Item information management system, item information management method, and program

Abstract

This invention provides an item information management system that allows for the efficient searching of similar items. [Solution] The article information management system in this embodiment includes: a shape data extraction unit that extracts some shape data representing the shape of an article from 3D model data representing the three-dimensional shape of the article; a feature quantity generation unit that generates feature quantities related to the shape of the article based on the shape data of the article extracted by the shape data extraction unit; a feature quantity storage unit that stores the feature quantities of a plurality of articles in association with each of these articles; and a search unit that searches for articles with similar three-dimensional shapes based on the feature quantities generated by the feature quantity generation unit and the feature quantities stored by the feature quantity storage unit.

Description

Technical Field

[0001] The present invention relates to an article information management system, an article information management method, and a program.

Background Art

[0002] For example, Patent Document 1 discloses a first member feature amount calculation unit that calculates a plurality of feature amounts for each of a plurality of members of a reference building, a second member feature amount calculation unit that calculates a plurality of feature amounts for a member of a target building, and a member cross-section estimation unit that estimates the cross-sectional shape of the member of the target building from the cross-sectional shapes of the plurality of members of the reference building using the feature amounts calculated by the first member feature amount calculation unit and the second member feature amount calculation unit. Each of the plurality of feature amounts for each of the plurality of members of the reference building and the plurality of feature amounts for the member of the target building includes at least one of a value indicating the floor on which the member is installed, a value indicating whether the member is a beam or a column, a value indicating a combination of joining directions of the beams and columns joined at both ends of the member, a value indicating the total length of the beams joined at one end of the member, a value indicating the total length of the beams joined at the other end of the member, and a value indicating the length of the member. A member cross-section estimation device is disclosed.

[0003] Furthermore, Patent Document 2 describes a tool identification device that identifies at least one tool attached to a machine tool, which is communicably connected to a tool specification database in which a general shape of the tool type and tool dimension feature items for identifying a tool belonging to the tool type are defined for each tool type, and a tool management data in which a tool ID that can uniquely identify the tool and tool dimension feature quantities of at least the tool dimension feature items corresponding to the tool type are associated for each tool, and which identifies at least one tool attached to a machine tool, and includes a tool image acquisition unit that acquires image data of the tool to be identified that is attached to the machine tool, and a unit that refers to the general shape of the tool type defined for each tool type stored in the tool specification database and identifies the tool to be identified from the image data A tool identification device is disclosed, comprising: a tool type identification unit that identifies the type of tool of the tool to be identified by recognizing its shape; a tool dimension feature quantity measuring unit that refers to the tool specification database to identify tool dimension feature items corresponding to the type of tool of the tool to be identified and measures the tool dimension feature quantity of the tool to be identified related to the identified tool dimension feature items from the image data; and a matching unit that compares the tool dimension feature quantity measured by the tool dimension feature quantity measuring unit with the tool database and calculates the likelihood that the tool to be identified is that tool for each tool registered in the tool database, wherein the device identifies which tool registered in the tool database the tool to be identified is based on the likelihood calculated by the matching unit. [Prior art documents] [Patent Documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2023-108332 [Patent Document 2] Japanese Patent Publication No. 2023-104771 [Overview of the project] [Problems that the invention aims to solve]

[0005] The present invention aims to provide an article information management system that can efficiently search for similar articles. [Means for solving the problem]

[0006] The article information management system according to the present invention comprises: a shape data extraction unit that extracts a portion of shape data representing the shape of an article from 3D model data representing the three-dimensional shape of the article; a feature quantity generation unit that generates feature quantities relating to the shape of the article based on the shape data of the article extracted by the shape data extraction unit; a feature quantity storage unit that stores the feature quantities of a plurality of articles in association with each of these articles; and a search unit that searches for articles with similar three-dimensional shapes based on the feature quantities generated by the feature quantity generation unit and the feature quantities stored in the feature quantity storage unit.

[0007] Preferably, the shape data extraction unit extracts shape data indicating the side shape or cross-sectional shape of the article, partial point cloud data indicating the three-dimensional shape of a part of the article, voxel data constituting the three-dimensional shape of the article, or 3D surface data indicating the surface shape of the article from the 3D model data, and the feature quantity generation unit generates the feature quantities using artificial intelligence that infers the three-dimensional shape based on the shape data extracted by the shape data extraction unit.

[0008] Preferably, the article is a product manufactured through a processing treatment, and the system further includes a product information storage unit that stores, in association with each of a plurality of products, a processing program used to manufacture each product, order history information for each product, information about subcontractors who cooperated in the manufacture of each product, inspection data related to the inspection of each product, feedback data regarding troubles or defects of each product, or photographs or illustrations of each product, and a product information presentation unit that presents information corresponding to the product retrieved by the search unit from the information stored in the product information storage unit.

[0009] Preferably, the shape data extraction unit extracts multiple two-dimensional image data of the article viewed from multiple viewpoints, or cross-sectional shape data of the article cut into multiple cross-sections, or a combination thereof, as shape data of the article, and further comprises a shape estimation unit that infers the three-dimensional shape of the article based on the shape data extracted by the shape data extraction unit, and the feature quantity generation unit generates feature quantities calculated for inferring the three-dimensional shape using the shape estimation unit.

