Sunteți pe pagina 1din 12

Chemical Engineering Science 62 (2007) 1839 – 1850

www.elsevier.com/locate/ces

A modified model of computational mass transfer for distillation column


Z.M. Sun, K.T. Yu, X.G. Yuan ∗ , C.J. Liu
State Key Laboratory for Chemical Engineering (Tianjin University) and School of Chemical Engineering and Technology, Tianjin University, Tianjin 300072,
People’s Republic of China

Received 9 October 2005; received in revised form 4 December 2006; accepted 11 December 2006
Available online 27 December 2006

Abstract
The computational mass transfer (CMT) model is composed of the basic differential mass transfer equation, closing with auxiliary equations,
and the appropriate accompanying CFD formulation. In the present modified CMT model, the closing auxiliary equations c2 .c [Liu, B.T.,
2003. Study of a new mass transfer model of CFD and its application on distillation tray. Ph.D. Dissertation, Tianjin University, Tianjin, China;
Sun, Z.M., Liu, B.T., Yuan, X.G., Liu, C.J., Yu, K.T., 2005. New turbulent model for computational mass transfer and its application to a
commercial-scale distillation column. Industrial and Engineering Chemistry Research 44, 4427–4434] are further simplified for reducing the
complication of computation. At the same time, the CFD formulation is also improved for better velocity field prediction. By this complex
model, the turbulent mass transfer diffusivity, the three-dimensional velocity/concentration profiles and the efficiency of mass transfer equipment
can be predicted simultaneously. To demonstrate the feasibility of the proposed simplified CMT model, simulation was made for distillation
column, and the simulated results are compared with the experimental data taken from literatures. The predicted distribution of liquid velocity
on a tray and the average mass transfer diffusivity are in reasonable agreement with the reported experimental measurement [Solari, R.B.,
Bell, R.L., 1986. Fluid flow patterns and velocity distribution on commercial-scale sieve trays. AI.Ch.E. Journal 32, 640–649; Cai, T.J., Chen,
G.X., 2004. Liquid back-mixing on distillation trays. Industrial and Engineering Chemistry Research 43, 2590–2597]. In applying the modified
model to a commercial scale distillation tray column, the predictions of the concentration at the outlet of each tray and the tray efficiency
are satisfactorily confirmed by the published experimental data [Sakata, M., Yanagi, T., 1979. Performance of a commercial scale sieve tray.
Institution of Chemical Engineers Symposium Series, vol. 56, pp. 3.2/21–3.2/34]. Furthermore, the validity of the present model is also shown
by checking the computed results with a reported pilot-scale tray column [Garcia, J.A., Fair, J.R., 2000. A fundamental model for the prediction
of distillation sieve tray efficiency. 1. Database development. Industrial and Engineering Chemistry Research 39, 1809–1817] in the bottom
concentration and the overall tray efficiency under different operating conditions. The modified CMT model is expected to be useful in the
design and analysis of distillation column.
䉷 2007 Published by Elsevier Ltd.

Keywords: Computational mass transfer (CMT); c2 .c model; Turbulent mass transfer diffusivity; Simulation; Sieve tray

1. Introduction efficiency, which might vary significantly from one to another


and is extremely influential to the technical–economical behav-
Distillation, the most commonly used separation process for iors of a column, has long been relying on experience, and the
the liquid mixture, has been widely used in the chemical and design of distillation columns is essentially empirical in nature
allied industries due to its reliability in large-size column ap- (Zuiderweg, 1982; Lockett, 1986). The lack of in-depth under-
plication and its maturity in engineering practice. Among the standing of the processes occurring inside a distillation column
distillation equipments, the tray column is popularly employed is known to be the major barrier to the proper estimation and
in the industrial production for its simple structure and low improvement of the column performance (AIChE, 1998).
cost of investment. However, the estimation of distillation tray With the development of computer technology and the
advancement in numerical methods, it becomes possible to
∗ Corresponding author. Tel.: +86 22 27404732; fax: +86 22 27404496. investigate the transfer process numerically with the chemi-
E-mail address: yuanxg@tju.edu.cn (X.G. Yuan). cal engineering and cross-disciplinary theories. The numerical
0009-2509/$ - see front matter 䉷 2007 Published by Elsevier Ltd.
doi:10.1016/j.ces.2006.12.021
1840 Z.M. Sun et al. / Chemical Engineering Science 62 (2007) 1839 – 1850

