Phase Characterisation for Recycling of Shredded Waste Printed Circuit Boards

Abstract

In this study, we adopt a geometallurgical analytical approach common in mineral processing in the characterization of samples of shredded waste printed circuit board (PCB) E-waste, originating from Europe. Conventionally, bulk chemical analysis provides a value for E-waste; however, chemical analysis alone does not provide information on the textural variability, phase complexity, grain size, particle morphology, phase liberation and associations. To address this, we have integrated analysis using binocular microscopy, manual scanning electron microscopy, phase, textural and compositional analyses by automated (SEM-EDS), phase analysis based on (Automated Material Identification and Classification System (AMICS) software, and elemental analysis using micro-XRF. All methods used have strengths and limitations, but an integration of these analytical tools allows the detailed characterization of the texture and composition of the E-waste feeds, ahead of waste reprocessing. These data can then be used to aid the design of optimized processing circuits for the recovery of the key payable components, and assist in the commercial trading of e-scrap.

1. Introduction

1.1. Waste Printed Circuit Board Recycling

Current estimates suggest that in 2019, 53.6 million tonnes of electronic waste (E-waste) was generated, and this is projected to increase by 2030 to an annual waste stream of 74.7 million tonnes [1]. Disposal of this waste through incineration or landfill poses significant environmental risks [2], and is also not a sustainable use of the rare raw materials needed in the energy industry and manufacturing processes. Primary sources of raw materials derived through mining are increasingly having to focus on lower-grade, finer-grained and more complex ore deposits, as many of these mineral resources that are nearer to the ground surface and easier won have already been exploited. What is more, the development of new ore deposits involves both very long lead times and significant upfront revenue, making the recycling of E-waste an increasingly important and cost-effective resource.

In comparison with many primary ore deposits, E-waste has a much higher metal content and as such is a clear target for recycling. The challenge, however, is to be able to cost-effectively recover the component materials, in an environmentally sustainable way, from a waste feed material that is extremely heterogeneous and complex [1,3,4]. In terms of the metal content, the most valuable components of E-waste are waste printed circuit boards (PCB), which are estimated to make up anywhere between 2 and 7% of the total E-waste [1,3,5,6,7,8]. This range in estimates for the proportion of waste PCB within the E-waste reflects the potential heterogeneity of the initial waste stream, which would have a clear impact on the commercial viability of a recycling operation [4]. PCBs are themselves multi-component and are typically composed of approximately 30–50% metals and 50–70% non-metals (e.g., glasses, plastics and ceramics) [3,5,6,9]. The metal fraction is dominated by Cu, along with Al, Fe, Ni, Pb, Sn, Zn, and precious metals such as Au, Ag and PGM [3]. One method currently used for assessing PCB waste involves the following steps: (a) selection of a primary sample of the raw material (on the scale of tens of kilograms); (b) calcination of the crushed material in a muffle furnace; (c) crushing, milling to <0.2 mm, homogenization, and the preparation of a representative subsample for final analysis (less than one gram); and (d) chemical analysis (e.g., ICP-OES/MS) of the samples for Cu, Ni, Sn, and precious metals (Au, Ag, Pd, Pt). Such an approach determines the overall chemical attributes of the bulk sample. However, whilst providing quantitative chemical data, many of the attributes of the waste material are not determined.

Some authors have classified PCB wastes based on their Au content into low (100 ppm), medium (100–400 ppm) and high (>400 ppm) grades [1]. For comparison, the grade of Au expected in most current mineral exploration projects is typically <5 ppm (although tonnages are significantly greater), thus making even the “low grade” waste PCB of potential economic interest.

Numerous authors have provided detailed reviews of the current processes used for waste PCB recycling [1,2,3,5,7,8,9,10,11] along with a review of the large number of patents awarded regarding different stages in the process [12]. In most reviews, four main stages of the recycling process are recognized, as follows (Figure 1): (1) disassembly, most commonly achieved manually, to remove the hazardous components; (2) size reduction, including the shredding and grinding of the waste PCBs to both reduce the overall waste size and also to help liberate the different components; (3) separation, commonly a multi-step process, which may include physical separation based on, for example, density or magnetic properties, but also chemical separation (e.g., chemical degradation of epoxy resin, etching, leaching, oxidative breakdown of the organic matrix and complexation of non-metallic additives), followed by (4) metal separation using pyrometallurgy, hydrometallurgy, biohydrometallurgy and electrochemical processing [1,3,5,7,8,9,13,14,15,16,17,18,19,20,21] (Figure 1). Metal refining (i.e., the purification of recovered metals) typically follows the recovery stage, but is beyond the scope of the schematic in Figure 1. Whilst most papers focus on metal recycling, other authors have focused on the recycling of the non-metallic fraction [2] or the potential to recover metals from coarse PCBs without crushing/milling [6]. There is little, if any, published information on the phase characteristics of E-waste PCBs.

