Computers and Chemical Engineering 66 (2014) 201213

Contents lists available at ScienceDirect

Computers and Chemical Engineering

journal homepage: www.elsevier.com/locate/compchemeng

A systematic methodology for design of tailor-made blended products

Nor Alaza Yunus a,b, , Krist V. Gernaey a , John M. Woodley a , Raqul Gani a
a
b

Department of Chemical and Biochemical Engineering, Technical University of Denmark, Sltofts Plads, 2800 Kongens Lyngby, Denmark
Process Systems Engineering Centre (PROSPECT), Faculty of Chemical Engineering, Universiti Teknologi Malaysia, 81310 Skudai, Johor, Malaysia

a r t i c l e

i n f o

Article history:
Received 2 November 2013
Received in revised form
21 December 2013
Accepted 24 December 2013
Available online 3 January 2014
Keywords:
Product design
Blended product
Gasoline blend
Lubricant base oil

a b s t r a c t
A systematic methodology for design of tailor-made blended products has been developed. In tailor-made
blended products, one identies the product needs and matches them by blending different chemicals.
The systematic methodology has four main tasks. First, the design problem is dened: the product needs
are identied, translated into target properties and the bounds for each target property are dened.
Secondly, target property models are retrieved from a property model library. Thirdly, a mixture/blend
design algorithm is applied to obtain the mixtures/blends that match the design targets. The result is
a set of blends that match the constraints, the composition of the chemicals present in the blend, and
the values of the target properties. Finally, the mixture target property values are veried by means
of rigorous models for the properties and the mixtures. In this paper, the methodology is highlighted
through two case studies involving gasoline blends and lubricant base oils.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
A blended liquid product is dened as a formulation of various chemicals in the liquid state to obtain a liquid mixture with
a particular set of desired characteristics and qualities. Examples of blended liquid products are synthetic fuels and lubricants.
Consumer-oriented liquid products, which are also blends of chemicals, are usually called formulated products, where a solid active
ingredient is dissolved and blended with other chemicals. For
example, the formulation of an insect repellent contains a solid
active ingredient that is responsible for the main function of the
product, solvents that deliver the active ingredient, and additives that enhance the quality of the product (Conte, Gani, & Ng,
2011). The blended liquid products, on the other hand, contain
one or more liquid chemicals that serve as the main ingredient and perform the main function of the product (for example,
release heat when combusted or absorb heat or release and take
up heat in a cyclic operation) and additives that enhance the quality of the product. For example, a lubricant blend may contain a

Abbreviations:
2BE, 2-butanone; 2MT, 2-methyltricosane; 3ET, 3ethyltetracosane; 3ME, 3-methyleicosane; 9ODA, cis-9-octadecenoic acid; ACE,
acetone; DFE, 1H-dibenzo[a,i]uorene,eicosahydro-; ETOH, ethanol; G, gasoline;
GLY, propane-1,2,3-triol; MeTHF, furan,tetrahydro-2-methyl-; MI, main ingredient;
MoT, modeling tool; PE, polyethylene; THF, tetrahydrofuran; WCO, waste cooking
oil.
Corresponding author at: Department of Chemical and Biochemical Engineering,
Technical University of Denmark, Sltofts Plads, 2800 Kongens Lyngby, Denmark.
Tel.: +45 45252910; fax: +45 45932906.
E-mail addresses: noy@kt.dtu.dk, alaza@gmail.com (N.A. Yunus).
0098-1354/$ see front matter 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compchemeng.2013.12.011

specic base oil as the main ingredient and a set of additives. The
base oil primarily determines the lubricant performance and the
additives enhance its quality. In this work, only the class of formulations that are blended liquid mixtures are considered, and
they will be referred to as tailor-made blended products throughout this article. A tailor-made blended product will refer to a
mixture of several chemicals tailored to match specic product
attributes.
In order to efciently design tailor-made blended products,
a systematic methodology is needed. Several efforts have been
reported on the development of systematic methodologies for
product design. Churi and Achenie (1997) proposed a mathematical programming approach to design refrigerant mixtures. A small
set of individual refrigerants were used as the building blocks
in the mixtures design. This approach is practical in obtaining
the best mixture by optimizing a performance criterion but the
approach only implies one type of ingredient in the mixtures. On
the other hand, Cheng, Lam, Ng, Ko, and Wibowo (2009), Conte,
Gani, and Ng (2011), Conte, Gani, Cheng, and Ng (2012) and Teixeira,
Rodriguez, Rodrigues, Martins, and Rodrigues (2012) designed consumer oriented chemicals based products that involve various
types of ingredients. Cheng et al. (2009) proposed an integrated
approach to design a skin care cream, taking into consideration both
technical as well as business-related factors. Meanwhile, Conte
et al. (2011) developed a model-based computer-aided methodology to design and verify formulated products (for example, paint
and insect repellent lotion). Conte et al. (2012) added an experimental component to their model-based approach. That is, the nal
validation, selection and adjustment of the design is made through
experiments.

202

N.A. Yunus et al. / Computers and Chemical Engineering 66 (2014) 201213

Nomenclature
Indexes
i
j
B
k
m

compound i in blend
compound j in blend
blend
target property
mixture

Notation

activity coefcient
solubility parameter (MPa1/2 )


target property

dynamic viscosity (cP)

kinematic viscosity (cSt)
0
kinematic viscosity of 0 VI oil at 40 C (cSt)
kinematic viscosity of blend oil at 40 C (cSt)
B
kinematic viscosity of 100 VI oil at 40 C (cSt)
100

density (g/cm3 )
amorphous density (g/cm3 )
a
acentric factor

heat of combustion (kJ/mol)

Hc
Gmix
energy of mixing
intrinsic viscosity
[]
blending index
BI
cost
C
HHV
higher heating value (MJ/kg)
log LC50 lethal concentration (mol/L)
Mw
molecular weight (g/mol)
NC
number of compounds
critical pressure (bar)
Pc
Psat
saturated vapor pressure (kPa)
PP
pour point (K)
gas constant
R
RON
research octane number
RVP
Reid vapor pressure (kPa)
SG
specic gravity
critical temperature
Tc
Tf
ash point (K)
melting point (K)
Tm
Tg
glass transition temperature (K)
reduced temperature (K)
Tr
Vvap
vapor loss (wt%)
VI
viscosity index
molar volume at critical point
Vc
WtO2
weight percent of oxygen (%)
k,m
x1,LB

k,m
x1,UB

ZRA

the lowest composition of component 1 in mixture,

m that satises target property, k
the highest composition of component 1 in mixture,
m that satises target property, k
constant of the Modied Rackett equation

Design of tailor-made blended products is challenging in different ways. Tailor-made blended products usually have a main
ingredient that is mixed with additives, to obtain the desired endproperties. The challenge in the design of these products is to nd
suitable chemicals and their compositions within the blend such
that the end-properties of the resulting product achieve the desired
performance. Chemical selection is an important step in blended
product design and has the potential to signicantly enhance the
likelihood of nding truly innovative products. Another challenge is
how to deal with the phase behavior issue since by denition, the
blended products must be stable liquid solutions. Therefore, efcient solution strategies are needed to deal with all the challenges.

