System and method for processing metal scraps

Abstract

A system for processing metal pieces, the system comprising: - a transportation system (202) configured to transport a plurality of metal pieces, - an image acquisition system (206) configured to acquire images of metal pieces transported by the transportation system, - an image recognition system configured to identify optical parameters of metal pieces based on images acquired by the image acquisition means, - a spectroscopy system configured to determine spectra of the metal pieces transported by the transportation system and configured to analyze at least a part of the metal pieces based on the optical parameters identified by the image recognition system, and - an electric arc furnace (212) configured to melt the metal pieces controlled by an electric arc furnace control system based on control parameters dependent on the determined spectra and / or the optical parameters of the metal pieces.

Description

[0001] SYSTEM AND METHOD FOR PROCESSING METAL

[0002] FIELD

[0003] The present disclosure relates to systems and methods for processing metal, in particular steel. In particular, this disclosure relates to steel recycling with the use of an electric arc furnace.

[0004] BACKGROUND

[0005] Metal processing, e.g. steel recycling, typically comprises melting metal in an electric arc furnace. The metal that is melted in the electric arc furnace typically originates from a plurality of metal pieces, e.g. a plurality of steel scrap pieces that may vary in geometry and composition.

[0006] Some of the metal pieces, e.g. some of the steel scrap pieces, may not be sufficiently pure / uncontaminated in order to obtain metal with the required quality after melting. In particular, metal pieces may be contaminated and / or pieces may include non-metallic materials and / or contaminants. Dangerous and / or contaminated metal pieces pose a significant risk for the final quality of the processed metal. In addition, the electric arc furnace has to be controlled to ensure the desired quality and efficiency also in terms of energy and economic operation.

[0007] A limited visual inspection of the metal pieces is traditionally used to filter out pieces that have visible anomalies. The use of a spectrometer, e.g. a laser spectroscopic apparatus, is very time and resource consuming and therefore not suitable for large quantities of e.g. steel scrap pieces and is in any case, not suitable for determining detailed geometric or visual characteristics of the metal pieces such that a visual inspection remains necessary when processing e.g. large quantities of steel scrap pieces to filter out dangerous scrap pieces.

[0008] For practical purposes the use of a spectrometer when processing / recy cling a significant amount of metal is therefore prohibitive.

[0009] There is therefore, the need to improve the efficiency of a determination of qualitative and quantitative parameters of metal scrap pieces when processing metal, and / or to efficiently filter out anomalous metal pieces and / or to improve a control of an electric arc furnace based on an improved determination of qualitative and quantitative parameters of the metal pieces.

[0010] SUMMARY The invention is defined by the independent claims. The dependent claims define further embodiments of the invention.

[0011] According to an aspect, the present disclosure provides a system for processing metal pieces, the system comprising:

[0012] - a transportation system configured to transport a plurality of metal pieces,

[0013] - an image acquisition system configured to acquire images of metal pieces transported by the transportation system,

[0014] - an image recognition system configured to identify optical parameters of metal pieces based on images acquired by the image acquisition means,

[0015] - a spectroscopy system configured to determine spectra of the metal pieces transported by the transportation system and configured to analyze at least a part of the metal pieces based on the optical parameters identified by the image recognition system,

[0016] - an electric arc furnace configured to melt the metal pieces controlled by an electric arc furnace control system based on control parameters dependent on the determined spectra and / or the optical parameters of the metal pieces.

[0017] According to an aspect, the present disclosure provides a method for processing metal pieces, the method comprising:

[0018] - transporting a plurality of metal pieces,

[0019] - acquiring images of the metal pieces,

[0020] - identifying optical parameters of the metal pieces based on the acquired images,

[0021] - determining spectra of the metal pieces using a spectroscopy system by analyzing at least a part of the metal pieces based on optical parameters, and

[0022] - melting the metal pieces with an electric arc furnace controlled based on parameters dependent on the spectra and / or the optical parameters of metal pieces.

[0023] According to another aspect, the present disclosure provides a system for processing metal pieces, the system comprising: - a transportation system configured to transport a plurality of metal pieces,

[0024] - a spectroscopy system configured to determine spectra of the metal pieces transported by the transportation system and configured to analyze at least a part of the metal pieces, and

[0025] - an electric arc furnace configured to melt the metal pieces controlled by an electric arc furnace control system based on control parameters dependent on the determined spectra; wherein the system further comprises a detection system configured to identify non-anomalous metal pieces based on the spectra of the metal pieces, wherein the identification of the non-anomalous metal pieces is based on a normal zone of spectra of substantially uncontaminated metal pieces, the spectra measured by the spectroscopy system of the system for processing metal pieces; and wherein the electric arc furnace is configured to melt only the non-anomalous metal pieces.

[0026] In another aspect, the present disclosure provides a method for processing metal pieces, the method comprising:

[0027] - transporting a plurality of metal pieces,

[0028] - determining spectra of the metal pieces using a spectroscopy system, and

[0029] - melting the metal pieces with an electric arc furnace controlled based on parameters dependent on the spectra; wherein the method further comprises:

[0030] - identifying non-anomalous metal pieces based on the spectra of the metal pieces, based on a normal zone of spectra of substantially uncontaminated metal pieces, the spectra measured using the spectroscopy system; melting only the non-anomalous metal pieces when melting metal pieces with the electric arc furnace.

[0031] Aspects and advantages of the present disclosure will be described in detail in the following detailed description and claims and drawings that illustrate the present disclosure.

[0032] BRIEF DESCRIPTION OF THE DRAWINGS Embodiments of the present disclosure are described with reference to the following figures:

[0033] FIG. 1 shows known processing of metal pieces,

[0034] FIG. 2A shows details of a system for processing metal pieces according to embodiments of the present disclosure,

[0035] FIG. 2B shows further details of a system for processing metal pieces according to embodiments of the present disclosure,

[0036] FIG. 3 shows known sampling patterns of a laser induced breakdown spectroscopy system,

[0037] FIG. 4 shows sampling patterns of a laser induced breakdown spectroscopy system according to embodiments of the present disclosure,

[0038] FIG. 5 shows details of a detection system configured to identify non-anomalous metal pieces according to embodiments of the present disclosure,

[0039] FIG. 6 shows a known spectrum of a steel alloy,

[0040] FIG. 7 shows a spectrum of steel alloy as measured according to embodiments of the present disclosure,

[0041] FIG. 8 shows a normal zone of spectra of substantially uncontaminated metal pieces according to embodiments of the present disclosure,

[0042] FIG. 9 shows a comparison between anomaly detection according to embodiments of the present disclosure and a known classification.

[0043] DETAILED DESCRIPTION OF EMBODIMENTS

[0044] A detailed description of embodiments of the present disclosure is presented in the following.

[0045] Different embodiments can be combined with each other, unless otherwise stated or unless they are mutually exclusive.

[0046] The expressions “in one embodiment”, “in some embodiments”, “according to an embodiment”, “according to embodiments” refer to general teachings of preferred features which are combinable (if not otherwise stated or mutually exclusive), i.e. combinable with other features and embodiments described herein, and said expressions are not intended (if not otherwise explicitly stated) as explicitly listing a specific combination of features. Common features of different embodiments may not be explicitly repeated in order to obtain a compact and readable description.

[0047] The expression “embodiment” refers in particular to embodiments of the present disclosure.

[0048] The present disclosure relates to systems and methods for processing metal pieces.

