] was published. The AIChE semi-
empirical correlation was based on the assumption that
point efficiencies were real and measurable. It is clear that
direct measurement of this quantity is difficult however.
The correct prediction of point efficiencies is subject of
question and there is not a unified model to predict its
variation along the tray. Point efficiency should be
properly based on vapor-liquid mass-transfer
fundamentals and transport between phases in the
turbulent two phase dispersion.
There have been few attempts to model sieve tray
hydrodynamics using CFD simulation, were reviewed in
references [3, 4]. Rahimi et al., [3] were studied the
hydrodynamics of sieve trays, by means of a 3D two fluid
CFD simulation. The velocity, concentration, and
temperature distributions were determined [4].
Nowadays, artificial neural networks (ANNs) method
has been found extensive application in the various fields
of chemical engineering. ANN is a model that attempts to
mimic simple biological learning processes and simulate
specific functions of human nervous system [5]. This
model creates a connection between input and output
variables and keeps the underlying complexity of the
process inside the system.
In this work a CFD-ANN model is developed to give
the predictions of sieve tray efficiency. The main
objective has been to find the extent to which CFD can be
used in combination with artificial neural network as a
prediction tool for efficiencies of industrial trays.

2. Mass-Transfer Efficiency
The three different efficiencies are defined as:
(i) Point Efficiency, The point efficiency is defined
as
point
1 n
*
n
1 n n
p
Y Y
Y Y
E
|
|
.
|

\
|

(1)
The vapor and liquid phase number of mass-transfer units
are defined as
(ii) Murphree Tray Efficiency. The Murphree tray
efficiency,
MV
E , is defined as
1]/ ) E [exp( E
p MV
= (2)
Modeling distillation mass transfer efficiency


166

Where
p
E is the point efficiency. The point and
Murphree efficiencies are the same when the phases in the
froth are completely mixed, if the liquid is in plug flow
the Murphree efficiency is greater than the point
efficiency. For deduction of point efficiency from
measured Murphree efficiency the mixing model is
required.
(iii) Column Efficiency. This efficiency is simply the
ratio of the number of theoretical plates to the actual
number of plates required for a desired separation. The
column efficiency is deduced from the Murphree tray
efficiency.

3. Model Development
In this work the hydrodynamic and mass transfer of
sieve trays is considered. The dispersed gas and the
continuous liquid phases are modelled in the Eulerian-
Eulerian frame work as two interpenetrating phases with
separate transport equations. A 3-D two-fluid CFD model
was developed to predict the hydrodynamics and mass
transfer of sieve trays. Interaction between the two phases
occurs via interphase momentum and mass transfer.
Continuity, momentum conservation and mass transfer
equations were used, they are not repeated here, they can
be seen in a previous paper [3]. The methodology for
point efficiency calculation from concentration
distribution results of CFD, inside the tray, has been
reported elsewhere [3] and not repeated here.
Artificial neural networks (ANNs) is an efficient
algorithm to approximate any function with finite number
of discontinuities by learning the relationships between
input and output vectors. The ANNs are able to deal with
non-linear problems, and once trained can perform
prediction and generalization rapidly. ANNs are
biological inspirations based on the various brain
functionality characteristics. They are composed of many
simple elements called neurons that are interconnected by
links and act like axons to determine an empirical
relationship between the inputs and outputs of a given
system. Multiple layers arrangement of a typical
interconnected neural network is consist of an input layer,
an output layer and one hidden layer with different roles.
Each connecting line has an associated weight. An ANN
is trained by adjusting these input weights (connection
weights), so that the calculated outputs may be
approximated by the desired values. The output from a
given neuron is calculated by applying a transfer function
to a weighted summation of its input to a give output,
which can serve as input to other neurons, as follows:

=

+ =
1
1
1
k
N
i
jk ) k ( i ijk k jk
) w ( F | o o (3)
Where
jk
o is neuron j s output fromk s layer
jk
is the
bias weight for neuron j in layer k . The neurons in the
hidden layer perform two tasks: summing the weighted
inputs connected to them and passing the result through a
non linear activation function to the output or adjacent
neurons of the corresponding hidden layer.
The model fitting parameters w
ijk
are the connection
weights. The nonlinear activation transfer functions
k
F

may have many different forms [6, 7,8].
The training process requires a proper set of data i.e.
input (I
i
) and target output (t
i
). During training the
weights and biases of the network are iteratively adjusted
to minimize the network error function. The typical error
function that is used is the AARE:

=
|
|
.
|

\
|
=
N
i i
i i
t
t
ABS
N
AARE
1
1 o
(4)
There are many different types of neural networks,
differing by their network topology and/or learning
algorithm. In this paper the Back Propagation (BP)
learning algorithm, which is one of the most commonly
used algorithms in ANN training, is applied to predict the
sieve tray efficiency. The simplest implementation of BP
learning is the network weights and biases updates in the
direction of the negative gradient that the performance
function decreases most rapidly. An iteration of this
algorithm can be written as follows [10]:
k k k k
g l x x =
+1
(5)
The neural nets learn to recognize the patterns of the data
sets during the training process. Neural nets learn the
patterns of the data set letting the analyst to perform more
interesting flexible work in a changing environment [5].
Although neural network may take some time to learn a
sudden drastic change, but it is excellent to adapt
constantly changing information. However the
programmed systems are constrained by the designed
situation and they are not valid otherwise. The neural
networks can easily model data even with very complex
interactions, which are too difficult to model with
traditional methods like nonlinear regressions [5]. The
neural networks built models are more reflective of the
data structure and are significantly faster.
The network inputs are liquid density (
l
p
), vapor density
(
g
p
), liquid viscosity (
l
u ), vapor viscosity (
g
u ), liquid
diffusivity (
l
D
), and vapor diffusivity (
v
D
), surface
tension (o ), slope of the equilibrium curve (m), hole
diameter (
h
d ), weir height (
h
w ), weir length (
w
L
) ,
liquid flux (
l
Q
), gas flux (
g
Q
), ratio of hole area to active
area of the tray (HA/AA) while the output is point
efficiency( E
P
).

4. Results and Discussion
The results show that the CFD model predicts point
efficiency close to the experimentally measurements. The
predicted liquid composition for a binary mixture
(MeOH/nPrOH) in a rectangular tray is close to
Modeling distillation mass transfer efficiency


167

experimental data as shown in Fig. 1. The methodology
for point efficiency calculation from concentration
distribution results of CFD, inside the tray, has been
reported elsewhere [3, 4] and not repeated here. An
averaging of such data is required for preparation of data
for ANN model.

Figure 1. Centerline liquid composition and liquid composition
profiles for the rectangular tray,MeOH/nPrOH binary system. xm =
0.2790.

Figures (2, 3) confirm the capability and accuracy of
the ANN model to predict the point efficiency of sieve
trays. In these figures the point efficiency predicted by net
(A) were plotted against experimental ones (T) in the two
main stages, learning and testing stages.

20 40 60 80 100
20
30
40
50
60
70
80
90
100
T
A
R = 0.909
Data Points
Best Linear Fit
A = T

Figure 2. The relation between net response and experimental data
in learning stage.

20 30 40 50 60 70 80 90 100
20
30
40
50
60
70
80
90
100
T
A
R = 0.91
Data Points
Best Linear Fit
A = T
Figure 3. The relation between net response and experimental data
in prediction stage.
The absolute error in learning stage is 0.329 and
R
2
=0.909, whereases the absolute error in testing stage is
3.37 and R
2
=0.91, where shown that model have good
prediction. The Prediction accuracy of this CFD-ANN
model is very high compared to conventional methods of
efficiency calculation. The results were shown that the
CFD-ANN model predicts point efficiency close to the
experimentally measurements.
5. Conclusion
The ability of CFD- ANN modeling to model and
predict sieve tray efficiency has been investigated in this
work. Predicted tray efficiencies have been compared
with experimental data. The results show a good
agreement between experimental data and those predicted
by CFD-ANN. It has been clearly shown that the CFD-
ANN calculates the point efficiency based on the
experimental data only, instead of using empirical
models. Therefore it is not necessary to use approximate
and complex analytical equations to calculate sieve tray
efficiency. In comparison to available models, the
proposed CFD-ANN model provides very accurate
results. This ANN model can be used for fast estimating
the point efficiency for new cases where be between the
Maximum and minimum limits of Table. 1. Because the
CFD runs are time consuming relative to ANN model
using this ANN model can predict the efficiencies in
shorter time with good accuracy that is better than
available theoretical methods.

REFERENCES

[1] K.T. Kelemola, J.K. Ilme, Distillation efficiencies of an industrial-
scale i-butane/n-butane fractionator, Ind. Eng. Chem. Res. 1996,
35, 4579-4586.

[2] AIChE, Bubble Tray Design Manual, Am.Inst.Chem.Engrs.,New
York 1958.

[3] R. Rahimi, M. R. Rahimi, F. Shahraki, M. Zivdar, efficiencies of
sieve tray distillation columns by CFD simulation, Chem. Eng and
Technol, 2006, 29, 3, 326-335.

[4] M. R. Rahimi, H. Karimi, CFD Simulation of Hydraulics of Sieve Trays
with Gas Mal-Distribution, Chemical product and process modeling,
Vol 5 (2010), issue1, article 2.

[5] A.R. Moghadassi, F. Parvizian, S.M. Hosseini, A.R. Fazlali, " A new
approach for estimation of PVT properties of pure gases based on
artificial neural network model" Braz. J. of Chem. Eng., 2009, l,
26, 199-206.

[6] R.I.W. Lang, A Future for Dynamic Neural Networks, Dept.
Cybernetics, University of Reading, UK, 2000.

[7] A.B. Bulsari, Neural Networks for Chemical Engineers, Elsevier
Science Press, Amsterdam, 1995.

[8] H. Karimi, N. Saghatoleslami, M. R. Rahimi, Prediction of water
activity coefficient in Triethylene glycol -Water system using
diffusion neural network (DNN), Chem. Biochem. Eng. Quar.,
2010, 2, 167-176

Das könnte Ihnen auch gefallen