approaches have many advantages, such as offering more in- to improve the velocity modeling, which is influential to the
depth information than that upon experiments, shortening the computed tray efficiency. To testify the validity of the sim-
cycles for process and equipment development by visualizing plification and improvement, the computed results are com-
and comparing the results of virtual trials and modifications. pared with the experimental data taken from literatures. The
The computational fluid dynamics (CFD) has been used suc- agreement between them demonstrates that the modified CMT
cessfully in the field of chemical engineering as a tool. method can be used effectively in analyzing the performance
In the simulation of distillation process and equipment, Yu of existing distillation column as well as assessing the tray ef-
(1992) and Zhang and Yu (1994) presented a two-dimensional ficiency before construction.
CFD model for simulating the liquid phase flow on a sieve
tray, in which the k. equations were employed to achieve 2. Proposed model for CMT
the closure of the equation system and a body force of vapor
was included in the source term of the momentum equation to 2.1. Simplification of c2 .c model
consider the interacting effect of vapor and liquid phases. On
this basis, Liu et al. (2000) developed a model describing the The instantaneous equation of turbulent mass transfer can be
liquid-phase flow on a sieve tray with consideration of both written as follows in the tensor form for avoiding complicated
the resistance and the bubbling effect generated by the uprising mathematical expression:
vapor. Later on, Wang et al. (2004) further developed a three-
dimensional model considering the effect of vapor by adding jC jC j2 C
+ Uj = D 2 + SC , (1)
drag force, lift force, virtual mass force and body force in the jt jxj jxj
model, and simulated a 1.2-m-diameter column with 10 sieve
trays under total reflux. The CFD application to distillation where U and C are the instantaneous velocity and concentration,
was also made by Krishna et al. (1999) and van Baten and respectively. If both U and C are expressed by the time average
Krishna (2000). They used fully three-dimensional transient values U and C, the foregoing equation is transformed to the
simulations to describe the hydrodynamics of trays, and gave following Reynolds average form for the transport of average
liquid volume fraction, velocity distribution and clear liquid concentration:
height for a rectangular tray and a circular tray, respectively,  
jC jC j jC
via a two-phase flow transient model. Also, Fischer and Quarini + Uj = D − uj c + S C . (2)
(1998) proposed a three-dimensional heterogeneous model for jt jxj jxj jxj
simulating tray hydraulics. Mehta et al. (1998) and Gesit et al.
Similar to Boussinesq’s assumption, the turbulent mass flux
(2003) predicted liquid velocity distribution, clear liquid height,
uj c in Eq. (2) can be expressed in terms of turbulent mass
froth height, and liquid volume fraction on trays using CFD
transfer diffusivity Dt and concentration gradient
techniques.
The idea of using CFD to incorporating the prediction on jC
tray efficiency relies on the fact that the hydrodynamics is an −uj c = Dt . (3)
jxj
essential influential factor for mass transfer in both interfacial
and bulk diffusions, which could be understood by the effect Since the turbulent mass transfer diffusivity Dt is regarded as
of velocity distribution on concentration profile. This in fact direct proportional to the product of the characteristic velocity
opens an issue on the computation for mass transfer prediction and the characteristic length, we have,
based on the fluid dynamics computation.
The key problem of this approach is the closure of the dif- Dt = Ct k 1/2 Lm . (4)
ferential mass transfer equation, as two unknown variables, the √
With the relationship Lm = k 1/2 m , m =  c , and the defini-
concentration and the turbulent mass transfer diffusivity, being
tion of two timescales (Colin and Benkenida, 2003)  = k/,
involved in one equation. The turbulent mass transfer diffu-
sivity depends not only on the fluid dynamic properties (e.g. c = c2 /c , the turbulent mass transfer diffusivity Dt can be
turbulence viscosity of the fluid) but also on the fluctuation written as
of concentration in turbulent flow. Liu (2003) proposed a two-  1/2
k c2
equation model with a concentration variance c2 equation and Dt = Ct k . (5)
 c
its dissipation rate c equation as a measure to the closure of the
differential mass transfer equation. Liu’s computational mass
transfer (CMT) model has been applied successfully to predict In Eq. (5), c2 is the concentration variance and c is the dissipa-
the turbulent mass transfer diffusivity and efficiency of a com- tion rate of concentration variance, which can be expressed as
mercial scaled distillation column by Sun et al. (2005). jc jc
However, Liu’s model is of prototype as its initial form is c = D . (6)
jxj jxj
complicated and the computation is tedious. In the present pa-
per, the c2 .c model is simplified and the model constants are For the closure of the turbulent mass transfer, or the elimi-
ascertained. At the same time, the CFD equation is modified in nation of diffusivity Dt , two auxiliary equations are developed
describing the interaction between the vapor and liquid phases as follows. Substituting C = C + c and U = U + u into Eq. (1)
Z.M. Sun et al. / Chemical Engineering Science 62 (2007) 1839 – 1850 1841

and subtracting Eq. (2), the transport equation for the concen- where  is the timescale and can be expressed as k/
tration fluctuation can be obtained. After mathematical treat- in CFD.
ments, the precise expression for c2 equation is as follows: Similarly, the production part of c equation can be modeled
  in the following manner:
jc2 jc2 j jc2 jC
+ Uj = D − uj c − 2uj c
2 − 2c . production part of c equation
jt jxj jxj jxj jxj
1
(7) = Cc1 × production part of c2 equation, (14)

