Imaging spectroscopy based strategies for ceramic glass contaminants removal in glass recycling

Transcription

1 Waste Management 26 (2006) Imaging spectroscopy based strategies for ceramic glass contaminants removal in glass recycling Giuseppe Bonifazi *, Silvia Serranti Dipartimento di Ingegneria Chimica, dei Materiali, delle Materie Prime e Metallurgia, UniversitaÕ Degli Studi Di Roma La Sapienza, Via Eudossiana, Roma, Italy Accepted 16 June 2005 Available online 19 August 2005 Abstract The presence of ceramic glass contaminants in glass recycling plants reduces production quality and increases production costs. The problem of ceramic glass inspection is related to the fact that its detectable physical and pictorial properties are quite similar to those of glass. As a consequence, at the sorting plant scale, ceramic glass looks like normal glass and is detectable only by specialized personnel. In this paper an innovative approach for ceramic glass recognition, based on imaging spectroscopy, is proposed and investigated. In order to define suitable inspection strategies for the separation between useful (glass) and polluting (ceramic glass) materials, reference samples of glass and ceramic glass presenting different colors, thicknesses, shapes and manufacturing processes have been selected. Reflectance spectra have been obtained using two equipment covering the visible and near infrared wavelength ranges ( and nm). Results showed as recognition of glass and ceramic glass is possible using selected wavelength ratios, in both visible and near infrared fields. Ó 2005 Elsevier Ltd. All rights reserved. 1. Introduction In the glass recycling industry, the development and application of sorting strategies for the separation of contaminants from the recyclable glass fraction is of great relevance. Since the early 1970s, glass recycling has experienced successive phases of growth due to the introduction of innovative technologies that enabled the automation of sorting process and the increase of recycling rates. Mechanical selectors and magnetic/metal detectors in the 1980s and, finally, glass cullet color sorting in the 1990s, represent the main technological innovations in the glass recycling sector (Leverenz and Kreith, 2002). Glass fragments resulting from both industrial and differentiated urban waste collection processes present * Corresponding author. Tel.: ; fax: address: giuseppe.bonifazi@uniroma1.it (G. Bonifazi). several polluting materials ranging from ferrous, metallic, wood, paper, plastic and ceramic glass or glass-like contaminants. Such materials can negatively affect both the production process and the resulting final products. The reduction of the level of contamination is thus essential to the financial viability of the glass recycling process. Automatic on-line sorting of contaminants is usually performed according to the physical characteristics of the material. Ferrous and metallic contaminants are detected and discarded by means of magnetic sorters. Paper, plastic and other light material are sorted by means of vacuum suction (i.e., different densities) and vibrating screens (i.e., different particle size class distribution and different mass). Infrared sensors are sometimes used to detect opaque cullet within the dirty recyclable glass in order to separate ceramic/ceramic glass and stone fragments from glass cullet. No entirely effective and low cost solution has been found until now for ceramic glass on-line automatic X/$ - see front matter Ó 2005 Elsevier Ltd. All rights reserved. doi: /j.wasman

2 628 G. Bonifazi, S. Serranti / Waste Management 26 (2006) Fig. 1. Glass fragments (cullets) characteristics are usually investigated adopting a laser beam technology. An on off logic is applied. The heart of the sorting system is constituted by a line of emitting laser diodes and by a corresponding set of receivers. The cullets pass through. The system analyzes their characteristics, usually according to the detected energy (transparent or opaque object) and in some cases to the spectra (object color attributes). On the bases of detected and analyzed signals, actuators provide to control specific nozzles, blowing air, that allow to modify the trajectory, usually the waste material, according to its characteristics. sorting in recycling plants. X-ray sorting techniques have been recently proposed as a solution for ceramic glass identification (Jong and Dalmijn, 2002). Anyway, it must be considered that the use of X-ray equipment in a plant requires appropriate shielding and must follow severe rules to protect workers from exposure, with an obvious increase in costs and environmental and safety problems. Glass sorting, originally based on analog devices, utilizing laser beam technology (Fig. 1), moved towards digital image techniques (Bonifazi, 2000; Bonifazi and Massacci, 2000). Scan line color cameras are thus more and more utilized in this sector to implement selection strategies addressed to identify opaque objects inside the flow stream and/or to separate cullets according to their color (Bonifazi and Massacci, 1998). Technology in this field, even if very sophisticated, remains practically blind versus the possible identification of ceramic glass materials. Spectrophotometers should be able, at least in principle, to identify these contaminants, however they are usually only able to work on a pointby-point basis and furthermore are not able to cope with real time sampling/sorting architectures as those required in glass recycling plants. They are used, in several industrial fields, mainly