You are on page 1of 12

Combustion and Flame 138 (2004) 308319

www.elsevier.com/locate/jnlabr/cnf

A model for predicting the flash point of ternary flammable


solutions of liquid
Horng-Jang Liaw , Chia-Ling Tang, Jim-Shoung Lai
Department of Occupational Safety and Health, China Medical University, Taichung, Taiwan, Republic of China
Received 8 August 2003; received in revised form 27 May 2004; accepted 1 June 2004

Abstract
A mathematical model that may be used to predict the flash point of multiple component solutions has been
proposed and is shown to be adequate for some specified systems, such as ternary solutions. The ternary solutions methanol + methyl acetate + methyl acrylate and methanol + ethanol + acetone were used to validate this
proposed model. The model is able to predict the flash point precisely over the entire composition range of such
ternary solutions by utilizing the flash points of the individual components. If the binary parameters for a ternary
solution are not accessible, a model based upon the binary parameters of binary solutions may provide a very
acceptable means of predicting the flash points for the ternary solution as revealed by a comparison between
predicted and experimental data. A further finding is that the minimum flash point exhibited by a binary highly
nonideal solution may disappear on addition of a specific third component.
2004 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
Keywords: Flash point; Vaporliquid equilibrium; Ternary solution; Binary parameter; Minimum flash point

1. Introduction
Waste materials have to be stored for various periods. To ensure the safe storage of such materials,
flash-point data for flammable liquid solutions are
important. The flash point of a liquid is the temperature (determined experimentally) at which such a substance emits sufficient vapor to form a combustible
mixture with air [1]. This is the temperature at which
the vapor will burn while in contact with an ignition
source, but will not continue to burn after the ignition
source has been removed [1]. The flash point is one
* Corresponding author. Fax: 886-4-22031843.

E-mail address: hjliaw@mail.cmu.edu.tw


(H.-J. Liaw).

of the major physical properties used to determine the


fire and explosion hazards of liquids [2]. The specific
flash point for a substance is generally considered to
be appropriately determined using a flash-point analyzer. There are two methods for measuring the flash
point, the closed cup test and the open cup test [3].
The flash point is one of the major safety data
items specified in a typical material safety data sheet
(MSDS). For a pure substance, the flash point can also
be found in, e.g., the SFPE (Society of Fire Protection
Engineers) handbook [4] or the Merck index [5]. Data
on the flash points for a variety of liquid mixtures are
scarce in the literature. If a model for predicting the
flash points of liquid mixtures could be successfully
developed, the flash points of various liquid solutions
could be predicted relatively easily based on a limited
number of initial basic data.

0010-2180/$ see front matter 2004 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.combustflame.2004.06.002

H.-J. Liaw et al. / Combustion and Flame 138 (2004) 308319

309

Nomenclature
A, B, C
Aij
G
g
LFL
l
MW
N
P
P sat
sat
Pi,fp
q
R
r
T
Ti,fp
u

Antoine coefficients
binary parameter
defined in Table 1
binary parameters of NRTL
equation . . . . . . . . . . . . . . . . . . . . J mol1
lower flammable limit
UNIQUAC parameter, defined in Table 1
molecular weight . . . . . . . . . . . . g mol1
number of data points
ambient pressure . . . . . . . . . . . . . . . . kPa
saturated vapor pressure . . . . . . . . . . kPa
saturated vapor pressure of component i
at flash point . . . . . . . . . . . . . . . . . . . . kPa
measure of molecular surface areas
gas constant, 8.314 J mol1 K1
measure of molecular van der Waals volumes
temperature . . . . . . . . . . . . . . . . . . . . . . . K
flash point temperature of pure
component i . . . . . . . . . . . . . . . . . . . . . . K
binary parameters of UNIQUAC
equation . . . . . . . . . . . . . . . . . . . . J mol1

Crowl and Louvar [2] have suggested that the flash


point of a liquid solution with only one flammable
component, such as a binary aqueousorganic solution, can be estimated using Raoults law. Our previous study [6] showed that Crowl and Louvars method
is only adequate for a composition range where the
flammable component approaches unity. A mathematical model to predict the flash point for a binary
aqueousorganic solution was proposed previously
[6] and was verified as being able to predict the experimental measurements successfully [6].
Affens and McLaren [7] developed a model to predict the flash points of binary hydrocarbon mixtures
using Raoults law. White et al. [8] reduced Affens
and McLarens model to a simpler equation by ignoring any dependence of the lower flammable limit
(LFL) on temperature; they used such an equation to
estimate the flash points of two aviation-fuel mixtures, JP-4/JP-8 and JP-5/JP-8 [8]. However, there
would appear to be some deviation between the measurements and their predictions. We have demonstrated [9] that Affens and McLarens model [7] and
White et al.s equation [8] were only able to explain
the measured flash point of an ideal solution. They
were unable to predict the flash point of a nonideal
solution [9]. Since no model for predicting the flashpoints of solutions with more than one flammable
component has been proposed previously, especially

vl
x
y
z

molar volume of liquid . . . . . m3 mol1


mole fraction of a species in liquid phase
mole fraction of a species in vapor phase
coordination number