It is widely recognized that there are several significant challenges regarding the reprocessing of waste PCBs, including (a) the waste stream heterogeneity, diversity and complexity; (b) the variability of the waste feed with time; (c) the presence of glasses, plastics, ceramics and metals with complex and fine resolution microstructures; (d) the presence of a wide range of metallic elements in the single feed [3,5,8] and (e) taking a fully representative and unbiased sample for assaying. Whilst many elements are potentially “payable” and have a value in recovery, others are “penalty elements” deleterious to the value components or incurring a penalty cost if a processed material is to be sold on, for example, for smelting. In mining and mineral processing, geometallurgical analysis is an important step used to characterize the ore feed to aid the design of the mineral processing circuits [22] to maximize recovery of the valuable components. Several authors have noted the importance of characterizing waste PCBs ahead of analysis—“characterisation in terms of types, structure, components and composition is important to establish the route and process for eco-recycling” [5]; however, this step is commonly missing. Some authors have proposed characterization based on the chemical analysis of leachates following acid digestion [17], whilst other authors have suggested the use of methods including X-ray diffraction (XRD), microCT, scanning electron microscopy (SEM) with energy dispersive spectrometers (EDS), transmission electron microscopy (TEM) and chemical/structural methods such as FTIR, NMR, ICP and gas chromatography [1]. Geometallurgy in mining and mineral processing has advanced significantly as a result of improved analytical instrumentation, enabling a considerably improved understanding of elemental distribution in both 3D but also in textural context. Such robust characterization of the waste feed has been proposed in the analysis of recycled Li-ion battery “black mass” [23].

1.2. Aims

At present, PCB wastes are primarily characterized chemically; for instance, by taking a bulk sample, followed by calcining, crushing, milling and then analyzing a representative subsample using, for example, ICP-OES/MS. However, chemical methods do not provide information on the occurrence and distribution of the metals (such as size, shape, morphology, particle size, liberation, textural context and elemental deportment), all of which are important factors in the sampling, preparation, processing and handling of the material. For instance, size and liberation are critical parameters for the optimization of the comminution (milling) strategy (one of the most significant stages in terms of cost, energy, and time).

In this paper, we utilize modern analytical methods, as are widely used in mining/mineral processing geometallurgical investigations, to characterize a range of shredded waste PCB samples. Such characterization can address a range of potential queries, such as the following: (1) Is the shredded/crushed material derived from waste PCBs, and if so, what types? (2) What payable elements are present and what particle types within the waste are they deporting to? (3) What are the particle morphologies, types and microstructures present? (4) What is the degree of liberation of the individual components in different size fractions? This material characterization can then be used to aid in the design of the recycling processes, and can also be used to evaluate the recycling process efficiency. Our overall aim is to propose a rapid, fit for purpose, cost-effective, and practicable analytical approach for waste PCB characterization as an important step ahead of reprocessing.

2. Overview of E-Scrap Samples Analyzed

Nine shredded waste PCB samples were available for analysis, as summarized in

Table 1. Summary of samples and analyses. No. blocks = the number of replicate sub-samples analyzed using automated SEM-EDS; the number of replicate samples is controlled by the particle size of the material provided for analysis.

Sample	Description	No. Blocks
(A)	Oversized fraction from the hammer mill—light fraction from air sifter > 0.5 mm	4
(B)	Oversized fraction from the hammer mill—light fraction from air sifter < 0.5 mm	3
(C)	Magnetic fraction pre milled < 40 mm	6
(D)	Non-ferrous discard fraction from eddy current separator < 40 mm	6
(E)	Oversized fraction from hammer mill—heavy fraction from air sifter > 2 mm	4
(F)	Undersized fraction from hammer mill—heavy fraction from air sifter < 0.5 mm	3
(G)	Middle fraction from hammer mill—heavy fraction from air sifter 0.5 mm < 2 mm	3
(H)	Medium undersize, raw	2
(I)	1 mm milled	1

and Figure 2. Waste PCB commodities will be highly heterogeneous, and it is likely that the feed supplied to a recycling process plant will also vary with time. Consequently, the analytical results presented here will only apply to the small suite of samples available for analysis, but the aim of this contribution is instead to propose a methodology for waste PCB characterization ahead of recycling.

2.1. Provenance

The waste PCB samples analyzed in this study were sourced from a supplier in the United Kingdom. These had undergone processing by shredding before receipt, and some of the fractions had also been milled to present a range of samples in terms of particle sizes.

2.2. Sub-Sampling

Due to the variable textures and complexity of the shredded PCBs, it was not possible to take a sample that was provably unbiased and fully representative of the original mass. To align with the “Theory of Sampling”, each particle must have an equal opportunity of being selected during the process of sampling [24]. Milling the PCBs to produce a finer material increases the likelihood of drawing a representative sample; however, milling can cause liberation, the loss of malleable metals and the destruction of the original textures and particle morphologies. As such, the nine samples were analyzed, to give an indication of the different textures, grades and particle characteristics.

