Introduction

The quantities of Waste Electrical and Electronic Equipments (WEEE) are increasing worldwide, by 2030 the global mass of electronic-waste generated is forecast to rise by 40% (Forti et al.). Modern solutions are needed to recycle this electronic-waste efficiently and with high recycling rates in order to decrease the need of primary raw materials. This helps to drop carbon dioxide emissions (Hintzmann et al. 2010) and reduces dependencies on imports of critical raw materials (Garbarino et al. 2018).

Current industrial recycling approaches for e-waste vary widely in their type, intensity and degree of automation depending on the plant (Kurth et al. 2018; Martens and Goldmann 2016). Basically, four main process steps can be identified that are found in most industrial recycling plants. A flow-chart of this main recycling process is shown in Fig. 1.

Fig. 1
figure 1

Main process scheme for E-Waste, in accordance to (Sander et al. 2019; Handke et al. 2019)

Full size image

The first process step, which in most cases is carried out completely manually (Bilitewski et al. 2017; Handke et al. 2019; Kurth et al. 2018), is the presorting of the e-waste. On the one hand, this should result in the removal of critical components of which the minimum level is defined in EU Directive 2012/19/EU (WEEE directive) (European Union 2012), and on the other hand, devices with a high resource potential should be sorted out (Sander et al. 2019; Handke et al. 2019). The necessary removal of critical components according to EU Directive 2012/19/EU (European Union 2012) is listed in the following:

  • Batteries

  • Printed circuit boards of mobile phones or with an area > 10 cm2

  • Toner cartridges, liquid and paste

  • Cathode ray tubes

  • Gas discharge lamps

  • Liquid crystal displays > 100 cm2

  • Plastic parts with Brominated flame retardants

  • Electrolyte capacitors containing substances of concern

  • Components containing: Asbestos, Mercury, PCB/PCT, refractory ceramic fibres, radioactive substances

In order to remove all critical components according to the EU Directive 2012/19/EU (European Union 2012) two more processing steps are implemented, a pre-crushing stage in which the objects are processed through a hammer mill or a cross-flow shredder (Kurth et al. 2018; Martens and Goldmann 2016) cracking the electronic devices housing, to expose the internal components such as batteries or printed circuit boards. Followed by further manual sorting for the removal of critical components (Kurth et al. 2018; Sander et al. 2019; Handke et al. 2019). As described in the EU Directive 2012/19/EU the removal of critical components “shall be applied in such a way that environmentally sound preparation for reuse and recycling of components or whole appliances is not hindered” (European Union 2012) therefore, cutting procedures are not used in the pre-crushing stage. However, the use of hammer mills or cross-flow shredders can also lead to damage of the critical components (Kurth et al. 2018) and result in contamination of further components. Damage of batteries in particular, poses a high fire risk (Bruns and Dinse 2018) and results in the release of highly toxic substances (Korthauer 2013).

After this second stage of critical components removal, the remaining mass flow is crushed in a multi-stage size reduction process by various shredders. The resulting fine fraction is sorted by different separation methods into primary materials, which can be reused as recyclates, and a residue fraction that cannot be reused any further (Martens and Goldmann 2016; Handke et al. 2019).

For the evaluation of current recycling and the use of recyclates in new products within the EU, the end-of-life recycling input rate (EOL-RIR) considers “the total material input into the production system that comes from recycling of post-consumer scrap” (Peiró et al. 2018) and is therefore a measurable indicator of the dependence on primary raw materials. Table 1 lists the EOL-RIR for some selected elements. This illustrates how heavily the EU relies on the use of primary raw materials for elementary metals such as aluminium, copper and iron. Even more critical is the limited use of recycled materials for elements that are currently extremely important for the European transition to renewable energies, such as neodymium and lithium (Marscheider-Weidemann et al. 2021) with an EOL-RIR of 1% or even 0% (Blengini et al. 2020).

Table 1 End-of-life-recycling-input rate compared to primary raw material input for selected materials in the EU (Blengini et al. 2020)
Full size table

In order to reduce this dependence on primary raw materials, the improvement of current recycling is an elementary component. Studies show that intensive manual presorting and disassembly before the multi-stage size reduction, can significantly increase the amount of reusable materials (Robert 2020; Zeller et al. 2016). Unfortunately, this intensive manual labour is not covered by the additional proceeds from the recycled materials. In order to achieve economic viability, the time for manual disassembly carried out must be reduced by more than 80% (Robert 2020). Such an increase, is not feasible with manual labour. Only an automated process is able to carry this out in an economical way.

