Resources, Conservation and Recycling 54 (2010) 571578

Contents lists available at ScienceDirect

Resources, Conservation and Recycling

journa l homepage: www.e lsev ier .com/

Autom y sthree-d

Shigeki KResearch Instit ial SciIbaraki 305-85

a r t i c l

Article history:Received 27 MReceived in reAccepted 21 O

Keywords:Automatic sorLightweight mApparent densThree-dimensMultivariate aEnd-of-life veh

ing ofem eensited arginges ofultivnd-ohe rece co

1. Introdu

The motor vehicle industry is paying a great deal of attention toimproving fuel efciency by reducing vehicle weight and therebyreducing emissions of greenhouse effect gases. As lightweightmet-als, such asto reduce vecic strengtsteel parts iautomotiveis now incretc.), in addused for enstrategic taamount of areach 250kaluminum)of aluminumminum recwrought anfor use in walloy, whichtively affect1983). As a

CorresponE-mail add

thecast aluminum is not speculated to increase comparedwith that forwrought aluminum in the future. On the other hand, magnesiumalloy parts have been introduced on a smaller scale than aluminumbecause of their low density (about 30% lower than aluminum).

0921-3449/$ doi:10.1016/j.aluminumandmagnesiumalloys, are suitablematerialshicleweight because of their low density and high spe-h, the substitution of light alloy parts for conventionals becoming more and more prevalent. With respect toapplications, the adoption of wrought aluminum partseasing for use in body panels (hoods, roofs, deck lids,ition to the fact that cast aluminum has been widelygine parts (Inaba, 2002; Sakurai, 2007). According to arget drawn up by the Japan Aluminum Association, theluminum parts per automobile in Japan is expected tog (130kg ofwrought aluminum, 120kg of cast or forgedin 2025 (Okubo, 2005). However, this rapid expansionparts in motor vehicles has caused anxiety in the alu-

ycling business: If scrap is recovered as a mixture ofd cast alloys, the recovered aluminum is not suitablerought alloy production, and its usage is limited to castallows for somewhat contaminant elements of nega-

ingmechanical andchemical properties (Ambroseet al.,consequence, this may lead to a supplydemand imbal-

ding author. Tel.: +81 29 861 8099; fax: +81 29 861 8457.ress: s-koyanaka@aist.go.jp (S. Koyanaka).

The use ofmagnesiumparts is expected to increase steadily againstthe background of the serious requirement for weight reduction inthe motor vehicle industry (Gesing et al., 2003). Currently, most ofthe magnesium scrap from end-of-life vehicles (ELV) seems to berecovered along with aluminum scrap, and recycled as an additivefor secondary aluminumalloy. Although this cascade recycling sys-temofmagnesiumscrap is effectiveprovided consumption is small,separation from aluminum scrap will become important so as torecycle it as secondary magnesium alloy when its consumptiongrows in the future.

In an ELV shredder facility, shredded lightweight metal frag-ments are concentrated in a mixture of nonferrous metals afterconventional separation processes such as magnetic separation,pneumatic separation, and eddy current separation. This metalmixture usually consists of copper, brass, zinc, lead, nonmagneticstainless steel, aluminumandmagnesium fragments, and thereforefurther separation is necessary to recycle each material. Althoughthese different metals are currently separated by manual sortingin most recycling facilities, several automatic separation or sortingtechniques are available for this purpose; color detection can beapplied for recovery of copper and brass, electromagnetic sensingcan be applied for recovery of nonmagnetic stainless steel (Mesinaet al., 2005), and X-ray transmission sensing and dense medium

see front matter 2009 Elsevier B.V. All rights reserved.resconrec.2009.10.014atic sorting of lightweight metal scrap bimensional shape

oyanaka , Kenichiro Kobayashiute for Environmental Management Technology, National Institute of Advanced Industr69, Japan