It is the intention to implement the model-based methodology proposed in this paper at the early stage of blend design. The proposed
approach is suitable to perform two of the four product design steps
of (Cussler & Moggridge, 2011): generate ideas and select ideas. The
use of computer-aided methods allows to quickly identify the most
suitable blend candidates and avoid spending efforts on infeasible
regions of the search space. The objective here is to save time and
resources by quickly identifying the promising blends and then verifying and/or improving them through focused experimental work.
The objective of this paper is to present a systematic
methodology for design of tailor-made blended products using
a computer-aided model-based technique. The methodology is
divided into four main tasks; problem denition, model identication, candidate (blends) generation and model-based verication.
Three supporting tools are developed for this work, namely,
a chemical database, a property model library and a blendcomposition optimizer. The methodology has been applied to two
case studies involving the design of gasoline and lubricant blends.
This paper is an extended version of a short ESCAPE-23 contribution
(Yunus, Gernaey, Woodley, & Gani, 2013).
2. Methodology
The systematic methodology employs the reverse design
approach (Gani & Pistikopoulos, 2002), where the targets of the
design problem are dened and blends that match the targets
are identied. In blend design, thousands of blend candidates are
systematically generated and screened. For most problems it is possible to generate and screen all potential feasible blend candidates.
The tailor-made blend design problem is formulated as a mathematical problem, which results in a Mixed Integer Non-Linear
Programming (MINLP) problem. The objective here is to optimize
a specic performance, subject to product stability and attributes
(target properties) as well as process specications. The result from
the solution of the problem is information on the blends in term of
the chemicals that are present in it, their compositions, and the
estimated target property values.
Considering the multiple types of constraint equations, the general tailor-made blend problem is formulated as follows:
min or max fobj (x, y, C, E, S, Q )

(1)

Subject to
g1 (x, y) 0

Mixture constraints :

Target property constraints :

Other constraints :
x
y

g3 (x) = 0

x|x Rn , 0 x 1

0, 1

LB g2 (x, y, ) UB

(2)
(3)
(4)
(5)
(6)

fobj is the objective function to minimize/maximize one or more of

the following parameters: the blend composition (x), the type of
mixture (y), cost (C), environmental impact (E), safety factor (S) or
product performance (Q); y is an integer variable, which is linked
to the type of mixtures; x is a continuous variable, which is related
to the mixture compositions;  corresponds to a vector of target
properties; subscripts UB and LB represent the upper and lower
limits, respectively; g1 corresponds to the mixtures constraints
with respect to blend miscibility and the solubility condition that
must be satised; g2 is a vector of target property constraints translated from product needs, for example, viscosity; g3 is a vector of
other constraints such as the denition of mole or weight or volume
fraction.

N.A. Yunus et al. / Computers and Chemical Engineering 66 (2014) 201213

2.1.1. Task 1 Problem denition

2.1.1.1. Task 1.1 Identify product needs. The needs for blended
products are primarily determined from the principal product functionality, which is the main reason for a products to be of interest
for potential customers. For example, for an engine lubricant, the
principal function of the blended product is to reduce the resistance
and prevent wear between two moving surfaces. A blended product
may have more than one principal function. Besides, requirements
from environmental regulation and safety are also considered as
additional constraints in the design of these blended products. A
knowledge base, a thorough literature search and legislation details
are used to determine the product needs in this work.

Tailor-made Blended Product Design

Task 1 Problem definition
1.1 Identify needs
1.2 Translate needs into target properties
1.3 Set the target values

Task 2 Property model identification

Property models library

3.1 Collect input data

3.2 Generate and screen blends using the
mixture/blend design algorithm
3.3 Rank blend candidates according to a
selection criterion

Chemical databases

STABILITY tool

2.1 Retrieve models from library

Task 3 Mixture/blend design

203

Task 4 Model-based verification

4.1 Verify using rigorous models

Blend formulations
Fig. 1. Work ow in the systematic methodology for tailor-made blended product
design.

The above-mentioned blend design problem, involving a large

database of chemicals and non-linear constraints, creates a combinatorial problem resulting in a very large search space. By
employing a systematic decomposition based solution approach
(Karunanithi, Achenie, & Gani, 2005), it is possible to manage the
complexity of this design problem efciently and to reduce the
search space. The decomposition based solution approach divides
the MINLP problem into several sub-problems that are simpler
and relatively easy to solve. First only the linear constraints are
considered and the compositions of the blends are characterized
in terms of lower- and upper-bounds within which all the linear constraints are satised (see steps 3.1 and 3.2 in Section 2.2).
Using the lower- and upper-bounds of compositions from the linear
constraints problem, the non-linear property constraints are now
considered. Solution of this problem gives new lower- and upperbounds but still within the range of the linear constraints (see step
4.1 in Section 2.2). In the nal step, the compositions within the
established bounds from the non-linear constraints that optimize
the objective function (Eq. (1)) are determined. As a nal verication step, the solution from the nal step is given as initial step
and the optimization problem dened by Eqs. (1)(6) is solved. The
work-ow of the decomposition based methodology is presented
in Fig. 1.
2.1. Tasks of the methodology
This systematic methodology consists of four tasks. The rst task
is to dene the design problem, where the product needs are identied, translated into target properties and the target values are
determined. In the second task, the required target property models are retrieved from the property model library. In the third task,
the blend problem is solved using a blend design algorithm to nd
the blends that best match the design targets. Finally, the mixture
target property values are veried by rigorous models for the properties and mixtures that require it. Here, collection of data in the
database helps to verify the predicted blend properties.

2.1.1.2. Task 1.2 Translate needs into physico-chemical properties. A

specically developed knowledge base is used to transform the
product needs into desirable target properties. Note that not all the
product needs, such as color, odor and shelf life, can be evaluated
using a model-based approach. However, when validated models
are available, it is easier and faster to test using the model than
performing experiments on the products.
2.1.1.3. Task 1.3 Set the target values. The target values for similar
products are retrieved from the knowledge base. The target values may also be changed with the aim of improving the product
performance criteria.
2.1.2. Task 2 Property model identication
2.1.2.1. Task 2.1 Retrieve the required property models from the
library. The model library contains property models for mixture
and pure component properties that dene the blend design problem (see Tables 34). Different blend design problems need a
different set of property models.
2.1.3. Task 3 Mixture/blend design
2.1.3.1. Task 3.1 Collect input data. The input data for this task are
the main ingredient properties and composition, and a list of chemicals with their associated properties.
2.1.3.2. Task 3.2 Generation and screening. Generate and screen for
all feasible blend alternatives using the mixture/blend design algorithm. Further explanation on this algorithm is given in Section 2.2.
Other tools employed in this task are, a STABILITY tool for miscibility test, and a chemicals database, for the list of chemicals that are
to be considered in the blend design. The results from this task are
the blend formulations, compositions, and the target values.
2.1.3.3. Task 3.3 Rank blend candidates. Rank the results of the mixture/blend design algorithm according to a selected criterion. The
selection criterion can be blend composition, any target property,
performance criterion or cost, if available.
2.1.4. Task 4 Model-based verication
2.1.4.1. Task 4.1 Verication. Here, the objective is to verify the
mixture property values by means of rigorous models for the properties and mixtures that require it. For example, a linear mixing
rule is used to estimate the viscosity of blends. The model gives
a good prediction for ideal mixtures. However, the linear models
may have signicant errors for non-ideal mixtures. Therefore, further verication using rigorous models is necessary. If the target
values are not matched with the rigorous property models, then
Task 3 is repeated for the corresponding blends by assigning new
compositions as input until a matched blend formulation is found.
The result from this task is a set of blends that satisfy all property
targets and that can now be further veried, if necessary. Table 1
lists the methods and tools required in the methodology.

204

N.A. Yunus et al. / Computers and Chemical Engineering 66 (2014) 201213

Table 1
List of methods and tools used in the blend design methodology.
Tasks

Description

Methods/tools

Task1

Problem denition
1.1 Identify needs
1.2 Translate needs into target
properties
1.3 Set the target values

Knowledge base
Knowledge base
Knowledge
base/legislation/safety

Task2

Property models identication

2.1 Retrieve models from
library

Property models
library

Task3

Mixture/blend design
3.1 Collect input data
3.2 Generate and screen blends
using the mixture/blend design
algorithm
3.3 Rank blend candidates
according to a selection
criterion

Chemicals database
Mixture/blend design
algorithm, chemicals
database, STABILITY
test