Greek letters
ij

NRTL parameter
activity coefficient
area fraction of component i
defined in Table 1
binary parameters of Wilson
equation . . . . . . . . . . . . . . . . . . . . J mol1
density . . . . . . . . . . . . . . . . . . . . . . g cm3
defined in Table 1
segment fraction

Subscripts
exp
fp
i
pred

experimental data
flash point
species i
prediction value

one for predicting the flash point of a nonideal solution, it has been suggested [1,2] that the flash points
of such solutions should be determined using any of
the available test methods. In our previous study [9],
a mathematical model to predict the flash points of
binary liquid solutions with two flammable components was proposed; this was verified as being able to
successfully predict the experimental results for both
ideal and nonideal solutions.
The above models for predicting the flash points of
liquid mixtures are all adequate for binary solutions.
However, the flammable liquid solutions present in
many real situations consist of more than two components. Garland and Malcolm [10] developed a statistical model for predicting the flash point of an
organic acidwater solution, acetic acid + propionic
acid + butyric acid + water. This model is able to
roughly match Garland and Malcolms measured
flash points of such solutions [10]. For a nonideal
solution, the main assumption of Garland and Malcolms model [10], the linear relationship between
the flash point and a mole fraction, appears not to
apply. Thus, it is somewhat questionable to apply Garland and Malcolms model [10] to a nonideal solution.
Since no model currently exists for the flash point of
a mixture with more than two components, there is a
great need to develop such a model for a multicomponent liquid solution. The objective of this manuscript

310

H.-J. Liaw et al. / Combustion and Flame 138 (2004) 308319

is to establish such a model for successfully predicting the flash point of mixtures of flammable components. Such a model has been developed based upon
Le Chateliers rule [11] and the theory of vaporliquid
equilibrium (VLE). It was subsequently verified using
experimental measurements for ternary flammable
liquid solutions using the closed cup test method.
Since methanol, ethanol, and acetone are frequently
used to manufacture semiconductors, a ternary solution of methanol + ethanol + acetone was selected
as one of the examples to validate this model. The
alcoholysis of methyl acetate with methanol is a possible manufacturing route for methyl acrylate [12],
and the ternary solution of methanol + methyl acetate + methyl acrylate is a strongly nonideal solution
[12]; thus, such a ternary solution was also selected
for investigation.

2. Experimental details
The flash-point analyzer (HFP 362-Tag, Walter Herzog GmbH, Germany) was used to measure
the flash points of a variety of ternary liquid mixtures, specifically mixtures of methanol + methyl
acetate + methyl acrylate and methanol + ethanol +
acetone with different compositions. The flash-point
analyzer incorporates control devices to program the
instrument to heat the sample at a specified rate (the
heating rate) to within a temperature range close to
the expected flash point. The flash point is automatically tested with the igniter at specified temperature
intervals (the test interval). If the expected flash point
is lower than or equal to the change temperature, the
value of it is set to be 60 C for the standard ASTM
D56 method [13], a heating rate of 1 C min1 is used
and the igniter is fired at test interval-1. If the expected
flash point is higher, a heating rate of 3 C min1 is
used and the igniter is fired at test interval-2. The
first time that the flash point is tested is at a temperature equivalent to the expected flash point minus
the value of the start test. The flash-point analyzers
heater cuts out as soon as the testing temperature exceeds the sum of the expected flash point plus the
end-of-test value, if the flash point is not determined.
The flash-point analyzer is operated according to a
standard test method, namely ASTM D56 [13], and
the following set of selected parameters: start test
5 C; end of test 20 C; test interval-1 0.5 C; test
interval-2 1.0 C. Both methanol and acetone were
HPLC/Spectro-grade reagents and supplied by the Tedia Co., Inc. (USA); the methyl acetate was obtained
from Lancaster (England); the methyl acrylate from
Acros Organics (USA), and the ethanol from NASA
(USA).

3. Model
3.1. Mathematical formulation
The flash point of any flammable liquid was defined as the temperature at which the vapor pressure
of the liquid is sufficient to produce a concentration of
vapor in the air, corresponding to the lower flammable
limit (LFL) in air [3].
At the lower flammable limit of a gas mixture,
Le Chateliers rule [11] for a flammable multicomponent mixture of vapor + air will be expressed as
 yi
,
1=
(1)
LFLi
where yi is the mole fraction of a flammable substance i in the vapor phase, and LFLi is the lower
flammable limit of the pure component i. From the
definition of the flash point for a pure substance [3],
the LFL of component i, LFLi , is expressed relative
sat ,
to its saturated vapor pressure at the flash point, Pi,fp
as
LFLi =

sat
Pi,fp

(2)
,
P
where P is the ambient pressure. The flash point of
a substance is generally measured under atmosphere
pressure, which can be deemed low enough for a gas
to be ideal and for yi to be that for vaporliquid equilibrium (VLE), so [9]
yi =

xi i Pisat

.
P
Substituting Eqs. (2) and (3) into Eq. (1),
1=

 xi i P sat
i

sat
Pi,fp

(3)