e i n f o

ay 2009vised form 20 October 2009ctober 2009

tingetal scrapityional shapenalysisicle

a b s t r a c t

A new method for the automatic sortcling of scrap metal. The sorting systto the differences in their apparent das width, height, volume, and projectare measured by means of a 3D imaassociated optics. The measured valudata-processing software that uses maluminum, and magnesium from an eput of the data-processing software. Thighly viable method that could repla

ction ance inlocate / resconrec

ensing apparent density and

ence and Technology (AIST), 16-1, Onogawa, Tsukuba,

lightweight metal scrap has been developed to aid the recy-nables separation of relatively large metal pieces accordingy and three-dimensional (3D) shape. Shape parameters suchea of irregular-shaped metal pieces moving along a conveyercamera system consisting of a linear laser and camera withthe weight and shape parameters are transferred to our ownariate analysis. Mixed fragments of cast aluminum, wroughtf-life vehicles shredder facility were sorted based on the out-sults show that the developed automatic sorting system is anventional dense medium separation and manual sorting.

2009 Elsevier B.V. All rights reserved.

secondary aluminum alloy market, because demand for

572 S. Koyanaka, K. Kobayashi / Resources, Conservation and Recycling 54 (2010) 571578

separation can be used in separation systems, based on the dif-ference in atomic number or density of materials. However, withregard to separation within the three types of lightweight metals,it is difcult to apply these techniques because of similarities inphysical properties (color, electrical conductivity, density, etc.). ADelft University of Technology group conducted a test using a dualenergy X-ray transmission (DE-XRT) sorting system, one of the lat-est technologies in the recycling industry, but the result was notgood enough to achieve separation between cast aluminum andwrought aluminum (Mesina et al., 2007). Other separation tech-niques based on laser-induced breakdown spectroscopy (Noll et al.,2008; Stepputat and Noll, 2003) or X-ray uorescence analysis arepotentially applicable, but their high sensitivity to contaminationon the surfaces of metal scrap and low processing speed becomeserious problems in an actual ELV shredder facility.

The purpose of this study is to develop a new automatic sortingtechnique that can overcome the above-mentioned problems, andwhich has high separation efciency and low processing cost. Inthis study,werst demonstrate a sorting systemthat combines twocutting-edge instruments: a three-dimensional (3D) imaging cam-era that enables various application developments by connectingit to a personal computer (PC); and, a weight meter that measuresthe weight of a moving object on a belt conveyer. This sorting tech-nique uses the weight information and the 3D shape parametersof an inspected piece of scrap, and an identication of the frag-ment is made by inputting these values into our own calculationprogram, which is based on multivariate analysis. In this paper, wepoint out tcast aluminthe equipmidenticatiolightweighttures of thetechniques,

2. Demand

Represechemical coalloy in JapAs seen in timpure elem

therefore it is difcult to recycle aluminum scrap into wrought alu-minum alloy. With regard to wrought aluminum scrap, only the3000 series used for the bodies of beverage cans and a part of the5000 series used for the lids of beverage cans are currently recy-cled as secondary wrought aluminum, all of the others are recycledas cast aluminum. In particular, ADC12 alloy which allows for arelatively high contamination of silicon, iron and copper is mostfrequently produced as secondary aluminum alloy. As mentionedin the previous section, a closed recycling system inwhichwroughtaluminumparts are recycled aswrought aluminumparts is the tar-get. For this purpose, it is necessary to separate cast aluminum andwrought aluminum scrap during their recycling process, in addi-tion to removing other contaminants. However, the differences inthe physical properties of these aluminum alloys are very small:The color tones of these alloys are basically the same. The densityand electrical conductivity of these alloys usually spans the range2.652.85g/cm3 and 3060% IACS, respectively. For suchmaterials,accurate separation cannot be expected using conventional colorseparation, dense medium separation or electromagnetic sensingseparation.

