0957–5820/06/$30.00+0.

00
# 2006 Institution of Chemical Engineers
www.icheme.org/journals Trans IChemE, Part B, January 2006
doi: 10.1205/psep.05041 Process Safety and Environmental Protection, 84(B1): 1 – 9

PREDICTION OF MINIMUM FLASH POINT BEHAVIOUR FOR

BINARY MIXTURES
M. VIDAL, W. J. ROGERS and M. S. MANNAN

Mary Kay O’Connor Process Safety Center, Chemical Engineering Department, Texas A&M University System, TX, USA

F
lash point is a primary property used to determine the fire and explosion hazards of a
liquid. Minimum flash point behaviour (MFPB) is exhibited when the flash point of a
mixture is below the flash points of the individual components. The identification of
this behaviour is critical, because a hazardous situation results from taking the lowest
component flash point value as the mixture flash point. Even for experimental measurements,
an estimate of the mixture flash point is needed. A procedure to estimate the flash point of
binary mixtures is discussed. Predictions for the aqueous mixtures methanol –water and
ethanol – water are presented and compared with experimental values as well as for the
flammable mixtures octane – ethanol and octane-1-butanol, which exhibit MFPB. Quantum
chemical calculation methods such as COSMO-RS and theoretical methods such as
UNIFAC were evaluated for the prediction of MFPB mixtures.

Keywords: minimum flash point behavior; flash point; UNIFAC; COSMO-RS; binary mix-
tures; flammability.

INTRODUCTION have a flash point value equivalent to the minimum flash

point value of the individual mixture components, and
The flash point of a liquid is one of the most important that estimate is considered to be a ‘conservative’ measure.
flammability properties that are used in hazard and risk This practice can lead to a very hazardous situation if the
assessments. It is also the parameter used by organizations mixture exhibits a minimum flash point behaviour
such as National Fire Protection Agency (NFPA) to cate- (MFPB) where the flash point of the mixture is lower
gorize flammability. Flash point is the lowest temperature than the corresponding flash point values of the individual
at which there is sufficient vapour to form an ignitable components. Liaw et al. (Liaw and Chiu, 2003; Liaw et al.,
mixture with air (Crowl and Louvar, 2002). 2002) developed a mathematical model for the prediction
Good experimental and estimation methods are needed of the flash point of binary mixtures that seems to be
to comply with the current regulations. Most of the very promising; however it depends on the availability of
methods available are focused primarily on pure some experimental data.
compounds; however, most of the chemicals handled in The purpose of this work is to develop a procedure for
industry are mixtures. Now, the question is, ‘Can we treat the estimation of flash points of binary mixtures, especially
mixtures as pure compounds and use the available method- the ones exhibiting MFPB. The procedure is intended to
ologies to evaluate the risks associated with them?’ The guide the user step by step and to highlight the basic infor-
answer to this question is generally NO. The behaviour of mation that is needed to evaluate the flash points of binary
mixtures can be extremely different compared to the beha- mixtures. Several methods are discussed to provide alterna-
viour of the individual components, and making such tives to be evaluated depending on the specific compounds
assumptions can be very dangerous. of the mixture. Even for experimental tests, an estimate of
The flash point of a mixture will vary depending on the the mixture flash point is needed primarily to identify a
composition. The total characterization of a mixture MFPB mixture. Therefore, the goal is to identify MFPB
requires significant experimental work and time. Probably, mixtures and then to estimate the flash points over the
that is why the available experimental flash point data of entire composition range of the mixture. Quantum chemical
mixtures are scarce. Sometimes mixtures are assumed to calculations have been included in the procedure to allow
for estimation of parameters or properties when insufficient


experimental data are available. Because of its wide appli-
Correspondence to: Dr M. S. Mannan, Mary Kay O’Connor Process
Safety Center, Chemical Engineering Department, Texas A&M University
cation as an estimation and prediction tool in the chemical
System, College Station, TX, 77843-3122, USA. industry, the UNIFAC model has been used for the predic-
E-mail: [email protected] tion of all mixture flash points.

1
2 VIDAL et al.

BACKGROUND flammable limit composition in air.