[0049] The metal of the metal pieces may be steel and / or aluminum and / or copper and / or lead and / or any metal alloy, in particular the metal may comprise steel. More particularly, the metal may consist essentially of steel or the metal may be steel.

[0050] The metal pieces may include contaminants as found commonly in scrap processing, in particular in typical steel scrap processing and recycling applications.

[0051] For example, steel pieces may be contaminated by copper and other contaminants. For example, metal pieces comprising steel pieces and / or consisting essentially of steel pieces and / or consisting of steel, may still comprise contaminated pieces for example including copper, such as in particular in the case of metal pieces that may include motor or generator parts still having coper wires or other materials attached and / or that are still formed of different parts and / or that are contaminated.

[0052] The present disclosure is in particular advantageous to process large quantities of steel pieces, e.g. in particular in the recycling of steel scrap pieces. In particular, according to the present disclosure, when the metal is steel or consists substantially of steel, large quantities of steel scrap can be processed improving the quality of the final processed / recycled steel after melting, and without requiring a prohibitive amount of time, while contaminants are still removed effectively.

[0053] Steel can either be produced from raw material (iron ore) or from recycled metal pieces, e.g. steel scrap. For example, when processing / recycling steel, steel scrap may be molten in an electric arc furnace. Steel made from metal pieces, e.g. from steel scrap pieces has a significantly lower carbon footprint than virgin steel and the demand for low-carbon steel is expected to rise due to climate regulations. Scrap based steel production will likely be limited by the availability of scrap and, therefore, an efficient use of all available metal pieces is important.

[0054] According to the present disclosure, metal pieces may be steel pieces and / or steel scrap pieces and / or pieces substantially comprising steel or comprising steel. Additional materials and contaminants may be present in the metal pieces. The metal pieces may be substantially homogeneous or made of different parts that may be different, i.e. with different levels of other materials and / or contaminants.

[0055] Alternatively, or in addition, the metal pieces may comprise or may consist of aluminum and / or copper and / or lead and / or steel pieces or parts including steel pieces.

[0056] FIG. 1 shows known processing 100 of metal pieces. Arrows in the FIG. 1 indicate a temporal relationship, i.e. an arrowhead points to a subsequent state reached in the process of metal processing, e.g. during recycling. Nevertheless, different states may also be simultaneously present and / or the processing may be carried out repeatedly in a loop.

[0057] In a first step, scrap shipping 102 brings metal pieces to be processed / recycled. For example, metal pieces formed by steel scrap pieces are brought to a mill for processing / recycling. The metal pieces may be of different grade or quality and have different compositions and / or include different types and quantities of contaminants or impurities. For example, the metal pieces may be of different compositions and mixtures 104, such as for example mixtures 104-1, 104-2, . . ., 104-n. Said mixtures may already be partitioned in different blocks or may be present together in an undifferentiated quantity of metal pieces to be processed.

[0058] The metal pieces and / or each mixture of metal pieces may be further contaminated and / or include further impurities or elements or compositions. For example, several different grades of metal pieces forming for example metal pieces of scrap may be defined, each defined by different upper limits on metals such as for example copper or lead in the case of steel pieces when recycling steel scrap pieces. There can be different quality levels defined. These requirements limit the amount of scrap that can be sourced and processed together in an arc furnace.

[0059] To filter out metal pieces that do not satisfy the requirements, according to known solutions, typically firstly a visual inspection 106 is performed. Such a known visual inspection 106 is done by human operators and is therefore only superficial. Usually, several tons of material are handled at once, e.g., by cranes or magnets, and an inspection of each individual piece is therefore not possible by human operators.

[0060] Laser-Induced Breakdown Spectroscopy (LIBS) is a sensor technology that can determine the material composition of samples in a minimally invasive fashion. It can, however, only determine the material composition in one small sampling spot on the surface of the sample at a time. Therefore, LIBS is used only on small and / or homogeneous samples today. More traditional spectroscopy systems would require even more time.

[0061] A known solution for analyzing metal pieces, is a laser-induced breakdown spectroscopy system (LIBS) or other spectroscopy system. But even using a LIBS system, a comprehensive analysis of the metal pieces is not possible since individual pieces may overlap, e.g. overlapping on a conveyor belt or other transportation system. In addition, the metal pieces may themselves be heterogenous and a LIBS system is only able to carry out a punctual analysis without knowledge of the geometry of the metal pieces. Geometrical characteristics and / or a volume of the pieces is not provided using LIBS alone. Analyzing all the metal pieces of a large scrap stream is furthermore not possible using LIBS because a thorough analysis of all the metal pieces with a LIBS system would require a prohibitive amount of time.

[0062] Therefore, according to known solutions, the metal pieces / scrap pieces needs to be analyzed in many different sampling spots to achieve a representative overview of the elemental composition which is not possible in a useful time for the quantities of metal pieces usually processed, and which would also require a prohibitive amount of time and energy for a throughout analysis of the metal pieces.

[0063] For example, according to the known solution illustrated in FIG. 1, molten metal of the electric arc furnace is sampled with a limited sampling 108 that typically is carried out a limited number of times The sampling 108 of the molten metal in the electric arc furnace is typically manually controlled.

[0064] In known solutions, the molten metal is analyzed and the information is used to change some parameter in the furnace i.e. to take corrective actions. For example, the metal may be heated to a higher temperature or additives may be added. Such steps cost time and / or money. It is one goal of system and methods according to the present disclosure to make the repeated sampling 108 unnecessary by obtaining sufficient information before the initial melting, further providing the possibility to filter out contaminated and dangerous metal pieces even before they enter the furnace.

[0065] According to known methods therefore, in addition or alternatively, at best a limited number of samples of very few small metal pieces (e.g. scrap chips) may be further manually analyzed also before entering the furnace and the whole mixture / batch, e.g. mixture 104-1 and / or 104-2 and / or ... 104-n is accepted or rejected based on said limited analysis that therefore, cannot efficiently filter the metal pieces to obtain processed metal pieces with a high degree of purity and homogeneity and / or of a desired quality and grade. Moreover, according to known solutions, if a contamination is detected, the whole batch of metal pieces is discarded assuming that the whole batch is contaminated, thereby also discarding those metal pieces in the batch that may not be contaminated and / or dangerous.

[0066] In known systems and methods for processing metal pieces according to FIG. 1, metal 110 is finally obtained by melting the metal pieces, typically with an electric arc furnace 107, as filtered out after the limited sampling and the limited analysis done on the samples resulting from the limited sampling and / or the visual inspection 106.

[0067] In known methods it is therefore not possible to accurately control the electric arc furnace based on the quantitative composition of the metal pieces, because it is not possible to adequately evaluate both a precise and accurate composition of the metal pieces together with a precise and accurate evaluation of a geometry or volume of the metal pieces such that the electric arc furnace receives pieces which compositions that can vary still with significant variance and with a still unknown volume / mass. According to known solutions, it is therefore not possible to provide a desired high quality and grade of the processed metal pieces, e.g. the recycled steel, processing and analyzing in detail a large quantities of metal pieces, e.g. steel scrap, in useful time.

[0068] The composition and variability of contaminants still present in the metal pieces when melted is therefore often very disadvantageous and it is not possible to obtain metal with a high purity, when using known solutions. In addition, contaminants can pose safety risks and reduce the overall quality of the recycled metal.