3. Analytical Methods and Sample Preparation

Four representative subsamples were taken from each sample and examined using (a) binocular microscopy, (b) manual scanning electron microscopy, (c) automated scanning electron microscopy with linked energy-dispersive spectrometers (SEM-EDS) using AMICS software, and (d) micro X-Ray Fluorescence (microXRF). The analytical methods are summarized here.

3.1. Binocular Microscopy and Manual SEM-EDS

The surface morphology of the particles in each sample was examined and imaged at a range of magnifications using binocular microscopy, and also using manual scanning electron microscopy. Scanning electron microscope (SEM) backscatter electron imaging (BSE) was carried out using either a Hitachi SU5000 field emission gun (FEG) or Hitachi SU3900 tungsten filament SEM. Only the finer-grained samples were imaged, but even then, some contained particles in excess of 3 mm, which required magnifications as low as x17 (field of view of approximately 7 mm). The SU3900 has a larger chamber size, which allows for the longer working distances required for very low-magnification analysis. Hence, the comparatively coarse samples were imaged with the SU3900 (field of view generally >2 mm) and finer samples imaged using the SU5000 (field of view generally <2 mm). All samples were prepared for analysis by sprinkling some of the powdered sample onto an adhesive carbon tag and carbon coating before imaging. An accelerating voltage of 20 kV was used for the lower-magnification images, whereas an accelerating voltage of 10 kV was used for the higher-magnification images. Similarly, the working distance used was dependent on the desired field of view, and therefore ranged from approximately 23 mm for coarse samples to 7 mm for finer samples.

3.2. Automated SEM-EDS

Automated SEM-EDS provides the rapid determination and quantification of the mineralogy/phase chemistry, particle size and shape of a variety of sample types. The technology was initially developed for use in the mining industry for the characterization of mineral processing products such as feeds, concentrates and tailings, but has also been used in the analysis of a wide range of materials and waste streams [25,26]. Data collection is operator-independent, with the acquisition of very large data sets; hence the results are statistically reliable and provide highly reproducible analyses. However, it should be noted that as the material particle size increases, the number of discrete particles analyzed will decrease, reducing the representativity of the modal analysis (see below).

Various automated SEM-EDS platforms exist (such as AMICS, QEMSCAN, and TIMA), but they all work on a broadly similar basis [27]. The analytical area is divided into a grid of higher-magnification fields; for each field, a BSE image is acquired and used to guide the placement of the electron beam; EDS spot chemical analyses are collected; and the resulting spectra are matched to a phase name. Signals are acquired very rapidly (typically milliseconds for each point), and this allows a high-resolution, large-area mineral or phase map to be efficiently generated.

For automated SEM-EDS, samples are usually mounted in 30 mm-diameter resin blocks and sectioned, polished and carbon-coated prior to analysis (Figure 3). Because of the coarse particle size of some of the samples, multiple replicate blocks were prepared, with the data for each sample combined. In addition, given the extremely platy nature of the majority of particles, and therefore the potential for bias through preferential alignment (and to a lesser extent, density settling), all samples were prepared as cross-sections. In total, 32 separate polished blocks were measured (Table 1). Therefore, whilst the binocular microscopy and manual SEM examination images the particle surface textures and particle morphologies, the automated SEM-EDS analysis characterizes the particles in cross-section. Analysis was undertaken with a Hitachi SU3900 scanning electron microscope fitted with dual large-area (60 mm²) Bruker silicon drift detectors (SDD) and energy-dispersive spectrometers, and running the AMICS automated mineralogy package. Beam conditions were optimized for analysis, and therefore an accelerating voltage of 20 kV coupled with a beam current of approximately 15 nA was used. The samples were measured using the segmented field image mode of analysis. Measurement parameters were optimized to the particle size and size of the features present (i.e., texture), and hence, given the comparatively coarse particle size, an effective image resolution of 3.1 µm was used for all samples.

The EDS spectra acquired during the measurement are compared with a library of measured and synthetic standards and a phase identification is made on a closest match basis. Phases that are not represented in the standards list at the time of measurement are added either by acquiring reference spectra directly from the sample, or by creating a reference spectrum from the measurement itself. As the standards list can comprise hundreds of reference spectra, the data are grouped into a final, more manageable, reported list of phases (Appendix A). During AMICS analysis, an SEM-BSE map of the area imaged is also collected.

3.3. Micro XRF

Micro X-ray fluorescence (micro-XRF) can provide information on elemental distribution and quantification, as well as potential phase or mineralogical analysis (AMICS). In addition, and importantly for these samples, micro-XRF analysis requires little or no sample preparation, which has led to the technique being used in a wide variety of applications, from geology to cultural heritage [28]. In this study, uncoated and unprepared sub-samples were dispersed as a layer approximately 1 cm deep in sample trays measuring approximately 10 cm × 8 cm (Figure 3B). The samples were scanned using a Bruker M6 Jetstream scanning micro-XRF.

To test optimal measurement parameters, the samples were scanned five times, with varying pixel sizes and analysis times per pixel (

Table 2. Summary of measurement parameters for micro-XRF analysis using the M6 Jetstream system for six of the PCB samples.