Experimental Psat
i,fp (Tf )
LFLi ¼ (1)
Flash points are determined experimentally by heating P
the liquid in a container and then introducing a small
where Psati,fp(T) is the vapour pressure at the flash point
flame just above the liquid surface. The temperature at
temperature and P is the ambient pressure. The vapour
which there is a flash or ignition is recorded as the flash
pressure of a mixture and consequently its flash point will
point (ILPI, 2004). Two general experimental methods
be affected by the predominant interactions (intermolecular
are the closed cup and open cup. In the closed cup
forces) among the molecules (see Figure 1).
method the fuel is enclosed, while in the open cup the
For example, a mixture of heptane/octane has properties
fuel is exposed to open air. The flash points determined
that can be predicted with a simple mixing rule that ignores
with the closed cup method are usually lower than the
interactions among the individual components because
open cup method values because the vapours are prevented
these chemicals are very similar. Both heptane and octane
from escaping.
are chain hydrocarbons composed of C and H atoms, and
Several standardized test methods exist for the exper-
the interactions among the molecules are weak. However,
imental determination of flash point. Most of them can be
when a hydrocarbon is mixed with an alcohol the inter-
obtained through the American Society for Testing and
actions are very strong due to differences in polarities.
Materials (ASTM) and are designated as E 502 (ASTM E
The degree of interaction depends on the sizes of the
502-84, 2000), D 56 (ASTM D 56, 2002a), D 92 (ASTM
respective molecules, and the larger their difference in
D 92, 2002b), D 93 (ASTM D 93, 2002c), D 1310
size, the stronger their interactions. In a liquid mixture of
(ASTM D 1310, 2001a), D 3278 (ASTM D 3278, 2004),
a hydrocarbon and an alcohol the molecules prefer to be
D 3828 (ASTM D 3828, 2002d), D 3934 (ASTM D
alone than together. Therefore due to the strong repulsive
3934, 2001c) and D 3941 (ASTM D 3941, 2001b, 1996).
interactions, more molecules escape to the vapour phase
According to a survey performed by the ASTM Committee
than for an ideal mixture, so the vapour pressure of the mix-
E27 (Ural et al., 2004) the flash point test method E 502,
ture is increased. Because in this case both chemicals are
which is a compilation of closed cup methods, is the
flammable, a higher vapour pressure results in a lower
most frequently used method in the United States.
flash point value to reach the lower flammable limit (LFL).
The problem is that most of these methods cover primarily
Therefore, when two flammable compounds are mixed
the determination of the flash point of pure chemicals and
together, the intermolecular forces between them are more
only contain comments on the flammability of mixtures.
important that their individual flash point values. On the
For example, the ASTM E 1232 (ASTM E 1232-02,
other hand, when the mixture is composed of a flammable
2002) specifies that:
compound and a non-flammable one, the flash point value
“Care must be exercised in the use of estimated flashpoints for of the flammable chemical can be assumed to represent the
temperature limit testing of mixtures. There are many mixtures flash point of the mixture. If the interactions between these
(non-ideal solutions) having flashpoints below the flashpoint of two chemicals are so repulsive that the vapour pressure is
any component. Unless detailed data are available it is difficult higher for the mixture than for the pure components, the
to estimate temperature limits of flammability1 of mixtures.’
non-combustible compound will serve as a diluent in the
The availability of flash point experimental values for liquid gas mixture. This is the case of water, which is not combus-
mixtures is extremely limited. Most of these data are presented tible. When water is mixed with any flammable chemical,
in graphs (Affens and McLaren, 1972; Gmehling and the resultant mixture is extremely non-ideal, but the resultant
Rasmussen, 1982; Liaw and Chiu, 2003; Liaw et al., 2002, mixture cannot have a flash point value lower than the flash
2003) or charts (Affens and McLaren, 1972; Choe and point of the flammable component.
Schecker, 1988; Ellis, 1976; Hanley, 1998; Lenoir, 1975; The flash point of a binary mixture can be estimated by
McGovern, 1992a, b; Rybicky and Stevens, 1981; Wu and the model developed by Liaw et al. (2002):
Finkelman, 1978), which can lead to different interpret-
ations. The flash point of a mixture is a function of the mix- X
2
xi gi Psati
ture composition. Therefore, a characterization of the liquid ¼1 (2)
Psat
i;fp
mixture over its entire composition range is required. i¼1

where xi , gi , Psat
i and Psat
i,fp are the mole fraction, activity
coefficient, vapour pressure at temperature T, and vapour
Theoretical pressure at the flash point T of the mixture components,
The properties of mixtures are difficult to predict, respectively. This model is a result of the combination of
because a simple mixing rule will not work when inter- equation (1) with the Le Chatelier’s rule for the flammable
actions among the mixture components are strong. The vapour –air mixture of two components as shown below:
flash point of a liquid mixture depends on the composition
of the mixture and on the types of chemicals involved. X2
Yi
Mathematically, the flash point is the temperature at ¼1 (3)
i¼1
LFL i
which the vapour pressure is equivalent to the lower
1
The temperature limit of flammability is the minimum temperature at
where y1 and y2 refer to the vapour mole fraction of
which liquid chemicals evolve sufficient vapours to form a flammable components 1 and 2, and LFL1 and LFL2 refer to the
mixture with air under equilibrium conditions. lower flammable limit of components 1 and 2. A proof of

Trans IChemE, Part B, Process Safety and Environmental Protection, 2006, 84(B1): 1–9
 PREDICTION OF MINIMUM FLASH POINT BEHAVIOUR FOR BINARY MIXTURES 3

Le Chatelier’s mixing rule based on thermodynamics is

presented in (Mashuga and Crowl, 2000). The assumptions
made in the derivation of the rule were found to be reason-
ably valid for the LFL at the flash point temperature, Tf.
Another assumption of the model is that the liquid and
vapour phase of the mixture are in equilibrium with the
vapour phase behaving as an ideal gas. From this equili-
brium condition, the compositions of the flammable com-
ponents in the vapour phase, yi, are obtained as shown in
equation (4):
Figure 2. Conditions to identify a mixture that exhibit MFPB as developed
xi gi Psat by Liaw et al. (2003) (MFPB: minimum flash point behaviour; FPBICS:
yi P ¼ xi gi Psat
i or yi ¼ i
(4) flash point between individual components).
P

