An innovative recycling process to obtain pure polyethylene [619603]

An innovative recycling process to obtain pure polyethylene
and polypropylene from household waste
Silvia Serrantia,⇑, Valentina Luciania, Giuseppe Bonifazia, Bin Hub, Peter C. Remb
aDICMA, Department of Chemical Engineering, Materials and Environment, Sapienza University of Rome, Via Eudossiana 18, 00184 Rome, Italy
bFaculty of Civil Engineering and Geosciences, Delft University of Technology, 2628 CN Delft, The Netherlands
article info
Article history:
Received 9 August 2014
Accepted 15 October 2014Available online xxxx
Keywords:Plastic recyclingHousehold wastePolyolefinMagnetic density separationHyperspectral imagingabstract
An innovative recycling process, based on magnetic density separation (MDS) and hyperspectral imaging
(HSI), to obtain high quality polypropylene and polyethylene as secondary raw materials, is presented.More in details, MDS was applied to two different polyolefin mixtures coming from household waste.
The quality of the two separated PP and PE streams, in terms of purity, was evaluated by a classification
procedure based on HSI working in the near infrared range (1000–1700 nm). The classification model wasbuilt using known PE and PP samples as training set. The results obtained by HSI were compared withthose obtained by classical density analysis carried in laboratory on the same polymers. The results
obtained by MDS and the quality assessment of the plastic products by HSI showed that the combined
action of these two technologies is a valid solution that can be implemented at industrial level.
/C2112014 Elsevier Ltd. All rights reserved.
1. Introduction
The constant growth of plastic consumption, in various applica-
tions, encouraged the research of new recycling procedures and
the improvement of the old ones.
Europe is the second plastic producer, after China, accounting
for the 20.4% of the world total production. The global production
reached 288 million tonnes in 2012, recording a 2.8% increase com-
pared with 2011 and, in general, in the last 50 years the plastic
market has grown constantly and, as a consequence, the amount
of polymer wastes increased as well. Every year the amount of
plastics delivered to landfill is decreasing with a positive trend in
the recovery of post-consumer plastics. In 2012, 61.9% of the plas-
tics were recovered, 35.6% were used for energy recovery and
26.3% were recycled. However, the recycling rate of plastic is still
at low levels ( Plastic the facts, 2013 ).
Packaging dominates the waste generated from plastics, cover-
ing 62.2% of the total. Polyolefins (POs) account for more than 50%
of the packaging production. Plastic recovery starts with the sepa-
rate collection of post-consumer waste ( Al-Salem et al., 2009 ). In
the recycling plants, the different types of plastics are sorted
mainly using two different technologies: the first one is based on
Optical Sensing Techniques (OST), the second one is based on sink-
float processes. The OST recognizes different types of polymerscomparing their spectra in the near infrared (NIR) range with a
library of reference spectra. The main limitations are linked to:
(i) separation efficiency, depending on the size of plastic flakes
(good detection is usually achieved for particles with size >5 cm),
(ii) sensing unit ‘‘blindness’’, depending on flakes color (dark plas-
tics, for their low reflectance, are not classified) and, (iii) flakes sur-
face status (the presence of labels, dirtyness or paint and coating
on the plastic surface does not allow a correct recognition) ( Di
Maio et al., 2010 ). For these reasons about 50% of the input mate-
rial is not correctly separated and ends up in the residues, which
are usually send to incineration.
Sink-float processes separate different materials taking advan-
tage on their difference in densities. These techniques use a process
medium with an intermediate density between those of the poly-
mers to be separated. Usually sink-float separation is quite simple
but it becomes difficult if the materials are characterized by a
slight different density. This method is often combined with flota-tion ( Burat et al., 2009; Pongstabodee et al., 2008; Dodbiba et al.,
2010 ).
Sink-float separation is commonly used to separate the light PO
from the plastics heavier than water (i.e. PVC and PET). The result-
ing PO mixed product contains both polyethylene (PE) and poly-
propylene (PP) and can be used only for low quality recycled
materials. To achieve the same physical and mechanical properties
of virgin materials, a further separation of PE from PP is needed and
their grade should be better than 97% ( Bakker et al., 2009 ).
In this paper, the application of two innovative technologies for
PO sorting and final pure products (PE and PP) quality control, i.e.:
http://dx.doi.org/10.1016/j.wasman.2014.10.017
0956-053X/ /C2112014 Elsevier Ltd. All rights reserved.⇑Corresponding author. Tel.: +39 0644585360; fax: +39 0644585618.
