PreprintPDF Available

LASER-BASED SORTING OF CONSTRUCTION AND DEMOLITION WASTE FOR THE CIRCULAR ECONOMY

Authors:
Preprints and early-stage research may not have been peer reviewed yet.

Abstract and Figures

Closed material cycles and unmixed material fractions are required to achieve high recovery and recycling rates in the building industry. The growing diversity of construction and demolition waste is leading to increasing difficulties in separating the individual materials. Manual sorting involves many risks and dangers for the executing staff and is merely based on obvious, visually detectable differences for separation. An automated, sensor-based sorting of these building materials could complement or replace this practice to improve processing speed, recycling rates, sorting quality, and prevailing health conditions. A joint project of partners from industry and research institutions approaches this task by investigating and testing the combination of laser-induced breakdown spectroscopy (LIBS) and VIS/NIR spectroscopy. Joint processing of information (data fusion) is expected to significantly improve the sorting quality of various materials like concrete, main masonry building materials, organic components, etc., and may enable the detection and separation of impurities such as SO3-cotaining building materials (gypsum, aerated concrete, etc.). Focusing on Berlin as an example, the entire value chain will be analyzed to minimize economic/technological barriers and obstacles at the cluster level and to sustainably increase recovery and recycling rates. First LIBS measurements show promising results in distinguishing various material types. A meaningful validation shall be achieved with further practical samples. Future works will investigate the combination of LIBS and VIS/NIR spectroscopy in a fully automated measurement setup with conveyor belt speeds of 3 m/s.
Content may be subject to copyright.
LASER-BASED SORTING OF CONSTRUCTION AND DEMOLITION WASTE FOR THE
CIRCULAR ECONOMY
Tim Klewe1, Tobias Völker1, Mirko Landmann2, Gerd Wilsch1, Sabine Kruschwitz1,3
1 Bundesanstalt für Materialforschung und -prüfung (BAM), Berlin, Germany
2 Institut für Angewandte Bauforschung Weimar gGmbH, Weimar, Germany
3 Technische Universität Berlin, Berlin, Germany
Keywords NDT, LIBS, data fusion, material classification, recycling
Corresponding author: tim.klewe@bam.de
ABSTRACT
Closed material cycles and unmixed material fractions are required to achieve high recovery and
recycling rates in the building industry. The growing diversity of construction and demolition waste is
leading to increasing difficulties in separating the individual materials. Manual sorting involves many
risks and dangers for the executing staff and is merely based on obvious, visually detectable differences
for separation. An automated, sensor-based sorting of these building materials could complement or
replace this practice to improve processing speed, recycling rates, sorting quality, and prevailing health
conditions.
A joint project of partners from industry and research institutions approaches this task by investigating
and testing the combination of laser-induced breakdown spectroscopy (LIBS) and visual (VIS)/ near-
infrared (NIR) spectroscopy. Joint processing of information (data fusion) is expected to significantly
improve the sorting quality of various materials like concrete, main masonry building materials, organic
components, etc., and may enable the detection and separation of impurities such as SO3-containing
building materials (gypsum, aerated concrete, etc.). Focusing on Berlin as an example, the entire value
chain will be analyzed to minimize economic/technological barriers and obstacles at the cluster level
and to sustainably increase recovery and recycling rates.
First LIBS measurements show promising results in distinguishing various material types. A meaningful
validation shall be achieved with further practical samples. Future works will investigate the
combination of LIBS and VIS/NIR spectroscopy in a fully automated measurement setup with conveyor
belt speeds of 3 m/s.
INTRODUCTION
To establish a closed-loop cycle in the building materials industry it is necessary to ensure a sufficient
purity of crushed and sorted recycled material. One of the major problems in processing of future
construction and demolition waste (CDW) is the increasing material diversity and amount of composite
materials and structures (Figure 1). Different foreign materials and impurities can be attached on usable
materials that may disturb subsequent processing and sorting techniques as well as the reuse.
Unfortunately, in CDW recycling, the preference to date has been to apply simple but proven techniques
in order to process large quantities of construction rubble in a short time. This contradicts with
increasingly complex composite materials and structures within the mineral building material industry.
