Accepted Manuscript

Development of composites based on recycled polypropylene for injection moulding

automobile parts using hierarchical clustering analysis and principal component
estimate
Fu Gu, Philip Hall, N.J. Miles
PII:

S0959-6526(16)30916-7

DOI:

10.1016/j.jclepro.2016.07.028

Reference:

JCLP 7600

To appear in:

Journal of Cleaner Production

Received Date: 31 July 2015

Revised Date:

29 June 2016

Accepted Date: 5 July 2016

Please cite this article as: Gu F, Hall P, Miles NJ, Development of composites based on recycled
polypropylene for injection moulding automobile parts using hierarchical clustering analysis and principal
component estimate, Journal of Cleaner Production (2016), doi: 10.1016/j.jclepro.2016.07.028.
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT
Development of composites based on recycled polypropylene
for injection moulding automobile parts using hierarchical
clustering analysis and principal component estimate
Fu GU, Philip HALL*, and N J MILES
Faculty of Science and Engineering, University of Nottingham, Ningbo, China

RI
PT

Abstract: Adapting recycled plastics into demanding manufacturing industry is

helpful in promoting recycling rate of waste plastic scrap, but yet only a few

researchers are tackling with this issue and recycled plastics are still ending up in
low-value products. This paper presents a novel method to adapt recycled plastic for
demanding industrial applications by designing suitable formulae which satisfy the
technical requirements of the applications. To illustrate the proposed method which

SC

combines hierarchical clustering analysis and principal component estimate, actual

requirements of some selected automobile parts were acquired and benchmarked in
this study. Two common particle fillers - talcum powder and organically modified

M
AN
U

montmorillonite were added in recycled polypropylene with maleic anhydride

grafted polypropylene as a compatibiliser. The initial compositions were selected
according to the specific composition selection rules which are adapted from Taguchi
method for reducing the number of trials. Corresponding mechanical, rheological,
and thermal properties were tested based upon the technical requirements of
selected automobile parts. The interrelationships between multiple objectives
(requirements) were analysed and classified by hierarchical clustering analysis, and

TE
D

then the total number of requirements were reduced. Effects of each component
were identified numerically by principal component estimate, and a corresponding
linear regression model was obtained. The linear regression model was compared to
other linear regression models which obtained by other mathematic techniques, and
it has been proved to be the best model which has the smallest gap between

EP

predicted values and experimental results. Optimal formulae were calculated via
linear programming with the objects of minimising material cost and satisfying the
reduced technical requirements of selected automobile parts. In verification tests,

AC
C

the experimental performance of the obtained formulae closely matched the model
predictions. The proposed formula design method is novel and original, and it is
shown to be effective and efficient in designing recycled plastic based composites
for demanding industrial applications.
Keyword: Manufacturing; Decision support system; Polypropylene; Principal
component estimate; Hierarchical cluster analysis
1. Introduction
Plastic, which is usually synthesized using non-renewable fossil resources, is one of
the major materials used in modern life. With a continuous growth over 50 years,
the global plastic production in 2013 rose to 299 million tonnes (PlasticsEurope,
2014). According to the statistics (PlasticsEurope, 2014), 25.2 million tonnes of
post-consumer plastics waste in 2012, only 26% is recycled and 36% is recovered
Tel: +86 574 8818 0018
Email:Philip.Hall@nottingham.edu.cn

ACCEPTED MANUSCRIPT
as energy resources while 38% still went to landfill. Toxic compounds including
persistent organic pollutants (POPs), such as dioxins and polychlorinated biphenyls
(PCBs), are emitted from the energy recovery processes with exhaust gases
(Menada et al., 1998; Caneghem et al., 2010; Font et al., 2011). The treatment of
POP emission in those processes is expensive and ineffective, as dioxins will be
reproduced afterward (Grosso, 2009). Thus, energy recovery is not encouraged due
to environmental concerns (Huang et al., 2013). Due to plastic's low density, landfill

RI
PT

of waste plastic takes up a huge amount of valuable land and causes other

environmental problems, such as POP leakage in surface soil (Melnyk et al., 2015)
and generation of greenhouse gases (Karanjekar et al., 2015). Despite odorous
emissions released during meltdown of plastics can affect environmental and

human health (Tsai et al., 2009; Huang et al., 2013), recycling of waste plastics is
economically viable (Al-Salem et al., 2009; Meng et al., 2015). From environmental

SC

protection perspective, recycling plastic could reduce 80% greenhouse gas emission
(Makuta et al., 2000) and 24% carbon footprint (Dormer et al., 2013) of material
production. Plastic scrap can be considered as a renewable source of material which

M
AN
U

fits with the idea of socio-ecological sustainability (Meran et al., 2008;

Sommerhuber et al., 2016). Thus, using recycled plastic in manufacturing is a
promising practice and is an essential consideration under the context of
sustainable development (Lindahla et al., 2014).

An important way to promote the recycling rate of plastic waste is to extend the use
of recycled plastic in manufacturing (Gu et al., 2016). Due to chemical and physical

TE
D

changes in the molecular structures or contaminations during the first life, recycled
plastic tends to have less desirable properties which limit their application (Ehrig,
1992; Gu et al., 2014; Sommerhuber et al., 2016). A number of different methods
were proposed to promote the performance of recycled plastic, such as additives
(Bahlouli et al., 2006; Putra et al., 2009; Sommerhuber et al., 2016),

EP

compatibilisation (Bertin and Robin, 2002; Forteln et al., 2002; Luzuriaga et al.,
2011) and procedure/formula optimisation (Khan et al., 2010; Gu et al., 2014;
Homkhiew et al., 2014). Before being used in manufacturing industry, material

AC
C

needs to meet a series of technical requirements, such as mechanical, rheological

and thermal performance . Yet, most of those existing methods only focus on
improving properties of particular aspect, such as tensile properties (Bahlouli et al.,
2006; Putra et al., 2009). Although there have been some reported works about
optimising the procedures (Gu et al., 2014) or formulas (Homkhiew et al., 2014) for
adapting recycled plastics in manufacturing, there is only limited research to design
recycled plastic based composites for real industrial applications. Homkhiew et al.
(2014) proposed a formula design method for designing recycled plastic based
composite via numerically modelling of the effects of each component. Although the
design output meet the proposed targets, the method requires a large number of
trials on each composition (Homkhiew et al., 2014) which is costly, time-consuming
and environmental-unfriendly. Same weakness was also found in other composite
design methods (Stark and Matuana, 2003; Soury et al., 2009; Zolfaghari et al.,

ACCEPTED MANUSCRIPT
2013). Thus, despite countless methods which aim at improving recycled plastic
have been reported, most of the recycled plastic in the current market still ends up
in producing low-value items, such as trash bags and bins (WRAP, 2010).
With the purpose of adapting recycled plastic in more demanding industrial
applications, a novel method is proposed to design recycled plastic based
composites using a combination of hierarchical clustering analysis (HCA) and

RI
PT

principal component estimate (PCE). Determination of proper formula for recycled

plastic based composite involves solving multi-objective collinear problems, in
which two or more compositions or properties might be highly correlated

(Homkhiew et al., 2014). Meanwhile, the requirements of demanding industrial

applications could be complicated and numerous. The proposed formula design

method is solving the multi-objective collinear problem by starting with reducing

SC

the dimensions, in this case, number of technical requirements for manufacturing

selected automobile parts. HCA is a method of cluster analysis which seeks to build
a hierarchy of clusters, which is used for identifying and presenting the

M
AN
U

interrelationships (similarities) between multiple properties in form of dendrograms

(Manly, 1994). PCE, or so-called principal component regression, is a biased
regression method based on principal component analysis (PCA) for handling the
collinear modelling problem among multi-dimensional variables (Massy, 1965;
Aguilera et al., 2006). Although these two methods have been frequently applied in
environmental science (Saleh et al., 2008; Sautot et al., 2015), agriculture (Chen et
al., 2015; Zafiriou et al., 2015), biomedical (Schumann et al., 2013), only a handful

TE
D

literatures focus on plastic engineering or manufacturing applications are found

(Yang and Gao, 2006; Park et al., 2015) and none of them is involving recycled
plastic. In our previous works (Gu et al., 2014; Gu et al., 2016), HCA was used for
preliminary analysis of the tested properties while PCA were used for evaluating the
performance of composites. Yet, both HCA and PCE are never being applied in the

EP

field of composite design.