(4)

The saturated vapor pressure of each pure component


i varies with temperature according to the Antoine
equation,
Bi
.
(5)
T + Ci
The vapor pressure of pure liquid i at its flash point
sat , as presented in Eq. (4), can be estimated by subPi,fp
stituting Ti,fp , the flash point of component i, into the
Antoine equation.
The activity coefficients i in Eq. (4), can be estimated using, e.g., the three-suffix Margules equation, the Wilsons equation [14], the NRTL equation
[15], or the UNIQUAC equation [16]. These equations were all used in this study and are listed in
Table 1.
For moderately nonideal binary solutions, the
three-suffix Margules equation is easier to handle
mathematically than the Wilson, NRTL, and

log Pisat = Ai

H.-J. Liaw et al. / Combustion and Flame 138 (2004) 308319

311

Table 1
Some models for activity coefficients of multicomponent systems using only pure-component and binary parameters
Name
Wilson

Activity coefficient for component i


N

N


ln i = ln
xj ij + 1

xk ki

N
k 

xj kj

where



v lj
ij ii
ij = l exp
RT
vi
N


NRTL

j i Gj i xj

ln i =

N


+
Gki xk

N

j

xj Gij
N


N



ij

Gkj xk

xk kj Gkj 

k
N


Gkj xk

where

gij gjj
RT
ln Gij = ij ij

ij =
UNIQUAC

i
z


+ qi ln i + li i
xj lj
xi
2
i
xi
j
N

N


j j i
qi ln
j j i + qi qi
N

j
j
k kj
N

ln i = ln

where

uij ujj
ln ij =
RT
xi ri
i =
N

xk rk
k

i =

xi qi
N


xk qk

z
(ri qi ) (ri 1)
2
z = 10


ln i = (1 xi )2 Ai + 2xi (Bi Ai )
where
N

xj Aij
Ai =
1 xi
li =

Three-suffix Margules

Bi =

j =1
N


j =1

UNIQUAC equations. Unlike the NRTL equation,


Wilsons equation contains only two adjustable parameters and is mathematically simpler to manipulate
than the UNIQUAC equation. Also, Wilsons equation is probably the most useful method for evaluating
the activity coefficient for a strongly nonideal binary
mixture, for which the three-suffix Margules equation
is likely to represent the measurements with much
less success [17]. Wilsons equation, however, is not

xj Aj i
1 xi

applicable to a mixture exhibiting a miscibility gap.


Unlike Wilsons equation, the NRTL and UNIQUAC
equations are applicable to both vaporliquid and
liquidliquid equilibria [17]. While the UNIQUAC
equation is mathematically more complex than the
NRTL equation, it does feature only two parameters
and these parameters often exhibit a smaller dependence on temperature than those of the NRTL and
Wilson equations [17].

312

H.-J. Liaw et al. / Combustion and Flame 138 (2004) 308319

3.2. The model predicting a flash point in the case of


a binary solution
For a binary mixture of flammable liquids, Eq. (4)
becomes
1=

 xi i P sat
i
sat
Pi,fp

x1 1 P1sat
sat
P1,fp

x2 2 P2sat
sat
P2,fp

(6)

Thus, the flash point for such a binary solution can


be estimated using Eqs. (5) and (6) and those listed in
Table 1. These equations constitute the model for a binary solution and are equivalent to the analogues developed and verified for predicting the experimental
measurements for binary flammable liquid solutions
in our previous work [9].

The predictions from Eqs. (9) and (10) were compared with those from Eq. (7) to highlight the need
for a nonideal model.
In this study, a flash-point-prediction model for a
mixture of flammable liquids has been derived; such a
model can be simplified for a binary solution, as proposed previously [9]. Such a reduced form has been
verified to predict the experimental data for various
binary solutions [9,18]. If the model for a multicomponent mixture of flammable liquids can predict the
flash point of ternary solutions, it has the potential
to predict all multicomponent flammable liquid solutions. Thus, the reduced form of such a model simplified for a ternary solution as referred to above was
verified by the experimentally derived data as part of
this study.

3.3. The model for a ternary flammable mixture


Eq. (6) is adequate for a binary liquid solution, but
not for a solution with more than two components.
For a ternary liquid solution, Eq. (4) reduces to
1=

 xi i P sat

sat
sat
i = x1 1 P1 + x2 2 P2
sat
sat
sat
Pi,fp
P1,fp
P2,fp
x3 3 P3sat
+
.
sat
P3,fp

(7)

Therefore, Eqs. (5), (7) and the equations listed in Table 1 constitute the flash-point-prediction model for a
ternary solution. The temperature from the solution of
these equations is deemed to be the flash point of such
a ternary solution.
3.4. The model for ideal solutions
For an ideal solution, Eq. (4) reduces to
1=

 xi P sat
i

sat
Pi,fp

(8)

So the flash point is the solution of Eqs. (5) and (8).


