Processing method, selecting device, and crushed piece

Abstract

To more easily perform selection of a crushed piece of a metal product on a part-by-part basis, and in turn prevent the generation of waste.SOLUTION: A processing method includes: marking every part of a metal product comprising a plurality of parts; crushing the marked metal product to obtain a first crushed piece; performing classification as to which of the plurality of parts the first crushed piece is on the basis of the marking; and generating, from information on a second crushed piece of the metal product crushed without performing marking, training data for training a machine learning model that outputs a result of classification as to which of the plurality of parts the second crushed piece is, on the basis of a result of the classification and information on the first crushed piece.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a processing method, a sorting device, and fragments. [Background technology]

[0002] Conventionally, metal recycling has been known to involve crushing metal products such as automobiles and sorting the resulting mixture according to their intended use. Patent Document 1 discloses a technique for identifying wrinkles from the surface information of the crushed pieces and sorting them according to the content of added metals contained in the crushed metal pieces based on the proportion of wrinkles on the surface. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6726753 Summary of the Invention [Problem to be solved by the invention]

[0004] When trying to separate the crushed pieces into product parts by material, it is necessary to perform processes such as dismantling the pieces into individual materials before crushing them, or analyzing the components of each crushed piece, which is a time-consuming process.

[0005] Therefore, an object of the present invention is to make it possible to more easily sort the crushed pieces of metal products into parts, thereby preventing the generation of waste. [Means for solving the problem]

[0006] According to the present invention, a marking step of marking a metal product made up of a plurality of parts on a part-by-part basis; a crushing step of crushing the marked metal product into first crushed pieces; a classification step of classifying the first crushed fragment as one of the plurality of parts based on the marking; a generation step of generating training data for training a machine learning model that outputs a classification result of which of the plurality of parts the second fragments are from information about second fragments of the metal product crushed without marking, based on the classification result and information about the first fragments; A processing method is provided, comprising: [Effects of the Invention]

[0007] According to the present invention, it is possible to more easily separate the crushed pieces of metal products into parts, thereby preventing the generation of waste. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram showing an example of a sorting system including a sorting device according to an embodiment. [Figure 2] 10 is a flowchart illustrating an example of a machine learning model generation process according to the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, the embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the scope of the invention as claimed, and not all combinations of features described in the embodiments are necessarily essential to the invention. Two or more of the features described in the embodiments may be combined in any desired manner. Furthermore, the same reference numerals are used to designate identical or similar components, and redundant descriptions will be omitted.

[0010] When sorting fragments of shredded metal products, it is possible to use a sorting process using a machine learning model. When sorting using a machine learning model, training data on the objects to be sorted must be prepared in order to train the machine learning model. In the learning stage of a machine learning model that sorts using captured images of the fragments, labels indicating the correct answers for the captured fragments are prepared in advance. However, there are usually a large number of fragments, and it is often difficult to label the fragments as parts of the metal product based only on information about the shape and appearance of the fragments.

[0011] In a processing method according to one embodiment of the present invention, a metal product made up of multiple parts is marked on a part-by-part basis, and the marked metal product is crushed into first fragments. Next, the first fragments are classified as one of the multiple parts based on the markings, and training data is generated based on the classification results and information about the first fragments to train a machine learning model that outputs a classification result as to which of the multiple parts the second fragments are, based on information about second fragments of a metal product crushed without marking.

[0012] The processing method according to this embodiment is used to generate training data for a machine learning model for selectively recovering (sorting) fragments of metal products. First, a sorting device for recovering such metal products will be described below.

[0013] Fig. 1 is a schematic diagram showing an example of the configuration of a sorting device 1 according to this embodiment. Fig. 1(A) is an overhead view of a sorting system using the sorting device 1, and Fig. 1(B) is a side view of the sorting system using the sorting device 1. Hereinafter, the sorting system using the sorting device 1 may be simply referred to as the "sorting system."

[0014] The sorting device 1 sorts out crushed pieces of metal products crushed by a crusher (shredder) (not shown). In Fig. 1, the crushed pieces are indicated by OB1 to OB3. The sorting system in Fig. 1 includes a conveying unit 11, an information acquiring unit 12, a sorting unit 13, and a control unit 14.

