liberation characteristics assessment for copper …

8
Liberation Characteristics Assessment for Copper Component in PCB Comminution Product by Image Analysis Seungsoo Park, Seongmin Kim, Seongsoo Han, Boram Kim, Byeongwoo Kim, Yosep Han and Jaikoo Park + Department of Earth Resources and Environmental Engineering, Hanyang University, 222, Wangsimni-ro, Seongdong-gu, Seoul, 04763, Korea Cu is one of the most valuable metals found in waste printed circuit boards (PCBs). In order to recover the metals from the e-waste, PCB recycling process exploiting physical concentration techniques have widely been applied. Size reduction steps including crushing and grinding are done for liberation of Cu from the non-metallic materials such as glass ber and epoxy resin. Degree of liberation is an important factor by which liberation performance of comminution process is evaluated. The primary purpose of this study was to investigate liberation characteristics of Cu particles in comminuted PCB by using image analysis method. The analysis procedures include image processing such as noise reduction, edge detection and component discrimination so that a set of images of ground PCB particles provides quantitative information such as Cu grade distribution and degree of Cu liberation. The Cu liberation phenomenon was also discussed in terms of its disassociation behavior from non-metallic material and grindability dierence. [doi:10.2320/matertrans.M2018113] (Received April 3, 2018; Accepted June 13, 2018; Published July 27, 2018) Keywords: copper, degree of liberation, image analysis, printed circuit board, recycling 1. Introduction Various small electronic devices such as laptops and mobile phones are being continuously disposed in bulk. Printed circuit board assemblies (PCBA) coming from discarded goods are referred to as urban ores that contain a variety of valuable metals. 1,2) Since PCBA does not only consist of valuable metals but toxic metals as well, there have been many related studies conducted due to its environmental concern even though PCBA only accounts to 3% of the entire e-waste. 3,4) PCBA is an electronic module populated with electronic components on printed circuit board (PCB). 5) The PCB is mainly composed of glass-reinforced epoxy laminate sheets (FR-4) and high grade Cu. 6) The amount of Cu in one sheet of PCB has the highest economic value only after that of gold in one PCB assembly module. 7) As a consequence, several studies have been made to recover Cu from PCBs by employing dierent techniques. The most dominant method for metal recovery from PCB is physical separation subsequent to size reduction through crushing and grinding. Such separation techniques include pneumatic separation, 8-10) magnetic separation, 11,12) eddy current separation, 13,14) electrostatic separation, 15-17) otation, 18-20) etc. In most cases, it is common to combine more than two separation techniques in order to acquire interest materials. The performance of Cu concentration process is represented by separation eciency, which is calculated by the subtraction of the recovery of FR-4 (recovery of gangue, R g ) from the recovery of Cu (recovery of valuable metal, R v ) into the concentrate. 21) The separation eciency relies on the liberation of Cu in PCB during the grinding process. Liberation refers to a phenomenon of free particles generation liberated from locked particles through comminution. 22) As emphasized in many previous studies, it is reasonable to say that the Cu grade distribution assessment and degree of liberation evaluation are signicant for estimating and improving the concentration performance of physical separation. 23,24) Degree of liberation is dened as a fraction of a phase which exists as liberated particles (or free particle). In a number of studies, PCBs were ground to the size where the degree of liberation was greatest in order to maximize separation eciency. The liberation size range varied from <100 μm, 25-27) >100 μm 10,28,29) to over a millimeter 30,31) according to the PCB characteristics and grinding mechanism. Among these studies which have analyzed the degree of liberation, most have done the calculation by dividing the number of free particles from that of entire particles. This is time-consuming and somewhat inaccurate because the analyzer should observe particles manually which is prone to human error. For natural ores, on the other hand, faster and more precise degree of liberation evaluation techniques have been facilitated which include SEM-based liberation analyzer 32,33) or optical imaging analysis algorithm. Optical image analysis, a simple and fast liberation assessment tool, has been adopted for the quantication of mineral liberation analysis in many studies. The grade of interest material is described with a fraction of the material coming from the concerned material which is estimated from information of spatial dimension (e.g. area, line and point, which correspond to two, one and zero dimension, respectively). 34) In relation to this, a large volume of research had been conducted and several representative methods include: 1) estimation of the liberation by measuring the linear intercept length of the target mineral using an optical microscope, 35,36) 2) measurement of the area ratio by using gray level dierence of each mineral with a module-type optical microscope 37) and 3) calculation of liberation degree after discriminating the minerals composition and compo- nents using gray level from the image of mined minerals taken from an optical microscope. 38) The purpose of this paper is to discuss the liberation phenomenon of Cu particles in PCB comminution nes. Firstly, the image processing sequence that we used for Cu + Corresponding author, E-mail: jkpark@hanyang.ac.kr Materials Transactions, Vol. 59, No. 9 (2018) pp. 1493 to 1500 © 2018 The Japan Institute of Metals and Materials