E-mail address: silvia.serranti@uniroma1.it (S. Serranti).Waste Management xxx (2014) xxx–xxx
Contents lists available at ScienceDirect
Waste Management
journal homepage: www.else vier.com/loc ate/wasman
Please cite this article in press as: Serranti, S., et al. An innovative recycling process to obtain pure polyethylene and polypropylene from househo ld waste.
Waste Management (2014), http://dx.doi.org/10.1016/j.wasman.2014.10.017

Magnetic Density Separation (MDS) and HyperSpectral Imaging
(HSI), is presented.
MDS is a physical separation method based on the differences in
density of the materials ( Murariu et al., 2005 ). It has been already
tested for the recovery of different materials: precious metals from
Municipal Solid Waste Incinerator (MSWI), bottom ash ( Bakker
et al., 2007 ),Waste from Electrical and Electronic Equipment (WEEE)
(Hu et al., 2011 ) and PolyEthylene Terephthalate (PET) ( Bakker and
Rem, 2006 ).
HSI analysis was introduced in the recycling process to monitor
the quality of the two separated product streams (PP and PE). This
fast and non-destructive technique is able to collect both spectral
and spatial information from an object and represents an attractive
solution for quality control in several industrial applications. In
fact, in the last ten years the use of this technique has rapidlygrown in different fields as in food and pharmaceutical sectors.
Several studies have been carried out also in waste recycling sec-
tor, i.e. glass recycling ( Bonifazi and Serranti, 2006 ), compost prod-
uct quality control ( Dall’Ara et al., 2012 ), characterization of end-of
life mobile phones ( Palmieri et al., 2014 ) and characterization of
different plastics ( Serranti et al., 2011, 2012; Ulrici et al., 2013;
Luciani et al., 2013 ). Both MDS and HSI were applied to the same
PO coming from Romanian and Dutch household waste (HW) to
assess and validate the performances of such technologies for PO
waste sorting and quality control.
2. Materials and methods
2.1. Polyolefin samples from household waste
PO samples were collected, as already mentioned, in two coun-
tries: Romania and the Netherlands. The Romanian sample (RO
HW) came from Valcea, a town with a population of approximately
80,000 inhabitants. A 27 kg sample, including both polyolefin and
non-polyolefin polymers, as well as a small amount of other
wastes, e.g. food garbage, was hand-picked from raw household
wastes. Foils and polymer objects smaller than 5 cm were not
selected, due to the high cost of hand-sorting from raw HW. Differ-
ent from the Romanian sample, the Dutch sample was separatelycollected from other HW according to the Dutch PlasticHero pro-
gram launched by Nedvang in 2008. In this program, the citizens
are encouraged to bring their plastic wastes into a separate trash
bin near their houses. The desired wastes include plastic bags, food
containers, lids of jars, bottles, etc., and exclude, for instance, fast
food packaging, meat packing materials or toys. The gathered poly-
mer wastes from the trash bin are transported to a sorting plant for
polymer recycling normally once a week. The investigated Dutch
sample (NL HW) came from Zeeland, a province in the south of
the Netherlands. Compared to the Romanian plastic waste, the
Dutch sample contains not only rigid plastics, but also plastic foils.
For this study, only the blown and injected PP and PE (including
both LDPE and HDPE) were selected, so the foils were removed. All
the samples, in order to be processed by MDS, were shredded
below 8 mm.
2.2. Magnetic density separation
2.2.1. MDS principle
MDS is a density-based sorting technology, similar to the con-
ventional sink-float method; but instead of using a medium with
a single cut density, it uses a liquid separation medium with a den-
sity gradient. Such liquid contains magnetic iron oxide particles
with a size of about 10–20 nm suspended in water. By applying
an artificial gravity, in the form of magnetic force, varying expo-
nentially in the vertical direction, the effective density of the liquidvaries in this direction as well. Plastic particles with the same den-
sity will float in the liquid at the same level: where the effective
density is equal to their own density.
When magnetic liquid is placed in a magnetic field, the weight
of the liquid becomes the sum of gravity and the vertical compo-
nent ( z) of the magnetic force. In such way, the separation medium
can be artificially lighter or heavier than would be expected on the
basis of its material density ( ql). In a gradient magnetic field ( B),
the total weight ( F) of a volume of magnetic liquid ( Vl) with mag-
netization Mis:
F¼q1gVlțMV lrzjBj¼q1țMrzjBj
g/C18/C19
gVl ð1Ț
When particles made of a non-magnetic material of material den-
sityqpare introduced into the liquid, their weight will be equal
to their gravity minus the weight of the same volume of liquid
(Archimedes’ Law):
F¼qpgVl/C0q1gVlțMV lrzjBj ðȚ ð 2Ț
In particular, the particles will be suspended (weightless) if
qp¼qeff¼ql/C6M
gdjBj
dzð3Ț
Where the gravity and the magnetic force work in opposite direc-
tions, the effective density ( qeff) becomes less than ql:
qeff¼qlțM
gdjBj
dzð4Ț
The gradient of | B| decreases in size with the distance to the
magnet that produces the field. Therefore particles of different
densities are suspended at different heights. In this way, the
MDS can be used to sort light polymers (polymers with different
densities less than that of ql), particularly polyolefins. The qlcan
be expressed as:
qeff¼ql/C0pMB 0
gpe/C0pz=pð5Ț
in which B0is the magnetic strength at the magnet surface.