[0069] For example, copper is a contaminant when recycling metal pieces comprising steel or consisting essentially of steel, like when recycling steel scrap pieces. A recycled steel with lots of copper is normally not desired and of less value than pure steel. According to known methods and systems, there is only a limited possibility to improve the quality of the metal once the metal pieces are in the electric arc furnace for being processed / recycled. For example, to some extent it is be possible to sample 108 the molten metal and to carry out some limited correction measures based on contaminants detected in the samples of the molten metal. For example, heat can be increased to remove lead or lime can be added to bind sulfur. These after-the-fact quality measures cost both time and energy and thereby reduce the production efficiency.

[0070] Higher-value metal need to be of a very specific composition that cannot normally be achieved by mixing variable-content scrap metal pieces.

[0071] The present disclosure improves discussed problems and limitations.

[0072] According to the present disclosure, metal pieces consisting e.g. essentially of steel may still comprise undesired contaminants, such as for example copper parts or pieces, as typically found for example in steel scrap recycling and the quantity of the contaminants is typically not negligible when the metal pieces consist for example essentially of steel when recycling steel scrap metal pieces.

[0073] To better select one or several representative sampling spots on a heterogeneous sample, some embodiments of the present disclosure, synergistically combine a spectroscopy system, in particular a LIBS system, with a camera and a suitable machine learning algorithm.

[0074] The present disclosure provides a better knowledge and selection of the input scrap metal pieces, provides an improved control of the electric arc furnace and allows the processing and recycling of metal with significant higher quality.

[0075] The present disclosure overcomes the limitations of known methods that cannot analyze the metal pieces with sufficient accuracy and precision in a practicable amount of time when processing a large quantity of metal pieces and that do not provide an accurate and precise qualitative and quantitative analysis.

[0076] The discussed problems and limitations of known solutions are overcome by the present disclosure, as described in detail in the following.

[0077] FIG. 2A shows details of a system for processing metal pieces 200 according to embodiments of the present disclosure. FIG. 2B shows further details of a system for processing metal pieces 200 according to embodiments of the present disclosure.

[0078] According to an embodiment, the present disclosure provides a system for processing metal pieces 200, the system comprising:

[0079] - a transportation system 202 configured to transport a plurality of metal pieces 204,

[0080] - an image acquisition system 206 configured to acquire images of metal pieces 204 transported by the transportation system,

[0081] - an image recognition system configured to identify optical parameters of metal pieces based on images acquired by the image acquisition means, for example configured to obtain optical parameters of the metal pieces with machine learning image recognition algorithm 210,

[0082] - a spectroscopy system configured to determine spectra of the metal pieces transported by the transportation system and configured to analyze at least a part of the metal pieces based on the optical parameters identified by the image recognition system,

[0083] - an electric arc furnace 212 configured to melt the metal pieces controlled by an electric arc furnace control system 214 based on control parameters dependent on the determined spectra and / or the optical parameters of the metal pieces.

[0084] In some embodiments, the electric arc furnace control system may be configured to control additives added when melting the metal pieces and / or a power and / or temperature of the electric arc furnace and / or a feed rate of metal pieces melted by the electric arc furnace based on the determined spectra and / or the optical parameters of the metal pieces.

[0085] Thus, according to the present disclosure, a spectroscopy system is synergistically combined with an image recognition system to control the electric arc furnace. The image recognition system can identify similar pieces that likely have similar spectra such that, if one piece is analyzed in detail by the spectroscopy system, other similar pieces as identified by the image recognition system do not need to be analyzed by the spectroscopy system with the same level of detail and may require less sampling spots or no sampling spots.

[0086] In some embodiments, the present disclosure further provides a method for processing metal pieces, the method comprising: - transporting a plurality of metal pieces,

[0087] - acquiring images of the metal pieces,

[0088] - identifying optical parameters of the metal pieces based on the acquired images,

[0089] - determining spectra of the metal pieces using a spectroscopy system by analyzing at least a part of the metal pieces based on optical parameters,

[0090] - melting the metal pieces with an electric arc furnace controlled based on parameters dependent on the spectra and / or the optical parameters of metal pieces.

[0091] In some embodiments, when melting the metal pieces, additives may be added and / or a power and / or a temperature of the electric arc furnace and / or a feed rate of metal pieces melted by the electric arc furnace may be controlled based on the spectra and / or the optical parameters of the metal pieces.

[0092] Methods according to the present disclosure can be carried out using the systems of the present disclosure and features of said systems of the present disclosure are combinable with corresponding features of the methods of the present disclosure and vice versa.

[0093] For example, in some embodiments the optical parameters may include information about the geometry of the metal pieces such that a position and / or a trajectory of the metal pieces can be tracked and such that said information about the geometry and position / trajectory is further integrated with information about the spectra of the metal pieces in a consistent way as provided by the spectroscopy system.

[0094] For example, in some embodiments a model of the metal pieces may be constructed and handled by the image recognition system and the spectroscopy system that consistently stores the optical parameters and / or the spectra of the metal pieces in a consistent manner and allows a tracking of the metal pieces and a corresponding control of the electric arc furnace based e.g. on the spectra and / or the optical parameters as stored in the model consistent with the actually obtained parameters and spectra of the metal pieces.

[0095] For example, the model may also allow an automatic or manual partitioning of the metal pieces such that only metal pieces with predetermined characteristics are melted by the electric arc furnace while remaining metal pieces are automatically sorted out. The model may be processed and stored on a computer of the image recognition system and / or of the spectroscopy system. According to some embodiments, the image acquisition system includes a machine learning image recognition algorithm 210.

[0096] The machine learning algorithm 210 can effectively identify similar metal pieces based on the optical parameters of the metal pieces. Similar metal pieces likely also have similar spectra, because they likely have a common origin in the stream of scrap such that if one metal piece is analyzed by the spectroscopy system with a high number of sampling spots, other similar metal pieces as identified by the machine learning algorithm 210 may require a low number of sampling spots or no sampling spots at all when subsequently analyzed by the spectroscopy system.

[0097] According to some embodiments, the spectroscopy system is a laser-induced breakdown spectroscopy (LIBS) system 208.

[0098] Some embodiments of the present disclosure, synergistically combine a spectroscopy system, such as for example a laser-induced breakdown spectroscopy (LIBS) system 208 with an image acquisition system 206, for example a camera system, for example an optical camera, a structural light camera, a stereo camera, etc.

[0099] The image acquisition system 206 may include at least one camera or may include multiple cameras of various types. The image acquisition system 206 may record images of metal pieces 204, e.g. of a group or of a plurality of metal pieces 204 or of individual metal pieces 204 that may or may not overlap with other metal pieces.

[0100] The metal pieces may be transported by a transportation system 202 configured to transport a plurality of metal pieces such as for example a conveyor belt.

[0101] In some embodiments, a machine learning image recognition algorithm 210 obtains optical parameters of the metal pieces to identify relevant and representative metal pieces and / or heterogeneous metal pieces and / or metal pieces that may be likely anomalous and / or trash that needs to be filtered out or e.g. non uniform pieces that should also be filtered out. For example, if the presence of a piece such as a motor, generator or transformer part is detected that includes or may include copper wires, the piece would be a dangerous piece that would disrupt the quality of the processed metal.

[0102] According to some embodiments, the optical parameters of the metal pieces are obtained with machine learning algorithms and include information about the geometry and the surface of the metal pieces, in particular about one or more of boundaries, textures, colors of the metal pieces, wherein in particular the machine learning algorithms comprise the use of convolutional neural networks or e.g. image recognition algorithms.