Upload: others

Post on 15-Nov-2021

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Liberation Characteristics Assessment for Copper …

Liberation Characteristics Assessment for Copper Componentin PCB Comminution Product by Image Analysis

Seungsoo Park, Seongmin Kim, Seongsoo Han, Boram Kim,Byeongwoo Kim, Yosep Han and Jaikoo Park+

Department of Earth Resources and Environmental Engineering, Hanyang University,222, Wangsimni-ro, Seongdong-gu, Seoul, 04763, Korea

Cu is one of the most valuable metals found in waste printed circuit boards (PCBs). In order to recover the metals from the e-waste, PCBrecycling process exploiting physical concentration techniques have widely been applied. Size reduction steps including crushing and grindingare done for liberation of Cu from the non-metallic materials such as glass fiber and epoxy resin. Degree of liberation is an important factor bywhich liberation performance of comminution process is evaluated. The primary purpose of this study was to investigate liberationcharacteristics of Cu particles in comminuted PCB by using image analysis method. The analysis procedures include image processing such asnoise reduction, edge detection and component discrimination so that a set of images of ground PCB particles provides quantitative informationsuch as Cu grade distribution and degree of Cu liberation. The Cu liberation phenomenon was also discussed in terms of its disassociationbehavior from non-metallic material and grindability difference. [doi:10.2320/matertrans.M2018113]

(Received April 3, 2018; Accepted June 13, 2018; Published July 27, 2018)

Keywords: copper, degree of liberation, image analysis, printed circuit board, recycling

1. Introduction

Various small electronic devices such as laptops andmobile phones are being continuously disposed in bulk.Printed circuit board assemblies (PCBA) coming fromdiscarded goods are referred to as urban ores that containa variety of valuable metals.1,2) Since PCBA does not onlyconsist of valuable metals but toxic metals as well, there havebeen many related studies conducted due to its environmentalconcern even though PCBA only accounts to 3% of the entiree-waste.3,4) PCBA is an electronic module populated withelectronic components on printed circuit board (PCB).5) ThePCB is mainly composed of glass-reinforced epoxy laminatesheets (FR-4) and high grade Cu.6) The amount of Cu in onesheet of PCB has the highest economic value only after thatof gold in one PCB assembly module.7) As a consequence,several studies have been made to recover Cu from PCBs byemploying different techniques.

The most dominant method for metal recovery fromPCB is physical separation subsequent to size reductionthrough crushing and grinding. Such separation techniquesinclude pneumatic separation,8­10) magnetic separation,11,12)

eddy current separation,13,14) electrostatic separation,15­17)

flotation,18­20) etc. In most cases, it is common to combinemore than two separation techniques in order to acquireinterest materials. The performance of Cu concentrationprocess is represented by separation efficiency, which iscalculated by the subtraction of the recovery of FR-4(recovery of gangue, Rg) from the recovery of Cu (recoveryof valuable metal, Rv) into the concentrate.21) The separationefficiency relies on the liberation of Cu in PCB during thegrinding process. Liberation refers to a phenomenon of freeparticles generation liberated from locked particles throughcomminution.22) As emphasized in many previous studies, itis reasonable to say that the Cu grade distribution assessmentand degree of liberation evaluation are significant for

estimating and improving the concentration performance ofphysical separation.23,24)

Degree of liberation is defined as a fraction of a phasewhich exists as liberated particles (or free particle). In anumber of studies, PCBs were ground to the size wherethe degree of liberation was greatest in order to maximizeseparation efficiency. The liberation size range varied from<100 µm,25­27) >100 µm10,28,29) to over a millimeter 30,31)

according to the PCB characteristics and grinding mechanism.Among these studies which have analyzed the degree ofliberation, most have done the calculation by dividing thenumber of free particles from that of entire particles. Thisis time-consuming and somewhat inaccurate because theanalyzer should observe particles manually which is proneto human error. For natural ores, on the other hand, fasterand more precise degree of liberation evaluation techniqueshave been facilitated which include SEM-based liberationanalyzer32,33) or optical imaging analysis algorithm.

