Information processing device, machine learning device, inference device, information processing method, machine learning method, and inference method

Abstract

An information processing device is provided that enables highly accurate retrieval of similar drawings from three-dimensional drawing data. [Solution] The information processing device 2 includes a new processed product information acquisition unit 203 that acquires new processed product information including 3D drawing data of the new processed product, a drawing data conversion unit 204 that converts the 3D drawing data acquired by the new processed product information acquisition unit 203 into 3D geometric data that extracts information related to the three-dimensional structure of the processed product, and a similar drawing extraction unit 205 that extracts drawing data of multiple existing processed products that are similar to the 3D geometric data converted from the 3D drawing data included in the new processed product information from a database 210 in which multiple pieces of existing processed product information including drawing data of existing processed products are registered.

Description

[Technical Field]

[0001] The present invention relates to an information processing device, a machine learning device, an inference device, an information processing method, a machine learning method, and an inference method. [Background technology]

[0002] Systems have been developed that search for and process drawing data similar to input drawing data. For example, Literature 1 discloses a system that searches a database for drawings similar to input drawing data, detects corresponding feature points between the two, and acquires attribute information linked to the corresponding feature points. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7004125 Summary of the Invention [Problem to be solved by the invention]

[0004] However, conventional techniques have not been able to search for similar drawing data with high accuracy for 3D drawing data.

[0005] The present invention has been made in light of the above-mentioned problems, and aims to provide an information processing device, a machine learning device, an inference device, an information processing method, a machine learning method, and an inference method that are capable of searching for similar drawings with high accuracy from 3D drawing data. [Means for solving the problem]

[0006] In order to achieve the above object, an information processing device according to one aspect of the present invention comprises: a new processed product information acquisition unit that acquires new processed product information including 3D drawing data of the new processed product; a drawing data conversion unit that converts the 3D drawing data acquired by the new processed product information acquisition unit into 3D geometric data that extracts information about the three-dimensional structure of the processed product; and a similar drawing extraction unit that extracts, from a database in which a plurality of pieces of existing processed product information including drawing data of the existing processed products is registered, drawing data of the existing processed products that is similar to the 3D geometric data converted from the 3D drawing data included in the new processed product information. Information processing device. [Effects of the Invention]

[0007] According to an information processing device according to an aspect of the present invention, it is possible to search for similar drawings with high accuracy in three-dimensional drawing data.

[0008] Problems, configurations, and effects other than those described above will become apparent from the detailed description of the invention that follows. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is an overall configuration diagram showing an overview of processing of a drawing search system 1 according to an embodiment. [Figure 2] FIG. 10 is a diagram showing an example of conversion from 3D drawing data D11 to 3D geometric data D13. [Figure 3] FIG. 2 is a block diagram showing an example of an information processing device 2. [Figure 4] FIG. 2 is a functional explanatory diagram showing an example of a feature quantity inference model learning unit 202. [Figure 5] 2 is a functional explanatory diagram showing an example of a new processed product information acquisition unit 203 and a drawing data conversion unit 204. FIG. [Figure 6] FIG. 10 is a functional explanatory diagram showing an example of a similar drawing extraction unit 205. [Figure 7] FIG. 9 is a hardware configuration diagram showing an example of a computer 900. [Figure 8] 10 is a flowchart showing an example of the operation (information processing method) of the information processing device 2. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, an embodiment for carrying out the present invention will be described with reference to the drawings. The scope necessary for the explanation to achieve the object of the present invention will be schematically shown, and the scope necessary for explaining the relevant parts of the present invention will be mainly explained, and the parts that are omitted from the explanation will be based on publicly known techniques.

[0011] Fig. 1 is a diagram showing an overall configuration of an example of a drawing search system 1 according to an embodiment. Fig. 2 is a diagram showing an example of conversion from 3D drawing data D11 to 3D geometric data D13.

[0012] 1, the drawing search system 1 is a system that receives order information (new processed product information D10) including new 3D (dimensional) drawing data from a user U and searches for similar drawing data (existing processed product information D20). The drawing search system 1 is also a system that registers and manages various information for searching for similar drawing data in a database 210.

[0013] As an example, the order information may be an order for a processed product that is manufactured by performing one or more processing steps on a material using various processing devices or by hand based on 3D drawing data. Specifically, examples of orders include machine-processed products, buildings that are constructed using architectural processing, and clothing that is cut and sewn. In this embodiment, the order is for a machine-processed product based on 3D drawing data, and a search for drawing data similar to this machine-processed product will be mainly described.

[0014] As shown in FIG. 1, the main data input to the information processing device 2 during a drawing search is new workpiece information D10 related to a new order. As an example, the new workpiece information D10 includes 3D drawing data D11 and additional information D12. The 3D drawing data D11 may be in a STEP format, DWG format, or DXF format, which are commonly used in CAD design, and is specified in a data format dependent on the CAD application. The 3D drawing data D11 may also be in a 3D PDF or eDrawings format, which is used for design review and sharing, and may be selected appropriately depending on the application.