Task4

Model-based verication
4.1 Verify using rigorous
models

Property models
library

Three tools are developed specically for this work in order to

solve the blending problem. They are; the mixture/blend design
algorithm for generating the blend candidates and for screening
them systematically; the chemicals database to store chemicals to
be used as building blocks in blend design; and the property models library to store the property models that are required in the
design. Also, the STABILITY tool developed by Conte et al. (2011) is
employed for identifying the miscible blends.
2.2. Mixture/blend design algorithm
The mixture/blend design algorithm employs a decomposition
method, where the problem is decomposed into four sub-problems
and solved accordingly as shown in Fig. 2. The rst level is for
screening the pure component properties, and the second level is to
analyze the mixture stability. The third and fourth levels are taking
into account the linear and non-linear target properties, respectively.
The mixture/blend design algorithm is described below for the
case of binary and ternary mixtures. It can also be extended to
multi-component mixtures. The rst compound in mixtures is
specied as the main ingredient (MI), and it must exist in all mixtures. It can be a single compound or a mixture of compounds. A
binary mixture is a combination of the MI and a compound i (Bi )
from the database (MI + Bi ), while a ternary mixture consists of MI
plus two compounds, i and j from the database (MI + Bi + Bj ). Subscripts i and j represent the number of both compounds. To avoid
any repetition of formulations of the ternary mixture, the value
of the subscript j must always be greater than the value of the
subscript i.
2.2.1. Level 1: Pure components constraints
At this level, the pure component properties of chemicals in the
database and MI are compared with respect to the target values.
Note that, this step is applied only for the linear target properties.
Step 1.1: Compare the target property,  k of the MI and the
chemical(s) in the mixture with the target value boundaries,  LB
and  UB for each target property k. Fig. 3 illustrates the comparison
of the binary mixture, where ik represents the target property of

k is the target property of MI;  k is the lower bound of

chemical i; MI
LB
k is the upper bound of the target propthe target property, k; and UB
erty, k. This step is done for all possible mixtures that are considered
in the design binary, ternary or multi-component mixtures.

Fig. 2. The mixture/blend design algorithm.

Rule 1: Reject a binary mixture if the property value of MI and

the pure component property value of the chemical i are both either
k <  k and  k <  k ), or
lower than the lower bound values (MI
LB
LB
i

k >  k and  k >  k ).

greater than the upper bound values (MI
UB
UB
i

N.A. Yunus et al. / Computers and Chemical Engineering 66 (2014) 201213

205

k is the upper bound of the target property, k;  k is the

where, UB
LB

lower bound of the target property, k; 1k and LB g4 x, , UB

are the target property values k of the chemicals 1 and 2, respectively. The specic composition, x1k,m for a dened mixture is given
by,

Target region

 k,m 2k

Fig. 3. Representation of the property comparison. The binary mixture of MI and

chemical i is infeasible.

x1k,m =

Rule 2: Reject a ternary mixture if the property value of MI and

pure component property values of the chemicals i and j are either
k <  k and  k <  k and  k <
lower than the lower bound values (MI
LB
LB
i
j

where  k,m is the specic target value of property, k.

k,m
m
and x1,UB
)
Step 3.2: Identify the overall composition range (x1,LB
for each binary mixture by comparing the composition ranges of all
k,m
target properties. The minimum and the maximum values of x1,LB

k ), or greater than the upper bound values ( k >  k and  k >

LB
MI
UB
i

(9)

1k 2k

k and  k >  k ).
UB
UB
j

k,m
and x1,UB
calculated by Eqs. (7)(8) for each property k are used as
follows:

2.2.2. Level 2: Stability analysis

Step 2.1: Collect input data for the stability test. The input
data consist of the UNIFAC-LLE group representation (Magnussen,
Rasmussen, & Fredenslund, 1981) of the chemicals involved in the
mixtures, and the temperature at which the stability test has to be
performed.
Step 2.2: Perform the stability test. The stability test of
binary mixtures is performed using the STABILITY tool. The
result obtained is the information on the miscibility of binary
pairs indicated as either totally miscible, partially miscible or
immiscible.
Step 2.3: Analyze the stability results. The result of the binary
mixtures is obtained directly from the stability test. Also the stability result for a multi-component mixture is analyzed by rst
listing all binary mixtures that represent the multi-component
mixture and then checking the stability of each binary pair. Total
and partially miscible mixtures are considered for the next level of
screening.
Rule 3: The multi-component mixtures are regarded as
immiscible if any of the binary pairs to form them is found to be
unstable. Reject the mixtures, which are immiscible to avoid any
phase split of the blends.

k,m
m
x1,LB
= max x1,LB

2.2.3. Level 3: Linear constraints

Step 3.1: Calculate the composition range for each linear target
property for all mixtures that satisfy the corresponding property
target values. The composition-property relation for a binary mixture is illustrated in Fig. 4.
k,m
k,m
and x1,UB
) for a
The composition ranges of chemical 1, (x1,LB
binary mixture, m is calculated as follows:
k,m
x1,LB
=

k,m
x1,UB
=

k,m
UB
2k

(7)

1k 2k
k,m
LB
2k

(8)

1k 2k

MIX

Target region of

Fig. 4. Representation of a binary mixture position.

k,m
m
x1,UB
= min x1,UB

(10)


(11)

m
m .
> x1,UB
Rule 4: Reject any binary mixture, if x1,LB
Rule 5: A ternary or multi-component mixture is assumed to
be infeasible if any of the binary mixtures representing it is found
infeasible. For example, if binary mixtures of MI + B1 , and MI + B2
are rejected due to the implementation of rule 4, then the combination of them to form a ternary mixture MI + B1 + B2 is regarded as
infeasible.
Steps 3.1 and 3.2 are combined and solved as a linear optimization problem to minimize and maximize the blend compositions
subject only to linear constraints as follows:

min or max fobj (x)

(12)

s.t.
LB g2 (x, ) UB
g3 (x) = 0
x

(13)
(14)

x|x Rn , 0 x 1

(15)

where g2 is a vector of inear constraints, and g3 is a vector of mole or

weight or volume fractions. The solution of the problem is a range
of compositions for each blend within which all the linear property
constraints are satised.
Step 3.3: Re-check the stability of the partially miscible mixtures
in the region of interest. The miscible region of partially miscible
mixtures is identied and compared with the region of interest. The
overlap region is dened as a new region of interest, that gives a
new composition range.
Rule 6: Reject the mixture that has a region of interest outside
of the miscible regions.
2.2.4. Level 4: Non-linear constraints
Step 4.1: Calculate the non-linear mixture properties,  k,m for
the remaining binary mixtures at the overall composition range
m
m
m
x1,LB
< x1m < x1,UB
and nd new composition ranges, (x1,LB
and
k,m
x1,UB
) that satisfy the non-linear constraints.
Rule 7: Reject the mixtures that did not match the non-linear
k and  k,m >  k
target values,  k,m < LB
UB
This step is solved as a non-linear optimization problem to minimize and maximize the blend composition subject to non-linear
constraints as follows:

min or max fobj (x)

(16)

206

N.A. Yunus et al. / Computers and Chemical Engineering 66 (2014) 201213

s.t.

LB g4 x, , UB

(17)

g5 (x) = 0

(18)

x|x Rn , xLB x xUB

(19)

where g4 is a vector of non-linear constraints whereas is the

additional parameter required for non-linear constraints, such as
temperature and activity coefcient, and g5 is a vector of mole
or weight or volume fractions. The composition, x, is restricted
by lower and upper limits. As a result, new composition ranges
are obtained, where the lower-bound value is not allowed to be
lower than the specied bound and vice versa for the upper-bound.
This ensures that all linear and non-linear property constraints are
satised.
Step 4.2: In this step, the mixture compositions within the established bounds from step 4.1 that minimizes (or maximizes) the
dened objective function Eq. (1) are determined. As a nal test, the
original optimization problem with all the constraints are solved
with the bounds from step 4.1 and the optimal solution from above
as the initial estimate.