Measurement	Pixel Size (µm)	Pixel Dwell Time (ms)	Total No. Pixels Analyzed	Total Analysis Time	Analytical Time Per Sample
1	1000	10	68,400	32 min	5.3 min
2	500	10	273,600	97 min	16.2 min
3	500	5	273,600	66 min	11 min
4	200	30	1,710,000	948 min	158 min
5	200	2	1,710,000	164 min	27.3 min

). Decreasing the pixel size and increasing the analysis time per pixel accordingly yields a longer analysis time. As shown in Table 2, using a 1000 µm pixel size and an analysis time of 10 ms allowed sample measurement in less than 6 min; decreasing the pixel size to 200 µm with an analysis time of 2 ms allowed sample measurement in less than 28 min. It should be noted that micro-XRF scanning was carried out on all six samples at once. The area scanned included both the area of the individual samples, but also the area between the samples, so the estimated analytical time per sample is a slight over-estimation. The effects on the results when varying these parameters are described below.

4. Phase Characterization

4.1. Particle Morphology

The morphology of the particles within the different waste PCB samples examined is a function of their respective preparation prior to analysis. The morphology was examined through both (a) rapid imaging using binocular microscopy and (b) manual scanning electron microscopy of the finer size fraction samples. Binocular microscopy has the advantages of being very low-cost and rapid, although the quantification of particle attributes, such as size, shape, material liberation, etc., is difficult to achieve by visual analysis alone. As the waste PCB is reduced in size, there is an increase in the apparent liberation of the separate major components (Figure 4), although it should be noted that this only holds true for the larger discrete components. Where, for example, fine Cu layers (approximately 100 µm thick) are present as an integral component within the PCB, interlayered with both latitudinally and longitudinally arranged glass fiber layers, then binocular microscopy will allow the identification of the PCB sheets, but not the liberation of the integral components.

Particle shape, size and composition are, however, critical when considering how a waste particle will behave during further processing operations, such as density, magnetic or electrostatic separation. In addition, shredding/grinding require significant energy input, and as such increase processing costs [6]; particle analysis can aid in determining the size fraction required for the most cost-effective component’s liberation for the planned separation and metal recovery. It is also apparent from binocular microscopy that the original shape of soft and malleable metal components within the original PCBs is affected during shredding, such that commonly original metal sheets become complexly folded and plastically deformed, and their particle shapes are modified during processing.

Within the finer fractions, the samples appear dominated by the plastic and ceramic components of the PCBs, and binocular microscopy lacks detailed resolution. Manual scanning electron microscopy does, however, allow the particle shapes to be imaged. Typically, within the finer waste PCBs, numerous individual glass fibers from the boards are generated during shredding and grinding (Figure 5). As noted above, the metal components, due to the commonly malleable metals used, are often folded or rounded during processing (Figure 5). Consequently, during the size reduction stage of reprocessing, the component particles will change from their original shapes, potentially affecting how the particles behave during the separation processing stage. In addition, the commonly liberated fine glass fibers may, if not also recovered by processing [2], become a greater environmental risk.

4.2. Particle Types

Based on the automated SEM-EDS analysis, the particles making up the crushed/shredded PCB samples are classified in this study into 34 compositional groups (Appendix A); of these, 17 are present in some, or all, of the analyzed samples at an abundance of >1% by area, and the remaining particle types have an abundance in all of the analyzed samples of <1% by area. The particle types present at an abundance of >1% include (a) metals (Cu, Sn, Al, brass, Cr steel, Al oxides and Pb phases), (b) plastics (Br-bearing plastics and Sb-Br plastics), (c) glasses (silica, Ca glass, Ca Mg glass) and (d) other phases (Mg silicates, barite, Mn Fe oxide, Ca oxide/carbonate, Al Si chloride) (Appendix A). Phases present at a measured area abundance of <1% include (a) metals (Zn, Ni, Au, Ag phases), (b) plastics (PVC), (c) glasses (Na Ca glass) and (d) other phases (silicon, Fe silicate, Ti oxides, Fe oxides, Ba Ni oxides, Mg oxide/carbonate, Ti REE, Ca sulphate, sulphur and “undifferentiated”). Particle classifications following automated SEM-EDS analysis can be redefined, depending on the level of detail required and the specific project requirements, such that the compositional groups presented in Appendix A could be simplified to report the main metal phases, plastics, ceramics and “others”. Here we present a detailed particle classification, in order to illustrate the analytical capability, yet this could be further refined and simplified. The relative abundance of the identified phases is presented in

Table 3. Relative abundance of metal phases present in the waste PCB samples analyzed.