[0015] In a drawing data conversion process (described in detail below), the 3D drawing data D11 is converted into 3D geometric data D13, which extracts information about the three-dimensional structure of the processed product. 3D geometric data is data that quantitatively expresses information about the three-dimensional structure, i.e., shape and dimensions, and is application-independent. Typical data formats for 3D geometric data include voxels, point clouds, and meshes. As an example, as shown in FIG. 2, the 3D drawing data D11 is converted into voxelized 3D geometric data D13, i.e., expressed as a combination of multiple boxes, in the drawing data conversion process. Note that FIG. 2 is a diagram that schematically illustrates voxelization, and the voxel size is appropriately set depending on the analysis accuracy and calculation load.

[0016] When searching for drawings, the main data referenced and output by the drawing search system 1 is existing processed product information D20 related to an existing order. As described above, as an example, the existing processed product information D20 includes drawing data D21 similar to the 3D drawing data D11 and additional information D22. These data are registered and stored in the database 210. Note that the drawing data D21 may be 3D data or 2D data. As an example, the drawing data D21 may be 3D geometric data D23.

[0017] The additional information D12 and D22 is information created together with the drawing data to supplement information about the processed product. The additional information D12 and D22 includes at least one of the following additional items: part name, shape category of the processed product, dimensions, tolerance, material, and processing quantity. The additional items of the additional information D12 and D22 are not limited to the above examples, as long as they indicate the characteristics of the processed product or details of the processing process. The additional information D12 and D22 may be, for example, CAD data (an example of vector format) output by various CAD software, or image data (an example of raster format) output by scanning a drawing printed on paper media. The additional information D22 may also include at least one of the following: estimate information such as the estimated date and estimated amount; performance information such as the order date and delivery date; and profit and loss information such as the drawing gross profit and the project gross profit, in addition to the information indicating the characteristics of the processed product and details of the processing process.

[0018] The shape category is a classification of the outer shape of a material before processing or the outer shape of a processed product after processing. Examples of shape categories include, but are not limited to, "plate," "rod," "pipe," and "box."

[0019] Dimensions are expressions of the outer shape of a material before processing or the outer shape of a processed product using two or more variables. Dimensions may be expressed using, for example, three variables (width x length x thickness), three variables (outer diameter x inner diameter x length), or two variables (outer diameter x length), but are not limited to these. Different dimension expressions may also be used depending on the shape category. For example, if the shape category is "plate," the dimension may be expressed using three variables (width x length x thickness), and if the shape category is "rod" or "tube," the dimension may be expressed using three variables (outer diameter x inner diameter x length).

[0020] Tolerances are the ranges of tolerances that are allowed for inevitable errors, such as general tolerances, dimensional tolerances, geometric tolerances, and fit tolerances, and are set appropriately to reduce processing costs while ensuring the required quality and reliability.

[0021] The material indicates the material of the raw material before processing. The material may be specified by either the formal name or the abbreviated name.

[0022] As shown in Fig. 1, the drawing search system 1 includes an information processing device 2 and a user terminal device 3. The information processing device 2 and the user terminal device 3 are connected to a wired or wireless network 4 and configured to be able to transmit and receive various data to and from each other. Note that the number of information processing devices 2 and user terminal devices 3 and the connection configuration of the network 4 are not limited to the example in Fig. 1 and may be changed as appropriate.

[0023] The information processing device 2 is a server-type computer or a cloud-type computer, and is configured by a general-purpose or dedicated computer (see FIG. 7 described later), etc. The information processing device 2 performs machine learning of a feature amount inference model 220. The information processing device 2 uses the trained feature amount inference model 220 to search for drawing data D21 similar to new 3D drawing data D11.

[0024] The user terminal device 3 is a client-type computer, and is configured as a general-purpose or dedicated computer (see FIG. 7 described later), etc. In order to input new 3D drawing data, the user terminal device 3 accepts various input operations via a display screen such as an app or browser, and outputs various information via the display screen or voice.

[0025] (Configuration of information processing device 2) FIG. 3 is a block diagram showing an example of the information processing device 2. As shown in FIG.

[0026] The information processing device 2 includes a control unit 20, a data storage unit 21, a trained model storage unit 22, a communication unit 23, an input unit 24, and an output unit 25.

[0027] The communication unit 23 is connected to an external device (e.g., the user terminal device 3, etc.) via the network 4, and functions as a communication interface for transmitting and receiving various types of data. The input unit 24 accepts various input operations, and the output unit 25 outputs various types of information via a display screen or voice, thereby functioning as a user interface. Note that the input unit 24 and the output unit 25 may be omitted.

[0028] The data storage unit 21 stores a database 210 and an information processing program 211. The database 210 registers a plurality of pieces of existing processed product information D20 (that is, drawing data D21 and additional information D22).

[0029] The trained model storage unit 22 stores trained feature quantity inference models 220. The feature quantity inference models 220 stored in the trained model storage unit 22 may be provided to other devices via the network 4, a recording medium, or the like. Furthermore, the number of feature quantity inference models 220 stored in the trained model storage unit 22 is not limited to one, and multiple inference models with different conditions, such as differences in machine learning methods or data, may be stored and made available selectively or in parallel.