One of the most promising techniques is sensor-based sorting. Through the installation of such sensor-
based single particle sorting devices in building materials recycling, the heterogeneous bulk materials
could be converted into homogeneous material fractions. Here, the performance limitations of the
prevalent sensor techniques, as well as economic aspects, must be considered. At the present level of
technology, only coarser aggregates can be technically and economically sorted by this technique.
Current investigations on sensor-based sorting technologies for CDW are based on analyzing the visual
(VIS) or near-infrared (NIR) spectrum [1].
The aim must be to detect and reject foreign materials and impurities before processing. Up to date, pre-
sorting technologies are not applied in processing of CDW. However, this is necessary to prevent further
spreading of unwanted materials in usable material streams. This applies in particular to sulphate-
containing building materials, such as gypsum plasters and others.
A current research project LIBS-ConSort aims to sort mineral CDW under the use of Laser-induced
breakdown spectroscopy (LIBS) combined with VIS/NIR spectroscopy. The extension of the
established camera-based techniques is expected to significantly improve the reliability of classifying
diverse building material groups. To emphasize this outlook, this study presents preliminary results by
using LIBS alone on various CDW representatives.
Application of LIBS
Laser-induced breakdown spectroscopy is a combination of laser evaporation and plasma excitation
followed by analysis of the emitted radiation by optical emission spectroscopy. The main advantages of
LIBS are (i) almost no sample preparation, (ii) all chemical elements can be measured, (iii) short
measurement times at high measurement frequency, (iv) real-time evaluation and process control, and
(v) non-contact measurement at distances ranging from a few millimeters to several meters.
The listed advantages make LIBS ideal for in-line process control. Application examples include process
control in the steel and glass industry, quality assurance in the pharmaceutical industry or for mix-up
control in metal processing. Other application examples show the potential for the identification of raw
and residual materials in a wide variety of application areas and material classes, such as for cements
[2], construction residues [3,4], soils [5], polymers [6], aluminum alloys [7] or scrap metal [8].
Compared to the hyperspectral technique mentioned in the introduction, LIBS has the advantage that
the direct chemical "fingerprint" of the material is used for classification instead of the surface color.
METHOD
For the purpose of examining the potential of LIBS measurements to distinguish various building
materials, common representatives were composed in a set of samples that is listed in Table 1. Each of
the seven material groups contain two to nine different specimens, resulting in a total amount of 29
samples. Figure 1 shows one example of each group.
On these materials, LIBS measurements were performed with a FiberLIBSLab system (SECOPTA
analytics GmbH) using a 3 mJ laser with a wavelength of 1064 nm. With a frequency of 50 Hz, the laser
produces 1.5 ns width pulses, which excite a plasma on the materials surface. The light emitted by the
plasma is then analyzed by a spectrograph ranging from 178 nm to 956 nm (covering the near ultraviolet
(NUV) through the visible light band to the near infrared (NIR) frequencies) with a resolution of
approximately 0.1 nm. On each specimen, a quadratic area of 20 mm x 20 mm was examined (see
Figure 1, left). The x- and y-grid spacing was chosen to be 1 mm, which resulted in 441 measurement
Figure 1 - Composite materials and structures in mineral CDW with attachments of foreign materials
and impurities
points. Therefore, the entire dataset contains 12.789 spectra with 7060 data points per spectrum. It forms
the basis for the subsequent training of a classifier, which is discussed in the following section.
Table 1- Variety and number of building materials investigated with LIBS
Material group
Number of specimens
Lime Sandstone
2
Light Concrete
2
Slag
2
Aerated Concrete
2
Facing Brick
8
Roof Tiles
4
Brick
9
Classification
To generate an input vector for the following classification task, the spectral line intensities of ten
different elements are extracted from each recorded spectrum (see Table 2). For each spectral line, a
baseline correction is performed followed by a calculation of the integral line intensity. An exemplary
spectrum with the spectral lines used is shown in Figure 3.
The reduced 12.789 × 10 data matrix (number of recorded spectra × number of extracted spectral line
intensities) is labeled according to the material groups in Table 1, which also define the seven desired
output classes for the supervised training process. Here, a random forest classifier, included in the scikit-
learn library (python) [9], is used in standard configuration (default parameters only). Such classifiers
combine multiple decision trees that are individually trained on randomly chosen sub-samples of the
dataset. Their individual decision for new input data is averaged in a voting process, where the majority
determines the final classification.