[0103] The information about the geometry may also be used to determine a volume and / or a mass of the metal pieces in particular used in an automated control of the electric arc furnace.

[0104] In some embodiments, the images acquired by the image acquisition system 206 may be analyzed by the machine learning image recognition algorithm 210 to identify homogeneous patterns within a heterogeneous group of metal pieces 204 and / or within a single metal piece 204.

[0105] The machine learning image recognition algorithm 210 may identify pieces that likely impact safety or quality during the metal processing, e.g. when recycling steel. For example, the machine learning image recognition algorithm 210 may be based on a convolutional neural network trained for object detection.

[0106] In some embodiments, the machine learning image recognition algorithm 210 may in particular be configured to find dangerous or important trash, such as e.g. trash contaminated with dangerous chemicals and / or precious metals that needs to be properly disposed, either to prevent danger or the reuse the precious metals. For instance, a dangerous chemical could also put the furnace and operators at risk if it would explode when heated and / or it may poison humans and / or the environment or damage the processing plant

[0107] In some embodiments, neural networks or other ML algorithms are configured to separate different objects in the images for a subsequent sampling in order to be able to sample all or most of the materials present, e.g. with the LIBS system.

[0108] In some embodiments, algorithms for image segmentation / edge detection use textures, colors and other information to decide which parts of the image constitute one object or one material.

[0109] In some embodiments, information from both optical parameters, as identified by the image recognition system e.g. based on the machine learning image recognition algorithm 210, and spectra, as determined by the spectroscopy system, e.g. using the LIBS spectroscopy system 208, is used as input for the control of an electric arc furnace 212. Therefore, the electric arc furnace 212 is configured to melt the metal pieces controlled by the electric arc furnace control system 214 based on control parameters dependent on the determined spectra and / or the optical parameters of the metal pieces.

[0110] For example, the electric arc furnace may be controlled with control parameters that comprise a power delivered to the electric arc furnace, a feed rate of the metal pieces and / or a total volume or mass of metal pieces transferred into the electric arc furnace in total or in a given time interval. Alternatively, or in addition, the electric arc furnace may be controlled adding additives to compensate for the presence of contaminants. For example, it is possible to add lime to bind sulfur.

[0111] In some embodiments, the control parameters may be determined based at least in part on information about a volume and a composition of each of the metal pieces obtained by combining the optical parameters with the spectra, for example based on a model of the metal pieces that includes the optical parameters and / or the spectra of the metal pieces, in particular wherein the optical parameters provide information about a volume of the metal scrap pieces and the spectra provide information about a composition of the metal scrap pieces. For example, the optical parameters as for example stored in the model, may allow a determination of a position and / or trajectory of the metal pieces that are then (e.g. automatically) sorted based on the optical parameters and / or the spectra and transferred into the electric arc furnace that is further controlled based on the optical parameters and / or the spectra as stored in the model for those pieces transferred to the furnace.

[0112] Therefore, in some embodiments, images from a camera are processed to determine input parameters of a LIBS sensor and data from both sensors is used to control the electric arc furnace, for example using a computer model to store and process the optical parameters and the spectra.

[0113] In some embodiments, a LIBS system is configured to shoot a tightly focused laser pulse in the range of tens of millijoules at some material. Small amounts of the material are ablated, and a plasma forms that is ignited by the laser energy. With the help of a spectrometer of the spectroscopy system that collects the light of the plasma emission, spectra of the metal pieces are determined and an elemental composition of the metal pieces can be determined accordingly. Depending on the material, different spectra of the metal pieces are observed.

[0114] FIG. 3 shows known sampling patterns of a laser induced breakdown spectroscopy (LIBS) system. According to know solutions, a laser beam of the LIBS system is moved across an aluminum scrap-stream in a zig-zag pattern and the full aluminum waste-stream must be covered thoroughly.

[0115] Therefore, known solutions require a very high number of sampling spots 302, typically in a zig-zag pattern over the scrap pieces, and known LIBS systems are therefore not used for a large quantity of metal pieces. In particular, analyzing large quantities of steel scrap pieces is not possible based on known LIBS systems.

[0116] FIG. 4 shows sampling patterns of a laser induced breakdown spectroscopy system according to embodiments. According to the present solution, that combines the LIBS system with the image acquisition system and image recognition system, a significantly lower number of sampling spots 302 is required, without affecting the final quality of the processed metal.

[0117] As illustrated, the LIBS spectroscopy system 208 is configured and / or used to analyze only relevant and representative spots on the samples as determined by the machine learning image recognition algorithm 210 according to embodiments described herein. This drastically reduces the effort of the analysis process and thereby enables an integration into many recycling or sorting processes such as for example, scrap-based metal making / recycling process.

[0118] In some embodiments, the speed and efficiency of a LIBS-based analysis are significantly improved, by reducing the required number of sampling spots 302 needed to analyze the metal pieces 204 in view of the additional consideration of the optical parameters as identified by the image recognition system, e.g. based on the machine learning image recognition algorithm 210 that may already partition the metal pieces into blocks that are sufficiently uniform.

[0119] As shown in FIG. 4, for a metal piece that has similar optical parameters (e.g. a similar form and / or surface) compared to another metal piece that was already sampled / analyzed with a high number of sampling spots 302, it is not necessary to have a high number of sampling spots also for the metal piece that may therefore require only a low number of sampling spots and / or that needs not to be analyzed by the LIBS system (the spectra of the piece may then be assumed to be as the spectra of another similar piece that was already sampled / analyzed). In fact, optically similar pieces often have a common origin and composition.

[0120] Therefore, for example, the sampling spots per piece is adapted based on the quantity of sampling spots already analyzed by the LIBS system for similar pieces, with a similarity detected by the image recognition algorithm / system with the use of machine learning. For example, the adaption may be an adaption of the number of sampling spots per piece subsequently analyzed, that may be lower than a higher number of sampling spots per piece for already previously analyzed metal pieces with similar optical parameters, wherein the similarity is determined by the image recognition algorithm / system with the use of machine learning.

[0121] For example, the lower number of sampling spots is lower than the higher number of sampling spots. In particular, the lower number of sampling spots may be zero.

[0122] In some embodiments, Alternatively or in addition, the low number of sampling spots and / or the high number of sampling spots may be further adapted in view of a surface measure of the metal piece, for example adjusting a surface density of the sampling spots.

[0123] In some embodiments, alternatively or in addition, for each metal piece the spectra are determined such that:

[0124] - if another metal piece with similar optical parameters compared to the metal piece was not sampled before by the laser induced breakdown spectroscopy system, the metal piece is sampled by the laser induced breakdown spectroscopy system with a high number of sampling spots per surface unit,

[0125] - and / or if another metal piece with similar optical parameters compared to the metal piece was sampled before by the laser induced breakdown spectroscopy system, the metal piece is sampled with a low number of sampling spots per surface unit or is not sampled and at least one spectrum of the metal piece is assumed to be identical to at least one spectrum of the another metal piece with similar optical parameters that was sampled before; wherein the low number of sampling spots per surface unit is lower than the high number of sampling spots per surface unit.

[0126] The similarity of the optical parameters may be evaluated or determined by machine learning.

[0127] In some embodiments, a multi-sensor system including one or more cameras and one or more LIBS sensors, is configured to select the sampling spots for the LIBS sensors based on the camera images by an image recognition algorithm e.g., a machine-learning algorithm.