[0010] The article information management method according to the present invention includes: a shape data extraction step in which a computer extracts some shape data indicating the shape of an article from 3D model data indicating the three-dimensional shape of the article; a feature generation step in which the computer generates feature quantities relating to the shape of the article based on the shape data of the article extracted in the shape data extraction step; a feature storage step in which the computer stores the feature quantities of a plurality of articles in association with each of these articles; and a search step in which the computer searches for articles with similar three-dimensional shapes based on the feature quantities generated in the feature generation step and the feature quantities stored in the feature storage step.

[0011] The program according to the present invention causes a computer to perform the following steps: a shape data extraction step of extracting some shape data representing the shape of an article from 3D model data representing the three-dimensional shape of the article; a feature generation step of generating feature quantities related to the shape of the article based on the shape data of the article extracted in the shape data extraction step; a feature storage step of storing the feature quantities of a plurality of articles in association with each of these articles; and a search step of searching for articles with similar three-dimensional shapes based on the feature quantities generated in the feature generation step and the feature quantities stored in the feature storage step.

[0012] The article information management system according to the present invention includes: an input receiving unit for receiving article information; a stage identification unit for identifying the manufacturing stage of the product to which the article information received by the input receiving unit belongs; a feature generation unit for generating feature quantities of the article information received by the input receiving unit; an article estimation unit for inferring an article based on the feature quantities generated by the feature generation unit, using a trained model corresponding to the manufacturing stage identified by the stage identification unit; an article management information storage unit for associating and storing the article at each manufacturing stage with the management information of the article at each manufacturing stage; and a related information retrieval unit for searching the article management information stored in the article management information storage unit based on the article estimated by the article estimation unit.

[0013] Preferably, the trained models include trained models corresponding to at least a component, which is the smallest constituent unit of the product; a semi-finished product, which is in the middle of the manufacturing process and is composed of such components; and a finished product, which is the completed product.

[0014] Preferably, the management information stored in the item management information storage unit includes drawing data for each manufacturing stage of the product, and the search unit searches for drawing data related to the item stored in the item management information storage unit based on the item inferred by the item analogy unit.

[0015] Preferably, the input receiving unit receives information about the shape of a part, a semi-finished product, or a finished product, and further includes a shape data extraction unit that extracts some shape data indicating the shape of the article, and a shape estimation unit that infers the three-dimensional shape of the article based on the shape data extracted by the shape data extraction unit, and the stage identification unit identifies the manufacturing stage from the three-dimensional shape inferred by the shape estimation unit based on the shape information received by the input receiving unit.

[0016] The article information management method according to the present invention comprises: an input reception step in which a computer receives article information; a stage identification step in which the computer identifies the manufacturing stage of the product to which the article information received in the input reception step belongs; a feature generation step in which the computer generates feature quantities of the article information received in the input reception step; an article inference step in which the computer uses a trained model corresponding to the manufacturing stage identified in the stage identification step to infer an article based on the feature quantities generated in the feature generation step; an article management information storage step in which the computer stores articles at each manufacturing stage in association with management information of the articles at each manufacturing stage; and an associated information retrieval step in which the computer searches for article management information stored in the article management information storage step based on the article inferred in the article inference step.

[0017] The program according to the present invention causes a computer to execute the following steps: an input reception step for receiving item information; a stage identification step for identifying the manufacturing stage of the product to which the item information received in the input reception step belongs; a feature generation step for generating feature quantities for the item information received in the input reception step; an item inference step for inferring an item based on the feature quantities generated in the feature generation step, using a trained model corresponding to the manufacturing stage identified in the stage identification step; an item management information storage step for storing items at each manufacturing stage in association with management information for items at each manufacturing stage; and a related information retrieval step for searching for item management information stored in the item management information storage step based on the item inferred in the item inference step. [Effects of the Invention]

[0018] According to the present invention, similar articles can be efficiently searched for. [Brief explanation of the drawing]

[0019] [Figure 1] This diagram illustrates the overall configuration of the item information management system 1. [Figure 2] It is a diagram illustrating the hardware configuration of the article information management device 3. [Figure 3] It is a diagram illustrating the functional configuration of the article information management device 3. [Figure 4] It is a diagram illustrating the AI used by the search unit 330 for searching. [Figure 5] It is a diagram illustrating the product information stored in the product information DB 620. [Figure 6] It is a flowchart for explaining the learning process (S10) by the article information management system 1. [Figure 7] It is a flowchart for explaining the extraction process (S20) of shape data from 3D model data by the shape data extraction unit 310. [Figure 8] It is a flowchart for explaining the feature quantity generation process (S30) by the feature quantity generation unit 320. [Figure 9] It is a flowchart for explaining the search process (S40) by the article information management system 1. [Figure 10] It is a flowchart for explaining the learning process (S50) by the learning unit 360. [Figure 11] It is a diagram illustrating the functional configuration of the article information management device 3 in Embodiment 2. [Figure 12] It is a table illustrating the article management information table stored in the article information management DB 630 in Embodiment 2. [Figure 13] It is a flowchart for explaining the search process (S60) by the article information management system 1 in Embodiment 2.

Mode for Carrying Out the Invention

[0020] [Embodiment 1] First, the background of the present invention will be described. In order to improve efficiency in manufacturing, sharing information between similar products leads to increased productivity and reduced labor. However, finding similar products among a vast number of products is difficult. For example, even when trying to use a quotation for a similar product as a reference, it was difficult to find similar products, and creating a quotation took a long time. Also, even if the product identification code was known, it was not possible to determine which products were similar to the product using only the product code. Therefore, in view of the above problems, the present invention makes it possible to easily search for similar products by searching for similar products based on 3D model data of the product, and further provides usable information by also presenting information corresponding to products that have been determined to be similar.