[0030] 3, the data storage unit 21 and the trained model storage unit 22 are shown as two storage units, but they may be configured as a single storage unit or three or more storage units. Furthermore, at least one of the data storage unit 21 and the trained model storage unit 22 may be configured as a storage unit of an external computer (for example, a server-type computer or a cloud-type computer).

[0031] By executing the information processing program 211 recorded in the data storage unit 21, the control unit 20 functions as a transmission / reception control unit 200, a database management unit 201, a feature inference model learning unit 202, a new processed product information acquisition unit 203, a drawing data conversion unit 204, and a similar drawing extraction unit 205.

[0032] (Transmission and reception control unit 200) The transmission / reception control unit 200 transmits and receives various types of data to and from an external device (such as the user terminal device 3). For example, the transmission / reception control unit 200 transmits display information for outputting various display screens to the user terminal device 3 to the user terminal device 3, and receives operation information for accepting input operations performed on the display screen of the user terminal device 3 from the user terminal device 3. The transmission / reception control unit 200 cooperates with the database management unit 201, the feature inference model learning unit 202, the new processed product information acquisition unit 203, the drawing data conversion unit 204, and the similar drawing extraction unit 205 to transmit display information and receive operation information.

[0033] (Database Management Department 201) When the database management unit 201 receives existing processed product information D20 from the user terminal device 3 for an existing order, the database management unit 201 associates the drawing data D21 and the additional information D22 included in the existing processed product information D20 and registers them in the database 210. As described above, the drawing data D21 may be 3D data or 2D data. As an example, for a new order, after 3D geometric data D13 is generated by a drawing data conversion process (described in detail later), the database management unit 201 registers the 3D geometric data D13 in the database 210 as drawing data D21. This increases the number of data registered in the database 210, thereby improving processing accuracy.

[0034] The various pieces of information registered in the database 210 may be made accessible to the user terminal device 3 by the transmission / reception control unit 200. In this case, editing operations such as addition, deletion, and modification of each piece of data may be performed on the display screen of the user terminal device 3. Furthermore, the information registered as the existing processed product information D20 in the database 210 is not limited to the drawing data D21, the incidental information D22, and the 3D geometric data D23, and any information may be associated as needed.

[0035] (Feature inference model learning unit 202) 4 is a functional explanatory diagram showing an example of the feature amount inference model learning unit 202. The feature amount inference model learning unit 202 includes a learning data acquisition unit 202A and a machine learning unit 202B.

[0036] The learning data acquisition unit 202A refers to the database 210 and acquires learning data D30 made up of input data and output data.

[0037] The input data constituting the learning data D30 is feature quantity inference model input information based on at least the 3D geometric data D23 of the existing processed product information D20. The feature quantity inference model input information may be based on the 3D geometric data D23 included in the existing processed product information D20, or may include, for example, the 3D geometric data D23 and the accompanying information D22. The feature quantity inference model input information may also include at least outer shape data obtained by cutting out an area in which the outer shape of an existing processed product is described.

[0038] The output data constituting the training data D30 are the features of the existing processed product information D20. The features can be extracted, for example, by treating each voxel of the voxelized 3D geometric data D13 as a three-dimensional binary array (e.g., occupied: 1, unoccupied: 0) and calculating statistics of the voxel occupancy rate. As an example of a process for extracting features, the voxelized 3D geometric data D13 may be treated as a voxel occupancy state, which may then be input into a convolutional neural network (CNN) to extract the features. Furthermore, features can be extracted efficiently by combining and applying 3DSIFT, shape descriptors such as moment descriptors, distance transforms, principal component analysis (PCA), and the like.

[0039] The learning data D30 is data used as teacher data (training data), verification data, and test data in supervised learning. The output data constituting the learning data D30 is data used as a correct answer label in supervised learning.

[0040] The learning data acquiring unit 202A may acquire the learning data D30 from data that satisfies a predetermined condition among the data registered in the database 210. The learning data acquiring unit 202A may also acquire the learning data D30 by another method instead of or in addition to the database 210. For example, the learning data acquiring unit 202A may acquire the learning data D30 in cooperation with an external device connected via the network 4, or may acquire the learning data D30 by accepting an input operation via the input unit 24 and the output unit 25.

[0041] The machine learning unit 202B performs machine learning to make the feature amount inference model 220 learn the correlation between the input data and the output data, using the plurality of sets of learning data D30 acquired by the learning data acquisition unit 202A.

[0042] The machine learning unit 202B causes the feature quantity inference model 220 to learn the correlation between the input data and the output data by performing distance learning so that the more similar the 3D geometric data D23 are, the smaller the distance when comparing the feature quantities, or the more dissimilar the 3D geometric data D23 are, the larger the distance when comparing the feature quantities. The trained feature quantity inference model 220 is stored in the trained model storage unit 22.

[0043] The feature inference model 220 may be, for example, a neural network, a vision transformer, a hash algorithm, or the like, but is not limited to these.

[0044] Furthermore, when the input data is 3D geometric data D23 and additional information D22, the feature inference model 220 may be configured as a single inference model that takes both of them as input data and outputs features, or may be configured as an inference model that takes 3D geometric data D23 as input data and outputs features, and an inference model that takes additional information D22 as input data and outputs features, and the features output by the two inference models are combined to output a final feature.