To avoid the scenario of overfitting and to simulate a realistic classification problem of unknown
samples, a cross validation is performed as follows: According to the total number of samples, we define
29 different training data sets, each excluding one individual sample, which in turn serves as the
respective test data. This shall test the capability of a trained classifier to recognize an unknown sample.
An exclusively correct classification on known samples and poor accuracies on others would indicate
Figure 2 - Exemplary pictures of each considered material group and the examined measuring area
for LIBS (left).
overfitting. Therefore, 29 models are trained and applied on their respective unkown test sample to
predict the material group. Instead of presenting 29 individual results, the achieved accuracies are
accumulated and evaluated in one confusion matrix (see later Figure 5). The results are shown and
discussed in the following section.
Table 2- Chemical elements and their center wavelength in the LIBS spectrum used for material
classification and allocation of the according building material class
RESULTS & DISCUSSION
A general overview of the measured data set is given in Figure 4 by showing the mean intensities and
standard deviations of the ten extracted element features for every material group. The Figure allows a
first impression of the feature-based separability of the examined samples and reveals characteristic
distributions between the material groups. Aerated concrete, generally made of quartz sand and burnt
lime, shows particularly prominent intensities for Si, Ca, and O, which are the dominant elements of the
named components. However, also for Mg and Ti the material group clearly stands out. Bricks and roof
tiles, which are based on clay minerals, show typical increased intensities for Fe and Al, but also for the
Element (symbol)
Iron (Fe)
Magnesium (Mg)
Silicon (Si)
Aluminum (Al)
Calcium (Ca)
Titanium (Ti)
Potassium (K)
Sodium (Na)
Oxygen (O)
Sulfur (S)
Fe Mg Si Al Ca Ti KNa OS
Intensity [a.u.]
Wavelength [nm]
Intensity [a.u.]
Wavelength [nm]
Wavelength [nm]
Intensity [a.u.]
Figure 3 - Full exemplary wavelength spectrum of a sandstone (top) and the extracted, element
specific spectral lines (bottom) used as features for classification.
alkali metals K and Na. Further, slag shows lower intensities for every extracted element, whereas lime
sandstone and light concrete mostly lie in between the given extremes.
The discussed distributions provide suitable information to understand the achieved classification
accuracies for each material group, which are shown by the confusion matrix in Figure 4. The scores are
given in the range from 0 to 1, standing for 0 % to 100 % of the respective test data. Here, the diagonal
includes the most meaningful values and shows accuracies between 60 % and 96 %. Brick and facing
brick can be classified best, followed by aerated concrete. Also, slag and roof tiles achieve satisfying
scores above 80 %, showing a good separability based on the chosen element features. Light concrete
Figure 4 - Mean intensities and standard deviations (black error bars) of the extracted element peaks
for each material group
Intensity [a.u.]
Element Feature
Lime
Roof Tiles
Brick
Lime Sandstone
Light Concrete
Aerated Concrete
Slag
Facing Brick
Roof Tiles
Brick
Lime Sandstone
Light Concrete
Aerated Concrete
Slag
Facing Brick
True Class
Predicted Class
0.17
0
0
0
0
0
00.01
0
0
0
0
0
00
0
0
0
0
0
0
0
0.01
0.01
0
0.01
0.01
0.01
0.01
0.01
0.02
0.02
0.03
0.02
0.06
0.06
0.11
0.92
0.84
0.11 0.89
0.15 0.73 0.09
0.60 0.19 0.14
0.96
0.82
Figure 5 - Confusion matrix of the performed cross validation with a random forest classifier,
showing the achieved accuracies. Rows contain the true classes (black) and columns the predicted
classes (blue).
and lime sandstone were less easily detectable. In 15 % or 19 % of the cases, respectively, both material
groups could not be distinguished from each other. As mentioned above, they barely stand out regarding
the element specific intensities in Figure 3. Further, light concrete and lime sandstone often show quite
similar and overlapping distributions across all features, which underlines the general danger of
confusion within this data set. Obviously, this is even emphasized by the limited amount of only two
specimens for each group. In contrast, aerated concrete and slag, which are as well represented by two
specimens, show a better separability by having the highest or lowest intensity over most features,
respectively. Until now it cannot be said if these findings are characteristic for the respective material
group. More representatives of each class will give additional insights into their separability. Machine
learning approaches generally require a large and diverse data basis to achieve accurate and robust
classification results on new samples. This is further underlined by the good results for brick and facing
brick, which included the highest number of specimens.