Table 2
Section of the database with numbers of available chemicals. The last column indicates the pure chemical property present in the database.
Section

Main
ingredient

Additives: number
of compounds

Pure chemical property

Gasoline

Conventional
gasoline

Bio-based: 22
Others: 660

Non temperature
dependence;
Mw, Tc, Pc, , ZRA , LC50 , Tf ,
Hc , ON, WtO2
Temperature dependence:
 (15 C),
(15 C), Psat
(38 C),

Lubricant

Base oil
- Mineral oil
- Glycerol
- WCO
- Waste PE

Hydrocarbons: 913
Bio-based: 25
Polymers: 150
Others: 624

Non temperature
dependence;
Mw, Tf , Tm , PP, , Tg , [], a
Temperature dependence:
 (100 C),  (40 C),

(100 C),
(40 C), Psat
(25 C)

Table 3
Target properties of blend and models used in this work.
Target property

Model

Function

Dynamic viscosity,

Linear mixing rule

GC(UNIFAC)-based method
(Cao, Knudsen,
Fredenslund, & Rasmussen,
1993)

Denition,  = 
Correlation (Rizvi, 2009)
Linear mixing rules

f (i , xi )
f (
i ,  i , xi )

Linear mixing rule (on

molar volume basis)
Modied Rackett equation
(Spencer & Danner, 1973)
Linear mixing rules

f (i , xi )
f (ZRAi , Vci , T, xi )

f (RONi , xi )

GC(UNIFAC)-based method

f (Pvap,i ,  i , xi )

Linear mixing rules

GC(UNIFAC)-based method
(Liaw, Lee, Tang, Hsu, & Liu,
2002; Liaw, Tang, & Lai,
2004)
Linear mixing rules
Linear mixing rules

f (O2,i , xi )
f (Tf,i ,  i , xi )

UNIFAC (Magnussen et al.,

1981)
Linear mixing rules (using
blending index) (Fahim,
Alsahhaf, & Elkilani, 2010)

f (GC, xi )

2.3. Database
A database of chemicals stores the physico-chemical properties
of each chemical. The database is divided into sections representing each specic type of blended product because different
products require different types of chemicals and sets of properties. The list of chemicals includes those that are commonly found
in blended products. Furthermore, additional chemicals are generated using the computer-aided molecular design (CAMD) technique
(Gani, Nielsen, & Fredenslund, 1991) to ll in the gaps. The generated chemicals that satisfy a set of target properties are stored in
the database. For the gasoline database, the target properties are
specied as follows: chemicals must be in liquid form at ambient temperature (Tm < 293 K), and must not be unstable, toxic or
corrosive. Olens are easily oxidized to form gums during storage,
thus they should not be used in gasoline blends. This section of
the database stores chemicals that have the potential to be used as
fuel substitutes. The gasoline database holds 660 chemicals of various types, for example ethers, esters, aldehydes, ketones, amines,
amides and furans.
The lubricant section of the database contains chemicals with
high viscosity that makes them suitable as base oil. The lubricant
section contains; 913 alkanes, 624 organic chemicals including
alcohols, ethers, esters, acids and derivatives, and 150 polymers. The information on the physico-chemical properties and
their parameters are also provided in the database. They are
retrieved from the CAPEC database (Nielsen, Abildskov, Harper,
Papaeconomou, & Gani, 2001) and from handbooks (Brandrup,
Immergut, Grulke, Abe, & Bloch, 1999; Yaws, 2003). The missing
pure component properties, for instance, ash point, lethal concentration, solubility, etc., are predicted using the property prediction
tool (Hukkerikar, Sarup, et al., 2012). Note that experimental data
are used if they are available. The details on the chemicals database
are given in Table 2.
2.4. Property model library
A library of property models used to estimate mixture and pure
component properties has been developed. Tables 3 and 4 list the
mixture and pure component properties covered by this library.
The models are collected from the literature, and some of them are
developed using the group contribution approach. Linear mixing

Kinematic viscosity, 
Viscosity Index, VI
Higher Heating Value,
HHV
Density, 

Research Octane
number, RON
Reid Vapor Pressure,
RVP
Oxygen content, WtO2
Open cup ash point, Tf

Cost, C
Toxicity parameter,
LC50
Energy of mixing,
Gmix
Pour point, PP

f (
, )
f ()
f (HHVi , xi )

f (Ci , xi )
f (LC50,i , xi )

f (BIPPi , xi )

Table 4
Property model to estimate the pure component property.
Property of pure
compound

Model

Function

Higher heating value,

HHVi
Dynamic viscosity,
i

GC method (developed)

f (GC)

M&G GC method (Conte,

Martinho, Matos, & Gani,
2008)
Denition, (i =
i /i )
Correlation (Yaws, 2003)
Modied Racket Equation
(Spencer & Danner, 1972)
C&G GC method
(Constantinou & Gani,
1994)
Correlation
M&G GC based method
(Hukkerikar, Kalakul, et al.,
2012)

f (GC)

Kinematic viscosity, i
Vapor pressure, Pvap,i
Density, i
Open cup ash point,
Tf,i
Cost, Ci
Lethal concentration,
LC50,i

f (
i , i )
f (T)
f (Tc,i , Pc,i , T)
f (GC, Tfp,i )

f (i )
f (GC)

N.A. Yunus et al. / Computers and Chemical Engineering 66 (2014) 201213

rules are used to estimate those mixture properties for which it can
be assumed that the given property is additive. Additive properties
include oxygen content, standard specic gravity and normal boiling point. However, properties like viscosity, density, ash point,
pour point, and Reid vapor pressure (RVP) are not additive. Nevertheless, the linear mixing rule is employed to some of them as a
simplication in the blend design methodology. The models have
been veried using experimental data and found to be of acceptable accuracy. The estimated values of viscosity and density are
compared with the experimental data obtained from the literature
(Chen, Shen, Ko, & Chang, 2011; Kermanpour, Jahani, & Iloukhani,
2009; Teja & Rice, 1981; Totchasov, Nikiforov, & Alper, 2002).
Table 5 shows the average absolute error (AAE) of the estimated
properties for several mixtures at various temperatures.
3. Applications
The systematic methodology and its implementation are highlighted through two case studies: tailor-made gasoline blends and
base oil (lubricant) blends. The optimization problems (steps 3.1,
4.1 and 4.2, in Section 2.2) are solved using MATLAB, where the linear problem is solved using the linprog solver, while, the function
fmincon is used to solve the non-linear problem with the maximum
number of interations set to 1000.
3.1. Case study 1: design of gasoline blends
For gasoline, the following two issues need to be considered
among others; the rst is related to the security (or availability)
of crude oil supply and the second is related to the presence of
toxic constituents in gasoline that are harmful to the environment
as well as to humans. To address these issues, potential chemicals
derived from renewable sources are being blended with conventional gasoline. Adding bio-based chemicals can reduce the crude
oil consumption and the amount of released toxic chemicals. The
new formulation of gasoline blends should have a good fuel performance, should be safe and have low environmental impacts.
The objective of this case study is to design gasoline blends that
are suitable for a car (spark-ignition type) engine and used in a
hot climate with average ambient temperature of 27 C. Gasoline
is chosen as the main ingredient and is blended with chemicals
derived from renewable sources called bio-based chemicals. The
bio-based chemicals selected in the database are alcohols with low
carbon number (C2C5), ethers, ketones, acid and furan derivatives.
The blends may consist of two or more chemicals (in addition to the
gasoline chemicals) to form either binary or ternary mixtures with
the base gasoline.

207

3.1.1. Task 1: Problem denition

3.1.1.1. Task 1.1 Identify product needs. Based on the knowledge
base the gasoline blend must have the following characteristics: can
be burnt and run the engine efciently; can ow continuously from
the fuel tank to the combustion chamber; have a suitable ammability limit; and have low toxicity. In addition, the gasoline blends
must be stable, meaning that the blends do not evaporate easily; do
not oxidize to form unwanted by-products, such as gums, sludge
and deposits during storage; and must not split into two liquid
phases.
3.1.1.2. Task 1.2 Translate needs into physico-chemical properties.
Using the knowledge base, the product needs are translated to
target properties. The ability to start up a combustion engine is
depends on the amount of vapor generated initially. Gasoline is
sprayed to form fuel droplets, and then vaporized to form the fuel
vapor. The vapor pressure of gasoline is measured at a temperature of 308 K in a chamber with a vapor/liquid volume ratio of 4:1,
which is called the Reid vapor pressure (RVP) (Anderson, Kramer,
Mueller, & Wallington, 2010). The potential power of a fuel is measured from its heating value. The heating value of a fuel is dened as
the amount of heat released during a complete combustion of a unit
of fuel (Felipe Ramirez-Verduzco, Esteban Rodriguez-Rodriguez, &
Rayo Jaramillo-Jacob, 2012). It is also called as the heat of combustion, gross caloric value or total heating value. On the other hand,
the octane rating is widely used to measure the gasoline quality.
Low octane gasoline might cause engine knocking. Knock is caused
by pre-ignition or unwanted chemical reactions in the combustion
chamber, resulting in loud noise inside the engine. Long exposure
to knock may cause engine damage. Therefore, higher octane rating is needed in order to increase the engine efciency. The other
needs are translated as follows:
Ability to be burned:
Engine efciency:
Consistency of fuel ow:
Flammability:
Toxicity:
Stability:
Low oxidation:
Environmental aspect:

Reid vapour pressure (RVP)

Octane rating (RON) and heating value (HHV)
Dynamic viscosity () and density ()
Flash point (Tf )
Lethal concentration (LC50 )
Gibbs energy of mixing (Gmix )
Choice of chemicals
Oxygen content (WtO2 )

3.1.1.3. Task 1.3 Set target values. Referring to the existing products, legislation and literature study (van Basshuysen & Schfer,
2004; Forsythe, 2003) the target values for each property are set as
follows.
Ability to be burned :
Engine efciency :

45 RVP (kPa) 60

RON 92

(20)
(21)

Table 5
Average absolute errors (AAE) of estimated viscosity and density at different temperature (K).
Mixtures

AAE

Viscosity
n-Hexane + benzene
n-Hexane + toluene
n-Hexane + ethylbenzene
Methanol + octane
Ethanol + octane

298.15
1.09E02
0.90E02
1.73E02
293.15
0.70E02
263.15
9.15E02

308.15
1.65E02
0.71E02
4.59E02
298.15
0.65E02
268.15
6.87E02

323.15
0.76E02
0.57E02
1.34E02
303.15
0.59E02
273.15
5.97E02

313.15
0.51E02
283.15
5.28E02

293.15
3.92E02

333.15
0.16E02
293.15
8.36E04
4.25E04
8.66E04

353.15
0.15E02
298.15
8.08E04
4.10E04
8.69E04

303.15
30.1E04
3.92E04
8.74E04

308.15

3.89E04
8.60E04

313.15

3.69E04
8.55E04

Density
Heptane + octane
2-Methyl-1-butanol + dibutyl ether
EGMBE + 2-ethyl-1-butanol
2-Ethyl-1-butanol + dibutyl ether

208

N.A. Yunus et al. / Computers and Chemical Engineering 66 (2014) 201213

Table 6
Pseudo-component of gasoline to represent the MI.
Chemicals

Composition, wt%

Butane
Heptane
Iso-octane
1-Pentene
Methylcyclopentane
Toluene

6.58
12.60
53.99
3.63
8.47
14.73

HHV (MJ/kg) 35

(22)
0.30  (cP) 0.60

Consistency of fuel ow :
0.720  (g/cm3 ) 0.775
Flammability :
Toxicity :
Stability :

Tf (K) 300

log LC50 (mol/L) < 3.08

G

mix

<0

Environmental aspect :

(23)
(24)
(25)
(26)
(27)

2 WtO2 (wt%) 20

(28)

Low oxidation: Chemicals choice (no target values).

3.1.2. Task 2 Property model identication
3.1.2.1. Task 2.1 Retrieve the required property models from the property models library. Six of the target properties are estimated using
linear mixing rules, which are
, , RON, HHV, log LC50 , and WtO2 .
The linear mixing model is represented by Eq. (29). The properties
RVP and Tf are predicted using non-linear models, Eqs. (30)(31).
k =

NC


xi ik

(29)

lethal concentration values of gasoline and 2-methylpropanal

are 3.33 mol/L and 3.94 mol/L, respectively. The target value to
be achieved is less than 3.08 mol/L. Applying rule 1, the blend
of gasoline and 2-methylpropanal is rejected as both the target
property values are greater than the upper limit.
Level 2: Stability analysis
Step 2.1: The UNIFAC-LLE group representation is obtained for
28 chemicals (including the MI components), and the temperature is set at the ambient temperature.
Step 2.2: The stability test is performed using the STABILITY
tool (Conte et al., 2011) and the results for 378 binary mixtures
are retrieved.
Step 2.3: The results for binary and ternary mixtures are analyzed. 8 of the binary mixtures are partially miscible. However,
all the mixtures are considered for Level 3.
Level 3: Linear constraints
Step 3.1: Using the list of mixtures resulting from Step 2.1, the
blend composition ranges are calculated for all linear target properties: heating value, viscosity, density, octane number, lethal
concentration and oxygen content. The composition ranges are
obtained for binary mixtures.
Step 3.2: Identify the overall composition range for multicomponent blends. The results of this step gives the feasible
mixtures with their composition ranges dened by lower- and
upper-bounds.
Steps 3.1 and 3.2 are combined and solved as a linear optimization problem with the objective to minimize and maximize
the blend composition subject to the linear constraints, Eq. (33)
which represents RON, HHV, , , log LC50 and WtO2 to match the
target given by Eqs. (21)(24), (26) and (28), respectively.
min or max fobj (x)
s.t.

i
NC

xi i P sat (308 K)
i

i=1


NC

i=1

RVPB

k
LB

=1

(30)

=1

(31)

In Eq. (30), for given x and T = 308 K, RVPB is calculated. In Eq.

(31), given x and saturated vapor pressure of component i, Pisat at
its ash point, Tf,i , Tf is calculated.
3.1.3. Task 3 Mixture/blend design
3.1.3.1. Task 3.1 Collect input data. Conventional gasoline is
selected as MI and its composition is given in Table 6. 22 bio-based
chemicals are selected from the gasoline database section, where
these chemicals are used as building blocks in the tailor-made gasoline blend design.
3.1.3.2. Task 3.2 Generate and screen blends using the mixture/blend
design algorithm. The problem is solved using the mixture/blend
design algorithm.
Level 1: Pure component constraints
Step 1.1: Screen all the possible blends by comparing the heating value, viscosity, density, octane number, lethal concentration
and oxygen content of pure chemicals with the target values.
The combinations of mixtures that have pure property values
out of the target value ranges are rejected. For instance, the

NC


k
xi ik UB

(33)

i
NC


xi i Pisat (Tf )
Pisat (Tf,i )

(32)

xi 1 = 0

(34)

i=1

0 < xi < 1

(35)

The solution of the problem leads to 8 blends and 112 ternary

blends (note that this means the compounds of the model gasoline
plus 1 for binary and 2 for ternary blends) being selected for the next
step, while 8 binary blends and 109 ternary blends are rejected.
Step 3.3: Re-check the stability of the partially miscibility of the
blends in the region of interest. All the partially miscible mixtures
were removed after considering the linear constraints in Steps 3.1
and 3.2.
Level 4: Non-linear constraints
Step 4.1: The non-linear constraints RVP is now considered as
the non-linear constraint. The compositions from level 3 are used
as input in this step, and new composition ranges are obtained.
This step is solved as a non-linear optimization problem where
the objective functions are to be minimized and/or maximized
subject to the non-linear constraint, Eq. (37), to match the target
given by Eq. (20).
min or max fobj (x)

(36)

N.A. Yunus et al. / Computers and Chemical Engineering 66 (2014) 201213

Binary mixture

Ternary mixture
Level 1

22

231

Level 2

16
16

221
221

Level 3
5

67

Fig. 5. Number of gasoline blend candidates reduced after screening using the mixture/blend design algorithm.

s.t.

 xi i P sat (308 K)
NC

i=1
NC


RVPB

xi 1 = 0

=1

(37)

(38)

i=1

xLB < xi < xUB

further. Among the proposed gasoline blends, MeTHF becomes one

of the favorable chemicals because it is present in most blend candidates. MeTHF has good fuel properties, which are high-energy
content, low vapor pressure, moderate oxygen level and considerably toxicity content. These blend candidates, however, need
further validation through experimental work.

