Method and apparatus for evaluation of value metals in electronic waste

Abstract

The present disclosure relates to the assessment of noble metal content in batches of materials, such as electronic waste, and in particular printed circuit boards, which is the necessary initial step when a recycle is assumed. It is proposed a method comprising the steps of:-imaging at least a statistically representative number of the plates of the batch; -processing the image to detect the printed circuit board; -for each detected printed circuit board, extracting a board-related feature vector using an image processing technique; and-providing a model taking at least the plate-related feature vector as input and calculating the precious metal content of the batch as output, where the model is calibrated for a batch-level precious metal assay. The present application relates directly to a batch of printed circuit boards, allowing a method in which a regression model is calibrated based on total batch measurements. And the printed circuit board component measurement is not required.

Description

[0001] When recycling is envisaged, assessment of the value of metals in batches of material such as electronic scrap is a necessary preliminary step.

[0002] In the present context, the electronic scrap of interest consists mainly of printed circuit boards (PCBs). Such electronic scrap is obtained in batches comprising hundreds of mixed boards.

[0003] Commercial transactions between suppliers of materials and recyclers require highly accurate measurements of each batch. Measurements are also necessary to closely monitor metallurgical processes, for example with respect to tracking any metal losses and optimizing process performance.

[0004] The value of electronic scrap is most often determined by the precious metal content, i.e. the amount of PGMs (Ru, Rh, Pd, Os, Ir, Pt), Au and Ag present in the material. Cu can also contribute significantly to the value and is also considered a precious metal in this context.

[0005] Since most of these recycled materials are complex and inhomogeneous, accurate and reliable determination of their composition can only be obtained through a 100% sampling scheme. This scheme is a rather complex multi-step process and includes a smelting operation performed on a representative sample to collect the precious metals in the metallic phase, which is then sampled again for chemical analysis. These operations are time consuming and result in delayed availability of critical information about the exact content, which in turn delays commercial transactions between suppliers and recyclers.

[0006] Therefore, an early assessment or prediction of precious metal content would be advantageous, even if it is not as precise as chemical analysis.

[0007] A system has been developed with this goal in mind.

[0008] A publication that addresses a similar problem is “Material value estimation for recycling of waste printed circuit boards by deep-learning-assisted approach on X-ray images”, M. Firsching et al., SBSC 2022, 9 thSensor-Based Sorting & Control, ShakerVerlag, Aachen 2022, pp. 161-178. This publication teaches a metal value estimation system for printed circuit boards. The estimation is based on a preliminary chemical determination of various categories of components such as ICs (particularly pin grid array ICs and ball grid array ICs), tantalum capacitors, and thickness measurements of gold coatings on connectors. The boards are subjected to automated component recognition and counting. The metal value of each board is derived from the count of components in each category. Component determination is necessary.

[0009] In contrast, the present application relates directly to batches of printed circuit boards, allowing a different approach in which the regression model is calibrated based on total batch measurements. No component measurements are required.

[0010] As mentioned above, a complete batch determination is always required and will always be performed independently of any other estimation. By using these determinations as calibration data, the delayed availability of accurate batch content is greatly mitigated. Furthermore, the fact that no additional effort is required to obtain calibration data makes continuous calibration more affordable.

[0011] A first embodiment of the present invention relates to a method for evaluating the precious metal content of a batch of printed circuit boards, the method comprising the following steps:

[0012] - imaging at least a statistically representative number of the printed circuit boards of the batch;

[0013] -processing the image to inspect the printed circuit board;

[0014] - for each detected printed circuit board, using image processing techniques to extract board-related feature vectors; and,

[0015] - providing a model taking at least the plate related feature vector as input and calculating the precious metal content of the batch as output, wherein the model is calibrated for batch level precious metal determination.

[0016] A batch usually corresponds to a collection of items obtained from a supplier within the framework of a single purchase contract. In relation to the present invention, a batch can contain any number of printed circuit boards, some of which may be damaged. It can also contain partially or completely separated individual components.

[0017] In the context of the present invention, the term "board" should be understood as a "printed circuit board" or "PCB", including all electronic or other components mounted on or attached to the board.

[0018] Detecting a plate means at least defining a corresponding bounding box, which is preferably supplemented with a mask identifying the contours more precisely. This makes it possible to perform plate-related feature extraction.

[0019] An "imaging system" may operate in the visible electromagnetic spectrum, but may also use IR, UV or X-ray sources and cameras.

[0020] Board-related features are, for example, the type of board (such as motherboard or sound card), and the type of detected components on the board (such as ICs or connectors). Features are mathematically represented by feature vectors.