[0128] In some embodiments, the camera system may include various image sensors, including for example RGB cameras, and / or IR cameras, and / or multispectral cameras and / or structural light cameras and / or stereo cameras. In some embodiments, the LIBS system may use mechano-optical components like moveable mirrors and lenses to direct the laser beam to the determined sampling points. The camera system and the laser system may be synchronized regarding time and position and / or they may use markers as reference points.

[0129] In some embodiments, the plurality of metal pieces may be transported on a transportation system 202 such as for example a conveyor belt, a chute or at another suitable location near the scrap-stream.

[0130] In some embodiments, the image recognition algorithm / system 210 identifies single scrap pieces in the scrap-stream. The algorithm / system may furthermore identify duplicates of previous metal pieces, known contaminants (e.g., gas cylinder, copper winding, copper pipe, ...) or other properties of the metal pieces (e.g., size, coatings, or welds). Based on the identification, the algorithm may determine suitable sampling points for the LIBS analysis. The algorithm may be configured in different ways, for example it may be configured to select one sampling spot per piece or it may be configured to only select sampling spots on suspicious scrap pieces (e.g., pieces that likely contain copper or lead), as identified by the image recognition system / algorithm.

[0131] Some embodiments, therefore overcome the limitations of known LIBS systems that have no knowledge of the scrap-stream and are configured to sample in a defined sampling pattern with the scrap pieces that need to be aligned carefully and / or may be sampled multiple times, producing erroneous results, e.g. when controlling the furnace, or resulting in unnecessary time consumption for the analysis.

[0132] Systems and methods of some embodiments, enable a LIBS analysis with significantly less sampling points than in previous LIBS systems.

[0133] According to some embodiments, a much higher throughput can therefore be achieved, making the analysis suitable for metal recycling where 50-200t of scrap may be charged into an oven per hour. Known LIBS systems (e.g., for aluminum) have a throughput of only a few tons per hour and are accordingly, not suitable for high throughput metal scrap.

[0134] The sampling patterns of FIG. 4 of a laser induced breakdown spectroscopy system according to embodiments show an advantage of the present disclosure. The number of sampling points can typically be decreased by a factor of 16 by using an individually determined sampling pattern, compared to the know sampling patterns of FIG. 3 and without affecting the quality of the processed metal.

[0135] In some embodiments, based on the optical parameters, identical or analogous pieces of scrap can be identified and a duplicate sampling of these can be avoided. Some other duplicate pieces may not be clearly identified, that may then be specifically selected for further analysis by the LIBS system. The speed with which the laser spot of the LIBS system can be moved across the stream of scrap may be further taken into account to increase the efficiency of the sampling pattern.

[0136] In some embodiments, the spectroscopy system is a laser induced breakdown spectroscopy system configured to determine spectra of metal pieces by sampling the metal pieces, the system in particular further configured such that for each metal piece the spectra are determined such that:

[0137] - if another metal piece with similar optical parameters compared to the metal piece was not sampled before by the laser induced breakdown spectroscopy system, the metal piece is sampled by the laser induced breakdown spectroscopy system with a high number of sampling spots,

[0138] - and / or if another metal piece with similar optical parameters compared to the metal piece was sampled before by the laser induced breakdown spectroscopy system, the metal piece is sampled with a low number of sampling spots or is not sampled, and at least one spectrum of the metal piece is assumed to be identical to at least one spectrum of the another metal piece with similar optical parameters that was sampled before; wherein the low number of sampling spots is lower than the high number of sampling spots.

[0139] In some embodiments, when a metal piece is not sampled, the LIBS sampling spots for the metal piece that is not sampled is zero.

[0140] In some embodiments of systems and methods according to the present disclosure, the optical parameters and / or the spectra of the metal pieces are further analyzed by the image recognition system and / or the spectroscopy system to determine non-homogeneous pieces, i.e. pieces that are composed by a varying type of material, for example pieces having different spectra in different parts that differ significantly or that may contain a plurality of materials according to varying optical parameters. In some embodiments, if a non-homogeneous piece is detected, the electric arc furnace may be configured to melt only homogeneous pieces (i.e. the non-homogeneous pieces are removed from the metal pieces).

[0141] In some embodiments, the number of sampling spots per piece and / or per unit surface of a metal piece may be adjusted based on the detected homogeneity or non-homogeneity of the piece.

[0142] The advantages of the solutions of some embodiments of the present disclosure become apparent when comparting the sampling patterns of a laser induced breakdown spectroscopy system according to embodiments as shown in FIG. 4, with known sampling patterns of a laser induced breakdown spectroscopy system as shown in FIG. 3.

[0143] Corresponding features and benefits are also combinable and obtainable for the analogous methods of the present disclosure.

[0144] Without a significant reduction of the quality of processed metal, system and methods of some embodiments of the present disclosure may provide at least one order of magnitude increase of the throughput of pieces analyzed by the LIBS system, for example a more than tenfold increase, for example a 16x increase of the throughput, further providing the possibility of estimating elemental composition by weight based on the piece and shape recognition of the metal pieces, the avoidance of a repeated analysis and sampling of at least some identical pieces that are identified by the image recognition system and not analyzed twice by the spectroscopy system. According to some embodiments of the present disclosure it is also possible to prevent that small pieces slip through the sampling, thereby preventing a contamination that said small pieces may still significantly cause in known solutions.

[0145] In some embodiments, the electric arc furnace configured to melt the metal pieces may be controlled by an electric arc furnace control system based on control parameters dependent on the determined spectra and / or the optical parameters of the metal pieces, for example as stored and processed in a computer model.

[0146] According to embodiments of the present disclosure, the information from both visual and elemental analysis may therefore be used to predict properties of the molten metal in an electric arc furnace and may furthermore be used to control such a furnace, controlling e.g., power, feed-rate, additives. For example, in some embodiments, a geometric analysis based on the images can determine the volume of each metal piece and the LIBS spectroscopy system may determine the elemental composition. In combination, the amount (weight) of each element in the metal pieces can be determined. The information may also be used to remove pieces of scrap from the stream if they are identified as a risk regarding safety or quality.

[0147] In particular for steel scrap processing, the present disclosure leverages the benefits of using a LIBS system, without requiring a prohibitive amount of time, given that e.g. a combined use of the LIBS system with the image acquisition system and the image recognition system allows an improved control without requiring a prohibitive number of sampling spots when using the LIBS system alone.

[0148] Some embodiments of the present disclosure also describe a detection system configured to identify non-anomalous metal pieces based on spectra of the metal pieces.

[0149] The detection system according to embodiments of the present disclosure may be used alone or in combination with other embodiments.

[0150] Some embodiments of the present disclosure also provide a detection system configured to identify non-anomalous metal pieces based on the spectra of the metal pieces and optionally also on optical parameters of the metal pieces, wherein the identification of the non-anomalous metal pieces is based on a normal zone of spectra of substantially uncontaminated metal pieces, and the spectra are measured by a laser induced breakdown spectroscopy system, such as for example, the LIBS system for processing metal pieces as previously described; and wherein an electric arc furnace is configured to melt only the non-anomalous metal pieces.

[0151] According to the present disclosure, substantially uncontaminated metal pieces correspond to metal pieces with an acceptable or desired composition of the metal in view of the desired quality of the finally processed steel.

[0152] Therefore, for substantially uncontaminated metal, for example for substantially uncontaminated steel, it has to be intended that a plurality of chemical elements can still be present in the metal, for example copper traces can be still present in the substantially uncontaminated steel, but in a maximal quantity or density that is still allowable in view of the desired final quality of the processed metal.