2.2.2. MDS setup and experiments
The MDS setup applied in this study consisted of four steps: (i)
wetting, (ii) feeding, (iii) separating and (iv) collecting. Fig. 1 shows
the scheme of the lab-scale MDS setup. All the components of the
MDS setup are submerged under the liquid surface. The process
liquid, magnetic fluid, circulates in the system: it moves from the
left chamber to the right side of the MDS by the pressure difference
generated by the pumps, and then flows back to the left side.
The shredded input materials (RO HW or NL HW) were first
wetted with boiling water for one minute in order to make thepolyolefin surface hydrophilic ( Hu et al., 2010 ) and as well as to
remove heavy plastics (>1000 kg/m
3) or other contaminants. The
wetted samples are fed into a box made of stainless steel wire
gauze with openings of 1 mm. Air in the feeding box was first dis-
charged before the box was placed in position, to avoid air caused
turbulence in the system. When the lid of the box is open, the poly-
olefin particles rose up and then flowed into the separation chan-
nel with the main flow stream, thanks to the fact that PO have a
density smaller than 1000 kg/m3. In general, a plastic flake with a
thickness of 1 mm takes three seconds to reach their equilibrium
height in MDS ( Bakker et al., 2009; Hu, 2014 ), therefore the flow
speed in the separation channel was set at 0.18 m/s during the
tests. The separation channel was 0.6 m long, so the residence time
of the particles in the channel was slightly longer than three sec-
onds. To avoid turbulence, the upper and lower belt speed was
the same as the flow speed. The magnet used in the MDS has a2 S. Serranti et al. / Waste Management xxx (2014) xxx–xxx
Please cite this article in press as: Serranti, S., et al. An innovative recycling process to obtain pure polyethylene and polypropylene from househo ld waste.
Waste Management (2014), http://dx.doi.org/10.1016/j.wasman.2014.10.017

B0of 0.6 T and a pole size of 0.12 m. The strength of the process
liquid was about 120 A/m. At the end of the separation channel,
there were four splitters which produced five outputs: P1, P2, P3,
P4 and residues. The corresponding cut density of each splitter is
shown in Table 1 .
The RO HW and NL HW were separately processed with MDS
(Table 2 ). The output products of each test were analyzed for their
density distribution with a series of liquid standards of different
densities, in order to study their density distribution. Mixtures of
de-mineralized water and 96% ethanol were used to make liquid
standards with 10 kg/m3intervals covering the density range
between 880 kg/m3and 1000 kg/m3.
2.3. Hyperspectral imaging quality control
2.3.1. Hyperspectral imaging architecture setup
The analyses were carried out using a specifically HSI designed
platform (DV srl, Italy), allowing to acquire images of the particle
waste flow streams transported on a conveyor belt.
The core of the utilized acquisition system is a NIR Spectral
Camera™ (Specim, Finland), embedding an ImSpector N17E™
imaging spectrograph, working in spectral range from 1000 to
1700 nm, coupled with a Te-cooled InGaAs photodiode array sen-
sor with pixel resolution of 12 bits. The spectral resolution was
7 nm, for a total of 100 investigated wavelengths. The device works
as a push-broom type line scan camera allowing the acquisition of
spectral information for each pixel in the line. The acquired image
is arranged in a ‘‘hypercube’’, a 3D dataset characterized by two
spatial dimensions and one spectral dimension.The analytical platform is controlled by a PC unit equipped with
specialized acquisition/processing software (SScanner software),handling the different units and the sensing device and performing
the acquisition and the collection of spectra.
Spectral data analysis was carried out utilizing the PLS_Toolbox
(Version 7.5.1, Eigenvector Research, Inc.) running inside Matlab/C210
(Version 7.11.1, The Mathworks, Inc.), adopting standard chemo-
metric techniques ( Geladi et al., 2007 ).