[0021] Embodiments of the present invention will be described below with reference to the drawings. However, the scope of the present invention is not limited to the illustrated examples. Figure 1 is a diagram illustrating the overall configuration of the item information management system 1. As illustrated in Figure 1, the item information management system 1 comprises an item information management device 3, a user terminal 5, and an existing system 7. The product information management device 3 is a computer that searches for similar products based on the imported 3D CAD data (hereinafter referred to as 3D model data). Specifically, the product information management device 3 searches for similar products based on the accumulated product features and the features based on the imported 3D model data. The product information management device 3 searches for similar products based on the overall similarity of the three-dimensional shape (hereinafter referred to as 3D shape) of the 3D model data, the similarity of the side shape or cross-sectional shape of the 3D shape, and the similarity of a part of the 3D shape. In addition, the product information management device 3 cooperates with the existing system 7 and extracts information on the searched products from the existing system 7. The product information management device 3 may also be a cloud server. User terminal 5 is a computer terminal used by the user, and it transmits 3D model data of the item to be searched for to the item information management device 3. User terminal 5 also displays the search results from the item information management device 3. Existing system 7 is an existing system that has information related to the item to be searched, such as a sales management system or a production management system.

[0022] Figure 2 is a diagram illustrating the hardware configuration of the item information management device 3. As illustrated in Figure 2, the item information management device 3 includes a CPU 500, a GPU 502, an NPU 504, an HDD 506, a memory 508, a display device 510, a network interface 512 (network IF 512), and an input device 514, and these components are connected to each other via a bus 516. CPU500 is, for example, a central processing unit. The GPU502 is a unit used, for example, to perform parallel processing of data. The NPU504 is, for example, a processor specifically designed for AI processing. The HDD506 is, for example, a hard disk drive, which stores computer programs and other data files as a non-volatile recording device. Memory 508 is, for example, volatile memory and functions as main memory. The display device 510 is, for example, a liquid crystal display. The network interface IF512 is an interface for wired or wireless communication, and enables communication, for example, between the user terminal 5 and the existing system 7. The input device 514 is, for example, a keyboard and a mouse.

[0023] Figure 3 is a diagram illustrating the functional configuration of the item information management device 3. As illustrated in Figure 3, the item information management device 3 in this example has an item information management program 30 installed, and is configured with a 3D model database 600 (3D model DB600), a feature database 610 (feature DB610), and a product information storage database (product information DB620). The item information management program 30 includes a data reading unit 300, a shape data extraction unit 310, a feature quantity generation unit 320, a search unit 330, a sorting unit 340, a product information presentation unit 350, and a learning unit 360. The shape data extraction unit 310 further includes a cross-section generation unit 312, an image generation unit 314, and a data conversion unit 316. The feature quantity generation unit 320 further includes an input reception unit 322, a shape analogy unit 324, and a feature extraction unit 326. The learning unit 360 further includes a reconstruction unit 362, a reconstruction 2D image generation unit 364, and a comparison unit 366. Furthermore, part or all of the item information management device program 30 may be implemented by hardware such as an ASIC. The item information management device program 30 is stored on a recording medium such as a CD-ROM and installed on the item information management device 3 via this recording medium.

[0024] In the item information management program 30, the data reading unit 300 reads 3D model data representing the 3D shape of the item to be searched from the 3D model DB 600. The items stored in the 3D model DB 600 are products manufactured through processing. The shape data extraction unit 310 extracts some shape data indicating the shape of the article read by the data reading unit 300. Specifically, the shape data extraction unit 310 creates images, point cloud data, voxel data, and surface data of the 3D model data read by the data reading unit 300. The images of the 3D model data include images of the 3D model taken from any direction, images of the side, and cross-sectional images. Specifically, the cross-section generation unit 312 generates 3D model data of the side shape or cross-sectional shape of the article read by the shape data reading unit 300 (hereinafter, the side shape or cross-sectional shape will be referred to as cross-sectional 3D model data). A cross-sectional shape is a partial shape obtained by virtually cutting the 3D model data at a specific position. A side shape is the contour of the cut surface that appears when the 3D model data is cut at a specific position. The cross-section generation unit 312 cuts the 3D model data with a specific plane or curved surface to visualize the internal structure and shape of the 3D model.

[0025] The 2D image generation unit 314 generates a 2D image of an article based on 3D model data stored in the 3D model DB 600, or cross-sectional 3D model data of the article generated by the cross-sectional generation unit 312. Specifically, the 2D image generation unit 314 generates images of the 3D model taken from arbitrary distances and angles, and images of the cross-sectional 3D model data taken from arbitrary distances and angles. The data conversion unit 316 converts the 3D model data of the item read by the data reading unit 300 into point cloud data, voxel data, and surface data. Point cloud data is data that represents the shape and surface of the 3D model data using coordinates in three-dimensional space. Voxel data is data that represents the 3D model data using the smallest unit, a cube. Surface data is data that visually represents the appearance and shape of the 3D model data.