[0153] According to the present disclosure, substantially uncontaminated does therefore, not refer to a minimal technical tolerance possible according to measurement methods, but to a tolerance still allowable for the metal that is finally processed, e.g. to the recycled steel.

[0154] In some embodiments, the image acquisition system and the spectroscopy system may provide a model of the metal pieces providing information about a geometry and spectrum of the metal pieces that may include a position or trajectory of the metal pieces and the detection system further identifies the non-anomalous metal pieces such that anomalous metal pieces (i.e. non non-anomalous metal pieces) can be sorted out, manually or automatically. In some embodiments, only the non-anomalous metal pieces are accordingly melted by the electric arc furnace, based on the identification of non-anomalous metal pieces carried out by the detection system based on the spectra and optionally, the optical parameters as stored in the computer model.

[0155] The detection system of some embodiments can be combined with embodiments of the present disclosure or form a separate system and embodiment to separately identify non-anomalous metal pieces.

[0156] Some embodiments of the present disclosure, also provides a method for identifying non- anomalous metal pieces based on spectra of metal pieces and optionally on optical parameters of the metal pieces, based on a normal zone of spectra of substantially uncontaminated metal pieces, the spectra measured for example using the laser induced breakdown spectroscopy system as previously described; the method further comprising melting only the non-anomalous metal pieces when melting metal pieces with the electric arc furnace.

[0157] The method for identifying non-anomalous metal pieces may be carried out as a standalone method or may be used for identifying non-anomalous metal pieces within the methods for processing metal pieces as presented in the present disclosure.

[0158] Methods of the present disclosure may use the systems of the present disclosure and features of said systems can be combined with the corresponding methods. FIG. 5 shows details of a detection system 502 configured to identify non-anomalous metal pieces according to embodiments of the present disclosure.

[0159] The detection system 502 allows a classification of metal pieces in non-anomalous 506 metal pieces and anomalous 508 metal pieces.

[0160] The detection system may be used for identifying non-anomalous metal pieces according to methods of the present disclosure and as used in methods of the present disclosure.

[0161] According to the present disclosure “non-anomalous” does not necessarily mean “substantially uncontaminated”, i.e. non-anomalous metal pieces may have a quantity or concentration of contaminants that is greater or equal than in substantially uncontaminated metal pieces, but that is considered still allowable for the particular metal pieces that are actually processed, in particular, still allowable in view of the quality of the actually processed steel obtained by melting the metal pieces.

[0162] In recycling applications (for example when processing metal) the ability to distinguish the exact composition of the feeding material is of vital importance. Today, LIBS results can be interpreted using machine learning classification algorithms or by manually providing rules to interpret the results. However, due to environmental factors, it is difficult to design a universal identification model.

[0163] Processing of metal pieces was already shown in FIG. 1.

[0164] In particular scrap is shipped 102, and several different grades of scrap exist, with upper limits on other metals such as copper or lead.

[0165] Today, the inspection of scrap is done by humans or at best with LIBS sensors that have patterns stored for specific materials. Human inspection can only be superficial at best. Automated solutions based on LIBS need to be highly standardized to work and any customization would be highly expensive and time consuming when based on the current state of the art.

[0166] Contaminants can pose safety risks during melting and reduce the overall quality of the recycled metal. For example, higher-value steels need to be of a very specific composition that cannot normally be achieved by mixing variable-content scrap. A better selection would provide metal of a better quality. Some embodiments of the present disclosure, combine LIBS with machine learning anomaly detection algorithms to detect contaminants and even inferior alloys in metal scrap.

[0167] In some embodiments, an anomaly detection system uses anomaly detection algorithms, e.g., Empirical-Cumulative-distribution-based Outlier Detection and / or Deep Support Vector Data Description, and / or Gaussian Mixture Models, trained on samples of the desired pure materials under the circumstances found in a particular installation, e.g. using a particular LIBS system.

[0168] Unlike classification or regression algorithms, the training of the anomaly detection algorithms used by the anomaly detection system can be done cost-effectively and time-effectively during commissioning of the installation.

[0169] In some embodiments, after training of the detection system 502 configured to identify non- anomalous metal pieces, the LIBS spectroscopy system 208 may be exposed to the metal pieces 204 in the scrap stream and sample them. The resulting spectrogram is passed on to the system 502 configured to identify non-anomalous metal pieces.

[0170] In some embodiments, the anomaly detection system / algorithm is trained to decide whether the metal pieces are of a non-anomalous material composition and / or to filter out anomalous (i.e. non non-anomalous) metal pieces. In other words, non-anomalous can be defined as regular / normal with respect to the desired quality of the finally processed metal.

[0171] In some embodiments, the detection system 502 configured to identify non-anomalous metal pieces is therefore able to distinguish non-anomalous 506 metal pieces from anomalous 508 metal pieces. Contaminated metal pieces may be anomalous 508 metal pieces.

[0172] In some embodiments, the detection system 502 configured to identify non-anomalous metal pieces may receive as an input a probability threshold 504 to discriminate between anomalous 508 and non-anomalous 506 metal pieces.

[0173] In some embodiments, the probability threshold 504 may be based on spectra of known substantially uncontaminated metal pieces and may be automatically determined by the detection system 502. For example, said probability threshold may in some embodiments be automatically determined based on a dispersion of the known spectra or may be alternatively determined or adjusted manually and / or in function of the (desired) quality of the metal after melting, for example in function of a measured or allowable contamination of the final metal obtained from known pure metal pieces and / or with respect to an allowable tolerance. The threshold may result from a maximal allowable deviation of the processed metal from the uncontaminated metal, e.g. the threshold resulting from a maximal allowable contamination of the processed metal. The threshold may be calculated automatically by a computer or by a human in function of the maximal allowable contamination.

[0174] In some embodiments, the detection system 502 configured to identify non-anomalous metal pieces may identify a normal zone of spectra of substantially uncontaminated metal pieces, the spectra measured for example, by the laser induced breakdown spectroscopy system of the system for processing metal pieces according to embodiments of the present disclosure.

[0175] In some embodiments, the normal zone is obtained by an anomaly detection system based on machine learning, in particular on algorithms based on Gaussian Mixture Models and / or Empirical-Cumulative-distribution-based Outlier Detection and / or Deep Support Vector Data Description.

[0176] In some embodiments, the detection system 502 configured to identify non-anomalous metal pieces may implement a machine learning anomaly detection algorithm and may include a probability threshold 504. Typical output of the algorithm can be interpreted as, or translated into, a probability that a metal piece is non-anomalous. The identification of non-anomalous metal pieces may be based on a threshold of acceptability that may be defined by an operator as a suitable probability threshold to achieve a desired purity of processed metal. A high threshold may result in a high amount of rejected material, erring on the side of caution. A low threshold results in a higher risk of producing the occasional batch of lower quality metal. The operator may therefore set the probability threshold 504 of the detection system 502 based on experimental results in function of the obtained metal. In some embodiments, the threshold may be automatically determined and / or be based on a heuristic in function of the quality of the metal after melting.

[0177] A LIBS system shoots a tightly focused laser pulse in the range of tens of millijoules at some material. Small amounts of material will be ablated and a plasma is ignited by the laser energy. With the help of a spectrometer that collects the light of the plasma emission, the elemental composition of the material can be determined. Depending on the material, different spectra are observed.