Taking the derivative of Eq. (1) with respect to xk and mul-
tiplying by 2Djc/jxk and averaging, the c equation is given where the concentration timescale c2 /c is used to express .
below: The production part of c2 equation is uj cjC/jxj , then the final
  form of the production part of c equation can be written as
jc jc j jc
+ Uj = D − uj c
jt jxj jxj jxj c jC
Pc = −Cc1 uj c . (15)
jc juk jC jc j2 C c2 jxj
− 2D − 2Duj
jxj jxj jxk jxk jxj jxk Since the dissipation part of  equation in CFD can be modeled
jc jc jU j jc juj jc as
− 2D − 2D
jxk jxj jxk jxj jxk jxk dissipation part of  equation
1
j2 c j2 c = C 2 × dissipation part of k equation. (16)
− 2D 2 . (8) 
jxj jxk jxj jxk
In the same way, the dissipation part of c equation can be
Because of the presence of unknown covariance terms, the written as
foregoing two equations cannot be used for direct computa-
tion unless they are further simplified. Applying the treatment dissipation part of c equation
similar to the Reynolds stress, the turbulent diffusion terms 1
= Cc2 × dissipation part of c2 equation. (17)
uj c2 and uj c can be expressed by the following gradient type 
equations:
According to the postulation by Launder (1976), the combi-
−uj c2 = (Dt /c )jc2 /jx j, (9) nation of two timescales of velocity (k/) and concentration
(c2 /c ) are used for expressing  for the case involving mass
−uj c = (Dt /c )jc /jxj . (10) transfer. Since the dissipation part of c2 equation is c , then
The complicated Equation (8) should be simplified to the Eq. (12) can be modeled by the following form:
form suitable for computation. In this paper, the method of
2c c
modeling is employed, giving a simplified new expression as c = −Cc2 − Cc3 . (18)
follows. Let the second, third, and fourth terms on the right- c2 k
hand side of Equation (8) to be the production part as shown The auxiliary equations for closing the differential mass
below: transfer equation are finally to be
jc juk jC jc j2 C   
Pc = − 2D − 2Duj jc2 jc2 j Dt jc2 jC
jxj jxj jxk jxk jxj jxk + Uj = D+ − 2uj c − 2c ,
jt jxj jxj c jxj jxj
jc jc jU j
− 2D . (11) (19)
jxk jxj jxk
 
jc jc j Dt jc
And the following two terms on the right-hand side of Eq. (8) + Uj = D+
be the dissipation part: jt jxj jxj  c jxj

jc juj jc j2 c j2 c c jC 2 c
c = −2D − 2D 2 . (12) − Cc1 uj c − Cc2 c − Cc3 . (20)
jxj jxk jxk jxj jxk jxj jxk c2 jxj c 2 k

The simplification of the c equation might resemble the Normally, the model constants are determined by experi-
treatment of  equation in the conventional CFD. The modeling ments. At the present, in view of lacking experimental data un-
of the production part of  equation in CFD by Zhang (2002) der the condition of mass transfer, as a substitute, the follow-
is given by ing manner of determination is employed. The value of Ct in
Eq. (5) was defined as follows:
production part of  equation
C
= C1
1
× production part of k equation, (13) Ct = √ , (21)
 Sct R
1842 Z.M. Sun et al. / Chemical Engineering Science 62 (2007) 1839 – 1850

where C is the coefficient in the CFD modeling equation t = where the model constants are: Ct = 0.11, Cc1 = 1.8, Cc2 = 2.2,
C k 2 /, which is adopted generally to be 0.09. In considering Cc3 = 0.72, Cc4 = 0.8, c = 1.0, c = 1.0.
the approximate turbulent Schmidt number is generally taken
as Sct = t /Dt = 0.7 and the timescale ratio R = c / = 2.2. Application of the proposed CMT model to distillation
(c2 /c )/(k/)=0.9. (Lemoine et al., 2000), we obtain Ct =0.14. column
By the analogy between mass transfer and heat transfer, the
constants c and c in Eqs. (9) and (10) are both assigned to be The proposed CMT model as applied to distillation is
unity, which are consistent with the assumption by Elghobashi composed of two parts. The first part is the respective differ-
and Launder (1983). According to the analogy and the research ential CFD equations describing the velocity distribution on
work of Colin and Benkenida (2003) on the concentration field the distillation tray. The second part is the mass transfer equa-
of a combustion device, we choose Cc1 to be 2.0. tions including the basic differential equation together with the
Similar to the treatment of Nagano and Kim (1988), the simplified c2 .c model for its closure as derived above. In this
constants Cc2 and Cc3 are related as follows: paper, the quasi-single liquid phase flow model is adopted for
the first part, the CFD computation.
Cc2 = R(C2 − 1), (22)
Cc3 = 2/R, (23) 2.2.1. The CFD equations
For simulating the velocity profile on distillation tray, the
where C2 =1.92, which is taken from standard k. model (Eqs. equations of the steady-state continuity and momentum for
(33) and (34)), and R=0.9, which is the timescale ratio between the liquid phase in two-phase flow are adopted. In the present
concentration and velocity as given above. Consequently, we model, the liquid volume fraction is considered and the inter-
obtain Cc2 and Cc3 to be 0.83 and 2.22, respectively. action between the vapor and liquid phases is attributed to the
In summary, the model constants in the present model are source term and is also implicitly involved in the velocity of the
given below: liquid phase, which is an improvement to the former pseudo-
single-phase model (Wang et al., 2004). The CFD model can
Ct = 0.14, Cc1 = 2.0, Cc2 = 0.83, be written as
Cc3 = 2.22, c = 1.0, c = 1.0. jL U j
= 0, (26)
Furthermore, by comparing the Cc2 and Cc3 and considering jxj
the value of timescale ratio R, it is found that the numerical
jL U j 1 jp
value of Cc3 c /k is about 3 times greater than that of Cc2 2c /c2 , Ui = − L + L g
and therefore the latter term may be neglected without affect- jxi L jxj
 