For a binary liquid solution, Eq. (8) becomes
1=

 xi P sat

x1 P sat x2 P sat
i
= sat1 + sat2 ,
sat
Pi,fp
P1,fp
P2,fp

(9)

as developed by White et al. [8] to predict the flash


points of the aviation-fuel mixtures JP-4/JP-8 and
JP-5/JP-8. We demonstrated previously [9] that such
an equation is limited to a binary ideal mixture of liquids.
For a ternary ideal mixture of liquids, Eq. (8) is
1=

 xi P sat

x1 P1sat x2 P2sat x3 P3sat


i
=
sat
sat + P sat + P sat .
Pi,fp
P1,fp
2,fp
3,fp

(10)

4. Results and discussion


4.1. The parameters used to predict the flash point of
ternary solutions
The model for a ternary mixture was used to predict the flash point of the solutions: methanol + methyl acetate + methyl acrylate and methanol + ethanol + acetone. The results obtained were compared
with the corresponding experimentally derived data.
Binary solutions of methanol + methyl acetate, methanol + methyl acrylate, methanol + acetone, and ethanol + acetone are nonideal and the first two solutions both exhibit minimum boiling azeotropic behavior at 101.3 kPa [12,19,20]. By contrast, the
binary solutions of methyl acetate + methyl acrylate and methanol + ethanol almost behave as ideal
solutions [12,21]. The liquid-phase activity coefficients for the mixture of methanol + methyl acetate + methyl acrylate were estimated using the four
different equations in Table 1. The NRTL, Wilson,
and UNIQUAC equations were used to estimate the
activity coefficients of the components for mixtures of
methanol + ethanol + acetone. These estimated activity coefficients were subsequently used in the model
for a ternary solution to predict the corresponding
flash points.
The parameters required for the model for a
ternary solution include the Antoine coefficients and
the parameters of certain equations necessary to estimate the activity coefficients. In addition, it is necessary to input the flash points of the pure components
into such a model to predict the flash point of a mixture. The Antoine coefficients were taken from the
literature [17,22,23] and are listed in Table 2. The parameters of the three-suffix Margules, Wilson, NRTL,
or UNIQUAC equations for these two ternary mixtures were also from the literature [12,20]; they were

H.-J. Liaw et al. / Combustion and Flame 138 (2004) 308319

obtained by regression of the experimental data for


such ternary solutions and are listed in Tables 3 and 4.
The relative van der Waals volume parameter (r) and
the surface-area parameter (q) for the pure components needed for the UNIQUAC equation were also
obtained from the relevant literature [17], and are
listed in Table 5, which also lists the specific volumes
for our pure components, and the formulae in Table 1.
The four equations for estimating activity coefficients for a multicomponent mixture only consider
two-body intermolecular interactions. Thus, such
equations for multicomponent systems require only
binary, i.e., two-body information, and no ternary (or
higher) constants appear [17]. These binary parameters were adopted from the literature [12,1921], and
are listed in Table 6.
The flash points of the pure substances used in this
study were measured with the flash-point analyzer.
Table 7 compares the experimental flash points for the
pure components with the values adopted from the literature. Except for methyl acrylate (provided by the
supplier), the flash points were taken from Merck [5]
and SFPE [4]. The experimentally derived value of
the flash point for ethanol is the same as that from various literature sources [4,5]. There appear to be slight,
but acceptable, deviations between our measurements
and the flash points reported for methanol, methyl
acrylate, and acetone in the literature; there are some
greater deviations for methyl acetate. The flash points
Table 2
Antoine coefficients of the components
Material

Ref.

Methanola

7.20519
16.1295
16.1088
7.24222
16.6513

1581.993
2601.92
2788.43
1595.811
2940.46

33.289
56.15
59.15
46.702
35.93

[22]
[17]
[17]
[23]
[17]

Methyl acetateb
Methyl acrylateb
Ethanola
Acetoneb

a log(P /kPa) = A B/[(T /K) + C].


b ln(P /mmHg) = A B/[(T /K) + C].

313

quoted by SFPE [4] and Merck [5] were measured


by the closed-cup method, although interestingly, the
standard test method is not mentioned in SFPE [4] or
Merck [5]. It is reported in SFPE that the measurement of the flash point depends upon the apparatus
employed [4]. The perceived difference in flash points
between the values applied for this work and the corresponding values reported in the SFPE handbook [4]
and the Merck index [5] might be attributable to existing differences in the standard test method. Moreover, using the definition of flash point [3], values of it
estimated from the vapor pressure using the Antoine
coefficients in Table 2 and the LFL for methanol, acetone, and methyl acetate are 7, 22, and 16 C,
respectively. These values are lower than those listed
in Merck [5] and our experimental data. This observation is consistent with the statement of Britton [24]
and may be due to inherent errors and limitations in
flash point test techniques. From the measured flash
points for methyl acetate, methanol and acetone, it is
suggested that the standard test method used in this
study is presumed to be closer to the vaporliquid
equilibrium assumption than that used to obtain the
data listed in Merck [5].
4.2. Comparison of predicted and measured flash
points
The flash points of ternary liquid mixtures covering the entire composition range of methanol + methyl
acetate + methyl acrylate were tested. Fig. 1 compares the predicted flash points, based upon the binary
parameters of a ternary solution as listed in Table 3,
with the corresponding measured values for binary
solutions of methanol, methyl acetate, and methyl
acrylate. Fig. 2 plots flash points for the same ternary
system at specified mole fractions of methyl acetate.
It is apparent that the predicted flash points do not
change significantly if a different equation is used to
estimate the relevant activity coefficients. Also, these
predicted values are all close to the measured values,