The success in the prediction of mixture flash points depends lowest flash point value is selected as component #1 to
on the accuracy of the parameters used in the model. apply these conditions.
The activity coefficient gi is a dimensionless parameter The rules are based on the estimation of the vapour
that measures the deviation from ideality of component i pressures of each mixture component at its own flash
in a mixture. For non-ideal solutions the components point temperature and at the flash point temperature of
have activity coefficients smaller or greater than 1. For a the other component. The activity coefficients at infinite
binary mixture with a positive deviation from the ideal dilution are also needed (g1
i ¼ gi jxi !0 ), which can be esti-
solution behaviour, the components have gi . 1, i ¼ 1, 2, mated using any of the Gex models or quantum mechanics
which also means that the partial pressure of the vapour methods illustrated in Figure 3.
is greater than that predicted by Raoult’s law:

P ¼ gi xi P0i (5)
FLASH POINT ESTIMATION PROCEDURE FOR
In these mixtures the attractive forces among the like mol- BINARY MIXTURES
ecules are stronger than the attractive forces among unlike The basic information needed for the estimation of a
molecules. These are the binary mixtures with a potential of binary mixture flash point is:
exhibiting MFPB because of the higher than ideal vapour
pressure generated when the two components are mixed. . flash points of the individual components of the mixture,
Quantum mechanics, which represents the energies and . vapour pressure of each mixture component, and
spatial distributions of small particles, can estimate the . activity coefficient of each component.
interaction energies that are needed for activity coefficient When vapour – liquid experimental data are available for
models. An excess Gibbs function model is a function of the mixture of interest, the flash point calculation is
temperature and composition, which allows for the calcu- straightforward by applying the model developed by Liaw
lation of activity coefficients that are needed to determine et al. (Liaw and Chiu, 2003; Liaw et al., 2002) and is
flash points at different mixture compositions. The relation presented in equation (2). However, sufficient data are
of the excess Gibbs energy function and the activity coeffi- often unavailable, and other ways of obtaining the basic
cient of component i is shown in equation (6): information are needed.
A procedure to estimate the flash point of binary

E
@G mixtures was developed and is presented in Figure 3.
RT ln gi ¼ (6) This procedure is based on obtaining the data needed for
@ni T,P,nj
the model of Liaw et al. (2002), and it includes the
option of determining some of the basic parameters from
A review of the available excess Gibbs energy function quantum chemical calculations or from correlations.
models can be found in (Orbey and Sandler, 1998). The flash point values of the individual mixture com-
To identify a binary mixture exhibiting MFPB, Liaw ponents can be obtained from experimental measurements
et al. (2003) developed necessary conditions, which are or from correlations with boiling point or with molecular
presented in Figure 2. The mixture component with the descriptors. A correlation of the flash point for organics
compounds with their boiling points was developed
by Satyanarayana and Rao (1992) and is presented in
equation (7).

b(c=Tb )2 ec=Tb
Tf ¼ a þ (7)
(1  ec=Tb )2

where Tf is the flash point temperature (K), a, b, and c are

Figure 1. Relation of molecular interaction to macroscopic properties, constants provided in Satyanarayana and Rao (1992), and
such as vapour pressure. Tb is the boiling point temperature (K).

Trans IChemE, Part B, Process Safety and Environmental Protection, 2006, 84(B1): 1–9
4 VIDAL et al.

Figure 3. Procedure to estimate the flash point of binary mixtures.

The equation was fit for over 1200 compounds with less can be obtained from collections such as (Boublik et al.,
than 1% absolute error with experimental data (based on 1984; Poling et al., 2001), or from online databases such as
K). The larger deviations, of approximately 108C, were the NIST Chemistry WebBook available at http://webbook.
found for the phosphorous chemical group. A comparison nist.gov/chemistry/. An excellent compilation of data and
of this correlation with other correlations is presented property calculation websites can be found at University
(Satyanarayana and Rao, 1992). of Illinois at Chicago—Thermodynamics Research
The flash point of pure components can also be obtained Laboratory, Nano, Quantum and Statistical Mechanics
from quantitative structure property relationships (QSPR), and Thermodynamics Data and Property Calculation
which are correlations that relate the flash point value with Websites. Available at: http://tigger.uic.edu/mansoori/
molecular descriptors. The QSPR approach finds quantitative Thermodynamic.Dataand.Property_html.
mathematical relationships between the intrinsic molecular Another correlation that can be used for the estimation of
structure and observable properties of the compounds. vapour pressure is the Wagner equation presented in (10):
Katritzky et al. (2001) used QSPR to develop the following
three-parameter correlation for the flash point: at þ bt1:5 þ ct3 þ dt6
ln Psat
i ¼ (10)
pffiffiffiffiffiffi Tr
3
TF ¼ (44:50 + 0:99) 3 Gb þ (16731 + 574:45)HDCA
þ (4:95 + 0:73)MR  (117:70 + 9:14) (8) where t ¼ 1  Tr and Tr corresponds to the reduced temp-
erature, which is defined by Tr ; T=Tc , where Tc is the
pffiffiffiffiffiffi critical temperature. This correlation is one of the most
where 3 Gb is the cubic root of the gravitational index, HDCA accurate ones, because it is constrained to generate a
is a hydrogen donor charged solvent accessible surface area, ‘reasonable shape’ for the vapour pressure curve from a
and MR is the molecular weight divided by the number of reduced temperature of 0.5 up to the critical point (Poling
atoms in the molecule. et al., 2001). The parameters for this correlation can be
The molecular descriptors can be obtained by using a found in Poling et al. (2001).
quantum chemical package including drawing and optimiz- If no experimental data are available, the vapour press-
ing the molecule. A review of molecular descriptors can be ures can be estimated by the Ambrose – Walton and/or
found in Karelson (2000). Riedel methods. Properties needed for these two methods
An equation for the saturated vapour pressure, Psat i , of are the critical temperature (Tc), critical pressure (Pc),
each mixture component as a function of temperature is and boiling point temperature (Tb) for Vetere’s modifi-
needed to estimate the mixture flash point. One of the cation of the Riedel method, and the accentric
most common correlations is the Antoine equation: factor (v), Tc and Pc for the Ambrose – Walton method.
Equations for both methods can be found in Poling
Bi et al. (2001).
log Psat
i ¼ Ai  (9)
T þ Ci The activity coefficients are the most important par-
ameters of the model because they determine whether a
where Ai, Bi and Ci are the parameters of compound i. This mixture will exhibit a MFPB. Values of these coefficients
correlation should not be used outside the temperature are also the most difficult to obtain due to the vast combi-
range at which the parameters where obtained. Usually in nation of possible mixtures. Some of the excess Gibbs
the range of 0.01 –2 bar, the Antoine equation provides energy function models that can be used to obtain the
excellent results. The parameters for the Antoine equation activity coefficients are: NRTL, Wilson, UNIQUAC, and