ing the numerical result of simulation as demonstrated in the j jU j
subsequent section. The final form of simplified c equation + L  − L ui uj + SMj , (27)
jxi jxi
becomes
 
jc jc j Dt jc where SMj is the source term representing the momentum
+ Uj = D+ exchange between vapor and liquid phases; and ui uj is the
jt jxj jxj c jxj
Reynolds stress, for which the Boussinisque’s relation is
c jC c applied:
− Cc1 uj c − Cc3 . (24)
c2 jxj k  
jU i jU j 2
In order to testify the validity of simplification, comparison −ui uj = t + −
ij k, (28)
jxj jxi 3
between the simplified and original models were made. The
simplified model by using Eqs. (19) and (20) is referred to where t = C k 2 /. We assume that the liquid volume fraction
as Model I hereinafter, and that with Eq. (24) is referred to L is not varying with the position, and is given by the corre-
as present model or Model II. The use of Eq. (19) and the lation of Bennett et al. (1983):
following equation for c is referred to as original model:  
0.91 
  G
jc jc j Dt jc L = exp −12.55 Us . (29)
+ Uj = D+ L − G
jt jxj jxj c jxj
c jC 2 For the source term in Eq. (27), Wang et al. (2004) considered
− Cc1 uj c − Cc2 c in their model the drag force, lift force, virtual mass force and
c2 jxj c2 body force. However, except for the drag force, which has been
jU i c c employed to interpret the interaction between individual bubble
− Cc3 ui uj − Cc4 and liquid by a number of researchers (Krishna et al., 1999;
jxj k k
Gesit et al., 2003; Wang et al., 2004), there are no common
j2 C consensus in the literatures on the uses of other forces i.e., lift
+ DD t , (25)
jxj jxk force, virtual mass force and body force. Liu et al. (2000) gave
Z.M. Sun et al. / Chemical Engineering Science 62 (2007) 1839 – 1850 1843

a fairly good prediction for a two-dimensional and two-phase where kL and kG are the film coefficients of mass transfer on
flow on distillation tray with only considering the body force liquid side and gas (vapor) side, respectively, m is the coefficient
given previously by Zhang and Yu (1994). Such consideration of distribution between two phases which can be obtained from
is adopted in the present work for the source term in the x and the vapor–liquid equilibrium data.
y coordinates: The simulated result by the proposed model depends on the
choice of mass transfer coefficients and effective vapor–liquid
G Us
SMj = − Uj (j = x, y), (30) interfacial area. A number of correlations developed for kL and
L h f kG can be found from the literatures. Several correlations have
where the froth height is estimated by the correlation hf = been used and checked the simulated results with the experi-
hL /L , in which the clear liquid height hL is calculated by mental data of a commercial scale distillation column reported
AIChE (1958) correlation: by Sakata and Yanagi (1979). It was found that applying the cor-
relations presented by Zuiderweg to calculate the mass trans-
hL = 0.0419 + 0.189hw − 0.0135Fs + 2.45qL / lw (31) fer coefficients for simulating the commercial scale distillation
column concerned gave the least deviation with the experimen-
and the liquid volume fraction L is estimated by Eq. (29). For tal data. It can be understood that the Zeiderweg’s correlations
the source term in the z coordinate, the drag force expressed are based on the data mostly from the commercial columns.
by Krishna et al. (1999) is chosen: The corresponding equations for kL and kG are given below:
(1 − L )3 0.13 0.065
SMz = g( L − G )|UG − UL |(Us − ULz ). (32) kG = − 2 (1.0 < G < 80 kg m−3 ), (38)
Us2 G G
In closing Eq. (27), the following standard k. method is used:  
1
  kL = − 1 mk G , (39)
jk jk j t jk jU i LPR
+ Uj = + − ui uj − , (33)
jt jxj jxj k jxj jxj where LPR is the liquid phase resistance which is 0.37
  (Zuiderweg, 1982), G is the vapor density. The average value
j j j t j of m covering the range of concentration under consideration
+ Uj = +
jt jxj jxj  jxj was found to be 0.0055.
The effective vapor–liquid interfacial area was calculated by
 jU i 2
− C1 ui uj − C 2 . (34) the correlation presented by Zuiderweg (1982).
k jxj k The numerical computation is begun from the top of the
The model parameters are customary chosen to be C = 0.09, column. As only the compositions of reflux and the vapor leav-
C1 = 1.44, C2 = 1.92, k = 1.0,  = 1.3. ing the top are known and the composition of entering vapor to
the top tray is unknown, the following trial-and-error method is
2.2.2. The mass transfer equation used to start the computation. An entering vapor composition is
The equation governing concentration profile of distillation assumed and then the trial value of C ∗ can be obtained, which
tray is is in equilibrium with the average vapor composition between
  entering and leaving. The amount of mass transfer in the top
jL C j jC tray is calculated by Eq. (34). By material balance, the liquid
Uj = L D −  L uj c + S C , (35) composition leaving the top tray can be found, which should
jxj jxj jxj
be equal to the assumed composition of entering vapor under
where SC is the source term for mass transfer between vapor the condition of total reflux. If not, make the trial again until
and liquid phases. The steady form of Eqs. (5), (19) and (20) the error is not more than 2%. For all the trays below, similar
(or Eq. (24)) are used to close Eq. (35), that is to eliminate method are used to obtain the compositions of vapor entering
the unknown mass transfer diffusivity Dt in order to obtain the the tray and the liquid leaving the tray.
concentration profile.
The source term SC in Eq. (35) is commonly known to be 2.2.3. The boundary conditions
The inlet conditions of the present CMT model are: U =
SC = KOL a(C ∗ − C), (36) U in , C = C in and that for the k. equations is followed the
2
where KOL is the overall liquid phase mass transfer coefficient, conventional formulas (Nallasamy, 1987) to be kin = 0.003U xin
∗ 3/2
a is the effective vapor–liquid interfacial area and C is the liq- and in = 0.09kin /(0.03 × W/2).
uid composition in equilibrium with the vapor passing through The inlet conditions of c2 .c equations, deducted by Liu
the liquid layer on the tray. (2003) and Sun et al. (2005), are given below:
The overall liquid phase mass transfer coefficient KOL can
be expressed by the conventional relationship: c2 in = [0.082 · (C in − C ∗ )]2 , (40)
 