The classification model was built according to the following
steps:
/C15selection of known PE and PP flakes from HW;
/C15flakes acquisition by NIR spectral camera;
/C15removal of the background noise, cutting the raw spectra at the
end of the wavelength range. The new investigated interval was
1000–1650 nm (93 wavelengths);
/C15spectra preliminary pre-processing in order to highlight the dif-
ferences between the two classes of materials;
/C15data exploration using a Principal Component Analysis (PCA)
based approach;
/C15classification and model validation utilizing a Partial Least
Square Discriminant Analysis (PLS-DA) based approach.
Selected flakes, clearly identified as PE and PP, were used as
training set to define the classification model. The entire surface
of the flakes was selected to build the dataset, including the edges
in order to reduce the classification error due to the border effect.
The model was validated applying it to two external sets of known
PE and PP flakes.
2.3.2. Spectra preprocessing and Principal Component Analysis (PCA)
Spectral data were pre-processed in order to highlight the dif-
ferences between the two materials. Several preprocessing algo-
rithms were applied to the raw spectra, the combination of
Baseline ,Derivative (2nd order) and Mean Center led to the best
results. After pre-processing, an exploratory analysis was carried
out applying PCA to the spectral data ( Wold et al., 1987 ). PCA com-
presses the data by projecting the samples into a low dimensional
subspace, whose axes (the principal components, PCs) point in the
directions of maximal variance. Looking at the distribution of the
samples into the PC space it is possible to analyze their common
features and/or their grouping.
Each principal component (PC) consists of scores and loadings:
the first indicate the association between the samples and second
illustrate how variables relate. The first few principal components
generated from PCA are usually used to analyze the common fea-
tures among samples: samples with similar spectra tend to aggre-
gate in the score plot of the first two or three components.Fig. 1. Scheme of the lab-scale MDS for polyolefin (PO) recycling. The wetting section is not shown.
Table 1
Split position (distance from magnet surface) of product and theoretical correspond-ing density.
Product Position range (mm) Corresponding density range (kg/m3)
P1 81–103 980–1000
P2 59–81 960–980
P3 37–59 930–960
P4 10–37 860–930Residue 0–10 –
Table 2
Plastic samples processed by MDS.
Test Sample name Mass (g)
T1 RO HW 2470
T2 NL HW 2400S. Serranti et al. / Waste Management xxx (2014) xxx–xxx 3
Please cite this article in press as: Serranti, S., et al. An innovative recycling process to obtain pure polyethylene and polypropylene from househo ld waste.
Waste Management (2014), http://dx.doi.org/10.1016/j.wasman.2014.10.017

2.3.3. Partial Least Square-Discriminant Analysis (PLS-DA)
PLS-DA is a regression extension of PCA that takes advantage of
class information to maximize the separation between groups of
observations ( Barker and Rayens, 2003 ). PLS-DA is a supervised
classification technique, requiring a prior knowledge of the data.
PLS-DA is used to classify samples into predefined groups by form-
ing discriminant functions from input variables (wavelengths) to
yield a new set of transformed values that provides a more accu-
rate discrimination than any variable (wavelength) alone. There-
fore when the model is built it can be applied for the
classification of new hypercubes. The result of a PLS-DA classifica-
tion is a prediction map in which each pixel is identified by the
color associated to the predicted class.
The PLS-DA model was built using the same preprocessing
adopted for the PCA analysis. After preprocessing, the original data,constituted by 93 spectral intensities, were divided into a calibra-
tion and a validation dataset. Contiguous block algorithm was uti-
lized for the cross validation, this algorithm separates the original
dataset into continuous subset alternatively utilized to build up
and validate the model ( Serranti et al., 2011 ).
2.3.4. Classification procedure
The PLS-DA model built in Matlab/C210was used for the direct clas-
sification of plastic flakes coming from the MDS. The quality con-
trol of the products separated by MDS was carried out according
to the following steps:
/C15flakes positioning on the conveyor belt;
/C15acquisition of the hyperspectral images;
/C15masking the background using a threshold mode: pixels having
reflectance lower than 0.2 at 1000 nm were hidden;
/C15application of the classification model to the masked image:
visualization of PP and PE flakes in two different colours (green
and red, respectively) and measurements of the area belongingto each class of material;
/C15application of the ‘‘ blob analysis ’’ to the classified images.
/C15area% measurement of PP and PE flakes.
The ‘‘ blob analysis ’’ was applied to eliminate misclassification
errors due to the border effects. The algorithm, starting from the
identified objects in an image, assigns the same class to the entire
object, based on the class associated to the majority of pixels
belonging to that particle. An example of the application of ‘‘ blob
analysis ’’ is reported in Fig. 2 .Based on the classification results obtained by HSI, PE and PP
flakes were first manually separated in two groups for each ana-
lyzed product and then weighted in order to compare these results
with those obtained by the ‘‘optical approach’’ (i.e.: area measure-
ment) and by classical density distribution carried out in
laboratory.