Trans IChemE, Part B, Process Safety and Environmental Protection, 2006, 84(B1): 1–9
 PREDICTION OF MINIMUM FLASH POINT BEHAVIOUR FOR BINARY MIXTURES 5

Table 1. Four Gex models.

Model Binary parameters Best fit data Comments

NRTL g122g22, g212g11 Aqueous organic mixtures

Wilson L12, L21 Alcohols, phenols, aliphatic hydrocarbons
UNIQUAC u122u22, u212u11 Hydrocarbons, ketones, esters, water, Requires pure component molar volumes as well
amines, alcohols, nitriles, and so on as surface area and volume parameters
UNIFAC Group interaction Hydrocarbons, ketones, esters, water, Based on the solution of groups concept and
parameters, amn and amn amines, alcohols, nitriles, and so on uses the group volumes (Rk) and group
surface areas (Qk) as input parameters

UNIFAC. A comparison of these four models is presented values of the two screening charge densities s and s 0 ,
in Table 1. which form a molecular contact. These screening charge
The first three Gex models presented in Table 1 to calcu- densities can be described as molecular descriptors and
late the activity coefficients depend on experimental binary provide information about the polarity of the molecules.
interaction parameters. Sometimes the values of these par- The s-profile of various compounds is presented in
ameters cannot be found for the mixture of interest and Figure 4.
alternatives are needed to supply the needed information. These molecular descriptors in conjunction with stat-
The UNIFAC model is the only one from the models in istical thermodynamics are used to calculate the chemical
Table 1 that does not need experimental binary interaction potential and therefore activity coefficient of a component
parameters. Instead, the contributions due to molecular in the mixture. This methodology is very useful for
interactions, parameters amn and anm, are obtained from a estimating the infinite dilution activity coefficient, g1 i .
database using a wide range of experimental results. Then the rules presented in Figure 2 are applied to
Some of these parameters can be found in Poling et al. evaluate if the mixture has a potential of exhibiting
(2001). The advantage of the UNIFAC method is that mix- MFPB.
tures composed of the same functional groups can be ana- Another approach is an indirect method to obtain infi-
lysed using the same binary parameters. nite dilution activity coefficients from solvation energies.
Quantum chemical (QC) calculations can be used to The electrostatic part of the free energy is obtained from
estimate values of the binary interaction parameters to various continuum solvation models, and the two energy
input into any of the activity coefficient models that require parameters in the UNIQUAC model are related to the
experimental data. This approach was recently proposed by attractive part of the solvation free energy. The fundamen-
Sum and Sandler (1999) and is based on a combination of tal equation that allows the calculation of g1 i from
ab initio quantum mechanical (QM) methods and the solvation energies is (Lin and Sandler, 1999):
activity coefficient model. Sum and Sandler implemented
this approach for the prediction of vapour –liquid equilibria r02
(VLE) for a number of hydrogen bonded binary mixtures. RT ln g1
sol
sol
sol
i ¼ DGi=j  DGi=i  DGi=i þ RT ln
r01
The molecular interactions in the liquids were determined
through construction of a molecular cluster model, which (11)
was normally made of eight molecules. If this approach
is applied, the cluster size depends on the sizes of where DG
sol
i=j is the free energy change of the solvation of
the molecules that compose the mixture. In other solute i in solvent j, DG
sol
i=i is the free energy change of
words, the critical cluster size must be addressed for each the solvation of solute i in solvent i, and r0i is the density
mixture. An assumption that is made when applying this of liquid i. More information on this approach can be
methodology to estimate the activity coefficients is that
the binary interactions parameters are not temperature
dependant.
Neiman et al. (2004) employed the same methodology,
but instead of using quantum mechanical methods, they
employed molecular dynamics (MD) simulations to evalu-
ate molecular interactions in the liquid. They claim that
MD simulations can describe very well the behavior of
liquids if a sufficiently large unit cell is used. In this case,
the calculations are dependent upon the size of the unit
cell, the simulation time, and the method used to evaluate
the interaction energies.
Another alternative is a general theoretical method
based on conductor-like screening model for real solvents
(COSMO-RS), which was proposed by Klamt and co-
workers (Klamt, 1995, 2003; Klamt and Eckert, 2000).
COSMO-RS is a theory that describes the interactions
in a fluid as local contact interaction of molecular sur- Figure 4. Screening charge profiles for various compounds obtained using
faces, and the interaction energies are quantified by the COSMO-RS.

