Property estimation for design, simulation and analysis of biodiesel process

systems: Review and Implementation

Fabrcio Rodrigues, Adilson J. Assis
a Faculdade

de Engenharia Qumica, Universidade Federal de Uberlandia, Av. Joao Naves de Avila,

2121, Bloco 1K, Campus Santa Monica,
CEP: 38400-901, Uberlandia, MG, Brazil

Abstract
Biodiesel is a very attractive alternative to fossil fuels in several aspects (e.g. sustainability), but requires economic
efficient process in order to be viable. Therefore, the application of systematic methods for process synthesis, design
and analysis is of special interest for biodiesel process systems. The success of most systematic approaches is based on
physicochemical properties values that are difficult (and sometimes impossible) to be experimentally obtained. Property estimation methods are a practical way to fulfill this gap, however the complex nature of the molecules involved
requires the application of more refined theories to capture the intrinsic physicochemical behavior. To overcome this
hudrle, an extensive review of the literature was performed in order to obtain the most representative compounds,
experimental data for several properties and the state-of-the-art property estimation methods resulting in a computeraided system to analyze (focusing on consistency checks) and selected the best estimation methods. The result is
a robust set of estimation methods that provide the basis for biodiesel process systems simulation. The generated
knowledge and computational tools are, thus, integrated to develop a package that extends the original ChemSep
compound database, including more than additional 200 compounds (acylglicerydes, fatty acids, fatty acid methyl
esters and fatty acid ethyl esters) typically present in biodiesel process systems and implements the selected property
estimation methods. The main advantage of the package is brought by the CAPE-OPEN capabilities of ChemSep that
enables the use of reliable physicochemical properties for biodiesel process systems in any process simulator that is
CAPE-OPEN compliant.
Keywords: Property estimation, Biodiesel process simulation, CAPE-OPEN

1. Introduction
Biodiesel fuel has many advantages in comparison to
fossil fuels. Biodiesel yields much more energy than
used in its production (372% more than bioethanol)
(Hill et al., 2006), has good transportation properties
due to its low volatility (Krawczyk, 1996), can be produced from many different renewable sources (Ma and
Hanna, 1999) and can be added, in small amounts
(1-2w%), to regular diesel fuel with low lubricant
properties to yield an acceptable fuel (Gerpen, 2005).
Nonetheless there are also many hurdles for its production and consumption, been the biodiesel price the main
hurdle as most existing processes can only deal with
high value raw materials (Ma and Hanna, 1999; Zhang
et al., 2003b).
author. Tel.: +55 3291 4189 ramal XXX
Email address: ajassis@feq.ufu.br (Adilson J. Assis)

Corresponding

Preprint submitted to Bioresource Technology

Although governments of many countries offer subsides to make the biodiesel production profitable there
is a very strong need to improve the process economics.
The application of process synthesis and design techniques (Biegler et al., 1997) can contribute very much
to process profitability whilst amplifying the social and
environmental benefits of biodiesel. Energy savings of
around 50% and net present cost reductions of 35% have
been achieved in industry (Siirola, 1996) and several
other examples of successful industrial applications and
best practices of process synthesis and design methods
and tools have been reported (Harmsen, 2004). However, most, if not all, of these techniques require the
knowledge of a set of physical and chemical properties that are currently not widely available for the compounds involved in biodiesel systems.
The complex nature of some molecules (typical
triglycerides have more than 50 carbons with many
unsaturations present) and the experimental difficulties
September 28, 2014

(and impossibilities) to obtain experimental data for

some properties (e.g. critical temperature due to degradation), requires the application of more refined theories
to capture the physicochemical behavior of the involved
molecules. Property estimation methods represent a
practical way to fulfill this gap. Several methods have
been developed and even though many authors have reviewed the field (Su et al., 2011; Daz-Tovar et al., 2011;
Chang and Liu, 2009), none of the discussions have
given the proper focus to the property models behavior. For example, some property models fit very well
experimental data for some compounds, but for other
molecules, that have no experimental data included, the
model presents inconsistencies (e.g. heat capacity decreases with temperature increase) and a purely statistical analysis of the methods deviations is, thus, not able
to identify.

Figure 2: General molecules of saturated (a) fatty acids, (b) simple

triglycerides, (c) fatty acid methyl esters and (d) fatty acid ethyl esters.

3. Property estimation model analysis and practical

implications
During the extensive literature review, several stateof-the-art property estimation methods were identified
for each of the selected properties. The methods can be
classified based on their approach to estimate the property, most of the methods are based on the group contribution approach and a complete classification of the
estimation methods is presented in Tables 2 and 10.
The compounds analyzed are also classified in fatty
acids (FA), acylglycerides, fatty acid methyl esters
(FAME) and fatty acid ethyl esters (FAEE). The four
classes of compounds can be divided in sub-classes
based on the number of unsaturations present in the
molecule. For each of these subclasses there is a general structure form for each of the property estimation approaches. This general structure allows to write
the property estimation methods in a simplified way
that can substantially increase our understanding of the
model.
To organize and simplify the discussion, properties
were classified in primary (depend only on the molecular structure), secondary (cannot be explicitly calculated
only from structural information and usually require the
knowledge of other properties) and temperature dependent properties.

Therefore, objectives of this paper are to discuss, assess and select property estimation methods, showing
that the analysis of the property models behavior provides a powerful tool to select a proper estimation methods in the face of scarce experimental data, and to implement the generated knowledge in a computational
tool that is CAPE-OPEN compliant providing a reliable
environment for the application of process synthesis and
design techniques to biodiesel process systems.