3. Results and discussion
3.1. Magnetic density separation
The mass distribution of products of each test is presented in
Table 3 . In addition to the heavy fraction (>1000 kg/m3), there were
also small amounts of residues (about 1% in both tests, see in
Table 3 ), mixtures of dust, dirt and few curved flakes, generated
Fig. 2. Example of application of ‘‘blob analysis’’ to the plastic flakes. classified PE (red) and PP (green) particles without (a) and with (b) blob analysis. ( For interpretation of
the references to colour in this figure legend, the reader is referred to the web version of this article.)Table 3
Mass distribution of products in the tests.
Product T1 (%) T2 (%)
P1 0.3 1.2%
P2 19.4 26.6P3 43.7 40.3
P4 35.1 31.0
Residue 1.1 0.9>1000 kg/m
30.4 –
Total 100.0 100.0
0% 20% 40% 60% 80% 100%
P4 P3 P2 P1Mass %
MDS Products
<920 kg/m3 920 -930 kg/m3 >930kg/m3
Fig. 3. Density grade of products obtained for RO HW sample (test T1).4 S. Serranti et al. / Waste Management xxx (2014) xxx–xxx
Please cite this article in press as: Serranti, S., et al. An innovative recycling process to obtain pure polyethylene and polypropylene from househo ld waste.
Waste Management (2014), http://dx.doi.org/10.1016/j.wasman.2014.10.017

during the MDS process. The MDS processes recovered most of the
PO (<1000 kg/m3). In T1 and T2, the recovery of PP and PE reached
around 99%. The product distributions of T1 and T2 among the out-
puts are comparable, because of the similarity of the input
materials.
3.1.1. Density grade
The density grade of each product of the two tests is presented
inFigs. 3 and 4 . In both T1 and T2, the grade of the <920 kg/m3frac-
tion was more than 93% in P4. In P3, this grade dropped to less than
4–5%. The concentration of this fraction in P1 and P2 was <2%.
According to the analysis of Hu et al. (2013) , PP, LDPE and HDPE
are mainly found in the density range of <920 kg/m3, 920–
930 kg/m3and >930 kg/m3respectively, therefore the concentra-
tion of the PP and PE can be well interpreted from the density
grade.3.2. Hyperspectral imaging
3.2.1. Exploratory analysis
An example of selected PE and PP flakes used as training set,
with the corresponding image acquired by HSI, showing the class
selection, is reported in Fig. 5 . The related average reflectance spec-
tra of PE and PP flakes are shown in Fig. 6 . It is clear that the two
polymers have different spectral signatures in the investigated
NIR spectral region (1000–1650 nm), in fact the absorption bands
due to the overtones or combination bands of carbon-hydrogen
(C–H) vibrations can be observed. The two types of PO can be
clearly recognized looking at the different spectra shape in the
regions between 1150–1250 nm, 1350–1450 nm and 1500–
1550 nm, due to their different chemical formula and structure.
The results of the pre-processing algorithms ( Baseline ,2nd
Derivative and Mean Centre ) applied to the mean spectra of the
two polymers are shown in Fig. 7 . The pre-processing procedure
clearly produces two spectral signatures that are quite different
in terms of trend from the original ones. PE and PP original col-
lected spectra are, in fact, characterized by the same trend, after
preprocessing they show a very different trend. Differences
between PE and PP are thus strongly enhanced. In fact, in the same
wavelength regions one processed spectra is characterized by
peaks (i.e. PE) the other by valleys (i.e. PP). This fact allows to per-
form a strong discrimination between PE and PP. The results of
PCA, applied to the selected PE and PP flakes, shown in Fig. 8 ,
clearly sustain what previously affirmed.
The PCA approach, decomposing spectral data into several Prin-
cipal Components (PCs) (linear combinations of the original spec-
tral data) embedding the spectral variations of each collected
spectral data set, allows to outline the difference existing among
the PE and PP samples. It appears as most of the variance was cap-
tured by the first three PCs (PC1, PC2 and PC3) explaining 72.63%,0% 20% 40% 60% 80% 100%
P4 P3 P2 P1Mass %
MDS Products
<920 kg/m3 920 – 930 kg/m3 >930kg/m3
Fig. 4. Density grade of products obtained for NL HW sample (test T2).
Fig. 5. Training set of known PE and PP flakes (left) and corresponding acquired image by HSI showing the two classes selection (right).
Fig. 6. Average reflectance spectra of PE and PP flakes in the range between 1000 nm and 1650 nm.S. Serranti et al. / Waste Management xxx (2014) xxx–xxx 5
Please cite this article in press as: Serranti, S., et al. An innovative recycling process to obtain pure polyethylene and polypropylene from househo ld waste.