Trans IChemE, Part B, Process Safety and Environmental Protection, 2006, 84(B1): 1–9
6 VIDAL et al.

Table 2. Input data used for COSMO-RS calculations.

Antoine coefficients
Gas phase
Chemical A B C energy (Ha)

Octane 6.90940 1349.820 209.385 2315.718761

Heptane 6.89385 1264.37 216.636 2276.4041106
Ethanol 8.21330 1652.050 231.480 2155.0456301
1-Butanol 7.47680 1362.390 178.730 2233.6744593

found from the work of Lin and Sandler (1999) and from
the work of Nanu and Loos (2004).

RESULTS
Figure 5. Prediction of the water–methanol mixture flash point.
The main purpose of this paper is to show how to esti-
mate flash points of binary mixtures when vapour –liquid mixture with FPBICS. This prediction will not present any
experimental data are limited and to identify mixtures hazard, since it will assume that the flash point for the mix-
that exhibit MFPB. Prediction tools such as UNIFAC as ture could be lower than the flash point of the individual
well as quantum chemical calculations have been used components. On the contrary, if the prediction were the
and applied to accomplish this goal. opposite, the results obtained with COSMO-RS would
First of all, the general theoretical method COSMO-RS result in a hazardous misclassification.
was used to estimate the g1 i , which were then used in the The advantage of using COSMO-RS is that the screening
rules presented in Figure 2 to analyse its usefulness for charge that is used as an input in the calculations must be
the identification of mixtures exhibiting MFPB. calculated just once per chemical, so the user can create a
The COSMO-RS calculations were performed using database of different chemicals and then apply the rules
C2 . DMol3 with the input data presented in Table 2. for different mixtures to screen the ones that have the
The Antoine coefficients were obtained from Wang potential of exhibiting MFPB.
(2002) and the gas phase energies were calculated using The flash points of two aqueous mixtures were estimated
Gaussian 03 (Frisch et al., 2004). The gas phase energies using interaction binary parameters that were obtained from
were obtained using the b3lyp level of theory and the quantum chemical calculations and from experimental data.
6-31 þ g(d) basis set. The results obtained for three The binary interaction parameters were used in the
mixtures are presented in Table 3. UNIQUAC Gex model to estimate the activity coefficients.
When compared with the original source (Liaw et al., The results for the water –methanol mixture flash point are
2003), where the activity coefficients were calculated using presented in Figure 5.
a Gex model, the results obtained with COSMO-RS are These predictions were made by using different para-
lower for Rule #1 and higher for Rule #2. However, the meters from the literature and by applying equation (2)
results comply with the inequality in Rule #1 (.1) and with modifications for aqueous mixtures presented in
they predict correctly the mixtures that exhibit MFPB. The Liaw and Chiu (2003). Each of the parameters, presented
flash points of the individual components were obtained in Table 4 was used in the UNIQUAC Gex model for the
from NFPA 325 (1994). It is important to note that uncertain- calculation of the activity coefficients.
ties in the parameters used for the vapour pressures cancel All the predictions agree with the experimental data2
out in both rules. Then, a successful identification of a obtained from (Liaw and Chiu, 2003). The trend of the
mixture with MFPB depends on the g1 i . experimental mixture flash point is predicted accurately
In general COSMO-RS can be used reliably to estimate with all of the parameters analysed. Better predictions are
the g1i , and then the rules in Figure 2 are used to identify obtained with experimental binary interactions parameters,
mixtures with a potential of exhibiting MFPB. If the results but reasonably results are obtained with QM binary
obtained for the g1 i with COSMO-RS are not accurate, the interaction parameters. The predictions assuming an ideal
results in the classification when applying the rules may be mixture are presented in Figure 5 for comparison.
the prediction of a mixture with MFPB when in fact it is a Larger deviations between the ideal predicted values and
experimental data are obtained as the water content is
Table 3. Results using COSMO-RS to obtain g1 increased. At a water mole fraction of 0.9, the difference
i to identify a mixture that
exhibits MFPB based on the conditions developed by Liaw et al. (2003). in the Tf from ideal behaviour and experimental data is
approximately 128C.
Rule #1 Rule #2 The same analysis was performed for the mixture of
g1 sat
g1 sat water – ethanol, and the results are presented in Figure 6.
1 P1 jT2,fp 2 P2 jT1,fp
.1 The parameters used in the predictions are presented in
Mixture Psat
1,fp Psat
2,fp Classification Table 5. The mixture flash point predictions obtained
Heptane (1) –Octane (2) 2.7555 0.3145 FPBICS 2
Octane (1) –1-Butanol (2) 15.5909 16.3828 MFPB Experimental data from (Liaw and Chiu, 2003; Liaw et al., 2002, 2003)
Octane (1) –Ethanol (2) 9.5680 294.4660 MFPB are presented in a graphical form. The image capabilities of Matlab
were used to obtain the experimental data points from the original source.

Trans IChemE, Part B, Process Safety and Environmental Protection, 2006, 84(B1): 1–9
 PREDICTION OF MINIMUM FLASH POINT BEHAVIOUR FOR BINARY MIXTURES 7

Table 4. Parameters used in the UNIQUAC Gex model for the prediction of Table 5. Parameters used in the UNIQUAC model for the prediction of the
the flash point of the water–methanol mixture. flash point of the water–ethanol mixture.