2. Computer-aided tools developed for the analysis

The compounds present in biodiesel process systems can be summarized in short-chain alcohols (e.g.
methanol, ethanol), glycerol, acylglycerides (AC), fatty
acids (FA) and alkyl esters (AE). The properties of
short-chain alcohols and glycerol are usually included
in common process simulators and databases, and a first
step for the analysis was the identification of 220 relevant compounds (29 fatty acids, 69 triglycerides, 40
diglycerides, 15 monoglycerides, 29 fatty acid methyl
esters and 29 fatty acid ethyl esters). The next and more
time consuming step was an extensive literature review
to gather information regarding compounds, important
properties and state-of-the-art estimation methods. To
manage and organize the information, a database was
created in Microsoft Access and implemented in C# resulting in a software named BioPES (BIOdiesel Property Estimation System) that offers aids as illustrated
in Figure 1 (the software can be obtained by the reader
through contact with the authors).

3.1. General Structure of the Compounds Involved in

Biodiesel Process Systems
A general structure for each subclass of the involved
compounds is derived as a basic means to understand
and extrapolate the behavior of the property estimation
methods. In Figure 2 some classes of molecules are
presented in a generalized way, from this representation is possible to draw generalized structures for each
of the approaches listed in Tables 2 and 10 as shown
in Table 1 for a group contribution method based on
the UNIFAC groups and considering only first-order
2

Figure 1: Concept of BioPES

explicitly written only in terms of the chemical structure and are often a function of other properties, the only
secondary property analyzed is the acentric factor ().
Most of the estimation methods in the literature for
primary properties are of the group contribution type.
They all assume that the groups have an additive contribution to the overall property value. Multi-level contributions are used as corrections for specific multigroup,
conformational or resonance effects.
The methods in Table 2 represent selected methods
for the estimation of primary properties. In Table 3 are
presented the experimental data points gathered for the
evaluation of primary properties in which it is clear the
lack of experimental data for acylglycerides for most
properties.

Table 1: General structure of saturated FA, simple TG, FAME and

FAEE for a first-order group contribution method.

Class
(Saturated)
FA
TG
FAME
FAEE

Group Contribution
Structure
1.CCH3 + n.CCH2 + 1.CCOOH
3.CCH3 + (3n + 2).CCH2 + 3.CCH2 COO + CCH
2.CCH3 + n.CCH2 + 1.CCH2 COO
2.CCH3 + (n + 1).CCH2 + 1.CCH2 COO

groups, such as Gani and Constantinou (1996); Marrero and Gani (2001). Other group-contribution methods that use different groups than those defined in the
UNIFAC group-contribution can easily be converted to
the proposed general structure as well as other classes
of compounds, more information is given in Tables S1
and S2.
At this point, it is possible to write any of the property
model functions in terms of general structures (such as
the ones presented in Table 1) which will greatly simplify the analysis of next section.

3.2.1. Critical Temperature

As presented in Table 2, five estimation methods are
analyzed for critical temperature, the property model
function of these methods is presented in Table 4. Applying the general structure of saturated FAME in Table 4 the general property model functions shown in
Table 5 are obtained (the property model functions for
other classes of compounds are presented in Tables S3
and S4). The estimation methods were checked against
22 experimental data points, distributed within each
class of compounds as presented in Table 3. The results in terms of average absolute deviation (AAD =
P
1/n ni=1 |Pexp Pcalc |) and average relative deviation
P
(ARD = 1/n ni=1 100|Pexp Pcalc |/Pexp ) are presented
in Table 6. The CG method provided the lowest ARD,
however no experimental data are available for acylglycerides further analysis is providential.

3.2. Primary and Secondary Properties

Primary properties are those that depend only of the
chemical structure of the molecule considered. The
primary properties considered are: critical temperature
(T c ), critical pressure (Pc ), critical volume (Vc ), boiling
point (T b ), melting point (T m ), standard Gibbs energy
of formation (G f ) and standard enthalpy of formation
(H f ). Secondary properties are those that cannot be
3

Table 2: Characteristics of the primary and secondary property estimation methods considered.

Property

Estimation Method

Approach

Fedors (1982)
Joback and Reid (1987) (JR)
Tc

Group-Contribution

Tu and Liu (1996) (TL)

Gani and Constantinou (1996) (CG)
Marrero and Gani (2001) (MG)

Vc , Pc , T b , T m ,
G f , H f

Joback and Reid (1987) (JR)

Group-Contribution

Gani and Constantinou (1996) (CG)

Marrero and Gani (2001) (MG)
Pitzer (1955)

Correlation

Pitzer (1955)

Definition

Lee and Kesler (1975)

Pitzer expansion/Corresponding-states

Ambrose and Walton (1989)

Pitzer expansion/Corresponding-states

Gani and Constantinou (1996) (CG)

Group-Contribution

Marrero and Gani (2001) (MG)

Group-Contribution

Table 3: Experimental data points per class of compounds for primary properties.

Number of data points

Tc

Vc

Pc

Tb

Tm

G f

H f

FA

18

19

24

71

18x

AG

11

29

ME

16

43

11

EE

15

43

Total

22

19

66

186

32

Table 4: Property model functions for the critical temperature.