[0015] The transport unit 11 is a transport mechanism that transports the crushed pieces. In the example of Fig. 1(B), the transport unit 11 is composed of a belt conveyor 111 and a motor 112, and OB1 to OB3 on the belt conveyor 111 are transported in the transport direction (X-axis direction) by driving the motor 112.

[0016] The information acquisition unit 12 acquires information about the fragments transported by the transport unit 11. In the example of Fig. 1(B), the information acquisition unit 12 is composed of an infrared image capture unit 12a that captures an infrared image using infrared rays and an optical image capture unit 12b that captures an optical image using an optical system, as information about the fragments.

[0017] The infrared image capturing unit 12a captures images of the fragments on the belt conveyor in the infrared region and generates an infrared image. The control unit 14, which will be described later, can obtain temperature information of the fragments based on the infrared image.

[0018] The sorting unit 13 sorts the crushed pieces transported by the transport unit 11. In the example of FIG. 1(B), the sorting unit 13 is composed of a kicker 131 for depositing the crushed pieces transported on the belt conveyor 111 into a corresponding tray depending on the sorting result, and a motor 132 for driving the kicker 131 to change the destination of the crushed pieces. Here, the kicker 131 is a plate-shaped member, and is configured to change the tray into which the transported crushed pieces are deposited by changing its orientation relative to the belt conveyor 111 through the driving of the motor 132. Also, in the example of FIG. 1, trays 19a and 19b are shown as trays to which the crushed pieces are transported, but the types of pieces to be sorted are not limited to two.

[0019] The control unit 14 controls the transport unit 11, the information acquisition unit 12, and the sorting unit 13. In particular, the control unit 14 acquires images captured by the information acquisition unit 12, and performs a classification process for the fragments based on the acquired images using a machine learning model that has been trained using training data generated by the processing method according to this embodiment. The classification process using the machine learning model will be described later.

[0020] The control unit 14 also controls the processes according to this embodiment shown in Fig. 2, which will be described later. Here, the control unit 14 is a control unit of a computer, and is assumed to include a CPU and memory. However, the control unit 14 is not limited to this as long as it can similarly control each process. For example, the control unit 14 may be a server device, or the functions performed by the control unit 14 may be implemented by multiple devices.

[0021] 1 is an example. The transport unit 11 is not limited to this configuration as long as it is capable of transporting the crushed fragments, and may, for example, have an arm that grips and transports the crushed fragments instead of the belt conveyor 111. The functional units included in the information acquisition unit 12 are not limited to the infrared image capture unit 12a and the optical image capture unit 12b, and may, for example, have only the optical image capture unit 12b, or there may be multiple capture units with the same function, and may include sensors (not shown) that acquire different information about the crushed fragments, such as an X-ray image capture unit that measures the thickness of the crushed fragments or a weight sensor that measures the weight of the crushed fragments.

[0022] In this embodiment, the process of separating the outer and inner panels of a vehicle body from the crushed pieces will be described as an example of parts constituting a metal product. For this purpose, a vehicle body is used as the metal product, and the crushed pieces OB1 to OB3 are treated as crushed pieces obtained by crushing a vehicle using a crusher, after which non-ferrous materials (especially resin materials) have been removed by pre-processing such as magnetic sorting. This pre-processing can be performed using known techniques for separating crushed metal products. For example, pre-processing such as wind sorting, electromagnetic induction sorting, gravity sorting, color sorting, or XRT may be performed to remove non-ferrous materials.

