# US 12,008,706 B2 · Method and device for generating image data for machine learning > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,008,706 B2 - Document ID: US12008706B2 - Status: Granted - Kind: B2 - Application number: 18/249,151 - Filed: 2020-11-30 - Published: 2023-10-12 - Granted: 2024-06-11 - Earliest claimed: 2020-10-21 - Assignees: T.I.M Solution Co., Ltd - Applicants: T.I.M Solution Co., Ltd - Inventors: Gye Hong Lee - Web page: https://patentgrep.com/patent/US12008706B2 ## CPC - G06T 15/10: Geometric effects - G06F 30/15: Vehicle, aircraft or watercraft design - G06F 30/17: Mechanical parametric or variational design - G06F 30/27: using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model - G06N 5/02: Knowledge representation; Symbolic representation - G06N 20/00: Machine learning - G06T 7/001: {using an image reference approach} - G06T 17/30: Polynomial surface description - G06T 2207/20081: Training; Learning - G06T 2207/30108: Industrial image inspection - G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract Provided are a method and an apparatus for generating image data for machine learning. The method may include extracting first image data for a product generated as a result of conducting a first step among a plurality of steps of a manufacturing process of a product based on 3D CAD data for the product, generating a plurality of posture conversion image sets corresponding to a plurality of postures for the product, respectively, based on the first image data, performing image processing through a rendering engine by applying actual image data to respective images of the plurality of posture conversion image sets, and generating a learning data set for the first image data through the image processing. ## Claims ### Claim 1 A method for generating image data for machine learning, which is performed by an apparatus for generating image data for machine learning, the method comprising: extracting first image data for a product generated as a result of conducting a first step among a plurality of steps of the manufacturing process of the product, based on 3D CAD data for the product; generating a plurality of posture conversion image sets corresponding to a plurality of postures for the product, respectively, based on the first image data; performing image processing through a rendering engine by applying actual image data to respective images of the plurality of posture conversion image sets; and generating a learning data set for the first image data through the image processing, wherein the performing the image processing through the rendering engine by applying the actual image data includes: receiving physical value information, environment value information, and noise information corresponding to the first step from an infrastructure network database storing a site information corresponding to the first step; and generating the actual image data based on the physical value information, the environment value information, and the noise information. ### Claim 2 (depends on 1) The method for generating image data for machine learning of claim 1, wherein the extracting the first image data includes, matching the 3D CAD data for the product with a first data group including identification information of the product, identification information of a finished product including the product, and information on each step of the manufacturing process. ### Claim 3 (depends on 1) The method for generating image data for machine learning of claim 1, wherein the extracting the first image data includes, matching the first image data with a second data group including identification information of the product, identification information of an intermediate output constituting the product, and information on a process progress rate of the first step compared with the plurality of steps of the manufacturing process. ### Claim 4 (depends on 1) The method for generating image data for machine learning of claim 1, wherein the generating the plurality of posture conversion image sets includes, performing predetermined viewpoint conversion for the first image data, and capturing the viewpoint-converted first image data. ### Claim 5 (depends on 1) The method for generating image data for machine learning of claim 1, wherein the performing the image processing through the rendering engine includes: determining a rendering model based on the plurality of posture conversion image sets and the actual image data; performing the image processing based on the rendering model; generating a third data group including the applied actual image data and information on the rendering model; and matching the third data group with a learning data set for the first image data. ### Claim 6 (depends on 5) The method for generating image data for machine learning of claim 5, wherein the site information includes process information corresponding to the first step, and the method includes: filtering the site information based on the process information; and performing reinforcement learning for the learning data set based on the site information, and the performing the reinforcement learning includes: generating simulation data to be applied to learning data through pre-registered reinforcement learning model based on the site information; and applying the simulation data to the learning data set. ### Claim 7 (depends on 6) The method for generating image data for machine learning of claim 6, wherein the process information includes at least one of a type of process corresponding to the first step, a progress direction of the process, and information on a work target surface of an intermediate output. ### Claim 8 (depends on 7) The method for generating image data for machine learning of claim 7, wherein the filtering the received site information based on the process information includes: extracting pre-matched noise information based on at least one of a type of process type and a progress direction of the process among the process information; and filtering the site information based on the pre-matched noise information. ### Claim 9 (depends on 1) The method for generating image data for machine learning of claim 1, wherein the extracting the first image data includes: identifying a hierarchy of the 3D CAD model for the product; performing a CAD light-weighting process of light-weighting an object determined as a merge target among objects constituting the hierarchy; and generating the 3D CAD data of the product by performing the CAD light-weighting process. ### Claim 10 (depends on 9) The method for generating image data for machine learning of claim 9, wherein the performing the CAD light-weighting process of light-weighting the object determined as the merge target among the objects constituting the hierarchy includes, determining the merge target among the objects based on the manufacturing process of the product. ### Claim 11 (depends on 1) The method for generating image data for machine learning of claim 1, further comprising: extracting second image data for a product generated as a result of conducting a second step among the plurality of steps of the manufacturing process of the product; generating a plurality of posture conversion image sets of the second image data corresponding to a plurality of postures for the product, respectively, based on the second image data; performing image processing through a rendering engine by applying actual image data to respective images of the plurality of posture conversion image sets of the second image data; and generating a learning data set for the second image data through the image processing. ### Claim 12 An apparatus for generating image data for machine learning, the apparatus comprising: one or more processors; a network interface receiving 3D CAD data for a product and a site information; a memory loading a computer program executed by the processor; and a storage storing the computer program, wherein the computer program includes: an operation of extracting first image data for a product generated as a result of conducting a first step among a plurality of steps of the manufacturing process of a product based on the 3D CAD data; an operation of generating a plurality of posture conversion image sets corresponding to a plurality of postures for the product, respectively, based on the first image data; an operation of performing image processing through a rendering engine by applying actual image data to respective images of the plurality of posture conversion image sets; and an operation of generating a learning data set for the first image data through the image processing, wherein the operation of the performing the image processing through the rendering engine by applying the actual image data includes: receiving physical value information, environment value information, and noise information corresponding to the first step from an infrastructure network database storing the site information corresponding to the first step; and generating the actual image data based on the physical value information, the environment value information, and the noise information.