Method
Fedors
JR
TL
CG
MG

Critical temperature model function

P
T c = 535log( i Ni T c1i )
P
P
T c = T b [0.584 + 0.965 i Ni T c1i ( i Ni T c1i )2 ]1
1
T c = [6.26897.106 + 2.56086.103 ( 2x
+ 1x )]1
P
where x = 0.160864 + i Ni T c1i
P
P
T c = 181.28ln[ i Ni T c1i + j M j T c2 j ]
P
P
P
T c = 181.6716ln[ i Ni T c1i + j M j T c2 j + k Mk T c3k ]

Table 5: General property model functions for the critical temperature estimation of saturated FAME.

Method

Generalized critical temperature model function for saturated FAME

Fedors

T c = 232.3475ln(10.12 + 1.34n)

JR
TL

T c = 2799.4737T b [(n + 25.5806)(n 70.0568)]1

T c = [6.26897.106 +

0.00256
]1
6.1438+0.9172n

CG

T c = 181.28ln[15.4897 + 3.492n]

MG

T c = 181.6716ln[16.7104 + 3.4607n]

recommended for use.

The CG and MG methods provided almost the same
estimations (the difference is always smaller than 8 K
for AG, ME and EE; and about 25 K for low NC FA,
the difference decreases as NC increases), TL and Fedors methods are qualitatively similar to CG and MG
predictions but overestimate CG and MG methods on
about 50 K. The same is also observed for other groups
of compounds as shown in Figure S1.
Based on this analysis, we recommend the CG
method for the T c estimation due to its confirmed usefulness as a thermodynamic parameter for the calculation of secondary and temperature dependent properties.

Table 6: T c estimation methods experimental analysis.

Method

AAD

ARD

Fedors

14.41 (5.46 - 27.59)

2.05 (0.77 - 4.16)

JR

13.18 (1.92 - 28.68)

1.81 (0.28 - 3.57)

TL

13.74 (5.94 - 25.05)

1.95 (0.84 - 3.77)

CG

4.39 (0.25 - 11.31)

0.64 (0.03 - 1.70)

MG

18.09 (5.17 - 33.63)

2.57 (0.73 - 5.09)

Regardless of experimental data, the generalized critical temperature model functions for each compound
class are a way to gain a deeper understanding of the
estimation method. Figure 3 presents the T c estimations vs the number of carbons (NC) (note that NC , n)
for saturated fatty acids (a) and saturated simple triglycerides (b) with carbon number. The T c vs. NC curve
of each estimation method deviate considerably in both
cases. For saturated fatty acids, there is a considerable
amount of experimental data suggesting the CG method
as the more accurate, the same happens for fatty acid
methyl and ethyl esters as shown in Figures S1 and S2.
It is clear from Figure 3 (b) that the JR method provides
a completely different curve compared to the other estimation methods even when experimental boiling points
are used (see Figure S2). The JR method generalized T c
model function for saturated simple triglycerides is:
Tc =

0.00128
6.1438+0.9172n

311.0526T b
(n + 11.67678)(n 20.202233)

3.2.2. Critical Pressure

Differently from T c , the Pc model functions of all
methods have the following form:
Pc = A +

1
(B +

Ni Pc1i )2

(2)

where, A and B are additional adjustable parameters of

the property estimation models. The generalized form
for each of the subclasses of compound has the following model form:
Pc = A +

1
(B1 + C.n)2

(3)

where, B1 combines the adjustable parameter B and the

group-contributions that are independent of n and C are
the group-contributions dependent of n.
The AAD and ARD for JR, CG and MG were calculated based on 19 experimental data points, and, as
presented in Table 3, only FA are represented by the
data. Figure 4 (a) presents the estimated Pc values for
saturated fatty acids showing that the estimation methods differ very little, the same happens for FAME and
FAEE as shown in Figure S2 and Table S2.

(1)

In this case NC FA = n + 3 and therefore the model

will return negative value for T c for triglycerides formed
from FA with NC higher than 23, such as Trilignocerin
(C24:0, C24:0, C24:0). Therefore, the JR method is not
5

Figure 3: Tc estimations behavior with NC for (a) saturated fatty acids and (b) saturated triglycerides.

The estimations differ significantly only for acylglycerides. Equations 4, 5 and 6 are the generalized model
functions for JR, CG and MG, respectively, applied for
saturated simple triglycerides.
Pc =

1
(0.2347 + 0.0288.n)2

Pc = 1.3705 +
Pc = 0.0519 +

Table 7: Critical pressure estimation methods analysis.

Method
JR
CG
MG

(4)

1
(0.24782 + 0.0318.n)2

(5)

1
(0.2325 + 0.0261.n)2

(6)

Pc AAD
0.72 (0.21 - 1.32)
0.66 (0.11 - 1.31)
1.07 (0.17 - 2.13)

Pc ARD
3.44 (0.91 - 8.80)
3.09 (0.64 - 7.75)
5.10 (0.96 - 9.88)

Table 8: Critical volume estimation methods analysis

Method
JR
CG
MG

where, in this case, n is equal to the number of carbons of the fatty acid that constitutes the triglyceride
subtracted by three. From these equations, it is possible
to identify two different regions:

Vc AAD
10.71 (0.17 - 22.5)
10.06 (0.32 - 21.85)
12.73 (0.32 - 24.77)

Vc ARD
1.81 (0.03 - 5.07)
1.73 (0.06 - 5.08)
2.04 (0.06 - 5.05)

if NC FA <= 9 then Pc (MG) > Pc (CG) > Pc (JR)

3.2.3. Critical Volume

else NC FA > 9 then Pc (CG) > Pc (MG) > Pc (JR)

All the critical volume model functions are linear and

there is essentially no difference from JR, CG and MG
as shown in Figures 5 and S4.