[0023] Furthermore, in order to preliminarily remove fragments that are neither outer nor inner panels from the fragments, a machine learning model that performs known image recognition processing may be used to remove fragments that are recognized as neither outer nor inner panels from the captured image (for example, cast materials, extruded materials, or resin materials) and treat the remaining fragments as fragments OB1 to OB3. Hereinafter, when simply referred to as "fragments," this refers to fragments from which some fragments have been removed by such pre-processing. Furthermore, here, "inner panels (inner panel skeleton)" refers to the body part that forms the skeleton of the vehicle, and "outer panels (outer panel skeleton)" refers to the exterior parts of the vehicle. Furthermore,

[0024] Here, the fragments take on various shapes depending on the material of the parts. For example, when a mild steel plate with a thickness of 1.0 mm or less (e.g., an ultra-low carbon rolled steel plate used for exterior panels) is crushed, the fragments will be crumpled at small intervals. When a mild steel plate with a thickness of 2.0 mm or more (e.g., a hot-rolled steel plate used for reinforcing members) is crushed, the fragments will be crushed at large intervals. When a high-tensile steel plate with a thickness of 1.0 mm or more (e.g., a frame part of a car body) is crushed, the fragments will be flat without being crushed. Furthermore, even for the same type of part (e.g., an exterior panel), the shape of the fragments may vary depending on whether the material is iron or aluminum. Furthermore, for example, structural steel such as Cu wire, brazed parts containing Cu, or bolts, which are not the fragments to be sorted, may be mixed in as fragments to be sorted. In the processing method according to this embodiment, training data is created to train a machine learning model to enable such sorting.

[0025] The processing method according to this embodiment will be described below. Fig. 2 is a flowchart showing an example of the processing for generating training data for a machine learning model performed in this embodiment. The processing method described below is assumed to be executed by a control unit 14 using a device similar to the sorting device shown in Fig. 1.

[0026] In S201, the control unit 14 performs a marking step in which metal products are marked on a part-by-part basis. Here, the marking step is performed, for example, by applying paint to the metal products. For example, fluorescent paint (any color can be selected) can be used as the paint, but the type of paint is not limited as long as it can impart information that can be identified by image recognition in the classification step (S207) described below. For example, when marking a pre-painted exterior panel, the marking may be performed with a paint different from the paint used to paint the exterior panel. By using fluorescent paint as the marking, the coloring by the fluorescent paint is easily recognized even when markings are only applied to some of the fragments, thereby improving the accuracy of classification by image recognition.

[0027] Here, the marking is performed on the outer panels of the vehicle body before shredding, but depending on the user's wishes, marking can also be performed on specific parts (which may be one or more) to be sorted after shredding. Here, the types of parts to be marked are, for example, aluminum outer panels, iron outer panels, inner panels, or high-tensile steel, and crushed pieces bearing such markings are classified as crushed pieces of specific parts by the classification process in S207, but the parts to be marked are not particularly limited to these. Marking on a part-by-part basis may mean marking one part, marking multiple parts with different marks, or marking multiple parts with the same marking.

[0028] The control unit 14 can perform the marking process by controlling a known painting device (not shown) that paints markings on predetermined parts. For example, the marking process can be performed using a painting head that sprays paint from the outer periphery of the vehicle body before crushing.

[0029] In S202, the control unit 14 controls a crusher (not shown) to perform a crushing process in which the marked metal products are crushed into crushed pieces. The crushing process according to this embodiment is performed in the same manner as the crushing process performed to generate the crushed pieces used in the sorting device described above. As the crusher, any crusher commonly used for crushing metal products can be used depending on the size or shape of the crushed pieces to be generated. For example, the crusher may be a compaction crusher such as a press crusher, a drilling crusher that performs cutting using a drill or the like, a refiner, a hammer crusher, or a shear crusher such as a uniaxial crusher or a biaxial crusher.

[0030] In this embodiment, after the shredding process, some of the crushed pieces are removed by pre-processing. In S203, the control unit 14 removes non-ferrous metals from the crushed pieces crushed in S202 by magnetic sorting. In S204, the control unit 14 captures an image (optical image) of each of the crushed pieces from which non-ferrous metals were removed in S203. In S205, the control unit 14 uses the images captured in S204 as input and uses a machine learning model to identify and remove crushed pieces that are recognized as neither outer nor inner plates. To this end, crushed pieces of various parts are prepared in advance without marking, and information about these crushed pieces (particularly the captured images) is used as training data to train a machine learning model that recognizes crushed pieces that are neither outer nor inner plates (or that recognize them as either outer or inner plates).

