Sorting device and sorting method

Abstract

This sorting device for sorting crushed pieces of specific materials from a plurality of crushed pieces obtained by crushing a vehicle body structure comprises: an infrared imaging unit for imaging the crushed pieces; and a processing unit for sorting the crushed pieces of the specific materials on the basis of the temperature distribution of the crushed pieces in an infrared image captured by the infrared imaging unit.

Description

Sorting device and sorting method

[0001] The present invention relates to a sorting device and a sorting method.

[0002] In recent years, efforts to significantly reduce waste generation through waste prevention, reduction, recycling, and reuse have been gaining momentum. To achieve this, research and development is being conducted into the recycling of metal materials that make up car bodies, such as steel and aluminum.

[0003] Patent Document 1 discloses a sorting device that sorts mild steel based on the shape of crushed pieces. The sorting device disclosed in Patent Document 1 identifies as wrinkles portions of the surface of the crushed pieces where the difference in index value compared to the surrounding area is equal to or greater than a threshold value based on surface information that indicates a height index value corresponding to the position of the crushed pieces, and sorts the crushed pieces based on the wrinkle ratio, which is the proportion of the surface of the crushed pieces that are wrinkled.

[0004] Patent No. 6726753 specification

[0005] The sorting device disclosed in Patent Document 1 uses a sorting element to separate crushed mild steel from uncrushed high tensile steel based on the proportion of wrinkles on the surface of the crushed pieces, and is unable to sort crushed pieces of a specified material regardless of the deformation state of the crushed pieces.

[0006] In view of the above problems, the present invention aims to provide an advantageous technology for sorting crushed pieces of a specified material regardless of the deformation state of the crushed pieces.

[0007] One embodiment of the sorting device of the present invention is a sorting device that sorts out fragments of a predetermined material from multiple fragments obtained by crushing a vehicle body structure, and is equipped with an infrared imaging unit that photographs the fragments, and a processing unit that sorts out the fragments of the predetermined material based on the temperature distribution of the fragments in the infrared image taken by the infrared imaging unit.

[0008] Another aspect of the sorting method of the present invention is a sorting method in a sorting device that sorts fragments of a specified material from multiple fragments obtained by crushing a vehicle body structure, and includes an infrared photography step of photographing the fragments using an infrared photography unit, and a processing step in which a processing unit sorts the fragments of the specified material based on the temperature distribution of the fragments in the infrared image taken in the infrared photography step.

[0009] According to the present invention, it is possible to select crushed pieces of a predetermined material regardless of the state of deformation of the crushed pieces.

[0010] 1 is a diagram showing an example of the configuration of a sorting device according to an embodiment; FIG. 2 is a diagram showing an overview of the processing of a sorting device according to an embodiment; FIG. 3 is a diagram showing an example of foreign matter detection using an optical image; FIG. 4 is a diagram showing an example of foreign matter detection using an infrared image; FIG. 5 is a diagram showing infrared image example 1 (thin plate with a blazed portion mixed in); FIG. 6 is a diagram showing the change in temperature distribution in the fragments over time; FIG. 7 is a diagram showing infrared image example 2 (extra-low carbon steel for outer panels); FIG. 8 is a diagram showing the change in temperature distribution in the fragments over time; FIG. 9 is a diagram showing infrared image example 3 (extra-low carbon steel for outer panels); FIG. 10 is a diagram showing the change in temperature distribution in the fragments over time; FIG. 11 is a diagram showing infrared image example 4 (thick high strength steel); FIG. 12 is an example of an optical image used for machine learning using optical images; FIG. 13 is a diagram showing a comparative example 1 between an optical image and a radiation image; FIG. 14 is a diagram illustrating the relationship between the distance in the radiation irradiation direction and the thickness of the fragments in the irradiation direction.

[0011] 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 arbitrarily combined. Furthermore, the same reference numerals are used for the same or similar components, and redundant explanations will be omitted.

[0012] (Configuration example of sorting device 100) Fig. 1 is a diagram showing a configuration example of a sorting device 100 according to an embodiment. The sorting device 100 sorts out fragments of a predetermined material from a plurality of fragments (OB1...OB5, etc.) obtained by shredding a vehicle body structure using a crusher (not shown) (hereinafter also referred to as a shredder). In the following description, the fragments will be collectively referred to as fragments OB.

[0013] In this embodiment, the target is crushed pieces OB obtained after crushing a vehicle body structure in a crusher and removing non-ferrous metals by magnetic sorting. The shapes of the crushed pieces OB can be classified into the following three types, for example, from the viewpoint of plate thickness and strength.

[0014] (a) Crumpled pieces at small intervals (for example, mild steel plates with a thickness of 1.0 mm or less, which are essentially ultra-low carbon steel for exterior panels) (b) Crushed pieces at large intervals (for example, mild steel plates with a thickness of 2.0 mm or more, which are essentially hot-rolled steel plates used for reinforcing members) (c) Flat pieces that are not crushed (high-strength steel (hereinafter also referred to as high-tensile steel) with a thickness of 1.0 mm or more, which are body frame parts) Furthermore, structural steel such as wiring (Cu wire), brazing parts containing Cu, and bolts may also be mixed in as foreign matter with the crushed fragments OB.