[0045] Furthermore, the feature inference model 220 may be, for example, an inference model pre-trained using a dataset for general image recognition, or an inference model with randomly initialized parameters. In this case, the machine learning unit 202B may perform additional learning, such as fine tuning or transfer learning, on the feature inference model 220. As a method of additional learning, the machine learning unit 202B may perform distance learning on the feature inference model 220, for example, so that the more similar the 3D geometric data D23 are, the smaller the distance between the features when comparing them, or the more dissimilar the 3D geometric data D23 are, the larger the distance between the features when comparing them. This allows the feature inference model 220 to extract similar 3D geometric data D23 by evaluating the distance between the features. The machine learning method is not limited to the above example and can be selected appropriately depending on the feature inference model 220.

[0046] The timing at which the machine learning unit 202B performs machine learning of the feature inference model learning unit 202 may be when the number of newly registered data in the database 210 exceeds a predetermined number, or when an instruction is received from the user U, but is not limited to these.

[0047] (New processed product information acquisition department 203) 5 is a functional explanatory diagram showing an example of the new processed product information acquisition unit 203 and the drawing data conversion unit 204. The new processed product information acquisition unit 203 acquires new processed product information D10 including 3D drawing data D11 and additional information D12 of a new processed product. The new processed product information acquisition unit 203 acquires the new processed product information D10 by receiving the new processed product information D10 from the user terminal device 3, for example, as a request for an estimate regarding a new processed product.

[0048] The new processed product information acquisition unit 203 may acquire part or all of the incidental information D12 from the 3D drawing data D11 by reading characters, numbers, symbols, notes, etc. contained in the 3D drawing data D11. For example, if text information is embedded in the 3D drawing data D11, the incidental information D12 is acquired by reading the text information. Alternatively, the new processed product information acquisition unit 203 may perform optical character recognition (OCR) on the 3D drawing data D11 to read the text information and acquire the incidental information D12. At this time, the acquired result of the incidental information D12 may be displayed on the display screen of the user terminal device 3, and the user U may perform an editing operation.

[0049] (Drawing data conversion unit 204) The drawing data conversion unit 204 performs a drawing data conversion step of converting the 3D drawing data D11 included in the new processed product information D10 acquired by the new processed product information acquisition unit 203 into 3D geometric data D13 extracted from information related to the three-dimensional structure of the new processed product. In the drawing data conversion step, the 3D drawing data D11 is converted into voxelized 3D geometric data D13, as shown in FIG. 5, for example.

[0050] Voxelization, a spatial partitioning method, divides 3D space into uniform cubic grids (voxels) and determines whether structures represented by 3D drawing data occupy those grids. Resolution can also be adjusted by adjusting the grid resolution (size). In the drawing data conversion process for voxelization, distance field analysis (DFT) can be used to add distance information from the surface to each voxel to determine whether each voxel is surface or interior, and evaluate the surface condition. Alternatively, ray casting can be used to determine whether each voxel is surface or interior by firing a ray in any direction and detecting the intersection of the ray with the surface, or by utilizing a sampling method to accurately identify the intersection. In this way, adjusting the resolution by combining voxelization with distance field analysis, ray casting, sampling, and other methods can improve computational efficiency and analysis accuracy.

[0051] Furthermore, if the voxelized 3D geometric data D13 is a surface model that retains information about a hollow shape using the above-mentioned ray casting method or the like, the drawing data conversion unit 204 may convert the voxelized grid in the hollow portion into a solid model that retains information about the solid shape, and output this. In this way, information about the interior of the three-dimensional structure indicated by the 3D drawing data, specifically, the internal structure, material, density, etc., can be reflected in the 3D geometric data, thereby improving the accuracy of subsequent feature acquisition.

[0052] In this embodiment, the 3D geometric data D13 is converted into voxel data, but the 3D geometric data D13 may be point cloud data or mesh data that are generally used to represent an object in a three-dimensional space. That is, the drawing data conversion unit 204 may convert the 3D drawing data D11 into point cloud data or mesh data.

[0053] (Similar drawing extraction unit 205) 6 is a functional explanatory diagram showing an example of the similar drawing extraction unit 205. The similar drawing extraction unit 205 includes a feature amount acquisition unit 205A and a similar processed product information extraction unit 205B.

[0054] The feature acquisition unit 205A acquires the features of the new processed product information D10 by inputting feature inference model input information based on 3D geometric data D13 converted from 3D drawing data D11 included in the new processed product information D10 into the feature inference model 220.

[0055] The feature amount inference model 220 is obtained by machine learning performed by the feature amount inference model learning unit 202 and stored in the learned model storage unit 22. The feature amount acquisition unit 205A performs preprocessing on the new processed product information D10 in accordance with the definition of input data in the feature amount inference model 220, and inputs the preprocessed information to the feature amount inference model 220. The feature amount acquisition unit 205A may input feature amount inference model input information based on at least the 3D geometric data D23 of the existing processed product information D20 to the feature amount inference model 220, thereby acquiring the features of the existing processed product information D20 and registering them in the database 210.