at the laboratory scale. The possibility to recognize glass and ceramic glass fragments by spectroscopic techniques has thus been investigated in this paper. In particular, a new hyperspectral imaging based architecture has been utilised. Properties of cullets are acquired as collection of spectra, one for each investigated pixel constituting the cullets itself, stored and handled as images. Different spectral ranges, in the visible and in the near infrared field, have been investigated. Originally utilized to optimize cullet sorting on the basis of color (Bonifazi et al., 2000), the proposed approach was then tested to recognize Pyrexe fragments inside of a cullet stream (Bonifazi and Massacci, 2001) and is currently the object of investigation in a European project (CRAFT HISPIM-GLASS Development of a novel and high-speed spectral imaging system to detect glass-like contaminants in the recyclable, cost effectively, increasing glass recycling and avoiding landfilling ). Once the possibility to recognize the two materials is verified, the further step will be the development and implementation of new sorting machines embedding an imaging based spectrometry logic able to realize an automatic on-line sorting system. 2. Glass recycling and ceramic glass contaminants The presence of ceramic glass inside of waste glass products rapidly increased in the last years, due to the introduction on the market of a large amount of ceramic glass manufactured goods, such as dishware, cookware, etc. (Höland and Beall, 2002). Such a material, even if quite similar, at least for our senses, to classic glass, is characterized by a different behavior (i.e., higher fusion point) when melted inside glass furnaces, where cullets are usually fed together with natural raw materials (quartz sands) (Pannhorst, 1997). As a consequence, the presence of ceramic glass reduces the production rates of the furnace, which needs to be shut down to be cleaned more frequently, and sometimes causes damages that require the furnace to be rebuilt or replaced. Another important effect is the damage to the glassproducing machines. Sometimes a bigger piece of glass ceramic can destroy the watercooled scissors, used for cutting the molten glass when bottles are produced. Such damage is very harmful as the scissors have to be changed and it is very dangerous because sometimes the whole machine can ignite. While producing flatglass, ceramic glass impurities can cause broken panes, and bigger glass-fragments can fall down often damaging the production line. Both effects, the damage of the scissors and the damage of the production line, are very expensive to rectify. The last damage concerns the finished products because the presence of ceramic glass affects the quality of the formed glass product (bottles, jas, etc.) that sometimes breaks during the bottling or handling process or can present some defects (Fig. 2). Actually the only two strategies extensively utilised and addressed to reduce the presence of ceramic glass

3 G. Bonifazi, S. Serranti / Waste Management 26 (2006) Fig. 2. Examples of damaged bottles due to ceramic glass contamination (courtesy of Reiling Glass Recycling GmbH, Marienfeld, Germany). digitally capture and handle spectra, as an image sequence, as they result along a pre-defined alignment on a surface sample properly energized. According to the different wavelength of the source and the different spectral sensitivity of the device, different physicalchemical superficial characteristics of the sample can be investigated and analyzed. The imaging based spectroscopy strategy adopted in this study is based on the utilization of two different devices belonging to the ImSpectore series spectrometer, developed by SpecIme Oy Equipment Fig. 3. Manual sorting of ceramic glass contaminants in a glass recycling plant (courtesy of Reiling Glass Recycling GmbH, Marienfeld, Germany). contaminants are source reduction and manual sorting. The problem with reduction at the source is that usually citizens, in spite of public education campaigns, confuse transparent-glass-like contaminants with normal glass, invalidating both curbside and door-by-door collection. As a consequence, some steps of the sorting process are still carried out manually by trained personnel (Fig. 3) that try to recognize ceramic glass fragments prior to crushing, looking at their shape or evaluating their reflective characteristics. Such an approach is expensive, not reliable, and represents a real and important problem for the whole glass recycling sector. 3. Digital imaging based spectroscopy Imaging spectroscopy is based on the utilization of an integrated hardware and software architecture able to The spectrograph (ImSpectore) is constituted by optics based on volume-type holographic transmission grating (Hyvarinen et al., 1998). The grating is used in patented prism-grating-prism construction (PGP element) which can be shown to have high diffraction efficiency and good spectral linearity, and it is nearly free of geometrical aberrations due to the on-axis operation principle. A collimated light beam is dispersed at the PGP so that the central wavelength passes symmetrically through the grating and prisms (so that it stays at the optical axis) and the short and longer wavelengths are dispersed up and down compared to the central wavelength. This results in minimum deviation from the ideal on-axis condition Architecture set-up The detection system is constituted by four basic elements: the optic, the spectrograph, the CCD camera, the light or excitation source.