SUMMARY & CONCLUSION
This paper showed first results of classifying various building material groups by the sole use of the
LIBS technology. Therefore, we measured the spectral information of 29 different CDW representatives
from which ten element specific features were extracted. These features were used to train numerous
random forest classifiers to be individually used on unknown test samples.
Brick, facing brick, aerated concrete, slag, and roof tiles achieved satisfying accuracies, ranging from
96 % to 82 %. Here, their characteristic distributions in the feature data set or the presence of a diverse
and quantitatively higher sample set led to good results. Light concrete and lime sandstone showed lower
accuracies of 73 % and 60 %, respectively, and were often confused with each other. Beside the fact
that only two samples of both groups were included in the data set, their features distributions mostly
overlapped, offering a difficult separability.
Since the presented work outlines only the first results of an ongoing research project, the achieved
accuracies can be generally seen as promising. By collecting more samples, the underlying data set is
expected to grow significantly and will serve as a sound basis for the planned methodological
combination with VIS/NIR-cameras. So far, the sole use of LIBS already showed the potential to sort
CDW with a satisfactory reliability.
Future works will also investigate the influence of different moisture contents and surface contamination
(dust, earth) on each individual measurement method as well as on the classification results. The findings
shall help to plan an effective preprocessing of CDW within the industrial prototype, which may include
washing the samples before the identification.
REFERENCES
1. Müller, A., “Erschließung der Ressourceneffizienzpotentiale im Bereich der Kreislaufwirtschaft
Bau“, Project report in the research program Zukunft Bau, Bundesinstitut für Bau-, Stadt- und
Raumforschung im Bundesamt für Bauwesen und Raumordnung, 2016
2. Völker, T.; Millar, S.; Strangfeld, C.; Wilsch, G., „Identification of type of cement through
laser-indusced breakdown spectroscopy”. Construction and Building Materials, vol. 258,
DOI: 10.1016/j.conbuildmat.2020.120345, 2020
3. Xia, H.; Bakker, M. C. M.; “Reliable classification of moving waste materials with LIBS in
concrete recycling”. Talanta, vol. 120, DOI: 10.1016/j.talanta.2013.11.082, 2014
4. Xia, H.; Bakker, M. C. N., "Single-shot LIBS spectral quality for waste particles in open air".
tm - Technisches Messen, vol. 82, pp. 606-615. DOI: 10.1515/teme-2015-0042, 2015
5. Pontes, M. J. C.; Cortez, J.; Galvão, R. K. H.; Pasquini, C.; Araújo, M. C. U.; Coelho, R. M.;
Chiba, M. K.; Ferreira de Abreu, M.; Madari, B. E., “Classification of Brazilian soils by using
LIBS and variable selection in the wavelet domain”. Analytica Chimica Acta, vol. 642, 1-2,
pp. 12-18, DOI: 10.1016/j.aca.2009.03.001, 2009
6. Costa, V. C.; Aquino, F. W. B.; Paranhos, C. M.; Pereira-Filho, E. R., “Identification and
classification of polymer e-waste using laser-induced breakdown spectroscopy (LIBS) and
chemometric tools”. Polymer Testing, vol. 59, pp. 390-395,
DOI: 10.1016/j.polymertesting.2017.02.017, 2017
7. Campanella, B.; Grifoni, E.; Legnaioli, S.; Lorenzetti, G.; Pagnotta, S.; Sorrentino, F.; Palleschi,
V., “Classification of wrought aluminum alloys by Artificial Neural Networks evaluation of
Laser Induced Breakdown Spectroscopy spectra from aluminum scrap samples”.