[0056] The similar processed product information extraction unit 205B extracts one or more pieces of existing processed product information D20 similar to the new processed product information D10 based on the feature quantities of the new processed product information D10 acquired by the feature quantity acquisition unit 205A from the database 210 in which a plurality of pieces of existing processed product information D20 including 3D geometric data D23 and accompanying information D22 of existing processed products are registered. Hereinafter, the existing processed product information D20 similar to the new processed product information D10 among the plurality of pieces of existing processed product information D20 will be referred to as similar processed product information.

[0057] As a method for extracting similar processed product information, the similar processed product information extraction unit 205B extracts similar processed product information based on the similarity when comparing the features of the new processed product information D10 with the features of the existing processed product information D20, for example, using the features of the existing processed product information D20.

[0058] If the feature amounts of the existing processed product information D20 are associated with the existing processed product information D20 and registered in the database 210, the similar processed product information extraction unit 205B may extract similar processed product information based on the similarity between the feature amounts of the new processed product information D10 and the feature amounts of the existing processed product information D20 registered in the database 210. If the feature amounts of the existing processed product information D20 are not associated with the existing processed product information D20 and registered in the database 210, the feature amount acquisition unit 205A may acquire the feature amounts of the existing processed product information D20 by inputting feature amount inference model input information based on at least the 3D geometric data D23 of the existing processed product information D20 to the feature amount inference model 220, and the similar processed product information extraction unit 205B may extract similar processed product information based on the similarity between the feature amounts of the new processed product information D10 and the feature amounts of the existing processed product information D20 acquired by the feature amount acquisition unit 205A.

[0059] The similarity is defined, for example, as the distance between the features when comparing them. When the features are defined as vector data, the distance between the features may be determined using, but is not limited to, a distance index such as Euclidean distance, Manhattan distance, Chebyshev distance, or Mahalanobis distance, or a similarity index such as cosine similarity.

[0060] The extraction method and number of similar processed product information items extracted from the database 210 can be set as appropriate. For example, a predetermined number or a predetermined percentage of existing processed product information items D20 may be extracted as similar processed product information items in descending order of similarity. Alternatively, a predetermined number or a predetermined percentage of existing processed product information items D20 may be selected as candidates in descending order of similarity, and then a narrowing down or rearrangement of the similarity rankings may be performed based on the additional information D12 included in the new processed product information D10 and the additional information D22 included in the existing processed product information D20. Based on the results, a predetermined number or a predetermined percentage of existing processed product information items D20 may be extracted as similar processed product information items. The narrowing down method is not particularly limited, but examples include a method of narrowing down the candidates to only those that match a specific additional item (e.g., shape category) in the additional information D12, or a method of using a specific additional item (e.g., dimension) in the additional information D12 as a criterion and narrowing down the candidates to only those that fall within a predetermined range based on the criterion. The method for changing the ranking of similarities is not particularly limited, but an example is a method of increasing the ranking of candidates that match specific auxiliary items in the auxiliary information D12.

[0061] The similar processed product information extraction unit 205B may limit the search range of the database 210 based on at least the incidental information D12 of the new processed product information D10, and extract similar processed product information based on the feature quantities of the new processed product information D10 from the database 210 limited by the search range. The similar processed product information extraction unit 205B may also output the similarity between the extracted similar processed product information and the new processed product information.

[0062] The method for limiting the search range of the database 210 is not particularly limited, but examples include a method for limiting the search range of the database 210 so that existing processed product information D20 that matches the additional items (e.g., shape category) of the additional information D12 is extracted. This makes it possible to remove from the search range existing processed products whose characteristics are significantly different from those of the new processed product, thereby enabling similar processed product information to be extracted with high accuracy.

[0063] (Hardware configuration of each device) 7 is a hardware configuration diagram showing an example of a computer 900. The information processing device 2 and the user terminal device 3 in the drawing search system 1 are configured by a general-purpose or dedicated computer 900.

[0064] 7, the computer 900 includes, as its main components, a bus 910, a processor 912, a memory 914, an input device 916, an output device 917, a display device 918, a storage device 920, a communication I / F (interface) unit 922, an external device I / F unit 924, an I / O (input / output) device I / F unit 926, and a media input / output unit 928. Note that the above components may be omitted as appropriate depending on the application of the computer 900.

[0065] The processor 912 is composed of one or more arithmetic processing devices (such as a central processing unit (CPU), a micro-processing unit (MPU), a digital signal processor (DSP), or a graphics processing unit (GPU)), and operates as a control unit that controls the entire computer 900. The memory 914 stores various data and programs 930, and is composed of, for example, a volatile memory (such as a DRAM or SRAM) that functions as a main memory, a non-volatile memory (ROM), a flash memory, etc.

[0066] The input device 916 is composed of, for example, a keyboard, a mouse, a numeric keypad, an electronic pen, a microphone, etc., and functions as an input unit. The output device 917 is composed of, for example, a sound (audio) output device, a vibration device, etc., and functions as an output unit. The display device 918 is composed of, for example, a liquid crystal display, an organic EL display, electronic paper, a projector, etc., and functions as an output unit. The input device 916 and the display device 918 may be integrated into one device, such as a touch panel display. The storage device 920 is composed of, for example, an HDD, an SSD, etc., and functions as a storage unit. The storage device 920 stores various data necessary for executing the operating system and the program 930.