3.2. Case study 2: design of base oil blends for lubrication

112

Level 4

209

(39)

The solution of the above problem gives new composition ranges

that satisfy the linear and the non-linear constraints. The ash point
is calculated using Eq. (31) and the composition that satised the
RVP. All the blend candidates are satised the ash point constraint,
Eq. (25). After this step, 45 ternary blends are removed, while a total
75 of binary plus ternary blends are retained. None of the blends
were removed due to the calculated ash-point temperatures as
they were all within the specied bound. The number of blends
generated and screened at each level is shown in Fig. 5 for both
binary and ternary blends.
Step 4.2: The objective here is to obtain the minimum gasoline composition in the blend formulations. Since value is already
known from Step 4.1, this step is not necessary.
3.1.3.3. Task 3.3 Rank blend candidates according to a selection criterion. The blends that satised all the constraints are ranked
according to the minimum amount of gasoline in the blends. Table 7
gives the results, where the blends are listed in terms of decreasing
gasoline composition.
3.1.4. Task 4 Model-based verication
This task was not necessary for the nal gasoline blends as
the property models used are already validated with experimental
data. What could be useful, however, to check these blends under
engine conditions, which is outside the scope of this work.
3.1.5. Product analysis
The quality of fuel is measured from its energy content (HHV).
Gasoline blends proposed in Table 4 have a high HHV value. Their
HHV values are actually quite close to the heating value of a conventional gasoline (4447 MJ/kg). The RON is commonly used to
measure the performance of fuel. Nevertheless, the RON value cannot be predicted for some blends due to missing octane number
for pure components, so these blends could not be considered any

The most important component in a lubricant is the base oil,

which comprises 6598% of the total composition of the lubricant.
The base oil properties inuence the lubricant attributes. Base oil
is derived from three sources: petroleum, synthetic and biological,
i.e. originating from plants or animals. Mineral oils derived from
petroleum are the most widely used base oils. On the other hand,
synthetic oils are man-made oils with superior properties, thereby
performing well in extreme conditions. Vegetable oils are easily
degraded, but have poor properties, making them suitable only for
low demand applications. Comparing various types of base oils, it
can be concluded that synthetic oils have excellent physical and
chemical properties, but they are expensive, while mineral oils are
cheaper but less environmentally friendly, and vegetable oils are
biodegradable but have poor oxidative stability and cold ow properties. In term of composition, mineral oil contains a mixture of
many hydrocarbons. As a consequence, the base oil composition
is mostly identied by groups or lumps of compounds that fall
into a similar chemical classication. The components of mineral
oil may comprise different percentages of parafns, naphthenes
and aromatics. A high content of parafns makes mineral oil waxy,
with high melting point and suitable for high-temperature applications. Nevertheless, aromatics and unsaturated chemicals should
be reduced in order to avoid excessive lubricant oxidation. The
basic chemicals present in lube feedstocks are classied according to their content of parafns, napthenes and aromatic groups.
These chemicals are used as building block in the base oil mixture
design.
The base oil mixture is designed as single grade oil for gasoline
engines to be used in a hot climate with average ambient temperatures of around 27 C. The components of mineral oil may comprise
different percentages of parafns, naphthenes and aromatics. The
objective of this work is to design mixtures of base oils from mineral
oils and organic chemicals that have good lubrication properties
with low environmental impact. In order to achieve the abovementioned objective, a set of chemicals from lube feedstocks and
bio-based chemicals are used as building blocks. The mixtures of
base oil may form binary, ternary or multi-component mixtures.

3.2.1. Task 1: Problem denition

3.2.1.1. Task 1.1 Identify product needs. The needs for base oil were
dened using both a literature survey and the knowledge base. The
main function of lubricant base oil is to lubricate and prevent wear
between two moving surfaces. In addition, it must be able to resist
at high temperature and ow continuously at low temperature.
It should have a suitable specic gravity for handling purposes,
should be non-ammable and with low vaporization rate to avoid
losses to the environment. Besides, the base oil mixtures must be
stable, meaning that the mixtures do not oxidize to form deposits
in the system to be lubricated, and must be a single-phase.

3.2.1.2. Task 1.2 Translate needs into physico-chemical properties.

Using the knowledge base the product needs are translated to properties. The following target properties are used to measure the
related needs:

210

N.A. Yunus et al. / Computers and Chemical Engineering 66 (2014) 201213

Table 7
Gasoline blended candidates with their composition and properties.
No.

Composition (vol%)

1
2
3
4
5

G (69)
G (67)
G (72)
G (75)
G (77)

Properties

THF (11)
ACE(13)
ACE(10)
2BE (13)
ETOH(12) (12)

MeTHF(20)
MeTHF(20)
2BE(18)
MeTHF(12)
MeTHF(11)

HHV

WtO2

RVP

RON

log LC50

41
41
40
43
42

7.2
7.8
7.3
5.5
6.7

46
46
49
45
45

96

0.48
0.47
0.48
0.50
0.57

2.7
2.7
2.7
2.9
2.8

Note: Property abbreviations are given in the nomenclature list.

Ability to lubricate and prevent wear:

Resist at high temperature:
Ability to ow at the ambient temperature:
Handling purpose:
Non-ammable:
Low vaporization rate:

kinematic viscosity ()
viscosity Index (VI)
pour point (PP)
density ()
ash point (Tf )
volatility (Vvap )

3.2.1.3. Task 1.3 Set target values. Referring to the existing products
(Anderson et al., 2010; Kramer et al., 1999) as benchmark and using
the knowledge base, the target values for each property are set as
follows.
Ability to lubricate and prevent wear :
Resist at high temperature :

 (cSt) 4.119

VI 80

Ability to ow at ambient temperature :

Handling purpose :
Non-ammable :

(41)
PP (K) 293

0.80  (g/cm3 ) 0.9

Tf (K) 490

Low vaporization rate :

Vvap (wt%) 15

(40)

(42)
(43)
(44)
(45)

3.2.2. Task 2: Property model identication

3.2.2.1. Task 2.1 Retrieve the required property models from the property models library. Kinematic viscosity and density are estimated
using liner mixing rules Eq. (29). Density is blended linearly with
respect to the molar volume. Pour point is blended linearly using
blending index, Eq. (46) and the pour point of the blend is estimated using Eq. (47). On the other hand, the viscosity index (see Eq.
(48)) and ash point (Eq. (31)) are estimated using non-linear constraints. The vaporization rate is estimated using ash calculation
(Smith, Van Ness, & Abbott, 2005).

BIB =

n


xvi BIPPi

(46)

BIPPi = PPi

1/0.08

(47)

o B
100 VILB
o 100

(48)

i=1

3.2.3.2. Task 3.2 Perform mixture/blend design algorithm. The problem is solved using the mixture/blend design algorithm. At Level 1
of the algorithm, the properties of viscosity, density and pour point
are considered. Then the blend stability is checked at Level 2. All
the linear constraints, viscosity and density are considered at Level
3; and at Level 4 the blends that satisfy the non-linear constraints,
(viscosity index, pour point, ash point and vaporization loss) are
used.
Level 1: Pure component constraints
Step 1.1: This step is applied to all chemicals by comparing their
viscosity, density and melting points with their target values. 946
binary mixtures are screened and 901 are rejected. One example
is the binary mixture of n-hexacosane + 1,1 -Biphenyl, which is
rejected due to the pure component viscosities.
Level 2: Stability analysis
Step 2.1: The UNIFAC-LLE group representation is obtained for
44 chemicals, and the temperature is set at the ambient temperature and at the operating condition, 100 C.
Step 2.2: The stability test is performed using the STABILITY tool
and the results for 946 binary mixtures are extracted.
Step 2.3: The results for binary, ternary and multi-component
mixtures are analyzed for both temperatures. 58 binary mixtures are partially miscible and 34 of the binary mixtures are
found totally immiscible at one or both temperatures, and
thus they are rejected. For example, mixture of propane-1,2,3triol + n-tetradecanoic acid is partially miscible at the ambient
temperature. However, it is completely immiscible at 100 C.
Therefore, this mixture is not considered in the blend design.
Level 3: Linear constraints
Step 3.1: Using the list of mixtures resulting from Step 2.1,
the blend composition ranges are calculated for all linear target properties: Viscosity and density. The composition ranges are
obtained for binary mixtures.
Step 3.2: Identify the overall composition range for multicomponent (ternary and quaternary) blends. The results of this
step are mixtures with their composition ranges.
Steps 3.1 and 3.2 are combined and solved as a linear optimization problem while the objective functions are to minimize
and maximize the blend composition subject to the linear constraint, Eq. (50), which represents viscosity and density to match
the target values given by Eqs. (40) and (43).
min or max