Table 3
Parameters of the Margules, Wilson, NRTL, and UNIQUAC equations for the ternary system containing methanol (1) + methyl
acetate (2) + methyl acrylate (3)
Parametera

Margules [12]

Wilson [12]

NRTL [12]

UNIQUAC [12]

A12
A13
A21
A23
A31
A32
12
13
23

0.92916
1.0477
0.82686
0.22378
1.0539
0.24691

452.06
457.90
99.919
115.45
42.665
88.173

101.12
174.96
227.65
57.685
215.18
83.755
0.27101
0.24841
0.29796

64.283
59.894
326.07
15.777
372.03
7.2038

a Wilson: A = ( )/R; NRTL: A = (g g )/R; UNIQUAC: A = (u u )/R.


ij
ij
ii
ij
ij
jj
ij
ij
jj

314

H.-J. Liaw et al. / Combustion and Flame 138 (2004) 308319

but there is some deviation between them, when the


three-suffix Margules equation is used to estimate an
activity coefficient. Tu et al. [12] and Reid et al. [17]

Table 4
Parameters of the Wilson, NRTL, and UNIQUAC equations
for the ternary system containing methanol (1) + ethanol
(2) + acetone (3)
Parametera

Wilson [20]

NRTL [20]

UNIQUAC [20]

A12
A13
A21
A23
A31
A32
12
13
23

206.34
352.57
186.27
113.82
112.60
58.76

257.99
39.61
361.35
244.98
194.43
61.79
0.300
0.296
0.308

79.98
61.60
112.48
129.00
246.14
42.17

a Wilson: A = ( )/R; NRTL: A = (g


ij
ij
ii
ij
ij
gjj )/R; UNIQUAC: Aij = (uij ujj )/R.

had previously concluded that the NRTL, Wilson, and


UNIQUAC equations were more effective than the
three-suffix Margules equation at describing the activity coefficients for strongly nonideal solutions.
The predicted flash-points using Eqs. (9) and (10),
which assume the liquid phase behaves as an ideal solution, are also plotted in Figs. 1 and 2, respectively.
The plots of Eq. (9) in Fig. 1 are equivalent to those
of White et al. equation [8]. It can be clearly seen
that the predicted flash points based upon White et
al. equation [8] for methanol + methyl acetate and
methanol + methyl acrylate solutions are substantially larger than the corresponding experimental data,
although White et al. equation does describe the measurements for methyl acetate + methyl acrylate well.
It was shown by Tu et al. [12] that the activity coefficients for binary solutions of methanol + methyl
acetate and methanol + methyl acrylate are all larger
than unity, and that for methyl acetate + methyl acrylate system they are close to unity. The deviation

Table 5
The relative van der Waals volumes (r) and surface areas (q) for the pure components for the UNIQUAC model and the specific
volume (vli ) for the pure components for the Wilson model
Component

vli (cm3 mol1 )a

MW [5]

(g cm3 )

r [17]

q [17]

Methanol
Methyl acetate
Methyl acrylate
Ethanol
Acetone

40.73
79.30
90.04
58.68
73.71

32.04
74.08
86.09
46.07
58.08

0.7867 [21]
0.9342 [5]
0.9561 [5]
0.7851 [21]
0.788 [5]

1.4311
2.8042
3.2485
2.1055
2.5735

1.432
2.576
2.904
1.972
2.336

a vl = MW / .
i i
i

Table 6
Binary parameters of the Wilson, NRTL, and UNIQUAC equations for binary systems of methanol, methyl acetate, methyl
acrylate, and binary systems of methanol, ethanol, and acetone
System

Parameter a

Wilson

NRTL

UNIQUAC

Methanol (1) + methyl acetate (2) [12]

A12
A21
12
A12
A21
12
A12
A21
12
A12
A21
12
A12
A21
12
A12
A21
12

451.90
116.30

468.17
49.467

122.29
174.97

68.35 [20]
66.46 [20]

280.508
64.814

96.9963
122.0260

86.237
224.99
0.27101
214.37
164.89
0.24841
3.8691
2.3936
0.29796
0.0 [21]
0.0 [21]
0.0 [21]
140.046
78.317
0.47
230.4142
16.1714
0.3025