Optical image analysis, a simple and fast liberationassessment tool, has been adopted for the quantification ofmineral liberation analysis in many studies. The grade ofinterest material is described with a fraction of the materialcoming from the concerned material which is estimatedfrom information of spatial dimension (e.g. area, line andpoint, which correspond to two, one and zero dimension,respectively).34) In relation to this, a large volume of researchhad been conducted and several representative methodsinclude: 1) estimation of the liberation by measuring thelinear intercept length of the target mineral using an opticalmicroscope,35,36) 2) measurement of the area ratio by usinggray level difference of each mineral with a module-typeoptical microscope37) and 3) calculation of liberation degreeafter discriminating the mineral’s composition and compo-nents using gray level from the image of mined mineralstaken from an optical microscope.38)

The purpose of this paper is to discuss the liberationphenomenon of Cu particles in PCB comminution fines.Firstly, the image processing sequence that we used for Cu+Corresponding author, E-mail: [email protected]

Materials Transactions, Vol. 59, No. 9 (2018) pp. 1493 to 1500©2018 The Japan Institute of Metals and Materials

Page 2: Liberation Characteristics Assessment for Copper …

liberation assessment is described. Secondly, liberationcharacteristic of Cu in PCB is discussed based on theanalyzed data acquired from image processing data.

2. Method and Materials

Laptop PCBAs received from Korean recycling facilityhave been used as the sample materials. Electroniccomponents such as CPU, socket slots, etc. were detachedfrom the boards using an apparatus specifically designed todisassemble electronic components and bare PCBs.39)

The bare PCBs were ground into particles smaller than4mm using a cut crusher (TOP-10-CC, Topcrusher, Korea)and then further ground to particles smaller than 1mm with agrinder that uses shearing force as the main grindingmechanism (KNSP-5 SYS, KOEN, Korea). Using vibratingscreens, the particles were classified to the following sizeranges: 177­250 µm, 250­354 µm, 354­500 µm, and 500­707 µm. The comminution fines were immobilized in amixture of epoxy resin (KEM90 Resin, ATM, Germany)and hardener (KEM90 Hardener, ATM, Germany) on thepolytetra-fluoroethylene mounting cup at room temperature.Surfaces of the specimens were ground with sand papers(#320, #600, #1200), polished with diamond suspension(1, 3, 6 µm) and subjected to optical analysis system. Theabove stated sample preparation procedure is schematicallydemonstrated in Fig. 1. The number of particles analyzedaccording to particle size ranges are tabulated in Table 1.

For image acquisition, a digital optical microscope(DVM2500, Leica) was used with magnification of 50©and white LED as the light source. A light diffuser wasattached at the tip of the LED source to evenly illuminate theentire specimen and minimize any reflection. The image ofthe specimen was then converted into a digital signal throughan in-built digital sensor so that the final image can be stored.Resolution of the acquired image was 1600 © 1200 and thedepth of color accuracy was 8-bit for each of red, green andblue. The acquired image file was input to MATLABμ forthe digital data processing and analysis. The loaded image istreated as three-dimensional matrix (1,200 © 1,600 © 3). Bymanipulating the elements in the matrix, images with desiredinformation are acquired. A serial operation of imageprocessing is described in detail in following sections.

The degree of liberation, DL, is calculated as follows:

DL ¼ pðg ¼ 1ÞZ 1

0

g � pðgÞdgð1Þ

where, g is grade of a phase (0 ¯ g ¯ 1), p(g) is a probabilitydensity function of weight fraction of particles with grade gand p(g = 1) is weight fraction of liberated particles (i.e.,the grade is equal to 1).

Generally, DL increases as the particle size decreases.Strictly speaking, the fraction of phase, i.e. grade ofphase, should be represented as weight fraction. However,volumetric, areal and linear fractions are also accepted

Fig. 1 Flowchart of specimen preparation.

Table 1 Analyzed conditions according to each particle size ranges.

S. Park et al.1494

Page 3: Liberation Characteristics Assessment for Copper …

as apparent degree of liberation. In this study, the degreeof liberation was calculated based on two dimensionalcalculations.