The same analysis is applied for all other classes

of compounds and the identified regions are described
in Table S4. Our final recommendation is to use the
CG method for FA, FAME and FAEE as it performed
slightly better to the experimental data. However, a
proper recommendation for acylglycerides is highly dependent of the situation, but based on the regions identified and presented in Table S4 it is possible to select
a models to provide a high, a low and a intermediate
estimation.

3.2.4. Boiling Temperature

Experimental data for T b and T m are much more
available than for the other primary properties. The estimations for the CG and MG methods are very good except with a reservation for acylglycerides, as shown in
Figure 6. Table 9 gives the calculated AAD and ARD,
suggesting the MG method as the most accurate.
6

Figure 4: Pc estimations behavior with NC for saturated fatty acids.

Figure 5: Vc estimations behavior with NC for saturated triglycerides.

Figure 6: Tb estimations behavior with NC for (a) saturated FA and (b) saturated triglycerides (JR estimations are above 1000 K).

Figure 7: Tm estimations behavior with NC for (a) saturated FA and (b) saturated triglycerides (JR estimations are above 1000 K).

Table 9: Boiling point estimation methods analysis.

Method
JR
CG
MG

T b AAD
0.72 (0.21 - 1.32)
0.66 (0.11 - 1.31)
1.07 (0.17 - 2.13)

T b ARD
3.44 (0.91 - 8.80)
3.09 (0.64 - 7.75)
5.10 (0.96 - 9.88)

3.2.5. Melting Temperature

Similarly to T b there is a considerable amount of experimental data for T m , the JR method strongly overestimate the experimental values while the estimations
obtained from CG and MG fit well the data except for
triglycerides (there is no data for di- and monoglycerides) as shown in Figure 7. It is difficult do determine
from a theoretically point of view the reasons for the deviations for triglycerides. The group-contribution models may be missing one or several important groups that
are very important for triglycerides. On the other hand
the definition of many other groups would also increase
the complexity of the models which is also a not desirable situation. Here we define two new second-order
groups for the MG method, one only to account for the
triglyceride form and another to incorporate a proper description of unsaturations present in triglycerides. The
modified MG-T m for TGs is:
T m = 143.5706 ln

X

T m1k + C1.NC + C2.NDB

Figure 8: New CG-Tm estimations for triglycerides with up to three

unsaturations.

0.643364 0.18376.NC , if NC is even

C1 =

2.559501 0.2099.NC , if NC is odd

5.21927, if NC is even
C2 =

4.08173, if NC is odd

(8)

(9)

3.2.6. Standard Enthalpy of Formation

A considerably good amount of experimental data is
available for H f regarding FA and FAME, the estimation methods agree well with the experimental data (see
Figure 9. For acylglycerides the estimations methods
deviate little with the deviation increasing with the number of unsaturations, the highest deviation was between
JR and MG: about 80 K for Trioctadecatrienoin (C18:3).


(7)

where, C1 and C2 are given by Equations 8 and 9,

NC is the number of carbons of the triglyceride and NDB
is the number of double bonds of the triglyceride. The
improved estimations are shown in Figure 8.
8

Our recommendation is to use CG method, however one

must consider that:
H f JR > H f CG > H f MG

(10)

3.2.7. Standard Gibbs Energy of Formation

No experimental data is available for G f , nevertheless it is demonstrated in Figures 10 that all the methods have precisely the same qualitative behavior. JR
provides a considerably lower estimation and there is
no significant difference between CG and MG methods
(the largest difference encountered is about 27 kJ/mol
found for TGs with many unsaturations).
3.2.8. Acentric Factor
The acentric factor is the only secondary property considered in this work, it was originally defined
by Pitzer (1955) as:


log10 lim (Pvap /Pc ) 1
(11)
T r 0.7

According to Poling et al. (2001) there are two useful procedures to estimate the acentric factor. One is to
use an accurate equation for Pvap and obtain (or estimate) the critical properties, the second approach is to
use group-contribution methods such as CG and MG.
Based on experimental data points for Pvap , T c and Vc ,
it is possible to derive 12 true experimental values for
the acentric factor from Equation 11 (11 for FA and 1
for FAME), Figure 11 presents the estimation curves
for FA (the calculated MAE and MAPE are also provided in Table S5) in which it is clear the accuracy of the
acentric factor calculations derived directly from Equation 11. The calculations used the Pvap model by Ceriani et al. (2013) and CG for the critical properties, Figure 12 shows the influence of the estimation methods
for T c and Pc demonstrating that the T c method chosen
is crucial while changing the Pc method has little effect
in the acentric factor calculations.
Based on experimental values for Pvap available for
all the classes of compounds analyzed, the Pvap model
by Ceriani et al. (2013) is considered a very accurate
model and our recommendation is to use this method
and CG for the critical properties in Equation 11. The
only relevant uncertainty is regarding T c , thus, in case
of inconsistencies the methods by Fedors (1982) and Tu
and Liu (1996) can offer alternative values.

Figure 11: Acentric factor estimation curves for FA.

Figure 12: Effect of the critical properties estimation method in the

Acentric factor estimation curves for FA.

3.3. Temperature-dependent Properties

Secondary properties are those that depend on
the chemical structure and also of other properties/variables. The secondary properties considered are
9

Figure 9: Hf estimations behavior with NC for (a) saturated FA and (b) saturated FAME.