[0015] The sorting device 100 of this embodiment has a configuration for acquiring image information of the crushed fragments OB, including an infrared image capturing unit 12A (thermo camera) that captures infrared images, an optical image capturing unit 12B (visible light camera) that captures optical images, and a radiation capturing unit (radiation generating unit 13A, radiation detecting unit 13B). These are collectively referred to as "image information acquiring units (12A, 12B, 13A, 13B)."

[0016] The sorting device 100 has an image information acquisition unit (12A, 12B, 13A, 13B) that acquires image information of the crushed fragments OB, a processing unit 110 that processes the information acquired by the image information acquisition unit, and a control unit 120 that controls the sorting mechanism 14A based on the judgment result of the processing unit 110.

[0017] The processing unit 110 of this embodiment is capable of performing inference processing using a trained machine learning model 150. In machine learning, optical images and infrared images prepared in advance are used as inputs to the learning model, and machine learning is performed to output a sorting judgment result as an output, thereby configuring the trained machine learning model 150. In the inference phase, optical images and infrared images of the fragments OB transported by the transport mechanism 11 described below become input data for the machine learning model 150. The processing unit 110 is capable of sorting out the fragments OB of the extra-low carbon steel for outer panels as classified in (a) above by performing inference processing based on the output data of the machine learning model 150. In addition to the inference processing using the machine learning model 150, the processing unit 110 may also sort out the fragments OB of the extra-low carbon steel for outer panels classified in (a) above by taking into account plate thickness information obtained from the radiographic image as needed.

[0018] The conveying mechanism 11 may be configured to convey a plurality of crushed pieces OB, and typically employs a belt conveyor in which a belt 11A is driven by a drive pulley 11B and a driven pulley 11C. A drive source such as a motor (not shown) is connected to the drive pulley 11B, and the rotational power of the drive source is transmitted to the drive pulley 11B, allowing the conveying mechanism 11 to convey the crushed pieces OB in a predetermined direction in the direction of the arrow. The belt 11A forms a conveying path for conveying each crushed piece OB, and is capable of conveying various crushed pieces OB1, OB2, and OB3 obtained by crushing a vehicle body structure using a crusher (shredder) in the direction of the arrows.

[0019] In Figure 1, the left side of the paper is the upstream side in the conveying direction, and the right side of the paper is the downstream side in the conveying direction. The conveying direction is the X direction, the up-down direction intersecting the X direction is the Z direction, and the direction perpendicular to the paper surface and intersecting the XZ direction is the Y direction. The crushed pieces OB discharged from the shredder are supplied to the upstream side of the conveying mechanism 11, and each crushed piece OB is conveyed from upstream to downstream by the conveying mechanism 11. The crushed pieces OB are photographed by image information acquisition units (12A, 12B, 13A, 13B) on the conveying path conveyed by belt 11A. The image information photographed by the image information acquisition units (12A, 12B, 13A, 13B) is input to the processing unit 110.

[0020] The infrared image capturing unit 12A (thermo camera) is configured to capture images of the fragments OB on the belt 11A in the infrared region and is used to detect the amount of infrared radiation emitted from the fragments OB. Image information (infrared image) captured by the infrared image capturing unit 12A is input to the processing unit 110. The processing unit 110 can obtain information on the temperature distribution of the fragments OB based on the image information (infrared image).

[0021] The optical image capturing unit 12B (visible light camera) is configured to capture images of the fragments OB on the belt 11A in the visible light range. The optical image capturing unit 12B may be, for example, a camera configured with a CCD / CMOS image sensor. The image information (optical image) captured by the optical image capturing unit 12B is input to the processing unit 110. The processing unit 110 can acquire appearance information of the fragments OB based on the image information (optical image).

[0022] The processing unit 110 also acquires position information of the fragments OB on the belt 11A based on the image information (optical image) acquired from the optical image capturing unit 12B. The position information of the fragments OB is linked to the image information (infrared image) of the infrared image capturing unit 12A and the radiation image information of the radiation capturing units (13A, 13B) described later. Furthermore, when switching the sorting destination (15A or 15B) of the sorting mechanism 14A, the fragments OB to be sorted are identified based on the position information of the fragments OB on the belt 11A.

[0023] The radiography unit has a radiation generation unit 13A that irradiates the fragments OB with radiation, and a radiation detection unit 13B that detects the radiation that has passed through the fragments OB.

[0024] The radiation generating unit 13A starts irradiating radiation in accordance with an irradiation start command from the processing unit 110. The radiation is typically X-rays, but may also be gamma rays or the like. The radiation irradiated from the radiation generating unit 13A passes through the fragments OB and enters the radiation detecting unit 13B. The radiation generating unit 13A stops irradiating radiation in accordance with an irradiation stop command from the processing unit 110.