1 in
KOL = , (37) cin = R c2 in , (41)
1/kL + 1/mk G kin
1844 Z.M. Sun et al. / Chemical Engineering Science 62 (2007) 1839 – 1850

where R represents the timescale ratio of concentration to ve-


locity and equals to 0.9 as shown in previous section.
At the outlet, we have p = 0, and jC/jx = 0.
The boundary conditions at the tray floor, the outlet weir and
the column wall are considered as non-slip, and the conven-
tional logarithm law expression is employed.
At the interface of the vapor and liquid, all the stresses are
equal to zero, so we have jU x /jz =0, jU y /jz =0, and U z =0.
Similarly, both at the wall and the interface, the concentration
flux is equal to zero.

3. Computational result of CMT model for distillation


column

3.1. Velocity distribution

To assess the validity of the CFD part of the proposed CMT


model, the velocity distribution on a 1.2-m-dia. sieve tray is
simulated for the comparison with experimental data reported
by Solari and Bell (1986). The model geometry and boundaries
are shown in Fig. 1. Solari and Bell (1986) measured the linear
liquid-velocity along two lines perpendicular to the liquid flow
direction on a plane 0.038 m above the tray floor. In the simu-
lated computation, air–water system is used. Figs. 2 and 3 show
the predicted liquid horizontal velocity and the experimental
data of Solari and Bell (1986). From the figures, we can see that
the predictions agree reasonably with the experimental data in
spite of having some deviations. The discrepancy between them
may be due to the following reasons. Firstly, the experimen-
tal work was under the condition of two-phase flow, while the
quasi-single-phase model is used for the simulation. Secondly,
the experiment is one-dimensional, namely the measured ve-
locity is the linear velocity of the tracer dye from one probe
to the next, while the present simulation is three-dimensional,
and the computed liquid phase velocity shown in Figs. 2 and 3
are the velocity component in x direction. Obviously, the com- Fig. 2. Liquid-velocity profile, QL = 6.94 × 10−3 m3 s−1 , FS = 1.015 m s−1
parison between experimental data and prediction is not ex- (kg m−3 )0.5 : (a) upstream profile; (b) downstream profile.
actly on the same basis. Thirdly, the inlet velocity distribu-
tion in present simulation is assumed to be uniform, while the
experimental condition might deviate from such assumption. Liu and Yuan, 2002). The existence of circulating flow can be
Fig. 4(a) and (b) show the liquid-velocity vector plot. It can be explained as follows. When the liquid passes through the inlet,
seen that the velocity is uniform in the main flow area. The cir- the flow area suddenly expands, leading to the separation of the
culating flow is found near the corner of the inlet weir, which boundary layer and forming the eddy current. The circulation
has been observed in many experimental works (Yu and Huang, flow increases the extent of fluid mixing, which is reflected on
1981; Porter et al., 1992; Biddulph, 1994; Yu et al., 1999, the increase of turbulent mass transfer diffusivity Dt as shown
in the later section.

3.2. Turbulent mass transfer diffusivity distribution

As a result of the present CMT simulation, Figs. 5–7 show


the turbulent mass transfer diffusivity profiles, which were
given separately by using the Original Model, Models I and II
(present model) for simulating a commercial scaled distillation
tray operated with cyclohexane-n-heptane system at 165 kPa
and outlet weir liquid load at 0.013 m−3 s−1 m−1 . Since the
Fig. 1. Flow geometry and boundary conditions. turbulent mass transfer diffusivity Dt , which is highly affected
Z.M. Sun et al. / Chemical Engineering Science 62 (2007) 1839 – 1850 1845

Fig. 4. Liquid-velocity vector plot on the x–y plane, z = 0.038 m. (b) Local
view of the circulation area (rectangle area in (a)).