4 630 G. Bonifazi, S. Serranti / Waste Management 26 (2006) The availability of a detector constituted by a linear array of sensing elements, each one able to detect the spectral components of the corresponding investigated constituting domains of the object, could permit the measurement of the optical spectrum components and the spatial location of an object surface (Hyvarinen et al., 1998). The equipment and the related measuring procedures are based on such a principle. This target has been realized by coupling together a dispersive stationary spectrograph and a CCD detector. The resulting information is thus constituted by a digital image where each column represents the discrete spectrum values of the corresponding element of the sensitive linear array. Such an architecture allows, with a simple arrangement of the detection device ( scan line perpendicular to the moving direction of the objects), the equipment to realize a full and continuous control. Differently from other detection systems, always based on spectroscopy, such a system offers several advantages: (i) reduced measurement time, (ii) no scanning movement and (iii) simultaneous measurement over a line area Measurements Line spectrometer is, as previously described, based on the detection, along a pre-defined alignment, of the spectra related to each element identified on the alignment itself (Hyvarinen et al., 1998). The length (L i ) and the width (W i ) of the scene line imaged at a time is thus determined by the slit length (L s ) and width (W s ), by the lens focal length (f) and by the distance between the object of the investigations and lens (D): L i = L s D/f and W i = W s D/f. Spatial resolution along the image line is determined by the camera pixel size and point spread size of the optics. Illumination is a crucial part of the real time spectral imaging system. Usually experimental tests are needed to define the optimal lighting. Line lighting, as the energizing source with uniform spatial distribution, is the most efficient solution. A necessary preliminary when an imaging spectrometry data handling and processing has to be applied is represented by calibration. Usually, it is performed in three steps, that is: (i) spectral axis calibration with spectral lamps (i.e., Hg, Ne, Ar and Kr); (ii) dark image acquisition and storage and (iii) measurements and storage of white reference image. After the calibration phase: (i) the image spectra is acquired and (ii) the reflectance (R ci ) (at wavelengths i and spatial pixels c of interest) is computed R ci ¼½ðsampleÞ ci ðdarkþ ci Š=½ðwhiteÞ ci ðdarkþ ci Š. ð1þ Such a procedure enables the equipment to compensate the offset due to CCD dark current and separates the sample reflectance from the system response. 4. Experimental 4.1. Investigated sample set Two different sets of samples have been analysed: one constituted by different typologies of ceramic glass pollutants, the other one constituted by glass fragments. All of the samples have been collected directly in glass recycling plants and can be considered representative of the cullet usually processed, being characterized by different colors, thicknesses, manufacturing processes and shapes Ceramic glass fragments The samples selected for spectral analyses are representative, according to information from glass recycling operators, of the most common and most problematic ceramic glass types found in the recycling plants (Fig. 4 and Table 1). They have been manually removed during glass processing. Ceramic glass fragments have been grouped into three classes of color: clear (8 samples: N0, R, C1, C3, C4, C5, C6, C7), opaque white (2 samples: AF, N11) and amber (3 samples: CC, CH, C2) Glass fragments The selected container-glass samples are shown in Fig. 5 and their description is reported in Table 2. The cullet can be grouped in the three standard color categories: clear (5 samples: V4, V7, V8, V9, V10), green (4 samples: V2, V5, V11, V12) and amber (3 samples: V1, V3, V6) Laboratory set-up for spectra acquisition The spectral analyses have been carried out using two different pieces of equipment working in two different wavelength ranges (Table 3): A desktop spectral scanner equipped with an imaging based spectrometer ImSpectore model V10, working in the visible-near infrared spectral range ( nm), named the VIS NIR field. A desktop spectral scanner, equipped with an imagingbased spectrometer ImSpectore model N17, working in the near infrared spectral range ( nm), named the NIR field. The desktop spectral scanner is a complete application designed for desktop use to produce a 2D spectral image of a sample (Fig. 6). The system scans spectral line images sequentially over the sample through movement of a table, providing full spectral and spatial information for each image pixel of the sample. The system includes a light source to provide the right level of energization to the sample.