[0021] According to another embodiment, the step of calculating the precious metal content of the batch is performed based on a combination of the board-related feature vector and batch-level features from a source other than the printed circuit board imaging.

[0022] The plate-related features are optionally combined with other batch-related features from sources unrelated to the plate imaging, such as the identity of the supplier or the geographical origin of the batch. In addition, historical data for batches from the same supplier may be used. Another example of a batch-related feature is the result of an optional rapid analysis method performed on the batch or on a statistically representative sample thereof. Prompt gamma neutron activation analysis is a technique that, when used alone, would not provide the required accuracy.

[0023] According to another embodiment, the calculation of the precious metal content of the batch is performed by either of options (1) and (2):

[0024] -(1) summing the plate-related feature vectors to obtain a batch-related feature vector; and calculating the precious metal content of the batch based on the batch-related feature vector; or,

[0025] - (2) for each printed circuit board, evaluating the precious metal content of the printed circuit board based on the feature vector associated with the board; summing the precious metal contents of the printed circuit boards to obtain the precious metal content of the batch.

[0026] When using non-linear regression models, the first options can have advantages as they will better capture the complex interactions between batches and plates. However, fitting a non-linear model will require more calibration data. The second option provides an assessment of the precious metal content of a single plate, allowing classification and sorting. While the prior art can achieve similar results, it relies on preliminary chemical determinations of many different components to calibrate the model, whereas according to the present invention, total batch determinations are used instead.

[0027] When using a linear model, both described options will calculate the same batch contents.

[0028] According to another embodiment, the method of assessing precious metal content is performed prior to metallurgical processing of the batch to recover the precious metals.

[0029] Knowing the precious metal content, even if approximate, allows the selection of an optimized metallurgical process. For example, batches rich in precious metals can be processed with high priority using methods that ensure maximum recovery of the precious metals.

[0030] According to another embodiment, a method of assessing precious metal content is used to qualify an advance payment to a supplier of a batch of printed circuit boards.

[0031] The advance payment corresponds to a preliminary estimate of the precious metal content using the above method. Once a more precise content is determined, this payment is regularized.

[0032] According to another embodiment, when the precious metal content evaluation process is selected according to option (2) above, the printed circuit boards are sorted into at least 2 categories based on the evaluated precious metal content of the printed circuit boards.

[0033] According to another embodiment, each printed circuit board type is individually metallurgically processed to recover precious metals.

[0034] A batch may contain printed circuit boards with widely varying precious metal contents. By dividing them into at least 2 categories, distinguishing between boards with an expected higher precious metal content and boards with an expected lower precious metal content allows optimization of the subsequent metallurgical processes for recovering and refining the precious metals. However, if the calculation scheme according to option (2) above is applied, only the metal content per board will be evaluated.

[0035] According to another embodiment, a batch level precious metal assay is obtained by smelting a statistically representative sample of the batch using the alloy as a collector and analyzing it.

[0036] A representative sample can be obtained by sampling the batch, then pulverizing the sample, and then resampling the pulverized product. This sampling, pulverizing and resampling scheme can be repeated until a reduced sample volume is obtained that is compatible with the needs of further steps leading to chemical determinations.

[0037] Another advantageous step is the smelting of the sample. The noble metal is thereby separated from the less noble metal and concentrated in an alloy. Such alloys are generally homogeneous and well suited for accurate chemical determinations. The determinations can be performed using any accurate analytical method. Inductively coupled plasma (ICP) is, for example, a suitable technique.

[0038] According to another embodiment, the total amount of flammable compounds in the batch is assessed.

[0039] The flammable compounds in electronic scrap may include plastics and certain metals, in particular aluminum. These compounds will contribute significantly to the enthalpy when a pyrometallurgical refining process is applied. Therefore, it is useful to make an early assessment of the approximate content of carbon and aluminum, i.e. before the refining process actually begins.

[0040] This assessment can be performed according to the same protocol as applied for precious metal assessment. The model is then additionally calibrated for batch level determinations of carbon and aluminum.

[0041] Another embodiment is directed to an apparatus for evaluating the precious metal content of a batch of printed circuit boards, the apparatus being configured to perform the above method.

[0042] A device "configured to execute" means that at least a portion of the software required to execute the model is present on the device or available to the device.

[0043] More specifically, a device is disclosed, the device comprising:

[0044] - a conveyor belt adapted to expose the printed circuit board to the imaging system;

[0045] - the imaging system comprising at least one camera observing the conveyor belt;

[0046] - one or more computing units connected to the imaging system and configured to perform the following tasks:

[0047] (a) processing images from the imaging system to inspect the printed circuit board;

[0048] (b) extracting board-related feature vectors for each detected printed circuit board;

[0049] (c) executing a model based on the vector as input and calculating the precious metal content of the batch as output, wherein the model is calibrated for batch level precious metal assays.