[0178] FIG. 6 shows a known spectrum 602 of a steel alloy. The spectrum 602 is a theoretical and ideal spectrum and does not account for variations and / or errors that occur when a spectrum is actually obtained by a particular LIBS system.

[0179] In theory, by comparing the intensities recorded at given wavelengths with the known elemental spectra, such as the known spectrum 602 of a steel alloy, it should be possible to determine the elemental composition of metal pieces. However, when comparing to an actual recorded spectrum of the same material from the literature, e.g. when comparing a spectrum as obtained by a given LIBS system to the known spectrum 602, it is hard for an observer to see actual similarities and / or clear differences, due to several factors such as technical specifications of the laser, environmental factors, and sample morphology that may cause a significant variance / differences in the actually observed spectra obtained with a given LIBS system.

[0180] As a result, any machine learning system must be tailored to a specific setup and specific LIBS system and use case of a given LIBS system.

[0181] FIG. 7 shows a spectrum of steel alloy as measured according to embodiments of the present disclosure.

[0182] The spectrum 702 of steel alloy as measured by an actual LIBS system, for example, by the LIBS spectroscopy system 208, which typically differs from the theoretical spectrum 602 of steel alloy.

[0183] In fact, the spectrum 702 of steel alloy as measured by an actual LIBS system differs from the theoretically known spectrum 602 of steel alloy, due to several factors, such as technical specifications of the laser, environmental factors, and sample morphology that may cause a significant variance / differences when an actual LIBS system is used, like the LIBS spectroscopy system 208.

[0184] The spectrum 702 of high alloy steel at 20mJ shows ionized iron, nickel and chromium, but with significant difference from the theoretically known spectrum 602 of steel alloy.

[0185] The machine learning algorithms using LIBS data encountered today are classification or regression algorithms.

[0186] A classification algorithm can identify specific materials, e.g. copper or specific steel alloys. For example, a classification algorithm may identify known spectra, such as specific steel alloys, for example a steel with composition Fe 70%, Ni 20%, Cr 10% as illustrated in FIG. 6 and distinguish them from other known materials as learned during a training phase of the classification algorithm, such as for example from another known spectrum of copper.

[0187] However, classification algorithms are not good at dealing with classes not seen before (e.g. aluminum) and give erroneous classification for those cases. In contrast, anomaly detection according to embodiments of the present disclosure is good at detecting such deviations, recognizing a deviation from normal known spectra. This is ideal for contaminant detection as is needed for recycling applications.

[0188] A regression algorithm can return the relative elemental composition of the sample. However, during training, it needs to have seen many different compositions to work properly. A working algorithm for highly contaminated streams as found in recycling is not economically feasible to train and would require a training that consumes a prohibitive amount of time and resources.

[0189] FIG. 8 shows a normal zone 810 of spectra 802 (represented in an n-dimensional space) of substantially uncontaminated metal pieces, e.g. steel pieces, according to embodiments.

[0190] In some embodiments, substantially uncontaminated metal pieces are used for training of the detection system 502 configured to identify non-anomalous metal pieces based on machine learning anomaly detection algorithms.

[0191] In some embodiments, the spectra 802 (or a suitable transformation thereof in an n-dimensional space) of said substantially uncontaminated metal pieces are shown in FIG. 8 as a white circle in a multidimensional space, representing parameters of the spectra and form training samples and are encapsulated into a hypersphere determining the normal zone 810.

[0192] In some embodiments, when new spectra 804 (or a suitable transformation thereof) of metal pieces analyzed by the LIBS system are analyzed, as indicated with black circles in FIG. 8, the spectra of the analyzed metal pieces (or a suitable transformation thereof) will either be within the hypersphere forming the normal zone 810 or not. The (transformed) spectra that are outside are considered anomalies, and those within the normal zone are considered non-anomalous. The metal pieces may then accordingly be partitioned into non-anomalous or anomalous pieces.

[0193] The spectra generated by LIBS can be seen as points in an n-dimensional space, for a suitable dimension n that captures the relevant properties of the spectra and / or filters out noise. In some embodiments, by defining a region of normality from the training data, it is possible to identify contaminants as they lie outside the normal zone 810. Additional transformation may operate on the spectra before the normal zone is defined, e.g. to filter out noise and / or to smooth the spectrum and / or to extract a power density or the like.

[0194] FIG. 9 shows a comparison between anomaly detection according to embodiments of the present disclosure and a known classification.

[0195] FIG. 9 also explains, why classification requires more training data then anomaly detection. In a classification, samples are needed for each category that is classified and not just the main class (for example steel). Furthermore, any element of a category not considered during a training of a classification algorithm will be classified in one of the categories present during training, i.e. a wrong classification would occur that is conversely prevented by the anomaly detection algorithms according to embodiments of the present disclosure.

[0196] A classification 920 algorithm is shown that would for example, classify steel samples as “steel” or “copper” based on training samples 902 that are either steel or copper spectra, i.e. only the two categories “steel” and “copper” are shown for simplicity in the shown example, but the conclusion extend e.g. also to an arbitrary finite number of categories.

[0197] For example, spectra of metal pieces 904 are submitted for classification and we assume the presence of an outlier W that may be highly contaminated steel and thereby forming an outlier W or may be for example of a category not seen during training.

[0198] The classifier 920 would possibly erroneously consider the outlier W as known and acceptable steel, since the outlier W would not be recognized as copper, falling outside of the “copper zone”. An erroneous classification would therefore occur, according to the shown example.

[0199] Conversely the normal zone 810 according to embodiments of the present disclosure provides an anomaly detection 910 that is specifically tailored to the spectra 802 of substantially uncontaminated metal pieces as seen during training of the anomaly detection system / algorithm and is not influenced by other categories not seen in training. As a consequence, according to embodiments of the present disclosure, a tight discrimination is possible and an outlier would be identified as anomalous, i.e. as not non-anomalous, thereby significantly improving the results obtainable with a classification 920. 1 Alternatively, the detection system configured to identify non-anomalous metal pieces may be present independently of other features described herein.

[0200] Alternatively, or in addition, identifying non-anomalous metal pieces may be carried out independently of other steps described herein.

[0201] Therefore, the present disclosure, independently form other embodiments, also provides further embodiments of systems and methods described in the following that may or may not be combined with embodiments of methods and systems previously described:

[0202] A system for processing metal pieces, the system comprising:

[0203] -a transportation system configured to transport a plurality of metal pieces,

[0204] -a spectroscopy system configured to determine spectra of the metal pieces transported by the transportation system and configured to analyze at least a part of the metal pieces,

[0205] -an electric arc furnace configured to melt the metal pieces controlled by an electric arc furnace control system based on control parameters dependent on the determined spectra; wherein the system further comprises a detection system configured to identify non-anomalous metal pieces based on the spectra of the metal pieces, wherein the identification of the non-anomalous metal pieces is based on a normal zone of spectra of substantially uncontaminated metal pieces, the spectra measured by the spectroscopy system of the system for processing metal pieces; and wherein the electric arc furnace is configured to melt only the non-anomalous metal pieces.

[0206] For example, the electric arc furnace control system may be configured to control additives added when melting the metal pieces and / or a power and / or a temperature of the electric arc furnace and / or a feed rate of metal pieces melted by the electric arc furnace based on the determined spectra.