Figure 10: Hf estimations behavior with NC for (a) saturated TG and (b) 3-unsaturations TG.

10

Table 11: Heat Capacity.

Classes of

Generalized formula for

Compounds

d from Eq. 12

Sat. FA

5.06 + 0.41 NC

Sat. TG

27.59 + 0.41 NC

Sat. FAME

5.39 + 0.41 NC

Sat. FAEE

4.98 + 0.41 NC

vapor pressure (Pvap ), liquid density (L ), liquid heat capacity (C Lp ), liquid viscosity (L ) and heat of vaporization (H vap ), which are all temperature dependent and
for some models also dependent of critical properties.
The methods considered are listed in Table 3.
Analysis regarding experimental data have already
been given by Su et al. (2011) for all the temperaturedependent properties considered here, similar results
based in our database is provided in Tables S6 and Figures S7-10. Our analysis is, therefore, focused on the
generality and inconsistencies of the property models
taking advantage of the general structure of the involved
compounds presented in Section 3.1.

Figure 14: Heat Capacity Stuff.

classes of compounds, specially when unsaturations are

present (see Figures S40-50).
The c p model by Morad et al. (2000) also presents inconsistencies. This method is based on the RowlinsonBondi equation, requiring knowledge of the critical
properties and the acentric factor and it is argued that
to apply this method to acylglycerols and vegetable oils
one can use a simple molar average of the required properties to represent the mixture properties. For example,
following the analysis the critical temperature of a simple triglyceride is:
X
xi T ci
(13)
T c,mix =

3.3.1. Heat Capacity

Four models are analyzed for c p and two of these
were found to be inconsistent.
The method by Kolska et al. (2008) is ultimately
given by the following quadratic function:
cp = a + b T + d T 2

where T ci is the critical temperature of the compound

i and xi is the composition of the compound i in the
mixture. This assumption does not introduces any problems for moderate temperatures, however note that the
critical temperature of a triglycerides is not simply the
molar average of the fatty acids that compose it. In Figure 14 one can see that about T r = 0.8, the value for c p
estimated using the method by by Morad et al. (2000) is
indeterminate (in this point T r,mix = 1).
The methods that do not present inconsistencies have
almost identical estimations and the method by Ceriani
et al. (2009) is recommended for use as it can describe
a larger number of molecules than the method by Zong
et al. (2009).

(12)

where a, b and d are given by the group contributions.

This model equation must not have a negative value for
the parameter d, otherwise it can potentially lead to c p
decreasing with an increase in temperature. Applying
the generalized structures, presented in 3.1, the generalized formulas for the parameter d are obtained and are
presented in Table 11.
It directly comes from the general formulas that the
model by Kolska et al. (2008) is potentially inconsistent
for:
simple saturated TG with NC smaller than 51.
saturated FAME with NC smaller than 14.

3.3.2. Density
Three models are analyzed for the density estimation (see Table 10), the model by Ihmels and Gmehling
(2003) (version GCVOL-OL-60) exhibits a very similar inconsistency found for the method by Kolska et al.

saturated FAME with NC smaller than 15.

Examples of the inconsistencies are presented in Figures 13. The same inconsistencies happen for the
11

Table 10: Secondary property estimation methods considered and their characteristics.

Property

cp

Estimation Method

Approach

Kolska et al. (2008)

Group-Contribution

Morad et al. (2000)

Rowlinson-Bondi Equation

Ceriani et al. (2009)

Group-Contribution

Zong et al. (2009)

Fragment-based

Tu and Liu (1996)

Hvap

Basarova and Svoboda (1995)

Group-Contribution

Ceriani et al. (2013)

Rackett Equation + Correction Term

Halvorsen et al. (1993)

Ihmels and Gmehling (2003)

Group-Contribution

Zong et al. (2009)

Pvap

Fragment-based

Ceriani et al. (2013)

Group-Contribution

Zong et al. (2009)

Fragment-based

Figure 13: Heat Capacity Stuff.

12

Table 12: Density inconsistency for the method by Ihmels and

Gmehling (2003).

Classes of

Generalized value for

Compounds

c from Eq. 14

Sat. FA

1-Unsat. FA

-0.0002878

2-Unsat. FA

-0.0005756

3-Unsat. FA

-0.0008634

4-Unsat. FA

-0.0011512

(2008). The density estimation model proposed by Ihmels and Gmehling (2003) can be represented in the following model equation:
=

MW
A + B T + C T2

(14)

The quadratic function in the denominator of Equation 14 is the molar volume of the substance and as the
volume must decrease with an increase in temperature,
a potential inconsistency may appear if the parameter
C is negative. Applying the general structure presented
in 3.1, the model equations in Table 12 are obtained.
Therefore if the model by Ihmels and Gmehling
(2003) wrong profiles like the one in Figure 15 can be
obtained.
A last consistency test for molar volume estimation
methods is:
lim (Vm ) = Vc
(15)
T r 1

It can be seen in Figure 15 that the method by Zong

et al. (2009) does not return the a Vc near from those predicted by the group-contribution methods and, thus, our
final recommendation is to use the model by Halvorsen
et al. (1993).
Figure 15: Density Stuff.