	A	B	C	D	E	F	G	H	I
Cu	25.59	10.89	10.65	21.53	17.34	52.47	34.00	3.67	4.99
Sn	1.51	1.00	1.50	1.08	2.84	13.12	15.90	0.46	0.67
Zn	0.00	0.08	0.06	0.00	0.00	0.03	0.50	0.00	0.74
Ni	0.01	0.05	0.01	0.00	0.02	0.13	0.09	0.02	0.01
Pb Phases	0.00	0.17	0.00	0.00	0.31	0.83	2.61	0.39	0.59
Gold	0.000	0.000	0.000	0.000	0.000	0.005	0.003	0.002	0.000
Ag Phases	0.01	0.13	0.00	0.00	0.00	0.04	0.13	0.00	0.00
Al	0.44	0.46	0.00	0.00	1.77	1.26	0.91	3.82	5.16
Brass	0.00	0.06	0.00	0.00	0.53	0.50	1.83	0.00	0.04
Cr Steel	0.00	0.12	18.03	0.00	0.72	0.00	0.00	0.00	0.48
Zn Oxide	0.13	0.02	0.00	0.00	0.00	0.04	0.03	0.00	0.03
Al Oxide	0.42	1.28	0.21	0.18	0.16	0.72	5.13	0.14	0.13

(metals) and

Table 4. Relative abundance of the non-metal phases present in the waste PCB samples analyzed.

	A	B	C	D	E	F	G	H	I
Silicon	0.00	0.34	0.00	0.00	0.21	0.01	0.00	0.01	0.02
Silica	4.76	6.68	0.30	4.48	10.55	4.66	7.61	1.50	7.35
Ca Glass	32.87	29.85	10.20	42.44	26.12	7.03	10.61	45.86	37.87
Ca Mg Glass	3.62	1.26	0.00	3.86	1.48	1.03	0.89	1.06	0.19
Na Ca Glass	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.03	0.12
Mg Silicates	3.45	2.04	1.20	1.54	0.71	1.96	1.48	0.03	0.04
Fe Silicate	0.00	0.00	0.00	0.00	0.00	0.03	0.60	0.00	0.00
Ti Oxides	0.04	0.10	0.00	0.02	0.01	0.05	0.12	0.01	0.24
Barite	1.45	5.32	0.60	2.93	0.76	2.76	0.30	2.30	1.92
Fe Oxide	0.00	0.10	0.01	0.00	0.00	0.14	0.62	0.01	0.72
Mn Fe Oxide	0.00	1.21	46.13	0.00	0.00	0.79	0.49	0.00	0.01
Ba Ni Oxide	0.00	0.56	0.20	0.00	0.00	0.17	0.73	0.01	0.09
Ca Oxide and Carbonate	0.01	0.07	0.67	0.04	0.02	0.02	1.07	0.07	0.56
Mg Oxide and Carbonate	0.01	0.05	0.00	0.12	0.02	0.01	0.02	0.40	0.05
Br Plastic	12.82	35.59	9.27	21.09	13.50	10.23	6.67	37.91	27.33
Sb- Br- Plastic	12.74	2.36	0.36	0.61	21.92	1.74	5.39	1.69	9.23
PVC	0.00	0.00	0.47	0.00	0.03	0.00	0.32	0.21	0.35
Al Si Chloride	0.00	0.00	0.00	0.00	0.00	0.04	1.01	0.01	0.03
Ti REE	0.00	0.00	0.00	0.00	0.00	0.00	0.63	0.00	0.01
Ca Sulfate	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.34
Sulfur	0.00	0.05	0.00	0.00	0.76	0.02	0.00	0.01	0.20
Undifferentiated	0.12	0.16	0.12	0.08	0.19	0.17	0.29	0.38	0.48

(non-metals) and also in Figure 6 and Figure 7.

There is a significant variation in the relative abundance of the phases present in the samples analyzed. However, it should be noted that in the coarser-grained samples, the reported relative area % abundance will be strongly influenced by the limited number of large particles analyzed, despite the preparation and analysis of multiple subsamples per sample (see Figure 2A). As particle size decreases, there is a corresponding increase in the number of individual particles analyzed and a corresponding improvement in data reproducibility and representivity. In addition, the samples were prepared as orientated polished sections, such that the imaged particles are cross-sections. This should minimize/mitigate any variance, especially for laminated samples such as these; however, the microstructure of the particle and its orientation within the polished section will influence the reported relative area percentage abundance.

The metal phases present in the samples analyzed comprise between 8.51% and 69.12% of the particle area measured, with an average abundance of 30.1%, which is directly comparable with previous estimates for metal abundance within waste PCBs [2,3,5], although, as discussed below, some plastic particles may not have been detected during the automated measurements, which implies that the actual metal abundance may be less than reported. Copper metal dominates the metal fraction in all except one of the analyzed samples (3.67% to 52.47%, average 20.1%), although Sn is also very abundant (13.12–15.90%) in samples F and G (Table 3). Cr steel particles dominate in Sample C; this sample was analyzed following magnetic separation. Typically, the order of metal abundance (excluding Cr steel) in the analyzed samples is Cu > Sn > Al > Al oxide > Pb phases > Brass > Zn > Ni > Ag phases > Au (Table 3, Figure 6). Whilst in part the relative abundance of the particles present will be controlled by the material size, and therefore, the number of particles analyzed, there is still considerable variation in the relative abundance of the different metals present in the analyzed samples (Figure 6).