[0067] The communication I / F unit 922 is connected to a network 940 (which may be the same as network 4 in FIG. 1 ) such as the Internet or an intranet via a wired or wireless connection, and functions as a communication unit that transmits and receives data to and from other computers in accordance with a predetermined communication protocol. The external device I / F unit 924 is connected to an external device 950 (such as a camera, printer, scanner, or reader / writer) via a wired or wireless connection, and functions as a communication unit that transmits and receives data to and from the external device 950 in accordance with a predetermined communication protocol. The I / O device I / F unit 926 is connected to I / O devices 960 (such as various sensors and actuators), and functions as a communication unit that transmits and receives various signals and data, such as detection signals from sensors and control signals to actuators, to and from the I / O devices 960. The media input / output unit 928 is formed by a drive device such as a DVD drive or a CD drive, and reads and writes data from and to media (non-transitory storage media) 970 (such as DVDs and CDs).

[0068] In the computer 900 having the above configuration, the processor 912 loads a program 930 stored in the storage device 920 into the memory 914, executes the program, and controls each unit of the computer 900 via the bus 910. The program 930 may be stored in the memory 914 instead of the storage device 920. The program 930 may be recorded on the medium 970 in an installable file format or an executable file format and provided to the computer 900 via the media input / output unit 928. The program 930 may be provided to the computer 900 by being downloaded via the network 940 via the communication I / F unit 922. Furthermore, the computer 900 may implement various functions realized by the processor 912 executing the program 930 using hardware such as an FPGA or an ASIC.

[0069] The computer 900 is an electronic device of any type, such as a desktop computer or a portable computer. The computer 900 may be a client computer, a server computer, a cloud computer, or an embedded computer such as a control panel or a controller (including a microcomputer, a programmable logic controller, or a sequencer).

[0070] (Operation of information processing device 2) Fig. 8 is a flowchart showing an example of the operation (information processing method) of the information processing device 2. Note that the information processing device 2 shown in Fig. 8 executes machine learning of the feature quantity inference model 220 in advance before accepting new processed product information related to a new order from the user terminal device 3 operated by the user U. The following description will be given assuming that the database 210 has a plurality of pieces of existing processed product information D20 registered by the database management unit 201 for a plurality of existing orders.

[0071] First, in step S100 (learning data acquisition step), the learning data acquisition unit 202A of the information processing device 2 acquires learning data D30 based on a plurality of pieces of existing processed product information D20 stored in the database 210. The learning data D30 includes supplementary information D22 (specifically, information such as the shape category, dimensions, and material of the processed product) and 3D geometric data D23.

[0072] In step S110 (machine learning step), the machine learning unit 202B uses the acquired learning data D30 to perform machine learning of the feature amount inference model 220. Through this learning step, a model for inferring feature amounts from the new processed product information D10 is constructed.

[0073] In step S120 (new processed product information acquisition step), the new processed product information acquisition unit 203 acquires new processed product information D10 from the user terminal device 3. The new processed product information D10 includes 3D drawing data D11 and additional information D12.

[0074] In step S130 (drawing data conversion step), the drawing data conversion unit 204 converts the 3D drawing data D11 included in the new processed product information D10 into 3D geometric data D13. If the converted 3D geometric data D13 is a surface model that holds information about a hollow shape, the drawing data conversion unit 204 may convert it into a solid model that holds information about a solid shape.

[0075] In step S140 (feature amount acquisition step), the feature amount acquisition unit 205A inputs feature amount inference model input information based on the 3D geometric data D13 to the trained feature amount inference model 220, and acquires the feature amounts of the new processed product information D10.

[0076] In step S150 (similar processed product information extraction step), the similar processed product information extraction unit 205B extracts similar processed product information from the existing processed product information D20 in the database 210 based on the feature amount of the acquired new processed product information D10.

[0077] In step S160 (output step), the extracted similar processed product information is output together with the degree of similarity with the new processed product information D10 to the user terminal device 3. This allows the user to check information on existing processed products similar to the new processed product information D10.

[0078] As described above, the information processing device 2 and information processing method according to this embodiment acquire new processed product information D10 including 3D drawing data D11 of a new processed product, convert the acquired 3D drawing data D11 into 3D geometric data D13, and extract, from a database in which multiple pieces of existing processed product information D20 including drawing data D21 of existing processed products are registered, one or more pieces of drawing data D21 of the existing processed product that are similar to the 3D geometric data D13 converted from the 3D drawing data D11 included in the new processed product information D10. This makes it possible to search for similar drawing data with high accuracy even when 3D drawing data is input.

[0079] (Other embodiments) The present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit and scope of the present invention, all of which are included in the technical concept of the present invention.