3.3.3. Vapor Pressure

Vapor pressure is one of the most important properties for process simulation, only two estimation models are analyzed, namely Zong et al. (2009) and Ceriani et al. (2013). Both methods have consistent estimations with experimental data for all the classes of compounds analyzed (see Table S8 and Figure S11). For
triglycerides, the methods are both in good agreement
in the reduced temperature range of 0.4-0.7, however
outside this range no experimental data is available and
the models estimations deviate very much from each
13

Table 13: Hvap Error (statistics).

Classes of

Generalized value

Compounds

for c from Eq. 14

Sat. FA

1-Unsat. FA

-0.0002878

2-Unsat. FA

-0.0005756

3-Unsat. FA

-0.0008634

4-Unsat. FA

-0.0011512

4. Computational Implementation of the Framework

Figure 17: Heat of Vaporization e etc/Octanoic acid.

The selected property estimation methods are of almost no use if these are not computationally available to be used in a process simulator or other related tool. Modern computational Chemical Engineering tools (software) are very much influenced by the
CAPE-OPEN interoperability standard, which has proposed a component (piece of software) architecture and
clear interface standards. The idea was to make BioPES
CAPE-OPEN compliant. Nevertheless, the simulator
used in our research group is COCO (CAPE-OPEN to
CAPE-OPEN) that is based on ChemSep van Baten
(2007) where an easiest work path was encountered.
ChemSep already has an efficient software architecture
that enables the incrementation of custom compounds
database.

other as shown in Figure 16 (in a representative example for Trioctanoin, for more information see Figures
S11-S30). The predictive power of Pvap for temperature ranges outside the experimental data is specially
cumbersome, as shown in the Appendix 2 fitting curves
to Pvap using a number of experimental data will potentially overestimate the Pvap for temperatures above
the range of experimental data. From Figures 16 and
S11-30, it is of no doubt that the model by Ceriani
et al. (2013) has used experimental data points for a
wider range of reduced temperatures than the model by
Zong et al. (2009), also Cerianis method is a groupcontribution model and is capable of predicting Pvap
for a much higher number of molecules than Zongs
method. Therefore, we recommend the model by Ceriani et al. (2013) to be used.

4.1. ChemSep Important Features

ChemSepTM has a component that manages the information regarding pure compounds, named PCDmanager, in which the user can perform several tasks such as
access pure compound properties, add a new compound,
insert and update correlation parameters for temperature
dependent properties. Among these tasks, some are of
critical importance for the accomplishment of this work:
export compound to file and import compound from file.
ChemSep-LiteTM has 432 compounds in PCDmanager
and by choosing one of these compounds and exporting
it to a file one encounters a simple text file that can be
divided in:

3.3.4. Heat of Vaporization

None of the methods analyzed exhibited inconsistencies. Su et al. (2011) analyzed the methods by Basarova
and Svoboda (1995) and Ceriani et al. (2009) and it is
claimed that the method by Ceriani et al. (2009) provides profiles in which the Hvap increases with temperature. Recently, a updated model has been proposed
by Ceriani et al. (2013) and using this model we did
not find any inconsistencies as shown in the profile of
Figure 17. The agreement of the estimations with experimental data is provided in Table XX suggesting the
model by Ceriani et al. (2013) as the best estimation
method.

1. Compound Identifiers (Name, Formula, CAS,

Smiles and Family)
2. Pure Compound Primary Properties (Critical Properties, Molecular Weight, Boiling Point and Melting Point, to name a few).
14

Figure 16: Vapor pressure e etc.

3. Parameters for 17 Pure Compound Temperature

Dependent Properties (Solid Density, Liquid Density, Vapor Pressure and Heat of Vaporization, to
name a few).
4. Additional Pure Compound Parameters for Calculation of Mixture Properties and Molecular Properties (UNIQUAC r and q, Chao-Seader acentric
factor, Chao-Seader solubility parameter)
5. Group Contribution Data (UNIFAC, UNIFACLLE, ASOG and Modified UNIFAC, to name a
few)
Therefore, to add compounds to ChemSepTM systematically, all that is required is to create such text files
holding all the required information and then add the
compounds through PCDmanager using the text file.
The process to increment ChemSepTM is further explained in Figure 18.
An additional difficulty to the process is that the temperature dependent correlations are based on a set of 35
defined model forms1 . Unfortunately, no estimations
methods have the defined model forms, for instance
some of the methods are based on group contribution
plus a correction term. To overcome this issue a parameter estimation procedure was developed in MS Excel.

Figure 19: Pamitoleic acid liquid density values generated using

Halvordensen et al. (1993) and a nonlinear
regression using the model

form number 105 y =

4.2. Parameter Estimation Procedure

Even though the estimation methods/models were not
directly constructed in any of the defined equation forms
available in ChemSep, all of them do follow at least
one of the 35 defined model forms as shown in the Figures 19 and 20 for liquid density and heat of vaporization, respectively.

A
B[1+(1T/C)D]

Figure 20: Pamitoleic acid heat of vaporization values generated using

Ceriani et al. (2013) and a nonlinear regression
 using the model form
number 106 y = A(1 T r )B+C T r +D T r +E T r .
2

1 The model forms are the same used in the DIPPR databank
and a complete list can be found at http://www.chemsep.com/

15

Figure 18: Process to export a file compound from BioPES to ChemSep.