Spectrochimica Acta Part B: Atomic Spectroscopy, vol. 134, pp. 52-57,
DOI: 10.1016/j.sab.2017.06.003, 2017
8. Merk, S.; Scholz, C., Florek, S.; Mory, D., „Increased identification rate of scrap metal using
Laser Induced Breakdown Spectroscopy Echelle spectra”, Spectrochimica Acta Part B: Atomic
Spectroscopy, vol. 112, pp. 10-15, DOI: 10.1016/j.sab.2015.07.009, 2015
9. Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M.;
Prettenhofer, P.; Weiss, R.; Dubourg, V.; Vanderplas, J.; Passos, A.; Cournapeau, D.; Brucher,
M.; Perrot, M.; Duchesnay, “E. Scikit-learn: Machine Learning in Python”. Journal of Machine
Learning Research 2011, 12, 28252830, 2011
ResearchGate has not been able to resolve any citations for this publication.
Technical Report
Full-text available
Utilization resource efficiency potentials in construction-related recycling Processing, Characteristics of primary construction materials, Characteristics of recycled construction materials, Recycling matrix and required technologies, Scenarios for recycling
Article
Full-text available
This work investigates the ability of LIBS to produce quality spectra from small particles of concrete demolition waste using single-shot spectra collected in open air. The 2-8mmmaterials are rounded river gravel, green glass shards, and plastic flakes. Considered are focal length, air, moisture, laser energy, and laser incidence angle (LIA). The research methodology is an experimental study using the so-called observation depth (OD) and spectral abundance as quality indicators. The relation between ablation volume, breakdown threshold, optical signal strength, and OD is captured in a simplified model to provide a better understanding of the dependence of the spectra on the LIA and material positioning in the laser beam. A 100 mm lens provided a compromise between spectral abundance, level of air interference and achievable OD. The study indicates LIBS can yield good quality data, even in cases of up to 3 mm surface roughness. Surfacemoisture did raise the percentage of bad spectra from an average 4% to 18%, but overall LIBS is still capable of providing quality data under challenging conditions.
Article
The composition of concrete determines its resistance to various degradation mechanisms such as ingress of ions, carbonation or reinforcement corrosion. Knowledge of the composition of the hardened concrete is therefore helpful to assess the remaining service life of an existing structure or evaluate the damage observed during inspections. For example, for most existing concrete structures the type of cement originally used is not known and must therefore be determined afterwards. This paper presents a preliminary study on the application of laser-induced breakdown spectroscopy (LIBS) to identify the type of cement. For this purpose, ten different types of cement were investigated. For every type, three cement paste prisms were produced: (i) prisms dried, ground and pressed into tablets, (ii) prisms dried and (iii) prisms untreated. LIBS measurements were performed with a diode-pumped low energy laser (1064 nm, 3 mJ, 1.5 ns, 100 Hz) in combination with two compact spectrometers which cover the UV and NIR spectral range. A reduced subset of spectral features was used to build a classification model based on linear discriminant analysis. The results show that the classification of homogenized pressed cement powder samples provides a high accuracy, however, factors such as a different sample matrix and moisture content can affect the accuracy of the classification. The study demonstrates that LIBS is a promising tool to identify the type of cement.
Article
Every year throughout the world > 50 million vehicles reach the end of their life, producing millions of tons of automotive waste. The current strategies for the separation of the non-ferrous waste fraction, contain mainly aluminum, magnesium, zinc and copper alloys, involve high investment and operational costs, and pose environmental concerns. The European project SHREDDERSORT, in which our research group was actively involved, aimed to overcome this issue by developing a new dry sorting technology for the shredding of non-ferrous automotive wastes. This work represents one step of the complex SHREDDERSORT project, dedicated to the development of a strategy based on Laser Induced Breakdown Spectroscopy (LIBS) for the sorting of light alloys.