[0025] As an example configuration, the radiation detection unit 13B may include a radiation detection panel having a pixel array in which pixels that detect radiation are arranged two-dimensionally, a conversion circuit that acquires charges corresponding to the irradiated dose, and an image generation unit that generates a radiographic image based on the charge information. Note that a line sensor in which pixels are arranged one-dimensionally may be used instead of the radiation detection panel having a two-dimensional pixel array. The image generation unit generates a radiographic image based on charge information corresponding to the radiation incident on the radiation detection panel and transmits the generated radiographic image to the processing unit 110.

[0026] The radiation dose that passes through the fragments OB varies depending on the thickness of the material that makes up the fragments OB and the state of crushing (for example, whether crushed into a spherical shape or wrinkled without being crushed, etc.), and this can result in shading due to the difference in dose in a radiological image showing the interior of the fragments OB. Based on the radiological image, the processing unit 110 can determine the thickness of the fragments OB or the distribution of thickness inside the crushed fragments OB (in the direction of radiation irradiation).

[0027] FIG. 10 shows a first comparative example of optical images (OM101-104) and radiographic images (XM101-104). Optical image OM101 is an optical image of a thin plate crushed into a spherical shape, and optical image OM102 is an optical image of a thick plate crushed into a spherical shape. Furthermore, radiographic image XM101 is a radiographic image of a thin plate crushed into a spherical shape, and radiographic image XM102 is a radiographic image of a thick plate crushed into a spherical shape. While it is difficult to determine the details of how the fragments OB are crushed simply by comparing the appearance of the optical images (OM101, OM102), by comparing the radiographic images XM101 and XM102, image analysis of how the fragments OB are crushed can be easily performed. While attempting to determine the details by image analysis of optical images alone would require a large number of optical images taken from various directions, by using radiographic images in combination, image analysis of how the fragments OB are crushed can be easily performed.

[0028] Optical image OM103 is an optical image of a thin plate that has been wrinkled without being crushed, and optical image OM104 is an optical image of a thick plate that has been slightly deformed. Furthermore, radiographic image XM103 is a radiographic image of a thin plate that has been wrinkled without being crushed, and radiographic image XM104 is a radiographic image of a thick plate that has been slightly deformed. The same is true for the optical images (OM103, OM104), and it is difficult to determine the details of how the fragments OB have been crushed simply by comparing their appearances. However, by comparing radiographic images XM103 and XM104, it is possible to easily perform image analysis of how the fragments OB have been crushed in both cases.

[0029] Figure 11 is a diagram illustrating the relationship between the distance in the radiation irradiation direction and the thickness of the fragments OB in the irradiation direction, as determined by analysis of a radiographic image. In the optical images (OM111, OM112) in Figure 11, the arrow directions 1101 and 1102 indicate the irradiation direction in which radiation is irradiated onto the fragments OB from a direction perpendicular to the paper surface. The plate thickness along the irradiation direction was measured from the intensity of the radiation that was irradiated from a direction perpendicular to the paper surface and transmitted through the fragments OB. The graph in Figure 11 illustrates the relationship between the distance along the irradiation direction and the thickness when radiation is irradiated onto the fragments OB in the optical images (OM111-112) from a direction perpendicular to the paper surface (arrow directions 1101 and 1102). The graph on the left of Figure 11 is a diagram illustrating the relationship between distance along the irradiation direction and thickness corresponding to optical image OM111, and the graph on the right of Figure 11 is a diagram illustrating the relationship between distance along the irradiation direction and thickness corresponding to optical image OM112.

[0030] The optical image OM111 in Fig. 11 is an optical image of a thin plate with almost no deformation, and the optical image 112 in Fig. 11 is an optical image of a thick plate with almost no deformation, and they are indistinguishable by comparing the appearances of the optical images, but the only difference is the plate thickness. In this way, although it is difficult to distinguish between thin and thick plates using optical images, it is possible to easily distinguish between thin and thick plates by using radiographic images.

[0031] (Configuration of Processing Unit 110) The processing unit 110 may be configured with a dedicated circuit such as a programmable logic device (PLD) such as a field programmable gate array (FPGA) or an application specific integrated circuit (ASIC). Alternatively, the processing unit 110 may be configured with a combination of a general-purpose processing circuit such as a processor and a storage circuit such as a memory. In this case, the general-purpose processing circuit may execute a program stored in the storage circuit, thereby realizing the functions of the processing unit 110.

[0032] The processing unit 110 is capable of performing material sorting processing based on inference processing using deep learning, neural networks, etc. The processing unit 110 is capable of sorting predetermined materials from the crushed fragments OB by applying a trained machine learning model 150 to an image (infrared image) acquired from the infrared image capturing unit 12A as an input image. Various materials can be selected as the material to be sorted, but it is also possible to select steel plates (ultra-low carbon mild steel plates) having a thickness of a predetermined plate thickness or less (for example, 1 mm or less).