5 G. Bonifazi, S. Serranti / Waste Management 26 (2006) Fig. 4. Ceramic glass fragments utilized for the spectral analysis. Table 1 Reference ceramic glass fragments utilized to perform the spectral analyses Sample Description Color class Average thickness (mm) AF Flat, scattered Opaque white 3.90 CC Flat, scattered Amber 3.94 CH Flat, scattered Amber 4.00 N0 Flat, smooth surfaces Clear 3.80 N11 Flat, scattered Opaque white 3.92 R Flat, scattered Clear 3.98 C01 Flat, smooth surfaces Clear 3.98 C02 Flat, scattered Amber 4.89 C03 Flat, smooth surfaces Clear 4.11 C04 Flat, smooth surfaces Clear 4.42 C05 Flat, smooth surfaces Clear 5.92 C06 Flat, smooth surfaces Clear 5.80 C07 Flat, smooth surfaces Clear 3.11 All of the measures have been carried out in an open environment, where the influence of solar radiation and that related to possible artificial lighting was always present. Such a choice was adopted to reproduce the actual operative conditions of the equipment in a glass recycling plant. The reflectance spectral images have been acquired and processed using the Spectral Scanner software v. 2.0 developed by DV s.r.l. (2003) in order to determine the corresponding spectral profiles. During the image acquisition phase, the ceramic glass and glass samples have been placed on a standard white surface. For calculation of the reflectance spectrum, the spectral images of the samples have been normalized using two reference standards: a white one to set-up the maximum reflectance conditions, and a black one to define the no (zero) reflectance condition. Eq. (1) was then applied. Starting from the spectral image of each sample, the reflectance spectrum has been obtained by averaging the reflectance values of the pixels located inside a rectangular region of interest (ROI) selected for the analysis (Fig. 6). 5. Results and discussion 5.1. Spectral signatures in the VIS NIR field The reflectance spectra in the VIS NIR field ( nm) of ceramic glass and glass samples are reported in Figs. 7 and 8, respectively. Comparing the reflectance spectra of different ceramic glass fragments (Fig. 7), it appears that most of the curves, representative of the spectral signature, show a similar shape but different reflectance levels, indicating a common behavior at the different wavelengths, with the exception of amber fragments (CC, CH, C2) and of clear sample C1. In fact, the reflectance values increase from 400 up to about 720 nm and then rapidly decrease, presenting a valley with a minimum value at about 760 nm. After such value, the spectra are characterized by constant or slightly increasing values up to the end of the investigated field (1000 nm). On the contrary, the spectra of C1 and amber (CC, CH, C2) samples show opposite behavior, presenting a peak in the region nm. The lower (0.0 to about 0.35) reflectance level in the spectra of amber samples, than that ( ) of clear and opaque white samples, is dependent on the different colors of the fragments. Clear and opaque white samples do not present relevant differences in their reflectance spectra. Considering the reflectance profiles of glass fragments (cullets) (Fig. 8), it is possible to distinguish two regions in the spectra: the visible field ( nm) and the NIR field ( nm). In the visible field, the spectral behavior is influenced by the color of the analyzed cullet (clear, green

6 632 G. Bonifazi, S. Serranti / Waste Management 26 (2006) Fig. 5. (a) Glass fragments (cullets) selected for the spectral analysis. According to the geometry surface of the fragments, presence of threaded surfaces (b), cleavage (c) or irregularities (d), the detected spectral response changes, accordingly (b.1, c.1 and d.1). or amber). For example, the spectra of the four green cullets (V2, V5, V11 and V12) show a characteristic peak at nm, corresponding to the green visible field. The other samples (both clear and amber) show spectral profiles with more or less constant values in the range nm, with differences in the reflectance level strictly related to differences in reflection power.