In this study, the combined method is proposed with the aim of designing suitable

AC
C

composites based on recycled materials for demanding industrial applications. In

this method, HCA is used for analysis and to reduce the number of the requirements
for the automobile applications. PCE is used to analysis the effects of each
composition in the materials and to obtain linear regression model, then
multi-objective linear programming is performed to find composites meet multiple
requirements. The requirements of the selected automobile parts were obtained
and used as objectives for composite design. Virgin and recycled polypropylene (PP)
were used as base materials; talcum powder (talc), organically modified
montmorillonite (OMMT) and maleic anhydride grafted polypropylene (MAPP) were
used as additives. The initial compositions were selected by the proposed rules, and
their mechanical and rheological properties, and thermal stability were tested
accordingly. Based upon the experimental results, the number of requirements was
reduced by HCA and the effects of each component were modelled. Design of

ACCEPTED MANUSCRIPT
composites was carried out based upon the models for satisfying both technical
requirements and minimum material cost criteria. A verification testing set was
carried out to prove the effective of the proposed method. By the utilisation of this
method, the use of recycled plastic can be extended to production of high-value
products, which will promote the overall recycling rate of plastic scrap and reduce
environmental impact resulting from landfill or combustion. The schematic diagram
of the method proposed in this study is shown as Fig.1. The proposed formula

RI
PT

design method is a multi-objective optimisation approach which aims at promoting

the use of recycled plastic in demanding industrial applications by achieving
following objectives:

(1). Reducing the number of initial compositions and experimental trials effectively
and efficiently;

(2). Reducing the multiple technical requirements of the selected industrial

SC

applications effectively reduced by Identifying and classifying their

interrelationships via HCA;

(3). Identifying the effects of each component in the composite numerically via the

M
AN
U

linear regression model of components' effects obtained by PCE, and verifying the
model by comparison of model predictions and experimental results;
(4). Verifying the optimal formulae acquired via the proposed method by verification
tests to ensure all desirable properties have met the technical requirements while

AC
C

EP

TE
D

minimum material cost is achieved.

ACCEPTED MANUSCRIPT
Initial Composition
Groups

Required

RI
PT

Experimental Data

Data

PCE

HCA

Linear Regression
Model

M
AN
U

Standardized Data

Clustered Multiple
Properties

Composition

Industrial

Formulae

Requirements

TE
D

Fig.1.

SC

Preprocessing

EP

The schematic procedure of the formula design method proposed

2. Requirements of Industrial Applications
Requirements for some common automobile parts which made from PP were

AC
C

obtained from our industrial collaborator with their testing standards and testing
parameters (Tianqiang, 2014), which are shown in Table 1. The examples of the
engine cover and the fenders are shown in Fig.2 and Fig.3 respectively.
Table 1.

The technical requirements and testing methods of selected automobile parts

Application

Tensile

Flexural

Flexural

Strength

Modulus

Strength

(MPa)

(MPa)

(MPa)

Charpy
Impact
Strength
-2

(KJ m )

Melt Flow
Index
(g 10min-1)

Temperature of
Deflection under
Load
(C)

Meter Holder

20

2400

35

40

85

Engine Cover

20

2200

35

40

92

Fender

22

2000

30

60

90

ACCEPTED MANUSCRIPT
Testing
Standards

ISO527[26]
Gauge

Testing

speed: 50
mm min

ISO178

[27]

[27]

Gauge

Gauge

speed: 2

speed: 2

mm

mm

min-1

min-1

-1

ISO1133

ISO179 [28]

[29]

Hammer

ISO75 [30]

Load:

selection: 7.5

230C,

-1

J, 3.8m sec

Load: 0.45 MPa

2.16kg

M
AN
U

SC

RI
PT

Parameters

ISO178

Fig.2.

AC
C

EP

TE
D

The engine cover

Fig.3.
The fenders
In this study, the technical requirements of meter holder, engine cover and fender
can be classified into three categories: mechanical, rheological and thermal. The
mechanical properties include tensile/flexural/Charpy impact strengths (TS/FS/IS)
and flexural modulus (FM), the rheological property is melt flow index (MFI), and the
6

ACCEPTED MANUSCRIPT
thermal property is referred to temperatures of deflection under load (TDL). The
material for meter holder to form the instrumental panel and requires the highest
level of surface rigidity (highest flexural modulus and strength), while the material
for fender requires more impact resistance. The engine cover is supposed to
withstand a higher working temperature, thus higher TDL value is expected.
Additionally, due to the size of the engine cover is the largest of all three parts, a
better rheological property is needed for injection moulding due to the

RI
PT

corresponding mould capacity and cycle time limit, which resulting in the highest
MFI required value.

The upper and lower limits of range for the possible compositions are shown in Table
2, which were obtained based upon past experience, previous literature (Bahlouli et

Table 2.
Restrictions of compositions in composites

SC

al., 2006) and communication with the industrial collaborator (Tianqiang, 2015).

Restriction (wt.%)

Recycled Content

60

Particle Filler Content

20

Compatibiliser

3Compatibiliser10

3. Experimental Approach
3.1. Materials

M
AN
U

Composition

TE
D

The virgin PP material (VPP) used in this study is a type of block co-polymer mainly
being used for manufacturing automobile parts electric and electronic devices or
pipelines, with a trade-name of PPB-MO2-V and manufactured by Yangzi
Petrochemical Co., Ltd. These pellets have an average of particle size of 3.0 mm,
density of 0.9 g cm-3, see Fig.4 (left), and were used as received. The recycled PP

EP

material (RPP) used in this study is grey pellet which recovered from white
post-customer storage boxes and other plastic furniture. The recyclates were
purchased from a Ningbo recycling plant. The pellets have an average of particle

AC
C

size of 3.0 mm, density of 1.0 g cm-3, see Fig.4 (right), and were used as received.

Fig.4.
7

ACCEPTED MANUSCRIPT
The virgin and recycled PP materials: left virgin PP; right recycled PP
There were two types of particle fillers selected in this study talc and OMMT, as the
practice of adding both materials into recycled plastic to promote the performance
has been carried out frequently (Bahlouli et al., 2006; Zhang et al., 2009). The talc
used in this study was acquired from a Ningbo factory, which has an average of
particle size of 12.5 m, a density of 2.7 g cm-3, as shown in Fig.5 (left), and is used

RI
PT

as received. The OMMT used in this study was acquired from Yoshida Chemical

co.ltd., Shenzhen, and has an average of size of 37.5 m, a density of 1.7 g cm-3, as

M
AN
U

SC

shown in Fig.5 (right), and was used as received.

Fig.5.

TE
D

The talc and OMMT materials used in this study: left talc; right OMMT
MAPP is one of the most frequently used compatibilisers which improves the
interface reactions between different layers, such as virgin-recycled PP and various
additives (Bahlouli et al., 2006; Gu et al., 2014; Homkhiew et al., 2014). The MAPP

EP

used in this study was acquired from Nanjing Deba Chemical Co.Ltd, which has an
average of particle size of 2.5 mm, a density of 0.9 g cm-3, with the grafted rate of

AC
C

0.8%, and was used as received.