Fig. 3. Liquid-velocity profile, QL = 6.94 × 10−3 m3 s−1 , FS = 1.464 m s−1


(kg m−3 )0.5 : (a) upstream profile; (b) downstream profile.

by the velocity and concentration fields, represents the inten-


sity of back-mixing, the larger local value of Dt corresponds
the lower local mass transfer efficiency. It can be seen from
the figures that the distribution of Dt is quite diverse. If we
take the volume average value of Dt , the order of magnitude
is about 10−2 .10−3 , which is close to those reported in the lit-
eratures (Barker and Self, 1962; Yu et al., 1990, Cai and Chen,
2004). Comparing the three figures, we can see that the shape
of Dt profile obtained by different models are similar, as seen in
Figs. 6 and 7. Fig. 8 shows that the volume average values of Fig. 5. Turbulent mass transfer diffusivity profile at 20 mm above the floor
Dt computed by Models I and II are in good agreement with (Original Model).
the average experimental data for commercial scaled column
reported by Cai and Chen (2004), while the computed results 3.3. Concentration distribution
by using Original Model are much lower. It demonstrates that
the simplified model can give better results than the original one The following computation aims at the simulation of a com-
as far as in predicting the turbulent mass transfer diffusivity is mercial scale distillation column reported by Sakata and Yanagi
concerned. (1979). The separating system is cyclohexane-n-heptane at the
1846 Z.M. Sun et al. / Chemical Engineering Science 62 (2007) 1839 – 1850

Fig. 8. Experimental vs. computational of turbulent mass transfer diffusivity.


Fig. 6. Turbulent mass transfer diffusivity profile at 20 mm above the floor
(Model I).

Fig. 9. Concentration profile of x–y plane on tray 2 at 20 mm above the floor


(Original Model).
Fig. 7. Turbulent mass transfer diffusivity profile at 20 mm above the floor
(Model II).
literature for the comparison. However, we may compare indi-
rectly by means of the outlet concentration of each tray. From
operating pressure of 165 kPa. The liquid rate is 30.66 m3 h−1 Fig. 12, it can be seen that the computed outlet concentration
and vapor rate is 5.75 kg s−1 . More detailed data about the col- of each tray is in good agreement with the experimental mea-
umn and the average physical properties of the systems are surement except for the tray 6. As we understand for the total
available in the literature (Sakata and Yanagi, 1979). The liq- reflux operation, the outlet concentration should form a smooth
uid in the downcomer is assumed to be completely mixed and curve on the plot. The deviation on tray 6 is likely to be due to
the computation followed a tray-by-tray scheme to simulate the experimental error or some other unknown reasons. The aver-
tray cascade. The grids and the coordinates for computation are age deviation of the outlet composition is 3.77%.
shown in Fig. 1. The trays should be numbered 2–9 from the The Murphree efficiency for each tray is also computed and
top of the column, while the reflux is designated as tray 1. compared with experimental data as shown in Fig. 13. Except
As a sample of the computed results, Figs. 9–11 show the for trays 6 and 7, the predicted results are in agreement with
computed concentration distribution on tray 2. It can be seen the measurement. The deviation at trays 6 and 7 is probably
that the concentration profiles computed by the three different coming from using different outlet concentration at tray 6 for
models are similar. Unfortunately, no experimental data on the calculating EMV . The overall tray efficiency can be evaluated
concentration field of a tray is available at the present in the by the Fenske–Underwood equation. The predicted overall tray
Z.M. Sun et al. / Chemical Engineering Science 62 (2007) 1839 – 1850 1847

Fig. 12. Predicted concentration vs. experimental measurement.


Fig. 10. Concentration profile of x–y plane on tray 2 at 20 mm above the
floor (Model I).
adopted for the simulation concerned:

kL = 8DL0.5 , (42)
 0.5
DG
kG = 0.625kL . (43)
DL

From Figs. 14 and 15, the computed bottom concentrations are


found to be somewhat less than the experimental measurements
and the overall tray efficiency is slightly higher. The average
deviation of the bottom concentration is 6.5%. The cause of
discrepancy may be attributed to the ideal operational condi-
tions concerned in the simulation, such as no weeping, no en-
trainment and perfect construction, which an existing column
may not achieve.

4. Conclusion

The original c2 −c model (Liu, 2003), which is used to close


Fig. 11. Concentration profile of x–y plane on tray 2 at 20 mm above the the differential mass transfer equation, is further simplified and
floor (Model II). the model constants are ascertained. An improved CFD equa-
tion is employed to predict the velocity field. To test the valid-
ity of the improvement, the proposed simplified CMT model
efficiency is 83.34% by Original Model, 81.46% by Model I is applied to two distillation columns. The computed results
and 80.68% by Model II, while the experimental measurement are compared with the respective experimental data taken from
is 89.4%. the literatures. The comparison with the experimental data for
To further demonstrate the feasibility of applying the sim- an industrial scale distillation column reported by Sakata and
plified Model II, simulation is also made for the bottom con- Yanagi (1979) reveal that the simplified models can give better
centration and overall tray efficiency of another distillation col- predictions on the turbulent mass transfer diffusivity than the
umn, a pilot-scale distillation column as described by Garcia original one, while the computed concentrations at the outlet of
and Fair (2000), which is 0.429 m in diameter with eight sieve each tray and the tray efficiency by these two models are in sat-
trays of 0.457 m tray spacing operated under total reflux at dif- isfactory agreement. In addition, the comparison is also made