3. Results and Discussion

3.1 Microscopic image of comminuted PCBFigure 2(a) shows the microscopic photograph of ground

PCB specimen acquired from the size range of 355­500 µm.Three main components such as Cu, FR-4 and etch resistwere found in the image. The Cu shows two different colorsof orange and dark brown, depending on its laying position.Etch resist, FR-4 and mounting epoxy resin reflect green,ivory and gray colors, respectively. In a number of previousresearches, phase discrimination was conducted by means ofintensity-based clustering. In order to segment each phasefrom the image, the gray level distribution of the microscopicimage was acquired (Fig. 2(b)). The epoxy resin and Cu withdark color are distributed within distinguishable intensityrange from other components. However, intensity range ofthe other materials including Cu, FR-4 and etch resist,overlapped each other as their brightness is somewhat similar.

Color component-based segmentation exploits a set ofmore than two color information such as RGB (red, greenand blue) or HSI (hue, saturation and intensity) per pixel.This overcomes a disadvantage of non-uniqueness problemthat intensity-based phase discrimination method possiblyconfronts. Figure 3 shows HSI-model-based color separatedimage of Fig. 2(a). The HSI model is one of the mostwidely used color models for image segmentation. Hue is anumerical value that represents the color spectrum. Saturationrepresents the intensity of a hue from gray tone (saturationequals to 0) to pure color. Intensity is brightness of a color. Asshown in Fig. 3, each phases of the particles can be notablyseparated by their hue values. Combining both saturation andintensity information additionally, the phases of the PCB finescan be discriminated from each other. In order to elaboratephase segmentation, the HSI-component based discriminationwas carried out followed by image enhancing process.

3.2 Image processing3.2.1 Noise reduction

When the camera sensor captures the photons reflectedfrom the subject, noises are unavoidably generated. Theseunwanted signals may result in inaccurate particle identi-fication or phase discrimination results. The first step ofimage processing is to reduce these noise signals fromacquired image data in order to discriminate PCB particleimages. The most commonly used tool for noise reductionis median filter by which every pixel in original image isreplaced by the median value of neighboring reference pixels(also called as filter mask or kernel) with specific size.The filtered image is obtained according to the followingexpression:

f2ðx; yÞ ¼ medianðf1ði; jÞjði; jÞ 2 wxyÞ ð2Þwhere, f2ðx; yÞ is a filtered image, f1ði; jÞ is an originalimage and wxy is a filter mask with size of i © j at positionof ðx; yÞ.

Fig. 3 HSI histogram of PCB comminution fines image (Fig. 2(a)): (a) Hue histogram, (b) Saturation histogram, (c) Intensity histogram(C: Cu, F: FR-4, Et: etch resist, Ep: epoxy resin).

Fig. 2 (a) Microscopic image of PCB comminution fines; (b) Intensityhistogram of the image (C: Cu, F: FR-4, Et: etch resist, Ep: epoxy resin).

Liberation Characteristics Assessment for Copper Component in PCB Comminution Product by Image Analysis 1495

Page 4: Liberation Characteristics Assessment for Copper …

The size of filter mask has an effect on the magnitude ofnoise reduction. The larger the mask size is, the more thenoise signals vanish. The size of noise signal in the acquiredoriginal image was as small as less than 3 pixels. Using a5 © 5 filter mask, all undesired noises were effectivelyremoved from the acquired image (Fig. 4(b)). Additionally,the edges of particles became smooth after the operation tofacilitate more accurate edge detection of particle border.3.2.2 Edge detection

Edge detection for particle segmentation was conducted bythresholding. This method substitutes pixels in original imageby black pixels if the intensities of the objective pixels aresmaller than a decisive value (“threshold”) or by white pixels

when the intensities of the pixels are greater than that value.Thresholding operation can be expressed by the followingequation:

f2ðx; yÞ ¼0 ðfor If1 ðx; yÞ � T Þ1 ðelseÞ

�ð3Þ

where, f2ðx; yÞ is the binarized image, If1 ðx; yÞ is intensity(brightness) of original image and T is criterion constant forthe thresholding.