Waste Management (2014), http://dx.doi.org/10.1016/j.wasman.2014.10.017

25.16% and 1.01% of the variance, respectively. The spectra related
to the two different materials are clustered in two well separated
regions of the score plot, indicating that the two polymers can be
recognized using the first three principal components.
3.2.2. Classification by PLS-DA
The PLS-DA model built for the classification of PE and PP flakes
from HW was applied to different sets of known plastic particles as
shown in Figs. 9 and 10 . The model can clearly identify PE and PP
flakes, classified in blue and red, respectively.
InTable 4 , the Sensitivity and Specificity of the classification
model are reported for both calibration (Cal) and cross validation
Fig. 7. Preprocessed spectra of PE and PP after the sequential application of Baselin e,Derivative (2nd order) and Mean Centre procedures.
Fig. 8. 3D score plot (PC1 vs PC2 vs PC3) in the NIR wavelength region related to PE
and PP flakes.
(a) (b) 3 cm
Fig. 9. (a) Set of plastic flakes from RO HW sample used for the validation of the
PLS-DA model and (b) corresponding classified image. PE particles are displayed inblue and PP particles in red. (For interpretation of the references to colour in thisfigure legend, the reader is referred to the web version of this article.)
(a) (b)3 cm
Fig. 10. (a) Set of plastic flakes from NL HW sample used for the validation of the
PLS-DA model and (b) corresponding classified image. PE particles are displayed inblue and PP particles in red. (For interpretation of the references to colour in this
figure legend, the reader is referred to the web version of this article.)
Table 4
Sensitivity and specificity related to the PLS-DA model built for
the classification of PE and PP from HW.
Class 1 (PE) Class 2 (PP)
Sensitivity (Cal) 1.00 1.00
Specificity (Cal) 1.00 1.00Sensitivity (CV) 1.00 1.00Specificity (CV) 1.00 1.006 S. Serranti et al. / Waste Management xxx (2014) xxx–xxx
Please cite this article in press as: Serranti, S., et al. An innovative recycling process to obtain pure polyethylene and polypropylene from househo ld waste.
Waste Management (2014), http://dx.doi.org/10.1016/j.wasman.2014.10.017

P1 P2 P3 P4
(a) Source Images
(b) Classified Images with Blob Analysis3 cm 3 cm 3 cm 3 cm
Fig. 11. (a) Source images of the four products (P1, P2, P3 and P4) obtained by MDS for the Romanian HW sample (T1), and (b) corresponding classified images. PE par ticles
are in red and PP particles in green. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of th is article.)
P2 P3 P4
(a) Source Images
(b) Classified Images with Blob Analysis3 cm 3 cm 3 cm
Fig. 12. (a) Source images of the three products (P2, P3 and P4), obtained by MDS for the Dutch HW sample (T2), and (b) corresponding classified images. PE particl es are in
red and PP particles in green. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this art icle.)S. Serranti et al. / Waste Management xxx (2014) xxx–xxx 7
Please cite this article in press as: Serranti, S., et al. An innovative recycling process to obtain pure polyethylene and polypropylene from househo ld waste.
Waste Management (2014), http://dx.doi.org/10.1016/j.wasman.2014.10.017

(CV). Sensitivity and Specificity are statistical parameters indicating
the performance of a binary classification test, also known in sta-
tistics as classification function. Sensitivity measures the propor-
tion of actual positives which are correctly identified as such,
while Specificity measures the proportion of negatives which are
correctly identified. A perfect predictor would be described as sen-
sitivity = 1 and specificity = 1.
Perfect results were obtained for the two classes of materials
having both Sensitivity and Specificity (for calibration and cross val-
idation) equal to 1, which means that the PLS-DA model can clearly
distinguish PE particles from PP particles.
3.2.3. Quality control by HSI
The PLS-DA model was applied to directly classify the plastic
flakes resulting from the MDS separation. More in details, HSI
was applied to the 4 different products (P1, P2, P3 and P4) obtained
by MDS for Romanian HW sample, as resulting from test T1 and to
the 3 different products (P2, P3 and P4) obtained by MDS for the
Dutch HW sample, as resulting from test T2. The classification
results are reported in terms of prediction maps in Figs. 11 and
12. The quality of the different products (i.e. presence of PP in a
PE streams and vice versa) results clearly defined according to their
recognized PE and PP content: PE and PP particles mapped in red
and green, respectively.For each product, the area% of classified
PE and PP flakes was computed after the application of ‘‘ blob anal-
ysis’’. The PE and PP flakes were then weighed. The results in terms
of computed area% and in terms of weight% (measured after HSI
classification and obtained by classical density distribution in lab-
oratory) for PE and PP flakes, with reference to the two different
investigated HW samples, are reported in Tables 5 and 6 . Looking
at the tables it is clear as the classification obtained by HSI reflects
the classification carried out checking density in laboratory, con-
sidering both area and weight measurements.