[0026] The feature generation unit 320 generates feature quantities related to the shape of an article based on the shape data of the article extracted by the shape data extraction unit 310. The feature generation unit 320 uses artificial intelligence for processing and includes a shape analogy unit 324 and a feature extraction unit 326. Specifically, the shape estimation unit 324 is artificial intelligence and estimates the 3D shape based on the shape data extracted by the shape data extraction unit 310. Specifically, the shape estimation unit 324 takes the shape data extracted by the shape data extraction unit 310 as input information and outputs the 3D shape based on the model learned by the learning unit 360. The feature extraction unit 326 extracts features from the 3D shape inferred by the shape inference unit 324. Features are variables generated during the process by which the shape inference unit 324 infers the 3D shape. Features are the number of features of the 3D shape and are a set of numerical combinations. Specifically, features are composed of multiple dimensions, and in this example, the features are 10-dimensional. Features generated by the feature generation unit 320 are stored in the feature database 610. The feature database 610 stores feature tables based on the input data. Features extracted using multiple partial images as input data are stored in the 2D partial feature table, and features extracted using six images as input data are stored in the 2D whole feature table. Features extracted using multiple images as input data are stored in the 3D feature table, and features extracted using point cloud data as input data are stored in the 3D point cloud feature table. Features extracted from voxel data are stored in the 3D voxel feature table, and features extracted from surface data are stored in the 3D surface feature table.

[0027] The search unit 330 searches for items with similar three-dimensional shapes based on the features generated by the feature generation unit 320 and the features stored in the feature database 610. Specifically, the search unit 330 determines the similarity between the features generated by the feature generation unit 320 and the features stored in the feature database 610. The search unit 330 calculates the similarity using cosine similarity. The search unit 330 extracts items that have features stored in the feature database 610 whose similarity is equal to or greater than a predetermined value. The sorting unit 340 displays items extracted by the search unit 330 with a similarity score equal to or greater than a predetermined value, sorted by similarity score, on the client terminal 5. The product information display unit 350 displays information corresponding to the item retrieved by the search unit 330 from the information stored in the product information DB 620. Specifically, the product information display unit 350 displays processing programs, order history information, information on subcontractors who cooperated in manufacturing, inspection data related to inspections, feedback data regarding troubles or defects, or photographs or illustrations corresponding to the item determined to be similar by the search unit 330.

[0028] The learning unit 360 constructs a learning model based on the features generated by the feature generation unit 320. Specifically, the reconstruction unit 362 generates 3D model data based on the features generated by the feature generation unit 320. Hereinafter, the 3D model data generated by the reconstruction unit 362 will be referred to as "generated 3D". The reconstructed 2D image generation unit 364 generates a 2D image based on the generated 3D image generated by the reconstruction unit 362. The comparison unit 366 compares the shape extracted by the shape data extraction unit 310 with the reconstructed shape. Specifically, it compares a 2D image obtained from 3D model data with a 2D image obtained from a generated 3D model reconstructed based on features. Specifically, it compares a 2D image generated by the 2D image generation unit 314 with a 2D image generated by the reconstructed 2D image generation unit 364. The comparison unit 366 learns to reduce the error in the comparison results.

[0029] Here, we will explain how the search unit 330 searches for similar products. As illustrated in Figure 4, the search unit 330 determines the similarity of products using 2D overall similarity AI, 2D partial similarity AI, 3D similarity AI, 3D point cloud similarity AI, 3D voxel similarity AI, and 3D surface similarity AI. The 2D overall similarity AI uses multiple side or cross-sectional images to determine the similarity of the side or cross-section of the 3D shape, and searches for products that look similar from the side or have a similar cross-sectional shape. The 2D partial similarity AI uses multiple cropped images from side or cross-sectional views to compare parts of 3D shapes, determine similarity, and search for products with similar 3D shapes. 3D Similarity AI uses multiple images from arbitrary viewpoints and arbitrary cross-sectional images created from a 3D model to compare 3D shapes, determine similarity, and search for products with similar overall part shapes. The 3D point cloud similarity AI uses point cloud data created from 3D models to compare parts of the 3D shapes and determine similarity, searching for products that have similar partial shapes such as bumps, holes, gears, etc. The 3D voxel similarity AI uses voxel data created from 3D models to compare the entire 3D shape to determine similarity and search for products with similar surfaces that make up the parts. 3D surface similarity AI determines similarity using surface data created from 3D models.

[0030] Figure 5 is an example of product information stored in the product information DB620. As illustrated in Figure 5, the product information DB620 stores product identification information, associated with processing programs used in product manufacturing, product order history information, information on subcontractors who cooperated in product manufacturing, inspection data related to product inspection, feedback data regarding product troubles or defects, and product photographs or illustrations. Inspection data related to product inspection includes, for example, shipping inspection, receiving inspection, and inspection management data, such as dimensional measurement data and surface roughness data. Feedback data is information necessary for quality control, such as trouble information and defect information. In Figure 5, the product information DB620 associates each data item with product identification information. However, it is not limited to this; for example, the storage location of each data item in the existing system 7, with which the item information management device 3 is linked, may also be associated with the product identification information and stored accordingly.