In a mixture, the weight proportions (expressed as wt.%) of compositions are not
independent, for they are none-negative values and must add up to 100%. In this
study, the compositions are denoted by VPP (c1), RPP (c2), Talc (c3), OMMT (c4), and
MAPP (c5).Based on the nature of a mixture and content in Table 2, the initial
constraints of designing such composites are obtained and shown in Table 3.
Table 3.
The initial constraints for designing composites
Constraint (wt.%)
c1+c2+c3+c4+c5=100
ci0, i=1,2,3,4,5
c260
8

ACCEPTED MANUSCRIPT
c3+c420
3c510
The initial groups of various proportional compositions are shown in Table 4. The
selected formulae are unlike the selected formulae in existing articles (Soury et al.,
2009; Zolfaghari et al., 2013; Homkhiew et al., 2014) which contain all possible
detailed compositions. There are two specific rules in this composition selection

RI
PT

which were proposed for promoting both computational efficiency and analysing
completeness:

(1). Each component must combine with every other component at least once.

(2). The total appearance of each component must no less than the total number of
components.

Table 4.
The initial compositions of blends (wt.%)
VPP(c1)

RPP(c2)

Talc(c3)

99

90

OMMT(c4)

M
AN
U

Trial.No

SC

The above rules were adapted from Taguchi method (Ross, 1996).

MAPP(c5)

96

15

10

80

15

70

30

80

3
4

70

8
9
10

10

10

50

40

10

55

15

15

20

30

25

20

TE
D

EP

These rules were originally proposed, as a novel adaptation of the principals of

Taguchi method (Ross, 1996). The compositions were used as basis for
experimental approach and design procedure. Compared to the reported methods

AC
C

(Stark and Matuana, 2003; Soury et al., 2009; Zolfaghari et al., 2013; Homkhiew et
al., 2014), the total number of experimental trials was significantly reduced, thus
the relating cost and environmental impact could also be reduced as well as the
computational complexity. The effectiveness of these composition selection rules
would be proved in following sections by comparingpredictions and experimental
results.

3.2. Sample Preparation

All materials were initially processed by an electrical mixer with speed of 60 rpm for
5 min according to proposed compositions in Table 4. Then mixed materials were
blended by a Kangrun KRSHJ-20 extruder, a co-rotating, intermeshing twin-screw
extruder, whose screw diameter is 20 mm and L/D=44. A single-screw feeder
attached to the hopper is used for all the PP pellets. The processing temperatures
9

ACCEPTED MANUSCRIPT
were increased from 175oC to 200oC from the hopper to the 3rd barrel, the
temperatures of the last three barrels were remained at 200oC, while the die
temperature was set at 200oC. The screw rotation speed was set at 180 rpm, and
the total mass flow rate was 5 kg h-1. Blended strands were extruded into a water
bath for cooling and pelletized.
The extruded pellets were dried in a dry oven at 85oC for 12 h with constant air flow

RI
PT

before being fed into the injection moulding machine. The purpose of oven drying is
to keep moisture content in plastic pellets below 1 wt.%. Then, these pellets are

injection moulded into ISO standard test specimens using a Haitian MA1200/370
injection moulding machine. There are 2 (tensile pieces) (ISO 527-2:2012) or 4

(flexural, impact and thermal stability pieces) (ISO 75-2:2004; ISO 178:2010; ISO
179-2:2010) test specimens per single injection moulding processing cycle. The

SC

temperatures of five heating barrels were set at 190oC-195oC-200oC-205oC-205oC.

The other processing parameters were: (1). an injection pressure of 50 MPa; (2). an
injection speed of 50 g s-1; (3). a packing pressure of 30 MPa for 5 s; (4). a cooling

M
AN
U

time in moulds of 5 s; (5). A mould temperature was set at 50oC. All these
processing parameters were set based real automobile production (Tianqiang,
2015), as a few necessary modifications were made to adapt the recycled materials.
During this study, all these processing parameters were maintained constant for
yielding comparable results.
3.3. Tests

TE
D

The tests were carried out in accordance with the requirements and methods in
Table 1, as selected mechanical, rheological and thermal properties were tested
respectively.
3.3.1. Mechanical

EP

The mechanical properties of composites were evaluated in terms of tensile, flexural,

and impact properties as shown in Table 1, and all sample specimens were
conditioned at 23oC and 50% R.H. for over 88 h before testing in accordance with

AC
C

ISO291 specifications (ISO 291:2008).

The tensile property tested is tensile strength (TS) accordingly, assessed in
accordance with ISO527 specifications (ISO 527-2:2012). The gauge distance is
110 mm with fixers' moving speed of 50 mm min-1 according to Table 1, sampling
rate at 200 points (pts) s-1, full-scale load range of 20 kN, performed on a Gotech
Universal Testing Machine (model TCS-2000NE) at room temperature of 23oC and at
50 % R.H.
The flexural properties tested are flexural modulus (FM) and flexural strength (FS),
assessed in accordance with ISO178 specifications (ISO 178:2010). The span is set
to 64 mm at a crosshead speed of 2 mm min-1 according to Table 1, sampling rate at
200 pts s-1, full-scale load range of 20 kN, performed on a Gotech Universal Testing

10

ACCEPTED MANUSCRIPT
Machine (model TCS-2000NE) at a room temperature of 23oC and at 50 % R.H.
The impact property tested is the Charpy impact strength (IS), gained in accordance
with ISO179 Charpy Impact specifications (ISO 179-2:2010). The span is set at 62
mm with the hammer of 7.5 J and impact speed of 3.8 m s-1 according to Table 1,
performed on a Gotech Impact Tester (model GT-7045-NHR) at a room temperature

RI
PT

of 23oC and at 50 % R.H.

8 sample pieces were tested for each mechanical property and the average result
taken if the coefficient of variance of tested results met the required limits (5 % in
accordance with ISO2602 specification, ISO 2602:1980).
3.3.2. Rheological

SC

Melt flow index test was conducted to determine the basic rheological behaviour of
the blends in accordance with ISO1133 Melt Mass-flow Rate specifications (ISO
1133-1:2011), measured in g/10 min under a load of 2.16 kg at 230C according

M
AN
U

to Table 1, using a Gotech melt flow indexer (model GT-7200-MIBH), at room

temperature of 23C and at 50 % R.H. Samples were pre-heated to 230C for 4
min.
3.3.3. Thermal Stability

In automobile applications, materials are required for thermal stability, i.e. to retain
their properties and structure unchanged for a longer time under elevated

TE
D

temperatures. Test for determining the temperature of deflection under load were
conducted using an HDT-VICAT test processor (model CEAST 6911.000,) according
to ISO 75 (ISO 75-2:2004), with a constant heating rate of 50oC h-1 and a load of
0.45 MPa according to Table 1. The samples were immersed into silicon oil and
preheated for 4 min at 40oC. 6 sample pieces were tested and the average result

EP

taken if the coefficient of variance met the required limits (5 % in accordance with
ISO2602 specifications, ISO 2602:1980).