[0031] In S206, the control unit 14 acquires information about the broken pieces that have been subjected to the exclusion process in S205. Here, the control unit 14 controls the information acquisition unit 12 to acquire images of the broken pieces as information about the broken pieces. The information about the broken pieces is not limited to images, etc., as long as it has features that can be input into a machine learning model and used to determine which parts the broken pieces are. When an image is used as information about the broken pieces, the captured image acquired in S204 may be used.

[0032] In step S208, which will be described later, training data is generated based on information about the broken pieces. For example, if optical images of the broken pieces are used directly as training data, the influence of coloring due to markings may affect the performance of the training data. From this perspective, for example, by using infrared images of the broken pieces, the shape of the broken pieces, the surface condition of the broken pieces, or the thickness of the broken pieces as information about the broken pieces, the influence of markings on the feature quantities can be reduced. By using infrared images, it is possible to generate training data using temperature information of the broken pieces, which is less affected by markings, as feature quantities. For example, the influence of markings (colorings) on the feature quantities may be reduced by correcting the optical images. Furthermore, instead of limiting the color of fluorescent paint applied to one type of part as markings to one color, marking with fluorescent paints of various colors (e.g., 20 types) may be performed, and the optical images may then be used as training data, thereby reducing the influence of color changes on the performance of the training data. For example, paint color information (previously acquired and registered) of the vehicle being processed may be used as information about the broken pieces.

[0033] In S207, the control unit 14 classifies the fragments that have been subjected to the exclusion process in S205 based on the markings to determine which parts of the metal product they belong to. Here, the control unit 14 classifies fragments that are determined to have fluorescent paint attached based on the captured image of the fragments as fragments of outer panels. However, the classification process is not limited to this as long as it is possible to confirm the presence or absence of markings on the fragments. For example, a sensor that can detect the presence or absence of markings on the fragments may be provided, and classification may be performed based on the detection results. The classification results are stored in a database in association with information about the fragments obtained in S206.

[0034] In S208, the control unit 14 generates training data to be used for training a machine learning model that outputs a classification result of which parts the second fragments are, based on information about fragments of a metal product crushed without marking (second fragments), based on information about the fragments (first fragments) acquired in S206 and the classification result in S207. Here, the training data is generated using optical images of the fragments, infrared images of the fragments, the shape of the fragments, the surface condition of the fragments, or the thickness of the fragments, vehicle paint color information, etc. as information about the fragments, so that the classification result in S207 can be output from fragments with similar characteristics.

[0035] In S209, the control unit 14 trains the machine learning model based on the training data generated in S208, and ends the processing of Fig. 2. As described above, the control unit 14 trains the machine learning model using the information about the crushed pieces acquired in S206 as input so that the classification result can be output in S207. In the sorting device 1, it is possible to perform a sorting process for crushed objects using a machine learning model trained using the training data generated in this manner.

[0036] 2 is an example, and some of the processes may be omitted. For example, steps S203 to S205 corresponding to the pre-processing may be omitted, and step S209 for performing learning may be omitted.

[0037] This process allows metal products to be marked on a part-by-part basis before being crushed, and training data can be generated by classifying the crushed pieces based on the markings. This allows for easy separation of the crushed pieces for efficient learning, and allows for the crushed pieces to be sorted into individual parts using image recognition processing.

[0038] [Summary of the embodiment] The above embodiment discloses at least the following processing method, sorting device, and crushed pieces.

[0039] 1. The processing method of the above embodiment is a marking step of marking a metal product made up of a plurality of parts on a part-by-part basis; a crushing step of crushing the marked metal product into first crushed pieces; a classification step of classifying the first crushed fragment as one of the plurality of parts based on the marking; a generation step of generating training data for training a machine learning model that outputs a classification result of which of the plurality of parts the second fragments are from information about second fragments of the metal product crushed without marking, based on the classification result and information about the first fragments; Equipped with. According to this embodiment, it is possible to mark a metal product on a part-by-part basis, then shred it, and generate training data by classifying the shredded pieces into parts based on the markings.

[0040] 2. In the treatment method of the above embodiment, the marking is the application of fluorescent paint. According to this embodiment, the accuracy of classification by image recognition processing can be improved.

[0041] 3. In the processing method of the above embodiment, the metal product includes an outer plate as the part. According to this embodiment, it is possible to select the outer panel as the desired member.