(49)

s.t.
k
LB

3.2.3. Task 3: Mixture/blend design

3.2.3.1. Task 3.1 Collect input data. 44 basic chemicals, serving as
base oil feedstocks, and organic chemicals are selected from the
lubricant database section, and these chemicals are used as building
block in the blend design. In this case, MI is not specied.

fobj (x)

NC


k
xi ik UB

(50)

i
NC

i=1

xi 1 = 0

(51)

N.A. Yunus et al. / Computers and Chemical Engineering 66 (2014) 201213

Table 8
The reduced number of blend candidates using mixture/blend design algorithm.
Level

N
N1
N2
N3
N4

Mixture
Ternary

Quaternary

946
60
45
45
14

13,244
1847
1400
1400
966

135,751
30,645
17,427
17,427
5483

x|x Rn , 0 x 1

(52)

Level 4: Non-linear constraints

Step 4.1: Calculate the non-linear constraints: Viscosity index,
pour point, ash point and vaporization loss have been estimated
and the new composition ranges are obtained. The compositions
from the previous task are used as input in this step.
This step is solved as a non-linear optimization problem where
the objective functions are to minimize and maximize the blend
compositions subject to the non-linear constraints, Eqs. (54)(56)
to satisfy the target property values, Eqs. (41)(42). The ash point,
Eq. (44) and vaporization loss, Eq. (45) were calculated for the
blends that satised the non-linear constraints.
min or max fobj (x)

(53)

s.t.
n


xvi BIPPi

(54)

BIPPi = PPi

1/0.08

(55)

o B
100 VILB
o 100

(56)

i=1

NC


xi 1 = 0

(57)

i=1

xLB < xi < xUB

Viscosity

Binary
Ternary
Quaternary

Step 3.3: Re-check the stability of the partially miscible mixtures

in the region of interest. Two partially miscible mixtures are identied and their stability regions are compared with the region of
interest. One of the mixtures is rejected because it was found unstable at the region of interest, while the other is accepted because it
was found to be totally miscible in the region of interest.

BIB =

Table 10
Comparison of the linear and rigorous viscosity values.

Number of candidates
Binary

(58)

211

-linear

-rigorous

4.12
4.12
4.12

4.4559
4.4547
5.0110

About 51%, 7% and 68% of the binary, ternary and quaternary

blends are removed at this level, respectively. A total number of
14, 966 and 5483 of the binary, ternary and quaternary blends
are found to satisfy all the constraints, respectively. The number
of blends generated and screened at each level is listed in Table 8
for three types of blends.
Step 4.2: The objective function, Eq. (59) is optimized to obtain a
low-cost base oil blend. The new composition ranges obtained from
Step 4.2 are used as the upper and lower boundary. Then, the target
property values for all feasible mixtures are recalculated using new
compositions obtained from the optimization.
fobj = min

NC


xi Ci

(59)

3.2.3.3. Task 3.3 Rank blend candidates according to a selection criterion. The price of the blend is used as the selection criterion.
Since many blend formulations satised all the constraints, they
are ranked according to the minimum price that is achievable for
the selected purpose. The blends with lowest price for each type of
mixture are given in Table 9.
3.2.4. Task 4: Model-based verication
3.2.4.1. Task 4.1 Validation. The viscosities of blends listed in
Table 8 need further validation because all of them are non-ideal
mixtures. The viscosities of the blends are estimated using a rigorous model and compared with the values estimated using a
linear model. The comparison results are given in Table 10. The
rigorous values are slightly higher than the linear values. Even so,
the mixtures are still acceptable because the values obtained with
the rigorous model are within the target range of 4.1212.5 cSt. It
should also be noted that, in general, viscosity has a wider range of
permissible values.
3.2.5. Product analysis
A mixture with low cost is selected for each type of mixture as
listed in Table 9. The bio-based chemical, propane-1,2,3-triol (GLY)
is present in all blends, while cis-9-Octadecenoic acid (9ODA) is
present in blend 3. It indicates that the consumption of mineral base

Table 9
The base oil blends at minimum price with their estimated property values.
Blend

xi

Cost ($/L)

Properties


Binary
2MT +
GLY
Ternary
3ME +
3ET +
GLY
Quaternary
2MT +
DFE +
GLY +
9ODA

VI

PP

SG

Tf

Vvap

0.56
0.44

7.80

4.12

95

263

0.9790

494

0.0787
0.5447
0.3765

6.88

4.12

101

273

0.9781

565

0.5820
0.0519
0.3561
0.0100

6.63

4.12

108

283

0.9778

566

Note: Property abbreviations are given in the notations list.

212

N.A. Yunus et al. / Computers and Chemical Engineering 66 (2014) 201213

oil can be reduced by replacing it with base oil derived from renewable sources. As the price of bio-based chemicals is currently higher
than mineral based oils, their blends also have higher prices. However, high viscosity indicates good quality of the lubricant, which
is achieved in the blends listed in Table 9.

4. Conclusions and future work

A systematic methodology for design of tailor-made blended
products has been developed, and was highlighted with case studies on design of tailor-made gasoline blends, and base oil lubricant
blends. A decomposition-based method has been applied to solve
the blending problem, where the objective was to quickly screenout a large number of alternatives and to reduce the search space
at each hierarchical step. The shortlisted blend candidates, however, need further experimental verication before moving in to
production.
The knowledge base method has been used for identication of
product needs, translation to properties and the setting of the target
values in Task 1. Information found in the literature and information on existing products has formed the basis of this work. The
availability of a proper knowledge base is very critical in product design. Currently, there is no established knowledge base or
database format for chemical product design. Product design is
still mostly relying on the expertise of the designer. For example,
technical understanding of the functionality of a designed product is required for translation of this functionality into a number
of target properties. This understanding is considered as knowledge, which should be transferred into a proper database. It should
be an open database, where laboratories, chemists, engineers and
researchers can share their knowledge and experiences. In order
to have a powerful database, input from various stakeholders is
required because the product design itself is a multi-disciplinary
task. A comprehensive database for chemicals based product design
is under development.
Tools such as chemical databases and a property models library
play important roles in the mixture/blend design as chemicals
in the database act as the starting building blocks. The size of
the chemicals database can be further increased by adding more
chemicals, especially bio-based chemicals. The choice of chemicals in the database is a key control issue with respect to the
complexity of the mixture/blend problems. The property model
selection can also inuence the performance of the blended products. The linear mixing rule applied in this work is only suitable
for the ideal mixtures. Thus, for the non-ideal mixtures, they need
to be veried using rigorous models. Some of the pure property
data are missing, for example, octane number (ON) and cloud
point (CPt). Models to predict the octane number for hydrocarbon compounds are available in literature (Albahri, 2003; Lapidus,
Smolenskii, Bavykin, Myshenkova, & Kondratev, 2008; Pasadakis,
Gaganis, & Foteinopoulos, 2006; Perdih & Perdih, 2006; Smolenskii,
Vlasova, & Lapidus, 2004). Nevertheless, none of them were found
to be suitable for the estimation of the ON for oxygenated compounds (ketone, alcohol, acid, ester and ether). Therefore, new
models need to be developed.
The mixture/blend design algorithm guided the screening of
alternatives wisely. At each level, rules are applied to eliminate
the undesired chemicals or blend candidates, thus reducing the
number alternatives. Selection criteria are introduced to rank the
blend candidates that satisfy all the constraints. The criteria are
set according to the product design objectives, which can be cost,
the product performance, the chemical composition, as well as the
target properties. For instance, gasoline is ranked according to the
composition of the gasoline in blends because the objective is to
reduce the fossil-fuel consumption. Besides, their heating value

also can be used to rank the blend selection since it determines

the energy content of the fuel.
The methodology can be used to design or tailor many liquid
blended products, where the scope and size depend on the chemicals available in the databases and the models available in the
property model library. Current work has considered the product
design using a model-based approach. A logical next step is to add
an experimental component to verify the designed products.