AC
C

4. Method Procedure

The proposed formula design method consisted of three major sections: data
preprocessing, HCA and PCE, and their MATLAB (version 2012b) routines were
developed to obtain the results.
4.1. Data Preprocessing
According to the schematic design procedure shown in Fig.1, data preprocessing is
the first step of the schematic procedure. In this process, the data preprocessing
procedure is consisting of three following steps.
(1). Data check
The Kaiser-Meyer-Olkin Measure of Sampling Adequacy (KMO) and Bartlett's Test of
Sphericity (p-value) on the original experimental data were performed as data

11

ACCEPTED MANUSCRIPT
check procedure to test partial correlation and dependence. Since PCE is designed
for handling the collinear problem (Massy, 1965; Aguilera et al., 2006), the potential
of non-independence of the original data must be excluded. For any applicable data
for PCE, KMO should demonstrate a value>0.5, and Bartlett's test of sphericity
should be significant (p<0.05).
(2). Standardisation of original data

RI
PT

Standardisation is a procedure that all the initial experimental results are

transferred into a comparable range, such as zero and one, which provides a basis
whereby all data can be equally weighted (Manly, 1994). Standardisation can be

carried out using several different approaches (Montgomery, 2009). In this study,
since the target values of all tested properties are supposed to have the

(1)

M
AN
U

Eq.(1) (Manly, 1994; Montgomery, 2009):

SC

characteristic of the-larger-the-better, the original sequence is normalized using

in which xij denotes the original sequence in original data matrix

is the sequence after the data preprocessing which will form the standardised

TE
D

x%ij

matrix

, max xj is the largest value of xj, min xj is the smallest

EP

value of xj, x1j is the initial value of xj, and x0 is the desired value.

AC
C

(3). Calculation of correlation coefficient matrix

Based on the standardised matrix, the correlation coefficient matrix

or so-called Pearsons correlation coefficient matrix, is obtained by Eq.(2) which

shown as follow (Manly, 1994; Joliffe, 2002),

(2)

in which

and

denote the mean values of xj and xk respectively, and the value

12

ACCEPTED MANUSCRIPT
range of rjk is [-1, 1].
4.2. Hierarchical Clustering Analysis
As one of the most frequently used clustering analysis (CA) methods, there are two
general strategies for HCA (Manly, 1994):
Agglomerative: a "bottom up" approach, in which each observation starts in its own
cluster and pairs of clusters are merged as one moves up the hierarchy.

RI
PT

Divisive: a "top down" approach, in which all observations start in one cluster and
splits are performed recursively as one moves down the hierarchy.

The results of hierarchical clustering are usually presented graphically in a

dendrogram which is a tree-like structure diagram, and representing the groups of

SC

patterns and the similarity level at which groupings change (Manly, 1994).

In this study, a hierarchical agglomerative algorithm for clustering analysis was

selected and applied, which would clearly show the interrelationships between

M
AN
U

multiple properties in a tree-like structured plot (Manly, 1994). Euclidean distance

was used, which the distance between points (or elements) a (a1ai) and b
(a1bi) equals to the square root of sum of all dimensional differences squares, as
expressed in Eq.(3),

(3)

TE
D

And minimum or so called "single-linkage clustering" criterion was used, which

performs the clustering procedure by selecting the minimum distance between
elements of each cluster.

In the proposed formula design method, HCA is used for analysing the

EP

interrelationships between multi-objective (or properties) and then reducing the

number of total objectives for reducing the overall computational workload. For
dimension reduction policy of the proposed method, only one of the properties in

AC
C

each class will be selected as the representative.

4.3. Principal Component Estimate
In this study, PCE is firstly used to numerically analyse the effect of each component
in the composites in form of linear regression model. In PCE, the dependent
variables (also known as the responses or outcomes) of a set of explanatory
variables (also known as predictors or independent variables) are expressed in form
of a linear regression model. PCA is applied for estimating the unknown regression
coefficients, as the principal components (PCs) of the explanatory variables are
used as regressors instead of regressing the dependent variable directly (Joliffe,
2002).
In this study, Ym*1=(y1, , ym)T denotes the matrix of observed dependent variable

13

ACCEPTED MANUSCRIPT
which is the set of experimental results, Xn*m denotes the matrix of independent
variables which is the set of formula compositions (as the compositions shown in
Table 4).The typical linear regression model is shown as follow (Joliffe, 2002),

Y=c+

(4)

in which, c denotes the proportional ratio of the composition (wt.%), denotes the
unknown parametric vector of regression coefficients, and denotes the vector of
random errors (also known as the constant). The primary goal of PCE is to obtain an
for the parameter .

RI
PT

efficient estimator vector

The major PCE procedure for obtaining desirable formulae after data preprocessing
is shown as follow (Joliffe, 2002):

in which, R is correlation coefficient;

T

SC

(1). Calculation of eigenvectors and explained variations (EV)

is eigenvalue which

(5)

, j=1, 2, ..., m;

M
AN
U

Vj=(aj1, aj2, ..., ajn) is the eigenvector corresponding to the eigenvalue.

(6)

in which, ap represents the accumulative explained variations(AEV) of the first pth

eigenvalues.

TE
D

(2). Calculation of principal component estimator

Wk=XVk=[XV1, , XVk] denotes m*k matrix having the first k PCs as its columns.
The number k is selected due the related ak is close to 1, as the kth PCs have already
explained the most of variation, thus kth eigenvectors are used as corresponding

EP

PCs.

Thus principal component estimator

can be calculated as follow,

=(WTW)-1WTY

AC
C

(7)

In this study, the formulation of principal component estimator

is modified to

increase its computation stability (especially for dealing with the fluctuations caused
by those smaller values close to 0) according to Yangs article (Yang, 1989),

=(WTAW)-1WTY

(8)

in which,
within the range of (k, 1).

, is called stabilizer whose value

Due to the nature of a mixture, the sum of percentage of each composition must

14

ACCEPTED MANUSCRIPT
equals to 100 and composition percentage cannot be negative, as previously shown
in Table 3. In this study, the least combined material cost was used as the main
objective - the minimum material cost objective was applied. Based on regressed
linear expression obtained by the PCE procedure and objectives, linear
programming would be performed to obtain the suitable formula to fulfill the
requirements of each industrial application listed in Table 1 and Table 3.

RI
PT

5. Results and Discussions

5.1. Experimental Results

The experimental results obtained are shown in Table 5, and some generic patterns
are found by observing the data. Addition of talc increased tensile and flexural

properties, but also reduced impact properties, as the experimental results shown in

Table 5.
The experimental results
Trial No.

TS

SC

Fig.6 and Fig.7.

FS

IS

MFI

TDL

22.8

1512

32.8

48.5

6.8

80.7

28.5

1555

39.9

133

2.4

86.1

23.9

1623

35.3

46.1

6.9

86.8

29.1

1704

40.8

78.3

2.8

88.5

26.7

1515

35.5

45.7

7.1

89.2

20.3

2643

36.7

24.2

9.2

86.3

24.7

1639

34.7

40.2

7.6

87.2

TE
D

M
AN
U

FM

25.8

2963

37.2

18.9

9.7

96.9

27.6

2473

36.4

46.8

8.6

94.5

10

19.4

3064

36.1

23.4

8.8

92.2

AC
C

EP

Fig.6.
Plot of FS (MPa) against percentage of talc in composites

15

Fig.7.

SC

Plot of IS (KJ m-2) against percentage of talc in composites

RI
PT

ACCEPTED MANUSCRIPT

The increments observed in rigidity (tensile and flexural properties) can be

explained by the formation of rigid inter-phase between the interfaces of the

M
AN
U

polymer matrix and the talc fillers (Dez-Gutirrez et al., 1999). The other likely
cause of this pattern could be the increase of the crystallinity of the composites, as
it was increased with the addition of talc (Wang et al., 2013). It has already been
shown that in PP/talc composites, both the nucleation and the growth PP crystal are
enhanced by the addition of talc (Naiki et al., 2001). It is also reported that in
particular, PP could be trans-crystallized on the talc cleavage surface with PP
crystals oriented perpendicular to the talc surface (Naiki et al., 2001), and this

TE
D

structure would reduce impact energy absorption which lead to inferior impact
properties. Similar pattern is also observed in the experimental results of
RPP/OMMT composites (Trial. No 1, 3, 5 in Table 4), as shown in Fig.8 and Fig.9.
However, there were some observable differences mechanical performance after
talc and OMMT additions. 5 wt.% of OMMT improved flexural strength by more than

EP

30 %, while 40% of talc only improved flexural strength by only less than 10 %.
Impact strength was initially increased with the addition of talc, then decreased with
the further talc addition, while impact strength was decreased immediately after

AC
C

introduction of OMMT.

