Computers them. Engng,Vol. 17, No. 9, pp. 879-884, 1993 0098-I 354/93 $6.00 + 0.

00
Printed in Great Britain. All rights reserved Copyright 0 1993 Pergamon Press Ltd

MULTIPLE LIQUID-VAPOUR EQUILIBRIUM

FLASH CALCULATION BY A GLOBAL
APPROACH

E. ECKERT and M. Kursfikc

Department of Chemical Engineering and Department of Mathematics, Prague Institute of Chemical
Technology, 166 28 Prague 6. Czech Republic

(Received 10 April 1992;finaI revision received 2 October 1992; received for publication 8 January 1993)

Abstract-A global approach to the calculation of nonadiabatic-isobaric multiple liquid-vapour equi-

librium flash is suggested. The structure of the system of equations is adaptively changed with respect to
suggested engineering-oriented rules. The resulting set of equations is solved by the Newton-Raphson
method. Results of two examples (equilibrium mixtures of the type G-L-L and G-L-L-L) show the
efficiency of the algorithm.

1. INTRODUCTION
A global Newton-Raphson method was rec-
The appearance of more liquid phases is generally an ommended many times for simulation of rectification
undesirable phenomenon in distillation processes; columns where only one liquid phase occurs, e.g.
however, it is a reality in practice. The problem of Kubieek et al. (1976). Eckert (1989b) also suggested
three-phase distillation has been discussed in the the use of the global approach for three-phase rectifi-
literature, a good survey can be found in the paper by cation columns with a variable dimension of the
Cairns and Furzer (1990). The prediction of the problem depending on the number of existing phases.
appearance of more liquid phases by using a minimiz- A modified Nelson’s algorithm is used to determine
ation of Gibbs energy (e.g. Gautam and Seider, 1979; the number of coexisting liquid phases on individual
Michelsen, 1982) seems to be superior to fugacity trays after each Newton-Raphson iteration.
matching because the latter condition does not guar- In this paper we present an algorithm for nonadi-
antee that the solution possesses a global minimum. abatic-isobaric flash calculations where the dimen-
Ohanomah and Thompson (1984) performed a com- sion (the number of equations) remains constant.
prehensive study of several techniques and found However, the structure of equations is changed adap-
the Gibbs energy miminization method not to be tively dependent on the number of coexisting phases.
competitive for multiphase equilibrium calculations_ The algorithm is formulated for an arbitrary number
Nelson (1987) suggested an algorithm for isother- of coexisting liquid phases.
mal-isobaric multiple-phase flash calculation based
2. THE ISOBARlC FLASH
on two nested iteration loops. Ratios of phases and
their number are corrected in the inner iteration loop, Consider the equipment schematically drawn in
compositions of the phases are iterated in the outer Fig. 1, usually called the isobaric flash separation
loop. Poorly chosen initial guesses can produce phys- unit. Assume that the feed specifications (feed flows
ically unacceptable situations in the course of iter- Fkr compositions z,,+, . . , zNc and enthalpies I+*,
ations. Some improvement of this algorithm in these k=l,... , m), heat supply (Q) and pressure in the
cases was suggested by Eckert (1989a). Recently Biinz equipment are given. The steady-state operation of
et al. (1991) used Nelson’s method for the calculation the unit can be described by the following equations:
of high-pressure ternary phase diagrams of two three-
(i) equilibrium equations for p liquid phases and
component mixtures. A modification of this method
a vapour phase, r is the index of a chosen
for a nonadiabatic flash needs an outer iteration loop
reference liquid phase:
also for temperature, which may cause convergence
problems. We found no quotations of an example O=y,--K:xj, i=l,...,n, (la)
using a generalization of Nelson’s method for the
O=K~x~-K{x+, i=l,...,n;
problems with the possible occurrence of more than
two coexisting liquid phases. j=l,...,p,j#r; (lb)

879
880 E. Ecxsar and M. KUB~CEK

A further p + 1 necessary equations have to be

5
I G chosen from the following 2p + 2 equations:

%7-
Q O=G or 0=1-gyi, (6)
---_- i= I

im ... o=Lj or 0=1-g&

i= I
j=l,...,p, (7)

ti L? LP depending on the phase situation (i.e. on the number

of coexisting liquid phases p and the existence of
Fig. 1. Multiple liquid-vapourflash unit.
vapour phase). This choice is demonstrated in Table I
for a possible coexistence of two liquid phases
(p = 2). Of course, the number of cases increases
(ii) component balances:
rapidly with p, generally, it equals to:

(p + 1) ! c [(p + 1 -j)!j !I--‘.