[0031] Figure 6 is a flowchart illustrating the learning process (S10) performed by the item information management system 1. As shown in Figure 6, in step 100 (S100), the data reading unit 300 reads the 3D model data of the product specified by the user from the 3D model DB 600. The read 3D model data becomes the criterion for the search. In step 105 (S105), the shape data extraction unit 310 extracts some shape data that indicates the shape of the product from the 3D model data read by the data reading unit 300. In step 110 (S110), the shape estimation unit 324 estimates the 3D shape based on the shape data extracted by the shape data extraction unit 310. The feature extraction unit 326 extracts the feature quantities of the 3D shape estimated by the shape estimation unit 324. In step 115 (S115), the reconstruction unit 362 generates a generated 3D based on the features generated by the feature generation unit 320. In step 120 (S120), the comparison unit 366 compares the shape data obtained from the 3D model data with the information restored by the restoration unit 362, and learns to reduce the error in the comparison result.

[0032] Figure 7 is a flowchart illustrating the process (S20) of extracting shape data from 3D model data by the shape data extraction unit 310. As shown in Figure 7, in step 200 (S200), the cross-section generation unit 312 generates cross-sectional 3D model data of the 3D model based on the 3D model data read by the data reading unit 300. In step 205 (S205), the 2D image generation unit 314 generates multiple 2D images of the 3D model viewed from multiple viewpoints, based on the cross-sectional 3D model data generated by the cross-sectional generation unit 312 and the 3D model data read by the reading unit 300. In step 210 (S210), the data conversion unit 316 converts the 3D model data read by the data reading unit 300 into point cloud data, voxel data, and surface data. The feature generation unit 320 receives the generated 2D image, converted point cloud data, voxel data, and surface data as input and extracts features.

[0033] Figure 8 is a flowchart illustrating the feature generation process (S30) performed by the feature generation unit 320. The shape estimation unit 324 estimates the 3D shape using a different learning model for each input information. The input information includes multiple partial images, one or more faces of a 3D shape, multiple images, point cloud data, voxel data, and surface data. The feature extraction process for each input will be described below. In step 300 (S300), the input receiving unit 322 receives input of multiple partial images generated by the 2D image generation unit 314. In step 305 (S305), the shape estimation unit 324 estimates a 2D partial shape from the multiple partial images received in S300, based on a learning model that has learned partial images as input information. In step 310 (S310), the feature extraction unit 326 extracts features generated in S305 during the process in which the shape estimation unit 324 estimates a 2D partial image from multiple partial images.

[0034] In step 315 (S315), the input unit 322 receives input of six images generated by the 2D image generation unit 314. In step 320 (S320), the shape estimation unit 324 estimates the overall 2D shape from the six images received in S315, based on a learning model that has learned the six images as input information. In step 325 (S325), the feature extraction unit 326 extracts features generated in S320 during the process in which the shape estimation unit 324 estimates the overall 2D shape from the six images. In step 330 (S330), the input unit 322 receives input of multiple images generated by the 2D image generation unit 314. In step 335 (S335), the shape estimation unit 324 estimates the 3D shape from the multiple images received in S330, based on a learning model that has learned from multiple images as input information. In step 340 (S340), the feature extraction unit 326 extracts features generated in S330 by the shape estimation unit 324 during the process of inferring a 3D shape from multiple images.

[0035] In step 345 (S345), the input unit 322 receives the point cloud data converted by the data conversion unit 316. In step 350 (S350), the shape estimation unit 324 estimates the 3D shape from the point cloud data received in S345, based on a learning model that has learned point cloud data as input information. In step 355 (S355), the feature extraction unit 326 extracts features generated in S350 during the process in which the shape estimation unit 324 infers a 3D shape from the point cloud data. In step 360 (S360), the input unit 322 receives the voxel data converted by the data conversion unit 316. In step 365 (S365), the shape estimation unit 324 estimates the 3D shape from the voxel data received in S360 based on a learning model that has learned the voxel data as input information. In step 370 (S370), the feature extraction unit 326 extracts features, which are variables generated in the process by which the shape estimation unit 324 infers the 3D shape from the voxel data in S365.

[0036] In step 375 (S375), the input unit 322 receives the surface data converted by the data conversion unit 316. In step 380 (S380), the shape estimation unit 324 estimates the 3D shape from the surface data received in S375, based on a learning model that has learned the surface data as input information. In step 385 (S385), the feature extraction unit 326 extracts features generated in S380 during the process in which the shape estimation unit 324 infers the 3D shape from the surface data. The extracted features are used for learning by the learning unit 360 and for retrieval processing by the retrieval unit 330, and are further stored in the feature database 610.

[0037] Figure 9 is a flowchart illustrating the search process (S40) performed by the item information management system 1. As shown in Figure 9, in step 400 (S400), the data reading unit 300 reads the 3D model data of the product specified by the user from the 3D model DB 600. The read 3D model data becomes the criterion for the search. In step 405 (S405), the shape data extraction unit 310 extracts some shape data that indicates the shape of the product from the 3D model data read by the data reading unit 300. In step 410 (S410), the shape estimation unit 324 estimates the 3D shape based on the shape data extracted by the shape data extraction unit 310. The feature extraction unit 326 extracts the features of the 3D model estimated by the shape estimation unit 324. In step 415 (S415), the search unit 330 calculates the similarity between the features extracted by the feature extraction unit 326 and the product features stored in the feature database 610. The search unit 330 identifies products that have features with a similarity of a predetermined value or higher. The sorting unit 340 sorts the products identified by the search unit 330 in order of similarity. In step 420 (S420), the product information provision unit 350 retrieves data from the product information DB 620 based on the product code of the product identified by the search unit 330, including processing programs used in the manufacture of the product, order history information for the product, information on subcontractors who cooperated in the manufacture of the product, inspection data related to product inspection, feedback data regarding product troubles or defects, and photos or illustrations of the product, and displays them on the user terminal 5 in association with the product.