[0080] For example, in the above embodiment, the similar drawing extraction unit 205 uses a trained model based on machine learning to extract drawing data of multiple existing processed products that are similar to the 3D geometric data converted from the 3D drawing data, but this is not limited to this. That is, the similar drawing extraction unit 205 may extract drawing data of multiple existing processed products that are similar to the 3D geometric data by using a known image processing technique without using a trained model based on machine learning.

[0081] In the above embodiment, the feature inference model learning unit 202 uses the features of existing processed product information as learning data, but this is not limited to this. That is, data such as classified class labels may be used as learning data without using features. Furthermore, the similar drawing extraction unit 205 may calculate the similarity of drawings using an autoencoder that uses 3D geometric data as input data, without acquiring features of 3D geometric data converted from 3D drawing data, and extract drawing data of multiple existing processed products that are similar to the 3D geometric data.

[0082] In the above embodiment, the information processing device 2 has been described as being configured as a single device, but it may be configured as a plurality of devices. For example, the units 202 to 205 of the information processing device 2 may be distributed across a plurality of devices, so that the information processing device 2 is configured as a machine learning device that includes the feature quantity inference model learning unit 202 and performs the feature quantity inference model learning step, and an information processing device that includes the new processed product information acquisition unit 203, the drawing data conversion unit 204, and the similar drawing extraction unit 205 and performs the similar drawing extraction step. In this case, each unit (each step) of each of the above devices may be realized by a program (information processing program or machine learning program) that can be executed by the computer 900.

[0083] Each of the above devices can be configured, for example, as follows: The configuration and operation of each part of each device and the various types of data handled by each device are the same as those in the above embodiment, and therefore detailed description thereof will be omitted.

[0084] The machine learning device that performs the feature inference model learning step includes a drawing data conversion unit 204 that converts 3D drawing data included in existing processed product information D20 into 3D geometric data D23, a learning data acquisition unit 202A that acquires multiple sets of learning data D30 composed of input data, a machine learning unit 202B that causes a feature inference model 220 to learn a correlation between the input data and output data using the multiple sets of learning data D30 acquired by the learning data acquisition unit 202A, and a learned model storage unit 22 that stores the feature inference model 220 that has learned the correlation by the machine learning unit 202B. The input data is feature inference model input information based on at least the 3D geometric data D23 of the existing processed product information D20. The output data is feature values ​​of the existing processed product information D20.

[0085] The information processing device that performs the similar drawing extraction process includes a new processed product information acquisition unit 203 that acquires new processed product information D10 including 3D drawing data, a drawing data conversion unit 204 that converts the 3D drawing data D11 acquired by the new processed product information acquisition unit 203 into 3D geometric data D13, a feature acquisition unit 205A that acquires the features of the new processed product information D10 by inputting feature inference model input information based on the 3D geometric data D13 converted by the drawing data conversion unit 204 to a feature inference model, and a similar processed product information extraction unit 205B that extracts multiple existing processed product information D20 that is similar to the new processed product information D10 based on the features of the new processed product information D10 acquired by the feature acquisition unit 205A from a database in which multiple existing processed product information D20 is registered.

[0086] The transmission / reception control unit 200 and the database management unit 201 may be provided in each of the above devices. The database 210 may be configured to be accessible from each of the above devices, and may be stored in a storage unit of one of the devices, or in a storage unit of an external computer. Furthermore, some of the above devices may be realized by the user terminal device 3.

[0087] (Inference device, inference method and inference program) The present invention can be provided not only in the form of the information processing device 2 (information processing method or information processing program) according to the above embodiment, but also in the form of an inference device (inference method or inference program) used to infer feature quantities. In this case, the inference device (inference method or inference program) can include a memory and a processor, and the processor executes a series of processes. The series of processes includes an information acquisition process (information acquisition step) for acquiring new processed product information D10 including 3D drawing data, a drawing data conversion process (drawing data conversion step) for converting the 3D drawing data D11 into 3D geometric data D13, and an inference process (inference step) for inferring feature quantities of the new processed product information based on the 3D geometric data D13 converted by the drawing data conversion process.

[0088] By providing it in the form of an inference device (inference method or inference program), it can be more easily applied to various devices than when an information processing device is implemented. It will be naturally understood by those skilled in the art that when the inference device (inference method or inference program) infers the feature quantities of new processed product information, the inference method implemented by the feature quantity acquisition unit 205A may be applied using a trained inference model generated by the machine learning device and machine learning method according to the above embodiments. [Explanation of symbols]

[0089] 1...drawing search system, 2...information processing device, 3...user terminal device, 4...network, 20...control unit, 21...data storage unit, 22...model storage unit, 23...communication unit, 24...input unit, 25...output unit, 200...transmission / reception control unit, 201...database management unit, 202...feature inference model learning unit, 202A...learning data acquisition unit, 202B... machine learning unit, 203... new processed product information acquisition unit, 204... drawing data conversion unit, 205... Similar drawing extraction unit, 205A... Feature acquisition unit, 205B... Similar processed product information extraction unit, 210... Database, 211... Information processing program, 220... Feature inference model