Acknowledgements
The nancial support for this PhD project, which was funded
by the Malaysian Ministry of Education (MoE) and Universiti
Teknologi Malaysia (UTM), and partially funded by the Qatar
National Research Fund (QNRF), is gratefully acknowledged.

References
Albahri, T. A. (2003). Structural group contribution method for predicting the octane
number of pure hydrocarbon liquids. Industrial & Engineering Chemistry Research,
42, 657662.
Anderson, J. E., Kramer, U., Mueller, S. A., & Wallington, T. J. (2010). Octane numbers
of ethanol- and methanol-gasoline blends estimated from molar concentrations.
Energy & Fuels, 24, 65766585.
Brandrup, J., Immergut, E. H., Grulke, E. A., Abe, A., & Bloch, D. R. (1999). Polymer
handbook (4th ed.). New York: John Wiley & Sons.
Cao, W., Knudsen, K., Fredenslund, A., & Rasmussen, P. (1993). Group-contribution
viscosity predictions of liquid-mixtures using UNIFAC-VLE parameters. Industrial & Engineering Chemistry Research, 32, 20882092.
Chen, W. K., Shen, Y. C., Ko, J. W., & Chang, C. M. (2011). Excess molar volumes,
densities and P-T-x phase equilibria for binary mixtures of benzene + toluene,
benzene + N-methylformamide (NMF), heptane + octane, toluene + m-xylene
and toluene + NMF at 333.15 K and 353.15 K. Journal-Taiwan Institute of Chemical
Engineers, 42, 441446.
Cheng, Y. S., Lam, K. W., Ng, K. M., Ko, R. K. M., & Wibowo, C. (2009). An integrative
approach to product developmentA skin-care cream. Computers and Chemical
Engineering, 33, 10971113.
Churi, N., & Achenie, L. E. K. (1997). The optimal design of refrigerant mixtures for
a two-evaporator refrigeration system. Computers & Chemical Engineering, 21,
S349S354.
Constantinou, L., & Gani, R. (1994). New group-contribution method for estimating
properties of pure compounds. AIChE Journal, 40, 16971710.
Conte, E., Martinho, A., Matos, H. A., & Gani, R. (2008). Combined group-contribution
and atom connectivity index-based methods for estimation of surface tension
and viscosity. Industrial & Engineering Chemistry Research, 47, 79407954.
Conte, E., Gani, R., & Ng, K. M. (2011). Design of formulated products: A systematic
methodology. AIChE Journal, 57, 24312449.
Conte, E., Gani, R., Cheng, Y. S., & Ng, K. M. (2012). Design of Formulated Products:
Experimental Component. AIChE Journal, 58, 173189.
Cussler, E. L., & Moggridge, G. D. (2011). Chemical product design (2nd ed.). New York:
Cambridge University Press.
Fahim, M. A., Alsahhaf, T. A., & Elkilani, A. (2010). Chapter 9 Product blending. In
Fundamentals of petroleum rening. Amsterdam: Elsevier.
Felipe Ramirez-Verduzco, L., Esteban Rodriguez-Rodriguez, J., & Rayo JaramilloJacob, A. (2012). Predicting cetane number, kinematic viscosity, density and
higher heating value of biodiesel from its fatty acid methyl ester composition.
Fuel, 91, 102111.
Forsythe, W. E. (2003). Smithsonian physical tables (9th ed.). New York: Knovel.
Gani, R., Nielsen, B., & Fredenslund, A. (1991). A group contribution approach to
computer-aided molecular design. AIChE Journal, 37, 13181332.
Gani, R., & Pistikopoulos, S. (2002). Property modelling and simulation for product
and process design. Fluid Phase Equilibria, 194197, 4359.
Hukkerikar, A. S., Kalakul, S., Sarup, B., Young, D. M., Sin, G., & Gani, R. (2012).
Estimation of environment-related properties of chemicals for design of sustainable processes: Development of group-contribution(+) (GC(+)) property
models and uncertainty analysis. Journal of Chemical Information and Modeling,
52, 28232839.
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.
Karunanithi, A. T., Achenie, L. E. K., & Gani, R. (2005). A new decomposition-based
computer-aided molecular/mixture design methodology for the design of optimal solvents and solvent mixtures. Industrial & Engineering Chemistry Research,
44, 47854797.
Kermanpour, F., Jahani, H., & Iloukhani, H. (2009). Excess molar volume and
derived thermodynamic properties of binary mixtures of 2-methyl-1-butanol
and 2-ethyl-1-butanol + different ethers at the temperature range of 293.15 to
313.15 K. Journal of Molecular Liquids, 146, 2934.

N.A. Yunus et al. / Computers and Chemical Engineering 66 (2014) 201213

Kramer, D. C., Ziemer, J. N., Cheng, M. T., Fry, C. E., Reynolds, R. N., Lok, B. K., et al.
(1999). Inuence of group II and III base oil composition on VI and oxidation
stability. In NLGI annual meeting (p. 66).
Lapidus, A. L., Smolenskii, E. A., Bavykin, V. M., Myshenkova, T. N., & Kondratev, L.
T. (2008). Models for the calculation and prediction of the octane and cetane
numbers of individual hydrocarbons. Petroleum Chemistry, 48, 277286.
Liaw, H., Lee, Y., Tang, C., Hsu, H., & Liu, J. (2002). A mathematical model for predicting the ash point of binary solutions. Journal of Loss Prevention in the Process
Industries, 15, 429438.
Liaw, H., Tang, C., & Lai, J. (2004). A model for predicting the ash point of ternary
ammable solutions of liquid. Combustion and Flame, 138, 308319.
Magnussen, T., Rasmussen, P., & Fredenslund, A. (1981). UNIFAC parameter table
for prediction of liquidliquid equilibriums. Industrial & Engineering Chemistry
Process Design and Development, 20, 331339.
Nielsen, T. L., Abildskov, J., Harper, P. M., Papaeconomou, I., & Gani, R. (2001). The
CAPEC database. Journal of Chemical and Engineering Data, 46, 10411044.
Pasadakis, N., Gaganis, V., & Foteinopoulos, C. (2006). Octane number prediction for
gasoline blends. Fuel Processing Technology, 87, 505509.
Perdih, A., & Perdih, F. (2006). Chemical interpretation of octane number. Acta Chimica Slovenica, 53, 306315.
Rizvi, S. Q. A. (2009). Comprehensive review of lubricant chemistry, technology, selection, and design. New York: ASTM International.
Smith, J. M., Van Ness, H. C., & Abbott, M. M. (2005). Introduction to chemical engineering thermodynamics. Boston: McGraw-Hill.

213

Smolenskii, E. A., Vlasova, G. V., & Lapidus, A. L. (2004). A Study of the

structureoctane number relationship for hydrocarbons. Doklady Physical Chemistry, 397, 145149.
Spencer, C., & Danner, R. (1973). Prediction of bubble-point density of mixtures.
Journal of Chemical and Engineering Data, 18, 230234.
Spencer, C., & Danner, R. (1972). Improved equation for prediction of saturated liquid
density. Journal of Chemical and Engineering Data, 17, 236241.
van Basshuysen, R., & Schfer, F. (2004). Internal combustion engine handbookBasics,
components, systems, and perspectives. United State: Society of Automotive Engineers, Inc.
Teja, A. S., & Rice, P. (1981). The measurement and prediction of the viscosities of
some binary liquid mixtures containing n-hexane. Chemical Engineering Science,
36, 710.
Teixeira, M. A., Rodriguez, O., Rodrigues, S., Martins, I., & Rodrigues, A. E. (2012). A
case study of product engineering: Performance of microencapsulated perfumes
on textile applications. AIChE Journal, 58, 19391950.
Totchasov, E. D., Nikiforov, M. Y., & Alper, G. A. (2002). Viscosity of the systems
methanol-n-octane and ethanol-n-octane. Russian Journal of General Chemistry,
72, 14981500.
Yaws, C. L. (2003). Yaws handbook of thermodynamic and physical properties of chemical compounds. New York: Knovel.
Yunus, N. A., Gernaey, K. V., Woodley, J. M., & Gani, R. (2013). Design of sustainable
blended products using an integrated methodology. Computer Aided Chemical
Engineering, 32, 835840.