[0038] Figure 10 is a flowchart illustrating the learning process (S50) performed by the learning unit 360. The learning unit 360 constructs a learning model based on the input information received by the input receiving unit 322. Specifically, the input receiving unit 322 accepts multiple partial images, six-sided images, multiple images, point cloud data, voxel data, and surface data as input data, and constructs a learning model for each of these input data types. Figure 10 shows that in step 500 (S500), the reconstruction unit 362 visualizes the 2D partial image based on the feature quantities of the partial image extracted by the feature extraction unit 326. In step 505 (S505), the 2D partial image extracted by the shape data extraction unit 310 is compared with the 2D partial image restored in S500, and the model of the 2D partial image is trained to reduce the difference. In step 510 (S510), the reconstruction unit 362 visualizes the entire 2D image based on the features of the entire 2D image extracted by the feature extraction unit 326. In step 515 (S515), the 2D overall image extracted by the shape data extraction unit 310 is compared with the 2D overall image restored in S510, and a model of the 2D overall image is trained to reduce the difference. In step 520 (S520), the reconstruction unit 362 generates a 3D overall image based on the features of the 3D overall image acquired by the feature extraction unit 326.

[0039] In step 525 (S525), the reconstructed 2D image generation unit 364 visualizes the generated 3D image produced by the reconstruction unit 362 as a 2D image. In step 530 (S530), multiple 2D images extracted by the shape data extraction unit 310 are compared with multiple 2D images visualized by the reconstructed 2D image generation unit 364, and a model of the overall 3D image is trained to reduce the difference. In step 535 (S535), the feature quantities of the point cloud data extracted by the feature extraction unit 326 are input to the classifier, which then classifies the shape from the input feature quantities of the point cloud data and learns a 3D point cloud model. In step 540 (S540), the reconstruction unit 362 generates a 3D shape based on the voxel data acquired by the feature extraction unit 326. In step 545 (S545), the voxel data extracted by the shape data extraction unit 310 is compared with the voxel data of the reconstructed 3D shape in S550, and a 3D voxel model is trained to minimize the difference. In step 550 (S550), the reconstruction unit 362 generates a 3D shape based on the surface data extracted by the feature extraction unit 326. In step 555 (S555), the surface data acquired by the shape data extraction unit 310 is compared with the surface data of the reconstructed 3D shape in S550, and the model of the 3D surface is trained to minimize the difference.

[0040] As explained above, the item information management system 1 has a learning model based on input data, and therefore has multiple methods for searching for items. Specifically, because it can compare parts of 3D models, it can search for similar products even if the whole model cannot be compared. Furthermore, because it can search even if the overall orientation of the 3D models is not aligned, it can include a larger number of products in the search target. Furthermore, the product information management system 1 presents various product information along with similar products in the search results, allowing users to obtain useful information.

[0041] [Embodiment 2] Next, another embodiment of the present invention, Embodiment 2, will be described. The item information management system 1 in Embodiment 2 retrieves and presents relevant information from the information managed by the BOM (Bill of Materials) system based on the manufacturing stage of the item. The BOM system is a system that manages the items that make up a product in a hierarchical structure of parent-child relationships between items, and is a system that manages design information, manufacturing information, procurement information, and maintenance information of items in association with the items.

[0042] In Embodiment 2, the article information management system 1 omits redundant explanations by assigning the same reference numerals to elements that have substantially the same functions and configuration as those in Embodiment 1.

[0043] Figure 11 is a diagram illustrating the functional configuration of the item information management device 3 in Embodiment 2. As illustrated in Figure 11, the item information management device 3 in this example has the item information management program 40 installed, and is configured with a 3D model database 600 (3D model DB600), a feature database 610 (feature DB610), and an item information management database (item information management DB630). The item information management program 40 includes an input reception unit 400, a shape data extraction unit 310, a feature quantity generation unit 410, a stage identification unit 420, an item analogy unit 430, a related information retrieval unit 440, and a learning unit 360. The feature quantity generation unit 410 further includes an input reception unit 322, a shape analogy unit 324, and a feature extraction unit 326. Furthermore, part or all of the item information management device program 40 may be implemented by hardware such as an ASIC. The item information management device program 40 is stored on a recording medium such as a CD-ROM and installed on the item information management device 3 via this recording medium.

[0044] The input reception unit 400 receives item information. Specifically, the input reception unit 400 receives information about the shape of an item entered by the user from the user terminal 5. This information about the shape of an item may be, for example, 3D model data.

[0045] The shape data extraction unit 310 extracts shape data indicating the shape of an object from the 3D model data received by the input receiving unit 400. Specifically, the shape data extraction unit 310 creates image, point cloud data, voxel data, and surface data from the received 3D model data. The shape estimation unit 324 is artificial intelligence that estimates a 3D shape based on the shape data extracted by the shape data extraction unit 310. Specifically, the shape estimation unit 324 takes the shape data extracted by the shape data extraction unit 310 as input information and outputs a 3D shape based on the model learned by the learning unit 360.