The relative abundances of the non-metal phases present in the analyzed samples are shown in Table 4 and Figure 7; on average, 69.9% of the measured particles are non-metallic, consistent with previous studies [2,3,5]. The dominant particle types present are a range of different glass compositions (dominated by Ca glass, silica, Ca Mg glass and trace NaCa glass in sample H) and plastics (Br plastic, Sb-Br plastic and PVC) (Table 4). On average, 33.8% of the particles can be categorized into the glass compositional groups, with 25.8% plastics (Table 4). With automated SEM-EDS analysis, to identify a particle, there needs to be a clear chemical difference between the resin mounting medium used and the embedded particles. The AMICS particle images reveal that there are particles present in which rods/cylinders of glass are surrounded by apparent voids, although the SEM-BSE images show that these are discrete particles, in which the glass fibers are themselves originally mounted in resin, interpreted to be of a similar composition to the mounting medium used during sample preparation (Figure 8). In this preliminary study, these resins have not been quantified; future methodological developments would enable the use of a different composition sample mounting medium, such that the resin present in the PCB could be differentiated from the mounting media. For example, in previous work focusing on coal fly ash, samples were mounted in carnauba wax to enable the carbon particles to be differentiated from the mounting media [29]. Therefore, in the reported particle abundance data, the resins in the PCB have not been quantified, hence the data over-estimate the relative abundance of the reported phases. A range of other phases are identified, although typically at an abundance of 1% or less, and these include silicon, Mg silicates, Fe silicates, Ti oxides, barite, Fe oxide, MnFe oxide, Ca oxide/carbonate, Mg oxide/carbonate, AlSi chloride, Ti-REE, Ca sulphate and sulfur (Table 4). In total, on average, these phases make up less than 4% of the analyzed particles by area. Mn Fe oxides are the exception in Sample C, in which they make up over 46% of the analyzed particles. The BaNi oxide category also includes particles of BaTi oxides, used in PCB ceramic caps.

During automated SEM-EDS analysis, further datasets and images are generated, in addition to determining the modal phase abundance. In mining/mineral processing applications, typical datasets may include particle size data, association data (based on recording the number of transitions from one phase to another) and liberation data. Liberation data, for a particular compositional group, calculate the proportions of particles that are either solely composed of that phase (i.e., are fully liberated) or fall into a series of defined classes, based on the relative abundance of the phase in question compared to other phases, typically with classes defined as >90%, >80%, >70%, etc., of the specific compositional phase. As an example, an individual particle composed only of Cu metal would report to the 95–100% class; one composed of 50% Cu metal and 50% other phases would fall into the 50–55% class. Therefore, liberation data quantify how successful the separation of different phases is following mechanical fragmentation. These data can then inform aspects such as the grind size required to achieve phase liberation, but can also aid in the design of the optimal processing circuit. Examples of liberation data for copper in Samples A, B, E and F are provided in

Table 5. Cu liberation data for samples A, B, E and F based on the automated SEM-EDS particle analysis.

Cu Liberation	A	B	E	F
0–5%	0.2	0.1	1.7	0.1
5–10%	0.6	0.1	1.8	0.3
10–15%	3.8	0.3	1.5	0.3
15–20%	1.4	1.3	2.7	0.4
20–25%	4.3	0.4	1.5	1.0
25–30%	2.0	1.2	2.4	2.1
30–35%	5.6	1.5	1.0	1.5
35–40%	0.4	3.1	2.1	3.3
40–45%	1.1	1.5	3.0	2.5
45–50%	1.8	1.9	2.6	4.8
50–55%	1.9	5.2	2.1	2.8
55–60%	1.0	5.8	1.5	3.8
60–65%	5.6	3.5	5.3	4.8
65–70%	2.0	6.7	3.6	4.6
70–75%	5.1	9.2	5.2	4.3
75–80%	3.8	7.1	4.4	6.9
80–85%	3.6	7.2	3.2	5.3
85–90%	6.8	7.7	5.4	4.4
90–95%	5.2	12.7	10.1	8.9
95–100%	43.9	23.4	38.8	38.0

. In the four samples reported in Table 5, Cu-bearing particles composed of >90% Cu form between 46.9 to 49.1% of the total Cu-bearing particles present in Samples A, E and F, but only 36.2% in Sample B. Conversely, between 11.5 and 21.2% of the total number of Cu particles are present in individual particles composed of <50% Cu.