Currently, the weight of magnesium parts per automobile inJapan is about 1kg,which is less than 1/100 that of aluminumparts.As long as theweight ofmagnesiumparts in an ELV ismuch smallerthan that of aluminum parts, it is reasonable that magnesium scrapwill be processed as a part of aluminum scrap. The amount of post-consumer magnesium scrap in Japan is speculated to reach around5000 t in 2015 (NEDO, 2008). Around that year, if magnesium scrap

epare comcomt then ofic, beto septio

teria

mple

prerepar

Table 1Representativ

Series hemic

Wrought alu1000 (Al) +Si

S. Koyanaka, K. Kobayashi / Resources, Conservation and Recycling 54 (2010) 571578 573

ments used in this study.

ed sor

conventiontromagnetiIn fact, alulightweightin most ELV

Samplefragments oalloy scrapsfragments,other materandmanualwrought alFig. 1 showtheir weighwrought alparts, respedie-cast partion, it waswere corrodAlthough sothe experimnation.

3.2. Sorting

Fig. 2 shtem, whichIndustrial SLtd., IVC-3Dand a PC. TFig. 1. Examples of lightweight scrap metal frag

Fig. 2. Schematic diagram of the developal separation techniques such as color detection, elec-c sensing, dense medium separation, or manual sorting.minum scrap that contains only the three types ofmetals stated above is currently recovered successfullyshredder facilities.

materials for the sorting experiments were st-sizedf cast aluminum, wrought aluminum, and magnesiumgenerated at an ELV shredder facility in Japan. These

sized between 10 and 200mm and well separated fromials, were taken at random from the aluminum scraply sorted into threegroups: cast aluminum(104pieces),uminum (192 pieces), and magnesium (246 pieces).s photographs of some of these pieces of scrap. Fromt and shape, it is believed that cast aluminum and

uminum mainly originated from the engine and bodyctively, and that all of the magnesium fragments werets, such as steering columns and seat frames. In addi-observed that the surfaces of the magnesium piecesed, whereas those of the aluminum pieces were not.me fragments were covered with mud, oil, or paint,ents were conducted without removing the contami-

system

ows a schematic diagram of the developed sorting sys-consists of a belt conveyor, weight meter (Anritsu

olutions Co. Ltd., KW6205), 3D imaging camera (Sick Co.), air compressor, air cylinders, electromagnetic valves,he 3D imaging camera is equipped with a linear laser

and an optiby laser linline on thecal sensor dis constructhe reectehaving a sizlength), andheight) witshape param

Fig. 3. Measuveyor.ting system.cal CCD, and the height of each fragment is determinede triangulation. As shown in Fig. 3, the laser draws asurface of a moving fragment when the attached opti-etects it, and a digital 3D image of the entire fragmentted in the 3D imaging camera from the movement ofd ray. The developed system can process a fragmente of 2250mm in the horizontal direction (width anda size of 260mm in the vertical direction (maximum

h a resolution of 0.04mm. Several dimensions or 3Deters of the fragment, i.e. volume, vertically projected

rement of 3D shape parameter of a fragment moving on a belt con-

574 S. Koyanaka, K. Kobayashi / Resources, Conservation and Recycling 54 (2010) 571578

Table 2Parameters for data analysis.

X0: material X5: width X10: X3/(X4 X5)X1: weight/volume X6: maximum height Xn: X7/X6X2: volume X7: height of the center of

gravityX12: X2/X3/X7

X3: vertically projected area X8: X2/X3 X13: X4/X5X4: length X9: X2/(X4 X5) X14: X5/X4

area, length,width,maximumheight andheight of the center of thegravity, aremeasured by calculation using the digital 3D image. Thesphericity of the fragment, which is usually taken into account inconventional image analysis, cannot bemeasured using the currentdeveloped system. These data are immediately transferred to thePC and analyzed by our own data-processing software using mul-tivariate analysis. After identication, the fragment is sorted by aburst of compressed air in accordance with an output signal fromthe PC.

3.3. Algorithm for identication

The sorting method we developed is based on the concept ofmultivariate analysis: That is, an unknown fragment is identiedby an algorithm using the discriminant functions determined froma database of weight and 3D shapes of sampled fragments. Themixture to be separated should rst be separated or sorted byother means when applying this method. However, that is notcritical problem if the samples are restricted to lightweight met-als, because most of these fragments can be identied by manualsorting, although this requires considerable time.