[0033] In machine learning for constructing the machine learning model 150, information on the temperature distribution of the fragments OB in an infrared image may be used as an input image (input data). Differences in temperature distribution over time after heating the fragments OB may be used as information on the temperature distribution of the fragments OB. Alternatively, differences in brightness due to temperature changes detected in an infrared image (differences in emissivity calculated from brightness) may be used as information on the temperature distribution of the fragments OB. Alternatively, differences in thermal conduction patterns or differences in thermal conductivity that can be calculated from changes in brightness due to temperature changes detected in an infrared image may be used as information on the temperature distribution of the fragments OB. Alternatively, differences in reflectance or brightness information when irradiated with light of a specific wavelength may be used as information on the temperature distribution of the fragments OB. Machine learning combining these data may be performed in advance as input images (input data), and a trained machine learning model 150 may be constructed.

[0034] 5A is a diagram showing Example 1 of infrared images (a thin plate containing blazed portions) showing the difference in temperature distribution over time, and FIG. 5B is a diagram showing the change in temperature distribution over time in the crushed pieces OB. Since heat is generated in the crushed pieces OB during processing by the shredder, the input data assumes heat generation due to processing, and the crushed pieces OB are heated to a predetermined temperature, and infrared images are obtained over time.

[0035] In Example 1 of Figure 5A, the broken piece OB includes a brazed portion 501. The brazed portion 501 is a joint portion brazed using, for example, a Cu-based brazing filler metal, and 502 indicates the surface (steel plate surface) of the broken piece OB. IR51 is an infrared image before heating, IR52 is an infrared image immediately after heating, IR53 is an infrared image TM1 hour after heating, and IR54 is an infrared image TM2 hours (>TM1) after heating.

[0036] 5B, the dashed line indicates the temperature change of the blaze portion 501, and the solid line indicates the temperature change of the surface 502 of the broken fragments OB. As shown in FIG. 5B, the temperature distribution changes such that before heating, the temperatures of the blaze portion 501 and the surface 502 of the broken fragments OB are the same. However, immediately after heating, after TM1 time has elapsed, and after TM2 time has elapsed, the surface 502 of the broken fragments OB (the steel plate surface) is characterized by being easier to heat and harder to cool than the blaze portion 501. The blaze portion 501, which contains Cu as a brazing filler metal, is characterized by being harder to heat and easier to cool than the surface 502 of the broken fragments OB. Broken fragments OB exhibiting such characteristics are reflected in the machine learning model 150 as being contaminated with foreign matter.

[0037] Figure 6A shows example 2 of an infrared image (ultra-low carbon steel for outer panels) showing the difference in temperature distribution over time, and Figure 6B shows the change in temperature distribution in the fragment OB over time.

[0038] In Example 2 of FIG. 6A , 601 indicates the interior of the crushed portion, and 602 indicates the surface of the crushed fragments OB. IR61 is an infrared image immediately after heating, IR62 is an infrared image TM1 hours after heating, and IR63 is an infrared image TM3 hours after heating (>TM2>TM1). In FIG. 6B , the dashed line indicates the temperature change of the interior of the crushed portion 601, and the solid line indicates the temperature change of the surface 602 of the crushed fragments OB. As shown in FIG. 6B , the temperature distribution changes in the interior of the crushed portion 601 and the surface 602 of the crushed fragments OB immediately after heating. However, as time passes, the surface 602 of the crushed fragments OB is characterized by being more likely to cool than the interior of the crushed portion 601. The material of the crushed fragments OB exhibiting such characteristics is reflected in the machine learning model 150 as ultra-low carbon steel for outer panels.

[0039] Figure 7A shows example 3 of an infrared image (ultra-low carbon steel for outer panels) showing the difference in temperature distribution over time, and Figure 7B shows the change in temperature distribution in the fragment OB over time.

[0040] In Example 3 of FIG. 7A , 701 indicates the interior of the crushed portion, and 702 indicates the surface of the crushed fragments OB. IR71 is an infrared image immediately after heating, IR72 is an infrared image TM1 hours after heating, and IR73 is an infrared image TM2 hours (>TM1) after heating. In FIG. 7B , the dashed line indicates the temperature change of the interior of the crushed portion 701, and the solid line indicates the temperature change of the surface 702 of the crushed fragments OB. As shown in FIG. 7B , the temperature distribution changes in the interior of the crushed portion 701 and the surface 702 of the crushed fragments OB immediately after heating. However, as time passes, the surface 702 of the crushed fragments OB is characterized as being more likely to cool than the interior of the crushed portion 701. The material of the crushed fragments OB exhibiting these characteristics is reflected in the machine learning model 150 as ultra-low carbon steel for outer panels.

[0041] FIG. 8 shows Example 4 of infrared images (thick high-strength steel plate) showing differences in temperature distribution over time. In Example 4 of FIG. 8, 801 indicates a hole on the surface of the broken fragment OB, and 802 indicates the edge of the plate material that constitutes the broken fragment OB. IR81 is an infrared image immediately after heating, IR82 is an infrared image TM1 hour after heating, and IR83 is an infrared image TM2 hours (>TM1) after heating. The plate edge 802 tends to heat up slowly immediately after heating and has a low temperature distribution. The material of the broken fragment OB that exhibits these characteristics is reflected in the machine learning model 150 as a thick plate.