71.429
329.34

54.217
358.19

53.603
47.919

162.67 [20]
240.45 [20]

37.323
197.251

73.0960
22.5570

Methanol (1) + methyl acrylate (2) [12]


Methyl acetate (1) + methyl acrylate (2) [12]
Methanol (1) + ethanol (2)
Methanol (1) + acetone (2) [19]
Ethanol (1) + acetone (2) [20]

a Wilson: A = ( )/R; NRTL: A = (g g )/R; UNIQUAC: A = (u u )/R.


ij
ij
ii
ij
ij
jj
ij
ij
jj

H.-J. Liaw et al. / Combustion and Flame 138 (2004) 308319

315

Table 7
A comparison of flash-point values adopted from the literature with experimentally derived data for the solution components used in this studya
Component

Experimental data ( C)

Literature ( C)

Methanol
Methyl acetate
Methyl acrylate
Ethanol
Acetone

10.0 0.4
14.4 0.3
2.1 0.4
13.0 0.3
18.9 0.2

12 [4,5]
10 [5]
3b
13 [4,5]
18 [4,5]

a Closed cup test.


b Provided by Acros Organics.

of our experimental measurements from the predictions of White et al. [8] for methanol + methyl acetate
and methanol + methyl acrylate solutions arises from
both binary solutions showing a positive deviation
from ideality; such behavior results in a reduction of
the solutions flash point from the value predicted for
an ideal solution. Ideal behavior, which is the main
assumption of White et al. [8], is valid for mixtures
of methyl acetate + methyl acrylate. Hence, there is
good agreement between the measured flash points
and the predictions from White et al. equation [8] for
such a solution. Fig. 2 demonstrated that the predictions from Eq. (10), which assumes ideal behavior,
deviates from the measurements of the ternary solution methanol + methyl acetate + methyl acrylate.
The activity coefficients for this solution were indicated by Tu et al. [12] to deviate significantly from
unity. Thus, Eq. (10) is not valid for such a nonideal
solution.
The flash points for the other ternary mixture,
methanol + ethanol + acetone, were computed. They
are compared with the experimental data in Figs. 3
and 4, which display good agreement, no matter
which equation, NRTL, Wilson, or UNIQUAC, is
used to estimate the relevant activity coefficients.
As was the other ternary mixture, the predictions
based upon ideal behavior were not satisfactory (see
Fig. 4). An analogous deviation for binary solutions
of methanol + acetone and ethanol + acetone is also
observed in Fig. 3. The reason for such substantial deviations is that the activity coefficients for these mixtures deviate notably from unity [12,20]. However,
mixtures of methanol + ethanol, as shown in Fig. 3
behave ideally [21].
Table 8 shows that there is no significant difference in the predictive capability when any one of the
NRTL, Wilson, or UNIQUAC equations is used to
estimate activity coefficients; the differences in flash
point are very small. Again, the three-suffix Margules
equation is not as good as the other three equations for
estimating activity coefficients. In general, the worst
predictions of flash point assume ideal solution behavior.

Fig. 1. Comparison of the flash-point-prediction curves


based upon binary parameters of a ternary solution with experimentally derived data for binary solutions of methanol
(1), methyl acetate (2), and methyl acrylate (3). For methanol
(1) + methyl acetate (2), the x-axis represents x2 (composition of methyl acetate), and for methanol (1) + methyl acrylate (3) and methyl acetate (2) + methyl acrylate (3) solutions, the x-axis represents x3 (composition of methyl acrylate).

Fig. 2. Comparison of the flash-point-prediction curves


based upon binary parameters of a ternary solution with experimentally derived data for methanol (1) + methyl acetate
(2) + methyl acrylate (3) solution.

The above computations assumed that the liquid


and vapor phases of a solution are in equilibrium.
Such an assumption holds for measurements by the
closed cup test method, but not from the open cup test
method, when conditions deviate from the assumption of vaporliquid equilibrium and presume the existence of a concentration gradient in the vapor.

316

H.-J. Liaw et al. / Combustion and Flame 138 (2004) 308319

Table 8
Deviation between calculated and experimental flash point Tfp a for binary and ternary solutions of methanol, methyl acetate,
and methyl acrylate and the analogues of methanol, ethanol, and acetone
Solution

Ideal solution

Margules

Wilson

NRTL

UNIQUAC

Methanol + methyl acetate + methyl acrylate

3.034

0.579b

0.157b
(0.247)c

0.177b
(0.204)c

Methanol + methyl acetate

3.460

0.726b

Methanol + methyl acrylate

3.748

0.743b

Methyl acetate + methyl acrylate

0.114

0.587b

Methanol + ethanol + acetone

2.977

Methanol + ethanol

0.117

Methanol + acetone

2.508

Ethanol + acetone

2.421

0.122b
(0.217)c
0.191b
(0.185)c
0.171b
(0.142)c
0.278b
(0.210)c
0.134b
(0.116)c
0.225b
(0.160)c
0.464b
(0.136)c

0.200b
(0.136)c
0.105b
(0.122)c
0.193b
(0.114)c
0.315b
(0.247)c
0.108b
(0.117)c
0.223b
(0.161)c
0.677b
(0.183)c

0.159b
(0.185)c
0.115b
(0.156)c
0.160b
(0.154)c
0.207b
(0.116)c
0.342b
(0.266)c
0.163b
(0.109)c
0.184b
(0.131)c
0.738b
(0.139)c

a Deviation of flash point: T =


fp

|Tfp,exp Tfp,pred |/N .