ferent F -factors (Fs = us G ). The separating system is the to a pilot-scale distillation column described by Garcia and
cyclohexane-n-heptane mixture at 165 kPa. As we know, the Fair (2000), the predicted bottom concentration and the overall
KOL is related with the structure and size of the sieve tray. It tray efficiency under different F -factors of a pilot-scale sieve
was found that the correlations of kL and kG by Hoogendoorn tray column are confirmed reasonably with the experimental
et al. (1988) is applicable to the pilot-scale column, and was data. The proposed simplified CMT model has demonstrated
1848 Z.M. Sun et al. / Chemical Engineering Science 62 (2007) 1839 – 1850

Fig. 13. Predicted EMV vs. experimental measurement.

Fig. 15. Overall tray efficiencies under different F -factors (0.429 m column).
Fig. 14. Predicted concentration vs. experimental measurement (0.429 m
column).

to be a prospective tool to predict the turbulent mass transfer C , C1 , turbulence model constants for the velocity field
diffusivity, concentration profile on a tray as well as the tray C 2
efficiency of a distillation column. C instantaneous concentration (mass fraction)
C time average concentration in liquid phase (mass
fraction)
Notation ∗
C time average concentration in liquid phase in
a effective vapor–liquid interfacial area, m2 m−3 equilibrium with concentration in gas phase
c fluctuating concentration (mass fraction) (mass fraction)
c2 concentration variance D molecular mass transfer diffusivity, m2 s−1
Ct , Cc1 , turbulence model constants for the concentration DG vapor-phase molecular mass transfer diffusivity,
Cc2 , Cc3 field m2 s−1
Z.M. Sun et al. / Chemical Engineering Science 62 (2007) 1839 – 1850 1849

DL liquid-phase molecular mass transfer diffusivity, and the assistance by the staffs in the State Key Laboratories
m2 s−1 of Chemical Engineering (Tianjin University).
Dt turbulent mass transfer diffusivity, m2 s−1
References
EMV Murphree efficiency of vapor phase

Fs F -factor (Fs = us G ) AIChE Research Committee, 1958. Bubble Tray Design Manual. AIChE,
g acceleration due to gravitation, m s−2 New York.
hf froth height, m AIChE and U.S. Department of Energy Office of Industrial technologies,
hL clear liquid height, m 1998. Vision 2020: 1998 Separations Roadmap. Center for Waste Reduction
Technologies of AIChE, New York.
hw weir height, m Barker, P.E., Self, M.F., 1962. The evaluation of liquid mixing effects on a
k turbulent kinetic energy, m2 s−2 sieve plate using unsteady and steady state tracer techniques. Chemical
kG vapor-phase mass transfer coefficient, m s−1 Engineering Science 17, 541–553.
kL liquid-phase mass transfer coefficient, m s−1 Bennett, D.L., Rakesh, A., Cook, P.J., 1983. New pressure drop correlation
for sieve tray distillation columns. A.I.Ch.E. Journal 29, 434–442.
KOL overall liquid phase mass transfer coefficient,
Biddulph, M.W., 1994. Mechanisms of recirculating liquid flow on
m s−1 distillation sieve plates. Industrial and Engineering Chemistry Research 33,
lw weir width, m 2706–2711.
Lm Prandtl mixing length, m Cai, T.J., Chen, G.X., 2004. Liquid back-mixing on distillation trays. Industrial
m distribution coefficient and Engineering Chemistry Research 43, 2590–2597.
Colin, O., Benkenida, A., 2003. A new scalar fluctuation model to predict
p time average pressure, Pa mixing in evaporating two-phase flows. Combustion and Flame 134,
P c production term in the c equation 207–227.
qL volumetric flow of liquid flow, m3 s−1 Elghobashi, S.E., Launder, B.E., 1983. Turbulent time scales and the
R timescale ratio dissipation rate of temperature variance in the thermal mixing layer. Physics
of Fluids 26, 2415–2419.
Re Reynolds number
Fischer, C.H., Quarini, J.L., 1998. Three-dimensional heterogeneous
SC source of interphase mass transfer modelling of distillation tray hydraulics. AIChE Meeting, Miami
SC time average source of interphase mass transfer Beach, FL.
SMj source of interphase momentum transfer Garcia, J.A., Fair, J.R., 2000. A fundamental model for the prediction of
t time, s distillation sieve tray efficiency. 1. Database development. Industrial and
Engineering Chemistry Research 39, 1809–1817.
u fluctuating velocity, m s−1 Gesit, G., Nandakumar, K., Chuang, K.T., 2003. CFD modeling of flow
U instantaneous velocity, m s−1 patterns and hydraulics of commercial-scale sieve trays. A.I.Ch.E. Journal
U time average velocity, m s−1 49, 910–924.
Us superficial vapor velocity, m s−1 Hoogendoorn, G.C., Abellon, R.D., Essens, P.J.M., Wesselingh, J.A., 1988.
Desorption of volatile electrolytes in a tray column (sour water stripping).
Greek letters Chemical Engineering Research & Design 66, 483–502.
Krishna, R., van Baten, J.M., Ellenberger, J., Higler, A.P., Taylor, R., 1999.
L liquid volume fraction CFD simulations of sieve tray hydrodynamics. Chemical Engineering