Automated SEM-EDS analysis also generates false color compositional maps of the analyzed particles, showing the textural association of the different defined phases present in a sample. Particle attributes such as shape and size are accurate depictions of the particles measured; within each particle, the composition is assigned on a pixel by pixel basis, and in this way a compositional map of the particle is generated [25]. A significant advantage to the AMICS analysis system is that a full-area SEM backscatter electron image is also generated during analysis, which can be compared with the compositional maps, and may provide additional textural data. Representative AMICS particle images are shown in Figure 9; these highlight a significant advantage of automated SEM-EDS analysis over other methods such as X-ray diffraction, in that the quantitative compositional data generated are in textural context. The visualization of the particles present allows their size, shape, compositional complexity and degree of liberation to be rapidly evaluated, and all of these parameters can be quantified from the single analytical measurement.

4.3. Micro-XRF Elemental Mapping

A potential limitation of the automated SEM-EDS particle composition characterization is that, particularly in the coarser-grained waste PCB samples, only a limited number of particles are measured. Micro-XRF has the advantage that no prior sample preparation is required, and a large-area sample can be measured rapidly (see Table 2). There are several modes in which the micro-XRF signal can be processed and displayed, which is very similar to an SEM-EDS system, in that measurements can be in point, line, or mapping (elemental and phase) mode. In this study, micro-XRF was used to provide element distribution maps for the waste PCB samples with no pre-treatment, scanning across an area 10 cm × 8 cm. In this case, the elemental mapping is not quantitative due to the inhomogeneity of the sample and the lack of available standards, and instead provides a rapid visual understanding of elemental abundance, size and shape, and also which particles (fragments) are hosting which elements (i.e., deportment and association). Quantification could be achieved if there was a suitable standard to compare against the PCB, taking into account all the plastic material.

As shown in Table 2, micro-XRF scanning was carried out five times using different measurement parameters; as the pixel size decreases, and the analysis time increases, the resolution and the analytical signal will improve, yielding higher-quality images for interpretation. As an example, Figure 10 provides element distribution maps for Au in Sample D for four different analytical resolutions. All the primary Au-bearing fragments identified in the highest-resolution measurement (Figure 10D) are also observed within the lowest-resolution measurement (Figure 10A), although the detailed particle image resolution is poor. There is, however, little apparent improvement in resolution between measurement 3 (Figure 10C), with a pixel size of 200 µm and a pixel measurement time of 2 ms, which took 164 min to measure six samples, and the highest-resolution measurement with a pixel size of 200 µm and a pixel measurement time of 30 ms, which took 948 min to measure six samples (Figure 10D) for gold. This is partly because Au has high-energy L-lines that are efficiently excited by X-rays. In addition, depending upon the nature of the matrix material and the energy line/element being detected, rather than purely being a surface analysis, the XRF beam will penetrate into the material. For example, with a plastic matrix, and the Au-L line, the beam will penetrate ~3 mm below the surface. This allows the Au L-lines to be detected within plastic coatings, as is noticeable for the Au square present in the right side of each image, which represents the gold wires inside a microchip.

A wide range of elements can be detected using micro-XRF, depending on concentration levels within the analyzed materials. Elemental information is acquired using EDS detectors, so that all elements are detected simultaneously, with the benefits of lower detection limits and an ability to acquire higher energy lines (up to 40 keV). This helps resolve potential overlaps at the lower energies, thus making it easier to identify key elements like Ag (K-alpha at 22.1 keV) or Sn (K-alpha at 25.27 keV), as examples. Given that the M6 is an open source system, and thus the samples are measured in air, only elements with an atomic number of 13 (Al) and higher will be detected. It would be possible to achieve an improved signal for lighter elements using the enclosed desktop system under vacuum, wherein elements down to atomic number 6 (C) can be detected, or using a He flush system on the M6. The resulting maps can be displayed for either single elements or any combination of elements, and representative maps are provided in Figure 11 and Figure 12 for six of the analyzed samples. In this case, whilst the micro-XRF data are not quantitative, the element distribution maps visually allow comparisons of the relative abundance of a particular element between different samples, and also provide a correlation between particle types and elements of interest. Whilst the bulk chemical analysis of a waste feed provides quantitative elemental abundance, these data do not provide an understanding of the deportment of those elements into the different components of the heterogeneous feed. In addition, chemical analysis requires sample dissolution, whilst micro-XRF can be carried out rapidly on unprepared materials. Figure 11 provides element maps for the key metals Au, Ag, Pt, Cu and Sn. For the low-abundance, high-value metals (Au, Ag, Pt), the element distribution maps allow the host particles to be identified. These data could then be used to design a process circuit to allow the separation of these particles from those that are barren of these elements, allowing a pre-concentration stage ahead of metallurgical processing. The analysis of samples after the separation stage could be used to evaluate process efficiency in the waste streams. For major components such as Cu, the distribution maps allow a visual comparison of metal abundance between different samples. It is also possible to perform automated mineralogy (AMICS) on samples measured using micro-XRF, as per the SEM (see Figure 3A), but on larger areas, and at lower relative resolutions (pixel size). However, due to the sample preparation method used here, and with the objective being the speed of measurement and information, this was not applied in this study, but it might be a future application for the larger size fractions.