N
b Based upon binary parameters of a ternary solution.
c Based upon binary parameters of binary solutions.

Fig. 3. Comparison of the flash-point-prediction curves


based upon binary parameters of a ternary solution with experimentally derived data for binary solutions of methanol
(1), ethanol (2), and acetone (3). For methanol (1) + ethanol
(2) and methanol (1) + acetone (3) solutions, the x-axis
represents x1 (composition of methanol), and for ethanol
(2) + acetone (3), the x-axis represents x3 (composition of
acetone).

4.3. The predictive results for ternary solutions


when binary parameters of binary solutions are used
The flash points for the two ternary mixtures
calculated using parameters for binary solutions,

Fig. 4. Comparison of the flash-point-prediction curves


based upon binary parameters of a ternary solution with experimentally derived data for methanol (1) + ethanol (2) +
acetone (3) solution.

as listed in Table 6, are compared with the corresponding experimental data in Figs. 58. It is apparent that these predicted flash points are very
similar to the values generated using any one of
the NRTL, Wilson, or UNIQUAC equations to estimate an activity coefficient. These predictions are all
in good agreement with the experimentally derived
data.

H.-J. Liaw et al. / Combustion and Flame 138 (2004) 308319

Fig. 5. Comparison of the flash-point-prediction curves


based upon binary parameters of binary solutions with experimentally derived data for binary solutions of methanol
(1), methyl acetate (2) and methyl acrylate (3). For methanol
(1) + methyl acetate (2), the x-axis represents x2 (composition of methyl acetate), and for methanol (1) + methyl acrylate (3) and methyl acetate (2) + methyl acrylate (3) solutions, the x-axis represents x3 (composition of methyl acrylate).

Fig. 6. Comparison of the flash-point-prediction curves


based upon binary parameters of binary solutions with experimentally derived data for methanol (1) + methyl acetate
(2) + methyl acrylate (3) solution.

Figs. 1, 3, 5, and 7 and Table 8 indicate that the


predictions based upon the binary parameters in Table 6 are in slightly better agreement with the measurements than based on data listed in Tables 3 and 4.
This may be attributed to the fact that the binary parameters listed in Table 6, which were obtained by fit-

317

Fig. 7. Comparison of the flash-point-prediction curves


based upon binary parameters of binary solutions with experimentally derived data for binary solutions of methanol
(1), ethanol (2), and acetone (3). For methanol (1) + ethanol
(2) and methanol (1) + acetone (3) solutions, the x-axis
represents x1 (composition of methanol), and for ethanol
(2) + acetone (3), the x-axis represents x3 (composition of
acetone).

Fig. 8. Comparison of the flash-point-prediction curves


based upon binary parameters of binary solutions with experimentally derived data for methanol (1) + ethanol (2) +
acetone (3) solution.

ting experimental data for binary solutions, are more


adequate for a binary solution than the more general
ones based on Tables 3 and 4, which were obtained
by fitting with a ternary solution.
Table 8 indicates that the flash points computed
from binary parameters of binary solutions are almost
as good as those calculated using binary parameters

318

H.-J. Liaw et al. / Combustion and Flame 138 (2004) 308319

of a ternary solution. Further, a model based upon


such binary parameters may provide a very acceptable
means of predicting flash points for ternary mixtures
if the binary parameters of such a ternary solution are
not accessible.
4.4. Disappearance of the minimum flash point for a
binary mixture
It was indicated in a previous study [18] that a binary solution of methanol + methyl acrylate exhibits
a minimum flash point, i.e., less than those of the pure
components. The experimental value of the minimum
flash point is 3 C at a mole fraction of methyl acrylate of 0.8. The flash points of methanol and methyl
acrylate are 10 and 2 C, respectively. The measurements for this solution are displayed in Figs. 1
and 2. Fig. 2 shows that the minimum flash point for
methanol + methyl acrylate becomes less conspicuous when a third component, methyl acetate, is added.
If the mole fraction of methyl acetate is increased to
0.3, the minimum flash point becomes significantly
less and disappears, when the mole fraction is increased to more than 0.4. It is suggested that the
minimum flash point for methanol + methyl acrylate
can be eliminated by adding sufficient methyl acetate.
This observation may be attributed to two factors.
One is that the flash point of methyl acetate is much
lower than for methanol or methyl acrylate. Hence,
the flash point of such a solution will be decreased by
adding methyl acetate, if the solution is ideal [9]. The
other is that the ternary mixture of methanol + methyl
acetate + methyl acrylate deviates positively from an
ideal solution [12]; thus the flash point will be lower
than that predicted for an ideal solution [9]. It is suggested that not only can the minimum flash point
be potentially eliminated, but also that the minimum
value of that flash point may be increased by adding
some specific third component. The flash point of the
third component should be higher than those of the
original two components. Also, the addition of such a
third component to the original solution should not
produce a mixture with a highly positive deviation
from ideality. However, this suggestion needs experimental verification. It had been concluded previously
[9] that mixtures with a minimum flash point appear
to be more hazardous than the individual components
alone. If such a suggestion is true, it may be possible
to reduce the fire and explosion hazard for a mixture
with a minimum flash point by addition of a specific
third component to such a solution.
5. Conclusions
The above model accurately predicts the flash
points for ternary mixtures. The reduced form of