ij Kronecker delta Research & Design, Transactions of the Institute of Chemical Engineers,
 turbulent dissipation, m2 s−3 Part A 77, 639–646.
Launder, B.E., 1976. Heat and mass transport. In: Bradshaw, P. (Ed.),
c dissipation rate of c2 , s−1 Turbulence—Topics in Applied Physics. Springer, Berlin, pp. 232–287.
t turbulent viscosity, m2 s−1 Lemoine, F., Antoine, Y., Wolff, M., Lebouche, M., 2000. Some experimental
density, kg m−3 investigations on the concentration variance and its dissipation rate in
a grid generated turbulent flow. International Journal of Heat and Mass
c , c , turbulence model constants for diffusion of c2 , Transfer 43, 1187–1199.
k ,  c , k,  Liu, B.T., 2003. Study of a new mass transfer model of CFD and its application
 c dissipation term in the c equation on distillation tray. Ph.D. Dissertation, Tianjin University, Tianjin, China.
Liu, C.J., Yuan, X.G., 2002. Computational fluid-dynamics of liquid phase
 time scale, s
flow on distillation column trays. Chinese Journal of Chemical Engineering
 , c timescales of velocity and concentration fields, s 10, 522–528.
m mean time scale, s Liu, C.J., Yuan, X.G., Yu, K.T., Zhu, X.J., 2000. A fluid-dynamics model
for flow pattern on a distillation tray. Chemical Engineering Science 55,
Subscripts 2287–2294.
Lockett, M.J., 1986. Distillation Tray Fundamentals. Cambridge University
G gas Press, Cambridge.
in inlet Mehta, B., Chuang, K.T., Nandakumar, K., 1998. Model for liquid phase flow
i, j, k tensor symbols on sieve trays. Chemical Engineering Research & Design, Transactions of
L liquid the Institute of Chemical Engineers, Part A 76, 843–848.
Nagano, Y., Kim, C., 1988. A two-equation model for heat transport in wall
x, y, z x, y, and z coordinates
turbulent shear flows. Journal of Heat Transfer, Transactions ASME 110,
583–589.
Acknowledgments Nallasamy, M., 1987. Turbulence models and their applications to the
prediction of internal flows. Computers & Fluids 15, 151–194.
Porter, K.E., Yu, K.T., Chambers, S., Zhang, M.Q., 1992. Flow patterns and
The authors wish to acknowledge the financial support by the temperature profiles on a 2.44 m diameter sieve tray. Chemical Engineering
National Natural Science Foundation of China (No. 20136010), Research & Design 70, 489–500.
1850 Z.M. Sun et al. / Chemical Engineering Science 62 (2007) 1839 – 1850

Sakata, M., Yanagi, T., 1979. Performance of a commercial scale sieve Yu, K.T., Huang, J., 1981. Simulation and efficiency of large tray (I)—eddy
tray. Institution of Chemical Engineers Symposium Series, vol. 56, pp. diffusion model with non-uniform liquid velocity field. Huagong Xuebao
3.2/21–3.2/34. 32, 11–19.
Solari, R.B., Bell, R.L., 1986. Fluid flow patterns and velocity distribution Yu, K.T., Huang, J., Li, J.L., Song, H.H., 1990. Two-dimensional flow
on commercial-scale sieve trays. A.I.Ch.E. Journal 32, 640–649. and eddy diffusion on a sieve tray. Chemical Engineering Science 45,
Sun, Z.M., Liu, B.T., Yuan, X.G., Liu, C.J., Yu, K.T., 2005. New turbulent 2901–2906.
model for computational mass transfer and its application to a commercial- Yu, K.T., Yuan, X.G., You, X.Y., Liu, C.J., 1999. Computational fluid-dynamics
scale distillation column. Industrial and Engineering Chemistry Research and experimental verification of two-phase two-dimensional flow on a
44, 4427–4434. sieve column tray. Chemical Engineering Research & Design, Transactions
van Baten, J.M., Krishna, R., 2000. Modelling sieve tray hydraulics using of the Institute of Chemical Engineers, Part A 77, 554–560.
computational fluid dynamics. Chemical Engineering Journal 77, 143–151. Zhang, M.Q., Yu, K.T., 1994. Simulation of two dimensional liquid phase
Wang, X.L., Liu, C.J., Yuan, X.G., Yu, K.T., 2004. Computational fluid flow on a distillation tray. Chinese Journal of Chemical Engineering 2,
dynamics simulation of three-dimensional liquid flow and mass transfer 63–71.
on distillation column trays. Industrial & Engineering Chemistry Research Zhang, Z.Sh., 2002. Turbulence. National Defence Industry Press, Beijing,
43, 2556–2567. p. 258.
Yu, K.T., 1992. Some progress of distillation research and industrial Zuiderweg, F.J., 1982. Sieve tray—a view on the state of the art. Chemical
applications in China. Institution of Chemical Engineers Symposium Engineering Science 37, 1441–1464.
Series, vol. 1, pp. A139–A166.

S-ar putea să vă placă și