Therefore, all that is necessary is to obtain the parameters for 5 temperature dependent properties for all
the 232 compounds that means to perform at least 1362
nonlinear regressions. All the necessary data can be
generated on BioPES and the nonlinear regressions are
easily done on Excel through the development and application of appropriate macros systematicaly calling
Excel Solver. The process to obtain the property estimation model parameters in a model form compatible
with ChemSepTM is summarized in Figure 21 with a few
more details.
The nonlinear regression was only accepted when the
coefficient of correlation was higher than 0.99 resulting
in 227 compounds with all the correlation parameters
determined. An example of the results for Trimyristin
is given in Table 14.

Figure 22: COCO flowsheet configuration GUI showing some of the

Compounds added in this work.

4.4. Results and Discussion for the Computational Implementation of the Framework
The implementation was successfully loaded on
COCO simulator as shown in Figure 22 and, as an initial
validation, the biodiesel production process from waste
cooking oil (triolein) proposed by Zhang et al. (2003a)
simulated on HYSYS was reproduced as represented in
Figure 23.
The simulation results were exact the same as those
reported by Zhang et al. (2003a). It should be mentioned that Triolein (the compound representing the
waste cooking oil) is usually the only triglyceride found
in most of the commercial simulators and the advantage in using the implementation developed here is that
other compounds present in the waste cooking oil can
be included in the analysis, resulting in a more detailed
description of the system. Furthermore, any other biodiesel production systems composed by the 228 compounds included in the framework can be approached
using the reliable property estimation methods.

4.3. Compound Text Files Generation and Compounds

Addition to ChemSep
After obtaining the property estimation model parameters in an model form compatible with ChemSepTM all
the requirements to accomplish the task defined in Figure 2 (create the compound text files and import them on
PCDmanager) are met. The compound text file generation was done through BioPES using the .NET Framework and, using these files, the compounds were finally
add to ChemSep using the Import from File feature.
At this point, the only step to be accomplished was to
save the work. The file generated contains 660 compounds (432 originally included and 228 add in this
work) with several reliable, constant and temperature
dependent, physical property values ready to be used in
a distillation column simulation using ChemSepTM or
an entire plant simulation using COCO simulator or any
other simulator that is CAPE-OPEN compliant such as
Aspen PlusTM . The final file containing all the information can be obtained via email contact to the authors.

5. Conclusions
The extensive review of the property estimation
methods for the compounds involved in biodiesel pro-

downloads/docs/ChemSepTutorial_PCDManager.pdf

16

Figure 21: Process to obtain the property estimation model parameters in an model form compatible with ChemSepTM in four steps.

Table 14: Nonlinear Estimation Results for Trimyristin.

Property
Liquid Density
Vapor Pressure

Eq. Number
105
101

Eq. Form
y= 
y=e

A
D

B 1+(1T/C)
A+ TB +C ln(T )+D T E

Heat of Vaporization

106



3
2
y = A(1 T r ) B+C Tr +D Tr +E Tr

Liquid Heat Capacity

y = A + BT

17

0.289

0.443

808.4

0.5848

194.2

-29241

-22.028

0.0002

-0.375

0.289

0.443

808.4

0.5848

1.01.106

1530.5

Figure 23: Simulation of the biodiesel production process from triolein proposed by Zhang et al. (2003) using COCO simulator and the framework
implemented in this work.

cess systems shows the importance of consistency tests

when building property models. Nevertheless a robust framework for the estimation of many properties
was successfully built. One subject of substantial importance not addressed in this work is phase equilibria and chemical reactions, but are addressed elsewhere
by Likozar and Levec (2014) and ??. The successful
computational implementation described shows the capabilities of integrating several different methods and
tools towards a common objective. One may realize that
the computational implementation developed here has a
considerable resemblance to the CAPE-OPEN standard,
even though the former is not as elegant as the later, each
of the tools used are specialized in a certain task and the
communication between the different tools were carefully controlled by the authors strategy.

Tb

Boiling Temperature (K)

Tc

Critical Temperature (K)

Tm

Melting Temperature (K)

Vc

Critical Volume (cm3 /mol)

AAD

Average Absolute Deviation

AE

Alkyl esters

AG

Acylglycerides

ARD

Average Relative Deviation (%)

BioPES A software developed in this work (Biodiesel

Property Estimation System)
CG

Gani and Constantinou (1996) property estimation method

Fatty acids

List of Symbols
G f

Standard Gibbs Energy of Formation (J/mol)

FA

H f

Standard Enthalpy of Formation (J/mol)

FAEE Fatty acid ethyl esters

Ci

Contribution of the group i in the property function

FAME Fatty acid methyl esters

Pc

Critical Pressure (bar)

JR
18

Joback and Reid (1987) property estimation

method

MG

Marrero and Gani (2001) property estimation method using the updated parameters by
Hukkerikar et al. (2012).