Since the background of the image has greater intensityvalue than particles, thresholding enables the segmentationbetween the background epoxy resin and other particlephases. Prior to the segmentation, extraction of particle

Fig. 4 Flowchart of image processing and analysis for liberation assessment: (a) original microscopic image of comminuted printed circuitboards, (b) image after noise reduction via spatial filtering, (c) background extracted from the microscopic image, (d) image afterbackground reduction, (e) particle shape acquisition by binarization with threshold, (f ) edge detected from the acquired particle, (g) Cucomposition measurement via image segmentation, (h) edge overlaid microscopic image, (i) information integrated final image ((g) and(h)).

S. Park et al.1496

Page 5: Liberation Characteristics Assessment for Copper …

images from surrounding epoxy resin was accomplished bymeans of additional median filtering operation. The size offilter mask was determined to be 1.5 times of the size ofthe largest particle in the image. As result of applying widefilter mask, plain background image without particles wasextracted from the original image (Fig. 4(c)). Subtracting thisbackground from original image yielded the pure particles(Fig. 4(d)) surrounded by white pixels. This enabled particlesegmentation by thresholding more accurate as backgroundintensity became equalized. The binarized image wasprocessed with further operations such as close/open methodand border removal in order to smooth the edge of particleand eliminate the particles within or crossing the boundary ofthe image (Fig. 4(e)). The border edge was acquired bydiluting the binary image and subtracting the eroded binaryimage from the expanded particle image (Fig. 4(f )). Not onlydo the particle shape and silhouette images (Fig. 4(e) andFig. 4(f ), respectively) function as verification images of thebinarization, but they also act as ROIs (region of interest) ofbinary images for subsequent phase segmentation steps.3.2.3 Phase discrimination

Consisting of different color combinations of HSI-basedcolor components, the phases of PCB could be discriminatedfrom each other by a sequential phase determinationprocedure as shown in Fig. 5. Every pixel of particleslocated in ROI underwent logical decision steps whether acolor component of pixel is greater or less than the specificvalue. Saturation was chosen to be the first color componentfor phase segmentation. Exclusive of epoxy resin, whichis surrounding of particles, dark Cu and FR-4 had lowersaturation and bright Cu and etch resistant had highersaturation. The criterion of saturation (S*) was chosen to bea value which laid on the bound of color componentdistributions of phases referred from saturation histogram(Fig. 3). Among the pixels with lower saturation than S*, thepixels with higher intensity were set to be pixels of FR-4 andthose with lower intensity were set to be pixels of dark Cu.

The criterion of intensity (I*) for the decision was determinedas of the set saturation. Pixels with higher saturation wereanalyzed according to their hue value and the value closeto that of red color (H*) were set to be pixels of Cu and theother were set to be pixels of etch resist. Once a pixel wasinvestigated as one of the PCB phases, the next pixel wasexamined until no pixel for a particle was remaining. Thecriterion values (H*, S* and I*) were calibrated until eachphases were properly mapped on particle images (Fig. 4(g)).Areal fraction of Cu for the particle was calculated fromnumber of Cu pixels divided by number of the particle’swhole pixels. After adjusting critical values for accuratephase segmentation (Fig. 4(g)) and particle discrimination(Fig. 4(h)), the areal fraction based Cu grade distribution wasobtained (Fig. 4(i)).

For the purpose of conversion of areal fraction gradedistribution to weight fraction grade distribution, followingcalculation was applied across entire particles.

gCu;w ¼ gCu;A � µCugCu;A � µCu þ gFR4;A � µFR4

ð4Þ

where, gCu,w is weight fraction of Cu in the particle, gCu,A isareal fraction of Cu in the particle, gFR4,A is areal fraction ofFR-4 in the particle, µCu is density of Cu and µFR4 is densityof FR-4. Integrating the mass fraction based grade values ofspecific particle size range, grade distribution and degree ofliberation were calculated by eq. (1).

3.3 Degree of liberation3.3.1 Cu grade distribution

Cumulative distribution of Cu grade according to differentparticle size range is shown in Fig. 6. Unlike the degree ofliberation, the distribution plot does not only provide theinformation regarding the amount of the free particles butthe locked particle as well. The slope between each intervalrepresents the weight fraction of given Cu grade class. Theslope of each particle size range is nearly constant within

Fig. 5 Component segmentation algorithm (H: hue, H*: hue criterion, S: saturation, S*: saturation criterion, I: intensity, I*: intensitycriterion, NV: pixel number of non-valuable matter, V: pixel number of valuable matter).