such a model for a binary solution is equivalent to


the model proposed previously [9], which predicts
the flash points of binary solutions [9,18]. This new
model can predict the flash points of multicomponent
solutions.
The predictions based on the binary parameters of
a binary instead of a ternary solution, agree well with
experiment. Since the former parameters are relatively easy to access, the flash-point-prediction model
proposed has the potential of being applied in real situations widely.
Another important finding is that the minimum
flash point of a highly nonideal mixture of methanol +
methyl acrylate disappeared by adding the third component, methyl acetate. Such a finding can reduce the
hazards of a mixture with a minimum flash point.
Acknowledgments
The authors thank the China Medical University for supporting this study financially under grant
#CMC-90-OSH-02.
References
[1] CCPS/AIChE, Guidelines for Engineering Design for
Process Safety, American Institute of Chemical Engineers, New York, 1993, p. 529.
[2] D.A. Crowl, J.F. Louvar, Chemical Process Safety:
Fundamentals with Applications, Prentice Hall, Englewood Cliff, NJ, 2002, pp. 230232.
[3] F.P. Lees, Loss Prevention in the Process Industries,
vol. 1, second ed., ButterworthHeinemann, Oxford,
UK, 1996, p. 16.
[4] SFPE, The SFPE Handbook of Fire Protection Engineering, second ed., Society of Fire Protection Engineers, Boston, 1995, pp. 2-1642-165.
[5] Merck, The Merck Index, twelfth ed., Merck, 1996, pp.
1018, 1029.
[6] H.-J. Liaw, Y.-Y. Chiu, J. Hazard. Mater. 101 (2) (2003)
83106.
[7] W.A. Affens, G.W. McLaren, J. Chem. Eng. Data 17
(1972) 482488.
[8] D. White, C.L. Beyler, C. Fulper, J. Leonard, Fire
Safety J. 28 (1997) 131.
[9] H.-J. Liaw, Y.-H. Lee, C.-L. Tang, H.-H. Hsu, J.-H. Liu,
J. Loss Prevent. Proc. 15 (2002) 429438.
[10] R.W. Garland, M.O. Malcolm, Process Safe. Progr. 21
(2002) 254260.
[11] H. Le Chatelier, Ann. Mines 19 (1891) 388395.
[12] C.H. Tu, Y.S. Wu, T.L. Liu, Fluid Phase Equilibr. 135
(1997) 97108.
[13] ASTM D56, Standard Test Method for Flash Point by
Tag Closed Tester, American Society for Testing and
Materials, West Conshohocken, PA, 2001.
[14] G.M. Wilson, J. Am. Chem. Soc. 86 (1964) 127130.
[15] H. Renon, J.M. Prausnitz, AIChE J. 14 (1968) 135
144.

H.-J. Liaw et al. / Combustion and Flame 138 (2004) 308319


[16] D.S. Abrams, J.M. Prausnitz, AIChE J. 21 (1975) 116
128.
[17] R.C. Reid, J.M. Prausnitz, T.K. Sherwood, The Properties of Gases and Liquids, third ed., McGrawHill,
New York, 1977, p. 299.
[18] H.-J. Liaw, T.-P. Lee, J.-S. Tsai, W.-H. Hsiao, M.-H.
Chen, T.-T. Hsu, J. Loss Prevent. Proc. 16 (2003) 173
186.
[19] M.C. Iliuta, F.C. Thyrion, Fluid Phase Equilibr. 103
(1995) 257284.

319

[20] J. Gmehling, U. Onken, VaporLiquid Equilibrium


Data Collection, vol. 1, part 2a, Dechema, Frankfurt,
1977, pp. 328, 617.
[21] K. Kurihara, M. Nakamichi, K. Kojima, J. Chem. Eng.
Data 38 (1993) 446449.
[22] A. Arce, A. Blanco, A. Soto, I. Vidal, Fluid Phase Equilibr. 128 (1997) 261270.
[23] T. Hiaki, K. Tatsuhana, T. Tsuji, M. Hongo, J. Chem.
Eng. Data 44 (1999) 323327.
[24] L.G. Britton, Process Safe. Prog. 21 (2002) 111.

You might also like