16

ACCEPTED MANUSCRIPT
Fig.8.

SC

RI
PT

Plot of FS (MPa) against percentage of OMMT in RPP/OMMT composites

Fig.9.

M
AN
U

Plot of IS (KJ m-2) against percentage of OMMT in RPP/OMMT composites

The experimental results shows addition of talc also increased MFI, as shown in
Fig.10, which could contribute to the increase in inter-phase delaminating and

AC
C

EP

TE
D

dispersion of talc particles or agglomerates (Wang et al., 2013).

Fig.10.

Plot of MFI (10 g/10min under 230oC 2.16kg) against percentage of talc in
composites

However, the detailed effect of each composition and the co-effects between them
cannot be readily found due to intertwining formulation, for there are up to five
components will affect the properties of end product. Based upon calculations on the
original experimental data matrix, the KMO result is 0.53>0.5 and Bartletts
p=0<0.05, which meant this data set had some internal dependences and overlaps
(Joliffe, 2002). Thus, the proposed PCE based formula designing procedure was
validly applied.

17

ACCEPTED MANUSCRIPT
5.2. Requirements Reduction
Based on experimental data in Table 5, standardization is performed to transfer raw
data into comparable data array (the standard data matrix) using Eq.(1), as shown
in Table 6. And the correlation coefficient matrix was calculated via Eq.(2), as shown
in Table 7. Then, HCA was applied in accordance with Section 4.2. The HCA results

Table 6.
The standardised experimental results

RI
PT

are presented in Fig.11.

Trial No.

TS

FM

FS

IS

MFI

0.35

0.26

0.60

0.94

0.03

0.88

0.46

0.07

0.31

0.24

0.62

0.12

0.52

0.05

0.48

0.75

0.01

0.34

0.23

0.64

0.52

0.09

0.73

0.49

0.05

0.93

0.35

0.55

0.08

0.24

0.19

0.71

0.40

0.66

0.93

0.55

0.85

0.62

0.45

0.24

0.85

0.85

10

0.41

0.04

0.88

0.71

0.33

SC

0.38

M
AN
U

Table 7.

TDL

The correlation coefficient matrix of the standardised experimental results

TE
D

0.56

0.07

0.65

-0.59

0.59

-0.62

0.68

-0.61

-0.92

0.18

0.70

0.28

-0.35

0.44

AC
C

EP

-0.47

18

Fig.11.

M
AN
U

The clustered results of tested properties

SC

RI
PT

ACCEPTED MANUSCRIPT

In Fig.11, the numbers in vertical axis represented the distances between two
tested properties, for a smaller figure indicates a closer relationship between the
two corresponding properties. Element 1 to 6 in horizontal axis are representing the
six tested properties:
No.1 represented TS;
No.3 represented FS
No.4 represented IS;

TE
D

No.2 represented FM;

No.5 represented MFI;

EP

No.6 represented TDL.

Fig.11 shows the six tested properties can be classified into two general categories:
(1). Flexural Modulus, Melt Flow Index and Temperature of Deflection under Load

AC
C

In which, FM and MFI are more closely related. The distances between any two
members of all these three properties are less than 0.4, thus they all can be
considered as a class according to Manly (1994) and Zafiriou et al (2012), and
flexural modulus was selected as the representative.
(2). Tensile Strength, Flexural Strength and Impact Strength
In which, TS and IS are more closely related, for their distance between TS and IS
is less than 0.4. Thus, they can be considered as a class according to Section 4.2,
and IS was selected as the representative.
According to Fig.11 and analysis above, properties whose distance less than 0.4 are
considered as one class, and the technical requirements in Table 1 were reduced: in
1st category, FM was selected; in 2nd category, FS and IS were selected. In total,
three properties were selected as the representatives, and the computation

19

ACCEPTED MANUSCRIPT
workload was greatly reduced. The reduced technical requirements based on Table
1 are shown in Table 8. Due to the nature of mixture, the constraints shown in Table
3 were kept as they were.
Table 8.
Applications

FM

FS

IS

Meter Holder

2400

35

40

Engine Cover

2200

35

40

Fender

2000

30

60

5.3. Linear Regression Model

RI
PT

The reduced requirements based on Table 1

SC

After the properties clustering, the technical requirements of the selected

automobile parts are greatly reduced, and the PCE procedure which described in
Section 4.3 was applied.

M
AN
U

4 PCs were selected in this study, since a4 obtained by Eq.(5) and Eq.(16) was 0.98
which is close to 1. It meant that the first four PCs have already explained most of
the variation. Then, the linear regression model of the three selected tested
properties (shown in Table 8) was obtained by Eq.(8) is shown in Table 9. The
calculation of coefficients in the model was based on the data preprocessed
matrixes of combination of both the initial compositions (Table 4) and the

Table 9.

TE
D

experimental results (Table 5).

The linear regression model obtained by PCE

Constant

VPP(c1)

RPP(c2)

Talc(c3)

OMMT(c4)

MAPP(c5)

FM

1736

-3.73

-4.68

46.07

98.56

-14.35

37.53

0.03

-0.03

0.03

0.22

-0.08

71.70

0.44

-0.21

-1.02

-2.70

0.12

FS
IS

EP

Tested Properties

AC
C

In Table 9, the effect of each composition is numerically identified via the

determination of parametric values in linear regression model. The effects of
particle fillers (talc and OMMT) addition are verified and confirmed, as the
coefficients of those particle fillers in flexural properties are positive while the
coefficient for impact strength is negative.
In this study, a least squares regression (LSR) method (Wang et al., 1990) and a
partial least square regression (PLSR) (Khajehsharifi et al., 2014) were applied as
comparatives to the PCE linear regression model. And the respective linear
regression models obtained by these two methods are shown in Table 10.
Table 10.
The linear regression models obtained by LSR and PLSR
20

ACCEPTED MANUSCRIPT
Tested Properties

Constant

VPP(c1)

RPP(c2)

Talc(c3)

OMMT(c4)

MAPP(c5)

FM

1279

0.65

0.03

50.35

102.07

-9.67

FS

34.89

0.05

-0.01

0.05

0.24

-0.10

IS

51.57

0.64

-0.06

-0.82

-2.54

0.32

FM

1041

-4.03

6.08

32.05

51.17

28.74

FS

36.22

0.01

-0.03

0.11

IS

76.69

0.26

-0.17

-1.234

-0.29

-0.73

LSR

PLSR

RI
PT

Methods

To evaluate the correlation accuracy of these three models (Table 9 and Table 10),
Eq.(9) was used.

Pr d Exp
100%
Exp

(9)

SC

Dif =

in which, Exp denotes the experimental result of a property and Prd denotes the
predicted value of the property by using Eq.(4) based on the linear regression

M
AN
U

models (Table 9 and Table 10) and the initial compositions (Table 4), Dif denotes the
difference between the experimental result (Exp) and the predicted value (Prd) is
expressed in proportional percentage. The smaller the Dif value, the more accurate
the model is. The obtained values of Difs of the three linear regression models are
shown in Table 11.

Table 11. The Difs of the reduced requirements using the three linear regression
models obtained by PCE, LSR and PLSR respectively
FM

FS

IS

FM

TE
D

Trial No.