Liberation Characteristics Assessment for Copper Component in PCB Comminution Product by Image Analysis 1497

Page 6: Liberation Characteristics Assessment for Copper …

a whole range of Cu grade class. The Y value of the firstpoint represents liberated particle with Cu grade of 0. Thedifference between Y value of the last point and 1 depictsliberated particle with Cu grade of 100%. These two valuescan be denoted by L0 (i.e., p(g = 0)) and L1 (i.e., p(g = 1)),respectively. The weight fraction of locked particles wascalculated by subtracting the sum of L0 and L1 from 1.3.3.2 L0, L1 and fraction of locked particles

L0, L1 and weight percent of locked particles accordingto particle size range are shown in Fig. 7, which wasreconstituted from Fig. 6. Notwithstanding slight bounce ofL0 at coarse size range, both L0 and L1 increased as particlesize range decreased, as expected. Naturally, as the particlesize decreased, fraction of locked particle also decreased.Meanwhile, as the particle size decreased, L1 convergestowards a value somewhat less than 30mass%, which isbelow the average grade of Cu in PCB. This diminution in

increment of free particle generation is due to relatively lowgrindability of Cu to that of FR-4, which is discussed asfollows.

Comparison between liberation degree of Cu and FR-4according to particle size range is shown in Fig. 8. Thedegree of liberation of both components increased as particlesize decreased. The DL of FR-4 was greater than that of Cuwherein the DL of FR-4 was almost double the DL of Cu atthe size range of 500­707 µm. The difference between the DL

of two materials decreased as the particle size of the PCBdecreased. The degree of liberation of both materials shouldeventually be equal to 100% at a specific particle size.

The disagreement of DL of two materials can be explainedby difference of the materials’ ductility as well as theirrelative volumetric fraction. The two components of PCB -Cu as metal and FR-4 as a composite material of ceramic andplastic - have significantly different ground behaviors. Likemany other metals, Cu is ductile so it is more likely not to bepulverized compared to FR-4 which is much more brittle thanCu. That is to say, the grindability of FR-4 is much higherthan Cu so that selective grind effect takes place when PCBis comminuted. Meanwhile, since FR-4 has lower density(µ µ 1.85 g/cm3) than Cu (µ µ 8.96 g/cm3), volumetricfraction of FR-4 in PCB is higher than weight fraction ofFR-4. In other words, FR-4 dominates about 10 timesvolumetric fraction than Cu in PCB. Being the predominantmaterial of the board, FR-4 tends to exist as free particlecompared to the metal, apart from the difference of itsgrindability.

Due to these reasons, liberation degree of Cu is always lessthan that of FR-4 throughout all particle size until both of DL

approach to 100%. Based on these considerations, it may beproper to remove as many coarse FR-4 particles as possibleduring grinding and further Cu concentration process toavoid unnecessary overgrinding of non-metallic material.There should be awareness of the possibility of loss of finemetal particles when removing coarse FR-4 particles bymeans of classification because their terminal velocities in aclassifier such as cyclone or air table are similar to each other.

Fig. 8 Degree of liberation of Cu in PCB comminution fines according tothe particle size range.

Fig. 7 L0, L1 and weight fraction of locked particles according to theparticle size range of PCB comminution fines.

Fig. 6 Cu grade distribution according to size ranges.

S. Park et al.1498

Page 7: Liberation Characteristics Assessment for Copper …

3.3.3 Liberation behavior of CuThe disagreement of grindability and density of two

materials in PCB causes an interesting variation in density.Figure 9 shows the density distribution of PCB comminutionfines acquired by a cut crusher. Based on the densityvariation, liberation of the materials in PCB was observed totake place in four stages. The first stage (stage 1, D > D1) iswhere no liberation but only size reduction occurs. Resultingfrom absence of liberation of Cu or FR-4, there is novariation in average Cu grade to any particles in specific sizeranges so that the density of the particles remain constantuntil any component of PCB is disassociated. In the nextstage (stage 2, D1 > D > D2), FR-4 starts to be liberated andamong the free particles, the polymeric particles begin to bepulverized faster than the metallic particles; consequently,metal content in the particle increase and so does averagedensity. This increment continues until the particle sizeattains a specific size. As particles are ground further, averagedensity starts to decrease, due to increment of FR-4 contents(stage 3, D2 > D > D3). In the last stage (stage 4, D3 > D),the density stabilized at a value which is lower than theaverage density of stage 1. This indicates that non-metalliccomponents are concentrated at fine particles. This is thestage where the most Cu is liberated; thus, excessive grinding

below this size should be avoided. Microscopic images ofPCB comminution fines with the different particle sizes areshown in Fig. 10.