7 G. Bonifazi, S. Serranti / Waste Management 26 (2006) Table 2 Reference glass fragments (cullets) utilized to perform the spectral analyses Sample Description Color class Average thickness (mm) V1 Bottom of bottle Amber 7.30 V2 Neck of bottle Green 3.18 V3 Bottom of bottle Amber 6.59 V4 Bottom of bottle Clear 8.54 V5 Bottle body Green 5.30 V6 Bottom of bottle Amber 5.80 V7 Bottom of bottle Clear 9.38 V8 Bottom of bottle Clear 5.32 V9 Neck of bottle Clear 3.64 V10 Bottom of bottle Clear 6.66 V11 Bottom of bottle Green 5.60 V12 Bottle body Green 3.18 With reference to the wavelength interval from 700 to 1000 nm, it can be noticed as the curves are characterized by similar shape for all samples. What is important to outline is that, all glass samples, independently from their color, show a characteristic peak in the range nm. After such value, the reflectance values decrease up to 1000 nm. Comparing the curves of Figs. 7 and 8, it appears that in most cases the two materials are characterized by a marked different spectral signature. From the color class point of view, clear glasses seem to be easily distinguished from the clear and opaque white ceramic glass samples. As previously described, the results are characterized by spectral curves presenting a different shape. Green glass, which cannot be compared with transparent ceramic glass of the same color as it probably does not exist in the market, can be distinguished from all of the analyzed ceramic glass samples as it present a different spectral signature. The amber samples seem to be the most difficult to recognize, as the spectral signatures present the same Table 3 Technical characteristics of the two desktop spectral scanners (DSS) DSS 1 DSS 2 Spectrometer type SPECIM V10 OEM-80 lm slit SPECIM N17 OEM-80 lm slit Spectral range nm nm Camera Panasonic CS8300 series Sensor Unlimited 128x128 Optic Carl Zeisss Jena Dor 9890 Tevidons 1.4/25 Electrophysics 25 mm f/1.4 macro lens Frame grabber Matrix Vision mv IMPACT Base Mutech MV bit Energizing system Visible IR Spectral scanner size Length: 70 cm; width: 53 cm; table height: 22 cm; column height: 70 cm Fig. 6. (a) Architecture set-up utilized to perform a progressive and continuous surface spectra acquisition and related spectral information as detected adopting the set-up based on the utilization of the desktop spectral scanner equipped with the ImSpectore spectrometer family devices. (b) Example of reconstructed image of scanned glass samples showing the rectangular ROI (region of interest) selected to compute the corresponding reflectance spectra (c).

8 634 G. Bonifazi, S. Serranti / Waste Management 26 (2006) Reflectance Wavelength (nm) C1 C2 C3 C4 C5 C6 C7 CC AF N0 N11 R CH Fig. 7. Reflectance spectra in the VIS NIR field ( nm) of ceramic glass samples Reflectance Wavelength (nm) V1 V2 V3 V4 V5 V6 V7 V8 V9 V10 V11 V12 Fig. 8. Reflectance spectra in the VIS NIR field ( nm) of glass samples. shape, with very small difference in terms of reflectance values. No direct correlation between sample thickness and spectral signature has been found Spectral signatures in the NIR field The results of the tests performed on ceramic glass fragments and glass cullet in the NIR field ( nm) are reported in Figs. 9 and 10, respectively. The region between 1650 and 1700 nm is affected by noise and, as a consequence, is not considered in the following discussion. Most of ceramic glass samples are characterized by a similar spectral signature (Fig. 9). In fact, the reflectance values are constant or slightly increasing from 1000 up to about 1360 nm, and then they slightly decrease up to the end of the investigated field (1650 nm). The exceptions are the two opaque white samples (AF and N11), presenting a valley with a minimum value at about 1220 nm, and the amber sample CC that is not reflecting at all. The reflectance levels are influenced by the fragment color: they are lower (up to 0.55) for amber samples (CH, CC and C2), intermediate ( ) for opaque white samples (AF and N11) and higher ( ) for clear fragments. The clear color class shows a spectral signature that is particularly homogeneous. The analysis of the plot related to glass fragments samples spectra (Fig. 10) shows that the shape of the curves is similar for most of them, with the exception of clear sample V7, green sample V11 and amber samples (V1, V3 and V6). In fact the reflectance values are quite constant up to 1280 nm and after such value they slightly increase up to the end of the investigated field. Sample V7, on the contrary, presents a small peak in the region between 1000 and 1200 nm. The amber cullets