[0050] The computing unit may include dedicated hardware and / or one or more general-purpose computers that contain or have access to the software required to perform the tasks.Some tasks may be performed by resources in the cloud.

[0051] Another embodiment relates to a computer program product comprising instructions that, when executed by a computer, cause the computer to implement the above-described model.

[0052] Another embodiment relates to a data carrier comprising instructions which, when executed by a computer, cause the computer to execute the above-described model.

[0053] The following examples are provided to further illustrate the present invention. Figures 1 to 4 It refers to the following aspects.

[0054] Figure 1 The imaging and inspection of PCBs on a conveyor belt is illustrated. Reference numeral 1 corresponds to the back of a computer motherboard, reference numeral 2 corresponds to the front of a sound card, and reference numeral 3 corresponds to the front of a RAM memory card.

[0055] Figure 2 The extraction of board-related features from the front side of an imaged motherboard is illustrated. Reference numeral 1 is a CPU, reference numeral 2 is a chipset IC, reference numeral 3 is a parallel port, and reference numeral 4 is a PCI slot.

[0056] Figure 3 The average mean square error (MSE) of Au in ppm for 10 different training scenarios is shown on the y-axis versus the number of training batches on the x-axis. Three linear models were fit using the usual least squares method (x), Lasso (+), and Ridge penalty (o).

[0057] Figure 4 Prediction (x) and training (o) results for Lasso regression for Au are shown. The y-axis is the predicted Au concentration (ppm) and the x-axis is the true / simulated Au concentration. One marker represents one batch. The continuous line represents the ideal response, where the predicted value is equal to the true value.

[0058] To illustrate the invention, a realistic simulation of a variety of printed circuit boards was used, accurately representing those typically encountered in the electronic waste recycling industry. A database of components and boards assembled from those components was created. Simulated batches of boards were then generated, each with an associated total precious metal content. Finally, an optical recognition system was simulated, including typical errors to be considered in the actual process.

[0059] Fit or train a regression model to relate the identified features to the precious metal content of each batch.

[0060] Parts Database

[0061] 25 component types are defined. Each component type is characterized by a mean value, for each of Ag, Pd and Au, a mean value between 5g and 50g is randomly selected for the component weight, and a mean value between 5mg and 1000mg is randomly selected for the precious metal content. For each generated component, a normal distribution is applied for weight and precious metal content. The standard deviation is set to 25% (for weight), 30% (for Ag), 35% (for Pd) and 25% (for Au). Therefore, each individual component is unique in its type, weight and composition. An offset of the normal distribution with a relative standard deviation of 10% is applied to each batch relative to the metal concentration of the component.

[0062] Board Database

[0063] 10 board types are defined. Each board type is characterized by the board weight distribution and the maximum allowed count of the components of each type. For each generated board, the number of components of each type is randomly selected from 0 to the maximum allowed. The board weight of each type is characterized by a normal distribution with a mean weight randomly selected between 100g and 1000g. The relative standard deviation is 25%. By multiplying the deviation by a random number in a normal distribution with a mean of 100% and a standard deviation of 10%, it is applied to the average weight for each type within the batch. Therefore, each individual board is unique in terms of its type, the number and selection of components, composition and weight. The component database and the board database are very similar to industrial reality.

[0064] Simulated batches of boards

[0065] 100 batches of plates were generated. Each batch consisted of a randomly selected subset of an average of 5 plate types from the 10 available plate types, with a random number of plates between 0 and 1000 for each selected type. On average, each batch thus included 2500 unique plates.

[0066] Optical identification of boards and components

[0067] In the simulation, the board type has a 90% probability of achieving both correct detection and correct classification, while also identifying if the board is presented upside down. A 5% probability of misclassification and a 5% probability of not being detected at all were built into the simulation. The probability of the board being presented upside down is 50%, in which case the board type is detectable, but the component is not.

[0068] When presented with a correctly oriented board, the part has a 95% probability of being detected and classified, regardless of whether the board is classified correctly or not. There is a 2.5% probability that the part is not detected, and a 2.5% probability that the part is classified incorrectly.

[0069] The above parameters were derived from our experience with camera-based feature detection using Mask R-CNN (“Mask R-CNN”, K. He, G. Gkioxari, P. Dollár, and R. Girshick, 2017 IEEE International Conference on Computer Vision (ICCV), 2017, pp. 2980-2988) on plates moving on a conveyor belt. Mask R-CNN was trained on plate annotations and part annotations. Python was used as the programming language, and TensorFlow, Keras, and imgaug were used to create and train the neural network. Pre- and post-processing of the images and results was performed with a combination of the following packages: sci-kitimage, PILLOW, sci-kit learn, scipy, numpy. The pre-processing step consisted of resizing the images and normalizing the color channels. The post-processing steps were image stitching and “non-maximum suppression” to count plates and parts only once.