[0046] The feature generation unit 410 generates feature quantities for the item indicated by the item information received by the input reception unit 400. Specifically, the feature generation unit 410 generates feature quantities related to the shape of the item based on the shape data extracted by the shape data extraction unit 310. More specifically, the feature extraction unit 326 generates feature quantities from the 3D shape inferred by the shape analogy unit 324, which is received by the input reception unit 322. Feature quantities are variables generated during the process in which the shape analogy unit 324 infers the 3D shape. Note that feature quantities are the number of features of the 3D shape and are a set of combinations of numbers. Specifically, feature quantities are composed of multiple dimensions, and in this example, the feature quantities are 10-dimensional.

[0047] The stage identification unit 420 identifies the manufacturing stage of the product to which the item information received by the input receiving unit 400 belongs. Specifically, the stage identification unit 420 identifies whether the item with the received information is a part, a semi-finished product, or a finished product. More specifically, based on the 3D shape output by the shape estimation unit 324, the stage identification unit 420 identifies whether the item indicated by the 3D shape is in the manufacturing stage of a part, a semi-finished product, or a finished product.

[0048] The item estimation unit 430 uses a trained model corresponding to the manufacturing stage identified by the stage identification unit 420 to estimate the item indicated by the item information received by the input reception unit 400, based on the feature quantities generated by the feature quantity generation unit 410. Specifically, if the item identified by the stage identification unit 420 is a part, the item estimation unit 430 estimates the item using a trained model for parts; if it is a semi-finished product, it uses a trained model for semi-finished products; or if it is a finished product, it uses a trained model for finished products.

[0049] The related information retrieval unit 440 searches for item management information stored in the item information management DB 630 based on the item inferred by the item inference unit 430. Specifically, the related information retrieval unit 440 searches for item management information stored in the item information management DB 630 based on the identification information of the item inferred by the item inference unit 430. For example, the related information retrieval unit 440 searches for drawing data associated with the item inferred by the item inference unit 430 and presents it to the user terminal 5.

[0050] Here, we will explain the item management information table stored in the item information management DB630, which is illustrated in Figure 12. The information stored in the inventory management information table is information managed within the BOM system. The item management information table contains information about the items that make up a product. Specifically, as illustrated in Figure 12, the item management information table stores information about the item, associated with the item ID, including configuration information (part number, name, number used, etc.), attribute information (dimension data, material data, etc.), design information (drawing data, specification data, etc.), manufacturing information (processing data, assembly procedure data, etc.), quality information (inspection data, defect history data, etc.), outsourcing information (cooperating company data, etc.), monetary information (quotation data, etc.), and image data of the item.

[0051] Figure 13 is a flowchart illustrating the search process (S60) performed by the item information management system 1 in Embodiment 2. In step 600 (S600), the input receiving unit 400 receives 3D model data from the user terminal 5. In step 605 (S605), the shape data extraction unit 310 extracts some shape data indicating the shape of an item from the 3D model data received by the input receiving unit 400. Specifically, the shape data extraction unit 310 creates image, point cloud data, voxel data, and surface data from the 3D model data received by the input receiving unit 400. The input receiving unit 322 receives the shape data extracted by the shape data extraction unit 310, and the shape estimation unit 324 estimates the 3D shape based on the shape data extracted by the shape data extraction unit 310. Specifically, the shape estimation unit 324 takes the shape data extracted by the shape data extraction unit 310 as input information and outputs the 3D shape based on the model learned by the learning unit 360. The stage identification unit 420 identifies whether the article indicated by the 3D shape is in the manufacturing stage of a part, semi-finished product, or finished product, based on the 3D shape output by the shape estimation unit 324. In step 610 (S610), the feature generation unit 410 generates feature quantities of the input data article received by the input reception unit 400. Specifically, it generates feature quantities related to the shape of the article based on the shape data extracted by the shape data extraction unit 310.

[0052] In step 615 (S615), the item estimation unit 430 uses a trained model corresponding to the manufacturing stage identified by the stage identification unit 420 to estimate the item represented by the 3D model data received by the input reception unit 400, based on the features generated by the feature generation unit 410. In step 620 (S620), the related information retrieval unit 440 searches for drawing data associated with the item stored in the item information management DB 630 based on the identification information of the item inferred by the item inference unit 430.

[0053] As described above, the article information management system 1 in Embodiment 2 can identify the manufacturing stage (parts, semi-finished products, finished products) to which an article belongs based on the shape information of the article, search for similar articles using a learning model corresponding to that manufacturing stage, and present information related to the articles in the search results. This allows for the acquisition of drawing data, manufacturing history, outsourcing information, inspection results, and other data associated with an item, based on its shape information, at any stage—parts, semi-finished products, or finished products—in design, manufacturing, and maintenance operations, significantly reducing the time and effort required for searching. [Explanation of Symbols]

[0054] 1…Item Information Management System 3…Article information management device 5…User terminal 7…Existing systems 30…Item Information Management Program 300...Data reading unit 310...Shape data extraction unit 312...Cross section generation part 314...2D Image Generation Unit 316...Data conversion unit 320...Feature generation unit 322... Input reception section 324…Shape analogy part 326...Feature extraction unit 330... Search section 340... Sort section 350…Product information provision department 360...Learning Department 362…Restoration Department 364...Restoration 2D Image Generation Unit 366...Comparison section 40…Item Information Management Program 400... Input reception section 410...Feature generation unit 420... Stage Specific Section 430…Article analogy section 440... Related Information Search Department

Claims

1. A shape data extraction unit extracts some shape data representing the shape of an article from 3D model data representing the three-dimensional shape of the article, A feature generation unit generates feature quantities related to the shape of an article based on the shape data of the article extracted by the shape data extraction unit, A feature storage unit that stores the feature quantities of multiple items in association with each of these items, A search unit searches for articles with similar three-dimensional shapes based on the feature quantities generated by the feature quantity generation unit and the feature quantities stored by the feature quantity storage unit. A product information management system having [specific features / features].