9 G. Bonifazi, S. Serranti / Waste Management 26 (2006) Reflectance R AF C5 C4 N11 CH C2 C6 C1 C7 C3 N CC Wavelength (nm) C1 C2 C3 C4 C5 C6 C7 CC AF N0 N11 R CH Fig. 9. Reflectance spectra in the NIR field ( nm) of ceramic glass samples. 1.0 Reflectance 0.8 V8 V7 0.6 V10 V9 0.4 V12 V4 0.2 V2 V5 V1 - V3 - V6 - V Wavelength (nm) V1 V2 V3 V4 V5 V6 V7 V8 V9 V10 V11 V12 Fig. 10. Reflectance spectra in the NIR field ( nm) of glass samples. (V1, V3 and V6) and the green V11 show a practically null flat spectrum, due to the absence of reflectance related to their dark color. Comparing the spectra of the clear and green cullet, they show a similar signatures shape, independently from their color, but different reflectance levels. In fact, color seems to affect the reflectance levels. Amber glasses (V1, V3 and V6), in fact, show reflectance values near zero, greenish glasses (V2, V5, V11 and V12) present reflectance values in the range of , whereas clear glasses (V7, V8, V9 and V10) display reflectance values from 0.5 to Clear sample V4 is characterized by a spectrum with a reflectance level ( ) more similar to the green samples due to its greenish component. The reflectance spectra seem not to be correlated with sample thickness, in agreement with the behavior in the VIS NIR field. Comparing the plots reported in Figs. 9 and 10, itis evident that ceramic glass samples and cullet (glass) are characterized by different spectral signatures in the NIR range, especially in the wavelength range between 1300 and 1650 nm. Glass spectra, in fact, after the peak corresponding more or less to k = 1350 nm, usually present higher reflectance values at the higher wavelengths, whereas the ceramic glass spectra, after the peak located in the same position, usually present decreasing reflectance values for increasing wavelengths Method for recognition of glass and ceramic glass fragments On the basis of the spectral analyses carried out, a simple and fast method for recognition of glass and ceramic glass fragments has been developed. In fact, for the implementation of an automated sorting system, the speed of data processing, due to the high feed rate, is an essential goal to fulfil, considering the short time allowed to perform sample recognition and its

10 636 G. Bonifazi, S. Serranti / Waste Management 26 (2006) eventual ejection. In this perspective, based on the analysis previously described of spectra shapes, three couples of reflectance values have been selected in the regions were the spectra of the two materials are more different, that is and nm in the VIS NIR field and nm in the NIR field. Their ratio has thus been computed, that is visible wavelength ratio 1 VW R1 = 500/680, visible wavelength ratio 2 VW R2 = 760/650 and near infrared wavelength ratio 1 NIRW R1 = 1360/1630. The results are graphically reported in Figs , respectively. From Fig. 11 it appears that the values assumed by VW R1 are <1 for ceramic glass and P1 for glass cullet, with the exception of two samples (opaque white N11 and amber V6). The values assumed by VW R2 (Fig. 12) are <1 for ceramic glass samples and >1 for glass cullet, with the exception of four ceramic glass samples (the amber ones, CC, CH, C2 and clear C1). Finally, in the NIR field, NIRW R1 (Fig. 13) is >1 for ceramic glass and <1 for glass cullet, with the exception of four samples (the amber CC, V1 and V6 and the green V11). Using all the three ratios, recognition of most of the ceramic glass and glass samples is performed, even if in some cases slight differences occurred (especially for VW R1 ). The correctness in classification is summarized in Table 4. From the table it appears that, without considering the amber color class, the three selected wavelength ratios produce only one error each. Moreover, a combination of two consecutive ratios can be utilized to refine the recognition. In fact, by selecting the VW R1 and then the VW R2, all of the samples are correctly identified. VW R1 500 nm/680 nm wavelength ratio C1 C2 C3 C4 C5 C6 C7 CC AF N0 N11 R CH V1 V2 V3 V4 V5 