PCE

9.31

4.65

3
4

IS

FM

LSR

FS

IS

PLSR

6.04

0.60

8.66

7.10

1.09

12.04

10.76

22.82

0.33

19.65

5.83

2.08

19.81

35.01

6.42

29.57

10.34

1.05

0.61

9.81

0.37

2.23

9.95

3.23

26.53

2.93

3.24

11.17

5.42

10.95

18.23

8.16

6.76

8.19

1.32

5.47

8.53

2.56

4.44

46.16

3.58

10.92

5.58

1.01

9.09

5.54

0.84

11.45

8.13

3.76

14.75

2.43

0.20

14.18

2.86

0.89

15.85

30.28

3.52

8.08

9.83

0.59

22.75

9.62

1.51

24.71

7.17

5.73

12.8

5
6

AC
C

EP

FS

2.55

0.36

19.76

2.71

1.54

19.29

1.21

0.05

16.6

10

6.05

3.82

60.68

6.41

0.66

60.56

9.44

0.33

27.74

Average

6.19

1.80

16.40

6.20

2.30

17.04

17.76

4.55

17.66

As shown in Table 11, the average Difs of the PCE linear regression model are
smaller than those of the other two models, which indicated the PCE linear
regression model has higher correlation accuracy. Compared to the existing formula
design methods (Soury et al., 2009; Homkhiew et al., 2014), this method is
particularly efficient for it obtains accurate linear regression model based on a
limited trial data instead of detailed experimental data on each composition.

21

ACCEPTED MANUSCRIPT
5.4. Optimal Formulation
As previously stated, the minimum material cost objective was applied in this study.
The prices of the materials used are shown in Table 12 which being expressed in
yuan per kilogram (average prices for small quantity purchase during the period of
Feb 2014 to Feb 2015, and they will be much cheaper in large order).
Table 12.

RI
PT

The average prices of the materials used in this study during the period of Feb 2014
to Feb 2015 (information obtained in Feb 2015 from material suppliers, Yangzi
Petrochemical Co., Ltd., Yoshida Chemical co.ltd., Deba Chemical co.ltd., et al)
VPP(c1)

RPP(c2)

Talc(c3)

OMMT(c4)

MAPP(c5)

12

18

20

SC

Price (in RMB per kg)

It is known that the total composition percentage is equal to 100 and no negative
value should be found in composition as shown in Table 3. The PCE linear regression
model shown in Table 9 was coupled with the minimum material cost objective in

M
AN
U

formulation calculation. The suitable compounding formulae which satisfy the

multiple objectives were obtained via using linear programming, and are presented
in Table 13.
Table 13.
The obtained formulae (wt.%)
VPP(c1)

RPP(c2)

Talc(c3)

OMMT(c4)

MAPP(c5)

Meter Holder

17

60

18

Engine Cover

17

60

20

Fender

38

39

20

TE
D

Application

In previous literatures which concerning composite design (Soury et al., 2009;

EP

Homkhiew et al., 2014), the use of fillers was extensive since mechanical
performance was the sole target. This study aims at achieving the minimum cost
objective along with multiple performance requirements, the use of cheap

AC
C

components is maximised, such as RPP and talc, as shown in Table 13. The
percentages of RPP in the formulae for meter holder and engine cover have reached
the upper limit, and the percentages of talc in the formulae for engine cover and
fender have also reached the upper limit. Consequently, the use of expensive
materials is minimised, such as OMMT and MAPP, as shown in Table 13. The
percentages of MAPP in the formulae for all the three applications have only up to
the lower limit.
Further, a verification test set was performed to compare the performance of
obtained optimal formulae with predicted values, in order to verify the effectiveness
of the proposed formula designing method.

22

ACCEPTED MANUSCRIPT
5.5. Verification
With the purpose of investigating the effectiveness of the formula design method
proposed, a verification testing set was carried out. The compositions of materials
were according to formulae in Table 12, and those composites were extruded, dried,
injection moulded, conditioned and tested under constant processing parameters as
same as the previous tests. The results of the verification test are summarized in
Table 14, the Exp values compared to the Prd values obtained by using linear

RI
PT

regression model in Table 9. In Table 14, the Difs are also calculated and presented
by using Eq.(9).
Table 14.

The performance of the obtained optimal formulae compared to the predicted

Cover
Fender

Exp

Prd

Dif

Required

Exp

Prd

Dif

Required

Exp

Prd

Dif

Required

2465

2375

3.65

2400

35.89

36.98

3.04

35

40.13

43.18

7.60

40

2214

2270

2.53

2200

36.12

36.60

1.3

35

44.82

46.54

3.84

40

2121

2290

47.97

2000

34.57

37.86

9.52

30

61.27

60.19

1.76

60

M
AN
U

Engine

IS

According to Table 14, all the Difs are within 10 % and satisfied the reduced
requirements in Table 8, which indicated that the feasibility of this proposed method
in designing composites based recycled plastic for potential automobile applications.

TE
D

Holder

FS

The effectiveness of the proposed method is verified numerically.

For the other three requirements - TS, MFI and TDL, the results are also obtained
and presented against to the full technical requirements (Table 1) in Table 15. In
Table 15, the experimental results have shown that the obtained optimal formulae

EP

Meter

FM

are fully satisfying the technical requirements in Table 1. Thus, it is proved that the
specific composition selection rules and the reduction of the requirements
performed by HCA are both effective and efficient, for they provide solid analysis

AC
C

Application

SC

values based on PCE linear regression model in Table 9

without dropping of any important information.

Table 15.

The experimental results of the other three properties of the obtained optimal
formulae compared to the requirements in Table 1
Application

TS

MFI

TDL

Exp

Required

Exp

Required

Exp

Required

Meter Holder

24.5

20

8.1

88.5

85

Engine Cover

23.8

20

8.4

92.7

92

Fender

25.3

22

7.2

90.9

90

23

ACCEPTED MANUSCRIPT
6. Conclusions
Using recycled plastic in manufacturing not only brings economic advantages, but
also reduces environmental impact. However, there is a lack of methods to make
recycled plastic meet the multiple requirements of high-end industrial applications.
Most of current researches only focus on improving one or few properties of recycled
plastic, and also require detailed tests on each possible formulation. This paper
presents a novel method that combines HCA and PCE for designing recycled plastic

RI
PT

based composites to meet the multi-objectives of manufacturing automobile parts.

It is a multi-objective optimisation method for composite design in real
manufacturing industry for promoting sustainability.

In this proposed formula design method, HCA is used for reducing the number of the
technical requirements of automobile parts by analysing the inter-relationships of

SC

multiple properties, and PCE is used for obtaining linear regression model of
composition based on selected PCs. With the chosen objective (in this study,
minimum material cost of the composites is selected as primary objective) and the

M
AN
U

reduced technical requirements, the formulae were obtained by linear programming

based on the PCE linear regression model. Real technical requirements of
automobile parts were used in this study. The composition selection rules were
proposed to reduce the number of initial composition trials. The PCE linear
regression model was compared to the regression models obtained via other means,
and was numerically proved to be the most suitable model. At last, the effectiveness
and the efficiency of the proposed method was verified via verification testing set,

TE
D

as the experimental results were closely matched the predictions. This method is
novel and original, which designs suitable formula for adapting recycled plastic into
high-value industrial applications without conducting detailed, repeated tests on
each possible formulation as other literatures did (Stark and Matuana, 2003; Soury
et al., 2009; Zolfaghari et al., 2013; Homkhiew et al., 2014). Thus, this

EP

multi-objective optimisationmethod approach can facilitate the recycling of waste

plastic scrap and reduces total environmental impact by expanding the use of
recycled plastic in high-value manufacturing industry. By conducting experiments