[0042] 4. In the treatment method of the above embodiment, the marking is formed by applying a paint different from the paint applied to the outer panel. According to this embodiment, it is possible to perform sorting even on pre-painted outer panels.

[0043] 5. In the treatment method of the above embodiment, the metal product is a vehicle. According to this embodiment, it is possible to sort the crushed vehicle debris.

[0044] 6. In the processing method of the above embodiment, the information regarding the first fragment includes vehicle paint color information, an infrared image of the first fragment, the shape of the first fragment, the surface condition of the first fragment, or the thickness of the first fragment. According to this embodiment, it is possible to generate training data in which the marking has only a small effect on the feature quantities.

[0045] 7. In the processing method of the above embodiment, the sorting step sorts the first crushed pieces into aluminum outer plates, iron outer plates, or high-tensile steel parts. According to this embodiment, it is possible to select crushed material of a desired material.

[0046] 8. In the processing method of the above embodiment, the machine learning model is trained using the training data. According to this embodiment, it is possible to easily separate the crushed pieces, perform efficient learning, and sort the crushed pieces into individual parts using image recognition processing.

[0047] 9. The sorting device (for example, 1) of the above embodiment classifies the fragments obtained by crushing the metal products using a machine learning model that has been trained using the processing method of the above embodiment. According to this embodiment, it is possible to easily separate the crushed pieces, perform efficient learning, and sort the crushed pieces into individual parts using image recognition processing.

[0048] 10. The crushed pieces of the metal products of the above embodiment are sorted by the sorting device of the above embodiment. According to this embodiment, fragments can be easily separated and sorted using a machine learning model that has undergone efficient learning to obtain fragments.

[0049] 11. The fragments of the above embodiment are a marking step of marking a metal product made up of a plurality of parts on a part-by-part basis; a crushing step of crushing the marked metal product into first crushed pieces; a classification step of classifying the first crushed fragment as one of the plurality of parts based on the marking; They are classified by: According to this embodiment, fragments can be easily separated and sorted using a machine learning model that has undergone efficient learning to obtain fragments.

[0050] Although the embodiments of the invention have been described above, the invention is not limited to the above-described embodiments, and various modifications and variations are possible within the scope of the gist of the invention. [Explanation of symbols]

[0051] 1: sorting device, 11: conveying unit, 12: information acquisition unit, 13: sorting unit, 14: control unit

Claims

1. a marking step of marking a metal product made up of a plurality of parts on a part-by-part basis; a crushing step of crushing the marked metal product into first crushed pieces; a classification step of classifying the first crushed fragment as one of the plurality of parts based on the marking; a generation step of generating training data for training a machine learning model that outputs a classification result of which of the plurality of parts the second fragments are from information about second fragments of the metal product crushed without marking, based on the classification result and information about the first fragments; A processing method comprising:

2. 2. The method of claim 1, wherein the marking is a coating of fluorescent paint.

3. The method of claim 1 , wherein the metal product includes a skin as the part.

4. 4. The method of claim 3, wherein the marking is a coating of paint different from the paint on the outer panel.

5. 2. The method of claim 1, wherein the metal product is a vehicle.

6. The processing method described in claim 5, characterized in that the information regarding the first fragments includes vehicle paint color information, an infrared image of the first fragments, the shape of the first fragments, the surface condition of the first fragments, or the thickness of the first fragments.

7. The processing method according to claim 1 , wherein the sorting step sorts the first fragments into aluminum outer plates, iron outer plates, or high-tensile steel parts.

8. The processing method according to claim 1 , further comprising a learning step of learning the machine learning model using the training data.

9. A sorting device that classifies the fragments obtained by crushing the metal product using a machine learning model that has been trained by the processing method according to claim 8.

10. Broken pieces of metal products sorted by the sorting device according to claim 9.

11. a marking step of marking a metal product made up of a plurality of parts on a part-by-part basis; a crushing step of crushing the marked metal product into first crushed pieces; a classification step of classifying the first crushed fragment as one of the plurality of parts based on the marking; Broken pieces of metal products, classified by:

Images