V6 V7 V8 V9 V10 V11 V12 Samples Fig. 11. Wavelength ratio VW R1 (500/680 nm) for all the analyzed samples (ceramic glass: white columns, glass: gray columns; errors: black columns). VW R2 760 nm/650 nm wavelength ratio C1 C2 C3 C4 C5 C6 C7 CC AF N0 N11 R CH V1 V2 V3 V4 V5 V6 V7 V8 V9 V10 V11 V12 Samples Fig. 12. Wavelength ratio VW R2 (760/650 nm) computed for all the analyzed samples (ceramic glass: white columns, glass: gray columns; errors: black columns). NIRW R nm/1630 nm wavelength ratio C1 C2 C3 C4 C5 C6 C7 CC AF N0 N10 R CH V1 V2 V3 V4 V5 V6 V7 V8 V9 V10 V11 V12 Samples Fig. 13. Wavelength ratio NIRW R1 (1360/1630 nm) computed for all the analyzed samples (ceramic glass: white columns, glass: gray columns, errors: black and absent columns).

11 G. Bonifazi, S. Serranti / Waste Management 26 (2006) Table 4 Errors in classification of ceramic glass and glass samples using the selected wavelength ratios (VW R1 = 500/680, VW R2 = 760/650 and NIRW R1 = 1360/1630 nm) (number of analyzed samples in brackets) Wavelength ratio (nm) Errors Ceramic glass (13) Glass (12) Clear (8) White (2) Amber (3) Clear (5) Green (4) Amber (3) 500/680 N11 V6 760/650 C1 C2, CC, CH 1360/1630 CC V11 V1, V6 6. Conclusions and further developments The possibility to apply a full digital imaging based spectrometry approach to recognize ceramic glass fragments inside a collected recycling flow stream was investigated. Different ceramic glass pollutants have been considered. Analyses have been carried out with reference to two possible detectable interval reflectance spectra, that is VIS NIR ( nm) and NIR ( nm). Results demonstrated that both of the approaches allow acquisition of useful information to utilize in the design and the implementation of innovative sorting logic and integrated hardware and software architecture that is aimed to eliminate the presence of ceramic glass contained in the final recycling glass product. More specifically, ceramic glass and glass are characterized by different reflectance spectral signature in the wavelength range nm. This is due to their different nature, influencing the spectral response of the two materials. Fig. 14. Architecture set-up of the automated on-line sorting system for ceramic glass contaminants (after CRAFT HISPIM-GLASS European project proposal).

12 638 G. Bonifazi, S. Serranti / Waste Management 26 (2006) With reference to the tests performed in the VIS NIR wavelength range, the results demonstrated that the best results are obtained for clear samples, whereas amber samples cannot be easily recognized, probably due to their low reflective characteristics. The fragment color affects the reflectance level for the entire spectral range and the spectrum shape in specific wavelength intervals, depending on the color itself. This is particular evident for green cullet. Concerning the results achieved in the NIR field, in most cases glass spectra are characterized by a typical shape, different from that of ceramic glass spectra, especially in the wavelength range of nm. The color of the sample affects the reflectance level of the spectrum but not the spectrum shape. In both cases, by selecting the right parameter representing the spectral signature, it is possible to recognize the two sample categories, glass and ceramic glass. To define an efficient sorting strategy, appropriate to the identification of optimal system requirements, the parameters related to specific wavelength ratios seem to best meet the objective for the intrinsic processing speed of the procedure. In particular, the use of a proper selected wavelength ratio could allow the system to recognize most of ceramic glass and glass samples in a fast way, simplifying the sorting architecture, thereby reducing equipment costs and allowing high production rates. The next step of the research will be the development of a prototype based on the imaging spectroscopy strategy. The proposed approach presents a big advantage of being characterised by a low industrial impact ; in fact it could be implemented minimising modification in currently adopted recognition-selection strategies for color glass fragments and opaque ceramic sorting. The final architecture set-up of the automated on-line sorting system is shown in Fig. 14. The system will be composed of: spectral imaging detection sub-system, including