AC
C

on limited trials, the required properties of plastic composites can be modelled,

predicted and optimised accordingly.
During the initial application of this multi-objective optimal formula design method,
the major findings were listed as follows:
(1). The novel proposed composition selection rules reduce the number of initial
compositions effectively and efficiently, as indicated by the experimental results of
verification testing set;
(2). The multiple technical requirements of industrial applications can be effectively
reduced by identifying and classifying their interrelationships via utilisation of HCA;
(3). The linear regression model of components' effects obtained by PCE is more
precise than the models obtained by LSR and PLSR, as the model predictions are
more closer to the experimental results;

24

ACCEPTED MANUSCRIPT
(4). The optimal formulae acquired via the proposed method were validly proved in
verification testing set, as all properties met the technical requirements while
minimum material cost was achieved.
Further, this work can be extended in several ways to expand its application:
(1). More demanding industrial applications will be taken into consideration, which
will include requirements of manufacturing details and structural design;

RI
PT

(2). Application of this method will be extended to the whole life-cycle of recycled
plastic, to model up the performance or behaviour of recycled plastic and related
composites in different stages of the life cycle of parts.
Acknowledgements

This work is financially supported by the Innovation Team of Ningbo Science and

SC

Technology Bureau (2011B81006), International Technological Cooperation Project

of the Ministry of Science and Technology (2012DFG91920) and Industrial
(2014A35001-2).
Abbreviations
Accumulative Explained Variations
Charpy Impact Strength
Cluster Analysis
Experimental Result
Flexural Modulus
Flexural Strength

TE
D

Explained Variation

M
AN
U

Technology Innovation and Industrialization of Science and Technology Project

AEV
IS
CA
Exp
EV
FM
FS
HCA

Kaiser-Meyer-Olkin Measure of Sampling Adequacy

KMO

Maleic Anhydride Grafted Polypropylene

MAPP

Melt Flow Index

MFI

Organically Modified Montmorillonite

OMMT

Polychlorinated Biphenyls

PCB

AC
C

EP

Hierarchical Clustering Analysis

Persistent Organic Pollutants

POP

Polypropylene

PP

Points

pts

Predicted Value

Prd

Principal Component

PC

Principal Component Analysis

PCA

Principal Component Estimate

PCE

Recycled Polypropylene

RPP

Tensile Strength

TS

Talcum Powder

Talc

Temperatures of Deflection under Load

TDL

Virgin Polypropylene

VPP

25

ACCEPTED MANUSCRIPT
References
Aguilera, M.A., Escabias, M., and Valderrama, J.M. (2006). Using principal
components for estimating logistic regression with high-dimensional
multicollinear data. Comput. Stat. Data. An. 50:1905-1924.
Al-Salem, S.M., Lettieri, P., and Baeyens, J. (2009). Recycling and recovery routes of
plastic solid waste (PSW): A review. Waste Manage. 29 2625-2643.

RI
PT

Bahlouli, N., Pessey, D., Ahzi, S. and Rmond, Y. (2006). Mechanical behavior of
composite based polypropylene: Recycling and strain rate effects. J. Phy. IV.
France. 134:1319-1323.

Bertin, S., and Robin, J.J. (2--2). Study and characterization of virgin and recycled
LDPE/PP blends. Eur. Polym. J. 38:2255-2264.

Caneghem, J.V., Block, C., Brecht, A.V., Wauters, G., and Vandecasteele, C. (2010).

SC

Mass balance for POPs in hazardous and municipal solid waste incinerators.
Chemosphere. 78:701-708.

Chen, X., Samson, E., Tocqueville, A., and Aubin, J. (2015). Environmental

M
AN
U

assessment of trout farming in France by life cycle assessment: using

bootstrapped principal component analysis to better define system classification.
J. Clean. Prod. 87: 87-95.

Dez-Gutirrez, S., Rodrguez-Prez, M.A., De Saja, J.A., and Velasco, J.I. (1999).
Dynamic mechanical analysis of injection-moulded discs of polypropylene and
untreated and silane-treated talc-filled polypropylene composites. Poly.
40:5345-5353.

TE
D

Dormer, A., Finn, D.P., Ward, P., and Cullen, J. (2013). Carbon footprint analysis in
plastics manufacturing. J. Cleaner. Prod. 51:133-141.
Ehrig, R.J. (1992). Plastics Recycling: Products and Process. New York: Hanser.
Font, R. Molt, J. Egea, S. and Conesa, J.A. (2011). Thermogravimetric kinetic
analysis and pollutant evolution during the pyrolysis and combustion of mobile

EP

phone case. Chemosphere. 85:516-524.

Forteln, I., Kruli, Z. and Michlkov, D. (2002). Use of additive compatibilisation
for recycling of municipal plastics iste. Polimery. 47:534-537.

AC
C

Grosso, M. (2009). Experimental evaluation of dioxins release and mass balance of

a WTE plant. ISWA APESB World Congress 2009, Lisbon, Portugal. 238:1-8.
Gu, F., Hall, P., Miles, N.J., Ding, Q., and Wu, T. (2014). Improvement of mechanical
properties of recycled plastic blends via optimising processing parameters using
the Taguchi method and principal component analysis. Mater. Des. 62:189-198.
Gu, F., Hall, P., and Miles, N.J. (2016). Performance evaluation for composites based
on recycled polypropylene using principal component analysis and cluster analysis.
J. Clean. Prod. 115:343-353.
Homkhiew, C., Ratanawilai, T., and Thongruang, W. (2014). The optimal formulation
of recycled polypropylene/rubberwood flour composites from experiments with
mixture design. Compos. Part. B. Eng. 56:350-357.
Huang, D., Zhou, S., Hong, W., Feng, W., and Tao, L. (2013). Pollution characteristics
of volatile organic compounds, polycyclic aromatic hydrocarbons and phthalate

26

ACCEPTED MANUSCRIPT
esters emitted from plastic wastes recycling granulation plants in Xingtan Town,
South China. Atmos. Environ. 71:327-334.
ISO 75-2:2004. Plastics - Determination of temperature of deflection under load Part 2: Plastics and ebonite published by the International Organization for
Standardisation (ISO), 2004.
ISO 178:2010. Plastics - Determination of flexural properties, published by the
International Organization for Standardisation (ISO), 2010.

RI
PT

ISO 179-2:2010. Plastics - Determination of Charpy impact properties - Part 2:

Instrumented impact test, published by the International Organization for
Standardisation (ISO), 2010.

ISO 291:2008. Plastics -- Standard atmospheres for conditioning and testing,

published by the International Organization for Standardisation (ISO), 2008.

ISO 527-2:2012. Plastics - Determination of tensile properties - Part 2: Test
Organization for Standardisation (ISO), 2012.

SC

conditions for moulding and extrusion plastics, published by the International

ISO 1133-1:2011. Plastics - Determination of the melt mass-flow rate (MFR) and

M
AN
U

the melt volume-flow rate (MVR) of thermoplastics, published by the International

Organization for Standardisation (ISO), 2011.

ISO 2602:1980. Statistical interpretation of test results - Estimation of the mean -Confidence interval, published by the International Organization for
Standardisation (ISO), 1980.

Karanjekar, R.V., Bhatt, A., Altouqui, S., Jangikhatoonabad, N., Durai, V., Sattler,
M.L., Hossain, M.D.S., and Chen, V. (2015). Estimating methane emissions from

TE
D

landfills based on rainfall, ambient temperature, and waste composition: The

CLEEN model. Waste Manage. 46:389-398.
Khan, Z.A., Kamaruddin, S., and Siddiquee, A.N. (2010). Feasibility study of use of
recycled high density polyethylene and multi response optimization of injection
moulding parameters using combined grey relational and principal component

EP

analyses. Mater. Des. 31(6):2925-2931.