ðu12  u22 Þ ðu21  u11 Þ ðu12  u22 Þ ðu21  u11 Þ

Parameters A12 ¼ A21 ¼ Reference Parameters A12 ¼ A21 ¼ Reference
R R R R
A 2271.26 736.01 (Khalfaoui et al., Experimental 2109.37 299.46 (Arce et al.,
1997) 1997b)
B 180.22 2117.34 (Arce et al., QM 131.57 24.49 (Sum and Sandler,
1997a) 1999)
QM 6.37 247.51 (Sum and Sandler,
1999)
Table 6. Antoine coefficients used in the prediction of mixtures exhibiting
MFPB log(P/kPa) ¼ A 2 B/[(T/K) þ C].

with the UNIQUAC Gex model, using binary interactions Compound A B C Reference
parameters either from experimental data or from QM n-octane 6.04394 1351.938 264.030 (Hiaki et al., 1996)
calculations, agree with the experimental data obtained ethanol 7.24222 1595.811 246.702 (Boublı́k et al., 1984)
from NFPA 325. Predictions made with the purely theoreti- 1-butanol 6.54068 1335.028 296.496 (Boublı́k et al., 1984)
cal model UNIFAC also agree with the experimental data.
The ideal solution model predicts higher mixture flash point Table 7. Group volumes and group surface
values of approximately 208C (see Figure 6 at a com- areas used in the UNIFAC model.
position of x1 ¼ 0.9).
Compound Rk Qk
The flash points of two mixtures exhibiting MFPB were
predicted using the UNIFAC Gex model for the activity n-octane 5.8486 4.936
coefficients. The UNIFAC Gex model is composed of ethanol 2.5755 2.588
two parts: the combinatorial and the residual parts. The 1-butanol 3.9243 3.668
combinatorial part represents the differences in size and
shape of the molecules in the mixture and requires for
calculation only pure component properties. The residual
According to the NFPA 325 the octane flash point value
part represents the energy interactions and is a function is 138C, but the experimental value presented in Liaw’s
of temperature. The temperature that satisfies equation work is 158C. The flash point value depends on the
(2), as well as the activity coefficient model and the
method and apparatus used in the determination (Wu and
vapour pressure as a function of temperature, is the Finkelman, 1978). Predictions using both flash point
one that is selected as the flash point of the mixture. values were made for the octane – ethanol and octane –
These iterative calculations were performed in an Excel
1-butanol mixtures. Evidently, the flash point predictions
spreadsheet. The mixtures studied were: octane – ethanol,
depend on the input data supplied to the model as presented
and octane –1-butanol, which were studied previously by in Figures 7 and 8.
Liaw et al. (2002, 2003) experimentally and theoretically.
The UNIFAC model predicts satisfactorily the MFPB of
The vapour pressures of the mixture components were
both mixtures. For the octane– ethanol mixture, the predicted
calculated using the Antoine equation with parameters flash point is a little lower than the experimental values. How-
listed in Table 6.
ever, the MFPB for almost the entire range of compositions
The parameters used in the UNIFAC model for the
(0.1 , x1 . 0.9) is predicted by UNIFAC. A decrease
studied mixtures are presented in Table 7. The flash of almost 108C is seen in the mixture flash point of
points for the octane –ethanol and octane – 1-butanol
mixtures are presented in Figures 7 and 8.

Figure 6. Prediction of the water–ethanol mixture flash point. Figure 7. Prediction of the octane–ethanol mixture flash point.

Trans IChemE, Part B, Process Safety and Environmental Protection, 2006, 84(B1): 1–9
8 VIDAL et al.

Non-idealities have a strong effect on mixture flash

point, and the assumption of ideal solution can lead to
erroneous estimates that conceal the risk associated with
a specific mixture. The theoretical UNIFAC group
contribution method described very well the MFPB. There-
fore, it can be used for the prediction of binary mixtures
flash points when the liquid mixture is non-ideal and
when experimental binary interaction parameters for the
mixture components are not available.