the spectrographic device (lens, PGP spectrograph, camera) capable of acquiring the spectral images of the cullet, the elaboration system (frame grabber, industrial PC) and an automatic spectrometer calibration device; air-jet sorting sub-system, with similar air-jet system developed to separate opaque ceramic contaminants from glass, or glasses of different color. The spectral imaging detection sub-system will be used to detect the position of the cullet and, in real time, command the air-jet sorting sub-system to blow the cullet from its normal falling line. As the detection system practically acts as a camera, its integration with compressed air based actuators should be relatively easy. Even if recognition is obtained for most of the samples at the laboratory scale, sometimes with slight differences between the two materials, especially considering the VW R1 ratio, obviously after the construction of a prototype working based on the proposed approach, the method should be tested using a wide set of samples in order to validate the results, to better tune the procedure and to calculate the statistical errors for recognition of both glass and ceramic glass fragments at the industrial scale. In any case, the investigated ceramic glass samples set belongs to families of products considered by the recycling industry to be the most harmful in terms of difficulty to be recognised and for this reason selected for this study. Finally, an improvement in the recognition of amber samples could be achieved by increasing the energizing source quantum efficiency (lamp power) in order to obtain better detectable reflectance spectra. Acknowledgments The work described was financially supported within the framework of the European Project CRAFT (Contract No. G1ST-CT ) Development of a Novel and High-Speed Spectral Imaging System to Detect Glass-Like Contaminants in the Recyclable, Cost Effectively, Increasing Glass Recycling and Avoiding Landfilling. A special thanks is given to Mr. Rohland Pohl (Reiling Glass Recycling GmbH, Marienfeld, Germany) for sample collection and for his valuable information concerning problems related with ceramic glass recognition and handling inside a glass recycling plant. References Bonifazi, G., Imaging based sorting logic in solid waste recycling. In: Proceedings of the 16th International Conference on Solid Waste Technology and Management, Philadelphia, PA, USA, vol. 6, pp Bonifazi, G., Massacci, P., Cullets (glass fragments) quality control by artificial vision: a color based approach. In: Proceedings of International Conference on Quality Control by Artificial Vision, Takamatsu, Japan, pp Bonifazi, G., Massacci, P., Cullets (glass fragments) quality control by artificial vision: a textural based approach. In: 4th World Congress R2000 Recovery, Recycling, Re-integration, Toronto, Ontario, Canada, CD-Paper 31, pp Bonifazi, G., Massacci, P., Classification of particulate solids materials by imaging based spectrometry. In: Proceedings of the Third International Conference on Intelligent Processing and Manufacturing of Materials, Richmond, BC, Canada. Bonifazi, G., Massacci, P., Pace, D., Real timing imaging spectrometry applied to bulk solids on the move: color fragments (cullets) selection for recycling. In: Proceedings of the 26th Conference Seminars Powder & Bulk Solids, Rosemont, Chicago, IL, USA, pp DV s.r.l., Spectral Scanner Software v. 2.0, Padova, Italy. Höland, W., Beall, G., Glass-Ceramic Technology. The American Ceramic Society, 372 pp.

13 G. Bonifazi, S. Serranti / Waste Management 26 (2006) Hyvarinen, T., Herrala, E., DallÕAva, A., Direct sight imaging spectrograph: a unique add-on component brings spectral imaging to industrial applications. In: SPIE Proceedings of Symposium on Electronic Imaging: Science and Technology, January, 1998, San Jose, California, vol. 3302, pp Jong, T.P.R., de Dalmijn, W.L., X-ray transmission imaging for process optimisation of solid resources. In: RÕ02, 6th World Congress on Integrated Resources Management, February 2002, Geneva, Switzerland, p Leverenz, H., Kreith, F., Markets and products for recycled material. In: Tchobanoglous, G., Kreith, F. (Eds.), Handbook of Solid Waste Management, second ed. McGraw-Hill Handbooks, pp Pannhorst, W., Glass ceramics: state of the art. Journal of Non- Crystalline Solids 219,