2. The shape data extraction unit extracts shape data indicating the side shape or cross-sectional shape of the article, partial point cloud data indicating the three-dimensional shape of a part of the article, voxel data constituting the three-dimensional shape of the article, or 3D surface data indicating the surface shape of the article from the 3D model data. The feature generation unit generates the feature quantities using artificial intelligence that infers the three-dimensional shape based on the shape data extracted by the shape data extraction unit. The article information management system according to claim 1.

3. The aforementioned article is a product manufactured through processing, A product information storage unit stores, associated with each of multiple products, the processing program used to manufacture each product, the order history information for each product, information about the outsourcing companies that cooperated in the manufacturing of each product, inspection data related to the inspection of each product, feedback data regarding troubles or defects of each product, or data of photographs or illustrations of each product. A product information display unit presents information corresponding to the product searched by the search unit from among the information stored in the product information storage unit. The article information management system according to claim 1, further comprising:

4. The shape data extraction unit extracts multiple two-dimensional image data of the article viewed from multiple viewpoints, or cross-sectional shape data of the article cut into multiple cross-sections, or a combination thereof, as the shape data of the article. A shape estimation unit infers the three-dimensional shape of an article based on the shape data extracted by the shape data extraction unit. It further possesses, The feature generation unit generates feature quantities calculated for inferring three-dimensional shapes using the shape analogy unit. The article information management system according to claim 1.

5. A shape data extraction step in which a computer extracts some shape data representing the shape of an article from 3D model data representing the three-dimensional shape of the article, A feature generation step in which a computer generates feature quantities related to the shape of an article based on the shape data of the article extracted in the shape data extraction step, A feature storage step in which a computer stores feature quantities associated with each of several items, A computer searches for articles with similar three-dimensional shapes based on the features generated in the feature generation step and the features stored in the feature storage step. A method for managing information on articles.

6. A shape data extraction step of extracting some shape data representing the shape of an article from 3D model data representing the three-dimensional shape of the article, A feature generation step, which generates feature quantities related to the shape of an article based on the shape data of the article extracted in the shape data extraction step, A feature storage step involves associating the feature quantities of multiple items with each of them and storing the feature quantities of these items, A search step that searches for articles with similar three-dimensional shapes based on the features generated in the feature generation step and the features stored in the feature storage step. A program that causes a computer to execute something.

7. An input reception unit that receives item information, A stage identification unit identifies the manufacturing stage of the product to which the item information received by the input receiving unit belongs, A feature quantity generation unit generates feature quantities for items based on the item information received by the input receiving unit, An item estimation unit uses a trained model corresponding to the manufacturing stage identified by the stage identification unit to estimate an item based on the features generated by the feature generation unit, A product management information storage unit stores products at each manufacturing stage in association with management information for products at each manufacturing stage. A related information retrieval unit searches for item management information stored in the item management information storage unit based on the item inferred by the item inferred by the item inferred unit. has Item information management system.

8. The aforementioned trained models include, at a minimum, trained models corresponding to parts, which are the smallest constituent units of a product; semi-finished products, which are in the process of being manufactured and assembled from such parts; and finished products, which are completed products. The article information management system according to claim 7.

9. The management information stored in the aforementioned product management information storage unit includes drawing data for each stage of product manufacturing, The related information retrieval unit retrieves drawing data related to the article stored in the article management information storage unit, based on the article inferred by the article inference unit. The article information management system according to claim 7.

10. The aforementioned input receiving unit receives information regarding the shape of a part, semi-finished product, or finished product. A shape data extraction unit that extracts some shape data indicating the shape of the aforementioned article, Based on the shape data extracted by the shape data extraction unit, a shape estimation unit infers the three-dimensional shape of the article. It further possesses, The aforementioned stage identification unit identifies the manufacturing stage from the three-dimensional shape inferred by the shape estimation unit, based on the shape information received by the input reception unit. The article information management system according to claim 7.

11. The computer receives an input acceptance step for item information, A computer identifies the manufacturing stage of the product to which the item information received in the input acceptance step belongs, A feature generation step in which a computer generates feature quantities for the item information received in the input reception step, The computer uses a trained model corresponding to the manufacturing stage identified in the stage identification step to infer an item based on the features generated in the feature generation step, and the computer infers an item. A computer stores items at each manufacturing stage in association with management information for those items at each manufacturing stage; A related information retrieval step in which the computer searches for item management information stored in the item management information storage step based on the item inferred in the item inferred step. has Product information management method.

12. An input acceptance step for receiving item information, A stage identification step that identifies the manufacturing stage of the product to which the item information received in the input acceptance step belongs, A feature generation step that generates feature quantities for the item information received in the input reception step, An item estimation step in which an item is estimated based on the features generated in the feature generation step, using a trained model corresponding to the manufacturing stage identified in the stage identification step, A step of storing item management information that associates and stores items at each manufacturing stage with management information for items at each manufacturing stage, A related information retrieval step, which searches for the item management information stored in the item management information storage step based on the item inferred in the item inferred step, Make the computer execute it. program.

Images