To achieve this, a robotic fractionation line for e-waste is being built in the Circular Digital Economy Lab (CDEL). With this line, small electrical devices will be automatically fractionated in order to improve the downstream mechanical recycling process. In this new approach, the disassembly will not be based on a reversible production process, but will generate splitted sub-fractions that will improve the downstream recycling process. These sub-fractions should avoid material compounds that can no longer be separated, for example magnets in steel, and at the same time produce material mixtures that are as simple as possible, such as fractions only containing steel and plastic, which are easy to separate and recycle in a conventional recycling process (Duddek and Freitas Seabra da Rocha 2022). In order to be able to carry out such automation, knowledge of the internal structure of the devices is required, which is not provided by the manufacturers of the devices.

X-ray

X-ray Technology

To gain knowledge of the devices inner structures before opening it, X-rays are an appropriate way. X-rays are light waves with a high-energy electromagnetic radiation of short wavelengths, typically 0.1–10 Å (MacDonald 2017). Contrary to light, these photons with high-energy and a short wavelength are only difficultly absorbed and therefore can pass through most materials. Depending on the electron density of the material more or less of the transmitted X-rays are absorbed by the material (MacDonald 2017). By capturing the X-rays behind the object, the internal structure can be displayed, as shown in Fig. 2 for a cordless screwdriver.

Fig. 2
figure 2

X-ray illustration of a cordless screwdriver

Full size image

X-ray Generation

The most common source for X-rays is a X-ray tube, in which free electrons are generated by a hot wire and then be accelerated with a high voltage to impact into a metal target. In this metal target the accelerated electron knocks out a lower orbital electron of the metal atoms. This gap is instantly filled with an electron from a higher orbital, through this change of orbitals the electron releases energy in form of a X-ray photon, as shown in Fig. 3 (MacDonald 2017).

Fig. 3
figure 3

Release of X-ray photon by electron intrusion (MacDonald 2017)

Full size image

In this particularly case, the X-ray radiation is generated by a pulsed X-ray generator with 270 kVp photon energy. This differs from the continuous X-ray tube in the fact that X-rays are not generated continuously, but only in short pulses. Figure 4 shows the electronic schematic of a pulsed X-ray generator. The power source in the form of a battery charges a capacitor, which is discharged in a time-controlled manner through a transformer. This discharge pulse is conducted via the transformer and the voltage is stepped up until a discharge tube switches through. This high-voltage pulse is then transferred to the X-ray flash tube. A tapered tungsten rod is arranged centrally in this tube, forming the anode. The tip of the tungsten rod is surrounded by a ring-shaped cathode. As a result of the high-voltage pulse, a flow of electrons is excited, which leads to the emission of X-rays in the direction of the anode rod tip (Osterloh et al. 2003). This kind of radiation is the so-called bremsstrahlung, it is generated through the loss of velocity of the electrons moving through the metal rod. In this process the kinetic energy is transformed to electromagnetic energy, which is transmitted as radiation (MacDonald 2017).

Fig. 4
figure 4

Electrical schematic for pulsed X-ray generator in accordance to (Osterloh et al. 2003)

Full size image

This method of radiation generation offers two major advantages. Firstly, the radiation dose generated is very low, as this is the product of incident intensity (I) multiplied by time (t) and absorption coefficient (μ) as shown in Eq. 1.

$$ Dose \approx \frac{I_{0}\mu_{1}t}{\rho} $$
(1)

The pulse duration of the pulsed X-ray device is 25 ns, of which 20–30 pulses are sufficient to X-ray one device. Classic continuously radiating X-ray units with a comparable acceleration voltage achieve significantly higher dose rates. This notable reduces the necessary protection measures to contain the occurring radiation. Secondly, the X-ray unit is also very economical due to the pulse operation, as the required high voltage only has to be generated for a very short time. This enables mobile operation with a rechargeable battery and reduces energy consumption.

X-ray Detection

For the digital processing of X-ray data, it is necessary to record and store the intensity of the X-rays that pass through a device. For this purpose, a digital X-ray detector is used. Within this detector, the X-rays are converted into visible light via caesium iodide scintillator crystals. These scintillator crystals are excited by collision processes when high-energy photons (X-rays) pass through them and release their excitation energy in the form of light (Lecoq et al. 2017). The intensity of the generated light is then captured by photodiodes. The caesium iodide scintillator crystals are grown directly on the photodiodes. This setup is shown in Fig. 5. This allows a pixel-precise representation of the intensity of the X-rays that pass through the object to be examined. The detector has a total of 4228 x 3524 pixels that can display the intensity of the X-rays with a resolution of 16 bits and therefore a total of 65,536 levels. The result is a matrix with 14,899,472 intensity values of the X-rayed object. This gives a direct indication of the density of the object at the various positions in front of the detector.

Fig. 5
figure 5

Sensing mechanism of direct deposition CsI X-ray detector

Full size image

X-ray Based Fractionation

To perform an automated robotic e-waste fractionation as described in “Introduction”, a cutting strategy has to be developed which, based on the generated X-ray data, cuts the object in such a way that optimal fractions are created for the further recycling process. A possible fractionation strategy for an electric screwdriver is shown in Fig. 6.