The stage transitions depend on particle size distributionand comminution mechanism. Under all circumstances,however, sum of product of densities and weight fractionsat each particle size equals to the density of uncomminutedPCB. If the particles are fine enough and all metals areliberated, stage 1 and/or even stage 2 may not appear andstage 3 and 4 might be the only steps. In that case, as particlesare ground finer, the particle size where the transition occurs(D3) possibly decreases. Reducing of the size at whichliberation starts to take place (D1) is the key factor forefficient physical concentration. Low-temperature pyrolysisof PCB,26,40) cryogenic treatment8,41) and microwaveinducing42) are some of the preprocesses which enables theCu to liberate in coarser size ranges. It should be noted thatliberation phenomenon alteration promoted by the priortreatments demand more detailed quantification in terms ofevaluation of the process performance.

4. Conclusion

The Cu liberation phenomenon of PCB comminution fineswas investigated by means of image analysis. The sequentialimage processing was utilized to examine the quantitativeassessment and liberation behavior of metal against FR-4.(1) Image processing techniques such as spatial filtering,

thresholding, HSI-based segmentation were appliedfor the image analysis. The analysis method coulddiscriminate comminuted PCB particles from each otherand calculate Cu grade of every single particle.

(2) Grade distribution of Cu was computed from the imageanalysis results. It was confirmed that liberation ofFR-4 occurs earlier than that of Cu. The reason forthis distinction is due to difference of grindabilitybetween FR-4 and Cu. In addition, their densitydifference results in variation of density according toparticle sizes.

(3) Liberation of Cu takes place in four stages. The averagegrade of Cu in a size range stabilizes while Cu is notliberated. Once Cu and FR-4 start to liberate, the Cugrade rises up to a value and then falls down to a valuebelow the average Cu grade of uncrushed PCB andremains stable.

Fig. 10 Microscopic images of PCB comminution fines with the different particle sizes.

Fig. 9 Density distribution according to particle size.

Liberation Characteristics Assessment for Copper Component in PCB Comminution Product by Image Analysis 1499

Page 8: Liberation Characteristics Assessment for Copper …

This unique behavior of Cu particle liberation phenome-non is originated from extreme gap between physicalproperties of two materials. Although these conditions aretoo extraordinary to be met in natural ores, the result of thisresearch can be suggestive for future selective grinding andliberation studies.

Acknowledgement

This work was supported by the Korea Institute of EnergyTechnology Evaluation and Planning (KETEP) and theKorean Ministry of Trade, Industry & Energy (MOTIE)(No. 20165020101200).

REFERENCES

1) B. Niu, Z. Chen and Z. Xu: J. Clean. Prod. 166 (2017) 512­518.2) P. Chancerel, M. Marwede, N.F. Nissen and K.-D. Lang: Resour.

Conserv. Recycling 98 (2015) 9­18.3) P. Hadi, M. Xu, C.S. Lin, C.W. Hui and G. McKay: J. Hazard Mater.

283 (2015) 234­243.4) C. Ning, P. Hadi, E. Aghdam, S. Zhu, D.C.-W. Hui, C.S.K. Lin and G.

McKay: Chem. Eng. J. 326 (2017) 594­602.5) H. Duan, K. Hou, J. Li and X. Zhu: J. Environ. Manage. 92 (2011)

392­399.6) M. Valix, Y.S. Loo, J. Bucknell, A.W.H. Cheung and Y. Hong:

Hydrometallurgy 173 (2017) 199­209.7) F. Cucchiella, I. D’Adamo, S.C. Lenny Koh and P. Rosa: Renew.