Khajehsharifi, H., Pourbasheer, E., Tavallali, H., Sarvi, S., and Sadeghi, M. (2014).
The comparison of partial least squares and principal component regression in

AC
C

simultaneous spectrophotometric determination of ascorbic acid, dopamine and

uric acid in real samples. Arab. J. Chem. In Press, Corrected Proof.
Lindahla, P., Robrta, K.H., Nya, H. and Bromana, G. (2014). Strategic sustainability
considerations in materials management. J. Clean. Prod. 64:98-103.
Luzuriaga, S.E., Kovov, J., and Forteln, I. (2011). Stability of model recycled
mixed plastic waste compatibilised with a cooperative compatibilisation system.
Polym. Degrad. Stabil. 96(5):751-755.
Joliffe, I.T. (2002). Principal component analysis (2nd ed.). New York:Springer.
Makuta, M., Moriguchi, Y., Yasuda, Y. and Sueno, S. (2000). Evaluation of the effect
of automotive bumper recycling by life-cycle inventory analysis. J. Mater. Cycles.
Waste. 2:125-137.
Manly, B.F.J. (1994). Multivariate Statistical Methods - A Primer(2nd ed.). London:
Chapman & Hall.

27

ACCEPTED MANUSCRIPT
Massy, W.F. (1965). Principal Components Regression in Exploratory Statistical
Research. J. Am. Stat. Assoc. 60:234-256.
Melnyk, A., Dettlaff, A., Kukliska, K., Namienik, J., and Wolska, L. (2015).
Concentration and sources of polycyclic aromatic hydrocarbons (PAHs) and
polychlorinated biphenyls (PCBs) in surface soil near a municipal solid waste
(MSW) landfill. Sci. Total Environ. 530-531:18-27.
Menada, N., Bjrkmana, B., and Allain, E.G. (1998). Combustion of plastics

RI
PT

contained in electric and electronic scrap. Resour. Conserv. Recy. 24:65-85.

Meng, T., Klepacka, A.M., Florkowski, W.J., and Braman, K. (2015). What drives an
environmental horticultural firm to start recycling plastics? Results of a Georgia
survey. Resour. Conserv. Recy. 102:1-8.

Meran, C., Ozturk, O., and Yuksel, M. (2008). Examination of the possibility of
recycling and utilising recycled polyethylene and polypropylene. Mater. Des.

SC

29:701-705.

Montgomery, D.C. (2009). Design and analysis of experiments (7th ed.). New York:
John Wiley & Sons.

M
AN
U

Naiki, M., Fukui, Y., Matsumura, T., Nomura, T., and Matsuda, M. (2001). The effect
of talc on the crystallization of isotactic polypropylene. J. Appl. Polym. Sci.
79:16931703.

Park, Y.S., Egilmez, G., and Kucukvar, E. (2015). A Novel Life Cycle-based Principal
Component Analysis Framework for Eco-efficiency Analysis: Case of the United
States Manufacturing and Transportation Nexus. J. Clean. Prod. 92:327-342.
PlasticsEurope. (2014). Plastics - the Facts 2014, an analysis of European plastics

TE
D

production, demand and waste data. Available

from:http://www.plastval.pt/conteudos/File/Publicacoes/Plastics%20-%20the%
20Facts%202014.pdf [retrieved on March 3, 2015].
Putra, H.D., Ngothai, Y., Ozbakkaloglu, T., and Seracino, R. (2009). Mineral filler
reinforcement for commingled recycled-plastic materials. J. Appl. Polym. Sci.,

EP

112:3470-3481.

Ross, P.J. (1996). Taguchi techniques for quality engineering. New York:
McGraw-Hill.

AC
C

Saleh, M.A., Sabah A.A.W., and Charles S.B. (2008). Combining principal
component regression and artificial neural networks for more accurate predictions
of ground-level ozone. Environ. Modell. Softw. 23:396-403.
Sautot, L., Faivre, B., Journaux, L., and Molin, P. (2015). The hierarchical
agglomerative clustering with Gower index: A method for automatic design of
OLAP cube in ecological data processing context. Ecol. Inform. 26:217-230.
Schumann, S., Nolte, L.P., and Zheng, G. (2013). Comparison of partial least
squares regression and principal component regression for pelvic shape prediction.
J. Biomech. 46:197-199.
Sommerhuber, P.F., Wang, T., and Krause, A. (2016). Wood-plastic composites as
potential applications of recycled plastics of electronic waste and recycled
particleboard. J. Clean. Prod. 121:176-185.
Soury, E., Behravesh, A.H., Esfahani, E.R., and Zolfaghari, A. (2009). Design,

28

ACCEPTED MANUSCRIPT
optimization and manufacturing of wood-plastic composite pallet. Mater. Des.
30(10):4183-4191.
Stark, N.M., and Matuana, L.M. (2003). Ultraviolet weathering of photostabilized
wood-flour-filled high-density polyethylene composites. J. Appl. Polym. Sci.
90(10):2609-2617.
Tsai, C.J., Chen, M.L., Chang, K.F, Chang, F.K., and Mao, I.F. (2009). The pollution
characteristics of odor, volatile organochlorinated compounds and polycyclic
74:1104-1110.

RI
PT

aromatic hydrocarbons emitted from plastic waste recycling plants. Chemosphere.

Tianqiang. (2014). Personal communication with workshop manager of Shanghai
Tianqiang Technical Co., Ltd. on 21/10/2014.

Wang, K., Bahlouli, N., Addiego, F., Ahzi, S., Rmond, Y., Ruch, D., and Muller, R.
(2013). Effect of talc content on the degradation of re-extruded

SC

polypropylene/talc composites. Poly. Degrad. Stab. 98:1275-1286.

Wang, S.G., Tse, S.K., and Chow, S.C. (1990). On the measures of multicollinearity
in least squares regression. Stat. Probabil. Lett. 9: 347-355.

M
AN
U

WRAP. (2010). China Market Sentiment Survey Report 2010,

http://www.wrap.org.uk/sites/files/wrap/China%20Market%20Sentiment%20S
urvey%20Report%202010.pdf [retrieved on March 3, 2015].
Yang, H. (1989). Single parameter principal component regression (in Chinese).
Appl.Math.Univ. A. 1:74-80.

Yang, Y. and Gao, F. (2006). Injection molding product weight: Online prediction and
control based on a nonlinear principal component regression model. Polym. Eng.

TE
D

Sci. 46:540-548.

Zafiriou, P., Mamolos, A.P., Menexes, G.C., Siomos, A.S., Tsatsarelis, C.A., Kalburtji,
K.L. (2012). Analysis of energy flow and greenhouse gas emissions in organic,
integrated and conventional cultivation of white asparagus by PCA and HCA: cases
in Greece. J. Clean. Prod. 29-30:20-27.

EP

Zhang, S.D., Ban, Q., Huang, H.X., and Li, Y.Z. (2009). Modified recycled glass fibre
reinforced flame retardant PBT with improved mechanical properties by
nanoparticles. The 5th ISFR, Chengdu, China, 171-177.Zolfaghari, A., Behravesh,

AC
C

A.H., and Adli, A. (2013). Continuous glass fiber reinforced wood plastic
composite in extrusion process: Mechanical properties. Mater. Des. 51:701-708.

29

ACCEPTED MANUSCRIPT
Highlights
(1). A new tool for formula design is proposed to adapt recycled plastic for
manufacturing;
(2). Multiple technical requirements and minimum material cost are achievable;
(3). Hierarchical cluster analysis is used to reduce technical requirements of auto
parts;
(4). Principal component estimate is used to identify effects of components in

RI
PT

composites;

(5). Verification tests confirmed effectiveness and efficiency of the proposed

AC
C

EP

TE
D

M
AN
U

SC

method.

Tel: +86 574 8818 0018

Email:Philip.Hall@nottingham.edu.cn