As the parameters for data analysis, sample fragment materials(X0) and 14 variables (X1X14) listed in Table 2 were recorded inthe databasare a combmeasuremefor the sam

of the fragment on the belt conveyor was different. Especially fora fragment having a complex shape, a lot of case data are neededbecause the identication of this method is a process in which ameasured case is checked against preliminarily recorded case datain the database. After sufcient datawere recorded in the database,all case data were sorted and categorized into several sectionsaccording to the size of the apparent density X1. Then a multitudeof discriminant analyses were carried out in each apparent den-sity section with the explained variable of X0 and the explanatoryvariable of X1X14. The reason of this procedure is discussed later.

Fig. 4 shows a schematic illustration of the algorithm usedto identify whether a fragment is cast aluminum (Alc), wroughtaluminum (Alw), or magnesium (Mg). The cases of X1 >3.6 g/cm3

and X1

S. Koyanaka, K. Kobayashi / Resources, Conservation and Recycling 54 (2010) 571578 575

Fig. 5. Weight distributions of fragments.

should be noted that most of the sorting results were obtainedagainst unrecorded case data, because it was very difcult to placeirregular-shaped fragments on the conveyor in the exact same ori-entation (i.e. 0.04mm accuracy) as its former measurement.

When we consider mass processing, number of variables fordata analysis, number of case data to register in the database (nc),number of apparent density sections (ns), number of discriminantfunctions (nd) and setting of the threshold values for distinc-tion are very important factors. Actually, these numbers must bedecidedaccording to thepropertyof thegroupof fragments that aresorted. In this study, these numbers were set at nc = 5453 (cast alu-minum), 7865 (wrought aluminum), 11056 (magnesium), ns = 21and nd =641.

4. Results and discussion

Distributions of fragment weight and several 3D shape param-eters that were claried during the process of making the databaseare shown in Figs. 5 and 6. As shown in Fig. 5, the average weightincreases in the order of magnesium, wrought aluminum, and castaluminum, but it is obvious that these materials cannot be identi-ed by the weight of each individual fragment. Fig. 6(a)(d) wereobtainedby15measurementsper fragmentby changing its postureon the belt conveyor. Except for the fact that the 3D shape parame-ters of themagnesium fragments tend to be smaller than that of thealuminum fragments, we cannot nd any clear difference betweenthese fragments, and therefore they cannot be identied from theseindividual parameters. In Fig. 6(a), the distributions of volume X2of both aluminum fragments are almost the same, however, thisresult contains the meaningful error as mentioned below.

Fig. 7 shows the distribution of the apparent density X1 calcu-lated from the data of Figs. 5 and 6(a). This gure clearly showsthat the apparent density of eachmaterialmeasured by this systemtends tobe lower than its actual density (aluminum2.7,magnesium1.7 g/cm3). When the laser beam scans the surface of a fragment,some blind spots are inevitably generated at the back or insidethe fragment. The 3D imaging camera counted these spots as partof the fragment volume and thus overestimated the volume. Animportant point to emphasize is the fact that the degree of thismeasurement error is related to the fragment material. Namely,wrought aluminum scrap characterized by fragments of twistedthin and plate-like parts tends to carry a larger error and havelower apparent density, compared to cast aluminum scrap char-acterized by fragments of broken thick and bulky parts (see Fig. 8).It is clear that the trend in the degree of measurement error shownin Fig. 8 is irrelevant to the scale of the fragments, and thereforethe separation among aluminum fragments based on a differencein apparent density is likely to be achieved regardless of their size,Fig. 6. Distributions of several 3D shape parameters of the fragments.

576 S. Koyanaka, K. Kobayashi / Resources, Conservation and Recycling 54 (2010) 571578

Fig. 7. Distribution of the apparent density calculated from the data contained inFigs. 4 and 5(a).

as far as thetem. In theas all the mWrought mof the difcthe differenbecomes cldistributionfragment wdensity.

Fig. 9(a)sity X1 andIn these geters at 0.2Here we rearelation beter: Wrougabout 0.5