Sustain. Energy Rev. 51 (2015) 263­272.8) C. Zhou, Y. Pan, M. Lu and C. Yang: J. Hazard Mater. 311 (2016) 203­

209.9) M. Sarvar, M.M. Salarirad and M.A. Shabani: Waste Manag. 45 (2015)

246­257.10) C. Guo, H. Wang, W. Liang, J. Fu and X. Yi: Waste Manag. 31 (2011)

2161­2166.11) J. Bacher, A. Mrotzek and M. Wahlstrom: Waste Manag. 45 (2015)

235­245.12) F.P. Silvas, M.M. Correa, M.P. Caldas, V.T. de Moraes, D.C. Espinosa

and J.A. Tenorio: Waste Manag. 46 (2015) 503­510.13) Y.J. Park and D.J. Fray: J. Hazard Mater. 164 (2009) 1152­1158.14) J. Li, Y. Jiang and Z. Xu: J. Clean. Prod. 141 (2017) 1316­1323.15) H.M. Veit, T.R. Diehl, A.P. Salami, J.S. Rodrigues, A.M. Bernardes and

J.A.S. Tenório: Waste Manag. 25 (2005) 67­74.16) J. Wu, J. Li and Z. Xu: Environ. Sci. Technol. 42 (2008) 5272­

5276.

17) H. Lu, J. Li, J. Guo and Z. Xu: J. Mater. Process. Technol. 197 (2008)101­108.

18) I.O. Ogunniyi and M.K.G. Vermaak: Miner. Eng. 22 (2009) 378­385.19) R.H. Estrada-Ruiz, R. Flores-Campos, H.A. Gamez-Altamirano and

E.J. Velarde-Sanchez: J. Hazard Mater. 311 (2016) 91­99.20) J. He and C. Duan: Waste Manag. 60 (2017) 618­628.21) B.A. Wills and J. Finch: Wills’ mineral processing technology: an

introduction to the practical aspects of ore treatment and mineralrecovery, Butterworth-Heinemann, 2015.

22) R.P. King: Modeling and simulation of mineral processing systems,Elsevier, 2012.

23) C.L. Duan, Z.J. Diao, Y.M. Zhao and W. Huang: Miner. Eng. 70 (2015)170­177.

24) Y. Fan, E. Shibata, A. Iizuka and T. Nakamura: Metall. Mater. Trans. B47 (2016) 2754­2760.

25) I.O. Ogunniyi, M.K.G. Vermaak and D.R. Groot: Waste Manag. 29(2009) 2140­2146.

26) G. Jie, L. Ying-Shun and L. Mai-Xi: J. Anal. Appl. Pyrolysis 83 (2008)185­189.

27) A. Das, A. Vidyadhar and S.P. Mehrotra: Resour. Conserv. Recycling53 (2009) 464­469.

28) A. Vidyadhar and A. Das: Separ. Purif. Tech. 118 (2013) 305­312.29) C. Duan, X. Wen, C. Shi, Y. Zhao, B. Wen and Y. He: J. Hazard Mater.

166 (2009) 478­482.30) S. Zhang and E. Forssberg: Powder Technol. 105 (1999) 295­301.31) S. Zhang and E. Forssberg: Resour. Conserv. Recycling 21 (1997) 247­

269.32) R. Fandrich, Y. Gu, D. Burrows and K. Moeller: Int. J. Miner. Process.

84 (2007) 310­320.33) Y. Gu: J. Miner. Mater. Charact. Eng. 2 (2003) 33.34) J. Park: Doctoral Thesis, The University of Tokyo, (1988).35) R. King: Int. J. Miner. Process. 6 (1979) 207­220.36) C.L. Lin, J.D. Miller, J.A. Herbst and K. Rajamani: Particle & Particle

Systems Characterization 4 (1987) 78­82.37) M. Bérubé, J. Marchand and B. Champagne: in Applied mineralogy:

proceedings of the Second International Congress on AppliedMineralogy in the Minerals Industry, Los Angeles, California,February 22­25, 1984, Metallurgical Society of Aime, 1985, p. 259.

38) E. Donskoi, S. Suthers, S. Fradd, J. Young, J. Campbell, T. Raynlynand J. Clout: Miner. Eng. 20 (2007) 461­471.

39) S. Park, S. Kim, Y. Han and J. Park: Int. J. Miner. Process. 144 (2015)11­15.

40) J. Li, H. Duan, K. Yu, L. Liu and S. Wang: Resour. Conserv. Recycling54 (2010) 810­815.

41) C.S. Tiwary, S. Kishore, R. Vasireddi, D.R. Mahapatra, P.M. Ajayanand K. Chattopadhyay: Mater. Today 20 (2017) 67­73.

42) J. Sun, W. Wang, Z. Liu and C. Ma: Ind. Eng. Chem. Res. 50 (2011)11763­11769.

S. Park et al.1500