Figure 12 provides element maps for Ni, Pb, Zn and Br for each of the six samples. In addition, a combined map for Ni+Zn+Au is also shown. Element mapping can clearly highlight the presence of key metal “payables” within a sample, but it can also be used to identify the presence of “penalty” elements and potential environmental contaminants. Some metals will be present as alloys within individual PCB components, and overlaying different element maps allows the identification of these composite materials. This information could be improved via automated mineralogical analysis, but would require additional sample preparation to get the best analytical results.

5. Discussion—Does Phase Characterization Aid Waste PCB Recycling?

In this preliminary study, we have used a range of analytical tools to characterize shredded waste PCB. Any analytical tool has both strengths and limitations, and commonly an integrated approach is required to characterize a material fully. One limitation to some of the methods used (binocular microscopy and manual SEM imaging) is that they only provide information about the surface attributes of the samples, whilst it is clear that individual particles are complex when viewed in 3D. Binocular microscopy is simple, low-cost, and can be used to rapidly scan a large area sample, but it provides only limited qualitative data on particle morphology, size and composition. Manual SEM imaging is typically restricted to either finer particle sizes, or a limited number of particles. It does, however, also enable quantitative spot chemical EDS analysis for particle characterization.

Automated SEM-EDS (AMICS) analysis provides detailed particle-by-particle characterization and quantitative data on the detailed material composition, along with other parameters such as liberation characteristics. The limitation to the current analysis is that, in coarser-grained waste PCB, the particle size of the waste requires multiple subsamples to be prepared, and even then, a relatively small number of particles are analyzed. As such, automated SEM-EDS analysis may provide more statistically reliable data for more finely shredded/ground waste streams, although it is interesting to note that the estimation of the metal content of the analyzed samples is directly comparable with previous estimates for metal abundance within waste PCBs [2,3,5]. Further work will compare results from automated SEM-EDS analysis with those from bulk chemical analysis. The analysis of coarse waste PCB is, however, of value in determining the optimal size reduction required for phase liberation. For example, the Cu liberation data shown in Table 5 are for two finer-grained samples (B and F) and two coarser samples (A and E). For a Cu liberation of >50% (i.e., where a particle is defined as “liberated” if over 50% of the particle by area is Cu), 78.8–79.6% of the coarser particles are liberated, whilst 83.7–88.5% of the finer samples are liberated; i.e., at 50% liberation, the decreased grain size results in an increased liberation of 5–10%. In contrast, at 70% liberation, all four samples are very similar in terms of Cu liberation (67.1–68.4%), hence the increased size reduction has not led to improved liberation.

Micro-XRF allows much larger samples to be analyzed, with no prior preparation. In addition, analysis is very rapid; even when using a combination of a fine pixel resolution and a longer analysis time, six samples were fully scanned in less than 30 min. Where samples are of low grade, for example for Au, the analytical speed would mean that a much larger area could be scanned to improve data acquisition. Depending on the specific element, the detected energy line and the host material, micro-XRF analysis is not purely a mode of surface analysis, in that the beam may penetrate into the host material—for example, Au may be detected at a depth of up to 3 mm within a host plastic. The data presented clearly allow the identification of the relative elemental abundance between different samples, and also allow a rapid visual understanding of the deportment of elements of interest to their host particles. The limitation is that the chemical data are not quantitative due to the minimal sample preparation and the fact that no standards were used in this study. The two independently derived datasets (automated SEM-EDS and micro-XRF) are complimentary and can be used in conjunction. For example, in the routine automated SEM-EDS analysis, gold grains were only detected at trace abundance in three samples. However, visually, gold appears to be more abundant than this, based on the micro-XRF analysis of larger, more representative samples, and also noting the penetration of the beam below the sample surface. As the Au occurs in a relatively small number of discrete particles, micro-XRF can allow the mapping and identification of these particles, which, if then pre-processed and concentrated, could be analyzed by automated SEM-EDS to characterize in detail the occurrences of Au.

The techniques used in this investigation could be of potential value to (a) determine the correct sampling and sample preparation strategy to ensure that samples are unbiased and fully representative, and (b) provide rapid, cost-effective, indicative data on the payable metal and penalty components to facilitate the trading and transactions of E-scrap.

6. Conclusions

E-Scrap waste streams are compositionally complex and markedly heterogeneous. Where these wastes are being traded there is a need for robust characterization to determine the nature and confirm the type, and therefore the potential value, of the waste. In addition, understanding the complexity of the feed can aid in designing the most cost-effective reprocessing pathway. We have used a “geometallurgy” approach, routinely used in mineral processing, to characterize a range of shredded waste PCB samples. Of the methods used, a combination of rapid micro-XRF element mapping along with detailed particle characterization through automated SEM-EDS analysis has the potential to (1) determine whether or not a shredded/crushed material is derived from waste PCBs, and if so, what types; (2) identify what payable elements are present and what particle types within the waste they are deporting to; (3) determine the particle morphologies, types and microstructures present, and (4) determine the degree of liberation of the individual components in different size fractions. Future work will aim to compare these methods with detailed chemical analysis.