Claims

1. a new processed product information acquisition unit that acquires new processed product information including 3D drawing data of the new processed product; a drawing data conversion unit that converts the 3D drawing data acquired by the new processed product information acquisition unit into 3D geometric data that extracts information about a three-dimensional structure of the new processed product; a similar drawing extraction unit that extracts, from a database in which a plurality of pieces of existing processed product information including 3D drawing data of existing processed products is registered, a plurality of pieces of existing processed product information including 3D drawing data that is converted into 3D geometric data having feature amounts similar to feature amounts of 3D geometric data converted from the 3D drawing data included in the new processed product information, The similar drawing extraction unit a feature amount acquisition unit that acquires feature amounts of the 3D geometric data converted from the 3D drawing data included in the new processed product information by inputting feature amount inference model input information including the 3D geometric data converted by the drawing data conversion unit to a feature amount inference model; a similar processed product information extraction unit that extracts from the database a plurality of pieces of existing processed product information including 3D drawing data converted into 3D geometric data having feature amounts similar to feature amounts of 3D geometric data converted from the 3D drawing data included in the new processed product information acquired by the feature amount acquisition unit, The feature quantity inference model is a trained inference model that has been trained by machine learning to determine a correlation between feature quantity inference model input information including 3D geometric data converted from 3D drawing data of the existing processed product among the existing processed product information, and the feature quantities of the 3D geometric data converted from the 3D drawing data of the existing processed product. Information processing device.

2. The 3D geometric data converted by the drawing data conversion unit is at least one of voxel data, point cloud data, and mesh data. The information processing device according to claim 1 .

3. When the converted 3D geometric data is a surface model that holds information about a hollow shape, the drawing data conversion unit converts the 3D geometric data into a solid model that holds information about a solid shape. The information processing device according to claim 1 .

4. the similar drawing extraction unit outputs a similarity of feature amounts between the extracted existing processed product information and the new processed product information. The information processing device according to claim 1 .

5. A machine learning device for a feature inference model used when searching for existing processed product information of an existing processed product having a feature similar to a feature of new processed product information of a new processed product, a drawing data conversion unit that converts 3D drawing data of the existing processed product included in existing processed product information of the existing processed product into 3D geometric data extracted with information on a three-dimensional structure of the existing processed product; a learning data acquisition unit that acquires a plurality of sets of learning data each composed of input data and output data; a machine learning unit that causes a feature inference model to learn a correlation between the input data and the output data using a plurality of sets of the learning data acquired by the learning data acquisition unit; a learned model storage unit that stores the feature quantity inference model that has learned the correlation by the machine learning unit, The input data is feature inference model input information including 3D geometric data converted from 3D drawing data of the existing processed product included in the existing processed product information of the existing processed product, and the output data is feature amounts of the 3D geometric data converted from the 3D drawing data of the existing processed product. Machine learning device.

6. An information processing method executed by an information processing device having a memory and a processor, a new processed product information acquisition step of acquiring new processed product information including 3D drawing data of the new processed product; a drawing data conversion step of converting the 3D drawing data acquired in the new processed product information acquisition step into 3D geometric data extracted with information about the three-dimensional structure of the new processed product; a similar drawing extraction step of extracting, from a database in which a plurality of pieces of existing processed product information including 3D drawing data of existing processed products is registered, a plurality of pieces of existing processed product information including 3D drawing data converted into 3D geometric data having feature amounts similar to feature amounts of 3D geometric data converted from the 3D drawing data included in the new processed product information, The similar drawing extraction step includes: a feature amount acquisition step of acquiring feature amounts of 3D geometric data converted from the 3D drawing data included in the new processed product information by inputting feature amount inference model input information including the 3D geometric data converted by the drawing data conversion step into a feature amount inference model; a similar processed product information extraction step of extracting from the database a plurality of pieces of existing processed product information including 3D drawing data converted into 3D geometric data having feature amounts similar to feature amounts of 3D geometric data converted from the 3D drawing data included in the new processed product information acquired in the feature amount acquisition step, The feature quantity inference model is a trained inference model that has been trained by machine learning to determine a correlation between feature quantity inference model input information including 3D geometric data converted from 3D drawing data of the existing processed product among the existing processed product information, and the feature quantities of the 3D geometric data converted from the 3D drawing data of the existing processed product. Information processing methods.

7. A machine learning method in which a machine learning device including a memory and a processor executes machine learning of a feature inference model used when searching for existing processed product information of an existing processed product having feature quantities similar to feature quantities of new processed product information of a new processed product, a drawing data conversion step of converting 3D drawing data of the existing processed product included in existing processed product information of the existing processed product into 3D geometric data extracted with information on the three-dimensional structure of the existing processed product; a learning data acquisition step of acquiring a plurality of sets of learning data each composed of input data and output data; a machine learning step of causing a feature inference model to learn a correlation between the input data and the output data using the plurality of sets of learning data acquired in the learning data acquisition step; a learned model storage step of storing the feature quantity inference model that has learned the correlation through the machine learning step in the memory, The input data is feature inference model input information including 3D geometric data converted from 3D drawing data of the existing processed product included in the existing processed product information of the existing processed product, and the output data is feature amounts of the 3D geometric data converted from the 3D drawing data of the existing processed product. Machine learning methods.

Images