Article
In the recycling of polymer e-waste, there is a pressing need for rapid measurement technologies for the simple identification and classification of these materials. The goal of this work was to instantly identify e-waste polymers by laser-induced breakdown spectrometry (LIBS). The studied polymers were acrylonitrile-butadiene-styrene (ABS), polystyrene (PS), polyethylene (PE), polycarbonate (PC), polypropylene (PP), and polyamide (PA). Emission lines were selected for C (247), H (656), N (742 + 744 + 747), and O (777), as well as the molecular band of C2 (516), and the theoretical ratios of the emission lines and molecular band were utilized. Classification models, k-nearest neighbors (KNN) and soft independent modeling of class analogy (SIMCA), were used to rank the polymers. Both constructed models gave satisfactory results for the validation samples, with average accuracies of 98% for KNN and 92% for SIMCA. These results prove the predictive analytical capabilities of the LIBS technique for plastic identification and classification.
Article
In this paper we address the application of an Echelle spectrometer for metal sorting tasks using laser induced breakdown spectroscopy in the industrial environment. Echelle spectrometers guarantee the simultaneous detection of broad spectral intervals with sufficient spectral resolution, which is highly desirable for LIBS applications. Until today, this benefit was overshadowed by the reduced speed of detection which typically could not exceed 10 frames per second. In this paper we present a newly developed high speed classification setup for scrap metal sorting, equipped with an Echelle spectrometer, capable of classifying 25 samples per second using a single burst double pulse excitation, a fast EMCCD camera and an externally triggered, chopper based timing. Different multivariate discriminatory techniques (PCA, PLS-DA) are applied and compared to gain maximum use of the spectra created by this setup. A simultaneous discrimination of 10 classes was done with > 90% correctness for most classes and an analysis of possible sources of error unique for non-laboratory experiments is presented.
Article
Effective discrimination between different waste materials is of paramount importance for inline quality inspection of recycle concrete aggregates from demolished buildings. The moving targeted materials in the concrete waste stream are wood, PVC, gypsum block, glass, brick, steel rebar, aggregate and cement paste. For each material, up to three different types were considered, while thirty particles of each material were selected. Proposed is a reliable classification methodology based on integration of the LIBS spectral emissions in a fixed time window, starting from the deployment of the laser shot. PLS-DA (multi class) and the hybrid combination PCA–Adaboost (binary class) were investigated as efficient classifiers. In addition, mean centre and auto scaling approaches were compared for both classifiers. Using 72 training spectra and 18 test spectra per material, each averaged by ten shots, only PLS-DA achieved full discrimination, and the mean centre approach made it slightly more robust. Continuing with PLS-DA, the relation between data averaging and convergence to 0.3% average error was investigated using 9-fold cross-validations. Single-shot PLS-DA presented the highest challenge and most desirable methodology, which converged with 59 PC. The degree of success in practical testing will depend on the quality of the training set and the implications of the possibly remaining false positives.
Article
This paper proposes a novel analytical methodology for soil classification based on the use of laser-induced breakdown spectroscopy (LIBS) and chemometric techniques. In the proposed methodology, linear discriminant analysis (LDA) is employed to build a classification model on the basis of a reduced subset of spectral variables. For the purpose of variable selection, three techniques are considered, namely the successive projection algorithm (SPA), the genetic algorithm (GA), and a stepwise formulation (SW). The use of a data compression procedure in the wavelet domain is also proposed to reduce the computational workload involved in the variable selection process. The methodology is validated in a case study involving the classification of 149 Brazilian soil samples into three different orders (Argissolo, Latossolo and Nitossolo). For means of comparison, soft independent modelling of class analogy (SIMCA) models are also employed. The best discrimination of soil types was attained by SPA-LDA, which achieved an average classification rate of 90% in the validation set and 72% in cross-validation. Moreover, the proposed wavelet compression procedure was found to be of value by providing a 100-fold reduction in computational workload without significantly compromising the classification accuracy of the resulting models.
  • F Pedregosa
  • G Varoquaux
  • A Gramfort
  • V Michel
  • B Thirion
  • O Grisel
  • M Blondel
  • P Prettenhofer
  • R Weiss
  • V Dubourg
  • J Vanderplas
  • A Passos
  • D Cournapeau
  • M Brucher
  • M Perrot
  • Duchesnay
Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.; Dubourg, V.; Vanderplas, J.; Passos, A.; Cournapeau, D.; Brucher, M.; Perrot, M.; Duchesnay, "E. Scikit-learn: Machine Learning in Python". Journal of Machine Learning Research 2011, 12, 2825-2830, 2011