Fig. 6
figure 6

Cutting lines for fractionation

Full size image

A separation of the motor (fraction A) from the drill chuck and gearbox (fraction B), as well as the separation of the handle (fraction C) and battery (fraction D), is carried out. By separating the cordless screwdriver into the four fractions as shown in Fig. 6, three goals are achieved. First, the fractions created avoid mixing inseparable materials, such as the neodymium magnets of the motor and the alloy steel of the drill chuck, with each other. Secondly, the simplest possible material fractions are generated, such as alloy steel and plastic in the case of the drill chuck (fraction B), and thirdly, removal of the critical components is achieved in accordance to the EU directive 2012/19/EU (European Union 2012) without potential damage to the critical components and without human intervention, by cleanly separating the critical components such as the batteries (fraction D) and circuit boards (fraction C) from the device without mechanically stressing them, such as by a cross-flow shredder (Duddek and Freitas Seabra da Rocha 2022).

In order to extract this cutting strategy automatically from the obtained X-ray data, several processing steps are required, which are performed using Matlab. The sequence of this processing is shown in Fig. 7. In these four steps, the data is processed in such a way that a matrix with cutting data is generated as a result. This matrix contains the start and end coordinates for each cut to be performed. On the basis of this matrix the object can be fractionated robotically. The individual process steps are described in detail below.

Fig. 7
figure 7

Process scheme for automated X-ray based fractionation

Full size image

Preprocessing

To generate an optimal fractionation of the object, the available data must be preprocessed. As already described in “X-ray Detection”, a matrix results as input value, which represents the different densities of the object in a resolution of 16 bit. Depending on the structure and the radiological density of the object, this depth of 16 bit is not always fully utilised. For a good segmentation, however, a delimitation of the different densities is crucial. Therefore, the data is globally adapted to the available 16 bit by stretching the existing value range to the available value range of 16 bit. This leads to a better contrast and an intensified delimitation of the denser ranges from the surrounding ones. The result of this process step is shown in Fig. 8. Global noise reduction is not necessary because there is little noise in the generated data and local noise reduction would distort the delimitation of the different elements.

Fig. 8
figure 8

Preprocessing stage

Full size image

Segmentation

Since the components relevant for fractionation (motor, gearbox, chuck and battery) are clearly distinguishable from the rest of the objects due to their higher density, it is possible to separate these elements from the rest of the object using a segmentation algorithm. Matlab provides various segmentation algorithms, of which six fitting options were compared in a simple test case. Two reliably achieved the required result with different datasets after qualitative preliminary investigations and optimisation of the adjustable parameters. The decisive factor here was that all relevant components were segmented reliably for a wide range of data while unimportant (small and/or low-density) components were segmented as little as possible. In the case of the batteries in particular, there was a big difference in the reliability of the algorithms. A bad fractionation and thus the damage of a battery could be dangerous. The results of the two algorithms, Imbinarize (based on Otsu’s global threshold method (Otsu 1979)) and Imsegfmm (based on Fast Marching Method (Sthian 1999)), are shown in Fig. 9.

The function Imbinarize sorts all pixels above a certain threshold into the foreground by setting them to 1, while all others become 0. Here, the threshold was determined by graythresh and multiplied by a defined factor, as only the densest parts of the dataset are of interest. Imsegfmm grows segmented areas from a seed (here, the densest 10% of the dataset), while each pixel is assigned a weight with graydiffweight based on its intensity difference to the seed mask. Processing the data using Imbinarize does retain significantly more irrelevant details in the cordless screwdriver handle than processing using Imsegfmm. In return, finer details such as the gap of the drive shaft behind the motor bearing, which is relevant for further fractionation, are not lost. In addition, the Imbinarize algorithm is 100 times faster than the second candidate Imsegfmm in various tests, which is highly relevant for the planned process of continuous disassembly, and therefore Imbinarize is chosen for fractionation.

Fig. 9
figure 9

Segmentation

Full size image

Postprocessing

The goal of postprocessing is to improve the previously created segmentation, this includes the removal of unnecessary elements, such as the cables and screws that are still in the segmentation, as well as the optimisation of the geometry of the segmentation. The workflow for postprocessing is shown in Fig. 10. In the first step, the segmentation is opened to remove small connections to pieces that are not part of a larger component but were segmented, especially cables and the motor shaft (a). Afterwards, all components smaller than 15,000 px are removed, such as screws or smaller metal parts (b). Next, all remaining holes are filled to obtain a closed segmentation (c). In a final step, the fragments are enlarged to ensure a complete segmentation of the components, to prevent components from being cut due to insufficient segmentation (d).