The area measurement can be used in an on-line system to
roughly estimate the purity of PE and PP products in weight%,
assuming a constant thickness and an average density value for
the each of the two PO.
Concerning the comparison between weight, it is interesting to
notice that, based on HSI classification, results are slightly better in
terms of quality of the separated products, which means an evenhigher grade of purity of PE and PP products obtained by MDS. It
is reasonable to assume that the classification carried out by HSI
can be used as a real quality control of the separation process,
being based on the spectral signature characteristics of each poly-
mer. On the contrary, the density check carried out at laboratory
scale, can be considered a first confirmation of the sorting effi-
ciency but, since the Gaussian density distributions of PP and PE
overlap each other, some misclassifications can occur. Further-
more, HSI quality control can be carried out in a faster and easier
way, without sampling and without waiting for laboratory results.
4. Conclusions
The trial separations of polyolefin mixtures coming from Roma-
nian and Dutch HW show promising results for using the MDS to
generate high quality PE and PP products. The grade of purity for
the RO HW reached 100% in two of the PE products (P1 and P2)
and 96.3% in P3; P4 had a PP content equal to 94.2%. Good results
were obtained also for the NL HW with a 98.8% PE grade in P2 and
95.7% PP grade in P4. These values allow to compare the recycled
PE and PP to the virgin polymers.
The quality control of the different products obtained by the
separation process was assessed by an HSI system working in the
NIR range. The results of the classification obtained by HSI were
consistent, reliable and comparable with those obtained by density
classification carried out in laboratory. The short elaboration time
for the plastic flakes classification (in the order of few seconds)
allows to perform an on-line quality check of the output products
obtained from the separation unit.
Acknowledgements
The study was realized thanks to the financial support of the
European Commission in the framework of the FP7 Collaborative
project ‘‘Magnetic Sorting and Ultrasound Sensor Technologies
for Production of High Purity Secondary Polyolefins from Waste
(W2Plastics)’’, Grant Agreement No. 212782.
References
Al-Salem, S.M., Lettieri, P., Baeyens, J., 2009. Recycling and recovery routes of
plastics solid waste (PSW): a review. Waste Manage. 29, 2625–2643 .
Bakker, E.J., Rem, P.C., 2006. Magneto-hydrostatic separation of PET. In: Proceedings
of the 5th International Conference for Conveying and Handling of Particulate
Solids, Sorrento, Italy. Aug 27–31.
Bakker, E.J., Muchová, L., Rem, P.C., 2007. Separation of precious metals from MSWI
bottom ash. In: Proceedings of the Sixth International Industrial Minerals
Symposium, February 1–3, Izmir, Turkey, pp. 322–327.
Bakker, E.J., Rem, P.C., Fraunholcz, N., 2009. Upgrading mixed polyolefin waste with
magnetic density separation. Waste Manage. 29, 1712–1717 .
Barker, M., Rayens, W., 2003. Partial least squares for discrimination. J. Chemom. 17,
166–173 .
Burat, F., Güney, A., Olgaç Kangal, M., 2009. Selective separation of virgin and post-
consumer polymers (PET and PVC) by flotation method. Waste Manage. 29,
1807–1813 .
Bonifazi, G., Serranti, S., 2006. Imaging spectroscopy based strategies for ceramic
glass contaminants removal in glass recycling. Waste Manage. 26, 627–639 .
Dall’Ara, A., Bonoli, A., Serranti, S., 2012. An innovative procedure to characterize
properties from tailored compost. Environ. Eng. Manage. J. 11, 1825–1832 .
Di Maio, F., Rem, P., Hu, B., Serranti, S., Bonifazi, G., 2010. The W2Plastics project:
exploring the limits of polymer separation. Open Waste Manage. J. 3, 90–98 .
Dodbiba, G., Haruki, N., Shibayama, A., Miyazaki, T., Fujita, T., 2010. Combination of
sink-float separation and flotation technique for purification of shredded PET-bottle from PE or PP flakes. Int. J. Miner. Process. 65, 11–29
.
Geladi, P., Grahn, H., Burger, J., 2007. Techniques and Applications of Hyperspectral
Image Analysis. John Wiley & Sons, England, pp. 1–15 .
Hu, B., Fraunholz, N., Rem, P.C., 2010. Wetting technologies for high-accuracy sink-
float separations in water-based media. Open Waste Manage. J. 3, 71–80 .