NOMENCLATURE
Ai, Bi and Ci Antoine parameters for compound i
a, b, and c constants for the correlation of flash point and boiling
Figure 8. Prediction of the octane–1-butanol mixture flash point. point [equation (7)]
Gex excess Gibbs
DG
sol
i=i free energy change of the solvation of solute i in
solvent i
octane – ethanol. Figure 7 clearly shows the erroneous predic- DG
sol
i=j free energy change of the solvation of solute i in
tion that could be made if an ideal solution is assumed. p ffiffiffiffiffiffi solvent j
3
Gb cubic root of the gravitational index
In the octane – 1-butanol mixture, a minimum flash HDCA hydrogen donor charged solvent accessible surface
point value is observed around x1 ¼ 0.9, which was also area
predicted by UNIFAC. The effect of assuming an ideal LFLi lower flammable limit of compound i
solution for the mixture flash point prediction is presented MR molecular weight divided by the number of atoms in
the molecule
in Figure 8. In general, the mixture flash point of an ideal P ambient pressure
mixture will have a linear trend, whereas the mixture flash Pc critical pressure
point of a non-ideal mixture will be curved. The shape of Psat
i vapour pressure at temperature T
this curve will depend on the components of the mixture. Psat
i,fp(Tf) vapour pressure at the flash point temperature
R gas constant
T temperature
Tb boiling point temperatue
Tc critical temperature
CONCLUSIONS Tf flash point temperature
The procedure presented here for the estimation of Tr reduced temperature
xi liquid mole fraction of compound i
binary mixture flash point is very useful for the assessment yi vapour mole fraction of compound i
of flammability hazards, because it saves time and effort by
providing the user with techniques and resources that can
be used to estimate parameters and obtain input data that Greek symbols
gi activity coefficient of compound i
are needed for determination of mixture flash points. This g1 activity coefficient of compound i at
i
procedure also allows for the estimation of a mixture infinite dilution
flash point when experimental data are limited or no exper- v acentric factor
imental vapour – liquid equilibrium data are available for r0i density of liquid i
the mixture of interest. Compared to estimation, exper-
imental flash point data are always preferred; however,
even for experimental tests estimates of flash point values REFERENCES
are needed. Also, a tool to predict minimum flash point
behaviour is a resource that can be used by the exper- Affens, W.A. and McLaren, G.W., 1972, Flammability properties of
hydrocarbon solutions in air, J Chem Eng Data, 17(4): 482–488.
imenter to perform experiments in a safer way, because Arce, A., Blanco, A., Soto, A. and Vidal, I., 1997a, Phase equilibria of
the possibility of a lower flash point value at a certain water þ methanol þ hexyl acetate mixtures, Fluid Phase Equilibr,
composition will then not be ignored. 128(1–2): 261–270.
COSMO-RS can be used to estimate mixture component Arce, A., Martinez-Ageitos, J., Mendoza, J. and Soto, A., 1997b, Water þ
ethanol þ 2-methoxy-2-methylbutane: properties of mixing at 298.15 K
activity coefficients at infinite dilution when binary inter- and isobaric vapour-liquid equilibria at 101.32 kPa, Fluid Phase
action parameters are not available. Due to its simplicity Equilibr, 141(1–2): 207–220.
COSMO-RS allows for the evaluation and screening of American Society for Testing and Materials (ASTM) D 56, 2002a,
several mixtures at the same time. As a result, more time Standard test method for flash point by tag closed cup tester. ASTM
and effort can be used to analyse the mixtures identified International.
American Society for Testing and Materials (ASTM) D 92, 2002b,
as MFPB mixtures. Standard test method for flash and fire points by Cleveland open cup
The calculation of binary interaction parameters from tester. ASTM International.
quantum chemical calculations appears to be very promis- American Society for Testing and Materials (ASTM) D 93, 2002c, Stan-
ing, even when this approach is still in its early stages. dard test methods for flash point by Pensky-Martens closed cup tester.
ASTM International.
Quantum chemical methods provide valuable tools for American Society for Testing and Materials (ASTM) D 1310, 2001a, Stan-
assessing the flammability of mixtures that are difficult to dard test method for flash point and fire point of liquids by tag open-cup
test experimentally. apparatus. ASTM International.

Trans IChemE, Part B, Process Safety and Environmental Protection, 2006, 84(B1): 1–9
 PREDICTION OF MINIMUM FLASH POINT BEHAVIOUR FOR BINARY MIXTURES 9