Fig. 10
figure 10

Postprocessing workflow

Full size image

Cutting Lines

In the last stage, the optimal cutting lines are generated to separate the previously captured fragments. In order to be able to generate these cutting lines, an outer boundary must be created first, which serves as the start and end point for each cutting line and as a size reference. For this purpose, the entire object, which is the largest object darker than the background is segmented. The resulting mask, representing the complete cordless screwdriver, is shown in Fig. 11.

Fig. 11
figure 11

Object mask

Full size image

Before the previously segmented fragments can be separated from the object, a cutting mask must be created around each fragment. This is accomplished by surrounding each individual fragment by the smallest possible bounding box, which is aligned parallel to the x- and y-axis. Since the individual fragments cannot be assumed to be parallel to the x- and y-axis, the individual fragments are rotated in 5∘ steps by a total of 85∘ and the smallest possible box parallel to the x- and y-axis is formed in each case. The rotation in which the smallest box is created therefore forms the ideal rotation for the fragment to be separated, since in this box, the content not being part of the relevant fragment is minimised. An illustration of this process of cutting box generation for the motor fragment can be found in Fig. 12.

Fig. 12
figure 12

Cutting box generation

Full size image

This cutting box is then transformed by the determined rotation and placed on the object mask from Fig. 11. To estimate if a line of this cutting box is a cutting line, the area of the object mask on all four sides outside the cutting box is calculated. If the area on one of the four sides is larger than half of the average of all four lateral areas and larger than 10,000 px, this side of the cutting box is assumed to be a cutting line. This ensures that a large area of the entire object is cut off and that no unnecessary cuts are made which, for example, only cut of a partial area of plastic. This process is repeated for all fragments consecutively. To ensure that the fragments are completely separated by the identified cutting lines and that no unnecessary long cuts are made, all identified cutting lines are subsequently extended or shortened to the size of the object mask. To avoid multiple cuts being made at close distances, as it might for example be the case between the motor and the gear with chuck, all cutting lines are checked for parallelism and removed as a cutting line, if the distance is less than 50 px to another parallel cutting line. In addition, cutting lines are checked for their collinearity with a tolerance of 5∘ and are joined together if the distance between them is less than 50 px, as it occurs, for example, for the horizontal cutting line under the motor and gear with chuck. The cutting lines automatically generated in this way, combined with the individual fragments can be seen in Fig. 13.

Fig. 13
figure 13

Generated cutting lines based on segmentation

Full size image

To generate the planned cuts, the start and end points of each cut are transferred to a robotic system, this robotic system moves the object to a high pressure waterjet cutter, to perform the cutting procedure, in order to prepare the object for a better material recovery in the subsequent recycling process.

Application on Further Products

The fractionation strategy shown for a cordless screwdriver is only one example for the use of this technology in the field of e-waste recycling. It can be easily transferred to similarly constructed devices. Especially for very packed devices, a simple cut for fractionation is only limited applicable, as here batteries or displays often cover the entire size of the device. But even in this case, the application of X-ray technology can be used to determine the internal structure. In addition, the implementation of a further fractionation stage would be necessary, such as the use of a mechanical or a thermal treatment to separate the different layers of the device in order to subsequently carry out a fractionation with a cutting process, for each individual layer. In addition, the guidance of the device with an industrial robot during the cutting process also allows cutting in several axes through the device, which also enables the separation of stacked components.

Conclusion

As illustrated by the example of a cordless screwdriver, it is possible to create an algorithm based fractionation strategy for small battery-powered electrical devices on the basis of X-ray data. Using X-ray technology, the internal structures of the devices can be visualised without opening them and without information from the manufacturer, as these are not provided. This allows simple fractionation for a downstream recycling process to be done safely without the risk of damaging sensitive components. This fractionation of small electronic devices simplifies downstream recycling by creating clean fractions, as in the example of gears and chucks (Fig. 6-B), a fraction of steel and plastics, without copper, without neodymium. Complex components such as the electric motor (Fig. 6-A) are cleanly separated and can be individually fed into a specialised recycling process designed for electric motors. Due to the significantly reduced mass flow, such a specialised mechanical recycling process can be carried out more economically than if all process steps for the recycling of a motor are applied to the entire device. In addition, a clean separation of all critical components such as the battery and the contained circuit boards takes place without the risk of damaging them and thereby without contaminating other valuable materials. Overall, more materials can be recovered from the electrical devices in a downstream mechanical recycling process due to the improved availability and purity of the individual fractions. Furthermore, the fractionation and removal of the critical components, which is required by EU-law for a conventional recycling process, can be performed fully robotised. For an industrial appliance, the economic viability of this procedure needs further investigation compared to a manual disassembly. In spite of the fact that the production of small battery-powered electrical devices is already partially or fully automated, automation is also needed when the devices have reached the end of their life cycle and represent a significantly lower economic value than at the time of their production.