j- I

(iii) enthalpy balance As a result we obtain a system of

0=1- Gh,+ 5 LJhJ, Iv = (n + l)(p + 1) + 1

it fih,+Q .
J-1 k-1 >
nonlinear equations:
(3)

Assuming that a component I does not condense or 0 = V(X), (8)

evaporate, the Zth equation in (la) is substituted by:
for Nunknowns y=(y,, . . . ,y,), [xj=(xj,. . ..x<).
0 =x; (4)
j=l,_.. ,p], G, L’, L2,. . . , Lp, T, denoted in a
or vector form as X. The composition of a nonexisting
phase is fictitious and does not fulfill the correspond-
o=y,, (5)
ing summation condition in (6) or (7). Instead,
respectively. equation 0 = G or 0 = LJ applies, see Table 1.

Table 1. Choice of additional equations for possible occurrence of two liquid phases @ = 2)
Existing phase(s) Case Equations Conditions
Single vapour G O=L’. O=LZ cx: -=z1. Exx:<I,
O=l-Cy, G>O

Single liquid I LI O-G. O=L’ ~xf<l orx’=x~.

o=I-~x: EY,< 1, LIZ0

Single liquid 2 L2 O-G. O=L’ xx; < 1 or x’ = XL,

o= t-xx: CY#C 1, L’>0

Liquids I and 2 Ll-L2 O=G, O=l-xx; cJJ,< 1, L’>O,

O=l-xx: #c*>o

Vapour and liquid 2 C-L2 O=L’. O=l-_cy, ~x)cl orYP=x*,

o=l-~x: L’>O. G z-0

Vapour and liquid 1 GLI o-L2, O=l-xy, xx:< 1 or x’=x*,

o- 1-zx: L’>O, G>O

Vapour, liquids I and 2 G-Ll-L2 0=1-x)‘, L’>O, L=>o,

OPI-~X: G>O

O-I-~X:
 Multiple liquid-vapour equilibrium flash calculation 881

An additional remark has to be made to the equations (l-3) and p + 1 equations from the set
problem presented. The subsystem (1) has always a (6-7). We used a Newton-Raphson method without
solution where: any decomposition of the system, i.e. we used the
so-called global approach. For the system (8) we thus
xi’ = xf, i=l,...,n, (9)
have:
for some t,s = I,... .p. t -c s. We shall denote such
X” + ’ = Xk - I [cp‘(Xk)] - ‘cp(XL). (10)
a solution as a trivial (parasitic) solulion. We can
accept it only if at most one value from L’, L” is The N x N Jacobi matrix q’(X) can be computed in
positive (i.e. nonzero). When L’> 0, L’ z- 0 and (9) each iteration by using simple two-point difference
are valid then the Jacobi matrix of the system (8) is formulas. An analytical method of evaluating partial
singular and the solution is not unique [the system (8) derivatives is also possible but the formulas for K{
is infinitely satisfied by many values of L’, L” with the and hJ,may be complicated. The value E. = 1 is chosen
same sum L’ + L”]. This singular case can be trans- when:
formed into a nonsingular one by the following
operations: L’ + L’+ L’, 0 -+ L’ (phase s does not IIV(X*+‘)/I < JIcp(Xk)ll (11)
really exist). By repeating these operations for every
and a standard version of the Newton-Raphson
t and s satisfying (9) and L’ > 0, L” > 0 we finally
method results. When the condition (11) is not fuhi-
reach a solution where L' . L” = 0 (only such nonsin-
lled, the value of 1 is halved and Xk + 1 reevaluated
gular solutions are considered in Table 1). The ques-
(dumped method).
tion of whether there exists another nonsingular case
A very important problem is the choice of p + 1
with a higher number of really (physically) existing
equations from (67) which corresponds to an ex-
liquid phases remains open.
pected set of coexisting phases. As we do not know
this real “phase situation” beforehand, we developed
3. THE ALGORITHM
an adaptive algorithm which enables us to switch to
An iteration technique must be used for solving the another “phase situation” in the course of the iter-
system of N nonlinear equations which consists of ation process (always after several, e.g. two, iteration

Table 2. Decision diagram for possible switching between different cases, p = 2. In mws 14 the conditions in the
last column are tested first. In
the last row conditions in the first three columns are tested first. Marks /. \: in
the algorithm realization S/l means e.g. S > 0.99999, S\O means S < 0