[0070] Figure 1 and Figure 2 A typical example of a photo recorded by a camera is provided. The plates occasionally overlap and can appear upside down half of the time. The machine learning model Mask R-CNN is used to detect the type, orientation and masks of plates and components. Masks of plates and components are used to connect them to each other. Metadata such as damaged or partially covered plates, plate color, region, etc. can be added.

[0071] Regression Model

[0072] The fitting / training process is performed on simulated batches. For each plate and each part, a feature vector is derived. The feature vector is clustered over the entire batch, and this clustered vector is used in a regression model designed to predict the precious metal content of the batch. Linear regression models are fit using least squares, Lasso, and ridge penalty methods. As the number of training batches increases, Lasso regression slightly outperforms Ridge regression and outperforms least squares. Least squares is sensitive to existing cross-correlations in the data set.

[0073] result

[0074] Using this scheme, such as Figure 3 As shown, meaningful predictions of precious metal content can be achieved after training on about 30 batches. This is remarkably fast given the many different boards and components involved. It is believed that this efficiency is due to the cross-correlations that exist between board type and the components mounted on it. Such cross-correlations are implied by the statistical values ​​used when creating the synthetic batches. These same cross-correlations are believed to exist in industrial reality.

[0075] The accuracy of precious metal forecasts is as follows Figure 4 This result is sufficient to provide early feedback on the material composition to the supplier of the batch, thereby enabling faster payment, or to plan further processing of the material without waiting for the results of the precise chemical determination.

[0076] Once such a system is actually utilized, the training database can be easily completed with actual data from each batch, thereby enhancing the correlation between the feature vectors and the chemical assays. A self-learning system is then obtained, which does not require any additional expense or effort other than periodically running the fitting or training algorithm of the regression model.

[0077] This implementation is particularly valuable because major e-waste recyclers may process several batches per day.

Claims

1. A method for evaluating the precious metal content of a batch of printed circuit boards, the method comprising the following steps: - imaging at least a statistically representative number of said printed circuit boards of said batch; - processing the image to inspect the printed circuit board; -For each detected printed circuit board, use image processing techniques to extract board-related feature vectors; as well as, - providing a model taking at least said plate related feature vector as input and calculating said precious metal content of said batch as output, wherein said model is calibrated for batch level precious metal determination.

2. The method of claim 1, wherein the batch precious metal content is calculated based on a combination of board-related feature vectors and batch-level features from a source other than printed circuit board imaging.

3. The method according to claim 1 or 2, wherein the calculation of the precious metal content of the batch is performed by any one of options (1) and (2): - (1) summing the plate-related feature vectors to obtain a batch-related feature vector; and calculating the batch precious metal content based on the batch-related feature vector; or, - (2) for each printed circuit board, evaluating the precious metal content of the printed circuit board based on the board-related feature vector; summing the precious metal contents of the printed circuit boards to obtain the batch precious metal content.

4. A method according to any one of claims 1 to 3, wherein assessing the precious metal content is performed prior to metallurgical processing of the batch to recover the precious metal.

5. The method according to claim 3 option (2), wherein the printed circuit boards are sorted into at least 2 categories based on the assessed precious metal content of the printed circuit boards.

6. The method of claim 5, wherein each printed circuit board type is individually metallurgically processed to recover the precious metals.

7. A method according to any one of claims 1 to 6, wherein the batch level precious metal assay is obtained by smelting a statistically representative sample of the batch using the alloy as a collector and analysing it.

8. The method according to any one of claims 1 to 7, wherein the total amount of flammable compounds of the batch is assessed.

9. An apparatus for assessing the precious metal content of a batch of printed circuit boards, the apparatus being configured to carry out the method according to claims 1 to 8.

10. The device according to claim 9, comprising: - a conveyor belt adapted to expose the printed circuit board to the imaging system; - the imaging system comprising at least one camera observing the conveyor belt; - one or more computing units connected to the imaging system and configured to perform the following tasks: (a) processing the image from the imaging system to inspect the printed circuit board; (b) extracting board-related feature vectors for each detected printed circuit board; (c) executing a model based on the vector as input and calculating the precious metal content of the batch as output, wherein the model is calibrated for batch level precious metal assays.

11. A computer program product comprising instructions which, when executed by a computer, cause the computer to perform a model of the method according to claims 1 to 3.

12. A data carrier comprising instructions which, when executed by a computer, cause the computer to execute a model of the method according to claims 1 to 3.

Images