[0042] In machine learning for constructing the machine learning model 150, image information of the broken pieces OB in the optical image may be used as the input image (input data). For example, an optical image such as that shown in FIG. 9 may be used as input data. OM91 illustrates an optical image of a thin mild steel plate for exterior panels. OM92 illustrates an optical image of high-strength steel (hi-tensile steel). OM93 illustrates an optical image of other components such as sashes, bolts, copper wires, etc. In machine learning using optical images, the machine learning model 150 may be configured so that output data is correctly output for these input images.

[0043] In the inference process using an infrared image, image information (infrared image) acquired from the infrared image capturing unit 12A is input to the trained machine learning model 150, whereby the processing unit 110 can acquire the results of sorting of the fragments OB as determination information as an output of the machine learning model 150. Specifically, this process may be executed in the fragments OB classification process (S33) or the overall determination process (S51), which will be described later in FIG. 2 . The same applies to the inference process using an optical image. By inputting image information (optical image) acquired from the optical image capturing unit 12B to the trained machine learning model 150, the processing unit 110 may acquire the results of sorting of the fragments OB as determination information as an output of the machine learning model 150. Specifically, this process may be executed in the fragments OB classification process (S33) or the overall determination process (S51), which will be described later in FIG. 2 .

[0044] The processing unit 110 transmits determination information indicating the material sorting results and position information of the crushed pieces OB that were selected to the control unit 120. Based on the position information of the crushed pieces OB and the determination information received from the processing unit 110, the control unit 120 transmits an operating signal to switch the sorting destination (15A or 15B) of the sorting mechanism 14A.

[0045] The sorting mechanism 14A adjusts the angle of the sorting panel 14B based on a movement signal received from the control unit 120, thereby switching the sorting destination of the crushed pieces OB transported by the belt 11A to either the recovery unit 15A or 15B. The example in Figure 1 shows a state in which crushed pieces OB4 selected as a predetermined material (ultra-low carbon mild steel plate) have been sorted and recovered in the recovery unit 15B. The example in Figure 1 also shows a state in which crushed pieces OB5 of a material other than the predetermined material have been sorted and recovered in the recovery unit 15A.

[0046] In the configuration example of the sorting mechanism 14A shown in FIG. 1, the sorting mechanism 14A changes the angle of the sorting panel 14B based on a movement signal, but other mechanisms capable of achieving equivalent functions may be used. For example, an air blow system or a robot arm system may be used. Boxes, trays, etc. are typically used for the collection units 15A and 15B. In the configuration example shown in FIG. 1, the number of collection units 15A and 15B is two, but three or more may be used.

[0047] (Outline of Processing by the Sorting Device 100) Fig. 2 is a diagram showing an outline of processing by the sorting device 100 according to an embodiment. In step S21, the optical image capturing unit 12B captures an image of the broken fragments OB on the belt 11A. The image information captured by the optical image capturing unit 12B is input to the processing unit 110. The processing unit 110 identifies the position information of the broken fragments OB on the belt 11A based on the image information acquired from the optical image capturing unit 12B (S61).

[0048] In step S22, the processing unit 110 performs a color recognition process on the surface of the broken fragments OB for the image acquired from the optical image capturing unit 12B. In step S23, the processing unit 110 refers to a vehicle information database (DB) and determines whether the surface color of the broken fragments OB based on the color recognition process (S22) matches the color information registered in the vehicle information DB. If the surface color of the broken fragments OB matches the registered color information, the processing unit 110 can classify the type of broken fragments OB as a component constituting an outer panel of a vehicle (S51).

[0049] On the other hand, if the surface color of the fragment OB does not match the registered color information, in step S24, the processing unit 110 detects characteristic features in the image acquired from the optical image capture unit 12B and determines whether the fragment OB contains any foreign objects, such as bolts or copper wires, that can be detected from the characteristic features of its appearance. Figure 3A shows an example of a detected bolt, and Figure 3B shows an example of a detected copper wire. In this way, fragments containing foreign objects, such as bolts or copper wires, are excluded from selection.

[0050] In step S31, the infrared image capturing unit 12A captures an image of the broken pieces OB on the belt 11A. The image information captured by the infrared image capturing unit 12A is input to the processing unit 110.

[0051] In step S32, the processing unit 110 uses the trained machine learning model 150 to acquire the temperature distribution and the like in the fragmented fragments OB from the image acquired from the infrared image capturing unit 12A and detects foreign objects contained in the fragmented fragments OB. FIG. 4A shows an example of detecting a blazed portion 401 contained in the fragmented fragments OB as an example of foreign object detection. The blazed portion 401 shown in FIG. 4A is a portion brazed using, for example, a Cu-based brazing filler metal, and the blazed portion 401 is distributed linearly in the fragmented fragments OB. FIG. 4B shows an example of detecting bolts 402, 403 contained in the fragmented fragments OB. By using the temperature distribution and the like in the fragmented fragments OB acquired from the infrared image, foreign objects contained in the fragmented fragments OB can be detected without being affected by dirt or remaining paint film on the surface of the fragmented fragments OB. In this way, fragments containing foreign objects such as bolts and copper wire are excluded from the selection targets.