Hu, B., Giacometti, L., Di Maio, F., Rem, P., 2011. Recycling of WEEE by magnetic
density separation. In: Selected proceedings of The Sixth International
Conference on Waste Management and Technology (ICWMT 6), 30 August–1
September 2011, Suzhou, China.Table 5
Comparison between the results obtained by density classification carried out inlaboratory and by HSI classification in terms of computed area% and after weighting
the PE and PP flakes, related to MDS test T1 (Romanian HW sample).
Product Classification by density Classification by HSI
PE PP PE PPMass% Mass% Area% Mass% Area% Mass%
P1 100.0 0.0 100 100.0 0 0.0
P2 100.0 0.0 100 100.0 0 0.0P3 96.0 4.0 96.4 96.3 3.6 3.7P4 7.0 93.0 7.8 5.8 92.2 94.2
Table 6
Comparison between the results obtained by density classification carried out in
laboratory and by HSI classification in terms of computed area% and after weightingthe PE and PP flakes, related to MDS test T2 (Dutch HW sample).
Product Classification by density Classification by HSI
PE PP PE PPMass% Mass% Area% Mass% Area% Mass%
P2 98.5 1.5 98.2 98.8 1.8 1.2
P3 93.9 6.1 95.8 95.5 4.2 4.5P4 6.1 93.9 3.9 4.3 96.1 95.78 S. Serranti et al. / Waste Management xxx (2014) xxx–xxx
Please cite this article in press as: Serranti, S., et al. An innovative recycling process to obtain pure polyethylene and polypropylene from househo ld waste.
Waste Management (2014), http://dx.doi.org/10.1016/j.wasman.2014.10.017

Hu, B., Serranti, S., Fraunholcz, N., Di Maio, F., Bonifazi, G., 2013. Recycling-oriented
characterization of polyolefin packaging waste. Waste Manage. 33, 574–584 .
Hu, B., 2014. Magnetic density separation of polyolefin waste. Doctoral thesis of
Delft University of Technology, Delft, the Netherlands, ISBN 978-94-6169-526-
0, 41. DOI: http://dx.doi.org/10.4233/uuid:0c3717fa-8000-4de0-a938-
d65605bf2a96 .
Luciani, V., Serranti, S., Bonifazi, G., Di Maio, F., Rem, P., 2013. Quality control in the
recycling stream of PVC from window frames by hyperspectral imaging, Optical
Sensors 2013Proc. SPIE 8774, 87741N, Prague, Czech Republic, April 15–17,
DOI: http://dx.doi.org/10.1117/12.2014755 .
Murariu, V., Svoboda, J., Sergeant, P., 2005. The modelling of the separation process
in a ferrohydrostatic separator. Miner. Eng. 18, 449–457 .
Palmieri, R., Bonifazi, G., Serranti, S., 2014. Recycling-oriented characterization of
plastic frames and printed circuit boards from mobile phones by electronic and
chemical imaging. Waste Manage. 34, 2120–2130 .
Plastics – the Facts, 2013. An analysis of European latest plastics production,
demand and waste data. < http://www.plasticseurope.org/documents/
document/20131018104201-plastics_the_facts_2013.pdf >.Pongstabodee, S., Kunachitpimol, N., Damronglerd, S., 2008. Combination of three-
stage sink-float method and selective flotation technique for separation of
mixed post-consumer plastic waste. Waste Manage. 28, 475–483 .
Serranti, S., Gargiulo, A., Bonifazi, G., 2011. Characterization of post-consumer
polyolefin wastes by hyperspectral imaging for quality control in recycling
processes. Waste Manage. 31, 2217–2229 .
Serranti, S., Gargiulo, A., Bonifazi, G., 2012. Hyperspectral imaging for process and
quality control in recycling plants of polyolefin flakes. J. Near Infrared Spectrosc.
20 (5), 573–581 .
Ulrici, A., Serranti, S., Ferrari, C., Cesare, D., Foca, G., Bonifazi, G., 2013. Efficient
chemometric strategies for PET–PLA discrimination in recycling plants usinghyperspectral imaging. Chemometr. Intell. Lab. Syst. 122, 31–39
.
Wold, S., Esbensen, K., Geladi, P., 1987. Principal component analysis. Chemometr.
Intell. Lab. Syst. 2, 37–52 .S. Serranti et al. / Waste Management xxx (2014) xxx–xxx 9
Please cite this article in press as: Serranti, S., et al. An innovative recycling process to obtain pure polyethylene and polypropylene from househo ld waste.
Waste Management (2014), http://dx.doi.org/10.1016/j.wasman.2014.10.017