TL

Tu and Liu (1996) method for critical temperature estimation

Ihmels, E. C., Gmehling, J., 2003. Extension and revision of the group
contribution method gcvol for the prediction of pure compound liquid densities. Industrial & engineering chemistry research 42 (2),
408412.
Joback, K. G., Reid, R. C., 1987. Estimation of pure-component properties from group-contributions. Chemical Engineering Communications 57 (1-6), 233243.
Kolska, Z., Kukal, J., Zabransky, M., Ruzickaa, V., 2008. Estimation
of the heat capacity of organic liquids as a function of temperature
by a three-level group contribution method. Industrial & Engineering Chemistry Research 47 (6), 20752085.
Krawczyk, T., 1996. Biodiesel-alternative fuel makes inroads but hurdles remain. inform 7 (8), 800815.
Lee, B. I., Kesler, M. G., 1975. A generalized thermodynamic correlation based on three-parameter corresponding states. AIChE Journal
21 (3), 510527.
Likozar, B., Levec, J., 2014. Transesterification of canola, palm,
peanut, soybean and sunflower oil with methanol, ethanol, isopropanol, butanol and i tert/i-butanol to biodiesel: Modelling
of chemical equilibrium, reaction kinetics and mass transfer based
on fatty acid composition. Applied Energy 123, 108120.
Ma, F., Hanna, M. A., 1999. Biodiesel production: a review. Bioresource technology 70 (1), 115.
Marrero, J., Gani, R., 2001. Group-contribution based estimation of
pure component properties. Fluid Phase Equilibria 183, 183208.
Morad, N. A., Kamal, A. M., Panau, F., Yew, T., 2000. Liquid specific heat capacity estimation for fatty acids, triacylglycerols, and
vegetable oils based on their fatty acid composition. JOURNALAMERICAN OIL CHEMISTS SOCIETY 77 (9), 10011006.
Pitzer, K. S., 1955. The volumetric and thermodynamic properties of
fluids. i. theoretical basis and virial coefficients1. Journal of the
American Chemical Society 77 (13), 34273433.
Poling, B. E., Prausnitz, J. M., John Paul, O., Reid, R. C., 2001. The
properties of gases and liquids. Vol. 5. McGraw-Hill New York.
Siirola, J. J., 1996. Industrial applications of chemical process synthesis. Advances in chemical Engineering 23, 162.
Su, Y.-C., Liu, Y., Diaz Tovar, C. A., Gani, R., 2011. Selection of
prediction methods for thermophysical properties for process modeling and product design of biodiesel manufacturing. Industrial &
Engineering Chemistry Research 50 (11), 68096836.
Tu, C.-H., Liu, C.-P., 1996. Group-contribution estimation of the enthalpy of vaporization of organic compounds. Fluid Phase Equilibria 121 (1), 4565.
van Baten, J. M., 2007. An introduction to coco. In: The 2007 Annual
Meeting.
Zhang, Y., Dube, M., McLean, D., Kates, M., 2003a. Biodiesel production from waste cooking oil: 1. process design and technological assessment. Bioresource technology 89 (1), 116.
Zhang, Y., Dube, M., McLean, D., Kates, M., 2003b. Biodiesel production from waste cooking oil: 2. economic assessment and sensitivity analysis. Bioresource technology 90 (3), 229240.
Zong, L., Ramanathan, S., Chen, C.-C., 2009. Fragment-based approach for estimating thermophysical properties of fats and vegetable oils for modeling biodiesel production processes. Industrial
& engineering chemistry research 49 (2), 876886.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, in the online version, at
ftp.feq.ufu.br/Adilson/supplementarydata.pdf.
References
Ambrose, D., Walton, J., 1989. Vapour pressures up to their critical
temperatures of normal alkanes and 1-alkanols. Pure and Applied
Chemistry 61 (8), 13951403.
Basarova, P., Svoboda, V., 1995. Prediction of the enthalpy of vaporization by the group contribution method. Fluid phase equilibria
105 (1), 2747.
Biegler, L. T., Grossmann, I. E., Westerberg, A. W., 1997. Systematic
methods of chemical process design. Vol. 796. Prentice Hall PTR
Upper Saddle River, NJ.
Ceriani, R., Gani, R., Liu, Y., 2013. Prediction of vapor pressure and
heats of vaporization of edible oil/fat compounds by group contribution. Fluid Phase Equilibria 337, 5359.
Ceriani, R., Gani, R., Meirelles, A. J., 2009. Prediction of heat capacities and heats of vaporization of organic liquids by group contribution methods. Fluid Phase Equilibria 283 (1), 4955.
Chang, A.-F., Liu, Y., 2009. Integrated process modeling and product design of biodiesel manufacturing. Industrial & Engineering
Chemistry Research 49 (3), 11971213.
Daz-Tovar, C.-A., Gani, R., Sarup, B., 2011. Lipid technology: Property prediction and process design/analysis in the edible oil and
biodiesel industries. Fluid Phase Equilibria 302 (1), 284293.
Fedors, R., 1982. A relationship between chemical structure and
the critical temperature. Chemical Engineering Communications
16 (1-6), 149151.
Gani, R., Constantinou, L., 1996. Molecular structure based estimation of properties for process design. Fluid Phase Equilibria
116 (1), 7586.
Gerpen, J. V., 2005. Biodiesel processing and production. Fuel processing technology 86 (10), 10971107.
Halvorsen, J., Mammel Jr, W., Clements, L., 1993. Density estimation
for fatty acids and vegetable oils based on their fatty acid composition. Journal of the American Oil Chemists Society 70 (9), 875
880.
Harmsen, G., 2004. Industrial best practices of conceptual process
design. Chemical Engineering and Processing: Process Intensification 43 (5), 671675.
Hill, J., Nelson, E., Tilman, D., Polasky, S., Tiffany, D., 2006. Environmental, economic, and energetic costs and benefits of biodiesel
and ethanol biofuels. Proceedings of the National Academy of Sciences of the United States of America 103 (30), 1120611210.
Hukkerikar, A. S., Sarup, B., Ten Kate, A., Abildskov, J., Sin, G.,
Gani, R., 2012. Group-contribution+ (gc+) based estimation of
properties of pure components: Improved property estimation and
uncertainty analysis. Fluid Phase Equilibria 321, 2543.

19