[0052] In step S33, the processing unit 110 classifies (shape classifies) the broken fragments OB. The processing unit 110 inputs the infrared image into the trained machine learning model 150 and performs inference processing to classify (shape classify) the broken fragments OB. The processing unit 110 may acquire the external characteristics of the broken fragments OB based on the optical image and classify the shape. Alternatively, the processing unit 110 may input the optical image into the machine learning model 150 and perform inference processing to classify the shape of the broken fragments OB.

[0053] In the inference process using infrared images, image information (infrared images) acquired from the infrared image capturing unit 12A is input to the trained machine learning model 150, whereby the processing unit 110 can acquire the sorting results of the fragments OB as judgment information as an output of the machine learning model 150. The time from processing by the shredder until the infrared image is captured by the infrared image capturing unit 12A may be the elapsed time after heating. When shredding the vehicle body structure using a shredder and subsequently capturing infrared images, the fragments are heated to a high temperature due to processing heating, so there is no need to heat the fragments. Furthermore, when sorting fragments that have been cooled to room temperature, they may be reheated using infrared rays, high frequency waves, or other methods.

[0054] In addition, in the inference processing using optical images, by inputting the image information (optical image) acquired from the optical image capturing unit 12B into the trained machine learning model 150, the processing unit 110 can acquire the sorting results of the fragments OB as judgment information as the output of the machine learning model 150.

[0055] In the classification process of step S33, if the end faces of the broken pieces OB are visible, detailed determination based on the plate thickness is possible, and the process proceeds to step S41.

[0056] In step S41, the processing unit 110 outputs an irradiation start command to the radiation generation unit 13A to start radiography. The radiation detection unit 13B generates a radiographic image of the fragments OB and transmits it to the processing unit 110.

[0057] In step S42, the processing unit 110 measures the thickness of the fragments OB based on the radiographic image. This thickness measurement can be performed, for example, as described in Fig. 11, by irradiating radiation from a direction perpendicular to the paper surface (arrow directions 1101 and 1102) and determining the thickness along the irradiation direction from the intensity of the radiation that has passed through the fragments OB.

[0058] In step S51, the processing unit 110 performs a comprehensive judgment process, which comprehensively selects the crushed fragments OB by taking into consideration the results of the classification process (inference process) in step S33, the results of the plate thickness measurement in step S42, and the results of the reference to the vehicle information DB (step S23).

[0059] The processing unit 110 may select fragments OB of a predetermined material using an infrared image, or may select fragments OB of a predetermined material by combining the results of analysis of a radiographic image or an optical image. The processing unit 110 may select fragments OB of a predetermined material based on the temperature distribution of the fragments OB in an infrared image captured by the infrared imaging unit 12A. The temperature distribution of the fragments OB includes changes in temperature distribution over time after heating the fragments OB, and the processing unit 110 may select fragments OB of a predetermined material based on the changes in temperature distribution (e.g., FIGS. 6A and 7A). Furthermore, the temperature distribution of the fragments OB includes differences in brightness due to temperature changes detected in the infrared image, and the processing unit 110 may select fragments OB of a predetermined material based on differences in thermal conduction patterns that can be calculated from the differences in brightness. In addition, the processing unit 110 may determine that the fragments OB contain a material different from a specified material based on differences in the tendency for the temperature of the fragments OB to rise or the tendency for the temperature of the fragments OB to fall due to differences in the thermal conductivity of the materials (for example, 4B in Figure 4), and may exclude fragments OB containing a different material (for example, bolts or wiring, etc.) from the sorting target.

[0060] In addition, the processing unit 110 may determine that an area in the infrared image where the temperature distribution is linear is a joint (e.g., 401, 501) that is brazed using a filler material (e.g., copper) different from a specified material (e.g., mild steel) (e.g., 4A in Figure 4, Figure 5A), and exclude fragments OB that include the joint from the selection target.

[0061] Furthermore, the processing unit 110 may determine that crushed fragments OB having a curvature rate equal to or greater than a predetermined value in the temperature distribution of the crushed fragments OB are mild steel, which is a predetermined material. In an infrared image of the crushed fragments OB determined to be mild steel, the processing unit 110 may select, as the crushed fragments OB, fragments whose surfaces are more likely to cool than their interiors (e.g., FIGS. 6A and 7A ).

[0062] In addition, as described in Figure 11, the processing unit 110 may irradiate radiation from a direction perpendicular to the paper surface (arrow directions 1101, 1102) and obtain the distribution of plate thickness along the irradiation direction from the intensity of the radiation that passes through the fragments OB, and may select fragments OB of a specified material using the temperature distribution of the fragments OB in the infrared image and the thickness distribution of the fragments OB.