To case
Fr0lll
case G LI L2 LI-L2 G-L2 G-L1 G-LI-L2

G *

Ll

L2

LI-L2

G-L2 LV.0 -4

GLI L’\O *
G’xO x.x:/1
x’ zx=
*
G-LI-L2 L’/O LWJ LYO

L\O G\O
or or

G’._O x’ =x2
882 E. ECKERTand M. KUB~EEK

Table 4. Specifications of Example I

steps). The algorithm can be described by means of
Feed comwsition (a) Ibb
a decision diagram. Such a diagram is presented
1. Butanol 0.15 0.05
schematically in Table 2 for p = 2, i.e. for possible 2. Water 0.80 0.65
existence of two liquid phases. After switching to 3. Propanol 0.05 0.30
Feed Rowrate (mol s ‘) 1 I
a new case it is useful to correct the vector of Feed temperature (K) 367 367
unknowns so that it fulfills the conditions required
for the new case (summation conditions by normal-
ization. zero flowrates for nonexisting phases). Blank EXAMPLE 1
boxes in Table 2 correspond to nongeneric switches
Consider the flash unit (condenser) with a pressure
which have no practical sense for the algorithm.
of 101,325 Pa, the single feed is a vapour mixture of
Conditions written in the decision diagram, cf. Table
butanol, water and propanol. Feed flowrate, tem-
2, are based on engineering intuition and for one-
perature and composition are presented in Table 4.
phase cases arc identical to conditions considered by
Such a mixture can form one vapour and two liquid
Nelson (1987).
phases (p = 2) as was shown by Block and Hegner
There is a basic problem, namely the choice of
(1976). Activity coefficients y, are evaluated by using
initial approximation for the Newton-Raphson
the NRTL equation (e.g. Reid et al., 1977). Binary
method. Not only for the unknowns but also for the
interaction coefficients u,, and Ag,,fR are listed in
initial “phase situation”. Our experience leads to a
Table 3. These values are taken or derived from the
conclusion that the case with the highest number of
paper by Block and Hegner (1976) for a temperature
expected phases serves as a good initial guess, see the
of 364 K.
examples below.
As an initial approximation the following values
have been used:
4. EXAMPLES
G = L ’ = L’ = (ZF)/3 [ = (CF)/(p + l)];
Calculation of physical properties
yi = xj = l/3 (= l/n), i = 1, 2,3;
The values of K, h, and h, depend generally on
composition, temperature and pressure. We used for x:=x:-o, x:= 1; T = TF
K-values the relation:
This initial guess led to convergence to a correct

(12) (physically acceptable) solution for both examples (a)

and (b) from Table 4 and for an arbitrary value of Q
where yi is the activity coefficient and the saturated chosen from the interval (- 46,000 W, 1000 W). The
vapour pressure Pp is given by the Antoine equation. results of computations in dependence on Q are
The molar enthalpy of vapour mixture has been presented for example (a) in Fig. 2. The phase
evaluated according to: specification of the outlet mixture (the case from

s T Table 1) varies depending on the value of Q as we can

b(T) = li 1Y,
;.. To
Cdl) dt, (13) see from the figure. A similar dependence has also
been computed for example (b); however, there exist
and molar enthalpy of liquid mixture analogously: only cases L 1, G-L 1 and G in the whole range of Q.

h,(T) = 5
#=I
xi
s =Cpi(t) dt
m

-hwMT - TciMTe, - Tc~)I~.~~. (14)

Consider
101,325
EXAMPLE 2

the flash unit (reboiler)

Pa with three feed streams (see Table 5). Such
with a pressure of

a mixture can form three liquid phases (p = 3) as was

Here C, is ideal gas heat capacity, hvi is heat of
shown by Null (1970). Activity coefficients yi are
vaporization at normal boiling point TB, and T,-, is
evaluated by using the Van Laar equation in the form
critical temperature, T,, = 298.15 K. All necessary
and with values of parameters given in the book by
data for examples presented below can be found in
the book of Reid et al. (1977).
Table 5. Specification of Example 2
Feed I 2 3

Table 3. NRTL narametersfor Exam& I Feed composition

I. Nitromethance I 0 0
i @,I % % 4% IR 4?lZlR &,lR 2. Dodecanol 0 I 0
I 0 0.48 0.30 0 327.19 260.79 3. Ethylene glycol 0 0
2 0.48 0 0.30 1278.85 0 998.34 Feed flow-ate (mol s-‘) 0.3 0.3 A.4
3 0.30 0.30 0 -223.0 - 26.024 0 Feed temperature(K) 295.2 295.2 295.2
 Multiple liquid-vapour equilibrium flash calculation 883

0.6- L2

- 40000 - 30000 - 20000 -10000 Q 6

Fig. 2. Liquid 1, liquid 2 and gas flowrate dependent on heat supply [Example 1, column (a) from Table 41.