[0207] In some embodiments, the system is configured such that the normal zone is obtained by an anomaly detection system based on machine learning, in particular on algorithms based on Gaussian Mixture Models and / or Empirical-Cumulative-distribution-based Outlier Detection and / or Deep Support Vector Data Description. In some embodiments, the spectroscopy system is a laser induced breakdown spectroscopy system configured to determine spectra of metal pieces by sampling the metal pieces.

[0208] Therefore, the present disclosure independently from other embodiments, also provides methods described in the following that may or may not be combined with embodiments of methods and systems previously described:

[0209] A method for processing metal pieces, the method comprising:

[0210] -transporting a plurality of metal pieces,

[0211] -determining spectra of the metal pieces using a spectroscopy system,

[0212] -melting the metal pieces with an electric arc furnace controlled based on parameters dependent on the spectra; wherein the method further comprises:

[0213] -identifying non-anomalous metal pieces based on the spectra of the metal pieces, based on a normal zone of spectra of substantially uncontaminated metal pieces, the spectra measured using the spectroscopy system; melting only the non-anomalous metal pieces when melting metal pieces with the electric arc furnace.

[0214] For example, when melting the metal pieces, additives may be added and / or a power and / or a temperature of the electric arc furnace and / or a feed rate of metal pieces melted by the electric arc furnace may be controlled based on the spectra of the metal pieces;

[0215] In some embodiments, the normal zone is obtained using an anomaly detection system based on machine learning, in particular on algorithms based on Gaussian Mixture Models and / or Empirical-Cumulative-distribution-based Outlier Detection and / or Deep Support Vector Data Description

[0216] In some embodiments, the spectroscopy system is a laser induced breakdown spectroscopy system configured to determine spectra of metal pieces by sampling the metal pieces; In some embodiments, the method further comprising identifying dangerous trash and excluding the dangerous trash from the non-anomalous metal pieces.

[0217] In some embodiments, which can be combined with other embodiments of methods and systems disclosed herein, the metal is steel and / or aluminum and / or copper and / or lead and / or a metal alloy. In particular, the metal comprises steel; more in particular the metal consists essentially of steel, and even more in particular the metal is steel.

Claims

CLAIMS1. A system for processing metal pieces, the system comprising:- a transportation system configured to transport a plurality of metal pieces,- an image acquisition system configured to acquire images of metal pieces transported by the transportation system,- an image recognition system configured to identify optical parameters of metal pieces based on images acquired by the image acquisition means,- a spectroscopy system configured to determine spectra of the metal pieces transported by the transportation system and configured to analyze at least a part of the metal pieces based on the optical parameters identified by the image recognition system,- an electric arc furnace configured to melt the metal pieces controlled by an electric arc furnace control system based on control parameters dependent on the determined spectra and / or the optical parameters of the metal pieces.

2. The system of claim 1, further comprising a detection system configured to identify non- anomalous metal pieces based on the spectra of the metal pieces and / or the optical parameters of the metal pieces, wherein the identification of the non-anomalous metal pieces is based on a normal zone of spectra of substantially uncontaminated metal pieces, the spectra measured by the spectroscopy system of the system for processing metal pieces; and wherein the electric arc furnace is configured to melt only the non-anomalous metal pieces.

3. The system of claim 2, further configured such that the normal zone is obtained by an anomaly detection system based on machine learning, in particular on algorithms based on Gaussian Mixture Models and / or Empirical-Cumulative-distribution-based Outlier Detection and / or Deep Support Vector Data Description.

4. The system of any of claims from 1 to 3, further configured such that the optical parameters of the metal pieces are obtained with machine learning algorithms and include information about the geometry and the surface of the metal pieces, in particular about one or more of boundaries, textures, colors of the metal pieces, wherein in particular, the machine learning algorithms comprise the use of convolutional neural networks and / or image recognition algorithms.

5. The system of any of claims from 1 to 4, wherein the spectroscopy system is a laser induced breakdown spectroscopy system configured to determine spectra of metal pieces by sampling the metal pieces, the system in particular further configured such that for each metal piece the spectra are determined such that:- if another metal piece with similar optical parameters compared to the metal piece was not sampled before by the laser induced breakdown spectroscopy system, the metal piece is sampled by the laser induced breakdown spectroscopy system with a high number of sampling spots;- and / or if another metal piece with similar optical parameters compared to the metal piece was sampled before by the laser induced breakdown spectroscopy system, the metal piece is sampled with a low number of sampling spots or is not sampled and at least one spectrum of the metal piece is assumed to be identical to at least one spectrum of the another metal piece with similar optical parameters that was sampled before; wherein the low number of sampling spots is lower than the high number of sampling spots.

6. The system of claim 5, further configured such that the similarity of the optical parameters is evaluated by machine learning.

7. The system of any of claims from 2 to 6, further configured such that the image recognition system is further configured to identify dangerous trash and such that the non-anomalous metal pieces do not include dangerous trash.

8. A method for processing metal pieces, the method comprising:- transporting a plurality of metal pieces,- acquiring images of the metal pieces,- identifying optical parameters of the metal pieces based on the acquired images,- determining spectra of the metal pieces using a spectroscopy system by analyzing at least a part of the metal pieces based on optical parameters,- melting the metal pieces with an electric arc furnace controlled based on parameters dependent on the spectra and / or the optical parameters of metal pieces.

9. The method of claim 8, further comprising- identifying non-anomalous metal pieces based on the spectra of the metal pieces and / or the optical parameters of the metal pieces, based on a normal zone of spectra of substantially uncontaminated metal pieces, the spectra measured using the spectroscopy system; melting only the non-anomalous metal pieces when melting metal pieces with the electric arc furnace.

10. The method of claim 9, wherein the normal zone is obtained using an anomaly detection system based on machine learning, in particular on algorithms based on Gaussian Mixture Models and / or Empirical-Cumulative-distribution-based Outlier Detection and / or Deep Support Vector Data Description.

11. The method of any of claims from 8 to 10, wherein the optical parameters of the metal pieces are obtained using machine learning algorithms and include information about the geometry and the surface of the metal pieces, in particular about one or more of boundaries,textures, colors of the metal pieces, wherein in particular, the machine learning algorithms comprise the use of convolutional neural networks and / or image recognition algorithms.

12. The method of any of claims from 8 to 11, wherein the spectroscopy system is a laser induced breakdown spectroscopy system configured to determine spectra of metal pieces by sampling the metal pieces; wherein in particular for each metal piece determining spectra of the metal piece comprises:- if another metal piece with similar optical parameters compared to the metal piece was not sampled before with the laser induced breakdown spectroscopy system, sampling the metal piece with the laser induced breakdown spectroscopy system with a high number of sampling spots,- and / or if another metal piece with similar optical parameters compared to the metal piece was sampled before with the laser induced breakdown spectroscopy system, sampling the metal piece with a low number of sampling spots or not sampling the metal piece and assuming at least one spectrum of the metal piece as identical to at least one spectrum of the another metal piece with similar optical parameters that was sampled before; wherein the low number of sampling spots is lower than the high number of sampling spots.

13. The method of claim 12, wherein the similarity of the optical parameters is evaluated by machine learning.

14. The method of any of claims from 8 to 13, further comprising identifying dangerous trash and excluding the dangerous trash from the non-anomalous metal pieces.

15. The system according to any of claims from 1 to 7 or the method of any of claims from 8 to 14, wherein the metal is steel and / or aluminum and / or copper and / or lead and / or a metalalloy, in particular wherein the metal comprises steel, more in particular wherein the metal consists essentially of steel, more in particular wherein the metal is steel.

Images