[0063] The processing unit 110 may acquire the outline of the fragment OB from a cooled region based on the temperature distribution of the fragment OB (e.g., 802 in FIG. 8 ). The processing unit 110 may acquire the appearance characteristics of the fragment OB based on the optical image, and sort out fragment OB of a predetermined material using the temperature distribution of the fragment OB and the appearance characteristics of the fragment OB. The processing unit 110 may determine that the fragment OB, which shows a tendency for a low temperature distribution in a region determined to be the edge of the plate material constituting the fragment OB based on the appearance characteristics of the fragment OB (e.g., 802 in FIG. 8 ), has a thickness greater than the plate thickness of the predetermined material (e.g., 1 mm for mild steel), and exclude the fragment OB from the sorting target (e.g., IR83 in FIG. 8 ).

[0064] In addition, in S23, if the surface color of the fragment OB matches the color information registered in the vehicle information DB, the fragment OB may be selected as a fragment OB of a material that constitutes the outer panel of the vehicle. In addition, if thickness information of the fragment OB is required, thickness information (S42) obtained from the radiographic image may be used in combination with the color information (S23) to select fragments OB of a predetermined material.

[0065] The processing unit 110 outputs the result of the comprehensive judgment process in S51 to the control unit 120. The processing unit 110 transmits to the control unit 120 judgment information indicating the sorting result of the fragments OB and position information of the fragments OB that were selected.

[0066] In step S62, the control unit 120 transmits a signal to switch the sorting destination (15A or 15B) of the sorting mechanism 14A based on the position information and determination information of the crushed fragments OB received from the processing unit 110. Based on the signal received from the control unit 120, the sorting mechanism 14A adjusts the angle of the sorting panel 14B to switch the sorting destination of the crushed fragments OB transported by the belt 11A to the recovery unit 15A or 15B. This allows for accurate recovery of crushed fragments OB of a specified material. According to this embodiment, it is possible to sort crushed fragments of a specified material regardless of the deformation state of the crushed fragments.

[0067] [Summary of the embodiment] Configuration 1. A sorting device (100) that sorts out fragments of a predetermined material from a plurality of fragments obtained by crushing a vehicle body structure includes: an infrared imaging unit (12A) that photographs the fragments; and a processing unit (110) that sorts out the fragments of the predetermined material based on the temperature distribution of the fragments in an infrared image photographed by the infrared imaging unit.

[0068] Configuration 2. The temperature distribution of the crushed pieces includes a change in temperature distribution over time after the crushed pieces are heated, and the processing unit (110) selects the crushed pieces of the specified material based on the change in temperature distribution.

[0069] Configuration 3. The temperature distribution of the fragments includes differences in brightness due to temperature changes detected in the infrared image, and the processing unit (110) sorts out the fragments of the predetermined material based on differences in heat conduction patterns that can be calculated from the differences in brightness.

[0070] Configuration 4. The processing unit (110) determines that the crushed fragments contain a material different from the predetermined material based on differences in the tendency of the crushed fragments to increase in temperature or differences in the tendency of the crushed fragments to decrease in temperature due to differences in the thermal conductivity of the materials, and excludes the crushed fragments containing the different material from the sorting targets.

[0071] Configuration 5: The different materials include bolt or wiring materials, and the processing unit (110) excludes the fragments containing the different materials from the sorting target.

[0072] Configuration 6. The processing unit (110) determines in the infrared image that a region where the temperature distribution is linear is a joint portion brazed using a filler material different from the predetermined material, and excludes fragments including the joint portion from the objects of sorting.

[0073] Configuration 7. The processing unit (110) selects, in the infrared image of the crushed fragments, fragments whose surfaces are more likely to cool than their interiors after crushing, as fragments of the predetermined material.

[0074] Configuration 8. The apparatus further comprises a generating unit (13A) that irradiates the fragments with radiation, and a detecting unit (13B) that detects the radiation that has passed through the fragments and acquires a radiological image based on the detected radiation dose, wherein the processing unit (110) acquires a thickness distribution of the fragments in the radiation irradiation direction based on the radiological image, and sorts out fragments of the predetermined material using the temperature distribution of the fragments in the infrared image and the thickness distribution of the fragments.

[0075] Configuration 9: The processing unit (110) obtains the outline of the fragment from a cooled region based on the temperature distribution of the fragment.

[0076] Configuration 10. The apparatus further includes an optical image capturing unit (12B) that captures optical images of the fragments, and the processing unit (110) acquires the external characteristics of the fragments based on the optical images, and sorts the fragments of the predetermined material using the temperature distribution of the fragments and the external characteristics of the fragments.

[0077] Configuration 11. The processing unit (110) determines that a fractured piece that shows a tendency for the temperature distribution to be low in an area determined to be an edge of the plate material constituting the fractured piece based on the appearance characteristics has a thickness greater than the plate thickness of the predetermined material, and excludes the fractured piece from being selected.