0.5- Ll

0.4- L2

0.3-

0.2-

L3

O.l-

CASE CASE
Ll-L2-L3 G-U-L2
0 ’

f5000 2.0000 25000 30000 61 35000

Fig. 3. Liquids 1-3 and gas flowrate dependent on the heat supply (Example 2).

Null (1970), pp. 216-219. The value of h, for dode- Table 2) has 15 rows and 15 columns and contains
canal, h, = 58,740 J mol-r, was estimated from 100 possible switches (nonblank boxes).
Fishtine’s modification of the Kistiakowski equation
(Reid and Sherwood, 1966).
5. CONCLUSIONS
We used the following initial approximation:
G = 0, L’ = L2 = L3 = l/3, y = (O,O, O), x’ = (0.005, The suggested computational algorithm is rela-
0.005, 0.99) x2 = (0.005,0.99,0.005), x3 = (0.99, tively simple, robust and not very sensitive to initial
0.005,0.005), T = TF = 295.2. This initial guess led to approximation. The convergence is generally fast, i.e.
convergence to a correct solution for an arbitrary only several (4-8) iterations usually lead to a solution.
value of Q E (0 W, 35,000 W). Computed flowrates of The algorithm can also be used for modelling of
products dependent on the value of Q are presented rectification columns, because each stage can be
in Fig. 3. It is obvious that the results for higher modelled as a flash.
values of Q, i.e. for higher temperatures than used by
Null (1970) can be far from reality. Acknowledgemenr-Part of this work was supported by
Alexander van Humboldt Foundation during the stay of
Let us mention that there are 15 cases in this one of the authors (EE) at the Institute for System
example. Therefore, the decision diagram (cf. Dynamics and Control, University of Stuttgart, Germany.
884 E. ECKERT and M. KUB~I~EK

NOMENCLATURE Bilnz A. P., R. Dohm and J. M. Prausnitz, Three-phase

flash calculations for multicomponent systems. Corn-
F= Molar Rowrate of feed to the flash puters them. Engng 15, 47 (1991).
G = Molar flowrate of vapour phase Cairns B. P. and 1. A. Furzer, Multicomponent three-phase
h = Molar enthalpy azeotropic distillation. 2. Phase-stability and phase-
K = Vapour-liquid distribution coefficient splitting algorithms. Ind. Engng Chem. Res. 29, 1364
L = Molar flowrate of liquid phase (1990).
m = Number of feed streams Eckert E., Calculation of rectification with possible occur-
n = Number of components rence of two liquid phases I. Equilibrium stage. Chem.
4 = Number of possible liquid phases Ind. (Prague) 39, 561 (1989a) (in Czech.).
P= Pressure Eckert E., Calculation of rectification with possible occur-
Q = Heat supply rence of two liquid phases II. Stage column. C/rem. fnd.
T = Temperature (Prague) 39, 619 (1989b) (in Czech.).
X = Vector of unknowns, cf. equation (8) Gautam R. and W. D. Seider, Computation of phase and
x = Mole fraction in liquid phase chemical equilibria: part I. Local and constrained minima
y = Mole fraction in vapour phase in Gibbs free energy; part II. Phase-splitting. AIChE Jl
z = Mole fraction in the feed 25, 991, 999 (1979).
y = Activity coelhcient KubiEek M.. V. Hlav%ek and F. ProchPska, Global modu-
lar Newton-Raphson technique for simulation of an
Subscripts interconnected plant applied to complex rectification
F=Feed columns. Chem. Engng Sci. 31, 277 (1976).
G = Vapour phase Michelsen M. L., The isothermal flash problem (parts I and
i = Component i II). Fluid Phase Equil. 9, I (1982).
k = Feed stream k Nelson P. A., Rapid phase determination in multiple-
L = Liquid phase phase flash calculations. Computers them. Engng 11, 581
(1987).
Superscripts Null H. R., Phase Equilibrium in Process Design. Wiley,
New York (1970).
j = Liquid phase j
Ohanomah M. 0. and D. W. Thompson, Computation of
r = Reference liquid phase
multi-component phase equilibria-part III. Multi-phase
equilibria. Computers them. Engng It, 163 (1984).
REFERENCES Reid R. C. and T. K. Sherwood. The Prozwrties ofGases and
Liquids. McGraw-Hill, New ‘York (1466). _
Block U. and B. Hegner, Development and application of Reid R. C., J. M. Prausnitz and T. K. Sherwood, The
a simulation model for three-phase distillation. AIChE JI Properties of Gases and Liquids. McGraw-Hill, New York
22, 382 (1976). (1977).