American Society for Testing and Materials (ASTM) D 3278, 2004, Liaw, H.-J. and Chiu, Y.-Y., 2003, The prediction of the flash point for
Standard test methods for flash point of liquids by small scale closed- binary aqueous-organic solutions, J Hazard Mater, A101: 83–106.
cup apparatus. ASTM International. Liaw, H.-J., Lee, T.-P., Tsai, J.-S., Hsiao, W.-H., Chen, M.-H. and Hsu,
American Society for Testing and Materials (ASTM) D 3828, 2002d, T.-T., 2003, Binary liquid solutions exhibiting minimum flash-point
Standard test methods for flash point by small scale closed cup tester. behavior, J Loss Prevent Proc, 16(3): 173–186.
ASTM International. Liaw, H.-J., Lee, Y.-H., Tang, C.-L., Hsu, H.-H. and Liu, J.-H., 2002, A
American Society for Testing and Materials (ASTM) D 3934, 2001c, mathematical model for predicting the flash point of binary solutions,
Standard test method for flash/no flash test-equilibrium method by a J Loss Prevent Proc, 15(6): 429–438.
closed-cup apparatus. ASTM International. Lin, S.-T. and Sandler, S.I., 1999, Infinite dilution activity coefficients
American Society for Testing and Materials (ASTM) D 3941, 2001b, from ab initio solvation calculations, AIChE J, 45(12): 2606–2618.
1996, Standard test method for flash point by the equilibrium method Mashuga, C.V. and Crowl, D.A., 2000, Derivation of le Chatelier’s mixing
with a closed-cup apparatus. ASTM International. rule for flammable limits, Process Safe Prog, 19(2): 112–117.
American Society for Testing and Materials (ASTM) E 502-84, 2000, McGovern, J.L., 1992a, 4 methods for predicting the flash-point of alkyd
Selection and use of ASTM standards for the determination of flash paints containing Vm- and P-naphtha and mineral spirits 1, J Coating
point of chemicals by closed cup methods. ASTM International. Technol, 64(810): 33–38.
American Society for Testing and Materials (ASTM) E 1232-02, 2002, McGovern, J.L., 1992b, A method for estimating the flash points of
Standard test method for temperature limit of flammability of chemicals. coatings containing mixtures of oxygenated and hydrocarbon solvents
ASTM International. and petroleum distillates 2, J Coating Technol, 64(810): 39–44.
Boublı́k, T., Fried, V.e. and Hála, E., 1984, The Vapour Pressure of Pure Nanu, D.E. and Loos, T.W.D., 2004, Limiting activity coefficients of
Substances: Selected Values of the Temperature Dependence of the aqueous flavour systems at 298 K by the group contribution (GCS)
Vapour Pressures of Some Pure Substances in the Normal and low model, Mol Phys, 10(3): 235–241.
Pressure Region, 2nd rev. (Elsevier Science Pub. Co.). Neiman, M., Cheng, H., Parekh, V., Peterson, B. and Klier, K., 2004, A
Choe, M.H. and Schecker, H.G., 1988, Flash points of multicomponent critical assessment on two predictive models of binary vapour-liquid
liquid-mixtures—comparison of experimental and calculated values, equilibrium, Phys Chem Chem Phys, 6(13): 3474–3483.
Chem-Ing-Tech, 60(2): 124 –125. NFPA 325, 1994, Fire hazard properties of flammable liquids, gases, and
Crowl, D.A. and Louvar, J.F., 2002, Chemical Process Safety: Fundamen- volatile solids (NFPA, Quincy, MA, USA).
tals with Applications, 2nd edition (Prentice Hall PTR). Orbey, H. and Sandler, S.I., 1998, Modeling Vapour-Liquid Equilibria:
Ellis, W.H., 1976, Solvent flash points—expected and unexpected, J Coat- Cubic Equations of State and Their Mixing Rules (Cambridge
ing Technol, 48(614): 44 –57. University Press, New York, USA).
Frisch, M.J. et al., 2004 Gaussian 03 (Gaussian, Inc., Wallingford, Poling, B.E., Prausnitz, J.M. and O’Connell, J.P., 2001, The Properties of
CT, USA). Gases and Liquids, 5th edition (McGraw-Hill, New York, USA).
Gmehling, J. and Rasmussen, P., 1982, Flash points of flammable liquid- Rybicky, J. and Stevens, J.R., 1981, Deceptive closed cup flash points of
mixtures using unifac, Ind Eng Chem Fund, 21(2): 186–188. solvent mixtures containing chlorinated solvents, J Coating Technol,
Hanley, B., 1998, A model for the calculation and the verification of 53(676): 40–42.
closed cup flash points for multicomponent mixtures, Process Saf Satyanarayana, K. and Rao, P.G., 1992, Improved equation to estimate
Prog, 17(2): 86 –97. flash points of organic-compounds, J Hazard Mater, 32(1): 81–85.
Hiaki, T., Taniguchi, A., Tsuji, T., Hongo, M. and Kojima, K., 1996, Sum, A.K. and Sandler, S.I., 1999, A novel approach to phase
Isobaric vapour-liquid equilibria of octane þ 1-butanol, þ2-butanol, equilibria predictions using ab initio methods, Ind Eng Chem Res, 38:
and þ2-methyl-2-propanol at 101.3 kPa, J Chem Eng Data, 41(5): 2849–2855.
1087–1090. University of Illinois at Chicago, Thermodynamics Research Laboratory.
ILPI (Interactive Learning Paradigms Incorporated), 2004, The MSDS Nano, Quantum & Statistical Mechanics & Thermodynamics.
HyperGlossary: Flash Point. Available at: http://www.ilpi.com/msds/ Ural, E.A., Fruirip, D., Britton, L., Fenlon, W., Going, J., Harrison, K. and
ref/flashpoint.html. Niemeier, J., 2004, The role of ASTM E27 methods in hazard assess-
Karelson, M., 2000, Molecular Descriptors QSAR/QSPR (John Wiley & ment, 38th Loss Prevention Symposium, AIChE Spring National
Sons, Inc., New York, USA). Meeting, New Orleans, USA.
Katritzky, A.R., Petrukhin, R., Jain, R. and Karelson, M., 2001, QSPR Wang, N.S., 2002, University of Maryland—Antoine coefficients data.
analysis of flash points, J Chem Inf Comp Sci, 41(6): 1521–1530. Available at: http://www.glue.umd.edu/nsw/ench250/antoine.dat.
Khalfaoui, B., Meniai, A.H. and Borja, R., 1997, Thermodynamic proper- Wu, D.T. and Finkelman, R., 1978, A mathematical model for the
ties of water þ normal alcohols and vapour-liquid equilibria for binary prediction of closed cup flash points, Organic Coatings and Plastics
systems of methanol or 2-propanol with water, Fluid Phase Equilibr, Chemistry, 38: 61–67.
127(1–2): 181 –190.
Klamt, A., 1995, Conductor-like screening model for real solvents—a new
approach to the quantitative calculation of solvation phenomena, J Phys
Chem, 99(7): 2224– 2235.
Klamt, A., 2003, COSMO-RS: a novel bridge from quantum chemistry to ACKNOWLEDGEMENT
fluid phase thermodynamics, Abstr Pap Am Chem S, 226: U432–U433.
Klamt, A. and Eckert, F., 2000, COSMO-RS: a novel and efficient method The authors wish to thank the National Science Foundation for the
for the a priori prediction of thermophysical data of liquids, Fluid Phase financial support provided.
Equilibr, 172(1): 43–72.
Lenoir, J.M., 1975, Predict flash points accurately, Hydrocarb Process, The manuscript was received 17 February 2005 and accepted for
54(1): 95–99. publication after revision 13 May 2005.

Trans IChemE, Part B, Process Safety and Environmental Protection, 2006, 84(B1): 1–9