[0078] Configuration 12. A sorting method for a sorting device that sorts out crushed fragments of a predetermined material from a plurality of crushed fragments obtained by crushing a vehicle body structure, comprising: an infrared photography step of photographing the crushed fragments using an infrared photography unit (12A), and a processing step of a processing unit (110) sorting out the crushed fragments of the predetermined material based on the temperature distribution of the crushed fragments in the infrared image photographed in the infrared photography step.

[0079] According to the first to twelfth configurations, it is possible to separate crushed pieces of a predetermined material regardless of the state of deformation of the crushed pieces. According to the configurations of the embodiments, it is possible to separate crushed pieces of high-purity iron scrap from scrap of car body structures. Furthermore, it is possible to separate copper (Cu) and other materials with high thermal conductivity. Until now, it was impossible to produce mild steel sheets for automotive exterior panels (essentially ultra-low carbon steel sheets) from commercial scrap containing large amounts of alloys such as carbon (C) and manganese (Mn). Therefore, the only option was to use mild steel scrap from steel mills and press scrap, mainly consisting of raw materials derived from iron ore (reduced iron or blast furnace pig iron). According to the present embodiment, it is possible to extract only ultra-low carbon mild steel sheets from crushed pieces of used car bodies. By using these raw materials to produce ultra-low carbon mild steel sheets in an electric furnace or the like, horizontal recycling of steel sheets from car bodies to car bodies becomes possible.

[0080] The invention is not limited to the above-described embodiment, and various modifications and variations are possible within the scope of the gist of the invention.

[0081] 12A: infrared image capturing unit, 12B: optical image capturing unit, 13A: radiation generating unit (generating unit), 13B: radiation detecting unit (detecting unit), 100: sorting device, 110: processing unit, 120: control unit, 150: machine learning model

Claims

1. A sorting device that sorts out fragments of a predetermined material from multiple fragments obtained by crushing a vehicle body structure, comprising: an infrared imaging unit that photographs the fragments; and a processing unit that sorts out the fragments of the predetermined material based on the temperature distribution of the fragments in the infrared image taken by the infrared imaging unit.

2. The sorting device described in claim 1, wherein the temperature distribution of the crushed pieces includes a change in temperature distribution over time after the crushed pieces are heated, and the processing unit sorts the crushed pieces of the specified material based on the change in temperature distribution.

3. The sorting device described in claim 1, wherein the temperature distribution of the fragments includes differences in brightness due to temperature changes detected in the infrared image, and the processing unit sorts the fragments of the specified material based on differences in heat conduction patterns that can be calculated from the differences in brightness.

4. The sorting device described in claim 1, wherein the processing unit determines that the fragments contain a material different from the specified material based on differences in the tendency of the fragments to increase in temperature or differences in the tendency of the fragments to decrease in temperature due to differences in the thermal conductivity of the materials, and excludes the fragments containing the different material from the sorting target.

5. The sorting device according to claim 4, wherein the different materials include bolt or wiring materials, and the processing unit excludes fragments containing the different materials from the sorting target.

6. The sorting device of claim 1, wherein the processing unit determines that an area in the infrared image where the temperature distribution is linear is a joint portion brazed using a filler material different from the specified material, and excludes fragments containing the joint portion from the sorting target.

7. The sorting device described in claim 1, wherein the processing unit sorts out, in the infrared image of the crushed fragments, fragments whose surfaces are more likely to cool than their interiors as fragments of the specified material.

8. A sorting device as described in claim 1, further comprising: a generating unit that irradiates the fragments with radiation; and a detecting unit that detects the radiation that has passed through the fragments and acquires a radiological image based on the detected radiation dose, wherein the processing unit acquires the thickness distribution of the fragments in the direction of irradiation with the radiation based on the radiological image, and sorts fragments of the specified material using the temperature distribution of the fragments in the infrared image and the thickness distribution of the fragments.

9. The sorting device according to claim 1, wherein the processing unit acquires the outline of the crushed pieces from the cooled areas in the temperature distribution of the crushed pieces.

10. A sorting device as described in claim 9, further comprising an optical image capturing unit that captures optical images of the fragments, wherein the processing unit acquires the external characteristics of the fragments based on the optical images, and sorts the fragments of the specified material using the temperature distribution of the fragments and the external characteristics of the fragments.

11. The sorting device described in claim 10, wherein the processing unit determines that a fragment that shows a tendency for the temperature distribution to be low in an area determined to be the edge of the plate material constituting the fragment based on the appearance characteristics has a thickness greater than the plate thickness of the specified material, and excludes the fragment from being sorted.

12. A sorting method for a sorting device that sorts out fragments of a specified material from multiple fragments obtained by crushing a vehicle body structure, comprising: an infrared photography step of photographing the fragments using an infrared photography unit; and a processing step of a processing unit sorting out the fragments of the specified material based on the temperature distribution of the fragments in the infrared image taken in the infrared photography step.

Images